text
stringlengths 254
1.16M
|
|---|
---
title: 'Medication Intake, Perceived Barriers, and Their Correlates Among Adults With
Type 1 and Type 2 Diabetes: Results From Diabetes MILES – The Netherlands'
authors:
- Stijn Hogervorst
- Marce C. Adriaanse
- Jacqueline G. Hugtenburg
- Mariska Bot
- Jane Speight
- Frans Pouwer
- Giesje Nefs
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2021
pmcid: PMC10012124
doi: 10.3389/fcdhc.2021.645609
license: CC BY 4.0
---
# Medication Intake, Perceived Barriers, and Their Correlates Among Adults With Type 1 and Type 2 Diabetes: Results From Diabetes MILES – The Netherlands
## Abstract
### Purpose
The purpose of this study is to investigate medication intake, perceived barriers and their correlates in adults with type 1 or type 2 diabetes.
### Methods
In this cross-sectional study, 3,383 Dutch adults with diabetes ($42\%$ type 1; $58\%$ type 2) completed the 12-item ‘Adherence Starts with Knowledge’ questionnaire (ASK-12; total score range: 12-60) and reported socio-demographics, clinical and psychological characteristics and health behaviors. Univariable and multivariable logistic regression analyses were used.
### Results
Adults with type 1 diabetes had a slightly lower mean ASK-12 score (i.e. more optimal medication intake and fewer perceived barriers) than adults with non-insulin-treated type 2 diabetes. After adjustment for covariates, correlates with suboptimal intake and barriers were fewer severe hypoglycemic events and more depressive symptoms and diabetes-specific distress. In type 2 diabetes, correlates were longer diabetes duration, more depressive symptoms and diabetes-specific distress.
### Conclusions
Adults with type 1 diabetes showed slightly more optimal medication intake and fewer perceived barriers than adults with non-insulin treated type 2 diabetes. Correlates differed only slightly between diabetes types. The strong association with depressive symptoms and diabetes-specific distress in both diabetes types warrants attention, as improving these outcomes in some people with diabetes might indirectly improve medication intake.
## Introduction
The global prevalence of both type 1 and type 2 diabetes is at an all-time high while incidence and prevalence rates continue to increase [1, 2]. Diabetes-related costs represent one of the highest expenditures in healthcare systems and can be attributed mainly to the high morbidity associated with diabetes (3–6).
The cornerstone of optimal glycemic outcomes in both type 1 and type 2 diabetes relies on diligent self-management, of which medication intake is a central element [7, 8]. However, only $50\%$ of adults with type 2 diabetes have an A1C of <$7\%$ (53 mmol/mol) [9]. For type 1 diabetes, the rate varies between 20-$60\%$, depending on country and age group [10]. The difficulty in meeting these target values is often attributed to suboptimal medication intake [3]. In adults with type 2 diabetes optimal medication intake ranges from $39\%$ to $93\%$ with large differences between studies, populations and measurement methods [11].
People have a myriad of reasons for not taking their medication as recommended, as illustrated by Kardas et al. who identified no less than 771 different factors in a review of systematic reviews [12]. There is a vast body of research exploring medication intake and its correlates in adults with type 2 diabetes. Using a model developed by the World Health Organization [13], the most relevant factors can be categorized as person-related (e.g. younger age, low health literacy, being male, having depression), socio-economic (e.g. affordability of medication, less social support), diabetes-related (e.g. shorter diabetes duration), treatment-related (e.g. more complicated medication regimen, more side-effects, insulin use) and healthcare-related (e.g. patient-clinician communication, lack of time for adequate care, lack of integrated care) (14–16). Far less is known about factors influencing medication intake in adults with type 1 diabetes. They were found to have a two-fold higher risk of “medication errors” at hospital admission as compared with adults with type 2 diabetes, defined by the authors as “unintentional medication discrepancies corrected by physicians” of very serious (potentially leading to life-threatening consequences) or serious (potentially causing harm or extended hospital stay) nature [17]. Of note, people with type 1 diabetes were more often admitted through the emergency department, more often had medication errors involving added medications and more medication errors per treatment compared to people with type 2 diabetes. Understanding the factors that influence medication intake will help clinicians and policy makers to provide better support to adults with diabetes in maintaining the medication intake necessary to achieve optimal glycemic control.
The characteristics of people with type 1 and type 2 diabetes differ on a number of important points. Around half of adults with type 1 diabetes have had the condition since childhood or adolescence whereas those with type 2 diabetes generally developed the condition later in life. In terms of etiology, type 1 diabetes is characterized by an absolute insulin deficiency resulting from an autoimmune process, whereas type 2 is characterized by insulin resistance and a relative insulin deficiency related to factors such as obesity, increasing age and genetic disposition [1]. In line with this, the treatment approaches for both types are different. First, type 1 diabetes is always treated with insulin therapy (through injections or pump) and monitoring of glucose levels with a blood glucose meter or a continuous glucose monitoring system (CGM). Type 2 diabetes can often be treated with a combination of a diet and oral medication. When oral medication is no longer sufficient, insulin injection therapy can be added. These differences also translate into different diabetes self-management activities. For example, people with type 1 diabetes often apply complex treatment regimens including multiple daily insulin doses and carbohydrate counting. On the other hand people with type 2 diabetes do not always have to inject insulin, but cope more often with multi-morbidity and multi-pharmacy. Consequently, medication intake, perceived barriers and their correlates might also differ between diabetes types.
A better understanding of medication intake, perceived barriers and their correlates will help clinicians, policy makers and researchers to tailor treatment and interventions to the needs of individual persons with diabetes from both target populations. Therefore, the aims of the present study were: (a) to investigate (self-reported) medication intake and perceived barriers in adults with type 1 and type 2 diabetes; and (b) to identify socio-demographic, psychological and clinical characteristics, and health behaviors that influence medication intake and perceived barriers.
## Participants and Procedure
Data were extracted from Diabetes MILES (Management and Impact for Long-term Empowerment and Success) – The Netherlands study. The MILES study is a national, online, cross-sectional observational study designed to examine the psychosocial aspects of living with diabetes. The study’s rationale, design and procedure has been described in detail elsewhere [18]. From September to October 2011, Dutch adults (aged ≥19 years, no upper age limit) with self-reported diabetes of any type were offered the opportunity through several media channels to participate in an online survey. The study protocol was approved by the Psychological Research Ethics Committee of Tilburg University, The Netherlands (EC-2011 5). Written informed consent was obtained digitally from all participants.
The subsample in the present study ($$n = 3$$,383) includes all participants with self-reported type 1 or type 2 diabetes who completed the Adherence Starts with Knowledge 12 (ASK-12) questionnaire and who indicated that they are prescribed medication for their diabetes (either insulin pump or injection, GLP-1 injection, blood glucose lowering tablets or a combination of those) [19].
## Medication Intake
Medication intake and perceived barriers were assessed using the ASK-12 [19]. The ASK-12 questionnaire has demonstrated validity and internal reliability consistency in people with type 2 diabetes and was validated both by making use of another questionnaire (the Morisky Medication Adherence Scale) and by making use of filled medication prescription data from pharmacies [19]. In the current study Cronbach’s alpha indicated that the ASK-12 is reliable for both adults with type 1 and type 2 diabetes (Cronbach’s alpha = 0.66 and 0.68 respectively). Additionally, lower ASK-12 total scores (i.e. more favorable medication intake and less perceived barriers) are associated with lower HbA1c in both type 1 diabetes (Logistic regression analysis, $$n = 1172$$, $B = 0.039$, p-value = 0.001) and type 2 diabetes with insulin (logistic regression analysis, $$n = 718$$, $B = 0.031$, p-value = 0.026) and type 2 diabetes without insulin (logistic regression analysis, $$n = 568$$, $B = 0.033$, p-value = 0.029).
The questionnaire includes twelve questions, rated from 1 (totally disagree) to 5 (totally agree) for the first seven questions and from 1 (in the past week) to 5 (never) for the final 5 questions. Responses are summed to generate a total score (range: 12-60). Three subscale scores can also be derived from the ASK-12: behavior (5 items, score range 5-25, e.g. ‘Have you not had medicine with you when it was time to take it?’), health beliefs (4 items, score range 4-20, e.g. ‘I feel confident that each of my medicines will help me’) and inconvenience/forgetfulness (3 items, score range 3-15, e.g. ‘Sometimes I simply forget to take my medications’). Higher ASK-12 total and subscale scores indicate suboptimal medication intake and more perceived barriers (i.e. more problematic beliefs and greater inconvenience/forgetfulness).
## Psychological Characteristics
Symptoms of depression and anxiety during the past two weeks were measured using the validated 9-item Patient Health Questionnaire (PHQ-9; total score range: 0–27) [20] and 7-item General Anxiety Disorder questionnaire (GAD-7; total score range: 0–21) [21]. For both measures, higher scores indicate higher levels of symptoms, and scores ≥10 indicate elevated symptoms of depression or anxiety. Diabetes-specific emotional distress was assessed with the validated 20-item Problem Areas in Diabetes scale (PAID; total score range: 0–100), in which higher scores indicate greater severity of diabetes-specific emotional distress [22].
## Socio-demographics, Clinical Characteristics and Health Behaviors
The Diabetes MILES – The Netherlands survey included several items on socio-demographic characteristics (i.e., sex, age, ethnic background, educational level, marital status and work status). Participants also self-reported their height/weight (enabling calculation of their Body Mass Index (BMI)), alcohol use (eight categories from 0 to ≥36 glasses per week), being a daily smoker, diabetes duration, current diabetes treatment, most recent A1C (continuous variable and dichotomized using the cut-off of ≥$7\%$/53 mmol/mol to indicate a sub-optimal glycemic outcome), the number of severe hypoglycemic events in the past year (defined as a low blood glucose level requiring assistance from another person for recovery), conditions that might be vascular complications of diabetes (e.g. myocardial infarction, stroke, peripheral arterial disease, nephropathy, retinopathy, neuropathy and/or foot ulcers) and somatic comorbid conditions (e.g. hypertension, high cholesterol, chronic heart failure, asthma/chronic obstructive pulmonary disease (COPD) and rheumatic disorders/joint problems). The number of medications for comorbidity was calculated by making an aggregation variable of medication taken by participants for 32 different diseases, including diabetes-related complications.
## Statistical Analyses
Socio-demographics, clinical and psychological characteristics and health behaviors were described as mean ± standard deviation (SD) for continuous variables and % (n/N) for categorical variables, stratified by diabetes type. First, differences in the ASK-12 total score and the three ASK-12 subscales were compared between adults with type 1 and insulin-treated and non-insulin-treated type 2 diabetes using independent sample t-tests. Sample size adjusted effect sizes (Cohen’s d statistic) were calculated.
Subsequently, the association of socio-demographics, psychological and clinical characteristics and health behaviors with less favorable medication intake and more perceived barriers was analyzed separately by diabetes type. A ‘suboptimal medication intake and more perceived barriers’ group was created based on the ASK-12 total score, by taking the least positive scoring quartile (i.e. the $25\%$ with the highest score) for both types of diabetes. Two separate univariable logistic regression analyses were used to determine risk markers in adults with type 1 diabetes and type 2 diabetes respectively. To avoid overfitting [23], we selected the following thirteen potential correlates a priori based upon clinical considerations, literature review and availability: sex, age, education level, having a partner, diabetes duration, number of medications for comorbid conditions, number of visits with a clinician in the past year, frequency of severe hypoglycemic events in the past year, diabetes-specific distress, anxiety symptoms, depressive symptoms, alcohol use and smoking behavior. Thereafter, two multiple logistic regression analyses were conducted to create an association model with correlates that contribute to suboptimal medication intake and more perceived barriers in adults with type 1 and type 2 diabetes, respectively. All factors that showed a p-value <0.10 in the univariate analyses were added to the multivariable association model. Analyses were performed using SPSS Statistics version 24 (IBM, Somers, NY, USA). A p-value <0.05 was considered to be statistically significant.
## Sample Characteristics
The total sample consisted of 3,383 adults with diabetes, of whom $42\%$ ($$n = 1$$,422) had type 1 diabetes and $58\%$ ($$n = 1$$,961) had type 2 diabetes. Table 1 shows the socio-demographics, psychological and clinical characteristics and health behaviors of these groups.
**Table 1**
| Unnamed: 0 | N Missing | Type 1 diabetes(n=1,422) | Type 2 diabetes(n=1,961) |
| --- | --- | --- | --- |
| SOCIO-DEMOGRAPHICS | | | |
| Female sex | 1.0 | 61 (860/1421) | 49 (952/1961) |
| Age, years | 49.0 | 47.6 (14.7) | 61.7 (10.1) |
| (non-Dutch) ethnic minority | 0.0 | 2 (31/1422) | 3 (51/1961) |
| Low educational level | 8.0 | 19 (264/1421) | 32 (630/1954) |
| Being single | 0.0 | 20 (278/1422) | 21 (403/1961) |
| Paid work | 6.0 | 63 (897/1420) | 36 (708/1957) |
| Body Mass Index, kg/cm2 | 36.0 | 25.5 (4.7) | 29.8 (5.9) |
| CLINICAL CHARACTERISTICS | | | |
| Diabetes duration, years | 3.0 | 23.5 (14.6) | 11.1 (8.1) |
| Primary diabetes treatment | 18.0 | | |
| - Insulin pump | | 49 (693/1422) | 6 (116/1943) |
| - Insulin injections | | 51 (729/1422) | 50 (933/1943) |
| - GLP-1 injections | | 0 | 2 (35/1943) |
| - Blood glucose lowering tablets | | 0 | 42 (787/1943) |
| Most recent A1C, mmol/mol | 924.0 | 58 (12) | 54 (12) |
| Most recent A1C, % | 924.0 | 7,5 (3.2) | 7,1 (3.2) |
| Suboptimal A1C (≥7%, ≥53 mmol/mol) | 924.0 | 70 (826/1172) | 50 (645/1287) |
| N° severe hypoglycemic events in past year | 22.0 | 1.4 (6.4) | 0.4 (2.9) |
| N° medications for comorbid conditions | 1.0 | 1.3 (1.5) | 2.1 (1.7) |
| N° appoints with clinicians in past year | 26.0 | 19.4 (20.7) | 20.2 (22.5) |
| At least one diabetes complication | 0.0 | 32 (448/1422) | 33 (641/1961) |
| Microvascular | | | |
| - Retinopathy | 0.0 | 19 (263/1422) | 6 (126/1961) |
| - Neuropathy | 0.0 | 17 (237/1422) | 18 (350/1961) |
| - Nephropathy | 0.0 | 5 (64/1422) | 4 (79/1961) |
| - Foot condition due to diabetes | 0.0 | 4 (63/1422) | 5 (106/1961) |
| Macrovascular | | | |
| - Myocardial infarction | 0.0 | 4 (52/1422) | 7 (130/1961) |
| - Stroke | 0.0 | 1 (18/1422) | 3 (54/1961) |
| - Peripheral arterial disease | 0.0 | 3 (39/1422) | 6 (116/1961) |
| Other co-morbid chronic conditions | | | |
| - Hypertension | 0.0 | 25 (353/1422) | 51 (993/1961) |
| - High cholesterol | 0.0 | 23 (328/1422) | 45 (877/1961) |
| - Chronic heart failure | 0.0 | 1 (18/1422) | 3 (54/1961) |
| - Asthma/COPD | 0.0 | 8 (109/1422) | 12 (231/1961) |
| - Rheumatic disorders/joint problems | 0.0 | 12 (169/1422) | 20 (387/1961) |
| PSYCHOLOGICAL CHARACTERISTICS | | | |
| Anxiety symptoms (GAD7 total score) | 194.0 | 3.1 (3.6) | 2.7 (3.6) |
| Depressive symptoms (PHQ9 total score) | 188.0 | 4.3 (4.7) | 4.3 (4.6) |
| Diabetes-specific distress (PAID total score) | 174.0 | 22.3 (19.3) | 19.5 (19.1) |
| HEALTH BEHAVIOURS | | | |
| Daily smoker | 115.0 | 11 (151/1376) | 8 (155/1892) |
| Alcohol >14 glasses/week | 110.0 | 9 (125/1380) | 6 (118/1893) |
## Medication Intake by Subgroup
Self-reported medication intake and perceived barriers (ASK-12 total and subscale scores) are shown for diabetes type 1, insulin-treated diabetes type 2 and non-insulin-treated diabetes type 2 in Table 2. Adults with type 1 diabetes reported slightly more optimal medication intake and fewer perceived barriers (i.e. lower total score ASK-12) than adults with non-insulin-treated type 2 diabetes (mean score: 21.2 ± 5.6 vs. 22.0 ± 6.0 respectively, $$p \leq 0.005$$, Cohen’s $d = 0.14$). No significant difference was found between people with type 1 diabetes and people with insulin-treated type 2 diabetes. With respect to the ASK-12 subscales, adults with type 1 diabetes reported more optimal medication intake behavior than adults with insulin-treated type 2 diabetes (7.2 ± 2.6 vs. 7.5 ± 2.9 resp., $$p \leq 0.022$$, Cohen’s $d = 0.11$) and than adults with non-insulin-treated type 2 diabetes (7.2 ± 2.6 vs. 7.7 ± 2.8 resp., p=<0,001, Cohen’s $d = 0.19$). No significant differences were found between groups on the subscales health beliefs and inconvenience/forgetfulness.
**Table 2**
| Unnamed: 0 | Type 1 (n=1,422) | Type 2 insulin-treated (n=1,049) | Type 2 non-insulin treated (n=822) | Cohen’s d* | Significance |
| --- | --- | --- | --- | --- | --- |
| Medication intake and perceived barriers (12-60) | 21.2 (5.6) | 21.7 (5.8) | 22.0 (6.0) | 0.14 | B |
| Behavior (5-25) | 7.2 (2.6) | 7.5 (2.9) | 7.7 (2.8) | 0.11, 0.19 | A, B |
| Health beliefs (4-20) | 8.3 (2.9) | 8.6 (3.0) | 8.6 (3.1) | | NS |
| Inconvenience/forgetfulness (3-15) | 5.7 (2.4) | 5.6 (2.4) | 5.8 (2.4) | | NS |
## Correlates of Medication Intake and Perceived Barriers
Table 3 shows correlates with less optimal medication intake and more perceived barriers (subgroups based on the dichotomized ASK-12 total score) in adults with type 1 and type 2 diabetes, based on univariable and multivariable regression analyses.
**Table 3**
| Unnamed: 0 | Type 1 diabetes (n=1,315) | Type 1 diabetes (n=1,315).1 | Type 2 diabetes (n=1,760) | Type 2 diabetes (n=1,760).1 |
| --- | --- | --- | --- | --- |
| | Univariable a | Multivariable b | Univariable a | Multivariable b |
| | OR (95% CI) | OR (95% CI) | OR (95% CI) | OR (95% CI) |
| Socio-demographic characteristics | | | | |
| Female sex | 1.12 (0.88 – 1.44) | | 1.32 (1.08 – 1.61)** | 1.02 (0.81 – 1.28) |
| Age, years | 0.99 (0.98 – 0.99)** | 0.99 (0.98 – 1.00) | 0.98 (0.97 – 0. 98)*** | 0.99 (0.98 – 1.00) |
| Higher education level | 1.07 (0.91 – 1.25) | | 1.06 (0.93 – 1. 19) | |
| Having a partner | 0.84 (0.63 – 1.13) | | 0.72 (0.57 – 0.92)** | 0.84 (0.64 – 1.10) |
| Clinical characteristics | | | | |
| Severe hypoglycemic events past year | 0.99 (0.96 – 1.01) | 0.97 (0.93 – 1.00)* | 0.97 (0.93 – 1.02) | 0.95 (0.90 – 1.01) |
| Diabetes duration, years | 0.99 (0.98 – 1.00)* | 0.97 (0.99 – 1.05) | 0.97 (0.96 – 0.99)*** | 0.98 (0.97 – 1.00)* |
| Number of appointments clinicians past year | 1.01 (1.00 – 1.01)** | 1.00 (1.00 – 1.01) | 1.00 (1.00 – 1.01) | 1.00 (0.99 – 1.00) |
| Number of medications for comorbidity f | 1.07 (0.99 – 1.15) | 1.07 (0.97 – 1.17) | 1.03 (0.97 – 1.10) | |
| Health behaviors | | | | |
| Daily smoker | 1.38 (0.96 – 1.99) | | 1.04 (0.71 – 1.50) | |
| Alcohol use | 1.01 (0.92 – 1.11) | | 0.97 (0.92 – 1.07) | |
| Psychological characteristics | | | | |
| Anxiety symptoms (GAD7 total score) c | 1.12 (1.08 – 1.15)*** | 0.97 (0.92 – 1.02) | 1.14 (1.11 – 1.17)*** | 1.01 (0.97 – 1.06) |
| Depressive symptoms (PHQ9 total score) d | 1.12 (1.09 – 1.15)*** | 1.06 (1.01 – 1.11)** | 1.13 (1.10 – 1.15)*** | 1.07 (1.03 – 1.11)*** |
| Diabetes-specific distress (PAID total score) e | 1.04 (1.04 – 1.05)*** | 1.03 (1.02 – 1.04)*** | 1.03 (1.03 – 1.04)*** | 1.02 (1.01 – 1.03)*** |
In the univariable analyses for adults with type 1 diabetes, less optimal medication intake and more perceived barriers were correlated with younger age ($$p \leq 0.001$$), shorter diabetes duration ($$p \leq 0.023$$), more appointments with clinicians in the past year ($p \leq 0.001$), more anxiety symptoms ($p \leq 0.001$), more depressive symptoms ($p \leq 0.001$) and more diabetes-specific distress ($p \leq 0.001$). In the multivariable analysis, less optimal medication intake and more perceived barriers were correlated with fewer severe hypoglycemic events in the past year ($$p \leq 0.042$$), more depressive symptoms ($$p \leq 0.012$$) and more diabetes-specific distress ($p \leq 0.001$).
In the univariable analyses for adults with type 2 diabetes, less optimal medication intake and more perceived barriers were correlated with being female ($$p \leq 0.007$$), younger age ($p \leq 0.001$), not having a partner ($$p \leq 0.008$$), shorter duration of diabetes ($p \leq 0.001$), more anxiety symptoms ($p \leq 0.001$), more depressive symptoms ($p \leq 0.001$) and higher diabetes-specific distress ($p \leq 0.001$). In the multivariate analysis, less optimal medication intake and more perceived barriers were correlated with shorter duration of diabetes ($$p \leq 0.047$$), more depressive symptoms ($$p \leq 0.001$$) and more diabetes-specific distress ($p \leq 0.001$).
## Key Findings
The present study showed that adults with type 1 diabetes had slightly more optimal medication intake and fewer perceived barriers (i.e. lower ASK-12 total score) than adults with non-insulin-treated type 2 diabetes. Additionally, they had slightly more optimal scores on the behavior subscales than adults with both insulin-treated and non-insulin-treated type 2 diabetes, but did not differ with respect to the subscales health beliefs or inconvenience/forgetfulness. Correlates of less optimal medication intake and more perceived barriers in type 1 diabetes were fewer severe hypoglycemic events in the past year, higher depressive symptoms and higher diabetes-specific distress. Correlates of less optimal medication intake and more perceived barriers in type 2 diabetes were a shorter duration of diabetes, more depressive symptoms and more diabetes-specific distress.
## Interpretations and Comparison to Literature
Adults with type 1 diabetes showed slightly more optimal medication intake and fewer perceived barriers than adults with non-insulin-treated type 2 diabetes and showed slightly more optimal medication intake behavior than adults with both insulin-treated and non-insulin-treated type 2 diabetes. To our knowledge, there has been no previous research comparing levels of medication intake and perceived barriers between adults with type 1 and type 2 diabetes. However, our findings need to be interpreted with care, as the absolute difference on the ASK-12 total score was very small and there is no clinically relevant cut-off point on the ASK-12 for less optimal medication intake and more perceived barriers in diabetes patients [19]. Moreover, Diabetes MILES – The *Netherlands is* an online, self-report survey, and probably less representative than a population-based study. The largest difference was found on the medication intake behavior subscale, which describes the degree to which a person intentionally misses a dose for various reasons.
The slightly more optimal medication intake and fewer perceived barriers in people with type 1 diabetes compared to people with type 2 diabetes could be explained by differences in pathophysiology. Specifically, people with type 1 diabetes experience more immediate risks of ketoacidosis when not administering insulin, whereas the direct consequences are less dire when people with type 2 diabetes miss multiple doses of glucose lowering tablets. Correlates of less optimal medication intake and more perceived barriers in type 1 diabetes were fewer severe hypoglycemic events in the past year, higher depressive symptoms and higher diabetes-specific distress. With regard to correlates of medication intake and perceived barriers in adults with type 1 diabetes, the literature is unclear. Main reasons are that most studies look at glycemic outcomes or HbA1c as outcome rather than medication intake. Additionally, most studies differ in terms of determinants included. Our findings for the correlates of less optimal medication intake and more perceived barriers in adults with type 2 diabetes are consistent with previous research showing the role of higher depressive symptoms (14–16). Clearly, in both types of diabetes many different factors play a role in medication intake behavior and perceived barriers. Therefore, clinicians need to openly and constructively discuss the various reasons underlying an individual’s suboptimal medication intake. Furthermore, the associations with severe hypoglycemia (in type 1 diabetes) and diabetes duration (in type 2 diabetes) warrant attention, as these factors might be indicative of worse medication intake and more perceived barriers.
Another important finding from this study is the strong association of medication intake and perceived barriers with two psychological factors (depressive symptoms and diabetes-specific distress), which was implicated both in adults with type 1 and type 2 diabetes. The relationship between diabetes and depressive symptoms is well-established and appears to be bi-directional. As compared to adults without diabetes, adults with diabetes have a two-to-threefold increased risk of depression [24]. Additionally, adults with depression have a 1.76 odds of suboptimal medication intake compared to adults without depression [25].People with T2DM and co-morbid depression show less optimal medication intake behavior, as well as more frequent hyperglycemia [26, 27]. The association between depressive symptoms and sub-optimal medication intake is possibly caused by various aspects of depression such as a lack of motivation, a lack of energy, difficulties in making decisions and a lower self-esteem. Comparable results have been found for diabetes-specific distress, since higher diabetes-specific distress was associated with less optimal medication intake, mediated through perceived control and self-efficacy [28]. Therefore, clinicians need to especially consider psychological factors as potential barriers for optimal medication intake. Additionally, clinicians might consider making use of the wide range of interventions developed specifically to target diabetes distress or depression in people with diabetes, as they have been shown to lower both depressive symptoms, diabetes distress and HbA1c [29, 30]. However, we need to be cautious with respect to glycemic effects, as improvement of the general medical condition including glycemic control is likely to require simultaneous attention to both conditions.”
## Strengths and Limitations
The strengths of this study include the use of a large dataset including adults with type 1 diabetes or type 2 diabetes, the extensive number of variables available in this dataset and the use of both univariable and multivariable analyses. Another strength is the use of a validated and reliable questionnaire to measure medication use [19], as well as depressive symptoms and diabetes-specific distress. However, analyses in this study were focused solely on this single self-report measurement of medication intake and perceived barriers, which may be subject to some social desirability bias. Additionally, the results from this study do not include any objective data on medication intake. The suitability of the ASK-12 as a measure of medication intake and perceived barriers for people with type 1 diabetes can be strengthened. Future rigorous psychometric testing of the instrument in this group is recommended. The validity of the results could therefore be improved by making use of an objective medication intake measurement such as medication event monitoring systems (MEMS) [31]. MEMS, however, are an expensive data collection method, which is not feasible to implement in an extensive project such as the MILES study.
Additionally, the Diabetes MILES – Netherlands sample of adults with type 1 diabetes and type 2 diabetes in the dataset is not fully representative of adults with type 1 or type 2 diabetes in the Netherlands. For example, adults with type 2 diabetes who do not use insulin and adults from ethnic minority groups were underrepresented [18]. Also people with lower HbA1c and emotional distress were overrepresented in the present sample [32, 33]. Moreover, the subpopulation evaluated in the present study showed higher medication intake scores and fewer perceived barriers as compared to the diabetes population participating in the ASK-12 publication study [19].
Diabetes MILES – the Netherlands aimed to measure life with diabetes across diabetes types. Comparisons between type 1 and type 2 diabetes are possible because the same instruments were used to measure constructs in both conditions. As a consequence, sometimes more condition-specific and detailed information is missing. For example, the ASK-12 questionnaire has not been specifically validated in people with type 1 diabetes, and detailed information on the insulin regimen is lacking.
Finally, this is a cross-sectional observational study so we cannot infer causality from these findings.
## Conclusions
In conclusion, adults with type 1 diabetes show slightly more optimal medication intake and fewer perceived barriers than adults with non-insulin-treated type 2 diabetes and more optimal medication intake behavior than adults with both insulin-treated and non-insulin-treated type 2 diabetes. However, these differences were not found on all subscales and absolute differences between groups are minimal. Correlates of less optimal medication intake and more perceived barriers in type 1 diabetes are fewer severe hypoglycemic events in the past year, higher depressive symptoms and higher diabetes-specific distress. Factors associated with less optimal medication intake and more perceived barriers in type 2 diabetes are a shorter duration of diabetes, higher depressive symptoms and higher diabetes-specific distress.
## Recommendations for Practice
These insights suggest practical ways in which clinicians can better support people with diabetes, tailoring their interventions to the specific reason(s) for suboptimal medication intake and unmet needs of the individual with diabetes. Especially the strong association with depressive symptoms and diabetes-specific distress in both diabetes types warrants attention, as improving these outcomes in some persons with diabetes might indirectly improve medication intake.
## Data Availability Statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics Statement
The study protocol was approved by the Psychological Research Ethics Committee of Tilburg University, The Netherlands (EC-2011 5). The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
All authors contributed to the conception and design of the study, analysis, and interpretation of the data. SH, MA, and JH drafted the first version of the manuscript. All authors (SH, MA, JH, MB, JS, FP, GN) critically revised the manuscript for important intellectual content. All authors contributed to the article and approved the submitted version.
## Funding
This study was supported by the Prof. dr. J. Terpstra Young Investigator Award 2010 from the Dutch Association for Diabetes Research (Nederlandse Vereniging voor Diabetes Onderzoek)/Lilly Diabetes to GN. The funding source had no role in the design, data collection, analysis or interpretation of the study, or in the decision to submit the manuscript for publication. JS is supported by core funding to the Australian Centre for Behavioural Research in Diabetes, derived from the collaboration between Diabetes Victoria and Deakin University.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Supplementary Material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcdhc.2021.645609/full#supplementary-material
## References
1. Huang Y. *IDF Diabetes Atlas* (2017)
2. Selvin E, Parrinello CM, Sacks DB, Coresh J. **Trends in Prevalence and Control of Diabetes in the United States, 1988–1994 and 1999–2010**. *Ann Internal Med* (2014) **160** 517. DOI: 10.7326/M13-2411
3. Currie CJ, Peyrot M, Morgan CL, Poole CD, Jenkins-Jones S, Rubin RR. **The impact of treatment non-compliance on mortality in people with type 1 diabetes**. *J Diabetes Its Complications* (2013) **27**. DOI: 10.1016/j.jdiacomp.2012.10.006
4. Lind M, Garcia-Rodriguez LA, Booth GL, Cea-Soriano L, Shah BR, Ekeroth G. **Mortality trends in patients with and without diabetes in Ontario, Canada and the UK from 1996 to 2009: a population-based study**. *Diabetologia* (2013) **56**. DOI: 10.1007/s00125-013-3063-1
5. Seuring T, Archangelidi O, Suhrcke M. **The Economic Costs of Type 2 Diabetes: A Global Systematic Review**. *PharmacoEconomics* (2015) **33**. DOI: 10.1007/s40273-015-0268-9
6. Dall TM, Yang W, Halder P, Pang B, Massoudi M, Wintfeld N. **The economic burden of elevated blood glucose levels in 2012: diagnosed and undiagnosed diabetes, gestational diabetes mellitus, and prediabetes**. *Diabetes Care* (2014) **37**. DOI: 10.2337/dc14-1036
7. **The Effect of Intensive Treatment of Diabetes on the Development and Progression of Long-Term Complications in Insulin-Dependent Diabetes Mellitus**. *New Engl J Med* (1993) **329**. DOI: 10.1056/NEJM199309303291401
8. **Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes**. *Lancet* (1998) **352**. DOI: 10.1016/S0140-6736(98)07019-6
9. Ali MK, McKeever Bullard K, Imperatore G, Barker L, Gregg EW. **Characteristics associated with poor glycemic control among adults with self-reported diagnosed diabetes–National Health and Nutrition Examination Survey, United States, 2007-2010**. *MMWR Suppl* (2012) **61**
10. McKnight JA, Wild SH, Lamb MJE, Cooper MN, Jones TW, Davis EA. **Glycaemic control of Type 1 diabetes in clinical practice early in the 21st century: an international comparison**. *Diabetic Med* (2015) **32**. DOI: 10.1111/dme.12676
11. Krass I, Schieback P, Dhippayom T. **Systematic Review or Meta-analysis Adherence to diabetes medication: a systematic review**. *Diabet Med* (2014) **32**. DOI: 10.1111/dme.12651
12. Kardas P, Lewek P, Matyjaszczyk M. **Determinants of patient adherence: A review of systematic reviews**. *Front Pharmacol* (2013) **4** 1-16. DOI: 10.3389/fphar.2013.00091
13. Brown MT, Bussell JK. **Medication adherence: WHO cares**. *Mayo Clin Proc* (2011) **86**. DOI: 10.4065/mcp.2010.0575
14. Capoccia K, Odegard PS, Letassy N. **Medication Adherence With Diabetes Medication: A Systematic Review of the Literature**. *Diabetes Educ* (2016) **42** 34-71. DOI: 10.1177/0145721715619038
15. Borgsteede SD, Westerman MJ, Kok IL, Meeuse JC, De Vries TPGM, Hugtenburg JG. **Factors related to high and low levels of drug adherence according to patients with type 2 diabetes**. *Int J Clin Pharmacy* (2011) **33**. DOI: 10.1007/s11096-011-9534-x
16. García-Pérez L-E, Álvarez M, Dilla T, Gil-Guillén V, Orozco-Beltrán D. **Adherence to Therapies in Patients with Type 2 Diabetes**. *Diabetes Ther* (2013) **4**. DOI: 10.1007/s13300-013-0034-y
17. Breuker C, Macioce V, Mura T, Audurier Y, Boegner C, Jalabert A. **Medication errors at hospital admission and discharge in Type 1 and 2 diabetes**. *Diabetic Med* (2017) **34**. DOI: 10.1111/dme.13531
18. Nefs G, Bot M, Browne JL, Speight J, Pouwer F. **Diabetes MILES–The Netherlands: rationale, design and sample characteristics of a national survey examining the psychosocial aspects of living with diabetes in Dutch adults**. *BMC Public Health* (2012) **12** 925. DOI: 10.1186/1471-2458-12-925
19. Matza LS, Park J, Coyne KS, Skinner EP, Malley KG, Wolever RQ. **Derivation and Validation of the ASK-12 Adherence Barrier Survey**. *Ann Pharmacother* (2009) **43**. DOI: 10.1345/aph.1M174
20. Kroenke K, Spitzer RL, Williams JBW. **The PHQ-9**. *J. Gen. Internal Med.* (2001) **16**. DOI: 10.1046/j.1525-1497.2001.016009606.x
21. Spitzer RL, Kroenke K, Williams JBW, Löwe B. **A Brief Measure for Assessing Generalized Anxiety Disorder**. *Arch Internal Med* (2006) **166** 1092. DOI: 10.1001/archinte.166.10.1092
22. Welch GW, Jacobson AM, Polonsky WH. **The problem areas in diabetes scale: An evaluation of its clinical utility**. *Diabetes Care* (1997) **20**. DOI: 10.2337/diacare.20.5.760
23. Babyak MA. **What You See May Not Be What You Get : A Brief, Nontechnical Introduction to Overfitting in Regression-Type Models**. *Psychosomatic Med* (2004) **66**. DOI: 10.1097/01.psy.0000127692.23278.a9
24. Bădescu SV, Tătaru C, Kobylinska L, Georgescu EL, Zahiu DM, Zăgrean AM. **The association between Diabetes mellitus and Depression**. *J Med Life* (2016) **9**
25. Grenard JL, Munjas BA, Adams JL, Suttorp M, Maglione M, McGlynn EA. **Depression and Medication Adherence in the Treatment of Chronic Diseases in the United States: A Meta-Analysis**. *J Gen Internal Med* (2011) **26**. DOI: 10.1007/s11606-011-1704-y
26. Gonzalez JS, Peyrot M, McCarl LA, Collins EM, Serpa L, Mimiaga MJ. **Depression and diabetes treatment nonadherence: A meta-analysis**. *Diabetes Care* (2008) **31**. DOI: 10.2337/dc08-1341
27. Lustman PJ, Anderson RJ, Freedland KE, De Groot M, Carney RM, Clouse RE. **Depression and poor glycemic control: A meta-analytic review of the literature**. *Diabetes Care* (2000) **23**. DOI: 10.2337/diacare.23.7.934
28. Gonzalez JS, Shreck E, Psaros C, Safren SA. **Distress and Diabetes Treatment Adherence: A Mediating Role for Perceived Control**. *Health Psychol* (2016) **34**. DOI: 10.1037/hea0000131
29. Schmidt CB, van Loon BJP, Vergouwen ACM, Snoek FJ, Honig A. **Systematic review and meta-analysis of psychological interventions in people with diabetes and elevated diabetes-distress**. *Diabetic Med* (2018) **35**. DOI: 10.1111/dme.13709
30. van der Feltz-Cornelis C, Allen SF, Holt RIG, Roberts R, Nouwen A, Sartorius N. **Treatment for comorbid depressive disorder or subthreshold depression in diabetes mellitus: Systematic review and meta-analysis**. *Brain Behav* (2021) **11** e01981. DOI: 10.1002/brb3.1981
31. El Alili M, Vrijens B, Demonceau J, Evers SM, Hiligsmannn M. **A scoping review of studies comparing the medication event monitoring system (MEMS ) with alternative methods for measuring medication adherence**. *Br J Clin Pharmacol* (2016) **82**. DOI: 10.1111/bcp.12942
32. Heintjes EM, Houben E, Beekman-Hendriks WL, Lighaam E, Cremers SM, Penning-Van Beest FJA. **Trends in mortality, cardiovascular complications, and risk factors in type 2 diabetes**. *Neth J Med* (2019) **77**
33. Skinner TC, Joensen L, Parkin T. **Twenty-five years of diabetes distress research**. *Diabetic Med* (2020) **37** 393-400. DOI: 10.1111/dme.14157
|
---
title: Western diet-induced shifts in the maternal microbiome are associated with
altered microRNA expression in baboon placenta and fetal liver
authors:
- Kameron Y. Sugino
- Ashok Mandala
- Rachel C. Janssen
- Sunam Gurung
- MaJoi Trammell
- Michael W. Day
- Richard S. Brush
- James F. Papin
- David W. Dyer
- Martin-Paul Agbaga
- Jacob E. Friedman
- Marisol Castillo-Castrejon
- Karen R. Jonscher
- Dean A. Myers
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2022
pmcid: PMC10012127
doi: 10.3389/fcdhc.2022.945768
license: CC BY 4.0
---
# Western diet-induced shifts in the maternal microbiome are associated with altered microRNA expression in baboon placenta and fetal liver
## Abstract
Maternal consumption of a high-fat, Western-style diet (WD) disrupts the maternal/infant microbiome and contributes to developmental programming of the immune system and nonalcoholic fatty liver disease (NAFLD) in the offspring. Epigenetic changes, including non-coding miRNAs in the fetus and/or placenta may also underlie this risk. We previously showed that obese nonhuman primates fed a WD during pregnancy results in the loss of beneficial maternal gut microbes and dysregulation of cellular metabolism and mitochondrial dysfunction in the fetal liver, leading to a perturbed postnatal immune response with accelerated NAFLD in juvenile offspring. Here, we investigated associations between WD-induced maternal metabolic and microbiome changes, in the absence of obesity, and miRNA and gene expression changes in the placenta and fetal liver. After ~8-11 months of WD feeding, dams were similar in body weight but exhibited mild, systemic inflammation (elevated CRP and neutrophil count) and dyslipidemia (increased triglycerides and cholesterol) compared with dams fed a control diet. The maternal gut microbiome was mainly comprised of Lactobacillales and Clostridiales, with significantly decreased alpha diversity ($$P \leq 0.0163$$) in WD-fed dams but no community-wide differences ($$P \leq 0.26$$). At 0.9 gestation, mRNA expression of IL6 and TNF in maternal WD (mWD) exposed placentas trended higher, while increased triglycerides, expression of pro-inflammatory CCR2, and histological evidence for fibrosis were found in mWD-exposed fetal livers. In the mWD-exposed fetus, hepatic expression levels of miR-204-5p and miR-145-3p were significantly downregulated, whereas in mWD-exposed placentas, miR-182-5p and miR-183-5p were significantly decreased. Notably, miR-1285-3p expression in the liver and miR-183-5p in the placenta were significantly associated with inflammation and lipid synthesis pathway genes, respectively. Blautia and Ruminococcus were significantly associated with miR-122-5p in liver, while Coriobacteriaceae and Prevotellaceae were strongly associated with miR-1285-3p in the placenta; both miRNAs are implicated in pathways mediating postnatal growth and obesity. Our findings demonstrate that mWD shifts the maternal microbiome, lipid metabolism, and inflammation prior to obesity and are associated with epigenetic changes in the placenta and fetal liver. These changes may underlie inflammation, oxidative stress, and fibrosis patterns that drive NAFLD and metabolic disease risk in the next generation.
## Introduction
Nonalcoholic fatty liver disease (NAFLD) is the most common liver disease worldwide. Characterized by simple steatosis (excess liver fat), NAFLD may progress to nonalcoholic steatohepatitis (NASH) with inflammation and fibrosis, leading to cirrhosis and increased risk for hepatocellular carcinoma [1]. The CDC estimates that >18 million reproductively competent people in the U.S. are overweight or obese, and maternal obesity is strongly linked to inflammatory and metabolic disorders in the offspring (2–7). Alarmingly, 1 in 5 preschoolers are obese [8] and $\frac{1}{3}$ of obese youth are diagnosed with NAFLD [9]. Despite evidence that maternal overnutrition adversely influences metabolic health in human [10] and animal (11–15) offspring, there is a fundamental lack of insight into molecular mechanisms by which maternal dietary exposures reprogram fetal immune development and NAFLD in utero, particularly in models that reflect the human condition.
The placenta acts as the primary interface between mother and fetus, allowing nutrient and oxygen transfer which supports fetal growth and development. Obesity-associated maternal-fetal inflammation contributes to various adverse pregnancy outcomes including placental dysfunction [16, 17], preeclampsia, neurodevelopment [18], intrauterine growth restriction [19, 20], and preterm labor [21, 22]. An investigation of the chronic pro-inflammatory milieu in placentas from obese pregnancies showed a two- to three-fold increase in resident macrophages (CD68+ and CD14+) and expression of pro-inflammatory cytokines compared with placentas from lean pregnancies [23]. However, the underlying causes of placental inflammation remain unclear. One possible mechanism is through inflammatory metabolites, such as lipopolysaccharides, which are produced by the microbiome, reach higher systemic levels in obesity [24], and can directly interact with placental Toll-like receptor 4 [25]. The role of a Western-style diet (WD) versus maternal obesity in disruption of placental function and inflammatory state remains to be elucidated, as most animal models of maternal obesity have relied upon a WD to attain and maintain obesity.
The maternal gut microbiome changes during pregnancy and is influenced by maternal diet, maternal obesity, excessive gestational weight gain, and gestational diabetes mellitus (GDM) (26–28). Animal and human studies (29–32) support a role for the maternal gut microbiome and bacterial metabolites in the pathophysiological changes accompanying maternal obesity [33], offspring NAFLD [34, 35], and infant inflammatory disorders [36]. WD-induced maternal obesity is associated with impaired microbiome and placental structural and functional changes, concomitant with oxidative stress and inflammation in the intrauterine environment (37–39). Trophoblasts recognize and respond to bacterial products [40] and integrate microbial-derived signals via pathways mediating the response to Toll-like receptors or epigenetic modifications [41]. Although mechanisms by which by-products of gut dysbiosis affect placental inflammation/function and fetal inflammation are not well understood, reduction in beneficial bacterial metabolites might play a role.
Emerging studies address the association of maternal diet and obesity with alterations to epigenetic signatures and microbiome function in offspring [13, 42]. The epigenome consists of DNA and histone modifications that produce heritable changes in transcription and cellular function [43]. One form of epigenetic modification are microRNAs (miRNAs). MiRNAs are small, noncoding RNAs that bind to the 3’ untranslated region of protein-coding mRNAs, degrading or repressing translation of the targeted mRNAs. In humans, miR-122, miR-34a, miR-21, and miR-29a were shown to regulate lipid metabolism, oxidative stress, and inflammation in the liver, with a crucial role in the pathophysiology of NAFLD (44–47). Studies investigating the impact of early changes in miRNAs in the fetal liver are sparse; however, using large-scale sequencing and pathway analysis in baboon fetal liver, Puppala et al. identified 11 miRNAs targeting 13 genes in metabolic pathways (TCA cycle, oxidative phosphorylation, and glycolysis), the proteasome, and WNT/β-catenin signaling [48]. In the placenta, miRNA expression levels are disrupted by hypoxia [49], maternal exposure to toxic agents such as cigarette smoke [50] and bisphenol A [51], and GDM [52]. In humans, elevated pre-pregnancy BMI has been associated with lower expression levels of placental miRNAs; these differences were modified by offspring sex and maternal gestational weight gain [53]. However, studies investigating associations between maternal diet, short-chain fatty acids (SCFAs), and epigenetic modifications in the placenta and fetal liver are lacking. Here, we leveraged our well-established olive baboon (Papio anubis) nonhuman primate (NHP) model of maternal WD consumption to investigate changes in the maternal microbiome and their association with placental and fetal liver miRNA expression profiles in non-obese pregnancy.
## Animal model
All experiments utilizing baboons were performed in compliance with guidelines established by the Animal Welfare Act for housing and care of laboratory animals as well as the U.S. National Institutes of Health Office of Laboratory Animal Welfare Public Health Service Policy on Humane Care and Use of Laboratory Animals. All animals received environmental enrichment. All experiments were conducted in accordance with and with approval from the University of Oklahoma Health Sciences Center Institutional Animal Care and Use Committee (IACUC; protocol 302043 #22-025-AH).
Nulliparous olive baboon females (Papio anubis; $$n = 11$$; 4-7 years old; puberty is ~4-5 years of age) with similar lean body scores were randomly separated into two primary cohorts with similar mean age and body weight, taking into consideration social stratification. Dams were housed in diet groups ($$n = 5$$/6 per corral) for the duration of the study. Control diet (CD) dams were fed standard monkey chow diet (5045, Purina LabDiets, St. Louis, MO; $30.3\%$ calories from protein, $13.2\%$ from fat, $56.5\%$ from carbohydrates) and the WD dams were fed a high-fat, high-simple sugar diet (TAD Primate diet, 5L0P, Purina; $18.3\%$ calories from protein, $36.3\%$ of calories from fat, $45.4\%$ from carbohydrates), matched to CD for micronutrients and vitamins, and supplemented with continuous access to a high-fructose beverage (100 g/L KoolAid™). Both groups were provided the same daily enrichment foods (fruits and peanuts). The TAD diet/high-fructose drink is widely used to study the role of an excess energy intake, high-saturated fat, high-fructose diet on physiological systems in NHPs [54, 55] and is consistent with human WDs. After an initial 3 months of WD to allow for collection of baseline samples and acclimation to WD, dams were bred to males that had been previously fed CD. Blood, fecal samples, and anthropometric measurements (body weight and sum of skin folds) were obtained under ketamine (10-20 mg/kg) and acepromazine (0.05-0.5 mg/kg) sedation administered via intramuscular injection. Following chemical sedation, cephalic or saphenous vein catheters were placed for blood draw. At 0.6 gestation (G; term is ~183 days gestation) baboons were fasted overnight. Intravenous glucose tolerance test (IVGTT) was performed under ketamine and acepromazine sedation and two appropriately sized intravenous catheters were placed into each saphenous or cephalic vein, or a combination thereof, one for infusion of dextrose and one for venous blood collection. A baseline blood draw was taken from one catheter and an intravenous bolus of $50\%$ dextrose (0.5 g/kg body weight) was administered over 30 seconds through the second catheter. Blood glucose was measured in venous blood using a glucometer at time 0, 2, 4, 8, 12, 16, 20, and 40 min post dextrose infusion.
## Maternal blood analyses
Complete blood counts (CBCs) were obtained for the dams from EDTA-anticoagulated whole blood samples. CBCs included analyses for red blood cells, white blood cell count (neutrophils, lymphocytes, monocytes, eosinophils, and basophils), platelets, and hemoglobin. Maternal serum samples collected at 0.6 G were analyzed for C-reactive protein (CRP) using an hsCRP ELISA kit (MP Biomedicals, Solon, OH) according to the manufacturer’s protocol with 1:100 serum dilution and were analyzed for IL-6 using an old world monkey IL-6 ELISA kit (U-CyTech Biosciences, the Netherlands) according to the manufacturer’s protocol. Maternal serum samples were analyzed for triglycerides (TGs) using a triglyceride colorimetric assay kit (Cayman Chemical, Ann Arbor, MI) according to the manufacturer’s protocol. High-density lipoprotein (HDL) and low-density lipoprotein/very low-density lipoprotein (LDL/VLDL) levels were quantified in serum samples taken at 0.6 G from fasted dams using EnzyChrom HDL and LDL/VLDL Assay Kits (BioAssay Systems, Hayward, CA). Samples were assayed in duplicate according to the manufacturer’s instructions.
## Cesarean section
At 0.9 G, dams were anesthetized using isoflurane and fetuses were delivered by cesarean section. Fetal and placental weights were obtained, and placental and liver tissue samples were processed for histology or flash-frozen in liquid nitrogen and stored at -80° C for subsequent analyses.
## Liver histology
Fetal liver tissue samples from the left lobe were fixed in $10\%$ formalin for 24 h followed by storage in $70\%$ EtOH. Histology was performed by the OUHSC Stephenson Cancer Tissue Pathology Core. In brief, samples were paraffin-embedded and sectioned for H&E and picrosirius red staining. Fresh-frozen liver from the left lobe was fixed in OCT compound, sectioned, and fixed with formalin for 5 min and washed with PBS. LipidSpot lipid droplet stain (Biotium, Fremont, CA) was added to the sections and incubated for 20 min. Sections were washed with PBS, counterstained with DAPI, and mounted using VectaMount AQ aqueous mounting medium (Vector Labs, Burlingame, CA). Slides were visualized using a Cytation 5 microscope and Gen5 imaging software (Agilent, Santa Clara, CA).
## Placenta immunofluorescence
Immediately upon delivery of the placenta during cesarean section, placental tissue was dissected from each cotyledon: one half of each sample was flash-frozen and stored at -80°C and the other half was fixed in $4\%$ paraformaldehyde for 48 h, transferred to $70\%$ EtOH, and paraffin embedded. Thin sections (5-micron thickness) were obtained every 150 microns from the paraffin blocks and placed onto slides. For immunofluorescence (IF) labelling, sections were selected to allow for a total of four sections per cotyledon at a minimum of 150 microns between sections. Slides were baked for 1 h at 56°C, deparaffinized and antigen retrieval was performed in a Retriever 2100 instrument with R-Universal epitope recovery buffer (Electron Microscopy Sciences, Hatfield, PA). After retrieval, slides were blocked in $5\%$ normal donkey serum for 1 h, then primary antibody (MAC387 [anti-S100A9 + Calprotectin], Abcam, Cambridge, UK) was added and slides were incubated for 16 h at 4°C with humidification. Slides were subsequently allowed to equilibrate to RT for 1 h. Secondary antibody (donkey anti-mouse IgG F(ab’)2 Alexa Fluor 594, Jackson ImmunoResearch, West Grove, PA) was added and slides were incubated for 1 h, covered, at RT. Slides were counterstained for 5 min with DAPI and coverslipped using Shur/Mount. Slides were visualized using a fluorescence microscope (Olympus BX43) and images were captured using CellSens imaging software (Olympus). Four 150x150 micron images were selected randomly per tissue section (700x500 microns) with no overlap between selected sections and macrophages were counted by a researcher blinded to the treatment group. Total macrophages were summed for all images per placenta and scored.
## Fetal liver and placental tissue analyses
RNA was extracted from flash-frozen fetal liver (left lobe) samples using a Direct-zol RNA miniprep kit (Zymo Research, Irvine, CA)) per manufacturer's instructions. cDNA was synthesized from 1 ug RNA using iScript Supermix (Bio-Rad, Hercules, CA) per manufacturer's instructions. Gene expression was measured using real time qPCR with PowerUp SYBR Green Master mix on a QuantStudio 6 instrument (Thermo Fisher Scientific, Waltham, MA). Results were normalized to RPS9 (ribosomal protein S9) using the comparative Ct method. For placenta, qPCR was performed similar to fetal liver except using a CFX96 RT-PCR Detection System (Bio-Rad, Hercules, CA). Placental results were normalized to ACTB using the comparative Ct method. Primers for qPCR are shown in Supplementary Table S1. Fetal liver TGs were extracted as described previously [32] and quantified using Infinity Triglycerides Reagent (Thermo Fisher) with normalization to starting tissue weight.
Total RNA, including miRNA, was isolated from flash-frozen fetal liver (left lobe) and placenta tissues (two samples/placenta from separate cotyledons) using an miRNeasy mini kit (Qiagen, Germantown, MD) following manufacturer’s instructions. TaqMan MicroRNA Assays (Thermo Fisher) were used for reverse transcription and real-time qPCR (miR-122-5p, assay ID 002245; miR-204-5p, assay ID 000508; miR-34a-5p, assay ID 000426; miR-21-5p, assay ID 000397; miR-183-5p, assay ID 000484; miR-29a-3p, assay ID 002112; miR-185-5p, assay ID 002271; miR-145-3p, assay ID 002149; miR-1285-3p, assay ID 002822; miR-199a-5p, assay ID 000498; miR-182-5p, assay ID 000597). Reverse transcription was carried out using TaqMan MicroRNA Reverse Transcription kit (Thermo Fisher), an RT primer from a specific TaqMan MicroRNA Assay, and 10 ng total RNA following manufacturer’s instructions. The miRNA-specific cDNA templates were placed on ice and used immediately for qPCR or stored at -20°C for 1-2 days prior to use. Real-time PCR reactions were performed in duplicate using TaqMan Universal Master Mix II, no UNG (Thermo Fisher) and the corresponding TaqMan MicroRNA Assay following manufacturer’s instructions. Assays were performed on a QuantStudio 6 Real-time PCR System. Ct values were calculated and the relative miRNA expression levels were quantitated with the comparative Ct method and using miR-92a-3p (assay ID 000431) for normalization.
## SCFA analysis
Frozen feces (100-200 mg) were added to a vial continuing 200 μg/L of deuterated butyric acid (internal standard), 0.2 g/ml NaH2PO4, and 0.8 g/ml ammonium sulfate (adjusted to pH 2.5 with phosphoric acid). External standards were prepared with 0.2 g/ml NaH2PO4, 0.8 g/ml ammonium sulfate, 200 μg/L deuterated butyric acid (internal standard), 200 μg/L 2:0 (acetic acid), 100 μg/L 3:0, and 50 μg/L of 4:0, 5:0, and 6:0 (adjusted to pH 2.5). All vials were quickly capped and vigorously agitated. A 7890A gas chromatograph equipped with 7697A headspace sampler, 5975C mass spectrometer detector, and DB-FATWAX UI 30 x 0.25 x 250 column was used for analysis (Agilent). The headspace sampler oven, loop, and transfer line temperatures were held at 50°C, 100°C, and 110°C, respectively. The vial equilibration time was 30 min and the injection duration was 1 min. The vial fill pressure was 15 psi and the loop final pressure was 1.5 psi. Loop equilibration time was 0.05 min. The GC inlet and MS interface were held at 250°C. The oven temperature was held at 120°C for 2 min, ramped at 5°C/min to 140°C, ramped at 20°C/min to 220°C, and held at 220°C for 1 min. Helium carrier gas flowed constantly at 1.2 ml/min and the split ratio was 10:1. SCFAs were detected in SIM ion mode using m/z values of 43, 45, 60, 63, 73, 74, 77, and 87. Quantities of 2:0, 3:0, 4:0, 5:0, and 6:0 were determined by comparison to external standards.
## Microbial DNA extraction and sequencing
DNA extraction from feces from dams employed the DNeasy PowerSoil Pro Kit (Qiagen) per manufacturer’s instructions with the following modification. Aliquots of 300 mg of fecal material were weighed, resuspended in 800 μL Solution CD1, and incubated at 60°C for 10 min. Library construction and DNA sequencing were performed by the OUHSC Laboratory for Molecular Biology and Cytometry Research. Library construction employed the Nextera XT Library Preparation Kit (Illumina, San Diego, CA). The V4 region of the 16S gene was targeted using Earth Microbiome Project primers 515F/806R. Samples were barcoded for multiplexing, and sequenced on an Illumina MiSeq using paired-end sequencing with a 600 cycle MiSeq Reagent Kit v3 (Illumina).
## Data analysis
Maternal data, fetal liver TGs, and qPCR data were analyzed for comparisons between CD and WD using an unpaired, 2-tailed Student’s t test; significance was determined as $P \leq 0.05.$ Initial processing of raw microbiome data employed QIIME2 2019.10 software [56]. Sequenced 16S rRNA raw fastq files were imported as demultiplexed paired end reads with a Phred score of 33. Sequences were trimmed, quality filtered and denoised into amplicon sequence variants (ASVs) using DADA2 [57]. ASVs were then aligned de novo using MAFFT [58] and structured into a rooted phylogenetic tree using FastTree2 [59]. Alpha diversity (e.g., Faith’s Phylogenic Diversity, Shannon) and beta diversity (e.g., Weighted UniFrac distance, Bray-Curtis) were compared between diet groups using Kruskal-Wallis and PERMANOVA, respectively. Taxonomy was assigned to each ASV using a sklearn-based Naïve Bayes taxonomy classifier [60] pre-trained on the Greengenes 13_8 $99\%$ OTUs reference database sequences [61]. Linear discriminant analysis effect size, LEfSe [62], assessed the raw taxonomic abundance table for significant taxa which were differentially abundant in the context of the experimental groups. The Firmicutes to Bacteroidetes ratio (F/B) was correlated to maternal measures using a Spearman correlation and the Wilcoxon rank-sum test was used to test for differences in the F/B ratio between diet groups.
Variable selection methods: Due to the small sample size ($$n = 11$$) and relatively large number of comparison variables for dam (11 measures) and fetus (63 measures), we performed Least Absolute Shrinkage and Selection Operator (LASSO) regularization [63] to utilize the technique’s variable selection properties before testing for significance between maternal and fetal measurements. Briefly, LASSO regression applies a shrinkage term (lambda) to coefficients in the model in order to improve (i.e., reduce) the model’s mean squared error. This type of penalty can reduce the value of some coefficients to zero, eliminating them from the model and providing a method of variable selection. To implement this procedure, we used the R package glmnet [64]. Maternal and fetal measures were standardized around the mean and standard deviation before building each model. Each fetal measure was used as the response variable and maternal measures as the explanatory variables; we included an interaction term between the maternal measures and fetal sex for each maternal measure within the model. Since complete data are needed for LASSO regularization, each model was estimated using only samples with complete data (minimum $$n = 5$$, median $$n = 7$$) and the best lambda value was selected based on the model with the lowest mean squared error. Maternal variables selected by the LASSO procedure were then run in univariate ANOVA models against the fetal variable using the full dataset ($$n = 11$$), P values were compiled and FDR correction was applied using the Benjamini-Hochberg procedure. A similar method was applied to compare the fetal measures to each other; however, since only four fetal samples had complete data, the fetal measures were broken into three separate datasets: miRNA (33 measures, both liver and placenta), fetal mRNA levels (27 measures), and other measures (fetal weight, heart weight, and liver TGs); fetal sex was included as an interaction term for all models. Comparisons between the fetal miRNA were performed using the liver miRNA measures as the response variable and placental miRNA as the predictor. After breaking up the fetal dataset, we performed the following comparisons: fetal miRNA vs. mRNA, miRNA vs. other measures, and mRNA vs. other measures; sample sizes used for each LASSO procedure reached a minimum of 5, median of 7 datapoints for these comparisons.
LASSO regularization was also used for variable selection to compare the maternal microbiota ($$n = 10$$) to maternal and fetal measures. To compare the maternal measures to the maternal microbiome, we utilized the R package mpath [65] to perform the LASSO procedure using a negative binomial regression model. Briefly, the microbiota were classified to the taxonomic family and genus levels and filtered to include taxa which were present in at least $50\%$ of the samples and reached at least $1\%$ relative abundance. The microbiota count data were used for input into the glmregNB function for LASSO regularization with default values, and the best model was selected using the Bayesian information criterion. Again, only complete sets of data were used in the LASSO procedure ($$n = 6$$). Variables selected for each taxon were run in a univariate negative binomial regression with the R package MASS [66] using the full dataset ($$n = 10$$). P values were compiled and FDR correction was applied using the Benjamini-Hochberg procedure. LASSO models were constructed using only complete datasets, while statistics were run on the full dataset. To compare fetal measures as an outcome of the maternal microbiota abundances, we used arcsin square root transformation of the microbiota relative abundance data [67] as predictors for the infant measures. LASSO regularization was used to build these models using glmnet and univariate ANOVA models were used to test for significance as described in the previous paragraph. The samples sizes used for LASSO regularization between maternal microbiota and fetal measures were as follows: fetal miRNA, $$n = 8$$; fetal mRNA, $$n = 6$$; and other fetal measures (weight, heart weight, and liver TGs), $$n = 8$.$ Fetal sex was included as an interaction term within these models.
## WD induces inflammation in dams, placenta, and fetuses and impairs lipid metabolism
The time (days) from initiation of WD to IVGTT (carried out at 0.6 G) ranged from 276 to 480 days (332 ± 39 d); the time from initiation of WD to cesarean section (carried out at 0.9 G) ranged from 324 to 535 days (386.5 ± 39 d). Despite this duration of exposure to WD/high fructose, at 0.6 G, no significant difference in maternal body weight (Figure 1A) or in adiposity index (sum of skin fold thickness, Figure 1B) was observed. These findings are consistent with those reported by the Nathanielsz lab [68, 69] using a similar diet and baboon model, demonstrating that attainment of maternal obesity requires a minimum of 9 mo. to ~3 years of WD feeding. Although WD-fed dams did not become obese, serum CRP (Figure 1C) and neutrophils (Figure 1D) increased, without a change in serum IL-6 (Figure 1E), indicative of mild, systemic inflammation in response to WD prior to attaining obesity. WD-fed dams also exhibited elevated serum TGs (Figure 1F). Serum HDL cholesterol and LDL/VLDL cholesterol were significantly higher in WD-fed dams compared with CD-fed dams; however, no significant change in total cholesterol was observed (Figure 1G). Blood glucose tolerance tests did not show a significant effect of diet (Figure 1H). Together, these findings demonstrate that a relatively short exposure to WD during pregnancy induces significant systemic inflammation and impaired lipid metabolism in baboon dams prior to a significant change in adiposity or insulin sensitivity.
**Figure 1:** *WD-fed dams exhibit alterations in inflammation and lipid metabolism at 0.6 gestation. Maternal body weight (A), sum of (Σ) skin folds as a measure of adiposity (B). Maternal serum levels of C-reactive protein (CRP, C), neutrophil count from complete blood count (D), serum IL-6 levels (E), and triglycerides (F). Cholesterol analysis of red blood cells (G) and IVGTT analysis (H). n = 5-6 CD and n = 5 WD. Unpaired 2-tailed Student’s t test was used to test significance. *P < 0.05, **P < 0.01.*
Given the markers for inflammation observed in dams on WD, we performed immunofluorescence on fixed placental sections obtained at cesarean section using an antibody to MAC387 to label infiltrating monocytes/macrophages. We observed very few MAC387+ macrophages in maternal CD (mCD)-exposed placentas (Figure 2A). Macrophage infiltration was more variable in the placenta of WD-fed dams, with notable macrophages evident in 3 of 5 placentas whilst macrophage presence in two placentas was comparable with those of mCD-exposed placentas (Figure 2A, right panel). Expression of cytokines/chemokines (IL1B, IL6, TNF, IL8) in placental tissue (Figure 2B) were elevated in maternal WD (mWD)-exposed placentas but differences did not reach significance when compared with mCD-exposed placentas. However, both IL6 ($$P \leq 0.08$$) and TNF ($$P \leq 0.068$$) exhibited trends for significance. No notable gross placental pathologies (calcifications, infarcts) were observed in mWD-exposed compared with mCD-exposed placentas.
**Figure 2:** *WD exposure increases monocyte infiltration of the placenta and induces fetal hepatic steatosis and fibrosis. Representative images for immunofluorescence in placenta tissue and quantitation (A). Red arrows point to MAC387-labeled macrophages (green). Blue staining - DAPI. (B) mRNA expression of cytokines in placenta using qPCR. ACTB was used for reference. Representative images of histological analysis of fetal liver tissue (C) with H&E, picrosirius red (PSR), and LipidSpot, taken at 100 um. Fetal liver triglycerides (TG) (D) and mRNA expression analysis using qPCR (E) with RPS9 used for normalization. n = 4-6 CD and n = 5 WD. Unpaired 2-tailed Student’s t test was used to test significance. *P < 0.05, ****P < 0.0001.*
We next tested whether maternal exposure to WD influenced fetal liver health. Liver lipids were histologically evident in mWD-exposed fetal livers (Figure 2C, upper panel) and verified by increased LipidSpot staining (Figure 2C, lower panel) and by TG analysis (Figure 2D). Picrosirus red staining, an indicator for fibrosis, was strikingly more evident in mWD-exposed fetal liver (Figure 2C, middle panel). Expression of mRNA for genes involved in inflammation (IL6), fibrosis (COL3A1), and monocyte infiltration (CCR2) was elevated in fetal livers from WD-fed dams (Figure 2D), although only CCR2 expression was significant between groups. Together, these data suggest that exposure to mWD in utero promotes TG storage, a trend for increased inflammation, and early fibrogenesis, which are hallmarks of NAFLD/NASH changes in the fetal liver. We also performed qPCR analysis for the differential expression of inflammatory (TNF, CCL2, TLR4, NOS2; M1 macrophage) as well as anti-inflammatory (IL10, TGFB1, PDGFA, IL12B; M2 macrophage) genes. However, our analysis did not reveal significant differences in the expression of these genes between mCD- and mWD-exposed fetuses (Supplementary Table S2).
## Maternal WD induces subtle microbiota changes in dams
At 0.6 G, maternal fecal SCFA levels did not differ between diets for propionate and butyrate, but there was an increase in acetate load in the feces of WD-fed dams (Figure 3A). Using 16S sequencing, we found that the maternal gut microbiome was mainly comprised of Lactobacillales and Clostridiales, with modest changes in microbial composition between groups (Figure 3B). Exposure to WD resulted in significantly lower alpha diversity (Figure 4A) but no community-wide differences between CD and WD groups as measured by beta diversity (PERMANOVA P value = 0.26; Figure 4B). We compared differences in maternal F/B ratio by diet group and maternal measures and found no significant associations (Supplementary Table S3, Supplementary Figure S1). Comparing microbiota abundances between diets, feces from WD-fed dams were enriched in Acidaminococcus and unclassified Betaproteobacteria, while those from CD-fed dams were characterized by higher Anaeroplasmatales abundance (Figure 4C). Using PICRUSt to predict functional pathways, we found 11 pathways potentially differentiating the effect of diet on the maternal gut microbes. WD-exposed microbiota were predicted to have enrichment of biosynthesis pathways such as gluconeogenesis and aspartate/asparagine synthesis, whereas CD feeding was associated with enrichment of pathways for rhamnose biosynthesis, lactose/galactose degradation, and bacterial-specific biosynthesis of peptidoglycans (Staphylococci) and O-antigen (Escherichia coli; Figure 4C).
**Figure 3:** *Short-duration exposure to WD induces few alterations in maternal SCFAs and microbiota. (A) Abundance of fecal SCFAs. n = 5/group. Unpaired 2-tailed Student’s t test was used to test significance. **P < 0.01. (B) Microbial abundances for each gut sample, clustered at the order level. Orders comprising less than 0.5% total abundance are displayed as “Other”.* **Figure 4:** *(A) Alpha diversity measured using Faith’s phylogenetic diversity. Significance of species richness was tested using the Kruskal-Wallis test (P = 0.02). (B) PCoA ordination displaying weighted Unifrac beta diversity. Percent variation explained is shown on each axis (PC1: 35% & PC2: 16%). PERMANOVA significance for the weighted Unifrac distances (P = 0.26). (C) Lefse histograms plotted for significant enrichment in taxa abundances (upper plot) and biochemical pathways (lower plot). n = 5/group.*
## Maternal WD alters placental and fetal miRNA profiles
Mounting evidence indicates that miRNAs mediate gut microbiome-host molecular communications [70]. We sought to explore differences in a targeted set of miRNAs in placentas and fetuses associated with maternal diet. We selected miRNAs that were previously found to be differentially expressed in fetal liver in a study in baboons in which the dams were obese, as well as miRNAs shown to be regulated by the gut microbiome (miR-122-5p, miR-204-5p, and miR-34a-5p) (71–74) or associated with NASH in rodents and humans (miR-21-5p and miR-29a-3p) (75–78). MiRNA expression analysis in fetal liver tissue (Figure 5A) revealed a significant downregulation of miR-204-5p and miR-145-3p expression upon mWD exposure. In the placenta, we observed a significant downregulation of miR-183-5p and miR-182-5p in the WD-fed group (Figure 5B). In placental tissue, miR-199a-5p showed a trend for downregulation ($$P \leq 0.057$$). The expression levels of miR-183-5p, miR-182-5p and miR-199a-5p were directionally similar in both fetal liver and in the placenta, as was the increase in expression of miR-1285-3p.
**Figure 5:** *MicroRNA expression analysis. Expression of miRNAs in mCD- and mWD-exposed fetal liver (A) and placental tissue (B). n = 6 CD and n = 5 WD. Unpaired 2-tailed Student’s t test was used to test significance. *P < 0.05, **P < 0.01.*
## Associations between microbiota, miRNA, metabolic features, and fetal liver gene expression
We first analyzed associations between maternal parameters of systemic and lipid metabolism to maternal microbiota and placental miRNAs (Table 1). After FDR correction, only unclassified Rickettsiales and unclassified Verrucomicrobia were significantly and positively associated with maternal HDL and Desulfovibrionaceae was positively associated with placental weight. Next, we analyzed associations between maternal parameters of systemic and lipid metabolism and fetal liver TG, miRNA, and mRNA levels (Table 2). Fetal liver TGs were positively associated with maternal TGs, HDL, and LDL/VLDL. After FDR correction, only the comparison between fetal liver TGs and maternal LDL/VLDL remained significant. We next compared fetal liver TGs to fetal liver/placental miRNAs and fetal liver mRNA levels (Table 3); however, after FDR correction, none of these comparisons were significant. In our analysis of fetal liver expression levels of miRNA and mRNA, we found several placental and fetal miRNAs associated with markers of lipid metabolism, oxidative stress, inflammation, and fibrosis (Table 4). The most common miRNA associated with liver mRNA levels was miR-1285-3p. Liver miR-1285-3p was negatively associated with NFE2L2, ICAM1, and VCAM1, but positively associated with IL6. Placental miR-1285-3p was positively associated with NFE2L2 and negatively associated with FAP. After FDR correction, only the association between miR-1285-3p and IL6 mRNA was significant, though several other associations were trending towards significance (Table 4). We next compared fetal liver miRNA levels and relative abundance of maternal microbiota at the taxonomic family level and found miR-122-5p was positively associated with Succinivibrionaceae (Table 5). Placental miR-1285-3p was positively associated with Coriobacteriaceae and Prevotellaceae. Both taxa displayed sex differences in association with placental miR-1285-3p, where males remained positively associated (Coriobacteriaceae, r2 = 0.57; Prevotellaceae, r2 = 0.70) but females showed no association (r2 = 0 for both taxa). After FDR correction, only placental miR-1285-3p remained significantly associated with Coriobacteriaceae, Prevotellaceae, and their interactions with sex. Fetal liver miR-122-5p and Succinivibrionaceae showed a trend toward association after FDR correction ($P \leq 0.1$). At the genus level, fetal liver miR-122-5p remained negatively associated with Blautia levels and positively associated with Ruminococcus after FDR correction. No other genera were selected for inclusion in our LASSO models for any of the other miRNA measured.
## Discussion
Epigenetic patterning in the placenta and fetus resulting from a Western-style maternal diet may be influenced by maternal gut microbes to promote the development of offspring NAFLD; however, these relationships are poorly described in human-relevant models. Our study using the olive baboon, in which dams were fed either a CD or WD prior to and during pregnancy, is the first to investigate early associations between the microbiota, placental and fetal miRNAs, and maternal-fetal metabolic dysregulation. Despite ~1 year of WD feeding, dams did not display significant weight gain or adipose deposition, consistent with similar studies in baboons [69] and Japanese macaques [79] where obesity was attained following 2 to 3 years of WD feeding. Strikingly, in the absence of obesity, we found that mWD adversely affected lipid metabolism and inflammation along the maternal-placental-fetal axis, concomitant with decreased miR-182-5p and miR-183-5p in mWD-exposed placentas compared with mCD-exposed placentas. Previous studies in NHP have not focused on miRNAs in placenta and their association with fetal liver function. A study in rats reported a role for miR-183-5p in mediating NF-κB signaling and reducing pro-inflammatory cytokine secretion [80], suggesting a potential role for the miR-183 family in regulating placental inflammation. We found the reduced placental expression levels of the miR-183 family (including mR-182-5p and miR-183-5p) to be positively associated with expression of a number of NAFLD/NASH-relevant genes in the fetal liver; this family was shown to attenuate pathophysiology in a mouse model of NASH [81] and may therefore be a novel target for future mechanistic studies in our rodent models. We also found significant associations between the maternal microbiota and placental miR-1285-3p, an miRNA associated with promotion of postnatal growth in infants born to mothers with obesity [82]. Livers from fetuses of WD-fed dams showed increased steatosis and elevated expression levels of genes involved in macrophage infiltration, inflammation, and fibrosis. These fetal NAFLD indices were increased concomitant with downregulation of hepatic expression of miR-204-5p and miR-145-3p. We and others have shown that mWD promotes oxidative stress in utero, reprogramming macrophages towards a pro-inflammatory “M1”-like phenotype. Previously, miR-204-5p, an miRNA known to be regulated by gut microbes [74], was reported to be involved in “M2” macrophage polarization by suppressing IL-6R signaling in tissue and cell lines [83, 84] and miR-145-3p altered IL-16 signaling in monocytes [85]. Although our data for miR-145-3p conflict with those of Puppala et al. [ 48], dams in our study were in a pre-obese state. Further studies investigating the mechanistic role of miR-204-5p and miR-145-3p in promoting hepatic inflammation are warranted.
Although mechanisms are not fully elucidated, the gut microbiome influences the development of chronic inflammatory diseases such as obesity through a variety of host pathways, including by mediating the expression of host miRNAs [86]. The effects of maternal obesity on the offspring microbiome were previously described in Japanese macaques [31, 87], but potential effects of the maternal microbiome on altering epigenetic marks in the placenta or fetus are undetermined. In our model of short-duration WD-feeding, the maternal microbiota were dominated by Firmicutes, primarily Lactobacilliales, Clostridiales, and Bacteroidales, strongly implicated in the maintenance of overall gut function and health [88, 89]. In the absence of maternal obesity, we found several associations between maternal gut microbes and maternal serum TGs, HDL, LDL/VLDL, and CRP, although the F/B ratio was not significantly associated with any maternal measure. The F/B ratio has been considered a marker for high-fat diet-related obesity and other metabolic disorders in human and animal models (90–93). However, a recent review challenges this idea and suggests that other factors, including differences in lipopolysaccharide and SCFA production, study methodologies and subject characteristics, as well as the complex etiology behind obesity and metabolic disorders, influence this relationship [94]. The F/B ratio relationship is also controversial within NHP studies. One study reported a higher F/B ratio in cynomolgus macaques on a high-fat diet compared with a Mediterranean diet [95], while another investigation reported a lower F/B ratio in NHP on a high-fat diet compared with chow diet [96]. In our cohort of non-obese, pregnant baboons, we found no associations between the F/B ratio at 0.6 G and any of the maternal metabolic parameters measured, including diet group.
Similar to our findings associating Desulfovibrionaceae abundance with placental weight in baboons, the genus *Desulfovibrio is* enriched in women with GDM [97, 98]; notably, GDM is associated with increased placental weight [99]. We found placental miR-1285-3p was associated with maternal gut microbiota abundances of Coriobacteriaceae and Prevotellaceae, both of which displayed a positive relationship with postnatal growth-promoting [82] miR-1285-3p expression in placentas from male fetuses, but no association in females. Both of these bacterial families are metabolically active members of the gut microbiome. MiR-122 accounts for around $70\%$ of all miRNAs in the adult liver and plays a role in regulation of innate immunity [100], proliferation and differentiation of hepatocytes [101], lipid accumulation, and cholesterol metabolism [102]. We found a differential association between fetal liver miR-122-5p and two maternal gut genera, Blautia and Ruminococcus, where Blautia negatively associated with miR-122-5p and Ruminococcus positively associated with miR-122-5p. To our knowledge, we are the first to report associations between maternal microbiota and fetal liver miRNA expression. Other studies have only investigated the relationship between the gut microbiome and miRNA in tumor tissues, which report a negative association between Blautia and tumor-related miRNAs such as miR-20a, miR-21, miR-96, miR-182, miR-183, and miR-7974 [103] and a positive association with miR-484 [104], which plays a role in tumorigenesis and apoptosis [105]. Interestingly, a recent paper reported that infections caused by enteric pathogens results in intestinal inflammation via aberrant expression of miR-122 and miR-21 [106]. Another study found *Bacteriodes uniformis* and *Phascolarctobacterium faecium* negatively correlated with miR-122-5p in the serum of patients with type 2 diabetes [73]. However, indirect evidence shows support for our findings that maternal Blautia may help regulate fetal liver lipid metabolism through miR-122 expression through unknown mechanisms. For example, decreased Blautia abundance was shown to be associated with obesity and intestinal inflammation in children [107] and Blautia species abundances have been associated with lowered visceral fat accumulation in human adults [108]. Conversely, *Ruminococcus is* positively associated with visceral fat accumulation [109] and negatively associated with miR-484 [104], suggesting Ruminococcus and Blautia have opposing influences on miRNA related to fetal lipid accumulation and tumorigenesis. Determining whether miR-1285-3p and miR-122-5p are early biomarkers for programmed obesity and NAFLD that may be regulated by maternal genera such as Blautia and *Ruminococcus is* an important avenue for future exploration.
Compared with CD-fed dams, WD-fed dams exhibited dyslipidaemia, characterized by elevated LDL‐cholesterol and TGs. However, based on the IVGTT at 0.6 G, these dams did not show indices of insulin resistance. These findings are supported by the observations of Short et al., wherein young (5-6 years of age) male baboons fed a WD (high in monosaccharides and saturated fatty acids) for eight weeks exhibited elevated levels of HDL and LDL/VLDL cholesterol and TGs, without a change in body weight or blood glucose [110]. Short et al. also noted significant effects on inflammatory indices, with enhanced CD14+ mononuclear cell chemotaxis as well as a significant increase in blood neutrophils [110]. We similarly found that WD-fed dams had mild, systemic inflammation, exemplified by elevated levels of CRP and increased neutrophil counts, which was consistent with some of the changes in cytokine profiles known to be exacerbated by maternal obesity. These observations suggest that upregulation of at least some pro-inflammatory programs during pregnancy are mediated by the diet alone.
Our finding of a trend for increased numbers of macrophages in the placentas of WD-fed dams is suggestive of enhanced chemoattraction of maternal monocytes or fetal monocytes/Hofbauer cells to the placenta, or activation of maternal peripheral monocytes that target the placenta, and is consistent with previous findings of monocyte priming and enhanced monocyte chemotaxis in male baboons fed a WD for eight weeks [110]. Peripheral monocytes from obese pregnancies displayed elevated chemokine receptor expression and enhanced migration capacity [23]. Further, placental resident CD14+ and CD68+ mononuclear cells (macrophages) increased two- to three-fold in obese pregnancies [23, 111]. Previously, Frias and colleagues noted increased placental inflammatory cytokine expression and placental infarctions in the Japanese macaque model of chronic WD feeding and maternal obesity [112]. We did not specifically address monocyte priming in the current study; however, we did note that mRNA expression of both TNF and IL8 mRNA was elevated in mWD-exposed placentas, consistent with WD-induced ‘priming’ of placental inflammation (in combination with recruitment of macrophages). We also did not find notable gross placental pathologies (calcifications, infarcts) in mWD-exposed compared with mCD-exposed placentas.
Our novel study has several strengths including our use of the olive baboon as a translational model for developmental programming. An advantage over rodents, baboons are similar to humans in gestational duration, placentation (hemochorial monodiscoid), singleton births, hormone profiles, and social behaviors. Further, our model allowed us to address the role of WD feeding independent from maternal obesity in nulliparous females. One drawback of this study is the small sample size, which limited the conclusions we could draw from our analyses. In a baboon model where maternal obesity was attained, Puppala et al. showed a trend toward increased hepatic lipid accumulation and more severe steatosis, assessed histologically, that did not progress toward NASH in the fetus [48]. This is in contrast to our findings and those of Wesolowski et al. [ 12] and McCurdy et al. [ 79] in obese Japanese macaques where fetuses from WD-fed dams had higher hepatic TGs associated with impaired mitochondrial function and increased fibrogenesis, which Nash et al. demonstrated was localized to the hepatic periportal region [113].
In contrast to our findings using a targeted analysis, Puppala et al. used an untargeted microarray-based approach and found that miR-145-3p was upregulated in livers from baboon fetuses exposed to maternal obesity and associated with a decrease in SMAD4; they additionally reported an increase in miR-182-5p and miR-183-5p [48]. Our findings are consistent with observations previously reported in obese Japanese macaques where a differential response to high-fat diet in dams was found, allowing segregation into insulin-sensitive and insulin-resistant subgroups [114]. Maternal insulin resistance (elevated TGs, insulin, and weight gain) led to activation of de novo lipogenic and pro-inflammatory pathways in liver from offspring at 1 year of age [114]. Moreover, Elsakr et al. showed that prolonged WD feeding, multiple diet switches, and increasing age and parity were associated with increased insulin resistance in dams [115]. The dams in our study, subjected to a relatively short-duration exposure to WD, were not insulin resistant and did not have significant weight gain compared with matched CD-fed dams; although the effects we observed in the fetal liver were modest, they are striking given that the maternal exposures were limited.
We conclude that, prior to the onset of obesity, a WD initiated several months preceding gestation and maintained over its course caused perturbed maternal lipid homeostasis and impacted maternal gut microbiota composition. Both maternal metabolic parameters and maternal gut microbiota were associated with expression of fetal liver miRNA and mRNA that are markers for lipid metabolism, oxidative stress, and inflammation, suggesting that maternal diet, in the absence of obesity, has significant consequences for epigenetic regulation of fetal and infant health.
## Data availability statement
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://www.ncbi.nlm.nih.gov/, 839382 (BioProject).
## Ethics statement
The animal study was reviewed and approved by University of Oklahoma Health Sciences Center Institutional Animal Care and Use Committee.
## Author contributions
JP, JF, KJ, and DM contributed to conception and design of the study. KS, AM, RJ, MC-C, and KJ contributed to writing the original draft of the manuscript. SG, JP, and DM collected maternal data. AM, RJ, SG, MD, and RB performed experiments. KS, MT, and DD performed microbiome and association analyses. DD, M-PA, JF, and DM contributed to supervision of the study. All authors contributed to the article and approved of the submitted version.
## Funding
Funding for this study was received from the Harold Hamm Foundation/Presbyterian Hospital Foundation, grant 20211471 (DM, JF, KJ, JP, DD, M-PA) and National Institute of Diabetes and Digestive and Kidney Diseases, grant R01-DK128416 (JF).
## Acknowledgments
We thank the OUHSC Stephenson Cancer Tissue Pathology Core, supported partly by NIGMS P20GM103639 and NCI P30CA225520, for the liver histology. We thank the OUHSC Laboratory for Molecular Biology and Cytometry Research core for library construction and sequencing.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcdhc.2022.945768/full#supplementary-material
## References
1. Pierantonelli I, Svegliati-Baroni G. **Nonalcoholic fatty liver disease: basic pathogenetic mechanisms in the progression from NAFLD to NASH**. *Transplantation* (2019) **103** e1-e13. DOI: 10.1097/tp.0000000000002480
2. Catalano P, deMouzon SH. **Maternal obesity and metabolic risk to the offspring: why lifestyle interventions may have not achieved the desired outcomes**. *Int J Obes (Lond)* (2015) **39**. DOI: 10.1038/ijo.2015.15
3. Chang E, Hafner H, Varghese M, Griffin C, Clemente J, Islam M. **Programming effects of maternal and gestational obesity on offspring metabolism and metabolic inflammation**. *Sci Rep* (2019) **9** 16027. DOI: 10.1038/s41598-019-52583-x
4. Segovia SA, Vickers MH, Gray C, Reynolds CM. **Maternal obesity, inflammation, and developmental programming**. *BioMed Res Int* (2014) **2014**. DOI: 10.1155/2014/418975
5. Parisi F, Milazzo R, Savasi VM, Cetin I. **Maternal low-grade chronic inflammation and intrauterine programming of health and disease**. *Int J Mol Sci* (2021) **22** 1732. DOI: 10.3390/ijms22041732
6. Cuzmar V, Alberti G, Uauy R, Pereira A, García C, De Barbieri F. **Early obesity: risk factor for fatty liver disease**. *J Pediatr Gastroenterol. Nutr* (2020) **70**. DOI: 10.1097/mpg.0000000000002523
7. Hanson MA, Gluckman PD. **Early developmental conditioning of later health and disease: physiology or pathophysiology**. *Physiol Rev* (2014) **94**. DOI: 10.1152/physrev.00029.2013
8. Cunningham SA, Kramer MR, Narayan KM. **Incidence of childhood obesity in the united states**. *N Engl J Med* (2014) **370**. DOI: 10.1056/NEJMoa1309753
9. Anderson EL, Howe LD, Jones HE, Higgins JP, Lawlor DA, Fraser A. **The prevalence of non-alcoholic fatty liver disease in children and adolescents: A systematic review and meta-analysis**. *PloS One* (2015) **10**. DOI: 10.1371/journal.pone.0140908
10. Freinkel N. **Banting lecture 1980. of pregnancy and progeny**. *Diabetes* (1980) **29**. DOI: 10.2337/diab.29.12.1023
11. Thompson MD, Derse A, Ferey J, Reid M, Xie Y, Christ M. **Transgenerational impact of maternal obesogenic diet on offspring bile acid homeostasis and nonalcoholic fatty liver disease**. *Am J Physiol Endocrinol Metab* (2019) **316**. DOI: 10.1152/ajpendo.00474.2018
12. Wesolowski SR, Mulligan CM, Janssen RC, Baker PR, Bergman BC, D'Alessandro A. **Switching obese mothers to a healthy diet improves fetal hypoxemia, hepatic metabolites, and lipotoxicity in non-human primates**. *Mol Metab* (2018) **18** 25-41. DOI: 10.1016/j.molmet.2018.09.008
13. Wankhade UD, Zhong Y, Kang P, Alfaro M, Chintapalli SV, Thakali KM. **Enhanced offspring predisposition to steatohepatitis with maternal high-fat diet is associated with epigenetic and microbiome alterations**. *PloS One* (2017) **12**. DOI: 10.1371/journal.pone.0175675
14. Mennitti LV, Carpenter AAM, Loche E, Pantaleão LC, Fernandez-Twinn DS, Schoonejans JM. **Effects of maternal diet-induced obesity on metabolic disorders and age-associated miRNA expression in the liver of male mouse offspring**. *Int J Obes (Lond)* (2022) **46**. DOI: 10.1038/s41366-021-00985-1
15. Schoonejans JM, Ozanne SE. **Developmental programming by maternal obesity: Lessons from animal models**. *Diabetes Med* (2021) **38**. DOI: 10.1111/dme.14694
16. Aye IL, Lager S, Ramirez VI, Gaccioli F, Dudley DJ, Jansson T. **Increasing maternal body mass index is associated with systemic inflammation in the mother and the activation of distinct placental inflammatory pathways**. *Biol Reprod* (2014) **90** 129. DOI: 10.1095/biolreprod.113.116186
17. Goldstein JA, Gallagher K, Beck C, Kumar R, Gernand AD. **Maternal-fetal inflammation in the placenta and the developmental origins of health and disease**. *Front Immunol* (2020) **11**. DOI: 10.3389/fimmu.2020.531543
18. Goeden N, Velasquez J, Arnold KA, Chan Y, Lund BT, Anderson GM. **Maternal inflammation disrupts fetal neurodevelopment**. *J Neurosci* (2016) **36**. DOI: 10.1523/jneurosci.2534-15.2016
19. Sauder MW, Lee SE, Schulze KJ, Christian P, Wu LSF, Khatry SK. **Inflammation throughout pregnancy and fetal growth restriction in rural Nepal**. *Epidemiol Infect* (2019) **147**. DOI: 10.1017/s0950268819001493
20. Burton GJ, Jauniaux E. **Pathophysiology of placental-derived fetal growth restriction**. *Am J Obstet Gynecol* (2018) **218**. DOI: 10.1016/j.ajog.2017.11.577
21. Pique-Regi R, Romero R, Tarca AL, Sendler ED, Xu Y, Garcia-Flores V. **Single cell transcriptional signatures of the human placenta in term and preterm parturition**. *Elife* (2019) **8**. DOI: 10.7554/eLife.52004
22. Miller D, Gershater M, Slutsky R, Romero R, Gomez-Lopez N. **Maternal and fetal T cells in term pregnancy and preterm labor**. *Cell Mol Immunol* (2020) **17** 693-704. DOI: 10.1038/s41423-020-0471-2
23. Challier JC, Basu S, Bintein T, Minium J, Hotmire K, Catalano PM. **Obesity in pregnancy stimulates macrophage accumulation and inflammation in the placenta**. *Placenta* (2008) **29**. DOI: 10.1016/j.placenta.2007.12.010
24. Trøseid M, Nestvold TK, Rudi K, Thoresen H, Nielsen EW, Lappegård KT. **Plasma lipopolysaccharide is closely associated with glycemic control and abdominal obesity: evidence from bariatric surgery**. *Diabetes Care* (2013) **36**. DOI: 10.2337/dc13-0451
25. Firmal P, Shah VK, Chattopadhyay S. **Insight into TLR4-mediated immunomodulation in normal pregnancy and related disorders**. *Front Immunol* (2020) **11**. DOI: 10.3389/fimmu.2020.00807
26. Collado MC, Isolauri E, Laitinen K, Salminen S. **Distinct composition of gut microbiota during pregnancy in overweight and normal-weight women**. *Am J Clin Nutr* (2008) **88**. DOI: 10.1093/ajcn/88.4.894
27. Dreisbach C, Morgan H, Cochran C, Gyamfi A, Henderson WA, Prescott S. **Metabolic and microbial changes associated with diet and obesity during pregnancy: What can we learn from animal studies**. *Front Cell Infect Microbiol* (2021) **11**. DOI: 10.3389/fcimb.2021.795924
28. Wu N, Zhou J, Mo H, Mu Q, Su H, Li M. **The gut microbial signature of gestational diabetes mellitus and the association with diet intervention**. *Front Cell Infect Microbiol* (2021) **11**. DOI: 10.3389/fcimb.2021.800865
29. Nuriel-Ohayon M, Neuman H, Koren O. **Microbial changes during pregnancy, birth, and infancy**. *Front Microbiol* (2016) **7**. DOI: 10.3389/fmicb.2016.01031
30. Chu DM, Meyer KM, Prince AL, Aagaard KM. **Impact of maternal nutrition in pregnancy and lactation on offspring gut microbial composition and function**. *Gut Microbes* (2016) **7**. DOI: 10.1080/19490976.2016.1241357
31. Ma J, Prince AL, Bader D, Hu M, Ganu R, Baquero K. **High-fat maternal diet during pregnancy persistently alters the offspring microbiome in a primate model**. *Nat Commun* (2014) **5** 3889. DOI: 10.1038/ncomms4889
32. Soderborg TK, Clark SE, Mulligan CE, Janssen RC, Babcock L, Ir D. **The gut microbiota in infants of obese mothers increases inflammation and susceptibility to NAFLD**. *Nat Commun* (2018) **9** 4462. DOI: 10.1038/s41467-018-06929-0
33. Coppola S, Avagliano C, Calignano A, Berni Canani R. **The protective role of butyrate against obesity and obesity-related diseases**. *Molecules* (2021) **26** 682. DOI: 10.3390/molecules26030682
34. Chen J, Vitetta L. **Gut microbiota metabolites in NAFLD pathogenesis and therapeutic implications**. *Int J Mol Sci* (2020) **21** 5214. DOI: 10.3390/ijms21155214
35. Chu H, Duan Y, Yang L, Schnabl B. **Small metabolites, possible big changes: a microbiota-centered view of non-alcoholic fatty liver disease**. *Gut* (2019) **68**. DOI: 10.1136/gutjnl-2018-316307
36. Sanidad KZ, Zeng MY. **Neonatal gut microbiome and immunity**. *Curr Opin Microbiol* (2020) **56**. DOI: 10.1016/j.mib.2020.05.011
37. Wang YW, Yu HR, Tiao MM, Tain YL, Lin IC, Sheen JM. **Maternal obesity related to high fat diet induces placenta remodeling and gut microbiome shaping that are responsible for fetal liver lipid dysmetabolism**. *Front Nutr* (2021) **8**. DOI: 10.3389/fnut.2021.736944
38. Bedell S, Hutson J, de Vrijer B, Eastabrook G. **Effects of maternal obesity and gestational diabetes mellitus on the placenta: Current knowledge and targets for therapeutic interventions**. *Curr Vasc Pharmacol* (2021) **19**. DOI: 10.2174/1570161118666200616144512
39. Gohir W, Kennedy KM, Wallace JG, Saoi M, Bellissimo CJ, Britz-McKibbin P. **High-fat diet intake modulates maternal intestinal adaptations to pregnancy and results in placental hypoxia, as well as altered fetal gut barrier proteins and immune markers**. *J Physiol* (2019) **597**. DOI: 10.1113/jp277353
40. Liong S, Lappas M. **Lipopolysaccharide and double stranded viral RNA mediate insulin resistance and increase system a amino acid transport in human trophoblast cells**. *Placenta* (2017) **51** 18-27. DOI: 10.1016/j.placenta.2017.01.124
41. Kato E, Yamamoto T, Chishima F. **Effects of cytokines and TLR ligands on the production of PlGF and sVEGFR1 in primary trophoblasts**. *Gynecol Obstet Invest* (2017) **82** 39-46. DOI: 10.1159/000446279
42. Seki Y, Suzuki M, Guo X, Glenn AS, Vuguin PM, Fiallo A. *Endocrinology* (2017) **158**. DOI: 10.1210/en.2017-00334
43. Goldberg AD, Allis CD, Bernstein E. **Epigenetics: a landscape takes shape**. *Cell* (2007) **128**. DOI: 10.1016/j.cell.2007.02.006
44. Wu J, Nagy LE, Liangpunsakul S, Wang L. **Non-coding RNA crosstalk with nuclear receptors in liver disease**. *Biochim Biophys Acta Mol Basis Dis* (2021) **1867**. DOI: 10.1016/j.bbadis.2021.166083
45. Gjorgjieva M, Sobolewski C, Dolicka D, Correia de Sousa M, Foti M. **miRNAs and NAFLD: from pathophysiology to therapy**. *Gut* (2019) **68**. DOI: 10.1136/gutjnl-2018-318146
46. Roderburg C, Urban GW, Bettermann K, Vucur M, Zimmermann H, Schmidt S. **Micro-RNA profiling reveals a role for miR-29 in human and murine liver fibrosis**. *Hepatology* (2011) **53**. DOI: 10.1002/hep.23922
47. Mandala A, Janssen RC, Palle S, Short KR, Friedman JE. **Pediatric non-alcoholic fatty liver disease: Nutritional origins and potential molecular mechanisms**. *Nutrients* (2020) **12**. DOI: 10.3390/nu12103166
48. Puppala S, Li C, Glenn JP, Saxena R, Gawrieh S, Quinn A. **Primate fetal hepatic responses to maternal obesity: epigenetic signalling pathways and lipid accumulation**. *J Physiol* (2018) **596**. DOI: 10.1113/jp275422
49. Doridot L, Houry D, Gaillard H, Chelbi ST, Barbaux S, Vaiman D. **miR-34a expression, epigenetic regulation, and function in human placental diseases**. *Epigenetics* (2014) **9**. DOI: 10.4161/epi.26196
50. Maccani MA, Avissar-Whiting M, Banister CE, McGonnigal B, Padbury JF, Marsit CJ. **Maternal cigarette smoking during pregnancy is associated with downregulation of miR-16, miR-21, and miR-146a in the placenta**. *Epigenetics* (2010) **5**. DOI: 10.4161/epi.5.7.12762
51. Avissar-Whiting M, Veiga KR, Uhl KM, Maccani MA, Gagne LA, Moen EL. **Bisphenol a exposure leads to specific microRNA alterations in placental cells**. *Reprod Toxicol* (2010) **29**. DOI: 10.1016/j.reprotox.2010.04.004
52. Shah KB, Chernausek SD, Teague AM, Bard DE, Tryggestad JB. **Maternal diabetes alters microRNA expression in fetal exosomes, human umbilical vein endothelial cells and placenta**. *Pediatr Res* (2020) **89**. DOI: 10.1038/s41390-020-1060-x
53. Tsamou M, Martens DS, Winckelmans E, Madhloum N, Cox B, Gyselaers W. **Mother's pre-pregnancy BMI and placental candidate miRNAs: Findings from the ENVIRONAGE birth cohort**. *Sci Rep* (2017) **7** 5548. DOI: 10.1038/s41598-017-04026-8
54. Li C, Jenkins S, Considine MM, Cox LA, Gerow KG, Huber HF. **Effect of maternal obesity on fetal and postnatal baboon (Papio species) early life phenotype**. *J Med Primatol* (2019) **48**. DOI: 10.1111/jmp.12396
55. Rivera HM, Kievit P, Kirigiti MA, Bauman LA, Baquero K, Blundell P. **Maternal high-fat diet and obesity impact palatable food intake and dopamine signaling in nonhuman primate offspring**. *Obes (Silver Spring)* (2015) **23**. DOI: 10.1002/oby.21306
56. Bolyen E, Rideout JR, Dillon MR, Bokulich NA, Abnet CC, Al-Ghalith GA. **Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2**. *Nat Biotechnol* (2019) **37**. DOI: 10.1038/s41587-019-0209-9
57. Callahan BJ, McMurdie PJ, Rosen MJ, Han AW, Johnson AJ, Holmes SP. **DADA2: High-resolution sample inference from illumina amplicon data**. *Nat Methods* (2016) **13**. DOI: 10.1038/nmeth.3869
58. Katoh K, Standley DM. **MAFFT multiple sequence alignment software version 7: improvements in performance and usability**. *Mol Biol Evol* (2013) **30**. DOI: 10.1093/molbev/mst010
59. Price MN, Dehal PS, Arkin AP. **FastTree 2–approximately maximum-likelihood trees for large alignments**. *PloS One* (2010) **5**. DOI: 10.1371/journal.pone.0009490
60. Bokulich NA, Kaehler BD, Rideout JR, Dillon M, Bolyen E, Knight R. **Optimizing taxonomic classification of marker-gene amplicon sequences with QIIME 2's q2-feature-classifier plugin**. *Microbiome* (2018) **6** 90. DOI: 10.1186/s40168-018-0470-z
61. McDonald D, Price MN, Goodrich J, Nawrocki EP, DeSantis TZ, Probst A. **An improved greengenes taxonomy with explicit ranks for ecological and evolutionary analyses of bacteria and archaea**. *Isme J* (2012) **6**. DOI: 10.1038/ismej.2011.139
62. Segata N, Izard J, Waldron L, Gevers D, Miropolsky L, Garrett WS. **Metagenomic biomarker discovery and explanation**. *Genome Biol* (2011) **12** R60. DOI: 10.1186/gb-2011-12-6-r60
63. Tibshirani R. **Regression shrinkage and selection**. *J R Statist Soc* (1996) **58**. DOI: 10.1111/j.2517-6161.1996.tb02080.x
64. Friedman J, Hastie T, Tibshirani R. **Regularization paths for generalized linear models**. *J Stat. Softw* (2010) **33** 1-22. DOI: 10.18637/jss.v033.i01
65. Wang Z. *Mpath: Regularized linear models* (2022)
66. Venables WN, Ripley BD. *Modern applied statistics with s* (2002)
67. Ahrens WH, Cox DJ, Budhwar G. **Use of the arcsine and square root transformations for subjectively determined percentage data**. *Weed Sci* (1990) **38**. DOI: 10.1017/S0043174500056824
68. Li C, Jenkins S, Huber HF, Nathanielsz PW. **Effect of maternal baboon (Papio sp.) dietary mismatch in pregnancy and lactation on post-natal offspring early life phenotype**. *J Med Primatol* (2019) **48**. DOI: 10.1111/jmp.12415
69. Maloyan A, Muralimanoharan S, Huffman S, Cox LA, Nathanielsz PW, Myatt L. **Identification and comparative analyses of myocardial miRNAs involved in the fetal response to maternal obesity**. *Physiol Genomics* (2013) **45** 889-900. DOI: 10.1152/physiolgenomics.00050.2013
70. Williams MR, Stedtfeld RD, Tiedje JM, Hashsham SA. **MicroRNAs-based inter-domain communication between the host and members of the gut microbiome**. *Front Microbiol* (2017) **8**. DOI: 10.3389/fmicb.2017.01896
71. Cossiga V, Lembo V, Nigro C, Mirra P, Miele C, D'Argenio V. **The combination of berberine, tocotrienols and coffee extracts improves metabolic profile and liver steatosis by the modulation of gut microbiota and hepatic miR-122 and miR-34a expression in mice**. *Nutrients* (2021) **13** 1281. DOI: 10.3390/nu13041281
72. Gaddam RR, Jacobsen VP, Kim YR, Gabani M, Jacobs JS, Dhuri K. **Microbiota-governed microRNA-204 impairs endothelial function and blood pressure decline during inactivity in db/db mice**. *Sci Rep* (2020) **10** 10065. DOI: 10.1038/s41598-020-66786-0
73. Li L, Li C, Lv M, Hu Q, Guo L, Xiong D. **Correlation between alterations of gut microbiota and miR-122-5p expression in patients with type 2 diabetes mellitus**. *Ann Transl Med* (2020) **8** 1481. DOI: 10.21037/atm-20-6717
74. Vikram A, Kim YR, Kumar S, Li Q, Kassan M, Jacobs JS. **Vascular microRNA-204 is remotely governed by the microbiome and impairs endothelium-dependent vasorelaxation by downregulating Sirtuin1**. *Nat Commun* (2016) **7**. DOI: 10.1038/ncomms12565
75. Zhao J, Tang N, Wu K, Dai W, Ye C, Shi J. **MiR-21 simultaneously regulates ERK1 signaling in HSC activation and hepatocyte EMT in hepatic fibrosis**. *PloS One* (2014) **9**. DOI: 10.1371/journal.pone.0108005
76. Lin H, Mercer KE, Ou X, Mansfield K, Buchmann R, Børsheim E. **Circulating microRNAs are associated with metabolic markers in adolescents with hepatosteatosis**. *Front Endocrinol (Lausanne)* (2022) **13**. DOI: 10.3389/fendo.2022.856973
77. Kim TH, Lee Y, Lee YS, Gim JA, Ko E, Yim SY. **Circulating miRNA is a useful diagnostic biomarker for nonalcoholic steatohepatitis in nonalcoholic fatty liver disease**. *Sci Rep* (2021) **11** 14639. DOI: 10.1038/s41598-021-94115-6
78. Fu J, Wu B, Zhong S, Deng W, Lin F. **miR-29a-3p suppresses hepatic fibrosis pathogenesis by modulating hepatic stellate cell proliferation**. *Biochem Biophys Res Commun* (2020) **529**. DOI: 10.1016/j.bbrc.2020.06.102
79. McCurdy CE, Bishop JM, Williams SM, Grayson BE, Smith MS, Friedman JE. **Maternal high-fat diet triggers lipotoxicity in the fetal livers of nonhuman primates**. *J Clin Invest* (2009) **119**. DOI: 10.1172/JCI32661
80. Xiang B, Zhong P, Fang L, Wu X, Song Y, Yuan H. **miR-183 inhibits microglia activation and expression of inflammatory factors in rats with cerebral ischemia reperfusion**. *Exp Ther Med* (2019) **18**. DOI: 10.3892/etm.2019.7827
81. Liang Q, Chen H, Xu X, Jiang W. **miR-182-5p attenuates high-fat -diet-induced nonalcoholic steatohepatitis in mice**. *Ann Hepatol* (2019) **18**. DOI: 10.5604/01.3001.0012.7902
82. Carreras-Badosa G, Bonmatí A, Ortega FJ, Mercader JM, Guindo-Martínez M, Torrents D. **Dysregulation of placental miRNA in maternal obesity is associated with pre- and postnatal growth**. *J Clin Endocrinol Metab* (2017) **102**. DOI: 10.1210/jc.2017-00089
83. Zhou T, He C, Lai P, Yang Z, Liu Y, Xu H. **miR-204-containing exosomes ameliorate GVHD-associated dry eye disease**. *Sci Adv* (2022) **8**. DOI: 10.1126/sciadv.abj9617
84. Li H, Wang J, Liu X, Cheng Q. **MicroRNA-204-5p suppresses IL6-mediated inflammatory response and chemokine generation in HK-2 renal tubular epithelial cells by targeting IL6R**. *Biochem Cell Biol* (2019) **97**. DOI: 10.1139/bcb-2018-0141
85. Huang Y, Du KL, Guo PY, Zhao RM, Wang B, Zhao XL. **IL-16 regulates macrophage polarization as a target gene of mir-145-3p**. *Mol Immunol* (2019) **107** 1-9. DOI: 10.1016/j.molimm.2018.12.027
86. Li M, Chen WD, Wang YD. **The roles of the gut microbiota-miRNA interaction in the host pathophysiology**. *Mol Med* (2020) **26** 101. DOI: 10.1186/s10020-020-00234-7
87. Prince AL, Pace RM, Dean T, Takahashi D, Kievit P, Friedman JE. **The development and ecology of the Japanese macaque gut microbiome from weaning to early adolescence in association with diet**. *Am J Primatol* (2019) **81**. DOI: 10.1002/ajp.22980
88. Lopetuso LR, Scaldaferri F, Petito V, Gasbarrini A. **Commensal clostridia: leading players in the maintenance of gut homeostasis**. *Gut Pathog* (2013) **5**. DOI: 10.1186/1757-4749-5-23
89. Kang TS, Korber DR, Tanaka T. **Regulation of dual glycolytic pathways for fructose metabolism in heterofermentative lactobacillus panis PM1**. *Appl Environ Microbiol* (2013) **79**. DOI: 10.1128/aem.02377-13
90. De Bandt JP, Waligora-Dupriet AJ, Butel MJ. **Intestinal microbiota in inflammation and insulin resistance: relevance to humans**. *Curr Opin Clin Nutr Metab Care* (2011) **14**. DOI: 10.1097/MCO.0b013e328347924a
91. Ley RE, Turnbaugh PJ, Klein S, Gordon JI. **Microbial ecology: human gut microbes associated with obesity**. *Nature* (2006) **444**. DOI: 10.1038/4441022a
92. Turnbaugh PJ, Hamady M, Yatsunenko T, Cantarel BL, Duncan A, Ley RE. **A core gut microbiome in obese and lean twins**. *Nature* (2009) **457**. DOI: 10.1038/nature07540
93. Bäckhed F, Ding H, Wang T, Hooper LV, Koh GY, Nagy A. **The gut microbiota as an environmental factor that regulates fat storage**. *Proc. Natl. Acad. Sci U.S.A* (2004) **101**. DOI: 10.1073/pnas.0407076101
94. Magne F, Gotteland M, Gauthier L, Zazueta A, Pesoa S, Navarrete P. **The Firmicutes/Bacteroidetes ratio: A relevant marker of gut dysbiosis in obese patients**. *Nutrients* (2020) **12** 1474. DOI: 10.3390/nu12051474
95. Nagpal R, Shively CA, Appt SA, Register TC, Michalson KT, Vitolins MZ. **Gut microbiome composition in non-human primates consuming a Western or Mediterranean diet**. *Front Nutr* (2018) **5**. DOI: 10.3389/fnut.2018.00028
96. Lang JM, Sedgeman LR, Cai L, Layne JD, Wang Z, Pan C. **Dietary and pharmacologic manipulations of host lipids and their interaction with the gut microbiome in non-human primates**. *Front Med (Lausanne)* (2021) **8**. DOI: 10.3389/fmed.2021.646710
97. Ferrocino I, Ponzo V, Gambino R, Zarovska A, Leone F, Monzeglio C. **Changes in the gut microbiota composition during pregnancy in patients with gestational diabetes mellitus (GDM)**. *Sci Rep* (2018) **8** 12216. DOI: 10.1038/s41598-018-30735-9
98. Crusell MKW, Hansen TH, Nielsen T, Allin KH, Rühlemann MC, Damm P. **Gestational diabetes is associated with change in the gut microbiota composition in third trimester of pregnancy and postpartum**. *Microbiome* (2018) **6** 89. DOI: 10.1186/s40168-018-0472-x
99. Taricco E, Radaelli T, Nobile de Santis MS, Cetin I. **Foetal and placental weights in relation to maternal characteristics in gestational diabetes**. *Placenta* (2003) **24**. DOI: 10.1053/plac.2002.0913
100. Morishita A, Oura K, Tadokoro T, Fujita K, Tani J, Masaki T. **MicroRNA interference in hepatic host-pathogen interactions**. *Int J Mol Sci* (2021) **22** 3554. DOI: 10.3390/ijms22073554
101. Hu Y, Peng X, Du G, Zhang Z, Zhai Y, Xiong X. **MicroRNA-122-5p inhibition improves inflammation and oxidative stress damage in dietary-induced non-alcoholic fatty liver disease through targeting FOXO3**. *Front Physiol* (2022) **13**. DOI: 10.3389/fphys.2022.803445
102. Raitoharju E, Seppälä I, Lyytikäinen LP, Viikari J, Ala-Korpela M, Soininen P. **Blood hsa-miR-122-5p and hsa-miR-885-5p levels associate with fatty liver and related lipoprotein metabolism-the young finns study**. *Sci Rep* (2016) **6**. DOI: 10.1038/srep38262
103. Yuan C, Burns MB, Subramanian S, Blekhman R. **Interaction between host microRNAs and the gut microbiota in colorectal cancer**. *mSystems* (2018) **3**. DOI: 10.1128/mSystems.00205-17
104. Dong J, Gu W, Yang X, Zeng L, Wang X, Mu J. **Crosstalk between polygonatum kingianum, the miRNA, and gut microbiota in the regulation of lipid metabolism**. *Front Pharmacol* (2021) **12**. DOI: 10.3389/fphar.2021.740528
105. Jia YZ, Liu J, Wang GQ, Song ZF. **miR-484: A potential biomarker in health and disease**. *Front Oncol* (2022) **12**. DOI: 10.3389/fonc.2022.830420
106. Rashid H, Siddiqua TJ, Hossain B, Siddique A, Kabir M, Noor Z. **MicroRNA expression and intestinal permeability in children living in a slum area of Bangladesh**. *Front Mol Biosci* (2021) **8**. DOI: 10.3389/fmolb.2021.765301
107. Benítez-Páez A, Gómez Del Pugar EM, López-Almela I, Moya-Pérez Á, Codoñer-Franch P, Sanz Y. **Depletion of blautia species in the microbiota of obese children relates to intestinal inflammation and metabolic phenotype worsening**. *mSystems* (2020) **5**. DOI: 10.1128/mSystems.00857-19
108. Ozato N, Yamaguchi T, Mori K, Katashima M, Kumagai M, Murashita K. **Two blautia species associated with visceral fat accumulation: A one-year longitudinal study**. *Biol (Basel)* (2022) **11** 318. DOI: 10.3390/biology11020318
109. Yan H, Qin Q, Chen J, Yan S, Li T, Gao X. **Gut microbiome alterations in patients with visceral obesity based on quantitative computed tomography**. *Front Cell Infect Microbiol* (2021) **11**. DOI: 10.3389/fcimb.2021.823262
110. Short JD, Tavakoli S, Nguyen HN, Carrera A, Farnen C, Cox LA. **Dyslipidemic diet-induced monocyte "priming" and dysfunction in non-human primates is triggered by elevated plasma cholesterol and accompanied by altered histone acetylation**. *Front Immunol* (2017) **8**. DOI: 10.3389/fimmu.2017.00958
111. Basu S, Haghiac M, Surace P, Challier JC, Guerre-Millo M, Singh K. **Pregravid obesity associates with increased maternal endotoxemia and metabolic inflammation**. *Obes (Silver Spring)* (2011) **19**. DOI: 10.1038/oby.2010.215
112. Frias AE, Morgan TK, Evans AE, Rasanen J, Oh KY, Thornburg KL. **Maternal high-fat diet disturbs uteroplacental hemodynamics and increases the frequency of stillbirth in a nonhuman primate model of excess nutrition**. *Endocrinology* (2011) **152**. DOI: 10.1210/en.2010-1332
113. Nash MJ, Dobrinskikh E, Newsom SA, Messaoudi I, Janssen RC, Aagaard KM. **Maternal Western diet exposure increases periportal fibrosis beginning**. *JCI Insight* (2021) **6**. DOI: 10.1172/jci.insight.154093
114. Thorn SR, Baquero KC, Newsom SA, El Kasmi KC, Bergman BC, Shulman GI. **Early life exposure to maternal insulin resistance has persistent effects on hepatic NAFLD in juvenile nonhuman primates**. *Diabetes* (2014) **63**. DOI: 10.2337/db14-0276
115. Elsakr JM, Zhao SK, Ricciardi V, Dean TA, Takahashi DL, Sullivan E. **Western-Style diet consumption impairs maternal insulin sensitivity and glucose metabolism during pregnancy in a Japanese macaque model**. *Sci Rep* (2021) **11** 12977. DOI: 10.1038/s41598-021-92464-w
|
---
title: 'Prevalence of Cardiometabolic Syndrome and its Association With Body Shape
Index and A Body Roundness Index Among Type 2 Diabetes Mellitus Patients: A Hospital-Based
Cross-Sectional Study in a Ghanaian Population'
authors:
- Enoch Odame Anto
- Joseph Frimpong
- Wina Ivy Ofori Boadu
- Valentine Christian Kodzo Tsatsu Tamakloe
- Charity Hughes
- Benjamin Acquah
- Emmanuel Acheampong
- Evans Adu Asamoah
- Stephen Opoku
- Michael Appiah
- Augustine Tawiah
- Max Efui Annani-Akollor
- Yaw Amo Wiafe
- Otchere Addai-Mensah
- Christian Obirikorang
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2022
pmcid: PMC10012128
doi: 10.3389/fcdhc.2021.807201
license: CC BY 4.0
---
# Prevalence of Cardiometabolic Syndrome and its Association With Body Shape Index and A Body Roundness Index Among Type 2 Diabetes Mellitus Patients: A Hospital-Based Cross-Sectional Study in a Ghanaian Population
## Abstract
Cardiometabolic syndrome (MetS) is closely linked to type 2 diabetes mellitus (T2DM) and is the leading cause of diabetes complications. Anthropometric indices could be used as a cheap approach to identify MetS among T2DM patients. We determined the prevalence of MetS and its association with sociodemographic and anthropometric indices among T2DM patients in a tertiary hospital in the Ashanti region of Ghana. A comparative cross-sectional study was conducted among 241 T2DM outpatients attending the Komfo Anokye Teaching Hospital (KATH) and the Kumasi South Hospital for routine check-up. Sociodemographic characteristics, clinicobiochemical markers, namely, systolic blood pressure (SBP), diastolic blood pressure (DBP), fasting blood glucose (FBG), and glycated hemoglobin (HbA1C) were measured. Anthropometric indices, namely, body mass index (BMI), Conicity index (CI), body adiposity index (BAI), A body shape index (ABSI), body roundness index (BRI), Waist-to-hip ratio (WHR), and Waist-to-height ratio (WHtR) were computed based on either the Height, Weight, Waist circumference (WC) or Hip circumference (HC) of the patients. Metabolic syndrome (MetS) was classified using the National Cholesterol Education Program (NCEP) Adult Treatment Panel III (ATP III) criteria. Data entry and analysis were done using Excel 2016 and SPSS version 25.0 respectively. Of the 241 T2DM patients, 99 ($41.1\%$) were males whereas 144 ($58.9\%$) were females. The prevalence of cardiometabolic syndrome (MetS) was $42.7\%$ with dyslipidemia and hypertension recording a prevalence of 6.6 and $36.1\%$, respectively. Being a female T2DM patient [aOR = 3.02, $95\%$CI (1.59–5.76), $$p \leq 0.001$$] and divorced [aOR = 4.05, $95\%$CI (1.22–13.43), $$p \leq 0.022$$] were the independent sociodemographic predictors of MetS among T2DM patients. The 4th quartile for ABSI and 2nd to 4th quartiles for BSI were associated with MetS on univariate logistic regression ($p \leq 0.05$). Multivariate logistic regression identified the 3rd quartile (aOR = 25.15 (2.02–313.81), $$p \leq 0.012$$) and 4th quartile (aOR = 39.00, $95\%$CI (2.68–568.49), $$p \leq 0.007$$) for BRI as the independent predictors of MetS among T2DM. The prevalence of cardiometabolic syndrome is high among T2DM patients and this was influenced by female gender, being divorced, and increased BRI. Integration of BRI as part of routine assessment could be used as early indicator of cardiometabolic syndrome among T2DM patients.
## Introduction
Diabetes mellitus (DM) has become a public health concern and its morbidity and mortality rates have continued to rise gradually [1]. It was estimated that the global prevalence of type 2 *Diabetes mellitus* (T2DM) in 2019 was $7.5\%$ (374 million) and is expected to reach $8.0\%$ (454 million) by 2030 and $8.6\%$ (548 million) by 2045 [2]. Furthermore, reports are that in the 21st century, developing countries will face the risk of this epidemic, with $80\%$ of all new DM cases due to occur in Sub-Saharan African countries like Ghana by 2025 [3]. In Africa alone, an estimated 15.5 million adults aged 20 to 79 had diabetes, which represents a regional prevalence of $3.3\%$ [4]. Moreover, T2DM accounted for over 298,160 deaths ($6\%$ of all deaths) in Africa region that same year [4]. An estimated 19.4 million adults (20–79 years) lived with diabetes in the International Diabetes Federation (IDF) Africa Region and this signifies a $3.9\%$ regional prevalence [5]. The Region with the highest fraction of undiagnosed diabetes is Africa, with $60\%$ of adults still living with diabetes and not aware of it [5]. In urban Ghana, at least $6\%$ were diagnosed with T2DM and were related to obesity, age, and low socioeconomic status, and often leading to cardiometabolic risk factors such as hypertension and dyslipidemia [6]. T2DM causes more havoc by its strong association with cardiometabolic risk factors such as dyslipidemia, metabolic syndrome and hypertension and its primary driving factor is overweight and obesity [7]. In a Ghanaian population, cardiometabolic risk factors were found to have increased among urban settlers as a result of increased physical inactiveness and unhealthy eating habits among the urban settlers and the association between obesity and T2DM has been well documented (8–10). Overweight and Obesity are linked to increased cardiometabolic risk but can differ significantly depending on gender, age, eating habits, and even among subjects with morbid obesity [9].
Anthropometric indices such as BMI used to access obesity have been accepted in clinical practices due to its simplicity and usefulness for the prediction of body fat distribution in Diabetes Mellitus [10, 11]. Body mass index (BMI) has been the traditional anthropometric index for general obesity diagnosis and reflects the total body fat distribution [12]. BMI is however, limited by its inability to differentiate fat and muscle mass, and also overall distribution of body fat [13]. A previous report has shown that the conventional anthropometric indices such as BMI could not differentiate muscle mass and body fat [14]. Other anthropometric indices such as Waist Circumference (WC), Weight to Height Ratio (WHtR), Waist to Hip Ratio (WHR), Body Adiposity Index (BAI), and Conicity Index (CI) have been used to predict the various cardiometabolic risk factors in T2DM patients (15–17). Due to endpoint dissimilarity between men and women and also different racial groups, the validity of WC has also been questioned for clinical use in cardiometabolic risk assessment [13, 18]. Likewise, WHR as a measure of fat distribution necessitates endpoints for ethnic group and sex [19]. The use of WHtR as a standardized tool for ascertaining central obesity between varied racial groups has also been questioned [20]. BAI is also limited by severe obesity and its validity has been questioned [21, 22] and optimal cut-off points of CI is also limited by sex and age [23].
The development of other anthropometric indices to improve the limitation of other anthropometric has been explored [24]. Two new body indices have been formulated lately [25, 26]. A new body index referred to as A Body Shape Index (ABSI) was introduced by Krakauer and Krakauer [25]. This index takes into account one’s waist circumference, height, and weight. Krakauer and Krakauer discovered that ABSI values and abdominal body fat were positively correlated and recent studies have used ABSI to predict premature mortality [27, 28]. Thomas et al. [ 26] introduced another new index called the Body Roundness Index (BRI) in 2013. This index takes WC and height into account. However, the controversy over the association of the two new body indices (ABSI and BRI) and cardiometabolic risk factors among diabetes patients is yet to be explored in a Ghanaian setting [16].
Previous studies conducted showed that the two new indices—ABSI and BRI were more related to cardiometabolic risk factors than WC and BMI (29–31). However, studies on the association of BRI and ABSI with cardiometabolic risk factors among Ghanaian Diabetes patients are non-existent. It is therefore imperative that this study is done to evaluate the relationship between the two new indices and cardiometabolic risk factors among Type 2 Diabetes patients in the Ghanaian population.
## Study Design
A hospital-based comparative cross-sectional study was conducted between March 2021 and June 2021 at the Komfo Anokye Teaching Hospital (KATH) and Kumasi South Hospital, Agogo after obtaining permission from the Institutional Ethics Committee.
## Study Setting
KATH is a 1,200-bed facility situated in Kumasi in the Ashanti region, Ghana. The Ashanti region is situated centrally in Ghana’s middle belt and lies between longitudes 0.15W and 2.25W, and latitudes 5.50N and 7.46N. Kumasi is second only to Accra in population density. Its strategic geographic position has granted it the status of the main transport depot and guaranteed its central role in an immense and lucrative distribution of goods not only in the country but beyond. This has made KATH one of the nation’s most assessable tertiary medical centers. In Kumasi, there are nine sub-metros, including the Bantama sub-metro, where KATH is situated. KATH is Ghana’s second-largest hospital and the only tertiary health institution of the Ashanti Region. It is the primary referral hospital for Ashanti, Northern, Brong Ahafo, and Western Regions in Ghana. It also receives referrals from other neighboring countries such as Burkina Faso and Ivory Coast. For easy management and specialization, the hospital has been divided into fifteen [15] Directorates. Out of the 15, two are non-clinical and thirteen are clinical. Several clinical and non-clinical supporting units are also there. Kumasi South *Hospital is* situated in Agogo and is the second largest in the Southern part of Ghana. The hospital was built in 1976, as an urban health center and was later changed to be the Kumasi South Hospital. It was upgraded to the status of Ashanti Regional Hospital in 2002. The *Kumasi metropolis* has a total population of 3,490,030 (2021 population census).
## Study Population and Sample Size Estimation
A total of 241 T2DM patients were recruited based on the inclusion criteria of the study until the required sample size was attained. The diagnosis of type 2 diabetes mellitus was made based on the American Diabetes Association (ADA) criteria [32]. The sample size was estimated using the formula n = Z2 × p (1 − p)/d2 (Charan & Biswas, 2013), where n = sample size, $Z = 1.96$, p = prevalence, and d = marginal error (0.05). Using a prevalence of $90.6\%$ obtained from a similar study conducted by Agyemang-Yeboah et al. [ 33] in the Bantama sub-metro, the estimated sample size (n) was 124. To increase statistical power and account for non-response distribution, 241 T2DM patients were sampled for the study.
## Inclusion and Exclusion Criteria
Outpatients T2DM patients who gave consent to the study were the only ones recruited. Outpatients 30 years to 78 years who had type 2 diabetes mellitus and consented to participate were recruited and included into the study. Pregnant women and outpatients diagnosed with gestational diabetes or type 1 diabetes mellitus were excluded from the study. Individuals below 30 years were also excluded from the study and those who suffered from chronic conditions (hypertension, stroke, HIV, tuberculosis and cancer) were as well excluded from this study.
## Ethical Consideration
Approval was sought from the Committee on Human Research, Publication and Ethics (CHRPE) at the School of Medical Sciences of the Kwame Nkrumah University of Science and Technology (KNUST) and the Komfo Anokye Teaching Hospital (KATH), Kumasi and the Kumasi South Hospital. Written informed consent was sought from each participant before the commencement of the study.
## Data Collection
Participants were first educated on the purpose of the study and only those who gave consent to participate in the study were recruited. A self-reported questionnaire was used to obtain information about the name, age, gender, and sociodemographic factors such as marital status, level of education, occupation, family history of diabetes, level of physical activity, smoking status, and alcohol status of the participants.
## Blood Pressure Measurement
Blood pressure was measured by qualified nurses using a mercury sphygmomanometer and stethoscope. Recommendation of the American Heart Association (AHA) was used to take measurements from the upper left arm after participants had sat for more than 5 min (Kirkendall, Burton, Epstein, & Freis, 1967). The average value for the two measurements (with a 5-minute break interval between measurements) was recorded to the nearest 2.0 mmHg.
## Anthropometric Measurements
Anthropometric measurements included height, weight, WC, HC, BMI, WHR, WHtR, BAI, CI, and the two new indices—ABSI and BRI. The height of subjects was measured to the nearest 0.1 cm without shoes and weight was also measured to the nearest 0.1 kg with participants in light clothing. A bathroom scale (Zhongshan Camry Electronic Co. Ltd., Guangdong, China) was used to weigh the participants and their height was measured with a stadiometer (Seca 213 mobile stadiometer, Germany). During height measurement, participants stood upright with back straight, heels together, and their feet slightly apart at a 60° angle. Waist circumference (to the nearest 0.1 cm) was measured with a Gulick II spring-loaded measuring tape (Gay Mills, WI) halfway between the inferior angles of the ribs and the suprailiac crests. The hip circumference was measured at the widest diameter around the gluteal protuberance to the nearest 0.1 cm. The other anthropometric indices were calculated as follows: BMI was calculated according to Quetelet’s formula [34]: CI was calculated from the formula [23]: The BAI was calculated from the formula [35]:
ABSI was calculated from the formula [25]: BRI was calculated by the formula [26]:
## Blood Sampling and Biochemical Analysis
A volume of five [5] milliliters (mls) venous blood samples were collected after an overnight fast; 4 ml was dispensed into a serum separator tube and 1 ml into fluoride-oxalate tubes. After centrifugation at 500g for 15 min, the serum and plasma were stored at −80°C until assayed. Parameters included *Fasting plasma* glucose (FPG), HbA1C, total cholesterol (TC), low-density lipoprotein (LDL), triglycerides (TG), high-density lipoprotein (HDL) cholesterol, and were assayed using the COBAS INTEGRA(R) 400 plus Automated Chemistry Analyzer. The protocol for the determination of the parameters was as indicated in the manufacturer’s instructions (Fortress Diagnostics Limited, Unit 2C Antrim Technology Park, Antrim BT41 1QS, United Kingdom).
## Definition of Clinicobiochemical Terms
Dyslipidemia was defined as follows: high TC (>5.0 mmol/L), high LDL-C (>3.0 mmol/L), high TG (>1.7 mmol/L), low HDL-C (<1.0 mml/L for men and <1.2 mmol/L for women) [36]. Atherogenic dyslipidemia was defined as high TG levels, low HDL cholesterol levels and an increase in LDL [37]. BMI was categorized into four groups according to the conventional WHO classification [38]: underweight (<18.5 kg/m2), normal weight (18.5–24.9 kg/m2), overweight (25–29.9 kg/m2), and obese (≥30 kg/m2). Metabolic syndrome (MetS) was defined according to the National Cholesterol Education Program (NCEP) Adult Treatment Panel III (ATP III), without waist circumference [39]; BP over $\frac{130}{85}$ mmHg, TG >1.7 mmol/L, HDL-C levels less than 1.03 mmol/L (men) or 1.29 mmol/L (women) and fasting blood glucose over 5.5 mmol/L. Blood pressure was defined as mean SBP ≥140 mmHg and/or a mean DBP ≥ 90mmHg or previously diagnosed hypertension or patient on blood pressure lowering drugs.
## Statistical Analysis
Entry and analysis of data were done using Microsoft Excel 2016 and SPSS version 25.0. Categorical variables were presented as frequencies with percentages while continuous variables were presented as means with standard deviations or medians with interquartile ranges after checking for normality. Chi-square analysis was used to determine the association of sociodemographic characteristics such as age categories, gender, marital status, occupation and level of physical activity with MetS among the type 2 diabetes mellitus patients. Parametric data were analyzed using independent t-test whereas nonparametric data were analyzed using Mann–Whitney U-test. To assess the strength of the association between continuous variables, partial Spearman’s correlation coefficient was used. Multivariate logistic regression analysis was conducted to compare the predictive capacities of the anthropometric indices for cardiometabolic risk among the participants. P-values less than 0.05 were considered statistically significant for all analyses.
## Socio-Demographic Characteristics of the Study Population
Table 1 shows the socio-demographic characteristics of Type 2 diabetes patients with and without metabolic syndrome. A total of 241 type 2 diabetes mellitus patients were recruited into the study of which 138 ($57.3\%$) had no metabolic syndrome (MetS) and 103 ($42.7\%$) had MetS.
**Table 1**
| Variables | Total(n = 241) | T2DM | T2DM.1 | T2DM.2 | T2DM.3 | Unnamed: 6 | Unnamed: 7 |
| --- | --- | --- | --- | --- | --- | --- | --- |
| | | Without MetS (n = 138) | Without MetS (n = 138) | With MetS (n = 103) | With MetS (n = 103) | p-value | p-value |
| Age (years) | | | | | | 0.625 | 0.625 |
| Median (IQR) | 58.00 (50.00–65.00) | 57.50 (50.00–66.00) | 57.50 (50.00–66.00) | 58.00 (49.00–63.00) | 58.00 (49.00–63.00) | | |
| Age Categories n (%) | | | | | | 0.068 | 0.068 |
| 30–49 | 59 (24.5) | 33 (23.9) | 33 (23.9) | 26 (25.2) | 26 (25.2) | | |
| 50–59 | 77 (32.0) | 47 (34.1) | 47 (34.1) | 30 (29.1) | 30 (29.1) | | |
| 60–69 | 74 (30.7) | 35 (25.4) | 35 (25.4) | 39 (37.9) | 39 (37.9) | | |
| 70–79 | 31 (12.9) | 23 (16.7) | 23 (16.7) | 8 (7.8) | 8 (7.8) | | |
| Sex n (%) | | | | | | <0.0001 | <0.0001 |
| Male | 99 (41.1) | 71 (51.4) | 71 (51.4) | 28 (27.2) | 28 (27.2) | | |
| Female | 142 (58.9) | 67 (48.6) | 67 (48.6) | 75 (72.8) | 75 (72.8) | | |
| Marital status n (%) | | | | | | 0.018 | 0.018 |
| Single | 4 (1.7) | 3 (2.2) | 3 (2.2) | 1 (1.0) | 1 (1.0) | | |
| Married | 164 (68.0) | 102 (73.9) | 102 (73.9) | 62 (60.2) | 62 (60.2) | | |
| Divorced | 18 (7.5) | 4 (2.9) | 4 (2.9) | 14 (13.6) | 14 (13.6) | | |
| Separated | 7 (2.9) | 3 (2.2) | 3 (2.2) | 4 (3.9) | 4 (3.9) | | |
| Widowed | 48 (19.9) | 26 (18.8) | 26 (18.8) | 22 (21.4) | 22 (21.4) | | |
| Educational level n (%) | | | | | | 0.765 | 0.765 |
| Tertiary | 36 (14.9) | 24 (17.4) | 24 (17.4) | 12 (11.7) | 12 (11.7) | | |
| Senior High School | 57 (23.7) | 33 (23.9) | 33 (23.9) | 24 (23.3) | 24 (23.3) | | |
| Junior High School | 78(32.4) | 42 (30.4) | 42 (30.4) | 36 (35.0) | 36 (35.0) | | |
| Lower Primary School | 28(11.6) | 15 (10.9) | 15 (10.9) | 13 (12.6) | 13 (12.6) | | |
| No former education | 42 (17.4) | 24 (17.4) | 24 (17.4) | 18 (17.5) | 18 (17.5) | | |
| Occupation n (%) | | | | | | 0.970 | 0.970 |
| Student | 1 (0.4) | 0 (0.0) | 0 (0.0) | 1 (1.0) | 1 (1.0) | | |
| Retired | 32 (13.3) | 19 (13.8) | 19 (13.8) | 13 (12.6) | 13 (12.6) | | |
| Keeping House | 23 (9.5) | 14 (10.1) | 14 (10.1) | 9 (8.7) | 9 (8.7) | | |
| Employed | 152 (63.1) | 87 (63.0) | 87 (63.0) | 65 (63.1) | 65 (63.1) | | |
| Unemployed | 31 (12.9) | 17 (12.3) | 17 (12.3) | 14 (13.6) | 14 (13.6) | | |
| Other | 2 (0.8) | 1 (0.7) | 1 (0.7) | 1 (1.0) | 1 (1.0) | | |
| Physical activity n (%) | | | | | | 0.915 | 0.915 |
| Primary sedentary | 59 (24.5) | 32 (23.2) | 32 (23.2) | 27 (26.2) | 27 (26.2) | | |
| Sedentary with frequent activity | 97 (40.2) | 56 (40.6) | 56 (40.6) | 41 (39.8) | 41 (39.8) | | |
| Primary physical | 79 (32.8) | 47 (34.1) | 47 (34.1) | 32 (31.1) | 32 (31.1) | | |
| Physical with high intensity activity | 6 (2.5) | 3 (2.2) | 3 (2.2) | 3 (2.9) | 3 (2.9) | | |
| Family history of T2DM n (%) | Family history of T2DM n (%) | | | | | 0.006 | 0.006 |
| Yes | 184 (76.7) | 99 (72.3) | 99 (72.3) | 85 (82.5) | 85 (82.5) | | |
| No | 56 (23.3) | 38 (27.7) | 38 (27.7) | 18 (17.5) | 18 (17.5) | | |
| Smoking n (%) | | | | | | 0.223 | 0.223 |
| Yes | 33 (13.8) | 22 (16.2) | 22 (16.2) | 11 (10.7) | 11 (10.7) | | |
| No | 206 (85.2) | 114 (83.8) | 114 (83.8) | 92 (89.3) | 92 (89.3) | | |
| Alcohol intake n (%) | | | | | | 0.991 | 0.991 |
| Yes | 102 (42.7) | 58 (42.6) | 58 (42.6) | 44 (42.7) | 44 (42.7) | | |
| No | 137 (57.3) | 78 (57.4) | 78 (57.4) | 59 (57.3) | 59 (57.3) | | |
The median age of the total participants was 58 years and statistically, there was no significant difference between the median ages of participants without MetS and those with MetS [57.50 versus 58.00, $$p \leq 0.625$$]. Majority of the participants were in the age categories 50–59 ($32.0\%$). Age categories was however not significantly associated with MetS status of participants ($$p \leq 0.068$$). The male:female ratio of the overall participants was 1:1.43. Of the 103 participants with MetS, $72.8\%$ were females and $28.2\%$ were males. Gender was significantly associated with MetS status ($p \leq 0.0001$). The highest proportion of participants was married ($68.0\%$). Marital status of participants was significantly associated with their MetS status ($$p \leq 0.018$$). Furthermore, higher proportion of the participants had completed junior high school ($32.4\%$), were employed ($63.1\%$), practice sedentary lifestyle with frequent exercise ($40.2\%$), had family history of T2DM ($76.7\%$), were non-smokers ($85.2\%$) and non-alcoholic beverage drinkers ($57.3\%$). On the contrary, participants’ educational level, occupation, physical activity status, family history of T2DM, smoking status and alcohol intake status were not proportionally significantly different in terms of participants with and without MetS status ($p \leq 0.05$).
## Clinical, Anthropometric, and Lipid Profile Variables of the Study Population
Table 2 shows the clinical, anthropometric and lipid profile variables of the study population. Participants with MetS had significantly higher median levels of SBP (148.00 mmHg versus 132.00 mmHg, $p \leq 0.0001$) and DBP (88.00 mmHg versus 78.00 mmHg, $p \leq 0.0001$) compared to the participants without MetS. Levels of FBG ($$p \leq 0.067$$) and HbA1C ($$p \leq 0.158$$) were not significantly different between the two groups. Also, except for height which was significantly taller among participants without MetS than that observed for participants with MetS [1.66 m versus 1.64 m, $$p \leq 0.025$$], all the other anthropometric indices, namely, Weight, BMI, CI, BAI, ABSI, BRI, WC, HC, and WHtR were significantly higher among participants with MetS compared to those without MetS ($p \leq 0.05$). Participants with MetS had significantly higher median concentrations of TG [1.37 mmol/L versus 1.05 mmol/L, $p \leq 0.0001$], Coronary risk [5.40 versus 4.68, $p \leq 0.0001$], and VLDL [0.62 mmol/L versus 0.48 mmol/L, $p \leq 0.0001$] than that observed among participants without MetS. Conversely, HDL-C concentration was significantly lower among participants with MetS compared to participants without MetS [1.20 mmol/L versus 1.32 mmol/L, $p \leq 0.0001$]. The median concentrations of TC and LDL-C measures were however not statistically significantly different between T2DM with and without MetS status of participants ($$p \leq 0.347$$ and $$p \leq 0.252$$ respectively).
**Table 2**
| Variable | Total (n = 241) | T2DM | T2DM.1 | Unnamed: 4 |
| --- | --- | --- | --- | --- |
| | | Without MetS (n = 138) | With MetS (n = 103) | p-value |
| Clinical | | | | |
| SBP (mmHg) | 136.00 (121.00–153.00) | 132.00 (119.00–144.50) | 148.00 (130.00–162.00) | <0.0001 |
| DBP (mmHg) | 81.00 (72.00–90.00) | 78.00 (70.00–84.00) | 88.00 (77.00–97.00) | <0.0001 |
| FBS (mmol/L) | 7.90 (6.30–11.40) | 7.60 (5.60–11.70) | 8.20 (6.98–11.00) | 0.067 |
| HbA1C (%) | 8.00 (6.60–9.60) | 7.80 (6.40–9.45) | 8.15 (7.00–10.03) | 0.158 |
| Anthropometrics | | | | |
| Height (m) | 1.65 (1.60–1.70) | 1.66 (1.62–1.71) | 1.64 (1.58–1.69) | 0.025 |
| Weight (kg) | 68.65 (60.95–80.55) | 66.78 (58.98–76.36) | 75.45 (65.35–84.000 | <0.0001 |
| BMI (kg/m2) | 25.51 (22.57–29.30) | 24.07 (21.62–27.90) | 27.72 (23.74–30.86) | <0.0001 |
| CI (m3/2/kg1/2) | 1.31 (1.25–1.37) | 1.29 (1.21–1.35) | 1.34 (1.29–1.40) | <0.0001 |
| BAI (%) | 29.14 (24.22–33.60) | 26.11 (22.61–31.33) | 31.66 (27.83–35.86) | <0.0001 |
| ABSI (m11/6kg−2/3) | 0.083 (0.079–0.088) | 0.082 (0.078–0.086) | 0.084 (0.080–0.089) | 0.001 |
| BRI | 4.70 (3.58–5.83) | 3.87 (3.14–4.92) | 5.18 (4.65–6.46) | <0.0001 |
| WC (cm) | 92.77 ± 13.05 | 88.10 ± 13.56 | 99.02 ± 9.20 | <0.0001 |
| HC (cm) | 99.41 ± 13.60 | 94.83 ± 13.97 | 105.55 ± 10.30 | <0.0001 |
| WHR | 0.94 ± 0.06 | 0.93 ± 0.07 | 0.94 ± 0.06 | 0.237 |
| WHtR | 0.56 ± 0.08 | 0.53 ± 0.09 | 0.60 ± 0.07 | <0.0001 |
| Lipid profile | | | | |
| TG (mmol/L) | 1.13 (0.89–1.51) | 1.05 (0.84–1.35) | 1.37 (0.95–1.79) | <0.0001 |
| TC (mmol/L) | 4.80 (3.77–5.50) | 4.70 (3.80–5.40) | 4.90 (93.70–5.65) | 0.347 |
| HDL-C (mmol/L) | 1.30 (1.10–1.50) | 1.32 (1.20–1.60) | 1.20 (1.10–1.40) | <0.0001 |
| LDL-C (mmol/L) | 2.81 (1.93–3.61) | 2.68 (1.94–3.43) | 2.97 (1.93–3.78) | 0.252 |
| Coronary Risk | 4.88 (3.83–5.96) | 4.68 (3.58–5.53) | 5.40 (4.21–6.49) | <0.0001 |
| VLDL-C (mmol/L) | 0.52 (0.41–0.68) | 0.48 (0.38–0.62) | 0.62 (0.44–0.82) | <0.0001 |
## Prevalence of Cardiometabolic Risk Factors Among the Study Population Stratified by Male and Female
Figure 1 shows the prevalence of cardiometabolic risk factors among the study population stratified by male and female. Of the 241 subjects, 16 had dyslipidemia, 103 had metabolic syndrome and 87 were hypertensive representing a prevalence of 6.6, 42.7, and $36.1\%$, respectively. Stratifying by gender, a proportion of $6.3\%$ ($\frac{9}{142}$) of female participants had dyslipidemia, $52.8\%$ ($\frac{75}{142}$) had MetS, and $35.2\%$ ($\frac{50}{142}$) had hypertension. Of the 99 males, $7.1\%$ had dyslipidemia, $28.3\%$ had metabolic syndrome, and $37.4\%$ were hypertensive. There was a statistically significant difference in the proportions between male and females in terms of their MetS status ($$p \leq 0.0002$$). On the contrary, there was no significant difference in the proportions between males and females in relation to dyslipidemia ($$p \leq 0.8085$$) and hypertension ($$p \leq 0.7309$$) status of the participants.
**Figure 1:** *Prevalence of cardiometabolic risk factors among the study population stratified by Male and Female.*
## Anthropometric Indices, Sociodemographic, Clinical indices and Lipid Measures Associated With MetS Among T2DM Patients
Table 3 shows the odds ratios of anthropometric indices, sociodemographic, clinical indices, and lipid profile measures associated with MetS. After adjusting for possible confounders in multivariate logistic regression, BRI quartiles—Q3[a OR = 25.15, $95\%$CI (2.02–313.81), $$p \leq 0.012$$], Q4 [aOR = 39.00, $95\%$CI (2.68–568.49), $$p \leq 0.007$$], being a female [aOR = 3.02, $95\%$CI (1.59–5.76), $$p \leq 0.001$$] and divorced [aOR = 4.05, $95\%$CI (1.22–13.43), $$p \leq 0.022$$], DBP [aOR = 1.07, $95\%$CI (1.03–1.10), $p \leq 0.0001$] and HDL-C [aOR = 0.10, $95\%$CI (0.03–0.35), $p \leq 0.0001$] were the independent predictors of MetS among T2DM.
**Table 3**
| Variables | cOR (95%CI) | p-value | aOR (95%CI)* | p-value.1 |
| --- | --- | --- | --- | --- |
| ABSI quartiles | | | | |
| Q1 | Ref (1) | | Ref (1) | |
| Q2 | 1.98 (0.93–4.20) | 0.078 | 1.2 (0.41–3.60) | 0.734 |
| Q3 | 1.12 (0.54–2.47) | 0.707 | 0.67 (0.22–2.03) | 0.482 |
| Q4 | 3.56 (1.61–7.90) | 0.002 | 1.80 (0.59-5.51) | 0.304 |
| BRI quartiles | | | | |
| Q1 | Ref (1) | | Ref (1) | |
| Q2 | 5.15 (1.9–13.96) | 0.001 | 8.04 (0.71–90.90) | 0.092 |
| Q3 | 14.18 (5.23–38.46) | <0.0001 | 25.15 (2.02–313.81) | 0.012 |
| Q4 | 18.78 (6.82–51.71) | <0.0001 | 39.00 (2.68–568.49) | 0.007 |
| BAI quartiles | | | | |
| Q1 | Ref (1) | | Ref (1) | |
| Q2 | 2.85 (1.21–6.76) | 0.017 | 0.50 (0.13–1.91) | 0.31 |
| Q3 | 7.30 (3.09–17.28) | <0.0001 | 0.65 (0.14–2.95) | 0.574 |
| Q4 | 7.86 (3.31–18.63) | <0.0001 | 0.33 (0.05–2.16) | 0.25 |
| BMI categories | | | | |
| Underweight | Ref (1) | | – | – |
| Normal weight | 1.37 (0.26–7.17) | 0.708 | – | – |
| Overweight | 3.32 (0.63–17.50) | 0.156 | – | – |
| Obese | 5.25 (0.95–29.147) | 0.058 | – | – |
| CI status | | | | |
| Normal | Ref (1) | | Ref (1) | |
| High risk | 8.33 (2.45–28.40) | 0.001 | 3.48 (0.59–20.60) | 0.17 |
| WHtR status | | | | |
| Normal | Ref (1) | | Ref (1) | |
| High risk | 10.38 (3.94–27.33) | <0.0001 | 0.60 (0.04–9.40) | 0.717 |
| WHR status | | | | |
| Normal | Ref (1) | | Ref (1) | |
| Overweight | 2.41 (0.98–5.93) | 0.055 | 1.02 (0.33–3.20) | 0.971 |
| Obese | 3.80 (1.89–7.62) | <0.0001 | 0.54 (0.10–2.83) | 0.462 |
| Sex | | | | |
| Male | Ref (1) | | Ref (1) | |
| Female | 2.84 (1.64–4.91) | <0.0001 | 3.02 (1.59–5.76) | 0.001 |
| Marital status | | | | |
| Single | Ref (1) | | Ref (1) | |
| Married | 0.55 (0.06–5.39) | 0.606 | 0.41 (0.04–4.28) | 0.457 |
| Divorced | 5.76 (1.81–18.28) | 0.003 | 4.05 (1.22–13.43) | 0.022 |
| Separated | 2.19 (0.48–10.13) | 0.314 | 1.42 (0.28–7.23) | 0.676 |
| Widowed | 1.39 (0.73–2.67) | 0.318 | 0.68 (0.31–1.53) | 0.353 |
| SBP (mmHg) | 1.03 (1.01–1.04) | <0.0001 | 1.01 (0.99–1.03) | 0.306 |
| DBP (mmHg) | 1.07 (1.04–1.09) | <0.0001 | 1.07 (1.03–1.10) | <0.0001 |
| TG (mmol/L) | 2.89 (1.69–4.93) | <0.0001 | 0.00 (0.00–Inf) | 0.353 |
| TC (mmol/L) | 1.13 (0.92–1.38) | 0.243 | – | – |
| HDL-C (mmol/L) | 0.18 (0.07–0.45) | <0.0001 | 0.10 (0.03–0.35) | <0.0001 |
| LDL-C (mmol/L) | 1.17 (0.94–1.47) | 0.159 | – | – |
| Coronary Risk | 1.40 (1.17–1.69) | <0.0001 | 1.10 (0.86–1.40) | 0.455 |
| VLDL (mmol/L) | 10.86 (3.32–35.53) | <0.0001 | >100 (0.00–Inf) | 0.324 |
## Partial Spearman Correlation Coefficients of Anthropometric Indices With Hemodynamic and Lipid Markers Among T2DM Patients
Table 4 illustrates the partial coefficients of Spearman correlation of the anthropometric indices (ABSI, BRI, BMI, WC, BAI, CI, WHtR, and WHR) with hemodynamic indices and the lipid markers among the T2DM patients. After controlling for age and gender, the new indices—ABSI and BRI were correlated moderately ($r = 0.406$, $p \leq 0.0001$). The BRI had a strong positive correlation with WHtR ($r = 0.992$, $p \leq 0.0001$), WC ($r = 0.940$, $p \leq 0.0001$), BAI ($r = 0.895$, $p \leq 0.0001$), BMI ($r = 0.709$, $p \leq 0.0001$), a moderate and weak correlation with CI ($r = 0.646$, $p \leq 0.0001$), and WHR ($r = 0.218$, $p \leq 0.0001$) respectively but showed a negative correlation with height (r = −0.273, $p \leq 0.0001$). However, ABSI only showed a strong positive relationship with CI ($r = 0.957$, $p \leq 0.0001$) but moderate correlation with WC ($r = 0.499$, $p \leq 0.0001$), WHR ($r = 0.461$, $p \leq 0.0001$), and BAI ($r = 0.355$, $p \leq 0.0001$). For BMI (r = −0.313, $p \leq 0.0001$), ABSI showed a moderate negative correlation. ABSI was not associated with WHR and Height of participants. Moreover, BRI showed a slight but significant positive correlation with blood pressure (DBP) and two lipid markers—TG and VLDL-C. This was however not true for ABSI as it was neither associated with blood pressure nor any of the lipid markers.
**Table 4**
| Unnamed: 0 | ABSI | BRI | BMI | WC | BAI | CI | WHtR | WHR |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| ABSI | 1 | 0.406** | −0.313** | 0.499** | 0.355** | 0.957** | 0.461** | 0.123 |
| BRI | 0.406** | 1 | 0.709** | 0.940** | 0.895** | 0.646** | 0.992** | 0.218** |
| BMI | −0.313** | 0.709** | 1 | 0.646** | 0.640** | −0.030 | 0.677** | 0.122 |
| WC | 0.499** | 0.940** | 0.646** | 1 | 0.824** | 0.725** | 0.958** | 0.197** |
| BAI | 0.355** | 0.895** | 0.640** | 0.824** | 1 | 0.573** | 0.908** | −0.207** |
| CI | 0.957** | 0.646** | −0.030 | 0.725** | 0.573** | 1 | 0.694** | 0.162* |
| WHtR | 0.461** | 0.992** | 0.677** | 0.958** | 0.908** | 0.694** | 1 | 0.188** |
| WHR | 0.123 | 0.218** | 0.122 | 0.197** | −0.207** | 0.162* | 0.188** | 1 |
| Height | 0.073 | −0.273** | −0.175* | 0.035 | −0.374** | 0.022 | −0.251** | −0.004 |
| SBP | −0.031 | 0.101 | 0.128 | 0.105 | 0.063 | 0.013 | 0.104 | 0.105 |
| DBP | 0.012 | 0.190** | 0.216** | 0.216** | 0.135* | 0.081 | 0.204** | 0.134 |
| FBS | −0.041 | −0.268** | −0.265** | −0.277** | −0.222** | −0.123 | −0.265** | −0.083 |
| HbA1C | −0.107 | −0.259** | −0.181** | −0.245** | −0.223** | −0.169* | −0.253** | −0.079 |
| TG | 0.144 | 0.199** | 0.070 | 0.151* | 0.130 | 0.171* | 0.190** | 0.189** |
| TC | 0.055 | −0.054 | −0.092 | −0.054 | −0.064 | 0.023 | −0.054 | 0.041 |
| HDL-C | −0.024 | −0.270** | −0.282** | −0.275** | −0.225** | −0.110 | −0.266** | −0.103 |
| LDL-C | 0.039 | −0.034 | −0.047 | −0.021 | −0.040 | 0.020 | −0.032 | 0.027 |
| CR | 0.060 | 0.092 | 0.060 | 0.089 | 0.057 | 0.076 | 0.087 | 0.094 |
| VLDL-C | 0.145 | 0.199** | 0.069 | 0.151* | 0.130 | 0.172* | 0.190** | 0.189** |
## Discussion
Association between type 2 diabetes mellitus (T2DM) and cardiometabolic syndrome have extensively been explored. However, for the first time, this study evaluated the prevalence of cardiometabolic syndrome and its association with two new anthropometric indices among T2DM patients in two selected hospitals in the Ashanti Region of Ghana.
The present study found $42.7\%$ of the T2DM patients with MetS and the prevalence was higher in females than in male participants (Figure 1). Similarly, a cross-sectional study by Yadav et al. [ 40] among Indian type 2 diabetes patients reported MetS prevalence of $57.7\%$ with females having a higher prevalence than males. The slightly higher prevalence of the previous study could be partly due to low sample size in the present study and genetic differences between Caucasians and Blacks. This study registered more females ($58.9\%$) as compared to males ($41.1\%$). In this present study, there was a significant association between sex of participants and their MetS status ($p \leq 0.0001$). Furthermore, being a female was significantly associated with increased odds of having MetS as compared to being a male. This finding is consistent with previous cross-sectional studies which also reported that female type 2 diabetes patients are at higher risk of having MetS when compared to males with type 2 diabetes mellitus (T2DM) [41, 42]. Less exercise, increased body weight, and an increased risk of dyslipidemia in women could be the possible reason for the higher odds of females having MetS than males [41]. In this study, marital status of participants was significantly associated with MetS status. Being divorced was associated with significant 5-times increased odds of having MetS compared to being single even after possible covariates were controlled. Chung and colleagues reported a similar finding in a cross-sectional study conducted among Korean adults [43]. The driving factor for this finding is however not well understood. Probable explanations could be lack of social support and living alone after being divorced which could compound the risk of having MetS [44].
Controversies still exist as to which anthropometric index best predict cardiometabolic risk among T2DM. BMI is the most widely used obesity marker and has been associated with cardiometabolic risk and type 2 diabetes [45]. In this present study, the median Body Mass Index (BMI), Waist-to-Hip Ratio (WHR), Waist-to-Height Ratio (WHtR), Body Adiposity Index (BAI), and Conicity Index (CI) were significantly higher among T2DM patients who had MetS as compared to those without MetS. However, none of these anthropometric were independent predictors of MetS after multivariate logistic regression. BMI is reported to be a poor indicator of cardiometabolic syndrome compared to the other obesity indices (46–48). BMI is unable to distinguish between fat and muscle mass and also between fat compartments such as visceral adipose tissue and subcutaneous adipose tissue, which are closely linked to cardiometabolic syndrome [49]. Additionally, the possible explanation for failure of BMI, BAI, CI, WC, WHR, and WHtR independently predict MetS could be due to its weaker correlation with cardiometabolic risk factors as compared to the other obesity indices as observed in this study.
In this current study, the two new indices [A Body Shape Index (ABSI) and Body Roundness Index (BRI)] were included to the traditional anthropometric indices in quest to compare their predictive capabilities for cardiometabolic risk among type 2 diabetes patients in a Ghanaian population. This current study found that ABSI could not independently predict MetS among the T2DM when it was compared to the other anthropometric indices in an adjusted multivariate logistic model. This finding is in consonant with previous studies conducted among the Caucasians (50–52). Maessen et al. [ 52] found that the ABSI was ineffective in identifying cardiometabolic risk factors among Netherland population. Similarly, a study conducted by Li et al. [ 51] showed that ABSI failed to significantly predict MetS and T2DM among overweight and obese adults The poor correlation between ABSI and MetS, on the other hand, is debatable in that previous studies have linked ABSI to some cardiometabolic risk factors [26, 53]. Disparities in anthropometric measures and races may have a significant impact on the predictive value of ABSI. Although the ABSI formula was adjusted for BMI, obesity status differs for Africans, Europeans and Asians since there is differences in WC across these races [25]. Furthermore, Asians are much shorter than Africans, Americans and Europeans, which may confound the predictive value of ABSI. Ethnicity has been a significant moderator in the relationship between these obesity indices and cardiometabolic risk, and it holds true for both genders. In a meta-analysis conducted by Rico-Martín et al. [ 54], the AUCs for all anthropometric indices in the non-Chinese population were better predictors of MetS than they were for the Chinese population. A probable reason for the failure of ABSI to superiorly predict MetS is that, it was originally formulated as a risk assessment tool to predict mortality risk in a follow-up study [25]. However, we employed it in a cross-sectional study to predict MetS among T2DM patients. There is a possibility that this resulted in the failure of the ABSI to show significant predictive power compared with the other indices. Also, ABSI was moderately correlated with WC in the current study but showed negative correlation with BMI from the partial Spearman’s correlation coefficients. A negative correlation with BMI suggests an inverse relationship between the two (ABSI increases with decreasing BMI). Despite limitations of BMI and WC, increasing measures of these indices are widely standard markers to predict MetS and other cardiometabolic risk (55–58). The moderate and negative correlation of ABSI with WC and BMI respectively may have accounted for its failure to predict MetS among the T2DM. Furthermore, ABSI showed no significant correlation with the hemodynamic indices and the lipid markers from the partial Spearman’s correlation test. These markers are known predictors of MetS [59, 60] and this could be among the reasons why ABSI was ineffective to predict MetS among the subjects.
The strength of this study is that BRI but not ABSI and other anthropometric indices (BMI, WHR, WHtR, BAI, and CI) was an independent predictor of cardiometabolic risk among the type 2 diabetes mellitus patients after controlling for age, gender, and marital status of participants. BRI estimates the human as an elliptical figure and improved body fat% and Visceral Adipose Tissue (VAT)% as compared to the traditional anthropometric indices. VAT% and MetS have a well-established association [61]. In the present study, BRI was the only independent predictor of MetS. This implies that only BRI was significantly associated with the higher odds of having MetS and is superior to the traditional indices in predicting MetS among the subjects. In keeping with our results, several studies have similarly reported the superior power of BRI over the traditional anthropometric indices in predicting MetS [51, 62, 63]. In a meta-analysis of data pooled from more than one fifty thousand people, increased BRI odds was significantly associated with increased risk of having MetS [54]. Additionally, a study conducted among obese and overweight Chinese adults found that BRI was a better predictor of MetS and T2DM [51].
The current study found that BRI had a strong positive correlation with WC and also with BMI regardless of the exclusion of BMI in the BRI formulation but a negative correlation with height after age and gender were controlled. This finding is consistent with a study by Li et al. [ 51]. In other words, for a constant WC, height decreases whereas BMI increases and the body assumes an elliptical shape. Elliptical shape and increased odds of cardiometabolic risks has been well established [64]. Also, BMI and WC have been globally accepted as a tool for predicting MetS despite some shortcomings (55–58). This may have accounted for superiority of BRI in predicting MetS owing to its strong correlation with these two markers after possible cofounders were controlled. Furthermore, partial Spearman’s correlation coefficients showed that BRI was marginal but significantly associated with blood pressure (increasing DBP) and the lipid markers (TG and VLDL-C). Several studies have linked increasing blood pressure and lipid markers to a significant increased likelihood of having MetS among T2DM [59, 60]. This could possibly be another reason why BRI showed a superior predictive capacity for MetS among the subjects over the other indices.
Despite the novel findings, this study had some limitations that are worth mentioning for consideration by future studies. First of all, the female participants outnumbered the males which could have introduced a bias in the prevalence of cardiometabolic syndrome. Also, all the participants were aged (30 and above), and we could not validate that the optimal anthropometric index (BRI) would be superior to the other indices in other age groups. Sample size was also small [241] which could have introduced a bias in our analysis.
## Conclusions
The prevalence of cardiometabolic syndrome was $42.7\%$ among the T2DM patients. Cardiometabolic syndrome was influenced by female gender, being divorced, and increased body roundness index (BRI). Integration of BRI as part of routine assessment could be used as early indicator of cardiometabolic syndrome among T2DM patients.
Further studies can be done with a larger population to establish the relationship between these two new but simple anthropometric indices and MetS among type 2 diabetes patients.
## Data Availability Statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics Statement
The studies involving human participants were reviewed and approved by The Committee on Human Research, Publication and Ethics, Kwame Nkrumah University of Science and Technology. The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
Conceptualization EOA; methodology, JF, and EOA; formal analysis, JF, EOA, and SO; investigation, JF, VCKTT, CH, and BA; original draft preparation, EOA, and JF supervision, EOA. All authors listed reviewed, edited have made a substantial, direct, and intellectual contribution to the work and approved it for publication.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Danaei G, Finucane MM, Lu Y, Singh GM, Cowan MJ, Paciorek CJ. **National, Regional, and Global Trends in Fasting Plasma Glucose and Diabetes Prevalence Since 1980: Systematic Analysis of Health Examination Surveys and Epidemiological Studies With 370 Country-Years and 2· 7 Million Participants**. *Lancet* (2011) **378** 31-40. DOI: 10.1016/S0140-6736(11)60679-X
2. Saeedi P, Petersohn I, Salpea P, Malanda B, Karuranga S, Unwin N. **Global and Regional Diabetes Prevalence Estimates for 2019 and Projections for 2030 and 2045: Results From the International Diabetes Federation Diabetes Atlas, 9(Th) Edition**. *Diabetes Res Clin Pract* (2019) **157** 107843. PMID: 31518657
3. Mumu SJ, Saleh F, Ara F, Haque MR, Ali L. **Awareness Regarding Risk Factors of Type 2 Diabetes Among Individuals Attending a Tertiary-Care Hospital in Bangladesh: A Cross-Sectional Study**. *BMC Res Notes* (2014) **7** 599. DOI: 10.1016/j.diabres.2019.107843
4. 4
IDF. IDF Africa Atlas. 8th edn. Brussels, Belgium: International Diabetes Federation (2017) p. 1–150. doi: 10.1186/1756-0500-7-599. *IDF Africa Atlas* (2017) 1-150. DOI: 10.1186/1756-0500-7-599
5. 5
International Diabetes Federation. Diabetes in Africa (2019). Available at: https://idf.org/our-network/regions-members/africa/diabetes-in-africa.html.. *Diabetes in Africa* (2019)
6. Danquah I, Bedu-Addo G, Terpe KJ, Micah F, Amoako YA, Awuku YA. **Diabetes Mellitus Type 2 in Urban Ghana: Characteristics and Associated Factors**. *BMC Public Health* (2012) **12** 210. DOI: 10.1186/1471-2458-12-210
7. Adua E, Roberts P, Wang W. **Incorporation of Suboptimal Health Status as a Potential Risk Assessment for Type II Diabetes Mellitus: A Case-Control Study in a Ghanaian Population**. *Epma J* (2017) **8**. DOI: 10.1007/s13167-017-0119-1
8. Obirikorang DN, Osakunor EO, Anto SO, Amponsah OK. **Obesity and Cardio-Metabolic Risk Factors in an Urban and Rural Population in the Ashanti Region-Ghana: A Comparative Cross-Sectional Study**. *PloS One* (2015) **10** e0129494. DOI: 10.1371/journal.pone.0129494
9. Nichols GA, Horberg M, Koebnick C, Young DR, Waitzfelder B, Sherwood NE. **Cardiometabolic Risk Factors Among 1.3 Million Adults With Overweight or Obesity, But Not Diabetes, in 10 Geographically Diverse Regions of the United States, 2012-2013**. *Prev Chronic Dis* (2017) **14** E22. DOI: 10.5888/pcd14.160438
10. Kumanyika SK, Obarzanek E, Stettler N, Bell R, Field AE, Fortmann SP. **Population-Based Prevention of Obesity: The Need for Comprehensive Promotion of Healthful Eating, Physical Activity, and Energy Balance: A Scientific Statement From American Heart Association Council on Epidemiology and Prevention, Interdisciplinary Committee for Prevention (Formerly the Expert Panel on Population and Prevention Science)**. *Circulation* (2008) **118**. DOI: 10.1161/CIRCULATIONAHA.108.189702
11. Qiao Q, Nyamdorj R. **Is the Association of Type II Diabetes With Waist Circumference or Waist-to-Hip Ratio Stronger Than That With Body Mass Index**. *Eur J Clin Nutr* (2010) **64**. DOI: 10.1038/ejcn.2009.93
12. Aguilar-Morales I, Colin-Ramirez E, Rivera-Mancía S, Vallejo M, Vázquez-Antona C. **Performance of Waist-To-Height Ratio, Waist Circumference, and Body Mass Index in Discriminating Cardio-Metabolic Risk Factors in a Sample of School-Aged Mexican Children**. *Nutrients* (2018) **10** 1-14. DOI: 10.3390/nu10121850
13. Millar SR, Perry IJ, Phillips CM. **Assessing Cardiometabolic Risk in Middle-Aged Adults Using Body Mass Index and Waist-Height Ratio: Are Two Indices Better Than One? A Cross-Sectional Study**. *Diabetol Metab Syndr* (2015) **7** 73. DOI: 10.1186/s13098-015-0069-5
14. Myint PK, Kwok CS, Luben RN, Wareham NJ, Khaw KT. **Body Fat Percentage, Body Mass Index and Waist-to-Hip Ratio as Predictors of Mortality and Cardiovascular Disease**. *Heart* (2014) **100**. DOI: 10.1136/heartjnl-2014-305816
15. Jenkins DA, Bowden J, Robinson HA, Sattar N, Loos RJF, Rutter MK. **Adiposity-Mortality Relationships in Type 2 Diabetes, Coronary Heart Disease, and Cancer Subgroups in the UK Biobank, and Their Modification by Smoking**. *Diabetes Care* (2018) **41**. DOI: 10.2337/dc17-2508
16. Obirikorang Y, Obirikorang E, Acheampong EO, Anto E, Toboh EA. **Association of Wrist Circumference and Waist-To-Height Ratio With Cardiometabolic Risk Factors Among Type II Diabetes Patients in a Ghanaian Population**. *J Diabetes Res* (2018) **2018** 1838162. DOI: 10.1155/2018/1838162
17. Zerga AA, Bezabih AM, Adhanu AK, Tadesse SE. **Obesity Indices for Identifying Metabolic Syndrome Among Type Two Diabetes Patients Attending Their Follow-Up in Dessie Referral Hospital, North East Ethiopia**. *Diabetes Metab Syndr Obes* (2020) **13**. DOI: 10.2147/DMSO.S242792
18. Katzmarzyk PT, Bray GA, Greenway FL, Johnson WD, Newton RL, Ravussin E. **Ethnic-Specific BMI and Waist Circumference Thresholds**. *Obes (Silver Spring)* (2011) **19**. DOI: 10.1038/oby.2010.319
19. Lear SA, James PT, Ko GT, Kumanyika S. **Appropriateness of Waist Circumference and Waist-to-Hip Ratio Cutoffs for Different Ethnic Groups**. *Eur J Clin Nutr* (2010) **64** 42-61. DOI: 10.1038/ejcn.2009.70
20. Bohr AD, Laurson K, McQueen MB. **A Novel Cutoff for the Waist-to-Height Ratio Predicting Metabolic Syndrome in Young American Adults**. *BMC Public Health* (2016) **16** 295. DOI: 10.1186/s12889-016-2964-6
21. Segheto W, Coelho FA, Cristina Guimarães da Silva D, Hallal PC, Marins JC, Ribeiro AQ. **Validity of Body Adiposity Index in Predicting Body Fat in Brazilians Adults**. *Am J Hum Biol* (2017) **29** 1-8. DOI: 10.1002/ajhb.22901
22. Geliebter A, Atalayer D, Flancbaum L, Gibson CD. **Comparison of Body Adiposity Index (BAI) and BMI With Estimations of % Body Fat in Clinically Severe Obese Women**. *Obes (Silver Spring)* (2013) **21**. DOI: 10.1002/oby.20264
23. Motamed N, Perumal D, Zamani F, Ashrafi H, Haghjoo M, Saeedian FS. **Conicity Index and Waist-To-Hip Ratio Are Superior Obesity Indices in Predicting 10-Year Cardiovascular Risk Among Men and Women**. *Clin Cardiol* (2015) **38**. DOI: 10.1002/clc.22437
24. de Quadros TMB, Gordia AP, Andaki ACR, Mendes EL, Mota J, Silva LR. **Utility of Anthropometric Indicators to Screen for Clustered Cardiometabolic Risk Factors in Children and Adolescents**. *J Pediatr Endocrinol Metab* (2019) **32** 49-55. DOI: 10.1515/jpem-2018-0217
25. Krakauer NY, Krakauer JC. **A New Body Shape Index Predicts Mortality Hazard Independently of Body Mass Index**. *PloS One* (2012) **7** e39504. DOI: 10.1371/journal.pone.0039504
26. Thomas C, Bredlau A, Bosy-Westphal M, Mueller W, Shen D. **Relationships Between Body Roundness With Body Fat and Visceral Adipose Tissue Emerging From a New Geometrical Model**. *Obes (Silver Spring)* (2013) **21**. DOI: 10.1002/oby.20408
27. He S, Chen X. **Could the New Body Shape Index Predict the New Onset of Diabetes Mellitus in the Chinese Population**. *PloS One* (2013) **8** e50573. DOI: 10.1371/journal.pone.0050573
28. Krakauer NY, Krakauer JC. **Dynamic Association of Mortality Hazard With Body Shape**. *PloS One* (2014) **9** e88793. DOI: 10.1371/journal.pone.0088793
29. Chang Y, Guo X, Li T, Li S, Guo J, Sun Y. **A Body Shape Index and Body Roundness Index: Two New Body Indices to Identify Left Ventricular Hypertrophy Among Rural Populations in Northeast China**. *Heart Lung Circ* (2016) **25**. DOI: 10.1016/j.hlc.2015.08.009
30. Gu Z, Li D, He H, Wang J, Hu X, Zhang P. **Body Mass Index, Waist Circumference, and Waist-to-Height Ratio for Prediction of Multiple Metabolic Risk Factors in Chinese Elderly Population**. *Sci Rep* (2018) **8** 385. DOI: 10.1038/s41598-017-18854-1
31. Zhang N, Chang Y, Guo X, Chen Y, Ye N, Sun Y. **A Body Shape Index and Body Roundness Index: Two New Body Indices for Detecting Association Between Obesity and Hyperuricemia in Rural Area of China**. *Eur J Intern Med* (2016) **29**. DOI: 10.1016/j.ejim.2016.01.019
32. **Classification and Diagnosis of Diabetes**. *Diabetes Care* (2015) **38**. DOI: 10.2337/dc15-S005
33. Agyemang-Yeboah F, Eghan BAJ, Annani-Akollor ME, Togbe E, Donkor S, Oppong Afranie B. **Evaluation of Metabolic Syndrome and Its Associated Risk Factors in Type 2 Diabetes: A Descriptive Cross-Sectional Study at the Komfo Anokye Teaching Hospital, Kumasi, Ghana**. *BioMed Res Int* (2019) **2019** 4562904. DOI: 10.1155/2019/4562904
34. Gysel C. **Adolphe Quetelet (1796-1874). The Statistics and Biometry of Growth**. *Orthod Fr* (1974) **45**
35. Bergman RN, Stefanovski D, Buchanan TA, Sumner AE, Reynolds JC, Sebring NG. **A Better Index of Body Adiposity**. *Obes (Silver Spring)* (2011) **19**. DOI: 10.1038/oby.2011.38
36. Graham I, Atar D, Borch-Johnsen K, Boysen G, Burell G, Cifkova R. **European Guidelines on Cardiovascular Disease Prevention in Clinical Practice: Executive Summary: Fourth Joint Task Force of the European Society of Cardiology and Other Societies on Cardiovascular Disease Prevention in Clinical Practice (Constituted by Representatives of Nine Societies and by Invited Experts)**. *Eur Heart J* (2007) **28**. DOI: 10.1097/01.hjr.0000277984.31558.c4
37. Semenkovich CF. **Insulin Resistance and Atherosclerosis**. *J Clin Invest* (2006) **116**. DOI: 10.1172/JCI29024
38. EP WHO. **Executive Summary of the Clinical Guidelines on the Identification, Evaluation, and Treatment of Overweight and Obesity in Adults**. *Arch Intern Med* (1998) **158**. DOI: 10.1001/archinte.158.17.1855
39. Grundy SM, Cleeman JI, Daniels SR, Donato KA, Eckel RH, Franklin BA. **Diagnosis and Management of the Metabolic Syndrome: An American Heart Association/National Heart, Lung, and Blood Institute Scientific Statement**. *Circulation* (2005) **112**. DOI: 10.1161/CIRCULATIONAHA.105.169404
40. Yadav D, Mahajan S, Subramanian SK, Bisen PS, Chung CH, Prasad GB. **Prevalence of Metabolic Syndrome in Type 2 Diabetes Mellitus Using NCEP-ATPIII, IDF and WHO Definition and its Agreement in Gwalior Chambal Region of Central India**. *Glob J Health Sci* (2013) **5**. DOI: 10.5539/gjhs.v5n6p142
41. Foroozanfar ZM, Najafipour HP, Khanjani NPM, Bahrampour APM, Ebrahimi HM. **The Prevalence of Metabolic Syndrome According to Different Criteria and its Associated Factors in Type 2 Diabetic Patients in Kerman, Iran**. *Iran J Med Sci* (2015) **40**. DOI: 10.4082/kjfm.2010.31.3.208
42. Rashidi H, Fardad F, Ghaderian B, Shahbazian HB, Latifi M, Karandish M. **Prevalence of Metabolic Syndrome and its Predicting Factors in Type 2 Diabetic Patients in Ahvaz**. *Jundishapur Sci Med J* (2012) **11**
43. Chung TH, Kim MC, Choi CH, Kim CS. **The Association Between Marital Status and Metabolic Syndrome in Korean Men**. *Korean J Family Med* (2010) **31**. DOI: 10.4082/kjfm.2010.31.3.208
44. Shumaker SA, Czajkowski SM. *Social Support and Cardiovascular Disease* (2013). DOI: 10.1007/978-1-4899-2572-5
45. Christian AH, Mochari H, Mosca LJ. **Waist Circumference, Body Mass Index, and Their Association With Cardiometabolic and Global Risk**. *J Cardiometab Syndr* (2009) **4**. DOI: 10.1111/j.1559-4572.2008.00029.x
46. Fan H, Li X, Zheng L, Chen X, Lan Q, Wu H. **Abdominal Obesity is Strongly Associated With Cardiovascular Disease and its Risk Factors in Elderly and Very Elderly Community-Dwelling Chinese**. *Sci Rep* (2016) **6** 21521. DOI: 10.1038/srep21521
47. Peer N, Steyn K, Levitt N. **Differential Obesity Indices Identify the Metabolic Syndrome in Black Men and Women in Cape Town: The CRIBSA Study**. *J Public Health (Oxf)* (2016) **38**. DOI: 10.1093/pubmed/fdu115
48. Zhang ZQ, Deng J, He LP, Ling WH, Su YX, Chen YM. **Comparison of Various Anthropometric and Body Fat Indices in Identifying Cardiometabolic Disturbances in Chinese Men and Women**. *PloS One* (2013) **8** e70893. DOI: 10.1371/journal.pone.0070893
49. Melmer A, Lamina C, Tschoner A, Ress C, Kaser S, Laimer M. **Body Adiposity Index and Other Indexes of Body Composition in the SAPHIR Study: Association With Cardiovascular Risk Factors**. *Obesity (Silver Spring)* (2013) **21**. DOI: 10.1002/oby.20289
50. Fujita M, Sato Y, Nagashima K, Takahashi S, Hata A. **Predictive Power of a Body Shape Index for Development of Diabetes, Hypertension, and Dyslipidemia in Japanese Adults: A Retrospective Cohort Study**. *PloS One* (2015) **10** e0128972. DOI: 10.1371/journal.pone.0128972
51. Li G, Wu HK, Wu XW, Cao Z, Tu YC, Ma Y. **The Feasibility of Two Anthropometric Indices to Identify Metabolic Syndrome, Insulin Resistance and Inflammatory Factors in Obese and Overweight Adults**. *Nutrition* (2019) **57** 194-201. DOI: 10.1016/j.nut.2018.05.004
52. Maessen MF, Eijsvogels TM, Verheggen RJ, Hopman MT, Verbeek AL, de Vegt F. **Entering a New Era of Body Indices: The Feasibility of a Body Shape Index and Body Roundness Index to Identify Cardiovascular Health Status**. *PloS One* (2014) **9** e107212. DOI: 10.1371/journal.pone.0107212
53. Cheung YB. **“A Body Shape Index” in Middle-Age and Older Indonesian Population: Scaling Exponents and Association With Incident Hypertension**. *PloS One* (2014) **9** e85421. DOI: 10.1371/journal.pone.0085421
54. Rico-Martín S, Calderón-García JF, Sánchez-Rey P, Franco-Antonio C, Martínez Alvarez M, Sánchez Muñoz-Torrero JF. **Effectiveness of Body Roundness Index in Predicting Metabolic Syndrome: A Systematic Review and Meta-Analysis**. *Obes Rev* (2020) **21** e13023. DOI: 10.1111/obr.13023
55. Cerhan JR, Moore SC, Jacobs EJ, Kitahara CM, Rosenberg PS, Adami HO. **A Pooled Analysis of Waist Circumference and Mortality in 650,000 Adults**. *Mayo Clin Proc* (2014) **89**. DOI: 10.1016/j.mayocp.2013.11.011
56. Dwivedi AK, Dubey P, Cistola DP, Reddy SY. **Association Between Obesity and Cardiovascular Outcomes: Updated Evidence From Meta-Analysis Studies**. *Curr Cardiol Rep* (2020) **22** 25. DOI: 10.1007/s11886-020-1273-y
57. Song X, Jousilahti P, Stehouwer CD, Söderberg S, Onat A, Laatikainen T. **Cardiovascular and All-Cause Mortality in Relation to Various Anthropometric Measures of Obesity in Europeans**. *Nutr Metab Cardiovasc Dis* (2015) **25** 295-304. DOI: 10.1016/j.numecd.2014.09.004
58. Yi SW, Ohrr H, Shin SA, Yi JJ. **Sex-Age-Specific Association of Body Mass Index With All-Cause Mortality Among 12.8 Million Korean Adults: A Prospective Cohort Study**. *Int J Epidemiol* (2015) **44**. DOI: 10.1093/ije/dyv138
59. Gonna H, Ray KK. **The Importance of Dyslipidaemia in the Pathogenesis of Cardiovascular Disease in People With Diabetes**. *Diabetes Obes Metab* (2019) **21 Suppl 1** 6-16. DOI: 10.1111/dom.13691
60. Tajeu GS, Booth JN, Colantonio LD, Gottesman RF, Howard G, Lackland DT. **Incident Cardiovascular Disease Among Adults With Blood Pressure <140/90 Mm Hg**. *Circulation* (2017) **136** 798-812. DOI: 10.1161/CIRCULATIONAHA.117.027362
61. Lee JJ, Pedley A, Hoffmann U, Massaro JM, Fox CS. **Association of Changes in Abdominal Fat Quantity and Quality With Incident Cardiovascular Disease Risk Factors**. *J Am Coll Cardiol* (2016) **68**. DOI: 10.1016/j.jacc.2016.06.067
62. Tian S, Zhang X, Xu Y, Dong H. **Feasibility of Body Roundness Index for Identifying a Clustering of Cardiometabolic Abnormalities Compared to BMI, Waist Circumference and Other Anthropometric Indices: The China Health and Nutrition Survey, 2008 to 2009**. *Med (Baltimore)* (2016) **95** e4642. DOI: 10.1097/MD.0000000000004642
63. Zhang J, Zhu W, Qiu L, Huang L, Fang L. **Sex- and Age-Specific Optimal Anthropometric Indices as Screening Tools for Metabolic Syndrome in Chinese Adults**. *Int J Endocrinol* (2018) **2018** 1067603. DOI: 10.1155/2018/1067603
64. Paajanen TA, Oksala NK, Kuukasjärvi P, Karhunen PJ. **Short Stature is Associated With Coronary Heart Disease: A Systematic Review of the Literature and a Meta-Analysis**. *Eur Heart J* (2010) **31**. DOI: 10.1093/eurheartj/ehq155
|
---
title: A Spanish-language translation for the U.S. of the type 2 diabetes stigma assessment
scale (DSAS-2 Spa-US)
authors:
- Kevin L. Joiner
- Mackenzie P. Adams
- Amani Bayrakdar
- Jane Speight
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2022
pmcid: PMC10012130
doi: 10.3389/fcdhc.2022.1057559
license: CC BY 4.0
---
# A Spanish-language translation for the U.S. of the type 2 diabetes stigma assessment scale (DSAS-2 Spa-US)
## Abstract
### Background
Diabetes stigma is recognized to negatively impact health-related outcomes for people living with type 2 diabetes (T2D), but there is a dearth of evidence among U.S. Latino adults with T2D. Our aim was to develop a Spanish-language translation of the Type 2 Diabetes Stigma Assessment Scale (DSAS-2) and examine its psychometric properties among U.S. Latino adults with T2D.
### Methods
The translation was developed through a multi-step process, including a focus group with community health workers ($$n = 5$$) and cognitive debriefing interviews with Latino adults with T2D ($$n = 8$$). It was field-tested in an online survey of U.S. Latino adults with T2D, recruited via Facebook (October 2018 to June 2019). Exploratory factor analysis examined structural validity. Convergent and divergent validity were assessed by testing hypothesized correlations with measures of general chronic illness stigma, diabetes distress, depressive and anxiety symptoms, loneliness, and self-esteem.
### Results
Among 817 U.S. Latino adults with T2D who participated in the online survey, 517 completed the Spanish-language DSAS-2 (DSAS Spa-US) and were eligible for the study (mean age 54 ± 10 years, and $72\%$ female). Exploratory factor analysis supported a single-factor solution (eigenvalue=8.20), accounting for $82\%$ of shared variance among the 19 items, all loading ≥ 0.5. Internal consistency reliability was high (α=0.93). As expected, strong, positive correlations were observed between diabetes stigma and general chronic illness stigma (rs=0.65) and diabetes distress (rs=0.57); medium, positive correlations, between diabetes stigma and depressive (rs=0.45) and anxiety (rs=0.43) symptoms, and loneliness (rs=0.41); and a moderate negative correlation between diabetes stigma and self-esteem (rs=-0.50). There was no relationship between diabetes stigma and diabetes duration (rs=0.07, ns).
### Conclusion
The DSAS-2 Spa-US is a version of the DSAS-2, translated into Spanish, that has good psychometric properties for assessing diabetes stigma in U.S. Latino adults with T2D.
## Introduction
An estimated 37 million U.S. adults are living with diabetes: the nationwide prevalence is nearly $15\%$, and the estimated yearly cost is $327 billion [1]. Many adults with type 2 diabetes (T2D), accounting for 90 to 95 percent of diabetes, encounter diabetes-related stigma and discrimination in social spaces, workplaces, and healthcare facilities (2–5). Stigma processes occur due to a characteristic that “marks” an individual as different from others, typically in contexts of culture and/or power [6]. There are two general forms of stigma: social stigma and self-stigma [6]. Social stigma can be experienced and perceived by an individual as blame, judgment, stereotyping, rejection, exclusion, and discrimination. Self-stigma occurs when society’s negative beliefs are internalized by an individual, manifesting as feelings of embarrassment, shame, reduced self-efficacy, and/or reduced self-esteem.
The Type 2 Diabetes Stigma Assessment Scale (DSAS-2) was developed to assess experienced and perceived diabetes-related stigma among adults with T2D [7]. Development of the DSAS-2 was informed by a comprehensive literature review and qualitative research [2, 8]. Subsequent research using the DSAS-2 in Australian adults with T2D showed that experiences and perceptions of diabetes stigma are associated with higher levels of diabetes distress and more depressive and anxiety symptoms [3]. A study using the DSAS-2 in a large sample of U.S. adults with T2D found that diabetes stigma is associated with higher diabetes distress, lower engagement in diabetes self-management, lower diabetes self-efficacy, and lower quality interactions with health care professionals [9].
While medical advancements and new technologies have transformed T2D health and health care over the last few decades, not all people with T2D have benefited equally [10]. People from minority racial and ethnic backgrounds are disproportionately affected by T2D, with higher rates of diabetes-related complications and mortality [11]. Adults who identify as Hispanic or Latino (hereto forth referred to as Latino) make up the largest ethnic minority group in the U.S., comprising $17.5\%$ of adults (6.5 million people) with diabetes [1]. Seventy percent of Latinos in the U.S. speak Spanish at home, and those with limited English proficiency are often excluded from research. This can be due to factors such as researchers not devoting sufficient resources to building trusted relationships with Spanish-speaking Latino populations, lack of professional interpreters, and Spanish-language study materials not being made available [12]. Furthermore, research suggests that individuals’ experiences and perceptions of stigma related to chronic illnesses vary by race and ethnicity [13, 14]. These deficiencies pose a significant threat to representing Latino adults in diabetes research, addressing disparities, and advancing the healthcare system’s capacity to meet the needs of diabetes disparities populations.
Therefore, the aims of this study were to conduct a cultural and linguistic validation of the DSAS-2, to create a Spanish-language version (DSAS-2 Spa-US), and to assess its psychometric properties in a sample of U.S. Latino adults with T2D. This will enable assessment of diabetes stigma in this population to track and evaluate gaps in health equity.
## Materials and methods
The study included two phases: 1) developing the DSAS-2 Spa-US, and 2) field-testing the DSAS-2 Spa-US in an online survey of Spanish-speaking U.S. Latino adults with T2D. The study was approved by the Health Sciences and Behavioral Sciences Institutional Review Board of the University of Michigan (HUM00139792 and HUM00142346).
## Characteristics of the DSAS-2
The DSAS-2 is comprised of 19 items, which form a total scale (19 items), and three subscales: (a) Treated Differently (six items), (b) Blame and Judgement (seven items), and (c) Self-Stigma (six items) [7]. Each item is presented as a statement with five Likert-type ratings ranging: 1 (strongly disagree), 2 (disagree), 3 (unsure), 4 (agree), and 5 (strongly agree). The scale and subscales are scored by summing the relevant items. Higher scores on the total scale are interpreted as more experienced or perceived diabetes stigma, and higher scores on the subscales indicate greater endorsement of experiencing or perceiving being treated differently, experiencing or perceiving blame or judgment, and experiencing or perceiving self-stigma.
## Translation of the DSAS-2 into Spanish for the U.S.
The translation team included the primary investigator, who holds a BA in Spanish and a Ph.D. in nursing, and three translators, who were doctoral-prepared or held a Ph.D. in Spanish. Two translators identify as Latino and are native Spanish speakers, one of Mexican heritage, and one of Peruvian heritage. One translator identifies as non-Hispanic White and is a native English speaker from the U.S. In the first step, the translators who are native Spanish speakers independently translated the DSAS-2 from English into Spanish, which resulted in two initial prototypes of the DSAS-2 Spa-US. In the second step, the team compared the initial versions, identified, and resolved discrepancies, and created a harmonized prototype of the DSAS-2 Spa-US [15]. In the third step, the third translator, a native English speaker, translated the harmonized prototype from Spanish into English. In the fourth step, the team presented the English translation of the harmonized prototype to a representative of the Australian team that developed the original English language DSAS-2, who liaised with the senior researcher on the Australian team. In this meeting, any translation challenges were discussed and resolved by consensus agreement. For example, there was some difficulty finding Spanish terms and phrases equivalent to the English terms and phrases used in the DSAS-2: “shame,” “I’m ashamed,” and “I feel embarrassed.” Although several possible Spanish terms and phrases were considered, it was decided, based on other widely used and well-respected healthcare resources that are translated from English into Spanish, to use the Spanish term “vergüenza” to translate the term “shame,” the Spanish phrase “me da vergüenza” to translate the English phrase “I’m ashamed,” and the Spanish phrase “me siento avergonzado/a” to translate the English phrase “I feel embarrassed” [16, 17].
In the fifth and sixth steps, the principal investigator conducted a focus group (April 2018) of Spanish-speaking community health workers ($$n = 5$$) who had professional experience providing diabetes education and support in Spanish for Latino adults with T2D and then conducted cognitive debriefing interviews in Spanish (May 2018) with Spanish-speaking Latino adults with T2D ($$n = 8$$), to elicit feedback on the prototype of the DSAS-2 Spa-US approved by the Australian team. The community health workers and the Latino adults with T2D were recruited from a health services organization serving communities in West Michigan. The cognitive debriefing interview participants were asked to reflect on and respond to each of the DSAS-2 Spa-US items and then share what they had thought about when reading and contemplating their responses to the DSAS-2 Spa-US items. The focus group participants each received a $50 gift card in appreciation of one hour of their time, while cognitive debriefing interviews participants each received a $25 gift card for 30 minutes of their time. The translation team reviewed the data for cases where participants consistently indicated that they had trouble with the instructions, the items, or the response options. Minor rewordings were discussed and agreed upon, changes were made. In the seventh and final step, two scientific experts in the field of diabetes-related psychosocial support who are both native Spanish speakers, one of Cuban heritage, and the other of Mexican heritage, assessed the DSAS-2 Spa-US for the use of appropriate language for Latino adults living with T2D in the U.S.
## Field testing and psychometric validation of the DSAS-2 Spa-US
In the second phase of the study, the DSAS-2 Spa-US was field-tested by Spanish-speaking U.S. Latino adults with T2D who participated in an online survey using Qualtrics XM software (Seattle, WA). The inclusion criteria were being 18 years or older, residing currently in the U.S., speaking Spanish, identifying as Hispanic or Latino, and having a current diagnosis of T2D. A total of 817 people responded to the survey, which was advertised through Facebook between October 2018 and June 2019. This study includes data on the 517 survey respondents who completed >$90\%$ of the 19-item DSAS-2 Spa-US. All participants provided consent before enrollment. Due to the anonymous nature of the online survey, participants did not receive compensation for their time, which might account for the large dropout before survey completion.
Participants provided sociodemographic and clinical information, including age, sex, country of origin, type of geographic area (urban/rural/suburban), educational level, relationship status, employment status, household income, diabetes duration, and diabetes treatment modality. Participants were asked to report their height and body weight to enable the calculation of body mass index (BMI) [weight (kg) divided by height (m2)].
Participants were administered Spanish-language versions of instruments measuring several psychological constructs, which are hypothesized to have relationships with diabetes stigma [7]. A Spanish-language version of the 8-item Stigma Scale for Chronic Illnesses (SSCI-8; α = 0.84), measures general chronic illness stigma [18]. Permission was granted for an instruction to be added before the items of the SSCI-8, which encouraged the respondents to respond regarding their diabetes (Jones, J. P., personal communication, June 1, 2018). A Spanish-language version [19] of the 17-item Diabetes Distress Scale (DDS-17; α = 0.96) measures diabetes-specific emotional distress [20]. A Spanish-language version [21] of the 8-item Patient Health Questionnaire (PHQ-8; α = 0.89) measures depressive symptoms [22]. A Spanish-language version [23] of the 7-item Generalized Anxiety Disorder-7 scale (GAD-7; α = 0.91), measures anxiety symptoms [24]. A Spanish-language version [25] of the University of California Los Angeles (UCLA) 3-item loneliness scale (α = 0.87), measures loneliness [26]. A Spanish-language version [27] of the Rosenberg Self-Esteem Scale (RSE; α = 0.79) measures general self-esteem [28].
## Data analysis
A descriptive approach was used to analyze the qualitative data [29]. Univariate analyses were used to describe the characteristics of the study sample. Exploratory factor analysis (unrotated) was used to determine if the full set of the 19 items of the DSAS-2 Spa-US clustered together into one or more factors. Internal consistency reliability was assessed with Cronbach’s alpha, α. Following convention, simple imputation was applied in cases where <$10\%$ of data were missing for scoring the DSAS-2 Spa-US, the SSCI-8, the DDS, the PHQ-8, the GAD-7, and the RSE. Convergent validity was assessed against the scores of the SSCI-8, the DDS, the PHQ-8, the GAD-7, the 3-item UCLA Loneliness Scale, and the RSE. Based on the theoretical absence of a relationship between duration of diabetes and diabetes stigma, discriminant validity was assessed against diabetes duration. Moderate-to-large positive or negative correlations were expected as evidence of support of convergent validity. Correlations were considered, negligible (rs<0.10), small (rs≥0.10-0.29), moderate (rs≥0.30-0.49), or large (rs≥0.50) [30]. A p-value of < 0.05 was considered statistically significant. Analyses were performed in STATA Version 16 (College Station, TX).
## Development of the DSAS-2 Spa-US
The primary issues raised by the community health worker participants in the focus group concerned the importance of the DSAS-2 Spa-US meeting the needs of individuals with varying levels of reading comprehension and health literacy. Based on the cognitive debriefing interview data, the translation of two items was identified as notable and needed further consideration. In the first case, the item stated in English reads, “Health professionals think that people with type 2 diabetes don’t know how to take care of themselves.” As it was translated into Spanish, it led some of the cognitive debriefing interview participants to interpret the phrase “don’t know how to” as “don’t know the ways” or “don’t know the information.” Thus, the item was interpreted as asking if it is perceived that health professionals think that people with diabetes need information and help with their self-care for their diabetes. According to the concept elaboration document provided by the Australian team, this contrasted with the original intent of the item, which was to ask whether it is perceived that healthcare professionals are generally unfair or unjustified in their judgments of individuals’ self-care of their diabetes. The second item that was identified for further consideration was “Because I have type 2 diabetes, some people judge me for my food choices.” Some of the cognitive debriefing interview participants took the interpretation of this item to mean that people had the desire for guests to eat what their hosts offered, in the sense that it would be rude or disrespectful not to eat what a host provided. Based on the explanation of the item in the concept elaboration document, this item was intended to convey a sense of judgment.
The focus group feedback was reviewed and discussed by the team, and their suggestions were incorporated into the DSAS-2 Spa-US. The team discussed the translation challenges that were uncovered in the cognitive debriefing interviews and decided to keep the translations of the two items as they were in the DSAS-2 Spa-US to maintain as much consistency as possible with the original English language DSAS-2. After minor rewording changes were made (Appendix 1) a consensus-derived determination was made that the final DSAS-2 Spa-US displayed the use of appropriate language linguistically and culturally for U.S. Latino adults living with diabetes.
## Psychometric properties of the DSAS-2 Spa-US in Latino adults with T2DM
The characteristics of the study sample of participants in the online survey are displayed in Table 1. The mean (SD) age was 54 [10] years old; $72\%$ ($$n = 374$$) were female, $54\%$ ($$n = 281$$) were born in Mexico, $62\%$ ($$n = 320$$) had an education level of high school or less, $51\%$ ($$n = 263$$) were not currently working, $71\%$ ($$n = 368$$) had a yearly household income of <$40,000, and $73\%$ ($$n = 375$$) were currently residing in urban areas. Fifty-eight percent ($$n = 302$$) reported a height and body weight indicating a BMI ≥30 kg/m2, the mean (SD) duration of T2D was 10 [9] years, and $30\%$ ($$n = 156$$) used insulin to manage their T2D.
**Table 1**
| Age (years), mean ± SD | 53.9 ± 10.1 |
| --- | --- |
| Sex, female, n (%) | 374 (72.3) |
| Education level, n (%) | Education level, n (%) |
| Primary school (grades 1-5) | 50 (9.7) |
| Middle school (grades 6-8) | 83 (16.1) |
| High school (grades 9-12) | 187 (36.2) |
| Technical school or university | 190 (36.8) |
| Employment status, n (%) | Employment status, n (%) |
| Full-time | 158 (30.6) |
| Part-time | 84 (16.3) |
| Not working | 263 (50.9) |
| Yearly household income, n (%) | Yearly household income, n (%) |
| < $10,000 | 155 (30.0) |
| $10,000 to < $20,000 | 127 (24.6) |
| $20,000 to < $40,000 | 86 (16.6) |
| ≥ $40,000 | 105 (20.3) |
| Don’t know/prefer not to say | 44 (8.5) |
| Relationship status, n (%) | Relationship status, n (%) |
| Single | 63 (12.2) |
| Married or living together | 324 (62.6) |
| Divorced or separated or widowed | 124 (24.0) |
| Place of birth, n (%) | Place of birth, n (%) |
| U.S. (50 states and District of Columbia) | 40 (7.7) |
| Mexico | 281 (54.4) |
| Caribbean | 74 (14.3) |
| Central America | 53 (10.3) |
| South America | 42 (8.1) |
| Geographic area, n (%) | Geographic area, n (%) |
| Urban | 375 (72.5) |
| Suburban | 62 (12.0) |
| Rural | 46 (8.9) |
| BMI category, n (%) | BMI category, n (%) |
| Underweight (<18.5 kg/m2) | 4 (0.8) |
| Normal weight (18.5-24.9 kg/m2) | 52 (10.1) |
| Overweight (25-29.9 kg/m2) | 159 (30.8) |
| Obesity (30-39.9 kg/m2) | 191 (36.9) |
| Severe obesity (≥40 kg/m2) | 111 (21.5) |
| Duration of diabetes diagnosis (years), Mean ± SD (range) | 10.2 ± 9.4 (0-59) |
| Primary diabetes treatment, n (%) | Primary diabetes treatment, n (%) |
| Lifestyle | 35 (7.6) |
| Oral hypoglycemic agents | 281 (61.2) |
| Insulin | 156 (30.2) |
| Diabetes specific distress (DDS-17 score) | Diabetes specific distress (DDS-17 score) |
| Mean ± SD, (range) | 2.9 ± 1.3 (1-6) |
| Median (25th - 75th percentile) | 2.7 (1.6, 3.9) |
| Depressive symptoms (PHQ-8 score) | Depressive symptoms (PHQ-8 score) |
| Mean ± SD, (range) | 7.3 ± 5.8 (0-24) |
| Median (25th - 75th percentile) | 6.0 (3.0, 11.0) |
| Anxiety symptoms (GAD-7 score) | Anxiety symptoms (GAD-7 score) |
| Mean ± SD, (range) | 6.1 ± 5.3 (0-21) |
| Median (25th - 75th percentile) | 5.0 (2.0, 9.0) |
| Loneliness (UCLA 3-item loneliness scale) | Loneliness (UCLA 3-item loneliness scale) |
| Mean ± SD, (range) | 1.7 ± 0.6 (1-3) |
| Median (25th - 75th percentile) | 1.7 (1.0 - 2.0) |
| Self-esteem (RSE score) | Self-esteem (RSE score) |
| Mean ± SD, (range) | 31.1 ± 4.8 (15-40) |
| Median (25th - 75th percentile) | 32.0 (28.0, 35.0) |
Exploratory factor analyses revealed that $82\%$ of the variance in the participant responses to the 19 items of the DSAS-2 Spa-US was explained by a single factor (Table 2), with all the items loading >0.5. The scree plot also suggested a single factor because the eigenvalues level off after one factor (Figure 1). A three-factor solution was not apparent, suggesting that subscale scores are not supported for the DSAS-2 Spa-US. The internal reliability of the total DSAS-2 Spa-US scale was high (α=0.93), supporting the calculation of a total DSAS-2 Spa-US score representing diabetes stigma. For all 19 items, the full range of response options was endorsed. The frequency distributions of the individual items had consistently positive skews, ranging from 0.21 to 1.45. The kurtosis for individual items was also consistently positive, ranging from 1.56 to 4.70. Missing response data for the individual items ranged from $1\%$ to $4\%$ (Table 3).
Convergent validity was demonstrated with strong, positive correlations observed between diabetes stigma and general chronic illness stigma (rs=0.65, $p \leq 0.001$) and diabetes distress (rs=0.57, $p \leq 0.001$), and medium, positive correlations observed between diabetes stigma and depressive symptoms (rs=0.45, $p \leq 0.001$), anxiety symptoms (rs=0.43, $p \leq 0.001$), and loneliness (rs=0.41, $p \leq 0.001$). There was a medium, negative correlation between diabetes stigma and general self-esteem (rs=-0.51, $p \leq 0.001$). Discriminant validity was supported by the lack of correlation between diabetes stigma and duration of diabetes ($r = 0.07$, $$p \leq 0.120$$).
## Discussion
The DSAS-2 was translated into Spanish, and the field test results of the DSAS-2 Spa-US among U.S. Latino adults with T2D indicate good psychometric performance. The factor analysis identified a single factor (explaining $82\%$ of the variance) with strong internal consistency reliability. Evidence of the convergent and divergent validity of the DSAS-2 Spa-US was demonstrated, confirming the hypothesized relationships between diabetes stigma and general chronic illness stigma, diabetes distress, depressive and anxiety symptoms, loneliness, and general self-esteem. The lack of association between diabetes stigma and duration of diabetes demonstrated evidence of discriminant validity.
The psychometric performance of the DSAS-2 Spa-US among U.S. Latino adults with T2D differed somewhat from that of the DSAS-2 among Australian adults with T2D. In the current study, there was strong support for a single unidimensional scale but no evidence for the three subscales (Treated Differently, Blame and Judgement, and Self-Stigma), which were identified during the original validation of the DSAS-2 among Australian adults with T2D [7]. This discrepancy may be due to the Australian participants having a higher education level than the U.S. Latino participants ($60\%$ versus $37\%$ with more than a high school education). Though it is unclear whether the differing factor structure can be attributed to education, it is important since adults with higher educational levels are more likely to report experiences and perceptions of health-related stigma [4, 31]. Another possible explanation is that despite the authors’ efforts, some concepts were challenging to translate from English into Spanish. A recently published study, which examined the psychometric properties of the DSAS-2 Spa-US using data from adults with T2D in Colombia, also provides evidence of support for a single unidimensional scale [32]. Taken together, the findings highlight the need for further attention to education, culture, and race/ethnicity in diabetes stigma research, as there may be differences in the ways that having T2D is construed and in the value that is placed on orientation toward others and to the self that results in diabetes stigma being enacted differently. For now, these findings suggest that there is no support for the three subscales of the DSAS-2, and that only the total DSAS-2 scale score should be calculated when using the DSAS-2 Spa-US in Spanish-speaking adults with T2D in the US and in Colombia.
Latino adults in the U.S. constitute a large and diverse ethnic group, although there are commonalities, including using Spanish as a shared language [33]. To an extent, the diversity is reflected in the backgrounds of the participants in the study sample, in which adults with Mexican heritage accounted for the largest proportion of the study participants in the online survey ($54\%$). Although for Spanish-speaking groups, there are differences among the languages spoken in different parts of the country, the multiphase process that was used to develop the DSAS-2 Spa-US was intended to capture the true meaning of the English language used in the DSAS-2 to produce a Spanish-language translation of the DSAS-2 of high-quality and integrity.
## Implications for practice and future research
The DSAS-2 Spa-US can be used to assess diabetes stigma in Latino U.S. adults with T2D as part of future studies to determine the impact of diabetes stigma on clinical and psychosocial outcomes, as well as to examine the impact of interventions designed to minimize diabetes stigma. Studies using the DSAS-2 in adults with T2D in the U.S. indicate that diabetes stigma is associated with higher diabetes distress, lower engagement in diabetes self-management, lower diabetes self-efficacy, and lower quality interactions with healthcare professionals [9]. Studies to determine if similar relationships exist between diabetes stigma and these constructs using the DSAS-2 Spa-US in Latino U.S. adults with T2D are needed.
## Limitations
This study has some limitations. Due to the cross-sectional nature of the study, we could not determine the test-retest reliability of the DSAS-2 Spa-US (i.e., how consistently individuals might respond if asked to repeat the DSAS-2 Spa-US within a short period). Evidence supporting the validation of inferences made with the DSAS-2 Spa-US may have been stronger if test-retest reliability was evaluated. However, longitudinal stigma studies are lacking, and the trajectory of stigma over time is unknown. Nor could we determine the predictive validity of the DSAS-2 Spa-US (i.e., the extent to which diabetes stigma at baseline predicts future clinical or psychosocial outcomes). The DSAS-2 Spa-US was used in a large sample in which participants self-reported their clinical data. Using the DSAS-2 Spa-US in a clinical setting where clinical data could be obtained from electronic medical records would be desirable. The responsiveness of the DSAS-2 Spa-US in Latino adults with T2D will need to be examined in future intervention studies. *The* generalizability of the current research is dependent on the sample being representative of U.S. Latino adults with T2D. Participants were recruited via a social media platform and completed the survey online using a smartphone, a desktop, a laptop, or a tablet computer. While there have been gains in technology adoption in the U.S., the ‘digital divide’ still exists, which may have created barriers to participation for Latino adults with T2D who are older, have lower incomes, and live in rural areas [34, 35].
## Future research
This study contributes to the public health imperative to address diabetes disparities experienced among ethnic minority communities in the U.S. with the development of the DSAS-2 Spa-US to measure diabetes stigma in Latino adults with T2D. Experiences and perceptions of diabetes stigma remain understudied among adults with T2D. Considering the high prevalence of T2D among Latino adults in the U.S., research needs to continue to examine the effects of diabetes stigma on Latino populations’ diabetes self-management and overall well-being.
## Conclusion
This study demonstrates that the DSAS-2 Spa-US is a valid and reliable assessment of diabetes stigma and is suitable for U.S. Latino adults with T2D. It is ready for use in research to examine the experience and impact of diabetes stigma in U.S. Latino adults with T2D, an underrepresented ethnic group, to accelerate health equity and eliminate disparities in diabetes health outcomes and health care.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by The University of Michigan Institutional Review Board. The patients/participants provided their written informed consent to participate in this study.
## Author contributions
Conceptualization, KJ and JS. Methodology, KJ and JS. Formal analysis and investigation, KJ and AB. Writing - original draft preparation, KJ and MA. Writing - review and editing, JS. Funding acquisition, JS. Resources, KJ and JS. Supervision, JS. All authors contributed to the article and approved the submitted version.
## Conflict of interest
JS is the Director of the Australian Centre for Behavioural Research in Diabetes, which owns the copyright of the DSAS-2, including all translations.
The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcdhc.2022.1057559/full#supplementary-material
## References
1. 1
Centers for Disease Control and Prevention. National diabetes statistics report website (2022). Available at: https://www.cdc.gov/diabetes/data/statistics-report/index.html.. *National diabetes statistics report website* (2022)
2. Browne JL, Ventura A, Mosely K, Speight J. **I Call it the blame and shame disease': a qualitative study about perceptions of social stigma surrounding type 2 diabetes**. *BMJ Open* (2013) **3**. DOI: 10.1136/bmjopen-2013-003384
3. Holmes-Truscott E, Browne JL, Ventura AD, Pouwer F, Speight J. **Diabetes stigma is associated with negative treatment appraisals among adults with insulin-treated type 2 diabetes: results from the second diabetes MILES - Australia (MILES-2) survey**. *Diabetic Med.* (2018) **35**. DOI: 10.1111/dme.13598
4. Liu NF, Brown AS, Folias AE, Younge MF, Guzman SJ, Close KL. **Stigma in people with type 1 or type 2 diabetes**. *Clin. Diabetes* (2017) **35** 27-34. DOI: 10.2337/cd16-0020
5. Stuckey H, Peyrot M. **Living with diabetes: literature review and secondary analysis of qualitative data**. *Diabetic Med.* (2020) **37** 493-503. DOI: 10.1111/dme.14255
6. Link BG, Phelan JC. **Conceptualizing stigma**. *Annu. Rev. Sociol.* (2001) **27**. DOI: 10.1146/annurev.soc.27.1.363
7. Browne JL, Ventura AD, Mosely K, Speight J. **Measuring the stigma surrounding type 2 diabetes: Development and validation of the type 2 diabetes stigma assessment scale (DSAS-2)**. *Diabetes Care* (2016) **39**. DOI: 10.2337/dc16-0117
8. Schabert J, Browne JL, Mosely K, Speight J. **Social stigma in diabetes : a framework to understand a growing problem for an increasing epidemic**. *Patient* (2013) **6** 1-10. DOI: 10.1007/s40271-012-0001-0
9. Puhl RM, Himmelstein MS, Hateley-Browne JL, Speight J. **Weight stigma and diabetes stigma in U.S. adults with type 2 diabetes: Associations with diabetes self-care behaviors and perceptions of health care**. *Diabetes Res. Clin. Pract.* (2020) **168**. DOI: 10.1016/j.diabres.2020.108387
10. Gabbay RA, Kendall D, Beebe C, Cuddeback J, Hobbs T, Khan ND. **Addressing therapeutic inertia in 2020 and beyond: A 3-year initiative of the American diabetes association**. *Clin. Diabetes* (2020) **38**. DOI: 10.2337/cd20-0053
11. Golden SH, Brown A, Cauley JA, Chin MH, Gary-Webb TL, Kim C. **Health disparities in endocrine disorders: biological, clinical, and nonclinical factors–an endocrine society scientific statement**. *J. Clin. Endocrinol. Metab.* (2012) **97**. DOI: 10.1210/jc.2012-2043
12. Yancey AK, Ortega AN, Kumanyika SK. **Effective recruitment and retention of minority research participants**. *Annu. Rev. Public Health* (2006) **27** 1-28. DOI: 10.1146/annurev.publhealth.27.021405.102113
13. Hatzenbuehler ML, Phelan JC, Link BG. **Stigma as a fundamental cause of population health inequalities**. *Am. J. Public Health* (2013) **103**. DOI: 10.2105/AJPH.2012.301069
14. Weiss MG, Ramakrishna J, Somma D. **Health-related stigma: rethinking concepts and interventions**. *Psychol. Health Med.* (2006) **11**. DOI: 10.1080/13548500600595053
15. Wild D, Grove A, Martin M, Eremenco S, McElroy S, Verjee-Lorenz A. **Principles of good practice for the translation and cultural adaptation process for patient-reported outcomes (PRO) measures: report of the ISPOR task force for translation and cultural adaptation**. *Value Health* (2005) **8** 94-104. DOI: 10.1111/j.1524-4733.2005.04054.x
16. Post PRTW, Grover S. **Patient education: Health risks of obesity (The basics)**. *UptoDate* (2021)
17. Post PRTW, Grover S. **Patient education: Social anxiety disorder (The basics)**. *UptoDate* (2021)
18. Molina Y, Choi SW, Cella D, Rao D. **The stigma scale for chronic illnesses 8-item version (SSCI-8): development, validation and use across neurological conditions**. *Int. J. Behav. Med.* (2013) **20**. DOI: 10.1007/s12529-012-9243-4
19. Martinez-Vega IP, Doubova SV, Aguirre-Hernandez R, Infante-Castañeda C. **Adaptation and validation of the distress scale for Mexican patients with type 2 diabetes and hypertension: a cross-sectional survey**. *BMJ Open* (2016) **6**. DOI: 10.1136/bmjopen-2015-009723
20. Polonsky WH, Fisher L, Earles J, Dudl RJ, Lees J, Mullan J. **Assessing psychosocial distress in diabetes: development of the diabetes distress scale**. *Diabetes Care* (2005) **28**. DOI: 10.2337/diacare.28.3.626
21. Huang FY, Chung H, Kroenke K, Delucchi KL, Spitzer RL. **Using the patient health questionnaire-9 to measure depression among racially and ethnically diverse primary care patients**. *J. Gen. Internal Med.* (2006) **21**. DOI: 10.1111/j.1525-1497.2006.00409.x
22. Kroenke K, Strine TW, Spitzer RL, Williams JB, Berry JT, Mokdad AH. **The PHQ-8 as a measure of current depression in the general population**. *J. Affect. Disord.* (2009) **114**. DOI: 10.1016/j.jad.2008.06.026
23. Mills SD, Fox RS, Malcarne VL, Roesch SC, Champagne BR, Sadler GR. **The psychometric properties of the generalized anxiety disorder-7 scale in Hispanic americans with English or Spanish language preference**. *Cultural Diversity Ethnic Minority Psychol.* (2014) **20**. DOI: 10.1037/a0036523
24. Spitzer RL, Kroenke K, Williams JBW, Löwe B. **A brief measure for assessing generalized anxiety disorder: The GAD-7**. *Arch. Internal Med.* (2006) **166**. DOI: 10.1001/archinte.166.10.1092
25. Foti SA, Khambaty T, Birnbaum-Weitzman O, Arguelles W, Penedo F, Espinoza Giacinto RA. **Loneliness, cardiovascular disease, and diabetes prevalence in the Hispanic community health Study/Study of latinos sociocultural ancillary study**. *J. Immigrant Minority Health* (2020) **22**. DOI: 10.1007/s10903-019-00885-7
26. Hughes ME, Waite LJ, Hawkley LC, Cacioppo JT. **A short scale for measuring loneliness in Large surveys: Results from two population-based studies**. *Res. Aging* (2004) **26**. DOI: 10.1177/0164027504268574
27. Gómez-Lugo M, Espada JP, Morales A, Marchal-Bertrand L, Soler F, Vallejo-Medina P. **Adaptation, validation, reliability and factorial equivalence of the Rosenberg self-esteem scale in Colombian and Spanish population**. *Spanish J. Psychol.* (2016) **19** E66. DOI: 10.1017/sjp.2016.67
28. Rosenberg M. *Society and the adolescent self-image* (2016)
29. Sandelowski M. **Whatever happened to qualitative description**. *Res. Nurs. Health* (2000) **23**. DOI: 10.1002/1098-240x(200008)23:4<334::aid-nur9>3.0.co;2-g
30. Levine TR. **Confirmatory factor analysis and scale validation in communication research**. *Communication Res. Rep.* (2005) **22**. DOI: 10.1080/00036810500317730
31. Himmelstein MS, Puhl RM. **At Multiple fronts: Diabetes stigma and weight stigma in adults with type 2 diabetes**. *Diabetic Med.* (2021) **38**. DOI: 10.1111/dme.14387
32. Pedrero V, Alonso LM, Manzi J. **Propiedades psicométricas de la versión en español del type 2 diabetes stigma assessment scale (DSAS 2) en población colombiana [Psychometric properties of the Spanish version of the type 2 diabetes stigma assessment scale (DSAS 2) in a Colombian population]**. *Biomedica Rev. del Instituto Nacional Salud* (2022) **42** 41-54. DOI: 10.7705/biomedica.6128
33. Lopez MH, Krogstad JM, Passel JS. *Who is Hispanic?* (2021)
34. Vogels E. *Some digital divides persist between rural, urban and suburban america, (2021)* (2021)
35. Atske S, Perrin A. *Home broadband adoption, computer ownership vary by race, ethnicity in the U.S* (2021)
|
---
title: Seeking Health Information and Social Support in the Diabetes Online Community
authors:
- Allyson Hughes
- Nazanin Heydarian
- Diana Gerardo
- Isabela Solis
- Osvaldo Morera
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2021
pmcid: PMC10012138
doi: 10.3389/fcdhc.2021.708405
license: CC BY 4.0
---
# Seeking Health Information and Social Support in the Diabetes Online Community
## Abstract
### Purpose
People with type 1 diabetes (T1D) search for health information online in the Diabetes Online Community (DOC), where individuals with diabetes, researchers and caregivers post and respond to health questions. The aims of this study were 1) to understand how people with T1D are seeking health information and engaging in health behaviors in the DOC, and 2) develop a measure of online health information seeking in adults with T1D.
### Research Method
Ninety-five adults with T1D completed qualitative prompts online.
### Results
Themes that emerged in this study included sense of community, and multiple types of social support that are necessary in disease management.
### Conclusions
This study used qualitative methods to develop a valid scale tailored for adults with T1D. Future research should seek to collect additional data to bolster validity and reliability.
## Introduction
Seeking health information and engagement in online health communities are significant and emerging public health phenomena. Health information is being exchanged continuously in the Diabetes Online Community (DOC; 1). The online community thrives on social media sites and provides largely anecdotal evidence (microblogging via tweets, Facebook posts, blog posts and discussion boards) regarding medical decision-making [2]. Type 1 Diabetes (T1D) is a chronic condition that requires checking blood glucose levels multiple times a day, multiple insulin injections daily, and/or use of durable medical equipment that provides the person with diabetes with insulin [3]. T1D management varies from one person to the next; health practices that work extremely well for one person may be ineffective for another person [4]. People with T1D, who are members of the DOC, can encounter health information (considered one directional), but also participate in health engagement (bidirectional information exchanges) [5].
The need for tailored health information often leads people with T1D to seek information from their peers in a phenomenon known as peer-to-peer healthcare [6]. Health information seeking has long been documented as a key coping strategy and is characterized as a monitoring behavior that leads to problem-focused coping strategies [7]. Findings show increased information seeking is associated with increased self-care and health promotion [8]. Additionally, sharing health information is positively associated with behavioral intentions to follow health recommendations [9]. It is well established that the DOC is a environment for peer support disease management support [10]. Although we know about peer health information seeking and its benefits, and the way the DOC is set up would lend itself nicely to peer information seeking, no one has yet examined peer health information seeking in the DOC before.
Benefits from being a member of the DOC include 1) increased positive emotional experiences, 2) increased positive attitudes toward T1D, and 3) increased engagement in T1D management behaviors [11]. Psychosocial benefits to participating in online support groups include empowerment and social support [12]. Indeed, in a previous study, diabetes bloggers perceived more social support the more they blogged [13]. In a recent qualitative study, participants with either T1D or type 2 diabetes described the benefits of DOC membership and one key benefit was the ability for the online community to feel like a tight knit family [14]. The diabetes online community can also promote facilitators of diabetes self-management such as positive individual strategies and social support from peers and healthcare providers [15]. The online community could be the key for providing diabetes self-management strategies outside of the clinic [16].
Although there is much existing research on psychosocial and behavioral benefits of the DOC, there is a dearth of information on health information seeking in the DOC. For this reason, the current qualitative study was exploratory in nature and sought to determine how people with T1D seek health information and participate in health engagement in the DOC and create a scale on seeking health information online for adults with T1D.
## Methods
This study’s qualitative data was collected solely online using Qualtrics and was analyzed using a thorough thematic analysis framework [17].
## Participants
Participants were recruited from the DOC via Facebook posts, tweets using the hashtags #doc, #type1 diabetes and #dsma, and peer to peer referrals. Participant eligibility required the following: 1) being 18 years of age or older, 2) being a member of the DOC and 3) having been diagnosed with T1D by a doctor. The study sample included 95 DOC stakeholders. Prior to data collection, the study underwent review by the Institutional Review Board at the University of Texas at El Paso.
## Measures
The demographics questionnaire asked participants to self-report their age, gender, marital status, ethnicity, and education level. Participants reported their diabetes duration, mode of insulin delivery and health outcomes. Participants were provided with qualitative questions in order to better understand their role in the DOC and their experience in the DOC in general. Participants answered 15 Likert scale questions from the Attitudes towards Seeking Health Information Online Scale. Sample items included: I frequently use the internet to gain health advice in the Diabetes Online Community; I review multiple internet sources in the Diabetes Online Community before making a health decision for myself; I do not follow the health information that I find on social media in the Diabetes Online Community; I trust the health information that I find in the Diabetes Online Community; I feel comfortable receiving health advice in the Diabetes Online Community; I trust the health information that my friends on social media (Facebook, Twitter, Instagram, discussion forums) provide in the Diabetes Online Community. The scale was shown to the participants and they provided qualitative feedback regarding item wording. Existing literature was used to develop the scale and the qualitative questions further refined the scale with an emphasis on improving validity.
## Qualitative Research Questions
Participants responded to open-ended questions about their experiences as members of the DOC. The goal of these questions was to determine: What are people with T1D’s perceptions of how the DOC assists them with their physical and mental health? The second question was: What characterizes a person with T1D’s experiences interacting with DOC members to make a treatment decision? The third question was: What elements of the DOC do people with T1D find to be most useful? These research questions were formulated in relation to the open-ended questions participants answered.
## Research Assistant Training
In order to promote validity and reliability of the findings, two research assistants (RA) were trained to code the qualitative data. The RA met with the principal investigator several times to practice coding. The research team discussed the benefits and limitations of mixed methods in the online setting and discussed assigned readings, YouTube videos and podcasts regarding the culture of T1D in adults.
## Thematic Analyses Plan
A codebook was developed based upon existing literature and exploratory themes derived from RA training materials. The codebook consisted of 1) proposed categories, 2) proposed themes, 3) proposed subthemes, 4) definitions, and 5) example quotes to illustrate the meaning of the themes. Categories were populated by themes and themes were populated by subthemes. Themes and subthemes were made up of codes and these were derived from participant quotes. There were two coders, authors ASH and DG. After each round of coding, the authors would compare their codes and discuss discrepancies and come to agreement. Themes were determined both inductively and deductively using not only existing literature but also the data to structure themes.
## Participant Characteristics
Ninety-five participants were included in this sample. Complete demographics are available in Table 1.
**Table 1**
| Age, Mean (SD) | 26.8 (7.18) |
| --- | --- |
| Gender, N (%) | |
| Female | 87 (91.6%) |
| Race, N (%) | |
| White | 75 (78.9%) |
| African-American | 2 (2.1%) |
| Hispanic or Latino | 12 (12.6%) |
| Other | 6 (6.3%) |
| Education, N (%) | |
| < Bachelor’s Degree | 34 (35.8%) |
| Bachelor’s Degree | 34 (35.8%) |
| > Bachelor’s Degree | 27 (28.4%) |
| Household Income, Mean (SD) | $66,283 (56,148) |
| Diabetes Demographics | |
| Insulin Pump, N (%) | 59 (62.1%) |
| Length of Diagnosis, Mean (SD) | 12.3 years (9.17) |
| Continuous Glucose Monitor, N (%) | 55 (57.9%) |
| A1c, Mean (SD) | 7.1% (1.5) |
## Thematic Analyses
Coding categories included 756 quotes, and 36 themes. Participants completed a detailed assessment of the Attitudes Toward Seeking Online Health Information scale. More than half of the participants ($56.6\%$) reported completing the scale in 15 minutes or less. Seventy-seven percent of the participants believed the questions were written by someone who had an accurate idea of T1D. The first theme was sense of community ($$n = 48$$), where participants reported experiencing an overall feeling of belonging to something “bigger” in the DOC. One example of this is: [ID 177]: “Reading other people’s stories whom I can relate with. No judgement and everyone understands each other”. The second theme was social interaction and support ($$n = 30$$), defined as DOC members engaging with other members and receiving social support during these interactions. The third theme was informational support ($$n = 29$$), participants reported exchanging advice, and receiving and sharing suggestions related to diabetes management and overall information. Participants also reported learning about new technologies and medications.
## Pros and Cons of Membership in the DOC
Participants were also asked to describe the pros of being a member of the DOC. Themes included gaining information and advice from other members of the DOC ($$n = 25$$), experiencing a “sense of community” within the DOC ($$n = 52$$), stating [ID 108]: “We are all going through this together, so that is the best part”. Participants also described cons of being a member of the DOC including comparing self to others ($$n = 10$$) and misinformation ($$n = 6$$).
## Physical Impact of the DOC
Overall participants stated that the DOC had a positive impact on their physical health ($$n = 64$$), stating improvements in self-care, exercise behaviors, improved nutrition, and access to healthcare and medication. Quotes included: [ID 109] “It’s improved. I’ve learned a few tidbits to apply to daily life, especially about alternative snacking habits and insulin dosing strategies”. Another participant stated that the DOC helped them adjust to a new lifestyle change: [ID 120]: “Being able to see what others do in regards to their diet and preventing highs and lows before, during and after workouts has been amazing”. Lastly, a participant described a dire experience [ID 155]: “At one point I had no insulin and as soon as I asked for help someone from the group quickly got in contact with me and sent me some right away”.
## Mental Health Impact of the DOC
The majority of participants endorsed that the DOC had a positive impact on their mental health including feeling less isolated. Mental health benefits included receiving encouragement, feeling less alone, having an improved mental health status, feeling a sense of community, and normalizing the diabetes experience. Of note, participants also stated the negative mental health impact that the DOC can impart such that being part of the DOC caused them to experience anxiety about T1D and that involvement in the DOC promoted negative behaviors. An example of this negative behavior occurs when a member of the DOC shares a picture of what their blood sugars have been in the last 24 hours. If they post a picture of blood sugars in target range then that may produce anxiety in DOC members who are not experiencing in range blood sugars.
## Seeking Health Information in the DOC
Several participants reported seeking out advice about dosing insulin ($$n = 11$$). Overall, participants described seeking advice about durable medical equipment ($$n = 16$$). There were few overlapping themes regarding topics participants sought advice and engagement about: insulin dosing while exercising ($$n = 9$$), allergic reactions ($$n = 1$$), blood glucose advice ($$n = 3$$), diabulimia treatment ($$n = 1$$), blood sugar meter advice ($$n = 1$$), nutrition ($$n = 2$$), and sick days ($$n = 1$$). Others described consulting not only their healthcare professional but also the online community [doctor and DOC ($$n = 1$$)] and giving advice ($$n = 5$$). A participant described seeking help from the DOC when they were in the middle of a medical emergency in a foreign country.
Participants also reported seeking existing advice in order to answer any health questions they may have. Members of the DOC may endorse the answer to the question or state how this piece of information has impacted them. Participants reported that the DOC provided them with information on how to use their medical devices, including managing sick days, emergency situations, and exercising. Some DOC members indicated that they provide advice to others but they do not request it. This dynamic has important implications for the way that individuals with chronic disease seek health information. Specifically, existing information seeking theories do not take into account the role of information brokering that occurs in the DOC. Furthermore, there is a difference between the types of information (by topic) that are being sought and how (actively seeking advice versus passively seeking advice) they are being sought.
## Stakeholder Assessment of Seeking Health Information Online Scale
Regarding additional comments about the questionnaire, participants reported support for the items and endorsed the cultural competency of the scale with statements such as: [ID 195]: “asked relevant questions for someone with T1D”, and [ID 149]: “Asking how the diabetes community has helped with physical and mental health. Those are 2 significant aspects that are affected by this condition”. Several participants endorsed Instagram as part of the DOC because they sought social support for T1D management on Instagram. Instagram had not been previously included in the survey materials.
Specific participant requests encompassed a need to improve clarity and change “treatment decisions” to “advice”. Many participants stated the importance and major impact of the DOC in how they make decisions about which medical devices they will be using. Much of the conversation in the DOC involves medical device usage, tips and tricks, navigating insurance and medical claims advice and overall conversations on accessibility. Breaking news about medical devices is often shared widely in the DOC such that when FDA approval is given to a new diabetes device, DOC members will find out from social media-based news outlets and other DOC members before they find out from their doctor. Of importance, participants stated that they began using specific types of durable medical equipment due to endorsements from DOC members.
Participants also reported seeking existing advice in order to answer any health questions they may have. Members of the DOC may endorse the answer to the question or state how this piece of information has impacted them. This dynamic challenges how online health information seeking was originally conceptualized for this set of studies. The initial conceptualization did not account for existing information but instead focused on sharing new information.
Participants requested more open-ended survey questions and more studies about various aspects of information seeking in the DOC. Participants in this community are very forthcoming in what they need and want to see in research. Overall, participants stated that the questions seemed relevant to T1D and DOC usage. They reported gathering information first in preparation for making a decision about whether or not to go to the doctor. Participants reported that the DOC provided them with information on how to use their medical devices, including information about how to address treatment management for sick days, emergencies situations, and exercising.
An interesting phenomenon within these data (and generalized to this community) is that some DOC members indicated that they provide advice to others but they do not request it. Key quotes included: [ID 189]: “I wouldn’t make a treatment decision online with someone who I do not know as that could result in poor treatment. I have made suggestions once in a while or advised how I would treat myself in that situation”. and [ID 173]: “A woman had asked about using her libre [continuous glucose monitor] to make decisions on the insulin, she was new to it and was hesitant about how to treat. I gave her several personal examples and showed successes and failures. Others did the same thing. She decided to try small changes and let her doctor know which I also advised.” Another key quote: [ID 108]: I have a family in CO who has a son about the same age as my son who was having a hard time reach out to me direct, and I was able to help them get some things set while their son had the flu. It was a really good feeling and we have been friends for some time now. As for me, I haven’t had to ask for help on anything in a long time since I have done most of it on my own for so long.
This dynamic has important implications for the way that individuals with chronic disease seek health information. Specifically, existing information seeking theories do not take into account the role of information brokering that occurs in the online community. Furthermore, there is a difference between the types of information (by topic) that are being sought and how actively seeking advice versus passively seeking advice) they are being sought.
Participants had a wide range of experience and weekly commitment giving medical advice in the DOC. An example of this range was that some participants stated that they spent 0 minutes weekly providing advice in the DOC but others stated that they spent 6 to 7 hours giving advice in the DOC per week. Of note, one participant indicated: [ID 103] “Not much. While I appreciate anecdotal advice but I prefer medical information to come from my endocrinologist. I rarely SEEK out medical advice … that doesn’t prevent it from being offered to me though….” Several participants stated that they were recipients of unsolicited health advice in the DOC. Another identified: [ID 109]: “whatever time I don’t spend looking, I am helping”. Another participant stated, [ID 179]: “Currently, I am not seeking advice, at least not $100\%$ of the time. Sometimes I’m just reading and stumble upon advice that I find useful”.
## Discussion
The study examined seeking health information online and behavior engagement in the DOC, a prominent diabetes focused health community where peers provide multiple types of social support and broker information. The DOC assists with a variety of issues (including information gathering, with medical devices, promoting social support and connecting others). Social support sought after in the DOC included emotional support, encouragement to get a continuous glucose monitor, informational support and inspiration. Existing DOC literature does not examine the phenomenon of health information seeking so this research was critical for developing and further validating the scale.
The majority of participants expressed that the DOC had a positive impact on their physical health, stating improvements in self-care, exercise behaviors, nutrition, and access to healthcare and medication. Participants also reported receiving encouragement, feeling less alone, having an improved mental health status, feeling a sense of community, and normalizing the diabetes experience. Participants endorsed that the DOC had a positive impact on their mental health. Importantly, a handful of participants reported experiencing anxiety related to the DOC which appears to be connected to the behavior of comparing oneself to other members of the DOC.
Many participants engaged in health behavior resources. Participants sought advice about medication dosing and using insulin pumps and continuous glucose monitors. There were very few overlapping topics which garnered further support for the complexity of the needs in the DOC and the difficulty of disease management. Several participants reported not seeking information online but instead providing information online. These diverse responses show that seeking health information in the DOC is not for everyone but those who do seek the health information benefit greatly. The DOC is capable of providing important, tailored information and assistance.
Participants were very expressive in what was the most useful part of the DOC such as informational support where DOC members are exchanging advice about disease management. Participants also expressed the importance of social interaction and support where DOC members are interacting with other members and receiving social support during these interactions. The final theme was the sense of community experienced by members of the DOC. Participants expressed feeling part of a larger group where they do not feel judged and they related to other members while reaping benefits of said membership. Overall, these examples and themes provide powerful support that the DOC has a beneficial impact on the amount of social support that individuals with T1D are experiencing. At the core of this research, is the need to further understand how individuals with T1D are gaining health information in the DOC and the impact that this support has on their health.
Regarding how information seeking is occurring in the DOC, participants also reported following advice that already existed in social media such that they are not generating a new post to find an answer to their question. Instead, they are seeking existing posts where their health question has been answered. Most social media sites have a search mechanism that makes this fast and easy to accomplish. Importantly, members of the DOC also reported on the phenomenon of endorsing existing answers which impacts the trustworthiness of the information. This dynamic greatly challenges how online health information seeking is presently studied in the literature (using the existing scales with vignettes based on hypothetical situations). Of importance, participants stated that they began using specific types of durable medical equipment because of endorsements from DOC members.
This project provides a view of the “real world” perspective T1D management outside of the health clinic. The project also sought to clarify how members of the DOC seek health information and what they perceive to be the benefits of being a member. Prior research has suggested evidence of benefits of membership include emotional support and informational support [18]. Previous research has also suggested anecdotal evidence of benefits of membership include increased positive emotional experiences, increased positive attitudes towards T1D, and increased engagement in T1D management behaviors [1].
These findings provide support for the four key types of social support: emotional support (e.g., providing caring endearments when needed), informational support (e.g., providing advice about how much insulin to dose during exercise), instrumental support (e.g., providing insulin pump training to individuals who do not have the local training resources), and appraisal support (e.g., members make other feel “normal”). Findings also provided support for social provisions: guidance (e.g., advice about treatment decisions), reliable alliance (e.g., guarantees that others will be there in a stressful situation such as being without insulin or when an insulin pump breaks), reassurance of worth (e.g., recognition of one’s competence found during times struggling with blood sugar readings that are out of range), attachment (e.g., emotional closeness with group members and group as a whole), social integration (e.g., a sense of belonging to a group of social media acquaintances), and opportunity for nurturance (e.g., providing assistance to others).
## Future Directions
Future research should investigate health information seeking across different social media platforms including Instagram for review of health information sharing endorsed by influencers who are sponsored by pharmaceutical companies. Additional research should examine information seeking by caregivers of adolescents with T1D as much of T1D management is shared with family.
## Limitations
Due to limitations of online qualitative data collection, some qualitative responses were very brief and some participants did not answer some prompts at all. Future studies should delineate between seeking advice versus providing advice as many participants stated that they did not seek advice but instead offered it. The sample was cross-sectional and used convenience sampling. The sample was recruited from the DOC, which introduces the possibility of sampling bias. As to be expected, the online sample was mostly white, well-educated and female. A potential bias may be sampling of more active versus less active users in the DOC. Importantly, these results should not be generalized to other types of diabetes because each type of diabetes differs in its treatment [19].
## Conclusions
In conclusion, this project’s findings provide support for the relationships between seeking health information online, social support and T1D related health outcomes and behaviors. This project adds to the information seeking knowledge base by characterizing how individuals with T1D are using social media. With a better understanding of the roles of online social support and seeking health information online on disease management, this project serves as the first of several series of studies to improve usage of the DOC and facilitate constructions of interventions that encourage or discourage specific aspects of each behavior. Future research should seek to collect additional data to bolster validity and reliability for the developing scale. Currently, the scale is being tested in varying groups of the DOC. Despite many established psychosocial benefits to participating in online support groups and also physical benefits to the information being brokered in the online community, this community (and precisely, particular subgroups) may not benefit everyone with T1D.
## Data Availability Statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics Statement
The studies involving human participants were reviewed and approved by The University of Texas at El Paso - 1216875-2. The participants provided their written informed consent to participate in this study.
## Author Contributions
AH and OM conceived of the presented idea. AH developed the study and recruited participants. DG and IS verified the analytical methods. All authors contributed to the article and approved the submitted version.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Hilliard M, Sparling K, Hitchcock J, Oser T, Hood K. **The Emerging Diabetes Online Community**. *Curr Diabetes Rev* (2015) **11**. DOI: 10.2174/1573399811666150421123448
2. Korda H, Itani Z. **Harnessing Social Media for Health Promotion and Behavior Change**. *Health Promotion Pract* (2013) **14** 15-23.). DOI: 10.1177/1524839911405850
3. 3
American Diabetes Association. Statistics About Diabetes (2018). Available at: https://www.diabetes.org/resources/statistics/statistics-about-diabetes.. *Statistics About Diabetes* (2018)
4. Brown A. **42 Factors That Affect Blood Glucose?! A Surprising Update**. *Diatribe* (2018)
5. Collins SE, Lewis DM. **Social Media Made Easy: Guiding Patients to Credible Online Health Information and Engagement Resources**. *Clin Diabetes* (2013) **31**. DOI: 10.2337/diaclin.31.3.137
6. Fox S. **After Dr. Google: Peer-to-Peer Health Care**. *Pediatrics* (2013) **131**. DOI: 10.1542/peds.2012-3786K
7. Glanz K, Schwartz MD. **Stress, Coping, and Health Behavior**. *Health Behav Health education: Theory Research Pract* (2008) **4**
8. Lambert SD, Loiselle CG. **Health Information—Seeking Behavior**. *Qual Health Res* (2007) **17**. DOI: 10.1177/1049732307305199
9. Crook B, Stephens KK, Pastorek AE, Mackert M, Donovan EE. **Sharing Health Information and Influencing Behavioral Intentions: The Role of Health Literacy, Information Overload, and the Internet in the Diffusion of Healthy Heart Information**. *Health Communication* (2016) **31** 60-71. DOI: 10.1080/10410236.2014.936336
10. Frey E, Bonfiglioli C, Frawley J. **Self-Management: Diabetes Care Is Getting Social**. *BMJ* (2021) **373** 1. DOI: 10.1136/bmj.n1333
11. Litchman ML, Walker HR, Ng AH, Wawrzynski SE, Oser SM, Greenwood DA. **State of the Science: A Scoping Review and Gap Analysis of Diabetes Online Communities**. *J Diabetes Sci Technol* (2019) **13**. DOI: 10.1177/1932296819831042
12. Van Uden-Kraan CF, Drossaert CHC, Taal E, Seydel ER, van de Laar MAFJ. **Self-Reported Differences in Empowerment Between Lurkers and Posters in Online Patient Support Groups**. *J Med Internet Res* (2008) **10** e18. DOI: 10.2196/jmir.992
13. Rains SA, Keating DM. **The Social Dimension of Blogging About Health: Health Blogging, Social Support, and Well-Being**. *Communication Monogr* (2011) **78**. DOI: 10.1080/03637751.2011.618142
14. Carrier MA, Beverly EA. **Focus on the Positive: A Qualitative Study of Positive Experiences Living With Type 1 or Type 2 Diabetes**. *Clin Diabetes* (2021) **39**. DOI: 10.2337/cd20-0082
15. Johansen CB, Pouwer F, Beck-Nielsen H, Rothmann MJ. **Positive Facilitators of Diabetes Management in Emerging Adults With Type 1 Diabetes—A Qualitative Analysis of Blogs**. *Endocrinol Diabetes Metab* (2020) **3** e00161. DOI: 10.1002/edm2.161
16. Myneni S, Lewis B, Singh T, Paiva K, Kim SM, Cebula AV. **Diabetes Self-Management in the Age of Social Media: Large-Scale Analysis of Peer Interactions Using Semiautomated Methods**. *JMIR Med Inf* (2020) **8** e18441. DOI: 10.2196/18441
17. Braun V, Clarke V. **Using Thematic Analysis in Psychology**. *Qual Res Psychol* (2006) **3** 77-101. DOI: 10.1191/1478088706qp063oa
18. Greene JA, Choudhry NK, Kilabuk E, Shrank WH. **Online Social Networking by Patients With Diabetes: A Qualitative Evaluation of Communication With Facebook**. *J Gen Intern Med* (2011) **26**
19. **Standards of Medical Care in Diabetes—2021**. *Diabetes Care* (2021) **44**. DOI: 10.2337/dc21-S015
|
---
title: 'Barriers to Uptake of Open-Source Automated Insulin Delivery Systems: Analysis
of Socioeconomic Factors and Perceived Challenges of Caregivers of Children and
Adolescents With Type 1 Diabetes From the OPEN Survey'
authors:
- Antonia Huhndt
- Yanbing Chen
- Shane O’Donnell
- Drew Cooper
- Hanne Ballhausen
- Katarzyna A. Gajewska
- Timothée Froment
- Mandy Wäldchen
- Dana M. Lewis
- Klemens Raile
- Timothy C. Skinner
- Katarina Braune
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2022
pmcid: PMC10012142
doi: 10.3389/fcdhc.2022.876511
license: CC BY 4.0
---
# Barriers to Uptake of Open-Source Automated Insulin Delivery Systems: Analysis of Socioeconomic Factors and Perceived Challenges of Caregivers of Children and Adolescents With Type 1 Diabetes From the OPEN Survey
## Abstract
### Background
As a treatment option for people living with diabetes, automated insulin delivery (AID) systems are becoming increasingly popular. The #WeAreNotWaiting community plays a crucial role in the provision and distribution of open-source AID technology. However, while a large percentage of children were early adopters of open-source AID, there are regional differences in adoption, which has prompted an investigation into the barriers perceived by caregivers of children with diabetes to creating open-source systems.
### Methods
This is a retrospective, cross-sectional and multinational study conducted with caregivers of children and adolescents with diabetes, distributed across the online #WeAreNotWaiting online peer-support groups. Participants—specifically caregivers of children not using AID—responded to a web-based questionnaire concerning their perceived barriers to building and maintaining an open-source AID system.
### Results
56 caregivers of children with diabetes, who were not using open-source AID at the time of data collection responded to the questionnaire. Respondents indicated that their major perceived barriers to building an open-source AID system were their limited technical skills ($50\%$), a lack of support by medical professionals ($39\%$), and therefore the concern with not being able to maintain an AID system ($43\%$). However, barriers relating to confidence in open-source technologies/unapproved products and fear of digital technology taking control of diabetes were not perceived as significant enough to prevent non-users from initiating the use of an open-source AID system.
### Conclusions
The results of this study elucidate some of the perceived barriers to uptake of open-source AID experienced by caregivers of children with diabetes. Reducing these barriers may improve the uptake of open-source AID technology for children and adolescents with diabetes. With the continuous development and wider dissemination of educational resources and guidance—for both aspiring users and their healthcare professionals—the adoption of open-source AID systems could be improved.
## Introduction
There are about 1.2 million children and adolescents <20 years of age worldwide who live with type 1 diabetes [1]. To reduce their risks of acute and long-term complications, therapeutic guidelines recommend target hemoglobin A1c levels of <7,$0\%$ [2, 3]. However, not all children and adolescents achieve these glycemic targets. Methods for treating diabetes vary widely [4, 5]. Technology is evolving rapidly and continuously, which is significant in improving health conditions. Medical devices, mobile technology, cloud computing, and social media make it possible—especially for patients—to improve, co-design, and co-develop new treatments [6]. This possibility is particularly important for children and adolescents living with diabetes, as well as their caregivers and families [7].
Automated Insulin Delivery (AID) systems represent an important advance in diabetes therapy. Given the limitations in access to this technology, the #WeAreNotWaiting community has created so-called “Do-It-Yourself Artificial Pancreas Systems” (DIYAPS) or ‘open-source AID’ systems and made the resources needed to build them available via open-source platforms (7–10). These systems are not approved by regulatory bodies and thus are used by people with diabetes at their own risk. However, devices that are approved and commercially available are needed as components (e.g. insulin pumps and continuous glucose monitoring (CGM) systems). There are several types of open-source AID systems each with multiple different branches. While OpenAPS runs the algorithm on a Linux-based minicomputer, AndroidAPS, Loop, and FreeAPS are smartphone applications. Depending on the setup, additional hardware (e.g. OpenAPS rig, Riley-, Orange- or EmaLink) and software components (e.g. xDrip+, Nightscout) may also be required [11, 12]. The algorithms for automated insulin delivery adjust insulin dosing according to the user’s glycemic levels. Previous studies have shown that open-source AID systems have the potential to improve clinical outcomes in users of several age groups (e.g. better HbA1c-level and time in range (TIR)) (13–23). Moreover, they reduce the individual burden of living with diabetes, such as improving quality of life and sleep quality (14, 24–28). A profound understanding of pump therapy and CGM systems, but also technical literacy are needed to successfully build and use open-source AID (25, 29–32). The questions that arise include who is actually able to use and even create this technology and who would benefit from it [33, 34].
Few studies have examined the perceived barriers to adopting open-source AID solutions. Schipp et al. looked at the perceived challenges of adults during the set-up of their AID system [25]. However, this work and most other research reports almost exclusively focus on the experiences and emotions of people with diabetes who have already successfully built and are using an open-source system. Among members of online support groups such as “Looped” on Facebook ($$n = 28$$,323), there are a number of people with diabetes and caregivers of children who have not yet built and used open-source AID. O’Donnell et al. previously identified barriers perceived by adult non-users [35], but evidence is currently lacking concerning children and adolescents with diabetes and their caregivers. Therefore, it might be possible that the barriers to building and using open-source AID are not completely identified yet. To fill such an evidence gap, this paper refers to the caregivers of children living with diabetes. One of the challenges to addressing this gap is that to respond to questions about the barriers to using open-source AID systems, it is necessary to recruit caregivers who know about these systems and have some understanding of what it entails to build and maintain them. Therefore, this study aimed to recruit caregivers from the #WeAreNotWaiting community who meet these criteria. Clearly, with members of this well-informed and pro-active community, it is to be expected that there are biases with this sample. Hence, our results will not include the barrier of not knowing about the systems.
The overall aim of this study is to 1) investigate the barriers to scale-up open-source AID systems in caregivers of children and adolescents in the #WeAreNotWaiting community who are non-users of open-source AID and 2) analyze the participants’ socioeconomic status in relation to the perceived barriers.
## Study Design
This survey was part of a large retrospective, multinational, web-based cross-sectional study conducted from September to November 2020 with users and non-users of open-source AID within the #WeAreNotWaiting community. Two questionnaires, titled “Your Thoughts about DIYAPS” (DIWHYnot) and “About you and your child” (socioeconomic factors) were distributed to caregivers of children with diabetes who were still non-users of open-source AID.
## Survey Tool
Questionnaires were designed by an interdisciplinary team of researchers living with type 1 diabetes and were both users and non-users of open-source AID [35]; some researchers had used open-source AID for several years, some were in the process of uptaking systems, and others were not interested in using open-source AID. TF—a non-user—provided statements about challenges regarding the set-up, which were reviewed and completed by users (SO, DL, KB, MW) and non-users (KAG) to generate a final list of items. The ‘DIWHYnot’ questionnaire comprised of a combination of check-box items with comments, and questions on a 5-point Likert scale (“Strongly Agree”, “Agree”, “Neither Agree or Disagree’’, “Disagree” and “Strongly Disagree”); respondents were able to choose “Other”, “I don’t know” or “I’d rather not say” in response to the questions (Appendix A). The questionnaire applied branching logic to address progressively more specific questions. “ About you and your child” used mostly check-box items and open-field inputs to collect information on socio-economic factors; respondents were again capable of answering “I’d rather not say”, “Other”, “None of the above’’ or “I don’t know”, allowing each participant to be included in the statistics (Appendix B).
## Participants and Recruitment
Caregivers (e.g. a parent, family member, or legal guardian) of children and adolescents under the age of 18 who are living with diabetes were eligible for participation. The participants were recruited via Facebook groups including the multinational “Looped” groups, “AndroidAPS users”, “CGM in the Cloud” and “Nightscout Germany”; through the OPEN website; and social media accounts such as “Diabetes Daily”. The survey was conducted using the REDCap electronic data capture tool hosted by Charité – Universitätsmedizin Berlin. Ethical approval for the survey—including all questionnaires—was granted by the Life Sciences Human Research Ethics Committee at University College Dublin (LS-20-37).
## Data Analysis
After data cleaning, analyses were conducted using IBM SPSS Statistics 27 (International Business Machines Corporation, Armonk, NY, United States) and Microsoft Office (Microsoft Corporation, Redmond, WA, United States). Validity and internal consistency (e.g. factor analysis, Cronbach’s alpha) tests of specific survey items were performed followed by descriptive and inferential analyses (e.g. Levene’s test, independent samples t-test).
## Participant Characteristics
Of the 1052 total participants of the OPEN study, 56 were caregivers of children with diabetes who were not using open-source AID at the time of data collection (Supplementary Figure 1). Responses from 49 participants were included in the analysis of socioeconomic factors (of children and caregivers). Overall, $59.2\%$ of the children were male with a mean age of 11 years (range: 1-18 years, SD: 3 years). The participants were from 13 different countries of which $67.3\%$ were from Germany ($$n = 12$$), Denmark ($$n = 11$$), the United Kingdom ($$n = 5$$), and the United States ($$n = 5$$). Of the participants, $87.8\%$ described the ethnicity of the children as “White”. $63.3\%$ of the caregivers were employed either full- or part-time, mostly in the science sector ($32.6\%$), most commonly with educational qualifications of a Master’s ($34.7\%$) or Bachelor’s degree ($30.6\%$). The majority reported annual household income ranged from 100 000 to 199 999 US dollars ($32.7\%$) and 50 000 to 99 999 US dollars ($20.4\%$). Supplementary Table 1 summarizes the demographic data in detail. Most participants ($84.6\%$; $$n = 22$$/26) expressed they wanted to learn more about open-source AID, especially regarding the support they would get if they decided to build a system. More than half of them ($64.3\%$; $$n = 18$$/28) have not yet created a system but could imagine doing so under certain conditions. Among those who could imagine building an open-source AID system, most ($30.4\%$; $$n = 17$$/56) were interested in “Loop” (app for Apple iPhones). While $61.5\%$ ($$n = 16$$/26) were convinced and wanted to create a system, a smaller group ($21.4\%$; $$n = 6$$/28) were already in the process of setting up a system but had not yet used it. $28.6\%$ ($$n = 16$$/56) did not report out-of-pocket expenses for the required diabetes equipment, while the remaining participants pay up to 50 USD per month ($7.1\%$; $$n = 4$$/56), rarely more. Expenses for insulin were reported most often ($14.3\%$; $$n = 8$$/56), followed by CGM sensors ($10.7\%$; $$n = 6$$/56). When asked how the participants had heard about open-source AID, the majority responded “I have heard of it through social media” ($48.2\%$; $$n = 27$$/56).
## Types of Barriers
Only a few respondents ($11.1\%$; $$n = 3$$/27) perceived the necessary components were too expensive. $33.3\%$ ($$n = 9$$/27) were interested in building an open-source AID but did not know where to source some of the components, especially additional components, such as the RileyLink and OpenAPS rig ($77.8\%$; $$n = 7$$/9) and loopable pumps ($66.7\%$; $$n = 6$$/9).
Overall, the result of the internal consistency analysis was acceptable (Cronbach’s alpha=0.741). 18 out of 20 items questioned all perceived barriers except for the procurement and the costs of necessary components to use open-source AID (Cronbach’s alpha=0.730). The reliability of the remaining 18 items was improved by deleting several statements, with Cronbach’s alpha increasing to 0.807. Following the elimination of the items “insufficient expertise of diabetes teams”, the “missing knowledge about pump therapy” and the “imagination to carry the required equipment”, a good internal consistency was achieved. The remaining 15 items were examined through exploratory factor analysis using principal component analysis and the Varimax rotation method, which indicated the point of inflection on the screen plot was at three factors and this generated a simple solution (factors only loading > 0.4 on one factor (Table 1). As the results suggested, the 15 items can be reduced to three components (cumulative proportion=$60.71\%$): Dimension 1 “building and maintenance” of a system, Dimension 2 “therapy knowledge and trust in technology” and Dimension 3 “support and liability”. The item “My child is currently using commercial automated delivery systems.” ( loading<0.4) did not fit into any of the dimensions. Thus, our final scale was best represented by the remaining 14 items in Table 1.
**Table 1**
| Items | Dimension 1 | Dimension 2 | Dimension 3 |
| --- | --- | --- | --- |
| I don’t have sufficient knowledge of CGM therapy. | | 0.619 | |
| My child is currently using commercial automated delivery systems. | | | |
| I don’t trust machines/technologies to take over the control of my child’s diabetes. | | 0.761 | |
| I don’t trust open-source technology. | | 0.709 | |
| I don’t trust products that are not approved by a regulatory body. | | 0.737 | |
| I don’t have the necessary programming knowledge to build the software on my own. | 0.867 | | |
| I can find help to build the DIYAPS but I am scared I won’t be able to maintain it. | 0.773 | | |
| I am afraid we might lose the support of my child’s healthcare provider if we start looping. | | | 0.647 |
| I am afraid we might lose my child’s health insurance if we start looping. | | | 0.776 |
| I feel it would increase my level of responsibility and I don’t want that. | | 0.522 | |
| I feel it would increase my level of liability and I don’t want that. | | | 0.503 |
| I don’t have the energy to do it myself. | 0.917 | | |
| I don’t have the time to do it myself. | 0.833 | | |
| I feel that the DIYAPS expertise and resources are too overwhelming to understand. | 0.871 | | |
| My child’s diabetes team discourages me from building a DIYAPS. | | | 0.845 |
Regarding “building and maintenance” of the open-source AID system, half of the participants reported ($$n = 14$$/28) that they did not perceive having the skills needed to build it. $42.9\%$ ($$n = 12$$/28) knew they could find help to set it up, but were not sure if they would be able to maintain the system. Lack of time ($33.3\%$; $$n = 9$$/27) and too little energy ($32.1\%$; $$n = 9$$/28) were concerns, in addition to that resources are too overwhelming to understand ($35.7\%$; $$n = 10$$/28). In terms of “therapy knowledge and trust in technology”, respondents were most likely to fear having to take on additional responsibility ($21.4\%$; $$n = 6$$/28). Insufficient knowledge about CGM was reported by $14.3\%$ ($$n = 4$$/28). Only a minority of the respondents ($7.1\%$; $$n = 2$$/28) reported a lack of trust in machines and technologies to take over the control of diabetes in general. Similarly, only one respondent reported a lack of trust in products that have not been approved by a regulatory body. As for “support and liability”, by far the biggest concern was a potential loss of support by the healthcare provider ($39.3\%$; $$n = 11$$/28), followed by discouraging the uptake of an open-source AID by the diabetes team ($25\%$; $$n = 7$$/28). There was less agreement on both the fear of losing health insurance if they start looping and the worry that liability would increase ($10.7\%$; $$n = 3$$/28). Additionally, it is important to mention that $44.4\%$ ($$n = 12$$/27) reported no available support from healthcare professionals (HCP) of the diabetes care team due to their limited expertise in diabetes technology in general (Figure 1).
**Figure 1:** *N=56; responses to statements regarding interest in building an open-source AID (“I would be interested in building a DIYAPS, but…”). Participants rated statements on a 5-point Likert scale (strongly agree, agree, neither agree nor disagree, disagree, strongly disagree). The responses were classified into three dimensions using principal component analysis and are labeled accordingly (*Building and Maintenance, **Knowledge and Trust in Technology, ***Support and Liability, † were excluded due to reduced reliability, ‡ was excluded after principal component analysis due to insufficient loading).*
## Encouragement
It is remarkable that many of the respondents would be encouraged to set up an AID system if their decision was officially supported. For example, $88.9\%$ ($$n = 24$$/27) confirmed “strongly agree” or “agree” that they would be motivated to start an open-source AID if their HCP recommended it. A similar number ($80.8\%$; $$n = 21$$/26) would be convinced if professional diabetes associations such as the International Diabetes Federation supported their use. Both the support of diabetes care teams ($77.8\%$; $$n = 21$$/27) and increased uptake in open-source AID in healthcare in general ($74.1\%$; $$n = 20$$/27) would encourage some caregivers to consider taking up a system. However, for slightly more than half of the participants ($55.6\%$; $$n = 15$$/27), it is also important that there is a company to provide warranty and support in case of technical errors.
## Differences by Education Level and Household Income
To determine whether perceptions of different barriers are related to educational attainment, participants with Bachelor’s and Master’s degrees, as well as different household income groups were compared (Table 2).
**Table 2**
| Unnamed: 0 | Education | Education.1 | Education.2 | Annual income | Annual income.1 | Annual income.2 |
| --- | --- | --- | --- | --- | --- | --- |
| Item | Bachelor’s degree | Master’s degree | t-test | USD50 000 to 99 999 | USD100 000 to 199 999 | t-test |
| …I want to build a DIYAPS system. | 7 | 10 | 0.398 | 8 | 10 | 0.358 |
| …it is too expensive to purchase all the components. | 8 | 10 | 0.654 | 8 | 10 | 0.833** |
| …I don’t know where to source all the components. | 8 | 10 | -0.105 | 8 | 10 | -0.274 |
| I don’t have sufficient knowledge of CGM therapy. | 8 | 11 | -0.589 | 8 | 10 | 0.147 |
| I am currently using commercial automated delivery systems. | 7 | 10 | 1.286** | 8 | 9 | 1.039 |
| I don’t trust machines/technologies to take over the control of my diabetes. | 8 | 11 | -0.408 | 8 | 10 | -2.074 |
| I don’t trust open-source technology. | 8 | 11 | -1.657 | 8 | 10 | 0.000 |
| I don’t trust products that are not approved by a regulatory body. | 8 | 11 | -0.986 | 8 | 10 | 0.622 |
| I don’t have the necessary programming knowledge to build the software on my own. | 8 | 11 | 1.050 | 8 | 10 | 1.805 |
| I can find help to build the DIYAPS but I am scared I won’t be able to maintain it. | 8 | 11 | 0.691 | 8 | 10 | 2.855* |
| I am afraid I might lose the support of my healthcare provider if I start looping. | 8 | 11 | -1.361 | 8 | 10 | 1.155 |
| I am afraid I might lose my health insurance if I start looping. | 8 | 11 | 0.975 | 8 | 10 | 0.457 |
| I feel it would increase my level of responsibility and I don’t want that. | 8 | 11 | -2.904* | 8 | 10 | 0.805 |
| I feel it would increase my level of liability and I don’t want that. | 8 | 11 | -1.095** | 8 | 10 | 1.042 |
| I don’t have the energy to do it myself. | 8 | 11 | -0.263 | 8 | 10 | 0.041 |
| I don’t have the time to do it myself. | 8 | 10 | 0.147 | 8 | 9 | -0.178 |
| I feel that the DIYAPS expertise and resources are too overwhelming to understand. | 8 | 11 | -0.392 | 8 | 10 | -1.452 |
| My diabetes team discourages me from building DIYAPS. | 8 | 11 | -1.288 | 8 | 10 | 0.989 |
| How much do you have to pay for the diabetes supplies monthly? | 8 | 11 | 1.966** | 8 | 10 | 0.300 |
The worry about having to take on additional responsibility when using open-source AID differed significantly between caregivers with a Bachelor’s and a Master’s degree (T-test, T=-2.904, df=17, $$p \leq 0.01$$), with participants with a Master’s degree being more aware. No significant differences were found regarding the caregivers’ motivation to build open-source AID (T-test, $T = 0.398$, df=15, $$p \leq 0.69$$) or concerns about sourcing components (T-test, T=-0.105, df=16, $$p \leq 0.92$$). Differences were also found between the income groups of USD 50 000 to 99 999 and USD 100 000 to 199 999 per year. The worry of not being able to regularly maintain the open-source AID system once they have successfully built it was perceived as more significant for caregivers with a higher income (T-test, $T = 2.855$, df=16, $$p \leq 0.01$$). Non-significant differences were found for insufficient programming skills (T-test, $T = 1.805$, df=16, $$p \leq 0.09$$) and fear of losing support from the healthcare provider (T-test, $T = 1.155$, df=16, $$p \leq 0.27$$).
## Discussion
Non-users of open-source AID reported several structural and individual barriers to the adoption of open-source AID. Structural barriers concerned the sourcing of compatible insulin pumps and additional components. Major individual barriers were limited perceived technical skills such as programming knowledge, limited support by medical professionals, and therefore the concerns of not being able to build and maintain the AID system by themselves. However, it was neither the confidence in the open-source technologies and lack of their regulatory approval nor the fear of digital technologies taking over the control of diabetes management that prevented non-users from creating and using an open-source AID system. Except for the two significant characteristics of annual household income and highest educational degree, socioeconomic status did not have a significant impact on the perceived barriers.
The structural problem of obtaining suitable insulin pumps could best be explained by the fact that not all compatible models are available via prescription. A compatible insulin pump refers to one that can interoperate with the algorithm and receive commands to adjust insulin delivery. Although the number of compatible insulin pump models has increased in recent years, some “loopable” pumps are only available on prescription in select countries, insurances often set time limits when a prescription can be renewed, and the availability of older out-of-warranty pump models, e.g. traded second-hand via online platforms, is limited. A previous study on barriers to the adoption of insulin pumps in Ireland has identified some people with diabetes having difficulties in orienting and understanding the health systems and their reimbursement principles [36]. In terms of individual challenges, the self-perceived insufficient programming skills emerged as very relevant.
Finally, non-users were concerned about their ability to maintain and service the system on their own. Previous studies describing the experiences of those who have successfully set up open-source AID have shown that peer-support can help overcome this barrier [25, 37] with the support of experienced or technically versed community members. Their fears and worries could be alleviated and their self-confidence and determination could be strengthened through the achievement of successfully setting up an AID system. Therefore, connecting with other members of the #WeAreNotWaiting community online or in-person, as well as utilizing the available resources (e.g., online documentation and tutorials) could help close pre-existing knowledge gaps.
Similarly, respondents reported not receiving the desired support from their diabetes care teams, e.g., as they have limited necessary expertise in diabetes technology in general and open-source AID in particular. Medical professionals are important gatekeepers when it comes to access to open-source AID and, in fact, all diabetes technology. Many families with children with diabetes experience difficulties in access to AID systems, i.e., because these are not yet approved in their countries, are not approved for children of a particular age or are not reimbursed. It may take many years until AID technology is fully affordable and accessible for everyone. Until such time, open-source AID will continue to “fill the gap” for some people with diabetes in accessing this life-enhancing technology and therefore merits the support of medical professionals as well as other stakeholders in the diabetes community. Previous work has highlighted that HCPs are caught in a dilemma between the uncertainty regarding liability, the lack of regulatory approval of open-source AID, and supporting the choice and best interest of their patients [12, 33, 38]. Meanwhile, individuals with diabetes may also face an associated dilemma between the advantages of using an AID system and the risk of losing support from HCPs. It seems that the open-source innovations have gained more acceptance among some HCPs, but not the majority of them. Due to the trend, more knowledge about the technology is available and can be used for supporting patients who are interested in applying an AID system [39]. Therefore, changing medical guidelines to support open-source AID could help to reduce concerns of people with diabetes and remove the two barriers “insufficient expertise of diabetes teams” and the associated “lack of support”. This point of view has already been supported by other authors who looked at the perspectives of HCPs (34, 39–42). A recently published international consensus statement provided practical guidance to HCPs and specifically addressed professional educational aspects but also ethical and legal issues [12]. For children and adolescents specifically, the consensus group recommended that the child’s welfare should always be considered by HCPs and caregivers who are setting up open-source AID for children, with the child’s assent and engagement [12]. Further research should investigate the experiences and thoughts of HCPs and particularly address the challenges of procuring necessary devices (insulin pumps, CGM) via prescription (12, 39, 43–47).
To the best of the authors’ knowledge, this is the first study that is determining the perceived challenges in detail that are seen by caregivers of children and adolescents with diabetes when setting up an open-source AID. Previous studies have specifically examined the experiences of adult users (less so children) with open-source AID [14, 17, 25, 37, 48, 49]. The fact that most of the involved researchers have personal experience with open-source AID as active users in addition to their professional roles, as well as the involvement of non-users in the study design underlines the public and patient engagement as a strength of this paper. Of further strength is the multinational character of this study. Nevertheless, several limitations apply as the sample size of 56 participants is relatively small compared to other sub-cohorts that participated in the OPEN survey. Furthermore, the participants identified predominantly as “White”, based in North America and Europe, and mostly had a high socioeconomic status. The number of participants with an educational level lower than a Bachelor’s degree was too small to be included in the factor analysis. Therefore, the sample is not representative of all caregivers of children with diabetes who are not using an open-source AID system, and the results may not reflect the non-user population of developing and emerging countries [50]. Finally, based on the results of the reliability and validity tests, the items of the questionnaires could be adapted and improved for further surveys. Given the smaller sample size and socioeconomic background of this specific population, it would be of interest to investigate barriers in non-users outside the #WeAreNotWaiting community. Previous studies described motivations, enablers, and sources of support in open-source AID users (adults and caregivers of children) [14, 25, 27]. While this paper describes barriers that prevent non-users from building open-source AID, it may be equally important to investigate what might encourage them to do so.
Lastly, as commercially developed AID systems have recently become available in select countries and can be made available via prescription, it would be of interest to investigate barriers to uptake with respect to commercial AID systems and outside the context of the #WeAreNotWaiting community as well.
As part of a retrospective, multinational, web-based cross-sectional study of the #WeAreNotWaiting community, this work has identified caregivers’ challenges for uptaking an open-source AID system. In order to increase the distribution of open-source AID, using online resources and community peer-support could be useful and complement support from medical professionals. Sharing problems that occurred during the build and use of open-source AID, and how these were encountered, could be insightful to aspiring users. The current open-source AID documentation already includes a comprehensive list of build errors and solution strategies. The findings of this study could help the #WeAreNotWaiting community to further extend these resources to better meet the needs of current non-users. In addition, providing educational resources to HCPs, such as the recently published consensus statement, could also help care teams to understand and better support current and future open-source AID users. Finally, regional differences and limitations in the availability of insulin pumps and CGM systems as AID components should be addressed by manufacturers, regulators, and policymakers. If access to diabetes technology would be more equal, many more people with diabetes would be able to benefit from digital innovations.
## Data Availability Statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics Statement
The studies involving human participants were reviewed and approved by Life Sciences Human Research Ethics Committee at University College Dublin. The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
TF, KG, SO’D, DL, MW and KB created the survey design. Ethics approval was sought by SO’D. AH, YC, DC, SO’D, HB and KB processed and analyzed the data. AH and KB wrote the initial draft of the manuscript. All co-authors had access to the full data set, and have critically reviewed and revised the manuscript, approved the final version of the manuscript and can confirm the integrity of the study.
## Funding
The authors declare that this study received funding from European Commission's Horizon 2020 Research and Innovation Program under the Marie-Sklodowska-Curie Action Research and Innovation Staff Exchange (RISE) grant agreement number 823902, the DFG-funded Digital Clinician Scientist Program of the Berlin Institute of Health (BIH), and the SPOKES Wellcome Trust Translational Partnership Program. The funders was not involved in the study design, collection, analysis, interpretation of data, the writing of this article or the decision to submit it for publication.
## Conflict of Interest
Authors HB, KG and TF were employed by company Dedoc Labs GmbH.
The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary Material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcdhc.2022.876511/full#supplementary-material
## References
1. Sun H, Saeedi P, Karuranga S, Pinkepank M, Ogurtsova K, Duncan BB. **IDF Diabetes Atlas: Global, Regional and Country-Level Diabetes Prevalence Estimates for 2021 and Projections for 2045**. *Diabetes Res. Clin. Pract.* (2022.0) **183** 109119. DOI: 10.1016/j.diabres.2021.109119
2. DiMeglio LA, Acerini CL, Codner E, Craig ME, Hofer SE, Pillay K. **ISPAD Clinical Practice Consensus Guidelines 2018: Glycemic Control Targets and Glucose Monitoring for Children, Adolescents, and Young Adults With Diabetes**. *Pediatr. Diabetes* (2018.0) **19 Suppl 27**. DOI: 10.1111/pedi.12737
3. Draznin B, Aroda VR, Bakris G, Benson G. **6. Glycemic Targets: Standards of Medical Care in Diabetes-2022**. *Diabetes Care* (2022.0) **45**. DOI: 10.2337/dc22-S006
4. Sherr JL, Tauschmann M, Battelino T, de Bock M, Forlenza G, Roman R. **ISPAD Clinical Practice Consensus Guidelines 2018: Diabetes Technologies**. *Pediatr. Diabetes* (2018.0) **19 Suppl 27**. DOI: 10.1111/pedi.12731
5. Danne T, Phillip M, Buckingham BA, Jarosz-Chobot P, Saboo B, Urakami T. **ISPAD Clinical Practice Consensus Guidelines 2018: Insulin Treatment in Children and Adolescents With Diabetes**. *Pediatr. Diabetes* (2018.0) **19 Suppl 27**. DOI: 10.1111/pedi.12718
6. Lee JM, Hirschfeld E, Wedding J. **A Patient-Designed Do-It-Yourself Mobile Technology System for Diabetes: Promise and Challenges for a New Era in Medicine**. *JAMA* (2016.0) **315**. DOI: 10.1001/jama.2016.1903
7. 7
LoopDocs. Available at: https://loopkit.github.io/loopdocs/ (Accessed February 7, 2022).. *LoopDocs*
8. 8
Welcome to the AndroidAPS Documentation — AndroidAPS 3.0 Documentation. Available at: https://androidaps.readthedocs.io/ (Accessed February 7, 2022).. *Welcome to the AndroidAPS Documentation — AndroidAPS 3.0 Documentation*
9. 9
Welcome to Openaps’s Documentation! — OpenAPS 0.0.0 Documentation. Available at: https://openaps.readthedocs.io/ (Accessed February 7, 2022).. *Welcome to Openaps’s Documentation! — OpenAPS 0.0.0 Documentation.*
10. 10LoopWorkspace. Github.
11. 11
Project N. Welcome to Nightscout (2019). Available at: http://www.nightscout.info/.. *Welcome to Nightscout* (2019.0)
12. Braune K, Lal RA, Petruželková L, Scheiner G, Winterdijk P, Schmidt S. **Open-Source Automated Insulin Delivery: International Consensus Statement and Practical Guidance for Health-Care Professionals**. *Lancet Diabetes Endocrinol* (2022.0) **10** 58-74. DOI: 10.1016/S2213-8587(21)00267-9
13. Knoll C, Peacock S, Wäldchen M, Cooper D, Aulakh SK, Raile K. **Real-World Evidence on Clinical Outcomes of People With Type 1 Diabetes Using Open-Source and Commercial Automated Insulin Dosing Systems: A Systematic Review**. *Diabetes Med.* (2021.0). DOI: 10.1111/dme.14741
14. Braune K, Gajewska KA, Thieffry A, Lewis DM, Froment T, O’Donnell S. **Why #WeAreNotWaiting—Motivations and Self-Reported Outcomes Among Users of Open-Source Automated Insulin Delivery Systems: Multinational Survey**. *J. Med. Internet Res.* (2021.0) **23**. DOI: 10.2196/25409
15. Braune K, O’Donnell S, Cleal B, Lewis D, Tappe A, Willaing I. **Real-World Use of Do-It-Yourself Artificial Pancreas Systems in Children and Adolescents With Type 1 Diabetes: Online Survey and Analysis of Self-Reported Clinical Outcomes**. *JMIR Mhealth Uhealth* (2019.0) **7**. DOI: 10.2196/14087
16. Street TJ. **Review of Self-Reported Data From UK Do-It-Yourself Artificial Pancreas System (DIYAPS) Users to Determine Whether Demographic of Population Affects Use or Outcomes**. *Diabetes Ther.* (2021.0) **12**. DOI: 10.1007/s13300-021-01071-5
17. Gawrecki A, Zozulinska-Ziolkiewicz D, Michalak MA, Adamska A, Michalak M, Frackowiak U. **Safety and Glycemic Outcomes of Do-It-Yourself AndroidAPS Hybrid Closed-Loop System in Adults With Type 1 Diabetes**. *PloS One* (2021.0) **16**. DOI: 10.1371/journal.pone.0248965
18. Petruzelkova L, Jiranova P, Soupal J, Kozak M, Plachy L, Neuman V. **Pre-School and School-Aged Children Benefit From the Switch From a Sensor-Augmented Pump to an AndroidAPS Hybrid Closed Loop: A Retrospective Analysis**. *Pediatr. Diabetes* (2021.0). DOI: 10.1111/pedi.13190
19. Kesavadev J, Srinivasan S, Saboo B, Krishna BM, Krishnan G. **The Do-It-Yourself Artificial Pancreas: A Comprehensive Review**. *Diabetes Ther.* (2020.0) **11**. DOI: 10.1007/s13300-020-00823-z
20. Petruzelkova L, Soupal J, Plasova V, Jiranova P, Neuman V, Plachy L. **Excellent Glycemic Control Maintained by Open-Source Hybrid Closed-Loop AndroidAPS During and After Sustained Physical Activity**. *Diabetes Technol. Ther.* (2018.0) **20**. DOI: 10.1089/dia.2018.0214
21. Patel R, Crabtree TSJ, Taylor N, Langeland L, Gazis AT, Mendis B. **Safety and Effectiveness of do-it-Yourself Artificial Pancreas System Compared With Continuous Subcutaneous Insulin Infusions in Combination With Free Style Libre in People With Type 1 Diabetes**. *Diabetes Med.* (2022.0). DOI: 10.1111/dme.14793
22. Lum J, Bailey R, Barnes-Lomen V, Naranjo D, Hood K, Lal RA. **A Real-World Prospective Study of the Safety and Effectiveness of the Loop Open Source Automated Insulin Delivery System**. *Diabetes Technol. Ther.* (2020.0). DOI: 10.1089/dia.2020.0535
23. Choi MS, Lee S, Kim J, Kim G, Park SM, Kim JH. **Do-It-Yourself Open Artificial Pancreas System in Children and Adolescents With Type 1 Diabetes Mellitus: Real-World Data**. *Diabetes Metab. J.* (2021.0). DOI: 10.4093/dmj.2021.0011
24. Wu Z, Luo S, Zheng X, Bi Y, Xu W, Yan J. **Use of a do-it-Yourself Artificial Pancreas System Is Associated With Better Glucose Management and Higher Quality of Life Among Adults With Type 1 Diabetes**. *Ther. Adv. Endocrinol. Metab.* (2020.0) **11** 2042018820950146. DOI: 10.1177/2042018820950146
25. Schipp J, Skinner T, Holloway E, Scibilia R, Langstrup H, Speight J. **How Adults With Type 1 Diabetes Are Navigating the Challenges of Open-Source Artificial Pancreas Systems: A Qualitative Study**. *Diabetes Technol. Ther.* (2021.0). DOI: 10.1089/dia.2020.0652
26. Suttiratana S, Wong J, Lanning MS, Dunlap AR, Hanes S, Hood K. **Qualitative Study of User Experiences With Loop, an Open-Source Automated Insulin Delivery (AID) System**. *Diabetes Technol. Ther.* (2022.0). DOI: 10.1089/dia.2021.0485
27. Braune K, Krug N, Knoll C, Ballhausen H, Thieffry A, Chen Y. **Emotional and Physical Health Impact in Children and Adolescents and Their Caregivers Using Open-Source Automated Insulin Delivery: Qualitative Analysis of Lived Experiences**. (2021.0). DOI: 10.2139/ssrn.3946995
28. Cleal B, Braune K, O’donnell S, Hauck B, Lewis DM, Skinner TC. **78-LB: Detailing the Experiences of People With Diabetes Using Do-It-Yourself Artificial Pancreas Systems—Qualitative Analysis of Responses to Open-Ended Items in an International Survey**. *Diabetes* (2019.0) **68**. DOI: 10.2337/db19-78-LB
29. Asarani NAM, Reynolds AN, Elbalshy M, Burnside M, de Bock M, Lewis DM. **Efficacy, Safety, and User Experience of DIY or Open-Source Artificial Pancreas Systems: A Systematic Review**. *Acta Diabetol.* (2020.0). DOI: 10.1007/s00592-020-01623-4
30. Lewis D. **History and Perspective on DIY Closed Looping**. *J. Diabetes Sci. Technol.* (2019.0) **13**. DOI: 10.1177/1932296818808307
31. Staszak W, Chromik J, Braune K, Arnrich B. **Impact of Custom Features of Do-It-Yourself Artificial Pancreas Systems (DIYAPS) on Glycemic Outcomes of People With Type 1 Diabetes**. *Conf. Proc. IEEE Eng. Med. Biol. Soc.* (2021.0) **2021**. DOI: 10.1109/EMBC46164.2021.9629487
32. Palmer W, Greeley SAW, Letourneau-Freiberg LR, Naylor RN. **Using a Do-It-Yourself Artificial Pancreas: Perspectives From Patients and Diabetes Providers**. *J. Diabetes Sci. Technol.* (2020.0) **14**. DOI: 10.1177/1932296820942258
33. Barnard KD, Ziegler R, Klonoff DC, Braune K, Petersen B, Rendschmidt T. **Open Source Closed-Loop Insulin Delivery Systems: A Clash of Cultures or Merging of Diverse Approaches**. *J. Diabetes Sci. Technol.* (2018.0) **12**. DOI: 10.1177/1932296818792577
34. Burnside M, Crocket H, Mayo M, Pickering J, Tappe A, de Bock M. **Do-It-Yourself Automated Insulin Delivery: A Leading Example of the Democratization of Medicine**. *J. Diabetes Sci. Technol.* (2020.0) **14**. DOI: 10.1177/1932296819890623
35. O’Donnell S, Lewis D, Marchante Fernández M, Wäldchen M, Cleal B, Skinner T. **Evidence on User-Led Innovation in Diabetes Technology (The OPEN Project): Protocol for a Mixed Methods Study**. *JMIR Res. Protoc.* (2019.0) **8**. DOI: 10.2196/15368
36. Gajewska KA, Biesma R, Bennett K, Sreenan S. **Barriers and Facilitators to Accessing Insulin Pump Therapy by Adults With Type 1 Diabetes Mellitus: A Qualitative Study**. *Acta Diabetol.* (2021.0) **58** 93-105. DOI: 10.1007/s00592-020-01595-5
37. Ahmed SH, Ewins DL, Bridges J, Timmis A, Payne N, Mooney C. **Do-It-Yourself (DIY) Artificial Pancreas Systems for Type 1 Diabetes: Perspectives of Two Adult Users, Parent of a User and Healthcare Professionals**. *Adv. Ther.* (2020.0) **37**. DOI: 10.1007/s12325-020-01431-w
38. de Bock M. **The “Do it Yourself” Type 1 Diabetes Dilemma for Medical Practitioners**. *Intern. Med. J.* (2019.0) **49**. DOI: 10.1111/imj.14286
39. Roberts JTF, Moore V, Quigley M. **Prescribing Unapproved Medical Devices? The Case of DIY Artificial Pancreas Systems**. *Med. Law Int.* (2021.0) **21** 42-68. DOI: 10.1177/0968533221997510
40. Marshall DC, Holloway M, Korer M, Woodman J, Brackenridge A, Hussain S. **Do-It-Yourself Artificial Pancreas Systems in Type 1 Diabetes: Perspectives of Two Adult Users, a Caregiver and Three Physicians**. *Diabetes Ther.* (2019.0) **10**. DOI: 10.1007/s13300-019-00679-y
41. Shaw D, Crabtree TSJ, Hammond P, McLay A, Wilmot EG. **The DIY Artificial Pancreas System: An Ethical Dilemma for Doctors**. *Diabetes Med.* (2020.0). DOI: 10.1111/dme.14270
42. Crabtree TSJ, Choudhary P, Lumb A, Hammond P, McLay A, Campbell F. **Association of British Clinical Diabetologists, Diabetes Technology Network UK and Association of Children’s Diabetes Clinicians Survey of UK Healthcare Professional Attitudes Towards Open-Source Automated Insulin Delivery Systems**. *Diabetes Ther.* (2022.0). DOI: 10.1007/s13300-022-01203-5
43. Shepard JA, Breton M, Nimri R, Roberts JTF, Street T, Klonoff D. **User and Healthcare Professional Perspectives on Do-It-Yourself Artificial Pancreas Systems: A Need for Guidelines**. *J. Diabetes Sci. Technol.* (2020.0) 1932296820957728. DOI: 10.1177/1932296820957728
44. DeSalvo DJ, Noor N, Xie C, Corathers SD, Majidi S, McDonough RJ. **Patient Demographics and Clinical Outcomes Among Type 1 Diabetes Patients Using Continuous Glucose Monitors: Data From T1D Exchange Real-World Observational Study**. *J. Diabetes Sci. Technol.* (2021.0) 19322968211049783. DOI: 10.1177/19322968211049783
45. Addala A, Hanes S, Naranjo D, Maahs DM, Hood KK. **Provider Implicit Bias Impacts Pediatric Type 1 Diabetes Technology Recommendations in the United States: Findings From The Gatekeeper Study**. *J. Diabetes Sci. Technol.* (2021.0) **15**. DOI: 10.1177/19322968211006476
46. Lawton J, Kirkham J, Rankin D, White DA, Elliott J, Jaap A. **Who Gains Clinical Benefit From Using Insulin Pump Therapy? A Qualitative Study of the Perceptions and Views of Health Professionals Involved in the Relative Effectiveness of Pumps Over MDI and Structured Education (REPOSE) Trial**. *Diabetes Med.* (2016.0) **33**. DOI: 10.1111/dme.12879
47. Puckett C, Wong JC, Daley TC, Cossen K. **How Organizations Shape Medical Technology Allocation: Insulin Pumps and Pediatric Patients With Type 1 Diabetes**. *Soc. Sci. Med.* (2020.0) **249** 112825. DOI: 10.1016/j.socscimed.2020.112825
48. Wong J, Suttiratana S, Lal RA, Lum J, Lanning MS, Dunlap A. **Discontinued Use of the Loop Insulin Dosing System: A Mixed-Methods Investigation**. *Diabetes Technol. Ther.* (2021.0). DOI: 10.1089/dia.2021.0362
49. Braune K, O’donnell S, Cleal B, Lewis DM, Tappe A, Hauck B. **117-LB: DIWHY: Factors Influencing Motivation, Barriers, and Duration of DIY Artificial Pancreas System Use Among Real-World Users**. *Diabetes* (2019.0) **68**. DOI: 10.2337/db19-117-LB
50. Kesavadev J, Saboo B, Kar P, Sethi J. **DIY Artificial Pancreas: A Narrative of the First Patient and the Physicians’ Experiences From India**. *Diabetes Metab. Syndr.* (2021.0) **15**. DOI: 10.1016/j.dsx.2021.02.033
|
---
title: Preoperative Electrophysiology in Patients With Ulnar Nerve Entrapment at the
Elbow-Prediction of Surgical Outcome and Influence of Age, Sex and Diabetes
authors:
- Ilka Anker
- Erika Nyman
- Malin Zimmerman
- Ann-Marie Svensson
- Gert S. Andersson
- Lars B. Dahlin
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2022
pmcid: PMC10012145
doi: 10.3389/fcdhc.2022.756022
license: CC BY 4.0
---
# Preoperative Electrophysiology in Patients With Ulnar Nerve Entrapment at the Elbow-Prediction of Surgical Outcome and Influence of Age, Sex and Diabetes
## Abstract
The impact of preoperative electrophysiology on outcome of surgical treatment in ulnar nerve entrapment at the elbow (UNE) is not clarified. Our aim was to evaluate influence of preoperative electrophysiologic grading on outcome and analyse how age, sex, and in particular diabetes affect such grading. Electrophysiologic protocols for 406 UNE cases, surgically treated at two hand surgery units reporting to the Swedish National Quality Register for Hand Surgery (HAKIR; 2010-2016), were retrospectively assessed, and graded as normal, reduced conduction velocity, conduction block or axonal degeneration. Outcome of surgery after primary and revision surgery was evaluated using QuickDASH and a doctor reported outcome measure (DROM) grading. No differences in QuickDASH or DROM were found between the four groups with different electrophysiologic grading preoperatively, or at three and 12 months or at follow up, respectively. When dichotomizing the electrophysiologic grading into normal and pathologic electrophysiology, cases with normal electrophysiology had worse QuickDASH than cases with pathologic electrophysiology preoperatively ($$p \leq 0.046$$). Presence of a conduction block or axonal degeneration indicated a worse outcome by DROM grading ($$p \leq 0.011$$). Primary surgeries had electrophysiologic more pronounced nerve pathology compared to revision surgeries ($$p \leq 0.017$$). Cases of older age, men, and those with diabetes had more severe electrophysiologic nerve affection ($p \leq 0.0001$). In the linear regression analysis, increasing age (unstandardized $B = 0.03$, $95\%$ CI 0.02-0.04; $p \leq 0.0001$) and presence of diabetes (unstandardized $B = 0.60$, $95\%$ CI 0.25-0.95; $$p \leq 0.001$$) were associated with a higher risk of a worse electrophysiologic classification. Female sex was associated with a better electrophysiologic grading (unstandardized B=-0.51, $95\%$ CI -0.75- -0.27; $p \leq 0.0001$). We conclude that older age, male sex, and concomitant diabetes are associated with more severe preoperative electrophysiologic nerve affection. Preoperative electrophysiologic grade of ulnar nerve affection may influence surgical outcome.
## Introduction
Ulnar nerve entrapment at the elbow (UNE) is mainly considered to be idiopathic. However, risk factors, such as age, sex, concomitant carpal tunnel syndrome (CTS), heavy manual work and multiple occasions of minor pressure at the retrocondylar groove, may predispose to the condition indicating surgery (1–4), but the factors also risks for UNE relapse requiring surgical revision (5–7). Furthermore, diabetes is a known risk factor for compression neuropathies, including UNE (3, 8–10).
The diagnosis of UNE is often based on patient history, symptoms, and clinical signs, supported by electrophysiologic findings (sensitivity 73-$96\%$) to localize the site and estimating the severity of nerve compression (11–13). In addition, electrophysiologic examination may predict surgical outcome according to some studies [14, 15]. However, outcome of primary simple decompression does not seem to differ between cases with solely clinical diagnosis compared to cases with a diagnosis supported by electrophysiology, indicating that clinical symptoms weigh heavily for diagnosis and treatment [14]. There is no clear consensus on optimal management of UNE, and the benefit of preoperative electrophysiology for diagnosis, and prognosis of surgery. There is also a debate about the impact of comorbidity, such as diabetes [16].
Outcome of surgical treatment for UNE seems to be similar (17–19) irrespective of surgical method, with respect to improvements in both clinical and electrophysiologic variables, and even regarding severity of UNE [16]. Diabetes does not affect patient reported outcome after simple decompression in primary UNE, but men with diabetes have a risk for more residual postoperative symptoms [20]. In addition, the relation between preoperative electrophysiologic grading and outcome in UNE patients with diabetes is not known.
There is a need for a clinically applicable preoperative electrophysiologic grading in UNE in order to predict outcome as related to patient characteristics as well as to comorbidities. Our aim was to evaluate the impact of the preoperative electrophysiologic grade of ulnar nerve pathology on outcome of surgery for UNE at the elbow with respect to age, sex, and diabetes.
## Methods
All surgically treated UNE cases between 2010-2016 [identified by ICD-10 diagnosis code G562 and surgical codes ACC53 (simple decompression), ACC43 (transposition) or NCK19 (medial epicondylectomy)], from two hand surgery departments (Malmö and Linköping) included in the Swedish National Quality Register for Hand Surgery (HAKIR; www.hakir.se) [21], were identified and cases with available preoperative electrophysiologic data were included in the study. Electrophysiologic protocols and medical charts were retrospectively assessed. The Swedish National Diabetes Registry (NDR; www.ndr.nu) for adults was merged with data from HAKIR to obtain data for patients with diabetes. The NDR contains data on type of diabetes treatment, complications, and associated risk factors [20, 22]. Each case was defined as a treated nerve. The expression UNE was consistently used in the present study, independently exactly where the ulnar nerve was affected; thus, possibly at the level of the medial epicondyle or by the ligament of Osborne in accordance with previous publications [20, 23]. The study was approved by the Regional Ethical Review Boards in Lund, Sweden (No $\frac{2016}{931}$ and $\frac{2018}{57}$) and Regional Ethics Review Board, Linköping, Sweden (register number $\frac{2016}{88}$-31).
## Data From National Registries and Medical Charts
Data from HAKIR consisted of age, sex, type of ulnar nerve surgery, other concomitant hand surgical procedures, operated side and date of surgery. Pre- and postoperative disability were in the register assessed using the Swedish version of the patient reported outcome measure (PROM) QuickDASH (shortened version of the DASH; Disability of Arm, Shoulder and Hand questionnaire; total calculated score 0-100, higher score indicating more disability). Outcome was scored at three and 12 months postoperatively, as earlier described (20, 23–25).
Additional clinical data, not registered in HAKIR, was retrospectively sampled from patient charts as previously described [23]. Doctor reported outcome measure (DROM) was based on the last out-patient visit (graded by IA; not treating surgeon in any case) and was graded into four groups; cured, improved, unchanged and worsened, and later dichotomized into two groups for statistical analyses (cured/improved and unchanged/worsened).
## Electrophysiology
Electrophysiologic examinations were performed on the ulnar nerves, in most cases bilaterally. The nerves were stimulated at the wrist, below and above elbow and a response was recorded from the abductor digiti minimi muscle. The patients in Lund, Sweden, were also examined with a short segment (2 cm) stimulation across the elbow segment. F-waves and orthodromic sensory response of the ulnar nerve to stimulation of the little finger were also recorded. The results were revised, assessed and graded by one of the authors (GS.A; specialist in neurophysiology; blinded to treatment and outcome) into four groups based on reference values at the Departments of Clinical Neurophysiology in Lund and Linköping, Sweden, respectively, with defined diagnostic criteria for the abnormal groups: i.e. [1] normal findings, [2] reduced conduction velocity across the elbow segment [if upper normal limits are exceeded for a single 2 cm segment (0.9 msec for men and 0.8 msec for women), two segments (1.3 msec men, 1.2 msec women) or all seven segments (3.3 msec men, 3.0 msec women)], [3] nerve conduction block (a $20\%$ amplitude drop over the elbow segment when stimulating above elbow compared with stimulating below elbow), or [4] axonal degeneration [sensory and/or motor amplitudes below normal limit (dependent on age, sex and body height) as earlier described [14, 26]. If a nerve showed both reduced conduction velocity and axonal degeneration, it was graded according to its most pathological parameter.
## Statistical Analyses
Data are presented as median [interquartile range; IQR; Q25-Q75]. Nominal data are presented as numbers (%). For nominal data, a Chi-squared test (Pearson or Fisher´s exact test) was used to compare differences between groups. Non-parametric Kruskal-Wallis test was used to compare differences between groups for continuous data, with subsequent post-hoc analyses (Mann-Whitney U test). Correlations were assessed by Point-Biserial correlation coefficient for dichotomous variables (r, with p-value). An r-value of ≥0.30 (positive or negative value) was interpreted as a correlation (0.30 – 0.7 = moderate correlation; >0.70 = strong correlation). Linear regression analyses were performed to analyse effects of nominal factors on QuickDASH score (unstandardized B [$95\%$ CI]; p-value). A linear regression analysis was performed to investigate the effect of another hand surgical procedure or surgery for another nerve entrapment performed at the same time as UNE surgery on QuickDASH results. All regressions were adjusted for age, sex and diabetes. A p-value <0.05 was considered statistically significant. IBM® SPSS® Statistics, version 26, 2019 (IBM Inc., Chicago, IL) was used for all calculations. Each treated arm was analysed as a separate case and statistical entity.
## Case Characteristics and Surgeries
A larger proportion of the patient cohort has been described earlier [23]. Characteristics of the cases grouped by surgical procedure are presented in Table 1. Out of the original population, consisting of 548 UNE surgeries, solely $\frac{406}{548}$ ($74\%$) surgeries, on which preoperative electrophysiologic data was available, were included in the present study (Figure 1). Out of the included 406 UNE surgeries, $\frac{356}{406}$ ($88\%$) were primary surgeries ($81\%$ simple decompression) and $\frac{50}{406}$ ($12\%$) were reoperations ($86\%$ transpositions). There were no cases surgically treated with a medial epicondylectomy or endoscopic decompression (Table 1).
Out of all surgically treated cases, $\frac{207}{406}$ ($51\%$) were females, with a median age of 50 [interquartile range; IQR 41-59] years for women and 53 [44-62] for men (Table 1). In total, $\frac{56}{406}$ ($14\%$) had concomitant diabetes [$\frac{16}{56}$ ($29\%$) with type 1 and $\frac{35}{56}$ ($62\%$) with type 2 diabetes, data missing or unclassified in 5 cases; $9\%$]. Another hand surgical procedure was performed at the same time as the surgery for UNE in $6\%$ of cases; i.e. surgery for trigger finger, thumb basal osteoarthritis, ganglion, de Quervain’s tenosynovitis or idiopathic synovitis. Another nerve entrapment surgery was performed at the same time as the surgery for UNE in $12\%$; i.e. carpal tunnel release, decompression of the ulnar nerve at wrist level (Guyon´s canal), decompression of the radial nerve or surgery on multiple nerves. These concomitant hand surgical and nerve related entrapment procedures (adjusted for age, sex, and diabetes) did not affect QuickDASH results at 3 or at 12 months (regression analysis; data not shown).
## Preoperative Electrophysiology and Surgical Procedures
Among primary surgeries, there were relatively more cases with reduced nerve conduction velocity and axonal degeneration, based on the electrophysiologic grading, than among revision surgeries ($$p \leq 0.017$$; Table 2). Electrophysiologic grading preoperatively did not differ when comparing primary simple decompressions with primary ulnar nerve transpositions ($$p \leq 0.07$$; results not shown). There were too few cases among revision simple decompression surgeries with electrophysiologic pathology [$\frac{3}{50}$ ($6\%$)] for adequate statistical analyses to be made on revision surgeries (data not shown).
**Table 2**
| Electrophysiologic grading | Primary surgery (n = 356) | Revision surgery (n = 50) | P-value |
| --- | --- | --- | --- |
| Normal (n = 134) | 108 (30%) | 26 (52%) | |
| Reduced nerve conduction velocity (n = 60) | 56 (16%) a | 4 (8%) | |
| Conduction block (n = 23) | 22 (6%) | 1 (2%) | |
| Axonal degeneration (n = 189) | 170 (48%) a | 19 (38%) | 0.017 |
## Responders, PROM, DROM and Electrophysiology
QuickDASH response rates were $\frac{92}{406}$ ($23\%$) preoperatively, $\frac{107}{406}$ ($27\%$) at three months postoperatively and $\frac{101}{406}$ ($25\%$) at 12 months postoperatively. DROM grading (median follow up time 3.0 months [IQR 1.5-6.0]), was possible to evaluate in $\frac{395}{406}$ ($97\%$; missing in 11 cases; $3\%$) of cases. In the remaining cases, no postoperative outcome was noted in the patient charts.
When analysing all surgically treated UNE cases, no difference in QuickDASH was found neither preoperatively, nor at three or 12 months postoperatively, between the four electrophysiology groups (Table 3). Similar results were found when analysing solely primary UNE surgeries, i.e. no significant differences in QuickDASH score in relation to electrophysiologic grading preoperatively or at three or 12 months postoperatively ($$p \leq 0.14$$ preoperatively; $$p \leq 0.79$$ at 3 months; $$p \leq 0.07$$ at 12 months postoperatively; data not shown). QuickDASH response rates were too low among revision surgeries for statistical analyses to be made.
**Table 3**
| QuickDASH | Normal (n = 134) | Reduced nerve conduction velocity (n = 60) | Conduction block (n = 23) | Axonal degeneration (n = 189) | P-values |
| --- | --- | --- | --- | --- | --- |
| Preoperatively | 61 [43-73] | 53 [27-62] | 34 [16-60] | 55 [39-68] | 0.09 |
| Preoperatively | (n = 27) | (n = 14) | (n = 4) | (n = 47) | 0.09 |
| 3 months postoperatively | 39 [22-60] | 35 [15-48] | 16 [10-72] | 31 [15-63] | 0.51 |
| 3 months postoperatively | (n = 33) | (n = 20) | (n = 8) | (n = 48) | 0.51 |
| 12 months postoperatively | 45 [25-64] | 24 [13-36] | 43 [15-60] | 48 [24-63] | 0.06 |
| 12 months postoperatively | (n = 27) | (n = 18) | (n = 13) | (n = 44) | 0.06 |
| DROM grading a | Normal (n = 134) | Reduced nerve conduction velocity (n = 60) | Conduction Block (n = 23) | Axonal degeneration (n = 189) | |
| Cured/improved (n = 295) | 105 (78%) | 48 (80%) | 15 (65%) | 127 (67%) | |
| Unchanged/worsened (n = 100) | 27 (20%) | 10 (17%) | 6 (26%) | 57 (30%) | 0.08 |
| Missing grading a | 2 (2%) | 2 (3%) | 2 (9%) | 5 (3%) | |
When analysing DROM grading in all surgically treated cases ($$p \leq 0.08$$; Table 3) and solely primary UNE cases ($$p \leq 0.07$$; data not shown), no differences were found in postoperative outcome in relation to the four electrophysiologic grades of nerve affection.
Furthermore, when dichotomizing the electrophysiologic grading into normal [$$n = 132$$; cured/improved=105 ($80\%$) and unchanged/worsened=27 ($20\%$)] and pathologic [$$n = 263$$; cured/improved=190 ($72\%$) and unchanged/worsened=73 ($28\%$)] electrophysiology, no difference was observed in the DROM grading ($$p \leq 0.14$$; Fisher’s exact test). Using the same dichotomizing procedure, the QuickDASH scores differed preoperatively (normal 61 [43-73], $$n = 27$$; pathologic 55 [34-64], $$n = 65$$; $$p \leq 0.046$$), but not at three (39 [22-60], $$n = 33$$; 30 [14-57], $$n = 76$$; respectively, $$p \leq 0.16$$) and 12 months (45 [25-64], $$n = 27$$; 41 [15-59], $$n = 75$$, respectively; $$p \leq 0.31$$).
When the electrophysiologic grading was divided into two other groups in accordance with a previous method [14], normal and reduced velocity [$$n = 190$$; cured/improved=153 ($81\%$) and unchanged/worsened=37 ($19\%$)] versus conduction block and axonal degeneration [$$n = 205$$; cured/improved=142 ($69\%$) and unchanged/worsened=63 ($31\%$)], a significant difference was observed in grading with DROM ($$p \leq 0.011$$; Fisher’s exact test). Using the same procedure, the QuickDASH scores did not differ preoperatively (normal/reduced velocity 59 [40-71], $$n = 41$$; conduction block/axonal degeneration 55 [39-68], $$n = 51$$; $$p \leq 0.38$$), at three (39 [19-58], $$n = 53$$; 30 [14-63], $$n = 56$$; respectively, $$p \leq 0.50$$) or 12 months (34 [18-55], $$n = 45$$; 45 [21-62], $$n = 57$$, respectively; $$p \leq 0.25$$).
No moderate or strong correlations were found between neither electrophysiologic grading and pre- or postoperative QuickDASH scores, nor DROM grading.
## Age, Sex, and Electrophysiology
Cases at older age (regardless of sex and both in all and solely primary UNE) had more severe electrophysiologic findings than cases at younger age ($p \leq 0.0001$; Table 4). A moderate positive correlation ($r = 0.38$, $p \leq 0.0001$) was found between age and electrophysiologic grade of nerve affection.
**Table 4**
| Age (years) | Normal (n = 134) | Reduced nerve conduction velocity (n = 60) | Conduction block (n = 23) | Axonal degeneration (n = 189) | P-values |
| --- | --- | --- | --- | --- | --- |
| | 45 (37-52) a | 53 (45-59) | 60 (58-65) | 56 (47-64) | <0.0001 |
| Sex | Normal (n = 134) | Reduced nerve conduction velocity (n = 60) | Conduction block (n = 23) | Axonal degeneration (n = 189) | |
| Male (n = 199) | 47 (24%) | 26 (13%) | 10 (5%) | 116 (58%) b | |
| Female (n = 207) | 87 (42%) | 34 (16%) | 13 (6%) | 73 (35%) | <0.0001 |
| Diabetes status (only primary surgeries included) | Normal (n = 108) | Reduced nerve conduction velocity (n = 56) | Conduction block (n = 22) | Axonal degeneration (n = 170) | |
| Diabetes (n = 53) | 2 (4%) | 11 (21%) | 5 (9%) | 35 (66%) | |
| No diabetes (n = 303) | 106 (35%) c | 45 (15%) | 17 (6%) | 135 (44%) | <0.0001 |
Men more often had axonal degeneration at the electrophysiologic examination than women ($p \leq 0.0001$; Table 4). No correlation (moderate or strong) was found between electrophysiologic grading and sex.
## Diabetes and Electrophysiology
Cases with diabetes who had undergone primary surgeries were older (58 [IQR 53-64] years; $$n = 53$$) compared to cases without diabetes (51 [IQR 42-61]) ($p \leq 0.0001$; $$n = 303$$; Mann-Whitney U-test), but with no differences in sex distribution ($$p \leq 0.13$$; Fisher´s exact test). In addition, there was no significant difference in sex distribution among all the cases concerning presence of diabetes (males with diabetes $\frac{32}{199}$ ($16\%$) and females with diabetes $\frac{24}{207}$ ($12\%$); $$p \leq 0.12$$; Fisher´s exact test). Most cases with diabetes were found among primary surgeries ($\frac{53}{56}$; $95\%$). Among revision surgeries only three cases had concomitant diabetes [$\frac{3}{56}$ ($5\%$); among revision with ulnar nerve transpositions only; Table 1] and due to the low frequency, further analyses on and including revision surgeries were not performed.
Primary UNE cases with diabetes (only primary cases analysed due to few revision cases among patients with diabetes) had significantly more severe electrophysiologic pathology, in the form of reduced nerve conduction velocity, nerve conduction block and axonal degeneration compared to cases without diabetes ($p \leq 0.0001$; Table 4), which was also found to be similar for men with diabetes ($$p \leq 0.012$$) and women with diabetes ($$p \leq 0.011$$; Table 5). No moderate or strong correlations were found between concomitant diabetes and electrophysiologic grade of nerve pathology.
**Table 5**
| Diabetes status and sex | Normal (n = 38) | Reduced nerve conduction velocity (n = 25) | Conduction block (n = 10) | Axonal degeneration (n = 106) | P-values |
| --- | --- | --- | --- | --- | --- |
| Males with diabetes (n = 31) | 0 (0%) | 4 (13%) | 2 (6%) | 25 (81%) | |
| Males with no diabetes (n = 148) | 38 (26%) a | 21 (14%) | 8 (5%) | 81 (55%) | 0.012 |
| Diabetes status and sex | Normal (n = 70) | Reduced nerve conduction velocity (n = 31) | Conduction block (n = 12) | Axonal degeneration (n = 64) | |
| Females with diabetes (n = 22) | 2 (9%) | 7 (32%) | 3 (14%) | 10 (45%) | |
| Females with no diabetes (n = 155) | 68 (44%) b | 24 (15%) | 9 (6%) | 54 (35%) | 0.011 |
## Linear Regression, Age, Sex, and Diabetes
Increasing age (unstandardized $B = 0.03$, $95\%$ CI 0.02-0.04; $p \leq 0.0001$) and concomitant diabetes (unstandardized $B = 0.60$, $95\%$ CI 0.25-0.95; $$p \leq 0.001$$) were associated with a higher risk of a worse electrophysiology classification, while female sex was associated with better grading in the electrophysiology classification (unstandardized B=-0.51, $95\%$ CI -0.75- -0.27; $p \leq 0.0001$).
## Discussion
In the present study, we found no significant differences in outcome, evaluated with QuickDASH or DROM, in all surgically treated cases or in solely primary cases, neither at three nor at 12 months postoperatively or at follow-up, respectively, when four different grades of electrophysiologic pathology were compared. This is in line with a previous systematic review reporting effectiveness and safety of treatment for UNE referencing few studies with a follow-up longer than 12 months [16]. However, in accordance with our previous retrospective study [14], dichotomizing patients with a preoperative nerve conduction block or axonal degeneration against normal findings and reduced conduction velocity, a higher risk of worse postoperative outcome after primary simple decompression was found when outcome was analysed with DROM [14], but not with QuickDASH, even though DROM and QuickDASH has been found to be related [27]. No statistical correlation analysis between individual nerve conduction velocities (in m/s) and QuickDASH scores was performed due to the limited number of cases. Electrophysiologic grading is not always considered in larger systematic reviews and meta-analysis when evaluating safety and outcome of surgical procedures for UNE [28]. However, it is still a debate if and how the preoperative electrophysiologic grading influence outcome of surgery, which may depend on the methods of evaluation [29, 30].
One cannot exclude that a relation exists between the different severities of electrophysiologic grading and outcome as evaluated by QuickDASH, although significance was not achieved among the four groups, which may be related to statistical power. An explanation of our findings, regarding outcome when QuickDASH was used, may be due to a limited number of cases in each group and being an effect of under-power. When a dichotomizing procedure was performed, dividing electrophysiologic grade into normal and pathologic findings, the former had a slightly higher Quick DASH score preoperatively (around 6 points), but with no differences at three or 12 months, indicating more disability preoperatively for cases with normal electrophysiologic grading. This is clinically a minor difference in disability, meaning that those findings should be interpreted with caution. Nevertheless, the DROM grading indicated that if the patients present with the two worst electrophysiology grades, there is a risk of worse outcome, irrespective of being a nerve conduction block or presence of axonal degeneration. However, a larger population is required to distinguish the outcome of surgery based on the electrophysiology grades nerve conduction block and axonal degeneration. Furthermore, we cannot explain the present observation that there was an initial improvement in QuickDASH at three months and a subsequent worsening at 12 months among the patients with preoperative electrophysiology findings of axonal degeneration. One may speculate that such an affected nerve, due to the lower number of functioning nerve fibres, may be more susceptible to further trauma, such as development of scar tissue around the nerve over time.
In the current study, we found more cases with electrophysiologic more severe nerve pathology among primary surgeries compared to revision surgeries. In primary UNE cases, simple decompression is usually the surgical gold standard treatment, regardless of electrophysiologic severity of nerve affection [16]. If an ulnar nerve dislocation is found pre- or perioperatively, an ulnar nerve transposition is commonly performed instead as the primary procedure. Even after a simple decompression a greater mobility of the ulnar nerve can be expected with a risk for dislocation of the ulnar nerve; a statement that is supported by a recent study [23]. A significantly higher presence of ulnar nerve dislocation was found among revision surgeries compared to primary surgeries, and at the same time significantly higher presence of ulnar nerve dislocation among primary transposition surgeries was observed compared to primary simple decompressions [23]. Hence, we interpret that our current findings might be reflecting a presence of ulnar nerve dislocation among revision surgeries and primary ulnar nerve transpositions, being the reason for these cases presenting an electrophysiological normal or less severe nerve pathology, i.e. due to these unstable ulnar nerves having normal electrophysiologic findings.
Further, we found that men and cases at older age had more severe electrophysiologic impact on nerve function compared to women and cases at younger age. For the latter, we also found a moderate positive correlation, altogether indicating that increasing age may affect electrophysiologic findings negatively and increase severity of nerve pathology. Some earlier studies point out that older age and male sex, among others, are risk factors to develop UNE (1–4). Men in the present study showed a higher proportion of axonal degeneration based on the electrophysiology examination, which also may indicate an increased susceptibility to compression. It has been shown that men have lower intraepidermal nerve fibre density in biopsies from skin at wrist level compared to women [31]. This can be interpreted as men having a more sensitive peripheral nervous system, with less reserve nerve fibre capacity, being more prone to be affected by compression, compared to women and might support our findings of men, and particularly those with diabetes, having electrophysiologic more severe impact on nerve function. Data from national registers also support the notion that men with diabetes may not have the same benefit as women with diabetes to improve by a simple decompression [20].
When analysing comorbidity in form of concomitant diabetes, we found that cases with diabetes were significantly older, but there were no differences in sex distribution. Cases with diabetes had more severe electrophysiologic nerve pathology. Diabetes may affect the peripheral nervous system and is a known risk factor for distal sensory polyneuropathy [32] and compression neuropathies, such as carpal tunnel syndrome [33]. Several studies have found diabetes to be a risk factor for primary UNE as well (3, 8–10), although it has not consistently been found to increase risk of UNE relapse (5–7). Diabetes affects peripheral nerves by inducing intraneural structural changes [34]. We interpret our findings, of cases with diabetes having more severe electrophysiologic nerve affection, as a reflection of this known peripheral nerve affection due to the mentioned structural changes in the nerves.
The interpretation of our combined results, with male sex and diabetes as comorbidity being related to electrophysiological more severe nerve affection, might be explained by men having a more sensitive peripheral nervous system with less reserve capacity when it comes to nerve fibre quantity [31] and the fact that men, as reported, seem to be affected by diabetic neuropathy to a greater extent and earlier compared to women [35, 36].
## Strengths and Limitations
The low response rate in QuickDASH scores is a limitation even if similar rates have been reported in earlier studies. The HAKIR register was at the time of data collection [2010-2016] also a rather new register with initial problems to include patients. Due to the coding system in HAKIR, we did not have data on which type of transposition that was performed and data on whether surgery was primary or revision was not appropriately specified in HAKIR. Hence, the latter data was added after the thorough retrospective evaluation made on each unique patient chart. A further weakness is that we could not in detail, based on the information from the patient charts, define the exact level of ulnar nerve affection; thus, being at or just proximal to the medial epicondyle or distally, exactly at the ligament of Osborne [37], although the latter location was probably the most common site. However, we defined the presently used expression UNE as a single entity, including both locations, in accordance with previous publications [20, 23]. A strength is the use of data from the two national quality registers (HAKIR and NDR), combined with data from each unique patient chart together with a validated outcome measure (QuickDASH), which enables analyses of outcome concerning a nationwide population.
## Conclusions
We conclude that older age, male sex, and diabetes are associated with more severe preoperative electrophysiologic nerve affection, which may be interpreted as more susceptible peripheral nerves in men and in diabetes that should be taken into account when surgically treating UNE patients. Preoperative electrophysiologic assessment and severe grade of ulnar nerve affection may influence surgical outcome.
## Data Availability Statement
The datasets generated and/or analysed during the current study are not publicly available. Public access to data is restricted by the Swedish Authorities (Public Access to Information and Secrecy Act; https://www.government.se/information-material/$\frac{2009}{09}$/public-access-to-information-and-secrecy-act/), but data can be available for researchers after a special review that includes approval of the research project by both an Ethics Committee at the national level (etikprövningsmyndigheten.se) and the authorities’ data safety committees (such as “KVB-decision”).
## Ethics Statement
The study was approved by the Regional Ethical Review Boards in Lund, Sweden (No $\frac{2016}{931}$ and $\frac{2018}{57}$) and Regional Ethics Review Board, Linköping, Sweden (register number $\frac{2016}{88}$-31). The patients/participants provided their written informed consent to participate in HAKIR and NDR.
## Author Contributions
IA, EN, MZ, and LD generated the hypothesis and outline of the project. All authors interpreted the data and critically reviewed the report. IA and EN collected the data from electrophysiologic examinations and patients’ charts. GA analysed and interpreted the electrophysiologic data. A-MS was responsible for collecting the data from the diabetes register. IA performed the initial analyses and drafted the first manuscript. LD performed the final statistical analyses. A-MS and GA contributed to hypothesis generation and to writing the manuscript. All authors fulfilled the criteria for authorship. All authors contributed to the article and approved the submitted version.
## Funding
This study was supported by grants from the Lund University, ALF [grant number 2018-Project 0104], Region Skåne (Funds from Skåne University Hospital Malmö-Lund), the Swedish Diabetes Foundation [grant number DIA2016-117 and DIA2020-492], the Swedish Research Council [grant number 2021-01942], Sydvästra Skånes Diabetesförening, Sweden, and ALF Grants [grant number LIO-823361], Region Östergötland, Sweden.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Osei DA, Groves AP, Bommarito K, Ray WZ. **Cubital Tunnel Syndrome: Incidence and Demographics in a National Administrative Database**. *Neurosurgery* (2017.0) **80**. DOI: 10.1093/neuros/nyw061
2. Omejec G, Podnar S. **What Causes Ulnar Neuropathy at the Elbow**. *Clin Neurophysiol* (2016.0) **127**. DOI: 10.1016/j.clinph.2015.05.027
3. Bartels RH, Verbeek AL. **Risk Factors for Ulnar Nerve Compression at the Elbow: A Case Control Study**. *Acta Neurochir (Wien)* (2007.0) **149** 669-74; discussion 74. DOI: 10.1007/s00701-007-1166-5
4. Descatha A, Leclerc A, Chastang JF, Roquelaure Y. **Incidence of Ulnar Nerve Entrapment at the Elbow in Repetitive Work**. *Scand J Work Environ Health* (2004.0) **30**. DOI: 10.5271/sjweh.784
5. Camp CL, Ryan CB, Degen RM, Dines JS, Altchek DW, Werner BC. **Risk Factors for Revision Surgery Following Isolated Ulnar Nerve Release at the Cubital Tunnel: A Study of 25,977 Cases**. *J Shoulder Elbow Surg* (2017.0) **26**. DOI: 10.1016/j.jse.2016.10.028
6. Krogue JD, Aleem AW, Osei DA, Goldfarb CA, Calfee RP. **Predictors of Surgical Revision After**. *J Shoulder Elbow Surg* (2015.0) **24**. DOI: 10.1016/j.jse.2014.12.015
7. Gaspar MP, Kane PM, Putthiwara D, Jacoby SM, Osterman AL. **Predicting Revision Following**. *J Handb Surg Am* (2016.0) **41**. DOI: 10.1016/j.jhsa.2015.12.012
8. Mondelli M, Aretini A, Rossi S. **Ulnar Neuropathy at the Elbow in Diabetes**. *Am J Phys Med Rehabil* (2009.0) **88**. DOI: 10.1097/PHM.0b013e318190b89d
9. Rota E, Zavaroni D, Parietti L, Iafelice I, De Mitri P, Terlizzi E. **Ulnar Entrapment Neuropathy in Patients With Type 2 Diabetes Mellitus: An Electrodiagnostic Study**. *Diabetes Res Clin Pract* (2014.0) **104**. DOI: 10.1016/j.diabres.2014.01.024
10. Rydberg M, Zimmerman M, Gottsater A, Nilsson PM, Melander O, Dahlin LB. **Diabetes Mellitus as a Risk Factor for Compression Neuropathy: A Longitudinal Cohort Study From Southern Sweden**. *BMJ Open Diabetes Res Care* (2020.0) **8** e001298. DOI: 10.1136/bmjdrc-2020-001298
11. Visser LH, Beekman R, Franssen H. **Short-Segment Nerve Conduction Studies in Ulnar Neuropathy at the Elbow**. *Muscle Nerve* (2005.0) **31**. DOI: 10.1002/mus.20248
12. Todnem K, Michler RP, Wader TE, Engstrom M, Sand T. **The Impact of Extended Electrodiagnostic Studies in Ulnar Neuropathy at the Elbow**. *BMC Neurol* (2009.0) **9**. DOI: 10.1186/1471-2377-9-52
13. Pelosi L, Mulroy E. **Diagnostic Sensitivity of Electrophysiology and Ultrasonography in Ulnar Neuropathies of Different Severity**. *Clin Neurophysiol* (2019.0) **130** 297-302. DOI: 10.1016/j.clinph.2018.11.018
14. Anker I, Zimmerman M, Andersson GS, Jacobsson H, Dahlin LB. **Outcome and Predictors in Simple Decompression of Ulnar Nerve Entrapment at the Elbow**. *Hand Microsurg* (2018.0) **7** 24-32. DOI: 10.5455/handmicrosurg.266188
15. Osterman AL, Davis CA. **Subcutaneous Transposition of the Ulnar Nerve for Treatment of Cubital Tunnel Syndrome**. *Handb Clin* (1996.0) **12**
16. Caliandro P, La Torre G, Padua R, Giannini F, Padua L. **Treatment for Ulnar Neuropathy at the Elbow**. *Cochrane Database Syst Rev* (2016.0) **11**. DOI: 10.1002/14651858.CD006839
17. Liu CH, Wu SQ, Ke XB, Wang HL, Chen CX, Lai ZL. **Subcutaneous Versus Submuscular Anterior Transposition of the Ulnar Nerve for Cubital Tunnel Syndrome: A Systematic Review and Meta-Analysis of Randomized Controlled Trials and Observational Studies**. *Med (Baltimore)* (2015.0) **94**. DOI: 10.1097/MD.0000000000001207
18. Biggs M, Curtis JA. **Randomized, Prospective Study Comparing Ulnar Neurolysis**. *Neurosurgery* (2006.0) **58** 296-304; discussion 296-304. DOI: 10.1227/01.NEU.0000194847.04143.A1
19. Gervasio O, Gambardella G, Zaccone C, Branca D. **Simple Decompression Versus Anterior Submuscular Transposition of the Ulnar Nerve in Severe Cubital Tunnel Syndrome: A Prospective Randomized Study**. *Neurosurgery* (2005.0) **56** 108-17; discussion 17. DOI: 10.1227/01.NEU.0000145854.38234.81
20. Zimmerman M, Anker I, Karlsson A, Arner M, Svensson AM, Eeg-Olofsson K. **Ulnar Nerve Entrapment in Diabetes: Patient-Reported Outcome After Surgery in National Quality Registries**. *Plast Reconstr Surg Glob Open* (2020.0) **8**. DOI: 10.1097/GOX.0000000000002740
21. Arner M. **Developing a National Quality Registry for Hand Surgery: Challenges and Opportunities**. *EFORT Open Rev* (2016.0) **1**. DOI: 10.1302/2058-5241.1.000045
22. Lind M, Bounias I, Olsson M, Gudbjornsdottir S, Svensson AM, Rosengren A. **Glycaemic Control and Incidence of Heart Failure in 20,985 Patients With Type 1 Diabetes: An Observational Study**. *Lancet* (2011.0) **378**. DOI: 10.1016/S0140-6736(11)60471-6
23. Anker I, Zimmerman M, Nyman E, Dahlin LB. **Ulnar Nerve Dislocation in Ulnar Nerve Entrapment at the Elbow. Influence on Surgical Outcome**. *Hand Surg Rehabil* (2022.0) **41** 96-102. DOI: 10.1016/j.hansur.2021.09.003
24. 24
The Swedish Translated Version of QuickDASH. Available at: http://www.dash.iwo.on.ca/assets/images/pdfs/OuickDASH_Swedish.pdf.. *The Swedish Translated Version of QuickDASH*
25. Zimmerman M, Eeg-Olofsson KMDP, Svensson Am Rn P, Astrom M, Arner M, Dahlin L. **Open Carpal Tunnel Release and Diabetes: A Retrospective Study Using PROMs and National Quality Registries**. *BMJ Open* (2019.0) **9**. DOI: 10.1136/bmjopen-2019-030179
26. Anker I, Andersson G, Zimmerman M, Jacobsson H, Dahlin LB. **Subcutaneous and Submuscular Transposition Due to Ulnar Nerve Entrapment at the Elbow – Analyses of 43 Primary and 44 Revision Cases**. *Hand Microsurg* (2019.0) **8** 9-18. DOI: 10.5455/handmicrosurg.299222
27. Dahlin E, Dahlin E, Andersson GS, Thomsen NO, Bjorkman A, Dahlin LB. **Outcome of Simple Decompression of the Compressed Ulnar Nerve at the Elbow - Influence of Smoking, Gender, and Electrophysiological Findings**. *J Plast Surg Handb Surg* (2017.0) **51**. DOI: 10.1080/2000656X.2016.1210520
28. Wade RG, Griffiths TT, Flather R, Burr NE, Teo M, Bourke G. **Safety and Outcomes of Different Surgical Techniques for Cubital Tunnel Decompression: A Systematic Review and Network Meta-Analysis**. *JAMA Netw Open* (2020.0) **3**. DOI: 10.1001/jamanetworkopen.2020.24352
29. Beekman R, Wokke JH, Schoemaker MC, Lee ML, Visser LH. **Ulnar Neuropathy at the Elbow: Follow-Up and Prognostic Factors Determining Outcome**. *Neurology* (2004.0) **63**. DOI: 10.1212/01.WNL.0000142535.24626.90
30. Shi Q, MacDermid J, Grewal R, King GJ, Faber K, Miller TA. **Predictors of Functional Outcome Change 18 Months After Anterior Ulnar Nerve Transposition**. *Arch Phys Med Rehabil* (2012.0) **93**. DOI: 10.1016/j.apmr.2011.08.040
31. Thomsen NO, Englund E, Thrainsdottir S, Rosen I, Dahlin LB. **Intraepidermal Nerve Fibre Density at Wrist Level in Diabetic and Non-Diabetic Patients**. *Diabetes Med* (2009.0) **26**. DOI: 10.1111/j.1464-5491.2009.02823.x
32. Barrell K, Smith AG. **Peripheral Neuropathy**. *Med Clin North Am* (2019.0) **103**. DOI: 10.1016/j.mcna.2018.10.006
33. Tseng CH, Liao CC, Kuo CM, Sung FC, Hsieh DP, Tsai CH. **Medical and Non-Medical Correlates of Carpal Tunnel Syndrome in a Taiwan Cohort of One Million**. *Eur J Neurol* (2012.0) **19**. DOI: 10.1111/j.1468-1331.2011.03440.x
34. Dahlin LB, Rix KR, Dahl VA, Dahl AB, Jensen JN, Cloetens P. **Three-Dimensional Architecture of Human Diabetic Peripheral Nerves Revealed by X-Ray Phase Contrast Holographic Nanotomography**. *Sci Rep* (2020.0) **10** 7592. DOI: 10.1038/s41598-020-64430-5
35. Aaberg ML, Burch DM, Hud ZR, Zacharias MP. **Gender Differences in the Onset of Diabetic Neuropathy**. *J Diabetes Complications* (2008.0) **22**. DOI: 10.1016/j.jdiacomp.2007.06.009
36. Ennis SL, Galea MP, O’Neal DN, Dodson MJ. **Peripheral Neuropathy in the Hands of People With Diabetes Mellitus**. *Diabetes Res Clin Pract* (2016.0) **119** 23-31. DOI: 10.1016/j.diabres.2016.06.010
37. Osborne G. **Ulnar Neuritis**. *Postgrad Med J* (1959.0) **35**. DOI: 10.1136/pgmj.35.405.392
|
---
title: 'Predictive Value of Immune Cells in the Risk of Gestational Diabetes Mellitus:
A Pilot Study'
authors:
- Adnette Fagninou
- Magloire Pandoua Nekoua
- Salomon Ezéchiel M. Fiogbe
- Kabirou Moutaïrou
- Akadiri Yessoufou
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2022
pmcid: PMC10012146
doi: 10.3389/fcdhc.2022.819164
license: CC BY 4.0
---
# Predictive Value of Immune Cells in the Risk of Gestational Diabetes Mellitus: A Pilot Study
## Abstract
### Aims
Immunological and biochemical parameters are gaining more and more importance in the prognosis of diabetes and its complications. Here, we assessed the predictive power of immune cells correlated with biochemical parameters in gestational diabetes mellitus (GDM).
### Materials and Methods
Immune cells and serum biochemical parameters were determined in women with GDM and pregnant controls. Receiver operating characteristics (ROC) curve analyses were conducted to assess the optimal cutoff and value of ratios of immune cells to biochemical parameters for predicting GDM.
### Results
Blood glucose, total cholesterol, LDL-cholesterol and triglycerides were significantly increased whereas HDL-cholesterol decreased in women with GDM compared to pregnant controls. Glycated hemoglobin, creatinine, transaminase activities did not significantly differ between both groups. Total leukocyte, lymphocyte and platelet numbers were significantly high in women with GDM. Correlation tests showed that ratios of lymphocyte/HDL-C, monocyte/HDL-C and granulocyte/HDL-C were significantly higher in women with GDM than in pregnant controls ($$p \leq 0.001$$; $$p \leq 0.009$$ and $$p \leq 0.004$$ respectively). Women with a lymphocyte/HDL-C ratio greater than 3.66 had a 4-fold increased risk of developing GDM than those with lower ratios (odds ratio 4.00; $95\%$ CI: 1.094 – 14.630; $$p \leq 0.041$$).
### Conclusion
Our study showed that ratios of lymphocyte, monocyte and granulocyte to HDL-C might represent valuable biomarkers for GDM and in particular, lymphocyte/HDL-C ratio exhibited a strong predictive power for GDM risk.
## Introduction
One of the major concerns of researchers is to find biological or clinical factors with prognostic or early diagnostic value of diseases in order to strengthen or improve prevention rather than cure. In this context, little is known about the use of immunological and/or biochemical parameters in the prediction of gestational diabetes mellitus (GDM) [1, 2]. Recently, we investigated the modulation of immune cell frequencies in gestational diabetes, and found that gestational diabetes mellitus (GDM) modulated the frequencies of total CD3+ and CD4+ T and B cells, suggesting that immune cells could play specific role in the prognosis of this disease [3]. GDM is defined as glucose intolerance arising for the first time during pregnancy with or without remission after the end of pregnancy [4, 5]. GDM, as one of major endocrine abnormalities, is the most common metabolic disease during pregnancy and its incidence is increasing worldwide [4, 5]. The global prevalence of GDM varies from 1 to $28\%$ depending on population characteristics, screening methods, and diagnostic criteria (6–8) with a great percentage reported in low and middle-income countries, where access to maternal care is often limited [9]. Sedentary and modern lifestyle in developing countries contribute to the increased prevalence of GDM [10, 11].
Evidently, immunological parameters including immune cell subpopulations and cytokines have been designated as predictors of endothelial dysfunction and inflammation [12]. Likewise, we have recently reported that immune cell frequencies, including neutrophils, eosinophils, monocytes, NK cells, and lymphocytes, can be modulated in type 1 diabetes and type 2 diabetes whether associated with pregnancy or not, suggesting that these cells can play important roles in the pathogenesis of this disease, on the one hand [3, 13, 14]. On the other hand, we have reported that GDM can induce disruption of several biochemical and immunological parameters [3, 15, 16]. Additionally, we have reported that biochemical parameters, including glycaemia, triglycerides (TG), high density lipoprotein-cholesterol (HDL-C), total cholesterol (TC), low density lipoprotein-cholesterol (LDL-C), known as metabolic biomarkers, are modulated during GDM and macrosomia (15, 17–20). Interestingly in the same way, several studies have found that immune parameters, including lymphocytes, neutrophils, monocytes, platelets, and the ratios between these cells and HDL-C, may be related to metabolic syndrome and atherosclerotic processes, as potential indicators of prothrombotic and pro-inflammatory states (21–24). Consequently, early diagnosis of gestational diabetes, based on biochemical and immunological parameters, could be crucial to anticipate the care of pregnant diabetic women and thus, prevent the wide range of adverse consequences on the offspring, including macrosomia, fetal death, prematurity, birth trauma, respiratory distress syndrome, obesity, impaired glucose tolerance, and type 2 diabetes in adulthood [15, 20]. Evidently, biochemical parameters can be easily determined in plasma and immune parameters can be easily measured from peripheral blood. Biochemical and immunological indicators, as discussed above, can be used as potential markers to predict GDM. Therefore, the principal objective of this study is to determine whether immune cells could be correlated with biochemical parameters to assess their predictive value for GDM.
## Study Participants
In this cross-sectional and descriptive study, two hundred and forty-six [246] pregnant women were firstly enrolled by specialist clinicians of the department of gynecology and obstetrics three national hospital centers in southern Benin. This sample size was calculated based on Dagnelie’s formula. Based on inclusion criteria including absence of preexisting type 1 or type 2 diabetes, infectious diseases including hepatitis, HIV and malaria after blood sample tests, 210 pregnant women, aged from 19 to 43 years, were selected and then screened for GDM (see protocol below). Anthropometric and socio-demographic data, risk factors and family history associated with diabetes were recorded and presented in Table 1.
**Table 1**
| Characteristics | Pregnant control women | Pregnant control women.1 | Women with GDM | Women with GDM.1 | Total |
| --- | --- | --- | --- | --- | --- |
| Characteristics | Number | Percentage (%) | Number | Percentage (%) | Total |
| Number of subjects | 185 | 88.10 | 25 | 11.90 | 210 |
| Age (A, years) | Age (A, years) | Age (A, years) | Age (A, years) | Age (A, years) | Age (A, years) |
| < 20 | 05 | 2.70 | 00 | 00 | 05 |
| 20 ≤ A < 30 | 104 | 56.22 | 09 | 36 | 113 |
| 30 ≤ A < 40 | 71 | 38.38 | 15 | 60 | 86 |
| ≥ 40 | 05 | 2.7 | 01 | 04 | 06 |
| Physical activity | Physical activity | Physical activity | Physical activity | Physical activity | Physical activity |
| Intense | 00 | 00 | 00 | 00 | 00 |
| Moderate | 136 | 73.51 | 17 | 68 | 153 |
| Inadequate | 49 | 26.49 | 08 | 32 | 57 |
| Menstrual cycle | Menstrual cycle | Menstrual cycle | Menstrual cycle | Menstrual cycle | Menstrual cycle |
| Regular | 80 | 43.24 | 10 | 40 | 90 |
| Irregular | 105 | 56.76 | 15 | 60 | 120 |
| Number of children (N) | Number of children (N) | Number of children (N) | Number of children (N) | Number of children (N) | Number of children (N) |
| 0 | 50 | 27.02 | 02 | 08 | 52 |
| 1 ≤ N ≤ 2 | 97 | 52.44 | 07 | 28 | 104 |
| N ≥ 3 | 38 | 20.54 | 16 | 64 | 54 |
| Previous disturbances | Previous disturbances | Previous disturbances | Previous disturbances | Previous disturbances | Previous disturbances |
| Miscarriage | 44 | 23.78 | 12 | 48 | 56 |
| Prematurity | 07 | 3.78 | 04 | 16 | 11 |
| Normal delivery | 134 | 72.43 | 09 | 36 | 143 |
The study was conducted in accordance with the Declaration of Helsinki 1964 (as revised in Edinburgh 2000) and was approved by the Ethics Committee on Research of the Institute of Applied Biomedical Sciences of Cotonou, Benin under the number Dec.n°100/CER/ISBA-2016. Prior to enrollment, written consent was obtained from each participant who were informed of the study aim. The privacy rights of human subjects were observed.
## Screening of Gestational Diabetes Mellitus
Gestational diabetes mellitus was diagnosed in pregnant women following the protocol of the International Association of Diabetes and Pregnancy Study Group (IADPSG) [25]. Briefly, women between 24 and 28 weeks of gestation after overnight fasting were submitted to an oral glucose tolerance test (OGTT) and given 75 grams of glucose. Subjects were declared as positive for GDM when overnight fasting plasma glucose was ≥ 92 mg/dL (5.1 mmol/L), or 1-hour OGTT plasma glucose level was ≥ 180 mg/dL (10.0 mmol/L), or 2-hours OGTT plasma glucose level was ≥ 153 mg/dL (8.5 mmol/L).
The GDM screening revealed that 25 pregnant women have gestational diabetes, representing a percentage of $11.90\%$, and considered as the cases’ group. Pairing of these 25 newly GDM diagnosed women with non-diabetic pregnant women, according to age, body mass index and gestational age, allowed us to select 35 pregnant women without GDM, and considered as control group. Therefore, both groups of participants, twenty-five women with GDM and thirty-five age-matched and body mass index-matched and gestational age-matched pregnant controls were selected and submitted for blood collection and biochemical and immunological assays.
## Blood Samples
Blood samples were collected from each selected participant in appropriate tubes and immediately transported to the laboratory for biological assays within 2 hours. Immune parameters and glycated hemoglobin (HbA1c) were determined in whole blood. Plasma samples were immediately used for glucose determination. Serum obtained by low-speed centrifugation was used for biochemical assays.
## Biochemical Assays
Plasma glucose, total cholesterol, HDL cholesterol, triglycerides were measured by colorimetric enzymatic method using ELITech reagents (ELITech Group, Puteaux, France) according to manufacturer’s instructions. LDL-cholesterol was calculated using Friedewald method [26]. Total protein levels were determined by direct Biuret colorimetric method (ELITech Group, Puteaux, France). Aspartate aminotransferase (AST) and alanine aminotransferase (ALT) enzymatic activities and creatinine levels were determined by enzymatic kinetic assay (DiaSys reagents, Diagnostic Système GmbH, Germany). HbA1c concentration was calculated using a percentage of total hemoglobin, according to the manufacturer’s instructions (Reference 41190, Labkit Chemelex SA, Barcelona, Spain).
## Determination of Immune Cells
Immune cells were determined through the complete blood formula count using an automatic blood cell analyzer (Cell Dyn 3500, Abbott, France). These cells included total leukocytes, lymphocytes, monocytes, granulocytes and platelets (PLT).
## Statistical Analysis
Data analyses were performed using Graph Pad Prism 6.0 (Graph Pad Inc., CA, USA) and IBM® SPSS® Statistics (version 25.0). Values are means ± standard deviation or medians with interquartile ranges. Student’s t-test, Mann–Whitney U test and Chi-squared (χ2) test were used when appropriate. Pearson and Spearman correlations were used to determine the association between immunological and biochemical parameters. Receiver Operating Characteristics (ROC) curve analysis was used to assess the value of immunological to biochemical parameter ratios for predicting gestational diabetes mellitus and to obtain the best cutoff value using Youden’s index (sensitivity + specificity – 1). The odds ratios (ORs) are presented with $95\%$ confidence intervals (CI). Differences were considered significant with a two-tailed p value < 0.05.
## Biochemical Parameters in Women With GDM and Pregnant Controls
Biochemical parameters of women with GDM and pregnant controls are presented in Table 2. We observed that plasma fasting glucose ($p \leq 0.001$), total cholesterol ($$p \leq 0.001$$), LDL cholesterol ($$p \leq 0.015$$), triglyceride ($$p \leq 0.035$$) and total protein ($$p \leq 0.005$$) and HDL cholesterol ($$p \leq 0.001$$) levels significantly increased, while HDL-cholesterol level decreased in women with GDM compared to pregnant controls (Table 2). However, HbA1c and creatinine levels, and transaminase (AST and ALT) activities did not significantly differ between the two groups of women (Table 2).
**Table 2**
| Parameters | Pregnant control women (n = 35) | Women with GDM (n = 25) | p-value |
| --- | --- | --- | --- |
| Glucose (g/L) | 0.81 ± 0.03 | 1.16 ± 0.04 | 0.001 |
| HbA1c (%) | 5.65 ± 0.23 | 6.47 ± 0.48 | 0.451 |
| TC (g/L) | 1.53 ± 0.11 | 2.11 ± 0.31 | 0.001 |
| HDL-C (g/L) | 1.21 ± 0.15 | 0.35 ± 0.10 | 0.001 |
| LDL-C (g/L) | 0.91 (0.14-1.36) | 1.53 (1.15-1.68) | 0.015 |
| TG (g/L) | 1.30 ± 0.08 | 1.63 ± 0.18 | 0.035 |
| AST (UI/L) | 26.63 ± 2.64 | 26.20 ± 6.46 | 0.951 |
| ALT (UI/L) | 10.50 (9.75-15.25) | 14.00 (12.00-20.00) | 0.425 |
| Creatinine (mg/L) | 7.30 ± 0.45 | 8.92 ± 0.30 | 0.125 |
| Total proteins (g/L) | 71.60 ± 1.585 | 82.23 ± 3.32 | 0.005 |
## Immune Cells in Women With GDM and Pregnant Controls
Immune cell percentages of women with GDM compared to pregnant controls are shown in Figure 1. Total leukocytes ($$p \leq 0.045$$), lymphocytes ($$p \leq 0.015$$) and platelet numbers ($$p \leq 0.033$$) were significantly higher in women with GDM than in pregnant controls. However, no significant difference was observed in the numbers of monocytes and granulocytes between both groups (Figures 1, 2).
**Figure 1:** *Immune cell numbers in women with GDM (n = 25) and pregnant women without GDM as control group (n = 35): (A) total leukocytes, lymphocytes, monocytes and granulocytes; (B) Platelet numbers in women with GDM and pregnant controls. Values are means ± SD. *p values (p < 0.05) indicate significant difference between women with GDM and pregnant controls. Statistical analyses were performed using the Student’s t-test or Mann-Whitney test.* **Figure 2:** *Receiver Operating Characteristics (ROC) curve analysis of the value of ratios lymphocytes/HDL-C, granulocytes/HDL-C and monocytes/HDL-C for predicting gestational diabetes mellitus in pregnant women. HDL-C, high-density lipoprotein - cholesterol. n = 25 women with GDM, n= 35 pregnant women without GDM as control group.*
## Correlation Between Immune Cells and Biochemical Parameters
In pregnant control women, the correlation tests revealed a positive correlation between blood glucose with lymphocytes ($r = 0.89$; $$p \leq 0.03$$) and between lymphocytes with total cholesterol ($r = 0.50$; $$p \leq 0.04$$) (Table 4). Moreover, a positive correlation was found between monocytes with triglycerides ($r = 0.58$; $$p \leq 0.04$$). In contrast, a negative correlation was observed between monocytes and HDL-cholesterol levels (r = − 0.68; $$p \leq 0.007$$) (Table 3).
**Table 3**
| Immune cell subtypes | Glucose | Glucose.1 | TC | TC.1 | HDL-C | HDL-C.1 | LDL-C | LDL-C.1 | TG | TG.1 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Immune cell subtypes | r | p | r | p | r | p | r | p | r | p |
| Leucocytes | -0.50 | 0.45 | 0.37 | 0.22 | -0.19 | 0.46 | -0.42 | 0.09 | 0.37 | 0.24 |
| Lymphocytes | 0.89 | 0.03* | 0.50 | 0.04* | 0.36 | 0.17 | 0.11 | 0.68 | 0.16 | 0.73 |
| Monocytes | -0.45 | 0.40 | -0.43 | 0.08 | -0.68 | 0.007* | 0.36 | 0.17 | 0.58 | 0.04* |
| Granulocytes | -0.81 | 0.07 | -0.27 | 0.30 | 0.22 | 0.39 | -0.25 | 0.35 | -027 | 0.39 |
| Platelets | 0.68 | 0.08 | -0.33 | 0.20 | -0.36 | 0.18 | 0.20 | 0.45 | 0.33 | 0.27 |
In women with GDM, there was a positive correlation between glucose with leukocytes ($r = 0.70$; $$p \leq 0.03$$) on the one hand and between glucose with lymphocytes ($r = 0.67$; $$p \leq 0.02$$) on the other hand (Table 4). Also, we noticed a positive correlation between serum triglycerides with monocytes ($r = 0.87$; $$p \leq 0.045$$). In contrast, a negative correlation between granulocytes with HDL cholesterol was noted (r = − 0.90; $$p \leq 0.026$$).
**Table 4**
| Immune cell subtypes | Glucose | Glucose.1 | TC | TC.1 | HDL-C | HDL-C.1 | LDL-C | LDL-C.1 | TG | TG.1 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Immune cell subtypes | r | p | r | p | r | p | r | p | r | p |
| Leucocytes | 0.7 | 0.03* | 0.20 | 0.76 | 0.66 | 0.26 | 0.61 | 0.30 | 0.21 | 0.76 |
| Lymphocytes | 0.6 | 0.02* | 0.12 | 0.99 | 0.81 | 0.13 | -0.12 | 0.95 | 0.16 | 0.95 |
| Monocytes | -0.23 | 0.66 | 0.66 | 0.28 | 0.37 | 0.15 | 0.66 | 0.26 | 0.87 | 0.04* |
| Granulocytes | 0.72 | 0.23 | -0.48 | 0.5 | -0.9 | 0.02* | -0.41 | 0.51 | -0.2 | 0.78 |
| Platelets | -0.51 | 0.45 | 0.73 | 0.23 | 0.8 | 0.13 | 0.70 | 0.23 | 0.11 | 0.97 |
## Ratios of Immune Cells to Biochemical Parameters for Predicting GDM
Table 5 shows ratios between immunological to biochemical parameters in women with GDM and pregnant controls. We found that ratios of lymphocytes/HDL-C, monocytes/HDL-C and granulocytes/HDL-C were significantly higher in women with GDM than in pregnant controls ($$p \leq 0.001$$; $$p \leq 0.009$$ and $$p \leq 0.004$$ respectively).
**Table 5**
| Variables | Pregnant control women (n = 35) | Women with GDM (n = 25) | p-value |
| --- | --- | --- | --- |
| Lymphocytes to Glucose ratio | 2.48 ± 0.43 | 2.03 ± 0.59 | 0.125 |
| Lymphocytes to HDL-C ratio | 1.65 ± 0.86 | 7.38 ± 3.21 | 0.001* |
| Lymphocytes to LDL-C ratio | 2.53 ± 0.58 | 1.62 ± 0.22 | 0.591 |
| Lymphocytes to TG ratio | 1.50 ± 0.27 | 1.25 ± 0.60 | 0.469 |
| Monocytes to Glucose ratio | 0.34 ± 0.09 | 0.47 ± 0.13 | 0.424 |
| Monocytes to HDL-C ratio | 0.39 ± 0.05 | 1.73 ± 0.49 | 0.009* |
| Monocytes to LDL-C ratio | 0.41 ± 0.27 | 0.39 ± 0.15 | 0.701 |
| Monocytes to TG ratio | 0.25 ± 0.07 | 0.34 ± 0.09 | 0.117 |
| Granulocytes to Glucose ratio | 5.13 ± 1.24 | 3.33 ± 0.73 | 0.082 |
| Granulocytes to HDL-C ratio | 3.55 ± 2.02 | 14.18 ± 5.70 | 0.004* |
| Granulocytes to LDL-C ratio | 5.16 ± 3.42 | 3.27 ± 0.95 | 0.657 |
| Granulocytes to TG ratio | 3.17 ± 0.61 | 2.91 ± 0.73 | 0.229 |
As shown in Figure 2, a ROC curve analysis was used to assess the accuracy, sensitivity, specificity and value of the ratios of lymphocytes/HDL-C, granulocytes/HDL-C and monocytes/HDL-C for predicting GDM. The analysis showed that the lymphocytes/HDL-C ratio had a higher accuracy in predicting gestational diabetes mellitus (AUC = 0.859; $p \leq 0.001$; $95\%$ CI: 0.752 - 0.966) than the granulocytes/HDL-C ratio (AUC = 0.787; $p \leq 0.01$; $95\%$ CI: 0.654 -0.921) or the monocytes/HDL-C ratio (AUC = 0.716; $p \leq 0.01$; $95\%$ CI: 0.576 - 0.855) (Table 6). The optimal cutoff values of lymphocytes/HDL-C ratio, granulocytes/HDL-C ratio and monocytes/HDL-C ratio for predicting GDM were, respectively, 3.66 (sensitivity = $80.0\%$; specificity = $50.1\%$); 5.50 (sensitivity = $70.3\%$; specificity = $59.4\%$) and 1.56 (sensitivity = $60.9\%$; specificity = $50.0\%$) (Table 6). Odds ratios were used to assess the risk of GDM. We observed that pregnant women with the lymphocytes/HDL-C ratio greater than 3.66 had a 4-fold increased risk of developing GDM than those with lower ratios (odds ratio 4.00; $95\%$ CI: 1.094 – 14.630; $$p \leq 0.041$$) (Table 7).
## Discussion
Increasingly, the identification of biological parameters that can facilitate the prediction and early prognosis of gestational diabetes mellitus (GDM) has become a major concern for researchers. Given the complications associated with GDM in mothers, fetuses, newborns and adult offspring, an early diagnosis of GDM could help anticipate the care of pregnant women and limit the adverse effects. Therefore, the aim of this study was to investigate whether immunological parameters like immune cells, in conjunction with biochemical parameters, could be used to predict the risk of GDM.
As far as metabolic aspect is concerned, diabetes is known to be associated with biochemical and metabolic disturbance [27]. In the present study, we observed that HbA1C levels were normal and did not significantly differ between women with GDM and pregnant controls, although glycaemia remained high in women with GDM. The normal level of HbA1c might suggest that women with GDM were under an adequate metabolic control [28, 29]. However, the fact that their glycemia remained high could suggest that women with GDM had poor glycemia control despite their normal HbA1c levels. In fact, it’s important to note that pregnancy can impact HbA1c levels independently of glycemia. In a study conducted in pregnant women without GDM, it has been reported that the HbA1c level was low in early pregnancy and even more reduced at the end of pregnancy compared to non-pregnant women of the same age, suggesting that the pregnancy can significantly influence HbA1c levels regardless of glycemia [17]. In addition, this dichotomy could also be explained by the fact that women with GDM are newly diagnosed and they have not yet been subjected to any anti-diabetic treatment [28, 29].
It is commonly believed that GDM is associated with the modulation of lipid profiles. Although the results describing lipid profiles during normal pregnancy and GDM are diverse and extensive, the results have been inconsistent (15, 17–20, 30, 31). The present study showed that serum TC, LDL-C and TG levels increased significantly, while HDL-C levels decreased in women with GDM compared to pregnant controls. These results are in agreement with others who have also shown a significant decrease in HDL levels in pregnant women with glucose intolerance compared to control women [32, 33]. However, other studies have noted a significant rise in all lipids including HDL-C in women with GDM from the middle of the 2nd trimester of pregnancy to reach their peak at childbirth [34]. A meta-analysis showed that GDM was associated with elevated serum TG in the 3rd trimester of pregnancy, while serum HDL-C levels were significantly low in the 2nd and 3rd trimesters [30]. Indeed, during normal pregnancy, circulating lipids markedly increase, due to estrogen stimulation and insulin resistance [35]. High maternal fat accumulation during pregnancy has also been shown to be associated with both overeating and increased fetal lipogenesis and energy demand, necessary for childbirth and lactation (27, 36–38). In GDM, the situation appears to be similar as lipid levels increased during pregnancy. In fact, the increased levels of TG, TC, and LDL-C observed in GDM in the present study could lead to increased lipid storage in women with GDM, due to decreased lipolytic clearance of TG and increased hepatic lipase activity which appears to lead to increased HDL catabolism [39, 40].
As far as immunological aspects are concerned, there is evidence that gestational diabetes induces a profound variation of immune parameters [41]. Indeed, we have recently demonstrated that GDM was associated with high frequencies of total CD3+ and CD4+ T lymphocytes and B cells, suggesting that GDM could induce a concomitant activation of cellular and humoral immunity [3]. Likewise in the present study, we found that total leukocytes and lymphocytes in particular significantly increased in women with GDM as compared to control pregnant. There was no significant difference in granulocyte and monocyte numbers in both groups of women. These results are in agreement with previous work which reported that increased inflammatory cellular markers were associated with impaired glucose metabolism, insulin resistance and GDM (42–44). Evidently, the increased numbers of leukocytes and lymphocytes in women with GDM were consistent with the increase of a wide range of inflammatory metabolic markers such as TG, TC and LDL-cholesterol which together lead to insulin resistance (42–44). In addition, we noticed a significant increase in the number of platelets in women with GDM compared to control women. These results were similar to those of Lim et al. [ 23] who have shown that high platelet numbers was associated with an increased prevalence and risk of metabolic syndrome in children and adolescents.
All these observations prompted us to investigate the correlations between immunological and metabolic parameters during GDM. In fact, we observed, in both pregnant controls as well as in women with GDM, a positive correlation between blood glucose and total lymphocytes, between TG and monocytes; between TC and lymphocytes in pregnant controls and between blood glucose and total leukocytes in women with GDM on the one hand. On the other hand, we found a negative correlation between HDL-cholesterol and monocytes in pregnant controls and between HDL-cholesterol and granulocytes in women with GDM. All these correlations suggested that these parameters could be useful in predicting GDM.
In order to determine whether both parameters could help in the prediction of GDM, we evaluated the ratios between immune cells (lymphocytes, granulocytes and monocytes) and biochemical parameters (glucose, TC, TG and LDL-C). Interestingly, we found that lymphocytes/HDL-C, monocytes/HDL-C, and granulocytes/HDL-C ratios were significantly higher in women with GDM than in pregnant controls, suggesting that these ratios may certainly have significant value in predicting GDM. In fact, analysis of odds ratios indicated that only pregnant women with a lymphocytes/HDL-C ratio greater than 3.66 have a 4.0-fold higher risk of developing GDM than those with a lymphocyte-to-HDL-C ratio lower (odds ratio 4.00; $95\%$ CI: 1,094 - 14,630; $$p \leq 0.041$$).
Moreover, we would like to highlight the role of HDL-C in the present results as this lipoprotein seems to represent a central parameter to which immune cell frequencies could be added to more reliably determine the pathogenesis of GDM. In fact, HDL-C, as an anti-atherogenic lipoprotein, is recognized as a protective factor in atherosclerosis and inflammation [45, 46]. It has also been reported that TG/HDL-C ratio is a better marker for evaluating insulin resistance and diabetes [47]. In addition, previous studies have shown that immune cells can be used as novel markers for predicting inflammation, metabolic syndromes, diabetes and atherosclerosis [48]. Indeed, Pattanathaiyanon et al. [ 49] demonstrated that increased leucocyte numbers in early pregnancy may lead to a significant risk of GDM. Wolf et al. [ 50] have also previously reported that leucocyte numbers greater than 9100 cells/μL in early pregnancy were significantly associated with a heightened risk of GDM.
To the best of our knowledge, our study is the first which analyzes the predictive power of immuno-biochemical markers in GDM, through ratios of lymphocytes, monocytes and granulocytes and HDL-C levels. Among these markers, the lymphocytes/HDL-C ratio seems to have a strong predictive power in the onset and development of GDM, and these parameters are easily accessible in patients. Even though the sample size was relatively small in this study, the causative effect of immune-metabolic biomarkers in GDM needs to be more investigated by including, in addition to immune cells, other inflammatory markers such as cytokines and chemokines. This aspect could be addressed in future investigations.
## Conclusion
The present results constitute a major advance in the use of biological parameters for prediction of GDM. Immune cells associated with biochemical parameters appear as valuable markers which can allow to predict GDM. The interest of this study lies in the fact that these markers can be easily assessed on automatic devices which are usually found in medical analysis laboratories and that the interpretation of data is relatively simple. Pending future investigations that may involve other markers, we hope that the present results may be useful to clinicians and biologists specializing in the care of pregnant women.
## Data Availability Statement
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author.
## Ethics Statement
The study was conducted in accordance with the Declaration of Helsinki 1964 (as revised in Edinburgh 2000) and was approved by the Ethics Committee on Research of the Institute of Applied Biomedical Sciences (CER-ISBA) of Cotonou, Benin under the number Dec.n°100/CER/ISBA-2016. Prior to enrollment, written consent was obtained from each participant who were informed of the study aim. The privacy rights of human subjects were observed. The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
AF was in charge of major parts of technical aspects of work and participated in the manuscript writing. SF and MN participated in the technical work and participated in the interpretation of data. KM participated in the manuscript writing. AY designed the study, supervised the work, wrote the manuscript and established the collaborative aspects. All authors read and approved the final manuscript.
## Funding
This work did not receive any funding from any organization. The work was financed from the authors’ own funds. Products and reagents were purchased by the contribution of the authors themselves.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Wang C, Zhu W, Wei Y, Su R, Feng H, Lin L. **The Predictive Effects of Early Pregnancy Lipid Profiles and Fasting Glucose on the Risk of Gestational Diabetes Mellitus Stratified by Body Mass Index**. *J Diabetes Res* (2016) **2016** 1-8. DOI: 10.1155/2016/3013567
2. Kang X, Liang Y, Wang S, Hua T, Cui J, Zhang M. **Prediction Model Comparison for Gestational Diabetes Mellitus With Macrosomia Based on Risk Factor Investigation**. *J Matern Fetal Neonatal Med* (2021) **34**. DOI: 10.1080/14767058.2019.1668922
3. Fagninou A, Nekoua MP, Sossou D, Moutairou K, Fievet N, Yessoufou A. **Th2-Immune Polarizing and Anti-Inflammatory Properties of Insulin Are Not Effective in Type 2 Diabetic Pregnancy**. *J Immunol. Res* (2020) **2020** e2038746. DOI: 10.1155/2020/2038746
4. Seshiah V, Das AK, Balaji V, Joshi SR, Parikh MN, Gupta S. **Gestational Diabetes Mellitus–Guidelines**. *J Assoc Physicians India* (2006) **54**. DOI: 10.18203/2320-1770.ijrcog20151271
5. Kalra P, Kachhwaha CP, Singh HV. **Prevalence of Gestational Diabetes Mellitus and its Outcome in Western Rajasthan**. *Indian J Endocrinol Metab* (2013) **17**. DOI: 10.4103/2230-8210.113760
6. Jiwani A, Marseille E, Lohse N, Damm P, Hod M, Kahn JG. **Gestational Diabetes Mellitus: Results From a Survey of Country Prevalence and Practices**. *J Matern Fetal Neonatal Med* (2012) **25**. DOI: 10.3109/14767058.2011.587921
7. Eades CE, Cameron DM, Evans JMM. **Prevalence of Gestational Diabetes Mellitus in Europe: A Meta-Analysis**. *Diabetes Res Clin Pract* (2017) **129**. DOI: 10.1016/j.diabres.2017.03.030
8. Lee KW, Ching SM, Ramachandran V, Yee A, Hoo FK, Chia YC. **Prevalence and Risk Factors of Gestational Diabetes Mellitus in Asia: A Systematic Review and Meta-Analysis**. *BMC Pregnancy Childbirth* (2018) **18** 494. DOI: 10.1186/s12884-018-2131-4
9. Ogurtsova K, da Rocha Fernandes JD, Huang Y, Linnenkamp U, Guariguata L, Cho NH. **IDF Diabetes Atlas: Global Estimates for the Prevalence of Diabetes for 2015 and 2040**. *Diabetes Res Clin Pract* (2017) **128** 40-50. DOI: 10.1016/j.diabres.2017.03.024
10. England LJ, Dietz PM, Njoroge T, Callaghan WM, Bruce C, Buus RM. **Preventing Type 2 Diabetes: Public Health Implications for Women With a History of Gestational Diabetes Mellitus**. *Am J Obstet Gynecol* (2009) **200**. DOI: 10.1016/j.ajog.2008.06.031
11. Yang H, Wei Y, Gao X, Xu X, Fan L, He J. **Risk Factors for Gestational Diabetes Mellitus in Chinese Women-a Prospective Study of 16 286 Pregnant Women in China**. *Diabetes Med* (2000) **26**. DOI: 10.1111/j.1464-5491.2009.02845.x
12. Hope O, Ifeanyi OE, Braxton N AQ. **Investigation of Some Haematological Parameters in Pregnant Women With Gestational Diabetes at Federal Medical Center, Owerri, Imo State, Nigeria**. *Ann Clin Lab Res* (2019) **7** 1–6
13. Nekoua MP, Fachinan R, Atchamou AK, Nouatin O, Amoussou-Guenou D, Amoussou-Guenou MK. **Modulation of Immune Cells and Th1/Th2 Cytokines in Insulin-Treated Type 2 Diabetes Mellitus**. *Afr Health Sci* (2016) **16** 712. DOI: 10.4314/ahs.v16i3.11
14. Nekoua MP, Fachinan R, Fagninou A, Alidjinou EK, Moutairou K, Hober D. **Does Control of Glycemia Regulate Immunological Parameters in Insulin-Treated Persons With Type 1 Diabetes**. *Diabetes Res Clin Pract* (2019) **157**. DOI: 10.1016/j.diabres.2019.107868
15. Ategbo J-M, Grissa O, Yessoufou A, Hichami A, Dramane KL, Moutairou K. **Modulation of Adipokines and Cytokines in Gestational Diabetes and Macrosomia**. *J Clin Endocrinol Metab* (2006) **91**. DOI: 10.1210/jc.2006-0980
16. Seck A, Hichami A, Doucouré S, Diallo Agne F, Bassène H, Ba A. **Th1/Th2 Dichotomy in Obese Women With Gestational Diabetes and Their Macrosomic Babies**. *J Diabetes Res* (2018) **2018** 1-7. DOI: 10.1155/2018/8474617
17. Guermouche B, Yessoufou A, Soulimane N, Merzouk H, Moutairou K, Hichami A. **N-3 Fatty Acids Modulate T-Cell Calcium Signaling in Obese Macrosomic Rats**. *Obes Res* (2004) **12**. DOI: 10.1038/oby.2004.216
18. Soulimane-Mokhtari NA, Guermouche B, Yessoufou A, Saker M, Moutairou K, Hichami A. **Modulation of Lipid Metabolism by N–3 Polyunsaturated Fatty Acids in Gestational Diabetic Rats and Their Macrosomic Offspring**. *Clin Sci* (2005) **109**. DOI: 10.1042/CS20050028
19. Yessoufou A, Soulaimann N, Merzouk SA, Moutairou K, Ahissou H, Prost J. **N-3 Fatty Acids Modulate Antioxidant Status in Diabetic Rats and Their Macrosomic Offspring**. *Int J Obes* (2006) **30**. DOI: 10.1038/sj.ijo.0803211
20. Yessoufou A, Moutairou K. **Maternal Diabetes in Pregnancy: Early and Long-Term Outcomes on the Offspring and the Concept of “Metabolic Memory”**. *Diabetes Res* (2011) **2011**. DOI: 10.1155/2011/218598
21. Akboga MK, Canpolat U, Sahinarslan A, Alsancak Y, Nurkoc S, Aras D. **Association of Serum Total Bilirubin Level With Severity of Coronary Atherosclerosis is Linked to Systemic Inflammation**. *Atherosclerosis* (2015) **240**. DOI: 10.1016/j.atherosclerosis.2015.02.051
22. Balci KG, Balci MM, Sen F, Canpolat U, Akboga MK, Unal S. **The Role of Baseline Indirect Inflammatory Markers in Prediction of Response to Cardiac Resynchronisation Therapy**. *Kardiol Pol* (2015) **2015**. DOI: 10.5603/KP.a2015.0142
23. Lim H-J, Seo M-S, Shim J-Y, Kim K-E, Shin YH, Lee Y-J. **The Association Between Platelet Count and Metabolic Syndrome in Children and Adolescents**. *Platelets* (2015) **26**. DOI: 10.3109/09537104.2014.995613
24. Ertem AG, Yayla C, Acar B, Kirbas O, Unal S, Uzel Sener M. **Relation Between Lymphocyte to Monocyte Ratio and Short-Term Mortality in Patients With Acute Pulmonary Embolism**. *Clin Respir J* (2018) **12**. DOI: 10.1111/crj.12565
25. **International Association of Diabetes and Pregnancy Study Groups (IADPSG) Recommendations on the Diagnosis and Classification of Hyperglycemia in Pregnancy**. *Diabetes Care* (2010) **33**. DOI: 10.2337/dc09-1848
26. Benlian P, Béréziat G. *Revue Française Des Laboratoires - Présentation - EM Consulte.Html* (2001) **2001**. DOI: 10.1016/S0338-9898(01)80058-5
27. Mankuta D, Elami-Suzin M, Elhayani A, Vinker S. **Lipid Profile in Consecutive Pregnancies**. *Lipids Health Dis* (2010) **9**. DOI: 10.1186/1476-511X-9-58
28. Mahmoud F, Abul H, Omu A, Haines D. **Lymphocyte Sub-Populations in Gestational Diabetes**. *Am J Reprod Immunol* (2005) **53**. DOI: 10.1111/j.1600-0897.2004.00241.x
29. Viardot A, Grey ST, Mackay F, Chisholm D. **Potential Antiinflammatory Role of Insulin**. *Endocrinology* (2007) **148**. DOI: 10.1210/en.2006-0686
30. Ryckman K, Spracklen C, Smith C, Robinson J, Saftlas A. **Maternal Lipid Levels During Pregnancy and Gestational Diabetes: A Systematic Review and Meta-Analysis**. *BJOG Int J Obstet Gynaecol* (2015) **122**. DOI: 10.1111/1471-0528.13261
31. Shen H, Liu X, Chen Y, He B, Cheng W. **Associations of Lipid Levels During Gestation With Hypertensive Disorders of Pregnancy and Gestational Diabetes Mellitus: A Prospective Longitudinal Cohort Study**. *BMJ* (2016) **6** e013509. DOI: 10.1136/bmjopen-2016-013509
32. Aziz R, Mahboob T. **Lipid Profile and Serum Insulin Levels in Gestational Diabetes**. *J Dow Univ Health Sci JDUHS* (2008) **2**
33. Suresh E, Prasad G. **Study of Biochemical Changes in Gestational Diabetes Mellitus**. *Int J Appl Res* (2017) **3** 03-8
34. Herrera Martínez A, Palomares Ortega R, Bahamondes Opazo R, Moreno-Moreno P, Molina Puerta M-J, Gálvez-Moreno A. **Hyperlipidemia During Gestational Diabetes and Its Relation With Maternal and Offspring Complications**. *Nutr Hosp* (2018) **35** 698-706. DOI: 10.20960/nh.1539
35. Butte NF. **Carbohydrate and Lipid Metabolism in Pregnancy: Normal Compared With Gestational Diabetes Mellitus**. *Am J Clin Nutr* (2000) **71**. DOI: 10.1093/ajcn/71.5.1256s
36. Murphy SP, Abrams BF. **Changes in Energy Intakes During Pregnancy and Lactation in a National Sample of US Women**. *Am J Public Health* (1993) **83**. DOI: 10.2105/AJPH.83.8.1161
37. Di Cianni G, Miccoli R, Volpe L, Lencioni C, Ghio A, Giovannitti MG. **Maternal Triglyceride Levels and Newborn Weight in Pregnant Women With Normal Glucose Tolerance**. *Diabetes Med* (2005) **22**. DOI: 10.1111/j.1464-5491.2004.01336.x
38. Ghio A, Bertolotto A, Resi V, Volpe L, Di Cianni G. **Triglyceride Metabolism in Pregnancy**. *Adv Clin. Chem* (2011) **55**. DOI: 10.1016/B978-0-12-387042-1.00007-1
39. Herrera E, Lasunción MA, Gomez-Coronado D, Aranda P, López-Luna P, Maier I. **Role of Lipoprotein Lipase Activity on Lipoprotein Metabolism and the Fate of Circulating Triglycerides in Pregnancy**. *Am J Obstet Gynecol* (1988) **158**. DOI: 10.1016/0002-9378(88)90193-7
40. Meyer BJ, Stewart FM, Brown EA, Cooney J, Nilsson S, Olivecrona G. **Maternal Obesity Is Associated With the Formation of Small Dense LDL and Hypoadiponectinemia in the Third Trimester**. *J Clin Endocrinol Metab* (2013) **98**. DOI: 10.1210/jc.2012-3481
41. Siddiqui S, Waghdhare S, Jhaa S, Dubey S. **Role of Immunological Markers in Gestational Diabetes Mellitus-A Brief Review**. *Diabetes Metab Syndr* (2019) **13**. DOI: 10.1016/j.dsx.2018.07.018
42. Kim S, Nguon C, Guillard B, Duong S, Chy S, Sum S. **Performance of the CareStart**. *PloS One* (2011) **6** e28357. DOI: 10.1371/journal.pone.0028357
43. Demirtas L, Degirmenci H, Akbas EM, Ozcicek A, Timuroglu A, Gurel A. **Association of Hematological Indicies With Diabetes, Impaired Glucose Regulation and Microvascular Complications of Diabetes**. *Int J Clin Exp Med* (2015) **8**
44. Yılmaz ZV, Yılmaz E, İçer B, Küçüközkan T. **Association of Complete Blood Count Parameters With Gestational Diabetes Mellitus**. *Gynecol Obstet Reprod Med* (2016) **23** 1-5. DOI: 10.21613/GORM.2016.649
45. Murphy AJ, Woollard KJ, Hoang A, Mukhamedova N, Stirzaker RA, McCormick SPA. **High-Density Lipoprotein Reduces the Human Monocyte Inflammatory Response**. *Arterioscler Thromb Vasc Biol* (2008) **28**. DOI: 10.1161/ATVBAHA.108.168690
46. Rohatgi A. **High-Density Lipoprotein Function Measurement in Human Studies: Focus on Cholesterol Efflux Capacity**. *Prog Cardiovasc Dis* (2015) **58** 32-40. DOI: 10.1016/j.pcad.2015.05.004
47. Lin D, Qi Y, Huang C, Wu M, Wang C, Li F. **Associations of Lipid Parameters With Insulin Resistance and Diabetes: A Population-Based Study**. *Clin Nutr* (2018) **37**. DOI: 10.1016/j.clnu.2017.06.018
48. Chen T, Chen H, Xiao H, Tang H, Xiang Z, Wang X. **Comparison of the Value of Neutrophil to High-Density Lipoprotein Cholesterol Ratio and Lymphocyte to High-Density Lipoprotein Cholesterol Ratio for Predicting Metabolic Syndrome Among a Population in the Southern Coast of China**. *Diabetes Metab Syndr Obes Targets Ther* (2020) **13** 597-605. DOI: 10.2147/DMSO.S238990
49. Pattanathaiyanon P, Phaloprakarn C, Tangjitgamol S. **Comparison of Gestational Diabetes Mellitus Rates in Women With Increased and Normal White Blood Cell Counts in Early Pregnancy: White Blood Cells and Gestational Diabetes**. *J Obstet Gynaecol Res* (2014) **40**. DOI: 10.1111/jog.12306
50. Wolf M, Sandler L, Hsu K, Vossen-Smirnakis K, Ecker JL, Thadhani R. **First-Trimester C-Reactive Protein and Subsequent Gestational Diabetes**. *Diabetes Care* (2003) **26**. DOI: 10.2337/diacare.26.3.819
|
---
title: The impact of ethnicity and intra-pancreatic fat on the postprandial metabolome
response to whey protein in overweight Asian Chinese and European Caucasian women
with prediabetes
authors:
- Aidan Joblin-Mills
- Zhanxuan Wu
- Karl Fraser
- Beatrix Jones
- Wilson Yip
- Jia Jiet Lim
- Louise Lu
- Ivana Sequeira
- Sally Poppitt
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2022
pmcid: PMC10012149
doi: 10.3389/fcdhc.2022.980856
license: CC BY 4.0
---
# The impact of ethnicity and intra-pancreatic fat on the postprandial metabolome response to whey protein in overweight Asian Chinese and European Caucasian women with prediabetes
## Abstract
The “Thin on the Outside Fat on the Inside” TOFI_Asia study found Asian Chinese to be more susceptible to Type 2 Diabetes (T2D) compared to European Caucasians matched for gender and body mass index (BMI). This was influenced by degree of visceral adipose deposition and ectopic fat accumulation in key organs, including liver and pancreas, leading to altered fasting plasma glucose, insulin resistance, and differences in plasma lipid and metabolite profiles. It remains unclear how intra-pancreatic fat deposition (IPFD) impacts TOFI phenotype-related T2D risk factors associated with Asian Chinese. Cow’s milk whey protein isolate (WPI) is an insulin secretagogue which can suppress hyperglycemia in prediabetes. In this dietary intervention, we used untargeted metabolomics to characterize the postprandial WPI response in 24 overweight women with prediabetes. Participants were classified by ethnicity (Asian Chinese, $$n = 12$$; European Caucasian, $$n = 12$$) and IPFD (low IPFD < $4.66\%$, $$n = 10$$; high IPFD ≥ $4.66\%$, $$n = 10$$). Using a cross-over design participants were randomized to consume three WPI beverages on separate occasions; 0 g (water control), 12.5 g (low protein, LP) and 50 g (high protein, HP), consumed when fasted. An exclusion pipeline for isolating metabolites with temporal (T0-240mins) WPI responses was implemented, and a support vector machine-recursive feature elimination (SVM-RFE) algorithm was used to model relevant metabolites by ethnicity and IPFD classes. Metabolic network analysis identified glycine as a central hub in both ethnicity and IPFD WPI response networks. A depletion of glycine relative to WPI concentration was detected in Chinese and high IPFD participants independent of BMI. Urea cycle metabolites were highly represented among the ethnicity WPI metabolome model, implicating a dysregulation in ammonia and nitrogen metabolism among Chinese participants. Uric acid and purine synthesis pathways were enriched within the high IPFD cohort’s WPI metabolome response, implicating adipogenesis and insulin resistance pathways. In conclusion, the discrimination of ethnicity from WPI metabolome profiles was a stronger prediction model than IPFD in overweight women with prediabetes. Each models’ discriminatory metabolites enriched different metabolic pathways that help to further characterize prediabetes in Asian Chinese women and women with increased IPFD, independently.
## Introduction
The prevalence of type 2 diabetes (T2D) in China has increased drastically in recent decades, reaching epidemic proportions [1]. As mainland China represents the highest number of T2D and prediabetic cases worldwide, most concerning to the population is the increasing prevalence in young and lean adults, a worse profile when compared to, for example, more resilient European Caucasians [2, 3]. The susceptibility of Asian Chinese to T2D can be attributed to both genetic and lifestyle risk factors, with decreased exercise and westernized diets important [4, 5]. Likely to play a role in exacerbating T2D onset and metabolic syndrome [6, 7] is the preferential accumulation of both visceral adipose tissue (VAT) and ectopic organ fat, far more so than permissive subcutaneous adipose tissue (SAT).
A high VAT+organ fat to SAT ratio in outwardly lean individuals has been termed the “Thin on the Outside, Fat on the Inside” (TOFI) phenotype, and may help to explain the high T2D risk among Asian countries relative to other parts of the world [8, 9]. A high VAT/SAT ratio among Asian cohorts has previously been associated with hyperglycemia, hyperinsulinemia and/or insulin resistance, high blood pressure, and increased levels of plasma uric acid and triglycerides (TGs), regardless of body mass index (BMI) and/or a diabetic diagnosis (10–12). The accumulation of SAT in overweight individuals has been proposed to provide a beneficial sink for free fatty acids and TGs, reducing the exposure of organs to lipotoxic stress [13]. Several authors have proposed that when the lipid storage capacity of SAT becomes oversaturated, individuals may be predisposed to excess VAT and increased accumulation of ectopic fat in skeletal muscle, epicardial tissue, liver and pancreas (14–16). Why some individuals are more susceptible to ectopic fat accumulation has not yet been established.
Magnetic resonance imaging and spectroscopy (MRI and MRS) shows differences in SAT, VAT, and ectopic fat depots to be poorly identified using standard anthropometry techniques and total body fat assessments [9, 17]. A previous study from our laboratory conducted by Wu et al. demonstrated that SAT, VAT, pancreas and liver fat could be characterized through untargeted lipidomics and metabolomics methods [18]. Using MRI and MRS to characterize body fat depots in healthy and pre-diabetic Caucasian and Chinese women in the TOFI_Asia study, we subsequently identified a set of metabolomic markers that could successfully predict the fat levels of high VAT/SAT ratios, intra-pancreatic and intra-liver fat depots, with greater predication accuracy (cum R2) than typical clinical markers, cardiovascular disease (CVD) risk factors, and anthropometric measurements. Using partial least squares discriminative analysis (PLS-DA) we also identified a clear and robust separation in lipid and polar metabolite profiles that characterized the two ethnic cohorts [19].
Whilst intra-pancreatic fat deposition (IPFD) has been linked to suppressed insulin secretion in participants with impaired glucose tolerance (IGT) and impaired fasting glycaemia (IFG) [20], a recent review has reported a series of findings showing low level IPFD to be common even in metabolically healthy individuals [21]. The authors propose that a clearer distinction between fatty pancreas disease (FPD) and the non-disease related IPFD is required. A large-scale MRI study conducted by Wong et al. assessed FPD in a cohort of 685 Hong Kong residents (≥18 years of age), using the 95th percentile of IPFD in individuals without metabolic syndrome or alcohol abuse as a cutoff, and proposed $10.4\%$ as the IPFD cut point for FPD. However, there is no international standard for MR-assessed IPFD established at the time of writing. Their cutoff point identified $16.1\%$ prevalence of FPD in the Asian Chinese cohort [22]. With such a high tendency for Asian Chinese to accumulate VAT/SAT and develop FPD, the role of IPFD in pathogenesis of T2D remains unclear. IPFD may be quite widespread throughout the world’s population [21]. It raises the question of whether IPFD is part of the TOFI phenotype, and whether it plays a significant role in T2D progression from prediabetes, or whether other factors associated with the Asian Chinese ethnicity are more pertinent.
Several dietary intervention studies investigating prevention of T2D, weight loss and postprandial satiety have shown promising results for dairy products, in particular the whey protein fraction (23–25). Rich in essential amino acids (EAAs), non-essential amino acids (NEAAs) and essential branch-chain amino acids (BCAAs), whey protein isolate (WPI) can decrease postprandial hyperglycemia and promote insulin secretion in both healthy and diabetic individuals [26, 27]. We questioned whether differences in postprandial response to WPI would be detectable using a broader untargeted metabolomics platform, to compare prediabetic (raised fasting plasma glucose, FPG, 5.6–6.9 mmol/L) European Caucasian and Asian Chinese women with varying degrees of IPFD.
Due to the natural variability that exists among cohorts, identifying phenotypic biomarkers from large-scale omics datasets can be a difficult task. One method metabolomic researchers have begun using is support vector machine-recursive feature elimination (SVM-RFE) algorithm [28]. SVM-RFE machine learning is an optimal method for identifying phenotypic features from small cohort studies, due to the implementation of a SVM kernel trick; which projects variables from a 2-dimensional space to a 3-dimensional space, providing more options for optimizing decision boundary parameters for the classification of a phenotype [29]. Use of SVM with RFE not only allows for optimal discrimination of different classes in a model, but also identifies the important variables contributing towards the classification model, complementing traditional multivariate analysis [30]. This has proven to be an effective method for classifying different cancers from large scale genomic datasets [31, 32].
With the development and curation of open-source databases such as the Human Metabolome database (HMDB) and Kyoto Encyclopedia of Genes and Genomes (KEGG) [33, 34], the detection and annotation of hundreds of metabolites via hydrophilic interaction chromatography tandem mass spectrometry (HILIC–MS) facilitates the use of more holistic approaches to data processing, such as network topology and bio-ontology enrichment analysis’ (35–37). These bioinformatic tools provide researchers with a method for discerning biological relevancy from data complexity. Thus, we aimed to model the WPI metabolome response firstly for European Caucasian and Asian Chinese participants, and secondly for participants with lower and higher IPFD than the median value for our current cohort ($4.66\%$ IPFD), to determine their impact on metabolic markers and metabolic pathways associated with prediabetes. Notably the IPFD cut point was comparable with that of Singh and colleagues in their 2017 systematic review and meta-analysis [38].
## Study design
The presented work is a continuation of previously reported TOFI_Asia studies [7, 18]. The recruitment procedures, study design, WPI composition, participant characteristics, appetite biomarkers and gluco-corticoid hormone measurements have been summarized previously by Lim et al. [ 39]. In brief, this was an acute, randomized, three treatment cross-over study which investigated the postprandial WPI response of 12 Asian Chinese females and 12 European Caucasian females, aged 20–69 years and BMI 19.6–36.8 kg/m2. At the time of screening, all participants had prediabetes based on ADA criteria, with raised FPG of 5.6–6.9 mmol/L [40]. Magnetic resonance imaging (MRI) was used to quantify pancreatic fat in 20 participants, as detailed by Wu et al. [ 18]. low and high IPFD were defined as < and ≥ the cohort median of $4.66\%$, respectively. Each participant attended the Human Nutrition Unit (HNU), University of Auckland, New Zealand for three study visits over a three-week duration, with a minimum seven-day wash-out period. At each visit, a fasted baseline ($T = 0$ min) plasma sample was collected. Following consumption of the 280 mL test drinks comprising 0 g (water control, WC), 12.5 g (low protein, LP) and 50 g (high protein, HP) WPI, postprandial plasma samples were collected via a venous cannula at $T = 30$, 60, 120 and 240 min. No other foods or beverages were consumed during the study morning and participants followed a sedentary protocol.
## Sample preparation
Blood samples were stored at -80°C and batch analysed at the end of the study. For each sample, 100 μL plasma was mixed with 800 μL pre-chilled (-20°C) CHCl3:MeOH (50:50, v/v), agitated for 30 s and placed in a -20°C freezer for 60 min to allow protein precipitation. 400 μL milliQ water was subsequently added to each sample, agitated for 30 s and centrifuged at 11,000 rpm at 4°C for 10 min. From each biphasic separation, 200 μL of the upper aqueous layer was transferred to 2 mL micro-centrifuges and dried under a nitrogen stream before being stored at -80°C. To account for batch-to-batch variations, pooled quality control (QC) samples were prepared by pooling aliquots from each sample into a clean glass vial and stored at -80°C [41]. Pooled samples were combined from each batch and dispensed into separate 200 µl aliquots for drying under a nitrogen stream and -80°C storage. Dried polar extracts were reconstituted in 200 μL acetonitrile:H2O (50:50, v/v) before injection [18].
## Liquid chromatography tandem mass spectrometry conditions
Polar metabolites were analysed with an Accela 1250 quaternary UHPLC pump coupled to an Exactive Orbitrap mass spectrometry (Thermo Fisher Scientific, USA). Chromatographic separation was carried out at 25°C on a SeQuant® ZIC®-pHILIC 5 µm 2.1 mm × 100 mm column (Merck, Darmstadt, Germany) using the following solvent system: $A = 10$ mM ammonium formate in milliQ water, $B = 0.1$% formic acid in acetonitrile at a gradient program flow rate of 250 µL/min: 3–$3\%$ A (0.0–1.0 min), 3–$30\%$ A (1.0–12.0 min), 30–$90\%$ A (12.0– 14.5 min). $90\%$ A was maintained for 3.5 min followed by re-equilibration with $3\%$ A for 7 min. An injection volume of 2 µL was used. The electrospray probe was operated at room temperature (25°C) to avoid degradation of thermally labile compounds. External mass calibration of the Orbitrap prior to sample analysis was performed by the flow injection of the calibration mix solution according to manufacturer’s instruction. High resolution [25,000] data were acquired by full scan (m/z 55 to 1100) with a source voltage of 4000 V for both ESI + and ESI - ion modes. A capillary temperature of 325°C was set, and sheath, auxiliary, and sweep gas flow rates of 40, 10, and five arbitrary units were applied, respectively [42].
## Peak processing
Raw datafiles were converted to mzXML format using ProteoWizard tool MSconvert (v 3.0.1818). Open mzXML data files were pre-processed for features by untargeted peak filtering, peak alignment, and peak filling parameters with the XCMS package (v3.0.2) in the R programming environment (v3.2.2) [43, 44]. Features not detected in $100\%$ of the QC samples were excluded from the analysis and resulting extracted ion chromatograms were manually examined to filter poorly integrated peaks generated by the diffreport function. Signal drift and batch effects were corrected for by LOESS algorithm in the online W4M Galaxy environment, and feature filtering with a < $30\%$ coefficient of variation limit among QC samples was applied [45, 46].
## Feature exclusion pipeline
Before pre-processing, a K-nearest neighbours’ algorithm was used to impute values for two missing samples. A Shapiro-Wilk normality test was performed over the dataset and features with p-value ≤ 0.05 were log transformed to reduce skewed distributions. The resulting data set was mean centred, and summary scaled to normalise and standardise value ranges [47]. To determine likely features impacted by WPI intake, linear mixed-effect (LME) modelling with 10,000 permutations was performed with Meal, Time, Meal*Time, Age and BMI as fixed effects, and participants ID’s as random effects [48]. LME modelling was performed using the nlme package in the R environment. Benjamini-Hochberg false discovery adjustments were applied to p-values and resulting features with q-values ≥ 0.05 were filtered out.
To further filter out LME false positives, incremental area-under-the-curve (iAUC) calculations were performed against remaining features using the trapezoidal method in Graphpad Prism (v9.0.0); values were obtained for each meal per participant using the mean of each feature at $T = 0$ as a baseline and ignoring peaks less than $75\%$ of the height from minimum to maximum Y, and peaks defined by fewer than three adjacent time points. Feature net area values were subsequently used to measure fold-change (FC) between each WPI concentration and water control in MetaboAnalyst (v5.0.0) [49, 50]. Features with a significant log2 FC for the comparison of 12.5 g/0 g WPI alone were removed and remaining features with a significant log2 FC for both WPI meal comparisons and features significant for the 50 g/0 g WPI comparison alone, were shortlisted for modelling.
## Support vector machine-recursive feature elimination and cross-validation
Support vector machine–recursive feature elimination (SVM-RFE) procedures were implemented using the Github repository code provided by John Colby [51], and the e1071 package in R environment. Linear kernel SVM was used to rank all AUC-log2FC features by weight across a 10-fold cross-validation (CV) set for the classification of ethnicity (Caucasian and Chinese, $$n = 24$$) and pancreatic fat (IPFD_Low and IPFD_High, $$n = 20$$) as postprandial WPI metabolome models [31]. Feature ranks were averaged across all training set folds and the lowest weights were removed by “multiple RFE”, wherein reducing the feature total by half before introducing traditional “one-by-one” SVM-RFE [32]. Final ranking scores for top features were presented as averages across all training folds per model. To obtain generalized error estimates for testing folds, a radial basis function (RBF) kernel SVM was first applied to training folds with optimal tuning of the SVM hyperparameters (Cost and Gamma combinations) by internal 10-fold CV. Optimal parameters were used to train the SVM of each training fold before predicting corresponding test folds to calculate generalized error estimates [52]. All testing fold generalization error estimates were averaged, and the process repeated with varying numbers of top features as input at each iteration to determine optimal number of features for a given classification model. For each model matrix, a comparative confusion (dummy) matrix was created by reassigning each feature-columns y values to a random class ID using the Kutools package in Excel [53]. Ethnicity and IPFD dummy matrices were subject to all SVM-RFE procedures to calculate average feature ranks and generalized error estimates for comparison to respective query models. Top 20 features were annotated as metabolites and presented with SVM-RFE average ranking value and their respective Meal*Time LME interaction significance.
## Multivariate analysis
All multivariate analysis was performed in SIMCA software version 16 (Sartorius, Umeå Sweden). Principle component analysis (PCA) and partial least squares-discriminant analysis (PLS-DA) was applied to evaluate each classification model’s performance through mSVM-RFE ranked features across all plasma samples (i.e Time and Meal). PLS-DA models were subject to 100-fold permutations to evaluate separation performances and visualized [30].
## Metabolite annotations
A subset of positive and negative mode features were annotated using accurate mass and retention time matching to an in-house library of authentic standards. These were used as a quality control measure for annotating features with the metID package (GitHub) in R [54]. Features were subsequently annotated using metID’s hilic 0.0.2 database with accurate mass and a retention time window shift of 420 s as calculated from initial in-house library matching. Features that lacked an annotation had their molecular and co-eluting ions manually inspected for pseudo MS2 fragmentation patterns created from in-source fragmentation. Pseudo fragments and chemometric features (e.g., isotopes, multiple adducts) representing the same metabolites were annotated accordingly. Remaining features were annotated using the online Human Metabolome Database (HMDB) using an m/z ppm error of 15 for positive and negative ion modes [33].
## Metabolic network analysis and pathway enrichment
Machine learning and multivariate processing of metabolomics features determines the weight of each metabolite through vector values alone, which can be advantageous when characterizing metabolites with unknown annotations, but becomes restricted in application without consideration of a given metabolite’s biological ontology [55, 56]. To determine the relevance of Ethnicity and IPFD model metabolites to established networks and metabolic pathways, the KEGG IDs of top-ranking metabolites for each query model were subject to network construction and topological analysis using the MetScape plugin (v3.1.3) within Cytoscape (v3.1.3) [57], and pathway enrichment with Metabolite Set Enrichment Analysis (MSEA) in MetaboAnalyst [58]. Each query model was visualized as a network through Metscapes pathway-based network build function, and topological parameters were extracted using the network analyzer tool [59]. To identify the most important metabolites of the network, a relative betweenness centrality algorithm was applied, measuring the number of shortest paths going through a node for a given network. This takes into consideration the global network structure, rather than the immediate neighbor of the query node [35].
To identify pathways enriched from input metabolites, MSEA was implemented with hypergeometric testing through over-representation analysis (ORA) using the KEGG database with 80 registered Human metabolic pathways [34]. Enriched pathways p-values were subject to FDR correction and presented with relative impact values. Impact values were calculated autonomously through pathway topology and presented as a cumulative percentage representing the importance of all matched metabolites for the enriched query pathway [60].
## Baseline characteristics of the cohorts
All 24 women enrolled completed the three treatment arms. A subset of 20 women had MRI-assessed IPFD, and a range of 2.13 to $12.7\%$ IPFD was calculated. Mean (SD) age, BMI, FPG and IPFD are presented for both ethnicity (European Caucasian and Asian Chinese) and IPFD (Low and High) cohorts (Table 1). In comparison to the Caucasian cohort ($$n = 12$$), the Chinese cohort ($$n = 12$$) had a significantly lower age and BMI, but similar FPG and IPFD. When comparing IPFD cohorts, the High IPFD cohort ($$n = 10$$) had a significantly higher mean age, but similar BMI and FPG to the Low IPFD cohort.
**Table 1**
| Unnamed: 0 | Ethnicity | Ethnicity.1 | Ethnicity.2 | Intra-pancreatic fat | Intra-pancreatic fat.1 | Intra-pancreatic fat.2 |
| --- | --- | --- | --- | --- | --- | --- |
| | Caucasian(n=12) | Chinese(n=12) | P value | Low(n = 10) | High(n = 10) | P value |
| Age (years) | 54.7 ± 15.6 | 42.0 ± 10.8 | 0.038 | 35.9 ± 12.5 | 53.9 ± 8.1 | 0.002 |
| BMI (kg/m2) | 31.4 ± 4.2 | 26.9 ± 3.8 | 0.014 | 28.7 ± 3.8 | 28.2 ± 5.1 | 0.803 |
| FPG (mmol/L) | 6.1 ± 0.4 | 5.9 ± 0.3 | 0.257 | 5.8 ± 0.4 | 6.1 ± 0.4 | 0.147 |
| IPFD (%) | 5 ± 1.5* | 4.8 ± 2.4 | 0.864 | 3.5 ± 0.8 | 6.2 ± 2.1 | 0.002 |
## Postprandial metabolome responses
In total, 524 features (positive and negative ionization mode) were detected by HILIC-MS metabolomics. After batch correction, filtration and removal of noise, a matrix of 367 features was subject to linear mixed effect modeling to determine potential metabolites impacted by WPI intake over time, accounting for age and BMI. 216 features were significant (q-value ≤ 0.01) for Meal*Time interaction and were further processed as incremental area under the curve (iAUC) per meal and fold change (FC) calculated between meals 0 g WC/12.5 g LP and 0 g WC/50 g HP respectively (Supplementary Figure 1). FC values identified 92 up-regulated and 8 down-regulated features with the consumption of 12.5 g LP. After 50 g HP, 14 additional features were up-regulated, 11 features were down-regulated relative to 12.5 g LP, leaving a total of 125 features of interest.
## Ethnicity and intra-pancreatic fat deposition modelling
Support vector machine-recursive feature elimination (SVM-RFE) in combination with 10-fold cross-validations was implemented to classify Ethnicity (Caucasian and Chinese; $$n = 24$$) and IPFD (Low and High; $$n = 20$$) models with the mass spectrometer features identified in the postprandial whey protein response. Both Ethnicity and IPFD SVM-RFE models with respective Ethnicity_dummy and IPFD_dummy confusion models were plotted to compare their relative success in classification by the number of top-ranking features relative to generalized error estimates (Figure 1). Neither dummy model identified an optimal generalized error estimate with any given number of input features, while Ethnicity as a model was classified with an error estimate of 0.042 from the input of four top-ranking features, and IPFD was classified with an error estimate of 0.047 from the top 19 features. Therefore, the top 20 features were annotated discriminating both Caucasian and Chinese cohorts and Low and High IPFD cohorts.
**Figure 1:** *Generalized ten-fold cross validation error estimates for each testing set iteration per number of top SVM-RFE input features. Ten-fold CV error estimates depict the prediction accuracy of ethnicity and intra-pancreatic fat deposition (IPFD) models relative to number of top-ranking features input at each testing fold iteration. Ethnicity_dummy and IPFD_dummy models represent respective query model classes against shuffled y axis values (feature vectors), reflecting the classification legitimacy of each query model. Horizontal dotted line indicates threshold for optimal features input for an acceptable error estimate value (< 0.05) for model prediction. Red circle denotes the minimum number of input features to obtain an optimal error estimate to classify Ethnicity as a model. Red square denotes the minimum number of input features to obtain an optimal error estimate to classify IPFD as a model.*
The input of imidazolelactic acid, uric acid, N(ϵ)-methyl-lysine and L-cystine as model metabolites alone was sufficient in discriminating Caucasian and Chinese cohorts (Table 2). A closer look at the top-ranking features for Ethnicity by plotting Time and Meal identified Caucasians as having a base ($T = 0$) two-fold higher separation of imidazolelactic acid and N(ϵ)-methyl-lysine levels than the Chinese cohort, regardless of WPI (Supplementary Figure 2A, 2C). While both models had citric acid, creatine, glycine, imidazolelactic acid, N(ϵ)-methyl-lysine, octopamine, ornithine and uric acid as top-ranking metabolites (Table 2), only the Ethnicity model presented branched-chain amino acids valine, isoleucine and leucine within the classification model. The top IPFD metabolite, Uric acid, presented an average ranking score of 4.1 across all testing folds, higher than the ranking scores of top four ranking Ethnicity metabolites (2 - 3.8). The smaller the average ranking score for a feature, the greater its contribution towards a SVM classification. This indicates that the features significant for Meal*Time interaction had a greater strength for predicting ethnicity as a model than IPFD, and that uric acid is a stronger predictor of ethnicity than it is for IPFD.
**Table 2**
| Ethnicity model | Ethnicity model.1 | Ethnicity model.2 | IPFD model | IPFD model.1 | IPFD model.2 |
| --- | --- | --- | --- | --- | --- |
| Annotation | AvgRank | Meal*Time | Annotation | AvgRank | Meal*Time |
| Imidazolelactic acid | 2 | 6.56E-03 | Uric acid | 4.1 | 2.40E-03 |
| Uric acid | 2.6 | 2.40E-03 | N(ϵ)-Methyl-Lysine | 9.4 | 6.89E-04 |
| N(ϵ)-Methyl-Lysine | 2.7 | 6.89E-04 | Octopamine | 12.8 | 6.89E-04 |
| L-Cystine | 3.8 | 2.81E-02 | Unk m/z 156.895: rt 762s | 13.2 | 1.41E-02 |
| Glycine | 6.5 | 6.89E-04 | L-Lysine | 17.8 | 6.89E-04 |
| L-Valine | 14.8 | 6.89E-04 | Ornithine | 19.5 | 6.89E-04 |
| L-Arginine | 15.1 | 6.89E-04 | L-Tyrosine | 20.6 | 6.89E-04 |
| Citric acid | 19.1 | 6.89E-04 | Imidazolelactic acid | 21.7 | 6.56E-03 |
| Octopamine | 19.9 | 6.89E-04 | Glyceric acid | 22.1 | 4.53E-03 |
| Ornithine | 21.9 | 6.89E-04 | L-Glutamine | 23.7 | 1.30E-03 |
| 4-Aminobutanoate | 23.3 | 1.07E-02 | L-Histidine | 25.3 | 6.89E-04 |
| L-Methionine | 23.5 | 6.89E-04 | Phenylalanyl-valine | 25.3 | 6.89E-04 |
| L-Asparagine | 25.6 | 6.89E-04 | 2-Aminobutyrate | 27.5 | 6.89E-04 |
| Creatine | 25.6 | 6.89E-04 | L-Threonine | 29.4 | 6.89E-04 |
| Urea | 26.2 | 6.89E-04 | Glycine | 29.9 | 6.89E-04 |
| Unk m/z 199.038: rt 591s | 27.1 | 4.53E-03 | Creatine | 30 | 6.89E-04 |
| L-Isoleucine | 27.1 | 6.89E-04 | 3-Oxopentanoic acid | 30.8 | 4.53E-03 |
| Leucine | 27.2 | 6.89E-04 | Citric acid | 31.8 | 6.89E-04 |
| L-Glutamic Acid | 29.4 | 6.89E-04 | Unk m/z 778.516: rt 650s | 37.8 | 1.30E-03 |
| L-Phenylalanine | 32.7 | 6.89E-04 | L-Serine | 44.2 | 6.89E-04 |
The strength of both SVM-RFE models was further validated through PCA and PLS-DA. While Ethnicity PCA modelling (R2X 0.526) had better separation than IPFD (R2X 0.428), both were low in separation as an un-supervised model from their respective cumulative R2X values (Q2 0.297, Q2 0.284) (Supplementary Figures 4A, 4B). PLS-DA resulted in a strong separation of Ethnicity with permutation testing (R2Y 0.788, Q2 0.75) (Figure 2A), while the IPFD model PLS-DA presented a moderate separation with ranking features (R2Y 0.501, Q2 0.354). ( Figure 2B)
**Figure 2:** *PLS−DA analysis of SVM-RFE metabolome models for Ethnicity cohorts and IPFD classes. PLS−DA score plot (top) and 100 permutation tests (bottom) showing A) good separation and robust model for SVM-RFE ranked metabolites between Caucasian and Asian Chinese, and B) moderate separation and predictive modelling for SVM-RFE ranked metabolites between Low and High IPFD. FPD, fatty pancreas disease > 10.4% IPFD).*
## Metabolic network analysis of ethnicity model metabolites
From the top-ranking metabolites discriminating participants Ethnicity by postprandial WPI response; 19 metabolites presented KEGG IDs, in which 16 (Table 2) were available for metabolic network analysis and pathway enrichment when using the Metscape and KEGG databases against 80 human metabolism pathways. Ethnicity model metabolites constructed a single component metabolic network with 492 nodes and 563 edges (Figure 3). Network centrality identified glycine, L-glutamate, and L-phenylalanine as hub nodes among the array of metabolites, enzymes and genes based on their measure of degree (>10) and betweenness centrality (>0.1) (Table 3). Over representation analysis (ORA) identified three pathways as significant with 16 ethnicity associated metabolites (Table 4), This included alanine, aspartate and glutamate metabolism with four metabolites, arginine biosynthesis with four metabolites, and arginine and proline metabolism with five metabolites.
**Figure 3:** *Global metabolic network pathways identified with top ranking Ethnicity model metabolites. Nodes correspond to an identified KEGG compound/gene/reaction, and edges indicate a significant correlation between nodes. Annotated red hexagons represent input metabolites, pink hexagons represent network metabolites, blue circles represent encoding genes, green quadrangles represent associated enzymes and grey diamonds represent metabolic reactions. Metabolites in bold are hub metabolites by degree and betweenness centrality measures.* TABLE_PLACEHOLDER:Table 3 TABLE_PLACEHOLDER:Table 4
## Metabolic network analysis of intra-pancreatic fat deposition model metabolites
Of the 20 top-ranking postprandial WPI response metabolites separating participants with low IPFD from high IPFD, 16 had KEGG IDs, of which 13 (Table 2) were processed for pathway enrichment and metabolic network construction. IPFD model metabolites constructed a four-component metabolic network with 454 nodes and 504 edges (Figure 4). Network centrality identified glycine, L-tyrosine, L-serine, and L-glutamine as hub nodes based on their measure of degree (>10) and betweenness centrality (>0.1) (Table 5). ORA identified three pathways as significant with 13 IPFD metabolites (Table 6), This included glycine, serine, and threonine metabolism with five metabolites, glyoxylate and dicarboxylate metabolism with five metabolites, and aminoacyl-TRNA biosynthesis with seven metabolites.
**Figure 4:** *Global metabolic network pathways identified with top ranking IPFD model metabolites. Nodes correspond to an identified KEGG compound/gene/reaction, and edges indicate a significant correlation between nodes. Annotated red hexagons represent input metabolites, pink hexagons represent network metabolites, blue circles represent encoding genes, green quadrangles represent associated enzymes and grey diamonds represent metabolic reactions. Metabolites in bold are hub metabolites by degree and betweenness centrality measures.* TABLE_PLACEHOLDER:Table 5 TABLE_PLACEHOLDER:Table 6
## Discussion
To the best of our knowledge, this is the first study to model the postprandial WPI response to determine differences in metabolomic profiles between ethnic groups and/or groups with various levels of IPFD. It is also the first reported comparison of the postprandial metabolome responses associated with IPFD and ethnicity as an estimate of prediabetic risk factors. While most postprandial response studies measure differences associated with beverage composition or the capacity of a beverage to elicit a response [27, 48, 61, 62], this study utilized prior knowledge of WPI response as a basis for discerning differences between cohorts likely not apparent when comparing only basal metabolite levels.
Characterizing a metabolic response to a dietary intervention can determine an individual’s risk of developing a disease outcome [63]. We hypothesized that differences in postprandial WPI response could provide insight into how IPFD and/or ethnicity may contribute to adverse metabolic health outcomes. Therefore, we measured the postprandial response to WPI in a cohort of overweight women with prediabetes using untargeted metabolomics. Our aim was to discern differences in ethnicity- and IPFD-associated biomarkers following WPI intake. Notably, in the TOFI_Asia study, using PLS-DA analysis we had previously identified a clear and robust separation in fasting lipid and polar metabolite profiles that characterized the two ethnicity cohorts [19]. In our current data set we confirmed this separation through fasting polar metabolites; 3-methoxytyrosine, dihydrothymine, asymmetric dimethylarginine, valeric acid, 1-methyl-L-histidine and succinic acid. Students t-test analysis of these plasma metabolites identified a significant difference at baseline ($T = 0$) between ethnicity cohorts, and no temporal response to increasing doses of WPI. Also of note, we had previously identified dihydrothymine, valeric acid and 1-methyl-L-histidine as significantly different between ethnicity cohorts in the larger data set of Asian Chinese and European Caucasian men and women from the TOFI_Asia study [19]. The difference in basal level metabolites between Caucasian and Chinese cohorts indicates a clear disparity in metabolic pathways. Although these metabolites may contribute to resistance or susceptibility to developing T2D, they were omitted from the postprandial WPI models due to their lack of temporal profile.
We developed a feature selection pipeline that first identified a set of relevant features by linear mixed effect (LME) modeling for Meal*Time interactions with the inclusion of participants age and BMI, then removing LME false positives by incremental area under the curve-fold change (iAUC-FC) analysis, then comparing each participant WPI concentration dependent response to their respective postprandial water control response. Resulting features were used to model the differences in WPI response associated with ethnicity (Caucasian and Chinese) or IPFD (low IPFD and high IPFD) classes through a SVM-RFE algorithm. SVM-RFE found that the postprandial WPI metabolome response was most impacted by differences associated with ethnicity rather than IPFD, as average ranking values for metabolites classifying ethnicity as a model were greater (lower in value) than those for IPFD modelling. Additionally, the ethnicity model required far less metabolites to discriminate Caucasian and Chinese participants, than participants with low or high IPFD. The discriminant power of these models was confirmed by PLS-DA, where a strong separation between Caucasian and Chinese participants was achieved, in agreement with our previous data from the TOFI_Asia study [19], while only a moderate separation of IPFD classes from SVM-RFE ranking metabolites was produced. We set a threshold of ≥ $4.66\%$ as cut point for high IPFD, based on the cohort median, which is $0.16\%$ higher than the weighted mean obtained in a 2017 meta-analysis [38]. A value of $10.4\%$ for IPFD has previously been proposed as the cut point for fatty pancreas disease (FPD) by Wong et al. [ 22], based on a large cohort of Hong Kong Chinese. Whilst differences in MRI techniques and post scan analysis methods prevents robust between-study comparison, we note contribution of one participant in our current cohort with an IPFD value of $12.17\%$ which may represent FPD, and which is presented as a strong outlier in the IPFD PLS-DA.
Network topology and metabolite enrichment set analysis (MESA) in our current cohort provided a weighted evaluation of discriminatory metabolites by their annotated function in human metabolism; characterizing key network hubs and enriched metabolic pathways that separate both Caucasians and Chinese participants, and low IPFD from high IPFD participants by WPI metabolome response. While both models enriched different metabolic pathways, central to each ethnicity- and IPFD-WPI response network was the amino acid glycine. Considered a semi-essential amino acid, glycine has a key role in many metabolic pathways, including protein biochemistry, nitrogen metabolism, bile acid conjugation, and central nervous system signaling as a neurotransmitter [64]. Genome wide-association studies (GWAS) using both single-nucleotide polymorphisms (SNPs) and exome sequencing data have linked plasma glycine levels to genetic variants in the carbamoyl phosphate synthase 1 (CPS1) gene, which encodes the rate-limiting step of the urea cycle [48, 65]. This aligns with our network topology analysis, which identified the urea cycle as a key metabolic subnetwork from discriminatory metabolites for ethnicity. This included glycine and glutamic acid as key network hubs, and urea, arginine, asparagine, glutamic acid, ornithine, and 4-aminobutyric acid as contributing urea network nodes. The most impacted pathway discriminating Caucasians from Chinese was arginine and proline metabolism which is fundamental to urea production from arginine via arginase-1 activity [66]. The representation of urea cycle metabolites among the discriminatory model for ethnicity indicates differences in postprandial ammonia and nitrogen metabolism pathways between Caucasians and Chinese presented with a WPI challenge.
A closer look at the glycine profiles within ethnicity and IPFD cohorts found a prominent depletion of levels relative to increasing WPI concentration. Though the WPI beverages contained trace amounts of glycine (0.2 – 0.9 g) [39], it was apparent that high IPFD participants fed with 12.5 g of WPI were more sensitive to glycine depletion than participants with low IPFD. Glycine depletion was also more pronounced in the Chinese cohort’s response to 50 g or 12.5 g WPI than Caucasians. These results contrast other postprandial studies, wherein men or women with a BMI within the healthy range (≤ 25 kg/m2) had increased levels of glycine in response to a protein supplement [48, 65]. Interestingly, the BCAAs; valine, leucine, and isoleucine, were identified as top-ranking metabolites for the discrimination of ethnicity as a WPI response in women with prediabetes. Increased BCAAs have been positively associated with insulin resistance, diabetic nephropathy, and dyslipidemia in epidemiological studies [67, 68]. Furthermore, glycine levels have been negatively associated with metabolic syndrome, obesity, and diabetic complications (69–71). By use of the Zucker-fatty rat (ZFR) and Zucker-lean rat (ZLR) models, White et al., demonstrated that the raised levels of BCAAs associated with obesity generates excess levels of ammonia from increased BCAA transamination, leading to the recruitment of glycine as a carbon donor for the pyruvate-alanine cycle [72]. Our results indicate that the TOFI phenotype contributes towards the depletion of glycine more so than BMI, as the mean BMI of both IPFD cohorts was not significantly different, but their glycine response differed. The Chinese cohort, with a mean BMI of 26.9 ± 1.1 had greater sensitivity to glycine depletion than the Caucasian cohort, whose mean BMI of 31.4 ± 1.32 was significantly greater.
Although the IPFD model presented glycine within the top-ranking metabolites, it lacked the ranking of BCAAs as seen within the ethnicity model. Instead, the IPFD model ranked the aromatic amino acid tyrosine, whose presence has been an established biomarker for the exacerbation of insulin dysregulation in patients with non-alcoholic fatty liver disease (NAFLD) and diabetic nephropathy (73–76). Interestingly, the most impacted pathway discriminating low IPFD from high IPFD participants was glycine, serine, and threonine metabolism, as serine and threonine are key metabolites involved in the de novo synthesis of glycine [64], and along with glutamine, which was also ranked with high IPFD, are all associated with purine metabolism and the formation of excess uric acid [64, 77, 78]. Uric acid was the top-ranking metabolite discriminating low IPFD participants from high IPFD through their WPI metabolome response. Commonly associated with Gout formation in joints, uric acid has long been considered as an inert end-product from purine degradation [79]. But recent studies have shed light on uric acid as a regulator of key metabolic signaling pathways; stimulating fat storage and insulin resistance through adenosine monophosphate (AMP) deaminase, or promoting fat degradation and the decrease of gluconeogenesis through AMP activated protein kinase (80–82). While these attributes were once advantageous during times of food scarcity, it has been hypothesized that they have become detrimental to modern humans who lack a functional urate oxidase enzyme, resulting in higher levels of serum uric acid during an era of obesity [83]. With a decrease in both glycine and serine in response to WPI, along with increased levels of plasma uric acid in high IPFD participants, an inappropriate signal of fat storage and insulin resistance could be perpetuated towards further metabolic complications.
Two metabolites impacted by WPI that significantly contributed towards both SVM-RFE models, in particular the separation of Caucasians from Chinese participants, were imidazolelactic acid and N(ϵ)-methyl-lysine. Imidazolelactic acid is formed from the reduction of imidazole-pyruvate, which represents a key branch point in the source production of aspartate from the histidine transaminase pathway in *Escherichia coli* [84]. N(ϵ)-methyl-lysine is a poorly characterized metabolite, first detected in small concentrations by chromatographing plasma from fasting humans [85, 86]. Production of N(ϵ)-methyl-lysine has been reported in *Proteus vulgaris* bacteria [87]. Both metabolites were higher at basal level within the Caucasian cohort, with imidazolelactic acid decreasing in response to WPI concentrations and N(ϵ)-methyl-lysine increasing in response to WPI. Due to their unique profile of complete separation at basal level, but with a postprandial WPI response, and absence from the human metabolic pathway database, we speculate that they are associated with ethnic differences relating to gut microbiota profiles, as both *Escherichia coli* and *Proteus vulgaris* bacteria have been associated with the human gastrointestinal microbiome previously [88].
In conclusion, our study used untargeted metabolomics and postprandial WPI responses to identify a set of metabolites both common and disparate between IPFD and ethnicity models, using SVM-RFE modelling in overweight women with prediabetes. The discriminant power of these models demonstrated a strong separation of metabolites between European Caucasian and Asian Chinese participants, in agreement with our prior data from the TOFI_Asia study. Network analysis and pathway enrichments revealed several metabolites of the urea cycle, and arginine and proline metabolism that could differentiate between Caucasian and Chinese participants. Previously we identified a strong association of creatine for the Chinese cohort in our larger TOFI_Asia study, which was further validated in this study as a contributing metabolite for the discrimination of both ethnicity and IPFD WPI metabolome response models. Metabolites of the glycine, serine, and threonine metabolism were used in the discrimination of low and high IPFD classes, therefore implicating purine synthesis and uric acid production with increased IPFD levels. Betweenness centrality identified glycine as a key network hub for both ethnicity and IPFD metabolome networks, representing a difference in contribution towards urea cycle and uric acid metabolism, respectively. Glycine depletion was most prominent in the Chinese cohort relative to the Caucasian cohort, the latter notably with significantly higher BMI. Furthermore, the high IPFD cohort had a more prominent glycine depletion profile than the low IPFD cohort, despite comparable BMI. These results further characterize the obesity associated postprandial glycine profile established in the literature, but in addition brings to light the relative contribution that VAT and ectopic fat deposition have over BMI as an exacerbator of glycine depletion in a cohort with impaired fasting glucose. This study identified several unknown features as potential metabolites, annotated by their respective retention time and mass charge. These and other metabolites within this study, such as imidazolelactic acid and N(ϵ)-methyl-lysine, will need to be further characterized before they can be considered for systems biology modelling in future cohorts.
## Data availability statement
The datasets presented in this study can be found in the EMBL-EBI MetaboLights database with the identifier MTBLS5568. You can access the study here: https://www.ebi.ac.uk/metabolights/MTBLS5568.
## Ethics statement
This study was reviewed and approved by Auckland Health and Disabilities Ethics Committee (HDEC, Reference: 17/NTA/172), New Zealand. Study was registered with the Australia New Zealand Clinical Trial Registry (ANZCTR, Reference: ACTRN12618000145202. The patients/participants provided their written informed consent to participate in this study.
## Author contributions
All listed authors meet the requirements for authorship. IS led the clinical part of this study. IS, WY, JL and LL contributed to participants recruitment, sample collection and the clinical data. IS, and WY contributed to IPFD measurements. KF led and supervised the metabolomics part of this study. ZW conducted sample extraction and metabolomics profile acquisition. BJ and KF advised on statistical analysis. AJ-M conducted data analyses, results interpretation and drafted the manuscript. IS, JL, BJ, KF and SP critically revised and commented on the first and subsequent drafts. SP was the principal investigator for the Metabolic Health platform within the National Science Challenge High Value Nutrition (HVN) program, and fundraiser, who conceptualized and designed this study. All authors contributed to the article and approved the submitted version.
## Funding
This study was funded by the New Zealand National Science Challenge High Value Nutrition Program, Ministry for Business, Innovation and Employment (MBIE), Grant # 3710040.
## Acknowledgments
Fonterra Co-operative Group Ltd, New Zealand provided the whey protein isolate for this intervention.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcdhc.2022.980856/full#supplementary-material
## References
1. Reed J, Bain S, Kanamarlapudi V. **A review of current trends with type 2 diabetes epidemiology, aetiology, pathogenesis, treatments and future perspectives**. *Diabetes Metab Syndr Obes: Targets Ther* (2021.0) **14**
2. Li Y, Teng D, Shi X, Qin G, Qin Y, Quan H. **Prevalence of diabetes recorded in mainland China using 2018 diagnostic criteria from the American diabetes association: national cross sectional study**. *BMJ* (2020.0) **369** m997. PMID: 32345662
3. Shen X, Vaidya A, Wu S, Gao X. **The diabetes epidemic in China: An integrated review of national surveys**. *Endocr Pract* (2016.0) **22**
4. Hu C, Jia W. **Diabetes in China: Epidemiology and genetic risk factors and their clinical utility in personalized medication**. *Diabetes* (2018.0) **67** 3-11. PMID: 29263166
5. Pan A, Malik VS, Hu FB. **Exporting diabetes mellitus to Asia**. *Circulation* (2012.0) **126**
6. Shuster A, Patlas M, Pinthus JH, Mourtzakis M. **The clinical importance of visceral adiposity: a critical review of methods for visceral adipose tissue analysis**. *Br J Radiol* (2012.0) **85** 1-10. PMID: 21937614
7. Sequeira IR, Yip W, Lu L, Jiang Y, Murphy R, Plank L. **Visceral adiposity and glucoregulatory peptides are associated with susceptibility to type 2 diabetes: The TOFI_Asia study**. *Obesity* (2020.0) **28**
8. Wulan SN, Westerterp KR, Plasqui G. **Ethnic differences in body composition and the associated metabolic profile: A comparative study between asians and caucasians**. *Maturitas* (2010.0) **65**
9. Thomas EL, Parkinson JR, Frost GS, Goldstone AP, Doré CJ, Mccarthy JP. **The missing risk: MRI and MRS phenotyping of abdominal adiposity and ectopic fat**. *Obesity* (2012.0) **20** 76-87. PMID: 21660078
10. Rattarasarn C, Leelawattana R, Soonthornpun S, Setasuban W, Thamprasit A, Lim A. **Relationships of body fat distribution, insulin sensitivity and cardiovascular risk factors in lean, healthy non-diabetic Thai men and women**. *Diabetes Res Clin pract* (2003.0) **60** 87-94. PMID: 12706316
11. Rattarasarn C, Leelawattana R, Soonthornpun S, Setasuban W, Thamprasit A, Lim A. **Regional abdominal fat distribution in lean and obese thai type 2 diabetic women: relationships with insulin sensitivity and cardiovascular risk factors**. *Metab - Clin Exp* (2003.0) **52**
12. Tang L, Zhang F, Tong N. **The association of visceral adipose tissue and subcutaneous adipose tissue with metabolic risk factors in a large population of Chinese adults**. *Clin Endocrinol* (2016.0) **85** 46-53
13. Frayn K. **Adipose tissue as a buffer for daily lipid flux**. *Diabetologia* (2002.0) **45**
14. Rattarasarn C. **Dysregulated lipid storage and its relationship with insulin resistance and cardiovascular risk factors in non-obese Asian patients with type 2 diabetes**. *Adipocyte* (2018.0) **7** 1-10
15. Neeland IJ, Ross R, Després J-P, Matsuzawa Y, Yamashita S, Shai I. **Visceral and ectopic fat, atherosclerosis, and cardiometabolic disease: a position statement**. *Lancet Diabetes Endocrinol* (2019.0) **7**
16. Levelt E, Pavlides M, Banerjee R, Mahmod M, Kelly C, Sellwood J. **Ectopic and visceral fat deposition in lean and obese patients with type 2 diabetes**. *J Am Coll Cardiol* (2016.0) **68** 53-63. PMID: 27364051
17. Thomas EL, Saeed N, Hajnal JV, Brynes A, Goldstone AP, Frost G. **Magnetic resonance imaging of total body fat**. *J Appl Physiol* (1998.0) **85**
18. Wu ZE, Fraser K, Kruger MC, Sequeira IR, Yip W, Lu LW. **Untargeted metabolomics reveals plasma metabolites predictive of ectopic fat in pancreas and liver as assessed by magnetic resonance imaging: the TOFI_Asia study**. *Obes* (2021.0) **45**
19. Wu ZE, Fraser K, Kruger MC, Sequeira IR, Yip W, Lu LW. **Metabolomic signatures for visceral adiposity and dysglycaemia in Asian Chinese and Caucasian European adults: the cross-sectional TOFI_Asia study**. *Nutr Metab* (2020.0) **17** 95
20. Heni M, Machann J, Staiger H, Schwenzer NF, Peter A, Schick F. **Pancreatic fat is negatively associated with insulin secretion in individuals with impaired fasting glucose and/or impaired glucose tolerance: a nuclear magnetic resonance study**. *Diabetes/Metabolism Res Rev* (2010.0) **26**
21. Petrov MS, Taylor R. **Intra-pancreatic fat deposition: bringing hidden fat to the fore**. *Nat Rev Gastroenterol Hepatol* (2022.0) **19**
22. Wong VW-S, Wong GL-H, Yeung DK-W, Abrigo JM, Kong AP-S, Chan RS-M. **Fatty pancreas, insulin resistance, and β-cell function: A population study using fat-water magnetic resonance imaging**. *Am Coll Gastroenterol | ACG* (2014.0) **109**
23. Mcgregor RA, Poppitt SD. **Milk protein for improved metabolic health: a review of the evidence**. *Nutr Metab* (2013.0) **10** 46
24. Mignone LE. **Whey protein: The “whey” forward for treatment of type 2 diabetes**. *World J Diabetes* (2015.0) **6** 1274. PMID: 26516411
25. Poppitt SD, Proctor J, Mcgill A-T, Wiessing KR, Falk S, Xin L. **Low-dose whey protein-enriched water beverages alter satiety in a study of overweight women**. *Appetite* (2011.0) **56**
26. Stevenson EJ, Allerton DM. **The role of whey protein in postprandial glycaemic control**. *Proc Nutr Soc* (2018.0) **77** 42-51. PMID: 28942740
27. Nilsson M, Holst JJ, Björck IM. **Metabolic effects of amino acid mixtures and whey protein in healthy subjects: studies using glucose-equivalent drinks**. *Am J Clin Nutr* (2007.0) **85** 996-1004. PMID: 17413098
28. Heinemann J, Mazurie A, Tokmina-Lukaszewska M, Beilman GJ, Bothner B. **Application of support vector machines to metabolomics experiments with limited replicates**. *Metabolomics* (2014.0) **10**
29. Cevikalp H. **Best fitting hyperplanes for classification**. *IEEE Trans Pattern Anal Mach Intell* (2017.0) **39**
30. Worley B, Powers R. **Multivariate analysis in metabolomics**. *Curr. Metab* (2012.0) **1** 92-107
31. Guyon I, Weston J, Barnhill S, Vapnik V. **Gene selection for cancer classification using support vector machines**. *Mach Learn* (2002.0) **46** 389-422
32. Duan KB, Rajapakse JC, Wang H, Azuaje F. **Multiple SVM-RFE for gene selection in cancer classification with expression data**. *IEEE Trans Nanobioscience* (2005.0) **4**
33. Wishart DS, Feunang YD, Marcu A, Guo AC, Liang K, Vázquez-Fresno R. **HMDB 4.0: the human metabolome database for 2018**. *Nucleic Acids Res* (2018.0) **46**
34. Kanehisa M, Goto S, Sato Y, Furumichi M, Tanabe M. **KEGG for integration and interpretation of large-scale molecular data sets**. *Nucleic Acids Res* (2012.0) **40**
35. Wang L, Hou E, Wang L, Wang Y, Yang L, Zheng X. **Reconstruction and analysis of correlation networks based on GC–MS metabolomics data for young hypertensive men**. *Anal Chim Acta* (2015.0) **854** 95-105. PMID: 25479872
36. Tebani A, Bekri S. **Paving the way to precision nutrition through metabolomics**. *Front Nutr* (2019.0) **6**
37. Kulmanov M, Smaili FZ, Gao X, Hoehndorf R. **Semantic similarity and machine learning with ontologies**. *Briefings Bioinf* (2021.0) **22**
38. Singh RG, Yoon HD, Wu LM, Lu J, Plank LD, Petrov MS. **Ectopic fat accumulation in the pancreas and its clinical relevance: A systematic review, meta-analysis, and meta-regression**. *Metabolism* (2017.0) **69** 1-13. PMID: 28285638
39. Lim JJ, Sequeira IR, Yip WCY, Lu LW, Barnett D, Cameron-Smith D. **Postprandial glycine as a biomarker of satiety: A dose-rising randomised control trial of whey protein in overweight women**. *Appetite* (2021.0) **169** 105871. PMID: 34915106
40. American Diabetes A. **2. classification and diagnosis of diabetes: Standards of medical care in diabetes-2021**. *Diabetes Care* (2021.0) **44**
41. Xu J, Begley P, Church SJ, Patassini S, Hollywood KA, Jüllig M. **Graded perturbations of metabolism in multiple regions of human brain in alzheimer's disease: Snapshot of a pervasive metabolic disorder**. *Biochim Biophys Acta (BBA)* (2016.0) **1862**
42. Cabrera D, Kruger M, Wolber F, Roy N, Totman J, Henry C. **Association of plasma lipids and polar metabolites with low bone mineral density in Singaporean-Chinese menopausal women: A pilot study**. *Environ Res Public Health* (2018.0) **15** 1045
43. Smith CA, Want EJ, O'Maille G, Abagyan R, Siuzdak G. **XCMS: processing mass spectrometry data for metabolite profiling using nonlinear peak alignment, matching, and identification**. *Analytical Chem* (2006.0) **78**
44. 44
Team RC. R: A language and environment for statisitical computing. Vienna, Austria: R Foundation for Statistical Computing (2021).. *R: A language and environment for statisitical computing* (2021.0)
45. van der Kloet FM, Bobeldijk I, Verheij ER, Jellema RH. **Analytical error reduction using single point calibration for accurate and precise metabolomic phenotyping**. *J Proteome Res* (2009.0) **8**
46. Giacomoni F, Le Corguille G, Monsoor M, Landi M, Pericard P, Petera M. **Workflow4Metabolomics: a collaborative research infrastructure for computational metabolomics**. *Bioinformatics* (2015.0) **31**
47. Lyman Ott ML. *An introduction to statistical methods and data analysis. 6th ed* (2010.0)
48. Chungchunlam SMS, Henare SJ, Ganesh S, Moughan PJ. **Dietary whey protein influences plasma satiety-related hormones and plasma amino acids in normal-weight adult women**. *Eur J Clin Nutr* (2015.0) **69**
49. Gannon MC, Nuttall FQ, Westphal SA, Neil BJ, Seaquist ER. **Effects of dose of ingested glucose on plasma metabolite and hormone responses in type II diabetic subjects**. *Diabetes Care* (1989.0) **12**
50. Xia J, Psychogios N, Young N, Wishart DS. **MetaboAnalyst: a web server for metabolomic data analysis and interpretation**. *Nucleic Acids Res* (2009.0) **37**
51. Colby J. *An r implementation of the multiple support vector machine recursive feature elimination feature ranking algorithm GitHub Repository2011*
52. Ambroise C, Mclachlan GJ. **Selection bias in gene extraction on the basis of microarray gene-expression data**. *Proc Natl Acad Sci* (2002.0) **99**
53. Arjaria SK, Rathore AS, Cherian JS, N P, Kautish S, Peng S-L. **Chapter 13 - kidney disease prediction using a machine learning approach: A comparative and comprehensive analysis**. *Demystifying big data, machine learning, and deep learning for healthcare analytics* (2021.0)
54. Shen X, Wu S, Liang L, Chen S, Contrepois K, Zhu ZJ. **metID: a r package for automatable compound annotation for LC-MS-based data**. *Bioinformatics* (2021.0) **38**
55. Dührkop K, Nothias L-F, Fleischauer M, Reher R, Ludwig M, Hoffmann MA. **Systematic classification of unknown metabolites using high-resolution fragmentation mass spectra**. *Nat Biotechnol* (2021.0) **39**
56. Park CY, Wong AK, Greene CS, Rowland J, Guan Y, Bongo LA. **Functional knowledge transfer for high-accuracy prediction of under-studied biological processes**. *PloS Comput Biol* (2013.0) **9**. PMID: 23516347
57. Karnovsky A, Weymouth T, Hull T, Tarcea VG, Scardoni G, Laudanna C. **Metscape 2 bioinformatics tool for the analysis and visualization of metabolomics and gene expression data**. *Bioinformatics* (2012.0) **28**
58. Xia J, Wishart DS. **Metabolomic data processing, analysis, and interpretation using MetaboAnalyst**. *Curr Protoc Bioinf* (2011.0) **34**
59. Basu S, Duren W, Evans CR, Burant CF, Michailidis G, Karnovsky A. **Sparse network modeling and metscape-based visualization methods for the analysis of large-scale metabolomics data**. *Bioinformatics* (2017.0) **33**
60. Chen H-H, Tseng YJ, Wang S-Y, Tsai Y-S, Chang C-S, Kuo T-C. **The metabolome profiling and pathway analysis in metabolic healthy and abnormal obesity**. *Int J Obes* (2015.0) **39**
61. Akhavan T, Luhovyy BL, Brown PH, Cho CE, Anderson GH. **Effect of premeal consumption of whey protein and its hydrolysate on food intake and postmeal glycemia and insulin responses in young adults**. *Am J Clin Nutr* (2010.0) **91**
62. Sridonpai P, Prachansuwan A, Praengam K, Tuntipopipat S, Kriengsinyos W. **Postprandial effects of a whey protein-based multi-ingredient nutritional drink compared with a normal breakfast on glucose, insulin, and active GLP-1 response among type 2 diabetic subjects: a crossover randomised controlled trial**. *J Nutr Sci* (2021.0) **10**
63. Loo RL, Zou X, Appel LJ, Nicholson JK, Holmes E. **Characterization of metabolic responses to healthy diets and association with blood pressure: application to the optimal macronutrient intake trial for heart health (OmniHeart), a randomized controlled study**. *Am J Clin Nutr* (2018.0) **107**
64. Razak MA, Begum PS, Viswanath B, Rajagopal S. **Multifarious beneficial effect of nonessential amino acid, glycine: A review**. *Oxid Med Cell Longev* (2017.0) **2017** 1-8
65. Skov K, Oxfeldt M, Thøgersen R, Hansen M, Bertram HC. **Enzymatic hydrolysis of a collagen hydrolysate enhances postprandial absorption rate–a randomized controlled trial**. *Nutrients* (2019.0) **11** 1064. PMID: 31086034
66. Schwingshackl L, Hoffmann G. **Comparison of high vs. Normal/Low protein diets on renal function in subjects without chronic kidney disease: A systematic review and meta-analysis**. *PloS One* (2014.0) **9**. PMID: 24852037
67. Yang P, Hu W, Fu Z, Sun L, Zhou Y, Gong Y. **The positive association of branched-chain amino acids and metabolic dyslipidemia in Chinese han population**. *Lipids Health Dis* (2016.0) **15**
68. Chen T, Ni Y, Ma X, Bao Y, Liu J, Huang F. **Branched-chain and aromatic amino acid profiles and diabetes risk in Chinese populations**. *Sci Rep* (2016.0) **6** 20594. PMID: 26846565
69. Li X, Sun L, Zhang W, Li H, Wang S, Mu H. **Association of serum glycine levels with metabolic syndrome in an elderly Chinese population**. *Nutr Metab* (2018.0) **15**
70. Adeva-Andany M, Souto-Adeva G, Ameneiros-Rodríguez E, Fernández-Fernández C, Donapetry-García C, Domínguez-Montero A. **Insulin resistance and glycine metabolism in humans**. *Amino Acids* (2018.0) **50** 11-27. PMID: 29094215
71. Alves A, Bassot A, Bulteau A-L, Pirola L, Morio B. **Glycine metabolism and its alterations in obesity and metabolic diseases**. *Nutrients* (2019.0) **11** 1356. PMID: 31208147
72. White PJ, Lapworth AL, Mcgarrah RW, Kwee LC, Crown SB, Ilkayeva O. **Muscle-liver trafficking of BCAA-derived nitrogen underlies obesity-related glycine depletion**. *Cell Rep* (2020.0) **33** 108375. PMID: 33176135
73. Adams SH. **Emerging perspectives on essential amino acid metabolism in obesity and the insulin-resistant state**. *Adv Nutr* (2011.0) **2**
74. Luo H-H, Li J, Feng X-F, Sun X-Y, Li J, Yang X. **Plasma phenylalanine and tyrosine and their interactions with diabetic nephropathy for risk of diabetic retinopathy in type 2 diabetes**. *BMJ Open Diabetes Res Care* (2020.0) **8**
75. Miwa Kawanaka M, Nishino K, Oka T, Urata N, Nakamura J, Suehiro M. **Tyrosine levels are associated with insulin resistance in patients with nonalcoholic fatty liver disease**. *Hepat Med: Evid Res* (2015.0) **29**
76. Vangipurapu J, Stancáková A, Smith U, Kuusisto J, Laakso M. **Nine amino acids are associated with decreased insulin secretion and elevated glucose levels in a 7.4-year follow-up study of 5,181 Finnish men**. *Diabetes* (2019.0) **68**
77. Mahbub M, Yamaguchi N, Takahashi H, Hase R, Amano H, Kobayashi-Miura M. **Alteration in plasma free amino acid levels and its association with gout**. *Environ Health Prev Med* (2017.0) **22**
78. Yü TA-F, Adler M, Bobrow E, Gutman AB. **Plasma and urinary amino acids in primary gout, with special reference to glutamine**. *J Clin Invest* (1969.0) **48**
79. Kanbay M, Jensen T, Solak Y, Le M, Roncal-Jimenez C, Rivard C. **Uric acid in metabolic syndrome: From an innocent bystander to a central player**. *Eur J Internal Med* (2016.0) **29** 3-8. PMID: 26703429
80. Choi Y-J, Shin H-S, Choi HS, Park J-W, Jo I, Oh E-S. **Uric acid induces fat accumulation**. *Lab Invest* (2014.0) **94**
81. Cicerchi C, Li N, Kratzer J, Garcia G, Roncal-Jimenez CA, Tanabe K. **Uric acid-dependent inhibition of AMP kinase induces hepatic glucose production in diabetes and starvation: evolutionary implications of the uricase loss in hominids**. *FASEB J* (2014.0) **28**
82. Lanaspa MA, Cicerchi C, Garcia G, Li N, Roncal-Jimenez CA, Rivard CJ. **Counteracting roles of AMP deaminase and AMP kinase in the development of fatty liver**. *PloS One* (2012.0) **7**. PMID: 23152807
83. Nyborg AC, Ward C, Zacco A, Chacko B, Grinberg L, Geoghegan JC. **A therapeutic uricase with reduced immunogenicity risk and improved development properties**. *PloS One* (2016.0) **11**. PMID: 28002433
84. Moro J, Tomé D, Schmidely P, Demersay T-C, Azzout-Marniche D. **Histidine: A systematic review on metabolism and physiological effects in human and different animal species**. *Nutrients* (2020.0) **12** 1414. PMID: 32423010
85. Perry T, Stedman D, Hansen S. **A versatile lithium buffer elution system for single column automatic amino acid chromatography**. *J Chromatogr* (1969.0) **38**
86. Perry TL, Diamond S, Hansen S. **ϵ-N-Methyl lysine: an additional amino-acid in human plasma**. *Nature* (1969.0) **222**
87. Ambler RP, Rees MW. **ε-N-Methyl-lysine in bacterial flagellar protein**. *Nature* (1959.0) **184**
88. Guentzel MN, Baron S. **Chapter 26: Escherichia, klebsiella, enterobacter, serratia, citrobacter, and Proteus**. *Medical microbiology 4th edition* (1996.0)
|
---
title: Low Risk for Developing Diabetes Among the Offspring of Individuals With Exceptional
Longevity and Their Spouses
authors:
- Iva Miljkovic
- Ryan Cvejkus
- Ping An
- Bharat Thyagarajan
- Kaare Christensen
- Mary Wojczynski
- Nicole Schupf
- Joseph M. Zmuda
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2022
pmcid: PMC10012150
doi: 10.3389/fcdhc.2022.753986
license: CC BY 4.0
---
# Low Risk for Developing Diabetes Among the Offspring of Individuals With Exceptional Longevity and Their Spouses
## Abstract
Little is known about the risk of type 2 diabetes (T2D) among the offspring of individuals with exceptional longevity. We determined the incidence of and potential risk and protective factors for T2D among the offspring of probands and offspring’s spouses (mean age=60 years, range 32-88 years) in the Long Life Family Study (LLFS), a multicenter cohort study of 583 two-generation families with a clustering of healthy aging and exceptional longevity. Incident T2D was defined as fasting serum glucose ≥126 mg/dl, or HbA1c of ≥$6.5\%$, or self-reported with doctor diagnosis of T2D, or the use of anti-diabetic medication during a mean follow-up 7.9 ± 1.1 years. Among offspring ($$n = 1105$$) and spouses ($$n = 328$$) aged 45-64 years without T2D at baseline visit, the annual incident rate of T2D was 3.6 and 3.0 per 1000 person-years, respectively, while among offspring ($$n = 444$$) and spouses ($$n = 153$$) aged 65+ years without T2D at baseline, the annual incident rate of T2D was 7.2 and 7.4 per 1000 person-years, respectively. By comparison, the annual incident rate of T2D per 1000 person-years in the U.S. general population was 9.9 among those aged 45-64, and 8.8 among those aged 65+ years (2018 National Health Interview Survey). Baseline BMI, waist circumference, and fasting serum triglycerides were positively associated with incident T2D, whereas fasting serum HDL-C, adiponectin, and sex hormone binding globulin were protective against incident T2D among the offspring (all $P \leq 0.05$). Similar associations were observed among their spouses (all $P \leq 0.05$, except sex hormone binding globulin). In addition, we observed that among spouses, but not offspring, fasting serum interleukin 6 and insulin-like growth factor 1 were positively associated with incident T2D ($P \leq 0.05$ for both). Our study suggests that both offspring of long-living individuals and their spouses, especially middle-aged, share a similar low risk for developing T2D as compared with the general population. Our findings also raise the possibility that distinct biological risk and protective factors may contribute to T2D risk among offspring of long-lived individuals when compared with their spouses. Future studies are needed to identify the mechanisms underlying low T2D risk among the offspring of individuals with exceptional longevity, and also among their spouses.
## Introduction
Preserved glucose tolerance and insulin sensitivity has been recognized as one of the major biological pathways to maintaining health and achieving exceptional longevity [1] [2]. Comparisons of nonagenarians and centenarians with “younger” individuals aged 65-89 years from the same study indicate that improved life expectancy is consistently associated with favorable fasting serum glucose and preserved insulin sensitivity (3–8). Previous reports have also shown that centenarians may have preserved glucose tolerance even comparable with that of healthy young individuals [9]. However, little is known about the incidence of and risk or protective factors for type 2 diabetes (T2D) among the offspring of individuals with exceptional longevity. Lifestyle, environmental, and genetic factors likely contribute to the complex phenotype of exceptional longevity. Although, several cross-sectional studies have reported a low prevalence of T2D among the offspring of individuals with exceptional longevity (10–12), similar prospective studies are lacking. When studying offspring of individuals with exceptional longevity, it is a challenge to select a proper comparison group that may adequately represent the general population that also have measured important confounders. The spouses of the offspring could offer an advantageous approach as they are expected to be close in age, and matched on environment, socioeconomic and geographical background [11, 13, 14]. Interestingly, previous studies have reported that the mortality of spouses marrying into longevity-enriched families is substantially lower than the mortality in the general population [15].
The Long Life Family Study (LLFS) is a multicenter cohort study of two-generation families with a clustering of healthy aging and exceptional survival. We have previously shown that diabetes prevalence was lower and glucose metabolism seem to be healthier in LLFS probands and offspring as well as their spouses compared to similar aged persons in the other epidemiologic cohorts [16], making the LLFS a valuable longitudinal cohort to study the glucose metabolism and diabetes and how they relate to longevity. Our primary objective was to compare the rate of T2D incidence among the offspring of LLFS probands and offspring spouses, as well as compare these rates with the rates observed in general population. Our secondary objective was to identify potential risk and protective factors for developing T2D among the offspring and compare them with their spouses.
## Study Population: The Long-Life Family Study
The Long-Life Family Study (LLFS) is a family-based cohort study of exceptional longevity that recruited families at four study centers (Boston, Massachusetts; New York, New York; Pittsburgh, Pennsylvania; and Denmark). The three U.S. field centers used Center for Medicare and Medicaid Services lists of Medicare enrollees to mail a recruitment brochure. The initial file included people who were at least 79 years old on January 1, 2005; had no recorded date of death; were not in the end-stage renal disease or hospice programs; and lived in zip codes within 3 hours driving distance one of the study centers. A pilot mailing tested the yield of families recruited from mailing to individuals in their 80’s and higher age strata. Based on these yields, subsequent mailings targeted those age 89 and older. Study participants were also recruited from local communities using mailed brochures, posters, web-based media and newspaper advertisements. Additional mailing lists were obtained through voter registries and purchased public domain lists from various commercial vendors. The University of Southern Denmark field center identified individuals who would be ages 90 and above during the study recruitment period through the Danish National Register of Persons, which contains current information on names, including past names such as maiden names for women, addresses, place of birth, marriages, and vital status [17]. Second, using information on the place of birth and the names, parish registers available in regional archives were searched to locate the parents of the elderly individuals in order to identify sibships. Based on the above information, potentially eligible families were identified, and contact was made with potential probands to further assess the family’s eligibility for and willingness to participate in the LLFS using criteria parallel to that used in the United States. The criteria for the final recruitment were based on having 2 or more siblings who were exceptionally long-lived (aged 80+ years in the US and 90+ years in Denmark). Families were primarily white and met the following eligibility criteria: 1) enrolled one long-lived participant (proband) aged ≥90, 2) enrolled ≥1 sibling of the proband, 3) enrolled ≥1 offspring of either the proband or the proband’s sibling, and 4) the proband generation had a clustering of members with exceptional survival based on a family longevity selection score [18]. Briefly, the Family Longevity Selection Score (FLoSS) was used to rank a potential proband sibship on their combined exceptionality of survival [18]. A family’s entry into LLFS required at least one living member of the proband sibship with “decisional capacity”, a living offspring, and a proband sibship FLoSS of at least 7. The FLoSS is designed to be negative for families with less than average longevity, with higher scores representing increasingly exceptional longevity. For example, the FLoSS for a five-person sibship with each sib at the 91st percentile of longevity for his or her birth cohort is about 7; and if all five sibs were at the 98th percentile, its FLoSS would be nearly 15. As an indication of the exceptionality of LLFS sibships, fewer than $1\%$ of families in the Framingham Heart Study sample have a FLoSS > 7 [18]. The two generations in the LLFS were labeled as the proband generation (long-lived individual and their enrolled siblings) and the offspring generation (all enrolled offspring of individuals in the proband generation). The LLFS also recruited as many spouses as possible. Spouses of the proband generation were recruited only if their biological children were enrolled in the study. Spouses of the offspring generation were recruited as spousal controls, if they lived in the same household as their offspring pair. In total, 4559 long-lived probands and their siblings ($$n = 1445$$), their offspring ($$n = 2329$$) and spouse controls ($$n = 785$$) were recruited from 2006 to 2009. Other characteristics of family eligibility, recruitment, and composition have been previously described [16, 18].
From 2014 to 2017, surviving participants were invited to take part in an in-home follow-up examination. In total, 3,198 individuals participated in the follow-up exam ($86\%$ of survivors). Within the offspring generation (including spousal controls), 3,114 individuals participated in the baseline exam, and 2,219 individuals participated in the follow-up exam ($74.4\%$ of survivors).
Written informed consent was obtained from each LLFS participant using forms and procedures approved by each participating institution’s Institutional Review Board.
## Interview and Measurements
Sociodemographic factors, including date of birth, gender, race, and education, smoking status, difficulty with activities of daily living, health status, and chronic conditions were determined by interview, in addition a blood sample was collected, in the participant’s home at both visits. History or presence of type 2 diabetes, heart disease, stroke, cancer, emphysema, and chronic obstructive pulmonary disease was based on self-report of a physician’s diagnosis. All prescription and non-prescription medications were examined in their original containers for a medication inventory. Weight, waist circumference, and systolic and diastolic blood pressure were measured at both visits.
Biological markers evaluated in the current analysis included: lipids (triglycerides, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), total cholesterol), pro- and anti- inflammatory biomarkers (adiponectin, interleukin 6, high-sensitivity C-reactive protein), insulin-like growth factor 1 (IGF-1), and sex hormone binding globulin (SHBG). All biomarkers were measured in fasting blood by a central laboratory at the University of Minnesota. Participants were asked to fast for at least eight hours prior to the blood draw.
T2D was defined as fasting serum glucose ≥126 mg/dl, or glycated hemoglobin (HbA1c) of ≥$6.5\%$, or self-reported with doctor diagnosis of T2D, or the use of anti-diabetic medication, and incident T2D was determined during a mean follow-up 7.9 ± 1.1 years.
## Statistical Analysis
Baseline and metabolic characteristics of offspring and their spousal controls were summarized and compared using age- and sex- adjusted means and standard errors for continuous traits, and frequencies and percent for categorical traits. Each trait was assessed for normality and transformed as needed before statistical comparison.
The associations of each of these traits with odds of incident T2D were determined in the offspring group using generalized linear models, incorporating a logit link and an exchangeable correlation structure within families to account for genetic relatedness of individuals. Associations of incident T2D were determined in the spousal control group using multiple logistic regression. Covariates included in the models were age, sex, BMI, field center, and lifestyle factors including alcohol intake and physical activity. These covariates were determined a priori and were forced into each model. Odds ratios of incident diabetes are presented per standard deviation of each continuous trait or per level of each categorical trait, in offspring and their spouses separately.
## Baseline Characteristics
For the current analyses, we focused on baseline offspring and their spouses without T2D only. 2,889 non-diabetics participated in the baseline exam and 2,080 of those individuals also participated in the follow-up exam ($74.9\%$ of survivors). Offspring and their spouses who participated in the follow-up exam were more likely to be women, physically active, and to consume alcohol, and less likely to be obese compared with those who did not participate in the follow-up (all $P \leq 0.05$, data not shown).
Table 1 shows selected baseline characteristics of LLFS offspring and their spouses without T2D who returned for Exam 2. Offspring and their spouses were on average 60 years old. At baseline, offspring were less likely to be men ($42\%$ vs $52\%$, $P \leq 0.0001$), to be married ($75.4\%$ vs $98.9\%$, $P \leq 0.001$), to drink four of more alcoholic drinks per week ($42.1\%$ vs $54.1\%$, $P \leq 0.0001$), and to walk three or more hours per week ($76.7\%$ vs. $81.1\%$, $$P \leq 0.016$$, but more likely to achieve a high school or higher education than their spouses ($91.1\%$ vs. $94.5\%$, $$P \leq 0.0036$$) Offspring also had slightly lower level of fasting plasma LDL-C at baseline compared to their spouses (123.6 vs. 127 mg/dl, $$P \leq 0.046$$).
**Table 1**
| Mean ± SD | Offspring (n=1585) | Offspring Spouses (n=495) | p-value* |
| --- | --- | --- | --- |
| Median [Q1, Q3] | Offspring (n=1585) | Offspring Spouses (n=495) | p-value* |
| Men | 679 (42.8%) | 260 (52.5%) | <0.0001 |
| Age (years) | 60.1 ± 8.0 | 60.5 ± 8.3 | 0.46 |
| (range) | (32–88) | (36–88) | 0.46 |
| BMI (kg/m2) | 27.0 ± 4.7 | 26.9 ± 4.3 | 0.33 |
| Waist Circumference (cm) | 93.0 ± 13.3 | 93.8 ± 12.2 | 0.36 |
| Married | 1195 (75.4%) | 488 (98.9%) | <0.0001 |
| High School Education or Higher | 1497 (94.5%) | 450 (91.1%) | 0.0036 |
| Current Smoker | 128 (8.1%) | 43 (8.7%) | 0.89 |
| Drinks per Week | 666 (42.1%) | 266 (54.1%) | |
| ≥ 4 | 666 (42.1%) | 266 (54.1%) | <0.0001 |
| Walking hours per week | 1208 (76.7%) | 400 (81.1%) | |
| ≥ 3 | 1208 (76.7%) | 400 (81.1%) | 0.016 |
| Triglycerides (mg/dl)** | 90.0 [66.0, 130.0] | 94.0 [67.0, 133.0] | 0.52 |
| HDL (mg/dl)** | 61.6 ± 17.7 | 59.5 ± 16.4 | 0.29 |
| LDL (mg/dl)** | 123.6 ± 33.0 | 127.0 ± 34.2 | 0.046 |
| Adiponectin (ng/mL) | 9977.0 [6580.0, 14408.0] | 9208.0 [6296.0, 13005.0] | 0.23 |
| IL-6 (pg/mL) | 0.7 [0.4, 1.1] | 0.7 [0.5, 1.2] | 0.71 |
| hsCRP (mg/L) | 1.2 [0.6, 2.4] | 1.1 [0.5, 2.4] | 0.46 |
| IGF1 (ng/mL) | 133.0 [105.0, 177.0] | 130.0 [107.0, 170.0] | 0.13 |
| SHBG (nmol/L | 53.0 [38.0, 75.0] | 54.0 [39.0, 74.0] | 0.22 |
## Total Cumulative and Annualized T2D Incidence
Among offspring ($$n = 1585$$) and spouses ($$n = 495$$) without diabetes at study entry, 58 ($3.7\%$) and 19 ($3.8\%$) developed incident T2D, respectively. Annual incidence rate was 4.6 cases per 1000 person-years among the offspring, and 4.7 cases per 1000 person-years among the spousal controls. Among offspring ($$n = 1105$$) and spouses ($$n = 328$$) without diabetes and aged 45-64 years the annual incident rate of T2D per 1000 person-years was 3.6 and 3.0, respectively, while among offspring ($$n = 444$$) and spouses ($$n = 153$$) without diabetes and aged 65 years and older at study entry, the annual incident rate of T2D per 1000 person-years was 7.2 and 7.4, respectively. There was no difference between the T2D incidence rate in the U.S. Study Centers vs. Denmark, in offspring nor spouses in the total sample or by age groups (data not shown).
## Risk Factors for Incident T2D in Offspring and Their Spouses
Table 2 present the odds of incident T2D per standard deviation greater level of baseline risk factor for continuous variables or presence of the risk factor for dichotomous variables. BMI (OR=2.55, $P \leq 0.0001$), waist circumference (OR=2.52, $P \leq 0.0001$), and fasting serum triglycerides (OR=1.56, $$P \leq 0.0042$$), at baseline visit were positively, whereas HDL-C (OR=0.56, $P \leq 0.0042$), adiponectin (OR=0.60, $P \leq 0.0024$), and sex hormone-binding globulin (OR=0.55, $P \leq 0.0019$), were inversely associated with incident T2D among the offspring. Similar associations were observed in their spouses, although there was no significant association between sex hormone-binding globulin and incident T2D (Table 3). Additionally, among the offspring spouses, we observed that circulating interleukin 6 (OR=1.63, $$P \leq 0.048$$) and insulin-like growth factor 1 (OR=1.79, $$P \leq 0.04$$) were positively associated with incident T2D (Table 3). Among both the offspring and their spouses none of the lifestyle factors measured in our study were associated with incident T2D.
## Discussion
We [16] and others [8, 19] have previously shown that individuals with exceptional longevity and their offspring exhibit a healthier metabolic profile including glycemic control when compared with the general population. A surprising finding from the Long Life Family *Study is* a marked survival advantage among spouses to offspring of long-lived families when compared with the general population [15, 20]. We have hypothesized that the risk of developing incident T2D among the offspring of long-lived individuals would be similar to the risk among their spouses, and lower than the risk in the general population. Indeed, we found that both offspring and their spouses, especially middle-aged, may share a similar, low rate of T2D, which is lower than the rates observed in general adult population from the most recent National Health Interview Survey [21]. Specifically, the U.S. National Health Interview Survey [21] reports T2D incidence as 4.3 per 1000 person-years among individuals aged 18-44 years, 9.9 per 1000 person-years among those aged 45-64 year, 8.8 per 1000 person-years in those 65 and older years, and 6.9 per 1000 person-years among their total sample (individuals aged 18 years and older). While hereditary influences, such as genetic and/or epigenetic mechanisms, are likely to play a significant role in T2D risk [22], shared environmental factors are also likely to contribute to our findings. Compatibility in lifestyle (23–26) and leisure preferences [27] among couples who live together have been previously documented. In our study, spouses were actually more likely to be physically active and to report moderate alcohol consumption when compared with the offspring, and both of these lifestyle factors may reduce T2D risk (28–30). Thus, it is possible that the protective familial genetic and biological factors impact glucose homeostasis among the offspring, whereas in their spouses the healthier lifestyles may in fact contribute to their lower T2D risk. Another possible explanation might be assortative mating - a nonrandom mating and sexual selection based on similar phenotypes and henceforth similar genotypes. It has been previously proposed that assortative mating of members from families enriched for longevity could preserve longevity across generations by increasing the likelihood of transmission of rare variants with a recessive effect [31]. Further studies in the offspring and their spouses in the LLFS could help us better understand a lower risk of T2D associated not only with being an offspring of a long-lived individuals, but also with being married into a long-lived family.
A second objective of our analysis was to identify potential risk and protective factors for T2D among the offspring of long-living individuals, and to compare these factors to those in spouses. Significant associations between the anthropometric measures, lipid and lipoproteins, and adiponectin and incident T2D were very similar among the offspring and their spouses. However, our findings raise the possibility that biological factors contributing to T2D might be different in the offspring as compared to their spouses. We found that pro-inflammatory and insulin-like growth factor signaling biomarkers may play a greater role in T2D among spouses than in the offspring of the exceptional survivors. In contrast, sex hormone-binding globulin might be uniquely protective against T2D among the offspring of exceptional survivors. Further research is needed to identify the molecular mechanisms and pathways that are underlying this low T2D risk and favorable glucose control in offspring of exceptionally long-lived individuals and their spouses, as the LLFS is currently generating transcriptomic, methylomic, proteomic, and metabolomic data longitudinally.
The biology underlying the association between glucose metabolism and longevity is still under extensive investigation, but lipid and lipoprotein metabolism [32] and adipokine signaling pathways [33, 34] have emerged as a possible mechanistic link. In our study, greater fasting serum TG and lower fasting serum HDL-C and adiponectin appear to have similar relationships with T2D in offspring and their spouses. Dysregulation of adipokines is associated with insulin resistance, hyperglycemia, dyslipidemia, and T2D [35], as well as with wasting syndromes, such as cachexia [36], suggesting that adipose tissue endocrine function might be essential for maintaining whole-body energy homeostasis with aging. Furthermore, genetic manipulation of adipose tissue promotes longevity in mice, suggesting a possible role in longevity [37]. We did not have measures of body composition in the current study to address the role of adipose tissue. Our findings suggest that the offspring of long-living individuals might be protected against the detrimental effects of IL-6 and IGF-1 on the risk of T2D, in contrast to their spouses. The observed association between IL-6 and incident T2D in spousal controls is consistent with a recent meta-analysis of 15 prospective studies which reported that higher levels of IL-6 were significantly associated with a higher risk of incident T2D [38]. The link between IL-6 and longevity is less explored, though some studies suggest that greater circulating IL-6 is associated with a higher risk of mortality, including among successfully aging individuals [39]. Although, it is known that IGF-1 is involved in glucose metabolism [40], prospective studies have been inconclusive, some reporting no association between serum IGF-1 and the risk of T2D [41], while others found positive [42, 43] or even inverse [44] associations. In animal models, down-regulation of the GH/IGF-1/insulin system significantly prolongs lifespan, but data in humans are inconsistent [45]. Finally, low circulating level of the sex hormone-binding globulin is another well-known predictor of the risk of T2D [46]. In our study, sex hormone-binding globulin was protective against T2D, but only among the offspring of long-lived individuals. The link between sex hormone-binding globulin and longevity is still not well understood (47–49).
There are several limitations of our study. Our analyses are potentially limited by the smaller sample size of the spousal control group, and the fact that family history of longevity was not collected for spousal controls. It is possible that some of the spousal controls may have come from long-lived families. Our cohort included only European ancestry individuals, and thus, findings may not apply to other ethnic/race groups. Furthermore, data on dietary intake and dietary patterns were not collected and specific foods and overall dietary patterns may be associated with T2D risk. Additionally, biomarkers were only measured at baseline. Finally, we only studied several candidate biomarkers, which were available in our study, but there are many other relevant biological factors which may be associated with T2D. On the other hand, a major strength of our study is that it includes the spouses of the offspring. By sharing the same environment, it is less likely that environmental factors would have confounded the observed differences between the offspring and their spouses. However, although our records indicate that the spouses shared the physical address with their offspring pair, it is possible that they were spending some time apart, and thus, might have not shared the same environment all the time.
In conclusion, our study suggests that offspring of exceptionally long-lived individuals and their spouses, especially at middle-age, may share a similar, low risk for developing T2D compared with the general population. Our findings also raise the possibility that distinct biological risk factors may contribute to T2D risk among offspring of exceptional survivors compared with their spouses. Additional studies are needed to identify the biological mechanisms underlying low T2D risk among the offspring of exceptionally long-lived individuals, but also among their spouses.
## Data Availability Statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics Statement
The studies involving human participants were reviewed and approved by each participating institution’s Institutional Review Board. The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
Manuscript concept: IM and JZ. Data analysis: RC. Manuscript writing: IM. Interpretation of data and manuscript editing and critical review: all authors. All authors contributed to the article and approved the submitted version.
## Funding
National Institute on Aging-National Institutes of Health (Grants U01-AG023746, U01-AG023712, U01-AG023749, U01-AG023755, U01-AG023744, and U19 AG063893).
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Longo VD, Finch CE. **Evolutionary Medicine: From Dwarf Model Systems to Healthy Centenarians**. *Science* (2003.0) **299**. DOI: 10.1126/science.1077991
2. De Tata V. **Age-Related Impairment of Pancreatic Beta-Cell Function: Pathophysiological and Cellular Mechanisms**. *Front Endocrinol* (2014.0) **5**. DOI: 10.3389/fendo.2014.00138
3. Atzmon G, Rincon M, Schechter CB, Shuldiner AR, Lipton RB, Bergman A. **Lipoprotein Genotype and Conserved Pathway for Exceptional Longevity in Humans**. *PloS Biol* (2006.0) **4** e113. DOI: 10.1371/journal.pbio.0040113
4. Baranowska B, Bik W, Baranowska-Bik A, Wolinska-Witort E, Szybinska A, Martynska L. **Neuroendocrine Control of Metabolic Homeostasis in Polish Centenarians**. *J Physiol Pharmacol: Off J Polish Physiol Soc* (2006.0) **57 Suppl 6** 55-61
5. Bonafe M, Olivieri F. **Genetic Polymorphism in Long-Lived People: Cues for the Presence of an Insulin/IGF-Pathway-Dependent Network Affecting Human Longevity**. *Mol Cell Endocrinol* (2009.0) **299**. DOI: 10.1016/j.mce.2008.10.038
6. Kojima H, Fujimiya M, Matsumura K, Nakahara T, Hara M, Chan L. **Extrapancreatic Insulin-Producing Cells in Multiple Organs in Diabetes**. *Proc Natl Acad Sci USA* (2004.0) **101**. DOI: 10.1073/pnas.0308690100
7. Rozing MP, Mooijaart SP, Beekman M, Wijsman CA, Maier AB, Bartke A. **C-Reactive Protein and Glucose Regulation in Familial Longevity**. *Age (Dordrecht Netherlands)* (2011.0) **33**. DOI: 10.1007/s11357-011-9206-8
8. Wijsman CA, Rozing MP, Streefland TC, le Cessie S, Mooijaart SP, Slagboom PE. **Familial Longevity Is Marked by Enhanced Insulin Sensitivity**. *Aging Cell* (2011.0) **10**. DOI: 10.1111/j.1474-9726.2010.00650.x
9. Paolisso G, Gambardella A, Ammendola S, D’Amore A, Balbi V, Varricchio M. **Glucose Tolerance and Insulin Action in Healthy Centenarians**. *Am J Physiol* (1996.0) **270**. DOI: 10.1152/ajpendo.1996.270.5.E890
10. Atzmon G, Schechter C, Greiner W, Davidson D, Rennert G, Barzilai N. **Clinical Phenotype of Families With Longevity**. *J Am Geriatr Soc* (2004.0) **52**. DOI: 10.1111/j.1532-5415.2004.52068.x
11. Westendorp RG, van Heemst D, Rozing MP, Frölich M, Mooijaart SP, Blauw GJ. **Nonagenarian Siblings and Their Offspring Display Lower Risk of Mortality and Morbidity Than Sporadic Nonagenarians: The Leiden Longevity Study**. *J Am Geriatr Soc* (2009.0) **57**. DOI: 10.1111/j.1532-5415.2009.02381.x
12. Adams ER, Nolan VG, Andersen SL, Perls TT, Terry DF. **Centenarian Offspring: Start Healthier and Stay Healthier**. *J Am Geriatr Soc* (2008.0) **56**. DOI: 10.1111/j.1532-5415.2008.01949.x
13. Yerges-Armstrong LM, Chai S, O’Connell JR, Curran JE, Blangero J, Mitchell BD. **Gene Expression Differences Between Offspring of Long-Lived Individuals and Controls in Candidate Longevity Regions: Evidence for PAPSS2 as a Longevity Gene**. *J Gerontol A Biol Sci Med Sci* (2016.0) **71**. DOI: 10.1093/gerona/glv212
14. Barral S, Cosentino S, Costa R, Matteini A, Christensen K, Andersen SL. **Cognitive Function in Families With Exceptional Survival**. *Neurobiol Aging* (2012.0) **33**. DOI: 10.1016/j.neurobiolaging.2011.02.004
15. Pedersen JK, Elo IT, Schupf N, Perls TT, Stallard E, Yashin AI. **The Survival of Spouses Marrying Into Longevity-Enriched Families**. *J Gerontol Ser A Biol Sci Med Sci* (2017.0) **72**. DOI: 10.1093/gerona/glw159
16. Newman AB, Glynn NW, Taylor CA, Sebastiani P, Perls TT, Mayeux R. **Health and Function of Participants in the Long Life Family Study: A Comparison With Other Cohorts**. *Aging (Albany NY)* (2011.0) **3** 63-76. DOI: 10.18632/aging.100242
17. Pedersen CB, Gøtzsche H, Møller JO, Mortensen PB. **The Danish Civil Registration System. A Cohort of Eight Million Persons**. *Dan Med Bull* (2006.0) **53**
18. Sebastiani P, Hadley EC, Province M, Christensen K, Rossi W, Perls TT. **A Family Longevity Selection Score: Ranking Sibships by Their Longevity, Size, and Availability for Study**. *Am J Epidemiol* (2009.0) **170**. DOI: 10.1093/aje/kwp309
19. Christensen K, Wojczynski MK, Pedersen JK, Larsen LA, Kløjgaard S, Skytthe A. **Mechanisms Underlying Familial Aggregation of Exceptional Health and Survival: A Three-Generation Cohort Study**. *Aging Cell* (2020.0) **19** e13228. DOI: 10.1111/acel.13228
20. Sebastiani P, Sun F, Andersen S, Lee J, Wojczynski M, Sanders J. **Families Enriched for Exceptional Longevity Also Have Increased Health-Span: Findings From the Long Life Family Study**. *Front Public Health* (2013.0) **1**. DOI: 10.3389/fpubh.2013.00038
21. **Centers for Disease Control and Prevention. National Diabetes Statistics Report website**
22. Raciti GA, Longo M, Parrillo L, Ciccarelli M, Mirra P, Ungaro P. **Understanding Type 2 Diabetes: From Genetics to Epigenetics**. *Acta Diabetol* (2015.0) **52**. DOI: 10.1007/s00592-015-0741-0
23. Nakosteen RA, Westerlund O, Zimmer MA. **Health-Related Disabilities and Matching of Spouses: Analysis of Swedish Population Data**. *J Popul Eco* (2005.0) **18** 491-507. DOI: 10.1007/s00148-005-0231-8
24. Venters MH, Jacobs DR, Luepker RV, Maiman LA, Gillum RF. **Spouse Concordance of Smoking Patterns: The Minnesota Heart Survey**. *Am J Epidemiol* (1984.0) **120**. DOI: 10.1093/oxfordjournals.aje.a113922
25. Macken LC, Yates B, Blancher S. **Concordance of Risk Factors in Female Spouses of Male Patients With Coronary Heart Disease**. *J Cardiopulm Rehabil* (2000.0) **20**. DOI: 10.1097/00008483-200011000-00005
26. Shiffman D, Louie JZ, Devlin JJ, Rowland CM, Mora S. **Concordance of Cardiovascular Risk Factors and Behaviors in a Multiethnic US Nationwide Cohort of Married Couples and Domestic Partners**. *JAMA Netw Open* (2020.0) **3** e2022119. DOI: 10.1001/jamanetworkopen.2020.22119
27. Houts RM, Robins E, Huston TL. **Compatibility and the Development of Premarital Relationships**. *J Marriage Family* (1996.0) **58** 7-20. DOI: 10.2307/353373
28. Carlsson S, Hammar N, Grill V, Kaprio J. **Alcohol Consumption and the Incidence of Type 2 Diabetes. A 20-Year Follow-Up of the Finnish Twin Cohort Study**. *Diabetes Care* (2003.0) **26**. DOI: 10.2337/diacare.26.10.2785
29. Wannamethee SG, Shaper AG, Perry IJ, Alberti KGMM. **Alcohol Consumption and the Incidence of Type II Diabetes**. *J Epidemiol Commun Health* (2002.0) **56**. DOI: 10.1136/jech.56.7.542
30. Hu FB, Sigal RJ, Rich-Edwards JW, Colditz GA, Solomon CG, Willett WC. **Walking Compared With Vigorous Physical Activity and Risk of Type 2 Diabetes in WomenA Prospective Study**. *JAMA* (1999.0) **282**. DOI: 10.1001/jama.282.15.1433
31. Govindaraju DR. **Evolutionary Genetic Bases of Longevity and Senescence**. *Adv Exp Med Biol* (2015.0) **847** 1-44. DOI: 10.1007/978-1-4939-2404-2_1
32. Johnson AA, Stolzing A. **The Role of Lipid Metabolism in Aging, Lifespan Regulation, and Age-Related Disease**. *Aging Cell* (2019.0) **18**. DOI: 10.1111/acel.13048
33. Rasouli N, Kern PA. **Adipocytokines and the Metabolic Complications of Obesity**. *J Clin Endocrinol Metab* (2008.0) **93**. DOI: 10.1210/jc.2008-1613
34. Blüher M. **Clinical Relevance of Adipokines**. *Diabetes Metab J* (2012.0) **36**. DOI: 10.4093/dmj.2012.36.5.317
35. Tilg H, Moschen AR. **Adipocytokines: Mediators Linking Adipose Tissue, Inflammation and Immunity**. *Nat Rev Immunol* (2006.0) **6**. DOI: 10.1038/nri1937
36. Unger RH. **Longevity, Lipotoxicity and Leptin: The Adipocyte Defense Against Feasting and Famine**. *Biochimie* (2005.0) **87** 57-64. DOI: 10.1016/j.biochi.2004.11.014
37. Blüher M, Kahn BB, Kahn CR. **Extended Longevity in Mice Lacking the Insulin Receptor in Adipose Tissue**. *Science* (2003.0) **299**. DOI: 10.1126/science.1078223
38. Bowker N, Shah RL, Sharp SJ, Luan J, Stewart ID, Wheeler E. **Meta-Analysis Investigating the Role of Interleukin-6 Mediated Inflammation in Type 2 Diabetes**. *EBioMedicine* (2020.0) **61**. DOI: 10.1016/j.ebiom.2020.103062
39. Puzianowska-Kuźnicka M, Owczarz M, Wieczorowska-Tobis K, Nadrowski P, Chudek J, Slusarczyk P. **Interleukin-6 and C-Reactive Protein, Successful Aging, and Mortality: The PolSenior Study**. *Immun Ageing* (2016.0) **13**. DOI: 10.1186/s12979-016-0076-x
40. Sandhu MS. **Insulin-Like Growth Factor-I and Risk of Type 2 Diabetes and Coronary Heart Disease: Molecular Epidemiology**. *Endocr Dev* (2005.0) **9** 44-54. DOI: 10.1159/000085755
41. Drogan D, Schulze MB, Boeing H, Pischon T. **Insulin-Like Growth Factor 1 and Insulin-Like Growth Factor-Binding Protein 3 in Relation to the Risk of Type 2 Diabetes Mellitus: Results From the EPIC-Potsdam Study**. *Am J Epidemiol* (2016.0) **183**. DOI: 10.1093/aje/kwv188
42. Rajpathak SN, He M, Sun Q, Kaplan RC, Muzumdar R, Rohan TE. **Insulin-Like Growth Factor Axis and Risk of Type 2 Diabetes in Women**. *Diabetes* (2012.0) **61**. DOI: 10.2337/db11-1488
43. Petersson U, Ostgren CJ, Brudin L, Brismar K, Nilsson PM. **Low Levels of Insulin-Like Growth-Factor-Binding Protein-1 (IGFBP-1) Are Prospectively Associated With the Incidence of Type 2 Diabetes and Impaired Glucose Tolerance (IGT): The Söderåkra Cardiovascular Risk Factor Study**. *Diabetes Metab* (2009.0) **35** 198-205. DOI: 10.1016/j.diabet.2008.11.003
44. Sandhu MS, Heald AH, Gibson JM, Cruickshank JK, Dunger DB, Wareham NJ. **Circulating Concentrations of Insulin-Like Growth Factor-I and Development of Glucose Intolerance: A Prospective Observational Study**. *Lancet* (2002.0) **359**. DOI: 10.1016/s0140-6736(02)08655-5
45. Reddy SSK, Chaiban JT. **The Endocrinology Of Aging: A Key To Longevity “Great Expectations”**. *Endocr Pract* (2017.0) **23**. DOI: 10.4158/ep171793.Ra
46. Ding EL, Song Y, Manson JE, Hunter DJ, Lee CC, Rifai N. **Sex Hormone-Binding Globulin and Risk of Type 2 Diabetes in Women and Men**. *N Engl J Med* (2009.0) **361**. DOI: 10.1056/NEJMoa0804381
47. Hyde Z, Norman PE, Flicker L, Hankey GJ, Almeida OP, McCaul KA. **Low Free Testosterone Predicts Mortality From Cardiovascular Disease But Not Other Causes: The Health in Men Study**. *J Clin Endocrinol Metab* (2012.0) **97**. DOI: 10.1210/jc.2011-1617
48. Corona G, Rastrelli G, Monami M, Guay A, Buvat J, Sforza A. **Hypogonadism as a Risk Factor for Cardiovascular Mortality in Men: A Meta-Analytic Study**. *Eur J Endocrinol* (2011.0) **165** 687-701. DOI: 10.1530/eje-11-0447
49. Schederecker F, Cecil A, Prehn C, Nano J, Koenig W, Adamski J. **Sex Hormone-Binding Globulin, Androgens and Mortality: The KORA-F4 Cohort Study**. *Endocr Connect* (2020.0) **9**. DOI: 10.1530/EC-20-0080
|
---
title: Empowered by Intertwined Theory and Practice – Experiences From a Diabetes
Sports Camp for Physically Active Adults With Type 1 Diabetes
authors:
- Stig Mattsson
- Peter Adolfsson
- Johan Jendle
- Viktor Bengtsson
- Carina Sparud-Lundin
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2021
pmcid: PMC10012152
doi: 10.3389/fcdhc.2021.655238
license: CC BY 4.0
---
# Empowered by Intertwined Theory and Practice – Experiences From a Diabetes Sports Camp for Physically Active Adults With Type 1 Diabetes
## Abstract
### Aims
To describe the experiences of individuals with diabetes type 1 (T1D) participating in diabetes sports camps and how acquired knowledge could be used in daily self-management.
### Methods
Semi-structured telephone interviews were conducted with 15 adults with T1D. A strategic sample procedure was chosen. The interviews were analyzed using qualitative content analysis.
### Results
The overarching theme ”Empowered by intertwined theory and practice”, included three main categories: Learning in a motivation-enhancing environment, incorporation of new habits and perceptions of glycemic control and health-related outcomes. The participants considered the camp to be an excellent opportunity to share feelings, ideas, and knowledge. They felt empowered by the camp atmosphere as well as supportive environment. After the camp, the general well-being was improved by incorporating new habits and improvements in glucose control.
### Conclusions
A diabetes sports camp constitutes an excellent, but resource-intensive, complimentary support in diabetes care and provides opportunities for T1D individuals to become more independent and autonomous. The findings indicate the need for more directed learning activities for individuals with type 1 diabetes and health care providers to increase their competence in the area of T1D and exercise in order to adequately manage counseling in various types of sports.
## Introduction
Well-controlled diabetes can delay or prevent the development of diabetes complications [1]. The effort involved in achieving good glycemic control places high demands on most people with type 1 diabetes (T1D). Physical exercise (PE) is of fundamental importance for both health and well-being [2] and is recommended (along with diet and insulin) as one of the cornerstones in the treatment of T1D [3]. However, PE is associated with an increased incidence of both hypo- and hyperglycemia [4]. In T1D, hypoglycemia is considered to be a major barrier to PE and the main reason for deteriorated glucose control [5]. In daily practice for people with diabetes severe fluctuations in glucose on days with PE is a common phenomenon. The increased glucose variability is often associated with frustration, as satisfactory glycemic control becomes difficult to achieve.
Achieving stable glucose regulation in relation to PE of different duration and intensity puts high demands on appropriate insulin adjustments and adapted carbohydrate intake [6]. For physically active T1D individuals, a high degree of educational support is therefore required [7]. Education and increased self-perceived knowledge in the area of T1D have been shown to have a major impact on an individual’s ability to manage their diabetes self-care [7]. Adolfsson et al. showed that education and feedback related to PE resulted in improved glucose control and increased levels of physical activity [8]. It has been observed that individuals with T1D who increase their knowledge of nutrition and insulin adjustments to minimize PE-induced hypoglycemia also reduce their barrier to physical activity [5]. However, most patients with poor glycemic control have already received diabetes self-management education (glucose testing, bolus calculation, insulin adjustments in relation to PE etc.) but failed to implement this into their daily regimen [7]. There can be many different reasons for not implementing learned self-management routines: difficulties in achieving glycemic goals, fear of hypoglycemia, feelings of failure or lack of support [9]. A fundamental problem regarding adequate counseling on sport and T1D is that health care professionals (HCPs) often lack sufficient knowledge [10]. The American Diabetes Association (ADA) suggests that patients with T1D may benefit from working with a physiologist experienced in diabetes and exercise to adequately assist them in formulating a safe and effective exercise prescription [3].
Diabetes camps with multidisciplinary participation, including physicians, nurses and dietitians, have been proven to contribute to learning principles of self-management positively [11]. Diabetes sports camps can thus enable theoretical and practical learning of principles related to the specifics of PE. Moreover, such camps offer an excellent opportunity to share and discuss experiences. Increased understanding of the participants’ experiences of structured education and support in T1D and PE can provide enhanced knowledge, which may lead to more targeted and tailored healthcare efforts for physically active patients with T1D who regularly perform PE. Our research group arranged eight sports camps for individuals with T1D, resulting in significant improvements of self-estimated knowledge in the area of insulin adjustments and carbohydrate intake, measured directly after the camp. Furthermore, improved glucose control was confirmed, measured as glycated hemoglobin (HbA1c), 3 and 12 months after the sports camp [12]. This study aimed to further explore the mechanisms behind these results by describing the individual experiences gained during a sports camp, and how acquired knowledge could be applied in daily self-management.
## Study Design
The study applies a qualitative design with an inductive approach whereby people’s perceptions and experiences are interpreted in the most unconditional way. The study is part of a larger project carried out by our research team, individuals with T1D participated in a three-day diabetes sports camp, including theoretical and practical education [12]. Informed consent was collected before the study, which was approved by the regional ethical review board in Uppsala, Sweden (DNR: $\frac{2012}{159}$).
## The Context of the Diabetes Sports Camps
In all, eight diabetes sports camps with a total of 105 participants were held. The participants were between 16 and 70 years of age, and all performed PE ≥3 sessions per week. During the camp, all participants received general education and personalized feedback by two physicians and a dietician. The training contained information on insulin adjustments and carbohydrate intake before, during and after PE. The participants practiced carbohydrate counting for all meals. Various types of sports were done one to two times daily. After each training session, the participants received individualized feedback based on downloaded glucose data from real-time Continuous Glucose Monitoring (CGM) devices and insulin doses from Continuous Subcutaneous Insulin Infusion (CSII) pumps or records of doses administered by Multiple Daily Injections (MDI) devices. Assessment of self-perceived knowledge regarding insulin and carbohydrate adjustments related to exercise was performed before and immediately after the camp.
## Sample and Recruitment for Interviews
Interviews were conducted 2–2.5 years after completion of the sports camps. The participants were interviewed about their experiences. A strategic sample procedure was chosen to obtain participants who provided a range of perceptions and experiences. Variation was sought regarding gender, age and sporting ambition (recreational/elite). It was estimated that between 15 and 20 people needed to participate. Of the 19 people who were asked to participate in the interview, 15 agreed to do so. Interviews were conducted by telephone.
## Data Collection and Analysis
The principal interviewer, a nurse with experience of T1D, did not attend any of the diabetes sports camps. Semi-structured telephone interviews were conducted, starting with open-ended questions. A pilot interview was undertaken to test and develop the interview guide, which focused on the participants’ experiences and whether increased knowledge affected their diabetes self-management in everyday life. Some questions concerned the general perception of having T1D adjacent to PE. The interviews were recorded digitally and then transcribed verbatim. According to Graneheim and Lundman, a descriptive qualitative content analysis was used to analyze data [13]. The interview text was read repeatedly to get a sense of the overall content. Meaning units were selected based on the study aim and after that, condensed further while keeping the essence intact. The condensed meaning units were sorted into codes that were still very close to data and limited interpretation of content. The next step was to compare codes and cluster them into categories and main categories, based on similar content and meaning. In this phase, a more interpretative process of abstracting data to a higher level was applied, especially in the following step when formulating overall theme (see an example of the analytic process in Table 1). Three authors selected meaning units (SM, VB, CSL) and two authors independently coded the data (SM, CSL). Disagreements in coding were resolved through a consensus approach among all authors until joint decisions were reached. Emerging categories and the overall theme were discussed among all the authors for consistency and to secure thrustworthiness.
**Table 1**
| Meaning unit | Code | Category | Main Category | Theme |
| --- | --- | --- | --- | --- |
| The sports camp gave me incredible joy and relief, to meet other people who are in exactly the same situation as me. I am not ashamed that I have diabetes, I have no problem that it beeps (CGM) (p5) | To meet people who are in the same situation | Strengthened by shared experiences | Learning in a motivation-enhancing environment | Empowered by intertwined theory and practice |
| The camp included education in carbohydrate counting, for me carbohydrate counting works great! I have been asking for a tool for a long time to help me make decisions (insulin dosage), here and now (p14) | Carbohydrate counting is a tool that helps me to decide an appropriate insulin dose for the meal | Carbohydrate counting | Incorporating of new habits | Empowered by intertwined theory and practice |
| When you notice that you are good (blood glucose), then you are super … you can do so much more in the workout, so there really is a difference, you dare to give it all (p3) | With good glucose control during exercise you dare to give it your all | Physical performance | Perception of glycemic control and health related outcomes | Empowered by intertwined theory and practice |
## Results
Fifteen individuals were included, of which nine were females and six males. The median age was 42.5 years (range: 28–63). The mean diabetes duration was 22.1 years (range: 0.5–52). The main characteristics of the participants are described in Table 2.
**Table 2**
| Number, n | 15 |
| --- | --- |
| Gender (female/male), n | 9/6 |
| Age (years), mean ± SD (range) | 41.9 ± 9.2 (28–63) |
| BMI1 (kg/m2) | 24.5 ± 2.1 (20.4–28.3) |
| Diabetes duration (years), mean ± SD (range) | 22.1 ± 14.9 (0.5–52) |
| Treatment regimen (CSII2/MDI3), n | 10/5 |
| Total daily insulin dose (IU/kg), mean (range) | 0.54 ± 0.1 (0.34–0.84) |
| HbA1c, pre-camp (NGSP4, %), mean (range) | 7.7 ± 0.9 (6.0–9.3) |
| HbA1c, pre-camp (IFCC5, mmol/mol), mean (range) | 60.3 ± 9.5 (42–78) |
| HbA1c, 12 mo. post-camp (NGSP4, %), mean (range) | 7.2 ± 0.8 (5.9–8.8) |
| HbA1c, 12 mo. post-camp (IFCC5, mmol/mol), mean (range) | 55.6 ± 8.6 (41–73) |
The analysis of the interviews resulted in an overarching theme: Empowered by intertwined theory and practice. The theme included three main categories: Learning in a motivation-enhancing environment, Incorporation of new habits and Perceptions of glycemic control and health-related outcomes. The three main categories further consisted of a total of ten categories (see Figure 1). For those interviewed, a recurring wish was to live a physically active life with good glucose control during exercise, thus enabling them to live a full life together with diabetes. For these participants, mastering their ordinary diabetes self-management in everyday life was not enough. The participants considered the sports camp to be an excellent opportunity to share feelings, ideas, and knowledge and cultivate a sense of belonging with other people with T1D, which was inspiring and motivating. They felt empowered by the camp atmosphere and the supportive environment generated by both participants and experienced HCPs, creating unique conditions to put theory into practice in a safe environment. After the camp, the participants’ general well-being was improved due to incorporating new habits in daily life and physical activity.
**Figure 1:** *Theme, main categories and categories of the participants’ experiences gained during a sports camp.*
Moreover, improvements in glucose control had given them more “energy” during the training sessions and in everyday life. Previously, with the experience of uncontrolled glucose control, they often felt cranky and angry, which also harmed their social life. In the following section, the three main categories are described, followed by exemplifying quotes from participants.
## Strengthened by Shared Experiences
The participants’ shared experience of attending the sports camp gave them a sense of freedom, as they felt there was nothing they had to hide. It did not matter if there was noise from the CSII or CGM-equipment, and it was no big deal if one of the participants reached hypoglycemic glucose levels, as everyone had been through the same thing before, many times. They described pleasure in being able to support others but also in receiving help from other participants experiencing the same problems and issues. They felt a sense of community and that they were like ordinary people and not different or alone. The camp generated the community spirit not just because the participants all had diabetes, but also because they loved PE and different kinds of sports and were eager to accept the challenges of living with T1D.
The sports camp gave this community, that you have made friends you can contact and ask, how do you solve this, and how do you do this? ( p5) One of the absolute best things about the sports camp has been getting a feeling of normality, that you’re not so weird, and that you’re not alone. ( p6)
## Reflective Approach Promoting Self-Efficacy
Being able to reflect on issues arising when performing physical activities with others in the same situation and accessing knowledgeable HCPs in the area of T1D and PE were described as motivating the participants to adopt a can-do attitude. Theoretical education and assisted analyses of how to make appropriate adjustments of insulin doses and carbohydrate intake before, during and after PE were perceived as new and valuable empowering support. The participants noticed that the methods they learned during the camp worked, which was an essential factor in achieving lasting behavioral change.
I might have hesitated more, but now I think—no, I actually can!—I will anyway! And I really took home that feeling from the camps. It’s not a question of whether I should do it, it’s just a question of how I should do it. It was a strong feeling that you got! ( p6) Eh, and it’s great to get a boost, and you get it at a camp like this, to be reminded—yeah sure! Here’s how to do it … now it’s important to give it your all! ( p10)
## Applying Knowledge-Based Tools
The participants described receiving a great deal of solid advice during the camp, such as how to adjust insulin doses and how to consume appropriate amounts of carbohydrates in conjunction with PE. They appreciated the constant feedback on how their actions affected glucose regulation during exercise. This feedback was something that the participants found very helpful in fine-tuning their diabetes self-management, both regarding PE and in everyday life after the sports camp.
To keep this preparatory protocol, and see for a week—what you eat—how much carbohydrate, how much insulin you take and what blood glucose values you have obtained after that (analysis of downloaded glucose data) (p9) Eating the right amount of carbohydrates at the right time, adjusting the insulin doses appropriately, and not like I did before. ( p14)
## Increased Awareness of Insufficient Support in Diabetes Care
The sports camp increased the participants’ understanding of the strengths and weaknesses of their regular diabetes health care. Most of the participants had experienced that HCPs were not knowledgeable enough in T1D and sport. *In* general, when asked questions, the caregivers at their diabetes centers had told them to try what suited them best in different situations. HCPs often claimed that, since all individuals are different, it is difficult to give concrete advice. Thus, the participants described often having to learn things through a trial and error strategy. After receiving experiences at the camp, many participants stated that they now expected more of their HCPs.
There’s a significant lack of knowledge in healthcare—specifically about sports, and I was advised by my doctor, some years ago, to quit sports. So, I haven’t really received any support, either from a diabetes nurse or a doctor. They have no knowledge about sports! I think it’s because they don’t practise sports themselves so much. We haven’t been able to communicate well. ( p11)
## Carbohydrate Counting and Consumption of an Appropriate Amount of Carbohydrates
Before the sports camp, most participants did not count carbs, but the majority continued with carb counting at home. They said it could be challenging at first, but over time it got easier. Carb counting was considered to make everyday life easier. It helped them choose a proper dose of insulin for the meal and enabled them to compose a meal that provided a sufficient amount of carbohydrates. After starting carb counting, some experienced improved glucose control in everyday life.
Yes, I count carbs, I didn’t before. So, I learned that at camp, a good way. I weigh pasta and rice after camp. There’s actually a difference. ( p9) Many participants claimed they did not know enough about the appropriate amount of carbohydrates they needed to consume before, during and after exercise. The participants described how the sports camp had increased their knowledge of how the body works and why a certain amount of carbohydrates may be needed in different situations to maintain stable glucose regulation. The participants also described how they had changed their eating habits in connection with exercise after the camp and how they had started to think more about their diet.
How to think in connection with exercise was very positive. How to think before, during and after training. I also got to learn about the carbohydrate-mind-set, when you train, to consume carbohydrates, it was a fantastic discovery. ( p3) On the days when you exercise, you should eat more, and vice versa. That, along with carbohydrate counting, changed my whole way of managing my diet. I have an incredible improvement in my diabetes control. ( p14)
## Insulin Adjustments Pre-, During- and Post-Exercise
The participants changed the way they adjusted their insulin doses in connection with exercise and claimed that it had reduced incidences of hypoglycemia after the camp. They reduced their basal insulin before exercise to reduce the risk of hypoglycemia during the workout. Likewise, they understood that they needed less insulin for a given amount of carbohydrates in the meal after exercise due to increased insulin sensitivity.
We received concrete advice on how we can adjust the insulin dose and to what percentage I have to reduce the insulin dose—and how long before. ( p1) *But this* thing, about being able to reduce the basal insulin and consuming carbohydrates, how I should handle everything and think. Yes, I have learned that! ( p12)
## Exploring Diabetes Technologies
Consequently, letting all participants use CGM during the sports camp was that most of them became more interested in using technical aids and started to use CGM in everyday life at home. It was not only the opportunity to see their glucose values in real-time that attracted them to using CGM but also that they could download data and subsequently analyze their glucose values. After the camp, some participants started setting aside one day a week to analyze their glucose values for the past week. The evaluation procedure helped them see what had gone well or less well during the previous week, enabling them to fine-tune their insulin dosage and dietary intake. Some participants who used insulin pens during the camp had switched to an insulin pump afterwards, convinced about its benefits by other insulin pump users during the camp.
The sports camp made me interested in technical aids. I had not received any information about insulin pumps before the camp. I had neither insulin pump nor CGM before the camp. After the camp I’ve been using both an insulin pump and CGM. ( p1) CGM has helped me a lot with basal doses. It’s helped me understand what happens during training and after training and so on. ( p11)
## Own and Others’ Well-Being
It was stated that knowledge and practice gained at the sports camp also improved diabetes self-management at home after the sports camp. Glucose control had been improved with PE but also in everyday life. Some stated that their quality of life had been affected positively; for example, improved glucose control and general well-being had made them feel more satisfied—something people in their social network had noticed and appreciated.
I feel that with this knowledge—it’s a different quality of life—absolutely! ( p11) It has positively affected them (how improved glucose control affected relatives). I have less mood swings, I am happier and more alert. Again, general health is much better. I’m alert, happy, things are better at work too. I no longer get the dips I had before, with low blood glucose when I’m doing sports. It’s now stable in another way. People in my environment have well received this as a result. My brother also has diabetes, so I gave him the information. He’s also starting to work it out and feel much better with his diabetes. So my parents and siblings and everyone are happy and my partner—they think it’s absolutely brilliant. ( p15)
## Sense of Safety
The camp leaders’ knowledge and experience in the field of T1D and sports generated trust and made the participants feel safe in the decisions made during camp activities. With knowledgeable HCPs on hand, the participants found carrying out all the different steps themselves a valuable exercise, as it meant they could also perform them at home. Now that they had learned and practiced all the necessary actions while at the camp, their self-confidence and sense of safety/security when performing PE once they got home increased.
After the camps, I have a completely different sense of security in practicing sports. ( p5) My wife probably feels a little safer as well, that I won’t get hypoglycemia and pass out somewhere. Especially if I’m out driving a car. ( p10)
## Physical Performance
The participants described physical performance during exercise as an important aspect: they wanted to feel good during the workout. They said that before coming to the camp, their bodies did not function as they should during exercise, with a loss of power if blood glucose was too low or high. Glucose excursions meant exercise was not a joyful experience and could negatively affect the rest of their day. After the camp, they said it was easier to achieve more stable glycemic control and that they had more power during their training workouts. The more stable glucose control had a positive effect on their well-being overall.
If you notice that you are at a good level (blood glucose), then you’re super, you have so much more energy during the workout, you can really give it your all. But as soon as your blood glucose goes down, you can do nothing. Actually it’s also very hard throughout the day then. ( p3) It’s more fun to exercise if your blood sugar is good, and that’s that! ( p4)
## Discussion
The current study is the first study to examine experiences from a diabetes sports camp for adults with T1D, where theoretical education was combined with practical exercise and the application of the acquired knowledge in daily self-management. The overall interpretation of the findings is that the intertwined theory and practice during the camps empowered the participants. All of them stated that the main strength was the camp as a whole. Firstly, both the participants and the experienced HCPs had a great interest in practicing different sports. Secondly, the participants and HCPs worked together as a team during the camp with both parties taking part in theoretical education and the practical execution of the exercise.
An important factor in obtaining lasting behavioral changes in an individual’s self-management routine is to do what is to be learned, i.e. hands-on skills training [14]. During the sports camp, the participants performed all adjustments with guidance from the HCPs. The participants’ autonomy, self-efficacy and self-management behavior were improved via theoretical education and practical execution as a learning principle. It has been shown that patients experience deficiencies due to the gap between theoretical diabetes education and practical implementation of the same [15]. The fear of PE-associated hypoglycemias may, for some, mean that they do not even dare to try new methods. The camp provided a safe and secure environment, which was an essential aspect in terms of giving the participants enough confidence to try new methods, such as increasing insulin doses before and during high-intensity anaerobic PE.
The participants said they appreciated the continuous feedback from the physicians, dietitians, diabetes nurses and other camp attendees during and after every workout. Any adjustments to insulin and diet were repeatedly evaluated and readjusted if necessary before the next training session. This continuous evaluation and feedback during and after PE was crucial for learning and would not have been possible during the camps without CGM. Dyck et al. [ 16] reported that of all the education tools (classroom teaching, real-time CGM, supervised exercise etc.) to learn glucose responses during and after exercise; it was real-time CGM that was considered the most valuable tool for understanding and improved knowledge. Our study confirms this. The participants stated that having CGM improved their knowledge of the importance of adjusting insulin and carbohydrate intake in connection with PE. The overall goal of the sports camps was for every participant to reach glycemic control during exercise. The fact that they were able to achieve stable glucose levels while performing PE was important in increasing self-efficacy and a lasting change in self-management behavior. It has been shown that repeatedly failing to achieve glucose target levels within target ranges is demoralizing and counteracts the participants’ initial feelings of motivation and empowerment [17].
Studies have shown that the participants themselves constitute an important factor for the development autonomy, self-efficacy and self-management in diabetes camps (18–20). The camp members can become advisors and role models, which in turn facilitates the individual’s acceptance of their condition and leads to improved engagement in self-management attitudes [19]. This is in line with the findings in our study, where all participants experienced that meeting other individuals with T1D was inspiring, motivating and rewarding. The participants stated that they were affected by the other participants’ attitudes, thoughts, values and reactions. They adopted more of a can-do-attitude, the question being not whether or not they should do something, but simply how they should do it.
In our study, all camp attendees participated in lectures and discussions, where the aim was to understand the physiology during and after exercise. They were also presented with a rationale for why insulin and/or carbohydrate intake should be adjusted in different situations to achieve stable glucose regulation. The goal was to enable them to make appropriate choices about insulin and carbohydrates in connection with different kinds of PE and thus help them achieve the desired result of proactive self-management at home, after the camp. In the interviews, the participants said they had incorporated many of these new habits in daily life, and the main reason they were maintaining their new diabetes self-management behavior was that they noticed it was working. It is known that for new self-management behaviors to be permanent, it is crucial for individuals that the result is in line with the intended outcome [17]. The participants reported that their glucose control had improved after the camp and that the incidence of hypoglycemia decreased in connection with both exercise and everyday life. They also said that glycemic control during their training sessions was often reflected in their mood even afterwards. If glucose control was poor during exercise, the rest of the day could feel worse as a result. Additionally, overall improved glucose control made them happier and more satisfied with life in general, which relatives and friends also confirmed.
In the current study, all HCPs had extensive experience in the field of T1D and sports. Previous studies have shown that, during a diabetes camp, a well-trained staff of HCPs is absolutely crucial to achieving an autonomy-supportive environment [18, 19]. Our results showed that the majority of the participants’ felt that their regular HCPs did not have sufficient knowledge within the field of T1D and sports to be able to provide them with guidance on appropriate insulin adjustments and carbohydrate intake associated with PE. These findings are in line with other studies that have examined the level of knowledge of health professionals in the field of exercise [10, 21, 22]. This is a major problem in regular diabetes care. In order for clinicians to give advice in the area of T1D and PE, they must have a certain understanding of the metabolic response that occurs during different types of PE and what requirements this places on both insulin adjustments and carbohydrate requirements. To arrange sports camps similar to the one in this study requires large resources, both in working hours and financially. However, the most important factor for the success in these sports camps was probably that all HCPs were well educated within the field of T1D and PE and the participants were eager to learn. This constellation of people can also be obtained in web-based education programs where experiences also can be exchanged between patients. Web-based education has been shown to improve knowledge, self-efficacy and self-management behaviors in health care (23–25). The most effective way to reach higher levels of knowledge about PE and T1D in both patients and HCPs may be to use web-based solutions in regular diabetes care [24, 26, 27].
Our results show that the sports camps are educational and rewarding for the patients, but could be considered costly. Another feasible solution, however, is to organize national diabetes sports camps annually where patients and HCPs from diabetes teams are invited. This approach would gradually increase the level of competence of both patients and HCPs.
## Limitations
A limitation of this study is that selection bias could exist because the study participants were not randomly selected from the TID population. Moreover, the participants themselves expressed their interest in participating in the camps. The inclusion criterion for participating at the camps was that the participants would exercise ≥3 workout sessions/week, which probably yielded a relatively homogeneous group of participants who were determined to improve glycemic control. Thus, the participants may not be a representative sample of adult subjects with T1D who regularly practice physical activity in general.
## Conclusion
The experiences from a dedicated diabetes sports camp focusing on achieving good glucose control during daily training sessions indicate that this kind of complimentary support in diabetes care can empower individuals to become more independent and autonomous and thus able to live a full life with diabetes and physical activity. The positive atmosphere that arose during the sports camps was primarily due to the fact that both HCPs and the patients had common interests—T1D and PE. These people could also meet and exchange experiences in other contexts. The post-pandemic digital development has provided new opportunities for web-based solutions. Thus, web-based education packages in the area of T1D and PE could be used to increase the level of knowledge for both patients and healthcare professional—physicians, dietitians and nurses in diabetes care.
## Data Availability Statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics Statement
The studies involving human participants were reviewed and approved by Regional ethical review board in Uppsala, Sweden (DNR: $\frac{2012}{159}$). The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
SM and CS-L conceived and designed the research. PA and JJ participated in the planning of the study. VB conducted all interviews. SM, CS-L, PA, and JJ conducted a descriptive qualitative content analysis of collected data. SM wrote the manuscript. CS-L, PA, JJ, and VB reviewed the manuscript. All authors contributed to the article and approved the submitted version.
## Funding
The study was funded by an unrestricted grant from Novo Nordisk AS, Bagsværd, Denmark. The funder was not involved in the study design, collection, analysis, interpretation of data, the writing of this article or the decision to submit it for publication.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Abbreviations
A1c, Glycated Hemoglobin; BMI, Body Mass Index; CGM, Continuous Glucose Monitoring; CSII, Continuous Subcutaneous Insulin Infusion; IFCC, International Federation of Clinical Chemistry; MDI, Multiple Daily Injections; NGSP, National Glycohemoglobin Standardization Program; PE, Physical Exercise; T1D, Type 1 Diabetes.
## References
1. Nathan DM, Genuth S, Lachin J, Cleary P, Crofford O, Davis M. **The Effect of Intensive Treatment of Diabetes on the Development and Progression of Long-Term Complications in Insulin-Dependent Diabetes Mellitus**. *N Engl J Med* (1993) **329**. DOI: 10.1056/NEJM199309303291401
2. Blair SN, Sallis RE, Hutber A, Archer E. **Exercise Therapy - the Public Health Message**. *Scand J Med Sci Sports Aug* (2012) **22**. DOI: 10.1111/j.1600-0838.2012.01462.x
3. Colberg SR, Sigal RJ, Yardley JE, Riddell MC, Dunstan DW, Dempsey PC. **Physical Activity/Exercise and Diabetes: A Position Statement of the American Diabetes Association**. *Diabetes Care* (2016) **39**. DOI: 10.2337/dc16-1728
4. Riddell MC, Perkins BA. **Type 1 Diabetes and Vigorous Exercise: Applications of Exercise Physiology to Patient Management**. *Can J Diabetes* (2006) **30** 63-71. DOI: 10.1016/S1499-2671(06)01010-0
5. Brazeau AS, Rabasa-Lhoret R, Strychar I, Mircescu H. **Barriers to Physical Activity Among Patients With Type 1 Diabetes**. *Diabetes Care* (2008) **31**. DOI: 10.2337/dc08-0720
6. Adolfsson P, Riddell MC, Taplin CE, Davis EA, Fournier PA, Annan F. **ISPAD Clinical Practice Consensus Guidelines 2018: Exercise in Children and Adolescents With Diabetes**. *Pediatr Diabetes* (2018) **19**. DOI: 10.1111/pedi.12755
7. Ozcan S, Amiel SA, Rogers H, Choudhary P, Cox A, de Zoysa N. **Poorer Glycaemic Control in Type 1 Diabetes is Associated With Reduced Self-Management and Poorer Perceived Health: A Cross-Sectional Study**. *Diabetes Res Clin Pract* (2014) **106** 35-41. DOI: 10.1016/j.diabres.2014.07.023
8. Adolfsson P, Strömgren A, Mattsson S, Chaplin J, Jendle J. **Education and Individualised Support Regarding Exercise and Diabetes Improves Glucose Control and Level of Physical Activity in Type 1 Diabetes Individuals**. *J Endocrine Diabetes Obes* (2015) **2**
9. Rankin D, Cooke DD, Heller S, Elliot J, Amiel S, Lawton J. **Experiences of Using Blood Glucose Targets When Following an Intensive Insulin Regimen: A Qualitative Longitudinal Investigation Involving Patients With Type1 Diabetes**. *Diabetic Med* (2012) **29**. DOI: 10.1111/j.1464-5491.2012.03670.x
10. Litchfield I, Andrews RC, Narendran P, Greenfield S. **Patient and Healthcare Professionals Perspectives on the Delivery of Exercise Education for Patients With Type 1 Diabetes**. *Front Endocrinology* (2019) **10** 76-89. DOI: 10.3389/fendo.2019.00076
11. Santiprabhob J, Likitmaskul S, Sriwijitkamol A, Peerapatdit T, Sawathiparnich P, Nitiyanant W. **Improved Glycemic Control Among Thai Children and Young Adults With Type 1 Diabetes Participating in the Diabetes Camp**. *J Med Assoc Thai* (2005) **88**
12. Mattsson S, Adolfsson P, Jendle J. **Diabetes Sports Camps for Individuals With Type 1 Diabetes Associated With Improved Glycemic Control and Self-Estimated Level of Knowledge**. *J Dia Res Ther* (2017) **3**. DOI: 10.16966/2380-5544.127
13. Graneheim UH, Lundman B. **Qualitative Content Analysis in Nursing Research: Concepts, Procedures and Measures to Achieve Trustworthiness**. *Nurse Educ Today* (2004) **24**. DOI: 10.1016/j.nedt.2003.10.001
14. Choi S, Song M, Chang SJ, Kim SA. **Strategies for Enhancing Information, Motivation, and Skills for Self-Management Behavior Changes: A Qualitative Study of Diabetes Care for Older Adults in Korea**. *Patient Prefer Adherence* (2014) **8)**. DOI: 10.2147/PPA.S58631
15. Wiley J, Westbrook M, Long J, Greenfield JR, Day RO, Braithwaite J. **Diabetes Education: The Experiences of Young Adults With Type 1 Diabetes**. *Diabetes Ther* (2014) **5** 299-321. DOI: 10.1007/s13300-014-0056-0
16. Dyck RA, Kleinman NJ, Funk DR, Yeung RO, Senior P, Yardley JE. **We Can Work (It) Out Together: Type 1 Diabetes Boot Camp for Adult Patients and Providers Improves Exercise Self-Efficacy**. *Can J Diabetes* (2018) **42**. DOI: 10.1016/j.jcjd.2018.02.006
17. Campbell F, Lawton J, Rankin D, Clowes M, Coates E, Heller S. **Follow-Up Support for Effective Type 1 Diabetes Self-Management (The FUSED Model): A Systematic Review and Meta-Ethnography of the Barriers, Facilitators and Recommendations for Sustaining Self-Management Skills After Attending a Structured Education Programme**. *BMC Health Serv Res* (2018) **18** 898. DOI: 10.1186/s12913-018-3655-z
18. Barone MT, Vivolo MA, Madden PB. **Are Diabetes Camps Effective**. *Diabetes Res. Clin. Practice* (2016) **114** 15-22. DOI: 10.1016/j.diabres.2016.01.013
19. Hill E, Sibthorp J. **Autonomy Support at Diabetes Camp: A Self Determination Theory Approach to Therapeutic Recreation**. *Ther Recreation J* (2006) **40**
20. Beesley T, Zaharieva DP, Fraser-Thomas JL, Riddell MC. **Exploring the Process of Psychosocial Development and Glycemic Control in Youth With Type 1 Diabetes Mellitus (T1DM) Attending a Unique Diabetes Sports Camp**. *Can J Diabetes* (2012) **36** S54. DOI: 10.1016/j.jcjd.2012.07.387
21. Weiler R, Chew S, Coombs N, Hamer M, Stamatakis E, Stamatakis E. **Physical Activity Education in the Undergraduate Curricula of All UK Medical Schools: Are Tomorrow's Doctors Equipped to Follow Clinical Guidelines**. *Br J Sports Med* (2012) **46**. DOI: 10.1136/bjsports-2012-091380
22. Thornton JS, Frémont P, Khan K, Poirier P, Fowles J, Wells GD. **Physical Activity Prescription: A Critical Opportunity to Address a Modifiable Risk Factor for the Prevention and Management of Chronic Disease: A Position Statement by the Canadian Academy of Sport and Exercise Medicine**. *Br J Sports Med* (2016) **50**. DOI: 10.1136/bjsports-2016-096291
23. Lorig K, Ritter Pl, Laurent DD, Plant K, Plant K, Green M, Jernigan VB. **Online Diabetes Self-Management Program: A Randomized Study**. *Diabetes Care* (2010) **33**. DOI: 10.2337/dc09-2153
24. Poduval SA-OX, Marston LA-O, Hamilton FA-O, Stevenson FA-O, Murray EA-O. **Feasibility, Acceptability, and Impact of a Web-Based Structured Education Program for Type 2 Diabetes: Real-World Study**. *JMIR Diabetes* (2020) **5**. DOI: 10.2196/15744
25. Pereira K, Phillips B, Johnson C, Vorderstrasse A. **Internet Delivered Diabetes Self-Management Education: A Review**. *Diabetes Technol Ther* (2015) **17** 55-63. DOI: 10.1089/dia.2014.0155
26. Wantland DJ, Portillo CJ, Holzemer WL, Slaughter R, McGhee EM. **The Effectiveness of Web-Based vs. Non-Web-Based Interventions: A Meta-Analysis of Behavioral Change Outcomes**. *J Med Internet Res* (2004) **6** e40. DOI: 10.2196/jmir.6.4.e40
27. Han J, Torok M, Gale N, Wong QJJ, Werner-Seidler A, Hetrick SE. **Use of Web Conferencing Technology for Conducting Online Focus Groups Among Young People With Lived Experience of Suicidal Thoughts: Mixed Methods Research**. *JMIR Ment Health* (2019) **6** e14191. DOI: 10.2196/14191
|
---
title: 'Intensification of Insulin Treatment With Insulin Degludec/Aspart in Type
2 Diabetic Patients: A 2-Year Real-World Experience'
authors:
- Hatice Oner
- Hatice Gizem Gunhan
- Dilek Gogas Yavuz
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2022
pmcid: PMC10012153
doi: 10.3389/fcdhc.2022.783277
license: CC BY 4.0
---
# Intensification of Insulin Treatment With Insulin Degludec/Aspart in Type 2 Diabetic Patients: A 2-Year Real-World Experience
## Abstract
### Aim
To evaluate the effects of insulin degludec/insulin aspart (IDegAsp) coformulation as an intensification of insulin treatment for glycemic control in patients with type 2 diabetes (T2D) in a long term real-world clinical setting.
### Materials and Methods
This retrospective non-interventional study, included 210 patients with T2D who to IDegAsp coformulation from prior insulin treatment in a tertiary endocrinology center between September 2017 and December 2019. The baseline data was taken as the index date and defined as the first IDegAsp prescription claim. Previous insulin treatment modalities, hemoglobin A1c (HbA1c), fasting plasma glucose (FPG), and body weight were recorded, respectively at the 3rd, 6th, 12th, and 24th months of the IDegAsp treatment.
### Results
Out of the total 210 patients, 166 patients under insulin treatment switched to twice-daily IDegAsp treatment, 35 patients switched to once daily IDegAsp and twice premeal short-acting insulin regimen as a modified basal-bolus (BB) treatment, and nine patients commenced with once-daily IDegAsp treatment. HbA1c decreased from $9.2\%$ ± $1.9\%$ to $8.2\%$ ± $1.6\%$ in 6 months, $8.2\%$ ± $1.7\%$ in the first year, and $8.1\%$ ± $1.6\%$ in the second year of the therapy ($p \leq 0.001$). FPG decreased from 209.0 ± 85.0 mg/dL to 147.0 ± 62.6 mg/dL in the second year ($p \leq 0.001$). The required total daily dose of insulin increased in the second year of IDegAsp treatment compared to baseline. However, there was a borderline significance increase in IDegAsp requirement for the whole group at the two-year follow-up ($$p \leq 0.05$$). Patients who were administered twice daily IDegAsp injections required more total insulin in the first and second years due to added premeal short-acting insulin injections ($p \leq 0.05$). The frequency of patients with HbA1c < $7\%$ was $31.8\%$ in first year and $35.8\%$ in second year under IDegAsp treatment. Insulin dose was de-escalated in $28.5\%$ of the patients under BB treatment, while $15\%$ under twice-daily IDegAsp required increased BB treatment.
### Conclusion
Intensification of insulin treatment with IDegAsp coformulation improved glycemic control in patients with T2D. The total daily insulin requirement increased but the IDegAsp requirement lightly increased at the two-year follow-up. Patients under BB treatment required de-escalation of insulin treatment.
## Introduction
Insulin degludec/insulin aspart (IDegAsp) is a fixed-ratio coformulation of insulin degludec (degludec, $70\%$) and insulin aspart (IAsp, $30\%$) in a single injection that provides both prandial and basal glycemic coverage (1–3); it is usually administered once- or twice-daily with the main meals. The basal component, Degludec,has a flat pharmacokinetic profile over 24 hours at a steady-state and provides a stable and long-lasting, glucose-lowering effect with less hypoglycemia than long-acting insulins (2–6).
IDegAsp coformulation is an insulin preparation whose dual action profile offers both prandial and basal glycemic coverage without the need for multiple injections (5, 7–9). Randomized clinical trials found that treatment intensification with IDegAsp BID reduces nocturnal severe hypoglycemic episodes while reaching the non-inferior limit mean reduction in HbA1c when compared to biphasic insulin Aspart 30 [8]. IDegAsp coformulation provides similar glycemic efficacy and less nocturnal hypoglycemia in type 2 diabetic patients when compared to those using premix, basal-bolus (BB), and basal + oral antidiabetic drug combinations (10–12). A Japanese study showed that switching from insulin to IDegAsp coformulation significiantly improved glycemic control and reduced hypreducedng the at a 12 month follow-up of patients with type 2 diabetes [12]. Premixed biphasic insulins can increase treatment compliance and achieve efficacy similar to that of basal-bolus therapy but with fewer injections [9, 11]. Access to medication is remains a concern for people with diabetes [11].
Although robust data from randomized controlled trials have been published in the past, data from observational and real-world settings are scarce. A small retrospective study from India showed that IDegAsp coformulation reduced the insulin dose requirement when compared to premixed insulin [13].
IDegAsp coformulation has been available in Turkey since 2017 and reimbursed for insulin treatment intensifications in type 1 and type 2 diabetic patients. This study aimed to evaluate the effects of Insulin Degludec/Insulin Aspart (IDegAsp) as an intensification treatment for glycemic control in patients with type 2 diabetes (T2D) in a real-world clinical setting.
## Patient Selection
This retrospective study included 210 patients T2D who underwent insulin treatment for at least one year and switched to IDegAsp coformulation as intensification treatment between September 2017 and December 2019 in the endocrinology outpatient clinic of Marmara University Pendik Research and Training Hospital.
Patients with the following conditions were excluded from the study: type 1 diabetes, active inflammatory and infectious disorders, active cancer treatment, end-stage renal and hepatic disease, female gestational diabetes, age under 18 years or over 85 years, insulin treatment for less than one year, acute hyperglycemia without insulin usage, and hospitalization.
The study protocol was approved by the local ethics committee of the Marmara University School of Medicine (09.2021.111). The study was performed according to the International Conference on Harmonization Guidelines for Good Clinical Practice and the Declaration of Helsinki.
## Clinical Evaluation
The following demographic, clinical, and laboratory data were recorded from the patients’ files: age; gender; duration of diabetes; duration of insulin treatment; daily insulin dose requirement; total, long, and short-acting insulin doses per day; daily IDegAsp dose requirement; injection frequencies, lipid parameters, oral antidiabetic medicine, and body weight.
All patients were undergoing analog insulin treatment, including Detemir, Glargine U100, Glargine U300 long-acting insulin, Lispro, Glulisine, and insulin Aspart as rapidly acting insulins.
Twice-daily IDegAsp regimen doses were calculated by converting between premix and basal insulin regiments, divided as two equal doses before morning and evening meals. IDegAsp OD/BID regimen was preferred for patients who reported nocturnal hypoglycemia and to reduce injection frequencies under basal-bolus treatment. IDegAsp dose calculations were performed for unit-to-unit basal and basal insulin requirements. Basal insulin regimens switched to IDegAsp OD if nocturnal hypoglycemia was a concern and if postprandial glucose control was needed.
Nocturnal hypoglycemia was defined as glycemia that occurred from midnight to 6 AM. Severe hypoglycemia was defined as an episode requiring another person to administer carbonhydrate, glucagon, or to take other corrective actions; plasma glucose concentrations may not be available.
The patients were organized into three categories according to the IDegAsp insulin regimen at the beginning of the study: once-daily IDegAsp (IDegAsp/OD) ($$n = 9$$, $4.3\%$), twice-daily IDegAsp (IDegAsp/DIB) ($$n = 166$$, $79\%$) and once-daily IDegAsp with two short-acting insulin injections (IDegAsp OD/BID) ($$n = 35$$, $16.7\%$).
In the two-year follow-up period, various changes in insulin treatment options categorized as de-escalation, intensification, and interchange to another type of insulin were recorded, at the two-year follow-up, and results were calculated by grouping patients in the relevant insulin regimen.
## Biochemical Parameters
Biochemical results in the 3rd, 6th, 12th, and 24th months were recorded from the patients’ files. Fasting plasma glucose levels (FPG) were measured using an enzymatic UV test (hexokinase method); total cholesterol, HDL, and triglycerides were also analyzed using an enzymatic color method, whereas HbA1c was analyzed with high-performance liquid chromatography in Premier Hb9210 (Trinity Biotech, USA).
## Statistical Analysis
All the statistical analyses were conducted using SPSS (Statistical Package for the Social Sciences) version 20.0. Descriptive data was stated as frequencies (%) for categorical data, means, and standard deviations (SD) for continuous data with a normal distribution. Mann-Whitney U test, Kruskal-Wallis, and ANOVA tests were used to compare groups. A chi-square test was used to compare categorical data. The results were evaluated at a $95\%$ confidence interval. The statistical significance level was accepted as $p \leq 0.05$, and the results were expressed as mean ± SD values.
## Results
Out of all the 210 T2D patients, the mean age was 59.6 ± 10.5 years, and 131 ($62.4\%$) patients were females. The mean duration of diabetes was 12.8 ± 7.1 years. The mean weight was 95.7 ± 19.6 kg, and HbA1c levels were $9.2\%$ ± $1.9\%$. The mean daily insulin requirement was 62.0 ± 35.0 U.
During the inclusion initially $45.5\%$ of the patients were receiving premixed insulin, $18.5\%$ basal insulin, $3.3\%$ basal plus, and $28\%$ BB insulin.
At the time of initiation of after beginning IDegAsp treatment, 132 patients were treated with metformin ($62.9\%$), 104 patients with DPP-4 inhibitors ($49\%$), 13 patients with SGLT-2 inhibitors ($6.2\%$), and 10 patients with sulfonylurea ($4.8\%$). Eight patients were using GLP-1 analogue ($3.8\%$), three patients were using alpha glucosidase inhibitors ($1.4\%$), and one patient was using thiazolidinedione ($0.5\%$). In the follow-up, metformin was initiated in 27 patients, SGLT-2 inhibitor in 15 patients, DPP-4 inhibitors in 31 patients, and thiazolidinedione in two patients. The drugs for patients using sulfonylurea and for patients using alpha glucosidase were discontinued. In addition, SGLT-2 inhibitor in five patients, the GLP-1 analogue in two patients, and the DPP-4 inhibitors in three patients were discontinued.
According to lack of follow-up or interchange to another insulin regimen, 132 patients in the first year and 81 patients in the second year were under IDegAsp treatment. Table 1 shows insulin requirement and glycemic parameters during the follow-up period. HbA1c levels decreased after switching to IDegAsp treatment in the third month of the therapy ($p \leq 0.001$) and remained at the same level for two years period for all the groups ($p \leq 0.001$). Glycaemic parameters and insulin requirements for the IDegAsp regimen treated patients are shown in Table 2. In patients who switched to IDegAsp OD, IDegAsp BID, IDegAsp OD/BID regimen’s HbA1c levels decreased at the 3rd, 6th, 12th, and 24th month of the therapy compared to before treatment levels.
During follow-up, there was a significant increase in the total daily insulin requirement significantly increased during follow-up only in the second year compared to baseline ($$p \leq 0.009$$). There was no statistically significant increase in the total daily insulin requirement of the patients who received IDegAsp as part of the BB treatment ($$p \leq 0.520$$). However, the IDegAsp requirement slightly increased during the two-year follow-up period in the whole group ($$p \leq 0.05$$). Patients who received twice-daily IDegAsp required 62.2 ± 32.3 U total daily insulin in the beginning, followed by 74.7 ± 37.6 U in the first year and 79.9 ± 42.6 U at the and of the second year ($p \leq 0.05$). The dosage in patients who received twice-daily IDegAsp was 58.3 ± 27.3 U at the baseline and 71.2 ± 32.8 at the end of the second year ($p \leq 0.05$) (Table 2). Similar glycemic control was not observed in the patients who received two doses as a part of basal-bolus therapy. Although HbA1c significantly decreased in patients using two doses of insulin ($p \leq 0.01$), the same decrease did not reach statistical significance in patients who participated in BB therapy ($$p \leq 0.236$$).
Table 3 shows the insulin regimen and frequency in patients T2D under IDegAsp treatment at follow-up. Eighty one patients completed the two-year follow-up. Thirty-three patients switched to other insulin treatment regimens. Twenty-five patients in the first year and 39 patients in the second year lost of follow-up. Two of the patients ($22.2\%$) using IDegAsp/OD were converted to twice-daily IDegAsp, and six patients ($3.6\%$) using twice-daily IDegAsp were converted to IdegAsp OD/ BID treatment. Twenty-five of the patients who received twice daily IDegAsp switched to other insulin regimens, and ten of the patients who received IDegAsp as of BB insulin therapy were de-escalated.
**Table 3**
| Unnamed: 0 | Once-daily | Twice daily | Part of basal-bolus | Total |
| --- | --- | --- | --- | --- |
| Initiation (n) | 9 | 166 | 35 | 210 |
| Intensification (n) | 2 | 6 | – | 8 |
| De-escalation (n) | – | 3 | 10 | 13 |
| Interchange (n) | 3 | 25 | 5 | 33 |
Patients with HbA1c less than $7\%$ were, $15\%$ at the third month, $24.6\%$ at the sixth month,$31.8\%$ in the first year, and $35.8\%$ in the second year. The weight of the patients did not significantly differ during the 2-year follow-up period ($$p \leq 0.178$$).
## Discussion
This retrospective, non-interventional study, monitored patients with T2D who switched from other insulin regimens to IDegAsp as part of routine clinical practice. Shifting to IDegAsp treatment for two years maintained glycemic control, increased total daily insulin but stable IDegAsp insulin dose requirements in comparison with the baseline data. IDegAsp treatment also reduced FPG an HbA1c values, which were consistent with the individual trials [14, 15].
Differences in glycemic control were statistically significant after switching from other insulin regimens to IDegAsp. A Japanese study also found that FPG also improved with IDegAsp versus BIAsp 30 in two treat-to-target RCTs of insulin-experienced patients with T2D [16, 17].
The mean baseline HbA1c was $9.2\%$, indicating that many patients in this study did not achieve optimal glycemic control with previous insulin regimens. This likely motivated their switch to the IDegAsp regimen. The decrease in HbA1c at 24 months ($1.1\%$) observed here is similar to the targeted HbA1c reduction after 6 months of IDegAsp treatment in Japanese patients with T2D whose insulin levels were insufficiently controlled by previous therapies ($1.4\%$) [16]. RCT baseline HbA1c ($8.3\%$) was lower in a previous study than in our study ($9.2\%$), and the proportion of patients achieving HbA1c <$7.0\%$ was $52.5\%$ [16], whereas our study found the same to be $33.3\%$. After it was reduced ($8.2\%$) in the first three months, the median HbA1c value did not significantly decrease at the follow-up, indicating that IDegAsp treatment lowers blood glucose within a few months.
Our findings on fasting blood glucose levels are consistent with the current literature. Twenty-two patients, with T2D were administered premixed insulin for the first 2 months, followed by IDegAsp for 2 months [18]; mean blood glucose levels (175.5 vs. 163.0 mg/dL; $$p \leq 0.004$$) were significantly lower in the IDegAsp phase when compared to the premixed phase measured before and after breakfast as well as before and after the evening meal [18]. Here, the insulin dose requirement increased by 14 U during two-year follow-up period when compared to the baseline level ($p \leq 0.0095$).
We also observed that there was no significant increase in, yet the total IDegAsp requirement did not significantly increase. The insulin requirement increased because the initial dose was reduced due to substituting previous insulin therapy with IDegAsp in all patients.
The IDegAsp coformulation simplifies patients’ lives and potentially improves glycemic control in the basal-bolus regimen when compared to concentrated insulin therapy [19]. Administrating fewer injections may better overcome barriers to insulin intensification and reduce clinical inertia when compared with BB insulin regimens. The basal and bolus components in IDegAsp offer flexibility in administration timings and a greater opportunity for individualized therapy when compared to traditional premixed preparations [20].
This study enrolled more patients, and included a longer follow-up time than previous real-world studies in Turkey [21, 22]. However, some key limitations should be considered when interpreting our study’s results. Firstly, whether the oral antidiabetic drug or IDegAsp coformulation affects glycemic control in patients who started with both oral antidiabetic drugs and the IDegAsp regimen remains unknown. Another limitation is that all the patients did not come to the follow-up until the second year, and some of them left the follow-up early in the sixth-month or the first year.
In conclusion, the usage of IDegAsp coformulation as a component of once-daily, twice-daily, and basal-bolus regimens over other insulin treatment options provides a therapeutic alternative to improve glycemic control, resulting in lowering of fasting blood glucose and HbA1c. Our findings provide important insights into the use of IDegAsp in a real-world setting in Turkey.
## Data Availability Statement
The original contributions presented in the study are included in the article/supplementary material. Further inquiries can be directed to the corresponding author.
## Ethics Statement
The studies involving human participants were reviewed and approved by Local ethics committee of the Marmara University School of Medicine. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.
## Author Contributions
HO, HG, and DY contributed equally to conception, design and writing of the manuscript. All authors revised the manuscript and approved the submitted version.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Ma Z, Parkner T, Christiansen JS, Laursen T. **IDegAsp: A Novel Soluble Insulin Analogs Combination**. *Expert Opin. Biol. Ther.* (2012) **12**. DOI: 10.1517/14712598.2012.722203
2. Haahr H, Fita EG, Heise T. **A Review of Insulin Degludec/Insulin Aspart: Pharmacokinetic and Pharmacodynamic Properties and Their Implications in Clinical Use**. *Clin Pharmacokinet* (2017) **56**. DOI: 10.1007/s40262-016-0455-7
3. Hirsch IB, Franek E, Mersebach H, Bardtrum L, Hermansen K. **Safety and Efficacy of Insulin Degludec/Insulin Aspart With Bolus Mealtime Insulin Aspart Compared With Standard Basal-Bolus Treatment in People With Type 1 Diabetes: 1-Year Results From a Randomized Clinical Trial (BOOST(®) T1)**. *Diabetes Med* (2017) **34**. DOI: 10.1111/dme.13068
4. Niskanen L, Leiter LA, Franek E, Weng J, Damci T, Muñoz-Torres M. **Comparison of a Soluble Co-Formulation of Insulin Degludec/Insulin Aspart vs Biphasic Insulin Aspart 30 in Type 2 Diabetes: A Randomised Trial**. *Eur Endocrinol* (2012) **167**. DOI: 10.1530/EJE-12-0293
5. Heise T, Nosek L, Roepstorff C, Chenji S, Klein O, Haahr H. **Distinct Prandial and Basal Glucose-Lowering Effects of Insulin Degludec/Insulin Aspart (IDegAsp) at Steady State in Subjects With Type 1 Diabetes Mellitus**. *Diabetes Ther* (2014) **5**. DOI: 10.1007/s13300-014-0070-2
6. Bebakar WM, Chaykin L, Hersløv ML, Rasmussen S. **Intensification of IDegAsp Twice Daily (Adding Insulin Aspart vs. Switching to Basal-Bolus): Exploratory Randomized Trial in Type 2 Diabetes**. *Diabetes Ther* (2017) **8** 197-205. DOI: 10.1007/s13300-016-0213-8
7. Havelund S, Ribel U, Hubálek F, Hoeg-Jensen T, Wahlund P-O, Jonassen I. **Investigation of the Physico-Chemical Properties That Enable Co-Formulation of Basal Insulin Degludec With Fast-Acting Insulin Aspart**. *Pharm Res* (2015) **32**. DOI: 10.1007/s11095-014-1614-x
8. Kaneko S, Chow F, Choi DS, Taneda S, Hirao K, Park Y. **Insulin Degludec/Insulin Aspart Versus Biphasic Insulin Aspart 30 in Asian Patients With Type 2 Diabetes Inadequately Controlled on Basal or Pre-/Self-Mixed Insulin: A 26-Week, Randomised, Treat-to-Target Trial**. *Diabetes Res Clin Pract* (2015) **107**. DOI: 10.1016/j.diabres.2014.09.026
9. Hirsch IB, Bode B, Courreges JP, Dykiel P, Franek E, Hermansen K. **Insulin Degludec/Insulin Aspart Administered Once Daily at Any Meal, With Insulin Aspart at Other Meals Versus a Standard Basal-Bolus Regimen in Patients With Type 1 Diabetes: A 26-Week, Phase 3, Randomized, Open-Label, Treat-to-Target Trial**. *Diabetes Care* (2012) **35**. DOI: 10.2337/dc11-2503
10. Fulcher GR, Christiansen JS, Bantwal G, Polaszewska-Muszynska M, Mersebach H, Andersen TH. **Comparison of Insulin Degludec/Insulin Aspart and Biphasic Insulin Aspart 30 in Uncontrolled, Insulin-Treated Type 2 Diabetes: A Phase 3a, Randomized, Treat-to-Target Trial**. *Diabetes Care* (2014) **37**. DOI: 10.2337/dc13-2908
11. Mehta R, Chen R, Hirose T, John M, Kok A, Lehmann R. **Practical Use of Insulin Degludec/Insulin Aspart in a Multinational Setting: Beyond the Guidelines**. *Diabetes Obes. Metab.* (2020) **22**. DOI: 10.1111/dom.14128
12. Kaneko S, da Rocha Fernandes JD, Yamamoto Y, Langer J, Faurby M. **A Japanese Study Assessing Glycemic Control With Use of IDegAsp Co-Formulation in Patients With Type 2 Diabetes in Clinical Practice: The JAGUAR Study**. *Adv Ther* (2021) **38**. DOI: 10.1007/s12325-021-01623-y
13. Kalra S, Baruah MP. **Insulin Degludec Aspart: One-Year Real World Experience**. *Indian J Endocrinol Metab* (2016) **20**. DOI: 10.4103/2230-8210.177416
14. Demir T, Turan S, Unluhizarci K, Topaloglu O, Tukek T, Gogas Yavuz D. **Use of Insulin Degludec/Insulin Aspart in the Management of Diabetes Mellitus: Expert Panel Recommendations on Appropriate Practice Patterns**. *Front Endocrinol (Lausanne)* (2021) **12**. DOI: 10.3389/fendo.2021.616514
15. Christiansen JS, Niskanen L, Rasmussen S, Johansen T, Fulcher G. **Lower Rates of Hypoglycemia During Maintenance Treatment With Insulin Degludec/Insulin Aspart Versus Biphasic Insulin Aspart 30: A Combined Analysis of Two Phase 3a Studies in Type 2 Diabetes**. *J Diabetes* (2016) **8**. DOI: 10.1111/1753-0407.12355
16. Taneda S, Hyllested-Winge J, Gall MA, Kaneko S, Hirao K. **Insulin Degludec/Insulin Aspart Versus Biphasic Insulin Aspart 30 Twice Daily in Insulin-Experienced Japanese Subjects With Uncontrolled Type 2 Diabetes: Subgroup Analysis of a Pan-Asian, Treat-to-Target Phase 3 Trial**. *J Diabetes* (2017) **9**. DOI: 10.1111/1753-0407.12407
17. Onishi Y, Yamada K, Zacho J, Ekelund J, Iwamoto Y. **Insulin Degludec/Insulin Aspart vs Biphasic Insulin Aspart 30 Twice Daily in Japanese Patients With Type 2 Diabetes: A Randomized Controlled Trial**. *J Diabetes Investig* (2017) **8**. DOI: 10.1111/jdi.12569
18. Fujimoto K, Iwakura T, Aburaya M, Matsuoka N. **Twice-Daily Insulin Degludec/Insulin Aspart Effectively Improved Morning and Evening Glucose Levels and Quality of Life in Patients Previously Treated With Premixed Insulin: An Observational Study**. *Diabetol Metab Syndr* (2018) **10** 64. DOI: 10.1186/s13098-018-0366-x
19. Rodbard HW, Cariou B, Pieber TR, Endahl LA, Zacho J, Cooper JG. **Treatment Intensification With an Insulin Degludec (IDeg)/insulin Aspart (IAsp) Co-Formulation Twice Daily Compared With Basal IDeg and Prandial IAsp in Type 2 Diabetes: A Randomized, Controlled Phase III Trial**. *Diabetes Obes Metab* (2016) **18**. DOI: 10.1111/dom.12609
20. Dardano A, Bianchi C, Del Prato S, Miccoli R. **Insulin Degludec/Insulin Aspart Combination for the Treatment of Type 1 and Type 2 Diabetes**. *Vasc Health Risk Manag* (2014) **10**. DOI: 10.2147/VHRM.S40097
21. Kisioglu SV, Demir AS, Tufekci D, Emur Gunay Y, Coskun H, Ucuncu O. **Clinical Research of Insulin Glargine U300 Basal-Bolus Therapy and Insulin Degludec/Aspart Co-Formulation in Type 2 Diabetes Mellitus: A Real World Experience**. *Int J Clin Pract* (2021) **75**. DOI: 10.1111/ijcp.14377
22. Onder CE, Kuşkonmaz SM, Koc G, Firat S, Omma T, Taskaldiran I. **Factors That Affect the Glycemic Control Achieved by Switching to Insulin Degludec/ Aspart in Insulin-Treated Patients With Type 1 and Type 2 Diabetes in a Real-World Setting: A non-Interventional, Retrospective Cohort Study**. *Acta Endocrinol. (Buchar.)* (2020) **16**. DOI: 10.4183/aeb.2020.443
|
---
title: 'Overcoming Barriers to Injectable Therapies: Development of the ORBIT Intervention
Within a Behavioural Change Framework'
authors:
- Karen McGuigan
- Alyson Hill
- Deirdre McCay
- Maurice O’Kane
- Vivien Coates
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2021
pmcid: PMC10012154
doi: 10.3389/fcdhc.2021.792634
license: CC BY 4.0
---
# Overcoming Barriers to Injectable Therapies: Development of the ORBIT Intervention Within a Behavioural Change Framework
## Abstract
It is estimated among individuals with type 2 diabetes (T2D) requiring injectable therapies to achieve optimal glycaemic control, one-third are reluctant to initiate therapies, with approximately $80\%$ choosing to discontinue or interrupt injectable regimens soon after commencement. Initiation of injectables is a complex issue, with effectiveness of such treatments undermined by non-adherence or poor engagement. Poor engagement and adherence are attributed to psychological aspects such as individuals’ negative perceptions of injectables, depression, anxiety, feelings of shame, distress and perceived lack of control over their condition. The aim of this study was to describe the development of a structured diabetes intervention to address psychological barriers to injectable treatments among a cohort of those with T2D; conducted within a behavioural change framework. An evidence base was developed to inform on key psychological barriers to injectable therapies. A systematic review highlighted the need for theory-based, structured diabetes education focussed on associated psychological constructs to inform effective, patient-centric provisions to improve injectable initiation and persistence. Findings from the focus groups with individuals who had recently commenced injectable therapies, identified patient-centric barriers to initiation and persistence with injectables. Findings from the systematic review and focus groups were translated via Behavioural Change Wheel (BCW) framework to develop an intervention for people with T2D transitioning to injectable therapies: Overcoming and Removing Barriers to Injectable Treatment in T2D (ORBIT). This article describes how psychological barriers informed the intervention with these mapped onto relevant components, intervention functions and selected behaviour change techniques, and finally aligned with behaviour change techniques. This article outlines the systematic approach to intervention development within the BCW framework; guiding readers through the practical application of each stage. The use of the BCW framework has ensured the development of the intervention is theory driven, with the research able to be evaluated and validated through replication due to the clarity around processes and tasks completed at each stage.
## Introduction
The complexities of treatment regimens for those living with Type 2 Diabetes (T2D) can be problematic, contributing to issues with medication adherence among this population [1]. This is particularly salient among those who require the use of injectable therapy i.e. GLP-1 receptor agonists or insulin, to achieve optimal glycaemic control [2]. It is estimated 1 in 3 people are reluctant to begin prescribed injectable therapy, with approximately $80\%$ believed to discontinue or interrupt injectable regimens quite quickly after commencement (3–5). Issues of non-adherence or poor engagement with injectables serves to undermine the effectiveness of these treatments. Poor injectable uptake and adherence have been attributed to “significant barriers in the minds of patients” [[6], s12]. Psychological aspects such as individuals’ perceptions of injectables, depression, anxiety, fear of injections, perceived pain, feelings of shame and failure can impact on engagement with therapies of this type (7–9). These psychological aspects are associated with poorer: clinical outcomes, initiation of injectable therapies, medication adherence and motivation which impact negatively on effective self-management behaviours (9–13).
Non-adherence to medication regimen, medical guidelines or treatment targets can be intentional or not, with people living with diabetes making conscious or sub-conscious determinations about the benefit of the treatment against the potential impact on daily functioning, wellbeing and quality of life [14]. The literature supports the links between individuals’ perceptions of their condition and their motivation to adhere to prescribed treatments [15, 16]. People with diabetes report concerns around treatment complexity, the restrictive nature of injectable regimens and the impact on day-to-day living [17]. Poor knowledge about the use of injectables, particularly insulin, can affect confidence in their use [9]. This, in turn, serves to increase the risk of associated complications, increase diabetes related distress, and adversely affect glycaemic control [9, 11].
As T2D is a condition that is primarily managed by the individual, efforts to address patient-related challenges of the condition have been recommended to improve self-management, medication adherence and outcomes [2, 18]. However, despite clear recognition of the importance of behavioural change to ensure effective self-management of T2D; behavioural change techniques and the psychological aspects which affect behaviour change have been overlooked in the development of structured diabetes interventions [19]. There has been significant underinvestment in interventions which target behaviour change with greater focus on the development of medications and devices to affect better outcomes [19]. Overcoming the challenges presented by injectable therapy would be best served through an educational intervention [6, 9]. However, the intervention must reflect best practice guidelines for structured diabetes education [20], respond to practical aspects of injectable use, and address psychological barriers to injectable uptake and maintenance [6, 9, 21].
When developing a new or novel intervention to change behaviour, few researchers provide a detailed description of the intervention development stages or processes, resulting in a lack of clarity for evaluation or replication of the intervention [22]. Whilst researchers suggest a theory or framework underpinning intervention design, this is often either poorly described or applied [23]. In other instances, intervention design is not guided by a theoretical framework, suggesting perhaps the existing options are not suited to the intervention aims [24]. The Medical Research Council (MRC) has provided guidance on the use of theoretical frameworks in intervention design, emphasising their effectiveness in identifying mechanisms for behavioural change [25]. Evidence-based interventions using appropriate theoretical frameworks are more likely to be successful in changing targeted behaviours [26].
Understanding and consequently attempting to change health behaviours is not a simple task. However, interventions aimed at affecting behaviour change are more likely to be effective if they are grounded in key psychological principles or theories of behaviour change [27]. Given the demands self-management places on the individual living with T2D, it is important to recognise the behavioural adaptations required to ensure optimal glycaemic control for those on injectables. This requires adherence to their treatment plan, monitoring blood glucose levels, improving diet and physical activity levels, as well as attending regular healthcare appointments [14]. Unsurprisingly, living with T2D has been described as a “chronic stressor for patients and families, affecting various life domains” (14,p.541). Accordingly, a behavioural change theory with a singular focus and little consideration for contextual factors may not be the best fit for an intervention to overcome barriers to injectable treatments for those with T2D. Roter et al. [ 28] were among some of the earlier researchers to assert that more comprehensive interventions with a combined focus would yield better outcomes, with single-focus interventions exhibiting less efficacy. They recommend interventions reflect psychological, behavioural and affective aspects to inprove effectiveness. Education provision for successful management of T2D using injectables cannot have a singular focus on injections or simple provision of information; instead requiring training on appropriate behavioural skills, coping strategies, individual practice, feedback and support [14, 29]. Indeed, whilst interventions based on a single theory may be easier to evaluate, they do not provide a comprehensive assessment of a clearly operationalised behavioural change problem [23]. A comprehensive theoretical framework for behavioural change is required.
The Behaviour Change Wheel (BCW) is a comprehensive theoretical model for behaviour change which was developed from a synthesis of 19 existing behavioural change frameworks, in essence ensuring a model that reflects their best practice from those [26, 30]. Accordingly, the BCW overcomes many of the issues that have hampered other frameworks or theories of behavioural change [31]. The BCW advocates a systematic approach for intervention design [32] offering a pragmatic, theoretical framework for health intervention development and evaluation that has been shown to successfully facilitate behavioural change [24].
The National Institute for Health and Care Excellence [33] guidelines for behavioural change advocate behaviours are a result of the interface between an individual’s capability and opportunity to perform that behaviour, and their motivation to do so. Reflecting this guidance, within the centre of the BCW, is the COM-B behaviour system. The COM-B system highlights the interaction between capability (C), opportunity (O) and motivation (M) necessary to perform a desired behaviour [34]. Behaviour arises as a function of someone’s physical and psychological capability (e.g.: skills and knowledge to perform the behaviour); the physical and social opportunity (e.g.: social cues/norms); and their automatic and reflective motivation (e.g.: impulsive response and cognitive evaluation of the benefit of performing the behaviour) [30, 35]. For a behavioural change intervention to be successful, one or more of these three factors in the COM-B system need to change [26, 31, 36].
The layer surrounding the central COM-B system comprises nine intervention functions through which behavioural change is promoted or encouraged [37]. Intervention functions are described as “broad categories by which an intervention can change behaviour”, emphasising an intervention may have “more than one function” [[31],p.166]. These functions are linked to the COM-B model, in essence showing more clearly which intervention functions are linked to desired behavioural change, e.g. education intervention affects change in psychological capability and reflective motivation.
The BCW also comprises seven policy categories [37]. These policy categories reflect the understanding that sometimes behavioural change occurs due to changes demanded or promoted by relevant authorities which serve to support or enable the adoption of new or revised behaviours [31]. For example, improved workplace health and safety practices as a result of legislation introduced by Government on safety and health at work. Lastly, the BCW allows linking of intervention function to behavioural change techniques, which are in essence observable and reproduceable components of the intervention [35]. A list of 93 behavioural change techniques have been listed and described for consideration in the behaviour change technique taxonomy to allow for appropriate alignment and operationalisation with intervention functions in the BCW [38].
With this in mind, this study sought to develop an intervention to overcome and remove the psychological barriers to injectable treatment in T2D, within a BCW framework.
## Preliminary Work
The importance of good primary research among those whom the intervention is developed to assist, is key, particularly in patient centric interventions or those targeting behavioural change [22]. A systematic approach has been taken to build an evidence base to understand the key psychological barriers to the initiation of, and adherence to, injectables. This approach reflects O’Cathain et al’s framework [39] to support implementation of the Medical Research Council guidance for development and evaluation of complex interventions [25].
A systematic review was undertaken [40] which reported on the need for theory-based, structured diabetes education to focus on associated psychological constructs to inform effective, patient-centric provisions to improve injectable initiation and persistence. More specifically the review found that successful diabetes education relied on facilitating change in participant cognition and behaviours, with psychosocial and behavioural change central in successful interventions. The review also confirmed diabetes education was more effective when led by Health care professionals (HCPs), with peer input, delivered in a group setting [40].
Involving people from a target population in the research development process allows a focus on aspects and experiences that are important for service users; and alignment with patient centric outcomes relevant to the target population [41]. To gain such insights, focus groups were conducted with individuals with T2D who had recently commenced injectable therapies. Focus group findings highlighted patient-centric issues, as well as the education requirements, to be addressed to increase uptake and adherence. The four main themes identified within the data were: 1. Beliefs about diabetes and injectable treatments. 2. Knowledge of diabetes and injectables. 3. Barriers to initiation and adherence. 4. Informing education design (Supplementary Information: Table of Results).
Findings from the systematic review and focus groups provided an evidence base to inform development of an intervention for people with T2D transitioning to injectable therapies: Overcoming and Removing Barriers to Injectable Treatment in T2D (ORBIT). The BCW, which captures the range of psychosocial and physical mechanisms necessary for optimal behavioural change, was used to provide a theoretical framework for the development of this intervention. The key stages in the design of interventions can be separated into three key aspects: i) Understanding the behaviour; ii) Identifying intervention options and iii) Identifying content and implementation options [30].
## Understanding the Behaviour
In this phase the foundations for successful intervention are laid, with each subsequent phase building on this initial phase. To address this, and in line with the central system in the BCW, the problem behaviour must be defined, then a target behaviour must be selected and specified with identification of the change(s) required [35, 42].
Define the problem in behavioural terms: This step requires definition of the problem, specifying the target group and the behaviour. In developing this intervention, identified gaps in the literature around addressing barriers to uptake and adherence to injectables, coupled with calls to provide interventions to tackle this issue, informed this definition. Therefore, suboptimal uptake and adherence to injectable therapies among T2Ds was defined as the problem to be addressed.
Select target behaviour: Although treatment targets exist for HbA1c and related physiological measures, the same consistency is not available for psychological and behavioural aspects. Accordingly, when developing a list of target behaviours for consideration, each should be considered in terms of impact, ease of change and measurement. For this intervention the targeted behaviour was increased uptake of injectable therapies and improved adherence.
Specify target behaviour: This aspect requires greater consideration of the target behaviour i.e. Who will perform this behaviour? What do they need to do differently? When will they do this? Where? With whom? How? These questions were addressed in relation to the existing literature and the views gathered from T2Ds using injectables.
Identify what needs to change: From the literature, systematic review [40] and focus group findings there are clear barriers and behaviours which need to be changed to address suboptimal uptake and adherence and improve same. These fall into key domains: Patient beliefs/perceptions, side effects, healthcare professional inertia, patient knowledge, psychological aspects, self-efficacy/mastery, control, daily life, education requirements and support (See Tables 1, 2).
## Identifying Intervention Functions
This stage of the design requires identification and selection of appropriate intervention functions and policy categories. The intervention functions are broad descriptors of methods by which the proposed intervention can serve to change behaviour in the target group [35]. Policy categories are external aspects which may help to support intervention delivery.
Identify intervention functions: Guided by the COM-B system, 6 of the 9 intervention functions to address problematic behaviour (barriers) were identified: Education, Persuasion, Modelling, Enablement, Training, and Coercion (See Table 1).
Identify policy categories: When exploring policy categories, it was evident that whilst the majority of the categories could be described as relevant to the intervention design, there were 4 key policy categories that best aligned with the selected intervention functions: guidelines, regulation, legislation and service provision. Guidelines and service provision were deemed to be most appropriate in relation to this intervention.
## Identifying Content and Implementation Options
In the final phase, behavioural change techniques need to be considered alongside a method of delivery for the proposed intervention.
Identify behaviour change techniques: Behavioural change techniques were identified using the behaviour change technique taxonomy [38]. The techniques were considered in conjunction with the identified barriers, intervention functions and policy categories. The capacity of the behaviour change technique to facilitate change and the potential for translation into the intervention were also used to guide selection (See Table 2).
Identify the mode of delivery: The mode of delivery is key to the effective translation of the behavioural change techniques into intervention content. The grouping of people with diabetes, their needs and the barriers to injectable initiation and continuation influenced the delivery mode, with face-to-face delivery in small groups identified as preferrable.
## Results
Key guidance and methodology from successful interventions have been used to inform the development of intervention functions, policy categories and behavioural change techniques to overcome and remove barriers to injectable treatment in T2D [31, 34, 43]. For this intervention the targeted behaviour was increased uptake of injectable therapies and improved adherence. Key barriers to be overcome in order to improve uptake of and adherence to injectable treatment in those with T2D have been examined within the existing literature and also captured via discussions with respondents with T2D currently prescribed injectable therapies. The identified barriers have been used to inform the intervention, within the BCW Framework, with these mapped onto the relevant COM-B components, intervention functions and selected behavioural change techniques (Table 1).
The final phase in this systematic process called for the identification of content and behaviour change techniques. Table 2 describes the behavioural change techniques drawn together from the COM-B system and intervention functions outlined in Table 1. In line with feedback from patients and guided by the literature, the newly developed ORBIT intervention consists of the delivery of the identified behavioural change techniques to people with diabetes prior to commencing injectable treatment.
## Discussion
The systematic process within the BCW framework provides a clear pathway for intervention design. The BCW has been used to develop interventions to facilitate a variety of behaviour change, in various contexts and across a range of populations (e.g. 34,35,43). At the centre of the model sits a deceptively simple behavioural system that takes cognisance of the key aspects required to engage in a given behaviour [32]. The COM-B system allowed for clear identification of the behaviour targeted for change, but also provided a basis for the selection of intervention functions. It is important that as the researcher moves through the related steps in the BCW they do not become overwhelmed by the number of options or the variety of behavioural change techniques. Key to this clarity is the preparatory work or primary research that is essential to inform intervention design when using the BCW [22]. This preparatory work is essential to ensure that patient-centric interventions, such as ORBIT, are developed in line with the needs of people with diabetes rather than relying on “practitioner or researcher intuition” (23,p.1). The use of the BCW framework in this process has ensured the development of the intervention is theory driven, with the research able to be more readily validated through replication due to the clarity around processes and tasks completed at each stage. The BCW responds to the call from the MRC for the use of theoretical frameworks in the development of behavioural change interventions to ensure accurate identification of the mechanisms for behavioural change [25]. It appears the BCW provides a simple and systematic framework for intervention design to affect behavioural change.
As noted in the Results, the newly developed ORBIT intervention will allow for delivery of the identified behavioural change techniques to people with diabetes prior to commencing injectable treatment. In line with the translation of the behavioural change techniques, best practice guidelines and clinical guidelines will be used to guide same [20, 33, 40]. Responding to this, the intervention should be delivered in a group setting, led by a trained HCP (Diabetes Specialist Dietitian), with opportunity for peer discussion and support. Participants should also receive related support materials. This intervention was designed to overcome and remove the psychological barriers to injectable treatment in T2D, to improve uptake and adherence to injectables. As such ORBIT has the capacity to improve outcomes for those living with T2D through better glycaemic control, by hastening adoption of injectables after prescription and helping ensure continued use.
## Limitations
Simply because the framework is systematic and well-devised does not ensure its effectiveness. This is dependent on the accurate operationalising of behaviours, identification of barriers, intervention categories and behavioural change techniques. However, even then success is not guaranteed as the skill is in the translation of behavioural change techniques into intervention content to ensure the intervention addresses the target behaviour within the grouping of people with diabetes. In this instance, although the intervention has been devised and the content developed, it needs to be evaluated with respect to the removal of the barriers it has been designed to overcome. In essence, the intervention should reduce or remove the identified barriers to injectable initiation and treatment adherence.
## Directions for Future Research
Trialling of the ORBIT intervention would be required. Appropriate reliable and valid individual reported outcome measures of identified barriers. The literature in the area and findings from the systematic review suggest these measures may assess people’s perceptions of their condition or condition related distress, knowledge, self-efficacy/mastery, anxiety, depression, and control should be utilised prior to, and following this intervention to evaluate efficacy. It should not be forgotten that the reduction in levels of depression, anxiety and distress, alongside improvements in knowledge, self-efficacy/mastery and control will not only evidence intervention efficacy; but should ultimately serve to improve injectable use and treatment persistence. To more objectively assess this, it would be important to monitor medication usage, weight (BMI) and HbA1c among those taking part in the intervention to add support for changes in the reported outcome measures by people with diabetes.
## Data Availability Statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics Statement
Ethical approval was granted by the Office of Research Ethics Committee in Northern Ireland, with governance from the Trust Research Governance Committee prior to the commencement of participant recruitment (15/NI/0091). All participants provided informed consent. The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
DM, AH and VC designed the study. DM, AH, VC, KM, and MO’K were active in the development of the preliminary work. KM and DM guided the development of the intervention via the BCW methodology. KM and DM wrote the first draft of the manuscript. AH and VC critically advised on important intellectual content and contributed to revising of the manuscript. All authors read and approved the manuscript for submission.
## Funding
This work was supported by the HSC Research and Development Division of the Public Health Agency, Northern Ireland (grant number: EAT/$\frac{4909}{13}$).
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary Material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcdhc.2021.792634/full#supplementary-material
## References
1. Zullig LL, Gellad WF, Moaddeb J, Crowley MJ, Shrank W, Granger BB. **Improving Diabetes Medication Adherence: Successful, Scalable Interventions**. *Patient Prefer Adherence* (2015) **9**. DOI: 10.2147/PPA.S69651
2. Gogas Yavuz D, Ozcan S, Deyneli O. **Adherence to Insulin Treatment in Insulin-Naïve Type 2 Diabetic Patients Initiated on Different Insulin Regimens**. *Patient Prefer Adherence* (2015) **9**. DOI: 10.2147/PPA.S87935
3. Peyrot M, Barnett AH, Meneghini LF, Schumm-Draeger PM. **Insulin Adherence Behaviours and Barriers in the Multinational Global Attitudes of Patients and Physicians in Insulin Therapy Study**. *Diabetes Med* (2012) **29**. DOI: 10.1111/j.1464-5491.2012.03605.x
4. Perez-Nieves M, Kabul S, Desai U, Ivanova JI, Kirson NY, Cummings AK. **Basal Insulin Persistence, Associated Factors, and Outcomes After Treatment Initiation Among People With Type 2 Diabetes Mellitus in the US**. *Curr Med Res Opin* (2016) **32**. DOI: 10.1185/03007995.2015.1135789
5. Perez-Nieves M, Ivanova JI, Hadjiyianni I, Zhao C, Cao D, Schmerold L. **Basal Insulin Initiation Use and Experience Among People With Type 2 Diabetes Mellitus With Different Patterns of Persistence: Results From a Multi-National Survey**. *Curr. Med. Res. Opin* (2017) **33**. DOI: 10.1080/03007995.2017.1341403
6. Edelman S, Pettus J. **Challenges Associated With Insulin Therapy in Type 2 Diabetes Mellitus**. *Am J Med* (2014) **127**. DOI: 10.1016/j.amjmed.2014.07.003
7. Polonsky WH, Fisher L, Guzman S, Villa-Caballero L, Edelman SV. **Psychological Insulin Resistance in Patients With Type 2 Diabetes: The Scope of the Problem**. *Diabetes Care* (2005) **28**. DOI: 10.2337/diacare.28.10.2543
8. Marrero DG. **Overcoming Patient Barriers to Initiating Insulin Therapy in Type 2 Diabetes Mellitus**. *Clin Cornerstone* (2007) **8** 33-43. DOI: 10.1016/s1098-3597(09)60006-5
9. Aghili R, Ridderstråle M, Farshchi A, Valojerdi AE, Banazadeh Z, Malek M. **Psychosocial Factors and Glycemic Control in Insulin-Naïve and Insulin-Experienced People With Type 2 Diabetes: A Path Analysis Model**. *Int. J Diabetes Dev Ctries* (2018) **38**. DOI: 10.1007/s13410-017-0581-2
10. Nefs G, Pop VJ, Denollet J, Pouwer F. **The Longitudinal Association Between Depressive Symptoms and Initiation of Insulin Therapy in People With Type 2 Diabetes in Primary Care**. *PloS One* (2013) **8** e78865. DOI: 10.1371/journal.pone.0078865
11. Lee KP. **Psycholosocial Factors Associated With Psychological Insulin Resistance in Primary Care Patients in Hong Kong**. *J Clin Transl Endocrinol* (2015) **2**. DOI: 10.1016/j.jcte.2015.10.001
12. McGuigan K, Hill A, Coates V, O'Kane M, Thompson DR, Ski CF. **Moderating the Relationship Between Diabetes Distress and Mastery: The Role of Depression and Empowerment**. *Psychol Health Med* (2021) **1** 1-10. DOI: 10.1080/13548506.2021.1894343
13. Peyrot M, Rubin RR, Kruger DF, Travis LB. **Correlates of Insulin Injection Omission**. *Diabetes Care* (2010) **33**. DOI: 10.2337/dc09-1348
14. Gonzalez JS, Tanenbaum ML, Commissariat PV. **Psychosocial Factors in Medication Adherence and Diabetes Self-Management: Implications for Research and Practice**. *Am Psychol* (2016) **71**. DOI: 10.1037/a0040388
15. Moss-Morris R, Weinman J, Petrie K, Horne R, Cameron L, Buick D. **The Revised Illness Perception Questionnaire (IPQ-R)**. *Psychol Health* (2002) **17** 1-16. DOI: 10.1080/08870440290001494
16. Hudson JL, Bundy C, Coventry PA, Dickens C. **Exploring the Relationship Between Cognitive Illness Representations and Poor Emotional Health and Their Combined Association With Diabetes Self-Care. A Systematic Review With Meta-Analysis**. *J Psychosom Res* (2014) **76**. DOI: 10.1016/j.jpsychores.2014.02.004
17. Allen NA, Zagarins SE, Feinberg RG, Welch G. **Treating Psychological Insulin Resistance in Type 2 Diabetes**. *J Clin Transl Endocrinol* (2016) **7** 1-6. DOI: 10.1016/j.jcte.2016.11.005
18. Sorli C, Heile MK. **Identifying and Meeting the Challenges of Insulin Therapy in Type 2 Diabetes**. *J Multidiscip Healthc* (2014) **7**. DOI: 10.2147/JMDH.S64084
19. McSharry J, Byrne M, Casey B, Dinneen SF, Fredrix M, Hynes L. **Behaviour Change in Diabetes: Behavioural Science Advancements to Support the Use of Theory**. *Diabetes Med* (2020) **37**. DOI: 10.1111/dme.14198
20. 20
National Institute for Health and Care Excellence. Diabetes in Adults. NICE Quality Standard [QS6]. London: NICE (2016).. *Diabetes in Adults. NICE Quality Standard [QS6]* (2016)
21. Weinger K, Beverly EA. **Barriers to Achieving Glycemic Targets: Who Omits Insulin and Why**. *Diabetes Care* (2010) **33**. DOI: 10.2337/dc09-2132
22. Owens C, Farrand P, Darvill R, Emmens T, Hewis E, Aitken P. **Involving Service Users in Intervention Design: A Participatory Approach to Developing a Text-Messaging Intervention to Reduce Repetition of Self-Harm**. *Health Expect* (2011) **14**. DOI: 10.1111/j.1369-7625.2010.00623.x
23. Cane J, O'Connor D, Michie S. **Validation of the Theoretical Domains Framework for Use in Behaviour Change and Implementation Research**. *Implement Sci* (2012) **7**. DOI: 10.1186/1748-5908-7-37
24. Michie S, van Stralen MM, West R. **The Behaviour Change Wheel: A New Method for Characterising and Designing Behaviour Change Interventions**. *Implement Sci* (2011) **6**. DOI: 10.1186/1748-5908-6-42
25. Craig P, Dieppe P, Macintyre S, Michie S, Nazareth I, Petticrew M. **Medical Research Council Guidance. Developing and Evaluating Complex Interventions: The New Medical Research Council Guidance**. *BMJ* (2008) **337**. DOI: 10.1136/bmj.a1655
26. Webb J, Foster J, Poulter E. **Increasing the Frequency of Physical Activity Very Brief Advice for Cancer Patients. Development of an Intervention Using the Behaviour Change Wheel**. *Public Health* (2016) **133** 45-56. DOI: 10.1016/j.puhe.2015.12.009
27. Abraham C, Kelly MP, West R, Michie S. **The UK National Institute for Health and Clinical Excellence Public Health Guidance on Behaviour Change: A Brief Introduction**. *Psychol Health Med* (2009) **14** 1-8. DOI: 10.1080/13548500802537903
28. Roter DL, Hall JA, Merisca R, Nordstrom B, Cretin D, Svarstad B. **Effectiveness of Interventions to Improve Patient Compliance: A Meta-Analysis**. *Med Care* (1998) **36**. DOI: 10.1097/00005650-199808000-00004
29. Norris SL, Engelgau MM, Narayan KM. **Effectiveness of Self-Management Training in Type 2 Diabetes: A Systematic Review of Randomized Controlled Trials**. *Diabetes Care* (2001) **24**. DOI: 10.2337/diacare.24.3.561
30. Atkins L, Michie S. **Designing Interventions to Change Eating Behaviours**. *Proc Nutr Soc* (2015) **74**. DOI: 10.1017/S0029665115000075
31. Lefevre CE. *The Behavioural Change Wheel in Action: Applying Behaviour Change Methods and Techniques to Environmental Issues* (2016)
32. Michie S, Atkins L, West R. *The Behaviour Change Wheel: A Guide to Designing Interventions* (2014) 322 p
33. 33
National Institute for Health and Care Excellence. Behaviour Change Individual Approaches. NICE Public Health Guideline [PH49]. London: NICE (2014).. *Behaviour Change Individual Approaches. NICE Public Health Guideline [PH49]* (2014)
34. Barker F, Atkins L, de Lusignan S. **Applying the COM-B Behaviour Model and Behaviour Change Wheel to Develop an Intervention to Improve Hearing-Aid Use in Adult Auditory Rehabilitation**. *Int J Audiol* (2016) **55**. DOI: 10.3109/14992027.2015.1120894
35. Mc Sharry J, Murphy PJ, Byrne M. **Implementing International Sexual Counselling Guidelines in Hospital Cardiac Rehabilitation: Development of the CHARMS Intervention Using the Behaviour Change Wheel**. *Implement Sci* (2016) **11** 134. DOI: 10.1186/s13012-016-0493-4
36. Smith CM, Griffiths F, Fothergill RT, Vlaev I, Perkins GD. **Identifying and Overcoming Barriers to Automated External Defibrillator Use by GoodSAM Volunteer First Responders in Out-of-Hospital Cardiac Arrest Using the Theoretical Domains Framework and Behaviour Change Wheel: A Qualitative Study**. *BMJ Open* (2020) **10** e034908. DOI: 10.1136/bmjopen-2019-034908
37. Carney R, Bradshaw T, Yung AR. **Physical Health Promotion for Young People at Ultra-High Risk for Psychosis: An Application of the COM-B Model and Behaviour-Change Wheel**. *Int J Ment Health Nurs* (2016) **25**. DOI: 10.1111/inm.12243
38. Michie S, Richardson M, Johnston M, Abraham C, Francis J, Hardeman W. **The Behavior Change Technique Taxonomy (V1) of 93 Hierarchically Clustered Techniques: Building an International Consensus for the Reporting of Behavior Change Interventions**. *Ann Behav Med* (2013) **46** 81-95. DOI: 10.1007/s12160-013-9486-6
39. O'Cathain A, Croot L, Duncan E, Rousseau N, Sworn K, Turner KM. **Guidance on How to Develop Complex Interventions to Improve Health and Healthcare**. *BMJ Open* (2019) **9** e029954. DOI: 10.1136/bmjopen-2019-029954
40. McCay D, Hill A, Coates V, O'Kane M, McGuigan K. **Structured Diabetes Education Outcomes: Looking Beyond HbA1c. A Systematic Review**. *Pract. Diabetes* (2019) **36** 86-90. DOI: 10.1002/pdi.2221
41. 41
Public Health Agency. HSC Research and Development Division: Strategy for Personal and Public Involvement in Health and Social Care Research. Belfast: PHA (2010).. *HSC Research and Development Division: Strategy for Personal and Public Involvement in Health and Social Care Research* (2010)
42. Sargent L, McCullough A, Del Mar C, Lowe J. **Using Theory to Explore Facilitators and Barriers to Delayed Prescribing in Australia: A Qualitative Study Using the Theoretical Domains Framework and the Behaviour Change Wheel**. *BMC Fam Pract* (2017) **18** 20. DOI: 10.1186/s12875-017-0589-1
43. Suntornsut P, Wongsuwan N, Malasit M, Kitphati R, Michie S, Peacock SJ. **Barriers and Recommended Interventions to Prevent Melioidosis in Northeast Thailand: A Focus Group Study Using the Behaviour Change Wheel**. *PloS Negl Trop Dis* (2016) **10** e0004823. DOI: 10.1371/journal.pntd.0004823
|
---
title: Elevated ACE Levels Indicate Diabetic Nephropathy Progression or Companied
Retina Impaired
authors:
- Kangkang Huang
- Yunlai Liang
- Kun Wang
- Yating Ma
- Jiahui Wu
- Huidan Luo
- Bin Yi
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2022
pmcid: PMC10012155
doi: 10.3389/fcdhc.2022.831128
license: CC BY 4.0
---
# Elevated ACE Levels Indicate Diabetic Nephropathy Progression or Companied Retina Impaired
## Abstract
### Objectives
Renin-angiotensin-aldosterone system plays important roles in the development of diabetic nephropathy (DN), and angiotensin converting enzyme (ACE) is the key factor in the process from angiotensin I to angiotensin II, but the variation and roles of serum ACE in DN patients are still unclear.
### Methods
Forty-four type 2 diabetes mellitus (T2DM) patients, 75 DN patients, and 36 age-gender-matched healthy volunteers were recruited who attended Xiangya Hospital of Central South University in this case control study. Serum ACE levels and other indexes were tested with commercial kit.
### Results
ACE levels in DN were significantly higher than T2DM and controls ($F = 9.66$, $P \leq 0.001$). Serum ACE levels significantly correlated with UmALB ($r = 0.3650$, $P \leq 0.001$), BUN ($r = 0.3102$, $P \leq 0.001$), HbA1c ($r = 0.2046$, $$P \leq 0.0221$$), ACR ($r = 0.4187$, $P \leq 0.001$), ALB (r = -0.1885, $$P \leq 0.0192$$), and eGFR (r = -0.3955, $P \leq 0.001$), and we got an equation that $Y = 2.839$ + 0.648X1 + 2.001X2 + 0.003X3 - 6.637X4 +0.416X5 - 0.134X6 (Y: ACE; X1: BUN; X2: HbA1C; X3: UmALB; X4: gender; X5: ALB; X6: eGFR, R2 = 0.655). When DN patients were divided into advanced-stage and early-stage with or without DR, ACE levels would increase when early-stage DN develops into advanced-stage or companied with DR.
### Conclusion
Elevated serum ACE levels may hint DN progression or retina impaired of DN patients.
## Introduction
Diabetes mellitus is a kind of metabolic disease with abnormal higher blood glucose. It was estimated that 463 million people were diagnosed with DM in 2019, and this number will arise to 700 million by 2045 among aged 18–99 years [1]. Long term hyperglycemia causes general vascular damage of kidneys, eyes, nerves, and heart and leads to microvascular complications [2], such as diabetic nephropathy (DN), diabetic retinopathy (DR), and diabetic peripheral neuropathy. Approximately 25–$35\%$ [3] of DM patients will develop into DN, which is one of the leading pathological causes of end-stage renal disease worldwide and is the most common cause of nephropathies requiring renal replacement therapy in many nations [4].
Except renal replacement therapy, researchers reported potential drugs such as empagliflozin [5], sirtuins 3 (SIRT3) [6], linagliptin [7], rho-associated kinase (ROCK) inhibitors [8], mineralocorticoid receptor antagonist [9], and peptide N-acetyl-seryl-aspartyl-lysyl-proline (AcSDKP) [10] could reduce the progression of DN in diabetic patients. The most common DN treatments are based on the RAAS system inactivation, precisely with the use of either the ACE inhibitors (ACEis) or angiotensin receptor blockers (ARBs) or their combination; however, ACEi could elevate AcSDKP level, whereas ARB does not [11]. Physiologically, renal epithelial cells are associated tightly with their neighbors, which prevent their potential for movement and dissociation from the epithelial layer. Under the effect of high glucose concentrations, glomerular podocyte would appear the phenotypic change of epithelial-to-mesenchymal transition (EMT) and endothelial to mesenchymal transition (EndMT) [12, 13]. When EMT and EndMT occurred, endothelial cells lost their typical phenotype to acquire mesenchymal features, characterized by the development of invasive and migratory abilities as well as the expression of typical mesenchymal products such as α-smooth muscle actin and type I collagen [12]. The glucocorticoid receptor (GR) is a nuclear hormone receptor that is expressed ubiquitously in most cell types. A previous study has reported that loss of endothelial GR activates Wnt signaling pathway. This pathway is known to be upregulated in renal fibrosis [14]. The most important inducer of kidney fibrosis is TGF-β, which could trigger EndMT by activating specific AKT and Smad signaling pathways [15]. Fibroblast growth factor 1 (FGFR1) as mitogen and insulin sensitizer could suppress inflammation and renal glomerular and tubular damage through inhibiting the activation of nuclear factor κB and c-Jun N-terminal kinase signaling pathways [16]. Notch signaling and Hedgehog signaling are also involved in the development of DN, Notch signaling promotes diabetic glomerulopathy [17] and tubulointerstitial fibrosis [18] and inhibits endothelial cell proliferation and migration [19], and Hedgehog signaling could reduce reactive oxygen species production through increasing superoxide dismutase and catalase production (20–22).
Several factors were proved associated with the occurrence of DN. DN is a prototype disease of the activation of renin-angiotensin-aldosterone system (RAAS) [23], including angiotensinogen, angiotensin receptor, Ang I, Ang II, ACE, and renin. Angiotensinogen is converted to Ang I by renin and Ang I transformed into Ang II through ACE. ACE gene polymorphisms have implications in the pathophysiology of diabetes developing into DN. It has been reported in several studies that D allele is a risk factor for DN and I allele is a protective factor for DN in Asian people [24, 25]. Further study implicated that D/D genotype is an independent risk factor for the development of DN in Chinese population with T2DM [25, 26]. In adults, plasma ACE level does not change with age, but it is affected by factors like environment or lifestyle [26, 27]. Plasma ACE level in I/D and D/D genotype people was reported to be $30\%$ and $60\%$ higher, respectively, than that in I/I genotype people [27].
Hence, what the variation of plasma ACE level in early and advanced stage DN patients with or without DR attracts us. In this research, we elaborated that serum ACE levels elevated in DN patients compared with T2DM and healthy. In the subgroup analysis, we separated DN patients with urinary microalbumin creatinine ratio (ACR) at the cut-off of 300 mg/g to early and advanced stage; meanwhile, we consider the DN patients whether retinal damage. As for DN patients with uninjured retina, increased serum ACE levels hint that early stage developed into advanced stage, and serum ACE levels raised are also a reminder of retina impaired in early-stage DN patients. The results of our study make a point that serum elevated ACE levels are a marker for progression of DN and early-stage DN patients with impaired retina.
## Participants
A total of 155 participants were recruited in this study from August 2018 to March 2020 in Xiangya Hospital of Central South University. Forty-four T2DM patients averaged 57.82 ± 9.56 years meeting the American Diabetes Association (ADA) standard in 2018 [28], including fasting blood-glucose (FPG) level of ≥126 mg/dl (7.0 mmol/L), a 2-h plasma glucose (2h-PG) level of ≥200 mg/dl (11.1 mmol/L) during OGTT, a hemoglobin A1c (HbA1c) level of ≥$6.5\%$ (48 mmol/mol), or a random plasma glucose ≥200 mg/dl (11.1 mmol/L) in a patient with classic symptoms of hyperglycemic crisis. Seventy-five patients averaged 56.60 ± 13.45 years were defined as DN on the condition of their ACR > 30 mg/g or eGFR < 60 ml•min-1•1.73 m2 [29]. Patients with type 1 diabetes mellitus (T1DM) or other kidney disease (like nephropathy syndrome or nephritis), cancer, cardiovascular disease, severe lung or liver disease, or infectious disease, or treated with nephrotoxic drugs or other drugs that could influence urinary albumin excretion, or receiving medications of angiotensin receptor blocker (ARB) or angiotensin-converting enzyme inhibitor (ACEI) were excluded. Meanwhile, 36 (53.33 ± 7.14 years) age-gender-matched healthy volunteers were introduced in this study. DR diagnosis was made by an ophthalmologist through direct ophthalmoscopy with retinal vascular structural changes such as microaneurysm, intraretinal hemorrhage, vascular circuity, and vascular malformation [30]. Participants in this study were given informed consent, and this study was permitted by the ethics committee of Xiangya Hospital of Central South University (No 202009119).
## Samples
Venous blood was collected after a minimum of 8 h fasting diet and then centrifuged at 3600 rpm for 5 min. The isolated serum samples were frozen at -20°C until test.
## Clinical and Laboratory Indexes
Clinical parameters of each participants were obtained including diagnostic message, gender, age, BMI, and laboratory indexes performed on blood samples containing ALB (albumin), TG (total triglycerides), TC (total cholesterol), HDL-C (high density lipoprotein cholesterol), LDL-C (low density lipoprotein cholesterol), TB (total bilirubin), DB (direct bilirubin), TBA (total bile acid), BUN (blood urea nitrogen), Scr (serum creatinine), UmALB (urine microalbumin), Ucr (urine creatinine), and HbA1c (glycosylated hemoglobin) measured on AU5800 automatic analyzer (Beckman Coulter, CA, USA). Serum ACE levels were tested with appropriate Commercial kit (DEROM Biomedical Engineering Co., LTD, Hunan, China). In addition, eGFR (estimated glomerular filtration rate) was computed by modified MDRD equation and ACR (urinary microalbumin creatinine ratio) was also calculated as below: eGFR [ml.min-1. ( 1.73m2)] = 175*[Scr (mg/dl)]-1.154* (age)-0.203, female multiple with 0.742, ACR (mg/g) = UmALB * (113.1* Ucr) -1*106.
## Statistical Analysis
Statistical analyses were performed using SPSS (Version 26, SPSS Inc., Chicago, IL, USA). The normal distribution quantitative statistics were presented as mean ± SD, the differences among groups were compared with ANOVA, and SNK test was used to verify the differences between two groups. Gender as dichotomous data was coded as “male=1” and “female=0”, and the difference among groups used chi-square test. Pearson correlation coefficients were reported for correlations between ACE and other indexes. Independent variable potentially influencing plasma ACE levels was tested by multiple linear regression analysis with input α = 0.05, output α = 0.1. All the tests were two-tailed, and $P \leq 0.05$ was considered statistically significant.
## Patient Baseline Characteristics
Baseline characteristics of the participants in our research were summarized in Table 1. ACE levels were 19.46 ± 7.67 (μmol/L) in the control group, 18.75 ± 12.96 (μmol/L) in T2DM, and 27.12 ± 11.93 (μmol/L) in DN, respectively. As shown in Supplementary Figure S1, the ACE levels in the DN group were remarkably elevated compared to that in the control and in T2DM ($P \leq 0.01$), whereas there was no significant difference of ACE levels between the T2DM group and the control. Meanwhile, the differences of age, BMI, duration, TBA, LDL, TC, and HDL/LDL between groups were not significant. Both the levels of BUN and Scr in DN patients were significantly higher than that in the control or T2DM patients, while the concentration of ALB, eGFR, TB, and DB in DN group was distinctly declined in contrast to the control or T2DM patients. HbA1c levels in DN or T2DM group were greatly higher than that in the control group.
**Table 1**
| Parameter | Control | T2DM | DN | F | P value (two-tailed) |
| --- | --- | --- | --- | --- | --- |
| Age (years) | 53.33 ± 7.14 | 57.82 ± 9.56 | 59.75 ± 9.02 | 2.06 | 0.13 |
| Gender, M/F, n | 20/16 | 26/18 | 50/25 | 0.001 | 0.973* |
| BMI (kg/m2) | 21.51 ± 2.98 | 22.44 ± 2.53 | 25.64 ± 23.22 | 0.99 | 0.376 |
| ACE (μmol/L) | 19.46 ± 7.68 | 18.75 ± 12.96 | 27.12 ± 11.93 | 9.66 | 0.000 |
| ALB (g/L) | 46.34 ± 3.38 | 39.64 ± 5.87 | 33.96 ± 7.61 | 47.07 | 0.000 |
| TG (mmol/L) | 1.47 ± 0.75 | 2.02 ± 1.58 | 2.70 ± 3.24 | 3.19 | 0.044 |
| TC (mmol/L) | 4.93 ± 0.88 | 4.34 ± 1.25 | 4.89 ± 1.86 | 2.16 | 0.119 |
| HDL (mmol/L) | 1.38 ± 0.36 | 1.05 ± 0.48 | 1.16 ± 0.83 | 2.55 | 0.082 |
| LDL (mmol/L) | 3.02 ± 0.68 | 2.74 ± 0.87 | 3.06 ± 1.24 | 1.4 | 0.25 |
| HDL/LDL. | 0.45 ± 0.19 | 0.47 ± 0.54 | 0.48 ± 0.74 | 0.02 | 0.978 |
| TB (μmol/L) | 10.98 ± 3.98 | 10.63 ± 4.53 | 7.98 ± 4.28 | 8.36 | 0.000 |
| DB (μmol/L) | 5.49 ± 1.98 | 4.68 ± 1.85 | 3.70 ± 2.05 | 10.57 | 0.000 |
| TBA (μmol/L) | 3.58 ± 2.18 | 6.65 ± 6.31 | 6.59 ± 8.27 | 2.78 | 0.065 |
| UmALB (mg/L) | / | 9.80 ± 7.78 | 970.91 ± 156.49 | 24.69 | 0.000 |
| BUN (μmol/L) | 4.59 ± 1.55 | 5.41 ± 1.89 | 9.19 ± 6.26 | 16.51 | 0.000 |
| Scr (μmol/L) | 74.33 ± 19.42 | 79.44 ± 18.97 | 230.46 ± 405.61 | 5.67 | 0.004 |
| HbA1c (%) | 5.36 ± 0.33 | 8.27 ± 3.07 | 8.07 ± 2.26 | 10.34 | 0.000 |
| UCr (mol/L) | / | 6704.52 ± 3713.47 | 5835.59 ± 3764.71 | 5.86 | 0.017 |
| eGFR [ml.min-1. (1.73m2)] | 82.96 ± 18.19 | 81.18 ± 17.97 | 48.99 ± 31.33 | 32.63 | 0.000 |
| ACR (mg/g) | / | 13.83 ± 7.74 | 2090.25 ± 2884.91 | 22.72 | 0.000 |
## Correlations Between ACE and Other Indexes
We used scatter diagram to explore the correlations between ACE and other laboratory indexes such as ALB, UmALB, BUN, Scr, GFR, BMI, TB, DB, TBA, HbA1c, and ACR. As a result, we found that ACE level significantly correlated with UmALB ($r = 0.3650$, $P \leq 0.001$), BUN ($r = 0.3102$, $P \leq 0.001$), HbA1c ($r = 0.2046$, $$P \leq 0.0221$$), ACR ($r = 0.4187$, $P \leq 0.001$), ALB (r = -0.1885, $$P \leq 0.0192$$), and eGFR (r = -0.3955, $P \leq 0.001$) as Table 2 described. Specifically, it had closely positive correlations with UmALB ($r = 0.4418$, $P \leq 0.0001$), BUN ($r = 0.3082$, $P \leq 0.0001$), HbA1c ($r = 0.2227$, $$P \leq 0.0129$$), or ACR ($r = 0.4094$, $P \leq 0.0001$), and negative correlations with ALB (r = -0.1885, $$P \leq 0.0192$$) or eGFR (r = -0.4091, $P \leq 0.0001$) as shown in Supplementary Figure S2, but no other significant correlations were observed between ACE levels and other laboratory data.
**Table 2**
| Parameter | UmALB | BUN | HbA1c | ACR | ALB | eGFR |
| --- | --- | --- | --- | --- | --- | --- |
| r | 0.3650 | 0.3102 | 0.2046 | 0.4187 | -0.1885 | -0.3955 |
| R2 | 0.1332 | 0.0962 | 0.0419 | 0.1753 | 0.0355 | 0.1564 |
| 95% CI | (0.1917–0.5163) | (0.1604–0.4460) | (0.0301–0.3671) | (0.2519–0.5613) | (0.3365, 0.03124) | (-0.5207–0.2537) |
| P value (two-tailed) | <0.001 | <0.001 | 0.0221 | <0.001 | 0.0192 | <0.001 |
## The Influence Factors of ACE
We used multiple logistic regression to investigate the influence factors of serum ACE level in serum. ACE level was defined as the dependent variable, and predictor variables include BUN, HbA1c, UmALB, gender, ALB, eGFR, BMI, input α = 0.05, and output α = 0.1. The equation was achieved as follows: $Y = 2.839$ + 0.648X1 + 2.001X2 + 0.003X3 - 6.637X4 + 0.416X5 - 0.134X6 (Y: ACE, X1: BUN, $$P \leq 0.093$$; X2: HbA1c, $$P \leq 0.000$$; X3: UmALB, $$P \leq 0.010$$; X4: gender, $$P \leq 0.006$$; X5: ALB, $$P \leq 0.014$$; X6: eGFR, $$P \leq 0.031$$, R2 = 0.655). The results given in Table 3 presented that these variables could influence $65.5\%$ of the ACE levels in the serum and only $35.5\%$ of the ACE affected by accidentia or other factors. The standardized regression coefficients of X1-X6 were 0.197, 0.387, 0.267, -0.227, 0.232, and -0.279, respectively, implicating that HbA1c could affect the ACE level most.
**Table 3**
| Parameter | BUN | HbA1c | UmALB | Gender | ALB | eGFR | Constant |
| --- | --- | --- | --- | --- | --- | --- | --- |
| β | 0.648 | 2.001 | 0.003 | -6.367 | 0.416 | -0.134 | 2.839 |
| β’ | 0.197 | 0.387 | 0.267 | -0.227 | 0.232 | -0.279 | |
| t | 1.697 | 4.722 | 2.621 | -2.789 | 2.497 | -2.196 | |
| P value (two-tailed) | 0.093 | 0.0 | 0.01 | 0.006 | 0.014 | 0.031 | |
## ACE Level in Patients With DN and DN Combined With DR
It has been widely accepted that ACR is a frequently-used laboratory index to diagnose DN and distinguish early-stage (30 mg/g < ACR < 300mg/g) and advanced-stage (ACR > 300 mg/g) DN [31]. Based on ACR levels, DN patients were divided into four subgroups of early-stage DN, early-stage DN combined with DR, advanced-stage DN, and advanced-stage DN patients DR. Key clinical and laboratory data of every subgroup were summarized in Table 4, and the ACE levels were exhibited in Figure 1. Statistical analysis proved that the ACE level in early-stage DN significantly dropped in comparison with the advanced-stage DN or early-stage DN combined with DR ($P \leq 0.05$) (Figure 1).
## Discussion
DN and DR is main microvascular complication of T2DM. DN is one of the most important causes of end-stage renal disease (ESRD), and DR would evolve into blind, at present, DN and DR rapidly increasing to be a popular disease in China [32]. In clinical practice, DN is characterized with proteinuria; however, diet, sports, and other factors could affect the levels of proteinuria [33]. Searching a biomarker to distinctively recognize DN to replace UmALB is helpful and valuable to diagnose DN. In our research, we aimed to assess the variation of ACE levels in DN patients compared with T2DM and healthy and the value of ACE levels to diagnose DN. Our results manifested that ACE level in the DN patients was significantly higher than that in the control or in the T2DM which may partly demonstrate that ACE levels may be a new marker to inflect DN. When we divided DN patients with ACR and considered whether DN patients companied with DR to subgroup, further subgroup analysis results showed that early-stage DN patients develop into advanced-stage or early-stage DN patients companied with DR, and serum ACE levels obviously increased. Above results of our study exhibited that elevated serum ACE levels may be a valuable biomarker to discern DN progression and early-stage DN patients with impaired retina.
Diabetic vascular complications are responsible for most of the mortality and morbidity in diabetic populations worldwide [34]. The complications are divided into macrovascular and microvascular. Macrovascular complications include coronary artery disease and cerebrovascular disease, and microvascular complications cover diabetic retinopathy (DR) and DN [35]. DR may have the identical pathologic change as DN; when T2DM patients have impaired retina, it is thought to be an indicator of DN [36]. In T2DM patients, DR and proteinuria could be decisive for the decline of renal function [37]. There is evidence that when T2DM patient was diagnosed with DR, it would be essential to assess their kidney function and DR may predict the renal outcomes of DN patients [38]. In our study, DR and DN coexisted in 35 patients, where the ACE level in these patients was higher than that in another 40 DN patients without DR; in particular, when ACR was <300 mg/g, ACE level in patients with DN concomitant DR was the time is prominently higher than that in DN patients without DR. It is reasonable to predict that ACE level could elevate in T2DM patients accompanied by retinopathy. Our study revealed that ACE level in DN patients was significantly higher than that in the control or T2DM; further subgroup analysis confirmed obvious elevation of ACE level in advanced-stage DN in comparison with the early-stage DN, and when early-stage DN patients companied with DR, serum ACE levels will also increase than early-stage DN patients.
As we all know that, RAAS plays vital roles to maintain plasma sodium, arterial blood pressure, and extracellular volume homeostasis. Angiotensinogen, angiotensin I, angiotensin II, and angiotensin converting enzyme are all important composition of RAAS, and Ang I cleaved into Ang II by ACE at lung capillaries, endothelial cells, and kidney epithelial cells (39–41). A study also reported that AngIIof intra-renal was 50–100 times higher than the circulatory AngII [42]. Proteinuria is a renal pathology and also the clinical characteristic of DN; however, high RAAS activity will cause or aggravate albuminuria [43]. Higher level of ACE reflects vast AngII, and a mass of AngIIindicates high RAAS activity. We imagined that higher serum ACE levels in DN patients were strongly correlated with proteinuria, and we probed into the correlation between serum ACE levels and renal function indexes such as BUN, UmALB, ACR, and eGFR, as you could see in Supplementary Figure S2. Strong relevance was gotten between ACE and UmALB ($r = 0.3566$), BUN ($r = 0.337$), ACR ($r = 0.4094$), and eGFR (r = -0.4091). The consequences also prove the idea that ACE is a potential biomarker to diagnose DN.
The present study has potential limitations. Firstly, as it was conducted in single Chinese population, ethic group difference should be regarded when popularizing the conclusion to other ethnics. Secondly, the number of patients recruited in this study was relatively small; therefore, enlarged sample size is required for further confirmation of the results.
In summary, our results imply that elevated serum ACE levels in DN patients may be an indicator for diabetic nephropathy, and continuously increased ACE is a possible signal of diabetic nephropathy progression. Additionally, increased ACE levels may be a single of retina impaired of early-stage DN patients.
## Data Availability Statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics Statement
The studies involving human participants were reviewed and approved by ethics committee of Xiangya Hospital of Central South University. The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
All authors participated in the design, interpretation of the studies, analysis of the data, and review of the manuscript. KH designed and analyzed the data and drafted the manuscript. YL and KW conducted the experiment. JW, HL, and YM collected the data. Bin Yi supplied critical reagents and reviewed and edited the text.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary Material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcdhc.2022.831128/full#supplementary-material
## References
1. Saeedi P, Petersohn I, Salpea P, Malanda B, Karuranga S, Unwin N. **Global and Regional Diabetes Prevalence Estimates for 2019 and Projections for 2030 and 2045: Results From the International Diabetes Federation Diabetes Atlas, 9(Th) Edition**. *Diabetes Res Clin Pract* (2019) **157**. DOI: 10.1016/j.diabres.2019.107843
2. **Global Health Risks**. *Mortality and Burden of Disease Attributable to Selected Major Risks* (2009)
3. Eppens MC, Craig ME, Cusumano J, Hing S, Chan AK, Howard NJ. **Prevalence of Diabetes Complications in Adolescents With Type 2 Compared With Type 1 Diabetes**. *Diabetes Care* (2006) **29**. DOI: 10.2337/dc05-2470
4. White SL, Chadban SJ, Jan S, Chapman JR, Cass A. **How can We Achieve Global Equity in Provision of Renal Replacement Therapy**. *Bull World Health Organ* (2008) **86**. DOI: 10.2471/blt.07.041715
5. Wanner C, Inzucchi SE, Lachin JM, Fitchett D, von Eynatten M, Mattheus M. **Empagliflozin and Progression of Kidney Disease in Type 2 Diabetes**. *N Engl J Med* (2016) **375**. DOI: 10.1056/NEJMoa1515920
6. Li J, Liu H, Takagi S, Nitta K, Kitada M, Srivastava SP. **Renal Protective Effects of Empagliflozin via Inhibition of EMT and Aberrant Glycolysis in Proximal Tubules**. *JCI Insight* (2020) **5**. DOI: 10.1172/jci.insight.129034
7. Awal HB, Nandula SR, Domingues CC, Dore FJ, Kundu N, Brichacek B. **Linagliptin, When Compared to Placebo, Improves CD34+ve Endothelial Progenitor Cells in Type 2 Diabetes Subjects With Chronic Kidney Disease Taking Metformin and/or Insulin: A Randomized Controlled Trial**. *Cardiovasc. Diabetol* (2020) **19** 72. DOI: 10.1186/s12933-020-01046-z
8. Shu A, Du Q, Chen J, Gao Y, Zhu Y, Lv G. **Catalpol Ameliorates Endothelial Dysfunction and Inflammation in Diabetic Nephropathy**. *Chem Biol Interact* (2021) **348**. DOI: 10.1016/j.cbi.2021.109625
9. Bakris GL, Agarwal R, Anker SD, Pitt B, Ruilope LM, Rossing P. **Effect of Finerenone on Chronic Kidney Disease Outcomes in Type 2 Diabetes**. *N Engl J Med* (2020) **383**. DOI: 10.1056/NEJMoa2025845
10. Srivastava SP, Goodwin JE, Kanasaki K, Koya D. **Metabolic Reprogramming by N-Acetyl-Seryl-Aspartyl-Lysyl-Proline Protects Against Diabetic Kidney Disease**. *Br J Pharmacol* (2020) **177**. DOI: 10.1111/bph.15087
11. Montinaro V, Cicardi M. **ACE Inhibitor-Mediated Angioedema**. *Int Immunopharmacol* (2020) **78**. DOI: 10.1016/j.intimp.2019.106081
12. Giordo R, Ahmed YMA, Allam H, Abusnana S, Pappalardo L, Nasrallah GK. **EndMT Regulation by Small RNAs in Diabetes-Associated Fibrotic Conditions: Potential Link With Oxidative Stress**. *Front Cell Dev Biol* (2021) **9**. DOI: 10.3389/fcell.2021.683594
13. Loeffler I, Wolf G. **Epithelial-To-Mesenchymal Transition in Diabetic Nephropathy: Fact or Fiction**. *Cells* (2015) **4**. DOI: 10.3390/cells4040631
14. Srivastava SP, Zhou H, Setia O, Liu B, Kanasaki K, Koya D. **Loss of Endothelial Glucocorticoid Receptor Accelerates Diabetic Nephropathy**. *Nat Commun* (2021) **12**. DOI: 10.1038/s41467-021-22617-y
15. Wang J, Liew OW, Richards AM, Chen YT. **Overview of MicroRNAs in Cardiac Hypertrophy, Fibrosis, and Apoptosis**. *Int J Mol Sci* (2016) **17**. DOI: 10.3390/ijms17050749
16. Liang G, Song L, Chen Z, Qian Y, Xie J, Zhao L. **Fibroblast Growth Factor 1 Ameliorates Diabetic Nephropathy by an Anti-Inflammatory Mechanism**. *Kidney Int* (2018) **93** 95-109. DOI: 10.1016/j.kint.2017.05.013
17. Fioretto P, Caramori ML, Mauer M. **The Kidney in Diabetes: Dynamic Pathways of Injury and Repair. The Camillo Golgi Lecture 2007**. *Diabetologia* (2008) **51**. DOI: 10.1007/s00125-008-1051-7
18. Bohle A, Wehrmann M, Bogenschütz O, Batz C, Müller CA, Müller GA. **The Pathogenesis of Chronic Renal Failure in Diabetic Nephropathy. Investigation of 488 Cases of Diabetic Glomerulosclerosis**. *Pathol Res Pract* (1991) **187**. DOI: 10.1016/s0344-0338(11)80780-6
19. Krebs LT, Xue Y, Norton CR, Shutter JR, Maguire M, Sundberg JP. **Notch Signaling Is Essential for Vascular Morphogenesis in Mice**. *Genes Dev.* (2000) **14**. DOI: 10.1101/gad.14.11.1343
20. Song N, Wang H, Gu T, Qi J, Yang J, Qiu Y. **Sonic Hedgehog-C-Jun N-Terminal Kinase-Zinc Finger Protein Gli1 Signaling Protects Against High Glucose Concentration-Induced Reactive Oxygen Species Generation in Human Fibroblasts**. *Exp Ther Med* (2018) **15**. DOI: 10.3892/etm.2018.6074
21. Al-Ayadhi LY. **Relationship Between Sonic Hedgehog Protein, Brain-Derived Neurotrophic Factor and Oxidative Stress in Autism Spectrum Disorders**. *Neurochem Res* (2012) **37** 394-400. DOI: 10.1007/s11064-011-0624-x
22. Zhang RY, Qiao ZY, Liu HJ, Ma JW. **Sonic Hedgehog Signaling Regulates Hypoxia/Reoxygenation-Induced H9C2 Myocardial Cell Apoptosis**. *Exp Ther Med* (2018) **16**. DOI: 10.3892/etm.2018.6678
23. Price DA, Porter LE, Gordon M, Fisher ND, De'Oliveira JM, Laffel LM. **The Paradox of the Low-Renin State in Diabetic Nephropathy**. *J Am Soc Nephrol JASN* (1999) **10**. DOI: 10.1681/ASN.V10112382
24. Zhong W, Jiang Z, Zhou TB. **RETRACTED: Association Between the ACE I/D Gene Polymorphism and T2DN Susceptibility: The Risk of T2DM Developing Into T2DN in the Asian Population**. *J Renin Angiotensin Aldosterone Syst* (2015) **16** NP35. DOI: 10.1177/1470320314566019
25. Wang F, Fang Q, Yu N, Zhao D, Zhang Y, Wang J. **Association Between Genetic Polymorphism of the Angiotensin-Converting Enzyme and Diabetic Nephropathy: A Meta-Analysis Comprising 26,580 Subjects**. *J Renin Angiotensin Aldosterone Syst* (2012) **13**. DOI: 10.1177/1470320311417655
26. Zhou TB, Yin SS, Qin YH. **Association Between Angiotensin-Converting Enzyme Insertion/Deletion Gene Polymorphism and End-Stage Renal Disease Susceptibility**. *J Renin Angiotensin Aldosterone Syst* (2014) **15** 22-31. DOI: 10.1177/1470320312460898
27. Glenn KL, Du ZQ, Eisenmann JC, Rothschild MF. **An Alternative Method for Genotyping of the ACE I/D Polymorphism**. *Mol Biol Rep* (2009) **36**. DOI: 10.1007/s11033-008-9313-5
28. American Diabetes A. **2. Classification and Diagnosis of Diabetes: Standards of Medical Care in Diabetes-2018**. *Diabetes Care* (2018) **41**. DOI: 10.2337/dc18-S002
29. **Microvascular Complications and Foot Care: Standards of Medical Care in Diabetes-2020**. *Diabetes Care* (2020) **43**. DOI: 10.2337/dc20-S011
30. **Microvascular Complications and Foot Care**. *Diabetes Care* (2017) **40**. DOI: 10.2337/dc17-S013
31. Bennett PH, Haffner S, Kasiske BL, Keane WF, Mogensen CE, Parving HH. **Screening and Management of Microalbuminuria in Patients With Diabetes Mellitus: Recommendations to the Scientific Advisory Board of the National Kidney Foundation From an**. *Am J Kidney Dis* (1995) **25**. DOI: 10.1016/0272-6386(95)90636-3
32. Xiong Y, Zhou L. **The Signaling of Cellular Senescence in Diabetic Nephropathy**. *Oxid Med Cell Longev* (2019) **2019**. DOI: 10.1155/2019/7495629
33. Szabóová E, Lisovszki A, Fatľová E, Kolarčik P, Szabó P, Molnár T. **Prevalence of Microalbuminuria and Its Association With Subclinical Carotid Atherosclerosis in Middle Aged, Nondiabetic, Low to Moderate Cardiovascular Risk Individuals With or Without Hypertension**. *Diagnostics (Basel)* (2021) **11**. DOI: 10.3390/diagnostics11091716
34. Naz I, Walters E, Akbari CM, Attinger CE, Kim PJ. **Noninvasive Vascular Assessment of Lower Extremity Wounds in Diabetics: Are We Able to Predict Perfusion Deficits**. *Surg Technol Int* (2017) **31** 66-74. PMID: 29310150
35. Mima A. **Incretin-Based Therapy for Prevention of Diabetic Vascular Complications**. *J. Diabetes Res* (2016) **2016** 1-12. DOI: 10.1155/2016/1379274
36. Wang X, Li J, Huo L, Feng Y, Ren L, Yao X. **Clinical Characteristics of Diabetic Nephropathy in Patients With Type 2 Diabetic Mellitus Manifesting Heavy Proteinuria: A Retrospective Analysis of 220 Cases**. *Diabetes Res Clin Pract.* (2019). DOI: 10.1016/j.diabres.2019.107874
37. Moriya T, Matsubara M, Kishihara E, Yoshida Y, Ouchi M. **Type 2 Diabetic Patients With Diabetic Retinopathy and Concomitant Microalbuminuria Showed Typical Diabetic Glomerulosclerosis and Progressive Renal Dysfunction**. *J. Diabetes Complications* (2016) **30**. DOI: 10.1016/j.jdiacomp.2016.04.007
38. Zhang J, Wang Y, Li L, Zhang R, Guo R, Li H. **Diabetic Retinopathy may Predict the Renal Outcomes of Patients With Diabetic Nephropathy**. *Ren Fail* (2018) **40**. DOI: 10.1080/0886022X.2018.1456453
39. Crisan D, Carr J. **Angiotensin I-Converting Enzyme: Genotype and Disease Associations**. *J Mol Diagn* (2000) **2**. DOI: 10.1016/s1525-1578(10)60624-1
40. Bernstein KE, Gonzalez-Villalobos RA, Giani JF, Shah K, Bernstein E, Janjulia T. **Angiotensin-Converting Enzyme Overexpression in Myelocytes Enhances the Immune Response**. *Biol Chem* (2014) **395**. DOI: 10.1515/hsz-2013-0295
41. Patel S, Rauf A, Khan H, Abu-Izneid T. **Renin-Angiotensin-Aldosterone (RAAS): The Ubiquitous System for Homeostasis and Pathologies**. *Biomed Pharmacother* (2017) **94**. DOI: 10.1016/j.biopha.2017.07.091
42. Kobori H, Nangaku M, Navar LG, Nishiyama A. **The Intrarenal Renin-Angiotensin System: From Physiology to the Pathobiology of Hypertension and Kidney Disease**. *Pharmacol Rev* (2007) **59**. DOI: 10.1124/pr.59.3.3
43. Basi S, Fesler P, Mimran A, Lewis JB. **Microalbuminuria in Type 2 Diabetes and Hypertension: A Marker, Treatment Target, or Innocent Bystander**. *Diabetes Care* (2008) **31 Suppl 2**. DOI: 10.2337/dc08-s249
|
---
title: A Changed Gut Microbiota Diversity Is Associated With Metabolic Improvements
After Duodenal Mucosal Resurfacing With Glucagon-Like-Peptide-1 Receptor Agonist
in Type 2 Diabetes in a Pilot Study
authors:
- Suzanne Meiring
- Annieke C. G. van Baar
- Nikolaj Sørensen
- Frits Holleman
- Maarten R. Soeters
- Max Nieuwdorp
- Jacques J. G. H. M. Bergman
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2022
pmcid: PMC10012157
doi: 10.3389/fcdhc.2022.856661
license: CC BY 4.0
---
# A Changed Gut Microbiota Diversity Is Associated With Metabolic Improvements After Duodenal Mucosal Resurfacing With Glucagon-Like-Peptide-1 Receptor Agonist in Type 2 Diabetes in a Pilot Study
## Abstract
### Introduction
The gut microbiota influences and interacts with the host metabolism through effects on nutrient metabolism and digestion. Duodenal Mucosal Resurfacing (DMR) is a novel endoscopic procedure involving duodenal mucosal ablation by the use of hydrothermal energy. DMR, when combined with a glucagon-like peptide-1 receptor agonist (GLP-1RA), resulted in discontinuation of exogenous insulin treatment in $69\%$ of patients with insulin dependent type 2 diabetes mellitus (T2DM) in the INSPIRE study. These patients also experienced improved glycaemic control and metabolic health. We thus investigated if these clinical effects were associated with a change in gut microbiota alpha and beta diversity.
### Methods
Faecal samples from the 16 patients were obtained for Illumina shotgun sequencing at baseline and 3 months after DMR. We assessed alpha and beta diversity of the gut microbiota in these samples and analysed its correlations with changes in HbA1c, body weight, and liver MRI proton density fat fraction (PDFF).
### Results
HbA1c correlated negatively with alpha diversity ($$p \leq 0.011$$, rho: -0.62) whereas changes in PDFF correlated significantly with beta diversity ($$p \leq 0.036$$, rho: 0.55) 3 months after initiation of the combined intervention. These correlations with metabolic parameters were observed despite finding no change in gut microbiota diversity at 3 months post DMR.
### Discussion
The correlation between gut microbiota richness (alpha diversity) and HbA1c as well as the change in PDFF and changed microbiota composition (beta diversity) suggests that changed gut microbiota diversity is associated with metabolic improvements after DMR in combination with glucagon-like-peptide-1 receptor agonist in type 2 diabetes. Larger controlled studies are however needed to find causal links between DMR with GLP-1RA, the gut microbiota, and improvements in metabolic health.
## Introduction
Diabetes mellitus is nowadays one of the most important public health challenges. About 1 in 11 adults worldwide has type 2 diabetes mellitus (T2DM) and its prevalence is still rising [1]. Many patients with T2DM eventually require insulin therapy to maintain glycaemic control, but this symptomatic therapy does not address insulin resistance, the root phenomenon of T2DM, and can contribute to a deterioration of metabolic health [2]. In this regard, exposure to a Western diet, rich in refined sugars and saturated fats, can lead to changes in gut microbial composition and physiology, which in turn are linked to development of T2DM [3]. In this regard, the gut microbiota interacts and influences the host metabolism through effects on nutrient metabolism and digestion, bile acid synthesis, energy metabolism, gut barrier function, the immune system, and xenobiotic metabolism. The interest for the role of the gut microbiota in the aetiology and potential treatment of obesity and T2DM has therefore grown over the last decades [3, 4]. Indeed there is consensus that T2DM patients are characterized by an overall lower gut microbiota diversity compared to healthy individuals [5].
Bariatric surgery is the most effective treatment for T2DM with a high long-term remission rate [6, 7] and major improvements in metabolic and cardiovascular health. Glycaemic improvements (6–9) occur overnight after surgery, even before any significant weight loss has occurred (10–12). Multiple studies have found an overall increase in gut microbial diversity in patients with T2DM after bariatric surgery [13]. Additionally, metabolic improvements in patients receiving a duodenal jejunal bypass liner are accompanied by an increase in microbial diversity [14]. It is therefore hypothesized that changes in gut microbiota composition play a regulating role in the positive glycaemic and metabolic effects after interventions involving the small bowel.
Duodenal Mucosal Resurfacing (DMR) is a minimally invasive endoscopic procedure that administers hydrothermal energy to ablate the duodenal mucosa [15]. Data from animal models and human studies suggest that a single DMR elicits improvements in insulin sensitivity, similar to the metabolic improvements seen after bariatric surgery, albeit in a lesser extent [16]. The physiological mechanism underlying the efficacy of DMR has yet to be elucidated, but we hypothesize that changes in the gut microbiota composition play a role in this effect, in line with other interventions involving the small bowel.
We recently published on the effect of a combination treatment of DMR and glucagon-like peptide-1 receptor agonist (GLP-1RA) administration in a pilot study of 16 patients with T2DM, treated with insulin and found an overall improvement in glycaemic and metabolic endpoints, leading to a discontinuation of insulin treatment in $69\%$ of the patients at 6 months. In the current spin-off study, we explored whether a change in gut microbiota diversity and composition was seen after initiation of the DMR and GLP-1RA combination treatment in these patients. We also explored whether these changes were associated with improved glycaemic and metabolic parameters, including liver fat fraction, and if change in specific bacterial species were seen.
## Study Design
The original pilot study was a single-centre, single-arm, prospective, open-label clinical study that investigated the effect of a single DMR procedure combined with GLP-1RA (liraglutide), in patients with T2DM, treated with insulin on glycaemic and metabolic health. The study protocol was approved by the medical ethics committee of the Amsterdam University Medical Center. The study was conducted in accordance with ICH Good Clinical Practice Guidelines and the Declaration of Helsinki. The study is registered under EudraCT number 2017-00349-30 at Clinicaltrialsregister.eu. The main clinical outcomes of this study have been reported previously. This report concerns a spin-off study, investigating the association between metabolic improvements and changes in gut microbiota after initiation of the combination treatment of DMR and GLP-1RA.
## Clinical Study Summary
We included 16 patients with T2DM using long-acting insulin, aged 28-75 years, with a body mass index of 24-40 kg/m2, and an HbA1c ≤ $8.0\%$ (64 mmol/mol). The list of inclusion and exclusion criteria can be found in Supplementary Table 1. Patients were primarily recruited via the general practitioner. Written informed consent was obtained from all patients. Endoscopic DMR was performed under deep sedation with propofol by a single endoscopist (JJGHM) with experience in endoscopic DMR procedures [15, 16]. The DMR procedure involved circumferential hydrothermal ablation of the duodenal mucosa using an over-the-guidewire catheter, as described previously [15, 16]. Exogenous insulin administration was discontinued the day before the DMR procedure according to the study protocol. Patients were instructed by a dietician to adhere to a tailored isocaloric 2-week post-procedural diet (i.e. gradual transition from liquid to solid food) to allow adequate healing of the duodenal mucosa. Patients began with self-administration of subcutaneous GLP-1RA, liraglutide (Victoza®, Novo Nordisk A/S) once daily at a standard dosage of 0.6 mg/day that was gradually increased to 1.8mg/day, as registered for treatment of T2D, after finishing the post-procedural diet. Standard mild nutritional counselling and lifestyle education were provided before DMR and during follow-up [17]. All oral glucose-lowering medications were continued in the same dosage throughout the study. Insulin was reintroduced (and liraglutide discontinued) when HbA1c was >$9.5\%$ at 3 months follow-up, >$7.5\%$ at 6, 12, 18 months follow-up or if self-measured fasting glucose was >270 mg/dl on 3 consecutive days. The primary endpoint of this pilot study was the percentage of patients free of exogenous insulin therapy with adequate glycaemic control, defined as HbA1c ≤ $7.5\%$ at the six-month follow-up (responders). Secondary endpoints were changes compared to baseline in glycaemic parameters during follow-up (HbA1c, HOMA-IR, FPG) and in metabolic parameters (BMI, ALT and PDFF) to evaluate additional metabolic benefits of the combination treatment applied in this pilot study. These primary and secondary endpoints have been published previously [17]. This additional report focusses on the changes in gut microbiota composition and diversity.
## Clinical and Anthropometric Evaluations
At baseline and at 6 months follow-up body weight and HbA1c levels were measured and magnetic resonance imaging (MRI; model clinical 3 Tesla scanner, Achieva, Philips) was used to calculate the liver proton density fat fraction (PDFF).
## Collection of Faecal Samples
Patients were instructed to collect a morning faeces sample prior to the visit at baseline (1 week before DMR) and 3 months after DMR. The sample was collected by the patient using gloves, put into a faeces collection tube and directly frozen at -20°C in their freezer at home. Once arrived at the hospital, all samples were immediately stored at -80°C until the actual analyses.
## Dna Extraction
Approximately 0.2 g of faecal material was used per extraction. DNA was extracted from samples using NucleoSpin® 96 Soil (Macherey-Nagel). Bead beating was done on a Vortex-Genie 2 horizontally for 5 min. A minimum of one negative control was included per batch of samples from the DNA extraction and throughout the laboratory process (including sequencing). A ZymoBIOMICS™ Microbial Community Standard (Zymo Research) was also included in the analysis.
## Shotgun Sequencing
Faecal microbiota sequencing was done at Clinical Microbiomics (Copenhagen, Denmark). Before sequencing, the quality of the DNA extractions was evaluated using agarose gel electrophoresis and the quantity by Qubit 2.0 fluorometer quantitation. The genomic DNA was randomly sheared into fragments of around 350 bp. The fragmented DNA was used for library construction using NEBNext Ultra Library Prep Kit for Illumina (New England Biolabs). The prepared DNA libraries were evaluated using Qubit 2.0 fluorometer quantitation and Agilent 2100 Bioanalyzer for the fragment size distribution. Quantitative real-time PCR (qPCR) was used to determine the concentration of the final library before sequencing. The library was sequenced using 2 × 150 bp paired-end sequencing on an Illumina platform.
## Gene Catalogue and Metagenomic Species Definitions
As a reference gene catalogue, we used the Clinical Microbiomics Human Gut 22M gene catalogue (22,459,186 genes), which was created from >5000 deep-sequenced human gut specimens. For MGS abundance profiling, we used the Clinical Microbiomics HGMGS v.2.3 set of 1273 metagenomic species (MGS), which have highly coherent abundance and base composition in a set of 1776 independent reference human gut samples. The approach is based on the metagenomic species concept [18]. To taxonomically annotate the MGSs, we blasted all the catalogue genes to the NCBI RefSeq genome database [2018-10-01] and used different levels of similarity (95, 95, 85, 75, 65, 55, 50 and $45\%$ for subspecies, species, genus, family, order, class, phylum, superkingdom, respectively) to annotate at the various taxonomic levels and requiring a minimum of $80\%$ sequence coverage. We calculated the percentage of genes of each MGS that mapped to each species and assigned species level taxonomy to a MGS if > $75\%$ of its genes could be annotated to a given species. For genus, family, order, class and phylum, we used 60, 50, 40, 30 and $25\%$ consistency, respectively. Furthermore, at species and at genus level, we allowed a maximum of $10\%$ of the genes belonging to a MGS to be annotated to an alternative taxon. Finally, we also used checkM to annotate the MGSs, and updated our annotations with checkM results in cases where checkM could annotate an MGS at higher resolution (lower taxonomic rank).
## Sequencing Data Pre-Processing
Quality control of raw FASTQ files was performed using KneadData (v. 0.6.1) to remove low-quality bases and reads derived from the host genome.
## MGS Relative Abundance Calculation
For each MGS, the signature gene set was defined as the 100 genes optimized for accurate abundance profiling of the MGS. A table of MGS counts was created based on the total gene counts for the 100 signature genes of each MGS. However, an MGS was considered detected only if read pairs were mapped to at least three of its 100 signature genes; counts for MGSs that did not satisfy this criterion were set to zero. This threshold has in internal benchmarks resulted in $99.6\%$ specificity.
## Functional Annotation and Profiling
Emapper software (v. 1.0.3, HMM mode) was used to compare each gene in the gene catalog to the EggNOG (v. 4.5) orthologous groups database (http://eggnogdb.embl.de/), resulting in annotations for $65\%$ of genes. *These* genes were then mapped from EggNOG to the Kyoto Encyclopedia of Genes and Genomes (KEGG) orthology (KO) database http://www.genome.jp/kegg/kegg1.html) using MOCAT2 lookup tables (http://mocat.embl.de/). Functional potential profiles based on KOs were calculated as the number of read pairs mapping to all genes annotated to a given KO, divided by the total number of mapped reads.
## Study Endpoints
The endpoints for this spin-off study were changes in gut microbiota alpha and beta diversity and abundance of bacterial species at 3 months after DMR compared to baseline. Additionally, we explored correlations between HbA1c, body weight and PDFF at 6 months and gut microbiota diversity at 3 months after DMR.
## Calculations and Statistical Analysis
Clinical parameters are expressed as medians with interquartile range [Q1-Q3]. Alpha diversity was calculated as richness (e.g. the number of MGSs observed in a sample). Beta diversity was calculated as dissimilarity in community composition between samples, using Bray-Curtis measurements. Bray-Curtis dissimilarity takes the abundance of species into account and can be 0–1, where 0 means that the two samples have identical compositions (they share all species at the same relative abundance), and 1 means that the two samples are completely different (they do not share any species). The abundance of MGSs in faecal samples before-treatment and after-treatment was compared using Wilcoxon signed-rank tests. Correlations between alpha diversity, beta diversity and abundance of MGSs and KOs were correlated with clinical variables (HbA1c, PDFF and body weight) using Spearman’s correlation. To correct for multiple hypothesis testing, false discovery rate was used for detecting false positives (using the conventional threshold of 0.10). Only MGSs with a prevalence of >$20\%$ in the investigated samples were included in the analyses.
## Patient Characteristics
All 16 enrolled patients underwent a successful DMR procedure. Patients were on average 61 years old, T2DM duration was 11 years, and patients used 31 units of insulin per day prior to DMR. All baseline characteristics can be found in Supplementary Table 2. At 6 months, $69\%$ of patients ($\frac{11}{16}$) met the primary endpoint of the study: i.e. off insulin therapy with an HbA1c ≤ $7.5\%$. Patients also demonstrated significant improvements in secondary endpoints regarding metabolic health: fasting plasma glucose, body weight, total body fat and average MRI liver PDFF decreased significantly. Details on these clinical outcomes have been published previously [17].
## Shotgun Sequencing Results
Of the 15-23 million high-quality non-host reads generated for each of the 32 faecal samples, 42-$89\%$ mapped to the gene catalogue [18] and the Clinical Microbiomics human gut MGS database. Approximately $50\%$ of the MGSs in all 32 faecal samples composed of Clostridiales, followed by Bacteroidales with ± $20\%$. Other taxa with high abundance were Acidaminococcales, Enterobacterales, Verrucomicrobiales and Selenomonadeles. Taxonomical profiles aggregated at order level are shown in Figure 1.
**Figure 1:** *Order level taxonomic composition of faecal samples by time of sampling before and after DMR. Each bar represents a single patient sample. DMR, duodenal mucosal resurfacing.*
## Microbiota Diversity and MGS Abundance Did Not Change Significantly Upon Intervention
We did not find a change in faecal microbiota alpha diversity (richness) nor beta diversity (community composition) at 3 months follow-up compared to baseline. Moreover, we did not find significant changes in specific bacterial species.
## Microbiota Diversity Correlated to Clinical Parameters
We observed an inverse correlation between the change in HbA1c and the change in gut microbiota alpha diversity ($$p \leq 0.011$$, rho: -0.62). Thus a decrease in HbA1c was correlated with an increase in gut microbiota richness and vice versa (Figure 2). Moreover, changes in PDFF correlated significantly with gut microbiota beta diversity (community composition as determined by Bray-Curtis dissimilarity of species, $$p \leq 0.036$$, rho: 0.55 and as determined by Bray-Curtis dissimilarity of KOs, $$p \leq 0.035$$, rho: 0.55). Thus, a large change in liver fat content is associated with a large change in the gut microbiota diversity (Figures 3A and 3B). No significant correlations between change in weight and change in gut microbiota diversity were found. Lastly, no significant correlations between changes in specific bacterial species and changes in other clinical variables were found.
**Figure 2:** *Change in HbA1c as a function of change in MGS richness in the faecal samples at baseline and 3 months after DMR. HbA1c, haemoglobin A1c; MGS, metagenomic species; DMR, duodenal mucosal resurfacing.* **Figure 3A:** *Change in PDFF as a function of MGS Bray-Curtis dissimilarity between faecal samples at baseline and 3 months after DMR. PDFF, proton density fat fraction; MGS, metagenomic species; DMR, duodenal mucosal resurfacing.* **Figure 3B:** *Change in PDFF as a function of KO Bray-Curtis dissimilarity between faecal samples at baseline and 3 months after DMR. PDFF, proton density fat fraction; KO, Kyoto Encyclopedia of Genes and Genomes (KEGG) orthology; DMR, duodenal mucosal resurfacing.*
## Discussion
We found that after initiation of DMR with GLP-1RA, a decrease in HbA1c was associated with an increase in gut microbiota richness in faecal samples. Moreover, larger changes in PDFF at 3 months after DMR was associated with a more changed gut microbiota diversity. We did not find a significant change in abundance of any bacterial species after compared to before DMR. However, the sample size of our pilot study was limited and the background variability is known to be high in gut microbiota studies, so these results do not exclude the possibility that duodenal ablation is associated with relevant changes in the gut microbiota. This is currently the only clinical study in which faecal samples of patients were obtained for gut microbiota analysis to unravel the underlying insulin-sensitizing mechanism of the novel DMR procedure for type 2 diabetes.
Our first key finding was the negative correlation between HbA1c and gut microbiota richness. This means that improvement in glycaemic control was correlated with increased gut microbiota diversity in our patients with T2DM. These findings are supported by a recently published systematic review of other metabolic procedures, reporting increased gut microbiota diversity in patients with metabolic improvements following endoscopic and surgical metabolic procedures across several studies [3]. However, we did not find increased diversity of the gut microbiota in our complete ($$n = 16$$) population after the intervention compared to baseline. This might be explained by the fact that $\frac{5}{16}$ of our patients did not develop glycaemic improvements post intervention, previously referred to as the non-responders. We can not exclude that this non-responder population affected the results in this rather small pilot population. Interestingly, in another spin-off report of this study, we observed increased postprandial unconjugated bile acids and increased secondary bile acids in the complete population [19]. These findings suggest that the gut microbiota composition did change considerably in our patients, since bile acid deconjugation, conversion, and active uptake of bile acids are executed by gut bacteria. These findings are a step forward in elucidating the mechanism behind DMR and thereby possibly improving efficacy of the DMR procedure. It also adds to the evidence that gut microbiota play a major role in glycaemic and metabolic health, and that a changed gut microbiota is associated with insulin resistance in the context of metabolic syndrome. It is possible that change in gut microbiota diversity in the complete population could not be captured by the current available techniques, and our small sample size can be explanatory as well.
Our second key finding was that a large change in liver MRI PDFF was correlated with a large change in gut microbiota diversity. Clinical and pre-clinical studies show that T2DM and non-alcoholic fatty liver disease (NAFLD) often co-exist as they share a common pathway of adipose tissue dysfunction and hepatic insulin resistance (20–22). Microbiota-derived-compounds reach the liver via the portal vein, where they can directly interfere with hepatocytes [23]. Our study results suggest an interesting interplay between gut microbiota richness and liver fat content. Currently there are no approved therapies for the treatment of NAFLD, therefore finding potential new targets (the gut microbiota) to treat NAFLD is desirable.
Since our combination treatment of DMR and GLP-1RA is offered as a package deal in this pilot study, we cannot identify the correlation of DMR or GLP-1RA individually with the gut microbiota. However, it is still interesting to speculate which of the two might be held largely responsible for the specific changes observed in the gut microbiota composition in the responders of the combination treatment. For GLP-1RA, there is no human data available of its influence on gut microbiota composition. In mice, treatment with liraglutide showed a trend toward increase in microbiota diversity, however this did not attain statistical significance [24]. We therefore hypothesize that the here presented changes in gut microbiota diversity are mainly attributable due to DMR. However, the exact mechanism on how DMR influences the gut microbiota remains unknown and a causal relationship has yet to be established.
As we mentioned previously, this spin-off study has some limitations. First, it is an observational uncontrolled proof-of-concept study with a limited sample size. Second, we cannot determine the effect of each individual treatment component on glycaemia and microbiota diversity. Third, as we did not monitor dietary intake, we cannot rule out that changes in microbiota diversity occurred as an epiphenomenon during metabolic improvements. Fourth, we did find significant correlations between a change in HbA1c and a change in gut microbiota alpha diversity, and a change in liver fat and gut microbiota beta diversity, but these correlations were less clear by visual examination of the scatterplots. Larger randomized controlled studies are necessary to elucidate causal links between metabolic improvements, the gut microbiota and DMR.
In conclusion, we found that improved glycaemic control in patients with T2DM after initiation of DMR with GLP-1RA was correlated with an increase in gut microbiota richness. Additionally, a large change in liver fat content was correlated with a large change in gut microbiota diversity. This data supports the important role of the gut microbiota in metabolic diseases and its possible potential to modify disease. However, causality cannot be proven with this study.
## Data Availability Statement
Publicly available datasets were analyzed in this study. This data can be found here: [https://www.ebi.ac.uk/ena/browser/view/PRJEB51010?show=reads].
## Ethics Statement
The studies involving human participants were reviewed and approved by Ethics committee of the Amsterdam UMC, location AMC. The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
Conceptualization, AB, FH, MS, MN, and JB. Investigation, SM and AB. Laboratory analyses, NS. Formal statistical analysis, SM, AB, and NS. Writing manuscript, SM and AB. Editing, AB, NS, FH, MN, and JB. Reviewing, AB, NS, FH, MS, MN, and JB. All authors contributed to the article and approved the submitted version.
## Funding
This study received funding from Fractyl Health Inc. The funder was not involved in the study design, collection, analysis, interpretation of data, the writing of this article or the decision to submit it for publication. MN is supported by a ZONMW VICI grant 2020 [09150182010020].
## Conflict of Interest
MN is in the Scientific Advisory Board of Caelus Pharmaceuticals, the Netherlands and Kaleido Biosciences, USA. However none of these are directly relevant to the current paper. FH reports speaker fees from Sanofi, Bioton, Astra Zeneca, and Boehringer Ingelheim. JB received research support from Fractyl Health Inc. for IRB-based studies and received a consultancy fee for a single advisory board meeting of Fractyl in September 2019.
The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary Material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcdhc.2022.856661/full#supplementary-material
## References
1. Zimmet P, Alberti KG, Magliano DJ, Bennett PH. **Diabetes Mellitus Statistics on Prevalence and Mortality: Facts and Fallacies**. *Nat Rev Endocrinol* (2016) **12**. DOI: 10.1038/nrendo.2016.105
2. Mudaliar S, Edelman SV. **Insulin Therapy in Type 2 Diabetes**. *Endocrinol. Metab. Clin. North Am.* (2001) **30**. DOI: 10.1016/S0889-8529(05)70222-X
3. van Olst SMea N. **Small Intestinall Physiology Relevant to Bariatric and Metabolic Endoscopic Therapies: Incretins, Bile Acid Signalling, and Gut Microbiome**. *Techniques Innovations Gastrointestinal Endoscopy* (2020) **22**. DOI: 10.1016/j.tige.2020.03.003
4. Debedat J, Amouyal C, Aron-Wisnewsky J, Clement K. **Impact of Bariatric Surgery on Type 2 Diabetes: Contribution of Inflammation and Gut Microbiome**. *Semin Immunopathol* (2019) **41**. DOI: 10.1007/s00281-019-00738-3
5. Li Q, Chang Y, Zhang K, Chen H, Tao S, Zhang Z. **Implication of the Gut Microbiome Composition of Type 2 Diabetic Patients From Northern China**. *Sci. Rep.* (2020) **10** 5450. DOI: 10.1038/s41598-020-62224-3
6. Adams TD, Davidson LE, Litwin SE, Kolotkin RL, LaMonte MJ, Pendleton RC. **Health Benefits of Gastric Bypass Surgery After 6 Years**. *JAMA* (2012) **308**. DOI: 10.1001/2012.jama.11164
7. Chang SH, Stoll CR, Song J, Varela JE, Eagon CJ, Colditz GA. **The Effectiveness and Risks of Bariatric Surgery: An Updated Systematic Review and Meta-Analysis, 2003-2012**. *JAMA Surg* (2014) **149**. DOI: 10.1001/jamasurg.2013.3654
8. Pucci A, Tymoszuk U, Cheung WH, Makaronidis JM, Scholes S, Tharakan G. **Type 2 Diabetes Remission 2 Years Post Roux-En-Y Gastric Bypass and Sleeve Gastrectomy: The Role of the Weight Loss and Comparison of DiaRem and DiaBetter Scores**. *Diabetes Med* (2018) **35**. DOI: 10.1111/dme.13532
9. Park JY. **Prediction of Type 2 Diabetes Remission After Bariatric or Metabolic Surgery**. *J Obes Metab Syndr* (2018) **27**. DOI: 10.7570/jomes.2018.27.4.213
10. Jackness C, Karmally W, Febres G, Conwell IM, Ahmed L, Bessler M. **Very Low-Calorie Diet Mimics the Early Beneficial Effect of Roux-En-Y Gastric Bypass on Insulin Sensitivity and Beta-Cell Function in Type 2 Diabetic Patients**. *Diabetes* (2013) **62**. DOI: 10.2337/db12-1762
11. Reed MA, Pories WJ, Chapman W, Pender J, Bowden R, Barakat H. **Roux-En-Y Gastric Bypass Corrects Hyperinsulinemia Implications for the Remission of Type 2 Diabetes**. *J Clin Endocrinol Metab* (2011) **96**. DOI: 10.1210/jc.2011-0165
12. Koliaki C, Liatis S, le Roux CW, Kokkinos A. **The Role of Bariatric Surgery to Treat Diabetes: Current Challenges and Perspectives**. *BMC Endocr Disord* (2017) **17** 50. DOI: 10.1186/s12902-017-0202-6
13. Kaska L, Sledzinski T, Chomiczewska A, Dettlaff-Pokora A, Swierczynski J. **Improved Glucose Metabolism Following Bariatric Surgery Is Associated With Increased Circulating Bile Acid Concentrations and Remodeling of the Gut Microbiome**. *World J. Gastroenterol* (2016) **22**. DOI: 10.3748/wjg.v22.i39.8698
14. de Jonge C, Fuentes S, Zoetendal EG, Bouvy ND, Nelissen R, Buurman WA. **Metabolic Improvement in Obese Patients After Duodenal-Jejunal Exclusion is Associated With Intestinal Microbiota Composition Changes**. *Int J Obes (Lond)* (2019) **43**. DOI: 10.1038/s41366-019-0336-x
15. Rajagopalan H, Cherrington AD, Thompson CC, Kaplan LM, Rubino F, Mingrone G. **Endoscopic Duodenal Mucosal Resurfacing for the Treatment of Type 2 Diabetes: 6-Month Interim Analysis From the First-In-Human Proof-Of-Concept Study**. *Diabetes Care* (2016) **39**. DOI: 10.2337/dc16-0383
16. van Baar ACG, Holleman F, Crenier L, Haidry R, Magee C, Hopkins D. **Endoscopic Duodenal Mucosal Resurfacing for the Treatment of Type 2 Diabetes Mellitus: One Year Results From the First International, Open-Label, Prospective, Multicentre Study**. *Gut* (2019) **69** 295-303. DOI: 10.1136/gutjnl-2019-318349
17. van Baar ACG, Meiring S, Smeele P, Vriend T, Holleman F, Barlag M. **Duodenal Mucosal Resurfacing Combined With Glucagon-Like Peptide-1 Receptor Agonism to Discontinue Insulin in Type 2 Diabetes: A Feasibility Study**. *Gastrointest Endosc* (2021) **94** 111-20.e3. DOI: 10.1016/j.gie.2020.12.021
18. Nielsen HB, Almeida M, Juncker AS, Rasmussen S, Li J, Sunagawa S. **Identification and Assembly of Genomes and Genetic Elements in Complex Metagenomic Samples Without Using Reference Genomes**. *Nat Biotechnol* (2014) **32**. DOI: 10.1038/nbt.2939
19. Meiring S, Meessen ECE, van Baar AC, Holleman F, Nieuwdorp M, Olde Damink SW. **Duodenal Mucosal Resurfacing With GLP-1 Receptor Agonism Increases Postprandial Unconjugated Bile Acids in Patients With Insulin-Dependent Type 2 Diabetes**. *Am J Physiol Endocrin Metabol* (2021) **322**. DOI: 10.1152/ajpendo.00337.2021
20. Rabot S, Membrez M, Bruneau A, Gerard P, Harach T, Moser M. **Germ-Free C57BL/6J Mice are Resistant to High-Fat-Diet-Induced Insulin Resistance and Have Altered Cholesterol Metabolism**. *FASEB J* (2010) **24**. DOI: 10.1096/fj.10-164921
21. Le Roy T, Llopis M, Lepage P, Bruneau A, Rabot S, Bevilacqua C. **Intestinal Microbiota Determines Development of Non-Alcoholic Fatty Liver Disease in Mice**. *Gut* (2013) **62**. DOI: 10.1136/gutjnl-2012-303816
22. Zeng H, Liu J, Jackson MI, Zhao FQ, Yan L, Combs GF. **Fatty Liver Accompanies an Increase in Lactobacillus Species in the Hind Gut of C57BL/6 Mice Fed a High-Fat Diet**. *J Nutr* (2013) **143**. DOI: 10.3945/jn.112.172460
23. Kolodziejczyk AA, Zheng D, Shibolet O, Elinav E. **The Role of the Microbiome in NAFLD and NASH**. *EMBO Mol Med* (2019) **11** e9302. DOI: 10.15252/emmm.201809302
24. Madsen MSA, Holm JB, Palleja A, Wismann P, Fabricius K, Rigbolt K. **Metabolic and Gut Microbiome Changes Following GLP-1 or Dual GLP-1/GLP-2 Receptor Agonist Treatment in Diet-Induced Obese Mice**. *Sci Rep* (2019) **9** 15582. DOI: 10.1038/s41598-019-52103-x
|
---
title: 'Clinical Team Response to the Impact of COVID-19 on Diabetes Self-Management:
Findings From a Qualitative Study'
authors:
- Lily Hale
- Thomas C. Cameron
- Katrina E. Donahue
- Maihan B. Vu
- Jennifer Leeman
- Asia Johnson
- Erica Richman
- Jennifer Rees
- Laura Young
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2022
pmcid: PMC10012158
doi: 10.3389/fcdhc.2022.835845
license: CC BY 4.0
---
# Clinical Team Response to the Impact of COVID-19 on Diabetes Self-Management: Findings From a Qualitative Study
## Abstract
The aims of this study were to explore providers’ perceptions of how COVID-19 affected patients’ psychological wellbeing and diabetes self-care and discover how providers responded to sustain and improve patients’ psychological health and diabetes management during the pandemic. Twenty-four semi-structured interviews were completed with primary care providers ($$n = 14$$) and endocrine specialty clinicians ($$n = 10$$) across sixteen clinics in North Carolina. Interview topics included: [1] current glucose monitoring approaches and diabetes management strategies for people with diabetes [2] barriers and unintended consequences encountered with respect to diabetes self-management, and [3] innovative strategies developed to overcome barriers. Interview transcripts were coded using qualitative analysis software and analyzed to identify cross-cutting themes and differences between participants. Primary care providers and endocrine specialty clinicians reported that people with diabetes experienced increased mental health symptoms, increased financial challenges and positive and negative changes in self-care routines due to COVID-19. To offer support, primary care providers and endocrine specialty providers focused discussions on lifestyle management and utilized telemedicine to connect with patients. Additionally, endocrine specialty clinicians helped patients access financial assistance programs. Findings indicate that people with diabetes experienced unique challenges to self-management during the pandemic and providers responded with targeted support strategies. Future research should explore the effectiveness of these provider interventions as the pandemic continues to evolve.
## Introduction
COVID-19 restrictions challenged patient-centered care for diabetes. Early in the pandemic, many ambulatory practices halted their in-person visits in favor of virtual appointments (1–3). This posed a problem for the management of diabetes, which typically requires routine in-office monitoring, lab testing, and medication management. In the first two months of the pandemic, rates of HbA1c tests fell by as much as $66\%$ [4]. There was concern among the diabetes healthcare community that the pandemic would result in poor diabetes outcomes, as interruptions in appointments and HbA1c testing have previously been associated with worsened glycemic control [5, 6].
In addition, people with diabetes have struggled with self-management practices in the face of COVID-19. People with diabetes report eating more and exercising less compared with prior to the pandemic (7–9). Members of the diabetes community also report concerns over finances and employment, which may have impacted insurance status and access to care [8]. Additionally, disruptions in diabetes supplies have affected individuals’ ability to adhere to medication and self-monitoring recommendations. A U.S. survey of people with diabetes found that one in six people needing insulin experienced a problem. A similar proportion of people had issues obtaining test strips, and $25\%$ had difficulties obtaining pumps or continuous glucose monitoring supplies [10].
In the face of these significant challenges, it is unsurprising that people with diabetes report higher levels of stress and depression than prior to the pandemic (10–12). Restrictions on gatherings limit individuals’ access to social support systems and may exacerbate these mental health effects [13]. Stress and depression affect glycemic control directly and through their impact on diabetes self-care [14]. Therefore, it is imperative for providers to recognize the diverse array of challenges COVID-19 presents to their patients with diabetes and develop targeted strategies for support. In this study, we [1] explore providers’ perceptions of how COVID-19 affected patients’ psychological wellbeing and diabetes self-care and [2] identify how providers responded to sustain and improve patients’ psychological health and diabetes management during the pandemic, with a focus on self-monitoring of blood glucose, A1c testing, dietary intake, and physical activity.
## Study Design
The study used a qualitative design to collect data through in-person and telephone interviews. These in-depth, semi-structured interviews provided rich and detailed descriptions regarding provider impressions of challenges facing patients with diabetes during COVID-19. This study was conducted in the context of a larger study, Re-Think the Strip (RTS), which aims to promote the de‐adoption of daily self‐monitoring of blood glucose (SMBG) among non‐insulin treated patients with controlled type 2 diabetes given the lower utility of this practice (15–18). Rethink the Strip involves 20 primary care clinics in the North Carolina. We were concerned that decreases in access to A1c test results would lead patients and providers to question the wisdom of de‐adopting SMBG during the pandemic. To address this concern, we conducted interviews with staff members within the existing RTS clinics and expanded the interviews to endocrinologists in North Carolina to obtain a more complete picture of diabetes care during COVID-19. Primary care providers focused their discussion on patients with type 2 diabetes, while endocrinologists commented on their experience treating patients with type 1 and type 2 diabetes.
## Study Population
Our study population comprised of 24 diabetes clinicians and staff. This included 9 primary care clinicians and 5 primary care ancillary staff members from the existing RTS practices, and 10 endocrine specialty clinicians. Participants are summarized in Table 1. Participants were identified and recruited using purposive sampling based on their ability to provide in-depth, detailed information about diabetes care during COVID-19.
**Table 1**
| Participant Role, Title | Number |
| --- | --- |
| Medical Director | 1 |
| Primary Care Physician (MD) | 2 |
| Nurse Practitioner (NP) | 1 |
| Practice Manager | 1 |
| Physician Assistant (PA) | 5 |
| Registered Dietician (RD) | 3 |
| Registered Nurse (RN) | 1 |
| Endocrinologist (MD) | 10 |
| Total | 24 |
## Data Collection
Semi-structured, in-depth telephone and in-person interviews were conducted from September 2020-August 2021. Before each interview, the researcher explained the goals of the study, reviewed confidentiality measures, and obtained verbal consent. All interviews were recorded using a field recorder with participant consent. The interviews were conducted in English. An interview guide (Table 2) was developed prior to participant interviews, based on literature review and previous work completed by the RTS team. Participants were asked a series of pre-determined open-ended questions and follow-up probes were generated based on participant responses. Each interview lasted approximately 35 minutes.
**Table 2**
| Dimensions | Open Ended Questions |
| --- | --- |
| Glucose Monitoring Approaches | As a health care provider, what have been the biggest changes in terms of patient care?How is your practice currently addressing A1c monitoring?How has your recommendation on glucose monitoring changes, if at all, since COVID-19? |
| Changes and Barriers in Patients Self-Management Practices | What concerns have your patients shared about their diabetes care during COVID-19?How has COVID-19 impacted your patients’ ability to adhere to your self-management recommendations?What are challenges your patients have shared about their new glucose monitoring approach or self-management? |
| COVID-19 Health and Outcomes | To get an overall sense of diabetes control, how often do other indicators like diet, weight, or blood pressure enter the discussions?During this time of COVID-19, for your patients with diabetes, what additional health issues or conditions have come to your attention?Since COVID-19, what changes, if any, have you noticed with your patients’ glycemic control? |
## Data Analysis
Interviews were professionally transcribed and reviewed by a team member for accuracy. Interview transcripts were independently analyzed by two investigators using content analysis. A codebook was developed by the research team, consisting of a priori codes derived from discussion questions and additional concepts that emerged from analysis. ATLAS.ti 9, a qualitative software program, was used to facilitate the analysis [19]. Differences and discrepancies were discussed and reconciled. Cross-cutting themes and differences between participants were identified.
## Results
Participant impressions of the impact of COVID-19 on patient self-management were categorized into five key themes, summarized in Table 3. Illustrative quotations from primary care providers and endocrine specialty clinicians are presented to highlight study findings.
**Table 3**
| Theme | Subtheme and Illustrative Quotation |
| --- | --- |
| Increase in mental health symptoms made self-care difficult. | Depression Then there’s been quite a bit of depression too. Not necessarily depression that needs medication, but definitely some decreased moods just with patients not being able to see their loved ones, or see their friends, or get out and do anything. (Primary Care Provider) |
| Increase in mental health symptoms made self-care difficult. | Anxiety We’ve seen more anxiety, a lot more anxiety, a lot more anxiety and depression as well. But I think it’s been, I would say, maybe 75/25 anxiety/depression for me because I only see women. I think women, especially women with young children or families, they’re taking this very seriously some of them. It’s really affecting them mentally because they’re the protectors of the children, and they can’t protect their children, and it’s really working their nerves very badly. (Primary Care Provider) |
| Increase in mental health symptoms made self-care difficult. | General stress and wellbeing In general I think more folks were just feeling down. In general you could feel that they were stressed. A lot of their normal routine was disrupted, and they were not able to keep up with their families just because everyone was somewhat staying at home. So, I think it was definitely a time where people were just not feeling well. (Endocrine Specialty Clinician) |
| Financial challenges impacted self-care. | Job or insurance loss I think they had more challenges because a lot of people did lose jobs and it was a big challenge to go to doctors and paying copays and everything. They lost insurances too. That way I think they’re also having issues paying bills for their medical health. (Endocrine Specialty Clinician) |
| Financial challenges impacted self-care. | New Expenses People who are not working their full hours, people who were having to pay for childcare that didn’t usually because the kids were in school … I think a lot of people were having to pinch a penny here or there, and if they weren’t really vested in doing [self-management] before, it was certainly one of the easier things to cut loose. (Primary Care Provider) |
| Restrictions disrupted self-care routines. | New family responsibilities The people who suffer the most are the same people who always suffer the most, women for sure, childcare. There was always kids in the background. They’re the ones for the most part stuck with childcare, and elder care, and everything else. (Endocrine Specialty Clinician) |
| Restrictions disrupted self-care routines. | Reduced safe exercise space They’re not able to go to the gym for a long time … Some of them are even afraid to go out and walk which I kind of told them that hey, you can go outdoors without a mask if you’re not around other people it’s okay but some were not willing to do that initially so many patients have gained weight. (Endocrine Specialty Clinician) |
| Restrictions disrupted self-care routines. | Reduced access to healthy food Of course, a lot of people are stress eating or tending to eat food because they really don’t want to go to the grocery store as often, so they’re not buying as many fresh fruits and vegetables. They’re tending to eat more processed, non-perishable foods. (Primary Care Provider) |
| Restrictions disrupted self-care routines. | Increased availability of unhealthy food Their access to food ended up being worse because they were doing drive-thru rather than going to the grocery store. (Endocrine Specialty Clinician) |
| Restrictions allowed patients to prioritize self-care. | Improved diet They would say, “Yeah, it was great to be home with the kids, and I also cooked all the time, so I did well with my blood sugar monitoring and carb intake,” (Endocrine Specialty Clinician) |
| Restrictions allowed patients to prioritize self-care. | Increased exercise I have occasional patients who now find themselves with more free time because of COVID. Maybe their work hours have been cut back or whatever, so they’re actually exercising more. (Primary Care Provider) |
| Restrictions allowed patients to prioritize self-care. | More time for self-management A lot of my patients cut down on one to two hours of commuting a day into their office … they were able to check for the first time three times a day before they ate their meals and could decide how much insulin to give which, in an office environment, sometimes, they wouldn’t check before lunch because they didn’t want to do it or didn’t have time. (Primary Care Provider) |
| Restrictions allowed patients to prioritize self-care. | Improved participation in care And that actually made my patients a little bit more participatory in their diabetes care because there was all these other setbacks meaning they couldn’t come in for routine bloodwork and evaluation in a clinic type setting. (Endocrine Specialty Clinician) |
| Clinical team utilized patient-centered strategies. | Prioritized lifestyle discussions With the gyms closing and now that they’re back open some people do feel comfortable to go back but some people are still saying “yeah, I’m still not feeling great about it” but then kind of having the discussion of “alright, I get it, I’m not going back to the gym either right now but it is still nice outside and you can walk” and then they’re like “oh yeah, I guess I could do that” … so having to make more of a conscious effort to go for a walk, to get up and move every hour or so to get some more steps in. (Primary Care Provider) |
| Clinical team utilized patient-centered strategies. | Utilized telemedicine I would say in the patient who is pretty tech savvy and can upload their CGM or their pump, the virtual visits actually work quite well because we have all of the data in front of us so especially if they’re on a CGM, we can estimate their A1C just from that. In that population, I think it actually went pretty well. (Endocrine Specialty Clinician) |
| Clinical team utilized patient-centered strategies. | Offered financial assistance We encouraged patients to apply for drug assistance, and sometimes helping them out with doing that. We directed them to clinics, pharmacies, clinics that have pharmacies for indigent patients that offer a drug program that they can access. All of that, and then we do have an active sampling system (Endocrine Specialty Clinician) |
## Theme 1: Providers Observed an Increase in Mental Health Symptoms Which Made Patient Self-Care More Difficult
Primary care and endocrine providers reported patients had increased mental health symptoms, mainly depression, anxiety, and general stress. These symptoms presented challenges to diabetes self-management.
## Depression
Both primary care providers and endocrine specialty clinicians noticed that more patients were struggling with depression compared with time prior to the pandemic. Some patients’ depression stemmed from isolation and lack of social support due to COVID-19 restrictions. One endocrine provider stated: I would say definitely mental health has suffered in all my patients including diabetes patients, lots of just social isolation and general isolation. I’m definitely seeing more depressive symptoms in that regard… A primary care provider echoed these sentiments: Then there’s been quite a bit of depression too. Not necessarily depression that needs medication, but definitely some decreased moods just with patients not being able to see their loved ones, or see their friends, or get out and do anything.
Isolation seemed to be especially difficult for patients with diabetes who had incorporated social support groups into their self-care routines. I think there was a component of maybe even like loneliness or what is it called, like the diabetes distress … I had an older population that did group classes and things like that. They didn’t have the gym to go out and socialize with, which caused them to be depressed, which would cause them to not take as good a care of themselves as they could. Patients that did walking groups, things like that, again, they didn’t have that support. ( Endocrine Specialty Clinician) Patients also experienced depression following the loss of family and friends during the pandemic. Depression is very common, stress and they have lost a lot of loved ones and their family which led to stress and stress eating which led to everything like the weight gain and poor control. ( Endocrine Specialty Clinician)
## Anxiety
Both endocrine specialty clinicians and primary care providers endorsed that patients experienced increased rates of anxiety during the pandemic. One source of anxiety was the risk of contracting COVID-19, especially given the high-risk status of people with diabetes. I’m definitely seeing more depressive symptoms in that regard, more anxiety as there was worry about contracting the virus and worry about going out in public, so I’ve definitely seen that. ( Endocrine Specialty Clinician) Another significant source of anxiety was financial concerns, as patients lost their job and were forced to juggle competing priorities. People are suddenly having to feel the financial losses and that sort of thing with COVID. I think people are definitely stress eating more or just distracted from their own health issues because they suddenly have – trying to figure out how to find another job or how they’re going to pay their rent. Dealing with their diabetes is sort of a lower thing on their priority list. ( Primary Care Provider) Finally, participants reported that caretakers experienced increased anxiety over the desire to protect their loved ones during the pandemic. We’ve seen more anxiety, a lot more anxiety, a lot more anxiety and depression as well. But I think it’s been, I would say, maybe $\frac{75}{25}$ anxiety/depression for me because I only see women. I think women, especially women with young children or families, they’re taking this very seriously some of them. It’s really affecting them mentally because they’re the protectors of the children, and they can’t protect their children, and it’s really working their nerves very badly. ( Primary Care Provider)
## General Stress and Well-Being
Primary care providers and endocrine participants noted that patients were also facing more general mental health challenges, such as “just feeling down”, or experiencing “high levels of stress”. These mood changes resulted in patients becoming increasingly isolated and, for many patients, translated to more difficulty with diabetes-self management practices. I see that the stress is impacting their eating, emotional eating. It’s impacting their blood pressure. It’s impacting their food choices which is ending up sometimes causing their cholesterol to go up or their blood sugars to be higher so emotional eating, stress eating and things like that. ( Primary Care Provider)
## Theme 2: Patients Encountered Financial Challenges Which Impacted Self-Care
Participants noted patients had trouble affording diabetes supplies, medications, or healthy food due to job or insurance loss or new expenses.
## Job or Insurance Loss
Both primary care providers and endocrine specialty clinicians explained that patients had trouble following self-management strategies due to “financial hardships with some people losing their job”. As one primary care provider explained, “patients who have lost their job often times also have lost their health insuranc”.. Financial pressures forced patients to choose what expenses to prioritize. Diabetes self-management practices “seem to be easier things to let go of because it’s not life-threatening to not test your blood sugar”. One endocrine specialty clinician summarized: We have several patients that had to start working two jobs because of COVID and a lot of them were laid off. So a lot of patients lost insurance during COVID. So then they stopped their medications and didn’t check their blood sugar because test strips are expensive.
## New Expenses
In addition to loss of income, primary care and endocrine participants noted that patients had new expenses, such as childcare. People who are not working their full hours, people who were having to pay for childcare that didn’t usually because the kids were in school … I think a lot of people were having to pinch a penny here or there, and if they weren’t really invested in doing [self-management] before, it was certainly one of the easier things to cut loose. ( Primary Care Provider) The cost of the medication in diabetes care has been an issue even as I said pre-pandemic and then the pandemic put more stress on it. People lost their jobs and things like that or they had some other expenses. Let’s say child care for example and things like that so the cost of diabetes medication I would say is probably … challenging. ( Endocrine Specialty Provider)
## Theme 3: Restrictions Created Disruptions to Patient Self-Care Routines
Participants discussed the ways that COVID-19 restrictions disrupted patient self-care routines, including new family responsibilities and changes to diet and exercise habits.
## New Family Responsibilities
Endocrine specialty clinicians noted that new “childcare” or “eldercare” responsibilities impacted patients’ ability to practice self-care. One clinician noted that for many patients, especially women, there were “always kids in the background” during virtual appointments. People with diabetes that were having distress related to care that had children at home and then were doing the homeschooling and things like that, their control definitely deteriorated because they were struggling to manage just day-to-day activities… (Endocrine Specialty Clinician) Another endocrine clinician wondered if family responsibilities may have played a role in the differences he observed between type 1 and type 2 diabetes patients: Actually, come to think of it, the majority of my type 1’s would be in their 30’s or maybe 40’s … And with the pandemic, schools were shut so everything was virtual. So now I can imagine if they had to help with their kids as well as thinking of themselves, that would have been a lot more challenging. With my type 2’s, a little bit older in general, and so their kids are already grown and that’s not a concern. So, maybe they had more me time. ( Endocrine Specialty Clinician)
## Reduced Safe Exercise Space
Endocrine and primary care providers discussed barriers to safe exercise during COVID-19. Gym closures disrupted patients’ exercise routine, and it was difficult to find a safe alternative. We have to talk a lot about weight and exercise even when the gym is closed or even when they don’t feel comfortable going to the gym even when it’s re- opened and when they’re trying to really socially distance, and so they don’t want to walk in the mall anymore, and that sort of thing. Of course, they all, by definition, are high risk if they got COVID, so it’s even trickier with that population to find ways that they can safely exercise. ( Primary Care Provider) For people with diabetes, their high-risk status created significant distress about leaving the house to exercise. One endocrine provider noted that some patients were “afraid to go out and walk”. A primary care provider with a similar experience said, “they feel like they have to wear a mask outside because somebody might walk too close to them and they can’t breathe with the mask outside because it was too hot”.
Additionally, working from home reduced patients’ level of activity. I’ve said multiple times to people, “You’re not working. You’re working from home. You’re sitting in front of a computer eight hours a day”. I have a teacher who she just absolutely couldn’t understand why she had gained weight, and I said, “You’re not chasing around a five-year-old anymore for eight hours a day”, and so we’ve seen a lot of that. ( Primary Care Provider) People did not walk. People didn’t go out to play or to exercise as much as they would. People didn’t go to the gym as much as they did. The lifestyle management aspect of diabetes management went out the window. ( Endocrine Specialty Clinician)
## Reduced Access to Healthy Food
Primary care and endocrine providers noted that concerns over COVID-19 impacted patients’ access to healthy food, because “These patients don’t want to get out and go to the grocery store, go to the farmers market or whatever”. Because patients were hesitant to make frequent trips to the grocery store, “they’re not buying as many fresh fruits and vegetables”. If patients did make it to the grocery store, healthy options were not always available. I also heard patients saying that they were eating what was available in the supermarkets and that they couldn’t always find what it is that they felt they should be eating, and they had to just take what was there at one point. ( Primary Care Provider)
## Increased Availability of Unhealthy Food
Primary care and endocrine providers discussed the increased accessibility of unhealthy food during the pandemic. An endocrine provider described that for some patients, “their access to food ended up being worse because they were doing drive-thru rather than going to the grocery store”. One primary care provider explained that if patients went to the grocery store, they tended to buy “more processed, non-perishable foods” to reduce the frequency of grocery trips. Additionally, a primary care provider explained that lockdowns and work from home placed many patients “within twenty feet of their refrigerator”, with constant access to less healthy foods.
Primary care and endocrine providers noted that this increased availability coupled with emotional eating led to an increase in unhealthy food consumption. Patients turned to food to deal with “stress”, “boredom” or to find “comfort”. As one primary care provider summarized, “People working at home and having easy access to snacks and being stressed and stress eating” often resulted in weight gain.
## Theme 4: COVID-19 Restrictions Allowed Patients to Prioritize Self-Care
More endocrine providers than primary care providers noted that for some patients, lockdowns allowed more time for self-management, encouraged healthier routines, and improved patient participation in care.
## Improved Diet
Endocrine and primary care providers noticed that lockdowns and working from home allowed patients to cook meals at home and pay more attention to their diet. One primary care provider stated that for many patients, eating out at restaurants “wasn’t an option, many more people were cooking more so I have noticed a change over time where a lot of people I was talking to were cooking more often, eating out less”. An endocrine provider with a similar experience recalled patients saying, “‘Yeah, it was great to be home with the kids, and I also cooked all the time, so I did well with my blood sugar monitoring and carb intake’”.
One endocrine provider noticed an improvement in patients with type 2 diabetes: I think it has gotten better for my type 2 because they were able to appreciate their caloric intake, and they were actually prepping the meals themselves. They no longer had access to eating at restaurants, which was the majority, 70 percent of their situation prior to Covid. And so I think it helps with betterment of glycemic control during the pandemic in the shutdown.
## Increased Exercise
Endocrine and primary care providers felt that for some patients, working from home led to increased opportunism for exercise. One endocrine provider stated, “a lot of my patients cut down on one to two hours of commuting time a day” which allowed them to “walk more” for exercise. A primary care provider observed how job loss allowed more time for exercise: I have occasional patients who now find themselves with more free time because of COVID. Maybe their work hours have been cut back or whatever, so they’re actually exercising more. I have a couple of patients who have actually lost a lot of weight during COVID.
## More Time for Self-Management
Endocrine and primary care providers discussed that some patients’ self-management improved because of “more free time”. An endocrine provider noted that patients “weren’t obligated to go somewhere to partake in or help out in other scenarios” so they “had more time to themselves”. One endocrine provider described how working from home made monitoring blood sugar more convenient: They were able to check for the first time three times a day before they ate their meals and could decide how much insulin to give which, in an office environment, sometimes, they wouldn’t check before lunch because they didn’t want to do it or didn’t have time, and they got home after dinner, and they were eating on the road, so they didn’t have time to check their sugar or give their insulin. For a small subset, maybe not small, but for a subset, I would say they increased their blood sugar monitoring because they had more time, and they were able to take better care of themselves.
## Improved Participation in Care
Endocrine providers discussed how aspects of the pandemic made patients “more participatory in their diabetes care”. One endocrine provider noted that because patients “couldn’t come in for routine bloodwork and evaluation in a clinic type setting”, they were more motivated to care for their diabetes. Another endocrine provider explained that the risk of COVID-19 encouraged more participation in care: *In* general, COVID was a very scary event for them. And so they tried really hard to get their sugars under control because they were very afraid of winding up in the hospital and getting very ill. And because there were fewer no-shows … because we did telemedicine they were more engaged. And our CDEs were excellent in having visits between the physician visits. And so overall, I think, after we all got through the craziness of COVID, we felt more connected. And patients felt more engaged, more a sense of responsibility over caring for themselves … And it’s not just a theoretical risk of harm if you don’t take your medications, but it was a very real risk. So I would say that there’s a lot of education that was achieved during COVID and then more ownership of their care.
## Theme 5: Clinical Team Members Utilized Patient-Centered Management Strategies
Primary care and endocrine providers discussed the strategies used to help patients maintain optimal self-care during the pandemic, including prioritizing lifestyle discussions, utilizing telemedicine, and helping with financial assistance.
## Prioritizing Lifestyle Discussions
Primary care and endocrine providers made lifestyle discussions a priority during COVID-19. Discussions with patients focused on creative ways to stay active despite pandemic restrictions. With the gyms closing and now that they’re back open some people do feel comfortable to go back but some people are still saying “yeah, I’m still not feeling great about it” but then kind of having the discussion of “alright, I get it, I’m not going back to the gym either right now but it is still nice outside and you can walk” and then they’re like “oh yeah, I guess I could do that” … so having to make more of a conscious effort to go for a walk, to get up and move every hour or so to get some more steps in. ( Primary Care Provider) When it came to weight, I tried to coach as much as possible because a lot of the gyms were shut down during COVID. What I would do is encourage them to go for short walks after the biggest meal of the day to help keep blood sugars under control as well as cholesterol. I encouraged them to start out at like just five minutes at a time and then build up to 30 minutes. That was probably the most common instructions I gave for weight management and weight control. ( Endocrine Specialty Clinician)
## Utilizing Telemedicine
Primary care and endocrine providers noted that the use of telemedicine had allowed them to stay connected to their patients during COVID-19. Although the physical exam was limited and accessing blood glucose data was challenging, providers felt that it was largely successful. I don’t think I’ve had anybody where we’ve kind of not been able to accomplish what we kind of want to get done [with a virtual visit]. Obviously we can’t do like foot exams and things like that. Obviously there are some cases where we want to know what the A1C is or we want to know what the blood sugar is and we can’t check it but I think in those cases where we’re kind of saying alright well let’s see how things are looking in another couple of months and we can do it then. But I think overall it’s been pretty good. I’ve been able to accomplish all I want to accomplish pretty much. ( Primary Care Provider) Endocrine participants noted that telemedicine visits were especially successful when patients were able to share glucose data through CGM devices or pumps. Yeah, my hope is that there will be a continued option for virtual because I do think it was helpful and worked well in certain circumstances. Yeah, I would hope that we would still have that as an option. Diabetes specifically, I would say in the patient who is pretty tech savvy and can upload their CGM or their pump, the virtual visits actually work quite well because we have all of the data in front of us so especially if they’re on a CGM, we can estimate their A1C just from that. In that population, I think it actually went pretty well. ( Endocrine Specialty Clinician) Endocrine providers felt telemedicine offered patients a convenient way to access their healthcare team. This seemed especially beneficial for patients in rural areas, or who were busy with work or family obligations. Now, it’s been great for patients that live very far away because I’m the only endocrinologist in this part or in this county that takes Medicare/Medicaid, so I have a lot of patients that travel greater than an hour to see me. They don’t necessarily have reliable transportation unfortunately, so being available by telehealth has allowed them to have their diabetes monitored more routinely. ( Endocrine Specialty Clinician) We have several patients that had to start working two jobs because of COVID and a lot of them were laid off. So they were doing you know a lot of temp jobs. And so they had no time to come into clinic. So when I called them for a tele-visit, it was almost always at work. And they were in their car or driving an Uber, for example, they would pull over and will conduct the visit. So we had extremely high satisfaction rates because we were able to offer tele medicines, you know, even up till now. And so actually our no-show rate was the lowest it’s been because people didn’t have to take public transportation in and they didn’t have to take out like a whole half day from work. ( Endocrine Specialty Clinician) Endocrine participants did note that the benefits of telemedicine were limited for older patients who had difficulty interfacing with the technology and for patients without reliable access to a computer or internet. So there’s definitely barriers when a patient doesn’t have access to technology, so if they don’t have a smartphone, or a computer, then you’re severely limited. And so it doesn’t work as well … But so I would say, you know, and that’s sad, because the patients that are economically disadvantaged, still have barriers to care, because this is mostly technology based. ( Endocrine Specialty Clinician) There just seems to be kind of an age barrier to the virtual as far as especially over 65 for a diabetes visit where they have to figure out how to upload [CGM] or something. I think that just becomes a significant burden. And sometimes even just trying to figure out how to get my microphone on or my camera on in that age group is just a big source of stress on their part. And they will be begging to come in even if we advise them not to. ( Endocrine Specialty Clinician)
## Offering Financial Assistance
Endocrine providers aimed to help patients afford diabetes supplies and medications by suggesting inexpensive options, offering product samples or helping patients apply for drug assistance programs. So I tell them to go to Walmart and buy a Walmart meter which I find that out of all the generic, cheap meters out there out there it’s probably more reliable than some of them and are more affordable and they can get it at a much cheaper price and the test strips are also a lot cheaper so I’m trying to encourage them to not worry about the insurance and get more testing supplies from Walmart. ( Endocrine Specialty Clinician) We encouraged patients to apply for drug assistance, and sometimes helping them out with doing that. We directed them to clinics, pharmacies, clinics that have pharmacies for indigent patients that offer a drug program that they can access. All of that, and then we do have an active sampling system (Endocrine Specialty Clinician)
## Discussion
The COVID-19 pandemic caused significant impacts on the lives of people with diabetes. Based on this qualitative study, primary care and endocrine providers noted that people with diabetes had increases in mental health symptoms, increased financial challenges and positive and negative changes in self-care routines. Primary care and endocrine providers discussed prioritizing lifestyle discussions and noted more use of telemedicine, specifically in blood glucose review. Results from this study can help guide management of people with diabetes as the pandemic evolves.
Our study found that most primary care and endocrine providers noted an increase in mental health concerns in their patients with diabetes. This included a subjective increase in the number of patients complaining of depression, anxiety and stress. These symptoms stemmed in part from social isolation and concern over the high risk of severe infection in the diabetes population. Diabetes was already associated with an increase in mental health conditions prior to the pandemic, and early in the pandemic concerns existed regarding worsening mental health during the pandemic in patients with diabetes [20]. A recent study found that $93\%$ of patients with diabetes showed signs of mental suffering and $43\%$ had signs of severe distress [12]. Additionally, one in 10 patients with diabetes had suicidal thoughts during the pandemic [21].
Primary care and endocrine providers noted disruptions in self-care in patients with diabetes during the pandemic. A significant theme was a decrease in physical activity and an increase in unhealthy eating patterns. People with diabetes struggled to find safe ways to exercise and obtain healthy foods while protecting themselves from the virus. However, a minority of providers reported patients had improved diets, exercise and time for self-management with a new change in routine. A recent qualitative study investigating the impact of the pandemic on Chinese patients with diabetes found decreases in physical activity, increased anxiety and feeling of lack of support from healthcare professionals [22]. However, distinct differences in this study included complaints of lack of access to blood glucose monitoring and lack of space to perform physical activity related to strict quarantine/isolation procedures. An additional study looking at the effect of COVID-19 on self-management in patients with type 2 diabetes using a DMSQ assessment found decreases in physical activity and decreased use of health services [23]. Therefore, the literature seems to support the negative changes in self-care routines most of our providers described.
An important question is how these changes in self-care ultimately effect glycemic control. Despite early concerns regarding glycemic control, numerous studies during the pandemic have shown no change in diabetes control in patients with diabetes [24, 25]. Additionally, a recent meta-analysis noted a modest improvement in many glucose control parameters in patients with type 1 diabetes during the pandemic [26]. The ability of some patients to improve their control may be due to improved diets and exercise regimen that a minority of providers noted in our study.
A finding in our study that may be unique to the US healthcare system is providers noting significant financial stress in patients with diabetes due to job loss and subsequent insurance loss. Early in the pandemic, a report estimated 7.3 million workers in the United States would lose health insurance coverage due to job loss [27]. A cross sectional survey conducted April 15-20, 2020 found $28\%$ of respondents had coronavirus related employment or earnings loss. Additionally, $45\%$ of respondents who had COVID job loss are not confident that they could pay for medical care or insurance premiums [28]. Though the number of total job losses is less than predicted, it is likely numerous people with diabetes have lost insurance and this has been noted by our participants. There is lack of ongoing research into insurance loss during the COVID-19 pandemic in people with diabetes.
Both primary care and endocrine providers in our study noted the widespread use of telemedicine early in the pandemic with a focus on CGM data review (endocrine specific providers) and increase in lifestyle-based discussions (all clinical team members). Providers did note difficulties with virtual visits in select elderly patients who did not have access to, or knowledge of, specific technologies. The beneficial effects of telemedicine during the COVID-19 pandemic have been noted in studies [29]. The perceived and actual benefit to clinicians and patients will need to be an area of further study and development.
Our study included primary care and endocrine providers. Providers reported similar experiences treating patients with diabetes during the early COVID-19 pandemic with a few notable differences. Endocrine providers discussed improvements in self-care in some patients with diabetes and emphasized the benefit of virtual care on CGM and insulin pump monitoring. These differences are likely due to the increased population of patients with type 1 diabetes and use of technology in endocrine specific clinics.
## Strengths/Limitations
To our knowledge, this is the first study to examine clinical team member impressions of challenges facing people with diabetes during the COVID-19 pandemic. Other studies have directly studied patients’ experiences and outcomes with various methods. It is important to study this issue from the clinical team perspective to gain insight into what challenges clinicians perceive to be impacting their patients with diabetes. Additionally, because each clinical team member interacts with multiple patients, they offer a more comprehensive overview of the topic than would be provided by individual patients. Using qualitative methods, we have been able to focus on overarching themes clinical team members have observed during this time.
That said, this study has several limitations. Because we did not directly interview patients with diabetes, our findings are representative of what care team members understood about their patients’ experiences during the pandemic. The comments from providers may not be completely accurate or comprehensive of their patients’ lived experiences. Although our principal interest was to investigate what providers perceived during this time, our overall understanding of the challenges facing people with diabetes during COVID-19 would have been enriched by directly interviewing patients with diabetes. A potential future direction of this work would include interviewing patients to obtain a more complete picture of living with diabetes during COVID-19.
Additionally, our method of purposive sampling has the potential to induce selection bias, as we selected participants based on their ability to provide robust information rather than collecting a random sample. To increase the external validity, we recruited participants from a diverse range of roles and practice settings. However, our study population was limited by high staff turnover and changes in staff roles in clinics due to COVID-19. Though clinical providers were solicited, only a select number responded and participated in the interview process. We were unable to engage every clinical team member, including social workers and mental health workers in clinics who could have added a unique perspective to our findings.
## Conclusion
Clinical team members of patients with diabetes described increases in mental health symptoms, financial stress and disrupted routines leading to self-care challenges during COVID-19. Participants additionally noted the increased use of telemedicine in the care of patients with diabetes during the pandemic. Our study of this clinical team cohort provides information regarding stressors facing patients with diabetes during the pandemic. It is important that these are noted as clinical team members continue to care for patients with diabetes as the pandemic evolves. Mental health support and lifestyle encouragement are critically important. Continued use of telemedicine may help many patients access these services.
## Data Availability Statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics Statement
The studies involving human participants were reviewed and approved by UNC Chapel Hill Office of Human Research Ethics. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.
## Author Contributions
LH: Recruited participants, conducted interviews, coded data, conducted analyses, wrote/edited manuscript. TC: Conducted interviews, coded data, wrote/edited manuscript. MV: Conducted interviews, coded data, conducted analyses, reviewed/edited the manuscript. JL: Developed interview guide, tested codebook, review/edited the manuscript. AJ: Coded data, reviewed/edited the manuscript. KD: Contributed to writing, reviewed/edited the manuscript. ER: Reviewed/edited the manuscript. JR: Study recruitment, reviewed/edited the manuscript. LY: Reviewed, edited manuscript. All authors contributed to the article and approved the submitted version.
## Funding
This work was supported by grant number T35-DK007386 from the National Institutes of Health and by a PCORI Dissemination and Implementation Contract: # DI-2018C1-10853 and NC Translational and Clinical Sciences (NC TraCS) Institute, which is supported by the National Center for Advancing Translational Sciences (NCATS), National Institutes of Health, through Grant Award Number UL1TR002489.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Hincapie MA, Gallego JC, Gempeler A, Pineros JA, Nasner D, Escobar MF. **Implementation and Usefulness of Telemedicine During the COVID-19 Pandemic: A Scoping Review**. *J Prim Care Community Health* (2020.0) **11** 1-7. DOI: 10.1177/2150132720980612
2. Ceccato F, Voltan G, Sabbadin C, Camozzi V, Merante Boschin I, Mian C. **Tele-Medicine Versus Face-to-Face Consultation in Endocrine Outpatients Clinic During COVID-19 Outbreak: A Single-Center Experience During the Lockdown Period**. *J Endocrinol Invest* (2020.0) **44**. DOI: 10.1007/s40618-020-01476-2
3. Mann DM, Chen J, Chunara R, Testa PA, Nov O, Mann D. **COVID-19 Transforms Health Care Through Telemedicine: Evidence From the Field**. *J Am Med Inf Assoc* (2020.0) **27**. DOI: 10.1093/jamia/ocaa072
4. Fragala MS, Kaufman HW, Meigs JB, Niles JK, McPhaul MJ. **Point of View Consequences of the COVID-19 Pandemic: Reduced Hemoglobin A1c Diabetes Monitoring**. *Popul Health Manage* (2021.0) **24** 8-9. DOI: 10.1089/pop.2020.0134
5. Karter AJ, Parker MM, Moffet HH, Ahmed A, Ferrara A, Liu J. **Missed Appointments and Poor Glycemic Control: An Opportunity to Identify High-Risk Diabetic Patients**. *Med Care* (2004.0) **42**. DOI: 10.1097/01.mlr.0000109023.64650.73
6. Hartmann-Boyce J, Morris E, Goyder C, Kinton J, Perring J, Nunan D. **Diabetes and COVID-19: Risks, Management, and Learnings From Other National Disasters**. *Diabetes Care* (2020.0) **43**. DOI: 10.2337/dc20-1192
7. Assaloni R, Pellino VC, Puci MV, Ferraro O, Lovecchio N, Girelli A. **Coronavirus Disease (Covid-19): How Does the Exercise Practice in Active People With Type 1 Diabetes Change? A Preliminary Survey**. *Diabetes Res Clin Pract* (2020.0) **166** 108297. DOI: 10.1016/J.DIABRES.2020.108297
8. Fisher L, Polonsky W, Asuni A, Jolly Y, Hessler D. **The Early Impact of the COVID-19 Pandemic on Adults With Type 1 or Type 2 Diabetes: A National Cohort Study**. *J Diabetes Complications* (2020.0) **34**. DOI: 10.1016/j.jdiacomp.2020.107748
9. Ruiz-Roso CK-TDM-EAGMS-NAD MB. **COVID-19 Lockdown and Changes of the Dietary Pattern and Physical Activity Habits in a Cohort of Patients With Type 2 Diabetes Mellitus**. *Nutrients* (2020.0) **12**. DOI: 10.3390/nu12082327
10. 10
Impact of COVID-19 on the Diabetes Community in the United States | Dq&a. Available at: https://d-qa.com/impact-of-covid-19-on-the-usa-diabetes-community/?utm_source=Closer+Look+Subscribers+2018&utm_campaign=4285f7ac19-2020-04-19_WIR_4%2F13-4%2F1704_18_2020&utm_medium=email&utm_term=0_c55d924bf1-4285f7ac19-409220105 (Accessed January 20, 2021).. *Impact of COVID-19 on the Diabetes Community in the United States | Dq&a*
11. Al-Sofiani ME, Albunyan S, Alguwaihes AM, Kalyani RR, Golden SH, Alfadda A. **Determinants of Mental Health Outcomes Among People With and Without Diabetes During the COVID-19 Outbreak in the Arab Gulf Region**. *J Diabetes* (2021.0) **13**. DOI: 10.1111/1753-0407.13149
12. Alessi J, De Oliveira GB, Franco DW, Brino Do Amaral B, Becker A, Knijnik C. **Mental Health in the Era of COVID-19: Prevalence of Psychiatric Disorders in a Cohort of Patients With Type 1 and Type 2 Diabetes During the Social Distancing**. *Diabetol Metab Syndr* (2020.0) **12** 76-86. DOI: 10.1186/s13098-020-00584-6
13. Sujan MSH, Tasnim R, Islam MS, Ferdous M, Apu M, Musfique M. **COVID-19-Specific Diabetes Worries Amongst Diabetic Patients: The Role of Social Support and Other Co-Variates**. *Prim Care Diabetes* (2021.0) **15**. DOI: 10.1016/j.pcd.2021.06.009
14. Snoek FJ, Bremmer MA, Hermanns N. **Constructs of Depression and Distress in Diabetes: Time for an Appraisal**. *Lancet Diabetes Endocrinol* (2015.0) **3**. DOI: 10.1016/S2213-8587(15)00135-7
15. Young LA, Buse JB, Weaver MA, Vu MB, Mitchell M, Blakeney T. **Glucose Self-Monitoring in Non-Insulin-Treated Patients With Type 2 Diabetes in Primary Care Settings: A Randomized Trial**. *JAMA Intern Med* (2017.0) **177**. DOI: 10.1001/jamainternmed.2017.1233
16. O’Kane MJ, Bunting B, Copeland M, Coates VE. **Efficacy of Self Monitoring of Blood Glucose in Patients With Newly Diagnosed Type 2 Diabetes (ESMON Study): Randomised Controlled Trial**. *BMJ* (2008.0) **336**. DOI: 10.1136/bmj.39534.571644.BE
17. Simon J, Gray A, Clarke P, Wade A, Neil A, Farmer A. **Cost Effectiveness of Self Monitoring of Blood Glucose in Patients With Non-Insulin Treated Type 2 Diabetes: Economic Evaluation of Data From the Digem Trial**. *Bmj* (2008.0) **336**. DOI: 10.1136/bmj.39526.674873.BE
18. Farmer A, Wade A, Goyder E, Yudkin P, French D, Craven A. **Impact of Self Monitoring of Blood Glucose in the Management of Patients With Non-Insulin Treated Diabetes: Open Parallel Group Randomised Trial**. *Br Med J* (2007.0) **335**. DOI: 10.1136/bmj.39247.447431.BE
19. Soratto J, de PDEP, Friese S. **Thematic Content Analysis Using ATLAS.Ti Software: Potentialities for Researchs in Health**. *Rev Bras Enferm* (2020.0) **73**. DOI: 10.1590/0034-7167-2019-0250
20. Mukhtar S, Mukhtar S. **Mental Health and Psychological Distress in People With Diabetes During COVID-19**. *Metabolism* (2020.0) **108** 154248. DOI: 10.1016/j.metabol.2020.154248
21. Alessi J, Scherer GDLG, Erthal IN, Teixeira J, de Oliveira G, Jaeger E. **One in Ten Patients With Diabetes Have Suicidal Thoughts After 1 Year of the COVID-19 Pandemic: We Need to Talk About Diabetes and Mental Health Not Only During Suicide Prevention Awareness Month**. *Acta Diabetol* (2021.0) **59**. DOI: 10.1007/s00592-021-01807-6
22. Shi C, Zhu H, Liu J, Zhou J, Tang W. **Barriers to Self-Management of Type 2 Diabetes During COVID-19 Medical Isolation: A Qualitative Study**. *Diabetes Metab Syndr Obes* (2020.0) **13**. DOI: 10.2147/DMSO.S268481
23. Utli H, Vural Doğru B. **The Effect of the COVID-19 Pandemic on Self-Management in Patients With Type 2 Diabetics**. *Prim Care Diabetes* (2021.0) **15** 799-805. DOI: 10.1016/j.pcd.2021.07.009
24. Dover AR, Ritchie SA, McKnight JA, Strachan MWJ, Zammitt NN, Wake DJ. **Assessment of the Effect of the COVID-19 Lockdown on Glycaemic Control in People With Type 1 Diabetes Using Flash Glucose Monitoring**. *Diabetes Med* (2021.0) **38** e14374. DOI: 10.1111/dme.14374
25. Patel SY, McCoy RG, Barnett ML, Shah ND, Mehrotra A. **Diabetes Care and Glycemic Control During the COVID-19 Pandemic in the United States**. *JAMA Intern Med* (2021.0) **181** 1412. DOI: 10.1001/JAMAINTERNMED.2021.3047
26. Garofolo M, Aragona M, Rodia C, Falcetta P, Bertolotto A, Campi F. **Glycaemic Control During the Lockdown for COVID-19 in Adults With Type 1 Diabetes: A Meta-Analysis of Observational Studies**. *Diabetes Res Clin Pract* (2021.0) **180**. DOI: 10.1016/j.diabres.2021.109066
27. Woolhandler S, Himmelstein DU. **Intersecting U.S. Epidemics: COVID-19 and Lack of Health Insurance**. *Ann Intern Med* (2020.0) **173**. DOI: 10.7326/M20-1491
28. Saloner B, Gollust SE, Planalp C, Blewett LA. **Access and Enrollment in Safety Net Programs in the Wake of COVID-19: A National Crosssectional Survey**. *PloS One* (2020.0) **5** e0240080. DOI: 10.1371/journal.pone.0240080
29. Predieri B, Leo F, Candia F, Lucaccioni L, Madeo S, Pugliese M. **Glycemic Control Improvement in Italian Children and Adolescents With Type 1 Diabetes Followed Through Telemedicine During Lockdown Due to the COVID-19 Pandemic**. *Front Endocrinol (Lausanne)* (2020.0) **11**. DOI: 10.3389/fendo.2020.595735
|
---
title: 'Prevalence and Risk Factors Associated With Gestational Diabetes Mellitus
Among Pregnant Women: A Cross-Sectional Study in Ghana'
authors:
- Wina Ivy Ofori Boadu
- Philomina Kugblenu
- Ebenezer Senu
- Stephen Opoku
- Enoch Odame Anto
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2022
pmcid: PMC10012159
doi: 10.3389/fcdhc.2022.854332
license: CC BY 4.0
---
# Prevalence and Risk Factors Associated With Gestational Diabetes Mellitus Among Pregnant Women: A Cross-Sectional Study in Ghana
## Abstract
Gestational diabetes mellitus (GDM) is a global public health issue that have serious consequences on mother and her child’s health. However, limited data is available on the prevalence of GDM and its associated risk factors in Ghana. This study investigated the prevalence and associated risk factors of GDM among women attending selected antenatal clinics in Kumasi, Ghana. This cross-sectional study included 200 pregnant women who attended antenatal clinics from Three-selected health facilities in the Ashanti Region, Ghana. Women already diagnosed of GDM were identified through their medical records and were confirmed based on the criteria of the International Association of Diabetes and Pregnancy Study Groups (IADPSG), which uses a fasting blood glucose of ≥ 5.1 mmol/L. A well-structured questionnaire was used to collect data on socio-demographic, obstetric, clinical and lifestyle risk factors. Multivariate logistic regression models were used to determine the independent risk factors of GDM. The overall prevalence of GDM among study participants was $8.5\%$. GDM was prevalent among age 26 and 30 years ($41.2\%$), married participants ($94.1\%$) with basic education ($41.2\%$) and being Akan by ethnicity ($52.9\%$). Previous history of oral contraceptive use (aOR: 13.05; $95\%$ CI: 1.43–119.23, $$p \leq 0.023$$), previous history of preeclampsia (aOR: 19.30; $95\%$ CI: 2.15-71.63; $$p \leq 0.013$$) and intake of soda drinks (aOR: 10.05, $95\%$ CI: 1.19–84.73, $$p \leq 0.034$$) were found to be independent risk factors of GDM. The prevalence of GDM was found to be $8.5\%$ and this was associated with the previous use of oral contraceptives, history of preeclampsia and intake of soda drinks. Public health education and dietary lifestyle modification may be required for pregnant women who are at risk of GDM.
## Introduction
Diabetes in pregnancy affects women in one of two ways: pregestational (which includes type 1 and type 2 diabetes) or gestational diabetes mellitus (GDM). By the 1950s, the word “gestational diabetes” had been coined to describe what was believed to be a temporary disease that had a negative impact on foetal outcomes and then disappeared after birth [1].
GDM, is carbohydrate resistance that starts or is first recognized throughout pregnancy [2]. This diagnosis does not extend to pregnant women who have already been diagnosed with diabetes before becoming pregnant. Gestational diabetes has a pathophysiology which is similar to that of type 2 diabetes. As the pregnancy progresses, tissue resistance to insulin increases, necessitating the use of more insulin [3]. The demand is easily met in the vast majority of pregnancies, so the balance between insulin resistance and insulin supply is maintained. Women, on the other hand, become hyperglycaemic if resistance becomes dominant. This typically happens in the second trimester of pregnancy, with insulin resistance steadily increasing before delivery, and usually disappears quickly after delivery [4]. In 2017, one in every seven births was diagnosed with GDM, according to the International Diabetes Federation (IDF) [5]. Globally, the prevalence of GDM is estimated to be about $15\%$, according to a systemic review [6]. GDM prevalence ranges from 2-6 percent in most racial/ethnic groups studied to, 10-20 percent in high-risk populations, with a growing trend across most racial/ethnic groups [7]. Since ethnicity has such a strong impact on GDM, the prevalence rate varies by race. The prevalence of GDM in Sub-Saharan *Africa is* around $14\%$ [8]. As at 2014 in Ghana, GDM was discovered in $10\%$ of pregnant women [9]. The prevalence of gestational diabetes in *Ghana is* increasing at a very fast rate. In 2004 and 2015, studies conducted in Ghana indicated a prevalence of $0.5\%$ and $9.3\%$ respectively [10, 11].
The prevalence of GDM is projected to continue to increase due to the rising obesity epidemic, delayed childbearing, and multiple pregnancies [12]. Gestational diabetes is known to have serious short and long-term effect on both the mother and foetus. GDM may cause pregnancy complications such as high blood pressure, heavy birth weight babies, and obstructed labour in the short term [11]. GDM tends to be a major risk factor for the development of type 2 diabetes (T2D) and cardiovascular diseases in women (13–15). Obesity and T2D are more likely to occur in children whose mothers had diabetes during pregnancy [16].
The development of gestational diabetes has been associated with several predisposing factors. These predisposing factors can be studied under lifestyle, obstetric, socio-demographic and clinical risk factors. Some of the obstetric risk factors associated with GDM includes previous abortion, parity and stillbirth [17]. Socio-demographic risk factors studied so far also include age, ethnicity and family history of diabetes [18, 19]. Clinical risk factors assessed are obesity and hypertension. Lifestyle risk factors like diet, physical activity, smoking and alcoholism are also known to have a link with GDM [19].
Despite the fact that urine dipstick has been accepted as the major screening criteria for GDM in Ghana, it is limited to renal threshold, hence most pregnant women who develop GDM later in their pregnancy are still excluded during regular screening programs for women. The risk factors of gestational diabetes can be used as a better diagnostic tool for early screening of women at risk hence it is necessary to identify the socio-demographic, obstetric, clinical and lifestyle risk factors of GDM. However, in Ghana and especially in greater Kumasi metropolis, there is no enough data on the risk factors associated with GDM. Therefore, this study is warranted.
## Study Design and Settings
This study is a hospital-based multi-centre cross-sectional study, conducted at the Antenatal Clinics of Kwame Nkrumah University of Science and Technology Hospital, Kumasi South Hospital and Saint Michael’s Hospital all in Kumasi Ghana, from April to July 2021.
## Study Population and Selection
A simple randomised sampling method was used to recruit a total of 200 pregnant women aged 16-45 years with gestational period between 16 and 40 weeks who were attending regular antenatal clinic at three selected hospitals in the Ashanti Region of Ghana namely, the Kwame Nkrumah University Hospital, Kumasi South Hospital and St. Michael’s Hospital. Ethical approval was sought from the Ethics Committee of the hospitals and The Committee on Human Research, Publication and Ethics, School of Medical Sciences, Kwame Nkrumah University of Science and Technology (CHRPE/AP/$\frac{204}{21}$). Thorough explanation of the study protocol and assurance of anonymity was made to the subjects. Written informed consent was also sought from participants and healthcare management prior to data and samples collection. Participants were first educated on the purpose of the study and only those who gave their consent to participate in the study were recruited. Diagnosis of GDMs was done by a Consultant/Specialist Obstetrician based on the criteria of the International Association of Diabetes and Pregnancy Study Groups (IADPSG), which uses a fasting blood glucose of ≥ 5.1 mmol/L [3]. Pregnant women who were already diagnosed of diabetes before pregnancy were excluded.
## Sample Size Justification
The sample size was obtained by the formula: n=z2p(1−p)e2, Where: Z is the standard normal variate at a confidence interval of $95\%$ = 1.96.
p is the prevalence = $10\%$ [10], e is the margin of error = 0.05.
Hence a minimum of one hundred and thirty-eighty participants were needed for the study A $95\%$ confidence level, $50\%$ response distribution, and $5\%$ margin of error at a statistical power of $80\%$ was employed in the calculation of the sample size. To increase statistical power, total of 200 participants were recruited for the study.
## Fasting Blood Glucose Measurement
A 3 millilitres of fasting blood samples were collected into fluoride tubes and centrifuged at 3000rpm at 5 minutes. Plasma was analysed for fasting blood glucose (FBG) levels using a fully automated clinical chemistry analyser (LE Scientific, China). Participants who gave their consent had their GDM status confirmed based on the IADPSG criteria using an FBG ≥ 5.1 and were asked to answer a standard questionnaire which provided information on their lifestyle, obstetrics, socio-demographics and clinical conditions.
## Questionnaire
A well-structured questionnaire was used which had four sections. The first section of the questionnaire was used to assess socio-demographics such as age, ethnicity, Occupational status, family history of diabetes level of education was also, marital status and previous use of oral contraceptives. The second section of the questionnaire assessed the obstetric risk factors associated with gestational diabetes mellitus. This section focused on factors such as history of abortion parity, previous perinatal outcomes, history of caesarean section and gravidity. The third section was used to assess the clinical risk factors of GDM such as obesity, history of hypertension, and history of intrauterine foetal death among others. The fourth section of the questionnaire was used to assess lifestyle risk factors such as physical activity, smoking alcoholism and diet.
## Anthropometric Measurements
Anthropometric measurements such as weight and height were taken to obtain body mass index (BMI) of participants using the weighing scale and stadiometer, respectively. BMI was calculated as weight in kilograms divided by the square of the height in meters. The BMI was classified into 4 categories in accordance with the WHO standard BMI criteria for adults. The categories into; underweight (BMI< 18.5 kg/m2), normal weight (BMI between 18.5 kg/m2 to 24.9 kg/m2), overweight (BMI: 25–29.9 kg/m2), and obese (BMI ≥ 30 kg/m2).
## Statistical Analysis
Collected data were entered in to Microsoft Excel 2016 and analysed using the Statistical Package for Social Sciences (SPSS) Version 23.0 (Chicago IL, USA) and Graph pad prism version 5.0 (Graph Pad software, San Diego California USA, www.graphpad.com).
A descriptive statistic was used to analyse the study variables. Continuous variables were represented by means (± standard deviations) whilst categorical variables by numbers (%).
A bar chart was used to illustrate the prevalence of gestational diabetes among study participants. Univariate logistic regression analysis was performed to screen for potential socio-demographic, obstetric, clinical and lifestyle risk factors associated with gestational diabetes. Multivariate logistic regression modal was used to determine independent risk factors of gestational diabetes.
A p-value of less than 0.05 ($p \leq 0.05$) and a confidence interval of $95\%$ were chosen as the statistical significance level and confidence interval, respectively.
## Sociodemographic and Clinical Characteristics of Study Participants
Table 1 shows the Sociodemographic and clinical characteristics of study participants. A total of 200 participants eligible for the study were included in the final statistical analysis. Of the total participants, one-third ($29.5\%$) were found to be widely distributed between age categories 16 to 25 years followed by 31 to 35 years ($28.5\%$). More than two-fifth of the participants had basic education ($44.5\%$), majority were married ($70.0\%$) and were Akans ($69.5\%$). In addition, two-third of study participants had informal occupation ($63.0\%$) with majority having no family history of hypertension ($87.5\%$) or abortion ($75.0\%$). Moreover, two-third of the participants had no record of previous oral contraceptives use ($63.5\%$), and majority did not have any pregnancy related disorder ($95.5\%$) and had not undergone any caesarean section ($83.0\%$). History on the participant’s previous obstetric risk factors shown that most of the pregnant women previously had termed and vertex delivery, live birth, no macrosomic baby and no previous history of GDM. This corresponds to ($65.5\%$), ($65.5\%$), ($93.5\%$), ($81.0\%$), ($96.5\%$) respectively (Table 1).
**Table 1**
| Variable | Frequency (n = 200) | Percentage (%) |
| --- | --- | --- |
| Age category (Years) | | |
| 16-25 | 59.0 | 29.5 |
| 26-30 | 54.0 | 27.0 |
| 31-35 | 57.0 | 28.5 |
| 36-45 | 30.0 | 15.0 |
| Educational level | | |
| No formal education | 8.0 | 4.0 |
| Basic education | 89.0 | 44.5 |
| Secondary education | 63.0 | 31.5 |
| Tertiary education | 40.0 | 20.0 |
| Marital status | | |
| Married | 140.0 | 70.0 |
| Cohabiting | 25.0 | 17.5 |
| Single | 35.0 | 17.5 |
| Ethnicity | | |
| Akan | 139.0 | 69.5 |
| Ewe | 3.0 | 1.5 |
| Ga-Adangbe | 6.0 | 3.0 |
| Northerners | 52.0 | 26.0 |
| Occupational status | | |
| Unemployed | 33.0 | 16.5 |
| Informal | 126.0 | 63.0 |
| Formal | 41.0 | 20.5 |
| Family history of hypertension | | |
| No | 175.0 | 87.5 |
| Yes | 25.0 | 12.5 |
| History of abortion | | |
| No | 150.0 | 75.0 |
| Yes | 50.0 | 25.0 |
| Previous use of oral contraceptive | | |
| No | 127.0 | 63.5 |
| Yes | 73.0 | 36.5 |
| Diagnosis of pregnancy related disorder (s) | | |
| No | 191.0 | 95.5 |
| Yes | 9.0 | 4.5 |
| Previous caesarean section | | |
| No | 166.0 | 83.0 |
| Yes | 34.0 | 17.0 |
| Previous delivery status | | |
| | 54.0 | 27.0 |
| Term | 131.0 | 65.5 |
| Preterm | 15.0 | 7.5 |
| Previous presentation at delivery | | |
| | 60.0 | 30.0 |
| Vertex | 131.0 | 65.5 |
| Breech | 9.0 | 4.5 |
| Previous Birth outcomes | | |
| Live birth | 187.0 | 93.5 |
| Stillbirth | 13.0 | 6.5 |
## Clinical and Lifestyle Factors
Table 2 shows the clinical and lifestyle characteristics of study participants. Majority of participants neither had preeclampsia ($78.5\%$) nor gestational hypertension ($93.0\%$). Clinically, two-fifth of the participants were overweight ($46.2\%$), majority with no history of intrauterine foetal death ($91.0\%$), threatened abortion ($95.5\%$) or infertility ($96.0\%$). Furthermore, the lifestyle risk factors assessed indicated more than half of the participants eat three times a day ($52.5\%$), take snacks in between meals ($56.6\%$), did not eat late night meals ($65.0\%$) but take in fruits and vegetables regularly ($81.5\%$). In addition, most of the participants exercise regularly ($54.0\%$), and had high preference for carbohydrate food ($61.0\%$) but did not take in neither soda drinks ($58.5\%$) nor fast food ($74.0\%$) (Table 2).
**Table 2**
| Variable | Frequency (n = 200) | Percentage (%) |
| --- | --- | --- |
| Previous macrosomic baby | | |
| No | 119.0 | 81.0 |
| Yes | 28.0 | 19.0 |
| Previous history of Preeclampsia | | |
| No | 157.0 | 78.5 |
| Yes | 43.0 | 21.5 |
| Previous history of Gestational hypertension | | |
| No | 186.0 | 93.0 |
| Yes | 14.0 | 7.0 |
| Previous history of GDM | | |
| No | 193.0 | 96.5 |
| Yes | 7.0 | 3.5 |
| History of intrauterine foetal death | | |
| No | 182.0 | 91.0 |
| Yes | 18.0 | 9.0 |
| Threatened abortion | | |
| No | 191.0 | 95.5 |
| Yes | 9.0 | 4.5 |
| History of infertility | | |
| No | 192.0 | 96.0 |
| Yes | 8.0 | 4.0 |
| Number of daily food intake | | |
| Once | 2.0 | 1.0 |
| Twice | 29.0 | 14.5 |
| Thrice | 105.0 | 52.5 |
| Four time | 45.0 | 22.5 |
| More Than Four Times | 19.0 | 9.5 |
| Snack in between meals | | |
| No | 87.0 | 43.5 |
| Yes | 113.0 | 56.6 |
| Late night meals | | |
| No | 130.0 | 65.0 |
| Yes | 70.0 | 35.0 |
| Regular intake of fruits and vegetables | | |
| No | 37.0 | 18.5 |
| Yes | 163.0 | 81.5 |
| Regular exercise | | |
| No | 92.0 | 46.0 |
| Yes | 108.0 | 54.0 |
| Number of exercises per week | | |
| Daily | 82.0 | 74.5 |
| Once | 4.0 | 3.6 |
| Twice | 13.0 | 11.8 |
| Thrice | 5.0 | 4.5 |
| More Than Four Times | 6.0 | 5.5 |
| Intake of soda drinks | | |
| No | 117.0 | 58.5 |
| Yes | 83.0 | 41.5 |
| Fast food intake | | |
| No | 148.0 | 74.0 |
| Yes | 52.0 | 26.0 |
| Preference for carbohydrate food | | |
| Low | 13.0 | 6.5 |
| Moderate | 65.0 | 32.5 |
| High | 122.0 | 61.0 |
| BMI status | | |
| Normal | 58.0 | 29.1 |
| Overweight | 92.0 | 46.2 |
| Obese | 49.0 | 24.6 |
## Prevalence of Gestational Diabetes Mellitus
As shown in Figure 1, of the 200 patients eligible for this study, 17 had gestational diabetes which indicated a prevalence of $8.5\%$ whereas 183 ($91.5\%$) were non-GDM patients (Figure 1).
**Figure 1:** *Prevalence of gestational diabetes mellitus (GDM) among study participants.*
## Risk Factors Associated With Gestational Diabetes Mellitus
Table 3 shows the risk factors associated with gestational diabetes mellitus. In the preliminary univariate analysis, the following study variables had $p \leq 0.05$ and were excluded from the final multivariate logistic regression analysis modal; educational level, ethnicity, occupational status, history of abortion, previous caesarean section, preeclampsia, previous history of gestational diabetes, history of intrauterine foetal death, history of infertility, number of times participants eat in a day, snack in between meals, late night meal, regular exercise, and fast food intake. However, after adjusting for possible confounders in the final multivariate logistic regression model, the use of oral contraceptives [aOR = 13.05, $95\%$CI = (1.43-119.23), $$p \leq 0.023$$] previous history of preeclampsia (aOR: 19.30; $95\%$ CI: 2.15-71.63; $$p \leq 0.013$$) and intake of soda drink [aOR =10.05, $95\%$CI = (1.19-84.73), $$p \leq 0.034$$] (Table 3).
**Table 3**
| Variable | cOR (95% CI) | p-Value | aOR (95% CI) | p-Value.1 |
| --- | --- | --- | --- | --- |
| Age category (Years) | | | | |
| 16-25 | Ref (1) | | Ref (1) | |
| 26-30 | 8.64 (1.03-72.71) | 0.047 | 2.96 (0.12-73.33) | 0.509 |
| 31-35 | 6.82 (0.80-58.59) | 0.08 | – | |
| 36-46 | 6.44 (0.64-64.85) | 0.114 | – | |
| Family history of hypertension | | | | |
| No | Ref (1) | | Ref (1) | |
| Yes | 3.396 (1.08-10.64) | 0.036 | 1.17 (0.10-14.37) | 0.903 |
| Use of oral contraceptive | | | | |
| No | Ref (1) | | Ref (1) | |
| Yes | 3.58 (1.26-10.13) | 0.016 | 13.05 (1.43-119.23) | 0.023 |
| Previous history of preeclampsia | | | | |
| No | Ref (1) | | Ref (1) | |
| Yes | 10.95 (2.62-45.79) | 0.001 | 19.30 (2.15-71.62) | 0.013 |
| Previous perinatal outcomes | | | | |
| | Ref (1) | | Ref (1) | |
| Live birth | 2.17 (0.46-10.26) | 0.327 | – | |
| Stillbirth | 16.56 (2.74-100.24) | 0.002 | 0.006 (0-0.51) | 0.025 |
| Previous macrosomic baby | | | | |
| No | Ref (1) | | Ref (1) | |
| Yes | 4.89 (1.69-14.18) | 0.003 | 3.38 (0.37-31.28) | 0.283 |
| Gestational hypertension | | | | |
| No | Ref (1) | | Ref (1) | |
| Yes | 5.32 (1.47-19.32) | 0.011 | 6.15 (0.34-110.06) | 0.217 |
| Threatened abortion | | | | |
| No | Ref (1) | | Ref (1) | |
| Yes | 18.65 (4.42-78.61) | 0.001 | 10.15 (0.53-196.42) | 0.125 |
| BMI | | | | |
| Normal | Ref (1) | | Ref (1) | |
| Overweight | 2.31 (0.46-11.50) | 0.308 | – | |
| Obese | 5.46 (1.10-27.09) | 0.038 | 8.35 (0.34-205.13) | 0.194 |
| Regular intake of fruits and vegetable | | | | |
| No | Ref (1) | | Ref (1) | |
| Yes | 0.21 (0.08-0.59) | 0.003 | 0.20 (0.02-1.61) | 0.13 |
| Intake of soda drinks | | | | |
| No | Ref (1) | | Ref (1) | |
| Yes | 7.71 (2.14-27.79) | 0.002 | 10.05 (1.19-84.73) | 0.034 |
## Discussions
Recently, the incidence of gestational diabetes mellitus (GDM) has increased globally and *Africa is* not an exception. This study therefore evaluated the prevalence and risk factors of GDM in Kumasi, Ghana. Our study observed a prevalence of $8.5\%$ gestational diabetes mellitus cases among study participants. Previous use of oral contraceptive, and intake of soda drinks as the independent risk factors of gestational diabetes mellitus.
The present study prevalence of $8.5\%$ gestational diabetes is in consistent with that of Anzaku et al. [ 20], who reported a prevalence of $8.3\%$ among Nigerians [20]. On the contrary, the observed prevalence of this study shows a decrease in the prevalence of gestational diabetes compared to that of Oppong et al. [ 10], who reported $10\%$ prevalence gestational diabetes among Ghanaians. The observed difference in prevalence may be attributed to the present study recruited participants from a clinic setting as opposed to Oppong et al. study conducted at a teaching hospital. Surprisingly, majority of study participants had normal BMI and exercise regularly which explain why we observed a decreased in prevalence of gestational diabetes as compare to previous study reported among Ghanaians.
In this current study, we observed previous use of oral contraceptive as independent risk factor of gestational diabetes. This is in consistent with a study by Kramer et al. [ 21], who reported the use of oral contraceptives is associated with gestational diabetes [21]. The agreement between the current and previous study provides evidence that hormonal contraceptive methods may increase a woman’s risk for GDM since most oral contraceptives are made up of oestrogen and progesterone which in excess induces hypercortisolism and therefore leading to insulin resistance and hyperglycaemia in pregnancy.
Our study observed that, patients with history of preeclampsia were 19-fold more likely to develop GDM. Lee et al. [ 22], found preeclampsia to be associated with GDM in subsequent pregnancy. This can be explained that the two disease conditions share a common pathophysiology and are characterized by systemic endothelial dysfunction [22].
Another finding of this study was that intake of soda drinks is independently associated with GDM. This finding confirms previous study by Donazar-Ezcurra et al. [ 23], who reported dietary intake of soda have a strong association with weight gain and metabolic syndrome. This could be explained as leading to spike in insulin which worsen insulin sensitivity overtime thereby enhancing insulin resistance leading to hyperglycaemia.
Previous studies have shown that lifestyle risk factors such as history of smoking, alcoholism and regular exercise are also highly associated with GDM [24]. On the contrary, this study did not show similar finding.
Even though the strength of the present study is that it is the first study to be reported in the metropolis, our use of a cross-sectional study design limits this study findings as the casual-effect relationship could not be established. Therefore, it is recommended that a cohort study design should be employed in subsequent studies to help assess more potential risk factors of GDM.
## Conclusion
This study evaluated the prevalence and risk factors associated with GDM in Kumasi, Ghana. We observed GDM prevalence of $8.5\%$ among study participants, which was influenced by previous history of oral contraceptives use, history of preeclampsia and consumption of soda drink *It is* recommended that pregnant women be educated on lifestyle modification and the need to reduce consumption of soda drinks and the use oral contraceptives.
## Data Availability Statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics Statement
The studies involving human participants were reviewed and approved by The Committee on Human Research, Publication and Ethics, School of Medical Sciences, Kwame Nkrumah University of Science and Technology. The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
EOA, WB, and PK. Methodology: EOA, WB, and PK. Formal analysis, PK, ES, and SO. Investigation, WB, EOA, andPK. Original draft preparation, EOA, PK, SO, and ES. Supervision,EOA and WB. All authors listed reviewed, edited have made a substantial, direct, and intellectual contribution to the work and approved it for publication. All authors contributed to the article and approved the submitted version.
## Funding
This study was funded by TiDi Foundation, a non-governmental funding body which supports student research with a seed fund of 140 dollars.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Buchanan TA, Xiang AH. **Gestational Diabetes Mellitus**. *J Clin Invest* (2005) **115**. DOI: 10.1172/JCI200524531
2. Association AD. **Gestational Diabetes Mellitus**. *Diabetes Care* (2004) **27** S88. DOI: 10.2337/diacare.27.2007.S88
3. Association AD. **Diagnosis and Classification of Diabetes Mellitus**. *Diabetes Care* (2014) **37**. DOI: 10.2337/dc14-S081
4. Ben-Haroush A, Yogev Y, Hod M. **Epidemiology of Gestational Diabetes Mellitus and Its Association With Type 2 Diabetes**. *Diabetic Med* (2004) **21**. DOI: 10.1046/j.1464-5491.2003.00985.x
5. Ogurtsova K, da Rocha Fernandes J, Huang Y, Linnenkamp U, Guariguata L, Cho NH. **IDF Diabetes Atlas: Global Estimates for the Prevalence of Diabetes for 2015 and 2040**. *Diabetes Res Clin Pract* (2017) **128** 40-50. DOI: 10.1016/j.diabres.2017.03.024
6. Mensah GP, van Rooyen DR, ten Ham-Baloyi W. **Nursing Management of Gestational Diabetes Mellitus in Ghana: Perspectives of Nurse-Midwives and Women**. *Midwifery* (2019) **71** 19-26. DOI: 10.1016/j.midw.2019.01.002
7. Galtier F. **Definition, Epidemiology, Risk Factors**. *Diabetes Metab* (2010) **36**. DOI: 10.1016/j.diabet.2010.11.014
8. Mwanri AW, Kinabo J, Ramaiya K, Feskens EJ. **Gestational Diabetes Mellitus in Sub-Saharan Africa: Systematic Review and Metaregression on Prevalence and Risk Factors**. *Trop Med Int Health* (2015) **20** 983-1002. DOI: 10.1111/tmi.12521
9. Asare-Anane H, Atbek O, Amanquah SD. **Risk Factors for Gestational Diabetes Mellitus Among Ghanaian Women at the Korle-Bu Teaching Hospital**. *Risk* (2014) **4**
10. Oppong SA, Ntumy MY, Amoakoh-Coleman M, Ogum-Alangea D, Modey-Amoah E. **Gestational Diabetes Mellitus Among Women Attending Prenatal Care at Korle-Bu Teaching Hospital, Accra, Ghana**. *Int J Gynecol Obstet* (2015) **131**. DOI: 10.1016/j.ijgo.2015.05.029
11. Quansah DY, Boateng D, Kwantwi LB. **Overview of Gestational Diabetes Mellitus (GDM) in Ghana; a Call for Action**. *Diabetes Care* (2018) **27** 88-90. DOI: 10.19080/CRDOJ.2018.06.555694
12. Durnwald C. **Gestational Diabetes: Linking Epidemiology, Excessive Gestational Weight Gain, Adverse Pregnancy Outcomes, and Future Metabolic Syndrome**. *Semin Perinatol* (2015) **39**. DOI: 10.1053/j.semperi.2015.05.002
13. Kaaja R, Rönnemaa T. **Gestational Diabetes: Pathogenesis and Consequences to Mother and Offspring**. *Rev Diabet Stud RDS* (2008) **5** 194. DOI: 10.1900/RDS.2008.5.194
14. Rautio N, Jokelainen J, Korpi-Hyövälti E, Oksa H, Saaristo T, Peltonen M. **Lifestyle Intervention in Prevention of Type 2 Diabetes in Women With a History of Gestational Diabetes Mellitus: One-Year Results of the FIN-D2D Project**. *J Womens Health* (2014) **23**. DOI: 10.1089/jwh.2013.4520
15. Vounzoulaki E, Khunti K, Abner SC, Tan BK, Davies MJ, Gillies CL. **Progression to Type 2 Diabetes in Women With a Known History of Gestational Diabetes: Systematic Review and Meta-Analysis**. *BMJ* (2020) **369** 1-10. DOI: 10.1136/bmj.m1361
16. Ornoy A. **Prenatal Origin of Obesity and Their Complications: Gestational Diabetes, Maternal Overweight and the Paradoxical Effects of Fetal Growth Restriction and Macrosomia**. *Reprod Toxicol* (2011) **32**. DOI: 10.1016/j.reprotox.2011.05.002
17. Lee KW, Ching SM, Ramachandran V, Yee A, Hoo FK, Chia YC. **Prevalence and Risk Factors of Gestational Diabetes Mellitus in Asia: A Systematic Review and Meta-Analysis**. *BMC Pregnancy Childbirth* (2018) **18** 1-20. DOI: 10.1186/s12884-018-2131-4
18. Khan R, Ali K, Khan Z. **Socio-Demographic Risk Factors of Gestational Diabetes Mellitus**. *Pakistan J Med Sci* (2013) **29** 843. DOI: 10.12669/pjms.293.3629
19. Carroll X, Liang X, Zhang W, Zhang W, Liu G, Turner N. **Socioeconomic, Environmental and Lifestyle Factors Associated With Gestational Diabetes Mellitus: A Matched Case-Control Study in Beijing, China**. *Sci Rep* (2018) **8** 1-10. DOI: 10.1038/s41598-018-26412-6
20. Anzaku AS, Musa J. **Prevalence and Associated Risk Factors for Gestational Diabetes in Jos, North-Central, Nigeria**. *Arch Gynecol Obstet* (2013) **287**. DOI: 10.1007/s00404-012-2649-z
21. Kramer BA, Kintzel J, Garikapaty V. **Peer Reviewed: Association Between Contraceptive Use and Gestational Diabetes: Missouri Pregnancy Risk Assessment Monitoring System, 2007–2008**. *Prev Chronic Dis* (2014) **11** 1-6. DOI: 10.5888/pcd11.140059
22. Lee J, Ouh Y-T, Ahn KH, Hong SC, Oh M-J, Kim H-J. **Preeclampsia: A Risk Factor for Gestational Diabetes Mellitus in Subsequent Pregnancy**. *PloS One* (2017) **12** e0178150. DOI: 10.1371/journal.pone.0178150
23. Donazar-Ezcurra M, Lopez-del Burgo C, Martinez-Gonzalez MA, Basterra-Gortari FJ, de Irala J, Bes-Rastrollo M. **Soft Drink Consumption and Gestational Diabetes Risk in the SUN Project**. *Clin Nutr* (2018) **37**. DOI: 10.1016/j.clnu.2017.02.005
24. McIntyre HD, Catalano P, Zhang C, Desoye G, Mathiesen ER, Damm P. **Gestational Diabetes Mellitus**. *Nat Rev Dis Primers* (2019) **5** 1-19. DOI: 10.1038/s41572-019-0098-8
|
---
title: 'Playful Communication and Care: Exploring Child-Centred Care of Young Children
With Type 1 Diabetes Through the Framework of Zone of Proximal Development'
authors:
- Patricia DeCosta
- Dan Grabowski
- Louise Norman Jespersen
- Timothy C. Skinner
journal: Frontiers in Clinical Diabetes and Healthcare
year: 2022
pmcid: PMC10012162
doi: 10.3389/fcdhc.2021.707553
license: CC BY 4.0
---
# Playful Communication and Care: Exploring Child-Centred Care of Young Children With Type 1 Diabetes Through the Framework of Zone of Proximal Development
## Abstract
### Introduction
Little is known about the psychosocial experiences and care needs of young children under the age of 7 years who have been diagnosed with type 1 diabetes. To address this knowledge gap, we examine children’s psychosocial care needs through the lens of child-centred care and the framework of Zone of Proximal Development.
### Objectives
To explore current care practices for young children with diabetes and identify aspects of child-centred care already successfully integrated into current practice.
### Method
Individual face-to-face, semi-structured interviews were conducted with 20 Healthcare Professionals, representing 11 of 17 paediatric diabetes clinics in Denmark.
### Results
Our data provided valuable insights into existing child-centred practices. Our analysis identified practices covering four main themes: 1. Accommodating immediate emotional needs, 2. Putting children before diabetes, 3. Encouraging meaningful participation, 4. Playful communication.
### Discussion
Healthcare Professionals provided child-centred care, largely through play-based approaches that make diabetes care meaningful and relevant. Such practices provide the scaffolding necessary to enable young children to gradually engage, comprehend and participate in their own care.
## Introduction
Type 1 diabetes is one of the most common chronic diseases diagnosed in childhood [1]. Over the past 20 years, the number of children diagnosed with type 1 diabetes has been increasing [2], with the greatest increase observed in children under 5 years of age [3]. However, little is known about the psychosocial needs and experiences of young children under the age of 7 years who have been diagnosed with type 1 diabetes [4]. The lack of knowledge regarding this age group is reflected in The Position Statement from the American Diabetes Association, on psychosocial care for people with diabetes, which does not address children under the age of 7 years [5].
In the adult as well as child population, the importance of being sensitive to the broader psychosocial dimensions of living with a chronic illness has gained increased recognition [6]. Accordingly, providing patient-centred care (PCC) has been a central objective in improving healthcare, as suggested by the US Institute of Medicine [7]. While a significant amount of research on PCC currently exists, including systematic reviews, showing that PCC processes are positively correlated with patient satisfaction, well-being and clinical outcomes [6], the literature on child-centred care (CCC) in healthcare is still relatively new [8]. Likewise, CCC is not mentioned in the Consensus Guidelines for psychological care of children and adolescents with type 1 diabetes [9]. However, viewing the psychological care of children through a child-centred lens is needed to ensure that guidelines represent both a child perspective and the child’s insider perspective [10].
CCC encompass being responsive to children’s views and preferences, eliciting the child’s perspectives, tailoring care to the individual child’s needs, building relationships and providing children with the time and opportunity for participation [11]. Thus, adapting a CCC approach in healthcare reflects a wider acknowledgement of children’s rights as well as their right to participate and be involved in their own healthcare [10]. Research suggests that children favour CCC, although this need may not currently be accommodated [12]. Consistently, a recent review, consolidating the views of at least 650 (mainly older) children with type 1 diabetes, found that children preferred individualised, collaborative, relationship-based diabetes care [13].
Accordingly, in the present study, we will address the gap in knowledge regarding young children’s (age <7 years) psychosocial care needs and do so through the lens of CCC. Specifically, we will accomplish this by exploring the knowledge and practical experience of healthcare professionals (HCPs) working with these young children.
Our objective is to explore current care practices for young children with type 1 diabetes and identify aspects of CCC already successfully integrated into current practice.
## Theoretical Framework
To ensure that our objectives are explored in an appropriate developmental context, we use Vygotsky’s theory of Zone of Proximal Development (ZPD) [14] and the concept of scaffolding [15] to frame our analysis. The ZPD can be described as ‘the distance between the actual developmental level as determined by independent problem solving and the level of potential development as determined through problem solving under adult guidance or in collaboration with more capable peers’ [14]. Scaffolding refers to the activities and tools provided by an adult or capable peer that serve to support the child as he or she is led through the ZPD [15]. The theory of ZPD acknowledges that development is intrinsically linked to learning, as opposed to the notion that learning can only take place when a certain developmental stage has been reached. From the ZPD perspective, even young children have the ability to understand, contribute to and participate in their own care, provided that interactions take place within the ZPD and are supported by the use of age-appropriate tools and activities.
In the context of young children, play is a fundamental part of the ZPD. Young children communicate, express emotion and comprehend information through play: ‘Play creates a zone of proximal development of the child. In play, the child always behaves beyond his average age, above his daily behaviour; in play, it is as if he is a head taller than himself. As in the focus of a magnifying glass, play contains all developmental tendencies in a condensed form and is itself a major source of development’ [14]. Play is widely used in therapeutic settings to strengthen the connection and positive interaction between young children and their caregiver as well as to examine young children’s attachment and relationships [16, 17]. Accordingly, we use the theory of ZPD, the concept of scaffolding and play as a framework both to identify CCC practices and to understand the mechanisms that make these practices appropriate from a developmental perspective.
## Method
HCPs from all paediatric diabetes clinics in Denmark ($$n = 17$$) were invited to participate in this qualitative study. A total number of 20 HCPs, representing 11 clinics, accepted the invitation. One clinic declined, while two clinics did not respond. Another three clinics initially accepted the invitation. However, they did not respond to further attempts to arrange the interview. Individual face-to-face interviews were conducted with 11 paediatric nurses, four paediatricians, three psychologists and two dietitians. The HCPs experience in paediatrics ranged from 10 to 40 years (median 19 years). All interviews took place at the home clinic of the respective HCP and lasted between 23-50 minutes. We used an open-ended, semi-structured interview guide, created with the objective to study current care practices, explore knowledge and experiences of HCPs and gain insights into the needs and experiences of young children with type 1 diabetes. The guide included 18 open-ended questions concerning young children’s psychosocial needs, experiences, challenges and strengths, as well as current care practices. E.g., how do you involve young children and elicit their perspectives at diagnosis and in subsequent consultations? *Have this* practice changed over time?
Prior to data collection, the first author (PD) passively observed a number of paediatric diabetes consultations at a clinic in Denmark. While these observations were not used in the analysis, they provided a contextual background for the subsequent interviews. All interviews were conducted by PD. The data were transcribed verbatim by PD and a student assistant. The study took place between December 2018 and January 2020.
## Ethics
The study was approved by The Institutional Ethical Review Board, University of Copenhagen, Department of Psychology. Approval number: IP-IRB/14112018.
## Analysis
In accordance with Braun and Clarke [18], we used thematic analysis to identify and report patterns – or themes – in our interview data. A preliminary analysis of the data was conducted in Phase 1 by listening to the interviews and rereading the written transcripts. In Phase 2, we transferred the transcribed dataset into the qualitative data analysis program Nvivo for detailed coding. At this point, analysis was narrowed to identify CCC practices, as described by the HCPs. Specifically, we gathered all data in which HCPs described practices that not only fulfilled young children’s medical needs, but also supported their psychosocial needs. In Phase 3, we searched for potential themes by gathering and grouping codes, as well as pulling together all data relevant to each potential theme. In Phase 4, we refocused the analysis to the broader level of themes, and we constructed a thematic map. In Phase 5, themes were refined, defined and ultimately named. All authors (PD), (LNJ), (DG) and (TS) took part in the discussion throughout the analysis, and consensus regarding final themes was reached jointly.
## Results
Our data revealed significant diversity in practices, views and beliefs concerning young children’s need and ability to participate in their own care. Some HCPs found that cognitive and language abilities were a barrier to eliciting young children’s views in a meaningful way. Others found that participation was unnecessary, as the responsibility for diabetes care rested fully with the parents. A few HCPs simply did not put much emphasis on young children’s psychosocial needs or experiences, viewing diabetes care in this particular age group as principally biomedical.
‘And in that way, it may be easy, because when we typically have 2- to 3-year-olds, or a 4-year-old for that matter too, then of course it can be about us having to talk to them about that; “now you need this here” and “I know that it’s not that fun” and stuff like that. But if you put it a little more bluntly, then it is a little like ‘a body’ that has to line up for this. But it is the parents who really have to do it, and it is the parents you talk to, if that makes sense?’ – Paediatrician [10].
At the same time, others provided great insight into how to elicit young children’s perspectives and actively engage children in their own care. They described a coordinated strategy for being sensitive to and meeting young children’s needs, providing numerous examples of specific child-centred practices. The data describing such practices are presented here.
We identified four main themes in our analysis: accommodating immediate emotional needs, putting children before diabetes, encouraging meaningful participation and playful communication.
## Accommodating Immediate Emotional Needs
At diagnosis, the HCPs found that young children reacted to the immediate pain of blood tests and insulin injections, to the cues of the emotional state of their parents, as well as to being surrounded by strangers in an unfamiliar environment. The HCPs explained that the immediate need of young children, in addition to medical treatment, was to feel safe and become familiar with their new surroundings and the people treating them; a prerequisite for learning and development to occur. For the majority of children admitted to hospital, medical intervention could safely be delayed, or kept to a minimum, in order to accommodate the psychological needs of young children and their families.
‘Their greatest need, here and now, is that it should be safe and calm, and not dangerous. That is, when they are admitted. Let’s do a minimal intervention on them. Something good will come out of it. And something good will come out of that in the long run, as well’. – Paediatrician [19].
For young children in particular, pricks, jabs and injections were a major source of fear, anxiety and distress immediately following diagnosis. The HCPs with many years of experience could describe how an increased focus on children’s psychological experiences had changed practices over time. Now, whenever possible, they would increase the time between admittance and treatment in order to minimize the negative consequences associated with the experience of pain.
‘I think it’s a huge advantage that we don’t have to prick the kids as much anymore. And I also think that the general attitude has become, -in the past when a child with newly diagnosed diabetes came, we immediately swarmed them […] there was not much awareness of that, all these many jabs, it actually wasn’t good. Where now, we are much more hesitant, which we can be, because they often come to us well before they are in ketoacidosis. So we try to hold back a little bit with all these jabs. So there may well be a diabetes child who comes to us, where our assessment is, that the situation is such that we can easily put an Emla [anaesthetic patch] on and wait that hour to one and a half hours, until it has worked, and the lab technicians can come, before we even do anything at all. Um, I think that’s a major advantage and a major change’. – Nurse [1].
In response to children’s need to feel safe, the HCPs tried to create a framework of predictability and continuity, from the time of admission through transferring to outpatient clinic appointments. They would explain, or in the case of young children demonstrate, what was going to happen. They acknowledge that without a framework and age-appropriate explanations, young children had little understanding of the new situation in which they found themselves.
‘Yes, so I think they need some, things like a sense of trust, a feeling of safety, and a framework so that they know what’s going on in here, but also that they know what’s going to happen, and so on, and that things are explained … And time, right; that we’ve got the time for them too’. – Nurse [16].
In the acute phase of diagnosis, where learning and development was not the main objective, care practices operating within young children’s ZPD, still proved valuable. By demonstrating procedures, rather than relying on solely verbal communication, HCPs actively assisted children’s understanding with the intention of providing a predictable framework and a sense of safety.
One HCP described how their clinic had a coordinated strategy for diabetes education to be delivered consistently by the same nurse in the team, from admission onwards. This approach differed from that used at other clinics, where initial diabetes education was delivered by the HCPs on the ward. For caregivers, this approached insured that consistent information was given at all time. Importantly, for young children, it meant being able to build relationships with the people treating them.
‘We actually see them quite frequently during the hospitalisation, and it provides a really good foundation for the future collaboration we have, when we will see them a few times a year, when they come to our outpatient clinic. So that beginning, it’s a good investment’. – Nurse [16].
## Putting Children Before Diabetes
By anchoring social interaction and learning within the context of children’s everyday life, HCPs were more likely to engage young children in the ZPD and ensure child-centered practices. The HCPs emphasised the importance of seeing and acknowledging the whole child first, before putting any focus on diabetes. Accordingly, they viewed diabetes in the context of children’s everyday lives. The rational for and proposed benefits of this approach were manifold.
‘We would like them to understand that it is them as a person who is exciting and interesting and funny and that we are curious about. […] If they can come and tell me, like all other children, what they’re doing and what they did yesterday, or what they are going to do today, without talking about diabetes. Then it’s a sign that things are going well’. – Paediatrician [19].
In this way, the HCPs wished to convey the message that, even in the hospital or clinic setting, diabetes is only one part of the child’s life – and not the most important part. At the same time, the children’s inclination to talk about their everyday life revealed important information about their experience of living with diabetes, as well as giving an initial indication of adjustment.
‘It’s about the person being important. And that diabetes is, in quotation marks, “is just the circumstance”. It’s something you have to carry, and that we have to work with, but it’s not what’s ultimately important. It’s the person you are, and who you are today and tomorrow and in the future. And the wishes and dreams that you have, that’s what’s important. Because somehow, it’s not just about blood sugar, or it’s actually not about that it at all, it’s about quality of life, right’. – Paediatrician [19].
Putting diabetes in the context of young children’s everyday life and showing interest in their life outside the clinic mattered to even young children. The HCPs found that this practice laid the foundation for building relationships and trust. In this way, such practice was consciously integrated into the clinic appointment.
‘In fact, it’s important to us; if the child has told us something –it could be that he was going on an overnight fieldtrip with the kindergarten, some still do that, that we then note in their medical records. Because when they return next time, then we remember to ask about it; how did it go? So that they will go – Ok! They actually know me! It can be difficult for us to remember all the kids, so that’s why we make sure to write it down. So they get a sense that – they actually know me well. […] And I think I establish a good relationship with the families. Therefore, I think if there is something that they find really difficult, then they also come to me and tell me about it’. – Nurse [12].
Additionally, showing an interest in the whole child and what was important to that individual was viewed in relation to the long-term perspective of these children’s continued contact with the HCPs and the clinic. Although the interaction seldom centred on diabetes, but rather on what was meaningful to young children in their everyday life, it entailed important learning about how to actively engage and interact with HCPs in the consultation – an important skill for children with a chronic illness.
‘We also do it with the purpose to teach them to speak for themselves in the consultation. Because at some point when they get a little older, then they’re used to talking to us. And then it might also be easier to bring up other issues, things that they’re struggling with, in relation to their diabetes’. – Nurse [12]. This practice illustrate how HCPs, who operate within young children’s immediate cognitive and psychological range, essentially utilize the ZPD. When HCPs encourage young children to express their understandings, likes and feelings concerning their everyday life, they guide and assist these children’s future ability to express their understandings, needs and feelings, in relation to diabetes management.
Integrating such practices as well as ensuring early and active engagement required active consideration and planning on the part of the diabetes team. The HCPs were aware that technology, and the instant data provided by it, could potentially direct attention away from the child sitting there in front of them.
‘So it takes a lot of time to get all these practical things done around the equipment itself. Then I also think that the more information you get from the pump and censor, and we read it, the more we become focused on this screen and the more focus moves away from the child. So we have to be really conscious to say: Well okay, it’s also important that we have time to look at the screen and adjust what needs to be adjusted, but it must not be at the expense of us forgetting the child and forgetting the family, and we forget to hear some of these soft values in the consultation as well. It’s all just very hard facts once you get things up on the screen with numbers’. – Dietitian [6].
The time invested in getting to know young children also gave the HCPs an insight into both the individual child’s emotional state and how such emotions were expressed. In turn, the HCPs could be responsive to the needs and preferences of even young children who lack the ability to fully express their experiences, verbally or nonverbally.
‘When I start to get to know the children, I know what works, that is, from child to child. So I know that, well there are some who may be quite hyper [behaviour] when they come. They don’t quite know, - I then know that we need some kind of diversion. There’s one boy for example, where I know that when he comes, he just needs to be allowed to say a lot of things, and sometimes there are some swear words and stuff like that. […] I think it’s his way of saying, that he thinks it’s a little annoying, that he has to come here’. – Nurse [13].
## Encouraging Meaningful Participation
There was consensus among the HCPs that young children quickly lost interest and got bored when adult attention was focused on the computer screen or exclusively ‘adult talk’. Evaluating data from the insulin pump or glucose meter is a fundamental part of the consultation. However, without modification, this activity is outside young children’s ZPD and does not offer them any real opportunity for participation. Children would show signs of boredom by becoming fidgety or restless. Because of this, some HCPs explained that they would make a clear distinction between the activities they sought to actively engage the children in and the activities clearly aimed at the caregivers.
‘So I often make a point of saying; you know what, now we’re actually done with this, so feel free to sit down with your iPad, or feel free to go out and play in the waiting room, or whatever you fancy. […] Because it often is the boring stuff, right? And it will be things they don’t understand. So I think it’s important that when we want their attention, it’s on activities they can participate in and find interesting. Also to keep up their spirits for the next time they have to come in for a consultation’. – Nurse [12].
In this way, the HCPs were responsive to young children’s preferences and only asked for their attention when they had a real opportunity to engage. They acknowledged that, if the activity did not actively include or engage the child, young children’s attention span was very limited. In addition, the HCPs emphasised that one prerequisite for making consultations inclusive or fun was to create a positive association with coming to the clinic.
‘I really want them to leave here, thinking that it was ok to be in the hospital. Because, of course I know, that sometimes they must have blood samples done as well. Luckily, it’s not here with me, it’s over at the lab, right. So it’s not always fun to come to the hospital, but it must be ok when they come into our [consultation] room, here with us. It needs to be a good place to be’. – Paediatrician [19].
The HCPs found that the education material they had available was largely not suitable for young children. Although the material included pictures and illustrations, it failed to capture young children’s attention in a way that was meaningful to them.
‘For preschool children, it can be difficult to show these pictures of the apple going down into the stomach and the pancreas situated there. […] So we try, as early as possible, to involve them in, in a very, very simple way, that there is some relationship between eating and what one’s blood sugar levels are’. – Nurse [1].
According to the HCPs, young children could still be actively included in education and treatment using other approaches. One HCP pointed out that all children had a wish and need to be involved in what was happening to them. The only question was how to do this appropriately.
‘They do really have a need to be involved. But involved where we have - that is, where it has been well reflected on; what are they to be involved in? And at what level? And at what point? […] Because they mirror what they see, and they snap up everything. That nonverbal communication, they pick it up straight away, right. So it’s really important, that when we plan our teaching and plan new initiatives, that we then also consider: How are we going to present this to the child? Yes, in my experience at least, I often find that the results are best if we’ve considered; how do we get the child involved? And in what way?’ – Nurse [13].
Many HCPs reported that fear in children was often associated with a lack of understanding of what was going to happen to them. A number of HCPs found it useful to demonstrate medical procedures on teddy bears, or even on themselves, prior to the procedure being done on the child. This practice resonated with young children, broadening their understanding and, thus, minimising fear.
‘I also often use the needle on myself, it sounds so dramatic. But if I have to do something to these children, where they’re having a sensor put on, or a pump or something, then I demonstrate it on myself. So that they can see what my face looks like when I put something into my body, that they’re about to have’. – Nurse [3].
While this approach did not always result in children happily complying with medical treatment, the HCPs found it helpful in that young children would engage in this type of interaction. Additionally, medical role-play was helpful when reconnecting with a child following an unpleasant or painful procedure. This also helped the child ‘play out’, or work through, the difficult experience as a means to move past the event.
‘And then I had to be cut on my leg with the, -because there was this knife in that doctor’s set, right? Then I had to have a bandage on. So I kind of had to go through-, I had to go through the same thing as him, right? And then we played a little bit, so that now it was his turn to kind of be the one to do things to me, right? And then it was ok. Then he was-, then he got over it quickly’. – Nurse [3].
The HCPs explained that getting young children to actively participate in the preparation of a medical procedure was a tangible way of including them in their own treatment as well as increasing their comprehension and feeling of agency. It was a way to make diabetes education meaningful and to encourage age-appropriate participation in their own care.
‘Well, it’s about, that you have to go down to the child’s level and explain the things that need to be explained, right. And why we do these things. And again, we need to have the parents on board too, because, so that the child feels more secure about the things that need to be done. We try to explain what needs to be done, nice and calmly. We’re trying to get them involved in preparing the blood glucose meter for example, put the strip in’. – Nurse [15].
While all clinics had a psychologist on the diabetes team, only a few worked directly with young children. Most would see the caregivers instead. However, psychologists who included young children in their consultation found that they added valuable insights and that children benefitted from ‘having a say’ in their own treatment.
According to the psychologists, given the right framework and tools, even young children could relate their feelings and express preferences for the support offered by their caregivers in difficult situations.
‘Yes, we draw a lot and maybe use the board, they also draw a bit on the board. We have some toys with us, but it’s not something I systematically include in the conversation, it’s a bit dependent on what it is. We have some Duplo with different emotions, so it may well be that we use them to express whether one felt sad or happy or whatever it was, in that age group’. – Psychologist [4].
‘There are also a lot of children who, for example, don’t like needles, so you work with that. We then work with drawings, and how do we get a handle on the fact that you have to be pricked or measured. So a lot is with drawings. We can also do it with play, with dolls and dollhouses, it depends on what age the child is […] Yes, and how do we help the child cope with pricks or injections next time. How to draw it, and who is in control? The fact that the child himself gets the experience that it is me who can accomplish this. […] so you know, participation within the boundaries is quite important –even for a very young child’. – Psychologist [7].
## Playful Communication
The HCPs used a combination of play, humour and positive feedback when interacting with young children in a way that was meaningful to them. Accordingly, our fourth theme describes how the HCPs used play-based approaches across all themes, whether the aim was to meet the immediate emotional needs of young children or to encourage meaningful participation. Some HCPs would get on the floor and play with cars, others would sit on children’s furniture or join young children in the playroom or designated play area. The common thread here was that they sought to meet young children in their own domain, by stepping into their world. Hence, the data revealed that the HCPs used play-based approaches to connect and interact with young children, within the ZPD, thereby making activities meaningful and relevant.
Additionally, the HCPs found that young children were very responsive to positive feedback, mainly in the form of praise, but tangible rewards such as small toys were also greatly appreciated by young children.
‘Many of them have this scanner, which they scan to get a tissue sugar and then we can ask them; how do you do it? And, can I see? and what do you do and –so we get something hands on, and they also get the opportunity to show us all the things they can do. In my opinion, that works really well and it’s completely undramatic and not scary. It can only be positive, because they can only receive praise. You can’t do anything wrong’. – Paediatrician [19].
Small toys were used as an incentive to comply with medical procedures, thus avoiding any use of force; they were also used to ensure that young children associated the clinic with something positive. The HCPs used positive feedback to reward and encourage engagement and participation. However, it was never contingent on an outcome.
‘They learn so fast, even when they’re quite young, in my experience, whether a blood sugar is good or bad, that is. And where I say, after all, they shouldn’t call it good or bad. Every blood sugar measured, I usually say to them, is a good blood sugar! But there are some we need to do something about, there are some that are higher, and we need to do something about them’. – Nurse [12].
The HCPs used a combination of play and humour to distract children before or during potentially painful procedures. They found that distraction, in combination with appropriate pain relief, could decrease fear and anxiety and defuse tension, for both the young children and their caregiver.
‘Diversions can be helpful […] I have this funny pen that can, there’s a kind of “TV” on, you can press a button and then Mickey appears. Then we can talk a little bit about it and maybe get a little preoccupied with it, right, and then they barely notice it. Because that “magic patch” actually works really well. Because often it’s not the pain, it’s the fear that they don’t know if it will hurt, and the thing about; if they have to be restrained. What is happening? They can also sense that mum is getting nervous and dad is nervous, right, and that they want this over and done with’. – Nurse [16].
The HCPs described how creating a playful and fun atmosphere was of great importance to young children’s experiences, as it counteracted fear and anxiety and could alter the perception of pain associated with treatment. For this purpose, the HCPs found that the assistance of hospital clowns (Hospital Clowns (Danske Hospitalsklovne) is a privately founded charity that supports hospitalised children and their families) was very helpful.
‘That you can get some things through, some procedure through without the cost being too high for the kids, regarding pain and suffering, in my opinion at least. […] I have some procedures, some things that are painful but that can be accomplished, with the help of [hospital] clowns. And I think that’s good. […] Yes, or get them laughing or something. Getting them in a good mood and stuff like that, I think’. – Paediatrician [18].
## Discussion
Analysing our data in a developmental perspective using the theory of ZPD proved valuable in understanding the care needs of young children diagnosed with type 1 diabetes. It also provided insights into why CCC matters even for young children with type 1 diabetes, and what this care may look like in practice.
The HCPs provided CCC, largely through play-based approaches and starting from the children’s tangible experiences and understandings, in this way providing the necessary scaffolding that is essential to young children’s ability to engage, comprehend and participate in their own care. Our findings are consistent with research indicating that even young children’s levels of understanding, knowledge, and skill gained from their experience of living with diabetes develop through experience rather than as a function of age [19].
The usefulness of ZPD in the area of nursing interventions for children has previously been proposed [20] and may be particularly pertinent when caring for young children. We found that operating within the ZPD, using appropriate scaffolding and playful communication were at the core of being responsive to the needs and preferences of young children, establishing therapeutic relationships, eliciting the child’s perspectives and encouraging participation.
As young children’s views and competencies in relation to diabetes care are largely missing from the literature [4, 13, 19], it is difficult to compare our result with findings from other studies. However, our findings suggest that young children, like older children and adults, have a need for and benefit greatly from CCC practices.
While the present study illustrates that HCPs can successfully accommodate young children’s preference for relationship-based care and enable young children to actively participate in their care, other research suggests that children (4-10 years of age) with type 1 diabetes do not experience the clinic environment as one in which they can usefully contribute [21]. This indicates that current practices may not adequately demonstrate a belief in the value and validity of children’s views or support children’s active participation [21]. Likewise, our results highlighted a disparity between clinics where HCPs were responsive to young children’s needs and preferences and clinics that waited for children to grow older before actively considering their need to participate.
This inconsistency in practices may reflect a lack of research or, more likely, a failure to consider the child’s perspective in the research and, thus, in guidelines and recommendations. While the present study represents a child perspective, it does not include the child’s perspective, which embodies the child’s insider perspective on experiences, perceptions and actions, based on what the child deems important [10]. Both perspectives are necessary if we are to better understand the value of CCC and should be given due consideration in guidelines and recommendations. Indeed, if we are to live up to the UN Convention on the Rights of the Child, at a minimum, children should be listened to, supported in expressing their views and have their views taken into account [22].
Research indicates that HCPs find it difficult to meet children’s preferences for collaborative, relationship-based care [13]. Therefore, identifying and describing existing, tangible practices that have been successfully integrated into clinics may be particularly useful for diabetes teams and HCPs working with young children. Our results showed that current educational material was primarily aimed at older children. Besides age-appropriate educational tools, tools that can aid communication between young children and HCPs are needed. Developing such tools in cooperation with young children and HCPs could be a useful strategy for encouraging and facilitating CCC in this age group. We suggest that future research include the perspective of young children, directly and through observation. We argue that including the voices of young children who have been diagnosed with type 1 diabetes will add valuable information about their experiences and needs.
## Implications for Practice
Our findings point to specific CCC practices that can be incorporated into diabetes care for young children. Such practices will encourage care that is responsive to the needs and preferences of young children, foster relationships, and frame diabetes within the context of children’s everyday lives.
## Strengths and Limitations
The main limitation of the present study is that our results are not based on actual observation of current practices. We recognize that there may be a discrepancy between described practices and actual practices. Likewise, self-selection of HCPs, with an interest in psychosocial aspect of diabetes care, is a potential source of selection bias. Therefore, our result may not accurately represent current diabetes practice. Further, this study solely explored the child perspective though the HCPs position. Young children and their parent’s perspective on successful CCC practices are not represented in this study. A key strength of the present article is the broad inclusion and geographical representation of all regions in Denmark. Through our approach of identifying existing practices, we ensure that our results are relevant and can be integrated into current clinical practice.
## Data Availability Statement
The datasets presented in this article are not readily available because consent for sharing interview data has not been obtained from participants. Requests to access the datasets should be directed to [email protected].
## Ethics Statement
The studies involving human participants were reviewed and approved by The Institutional Ethical Review Board, University of Copenhagen, Department of Psychology. The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
PD, DG, and TS contributed to the conception and design of the study, PD collected and transcribed the data and drafted the manuscript. PD led the analysis and interpretation of data in collaboration with DG, LJ, and TS. DG, LJ and TS critically revised the article for intellectual content and gave their approval of the final version to be published.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Association AD. **2. Classification and Diagnosis of Diabetes**. *Diabetes Care* (2015) **38** S8-16. DOI: 10.2337/dc15-S005
2. Patterson CC, Harjutsalo V, Rosenbauer J, Neu A, Cinek O, Skrivarhaug T. **Trends and Cyclical Variation in the Incidence of Childhood Type 1 Diabetes in 26 European Centres in the 25 Year Period 1989–2013: A Multicentre Prospective Registration Study**. *Diabetologia* (2019) **62**. DOI: 10.1007/s00125-018-4763-3
3. Patterson CC, Dahlquist GG, Gyürüs E, Green A, Soltész G. **Incidence Trends for Childhood Type 1 Diabetes in Europe During 1989–2003 and Predicted New Cases 2005–20: A Multicentre Prospective Registration Study**. *Lancet* (2009) **373**. DOI: 10.1016/S0140-6736(09)60568-7
4. DeCosta P, Grabowski D, Skinner T. **The Psychosocial Experience and Needs of Children Newly Diagnosed With Type 1 Diabetes From Their Own Perspective: A Systematic and Narrative Review**. *Diabetic Med* (2020) **37**. DOI: 10.1111/dme.14354
5. Young-Hyman D, De Groot M, Hill-Briggs F, Gonzalez JS, Hood K, Peyrot M. **Psychosocial Care for People With Diabetes: A Position Statement of the American Diabetes Association**. *Diabetes Care* (2016) **39**. DOI: 10.2337/dc16-2053
6. Rathert C, Wyrwich MD, Boren SA. **Patient-Centered Care and Outcomes: A Systematic Review of the Literature**. *Med Care Res Rev* (2013) **70**. DOI: 10.1177/1077558712465774
7. Baker A. **Crossing the Quality Chasm: A New Health System for the 21st Century**. *Br Med J Publishing Group* (2001) **323**. DOI: 10.1136/bmj.323.7322.1192
8. Ford K, Dickinson A, Water T, Campbell S, Bray L, Carter B. **Child Centred Care: Challenging Assumptions and Repositioning Children and Young People**. *J Pediatr Nurs* (2018) **43**. DOI: 10.1016/j.pedn.2018.08.012
9. Delamater AM, de Wit M, McDarby V, Malik JA, Hilliard ME, Northam E. **ISPAD Clinical Practice Consensus Guidelines 2018: Psychological Care of Children and Adolescents With Type 1 Diabetes**. *Pediatr Diabetes* (2018) **19**. DOI: 10.1111/pedi.12736
10. Söderbäck M, Coyne I, Harder M. **The Importance of Including Both a Child Perspective and the Child’s Perspective Within Health Care Settings to Provide Truly Child-Centred Care**. *J Child Health Care* (2011) **15** 99-106. DOI: 10.1177/1367493510397624
11. Coyne I, Hallström I, Söderbäck M. **Reframing the Focus From a Family-Centred to a Child-Centred Care Approach for Children’s Healthcare**. *J Child Health Care* (2016) **20** 494-502. DOI: 10.1177/1367493516642744
12. Coyne I. **Children’s Participation in Consultations and Decision-Making at Health Service Level: A Review of the Literature**. *Int J Nurs Stud* (2008) **45**. DOI: 10.1016/j.ijnurstu.2008.05.002
13. Curtis-Tyler K, Arai L, Stephenson T, Roberts H. **What Makes for a ‘Good’or ‘Bad’paediatric Diabetes Service From the Viewpoint of Children, Young People, Carers and Clinicians? A Synthesis of Qualitative Findings**. *Arch Dis Childhood* (2015) **100**. DOI: 10.1136/archdischild-2014-307517
14. Vygotsky LS. **Mind in Society**. *The Development of Higher Psychological Processes* (1978)
15. Wood D, Bruner JS, Ross G. **The Role of Tutoring in Problem Solving**. *J Child Psychol Psych* (1976) **17** 89-100. DOI: 10.1111/j.1469-7610.1976.tb00381.x
16. Thomas R, Zimmer-Gembeck MJ. **Behavioral Outcomes of Parent-Child Interaction Therapy and Triple P—Positive Parenting Program: A Review and Meta-Analysis**. *J Abnormal Child Psychol* (2007) **35**. DOI: 10.1007/s10802-007-9104-9
17. Kelly KR, Bailey AL. **Narrative Story Stem Methodologies: Use and Utility of Quantitative and Qualitative Approaches Across the Lifespan**. *Narrative Inquiry* (2020) **31**. DOI: 10.1075/ni.20088.kel
18. Braun V, Clarke V. **Using Thematic Analysis in Psychology**. *Qual Res Psychol* (2006) **3** 77-101. DOI: 10.1191/1478088706qp063oa
19. Alderson P, Sutcliffe K, Curtis K. **Children as Partners With Adults in Their Medical Care**. *Arch Dis Childhood* (2006) **91**. DOI: 10.1136/adc.2005.079442
20. Holaday B, Lamontagne L, Marciel J. **Vygotsky’s Zone of Proximal Development: Implications for Nurse Assistance of Children’s Learning**. *Issues Compr Pediatr Nurs* (1994) **17** 15-27. DOI: 10.3109/01460869409078285
21. Curtis-Tyler K. **Facilitating Children’s Contributions in Clinic? Findings from an in-depth qualitative study with children with Type 1 diabetes**. *Diabet Med* (2012) **29**. DOI: 10.1111/j.1464-5491.2012.03714.x
22. Shier H. **Pathways to Participation: Openings, Opportunities and Obligations**. *Children Soc* (2001) **15**. DOI: 10.1002/chi.617
|
---
title: 'Preconception, Interconception, and reproductive health screening tools: A
systematic review'
authors:
- Megan Ren
- Hannah Shireman
- Emily White VanGompel
- Jennifer K. Bello
- Francesca Carlock
- Ashley McHugh
- Debra Stulberg
journal: Health Services Research
year: 2023
pmcid: PMC10012234
doi: 10.1111/1475-6773.14123
license: CC BY 4.0
---
# Preconception, Interconception, and reproductive health screening tools: A systematic review
## Body
What is known on this topic A barrier to preconception and interconception health screening in general outpatient practices is the number of competing clinical needsMany standardized tools exist that proactively screen for these reproductive health needsClinicians seeking guidance on these tools lack a structured review of different approaches *What this* study adds This study provides a structured review of different approaches to preconception and interconception health screeningThe search yielded 53 eligible studies representing 22 tools/standardized approaches, of which 10 had evidence from randomized clinical trials. Tools ranged widely in design, setting, and population of study.
## Abstract
### Objective
To identify and describe the standardized interconception and preconception screening tools for reproductive health needs that are applicable in general outpatient clinical practice.
### Data Sources and Study Setting
This systematic review identifies research on pregnancy intention screening and counseling tools, and standardized approaches to preconception and interconception care. We focus on tools designed for clinical settings, but also include research tools with potential for clinical implementation. These tools may include a component of contraceptive counseling, but those focusing solely on contraceptive counseling were excluded. Data were collected from studies done in the United States between January 2000 and March 2022.
### Study Design
We performed a systematic literature search to generate a list of unique tools, assessed the quality of evidence supporting each tool, and described the peer‐reviewed clinical applications of each. We used the Mixed Methods Appraisal Tool to appraise the quality of individual studies.
### Data Collection/Extraction Methods
We searched PubMed, Web of Science, and CINAHL databases for standardized preconception and interconception health screening tools published in English from January 2000 through March 2022. We used keywords “preconception care,” “interconception care,” “family planning,” “contraception,” “reproductive health services,” and “counseling.” Utilizing the Preferred Reporting Items for Systematic Reviews guidelines, we screened titles and abstracts to identify studies for full text review.
### Principal Findings
The search resulted in 15,399 studies. After removing 4172 duplicates, we screened 11,227 titles/abstracts and advanced 207 for full‐text review. From these, we identified 53 eligible studies representing 22 tools/standardized approaches, of which 10 had evidence from randomized clinical trials. These ranged widely in design, setting, and population of study.
### Conclusions
Clinicians have a choice of tools when implementing standard reproductive screening services. A growing body of research can inform the selection of an appropriate tool, and more study is needed to establish effects on long‐term patient outcomes.
## What is known on this topic
A barrier to preconception and interconception health screening in general outpatient practices is the number of competing clinical needsMany standardized tools exist that proactively screen for these reproductive health needsClinicians seeking guidance on these tools lack a structured review of different approaches
## What this study adds
This study provides a structured review of different approaches to preconception and interconception health screeningThe search yielded 53 eligible studies representing 22 tools/standardized approaches, of which 10 had evidence from randomized clinical trials. Tools ranged widely in design, setting, and population of study.
## INTRODUCTION
Many factors associated with perinatal morbidity and mortality are chronic conditions that begin prior to the start of pregnancy. To address these upstream risks and help people control the timing and conditions of pregnancy, the Centers for Disease Control and Prevention (CDC) and the American College of Obstetricians and Gynecologists (ACOG) recommend routine preconception health counseling, defined as “the health of people during their reproductive years, or the years they can have a child”. 1, 2 However, only $14\%$ of ambulatory visits in the United States include any preconception or contraceptive counseling. Among women at particular risk of increased pregnancy‐related morbidity, including those with a recent birth who had diabetes, hypertension, or both prior to becoming pregnant, fewer than half report receiving preconception health counseling. 3, 4 Additionally, low‐income women are more likely to receive reproductive health services in primary care (vs. dedicated women's health settings) and are disproportionately cared for in federally qualified health centers (FQHCs). 5, 6 Women who have Medicaid insurance or are uninsured have been shown to have increased rates of adverse birth outcomes, 7 hence, it is particularly important that general outpatient practices have a means of identifying those that need preconception/interconception care. In this paper, we use “general outpatient practice” to include primary care and other outpatient practices not specifically designed for family planning.
Patients have expressed interest in receiving preconception/interconception health screening and counseling from their primary care provider (PCP). 8, 9 However, numerous competing demands, lack of preconception health knowledge by both patients and clinicians, and lack of ownership in delivery of preconception care are well‐documented barriers. 10, 11 Often, clinicians do not even recognize that a patient would be eligible for preconception care. 12 Despite the routine use of screening tools to identify other health needs such as depression, screening for preconception and interconception health needs is not widely adopted and there is a lack of consensus on the best approach. 13 To help clinicians in overcoming barriers to incorporating preconception counseling, we conducted this systematic review of preconception, interconception, and pregnancy intention screening tools and standardized (i.e., replicable) approaches relevant to general outpatient practice in the United States. Many of these standardized approaches involved the use of a specific tool, such as a screening question or form, but any study adopting a standardized approach (with or without a specific tool) was included in this review. In describing specific studies, we mirrored the language authors chose for framing their research (i.e., “interconception counseling” vs. “reproductive counseling”, “pregnant person” vs. “woman”); however, we recognize that people of all genders, including nonbinary and transgender individuals, can become pregnant and give birth. The study's objective is to describe these standardized approaches and tools, the settings in which they have been studied, and the published findings about their uses and limitations.
## Search strategy and screening
We conducted a structured search of the PubMed, Web of Science, and CINAHL databases for standardized preconception and interconception health screening tools published in English between January 1, 2000 and March 1, 2022. Our search strategy utilized a combination of phrases and keywords, including: “preconception care,” “interconception care,” “family planning,” “contraception,” “reproductive health services,” and “counseling.” The University of Chicago biomedical librarians provided guidance throughout the development of our search strategy and a complete list of search terms and criteria can be found in (Table S1). To maintain a high degree of search sensitivity, no additional database filters were applied. Studies with titles clearly indicating that they were conducted outside the United States were removed from the results to be screened. The search process was conducted from March to April 2022 and resulted in a total of 15,399 studies. After removing 4172 duplicates, we reviewed the remaining 11,227 (Figure 1).
**FIGURE 1:** *PRISMA diagram of included studies. Flow diagram of our literature search from three databases (PubMed, Web of Science, CINAHL).*
Our inclusion criteria centered on standardized approaches or tools with potential for implementation within clinical settings in the United States. Tools tested in family planning clinics were accepted if they were also applicable in general outpatient settings. To ensure included studies outline tools with potential for clinical implementation, we excluded approaches that required resources beyond the typical clinical setting and those that focused solely on skills development for clinical staff or students. We considered contraceptive needs assessment and counseling a component of preconception and interconception care. Given our study's objective, we excluded studies that focused exclusively on contraceptive use, continuation, or method choice, unless the tool had clear applicability for preconception care more broadly defined. Table 1 provides the full list of inclusion and exclusion criteria used.
**TABLE 1**
| Inclusion criteria | Exclusion criteria |
| --- | --- |
| StudyPublication between January 2000 and March 2022Original research articlesRandomized control trials, observational assessments | StudyStudies that take place outside the United StatesNon‐English language articlesSystematic, scoping, or literature reviews; dissertations, editorials, commentaries, conference abstracts |
| Tool/ApproachAddresses preconception/ interconception careDesigned for clinical settingResearch tools with potential for clinical implementation | Tool/ApproachExclusively about patient choice of contraception method or use of contraceptionCurricula for clinicians or clinical staffRequires resources (including personnel) beyond a typical clinical setting |
Using the Covidence platform to manage and track our screening process, each title/abstract was reviewed by two study team members to determine whether the study could be eligible for inclusion. 14, 15 If the first two reviewers' eligibility decisions conflicted, a separate reviewer performed a third screening to determine inclusion eligibility. Once initial eligibility was assessed, full text manuscripts were reviewed to determine final inclusion. Full text review was similarly conducted by two separate authors. If eligibility assessments conflicted, the two reviewing authors discussed the article and made a final decision together.
## Data abstraction
We exported the included studies from Covidence to Excel and identified the following characteristics of each: the study objectives, the setting and/or population studied, the primary methods, and the key findings.
## Quality assessment
To assess the methodological quality of included studies, we used the Mixed Methods Appraisal Tool (MMAT) designed for critical appraisal of qualitative research, randomized controlled trials, non‐randomized studies, quantitative descriptive studies, and mixed methods studies. 16 Each study was independently assessed by two reviewers. Reviewers did not assess studies for which they were included as authors. For each MMAT question, reviewers responded “Yes” if the study met relevant criteria, “No” if the study did not meet relevant criteria, or “Cannot tell” if the information reported was not adequate to determine whether criteria were met. Of the 53 included studies, 5 were qualitative, 15 were randomized controlled trials, 16 were non‐randomized, 11 were quantitative descriptive, 5 were mixed‐methods studies, and 1 was a corrigendum.
To identify the most rigorous level of research supporting individual tools/approaches, we assessed the overall quality and robustness of the body of evidence available for each. Based on the list of identified studies about each tool, we categorized each as: [1] tested with a randomized controlled trial (RCT); [2] observational study/studies, with assessment of patient outcomes (which could include rates of counseling or patient‐reported satisfaction with care); these encompassed cohort studies and case–control studies; [3] clinical feasibility/acceptability/patient experience studies or validated measure. We assigned a tool the highest level of published quality, for example, we assigned a study that had both a randomized control trial and observational studies as “tested with RCT” based on the levels of evidence as defined by the Oxford Centre of Evidence‐Based Medicine. 17
## Search outcome
Of the 11,227 studies that underwent title and abstract screening, we excluded 11,020 and advanced 207 for full‐text review. We excluded an additional 154 studies during full‐text review, yielding 53 for final inclusion and analysis (Figure 1).
## Article characteristics
Table 2 provides a descriptive summary of the studies included in this review. The majority of articles are published after 2010. Methods included feasibility assessments, pre/post comparisons, and randomized controlled trials (RCTS). The 53 studies encompassed 22 distinct tools or approaches (Table 3), with a focus on different populations and implementation settings.
## Results from the mixed methods appraisal tool quality assessment
Each included study collected data to address clearly defined research questions and therefore met the baseline criteria for assessing quality using the Mixed Methods Appraisal Tool. Assessment findings for each category of study are as follows: the qualitative research and mixed methods studies we assessed met $80\%$ or more of the included criteria. Most randomized controlled trials reported complete outcome data, included groups that were comparable at baseline, and study participants adhered to the assigned intervention. Nine of 14 studies did not provide adequate information to determine whether outcome assessors were blinded to the intervention. All quantitative non‐randomized studies used appropriate measurements for the intervention and the majority reported complete outcome data. However, only $\frac{3}{16}$ studies provided enough information to indicate that the study sample was representative of the target population. Overall, the quantitative descriptive studies utilized appropriate sampling strategies ($\frac{11}{11}$), measurements ($\frac{10}{11}$), and statistical analysis ($\frac{10}{11}$) but only one provided enough information to determine that the sample was representative of the target population.
## Specific tools and other standardized approaches
We found 17 specific tools, and an additional five standardized approaches not associated with a specific tool. Many of the specific tools incorporated a combination of approaches. For example, the Contraceptive Vital Sign used a patient intake questionnaire along with a clinical decision support prompt in the clinic's electronic medical record. Other tools constituted a single approach, such as the Desire to Avoid Pregnancy scale questionnaire, or the Pregnancy Attitudes How Important Timing (PATH) counseling framework.
A number of these various tools and approaches have been tested in specific primary care settings such as Federally Qualified Health Centers (One Key Question, Reproductive Health Self Assessment Tool, electronic medical record reminder) and the Veterans Administration (My Path, One Key Question). Others were designed for specialty care outpatient settings, like READY‐Girls (Reproductive health Education and Awareness of Diabetes in Youth for Girls). Table 3 includes a comprehensive list of specific tools and standardized approaches, as well as all the settings in which they have been tested.
## DISCUSSION
We identified 22 standardized tools and approaches for preconception, interconception, or pregnancy intention screening in U.S. medical care settings. Some approaches have been studied with randomized controlled trials, including two web‐based tools (MyFamilyPlan and Gabby), one patient‐facing pre‐visit tool (MyPath), one in‐clinic screening tool (One Key Question), three standardized counseling approaches (READY‐Girls preconception counseling for patient with diabetes, Computer‐Assisted Motivational Interviewing, and a one‐session motivational interviewing on alcohol use), two electronic prompts (Contraceptive Vital Sign and a CDS focused on teratogen prescribing), and one maternal health screening embedded in the well child visit (Healthy Moms Healthy Babies). Others have been described in multiple observational studies with post‐implementation assessments of outcomes, such as folic acid use (IMPLICIT) and family planning counseling (Family Planning Services prompt). Other approaches are based on sound theory and have data indicating that they are acceptable and appreciated by patients (Reproductive Health Self‐Assessment Tool, Reproductive Life Index, Reproductive Health Service need question) or are associated with contraceptive method selection (Pregnancy Attitudes, Timing, and How important is pregnancy prevention). And finally, some are validated research tools with potential application in clinical settings (Attitude Toward Potential Pregnancy Scale, Reproductive Health Attitudes and Behavior, checklist for preconception care needs of people living with HIV).
For specific health conditions and behaviors, where there is strong evidence that providing preconception care improves pregnancy outcomes, such as preconception glycemic control in patients with diabetes reducing the risk of major and minor congenital anomalies, and preconception use of folic acid reducing the risk of neural tube defects and related anomalies, 18, 19 it is important to use validated tools and approaches found to be effective in clinical practice. Interventions such as READY‐Girls for adolescents with diabetes, or the IMPLICIT model for embedding interconception care within well‐child visits, provide evidence‐based strategies for implementing these practices.
For many other approaches, the evidence indicates that implementing a standardized prompt or screening tool leads to higher rates of screening and counseling, but these have yet to demonstrate changes in patient behavior or clinical outcomes. Similar to our findings, prior systematic reviews have found insufficient evidence on the effectiveness of pregnancy intention screening to change patient outcomes, although with our updated review we add multiple recent studies that indicate pregnancy intention screening leads to changes in counseling rates or patient satisfaction. 20 Future research will ideally follow patients longitudinally and track multiple outcomes, including prevention of undesired pregnancy, better pre‐pregnancy health, and improved perinatal outcomes.
Our approach had several limitations that clinicians should consider when applying these tools: First, we only reviewed studies in English that tested tools in the United States. It is possible that tools developed in other languages and countries could be adapted and prove clinically useful in the United States. Our strategy may have disproportionately excluded tools that would be useful in U.S.‐based patient populations for whom *English is* not the primary language. A second limitation is the exclusion of studies that only looked at contraceptive use (including method selection, continuation, etc.), unless it was clear the tool would also be applicable for preconception counseling. In applying this exclusion criterion, it is possible we inadvertently excluded tools that could be useful for interconception/preconception care.
Nonetheless, the findings of this systematic review are significant in summarizing a diverse and growing body of research: the implementation of clinical tools for preconception and interconception care, and pregnancy intention screening. Using our findings, clinicians will be able to identify applicable tools based on their practice setting and patient population and understand the existing research about each tool. Researchers wishing to build on existing evidence, and those interested in adapting known tools to new settings, may also find this review useful. Routine use of evidence‐based preconception tools has the potential to help patients control the timing of conditions of future pregnancies and improve perinatal outcomes – especially for those with pre‐pregnancy chronic conditions or other risk factors for adverse outcomes. To achieve this potential benefit, ongoing research is needed to further assess the use of tools in everyday clinical practice.
## CONFLICT OF INTEREST
The authors declare that there is no conflict of interest.
## References
1. **Prepregnancy counseling**. *Am Coll Obstet Gynecol* (2019.0) **133** 78-89
2. Johnson K, Posner SF, Biermann J, Cordero JF. *Recommendations to Improve Preconception Health and Health Care ‐‐‐ United States A Report of the CDC/ATSDR Preconception Care Work Group and the Select Panel on Preconception Care* (2006.0)
3. Bello JK, Rao G, Stulberg DB. **Trends in contraceptive and preconception care in United States ambulatory practices**. *Fam Med* (2015.0) **47** 264-271. PMID: 25853596
4. Marshall CJ, Huma Z, Deardorff J, Britton LE. **Prepregnancy counseling among U.S. women with diabetes and hypertension, 2016–2018**. *Am J Prev Med* (2021.0) **61** 529-536. DOI: 10.1016/j.amepre.2021.03.021
5. Hall KS, Patton EW, Crissman HP, Zochowski MK, Dalton VK. **A population‐based study of US women's preferred versus usual sources of reproductive health care**. *Am J Obstet Gynecol* (2015.0) **213** 352.e1-352.e14. DOI: 10.1016/j.ajog.2015.04.025
6. Wood SF, Strasser J, Sharac J. *Community Health Centers and Family Planning in an Era of Policy Uncertainty* (2018.0)
7. Taylor YJ, Liu TL, Howell EA. **Insurance differences in preventive care use and adverse birth outcomes among pregnant women in a Medicaid nonexpansion state: a retrospective cohort study**. *J Women's Health 2002* (2020.0) **29** 29-37. DOI: 10.1089/jwh.2019.7658
8. Stones RE, Stulberg DB, Bello Kottenstette JK. **Patient experiences with pregnancy planning and perspectives on reproductive Care in Community Health Centers: a qualitative study of African American women in Chicago**. *Womens Health Issues* (2017.0) **27** 322-328. DOI: 10.1016/j.whi.2016.10.008
9. Manze MG, Romero DR, Sumberg A, Gagnon M, Roberts L, Jones H. **Women's perspectives on reproductive health Services in Primary Care**. *Fam Med* (2020.0) **52** 112-119. DOI: 10.22454/FamMed.2020.492002
10. Chuang CH, Hwang SW, McCall‐Hosenfeld JS, Rosenwasser L, Hillemeier MM, Weisman CS. **Primary care Physicians' perceptions of barriers to preventive reproductive health care In rural communities**. *Perspect Sex Reprod Health* (2012.0) **44** 78-83. DOI: 10.1363/4407812
11. Akers AY, Gold MA, Borrero S, Santucci A, Schwarz EB. **Providers' perspectives on challenges to contraceptive counseling in primary care settings**. *J Women's Health* (2010.0) **19** 1163-1170. DOI: 10.1089/jwh.2009.1735
12. Bello JK, Adkins K, Stulberg DB, Rao G. **Perceptions of a reproductive health self‐assessment tool (RH‐SAT) in an urban community health center**. *Patient Educ Couns* (2013.0) **93** 655-663. DOI: 10.1016/j.pec.2013.09.004
13. Manze MG, Calixte C, Romero DR. **Physician perspectives on routine pregnancy intention screening and counseling in primary care**. *Contraception* (2020.0) **101** 91-96. DOI: 10.1016/j.contraception.2019.11.004
14. **Covidence systematic review software**
15. Page MJ, McKenzie JE, Bossuyt PM. **The PRISMA 2020 statement: an updated guideline for reporting systematic reviews**. *Int J Surg* (2021.0) **88**. DOI: 10.1016/j.ijsu.2021.105906
16. Hong QN, Pluye P, Fàbregues S. **Mixed Methods Appraisal Tool (MMAT) Version 2018: User guide**
17. **Levels of Evidence**. (2011.0)
18. Ray JG. **Preconception care and the risk of congenital anomalies in the offspring of women with diabetes mellitus: a meta‐analysis**. *QJM* (2001.0) **94** 435-444. DOI: 10.1093/qjmed/94.8.435
19. Temel S, van Voorst SF, Jack BW, Denktaş S, Steegers EAP. **Evidence‐based Preconceptional lifestyle interventions**. *Epidemiol Rev* (2014.0) **36** 19-30. DOI: 10.1093/epirev/mxt003
20. Burgess CK, Henning PA, Norman WV, Manze MG, Jones HE. **A systematic review of the effect of reproductive intention screening in primary care settings on reproductive health outcomes**. *Fam Pract* (2018.0) **35** 122-131. DOI: 10.1093/fampra/cmx086
21. Barnet B, Liu J, DeVoe M, Duggan AK, Gold MA, Pecukonis E. **Motivational intervention to reduce rapid subsequent births to adolescent mothers: a community‐based randomized trial**. *Ann Fam Med* (2009.0) **7** 436-445. DOI: 10.1370/afm.1014
22. Batra P, Mangione CM, Cheng E. **A cluster randomized controlled trial of the MyFamilyPlan online preconception health education tool**. *Am J Health Promot* (2018.0) **32** 897-905. DOI: 10.1177/0890117117700585
23. Bello JK, Chavez J, Liederbauer V, Poston L, Stones R, Stulberg DB. **Perceptions of a Spanish language reproductive health self‐assessment tool among Spanish‐speaking women at a federally qualified health center**. *J Immigr Minor Health* (2020.0) **22** 691-700. DOI: 10.1007/s10903-020-00988-6
24. Bernstein PS, Sanghvi T, Merkatz IR. **Improving preconception care**. *J Reprod Med* (2000.0) **45** 546-552. PMID: 10948464
25. Bickmore T, Zhang Z, Reichert M, Julce C, Jack B. **Promotion of preconception care among adolescents and young adults by conversational agent**. *J Adolesc Health* (2020.0) **67** S45-S51. DOI: 10.1016/j.jadohealth.2019.09.006
26. Callegari LS, Nelson KM, Arterburn DE. **Development and pilot testing of a patient‐centered web‐based reproductive decision support tool for primary care**. *J Gen Intern Med* (2021.0) **36** 2989-2999. DOI: 10.1007/s11606-020-06506-6
27. Caskey R, Stumbras K, Rankin K, Osta A, Haider S, Handler A. **A novel approach to postpartum contraception: a pilot project of Pediatricians' role during the well‐baby visit**. *Contracept Reprod Med* (2016.0) **1** 7. DOI: 10.1186/s40834-016-0018-1
28. Charron‐Prochownik D, Ferons‐Hannan M, Sereika S, Becker D. **Randomized efficacy trial of early preconception counseling for diabetic teens (READY‐girls)**. *Diabetes Care* (2008.0) **31** 1327-1330. DOI: 10.2337/dc07-1266
29. Charron‐Prochownik D, Rodgers Fischl A, Choi J. **Mother‐daughter dyadic approach for starting preconception counseling at puberty in girls with diabetes**. *Res J Womens Health* (2014.0) **1** 2. DOI: 10.7243/2054-9865-1-2
30. Charron‐Prochownik D, Sereika SM, Becker D. **Long‐term effects of the booster‐enhanced READY‐girls preconception counseling program on intentions and behaviors for family planning in teens with diabetes**. *Diabetes Care* (2013.0) **36** 3870-3874. DOI: 10.2337/dc13-0355
31. Charron‐Prochownik D, Wang SL, Sereika SM, Kim Y, Janz NK. **A theory‐based reproductive health and diabetes instrument**. *Am J Health Behav* (2006.0) **30** 208-220. DOI: 10.5993/AJHB.30.2.10
32. Chilukuri N, Cheng TL, Psoter KJ. **Effectiveness of a pediatric primary care intervention to increase maternal folate use: results from a cluster randomized controlled trial**. *J Pediatr* (2018.0) **192** 247-252.e1. DOI: 10.1016/j.jpeds.2017.09.020
33. DeMarco MP, Shafqat M, Horst MA. **Interventions to increase multivitamin use among women in the Interconception period: an IMPLICIT network study**. *Matern Child Health J* (2021.0) **25** 207-213. DOI: 10.1007/s10995-020-03055-6
34. DiPietro MN, Mills C, Snelling A. **Utility of reproductive life plans in identification of potentially teratogenic medication use: a pilot study**. *Birth* (2018.0) **45** 50-54. DOI: 10.1111/birt.12318
35. Dunlop AL, Logue KM, Miranda MC, Narayan DA. **Integrating reproductive planning with primary health care: an exploration among low‐income, minority women and men**. *Sex Reprod Healthc* (2010.0) **1** 37-43. DOI: 10.1016/j.srhc.2010.01.001
36. Dunlop AL, Dretler AW, Badal HJ, Logue KM. **Acceptability and potential impact of brief preconception health risk assessment and counseling in the WIC setting**. *Am J Health Promot* (2013.0) **27** S58-S65. DOI: 10.4278/ajhp.120109-QUAL-7
37. Dunlop AL, Logue KM, Thorne C, Badal HJ. **Change in Women's knowledge of general and personal preconception health risks following targeted brief counseling in publicly funded primary care settings**. *Am J Health Promot* (2013.0) **27** S50-S57. DOI: 10.4278/ajhp.120116-QUAN-39
38. Ferketa M, Schueler K, Song B, Carlock F, Stulberg DB, White VGE. **Facilitators of and barriers to successful implementation of the one key question pregnancy intention screening tool**. *Womens Health Rep* (2022.0) **3** 326-334. DOI: 10.1089/whr.2021.0100
39. Fischl AFR, Herman WH, Sereika SM. **Impact of a preconception counseling program for teens with type 1 diabetes (READY‐girls) on patient‐provider interaction, resource utilization, and cost**. *Diabetes Care* (2010.0) **33** 701-705. DOI: 10.2337/dc09-1821
40. Frayne D, Hughes P, Lugo B. **Interconception Care for Mothers at well child visits after implementation of the IMPLICIT model**. *Matern Child Health J* (2021.0) **25** 1193-1199. DOI: 10.1007/s10995-021-03137-z
41. Gardiner P, Bickmore T, Yinusa‐Nyahkoon L. **Using health information technology to engage African American women on nutrition and supplement use during the preconception period**. *Front Endocrinol* (2021.0) **11**. DOI: 10.3389/fendo.2020.571705
42. Gawron LM, Simonsen S, Millar MM, Lewis‐Caporal J, Patel S, Simmons RG. **Pregnancy risk screening and counseling for women veterans: piloting the one key question in the veterans healthcare administration**. *South Med J* (2021.0) **114** 150-155. DOI: 10.14423/SMJ.0000000000001218
43. Geist C, Aiken AR, Sanders JN. **Beyond intent: exploring the association of contraceptive choice with questions about pregnancy attitudes, timing and how important is pregnancy prevention (PATH) questions**. *Contraception* (2019.0) **99** 22-26. DOI: 10.1016/j.contraception.2018.08.014
44. Gipson JD, Bornstein M, Berger A, Rocca CH. **Desire to avoid pregnancy and contraceptive use among female methadone patients in Los Angeles**. *Contraception* (2021.0) **103** 322-327. DOI: 10.1016/j.contraception.2021.01.019
45. Grotell LA, Bryson L, Florence AM, Fogel J. **Postpartum note template implementation demonstrates adherence to recommended counseling guidelines**. *J Med Syst* (2021.0) **45** 14. DOI: 10.1007/s10916-020-01692-6
46. Ingersoll KS, Ceperich SD, Hettema JE, Farrell‐Carnahan L, Penberthy JK. **Preconceptional motivational interviewing interventions to reduce alcohol‐exposed pregnancy risk**. *J Subst Abus Treat* (2013.0) **44** 407-416. DOI: 10.1016/j.jsat.2012.10.001
47. Jack B, Bickmore T, Hempstead M. **Reducing preconception risks among African American women with conversational agent technology**. *J Am Board Fam Med* (2015.0) **28** 441-451. DOI: 10.3122/jabfm.2015.04.140327
48. Jack BW, Bickmore T, Yinusa‐Nyahkoon L. **Improving the health of young African American women in the preconception period using health information technology: a randomised controlled trial**. *Lancet Digit Health* (2020.0) **2** e475-e485. DOI: 10.1016/S2589-7500(20)30189-8
49. Kavanaugh ML, Schwarz EB. **Prospective assessment of pregnancy intentions using a single‐ versus a multi‐item measure**. *Perspect Sex Reprod Health* (2009.0) **41** 238-243. DOI: 10.1363/4123809
50. Kohn JR, Hilliard ME, Lyons SK, Fox KA, Kushner JA. **Identifying and addressing gaps in reproductive health education for adolescent girls with type 1 diabetes**. *PLoS One* (2018.0) **13**. DOI: 10.1371/journal.pone.0206102
51. Kvach E, Lose J, Marcus H, Loomis L. **Routine screening for pregnancy intention to address unmet reproductive health needs in two urban federally qualified health centers**. *J Health Care Poor Underserved* (2017.0) **28** 1477-1486. DOI: 10.1353/hpu.2017.0128
52. Madrigal JM, Stempinski‐Metoyer K, McManus AE, Zimmerman L, Patel A. **The family planning quotient and reproductive life index (FPQ/RepLI) tool: a solution for family planning, reproductive life planning and contraception counseling**. *Reprod Health* (2019.0) **16** 125. DOI: 10.1186/s12978-019-0787-5
53. Mirabal‐Beltran R, Anderson J, Dariotis JK, Finocchario‐Kessler S. **A checklist to assess childbearing intentions and promote referral to preconception care or contraception: a multi‐site study**. *Matern Child Health J* (2021.0) **25** 786-795. DOI: 10.1007/s10995-020-03051-w
54. Mittal P. **Use of a modified reproductive life plan to improve awareness of preconception health in women with chronic disease**. *Perm J* (2014.0) **18** 28-32. DOI: 10.7812/TPP/13-146
55. Paterno MT, Han H. **Development and psychometric testing of the attitude toward potential pregnancy scale**. *J Obstet Gynecol Neonatal Nurs* (2014.0) **43** 710-718. DOI: 10.1111/1552-6909.12511
56. Rosener SE, Barr WB, Frayne DJ, Barash JH, Gross ME, Bennett IM. **Interconception Care for Mothers during Well‐Child Visits with Family Physicians: an IMPLICIT network study**. *Ann Fam Med* (2016.0) **14** 350-355. DOI: 10.1370/afm.1933
57. Samari G, Foster DG, Ralph LJ, Rocca CH. **Pregnancy preferences and contraceptive use among US women**. *Contraception* (2020.0) **101** 79-85. DOI: 10.1016/j.contraception.2019.10.007
58. Schwarz EB, Parisi SM, Handler SM. **Clinical decision support to promote safe prescribing to women of reproductive age: a cluster‐randomized trial**. *J Gen Intern Med* (2012.0) **27** 831-838. DOI: 10.1007/s11606-012-1991-y
59. Schwarz EB, Parisi SM, Handler SM, Koren G, Shevchik G, Fischer GS. **Counseling about medication‐induced birth defects with clinical decision support in primary care**. *J Women's Health* (2013.0) **22** 817-824. DOI: 10.1089/jwh.2013.4262
60. Schwarz EB, Parisi SM, Williams SL, Shevchik GJ, Hess R. **Promoting safe prescribing in primary care with a contraceptive vital sign: a cluster‐randomized controlled trial**. *Ann Fam Med* (2012.0) **10** 516-522. DOI: 10.1370/afm.1404
61. Schwarz EB, Sobota M, Gonzales R, Gerbert B. **Computerized counseling for folate knowledge and use**. *Am J Prev Med* (2008.0) **35** 568-571. DOI: 10.1016/j.amepre.2008.06.034
62. Shah SD, Prine L, Waltermaurer E, Rubin SE. **Feasibility study of family planning services screening as clinical decision support at an urban federally qualified health center network**. *Contraception* (2019.0) **99** 27-31. DOI: 10.1016/j.contraception.2018.10.004
63. Shlay JC, McEwen D, Bell D. **Integration of family planning services into a sexually transmitted disease clinic setting**. *Sex Transm Dis* (2013.0) **40** 669-674. DOI: 10.1097/OLQ.0b013e318294ff6a
64. Song B, White VanGompel E, Wang C. **Effects of clinic‐level implementation of one key question on reproductive health counseling and patient satisfaction**. *Contraception* (2021.0) **103** 6-12. DOI: 10.1016/j.contraception.2020.10.018
65. Song B, White VanGompel E, Wang C. **Corrigendum to “effects of clinic‐level implementation of one key question on reproductive health counseling and patient satisfaction” [contraception 103 (2021): 6‐12]**. *Contraception* (2021.0) **104** 324-325. DOI: 10.1016/j.contraception.2021.04.008
66. Srinivasulu S, Shah SD, Schechter CB, Prine L, Rubin SE. **Effectiveness of clinical decision support to enhance delivery of family planning services in primary care settings**. *Contraception* (2020.0) **101** 199-204. DOI: 10.1016/j.contraception.2019.11.002
67. Stulberg DB, Dahlquist IH, Disterhoft J, Bello JK, Hunter MS. **Increase in contraceptive counseling by primary care clinicians after implementation of one key question at an Urban Community health center**. *Matern Child Health J* (2019.0) **23** 996-1002. DOI: 10.1007/s10995-019-02754-z
68. Stulberg DB, Datta A, White VanGompel E, Schueler K, Rocca CH. **One key question and the desire to avoid pregnancy scale: a comparison of two approaches to asking about pregnancy preferences**. *Contraception* (2020.0) **101** 231-236. DOI: 10.1016/j.contraception.2019.12.010
69. Thorman A, Engle A, Brintz B. **Quantitative and qualitative impact of one key question on primary care providers' contraceptive counseling at routine preventive health visits**. *Contraception* (2022.0) **109** 73-79. DOI: 10.1016/j.contraception.2022.01.004
70. Upadhya KK, Psoter KJ, Connor KA, Mistry KB, Levy DJ, Cheng TL. **Cluster randomized trial of a pre/Interconception health intervention for mothers in pediatric visits**. *Acad Pediatr* (2020.0) **20** 660-669. DOI: 10.1016/j.acap.2019.10.003
71. Wade GH, Herrman J, McBeth‐Snyder L. **A preconception care program for women in a college setting**. *MCN Am J Matern Nurs* (2012.0) **37** 164-170. DOI: 10.1097/NMC.0b013e31824b59c7
|
---
title: 'Salivary Antimicrobial Peptide Histatin-5 Does
Not Display Zn(II)-Dependent or -Independent Activity against Streptococci'
authors:
- 'Louisa
J. Stewart'
- YoungJin Hong
- Isabel R. Holmes
- Samantha J. Firth
- Yasmin Ahmed
- Janet Quinn
- Yazmin Santos
- Steven L. Cobb
- Nicholas S. Jakubovics
- Karrera Y. Djoko
journal: ACS Infectious Diseases
year: 2023
pmcid: PMC10012264
doi: 10.1021/acsinfecdis.2c00578
license: CC BY 4.0
---
# Salivary Antimicrobial Peptide Histatin-5 Does
Not Display Zn(II)-Dependent or -Independent Activity against Streptococci
## Body
Antimicrobial peptides are short, often cationic peptides that are secreted by diverse organisms from across the domains of life.1 These peptides usually act as immune effectors that kill invading microbes as part of the host innate immune system, but many also play key functions in the normal biology of the host organism. A subfamily of antimicrobial peptides binds metals. Some of these metallo-peptides become activated upon metal binding,2−4 for instance, by folding into an optimal conformation for disrupting microbial membranes or for acting on their targets (e.g., clavanin A from tunicates4 and piscidin from fish2). Other metallo-peptides bind metals and withhold these essential nutrients away from microbes, causing them to starve (e.g., microplusin from cattle ticks5).
Histatins comprise a family of cationic, His-rich, metallo-peptides in the saliva and tears of humans and some higher primates.6−8 These peptides are derived from two parent peptides, namely, Histatin-1 and Histatin-3.6,9 Both parent histatins are expressed by the salivary and tear glands.10,11 Upon secretion in saliva into the oral cavity, the parent histatins are rapidly processed into shorter fragments12−14 by unidentified human salivary proteases or proteases from resident oral microbes. Whether the parent histatins are proteolytically degraded in tears is currently unknown. Of the various salivary fragments, Histatin-5 (Hst5; Table 1) is the best characterized in vitro.
Hst5 is noted for its ability to kill the fungus Candida albicans[15,16], and several pathogenic bacterial species, namely, Pseudomonas aeruginosa, Staphylococcus aureus, Acinetobacter baumanii, Enterococcus faecium, and Enterobacter cloacae.16 Unlike other antimicrobial peptides, Hst5 does not appear to permeabilize fungal membranes, although it does destabilize some bacterial membranes.16 Beyond its direct action on membranes, the antimicrobial activity of Hst5 requires the peptide to be internalized into the cytoplasm, usually via energy-dependent pathways for peptide uptake.16,17 Once in the cytoplasm, Hst5 is thought to encounter its targets, which in C. albicans include the mitochondria18 but in bacteria remain unidentified, and causes toxicity via multiple pathways that are not fully elucidated.15,18 Hst5 contains a characteristic Zn(II)-binding motif, His-Glu-x-His-His (Table 1), but whether Hst5 associates with Zn(II) in saliva is unknown. Likewise, whether Zn(II) binding is essential for the antimicrobial activity of Hst5 is unclear. Synthetic Hst5 derivatives that lack one or all three putative Zn(II)-binding His residues remain active against C. albicans.19 In addition, conflicting reports show that addition of Zn(II) can both enhance20 and suppress21 Hst5 activity against this fungus. However, a recent report indicates that the role of Zn(II) is concentration-dependent: low concentrations of added Zn(II) enhance the antimicrobial activity of Hst5 against C. albicans (compared with the control without any added Zn(II)), while high concentrations of added Zn(II) suppress it.22 Beyond histatins and Zn(II)-binding metallo-peptides, Zn(II)-dependent host innate immune responses are well described. In response to microbial infection, Zn(II) levels and those of Zn(II)-binding or Zn(II)-transporting proteins within a host organism can rise and fall, leading to fluctuations in Zn(II) availability within different niches in the infected host. Increases in Zn(II) availability promote microbial poisoning while decreases in Zn(II) availability promote microbial starvation. These antagonistic host responses, known as “nutritional immunity”,23 suppress microbial growth in the host and inhibit the progress of infectious disease. Although Zn(II) influences the activity of Hst5,22 it is unclear whether histatins themselves participate in nutritional immunity by modulating Zn(II) availability to microbes.
The healthy human oral cavity and oropharynx are colonized by a mixture of microbial species, with *Streptococcus as* the most abundant taxon.24−28 Some species, such as S. gordonii and S. sanguinis, are considered commensals. These species contribute to oral health, for example, by inhibiting colonization by competitor species.29,30 Some streptococcal species are considered pathogenic. For example, S. mutans and S. pyogenes are associated with dental caries and pharyngitis,31 respectively. Nevertheless, asymptomatic carriage of these pathogenic species is common32 and these species are generally considered normal components of the healthy oral and oropharyngeal microflora. Importantly, all streptococci are opportunistic pathogens that can cause disseminated infections, such as bacterial infective endocarditis.33 The goals of this study were to determine the antibacterial activity of Hst5 against oral and oropharyngeal streptococci, and to investigate the potential role of this peptide in influencing Zn(II) availability to the streptococci as a component of nutritional immunity. Based on the established features of nutritional immunity, we specifically examined whether Hst5 limits Zn(II) availability (and promotes microbial Zn(II) starvation) and/or raises Zn(II) availability (and promotes Zn(II) poisoning).
## Abstract
Histatin-5 (Hst5) is a member of the histatin superfamily of cationic, His-rich, Zn(II)-binding peptides in human saliva. Hst5 displays antimicrobial activity against fungal and bacterial pathogens, often in a Zn(II)-dependent manner. In contrast, here we showed that under in vitro conditions that are characteristic of human saliva, Hst5 does not kill seven streptococcal species that normally colonize the human oral cavity and oropharynx. We further showed that Zn(II) does not influence this outcome. We then hypothesized that Hst5 exerts more subtle effects on streptococci by modulating Zn(II) availability. We initially proposed that Hst5 contributes to nutritional immunity by limiting nutrient Zn(II) availability and promoting bacterial Zn(II) starvation. By examining the interactions between Hst5 and *Streptococcus pyogenes* as a model Streptococcus species, we showed that Hst5 does not influence the expression of Zn(II) uptake genes. In addition, Hst5 did not suppress growth of a ΔadcAI mutant strain that is impaired in Zn(II) uptake. These observations establish that Hst5 does not promote Zn(II) starvation. Biochemical examination of purified peptides further confirmed that Hst5 binds Zn(II) with high micromolar affinities and does not compete with the AdcAI high-affinity Zn(II) uptake protein for binding nutrient Zn(II). Instead, we showed that Hst5 weakly limits the availability of excess Zn(II) and suppresses Zn(II) toxicity to a ΔczcD mutant strain that is impaired in Zn(II) efflux. Altogether, our findings led us to reconsider the function of Hst5 as a salivary antimicrobial agent and the role of Zn(II) in Hst5 function.
## Hst5 Does Not Kill Oral
or Oropharyngeal Streptococci
There is little consensus regarding the antibacterial activity of Hst5 against streptococci—it varies depending on the species or experimental conditions,34−40 but the chemical and molecular reasons for these discrepancies have not been identified. In this work, the ability of Hst5 to kill seven oral or oropharyngeal streptococci, namely, S. anginosus, S. gordonii, S. mutans, S. oralis, S. pyogenes, S. salivarius, and S. sanguinis, was examined in parallel. Following the approach used previously for C. albicans and ESKAPE pathogens, these kill assays were performed for several hours in dilute phosphate buffer (10 mM).16,20 Under these conditions, up to 50 μM Hst5 (ca. total histatin concentrations in fresh salivary secretions13) did not promote killing of the streptococcal species (Figure 1A), even when the assay was extended to 24 h (Figure S1). Consistent with a previous report,16 parallel control experiments showed that Hst5 killed P. aeruginosa and C. albicans (Figure S2), confirming that our peptide preparations were active.
**Figure 1:** *Effects of Hst5 on survival of streptococci
in (A) phosphate buffer
and (B) artificial saliva buffer. Bacteria were incubated in phosphate
buffer (10 mM, pH 7.4; N = 3) or artificial saliva
buffer (pH 7.2–7.4; N = 3), with (○)
or without (●) Hst5 (50 μM), and sampled at t = 0 and 3 h for enumeration. Hst5 did not affect the survival of
any species in either buffer (P = 0.73, 0.99, 0.57,
0.72, 0.85, 0.71, 0.50 in phosphate buffer, and 0.72, 0.71, 0.43,
0.56, 0.52, 0.48, and 0.86 in artificial saliva buffer, for S. anginosus, S. gordonii, S. mutans, S. oralis, S. pyogenes, S. salivarius, and S. sanguinis, respectively).*
Like other cationic antimicrobial peptides, the antimicrobial activity of Hst5 is influenced by pH and ionic strength.16,19,41−45 To better reflect the physiological context in which Hst5 plays a role, the kill assays were repeated in an artificial, synthetic “saliva buffer”, whose pH and ionic composition approximate that of saliva (Table S1A). Again, Hst5 did not kill any of the streptococci (Figures 1B and S1). Interestingly, under these new conditions, Hst5 did not kill the control organisms P. aeruginosa and C. albicans (Figure S2). The high ionic strength of the saliva buffer likely interferes with electrostatic binding of the peptide to surface proteins or membranes of these control organisms,16,46 and subsequent internalization and killing. To better understand the activity of Hst5 under conditions that are more characteristic of saliva, further kill assays below used the artificial saliva buffer.
## Zn(II) Does Not Influence the Activity of Hst5 against Streptococci
Saliva typically contains low micromolar levels of total Zn(II) (between 0.2 and 3 μM have been reported47), although the speciation or bioavailability of this metal ion is poorly defined. Our artificial saliva buffer is Zn(II)-deplete (low nanomolar concentrations of Zn(II) are routinely detected by inductively coupled plasma mass spectrometry (ICP MS)). Thus, to determine if the activity of Hst5 against streptococci is Zn(II)-dependent, the kill assays were repeated in the presence of added Zn(II). The results showed that added Zn(II), whether substoichiometric (5 μM), stoichiometric (50 μM), or super-stoichiometric (100 μM) relative to Hst5 (50 μM), neither suppresses nor enhances killing of the seven streptococcal species by Hst5 (Figure 2).
**Figure 2:** *Effects of Zn(II) and Hst5 on survival of streptococci
in artificial
saliva buffer. Bacteria (N = 2) were incubated in
artificial saliva buffer in the presence of added Zn(II) (0, 5, 50,
or 100 μM), with (○) or without (●) Hst5 (50 μM),
and sampled at t = 3 h for enumeration. Addition
of Zn(II) did not influence the effects of Hst5 on the survival of
any species (P values for the interaction between
Zn(II) and Hst5 = 0.40, 0.46, 0.96, 0.98, 0.69, 0.45, and 0.09 for S. anginosus, S. gordonii, S. mutans, S. oralis, S. pyogenes, S. salivarius, and S. sanguinis, respectively).*
## Hst5 Does Not Contribute to Zn(II)-Dependent
Nutritional Immunity
To determine whether Hst5 contributes to Zn(II)-dependent nutritional immunity against streptococci, either by promoting Zn(II) starvation or Zn(II) poisoning, we examined the effects of Hst5 on transcription of Zn(II)-responsive genes. S. pyogenes (Group A *Streptococcus or* GAS) was used as a model Streptococcus, since the transcriptional responses of this species to varying Zn(II) availability is understood (Figure 3), mutant strains lacking key Zn(II) transport proteins are available in our laboratory, and the phenotypes of these mutant strains are known.48
**Figure 3:** *Zn(II)
homeostasis in GAS and hypothesized actions of Hst5. Zn(II)
uptake: AdcAI and AdcAII capture extracellular Zn(II) and transfer
this metal to AdcBC for import into the cytoplasm. These proteins
are transcriptionally upregulated in response to decreases in Zn(II)
availability and Zn(II) starvation (and downregulated in response
to increases in Zn(II) availability).49 Alternatively, Zn(II) may enter the cytoplasm via nonspecific cation transporters (wavy arrow). Zn(II) efflux: CzcD
exports excess Zn(II) out of the cytoplasm. It is transcriptionally
upregulated by GczA in response to increases in Zn(II) availability
and Zn(II) poisoning.50 Alternatively,
Zn(II) may exit the cytoplasm via nonspecific cation
transporters (wavy arrow). Hypothesized actions of Hst5: Hst5 may
bind extracellular Zn(II) and either remain extracellular to suppress
Zn(II) availability or become internalized as the Zn(II)–Hst5
complex and increase Zn(II) availability. Alternatively, Hst5 may
enter the cytoplasm (dotted arrow), bind intracellular Zn(II), and
suppress intracellular Zn(II) availability.*
In response to decreases in Zn(II) availability and Zn(II) starvation, GAS upregulates transcription of the AdcR regulon, including adcAI and adcAII. Conversely, in response to increases in Zn(II) availability and Zn(II) poisoning, GAS upregulates transcription of the GczA regulon, including czcD. Expression of adcAI, adcAII, and czcD, with and without Hst5, was thus examined here. However, poor RNA yields were obtained from the static (nongrowing) bacterial suspensions used in the kill assays. As an alternative approach, GAS was grown in a metal-deplete (low nanomolar concentrations of Zn(II) are routinely detected by ICP MS), chemically defined medium (CDM).51 GAS displayed the same phenotypes in CDM and in artificial saliva buffer, i.e., addition of up to 50 μM Hst5 did not affect the growth of this streptococcus and addition of Zn(II) did not influence this outcome (Figure 4), thus validating the approach.
**Figure 4:** *Effects
of Zn(II) and Hst5 on growth of GAS. Bacteria (N =
3) were cultured in CDM in the presence of Zn(II) (0,
5, 25, or 50 μM), with (○) or without (●) Hst5
(50 μM), and sampled every 20 min for a total of 10 h. While
addition of Zn(II) inhibited bacterial growth (P =
1.0, <0.0001, and <0.0001 for 5, 25, and 50 μM Zn(II),
respectively), addition of Hst5 did not influence this effect (P = 0.88, 0.82, 0.83, and 0.56 for 0, 5, 25, and 50 μM
Zn(II), respectively).*
In the control experiment, adding Zn(II) alone did not perturb transcription of adcAI and adcAII in wild-type GAS, but it did induce expression of czcD (Figure S3A), consistent with an increase in cellular Zn(II) availability or Zn(II) poisoning. Conversely, adding the Zn(II) chelator TPEN induced expression of adcAI and adcAII, consistent with a decrease in cellular Zn(II) availability or Zn(II) starvation, but it did not perturb transcription of czcD (Figure S3B). By contrast, adding Hst5 perturbed neither the basal expression of adcAI or adcAII (Figure 5A) nor the Zn(II)-dependent expression of czcD (Figure 5B). These results indicate that Hst5 promotes neither Zn(II) starvation nor Zn(II) poisoning to GAS and that Hst5 does not contribute to Zn(II)-dependent nutritional immunity against GAS.
**Figure 5:** *Effects of Hst5 on expression of Zn(II)-responsive
genes in GAS.
(A) Background expression of all genes. Bacteria (N = 7) were cultured in CDM with (○) or without (●)
Hst5 (50 μM). Levels of adcAI, adcAII, and czcD mRNA were determined by quantitative
real-time polymerase chain reaction (qRT-PCR) and normalized to holB. Addition of Hst5 did not affect the background expression
of any of the three genes (P = 0.35, 0.74, and 0.08
for adcAI, adcAII, and czcD, respectively). (B) Zn(II)-dependent expression of czcD. Bacteria (N = 3) were cultured in CDM with or
without added Zn(II) (2 or 5 μM), with (○) or without
(●) Hst5 (50 μM). Levels of czcD mRNA
were measured by qRT-PCR, normalized to holB, and
compared with normalized mRNA levels of the corresponding untreated
controls (0 μM added Zn(II)). Addition of Hst5 did not affect
Zn(II)-dependent expression of czcD (P = 0.21 and 0.71 for 2 and 5 μM Zn(II), respectively).*
## Hst5 Weakly Suppresses Zn(II) Toxicity
To further explore the hypothesized role of Hst5 in modulating Zn(II) availability, the effects of Hst5 were examined using GAS ΔadcAI and ΔczcD mutant strains that are deficient in Zn(II) uptake and Zn(II) efflux, respectively (Figure 3). These mutant strains were validated to be sensitive to growth inhibition by the Zn(II) chelator TPEN52,53 and added Zn(II),50,53 respectively (Figure S4). Although additional Zn(II)-binding lipoproteins such as AdcAII contribute to Zn(II) uptake, AdcAI is thought to act as the primary Zn(II) uptake lipoprotein.52,53 Therefore, only the ΔadcAI mutant was employed here.
The ΔadcAI mutant strain displayed wild-type survival and growth phenotypes in the presence of Hst5 (Figure 6A,B), strengthening our proposal that Hst5 does not starve GAS of nutrient Zn(II). Similarly, the ΔczcD mutant strain displayed wild-type survival phenotype (Figure 6C). However, mild differences between the ΔczcD mutant and wild-type strains were observed in growth experiments. While Hst5 did not influence the growth of Zn(II)-treated wild-type organism (see Figure 4), Hst5 weakly but reproducibly improved the growth of the Zn(II)-treated ΔczcD mutant strain (Figure 6D). This effect was observed most clearly upon comparing final culture densities after 10 h of growth since the exponential growth rates were unaffected (Figure S5). This growth-promoting effect of Hst5 appeared to require the predicted Zn(II)-binding ligands His15, His18, and His1954,55 since the ΔH15,18,19 variant of Hst5 did not rescue the growth of the Zn(II)-treated ΔczcD mutant strain (Figure S6, see Table 1 for peptide sequences). These results suggest that Hst5 binds to Zn(II) and suppresses (instead of enhances) the toxicity of an excess of this metal ion to GAS.
**Figure 6:** *Effects
of Hst5 on Zn(II) availability. (A) Survival of ΔadcAI. Bacteria (N = 3) were incubated
in artificial saliva buffer, with (○) or without (●)
Hst5 (50 μM), and sampled at t = 0 and 3 h
for enumeration. Hst5 did not affect the time-dependent survival of
the ΔadcAI mutant (P = 0.90).
(B) Growth of ΔadcAI. Bacteria (N = 2) were cultured in CDM with or without Hst5 (50 μM). Hst5
did not affect the growth of the ΔadcAI mutant
(P = 0.26). (C) Survival of ΔczcD. Bacteria (N = 3) were incubated in artificial
saliva buffer, with or without added Zn(II) (0, 5, 50, or 100 μM),
with (○) or without (●) Hst5 (50 μM). Hst5 did
not affect the Zn(II)-dependent survival of the ΔczcD mutant (P value for the interaction between Hst5
and Zn(II) = 0.73). (D) Growth of ΔczcD. Bacteria
(N = 3) were cultured in CDM in the presence of Zn(II)
(0–20 μM), with (○) or without (●) Hst5
(50 μM). Hst5 did not affect the growth of the ΔczcD mutant in the absence of Zn(II) (P = 0.61) but it did affect growth in the presence of Zn(II) (P = 0.07, 0.02, and 0.01 for 5, 10, and 20 μM Zn(II),
respectively). (E) Levels of cell-associated Zn(II) in ΔczcD. Bacteria (N = 3) were cultured in
CDM in the presence of Zn(II) (0–5 μM), with (○)
or without (●) Hst5 (50 μM), and sampled at t = 4 h. Levels of cell-associated Zn(II) were measured by ICP MS
and normalized to colony counts. Addition of Hst5 had a negative effect
on cellular Zn(II) levels (P = 0.005).*
Two mechanisms are plausible (see Figure 3): (i) Hst5 binds extracellular Zn(II) and suppresses accumulation of this metal ion in the cytoplasm, leading to less Zn(II) toxicity, or (ii) Hst5 binds cellular Zn(II) and enables more Zn(II) to accumulate in the cytoplasm, but with less toxicity. To distinguish between these models, total cell-associated Zn(II) levels in the ΔczcD mutant strain were assessed by ICP MS. Only up to 5 μM Zn(II) was used, since adding 10 μM Zn(II) or more into the cultures inhibited the growth of the ΔczcD mutant and did not produce sufficient biomass for metal analyses. Only wild-type Hst5 peptide was used, owing to the large culture volumes required and the high cost of peptide synthesis. Figure 6E shows that Zn(II) treatment increased cell-associated Zn(II) levels in the ΔczcD mutant, but co-treatment with Hst5 suppressed this effect. These results are consistent with model (i) above, in which Hst5 binds extracellular Zn(II) and suppresses accumulation of Zn(II) in GAS.
## Hst5 Binds Zn(II) with
Micromolar Affinities
To understand how Hst5 weakly modulates Zn(II) availability to GAS and suppresses the toxicity of excess Zn(II) without promoting nutrient Zn(II) starvation, we examined the ability of this peptide to bind Zn(II). Hst5 is thought to bind up to three Zn(II) ions. Previous measurements by isothermal titration calorimetry (ITC) yielded log KZn(II) values of 5.1, 5.0, and 4.0,56 indicating that each Zn(II) ion binds to Hst5 with a high micromolar affinity. In agreement with this proposal, a high micromolar concentration of the Zn(II)–Hst5 complex readily dissociated upon passage through a desalting column (Figure 7A). The affinities of Hst5 to Zn(II) were further re-examined here by competing the peptide with the colorimetric Zn(II) indicator Zincon (log KZn(II) ∼ 6.0) in (Mops) buffer and by monitoring solution absorbances of apo-Zincon (466 nm) and Zn(II)-Zincon (620 nm) (Figure 7B). The competition curve (in the presence of Hst5) was nearly indistinguishable from the control (in the absence of Hst5) (Figure 7C). Moreover, a new peak at 650 nm appeared in the presence of Hst5 (Figure 7B(iii)), indicating the formation of a new species, likely a ternary complex between Hst5, Zincon, and Zn(II). This peak did not disappear upon adding excess Hst5 (10 molar equiv; Figure 7B(iv)). These results indicate that Hst5 does not compete effectively with Zincon and that this peptide binds Zn(II) with high micromolar affinities, as previously estimated by ITC.56
**Figure 7:** *Zn(II) affinity of Hst5.
(A) Separation of Hst5 (○) and
Zn(II) (●) on a polyacrylamide desalting column. (B) Representative
spectral changes upon addition of Zn(II) (0–50 μM) into apo-Zincon (20 μM): (i) in the absence (solid traces)
or (ii) presence (dashed traces) of Hst5 (20 μM). (iii) Overlaid
spectra for 0 and 50 μM Zn(II) from panels (i) and (ii). The
new peak at 650 nm is indicated with a star. (iv) Representative spectral
changes upon addition of excess Hst5 (0–200 μM) into
a solution of Zn(II) (20 μM) and apo-Zincon
(25 μM). (C) Normalized plot of the absorbance intensities of apo-Zincon at 467 nm upon addition of Zn(II), in the absence
(●) or presence (○) of Hst5 (20 μM).*
The lack of competition between Hst5 and Zincon as shown in Figure 7 contrasts with a previous study showing an effective competition between Hst5 and Zincon in phosphate buffer, with Hst5 removing 2 molar equiv of Zn(II) from Zincon.20 Here it is important to highlight that phosphate binds to Zn(II). Although the affinity of phosphate to Zn(II) is relatively low (log KZn(II) ∼ 2.4),57 when used at millimolar concentrations, phosphate can interfere with Zn(II) binding studies by competing for Zn(II). Addition of Zn(II) to apo-Zincon in phosphate buffer (50 mM) instead of Mops buffer led to incomplete formation of Zn(II)-Zincon (monitored at 620 nm), suggesting that Zn(II) partitioned between Zincon and phosphate (Figure S7A,B). Conversely, prolonged incubation (>10 min) of a pre-formed Zn(II)-Zincon complex in phosphate buffer led to a loss of the characteristic blue color (Figure S7C), indicating removal of Zn(II) from Zn(II)-Zincon by phosphate alone (without adding Hst5). Therefore, our studies of Zn(II) binding by Hst5 in Mops buffer are likely to be more reliable.
## AdcAI from GAS Binds Zn(II) with Sub-Nanomolar Affinity
The low affinity of Hst5 to Zn(II) was clearly insufficient to starve wild-type GAS of nutrient Zn(II) (see Figure 5A), indicating that this peptide does not compete with the high-affinity, Zn(II)-specific uptake protein AdcAI (see Figure 3). Therefore, the Zn(II) affinities of AdcAI were examined here by competition with the colorimetric Zn(II) indicator Mag-fura2 (Mf2). The competition curve, generated by monitoring the solution absorbance of apo-Mf2 at 377 nm (Figure 8A(i)), clearly showed two Zn(II) binding sites in AdcAI as anticipated.58 The high-affinity Zn(II) binding site outcompeted Mf2, as evidenced by the lack of spectral changes upon adding up to 1 molar equiv of Zn(II) vs AdcAI (Figure 8A(ii)). The low-affinity site competed effectively with Mf2 with a log KZn(II) = 8.5 (±0.2). The high-affinity site was better estimated using Quin-2 (Q2) as a competitor. By monitoring the absorbance of apo-Q2 at 266 nm, log KZn(II) = 12.5 (±0.2) was obtained for this site (Figure 8B).
**Figure 8:** *Zn(II)
affinity of AdcAI. (A) Low-affinity site. (i) Representative
spectral changes upon titration of Zn(II) (0–25 μM) into
a mixture of apo-Mf2 (10 μM) and AdcAI (5 μM).
(ii) Normalized plot of the absorbance intensities of apo-MF2 (10 μM) at 377 nm upon addition of Zn(II), in the absence
(●) or presence (○) of AdcAI (5 μM). Competition
with Hst5 (X; 10 μM) is shown for comparison. (B) High-affinity
site. (i) Representative spectral changes upon titration of Zn(II)
(0–25 μM) into a mixture of apo-Q2 (7.5
μM) and AdcAI (10 μM). (ii) Normalized plot of the absorbance
intensities of apo-Q2 (7.5 μM) at 262 nm upon
addition of Zn(II), in the absence (●) or presence (○)
of AdcAI (10 μM).*
The log KZn(II) values for AdcAI determined here were each ∼1000-fold higher than those determined previously by ITC.58 ITC can underestimate high metal binding affinities due to lack of sensitivity.59 Crucially, Hst5 did not compete with Mf2 for Zn(II) (Figure 8A(ii)). Thus, the relative affinities between Hst5 and AdcAI, determined using the same approach under the same conditions, support the hypothesis that Hst5 does not compete with AdcAI for binding Zn(II). These relative affinities also provide a molecular explanation for why Hst5 does not suppress the availability of nutrient Zn(II) to wild-type GAS.
Hst5 did not affect the growth of GAS even when AdcAI was deleted by mutagenesis (see Figure 6A,B), suggesting that this peptide does not compete with other high-affinity Zn(II) uptake proteins such as AdcAII (see Figure 3). AdcAII was also expressed here for measurements of Zn(II) affinity. However, consistent with a previous report,60 recombinant AdcAII co-purified with 1 molar equiv of bound Zn(II), which could not be removed without denaturing the protein. Nevertheless, the reported apparent affinity of the S. pneumoniae homologue to Zn(II) (log KZn(II) = 7.7; $67\%$ identity, $81\%$ similarity), determined via competition with Mf2,61 is ∼100-fold higher than that of Hst5, consistent with our proposal that Hst5 does not compete effectively with AdcAII for binding Zn(II).
## Role of
Histatins as Salivary Antimicrobial Agents
The oral cavity is rich in saliva, and interactions between with the components of this host fluid are key for colonization, maintenance, infection, and subsequent transmission of streptococci.62−64 For example, exposure to saliva promotes aggregation of some streptococci and blocks adherence to mucosal epithelia.65,66 Saliva also contains polysaccharides and glycoproteins that may serve as sources of nutrients. Finally, antimicrobial peptides and enzymes such as lysozyme, lactoperoxidase, and chitinase directly inhibit or kill streptococci.67 Given the widely reported antimicrobial activity of Hst5, histatins are thought to function as salivary antimicrobial peptides. Yet, our work shows that Hst5 does not kill seven oral and oropharyngeal streptococcal species under in vitro experimental conditions that are characteristic of saliva. It is tempting to speculate that histatins help shape the microbial composition in the healthy oral cavity by suppressing the viability of some microbes (e.g., C. albicans) but not others (e.g., streptococci). Future work should carefully examine this potential for histatins to exert a selective antimicrobial activity, to verify that it is not associated only with low ionic strength conditions that are not characteristic of saliva. For example, our work showed that antimicrobial activity of Hst5 against C. albicans and P. aeruginosa disappears when examined in our artificial saliva buffer (see Figure S2).
To date, there is no consensus as to whether histatin levels in saliva correlate with infection levels in the oral cavity. Comparisons of children or adult patients with and without dental caries have found no variation in salivary histatin levels,68,69 higher salivary histatin levels in patients with caries,70,71 and lower salivary histatin levels in patients with caries.72−74 Similarly, there is no clear correlation between histatin levels and the prevalence of oral C. albicans in healthy people75 but high histatin levels do correlate with high prevalence of oral candidiasis in immunocompromised patients.76 *It is* important to note that distinct ecological niches exist within the oral cavity. These niches differ in, among many variables, nutrient content, pH, and oxygen tension. Our work does not discount the possibility that histatins exert strong and selective antimicrobial activity in some niches.
## Interactions between Zn(II) and Histatins
Systems for the uptake and efflux of metals such as Zn(II) are important for the survival of streptococci in the oral cavity and oropharynx since salivary concentrations of metals can fluctuate, for example, during and between meals, disease, or human hygiene and dental interventions. In addition, salivary components such as lactoferrin and calprotectin sequester metals and restrict microbial growth.
Our work showed that Hst5 does not contribute to Zn(II)-dependent nutritional immunity against streptococci, since this peptide neither starves our model Streptococcus (S. pyogenes or GAS) of nutrient Zn(II) nor enhances Zn(II) toxicity to this bacterium. These findings are consistent with results from a genome-wide screen of a GAS mutant library, which did not identify genes involved in Zn(II) uptake or Zn(II) efflux as essential for growth in saliva.77 Given the general conservation of Zn(II) homeostasis mechanisms among the streptococci, we anticipate that Hst5 does not contribute to Zn(II)-dependent nutritional immunity against other streptococci.
The low affinity of Hst5 to Zn(II), particularly compared with the high affinities of the Zn(II) uptake lipoproteins AdcAI and AdcAII, explains why Hst5 does not starve GAS (and, presumably, other streptococci) of nutrient Zn(II). Here, the antimicrobial protein calprotectin provides a useful comparison. Calprotectin binds two Zn(II) ions with affinities (log KZn(II) > 11 and >9.6)78 that are comparable to those of AdcAI and higher than that of AdcAII. Indeed, adding calprotectin induces a robust Zn(II) starvation response in streptococci,79,80 consistent with its established role in nutritional immunity.
Its low affinity to Zn(II) also explains why Hst5 only weakly suppresses the availability of excess (toxic) Zn(II) to GAS in vitro. Like most culture media, our CDM51 contains phosphate (∼6 mM) and amino acids (∼6 mM total), which would outcompete Hst5 (50 μM) for binding Zn(II).57 However, if these competing ligands become depleted, for example as a result of bacterial growth, then Hst5 may become competitive and bind Zn(II), particularly when Zn(II) concentrations are high. Such shifts in Zn(II) speciation likely explain why the protective effect of Hst5 on the GAS ΔczcD mutant strain during conditions of Zn(II) stress became apparent only at the later stages of growth (see Figure S5). The increased binding of Zn(II) to Hst5 in these later stages of growth may suppress nonspecific Zn(II) import into the GAS cytoplasm, for instance by outcompeting promiscuous divalent metal transporters.
Unlike in vitro growth media, saliva and its components are continuously replenished in vivo. Saliva contains ∼10 mM phosphate81,82 and proteinaceous components that also bind Zn(II).83 Thus, in vivo, Hst5 is unlikely to be competitive for binding Zn(II). Nonetheless, synergistic effects between Zn(II) and Hst5 may occur in vivo, but likely via indirect mechanisms that do not rely on direct binding of Zn(II) to Hst5 and formation of a Zn(II)–Hst5 complex. Zn(II) and Hst5 may separately target the same cellular pathways in a microbe, leading to the enhancement of the antimicrobial activity of Hst5 by Zn(II). Alternatively, Zn(II) may disable cellular pathways that render the target microbe more susceptible to the separate action of Hst5 on a different cellular pathway (or vice versa), again leading to the enhancement of microbial killing. Indirect interactions between Zn(II) and Hst5 may also exert subtle effects on microbial physiology that do not lead to a direct antimicrobial action and thus are not captured by the assays described here. For example, a combination of Zn(II) and Hst5 at nonlethal doses is thought to reduce the virulence of C. albicans.84 Whether Hst5 reduces the virulence of streptococci and subsequently enables these organisms to become the dominant commensal microorganisms in the oral cavity and oropharynx is an intriguing concept that warrants further investigation.
## Data Presentation
Except growth curves, individual replicates from microbiological experiments are plotted, with shaded columns representing the means and error bars representing standard deviations. Growth curves show the means, with shaded regions representing standard deviations. The number of biological replicates (independent experiments, using different starter cultures and different medium or buffer preparations, performed on different days; N) is stated in figure legends. In the case of metal–protein and metal–peptide titrations, individual data points from two technical replicates (independent experiments performed on different days but using the same protein or peptide preparation) are plotted, but only representative spectra are shown for clarity of presentation.
## Statistical Analyses
Descriptive statistics are displayed on all graphical plots. Inferential statistics have been computed for all data and the relevant P values are listed in figure legends. Unless otherwise stated, tests of significance used two-way analysis of variance using the statistical package in GraphPad Prism 8.0. All analyses were corrected for multiple comparisons.
## Reagents
The nitrate salt of Zn(II) was used in experiments. Numerous additional tests did not identify any observable difference in the results when the chloride or sulfate salts of Zn(II) were used. Peptides were synthesized commercially with free N- and C-termini as the acetate salt, purified to >$95\%$ (GenScript), and confirmed to be metal-free by ICP MS. Concentrations of stock peptide solutions were estimated using solution absorbances at 280 nm in Mops buffer (50 mM, pH 7.4; ε280 = 2667 cm–1). Concentrations of fluorometric and colorimetric metal indicators (Zincon, PAR, Mf2, Q2) were standardized using a commercial standard solution of copper chloride. Concentrations of optically silent chelators (NTA) were standardized by competition with a standardized solution of Zn(II)-Zincon.
## Strains and Culture Conditions
All bacterial strains (Table S1B) were propagated from frozen glycerol stocks onto solid THY (Todd Hewitt + $0.2\%$ yeast extract) medium without any antibiotics and incubated overnight in the presence of $5\%$ v/v of atmospheric CO2. Liquid cultures were prepared in THY or CDM.51 All solid and liquid growth media contained catalase (50 μg/mL).
## Streptococcal Kill Assays
Fresh colonies from an overnight THY agar were resuspended to 106–107 CFU/mL in either potassium phosphate buffer (10 mM, pH 7.4) or artificial saliva buffer (pH 7.2; Table S1A). The cultures were incubated at 37 °C with or without Hst5 and/or Zn(II) as required. At $t = 0$ and 3 h, cultures were sampled and serially diluted in CDM. Exactly 10 μL of each serial dilution was spotted onto THY agar. Colonies were enumerated after overnight incubation at 37 °C.
## C. albicans Kill Assays
Cells from a fresh YPD plate were harvested, washed three times in phosphate-buffered saline (PBS), and resuspended in either potassium phosphate buffer (10 mM, pH 7.4) or saliva salts (pH 7.2) to an OD600 of 0.4 (∼5 × 106 CFU/mL). Cultures were incubated with or without Hst5 at 37 °C. Tubes were inverted every 20 min to maintain cell suspension. At $t = 0$, 1, and 3 h, samples were taken, serially diluted, and plated onto YPD agar. Colonies were numerated following overnight incubation at 30 °C.
## Growth Assays
Colonies from an overnight THY agar were resuspended in CDM to an OD600 = 0.02 and dispensed into wells in flat-bottomed 96-well plates (200 μL per well) containing Hst5 and/or Zn(II) as required. Bacterial growth was monitored using an automated microplate shaker and reader. Each plate was sealed with a gas-permeable, optically clear membrane (Diversified Biotech). OD600 values were measured every 20 min for 10 h. The plates were shaken immediately before each reading (200 rpm, 1 min, double-orbital mode). OD600 values were not corrected for path length (ca. 0.58 cm for a 200 μL culture).
## RNA Extraction
Colonies from an overnight THY agar were resuspended in CDM to an OD600 = 0.02 and incubated in 24-well plates (1.6 mL per well), with or without Hst5 or Zn(II) as required, without shaking, at 37 °C. Each plate was sealed with a gas-permeable, optically clear membrane (Diversified Biotech). At $t = 4$ h, cultures were centrifuged (4000g, 4 °C, 5 min) and the resulting bacterial pellets were resuspended immediately in RNAPro Solution (0.5 mL; MP Biomedicals). Bacteria were lysed in Lysing Matrix B and total RNA was extracted following the manufacturer’s protocol (MP Biomedicals). Crude RNA extracts were treated with RNase-Free DNase I (New England Biolabs). Removal of gDNA was confirmed by PCR using gapA-check-F/R primers (Table S1C). gDNA-free RNA was purified using Monarch RNA Clean-up Kit (New England Biolabs) and visualized on an agarose gel.
## qRT-PCR Analyses
cDNA was generated from RNA (1.6 μg) using SuperScript IV First-Strand Synthesis System (Invitrogen). Each qRT-PCR reaction (20 μL) contained cDNA (5 ng) as template and the appropriate primer pairs (0.4 μM; Table S1C). Samples were analyzed in technical duplicates. Amplicons were detected with Luna Universal qRT-PCR Master Mix (New England Biolabs) in a CFXConnect Real-Time PCR Instrument (Bio-Rad Laboratories). Cq values were calculated using LinRegPCR85 after correcting for amplicon efficiency. Cq values of technical duplicates were typically within ±0.25 of each other. holB, which encodes DNA polymerase III, was used as reference gene. Its transcription levels were verified to remain constant in the experimental conditions tested here.
## Cellular Metal Content
Colonies from an overnight THY agar were resuspended in CDM to an OD600 = 0.02 and incubated at 37 °C with or without Hst5 and/or Zn(II) as required. At $t = 4$ h, an aliquot was collected for the measurement of plating efficiency (colony counts). The remaining cultures were centrifuged (5000g, 4 °C, 10 min). The resulting bacterial pellets were washed once with ice-cold wash buffer (1 M d-sorbitol, 50 mM Tris–HCl, 10 mM MgCl2, 1 mM ethylenediaminetetraacetic acid (EDTA), pH 7.4) and twice with ice-cold PBS. The final pellets were dissolved in concentrated nitric acid (100 μL), heated (85 °C, 1.5 h), and diluted to 3.5 mL with $2\%$ nitric acid. Total metal levels were determined by ICP MS and normalized to colony counts.
## Elution of Zn(II)–Hst5 on a Desalting Column
Apo-Hst5 (100 μM) was incubated with 1.5 molar equiv of Zn(II) for 15 min at the bench and loaded onto a polyacrylamide desalting column (1.8 kDa molecular weight cutoff, Thermo Scientific). Peptide and Zn(II) were eluted from the column using Mops buffer (50 mM, pH 7.4). The concentration of Hst5 in each fraction was determined using QuantiPro BCA Assay Kit (Merck) and known quantities of Hst5 as standards. The concentration of Zn(II) was determined using the colorimetric Zn(II) ligand PAR against a standard curve.
## Equilibrium
Competition Reactions
Our approach to determine metal-binding affinities followed that described by Young and Xiao.59 For each competition (eq 1), a master stock was prepared to contain both competing ligands (L1 and L2) in Mops buffer (50 mM, pH 7.4). Serial dilutions of the metal (M) were prepared separately in deionized water. Exactly 135 μL of the master stock was dispensed into an Eppendorf UVette and 15 μL of the appropriate metal stock was added. Solution absorbances were recorded and used to calculate concentrations of apo- and metalated forms of each ligand. These concentrations were plotted against metal concentrations and fitted in DynaFit86 using binding models as described in the text. The known association or dissociation constants for all competitor ligands are listed in Table S1D1
## Overexpression and Purification of AdcAI and AdcAII
Nucleic acid sequences encoding the soluble domains of AdcAI (from Thr21) and AdcAII (from Thr31) from M1GAS strain 5448 were subcloned into vector pSAT1-LIC using primers listed in Table S1C. This vector generates N-terminal His6-SUMO fusions with the target proteins. The resulting plasmids were propagated in *Escherichia coli* Dh5α, confirmed by Sanger sequencing, and transformed into E. coli BL21 Rosetta 2(DE3).
To express the proteins, transformants were plated onto Lysogeny Broth (LB) agar. Fresh colonies were used to inoculate LB (1 L in 2 L baffled flasks) to an OD600 of 0.01. The culture medium contained ampicillin (100 μg/mL) and chloramphenicol (33 μg/mL). Cultures were shaken (200 rpm, 37 °C) until an OD600 of 0.6–0.8 was reached, and expression was induced by adding isopropyl β-d-1-thiogalactopyranoside (IPTG) (0.1 mM). After shaking for a further 16 h at 20 °C, the cultures were centrifuged (4000g, 4 °C) and the pellets were resuspended in buffer A500 (20 mM Tris–HCl, pH 7.9, 500 mM NaCl, 5 mM imidazole, $10\%$ glycerol).
To purify proteins, bacteria were lysed by sonication (40 kpsi), centrifuged (20,000g, 4 °C), and filtered through a 0.45 μm poly(ether sulfone) (PES) membrane filtration unit. Clarified lysates were loaded onto a HisTrap HP column (Cytiva). The column was washed with 10 column volumes (CV) of buffer A500 followed by 10 CV of buffer A100 (20 mM Tris–HCl, pH 7.9, 100 mM NaCl, $10\%$ w/v glycerol) containing imidazole (5 mM). Both AdcAI and AdcAII were bound to the column and subsequently eluted with 3 CV of buffer A100 containing 250 mM imidazole followed by 5 CV of 500 mM imidazole. Protein-containing fractions were loaded onto a Q HP column (Cytiva). The column was washed with 5 CV of buffer A100 and bound proteins were eluted using a step gradient of 0, 10, 15, and $20\%$ buffer C1000 (20 mM Tris–HCl, pH 7.9, 1000 mM NaCl, $10\%$ w/v glycerol). Eluted proteins were incubated overnight at 4 °C with hSENP2 SUMO protease to cleave the His6-SUMO tag from the target protein. Samples were passed through a second Q HP column and the flowthrough fractions containing untagged target protein were collected.
## References
1. Zhang L. J., Gallo R. L.. **Antimicrobial peptides**. *Curr.
Biol.* (2016) **26** R14. DOI: 10.1016/j.cub.2015.11.017
2. Kim S. Y., Zhang F., Gong W., Chen K., Xia K., Liu F., Gross R., Wang J. M., Linhardt R. J., Cotten M. L.. **Copper regulates the interactions of antimicrobial piscidin peptides from fish mast cells with formyl peptide receptors and heparin**. *J. Biol. Chem.* (2018) **293** 15381. DOI: 10.1074/jbc.RA118.001904
3. Portelinha J., Heilemann K., Jin J., Angeles-Boza A. M.. **Unraveling the implications of multiple histidine residues in the potent antimicrobial peptide Gaduscidin-1**. *J. Inorg. Biochem.* (2021) **219**. DOI: 10.1016/j.jinorgbio.2021.111391
4. Juliano S. A., Pierce S., deMayo J. A., Balunas M. J., Angeles-Boza A. M.. **Exploration of the Innate Immune System of**. *Biochemistry* (2017) **56** 1403. DOI: 10.1021/acs.biochem.6b01046
5. Silva F. D., Rezende C. A., Rossi D. C., Esteves E., Dyszy F. H., Schreier S., Gueiros-Filho F., Campos C. B., Pires J. R., Daffre S.. **Structure and mode of action of microplusin, a copper II-chelating antimicrobial peptide from the cattle tick**. *J. Biol. Chem.* (2009) **284** 34735. DOI: 10.1074/jbc.M109.016410
6. Sabatini L. M., Azen E. A.. **Histatins, a family of salivary histidine-rich proteins, are encoded by at least two loci (HIS1 and HIS2)**. *Biochem. Biophys. Res. Commun.* (1989) **160** 495. DOI: 10.1016/0006-291X(89)92460-1
7. Azen E. A., Leutenegger W., Peters E. H.. **Evolutionary and dietary aspects of salivary basic (Pb) and post Pb (PPb) proteins in anthropod primates**. *Nature* (1978) **273** 775. DOI: 10.1038/273775a0
8. Kalmodia S., Son K. N., Cao D., Lee B. S., Surenkhuu B., Shah D., Ali M., Balasubramaniam A., Jain S., Aakalu V. K.. **Presence of Histatin-1 in Human Tears and Association with Aqueous Deficient Dry Eye Diagnosis: A Preliminary Study**. *Sci. Rep.* (2019) **9**. DOI: 10.1038/s41598-019-46623-9
9. Troxler R. F., Offner G. D., Xu T., Vanderspek J. C., Oppenheim F. G.. **Structural relationship between human salivary histatins**. *J. Dent. Res.* (1990) **69** 2. DOI: 10.1177/00220345900690010101
10. Shah D., Ali M., Pasha Z., Jaboori A. J., Jassim S. H., Jain S., Aakalu V. K.. **Histatin-1 Expression in Human Lacrimal Epithelium**. *PLoS
One* (2016) **11**. DOI: 10.1371/journal.pone.0148018
11. Huang L. C., Jean D., Proske R. J., Reins R. Y., McDermott A. M.. **Ocular surface expression and in vitro activity of antimicrobial peptides**. *Curr. Eye Res.* (2007) **32** 595. DOI: 10.1080/02713680701446653
12. Helmerhorst E. J., Alagl A. S., Siqueira W. L., Oppenheim F. G.. **Oral fluid proteolytic effects on histatin 5 structure and function**. *Arch. Oral Biol.* (2006) **51** 1061. DOI: 10.1016/j.archoralbio.2006.06.005
13. Campese M., Sun X., Bosch J. A., Oppenheim F. G., Helmerhorst E. J.. **Concentration and fate of histatins and acidic proline-rich proteins in the oral environment**. *Arch. Oral Biol.* (2009) **54** 345. DOI: 10.1016/j.archoralbio.2008.11.010
14. Baum B. J., Bird J. L., Millar D. B., Longton R. W.. **Studies on histidine-rich polypeptides from human parotid saliva**. *Arch.
Biochem. Biophys.* (1976) **177** 427. DOI: 10.1016/0003-9861(76)90455-0
15. Puri S., Edgerton M.. **How does it kill?: understanding the candidacidal mechanism of salivary histatin 5**. *Eukaryotic Cell* (2014) **13** 958. DOI: 10.1128/EC.00095-14
16. Du H., Puri S., McCall A., Norris H. L., Russo T., Edgerton M.. **Human Salivary Protein Histatin 5 Has Potent Bactericidal Activity against ESKAPE Pathogens**. *Front. Cell.
Infect. Microbiol.* (2017) **7**. DOI: 10.3389/fcimb.2017.00041
17. Mochon A. B., Liu H.. **The antimicrobial peptide histatin-5 causes a spatially restricted disruption on the**. *PLoS Pathog.* (2008) **4**. DOI: 10.1371/journal.ppat.1000190
18. Helmerhorst E. J., Breeuwer P., van’t Hof W., Walgreen-Weterings E., Oomen L. C. J. M., Veerman E. C. I., Nieuw Amerongen A. V., Abee T.. **The cellular target of histatin 5 on**. *J. Biol. Chem.* (1999) **274** 7286. DOI: 10.1074/jbc.274.11.7286
19. Helmerhorst E. J., van’t Hof W., Veerman E. C. I., Simoons-Smit I., Nieuw Amerongen A. V.. **Synthetic histatin analogues with broad-spectrum antimicrobial activity**. *Biochem. J.* (1997) **326** 39. DOI: 10.1042/bj3260039
20. Norris H.
L., Kumar R., Ong C. Y., Xu D., Edgerton M.. **Zinc Binding by Histatin 5 Promotes Fungicidal Membrane Disruption in**. *J. Fungi* (2020) **6**. DOI: 10.3390/jof6030124
21. Puri S., Li R., Ruszaj D., Tati S., Edgerton M.. **Iron binding modulates candidacidal properties of salivary histatin 5**. *J. Dent. Res.* (2015) **94** 201. DOI: 10.1177/0022034514556709
22. Campbell J. X., Gao S., Anand K. S., Franz K. J.. **Zinc Binding Inhibits Cellular Uptake and Antifungal Activity of Histatin-5 in**. *ACS Infect. Dis.* (2022) **8** 1920. DOI: 10.1021/acsinfecdis.2c00289
23. Hood M. I., Skaar E. P.. **Nutritional immunity: transition metals at the pathogen-host interface**. *Nat. Rev. Microbiol.* (2012) **10** 525. DOI: 10.1038/nrmicro2836
24. Bik E. M., Long C. D., Armitage G. C., Loomer P., Emerson J., Mongodin E. F., Nelson K. E., Gill S. R., Fraser-Liggett C. M., Relman D. A.. **Bacterial diversity in the oral cavity of 10 healthy individuals**. *ISME J.* (2010) **4** 962. DOI: 10.1038/ismej.2010.30
25. Zaura E., Keijser B. J., Huse S. M., Crielaard W.. **Defining the healthy ″core microbiome″ of oral microbial communities**. *BMC Microbiol.* (2009) **9**. DOI: 10.1186/1471-2180-9-259
26. Dewhirst F. E., Chen T., Izard J., Paster B. J., Tanner A. C., Yu W. H., Lakshmanan A., Wade W. G.. **The human oral microbiome**. *J.
Bacteriol.* (2010) **192** 5002. DOI: 10.1128/JB.00542-10
27. Aas J. A., Paster B. J., Stokes L. N., Olsen I., Dewhirst F. E.. **Defining the normal bacterial flora of the oral cavity**. *J. Clin. Microbiol.* (2005) **43** 5721. DOI: 10.1128/JCM.43.11.5721-5732.2005
28. Bach L. L., Ram A., Ijaz U. Z., Evans T. J., Lindström J.. **A Longitudinal Study of the Human Oropharynx Microbiota Over Time Reveals a Common Core and Significant Variations With Self-Reported Disease**. *Front. Microbiol.* (2021) **11**. DOI: 10.3389/fmicb.2020.573969
29. Herrero E. R., Slomka V., Bernaerts K., Boon N., Hernandez-Sanabria E., Passoni B. B., Quirynen M., Teughels W.. **Antimicrobial effects of commensal oral species are regulated by environmental factors**. *J. Dent.* (2016) **47** 23. DOI: 10.1016/j.jdent.2016.02.007
30. Abranches J., Zeng L., Kajfasz J. K., Palmer S. R., Chakraborty B., Wen Z. T., Richards V. P., Brady L. J., Lemos J. A.. **Biology of Oral Streptococci**. *Microbiol. Spectrum* (2018) **6** 6.5.11. DOI: 10.1128/microbiolspec.GPP3-0042-2018
31. Carapetis J. R., Steer A. C., Mulholland E. K., Weber M.. **The global burden of group A streptococcal diseases**. *Lancet Infect.
Dis.* (2005) **5** 685. DOI: 10.1016/S1473-3099(05)70267-X
32. Shaikh N., Leonard E., Martin J. M.. **Prevalence of streptococcal pharyngitis and streptococcal carriage in children: a meta-analysis**. *Pediatrics* (2010) **126** e557-e564. DOI: 10.1542/peds.2009-2648
33. Pant S., Patel N. J., Deshmukh A., Golwala H., Patel N., Badheka A., Hirsch G. A., Mehta J. L.. **Trends in infective endocarditis incidence, microbiology, and valve replacement in the United States from 2000 to 2011**. *J. Am. Coll.
Cardiol.* (2015) **65** 2070. DOI: 10.1016/j.jacc.2015.03.518
34. Fernández-Presas A., Marquez
Torres Y., Garcia Gonzalez R., Reyes Torres A., Becker Fauser I., Rodriguez Barrera H., Ruiz Garcia B., Toloza Medina R., Delgado Dominguez J., Molinari Soriano J. L.. **Ultrastructural damage in**. *Arch. Oral Biol.* (2018) **87** 226. DOI: 10.1016/j.archoralbio.2018.01.004
35. MacKay B. J., Denepitiya L., Iacono V. J., Krost S. B., Pollock J. J.. **Growth-inhibitory and bactericidal effects of human parotid salivary histidine-rich polypeptides on**. *Infect. Immun.* (1984) **44** 695. DOI: 10.1128/iai.44.3.695-701.1984
36. Tian X. L., Salim H., Dong G., Parcells M., Li Y. H.. **The BceABRS four-component system that is essential for cell envelope stress response is involved in sensing and response to host defence peptides and is required for the biofilm formation and fitness of**. *J. Med. Microbiol.* (2018) **67** 874. DOI: 10.1099/jmm.0.000733
37. Andrian E., Qi G., Wang J., Halperin S. A., Lee S. F.. **Role of surface proteins SspA and SspB of**. *Microbiology* (2012) **158** 2099. DOI: 10.1099/mic.0.058073-0
38. Krzyściak W., Jurczak A., Piątkowski J., Kościelniak D., Gregorczyk-Maga I., Kołodziej I., Papież M. A., Olczak-Kowalczyk D.. **Effect of histatin-5 and lysozyme on the ability of**. *Postepy Hig Med.
Dosw.* (2015) **69** 1056
39. Helmerhorst E. J., Hodgson R., van’t
Hof W., Veerman E. C. I., Allison C., Nieuw Amerongen A. V.. **The Effects of Histatin-derived Basic Antimicrobial Peptides on Oral Biofilms**. *J. Dent. Res.* (1999) **78** 1245. DOI: 10.1177/00220345990780060801
40. Moussa D. G., Siqueira W. L.. **Bioinspired caries preventive strategy via customizable pellicles of saliva-derived protein/peptide constructs**. *Sci. Rep.* (2021) **11**. DOI: 10.1038/s41598-021-96622-y
41. Jang W. S., Li X. S., Sun J. N., Edgerton M.. **The P-113 fragment of histatin 5 requires a specific peptide sequence for intracellular translocation in**. *Antimicrob.
Agents Chemother.* (2008) **52** 497. DOI: 10.1128/AAC.01199-07
42. Yu H. Y., Tu C. H., Yip B. S., Chen H. L., Cheng H. T., Huang K. C., Lo H. J., Cheng J. W.. **Easy strategy to increase salt resistance of antimicrobial peptides**. *Antimicrob. Agents Chemother.* (2011) **55** 4918. DOI: 10.1128/AAC.00202-11
43. Han J., Jyoti M. A., Song H. Y., Jang W. S.. **Antifungal Activity and Action Mechanism of Histatin 5-Halocidin Hybrid Peptides against Candida ssp**. *PLoS One* (2016) **11**. DOI: 10.1371/journal.pone.0150196
44. Helmerhorst E. J., Flora B., Troxler R. F., Oppenheim F. G.. **Dialysis unmasks the fungicidal properties of glandular salivary secretions**. *Infect. Immun.* (2004) **72** 2703. DOI: 10.1128/IAI.72.5.2703-2709.2004
45. Meurer M., O’Neil D. A., Lovie E., Simpson L., Torres M. D. T., de
la Fuente-Nunez C., Angeles-Boza A. M., Kleinsorgen C., Mercer D. K., von Kockritz-Blickwede M.. **Antimicrobial Susceptibility Testing of Antimicrobial Peptides Requires New and Standardized Testing Structures**. *ACS Infect.
Dis.* (2021) **7** 2205. DOI: 10.1021/acsinfecdis.1c00210
46. Li X. S., Sun J. N., Okamoto-Shibayama K., Edgerton M.. *J. Biol. Chem.* (2006) **281** 22453. DOI: 10.1074/jbc.M604064200
47. Lynch R. J. M.. **Zinc in the mouth, its interactions with dental enamel and possible effects on caries; a review of the literature**. *Int.
Dent. J.* (2011) **61** 46. DOI: 10.1111/j.1875-595X.2011.00049.x
48. Turner A. G., Ong C. Y., Walker M. J., Djoko K. Y., McEwan A. G.. **Transition Metal Homeostasis in**. *Microbiology of Metal Ions* (2017) **70** 123. DOI: 10.1016/bs.ampbs.2017.01.002
49. Sanson M., Makthal N., Flores A. R., Olsen R. J., Musser J. M., Kumaraswami M.. **Adhesin competence repressor (AdcR) from**. *Nucleic Acids Res.* (2015) **43** 418. DOI: 10.1093/nar/gku1304
50. Ong C. L., Gillen C. M., Barnett T. C., Walker M. J., McEwan A. G.. **An antimicrobial role for zinc in innate immune defense against group A streptococcus**. *J. Infect. Dis.* (2014) **209** 1500. DOI: 10.1093/infdis/jiu053
51. Stewart L. J., Ong C. L., Zhang M. M., Brouwer S., McIntyre L., Davies M. R., Walker M. J., McEwan A. G., Waldron K. J., Djoko K. Y.. **Role of Glutathione in Buffering Excess Intracellular Copper in**. *mBio* (2020) **11**. DOI: 10.1128/mBio.02804-20
52. Tedde V., Rosini R., Galeotti C. L.. **Zn2+ Uptake in**. *PLoS One* (2016) **11**. DOI: 10.1371/journal.pone.0152835
53. Ong C. L., Berking O., Walker M. J., McEwan A. G.. **New Insights into the Role of Zinc Acquisition and Zinc Tolerance in Group A Streptococcal Infection**. *Infect. Immun.* (2018) **86**. DOI: 10.1128/IAI.00048-18
54. Grogan J., McKnight C. J., Troxler R. F., Oppenheim F. G.. **Zinc and copper bind to unique sites of histatin 5**. *FEBS
Lett.* (2001) **491** 76. DOI: 10.1016/S0014-5793(01)02157-3
55. Melino S., Stefano Rufini S., Sette M., Morero R., Grottesi A., Paci M., Petruzzelli R.. **Zn(2+) ions Selectively Induce Antimicrobial Salivary Peptide Histatin-5 To Fuse Negatively Charged Vesicles. Identification and Characterization of a Zinc-Binding Motif Present in the Functional Domain**. *Biochemistry* (1999) **38** 9626. DOI: 10.1021/bi990212c
56. Gusman H., Lendenmann U., Grogan J., Troxler R. F., Oppenheim F. G.. **Is salivary histatin 5 a metallopeptide?**. *Biochim. Biophys.
Acta, Protein Struct. Mol. Enzymol.* (2001) **1545** 86. DOI: 10.1016/S0167-4838(00)00265-X
57. Krężel A., Maret W.. **The biological inorganic chemistry of zinc ions**. *Arch. Biochem.
Biophys.* (2016) **611** 3. DOI: 10.1016/j.abb.2016.04.010
58. Cao K., Li N., Wang H., Cao X., He J., Zhang B., He Q. Y., Zhang G., Sun X.. **Two zinc-binding domains in the transporter AdcA from**. *J. Biol. Chem.* (2018) **293** 6075. DOI: 10.1074/jbc.M117.818997
59. Young T. R., Xiao Z.. **Principles and practice of determining metal-protein affinities**. *Biochem.
J.* (2021) **478** 1085. DOI: 10.1042/BCJ20200838
60. Linke C., Caradoc-Davies T. T., Young P. G., Proft T., Baker E. N.. **The laminin-binding protein Lbp from**. *J. Bacteriol.* (2009) **191** 5814. DOI: 10.1128/JB.00485-09
61. Župan M. L., Luo Z., Ganio K., Pederick V. G., Neville S. L., Deplazes E., Kobe B., McDevitt C. A.. **Conformation of the Solute-Binding Protein AdcAII Influences Zinc Uptake in**. *Front. Cell. Infect.
Microbiol.* (2021) **11**. DOI: 10.3389/fcimb.2021.729981
62. Hamburger M., Robertson O. H.. **Expulsion of group a hemolytic streptococci in droplets and droplet nuclei by sneezing, coughing and talking**. *Am. J. Med.* (1948) **4** 690. DOI: 10.1016/S0002-9343(48)90392-1
63. Hamburger M.. **Studies on the transmission of hemolytic streptococcus infections: II. Beta hemolytic streptococci in the saliva of persons with positive throat cultures**. *J. Infect. Dis.* (1944) 71. DOI: 10.1093/infdis/75.1.71
64. Courtney H. S., Hasty D. L.. **Aggregation of group A streptococci by human saliva and effect of saliva on streptococcal adherence to host cells**. *Infect. Immun.* (1991) **59** 1661. DOI: 10.1128/iai.59.5.1661-1666.1991
65. Courtney H. S., Hasty D.. **Aggregation of group A streptococci by human saliva and effect of saliva on streptococcal adherence to host cells**. *Infect. Immun.* (1991) **59** 1661. DOI: 10.1128/iai.59.5.1661-1666.1991
66. Wollein
Waldetoft K., Mohanty T., Karlsson C., Morgelin M., Frick I. M., Malmstrom J., Bjorck L.. **Saliva-Induced Clotting Captures Streptococci: Novel Roles for Coagulation and Fibrinolysis in Host Defense and Immune Evasion**. *Infect.
Immun.* (2016) **84** 2813. DOI: 10.1128/IAI.00307-16
67. Amerongen A. N., Veerman E. C.. **Saliva--the defender of the oral cavity**. *Oral Dis.* (2002) **8** 12. DOI: 10.1034/j.1601-0825.2002.1o816.x
68. Bhadbhade S. J., Acharya A. B., Thakur S. L.. **Salivary and gingival crevicular fluid histatin in periodontal health and disease**. *J. Clin. Exp. Dent.* (2013) **5**. DOI: 10.4317/jced.51106
69. Ribeiro T. R., Dria K. J., de Carvalho C. B. M., Monteiro A. J., Fonteles M. C., de Moraes Carvalho K., Fonteles C. S. R.. **Salivary peptide profile and its association with early childhood caries**. *Int.
J. Paediatr. Dent.* (2013) **23** 225. DOI: 10.1111/j.1365-263X.2012.01258.x
70. Jurczak A., Koscielniak D., Papiez M., Vyhouskaya P., Krzysciak W.. **A study on beta-defensin-2 and histatin-5 as a diagnostic marker of early childhood caries progression**. *Biol. Res.* (2015) **48**. DOI: 10.1186/s40659-015-0050-7
71. Bielawski K., Tokajuk G., Bielawska A., Maciorkowska E., Jablonski R., Wojcicka A., Bielawski K.. **The assessment of sIgA, histatin-5, and lactoperoxidase levels in saliva of adolescents with dental caries**. *Med. Sci. Monit.* (2014) **20** 1095. DOI: 10.12659/MSM.890468
72. Sun X., Huang X., Tan X., Si Y., Wang X., Chen F., Zheng S.. **Salivary peptidome profiling for diagnosis of severe early childhood caries**. *J. Transl. Med.* (2016) **14**. DOI: 10.1186/s12967-016-0996-4
73. Munther S.. **The impact of salivary lactoperoxidase and histatin-5 on early childhood caries severity in relation to nutritional status**. *Saudi Dent. J.* (2020) **32** 410. DOI: 10.1016/j.sdentj.2020.01.010
74. Vitorino R., Lobo M. J. C., Duarte J. R., Ferrer-Correia A. J., Domingues P. M., Amado F. M. L.. **The role of salivary peptides in dental caries**. *Biomed. Chromatogr.* (2005) **19** 214. DOI: 10.1002/bmc.438
75. Jainkittivong A., Johnson D. A., Yeh C. K.. **The relationship between salivary histatin levels and oral yeast carriage**. *Oral
Microbiol. Immunol.* (1998) **13** 181. DOI: 10.1111/j.1399-302X.1998.tb00730.x
76. Atkinson J. C., Yeh C., Oppenheim F. G., Bermudez D., Baum B. J., Fox P. C.. **Elevation of salivary antimicrobial proteins following HIV-1 infection**. *J. Acquired Immune Defic. Syndr.* (1990) **3** 41. PMID: 2293641
77. Zhu L., Charbonneau A. R. L., Waller A. S., Olsen R. J., Beres S. B., Musser J. M.. **Novel Genes Required for the Fitness of**. *mSphere* (2017) **2**. DOI: 10.1128/mSphereDirect.00460-17
78. Brophy M. B., Hayden J. A., Nolan E. M.. **Calcium ion gradients modulate the zinc affinity and antibacterial activity of human calprotectin**. *J. Am. Chem. Soc.* (2012) **134** 18089. DOI: 10.1021/ja307974e
79. Makthal N., Nguyen K., Do H., Gavagan M., Chandrangsu P., Helmann J. D., Olsen R. J., Kumaraswami M.. **A Critical Role of Zinc Importer AdcABC in Group A Streptococcus-Host Interactions During Infection and Its Implications for Vaccine Development**. *EBioMedicine* (2017) **21** 131. DOI: 10.1016/j.ebiom.2017.05.030
80. Burcham L. R., Le Breton Y., Radin J. N., Spencer B. L., Deng L., Hiron A., Ransom M. R., Mendonca J. D. C., Belew A. T., El-Sayed N. M., McIver K. S., Kehl-Fie T. E., Doran K. S.. **Identification of Zinc-Dependent Mechanisms Used by Group B Streptococcus To Overcome Calprotectin-Mediated Stress**. *mBio* (2020) **11**. DOI: 10.1128/mBio.02302-20
81. Hartman M. L., Groppo F., Ohnishi M., Goodson J. M., Hasturk H., Tavares M., Yaskell T., Floros C., Behbehani K., Razzaque M. S.. **Can Salivary Phosphate Levels Be an Early Biomarker to Monitor the Evolvement of Obesity?**. *Contributions to Nephrology* (2013) **180** 138. PMID: 23652556
82. Savica V., Calo L. A., Caldarera R., Cavaleri A., Granata A., Santoro D., Savica R., Muraca U., Mallamace A., Bellinghieri G.. **Phosphate salivary secretion in hemodialysis patients: implications for the treatment of hyperphosphatemia**. *Nephron Physiol.* (2007) **105** p52. DOI: 10.1159/000098544
83. Lynge
Pedersen A. M., Belstrom D.. **The role of natural salivary defences in maintaining a healthy oral microbiota**. *J.
Dent.* (2019) **80** S3. DOI: 10.1016/j.jdent.2018.08.010
84. Norris H. L., Kumar R., Edgerton M.. **A Novel Role for Histatin 5 in Combination with Zinc to Promote Commensalism in**. *Pathogens* (2021). DOI: 10.3390/pathogens10121609
85. Ramakers C., Ruijter J. M., Deprez R. H., Moorman A. F.. **Assumption-free analysis of quantitative real-time polymerase chain reaction (PCR) data**. *Neurosci. Lett.* (2003) **339** 62. DOI: 10.1016/S0304-3940(02)01423-4
86. Kuzmič P.. **DynaFit--A Software Package for Enzymology**. *Methods in
Enzymology* (2009) **467** 247. PMID: 19897096
|
---
title: 'Co-morbidity associated with development of severe COVID-19 before vaccine
availability: a retrospective cohort study in the first pandemic year among the
middle-aged and elderly in Jönköping county, Sweden'
authors:
- Dennis Nordvall
- Dan Drobin
- Toomas Timpka
- Robert G. Hahn
journal: BMC Infectious Diseases
year: 2023
pmcid: PMC10012282
doi: 10.1186/s12879-023-08115-0
license: CC BY 4.0
---
# Co-morbidity associated with development of severe COVID-19 before vaccine availability: a retrospective cohort study in the first pandemic year among the middle-aged and elderly in Jönköping county, Sweden
## Abstract
### Background
In preparation of future pandemics, it is important to recognise population-level determinants associated with development of severe illness before efficient vaccines and evidence-based therapeutic measures are available. The aim of this study was to identify pre-pandemic diagnoses recorded in a middle-aged and elderly population that were associated with development of severe COVID-19 during the first pandemic year.
### Methods
A cohort study design was used. Severe COVID-19 was defined as a course of illness that resulted in hospital admission or death. A retrospective analysis was performed that comprised all individuals aged 39 years and older ($$n = 189$$,951) living in Jönköping County, Sweden. All diagnosed morbidity recorded in contacts with health care during the pre-pandemic year 2019 was used to identify which diagnoses that were associated with development of severe COVID-19 in the first pandemic year 2020. The analyses were performed separately for each diagnosis using binary logistic regression with adjustment for sex and age.
### Results
Severe COVID-19 was suffered by $0.67\%$ ($$n = 1$$,280) of the middle-aged and elderly population in the first pandemic year. Individuals previously diagnosed with dementia, cerebral palsy, kidney failure, type 2 diabetes mellitus, hypertension, and obesity were at higher risk of developing severe COVID-19. For patients with Type 2 diabetes mellitus, the odds ratio (OR) was 2.18 ($95\%$ confidence interval, 1.92–2.48). Type 1 diabetes mellitus was not associated with increased risk.
### Conclusion
Diagnoses suggesting service provision at long-term healthcare facilities and co-morbidity with components of the metabolic syndrome were associated with an increased risk of developing severe COVID-19 in a middle-aged and elderly population before vaccines were available.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12879-023-08115-0.
## Background
The COVID-19 pandemic has profoundly influenced social life, economy and health services worldwide. Although individuals infected with SARS-CoV-2 may suffer only mild illness, many develop severe COVID-19 requiring hospital admission and possibly leading to death. The course of severe illness is characterized by fever, cough, and dyspnoea followed by dysregulation of the immunological response (“cytokine storm”) causing sudden deterioration [1]. Much effort has been made to limit the spread of the SARS-CoV-2 virus through vaccination programs [2] and to improve the clinical treatment of COVID-19 with, e.g., provision of nirmatrelvir-ritonavir [3] during the early course of illness and corticosteroids in the later phases [4]. However, in preparation for next pandemic, it is important to identify determinants that were associated with development of severe illness before efficient preventive and therapeutic measures are available.
Male sex and advanced age have in previous research been found to be associated with increased risk of developing severe COVID-19 in the early pandemic stages [5, 6]. Many studies have also focused on selected diagnoses and their association with fatal outcomes [7, 8] or clinical mortality risk scores for in-hospital patients created based on vital signs, blood chemistry, and limited sets of health conditions [9–11]. In consequence, heterogeneity of the early population-based studies has made it difficult to identify meaningful associations between development of severe COVID-19 and pre-existing morbidity during the initial pandemic phases [5].
The aim of this study was to identify pre-pandemic diagnoses among the middle-aged and elderly in a representative Swedish county associated with developing severe COVID-19 during the first pandemic year. We considered all clinical diagnoses as potential determinants of developing severe COVID-19. The results are to be used to inform planning non-pharmaceutical interventions and the initial stages of vaccination programs during future pandemics.
## Materials and methods
A cohort study design was used for the analysis. The study cohort consisted of all individuals aged 39 years and older residing in Jönköping County, Sweden on December 31st 2019 ($$n = 189$$,951). The primary endpoint measure was incident cases of severe covid-19 during 2020.
Severe covid-19 was defined as having been diagnosed with COVID-19 confirmed by Reverse Transcription-Polymerase Chain Reaction (RT-PCR) testing (ICD-10 U07.1) and been admitted to hospital or died.
## Data collection
Demographic data (age, sex) for formal Jönköping County residents aged 39 years and older by December 31st 2019 ($$n = 189$$,237) were collected from Statistics Sweden (SCB). Data on all diagnoses recorded in 2019 and on hospital admissions and deaths due to covid-19 in 2020 were collected from the Primary Health Care Registry (44 primary care units) and the Specialised Care Registry (three hospitals) in Region Jönköping Län, the health care provider in Jönköping County. The only missed caregivers were a small number of practicing private physicians.
When comparing the demographic and healthcare records, individuals not included in the formal demographic records were found to have been provided health care in Region Jönköping Län during 2019. These individuals were assumed to remain residing in the county. This resulted in a total study cohort ($$n = 189$$,951) that was $0.38\%$ ($$n = 714$$) larger than the formally defined demographic number.
The ICD-10-SE diagnoses recorded in the study cohort during 2019 comprised 2,028 different diagnosis codes. The 330 most frequently recorded codes accounted for $90\%$ of all registered diagnoses in the study county during 2019 (Supplement Table S1). Binary variables were created for each of these diagnosis codes at the ICD-10-SE Category level (i.e., three characters). When necessary, the Diagnosis level, i.e. four characters, was also used. These 330 variables were the only representations of diagnosis codes used in the study.
Data on severe COVID-19 were gathered from individuals having in 2020 been admitted to hospital and/or died diagnosed with COVID-19 confirmed by Reverse Transcription-Polymerase Chain Reaction (RT-PCR) test (ICD-10 U07.1).
## Data analyses
The study population was first disaggregated according to sex, age, morbidity in 2019, and severe COVID-19 in 2020 and descriptive statistics presented. Binary logistic regression models were then applied to identify determinants of the primary outcome severe COVID-19 (1, yes; 0, no). The determinant variables included in the models were sex, age, and medical diagnoses recorded in 2019. Models were first produced using age (centiles from 39 years of age; reference category 39–48 years) and sex (1, male; 0, female), respectively, as determinant variables and severe COVID-19 in 2020 as outcome variable. The model using sex as determinant variable was adjusted for age and vice versa. Thereafter separate adjusted models were produced for each and every of the 330 selected diagnoses recorded in 2019 (1, present; 0, not present). These models were adjusted for sex and age and used severe COVID-19 in 2020 (1, present; 0, not present) as outcome variable.
The threshold for significance was set at $P \leq 0.0001.$ The results are reported as the odds ratio (OR) and the $95\%$ confidence interval ($95\%$ CI). The statistical analyses were performed using the R software package version 4.0 [12].
## Results
The average age of the middle-aged and elderly study cohort was 61.1 years; the proportion of women was $51\%$. Eigthy-six percent ($$n = 162$$,967) of the population had a diagnosis recorded in the healthcare information system during the pre-pandemic year. Of these, 6,715 ($4.1\%$) were diagnosed with covid-19 and 1,218 ($0.74\%$) developed severe covid-19 in the first pandemic year. Additionally, 62 ($0.2\%$) of the 26,984 individuals with no recorded contact with the healthcare system in 2019 developed severe COVID-19. In total, 1,280 individuals ($0.67\%$) in the study cohort developed severe COVID-19 during 2020. The fatality proportion was $32\%$ ($$n = 405$$).
Overall, more men than women in the study cohort were admitted to hospital or died, with an age-adjusted OR of 1.39 ($95\%$ CI, 1.24–1.55) (Table 1; Fig. 1). Individuals aged 80 years or older had a 7 to 13-fold higher risk of developing severe covid-19 compared with those aged 39–48 years (Table 1).
Table 1Binary logistic regression models of severe COVID-19 displayed by age and sex. The model of sex was adjusted for age and vice versa. Risk estimates were based on 189,951 individuals; 1,280 of these developed severe COVID-19 ($0.67\%$)Sex / Age groupOdds ratio ($95\%$ confidence interval)Sex1.39 (1.24–1.55)Age group39–48 years149–58 years1.53 (1.21– 1.93)59–68 years2.09 (1.67–2.63)69–78 years2.83 (2.28–3.54)79–88 years6.95 (5.64–8.64)89–98 years11.93 (9.37–15.25)99–108 years12.85 (4.98–27.18)Male sex and age category 39–48 years are used as reference.
Fig. 1Incidence of severe COVID-19 in the study cohort. Data for individuals ages 39 and older are shown. The data for men and women are separated Diagnoses associated with increased risk of severe covid-19 in the age- and sex-adjusted models are listed in Table 2; these included dementia (various forms), kidney failure, cerebral palsy, history of self-harm, type 2 diabetes mellitus, obesity, atherosclerosis, and heart failure.
Table 2Diagnosis codes recorded in 2019 that were associated with severe COVID-19 in 2020 in binary logistic regression models adjusted for age and sexICD10 codeDiagnosisNOR for severe covid-19 ($$n = 1$$,280)P-valueDiseases of the circulatory systemI10Essential hypertension59,9961.48 (1.31–1.67)3 × 10− 10I50Heart failure88332.08 (1.77–2.44)2 × 10− 16I70Atherosclerosis9242.63 (1.80–3.72)2 × 10− 7Diseases of the endocrine system and metabolismE11Diabetes mellitus type 218,8972.18 (1.92– 2.48)2 × 10− 16E66Obesity74942.25 (1.83–2.73)4 × 10− 15E63Nutritional deficiencies40892.12 (1.70– 2.62)1 × 10− 11Mental disordersF41Anxiety13,4361.54 (1.29–1.82)2 × 10− 6F03Unspecified dementia22292.11 (1.64–2.67)2 × 10− 9F01Vascular dementia11332.69 (1.97– 3.58)7 × 10− 11Diseases of the respiratory systemJ45Asthma10,6551.52 (1.24–1.85)3 × 10− 5J44Chronic obstructive pulmonary disease64841.91 (1.57– 2.30)4 × 10− 11Digestive diseasesK59Obstipation95042.16 (1.84– 2.53)2 × 10− 16Diseases of the genitourinary systemN39Urinary tract infection/incontinence93741.94 (1.64–2.28)3 × 10− 15N30Cystitis86721.83 (1.52–2.20)1 × 10− 10N19Unspecified kidney failure30971.90 (1.51– 2.37)2 × 10− 8N18Chronic kidney failure21512.44 (1.88–3.13)8 × 10− 12Diseases of the nervous systemG47Sleep disorders90331.68 (1.38– 2.03)8 × 10− 8G30Alzheimer’s disease11972.66 (1.94– 3.56)3 × 10− 10G80Cerebral palsy1658.39 (3.53– 16.81)6 × 10− 8Factors influencing health statusZ921Long-term use of anticoagulants81471.77 (1.49– 2.10)4 × 10− 11Z867History of pulmonary embolism35511.86 (1.45–2.35)6 × 10− 7Z99Dependence on enabling machines22272.23 (1.57– 3.07)3 × 10− 6Z915Personal history of self-harm12633.48 (2.09– 5.41)3 × 10− 7Z94Transplanted organ5874.22 (2.45–6.74)3 × 10− 8Diseases of the blood and immune mechanismD64Anemia47501.90 (1.54– 2.31)8 × 10− 10Diseases of the musculoskeletal systemM10Gout38472.10 (1.67–2.60)7 × 10− 11 Secondary conditions and symptoms were also found to be associated with increased risk of severe COVID-19, e.g. dependence on enabling technologies and obstipation.
## Discussion
We found that $0.67\%$ of the middle-aged and elderly population in a Swedish county developed severe COVID-19 in the first pandemic year. Old age, male sex, conditions associated with daily care support needs, and diagnoses included in the metabolic syndrome were associated with developing severe COVID-19 before the vaccination program was initiated. The increased risk of severe COVID-19 in individuals with dementia, cerebral palsy, and a history of self-harm may be explained by the high prevalence of long-term care facility residents in these groups. Swedish nursery homes suffered substantial problems with widespread dissemination of SARS-CoV-2 during 2020 [13]. In contrast, the diagnoses associated with the metabolic syndrome, comprising type 2 diabetes mellitus, obesity, and hypertension, indicated an individual frailty for developing a severe course of illness following SARS-CoV-2 infection [14–16]. This observation suggests that in the period before vaccines were available, co-morbidity with disorders having lifestyle-related etiologies were associated with development of severe covid-19. However, this hypothesis requires confirmation in studies accounting for causal mechanisms in the analysis models.
## Determinants of COVID-19 in unvaccinated populations
In a meta-analysis of 76 population-based studies performed before vaccine availability, Booth et al. found increased risk estimates for male sex, advanced age, severe obesity, and active cancer [5]. Some diagnoses we observed to be associated with increased risk (such as diabetes type 2, hypertension, and kidney disease) may not have reached statistical significance in the meta-analysis due to the heterogeneity of the studies included. In comparison, the OpenSAFELY study based on electronic medical record data from $40\%$ of all patients in England reported findings similar to our observations, i.e. that advanced age, male sex, asthma, hypertension, diabetes, recent cancer diagnosis, and reduced kidney function were associated with development of severe COVID-19 [6]. However, it is noteworthy that cancer was not included among the determinants identified in our study. One explanation is that we, due to interpretation issues associated with “Table 2 Fallacy” [17], did not adjust for co-morbidity in our analyses. Other explanations include that the non-pharmaceutical infectious disease control interventions implemented among the cancer patients were successful in the study county, and that the cancer diagnoses was recorded during the pre-pandemic year and many patients with the diagnosis had completed their cancer treatments during the first pandemic year.
## Study strengths and limitations
An important strength of our study is that the data on pre-existing morbidity were collected before the outbreak of the pandemic, which prevents diagnoses caused by COVID-19 to be included in the results. Moreover, the data were derived from both specialised (hospital) care and primary care settings. A large majority of the middle-aged and elderly study population ($86\%$) had data on diagnoses recorded the healthcare system during the pre-pandemic year. We have not been able to identify studies that use pre pandemic diagnosis data from a total middle-aged and elderly population for analyses of determinants of hospitalisation and death due to COVID-19 in the first pandemic year. Regarding limitations, there can have been diseases that remained undiagnosed in the population. Our data are probably more complete for severe diseases than for the less severe diseases. Moreover, morbidity that is diagnosed according to well-defined criteria may also be more complete. For example, physicians may not have registered all diagnoses or been reluctant to record behavioural problems or lifestyle issues, e.g. substance abuse. Unobserved confounding may also have been present. We did not adjust the models with the separate diagnoses as determinants for severe COVID-19 with regards to concurrent co-morbidity or socioeconomic factors, e.g., country of birth, education, or income. Associations between socioeconomic factors and propensity of need for intensive care COVID-19 treatment during the early pandemic has been reported from Sweden [18]. The influence from comorbidity and socioeconomic factors on COVID-19 morbidity during the early pandemic phase before vaccines were available warrant further research.
An important limitation of our study design based on data retrieved from electronic medical records. Criticism has been raised against the OpenSafely study that used a study design similar to ours [6, 17, 19]. The authors collected data on patients who had deceased following that a validated or suspected SARS-CoV-2 infection was recorded in their electronic medical record. Hence, PCR testing had not been uniformly applied to validate the presence of COVID-19 infection. In our study, the endpoint measure included having a positive SARS-CoV-2 test recorded in the electronic medical record. Nonetheless, it was not ascertained whether the infection was the cause of hospitalisation or death. Also, the generalization of our results to other setting must be made with caution, although we believe that the strengths of the study (completeness of data collected from a total population) provide satisfactory external validity. We thus infer that our results are representative of comparable counties.
## Conclusion
Electronic medical record data from primary and specialised care in Region Jönköping County, Sweden, showed that diagnoses suggesting service provision at long-term healthcare facilities and co-morbidity with components of the metabolic syndrome were associated with an increased risk of developing severe COVID-19 among the middle-aged and elderly in the period before vaccines were available. Our results can be used to inform planning of non-pharmaceutical interventions during the early stages of future pandemics caused by pathogens similar to SARS-CoV-2.
## Electronic supplementary material
Below is the link to the electronic supplementary material.
Supplementary Material 1
## References
1. Hu B, Huang S, Yin L. **The cytokine storm and COVID-19**. *J Med Virol* (2021.0) **93** 250-6. DOI: 10.1002/jmv.26232
2. Korang SK, von Rohden E, Veroniki AA, Ong G, Ngalamika O, Siddiqui F, Juul S, Nielsen EE, Feinberg JB, Petersen JJ, Legart C, Kokogho A, Maagaard M, Klingenberg S, Thabane L, Bardach A, Ciapponi A, Thomsen AR, Jakobsen JC, Gluud C. **Vaccines to prevent COVID-19: a living systematic review with Trial Sequential Analysis and network meta-analysis of randomized clinical trials**. *PLoS ONE* (2022.0) **17** e0260733. DOI: 10.1371/journal.pone.0260733
3. 3.Shah MM, Joyce B, Plumb ID, Sahakian S, Feldstein LR, Barkley E, Paccione M, Deckert J, Sandmann D, Gerhart JL, Hagen MB. Paxlovid Associated with Decreased Hospitalization Rate Among Adults with COVID-19 - United States, April-September 2022. MMWR Morb Mortal Wkly Rep. 2022 Dec 2;71(48):1531–1537.
4. Bafadhel M, Faner R, Taillé C, Russell REK, Welte T, Barnes PJ, Agustí A. **Inhaled corticosteroids for the treatment of COVID-19**. *Eur Respir Rev* (2022.0) **31** 220099. DOI: 10.1183/16000617.0099-2022
5. Booth A, Reed AB, Ponzo S, Yassaee A, Aral M, Plans D, Labrique A, Mohan D. **Population risk factors for severe disease and mortality in COVID-19: a global systematic review and meta-analysis**. *PLoS ONE* (2021.0) **16** e247461. DOI: 10.1371/journal.pone.0247461
6. Williamson EJ, Walker AJ, Bhaskaran K, Bacon S, Bates C, Morton CE, Curtis HJ, Mehrkar A, Evans D, Inglesby P, Cockburn J, McDonald HI, MacKenna B, Tomlinson L, Douglas IJ, Rentsch CT, Mathur R, Wong AYS, Grieve R, Harrison D, Forbes H, Schultze A, Croker R, Parry J, Hester F, Harper S, Perera R, Evans SJW, Smeeth L, Goldacre B. **Factors associated with COVID-19-related death using OpenSAFELY**. *Nature* (2020.0) **584** 430-6. DOI: 10.1038/s41586-020-2521-4
7. Behnoush AH, Bazmi E, Forouzesh M, Behnoush B. **Risk of COVID-19 infection and the associated hospitalization, ICU admission and mortality in opioid use disorder: a systematic review and meta-analysis**. *Addict Sci Clin Pract* (2022.0) **17** 68. DOI: 10.1186/s13722-022-00349-8
8. Oldani S, Petrelli F, Dognini G, Borgonovo K, Parati MC, Ghilardi M, Dottorini L, Cabiddu M, Luciani A. **COVID-19 and lung cancer survival: an updated systematic review and meta-analysis**. *Cancers (Basel)* (2022.0) **14** 5706. DOI: 10.3390/cancers14225706
9. Clift AK, Coupland CAC, Keogh RH, Diaz-Ordaz K, Williamson E, Harrison EM, Hayward A, Hemingway H, Horby P, Mehta N, Benger J, Khunti K, Spiegelhalter D, Sheikh A, Valabhji J, Lyons RA, Robson J, Semple MG, Kee F, Johnson P, Jebb S, Williams T, Hippisley-Cox J. **Living risk prediction algorithm (QCOVID) for risk of hospital admission and mortality from coronavirus 19 in adults: national derivation and validation cohort study**. *BMJ* (2020.0) **371** m3731. DOI: 10.1136/bmj.m3731
10. 10.Knight SR, Ho A, Pius R, Buchan I, Carson G, Drake TM, Dunning J, Fairfield CJ, Gamble C, Green CA, Gupta R, Halpin S, Hardwick HE, Holden KA, Horby PW, Jackson C, Mclean KA, Merson L, Nguyen-Van-Tam JS, Norman L, Noursadeghi M, Olliaro PL, Pritchard MG, Russell CD, Shaw CA, Sheikh A, Solomon T, Sudlow C, Swann OV, Turtle LC, Openshaw PJ, Baillie JK, Semple MG, Docherty AB, Harrison EM. ; ISARIC4C investigators. Risk stratification of patients admitted to hospital with covid-19 using the ISARIC WHO Clinical Characterization Protocol: development and validation of the 4 C Mortality Score. BMJ 2020; 370: m3339.
11. Stachel A, Daniel K, Ding D, Francois F, Phillips M, Lighter J. **Development and validation of a machine learning model to predict mortality risk in patients with COVID-19**. *BMJ Health Care Inform* (2021.0) **28** e100235. DOI: 10.1136/bmjhci-2020-100235
12. 12.R Core Team. : The R project for statistical computing. The R Foundation, Vienna, Austria. URL: https://www.R-project.org/).
13. 13.Nilsson L, Andersson C, Kastbom L, Sjödahl R. Association between vaccination and preventive routines on COVID-19-related mortality in nursing home facilities: a population-based systematic retrospective chart review. Prim Health Care Res Dev. 2022 Nov 18;23:e75. 10.1017/S1463423622000640
14. Samson SL, Garber AJ. **Metabolic syndrome**. *Endocrinol Metab Clin North Am* (2014.0) **43** 1-23. DOI: 10.1016/j.ecl.2013.09.009
15. Gupta A, Gupta V. **Metabolic syndrome: what are the risks for humans?**. *Biosci Trends* (2010.0) **4** 204-12. PMID: 21068471
16. Kassi E, Pervanidou P, Kaltsas G, Chrousos G. **Metabolic syndrome: definitions and controversies**. *BMC Med* (2011.0) **9** 48. DOI: 10.1186/1741-7015-9-48
17. 17.Westreich D, Edwards JK, van Smeden M et al. Comment on Williamson. (OpenSAFELY): The Table 2 Fallacy in a Study of COVID-19 Mortality Risk Factors. Epidemiology. 2021 Jan;32(1):e1-e2.
18. 18.Nordberg P, Jonsson M, Hollenberg J, Ringh M, Kiiski Berggren R, Hofmann R, Svensson P. Immigrant background and socioeconomic status are associated with severe COVID-19 requiring intensive care. Sci Rep. 2022 Jul 15;12(1):12133. 10.1038/s41598-022-15884-2
19. Tennant PWG, Murray EJ. **The Quest for timely insights into COVID-19 should not come at the cost of scientific rigor**. *Epidemiology* (2021.0) **32** e2. DOI: 10.1097/EDE.0000000000001258
|
---
title: Chronic Strongyloides stercoralis infection increases presence of the Ruminococcus
torques group in the gut and alters the microbial proteome
authors:
- Na T. D. Tran
- Apisit Chaidee
- Achirawit Surapinit
- Manachai Yingklang
- Sitiruk Roytrakul
- Sawanya Charoenlappanit
- Porntip Pinlaor
- Nuttanan Hongsrichan
- Sirirat Anutrakulchai
- Ubon Cha’on
- Somchai Pinlaor
journal: Scientific Reports
year: 2023
pmcid: PMC10012286
doi: 10.1038/s41598-023-31118-5
license: CC BY 4.0
---
# Chronic Strongyloides stercoralis infection increases presence of the Ruminococcus torques group in the gut and alters the microbial proteome
## Abstract
We explored the impact of chronic *Strongyloides stercoralis* infection on the gut microbiome and microbial activity in a longitudinal study. At baseline (time-point T0), 42 fecal samples from matched individuals (21 positive for strongyloidiasis and 21 negative) were subjected to microbiome 16S-rRNA sequencing. Those positive at T0 (untreated then because of COVID19 lockdowns) were retested one year later (T1). Persistent infection in these individuals indicated chronic strongyloidiasis: they were treated with ivermectin and retested four months later (T2). Fecal samples at T1 and T2 were subjected to 16S-rRNA sequencing and LC–MS/MS to determine microbial diversity and proteomes. No significant alteration of indices of gut microbial diversity was found in chronic strongyloidiasis. However, the Ruminococcus torques group was highly over-represented in chronic infection. Metaproteome data revealed enrichment of Ruminococcus torques mucin-degrader enzymes in infection, possibly influencing the ability of the host to expel parasites. Metaproteomics indicated an increase in carbohydrate metabolism and Bacteroidaceae accounted for this change in chronic infection. STITCH interaction networks explored highly expressed microbial proteins before treatment and short-chain fatty acids involved in the synthesis of acetate. In conclusion, our data indicate that chronic S. stercoralis infection increases Ruminococcus torques group and alters the microbial proteome.
## Introduction
The nematode *Strongyloides stercoralis* is a neglected soil-transmitted helminth (STH) affecting approximately 600 million people around the world1. Adult female S. stercoralis inhabits in the small intestine. Infection usually occurs when filariform larvae penetrate through the skin to initiate the parasitic cycle2. The South-East Asian, African, Latin American and Western Pacific regions have the highest prevalence of strongyloidiasis. This can be a chronic, lifelong infection because S. stercoralis has the unique ability to undertake autoinfection of its host3. In chronic infection, there is reduced fecal excretion of larvae: the parasite persists with low worm burden and generally causes no symptoms (but a variety of gastrointestinal manifestations, pulmonary and cutaneous manifestations may occur)2,4. Chronic strongyloidiasis can remain unnoticed for decades. In Northeast Thailand, the prevalence of strongyloidiasis has remained high (around $25\%$) over a long time period5–7, predominantly among those over the age of 60. Hyperinfection or disseminated infections of S. stercoralis typically affect immunosuppressed hosts, such as those undergoing corticosteroid treatment, infected with human T-cell lymphotropic virus type 1 (HTLV-1) or elderly individuals, and can cause fatal disease8–10.
Helminths have a range of effects on the human gut microbiome11. The gut microflora has important roles in human health via fermentation of nondigestible components of food, protection of the host from pathogenic bacteria, and regulation of immunity. Helminth infections (Trichuris trichiura, Ascaris spp. and hookworm) might have either positive or negative roles in maintaining gut homeostasis via modulation of the gut microbiota in both human and animal hosts12. In humans, Strongyloides spp. adults in the digestive tract interact with the host gut microbiota, which can have an impact on gut homeostasis13–15. A longitudinal study in non-endemic areas indicated an increase of pathogenic bacteria after treatment16, while a cross-sectional study in Thailand showed the ability of infection to increase pathogenic microorganisms14. Thus, reported effects of S. stercoralis infection appear to vary depending on study design and geographic location. A cross-sectional study has limited explanatory power to clarify the association between infection and alteration of gut microbiome. Longitudinal studies of S. stercoralis infections and follow-up after treatment provide an ideal chance to assess the effects and interactions of chronic S. stercoralis colonization on the composition of the human gut microbiota.
Among the “-omics” approaches available for the investigation of microbial communities, 16S-rRNA next-generation sequencing (16S-rRNA NGS) is a useful tool to explore complex and diverse microbial communities. However, these sequences cannot determine cause-and-effect relationships between host and infection. Determination of the functional activity of microbial communities provides direct insight at the molecular level through identification and quantification of proteins produced by the microbes. Liquid chromatography with tandem mass spectrometry (LC–MS/MS) has emerged as a new tool that allows reliable, fast, and cost-effective identification of activities of microorganisms through identification of their proteins in host fecal material17,18.
In this study, we explore the impact of natural chronic infections by S. stercoralis on the fecal microbiota of a cohort of participants in an endemic area for parasites in Northeast Thailand. Analyses of bacterial 16S-rRNA high-throughput sequencing data was applied to compare the fecal microbial profiles of positive individuals compared to negative individuals at baseline. After 1 year, we undertook 16S-rRNA NGS and fecal proteomic microbiota profiling (using LC–MS/MS) of a subset of baseline subjects with chronic infection, both prior to and following treatment with anti-helminthic treatment. This study will contribute to knowledge of the effects of chronic helminth infection before and after treatment on the gut microbial community in endemic areas. Our work provides the first metaproteomic data to detect the effects of chronic helminth infection on the structure and function of the gut microbial community in individuals.
## Study population characteristics and biochemical parameters
Supplementary Table S1 displays the demographic and biochemical information of participants at the baseline (T0). Sample were coded as Pos (positive) or Neg (negative) for S. stercoralis infection. All samples were matched at T0 between Pos and Neg groups in terms of sex, age, body-mass index, and all biochemical test parameters were within the normal range. As a result, there were no differences in blood pressure, eGFR, HbA1c, glucose, or LDL cholesterol levels between the two groups. Only eosinophilia, typically seen in parasitic infections and allergic reactions, showed a significant increase in the Pos group.
## Gut microbiome results
The gut microbiome was evaluated from 16S-rRNA sequences obtained from fecal samples. These data were treated as two separate sets for the comparisons done. The first set of comparisons was between Pos and Neg samples at baseline (T0). The second set was collected 1 + years after baseline and comparisons were made between ten samples from chronically infected individuals before (T1 or Ss+PreT) and after (T2 or Ss+PostT) treatment. In which, samples from Neg and Ss+PostT were defined to be negative with S. stercoralis infection by conventional PCR.
## Gut bacterial diversity
The intestinal microbial diversity of all fecal samples was explored using 16S-rRNA NGS data to determine whether this could be related to S. stercoralis infection. All alpha-diversity indices were slightly higher in Neg compared to Pos, but there were no significant differences between these groups according to Chao1, Shannon or Simpson indices at baseline (T0). Alpha diversity was compared between samples collected at T1 and T2 (Ss+PreT and Ss+PostT) but revealed no statistically significant differences between these two time points. Figure 1 depicts the alpha diversity with p values for all groups. Figure 1Alpha diversity of groups compared at based line between Pos and Neg (T0), and at 1 + years later between before (Ss+PreT, T1) and after (Ss+PostT, T2) presented in boxplots for Chao 1 (A,D), Shannon (B,E) and Simpson indices (C,F) with p values.
Beta diversity was evaluated using non-metric distance scaling (NMDS) analysis of the Bray–Curtis dissimilarity distance (Fig. 2a). The weighted-Unifrac distance matrix showed similar bacterial communities in participants of both groups (index = 0.11). This suggested a similar species composition between Pos and Neg samples at T0. The unweighted-Unifrac distance matrix was most efficient in detecting abundance changes in rare lineages (index = 0.64) (Fig. 2b). The result from Adonis showed no significant difference in the weighted-Unifrac distance matrix (R2 = 0.038, $$P \leq 0.106$$) between Pos and Neg at baseline, but there was a significant difference according to the unweighted-Unifrac distance matrix (R2 = 0.037, $$P \leq 0.028$$). The same analysis was also done between T1 and T2 (Ss+PreT and Ss+PostT), but there were no significant differences detected in any analyses (Supplementary Fig. S1 and Table S3).Figure 2Beta diversity of samples at T0 of 16S-rRNA sequences demonstrated the distribution of samples in red dots (Neg) and green dots (Pos) in an NMDS plot (a). Heatmap of unweighted-Unifrac distance matrix showed a presence of some bacterial taxa only in the Pos group (index = 0.64) (b).
## Relative abundance of the most-represented microbial taxa
The five most-represented phyla and ten most-represented families (based on 16S-rRNA NGS data) are shown in Fig. 3a,c. Firmicutes, Bacteroidetes, Actinobacteriota, Fusobacteriota, and Proteobacteria were the five major phyla present in all groups, especially in the infected group. Firmicutes dominated ($78.5\%$ and $63.7\%$ in Pos at T0 and Ss+PreT, respectively), followed by Bacteroidetes ($10.2\%$ and $25.2\%$ in Pos at T0 and Ss+PreT, respectively). Proteobacteria had the lowest abundance among the top five phyla ($1\%$ and $0.7\%$ in Pos and Neg, respectively). Only Proteobacteria were significantly enriched in individuals infected with S. stercoralis at baseline (Pos group) according to LEfSe analysis (Supplementary Fig. S2). At the family level (Fig. 3b,d), Lachnospiraceae, Ruminococcaceae, Prevotellaceae, Bacteroidaceae, Clostridiaceae and Peptostreptococcaceae were prominent families in the human gut microbiome. There was enrichment in the relative abundances of Fusobacteriaceae, Muribaculaceae and Oscillospiraceae at T1 and T2 (immediately before and four months after treatment), replacing Coriobacteriaceae, Erysipelatoclostridiaceae, Erysipelotrichaceae at baseline T0. However, none of the differences among the most-represented bacterial families reported here reached the level of statistical significance. Figure 3Relative abundances of the five most-represented phyla (a,c) ten most-represented families based on 16S-rRNA NGS data (b,d) in each dataset at based line between Pos and Neg (T0), and at 1 + years later between before (Ss+PreT, T1) and after (Ss+PostT, T2) treatment. Relative abundance of top 35 genera presented as heatmaps (e,f).
The 35 most-represented genera of bacteria are shown in a heatmap (Fig. 3e,f). *Five* genera diminished in abundance in S. stercoralis-infected individuals. These were Eubacterium coprostanoligenes group, Bacteroides, Coprococcus, Fusobacterium and Roseburia. *Five* genera that increased in relative abundance in S. stercoralis-infected individuals were Blautia, the Ruminococcus torques group, *Clostridium sensu* stricto 1, Erysipelotrichaceae UCG-003 and Alloprevotella. Differences at the genus level were also analyzed using LefSe (Fig. 4 and Supplementary Fig. S2). This also revealed that abundance of the R. torques group and of Alloprevotella was significantly boosted in the presence of S. stercoralis (R. torques group (Fig. 5a, $p \leq 0.05$; Fig. 5e, $$p \leq 0.048$$) and Alloprevotella (Fig. 5b, $$p \leq 0.04$$; Fig. 5f, $p \leq 0.05$). In contrast, Roseburia and the E. coprostanoligenes group exhibited significantly reduced abundance during helminth infection (Roseburia (Fig. 5c, $$p \leq 0.0057$$; Fig. 5g, $p \leq 0.05$) and E. coprostanoligenes (Fig. 5d, $$p \leq 0.0146$$; Fig. 5h, $p \leq 0.05$). Among the top four most-represented genera, R. torques showed the greatest changes in abundance between infected and uninfected individuals: abundance of this genus was reduced in nine out of ten samples after treatment. Figure 4Histograms of linear discriminant analysis (LDA) effect size (LEfSe) comparison between stool microbiota at the genus level between Ss+PreT ($$n = 10$$) and Ss+PostT ($$n = 10$$). Log-level changes in LDA score are displayed on the x axis. Green and red bars taxa found in greater relative abundance in Ss+PreT and Ss+PostT, respectively. Figure 5Top four genera significantly differs between uninfected groups and those with chronic infection of S. stercoralis based on 16S-rRNA sequences. Data at T0 are presented as box plots [(a,b,c,d) are presented for Ruminococus torques group, Alloprevotella, Roseburia, Eubacterium coprostanoligenes group, respectively] and those of the before and after treatment groups (T1 and T2) are demonstrated as lines connecting points Ss+PreT and Ss+PostT [(e,f,g,h) are presented for Ruminococus torques group, Alloprevotella, Roseburia, Eubacterium coprostanoligenes group, respectively].
## Fecal proteins derived from members of the ten most-abundant bacterial families before and after treatment of chronic strongyloidiasis
To compare levels of protein expression in the ten most-abundant bacterial families between before (Ss+PreT) and after treatment (Ss+PostT) of chronic strongyloidiasis, the maximum intensity of protein in each sample was counted to determine total number of proteins representation at the family level. A total of 52,798 proteins, including uncharacterized proteins, derived from the top 10 families were assessed (Table 1). The relative decrease in protein numbers in the top 10 bacterial families after treatment is shown in the heatmap (Fig. 6a). However, there were no significant differences among all families after applying false discovery rate, FDR test (FDR = $1\%$) (Table 1).Table 1Summary of number of fecal proteins ascribed to the ten most abundant bacterial families. Adjusted p value < 0.01 is considered as indicating a significant difference (n = sample size of each group).FamilySs+PreT ($$n = 10$$)Ss+PostT ($$n = 10$$)Adjusted p valueBacteroidaceae2724.6 ± 419.42291 ± 211.30.012Clostridiaceae2987.2 ± 458.62468.4 ± 250.60.012Erysipelatoclostridiaceae2890.4 ± 449.92435.7 ± 228.10.012Fusobacteriaceae2915.7 ± 454.12432.5 ± 226.80.012Lachnospiraceae2809.5 ± 442.32335.1 ± 224.20.012Muribaculaceae2659.8 ± 4072198.3 ± 215.50.012Oscillospiraceae3030.1 ± 470.62498.6 ± 237.10.012Peptostreptococcaceae2907.1 ± 467.72459 ± 243.50.015Prevotellaceae3005.9 ± 460.72462.9 ± 257.20.012Ruminococcaceae2653.7 ± 422.82214.3 ± 213.30.012Figure 6Metaproteomic analysis result. ( a) Heatmap of abundance of fecal proteins derived from the top 10 bacterial families for ten individuals both before (Ss+PreT (T1)) and after ivermectin treatment (Ss+PostT (T2)). ( b) Relative levels of the beta-N-acetylglucosaminidase enzyme of Ruminococcus torques expressed in Ss+PreT and Ss+PostT groups by log2 max intensity peptide count. Data are presented as line graphs pre- and post-treatment of individual samples ($\frac{7}{10}$ samples contained this protein) and as a bar chart. Each column shows data from a single individual. ( c) Jvenn diagram showing numbers of highly expressed proteins (HEPs) in Ss+PreT and Ss+PostT groups and (d) functional categories of HEPs in Ss+PreT in a pie chart.
## Differentially expressed proteins compared before and after treatment of chronic strongyloidiasis
Differentially expressed proteins (DEPs) analysis with FDR testing (p-adjusted < 0.05) revealed no significant differences among the top ten families. Ruminococcus torques was conspicuously the bacterium with the most-altered abundance during S. stercoralis infection according to 16S-rRNA NGS data. This taxon is well-known for its mucin-degrading properties. All metaproteomic data of Ruminococcus spp. were therefore examined for the presence of proteins related to mucus glycoproteins without FDR testing. We found one protein, named F$\frac{5}{8}$ type C domain-containing protein of the R. torques group, which was classified by GO function as having beta-N-acetylglucosaminidase activity [GO:0016231] (Supplementary file Excel 1). This protein was significantly reduced after treatment by ivermectin (seven out of ten samples contained this protein (see Fig. 6b)).
## Functional categories of highly expressed proteins in chronic S. stercoralis infection
Because DEPs analysis with FDR testing (p-adjusted < 0.05) revealed no significant differences among the top ten families, we focused on highly expressed proteins (HEPs) in chronic S. stercoralis infection. Only proteins present in at least $50\%$ of samples in each group were selected to be analyzed by Venn diagrams (Fig. 6c). A total of 1335 proteins were highly represented in Ss+PreT samples across all 10 bacterial families (Supplementary Excel file 2). Among these, 647 HEPs were assigned to biological processes according to Uniprot and subcategorized into different functions (see Fig. 6d and Supplementary Excel file 3). Proteins associated with carbohydrate metabolism were markedly enriched in samples from people with S. stercoralis infection (Fig. 6d) and most of these were produced by members of the Bacteroidaceae. Indeed, the most-represented category of HEPs produced by this family of bacteria was associated with carbohydrate metabolism (Supplementary file Excel 3). Thus, proteins produced by this family were selected for further analysis using STITCH, focusing on carbohydrate metabolism.
## STITCH analysis network of HEPs of Bacteroidaceae during chronic S. stercoralis infection
The interesting proteins from Bacteroidaceae associated with carbohydrate metabolism were subjected to STITCH analysis to identify the enrichment pathway in S. stercoralis infection. The results demonstrated that carbohydrate metabolism from this family of bacteria is mainly involved in cellulose metabolism (Supplementary Fig. S3 and Table S4). Humans cannot digest cellulose, but intestinal bacteria do so to produce short-chain fatty acids (SCFAs) such as acetate. HEPs assigned to carbohydrate and fatty-acid metabolism were highly represented in Ss+PreT samples. STITCH analysis revealed interactions between known or putative proteins of Ss+PreT and acetate production (*Enterococcus casseliflavus* was used as a template) (see Fig. 7 and Supplementary file Excel 4).Figure 7STITCH analysis network of HEPs and SCFA molecules with S. stercoralis infection. Know/putative proteins among the highly expressed proteins (HEPs) of the Ss+PreT group of samples are marked in red boxes.
## Discussion
In this study, we investigated the impact of S. stercoralis infection on gut microbiota in participants considered otherwise healthy in a community at Nam Phong and Ubonrat Districts, Khon Kaen Province, Northeast Thailand. Longitudinal evaluation during chronic infection with S. stercoralis and following deworming treatment allowed us to demonstrate the real effect of S. stercoralis on the host’s gut microbiome. This is also the first study to explore metaproteomic-based analysis in research on the human gut microbiome after deworming. Based on this, we discovered that helminth infection alters not only the composition of the gut microbiome but also the metabolic function of that community. In our study, ivermectin was given as a single dose, so the alterations in our study, especially four months posttreatment, were considered a consequence of worm clearance rather than of the anthelmintic treatment itself11.
Our research focused on otherwise healthy individuals with chronic mono-infection with S. stercoralis in endemic areas and did not find any significant alteration in alpha diversity. Intestinal helminths inhabiting the large intestine, have a stronger effect on gut microbiota compared to those in the small intestine11. In our study, chronic S. stercoralis infection, persisting in the small intestine with a low worm burden (no larvae were found by FECT at T1) might not significantly alter measures of gastrointestinal microbial diversity. For beta diversity, we found only dissimilarity of taxonomic content between Pos and Neg groups at baseline (T0) according to the unweighted-Unifrac distance matrix, demonstrating the presence of some specific bacterial taxa associated with parasite infection. Unweighted-UniFrac is more sensitive to differences in low-abundance features. Other reports from Thailand concerning soil-transmitted helminths (Ascaris sp, hookworm and Trichiuris trichiura) in children showed similar results with no difference between infected and uninfected groups in bacterial alpha-diversity but did observe differences in beta-diversity19. In previous research we conducted in Don Chang District, Khon Kaen Province, alpha diversity was found to be greater in infected individuals. Another recent study in Thailand showed that bacterial diversity differed between subjects from different schools, possibly due to inter-school variation in diet19. Thus, different districts and time-points of sample collection could affect gut diversity in the same general area. The impact of helminths on the gut microbiota likely varies due to the large range of environmental variables (diet, sanitation, species of parasite, worm burden, etc.) and the batch effect produced by sample numbers and collection11.
Analysis of the 35 most-represented bacterial genera demonstrated many shifts in relative abundance of these between *Pos versus* Neg at baseline and 1 + years later between the Ss+PreT and Ss+PostT groups. Among these changes, the Ruminococcus torques group considerably increased in S. stercoralis infection. This taxon is enriched in various clinical situations such as visceral obesity, hemodialysis, and irritable bowel syndrome (IBS) or children’s autism-spectrum disorders20–23. In contrast, depletion of Ruminococcus during S. stercoralis infection has been reported in another study in a non-endemic area15. Our previous study with S. stercoralis infection in Donchang did not find a similar alteration in the R. torques group14. Different districts, the time of sample collection, and varying levels of infection could all account for the variation in the findings. Furthermore, in our previous study, samples from 26 individuals had been pooled whereas we analyzed samples individually in this study. In addition, we also evaluated samples from the same ten individuals at T0 and T1 to investigate alterations in bacterial genera, particularly the R. torques group, during chronic infection. However, in general, there was considerable difference in bacterial genera present at T0 and T1 (data not shown). Factors other than infection could have been affecting R. torques populations during the long and enforced gap between sample collections during the COVID-19 pandemic24. Thus, in this study, we decided to analyze our material as two separate sets of 16S-rRNA, one at baseline (T0), comparing Pos and Neg, and the other at T1 plus T2, comparing chronically infected individuals before and after worm clearance.
The Ruminococcus torques group became enriched during chronic S. stercoralis infection, more so than any other taxon according to the 16S-rRNA NGS data, and showed a significant reduction after treatment in nine out of ten samples (Fig. 4). Members of this group are known to produce substances that strongly degrade gastrointestinal mucin and play a role in the pathophysiology of Crohn’s disease25–27. Mucin is a major structural and functional component present in the mucus at concentration of 1–$5\%$28. Large amounts of mucus in stool are associated with diarrhea in conditions such as Crohn’s disease or IBS29. Similarly, infections with S. stercoralis or other helminths also increase mucus production with frequent diarrhea. The mucosal integrity of the entire gastrointestinal (GI) tract is vital for maintaining health. Given the known mucin-degrading ability of the R. torques group, we investigated in more detail the metaproteomic data from this taxon. One protein linked to beta-N-acetylglucosaminidase from R. torques CAG 126:1 was found to decrease significantly after treatment (Fig. 6b). Beta-N-acetylglucosaminidase is putatively involved in mucin degradation28. Mucus provides a crucial innate defense against invading gastrointestinal parasitic nematodes30,31, so we hypothesize that an increase of R. torques during chronic infection might aid prolonged worm survival in the host by degradation of mucus. An increase of the R. torques group may facilitate the establishment of new parthenogenetic females because of reduced mucus-mediated expulsion. While the same group of mucus-associated bacteria28 has been found in Crohn’s disease, there is a negative correlation between A. muciniphilia and R. torques25,32. Unlike Akkermansia, a well-known beneficial bacterial taxon33,34, an increase of R. torques has been highlighted in other diseases25–29. Further studies should be done to confirm the important roles of R. torques during in chronic S. stercoralis infection.
In this study, we also analyzed metaproteomic data before and after treatment of chronic S. stercoralis infection. There remain many knowledge gaps in our ability to evaluate effects of worm clearance on gastrointestinal microbes. Most studies have used 16S-rRNA sequences, which is unable to provide information about the functional roles of the bacterial community. Metaproteomics serves as a novel useful tool to investigate treatment effects on human gut microbes and has been applied here for the first time. Because no differentially expressed proteins (DEPs) were found by the TM4 suite of software, we focused on 1335 highly expressed bacterial proteins (HEPs) that were present in participants prior to treatment. We found that many of these were associated with carbohydrate metabolic function (Fig. 6d). This was especially the case for members of the Bacteroidaceae. A previous study using metagenomic profiles also predicted an enrichment of carbohydrate metabolism during infection with S. stercoralis15. HEPs present in helminth infection and originating from members of the Bacteroidaceae were recruited for STITCH analysis linked to glucose metabolism derived from cellulose digestion. Cellulose is a major building block of plant cell walls, consisting of molecules linked together into solid fibers. For humans, cellulose is indigestible, so gut bacterial enzymes are required to break down this material. One of the final products from cellulose digestion of bacteria is short-chain fatty acids (SCFAs). Bacterial interactions, particularly cross-feeding, which is the use of substrates or metabolites generated by other bacteria, are important in determining the final amounts of SCFAs in the colon because bacterial taxa do not act in isolation35. Many of the HEPs related to carbohydrate metabolism from infected individuals identified in this study were highly represented among enzymes related to acetate according to our STITCH network. When compared to other SCFAs, acetate often reaches the highest levels in the colon and can be produced by many gut microbial species35. In contrast, analysis of fecal samples found higher levels of acetate in non-infected subjects than in S. stercoralis-infected subjects in non-endemic areas13. Relevantly, decreased serum acetate level was also observed in strongyloidiasis in an endemic area14. Contradictory findings of acetate level between serum and fecal samples may be due to a variety of other factors related to SCFA production, not only S. stercoralis infection16,35–37. In addition, Bacteroidaceae also harbors several mucin-degrading groups and several of the glycosyl hydrolases identified in Supplementary Table S4 degrade dietary components as well as mucin glycans38,39. This result supports our hypothesis, mentioned earlier, that altered mucin profiles may support parasite persistence.
In conclusion, chronic S. stercoralis infection in infected participants at Nam Phong and Ubonrat districts did not lead to altered gut bacterial diversity. When compared to other taxa, the R. torques group has the largest alteration in abundance during chronic S. stercoralis infection. Interestingly, this, together with an increase in mucin degrader enzyme produced by this taxon, may facilitate prolonged worm persistence. Furthermore, anti-helminthic treatment and worm clearance with ivermectin depleted carbohydrate metabolism, with Bacteroidaceae being the primary culprits. Acetate production was likely to be increased during S. stercoralis infection based on STITCH analysis networks. Taken together, our results indicate that chronic S. stercoralis infection not only changes the human gut microbiota but also alters metabolic homeostasis.
## Human ethics statement
This study was approved by the Human Ethical Review Committee of Khon Kaen University, Thailand (HE641434) following the principles of the Declaration of Helsinki. Before stool samples were collected, participants were required to complete and sign written informed-consent forms.
## Participants and samples
All samples came from residents of Nam Phong and Ubonrat Districts, Khon Kaen Province, Northeast Thailand, under the Chronic Kidney Disease Northeastern Thailand (CKDNET) program. Intestinal parasitic infections were diagnosed from stool samples using the modified agar plate culture (mAPC) method and the formalin-ether concentration technique (FECT) as previously reported40. The inclusion criteria for the current investigation were infection with S. stercoralis alone (no additional intestinal parasites) and no recent usage of antibiotics or ivermectin as determined by questionnaires. At the times of sample collection, none of the patients was using medications such as corticosteroids, immunomodulators, or biological agents and were apparently otherwise healthy based on the questionnaires. None of the patients had underlying conditions such as diabetes and hypertension. The summary of biochemical tests of participants and sample details at each time-point are presented in Supplementary Tables S1 and S2, respectively.
The baseline samples we used were leftover fecal specimens from 42 individuals collected as part of an earlier study, in which a single stool sample was screened from each of 704 subjects40. From among these 704 subjects, we selected 21 positive cases and 21 matched negative cases. Of these specimens (baseline, or time-point T0), the samples were matched between groups by gender, age and were in the normal range for all biochemical tests done. Infected individuals at T0 could not be treated following this initial diagnosis because of the extended lockdowns imposed by the Thai Government as a result of the COVID19 pandemic. A stool sample was collected from each of the positive subjects 12 months later (T1), and examined using both diagnostic methods. Ten of these were collected and found to still be infected with S. stercoralis with low intensity of infection (no larvae were detected by FECT but larvae were still found using the agar plate technique). No other species of parasite was found. Since no subjects had previously received any ivermectin treatment, we regarded them as suffering chronic infection. Those fecal samples were coded as “Ss+PreT” and their donors were treated with ivermectin (200 μg/kg body weight). Around four months later (time-point T2), fecal samples were again obtained from these individuals and coded as “Ss+PostT”. Fecal examination showed all ten to be free of S. stercoralis infection. After being collected in the field, fecal samples (Ss+PreT and Ss+PostT) were quickly transferred to ice and frozen at -80 °C until analysis.
## Polymerase chain reaction (PCR) confirmation of presence or absence of S. stercoralis infection
Conventional PCR was used to confirm the findings of fecal examination, including for samples that were fecal-negative. Primers were those used in a previous study to amplify a 230-bp fragment from a dispersed repetitive sequence of S. stercoralis, GenBank accession no. M84229.141. The forward primer was 5′-CCGGACACTATAAGGATTGA-3′, and the reverse was 5′-ACAGACCTGTTATCGCTCTC-3′. The amplification profile was initial denaturation at 94 °C for 5 min; and 35 cycles at 94 °C for 40 s, 52.8 °C for 30 s, 72 °C for 2 min; and a final extension at 72 °C for 10 min. To confirm amplification and amplicon size, the PCR products were resolved on a $1.5\%$ agarose gel stained with a non-toxic dye.
## Fecal DNA extraction and Illumina sequencing
DNA was extracted from all 62 stool samples using QIAamp PowerFecal Pro DNA Kit (QIAGEN, Hilden, Germany) according to the manufacturer’s instructions. Samples were checked by agarose gel electrophoresis and an Agilent5400 fragment-analyzer system before library construction. The V3-V4 region of the prokaryotic 16S-rRNA gene was PCR-amplified and sequenced on an Illumina platform. In summary, targeted regions were PCR-amplified using specific primers connected to barcodes42. PCR products of the proper size were selected by agarose gel electrophoresis and equal quantities from each sample were pooled and Illumina adapters ligated after being end-repaired. The library was examined using a Qubit fluorometer, real-time PCR, and Bioanalyzer to check size distribution and for quantification. Libraries were sequenced on a paired-end Illumina platform to generate 250-bp paired-end raw reads.
## Sample preparation for shotgun proteomics
A volume of 500 µl of $0.5\%$ sodium dodecyl sulphate (SDS) was added to 100 mg feces, mixed well by pipetting, vortexed and centrifuged at 10,000g for 15 min. The supernatant was transferred to a new tube, mixed well with 2 volumes of cold acetone, and incubated overnight at − 20 °C. The mixture was centrifuged at 10,000g for 15 min to remove the supernatant; the protein pellet, after precipitation, was then dried and stored at − 80 °C prior to analysis. The protein concentration of each sample was then determined by the Lowry assay using bovine serum albumin as a standard protein43. Disulfide bonds were broken down in the presence of 10 mM dithiothreitol in 10 mM ammonium bicarbonate, and the reformation of disulfide bonds in the proteins was prevented by alkylation with 30 mM iodoacetamide in 10 mM ammonium bicarbonate. Sequencing-grade porcine trypsin was used to digest the protein samples for 16 h at 37 °C. For analysis by nano-liquid chromatography tandem mass spectrometry (nano LC–MS/MS), the tryptic peptides were dried with a speed vacuum concentrator and resuspended in $0.1\%$ formic acid.
## Liquid chromatography-tandem mass spectrometry (LC/MS–MS)
Thermo Scientific's Ultimate3000 Nano/Capillary LC System (Thermo Scientific, UK) was used to analyze the tryptic peptide samples. It was connected to a hybrid quadrupole Q-ToF impact II (Bruker Daltonics Ltd; Hamburg, Germany) with a Nano-captive spray ion source. Peptide digests were packed with Acclaim PepMap RSLC C18, 2 µm, 100 Å, and nanoViper after being enriched on a µ-Precolumn 300 m i.d. X 5 mm C18 PepMap 100, 5 µm, 100 Å (Thermo Scientific, UK) (Thermo Scientific, UK). A thermostated column oven with a 60 °C setting was used to surround the C18 column. On the analytical column, solvents A and B. containing $0.1\%$ formic acid in water and $0.1\%$ formic acid in $80\%$ acetonitrile, respectively, were provided. The peptides were eluted using a gradient of 5–$55\%$ solvent B for 30 min at a constant flow rate of 0.30 µl/min. The CaptiveSpray was used for 1.6 kV electrospray ionization. About 50 l/h of nitrogen was employed as a drying gas. When nitrogen gas served as the collision gas, collision-induced-dissociation (CID) product ion mass spectra were obtained. Positive-ion mode mass spectra (MS) and MS/MS spectra were obtained at 2 Hz over the (m/z) 150–2200 range. Based on the m/z value, the collision energy was modified to 10 eV. Each sample was examined in triplicate using LC–MS.
## 16S-rRNA NGS bioinformatic analysis
Raw data were further processed using Quantitative Insights into Microbial Ecology 2 (QIIME2) version qiime2-2021.11, a software package that performs microbial community analysis and taxonomic classification of microbial genomes44. Demultiplexing was conducted to remove primer sequences and paired-end reads were assigned to individual samples based on their unique barcode. Quality filtering on the raw reads was performed under specific filtering conditions (Q Score > 33) to obtain the high-quality clean reads. Denoised paired-end reads, after quality filtering by DADA2, were clustered into operational taxonomic units (OTUs) based on a criterion of $97\%$ similarity. Naïve Bayes Classifiers of the Silva full-length database were applied to classify taxonomy. For diversity analysis, samples were normalized so that all could be compared. Alpha diversity of OTU libraries was described using the Chao1, Simpson and Shannon indices using Kruskal–Wallis tests. Non-metric distance scaling (NMDS) analysis of the Bray–Curtis dissimilarity distance of samples among groups. The distance matrices were constructed using the unweighted and weighted UniFrac algorithms in QIIME2 from the whole-community phylogenetic tree. Adonis in QIIME2 was applied to calculate statistically significant differences for beta diversity among groups. Gut diversity was visualized using R studio version 2022.7.2.576 (http://www.rstudio.com/). The most abundant bacteria phyla and families were entered into Excel spreadsheets. The heatmap showing the top 35 bacterial genera was constructed using GraphPad Prism version 8.4 (www.graphpad.com).
A linear discriminant analysis (LDA) effect size (LEfSe) score criterion of > 2.0 was used to identify features that were significantly different in abundance between the groups in order to identify possible biomarker OTUs45. All significance thresholds were set at a two-sided p value of 0.05. Data of population characteristics and bacterial compositions (p value) were analyzed and visualize using GraphPad Prism version 8.4 and Excel.
## Metaproteomic analysis of samples before and after treatment
Proteins in individual samples were bioinformatically quantified by MaxQuant 2.1.0.0 using the Andromeda search engine to correlate MS/MS spectra to proteins produced by the ten most abundant bacterial families (Lachnospiraceae, Ruminococcaceae, Prevotellaceae, Bacteroidaceae, Fusobacteriaceae, Muribaculaceae, Peptostreptococcaceae, Oscillospiraceae, Clostridiaceae, Erysipelatoclostridiaceae) and one particular genus (Ruminococcus). The Uniprot database, which was accessed on 15 June 2022, was used for this. Label-free quantitation with MaxQuant’s standard settings was performed using a maximum of two missed cleavages, mass tolerance of 0.6 Dalton for the main search, trypsin as the digesting enzyme, carbamidomethylation of cysteine as a fixed modification, and oxidation of methionine and acetylation of the protein N-terminus as variable modifications. Peptides with a minimum of seven amino acids and at least one unique peptide were considered for protein identification and used for further data analysis. The protein FDR was set at $1\%$. The maximum number of modifications per peptide was set to 5. Peptides with maximum intensity from three injections were detected as spectral data of the total proteins expressed from each taxon.
Maximum peptide intensities were log2 transformed in Microsoft Excel, providing the protein expression levels for quantification of protein numbers and analysis of differentially expressed proteins (DEPs) in the ten most abundant bacterial families. Multiple t-tests using the two-stage linear step-up procedure of Benjamini, Krieger and Yekutieli46 (false discovery rate, FDR = $1\%$), carried out in GraphPad Prism, were applied to analysis of number of proteins of top 10 bacterial families of the 2 groups at T1 and T2. Numbers of proteins from each family were visualized in a heatmap using Morpheus from the Broad Institute (https://software.broadinstitute.org/morpheus). Log2 of max protein intensity from each sample was then subjected to statistical analysis to determine DEPs by using the MultiExperiment Viewer (MeV) in the TM4 suite software (adjusted p value < 0.05) in the 10 selected bacterial families47. The Wilcoxon rank-sum test with multiple-testing FDR correction was performed. The same analysis was applied for the genus Ruminococcus alone without applying a FDR correction. Only proteins found in at least $50\%$ of all samples in each group were further selected for analysis of highly expressed proteins (HEPs) using the Jvenn viewer48. Proteins found only in Ss+ PreT were considered as enriched proteins or HEPs. Those HEPs, present during chronic S. stercoralis infection, were classified further into different metabolic categories through GO biological function categories from Uniprot. The STITCH database version 5 was used to forecast functional interaction networks between identified proteins and small molecules and to generate figures (http://stitch.embl.de/)49.
## Supplementary Information
Supplementary Information 1.Supplementary Information 2. The online version contains supplementary material available at 10.1038/s41598-023-31118-5.
## References
1. Buonfrate D. **The global prevalence of**. *Pathogens* (2020) **9** 468. DOI: 10.3390/pathogens9060468
2. Grove DI. **Human strongyloidiasis**. *Adv. Parasitol.* (1996) **38** 251-309. DOI: 10.1016/s0065-308x(08)60036-6
3. Kalambay JD, Zaman R, Zaman MH, Zaman T. **Twenty-five years of chronic strongyloidiasis in an immigrant**. *Clin. Med. Insights Case Rep.* (2017) **10** 1179547616684828. DOI: 10.1177/1179547616684828
4. Neva FA. **Biology and immunology of human strongyloidiasis**. *J. Infect. Dis.* (1986) **153** 397-406. DOI: 10.1093/infdis/153.3.397
5. Sithithaworn P. **Epidemiology of**. *Trans. R. Soc. Trop. Med. Hyg.* (2003) **97** 398-402. DOI: 10.1016/s0035-9203(03)90069-1
6. Jongsuksuntigul P. **Prevalence of**. *J. Med. Assoc. Thail.* (2003) **86** 737-741
7. Laoraksawong P. **Current high prevalences of**. *BMC Public Health* (2018) **18** 940. DOI: 10.1186/s12889-018-5871-1
8. Marcos LA, Terashima A, Canales M, Gotuzzo E. **Update on strongyloidiasis in the immunocompromised host**. *Curr. Infect. Dis. Rep.* (2011) **13** 35-46. DOI: 10.1007/s11908-010-0150-z
9. Barroso M, Salvador F, Sanchez-Montalva A, Bosch-Nicolau P, Molina I. *PLoS Negl Trop Dis* (2019) **13** e0007230. DOI: 10.1371/journal.pntd.0007230
10. Luvira V. *Trop. Med. Infect. Dis.* (2022) **7** 310. DOI: 10.3390/tropicalmed7100310
11. Kupritz J, Angelova A, Nutman TB, Gazzinelli-Guimaraes PH. **Helminth-induced human gastrointestinal dysbiosis: A systematic review and meta-analysis reveals insights into altered taxon diversity and microbial gradient collapse**. *MBio* (2021) **12** e0289021. DOI: 10.1128/mBio.02890-21
12. Brosschot TP, Reynolds LA. **The impact of a helminth-modified microbiome on host immunity**. *Mucosal. Immunol.* (2018) **11** 1039-1046. DOI: 10.1038/s41385-018-0008-5
13. Jenkins TP. **Author Correction: A comprehensive analysis of the faecal microbiome and metabolome of**. *Sci. Rep.* (2019) **9** 8571. DOI: 10.1038/s41598-019-43508-9
14. Nguyen HT. **Investigation of gut microbiota and short-chain fatty acids in**. *Biosci. Microbiota Food Health* (2022) **41** 121-129. DOI: 10.12938/bmfh.2021-054
15. Pane S, Sacco A, Iorio A, Romani L, Putignani L. *Int. J. Mol. Sci.* (2021) **22** 2131. DOI: 10.3390/ijms22042131
16. Lin CH, Chen SJ, Kuo CH. **Author Response: Association of fecal and plasma levels of short-chain fatty acids with gut microbiota and clinical severity in patients with Parkinson disease**. *Neurology* (2022) **99** 404. DOI: 10.1212/WNL.0000000000201131
17. Salvato F, Hettich RL, Kleiner M. **Five key aspects of metaproteomics as a tool to understand functional interactions in host-associated microbiomes**. *PLoS Pathog.* (2021) **17** e1009245. DOI: 10.1371/journal.ppat.1009245
18. Xiong W, Abraham PE, Li Z, Pan C, Hettich RL. **Microbial metaproteomics for characterizing the range of metabolic functions and activities of human gut microbiota**. *Proteomics* (2015) **15** 3424-3438. DOI: 10.1002/pmic.201400571
19. Stracke K. **Field evaluation of the gut microbiome composition of pre-school and school-aged children in Tha Song Yang, Thailand, following oral MDA for STH infections**. *PLoS Negl. Trop. Dis.* (2021) **15** e0009597. DOI: 10.1371/journal.pntd.0009597
20. Shi X. **Alterations of gut microbial pathways and virulence factors in hemodialysis patients**. *Front. Cell Infect. Microbiol.* (2022) **12** 904284. DOI: 10.3389/fcimb.2022.904284
21. Yan H. **Gut microbiome alterations in patients with visceral obesity based on quantitative computed tomography**. *Front Cell Infect Microbiol* (2021) **11** 823262. DOI: 10.3389/fcimb.2021.823262
22. Wang L. **Increased abundance of**. *Mol. Autism* (2013) **4** 42. DOI: 10.1186/2040-2392-4-42
23. Malinen E. **Association of symptoms with gastrointestinal microbiota in irritable bowel syndrome**. *World J. Gastroenterol.* (2010) **16** 4532-4540. DOI: 10.3748/wjg.v16.i36.4532
24. Hasan N, Yang H. **Factors affecting the composition of the gut microbiota, and its modulation**. *PeerJ* (2019) **7** e7502. DOI: 10.7717/peerj.7502
25. Halmos EP. **Consistent prebiotic effect on gut microbiota with altered FODMAP intake in patients with Crohn's Disease: A randomised, controlled cross-over trial of well-defined diets**. *Clin. Transl. Gastroenterol.* (2016) **7** e164. DOI: 10.1038/ctg.2016.22
26. Kwak MS, Cha JM, Shin HP, Jeon JW, Yoon JY. **Development of a novel metagenomic biomarker for prediction of upper gastrointestinal tract involvement in patients With Crohn's disease**. *Front. Microbiol.* (2020) **11** 1162. DOI: 10.3389/fmicb.2020.01162
27. Takahashi K. **Reduced abundance of butyrate-producing bacteria species in the fecal microbial community in Crohn's disease**. *Digestion* (2016) **93** 59-65. DOI: 10.1159/000441768
28. Tailford LE, Crost EH, Kavanaugh D, Juge N. **Mucin glycan foraging in the human gut microbiome**. *Front. Genet.* (2015) **6** 81. DOI: 10.3389/fgene.2015.00081
29. Nobrega VG. **The onset of clinical manifestations in inflammatory bowel disease patients**. *Arq. Gastroenterol.* (2018) **55** 290-295. DOI: 10.1590/S0004-2803.201800000-73
30. Onah DN, Nawa Y. **Mucosal immunity against parasitic gastrointestinal nematodes**. *Korean J. Parasitol.* (2000) **38** 209-236. DOI: 10.3347/kjp.2000.38.4.209
31. Iriemenam NC, Sanyaolu AO, Oyibo WA, Fagbenro-Beyioku AF. *Parasitol. Int.* (2010) **59** 9-14. DOI: 10.1016/j.parint.2009.10.009
32. Png CW. **Mucolytic bacteria with increased prevalence in IBD mucosa augment in vitro utilization of mucin by other bacteria**. *Am. J. Gastroenterol.* (2010) **105** 2420-2428. DOI: 10.1038/ajg.2010.281
33. Shin J. **Elucidation of**. *Front. Microbiol.* (2019) **10** 1137. DOI: 10.3389/fmicb.2019.01137
34. van der Lugt B. *Immun. Ageing* (2019) **16** 6. DOI: 10.1186/s12979-019-0145-z
35. Blaak EE. **Short chain fatty acids in human gut and metabolic health**. *Benef. Microbes* (2020) **11** 411-455. DOI: 10.3920/BM2020.0057
36. Fernandes J, Vogt J, Wolever TM. **Kinetic model of acetate metabolism in healthy and hyperinsulinaemic humans**. *Eur. J. Clin. Nutr.* (2014) **68** 1067-1071. DOI: 10.1038/ejcn.2014.136
37. Werneck-Silva AL. **Intestinal permeability in strongyloidiasis**. *Braz. J. Med. Biol. Res.* (2001) **34** 353-357. DOI: 10.1590/s0100-879x2001000300009
38. Ravcheev DA, Thiele I. **Comparative genomic analysis of the human gut microbiome reveals a broad distribution of metabolic pathways for the degradation of host-synthetized mucin glycans and utilization of mucin-derived monosaccharides**. *Front. Genet.* (2017) **8** 111. DOI: 10.3389/fgene.2017.00111
39. Glover JS, Ticer TD, Engevik MA. **Characterizing the mucin-degrading capacity of the human gut microbiota**. *Sci. Rep.* (2022) **12** 8456. DOI: 10.1038/s41598-022-11819-z
40. Yingklang M. **Association of**. *PLoS ONE* (2022) **17** e0269080. DOI: 10.1371/journal.pone.0269080
41. Iamrod K. **Development and efficacy of droplet digital PCR for detection of**. *Am. J. Trop. Med. Hyg.* (2021) **106** 312-319. DOI: 10.4269/ajtmh.21-0729
42. Itthitaetrakool U. **Chronic**. *PLoS ONE* (2016) **11** e0165798. DOI: 10.1371/journal.pone.0165798
43. Lowry OH, Rosebrough NJ, Farr AL, Randall RJ. **Protein measurement with the Folin phenol reagent**. *J Biol Chem* (1951) **193** 265-275. DOI: 10.1016/S0021-9258(19)52451-6
44. Navas-Molina JA. **Advancing our understanding of the human microbiome using QIIME**. *Methods Enzymol.* (2013) **531** 371-444. DOI: 10.1016/b978-0-12-407863-5.00019-8
45. Segata N. **Metagenomic biomarker discovery and explanation**. *Genome Biol.* (2011) **12** R60. DOI: 10.1186/gb-2011-12-6-r60
46. Benjamini Y, Krieger AM, Yekutieli D. **Adaptive linear step-up procedures that control the false discovery rate**. *Biometrika* (2006) **93** 491-507. DOI: 10.1093/biomet/93.3.491%JBiometrika
47. Howe EA, Sinha R, Schlauch D, Quackenbush J. **RNA-Seq analysis in MeV**. *Bioinformatics* (2011) **27** 3209-3210. DOI: 10.1093/bioinformatics/btr490
48. Bardou P, Mariette J, Escudie F, Djemiel C, Klopp C. **jvenn: An interactive Venn diagram viewer**. *BMC Bioinformatics* (2014) **15** 293. DOI: 10.1186/1471-2105-15-293
49. Szklarczyk D. **STITCH 5: augmenting protein-chemical interaction networks with tissue and affinity data**. *Nucleic Acids Res.* (2016) **44** D380-384. DOI: 10.1093/nar/gkv1277
|
---
title: A protein subunit vaccine elicits a balanced immune response that protects
against Pseudomonas pulmonary infection
authors:
- Debaki R. Howlader
- Sayan Das
- Ti Lu
- Rahul Shubhra Mandal
- Gang Hu
- David J. Varisco
- Zackary K. Dietz
- Siva Sai Kumar Ratnakaram
- Robert K. Ernst
- William D. Picking
- Wendy L. Picking
journal: NPJ Vaccines
year: 2023
pmcid: PMC10012293
doi: 10.1038/s41541-023-00618-w
license: CC BY 4.0
---
# A protein subunit vaccine elicits a balanced immune response that protects against Pseudomonas pulmonary infection
## Abstract
The opportunistic pathogen *Pseudomonas aeruginosa* (Pa) causes severe nosocomial infections, especially in immunocompromised individuals and the elderly. Increasing drug resistance, the absence of a licensed vaccine and increased hospitalizations due to SARS-CoV-2 have made *Pa a* major healthcare risk. To address this, we formulated a candidate subunit vaccine against Pa (L-PaF), by fusing the type III secretion system tip and translocator proteins with LTA1 in an oil-in-water emulsion (ME). This was mixed with the TLR4 agonist (BECC438b). Lung mRNA sequencing showed that the formulation activates genes from multiple immunological pathways eliciting a protective Th1-Th17 response following IN immunization. Following infection, however, the immunized mice showed an adaptive response while the PBS-vaccinated mice experienced rapid onset of an inflammatory response. The latter displayed a hypoxic lung environment with high bacterial burden. Finally, the importance of IL-17 and immunoglobulins were demonstrated using knockout mice. These findings suggest a need for a balanced humoral and cellular response to prevent the onset of Pa infection and that our formulation could elicit such a response.
## Introduction
Pseudomonas aeruginosa (Pa)1,2 is a significant cause of nosocomial infections in health care settings. Immunocompromised individuals, burn victims, cystic fibrosis patients, and cancer patients are at a further elevated risk of infection by Pa3,4. A wide range of virulence factors are used by this opportunistic pathogen to initiate infection and subsequently adapt to its environment leading to chronic biofilm formation and acquisition of drug-resistance5. Recent reports have also acknowledged a significant increase in the identification of Multi-Drug Resistant (MDR) Pa strains and Extremely Drug Resistant (XDR) Pa strains5. The Centers for Disease Control and Prevention estimated attributable healthcare costs of $757 M with nearly 32,600 cases of MDR Pa infection in 20176. Furthermore, ventilator-associated pneumonia (VAP) caused by *Pa is* a common occurrence in 3–$5\%$ of adults ventilated for more than 2 days7. Likewise, Pa infection was the most common infection in military troops in 20162. Although certain diseases and occupational health hazards increase the risking for Pa infection, aging is the most common factor responsible for a lethal Pa infection2.
Faced with the elevated risks posed by MDR and XDR Pa, the most efficient way to move forward is to develop a broadly protective prophylactic vaccine to prevent the onset of infection. Several approaches have been used in the past, but no licensed Pa vaccine has reached the market. Vaccine formulation incorporating LPS O-antigen8, outer membrane proteins9, live-attenuated vaccines10,11, or whole cell vaccines12 are being developed, yet there is still no licensed Pa vaccine. Toward this end, type-III secretion system (T3SS) proteins present at the exposed tip of the T3SS apparatus (T3SA) needle have been tested for their efficacy recently with promising results13–16.
Our group previously demonstrated that a fusion of LTA-1 (the active moiety of the A subunit from the heat-labile enterotoxin of Enterotoxigenic Escherichia coli) with the Pa T3SA needle tip protein (PcrV) and first translocator protein (PopB) to give L-PaF was an effective vaccine against Pa infection in BALB/c mice17. In this study, L-PaF was used at multiple doses, with and without additional adjuvants to determine its effectiveness in protecting against Pa infection in the more genetically and immunologically diverse CD-1 mouse model. The formulations’ effects on cellular and humoral immune responses were then evaluated in terms of immune correlates. Lung mRNA seq analysis was also done to determine the immune pathways activated by the vaccine and how they compare to those activated by the actual Pa infection in non-immunized mice. Lastly, the effects of IL-17 and immunoglobulins, thought to be important for rapid Pa clearance prior to establishing chronic infection, were evaluated in vivo using il17−/− and muMt- knockout (KO) mice.
## Specific vaccine formulations induce serum immunoglobulins with high serum opsonophagocytic activity
Based on our previous work with L-PaF13, CD-1 mice were vaccinated with two concentrations of L-PaF (20 or 10 µg) alone, with ME or with BECC/ME. PBS was used as a vaccinated negative control. Mice were immunized intranasally (IN) starting at 6–8 weeks of age on days 0, 14 and 28. Regardless of formulation, L-PaF elicited significant systemic humoral immune responses (i.e., IgG and IgA) against both PcrV and PopB while PBS and Whole Cell Killed Pa (WCK) did not (Fig. 1). IgG subclasses were also measured and all L-PaF formulations were found to be significantly higher for all subclasses than the PBS control. IgG1 was the major subtype observed, followed by IgG2a and IgG3 (Supplementary Fig. 1). Significant differences in the generation of IgG1, IgG2a, and IgG3 were observed for the 10 μg L-PaF BECC/ME mice relative to the PBS group and 20 μg L-PaF BECC/ME produced significantly higher IgG2a and IgG3 relative to PBS.Fig. 1Kinetics of serum IgG and IgA.CD-1 mice were vaccinated on days 0, 14 and 28 and their sera were assessed for anti-PcrV and anti-PopB immunoglobulins as a function of time and boosting. Anti-PcrV IgG (A) and IgA (C), along with anti-PopB IgG (B) and IgA (D), are shown. Individual titers are represented as EU/ml. Each point denotes a mean and error bars represent SD for each group ($$n = 10$$ mice/group).
Opsonophagocytic killing (OPK) assays were performed as an in vitro functional assay to determine whether the secreted immunoglobulins promote specific macrophage killing of Pa (Fig. 2). In comparison to the PBS vaccinated mice, all other groups possessed significantly higher OPK activity except for the WCK and 20 μg L-PaF/ME groups. Both doses of L-PaF alone and with ME possessed 34–$47\%$ OPK activity (Fig. 2A) with the L-PaF formulations containing BECC438b and ME showing the highest killing ability with 10 µg and 20 μg L-PaF BECC/ME exhibiting 61 and $55\%$ OPK activity, respectively. Fig. 2Serum immunoglobulins possess opsonophagocytic killing (OPK) activity. The day 42 sera from Fig. 1 were heat-treated at 56 °C for 30 min for subsequent use in OPK assays. A Relative OPK activities are shown for sera collected from each CD-1 mouse group are shown. B A correlation between OPK and either anti-PcrV or anti-PopB immunoglobulin titers is shown. C A correlation between OPK and specific subclasses of IgG against PcrV and PopB is shown. The points in (A) represent individual OPK values obtained from the experiment (error bars represent SD for each group), whereas the points in (B) and (C) represent correlation values obtained between OPK and pooled D42 serum. Pearson’s r coefficient and simple linear regression ($95\%$ confidence level) were calculated for (B) and (C). * $p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001.$ Exact p values can be found in Supplementary Tables 1, and 2.
To determine whether the observed OPK levels correlate with immunoglobulin titers for day 42 serum, a correlation analysis was carried out. OPK values positively correlated with the amount of serum immunoglobulins present (r > 0.80 to <0.87, $p \leq 0.01$) (Fig. 2B and Supplementary Table 1). This was also positively correlated with the amounts of IgG sub-types (Fig. 2C and Supplementary Table 2). OPK activity appears to be an important determinant for reducing infection in vivo since mucoid Pa strains have been reported to be more resistant to non-opsonic phagocytosis than their non-mucoid counterparts1. Immunoglobulins present in the serum participate in OPK activity to kill mucoid Pa (mPa08-31).
## PcrV and PopB can stimulate Th1 and Th17 cells in the lung: a potentially important determinant of protection
At 56 days post the first immunization (DPIm), lung cells were isolated and stimulated with either PcrV or PopB and the cytokine secreting cells were enumerated using ELISpot (ImmunoSpot) analysis (Fig. 3). While control groups showed little to no response to stimulation with either antigen, IL-17A secreting cells were found to be significantly upregulated in the PcrV-stimulated lung cells derived from the 20 μg L-PaF, 20 μg L-PaF BECC/ME and 10 μg L-PaF BECC/ME vaccinated groups (Fig. 3A). In contrast, PopB stimulation of lung cells only resulted in a significant upregulation of IL-17A for the 20 μg L-PaF BECC/ME and 10 μg L-PaF BECC/ME vaccine groups. The number of cells secreting IFN-γ following antigen stimulation was smaller than the number of IL-17A secreting cells and appeared to only be significant for 20 μg L-PaF group (Fig. 3B), suggesting that IFN-γ is less of a player for the vaccine formulations being explored here. Interestingly, unstimulated (incubated with cell culture media only) cells from the 20 μg L-PaF BECC/ME and 10 μg L-PaF BECC/ME groups showed relatively higher frequencies of IFN-γ secreting cells indicating a Th1 response upon vaccination (Fig. 3B).Fig. 3IL-17 and IFN-γ secretion following ex vivo stimulation of lung cells. Cell suspensions were prepared and treated with either PcrV or PopB or left untreated (media). IL-17 (A) and IFN-γ (B) secreting cells were quantified after 24 h stimulation and are shown as secreting cells per million total cells. Values were plotted as individual points ± SD ($$n = 5$$/group). Error bars represent SD. Statistical significance was calculated by comparing PBS vaccinated group with each of the other vaccinated groups using Dunnett’s test. * $p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001.$
Stimulation of antigen-specific mucosal immunity is an important factor for protection from mucosal pathogens18. Although IL-17A (and IFN-γ to a lesser extent) secreting cells were found to be present in the lungs of some vaccinated mouse groups, ELISpot does not quantify the absolute amounts of cytokines secreted by these cells. To assess this, lung cells were incubated without or with either PcrV or PopB and the cytokine levels in the culture supernatants was quantified (Supplementary Fig. 2). Both proteins stimulated lung cells from L-PaF vaccinated mice to produce significant levels of IL-17A relative to the PBS and WCK vaccinated mice. The exception was the PcrV-stimulated 10 μg L-PaF and 10 μg L-PaF with ME groups (Supplementary Fig. 2A and B). For these, IFN-γ was found to be elevated to a significant level in the 20 μg L-PaF, as expected based on the data from Fig. 3, but also in the 10 μg L-PaF BECC/ME mice (Supplmentary Fig. 2D and E). Supplementary Fig. 2C, and F shows IL-17A and IFN-γ profiles in unstimulated lung cells, respectively.
## Vaccinated mice efficiently cleared Pa from lungs
Ultimately, rapid clearance of Pa from the lungs of challenged mice is a key hallmark of an effective vaccine. To assess this, vaccinated groups of mice were challenged with 1 × 108 CFU of mPa08-31 and at 16 HPI the lungs were harvested to enumerate the relative bacterial burdens. A significant reduction of lung burden was seen for 20 μg L-PaF, 20 μg L-PaF/ME, 10 μg L-PaF/ME and 10 μg L-PaF BECC/ME vaccinated mice (Fig. 4). A noteworthy observation is that the 20 μg L-PaF BECC/ME group only failed to be statistically significant because of a single outlier. Twenty percent of mice ($\frac{1}{5}$) completely cleared Pa from the lungs of the 20 μg L-PaF, 10 μg L-PaF, and 10 μg L-PaF BECC/ME vaccinated mice. Meanwhile, $40\%$ had completely cleared mPa08-31 in 10 μg L-PaF/ME and even the 20 μg L-PaF BECC/ME. The highest degree of complete bacterial clearance ($80\%$) was observed for the 20 μg L-PaF/ME group. Nevertheless, the 10 μg L-PaF BECC/ME vaccinated group showed the greatest overall average reduction of Pa burden, and at a lower dose of antigen for the degree of clearance seen. Fig. 4The in vivo protective efficacy of each vaccine formulation was determined. Mice were challenged with 1 × 108 CFU/30 µl/mouse of mPa08-31 and lung burden was determined after 16 h post infection (HPI) ($$n = 5$$). The points represent individual CFU/lung values, and the SDs are denoted by error bars. Lung burden was compared between PBS and all the other groups using a Dunnett’s test. * $p \leq 0.05.$
Interestingly, the CD-1 mice were also found to be protected from infection after 180 DPIm (Table 1). The results were comparable with the lung burden reductions seen in Fig. 4, however, only the 10 μg L-PaF/ME group was found to eliminate the Pa lung burden at 180 DPIm challenge. The greatest reductions in Pa burden for these mice appeared to require the presence of ME in these experiments, suggesting that a multimeric or particulate presentation of the antigen is important for long-term protection. Table 1Long term (180 DPIm) protection in CD-1 mice. Mean CFU/lung% compared to PBS controlPBS1550100WCK80051.6L-PaF 201409.0L-PaF 1035022.6L-PaF 20 in ME1258.1L-PaF 10 in ME00L-PaF 20 BECC/ME1207.7L-PaF 10 BECC/ME704.5Mice were challenged with 5 × 106 CFU/30 µl/mouse of mPa08-31 and their lungs were harvested at 16 HPI for bacterial enumeration.
## Lung cytokine levels as related to lung bacteria burden
Growing evidence suggests that a Th17 response is required for the clearance of Pa from lung19. Furthermore, the presence of Th1 cytokines, such as IL-6, also helps reduce Pa burden20. Based on these previous observations, we chose to look at markers related to a balanced Th1/Th17 response in the CD1 mouse model following IN immunization with L-PaF alone or in our formulations containing ME with or without BECC438b. IL-17A (Th17), IFN-γ, TNF-α, and IL-6 (Th1) lung cytokines were measured both pre- and post-challenge to determine if there were specific correlations with lung burden for these Th17 and Th1 markers (Fig. 5). Pre-challenge cytokines were measured upon stimulation with either PcrV or PopB, whereas post-challenge cytokines were measured in the absence of any further stimulation (as discussed above). Pre-challenge cytokines would thus mimic the effect of an actual vaccination/booster, while their post-challenge counterparts would show the cytokine profile following an in vivo infection. A negative correlation was observed between pro-inflammatory cytokines and the Pa lung burden (i.e., the higher the pro-inflammatory response pre-challenge characterized by TNF-α and IL-6, lower the lung burden is) (Fig. 5A–D and Supplementary Table 3). This observation is consistent with prior findings by others that Th1 and Th17 responses are involved in rapid pathogen clearance16,21. The levels of post-challenge cytokines reveal an interesting observation that a more sustained presence of controlled pro-inflammation is required as is seen by vaccination rather than the sudden onset caused by acute infection in the absence of vaccination. None of the groups vaccinated with L-PaF formulations showed a marked increase in these cytokines as a result of infection, while the PBS and WCK vaccinated groups showed a significant pro-inflammatory cytokine increase in response to the mPa08-31 infection (Fig. 5E and Supplementary Table 4). The exact fold-change values are shown in Supplementary Table 4. B.Fig. 5Correlation between lung burden and lung cytokines. Cytokine determination assays were performed to allow for identification of potential correlations between Pa lung burden and PcrV- or PopB-stimulated pre-challenge lung cytokines (IL-17A (A), IFN-γ (B), TNF-α (C) and IL-6 (D)). Fold changes between pre- and post-challenge untreated lung cytokines were evaluated to analyze the correlation with lung burden (E). Pearson’s r coefficient and simple linear regression ($95\%$ confidence level) were calculated. The exact values can be found in Supplementary Tables 3, and 4. The separation of points in A., B., and C. are readily seen in Supplementary Fig. 3A, B, and C, respectively.
The data shown in Fig. 5 and Supplementary Figs. 2, and 3 suggest that increased levels of pre-challenge immunoglobulins, cytokine secreting cells, and cytokines correlate with a rapid clearance of Pa for groups vaccinated with L-PaF formulations. This fits with the earlier observation that the relationship between IgG subtypes and IL-17A is an important determinant for elevated OPK activity. This is further shown as a function of serum antibodies vs. lung burden, serum IgG subtypes vs. lung burden and serum IgG subtypes vs. pre-challenge mean IL-17A (Supplementary Fig. 4A–C, respectively). Mice with higher serum antibodies showed lower lung burden (Supplementary Fig. 4A, and B, Supplementary Tables 5, and 6) and higher IL-17A (Supplementary Fig. 4C and Supplementary Table 7). OPK activity is demonstrated as one indicator of protection against Pa22 and, as such, it is shown to directly correlate with in vivo lung burden (i.e., higher the OPK, lower the lung burden) (Fig. 6). In line with this, the 10 μg L-PaF BECC/ME group showed highest OPK activity in vitro with the lowest overall lung burden in vivo (Fig. 6).Fig. 6Correlation between in vitro (OPK) and in vivo reduction in lung burden. Groups with high in vitro killing ability harbor fewer lung bacteria in vivo. Pearson’s r coefficient and simple linear regression ($95\%$ confidence level) were calculated. r = −0.7759, $95\%$ confidence interval = −0.9572 to −0.1572, R squared = 0.6020, p value (two-tailed) = 0.024.
## Both humoral and cellular responses are important for protection: a wide range of pathways and their associated genes are activated following PcrV or PopB treatment for lung cells in vitro from vaccinated mice
Stimulation of lung cells with PcrV or PopB collected from PBS (negative control) and 10 μg L-PaF BECC/ME (the formulation showing the highest reduction of Pa lung burden) vaccinated (but not infected) mice followed by mRNA-seq analysis showed an elevated expression for genes related to the immune system (Supplementary Figs. 5, and 6). Upregulation of genes related to defense responses, innate immune responses, immune system processes, cytokines, and other effectors were observed following PcrV treatment of lung cells collected from PBS vaccinated mice. Conversely, groups vaccinated with 10 μg L-PaF BECC/ME had a slightly different gene activation pattern, most notably they had genes from the defense response, leukocyte activation, and positive responses involving different immune pathways, among other changes. Stimulation with PopB resulted in the upregulation of various other immune pathways that were not seen following PcrV stimulation (Supplementary Figs. 5, and 6). Genes from inflammatory responses, cytokine production (IL-6, IL-1β), and other immune effector genes were observed in PopB-stimulated lung cells collected from PBS vaccinated mice (Supplementary Fig. 6). Conversely, the 10 μg L-PaF BECC/ME vaccinated mice showed upregulation of genes involved in the inflammatory response pathways, immune regulation and activation, and leukocyte activation. It should be noted that, the aforementioned gene activation profiles resulted from ex vivo restimulation of uninfected but vaccinated (or PBS vaccinated negative control) lungs. When infected with Pa, they show a different activation/deactivation profile as described below.
## The lungs of 10 μg L-PaF BECC/ME vaccinated mice elicited a different gene activation profile than PBS vaccinated mice following Pa infection
Mice vaccinated with 10 μg L-PaF BECC/ME and then challenged with Pa, showed a differential gene expression patterns relative to their PBS vaccinated counterparts to counteract as a result of the infection as determined by mRNA seq (Fig. 7A, C). Genes involved in adaptive immunity were found to be preferentially expressed in 10 μg L-PaF BECC/ME vaccinated mice, whereas genes involved with acute innate responses were preferentially expressed in the PBS vaccinated mice (Fig. 7B) with the latter leading to increased pro-inflammatory responses. These mice had experienced similar bacterial burden compared to the mice that were used for protective efficacy studies. Preferential expression of cytokine-cytokine receptor interactions, TCR signaling, chemokine signaling, JAK-STAT, Th1, Th17, cell adhesion molecules, antigen processing and presentation, and NOD-like receptor signaling pathways were observed in the lungs from the 10 μg L-PaF BECC/ME vaccinated and infected mice (Supplementary Figs. 7–12). On the other hand, the lungs of immunized mice (Fig. 7D) showed no such high expression levels for HIF-1 signaling, IL-17 signaling, TNF signaling, and MAPK (Supplementary Figs. 13–17). The differential expression of these genes is shown with graded red and green colors, respectively. Genes that were preferentially expressed in 10 μg L-PaF BECC/ME mice were downregulated, relatively speaking, in PBS-vaccinated mice and vice-versa. Notably, the highly expressed HIF-1 signaling in infected PBS-vaccinated mice suggests that lung damage is occurring, and there is an associated pro-inflammatory storm, that leads to higher bacterial burden in the lungs23,24.Fig. 7Pattern of gene up/down regulation following 16 HPI with mPa08-31 in 10 μg L-PaF BECC/ME vaccinated and PBS vaccinated mice lung cells. CI_1 = Lung cells from control or PBS vaccinated, infected group_sample #1, CI_2 = Lung cells from control or PBS vaccinated, infected group_sample #2, II_1 = Lung cells from immunized or 10 μg L-PaF BECC/ME infected group_sample #1, II_2 = Lung cells from immunized or 10 μg L-PaF BECC/ME infected group_sample #2. *Upregulated* genes in immunized or 10 μg L-PaF BECC/ME mice are shown in (A), downregulated genes in immunized or 10 μg L-PaF BECC/ME mice are shown in (C). B Pathways upregulated in infected 10 μg L-PaF BECC/ME (or downregulated in infected PBS vaccinated) mice shown via a cnet plot. *Upregulated* genes were found in cytokine-cytokine interaction, TCR signaling, chemokine signaling, JAK-STAT signaling, Th17, Th1/Th2 differentiation, cell adhesion molecules, antigen processing, NLR signaling pathways. D Pathways downregulated in infected 10 μg L-PaF BECC/ME (or upregulated in infected PBS vaccinated) mice shown via a cnet plot. *Downregulated* genes were found in cytokine-cytokine interaction, IL-17 signaling, TNF signaling, HIF-1 signaling and MAPK signaling pathways. Fold increase/decrease in (B and D), have been shown by different color according to the log2 scale shown in the figure. Size of the gene clusters have also been shown.
## IL-17 and immunoglobulins are important determinants of protection
Infection with mPa08-31 was successfully hindered by 10 μg L-PaF BECC/ME (“v” groups) vaccination in the WT group (Fig. 8). Both PBS (“u” groups) and 10 μg L-PaF BECC/ME vaccinated il17−/− groups showed a high lung burden (Fig. 8A). When the 10 μg L-PaF BECC/ME vaccinated il17−/− mice were complemented with lung cells from 10 μg L-PaF BECC/ME vaccinated WT, the infection was significantly reduced. Administration of rat anti-mouse IL-17A (IL-17A Rat anti-Mouse, NA/LE, Unlabeled, clone: TC11-18H10, catalogue #: BDB560268, 1:50 dilution) antibody does not dampen the host immune response. High levels of immunoglobulins and in vivo IL-17A were observed in these mice (Supplementary Figs. 18, and 19). High levels of IL-17A post-challenge correlated with lower bacterial burden (Supplementary Fig. 19A, and C), whereas higher IL-6 correlated with higher lung burden (Supplementary Fig. 19B, and D).Fig. 8In vivo protective efficacy study in knock out (KO) mice. A IL-17 KO or (B) muMt- KO mice were infected with 1 × 108 CFU/30 µl/mouse and their lungs processed at 16 HPI to evaluate lung bacterial burden ($$n = 3$$). The points represent individual CFU/lung values, and the SDs are denoted by error bars. Lung burden was compared between uWT and other groups using Dunnett’s test. * $p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001.$ uWT = PBS vaccinated wild type; vWT = 10 μg L-PaF BECC/ME vaccinated WT; u il17 KO = PBS vaccinated IL-17 K; v il17 KO = 10 μg L-PaF BECC/ME vaccinated IL-17 KO; u muMt KO = PBS vaccinated muMt KO; v muMt KO = 10 μg L-PaF BECC/ME vaccinated muMt KO mice.
Colonization with mPa08-31 was hindered in 10 μg L-PaF BECC/ME vaccinated WT mice, as expected. Conversely, PBS and 10 μg L-PaF BECC/ME vaccinated muMt- KO mice had higher bacterial burdens in their lung compared to the 10 μg L-PaF BECC/ME vaccinated WT group (Fig. 8B). Although not statistically significant, a clear difference was observed in the bacterial burden between PBS and 10 μg L-PaF BECC/ME vaccinated muMt- KO groups. Upon receiving serum from 10 μg L-PaF BECC/ME vaccinated WT mice, lung burden in the 10 μg L-PaF BECC/ME vaccinated muMt- KO mice were found to be comparable with 10 μg L-PaF BECC/ME vaccinated WT group. Addition of IL-17 neutralizing antibody showed no effect on 10 μg L-PaF BECC/ME vaccinated muMt- KO mice. As expected, the muMt- KO mice could not produce any immunoglobulins (Supplementary Fig. 20). They did produce IL-17A and IL-6, as well as other cytokines, leading to a scrambled immune response (with no humoral response and somewhat activated cellular response) (Supplementary Fig. 21A, and B). Although not statistically significant, a trend was observed when lung IL-17A and IL-6 post-challenge was compared with the lung burden with a visual correlation observed with higher IL-17A leading to a better protection (Supplementary Fig. 21C, and D).
## Discussion
Pa is a ubiquitous opportunistic pathogen that is an important cause of nosocomial infections in immune-compromised individuals, burn and wound victims, patients with cystic fibrosis, and the elderly. A licensed vaccine remains an unmet need. Different approaches have been devised to construct a vaccine against Pa, but with limited success thus far8–12. Recently, our group has assessed the efficacy of a new T3SS subunit vaccine using a mouse acute lung model13. The T3SS is an important virulence factor involved in initiating Pa infection. The Pa T3SS injects effectors into host cells, impairing early innate immune responses to allow initiation of infection25, but this requires a functional secretion apparatus (T3SA) with functional tip complex and translocator proteins. Since the T3SS is employed in the early stages of infection, proteins associated with T3SA can potentially be important immune targets for preventing the onset of infection26.
In this study, the T3SA tip and first translocator proteins, PcrV and PopB, respectively, were genetically fused with E. coli LTA1 to produce L-PaF. Mice were vaccinated with two concentrations of L-PaF with or without other adjuvants, i.e., MedImmune Emulsion (ME) and BECC438b (BECC). ME is an oil-in-water emulsion that we use to create a nanoparticle that enhances multimeric presentation of the antigen to increase the resulting immune response17,27. While monomeric protein antigens elicit robust immune responses and are often efficacious in mice, they often fail in human trials unless formulated into an oligomeric presentation with an appropriate adjuvant28,29. BECC438b is a TRL4 agonist that is a biosimilar of monophosphoryl lipid A (MPLA)30. Immunogenicity of L-PaF was found to increase with the addition of ME and BECC. Serum immunoglobulins were elevated at 42 DPIm, with a small decrease on 56 DPIm. Serum from mice vaccinated with L-PaF formulations showed enhanced killing capacity for wild-type Pa in in vitro OPK studies. This bacterial killing ability was further enhanced for formulations involving ME and BECC. Additionally, OPK activity had a positive correlation with serum IgG, IgA, and certain IgG subtypes.
Pa has been reported to be subjected to killing in the presence of a balanced Th1/Th2 antibody and opsonophagocytic response13,31,32. A marker for a balanced Th1/Th2 response is a balanced IgG1/IgG2a ratio33. Membrane proteins and soluble protein antigens usually upregulate IgG1, while responses against capsular polysaccharide are restricted to IgG2. On the other hand, IgG3 is a potent inflammatory antibody with comparatively shorter half-life but with the ability to effectively direct OPK34. In our study, a steady increase was observed for IgG1, IgG2a and IgG3 following vaccination. Studies have uncovered the ability of different IgG subtypes to bind with components of the immune system, such as binding with FcγRs and complement C1q components, to name a few35. A steady presence of different immunoglobulins, as well as their involvement with OPK activity, provides a primary indication that our L-PaF formulations are indeed steering the host immune response in a desirable direction.
Many bacterial infections are dependent on the host’s immune status. A Th2-dominant pulmonary response in mice is seen to be susceptible to Pa infection. Meanwhile, a Th1-dominant response has a better prognosis against Pa36. Involvement of a Th17 cell response in the form of IL-17 has also been shown useful for Pa clearance13,16,21. Secretion of the Th1 cytokine IFN-γ, albeit modestly, and IL-17 in the present study suggest a favorable outcome post infection. Indeed, reduced lung Pa burden at 16 HPI shows a clear correlation with the addition of the adjuvants used here. More importantly, this protective efficacy shows a dose-dependent response with the presence of ME and BECC. Mice were further checked for long-term protective efficacy at 6 months post first vaccination with significant success for the formulated antigen.
Th1-Th17 response plays an important role in determining the fate of an infection. Differentiation of naïve T cells into Th1/Th2 and/or Th17 depends on direct interaction with APCs, cytokine environment and coreceptors signals, among others37. In our study, a clear Th1 differentiation was observed in vaccinated and infected mouse lungs (Supplementary Fig. 9). We hypothesize this effect is elicited by the BECC438b adjuvant (a TLR4 agonist) in conjunction with the multimeric presentation of the antigens. Historically, LPS has been associated with eliciting a Th1-biased immune response38 and our study would agree with those findings. Activation of various upstream TCR signaling pathways (Supplementary Fig. 8) leading to the production of IFN-γ, as well as STAT1 and IL-12, in mice vaccinated with the L-PaF formulation, but not in PBS vaccinated mice, is an indication that Pa infection proceeds differently in the vaccinated mice. Moreover, upregulation of RORγt, IL-21 and IL-22 indicates a pronounced Th17 immune response following infection in vaccinated mice. IL-21 and IL-22 are known to mount an immune response against extracellular pathogens and play a protective role in mucosal anti-microbial host defenses39. While a protective immune response was observed in mice vaccinated with the L-PaF formulation, a more pronounced pro-inflammatory response was observed in PBS vaccinated mice subsequently infected with Pa. Genes from TNF, MAPK and other pro-inflammatory cytokines were downregulated in mice vaccinated with the L-PaF formulation, while these genes were upregulated, leading an uncontrolled inflammatory response, in PBS vaccinated and subsequently infected mice. Hypoxia inducible genes were also found to be upregulated in PBS vaccinated mice after infection. Hypoxia is a factor known to activate NF-κB, leading to an uncontrolled release of pro-inflammatory cytokines, such as IL-6, TNF-α and MCP-1, to name a few40. Downregulation of TNF, IL-17 and MAPK downstream pathways following infection in mice vaccinated with the L-PaF formulation, but not in PBS vaccinated mice indicates that PBS vaccinated mice reacted severely to the infection, while the mice vaccinated with the L-PaF formulation showed a more protective effect after infection. PBS vaccinated mice tried to circumvent the infection by activating their innate response gene manifold, while mice vaccinated with the L-PaF formulation were more “prepared” for the infection and eliminated it more rapidly.
Although several novel pathways were found to provide the observed anti-Pa immunity that have not been shown before, it was not possible to assess all of them within the scope of the present work. A controlled IL-17 response, along with serum immunoglobulins, were found to be associated with anti-Pa immunity in our previous studies, as well17. Thus, we chose these two main protective immune components to assess their effects in specific KO mice. In both cases, a dampened anti-Pa response suggests that both IL-17 and immunoglobulins are needed to generate a protective anti-Pa immunity in mice.
Additional mutational and molecular approaches are needed to fully understand the immune pathways and responses needed to mount effective anti-Pa immunity. Nevertheless, this work provides a preliminary view of the basic pathways required to generate an anti-Pa immunity and to eliminate the infection within the context of the formulations used here. How alternative formulations might affect the efficacy and immune pathways elicited remains to be determined. Future work can build upon the present work as a knowledge base for generating new information.
## Materials
Squalene was purchased from Echelon Biosciences (Salt Lake City, UT). BECC438b was extracted from *Yersinia pestis* after the introduction of lipid A modifying enzymes. It is a bis-phosphorylated lipid A analogue generated using a Bacterial Enzymatic Combinatorial Chemistry (BECC) platform in the *Yersinia pestis* KIM6 strain30. Candidate number 438 was given the designation BECC438b to indicate that it is purified from a biological (b) origin. The BECC438b used here has been tested and found to be nontoxic in rabbits. All other buffers, and chemicals were reagent grade.
## Proteins and formulations
L-PaF was purified using standard IMAC and Q anion exchange column chromatography steps13,17. L-PaF was dialyzed into PBS with $0.05\%$ LDAO and stored at −80 °C. The formulations were prepared as oil-in-water emulsions13,17. Briefly, squalene ($8\%$ by weight) and polysorbate 80 ($2\%$ by weight) were mixed to achieve a homogenous oil phase. Twenty percent sucrose and 40 mM histidine (pH 6) were added to the oil phase to generate a milky emulsion of 4XME (MedImmune Emulsion). BECC438b (2 mg/ml) was prepared in $0.5\%$ triethanolamine and adjusted to pH 7.2 with 1 M HCl. To make the L-PaF with ME formulation, the protein was added directly to the ME with a final concentration of 0.67 mg/ml. To make the L-PaF with ME and BECC438b formulation, ME and BECC438b were first mixed and then L-PaF was mixed with ME-BECC438b solution at a volumetric ratio of 1:1 to achieve the desired final antigen concentration.
## Mice and immunizations
Six- to eight-week-old CD-1, C57BL/6 (B6), B6.Cg-Il17a/Il17ftm1.1Impr Thy1a/J (il17−/− or IL-17 KO) and B6.129S2-Ighmtm1Cgn/J (muMt− KO) mice were purchased from Charles River Laboratories (Wilmington, MA) or The Jackson Laboratory (ME, USA). Mice were left to acclimate for 1 week following their delivery. Eight groups of CD-1 mice ($$n = 10$$/group) were vaccinated intranasal (IN) with 30 µl containing PBS, Whole Cell Killed (WCK) Pa, 20 μg L-PaF (20 μg L-PaF), 10 μg L-PaF (10 μg L-PaF), 20 μg L-PaF in ME (20 μg L-PaF/ME), 10 μg L-PaF in ME (10 μg L-PaF/ME), 20 μg L-PaF+BECC438b in ME (20 μg L-PaF BECC/ME), and 10 μg L-PaF+BECC438b in ME (10 μg L-PaF BECC/ME). Formulations were administered in a prime-boost-boost manner with the prime dose on day 0, followed by boosters on days 14 and 28. Blood was collected on days 0, 28, 42 and 56. il17−/−, muMt- and C57BL/6 were only vaccinated with PBS or the 10 μg L-PaF BECC/ME formulation and assessed for protection since this formulation was found to be protective in a preliminary trial ($$n = 3$$).
## Ethical statement
Animal works were carried out according to the University of Kansas (KU, Lawrence) IACUC animal use statement (AUS 222-03, valid until March 9, 2025). The Institution’s Animal Welfare Assurance number is D16-00220 (A3339-01).
## ELISA
Serum anti-PcrV, anti-PopB IgG and IgA were measured by coating the 96-well microtiter plates with 100 µl of either PcrV (1 µg/ml) or PopB (1 µg/ml) and incubated for 3 h41. Plates were then blocked with $10\%$ non-fat dry milk overnight at 4 °C. After washing, primary (serum) and secondary antibodies (1:1000 dilution for anti-IgG, and 1:4000 for anti-IgA), [Catalogue #: OB1040-05 for IgA from Southern Biotech, and catalogue #: 5450-0011 [474-1806] for IgG from Sera Care] were added and incubated for 1 h each followed by washes with PBS-Tween 20 ($5\%$). All secondary antibodies were horseradish peroxidase (HRP)-labeled goat anti-mouse IgG/IgA, that were human-adsorbed. TMB substrate (3, 3′, 5, 5′-tetramethylenebenzidine) was added and the reaction was stopped with phosphoric acid. Endpoint titers were calculated as ELISA units per ml (EU/ml). IgG1, IgG2a and IgG3 subclasses were measured similarly with the appropriate secondary antibody.
## Opsonophagocytic killing (OPK) assay
OPK assay was carried out using the Pa strain mPa08-31 that was grown overnight13. The following morning, a new culture was started using 1:100 inoculum in low salt LB broth ($0.5\%$ w/v NaCl) and the absorbance (600 nm) was adjusted to 0.3. A bacterial concentration of 2 × 107 cells/ml was adjusted in $10\%$ bovine serum albumin (BSA, Sigma, St. Louis, MO) containing Minimal Essential Medium (MEM, ThermoFisher, Waltham, MA). The J774.1 macrophage cell line (ATCC, Manassas, MA) was grown in Dulbecco’s Modified Eagle’s Medium (DMEM, ThermoFisher, Waltham, MA) until $90\%$ confluence was achieved. mPa08-31 was adjusted to a final Multiplicity of Infection (MOI) of 0.1. Sera from each mouse group at 42 DPIm were heat inactivated at 56 °C for 30 min. The serum was diluted 1:500 and then mixed with the bacteria and macrophages at a ratio of 1:1:1 to a final volume of 300 µl. This mixture was incubated for 30 min at 37 °C and serial dilutions were prepared for all the technical quadruples and plated on PIA. Percent killing was measured using the following formula:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left[{\left\{ {\left({{{{\mathrm{CFU}}}}\,{{{\mathrm{from}}}}\,{{{\mathrm{T}}}}_0} \right) - \left({{{{\mathrm{CFU}}}}\,{{{\mathrm{from}}}}\,{{{\mathrm{T}}}}_{30}} \right)} \right\}/\left({{{{\mathrm{CFU}}}}\,{{{\mathrm{from}}}}\,{{{\mathrm{T}}}}_0} \right) \times 100} \right].$$\end{document}CFUfromT0−CFUfromT30/CFUfromT0×100.
## Adoptive transfer
Adoptive transfer of lung cells and serum from vaccinated mice was done for il17−/− and muMt− mice, respectively. Lungs were collected from B6 mice vaccinated with 10 μg L-PaF BECC/ME and a single cell suspension (1 × 106 lung cells/ml) was prepared. Each mouse received 100 µl (1 × 105 lung cells) 3 h prior to the bacterial challenge. Serum was collected from B6 mice vaccinated with 10 μg L-PaF BECC/ME as described above. Serum (100 µl) was transferred to each of the recipient mice. Both cells and serum were introduced intraperitoneally42.
## Bacterial challenge experiment
The Pa strain mPa08-3143,44 was grown overnight and used to inoculate fresh low salt LB medium at a 1:100 ratio. The bacterial culture was incubated at 37 °C until it reached an OD600 of 0.3. Pa was then centrifuged and adjusted to a concentration of 1 × 108 CFU/30 µl. Two months after the first immunization, or 4 weeks after the last booster, mice were anesthetized using isoflurane and challenged intranasally (IN) with 30 µl of the bacterial suspension. The waiting period of 4 weeks ensures that the innate response has subsided and that any observed protection is from adaptive immunity. At 16 HPI, mice were euthanized, and the organs were processed as described below. The lung cells were then used to evaluate bacterial burden. The same bacterial concentration was used for the CD-1, B6 and the different KO mice. For mRNA seq. of the infected mice, a separate experiment was carried out ($$n = 2$$/group) with two groups. The groups used here were the PBS immunized or control and L-PaF 10 BECC/ME immunized or immunized. They were challenged at the same dose of Pa, and RNA was isolated at 16 HPI. For long-term protective efficacy, mice were challenged with 5 × 106 CFU bacterial inoculum per mouse at 180 DPIm. Lungs were assessed at 16 HPI for the analysis of lung burden as described above.
## Organ collection
For immunological analysis of the lung, they were collected aseptically in MACS® Tissue Storage Solution (Miltenyi Biotec, USA) and processed with a lung dissociation kit (Miltenyi Biotec, USA). An erythrocyte lysis step was carried out followed by adjusting the cell number to 1 × 107 cells/ml. These cells were then further processed for cytokine and ELISpot studies. Pre-challenge necropsies were carried out at 56 DPIm, while post-challenge necropsies were carried out at 16 HPI.
## T cell ELISpot assay
Lung cells from the previous step were incubated for 24 h at 37 °C in the presence or absence of 5 µg/ml of either PcrV or PopB. ELISpot plates were coated with capture antibodies against IFN-γ or IL-17A for a T cell double-color enzymatic assay according to the manufacturer’s instructions (ImmunoSpot, Cellular Technology Limited, USA). Cytokine secreting cells were quantified using an ImmunoSpot analyzer with Professional DC software. A quality control step was carried out before finalizing the data.
## Cytokine analysis
Lung cells were incubated for 48 h at 37 °C in the presence or absence of 10 µg/ml of either PcrV or PopB. Supernatants were analyzed with U-PLEX kits for the following cytokines: IFN-γ, IL-17A, IL-6 and TNF-α. Cytokine concentrations were measured using a Meso Scale Discovery (MSD, Rockville, MD) plate reader using DISCOVERY WORKBENCH® analytical software. Out of the 10 cytokines measured, only IL-17A, IFN-γ (pre-challenge) and IL-17A, IFN-γ and TNF-α (post-challenge) levels were found to be significantly changed when compared to their respective control groups.
## Correlation studies
Linear correlation studies were carried out using bivariate correlation in the form of Pearson r. They were further analyzed via simple linear regression. P value designations are discussed in the statistical analysis section. All fold change studies were done using the pre-challenge unstimulated lung cytokines as the denominator.i. OPK and immunoglobulins. The day 42 serum samples’ in vitro bacterial killing ability and serum immunoglobulin responses were analyzed for linear correlation using the methods described above. OPK was also analyzed against IgG subtypes.ii. Cytokines and lung burden. Pre- and post-challenge lung cytokines and lung burden were analyzed for linear correlation as described above.iii. OPK and lung burden. The in vitro bacterial killing ability of serum samples was cross-checked with in vivo lung burden to determine whether there is a possible correlation between the two.iv. IgG subtypes and IL-17A. Serum IgG subtypes were cross-checked with the protective cytokine IL-17A to determine whether there is a correlation between them.v. IgG subclasses and lung burden. IgG subclasses were also analyzed to determine whether there is any correlation between these with the in vivo lung burden.
## mRNA-seq analysis
RNA was isolated from lung cells using RNeasy® Mini kit according to the manufacturer’s instructions (QIAGEN, Hilden, Germany). RNAs with an RNA Integration Number (RIN) of >7 were shipped to Novogene for mRNA sequencing. Another round of QC was carried at their facility prior to the actual mRNA sequencing. mRNA-seq read count data was provided by Novogene which is further processed using the iDEP web server (PMID: 30567491) for differential expression and pathway analysis. Initially, only the genes with at least 0.5 counts per million (CPM) reads in at least one sample were considered. Read counts were then transformed as log2(CPM) using the EdgeR method. The differentially expressed genes (DEGs) were identified using DESeq2 with an FDR cutoff 0.05 and a minimum fold change of 2. Finally, the enriched pathways in DEGs for the selected comparisons were identified through GO Biological Process analysis. Up-regulated pathways which most likely play a critical role in immune modulation were manually selected for each comparison. The fold change of individual genes associated with selected pathways were visualized as a heatmap using R. Gene Set Enrichment Analysis of Gene Ontology and KEGG pathway analysis was performed using R clusterProfiler (version 3.0.4).
## Statistical analyses
GraphPad Prism 8.1.2 was used to prepare data and perform statistical analyses. PBS vaccinated groups were compared with the other vaccinated groups using Dunnett’s multiple comparison test. A p value of <0.05 was considered significant (*$p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001$). Pearson’s r values and R squared values are mentioned as deemed appropriate. mRNA sequencing data were prepared as stated above.
## Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
## Supplementary information
Supplemental Materials. REPORTING SUMMARY The online version contains supplementary material available at 10.1038/s41541-023-00618-w.
## References
1. Krieg DP, Helmke RJ, German VF, Mangos JA. **Resistance of mucoid Pseudomonas aeruginosa to nonopsonic phagocytosis by alveolar macrophages in vitro**. *Infect. Immun.* (1988.0) **56** 3173-3179. DOI: 10.1128/iai.56.12.3173-3179.1988
2. Merakou C, Schaefers MM, Priebe GP. **Progress Toward the Elusive Pseudomonas aeruginosa Vaccine**. *Surg. Infect. (Larchmt.)* (2018.0) **19** 757-768. DOI: 10.1089/sur.2018.233
3. Migiyama Y. **Pseudomonas aeruginosa Bacteremia among Immunocompetent and Immunocompromised Patients: Relation to Initial Antibiotic Therapy and Survival**. *Jpn J. Infect. Dis.* (2016.0) **69** 91-96. DOI: 10.7883/yoken.JJID.2014.573
4. Ohmagari N. **Risk factors for infections with multidrug-resistant Pseudomonas aeruginosa in patients with cancer**. *Cancer* (2005.0) **104** 205-212. DOI: 10.1002/cncr.21115
5. Horcajada JP. **Epidemiology and Treatment of Multidrug-Resistant and Extensively Drug-Resistant Pseudomonas aeruginosa Infections**. *Clin. Microbiol. Rev.* (2019.0) **32** e00031-19. DOI: 10.1128/CMR.00031-19
6. 6.Antibiotic resistance threats in the United States, 2019. National Center for Emerging and Zoonotic Infectious Diseases (NCEZID), Division of Healthcare Quality Promotion (DHQP), Centres for Disease Control and Prevention (2019).
7. Kollef MH. **Global prospective epidemiologic and surveillance study of ventilator-associated pneumonia due to Pseudomonas aeruginosa**. *Crit. Care Med.* (2014.0) **42** 2178-2187. DOI: 10.1097/CCM.0000000000000510
8. Hatano K, Goldberg JB, Pier GB. **Biologic activities of antibodies to the neutral-polysaccharide component of the Pseudomonas aeruginosa lipopolysaccharide are blocked by O side chains and mucoid exopolysaccharide (alginate)**. *Infect. Immun.* (1995.0) **63** 21-26. DOI: 10.1128/iai.63.1.21-26.1995
9. Rello J. **A randomized placebo-controlled phase II study of a Pseudomonas vaccine in ventilated ICU patients**. *Crit. Care* (2017.0) **21** 22. DOI: 10.1186/s13054-017-1601-9
10. Kamei A, Coutinho-Sledge YS, Goldberg JB, Priebe GP, Pier GB. **Mucosal vaccination with a multivalent, live-attenuated vaccine induces multifactorial immunity against Pseudomonas aeruginosa acute lung infection**. *Infect. Immun.* (2011.0) **79** 1289-1299. DOI: 10.1128/IAI.01139-10
11. Priebe GP, Meluleni GJ, Coleman FT, Goldberg JB, Pier GB. **Protection against fatal Pseudomonas aeruginosa pneumonia in mice after nasal immunization with a live, attenuated aroA deletion mutant**. *Infect. Immun.* (2003.0) **71** 1453-1461. DOI: 10.1128/IAI.71.3.1453-1461.2003
12. Li Y. **X-ray Irradiated Vaccine Confers protection against Pneumonia caused by Pseudomonas aeruginosa**. *Sci. Rep.* (2016.0) **6** 18823. DOI: 10.1038/srep18823
13. Das S. **Development of a Broadly Protective, Self-Adjuvanting Subunit Vaccine to Prevent Infections by Pseudomonas aeruginosa**. *Front. Immunol.* (2020.0) **11** 583008. DOI: 10.3389/fimmu.2020.583008
14. Schaefers MM. **PLGA-encapsulation of the Pseudomonas aeruginosa PopB vaccine antigen improves Th17 responses and confers protection against experimental acute pneumonia**. *Vaccine* (2018.0) **36** 6926-6932. DOI: 10.1016/j.vaccine.2018.10.010
15. Wang Y. **Development of a Chimeric Vaccine Against Pseudomonas aeruginosa Based on the Th17-Stimulating Epitopes of PcrV and AmpC**. *Front. Immunol.* (2021.0) **11** 601601. DOI: 10.3389/fimmu.2020.601601
16. Sawa T. **Active and passive immunization with the Pseudomonas V antigen protects against type III intoxication and lung injury**. *Nat. Med* (1999.0) **5** 392-398. DOI: 10.1038/7391
17. Howlader DR. **Effect of Two Unique Nanoparticle Formulations on the Efficacy of a Broadly Protective Vaccine Against Pseudomonas Aeruginosa**. *Front. Pharm.* (2021.0) **12** 706157. DOI: 10.3389/fphar.2021.706157
18. Fujimoto K. **Antigen-Specific Mucosal Immunity Regulates Development of Intestinal Bacteria-Mediated Diseases**. *Gastroenterology* (2019.0) **157** 1530-43.e4. DOI: 10.1053/j.gastro.2019.08.021
19. Omar T, Ziltener P, Chamberlain E, Cheng Z, Johnston B. **Mice Lacking γδ T Cells Exhibit Impaired Clearance of Pseudomonas aeruginosa Lung Infection and Excessive Production of Inflammatory Cytokines**. *Infect. Immun.* (2020.0) **88** e00171-20. DOI: 10.1128/IAI.00171-20
20. Baker SM, McLachlan JB, Morici LA. **Immunological considerations in the development of Pseudomonas aeruginosa vaccines**. *Hum. Vaccin Immunother.* (2020.0) **16** 412-418. DOI: 10.1080/21645515.2019.1650999
21. Bayes HK, Ritchie ND, Evans TJ. **Interleukin-17 Is Required for Control of Chronic Lung Infection Caused by Pseudomonas aeruginosa**. *Infect. Immun.* (2016.0) **84** 3507-3516. DOI: 10.1128/IAI.00717-16
22. Hassan R. **Immunization with outer membrane proteins (OprF and OprI) and flagellin B protects mice from pulmonary infection with mucoid and nonmucoid Pseudomonas aeruginosa**. *J. Microbiol. Immunol. Infect.* (2018.0) **51** 312-320. DOI: 10.1016/j.jmii.2016.08.014
23. Malkov MI, Lee CT, Taylor CT. **Regulation of the Hypoxia-Inducible Factor (HIF) by Pro-Inflammatory Cytokines**. *Cells* (2021.0) **10** 2340. DOI: 10.3390/cells10092340
24. AbdelMassih A. **Hypoxia-inducible factor (HIF): The link between obesity and COVID-19**. *Obes. Med.* (2021.0) **22** 100317. DOI: 10.1016/j.obmed.2020.100317
25. Hauser AR. **The type III secretion system of Pseudomonas aeruginosa: infection by injection**. *Nat. Rev. Microbiol.* (2009.0) **7** 654-665. DOI: 10.1038/nrmicro2199
26. Rashid MI, Naz A, Ali A, Andleeb S. **Prediction of vaccine candidates against Pseudomonas aeruginosa: An integrated genomics and proteomics approach**. *Genomics* (2017.0) **109** 274-283. DOI: 10.1016/j.ygeno.2017.05.001
27. Iyer V. **Impact of formulation and particle size on stability and immunogenicity of oil-in-water emulsion adjuvants**. *Hum. Vaccin Immunother.* (2015.0) **11** 1853-1864. DOI: 10.1080/21645515.2015.1046660
28. Denis J. **Immunogenicity of papaya mosaic virus-like particles fused to a hepatitis C virus epitope: evidence for the critical function of multimerization**. *Virology* (2007.0) **363** 59-68. DOI: 10.1016/j.virol.2007.01.011
29. Ratanji KD, Derrick JP, Dearman RJ, Kimber I. **Immunogenicity of therapeutic proteins: influence of aggregation**. *J. Immunotoxicol.* (2014.0) **11** 99-109. DOI: 10.3109/1547691X.2013.821564
30. Gregg KA. **Rationally Designed TLR4 Ligands for Vaccine Adjuvant Discovery**. *mBio* (2017.0) **8** e00492-17. DOI: 10.1128/mBio.00492-17
31. Rashid MI, Naz A, Ali A, Andleeb S. **Prediction of vaccine candidates against Pseudomonas aeruginosa: An integrated genomics and proteomics approach**. *Genomics* (2017.0) **109** 274-283. DOI: 10.1016/j.ygeno.2017.05.001
32. Laghaei P. **Immunogenicity and protective efficacy of Pseudomonas aeruginosa type a and b flagellin vaccines in a burned mouse model**. *Mol. Immunol.* (2016.0) **74** 71-81. DOI: 10.1016/j.molimm.2016.04.008
33. Pimentel LS. **Balanced Th1/Th2 immune response induced by MSP1a functional motif coupled to multiwalled carbon nanotubes as anti-anaplasmosis vaccine in murine model**. *Nanomedicine* (2020.0) **24** 102137. DOI: 10.1016/j.nano.2019.102137
34. Vidarsson G, Dekkers G, Rispens T. **IgG subclasses and allotypes: from structure to effector functions**. *Front Immunol.* (2014.0) **5** 520. DOI: 10.3389/fimmu.2014.00520
35. Irani V. **Molecular properties of human IgG subclasses and their implications for designing therapeutic monoclonal antibodies against infectious diseases**. *Mol. Immunol.* (2015.0) **67** 171-182. DOI: 10.1016/j.molimm.2015.03.255
36. Moser C. **Early immune response in susceptible and resistant mice strains with chronic Pseudomonas aeruginosa lung infection determines the type of T-helper cell response**. *Apmis* (1997.0) **107** 1093-1100. DOI: 10.1111/j.1699-0463.1999.tb01514.x
37. Kaiko GE, Horvat JC, Beagley KW, Hansbro PM. **Immunological decision-making: how does the immune system decide to mount a helper T-cell response?**. *Immunology* (2008.0) **123** 326-338. DOI: 10.1111/j.1365-2567.2007.02719.x
38. McAleer JP, Vella AT. **Understanding how lipopolysaccharide impacts CD4 T-cell immunity**. *Crit. Rev. Immunol.* (2008.0) **28** 281-299. DOI: 10.1615/CritRevImmunol.v28.i4.20
39. Yeste A. **IL-21 induces IL-22 production in CD4+ T cells**. *Nat. Commun.* (2014.0) **5** 3753. DOI: 10.1038/ncomms4753
40. Cuomo F. **Pro-inflammatory cytokines activate hypoxia-inducible factor 3α via epigenetic changes in mesenchymal stromal/stem cells**. *Sci. Rep.* (2018.0) **8** 5842. DOI: 10.1038/s41598-018-24221-5
41. Martinez-Becerra FJ. **Broadly protective Shigella vaccine based on type III secretion apparatus proteins**. *Infect. Immun.* (2012.0) **80** 1222-1231. DOI: 10.1128/IAI.06174-11
42. Lorè NI. **IL-17A impairs host tolerance during airway chronic infection by Pseudomonas aeruginosa**. *Sci. Rep.* (2016.0) **6** 25937. DOI: 10.1038/srep25937
43. McDaniel MS, Schoeb T, Swords WE. **Cooperativity between Stenotrophomonas maltophilia and Pseudomonas aeruginosa during Polymicrobial Airway Infections**. *Infect. Immun.* (2020.0) **88** e00855-19. DOI: 10.1128/IAI.00855-19
44. Lindgren NR, Novak L, Hunt BC, McDaniel MS, Swords WE. **Nontypeable Haemophilus influenzae Infection Impedes Pseudomonas aeruginosa Colonization and Persistence in Mouse Respiratory Tract**. *Infect. Immun.* (2022.0) **90** e0056821. DOI: 10.1128/iai.00568-21
|
---
title: Identification of potential biomarkers and therapeutic targets for posttraumatic
acute respiratory distress syndrome
authors:
- Peng Qi
- Mengjie Huang
- Tanshi Li
journal: BMC Medical Genomics
year: 2023
pmcid: PMC10012314
doi: 10.1186/s12920-023-01482-2
license: CC BY 4.0
---
# Identification of potential biomarkers and therapeutic targets for posttraumatic acute respiratory distress syndrome
## Abstract
### Background
Despite improved supportive care, posttraumatic acute respiratory distress syndrome (ARDS) mortality has improved very little in recent years. Additionally, ARDS diagnosis is delayed or missed in many patients. We analyzed co-differentially expressed genes (co-DEGs) to explore the relationships between severe trauma and ARDS to reveal potential biomarkers and therapeutic targets for posttraumatic ARDS.
### Methods
*Two* gene expression datasets (GSE64711 and GSE76293) were downloaded from the Gene Expression Omnibus. The GSE64711 dataset included a subset of 244 severely injured trauma patients and 21 healthy controls. GSE76293 specimens were collected from 12 patients with ARDS who were recruited from trauma intensive care units and 11 age- and sex-matched healthy volunteers. Trauma DEGs and ARDS DEGs were identified using the two datasets. Subsequently, Gene Ontology, Kyoto Encyclopedia of Genes and Genomes, and protein–protein interaction network analyses were performed to elucidate the molecular functions of the DEGs. Then, hub genes of the co-DEGs were identified. Finally, to explore whether posttraumatic ARDS and septic ARDS are common targets, we included a third dataset (GSE100159) for corresponding verification.
### Results
90 genes were upregulated and 48 genes were downregulated in the two datasets and were therefore named co-DEGs. These co-DEGs were significantly involved in multiple inflammation-, immunity- and neutrophil activation-related biological processes. Ten co-upregulated hub genes (GAPDH, MMP8, HGF, MAPK14, LCN2, CD163, ENO1, CD44, ARG1 and GADD45A) and five co-downregulated hub genes (HERC5, IFIT2, IFIT3, RSAD2 and IFIT1) may be considered potential biomarkers and therapeutic targets for posttraumatic ARDS. Through the verification of the third dataset, posttraumatic ARDS may have its own unique targets worthy of further exploration.
### Conclusion
This exploratory analysis supports a relationship between trauma and ARDS pathophysiology, specifically in relationship to the identified hub genes. These data may serve as potential biomarkers and therapeutic targets for posttraumatic ARDS.
## Background
Despite the continuous progress in the establishment of safety measures, trauma still accounts for more than one-tenth of deaths worldwide [1]. The burden is highest in individuals < 50years of age, among whom injury as a cause of death is second only to infectious diseases [2]. In addition to directly causing human tissue damage, trauma can cause a series of reactions both locally and systemically that lead to aggravation of the injury. The lung is the only organ that receives the entire cardiac output. In addition to being damaged by in situ-produced inflammatory mediators, it can be damaged by circulating inflammatory mediators produced by distant organs. Thus, lung injury occurs early, and severe injury can be caused by indirect traumatic factors. In 1967, Ashbaugh described 12 patients, including 7 patients with severe trauma who appeared to have acute hypoxemia, noncardiogenic pulmonary edema, reduced lung compliance (increased lung stiffness), increased work of breathing and the need for positive-pressure ventilation [3]. This series of syndromes was first named adult respiratory distress syndrome. It was subsequently renamed acute respiratory distress syndrome (ARDS) in a number of studies. Globally, ARDS affects approximately 3 million patients annually and accounts for $10\%$ of intensive care unit (ICU) admissions, and $24\%$ of patients with ARDS receive mechanical ventilation in the ICU [4]. In patients with traumatic injuries, inflammatory responses at the local and systemic levels affect the lung both directly and indirectly and are the common cause of ARDS. Approximately 5–$10\%$ of adult trauma patients develop ARDS [5], and up to $19\%$ of these patients are admitted to the ICU. Studies of patients with posttraumatic ARDS have identified a mortality of between 16 and $24\%$ [6], and the mortality of severely injured trauma patients with ARDS can reach 35–$45\%$ [7]. Despite decades of research and considerable advances in our understanding of the pathogenesis, risk factors and complication management of ARDS, there has been no change in the mortality rate of posttraumatic ARDS over the last four decades [8]. This observation suggests that posttraumatic ARDS has properties distinct from those of other forms of ARDS. Furthermore, although the definition and diagnosis of ARDS have been continuously improved, clinicians miss the diagnosis of $40\%$ of ARDS cases [8] because the assessment and diagnosis of ARDS are operator-dependent and partially subjective, leading to high interobserver variability and compromising diagnostic accuracy. In recent years, great effort has been dedicated to identifying biomarkers of ARDS. Precise diagnostic biomarkers and biomarkers that suggest the severity of ARDS may improve early diagnosis [9]. However, at present, diagnosis is still made in the absence of established biomarkers [10].
To identify and develop precise diagnostic biomarkers and therapeutic strategies for posttraumatic ARDS, a better understanding of the mechanisms leading to lung damage as well as recovery in trauma patients is essential. In our study, two datasets were downloaded from the Gene Expression Omnibus (GEO) and analyzed to identify co-differentially expressed genes (co-DEGs) associated with severe injury and ARDS. Then, we elucidated the molecular mechanisms of trauma-related DEGs and ARDS-related DEGs through functional and pathway analyses and protein–protein interaction (PPI) network analysis. Then, we screened potential hub genes and refer to common public databases for analysis one by one. Finally, since previous studies on ARDS were mostly focused on sepsis, there were few studies on posttraumatic ARDS targets. It is unclear whether they have a common pathway or a special target. However, the latest research suggests that the mortality rate of post-traumatic ARDS has not decreased due to the improvement of medical technology, but has increased [11]. Based on this, we introduced the third sepsis related ARDS database for parallel verification to explore whether the ARDS caused by the two causes is caused by a common target, so as to verify whether the specificity of posttraumatic ARDS is worthy of further independent research.
## Microarray data
Microarray studies in this paper were searched from the GEO database [12] (https://www.ncbi.nlm.nih.gov/geo/) using the terms “acute respiratory distress syndrome” and “trauma”. Two datasets (GSE64711 [13] and GSE76293 [14]) were selected for subsequent analysis. We used the “GEOquery” package [15] of R software (version 4.0.2, http://r-project.org/) to download the expression profile datasets GSE64711 and GSE76293 from the GEO database. The two datasets included all necessary information, and no samples had to be taken on site. The GSE64711 dataset included a subset of 244 severely injured trauma patients aged 16 to 90 years old, and an additional 21 healthy controls were enrolled for blood sampling for enriched polymorphonuclear neutrophil genomic analysis. Among them, the specimen collection time of trauma patients is within 4 days after the trauma, which also meets the requirements of the Berlin standard of ARDS. The inclusion criteria of GSE64711 included adult patients (age ≥ 16 years) who had been severely injured (injury severity score (ISS) > 15) after having undergone blunt trauma without severe traumatic brain injury (TBI) and with evidence of hemorrhagic shock (systolic blood pressure (SBP) < 90 mmHg or base deficit ≥ 6 mEq/L and requiring blood transfusion). Specimens of GSE76293 were collected from 12 patients with ARDS who were recruited from trauma ICUs in a U.K. teaching hospital and from 11 age- and sex-matched healthy volunteers. GSE64711 was profiled on a GPL19607 [hGlue1_0.r3] custom Affymetrix Human Transcriptome Array, and GSE76293 was profiled on a GPL570 [HG-U133_Plus_2] Affymetrix Human Genome U133 Plus 2.0 Array. All RNA information of the selected samples was downloaded and the original microarray datasets were normalized and preprocessed with R software.
## Identification of DEGs
We used the “limma” [16] package of R software to screen DEGs. A p value of < 0.05 and an absolute value of the log2 (fold change) of > 1 were the thresholds for identifying differences in gene expression as significant [16, 17]. Volcano plots were generated using the R package “ggplot2” [18].
The “limma” package of R software was used to screen DEGs in the three datasets. In the trauma dataset, the total number of IDs after removing the null values was 19,844, of which 337 IDs met the threshold of |log2(FC)|> 1 and p.adj < 0.05. The number of upregulated DEGs (log2(FC) > 1) was 172, and the number of downregulated DEGs (log2(FC) < − 1) was 165. In the ARDS dataset, the total number of IDs after removing the null values was 45,118, of which 894 IDs met the threshold of |log2(FC)|> 1 and p.adj < 0.05. The number of upregulated DEGs (log2(FC) > 1) was 541, and the number of downregulated DEGs (log2(FC) < − 1) was 353. The DEGs were visualized using the “ggplot2” package (Fig. 1).Fig. 1DEGs in the two datasets. A Volcano plot of GSE64711, red represents upregulated genes, blue represents downregulated genes, and gray represents no significantly expressed genes; B Volcano plot of GSE76293. The criteria for statistically significant difference of DEGs was adjusted |log2(FC)|> 1 and p.adj < 0.05 in expression
## Gene ontology (GO) and Kyoto encyclopedia of genes and genomes (KEGG) pathway enrichment analyses of DEGs
DEGs from the two datasets were screened for subsequent GO and KEGG pathway enrichment analyses. GO function analysis (with the cellular component [CC], biological process [BP], and molecular function [MF] categories) is a powerful bioinformatics tool to classify gene expression and its properties [19]. KEGG pathway analysis was used to identify the cell pathways in which the DEGs might be involved [20]. GO and KEGG pathway enrichment analyses were performed using the clusterProfiler [21] routine in R. $P \leq 0.05$ was considered statistically significant.
## PPI network construction and hub gene identification
We used the “VennDiagram” [22] package of R software to screen the co-upregulated and co-downregulated genes in the two datasets, which were constructed using the Search Tool for the Retrieval of Interacting Genes (STRING; http://string-db.org/) [23]. A minimum required interaction score > 0.4 was set as the cutoff point. Subsequently, Cytoscape software was used to construct and visualize molecular interaction networks. The hub genes in the PPI networks were selected using the Cyto-Hubba plug-in of Cytoscape.
The PPI network was constructed with the co-upregulated DEGs and co-downregulated DEGs, visualized and analyzed with Cytoscape software. Based on the identified DEGs, we constructed a PPI network for co-upregulated DEGs consisting of 57 nodes and 72 edges (Fig. 5A) and a PPI network for co-downregulated DEGs consissting of 21 nodes and 20 edges (Fig. 5B). The hub genes were selected by the maximal clique centrality (MCC) algorithm with the Cyto-Hubba plug-in of Cytoscape software and included glyceraldehyde-3-phosphate dehydrogenase (GAPDH), matrix metallopeptidase 8 (MMP8), hepatocyte growth factor (HGF), mitogen-activated protein kinase 14 (MAPK14), lipocalin 2 (LCN2), CD163 molecule (CD163), enolase 1 (EN01), CD44 molecule (CD44), arginase 1 (ARG1), and growth arrest and DNA-damage-inducible protein 45 alpha (GADD45A) which were co-upregulated hub DEGs. Besides, HECT and RLD domain containing E3 ubiquitin protein ligase 5 (HERC5), Interferon induced protein with tetratricopeptide repeats 2(IFIT2), Interferon induced protein with tetratricopeptide repeats 3(IFIT3), Radical S-adenosyl methionine domain containing 2(RSAD2), Interferon induced protein with tetratricopeptide repeats 1(IFIT1) were co-downregulated hub genes. Fig. 5PPI network construction A PPI network of the co-upregulated DEGs. Each node stands for a gene or a protein, and edges represent the interactions between the nodes; B PPI network of the down-upregulated DEGs
## Functional enrichment analysis of hub genes
The selected hub genes were analyzed by the PANTHER [24] classification system (http://pantherdb.org/), and the basic classification of each gene was obtained. Then, the Human Protein Atlas [25] (HPA) (https://www.proteinatlas.org/) was used to explore the expression profile of each hub gene in human tissue. Subsequently, the Deeply Integrated human Single-Cell Omics (DISCO) database [26] (https://www.immunesinglecell.org/) was used to clarify the expression level of the hub genes in lung. Finally, the “clusterProfiler” [21] package of R software was used for the enrichment analysis of the hub genes, and the “ggplot2” package of R software was used to visually present the results of the enrichment analysis. The results are displayed in a bubble plot and table.
Since the up-regulated DEG is a risk factor for the disease group, it is of greater significance for the occurrence and development of the disease. Therefore, we once again performed enrichment analysis on the co-upregulated hub genes to find out the commonness of this group of specific gene sets in biological composition / function / process. GO and KEGG pathway enrichment analyses of the ten co-upregulated hub genes (GAPDH, MMP8, HGF, MAPK14, LCN2, CD163, ENO1, CD44, ARG1 and GADD45A) were performed and the results showed that the main enriched BP terms were the neutrophil mediated immunity, neutrophil activation, neutrophil activation involved in immune response and neutrophil degranulation. The main enriched CC terms were the secretory granule lumen, cytoplasmic vesicle lumen and vesicle lumen, and the main enriched MF term were the serine-type endopeptidase activity, serine-type peptidase activity and serine hydrolase activity. The KEGG pathway analysis mainly showed enrichment of the terms biosynthesis of amino acids, Epstein–*Barr virus* infection and proteoglycans in cancer. All results are displayed in a bubble plot (Fig. 8A–D) and table (Table 6).Fig. 8Functional enrichment analysis of the ten co-upregulated hub genes and the Venn diagrams. A Shows the results of biological process terms enriched by BP analysis; B Shows the results of biological process terms enriched by MF analysis; C Shows the results of biological process terms enriched by CC analysis; D Shows the enriched pathway by KEGG analysis; E the Venn diagrams between the posttraumatic ARDS and the septic ARDS groups in hub genesTable 6Details of GO terms and KEGG pathway enrichment in co-upregulated hub genesOntologyIDDescriptionGeneRatioBgRatiopvaluep.adjustqvalueBPGO:0038066p38MAPK cascade$\frac{3}{1053}$/186702.56e−065.28e−042.04e−04BPGO:0043312Neutrophil degranulation$\frac{5}{10485}$/186702.62e−065.28e−042.04e−04BPGO:0002283Neutrophil activation involved in immune response$\frac{5}{10488}$/186702.70e−065.28e−042.04e−04BPGO:0042119Neutrophil activation$\frac{5}{10498}$/186702.99e−065.28e−042.04e−04BPGO:0002446Neutrophil mediated immunity$\frac{5}{10499}$/186703.02e−065.28e−042.04e−04CCGO:0034774Secretory granule lumen$\frac{5}{10321}$/197172.61e−076.17e−064.57e−06CCGO:0060205Cytoplasmic vesicle lumen$\frac{5}{10338}$/197173.38e−076.17e−064.57e−06CCGO:0031983Vesicle lumen$\frac{5}{10339}$/197173.43e−076.17e−064.57e−06CCGO:0035580Specific granule lumen$\frac{3}{1062}$/197173.50e−064.72e−053.50e−05CCGO:0042581Specific granule$\frac{3}{10160}$/197176.04e−056.52e−044.83e−04MFGO:0004252Serine-type endopeptidase activity$\frac{2}{10160}$/176970.0030.0770.042MFGO:0008236Serine-type peptidase activity$\frac{2}{10182}$/176970.0040.0770.042MFGO:0017171Serine hydrolase activity$\frac{2}{10186}$/176970.0050.0770.042MFGO:0016813Hydrolase activity, acting on carbon–nitrogen (but not peptide) bonds, in linear amidines$\frac{1}{1011}$/176970.0060.0770.042MFGO:0004707MAP kinase activity$\frac{1}{1014}$/176970.0080.0770.042KEGGhsa01230Biosynthesis of amino acids$\frac{3}{875}$/80764.17e−050.0040.003KEGGhsa05169Epstein–*Barr virus* infection$\frac{3}{8202}$/80767.87e−040.0250.020KEGGhsa05205Proteoglycans in cancer$\frac{3}{8205}$/80768.21e−040.0250.020KEGGhsa00010Glycolysis/gluconeogenesis$\frac{2}{867}$/80760.0020.0300.024KEGGhsa05218Melanoma$\frac{2}{872}$/80760.0020.0300.024
## Verification of hub genes
To explore whether posttraumatic ARDS and septic ARDS are common targets, or posttraumatic ARDS has its unique pathways and targets. We included a third data set for corresponding verification. We searched the GEO database with the keyword “sepsis”. Then searched using keywords and restricted the screening conditions to “Homo sapiens” and “Expression profiling by array” and limited the sample to peripheral blood. After reading the original studies on the data sets that met the screening conditions individually, we focused on the experimental design and methodology. The data set that met the principle of randomized control and did not violate ethical guidelines was taken as the research material.
Through conditional screening, we ultimately selected GSE100159 as the research object. The GSE100159 data set includes 35 sepsis patients and 12 healthy controls. The chip platform used is a GPL6884 Illumina HumanWG-6 v3.0 expression beadchip. Using the aforementioned methodology, we performed background correction and data normalization on the GSE100159 data set and then used the "limma" package of R software to analyze the disease group and the healthy control group to obtain DEGs. The inclusion criteria for DEGs were the same as those for the above two data sets. The DEGs of the GSE100159 and GSE76293 data sets were merged and analyzed, the co-DEGs were screened out, a PPI network was constructed, and the hub genes were extracted by Cytoscape software. The final results showed that the hub genes in the two data sets were FCGR1A, MPO, CR1, CEACAM8, CD163, ITGA2B, ITGB3, CD44, THBS1, PKM, RPL18A, RPL19, RPL8, RPL30, RPL18, RPL17, RPL27, RPL13, RPL10A and RPL22. It can be seen from the results that the hub genes of trauma and ARDS are not completely the same as those of sepsis and ARDS (Fig. 8E), which on the other hand shows that posttraumatic ARDS has its own characteristic targets and is worthy of further exploration.
## Data details and preprocessing
We analyzed peripheral blood samples from three independent GEO datasets. A total of 244 severely injured trauma patients, 12 patients with ARDS, 35 patients with sepsis and 44 healthy controls were included in this study. Ethical approval was not necessary because our study was a bioinformatic analysis.
## GO and KEGG pathway enrichment analyses of DEGs
GO and KEGG pathway enrichment analyses of the DEGs of GSE64711 and GSE76293 were performed using the clusterProfiler routine in R, and enrichment analysis results were visualized using the R package “ggplot2” (Figs. 2, 3). The significantly enriched CC, BP, and MF terms and metabolic pathways of the DEGs of severe injury trauma and posttraumatic ARDS are shown (Tables 1, 2).Fig. 2GO and KEGG enrichment analysis of DEGs in GSE64711. A Shows the results of biological process terms enriched by BP analysis; B Shows the results of biological process terms enriched by MF; C Shows the results of biological process terms enriched by CC analysis; D Shows the enriched pathway by KEGG analysis. The coloured dots represent the P-value for that term, with red representing greater significance. The size of the dots represents the number of involved genesFig. 3GO and KEGG enrichment analysis of DEGs in GSE76293. A Shows the results of biological process terms enriched by BP analysis; B Shows the results of biological process terms enriched by MF; C Shows the results of biological process terms enriched by CC analysis; D Shows the enriched pathway by KEGG analysisTable 1Details for GO terms and KEGG pathway enrichment in GSE64711OntologyIDDescriptionGeneRatioBgRatiopvaluep.adjustqvalueBPGO:0042119Neutrophil activation$\frac{41}{269498}$/186701.10e−193.90e−163.15e−16BPGO:0002446Neutrophil mediated immunity$\frac{40}{269499}$/186708.35e−191.48e−151.19e−15BPGO:0043312Neutrophil degranulation$\frac{39}{269485}$/186702.13e−182.34e−151.89e−15CCGO:0042581Specific granule$\frac{21}{276160}$/197179.19e−153.20e−122.91e−12CCGO:0070820Tertiary granule$\frac{20}{276164}$/197171.68e−132.93e−112.66e−11CCGO:0035580Specific granule lumen$\frac{11}{27662}$/197178.95e−101.04e−079.46e−08MFGO:0004953Icosanoid receptor activity$\frac{4}{26815}$/176976.15e−050.0310.029MFGO:0030246Carbohydrate binding$\frac{13}{268271}$/176972.63e−040.0670.061MFGO:0016810Hydrolase activity, acting on carbon–nitrogen (but not peptide) bonds$\frac{8}{268123}$/176975.72e−040.0810.074KEGGhsa05140Leishmaniasis$\frac{7}{14177}$/80763.70e−040.0700.063KEGGhsa04066HIF-1 signaling pathway$\frac{8}{141109}$/80766.10e−040.0700.063KEGGhsa04621NOD-like receptor signaling pathway$\frac{10}{141181}$/80760.0010.0730.066Table 2Details for GO terms and KEGG pathway enrichment in GSE76293OntologyIDDescriptionGeneRatioBgRatiopvaluep.adjustqvalueBPGO:0042119Neutrophil activation$\frac{21}{86498}$/186706.74e−151.27e−111.05e−11BPGO:0043312Neutrophil degranulation$\frac{20}{86485}$/186704.98e−143.52e−112.89e−11BPGO:0002283Neutrophil activation involved in immune response$\frac{20}{86488}$/186705.59e−143.52e−112.89e−11BPGO:0002446Neutrophil mediated immunity$\frac{20}{86499}$/186708.48e−144.01e−113.29e−11BPGO:0001819Positive regulation of cytokine production$\frac{10}{86464}$/186705.34e−050.0180.015 The “VennDiagram” package of R was used to screen 90 co-upregulated DEGs and 48 co-downregulated DEGs from the two datasets (Fig. 4A, B). GO term enrichment analysis was performed on the common DEGs (Fig. 4C, D), and the results showed that the BPs of co-upregulated DEGs were mainly related to neutrophil activation, neutrophil degranulation, neutrophil activation involved in the immune response, neutrophil-mediated immunity and positive regulation of cytokine production, These BP processes are also the current focus of intense attention in ARDS. At the same time, we tabulated the results of the enrichment analysis of co-upregulated differential genes and co-downregulated differential genes in the form of a third line table (Tables 3, 4).Fig. 4Venn diagrams and GO term enrichment analysis on common DEGs. A 90 co-upregulated DEGs were screened from the two datasets; B 48 co-downregulated DEGs were screened from the two datasets; C Shows the results of GO term enrichment analysis perform on the 90 co-upregulated DEGs; D Shows the results of GO term enrichment analysis perform on the 48 down-upregulated DEGsTable 3Details for co-upregulated DEGs of GO terms enrichment in GSE64711 and GSE76293OntologyIDDescriptionGeneRatioBgRatiopvaluep.adjustqvalueBPGO:0042119Neutrophil activation$\frac{21}{86498}$/186706.74e−151.27e−111.05e−11BPGO:0043312Neutrophil degranulation$\frac{20}{86485}$/186704.98e−143.52e−112.89e−11BPGO:0002283Neutrophil activation involved in immune response$\frac{20}{86488}$/186705.59e−143.52e−112.89e−11BPGO:0002446Neutrophil mediated immunity$\frac{20}{86499}$/186708.48e−144.01e−113.29e−11BPGO:0001819Positive regulation of cytokine production$\frac{10}{86464}$/186705.34e−050.0180.015CCGO:0042581Specific granule$\frac{13}{88160}$/197172.87e−134.36e−113.75e−11CCGO:0070820Tertiary granule$\frac{10}{88164}$/197173.11e−092.36e−072.03e−07CCGO:0035580Specific granule lumen$\frac{7}{8862}$/197171.12e−085.65e−074.85e−07CCGO:0034774Secretory granule lumen$\frac{12}{88321}$/197171.92e−087.29e−076.26e−07CCGO:0060205Cytoplasmic vesicle lumen$\frac{12}{88338}$/197173.39e−088.87e−077.62e−07Table 4Details for co-downregulated DEGs of GO terms enrichment in GSE64711 and GSE76293OntologyIDDescriptionGeneRatioBgRatiopvaluep.adjustqvalueBPGO:0051607Defense response to virus$\frac{8}{43238}$/186706.12e−088.26e−056.81e−05BPGO:0009615Response to virus$\frac{8}{43326}$/186706.76e−074.56e−043.76e−04BPGO:0043393Regulation of protein binding$\frac{6}{43217}$/186709.80e−060.0040.004BPGO:0002460Adaptive immune response based on somatic recombination of immune receptors built from immunoglobulin superfamily domains$\frac{7}{43361}$/186701.69e−050.0060.005BPGO:0060337Type I interferon signaling pathway$\frac{4}{4395}$/186706.67e−050.0130.011MFGO:0004966Galanin receptor activity$\frac{2}{4410}$/176972.68e−040.0140.011MFGO:0001608G protein-coupled nucleotide receptor activity$\frac{2}{4413}$/176974.63e−040.0140.011MFGO:0045028G protein-coupled purinergic nucleotide receptor activity$\frac{2}{4413}$/176974.63e−040.0140.011MFGO:0008528G protein-coupled peptide receptor activity$\frac{4}{44146}$/176974.67e−040.0140.011MFGO:0001653Peptide receptor activity$\frac{4}{44152}$/176975.44e−040.0140.011
## Analysis of hub genes
The PANTHER [24] classification system was used to analyze the selected hub genes, and the basic classification of each gene was obtained. The results are listed in a table (Table 5). To explore the expression of the hub genes in human tissue, the HPA database was used to analyze each hub gene (Fig. 6). The HPA RNA-seq tissue data is reported as nTPM (normalized protein-coding transcripts per million), corresponding to mean values of the different individual samples from each tissue. Color-coding is based on tissue groups, each consisting of tissues with functional features in common [25]. The DISCO [26] database was used annotate the single-cell sequencing data of cells from lung. The expression level of each hub gene in each cell is shown. The deeper the color is, the higher the expression of the gene in the cell (Fig. 7).Table 5The PANTHER classification system of hub genesIDGene nameGene IDPANTHER family/subfamilyPANTHER protein classENO1Alpha-enolase ENO1 orthologHUMAN|HGNC = 3350|UniProtKB = P06733Alpha-enolase (PTHR11902:SF12)LyaseMMP8Neutrophil collagenase MMP8 orthologHUMAN|HGNC = 7175|UniProtKB = P22894Neutrophil collagenase (PTHR10201:SF137)MetalloproteaseCD44CD44 antigen CD44 orthologHUMAN|HGNC = 1681|UniProtKB = P16070CD44 antigen (PTHR10225:SF6)Transmembrane signal receptorHGFHepatocyte growth factor HGF orthologHUMAN|HGNC = 4893|UniProtKB = P14210Hepatocyte growth factor (PTHR24261:SF8)Serine proteaseGADD45AGrowth arrest and DNA damage-inducible protein GADD45 alpha GADD45A orthologHUMAN|HGNC = 4095|UniProtKB = P24522Growth arrest and DNA damage-inducible protein GADD45 alpha (PTHR10411:SF3)–MAPK14Mitogen-activated protein kinase 14 MAPK14 orthologHUMAN|HGNC = 6876|UniProtKB = Q16539Mitogen-activated protein kinase 14 (PTHR24055:SF110)Non-receptor serine/threonine protein kinaseCD163Scavenger receptor cysteine-rich type 1 protein M130 CD163 orthologHUMAN|HGNC = 1631|UniProtKB = Q86VB7Scavenger receptor cysteine-rich type 1 protein M130 (PTHR19331:SF441)Serine proteaseLCN2Neutrophil gelatinase-associated lipocalin LCN2 orthologHUMAN|HGNC = 6526|UniProtKB = P80188Neutrophil gelatinase-associated lipocalin (PTHR11430:SF13)Transfer/carrier proteinGAPDHGlyceraldehyde-3-phosphate dehydrogenase GAPDH orthologHUMAN|HGNC = 4141|UniProtKB = P04406Glyceraldehyde-3-phosphate dehydrogenase (PTHR10836:SF111)DehydrogenaseARG1Arginase-1 ARG1 orthologHUMAN|HGNC = 663|UniProtKB = P05089Arginase-1 (PTHR43782:SF2)HydrolaseHERC5E3 ISG15–protein ligase HERC5HERC5HUMAN|HGNC = 24368|UniProtKB = Q9UII4E3 ISG15–protein ligase HERC5 (PTHR45622:SF7)Ubiquitin-protein ligaseIFIT2Interferon-induced protein with tetratricopeptide repeats 2 IFIT2HUMAN|HGNC = 5409|UniProtKB = P09913Interferon-induced protein with tetratricopeptide repeats 2 (PTHR10271:SF4)Defense/immunity proteinIFIT3Interferon-induced protein with tetratricopeptide repeats 3 IFIT3HUMAN|HGNC = 5411|UniProtKB = O14879Interferon-induced protein with tetratricopeptide repeats 3 (PTHR10271:SF3)Defense/immunity proteinRSAD2Radical S-adenosyl methionine domain-containing protein 2 RSAD2HUMAN|HGNC = 30908|UniProtKB = Q8WXG1Radical S-adenosyl methionine domain-containing protein 2 (PTHR21339:SF0)–IFIT1Interferon-induced protein with tetratricopeptide repeats 1 IFIT1HUMAN|HGNC = 5407|UniProtKB = P09914Interferon-induced protein with tetratricopeptide repeats 1 (PTHR10271:SF16)Defense/immunity proteinFig. 6RNA expression overview shows RNA-seq tissue data from internally generated Human Protein Atlas (HPA) data. Color-coding is based on tissue groups, each consisting of tissues with functional features in common. A GAPDH B MMP8 C HGF D MAPK14 E LCN2 F CD163 G ENO1 H CD44 I ARG1 J GADD45A K HERC5 L IFIT2 M IFIT3 N RSAD2 O IFIT1Fig. 7The DISCO database was used annotate the single-cell sequencing data of cells from lung. The deeper the color is, the higher the expression of the gene in the cell. A GAPDH B MMP8 C HGF D MAPK14 E LCN2 F CD163 G ENO1 H CD44 I ARG1 J GADD45A K HERC5 L IFIT2 M IFIT3 N RSAD2 O IFIT1
## Discussion
Trauma associated with severe injury not only leads to direct damage in patients but also causes a secondary systemic inflammatory response, which could result in systemic inflammatory response syndrome and is strongly correlated with the development of severe posttraumatic multiple organ dysfunction syndrome (MODS) [1]. Due to the uncontrollable inflammatory reactions in the lung tissue, a large number of neutrophils, macrophages, and other inflammatory cells accumulate in the alveolus [27, 28]. The incidence of posttraumatic ARDS was significantly higher than that of other organs. Some studies have even reported that the probability of ARDS in trauma patients is $29\%$ within 24 h, and on the fifth day after trauma, more than $90\%$ of patients have ARDS [29]. In recent years, although great strides have been made in medicine, there was no change in the mortality rate from posttraumatic ARDS [8]. Early diagnosis, risk assessment and timely treatments in the initial periods of posttraumatic ARDS play a crucial role in reducing mortality [30]. The identification and development of new biomarkers can provide major insights into the pathophysiologic mechanisms underlying posttraumatic ARDS and can be helpful for the diagnosis, risk stratification and identification of candidate therapeutic targets [10, 31]. Bioinformatic analyses enable us to understand the molecular mechanisms of disease occurrence and development, providing a novel and effective way to identify potential diagnostic biomarkers and therapeutic targets as early-warning signals and for the timely treatment of posttraumatic ARDS [32]. In the course of this study, the hub genes were identified and analyzed to screen ten co-upregulated genes (GAPDH, MMP8, HGF, MAPK14, LCN2, CD163, ENO1, CD44, ARG1 and GADD45A) and five co-downregulated genes (HERC5, IFIT2, IFIT3, RSAD2 and IFIT1). Because the up-regulated DEGs are risk factors for the disease group, we have demonstrated the existing research results and related possible inferences of these ten genes one by one through previous literature research. First of all, GAPDH is well established as one of the molecules promoting apoptotic signaling in the cell nucleus. The role of GAPDH in regulating inflammation has been demonstrated by several studies [33, 34]. The damage to the heme chaperone caused by GAPDH nitrosylation leads to a decrease in catalase activity, which is a typical feature of the inflammatory process. GAPDH might also affect the inflammatory process through the regulation of tumor necrosis factor synthesis [35], with GAPDH-mediated proinflammatory cascades occurring after severe injury and sepsis [36, 37]. In such cases, blocking the inflammatory response is an important part of effective treatment. Therefore, in theory, GAPDH is a potential drug target. This speculation was confirmed in this study. This study provides a therapeutic target for the treatment of posttraumatic ARDS. Matrix metalloproteinases (MMPs) play major roles in cell differentiation, proliferation, wound healing, apoptosis and angiogenesis [38]. They also contribute to the pathogenesis of various diseases and conditions, such as inflammation, atherosclerosis and myocardial infarction [39, 40]. In previous decades, MMPs were considered to play only extracellular roles; however, this concept has been challenged in recent years. Further research is needed to clarify the functions of MMP8 within the cell. Understanding the biological functions of MMPs in cells is essential not only for understanding their physiological functions but also for discovering new therapeutic targets for the treatment of various pathologies. In identifying MMPs as a target for posttraumatic ARDS, this study suggests a new research direction [41]. HGF was first defined as the mitogenic protein of mature liver cells in 1984. It is a multifunctional cytokine that participates in cell morphogenesis, survival and proliferation and has anti-inflammatory effects [38, 39]. HGF seems to be related to secondary inflammation or anti-inflammatory effects, but the mechanism by which HGF regulates the immune response has not been resolved, and further research is needed. MAPK14, also known as cytokine inhibitory anti-inflammatory drug binding protein, is an osmotic regulator protein kinase that can be activated by exposure to many types of cellular stress. It plays a key role in triggering different disease states, such as inflammatory diseases, neurodegenerative diseases, cardiovascular diseases and cancer [42]. The MAPK14 pathway is closely related to many chronic inflammatory factors. These factors contribute to the production of proinflammatory cytokines and are essential in diseases such as Crohn's disease and chronic asthma [43, 44]. Some studies have shown that activated MAPK14 is highly expressed in the alveoli of smokers with chronic obstructive pulmonary disease (COPD). Therefore, inhibition of MAPK14 may be a valuable drug target for the treatment of COPD, and such inhibition for the treatment of ARDS, which involves the same pathological state of the respiratory system, also warrants study [45]. In addition, some studies have reported that MAPK14 is a valuable therapeutic target for acute or chronic inflammatory diseases [46]. LCN2, also known as neutrophil gelatinase-associated lipocalin, is a new type of adipocyte factor with 198 amino acids [47]. Several studies have shown that TNFα induces the expression and secretion of LCN2 [48], and lipopolysaccharide is a strong inducer of LCN2 expression in various tissues [49]. In addition, the expression of LCN2 mRNA in bronchial epithelium and type II lung cells has been found to be significantly increased in patients with lung inflammation [50]. These findings indicate that there is a direct or indirect link between LCN2 and posttraumatic ARDS. The expression of CD163 is upregulated in many diseases, but our understanding of the pathological role of this receptor in diseases seems incomplete [51]. CD163 binds to and degrades inflammatory cytokines, is a weak inducer of tumor necrosis factor-like cell apoptosis [52], and recognizes and mediates local immune responses to bacteria [53] and to internalize viruses [54]. Therefore, the macrophage scavenger receptor CD163, which is upregulated in many inflammatory and malignant diseases, is a promising target. As a part-time protein, ENO1 has a variety of biochemical functions. ENO1 catalyzes the conversion of 2-phosphoglycerate to phosphoenolpyruvate, which is an important step in glycolysis, and it plays an important role in multiple pathophysiological processes [55]. ENO1 binds to guanylate-binding protein and negatively regulates T cell signaling by interfering with early T cell receptor signaling [56]. Posttranslational modification activities are also important for the function of ENO1 in immunity [57, 58]. Inflammatory stimulation can induce the translocation of ENO1 from the cytoplasm to the cell membrane [59, 60]. Therefore, the conversion of ENO1 localization is related to inflammatory pathology, and its role in the pathology of posttraumatic ARDS warrants further study. The nonkinase transmembrane glycoprotein CD44 was first described as a lymphocyte homing receptor in 1983 and has attracted considerable interest recently. Generally, CD44 is widely expressed on vertebrate cells, and its ability to regulate tumor progression, metastasis, and disease prognosis has been extensively explored [61]. CD44 is a multifunctional transmembrane glycoprotein receptor that binds to hyaluronic acid. Extracellular and intracellular hyaluronic acid binds to CD44 and affects cell behavior. As a receptor, CD44 can trigger signal cascades that regulate cell functional properties, such as proliferation, migration, angiogenesis, and wound healing. The regulation of these pathways may be critical to the development of pathological conditions such as inflammation and cancer [62]. ARG1 is a cytoplasmic enzyme that is expressed in macrophages, bone marrow-derived suppressor cells, dendritic cells, and innate lymphoid group 2 cells in response to Th2-type cytokines (IL-4 and IL-13) [63, 64] and infection with pathogens related to other signaling factors [64]. In humans, ARG1 exists in the granulocyte granular compartment of healthy subjects [65], the peripheral blood mononuclear cells of patients after injury, and the activated monocytes of patients with autoimmune diseases. For patients with autoimmune imbalance after trauma, ARG1 has research value. GADD45A exhibits a variety of important functions in cells, including the inhibition of cell growth, the mediation of cell cycle arrest at G2/M, the induction of apoptosis, and interaction with p53, cyclin-dependent kinase 1 (CDK1), Cdc2 and cyclin B1 [66]. In summary, these 10 hub genes obtained through screening are theoretically related to posttraumatic ARDS. However, further research is needed to predict and treat the exact mechanisms underlying ARDS.
Our study has several limitations. The first limitation is that this study is exploratory in nature. Thus, further experimental studies and clinical trials should be carried out to obtain accurate verification and to validate our results. Another limitation to this research is that gene expression patterns are also dependent on important underlying comorbid conditions and can additionally be dependent on age. Specifically, the upregulated (or downregulated) genes are not specific to posttraumatic ARDS and may be upregulated (or downregulated) even more in certain patient populations, and/or the gene expression findings may be “driven” by cohorts of patients with certain chronic inflammatory processes or age-dependent inflammatory processes. The linkage of the acute and chronic inflammatory processes is imperative for further identifying the patients with “treatable traits” that would be most responsive and potentially least harmful to the treatment being tested in clinical trials. Because this study included information from a public dataset, secondary classification of subtypes could not be performed, which may result in inapplicability of the results to certain groups. Further clinical trials are needed to validate the subtypes. In addition, although the data analyzed represent RNA expression patterns within the first 4 days of a critical illness, the time course of the first days after onset of a critical illness, whether sepsis, ARDS or trauma, is often associated with a fluctuating inflammatory response. Therefore, future studies of similar RNA expression patterns should include more discrete time points for collecting RNA in relation to the onset of critical illness.
## Conclusion
In combination with data from previous studies and bioinformatic analyses, our study found that GAPDH, MMP8, HGF, MAPK14, LCN2, CD163, ENO1, CD44, ARG1, GADD45A, HERC5, IFIT2, IFIT3, RSAD2 and IFIT1 were related to the potential common mechanisms between severe injury trauma and ARDS. In addition, it can be seen from the results that the hub genes of trauma and ARDS are not the same as those of sepsis and ARDS, which on the other hand shows that posttraumatic ARDS has its own characteristic targets and is worthy of further exploration. These findings shed new light on the diagnosis of posttraumatic ARDS and identify candidate targets for therapeutic intervention. Further research will be needed to explore these possibilities.
## References
1. Messerer DAC, Halbgebauer R, Nilsson B, Pavenstädt H, Radermacher P, Huber-Lang M. **Immunopathophysiology of trauma-related acute kidney injury**. *Nat Rev Nephrol* (2021.0) **17** 91-111. DOI: 10.1038/s41581-020-00344-9
2. Moore EE, Moore HB, Kornblith LZ, Neal MD, Hoffman M, Mutch NJ. **Trauma-induced coagulopathy**. *Nat Rev Dis Primers* (2021.0) **7** 30. DOI: 10.1038/s41572-021-00264-3
3. Matthay MA, Zemans RL, Zimmerman GA, Arabi YM, Beitler JR, Mercat A. **Acute respiratory distress syndrome**. *Nat Rev Dis Primers* (2019.0) **5** 18. DOI: 10.1038/s41572-019-0069-0
4. Fan E, Brodie D, Slutsky AS. **Acute respiratory distress syndrome: advances in diagnosis and treatment**. *JAMA* (2018.0) **319** 698-710. DOI: 10.1001/jama.2017.21907
5. Pfeifer R, Heussen N, Michalewicz E, Hilgers RD, Pape HC. **Incidence of adult respiratory distress syndrome in trauma patients: a systematic review and meta-analysis over a period of three decades**. *J Trauma Acute Care Surg* (2017.0) **83** 496-506. DOI: 10.1097/TA.0000000000001571
6. Afshar M, Smith GS, Cooper RS, Murthi S, Netzer G. **Trauma indices for prediction of acute respiratory distress syndrome**. *J Surg Res* (2016.0) **201** 394-401. DOI: 10.1016/j.jss.2015.11.050
7. Howard BM, Kornblith LZ, Hendrickson CM, Redick BJ, Conroy AS, Nelson MF. **Differences in degree, differences in kind: characterizing lung injury in trauma**. *J Trauma Acute Care Surg* (2015.0) **78** 735-741. DOI: 10.1097/TA.0000000000000583
8. Birkner DR, Halvachizadeh S, Pape HC, Pfeifer R. **Mortality of adult respiratory distress syndrome in trauma patients: a systematic review over a period of four decades**. *World J Surg* (2020.0) **44** 2243-2254. DOI: 10.1007/s00268-020-05483-5
9. Beitler JR, Goligher EC, Schmidt M, Spieth PM, Zanella A, Martin-Loeches I. **Personalized medicine for ARDS: the 2035 research agenda**. *Intensive Care Med* (2016.0) **42** 756-767. DOI: 10.1007/s00134-016-4331-6
10. Jabaudon M, Blondonnet R, Ware LB. **Biomarkers in acute respiratory distress syndrome**. *Curr Opin Crit Care* (2021.0) **27** 46-54. DOI: 10.1097/MCC.0000000000000786
11. Kasotakis G, Stanfield B, Haines K, Vatsaas C, Alger A, Vaslef SN. **Acute respiratory distress syndrome (ARDS) after trauma: Improving incidence, but increasing mortality**. *J Crit Care* (2021.0) **64** 213-218. DOI: 10.1016/j.jcrc.2021.05.003
12. Barrett T, Wilhite SE, Ledoux P, Evangelista C, Kim IF, Tomashevsky M. **NCBI GEO: archive for functional genomics data sets–update**. *Nucleic Acids Res* (2013.0) **41** D991-D995. PMID: 23193258
13. Vanzant EL, Hilton RE, Lopez CM, Zhang J, Ungaro RF, Gentile LF. **Advanced age is associated with worsened outcomes and a unique genomic response in severely injured patients with hemorrhagic shock**. *Crit Care* (2015.0) **19** 77. DOI: 10.1186/s13054-015-0788-x
14. Juss JK, House D, Amour A, Begg M, Herre J, Storisteanu DM. **Acute respiratory distress syndrome neutrophils have a distinct phenotype and are resistant to phosphoinositide 3-kinase inhibition**. *Am J Respir Crit Care Med* (2016.0) **194** 961-973. DOI: 10.1164/rccm.201509-1818OC
15. Davis S, Meltzer PS. **GEOquery: a bridge between the gene expression omnibus (GEO) and BioConductor**. *Bioinformatics* (2007.0) **23** 1846-1847. DOI: 10.1093/bioinformatics/btm254
16. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W. **limma powers differential expression analyses for RNA-sequencing and microarray studies**. *Nucleic Acids Res* (2015.0) **43** e47. DOI: 10.1093/nar/gkv007
17. Shankar R, Leimanis ML, Newbury PA, Liu K, Xing J, Nedveck D. **Gene expression signatures identify paediatric patients with multiple organ dysfunction who require advanced life support in the intensive care unit**. *EBioMedicine* (2020.0) **62** 103122. DOI: 10.1016/j.ebiom.2020.103122
18. Ito K, Murphy D. **Application of ggplot2 to pharmacometric graphics**. *CPT Pharmacomet Syst Pharmacol* (2013.0) **2** e79. DOI: 10.1038/psp.2013.56
19. Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM. **Gene ontology: tool for the unification of biology**. *Gene Ontol Consort Nat Genet* (2000.0) **25** 25-29. DOI: 10.1038/75556
20. Altermann E, Klaenhammer TR. **PathwayVoyager: pathway mapping using the Kyoto Encyclopedia of genes and genomes (KEGG) database**. *BMC Genom* (2005.0) **6** 60. DOI: 10.1186/1471-2164-6-60
21. Yu G, Wang LG, Han Y, He QY. **clusterProfiler: an R package for comparing biological themes among gene clusters**. *OMICS* (2012.0) **16** 284-287. DOI: 10.1089/omi.2011.0118
22. Chen H, Boutros PC. **VennDiagram: a package for the generation of highly-customizable Venn and Euler diagrams in R**. *BMC Bioinformatics* (2011.0) **12** 35. DOI: 10.1186/1471-2105-12-35
23. Szklarczyk D, Gable AL, Lyon D, Junge A, Wyder S, Huerta-Cepas J. **STRING v11: protein-protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets**. *Nucleic Acids Res* (2019.0) **47** D607-D613. DOI: 10.1093/nar/gky1131
24. Mi H, Muruganujan A, Huang X, Ebert D, Mills C, Guo X. **Protocol update for large-scale genome and gene function analysis with the PANTHER classification system (v.14.0)**. *Nat Protoc* (2019.0) **14** 703-721. DOI: 10.1038/s41596-019-0128-8
25. Uhlen M, Karlsson MJ, Zhong W, Tebani A, Pou C, Mikes J. **A genome-wide transcriptomic analysis of protein-coding genes in human blood cells**. *Science* (2019.0) **366** eaax9198. DOI: 10.1126/science.aax9198
26. Li M, Zhang X, Ang KS, Ling J, Sethi R, Lee NYS. **DISCO: a database of deeply integrated human single-cell omics data**. *Nucleic Acids Res* (2022.0) **50** D596-d602. DOI: 10.1093/nar/gkab1020
27. Vassallo A, Wood AJ, Subburayalu J, Summers C, Chilvers ER. **The counter-intuitive role of the neutrophil in the acute respiratory distress syndrome**. *Br Med Bull* (2019.0) **131** 43-55. DOI: 10.1093/bmb/ldz024
28. Chen X, Tang J, Shuai W, Meng J, Feng J, Han Z. **Macrophage polarization and its role in the pathogenesis of acute lung injury/acute respiratory distress syndrome**. *Inflamm Res* (2020.0) **69** 883-895. DOI: 10.1007/s00011-020-01378-2
29. Cheng Q, Lai X, Yang L, Yang H, Luo Y. **Serum CD5L predicts acute lung parenchymal injury and acute respiratory distress syndrome in trauma patients**. *Medicine* (2021.0) **100** e27219. DOI: 10.1097/MD.0000000000027219
30. 30.Tran A, Fernando SM, Brochard LJ, Fan E, Inaba K, Ferguson ND, et al. Prognostic factors for development of acute respiratory distress syndrome following traumatic injury—a systematic review and meta-analysis. Eur Respir J. 2021.
31. van der Zee P, Rietdijk W, Somhorst P, Endeman H, Gommers D. **A systematic review of biomarkers multivariately associated with acute respiratory distress syndrome development and mortality**. *Crit Care* (2020.0) **24** 243. DOI: 10.1186/s13054-020-02913-7
32. Tang Y, Yang X, Shu H, Yu Y, Pan S, Xu J. **Bioinformatic analysis identifies potential biomarkers and therapeutic targets of septic-shock-associated acute kidney injury**. *Hereditas* (2021.0) **158** 13. DOI: 10.1186/s41065-021-00176-y
33. Galván-Peña S, Carroll RG, Newman C, Hinchy EC, Palsson-McDermott E, Robinson EK. **Malonylation of GAPDH is an inflammatory signal in macrophages**. *Nat Commun* (2019.0) **10** 338. DOI: 10.1038/s41467-018-08187-6
34. Chakravarti R, Gupta K, Majors A, Ruple L, Aronica M, Stuehr DJ. **Novel insights in mammalian catalase heme maturation: effect of NO and thioredoxin-1**. *Free Radic Biol Med* (2015.0) **82** 105-113. DOI: 10.1016/j.freeradbiomed.2015.01.030
35. Millet P, Vachharajani V, McPhail L, Yoza B, McCall CE. **GAPDH binding to TNF-α mRNA contributes to posttranscriptional repression in monocytes: a novel mechanism of communication between inflammation and metabolism**. *J Immunol* (2016.0) **196** 2541-2551. DOI: 10.4049/jimmunol.1501345
36. Yang J, Zhang S, Zhang J, Dong J, Wu J, Zhang L. **Identification of key genes and pathways using bioinformatics analysis in septic shock children**. *Infect Drug Resist* (2018.0) **11** 1163-1174. DOI: 10.2147/IDR.S157269
37. Takaoka Y, Goto S, Nakano T, Tseng HP, Yang SM, Kawamoto S. **Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) prevents lipopolysaccharide (LPS)-induced, sepsis-related severe acute lung injury in mice**. *Sci Rep* (2014.0) **4** 5204. DOI: 10.1038/srep05204
38. Page-McCaw A, Ewald AJ, Werb Z. **Matrix metalloproteinases and the regulation of tissue remodelling**. *Nat Rev Mol Cell Biol* (2007.0) **8** 221-233. DOI: 10.1038/nrm2125
39. Hu J, Van den Steen PE, Sang QX, Opdenakker G. **Matrix metalloproteinase inhibitors as therapy for inflammatory and vascular diseases**. *Nat Rev Drug Discov* (2007.0) **6** 480-498. DOI: 10.1038/nrd2308
40. DeLeon-Pennell KY, Meschiari CA, Jung M, Lindsey ML. **Matrix metalloproteinases in myocardial infarction and heart failure**. *Prog Mol Biol Transl Sci* (2017.0) **147** 75-100. DOI: 10.1016/bs.pmbts.2017.02.001
41. 41.Bassiouni W, Ali MAM, Schulz R. Multifunctional intracellular matrix metalloproteinases: implications in disease. FEBS J. 2021.
42. Madkour MM, Anbar HS, El-Gamal MI. **Current status and future prospects of p38α/MAPK14 kinase and its inhibitors**. *Eur J Med Chem* (2021.0) **213** 113216. DOI: 10.1016/j.ejmech.2021.113216
43. Burton JC, Antoniades W, Okalova J, Roos MM, Grimsey NJ. **Atypical p38 signaling, activation, and implications for disease**. *Int J Mol Sci* (2021.0) **22** 4183. DOI: 10.3390/ijms22084183
44. Barnes PJ. **Kinases as novel therapeutic targets in asthma and chronic obstructive pulmonary disease**. *Pharmacol Rev* (2016.0) **68** 788-815. DOI: 10.1124/pr.116.012518
45. Gaffey K, Reynolds S, Plumb J, Kaur M, Singh D. **Increased phosphorylated p38 mitogen-activated protein kinase in COPD lungs**. *Eur Respir J* (2013.0) **42** 28-41. DOI: 10.1183/09031936.00170711
46. Zheng T, Zhang B, Chen C, Ma J, Meng D, Huang J. **Protein kinase p38α signaling in dendritic cells regulates colon inflammation and tumorigenesis**. *Proc Natl Acad Sci USA* (2018.0) **115** E12313-E12322. DOI: 10.1073/pnas.1814705115
47. Jaberi SA, Cohen A, D'Souza C, Abdulrazzaq YM, Ojha S, Bastaki S. **Lipocalin-2: structure, function, distribution and role in metabolic disorders**. *Biomed Pharmacother* (2021.0) **142** 112002. DOI: 10.1016/j.biopha.2021.112002
48. Yan QW, Yang Q, Mody N, Graham TE, Hsu CH, Xu Z. **The adipokine lipocalin 2 is regulated by obesity and promotes insulin resistance**. *Diabetes* (2007.0) **56** 2533-2540. DOI: 10.2337/db07-0007
49. Kang SS, Ren Y, Liu CC, Kurti A, Baker KE, Bu G. **Lipocalin-2 protects the brain during inflammatory conditions**. *Mol Psychiatry* (2018.0) **23** 344-350. DOI: 10.1038/mp.2016.243
50. Cowland JB, Sørensen OE, Sehested M, Borregaard N. **Neutrophil gelatinase-associated lipocalin is up-regulated in human epithelial cells by IL-1 beta, but not by TNF-alpha**. *J Immunol* (2003.0) **171** 6630-6639. DOI: 10.4049/jimmunol.171.12.6630
51. Skytthe MK, Graversen JH, Moestrup SK. **Targeting of CD163(+) macrophages in inflammatory and malignant diseases**. *Int J Mol Sci* (2020.0) **21** 5497. DOI: 10.3390/ijms21155497
52. Liu H, Lin D, Xiang H, Chen W, Zhao S, Peng H. **The role of tumor necrosis factor-like weak inducer of apoptosis in atherosclerosis via its two different receptors**. *Exp Ther Med* (2017.0) **14** 891-897. DOI: 10.3892/etm.2017.4600
53. Fabriek BO, van Bruggen R, Deng DM, Ligtenberg AJ, Nazmi K, Schornagel K. **The macrophage scavenger receptor CD163 functions as an innate immune sensor for bacteria**. *Blood* (2009.0) **113** 887-892. DOI: 10.1182/blood-2008-07-167064
54. Burkard C, Lillico SG, Reid E, Jackson B, Mileham AJ, Ait-Ali T. **Precision engineering for PRRSV resistance in pigs: Macrophages from genome edited pigs lacking CD163 SRCR5 domain are fully resistant to both PRRSV genotypes while maintaining biological function**. *PLoS Pathog* (2017.0) **13** e1006206. DOI: 10.1371/journal.ppat.1006206
55. Qiao G, Wu A, Chen X, Tian Y, Lin X. **Enolase 1, a moonlighting protein, as a potential target for cancer treatment**. *Int J Biol Sci* (2021.0) **17** 3981-3992. DOI: 10.7150/ijbs.63556
56. Forster F, Paster W, Supper V, Schatzlmaier P, Sunzenauer S, Ostler N. **Guanylate binding protein 1-mediated interaction of T cell antigen receptor signaling with the cytoskeleton**. *J Immunol* (2014.0) **192** 771-781. DOI: 10.4049/jimmunol.1300377
57. Gemta LF, Siska PJ, Nelson ME, Gao X, Liu X, Locasale JW. **Impaired enolase 1 glycolytic activity restrains effector functions of tumor-infiltrating CD8(+) T cells**. *Sci Immunol* (2019.0) **4** eaap9520. DOI: 10.1126/sciimmunol.aap9520
58. Zakrzewicz D, Didiasova M, Krüger M, Giaimo BD, Borggrefe T, Mieth M. **Protein arginine methyltransferase 5 mediates enolase-1 cell surface trafficking in human lung adenocarcinoma cells**. *Biochim Biophys Acta Mol Basis Dis* (2018.0) **1864** 1816-1827. DOI: 10.1016/j.bbadis.2018.02.021
59. Cho H, Um J, Lee JH, Kim WH, Kang WS, Kim SH. **ENOblock, a unique small molecule inhibitor of the non-glycolytic functions of enolase, alleviates the symptoms of type 2 diabetes**. *Sci Rep* (2017.0) **7** 44186. DOI: 10.1038/srep44186
60. Wygrecka M, Marsh LM, Morty RE, Henneke I, Guenther A, Lohmeyer J. **Enolase-1 promotes plasminogen-mediated recruitment of monocytes to the acutely inflamed lung**. *Blood* (2009.0) **113** 5588-5598. DOI: 10.1182/blood-2008-08-170837
61. 61.Guo Q, Yang C, Gao F. The state of CD44 activation in cancer progression and therapeutic targeting. FEBS J. 2021.
62. Tavianatou AG, Caon I, Franchi M, Piperigkou Z, Galesso D, Karamanos NK. **Hyaluronan: molecular size-dependent signaling and biological functions in inflammation and cancer**. *FEBS J* (2019.0) **286** 2883-2908. DOI: 10.1111/febs.14777
63. Monticelli LA, Buck MD, Flamar AL, Saenz SA, Tait Wojno ED, Yudanin NA. **Arginase 1 is an innate lymphoid-cell-intrinsic metabolic checkpoint controlling type 2 inflammation**. *Nat Immunol* (2016.0) **17** 656-665. DOI: 10.1038/ni.3421
64. Dzik JM. **Evolutionary roots of arginase expression and regulation**. *Front Immunol* (2014.0) **5** 544. DOI: 10.3389/fimmu.2014.00544
65. Rotondo R, Bertolotto M, Barisione G, Astigiano S, Mandruzzato S, Ottonello L. **Exocytosis of azurophil and arginase 1-containing granules by activated polymorphonuclear neutrophils is required to inhibit T lymphocyte proliferation**. *J Leukoc Biol* (2011.0) **89** 721-727. DOI: 10.1189/jlb.1109737
66. Salvador JM, Brown-Clay JD, Fornace AJ. **Gadd45 in stress signaling, cell cycle control, and apoptosis**. *Adv Exp Med Biol* (2013.0) **793** 1-19. DOI: 10.1007/978-1-4614-8289-5_1
|
---
title: 'Trends in mortality and comorbidities in hemodialysis patients between 2012
and 2017 in an East-European Country: a retrospective study'
authors:
- Lazar Chisavu
- Adelina Mihaescu
- Flaviu Bob
- Alexandru Motofelea
- Oana Schiller
- Luciana Marc
- Razvan Dragota-Pascota
- Flavia Chisavu
- Adalbert Schiller
journal: International Urology and Nephrology
year: 2023
pmcid: PMC10012315
doi: 10.1007/s11255-023-03549-6
license: CC BY 4.0
---
# Trends in mortality and comorbidities in hemodialysis patients between 2012 and 2017 in an East-European Country: a retrospective study
## Abstract
### Purpose
The aim of this study was to evidence trends and changes in mortality, comorbid conditions, prognosis, and causes of death after 5 years of continuous evolution of hemodialysis (HD) patients in Romania.
### Methods
We included two cohorts of stable HD patients (901 from 2012 and 1396 from 2017). Both cohorts were followed up for 1 year. The 5-year survivors of the 2012 cohort were identified in 2017 and their data changes were assessed.
### Results
The 2017 patients were older, with longer time on dialysis, higher serum creatinine and urea levels, and required higher ultrafiltration volume per dialysis. They also had lower hemoglobin, lower C-reactive protein, higher albumin, higher calcium bicarbonate, and higher parathyroidectomy prevalence. The 2017 cohort presented with lower average dialysis flow, less administration of iron sucrose, had more catheters, lower hepatitis C prevalence, higher diabetes mellitus prevalence, higher heart valve calcifications, higher heart rate disorders, higher prevalence of left ventricular hypertrophy, and lower ejection fraction. Cardiovascular disease was the main cause of death in both years ($50\%$ in 2012 and $45.6\%$ in 2017), followed by sepsis and cancer. The mortality was higher in 2017 compared to 2012 (14.1 vs $6.6\%$). The 5-year mortality was $37.2\%$ with an average of $7.44\%$/year. The risk of death increased with age, higher C-reactive protein, higher phosphate, lower hemoglobin, and lower albumin.
### Conclusion
Cardiovascular disease remains the main causes of death in HD-treated patients but with decreasing trend. Developing regional therapeutic strategies for quality care with early intervention will most likely improve mortality.
## Introduction
The advances of chronic kidney disease (CKD), end-stage kidney disease (ESKD), and their comorbidities/complications treatment over the last decades increased the survival of patients on hemodialysis therapy. The mortality, however, remained high, being 10 to 30-fold, higher than in the general population [1]. The major renal registries, European Renal Association-European Dialysis and Transplant Association (ERA-EDTA) and United States Renal Data System (USRDS), reported the annual mortality, in 2019, as $14.6\%$ in Europe [2] and $15.6\%$ in the United States, respectively [3].
Cardiovascular disease (CVD) remains the first cause of mortality in hemodialysis (HD)-treated patients. In 2019, $55\%$ of deaths were attributable to CVD according to USRDS Annual Report (i.e., arrhythmia/cardiac arrest, acute myocardial infarction, congestive heart failure, and stroke) [3]. The prevalence of cardiovascular disease is very high in ESKD patients with up to two-third of them presenting some form of CVD [4]. Although CVD remains the leading cause of death, in the recent years, several studies showed a positive trend for non-CVD deaths [5, 6] and a negative trend for CVD mortality in HD patients.
The aim of our multicenter study was to compare two cohorts of HD-treated patients (2012 and 2017) and to evidence trends and changes in mortality, comorbid conditions, prognosis, and causes of death after 5 years of continuous evolution of CKD management and treatment in Romania.
## Materials and methods
We included two cohorts of stable HD patients (901 from 2012 and 1396 from 2017) treated in the same 8 hemodialysis centers in Romania. Both cohorts had been followed up for 1 year. In all cases, HD therapy was performed using high flux, high surface dialyzers, three times/week (≥ 12 h/week). The assessment of therapy and follow-up of anemia and chronic kidney disease-mineral and bone disorder (CKD-MBD) were performed according to KDIGO guidelines [7, 8]. Personal data, history of disease, and HD therapy parameters were retrieved from patients’ dialysis files. Besides the clinical evaluation, the patients also underwent a cardiology assessment. A two-dimensional and M-mode continuous and Pulse Doppler echocardiography were performed in accordance with the recommendations of the European Association of Cardiovascular Imaging (EACI), between the second and third hour of the dialysis session, by the same operator for each cohort, using similar devices (avoid inter-observer differences). The operator assessed the left ventricular ejection fraction (LVEF) using the Simpson method, left ventricular hypertrophy and noted the presence and number of heart valve calcifications.
The 5-year survivors from 2012 cohort have been identified in 2017 and their data changes were assessed.
The Ethics Committee of each Dialysis Center approved the study and all patients signed a written informed consent to permit the investigators to use their personal data for scientific purposes. The study was performed in accordance with the Ethics Code of the World Medical Association.
## Statistical analysis
Data are presented as mean ± standard deviation (numerical variables with Gaussian distribution), median and interquartile range (numerical variables with non-Gaussian distributions), respectively, percentage from the sub-group total and number of individuals. Continuous variables distribution was tested for normality using Shapiro–Wilk test and for equality of variances using Levene’s test. Adjusted risk estimates for all-cause mortality were calculated using univariable and multivariable Cox regression. In this study, a p value of 0.05 was considered the threshold for statistical significance. Data were analyzed using SPSS v26 statistical software package (SPSS Inc, Chicago, IL, USA).
## Results
Two cohorts of ESKD patients, treated with HD in the same HD centers (901 patients from 2012 and 1396 from 2017) were analyzed. The baseline characteristics (Table 1) showed that the patients from 2017 were older (59.8 ± 13.1 vs 57.9 ± 13.3 years, $$p \leq 0.001$$) and had a longer time on dialysis therapy (5.8 ± 4.7 vs 4.7 ± 4.1 years, $p \leq 0.001$). The 2017 cohort had higher average serum creatinine (8.5 ± 2.4 vs 8.3 ± 2.3 mg/dl, $$p \leq 0.004$$) and urea (127 ± 34.6 vs 121.4 ± 38.9 mg/dl, $p \leq 0.001$) levels before dialysis and needed higher average ultrafiltration volume/dialysis (2467.8 ± 956.2 vs 2062.5 ± 1107.2 ml, $p \leq 0.001$). The average eKt/V, 1.4, as measured by ADIMEA, did not differ. Table 1Baseline data of the investigated cohorts2012 ($$n = 901$$)2017 ($$n = 1396$$)pAge (A + SD)57.9 (13.2)59.8 (13.1))0.0011Gender (female)381 ($42.3\%$)575 ($41.2\%$)0.633Dialysis treatment (years) (A + SD)4.7 (4.1)5.8 (4.7) < 0.001Deaths (%)60 ($6.6\%$)197 ($14.1\%$) < 0.001Creatinine (mg/dl) (A + SD)8.3 (2.3)8.5 (2.4)0.0041Urea (mg/dl) (A + SD)121.4 (38.9)127.0 (34.6)0.032eKt/V (A + SD)1.4 (0.3)1.4 (0.4)0.909Dry weight (kg) (A + SD)73.8 (29.0)74.4 (17.4)0.5731Ultrafiltration volume (ml) (A + SD)2062.5 (1107.2)2467.8 (956.2) < 0.0011Dialyzer surface (A + SD)2.1 (0.2)2.1 (0.2)0.6751Qb (ml/min) (A + SD)318.3 (47.0)321.8 (44.3)0.0661Qd (ml/min) (A + SD)673.5 (86.5)580.9 (63.3) < 0.0011Vascular access (catheters) (%)87 ($9.7\%$)255 (18.3) < 0.0012Hemoglobin (g/dl) (A + SD)11.3 (1.3)11.1 (1.4) < 0.0121Serum ferritin (ng/ml) (A + SD)743.4 (483.1)1025.2 (685.9) < 0.0011Transferrin saturation (%)(A + SD)35 (16.5)33.0 (14.5)0.0021Serum C-reactive protein (mg/dl) (A + SD)7.9 (18.3)1.2 (2.0) < 0.0011Serum albumin (g/dl) (A + SD)4.0 (0.4)4.1 (0.4) < 0.0011Iron sucrose (100 mg/vial) (vials/month) (A + SD)2.0 (1.7)1.2 (1.2) < 0.0011Darbepoetin alfa (micrograms/month) (A + SD)67.2 (56.8)68.1 (60.5)0.7231Serum calcium (mg/dl) (A + SD)8.5 (0.7)8.8 (0.7) < 0.0011Serum phosphate (mg/dl) (A + SD)4.7 (1.4)4.8 (1.5)0.0701Ca × P (mg2/dl2) (A + SD)40 (12.1)42.9 (13.4) < 0.0011Serum bicarbonate (mmol/l) (A + SD)21.1 [4]21.4 (2.8)0.0181iPTH (pg/ml) (A + SD)551.5 (627.9)528.1 (526.9)0.3361Hepatitis B (%)46 ($5.1\%$)75 ($5.4\%$)0.8542Hepatitis C (%)199 ($22.1\%$)183 ($13.1\%$) < 0.0012Parathyroidectomy (%)50 ($5.5\%$)153 ($11\%$) < 0.0012Peripheral vascular disease (%)260 ($28.9\%$)365 ($26.1\%$)0.1682Cerebrovascular disease (%)183 ($20.3\%$)256 ($18.3\%$)0.2632Diabetes mellitus (%)188 ($20.9\%$)358 ($25.6\%$)0.0102Heart rate disorders (EKG) (%)151 ($16.8\%$)286 ($20.5\%$)0.0302Myocardial infarction history (%)102 ($11.3\%$)143 ($10.2\%$)0.4552Ejection fraction (%) (A + SD)58.2 (9.6)56.8 (8.8) < 0.0012Heart valve calcifications (%) -1 valve501 ($55.6\%$)1056 ($75.6\%$) < 0.0012Heart valve calcifications (%) -2 Valves61 ($6.7\%$)48 ($3.4\%$)0.0003Left ventricular hypertrophy (%)625 ($69.4\%$)1188 ($85.1\%$) < 0.0012A average, SD standard deviation, kg kilograms, ml milliliters, min minute, Qb blood flow, *Qd dialysis* solution flow, g gram, mmol millimoles, l liters, dl deciliters, ng nanograms, mg milligrams, EKG electrocardiogram;1t test2Pearson’s Chi-squared testBolded numbers emphasize statistical significant results In the 2017 cohort, the patients had lower average serum hemoglobin ($$p \leq 0.012$$), higher ferritin ($p \leq 0.001$), lower transferrin saturation ($$p \leq 0.002$$), lower C-reactive protein ($p \leq 0.001$), and higher albumin levels ($p \leq 0.001$). Regarding mineral bone disorders, the 2017 cohort presented higher average calcium levels ($p \leq 0.001$) and higher serum bicarbonate ($$p \leq 0.018$$) but phosphate and intact parathyroid hormone (iPTH) did not significantly differ. The prevalence of parathyroidectomy, as expected, was higher ($p \leq 0.001$).
Statistical differences were identified between HD specifications, medication, and vascular access. The 2017 cohort had lower average dialysis solution flow (Qd) ($p \leq 0.001$), patients were treated with lower iron sucrose doses ($p \leq 0.001$), and HD was performed more on catheters (18.3 vs. $9.7\%$, $p \leq 0.001$). The differences between 2017 and 2012 cohorts concerning comorbidities were: lower hepatitis C virus infection prevalence ($p \leq 0.001$), higher prevalence of diabetes mellitus (DM) as primary cause of ESKD ($$p \leq 0.01$$), higher prevalence of heart valve calcifications ($p \leq 0.001$), left ventricular hypertrophy ($p \leq 0.001$), heart rate disorders (electrocardiography proven) ($$p \leq 0.03$$) as well as lower average LVEF ($p \leq 0.001$) (Table 1).
Each cohort was followed up for 1 year. The 1-year mortality rate among stable dialysis patients was higher in the 2017 cohort (14.1 vs $6.6\%$, $p \leq 0.001$). In the 2012 cohort, the deceased patients were older (63.1 ± 12.3 vs. 57.9 ± 13.2 years $$p \leq 0.0031$$), had lower average Hb levels (10.3 ± 1.9 vs. 11.3 ± 1.3 g/dl, $p \leq 0.0001$), and higher average CRP levels (28.2 ± 31 vs. 7.9 ± 18.3 mg/dl, $p \leq 0.0001$) as compared with the average baseline values.. The deceased patients from the 2017 cohort were older (67.5 ± 10.6 vs. 59.8 ± 13.1 years, $p \leq 0.0001$), were treated with HD for a shorter period of time (4.6 ± 4.1 vs. 5.8 ± 4.7 years, $$p \leq 0.0007$$), had a lower average Hb level (9.9 ± 1.9 vs. 11.1 ± 1.4 g/dl, $p \leq 0.0001$), higher CRP levels (4.2 ± 5.7 vs. 1.2 ± 2.0 mg/dl, $p \leq 0.0001$), lower serum albumin levels (3.6 ± 0.6 vs. 4.1 ± 0.4 g/dl, $p \leq 0.0001$), and lower calcium levels (8.6 ± 0.8 vs 8.8 ± 0.7 mg/dl, $$p \leq 0.0002$$) as compared with the baseline data.
Concerning the causes of death, there were no statistical differences among the two cohorts. Cardiovascular disease was the leading cause of death (acute myocardial infarction, heart failure, arrhythmias, stroke, and sudden death), accounting for $50\%$ of the patients from the 2012 cohort and $45.6\%$ patients from the 2017 cohort. Sepsis remained the second most important cause of death ($23.3\%$ in 2012 and $20.8\%$ in 2017). In the 2017 cohort, there was a downward trend in mortality from CVD with an increase in mortality caused by cancer without statistically significant differences (8.3–$13.2\%$) (Table 2).Table 2Causes of death in the 1-year follow-up in 2012 and 2017Causes of death2012 $$n = 60$$ (%)2017 $$n = 197$$ (%)pAcute myocardial infarction5 (8.3)20 (10.2)0.6771Hearth failure8 (13.3)17 (8.6)0.2821Arrythmias2 (3.3)10 (5.1)0.5751Sudden death6 (10.0)9 (4.6)0.1161Stroke9 (15.0)34 (17.3)0.6811Sepsis14 (23.3)41 (20.8)0.6771Cancer5 (8.3)26 (13.2)0.3111Hyperkalemia2 (3.3)5 (2.5)0.7401Withdraw from dialysis0 (0.0)3 (1.5)0.3361Others/unknown9 (15.0)32 (16.2)0.81811t test We analyzed several parameters that could modify mortality risk. Mortality risk in both cohorts was increased with: dialysis duration (OR = 1.179 CI 1.086–1.28) ($p \leq 0.01$), age (in both univariable and multivariate models) (OR = 1.05 CI 1.04–1.07, $p \leq 0.001$), the odds of death were 1.05 times higher for each 1-year increase in age, C-reactive protein (in multivariate but not univariate model) (OR = 1.02 CI 1.01–1.03, $p \leq 0.001$), the odds of death were 1.02 times higher for each one-unit increase in C-reactive protein level. For phosphate level (in multivariate but not univariate model), the odds of death are 1.19 times higher for each one-unit increase (CI 1.05–1.35, $$p \leq 0.007$$). For hemoglobin levels (in both univariable and multivariate model), the odds of death are 0.60 times lower (CI: 0.54–0.65, $p \leq 0.001$) for each one-unit increase. For albumin levels (in both univariable and multivariate models), the odds of death are 0.11 times lower for each one-unit increase (CI 0.08–0.15, $p \leq 0.001$) (Table 3).Table 3Univariate and multivariate analysis of factors associated with mortality in hemodialysis patientsParameterOR (univariate)OR (multivariate)Duration of dialysis1.179 (1.086–1.28), $p \leq 0.011.209$ (1.09–1.31), $p \leq 0.001$Age1.05 (1.04–1.07), $p \leq 0.0011.06$ (1.04–1.07), $p \leq 0.001$CRP1.02 (1.01–1.03), $p \leq 0.0010.96$ (0.93–0.99), $$p \leq 0.012$$Albumin0.11 (0.08–0.15), $p \leq 0.0010.18$ (0.12–0.27), $p \leq 0.001$Po41.02 (0.92–1.12), $$p \leq 0.7341.19$$ (1.05–1.35), $$p \leq 0.007$$Hemoglobin0.60 (0.54–0.65), $p \leq 0.0010.65$ (0.58–0.73), $p \leq 0.001$iPTH1.00 (1.00–1.00), $$p \leq 0.5031.00$$ (1.00–1.00), $$p \leq 0.090$$Ferritin1.00 (1.00–1.00), $$p \leq 0.0851.00$$ (1.00–1.00), $$p \leq 0.079$$CRP C-reactive protein, PO4 seric phosphate, iPTH intact parathormone, OR odds ratio In 2017, we identified 566 survivors out of the 901 stable HD patients from 2012 cohort. The 5-year survival rate was $62.8\%$ with the average mortality rate of $7.44\%$/year. All the patients that changed HD center or underwent renal transplant in the 5-year period were excluded. The patients’ characteristics were analyzed to determine the changes after 5 years of HD therapy. t test and Chi-square tests showed that there are several significant statistical differences concerning patient’s medical data in the 5-year survival time, 56 ($9.8\%$) patients died in 2017. As expected, the prevalence of catheter increased in HD (from 6.71 to $13.95\%$, $p \leq 0.001$). In 2017, the patients presented (before the dialysis session) higher average serum creatinine ($p \leq 0.004$), similar average dry weight and serum bicarbonate levels, increased average albumin ($p \leq 0.001$), and lower C-reactive protein ($p \leq 0.001$) levels, suggesting a good nutrition status and decreasing inflammation. Similar hemoglobin, transferrin saturation levels and similar average epoetin doses but with higher average ferritin levels ($p \leq 0.0001$) could be related to higher on average monthly doses of intravenous iron therapy ($p \leq 0.001$). The higher on average serum calcium ($p \leq 0.001$), higher calcium-phosphate product ($$p \leq 0.004$$), and higher PTH levels ($$p \leq 0.022$$) with similar phosphate levels was a result of excess calcium-based phosphate binders and/or vitamin D (and/or vitamin D analogs). One should also mention the higher number of patients treated with parathyroidectomy ($p \leq 0.001$) (Table 4). In time, eKt/V remained similar although the filter surfaces and Qb increased. Table 4Characteristics of 5-year HD therapy survivors from 2012 to 20172012 ($$n = 566$$)2017 ($$n = 566$$)pAge (A + SD)55.6 (12.8)59.8 (12.8) < 0.0013Creatinine (mg/dl) (A + SD)8.6 (2.3)9.0 (2.3) < 0.0043Urea (mg/dl) (A + SD)125.1 (35.8)130.0 (34.6) = 0.0243eKt/V (A + SD)1.4 (0.3)1.4 (0.2) = 0.5283Dry weight (kg) (A + SD)73.6 (16.8)72.6 (17.0) = 0.3563Ultrafiltration volume (ml) (A + SD)2102.5 (1118.0)2497.7 (930.5) < 0.0013Dialyzer surface (A + SD)2.1 (0.2)2.2 (0.2) < 0.0013Qb (ml/min) (A + SD)325.0 (46.8)334.0 (42.7) = 0.0013Qd (ml/min) (A + SD)680.6 (86.7)600.0 (68.8) < 0.0013Vascular access (catheters) (%)38 ($6.71\%$)79 ($13.95\%$) < 0.0012Hemoglobin (g/dl) (A + SD)11.3 (1.2)11.2 (1.4) = 0.1843Serum ferritin (ng/ml) (A + SD)756.2 (502.7)1202.6 (819.6) < 0.0013Transferrin saturation (%) (A + SD)35.7 (16.9)35.0 (15.4) = 0.4713Serum C-reactive protein (mg/dl) (A + SD)6.2 (15.3)1.1 (1.9) < 0.0013Serum albumin (g/dl) (A + SD)4.1 (0.4)4.2 (0.4) < 0.0013Iron sucrose (100 mg/vial) (vials/month) (A + SD)1.9 (1.6)1.1 (0.9) < 0.0013Darbepoetin alfa (micrograms/month) (A + SD)64.3 (56.3)64.9 (59.8) = 0.8693Serum calcium (mg/dl) (A + SD)8.5 (0.7)8.9 (0.8) < 0.0013Serum phosphate (mg/dl) (A + SD)4.8 (1.4)4.8 (1.5) = 0.7573Ca x Pi (mg[2]/dl[2]) (A + SD)40.8 (12.4)43.1 (14.5) = 0.0043Serum bicarbonate mmol/l (A + SD)21.0 (3.9)21.4 (2.8) = 0.0983iPTH (pg/ml) (A + SD)555.1 (592.5)640.5 (631.5) = 0.0223Parathyroidectomy (%)34 ($6\%$)106 ($18.72\%$) < 0.0012Hepatitis B (%)25 ($4.41\%$)28 ($4.94\%$) = 0.672Hepatitis C (%)117 ($20.67\%$)124 ($21.9\%$) = 0.612Coronary artery disease (%)354 ($60.9\%$)394 ($69.6\%$) = 0.0042Peripheral vascular disease (%)123 ($21.7\%$)157 ($27.7\%$) = 0.0232Cerebrovascular disease (%)87 ($15.3\%$)111 ($19.6\%$) = 0.0752Diabetes mellitus (%)95 ($16.7\%$)114 ($20.1\%$) = 0.1732Heart rate disorders (EKG) (%)77 ($13.6\%$)130 ($22.9\%$) < 0.0012History of myocardial infarction (%)48 ($8.4\%$)70 ($12.3\%$) = 0.0372Ejection fraction % (A + SD)59.2 (9.4)57.5 (8.8) = 0.0033Heart valve calcifications (%) -1 valve290 ($51.2\%$)434 ($76.6\%$) < 0.0012Heart valve calcifications (%) -2 valves37 ($6.5\%$)39 ($6.8\%$)Left ventricular hypertrophy (%)371 ($65.5\%$)508 ($89.7\%$) < 0.0012A average, SD standard deviation, kg kilograms, ml milliliters, min minute, Qb blood flow, *Qd dialysis* solution flow, g gram, mmol millimoles, l liters, dl deciliters, ng nanograms, mg milligrams, EKG electrocardiogram1t test2Pearson's Chi-squared testBolded numbers emphasize statistical significant results Also, as expected, in 2017, the patients presented more comorbid conditions compared to 2012: more coronary artery disease ($$p \leq 0.004$$), more peripheral vascular disease ($$p \leq 0.023$$), more heart rate disorders, ($p \leq 0.001$), more history of myocardial infarction ($$p \leq 0.037$$), more LVH ($p \leq 0.001$), more valvular calcifications ($p \leq 0.001$), and lower average ejection fraction ($$p \leq 0.003$$). ( Table 4).
## Discussion
ESKD patients treated with HD are at high risk for CVD with increased mortality under dynamic changes induced by the cumulative risk factors of CKD, increased pre-dialysis and dialysis survival as a result of better medical care (i.e., lowering CKD progression, preventing and treating CVD, and advanced CKD complications) and personalized HD therapy [9, 10]. Under these conditions, assessing the changes in the HD population, the trends in evolution, mortality and causes of death represent an important tool for future medical strategies [11, 12]. Therefore, in our multicenter study, we compared the baseline data of HD patients (at entry in the study) in two cohorts, 2012 and 2017, performing HD therapy in the same eight centers from Romania. In 2012, according to the Romanian Renal Registry (RRR), 9551 patients were performing HD therapy for ESKD from which $9.4\%$ [901] were represented by our cohort [13]. The annual increase of HD patients was $7.8\%$/year reported by the same registry. In 2017, the cohort from the same 8 centers was represented by 1396 patients. RRR reported that 13,362 patients were on HD in 2017, with $10.4\%$ [1396] representing our second cohort. The average annual increase of patients in the eight centers was $10.9\%$ (being higher than the national average). One should evidence the fact that the prevalence of HD patients decreased in Romania in 2020 with $1.4\%$ most probably related to the *Corona virus* pandemic [14].
The baseline data of the two cohorts (2012 and 2017) had many differences. In 2017, the patients were significantly older (59.8 vs 57.9 years, $$p \leq 0.001$$) and were, on average, on dialysis treatment for a longer time (5.8 vs 4.7 years, $p \leq 0.001$). In our cohorts, the mean age of HD patients increased, in 2017 being close to the mean age reported by ERA-EDTA (60.7 years). One should mention the fact that 566 ($40.5\%$) patients of the 2017 cohort were 5-year survivals from 2012. Therefore, age differences, HD duration time, and mortality are influenced by these high number of 5-year survivals. There were, however, important differences between European countries concerning mean age of prevalent HD patients (Albania 49.5 years, Ukraine 50, Scotland 56.8, Spain 59.5, Denmark 59, Portugal 67.9) [15]. In Europe (and in our cohorts also), there is a trend for increasing age in prevalent HD patients (mean age being 61.8 years in the 2019 ERA-EDTA registry). Age and duration of HD therapy have been associated to higher risk of mortality in these patients. Though a decrease in mortality was registered among USRDS reporting patients between 2010 and 2019, the 2020 mortality increased in all age groups. The largest absolute increase was evidenced in the older patient’s group [17, 18].
Average pre-dialysis serum creatinine and urea were significantly higher in the 2017 cohort ($p \leq 0.004$ and $p \leq 0.032$, respectively) but urea to creatinine ratio (UCR) did not differ. C-reactive protein (CRP) was lower ($p \leq 0.001$) and serum albumin was higher ($p \leq 0.001$) in the same cohort suggesting a lower inflammation and improved nutrition status but average dry weight did not significantly differ. Higher pre-dialysis serum creatinine and UCR were associated with increased risk of death in HD patients [19, 20] more or less related to malnutrition inflammation complex syndrome. In our 2017 cohort, this relation could not be evidenced since average CRP values and average dry body weight were not modified as expected. Most probably our results could be related to a better quality of the diet.
Lower average hemoglobin and transferrin saturation values and higher ferritin levels in context of near to normal average CRP (12 mg/l) were the characteristics of the 2017 cohort (as compared to 2012). It seems that the average higher ferritin levels are the result of higher intravenous iron therapy similar to that proposed by the PIVOTAL study [21]. Higher ferritin levels seem to be associated with increased mortality risk in HD patients but when corrected to malnutrition and inflammation, the risk seems to be attenuated [22]. Nevertheless, the average intravenous iron dose was adapted to ferritin levels and was reduced significantly in the 2017 cohort (from 200 mg/month in 2012 to 120 mg/month in 2017, $p \leq 0.001$).
In the 2017 cohort, the average serum calcium and CaxP product were significantly higher (8.8 ± 0.7 vs 8.5 ± 0.7 mg/dl $p \leq 0.001$ and 42.9 ± 13.4 vs 40 ± 12.1 mg2/dl2, $p \leq 0.001$, respectively) but average phosphate, iPTH, and vitamin D levels did not differ. These results could be related to higher need for phosphate binders and excessive use of calcium-based ones. The influence of calcium-based vs. non-calcium-based phosphate binders on the risk of death in HD patients is a still debated issue. Papers supporting decreased risk of all-cause mortality in HD patients using non-calcium-based phosphate binders vs. calcium-based ones [23, 24] to no effect or differences [25] between treatments have been published in the recent years. In our opinion, the conclusions of a 2018 Cochrane meta-analysis deserve to be mentioned [26]. Though Sevelamer may lower the risk of death compared to calcium-based binders, the authors find no benefits for any phosphate binder concerning cardiovascular death, myocardial infarction, coronary artery calcifications or stroke. In our cohorts, the differences mentioned above between the two cohorts did not influence the risk of death.
The dialysis adequacy in the 2017 cohort (as compared to 2012) was similar. The average targeted eKt/V of 1.4 was reached in both cohorts. Both cohorts were treated with high flux dialyzers, ≥ 12 h/week. Higher ultrafiltration was needed in 2017 (on average 8.29 vs. 6.98 ml/kg/h) and the average dialysis fluid flux (Qd) was lower. Prolonged fluid overload as well as higher ultrafiltration volumes have been related to increased mortality in HD patients [27–29], though no clear cutoff levels for optimal ultrafiltration are documented until now. As we will discuss further, higher ultrafiltration rates in the 2017 cohort did not increase mortality risk.
The baseline data from 2017 reveal a higher prevalence of some comorbid conditions and complications. The prevalence of DM as primary condition for ESKD increased (25.6 vs. $20.9\%$, $$p \leq 0.010$$). It is an expected result since prevalence of DM is increasing worldwide and ESRD related to DM, also [30, 31], though the incidence rates for DM-related ESKD are significantly lower in Europe as compared to US. The mortality of DM-related ESKD patients seems to be also significantly higher as compared to the no DM ESKD patients and this may influence the survival estimates in HD patients in the future [32–34]. The prevalence of hepatitis B virus infection did not change in the two cohorts being around $5\%$, but the hepatitis C virus infection decreased as a result of a nationwide prevention program applied since 2000 and from 2015 a nationwide treatment program with the novel direct oral antivirals which included HD patients also. In our two cohorts hepatitis virus infection did not increase mortality risk.
The prevalence of cerebrovascular disease, peripheral vascular disease, and history of myocardial infarction was not higher among the 2017 cohort HD patients. Nevertheless, patients presented higher prevalence of left ventricular hypertrophy, lower average ejection fraction, and higher prevalence of heart valve calcification (one valve/two valves) (see Table 1). The causes seem to be complex: higher age of the HD-treated patients, on average longer period on HD therapy, increased prevalence of DM 2 related ESKD. One should also evidence the fact that 566 HD patients from the 2012 cohort ($62.8\%$) survived and could be identified in the 2017 cohort (representing $40.5\%$ of the 2017 patients) (Table 3.). The 5-year unadjusted survival rate was very high if compared with the data from USA, Europe, and Japan Registries for patients initiating HD between 2004 and 2008 ($39\%$, $41\%$, and $60\%$, respectively) [35, 36]. The possible explanations should be that the 5-year survivors from 2012 were significantly younger at the inclusion in our study (around 55 years), they performed HD using mainly Cimino type shunts (only $6.7\%$ on catheters), had higher Hb levels, lower prevalence of DM, lower prevalence of cardiovascular complications, and higher average LVEF (Table 3). Similar results were presented by the ANZDATA for non-indigenous patients initiating HD between 2009 and 2018 for the age groups between 45 and 64 [37]. During the 5-year survival, as expected, all those characteristics became significantly worse, influencing the baseline data of the 2017 cohort and the mortality also (the 1-year mortality being $6.6\%$ in 2012 vs. $14.1\%$ in 2017). On the other hand, in the last 2 years, increasing mortality was reported by the USRDS also [18].
Concerning main causes of death, we found no significant differences between the two cohorts. Cardiovascular disease remained the main cause of death. Though a decreasing trend was registered ($45.6\%$ in 2017 of all causes of death vs. $50\%$ in 2012), the difference did not reach statistical significance. Decreasing trends in cardiovascular mortality among HD patients have been reported by many national and regional registries. However, cardiovascular disease remains the main cause of death in HD-treated ESRD patients (Australia $30.6\%$, New Zeeland $36.2\%$, USA 2020 $51.5\%$, Europe $39\%$) [16, 18, 37]. The 2020 Romanian Renal Registry evidenced a cardiovascular cause mortality of $53\%$ in Romanian HD patients [14]. In our cohorts, sepsis was the second cause of death with a decreasing prevalence from 2012 to 2017 (23.3–$20.8\%$—similar with the Romanian Renal Registry in 2018 i.e., $19\%$), still, remaining higher than reported by other registries (Australia $8.9\%$, USA $9.1\%$, New Zeeland $10.8\%$, and $16.2\%$ Europe 2015) [14, 18, 37, 38]. Cancer-related deaths in our cohort had an increasing trend, from $8.3\%$ in 2012 to $13.2\%$ in 2017. Cancer-related deaths varied across the national/regional registries also, being $1.4\%$ in New Zeeland, $2.3\%$ in the USA, 4. $4\%$ in Australia, and $7.8\%$ in Europe [18, 37, 38]. Our data are higher than those reported by the Romanian Renal Registry ($4\%$). It is evident that mortality and causes of death vary across countries/regions depending on socioeconomic status, renal replacement therapy (RRT) strategies, RRT practice patterns, disparities in access to treatment, detection of CKD and pre-dialysis care and treatment of CKD, and so on [36]. Even coding and reporting collected data to the registries may influence results as it was recently evidenced by comparing data published by USRDS and the Kaiser Permanente integrated health care system from California [39].
Multiple risk factors influence mortality in HD patients. In a 2017 meta-analysis (23 studies included), risk for all-cause mortality in HD patients was increased by age, presence of DM, previous CVD, higher CRP levels, higher levels of ferritin, higher levels of HbA1c, TnT and BNP, while higher BMI, hemoglobin, albumin, TIBC, ApoA2 and ApoA3 levels were associated with lower risk. Cardiovascular mortality risk was increased by age, gender (women vs. men), DM, previous CVD, HD duration, higher levels of ferritin, HDL, and HbA1c. Higher albumin, TIBC, ApoA2 levels turned out to be associated with lower risk. Worth to be mentioned the fact that mortality risk factors differ in Western and Eastern countries [40]. Differences concerning mortality risk exist between countries, HD centers, geographic areas, mainly dependent of group of investigated patients and collected data. For example, in Japan [2007], mortality risk was increased by high pulse pressure, presence of cerebrovascular disease, lower serum creatinine levels and low eKt/V [41]; in Spain [2021] in a community HD center, risk of mortality was associated to older age, acute deterioration of chronic kidney disease, use of catheters and hypoalbuminemia [42]. In our two cohorts, the risk of mortality increased related to dialysis duration, age, high phosphate and C-reactive protein levels. Higher hemoglobin and albumin levels turned out to be protective.
## Conclusion
The prevalence of ESRD patients needing HD therapy is increasing in our East-European population. More and more HD is performed at older ages and in higher number of DM patients. The 5-year survival is high in our cohorts (more than $62\%$ of the cases) based mainly on good-quality HD therapy and care. There are decreasing trends in cardiovascular mortality and sepsis-related deaths and increasing cancer-related mortality. Nevertheless, cardiovascular disease remains the main causes of death. If compared to other studies and annual data reports of regional and country registries, results are widely differing related to socioeconomic status, quality of care and therapy, and different reporting systems. Addressing modifiable risk of mortality and unifying reporting systems should decrease mortality in HD-treated ESRD patients and would permit developing regional therapeutic strategies to provide a better survival of HD patients.
## References
1. Pippias M, Jager KJ, Kramer A, Leivestad T, Sánchez MB, Caskey FJ, Collart F, Couchoud C, Dekker FW, Finne P, Fouque D, Heaf JG, Hemmelder MH, Kramar R, De Meester J, Noordzij M, Palsson R, Pascual J, Zurriaga O, Wanner C, Stel VS. **The changing trends and outcomes in renal replacement therapy: data from the ERA-EDTA registry**. *Nephrol Dial Transplant* (2016.0) **31** 831-841. DOI: 10.1093/ndt/gfv327
2. Liyanage T, Ninomiya T, Jha V, Neal B, Patrice HM, Okpechi I, Zhao MH, Lv J, Garg AX, Knight J, Rodgers A, Gallagher M, Kotwal S, Cass A, Perkovic V. **Worldwide access to treatment for end-stage kidney disease: a systematic review**. *Lancet* (2015.0) **385** 1975-1982. DOI: 10.1016/S0140-6736(14)61601-9
3. Thomas B, Wulf S, Bikbov B, Perico N, Cortinovis M, de Courville Vaccaro K, Flaxman A, Peterson H, Delossantos A, Haring D, Mehrotra R, Himmelfarb J, Remuzzi G, Murray C, Naghavi M. **Maintenance dialysis throughout the world in years 1990 and 2010**. *J Am Soc Nephrol* (2015.0) **26** 2621-2633. DOI: 10.1681/ASN.2015060660
4. van der Tol A, Lameire N, Morton RL, Van Biesen W, Vanholder R. **An international analysis of dialysis services reimbursement**. *CJASN* (2019.0) **14** 84-93. DOI: 10.2215/CJN.0815071
5. 5.Chantrel F, de Cornelissen F, Deloumeaux J, Lange C, Lassalle M; registre REIN. Survie et mortalité des patients en IRCT [Survival and mortality in ESRD patients]. Nephrol Ther. 2013 Sep;9 Suppl 1:S127-37. French. 10.1016/S1769-7255(13)70042-7. PMID: 24119578
6. 6.ERA-EDTA Registry: ERA-EDTA Registry Annual Report 2018, Amsterdam UMC, location AMC, Department of Medical Informatics, Amsterdam, the Netherlands, 2021
7. 7.United States Renal Data System (2021) USRDS Annual Data Report: Epidemiology of kidney disease in the United States. National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases, Bethesda, MD, 2021
8. Thompson S, James M, Wiebe N. **Cause of death in patients with reduced kidney function**. *J Am Soc Nephrol* (2015.0) **26** 2504-2511. DOI: 10.1681/ASN.2014070714
9. den Hoedt CH, Bots ML, Grooteman MP. **Should we still focus that much on cardiovascular mortality in end stage renal disease patients? The convective transport study**. *PLoS ONE* (2013.0) **8** e61155. DOI: 10.1371/journal.pone.0061155
10. Vogelzang JL, van Stralen KJ, Jager KJ. **Trend from cardiovascular to non-cardiovascular late mortality in patients with renal replacement therapy since childhood**. *Nephrol Dial Transplant* (2013.0) **28** 2082-2089. DOI: 10.1093/ndt/gft048
11. 11.Kidney Disease: Improving Global Outcomes (KDIGO) Anemia Work Group (2012) KDIGO clinical practice guideline for anemia in chronic kidney disease. Kidney Int Suppl. 2:279–335
12. 12.Kidney Disease: Improving Global Outcomes (KDIGO) CKD–MBD Work, Group (2017) KDIGO clinical practice guideline for the diagnosis, evaluation, prevention, and treatment of chronic kidney disease–mineral and bone disorder (CKD–MBD). Kidney Int Suppl. 7:1–59
13. 13.Mircescu G, Garbeata L, Stefan G, Podgoreanu E Romanian Renal Registry: 2014 Preliminary Annual Report https://www.registrulrenal.ro/resurse_statistica/2014_Raport-RO-2014.pdf. Accessed Nov 2022
14. 14.Ștefan G, Gârneață L, Podgoreanu E, Mircescu G Romanian Renal Registry Report 2019-2920 (Romanian) https://www.registrulrenal.ro/resurse_statistica/Raport_RRR_2019_2020.pdf. Accessed Nov 2022
15. 15.ERA-EDTA Registry: ERA-EDTA Registry Annual Report 2017. Amsterdam UMC, location AMC, Department of Medical Informatics, Amsterdam, the Netherlands, 2019
16. 16.ERA-EDTA Registry: ERA-EDTA Registry Annual Report 2019. Amsterdam UMC, location AMC, Department of Medical Informatics, Amsterdam, the Netherlands, 2021
17. Okechukwu CH, Lopez AA, Stack AG, Feng S, Wolfe RA. **Port FK Impact of years of dialysis therapy on mortality risk and the characteristics of longer term dialysis survivors**. *Am J Kidney Dis* (2002.0) **39** 533-538. DOI: 10.1053/ajkd.2002.31403
18. 18.United States Renal Data System (2022) USRDS Annual Data Report: Epidemiology of kidney disease in the United States. National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases, Bethesda, MD, 2022
19. Walther CP, Carter CW, Low CL, Williams P, Rifkin DE, Steiner RW, Ix GH. **Interdialytic creatinine change versus predialysis creatinine as indicators of nutritional status in maintenance hemodialysis**. *Nephrol Dial Transplant* (2012.0) **27** 771-776. DOI: 10.1093/ndt/gfr389
20. Tanaka S, Ninomiya T, Taniguchi M, Masanori TM, Masutani K, Ooboshi H, Kitazono H, Tsuruya K. **Impact of blood urea nitrogen to creatinine ratio on mortality and morbidity in hemodialysis patients: the Q-Cohort Study**. *Sci Rep* (2017.0) **7** 14901. DOI: 10.1038/s41598-017-14205-2
21. Macdougall IC, White C, Anker SD, Bhandari S, Farrington K, Kalra PA, McMurray JJV, Murray H, Tomson CRV, Wheeler DC, Winearls CG, Ford I. **Intravenous iron in patients undergoing maintenance hemodialysis**. *N Engl J Med* (2019.0) **380** 447-458. DOI: 10.1056/NEJMoa1810742
22. Karaboyas A, Morgenstern H, Pisoni RL, Zee J, Vanholder R, Jacobson SH, Inaba M, Loram LC, Port FK, Robinson BM. **Association between serum ferritin and mortality: findings from the USA, Japan and European Dialysis Outcomes and Practice Patterns Study**. *Nephrol Dial Transplant* (2018.0) **33** 2234-2244. DOI: 10.1093/ndt/gfy190
23. Patel L, Bernard LM, Elder GJ. **Sevelamer versus calcium-based binders for treatment of hyperphosphatemia in CKD: a meta-analysis of randomized controlled trials**. *Clin J Am Soc Nephrol* (2016.0) **11** 232-244. DOI: 10.2215/CJN.06800615
24. Komaba H, Wang M, Taniguchi M, Yamamoto S, Nomura T, Schaubel DE, Smith AR, Zee J, Karaboyas A, Bieber B, Fukagawa M, Tentori F. **Initiation of sevelamer and mortality among hemodialysis patients treated with calcium-based phosphate binders**. *Clin J Am Soc Nephrol* (2017.0) **12** 1489-1497. DOI: 10.2215/CJN.13091216
25. Zhang P, Sang S, Huang J, Feng S, Feng C, Wang R. **Effect of calcium-based phosphate binders versus sevelamer on mortality of patients with hemodialysis: a meta-analysis**. *Iran J Kidney Dis* (2022.0) **16** 215-227. PMID: 35962636
26. Ruospo M, Palmer SC, Natale P, Craig JC, Vecchio M, Elder GJ, Strippoli GF. **Phosphate binders for preventing and treating chronic kidney disease-mineral and bone disorder (CKD-MBD)**. *Cochrane Database Syst Rev* (2018.0) **8** CD006023. DOI: 10.1002/14651858.CD006023.pub3
27. Zoccali C, Moissl U, Chazot C, Mallamaci F, Tripepi G, Arkossy O, Wabel P. **Stuard S chronic fluid overload and mortality in ESRD**. *J Am Soc Nephrol* (2017.0). DOI: 10.1681/ASN.2016121341
28. Navarrete JE, Rajabalan A, Cobb J, Lea JP. **Proportion of hemodialysis treatments with high ultrafiltration rate and the association with mortality**. *Kidney 360* (2022.0) **3** 1359-1366. DOI: 10.34067/KID.0001322022
29. Slinin Y, Babu M, Ishani A. **Ultrafiltration rate in conventional hemodialysis: where are the limits and what are the consequences?**. *Semin Dial* (2018.0) **31** 544-550. DOI: 10.1111/sdi.12717
30. Khan MAB, Hashim MJ, King JK, Govender RD, Mustafa H, Al KJ. **Epidemiology of type 2 diabetes - global burden of disease and forecasted trends**. *J Epidemiol Glob Health* (2020.0) **10** 107-111. DOI: 10.2991/jegh.k.191028.001
31. Cheng HT, Xu X, Lim PS, Hung KY. **Worldwide epidemiology of diabetes-related end-stage renal disease, 2000–2015**. *Diabetes Care* (2021.0) **44** 89-97. DOI: 10.2337/dc20-1913
32. Harding JL, Morton JI, Shaw JE, Patzer RE, McDonald SP, Magliano DJ. **Changes in excess mortality among adults with diabetes-related end-stage kidney disease: a comparison between the USA and Australia**. *Nephrol Dial Transplant* (2022.0) **37** 2004-2013. DOI: 10.1093/ndt/gfab315
33. Lim WH, Johnson DW, Hawley C, Lok C, Polkinghorne KR, Roberts MA, Boudville N, Wong G. **Type 2 diabetes in patients with end-stage kidney disease: influence on cardiovascular disease-related mortality risk**. *Med J Aust* (2018.0) **209** 440-446. DOI: 10.5694/mja18.00195
34. González-Pérez A, Saez M, Vizcaya D. **Incidence and risk factors for mortality and end-stage renal disease in people with type 2 diabetes and diabetic kidney disease: a population-based cohort study in the UK**. *BMJ Open Diabetes Res Care* (2021.0) **9** e002146. DOI: 10.1136/bmjdrc-2021-002146
35. Robinson BM, Akizawa T, Jager KJ, Kerr PG, Saran R, Pisoni RL. **Factors affecting outcomes in patients reaching end-stage kidney disease worldwide: differences in access to renal replacement therapy, modality use, and haemodialysis practices**. *Lancet* (2016.0) **388** 294-306. DOI: 10.1016/S0140-6736(16)30448-2
36. Thurlow JS, Joshi M, Yan G, Norris KC, Agodoa LY, Yuan CM, Nee R. **Global epidemiology of end-stage kidney disease and disparities in kidney replacement therapy**. *Am J Nephrol* (2021.0) **52** 98-107. DOI: 10.1159/000514550
37. 37.ANZDATA Registry. 42nd Report, Chapter 3: mortality in end stage kidney disease. Australia and New Zealand Dialysis and Transplant Registry, Adelaide, Australia. 2019. Available at: http://www.anzdata.org.au. Accessed Nov 2022
38. Vogelzang JL, van Stralen KJ, Noordzij M, Diez JA, Carrero JJ, Couchoud C. **Mortality from infections and malignancies in patients treated with renal replacement therapy: data from the ERA-EDTA registry**. *Nephrol Dial Transplant* (2015.0) **30** 1028-1037. DOI: 10.1093/ndt/gfv007
39. Bhandari SK, Zhou H, Shaw SF, Shi J, Tilluckdharry NS, Rhee CM, Jacobsen SJ, Sim JJ. **Causes of death in end-stage kidney disease: comparison between the United States renal data system and a large integrated health care system**. *Am J Nephrol* (2022.0) **53** 32-40. DOI: 10.1159/000520466
40. Ma L, Zhao S. **Risk factors for mortality in patients undergoing hemodialysis: a systematic review and meta-analysis**. *Int J Cardiol* (2017.0) **238** 151-158. DOI: 10.1016/j.ijcard.2017.02.095
41. Ajiro J, Alchi B, Narita I, Omori K, Kondo D, Sakatsume M, Kazama JJ, Akazawa K, Gejyo F. **Mortality predictors after 10 years of dialysis: a prospective study of Japanese hemodialysis patients**. *CJASN* (2007.0) **2** 653-660. DOI: 10.2215/CJN.03160906
42. de Arriba G, Avila GG, Guinea MT, Alia IM, Herruzo J, Ruiz BR, Tejeiro RD, Rubio MEL, Poyatos CV, Roldán C. **Mortality of hemodialysis patients is associated with their clinical situation at the start of treatment**. *Nefrología (English Edition)* (2021.0) **41** 461-466. DOI: 10.1016/j.nefroe.2021.10.006
|
---
title: 'Effects of Alternate Insulin Pump Settings in Patients With Type 1
Diabetes During Ramadan: A Randomized Pilot Study'
authors:
- Ghufran AlGhatam
- Derek O’Keeffe
- Husain Taha
journal: Journal of Diabetes Science and Technology
year: 2021
pmcid: PMC10012352
doi: 10.1177/19322968211059217
license: CC BY 4.0
---
# Effects of Alternate Insulin Pump Settings in Patients With Type 1
Diabetes During Ramadan: A Randomized Pilot Study
## Abstract
### Background:
Various studies have evaluated the safety and efficacy of using insulin pumps during Ramadan; some of them demonstrated favorable outcomes in reducing hypoglycemia and hyperglycemia. However, there is no consensus on the recommendations for basal insulin adjustments and the utilization of technical features of insulin pumps to improve glycemic control.
### Objectives:
We aimed to investigate the effects of different insulin pump settings on time in range in patients with type 1 diabetes during Ramadan.
### Methods:
In this randomized pilot study, 30 patients classified to have low to moderate risk for fasting were assigned to either a control group to receive basal insulin adjustments only or an intervention group to use the temporary basal rate and extended bolus features in addition to the basal insulin modifications. The percentage of time spent at different glucose ranges was measured by continuous glucose monitoring.
### Results:
The percentage of time spent within target (70-180 mg/dL) increased significantly in the intervention group from 63.0 ± 10.7 to 76 ± $16.2\%$ (mean difference, $27\%$ points; $P \leq .001$). The percentage of time spent in hyperglycemia level 1 (>180 mg/dL) and level 2 (>250 mg/dL) met the criterion of significance, indicating that the intervention group spent less time in hyperglycemia. However, there was no significant difference in the percentage of time spent in hypoglycemia ranges.
### Conclusions:
Incorporating technological approaches of pump therapy with clinical practice guidelines could improve glycemic control during Ramadan.
## Introduction
Ramadan fasting (RF) is one of the five pillars of Islam. Muslims are refrained from eating, drinking, smoking, and sexual intercourse, and abstained from consuming any oral medications, starting from sunrise until sunset for 29 to 30 days each year. Around 1.8 billion Muslims worldwide devote themselves to fasting during Ramadan annually 1 because it is a compulsory deed of worship for all healthy Muslims after puberty. Nevertheless, individuals with chronic conditions, including some people with diabetes (PWD), are religiously and medically exempted from fasting. Despite this, the estimated number of fasting Muslims with diabetes is at least 50 million globally,2,3 and many insist on fasting sometimes against the medical recommendations.
Ramadan fasting entails distinctive changes in food and fluids consumption 4 that could potentially induce metabolic alterations in glucose metabolism and insulin sensitivity.5-7 There are also inter-individual variabilities in the glycemic parameters, which could be attributed to the cultural differences, dietary patterns, and the fasting duration across different geographical regions.6,8 In PWD, the process of glucose hemostasis is complex due to the pharmacokinetics and pharmacodynamics of different medications, including insulin. Patients who fast, especially those with type 1 diabetes mellitus (T1DM), are predisposed to excessive glycogenolysis, gluconeogenesis, and increased ketogenesis,9-11 resulting in increased risk for hypoglycemia, hyperglycemia, diabetic ketoacidosis, and dehydration.10-12 Still, only a few studies investigated glucose excursions during Ramadan using continuous glucose monitoring (CGM).13-15 These studies suggest that RF causes higher rates of hyperglycemia than hypoglycemia, as CGM profiles revealed typical patterns of a rapid spike after iftar (sunset meal) that last overnight, followed by a second rise after the suhoor (pre-dawn meal), with prolonged glucose decline over fasting hours.
Individuals with T1DM who are fasting throughout Ramadan constitute a unique population. Continuous subcutaneous insulin infusion (CSII) is an established therapy option offered for this category. This form of therapy has several advantages over multiple dose injection therapy for fasting individuals because of the capability to adjust insulin doses according to the individual’s physiological requirements during fasting hours. In addition, sensor augmented pumps (SAPs) are an advanced form of CSII therapy, which controls insulin delivery by a glucose sensor with a relevant algorithm. Sensor augmented pumps’ superiority to traditional CSII has been demonstrated in randomized controlled clinical trials; therefore, these devices provide innovative protection against the risks associated with fasting. A cluster of studies revealed the safety of using CSII therapy to reduce hypoglycemia 2,16-18 and improve glycemic variability17,18 during RF. Although these data suggest the favorable effects of insulin pumps on glucose control, achieving glycemic control and time-in-range [TIR] goals during Ramadan remain challenging for many patients. 19 *There is* a dearth in the research field examining the optimal clinical uses of insulin pump technology in diabetes management during RF, and very few studies have evaluated TIR in SAPs treated patients during Ramadan. In the present study, we propose that the potential variabilities in glucose patterns can be further managed by expanding the use of different pump technology features. The temporary basal rates (TBR) and extended bolus (EB) options were used to investigate their effect on TIR among T1DM patients during RF.
## Selection and Description of Participants
Thirty individuals participated in this study, who were selected from the outpatient clinics at Salmaniya Medical Complex, which is the largest public secondary and tertiary care hospital in Bahrain. We recruited men and women above 18 years old with T1DM who use SAP therapy (MiniMed 640G) with Guardian Link and Enlite 2 sensor (Medtronic, USA). The patients intended to fast during the month of Ramadan between April 13, 2021, and May 12, 2021. The inclusion was limited to participants with sufficient technical knowledge to communicate with the research team online due to Covid-19 restrictions (Figure 1). All candidates received pre-fasting assessment using the International Diabetes Federation–Diabetes and Ramadan (IDF-DAR) risk stratification calculator, which is a scoring system to assess fasting eligibility for PWD by considering various factors affecting fasting 20; only participants with low to moderate risk were included. The exclusion criteria included pregnancy, diabetes complications, history of severe hypoglycemia, or diabetic ketoacidosis in the last six months. Volunteers with high fasting risk scores were also excluded. An informed consent form was obtained from all participants, and the hospital’s research ethics committee approved the study (SHCRSC Ref. No. 24250221).
**Figure 1.:** *Consort flow diagram.*
## Experimental Protocol
This study used the IDF-DAR 2021 practice guidelines for diabetes management in CSII users. Both groups received general education about fasting and healthy eating habits during Ramadan. According to each participant’s individual needs and pre-Ramadan glucose control, basal insulin was reduced by $20\%$ to $40\%$ in the last four hours of fast and increased by $10\%$ to $30\%$ in the first three hours of iftar. The bolus insulin ratio remained unchanged as before Ramadan. The smartguard feature, which suspends insulin when sensor glucose approaches a pre-defined low limit (65 mg/dL), was activated for both groups to prevent hypoglycemia during fasting. The data from the devices were downloaded and reviewed. Throughout the trial, participants were encouraged to report any serious adverse events like pump malfunctioning, severe hypoglycemia (defined as hypoglycemia that necessitates other person’s assistance due to altered consciousness), and hyperglycemia with ketones.
## The TBR Feature
Patients in the intervention group received additional training on TBR feature to adjust basal insulin by ±$10\%$-$30\%$ for two to three hours to optimize glucose control with recurrent hypoglycemia or suspend before low before breaking the fast or with persistent hyperglycemia two hours post-iftar.
## The EB Feature
Further education was delivered to the intervention group about the optimal utilization of EB delivery in the form of the dual-wave bolus, which delivers insulin instantly followed by an extended delivery over several hours, to match insulin delivery with the high in fat and protein content of traditional meals. Participants were instructed to administer the EB 10 minutes before the meal as a $50\%$:$50\%$ or $60\%$:$40\%$ bolus: square-wave over two hours, according to the glucose reading pre-meal and meal composition.
## Outcomes
Time in range is a parameter that evaluates glucose control by the percentage of time a person with diabetes spends within the target range of 70 to 180 mg/dL. The primary outcome of this study is the time spent within the target range (70-180 mg/dL), which was measured before and after RF and compared between the two study groups. The secondary outcomes were the average glucose and the percentage of time spent in the hypoglycemic range defined as level 1 (<70 mg/dL) and level 2 (<54 mg/dL), and percentage of time spent in hyperglycemic range in level 1 (>180 mg/dL) and level 2 (>250 mg/dL).
## Statistics
The SPSS v27 software (SPSS Inc, Chicago, IL) was used to perform the statistical analyses. The descriptive statistics are presented as mean ± standard deviation (SD) and percentages, depending on data distribution. All variables were tested for normal distribution by the Shapiro-Wilk test, Levene’s test, and Box’s test. The mixed-design analysis if variance (ANOVA) test and the independent t-test were used to compare the difference in the percentage of time spent at different glucose ranges between the two study groups. A P value <.05 was considered statistically significant, and all tests were two-tailed.
## Results
A total of 36 patients were recruited, and 30 completed the study between April 10, 2021, and May 15, 2021. They were randomly assigned to either the control group (18 patients) or the experimental group (18 patients); the statistical analysis includes data for the participants who completed the study (15 participants in each group; Figure 1). In all, $42\%$ of the participants were men, the patients’ mean age was 22.4 ± 3.9 years, and baseline glycated hemoglobin level ranged from $5.5\%$ to $9.3\%$, with a mean of ($7.5\%$ ± 1.1). The baseline characteristics of the study group are listed in Table 1.
**Table 1.**
| Characteristic | Overall | Control (n = 12) | Experiment (n= 12) | P value* |
| --- | --- | --- | --- | --- |
| Age | 22.4 ± 3.9 | 22.6 ± 4.2 | 22.2 ± 3.8 | NA* |
| Sex: n (%) | Sex: n (%) | Sex: n (%) | Sex: n (%) | Sex: n (%) |
| Female | NA* | 7 (58%) | 7 (58%) | NA* |
| Male | NA* | 5 (42%) | 5 (42%) | NA* |
| HbA1c % | 58.5 ± 12.1 mmol/mol | 63.9 ± 9.8 mmol/mol | 53 ± 12 mmol/mol | NA* |
| HbA1c % | 7.5% ± 1.1 | 8% ± 0.9 | 7% ± 1.1 | |
| FD | 26 ± 3.9 | 25 (83%) | 28 (93%) | .07 |
| PPH events | 8 ± 5.4 | 12 ± 4.9 | 5 ± 3.2 | <.001 |
The CGM recordings were processed in accordance with international consensus recommendations on the use of CGM. 21 This study included data for 14 consecutive days before fasting and 30 days during the study period in Ramadan with a minimum of $70\%$ of CGM data capture. The mean glucose improved significantly in the intervention, whereas it declined in the control group at the end of Ramadan. The percentage of time spent within the target range (70-180 mg/dL) increased significantly in the intervention group from $63.0\%$ ± $10.7\%$ at baseline to $76\%$ ± $16.2\%$ at the end of Ramadan ($P \leq .001$). Conversely, the percentage of time spent in the level of the hyperglycemic range 1 (>180 mg/dL) and level 2 (>250 mg/dL) reduced significantly ($P \leq .001$). Moreover, no significant difference was observed between the groups in the percentage of time spent in the hypoglycemic ranges (Figure 2). There was a slight reduction in the total daily dose of insulin at the end of Ramadan comparing with before fasting total dose; however, the difference was not found to be significant when compared between groups (Table 2).
**Figure 2.:** *Comparison between different glucose ranges pre- and post-intervention.* TABLE_PLACEHOLDER:Table 2.
## Adverse Events
The participants did not report any adverse events during the study period; however, one patient from the control group reported pump malfunctioning, which was replaced on the same day. Neither group reported any events of severe hypoglycemia nor diabetic ketoacidosis during Ramadan. The mean of post-iftar hyperglycemia events was statistically significant (P value <.001) among the control and the experimental group 12 ± 4.9 versus 5 ± 3.2, respectively, in which about $21\%$ of all participants (of which one event occurred in the experimental group) reported the need to take exogenous insulin dose to correct level 2 hyperglycemia (>250 mg/dL). The number of hyperglycemia events is associated with the number of fasting days as all participants were instructed to break their fast when blood glucose values are below 70 mg/dL or above 300 mg/dL, even though participants in the experimental group fasted more days 28 ± 2.8 versus 25 ± 4.5, the difference was not statically significant.
## Discussion
The current study met its primary and secondary endpoints, demonstrating that people with T1DM treated with insulin pump therapy could safely improve the percentage of time spent within the target range without an increase in hypoglycemia or significant hyperglycemia, if they use the TBR and EB features to adjust insulin according to their requirements during the month of Ramadan.
Ramadan fasting creates changes in lifestyle, causing biochemical alterations that affect the overall glucose management, even in the presence of best practice guidelines. Our study proved that fasting during *Ramadan is* associated with radical changes in glucose profiles and TIR. We showed that the percentage of time spent within 70 to 180 mg/dL was notably prominent in the experimental group who received our intensified treatment protocol. Glucose at target range increased $27\%$ points with the advanced technology use compared over the use of standard practice guidelines. There were observed increases in glucose overall, which occurs mostly after iftar meal, yet significant reductions were also seen in hyperglycemic ranges in the intervention group when compared with controls.
Previous CGM studies on insulin-treated patients have reported that RF is associated with an increased percentage of time spent in hyperglycemia and a reduced percentage of time spent in the target range.19,22 Nonetheless, several studies reported fewer hypoglycemia episodes and improved glycemic variabilities among CSII-treated patients,2,17 which could be attributed to the benefits of reducing basal insulin infusion rates or suspending it to avoid hypoglycemia episodes during fasting. 16 Time in range has been recently approved as an outcome measure for glycemic control in clinical trials, 23 which enhances the effectiveness of CGM metrics by establishing treatment goals for the patients. Therefore, it could help in understanding the physiological changes associated with RF, as well as facilitating patients’ education on how to overcome glucose control barriers linked to Ramadan rituals. According to the International Consensus on CGM, each $5\%$ increase in TIR is associated with clinically significant benefits for individuals with T1DM. 21 Therefore, the $27\%$ points increase in TIR shown in our study has clinical relevance in that it would correspond to improved glucose management during Ramadan. There was also a significant difference in the mean glucose and the hyperglycemic ranges among the groups, which was mostly but not solely repeated after the iftar meal, despite the basal insulin increments during the fast-breaking period. The risk of hypoglycemia, which is of more clinical concern, was prominently low, with the basal insulin reduction in the last four to five hours of fasting, in the two groups.
The glucose variability is of paramount importance in glycemic management as emerging evidence suggests the potential association with several acute and chronic complications.19,24 Therefore, improving post-prandial hyperglycemia should be considered a strategy for preventing and managing diabetes complications.24,25 *In this* study, the experimental group experienced less average glucose and lower hyperglycemia episodes than the control group in both hyperglycemic levels 1 and 2. We speculate that the superior clinical outcomes in our study were probably driven by the benefits of using TBR and EB features for glucose management and reduction of post-iftar hyperglycemic events during Ramadan. This is supported by the fact that the number of post-iftar hyperglycemic events that occurred between 6 and 9 pm was considerably greater among participants in the control group, in which patients did not use these features and spent more time in the hyperglycemic range.
Other studies that investigated the impact of EB delivery on glycemic control have reported that the dual-wave bolus feature is particularly helpful to prevent prolonged post-prandial hyperglycemia resulting from the consumption of meals high in fat and protein.26,27 Klupa et al 30 have shown that frequent users of dual-wave bolus achieved improvements in HbA1c levels by $0.45\%$ ($$P \leq .0009$$) in two years of clinical observation. Similarly, Chase et al 31 verified that the dual-wave bolus was effective in achieving lower glucose levels four hours post-prandial with high carbohydrate, fat, and calories consumption. Consistently, our results advocate that dual-wave bolus can be an effective method for optimizing post-prandial glucose during Ramadan. Although there are no clear guidelines for administering dual-wave bolus for different meals, experts recommend extended insulin delivery for meals rich in complex carbohydrates, fat, and protein, 28 which is the case in Ramadan’s meals composition.
The TIR improvements seen in our protocol that employs different predictors affecting glucose are similar to that seen in a previous study, involving 150 participants aged 5 to 20 years treated with CSII therapy, in whom HbA1c decreased with the use of TBRs ($$P \leq .01$$). 26 This feature is very useful to manage unplanned high or low blood glucose levels resulting from exercise or other events. 29 Such findings suggest that patients may likely benefit from TBRs to achieve a higher TIR during Ramadan, while basal insulin adjustments are still required from the health care provider before starting the fast.
In addition, patients’ education has been identified as an important component to enhance glucose control during Ramadan.32,33 Al-Ozairi et al 34 showed that patients with T1DM, whether treated with injections or CSII, who underwent diabetes education involving Dose Adjustment for Normal Eating (DAFNE) and basal insulin reduction in a controlled fashion, were able to improve TIR during Ramadan. Correspondingly, consolidating practice guidelines with structured education on optimal clinical uses of insulin pump technology during Ramadan allows properly trained patients to monitor their glucose levels and adjust basal insulin infusion and/or insulin delivery according to carbohydrate consumption and meals composition to avoid hypoglycemia or hyperglycemia in Ramadan.
To our knowledge, this was the first study that examined the effects of employing different insulin pump settings to manage glucose during Ramadan. Importantly, the presented study included patients treated with SAPs who represent a particular category of T1DM patients. Therefore, the current findings provide insight into how to advise and manage this category of patients to avoid glycemic excursions associated with RF. Some of the strengths that contributed to the favorable outcomes in our study include the careful selection of participants, the patient retention rate, and the structured education session delivered before Ramadan that highlighted food choices usually consumed throughout Ramadan, which empowered participants to adhere to the assigned treatment protocols. Nevertheless, our study has certain limitations, including the small patient number and the wide range of lifestyle factors that might affect the results, which are beyond what we have covered in this study.
The SAPs provide a wide range of features for diabetes management, which could lead to better glycemic control when considering the dietary alterations encountered during Ramadan. Despite the small number of participants, the safety outcomes in this cohort were novel and promising. Extended bolus delivery and TBRs may indicate a decreased percentage of time spent in the hyperglycemic ranges and an increased percentage of time spent in the target range during the month of Ramadan. Yet, randomized controlled trials are needed to validate these findings.
## Conclusions
The present study created promising clinical data on glucose management during Ramadan in patients with T1DM treated with insulin pumps. It endorses the importance of following practice guidelines to optimize glycemic control during Ramadan. Nevertheless, our data add to the existing body of evidence and provide support for reviewing the current therapy guidelines. Our findings also point that current practice guidelines can incorporate technological approaches like TBRs and EB delivery to support glycemic control during the Ramadan fast. This study further emphasizes the influential role that health care providers can play in educating patients on how to fully benefit from all pump features during the month of Ramadan.
## References
1. Lipka M. **Muslims, and Islam: key findings in the U.S. and around the world**
2. Al-Arouj M, Assaad-Khalil S, Buse J. **Recommendations for management of diabetes during Ramadan: update 2010**. *Diabetes Care* (2010.0) **33** 1895-1902. DOI: 10.2337/dc10-0896.
3. Saeedi P, Petersohn I, Salpea P. **Global and regional diabetes prevalence estimates for 2019 and projections for 2030 and 2045: results from the International Diabetes Federation Diabetes Atlas**. *Diabetes Res Clin Pract* (2019.0) **157** 107843. DOI: 10.1016/j.diabres.2019.107843.
4. Alamoudi R, Alsubaiee M, Alqarni A. **Attitudes and habits of patients with type 1 diabetes during fasting Ramadan**. *J Clin Transl Endocrinol* (2018.0) **14** 1-4. DOI: 10.1016/j.jcte.2018.09.001.
5. Afandi B, Kaplan W, Al Hassani N, Hadi S, Mohamed A.. **Correlation between pre-Ramadan glycemic control and subsequent glucose fluctuation during fasting in adolescents with Type 1 diabetes**. *J Endocrinol Invest* (2017.0) **40** 741-744. DOI: 10.1007/s40618-017-0633-y.
6. Kaplan W, Afandi B.. **Blood glucose fluctuation during Ramadan fasting in adolescents with Type 1 diabetes: findings of continuous glucose monitoring**. *Diabetes Care* (2015.0) **38** e162-e163. DOI: 10.2337/dc15-1108.
7. Lessan N, Saadane I, Alkaf B. **The effects of Ramadan fasting on activity and energy expenditure**. *Am J
Clin Nutr* (2018.0) **107** 54-61. DOI: 10.1093/ajcn/nqx016.
8. Vasan SK, Karol R, Mahendri NV, Arulappan N, Jacob JJ, Thomas N.. **A prospective assessment of dietary patterns in Muslim subjects with type 2 diabetes who undertake fasting during Ramadan**. *Indian J Endocrinol Metab* (2012.0) **16** 552-557. DOI: 10.4103/2230-8210.98009.
9. Karamat MA, Syed A, Hanif W.. **Review of diabetes management and guidelines during Ramadan**. *J R Soc Med* (2010.0) **103** 139-147. DOI: 10.1258/jrsm.2010.090254.
10. Hassanein M, Al-Arouj M, Hamdy O. **Diabetes and Ramadan: practical guidelines**. *Diabetes Res Clin
Pract* (2017.0) **126** 303-316. DOI: 10.1016/j.diabres.2017.03.003.
11. Meo SA, Hassan A.. **Physiological changes during fasting in Ramadan**. *J Pak Med Assoc* (2015.0) **65**
12. Salti I, Bénard E, Detournay B. **A population-based study of diabetes and its characteristics during the fasting month of Ramadan in 13 countries: results of the epidemiology of diabetes and Ramadan 1422/2001 (EPIDIAR) study**. *Diabetes Care* (2004.0) **27** 2306-2311. DOI: 10.2337/diacare.27.10.2306.
13. Lessan N, Hannoun Z, Hasan H, Barakat MT.. *Diabetes Metab* (2015.0) **41** 28-36. DOI: 10.1016/j.diabet.2014.11.004.
14. Alabbood MH, Ho KW, Simons MR.. **The effect of Ramadan fasting on glycaemic control in insulin dependent diabetic patients: a literature review**. *Diabetes Metab Syndr* (2017.0) **11** 83-87. DOI: 10.1016/j.dsx.2016.06.028.
15. Alfadhli EM.. **Higher rate of hyperglycemia than hypoglycemia during Ramadan fasting in patients with uncontrolled type 1 diabetes: insight from continuous glucose monitoring system**. *Saudi Pharm
J* (2018.0) **26** 965-969. DOI: 10.1016/j.jsps.2018.05.006.
16. Bin-Abbas BS.. **Insulin pump therapy during Ramadan fasting in type 1 diabetic adolescents**. *Ann Saudi Med* (2008.0) **28** 305-306. DOI: 10.5144/0256-4947.2008.305.
17. Alamoudi R, Alsubaiee M, Alqarni A. **Comparison of insulin pump therapy and multiple daily injections insulin regimen in patients with type 1 diabetes during Ramadan fasting**. *Diabetes Technol
Ther* (2017.0) **19** 349-354. DOI: 10.1089/dia.2016.0418.
18. Elbarbary NS.. **Effectiveness of the low-glucose suspend feature of insulin pump during fasting during Ramadan in type 1 diabetes mellitus**. *Diabetes Metab Res Rev* (2016.0) **32** 623-633. DOI: 10.1002/dmrr.2781.
19. Saadane I, Ali T, El-Laboudi A, Lessan N.. **Ramadan fasting in insulin-treated patients is associated with potentially unfavourable changes in glucose metrics: a flash glucose monitoring (FGM) study**. *Diabetes Res Clin
Pract* (2021.0) **172** 108592. DOI: 10.1016/j.diabres.2020.108592.
20. Hassanein M, Afandi B, Al-Arouj M, Shaikh S. **Risk stratification of people with diabetes before Ramadan**. *Diabetes and Ramadan: Practical
Guidelines 2021* **2021** 91-97
21. Danne T, Nimri R, Battelino T. **International consensus on use of continuous glucose monitoring**. *Diabetes
Care* (2017.0) **40** 1631-1640. DOI: 10.2337/dc17-1600.
22. Saadane I, Ashraf T, Ali T, Lessan N.. **Diabetes and Ramadan: utility of flash-glucose monitoring derived markers of glycaemic control and comparison with glycosylated haemoglobin**. *Diabetes Res Clin Pract* (2019.0) **153** 150-156. DOI: 10.1016/j.diabres.2019.05.020.
23. Gabbay MAL, Rodacki M, Calliari LE. **Time in range: a new parameter to evaluate blood glucose control in patients with diabetes**. *Diabetol Metab Syndr* (2020.0) **12** 22. DOI: 10.1186/s13098-020-00529-z.
24. Ceriello A.. **Glucose variability and diabetic complications: is it time to treat?**. *Diabetes Care* (2020.0) **43** 1169-1171. DOI: 10.2337/dci20-0012.
25. Ceriello A.. **Postprandial hyperglycemia and diabetes complications. Is it time to treat?**. *Diabetes* (2005.0) **54** 1-7. DOI: 10.2337/diabetes.54.1.1.
26. Wilkinson J, McFann K, Chase HP.. **Factors affecting improved glycaemic control in youth using insulin pumps**. *Diabet Med* (2010.0) **27** 1174-1177. DOI: 10.1111/j.1464-5491.2010.03068.x.
27. Burdick J, Chase HP, Slover RH. **Missed insulin meal boluses and elevated hemoglobin A1c levels in children receiving insulin pump therapy**. *Pediatrics* (2004.0) **113**. DOI: 10.1542/peds.113.3.e221.
28. Lee S, Cao M, Sajid S, Hayes M, Choi L, Rother C.. **The dual-wave bolus feature in continuous subcutaneous insulin infusion pumps controls prolonged post-prandial hyperglycemia better than standard bolus in Type 1 diabetes**. *Diabetes Nutr
Metab* (2004.0) **17** 211-216. PMID: 15575341
29. Pańkowska E, Szypowska A, Lipka M, Szpotańska M, Błazik M, Groele L.. **Application of novel dual wave meal bolus and its impact on glycated hemoglobin A1c level in children with type 1 diabetes**. *Pediatr Diabetes* (2009.0) **10** 298-303. DOI: 10.1111/j.1399-5448.2008.00471.x.
30. Klupa T, Skupien J, Cyganek K, Katra B, Sieradzki J, Malecki MT.. **The dual-wave bolus feature in type 1 diabetes adult users of insulin pumps**. *Acta Diabetol* (2011.0) **48** 11-14. DOI: 10.1007/s00592-009-0173-9.
31. Chase HP, Saib SZ, MacKenzie T, Hansen MM, Garg SK.. **Post-prandial glucose excursions following four methods of bolus insulin administration in subjects with type 1 diabetes**. *Diabet Med* (2002.0) **19** 317-321. DOI: 10.1046/j.1464-5491.2002.00685.x.
32. Al-Agha AE, Kafi SE, Zain Aldeen AM, Khadwardi RH.. **Flash glucose monitoring system may benefit children and adolescents with type 1 diabetes during fasting at Ramadan**. *Saudi Med J* (2017.0) **38** 366-371. DOI: 10.15537/smj.2017.4.18750.
33. Mohamed K, Al-Abdulrazzaq D, Fayed A. **Fasting during the holy month of Ramadan among older children and adolescents with type 1 diabetes in Kuwait**. *J Pediatr Endocrinol Metab* (2019.0) **32** 843-849. DOI: 10.1515/jpem-2019.
34. Al-Ozairi E, El Samad A, Al Kandari J, Aldibbiat AM.. **Intermittent fasting could be safely achieved in people with type 1 diabetes undergoing structured education and advanced glucose monitoring**. *Front Endocrinol* (2019.0) **10** 849. DOI: 10.3389/fendo.2019.00849.
|
---
title: Physiological Condition-Dependent Changes in Ciliary GPCR Localization in the
Brain
authors:
- Kathryn M. Brewer
- Staci E. Engle
- Ruchi Bansal
- Katlyn K. Brewer
- Kalene R. Jasso
- Jeremy C. McIntyre
- Christian Vaisse
- Jeremy F. Reiter
- Nicolas F. Berbari
journal: eNeuro
year: 2023
pmcid: PMC10012409
doi: 10.1523/ENEURO.0360-22.2023
license: CC BY 4.0
---
# Physiological Condition-Dependent Changes in Ciliary GPCR Localization in the Brain
## Abstract
Primary cilia are cellular appendages critical for diverse types of Signaling. They are found on most cell types, including cells throughout the CNS. Cilia preferentially localize certain G-protein-coupled receptors (GPCRs) and are critical for mediating the signaling of these receptors. Several of these neuronal GPCRs have recognized roles in feeding behavior and energy homeostasis. Cell and model systems, such as Caenorhabditis elegans and Chlamydomonas, have implicated both dynamic GPCR cilia localization and cilia length and shape changes as key for signaling. It is unclear whether mammalian ciliary GPCRs use similar mechanisms in vivo and under what conditions these processes may occur. Here, we assess two neuronal cilia GPCRs, melanin-concentrating hormone receptor 1 (MCHR1) and neuropeptide-Y receptor 2 (NPY2R), as mammalian model ciliary receptors in the mouse brain. We test the hypothesis that dynamic localization to cilia occurs under physiological conditions associated with these GPCR functions. Both receptors are involved in feeding behaviors, and MCHR1 is also associated with sleep and reward. Cilia were analyzed with a computer-assisted approach allowing for unbiased and high-throughput analysis. We measured cilia frequency, length, and receptor occupancy. We observed changes in ciliary length, receptor occupancy, and cilia frequency under different conditions for one receptor but not another and in specific brain regions. These data suggest that dynamic cilia localization of GPCRs depends on properties of individual receptors and cells where they are expressed. A better understanding of subcellular localization dynamics of ciliary GPCRs could reveal unknown molecular mechanisms regulating behaviors like feeding.
## Significance Statement
Often, primary cilia localize specific G-protein-coupled receptors (GPCRs) for subcellular signaling. Cell lines and model systems indicate that cilia deploy dynamic GPCR localization and change their shape or length to modulate signaling. We used mice to assess neuronal cilia GPCRs under physiological conditions associated with the known functions of receptors and ciliopathy clinical features like obesity. We show that particular cilia with specific GPCRs appear to dynamically alter their length, while others appear relatively stable under these conditions. These results implicate multiple themes across cilia GPCR-mediated signaling and indicate that not all cilia modulate GPCR signaling using the same mechanisms. These data will be important for potential pharmacological approaches to target cilia GPCR-mediated signaling.
## Introduction
Cilia are nearly ubiquitous, small microtubule-based cellular appendages critical for proper development and homeostasis where they coordinate specific signaling pathways (Reiter and Leroux, 2017). Thus, cilia structure or function defects can result in many disorders with a broad array of clinical features (Reiter and Leroux, 2017). Collectively known as ciliopathies, these disorders are often associated with neural developmental or behavioral deficits. In addition, certain ciliopathies are associated with increased feeding behavior and obesity (Vaisse et al., 2017; Engle et al., 2021; Lee et al., 2022). Altered hypothalamic cilia signaling has been implicated in ciliopathies associated with obesity (Davenport et al., 2007; Loktev and Jackson, 2013; Sun et al., 2021; Wang et al., 2021b,c).
Despite their clinical relevance and an understanding of cilia-mediated signaling in development, little is known about the roles of cilia on terminally differentiated neurons in vivo and how they influence mammalian behaviors. A diverse set of G-protein-coupled receptors (GPCRs) appear to preferentially localize to cilia, including specific GPCRs with known roles in feeding behavior and energy homeostasis, such as melanin-concentrating hormone receptor 1 (MCHR1) and neuropeptide-Y receptor 2 (NPY2R) (Berbari et al., 2008a,b; Loktev and Jackson, 2013).
During embryonic development, dynamic localization of signaling machinery and a GPCR (GPR161) to the ciliary compartment in a ligand-dependent manner is critical for proper hedgehog signaling (Mukhopadhyay et al., 2013; Hwang and Mukhopadhyay, 2015; Pal et al., 2016). In addition, Chlamydomonas and Caenorhabditis elegans use cilia length, shape, vesicular shedding, and receptor localization changes to mediate signaling (Mukhopadhyay et al., 2008; Olivier-Mason et al., 2013; Wang et al., 2020, 2021a). Mammalian cell line data also clearly demonstrate the dynamic localization of ciliary GPCRs as a potential mechanism to mediate signaling, and ciliopathy mutations are associated with deficits in these processes (Ye et al., 2013; Nager et al., 2017; Phua et al., 2017; Shinde et al., 2020).
In mammalian adult homeostasis, less is understood about how cilia mediate GPCR signaling in the CNS. The most well studied examples are the photoreceptor and olfactory sensory neuron cilia, which mediate opsin/rhodopsin and odorant receptor signaling for vision and olfaction (Singla and Reiter, 2006; Berbari et al., 2009). Here, we sought to determine whether cilia GPCR localization, frequency, and length dynamics change within brain regions associated with both the specific GPCR function and ciliopathy-associated clinical features such as obesity. We focused on two ciliary GPCRs: MCHR1 and NPY2R. Both are expressed in the brain, including hypothalamic feeding centers. MCHR1 has also been implicated in sleep and reward (Pissios et al., 2008; Presse et al., 2014; Blanco-Centurion et al., 2019; Dilsiz et al., 2020). To determine whether these GPCRs dynamically localize to cilia in vivo, we assessed their localization under different feeding conditions. We hypothesized that cilia GPCRs throughout the CNS would dynamically localize to the compartment based on changes in signaling, similar to other model systems and cell line data.
## Mice
All procedures were approved by the Institutional Animal Care and Use Committee at Indiana University-Purdue University Indianapolis. Adult C57BL6/J mice were obtained from The Jackson Laboratory (stock #022409). Unless identified within the figure (see Fig. 2), all experiments were conducted in male animals. Unless stated otherwise, mice were housed on a standard 12 h light/dark cycle with ad libitum food and water.
## Feeding conditions
Fed mice were allowed ad libitum access to food, fasted mice had no food overnight (∼16 h), and Refed mice were given 4 h of ad libitum access to food immediately after an overnight fast.
## Diet-induced obesity
Mice were fed either a standard chow diet consisting of $13\%$ fat, $58\%$ carbohydrate, and $28.5\%$ protein caloric content (catalog #5001, LabDiet) or a calorie-rich, high-fat diet (HFD) consisting of $60\%$ fat, $20\%$ carbohydrate, and $20\%$ protein caloric content beginning at 8 weeks of age (catalog #D12492, ResearchDiets). Mice were weighed weekly before proceeding to tissue analysis after 11 weeks on these diets and the onset of obesity.
## Circadian time point conditions
Mice were randomly assigned to light or dark cycle perfusion groups. One hour before the light cycle [zeitgeber time 23 (ZT23)] and 4 h before the dark cycle (ZT8), mice were anesthetized and perfused under their respective dark/light conditions.
## MCHR1 antagonist treatment
As previously described, mice were given an injection of the MCHR1 antagonist GW803430 (GW; 3 mg/kg, i.p.; catalog #4242, Tocris Bioscience) or vehicle control for 7 d, 3 h after the start of the light cycle (Alhassen et al., 2022). One week before the start of injections, mice were singly housed. Body weights were measured on the first day before injections to calculate the correct vehicle volume and dosage of GW treatment. MCHR1 antagonist was made fresh daily at a concentration of 0.5 mg/ml in 2 ml aliquots, (1 mg of GW, 8 µl of acetic acid, 1.6 ml of water, 125 µl of $2\%$ Tween 80, and 100 µl of 1N NaOH). Mice were weighed on the morning of the last treatment day (day 7) and perfused 60–90 min after the last injection.
## Fixation and tissue processing
Mice were anesthetized with a 0.1 ml/10 g body weight dose of $2.0\%$ tribromoethanol (Sigma-Aldrich) and transcardially perfused with PBS, followed by $4\%$ paraformaldehyde in PBS (catalog #15710, Electron Microscopy Sciences). Brains were postfixed in $4\%$ paraformaldehyde for 4 h at 4°C and then cryoprotected using $30\%$ sucrose in PBS for 16–24 h. Cryoprotected brains were embedded in optimal cutting temperature compound (catalog #4585, Thermo Fisher Scientific) and sectioned at 15 µm.
## Immunofluorescence
Sections were washed with PBS for 5 min, then permeabilized and blocked in a PBS solution containing $1\%$ BSA, $0.3\%$ Triton X-100, $2\%$ (v/v) donkey serum, and $0.02\%$ sodium azide for 30 min at room temperature. Sections were incubated with primary antibodies in blocking solution overnight at 4°C. Primary antibodies include anti-MCHR1 (rabbit pAB; 1:250 dilution; catalog #711649, Thermo Fisher Scientific), anti-adenylate cyclase 3 [ACIII; 1:1000 dilution; chicken polyclonal antibody (pAb); CPCA-ACIII, Encor], anti-mCherry (chicken pAb; 1:1000 dilution; catalog NBP2-25158, Novus), anti-MCH (1:200 dilution; rabbit mAb; catalog #274415, Abcam). Sections were then washed with PBS before incubating with secondary antibodies for 1 h at room temperature. Secondary antibodies include donkey conjugated Alexa Fluor 647 and 488 (1:1000; Thermo Fisher Scientific) against appropriate species according to the corresponding primary. All primary and secondary solutions were made in the blocking solution described above. Slides were then washed in PBS and stained with Hoechst nuclear stain (catalog #H3570, Thermo Fisher Scientific) for 5 min at room temperature. Coverslips were mounted using SlowFade Diamond Antifade Mountant (catalog #S36972, Thermo Fisher Scientific).
## Mchr1 antibody validation
Brain sections from previously described Mchr1 knock-out mice (Mchr1KO) and fluorescent reporter mice (Mchr1mCherry) were used for immunofluorescence to confirm the fidelity of the anti-MCHR1 antibody used throughout (Fig. 1A,B; Jasso et al., 2021).
**Figure 1.:** *Antibody validation in MCHR1KO and Mchr1mCherry fusion allele animals. A, MCHR1 knock-out mice show ACIII-positive cilia but show no MCHR1-positive cilia. B, MCHR1 mCherry-tagged mice show colocalization of MCHR1 and mCherry tag-positive cilia. Scale bars, 10 µm. Hoechst nuclei blue stain was used. Arrows indicate example cilia. N = 3 animals/genotype.*
## Confocal imaging
All images were acquired using a Leica SP8 confocal microscope in resonant scanning mode using a 63×, numerical aperture 1.4 objective. For all images collected, 16 bit image files were used for subsequent analysis.
## Image analysis
Cilia analysis was performed as previously described (Bansal et al., 2021). Briefly, sum projection images from captured z-stacks were analyzed using the artificial intelligence module, which had been trained to recognize cilia in brain section images. As part of the GA3 recipe, objects <1 μm in length were removed from the analysis. There were four to five mice per experimental condition, with four images captured per brain nucleus.
## Statistical analysis
All statistical tests were performed using GraphPad Prism. All statistically significant observations are noted in the figures and specific tests used are named within the legends.
## Results
To understand whether cilia GPCRs dynamically localize in vivo under physiological contexts associated with receptor activity, we initially chose to assess the known ciliary GPCR MCHR1. We assessed its ciliary localization in conjunction with the broadly expressed CNS ciliary membrane-associated ACIII (Bishop et al., 2007; Berbari et al., 2008b; Hsiao et al., 2021; Kobayashi et al., 2021; Alhassen et al., 2022). We confirmed our MCHR1 antibody immunofluorescence specificity by observing the loss of ciliary staining in a Mchr1 knock-out allele mouse brain and colocalization with a Mchr1-mCherry knock-in fusion allele mouse (Fig. 1; Jasso et al., 2021). For our broader analysis of cilia localization, we used our recently reported computer-assisted approach for measuring cilia frequency, length, and fluorescence intensity (Bansal et al., 2021). This approach offers the advantages of being less biased and having higher throughput.
As the MCH and MCHR1 signaling axis displays sexual dimorphism, our initial analysis compared cilia frequency, length, and fluorescence intensity in adult male and female mice (Messina et al., 2006; Santollo and Eckel, 2008). Surprisingly, we did not observe differences in cilia frequency, length, or MCHR1 intensity in any of the brain regions assessed, including the hypothalamic arcuate (ARC) and paraventricular nucleus (PVN), and the nucleus accumbens (shell and core) between males and females (Fig. 2). Interestingly, we did observe length differences between MCHR1-only, (ACIII negative) positive cilia and MCHR1|ACIII double-positive cilia, where MCHR1|ACIII colocalized cilia were significantly longer (Fig. 2B). This length difference between the two cilia populations was observed throughout our data. As we did not observe differences between males and females, we continued the remaining studies using adult males.
**Figure 2.:** *MCHR1 cilia localization is similar in adult male and female mice. A, Representative immunofluorescence images of neuronal cilia (ACIII, green) and MCHR1 (red) in the Shell of males and females. Scale bars, 10 µm. Hoechst nuclei blue stain was used. Arrows indicate example cilia. B, Mean MCHR1 cilia frequency per animal in the ARC, PVN, and the core and shell of the nucleus accumbens for cilia that have only MCHR1 [MCHR1 (ACIII Negative)] and cilia that have both MCHR1 and ACIII (MCHR1|ACIII Colocalized). Mean MCHR1 cilia length and intensity in MCHR1 (ACIII Negative) cilia or in MCHR1|ACIII colocalized cilia in the ARC, PVN, and the core and shell (nested t test, p > 0.05 for all male vs female comparisons in each region). N = 5 animals/group with an average of 250 cilia per brain nuclei of each animal analyzed. *p < 0.05.*
MCHR1 function has been extensively implicated in feeding behaviors, body weight, and energy homeostasis (for recent review, see Al-Massadi et al., 2021). Its ligand, MCH, is increased following acute fasting (Simon et al., 2018). Upon a 16 h fast, we observed an increase in MCH ligand immunostaining in the lateral hypothalamus, the known site of MCH expression (Fig. 3A; Zamir et al., 1986). We next assessed the impact of fasting on ciliary MCHR1 in hypothalamic nuclei associated with this behavior, the ARC and PVN, and the nucleus accumbens, a site of high MCHR1 ciliary localization (Berbari et al., 2008a). We did not observe changes in cilia frequency or MCHR1 intensity (Fig. 3B,C). Surprisingly, we only observed significant fasting-associated increases in MCHR1|ACIII colocalized cilia length within the PVN (Fig. 3C). To determine whether body weight and obesity can influence MCHR1 ciliary localization, we assessed the brains of high-fat diet-induced obese mice (Fig. 4A). Obesity did not influence cilia frequency, length, or MCHR1 fluorescence intensity in the ARC, PVN, or accumbens (Fig. 4B,C). MCHR1 signaling has also been implicated in sleep/wake cycles (Blanco-Centurion et al., 2019). To determine whether MCHR1 cilia localization changes with the light cycle, we assessed brains at ZT8 (light) and ZT23 (dark). We initially assessed the suprachiasmatic nucleus (SCN), the classic region involved in circadian rhythms and where light cycle-associated cilia length changes have recently been implicated (Hastings et al., 2018; Tu et al., 2022). While we do not observe MCHR1-positive cilia in the SCN, we did note changes in ACIII cilia similar to those observed by Tu et al. ( 2022; Fig. 5A). Staining for the MCH ligand at both ZT8 and ZT23 did not show changes (Fig. 5B). Interestingly, we also observed changes in MCHR1 cilia frequency in the ARC and PVN during the light/dark cycle with more cilia being observed in the dark (ZT23; Fig. 6A,B). In addition, MCHR1|ACIII colocalized cilia length in the shell of the accumbens appeared shorter in the dark cycle (ZT23; Fig. 6B). In the ARC, the average MCHR1 fluorescence intensity was significantly reduced in both populations of cilia at ZT23 (Fig. 6B).
**Figure 3.:** *Acute feeding status alters MCHR1 length specifically in the PVN. A, MCH immunofluorescence staining (red) and intensity measurement (Heat) significantly increased under fasted conditions in the lateral hypothalamus (Student’s t test, p = 0.0024, 0.415 ± 0.120 a.u.). B, Representative immunofluorescence images of neuronal cilia (ACIII, green) and MCHR1 (red) in the PVN of ad libitum-fed (Fed) and fasted (Fast) animals. Scale bars, 10 µm. Hoechst nuclei blue stain was used. Arrows indicate example cilia. C, Mean MCHR1 cilia frequency per animal in the ARC, PVN, and the core and shell of the nucleus accumbens for cilia that have only MCHR1 [MCHR1 (ACIII Negative)] and cilia that have both MCHR1 and ACIII (MCHR1|ACIII Colocalized). Mean MCHR1 cilia length and intensity in cilia with just MCHR1 (ACIII Negative) or in MCHR1|ACIII colocalized cilia. Significant changes in MCHR1|ACIII colocalized cilia length were observed in the PVN (nested t test, p = 0.020, 0.62 ± 0.21 μm). N = 5 animals/treatment group with an average of 200 cilia/brain nucleus of each analyzed. *p < 0.05.* **Figure 4.:** *HFD-induced obesity does not influence MCHR1 cilia localization. A, High-fat diet-induced obese and chow-fed control animal body weights (Student’s t test; p = 0.008 at 2 weeks and is <0.0001 onward). B, Representative immunofluorescence images of neuronal cilia (ACIII, green) and MCHR1 (red) in the Shell of control diet (Chow) and HFD-induced obese males. Scale bars, 10 µm. Hoechst nuclei blue stain was used. Arrows indicate example cilia. C, Mean MCHR1 cilia frequency per animal in the ARC, PVN, and the core and shell of the nucleus accumbens for cilia that have only MCHR1 [MCHR1 (ACIII Negative)] and cilia that have both MCHR1 and ACIII (MCHR1|ACIII Colocalized). Mean MCHR1 cilia length and intensity in cilia with just MCHR1 [MCHR1 (ACIII Negative)] or in MCRH1|ACIII colocalized cilia (nested t test, p > 0.05). N = 5 animals per treatment group with an average of 250 cilia/animal and nuclei analyzed. *p < 0.05.* **Figure 5.:** *ACIII ciliary localization is altered at ZT23, while MCH levels do not change. A, SCN ACIII cilia length at ZT23 (dark cycle; nested t test, p = 0.0291, 0.74 ± 0.26 μm). B, MCH immunofluorescence staining (red) and intensity measurement (Heat) is not significantly different in the lateral hypothalamus between ZT8 and ZT23. *p < 0.05.* **Figure 6.:** *MCHR1 cilia localization is influenced by circadian rhythm. A, Representative immunofluorescence images of neuronal cilia (ACIII, green) and MCHR1 (red) in the shell at ZT8 (light cycle) and ZT23 (dark cycle) timepoints. Scale bars, 10 µm. Hoechst nuclei blue stain was used. Arrows indicate example cilia. B, MCHR1 cilia frequency per animal in the ARC, PVN, Core, and Shell at ZT8 and ZT23 for cilia that have only MCHR1 [MCHR1 (ACIII Negative)] and cilia that have both MCHR1 and ACIII (MCHR1|ACIII Colocalized; two-way ANOVA; ARC: p = 0.004, 73 ± 20 cilia; PVN: p = 0.005, 70 ± 20 cilia). Mean MCHR1 cilia length and intensity in MCHR1 (ACIII Negative) and MCHR1|ACIII colocalized cilia. Significant decreases in MCHR1 cilia length in MCHR1|ACIII cilia in the shell and significant decreases in MCHR1 (ACIII Negative) cilia fluorescence intensity in the ARC at ZT23 (nested t test; accumbens shell: p = 0.0089, −0.94 ± 0.23 µm; ARC: p = 0.0168, 0.386 ± 1.10 a.u.; p = 0.0147, −0.454 ± 1.24 a.u., respectively). N = 5 and 4 animals/treatment group, respectively, with an average of 200 cilia/animal and nuclei analyzed. *p < 0.05, **p < 0.01.*
After assessing multiple physiological conditions where MCHR1 function has been implicated, we next looked to see whether overt pharmacological antagonism could influence MCHR1 ciliary localization. Injection of the antagonist GW803430 for 7 d resulted in significant loss in body weight (Fig. 7A; Alhassen et al., 2022). MCHR1 antagonism increased the frequency of MCHR1|ACIII colocalized cilia in the ARC and in the PVN (Fig. 7C). Antagonism also increased ciliary length in the accumbens core and shell for both cilia populations (Fig. 7C). Interestingly, in the ARC cilia length increases were observed only in MCRH1 (ACIII negative) cilia (Fig. 7C).
**Figure 7.:** *Antagonism alters MCHR1 length in the ARC and NA. A, Antagonist treatment causes significant weight loss (Student’s t test; p = 0.002, 1.860 ± 0.4368 g). B, Representative immunofluorescence images of neuronal cilia (ACIII, green) and MCHR1 (red) in the shell of control animals (Vehicle) and MCHR1 antagonist-treated animals (Antagonist). Scale bars, 10 µm. Hoechst nuclei blue stain was used. Arrows indicate example cilia. C, MCHR1 cilia frequency in the ARC, PVN, Core, and Shell after vehicle and antagonist treatment. Significant increase in MCHR1 only cilia [MCHR1 (ACIII Negative)] after antagonist treatment in the ARC (two-way ANOVA; p = 0.008, 96 ± 28 cilia). Mean MCHR1 cilia length and fluorescence intensity in cilia with MCHR1 (ACIII Negative) and with MCHR1|ACIII-colocalized cilia. Significant changes in cilia length for both cilia populations in the ARC, Core, and Shell [MCHR1 (ACIII Negative) nested t test; ARC: p = 0.0089, 0.94 ± 0.23 µm; accumbens core: p = 0.0033, 0.97 ± 0.31 µm; accumbens shell: p = 0.0224, 0.89 ± 0.31 µm; MCRH1|ACIII Colocalized cilia: nested t test; accumbens core: p = 0.0003, 1.47 ± 0.24 µm; accumbens shell: p < 0.0001, 1.70 ± 0.22 µm, respectively]. N = 5 animals/treatment group with an average of 250 cilia/animal and nuclei analyzed. *p < 0.05, **p < 0.01.*
To determine whether these results are specific to MCHR1 or perhaps applicable to multiple neuronal ciliary GPCRs, we assessed the localization of NPY2R, another GPCR known to localize to cilia (Loktev and Jackson, 2013). We focused our analysis on the ARC as we did not observe NPY2R cilia localization in other brain regions of interest in males or females (Fig. 8). Within the ARC, we also did not observe changes in NPY2R cilia between sexes, in HFD-induced obesity or at different circadian times (Fig. 9). Similar to MCH, acute fasting also increases the levels of the NPY2R ligand NPY (Yasrebi et al., 2016). Thus, we sought to assess both MCHR1 and NPY2R on fasting and refed states (Fig. 10). We only observed significantly longer MCHR1 (ACIII-negative) cilia lengths in the refed condition compared with the fasted (Fig. 10A,B). However, we observed significant cilia length changes in NPY2R (ACIII-negative) and NPY2R|ACIII colocalized cilia. NPY2R cilia were significantly longer in both the ad libitum fed and refed conditions compared with the fasted condition (Fig. 10C,D). These results demonstrate that dynamic localization to cilia is dependent on properties of the individual receptor and the brain region of expression in vivo.
**Figure 8.:** *NPY2R does not localize to cilia in the PVN or nucleus accumbens. A–C, Representative immunofluorescence images of neuronal cilia (ACIII, green) and NPY2R (white) within the PVN (A), nucleus accumbens core (B), and nucleus accumbens shell (C). Scale bars, 10 µm. Hoechst nuclei blue stain was used. Arrows indicate example cilia.* **Figure 9.:** *NPY2R cilia localization in the ARC is unchanged among sexes, obesity, and circadian times. A, C, E, Representative immunofluorescence images of neuronal cilia (ACIII, green) and NPY2R (white) within the ARC between male and female, HFD obese and control chow, and at ZT8 (light) and ZT23 (dark). Scale bars, 10 µm. Hoechst nuclei blue stain was used. Arrows indicate example cilia. B, D, F, Mean NPY2R cilia frequency per animal for cilia that have only NPY2R [NPY2R (ACIII Negative)] and cilia that have both NPY2R and ACIII (NPY2R|ACIII Colocalized). Mean NPY2R cilia length and intensity in NPY2R (ACIII Negative) cilia or in NPY2R|ACIII colocalized cilia in the ARC under the conditions (Sex, HFD, Circadian).* **Figure 10.:** *NPY2R changes under different feeding conditions in the ARC. A, Representative immunofluorescence images of neuronal cilia (ACIII, green) and MCHR1 (red) in the ARC of ad libitum-fed (Fed), overnight fasted (Fast), and 4 h postrefeeding after fast (Refed) conditions. Scale bars, 10 µm. Hoechst nuclei blue was used. Arrows indicate example cilia. B, Mean MCHR1 cilia length and intensity in cilia that have only MCHR1 [MCHR1 (ACIII Negative)] and in cilia with both MCHR1 and ACIII (MCRH1|ACIII Colocalized Cilia) in Fed, Fast, and Refed animals. Significant increase in MCHR1 (ACIII Negative) cilia length on refeeding (nested one-way ANOVA: p = 0.004, 0.654 ± 0.201 µm). C, Representative immunofluorescence images of neuronal cilia (ACIII, green) and NPY2R (white) in the ARC of Fed, Fast, and Refed conditions. D, Mean NPY2R cilia length and intensity in cilia with only NPY2R [NPY2R (ACIII Negative)] or in cilia with both NPY2R and ACIII (NPY2R|ACIII Colocalized) in Fed, Fast, and Refed animals. NPY2R (ACIII negative) cilia in the ARC significantly change length on Fed, Fast, and Refed conditions (nested one-way ANOVA, p < 0.001, −2.47 ± 0.29 µm; p < 0.00011.45 ± 0.28 µm; p = 0.0002, −1.02 ± 0.25 µm, respectively). NPY2R|ACIII colocalized cilia are also significantly shorter on fasting and remain slightly shorter in the Refed condition compared with Fed (nested t test, p < 0.0001, −1.80 ± 0.41 µm; p = 0.0003, 1.43 ± 0.37 µm, respectively). Each data point represents a cilium. Scale bars, 10 µm. Hoechst nuclei blue stain was used. N = 5 animals/group with an average of 250 cilia/animal. *p < 0.05.*
## Discussion
Cilia are recognized as mediators of diverse signaling pathways, yet many questions remain unanswered regarding how they coordinate signaling. In cell line and heterologous expression systems in vitro, dynamic localization of receptors to the cilia membrane has been reported for a number of ciliary GPCRs, including MCHR1 (Ye et al., 2018). In vivo dynamic localization to the cilia as a means of signaling control has been best described for cilia-mediated hedgehog signaling during development (Bangs and Anderson, 2017). We sought to determine whether cilia broadly deploy dynamic GPCR localization in vivo to mediate signaling. We chose a ciliary receptor associated with several physiological states and phenotypes, including sexual dimorphic expression, acute feeding behavior, energy homeostasis, and sleep (Al-Massadi et al., 2021). MCHR1 also has the advantage of being the only known receptor for MCH in mice (Diniz and Bittencourt, 2019). In contrast, many other ciliary GPCRs are within a family of receptors for certain neuropeptides. For example, the ciliary somatostatin receptor 3 (SSTR3) is one of five receptors (SSTR1-5) for the ligand somatostatin (Yamada et al., 1992a, b, 1993).
Our initial assessment of MCHR1 focused on the hypothalamus for a number of reasons. Ciliopathies are known to have deficits in hypothalamic control of energy homeostasis (Davenport et al., 2007; Sun et al., 2021; Wang et al., 2021c). MCHR1 fails to localize properly in obese ciliopathy models of Bardet–Biedl syndrome (BBS; Berbari et al., 2008a). In addition, Mchr1 expression is observed in several hypothalamic nuclei under baseline conditions (Engle et al., 2018). MCHR1 signaling has also been extensively implicated in feeding behavior, energy homeostasis, and metabolism. Agonism or activation of the pathway is associated with increases in food intake, and loss-of-function alleles or pharmacological antagonism associated with weight loss (for recent review of MCH and MCHR1 signaling, see Al-Massadi et al., 2021).
We chose an antibody staining approach combined with a computer-assisted analysis as this combination was the best way to detect endogenous ciliary MCHR1 in an unbiased and high-throughput manner. It also allows us to readily observe hundreds of cilia per animal (Bansal et al., 2021; Jasso et al., 2021).
We were surprised to find that our analysis revealed that MCHR1 ciliary localization remained largely fixed across males and females, on fasting and diet-induced obesity, with only subtle significant changes observed in cilia length. We also observed that MCHR1|ACIII-colocalized cilia were significantly longer than MCHR1 only (ACIII negative) cilia. Our observation that ACIII cilia length changes within the SCN depending on the light or dark cycle as recently reported in a preprint (Tu et al., 2022), assured us that our analysis could detect broad-scale changes in cilia lengths, frequency, and localization. It was interesting that we also detected length decreases in MCHR1|ACIII colocalized cilia in the shell of the nucleus accumbens in the dark cycle (Becker-Krail et al., 2022). This suggests the potential for cilia-mediated signaling changes broadly in the brain based on light conditions.
Pharmacological MCHR1 antagonism demonstrated the most substantial changes in both cilia length and intensity across different brain regions, but this approach may not be physiologically relevant. However, this result is in line with what cilia have been proposed to do when their GPCR-associated signaling system is saturated or overwhelmed by changing their lengths and shedding cilia-specific vesicles (Nager et al., 2017; Phua et al., 2017). These phenomena have been directly observed for cilia in BBS cell models (Nager et al., 2017). It remains to be seen how common cilia length regulation and vesicular shedding is deployed as a means of cilia-mediated signaling in vivo. It is possible that both are important processes, but that under normal physiological conditions they remain challenging to detect in mammalian systems in vivo with currently available tools.
To further explore the possibility that other cilia GPCRs could be relatively stationary in vivo, we investigated another hypothalamic ciliary GPCR under physiological conditions in which it has been implicated: NPY2R and feeding status (Loktev and Jackson, 2013). Interestingly, for NPY2R, we observed significant changes in length for both cilia populations with fasted cilia being shorter and refed cilia being longer compared with ad libitum-fed animals. These data suggest that NPY2R cilia are more dynamic on acute changes in feeding when compared with MCHR1 cilia. At the neuroanatomical level, our data reveal that specific brain regions independently localize certain receptors to their cilia. In other words, the MCHR1/MCH signaling axis localization behaves differently dependent on the anatomic context. This opens up the possibility that ciliary GPCRs may be dynamic depending on what tissue is being investigated. For example, MCHR1 is potentially expressed in peripheral tissues, and its ciliary localization in these contexts is unclear (Balber et al., 2019). Overall, these data further point to the potential that many ciliary GPCRs may need to be assessed independently and in tissues and cells of interest to learn how their signaling is mediated in vivo.
At the receptor level, our data point to the potential for specific G-protein coupling being important for dynamic localization to cilia. MCHR1 is thought to be Gαi coupled while NPY2R is Gαs coupled. However, coupling at the cilia for most nonodorant ciliary GPCRs is undetermined (Loktev and Jackson, 2013; Saito et al., 2013). Our data also may reflect the inherent nature of some GPCRs being more dynamic at membranes compared with others (Schmidt et al., 2014). It is also possible that in some cases the pool of receptors that is critical for signaling is on the plasma membrane and not the ciliary membrane, and thus cilia localization appears stable for a given GPCR. Future studies will assess how G-protein coupling and other pools of receptors may specifically influence ciliary GPCR localization. For example, Gαs (e.g., NPY2R) ciliary receptors may be generally more dynamic to the compartment compared with those that couple to other Gα subunits (e.g., MCHR1).
Together our results demonstrate that dynamic localization to the ciliary compartment may not apply to some physiological conditions in vivo or be a common theme across ciliary GPCRs. Our results also suggest that only specific ciliary GPCRs use length control as a mechanism to mediate signaling, as may be the case for NPY2R but not MCHR1. Finally, our results also demonstrate that localization across different brain regions and nuclei that all possess the same ciliary GPCR are dynamically regulated differentially. For example, even on supraphysiological antagonism of MCHR1, we did not observe the same changes in cilia length and localization in all brain regions analyzed. Ultimately, a comprehensive understanding of how cilia mediate GPCR signaling could provide therapeutic opportunities for cilia-receptor ligands in conditions like obesity.
## Synthesis
appended below
## Author Response
We thank the reviewers for their feedback. Our updated manuscript attempts to address all of the critiques with editorial changes, revised figures, and additional data. We believe these changes improve the manuscript. Our responses to reviewer comments follow.
The writing of the paper could use improvement, particularly from an organizational standpoint. It was unclear from the abstract and even the introduction exactly what I was going to learn from this paper. For example, is this paper about the trafficking of specific GPCRs to and from the cilia in response to physiological conditions, or about physical regulation of cilia by these pathways in response to physiological conditions, or about cilia heterogeneity? Or maybe about each of these things? Some further clarity or perhaps a clear hypothesis would help this.
We appreciate this suggestion. While our work addresses aspects of all these questions, we have elected to focus on a couple of them in the abstract and introduction and leave the other parts for the discussion. For example, we now focus on GPCR dynamic localization to cilia in vivo to launch the results section. We offer a revised abstract and introduction to support this focus.
The abstract itself states that “no consistent theme” emerges from the study, which lessened my enthusiasm for the paper. This is an overly negative way of framing things right from the start, and is also ambiguous- what even were the themes being tested/examined?. I suspect this could be reframed with more careful consideration to what these findings *do* tell us, and by making more of an attempt to cohesively present this data.
We agree our framing needed improvement. We address this in the following ways: Abstract: We now explicitly state, “*These data* suggest that dynamic cilia localization of GPCRs may depend on properties of individual receptors.” Introduction: “To determine if these GPCRs dynamically localize to cilia in vivo, we assessed their localization under different feeding conditions. We hypothesized that cilia GPCRs throughout the CNS would dynamically localize to the compartment based upon changes in signaling, similar to other model systems and heterologous cell line data.”
Results: • We highlight two types of cilia throughout GPCR (ACIII negative) and GPCR|ACIII colocalized populations.
• We have incorporated more NPY2R data into new Figures 8, 9 and 10.
• Cilia frequency data is now throughout the figures as a separate graph.
• We now compare GPCR (ACIII Negative) and GPCR|ACIII *Colocalized cilia* population data across conditions within the same graphs.
• We explicitly indicate increases and decreases in frequency, length, and intensity. 2 Discussion: • We have added additional rationale for why we initially focused on MCHR1. It is the only known receptor for the MCH system in mice, whereas most other ciliary GPCRs lie within a family of receptor isoforms (i.e., somatostatin receptors 1-5).
• We have added ideas as to why one GPCR may be more dynamic than the other. For example, perhaps G-protein coupling preference is associated with dynamic localization (i.e., MCHR1-Gαi is stable at cilia while NPY2R-Gαs is dynamic at cilia). The NPY2R data had not been fully considered and felt a bit tacked on at the end, especially given that it shows some interesting responses. The authors should look at NPY2R under similar conditions to MCHR1 and compare and contrast these two GPCRs more extensively.
We agree and have looked at NPY2R in different conditions and brain regions; we have added these data to new Figures 8 and 9.
The distinction between MCHR1 and MCHR1/ACIII cilia was also somewhat ambiguous. Are the MCHR1 alone cilia ACIII negative or was ACIII just not evaluated there? If there are these distinct populations of cilia, it might be worth commenting further on what may be different between these populations.
We apologize for our lack of clarity. In all contexts, staining was simultaneously performed for the GPCR of interest (MCHR1 or NPY2R) and ACIII. To clarify for the reader, we have revised our wording within the results and the labels on the figures and graphs. The terms MCHR1 (ACIII negative) and MCHR1|ACIII colocalized cilia are used throughout. We also highlight the differences between these cilia populations in revised Figure 2, where we observe length differences. MCHR1|ACIII colocalized cilia are longer across all experiments compared to MCHR1 (ACIII negative) cilia. We have also included this as a discussion point in the revised manuscript.
Ciliary MCHR expression was analyzed in MCHR1 and MCHR1+ACII cilia. This appears to be an important addition. However, please also provide direct comparisons between MCHR1 vs MCHR1+ACII cilia. It would also be helpful to indicate whether MCHR1 only cilia represent a unique subpopulation of cilia - what might be some other markers available to classify them.
We have added the requested analysis throughout. Unfortunately, the prototypical markers for assessing cilia do not work in the adult brain (i.e., acetylated α-tubulin). In addition, markers thought to be ubiquitous to cilia (i.e., ARL13b) do not appear so in the adult brain. Thus, our analysis is limited to known ciliary GPCRs with established antibodies and ACIII. We considered assessing cilia using a cilia-tagged allele, but felt that this approach would come with its own caveats, so we pursued assessing endogenous ciliary GPCRs using antibody staining with appropriate knockout and transgenic controls for MCHR1. 3 The authors should comment on differences in cilia parameters between MCHR1 cilia and MCHR1+ACII cilia? It appears from Figure 1 that MCHR1 cilia might be shorter than MCHR1+ACII cilia.
We now highlight the differences in revised Figure 2. The authors found that there were no sex differences in ciliary characteristics between cells from male vs female brains. This is an important analysis, but in addition to ciliary parameters, please also indicate if there were sex differences in the proportion of cells in the ARC, PVN, NAc - for example, is there differences in ciliary frequency.
We have provided cilia frequency data throughout the revised manuscript.
Is there a difference in the intensity of MCH expression at ZT8 and ZT23?
We have provided this analysis in new Figure 5B. We do not observe significant changes in MCH staining in the LH between ZT8 and ZT23.
There were a few instances in reporting the Results where the authors indicate there was a significant difference in XXX (e.g., XXX = MCHR1 cilia length in PVN; changes in cilia frequency in ARC and PVN; etc) but the Results do not report the direction of the change (increase/decrease). Such revisions would be a better companion for the figures.
We agree and have added these changes directly to the results section.
Given that ciliary MCHR1 expression is resilient to physiological or perhaps even pharmacological manipulation, it would be appropriate to speculate on the role of ciliary MCHR1 expression in the Discussion. Is ciliary MCH receptor expression required for signaling? Is the receptor expressed elsewhere on the cell?
We have added several ideas to the revised discussion regarding this point. For example, the following has been added: *It is* possible that in some cases, the pool of receptor that is signaling is on the plasma membrane and not the cilia membrane and thus, cilia localization appears stable for a given GPCR. Minor: Figure 2A is labeled MCHR1 when it appears to be the ligand, MCH We have corrected this error.
Figure format - In the sample images of MCHR1, please add a pointer to indicate some examples of cilia in the image. Also, please outline the area that is shown in the higher magnification image. 4 We have added arrows and the outline of the inset to all data.
At the end of the third paragraph of the Results section the authors mention cilia frequency being unchanged under fasting or obesity and refer to data not shown. All data referenced in a manuscript should be shown for transparency and rigor.
We have added frequency data to all revised figures and have made all data available.
## References
1. Al-Massadi O, Dieguez C, Schneeberger M, López M, Schwaninger M, Prevot V, Nogueiras R. **Multifaceted actions of melanin-concentrating hormone on mammalian energy homeostasis**. *Nat Rev Endocrinol* (2021) **17** 745-755. DOI: 10.1038/s41574-021-00559-1
2. Alhassen W, Kobayashi Y, Su J, Robbins B, Nguyen H, Myint T, Yu M, Nauli SM, Saito Y, Alachkar A. **Regulation of brain primary cilia length by MCH signaling: evidence from pharmacological, genetic, optogenetic, and chemogenic manipulations**. *Mol Neurobiol* (2022) **59** 245-265. DOI: 10.1007/s12035-021-02511-w
3. Balber T, Bencurova K, Kiefer FW, Kulterer OC, Klebermass EM, Egger G, Tran L, Wagner KH, Viernstein H, Pallitsch K, Spreitzer H, Hacker M, Wadsak W, Mitterhauser M, Philippe C. **In vitro radiopharmaceutical evidence for MCHR1 binding sites in murine brown adipocytes**. *Front Endocrinol (Lausanne)* (2019) **10** 324. PMID: 31244769
4. Bangs F, Anderson KV. **Primary cilia and mammalian hedgehog signaling**. *Cold Spring Harb Perspect Biol* (2017) **9** a028175. DOI: 10.1101/cshperspect.a028175
5. Bansal R, Engle SE, Kamba TK, Brewer KM, Lewis WR, Berbari NF. **Artificial intelligence approaches to assessing primary cilia**. *J Vis Exp* (2021) e62521
6. Becker-Krail DD, Walker WH, Nelson RJ. **The ventral tegmental area and nucleus accumbens as circadian oscillators: implications for drug abuse and substance use disorders**. *Front Physiol* (2022) **13** 886704. PMID: 35574492
7. Berbari NF, Lewis JS, Bishop GA, Askwith CC, Mykytyn K. **Bardet-Biedl syndrome proteins are required for the localization of G protein-coupled receptors to primary cilia**. *Proc Natl Acad Sci U S A* (2008a) **105** 4242-4246. DOI: 10.1073/pnas.0711027105
8. Berbari NF, Johnson AD, Lewis JS, Askwith CC, Mykytyn K. **Identification of ciliary localization sequences within the third intracellular loop of G protein-coupled receptors**. *Mol Biol Cell* (2008b) **19** 1540-1547. DOI: 10.1091/mbc.e07-09-0942
9. Berbari NF, O’Connor AK, Haycraft CJ, Yoder BK. **The primary cilium as a complex signaling center**. *Curr Biol* (2009) **19** R526-R535. DOI: 10.1016/j.cub.2009.05.025
10. Bishop GA, Berbari NF, Lewis J, Mykytyn K. **Type III adenylyl cyclase localizes to primary cilia throughout the adult mouse brain**. *J Comp Neurol* (2007) **505** 562-571. DOI: 10.1002/cne.21510
11. Blanco-Centurion C, Luo S, Spergel DJ, Vidal-Ortiz A, Oprisan SA, Van den Pol AN, Liu M, Shiromani PJ. **Dynamic network activation of hypothalamic MCH neurons in REM sleep and exploratory behavior**. *J Neurosci* (2019) **39** 4986-4998. DOI: 10.1523/JNEUROSCI.0305-19.2019
12. Davenport JR, Watts AJ, Roper VC, Croyle MJ, van Groen T, Wyss JM, Nagy TR, Kesterson RA, Yoder BK. **Disruption of intraflagellar transport in adult mice leads to obesity and slow-onset cystic kidney disease**. *Curr Biol* (2007) **17** 1586-1594. DOI: 10.1016/j.cub.2007.08.034
13. Dilsiz P, Aklan I, Sayar Atasoy N, Yavuz Y, Filiz G, Koksalar F, Ates T, Oncul M, Coban I, Ates Oz E, Cebecioglu U, Alp MI, Yilmaz B, Atasoy D. **MCH neuron activity is sufficient for reward and reinforces feeding**. *Neuroendocrinology* (2020) **110** 258-270. DOI: 10.1159/000501234
14. Diniz GB, Bittencourt JC. **The melanin-concentrating hormone (MCH) system: a tale of two peptides**. *Front Neurosci* (2019) **13** 1280. PMID: 31849590
15. Engle SE, Antonellis PJ, Whitehouse LS, Bansal R, Emond MR, Jontes JD, Kesterson RA, Mykytyn K, Berbari NF. **A CreER mouse to study melanin concentrating hormone signaling in the developing brain**. *Genesis* (2018) **56** e23217. DOI: 10.1002/dvg.23217
16. Engle SE, Bansal R, Antonellis PJ, Berbari NF. **Cilia signaling and obesity**. *Semin Cell Dev Biol* (2021) **110** 43-50. DOI: 10.1016/j.semcdb.2020.05.006
17. Hastings MH, Maywood ES, Brancaccio M. **Generation of circadian rhythms in the suprachiasmatic nucleus**. *Nat Rev Neurosci* (2018) **19** 453-469. DOI: 10.1038/s41583-018-0026-z
18. Hsiao YC, Muñoz-Estrada J, Tuz K, Ferland RJ. **The transition zone protein AHI1 regulates neuronal ciliary trafficking of MCHR1 and its downstream signaling pathway**. *J Neurosci* (2021) **41** 3932-3943. DOI: 10.1523/JNEUROSCI.2993-20.2021
19. Hwang SH, Mukhopadhyay S. **G-protein-coupled receptors and localized signaling in the primary cilium during ventral neural tube patterning**. *Birth Defects Res A Clin Mol Teratol* (2015) **103** 12-19. DOI: 10.1002/bdra.23267
20. Jasso KR, Kamba TK, Zimmerman AD, Bansal R, Engle SE, Everett T, Wu CH, Kulaga H, Reed RR, Berbari NF, McIntyre JC. **An N-terminal fusion allele to study melanin concentrating hormone receptor 1**. *Genesis* (2021) **59** e23438. DOI: 10.1002/dvg.23438
21. Kobayashi Y, Tomoshige S, Imakado K, Sekino Y, Koganezawa N, Shirao T, Diniz GB, Miyamoto T, Saito Y. **Ciliary GPCR-based transcriptome as a key regulator of cilia length control**. *FASEB Bioadv* (2021) **3** 744-767. DOI: 10.1096/fba.2021-00029
22. Lee CH, Kang GM, Kim MS. **Mechanisms of weight control by primary cilia**. *Mol Cells* (2022) **45** 169-176. DOI: 10.14348/molcells.2022.2046
23. Loktev AV, Jackson PK. **Neuropeptide Y family receptors traffic via the Bardet-Biedl syndrome pathway to signal in neuronal primary cilia**. *Cell Rep* (2013) **5** 1316-1329. DOI: 10.1016/j.celrep.2013.11.011
24. Messina MM, Boersma G, Overton JM, Eckel LA. **Estradiol decreases the orexigenic effect of melanin-concentrating hormone in ovariectomized rats**. *Physiol Behav* (2006) **88** 523-528. DOI: 10.1016/j.physbeh.2006.05.002
25. Mukhopadhyay S, Lu Y, Shaham S, Sengupta P. **Sensory signaling-dependent remodeling of olfactory cilia architecture in C. elegans**. *Dev Cell* (2008) **14** 762-774. DOI: 10.1016/j.devcel.2008.03.002
26. Mukhopadhyay S, Wen X, Ratti N, Loktev A, Rangell L, Scales SJ, Jackson PK. **The ciliary G-protein-coupled receptor Gpr161 negatively regulates the Sonic hedgehog pathway via cAMP signaling**. *Cell* (2013) **152** 210-223. DOI: 10.1016/j.cell.2012.12.026
27. Nager AR, Goldstein JS, Herranz-Pérez V, Portran D, Ye F, Garcia-Verdugo JM, Nachury MV. **An actin network dispatches ciliary GPCRs into extracellular vesicles to modulate signaling**. *Cell* (2017) **168** 252-263.e14. DOI: 10.1016/j.cell.2016.11.036
28. Olivier-Mason A, Wojtyniak M, Bowie RV, Nechipurenko IV, Blacque OE, Sengupta P. **Transmembrane protein OSTA-1 shapes sensory cilia morphology via regulation of intracellular membrane trafficking in C. elegans**. *Development* (2013) **140** 1560-1572. DOI: 10.1242/dev.086249
29. Pal K, Hwang SH, Somatilaka B, Badgandi H, Jackson PK, DeFea K, Mukhopadhyay S. **Smoothened determines β-arrestin-mediated removal of the G protein-coupled receptor Gpr161 from the primary cilium**. *J Cell Biol* (2016) **212** 861-875. DOI: 10.1083/jcb.201506132
30. Phua SC, Chiba S, Suzuki M, Su E, Roberson EC, Pusapati GV, Schurmans S, Setou M, Rohatgi R, Reiter JF, Ikegami K, Inoue T. **Dynamic remodeling of membrane composition drives cell cycle through primary cilia excision**. *Cell* (2017) **168** 264-279.e15. DOI: 10.1016/j.cell.2016.12.032
31. Pissios P, Frank L, Kennedy AR, Porter DR, Marino FE, Liu FF, Pothos EN, Maratos-Flier E. **Dysregulation of the mesolimbic dopamine system and reward in MCH-/- mice**. *Biol Psychiatry* (2008) **64** 184-191. DOI: 10.1016/j.biopsych.2007.12.011
32. Presse F, Conductier G, Rovere C, Nahon JL. **The melanin-concentrating hormone receptors: neuronal and non-neuronal functions**. *Int J Obes Suppl* (2014) **4** S31-S36. DOI: 10.1038/ijosup.2014.9
33. Reiter JF, Leroux MR. **Genes and molecular pathways underpinning ciliopathies**. *Nat Rev Mol Cell Biol* (2017) **18** 533-547. DOI: 10.1038/nrm.2017.60
34. Saito Y, Hamamoto A, Kobayashi Y. **Regulated control of melanin-concentrating hormone receptor 1 through posttranslational modifications**. *Front Endocrinol (Lausanne)* (2013) **4** 154. PMID: 24155742
35. Santollo J, Eckel LA. **The orexigenic effect of melanin-concentrating hormone (MCH) is influenced by sex and stage of the estrous cycle**. *Physiol Behav* (2008) **93** 842-850. DOI: 10.1016/j.physbeh.2007.11.050
36. Schmidt P, Thomas L, Müller P, Scheidt HA, Huster D. **The G-protein-coupled neuropeptide Y receptor type 2 is highly dynamic in lipid membranes as revealed by solid-state NMR spectroscopy**. *Chemistry* (2014) **20** 4986-4992. DOI: 10.1002/chem.201304928
37. Shinde SR, Nager AR, Nachury MV. **Ubiquitin chains earmark GPCRs for BBSome-mediated removal from cilia**. *J Cell Biol* (2020) **219** e202003020. PMID: 33185668
38. Simon A, Németh J, Jávor A, Komlósi I, Bai P, Oláh J, Juhász B, Kiss R, Szilvássy Z, Czeglédi L. **Feeding state and age dependent changes in melanin-concentrating hormone expression in the hypothalamus of broiler chickens**. *Acta Biochim Pol* (2018) **65** 251-258. DOI: 10.18388/abp.2017_2362
39. Singla V, Reiter JF. **The primary cilium as the cell's antenna: signaling at a sensory organelle**. *Science* (2006) **313** 629-633. DOI: 10.1126/science.1124534
40. Sun JS, Yang DJ, Kinyua AW, Yoon SG, Seong JK, Kim J, Moon SJ, Shin DM, Choi YH, Kim KW. **Ventromedial hypothalamic primary cilia control energy and skeletal homeostasis**. *J Clin Invest* (2021) **131** e138107. PMID: 33021968
41. Tu H-Q. **Rhythmic cilium in SCN neuron is a gatekeeper for the intrinsic circadian clock**. *bioRxiv* (2022). DOI: 10.1101/2022.01.26.477948
42. Vaisse C, Reiter JF, Berbari NF. **Cilia and obesity**. *Cold Spring Harb Perspect Biol* (2017) **9** a028217. DOI: 10.1101/cshperspect.a028217
43. Wang J, Nikonorova IA, Gu A, Sternberg PW, Barr MM. **Release and targeting of polycystin-2-carrying ciliary extracellular vesicles**. *Curr Biol* (2020) **30** R755-R756. DOI: 10.1016/j.cub.2020.05.079
44. Wang J, Nikonorova IA, Silva M, Walsh JD, Tilton PE, Gu A, Akella JS, Barr MM. **Sensory cilia act as a specialized venue for regulated extracellular vesicle biogenesis and signaling**. *Curr Biol* (2021a) **31** 3943-3951.e3. DOI: 10.1016/j.cub.2021.06.040
45. Wang L, Liu Y, Stratigopoulos G, Panigrahi S, Sui L, Zhang Y, Leduc CA, Glover HJ, De Rosa MC, Burnett LC, Williams DJ, Shang L, Goland R, Tsang SH, Wardlaw S, Egli D, Zheng D, Doege CA, Leibel RL. **Bardet-Biedl syndrome proteins regulate intracellular signaling and neuronal function in patient-specific iPSC-derived neurons**. *J Clin Invest* (2021b) **131** e146287. PMID: 33630762
46. Wang Y, Bernard A, Comblain F, Yue X, Paillart C, Zhang S, Reiter JF, Vaisse C. **Melanocortin 4 receptor signals at the neuronal primary cilium to control food intake and body weight**. *J Clin Invest* (2021c) **131** e142064. PMID: 33938449
47. Yamada Y, Post SR, Wang K, Tager HS, Bell GI, Seino S. **Cloning and functional characterization of a family of human and mouse somatostatin receptors expressed in brain, gastrointestinal tract, and kidney**. *Proc Natl Acad Sci U S A* (1992a) **89** 251-255. DOI: 10.1073/pnas.89.1.251
48. Yamada Y, Reisine T, Law SF, Ihara Y, Kubota A, Kagimoto S, Seino M, Seino Y, Bell GI, Seino S. **Somatostatin receptors, an expanding gene family: cloning and functional characterization of human SSTR3, a protein coupled to adenylyl cyclase**. *Mol Endocrinol* (1992b) **6** 2136-2142. DOI: 10.1210/mend.6.12.1337145
49. Yamada Y, Kagimoto S, Kubota A, Yasuda K, Masuda K, Someya Y, Ihara Y, Li Q, Imura H, Seino S. **Cloning, functional expression and pharmacological characterization of a fourth (hSSTR4) and a fifth (hSSTR5) human somatostatin receptor subtype**. *Biochem Biophys Res Commun* (1993) **195** 844-852. DOI: 10.1006/bbrc.1993.2122
50. Yasrebi A, Hsieh A, Mamounis KJ, Krumm EA, Yang JA, Magby J, Hu P, Roepke TA. **Differential gene regulation of GHSR signaling pathway in the arcuate nucleus and NPY neurons by fasting, diet-induced obesity, and 17β-estradiol**. *Mol Cell Endocrinol* (2016) **422** 42-56. DOI: 10.1016/j.mce.2015.11.007
51. Ye F, Breslow DK, Koslover EF, Spakowitz AJ, Nelson WJ, Nachury MV. **Single molecule imaging reveals a major role for diffusion in the exploration of ciliary space by signaling receptors**. *Elife* (2013) **2** e00654. DOI: 10.7554/eLife.00654
52. Ye F, Nager AR, Nachury MV. **BBSome trains remove activated GPCRs from cilia by enabling passage through the transition zone**. *J Cell Biol* (2018) **217** 1847-1868. DOI: 10.1083/jcb.201709041
53. Zamir N, Skofitsch G, Bannon MJ, Jacobowitz DM. **Melanin-concentrating hormone: unique peptide neuronal system in the rat brain and pituitary gland**. *Proc Natl Acad Sci U S A* (1986) **83** 1528-1531. DOI: 10.1073/pnas.83.5.1528
|
---
title: 'Spinal muscular atrophy and anorexia nervosa: a case report'
authors:
- Siu Tsin Au Yeung
- Colleen Alford
- Daniel You
journal: BMC Pediatrics
year: 2023
pmcid: PMC10012429
doi: 10.1186/s12887-023-03915-4
license: CC BY 4.0
---
# Spinal muscular atrophy and anorexia nervosa: a case report
## Abstract
### Background
Spinal muscular atrophy (SMA) is an autosomal recessive condition affecting lower motor neurons causing progressive muscle atrophy. Anorexia nervosa (AN) is a psychiatric disorder characterised by intense fear of weight gain, restriction of energy intake, and preoccupation with body weight and shape. Low weight, gastrointestinal dysmotility, and respiratory infections are common in SMA but may mask AN. No paediatric cases of AN in SMA have been reported to date.
### Case presentation
A 14-year-old female with SMA2 presented with 12 months of declining body weight to a nadir of 24.8 kg (BMI 11). This was initially attributed to medical complications including pneumonia and gastroenteritis, and chronic gut dysmotility associated with SMA. Despite almost 2 years of dietetic input and nutritional supplementation due to the weight plateauing from age 11, no significant restoration or gain was achieved. The Eating Disorder Examination-Questionnaire (EDE-Q) indicated a possible eating disorder and psychiatric evaluation confirmed AN.
Initial management prioritised close medical monitoring and outpatient weight restoration on an oral meal plan. Skin fold anthropometric measurement was conducted to determine a minimum healthy weight. Individual psychological therapy and family sessions were undertaken. The patient developed major depression and a brief relapse with weight loss to 28 kg. Since then, the patient has maintained a weight of around 35 kg with stable mood.
### Conclusions
Low body weight, feeding issues, gastrointestinal dysmotility, and respiratory infections are common in SMA and diagnostic overshadowing can lead to delayed recognition of anorexia nervosa. Change to growth trajectory and prolonged weight loss should prompt consideration of comorbid psychiatric issues. Screening measures such as the EDE-Q and DASS may be helpful in this population. Close liaison between the neurogenetics and psychiatry teams is helpful. Skin fold anthropometry can assist in identifying a minimum healthy weight range.
## Background
Spinal muscular atrophy (SMA) is an autosomal recessive condition affecting lower motor neurons causing progressive muscle atrophy [1]. The incidence is 1 per 10,000 [2]. The disease is typically classified from spinal muscular atrophy type 1 (SMA1) to SMA4 by age of onset of muscle weakness and impaired achievement of motor milestones. SMA1 presents in the early infancy with diffuse proximal muscle weakness and atrophy. SMA4 has milder adult-onset weakness [1]. The cause is genetic abnormality in the survival motor neuron 1 (SMN1) gene and therapies targeting gene expression are now available. As a result, the prognosis of SMA is now more hopeful, particularly for those with earlier onset [1]. Supportive therapy aimed at nutrition and respiratory support and preventing complications of muscle weakness are cornerstones of treatment [1].
Patients with SMA often have low body weight, feeding and swallowing problems [3]. Optimal nutrition support and close monitoring of anthropometry is recommended [4]. The altered body composition, particularly lower lean body mass, in SMA makes interpretation of standard growth charts difficult [5].
Mental health comorbidity of paediatric patients with SMA has not been extensively studied. Evidence indicates that there is a high prevalence of anxiety and depression in school-aged children in China [6]. Qualitative research suggests that patients with SMA report their disease contributes to negative mental health including anxiety and depression [7, 8]. Anorexia nervosa (AN) is a psychiatric disorder characterised by intense fear of weight gain or becoming fat, restriction of energy intake, and preoccupation with and distorted perception of body weight and shape [9].
There have been no published case reports regarding anorexia nervosa or other eating disorders in paediatric SMA. Here we report a rare clinical case of AN in SMA and the complex interactions between both disorders that contributed to delayed diagnosis and impacted management.
## Case presentation
A 14-year-old female with SMA2 was referred for psychiatric review with progressively declining body weight over 12 months. Despite incremental weight gain in the decade prior, the patient’s weight plateaued at approximately 29 kg from age 12 (Basal Metabolic Index (BMI) 18) and fell to a nadir of 24.8 kg (BMI 11, < 3rd centile) at age 14.
Due to the SMA, the patient was wheelchair-bound and dependent on full-time assistance since early childhood. She had multiple complications including bilateral lower limb contractures, scoliosis, osteoporosis, obstructive sleep apnoea requiring respiratory support, and gastroparesis. The patient was able to attend mainstream schooling with some educational support. Her 12-year-old brother also had SMA2, adding to the significant burden on the caregivers. The patient had no previous psychiatric history nor was there any family history of mental health issues.
Initially, the patient’s poor weight gain and subsequent loss was attributed to several medical complications including pneumonia and gastroenteritis, and chronic gut dysmotility associated with SMA. Over a 2-year period, this was managed with dietetic input and with minimal improvement. With increasingly severe weight loss by age 14, the patient also began to demonstrate increasing social withdrawal and decreased interest in usual activities prompting the neurology team to administer the Eating Disorder Examination-Questionnaire (EDE-Q) and the Depression Anxiety Stress Scale (DASS) which indicated possible AN. This triggered a referral to the psychiatry team. There was no previous involvement of mental health teams.
On psychiatric review, the patient was vague and guarded regarding her diet. However, with prompting around the EDE-Q responses, she acknowledged extreme restrictive eating practices, body image preoccupation/distortion, intense fear of weight gain, and calorie counting. A diagnosis of AN and comorbid depression was made. It was formulated that lifelong disability, chronic carer burden with two disabled children, the life-limiting nature of SMA, the restrictions on adolescent individuation, and subsequent loneliness contributed to the development of the patient’s depression. Together with the undiagnosed and untreated depression, chronic feeding and swallowing difficulties associated with SMA made the patient more vulnerable to developing an eating disorder. For the patient, food restriction was seen as a powerful way to exert control and independence, particularly in association with tendencies to internalise emotions, perfectionistic traits, and low self-esteem. The patient was unsurprised by the AN diagnosis and described relief at having it finally recognised because for so long the SMA had been her family’s overriding focus. The parents were extremely surprised by the diagnosis.
Initial management included close medical monitoring and outpatient weight restoration. Following principles of Family Based Treatment, the parents were initially empowered to take responsibility for ensuring the patient undertook increased oral intake. Skin fold anthropometry was used to determine a minimum healthy weight of approximately 30 kg. Body composition and nutritional assessment by anthropometry is used clinically in SMA, as opposed to standard growth charts which are not appropriate in this population. Family sessions initially prioritised helping the family to find their roles in supporting recovery, psychoeducation, externalising the eating disorder, and meal support skills, and later on emotional attunement. Individual psychological therapy focussed on skills in emotional regulation, challenging eating-disordered cognitions and behaviours, and improving insights on the function of AN as a motivation to rely on family as emotional support. The supplementary individual component intended to provide a private therapeutic space for the patient to promote the process of individuation. Such process was hypothesised to have been interrupted by SMA due to the patient’s reliance of caregivers for everyday needs, which then have contributed to the maintenance of AN. The patient developed major depression, likely worsened by the removal of her usual coping strategy of restriction, and was commenced on fluoxetine and olanzapine. Following a 9-month period of weight restoration and stability, there was a relatively brief relapse with weight loss to 28 kg requiring brief inpatient admission and further intensive outpatient care. Since this time, the patient maintained a weight of around 35 kg with stable mood and was discharged by the psychiatric service. Given the multi-dimensional nature of the formulation, it was believed that the combination of family work, individual psychotherapy, and pharmacotherapy were all vital and contributed to the final positive outcome of the patient’s AN and depression.
## Discussion and conclusions
This case report is the first to describe AN or any eating disorder in paediatric SMA. It highlights the interplay of multiple aspects of each disease which have implications for diagnosis, monitoring, and management. As a single case report, this is limited in contributing to the literature, and further case series and research into the interplay of these two conditions is needed.
Low body weight, feeding issues, gastrointestinal dysmotility, and respiratory infections are common in SMA [1] and the resultant diagnostic overshadowing in this case lead to a delayed recognition of AN. Significant change to growth chart trajectory, prolonged weight loss (or failure of gain) despite intensive dietetic input and nutritional supplementation should prompt early consideration of comorbid mental health issues. It is increasingly recognised that psychiatric illness is more common in the SMA population [6–8]. This case highlights the need for increased awareness of AN in SMA patients with depression. Screening measures such as the EDE-Q and DASS may be helpful in this population. When indicated, psychiatric follow-up and intervention should be included as standard of care for SMA patients for the prevention of severe and life-threatening psychiatric disorders such as AN and depression.
Due to the complex medical aspects and complications associated with SMA, close liaison between and joint review by the neurogenetics team and psychiatry team is helpful. The muscular atrophy and low lean muscle mass in SMA mean that standard growth charts are less helpful [10]. Instead, skin fold anthropometry may be helpful to identify a minimum healthy weight range and guide refeeding. As occurred in this case, families may be surprised by the diagnosis of AN and instead attribute weight loss to SMA-related complications. The joint medical-psychiatric approach is helpful in reinforcing the AN diagnosis and emphasising the urgency of refeeding. Furthermore, involvement by the specialist eating disorder team at the tertiary children’s hospital reinforced this and helped to reinvigorate their efforts.
The multitude of supportive cares and complications associated with SMA mandate a high carer burden and often leads to burnout [8]. As such, it was at times hard for the family to externalise the AN and they could fall into inadvertently blaming the patient for “keeping” AN. The typical, structured Family Based Treatment approach used for paediatric AN needed to be tailored and made flexible to accommodate the competing priorities of SMA care, such as daily physiotherapy. For example, the patient’s brother (who also had SMA) and one parent caring for him were usually unable to attend weekly family sessions. An important theme that emerged was pre-existing guilt held by the parents for their role in the genetic inheritance of the patient’s SMA. This was compounded further by the guilt of the patient now having another severe diagnosis of AN.
The urgency of treatment for AN is often framed by therapists as life-threatening in order to spur the family into action [11]. However, in the context of a life-limiting medical condition such as SMA, this framing had to be tempered somewhat as both conditions had their own competing medical priorities. The therapists supported the parents to shift some focus from the pragmatics of medical treatment, to also attending to the patient’s emotional needs. Furthermore, the in-depth involvement of the family in therapy and focus on promoting communication and emotional attunement within the family system was greatly appreciated by both the patient and parents.
The chronic disability and dependence associated with SMA has a significant impact on AN treatment in teenagers. The patient was unable to socialise typically or undertake the same usual teenager activities as her friends, particularly in relation to part time employment and learning to drive. This resulted in the patient feeling increasingly lonely and turning more to AN. Phase 3 of Family Based Treatment for AN usually involves an emphasis on adolescent individuation and independence [12]. This was harder to foster in the patient due to the impairments imposed by the SMA. However, gains in this domain were made with referral to an adolescent chronic illness support group and family encouragement for the patient to pursue interests beyond her physical disability, such as career-related school subjects.
Two complex and severe conditions interfaced in this patient, and it was critical that both sets of challenges were recognised and addressed in adaptable ways.
## References
1. Nance JR. **Spinal muscular atrophy**. *Continuum (Minneap Minn)* (2020.0) **26** 1348-1368. DOI: 10.1212/CON.0000000000000918
2. Sugarman EA, Nagan N, Zhu H, Akmaev VR, Zhou Z, Rohlfs EM. **Pan-ethnic carrier screening and prenatal diagnosis for spinal muscular atrophy: clinical laboratory analysis of >72,400 specimens**. *Eur J Hum Genet* (2012.0) **20** 27-32. DOI: 10.1038/ejhg.2011.134
3. Martinez EE, Quinn N, Arouchon K, Anzaldi R, Tarrant S, Ma NS. **Comprehensive nutritional and metabolic assessment in patients with spinal muscular atrophy: opportunity for an individualized approach**. *Neuromuscul Disord* (2018.0) **28** 512-519. DOI: 10.1016/j.nmd.2018.03.009
4. Wang CH, Finkel RS, Bertini ES, Schroth M, Simonds A, Wong B. **Consensus statement for standard of care in spinal muscular atrophy**. *J Child Neurol* (2007.0) **22** 1027-1049. DOI: 10.1177/0883073807305788
5. Bertoli S, De Amicis R, Mastella C, Pieri G, Giaquinto E, Battezzati A. **Spinal Muscular Atrophy, types I and II: what are the differences in body composition and resting energy expenditure?**. *Clin Nutr* (2017.0) **36** 1674-1680. DOI: 10.1016/j.clnu.2016.10.020
6. Yao M, Xia Y, Feng Y, Ma Y, Hong Y, Zhang Y. **Anxiety and depression in school-age patients with spinal muscular atrophy: a cross-sectional study**. *Orphanet J Rare Dis* (2021.0) **16** 385. DOI: 10.1186/s13023-021-02008-8
7. Mazzella A, Curry M, Belter L, Cruz R, Jarecki J. **“I have SMA, SMA doesn’t have me”: a qualitative snapshot into the challenges, successes, and quality of life of adolescents and young adults with SMA**. *Orphanet J Rare Dis* (2021.0) **16** 96. DOI: 10.1186/s13023-021-01701-y
8. Wan HWY, Carey KA, D’Silva A, Vucic S, Kiernan MC, Kasparian NA. **Health, wellbeing and lived experiences of adults with SMA: a scoping systematic review**. *Orphanet J Rare Dis* (2020.0) **15** 70. DOI: 10.1186/s13023-020-1339-3
9. 9.American Psychiatric Association. Diagnostic and statistical manual of mental disorders. 5th ed. 2013. doi:10.1176/appi.books.9780890425596.
10. Ferrantini G, Coratti G, Onesimo R, Lucibello S, Bompard S, Turrini I. **Body mass index in type 2 spinal muscular atrophy: a longitudinal study**. *Eur J Pediatr* (2022.0) **181** 1923-1932. DOI: 10.1007/s00431-021-04325-3
11. Rienecke RD, Le Grange D. **The five tenets of family-based treatment for adolescent eating disorders**. *J Eat Disord* (2022.0) **10** 60. DOI: 10.1186/s40337-022-00585-y
12. Rienecke RD. **Family-based treatment of eating disorders in adolescents: current insights**. *Adolesc Health Med Ther* (2017.0) **8** 69-79. DOI: 10.2147/AHMT.S115775
|
---
title: 'Lipid metabolism, BMI and the risk of nonalcoholic fatty liver disease in
the general population: evidence from a mediation analysis'
authors:
- Song Lu
- Qiyang Xie
- Maobin Kuang
- Chong Hu
- Xinghui Li
- Huijian Yang
- Guotai Sheng
- Guobo Xie
- Yang Zou
journal: Journal of Translational Medicine
year: 2023
pmcid: PMC10012451
doi: 10.1186/s12967-023-04047-0
license: CC BY 4.0
---
# Lipid metabolism, BMI and the risk of nonalcoholic fatty liver disease in the general population: evidence from a mediation analysis
## Abstract
### Background
Body mass index (BMI) and lipid parameters are the most commonly used anthropometric parameters and biomarkers for assessing nonalcoholic fatty liver disease (NAFLD) risk. This study aimed to assess and quantify the mediating role of traditional and non-traditional lipid parameters on the association between BMI and NAFLD.
### Method
Using data from 14,251 subjects from the NAGALA (NAfld in the Gifu Area, Longitudinal Analysis) study, mediation analyses were performed to explore the roles of traditional [total cholesterol (TC), triglyceride (TG), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C)] and non-traditional [non-HDL-C, remnant cholesterol (RC), TC/HDL-C ratio, LDL-C/HDL-C ratio, TG/HDL-C ratio, non-HDL-C/HDL-C ratio, and RC/HDL-C ratio] lipid parameters in the association of BMI with NAFLD and quantify the mediation effect of these lipid parameters on the association of BMI with NAFLD using the percentage of mediation.
### Result
After fully adjusting for confounders, multivariate regression analysis showed that both BMI and lipid parameters were associated with NAFLD (All P-value < 0.001). Mediation analysis showed that both traditional and non-traditional lipid parameters mediated the association between BMI and NAFLD (All P-value of proportion mediate < 0.001), among which non-traditional lipid parameters such as RC, RC/HDL-C ratio, non-HDL-C/HDL-C ratio, and TC/HDL-C ratio accounted for a relatively large proportion, $11.4\%$, $10.8\%$, $10.2\%$, and $10.2\%$, respectively. Further stratified analysis according to sex, age, and BMI showed that this mediation effect only existed in normal-weight (18.5 kg/m2 ≤ BMI < 25 kg/m2) people and young and middle-aged (30–59 years old) people; moreover, the mediation effects of all lipid parameters except TC accounted for a higher proportion in women than in men.
### Conclusion
The new findings of this study showed that all lipid parameters were involved in and mediated the risk of BMI-related NAFLD, and the contribution of non-traditional lipid parameters to the mediation effect of this association was higher than that of traditional lipid parameters, especially RC, RC/HDL-C ratio, non-HDL-C/HDL-C ratio, and TC/HDL-C ratio. Based on these results, we suggest that we should focus on monitoring non-traditional lipid parameters, especially RC and RC/HDL-C ratio, when BMI intervention is needed in the process of preventing or treating NAFLD.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12967-023-04047-0.
## Background
With the rapid economic growth, the changes in lifestyle, and the prevalence of obesity, the prevalence of NAFLD is increasing all over the world. It is estimated that about $\frac{1}{4}$ of adults in the world have suffered from NAFLD [1–3], and in about ten years, the global prevalence of the disease is expected to further rise to $\frac{1}{3}$ [4], which will bring huge public health problems and economic burden to society [5]. Therefore, it is necessary to carry out early prevention and risk factor screening for susceptible people for NAFLD.
High BMI and dyslipidemia are known to be the most important modifiable risk factors for the onset of NAFLD. A series of studies have elucidated the associations between BMI and dyslipidemia and NAFLD in the past, among which elevated body weight and dyslipidemia were the most important risk factors for the increased risk of NAFLD [6–14]. In addition, it is worth noting that with the rapid development of global industrialization and economy, many chemicals are released into the environment, and long-term to exposure of these pollutants will cause harm to human health, further leading to lipid metabolism disorders and weight gain [15, 16], which in turn further aggravates the disease burden of NAFLD. Toxicology-based evidence also supported this notion, with the most pronounced effect of increased chemical exposure on liver cells as a redox imbalance, often manifested in mitochondrial damage, enhancement of lipid peroxidation, elevation in reactive oxygen species formation, and depletion of intracellular reduced glutathione [17–21]. In the context of today's rapid global development, it is particularly important to be able to assess NAFLD risks more efficiently and accurately. Considering that the measurement of BMI is simple, while the inspection of lipid parameters is relatively complex, it may be a useful combination to use BMI as an indicator of daily NAFLD risk assessment and lipid parameters as indicators of regular NAFLD risk assessment; further clarification is needed which lipid parameters work best in combination with BMI for routine risk assessment of NAFLD. To address this issue, in the current study, we tried to explore the influence of all non-traditional and traditional lipid parameters on the association between BMI and NAFLD through the meditation analysis system.
## Data sources and study population
To elucidate the effect of lipid parameters on the association of BMI with NAFLD, we performed a secondary analysis using the NAGALA study dataset, which has been uploaded to the Dryad database for public share by the Okamura team (https://doi.org/10.5061/dryad.8q0p192). The NAGALA study, a population-based longitudinal cohort study, aimed at assessing risk factors for common chronic diseases, thereby reducing the burden on the public health system and promoting population health.
Detailed information about the NAGALA cohort study has been described in a previously published article [22]. Based on a new study objective, this study extracted physical examination data from 20,944 subjects in the NAGALA dataset between May 1994 and December 2016. According to the inclusion and exclusion criteria, we excluded subjects who were on medication at baseline, subjects with missing covariate data or excessive drinking or unexplained withdrawal from the study, and subjects with diabetes or impaired fasting glucose [fasting blood glucose (FPG) above 6.1 mmol/L] or liver disease (except fatty liver) at baseline. A total of 14,251 subjects were included in the final analysis of this study (Fig. 1).Fig. 1Flowchart of the selection process of study subjects
## Ethics approval
The previous study was approved by the Murakami Memorial Hospital Ethics Committee [22], and written informed consent was obtained from each subject. The current study is a secondary analysis of the previous study, and the study protocol was approved by the Institutional Ethics Review Committee of Jiangxi Provincial People's Hospital, and the whole study process followed the Helsinki Declaration.
## Data collection and measurement
As mentioned earlier, standardized questionnaires were used to obtain subjects' age and sex, smoking and drinking status, the habit of exercise, disease history, and drug use history. Body weight, height, waist circumference (WC), and arterial blood pressure were professionally measured and recorded by professional medical workers. Where BMI was calculated as weight (kg)/[height(m)]2. A habit of exercise was defined as subjects participating in any form of physical activity more than one time per week. Drinking status was defined according to the amount of past weekly alcohol consumption of the subjects, and the subjects were classified as no or little drinkers, light drinkers, and moderate drinkers [23]. Smoking status was divided into three categories: none, former, and current. Blood specimens were collected from each subject after fasting for at least 8 h, and the following laboratory biochemical parameters, including gamma-glutamyl transferase (GGT), aspartate aminotransferase (AST), glycosylated hemoglobin (HbA1c), FPG and alanine aminotransferase (ALT), HDL-C, TC, and TG were analyzed and measured by an automated analyzer according to the standard method.
LDL-C and non-traditional lipid parameters are calculated as follows: LDL-C (mg/dL) = $90\%$non-HDL-C – $10\%$TG [24]; Non-HDL-C = TC – HDL-C [25]; RC = non-HDL-C – LDL-C [26];
TC/HDL-C ratio = TC/HDL-C [27]; TG/HDL-C ratio = TG/HDL-C [28]; LDL-C/HDL-C ratio = LDL-C/HDL-C [29]; Non-HDL-C/HDL-C ratio = Non-HDL-C/HDL-C [30];
RC/HDL-C ratio = RC/HDL-C [31];
## Diagnosis of NAFLD
Trained professional technicians performed abdominal ultrasound examinations on the subjects. The ultrasound images were evaluated by experienced gastroenterologists without knowing the subjects' personal health data, and a final diagnosis was made based on a combination of four ultrasound sonogram features, including vascular blurring, liver and kidney echo contrast, depth attenuation and liver brightness [32].
## Statistical analysis
Descriptive data for baseline characteristics were expressed as mean (standard deviation) or median (interquartile range) for continuous variables and frequency (percentage) for categorical variables, and whether there were differences between groups were compared by chi-square test or Mann–Whitney U test or t test. In addition, we also calculated standardized difference values between the NAFLD group and non-NAFLD group to assess and quantify the magnitude of differences between groups (standardized difference values > $10\%$ were considered significant), where skewed continuous variables were subjected to Box-COX normal transformation before data analysis [33, 34].
Before performing regression analysis and mediation analysis, we first evaluated the collinearity of lipid parameters/BMI and other covariates by multiple linear regression, and calculated the variance inflation factor as an evaluation tool, in which covariates with variance inflation factor greater than 10 were considered as collinear variable [35]. Collinearity screening results showed no high collinearity between all lipid parameters and baseline variables, but high collinearity between BMI and body weight (Additional file 1: Tables S1–S12). Based on these results of collinearity screening, we will exclude body weight in the subsequent regression analysis model and mediation analysis model, and further adjust the model according to epidemiological research evidence, and show the process of gradual adjustment in the main analysis [36].
We performed a mediation analysis according to the method suggested by Professor VanderWeele [37] to examine how lipid parameters as mediator variables affect the relationship between BMI (independent variable) and NAFLD (outcome variable). According to the requirements of the mediation analysis, the research data in this study must meet the following criteria: [1] BMI/lipid parameters must be associated with NAFLD: *As a* part of the study, we used multivariate logistic regression model to analyze the relationship between BMI/ lipid parameters and NAFLD, and considered the potential effects of sex, age, WC, habit of exercise, smoking status, drinking status, liver enzyme related factors (ALT, AST, GGT), blood pressure-related factors [systolic blood pressure (SBP), diastolic blood pressure (DBP)] and blood glucose metabolism factors (FPG, HbA1c) [1, 2, 38–40]; [2] BMI must be associated with lipid parameters: *In this* validation analysis, we assessed the association by multivariate linear regression in which we considered the potential effects of sex, age, WC, habit of exercise, smoking status, drinking status, blood pressure-related factors (SBP, DBP), and glucose metabolism factors (FPG, HbA1c) [2, 8, 40, 41]; [3] The association between BMI and NAFLD must to be attenuated when lipid parameters were included in the multivariate logistic regression models. If the above conditions are met then a mediation analysis can be performed to determine whether the effect of BMI on NAFLD is mediated by lipid parameters. In addition, the mediation percentage was obtained by calculating the ratio of the indirect effect to the total effect, thereby quantifying the size of the mediation effect. The significance of the mediation effect was tested using Bootstrap sampling (times = 1000) [42]. To reduce the effect of confounding factors, we adjusted for the covariates sex, age, WC, DBP, ALT, AST, GGT, FPG, SBP, HbA1c, the habit of exercise, smoking status, and drinking status in the mediation analysis [1, 2, 38–40]. Finally, we performed further exploratory stratified analyses for sex, age, and BMI, in which the age classification refers to the World Health Organization's classification standards (young people within 45 years old, middle-aged people between 45 and 59 years old, and elderly people over 60 years old), and the BMI classification refers to the World Health Organization's recommended BMI classification standards for Asian people [low weight (BMI < 18.5 kg/m2), normal weight (18.5 kg/m2 ≤ BMI < 25 kg/m2), and overweight/obese (BMI ≥ 25 kg/m2)] [43]. All analyses were done using R language version 3.4.3 and Empower(R) version 4. 0. All tests were two-tailed and statistical significance was set at $P \leq 0.05.$
## Characteristics of the study subjects
This study included 14,251 subjects with a mean age of 43.7 years, $48\%$ of whom were women, and $17.6\%$ of subjects were diagnosed with NAFLD. Subjects were grouped according to whether or not suffering from NAFLD, and Table 1 presents the differences in baseline characteristics of the study population. Except for drinking status, almost all baseline demographic data, anthropometric data, and serum biomarkers were significantly different between the NAFLD group and the non-NAFLD group (all $P \leq 0.05$). It is noteworthy that the BMI of the NAFLD group was significantly higher than that of the non-NAFLD group (25.5 vs 21.3), and the standardized difference value between the two groups was as high as $145\%$. In terms of lipid parameters, the standardized difference values between the two groups were generally higher for non-traditional lipid parameters than for traditional lipid parameters. Table 1Characteristics of the study subjects with and without NAFLDNon-NAFLDNAFLDStandardized difference (%)P-valueNo. of subjects11,7442507Sex78 [74, 83]< 0.001 Women6362 ($54.17\%$)478 ($19.07\%$)< 0.001 Men5382 ($45.83\%$)2029 ($80.93\%$)< 0.001Age, years42.00 (18.00–79.00)44.00 (19.00–72.00)18 [13, 22]< 0.001Weight, kg57.72 (9.98)72.18 (11.33)135 [131, 140]< 0.001Height, cm164.11 (8.44)168.03 (7.90)48 [44, 52]< 0.001BMI, kg/m221.33 (2.61)25.50 (3.13)145 [140, 149]< 0.001WC, cm74.09 (7.92)85.98 (7.79)151 [147, 156]< 0.001ALT, U/L15.00 (2.00–856.00)27.00 (6.00–220.00)96 [91, 100]< 0.001AST, U/L17.00 (3.00–590.00)20.00 (6.00–140.00)56 [51, 60]< 0.001GGT, U/L14.00 (3.00–259.00)23.00 (6.00–375.00)61 [57, 66]< 0.001TC, mmol/L5.06 (0.85)5.44 (0.87)45 [41, 49]< 0.001TG, mmol/L0.65 (0.07–10.27)1.24 (0.16–7.69)96 [91, 100]< 0.001HDL-C, mmol/L1.52 (0.40)1.19 (0.29)96 [92, 101]< 0.001LDL-C. mmol/L2.95 (0.95–9.72)3.49 (1.16–6.59)68 [64, 73]< 0.001Non-HDL-C, mmol/L3.47 (1.22–10.93)4.25 (1.38–7.71)83 [79, 88]< 0.001RC, mmol/L0.50 (0.17–2.72)0.71 (0.18–2.37)107 [102, 111]< 0.001TC/HDL-C ratio3.33 (1.51–13.02)4.71 (1.59–10.75)111 [106, 115]< 0.001TG/HDL-C ratio0.43 (0.03–16.55)1.07 (0.12–11.67)92 [87, 96]< 0.001LDL-C/HDL-C ratio1.99 (0.43–10.35)3.04 (0.50–8.01)106 [101, 110]< 0.001Non-HDL-C/HDL-C ratio2.33 (0.51–12.02)3.71 (0.59–9.75)111 [106, 115]< 0.001RC/HDL-C0.33 (0.07–4.39)0.62 (0.09–3.57)107 [103, 112]< 0.001FPG, mmol/L5.09 (0.40)5.39 (0.36)78 [74, 82]< 0.001HbA1c, %5.15 (0.31)5.30 (0.33)46 [42, 51]< 0.001SBP, mmHg111.91 (14.02)123.41 (14.83)80 [75, 84]< 0.001DBP, mmHg69.69 (9.85)77.81 (10.19)81 [77, 85]< 0.001Habit of exercise2093 ($17.82\%$)377 ($15.04\%$)8 [3, 12]< 0.001Drinking status4 (-0.01, 9)0.162 No or little9717 ($82.74\%$)2088 ($83.29\%$) Light1472 ($12.53\%$)286 ($11.41\%$) Moderate555 ($4.73\%$)133 ($5.31\%$)Smoking status35 [31, 39]< 0.001 None7561 ($64.38\%$)1185 ($47.27\%$) Former1920 ($16.35\%$)639 ($25.49\%$) Current2263 ($19.27\%$)683 ($27.24\%$)Values were expressed as mean (SD) or medians (quartile interval) or n (%). The differences between groups were compared by chi-square test or Mann–Whitney U test or t test (P-value < 0.05 indicates significance). Standardized difference values were used to evaluate and quantify the difference between groups (> $10\%$ were considered significant)NAFLD: Nonalcoholic fatty liver disease; BMI: body mass index; WC: Waist circumference; ALT: alanine aminotransferase; AST: aspartate aminotransferase; GGT: gamma-glutamyl transferase; HDL-C: high-density lipoprotein cholesterol; TC: total cholesterol; TG: triglyceride; LDL-C: low density lipoprotein cholesterol; Non-HDL-C: non-high-density lipoprotein cholesterol; RC: remnant cholesterol; HbA1c: hemoglobin A1c; FPG: fasting plasma glucose; SBP: systolic blood pressure; DBP: diastolic blood pressure
## Relationship between BMI/lipid parameters and NAFLD
After adjusting for potential confounders (Tables 2 and 3), there were significant associations between BMI or lipid parameters and NAFLD (all $P \leq 0.001$). The association between BMI and NAFLD was significantly positive in models 1–3 with Odds ratios (ORs) of 1.37 [$95\%$ confidence interval (CI)1.32–1.42, $P \leq 0.001$], 1.33 (1.29–1.38, $P \leq 0.001$), and 1.27 (1.22–1.31, $P \leq 0.001$), respectively; after further adjusting the lipid parameters (mediator variables), we found that the correlation between BMI and NAFLD weakened (Table 3, models 4–14), suggesting a mediation effect. In addition, with regard to lipid parameters, all lipid parameters were significantly positively associated with NAFLD except HDL-C. It is worth mentioning that the RC and the RC/HDL-C ratio were much riskier for NAFLD than other lipid parameters (RC: OR = 9.18, $95\%$ CI 6.88–12.26, $P \leq 0.001$; RC/HDL-C ratio: OR = 5.01, $95\%$ CI 4.02–6.24, $P \leq 0.001$).Table 2Relationship between lipid parameters and NAFLDOR ($95\%$ CI)Model 1Model 2Model 3BMI1.37 (1.32, 1.42)*1.33 (1.29, 1.38)*1.27 (1.22, 1.31)*TC1.41 (1.33, 1.50)*1.37 (1.29, 1.46)*1.17 (1.09, 1.25)*LDL-C1.61 (1.50, 1.73)*1.54 (1.43, 1.66)*1.29 (1.19, 1.39)*TG2.27 (2.09, 2.47)*2.25 (2.07, 2.45)*1.93 (1.77, 2.11)*HDL-C0.25 (0.21, 0.31)*0.25 (0.21, 0.31)*0.30 (0.24, 0.36)*Non-HDL-C1.67 (1.57, 1.77)*1.61 (1.51, 1.72)*1.36 (1.27, 1.46)*RC18.53(14.17, 24.23)*17.16(13.07, 22.54)*9.18 (6.88, 12.26)*TC/HDL-C ratio1.57 (1.50, 1.65)*1.56 (1.49, 1.64)*1.40 (1.33, 1.47)*TG/HDL-C ratio1.94 (1.80, 2.08)*1.92 (1.79, 2.07)*1.68 (1.56, 1.82)*LDL-C/HDL-C ratio1.68 (1.58, 1.78)*1.65 (1.56, 1.76)*1.45 (1.36, 1.54)*Non-HDL-C/HDL-C ratio1.57 (1.50, 1.65*1.56 (1.49, 1.64)*1.40 (1.33, 1.47)*RC/HDL-C ratio7.94 (6.47, 9.74)*7.79 (6.31, 9.62)*5.01 (4.02, 6.24)*OR: Odds ratios; CI: confidence interval; other abbreviations as in Table 1*$P \leq 0.001$Model 1 adjusted sex, age and WCModel 2 adjusted model 1 + SBP, DBP, habit of exercise, smoking status and drinking statusModel 3 adjusted model 2 + ALT, AST, GGT, FPG and HbA1cTable 3Relationship between BMI and NAFLDOR ($95\%$ CI)P-valueModel 11.37 (1.32, 1.42)< 0.001Model 21.33 (1.29, 1.38)< 0.001Model 31.27 (1.22, 1.31)< 0.001Model 41.26 (1.22, 1.31)< 0.001Model 51.24 (1.20, 1.29)< 0.001Model 61.25 (1.21, 1.30)< 0.001Model 71.26 (1.21, 1.31)< 0.001Model 81.25 (1.21, 1.30)< 0.001Model 91.25 (1.20, 1.30)< 0.001Model 101.24 (1.20, 1.29)< 0.001Model 111.25 (1.21, 1.30)< 0.001Model 121.25 (1.20, 1.29)< 0.001Model 131.24 (1.20, 1.29)< 0.001Model 141.25 (1.20, 1.29)< 0.001OR: Odds ratios; CI: confidence interval; other abbreviations as in Table 1Model 1 adjusted sex, age and WCModel 2 adjusted model I + SBP, DBP, habit of exercise, smoking status and drinking statusModel 3 adjusted model II + ALT, AST, GGT, FPG and HbA1cModel 4 adjusted model II + TC; Model 5 adjusted model II + HDL-C; Model 6 adjusted model II + TG; Model 7 adjusted model II + LDL-C; Model 8 adjusted model II + non-HDL-C; Model 9 adjusted model II + RC; Model 10 adjusted model II + TC/HDL-C ratio; Model 11 adjusted model II + TG/HDL-C ratio; Model 12 adjusted model II + LDL-C/HDL-C ratio; Model 13 adjusted model II + non-HDL-C/HDL-C ratio; Model 14 adjusted model II + RC/HDL-C ratioModels 4–14 show the correlation between BMI and NAFLD when lipid parameters are included in the regression model
## Relationship between BMI and lipid parameters
The results of multivariate linear regression models showed that all lipid parameters were significantly associated with BMI after fully adjusting for confounders (Additional file 1: Table S13); furthermore, it is worth mentioning that among all lipid parameters, RC and RC/HDL-C ratio were most positively associated with BMI (RC: β = 1.03, $95\%$ CI 0.88–1.17, $P \leq 0.001$; RC/HDL-C ratio: β = 0.80, $95\%$ CI 0.69–0.92, $P \leq 0.001$), while HDL-C was negatively associated with BMI.
## Mediation effect of lipid parameters on the association between BMI and NAFLD
Table 4 shows the mediation analysis results of lipid parameters on the relationship between BMI and NAFLD. In the general population, we found that all 11 lipid parameters partially mediated the association between BMI and NAFLD risk, with RC/HDL-C ratio, RC, non-HDL-C/HDL-C ratio, and TC/HDL-C ratio mediating the largest mediation effects at $11.4\%$, $10.8\%$, $10.2\%$, and $10.2\%$, respectively (Fig. 2); while lipid parameters such as TC, HDL-C, LDL-C, and non-HDL-C played a weak role in mediating the association between BMI and NAFLD risk. Overall, non-traditional lipid parameters mediated the association between BMI and NAFLD more than traditional lipid parameters. Table 4Mediation analysis for BMI and NAFLD via lipid parameters in the whole populationMediatorTotal effectMediation effectDirect effectPM (%)P-value of PMTC0.110 (0.097, 0.123)0.002 (0.001, 0.002)0.109 (0.095, 0.122)1.3< 0.001TG0.110 (0.097, 0.123)0.009 (0.007, 0.012)0.101 (0.088, 0.114)8.3< 0.001HDL-C0.110 (0.097, 0.123)0.006 (0.004, 0.008)0.105 (0.091, 0.117)5.4< 0.001LDL-C0.110 (0.097, 0.123)0.003 (0.002, 0.004)0.107 (0.094, 0.120)2.6< 0.001Non-HDL-C0.110 (0.097, 0.123)0.005 (0.004, 0.007)0.105 (0.091, 0.118)4.9< 0.001RC0.110 (0.097, 0.123)0.012 (0.010, 0.014)0.098 (0.085, 0.111)10.8< 0.001TC/HDL-C ratio0.110 (0.097, 0.123)0.011 (0.009, 0.014)0.099 (0.086, 0.112)10.2< 0.001TG/HDL-C ratio0.110 (0.097, 0.123)0.009 (0.007, 0.011)0.101 (0.088, 0.114)8.2< 0.001LDL/HDL-C ratio0.110 (0.097, 0.123)0.010 (0.008, 0.012)0.101 (0.087, 0.114)8.7< 0.001Non-HDL-C/HDL-C ratio0.110 (0.097, 0.123)0.011 (0.009, 0.014)0.099 (0.086, 0.112)10.2< 0.001RC/HDL-C ratio0.110 (0.097, 0.123)0.013 (0.010, 0.015)0.098 (0.085, 0.111)11.4< 0.001PM: propotion mediate; other abbreviations as in Table 1Adjusting variables: sex, age, WC, SBP, DBP, ALT, AST, GGT, FPG, HbA1c, habit of exercise, smoking status and drinking statusFig. 2Lipid parameters mediation models of the relationship between BMI and NAFLD. ME: Mediation effect; DE: Direct effect; BMI: Body mass index; NAFLD: Nonalcoholic fatty liver disease
## Stratified analysis
We further stratified all subjects according to BMI, sex, and age, and explored the mediating role of lipid parameters in the association between BMI and NAFLD in different populations.
Table 5 shows the mediation effect of lipid parameters on the association between BMI and NAFLD in different BMI groups. The results showed that the mediation effect of lipid parameters on the association between BMI and NAFLD was only observed in the normal-weight group, while no significant mediation effect was found in the low-weight group and the overweight/obese group. *In* general, in people with normal BMI, the mediation effect of non-traditional lipid parameters was greater than that of traditional lipid parameters, among which the meditation percentages of non-traditional lipid parameters such as RC/HDL-C ratio, RC, TC/HDL-C ratio, non-HDL-C/HDL-C ratio, LDL-C/HDL-C ratio, TG/HDL-C ratio were $17.2\%$, $16.2\%$, $15.9\%$, $15.9\%$, $13.9\%$, and $12.2\%$, respectively. Table 5Mediation analysis for BMI and NAFLD via lipid parameters in the whole population stratified by BMIMediatorTotal effectMediation effectDirect effectPM (%)P-value of PMBMI < 18.5 kg/m2 ($$n = 1545$$) TC− 0.000 (− 0.004, 0.003)0.000 (− 0.000, 0.000)− 0.001 (− 0.004, 0.003)–0.912 TG− 0.000 (− 0.004, 0.003)0.000 (− 0.000, 0.000)− 0.001 (− 0.004, 0.003)–0.878 HDL-C− 0.000 (− 0.004, 0.003)− 0.000 (− 0.000, 0.000)− 0.000 (− 0.004, 0.003)–0.910 LDL-C− 0.000 (− 0.004, 0.003)0.000 (− 0.000, 0.000)− 0.001 (− 0.004, 0.003)–0.878 Non-HDL-C− 0.000 (− 0.004, 0.003)0.000 (− 0.000, 0.000)− 0.001 (− 0.004, 0.003)–0.858 RC− 0.000 (− 0.004, 0.003)0.000 (− 0.000, 0.001)− 0.001 (− 0.004, 0.003)–0.854 TC/HDL-C ratio− 0.000 (− 0.004, 0.003)0.000 (− 0.000, 0.000)− 0.000 (− 0.004, 0.003)–0.974 TG/HDL-C ratio− 0.000 (− 0.004, 0.003)0.000 (− 0.000, 0.000)− 0.000 (− 0.004, 0.003)–0.970 LDL/HDL-C ratio− 0.000 (− 0.004, 0.003)0.000 (− 0.000, 0.000)− 0.000 (− 0.004, 0.003)–0.966 Non-HDL-C/HDL-C ratio− 0.000 (− 0.004, 0.003)0.000 (− 0.000, 0.000)− 0.000 (− 0.004, 0.003)–0.974 RC/HDL-C ratio− 0.000 (− 0.004, 0.003)0.000 (− 0.000, 0.000)− 0.000 (− 0.004, 0.003)–0.972BMI: 18.5–25 kg/m2 ($$n = 10$$,442) TC0.062 (0.048, 0.076)0.001 (0.001, 0.002)0.060 (0.046, 0.074)2.4< 0.001 TG0.062 (0.048, 0.076)0.007 (0.005, 0.010)0.054 (0.040, 0.068)12.1< 0.001 HDL-C0.062 (0.048, 0.076)0.005 (0.004, 0.007)0.057 (0.043, 0.071)8.3< 0.001 LDL-C0.062 (0.048, 0.076)0.003 (0.002, 0.005)0.059 (0.045, 0.073)5.1< 0.001 Non-HDL-C0.062 (0.048, 0.076)0.005 (0.004, 0.007)0.057 (0.043, 0.071)8.3< 0.001 RC0.062 (0.048, 0.076)0.010 (0.008, 0.012)0.052 (0.038, 0.066)16.2< 0.001 TC/HDL-C ratio0.062 (0.048, 0.076)0.010 (0.008, 0.012)0.052 (0.038, 0.066)15.9< 0.001 TG/HDL-C ratio0.062 (0.048, 0.076)0.008 (0.005, 0.010)0.054 (0.040, 0.068)12.2< 0.001 LDL/HDL-C ratio0.062 (0.048, 0.076)0.009 (0.007, 0.011)0.053 (0.039, 0.067)13.9< 0.001 Non-HDL-C/HDL-C ratio0.062 (0.048, 0.076)0.010 (0.008, 0.012)0.052 (0.038, 0.066)15.9< 0.001 RC/HDL-C ratio0.062 (0.048, 0.076)0.011 (0.008, 0.013)0.051 (0.037, 0.065)17.2< 0.001BMI: ≥ 25 kg/m2 ($$n = 2264$$) TC0.078 (0.045, 0.113)0.000 (− 0.001, 0.002)0.077 (0.045, 0.113)–0.826 TG0.077 (0.044, 0.113)− 0.002 (− 0.009, 0.004)0.080 (0.047, 0.113)–0.478 HDL-C0.077 (0.045, 0.111)0.002 (− 0.003, 0.008)0.075 (0.043, 0.110)–0.410 LDL-C0.077 (0.044, 0.113)0.000 (− 0.001, 0.002)0.077 (0.044, 0.112)–0.558 Non-HDL-C0.077 (0.044, 0.112)0.001 (− 0.002, 0.004)0.077 (0.044, 0.112)–0.666 RC0.077 (0.044, 0.113)− 0.001(− 0.007, 0.005)0.079 (0.046, 0.112)–0.712 TC/HDL-C ratio0.077 (0.043, 0.112)0.001 (− 0.003, 0.007)0.075 (0.043, 0.109)–0.556 TG/HDL-C ratio0.077 (0.043, 0.112)− 0.002 (− 0.008, 0.005)0.078 (0.046, 0.112)–0.676 LDL/HDL-C ratio0.077 (0.043, 0.111)0.002 (− 0.002, 0.006)0.075 (0.042, 0.109)–0.442 Non-HDL-C/HDL-C ratio0.077 (0.043, 0.112)0.001 (− 0.003, 0.007)0.075 (0.043, 0.109)–0.556 RC/HDL-C ratio0.077 (0.043, 0.112)− 0.000 (− 0.006, 0.006)0.077 (0.045, 0.111)–0.930PM: propotion mediate; other abbreviations as in Table 1Adjusting variables: sex, age, WC, SBP, DBP, ALT, AST, GGT, FPG, HbA1c, habit of exercise, smoking status and drinking status In the mediation analysis stratified by sex (Table 6), we found that the mediation effect of lipid parameters on the association between BMI and NAFLD was generally higher in women than in men; among them, the mediation percentages of lipid parameters such as RC/HDL-C ratio, RC, TG/HDL-C ratio, TG, TC/HDL-C ratio and non-HDL-C/HDL-C ratio were more than $10\%$ in women ($15.4\%$, $12.9\%$, $12.5\%$, $11.7\%$, $11.5\%$, $11.5\%$, respectively, $P \leq 0.001$). While the mediation effect of the RC/HDL-C ratio was the largest in men, accounting for $7\%$.Table 6Mediation analysis for BMI and NAFLD via lipid parameters in the whole population stratified by sexMediatorTotal effectMediation effectDirect effectPM (%)P-value of PMMen ($$n = 7441$$) TC0.110 (0.088, 0.132)0.001 (0.000, 0.002)0.109 (0.087, 0.131)10.006 TG0.110 (0.087, 0.132)0.005 (0.001, 0.008)0.105 (0.083, 0.128)4.4< 0.001 HDL-C0.111 (0.088, 0.133)0.004 (0.002, 0.007)0.106 (0.084, 0.128)3.9< 0.001 LDL-C0.110 (0.088, 0.132)0.002 (0.001, 0.003)0.108 (0.086, 0.131)1.7< 0.001 Non-HDL-C0.110 (0.087, 0.132)0.004 (0.002, 0.006)0.107 (0.084, 0.129)3.2< 0.001 RC0.110 (0.087, 0.132)0.007 (0.003, 0.010)0.104 (0.081, 0.126)5.9< 0.001 TC/HDL-C ratio0.111 (0.088, 0.133)0.008 (0.005, 0.011)0.103 (0.080, 0.125)7< 0.001 TG/HDL-C ratio0.111 (0.088, 0.132)0.006 (0.003, 0.010)0.105 (0.082, 0.127)5.5< 0.001 LDL/HDL-C ratio0.111 (0.088, 0.132)0.007 (0.004, 0.009)0.104 (0.082, 0.127)5.9< 0.001 Non-HDL-C/HDL-C ratio0.111 (0.088, 0.133)0.008 (0.005, 0.011)0.103 (0.080, 0.125)7< 0.001 RC/HDL-C ratio0.111 (0.088, 0.132)0.008 (0.005, 0.012)0.102 (0.080, 0.125)7< 0.001Women ($$n = 6840$$) TC0.084 (0.073, 0.097)0.001 (0.000, 0.002)0.083 (0.072, 0.096)10.046 TG0.084 (0.073, 0.096)0.010 (0.008, 0.012)0.074 (0.063, 0.087)11.7< 0.001 HDL-C0.084 (0.073, 0.097)0.005 (0.003, 0.006)0.080 (0.069, 0.092)5.4< 0.001 LDL-C0.084 (0.073, 0.097)0.003 (0.001, 0.004)0.082 (0.070, 0.094)3.10.002 Non-HDL-C0.084 (0.073, 0.096)0.004 (0.002, 0.006)0.080 (0.069, 0.093)4.9< 0.001 RC0.084 (0.073, 0.096)0.011 (0.009, 0.013)0.073 (0.062, 0.086)12.9< 0.001 TC/HDL-C ratio0.084 (0.073, 0.096)0.010 (0.007, 0.012)0.075 (0.063, 0.087)11.5< 0.001 TG/HDL-C ratio0.084 (0.073, 0.096)0.011 (0.008, 0.013)0.074 (0.062, 0.086)12.5< 0.001 LDL/HDL-C ratio0.084 (0.073, 0.096)0.008 (0.006, 0.011)0.076 (0.064, 0.089)9.8< 0.001 Non-HDL-C/HDL-C ratio0.084 (0.073, 0.096)0.010 (0.007, 0.012)0.075 (0.063, 0.087)11.5< 0.001 RC/HDL-C ratio0.084 (0.073, 0.096)0.013 (0.010, 0.016)0.071 (0.060, 0.084)15.4< 0.001PM: propotion mediate; other abbreviations as in Table 1Adjusting variables: age, WC, SBP, DBP, ALT, AST, GGT, FPG, HbA1c, habit of exercise, smoking status and drinking status In the mediation analysis stratified by age (Additional file 1: Table S14), we only found significant mediation effects in the 30–44 years old and 45–59 years old subgroups, and the results were basically consistent with the results of the whole population. The contribution of non-traditional lipid parameters to the risk of NAFLD associated with BMI was higher than that of traditional lipid parameters, especially RC, RC/HDL-C ratio, non-HDL-C/HDL-C ratio, and TC/HDL-C ratio.
## Discussion
In this data analysis of 14,251 general subjects based on the NAGALA cohort, we have found that both traditional and non-traditional lipid parameters played an important role in the association between BMI and NAFLD, among which the non-traditional lipid parameters RC/HDL-C ratio, RC, non-HDL-C/HDL-C ratio and TC/HDL-C ratio accounted for a large proportion in mediating the association between BMI and NAFLD, which were $11.4\%$, $10.8\%$, $10.2\%$, and $10.2\%$, respectively. The further stratified analysis also found that this mediation effect existed only in normal-weight people and young and middle-aged people (30–59 years old); in addition, compared with men, all lipid parameters except TC accounted for a higher proportion of mediation effects in women.
NAFLD has become a global health problem, affecting $\frac{1}{4}$ of the global population, and its rising prevalence is synchronized with the prevalence of metabolic syndrome, obesity, and type 2 diabetes [3]. In addition, a large number of epidemiological evidence showed that BMI and dyslipidemia were closely related to the occurrence and development of NAFLD [8, 11–14, 44]. NAFLD is a complex metabolic-related disease, and the latest research has proposed to change the name of NAFLD to metabolic-associated fatty liver disease [45] because this name was more in line with the pathophysiological process of NAFLD. Early management of these controllable risk factors for metabolic disorders may be useful and efficient in reducing the risk of NAFLD [46].
Similar to the results of some previous studies [11–14, 47–52], we have also reported the relationship between BMI and lipid parameters and NAFLD in the current study. The results showed that there was a significant correlation between BMI and lipid parameters and NAFLD after adjusting for potential confounding factors. In addition, after further adjustment of lipid parameters, we observed that the magnitude of the association between BMI and NAFLD had decreased, which suggested that lipid parameters may play a role in mediating the association, thereby supporting the next step of mediation analysis in this study. At present, little is known about the relationship between BMI and NAFLD mediated by lipid parameters. As far as we know, there is only one study to explore the relationship between BMI and NAFLD mediated by TG [53]. In the study by Xing et al., who analyzed data from 15,943 employees from the Kailuan Group in China, they showed that approximately $26\%$ of the effect of BMI on NAFLD was mediated through TG levels in the high BMI group (BMI ≥ 24 kg/m2), while in the low BMI group (BMI < 24 kg/m2), no significant mediation effect was observed. In this study, we evaluated the meditation percentage of TG-mediated BMI and NAFLD association in the whole population to be $8.3\%$, while further stratification according to the cut-off point of BMI in the Asian population, we found that the mediation effect of TG-mediated BMI and NAFLD association was only observed in normal-weight people, but it was not significant in low-weight people and overweight/obese people. In order to explore whether the inconsistency between the research results of Xing et al. and the current research results is related to the difference in BMI cut-off points, we continued to use the same BMI cut-off points as that of Xing et al. for further exploratory analysis [53]. The results still showed that the meditation effect of TG on the association between BMI and NAFLD existed only in normal-weight people (Additional file 1: Table S15). On the reasons why the results remain inconsistent after further analysis, we considered that it may be related to the different constituent ratios of the study population. In the current study, overweight/obese people only accounted for $15.89\%$, even if using 24 kg/m2 as the BMI cut-off point, the proportion of overweight/obese people in our study was only $23.65\%$, but in the study of Xing et al., the proportion of overweight/obesity people was $45.78\%$ [53]. In addition, it is worth mentioning that Xing et al. studied coal miners with an average age of 51 years old in northern China (average TG was 1.10 mmol/L), while our subjects were general people with an average age of 43 years (average TG was 0.89 mmol/L). It is well known that coal workers have been exposed to a large amount of coal dust for a long time, which can lead to many health problems [54], and the most common of which are lung disease and heart disease (arrhythmia and myocardial ischemia). Coal dust contains a variety of toxic substances, including acrolein, which can have a serious impact on lipid metabolism, especially glycerol phospholipid metabolism [55–57]. In further animal-based studies, Gasparotto et al. also found that coal dust can further enhance the pro-inflammatory characteristics of obese rats and induce microvesicular steatosis [58]. Correlative evidence based on toxicology studies further suggested that hepatotoxicity induced by increased chemical exposure may be the result of oxidative hazards that lead to mitochondrial/lysosomal toxic connection and disorders in biochemical markers, and antioxidants can play an important role in reversing this process [17–21]. Some clinical evidence also supported this important discovery [59, 60]. These findings may explain the higher proportion of TG-mediated BMI and NAFLD association in the study by Xing et al., and provide useful data for the prevention and treatment of NAFLD. Considering the uniqueness of the human body, combined with the existing research results based on human samples, we suggested that people at high risk of NAFLD should pay attention to increasing the intake of natural antioxidants and trace element antioxidants from dietary sources at an early stage, such as anthocyanins, curcumin and resveratrol as well as coffee, tea, soy, vitamin C, vitamin E and astaxanthin [2, 59, 61, 62]. For NAFLD patients with dyslipidemia or high-risk groups, it is recommended to use antioxidant vitamin C and vitamin E in combination on the basis of statin therapy. The recommended dose is 20 mg of atorvastatin, 1 g of vitamin C and 1000 IU of vitamin E daily mix of [60]. In addition, therapeutic drugs acting on hepatocyte channels should also be paid attention to, because dysfunctional gap junctions are closely related to the pathogenesis of NAFLD, and normally functioning channels contribute to the maintenance of tissue homeostasis and liver function [63].
In addition to the BMI subgroup, we also conducted the same analysis in the sex subgroup and age subgroup. *In* general, even after further stratification, we can still observe that the contribution of non-traditional lipid parameters to the association of BMI with NAFLD is higher than that of traditional lipid parameters in subgroups. Similar findings has been reported in several previous studies where non-traditional lipid parameters had better value than traditional lipid parameters in the assessment of the risk of NAFLD onset/prevalence [25–31, 64]. These new findings may provide targeted monitoring recommendations for BMI intervention in different populations. In the current study, we also found that almost all lipid parameters had a higher mediation effect on BMI and NAFLD association in women than in men, and the two non-traditional lipid parameters, RC/HDL-C ratio and RC, were the most noteworthy, which played a more important role in BMI and NAFLD association than other lipid parameters. The reason for this special phenomenon is currently unknown, and it may be related to the function of BMI-related but obviously sex-specific substances in the body. Based on a recent targeted lipidomic analysis of 10,339 adults in the AusDiab study, Beyene et al. measured 706 lipids from 36 different lipid classes and assessed whether the relationship between BMI and lipidomic profiles differed by sex. The results showed that acylcarnitine species exhibited opposite associations in men and women [65], a finding that may provide a potential direction to further explore the findings of the current study. According to the findings of the current study, we recommend that women should focus on monitoring non-traditional lipid parameters, especially RC/HDL-C ratio and RC if they need to prevent or treat NAFLD through the intervention of modifiable risk factors. Moreover, we further evaluated the effect of lipid parameters on BMI and NAFLD association in different age groups, and finally found significant mediation effects in 30–44 years old and 45–59 years old subgroups. The results also supported that the mediation effects of non-traditional lipid parameters in BMI and NAFLD association were greater than that of traditional lipid parameters, in which RC/HDL-C ratio and RC played a key role in mediating the association in people aged 30–44, while in 45–59 years old, only RC/HDL-C ratio played a role in mediating more than $10\%$ of the association. Overall, regardless of sex, age, fatness or thinness, we recommend that at least attention should be paid to the monitoring of the RC/HDL-C ratio, a non-traditional lipid parameter.
The mechanism of BMI and NAFLD association mediated by lipid parameters is still unclear. Given the relationship between BMI and lipid parameters and NAFLD, we speculated that insulin resistance (IR) may have an important influence on this mediation effect [8, 11–14]. Additionally, the findings of the current study may be of great significance for understanding how BMI leads to the onset of NAFLD and provide new and useful references for NAFLD risk intervention. At present, some studies have described the evidence that intervention against both BMI and dyslipidemia can reduce the risk of NAFLD. In the study of Magkos et al. [ 66], they found that progressive weight loss will improve human adipose disease and IR, and weight loss of $5\%$, $11\%$, and $16\%$ can reduce TG levels in the liver by $13\%$, $52\%$, and $65\%$, respectively; not only that, serum TG levels also showed a significant downward trend. Moreover, Promrat and Lazo et al. also had similar findings that weight loss will greatly improve liver steatosis [67, 68]. The above findings were very useful, and based on the results of this study, we suggested that we also need to strengthen the monitoring of non-traditional lipid parameters, especially the RC/HDL-C ratio and RC levels, which may be more valuable simple parameters that can improve compliance and success rate for health behavior changes. For future work, we suggested: [1] Further follow-up studies to verify the effectiveness of monitoring traditional/non-traditional lipid parameters in different populations for future NAFLD risk assessment. [ 2] Further research on drug intervention, including antioxidants, lipid-lowering drugs and cell channel modification drugs. [ 3] Medical workers or public health decision makers incorporate the joint management of BMI and lipid parameters into the disease prevention or treatment programs of NAFLD, and special attentions should be paid to the monitoring role of unconventional lipid parameters.
## Limitations
There were several limitations in the current study that need to be acknowledged: [1] Lipid parameters explained only part of the association between BMI and NAFLD, and further studies are needed to evaluate other potential mediators that mediate the association between BMI and NAFLD. [ 2] The study subjects were the *Japanese* general population, and BMI was classified according to the criteria suitable for Asians, so the results of this study should be used with caution for other ethnic groups. [ 3] Serum insulin was not measured in this study, and IR status could not be assessed, so the important role of IR in the association could not be further clarified. [ 4] The mechanisms behind sex and age affecting the mediation effects of lipid parameters remain unclear, and further longitudinal studies are needed to clarify the potential associations and effects. [ 5] NAFLD was not graded in the current study, so it was not possible to further assess whether the association of BMI with different grades of NAFLD was mediated by lipid parameters. [ 6] Although we have excluded the subjects who are taking drugs at baseline at the beginning of the study, and adjusted the factors including exercise habits, smoking and drinking status, blood pressure, blood glucose, liver enzyme metabolism, sex, age and WC in the subsequent data analysis, it is undeniable that there are still many life-related factors that have not been considered in the current study, which will inevitably lead to some residual confounding.
## Conclusion
In conclusion, all lipid parameters were found to be involved in and mediate the risk of NAFLD associated with BMI in this study, with non-traditional lipid parameters contributing more to the mediation effect of the association than traditional lipid parameters, especially RC, RC/HDL-C ratio, non-HDL-C/HDL-C ratio, and TC/HDL-C ratio; These findings provided new ideas for the prevention and treatment of NAFLD, we call for more attention to non-traditional lipid parameters when intervening on BMI as a controllable risk factor.
## Supplementary Information
Additional file 1: Table S1. Collinearity diagnostics steps of TC with other covariates. Table S2. Collinearity diagnostics steps of HDL-C with other covariates. Table S3. Collinearity diagnostics steps of TG with other covariates. Table S4. Collinearity diagnostics steps of LDL-C with other covariates. Table S5. Collinearity diagnostics steps of non-HDL-C with other covariates. Table S6. Collinearity diagnostics steps of RC with other covariates. Table S7. Collinearity diagnostics steps of TC/HDL-C ratio with other covariates. Table S8. Collinearity diagnostics steps of TG/HDL-C ratio with other covariates. Table S9. Collinearity diagnostics steps of LDL/HDL-C ratio with other covariates. Table S10. Collinearity diagnostics steps of non-HDL/HDL-C ratio with other covariates. Table S11. Collinearity diagnostics steps of RC/HDL-C ratio with other covariates. Table S12. Collinearity diagnostics steps of BMI with other covariates. Table S13. Association of BMI with lipid parameters. Table S14. Mediation analysis for BMI and NAFLD via lipid parameters in the whole population stratified by age. Table S15. Mediation analysis for BMI and NAFLD via lipid parameters in the whole population stratified by BMI.
## References
1. Milić S, Lulić D, Štimac D. **Non-alcoholic fatty liver disease and obesity: biochemical, metabolic and clinical presentations**. *World J Gastroenterol* (2014.0) **20** 9330-9337. PMID: 25071327
2. Hernandez-Rodas MC, Valenzuela R, Videla LA. **Relevant aspects of nutritional and dietary interventions in non-alcoholic fatty liver disease**. *Int J Mol Sci* (2015.0) **16** 25168-25198. DOI: 10.3390/ijms161025168
3. Younossi Z, Anstee QM, Marietti M, Hardy T, Henry L, Eslam M. **Global burden of NAFLD and NASH: trends, predictions, risk factors and prevention**. *Nat Rev Gastroenterol Hepatol* (2018.0) **15** 11-20. DOI: 10.1038/nrgastro.2017.109
4. Estes C, Razavi H, Loomba R, Younossi Z, Sanyal AJ. **Modeling the epidemic of nonalcoholic fatty liver disease demonstrates an exponential increase in burden of disease**. *Hepatology* (2018.0) **67** 123-133. DOI: 10.1002/hep.29466
5. Younossi ZM, Blissett D, Blissett R, Henry L, Stepanova M, Younossi Y. **The economic and clinical burden of nonalcoholic fatty liver disease in the United States and Europe**. *Hepatology* (2016.0) **64** 1577-1586. DOI: 10.1002/hep.28785
6. Soares ALG, Banda L, Amberbir A, Jaffar S, Musicha C, Price AJ. **A comparison of the associations between adiposity and lipids in Malawi and the United Kingdom**. *BMC Med* (2020.0) **18** 181. DOI: 10.1186/s12916-020-01648-0
7. Musso G, Cassader M, Paschetta E, Gambino R. **Bioactive lipid species and metabolic pathways in progression and resolution of nonalcoholic steatohepatitis**. *Gastroenterology* (2018.0) **155** 282-302.e8. DOI: 10.1053/j.gastro.2018.06.031
8. Katsiki N, Mikhailidis DP, Mantzoros CS. **Non-alcoholic fatty liver disease and dyslipidemia: an update**. *Metabolism* (2016.0) **65** 1109-1123. DOI: 10.1016/j.metabol.2016.05.003
9. Gaggini M, Morelli M, Buzzigoli E, DeFronzo RA, Bugianesi E, Gastaldelli A. **Non-alcoholic fatty liver disease (NAFLD) and its connection with insulin resistance, dyslipidemia, atherosclerosis and coronary heart disease**. *Nutrients* (2013.0) **5** 1544-1560. DOI: 10.3390/nu5051544
10. Speliotes EK, Massaro JM, Hoffmann U, Vasan RS, Meigs JB, Sahani DV. **Fatty liver is associated with dyslipidemia and dysglycemia independent of visceral fat: the Framingham Heart Study**. *Hepatology* (2010.0) **51** 1979-1987. DOI: 10.1002/hep.23593
11. Chen TP, Lin WY, Chiang CH, Shen TH, Huang KC, Yang KC. **Metabolically healthy obesity and risk of non-alcoholic fatty liver disease severity independent of visceral fat**. *J Gastroenterol Hepatol* (2021.0) **36** 2903-2910. DOI: 10.1111/jgh.15544
12. Lonardo A, Mantovani A, Lugari S, Targher G. **Epidemiology and pathophysiology of the association between NAFLD and metabolically healthy or metabolically unhealthy obesity**. *Ann Hepatol* (2020.0) **19** 359-366. DOI: 10.1016/j.aohep.2020.03.001
13. Tang Z, Pham M, Hao Y, Wang F, Patel D, Jean-Baptiste L. **Sex, Age, and BMI modulate the association of physical examinations and blood biochemistry parameters and NAFLD: a retrospective study on 1994 cases observed at Shuguang Hospital**. *China Biomed Res Int* (2019.0) **2019** 1246518. PMID: 31341886
14. Cho HC. **Prevalence and factors associated with nonalcoholic fatty liver disease in a nonobese Korean Population**. *Gut Liver* (2016.0) **10** 117-125. DOI: 10.5009/gnl14444
15. Zheng S, Yang Y, Wen C, Liu W, Cao L, Feng X. **Effects of environmental contaminants in water resources on nonalcoholic fatty liver disease**. *Environ Int* (2021.0) **154** 106555. DOI: 10.1016/j.envint.2021.106555
16. Thayer KA, Heindel JJ, Bucher JR, Gallo MA. **Role of environmental chemicals in diabetes and obesity: a National Toxicology Program workshop review**. *Environ Health Perspect* (2012.0) **120** 779-789. DOI: 10.1289/ehp.1104597
17. Videla LA, Valenzuela R. **Perspectives in liver redox imbalance: toxicological and pharmacological aspects underlying iron overloading, nonalcoholic fatty liver disease, and thyroid hormone action**. *BioFactors* (2022.0) **48** 400-415. DOI: 10.1002/biof.1797
18. Ahmadian E, Eftekhari A, Fard JK, Babaei H, Nayebi AM, Mohammadnejad D. **In vitro and in vivo evaluation of the mechanisms of citalopram-induced hepatotoxicity**. *Arch Pharm Res* (2017.0) **40** 1296-1313. DOI: 10.1007/s12272-016-0766-0
19. Ahmadian E, Babaei H, Mohajjel Nayebi A, Eftekhari A, Eghbal MA. **Mechanistic approach for toxic effects of bupropion in primary rat hepatocytes**. *Drug Res* (2017.0) **67** 217-222. DOI: 10.1055/s-0042-123034
20. Eftekhari A, Hasanzadeh A, Khalilov R, Hosainzadegan H, Ahmadian E, Eghbal MA. **Hepatoprotective role of berberine against paraquat-induced liver toxicity in rat**. *Environ Sci Pollut Res Int* (2020.0) **27** 4969-4975. DOI: 10.1007/s11356-019-07232-1
21. Eghbal MA, Taziki S, Sattari MR. **Mechanisms of phenytoin-induced toxicity in freshly isolated rat hepatocytes and the protective effects of taurine and/or melatonin**. *J Biochem Mol Toxicol* (2014.0) **28** 111-118. DOI: 10.1002/jbt.21542
22. Okamura T, Hashimoto Y, Hamaguchi M, Obora A, Kojima T, Fukui M. **Ectopic fat obesity presents the greatest risk for incident type 2 diabetes: a population-based longitudinal study**. *Int J Obes (Lond)* (2019.0) **43** 139-148. DOI: 10.1038/s41366-018-0076-3
23. Hashimoto Y, Hamaguchi M, Kojima T, Ohshima Y, Ohbora A, Kato T. **Modest alcohol consumption reduces the incidence of fatty liver in men: a population-based large-scale cohort study**. *J Gastroenterol Hepatol* (2015.0) **30** 546-552. DOI: 10.1111/jgh.12786
24. Chen Y, Zhang X, Pan B, Jin X, Yao H, Chen B. **A modified formula for calculating low-density lipoprotein cholesterol values**. *Lipids Health Dis* (2010.0) **9** 52. DOI: 10.1186/1476-511X-9-52
25. Zelber-Sagi S, Salomone F, Yeshua H, Lotan R, Webb M, Halpern Z. **Non-high-density lipoprotein cholesterol independently predicts new onset of non-alcoholic fatty liver disease**. *Liver Int* (2014.0) **34** e128-e135. DOI: 10.1111/liv.12318
26. Zou Y, Lan J, Zhong Y, Yang S, Zhang H, Xie G. **Association of remnant cholesterol with nonalcoholic fatty liver disease: a general population-based study**. *Lipids Health Dis* (2021.0) **20** 139. DOI: 10.1186/s12944-021-01573-y
27. Wu KT, Kuo PL, Su SB, Chen YY, Yeh ML, Huang CI. **Nonalcoholic fatty liver disease severity is associated with the ratios of total cholesterol and triglycerides to high-density lipoprotein cholesterol**. *J Clin Lipidol* (2016.0) **10** 420-5.e1. DOI: 10.1016/j.jacl.2015.12.026
28. Fan N, Peng L, Xia Z, Zhang L, Song Z, Wang Y. **Triglycerides to high-density lipoprotein cholesterol ratio as a surrogate for nonalcoholic fatty liver disease: a cross-sectional study**. *Lipids Health Dis* (2019.0) **18** 39. DOI: 10.1186/s12944-019-0986-7
29. Zou Y, Zhong L, Hu C, Zhong M, Peng N, Sheng G. **LDL/HDL cholesterol ratio is associated with new-onset NAFLD in Chinese non-obese people with normal lipids: a 5-year longitudinal cohort study**. *Lipids Health Dis* (2021.0) **20** 28. DOI: 10.1186/s12944-021-01457-1
30. Wang K, Shan S, Zheng H, Zhao X, Chen C, Liu C. **Non-HDL-cholesterol to HDL-cholesterol ratio is a better predictor of new-onset non-alcoholic fatty liver disease than non-HDL-cholesterol: a cohort study**. *Lipids Health Dis* (2018.0) **17** 196. DOI: 10.1186/s12944-018-0848-8
31. Zou Y, Hu C, Kuang M, Chai Y. **Remnant cholesterol/high-density lipoprotein cholesterol ratio is a new powerful tool for identifying non-alcoholic fatty liver disease**. *BMC Gastroenterol* (2022.0) **22** 134. DOI: 10.1186/s12876-022-02216-x
32. Hamaguchi M, Kojima T, Itoh Y, Harano Y, Fujii K, Nakajima T. **The severity of ultrasonographic findings in nonalcoholic fatty liver disease reflects the metabolic syndrome and visceral fat accumulation**. *Am J Gastroenterol* (2007.0) **102** 2708-2715. DOI: 10.1111/j.1572-0241.2007.01526.x
33. Sato T, Matsuyama Y. **Marginal structural models as a tool for standardization**. *Epidemiology* (2003.0) **14** 680-686. DOI: 10.1097/01.EDE.0000081989.82616.7d
34. Muanda FT, Weir MA, Bathini L, Blake PG, Chauvin K, Dixon SN. **Association of Baclofen with encephalopathy in patients with chronic kidney disease**. *JAMA* (2019.0) **322** 1987-1995. DOI: 10.1001/jama.2019.17725
35. Kim JH. **Multicollinearity and misleading statistical results**. *Korean J Anesthesiol* (2019.0) **72** 558-569. DOI: 10.4097/kja.19087
36. Fitchett EJA, Seale AC, Vergnano S, Sharland M, Heath PT, Saha SK. **Strengthening the Reporting of Observational Studies in Epidemiology for Newborn Infection (STROBE-NI): an extension of the STROBE statement for neonatal infection research**. *Lancet Infect Dis* (2016.0) **16** e202-e213. DOI: 10.1016/S1473-3099(16)30082-2
37. VanderWeele TJ. **Mediation analysis: a practitioner's guide**. *Annu Rev Public Health* (2016.0) **37** 17-32. DOI: 10.1146/annurev-publhealth-032315-021402
38. Huang TD, Behary J, Zekry A. **Non-alcoholic fatty liver disease: a review of epidemiology, risk factors, diagnosis and management**. *Intern Med J* (2020.0) **50** 1038-1047. DOI: 10.1111/imj.14709
39. Åberg F, Färkkilä M. **Drinking and obesity: alcoholic liver disease/nonalcoholic fatty liver disease interactions**. *Semin Liver Dis* (2020.0) **40** 154-162. DOI: 10.1055/s-0040-1701443
40. Hamabe A, Uto H, Imamura Y, Kusano K, Mawatari S, Kumagai K. **Impact of cigarette smoking on onset of nonalcoholic fatty liver disease over a 10-year period**. *J Gastroenterol* (2011.0) **46** 769-778. DOI: 10.1007/s00535-011-0376-z
41. Vekic J, Zeljkovic A, Stefanovic A, Jelic-Ivanovic Z, Spasojevic-Kalimanovska V. **Obesity and dyslipidemia**. *Metabolism* (2019.0) **92** 71-81. DOI: 10.1016/j.metabol.2018.11.005
42. Imai K, Keele L, Tingley D, Yamamoto T. **Causal mediation analysis using R**. *Lecture Notes Statist* (2010.0) **196** 129-154. DOI: 10.1007/978-1-4419-1764-5_8
43. **Appropriate body-mass index for Asian populations and its implications for policy and intervention strategies**. *Lancet* (2004.0) **363** 157-163. DOI: 10.1016/S0140-6736(03)15268-3
44. Zhang QQ, Lu LG. **Nonalcoholic fatty liver disease: dyslipidemia, risk for cardiovascular complications, and treatment strategy**. *J Clin Transl Hepatol* (2015.0) **3** 78-84. DOI: 10.14218/JCTH.2014.00037
45. Eslam M, Newsome PN, Sarin SK, Anstee QM, Targher G, Romero-Gomez M. **A new definition for metabolic dysfunction-associated fatty liver disease: an international expert consensus statement**. *J Hepatol* (2020.0) **73** 202-209. DOI: 10.1016/j.jhep.2020.03.039
46. Wong VW, Wong GL, Chan RS, Shu SS, Cheung BH, Li LS. **Beneficial effects of lifestyle intervention in non-obese patients with non-alcoholic fatty liver disease**. *J Hepatol* (2018.0) **69** 1349-1356. DOI: 10.1016/j.jhep.2018.08.011
47. Quirós-Tejeira RE, Rivera CA, Ziba TT, Mehta N, Smith CW, Butte NF. **Risk for nonalcoholic fatty liver disease in Hispanic youth with BMI > or =95th percentile**. *J Pediatr Gastroenterol Nutr* (2007.0) **44** 228-236. DOI: 10.1097/MPG.0b013e31802d4acc
48. Chang Y, Jung HS, Cho J, Zhang Y, Yun KE, Lazo M. **Metabolically healthy obesity and the development of nonalcoholic fatty liver disease**. *Am J Gastroenterol* (2016.0) **111** 1133-1140. DOI: 10.1038/ajg.2016.178
49. Liu M, Wang J, Zeng J, Cao X, He Y. **Association of NAFLD with diabetes and the impact of BMI changes: a 5-year cohort study based on 18,507 elderly**. *J Clin Endocrinol Metab* (2017.0) **102** 1309-1316. DOI: 10.1210/jc.2016-3440
50. Rong Y, Chun-Yan N, Hong-Xin Z, Lu Y, Wen W, Yu T. **Association of adolescent obesity with nonalcoholic fatty liver disease and related risk factors in Xi'an, China**. *Ann Hepatol* (2018.0) **17** 85-91. DOI: 10.5604/01.3001.0010.7538
51. Ballestri S, Nascimbeni F, Romagnoli D, Lonardo A. **The independent predictors of non-alcoholic steatohepatitis and its individual histological features: insulin resistance, serum uric acid, metabolic syndrome, alanine aminotransferase and serum total cholesterol are a clue to pathogenesis and candidate targets for treatment**. *Hepatol Res* (2016.0) **46** 1074-1087. DOI: 10.1111/hepr.12656
52. McCullough A, Previs SF, Dasarathy J, Lee K, Osme A, Kim C. **HDL flux is higher in patients with nonalcoholic fatty liver disease**. *Am J Physiol Endocrinol Metab* (2019.0) **317** E852-E862. DOI: 10.1152/ajpendo.00193.2019
53. Xing J, Guan X, Zhang Q, Chen S, Wu S, Sun X. **Triglycerides mediate body mass index and nonalcoholic fatty liver disease: a population-based study**. *Obes Facts* (2021.0) **14** 190-196. DOI: 10.1159/000514848
54. Liu T, Liu S. **The impacts of coal dust on miners' health: a review**. *Environ Res* (2020.0) **190** 109849. DOI: 10.1016/j.envres.2020.109849
55. 55.Fomenko DV, Gorokhova LG, Panev NI, Kazitskaia AS, Bondarev OI. Clinical and experimental studies of metabolic response to chronic exposure to coal dust. Med Tr Prom Ekol. 2011;15–21.
56. Conklin DJ, Barski OA, Lesgards JF, Juvan P, Rezen T, Rozman D. **Acrolein consumption induces systemic dyslipidemia and lipoprotein modification**. *Toxicol Appl Pharmacol* (2010.0) **243** 1-12. DOI: 10.1016/j.taap.2009.12.010
57. 57.Peng F, Dai J, Qian Q, Cao X, Wang L, Zhu M, et al. Serum metabolic profiling of coal worker's pneumoconiosis using untargeted lipidomics. Environ Sci Pollut Res Int. 2022.
58. Gasparotto J, Chaves PR, da Boit MK, da Rosa-Siva HT, Bortolin RC, Silva LFO. **Obese rats are more vulnerable to inflammation, genotoxicity and oxidative stress induced by coal dust inhalation than non-obese rats**. *Ecotoxicol Environ Saf* (2018.0) **165** 44-51. DOI: 10.1016/j.ecoenv.2018.08.097
59. Chen L, Fan Z, Sun X, Qiu W, Mu W, Chai K. **Diet-derived antioxidants and nonalcoholic fatty liver disease: a Mendelian randomization study**. *Hepatol Int* (2022.0). DOI: 10.1007/s12072-022-10443-3
60. Foster T, Budoff MJ, Saab S, Ahmadi N, Gordon C, Guerci AD. **Atorvastatin and antioxidants for the treatment of nonalcoholic fatty liver disease: the St Francis Heart Study randomized clinical trial**. *Am J Gastroenterol* (2011.0) **106** 71-77. DOI: 10.1038/ajg.2010.299
61. Salomone F, Godos J, Zelber-Sagi S. **Natural antioxidants for non-alcoholic fatty liver disease: molecular targets and clinical perspectives**. *Liver Int* (2016.0) **36** 5-20. DOI: 10.1111/liv.12975
62. Chen G, Ni Y, Nagata N, Xu L, Ota T. **Micronutrient antioxidants and nonalcoholic fatty liver disease**. *Int J Mol Sci* (2016.0) **17** 1379. DOI: 10.3390/ijms17091379
63. Hernández-Guerra M, Hadjihambi A, Jalan R. **Gap junctions in liver disease: implications for pathogenesis and therapy**. *J Hepatol* (2019.0) **70** 759-772. DOI: 10.1016/j.jhep.2018.12.023
64. Lu S, Kuang M, Yue J, Hu C, Sheng G, Zou Y. **Utility of traditional and non-traditional lipid indicators in the diagnosis of nonalcoholic fatty liver disease in a Japanese population**. *Lipids Health Dis* (2022.0) **21** 95. DOI: 10.1186/s12944-022-01712-z
65. Beyene HB, Olshansky G, Smith AAT, Giles C, Huynh K, Cinel M. **High-coverage plasma lipidomics reveals novel sex-specific lipidomic fingerprints of age and BMI: evidence from two large population cohort studies**. *PLoS Biol* (2020.0) **18** e3000870. DOI: 10.1371/journal.pbio.3000870
66. Ryan DH, Yockey SR. **Weight loss and improvement in comorbidity: differences at 5%, 10%, 15%, and Over**. *Curr Obes Rep* (2017.0) **6** 187-194. DOI: 10.1007/s13679-017-0262-y
67. Promrat K, Kleiner DE, Niemeier HM, Jackvony E, Kearns M, Wands JR. **Randomized controlled trial testing the effects of weight loss on nonalcoholic steatohepatitis**. *Hepatology* (2010.0) **51** 121-129. DOI: 10.1002/hep.23276
68. Lazo M, Solga SF, Horska A, Bonekamp S, Diehl AM, Brancati FL. **Effect of a 12-month intensive lifestyle intervention on hepatic steatosis in adults with type 2 diabetes**. *Diabetes Care* (2010.0) **33** 2156-2163. DOI: 10.2337/dc10-0856
|
---
title: Kai-Xin-San protects against mitochondrial dysfunction in Alzheimer’s disease
through SIRT3/NLRP3 pathway
authors:
- ShiJie Su
- Gongcan Chen
- Minghuang Gao
- Guangcheng Zhong
- Zerong Zhang
- Dongyun Wei
- Xue Luo
- Qi Wang
journal: Chinese Medicine
year: 2023
pmcid: PMC10012453
doi: 10.1186/s13020-023-00722-y
license: CC BY 4.0
---
# Kai-Xin-San protects against mitochondrial dysfunction in Alzheimer’s disease through SIRT3/NLRP3 pathway
## Abstract
### Background
Kai-Xin-San (KXS) has been reported to have a good curative impact on dementia. The purpose of the study was to determine whether KXS might ameliorate cognitive deficits in APP/PS1 mice and to evaluate its neuroprotective mechanism.
### Methods
APP/PS1 mice were employed as an AD animal model; Aβ1–42 and KXS-containing serum were used in HT22 cells. Four different behavioral tests were used to determine the cognitive ability of mice. Nissl staining was utilized to detect hippocampal neuron changes. ROS, SOD, and MDA were used to detect oxidative stress levels. Transmission electron microscopy and Western blot were used to evaluate mitochondrial morphology, mitochondrial division, and fusion state. Western blotting and immunofluorescence identified PSD95, BDNF, NGF, SYN, SIRT3, and NLRP3 inflammasome levels.
### Results
The results indicated that KXS protected APP/PS1 mice against cognitive impairments. KXS suppressed neuronal apoptosis and oxidative stress among APP/PS1 mice. KXS and KXS-containing serum improved mitochondrial dysfunction and synaptic and neurotrophic factors regarding APP/PS1 mice. In addition, KXS and KXS-containing serum enhanced mitochondrial SIRT3 expression and reduced NLRP3 inflammasome expression in APP/PS1 mice.
### Conclusion
KXS improves cognitive dysfunction among APP/PS1 mice via regulating SIRT3-mediated neuronal cell apoptosis. These results suggested that KXS was proposed as a neuroprotective agent for AD progression.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s13020-023-00722-y.
## Introduction
Alzheimer’s disease (AD) is a progressive neurological disorder that causes impairments and memory loss [1]. This debilitating disease exacts a heavy toll on clinically infected persons and family cares, as well as an enormous financial burden on society. The current understanding of AD pathogenic mechanisms is incomplete [2, 3]. The malfunction of clinical trials attacking the amyloid-peptide (Aβ) [4], a pathological hallmark of AD, has highlighted the critical need to investigate alternative features of AD pathology.
Reactive oxygen species (ROS) contribute to AD. Extensive oxidative damage resulting from ROS production is detected within the AD brain [5]. Overproduction of ROS due to depletion of antioxidant enzymes may exacerbate neuronal apoptosis by damaging cellular components such proteins, fats, and DNA. It has been hypothesized that mitochondrial malfunction is an underlying cause of AD and other forms of neurodegeneration that are related to advancing age [6]. Mitochondrial dysfunction had been detected among AD patient’s brains and in mouse models of the disease. This dysfunction characterized by a decline in bioenergetics and ATP synthesis, morphological disturbances, a mitochondrial dynamics imbalance, besides mitochondria redistribution [6]. Mitochondrial dysfunction and ROS generation have been identified as a trigger of neuroinflammation [7–9] through the cGAS-STING pathway and/or the NOD-like receptor-related protein 3 (NLRP3) inflammasome [10]. Mitochondrial damages or generation of mitochondrial ROS (mtROS) are important modulator of NLRP3 activation [11]. Moreover, the NLRP3 inflammasome could be activated by mitochondrial DNA (mtDNA) [12]. Recent findings supposed that the NLRP3 inflammasome is triggered among the of AD brains and MCI patients, and APP/PS1 mice [13]. Reducing Aβ deposition and memory loss protection can be achieved with NLRP3 suppression in APP/PS1 mice [14, 15], which implies NLRP3 inflammasome plays a critical role in AD pathology via regulating neuroinflammation [16]. There are reasons to believe that alleviating oxidative stress, mitochondrial dysfunction and neuroinflammation has the therapeutic effect of improving cognitive impairment and memory dysfunction.
The brain has high levels of sirtuin expression. Evidence suggests that sirtuins play a significant role in protecting neuronal health while aging, and that their levels change with age and are variably expressed throughout various brain regions [17]. Multiple mechanisms linked with AD pathophysiology, as APP processing, Tau protein aggregation, mitochondrial disorders, neuroinflammation, and oxidative damage, could be modulated by SIRTs [18, 19]. Mitochondrial sirtuin 3 (SIRT3), a mitochondrial deacetylase, is essential for energy balance, mitochondrial biogenesis, and oxidative stress regulation [20]. Downregulation of SIRT3 in AD patients’ brains has been documented [21] and experimental AD mouse models [22]. In addition, studies have also shown that Aβ rises Tau by Mediating Sirtuin 3 in AD [23]. Overexpression of SIRT3 decreased Tau acetylation, whereas downregulation of SIRT3 increased Tau acetylation in hippocampal neurons of mice [24]. Furthermore, reduced SIRT3 levels cause p53-mediated mitochondrial malfunction and brain damage in AD [18].
Kai-Xin-San (KXS), a classical Chinese medicinal formula composed of Panax ginseng C. A. Meyer, *Polygala tenuifolia* Willd., *Acorus tatarinowii* Schott. and Poria Cocos (Schw.) Wolf. Emerging evidence indicated that KXS has a good curative effect on various neuropsychiatric diseases, especially dementia and depression [25]. Previous research has demonstrated that neuronal apoptosis and tau pathology are both ameliorated in elderly SAMP8 mice by KXS treatment [26]. KXS reduces inflammation, oxidative stress, and neuronal degeneration to mitigate doxorubicin-induced cognitive impairment [27]. KXS contributes to the antidepressant effect by suppressing NLRP3 inflammasome activation and enhancing autophagy [28]. Our previous studies indicated that KXS could ameliorate scopolamine-induced cognitive dysfunction [29]. Additionally, we verified the anti-inflammatory properties of KXS in an AD model [30]. However, the underlying mechanisms of KXS for neuroprotection remain largely unknown. We aimed to explore whether KXS improves AD model learning and memory ability via the SIRT3-dependent pathway. This study employed APP/PS1 double transgenic mice as an AD animal model.
## Reagents and antibodies
Amyloid beta (Aβ1–42) peptide was acquired from (GL Biochem, Shanghai, China). Yuanye Biotechnology procured hexafluoroisopropanol (HFIP). Dimethyl Sulfoxide (DMSO) was imported from (Biomedicals, Santa Ana, USA). 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) was obtained from Sigma Aldrich (Saint Louis, MO, USA). Dulbecco’s Modified Eagle Medium (DMEM), Phosphate Buffered Saline (PBS), Dulbecco’s Modified Eagle Medium/Nutrient Mixture F-12 (DMEM/F-12), Fetal Bovine Serum (FBS), penicillin (100 U/ml) and streptomycin (100 μg/ml) was obtained from Gibco BRL, Gaithersburg (MD, USA). Dye 4,6-diamidino-2-phenylindole (DAPI) was obtained from Roche Diagnosis Co., Ltd. (Shanghai, China). Nissl Staining Solution (C0117) was purchased from the Beyotime Institute of Biotechnology (Jiangsu, China). Total Superoxide Dismutase (T-SOD) Assay Kit (A001-1-1), Malondialdehyde (MDA) Assay Kit (A003-1-1), Reactive oxygen species (ROS) Assay Kit (E004-1-1) were acquired from (Jiancheng Bioengineering, Nanjing, China); NAD+/NADH Assay Kit (S0175, Bbeyotime); *Bovine serum* albumin (BSA) was purchased from (Roche, Shanghai, China). Primary antibodies: anti-Synaptophysin (ab8049), anti-PSD95 (ab13552), anti-NGF (ab6199), anti-BDNF (ab108319), anti-p-Drp1(ab193216) were obtained from Abcam, Inc; anti-SIRT3 (bs-6105R) were obtained from BIOSS, Inc; anti-NLRP3 (15101S), anti-IL-18 (DF6252), anti-Caspase-1 (AF5418), anti-Cleaved-Caspase 1 (AF4005) and anti-β-actin (AF7018) were purchased from Affinity, Inc; anti-IL-1β (12242S), anti-Mitofusin-1 [14,739], anti-Mitofusin-2 [9482], anti-Drp1 [8570] and anti-ASC (67824S) were obtained from Cell Signaling Tech., Inc. Secondary antibodies: Goat Anti-Rabbit IgG H&L (HRP) (ab6721), Goat Anti-Mouse IgG H&L (HRP) (ab6789) and Goat Anti-Rabbit IgG H&L (Alexa Fluor® 555) (ab150078) were obtained from Abcam, Inc.
## Preparation of KXS
Panax ginseng C. A. Meyer, *Polygala tenuifolia* Willd., *Acorus tatarinowii* Schott. and Poria Cocos (Schw.) Wolf. ( Lot number: 190827, 20191101, C20053007 and 045200303) were purchased from Yuzhou Kaixuan Pharmaceutical Co., Ltd., and Anguo Runde Pharmaceutical Co., Ltd., respectively. Four raw herbs were mixed in a 3:2:2:3 ratio. According to our previous reports, the extraction of KXS and related quality control methods are prepared. The extracts were concentrated, lyophilized, and stored at − 20 °C (Table 1).Table 1Composition of drugs and their ratio of Kai-Xin-SanHerbal medicine nameLatin nameRatioRen ShenPanax ginseng C. A. Meyer3Yuan ZhiPolygala tenuifolia Willd2Shi Chang PuAcorus tatarinowii Schott2Fu LingPoria Cocos(Schw.) Wolf3
## Animals and drug administration
The Center for Experimental Animals at the Guangzhou University of Chinese Medicine maintained a specified pathogen-free (SPF) environment for male APP/PS1 mice that were 2 months old and age-matched C57BL/6 mice were acquired from the Guangdong Medical Animal Center. The source and specific information of APP/PS1 mice is B6; C3-Tg (APPswe, PSEN1dE9)85Dbo/Mmjax; MMRRC Strain #034829-JAX [31]. All animals were given unrestricted access to food and water, and they were maintained on a light/dark cycle that lasted for 12 h continuously. All experiments were performed with the authorization of the animal ethics committee and according to the Health Guide National Institutes for the Laboratory Animals Care and Use (Bethesda, MD, USA).
APP/PS1 animals were randomly assigned into four groups till all mice reached 9 months of age: APP/PS1 group, APP/PS1 + KXS-L (2.5 g/kg/day) group, APP/PS1 + KXS-M (5 g/kg/day) group and APP/PS1 + KXS-H (10 g/kg/day) group ($$n = 8$$). Two groups of C57BL/6 mice were allocated randomly: Control and Control + KXS-H (10 g/kg/day) groups ($$n = 8$$–10). KXS were given orally once/day for a month, while mice among Control and APP/PS1 group were given oral gavage with saline.
## Morris water maze test
The circular pool, which was 120 cm in diameter, was divided into quarters, with a platform measuring 10 cm in diameter set at the bottom of the fourth region level. The mice were tested with the Morris water maze after being treated for 4 weeks. During the adaptive training phase, a 1-min swimming period was provided for the mice. The mice were trained from four different water inlets every day for the following 5 consecutive days. If they could not find the platform within the 60 s, each mice were instructed to stand on the platform for 20 s. The mice will have 60 s of platform-free swimming time to investigate the pool on day 7. All the experimental results will be documented by the SuperMaze® software.
## Open field test
The open-field test contains a square box (50 × 50 cm) split into 25 squares (16 peripheral and 9 central). The camera system can automatically identify the movement track of the animal and record various activity parameters, including the movement distance of the central area and the time spent in the central area. At the beginning of the experiment, the mice were placed from a corner to the floor. The camera synchronously recorded the movement track and related parameters of the mice within 5 min. Keep the test room quietly. After each mouse was gently placed back in its cage, the chamber was sprayed with $70\%$ alcohol to remove foreign bodies and odors.
## Novel object recognition test
Before the experiment, each mouse was given three minutes to acclimate to the laboratory setting by moving freely in the empty chamber. In the formal experiment, two identical objects were put into the set position in the chamber. Then the mice were respectively put into the chamber from a fixed corner in the test box for 5 min exploration time. The following day, replace one of the pair of duplicates with a new object, leaving the same position. Mice were put into the laboratory box, and each mouse was given an exploration time of 5 min respectively. The total time exploring new and old objects was recorded as New (N) and Familiar (F), respectively. Novel object preference index = N/(N + F) was taken as the index to judge the ability of mice’s episodic memory. Before each mouse learns, it must first wipe the inside and bottom of the chamber with $70\%$ alcohol and remove animal waste and odors.
## Elevated zero maze test
Two open arm-areas and two closed arm-areas comprise the elevated zero maze. The diameter of the outer circle is 50 cm, and the diameter of the inner circle is 45 cm. The maze was 50 cm above the ground. During the experiment, the operator placed the mice in the open-arm area with the head facing the inner ring and recorded the animals’ behavior changes for 5 min with the camera system, including the entry time and entry times of the two arms areas. All four paws entered an arm through the central region before recording the experimental parameters. To prevent the experiment from being influenced by the lingering odour of a prior animal, we removed its faeces and doused the track with $70\%$ ethanol before wiping it dry.
## Animal anesthesia and euthanasia
To minimize pain and humanely sacrifice the mice, sodium pentobarbital (50 mg/kg, IP) was used to anesthetize each subject. Random cervical dislocation was used for half of the mice scarification among each group. The hippocampus was quickly removed on ice, cleaned in PBS buffer, and stored at – 80 °C for further detection. PBS was used to perfuse the other anesthetic mice, and then PBS containing $4\%$ paraformaldehyde has been used (PFA). The brain samples were taken from mice and fixed overnight with $4\%$ PFA in PBS for further analysis.
## Nissl stain
4 μm paraffin-embedded brain samples were sectioned, dewaxed, and hydrated. After that, Nissl solution was used to stain brain slides for 10 min. Finally, the slices were sealed by neutral resin after dehydrating, and dimethyl benzene transparency, and a light microscope equipped with LEICA QWin+ was then utilized to evaluate the slides (Wetzlar, Germany).
## Transmission electron microscopy (TEM)
The hippocampal tissue of the mice that had just been sacrificed was cut into small pieces (1 mm × 1 mm × 1 mm) and fixed at 4 °C in $2.5\%$ glutaraldehyde (PBS buffer). Then hippocampal samples were sent to Guangzhou KingMed Center for Clinical Laboratory Co., Ltd. (Guangzhou, China) for transmission electron microscopy. Mitochondrial ultrastructure and shape were assessed by TEM.
## Measurement of oxidative stress
The level of ROS, MDA along with T-SOD activity, and the ratio of NAD+/NADH were determined as per the related protocols of instructions.
## Immunofluorescence stain
The frozen sections of 20 μm were used for immunofluorescence stain. Tissue sections were placed in $0.5\%$Triton X-100 for permeability at 37 °C for 1 h; $3\%$ catalase was added to the slices to infiltrate the tissues and incubate them for 10 min to eliminate the endogenous oxidase activity of the tissues; $10\%$ goat serum was added to seal at 37 °C for 1 h; primary antibody (SIRT3/NLRP3) was added to the tissues. The slices were stored overnight in a moist box at 4 °C. The fluorescent secondary antibody was incorporated the following day and incubated for 1 h at 37 °C; DAPI dyed the nucleus; A fluorescence microscope (Nikon 80i) was utilized for observation.
## KXS-containing serum preparation
The Animal Center at Guangzhou University provided 20 male SD rats weighing 220 20 g. The experiment was splitted into the normal group ($$n = 10$$) and the KXS-containing serum group ($$n = 10$$). The KXS-containing serum group was given Kai-Xin-San (10 g/kg) twice for 7 consecutive days. Another 10 rats received an identical dose of saline solution. The related serum was centrifuged at 3000×g for 15 min at 4 °C. The upper serum was taken and bathed at 56 °C for 30 min, mixed in the same group. The serum was filtered and inactivated by a 0.22 μm filter and stored at − 20 °C.
## Aβ1–42 preparation and cell culture
1.1 mL HFIP was added to 5 mg Aβ1–42, mixed and stood at 37 °C for 1 h. each 100 μL dissolved Aβ1–42 was added to 22 μL DMSO and 978 μL DMEM/F12 phenol-free red medium. The solution was kept in an incubator at 37 °C for a week. After 7 days, the samples were centrifuged at 14,000×g for 30 min at 4 °C, and the 100 mM Aβ1–42 supernatant was taken. Briefly, HT22 cells were grown in DMEM supplemented with $10\%$ (v/v) FBS, 100 U/ml of penicillin, and 100 g/ml of streptomycin (Gibco BRL, Gaithersburg, MD, USA). At 37 °C and a constant atmosphere with $5\%$ CO2, the cells were kept.
## Cell treatment and cell viability assay
The quantity of crystallization of MTT is inversely proportional to the number of cells within a given concentration range, making the MTT assay a useful indirect indicator of cell viability. HT22 cells were spread in 96-well plate with 5000 cells/well. Following 24 h, Aβ1–42 (0 mM, 1 mM, 5 mM, 10 mM, 20 mM, 40 mM, 50 mM) and Blank/KXS-containing serum ($10\%$, $15\%$, $20\%$, $25\%$, $30\%$) were added. After different intervention times (12 h, 24 h, 36 h, 48 h, 60 h), 10 μL MTT was added to each well for 4 h, 50 μL DMSO were applied to each well. The micro-oscillator vibrated for 10 min so that the light absorption value of each hole was fully determined by the micrometer at the wavelength of 490 nm. The effect of Aβ1–42 and the blank/drug-containing serum itself on the activity of HT22 cells was identified to determine the optimal modeling concentration.
## Western blotting analysis
RIPA buffer lysis was used to collect samples of hippocampus or HT22 cells. After centrifuging the lysate for 15 min at 12,000×g (4 °C), we retrieved the supernatant. To determine the precise amount of proteins, a measurement was taken using a NanoPhotometer® NP80 (Implen, Germany). SDS-PAGE was used for protein separation, and then the proteins were transferred to a PVDF membrane (Millipore, Billerica, MA). Following a 1-h incubation at room temperature with $5\%$ BSA as a blocking agent, an overnight incubation at 4 °C with primary antibodies specific for target proteins, and a final 1-h incubation at RT with the secondary antibody, membranes were probed with antisera. We utilized the ECL + kit (Applygen, Beijing, China) for detection, and the Bio-Rad Image Lab 5.2.1 software for quantification (Ca, USA).
## Transfection of SIRT3 siRNA
SIRT3 siRNA and negative control siRNA (NC siRNA) sequence was purchased from (Suzhou GenePharma Co., Ltd). Fresh media was plated in six-well plates containing HT22 cells. SIRT3 siRNA or NC siRNA transfection was conducted using the Lipofectamine 3000 kit (Invitrogen, Carlsbad, CA, USA). After a day of incubation for serum-free culture, cells were supplemented with 10 mM Aβ1–42 and $20\%$Blank/KXS-containing serum for 24 h.
## Statistical analysis
We performed our statistical analyses in SPSS 19.0 and GraphPad Prism 5. We utilized one-way analysis of variance with post hoc to analyze the data, and we defined statistical significance at $P \leq 0.05.$ For the statistical description of measurement data, we used mean ± SEM for data that followed a normal distribution, and we used M (P25, P75) for data that did not.
## KXS protects against cognitive deficit in APP/PS1 mice
APP/PS1 mice were tested utilizing Morris water maze, open field, new object recognition, and raised O maze to determine the influence of KXS on their learning and memory abilities. Figure 1A showed the experimental schedule on animals. As shown in Fig. 1B, KXS-H (10 g/kg/days) exhibited no discernible impact on C57BL/6 mice’s learning and memory abilities throughout the course of 5 days of escape latency. The ability of 10-month-old APP/PS1 mice to learn and memory was significantly reduced in comparison to control group. The learning and memory abilities of APP/PS1 mice treated with varying dosages of KXS was significantly enhanced and the period required to locate the platform decreased as training days increased ($P \leq 0.05$, $P \leq 0.001$). In spatial probe test, APP/PS1 mice treated with KXS-H spent significantly more time on the target platform and crossed it more frequently than untreated mice ($P \leq 0.05$) (Fig. 1C–E). The swimming speed of APP/PS1 negative mice was slightly faster than APP/PS1 mice, but the difference was not statistically significant ($P \leq 0.05$) (Fig. 1F). KXS-H (10 g/kg/day) exhibited no significant influence on the behavioral performance of APP/PS1-negative mice in the open field during the new object recognition and raised O maze tests. Compared with C57BL/6 mice, the central zone distance traveled of 10-month-old APP/PS1 mice (Fig. 2A, B), the new object recognition index (Fig. 2C, D) significantly reduced ($P \leq 0.05$, $P \leq 0.05$), although open arm entries number is a downward trend, there is no statistical difference ($P \leq 0.05$) (Fig. 2E, F). However, APP/PS1 mice with different doses of KXS showed various degrees of improvement in related behavioral indicators, especially in the treating with KXS-H of the new object recognition index ($P \leq 0.05$).Fig. 1Kai-Xin-San (KXS) protects against cognitive deficit in APP/PS1 mice. Morris water maze test. A Experiment schedule; B Escape latency training for 5 consecutive days A in PP/PS1 mice; C Representative Diagram of swimming paths in the spatial probe trial; D Tim spent in the target quadrants; E Cross time of the platform location; F Swimming speed. KXS-L: 2.5 g/kg/day; KXS-M: 5 g/kg/day; KXS-H: 10 g/kg/day. The experiment data are expressed as means ± SEM. $$n = 8$$ in each group. # $P \leq 0.05$, ##$P \leq 0.01$ vs. Ctrl group; ∗$P \leq 0.05$ vs. APP/PS1 groupFig. 2KXS protects against cognitive deficits and anxiety states in APP/PS1 mice. A The moving tracks of the Open Filed test; B The central zone distance traveled in the Open Filed test; C The moving tracks of the New Object Recognition; D Novel object preference index; E The moving tracks of the Elevated O Maze test; F The number of open arm entries in the Elevated O Maze. KXS-L: 2.5 g/kg/day; KXS-M: 5 g/kg/day; KXS-H: 10 g/kg/day. The experiment data are expressed as means ± SEM. $$n = 8$$ in each group. # $P \leq 0.05$ vs. Ctrl group; *$P \leq 0.05$ vs. APP/PS1 group
## KXS suppresses neuronal apoptosis and oxidative stress in APP/PS1 mice
Nissl staining revealed that KXS-H dosage (10 g/kg/day) administered to C57BL/6 mice in hippocampus CA1 and CA3 regions did not have a significant impact on the number of healthy neurons and neurons that had survived. The number of healthy and surviving neurons in hippocampal CA1 and CA3 regions of 10-month-old APP/PS1 mice significantly reduced, that resulted in typical neuropathy, involving loss of Nissl body and nuclear disappearance in comparison to C57BL/6 mice. However, KXS-H (10 g/kg/day) improved hippocampal neurons’ survival and prevented the loss of neurons in CA1 and CA3 areas and the Nissl bodies in APP/PS1 mice (Fig. 3, Additional file 1: Figure S1). The findings revealed that KXS could prevent hippocampal neurons apoptosis in APP/PS1 mice. Furthermore, to evaluate the improvement influence of KXS on oxidative stress in APP/PS1 mice, ROS, T-SOD, MDA and the ratio of NAD+/NADH were used to detect the hippocampal tissues of APP/PS1 mice. As displayed in Fig. 4A–D, ROS and MDA levels in APP/PS1 mice were significantly higher than those in C57BL/6 mice, whereas T-SOD activity and the ratio of NAD+/NADH were significantly diminished. The findings suggested that KXS-H could decrease ROS and MDA level and raise T-SOD activity and the ratio of NAD+/NADH in hippocampus of APP/PS1 mice. Fig. 3Representative Nissl staining of hippocampal CA1 and CA3 regions of APP/PS1 mice. $$n = 3$$ in each groupFig. 4KXS decreases oxidative stress and mitochondrial morphology in hippocampus of APP/PS1 mice. A ROS level; B SOD activity; C MDA level; D NAD+/NADH; E Representative transmission electron microscope (TEM) photos. KXS-H: 10 g/kg/day. The experiment data are expressed as means ± SEM. $$n = 3$$ in each group. # $P \leq 0.05$, ##$P \leq 0.01$, ###$P \leq 0.001$ vs. Ctrl group; *$P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$ vs. APP/PS1 group
## KXS improves mitochondrial dysfunction in the hippocampus of APP/PS1 mice
To investigate KXS impacts on mitochondrial morphology of hippocampal neurons in APP/PS1 mice, we conducted TEM to observe mitochondrial morphology (Fig. 4E). The results of TEM exposed that the mitochondrial size of hippocampal neurons in C57BL/6 mice was basically normal, the stromal was uniform, the boundary and structure were clear, the mitochondrial cristae were dense and full, in the shape of a short stick or dumbbell, and no obvious mitochondrial swelling or vacuolar changes were detected. The hippocampus neuron mitochondria of C57BL/6 mice after KXS-H (10 g/kg/day) did not differ significantly from those of control group. In APP/PS1 mice group, the mitochondria of hippocampal neurons were swollen; most of them had swelling and vacuole-like degeneration, the matrix was uneven, the structure was fuzzy, and the swollen mitochondrial cristae had obvious rupture or even swelling rupture. The mitochondria morphology improved after the treatment of KXS-H, and the number of swollen or vacuolated mitochondria was significantly reduced. Additionally, we discovered by Western blot that mitochondrial division and fusion balance was disrupted in 10-month-old APP/PS1 mice KXS-H upregulated the expression of mitochondrial fusion proteins Mfn1 and Mfn2 and downregulated the mitochondrial mitogens Drp1 and p-Drp1, restoring the integrity of the mitochondrial structure (Fig. 5A, C–E). Therefore, we postulated that KXS-H alleviate mitochondrial dysfunction and thereby avoid cognitive impairment in APP/PS1 mice. Fig. 5KXS decreases the expression of mitochondrial fission and fusion-related proteins and increases the expression of neurotrophic factors and synaptic proteins in the hippocampus of APP/PS1 mice. A Expression of Mfn1, Mfn2, p-Drp1 and Drp1; B Expression of BDNF, NGF, SYN and PSD95; C–E Histogram of relative expression of Mfn1, Mfn2, p-Drp1 and Drp1; F–I Histogram of relative expression of BDNF, NGF, SYN and PSD95. KXS-H: 10 g/kg/day. The experiment data are expressed as means ± SEM. $$n = 3$$ in each group. # $P \leq 0.05$, ##$P \leq 0.01$, ###$P \leq 0.001$ vs. Ctrl group; *$P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$ vs. APP/PS1 group
## KXS increases the expression of synaptic and neurotrophic factors in the hippocampus of APP/PS1 mice
We identified lower levels of neurotrophic factor (BDNF, NGF, SYN) and synaptic protein (PSD95) in APP/PS1 mice aged 10 months compared to C57BL/6 mice. Nevertheless, the treatment of APP/PS1 mice with KXS-H elevated neurotrophic factor expressions (BDNF, NGF, SYN) and synaptic protein (PSD95) (Fig. 5B, F–I). The results showed that KXS-H had an obvious neuroprotective effect.
## KXS increases the expression of mitochondrial SIRT3 and reduces neuroinflammation in the hippocampus of APP/PS1 mice
Western blotting (Fig. 6A–G) and immunofluorescence (Fig. 6H–I, Additional file 1: Figure S2) results showed that 10-month-old APP/PS1 mice exposed down-regulated SIRT3 protein expression and up-regulated NLRP3 inflammasome-associated proteins (NLRP3, ASC, pro-Caspase1, cleaved-Caspase1, IL-1β, IL-18), compared with C57BL/6 mice. After KXS-H administration, SIRT3 expression in hippocampus of APP/PS1 mice was up-regulated, while NLRP3 inflammasome-associated proteins expression was down-regulated. These results indicated that KXS-H might improve learning and memory impairment by modifying SIRT3/NLRP3 expression in the APP/PS1 mice hippocampal. Fig. 6KXS up-regulated the expression of SIRT3 proteins and down-regulated the expression of neurotrophic NLRP3 inflammasome-related proteins in the hippocampus of APP/PS1 mice. A Expression of SIRT3 and NLRP3 inflammasome-related proteins; B Histogram of relative expression of SIRT3 proteins; C–G Histogram of relative expression of NLRP3 inflammasome-related proteins; H Representative immunofluorescent imaging of SIRT3 (red) and DPAI (blue); I Representative immunofluorescent imaging of NLRP3 (red) and DPAI (blue). KXS-H: 10 g/kg/day. The experiment data are expressed as means ± SEM. $$n = 3$$ in each group. # $P \leq 0.05$, ##$P \leq 0.01$, ###$P \leq 0.001$ vs. Ctrl group; *$P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$ vs. APP/PS1 group
## KXS-containing serum attenuates Aβ1–42-induced cytotoxicity in HT22 cells
We evaluated the cytotoxicity of KXS-containing serum on Aβ-induced HT22 cells by morphological observation and MTT assay. Preliminary experiments showed that the activity of Aβ1–42-induced HT22 cells reduced in a concentration-dependent manner within 1–50 mM and 12–60 h of HT22 cells, and the 10 mM of Aβ-treated HT22 cells showed moderate damage to HT22 cells within 24 h (Additional file 1: Figure S3). Hence, this concentration was utilized in all subsequent experiments. Additionally, 12–60 h of treatment with different doses of KXS-containing serum lowered the activity of HT22 cells in a concentration-dependent manner, and all HT22 cells were moderately damaged by $20\%$ KXS-containing serum (Additional file 1: Figure S4). Therefore, this concentration was used in all further experiments. As shown in Fig. 7A, cell morphology observation showed that $20\%$ KXS-containing serum prevented the death of HT22 cells and reversed the morphological changes of HT22 cells, including cell shape treated with 10 mM Aβ1–42. MTT assay results showed that after 24 h of treatment with 10 mM Aβ1–42, the activity of HT22 cells decreased, and $20\%$ KXS-containing serum m increased the survival rate of HT22 cells (Additional file 1: Figure S5).Fig. 7The effect of KXS-containing serum in morphological changes, mitochondrial fission and fusion-related proteins, neurotrophic factors and synaptic-related proteins and SIRT3/NLRP3 pathway in HT22 cells. A Morphological changes in HT22 cells treated with Aβ and KXS-containing serum. B Expression of Mfn1, Mfn2, p-Drp1 and Drp1; C–E Histogram of relative expression of Mfn1, Mfn2, p-Drp1/Drp1; F Expression of BDNF, NGF, SYN and PSD95; G–J Histogram of relative expression of BDNF, NGF, SYN and PSD95. K Expression of SIRT3 and NLRP3 inflammasome-related proteins; B Histogram of relative expression of SIRT3 proteins; L–Q Histogram of relative expression of NLRP3 inflammasome-related proteins. KXS-H: 10 g/kg/day. The experiment data are expressed as means ± SEM. $$n = 3$$ in each group. # $P \leq 0.05$, ##$P \leq 0.01$, ###$P \leq 0.001$ vs. Ctrl group; *$P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$ vs. APP/PS1 group
## KXS-containing serum mitochondrial dysfunction and the expression of synaptic and neurotrophic factors
We found that $20\%$ blank serum and $20\%$ KXS-containing serum had no significant influence on mitochondrial division and fusion balance of HT22 cells by western blotting. The mitochondrial division and fusion balance of HT22 cells by 10 mM Aβ1–42-induced was destroyed, and the activity of the HT22 cells was impaired. The $20\%$ KXS-containing serum can upregulate Aβ1–42-induced mitochondrial fusion proteins Mfn1 and Mfn2 expression and down-regulate the mitochondrial mitogenic proteins Drp1 and p-Drp1 (Fig. 7B–E). In addition, $20\%$ blank serum and $20\%$ KXS-containing serum did not significantly affect neurotrophic factor and synaptic protein expression in HT22 cells. The expressions of 10 mM Aβ1–42-induced HT22 cells neurotrophic factor (BDNF, NGF, SYN) and synaptic protein (PSD95) were significantly decreased. However, the expressions of Aβ1–42-induced HT22 cell neurotrophic factor (BDNF, NGF, SYN) and synaptic protein (PSD95) significantly increased after administering $20\%$ KXS-containing serum (Fig. 7F–J). The results showed that $20\%$ KXS-containing serum had an obvious neuroprotective effect on Aβ1–42-induced HT22 cells.
## SIRT3 is involved in the anti-neuroinflammation effect of KXS-containing serum on HT22 cells
We studied $20\%$ KXS-containing serum impact on 10 mM Aβ1–42-induced HT22 cells in the SIRT3/NLRP3 pathway. Compared with the $20\%$ blank serum or $20\%$ KXS-containing serum, the expression of SIRT3 protein in 10 mM Aβ1–42-induced HT22 cells was down-regulated, and NLRP3 inflammasome-associated proteins (NLRP3, ASC, pro-Caspase1, cleaved-Caspase1, IL-1β, IL-18) expression was up-regulated. SIRT3 protein expression was up-regulated and NLRP3 inflammasome-associated protein expression was down-regulated in 10 mM A1–42-induced HT22 cells following treatment with $20\%$ KXS-containing serum (Fig. 7K–Q). To examine SIRT3 function in Aβ-induced HT22 cells treated with $20\%$ serum containing KXS. The indicated plasmids were transfected into HT22 cells, which make the SIRT3 expression was reduced in HT22 cells (Fig. 8A, C). Western blot showed that $20\%$ KXS-containing serum reduced NLRP3 inflammasome activation in HT22 cells. In addition, SIRT3 siRNA transfection further reversed the upregulation of NLRP3 inflammasome treated with $20\%$ KXS-containing serum in HT22 cells (Fig. 8B, D, E). These results indicate that $20\%$ KXS-containing serum plays an anti-neuroinflammatory function by activating SIRT3.Fig. 8SIRT3 is involved in the antioxidant effect of KXS-containing serum on HT22 cells. A The expression of SIRT3; B The expression of NLRP3, ASC, Pro-Caspase1 and Cleaved-Caspase1; C Histogram of relative expression of SIRT3 proteins; D–F Histogram of relative expression of NLRP3 inflammasome-related proteins. The experiment data are expressed as means ± SEM. $$n = 3$$ in each group. ### $P \leq 0.001$ vs. Ctrl group; **$P \leq 0.01$, ***$P \leq 0.001$
## Discussion
The effects of the Chinese herb Kai-Xin-San on the brain’s memory and learning processes in APP/PS1 double transgenic mice were studied. KXS could improve learning and memory dysfunction in 10-month-old APP/PS1 mice, according to the study’s behavioral tests. Simultaneously, KXS possessed he functions of anti-neuronal apoptosis, anti-oxidant stress, and improvement of mitochondrial dynamics and morphological malfunction properties. Furthermore, KXS could increase the expression of SIRT3 protein and decrease NLRP3 inflammasome-related protein expression in hippocampus. These results suggested that KXS has promise as a treatment for AD.
Cognitive deficits are associated with progressive neuronal apoptosis and synaptic loss in AD [32, 33]. Hippocampal neurogenesis is crucial for cognition, and BDNF and NGF are involved in the neurogenesis process [34]. The consistent expression of synaptic proteins like PSD95 and SYN is necessary for appropriate synaptic function [35]. Enhancing synapse-associated proteins expression has the ability to sustain synaptic connections, which could alleviate memory and cognitive impairment [36]. The results of the Nissl stain and Western blot suggested that KXS could improve hippocampal neurons apoptosis in APP/PS1 mice, by increasing synaptic connections, and upregulating neurotrophic factor levels. The term “oxidative stress” refers to the imbalance between free radical and antioxidant production. Since neurons in the brain generate energy and consume oxygen at rapid rates, they are particularly vulnerable to excessive ROS production and oxidative stress. As neurons are so easily damaged by free radicals, abnormal ROS production and signaling can have far-reaching effects on brain function and behavior [37]. Several studies have provided evidence that oxidative stress is early pathogenesis of AD, which leads to neuronal apoptosis and synaptic loss [38]. Furthermore, oxidative stress can be caused by the free radicals produced by Aβ and tau clumps in AD patients and animal models [39, 40]. Numerous ROS scavengers that could ameliorate oxidative stress-mediated synaptic and cognitive performance in AD patients and animal models have been discovered, given the importance of ROS to the etiology of Alzheimer’s disease [41–43]. Thus, antioxidant treatment is considered a potentially useful strategy for AD prevention and therapeutic management. Our study demonstrated that KXS might enhance SOD activity and reduce ROS and MDA levels. The results suggested that KXS could reduce the ROS level, protecting the function of neuronal cells from the damage caused by excessive ROS.
The activation of neuronal apoptotic cascade, cognitive problems, and behavior functions may be exacerbated by persistent oxidative stress and mitochondrial malfunction. Mitochondria fission and fusion proteins are principally responsible for controlling the dynamics of mitochondria [44]. A variety of human illnesses, as AD and other neurodegenerative diseases, are related with abnormal mitochondrial dynamics and the resulting abnormalities in mitochondrial organization [45]. The fission and fusion mechanisms that directly alter the structure and morphology of mitochondria are the driving behind mitochondrial dynamics. As a result, the structure and function of mitochondria are inextricably intertwined [46]. Large GTPases mediate mitochondrial fission and fusion, with Drp1 being responsible for controlling mitochondrial fission and mitofusin (MFN) being responsible for regulating mitochondrial fusion [47]. There is a noticeable disruption in the balance between mitochondrial fusion and fission in AD neurons. Excessive mitochondrial fission, such as Mfn1, Mfn2, and OPA1, is responsible for impaired mitochondrial function and neuronal death in AD [48]. Downregulation of Drp1 expression decreases mitochondrial fragmentation and improves mitochondrial fusion and mitochondrial function in AD neurons [49]. Our study found that KXS could improve the swelling and rupture of mitochondria in the hippocampus of APP/PS1 mice by blocking excessive mitochondrial fission Drp1 or recruiting fusion MFN$\frac{1}{2.}$ The results indicated that KXS could improve mitochondrial morphological abnormalities and mitochondrial dysfunction in AD models.
It has been demonstrated that oxidative stress and mitochondrial dysfunction are the triggers that activate NLRP3 inflammasomes [10, 50]. Thioredoxin is liberated from TXNIP when oxidative stress rises, making it available to connect with the NLRP3 inflammasome [51, 52]. It has also been proposed that signals originating from mitochondrial malfunction, such as reactive oxygen species (mtROS), oxidized mitochondrial DNA (mtDNA), or the externalization of the phospholipid cardiolipin, induce NLRP3 activation [53]. NLRP3 inflammasomes are essential molecules in neuroinflammation, and Aβ was responsible for AD pathogenesis in AD models [54]. Neuroinflammation is postulated to be a critical component of AD pathogenesis. NLRP3 inflammasome is activated in AD and mild cognitive impairment (MCI) brains and APP/PS1 mice [13]. NLRP3 inflammasome activation leads to Caspase-1-mediated cleavage of the pro-inflammatory cytokines IL-1 and IL-18, which enhance neuroinflammation. NLRP3 deficiency or NLRP3 inhibitor MCC950 substantially attenuates AD phenotypes, involving spatial memory loss, in aged APP/PS1 mice [55, 56]. Interestingly, our experiments demonstrated that KXS could alleviate neuroinflammation mediated by NLRP3 inflammasome activation which improved cognitive impairment in APP/PS1 mice.
Our results shown that KXS can upregulate SIRT3 in APP/PS1 mice hippocampus. SIRT3 is a mitochondrial nicotinamide adenine dinucleotide (NAD+)-dependent deacetylase. SIRT3 is an essential part of different physiological and pathological processes, including energy metabolism [57, 58], neuronal death [59], oxidative stress [60], mitochondrial dysfunction [61], inflammatory [62] and so on. In AD mice models and post-mortem AD patient brains, changes in SIRT3 levels have been detected [22].
Knockout of SIRT3 significantly exacerbated hippocampal neuron death by ROS [63]. SIRT3 overexpression could ameliorate mitochondrial ROS levels and cell apoptosis, which improves ROS detoxification through activation of MnSOD and catalase. Nicotinamide riboside, a SIRT3 agonist, protects transgenic mouse models of AD against cognitive impairment, synaptic degeneration, and neuronal death [64, 65]. Additionally, a lower SIRT3 protein was associated with mitochondrial dysfunction in AD brain [66]. It has been reported that SIRT3 activities were ameliorated by melatonin and icariin administration, resulting in improved mitochondrial function [67]. In addition, SIRT3 inhibits neuroinflammation by removing damaged pro-inflammatory mitochondria and inhibiting the NLRP3 inflammasome [16]. Activating SIRT3 regulation of mitochondrial dysfunction and neuroinflammation leads to ameliorating cognitive decline in mice [21]. *In* general, our results indicated that KXS has the pharmacological effect of up-regulating the expression of SIRT3 protein. Up-regulating SIRT3 may play a role of anti-oxidative stress, anti-mitochondrial dysfunction, and anti-neuroinflammation, relieving the hippocampal neurons apoptosis and improving cognitive impairment.
To more explore the involvement of SIRT3 in KXS neuroprotective impacts, KXS-containing serum and SIRT3 siRNA were used for cell intervention in our study. In vitro, we also confirmed that KXS-containing serum protects neurotrophic factors and improves mitochondrial dysfunction in the Aβ1–42-induced HT22 cells. Furthermore, transfection with SIRT3 siRNA resulted in the reversal of all NLRP3 inflammasome-related proteins. These results indicate that KXS-containing serum may be a probable SIRT3 activator that can preserve neurons from oxidative stress, mitochondrial dysfunction and neuroinflammation mediated neuronal apoptosis.
## Conclusions
In conclusion, we found that KXS had neuroprotective characteristics, as it prevented cognitive deficits in APP/PS1 mice. KXS impact on mitochondrial dysfunction and oxidative stress-mediated neuronal cell apoptosis is associated with SIRT3/NLRP3 pathway regulation. KXS has neuroprotective properties that may make it a useful therapeutic Chinese medicine for treating AD (Fig. 9). However, KXS contains several components with complicated network regulatory mechanisms in AD, and the precise molecular targets and processes of KXS in AD have yet to be understood. Fig. 9Proposed mechanism of KXS improving cognitive ability in APP/PS1 mice
## Supplementary Information
Additional file 1: Figure S1. The histogram of the number of survival neurons in hippocampal CA1 and CA3 regions of APP/PS1 mice. KXS-H: 10g/kg/day. The experiment data are expressed as means ± SEM. $$n = 3$$ in each group. ### $P \leq 0.001$ vs. Ctrl group; ∗$P \leq 0.05$ vs. APP/PS1 group. Figure S2. The histogram of immunofluorescent about SIRT3 and NLRP3 in the hippocampus of APP/PS1 mice. $$n = 3$$ in each group. ### $P \leq 0.001$ vs. Ctrl group; ∗$P \leq 0.05$ vs. APP/PS1 group. Figure S3. The viability of HT22 cells in different concertation Aβ and at different times. $$n = 3$$ in each group. * $P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$ vs. Ctrl group; Figure S4. The viability of HT22 cells in different concertation serum (blank serum or KXS-containing serum) and at different times. $$n = 3$$ in each group.*$P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$, ###$P \leq 0.001.$ Figure S5. The viability of HT22 cells in $20\%$ KXS-concertation serum or $20\%$ blank serum and 10 mM Aβ. $$n = 3$$ in each group. ### $P \leq 0.001$ vs. Ctrl group; ∗∗∗$P \leq 0.0001$ vs. APP/PS1 group.
## References
1. Blanchard JW, Victor MB, Tsai LH. **Dissecting the complexities of Alzheimer disease with in vitro models of the human brain**. *Nat Rev Neurol* (2022) **18** 25-39. DOI: 10.1038/s41582-021-00578-6
2. Koutsodendris N, Nelson MR, Rao A. **Apolipoprotein E and Alzheimer's disease: findings, hypotheses, and potential mechanisms**. *Annu Rev Pathol* (2022) **17** 73-99. DOI: 10.1146/annurev-pathmechdis-030421-112756
3. Ryder MI, Xenoudi P. **Alzheimer disease and the periodontal patient: new insights, connections, and therapies**. *Periodontol 2000* (2021) **87** 32-42. DOI: 10.1111/prd.12389
4. Boopathi S, Poma AB, Garduño-Juárez R. **An Overview of Several inhibitors for Alzheimer's disease: characterization and failure**. *Int J Mol Sci* (2021) **22** 10798. DOI: 10.3390/ijms221910798
5. Hambright WS, Fonseca RS, Chen L. **Ablation of ferroptosis regulator glutathione peroxidase 4 in forebrain neurons promotes cognitive impairment and neurodegeneration**. *Redox Biol* (2017) **12** 8-17. DOI: 10.1016/j.redox.2017.01.021
6. Calvo-Rodriguez M, Hou SS, Snyder AC. **Increased mitochondrial calcium levels associated with neuronal death in a mouse model of Alzheimer's disease**. *Nat Commun* (2020) **11** 2146. DOI: 10.1038/s41467-020-16074-2
7. Simon DW, McGeachy MJ, Bayır H. **The far-reaching scope of neuroinflammation after traumatic brain injury**. *Nat Rev Neurol* (2017) **13** 171-191. DOI: 10.1038/nrneurol.2017.13
8. Tyrrell DJ, Goldstein DR. **Ageing and atherosclerosis: vascular intrinsic and extrinsic factors and potential role of IL-6**. *Nat Rev Cardiol* (2021) **18** 58-68. DOI: 10.1038/s41569-020-0431-7
9. Mattson MP, Arumugam TV. **Hallmarks of brain aging: adaptive and pathological modification by metabolic states**. *Cell Metab* (2018) **27** 1176-1199. DOI: 10.1016/j.cmet.2018.05.011
10. Sliter DA, Martinez J, Hao L. **Parkin and PINK1 mitigate STING-induced inflammation**. *Nature* (2018) **561** 258-262. DOI: 10.1038/s41586-018-0448-9
11. Hooftman A, Angiari S, Hester S. **The immunomodulatory metabolite itaconate modifies NLRP3 and inhibits inflammasome activation**. *Cell Metab* (2020) **32** 468-478.e7. DOI: 10.1016/j.cmet.2020.07.016
12. Sun N, Youle RJ, Finkel T. **The mitochondrial basis of aging**. *Mol Cell* (2016) **61** 654-666. DOI: 10.1016/j.molcel.2016.01.028
13. Long JM, Holtzman DM. **Alzheimer disease: an update on pathobiology and treatment strategies**. *Cell* (2019) **179** 312-339. DOI: 10.1016/j.cell.2019.09.001
14. Tang L, Zhou F. **Inflammasomes in common immune-related skin diseases**. *Front Immunol* (2020) **11** 882. DOI: 10.3389/fimmu.2020.00882
15. Siew JJ, Chen HM, Chen HY. **Galectin-3 is required for the microglia-mediated brain inflammation in a model of Huntington's disease**. *Nat Commun* (2019) **10** 3473. DOI: 10.1038/s41467-019-11441-0
16. Lautrup S, Sinclair DA, Mattson MP. **NAD(+) in brain aging and neurodegenerative disorders**. *Cell Metab* (2019) **30** 630-655. DOI: 10.1016/j.cmet.2019.09.001
17. Herskovits AZ, Guarente L. **SIRT1 in neurodevelopment and brain senescence**. *Neuron* (2014) **81** 471-483. DOI: 10.1016/j.neuron.2014.01.028
18. Lee J, Kim Y, Liu T. **SIRT3 deregulation is linked to mitochondrial dysfunction in Alzheimer's disease**. *Aging Cell* (2018) **17** 12679. DOI: 10.1111/acel.12679
19. Mohamad Nasir NF, Zainuddin A, Shamsuddin S. **Emerging roles of Sirtuin 6 in Alzheimer's disease**. *J Mol Neurosci* (2018) **64** 157-161. DOI: 10.1007/s12031-017-1005-y
20. Pereira AS, Gouveia AM, Tomada N. **Cumulative effect of cardiovascular risk factors on regulation of AMPK/SIRT1-PGC-1α-SIRT3 pathway in the human erectile tissue**. *Oxid Med Cell Longev* (2020) **2020** 1525949. DOI: 10.1155/2020/1525949
21. Li Y, Lu J, Cao X. **A newly synthesized rhamnoside derivative alleviates Alzheimer's amyloid-β-induced oxidative stress, mitochondrial dysfunction, and cell senescence through upregulating SIRT3**. *Oxid Med Cell Longev* (2020) **2020** 7698560. PMID: 32104538
22. Zhou ZD, Tan EK. **Oxidized nicotinamide adenine dinucleotide-dependent mitochondrial deacetylase sirtuin-3 as a potential therapeutic target of Parkinson's disease**. *Ageing Res Rev* (2020) **62** 101107. DOI: 10.1016/j.arr.2020.101107
23. Yin J, Han P, Song M. **Amyloid-β increases tau by mediating Sirtuin 3 in Alzheimer's disease**. *Mol Neurobiol* (2018) **55** 8592-8601. DOI: 10.1007/s12035-018-0977-0
24. Li S, Yin J, Nielsen M. **Sirtuin 3 mediates Tau deacetylation**. *J Alzheimers Dis* (2019) **69** 355-362. DOI: 10.3233/JAD-190014
25. Fu H, Xu Z, Zhang XL. **Kaixinsan, a well-known Chinese herbal prescription, for Alzheimer's disease and depression: a preclinical systematic review**. *Front Neurosci* (2019) **13** 1421. DOI: 10.3389/fnins.2019.01421
26. Jiao YN, Zhang JS, Qiao WJ. **Kai-Xin-San inhibits tau pathology and neuronal apoptosis in aged SAMP8 mice**. *Mol Neurobiol* (2022) **59** 3294. DOI: 10.1007/s12035-021-02626-0
27. Lyu W, Ouyang M, Ma X. **Kai-Xin-San attenuates doxorubicin-induced cognitive impairment by reducing inflammation, oxidative stress, and neural degeneration in 4T1 breast cancer mice**. *Evid Based Complement Alternat Med* (2021) **2021** 5521739. DOI: 10.1155/2021/5521739
28. Yu S, Liu S, Wang N. **Novel insights into antidepressant mechanism of Kai Xin San formula: Inhibiting NLRP3 inflammasome activation by promoting autophagy**. *Phytomedicine* (2021) **93** 153792. DOI: 10.1016/j.phymed.2021.153792
29. Xu YM, Wang XC, Xu TT. **Kai Xin San ameliorates scopolamine-induced cognitive dysfunction**. *Neural Regen Res* (2019) **14** 794-804. DOI: 10.4103/1673-5374.249227
30. Luo Y, Li D, Liao Y. **Systems pharmacology approach to investigate the mechanism of Kai-Xin-San in Alzheimer's disease**. *Front Pharmacol* (2020) **11** 381. DOI: 10.3389/fphar.2020.00381
31. Reiserer RS, Harrison FE, Syverud DC. **Impaired spatial learning in the APPSwe + PSEN1DeltaE9 bigenic mouse model of Alzheimer's disease**. *Genes Brain Behav* (2007) **6** 54-65. DOI: 10.1111/j.1601-183X.2006.00221.x
32. Yu Q, Wang Y, Du F. **Overexpression of endophilin A1 exacerbates synaptic alterations in a mouse model of Alzheimer's disease**. *Nat Commun* (2018) **9** 2968. DOI: 10.1038/s41467-018-04389-0
33. Chen MK, Mecca AP, Naganawa M. **Assessing synaptic density in Alzheimer disease with synaptic vesicle glycoprotein 2A positron emission tomographic imaging**. *JAMA Neurol* (2018) **75** 1215-1224. DOI: 10.1001/jamaneurol.2018.1836
34. Zhu F, Cizeron M, Qiu Z. **Architecture of the mouse brain synaptome**. *Neuron* (2018) **99** 781-799.e10. DOI: 10.1016/j.neuron.2018.07.007
35. Wei Z, Mahaman YAR, Zhu F. **GSK-3β and ERK1/2 incongruously act in tau hyperphosphorylation in SPS-induced PTSD rats**. *Aging (Albany NY)* (2019) **11** 7978-7995. DOI: 10.18632/aging.102303
36. Zhang W, Wu Q, Lu YL. **Protective effects of**. *Neural Regen Res* (2017) **12** 1131-1136. DOI: 10.4103/1673-5374.211193
37. Adeluyi A, Guerin L, Fisher ML. **Microglia morphology and proinflammatory signaling in the nucleus accumbens during nicotine withdrawal**. *Sci Adv* (2019) **5** eaax7031. DOI: 10.1126/sciadv.aax7031
38. Jadiya P, Kolmetzky DW, Tomar D. **Impaired mitochondrial calcium efflux contributes to disease progression in models of Alzheimer's disease**. *Nat Commun* (2019) **10** 3885. DOI: 10.1038/s41467-019-11813-6
39. Ballance WC, Qin EC, Chung HJ. **Reactive oxygen species-responsive drug delivery systems for the treatment of neurodegenerative diseases**. *Biomaterials* (2019) **217** 119292. DOI: 10.1016/j.biomaterials.2019.119292
40. Picón-Pagès P, Gutiérrez DA, Barranco-Almohalla A. **Amyloid beta-peptide increases BACE1 translation through the phosphorylation of the eukaryotic initiation factor-2α**. *Oxid Med Cell Longev* (2020) **2020** 2739459. DOI: 10.1155/2020/2739459
41. Ryu JK, Rafalski VA, Meyer-Franke A. **Fibrin-targeting immunotherapy protects against neuroinflammation and neurodegeneration**. *Nat Immunol* (2018) **19** 1212-1223. DOI: 10.1038/s41590-018-0232-x
42. Young ML, Franklin JL. **The mitochondria-targeted antioxidant MitoQ inhibits memory loss, neuropathology, and extends lifespan in aged 3xTg-AD mice**. *Mol Cell Neurosci* (2019) **101** 103409. DOI: 10.1016/j.mcn.2019.103409
43. Saini JS, Corneo B, Miller JD. **Nicotinamide ameliorates disease phenotypes in a human iPSC model of age-related macular degeneration**. *Cell Stem Cell* (2017) **20** 635-647.e7. DOI: 10.1016/j.stem.2016.12.015
44. Kim MH, Lee HJ, Lee SR. **Peroxiredoxin 5 inhibits glutamate-induced neuronal cell death through the regulation of calcineurin-dependent mitochondrial dynamics in HT22 cells**. *Mol Cell Biol* (2019) **39** e00148-19. DOI: 10.1128/MCB.00148-19
45. Pouli D, Balu M, Alonzo CA. **Imaging mitochondrial dynamics in human skin reveals depth-dependent hypoxia and malignant potential for diagnosis**. *Sci Transl Med* (2016) **8** 367ra169. DOI: 10.1126/scitranslmed.aag2202
46. Ruiz K, Thaker TM, Agnew C. **Functional role of PGAM5 multimeric assemblies and their polymerization into filaments**. *Nat Commun* (2019) **10** 531. DOI: 10.1038/s41467-019-08393-w
47. Song W, Chen J, Petrilli A. **Mutant huntingtin binds the mitochondrial fission GTPase dynamin-related protein-1 and increases its enzymatic activity**. *Nat Med* (2011) **17** 377-382. DOI: 10.1038/nm.2313
48. Gong M, Yuan M, Meng L. **Wenxin Keli regulates mitochondrial oxidative stress and homeostasis and improves atrial remodeling in diabetic rats**. *Oxid Med Cell Longev* (2020) **2020** 2468031. DOI: 10.1155/2020/2468031
49. Pradeepkiran JA, Reddy PH. **Defective mitophagy in Alzheimer's disease**. *Ageing Res Rev* (2020) **64** 101191. DOI: 10.1016/j.arr.2020.101191
50. Doridot L, Jeljeli M, Chêne C. **Implication of oxidative stress in the pathogenesis of systemic sclerosis via inflammation, autoimmunity and fibrosis**. *Redox Biol* (2019) **25** 101122. DOI: 10.1016/j.redox.2019.101122
51. Brocker CN, Kim D, Melia T. **Long non-coding RNA Gm15441 attenuates hepatic inflammasome activation in response to PPARA agonism and fasting**. *Nat Commun* (2020) **11** 5847. DOI: 10.1038/s41467-020-19554-7
52. Yin C, Liu B, Wang P. **Eucalyptol alleviates inflammation and pain responses in a mouse model of gout arthritis**. *Br J Pharmacol* (2020) **177** 2042-2057. DOI: 10.1111/bph.14967
53. He Y, Hara H, Núñez G. **Mechanism and regulation of NLRP3 inflammasome activation**. *Trends Biochem Sci* (2016) **41** 1012-1021. DOI: 10.1016/j.tibs.2016.09.002
54. Zhu F, Li C, Chu F. **Target dysbiosis of gut microbes as a future therapeutic manipulation in Alzheimer's disease**. *Front Aging Neurosci* (2020) **12** 544235. DOI: 10.3389/fnagi.2020.544235
55. Brown MR, Radford SE, Hewitt EW. **Modulation of β-amyloid fibril formation in Alzheimer's disease by microglia and infection**. *Front Mol Neurosci* (2020) **13** 609073. DOI: 10.3389/fnmol.2020.609073
56. Lee E, Hwang I, Park S. **MPTP-driven NLRP3 inflammasome activation in microglia plays a central role in dopaminergic neurodegeneration**. *Cell Death Differ* (2019) **26** 213-228. DOI: 10.1038/s41418-018-0124-5
57. Yang W, Nagasawa K, Münch C. **Mitochondrial Sirtuin network reveals dynamic SIRT3-dependent deacetylation in response to membrane depolarization**. *Cell* (2016) **167** 985-1000.e21. DOI: 10.1016/j.cell.2016.10.016
58. Liu L, Xie B, Fan M. **Low-level saturated fatty acid palmitate benefits liver cells by boosting mitochondrial metabolism via CDK1-SIRT3-CPT2 cascade**. *Dev Cell* (2020) **52** 196-209.e9. DOI: 10.1016/j.devcel.2019.11.012
59. Cheng A, Yang Y, Zhou Y. **Mitochondrial SIRT3 mediates adaptive responses of neurons to exercise and metabolic and excitatory challenges**. *Cell Metab* (2016) **23** 128-142. DOI: 10.1016/j.cmet.2015.10.013
60. Palomer X, Román-Azcona MS, Pizarro-Delgado J. **SIRT3-mediated inhibition of FOS through histone H3 deacetylation prevents cardiac fibrosis and inflammation**. *Signal Transduct Target Ther* (2020) **5** 14. DOI: 10.1038/s41392-020-0114-1
61. Tewari D, Stankiewicz AM, Mocan A. **Ethnopharmacological approaches for dementia therapy and significance of natural products and herbal drugs**. *Front Aging Neurosci* (2018) **10** 3. DOI: 10.3389/fnagi.2018.00003
62. Dikalova AE, Pandey A, Xiao L. **Mitochondrial deacetylase Sirt3 reduces vascular dysfunction and hypertension while Sirt3 depletion in essential hypertension is linked to vascular inflammation and oxidative stress**. *Circ Res* (2020) **126** 439-452. DOI: 10.1161/CIRCRESAHA.119.315767
63. Zhou S, Tang X, Chen HZ. **Sirtuins and insulin resistance**. *Front Endocrinol (Lausanne)* (2018) **9** 748. DOI: 10.3389/fendo.2018.00748
64. Cunnane SC, Trushina E, Morland C. **Brain energy rescue: an emerging therapeutic concept for neurodegenerative disorders of ageing**. *Nat Rev Drug Discov* (2020) **19** 609-633. DOI: 10.1038/s41573-020-0072-x
65. Li CC, Chen WX, Wang J. **Nicotinamide riboside rescues angiotensin II-induced cerebral small vessel disease in mice**. *CNS Neurosci Ther* (2020) **26** 438-447. DOI: 10.1111/cns.13276
66. Zheng J, Akbari M, Schirmer C. **Hippocampal tau oligomerization early in tau pathology coincides with a transient alteration of mitochondrial homeostasis and DNA repair in a mouse model of tauopathy**. *Acta Neuropathol Commun* (2020) **8** 25. DOI: 10.1186/s40478-020-00896-8
67. Kaur N, Raja R, Ruiz-Velasco A. **Cellular protein quality control in diabetic cardiomyopathy: from bench to bedside**. *Front Cardiovasc Med* (2020) **7** 585309. DOI: 10.3389/fcvm.2020.585309
|
---
title: Testing effects of partner support and use of oral contraception during relationship
formation on severity of nausea and vomiting in pregnancy
authors:
- Kateřina Roberts
- Jan Havlíček
- Šárka Kaňková
- Kateřina Klapilová
- S. Craig Roberts
journal: BMC Pregnancy and Childbirth
year: 2023
pmcid: PMC10012454
doi: 10.1186/s12884-023-05468-x
license: CC BY 4.0
---
# Testing effects of partner support and use of oral contraception during relationship formation on severity of nausea and vomiting in pregnancy
## Abstract
### Background
A recent study focusing on dietary predictors of nausea and vomiting in pregnancy (NVP) found that women with higher levels of partner support, and those who had used oral contraception (OC) when they met the father, both tended to report less severe NVP compared with previous non-users or those with less supportive partners. We provide a further test of these factors, using a large sample of women from four countries who retrospectively scored their NVP experience during their first pregnancy.
### Methods
We recruited women who had at least one child to participate in a retrospective online survey. In total 2321 women completed our questionnaire including items on demographics, hormonal contraception, NVP, and partner support. We used general linear models and path analysis to analyse our data.
### Results
Women who had used OC when they met the father of their first child tended to report lower levels of NVP, but the effect size was small and did not survive adding the participant’s country to the model. There was no relationship between NVP and partner support in couples who were still together, but there was a significant effect among those couples that had since separated: women whose ex-partner had been relatively supportive reported less severe NVP. Additional analyses showed that women who were older during their first pregnancy reported less severe NVP, and there were also robust differences between countries.
### Conclusions
These results provide further evidence for multiple influences on women’s experience of NVP symptoms, including levels of perceived partner support.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12884-023-05468-x.
## Introduction
Nausea and vomiting in pregnancy (NVP) affects women all around the world [1–3]. According to a recent meta-analysis, almost $70\%$ of women experience NVP to at least some extent [4]. Although most common during the first trimester, symptoms very often persist throughout pregnancy [5]. Despite its prevalence, causes and mechanisms of NVP remain poorly understood. Some consider it to be a simple by-product of intense hormonal changes in pregnancy [6], including progesterone and estradiol but especially human chorionic gonadotropin (hCG) [2], which is produced by the trophoblast and subsequently by the placenta, with levels rising rapidly through the first trimester. However, in view of associations with beneficial effects, including higher birth weight and a lower probability of miscarriages, birth defects, pre-term deliveries and perinatal deaths [7], other authors suggest it has an adaptive function, such as causing compensatory placental growth or by reducing ingestion of harmful foods [8–11].
Recently, it was reported that women who used oral contraceptives (OC) when they met their partner experience lower NVP than non-users [12]. Fiurašková et al. [ 12] hypothesised that this could be due to within-couple genetic similarity, because OC-users may select relatively HLA-similar partners [13–15]. Conception with HLA-similar men could influence a cascade of responses, including reduced maternal immune response to the foetus [16, 17], less extensive uterine vasculature remodelling during placentation, reduced placental growth [18] and hence placental hCG production [19], and finally lower NVP [2, 20]. Alternatively, the effect of previous OC use on NVP might be mediated via partner support, based on previous evidence that women who met their partner while using OC were more generally satisfied with their relationship [21, 22]. Such support is important when dealing with distressful health issues [23], and good communication and perceived partner support are both connected with lower NVP [12, 24].
It should be noted that although Fiurašková et al. [ 12], reported that women’s OC use when meeting their partner (and father of their child) was associated with reduced NVP severity, this relationship was found during exploratory analysis as part of a larger study on dietary predictors of NVP [12]. Consequently, its robustness needs to be established by further confirmatory studies. The main aim of this study was therefore to test whether women using OC during partner choice do go on to experience lower levels of NVP. Additionally, we explored the possible role of partner support in NVP symptom severity. Based on the findings of previous studies, we expected that enhanced partner support would be connected with a lower level of NVP. To test these predictions, we used a large sample of women from whom we could collect the necessary data. These women were from the Czech Republic, Slovakia, UK, USA and Canada, and were part of a broader study primarily focusing on patterns of OC use and relationship satisfaction [21].
## Participants
We used an open survey, available to each visitor to our survey site. To this site, we recruited women who had at least one child and we asked them a series of questions about their first pregnancy and the biological father of their first child. We used a variety of recruitment methods to maximise the sample size, including personal contacts and advertisements on pregnancy and parenthood discussion websites, for which there was no financial reward for taking part. In addition, to further boost sample size, we recruited a proportion of the participants through a research panel administered by Qualtrics.com. These panel participants received a small amount ($7) as compensation for their time. The Qualtrics system prevented participants from potentially attempting to create duplicate entries for financial reward so that there were no duplicate responses from the same IP address. The questionnaires were completed online via the Qualtrics platform and were in the Czech language for Czech and Slovakian participants, and in English for participants from the UK, USA and Canada. The survey was constructed in English, translated into Czech by a bilingual speaker, and then translated back into English by a different bilingual speaker. Based on this back-translation validation step, a small number of unclear items were identified and fixed before the surveys were launched. Before recruitment began, we pre-tested the surveys in both English and Czech with a small sample ($$n = 5$$ women in each language) to ensure that questions were understandable and to estimate the time taken to complete the survey.
Consistent with our ethical permission from the University of Liverpool Psychology Ethics Committee, participants provided informed consent by a mouse-click on the “I consent to take part” button at the end of the information sheet which formed the landing page of the survey. The information sheet explained that participation was anonymous and that there was no way to trace any information back to individual participants. All data were stored and coded according to the unique personal identifier automatically generated by the survey software.
In total, 3678 women participated in the study. Of these, 874 did not answer the question about hormonal contraception and were excluded. We further excluded women who reported another type of hormonal contraception than combined oral contraception ($$n = 273$$), women with lower age at pregnancy than 18 ($$n = 69$$), women who reported having a multiple birth ($$n = 91$$) and women whose country of origin was represented in our data by only one or very few participants ($$n = 50$$). In total, 2321 women who reported both their OC usage when meeting their partner and their level of NVP when pregnant with their first child were included in the analysis. Only women using combined oral contraception were counted as OC users, while non-users were women who reported not using any form of hormonal contraception and/or using other non-hormonal contraceptive methods such as condoms or a diaphragm. From these, 945 were OC users and 1376 were non-users. Participants were from the USA ($$n = 1122$$), the Czech Republic/Slovakia ($$n = 955$$), the UK ($$n = 153$$) and Canada ($$n = 91$$). At the time of testing, 1634 women were still in a relationship with the father of their child and 685 women were already separated (2 did not report this). Their average age was 37.8 and the average age at which the pregnancy occurred was 26.8 (18–48). In total, 1190 women gave birth to a boy and 1105 to a girl (48 did not report this).
## Measures
We asked participants if they used any type of hormonal contraception at the time when they began the relationship with the biological father of their first child. We also asked specifically what type and what brand of hormonal contraception they used. Their experience of NVP when expecting their first child was retrospectively assessed by two items, which were formulated as recall items based heavily on the wording and items in the Rhodes Index of Nausea, Vomiting and Retching [25], and adapting each of its main two forms of response option. The first item was: “During your pregnancy, did you experience feelings of nausea, retching or vomiting which you attributed to “morning sickness“ or “pregnancy sickness”?” with possible answers “No”, “Mild”, “Moderate”, “Great”, “Severe”. The second question focused on a frequency of vomiting in a typical day: “At its peak, how often did you experience vomiting/retching in a typical day?” with possible answers “Never”, “Once”, “2–5 times”, and “More than 5 times”. We coded the answers to both questions (1–5 and 1–4, respectively). Reliability analysis indicated high internal reliability (Cronbach’s α = 0.846). On this basis we summed these items to create a composite score of retrospectively scored NVP severity.
To measure partner support, we used Brown’s [26] measure of partner support behaviour. The questionnaire consists of 11 items and it is introduced as follows: “How satisfied are you with the following aspects of the quality of your relationship with your partner?”. The individual items include following statements: “He shares similar experiences as me”, “He helps keep up my morale”, “He helps me out when I am in a pinch”, “He shows interest in my daily activities and problems”, “He goes out of his way to do special or thoughtful things for me”, “He allows me to talk about things that are very personal or private”, “He lets me know I am appreciated for the things I do for him”, “He tolerates my ups and downs and unusual behaviours”, “He takes me seriously when I have concerns”, “He says things that make my situation clearer and easier to understand”, and “He lets me know that he will be around if I need assistance”. Each item is rated on a scale from 1 (“Very dissatisfied”) to 6 (“Very satisfied”) and then summed. For women still in a relationship with the father of their first child we used standard versions of the questionnaire, but for women who were separated we modified the wording of items by adding “Thinking back about my ex-partner…” at the beginning of each item.
## Statistical analysis
We first analysed potential predictors of NVP separately. As NVP severity was not normally distributed (Kolmogorov-Smirnov test, $P \leq 0.001$), we used non-parametric Spearman rank correlations where appropriate, or Mann-Whitney tests to analyse the effect of OC and sex of the child on women’s NVP scores. Next, for a more comprehensive analysis that accounted for other potential predictors, we used a univariate general linear model (GLM) with NVP score as the dependent variable. We included OC use when meeting the father as a fixed factor and added to the model other predictors that were found to have significant effects in the exploratory analysis. Although F-tests are often considered robust to deviation from normality only in certain circumstances [27, 28], recent simulation studies demonstrate that it is robust to Type 1 error regardless of either the severity of deviation from a normal distribution, the sample size, or unequal distribution between groups [29, 30]. Nonetheless, as recommended by Field [31], we checked and confirmed that the conclusions of our analyses were unchanged when we used a robust test with $20\%$ trimmed means (for this, we used the WALRUS package in Jamovi). Note that because this technique permits inclusion of factors but not covariates, we controlled for Age when pregnant by regressing NVP severity on Age when pregnant (r = − 0.137, $P \leq 0.001$) and computing the standardised residuals for use as the dependent variable (i.e. NVP severity relative to Age when pregnant).
Additionally, we used path analysis to test possible direct or indirect effects of OC and Age when pregnant. For this we used GLM Mediation Model with NVP as dependent variable, age when pregnant as a mediator and OC when met as a factor. To test the effect of partner support (Brown’s measure of partner support behaviour), we performed further GLMs separately for women who were still with their partner and those who had separated from their partner, because their satisfaction scores cannot be directly compared.
Our sample sizes exceeded 787, which was the minimum identified by power analysis (G*Power) to detect a small effect ($d = 0.1$) with $80\%$ power. All statistical tests were performed using Jamovi version 1.6.23. All P values were two-sided and we defined statistical significance with an alpha of 0.05.
## Predictors of NVP
Initial exploratory analyses revealed several significant associations with NVP severity (Table 1). Women who used OC when they met their partner reported less severe NVP than women who did not use OC at that time, and women pregnant with a girl reported more severe NVP compared to those who were pregnant with a boy. There was a negative correlation between NVP severity and Age when pregnant. Finally, there was a positive correlation between NVP severity and the time elapsed since the pregnancy. Table 1Descriptives and exploratory testsMean (SD)nrhopOC use when metUser4.28 (2.09)944< 0.001Non-user4.59 (2.14)1375Sex of childBoy4.38 (2.11)11880.040Girl4.55 (2.13)1105Age when pregnant2311−0.147< 0.001Time since pregnancy23110.130< 0.001Differences in NVP severity during the pregnancy leading to a woman’s first child were compared in relation to her OC use when she met the father, and the sex of the child (using Mann-Whitney U tests). In addition, age while pregnant and the time from the pregnancy to completing the survey were tested using Spearman correlations We then used GLM to estimate the independent effect of OC use during relationship formation on NVP severity, while controlling for other variables. We did not include Time since pregnancy in this analysis, both because it was negatively correlated with Age when pregnant (rho = − 0.379, $P \leq 0.001$) and because a path analysis showed that the direct effect of Age when pregnant was far more influential on NVP severity than either its indirect effect via Time since pregnancy or the direct effect of Time since pregnancy (Supplemental Table 1, Supplemental Fig. 1). In the GLM model, the independent effects of OC use, Sex of child and Age when pregnant all remained statistically significant, while there was no significant OC use x Sex of child interaction (Table 2). We checked that these results were unlikely to be caused by non-normal distribution of the dependent variable by conducting a robust analysis using $20\%$ trimmed means; in this analysis, both the effects of OC use ($Q = 8.12$, $$P \leq 0.005$$) and Sex of child ($Q = 3.96$, $$P \leq 0.047$$) remained significant. Table 2Outcome of a GLM to test independent effects on reported levels of NVP severityEffectMean SquaredfFpη2OC use when met26.901, 22816.130.0130.003Sex of child19.391, 22814.420.0360.002Age when pregnant176.511, 228140.21<.0010.017OC use when met x Sex of child0.021, 22810.010.9430.000OC use when met refers to a woman’s use or non-use of oral contraception at the time when she met her partner. Statistically significant results are marked in bold Although we found a relationship between women’s previous OC use and NVP, the effect of Age when pregnant was also significant and appeared stronger. However, we noticed that OC use when couples met also varied with Age when pregnant, with users being older on average (mean, SD: 27.6 years, 4.6, $$n = 939$$) than non-users (26.2 years, 5.4, $$n = 1374$$; $t = 6.60$, $P \leq 0.001$). For this reason, we also conducted a path analysis to investigate the relationships between these variables (Fig. 1). The results showed a significant direct effect of OC use, as well as an indirect effect of OC use through Age when pregnant, although the direct effect is much more pronounced (Table 3). This confirms that the effect of OC when couples met is not confounded by differences in age. Fig. 1The scheme of mediation model. OC is the predictor variable, Age when pregnant is a mediator variable and NVP is the dependent variableTable 3Mediation estimates of the path analysis shown in Fig. 1EffectLabelEstimateSEZp% MediationIndirecta × b− 0.07690.0169− 4.55<.00125.4Directc−0.22580.0897−2.520.01274.6Totalc + a × b−0.30270.0896−3.38<.001100.0 Finally, we tested for possible effects of the country from which participants came. Including Country in the model meant that the effect of previous OC use was no longer significant (F1,2269 = 1.77, $$P \leq 0.183$$); the only significant effects were now Age when pregnant (F1,2269 = 29.50, $P \leq 0.001$) and Country (F3,2269 = 31.51, $P \leq 0.001$). Indeed, comparison of NVP rates across countries showed that women from the Czech Republic/Slovakia (CZ/SK) reported significantly lower NVP severity (mean, SD; 3.90 ± 1.96) than women from either the UK (5.00 ± 2.27, $P \leq 0.001$) the USA (4.84 ± 2.12, post hoc Tukey test, $P \leq 0.001$) or Canada (4.78 ± 2.23, $P \leq 0.001$), while there were no significant differences between participants from the UK, USA and Canada. Based on these results, we divided the sample into two sub-samples (CZ/SK and UK/USA/Canada) and added this as a fixed factor (Country) to the original GLM. In this final analysis, which we believe to be the most robust test of our data, only the effect of Age when pregnant and Country remained significant (Table 4). We also ran the analyses separately for women from the two sub-samples. The results showed a robust and significant effect of Age when pregnant in both sub-samples, but the effects of OC and Sex of the child were no longer significant (Table 4).Table 4Effects on NVP severity in the whole sample (above) and the two geographical sub-samples (below)Mean SquareFdfpη2Whole sample OC use when met3.620.861, 22770.3550.000 Age when pregnant109.7126.001, 2277<.0010.011 Sex of child14.753.501, 22770.0620.001 Country381.2390.341, 2277<.0010.038 OC use when met x Sex of child0.140.031, 22770.8540.000 OC use when met x Country3.270.781, 22770.3790.000 Sex of child x Country1.020.241, 22770.6230.000 OC use when met x Sex of child x Country0.700.171, 22770.6840.000Czech Republic / Slovakia OC use when met0.050.011, 9200.9040.000 Age when pregnant57.9415.561, 920<.0010.017 Sex of child10.352.781, 9200.0960.003 OC use when met x Sex of child0.050.011, 9200.9050.000UK / USA / Canada OC use when met8.741.921, 13560.1660.001 Age when pregnant58.2012.781, 1356<.0010.009 Sex of child4.641.021, 13560.3130.001 OC use when met x Sex of child0.840.181, 13560.6680.000Statistically significant results are marked in bold
## Partner support
Of the women included in the analyses above, 1634 were still in a relationship with the father and provided scores of current support, while 685 were no longer in that relationship and provided retrospective scores of partner support. For this reason, we analysed the association between NVP severity and levels of perceived partner support separately for these two groups. The same variables as in the previous models were included in this analysis, plus an additional variable – Partner support.
Among couples still together, there was a significant effect of Age when pregnant and Country on NVP severity, but we found no significant effect of Partner support, nor of either OC use during relationship formation or Sex of child (Table 5). Among separated couples, however, there was a significant effect of Partner support ($$P \leq 0.030$$), which was independent of the significant effect of Country. Women who reported relatively high levels of support from their ex-partner during pregnancy reported lower NVP severity. As before, the effects of OC use during relationship formation and Sex of child were not significant. Finally, in this sub-sample, the effect of Age when pregnant did not quite reach statistical significance ($$P \leq 0.082$$; Table 5).Table 5Effects on NVP severity in couples still together and couples who subsequently separatedMean SquareFdfpη2Couples still together OC use when met6.171.4811,5740.2250.001 Sex of child15.143.6211,5740.0570.002 Age when pregnant86.9820.8011,574<.0010.012 Partner support0.110.02611,5740.8720.000 Country277.9766.4711,574<.0010.040 OC use when met x Sex of child2.260.5411,5740.4630.000 OC use when met x Country1.200.2911,5740.5920.000 Sex of child x Country1.100.2611,5740.6080.000 OC use when met x Sex of child x Country1.900.4511,5740.5010.000Couples who separated OC use when met0.730.1716530.6840.000 Sex of child0.650.1516530.7000.000 Age when pregnant13.403.0416530.0820.005 Partner support20.724.7016530.0300.007 Country49.4611.231653< 0.0010.017 OC use when met x Sex of child3.400.7716530.3800.001 OC use when met x Country3.880.8816530.3490.001 Sex of child x Country0.230.0516530.8180.000 OC use when met x Sex of child x Country5.961.3516530.2450.002Statistically significant results are marked in bold
## Discussion
In a large sample of women, we tested for possible associations between their OC use during relationship formation, or their satisfaction with the support they receive from their partner, and the perceived severity of NVP that they experienced in their first pregnancy. Based on previous findings, we expected that women who used OC during relationship formation and who expressed satisfaction with their partner’s support would report lower NVP severity. We found some initial support for the first prediction, but the effect was weak. The pattern was not present when we accounted for the women’s country of origin, nor when we tested it separately in samples of Czech/Slovakian women or women from the UK, the USA and Canada. Regarding the second prediction, we did find that women who reported higher satisfaction with their level of partner support had lower NVP, but this was only the case among women who had since separated from their partners; the effect was not present in those couples who remained together.
We found that there was a difference in levels of NVP between Czech/Slovakian women and those from the UK, the USA and Canada, with the former group having significantly lower NVP levels. We had not originally hypothesised a difference between these countries, but as the effect appeared to be strong, we included it in additional analyses. The discovery of this difference was important because it altered the conclusions we were able to draw about the apparent effect of oral contraceptive use at relationship formation. The initial analyses suggested that women experienced less severe NVP if they had met the father while using oral contraception, which would be consistent with our hypothesis that this may lead to relatively genetically similar partners, with potential corollary effects on NVP severity. The effect size was small, however, and this may be because it appears to be confounded by sample differences, such that Czech/Slovakian women who reported relatively less severe NVP were also more likely to have met the father while using OC (Supplemental Table 2), compared to women from the other countries. Our subsequent analyses in which we either include Country as a fixed factor in the model or analyse the two geographical sub-samples separately, confirm this. Our results thus expose and highlight the need for great caution in interpreting cross-national samples. If we had not included the effects of country, we would have reached a very different conclusion.
What lies behind this difference in perceived NVP severity is unclear but we can make some speculations. One suggestion is dietary differences between countries. For example, consumption of sugar-sweetened beverages is higher in the UK, the USA and Canada compared to the Czech Republic [32], and their consumption is associated with higher NVP levels [33]. Additionally, BMI could play a role, as higher pre-pregnancy BMI is associated with NVP levels [33, 34] and average BMI in the Czech Republic and *Slovakia is* lower than in the English-speaking countries in our sample, especially compared to the USA which represents the majority of this sub-sample [35]. An alternative form of explanation is a reporting difference, such that women from the Czech Republic and Slovakia experience an equivalent degree of NVP but tend to report it as less severe than those from the UK, the USA and Canada.
Although the apparent effect of previous OC use on subsequent NVP appears to be explained by other factors, this is not to say that it must be entirely non-existent. First, the confounds we discussed above are unlikely to be responsible for the positive association between NVP severity and OC use at relationhip formation reported by Fiurašková et al. [ 12], as their sample was predominantly from the UK or North America. Second, the literature shows clearly that NVP is affected by a wide variety of factors and, if OC use during partner choice does have some effect, this would likely be relatively small and could be easily overshadowed by other factors. Furthermore, we should bear in mind that OC use during partner choice, if it is a factor, is only (at best) a crude proxy for the hypothesised underlying mechanism, which is the level of HLA similarity between partners. There remains a need for further investigation of NVP levels in HLA-genotyped couples.
Our second focus of interest was whether NVP levels might be inversely associated with perceived levels of partner support [12], whether this is due to a causative influence of partner support on NVP level or to women with higher support tending to score their NVP as less severe. To test the effect of partner support, we analysed responses separately for women who were still in a relationship with their partner and for those who were no longer with their partner. Women who were still together with their partner reported their current level of partner’s support while those who were separated reported their received support retrospectively. We did not find a significant effect of partner support on NVP levels in women who were still together with their partners, but there was a protective effect on NVP of recalled partner support among those who had separated.
The fact that we did not find any effect of partner support in the group of women who were still with their partner could be caused by a ceiling effect, with relatively high scores of partner support across the sample (still together – mean, SD: 49.4, 11.9; apart – 30.1, 13.5), as well as lower variability in these scores compared to the sub-sample of women who had separated from their father (Levene’s test, $P \leq 0.05$). In the latter group, perhaps there was more variability in support even at the time of the pregnancy, which may have influenced women’s experience of NVP. Altogether then, our results indicated that partner support may play a role, at least to some extent, in either affecting the level of NVP or in the subjective perception of NVP symptoms.
We also found several other factors to be correlated with NVP severity scores. First, women’s age when pregnant was consistently a significant predictor of NVP. Women who were pregnant at a younger age had higher NVP severity scores than women who were pregnant at an older age. This inverse relationship between age and NVP is consistent across many studies [36–38]. Second, sex of the child was significantly associated with NVP scores in the initial analysis, although the effect size was small and disappeared in analyses that separately examined women still, or no longer, in a relationship with the father. This pattern matches previous evidence which is also somewhat inconsistent: several studies suggest that bearing a female foetus is associated with higher NVP ([39–42], but other studies do not find a sex difference [43].
## Criticisms
Although our sample was relatively large and targeted specifically at women’s first pregnancy, our approach introduced certain limitations which must be acknowledged. First, women provided retrospective scores of their NVP symptoms and severity. Retrospective reports are prone to memory-related biases [44] and an alternative approach would be to sample currently pregnant women, as we did previously [12]. There is, however, evidence that women can accurately and reliably report distant (10 to 15 years) events in their pregnancy [45] (although NVP was not among the tested variables in that study). We are aware that because of the possibility of memory-related bias, we need to interpret our results carefully.
Second, this retrospective approach meant that we could not use a standard measure of NVP severity such as the Rhodes Index [25], because this asks women about their symptoms over the preceding 12 hours. Instead, we asked women to provide scores on two items that dealt with their recalled experience over the entire pregnancy. ( It would be interesting, in a future study, to examine the correlation between Rhodes Index scores measured in the first trimester with scores on our retrospective items measured sometime after the pregnancy, but this was beyond the scope of this study.) Although these two criticisms compel us to be cautious about our results, the finding that age of pregnancy consistently predicted NVP scores, as it does in many previous studies, provides some reassurance about the accuracy of recall. Furthermore, an advantage of looking back over the whole pregnancy, compared to Rhodes Index responses, is that we circumvent the problem of varying onset and persistence of NVP symptoms: although NVP is most common in the first trimester, some women experience symptoms later or even throughout their pregnancy [12].
Another criticism is that possible recall bias may also have impacted on women’s ratings of partner support. Approximately $30\%$ of the sample were no longer in a relationship with the father of the first child. This meant that we had to perform the analysis of partner support separately for women who were still with, or no longer with, their partners. It also meant that women rated partner support slightly differently: those who were still together provided a rating of current support, while those who were separated assessed support retrospectively. We were therefore careful to analyse these sub-samples separately. An advantage of the approach, however, was that it revealed an interesting difference between the subsamples in the effect of partner support, such that an effect of reduced partner support may be easier to detect in couples whose relationship was destined soon to end.
## Conclusion
Although NVP is a widely occurring phenomenon in pregnancy, its causes and mechanisms are still not clear. Our study was motivated by two new potential predictors revealed in a previous exploratory analysis [12], but we did not find strong supporting evidence for these in this sample. We did not find strong evidence for an effect of OC use during relationship formation once the confounding effect of country of origin was taken into account. We did find some support for an effect of poor partner support on NVP levels, but only in couples who had separated when the survey was competed. In any case, it is likely that such effects would be rather small, considering how complex a phenomenon NVP is. More investigation is needed, including in samples of women who are currently pregnant and with more sensitive questionnaires, while also considering possible underlying mechanisms such as hormone levels and HLA similarity of partners.
## Supplementary Information
Additional file 1: Supplemental Table 1. Results of mediation model. Supplemental Fig. 1. The scheme of the mediation model. Supplemental Table 2. Descriptives of two sub-samples of women according to country.
## References
1. Kramer J, Bowen A, Stewart N, Muhajarine N. **Nausea and vomiting of pregnancy: prevalence, severity and relation to psychosocial health**. *Amer J Matern Child Nurs* (2013.0) **38** 21-27. DOI: 10.1097/NMC.0b013e3182748489
2. Lee NM, Saha S. **Nausea and vomiting of pregnancy**. *Gastroenterol Clin N Am* (2011.0) **40** 309-334. DOI: 10.1016/j.gtc.2011.03.009
3. Pepper G, Roberts SC. **Rates of nausea and vomiting in pregnancy and dietary characteristics across populations**. *Proc R Soc B Biol Sci* (2006.0) **273** 2675-2679. DOI: 10.1098/rspb.2006.3633
4. Einarson TR, Piwko C, Koren G. **Quantifying the global rates of nausea and vomiting of pregnancy: a meta analysis**. *J Popul Ther Clin Pharmacol* (2013.0) **20** 171-183
5. Lindseth G, Vari P. **Nausea and vomiting in late pregnancy**. *Health Care Women Int* (2005.0) **26** 372-386. DOI: 10.1080/07399330590933926
6. Lagiou P, Tamimi R, Mucci LA, Trichopoulos D, Adami HO, Hsieh CC. **Nausea and vomiting in pregnancy in relation to prolactin, estrogens, and progesterone: a prospective study**. *Obstet Gynecol* (2003.0) **101** 639-644. PMID: 12681864
7. Patil CL, Abrams ET, Steinmetz AR, Young SL. **Appetite sensations and nausea and vomiting in pregnancy: an overview of the explanations**. *Ecol Food Nutr* (2012.0) **51** 394-417. DOI: 10.1080/03670244.2012.696010
8. Huxley RR. **Nausea and vomiting in early pregnancy: its role in placental development**. *Obstet Gynecol* (2000.0) **95** 779-782. PMID: 10775746
9. Lumey LH. **Compensatory placental growth after restricted maternal nutrition in early pregnancy**. *Placenta.* (1998.0) **19** 105-111. DOI: 10.1016/S0143-4004(98)90105-9
10. Hook EB. **Dietary cravings aversions during**. *Am J Clin Nutr* (1978.0) **31** 1355-1362. DOI: 10.1093/ajcn/31.8.1355
11. 11.Profet M. Pregnancy sickness as adaptation: a deterrent to maternal ingestion of teratogens. In: Barkow JH, Cosmides L, Tooby J, editors. The Adapted Mind: Evolutionary Psychology and the generation of culture. Oxford: Oxford University Press; 1992. p. 327–65.
12. Fiurašková K, Havlíček J, Roberts SC. **Dietary and psychosocial correlates of nausea and vomiting in pregnancy**. *Food Qual Prefer* (2021.0) **93** 104266. DOI: 10.1016/j.foodqual.2021.104266
13. Havlíček J, Roberts SC. **MHC-correlated mate choice in humans: a review**. *Psychoneuroendocrinology.* (2009.0) **34** 497-512. DOI: 10.1016/j.psyneuen.2008.10.007
14. Wedekind C, Seebeck T, Bettens F, Paepke AJ. **MHC-dependent mate preferences in humans**. *Proc R Soc B Biol Sci* (1995.0) **260** 245-249. DOI: 10.1098/rspb.1995.0087
15. Roberts SC, Gosling LM, Carter V, Petrie M. **MHC-correlated odour preferences in humans and the use of oral contraceptives**. *Proc R Soc B Biol Sci* (2008.0) **275** 2715-2722. DOI: 10.1098/rspb.2008.0825
16. Havlíček J, Winternitz J, Roberts SC. **Major histocompatibility complex-associated odour preferences and human mate choice: near and far horizons**. *Philos Trans R Soc B: Biol Sci* (2020.0) **375** 20190260. DOI: 10.1098/rstb.2019.0260
17. Moffett A, Loke C. **Immunology of placentation in eutherian mammals**. *Nat Rev Immunol* (2006.0) **6** 584-594. DOI: 10.1038/nri1897
18. Madeja Z, Yadi H, Apps R, Boulenouar S, Roper SJ, Gardner L. **Paternal MHC expression on mouse trophoblast affects uterine vascularization and fetal growth**. *Proc Natl Acad Sci U S A* (2011.0) **108** 4012-4017. DOI: 10.1073/pnas.1005342108
19. Korevaar TIM, Steegers EAP, de Rijke YB, Schalekamp-Timmermans S, Visser WE, Hofman A. **Reference ranges and determinants of total hCG levels during pregnancy: the generation R study**. *Eur J Epidemiol* (2015.0) **30** 1057-1066. DOI: 10.1007/s10654-015-0039-0
20. Forbes S. **Pregnancy sickness and embryo quality**. *Trends Ecol Evol* (2002.0) **17** 115-120. DOI: 10.1016/S0169-5347(01)02428-4
21. Roberts SC, Klapilová K, Little AC, Burriss RP, Jones BC, DeBruine LM. **Relationship satisfaction and outcome in women who meet their partner while using oral contraception**. *Proc R Soc B Biol Sci* (2012.0) **279** 1430-1436. DOI: 10.1098/rspb.2011.1647
22. Roberts SC, Little AC, Burriss RP, Cobey KD, Klapilová K, Havlíček J. **Partner choice, relationship satisfaction, and oral contraception: the congruency hypothesis**. *Psychol Sci* (2014.0) **25** 1497-1503. DOI: 10.1177/0956797614532295
23. Revenson TA. **Social support and marital coping with chronic illness**. *Ann Behav Med* (1994.0) **16** 122-130
24. Iatrakis GM, Sakellaropoulos GG, Kourkoubas AH, Kabounia SE. **Vomiting and nausea in the first 12 weeks of pregnancy**. *Psychother Psychosom* (1988.0) **49** 22-24. DOI: 10.1159/000288062
25. Rhodes VA, McDaniel RW. **The index of nausea, vomiting, and retching: a new format of the index of nausea and vomiting**. *Oncol Nurs Forum* (1999.0) **26** 889-894. PMID: 10382187
26. Brown MA. **Social support during pregnancy: a unidimensional or multidimensional construct?**. *Nurs Res* (1986.0) **35** 4-9. DOI: 10.1097/00006199-198601000-00002
27. Montgomery DC. *Design and analysis of experiments* (2017.0)
28. Winer BJ, Brown DR, Michels KM. *Statistical principles in experimental design* (1991.0)
29. Schmider E, Ziegler M, Danay E, Beyer L, Bühner M. **Is it really robust? Reinvestigating the robustness of ANOVA against violations of the normal distribution assumption**. *Methodol Eur J Res Methods Behav Soc Sci* (2010.0) **6** 147-151
30. Blanca Mena MJ, Alarcón Postigo R, Arnau Gras J, Bono Cabré R, Bendayan R. **Non-normal data: Is ANOVA still a valid option?**. *Psicothema* (2017.0) **29** 552-557. PMID: 29048317
31. Field A. *Discovering statistics using IBM SPSS statistics* (2018.0)
32. Singh GM, Micha R, Khatibzadeh S, Shi P, Lim S, Andrews KG. **Global, regional, and national consumption of sugar-sweetened beverages, fruit juices, and milk: a systematic assessment of beverage intake in 187 countries**. *PLoS One* (2015.0) **10** 1-20. DOI: 10.1371/journal.pone.0124845
33. Chortatos A, Haugen M, Iversen PO, Vikanes Å, Magnus P, Veierød MB. **Nausea and vomiting in pregnancy: associations with maternal gestational diet and lifestyle factors in the Norwegian mother and child cohort study**. *BJOG.* (2013.0) **120** 1642-1653. DOI: 10.1111/1471-0528.12406
34. Klebanoff MA, Koslowe PA, Kaslow R, Rhoads GG. **Epidemiology of vomiting in early pregnancy**. *Obstet Gynecol* (1985.0) **66** 612-616. PMID: 3903578
35. Abarca-Gómez L, Abdeen ZA, Hamid ZA, Abu-Rmeileh NM, Acosta-Cazares B, Acuin C. **Worldwide trends in body-mass index, underweight, overweight, and obesity from 1975 to 2016: a pooled analysis of 2416 population-based measurement studies in 128.9 million children, adolescents, and adults**. *Lancet* (2017.0) **390** 2627-2642. DOI: 10.1016/S0140-6736(17)32129-3
36. Källen B, Lundberg G, Åberg A. **Relationship between vitamin use, smoking, and nausea and vomiting of pregnancy**. *Acta Obstet Gynecol Scand* (2003.0) **82** 916-920. PMID: 12956841
37. Dekkers GWF, Broeren MAC, Truijens SEM, Kop WJ, Pop VJM. **Hormonal and psychological factors in nausea and vomiting during pregnancy**. *Psychol Med* (2019.0) **50** 229-236. DOI: 10.1017/S0033291718004105
38. Fejzo MS, Trovik J, Grooten IJ, Sridharan K, Roseboom TJ, Vikanes Å. **Nausea and vomiting of pregnancy**. *Nat Rev Dis Primers* (2019.0) **5** 1-17. DOI: 10.1038/s41572-019-0110-3
39. Danzer H, Braustein GD, Rasor J, Forsythe A, Wade ME. **Maternal serum human chorionic gonadotropin concentrations and fetal sex prediction**. *Fertil Steril* (1980.0) **34** 336-340. DOI: 10.1016/S0015-0282(16)45027-2
40. Czeizel AE, Puhó E. **Association between severe nausea and vomiting in pregnancy and lower rate of preterm births**. *Cent Eur J Public Health* (2004.0) **12** 253-259
41. Mitsuda N, Eitoku M, Maeda N, Fujieda M, Suganuma N. **Severity of nausea and vomiting in singleton and twin pregnancies in relation to fetal sex: the Japan environment and Children’s study (JECS)**. *J Epidemiol* (2019.0) **29** 340-346. DOI: 10.2188/jea.JE20180059
42. Young NR, La Rosa M, Mehr SA, Krasnow MM. **Does greater morning sickness predict carrying a girl? Analysis of nausea and vomiting during pregnancy from retrospective report**. *Arch Gynecol Obstet* (2021.0) **303** 1161-1166. DOI: 10.1007/s00404-020-05839-1
43. Louik C, Hernandez-Diaz S, Werler MM, Mitchell AA. **Nausea and vomiting in pregnancy: maternal characteristics and risk factors**. *Paediatr Perinat Epidemiol* (2006.0) **20** 270-278. DOI: 10.1111/j.1365-3016.2006.00723.x
44. Coughlin SS. **Recall bias in epidemiologic studies**. *J Clin Epidemiol* (1990.0) **43** 87-91. DOI: 10.1016/0895-4356(90)90060-3
45. Yawn BP, Suman VJ, Jacobsen SJ. **Maternal recall of distant pregnancy events**. *J Clin Epidemiol* (1998.0) **51** 399-405. DOI: 10.1016/S0895-4356(97)00304-1
|
---
title: Exosomal USP13 derived from microvascular endothelial cells regulates immune
microenvironment and improves functional recovery after spinal cord injury by stabilizing
IκBα
authors:
- Xuhui Ge
- Zheng Zhou
- Siting Yang
- Wu Ye
- Zhuanghui Wang
- Jiaxing Wang
- Chenyu Xiao
- Min Cui
- Jiawen Zhou
- Yufeng Zhu
- Rixiao Wang
- Yu Gao
- Haofan Wang
- Pengyu Tang
- Xuhui Zhou
- Ce Wang
- Weihua Cai
journal: Cell & Bioscience
year: 2023
pmcid: PMC10012460
doi: 10.1186/s13578-023-01011-9
license: CC BY 4.0
---
# Exosomal USP13 derived from microvascular endothelial cells regulates immune microenvironment and improves functional recovery after spinal cord injury by stabilizing IκBα
## Abstract
Spinal cord injury (SCI) can result in irreversible sensory and motor disability with no effective treatment currently. After SCI, infiltrated macrophages accumulate in epicenter through destructed blood-spinal cord barrier (BSCB). Further, great majority of macrophages are preferentially polarized to M1 phenotype, with only a few transient M2 phenotype. The purpose of this study was to explore roles of vascular endothelial cells in microglia/macrophages polarization and the underlying mechanism. Lipopolysaccharide (LPS) was used to pretreat BV2 microglia and RAW264.7 macrophages followed by administration of conditioned medium from microvascular endothelial cell line bEnd.3 cells (ECM). Analyses were then performed to determine the effects of exosomes on microglia/macrophages polarization and mitochondrial function. The findings demonstrated that administration of ECM shifted microglia/macrophages towards M2 polarization, ameliorated mitochondrial impairment, and reduced reactive oxygen species (ROS) production in vitro. Notably, administration of GW4869, an exosomal secretion inhibitor, significantly reversed these observed benefits. Further results revealed that exosomes derived from bEnd.3 cells (Exos) promote motor rehabilitation and M2 polarization of microglia/macrophages in vivo. Ubiquitin-specific protease 13 (USP13) was shown to be significantly enriched in BV2 microglia treated with Exos. USP13 binds to, deubiquitinates and stabilizes the NF-κB inhibitor alpha (IκBα), thus regulating microglia/macrophages polarization. Administration of the selective IκBα inhibitor betulinic acid (BA) inhibited the beneficial effect of Exos in vivo. These findings uncovered the potential mechanism underlying the communications between vascular endothelial cells and microglia/macrophages after SCI. In addition, this study indicates exosomes might be a promising therapeutic strategy for SCI treatment.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s13578-023-01011-9.
## Background
Spinal cord injury (SCI) is a central nervous system (CNS) disease characterized by long-term sensory and motor disorders. Due to high morbidity and mortality, SCI poses a major health burden worldwide [1, 2]. Primary injury of spinal cord is mainly related to injury of neurons and axons directly, whereas secondary injury is mainly caused by neuroinflammation that may result in cavitation and edema [3, 4]. The blood-spinal cord barrier (BSCB) or blood-brain barrier (BBB) is broken within 5 min after SCI, resulting in inflammatory response, excessive edema, progressive neuronal death, axonal dieback and glia activation. BSCB comprises continuous endothelial cells connected by molecular junctions physiologically, and functions as a barrier to suppress paracellular and transcellular transport. Notably, peripheral inflammatory cells, factors and other harmful cytokines infiltrate and accumulate in the lesion epicenter through the broken BSCB after SCI [5, 6]. In addition, activated resident microglia and infiltrated macrophages accumulate in the epicenter, resulting in a severer inflammatory response. Activated microglia and infiltrated macrophages can polarize to classical pro-inflammatory (M1) or alternative anti-inflammatory (M2) phenotype due to diverse signals in SCI microenvironment. Studies reported that M1 microglia/macrophages are harmful and M2 polarization promotes neurogenesis after SCI [7–9]. Therefore, studies should investigate the underlying mechanism regarding the cell crosstalk in the SCI microenvironment and shift the microglia/macrophages phenotype balance from M1 to M2 thus inhibiting detrimental neuroinflammation. We previously reported that a large number of microglia/macrophages surround the newborn blood vessels in the injured area after spinal cord injury, and microglia/macrophages play a crucial role in the tight junction formation of vascular endothelial cells through exosomal transfer of miR-155 [10]. However, whether vascular endothelial cells affect microglia/macrophages polarization and the role of its exosomes in microglia/macrophages is still unclear.
Exosomes are a kind of nanosized extracellular vesicles with a particle diameter of 50–150 nm. They are crucial elements of the paracrine secretions and implicated in mediation of the communications between different cells via transfer of genetic material messages including non-coding RNAs, mRNAs as well as proteins and preventing them from degrading [11, 12]. Several studies have explored the important role of exosomes in different biological process including regulation of intercellular signaling, angiogenesis, neurogenesis, inflammation and tumor progression [13–15]. However, only a few studies have explored interactions and communications between diverse cells in the microenvironment of SCI. Interactions and detailed mechanisms between the vascular endothelial cells and macrophages/microglia should be explored.
This current study demonstrates that exosomes derived from microvascular endothelial cells promote motor rehabilitation and M2 polarization of microglia/macrophages through transferring ubiquitin-specific protease 13 (USP13) post-injury. The delivered exosomal USP13 regulates microglia/macrophages polarization after SCI by stabilizing IκBα thus inhibiting NF-κB signaling pathway. Our work elucidates the potential mechanism between vascular endothelial cells and microglia/macrophages, which may contribute to a better understanding of the regulation in the microenvironment after SCI. Moreover, these findings support that exosomes derived from vascular endothelial cells are promising therapeutic agents for treatment of SCI.
## SCI model and treatment
All animals were raised and housed under the guidelines of the Animal Committee of the First Affiliated Hospital of Nanjing Medical University. SCI model was performed as previously described [16]. Isoflurane inhalation was used to anesthetize mice, and then laminectomy was carried out at T8 level to expose spinal cord. We established a mice SCI model via dropping a 5 g rod from a height of 6.5 cm to the spinal cord using spinal cord impactor (RWD). The muscles and skin were sutured immediately after injury. The mice bladders were evacuated 3 times a day until the bladder function was restored.
Mice subjected to SCI were randomly divided into several groups and administrated with bEnd.3 cells derived exosomes (Exos) or USP13-deleted exosomes (shUSP13-Exos) or corresponding control exosomes (shNC-Exos, 200 µg total protein of exosomes in 200 µL PBS), or an equal volume of PBS (200 µL) through the tail-vein injection post-injury as described previously [17, 18].
## Functional behavioral analysis
Mice were housed in a 12-h light-dark cycle and provided with food and water ad libitum. Prior to performing behavioral tests, all animals were acclimatized to the testing room or apparatus for 1 h.
The Basso Mouse Scale (BMS) score was evaluated by two investigators blinded to the groups at 1, 3, 7, 14, and 28 days post-injury according to the hindlimb locomotor function.
A rotarod test was carried out to evaluate balance and motor coordination after SCI. Mice were placed on an accelerating rotarod from 0 to 40 r.p.m. Each mouse was allowed to practice for one trial before the two test trials. The interval between each trial is 20 min. Time taken to fall was averaged from two individual test for a final score per mouse.
The (motor evoked potentials) MEPs were examined 28 days after SCI by electromyography test. The stimulation, recording, reference, and grounding electrode were placed at the rostral ends of spinal cord, flexor of biceps femoris, distal tendon of hindlimb muscle, and under skin, respectively. A single stimulation (0.5 mA, 0.5 ms, 1 Hz) was used to induce MEPs, then amplitude and latency were quantified to determine the hindlimb nerve conduction function.
## Immunofluorescence staining
The spinal cords of the injured areas were dissected and were then fixed in $4\%$ paraformaldehyde overnight, followed by gradient dehydration in $15\%$ and $30\%$ sucrose solutions. Samples were then embedded in OCT compound and dissected into 14-µm thick sections. The frozen sections were washed with PBS and blocked with $5\%$ BSA plus $0.3\%$ Triton X-100 at room temperature for 1 h, and then incubated with primary antibodies at 4 ℃ overnight. Conjugated secondary antibodies were then used and DAPI was added. The images were photographed under a confocal microscope.
For cultured cell staining, the cells were fixed in $4\%$ paraformaldehyde and permeabilized with $0.3\%$ Triton X-100. After that, cells were blocked with $5\%$ BSA for 1 h at room temperature and incubated with primary antibodies overnight at 4 ℃. Conjugated secondary antibodies were then used and DAPI was added. The images were photographed under a confocal microscope.
## Intralumbar delivery
Betulinic acid (BA) was administered through intralumbar injection as described previously [19, 20]. A 30-gauge beveled needle was connected to a 10 µL syringe and the needle was inserted between L4 and L5. 2 µg BA in 5 µL $0.1\%$ DMSO (diluted in normal saline) was injected in the intralumbar using a syringe pump. Administration was carried out for more than 5 min to allow diffusion, and the cannula was left in place for 5 min. Administration of BA was performed twice a week and each mouse was given a total of 4 doses.
## Cell culture and transfection
bEnd.3 vascular endothelial cell line, BV2 microglia cell line, RAW264.7 macrophage cell line and HEK 293T cell line were purchased from Cell Bank of the Chinese Academy of Science. Lipopolysaccharide (LPS, 1 µg/ml) was added to stimulate BV2 microglia and RAW264.7 macrophages for 24 h followed by co-culturing with exosomes (200 µg/ml) of indicated groups. Cells were transfected using Lipofectamine 3000 reagents according to the manufacturer’s instructions.
Primary microglial cells were prepared as previously described [17]. Brain tissues of neonatal mouse were cut into 1 mm3 pieces and incubated with $0.125\%$ trypsin (Gibco, USA) with gentle shaking at 37 ℃ for 10 min. The digested tissues were filtered with a 100-µm nylon mesh and cell suspension were then cultured in T75 flasks pre-coated with ploy-l-lysine (Beyotime, China) to obtain primary mixed glial cells. After 14 days of culture in vitro, the mature microglial cells were separated by shaking at 200 rpm for 2 h at room temperature for immunofluorescence staining.
## The shRNA and plasmid construction
The shRNA-control and shUSP13 were constructed by Genebay Biotech (Nanjing, China). The Flag-tagged USP13 (either WT or C345A mutant) expressing plasmid, Myc-tagged IκBα expression plasmid, and a series of HA-tagged Ubiquitin expression plasmids were generated by cloning their open reading frame with the N-terminal tag sequence into the vectors (Genebay Biotech, China). Scrambled lentiviral construct was used as a negative control.
## Exosome isolation and identification
Once bEnd.3 cells achieving $80\%$ confluency, the culture medium was replaced with exosome-free fetal bovine serum for 48 h. The medium was then acquired and centrifuged at 300×g for 10 min and 2000×g for 20 min at 4 ℃. Supernatant was collected and centrifuged at 10,000×g for 30 min followed by filtered using a 0.22-µm sterile filter (Steritop™ Millipore). For exosomes collection and purification, filtered medium was subsequently ultra-centrifugated at 100,000×g for 60 min at 4 ℃. Then the supernatant was discarded and sterile PBS was used to wash the pellet of exosomes, and another ultracentrifugation procedure (100,000×g, 60 min) was continued. Exosomes were then resuspended, aliquoted and stored under − 80 °C or used immediately for downstream experiments.
Size distribution of exosomes was explored using Nanoparticle Tracking Analysis (NTA). Morphological analysis of exosomes was performed using a transmission electron microscope (TEM).
The detection of exosomes uptake in BV2 microglia was performed using exosomes-containing PBS medium which was incubated with Dil solution. The images were photographed under a confocal microscope.
## In vitro
detection of USP13 transfer
To detect the direct transfer of exosomal USP13, bEnd.3 cells were transfected with a GFP-USP13 fusion mRNA construct (Genebay biotech, China), after which exosomes were purified and applied to BV2 microglia as previously reported [21]. Following incubation, BV2 cells were fixed with $4\%$ PFA and permeabilized with $0.3\%$ Triton X-100, and stained with DAPI. The GFP fluorescence in the target BV2 cells were observed under a confocal microscope.
## Quantitative real-time PCR (qRT-PCR)
Total RNA was collected using TRIzol reagent (Invitrogen, USA). The RNA was then reverse transcribed into cDNA using a PrimeScript RT Reagent Kit (Takara, Japan). qRT-PCR was performed using SYBR Green PCR master mix. Relative expression levels of target genes were normalized to GAPDH and quantified using the 2−ΔΔCT method.
## RNA sequence (RNA-seq)
RNA-seq analysis was performed by Genminix Information Co., Ltd., (Shanghai, China). Total RNAs from BV2 microglia treated with Exos and PBS ($$n = 3$$/group) were extracted. Quality RNA samples were converted into cDNA libraries following previously described methods [22]. Following fragments purification, the purified products were amplified with 12–15 cycles of PCR to create the final cDNA library. Finally, Libraries were sequenced on the Illumina Hiseq X Ten following the manufacturer’s protocols. Data analysis was performed in R studio software after normalization, log2 transformation and probe annotation. Fold changes > 1.5 and $P \leq 0.05$ represented differentially expressed genes (DEGs).
## Flow cytometry
Flow cytometry was used to evaluate the polarization of BV2 microglia and RAW264.7 macrophages. Cells in indicated groups were collected and incubated with specific antibodies (F$\frac{4}{80}$, iNOS and CD206) according to the manufacturer’s instructions.
## Oxygen consumption rate (OCR) measurement
Measurement of OCR was determined using a Seahorse XF96 Metabolic Flux Analyzer as described previously [10]. OCR was determined by addition of 2 µM oligomycin, 1 µM carbonyl cyanide 4-(trifluoromethoxy) phenylhydrazone (FCCP), 1 µM antimycin A as well as 1 µM rotenone (A&R) sequentially. XFe Data was determined using wave software (Seahorse Biosciences, USA). The basal respiration, ATP production, respiratory capacity and respiratory reserve were determined following the manufacturer’s protocol.
## ROS and mitochondrial membrane potential evaluation
ROS detection Kit (Beyotime, China) was used to determine ROS levels using 2′,7′-dichlorofluorsecein-diacetate (DCFH-DA) by flow cytometry analysis. Mitochondrial membrane potential was examined by JC-1 evaluation Kit (Beyotime, China) and quantified by aggregate-to-monomer fluorescence intensity ratio.
## Western blot
Cells and spinal cord samples were lysed using RIPA lysis buffer. Equal amounts of protein were separated by SDS-PAGE, transferred to PVDF membranes, blocked with $5\%$ BSA and incubated with specific primary antibodies at 4 ℃ overnight. Corresponding secondary antibodies were then incubated with membranes for 2 h followed by enhanced chemiluminescent reagent treatment. The expression of protein was evaluated using Image J software (National Institutes of Health, USA).
## Immunoprecipitation (IP)
RIPA lysis buffer containing protease inhibitors (Beyotime, China) were used to collect and lyse cells. The protein concentration was determined using bicinchoninic acid (BCA) assay (Thermo Fisher, USA). Control IgG together with 20 µL protein A/G plus-agarose (Santa Cruz Biotechnology, USA) were used to preclear cell lysates for 1 h followed by immunoprecipitated with specific antibodies and protein A/G plus-agarose at 4 ℃ overnight. After that, immunocomplexes were collected and washed five times with lysis buffer and boiled. At last, the bound protein was separated by SDS-PAGE for immunoblotting.
## In vivo
ubiquitylation assays
Endogenous IκBα ubiquitination was determined by immunoprecipitation with anti-IκBα antibody firstly followed by immunoblotted with anti-ubiquitin antibody. Ubiquitylation of exogenous IκBα was evaluated after transfection of Flag-tagged USP13 (WT or C345A mutant), Myc-tagged IκBα and HA-tagged ubiquitin in HEK 293T cells. Lysate proteins were precipitated and evaluated by immunoblotted with specific antibodies.
## Statistical analysis
All data are shown as mean ± standard deviation, and statistical analysis was performed in GraphPad Prism (version 8.0, GraphPad Software Inc., USA). Student’s t-test was used for comparisons between two groups, and one-way or two-way ANOVA followed by the post-hoc Tukey correction was used for multivariate analysis. P value < 0.05 was considered to be significant.
## Identification and characterization of exosomes derived from vascular endothelial cells
To elucidate the potential role of exosomes in the crosstalk between vascular endothelial cells and microglia/macrophages in SCI microenvironment, exosomes of microvascular endothelial cell line bEnd.3 cells were extracted, characterized and named Exos. TEM images showed that exosomes had cup-shaped or spherical morphology approximately 100 nm in size (Additional file 1: Fig. S1a). NTA results were used to generate the distribution curve of the nanoparticle size of exosomes (Additional file 1: Fig. S1b). Western blot assay was performed to confirm the exosomal surface markers (Additional file 1: Fig. S1c). Moreover, a Dil dye-labelled exosomes was applied to co-culture with target BV2 microglia for 24 h. Images showed that exosomes were present in cytoplasm, indicating successful uptake of exosomes by target cells (Additional file 1: Fig. S1d).
## Exos treatment promotes M2 polarization and ameliorates mitochondrial impairment of LPS-pretreated microglia/macrophages
in vitro
Microglia and macrophages are markers of SCI pathology and exert corresponding functions in response to neuroinflammation that subsequently contribute to secondary injury. Therefore, the role of exosomes in altering the polarization of LPS-pretreated microglia/macrophages was explored. bEnd.3 cells were cultured and endothelial cells conditional medium (ECM) was obtained. To further explore the underlying effects of bEnd.3 cells on microglia, BV2 microglia pretreated with LPS for 24 h was cultured with ECM. The findings showed that administration of ECM significantly decreased expression levels of M1 markers (iNOS, TNF-α, IL-1β) whereas increased expression levels of M2 markers (Arg1, CD206, YM$\frac{1}{2}$) (Fig. 1a). Moreover, flow cytometry analysis demonstrated a decrease of F$\frac{4}{80}$+ iNOS+ cells as well as an increase of F$\frac{4}{80}$+ CD206+ cells in ECM group compared to corresponding control (Fig. 1b–e). Given that exosomes are implicated in paracrine effects between cell crosstalk, and to further explore if bEnd.3 cells regulated microglia M2 polarization through exosomes, bEnd.3 cells were pretreated with GW4869, which is an exosomal secretion inhibitor. We further confirmed that inhibition of exosomal secretion obviously reversed ECM-mediated microglia M2 polarization (Fig. 1a–e). We also performed above related experiments in RAW264.7 macrophages and primary microglial cells and found similar results (Additional file 1: Fig. S2a–f). These findings show that exosomes might be involved in cell-to-cell communications between vascular endothelial cells and microglia/macrophages and are implicated in vascular endothelial cells-induced M2 polarization in microglia/macrophages in vitro.
Fig. 1Exos treatment promotes M2 polarization and ameliorates mitochondrial impairment of LPS-pretreated microglia/macrophages in vitro. a Detection of mRNA expression of M1 and M2 markers in indicated groups. ** $P \leq 0.01$, ***$P \leq 0.001.$ b, c Detection and quantification of microglia M1 polarization by flow cytometry. *** $P \leq 0.001.$ d, e Measurement and quantification of microglia M2 polarization by flow cytometry. *** $P \leq 0.001.$ f, g Flow cytometry detection and quantification of ROS level. *** $P \leq 0.001.$ h, i Detection and quantification of mitochondrial potential by JC-1 staining. *** $P \leq 0.001.$ j, k Measurement of OCR and quantification of basal respiration, ATP production, respiratory capacity and respiratory reserve. ** $P \leq 0.01$, ***$P \leq 0.001$ Several studies have investigated the link between microglia/macrophages polarization and mitochondrial functions [23, 24]. In the present study, we further investigated ROS levels which is also another M1 microglia/macrophages feature. The findings showed that ECM administration significantly reduced ROS level in BV2 microglia and this anti-ROS effect was abolished by inhibition of exosomal secretion (Fig. 1f, g). JC-1 staining results showed that mitochondrial potential was significantly restored after treating cells with ECM, and this beneficial effect was partly inhibited by administration of GW4869 (Fig. 1h, i). Moreover, ECM administration significantly increased OCR which is an oxidative phosphorylation biomarker, however, GW4869 administration downregulated OCR expression (Fig. 1j). In addition, mitochondrial activities including basal respiration, ATP production, respiration capacity, and respiration reverse in BV2 microglia were markedly upregulated by ECM administration but were reversed by addition of GW4869 (Fig. 1k).
## Exos treatment promotes better functional recovery post-injury
in vivo
To further explore the functions of vascular endothelial cells in SCI microenvironment and their potential work on microglia/macrophages, Exos were extracted and injected immediately post-injury. Functional motor analyses were performed at indicated timepoint. BMS score after SCI indicated that mice administered with Exos had markedly greater motor recovery in the course of the 28-day recovery period in contrast to those administered with PBS (Fig. 2a). Moreover, results from rotarod testing showed better functional motor rehabilitation in Exos group in comparison with the control group (Fig. 2b). Furthermore, electrophysiological analyses showed that MEPs in mice administered with Exos exhibited higher amplitude as well as shorter latency (Fig. 2c, d). These findings on behavioral tests indicate that administration of Exos promotes functional motor rehabilitation in mice post-injury.
Fig. 2Exos treatment promotes better functional recovery post-injury in vivo. a BMS scores in the course of the 28-day recovery period in indicated groups. *** $P \leq 0.001.$ b Evaluation of motor recovery by rotarod tests at day 28 after SCI. ** $$P \leq 0.0086$$, ***$$P \leq 0.0008.$$ c, d Representative images and quantification of MEP amplitude and latencies in indicated groups at day 28 post-injury. *** $P \leq 0.001.$ e, f Representative immunofluorescence staining of NF in spinal cords at day 28 post-injury and quantification of NF+ area to the total area of uninjured axons at indicated distances from the SCI lesion core (Scale bar = 1000 μm). * $$P \leq 0.036$$, **$$P \leq 0.0069$$, ***$P \leq 0.001.$ g, h Immunostaining of NeuN in spinal cords at day 28 post-injury and quantification of viable neurons in Z1–Z4 zones adjacent to lesion core (Scale bar = 1000 μm). *** $P \leq 0.001$ Also, axons adjacent the injured area was further investigated. Neurofilament positive (NF+) axons were assessed for evaluation of axonal regeneration. A considerably greater quantity of NF+ axons was found in Exos group at day 28 post injury compared to PBS group (Fig. 2e, f). NeuN, which is the neuronal marker, was applied to represent viable neurons in particular areas (Z1–Z4) located at different distances as previous described [16, 25]. Administration of Exos caused an obvious increase in number of NeuN+ neurons in the Z1–Z3 zones compared to the PBS group (Fig. 2g, h). These findings demonstrate that Exos administration promotes functional recovery and axonal regeneration in mice after SCI.
## Exos treatment promotes M2 polarization of microglia/macrophages in vivo
Expression level of M1 markers in Exos group was significantly decreased whereas M2 markers increased compared with PBS group (Fig. 3a). In addition, western blot analysis showed that lower protein levels of M1 markers and higher protein levels of M2 markers in the Exos group (Fig. 3b, c). Moreover, we used CD68, which represents activated microglia/macrophages, together with iNOS or Arg1 respectively to explore the M1/M2 microglia/macrophages polarization via double immunofluorescence staining in indicated groups post-injury. No significant difference was found in quantification of CD68+ cells (Fig. 3d–g). However, an obvious reduction in M1 microglia/macrophages and an obvious increase in M2 microglia/macrophages were observed after administration of Exos compared with the PBS group at day 7 post-injury in vivo (Fig. 3d–g). These findings indicate that Exos play key role in regulating microglia/macrophages M1/M2 polarization in vivo.
Fig. 3Exos treatment promotes M2 polarization of microglia/macrophages in vivo. a Detection of mRNA levels of M1 and M2 markers in injured spinal cords at day 7 post-injury.**$$P \leq 0.0016$$, ***$P \leq 0.001.$ b, c Detection and quantification of iNOS and Arg1 protein expression level at day 7 post-injury. ** $$P \leq 0.0058$$, ***$$P \leq 0.0005.$$ d, e Double-staining of CD68 and iNOS at day 7 after SCI and quantification of CD68+ cells and M1 microglia/macrophages (Scale bar = 100 μm). *** $$P \leq 0.0003.$$ f, g Double-staining of CD68 and Arg1 at day 7 after SCI and quantification of CD68+ cells and M2 microglia/macrophages (Scale bar = 100 μm). *** $$P \leq 0.0007$$
## Exos treatment promotes functional rehabilitation and M2 polarization of microglia/macrophages via delivering USP13 post-injury
Exosomes contain mRNAs, therefore, exosomal mRNAs may exert biological beneficial functions in vitro and in vivo as observed above in this study. To explore the mechanism underlying Exos functions, RNAs were extracted from BV2 microglia treated with Exos or PBS, and RNA-seq was performed. Analysis showed that the deubiquitinase USP13 was the most significantly upregulated mRNA in BV2 microglia treated with Exos (Fig. 4a). To verify the USP13 mRNA profile data, expression level of USP13 was analyzed and confirmed by qRT-PCR in vitro (Fig. 4b).
Fig. 4Exos treatment promote functional rehabilitation and M2 polarization of microglia/macrophages post-injury by delivering USP13. a Volcano plot of genes of BV2 microglia treated with PBS and Exos. Red and green dots represent up- and down-regulated DEGs, respectively. b mRNA expression level of USP13 in BV2 microglia treated with PBS and Exos. *** $$P \leq 0.0003.$$ c BMS scores in the course of the 28-day recovery period in indicated groups. *** $P \leq 0.001.$ d Evaluation of motor recovery by rotarod tests at day 28 after SCI. ** $P \leq 0.01.$ e Assessment of electromyography using MEP at day 28 after SCI. f Quantification of MEP amplitudes and latencies in indicated groups. ** $$P \leq 0.0010$$, ***$$P \leq 0.0005.$$ g, h Double-staining of CD68 and iNOS at day 7 after SCI and quantification of CD68+ cells and M1 microglia/macrophages (Scale bar = 100 μm). ** $$P \leq 0.0012.$$ i, j Double-staining of CD68 and Arg1 at day 7 after SCI and quantification of CD68+ cells and M2 microglia/macrophages (Scale bar = 100 μm). *** $P \leq 0.001$ To further explore the role of exosomal USP13, lentiviral-based methods were used to knockdown USP13 and the corresponding negative control in bEnd.3 cells (Additional file 1: Fig. S3a). Exosomes were then isolated followed by co-cultured with target BV2 microglia. USP13 was significantly silenced in shUSP13-Exos in comparison with shNC-Exos (Additional file 1: Fig. S3b). Moreover, USP13 was found to be obviously decreased in BV2 microglia treated with shUSP13-Exos (Additional file 1: Fig. S3c). Additionally, robust GFP fluorescence was observed in BV2 microglia incubated with exosomes from plasmid-loaded bEnd.3 cells, whereas no detectable fluorescence was found in microglia treated with exosomes from empty-vector bEnd.3 cells (Additional file 1: Fig. S3d). These results showed that exosomal USP13 was delivered to target BV2 microglia and could be associated with the beneficial functions of Exos post-injury.
Furthermore, shNC-Exos and shUSP13-Exos were administrated in vivo to explore the functional role of exosomal USP13 in Exos-regulated benefits after SCI. Analysis based on functional experiments as described above showed that mice administrated with shUSP13-Exos exhibited worse functional recovery compared with shNC-Exos (Fig. 4c–f). Moreover, administration of shUSP13-Exos markedly suppressed M2 microglia/macrophages polarization post-injury through qRT-PCR and western blot analysis (Additional file 1: Fig. S4a–c). The double-immunofluorescence experiments demonstrated consistent results (Fig. 4g–j). These findings indicate that Exos promote functional recovery and M2 polarization of microglia/macrophages via delivering USP13.
## Exos treatment regulates microglia M2 polarization and modulates mitochondrial function through transferring USP13
in vitro
shNC-Exos and shUSP13-Exos were administered to BV2 microglia to confirm the function of USP13. Silencing of USP13 in Exos evidently suppressed the M2 shift and upregulated ROS production in microglia in comparison with the control by qRT-PCR and flow cytometry analysis (Additional file 1: Fig. S5a–g). Notably, mitochondrial potential, OCR, and mitochondrial activities were decreased after administration of shUSP13-Exos (Additional file 1: Fig. S5h–k). These in vitro results demonstrate that exosomal USP13 could regulate microglia polarization and mitochondrial function.
## USP13 binds to and stabilizes IκBα
To further explore the underlying mechanism in USP13-regulated microglia/macrophages polarization, IP coupled with mass spectrum (IP/MS) was performed to determine which protein interacts with USP13. IκBα, which is a key player in inhibiting NF-κB signaling pathway activation, was identified as a protein which interacts with USP13 (Fig. 5a). Co-immunoprecipitation (Co-IP) analysis showed that endogenous IκBα could be precipitated by USP13 antibody and endogenous USP13 was precipitated by IκBα antibody as well in BV2 microglia (Fig. 5b). Furthermore, the interaction between exogenous USP13 and IκBα was investigated and confirmed in HEK 293T cells (Fig. 5c, d). In all, above findings show that USP13 binds to IκBα.
Fig. 5USP13 binds to and stabilizes IκBα. a Analysis by IP/MS indicated IκBα interacts with USP13. b Detection of endogenous protein interactions between USP13 and IκBα in BV2 microglia by Co-IP. c, d Detection of exogenous protein interactions between USP13 and IκBα in HEK 293T cells by Co-IP. Flag-tagged USP13 and Myc-tagged IκBα plasmids were transfected in HEK 239T cells. e IκBα mRNA level in indicated groups. f, g Detection and quantification of IκBα and USP13 protein level in indicated groups with or without proteasome inhibitor MG132 treatment by western blot analysis. *** $P \leq 0.001.$ h Increasing amounts of Flag-tagged USP13 (WT or C345A mutant) were transfected and the expression levels of IκBα and Flag-tagged USP13 were detected by western blot analysis. i, j Measurement and quantification of IκBα protein expression in indicated groups with cycloheximide (CHX, 10 µg/ml) treatment by western blot analysis. ** $$P \leq 0.0021$$, ***$P \leq 0.001$ USP13 binds to IκBα, therefore, the effect of downregulating USP13 in BV2 microglia on IκBα expression level was further investigated. As shown in Fig. 5e and Additional file 1: Fig. S6a, downregulation of USP13 significantly decreased IκBα protein level while had no significant effect on IκBα mRNA level. This result indicates that USP13 regulates the protein level instead of mRNA level of IκBα. Interestingly, addition of MG132 which is the proteasome inhibitor significantly reverse the decreased IκBα protein level when silencing of USP13 (Fig. 5f, g). Furthermore, knockdown of USP13 upregulated expression level of nuclear accumulation of p65 in BV2 microglia, implying that it activated the NF-κB signaling pathway (Additional file 1: Fig. S6a, b). Further analysis was performed to explore if IκBα can be stabilized by USP13. IκBα protein level was increased after overexpression of USP13, whereas the level was not influenced by overexpression of catalytically inactive C345A mutant USP13 (Fig. 5h). Next, we investigated the potential role of USP13 on the endogenous IκBα protein stability in the presence of protein synthesis inhibitor cycloheximide (CHX). As shown in Fig. 5i, j, USP13 overexpression significantly suppressed IκBα degradation while USP13 downregulation accelerated IκBα degradation.
USP13 is a deubiquitinase (DUB) and regulates IκBα stability, therefore, analyses were performed to explore if USP13 regulates IκBα ubiquitination. Silencing of USP13 significantly enhanced the ubiquitination level of IκBα but decreased the protein level compared with shCtrl (Fig. 6a, b). To further confirm the role of USP13 in the regulation on IκBα ubiquitination, Flag-USP13 (either WT or C345A mutant), Myc-IκBα as well as HA-Ub were co-transfected in HEK 293T cells. Analysis showed that overexpression of WT USP13 decreased IκBα ubiquitination while transfection of C345A mutant USP13 showed no significant influence (Fig. 6c, d). Effect of USP13 on IκBα ubiquitination was further investigated in vivo. Notably, ubiquitination level of spinal cords IκBα was remarkably inhibited in Exos group compared with PBS group, whereas it was promoted in shUSP13-Exos group compared with corresponding control after SCI (Fig. 6e, f). It’s recognized that Lys48- and Lys63-linked chains are two main types of polyubiquitin chains. Therefore, the form of polyubiquitin modification of IκBα protein regulated by USP13 was explored. As shown in Fig. 6g, USP13 cleaved Lys48-linked polyubiquitin chains instead of Lys63-linked on IκBα protein. To further verify that Lys48-linked polyubiquitination is essential for USP13-mediated IκBα protein stability, we transfected a Lys48 mutant (Lys48R) type of ubiquitin in USP13-silenced HEK 293T cells. It was shown that transfection of Lys48R ubiquitin reversed the influence of silencing of USP13 in decreasing IκBα protein level (Fig. 6h). These findings indicate that USP13 regulates IκBα stability in microglia as well as injured spinal cord.
Fig. 6USP13 suppresses IκBα ubiquitination. a, b Evaluation and quantification of endogenous IκBα ubiquitination in indicated groups in BV2 microglia. *** $P \leq 0.001.$ c, d Evaluation and quantification of exogenous IκBα ubiquitination in HEK 293T cells co-transfected with Flag-tagged USP13 (WT or C345A), HA-tagged Ub and Myc-tagged IκBα. *** $P \leq 0.001.$ e, f Evaluation and quantification of IκBα ubiquitination in spinal cords of indicated groups. *** $P \leq 0.001.$ g Evaluation of the IκBα ubiquitylation linkage in HEK 293T cells co-transfected with Flag-USP13, Myc-IκBα and the specific HA-Ub, Lys0, Lys48, or Lys63 plasmids. h Evaluation of IκBα and USP13 protein expression levels in HEK 293T cells transfected with Ub WT or Ub Lys48R in indicated groups
## Exosomal USP13 regulates microglia polarization and mitochondrial function through stabilizing IκBα
Next, IκBα was overexpressed in BV2 microglia and in vitro analyses were performed to further investigate the potential relationship between exosomal USP13 and IκBα in regulating microglia polarization. As shown in Fig. 7a–e, results showed that ectopic IκBα overexpression significantly reversed microglia polarization caused by administration of shUSP13-Exos and promoted M2 polarization in BV2 microglia. Furthermore, overexpression IκBα decreased ROS production induced by administration of shUSP13-Exos (Fig. 7f, g). Also, downregulation in mitochondrial potential, OCR and mitochondrial activities after administration of shUSP13-Exos were reversed when overexpression of IκBα (Fig. 7h–k). These results show that exosomal USP13 stabilizes IκBα and thus regulates microglia polarization as well as mitochondrial function in vitro.
Fig. 7Exosomal USP13 regulates microglia polarization and mitochondrial function through stabilizing IκBα. a–k Rescue experiments were performed to explore the functions of exosomal USP13/IκBα in microglia. Rescue experiments for exosomal USP13 deletion were conducted by ectopic overexpressing IκBα. a Detection of mRNA expression of M1 and M2 markers in indicated groups. *** $P \leq 0.001.$ b–e Examination and quantification of M1 and M2 microglia by flow cytometry in indicated groups. ** $$P \leq 0.0043$$, ***$$P \leq 0.0009.$$ ( f, g) Flow cytometry measurement and quantification of ROS level in indicated groups. *** $$P \leq 0.0004.$$ h, i Detection and quantification of mitochondrial potential by JC-1 staining. *** $$P \leq 0.0001.$$ j Measurement of OCR. k Quantification of mitochondrial activities. ** $$P \leq 0.0055$$, ***$P \leq 0.001$
## Exos treatment promotes functional rehabilitation and M2 polarization of microglia/macrophages through stabilizing IκBα thus inhibiting NF-κB signaling activation
To further confirm the functional role of IκBα in Exos-mediated benefits in vivo, the selective inhibitor of IκBα BA was used. BA or vehicle ($0.1\%$ DMSO) was administered through the intralumbar post-injury. Functional in vivo analyses showed that administration of BA significantly suppressed motor function recovery of mice treated with Exos in comparison with administration of DMSO (Fig. 8a–d). Furthermore, qRT-PCR, western blot and immunofluorescence results revealed that administration of BA significantly suppressed microglia/macrophages M2 polarization in mice treated with Exos compared to those administered with DMSO (Additional file 1: Fig S7a–c and Fig. 8e–h). These findings indicate that Exos promote functional rehabilitation and M2 polarization of microglia/macrophages via upregulating IκBα and subsequently inhibiting NF-κB signaling pathway in vivo.
Fig. 8Exos treatment promotes functional rehabilitation and M2 polarization of microglia/macrophages through stabilizing IκBα thus inhibiting NF-κB signaling activation in vivo. a BMS scores of Exos-administrated mice which were treated with DMSO or BA, a selective inhibitor of IκBα. *** $P \leq 0.001.$ b Evaluation of motor recovery by rotarod tests at day 28 after SCI. * $$P \leq 0.0302$$, **$$P \leq 0.0083.$$ c, d Representative images and quantification of MEP amplitude and latencies in indicated groups at day 28 post-injury. ** $P \leq 0.01.$ e, f Double-staining of CD68 and iNOS at day 7 after SCI and quantification of CD68+ cells and M1 microglia/macrophages (Scale bar = 100 μm). ** $$P \leq 0.0046.$$ g, h Double-staining of CD68 and Arg1 at day 7 after SCI and quantification of CD68+ cells and M2 microglia/macrophages (Scale bar = 100 μm). ** $$P \leq 0.0015$$
## Discussion
SCI causes irreversible motor and sensory disabilities with high mortality. BSCB is disrupted immediately after SCI resulting in subsequent spinal edema, neuroinflammation, neuronal cell death and glial activation [5, 6, 26]. Moreover, macrophages infiltrate into the central lesion after SCI. Infiltrated macrophages and the activated microglia are highly sensitive to SCI microenvironment and can be polarized to either M1 or M2 phenotype [9]. Therefore, it is essential to explore the underlying mechanism of the cell-to-cell communication in the SCI microenvironment and augment microglia/macrophages M2 polarization to inhibit detrimental neuroinflammation. In the present study, our results demonstrated that exosomes derived from vascular endothelial cells improve functional recovery and augment microglia/macrophages M2 polarization in vivo and in vitro. Notably, USP13 was highly upregulated in Exos and microglia/macrophages treated with Exos. Knockdown of USP13 in Exos reversed the beneficial functional effects of Exos. USP13 positively regulates microglia/macrophages polarization via suppressing ubiquitination-mediated degradation of IκBα. Addition of selective IκBα inhibitor BA abrogated the effects of Exos in vivo. These findings indicate a potential underlying mechanism in cell-to-cell communications between vascular endothelial cells and microglia/macrophages in SCI microenvironment. Moreover, the current study shows that exosomes derived from vascular endothelial cells are a promising therapeutic way for treating SCI and elucidates the underlying mechanisms.
Therapeutic effects of exosomes in promoting tissue regeneration as well as treating several disorders have been reported previously. Moreover, exosomes can cross the BBB easily and are promising agents in treating CNS diseases owing to their unique advantages including nano-sized and membrane-permeable characteristics [13, 27]. In addition, previous studies have explored the potential of using exosomes as genetic material or drug delivery vectors across BBB [28–30]. Previous studies on CNS injury reported that exosomes of stem cells remarkedly promote functional recovery after SCI, brain injury and stroke in mice [13, 17, 27, 31, 32]. Moreover, exosomes exhibited protective effect in treatment of myocardial infarction, cardiomyopathy and ischemia-reperfusion injury in nonhuman primates and preclinical trails [33–36]. Notably, clinical trial was performed to explore role of exosomes in the potential therapy of type I diabetes mellitus. In fact, the first patient with graft versus host disease was effectively treated with exosomes [37].
Microglia/macrophages play a dual role after SCI by exhibiting pro-inflammatory and anti-inflammatory phenotypes. Excessive M1 microglia/macrophages activation induces production of high-level proinflammatory cytokines resulting in severer chronic inflammation [9]. Therefore, recent publications have reported several approaches including pharmaceutic administration, gene therapy, MSCs transplantation and exosomes injection to regulate microglia/macrophages polarization [38]. This present study showed that conditional medium from vascular endothelial cells promotes M2 polarization, reduces ROS production and regulates mitochondrial function in BV2 microglia as well as RAW264.7 macrophages. Notably, these beneficial effects were significantly inhibited by the presence of GW4869 which is an exosomal secretion inhibitor. In vitro analyses showed that exosomes from endothelial cells increase microglia/macrophages M2 polarization and exosome-based signals are essential in crosstalk between vascular endothelial cells and microglia/macrophages. Furthermore, in vivo analysis showed that administration of exosomes improves functional recovery after SCI. Moreover, the potential mechanism in exosomes-based therapy in SCI were explored. Exosomes contain mRNAs, proteins and miRNAs which can be transferred to target cells and play their corresponding biological effects [39]. A recent study reported that LCP1 derived from BMSCs can be packed into exosomes and transmitted to osteosarcoma cells, thus promoting tumorigenesis and metastasis [40]. In addition, serum exosomes contain ECRG4 could suppress tumor growth [41]. Lv et al. reported that exosomal CCL2 from tubular epithelial cells plays an important role in tubulointerstitial inflammation [21]. Moreover, exosomal CCL2 mRNA is a biomarker of active histological injury in IgA nephropathy [42]. Therefore, RNA-seq was performed in BV2 microglia treated with Exos or PBS. The findings showed that USP13 is highly upregulated in microglia treated with Exos. Furthermore, downregulation of USP13 in Exos partially abrogated the beneficial effects of Exos, which indicates that Exos-mediated favorable effects in treatment of SCI through transferring USP13.
Due to the presence of deubiquitinating enzymes (known as DUBs), ubiquitination is a reversible process. DUBs can cleave ubiquitin molecules from modified proteins. It has been reported that microglia/macrophages polarization is regulated by ubiquitination [15]. A recent study showed that USP19/NLRP3 axis shift macrophages polarization to M2 and thus promote anti-inflammatory response [43]. Wang et al. reported M2 macrophages polarization was inhibited by suppressing Ubc9 [44]. Another study proved that USP10 promotes M2 macrophages polarization by deubiquitinating NLRP7 [45]. Furthermore, other studies reported that USP8, A20 as well as PHLDA1 regulate microglia polarization through different mechanisms [46–48]. This present study showed that USP13 was markedly increased in Exos and regulated the polarization of macrophages/microglia. USP13 is a member of the DUBs family and plays key roles in various biological processes including tumor promotion, inflammation, apoptosis, drug resistance and anti-viral responses by cleaving ubiquitin molecules from associated proteins including STING, Myc, PTEN and Mcl-1 [49–52]. However, the detailed function of USP13 in neuroscience, mainly in regulating macrophages/microglia polarization after CNS injury has not been fully explored.
To investigate the potential mechanisms underlying the effects of USP13 on regulating M2 microglia/macrophages polarization, IP/MS was performed and IκBα was identified as a potential target that binds to USP13 in microglia. IκBα is a subunit in the IKK complex. Transcriptional activation of NF-κB occurs when the IKK complex is activated and IκB proteins are phosphorylated. IκBα strongly binds to and sequesters NF-κB in the cytoplasm of resting cells. Stimulation of the IκBα kinase/IKK complex promotes phosphorylation of serine $\frac{32}{36}$ of IκBα. Phosphorylation of IκBα promotes degradation of IκBα and releases NF-κB, which translocates to the nucleus [53–55]. Previous studies reported several proteins which target and control degradation and stability of IκBα. For example, Ji et al. reported that TRIM22 is implicated in promoting degradation of IκBα, thus activating NF-κB signaling [56]. A previous study reported that β-TrCP1 is involved in proteasomal-mediated degradation of IκBα [57]. Li et al. reported that USP34 inhibits NF-κB signaling activation by stabilizing IκBα [58]. In the present study, a novel DUB named USP13 which stabilizes IκBα was identified. The findings showed that USP13 interacts with IκBα and inhibits the ubiquitination of IκBα, leading to reduced degradation of IκBα in vitro and in vivo. Moreover, administration of IκBα inhibitor in vivo abolished M2 microglia/macrophages polarization resulted from Exos administration, suggesting that the favorable benefits of Exos on regulating microglia/macrophages polarization depends on the stability of IκBα. In all, this study suggests that USP13 regulates the polarization of microglia/macrophages, in part through promoting IκBα stability.
## Conclusion
In summary, this current study uncovered that exosomes derived from vascular endothelial cells improve functional recovery and augment microglia/macrophages M2 polarization via delivering USP13, which subsequently inhibits IκBα ubiquitination and degradation (Additional file 1: Fig. S8). This study provides a potential underlying mechanism in cell-to-cell communication between vascular endothelial cells and microglia/macrophages and show that exosomes are a promising approach for SCI treatment.
## Supplementary Information
Additional file 1. Additional figures.
## References
1. Simpson LA, Eng JJ, Hsieh JT, Wolfe DL. **Spinal cord injury rehabilitation evidence Scire research, the health and life priorities of individuals with spinal cord injury: a systematic review**. *J Neurotrauma* (2012) **29** 1548-55. DOI: 10.1089/neu.2011.2226
2. Ahuja CS, Wilson JR, Nori S, Kotter MRN, Druschel C, Curt A, Fehlings MG. **Traumatic spinal cord injury**. *Nat Rev Dis Primers* (2017) **3** 17018. DOI: 10.1038/nrdp.2017.18
3. Anwar MA, Al Shehabi TS, Eid AH. **Inflammogenesis of secondary spinal cord Injury**. *Front Cell Neurosci* (2016) **10** 98. DOI: 10.3389/fncel.2016.00098
4. Silva NA, Sousa N, Reis RL, Salgado AJ. **From basics to clinical: a comprehensive review on spinal cord injury**. *Prog Neurobiol* (2014) **114** 25-57. DOI: 10.1016/j.pneurobio.2013.11.002
5. Bartanusz V, Jezova D, Alajajian B, Digicaylioglu M. **The blood-spinal cord barrier: morphology and clinical implications**. *Ann Neurol* (2011) **70** 194-206. DOI: 10.1002/ana.22421
6. Jin LY, Li J, Wang KF, Xia WW, Zhu ZQ, Wang CR, Li XF, Liu HY. **Blood-spinal cord barrier in spinal cord Injury: a review**. *J Neurotrauma* (2021) **38** 1203-24. DOI: 10.1089/neu.2020.7413
7. Hu X, Leak RK, Shi Y, Suenaga J, Gao Y, Zheng P, Chen J. **Microglial and macrophage polarization-new prospects for brain repair**. *Nat Rev Neurol* (2015) **11** 56-64. DOI: 10.1038/nrneurol.2014.207
8. Kong X, Gao J. **Macrophage polarization: a key event in the secondary phase of acute spinal cord injury**. *J Cell Mol Med* (2017) **21** 941-54. DOI: 10.1111/jcmm.13034
9. Zhou X, He X, Ren Y. **Function of microglia and macrophages in secondary damage after spinal cord injury**. *Neural Regen Res* (2014) **9** 1787-95. DOI: 10.4103/1673-5374.143423
10. Ge X, Tang P, Rong Y, Jiang D, Lu X, Ji C, Wang J, Huang C, Duan A, Liu Y, Chen X, Chen X, Xu Z, Wang F, Wang Z, Li X, Zhao W, Fan J, Liu W, Yin G, Cai W. **Exosomal miR-155 from M1-polarized macrophages promotes EndoMT and impairs mitochondrial function via activating NF-kappaB signaling pathway in vascular endothelial cells after traumatic spinal cord injury**. *Redox Biol* (2021) **41** 101932. DOI: 10.1016/j.redox.2021.101932
11. Hessvik NP, Llorente A. **Current knowledge on exosome biogenesis and release**. *Cell Mol Life Sci* (2018) **75** 193-208. DOI: 10.1007/s00018-017-2595-9
12. Chevillet JR, Kang Q, Ruf IK, Briggs HA, Vojtech LN, Hughes SM, Cheng HH, Arroyo JD, Meredith EK, Gallichotte EN, Pogosova-Agadjanyan EL, Morrissey C, Stirewalt DL, Hladik F, Yu EY, Higano CS, Tewari M. **Quantitative and stoichiometric analysis of the microRNA content of exosomes**. *Proc Natl Acad Sci U S A* (2014) **111** 14888-93. DOI: 10.1073/pnas.1408301111
13. Chen J, Chopp M. **Exosome therapy for stroke**. *Stroke* (2018) **49** 1083-90. DOI: 10.1161/STROKEAHA.117.018292
14. Wortzel I, Dror S, Kenific CM, Lyden D. **Exosome-mediated metastasis: communication from a distance**. *Dev Cell* (2019) **49** 347-60. DOI: 10.1016/j.devcel.2019.04.011
15. Liu W, Tang P, Wang J, Ye W, Ge X, Rong Y, Ji C, Wang Z, Bai J, Fan J, Yin G, Cai W. **Extracellular vesicles derived from melatonin-preconditioned mesenchymal stem cells containing USP29 repair traumatic spinal cord injury by stabilizing NRF2**. *J Pineal Res* (2021) **71** e12769. DOI: 10.1111/jpi.12769
16. Liu W, Ge X, Zhou Z, Jiang D, Rong Y, Wang J, Ji C, Fan J, Yin G, Cai W. **Deubiquitinase USP18 regulates reactive astrogliosis by stabilizing SOX9**. *Glia* (2021) **69** 1782-98. DOI: 10.1002/glia.23992
17. Jiang D, Gong F, Ge X, Lv C, Huang C, Feng S, Zhou Z, Rong Y, Wang J, Ji C, Chen J, Zhao W, Fan J, Liu W, Cai W. **Neuron-derived exosomes-transmitted mir-124-3p protect traumatically injured spinal cord by suppressing the activation of neurotoxic microglia and astrocytes**. *J Nanobiotechnol* (2020) **18** 105. DOI: 10.1186/s12951-020-00665-8
18. Liu W, Rong Y, Wang J, Zhou Z, Ge X, Ji C, Jiang D, Gong F, Li L, Chen J, Zhao S, Kong F, Gu C, Fan J, Cai W. **Exosome-shuttled miR-216a-5p from hypoxic preconditioned mesenchymal stem cells repair traumatic spinal cord injury by shifting microglial M1/M2 polarization**. *J Neuroinflammation* (2020) **17** 47. DOI: 10.1186/s12974-020-1726-7
19. Bellampalli SS, Ji Y, Moutal A, Cai S, Wijeratne EMK, Gandini MA, Yu J, Chefdeville A, Dorame A, Chew LA, Madura CL, Luo S, Molnar G, Khanna M, Streicher JM, Zamponi GW, Gunatilaka AAL, Khanna R. **Betulinic acid, derived from the desert lavender Hyptis emoryi, attenuates paclitaxel-, HIV-, and nerve injury-associated peripheral sensory neuropathy via block of N- and T-type calcium channels**. *Pain* (2019) **160** 117-35. DOI: 10.1097/j.pain.0000000000001385
20. Li M, Rong ZJ, Cao Y, Jiang LY, Zhong D, Li CJ, Sheng XL, Hu JZ, Lu HB. **Utx regulates the NF-kappaB signaling pathway of natural stem cells to modulate macrophage migration during spinal cord Injury**. *J Neurotrauma* (2021) **38** 353-64. DOI: 10.1089/neu.2020.7075
21. Lv LL, Feng Y, Wen Y, Wu WJ, Ni HF, Li ZL, Zhou LT, Wang B, Zhang JD, Crowley SD, Liu BC. **Exosomal CCL2 from tubular epithelial cells is critical for albumin-induced tubulointerstitial inflammation**. *J Am Soc Nephrol* (2018) **29** 919-35. DOI: 10.1681/ASN.2017050523
22. Zhao SJ, Kong FQ, Jie J, Li Q, Liu H, Xu AD, Yang YQ, Jiang B, Wang DD, Zhou ZQ, Tang PY, Chen J, Wang Q, Zhou Z, Chen Q, Yin GY, Zhang HW, Fan J. **Macrophage MSR1 promotes BMSC osteogenic differentiation and M2-like polarization by activating PI3K/AKT/GSK3beta/beta-catenin pathway**. *Theranostics* (2020) **10** 17-35. DOI: 10.7150/thno.36930
23. Weinberg SE, Sena LA, Chandel NS. **Mitochondria in the regulation of innate and adaptive immunity**. *Immunity* (2015) **42** 406-17. DOI: 10.1016/j.immuni.2015.02.002
24. Formentini L, Santacatterina F, Nunez de Arenas C, Stamatakis K, Lopez-Martinez D, Logan A, Fresno M, Smits R, Murphy MP, Cuezva JM. **Mitochondrial ROS production protects the intestine from inflammation through functional M2 macrophage polarization**. *Cell Rep* (2017) **19** 1202-13. DOI: 10.1016/j.celrep.2017.04.036
25. Wanner IB, Anderson MA, Song B, Levine J, Fernandez A, Gray-Thompson Z, Ao Y, Sofroniew MV. **Glial scar borders are formed by newly proliferated, elongated astrocytes that interact to corral inflammatory and fibrotic cells via STAT3-dependent mechanisms after spinal cord injury**. *J Neurosci* (2013) **33** 12870-86. DOI: 10.1523/JNEUROSCI.2121-13.2013
26. Rong Y, Fan J, Ji C, Wang Z, Ge X, Wang J, Ye W, Yin G, Cai W, Liu W. **USP11 regulates autophagy-dependent ferroptosis after spinal cord ischemia-reperfusion injury by deubiquitinating beclin 1**. *Cell Death Differ* (2022) **29** 1164-75. DOI: 10.1038/s41418-021-00907-8
27. Liu W, Wang Y, Gong F, Rong Y, Luo Y, Tang P, Zhou Z, Zhou Z, Xu T, Jiang T, Yang S, Yin G, Chen J, Fan J, Cai W. **Exosomes Derived from Bone mesenchymal stem cells repair traumatic spinal cord Injury by suppressing the activation of A1 neurotoxic reactive astrocytes**. *J Neurotrauma* (2019) **36** 469-84. DOI: 10.1089/neu.2018.5835
28. Didiot MC, Hall LM, Coles AH, Haraszti RA, Godinho BM, Chase K, Sapp E, Ly S, Alterman JF, Hassler MR, Echeverria D, Raj L, Morrissey DV, DiFiglia M, Aronin N, Khvorova A. **Exosome-mediated delivery of hydrophobically modified siRNA for huntingtin mRNA silencing**. *Mol Ther* (2016) **24** 1836-47. DOI: 10.1038/mt.2016.126
29. Yuan D, Zhao Y, Banks WA, Bullock KM, Haney M, Batrakova E, Kabanov AV. **Macrophage exosomes as natural nanocarriers for protein delivery to inflamed brain**. *Biomaterials* (2017) **142** 1-12. DOI: 10.1016/j.biomaterials.2017.07.011
30. Yang J, Wu S, Hou L, Zhu D, Yin S, Yang G, Wang Y. **Therapeutic eEffects of simultaneous delivery of nerve growth factor mRNA and protein via exosomes on cerebral ischemia**. *Mol Ther Nucleic Acids* (2020) **21** 512-22. DOI: 10.1016/j.omtn.2020.06.013
31. Venkat P, Chopp M, Chen J. **Cell-based and exosome therapy in diabetic stroke**. *Stem Cells Transl Med* (2018) **7** 451-5. DOI: 10.1002/sctm.18-0014
32. Xiong Y, Mahmood A, Chopp M. **Emerging potential of exosomes for treatment of traumatic brain injury**. *Neural Regen Res* (2017) **12** 19-22. DOI: 10.4103/1673-5374.198966
33. Li Q, Xu Y, Lv K, Wang Y, Zhong Z, Xiao C, Zhu K, Ni C, Wang K, Kong M, Li X, Fan Y, Zhang F, Chen Q, Li Y, Li Q, Liu C, Zhu J, Zhong S, Wang J, Chen Y, Zhao J, Zhu D, Wu R, Chen J, Zhu W, Yu H, Ardehali R, Zhang JJ, Wang J, Hu X. **Small extracellular vesicles containing miR-486-5p promote angiogenesis after myocardial infarction in mice and nonhuman primates**. *Sci Transl Med* (2021) **13** eabb0202. DOI: 10.1126/scitranslmed.abb0202
34. Antounians L, Catania VD, Montalva L, Liu BD, Hou H, Chan C, Matei AC, Tzanetakis A, Li B, Figueira RL, da Costa KM, Wong AP, Mitchell R, David AL, Patel K, De Coppi P, Sbragia L, Wilson MD, Rossant J, Zani A. **Fetal lung underdevelopment is rescued by administration of amniotic fluid stem cell extracellular vesicles in rodents**. *Sci Transl Med* (2021) **13** eaax5941. DOI: 10.1126/scitranslmed.aax5941
35. Hirai K, Ousaka D, Fukushima Y, Kondo M, Eitoku T, Shigemitsu Y, Hara M, Baba K, Iwasaki T, Kasahara S, Ohtsuki S, Oh H. **Cardiosphere-derived exosomal microRNAs for myocardial repair in pediatric dilated cardiomyopathy**. *Sci Transl Med* (2020) **12** eabb3336. DOI: 10.1126/scitranslmed.abb3336
36. Yadid M, Lind JU, Ardona HAM, Sheehy SP, Dickinson LE, Eweje F, Bastings MMC, Pope B, O’Connor BB, Straubhaar JR, Budnik B, Kleber AG, Parker KK. **Endothelial extracellular vesicles contain protective proteins and rescue ischemia-reperfusion injury in a human heart-on-chip**. *Sci Transl Med* (2020) **12** eaax8005. DOI: 10.1126/scitranslmed.aax8005
37. Kordelas L, Rebmann V, Ludwig AK, Radtke S, Ruesing J, Doeppner TR, Epple M, Horn PA, Beelen DW, Giebel B. **MSC-derived exosomes: a novel tool to treat therapy-refractory graft-versus-host disease**. *Leukemia* (2014) **28** 970-3. DOI: 10.1038/leu.2014.41
38. Kanazawa M, Ninomiya I, Hatakeyama M, Takahashi T, Shimohata T. **Microglia and monocytes/macrophages polarization reveal novel therapeutic mechanism against stroke**. *Int J Mol Sci.* (2017) **18** 2135. DOI: 10.3390/ijms18102135
39. Valadi H, Ekstrom K, Bossios A, Sjostrand M, Lee JJ, Lotvall JO. **Exosome-mediated transfer of mRNAs and microRNAs is a novel mechanism of genetic exchange between cells**. *Nat Cell Biol* (2007) **9** 654-9. DOI: 10.1038/ncb1596
40. Ge X, Liu W, Zhao W, Feng S, Duan A, Ji C, Shen K, Liu W, Zhou J, Jiang D, Rong Y, Gong F, Wang J, Xu Z, Li X, Fan J, Wei Y, Bai J, Cai W. **Exosomal transfer of LCP1 promotes osteosarcoma cell tumorigenesis and metastasis by activating the JAK2/STAT3 signaling pathway**. *Mol Ther Nucleic Acids* (2020) **21** 900-15. DOI: 10.1016/j.omtn.2020.07.025
41. Mao L, Li X, Gong S, Yuan H, Jiang Y, Huang W, Sun X, Dang X. **Serum exosomes contain ECRG4 mRNA that suppresses tumor growth via inhibition of genes involved in inflammation, cell proliferation, and angiogenesis**. *Cancer Gene Ther* (2018) **25** 248-59. DOI: 10.1038/s41417-018-0032-3
42. Feng Y, Lv LL, Wu WJ, Li ZL, Chen J, Ni HF, Zhou LT, Tang TT, Wang FM, Wang B, Chen PS, Crowley SD, Liu BC. **Urinary Exosomes and Exosomal CCL2 mRNA as biomarkers of active histologic Injury in IgA Nephropathy**. *Am J Pathol* (2018) **188** 2542-52. DOI: 10.1016/j.ajpath.2018.07.017
43. Liu T, Wang L, Liang P, Wang X, Liu Y, Cai J, She Y, Wang D, Wang Z, Guo Z, Bates S, Xia X, Huang J, Cui J. **USP19 suppresses inflammation and promotes M2-like macrophage polarization by manipulating NLRP3 function via autophagy**. *Cell Mol Immunol* (2021) **18** 2431-42. DOI: 10.1038/s41423-020-00567-7
44. Wang F, Sun F, Luo J, Yue T, Chen L, Zhou H, Zhang J, Yang C, Luo X, Zhou Q, Zhu H, Li J, Yang P, Xiong F, Yu Q, Zhang H, Zhang W, Xu A, Zhou Z, Lu Q, Eizirik DL, Zhang S, Wang CY. **Loss of ubiquitin-conjugating enzyme E2 (Ubc9) in macrophages exacerbates multiple low-dose streptozotocin-induced diabetes by attenuating M2 macrophage polarization**. *Cell Death Dis* (2019) **10** 892. DOI: 10.1038/s41419-019-2130-z
45. Li B, Qi ZP, He DL, Chen ZH, Liu JY, Wong MW, Zhang JW, Xu EP, Shi Q, Cai SL, Sun D, Yao LQ, Zhou PH, Zhong YS. **NLRP7 deubiquitination by USP10 promotes tumor progression and tumor-associated macrophage polarization in colorectal cancer**. *J Exp Clin Cancer Res* (2021) **40** 126. DOI: 10.1186/s13046-021-01920-y
46. Han C, Yan P, He T, Cheng J, Zheng W, Zheng LT, Zhen X. **PHLDA1 promotes microglia-mediated neuroinflammation via regulating K63-linked ubiquitination of TRAF6**. *Brain Behav Immun* (2020) **88** 640-53. DOI: 10.1016/j.bbi.2020.04.064
47. Kinsella S, Fichtner M, Watters O, Konig HG, Prehn JHM. **Increased A20-E3 ubiquitin ligase interactions in bid-deficient glia attenuate TLR3- and TLR4-induced inflammation**. *J Neuroinflammation* (2018) **15** 130. DOI: 10.1186/s12974-018-1143-3
48. Zhao J, Bi W, Zhang J, Xiao S, Zhou R, Tsang CK, Lu D, Zhu L. **USP8 protects against lipopolysaccharide-induced cognitive and motor deficits by modulating microglia phenotypes through TLR4/MyD88/NF-kappaB signaling pathway in mice**. *Brain Behav Immun* (2020) **88** 582-96. DOI: 10.1016/j.bbi.2020.04.052
49. Zhang J, Zhang P, Wei Y, Piao HL, Wang W, Maddika S, Wang M, Chen D, Sun Y, Hung MC, Chen J, Ma L. **Deubiquitylation and stabilization of PTEN by USP13**. *Nat Cell Biol* (2013) **15** 1486-94. DOI: 10.1038/ncb2874
50. Fang X, Zhou W, Wu Q, Huang Z, Shi Y, Yang K, Chen C, Xie Q, Mack SC, Wang X, Carcaboso AM, Sloan AE, Ouyang G, McLendon RE, Bian XW, Rich JN, Bao S. **Deubiquitinase USP13 maintains glioblastoma stem cells by antagonizing FBXL14-mediated myc ubiquitination**. *J Exp Med* (2017) **214** 245-67. DOI: 10.1084/jem.20151673
51. Sun H, Zhang Q, Jing YY, Zhang M, Wang HY, Cai Z, Liuyu T, Zhang ZD, Xiong TC, Wu Y, Zhu QY, Yao J, Shu HB, Lin D, Zhong B. **USP13 negatively regulates antiviral responses by deubiquitinating STING**. *Nat Commun* (2017) **8** 15534. DOI: 10.1038/ncomms15534
52. Zhang S, Zhang M, Jing Y, Yin X, Ma P, Zhang Z, Wang X, Di W, Zhuang G. **Deubiquitinase USP13 dictates MCL1 stability and sensitivity to BH3 mimetic inhibitors**. *Nat Commun* (2018) **9** 215. DOI: 10.1038/s41467-017-02693-9
53. Chen ZJ, Parent L, Maniatis T. **Site-specific phosphorylation of IkappaBalpha by a novel ubiquitination-dependent protein kinase activity**. *Cell* (1996) **84** 853-62. DOI: 10.1016/S0092-8674(00)81064-8
54. Chen ZJ. **Ubiquitination in signaling to and activation of IKK**. *Immunol Rev* (2012) **246** 95-106. DOI: 10.1111/j.1600-065X.2012.01108.x
55. Liu S, Chen ZJ. **Expanding role of ubiquitination in NF-kappaB signaling**. *Cell Res* (2011) **21** 6-21. DOI: 10.1038/cr.2010.170
56. Ji J, Ding K, Luo T, Zhang X, Chen A, Zhang D, Li G, Thorsen F, Huang B, Li X, Wang J. **TRIM22 activates NF-kappaB signaling in glioblastoma by accelerating the degradation of IkappaBalpha**. *Cell Death Differ* (2021) **28** 367-81. DOI: 10.1038/s41418-020-00606-w
57. Hatakeyama S, Kitagawa M, Nakayama K, Shirane M, Matsumoto M, Hattori K, Higashi H, Nakano H, Okumura K, Onoe K, Good RA, Nakayama K. **Ubiquitin-dependent degradation of IkappaBalpha is mediated by a ubiquitin ligase Skp1/Cul 1/F-box protein FWD1**. *Proc Natl Acad Sci U S A* (1999) **96** 3859-63. DOI: 10.1073/pnas.96.7.3859
58. Li Q, Wang M, Xue H, Liu W, Guo Y, Xu R, Shao B, Yuan Q. **Ubiquitin-specific protease 34 inhibits Osteoclast differentiation by regulating NF-kappaB signaling**. *J Bone Miner Res* (2020) **35** 1597-608. DOI: 10.1002/jbmr.4015
|
---
title: 'Outcomes in critically Ill HIV-infected patients between 1997 and 2020: analysis
of the OUTCOMEREA multicenter cohort'
authors:
- Antoine Gaillet
- Elie Azoulay
- Etienne de Montmollin
- Maité Garrouste-Orgeas
- Yves Cohen
- Claire Dupuis
- Carole Schwebel
- Jean Reignier
- Shidasp Siami
- Laurent Argaud
- Christophe Adrie
- Bruno Mourvillier
- Stéphane Ruckly
- Jean-Marie Forel
- Jean-Francois Timsit
journal: Critical Care
year: 2023
pmcid: PMC10012467
doi: 10.1186/s13054-023-04325-9
license: CC BY 4.0
---
# Outcomes in critically Ill HIV-infected patients between 1997 and 2020: analysis of the OUTCOMEREA multicenter cohort
## Abstract
### Purpose
Despite antiviral therapy (ART), 800,000 deaths still occur yearly and globally due to HIV infection. In parallel with the good virological control and the aging of this population, multiple comorbidities [HIV-associated-non-AIDS (HANA) conditions] may now be observed.
### Methods
HIV adult patients hospitalized in intensive care unit (ICU) from all the French region from university and non-university hospital who participate to the OutcomeRea™ database on a voluntary basis over a 24-year period.
### Results
Of the 24,298 stays registered, 630 ($2.6\%$) were a first ICU stay for HIV patients. Over time, the mean age and number of comorbidities (diabetes, renal and respiratory history, solid neoplasia) of patients increased. The proportion of HIV diagnosed on ICU admission decreased significantly, while the median duration of HIV disease as well as the percentage of ART-treated patients increased. The distribution of main reasons for admission remained stable over time (acute respiratory distress > shock > coma). We observed a significant drop in the rate of active opportunistic infection on admission, while the rate of active hemopathy (newly diagnosed or relapsed within the last 6 months prior to admission to ICU) qualifying for AIDS increased—nonsignificantly—with a significant increase in the anticancer chemotherapy administration in ICU. Admissions for HANA or non-HIV reasons were stable over time. In multivariate analysis, predictors of 60-day mortality were advanced age, chronic liver disease, past chemotherapy, sepsis-related organ failure assessment score > 4 at admission, hospitalization duration before ICU admission > 24 h, AIDS status, but not the period of admission.
### Conclusion
Whereas the profile of ICU-admitted HIV patients has evolved over time (HIV better controlled but more associated comorbidities), mortality risk factors remain stable, including AIDS status.
## Introduction
The Human Immunodeficiency Virus (HIV) pandemic remains a major public health issue with 1.8 million new infections and 800,000 deaths per year [1].
With the development of triple antiretroviral therapy (ART) from 1996 onwards, which allows a better control of HIV, and the improvement of resuscitation techniques (especially ventilation), the prognosis of HIV patients has dramatically improved over the last 25 years [2]. As a result, the HIV population is becoming older, with increasing multiple comorbidities (cirrhosis, chronic obstructive pulmonary disease, renal failure, atherosclerosis, neoplasia), grouped under the term “HIV-Associated Non-Acquired immunodeficiency syndrome (AIDS)” (HANA) conditions. These patients may also be admitted for various symptoms and disease not specific to HIV infection (intoxications, community-acquired co-infections), or related to specific therapies (ART toxicity); as such, they may be present while HIV replication is low or undetectable [2–6].
Although the hospitalization rate of HIV patients has decreased over time (600 vs. 140 per 1000 patient-years in 1999 vs. 2007) [7–9], compared to non-HIV patients, they remain at higher risk ($50\%$ excess risk) of admission to intensive care unit (ICU) [6, 10]. Nevertheless, their mortality in ICU in Western countries tends to be similar to that of non-HIV patients [11–19].
The persistence of the pandemic and the phenotypic evolution of HIV patients over time makes relevant to investigate the epidemiology of HIV patients in ICU.
The objectives of this observational study were first to describe the phenotypic characteristics of unselected HIV patients admitted in ICUs from 1997 to 2020, using the French prospective cohort OutcomeRea™, and then to investigate the risk factors for the 60-day mortality after admission to the ICU.
## Methods
The reporting used in this article follows the STROBE recommendations [20].
## Study design and data sources
We conducted an analysis using the prospectively collected data from 1997 to 2020 from all the French region from university and non-university hospital who participate to the OutcomeRea™ database on a voluntary basis ($$n = 23$$ centers). The OutcomeRea™ database contains data on admission features and diagnosis, daily disease severity, iatrogenic events, nosocomial infections, vital status and decision to forgo life-sustaining therapy (DFLST). Each participating ICU chose to perform sampling by taking either consecutive admissions to randomly selected ICU beds throughout the year or randomly consecutive admissions to all ICU beds over a single month. The data-capture software automatically conducted multiple checks for internal consistency of most of the variables at entry in the database. *Queries* generated by these checks were resolved with the source ICU before incorporation of the new data into the database. At each participating ICU, data quality was controlled by having a senior physician from another participating ICU check a $2\%$ random sample of the study data. A 1-day coding course is organized annually with the study investigators and contrast research organization monitors. Further details on data collection and quality were described elsewhere [21]. Note that some additional variables were deduced from the database and constructed secondarily (compliance, precariousness).
The OutcomeRea™ database was declared to the “Comité consultatif français de l'informatique pour la recherche en santé” (CCTIRS) et la “Commission française de l'informatique et des libertés” (CNIL, #8,999,262), in accordance with French law, and this study was approved by the ethical committee of the French Society of Intensive Care (SRLF). Waiver for informed consent was granted because the study does not modify patients’ management and the data are anonymously collected.
## Study population
All adult (≥ 18 years) patients diagnosed with HIV or AIDS and registered in the OutcomeRea™ database from 1997 to 2020 were included. The first ICU stay of a patient during the same hospitalization was the only included.
## Definitions
HIV or AIDS, and preexisting chronic organ failures (including respiratory, cardiac, hepatic, renal replacement therapy) were defined according to the Knaus classification [22]. AIDS status was defined as the late stage of HIV infection, i.e., when the number of their CD4 cells fell < 200 cells/mm3 or if an “opportunistic affection” (infectious disease (such as pneumocystis or toxoplasmosis) or hemopathy (such as non-*Hodgkin lymphoma* or Kaposi's sarcoma) [23]) qualifying for AIDS was developed regardless of their CD4 count. HANA conditions were defined as chronic obstructive pulmonary disease, coronary artery disease, chronic kidney disease, liver cirrhosis and non-AIDS-defining malignancies [18]. The distinction between AIDS, HANA or other classifying conditions was made according to the current classification [23]. An opportunistic infection was considered as a past medical history if it was controlled by > 1 month of effective treatment, whereas a hematological disease required a remission for > 6 months for being considered as past medical history (considered active otherwise). Only CD4 and HIV viral load assays prior to ICU admission and within the last 6 months were considered. A patient was classified as “de novo” HIV if the infection diagnosis was made during the ICU stay or during the prior hospitalization period, otherwise he was classified as “known” HIV. Only “known” HIV patients with a CD4 count > 200/mm3 and a negative viral load within the last 6 months were considered “controlled.” Autonomy was assessed by the Katz scale (ADL) [24]. Disease severity was measured daily by the sepsis-related organ failure assessment (SOFA) [25]. Diagnoses at admission and during the stay were coded using the 10th International Classification of Diseases (ICD). Organ replacement was coded using the Common Classification of Medical Procedures (CCAM).
The ART compliance collected was that reported by the patient in ICU or at the last medical contact when the patient was not able to express himself. Precariousness was based on the few items of the “Agence technique de l'information hospitalière” (ATIH) [26]. Any patient having at least one criterion of precariousness was thus considered precarious.
Finally, DFLST corresponded to a withholding and/or withdrawing treatments aimed at supporting or replacing failing organs (dialysis, vasopressors, mechanical ventilation and cardiopulmonary resuscitation), antibiotics and blood products.
The full study period (1997–2020) was divided into three according to two previously defined dates of interest: 2007, the availability of integrase inhibitors [27]; and 2016, the WHO international recommendation to routinely treat HIV patients regardless of their CD4 count and AIDS-classifying conditions [28, 29].
## Statistical analyses
Characteristics of patients were described as mean (standard deviation), median (interquartile range) or count (percent) for quantitative and qualitative variables, as appropriate. Patient characteristics were compared using the Chi-square test or Fisher’s exact test for categorical variables and the nonparametric Student or Wilcoxon’s rank sum test for continuous variables, as appropriate. The trend tests used to evaluate the periodic evolution were, respectively, a Cochran–Armitage test (or Jonckheere-Terpstra if more than 2 modalities) and an Anova for the categorical and quantitative variables, taking into account the possible center effect.
Factors associated with 60-day mortality were investigated by performing univariate then multiple (with variables yielding p ≤ 0.1 in univariate analyses, and/or those of a priori clinical interest according to the literature, included in the model) Cox proportional hazards regression analyses [expressed as hazard ratios (HRs) and $95\%$ confidence intervals ($95\%$ CIs)]. Missing data for quantitative and categorical variables were imputed when they represented < $30\%$ of total data by median and mode, respectively. These variables were discarded if there was > $30\%$ missing data (example: last CD4 count). Patients lost to follow-up before day 60 because discharged from hospital were considered alive at day 60. Two-by-two interactions between clinically relevant explanatory variables were tested. Models were stratified by center. Proportional hazards assumption was evaluated using Shoenfeld’s residuals [30].
We used SAS 9.4 (NC, USA) and R software. All tests were two-sided, and p values less than 0.05 were considered significant without taking into account the alpha risk inflation related to multiple comparisons due to the exploratory nature of the analysis.
## Results
Of the 24,298 ICU stays in the OutcomeRea™ cohort, 677 ($2.8\%$) involved HIV patients, including 47 readmissions. The study cohort thus comprises 630 first stays (Fig. 1).Fig. 1HIV cohort from OutcomeRea™ flowchart. Abbreviations: D60 (day 60 after ICU admission); HIV (Human Immunodeficiency Virus) The median age of patients was 46.7 years [38; 55] and $69.8\%$ were men. Approximately $7\%$ of patients had each of the four main comorbidities of Knaus (hepatic, cardiovascular, renal, respiratory); $14.4\%$ of the patients had received prior anticancer chemotherapy. The median SOFA at admission was 5 [2; 8]. The main reasons for admission were acute respiratory distress ($35.6\%$), shock ($18.7\%$) and coma ($17.4\%$); an infection was diagnosed in $54.3\%$ of cases, mostly pneumonia ($52.3\%$ of infections) (Table 1).Table 1General baseline data, overall and by period, of the HIV cohort from OutcomeRea™Global($$n = 630$$)1997–2006($$n = 215$$)2007–2015($$n = 336$$)2016–2020($$n = 79$$)p valueAge (year)46.7 [38; 55]41.8 [36; 51]48.1 [39; 55]54.3 [44; 58] < 0.001Body mass index21.8 [19; 25]20.8 [19; 23]22.7 [20; 25]23 [20; 27] < 0.001Katz independence scale6 [6; 6]NA6 [6; 6]6 [6; 6]0.991Sex (male)440 ($69.8\%$)143 ($66.5\%$)242 ($72\%$)55 ($69.6\%$)0.350Diabetes48 ($7.6\%$)3 ($1.4\%$)35 ($10.4\%$)10 ($12.7\%$) < 0.001Obesity34 ($5.9\%$)4 ($2.5\%$)18 ($5.4\%$)12 ($15.2\%$) < 0.001Substance abuse79 ($13.7\%$)30 ($18.6\%$)40 ($12\%$)9 ($11.4\%$)0.464Precariousness ($$n = 55$$, 206, 59)221 ($69.1\%$)40 ($72.7\%$)142 ($68.9\%$)40 ($66.1\%$)0.446Hepatitis B25 ($4\%$)6 ($2.8\%$)18 ($5.4\%$)1 ($1.3\%$)0.901 C66 ($10.5\%$)13 ($6\%$)46 ($13.7\%$)7 ($8.9\%$)0.098Chronic disease (KNAUS) Hepatic47 ($7.5\%$)11 ($5.1\%$)30 ($8.9\%$)6 ($7.6\%$)0.229 Cardiovascular45 ($7.1\%$)8 ($3.7\%$)31 ($9.2\%$)6 ($7.6\%$)0.066 Renal41 ($6.5\%$)7 ($3.3\%$)27 ($8\%$)7 ($8.9\%$)0.027 Respiratory43 ($6.8\%$)4 ($1.9\%$)31 ($9.2\%$)8 ($10.1\%$)0.001Solid neoplasia23 ($3.6\%$)1 ($0.5\%$)15 ($4.5\%$)7 ($8.9\%$) < 0.001Non-AIDS hemopathy8 ($1.5\%$)1 ($0.7\%$)5 ($1.6\%$)2 ($2.7\%$)0.134Pre-admission immunosuppression (excluding HIV/AIDS) Aplasia28 ($4.4\%$)11 ($5.1\%$)14 ($4.2\%$)3 ($3.8\%$)0.560 Corticoid21 ($3.3\%$)5 ($2.3\%$)12 ($3.6\%$)4 ($5.1\%$)0.226 Anticancer chemotherapy91 ($14.4\%$)26 ($12.1\%$)56 ($16.7\%$)9 ($11.4\%$)0.644 SOT6 ($0.9\%$)04 ($1.2\%$)2 ($2.5\%$)0.037 Other16 ($2.5\%$)010 ($3\%$)6 ($7.6\%$) < 0.001Pre-ICU hospitalization stay (day)1 [1; 3]1 [1; 5]1 [1; 3]1 [1; 2]0.324Medical reason for ICU admission591 ($94\%$)199 ($92.6\%$)318 ($94.6\%$)74 ($94.9\%$)0.309SOFA upon ICU admission5 [2; 8]5 [3; 8]5 [2; 8]6 [1; 9]0.668Main purpose for ICU admission*0.626 Multivisceral failure12 ($1.9\%$)6 ($2.8\%$)3 ($0.9\%$)3 ($3.8\%$) Septic shock77 ($12.3\%$)19 ($8.9\%$)48 ($14.4\%$)10 ($12.8\%$) Hemorrhagic shock14 ($2.2\%$)6 ($2.8\%$)6 ($1.8\%$)2 ($2.6\%$) Cardiogenic shock8 ($1.3\%$)4 ($1.9\%$)2 ($0.6\%$)2 ($2.6\%$) Shock (other)18 ($2.9\%$)6 ($2.8\%$)10 ($3\%$)2 ($2.6\%$) Acute respiratory distress223 ($35.6\%$)85 ($39.7\%$)106 ($31.7\%$)32 ($41\%$) COPD decompensation4 ($0.6\%$)04 ($1.2\%$)0 Acute renal failure53 ($8.5\%$)14 ($6.5\%$)34 ($10.2\%$)5 ($6.4\%$) Coma109 ($17.4\%$)40 ($18.7\%$)57 ($17.1\%$)12 ($15.4\%$) Continuous monitoring103 ($16.4\%$)33 ($15.4\%$)60 ($18\%$)10 ($12.8\%$) Scheduled surgery5 ($0.8\%$)1 ($0.5\%$)4 ($1.2\%$)0Syndromic diagnosis on admission* Infection342 ($54.3\%$)131 ($60.9\%$)166 ($49.5\%$)45 ($57\%$)0.133 Bacteremia29 ($4.6\%$)15 ($7\%$)9 ($2.7\%$)5 ($6.3\%$) Pneumonia179 ($28.4\%$)80 ($37.2\%$)74 ($22\%$)25 ($31.6\%$) Meningitis52 ($8.2\%$)27 ($12.6\%$)16 ($4.8\%$)9 ($11.4\%$) Cardiovascular40 ($6.3\%$)14 ($6.5\%$)22 ($6.5\%$)4 ($5.1\%$)0.882 Cardiorespiratory arrest19 ($3\%$)6 ($2.8\%$)9 ($2.7\%$)4 ($5.1\%$) Acute lung edema13 ($2.1\%$)4 ($1.9\%$)9 ($2.7\%$)0 Myocardial infarction3 ($0.5\%$)3 ($1.4\%$)00 Stroke5 ($0.8\%$)1 ($0.5\%$)4 ($1.2\%$)0Bold indicates the significance of the result ($p \leq 0.05$)AIDS acquired immunodeficiency syndrome, COPD chronic obstructive pulmonary disease, HIV human immunodeficiency virus, SOFA sepsis-related organ failure assessment, SOT solid organ transplant*Only one proposition for each patient Among these admissions, 199 ($37.8\%$) were related to an AIDS-classifying condition, and 59 ($11.2\%$) to a HANA disease, while 268 ($51\%$) were not directly related to HIV. Overall, 468 ($74.3\%$) patients had a confirmed AIDS, and 232 ($51.1\%$) were not controlled, despite the administration of ART on admission in 313 ($58.9\%$) cases. The median duration of HIV disease prior to admission was 11 years [3; 17], the median last CD4 count and median last viral load were 242/mm3 [90; 437] and 2 Log [0; 4.6], respectively (Table 2).Table 2HIV-related data, overall and by period, of the HIV cohort from OutcomeRea™Global($$n = 630$$)1997–2006($$n = 215$$)2007–2015($$n = 336$$)2016–2020($$n = 79$$)p valueAIDS468 ($74.3\%$)166 ($76.7\%$)166 ($76.7\%$)53 ($67.1\%$)0.123HIV status ($$n = 454$$) < 0.001 De novo78 ($17.2\%$)33 ($28.4\%$)38 ($14.2\%$)7 ($9.8\%$) Known, uncontrolled232 ($51.1\%$)57 ($49.1\%$)151 ($56.5\%$)24 ($33.8\%$) Known, controlled144 ($31.7\%$)26 ($22.4\%$)78 ($29.2\%$)40 ($56.3\%$)Duration of HIV progression ($$n = 411$$)11 [3; 17]5 [2; 13]12 [5; 17]18 [7; 25] < 0.001Last CD4 count ($$n = 282$$)242 [90; 437]223 [104; 400]228 [81; 400]324 [130; 539]0.014Last HIV viral load ($$n = 260$$)2 [0; 4.6]3.1 [0; 5.1]2.3 [0; 4.6]0 [0; 2]0.004Antiretroviral treatment at admission ($$n = 313$$)313 ($58.9\%$)71 ($47.6\%$)188 ($61\%$)54 ($72\%$) < 0.001 Therapeutic class NRTI298 ($56\%$)70 ($47\%$)175 ($56.8\%$)53 ($70.7\%$) PI185 ($34.8\%$)41 ($27.5\%$)127 ($41.2\%$)17 ($22.7\%$) INI68 ($12.8\%$)039 ($12.7\%$)29 ($38.7\%$) Pre-resuscitation patient attitude Non-compliance ($$n = 370$$)137 ($37\%$)24 ($30.4\%$)89 ($39.4\%$)24 ($36.9\%$) ART for > 6 months ($$n = 364$$)255 ($70\%$)60 ($72.3\%$)148 ($67.9\%$)47 ($74.6\%$)History of AIDS-classifying condition* Infection209 ($39.3\%$)52 ($34.9\%$)135 ($43.8\%$)22 ($29.3\%$)0.896 Pneumocystis49 ($9.2\%$)11 ($7.4\%$)33 ($10.7\%$)5 ($6.7\%$) Tuberculosis86 ($16.2\%$)20 ($13.4\%$)58 ($18.8\%$)8 ($10.7\%$) Toxoplasmosis38 ($7.1\%$)2 ($1.3\%$)31 ($10.1\%$)5 ($6.7\%$) Cytomegalovirus33 ($6.2\%$)7 ($4.7\%$)23 ($7.5\%$)3 ($4\%$) Cryptococcosis5 ($0.9\%$)04 ($1.3\%$)1 ($1.3\%$) Candidiasis46 ($8.6\%$)11 ($7.4\%$)34 ($11\%$)1 ($1.3\%$) Varicella-Zona virus77 ($14.5\%$)18 ($12.1\%$)51 ($16.5\%$)8 ($10.7\%$) Cryptosporidiosis/Microsporidiosis11 ($2.1\%$)5 ($3.3\%$)6 ($1.9\%$)0 Hematologic disease71 ($13.3\%$)17 ($11.4\%$)46 ($14.9\%$)8 ($10.7\%$)0.860 Non-Hodgkin's lymphoma10 ($1.9\%$)2 ($1.3\%$)7 ($2.3\%$)1 ($1.3\%$) T lymphoma1 ($0.2\%$)001 ($1.3\%$) Kaposi55 ($10.3\%$)13 ($8.7\%$)37 ($12\%$)5 ($6.7\%$) Castelman21 ($3.9\%$)8 ($5.4\%$)12 ($3.9\%$)1 ($1.3\%$) Serous lymphoma5 ($0.9\%$)2 ($1.3\%$)3 ($1\%$)0Admission by AIDS diagnosisAIDS-classifying conditions*199 ($37.8\%$)63 ($42.3\%$)107 ($35.7\%$)29 ($38.7\%$)0.372 Opportunistic infections at admission135 ($25.7\%$)54 ($36.2\%$)63 ($20.4\%$)18 ($24\%$)0.007 Pneumocystis49 ($9.3\%$)21 ($14.1\%$)19 ($6.2\%$)9 ($12\%$) Tuberculosis37 ($7\%$)20 ($13.4\%$)12 ($3.9\%$)5 ($6.7\%$) Toxoplasmosis17 ($3.2\%$)7 ($4.7\%$)8 ($2.6\%$)2 ($2.7\%$) Cytomegalovirus40 ($7.6\%$)11 ($7.4\%$)22 ($7.1\%$)7 ($9.3\%$) Cryptococcosis4 ($0.8\%$)2 ($1.3\%$)1 ($0.3\%$)1 ($1.3\%$) Candidiasis31 ($5.8\%$)10 ($6.7\%$)17 ($5.5\%$)4 ($5.3\%$) Varicella-Zona virus10 ($1.9\%$)1 ($0.7\%$)7 ($2.3\%$)2 ($2.7\%$) Cryptopsoridiosis2 ($0.4\%$)02 ($0.6\%$)1 ($1.3\%$) PML3 ($0.6\%$)1 ($0.7\%$)2 ($0.6\%$)0 Other10 ($1.9\%$)3 ($2\%$)4 ($1.3\%$)3 ($4\%$) Active hemopathy on admission146 ($27.7\%$)29 ($19.5\%$)99 ($32.1\%$)18 ($24\%$)0.154 Non-Hodgkin lymphoma76 ($14.3\%$)11 ($7.4\%$)54 ($17.5\%$)11 ($14.7\%$) T lymphoma7 ($1.3\%$)06 ($1.9\%$)1 ($1.3\%$) Kaposi19 ($3.6\%$)5 ($3.3\%$)12 ($3.9\%$)2 ($2.7\%$) Castelman18 ($3.4\%$)4 ($2.7\%$)14 ($4.5\%$)0 Serous lymphoma9 ($1.7\%$)3 ($2\%$)5 ($1.6\%$)1 ($1.3\%$) Hodgkin lymphoma16 ($3\%$)5 ($3.3\%$)10 ($3.2\%$)1 ($1.3\%$) Other9 ($1.7\%$)1 ($0.7\%$)5 ($1.6\%$)3 ($4\%$)HANA-classifying condition59 ($11.2\%$)10 ($6.7\%$)43 ($14.2\%$)6 ($8\%$)0.352Not associated with HIV268 ($51\%$)76 ($51\%$)152 ($49.3\%$)40 ($53.3\%$)0.861Bold indicates the significance of the result ($p \leq 0.05$)AIDS acquired immunodeficiency syndrome, ART antiretroviral therapy, HANA HIV-associated non-AIDS, HIV human immunodeficiency virus, INI integrase inhibitor, PML progressive multifocal leukoencephalopathy*Several possible proposals per patient The ICU stay lasted 5 [3; 11] days in median; $45.6\%$ of the patients were mechanically ventilated during their stay, $29\%$ received vasopressors, and $18.9\%$ required renal replacement therapy. Of note, ART was maintained in two cases out of three ($68.6\%$). Finally, 169 ($26.8\%$) patients died before day 60, including 56 ($8.9\%$) DFLST.
## Impact of periods of ICU stay
Over the three periods, there was a significant decrease in the proportion of HIV patients admitted to intensive care ($3.2\%$ from 1997 to 2006, $2.6\%$ from 2007 to 2015 and $2.3\%$ from 2016 to 2020, $$p \leq 0.001$$ for trend test, adjusted for centers).
As reported in Table 1, the mean age of patients at admission increased over time (44.2 years vs. 51.6 years, $p \leq 0.001$). There was also an increase over time in the prevalence of most comorbidities, such as diabetes, obesity, solid neoplasia or cardiovascular, renal and respiratory diseases. The SOFA prognostic score at admission to ICU remained stable over time, as did the distribution of the main reasons for admission.
While the proportion of AIDS patients on admission to ICU remained stable over time ($$p \leq 0.123$$), the proportion of HIV patients who were controlled on admission almost tripled ($22.4\%$ vs. $56.3\%$), with only $9.8\%$ HIV discovery on admission to ICU in the third period (vs. $28.4\%$ in period 1) (Table 2). Complementarily, there was an increase in the median duration of HIV disease (5 vs. 18 years, $p \leq 0.001$) and ART coverage at admission ($48\%$ vs. $72\%$, $p \leq 0.001$) between periods 1 and 3. This was also associated with an improvement in biological markers of disease control, with an increase in median of the last pre-admission CD4 count (223/mm3 vs. 324/mm3, $$p \leq 0.014$$) and a decrease in the median viral load at admission (3.1 Log vs. 0 Log, $$p \leq 0.004$$). The rate of opportunistic infection at admission decreased over time ($36.2\%$ vs. $24\%$, $$p \leq 0.007$$), while the rate of AIDS-classifying hemopathy increased, although nonsignificantly ($19.5\%$ vs. $24\%$, $$p \leq 0.154$$), and the rates of admission for HANA or non-HIV-related were stable (respectively, $$p \leq 0.352$$ and $$p \leq 0.861$$). Finally, the management of ART evolved over time, with an increase in the rate of ART resumption and initiation between periods 1 and 3 (respectively, $11.8\%$ vs. $40\%$, $$p \leq 0.053$$, and $0\%$ vs. $8.2\%$, $$p \leq 0.032$$).
Regarding organ supplements therapies during ICU stay, the use of mechanical ventilation and renal replacement therapy were stable over time (respectively, from $48.4\%$ to $49.4\%$, $$p \leq 0.707$$, and from $14.9\%$ to $20.2\%$, $$p \leq 0.128$$), while vasopressors were administered significantly more frequently ($14.4\%$ vs. $44.3\%$, $p \leq 0.001$, with comparable initial SOFA, reason for admission and global amines use over time). Moreover, $12.7\%$ of patients received anticancer chemotherapy during their ICU stay in the third period, compared with $1.4\%$ and $8.3\%$, respectively, in periods 1 and 2 ($p \leq 0.001$). Importantly, DFLST rate in ICU was stable over periods ($$p \leq 0.505$$) although differences could be seen according to known/controlled HIV status (decrease for de novo HIV ($12.1\%$ vs. $0\%$ for period 1 and 3, respectively, $$p \leq 0.089$$), increase for known/uncontrolled HIV ($3.5\%$ vs. $16.7\%$ for period 1 and 3, respectively, $$p \leq 0.052$$) and for known/controlled HIV ($11.5\%$ vs. $15\%$ for period 1 and 3, respectively, $$p \leq 0.050$$). Finally, the in-ICU and 60-day mortality rates were also stable over time (respectively, $15.8\%$ to $16.5\%$, $$p \leq 0.992$$, and $22.3\%$ to $19\%$, $$p \leq 0.382$$).
## Risk factors for 60-day mortality on ICU admission
Predictors of 60-day mortality are reported in Table 3. Decedents were older, more likely to be men, and had more chronic liver disease and past history of anticancer chemotherapy. Decedents had a higher SOFA score and were more frequently hospitalized for more than 24 h prior to ICU admission. AIDS status, but not the duration of the disease or the last biological activity markers (CD4 count or viral load), was associated with prognosis. We did not find prognostic influence of ART coverage. Finally, 60-day mortality was higher in patients admitted with an active AIDS-classifying hemopathy or HANA, compared with patients admitted to the ICU without HIV involvement. Table 3Predictors of 60-day after ICU admission mortality in the HIV cohort from OutcomeRea™Alive at D60($$n = 495$$)Dead at D60($$n = 135$$)UnivariateMultivariateHR, CI $95\%$p valueHR, CI $95\%$p valueAge (years) < 38126 ($25.5\%$)29 ($21.5\%$)Ref.0.010Ref.0.029 38 to 54249 ($50.4\%$)57 ($42.2\%$)0.96 [0.61; 1.50]0.87 [0.54; 1.39] > 54119 ($24.1\%$)49 ($36.3\%$)1.68 [1.06; 2.67]1.47 [0.91; 2.36]Katz independence scale6 [6; 6]6 [6; 6]0.95 [0.76; 1.19]0.677Sex (male)333 ($67.4\%$)107 ($79.3\%$)1.70 [1.11; 2.58]0.0131.33 [0.85; 2.07]0.206Diabetes37 ($7.5\%$)11 ($8.1\%$)1.09 [0.59; 2.02]0.786Obesity29 ($5.9\%$)5 ($3.7\%$)0.67 [0.27; 1.63]0.375Chronic disease (KNAUS) Hepatic32 ($6.5\%$)15 ($11.1\%$)1.66 [0.97; 2.85]0.0662.07 [1.15; 3.73]0.015 Cardiovascular35 ($7.3\%$)10 ($7.4\%$)1.07 [0.56; 2.03]0.846 Renal29 ($5.9\%$)12 ($8.9\%$)1.59 [0.87; 2.88]0.128 Respiratory36 ($7.3\%$)7 ($5.2\%$)0.77 [0.36; 1.66]0.507Solid neoplasia17 ($3.4\%$)6 ($4.4\%$)1.23 [0.54; 2.80]0.619History of chemotherapy52 ($10.5\%$)39 ($28.9\%$)3.09 [2.04; 4.68] < 0.0012.48 [1.54; 4.00] < 0.001Inclusion period 1(1997–2006)167 ($33.8\%$)48 ($35.6\%$)Ref.0.929Ref.0.578 2(2007–2015)263 ($53.2\%$)72 ($53.3\%$)0.93 [0.63; 1.38]0.81 [0.54; 1.22] 3(2016–2020)64 ($13\%$)15 ($11.1\%$)1.00 [0.55; 1.82]0.82 [0.44; 1.53]Medical reason for ICU admission467 ($94.7\%$)123 ($91.1\%$)0.61 [0.33; 1.12]0.110Main symptom on admission Shock96 ($19.4\%$)33 ($24.4\%$)1.28 [0.79; 2.07] Acute respiratory distress184 ($37.2\%$)42 ($31.2\%$)0.93 [0.59; 1.47] Coma82 ($16.6\%$)27 ($20\%$)1.31 [0.78; 2.18] Other133 ($26.7\%$)33 ($24.4\%$)Ref.0.392SOFA upon ICU admission > 4262 ($53\%$)104 ($77\%$)2.60 [1.73; 3.88] < 0.0012.35 [1.56; 3.56] < 0.001Pre-ICU hospitalization stay > 24 h172 ($34.8\%$)72 ($53\%$)1.92 [1.36; 2.70] < 0.0011.47 [1.03; 2.11]0.033AIDS355 ($71.9\%$)112 ($83\%$)1.77 [1.12; 2.80]0.0141.79 [1.11; 2.89]0.017HIV status ($$n = 528$$) De novo63 ($15.1\%$)15 ($13.6\%$)Ref.0.613 Known, uncontrolled113 ($27\%$)26 ($23.7\%$)1.20 [0.68; 2.12] Known, controlled242 ($57.9\%$)69 ($62.7\%$)0.98 [0.52; 1.86]Duration of HIV progression > 10 years ($$n = 409$$)166 ($51.7\%$)40 ($45.4\%$)0.82 [0.54; 1.26]0.375Last CD4 count > 250/mm3 ($$n = 282$$)109 ($22.1\%$)35 ($25.9\%$)1.58 [0.92; 2.70]0.096 < 503 ($1.5\%$)3 ($6.1\%$)1.63 [0.80; 3.35] 50 to 20060 ($30.8\%$)20 ($40.8\%$)1.59 [0.88; 2.85] > 200132 ($67.7\%$)26 ($53.1\%$)Ref. Last HIV viral load > 2 Log ($$n = 260$$)107 ($50.7\%$)22 ($44.9\%$)0.78 [0.44; 1.38]0.394ART at ICU admission ($$n = 531$$)244 ($58\%$)69 ($62.7\%$)1.21 [0.82; 1.79]0.329History of AIDS-classifying condition Infection155 ($36.8\%$)53 ($48.2\%$)1.47 [1.00; 2.17]0.048 Hematologic disease54 ($12.8\%$)20 ($18.2\%$)1.18 [0.69; 2.01]0.541Diagnosis admission according to HIV/AIDS AIDS-classifying condition145 ($34.4\%$)53 ($48.2\%$) Active opportunistic infections105 ($24.9\%$)29 ($26.4\%$)0.98 [0.65; 1.49]1.39 [0.81; 2.39] Active hemopathy96 ($22.8\%$)50 ($45.5\%$)2.59 [1.68; 3.98]1.52 [0.94; 2.46] HANA43 ($10.2\%$)16 ($14.5\%$)1.90 [1.06; 3.39]1.49 [0.84; 2.64] Not associated with HIV233 ($55.3\%$)41 ($37.3\%$)Ref.0.010Ref.0.203ART management in ICU ($$n = 316$$) Suspension73 ($28.1\%$)21 ($38.2\%$)0.98 [0.58; 1.66]0.942 Resume20 ($7.7\%$)0–– Continued167 ($64.5\%$)33 ($57.9\%$)0.51 [0.32; 0.83]0.006 Introduction22 ($8.6\%$)4 ($7.3\%$)0.70 [0.21; 2.27]0.550Mechanical ventilation during the ICU stay179 ($36.2\%$)108 ($80\%$)5.98 [3.91; 9.14] < 0.001Vasopressor during the ICU stay113 ($22.9\%$)70 ($51.8\%$)3.08 [2.19; 4.34] < 0.001RRT during the ICU stay71 ($14.4\%$)48 ($35.6\%$)2.84 [1.98; 4.06] < 0.001Use of anticancer chemotherapy during the ICU stay25 ($5.1\%$)16 ($11.8\%$)2.22 [1.27; 3.91]0.005Bold indicates the significance of the result ($p \leq 0.05$)AIDS acquired immunodeficiency syndrome, ART antiretroviral therapy, HANA HIV-associated non-AIDS, HIV human immunodeficiency virus, ICU intensive care unit, RRT renal replacement therapy, SOFA sepsis-related organ failure assessment By multivariate analysis, age > 54 years (HR 1.47 [0.91; 2.36]), chronic liver disease (HR 2.07 [1.15; 3.73], $$p \leq 0.015$$), history of anticancer chemotherapy (HR 2.48 [1.54; 4.0], $p \leq 0.001$), SOFA score > 4 (HR 2.35 [1.56; 3.56], $p \leq 0.001$), pre-ICU hospitalization duration of stay > 24 h (HR 1.47 [1.03; 2.11], $$p \leq 0.033$$) and AIDS status (HR 1.79 [1.11; 2.89], $$p \leq 0.017$$) were associated with 60-day mortality. There was a nonsignificant trend toward an increased risk of 60-day mortality for patients admitted for an AIDS-classifying opportunistic infection (HR 1.39 [0.81; 2.39]) or active hemopathy (HR 1.52 [0.94; 2.46]) or for HANA (HR 1.49 [0.84; 2.64]), compared with patients admitted to ICU with no HIV involvement. Of note, the period of care was not associated with the risk of 60-day mortality in univariate ($$p \leq 0.929$$) and multivariate ($$p \leq 0.578$$) analyses (Fig. 2).Fig. 2Kaplan–Meier survival curves of HIV patients from the OutcomeRea™ cohort according to the period of care
## Discussion
This cohort study confirms and updates epidemiological data on ICU patients with HIV, i.e., an increase in the burden of comorbidities of HIV patients as well as an improvement in the control of the viral infection, and a stability over time of the risk factors for short-term death (more or less directly associated with HIV).
Consistent with the progressive decrease in the level of hospitalization of HIV patients over time [7, 8], the rate of admission to ICU for HIV patients decreased over the study period. The latter is potentially explained by the improvement in HIV control over time, as illustrated by the increase in CD4 count, the decrease in HIV viral load and the increase in ART coverage (up to $70\%$) at admission, and, as previously reported by Barbier and coll in 2014, the decrease in admissions related to opportunistic infections [6, 10, 18]. Of note, the stability over time and at a high rate (three times higher than described) of the AIDS-classifying hemopathies prevalence is probably related to the center effect induced by a major hematological center (corresponding to $46.3\%$ of the inclusions of the cohort).
The aging of HIV patients, a corollary of their care improvement, is associated with a greater clinical impact of any intoxication, co-infection or of HIV itself (chronic low-level viremia) [2–6]. Consequently, the expected survival benefit of an improved HIV control over time is probably partially offset by the increased burden of comorbidity (mainly respiratory, renal, metabolic diseases and non-AIDS neoplasia) of these patients. Importantly, in accordance with the data in the literature [31, 32] and remaining stable over time, more than half of the patients were admitted to ICU for reason not or indirectly linked to HIV. Moreover, the distribution of the main reasons for admission (in proportion and hierarchy) remains stable over time and overlaps with that of non-HIV patients, as described [31, 32].
Likely indicative of improved specific management of HIV patients in the ICU, the rate of ART resumption and introduction in the ICU increased over time. Although data as to the morbidity-mortality benefit of early HIV treatment is mostly demonstrated in non-critically ill patients [33], there are data to support the same benefit in ICU on short- and long-term prognosis, as outlined in the meta-analysis that Andrade and coll published in 2017 [34]. Meanwhile, the use of organ replacement was stable over time [vasopressor support probably artificially increased because of greater inotrope use during the first period ($18.6\%$ vs. $5.1\%$ in period 3, $p \leq 0.001$, data not shown) [17, 32]. Ultimately, reflecting the paradigm shift in HIV patient care, DFLST rate in ICU has inversely evolved over time, depending on whether HIV was discovered or known (decrease for de novo HIV, increase for known HIV). Knowledge on this subject remains scarce and future studies in view of the phenotypic evolution of the HIV population are warranted [35].
Previous publications on the subject have focused on short-term (ICU/hospital) mortality. After a significant decrease in mortality at the end of the 1990’s, the latest studies reported a stagnation of mortality in intensive care, with values ranging from 16 to $37\%$ according to the region of care [10, 17, 18, 36]. In line with these works, the 60-day ICU mortality rate of the OutcomeRea™ cohort is stable over time. In multivariate analysis, the main risk factors for mortality already described in the past were identified, namely age, history of liver disease or anticancer chemotherapy, AIDS status, severity at admission (estimated by the SOFA score) and duration of the hospital stay before ICU admission. Although there are tendencies for a poorest prognosis of patients with HIV-related reason for hospitalization (mainly active hemopathy), this parameter was no longer associated with prognosis in the multivariate model. When adjusted on all prognostic covariates, the period did not influence the 60-day mortality risk. The persistence over more than 20 years of modifiable risk factors or detectable risk factors invites us to optimize the overall management of HIV patients. Early identification of vulnerable patients would allow an early adaptation of the intensity of care.
The main strengths of this study are its prospective collection and broad and national inclusion period, which provide an accurate evolutionary perspective of the phenotype of HIV patients. In addition, it reassesses and confirms the risk factors for short-term mortality in HIV patients, some of which are avoidable or detectable, and reminds us of the margins for improvement in the management of this population.
The accuracy and completeness of the collection of comorbidities is one of this study limitations. Indeed, neuropsychiatric disorders were not recorded, and the Knaus classification is not very sensitive for the burden of comorbidities. Secondly, the HIV-related biological data could not be fully explored because of a significant lack of collected data (around $50\%$). Then, the small number of patients in the last period prohibited some subgroup analyses, due to lack of events or power. Finally, the 60-day timepoint for the mortality assessment might have been too early for a reliable picture of the overall risk of death related to ICU admission in these patients.
In conclusion, the phenotype of HIV patients admitted to intensive care is still evolving over time, with an improved control of HIV but an increase in the overall burden of comorbidity. Nevertheless, the medium-term prognosis remains stable over time. Several questions remain unanswered; the long-term post-resuscitation outcome, particularly in terms of quality of life, and the management of antiretroviral drugs require further explorations.
## References
1. 1.World Health Organization. Global HIV and AIDS statistics, 2019. http://www.unaids.org. Accessed 1 Nov 2019
2. Deeks SG, Lewin SR, Havlir DV. **The end of AIDS: HIV infection as a chronic disease**. *Lancet* (2013.0) **382** 1525-1533. DOI: 10.1016/S0140-6736(13)61809-7
3. Freiberg CCC, Kuller LH. **HIV Infection and the risk of acute myocardial infarction**. *JAMA Intern Med* (2013.0) **173** 4
4. Crothers K, Huang L, Goulet JL. **HIV infection and risk for incident pulmonary diseases in the combination antiretroviral therapy era**. *Am J Respir Crit Care Med* (2011.0) **183** 388-395. DOI: 10.1164/rccm.201006-0836OC
5. Lanoy E, Spano JP, Bonnet F. **The spectrum of malignancies in HIV-infected patients in 2006 in France: the ONCOVIH study**. *Int J Cancer* (2011.0) **129** 467-475. DOI: 10.1002/ijc.25903
6. Akgun KM, Gordon K, Pisani M. **Risk factors for hospitalization and medical intensive care unit (MICU) admission among HIV-infected veterans**. *J Acquir Immune Defic Syndr* (2013.0) **62** 52-59. DOI: 10.1097/QAI.0b013e318278f3fa
7. Paul S, Gilbert HM, Ziecheck W, Jacobs J, Sepkowitz KA. **The impact of potent antiretroviral therapy on the characteristics of hospitalized patients with HIV infection**. *AIDS* (1999.0) **13** 415-418. DOI: 10.1097/00002030-199902250-00015
8. Crum-Cianflone NF, Grandits G, Echols S. **Trends and causes of hospitalizations among HIV-infected persons during the late HAART era: what is the impact of CD4 counts and HAART use?**. *J Acquir Immune Defic Syndr* (2010.0) **54** 248-257. DOI: 10.1097/QAI.0b013e3181c8ef22
9. Yehia BR, Fleishman JA, Hicks PL, Ridore M, Moore RD, Gebo KA. **Inpatient health services utilization among HIV-infected adult patients in care 2002–2007**. *J Acquir Immune Defic Syndr* (2010.0) **53** 397-404. DOI: 10.1097/QAI.0b013e3181bcdc16
10. Akgun KM, Tate JP, Pisani M. **Medical ICU admission diagnoses and outcomes in human immunodeficiency virus-infected and virus-uninfected veterans in the combination antiretroviral era**. *Crit Care Med* (2013.0) **41** 1458-1467. DOI: 10.1097/CCM.0b013e31827caa46
11. Kwizera A, Nabukenya M, Peter A, Semogerere L, Ayebale E, Katabira C. **Clinical characteristics and short-term outcomes of HIV patients admitted to an african intensive care unit**. *Crit Care Res Pract* (2016.0) **2016** 2610873. PMID: 27800179
12. Botha J, Fabian J, Etheredge H, Conradie F, Tiemessen CT. **HIV and solid organ transplantation: where are we now**. *Curr HIV AIDS Rep* (2019.0) **16** 404-413. DOI: 10.1007/s11904-019-00460-7
13. Madan S, Patel SR, Saeed O, Sims DB, Shin JJ, Goldstein DJ. **Outcomes of heart transplantation in patients with human immunodefciency virus**. *Am J Transplant* (2019.0). DOI: 10.1111/ajt.15257
14. Dickson SJ, Batson S, Copas AJ, Edwards SG, Singer M, Miller RF. **Survival of HIV-infected patients in the intensive care unit in the era of highly active antiretroviral therapy**. *Thorax* (2007.0) **62** 964-968. DOI: 10.1136/thx.2006.072256
15. Turtle L, Vyakernam R, Menon-Johansson A. **Intensive care usage by HIV-positive patients in the HAART era**. *Interdiscip Perspect Infect Dis* (2011.0) **2011** 847835. DOI: 10.1155/2011/847835
16. van Lelyveld SF, Wind CM, Mudrikova T. **Short- and long-term outcome of HIV-infected patients admitted to the intensive care unit**. *Eur J Clin Microbiol Infect Dis* (2011.0) **30** 1085-1093. DOI: 10.1007/s10096-011-1196-z
17. Morquin D, Le Moing V, Mura T. **Short- and long-term outcomes of HIV-infected patients admitted to the intensive care unit: impact of antiretroviral therapy and immunovirological status**. *Ann Intensive Care* (2012.0) **2** 25. DOI: 10.1186/2110-5820-2-25
18. Barbier F, Roux A, Canet E. **Temporal trends in critical events complicating HIV infection: 1999–2010 multicentre cohort study in France**. *Intensive Care Med* (2014.0) **40** 1906-1915. DOI: 10.1007/s00134-014-3481-7
19. Adlakha A, Pavlou M, Walker DA. **Survival of HIV-infected patients admitted to the intensive care unit in the era of highly active antiretroviral therapy**. *Int J STD AIDS* (2011.0) **22** 498-504. DOI: 10.1258/ijsa.2011.010496
20. von Elm E, Altman DG, Egger M, Pocock SJ, Gotzsche PC, Vandenbroucke JP. **The strengthening the reporting of observational studies in epidemiology (STROBE) statement: guidelines for reporting observational studies**. *J Clin Epidemiol* (2008.0) **61** 344-349. DOI: 10.1016/j.jclinepi.2007.11.008
21. Zahar JR, Timsit JF, Garrouste-Orgeas M, Francais A, Vesin A, DescorpsDeclere A, Dubois Y, Souweine B, Haouache H, Goldgran-Toledano D, Allaouchiche B, Azoulay E, Adrie C. **Outcomes in severe sepsis and patients with septic shock: pathogen species and infection sites are not associated with mortality**. *Crit Care Med* (2011.0) **39** 1886-1895. DOI: 10.1097/CCM.0b013e31821b827c
22. Knaus WA, Wagner DP, Draper EA, Zimmerman JE, Bergner M, Bastos PG. **The APACHE III prognostic system. Risk prediction of hospital mortality for critically ill hospitalized adults**. *Chest* (1991.0) **100** 1619-1636. DOI: 10.1378/chest.100.6.1619
23. **Revised classification system for HIV infection and expanded surveillance case definition for AIDS among adolescents and adults**. *MMWR Morb Mortal Wkly Rep* (1993.0) **41** 1-19
24. Katz S, Downs TD, Cash HR, Grotz RC. **Progress in development of the index of ADL**. *Gerontologist Spring* (1970.0) **10** 20-30. DOI: 10.1093/geront/10.1_part_1.20
25. Vincent JL, Moreno R, Takala J, Willatts S, De Mendonça A, Bruining H. **The SOFA (sepsis-related organ failure assessment) score to describe organ dysfunction/failure. On behalf of the working group on sepsis-related problems of the European society of intensive care medicine**. *Intensive Care Med* (1996.0) **22** 707-710. DOI: 10.1007/BF01709751
26. Bréchat PH, Lansac C, Hasni S. **Étude de l’ATIH : prise en charge de la précarité dans les établissements de santé**. *Gest Hosp* (2009.0) **486** 269-274
27. Grinsztejn B, Nguyen B, Katlama C, Gatell J, Lazzarin A, Vittecoq D. **Safety and efficacy of the HIV-1 integrase inhibitor raltegravir (MK-0518) in treatment-experienced patients with multidrug-resistant virus: a phase II randomised controlled trial**. *Lancet* (2007.0) **369** 1261-1269. DOI: 10.1016/S0140-6736(07)60597-2
28. 28.http://apps.who.int/iris/bitstream/handle/10665/250576/WHO-HIV-2016.05-fre.pdf;jsessionid=6F1692E720C7ACA966C5C306F4D5D308?sequence=1
29. Lundgren J, Babiker A, Gordin F, Emery S, Grund B, Sharma S. **INSIGHT START study group. Initiation of antiretroviral therapy in early asymptomatic HIV infection**. *N Engl J Med* (2015.0) **373** 795-807. DOI: 10.1056/NEJMoa1506816
30. Grambsch PM, Therneau TM. **Proportional hazards tests and diagnostics based on weighted residuals**. *Biometrika* (1994.0) **81** 515-526. DOI: 10.1093/biomet/81.3.515
31. Barbier F, Mer M, Szychowiak P, Miller R, Mariotte E, Galicier L. **Management of HIV-infected patients in the intensive care unit**. *Intensive Care Med* (2020.0) **46** 329-342. DOI: 10.1007/s00134-020-05945-3
32. Coquet I, Pavie J, Palmer P, Barbier F, Legriel S, Mayaux J. **Survival trends in critically ill HIV-infected patients in the highly active antiretroviral therapy era**. *Crit Care* (2010.0) **14** R107. DOI: 10.1186/cc9056
33. Zolopa A, Andersen J, Powderly W. **Early antiretroviral therapy reduces AIDS progression/death in individuals with acute opportunistic infections: a multicenter randomized strategy trial**. *PLoS ONE* (2009.0) **4** e5575. DOI: 10.1371/journal.pone.0005575
34. Andrade HB, Shinotsuka CR, da Silva IRF, Donini CS, Yeh Li H, de Carvalho FB. **Highly active antiretroviral therapy for critically ill HIV patients: a systematic review and meta-analysis**. *PLoS ONE* (2017.0) **12** e0186968. DOI: 10.1371/journal.pone.0186968
35. Harding R. **Palliative care as an essential component of the HIV care continuum**. *Lancet HIV* (2018.0) **5** e524-e530. DOI: 10.1016/S2352-3018(18)30110-3
36. Morris A, Creasman J, Turner J, Luce JM, Wachter RM, Huang L. **Intensive care of human immunodeficiency virus-infected patients during the era of highly active antiretroviral therapy**. *Am J Respir Crit Care Med* (2002.0) **166** 262-267. DOI: 10.1164/rccm.2111025
|
---
title: Mechanical force induces macrophage-derived exosomal UCHL3 promoting bone marrow
mesenchymal stem cell osteogenesis by targeting SMAD1
authors:
- Panjun Pu
- Shengnan Wu
- Kejia Zhang
- Hao Xu
- Jiani Guan
- Zhichun Jin
- Wen Sun
- Hanwen Zhang
- Bin Yan
journal: Journal of Nanobiotechnology
year: 2023
pmcid: PMC10012474
doi: 10.1186/s12951-023-01836-z
license: CC BY 4.0
---
# Mechanical force induces macrophage-derived exosomal UCHL3 promoting bone marrow mesenchymal stem cell osteogenesis by targeting SMAD1
## Abstract
### Background
Orthodontic tooth movement (OTM), a process of alveolar bone remodelling, is induced by mechanical force and regulated by local inflammation. Bone marrow-derived mesenchymal stem cells (BMSCs) play a fundamental role in osteogenesis during OTM. Macrophages are mechanosensitive cells that can regulate local inflammatory microenvironment and promote BMSCs osteogenesis by secreting diverse mediators. However, whether and how mechanical force regulates osteogenesis during OTM via macrophage-derived exosomes remains elusive.
### Results
Mechanical stimulation (MS) promoted bone marrow-derived macrophage (BMDM)-mediated BMSCs osteogenesis. Importantly, when exosomes from mechanically stimulated BMDMs (MS-BMDM-EXOs) were blocked, the pro-osteogenic effect was suppressed. Additionally, compared with exosomes derived from BMDMs (BMDM-EXOs), MS-BMDM-EXOs exhibited a stronger ability to enhance BMSCs osteogenesis. At in vivo, mechanical force-induced alveolar bone formation was impaired during OTM when exosomes were blocked, and MS-BMDM-EXOs were more effective in promoting alveolar bone formation than BMDM-EXOs. Further proteomic analysis revealed that ubiquitin carboxyl-terminal hydrolase isozyme L3 (UCHL3) was enriched in MS-BMDM-EXOs compared with BMDM-EXOs. We went on to show that BMSCs osteogenesis and mechanical force-induced bone formation were impaired when UCHL3 was inhibited. Furthermore, mothers against decapentaplegic homologue 1 (SMAD1) was identified as the target protein of UCHL3. At the mechanistic level, we showed that SMAD1 interacted with UCHL3 in BMSCs and was downregulated when UCHL3 was suppressed. Consistently, overexpression of SMAD1 rescued the adverse effect of inhibiting UCHL3 on BMSCs osteogenesis.
### Conclusions
This study suggests that mechanical force-induced macrophage-derived exosomal UCHL3 promotes BMSCs osteogenesis by targeting SMAD1, thereby promoting alveolar bone formation during OTM.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12951-023-01836-z.
## Background
Malocclusion refers to irregular teeth or abnormal occlusion, with a high prevalence ranging from 20 to $100\%$ reported by different researchers [1–3]. Orthodontic treatment is an effective way to correct malocclusion. It is based on the principle that tooth movement occurs when the tooth and its periodontal tissues, such as alveolar bone, are subjected to long-term mechanical force. In essence, orthodontic tooth movement (OTM) is a process of alveolar bone remodelling induced by mechanical force and regulated by local aseptic inflammation, whose underlying mechanism is critical to dissect [4, 5].
In the force-induced alveolar bone remodelling microenvironment, multiple types of cells sense mechanical stimulation and regulate bone remodelling, including macrophages, bone marrow-derived mesenchymal stem cells (BMSCs), periodontal ligament stem cells (PDLSCs), osteoblasts, osteoclasts and osteocytes [6–10]. Macrophages, as mechanosensitive cells [11–13], play a vital role in OTM by secreting cytokines and modulating local inflammation [5, 14, 15]. Previous studies have shown that macrophages are modulated by mechanical force to regulate tooth movement, whereas the mechanism by which macrophages regulate alveolar bone remodelling, especially alveolar bone formation during OTM, is still not clear.
It has been demonstrated that BMSCs, progenitors of osteoblasts, can directly respond to mechanical forces and promote alveolar bone formation during OTM [6]. Macrophages are important modulators of BMSCs by secreting a variety of inflammatory cytokines [16, 17]. These cytokines can regulate the behaviors of BMSCs, including homing, proliferation, and osteogenic differentiation, and further promote bone regeneration [18–20]. Therefore, macrophages modulating BMSCs osteogenesis under mechanical force may be an important component of mechanical force-induced alveolar bone formation during OTM.
Recently, cumulative evidence suggests that exosomes mediate the communication between macrophages and BMSCs in the bone remodelling microenvironment [21, 22]. The bilayer membrane structure of exosomes can protect substances such as nucleic acids, proteins and lipids within it [23]. In addition, exosomes can be directly endocytosed by BMSCs, instead of being inhibited by certain feedback mechanisms like cytokines [24]. However, whether and how exosomes derived from macrophages under mechanical force regulate BMSCs osteogenesis as well as alveolar bone formation remains unknown. Lack of such knowledge is an important problem, since, without it, acquiring the ability to modulate key signaling processes pharmacologically in OTM and malocclusion treatment is highly unlikely.
In this study, we show that mechanical force modulates bone marrow-derived macrophage (BMDM)-derived exosomal ubiquitin carboxyl-terminal hydrolase isozyme L3 (UCHL3), which promotes BMSCs osteogenesis by targeting mothers against decapentaplegic homologue 1 (SMAD1), thereby promoting alveolar bone formation during OTM.
## Mechanical force promotes BMDM-mediated BMSCs osteogenesis
Previous studies have shown that mechanical force could promote MSCs osteogenesis through the modulation of macrophages [25, 26]. To mimic BMSCs osteogenesis during OTM in vitro, we generated a mechanical force response culture system in which BMDMs were stimulated with different mechanical stimulations (MS), and their supernatants were then collected to treat BMSCs (Fig. 1a; Additional file 1: Fig. S1). We first measured the viability of BMDMs under the influences of different strain levels (0, 5, 10, and $15\%$) or durations (0, 2, 6, and 10 h) by CCK-8 analysis. After BMDMs were treated with different strain levels (0, 5, 10, and $15\%$) for 2 h, their cell viability was significantly increased under the $10\%$ strain compared with other groups (Additional file 1: Fig. S2a). No significant differences were observed in BMDMs viability when they were treated with $10\%$ mechanical force for 2, 6, and 10 h, respectively (Additional file 1: Fig. S2b). Therefore, $10\%$ strain was used as the standard mechanical force for further experiments. Fig. 1Mechanical force promotes BMDM-mediated BMSCs osteogenesis. a Schematic illustration. b-f qRT‒PCR analysis of the mRNA expression of osteogenic genes and Western blotting of RUNX2 protein levels in BMSCs after treatment with conditioned medium from MS-BMDMs ($$n = 3$$). g Relative intensity analysis from f ($$n = 3$$). h, i ALP staining and Alizarin red staining of BMSCs after treatment with conditioned medium from MS-BMDMs. Scale bar: 100 μm. j Representative TEM images of MS-BMDM-EXOs. Scale bar: 200 nm. k Particle size distribution of MS-BMDM-EXOs from NTA. l Western blotting of the exosomal markers CD63, and TSG101, and the cell marker Calnexin. m Concentration of BMDM-EXOs and MS-BMDM-EXOs from NTA ($$n = 3$$). Data are shown as the mean ± SD. Two-tailed unpaired Student’s t test or one-way ANOVA followed by Tukey’s post hoc multiple comparisons were performed. * $P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$ Next, we collected the supernatants from BMDMs pretreated with $10\%$ strain for 0, 2, 6, and 10 h and prepared conditioned medium to culture BMSCs (Fig. 1a). qRT‒PCR showed that MS ($10\%$, 6 h)-BMDMs-derived conditioned medium significantly increased osteogenic gene expression (Alp, Runx2 Col1 and Ocn) in BMSCs compared with that in other groups (Fig. 1b–e). Western blotting further confirmed that RUNX2, a central regulator of osteogenesis [27], was significantly upregulated in BMSCs after treatment with conditioned medium derived from MS ($10\%$, 6 h)-BMDMs (Fig. 1f, g). Consistently, ALP staining also showed an increased ALP level in BMSCs treated with conditioned medium from MS ($10\%$, 6 h)-BMDMs (Fig. 1h). Alizarin red staining revealed enhanced extracellular matrix mineralization deposition in the MS ($10\%$, 10 h)-BMDM-treated BMSCs (Fig. 1i). In summary, these results confirmed that modest mechanical force promoted BMDM-mediated BMSC osteogenesis, and thus, MS ($10\%$, 6 h) was used for the following experiment.
We set out to explore the player that may mediate BMSCs osteogenesis in the supernatants of MS-BMDMs. As a key intercellular mediator, exosomes derived from macrophages have been demonstrated to play an important role in modulating BMSCs osteogenesis [22, 28]. Therefore, we focused on the exosomes and collected the supernatants of the BMDMs to isolate exosomes after they were treated with MS ($10\%$, 6 h). Transmission electron microscopy (TEM) revealed that these purified vesicles have cup- or sphere-shaped morphology (Fig. 1j), similar to the exosomes described previously [29]. Nanoparticle tracking analysis (NTA) demonstrated that the diameter of these particles predominantly ranged from 30 to 200 nm (Fig. 1k), which was consistent with the previously reported size distribution of exosomes [30]. Western blotting further verified that exosomal surface markers such as CD63 and TSG101 were present on these particles, while calnexin, a cytoplasmic protein, was not detected (Fig. 1l). NTA also showed that the supernatants of MS-BMDMs had a higher number of exosomes than BMDMs (Fig. 1m). All these data showed that these nanoparticles identified here were exosomes, and mechanical force promoted the secretion of exosomes derived from BMDMs. Based on these findings, we speculated that exosomes may be the key mediators secreted by MS-BMDMs to influence BMSCs osteogenesis.
## Exosomes derived from MS-BMDMs promote BMSCs osteogenesis in vitro
To confirm the potential effects of exosomes derived from MS-BMDMs in mediating BMSCs osteogenesis, we first labelled MS-BMDM-EXOs with the fluorescent dye Dil and treated BMSCs with these exosomes for 12 h. Fluorescence microscope imaging showed that these exosomes surrounded BMSCs nuclei, suggesting that they had been taken up by BMSCs (Fig. 2a). Next, BMSCs were treated with PBS, BMDM-EXOs or MS-BMDM-EXOs. qRT‒PCR and Western blotting showed that MS-BMDM-EXOs significantly promoted the mRNA expression of osteogenic genes (Alp, Runx2, Col1 and Ocn) and RUNX2 protein levels compared with PBS or BMDM-EXOs (Fig. 2b–g). ALP staining and Alizarin red staining also revealed that ALP level and extracellular matrix mineralization deposition were enhanced in BMSCs after treatment with MS-BMDM-EXOs compared with PBS or BMDM-EXOs (Fig. 2h, i). These results suggested that exosomes derived from MS-BMDMs promote osteogenesis in BMSCs. Fig. 2Exosomes derived from MS-BMDMs promote BMSCs osteogenesis in vitro. a Immunofluorescence images of the internalization of Dil-labelled MS-BMDM-EXOs in BMSCs after treatment for 12 h. MS-BMDM-EXOs and nuclei were stained with Dil (red) and DAPI (blue), respectively. Scale bar: 50 μm. b-f qRT‒PCR analysis of the mRNA expression of osteogenic genes and Western blotting of RUNX2 protein levels in BMSCs after treatment with PBS, BMDM-EXOs and MS-BMDM-EXOs ($$n = 3$$). g Relative intensity analysis of f ($$n = 3$$). h, i ALP staining and Alizarin red staining in BMSCs after treatment with PBS, BMDM-EXOs and MS-BMDM-EXOs. Scale bar: 100 µm. j Schematic illustration. k, l qRT‒PCR analysis of the mRNA expression of osteogenic genes and Western blotting of RUNX2 protein level in BMSCs after culture with conditioned medium from MS-BMDMs treated with GW4869 or DMSO as a control ($$n = 3$$). m Relative intensity analysis from l. n, o ALP staining and Alizarin red staining of BMSCs after culture with conditioned medium from MS-BMDMs treated with GW4869 or DMSO as a control. Scale bar: 100 μm. Data are shown as the mean ± SD. Two-tailed unpaired Student’s t test or one-way ANOVA followed by Tukey’s post hoc multiple comparisons were performed. * $P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$ Furthermore, GW4869, a widely used exosome secretion inhibitor [31, 32], was added to the culture medium of MS-BMDMs, and DMSO was used as a control (Fig. 2j). In BMSCs treated with conditioned medium from MS-BMDMs, GW4869 significantly decreased osteogenic gene (Alp, Runx2, Col1 and Ocn) and RUNX2 protein levels (Fig. 2k–m). Additionally, ALP staining and Alizarin red staining showed that ALP level and extracellular matrix mineralization deposition were also impaired when exosomes were blocked in MS-BMDMs (Fig. 2n, o). These results suggested that the key positive modulation of exosomes in MS-BMDMs regulated BMSCs osteogenesis.
## Exosomes derived from MS-BMDMs promote alveolar bone formation during OTM
To explore the role of exosomes derived from MS-BMDMs in alveolar bone formation during OTM, we established an OTM model in 2-month-old mice as previously described along with intraperitoneal injection of GW4869 every 2 days [33] (Additional file 1: Fig. S3). After force was applied for 14 days, tooth movement was significantly suppressed by administration of GW4869 (Fig. 3a, b). Micro-CT analysis showed that the BV/TV of alveolar bone in the interradicular region of the loaded tooth was significantly decreased after GW4869 injection compared with the OTM group (Fig. 3c, d). These results were consistent with the HE staining results (Fig. 3e, f). In addition, ALP staining showed that the ALP-positive surface in the interradicular region and tension side of loaded alveolar bone was increased in the OTM group compared with the Control group, and it was suppressed after GW4869 injection (Fig. 3g, h; Additional file 1: Fig. S4), thereby indicating that exosome inhibition could impair alveolar bone formation during OTM. These findings suggest that exosomes are essential for mechanical force-induced alveolar bone formation during OTM.Fig. 3Exosomes derived from MS-BMDMs promote alveolar bone formation during OTM. a Representative 3D Micro-CT reconstruction images of the murine OTM model at day 14. Arrow: direction of tooth movement. b Quantification of tooth movement distance at day 14 after OTM treatment ($$n = 4$$). c Representative 3D scanned sections of the first molar at day 14 after OTM treatment. d Quantification from c. BV/TV of the interradicular region of the first molar ($$n = 6$$). e Representative HE staining images of the first molar at day 14 after OTM treatment. Scale bar: 200 μm. f Quantification from e. BV/TV of the interradicular region of the first molar ($$n = 6$$). g Representative ALP staining images of the interradicular region of the first molar at day 14 after OTM treatment. D: dentin. Scale bar: 200 μm. h Quantification from g. ALP-positive surface relative to bone surface (%) in the interradicular region of the first molar ($$n = 6$$). i Representative 3D scanned sections of the first molar at day 14 after OTM treatment. Arrow: direction of tooth movement. j Quantification from i. BV/TV of the interradicular region of the first molar ($$n = 6$$). k Representative HE staining images of the first molar at day 14 after OTM treatment. Scale bar: 200 μm. l Quantification from k. BV/TV of the interradicular region of the first molar ($$n = 6$$). m Representative ALP staining images of the interradicular region of the first molar at day 14 after OTM treatment. D: dentin. Scale bar: 200 μm. n Quantification from m. ALP-positive surface relative to bone surface (%) in the interradicular region of the first molar ($$n = 6$$). Data are shown as the mean ± SD. One-way ANOVA followed by Tukey’s post hoc multiple comparisons was performed. * $P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$; ns, not significant To further evaluate the function of exosomes derived from MS-BMDMs in alveolar bone formation during OTM, PBS, BMDM-EXOs and MS-BMDM-EXOs were locally injected into the palatal gingiva of the loaded tooth every 3 days after the OTM model was established (Additional file 1: Fig. S5). After force was applied for 14 days, Micro-CT analysis revealed that BMDM-EXOs treatment greatly increased the BV/TV of alveolar bone in the interradicular region of the loaded tooth compared with PBS, and it was further enhanced by MS-BMDM-EXOs (Fig. 3i, j). These results were consistent with the HE staining results (Fig. 3k, l). In addition, through ALP staining, we found that the ALP-positive surface in the interradicular region and tension side of the loaded alveolar bone was also significantly increased after treatment with BMDM-EXOs compared with PBS, and this increase was further promoted by MS-BMDM-EXOs (Fig. 3m, n; Additional file 1: Fig. S6). These results suggested that exosomes derived from MS-BMDMs promoted alveolar bone formation during tooth movement.
## UCHL3 is enriched in MS-BMDM-derived exosomes
We next analyzed the protein expression profiles in MS-BMDM-EXOs and BMDM-EXOs by proteomic analysis to explore the mechanism by which MS-BMDM-EXOs mediate BMSC osteogenesis. Comparing the protein expression levels in MS-BMDM-EXOs and BMDM-EXOs with a fold change > 2 or < 0.5 and $P \leq 0.05$ as the threshold cut-off, we identified 45 upregulated and 60 downregulated proteins, respectively, in the MS-BMDM-EXOs group (Fig. 4a, b). Subsequently, differentially expressed proteins were selected for bioinformatic analysis. Gene ontology (GO) enrichment analysis revealed that these proteins were involved in multiple biological processes in which cell protein modification and MAPK cascades were related to osteogenesis (Fig. 4c). Among all the 45 upregulated proteins, we identified UCHL3, which is ranked 11th among the upregulated proteins and participates in the cellular protein modification process (Fig. 4b). Previous studies demonstrated that UCHL3, a deubiquitinase, is a critical regulator of osteogenesis [34]. Our Western blotting results further revealed that the UCHL3 protein level was higher in BMDMs than in BMSCs, mechanical force increased the protein level of UCHL3 in BMDMs and BMDM-EXOs (Fig. 4d–f). And after treatment with BMDM-EXOs and MS-BMDM-EXOs, the UCHL3 protein level in BMSCs was improved (Additional file 1: Fig. S7). In addition, immunofluorescence staining showed that the number of UCHL3+F$\frac{4}{80}$+ macrophages was significantly increased after force application (Fig. 4g, h). Therefore, UCHL3 may be one of the most important factors in MS-BMDM-EXOs, and we selected it for further analyses (Additional file 1: Fig. S8).Fig. 4UCHL3 is enriched in MS-BMDM-derived exosomes. a Differentially expressed proteins between MS-BMDM-EXOs and BMDM-EXOs. Red, increased expression; blue, decreased expression; grey, no difference. $P \leq 0.05$ and fold change > 2 or < 0.5 were considered significant. b Heatmap diagram of differential protein expression between MS-BMDM-EXOs and BMDM-EXOs. Red, increased expression; blue, decreased expression; grey, no difference. Top 15 upregulated proteins in MS-BMDM-EXOs (UCHL3 was ranked eleventh). c GO Biological Process enrichment analysis of the differentially expressed proteins between MS-BMDM-EXOs and BMDM-EXOs. d Western blotting of UCHL3 protein levels in BMDMs and BMSCs. e Western blotting of UCHL3 protein levels in BMDMs and MS-BMDMs. f Western blotting of UCHL3 protein levels in BMDM-EXOs and MS-BMDM-EXOs. g, h Representative immunofluorescence staining and quantification of F$\frac{4}{80}$+ (red) UCHL3+ (green) cells in the control and OTM groups at 14 d after OTM treatment ($$n = 6$$). White arrows indicate F$\frac{4}{80}$+UCHL3+ cells. D: dentin, AB: Alveolar bone, PDL: periodontal ligament. Scale bar: 100 μm. Data are shown as the mean ± SD. Two-tailed unpaired Student’s t test was performed. *** $P \leq 0.001$; ns, not significant
## UCHL3 is required for MS-BMDM-EXO-mediated BMSCs osteogenesis in vitro
To examine the role of UCHL3 in MS-BMDM-EXO-mediated BMSC osteogenesis, we first used TCID, a small molecule inhibitor which could inhibit the deubiquitinase activity of UCHL3 [35], to treat BMSCs. qRT‒PCR and Western blotting showed that TCID treatment significantly reduced osteogenic gene (Alp, Runx2 Col1 and Ocn) and RUNX2 protein levels (Fig. 5a–c). ALP staining and Alizarin red staining revealed that TCID also diminished ALP level and extracellular matrix mineralization deposition in BMSCs (Fig. 5d, e). These results suggested that UCHL3 was critical for the osteogenesis of BMSCs. Fig. 5UCHL3 is required for MS-BMDM-EXO-mediated BMSC osteogenesis in vitro. a, b qRT‒PCR of the mRNA expression of osteogenesis genes and Western blotting of RUNX2 and UCHL3 protein levels in BMSCs after treatment with UCHL3 inhibitor TCID, and DMSO as a control ($$n = 3$$). c Relative intensity analysis of b ($$n = 3$$). d, e ALP staining and Alizarin red staining in BMSCs after treatment with TCID, and DMSO as a control. Scale bar: 100 μm. f, g qRT‒PCR of the mRNA expression of Uchl3 and Western blotting of UCHL3 protein level in MS-BMDMs after treatment with siCon and siUchl3 ($$n = 3$$). h Western blotting of UCHL3 protein level in exosomes derived from MS-BMDMs after treatment with siCon and siUchl3. i, j qRT‒PCR of the mRNA expression of osteogenic genes and Western blotting of UCHL3 and RUNX2 protein levels in BMSCs after treatment with MS-BMDMssiCon-EXOs and MS-BMDMssiUchl3-EXOs. k Relative intensity analysis of j ($$n = 3$$). l, m ALP staining and Alizarin red staining in BMSCs after treatment with MS-BMDMssicon-EXOs and MS-BMDMssiUchl3-EXOs. Two-tailed unpaired Student’s t test was performed. ** $P \leq 0.01$, ***$P \leq 0.001$; ns, not significant To further confirm the function of UCHL3 in MS-BMDM-EXO-mediated BMSC osteogenesis, we used siUchl3 to knockdown Uchl3 expression in MS-BMDMs. qRT‒PCR and Western blotting showed that it had an effective inhibitory efficiency on the mRNA and protein levels of UCHL3 (Fig. 5f, g). After MS-BMDMs were transfected with siUchl3, their supernatants were collected to isolate exosomes. Western blotting also revealed an effective inhibitory efficiency of siUchl3 for UCHL3 protein levels in MS-BMDMsiUchl3-EXOs (Fig. 5h). BMSCs were then treated with MS-BMDMsiCon-EXOs and MS-BMDMsiUchl3-EXOs. qRT‒PCR and Western blotting showed that after UCHL3 was downregulated in MS-BMDM-EXOs, osteogenic gene expression (Alp, Runx2, Col1 and Ocn) and the protein level of RUNX2 in BMSCs were also significantly downregulated compared with those in the MS-BMDMsiCon-EXOs group (Fig. 5i–k). Additionally, through ALP staining and Alizarin red staining, it was found that ALP level and extracellular matrix mineralization deposition were also impaired when UCHL3 was inhibited in MS-BMDM-EXOs (Fig. 5l, m). These results indicated that UCHL3 was required for MS-BMDM-EXO-mediated BMSC osteogenesis.
## UCHL3 is required for MS-BMDM-EXO-mediated alveolar bone formation during OTM
To identify the unique role of UCHL3 derived from MS-BMDM-EXOs in mediating alveolar bone formation during OTM, we first established an OTM model and performed intraperitoneal injection of the UCHL3 inhibitor TCID every 2 days during loading (Additional file 1: Fig. S9). After 14 days of force application, the tooth movement distance significantly decreased after TCID injection (Fig. 6a, b). Micro-CT analysis showed that compared with the OTM group, the BV/TV of alveolar bone in the interradicular region of the loaded tooth decreased after TCID injection (Fig. 6c, d). This phenomenon was consistent with the HE staining results (Fig. 6e, f). In addition, ALP staining showed that the ALP-positive surface in the interradicular region and tension side of loaded alveolar bone were significantly increased in the OTM group compared with the Control group, and they were suppressed after TCID injection (Fig. 6g, h; Additional file 1: Fig. S10). These findings indicated that UCHL3 was required for mechanical force-induced bone formation during OTM.Fig. 6UCHL3 is required for MS-BMDM-EXO-mediated alveolar bone formation during OTM. a Representative 3D Micro-CT reconstruction images of the murine OTM model at day 14 after OTM treatment. Arrow: direction of tooth movement. b Quantification of tooth movement distance at day 14 after OTM treatment ($$n = 4$$). c Representative 3D scanned sections of the first molar at day 14 after OTM treatment. d Quantification from c. BV/TV of the interradicular region of the first molar ($$n = 6$$). e Representative HE staining images of the first molar at day 14 after OTM treatment. Scale bar: 200 μm. f Quantification from e. BV/TV of the interradicular region of the first molar ($$n = 6$$). g Representative ALP staining images of the interradicular region of the first molar at day 14 after OTM treatment. D: dentin. Scale bar: 200 μm. h Quantification from g. ALP-positive surface relative to bone surface (%) in the interradicular region of the first molar ($$n = 6$$). i Representative 3D scanned sections of the first molar at day 14 after OTM treatment. Arrow: direction of tooth movement. j Quantification from i. BV/TV of the interradicular region of the first molar ($$n = 6$$). k Representative HE staining images of the first molar at day 14 after OTM treatment. Scale bar: 200 μm. l Quantification from k. BV/TV of the interradicular region of the first molar ($$n = 6$$). m Representative ALP staining images of the interradicular region of the first molar at day 14 after OTM treatment. D: dentin. Scale bar: 200 μm. n Quantification from m. ALP-positive surface relative to bone surface (%) in the interradicular region of the first molar. Data are shown as the mean ± SD. One-way ANOVA followed by Tukey’s post hoc multiple comparisons was performed. * $P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$; ns, not significant To further explore the function of UCHL3 in MS-BMDM-EXO-mediated alveolar bone formation during OTM, PBS, MS-BMDMsiCon-EXOs, and MS-BMDMsiUchl3-EXOs were locally injected into the palatal gingiva of the loaded tooth every 3 days after the OTM model was established (Additional file 1: Fig. S11). After 14 days, Micro-CT analysis revealed that after UCHL3 was downregulated in MS-BMDM-EXOs, the BV/TV of alveolar bone in the interradicular region of the loaded tooth was also reduced compared with MS-BMDMsiCon-EXOs (Fig. 6i, j). These results were consistent with the HE staining results (Fig. 6k, l). In addition, ALP staining showed that the ALP-positive surface in the interradicular region and tension side of loaded alveolar bone was also significantly impaired when UCHL3 was inhibited in MS-BMDM-EXOs (Fig. 6m, n; Additional file 1: Fig. S12). Collectively, these results indicated that UCHL3 was required for MS-BMDM-EXO-mediated alveolar bone formation during OTM.
## MS-BMDM‑derived exosomal UCHL3 promotes BMSCs osteogenesis by targeting SMAD1
We next explored the mechanism by which MS-BMDM-derived exosomal UCHL3 modulates BMSCs osteogenesis. It has been reported that UCHL3 can interact with SMAD1, a classical osteogenesis-related molecule involved in BMSC osteogenesis [34, 36, 37]. Thus, we hypothesized that MS-BMDM-derived exosomal UCHL3 could mediate BMSC osteogenesis through the regulation of SMAD1. To verify the correlation between UCHL3 and SMAD1, we used TCID and MS-BMDMsiUchl3-EXOs to treat BMSCs and found that the protein level of SMAD1 was downregulated when UCHL3 was knocked down (Fig. 7a–c). Next, BMSCs overexpressing SMAD1 with lentiviruses were treated with MS-BMDMsiCon-EXOs and MS-BMDMsiUchl3-EXOs. qRT‒PCR and Western blotting revealed an effective overexpression efficiency of SMAD1 in BMSCs (Additional file 1: Fig. S13). Overexpression of SMAD1 significantly reversed the downregulated osteogenic gene expression (Alp, Runx2, Col1, and Ocn) and protein level of RUNX2 caused by UCHL3 inhibition in MS-BMDM-EXOs (Fig. 7d–f). ALP staining and Alizarin red staining further confirmed a similar rescue effect of SMAD1 overexpression in promoting the decreased ALP level and extracellular matrix mineralization deposition caused by UCHL3 inhibition in MS-BMDM-EXOs (Fig. 7g, h).Fig. 7MS-BMDM‑derived exosomal UCHL3 promotes BMSCs osteogenesis by targeting SMAD1. a Western blotting of SMAD1 and UCHL3 protein levels in BMSCs after treatment with the UCHL3 inhibitor TCID or MS-BMDMsiUchl3-EXOs. b, c Relative intensity analysis of a ($$n = 3$$). d, e qRT‒PCR of the mRNA expression of osteogenesis genes and Western blotting of RUNX2, UCHL3 and SMAD1 protein levels in BMSCs after overexpression of Smad1 with lentiviruses and treatment with MS-BMDMsiUchl3-EXOs ($$n = 3$$). f Relative intensity analysis of e ($$n = 3$$). g, h ALP staining and Alizarin red staining in BMSCs after overexpression of Smad1 with lentiviruses and treatment with MS-BMDMsiUchl3-EXOs. Scale bar: 100 μm. i 3D modelling of the interaction between UCHL3 (blue) and SMAD1 (grey) proteins predicted by protein‒protein docking from Alphafold 2. The stick structure represents amino acid residues, and the yellow dashed lines represent hydrogen bonds. j Immunofluorescence images of the colocalization of SMAD1 and UCHL3 in BMSCs (green UCHL3, red SMAD1, blue DAPI). Scale bar: 10 μm. k Endogenous UCHL3 and SMAD1 proteins interacted in BMSCs. UCHL3 protein was immunoprecipitated with anti-UCHL3 antibody. IgG served as a negative control, and endogenous SMAD1 was measured by Western blotting. l Endogenous UCHL3 and SMAD1 proteins interacted in BMSCs. SMAD1 protein was immunoprecipitated with anti-SMAD1 antibody. IgG served as a negative control, and endogenous UCHL3 was measured by Western blotting. m Western blotting of the polyubiquitinated level of SMAD1 in BMSCs after treatment with TCID. SMAD1 protein was immunoprecipitated, and polyubiquitinated SMAD1 protein was measured by Western blotting using anti-Ub antibody. Data are shown as the mean ± SD. Two-tailed unpaired Student’s t test or one-way ANOVA followed by Tukey’s post hoc multiple comparisons were performed. * $P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$; ns, not significant Furthermore, we wondered how UCHL3 modulated the function of SMAD1 in BMSCs. The protein‒protein docking results from Alphafold 2 revealed that UCHL3 could directly bind with SMAD1 through hydrogen bonds (Fig. 7i). Since UCHL3 is a regulator of ubiquitin, we considered that UCHL3 may increase SMAD1 levels by regulating its degradation. Further immunofluorescence and co-immunoprecipitation (Co-IP) verified the interaction between UCHL3 and SMAD1 in BMSCs (Fig. 7j–l). In addition, inhibition of endogenous UCHL3 by TCID elevated the ubiquitination level of endogenous SMAD1 in BMSCs (Fig. 7m), which indicated that UCHL3 could elevate SMAD1 protein stabilization through deubiquitination. Meanwhile, the immunohistochemical staining experiment showed that UCHL3 inhibitor TCID reduced the SMAD1 level in the loaded alveolar bone in the tooth movement mice (Additional file 1: Fig. S14). In summary, these results indicated that SMAD1 was necessary for MS-BMDM-EXO-derived UCHL3 to promote BMSCs osteogenesis.
## Discussion
Orthodontic treatment is an effective way to correct malocclusion. The essence of orthodontic treatment is a local aseptic inflammation-associated alveolar bone remodelling process induced by mechanical force [5]. Previous studies have shown that during tooth movement, macrophages secrete inflammatory cytokine and modulate the alveolar bone remodeling inflammatory microenvironment [14, 15]. Nevertheless, the mechanism by which macrophages regulate mechanical force-induced alveolar bone remodelling during OTM is not clear. In the present study, we demonstrated that the exosomes derived from mechanical force-stimulated macrophages promote BMSCs osteogenesis, thereby promoting alveolar bone formation during OTM. Furthermore, we verified that exosomal UCHL3, by targeting SMAD1, plays a prominent role in MS-BMDM-EXO-mediated pro-osteogenesis. In summary, this study demonstrated the mechanism of mechanical force-induced alveolar bone formation during OTM from the perspective of macrophages regulating BMSCs osteogenesis through exosomes.
Macrophages can regulate BMSCs osteogenesis by secreting proinflammatory factors, e.g., TNF-α, or anti-inflammatory factors, e.g., IL-10 [38]. Wei et al. found that under mechanical stimulation, macrophages polarized to the M2 phenotype and produced anti‐inflammatory cytokines such as IL‐10 and TGF‐β to regulate the local inflammatory microenvironment and promote BMSC osteogenesis [25, 26]. Our study confirmed the promotive effect of force stimulation on BMDM-mediated BMSC osteogenesis. Exosomes, which contain miRNA, protein and nucleic acids, are important mediators between macrophages and BMSCs [39]. Cumulative studies have recently reported that under the stimulation of some biomimetic materials, exosomes derived from macrophages promote BMSCs osteogenesis [32, 40]. However, whether mechanical force regulates BMSCs osteogenesis via macrophage-derived exosomes remains unclear. Our study showed that BMSCs osteogenesis was reduced when exosomes derived from MS-BMDMs were inhibited. Compared with exosomes derived from BMDMs, exosomes derived from MS-BMDMs showed a stronger ability to promote BMSCs osteogenesis. Therefore, it is reasonable to conclude that mechanical force could regulate macrophage-derived exosomes to promote BMSCs osteogenesis.
Orthodontic tooth movement follows Wolff's law—the remodelling of bone and surrounding tissues is an adaptive biological response to the external force [41]. Xu et al. recently found that in mechanical force-induced alveolar bone remodelling, macrophages could respond to force stimulation through Piezo1 [42]. However, knowledge of how force-mediated macrophages regulate alveolar bone remodeling is still insufficient. Evidence suggests that exosomes derived from macrophages are involved in bone formation in fracture models [22], periodontitis models [43], and osteoporosis models [44]. In this study, using the exosome inhibitor GW4869 and exosomes derived from macrophages to treat tooth movement mouse models, we demonstrated that mechanical force-regulated macrophage-derived exosomes could promote alveolar bone formation during OTM.
Accumulating evidence has proved that exosomes are mechanosensitive [45]. Our study confirmed that exosomes are indeed mechanosensitive and discovered that there was a higher concentration and different cargo of macrophage-derived exosomes under mechanical stimulation. To explore the mechanism by which MS-BMDM-EXOs mediate BMSCs osteogenesis, we performed a proteomic study and found that MS-BMDM-EXOs contain much higher levels of the deubiquitinase UCHL3. Deubiquitinases, which oppose protein ubiquitination by hydrolyzing ubiquitin linkages, control the function or abundance of targeted proteins and influence physiological or pathological processes [46, 47]. While UCHL3 has been identified as a novel mediator in DNA repair [48], aerobic glycolysis [49], and cancer progression [50], its role in bone formation has not been clarified. Kim et al. verified the positive effect of UCHL3 on BMSCs osteogenesis [34]. However, whether UCHL3 derived from MS-BMDM-EXOs plays a role in BMSCs osteogenesis and mechanical force-induced alveolar bone formation remains unknown. We used the UCHL3 inhibitor TCID and UCHL3low exosomes from MS-BMDMs to treat BMSCs and tooth movement mouse models and observed that BMSCs osteogenesis and alveolar bone formation were reduced by TCID and UCHL3low exosomes.
SMAD1, an immediate downstream molecule of BMP receptors, is involved in the regulation of BMSCs osteogenesis by affecting BMP signal transduction [51–53]. It has been reported that UCHL3, which decreases the amount of polyubiquitinated SMAD1, might be critical for osteogenesis [34]. Consistently, our data showed that UCHL3 physically interacted with SMAD1 and that its expression in BMSCs was positively correlated with SMAD1. Overexpression of SMAD1 rescued the effect of inhibiting MS-BMDM-derived exosomal UCHL3 on BMSCs osteogenesis. Moreover, using a ubiquitination assay, we confirmed that UCHL3 decreases the amount of polyubiquitinated SMAD1. This finding indicated that UCHL3 promotes BMSCs osteogenesis by decreasing SMAD1 ubiquitination degradation.
Notably, we only characterized the protein profile in MS-BMDM-EXOs. It is possible that mRNAs, miRNAs or other noncoding RNAs in exosomes also play a key role in this pro-osteogenesis process. Moreover, although we found that MS-BMDM-derived exosomal UCHL3 promotes BMSCs osteogenesis by targeting SMAD1, additional questions, such as binding sites, still require clarification. Therefore, further studies are needed to evaluate the mechanism by which UCHL3 regulates SMAD1.
## Conclusions
In summary, we showed that mechanical force could alter the protein composition in macrophage-derived exosomes. Transfer of UCHL3 by exosomes from mechanically stimulated macrophages to BMSCs improved SMAD1 levels in BMSCs, which was beneficial for BMSCs osteogenesis, thereby promoting alveolar bone formation during OTM. These findings therefore demonstrated the mechanism of mechanical force-induced osteogenesis during OTM from the perspective of macrophage-derived exosomes regulating BMSCs osteogenesis, which may contribute to the development of new solutions to clinical issues in orthodontic treatment.
## Animals
C57BL/6 J mice (male, 8 weeks old) that used in the present study were from the Laboratory Animal Center of Nanjing Medical University. All experimental protocols were approved by the Committee of Nanjing Medical University for Animal Resources (approval ID: IACUC1601118).
## Application of orthodontic devices
Appropriate mechanical force was exerted on the molars of these mice for 14 days as described previously [33]. Briefly, the left maxillary first molar was ligated to the maxillary incisors by utilizing a nickel-titanium coil spring. The springs were activated to deliver the orthodontic force of 10 g. Every 24 h, the devices were reviewed and, if necessary, reinstalled. To alleviate pain, the mice were offered soft food. After 14 days, the mice were sacrificed for following experiments.
## Drugs and exosomes administration
For exosomes inhibitor treatment, mice were injected intraperitoneally with GW4869 (#HY-19363, MCE, USA) at one dose of 2.5 mg/kg every 2 days following the orthodontic device application. The same volume of DMSO was injected as that for the controls [54, 55].
For UCHL3 inhibitor treatment, mice were injected intraperitoneally with TCID (#HY-18638, MCE, USA) at one dose of 10 mg/kg every 2 days following the orthodontic device application. The same volume of DMSO was injected as that for the controls.
For exosomes treatment, injection of 10 μl (2 μg/μl) exosomes were given locally into the palatal gingiva of the loaded first maxillary molar every 3 days utilizing a 33-gauge needle Hamilton syringe (Hamilton Company, USA) [43, 56].
## Micro‑CT analysis
After being separated and fixed for 2 days in $4\%$ paraformaldehyde, maxillae of mice were scanned by a Micro-CT scanner (Skyscan 1176, Bruker) at a resolution of 10 μm. The Micro-CT data were input into DataViewer (Bruker) to carry out 3D reconstruction, and meanwhile they were imported to CTAn (Bruker) to perform quantitative analysis. After that, the region of interest was selected from the inter-radicular region of the maxillary first molar and bone structures parameter BV/TV was analyzed [6, 57]. Then the Micro-CT data were imported into mimics (V20.0, Materialise, Belgium) for reconstruction and tooth movement distance measurement. Using the occlusal view of the 3D-reconstructed image, we obtained the midpoints of the distal marginal ridge of the first molar and the mesial marginal ridge of the second molar. The gap between the two points was the tooth movement distance [58].
## Histologic and immunofluorescent staining
After being isolated, maxilla was fixed for 2 days in $4\%$ paraformaldehyde and decalcified for 28 days in $10\%$ EDTA solution. Then, the samples were embedded in paraffin, where they were sectioned into 4 μm sections. Subsequently, the paraffin sections were stained using H&E for routine histology and bone structures parameter BV/TV analysis. ALP staining was also performed to identify osteoblasts as previously described [59].
For tissue immunofluorescent staining, the samples were embedded in Tissue-Tek O.C.T and sectioned into 8 μm sections. After being washed with PBS for 30 min and blocked with $10\%$ BSA for 30 min at room temperature, the sections were incubated with primary antibodies against UCHL3 (1:200, #D25E6, Cell Signaling Technology, USA) overnight at 4 °C. Then the sections were washed with PBS, and incubated with fluorescein Cy3-conjugated secondary antibody (Beyotime, China) and fluorescein CoraLite488-conjugated secondary antibody (#SA00013-2, Proteintech, USA) for 1.5 h. Nuclei were counterstained with 4′,6-diamidino-2-phenylindole (DAPI, # H-1200, Vectorlabs, USA). Images were captured with a fluorescent microscope (Carl Zeiss).
There were two regions of interest: [1] the distal side of the mesial root of the maxillary first molar was used to characterize bone formation on the tension-side bone; [2] the inter-radicular region of the maxillary first molar was used to characterize general bone formation [6]. Quantification of the ALP-positive surface relative to the bone surface (%) was determined by ImageJ software (V1.8.0, National Institutes of Health, USA) [60].
## Cell culture
The bone marrow‐derived macrophages were isolated and cultured as described previously [60]. Specifically, the bone marrow cells isolated from the femurs and tibias of 4- to 6-week-old mice were cultured in the α-MEM medium added with $10\%$ fetal bovine serum (ScienCell, USA), $1\%$ penicillin/streptomycin (Gibco), and 25 ng/ml M-CSF (Peprotech, USA) for 5 days.
The bone marrow-derived mesenchymal stem cells (BMSCs) were isolated and cultured as described previously [61]. In brief, the bone marrow cells were first flushed out from the femurs and tibias of 4-week-old mice using α-MEM medium containing $12\%$ FBS (ScienCell, USA) and $1\%$penicillin/streptomycin (Gibco, USA). After that, the flushing fluid was moved into a 6 cm-culture dish and cultured with BMSCs normal culture medium above. BMSCs in the second or the third passages were used in the subsequent experiments.
## Mechanical stretch of BMDMs and BMDMs transfection
For the mechanical stretch studies, BMDMs were cultured in flexible-bottom six-well culture plates (Flexcell International Corporation, USA) with a density of 1 × 106 cells for each well. At the fifth day, BMDMs were treated with cyclic stretching at 0.5 Hz with different strains for specific durations using the FX-5000 T Flexcell Tension System (Flexcell International Corporation, USA). After mechanical stimulation, culture medium of BMDMs was refreshed. Two days later, the supernatants were collected for the following experiments.
For BMDMs transfection, we transfected BMDMs with 100 nM siRNA using Lipofectamine 2000 (#11668-019, Invitrogen, USA) following the instruction of the manufacturer after they were mechanically stimulated. Then 48 h later, the medium was replaced with medium containing EXOs-free FBS. And after another 48 h, supernatants were collected for isolating exosomes. The sequences were as follows: Uchl3-siRNA, 5′-GAACAGAAGAGGAAGAAAATT-3′ and 5′-UUUUCUUCCUCUUCUGUUCTT-3′, negative control Con-siRNA, 5′-UUCUCCGAACGUGUCACGUTT-3′ and 5′-ACGUGACACGUUCGGAGAATT-3′.
## Conditioned medium and exosomes treatment of BMSCs
After mechanical stimulation, culture medium of BMDMs was refreshed with medium containing EXOs-free FBS. Two days later, the supernatants were collected and conditioned medium was prepared by mixing the collecting BMDMs supernatants with normal BMSCs culture/osteogenic medium (#MUXMX-90021, Cyagen Biosciences, USA) at a ratio of 1:3. BMSCs were cultured with conditioned medium or medium containing exosomes (50 μg/ml) [43, 62]. After 7-day treatment, the cells were used for qRT-PCR, Western blot, and ALP staining. After 14-day treatment, these cells were used for alizarin red staining.
## Drugs treatment of cells
For exosomes inhibitor treatment, BMDMs were treated with GW4869 (#HY-19363, MCE, USA, 10 μM) [32, 63] after being mechanically stimulated. The same volume of DMSO was added as that for the controls. Two days later, the supernatants were collected for preparing conditioned medium.
For UCHL3 inhibitor treatment, BMSCs were treated with TCID (#HY-18638, MCE, USA, 10 μM) [35]. The same volume of DMSO was added as that for the controls. After being treated for 7 days, the cells were used for qRT-PCR, Western blot, and ALP staining. After being treated for 14 days, the cells were used for alizarin red staining.
## Lentivirus construction and infection
Lentivirus were constructed and infected as previously [64]. Briefly, overexpression lentiviruses of SMAD1 and their vector control (ov-Smad1, ov-NC) were purchased from Genechem (Shanghai, China). Lentiviruses and polybrene (5 μg/ml, Sigma) were added to the medium and incubated with BMSCs for 24 h at a multiplicity of infection (MOI) of 50. After 24 h, the medium was refreshed and the cells were used for the following experiments.
## CCK‑8 (Cell counting kit‑8) assay
After mechanical stimulation, BMDMs were incubated with 2 ml fresh medium and 200 μl CCK-8 reagent (#96,992, Sigma-Aldrich, USA) under 37 °C for 30 min, and then the absorbance was measured at 450 nm.
## Exosome purification, characterization and uptake
Exosomes were isolated and purified from macrophage culture supernatants following previously established protocol [65, 66]. Briefly, after culture for 48 h, the culture supernatants were collected and centrifuged at 300 g for 10 min, 2000 g for 10 min, and 10,000 g for 30 min to eliminate cells and debris. Subsequently, the supernatants were ultracentrifuged at 100,000 g for 70 min for two times, and the pellets were resuspended in PBS and kept under -80℃ (Additional file 1: Fig. S15).
Transmission electron microscopy (TEM) (JEM, Japan) and Western blot analysis were applied to assess the morphology and surface markers expression of exosomes. Exosomes concentration and size distribution were determined using Nanoparticle tracking analysis (NTA) (ZetaView, Particle Metrix, Germany).
The uptake of exosomes by BMSCs was confirmed through labeling exosomes using Dil as previously described [67]. Briefly, the purified exosomes were labeled with the membrane-labeling dye Dil (#V22888, Thermo Fisher Scientific, USA) following the manufacturer’s instructions, and these labeled exosomes were re-suspended in sterile PBS. Then with ultracentrifugation at 100,000 g for 70 min, the Dil-labeled exosomes were isolated. Next, Dil-EXOs were co-cultured with BMSCs for 12 h, after which these cells were fixed with $4\%$ paraformaldehyde and photographed using fluorescence microscope (Leica Microsystems, Germany).
## RNA preparation and quantitative real-time PCR (qRT‒PCR)
In accordance with the operation instruction, the total RNA within the cells was isolated by RNA Isolation Kits (Omega, Guangzhou, China), and cDNA was reversely transcribed with the application of HiScript II Q RT SuperMix (Vazyme, China). Next, qRT-PCR was conducted by AceQ qPCR SYBR Green Master Mix (Vazyme, China) on the ABI QuantStudio7 real-time PCR system (Applied Biosystems, USA), and all the primer sequences were given in Table 1. Finally, all data were normalized to Gapdh expression and quantified with the 2−ΔΔCT method. Table 1The sequences of the primersqRT‒PCR primer namePrimer sequence (5′-3′)Gapdh (forward)TCCATGACAACTTTGGTATCGGapdh (reserve)TGTAGCCAAATTCGTTGTCAAlp (forward)ATAACGAGATGCCACCAGAGGAlp (reserve)TTCCACATCAGTTCTGTTCTTCGRunx2 (forward)AGAATGGACGTGCCCCCTARunx2 (reserve)CTGGGGAAGCAGCAACACTACol1 (forward)CTGACTGGAAGAGCGGAGAGCol1 (reserve)CGGCTGAGTAGGGAACACACOcn (forward)TTCTGCTCACTCTGCTGACCCOcn (reserve)CTGATAGCTCGTCACAAGCAGGUchl3 (forward)AGCAATGCCTGTGGAACGATUchl3 (reserve)TTTGGCTCTCTCTTCAGGGCSmad1 (forward)ACCCCTACCACTATAAGCGAGSmad1 (reserve)TGCTGGAAAGAGTCTGGGAAC
## Western blotting
Western blotting was carried out as described previously [68]. Firstly, the total proteins in cell lysates were harvested, and then they were separated using SDS-PAGE gels. Later, the proteins that had been separated were transferred to polyvinylidene difluoride membranes, which were then blocked in the fat free milk (concentration of $5\%$). Subsequently, the following antibodies were incubated under 4 °C overnight: anti-RUNX2 (#12556, Cell Signaling Technology, 1:1000), anti-TSG101 (#ab125011, Abcam, 1:1000), anti-CD63 (#ab217345, Abcam, 1:1000), anti-Calnexin (#ab213243, Abcam, 1:1000), anti-UCHL3 (#D25E6, Cell Signaling Technology, 1:1000), anti-SMAD1 (#AP20642c, Abcepta, 1:1000), and anti-β-ACTIN (#4976, Cell Signaling Technology, 1:1000). After incubated with horseradish peroxidase‐conjugated secondary antibodies (Shengxing Biological, Nanjing, China) for 1 h, the specific bands were visualized with the enhanced chemiluminescence (Tanon). The relative density was measured using ImageJ software (V1.8.0, National Institutes of Health, USA).
## ALP staining
ALP staining was carried out using the 1-step NBT/BCIP reagent (#34042, Thermo Fisher Scientific, USA) after BMSCs were cultured for 7 days. Briefly, the cells were washed for two times by PBS, and then they were fixed in $4\%$ paraformaldehyde for 30 min. After that, BMSCs were stained using BCIP/NBT substrate for 10 min and washed with PBS.
## Alizarin red staining
To detect matrix mineralization, alizarin red staining was conducted after BMSCs were cultured for 14 days. Later BMSCs was washed for two times by PBS and fixed in $4\%$ paraformaldehyde for 30 min. Finally, they were stained using $2\%$ alizarin red (#DS0072, Leagene, China) for 10 min and washed with deionized water.
## Cell immunofluorescent staining
BMSCs were fixed with $4\%$ paraformaldehyde for 30 min and permeabilized with $0.5\%$ Triton X-100 for 20 min. Then they were treated with goat serum for 30 min and incubated with the following primary antibodies overnight at 4 °C: anti-UCHL3 (#D25E6, Cell Signaling Technology, 1:200), anti-SMAD1 (sc-7965, Santa Cruz, 1:200). Subsequently, the species-matched secondary antibodies were used, and the nucleus were stained with DAPI.
## Coimmunoprecipitation (Co-IP) assay
Cells were harvested and lysed in IP buffer. Then, 50 μl protein-G agarose beads (#sc-2003, Santa Cruz) were added to 1 mg total protein, and incubated with rocking at 4 °C for 15 min. After centrifugation at 12,000 rpm for 10 min, the supernatant was collected and incubated with 2 μg anti-UCHL3 (#D25E6, Cell Signaling Technology), anti-SMAD1 (#AP20642c, Abcepta) or IgG (#A7028, Bytotime) antibodies at 4 °C for 2 h. Then 60 μl protein-G agarose beads were added, and the mixture was incubated with rocking at 4 °C overnight. After the agarose beads were collected, washed, and resuspended in 60 μl of $2\%$ protein loading buffer, the samples were boiled for 10 min and used for Western blotting.
## Protein–protein docking
Protein–protein docking was performed using Alphafold2. The full-length sequence of protein was downloaded from UniProt (https://www.uniprot.org/). Then the two protein sequences were input into Alphafold2 (https://github.com/sokrypton/ColabFold) for docking.
## Proteomic analysis
Exosomes were isolated from BMDMs and MS-BMDMs supernatants. The protein extracted from exosomes was used for proteomic analysis by HOOGEN BIOTECH (Shanghai, China). In brief, the extracted protein from exosomes was subjected to reduction, alkylation and trypsin digestion, and then the digested protein was used for LC‒MS/MS analyses (Thermo Fisher Scientific, USA). After LC‒MS/MS, protein identification and quantitative analysis, as well as differential protein screening with a cut-off of fold change > 2.0 or < 0.5 and P values < 0.05, were performed. Then, bioinformatics analysis was conducted, and Gene Ontology and hierarchical clusters with corrected P values < 0.05 were considered significant.
## Statistical analysis
The data were expressed in the form of mean ± SD. We carried out statistical analysis using GraphPad Prism 8 software (GraphPad, USA). Two-tailed unpaired Student’s t-test were used for comparisons between 2 groups. And one-way ANOVA followed by Tukey's post-hoc multiple comparisons was conducted to comparing among 3 or more groups. P values < 0.05 were considered statistically significant.
## Supplementary Information
Additional file 1: Fig. S1 Identification of BMSCs. a Osteogenic differentiation capability of third‐generation BMSCs detected by Alizarin red staining. Scale bar: 100 μm. b Adipogenic differentiation capability of third‐generation BMSCs detected by Oil red staining. Scale bar: 100 μm. c Chondrogenic differentiation capability of third‐generation BMSCs detected by Toluidine blue staining. Scale bar: 100 μm. Fig. S2 a Cell viability of BMDMs under the influences of different strain levels examined by CCK-8 assay ($$n = 3$$). b Cell viability of BMDMs under the influences of different strain durations examined by CCK-8 assay ($$n = 3$$). Data are shown as the mean ± SD. One-way ANOVA followed by Tukey’s post hoc multiple comparisons was performed. * $P \leq 0.05$, **$P \leq 0.01$; ns, not significant. Fig. S3 Experimental design. The OTM model was generated in 2-month-old mice and exosome level was blocked by intraperitoneal injection of GW4869 every 2 days during loading. After 14 days of OTM, the maxillary was harvested. Fig. S4, S6, S10, S12 a Representative ALP staining images of the interradicular region of the first molar at day 14 after OTM treatment. The square frame represents the alveolar bone on the tension side of the first molar D: dentin. Scale bar: 200 μm. b Quantification from the square frame of a. ALP-positive surface relative to bone surface (%) on the tension side of the first molar ($$n = 6$$). Data are shown as the mean ± SD. One-way ANOVA followed by Tukey’s post hoc multiple comparisons was performed. * $P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$; ns, not significant. Fig. S5 Experimental design. The OTM model was generated in 2-month-old mice. The PBS, BMDM-EXOs, and MS-BMDM-EXOs were locally injected into the palatal gingiva of the loaded first molar every 3 days. After 14 days of OTM, the maxillary was harvested. Fig. S7 Western blotting of UCHL3 protein level in BMSCs after treatment with the PBS, BMDM-EXOs and MS-BMDM-EXOs. Fig. S8 The rule of finding the key functional protein in this study. Fig. S9 Experimental design. The OTM model was generated in 2-month-old mice and UCHL3 level was inhibited by intraperitoneal injection of TCID every 2 days during loading. After 14 days of OTM, the maxillary was harvested. Fig. S11 Experimental design. The OTM model was generated in 2-month-old mice. The PBS, MS-BMDMsiCon-EXOs, and MS-BMDMsiUchl3-EXOs were locally injected into the palatal gingiva of the loaded first molar every 3 days. After 14 days of OTM, the maxillary was harvested. Fig. S13 a qRT‒PCR of the mRNA expression of Smad1 in BMSCs after treatment with the lentiviruses of Ov-NC and Ov-Smad1 ($$n = 3$$). b Western blotting of SMAD1 protein level in BMSCs after treatment with the lentiviruses of Ov-NC and Ov-Smad1. Fig. S14 a Representative immunohistochemical staining of SMAD1 in the loaded alveolar bone at 14 d after OTM treatment ($$n = 6$$). D: dentin, Scale bar: 200 μm. b Quantification from a. SMAD1-positive surface relative to bone surface (%) in the loaded alveolar bone. Data are shown as the mean ± SD. One-way ANOVA followed by Tukey’s post hoc multiple comparisons was performed. *** $P \leq 0.001$; ns, not significant. Fig. S15 The procedures of isolating and purifying the exosomes.
## References
1. Akbari M, Lankarani KB, Honarvar B, Tabrizi R, Mirhadi H, Moosazadeh M. **Prevalence of malocclusion among Iranian children: a systematic review and meta-analysis**. *Dent Res J* (2016) **13** 387-395. DOI: 10.4103/1735-3327.192269
2. Mtaya M, Brudvik P, Astrom AN. **Prevalence of malocclusion and its relationship with socio-demographic factors, dental caries, and oral hygiene in 12- to 14-year-old Tanzanian schoolchildren**. *Eur J Orthod* (2009) **31** 467-476. DOI: 10.1093/ejo/cjn125
3. Zou J, Meng M, Law CS, Rao Y, Zhou X. **Common dental diseases in children and malocclusion**. *Int J Oral Sci* (2018) **10** 7. DOI: 10.1038/s41368-018-0012-3
4. Krishnan V, Davidovitch Z. **On a path to unfolding the biological mechanisms of orthodontic tooth movement**. *J Dent Res* (2009) **88** 597-608. DOI: 10.1177/0022034509338914
5. Chaushu S, Klein Y, Mandelboim O, Barenholz Y, Fleissig O. **Immune changes induced by orthodontic forces: a critical review**. *J Dent Res* (2022) **101** 11-20. DOI: 10.1177/00220345211016285
6. Jin A, Hong Y, Yang Y, Xu H, Huang X, Gao X, Gong X, Dai Q, Jiang L. **FOXO3 mediates tooth movement by regulating force-induced osteogenesis**. *J Dent Res* (2022) **101** 196-205. DOI: 10.1177/00220345211021534
7. Jeon HH, Yang CY, Shin MK, Wang J, Patel JH, Chung CH, Graves DT. **Osteoblast lineage cells and periodontal ligament fibroblasts regulate orthodontic tooth movement that is dependent on Nuclear Factor-kappa B (NF-kB) activation**. *Angle Orthod* (2021) **91** 664-671. DOI: 10.2319/031520-182.1
8. He D, Liu F, Cui S, Jiang N, Yu H, Zhou Y, Liu Y, Kou X. **Mechanical load-induced H2S production by periodontal ligament stem cells activates M1 macrophages to promote bone remodeling and tooth movement via STAT1**. *Stem Cell Res Ther* (2020) **11** 112. DOI: 10.1186/s13287-020-01607-9
9. Chen X, Liu Y, Ding W, Shi J, Li S, Liu Y, Wu M, Wang H. **Mechanical stretch-induced osteogenic differentiation of human jaw bone marrow mesenchymal stem cells (hJBMMSCs) via inhibition of the NF-kappaB pathway**. *Cell Death Dis* (2018) **9** 207. DOI: 10.1038/s41419-018-0279-5
10. Xu H, Xia M, Sun L, Wang H, Zhang WB. **Osteocytes enhance osteogenesis by autophagy-mediated FGF23 secretion under mechanical tension**. *Front Cell Dev Biol* (2021) **9** 782736. DOI: 10.3389/fcell.2021.782736
11. Atcha H, Meli VS, Davis CT, Brumm KT, Anis S, Chin J, Jiang K, Pathak MM, Liu WF. **Crosstalk between CD11b and Piezo1 mediates macrophage responses to mechanical cues**. *Front Immunol* (2021) **12** 689397. DOI: 10.3389/fimmu.2021.689397
12. Li R, Serrano JC, Xing H, Lee TA, Azizgolshani H, Zaman M, Kamm RD. **Interstitial flow promotes macrophage polarization toward an M2 phenotype**. *Mol Biol Cell* (2018) **29** 1927-1940. DOI: 10.1091/mbc.E18-03-0164
13. Wang Y, Zeng Z, Cheng Y, Zhao L, Yan Q, Qiu Y, Hu J, Guo J. **Orthodontic compressive force modulates Ets-1/Tks5 pathway to promote the formation of circumferential invadopodia and the fusion of osteoclast precursors**. *J Cell Physiol* (2019) **234** 12685-12691. DOI: 10.1002/jcp.27879
14. He D, Kou X, Yang R, Liu D, Wang X, Luo Q, Song Y, Liu F, Yan Y, Gan Y, Zhou Y. **M1-like Macrophage polarization promotes orthodontic tooth movement**. *J Dent Res* (2015) **94** 1286-1294. DOI: 10.1177/0022034515589714
15. Zhang S, Zhang H, Jin Z, Wang S, Wang Y, Zhu L, Sun W, Yan B. **Fucoidan inhibits tooth movement by promoting restorative macrophage polarization through the STAT3 pathway**. *J Cell Physiol* (2020) **235** 5938-5950. DOI: 10.1002/jcp.29519
16. Sun X, Ma Z, Zhao X, Jin W, Zhang C, Ma J, Qiang L, Wang W, Deng Q, Yang H, Zhao J, Liang Q, Zhou X, Li T, Wang J. **Three-dimensional bioprinting of multicell-laden scaffolds containing bone morphogenic protein-4 for promoting M2 macrophage polarization and accelerating bone defect repair in diabetes mellitus**. *Bioact Mater* (2021) **6** 757-769. DOI: 10.1016/j.bioactmat.2020.08.030
17. Dai X, Heng BC, Bai Y, You F, Sun X, Li Y, Tang Z, Xu M, Zhang X, Deng X. **Restoration of electrical microenvironment enhances bone regeneration under diabetic conditions by modulating macrophage polarization**. *Bioact Mater* (2021) **6** 2029-2038. DOI: 10.1016/j.bioactmat.2020.12.020
18. Wang M, Chen F, Wang J, Chen X, Liang J, Yang X, Zhu X, Fan Y, Zhang X. **Calcium phosphate altered the cytokine secretion of macrophages and influenced the homing of mesenchymal stem cells**. *J Mater Chem B* (2018) **6** 4765-4774. DOI: 10.1039/C8TB01201F
19. Tian G, Jiang S, Li J, Wei F, Li X, Ding Y, Yang Z, Sun Z, Zha K, Wang F, Huang B, Peng L, Wang Q, Tian Z, Yang X, Wang Z, Guo Q, Guo W, Liu S. **Cell-free decellularized cartilage extracellular matrix scaffolds combined with interleukin 4 promote osteochondral repair through immunomodulatory macrophages: In vitro and in vivo preclinical study**. *Acta Biomater* (2021) **127** 131-145. DOI: 10.1016/j.actbio.2021.03.054
20. Zheng ZW, Chen YH, Wu DY, Wang JB, Lv MM, Wang XS, Sun J, Zhang ZY. **Development of an accurate and proactive immunomodulatory strategy to improve bone substitute material-mediated osteogenesis and angiogenesis**. *Theranostics* (2018) **8** 5482-5500. DOI: 10.7150/thno.28315
21. Kang M, Huang CC, Lu Y, Shirazi S, Gajendrareddy P, Ravindran S, Cooper LF. **Bone regeneration is mediated by macrophage extracellular vesicles**. *Bone* (2020) **141** 115627. DOI: 10.1016/j.bone.2020.115627
22. Xiong Y, Chen L, Yan C, Zhou W, Yu T, Sun Y, Cao F, Xue H, Hu Y, Chen D, Mi B, Liu G. **M2 Macrophagy-derived exosomal miRNA-5106 induces bone mesenchymal stem cells towards osteoblastic fate by targeting salt-inducible kinase 2 and 3**. *J Nanobiotechnology* (2020) **18** 66. DOI: 10.1186/s12951-020-00622-5
23. Pegtel DM, Gould SJ. **Exosomes**. *Annu Rev Biochem* (2019) **88** 487-514. DOI: 10.1146/annurev-biochem-013118-111902
24. Fitzgerald W, Freeman ML, Lederman MM, Vasilieva E, Romero R, Margolis L. **A system of cytokines encapsulated in extracellular vesicles**. *Sci Rep* (2018) **8** 8973. DOI: 10.1038/s41598-018-27190-x
25. Liang W, Ding P, Qian J, Li G, Lu E, Zhao Z. **Polarized M2 macrophages induced by mechanical stretching modulate bone regeneration of the craniofacial suture for midfacial hypoplasia treatment**. *Cell Tissue Res* (2021) **386** 585-603. DOI: 10.1007/s00441-021-03533-5
26. Dong L, Song Y, Zhang Y, Zhao W, Wang C, Lin H, Al-Ani MK, Liu W, Xue R, Yang L. **Mechanical stretch induces osteogenesis through the alternative activation of macrophages**. *J Cell Physiol* (2021) **236** 6376-6390. DOI: 10.1002/jcp.30312
27. Tang CY, Wu M, Zhao D, Edwards D, McVicar A, Luo Y, Zhu G, Wang Y, Zhou HD, Chen W, Li YP. **Runx1 is a central regulator of osteogenesis for bone homeostasis by orchestrating BMP and WNT signaling pathways**. *Plos Genet* (2021) **17** e1009233. DOI: 10.1371/journal.pgen.1009233
28. Liu K, Luo X, Lv ZY, Zhang YJ, Meng Z, Li J, Meng CX, Qiang HF, Hou CY, Hou L, Liu FZ, Zhang B. **Macrophage-derived exosomes promote bone mesenchymal stem cells towards osteoblastic fate through microRNA-21a-5p**. *Front Bioeng Biotechnol* (2021) **9** 801432. DOI: 10.3389/fbioe.2021.801432
29. Guo S, Chen J, Chen F, Zeng Q, Liu WL, Zhang G. **Exosomes derived from Fusobacterium nucleatum-infected colorectal cancer cells facilitate tumour metastasis by selectively carrying miR-1246/92b-3p/27a-3p and CXCL16**. *Gut* (2020) **70** 1507-1519. DOI: 10.1136/gutjnl-2020-321187
30. Xiong H, Huang Z, Yang Z, Lin Q, Yang B, Fang X, Liu B, Chen H, Kong J. **Recent progress in detection and profiling of cancer cell-derived exosomes**. *Small* (2021) **17** e2007971. DOI: 10.1002/smll.202007971
31. Zhu J, Liu B, Wang Z, Wang D, Ni H, Zhang L, Wang Y. **Exosomes from nicotine-stimulated macrophages accelerate atherosclerosis through miR-21-3p/PTEN-mediated VSMC migration and proliferation**. *Theranostics* (2019) **9** 6901-6919. DOI: 10.7150/thno.37357
32. Liu A, Jin S, Fu C, Cui S, Zhang T, Zhu L, Wang Y, Shen S, Jiang N, Liu Y. **Macrophage-derived small extracellular vesicles promote biomimetic mineralized collagen-mediated endogenous bone regeneration**. *Int J Oral Sci* (2020) **12** 33. DOI: 10.1038/s41368-020-00100-6
33. Xu H, Zhang S, Sathe AA, Jin Z, Guan J, Sun W, Xing C, Zhang H, Yan B. **CCR2(+) macrophages promote orthodontic tooth movement and alveolar bone remodeling**. *Front Immunol* (2022) **13** 835986. DOI: 10.3389/fimmu.2022.835986
34. Kim JY, Lee JM, Cho JY. **Ubiquitin C-terminal hydrolase-L3 regulates Smad1 ubiquitination and osteoblast differentiation**. *Febs Lett* (2011) **585** 1121-1126. DOI: 10.1016/j.febslet.2011.03.053
35. Ouyang L, Yan B, Liu Y, Mao C, Wang M, Liu N, Wang Z, Liu S, Shi Y, Chen L, Wang X, Cheng Y, Cao Y, Xiao D, Zhang L, Liu S, Tao Y. **The deubiquitylase UCHL3 maintains cancer stem-like properties by stabilizing the aryl hydrocarbon receptor**. *Signal Transduct Target Ther* (2020) **5** 78. DOI: 10.1038/s41392-020-0181-3
36. Chen CN, Chang HI, Yen CK, Liu WL, Huang KY. **mechanical stretch induced osteogenesis on human annulus fibrosus cells through upregulation of BMP-2/6 heterodimer and activation of P38 and SMAD1/5/8 signaling pathways**. *Cells-Basel* (2022) **11** 2600. DOI: 10.3390/cells11162600
37. Li Z, Xiang S, Lin Z, Li EN, Yagi H, Cao G, Yocum L, Li HT, Bruce KK, Fritch MR, Hu H, Wang B, Alexander PG, Khor KA, Tuan RS, Lin H. **Graphene oxide-functionalized nanocomposites promote osteogenesis of human mesenchymal stem cells via enhancement of BMP-SMAD1/5 signaling pathway**. *Biomaterials* (2021) **277** 121082. DOI: 10.1016/j.biomaterials.2021.121082
38. Valles G, Bensiamar F, Maestro-Paramio L, Garcia-Rey E, Vilaboa N, Saldana L. **Influence of inflammatory conditions provided by macrophages on osteogenic ability of mesenchymal stem cells**. *Stem Cell Res Ther* (2020) **11** 57. DOI: 10.1186/s13287-020-1578-1
39. Zhang D, Wu Y, Li Z, Chen H, Huang S, Jian C, Yu A. **MiR-144-5p, an exosomal miRNA from bone marrow-derived macrophage in type 2 diabetes, impairs bone fracture healing via targeting Smad1**. *J Nanobiotechnology* (2021) **19** 226. DOI: 10.1186/s12951-021-00964-8
40. Zhu Y, Zhao S, Cheng L, Lin Z, Zeng M, Ruan Z, Sun B, Luo Z, Tang Y, Long H. **Mg(2+) -mediated autophagy-dependent polarization of macrophages mediates the osteogenesis of bone marrow stromal stem cells by interfering with macrophage-derived exosomes containing miR-381**. *J Orthop Res* (2022) **40** 1563-1576. DOI: 10.1002/jor.25189
41. Uda Y, Azab E, Sun N, Shi C, Pajevic PD. **Osteocyte mechanobiology**. *Curr Osteoporos Rep* (2017) **15** 318-325. DOI: 10.1007/s11914-017-0373-0
42. Xu H, Guan J, Jin Z, Yin C, Wu S, Sun W, Zhang H, Yan B. **Mechanical force modulates macrophage proliferation via Piezo1-AKT-Cyclin D1 axis**. *Faseb J* (2022) **36** e22423. DOI: 10.1096/fj.202200314R
43. Chen X, Wan Z, Yang L, Song S, Fu Z, Tang K, Chen L, Song Y. **Exosomes derived from reparative M2-like macrophages prevent bone loss in murine periodontitis models via IL-10 mRNA**. *J Nanobiotechnology* (2022) **20** 110. DOI: 10.1186/s12951-022-01314-y
44. Yu L, Hu M, Cui X, Bao D, Luo Z, Li D, Li L, Liu N, Wu Y, Luo X, Ma Y. **M1 macrophage-derived exosomes aggravate bone loss in postmenopausal osteoporosis via a microRNA-98/DUSP1/JNK axis**. *Cell Biol Int* (2021) **45** 2452-2463. DOI: 10.1002/cbin.11690
45. Yan L, Liu G, Wu X. **Exosomes derived from umbilical cord mesenchymal stem cells in mechanical environment show improved osteochondral activity via upregulation of LncRNA H19**. *J Orthop Translat* (2021) **26** 111-120. DOI: 10.1016/j.jot.2020.03.005
46. Tang Y, Lv L, Li W, Zhang X, Jiang Y, Ge W, Zhou Y. **Protein deubiquitinase USP7 is required for osteogenic differentiation of human adipose-derived stem cells**. *Stem Cell Res Ther* (2017) **8** 186. DOI: 10.1186/s13287-017-0637-8
47. Sun T, Liu Z, Yang Q. **The role of ubiquitination and deubiquitination in cancer metabolism**. *Mol Cancer* (2020) **19** 146. DOI: 10.1186/s12943-020-01262-x
48. Liu M, Chen H, Chen X, Xiong J, Song Z. **Silencing UCHL3 enhances radio-sensitivity of non-small cell lung cancer cells by inhibiting DNA repair**. *Aging* (2021) **13** 14277-14288. DOI: 10.18632/aging.203043
49. Fan Y, Hu D, Li D, Ma C, Tang Y, Tao Q, Deng L, Tang D. **UCHL3 promotes aerobic glycolysis of pancreatic cancer through upregulating LDHA expression**. *Clin Transl Oncol* (2021) **23** 1637-1645. DOI: 10.1007/s12094-021-02565-1
50. Song Z, Li J, Zhang L, Deng J, Fang Z, Xiang X, Xiong J. **UCHL3 promotes pancreatic cancer progression and chemo-resistance through FOXM1 stabilization**. *Am J Cancer Res* (2019) **9** 1970-1981. PMID: 31598398
51. Lee JS, Kim ME, Seon JK, Kang JY, Yoon TR, Park YD, Kim HK. **Bone-forming peptide-3 induces osteogenic differentiation of bone marrow stromal cells via regulation of the ERK1/2 and Smad1/5/8 pathways**. *Stem Cell Res* (2018) **26** 28-35. DOI: 10.1016/j.scr.2017.11.016
52. Liu C, Liu AS, Zhong D, Wang CG, Yu M, Zhang HW, Xiao H, Liu JH, Zhang J, Yin K. **Circular RNA AFF4 modulates osteogenic differentiation in BM-MSCs by activating SMAD1/5 pathway through miR-135a-5p/FNDC5/Irisin axis**. *Cell Death Dis* (2021) **12** 631. DOI: 10.1038/s41419-021-03877-4
53. Yang W, Zhu W, Yang Y, Guo M, Qian H, Jiang W, Chen Y, Lian C, Xu Z, Bai H, Chen T, Zhang J. **Exosomal miR-100-5p inhibits osteogenesis of hBMSCs and angiogenesis of HUVECs by suppressing the BMPR2/Smad1/5/9 signalling pathway**. *Stem Cell Res Ther* (2021) **12** 390. DOI: 10.1186/s13287-021-02438-y
54. Faict S, Muller J, De Veirman K, De Bruyne E, Maes K, Vrancken L, Heusschen R, De Raeve H, Schots R, Vanderkerken K, Caers J, Menu E. **Exosomes play a role in multiple myeloma bone disease and tumor development by targeting osteoclasts and osteoblasts**. *Blood Cancer J* (2018) **8** 105. DOI: 10.1038/s41408-018-0139-7
55. Dinkins MB, Dasgupta S, Wang G, Zhu G, Bieberich E. **Exosome reduction in vivo is associated with lower amyloid plaque load in the 5XFAD mouse model of Alzheimer's disease**. *Neurobiol Aging* (2014) **35** 1792-1800. DOI: 10.1016/j.neurobiolaging.2014.02.012
56. Nakao Y, Fukuda T, Zhang Q, Sanui T, Shinjo T, Kou X, Chen C, Liu D, Watanabe Y, Hayashi C, Yamato H, Yotsumoto K, Tanaka U, Taketomi T, Uchiumi T, Le AD, Shi S, Nishimura F. **Exosomes from TNF-alpha-treated human gingiva-derived MSCs enhance M2 macrophage polarization and inhibit periodontal bone loss**. *Acta Biomater* (2021) **122** 306-324. DOI: 10.1016/j.actbio.2020.12.046
57. Guo S, Ni Y, Ben J, Xia Y, Zhou T, Wang D, Ni J, Bai H, Wang L, Ma J, Chen Q. **Class A scavenger receptor exacerbates osteoclastogenesis by an interleukin-6-mediated mechanism through ERK and JNK signaling pathways**. *Int J Biol Sci* (2016) **12** 1155-1167. DOI: 10.7150/ijbs.14654
58. Cao H, Kou X, Yang R, Liu D, Wang X, Song Y, Feng L, He D, Gan Y, Zhou Y. **Force-induced Adrb2 in periodontal ligament cells promotes tooth movement**. *J Dent Res* (2014) **93** 1163-1169. DOI: 10.1177/0022034514551769
59. Wang H, Chen Y, Li W, Sun L, Chen H, Yang Q, Zhang H, Zhang W, Yuan H, Zhang H, Xing L, Sun W. **Effect of VEGFC on lymph flow and inflammation-induced alveolar bone loss**. *J Pathol* (2020) **251** 323-335. DOI: 10.1002/path.5456
60. Yang Q, Zhao W, Chen Y, Chen Y, Shi J, Qin R, Wang H, Wang R, Yuan H, Sun W. **RelA/MicroRNA-30a/NLRP3 signal axis is involved in rheumatoid arthritis via regulating NLRP3 inflammasome in macrophages**. *Cell Death Dis* (2021) **12** 1060. DOI: 10.1038/s41419-021-04349-5
61. Yang L, Li Y, Gong R, Gao M, Feng C, Liu T, Sun Y, Jin M, Wang D, Yuan Y, Yan G, He M, Idiiatullina E, Ma W, Han Z, Zhang L, Huang Q, Ding F, Cai B, Yang F. **The long non-coding RNA-ORLNC1 regulates bone mass by directing mesenchymal stem cell fate**. *Mol Ther* (2019) **27** 394-410. DOI: 10.1016/j.ymthe.2018.11.019
62. Song X, Xue Y, Fan S, Hao J, Deng R. **Lipopolysaccharide-activated macrophages regulate the osteogenic differentiation of bone marrow mesenchymal stem cells through exosomes**. *PeerJ* (2022) **10** e13442. DOI: 10.7717/peerj.13442
63. Liu P, Wang S, Wang G, Zhao M, Du F, Li K, Wang L, Wu H, Chen J, Yang Y, Su G. **Macrophage-derived exosomal miR-4532 promotes endothelial cells injury by targeting SP1 and NF-kappaB P65 signalling activation**. *J Cell Mol Med* (2022) **26** 5165-5180. DOI: 10.1111/jcmm.17541
64. Li J, Wang P, Xie Z, Wang S, Cen S, Li M, Liu W, Tang S, Ye G, Zheng G, Su H, Ma M, Wu X, Wu Y, Shen H. **TRAF4 positively regulates the osteogenic differentiation of mesenchymal stem cells by acting as an E3 ubiquitin ligase to degrade Smurf2**. *Cell Death Differ* (2019) **26** 2652-2666. DOI: 10.1038/s41418-019-0328-3
65. Xia B, Gao J, Li S, Huang L, Zhu L, Ma T, Zhao L, Yang Y, Luo K, Shi X, Mei L, Zhang H, Zheng Y, Lu L, Luo Z, Huang J. **Mechanical stimulation of Schwann cells promote peripheral nerve regeneration via extracellular vesicle-mediated transfer of microRNA 23b–3p**. *Theranostics* (2020) **10** 8974-8995. DOI: 10.7150/thno.44912
66. Guo S, Gu J, Ma J, Xu R, Wu Q, Meng L, Liu H, Li L, Xu Y. **GATA4-driven miR-206-3p signatures control orofacial bone development by regulating osteogenic and osteoclastic activity**. *Theranostics* (2021) **11** 8379-8395. DOI: 10.7150/thno.58052
67. Qu Q, Wang L, Bing W, Bi Y, Zhang C, Jing X, Liu L. **miRNA-126-3p carried by human umbilical cord mesenchymal stem cell enhances endothelial function through exosome-mediated mechanisms in vitro and attenuates vein graft neointimal formation in vivo**. *Stem Cell Res Ther* (2020) **11** 464. DOI: 10.1186/s13287-020-01978-z
68. Chen Y, Yang Q, Lv C, Chen Y, Zhao W, Li W, Chen H, Wang H, Sun W, Yuan H. **NLRP3 regulates alveolar bone loss in ligature-induced periodontitis by promoting osteoclastic differentiation**. *Cell Prolif* (2021) **54** e12973. DOI: 10.1111/cpr.12973
|
---
title: Metformin regulates chondrocyte senescence and proliferation through microRNA-34a/SIRT1
pathway in osteoarthritis
authors:
- Shiju Yan
- Wenjing Dong
- Zhirui Li
- Junqiang Wei
- Tao Han
- Junliang Wang
- Feng Lin
journal: Journal of Orthopaedic Surgery and Research
year: 2023
pmcid: PMC10012483
doi: 10.1186/s13018-023-03571-5
license: CC BY 4.0
---
# Metformin regulates chondrocyte senescence and proliferation through microRNA-34a/SIRT1 pathway in osteoarthritis
## Abstract
### Background
Osteoarthritis (OA) is the most common degenerative disease in joints among elderly patients. Senescence is deeply involved in the pathogenesis of osteoarthritis. Metformin is widely used as the first-line drug for Type 2 diabetes mellitus (T2DM), and has great potential for the treatment of other aging-related disorders, including OA. However, the role of metformin in OA is not fully elucidated. Therefore, our aim here was to investigate the effects of metformin on human chondrocytes.
### Methods
After metformin treatment, expression level of microRNA-34a and SIRT1 in chondrocyte were detected with quantitative real-time PCR and immunofluorescence staining. Then, microRNA-34a mimic and small interfering RNA (siRNA) against SIRT1 (siRNA-SIRT1) were transfected into chondrocyte. Senescence-associated β-galactosidase (SA-β-gal) staining was performed to assess chondrocyte senescence. Chondrocyte viability was illustrated with MTT and colony formation assays. Western blot was conducted to detect the expression of P16, IL-6, matrix metalloproteinase-13 (MMP-13), Collagen type II (COL2A1) and Aggrecan (ACAN).
### Results
We found that metformin treatment (1 mM) inhibited microRNA-34a while promoted SIRT1 expression in OA chondrocytes. Both miR-34a mimics and siRNA against SIRT1 inhibited SIRT1 expression in chondrocytes. SA-β-gal staining assay confirmed that metformin reduced SA-β-gal-positive rate of chondrocytes, while transfection with miR-34a mimics or siRNA-SIRT1 reversed it. MTT assay and colony formation assay showed that metformin accelerated chondrocyte proliferation, while miR-34a mimics or siRNA-SIRT1 weakened this effect. Furthermore, results from western blot demonstrated that metformin suppressed expression of senescence-associated protein P16, proinflammatory cytokine IL-6 and catabolic gene MMP-13 while elevated expression of anabolic proteins such as Collagen type II and Aggrecan, which could be attenuated by transfection with miR-34a mimics.
### Conclusion
Overall, our data suggest that metformin regulates chondrocyte senescence and proliferation through microRNA-34a/SIRT1 pathway, indicating it could be a novel strategy for OA treatment.
## Introduction
Osteoarthritis (OA) is a prevalent age-related joint disorder characterized by progressive cartilage destruction, synovitis, subchondral bone remodeling and osteophyte formation [1]. OA causes pain and joint dysfunctions in the affected patients, leading to massive burden on social health care system. Various factors, including age, obesity, trauma history, abnormal mechanical stimulation, microRNAs and genetic predisposition, are closely related with both onset and progression of OA [2, 3]. However, there are no effective disease-modifying treatments that could reverse or delay OA development due to specific mechanisms leading to OA have not been fully elucidated. And more often, surgical treatments are essential for most patients with late-stage OA to improve life quality.
At present, etiology and pathogenesis of OA are complex and still remain not completely understood. Chondrocyte, the only cell type present in articular cartilage, plays a vital role in maintaining the dynamic homeostasis of anabolism and catabolism of the cartilage extracellular matrix (ECM) [4]. It is well accepted that chondrocyte senescence has been noted as a major event contributing to age-related changes in cartilage homeostasis, integrity and physiological function [5]. However, the exact mechanism linking senescence with OA pathogenesis remains to be further investigated and a better understanding of mechanisms of senescence underlying OA could provide new therapeutic targets for OA prevention and treatment.
Metformin has been widely used as the first-line medication for Type 2 diabetes mellitus (T2DM) by activating adenosine monophosphate-activated protein kinase (AMPK) [6], which lay a solid foundation for metformin to regulate metabolic and cellular processes, such as senescence, inflammation, oxidative stress and apoptosis [7, 8]. Evidence suggests that metformin has high safety profile and could be promising for a number of age-related diseases, such as degenerative musculoskeletal diseases, cardiovascular diseases and neurodegenerative diseases [9]. Recent study has been discussing the potential use of metformin in OA management. Metformin attenuates cartilage degradation and modulates pain-related behavior in a destabilization of the medial meniscus (DMM) model of OA in mice [10]. Latest research shows that metformin treatment decreases cartilage degradation, inhibits cartilage matrix catabolism and enhances cartilage matrix anabolism in both the early and late stages of OA [11]. Thus, metformin could be prospective in OA treatment.
MicroRNAs are a family of endogenous non-coding RNAs with 18 ~ 24 nucleotides in length [12]. MicroRNAs could regulate gene expression at post-translational level by binding to the 3′-untranslated region (3′-UTR) of target mRNAs [13]. Numerous studies have shown that microRNAs are involved with various cellular processes, such as senescence, cell proliferation, apoptosis and migration [3, 12, 14]. MicroRNA-34a (miR-34a) has been reported as a potential indicator of senescence, since its expression is increased in several aging tissue and cells, including chondrocytes [15–17]. Moreover, microRNA-34a is a target gene of metformin, which regulates multiple biological processes by regulating the expression of microRNA-34a [18, 19]. Therefore, exploring the roles of microRNAs and their potential target genes is critical for understanding the molecular mechanisms of OA pathogenesis.
Sirtuins (silent information regulator proteins) are a family of nicotinamide dinucleotide (NAD+)-dependent deacylases and highly conserved from bacteria to humans, which control a variety of cellular processes, including DNA repair, cell cycle, mitochondria homeostasis and cellular senescence [20, 21]. SIRT1 is the best studied sirtuins in bone and cartilage and is critical in maintaining cartilage health by promoting chondrocyte survival and ECM homeostasis [22]. Furthermore, evidence suggests that SIRT1 is one of the direct targets of microRNA-34a and microRNA-34a promotes chondrocyte apoptosis by regulating SIRT1, contributing to OA pathogenesis [17, 23]. Thus, this study is aimed to explore the effects of metformin in OA, specifically on chondrocyte senescence and proliferation, and investigate the molecular mechanisms of it by focusing on its target gene microRNA-34a and the downstream microRNA-34a/SIRT1 signaling pathway.
## Patient and cohort description
This study was approved by Ethics Committee of the Hainan Hospital of Chinese PLA General Hospital (approval number: S2022-10), and informed consent was obtained from all patients. OA articular cartilage samples were aseptically harvested from femoral condyles and tibial plateaus of 14 patients diagnosed with osteoarthritis with a Kellgren/Lawrence (K/L) grade of 3 or 4 who received total knee arthroplasty (mean age ± SD: 69.6 ± 3.1 years) at Hainan Hospital of Chinese PLA General Hospital from January 2020 to December 2020. Healthy cartilage samples were obtained from 12 trauma amputees with no history of OA or other musculoskeletal diseases (38.6 ± 6.7 years).
## Isolation and cell culture of primary chondrocytes
Cartilage samples were minced into small pieces and then digested with $0.25\%$ trypsin (Invitrogen, Carlsbad, USA) for 30 min and $0.2\%$ collagenase Type II (Millipore, Billerica, USA) for 10 h in a shaking water bath at 37 °C. Isolated chondrocytes were filtered through 100 μm nylon filters, washed twice with sterile PBS and seeded into culture flasks in DMEM (Gibco, NY, USA) medium supplemented with $10\%$ FBS (Hyclone, Thermo Scientific, USA), 100 U/mL penicillin and 100 μg/mL streptomycin (Gibco) at 37 °C in a humidified atmosphere with $5\%$ CO2. Primary chondrocytes and cells from passage 1 were used for this serial experiments.
## Cell treatments and transfection
miR-34a mimic, negative control (NC) and siRNA against SIRT1 (siRNA-SIRT1) were designed and synthesized by RiboBio (Guangzhou, China). At $80\%$ confluence, chondrocytes were treated with 1 mM metformin (Sigma-Aldrich, USA) for 48 h. After metformin treatment, chondrocytes were transfected with miR-34a mimics, siRNA against SIRT1 or negative control (NC), respectively, at working concentration of 100 nM for 48 h using Lipofectamine 2000 (Life technologies, USA) according to the manufacturer's protocol.
## Immunofluorescence staining
Metformin-treated chondrocytes were seeded into on 15 mm cell slides (Nest Biotechnology) in 6-well plates. Cells were fixed with $4\%$ paraformaldehyde (Beyotime, China) for 15 min and permeabilized with $0.1\%$ Trion X-100 in PBS for 20 min. Chondrocytes were blocked with $10\%$ BSA for 30 min and then incubated with anti-SIRT1 primary antibody (1:100, #8469, Cell Signaling Tech) overnight at 4 °C. Subsequently, cells were washed and incubated with Alexa Fluor 555-conjugated secondary antibody (1:500, A32727, Invitrogen) for 1 h at room temperature. Finally, nuclei were stained with DAPI in the dark, and fluorescence images were obtained using Fluorescence microscope (IX70, Olympus, Japan).
## RNA extraction and real-time quantitative RT-PCR
Total RNA was extracted from chondrocytes using TRIzol reagent (Invitrogen, CA) according to the manufacturer's instructions. Total RNA was reverse transcribed with PrimeScript RT Master Mix (TAKARA, Japan). Quantitative Real-Time PCR was performed using Power SYBR Green PCR Master Mix (Applied Biosystems, Foster city, CA, USA) according to manufacturer's protocol. Specific primers used for mRNAs were listed below: for SIRT1, 5′- TGGCAAAGGAGCAGATTAGTAGG -3′ (forward) and 5′- CTGCCACAAGAACTAGAGGATAAGA -3′ (reverse); for GAPDH, 5′-ATTCCACCCATGGCAAATTC-3′ (forward) and 5′-TGGGATTTCCATTGATGACAAG-3′ (reverse) (Table 1). GAPDH mRNA was used as a housekeeping gene for normalization with the comparative 2−ΔΔCt method. Table 1Primer sequenceGenePrimer sequence (5′-3′)SIRT1Forward:TGGCAAAGGAGCAGATTAGTAGGReverse:CTGCCACAAGAACTAGAGGATAAGAmicrorna-34aForward:CGTCACCTCTTAGGCTTGGAReverse:CATTGGTGTCGTTGTGCTGAPDHForward:ATTCCACCCATGGCAAATTCReverse:TGGGATTTCCATTGATGACAAGU6Forward:CTCGCTTCGGCAGCACATATACTReverse:ACGCTTCACGAATTTGCGTGTC For relative quantification of microRNA-34a, RNA was reverse transcribed using the TaqMan MicroRNA Reverse Transcription kit (Applied Biosystems, USA) according to the manufacturer’s protocol and Real-Time quantitative-PCR was performed with the TaqMan MicroRNA assays (Applied Biosystems, USA). The reactions were performed in 7500 Fast Real-Time System (Applied Biosystems, USA). Small nuclear RNA U6 was used as housekeeping gene for normalization (Table 1).
## SA-β-galactosidase staining
48 h after transfection, OA chondrocytes were fixed and stained with Senescence β-Galactosidase Staining Kit (#9860, Cell Signaling Technologies, USA) according to the manufacturer’s protocol. Senescent cells were identified by formation of blue precipitates at pH 6 and counted from three different fields of view under microscopy. Percentage of SA-β-gal-positive chondrocytes were calculated.
## MTT assay
Cell proliferation was evaluated with MTT assay. 24 h after transfection, chondrocytes were seeded into 96-well plates at a density of 5 × 103/well. 20 µl of MTT (5 mg/ml) (Sigma-Aldrich, USA) was added to each well every 24 h, followed by incubation at 37 °C for another 4 h. Culture medium was removed, and formazan precipitate was dissolved with 100 µl of dimethyl sulfoxide for 20 min at 37 °C. The optical density (OD) was determined by measuring the absorbance at 490 nm with a reference wavelength of 630 nm using a microplate reader (ThermoFisher, USA).
## Colony formation assay
Cell viability was evaluated with Colony formation assay. Transfected chondrocytes were seeded into 6-well plates (300/well) and cultured for 14 days in DMEM medium containing $10\%$ FBS. Colonies were fixed with methanol for 20 min and stained with crystal violet (KeyGen, China) for 20 min at room temperature. Cell colony was observed under microscope (Olympus, Japan) and counted.
## Western blot analysis
48 h after transfection, chondrocytes were rinsed with PBS and treated with RIPA lysis buffer (Beyotime, China) containing enzyme inhibitor cocktail (Roche, Switzerland) on ice for 30 min. Protein concentrations were quantified using BCA protein assay kit (Beyotime, China). 25 μg of total protein for each sample was loaded and separated by $10\%$ sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) and transferred to PVDF membranes (Millipore, USA). Membranes were blocked with $5\%$ skim milk and then incubated with primary antibodies against P16 (1:1000, ab51243, Abcam), ACAN (1:500, ab3778, Abcam), MMP13 (1:1000, #69,926, CST), IL-6 (1:1000, 66,146-1-Ig, Proteintech), COL2A1 (1:1000, ab34712, Abcam), SIRT1 (1:1000, #8469, CST), GAPDH (1:1000, ab9485, Abcam) overnight at 4 °C. Then, second antibodies were incubated for 2 h at room temperature after membranes were washed with TBST. Signals obtained with the ECL luminescent reagent (Millipore, USA) were detected and analyzed using ChemiDoc Imaging System (Bio-Rad, USA) with Quantity One analyzing system.
## Statistical analysis
Continuous variables were presented as mean ± standard deviation (SD). Comparisons between two groups were analyzed by unpaired Student's t test or Mann–Whitney U test. One-way analysis of variance (ANOVA) followed by post hoc LSD test was used for difference statistical analysis between multiple groups. All testing was performed using SPSS26.0 (IBM, USA). A difference with $P \leq 0.05$ was considered to be significant.
## Metformin regulated microRNA-34a and SIRT1 expression in chondrocytes
To investigate the mechanism and effects of metformin on OA, OA chondrocytes were treated with 1 mM metformin. As revealed in Fig. 1, metformin treatment significantly inhibited expression level of microRNA-34a and promoted mRNA level of SIRT1 in OA chondrocytes ($P \leq 0.01$).Fig. 1Metformin regulated microRNA-34a and SIRT1 expression in chondrocytes. OA chondrocytes were treated with 1 mM metformin. Metformin downregulated expression level of microRNA-34a (a) and promoted mRNA level of SIRT1 (b) in OA chondrocytes. ( c)–(h) immunofluorescence staining for SIRT1. ( c) SIRT1, (d) nuclei stained with DAPI and (e) merged field for chondrocytes in control group. ( f) SIRT1, (g) nuclei stained with DAPI and (h) merged field for metformin-treated chondrocytes in Met group. All experiments were repeated three times. ** $P \leq 0.01$ As consistently shown in immunofluorescence staining (Fig. 1c–h), SIRT1 mainly located in the nuclei and its expression level was promoted ($P \leq 0.05$) by metformin treatment. Collectively, these results suggested microRNA-34a and SIRT1 were regulated by metformin in OA chondrocytes and could be related with biofunction of metformin.
## MicroRNA-34a regulated SIRT1 expression in chondrocytes
Chondrocytes were transfected with miR-34a mimics, siRNA against SIRT1 or negative control (NC), respectively. Western blot assay was performed to evaluate SIRT1 expression. Results demonstrated that both miR-34a mimics and siRNA against SIRT1 inhibited SIRT1 in chondrocytes, as revealed in Fig. 2.Fig. 2MicroRNA-34a regulated SIRT1 expression in chondrocytes. Chondrocytes were transfected with miR-34a mimics, siRNA against SIRT1 (si-SIRT1) or negative control (NC) at a working concentration of 100 nM, respectively. ( a)–(b) miR-34a mimics and siRNA-SIRT1 inhibited SIRT1 expression in chondrocytes. All experiments were repeated three times. ** $P \leq 0.01$
## Metformin attenuated chondrocytes senescence
We performed SA-β-galactosidase staining to evaluate the effects of metformin on chondrocyte senescence. Compared with NC group (Fig. 3a), metformin treatment significantly decreased percentage of SA-β-gal-positive OA chondrocytes in Metformin group (Fig. 3b). Importantly, this decrease was reversed by transfection of microRNA-34a mimics in Met + miR-34a group (Fig. 3c) or siRNA against SIRT1 in Met + si-SIRT1 group (Fig. 3d). These results indicated that metformin regulated chondrocytes senescence in vitro via microRNA-34a and SIRT1.Fig. 3Metformin attenuated OA chondrocytes senescence in vitro. Chondrocytes were divided into Metformin treatment group (Metformin), Metformin + miR-34a group (Met + miR-34a), Negative Control group (NC) and Metformin + siRNA-SIRT1 group (Met + si-SIRT1). After treated with 1 mM metformin for 48 h, chondrocytes were transfected with 100 nM miR-34a negative control (b), miR-34a mimics (c) or siRNA-SIRT1 (d) for 48 h, respectively. NC group (a) was treated with culture medium as blank control. Percentage of SA-β-gal-positive chondrocytes were calculated. ( e) metformin attenuated OA chondrocytes senescence and this effect was reversed by transfection of microRNA-34a mimics or siRNA against SIRT1. All experiments were repeated three times. * $P \leq 0.05.$ ** $P \leq 0.01$
## Metformin promoted cell proliferation and viability of chondrocytes
Next, MTT assay was performed and proliferation curve was recorded to explore the effects of metformin on proliferation of human chondrocytes. Figure 4 shows that chondrocytes incubated with metformin in Metformin group exhibited a significant increase in proliferation capacity compared with chondrocytes in the NC group. Moreover, when transfected with microRNA-34a mimics or siRNA against SIRT1 after metformin treatment, chondrocytes in Met + miR-34a group and in Met + si-SIRT1 group grew at a lower rate than chondrocytes in Metformin group. No statistical significance was observed in proliferation rate among NC group, Met + miR-34a group and Met + si-SIRT1 group (Fig. 4). Fig. 4Metformin promoted chondrocytes proliferation in vitro. After treated with 1 mM metformin for 48 h, chondrocytes were transfected with 100 nM miR-34a negative control, miR-34a mimics or siRNA-SIRT1 for 48 h, respectively. NC group was treated with culture medium as blank control. 24 h after transfection, chondrocytes were seeded into 96-well plates and cell number was evaluated as the absorbance at 490 nm with a reference wavelength of 630 nm. All experiments were repeated three times. ** $P \leq 0.01$ To further test the effect of metformin on chondrocyte viability, we performed colony formation assay. Chondrocytes treated with metformin in Metformin group (Fig. 5b) formed increased cell colonies compared with NC group (Fig. 5a), while transfection of microRNA-34a mimic or siRNA against SIRT1 following metformin treatment inhibited cell colony formation in Met + miR-34a group (Fig. 5c) or Met + si-SIRT1 group (Fig. 5d).Fig. 5Metformin increased chondrocytes colony formation in vitro. After treated with 1 mM metformin for 48 h, chondrocytes were transfected with 100 nM miR-34a negative control (b), miR-34a mimics (c) or siRNA-SIRT1 (d) for 48 h, respectively. NC group (a) was treated with culture medium as blank control. Cell colony was observed and counted. ( e) Chondrocytes treated with metformin in Metformin group formed increased cell colonies compared with NC group, while transfection of miR-34a mimic or siRNA for SIRT1 following metformin treatment inhibited cell colony formation in Met + miR-34a group and Met + si-SIRT1 group. All experiments were repeated three times. * $P \leq 0.05.$ ** $P \leq 0.01$
## Metformin ameliorated chondrocytes senescence and OA through microRNA-34a/SIRT1 pathway
Western blot was performed to investigate the mechanism of metformin on senescence and metabolism homeostasis of chondrocytes. Results from western blot assay demonstrated that metformin treatment significantly inhibited P16 expression in Met group compared with NC group. Furthermore, metformin promoted expression level of COL2A1 (collagen II) and ACAN (aggrecan) while MMP-13, which regulates cartilage matrix degradation, and IL-6 were downregulated, suggesting that metformin cast a protective effect on chondrocytes against OA. On the contrary, microRNA-34a mimic transfection after metformin treatment counteracted this favorable effect by increasing expression level of P16, IL-6 and MMP-13 while decreasing expression level of COL2A1 and ACAN (Fig. 6). Fig. 6Metformin ameliorated chondrocytes senescence and OA through microRNA-34a/SIRT1 pathway. After treated with 1 mM metformin for 48 h, chondrocytes were transfected with 100 nM miR-34a negative control and miR-34a mimics for 48 h, respectively. NC group was treated with culture medium as blank control. Metformin treatment significantly inhibited P16 (b) expression in Metformin group compared with NC group. Furthermore, metformin promoted expression level of Collagen II (f) and Aggrecan (c) while downregulated expression level of MMP-13 (e) and IL-6 (d). Transfection of microRNA-34a mimic after metformin treatment reversed this favorable effect by increasing expression level of P16, IL-6 and MMP-13 while decreasing expression level of Collagen II and Aggrecan (a). All experiments were repeated three times. * $P \leq 0.05.$ ** $P \leq 0.01$
## Discussion
Although much progress has been made in the pathogenesis research and clinical treatment of OA, OA is still a major threaten to human health for elderly patients worldwide due to disability and pain of joints [24]. Over the past few years, metformin has received growing attention for its protective effects against OA and other age-related disease besides its hypoglycemic role. Metformin is reported to suppress IL-1β-induced oxidative and OA-like inflammatory changes by enhancing the SIRT3/PINK1/Parkin signaling pathway [25]. Administration of metformin gives a beneficial effect on long-term knee joint outcomes in obese OA patients by a reduced rate of medial cartilage volume loss and a reduction in risk of total knee replacement [26]. Therefore, revealing the roles of metformin and potential downstream genes of it is essential for elucidating the molecular mechanisms of OA and identifying new therapeutic targets for it.
Accumulating evidences show that microRNA-34a could be regulated by metformin [18, 19]. In this present study, the connection between metformin and microRNA-34a in OA was investigated and quantitative RT-PCR results showed that metformin treatment inhibited expression level of microRNA-34a in OA chondrocytes. However, conflicting results related with metformin and microRNA-34a have been reported. Several studies reported metformin could elevate microRNA-34a expression [27, 28], while more researchers demonstrated metformin could inhibit microRNA-34a expression [18, 19, 29], indicating that the interplay between metformin and microRNA-34a is more complicated than we thought and more efforts are needed to investigate the mechanisms of metformin on microRNA-34a. On the other hand, we found metformin incubation elevated SIRT1 expression in OA chondrocytes, which was also demonstrated by immunofluorescence staining.
It is known that microRNAs tend to exert their roles by regulating translation or stability of mRNAs of downstream target gene [13]. Specifically, microRNA-34a has been reported to directly target SIRT1 [30], which coincided with what we observed in this study that transfection of microRNA -34a mimic could inhibit SIRT1 expression in chondrocytes. Moreover, the siRNA we ordered could effectively inhibit SIRT1 expression, which is necessary for the following experiments. Thus, we hypothesized microRNA-34a and its well-studied target, the longevity-associated protein SIRT1, were involved in the biological function of metformin in OA.
Since aging is crucial to OA, any drug targeting cell senescence and associated chronic low-grade inflammation and metabolic disorder could be utilized to slow down OA progression. Metformin is thought to be an “anti-aging” drug, based on in vitro and animal experiments and numerous retrospective analyses on beneficial outcomes for type 2 diabetics and is being tested in several clinical trials evaluating the impact of metformin on aging [31–33]. Previous studies document microRNA-34a as an important regulator of age-dependent tissue changes and a cell senescence inducer in many diseases [34]. Meanwhile, SIRT1 has been reported to counteract multiple aging-associated diseases on account of its capacity to modulate a variety of cellular processes such as DNA repair, apoptosis, mitochondrial biogenesis and cell stress responses [35, 36]. Therefore, we assume that metformin ameliorates OA by regulating chondrocytes senescence through microRNA-34a/SIRT1 pathway. Firstly, we examined the effect of metformin on senescence in OA chondrocytes and results from SA-β-galactosidase staining suggest that metformin treatment attenuated chondrocytes senescence and overexpression of microRNA-34a or silencing SIRT1 by siRNA could reverse this effect. At the same time, we also detected that overexpression of microRNA-34a or silencing SIRT1 could counteract the stimulative effect of metformin on chondrocyte proliferation and cell clone formation. All these suggest metformin provided a protective effect to chondrocytes, while microRNA-34a/SIRT1 pathway is deeply involved.
Cellular senescence is a permanent state of cell cycle arrest and plays a significant role in the pathology of OA as OA chondrocytes shows elevated senescence-associated- β-galactosidase (SA-β-Gal) activity and accumulation of cyclin-dependent kinase (CDK) inhibitors P16, P21 [37]. Senescent chondrocytes exhibit inhibited proliferation capacity and loss of original phenotype, leading to imbalance of ECM metabolism [38]. In addition, OA chondrocytes undergo chronic and dynamic transformations into senescence-associated secretory phenotypes (SASP), characterized by increased secretion of proinflammatory cytokines, such as IL-1β, IL-6, TNF-α, and matrix metalloproteinases (MMP), such as MMP-13 [39]. In this study, we investigated the mechanisms of metformin on chondrocytes senescence. Western blot results demonstrated that metformin inhibited expression of P16, IL-6 and MMP-13 while elevated Collagen II and Aggrecan expression, indicating metformin ameliorated chondrocytes senescence through regulating P16 and attenuated OA by inhibiting SASP production and promoting extracellular matrix synthesis. Moreover, the restoration of microRNA-34a markedly counteracted this chondroprotective effect of metformin. Taken together, our study demonstrated that metformin inhibited chondrocyte senescence and promoted chondrocyte proliferation through microRNA-34a/SIRT1 pathway in osteoarthritis.
In summary, we suggest that metformin downregulated microRNA-34a and elevated SIRT1 in OA chondrocytes. Metformin ameliorated OA by inhibiting chondrocyte senescence and SASP secretion, as well as promoting chondrocyte proliferation and extracellular matrix synthesis via microRNA-34a/SIRT1 axis. These findings give us a new perspective of relationship between metformin and OA and could make metformin a novel choice for OA therapeutic strategy.
## References
1. Glyn-Jones S, Palmer AJ, Agricola R, Price AJ, Vincent TL, Weinans H, Carr AJ. **Osteoarthritis**. *Lancet* (2015) **386** 376-387. DOI: 10.1016/S0140-6736(14)60802-3
2. Abramson SB, Attur M. **Developments in the scientific understanding of osteoarthritis**. *Arthritis Res Ther* (2009) **11** 227. DOI: 10.1186/ar2655
3. Oliviero A, Della PG, Peretti GM, Maffulli N. **MicroRNA in osteoarthritis: physiopathology, diagnosis and therapeutic challenge**. *Br Med Bull* (2019) **130** 137-147. DOI: 10.1093/bmb/ldz015
4. Goldring MB. **The role of the chondrocyte in osteoarthritis**. *Arthritis Rheum* (2000) **43** 1916-1926. DOI: 10.1002/1529-0131(200009)43:9<1916::AID-ANR2>3.0.CO;2-I
5. Aigner T, Haag J, Martin J, Buckwalter J. **Osteoarthritis: aging of matrix and cells–going for a remedy**. *Curr Drug Targets* (2007) **8** 325-331. DOI: 10.2174/138945007779940070
6. Hawley SA, Gadalla AE, Olsen GS, Hardie DG. **The antidiabetic drug metformin activates the AMP-activated protein kinase cascade via an adenine nucleotide-independent mechanism**. *Diabetes* (2002) **51** 2420-2425. DOI: 10.2337/diabetes.51.8.2420
7. Teodoro JS, Nunes S, Rolo AP, Reis F, Palmeira CM. **Therapeutic options targeting oxidative stress, mitochondrial dysfunction and inflammation to hinder the progression of vascular complications of diabetes**. *Front Physiol* (2018) **9** 1857. DOI: 10.3389/fphys.2018.01857
8. Chen D, Xia D, Pan Z, Xu D, Zhou Y, Wu Y, Cai N, Tang Q, Wang C, Yan M. **Metformin protects against apoptosis and senescence in nucleus pulposus cells and ameliorates disc degeneration in vivo**. *Cell Death Dis* (2016) **7** e2441. DOI: 10.1038/cddis.2016.334
9. Campisi J, Kapahi P, Lithgow GJ, Melov S, Newman JC, Verdin E. **From discoveries in ageing research to therapeutics for healthy ageing**. *Nature* (2019) **571** 183-192. DOI: 10.1038/s41586-019-1365-2
10. Li H, Ding X, Terkeltaub R, Lin H, Zhang Y, Zhou B, He K, Li K, Liu Z, Wei J. **Exploration of metformin as novel therapy for osteoarthritis: preventing cartilage degeneration and reducing pain behavior**. *Arthritis Res Ther* (2020) **22** 34. DOI: 10.1186/s13075-020-2129-y
11. Yan J, Ding D, Feng G, Yang Y, Zhou Y, Ma L, Guo H, Lu Z, Jin Q. **Metformin reduces chondrocyte pyroptosis in an osteoarthritis mouse model by inhibiting NLRP3 inflammasome activation**. *Exp Ther Med* (2022) **23** 222. DOI: 10.3892/etm.2022.11146
12. Bartel DP. **MicroRNAs: genomics, biogenesis, mechanism, and function**. *Cell* (2004) **116** 281-297. DOI: 10.1016/S0092-8674(04)00045-5
13. Ambros V. **The functions of animal microRNAs**. *Nature* (2004) **431** 350-355. DOI: 10.1038/nature02871
14. Giordano L, Porta GD, Peretti GM, Maffulli N. **Therapeutic potential of microRNA in tendon injuries**. *Br Med Bull* (2020) **133** 79-94. DOI: 10.1093/bmb/ldaa002
15. Huang Q, Ou Y, Xiong H, Yang H, Zhang Z, Chen S, Ye Y, Zheng Y. **The miR-34a/Bcl-2 pathway contributes to auditory cortex neuron apoptosis in age-related hearing loss**. *Audiol Neurootol* (2017) **22** 96-103. DOI: 10.1159/000454874
16. Banerjee J, Roy S, Dhas Y, Mishra N. **Senescence-associated miR-34a and miR-126 in middle-aged Indians with type 2 diabetes**. *Clin Exp Med* (2020) **20** 149-158. DOI: 10.1007/s10238-019-00593-4
17. Yan S, Wang M, Zhao J, Zhang H, Zhou C, Jin L, Zhang Y, Qiu X, Ma B, Fan Q. **MicroRNA-34a affects chondrocyte apoptosis and proliferation by targeting the SIRT1/p53 signaling pathway during the pathogenesis of osteoarthritis**. *Int J Mol Med* (2016) **38** 201-209. DOI: 10.3892/ijmm.2016.2618
18. Li W, Jin S, Hao J, Shi Y, Li W, Jiang L. **Metformin attenuates ischemia/reperfusion-induced apoptosis of cardiac cells by downregulation of p53/microRNA-34a via activation of SIRT1**. *Can J Physiol Pharmacol* (2021) **99** 875-884. DOI: 10.1139/cjpp-2020-0180
19. Feng L, Liu T, Yang Y, Xiao W, Shi J, Mei X, Tian S, Liu X, Huang H, Bai Y. **Metformin promotes proliferation and suppresses apoptosis in Ox-LDL stimulated macrophages by regulating the miR-34a/Bcl2 axis**. *Rsc Adv* (2019) **9** 14670-14676. DOI: 10.1039/C9RA00705A
20. Haigis MC, Sinclair DA. **Mammalian sirtuins: biological insights and disease relevance**. *Annu Rev Pathol* (2010) **5** 253-295. DOI: 10.1146/annurev.pathol.4.110807.092250
21. van de Ven R, Santos D, Haigis MC. **Mitochondrial sirtuins and molecular mechanisms of aging**. *Trends Mol Med* (2017) **23** 320-331. DOI: 10.1016/j.molmed.2017.02.005
22. Matsuzaki T, Matsushita T, Takayama K, Matsumoto T, Nishida K, Kuroda R, Kurosaka M. **Disruption of Sirt1 in chondrocytes causes accelerated progression of osteoarthritis under mechanical stress and during ageing in mice**. *Ann Rheum Dis* (2014) **73** 1397-1404. DOI: 10.1136/annrheumdis-2012-202620
23. Ni S, Xu C, Zhuang C, Zhao G, Li C, Wang Y, Qin X. **LncRNA LUADT1 regulates miR-34a/SIRT1 to participate in chondrocyte apoptosis**. *J Cell Biochem* (2020) **122** 1003-1008. DOI: 10.1002/jcb.29637
24. Vina ER, Kwoh CK. **Epidemiology of osteoarthritis: literature update**. *Curr Opin Rheumatol* (2018) **30** 160-167. DOI: 10.1097/BOR.0000000000000479
25. Wang C, Yang Y, Zhang Y, Liu J, Yao Z, Zhang C. **Protective effects of metformin against osteoarthritis through upregulation of SIRT3-mediated PINK1/Parkin-dependent mitophagy in primary chondrocytes**. *Biosci Trends* (2019) **12** 605-612. DOI: 10.5582/bst.2018.01263
26. Wang Y, Hussain SM, Wluka AE, Lim YZ, Abram F, Pelletier JP, Martel-Pelletier J, Cicuttini FM. **Association between metformin use and disease progression in obese people with knee osteoarthritis: data from the Osteoarthritis Initiative-a prospective cohort study**. *Arthritis Res Ther* (2019) **21** 127. DOI: 10.1186/s13075-019-1915-x
27. Noren HN, Martin-Montalvo A, Dluzen DF, Zhang Y, Bernier M, Zonderman AB, Becker KG, Gorospe M, de Cabo R, Evans MK. **Metformin-mediated increase in DICER1 regulates microRNA expression and cellular senescence**. *Aging Cell* (2016) **15** 572-581. DOI: 10.1111/acel.12469
28. Zhang Y, Chen R, Deng L, Shuai Z, Chen M. **The effect of metformin on the proliferation, apoptosis and CD133 mRNA expression of colon cancer stem cells by upregulation of miR 342–3p**. *Drug Des Devel Ther* (2021) **15** 4633-4647. DOI: 10.2147/DDDT.S336490
29. Wu C, Qin N, Ren H, Yang M, Liu S, Wang Q. **Metformin regulating miR-34a pathway to inhibit Egr1 in rat mesangial cells cultured with high glucose**. *Int J Endocrinol* (2018) **2018** 6462793. DOI: 10.1155/2018/6462793
30. Yamakuchi M, Ferlito M, Lowenstein CJ. **miR-34a repression of SIRT1 regulates apoptosis**. *Proc Natl Acad Sci USA* (2008) **105** 13421-13426. DOI: 10.1073/pnas.0801613105
31. Kulkarni AS, Gubbi S, Barzilai N. **Benefits of metformin in attenuating the hallmarks of aging**. *Cell Metab* (2020) **32** 15-30. DOI: 10.1016/j.cmet.2020.04.001
32. Glossmann HH, Lutz O. **Metformin and aging: a review**. *Gerontology* (2019) **65** 581-590. DOI: 10.1159/000502257
33. Zajda A, Huttunen KM, Sikora J, Podsiedlik M, Markowicz-Piasecka M. **Is metformin a geroprotector? A peek into the current clinical and experimental data**. *Mech Ageing Dev* (2020) **191** 111350. DOI: 10.1016/j.mad.2020.111350
34. Raucci A, Macri F, Castiglione S, Badi I, Vinci MC, Zuccolo E. **MicroRNA-34a: the bad guy in age-related vascular diseases**. *Cell Mol Life Sci* (2021) **78** 7355-7378. DOI: 10.1007/s00018-021-03979-4
35. Deng Z, Li Y, Liu H, Xiao S, Li L, Tian J, Cheng C, Zhang G, Zhang F. *Biosci Rep* (2019). DOI: 10.1042/BSR20190189
36. Almeida M, Porter RM. **Sirtuins and FoxOs in osteoporosis and osteoarthritis**. *Bone* (2019) **121** 284-292. DOI: 10.1016/j.bone.2019.01.018
37. Loeser RF. **Aging and osteoarthritis: the role of chondrocyte senescence and aging changes in the cartilage matrix**. *Osteoarthritis Cartilage* (2009) **17** 971-979. DOI: 10.1016/j.joca.2009.03.002
38. Cao X, Luo P, Huang J, Liang C, He J, Wang Z, Shan D, Peng C, Wu S. **Intraarticular senescent chondrocytes impair the cartilage regeneration capacity of mesenchymal stem cells**. *Stem Cell Res Ther* (2019) **10** 86. DOI: 10.1186/s13287-019-1193-1
39. Coryell PR, Diekman BO, Loeser RF. **Mechanisms and therapeutic implications of cellular senescence in osteoarthritis**. *Nat Rev Rheumatol* (2021) **17** 47-57. DOI: 10.1038/s41584-020-00533-7
|
---
title: Mycobacterium tuberculosis lineage 4 associated with cavitations and treatment
failure
authors:
- Anabel Ordaz-Vázquez
- Pedro Torres-González
- Leticia Ferreyra-Reyes
- Sergio Canizales-Quintero
- Guadalupe Delgado-Sánchez
- Lourdes García-García
- Alfredo Ponce-De-León
- José Sifuentes-Osornio
- Miriam Bobadilla-Del-Valle
journal: BMC Infectious Diseases
year: 2023
pmcid: PMC10012486
doi: 10.1186/s12879-023-08055-9
license: CC BY 4.0
---
# Mycobacterium tuberculosis lineage 4 associated with cavitations and treatment failure
## Abstract
### Background
Mycobacterium tuberculosis genotyping has been crucial to determining the distribution and impact of different families on disease clinical presentation. The aim of the study was to evaluate the associations among sociodemographic and clinical characteristics and M. tuberculosis lineages from patients with pulmonary tuberculosis in Orizaba, Veracruz, Mexico.
### Methods
We analyzed data from 755 patients whose isolates were typified by 24-loci mycobacterial interspersed repetitive unit–variable number of tandem repeats (MIRU–VNTR). The associations among patient characteristics and sublineages found were evaluated using logistic regression analysis.
### Results
Among M. tuberculosis isolates, $\frac{730}{755}$ ($96.6\%$) were assigned to eight sublineages of lineage 4 (Euro-American). Alcohol consumption (adjusted odds ratio [aOR] 1.528, $95\%$ confidence interval (CI) 1.041–2.243; $$p \leq 0.030$$), diabetes mellitus type 2 (aOR 1.625, $95\%$ CI 1.130–2.337; $$p \leq 0.009$$), sputum smear positivity grade (3+) (aOR 2.198, $95\%$ CI 1.524–3.168; $p \leq 0.001$) and LAM sublineage isolates (aOR 1.023, $95\%$ CI 1.023–2.333; $$p \leq 0.039$$) were associated with the presence of cavitations. Resistance to at least one drug (aOR 25.763, $95\%$ CI 7.096–93.543; $p \leq 0.001$) and having isolates other than Haarlem and LAM sublineages (aOR 6.740, $95\%$ CI 1.704–26.661; $$p \leq 0.007$$) were associated with treatment failure. In a second model, multidrug resistance was associated with treatment failure (aOR 31.497, $95\%$ CI 5.119–193.815; $p \leq 0.001$). Having more than 6 years of formal education was not associated with treatment failure.
### Conclusions
Knowing M. tuberculosis genetic diversity plays an essential role in disease development and outcomes, and could have important implications for guiding treatment and improving tuberculosis control.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12879-023-08055-9.
## Background
The agent responsible for tuberculosis belongs to *Mycobacterium tuberculosis* complex (MTBC). Pulmonary tuberculosis is the most common disease presentation, reported in 4.8 million cases worldwide [1]. The state of Veracruz in southern Mexico reports the highest number of cases [2198] nationwide [2]. The incidence in the municipality of Orizaba, Veracruz was 16–38 cases/100,000 inhabitants during the period 1995–2010 surpassing the national incidence [3].
Nowadays, nine M. tuberculosis lineages have been identified strongly associated with particular geographic regions [4, 5]. In the Americas, tuberculosis is mainly caused by lineage 4 also known as Euro-American lineage [6].
Mycobacterium tuberculosis genotyping is important because it contributes to knowledge regarding its genetic diversity [7, 8]. The current gold standard for genotyping is mycobacterial interspersed repetitive unit–variable number of tandem repeat (MIRU–VNTR) method. Currently, MIRUs are used as markers for strains classification and sub-classification. For example, within the Latin American & Mediterranean (LAM) family, a single repeat of MIRU40 has been proposed as a marker of RDRio sub-lineage [9].
Risk factors related to M. tuberculosis genetics help in the early identification of patients infected with lineages associated with increased risk of treatment failure, relapse, drug resistance and death [10]. External risk factors associated with active tuberculosis development are poverty, overpopulation, overcrowding and malnutrition, in addition to comorbidities such as human immunodeficiency virus (HIV) coinfection, diabetes mellitus type 2 (DM2), chronic kidney failure, silicosis, immunosuppressive therapies and addictions such as smoking and drinking [11, 12].
In addition to host and environmental risk factors, tuberculosis epidemiology can also be influenced by M. tuberculosis genetic diversity [13]. Some lineages have shown differences in their virulence phenotypes, affecting transmissibility and pathogenesis and having implications in treatment outcomes and failure in the effectiveness of the BCG vaccine [6, 14].
The aim of this study was to evaluate the association among sociodemographic and clinical characteristics and M. tuberculosis lineages from isolates of patients with pulmonary tuberculosis obtained in a population-based study conducted in Orizaba, Veracruz, Mexico from 1995 to 2010.
## Study population and data collection
Between March 1995 and April 2010, pulmonary tuberculosis cases passive search was carried out in people over 15 years of age who had respiratory symptoms for more than two weeks in 12 health jurisdictions municipality of Orizaba, Veracruz, Mexico. During this period, 1132 patients were diagnosed and for this study 612 M. tuberculosis isolates were recovered from a strain collection and 143 more from a DNA collection using samples from these patients. We used the population-based cohort data from patients diagnosed with pulmonary tuberculosis from August 1, 1997, to April 30, 2010. The study was approved by the Ethics Committee (Ref. No. 1515). All participating patients signed informed consent forms.
As part of the cohort investigation, isolates were genotyped by 24-loci MIRU–VNTR and susceptibility tests were performed as previously described [15]. LAM RDRio and RD115 sublineages were classified according to the presence of a single repeat in MIRU40 and MIRU02 respectively.
## Definitions
The following sociodemographic variables were considered: sex, age, education level, dirt-floor home, and rural residence locality, nearest health center distance, social security access, and consumption of alcohol, tobacco and illicit drugs. DM2 and HIV diagnosis was also considered. Presence of acid-fast bacilli (AFB) in sputum samples information was considered and was graded as follows: 1 + (1–9 bacilli per 100 observed fields), 2 + (1–9 bacilli per 10 observed fields) or 3 + (1–9 bacilli per observed field). We included fever, hemoptysis and presence of cavitations variables, each used dichotomously. Body mass index (BMI) and number of days between symptom onset and start of treatment were calculated.
We used tuberculosis prevention and control program (NOM-SSA-006) operational definitions for treatment outcomes, except failure and death, which were defined according to international definitions [16, 17]: cure, patient who completed treatment, with signs and symptoms disappearance, or patient who had smear or culture negative at the end of treatment; failure, patient who had smear or culture positive after five months or later during treatment; and treatment completion, patient who completed his/her treatment regimen with signs and symptoms disappearance and smear or culture were not performed. Patients who did not complete treatment were classified into the following two categories: abandon, patient who interrupts treatment for 30 days or more; and death during treatment, patient who died due to any other cause during treatment.
Lineage variable was operationalized in disaggregated and aggregate way according to MIRU–VNTR genotyping. Disaggregated variable considers each identified sublineage, Haarlem, LAM, Cameroon, UgandaI, Ghana, S, X, TUR, EAI, Beijing and unknown. Aggregate variable considers lineage frequency, Haarlem, LAM and lineages other than Haarlem and LAM, because of the small frequency of each other lineages.
## Statistical analysis
We calculated the distributions percentage for qualitative variables as well as medians and interquartile ranges (IQR) for quantitative variables. We used Pearson chi-square test for dichotomous variables, binomial test for categorical variables and Kruskal–Wallis test for quantitative variables. Unconditional logistic regression models were adjusted to explain treatment failure and the presence of cavitation on radiography. Two models were adjusted to explain treatment failure: one included resistance variable to at least one drug, and the other included MDR. To include variables in a multivariate model were considered those that in the bivariate analysis had values of p ≤ 0.20 and biological plausibility. We estimated adjusted odds ratio (aOR) and $95\%$ confidence intervals (CIs).
Analyses were performed using STATA® v15 statistical software package (StataCorp LP, College Station, TX, USA).
## Results
The characteristics of the studied patients are shown in Table 1. The proportion of individuals among the population examined was similar to the proportion represented by this same group. A total of 755 patients were included in the study, 442 ($59\%$) of whom were men, with a median age of 45 years (IQR 32–59). There were 507 ($67\%$) patients with more than six years of formal education, and 174 ($23\%$) lived in dirt-floor homes. Comorbidity with DM2 was reported in 250 ($33\%$) patients. HIV status was known for 739 patients, of whom 13 ($2\%$) were positive. Resistance to any drug was present in $\frac{116}{612}$ ($19\%$) isolates, and 20 ($3\%$) were MDR. The most common clinical findings were fever and cavitation in $\frac{531}{752}$ ($71\%$) and $\frac{282}{626}$ ($45\%$) patients respectively. Cure was recorded in $\frac{532}{755}$ ($70\%$) patients. Table 1Demographic and clinical characteristics of the patients with pulmonary tuberculosis in Orizaba, Veracruz, 1995–2010 ($$n = 1132$$)VariablesTotal casesSample percentagen/totala (%)n/totala (%)Male$\frac{654}{1132}$ (58.0)$\frac{442}{755}$ (59.0)Age (years) (median [IQR])47 [32–60]45 [32–59] > 6 years of formal schooling$\frac{789}{1131}$ (70.0)$\frac{507}{754}$ (67.0)House with dirt floor$\frac{218}{1132}$ (19.0)$\frac{174}{755}$ (23.0)Rural residence$\frac{134}{1002}$ (13.0)$\frac{99}{732}$ (14.0)Nearest health center distance (meters) (median [IQR])698 [412–1073]708 [414–1099]Access to social security$\frac{400}{1132}$ (35.0)$\frac{254}{755}$ (34.0)Used alcohol$\frac{468}{1130}$ (41.0)$\frac{330}{753}$ (44.0)Used tobacco$\frac{222}{1129}$ (20.0)$\frac{164}{753}$ (22.0)Used illegal drug$\frac{50}{1131}$ (4.0)$\frac{35}{754}$ (5.0)Homelessness or residence in shelters$\frac{33}{1129}$ (3.0)$\frac{20}{754}$ (3.0)*Diabetes mellitus* type $\frac{2386}{1132}$ (34.0)$\frac{250}{755}$ (33.0)HIV coinfection$\frac{19}{1095}$ (2.0)$\frac{13}{739}$ (2.0)New case$\frac{934}{1131}$ (83.0)$\frac{688}{754}$ (91.0)AFB sputum positivity grade of 3+ $\frac{281}{1132}$ (24.8)$\frac{211}{755}$ (28.0)Resistance to at least one drug$\frac{176}{826}$ (21.0)$\frac{116}{612}$ (19.0)MDR$\frac{47}{826}$ (6.0)$\frac{20}{612}$ (3.0)Fever$\frac{744}{1129}$ (66.0)$\frac{531}{752}$ (71.0)Haemoptysis$\frac{352}{1126}$ (31.0)$\frac{250}{753}$ (33.0)Cavitations presence on chest radiograph$\frac{398}{927}$ (43.0)$\frac{282}{626}$ (45.0)BMI (median [IQR])21.2 [18.6–24.0]20.9 [18.4–23.8]Number of days between symptom onset and start of treatment (median [IQR])104 [63–186]105 [67–182]Treatment outcomeAbandon$\frac{86}{1132}$ (8.0)$\frac{49}{755}$ (6.0)Cure$\frac{756}{1132}$ (67.0)$\frac{532}{755}$ (70.0)Treatment completion$\frac{136}{1132}$ (12.0)$\frac{93}{755}$ (12.0)Failure$\frac{26}{1132}$ (2.0)$\frac{20}{755}$ (3.0)Death during treatment$\frac{49}{1132}$ (4.0)$\frac{26}{755}$ (3.0)No data$\frac{79}{1132}$ (7.0)$\frac{35}{755}$ (5.0)Treatment failureFailure$\frac{26}{918}$ (2.8)$\frac{20}{645}$ (3.1)Cure or treatment completion$\frac{892}{918}$ (97.2)$\frac{625}{645}$ (96.9)No treatment successAbandon, failure or death during treatment$\frac{161}{1053}$ (15.3)$\frac{95}{720}$ (13.2)Cure or treatment completion$\frac{892}{1053}$ (84.7)$\frac{625}{720}$ (86.8)Death during treatmentDeath$\frac{49}{1053}$ (4.7)$\frac{26}{720}$ (3.6)Cure or treatment completion, abandon or failure$\frac{1004}{1053}$ (95.4)$\frac{794}{720}$ (96.4)IQR, interquartile range; HIV, human immunodeficiency virus; AFB, acid fast bacilli; MDR, multidrug resistance; BMI, body mass indexaBecause there were missing values for the characteristics of some of the tuberculosis patients, several of the numbers below do not sum to the group total We identified ten sublineages among the 755 M. tuberculosis isolates (one from each patient) genotyped by 24-loci MIRU–VNTR. The most frequent sublineages were Haarlem (419, $55.5\%$) and LAM (163, $21.6\%$) which was sub classified in RDRio ($\frac{114}{163}$, $69.9\%$) characterized by one repeat in MIRU40 and RD115 ($\frac{31}{163}$, $19\%$) characterized by one repeat in MIRU02, followed by Cameroon (49, $6.5\%$), Uganda I (28, $3.7\%$), Ghana (23, $3\%$), S (18, $2.4\%$), X (15, $2\%$), TUR (15, $2\%$), EAI (15, $2\%$) and Beijing (2, $0.2\%$). It was not possible to determine lineage in eight isolates ($1.1\%$) and therefore, these were considered unknown, the distribution of M. tuberculosis isolates in the jurisdiction of Orizaba, *Veracruz is* shown in Fig. 1.Fig. 1Map of the distribution of M. tuberculosis isolates in the jurisdiction of Orizaba, Mexico *Data analysis* revealed that Haarlem sublineage had the highest proportion ($\frac{318}{419}$, $75.9\%$) of clustered patients compared with other sublineages. Patients with Cameroon sublineage isolates had more days between symptom onset and start of treatment (median 129, IQR 83–198), and patients isolates with Ghana sublineage presented hemoptysis ($\frac{11}{23}$, $47.8\%$) with greater frequency compared with the other sublineages. EAI sublineage isolates were more frequent in men ($\frac{12}{15}$, $80.0\%$), in DM2 patients ($\frac{10}{15}$, $66.7\%$) and in patients who had ever smoked ($\frac{7}{15}$, $46.7\%$). Patients with Beijing sublineage isolates were older (median 59 years, IQR 57–60) than patients with other sublineages; one of the two patients had HIV ($\frac{1}{2}$, $50\%$), the other had DM2 besides the isolate was MDR ($\frac{1}{2}$, $50\%$), both showed higher BMI (median 25.4, IQR 21.6–29.3) compared to patients with other sublineages (Additional file 1: Table S1).
The possible association between sublineages and population clinical characteristics was explored, grouping patients with Haarlem, LAM and lineages other than Haarlem and LAM (Table 2). Significantly, most of the patients with Haarlem lineage were men ($61.3\%$, $\frac{247}{419}$); $25.4\%$ ($\frac{106}{417}$) had ever smoked, had a median BMI of 20.5 (IQR 18.1–23.4), lower than the average, and $75.9\%$ ($\frac{318}{419}$) were found clustered ($p \leq 0.05$). Significantly more patients with LAM lineage presented cavitation $54.5\%$ ($\frac{74}{136}$) as radiographic finding ($$p \leq 0.023$$). In patients with sublineages other than Haarlem and LAM $40.5\%$ ($\frac{70}{173}$) had hemoptysis as the most common clinical feature ($$p \leq 0.005$$).Table 2Sociodemographic and clinical characteristics of patients with pulmonary tuberculosis according to sublineage aggregated, determined by 24-loci MIRU–VNTR, Orizaba Veracruz 1997–2010 ($$n = 755$$)VariablesTotalHaarlemLAMOther than Haarlem and LAMp-value*n/total (%)n/total (%)n/total (%)n/total (%)Male$\frac{442}{755}$ (58.5)$\frac{257}{419}$ (61.3)$\frac{81}{163}$ (49.7)$\frac{104}{173}$ (60.1)0.034Age (years) (median [IQR])45 (32–59)45 (33–59)43 (30–55)47 (31–60)0.503† > 6 years of formal schooling$\frac{507}{754}$ (67.2)$\frac{277}{418}$ (66.3)$\frac{115}{163}$ (70.6)$\frac{115}{173}$ (24.3)0.595House with dirt floor$\frac{174}{755}$ (23.1)$\frac{88}{419}$ (21.0)$\frac{44}{163}$ (27.0)$\frac{42}{173}$ (24.0)0.277Rural residence$\frac{99}{732}$ (13.5)$\frac{47}{409}$ (11.5)$\frac{30}{159}$ (18.9)$\frac{22}{164}$ (13.4)0.070Nearest health center distance (meters) (median [IQR])708 (414–1099)775 (432–1184)775 (432–1183.7)737 (455–1118)0.241†Access to social security$\frac{254}{755}$ (33.6)$\frac{152}{419}$ (36.3)$\frac{48}{163}$ (29.5)$\frac{54}{173}$ (31.2)0.218Used alcohol$\frac{330}{753}$ (43.8)$\frac{194}{417}$ (46.5)$\frac{64}{163}$ (39.3)$\frac{72}{173}$ (41.6)0.229Used tobacco$\frac{164}{753}$ (21.8)$\frac{106}{417}$ (25.4)$\frac{22}{163}$ (13.5)$\frac{36}{173}$ (20.8)0.007Used illegal drugs$\frac{35}{754}$ (4.6)$\frac{19}{418}$ (4.6)$\frac{5}{163}$ (3.1)$\frac{11}{173}$ (6.4)0.355Homelessness or residence in shelters$\frac{20}{754}$ (2.7)$\frac{12}{419}$ (2.9)$\frac{4}{162}$ (2.5)$\frac{4}{173}$ (2.3)0.918Diabetes mellitus type $\frac{2250}{755}$ (33.1)$\frac{138}{419}$ (32.9)$\frac{52}{163}$ (31.9)$\frac{60}{173}$ (34.7)0.858HIV coinfection$\frac{13}{739}$ (1.8)$\frac{7}{410}$ (1.7)$\frac{3}{163}$ (1.8)$\frac{3}{166}$ (1.8)0.993New case$\frac{688}{754}$ (91.3)$\frac{384}{418}$ (91.9)$\frac{145}{163}$ (89.0)$\frac{159}{173}$ (91.9)0.505AFB sputum positivity grade of 3+ $\frac{211}{755}$ (28.0)$\frac{123}{419}$ (29.4)$\frac{47}{163}$ (28.8)$\frac{41}{173}$ (23.7)0.363Resistance to at least one drug$\frac{117}{612}$ (19.1)$\frac{68}{337}$ (20.2)$\frac{25}{131}$ (19.1)$\frac{24}{144}$ (16.7)0.669MDR$\frac{21}{612}$ (3.4)$\frac{8}{337}$ (2.4)$\frac{8}{131}$ (6.1)$\frac{5}{144}$ (3.5)0.138Fever$\frac{531}{752}$ (70.6)$\frac{292}{417}$ (70.0)$\frac{114}{162}$ (70.4)$\frac{125}{173}$ (72.3)0.861Haemoptysis$\frac{250}{753}$ (33.2)$\frac{141}{417}$ (33.8)$\frac{39}{163}$ (23.9)$\frac{70}{173}$ (40.5)0.005Cavitations presence on chest radiograph$\frac{282}{626}$ (45.1)$\frac{152}{344}$ (44.2)$\frac{74}{136}$ (54.5)$\frac{56}{146}$ (38.4)0.023BMI (median [IQR])20.9(18.4–23.8)20.5(18.1–23.4)21.0(18.9–24.1)21.6(19.0–23.9)0.016†Number of days between symptom onset and start of treatment (median [IQR])105 (67–182)106 (63–180)119 (80–196)97 (66.5–175)0.098†Cluster belonging$\frac{519}{755}$ (68.7)$\frac{318}{419}$ (75.9)$\frac{95}{163}$ (58.3)$\frac{106}{173}$ (61.3) < 0.001IQR, interquartile range; HIV, human immunodeficiency virus; AFB, acid fast bacilli; MDR, multidrug resistance; BMI, body mass index*Pearson chi-square test†Kruskal–Wallis test Regarding treatment outcomes, we compared cure or treatment completion with treatment failure in all identified lineages. It was observed that patients with X sublineage isolates ($\frac{2}{11}$, $18.2\%$) and unknown isolates ($\frac{1}{7}$, $14.3\%$) had the highest failure rates, followed by Cameroon ($\frac{3}{41}$, $7.3\%$), EAI ($\frac{1}{15}$, $6.7\%$), S ($\frac{1}{18}$, $5.6\%$), LAM ($\frac{6}{135}$, $4.4\%$), Uganda I ($\frac{1}{26}$, $3.9\%$) and Haarlem ($\frac{5}{358}$, $1.4\%$). Patients with Beijing, Ghana and TUR lineages did not experience treatment failure (Additional file 1: Table S2).
Treatment outcome according to sublineage is summarized in Additional file 1: Table S3. When comparing cure or completion with treatment failure, patients with sublineages other than Haarlem and LAM showed higher proportion of treatment failure ($\frac{9}{152}$, $5.9\%$) than patients with Haarlem ($1.4\%$) and LAM ($4.4\%$) lineages ($$p \leq 0.016$$).
The comparison among clinical characteristics according to cure or treatment completion compared to failure, revealed higher proportion of treatment failure in patients who had ever smoked ($\frac{8}{20}$, $40\%$ vs $\frac{131}{624}$, $21.0\%$; $$p \leq 0.034$$), showed resistance to at least one drug ($\frac{13}{17}$, $76.0\%$ vs $\frac{80}{505}$, $16.0\%$; $p \leq 0.001$), showed MDR presence ($\frac{5}{17}$, $2.0\%$ vs $\frac{6}{505}$, $1.0\%$, $p \leq 0.001$) and had sublineages other than Haarlem and LAM ($\frac{9}{20}$, $45.0\%$ vs $\frac{143}{625}$, $23.0\%$; $$p \leq 0.022$$) (Table 3). We observed higher proportion of cure in patients who had formal education > 6 years ($\frac{423}{625}$, $68.0\%$ vs $\frac{9}{20}$, $45\%$; $$p \leq 0.034$$), had Haarlem sublineage ($\frac{353}{625}$, $56.5\%$ vs $\frac{5}{20}$, $25\%$; $$p \leq 0.005$$) and presented LAM RDRio sublineage ($\frac{1}{6}$, $16.7\%$ vs $\frac{92}{129}$, $71.3\%$; $$p \leq 0.005$$).Table 3Sociodemographic and clinical characteristics of patients with pulmonary tuberculosis by treatment outcome, Orizaba Veracruz 1997–2010 ($$n = 755$$)VariablesTotalFailureCure or treatment completionp-value*n/totala (%)n/total (%)n/total (%)Male$\frac{363}{645}$ (56.0)$\frac{15}{20}$ (75.0)$\frac{348}{625}$ (56.0)0.086Age (years) (median [IQR])56 (45–35)51 (33–57)45 (32–59)0.787† > 6 years of formal schooling$\frac{432}{645}$ (67.0)$\frac{9}{20}$ (45.0)$\frac{423}{625}$ (68.0)0.034House with dirt floor$\frac{154}{645}$ (24.0)$\frac{3}{20}$ (15.0)$\frac{151}{625}$ (24.0)0.344Rural residence$\frac{81}{624}$ (13.0)$\frac{3}{18}$ (17.0)$\frac{78}{606}$ (13.0)0.637Nearest health center distance (meters) (median [IQR])689 (412–1095)848 (274–1260)692 (412–1092)0.717†Access to social security$\frac{225}{645}$ (35.0)$\frac{3}{20}$ (15.0)$\frac{222}{625}$ (36.0)0.058Used alcohol$\frac{269}{644}$ (42.0)$\frac{10}{20}$ (50.0)$\frac{259}{624}$ (42.0)0.448Used tobacco$\frac{139}{644}$ (22.0)$\frac{8}{20}$ (40.0)$\frac{131}{624}$ (21.0)0.042Used illegal drugs$\frac{26}{645}$ (4.0)$\frac{0}{20}$ (0.0)$\frac{26}{625}$ (4.0)0.352Homelessness or residence in shelters$\frac{14}{644}$ (2.0)$\frac{0}{20}$ (0.0)$\frac{14}{624}$ (2.0)0.498Diabetes mellitus type $\frac{2221}{645}$ (34.0)$\frac{7}{20}$ (35.0)$\frac{214}{625}$ (34.0)0.944HIV coinfection$\frac{6}{635}$ (1.0)$\frac{0}{19}$ (0.0)$\frac{6}{616}$ (1.0)0.666New case$\frac{592}{644}$ (92.0)$\frac{17}{20}$ (85.0)$\frac{575}{624}$ (92.0)0.248AFB sputum positivity grade of 3 + $\frac{174}{645}$ (27.0)$\frac{7}{20}$ (35.0)$\frac{167}{625}$ (27.0)0.411Resistance to at least one drug$\frac{93}{522}$ (18.0)$\frac{13}{17}$ (76.0)$\frac{80}{505}$ (16.0) < 0.001MDR$\frac{11}{522}$ (2.0)$\frac{5}{17}$ (29.0)$\frac{6}{505}$ (1.0) < 0.001Fever$\frac{462}{642}$ (72.0)$\frac{11}{20}$ (55.0)$\frac{451}{622}$ (73.0)0.086Haemoptysis$\frac{225}{645}$ (35.0)$\frac{6}{20}$ (30.0)$\frac{219}{625}$ (35.0)0.642Cavitations presence on chest radiograph$\frac{244}{545}$ (45.0)$\frac{8}{13}$ (62.0)$\frac{236}{532}$ (44.0)0.218BMI (median [IQR])21.0 (18.7–23.8)19.1 (17.6–22.5)21.1 (18.7–22.5)0.181†Number of days between symptom onset and start of treatment (median [IQR])105 (67–197)152 (79–242)105 (66–179)0.120†Lineage by MIRUHaarlem$\frac{358}{645}$ (55.5)$\frac{5}{20}$ (25.0)$\frac{353}{625}$ (56.5)0.005‡LAM$\frac{135}{645}$ (20.9)$\frac{6}{20}$ (30.0)$\frac{129}{625}$ (20.6)0.311‡RDRio$\frac{93}{135}$ (68.9)$\frac{1}{6}$ (16.7)$\frac{92}{129}$ (71.3)0.005‡RD$\frac{11526}{135}$ (19.3)$\frac{2}{6}$ (33.3)$\frac{24}{129}$ (18.6)0.371‡Other than Haarlem and LAM$\frac{152}{645}$ (24.0)$\frac{9}{20}$ (45.0)$\frac{143}{625}$ (23.0)0.022‡Cluster belonging$\frac{447}{645}$ (69.3)$\frac{10}{20}$ [50]$\frac{437}{625}$ (69.9)0.057IQR, interquartile range; HIV, human immunodeficiency virus; AFB, acid fast bacilli; MDR, multidrug resistance; BMI, body mass index; RD, region of difference*Pearson chi-square test†Kruskal–Wallis test‡Binomial testaBecause there were missing values for the characteristics of some of the tuberculosis patients, several of the numbers below do not sum to the group total We compared cavitation presence or absence with population clinical characteristics (Table 4). We found that patients with cavitation had DM2 as comorbidity ($\frac{106}{282}$, $38.0\%$ vs $\frac{95}{344}$, $28.0\%$; $$p \leq 0.008$$), presented greater number of AFB in sputum samples ($\frac{101}{282}$, $36.0\%$ vs $\frac{71}{344}$, $21.0\%$, $p \leq 0.001$) and presented LAM sublineage isolates ($\frac{74}{282}$, $26.2\%$ vs $\frac{62}{344}$, $18.0\%$; $$p \leq 0.013$$).Table 4Sociodemographic and clinical characteristics of patients with pulmonary tuberculosis by cavitation presentation, Orizaba Veracruz 1997–2010 ($$n = 755$$)VariablesTotalCavitations presenceCavitations absencep-value*n/total (%)n/total (%)n/total (%)Male$\frac{362}{626}$ (58.0)$\frac{160}{282}$ (57.0)$\frac{202}{344}$ (59.0)0.617Age (years) (median [IQR])46 (32–59)47 (33–60)45 (31–580.466†> 6 years of formal schooling$\frac{429}{626}$ (69.0)$\frac{191}{282}$ (68.0)$\frac{238}{344}$ (69.0)0.696House with dirt floor$\frac{151}{626}$ (24.0)$\frac{66}{282}$ (23.0)$\frac{85}{344}$ (25.0)0.704Rural residence$\frac{81}{610}$ (13.0)$\frac{36}{273}$ (13.0)$\frac{45}{337}$ (13.0)0.952Nearest health center distance (meters) (median [IQR])702 (434.0–1089)749 (464.1–1109)683 (412–1073)0.114†Access to social security$\frac{224}{626}$ (36.0)$\frac{108}{282}$ (38.0)$\frac{116}{344}$ (34.0)0.235Used alcohol$\frac{265}{626}$ (42.0)$\frac{129}{282}$ (46.0)$\frac{136}{344}$ (40.0)0.118Used tobacco$\frac{135}{626}$ (22.0)$\frac{54}{282}$ (19.0)$\frac{81}{344}$ (24.0)0.183Used illegal drugs$\frac{30}{626}$ (5.0)$\frac{12}{282}$ (4.0)$\frac{18}{344}$ (5.0)0.569Homelessness or residence in shelters$\frac{18}{626}$ (3.0)$\frac{10}{282}$ (4.0)$\frac{8}{344}$ (2.0)0.363Diabetes mellitus type $\frac{2201}{626}$ (32.0)$\frac{106}{282}$ (38.0)$\frac{95}{344}$ (28.0)0.008HIV coinfection$\frac{11}{616}$ (2.0)$\frac{2}{276}$ (1.0)$\frac{9}{340}$ (3.0)0.073New case$\frac{566}{625}$ (91.0)$\frac{249}{281}$ (89.0)$\frac{317}{344}$ (92.0)0.132AFB sputum positivity grade of 3+ $\frac{172}{626}$ (27.0)$\frac{101}{282}$ (36.0)$\frac{71}{344}$ (21.0) < 0.001Resistance to at least one drug$\frac{101}{524}$ (19.0)$\frac{47}{220}$ (21.0)$\frac{54}{304}$ (18.0)0.302MDR$\frac{18}{524}$ (3.0)$\frac{8}{220}$ (4.0)$\frac{10}{304}$ (3.0)0.830Fever$\frac{438}{623}$ (70.0)$\frac{206}{280}$ (74.0)$\frac{232}{343}$ (68.0)0.107BMI (median [IQR])21.0 (18.4–13.8)20.8 (18.5–23.5)21.2 (18.4–23.90.356†Number of days between symptom onset and start of treatment (median [IQR])104 (66–177)108 (68–195)100 (63–161)0.064†Haemoptysis$\frac{211}{625}$ (34.0)$\frac{95}{282}$ (34.0)$\frac{116}{343}$ (34.0)0.972Lineage by MIRUHaarlem$\frac{344}{626}$ (55.0)$\frac{152}{282}$ (53.9)$\frac{192}{344}$ (55.8)0.632‡LAM$\frac{136}{626}$ (21.7)$\frac{74}{282}$ (26.2)$\frac{62}{344}$ (18.0)0.013‡RDRio$\frac{97}{136}$ (71.3)$\frac{55}{74}$ (74.3)$\frac{42}{62}$ (67.7)0.398‡RD$\frac{11524}{136}$ (17.7)$\frac{11}{74}$ (14.9)$\frac{13}{62}$ (21.0)0.352‡Other than Haarlem and LAM$\frac{146}{626}$ (23.0)$\frac{56}{282}$ (20.0)$\frac{90}{344}$ (26.0)0.063‡‡Cluster belonging$\frac{433}{626}$ (69.2)$\frac{198}{282}$ (70.2)$\frac{235}{344}$ (68.3)0.609‡IQR, interquartile range; HIV, human immunodeficiency virus; AFB, acid fast bacilli; MDR, multidrug resistance; BMI, body mass index; RD, region of difference*Pearson chi-square test†Kruskal–Wallis test‡Binomial test Using logistic regression models, we performed two adjusted models to explain treatment failure compared to cure and treatment completion; in one we included resistance variable to at least one drug, and in the other we included MDR variable (Table 5). In the first model adjusted for covariates, treatment failure was associated with resistance to at least one drug (aOR 25.763, $95\%$ CI 7.096–93.543; $p \leq 0.001$) and having lineage other than Haarlem and LAM (aOR 6.740, $95\%$ CI 1.704–26.661; $$p \leq 0.007$$). In the model that included MDR variable adjusted for covariates, failure was associated with MDR (aOR 31.497, $95\%$ CI 5.119–193.815; $p \leq 0.001$), in both models having > 6 years of formal education was not associated with treatment failure (aOR 0.166, $95\%$ CI 0.045–0.615; $$p \leq 0.007$$), (aOR 0.248, $95\%$ CI 0.069–0.885; $$p \leq 0.032$$) respectively. Table 5Characteristics associated with treatment failure Orizaba, Veracruz, 1997–2010 ($$n = 522$$)VariablesaOR$95\%$ CIp-value*aOR$95\%$ CIp-value*L.IU.IL.IU.ISex (male)3.0300.76611.9840.1143.1520.81012.2660.098Age (years)1.0421.0001.0860.0511.0340.9951.0740.085> 6 years of formal schooling0.1660.0450.6150.0070.2480.0690.8850.032Access to social security0.2460.0561.0850.0640.2640.0621.1230.071Diabetes mellitus type 20.7160.1952.6340.6150.7500.2222.5320.643New case0.8600.1744.2570.8531.0470.2015.4420.957Resistance to at least one drug25.7637.09693.543 < 0.001––––MDR––––31.4975.119193.815 < 0.001LineageHaarlemReferenceReferenceLAM2.8520.57414.1810.2002.0090.4269.4750.378Other than Haarlem and LAM6.7401.70426.6610.0073.3420.90612.3200.070aOR, adjusted odds ratio; MDR, multidrug resistance; L.I, lower interval; U.I, upper interval*Adjusted logistic regression We performed an adjusted logistic regression model to identify variables associated with cavitations presence (Table 6). Cavitation presence was associated with having ever consumed alcohol (aOR 1.528, $95\%$ CI 1.041–2.243; $$p \leq 0.030$$), having DM2 (aOR 1.625, $95\%$ CI 1.130–2.337; $$p \leq 0.009$$), AFB sputum positivity grade of 3 + (aOR 2.198, $95\%$ CI 1.524–3.168; $p \leq 0.001$) and having the LAM sublineage (aOR 1.023, $95\%$ CI 1.023–2.333; $$p \leq 0.039$$).Table 6Characteristics associated with the presence of cavitations in chest radiographs of patients with pulmonary TB in Orizaba, Veracruz, 1997–2010 ($$n = 626$$)VariablesaOR$95\%$ CIp-value*L.IU.ISex (male)0.7810.5291.1520.212Age (years)1.0030.9931.0130.573Used alcohol1.5281.0412.2430.030Diabetes mellitus type 21.6251.1302.3370.009AFB sputum positivity grade of 3+ 2.1981.5243.168 < 0.001LineageHaarlemReferenceLAM1.5451.0232.3330.039Other than Haarlem and LAM0.8060.5371.2100.298aOR, adjusted odds ratio; MDR, multidrug resistance; L.I, lower interval; U.I, upper interval*Adjusted logistic regression
## Discussion
This study describes the association among clinical and sociodemographic characteristics of patients with pulmonary tuberculosis and little described M. tuberculosis sublineages lineage 4 in the health jurisdiction of Orizaba, Veracruz, Mexico between 1995 and 2010. Our study population presented high frequency of lineage 4, Euro-American isolates. In addition, associated characteristics with treatment failure and cavitation presence were identified.
In this study, lineage 4 (Euro-American) was the most common (~ $96\%$) lineage identified, consistent with previous reports that have shown that isolates with this lineage are predominant in Mexico [18]. We were also able to observe that isolates with LAM lineage [163], the proportion of RDRio was $69.9\%$, higher compared to the $63.1\%$ recently described in isolates from Northern Mexico and in isolates from Venezuela ($55\%$), Argentine ($11\%$) and Paraguay ($10\%$) [19, 20]. Therefore, our results support that these lineages are endemic and that strains spread regionally with different rates of distribution.
We found that compared with other sublineages, cases with Haarlem sublineage isolates had higher proportion of clustered patients. A previous study showed similar results; the authors found that Haarlem sublineage isolates were more likely to belong to clusters [21]. This result confirms the wide distribution and genetic diversity of lineage 4 due to its virulence, which is reflected in cluster formation and its transmission success among the population [22].
On the other hand, we found that patients with Cameroon sublineage isolates showed more days between symptom onset and treatment start. Similar result have been described in patients with lineage 7 isolates in Ethiopia, where the time was longer between symptom onset and treatment start was attributed to lineage 7 strains slow growth [21]. Because treatment initiation is important to cut transmission chains, it is necessary to phenotypically confirm Cameroon sublineage isolates growth rate. To confirm this hypothesis, we cultured 45 isolates with Cameroon lineage on MGIT medium and determined the time and units of growth. We observed, that the Cameroon isolates grew less (14.6 CFU/h) compared to H37Rv (24.7 CFU/h).
Respect to Ghana sublineage, we found the majority of patients presented haemoptysis; this finding has not been reported thus far in literature. However, more data are needed.
Another interesting result was that cases with isolates of EAI lineage were more frequent in men, in patients with DM2 and patients who had ever smoked. It has been described that DM2 alone is associated with M. tuberculosis infection and progression to active disease with severe disease presentation [23]. Furthermore, decreased lung function has been observed in smokers with DM2 compared to nonsmokers [24]. Therefore, it is likely that social factors contribute to EAI dissemination, also these patients showed higher cavitations proportion ($69.2\%$), without statistical significance. Previously, a study that evaluated host–pathogen relationship and its association with clinical outcomes in patients with tuberculosis described that patients infected with strains that originated in geographic regions other than the patient’s origin (allopatric) such as EAI lineage in America presented an increased lung damage risk [25]. As observed in our results, it has been suggested that although these lineages are less adapted to transmit and cause disease in fully competent members of allopatric human populations, they can do so in the context of impaired host immune resistance [26]. However, it would be necessary to perform whole genome sequence on EAI lineage isolates to determine pathogen genetic characteristics that facilitate its possible adaptation to the host and transmission.
Furthermore, East Asia (Beijing) lineage was found in two elderly patients, one of them had HIV and the other had DM2 and MDR. Beijing isolates were genetically distinct, with $\frac{9}{15}$ different alleles by 24-loci MIRU–VNTR; these cases were probably due to reactivation. MDR has been associated with Beijing family; however, in this study, data are not conclusive because there were only two isolates [11, 27]. However, it is very likely that MDR isolate is due to antibiotics selective pressure because patient had received treatment previously.
We also observed higher proportion of treatment failure in patients with isolates of sublineages other than Haarlem and LAM, in patients who had ever smoked and in patients with isolates resistant to at least one drug or MDR. A greater proportion of resistance was found in Cameroon ($\frac{13}{49}$, $30.2\%$), UgandaI ($\frac{5}{22}$, $22.7\%$) and Ghana ($\frac{2}{16}$, $12.5\%$) sublineage isolates. A recent study conducted in Niger reported that $75\%$ of Cameroon and Ghana sublineage isolates were resistant to RIF and MDR [28]. However, treatment failure could be also the result of "antibiotic resilience" as recently described by Quingyun et al., they found that mutations in resR (Rv1830) gene do not show canonical drug resistance or drug tolerance but instead shorten the post-antibiotic effect, meaning that they enable M. tuberculosis to resume growth after drug exposure substantially faster than wild-type strains, and these mutations are associated to treatment failure acting in a regulatory cascade with other transcription factors controlling cell growth and division. Furthermore, they described that up to $10\%$ of strains from high-tuberculosis-burden countries showed fixed mutations in these regions [29].According to our results Cameroon and Ghana sublineages geographically restricted within Euro-American lineage, seem to have adapted to the study population and contribute significantly to the resistance generation and treatment failure. Therefore, it is necessary to genotype a greater number of isolates and performs susceptibility tests to determine the real impact on the resistance of lineages little described in Mexico and to perform whole genome sequencing to explore the possible association between resR mutation, treatment failure and whether any lineage is prone to acquire it.
Interestingly, having > 6 years of formal education was not associated with treatment failure. We believe that having higher education level probably implies that patients better understand treatment adherence and completion importance.
The LAM RDRio lineage has been described in other Latin American countries where it has been associated with the presence of cavitations, increased transmissibility and MDR [19]. However, in the present study we observed more proportion of LAM RDRio isolates in cured patients, previously in this study population was obtained that previous treatment (aOR 9.05, $95\%$ CI 3.6–22.5, $p \leq 0.001$) and LAM lineage (aOR 4.25, $95\%$ CI 1.4–12.7, $$p \leq 0.010$$) were associated with tuberculosis MDR [15]. These results have important implications in the tuberculosis control program, although isolates with LAM RDRio sub lineage are more prone to develop MDR following a previous treatment, patients seem to respond favorably to the second treatment.
Cavitations presence was associated with LAM sublineage, alcohol consumption, DM2 and AFB positivity grade 3+. Similar results have been previously described regarding the presence of more severe manifestations in patients with DM2 and tuberculosis [30, 31]. Moreover, it has been reported that cavitations presence in pulmonary tuberculosis is associated with higher contagiousness/transmissibility due to high AFB load [32]. In addition, these results support those described by Pasopanodya et al., who report that modern lineages strains, such as Euro-American lineages, developed nonlethal properties; however, they cause lung damage, which increases their dissemination capacity among the population [25]. Therefore, the increase in the number of people with DM2 in Mexico could result in greater transmission of tuberculosis due to lung damage associated with the presence of LAM sublineage. We thus suggest implementing genotyping of M. tuberculosis isolates with the use of 24-loci MIRU–VNTR in Mexico and determining the impact of LAM sublineage.
In conclusion, this study provides relevant results in relation to the association between the presence of cavitations, comorbidities and LAM sublineage isolates. Additionally, treatment failure associated with sublineages other than Haarlem and LAM. Furthermore, we found the possible EAI sublineage isolates association in patients with DM2 and cavitation. We describe that the genetic diversity of M. tuberculosis lineage 4 (Euro-American) probably plays an essential role in disease presentation, which could have important implications for treatment management and to improve tuberculosis control in Mexico.
## Supplementary Information
Additional file 1: Table S1. Sociodemographic and clinical characteristics of patients with pulmonary tuberculosis according to disaggregated sublineage, determined by 24-loci MIRU-VNTR, Orizaba Veracruz 1997-2010 ($$n = 755$$). Table S2. Treatment outcome of patients with pulmonary tuberculosis according to disaggregated lineage by 24-loci MIRU-VNTR, Orizaba Veracruz 1997-2010 ($$n = 755$$). Table S3. Treatment outcome of patients with pulmonary tuberculosis according to aggregated lineage by 24-loci MIRU-VNTR, Orizaba Veracruz 1997-2010 ($$n = 755$$).
## References
1. 1.World Health Organization. Global tuberculosis report 2021. https://www.who.int/publications/i/item/9789240037021.
2. 2.CENAPRECE, 2022. Programas Preventivos. Direccion de Micobacteriosis. Casos Nuevos de Tuberculosis Pulmonar. Estados Unidos Mexicanos 1990–2016. http://www.cenaprece.salud.gob.mx/programas/interior/micobacteriosis/tuberculosis/cifras_oficiales.html.
3. Blanco-Guillot F, Castaheda-Cediel ML, Cruz-Hervert P, Ferreyra-Reyes L, Delgado-Sanchez G, Ferreira-Guerrero E. **Genotyping and spatial analysis of pulmonary tuberculosis and diabetes cases in the state of Veracruz, Mexico**. *PLoS ONE* (2018.0) **13** e0193911. DOI: 10.1371/journal.pone.0193911
4. Coll F, McNerney R, Guerra-Assunção JA, Glynn JR, Perdigão J, Viveiros M. **A robust SNP barcode for typing**. *Nat Commun* (2014.0) **5** 4812. DOI: 10.1038/ncomms5812
5. Coscolla M, Gagneux S, Menardo F, Loiseau C, Ruiz-Rodriguez P, Brites D. **Phylogenomics of**. *Microb Genom* (2021.0) **7** 000477. PMID: 33555243
6. Gagneux S. **Host-pathogen coevolution in human tuberculosis**. *Philos Trans R Soc Lond B Biol Sci* (2012.0) **367** 850-859. DOI: 10.1098/rstb.2011.0316
7. Kato-Maeda M, Metcalfe JZ, Flores L. **Genotyping of**. *Future Microbiol* (2011.0) **6** 203-216. DOI: 10.2217/fmb.10.165
8. Couvin D, Rastogi N. **Tuberculosis—a global emergency: tools and methods to monitor, understand, and control the epidemic with specific example of the Beijing lineage**. *Tuberculosis.* (2015.0). DOI: 10.1016/j.tube.2015.02.023
9. Mokrousov I, Vyazovaya A, Iwamoto T, Skiba Y, Pole I, Zhdanova S. **Latin–American–Mediterranean lineage of**. *Mol Phylogenet Evol* (2016.0) **99** 133-143. DOI: 10.1016/j.ympev.2016.03.020
10. Wiens KE, Woyczynski LP, Ledesma JR, Ross JM, Zenteno-Cuevas R, Goodridge A. **Global variation in bacterial strains that cause tuberculosis disease: a systematic review and meta-analysis**. *BMC Med* (2018.0) **16** 1-13. DOI: 10.1186/s12916-018-1180-x
11. Gagneux S. **Ecology and evolution of**. *Nat Rev Microbiol* (2018.0) **16** 202-213. DOI: 10.1038/nrmicro.2018.8
12. Dheda K, Barry CE, Maartens G. **Tuberculosis**. *Lancet* (2016.0) **387** 1211-1226. DOI: 10.1016/S0140-6736(15)00151-8
13. Chae H, Shin SJ. **Importance of differential identification of**. *J Microbiol* (2018.0) **56** 300-311. DOI: 10.1007/s12275-018-8041-3
14. Coscolla M, Gagneux S. **Consequences of genomic diversity in**. *Semin Immunol* (2014.0) **26** 431-444. DOI: 10.1016/j.smim.2014.09.012
15. Ordaz-Vázquez A, Torres-González P, Cruz-Hervert P, Ferreyra-Reyes L, Delgado-Sánchez G, García-García L. **Genetic diversity and primary drug resistance transmission in**. *Infect Genet Evol* (2021.0) **93** 104994. DOI: 10.1016/j.meegid.2021.104994
16. 16.NORMA Oficial Mexicana NOM-006-SSA2-1993, para la Prevención y control de la tuberculosis.
17. 17.Varaine F, Rich M (eds). Tuberculosis: practical guide for clinicians, nurses, laboratory technicians and medical auxiliaries. Médecins San Frontières and Partners In Health; 2014.
18. Flores-Treviño S, Mendoza-Olazarán S, Garza-González E. **Drug resistance and molecular epidemiology of**. *Salud Publica Mex* (2014.0) **56** 63-77. DOI: 10.21149/spm.v56i1.7324
19. Bocanegra-García V, Cortez-de-la-Fuente LJ, Nakamura-López Y, González GM, Rivera G, Palma-Nicolás JP. **RD**. *Enferm Infecc Microbiol Clin* (2021.0) **39** 399-402. DOI: 10.1016/j.eimc.2020.07.003
20. DíazAcosta CC, Russomando G, Candia N, Ritacco V, Vasconcellos SEG, De BerrêdoPinhoMoreira M. **Exploring the “Latin American Mediterranean” family and the RD**. *BMC Microbiol.* (2019.0) **19** 131. DOI: 10.1186/s12866-019-1479-6
21. Yimer SA, Norheim G, Namouchi A, Zegeye ED, Kinander W, Tønjum T. *J Clin Microbiol* (2015.0) **53** 1301-1309. DOI: 10.1128/JCM.03566-14
22. Stucki D, Brites D, Jeljeli L, Coscolla M, Liu Q, Trauner A. *Nat Genet* (2016.0) **48** 1535-1543. DOI: 10.1038/ng.3704
23. Restrepo BI, Schlesinger LS. **Host-pathogen interactions in tuberculosis patients with type 2 diabetes mellitus**. *Tuberculosis.* (2013.0). DOI: 10.1016/S1472-9792(13)70004-0
24. Kolahian S, Leiss V, Nürnberg B. **Diabetic lung disease: fact or fiction?**. *Rev Endocr Metab Disord* (2019.0) **20** 303-319. DOI: 10.1007/s11154-019-09516-w
25. Pasipanodya JG, Moonan PK, Vecino E, Miller TL, Fernandez M, Slocum P. **Allopatric tuberculosis host–pathogen relationships are associated with greater pulmonary impairment**. *Infect Genet Evol* (2013.0) **16** 433-440. DOI: 10.1016/j.meegid.2013.02.015
26. Gagneux S, DeRiemer K, Van T, Kato-Maeda M, De Jong BC, Narayanan S. **Variable host-pathogen compatibility in**. *Proc Natl Acad Sci USA* (2006.0) **103** 2869-2873. DOI: 10.1073/pnas.0511240103
27. Hu Y, Ma X, Graviss EA, Wang W, Jiang W, Xu B. **A major subgroup of Beijing family**. *Epidemiol Infect* (2011.0) **139** 130-138. DOI: 10.1017/S0950268810000890
28. Ejo M, Hassane-Harouna S, Souleymane MB, Lempens P, Dockx J, Uwizeye C. **Multidrug-resistant patients receiving treatment in Niger who are infected with**. *Tuberculosis.* (2020.0) **122** 101922. DOI: 10.1016/j.tube.2020.101922
29. Liu Q, Zhu J, Dulberger CL, Stanley S, Wilson S, Chung ES. **Tuberculosis treatment failure associated with evolution of antibiotic resilience**. *Science* (2022.0) **378** 1111-1118. DOI: 10.1126/science.abq2787
30. Jiménez-Corona ME, Cruz-Hervert LP, García-García L, Ferreyra-Reyes L, Delgado-Sánchez G, Bobadilla-Del-Valle M. **Association of diabetes and tuberculosis: impact on treatment and post-treatment outcomes**. *Thorax* (2013.0) **68** 214-220. DOI: 10.1136/thoraxjnl-2012-201756
31. Armstrong LR, Steve Kammerer J, Haddad MB. **Diabetes mellitus among adults with tuberculosis in the USA, 2010–2017**. *BMJ Open Diabetes Res Care* (2020.0) **8** e001275. DOI: 10.1136/bmjdrc-2020-001275
32. Palaci M, Dietze R, Hadad DJ, Ribeiro FKC, Peres RL, Vinhas SA. **Cavitary disease and quantitative sputum bacillary load in cases of pulmonary tuberculosis**. *J Clin Microbiol* (2007.0) **45** 4064-4066. DOI: 10.1128/JCM.01780-07
|
---
title: Medical ID use by international patients with Aspirin-Exacerbated Respiratory
Disease
authors:
- Mohammed Alqabasani
- Andrea Lasso
- Shaun Kilty
journal: 'Allergy, Asthma, and Clinical Immunology : Official Journal of the Canadian
Society of Allergy and Clinical Immunology'
year: 2023
pmcid: PMC10012488
doi: 10.1186/s13223-023-00766-7
license: CC BY 4.0
---
# Medical ID use by international patients with Aspirin-Exacerbated Respiratory Disease
## Abstract
### Background
Patients widely use medical identification (ID) to indicate their food and drug allergies, and chronic medical conditions. One chronic condition for which patients are recommended to use a form of medical ID is Aspirin-Exacerbated Respiratory Disease (AERD), a disease characterized by the presence of asthma, chronic rhinosinusitis with nasal polyps and sensitivity to aspirin and other COX-1 inhibitors, including nonsteroidal anti-inflammatory drugs (NSAIDs). The uptake of medical ID use in AERD is unknown and has not been widely studied in this population.
### Methods
We conducted a cross-sectional survey study to measure the perception of the need to use a medical ID and its use by patients with AERD internationally.
### Results
245 members of an online AERD support group completed an online survey. The majority ($80\%$, $$n = 198$$) of the participants did not use any form of medical ID. The participants reported that the lack of knowledge and awareness about the importance of using a medical ID was the most common reason for not using it.
### Conclusion
This international survey found that the majority of the AERD patient respondents did not use a medical ID. The most common reasons for nonuse were not knowing that it is recommended for their condition and that the patients did not consider it necessary. The results highlight the need for further patient and health care provider education.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s13223-023-00766-7.
## Background
Patients widely use medical identification (ID) to indicate their food and medication allergies as well as any chronic medical conditions. One of the chronic conditions for which patients are recommended to use a form of medical ID is Aspirin-Exacerbated Respiratory Disease (AERD) [1]which is diagnosed by the presence of asthma, chronic rhinosinusitis with nasal polyps and a sensitivity to aspirin and other cyclooxygenase-1 (COX-1) inhibitors, including nonsteroidal anti-inflammatory drugs (NSAIDs). AERD usually presents during the third and fourth decades of life and is more common in females [1]. Almost half of those with AERD have severe asthma, and exposure to NSAIDs can lead to a serious acute asthma exacerbation. [ 2, 3] AERD patients who have not been desensitized must avoid aspirin and NSAIDs. However, total avoidance has been shown to be difficult to achieve. In a survey of AERD patients, $24\%$ reported that they had accidentally ingested an NSAID after they were diagnosed with AERD and were made aware that they should avoid these medications [4]. In twenty five percent of the cases, the medication had been prescribed, with the highest incidence of NSAID prescription by emergency room physicians and other non-ENT surgeons. In an emergency, when patients cannot provide medical history, they are at risk of unintentional exposure to these medications. One example is the use of aspirin as a pre-hospital treatment for chest pain of suspected cardiac origin [5, 6]. Accidental NSAID exposure in emergency circumstances may be avoided by using a medical ID, considering that $97\%$ of ambulance staff and $71\%$ of emergency room personnel check routinely for a medical alert ID [7].
Recently, our team evaluated the prevalence of use of medical ID among patients with AERD at our institution [8]. We surveyed 21 patients with AERD and found that only $19\%$ of patients wore a medical ID. Sixty percent of those who did not wear an ID cited not knowing they should wear one as the primary reason for nonuse. Although this survey was small, to our knowledge, this was the first attempt to determine the extent of use of medical ID in this at risk population. With these results, we sought to evaluate medical ID use in a larger population of AERD patients.
The goal of the present study was to determine the proportion of an international group of patients previously diagnosed with AERD who are currently using a medical ID and to explore patients’ attitudes toward the use of medical ID.
## Ethical considerations
Ethical approval for this study was obtained from the Ottawa Health Sciences Network Research Ethics Board (REB# 20200627-01H). This cross-sectional study used an anonymous online survey created on LimeSurvey [9] hosted at the Ottawa Hospital. The questionnaire was designed to measure the perception of the need to use medical ID and the actual behaviour, use of a medical ID by a patient diagnosed with AERD.
The questionnaire included two parts. Forced choice questions were used for most of the survey. The first part of the questionnaire collected participant demographic data, the perception of medical ID need, its use, the type of ID used and the duration of use. The second part of the questionnaire was for the patients who reported not using a medical ID. The reasons for not using a medical ID were recorded using both forced response and open response options. A link to the survey was posted on the Samter’s Society: AERD Samter’s Triad Support Group on Facebook with the administrator's approval. The survey was open between November 17, 2021, and March 2, 2022, and 2 reminders were posted on the group’s site during this time. The goals of the study were explained on the first screen, and completion of the survey implied consent to participate in the study. We present a descriptive analysis of the participants’ characteristics. Chi-square tests were used to test associations between demographic characteristics and the use of a medical ID. Analysis was done using STATA software (Version 12. College Station, TX: StataCorp LP).
## Results
A total of 245 members of the support group completed the survey. The average age at the time of the survey was 51 years (SD = 12.51); 215 participants ($87.76\%$) were female. Most respondents were located in North America ($77.96\%$) and Europe ($14\%$). Sixty Five percent of respondents had a bachelor’s degree or higher, and 125 ($51\%$) reported a household income of > 80,000 USD per year (see Table 1). Sixty-one ($25\%$) of respondents reported being aware that for people with AERD, it is recommended to use a medical ID; however, only 47 ($19\%$) of the patients were using a medical ID at the time of the survey. Of these, 28 ($59.5\%$) reported using a bracelet, 15 ($25.5\%$) reported using smartphone technology (see Table 2). Twenty-nine ($61.7\%$) of those who were wearing an ID had been wearing it for 1–5 years, while 11 ($23.4\%$) had been wearing the ID for 5–10 years; 7 respondents ($14.8\%$) had been wearing the ID for less than a year. The majority of the participants ($$n = 198$$, $81\%$) were not using any form of medical ID at the time of the survey. The most common reason for not using medical ID was a lack of knowledge regarding the importance of its use. Other patient reasons are listed in Table 3. Twenty-two participants ($9\%$) reported a visit to an emergency room due to ingestion of aspirin or NSAID in the year prior to the survey. Table 1Demographic characteristic of survey respondentsSex n (%)Female215 (87.76)Male27 (11.02)Do not want to answer/no answer3 (1.22)Age—mean (SD)51.13 (12.51)Geographic LocationNorth America191 (77.96)Europe36 (14.69)Asia4 (1.63)Africa2 (0.82)Oceania6 (2.45)No answer6(2.45)Highest Level of EducationGraduate High school20 (8.16)Trade certificate/apprenticeship6 (2.45)Community College/Other non-university program55 (22.45)University Degree (Bachelor's)85 (34.69)Postgraduate or Higher Degree76 (31.02)No answer3(1.22)Household income > $80,000 USD per year125 (51.02)40,000–80,000 USD per year60 (24.49)20,000–40,000 USD per year23 (9.39) < $20, 000 USD per year10 (4.08)No answer27 (11.02)Table 2Types of Medical ID used by survey respondentsType of IDN (%) –multiple answers were possibleBracelet28 (59.57)Smartphone technology12 (25.53)Wallet card7 (14.89)Watch8 (17.02)Necklace6 (12.77)Wristband3 (6.38)Tattoo1 (2.13)Table 3Reason for not wearing an IDReason for not wearing an IDn (%)I did not know I should be wearing one138 (69.70)I don't think it is necessary18 (9.09)The cost of medical ID s too high17 (8.59)They are uncomfortable13 (6.57)I don’t want other people to know that I have a medical condition8 (4.04)Other (includes desensitization)50 (25.25) No associations were found between demographic characteristics and the use of medical ID ($p \leq 0.05$) or visits to the emergency room in the previous year due to accidental exposure to NSAIDs (> 0.05).
## Discussion
Accidental exposure to NSAIDs is not uncommon for patients with AERD, and in some instances it can lead to life-threatening reactions [4]. Patients diagnosed with AERD who are not desensitized must be educated on the names of the medications they must avoid. They should also be trained on the importance of relating this information to health care providers that may dispense or prescribe medications to them. It is also recommended that patients with AERD wear a medical ID. [ 1] *This is* a simple prevention strategy that can avoid accidental exposure to NSAIDs in an emergency situation. In our study, nearly $10\%$ of the respondents had visited an emergency room within 12 months of the survey due to an adverse reaction to an NSAID. This suggests that annually, over $10\%$ of patients with AERD may accidentally ingest an NSAID thereby putting them at risk for a severe medication adverse reaction.
Further, our study results demonstrated that $25\%$ of respondents were aware that the use of medical ID is recommended for AERD patients, while only $19\%$ respondents were using a medical ID at the time of the survey. The majority ($81\%$) of respondents were not using any form of medical identification. Our current findings with this international cohort of AERD patients aligns with previous study results about medical ID use at our local institution, with an AERD population [8]. Despite differences in patient location, the attitudes towards medical ID use and the level of nonuse of medical ID are compellingly similar. The low prevalence of medical ID use could be explained by insufficient patient education by healthcare providers about the advantages of medical ID use for their NSAID sensitivity, representing a deficiency in patient care and a potential area for improvement. In this study, other reasons reported by survey respondents for not using a medical ID were high cost, not wanting others to know about their medical condition, appearance, and discomfort. Such reasons could be addressed and discussed with AERD patients with their health care provider during the education and counselling of the importance of using medical ID for their disease.
There was no association between age, income, geographical location, or education level and the use of medical ID in this study. However, there was an association between wearing medical ID and emergency room visits in the last year due to accidental exposure to NSAIDs. One potential explanation for this observation is the non-adherence phenomenon [10]. That is, the non-users of medical ID may have fewer health concerns and invest less in their in their own health. It is well known that nonadherence is a significant problem in chronic disease management [11] and non-adherence is directly related to elevated health care costs, hospitalizations, and patient mortality [12–14]. Despite this possible explanation, it remains a challenge to identify the exact cause for our results without having adequate information regarding the context of the visits to the emergency room or how the survey respondents were exposed to aspirin and NSAIDs.
The participants in this survey reported using many types of medical ID. The most commonly form of ID used was bracelets ($60\%$) followed by smart phone technology ($26\%$); other forms of ID such as a wallet card, wristbands and tattoos were also reported. The wide variety of options for medical ID can make it easier for patients to obtain one, but it can also mean that in an emergency, they are not easily located or accessible by health care providers. In our study, we did not ask for the type of information in the medical ID and therefore there may be limitations on the type and quality of the information provided. A recent study about the use of medical ID to communicate allergy information using of the MedicAlert Foundation database in Australia [15], found that the quality of the information on the ID was variable and non-standardized. The authors suggest that the specific allergen and nature of the reaction be recorded on the ID. Currently, there are no guidelines specifying what information AERD patients should have on a medical ID, therefore, this should be part of the education given to patients by their physicians.
There are clear advantages of using a medical ID for the AERD patient population to avoid or lower the risk of accidental exposure to aspirin and NSAIDs. Medical ID can serve as a reminder to AERD patients of their hypersensitivity when purchasing over counter medications or when getting a new physician prescription. The essential role of medical identification is clearly shown in emergencies when a patient cannot communicate to provide their medical history to health care providers, but they may [15] also expedite emergency treatment once an accidental exposure has occurred by providing information to emergency responders or emergency room personnel about possible cause for the hypersensitivity reaction. In our study, $9\%$ of patients reported a visit to an emergency room due to exposure to aspirin or another NSAID, however the survey did not collect details about the circumstances around the accidental exposure and therefore we cannot determine if the use of medical ID could have prevented the exposure. The real impact of medical ID use by the AERD population remains unclear on their care. The present study demonstrated that AERD patients require more education regarding the importance of using medical ID, which in turn can be sign that health care providers need to be trained themselves on AERD, NSAID avoidance and the need for medical ID use in this population. Additionally, this study can be used as an impetus for more investment into patient safety initiatives and as a guide for additional work to prevent accidental exposure to aspirin and NSAIDs in the AERD patient population. Furthermore, future studies may be needed to evaluate both the real impact of medical ID in preventing accidental exposures or expediting treatment is needed as well as the cost-effectiveness of medical ID use by the AERD patient population.
Our study has some limitations. Although we asked those who answered the survey to confirm that they had been diagnosed with AERD, we could not verify their diagnosis. The Facebook support group has approximately 4500 members. We only obtained 245 complete responses to the survey (response rate of $5.4\%$). However, we do not know how many members of the group are actual patients with AERD, and how many may be family members or simply other interested members from the community. Given that AERD is more prevalent in females than males at a 3:1 ratio, we expected a similar distribution among survey respondents. However, the proportion of female respondents to our survey was higher than expected ($88\%$). This may be explained by the fact that women have been found to use the internet more often than men to search health-related information and in particular, women use health forums and blogs more often than men [16]. Our survey was posted on the AERD Samter’s Triad Support Group on Facebook and it is possible that the majority of the members of the group are female.
## Conclusions
The findings of this study allow us to conclude that medical ID use in the AERD patient population is low. Further, patients lack the knowledge regarding the importance and advantages of using medical identification. The results of this study could be used for future research to improve patient safety and to evaluate the real impact and cost-effectiveness in the care of those with AERD.
## Supplementary Information
Additional file 1. Survey
## References
1. Kowalski ML, Agache I, Bavbek S. **Diagnosis and management of NSAID-Exacerbated Respiratory Disease (N-ERD)-a EAACI position paper**. *Allergy* (2019.0) **74** 28-39. DOI: 10.1111/all.13599
2. Szczeklik A, Nizankowska E, Sanak M, Swierczynska M. **Aspirin-induced rhinitis and asthma**. *Curr Opin Allergy Clin Immunol* (2001.0) **1** 27-33. DOI: 10.1097/01.all.0000010981.95396.58
3. Berges-Gimeno MP, Simon RA, Stevenson DD. **The natural history and clinical characteristics of aspirin-exacerbated respiratory disease**. *Ann Allergy Asthma Immunol* (2002.0) **89** 474-478. DOI: 10.1016/s1081-1206(10)62084-4
4. Kiladejo A, Palumbo M, Laidlaw TM. **Accidental ingestion of aspirin and nonsteroidal anti-inflammatory drugs is common in patients with aspirin-exacerbated respiratory disease**. *J Allergy Clin Immunol Pract* (2019.0) **7** 1656-1658.e2. DOI: 10.1016/j.jaip.2018.11.017
5. Savino PB, Sporer KA, Barger JA. **Chest pain of suspected cardiac origin: current evidence-based recommendations for prehospital care**. *West J Emerg Med* (2015.0) **16** 983-995. DOI: 10.5811/westjem.2015.8.27971
6. Zideman DA, Singletary EM, De Buck ED. **Part 9: first aid: 2015 international consensus on first aid science with treatment recommendations**. *Resuscitation* (2015.0) **95** e225-e261. DOI: 10.1016/j.resuscitation.2015.07.047
7. Morton L, Murad S, Omar RZ, Taylor K. **Importance of emergency identification schemes**. *Emerg Med J* (2002.0) **19** 584-586. DOI: 10.1136/emj.19.6.584
8. Alqabasani M, Alkherayf N, Lasso A, Kilty S. **Medical identification use in patients with aspirin-exacerbated respiratory disease**. *Eur J Rhinol Allerg* (2022.0) **5** 31-34. DOI: 10.5152/ejra.2022.22010
9. 9.Limesurvey GmbH. LimeSurvey: An Open Source survey tool. LimeSurvey GmbH. Hamburg, Germany. https://www.limesurvey.org
10. Sultan S, Bungener C, Andronikof A. **Individual psychology of risk-taking behaviours in non-adherence**. *J Risk Res* (2002.0) **5** 137-145. DOI: 10.1080/13669870110099802
11. Zhang NJ, Terry A, McHorney CA. **Impact of health literacy on medication adherence: a systematic review and meta-analysis**. *Ann Pharmacother* (2014.0) **48** 741-751. DOI: 10.1177/1060028014526562
12. Mantwill S, Schulz PJ. **Low health literacy associated with higher medication costs in patients with type 2 diabetes mellitus: Evidence from matched survey and health insurance data**. *Patient Educ Couns* (2015.0). DOI: 10.1016/j.pec.2015.07.006
13. Sokol MC, McGuigan KA, Verbrugge RR, Epstein RS. **Impact of medication adherence on hospitalization risk and healthcare cost**. *Med Care* (2005.0) **43** 521-530. DOI: 10.1097/01.mlr.0000163641.86870.af
14. Trichopoulou A, Costacou T, Bamia C, Trichopoulos D. **Adherence to a Mediterranean diet and survival in a Greek population**. *N Engl J Med* (2003.0) **348** 2599-2608. DOI: 10.1056/NEJMoa025039
15. Le T-TA, Foreman C, Smith WB. **The use of medical alert jewelry to communicate allergy information**. *J Allergy Clin Immunol. In Practice* (2019.0) **7** 1083-1085. DOI: 10.1016/j.jaip.2018.08.025
16. Bidmon S, Terlutter R. **Gender differences in searching for health information on the internet and the virtual patient-physician relationship in Germany: exploratory results on how men and women differ and why**. *J Med Internet Res* (2015.0) **17** e156. DOI: 10.2196/jmir.4127
|
---
title: 'Neutral effects of SGLT2 inhibitors in acute coronary syndromes, peripheral
arterial occlusive disease, or ischemic stroke: a meta-analysis of randomized controlled
trials'
authors:
- Pei-Chien Tsai
- Wei-Jung Chuang
- Albert Min-Shan Ko
- Jui-Shuan Chen
- Cheng-Hsun Chiu
- Chun-Han Chen
- Yung-Hsin Yeh
journal: Cardiovascular Diabetology
year: 2023
pmcid: PMC10012509
doi: 10.1186/s12933-023-01789-5
license: CC BY 4.0
---
# Neutral effects of SGLT2 inhibitors in acute coronary syndromes, peripheral arterial occlusive disease, or ischemic stroke: a meta-analysis of randomized controlled trials
## Abstract
### Background
Patients with type 2 diabetes are at increased risk for cardiovascular diseases. Sodium-glucose transport 2 inhibitors (SGLT2i) have been shown to enhance cardiovascular health since their debut as a second-line therapy for diabetes. Acute coronary syndrome (ACS), peripheral arterial occlusive disease (PAOD), and ischemic stroke (IS) are types of atherosclerotic cardiovascular disease (ASCVD), although the benefits of treating these disorders have not been shown consistently.
### Methods
We searched four databases (PubMed, Embase, the Cochrane library, and clinicaltrial.gov) for randomized clinical trials (RCTs) until November of 2022. Comparisons were made between SGLT2i-treated and control individuals with type 2 diabetes. Primary outcomes were ACS, PAOD, and IS; secondary outcomes included cardiovascular mortality and all-cause mortality. Risk ratio (RR) and $95\%$ confidence intervals (CI) were determined using a fixed effects model. Cochrane's risk-of-bias (RoB2) instrument was used to assess the validity of each study that met the inclusion criteria.
### Results
We enrolled 79,504 patients with type 2 diabetes from 43 RCTs. There was no difference in the risk of ACS (RR = 0.97, $95\%$ CI 0.89–1.05), PAOD (RR = 0.98, $95\%$ CI 0.78–1.24), or IS (RR = 0.95, $95\%$ CI 0.79–1.14) among patients who took an SGLT2i compared to those who took a placebo or oral hypoglycemic drugs. Subgroup analysis revealed that none of the SGLT2i treatments (canagliflozin, dapagliflozin, empagliflozin, and ertugliflozin) significantly altered outcomes when analyzed separately. Consistent with prior findings, SGLT2i reduced the risk of cardiovascular mortality (RR = 0.85, $95\%$ CI 0.77–0.93) and all-cause mortality (RR = 0.88, $95\%$ CI 0.82–0.94).
### Conclusion
Our results appear to contradict the mainstream concepts regarding the cardiovascular effects of SGLT2i since we found no significant therapeutic benefits in SGLT2i to reduce the incidence of ACS, PAOD, or IS when compared to placebo or oral hypoglycemic drugs.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12933-023-01789-5.
## Introduction
Treatment for type 2 diabetes should begin with metformin and other lifestyle adjustments, as recommended by the American Diabetes Association [1]. Sodium-glucose transport 2 inhibitors (SGLT2i) and other second-line therapeutic agent combinations may be necessary if first-line treatment fails to bring blood glucose under control. SGLT2 is a sodium-glucose transporter that is found in the S1 segment of the proximal tubule. SGLT2i aids in maintaining healthy blood glucose levels by blocking SGLT2 reabsorption [2]. The four most widely used SGLT2i, canagliflozin, dapagliflozin, empagliflozin, and ertugliflozin, all bind to the SGLT2 protein with varying degrees of affinity [3].
Stable control of blood glucose is just one of the benefits of SGLT2i. It has been reported that among adults with diabetic kidney disease, SGLT2i are associated with reduced risks of major adverse cardiovascular events (MACE), kidney outcomes, hospitalization for heart failure, and death [4, 5]. In addition, SGLT2i decreases systolic blood pressure in patients with heart failure [6], yielding also benefits in patients with heart failure with preserved ejection fraction [7]. SGLT2i act as anti-inflammatory agents by either indirectly improving metabolism and reducing stress conditions or via direct modulation of inflammatory signaling pathways [8]; the direct cardiac effects seem to be mediated by modulation of intracellular sodium concentration via the sodium-interactome [9].
Up to two thirds of patients with type 2 diabetes have atherosclerotic cardiovascular disease (ASCVD) [10], making them less manageable and leading to worse outcomes than the general population [11]. Animal models propose that SGLT2i prevents ASCVD by lowering serum levels of inflammatory factors linked to atherosclerosis, stopping the proliferation and migration of vascular smooth muscle cells (VSMCs), blocking foam cell formation, preventing platelet activation, and improving autophagy impairment [12], but human clinical data is less conclusive. To further understand the relationship between SGLT2i cardiovascular impact and ASCVD events, especially ACS, PAOD, IS, and mortality outcomes in individuals with type 2 diabetes, we conducted a meta-analysis.
## Database sources and search strategy
This study followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analysis) guidelines [13] and the Cochrane Handbook (Version 6.1) [14] in terms of its methodology, including data sources, inclusion and exclusion criteria, outcome assessment, quality assessment, and use of statistical methods. Four international databases (PubMed, Embase, Cochrane Library, and ClinicalTrials.gov) were searched. Terminology used to described “type 2 diabetes”, “sodium-glucose cotransporter-2 inhibitors,” and terms relevant to “acute coronary syndrome,” “peripheral arterial occlusive disease,” “ischemic stroke,” “cardiovascular mortality” and “all-cause mortality” were searched in the databases. The database search algorithm is provided in Additional file 1. The data collection workflow is shown in Fig. 1. The last search time was conducted in November 2022. In the first phase of the literature search, we retrieved a total of 729 records (292 from databases and 437 from registries), after removing 241 duplicates, we screened 488 records; then, 426 records were excluded based on the exclusion criteria for this study; finally, we retrieved 62 records and aseesed their eligibility, and we ended up including 43 studies. Fig. 1Study workflow of finding and including literature
## Inclusion and exclusion criteria
Our inclusion criteria were studies reported in English, have a comprehensive documentation of their outcomes, and patients with type 2 diabetes who were 18 or older. The exclusion criteria were studies involving patients with type 1 diabetes or malignancies, letters to the editor, editorials, case reports, review articles, and literature based on animal model. As glucagon-like peptide-1 (GLP-1) is an effective treatment for managing blood glucose levels, we also ruled out trials in which GLP-1 drugs were used as a control. Our control group was defined as those receiving placebo or active therapy using oral hypoglycemic drugs. In order to analyze the occurrence of adverse events in a larger pool of patients with type 2 diabetes, the placebo- and active-controlled trials were merged.
Two independent reviewers (WJC and RXC) were involved in the literature search and citation eligibility review, while a third reviewer (CHC) crossed-checked all eligible references. Final eligibility of references was determined by two senior authors (PCT and YYH). YYH carefully reviewed the definition of results and the use of SGLT2i in each study.
## Outcome measures
The primary outcome was the incidence of ACS (defined as acute myocardial infarction and/or unstable angina), PAOD (not including other related events such as peripheral artery ischemia or peripheral artery embolism), and IS. Secondary outcomes were cardiovascular mortality and all-cause mortality. The incidence of adverse events was retrieved from the clinicaltrials.gov registry and the published studies.
## Data extraction and quality assessment
Primary and secondary outcomes, study characteristics (sample size, trial name, ClinicalTrials.gov identifier), treatment details (dose, follow-up duration, protocol), patient characteristics (age, sex) were extracted from all included studies. Different doses of the same drug were pooled into one treatment group. To avoid duplication from the same population, only the most recent randomized controlled trials (RCTs) with the largest sample size were considered. Methodological quality was assessed using version 2 of the Cochrane risk-of-bias tool for randomized trials (RoB2) [15]. Selective reporting, random sequence generation, other sources of bias, incomplete outcome data, blinded outcome assessment, and allocation concealment were all identified as potential sources of bias. Each risk of bias category was rated as low, high, and unclear. The RoB 2 was used to evaluate the reliability of the evidence. Study selection, data extraction, and quality assessment of data extraction were carried out by three independent reviewers (WJC, RXC, and CHC). The data gathered from the publications was analyzed for potential bias by two additional researchers (PCT and YYH), who discussed their contrasting findings until a consensus is reached.
## Statistical analysis
Funnel plots and Egger’s test [16] were used to look for signs of publication bias. Pooled risk ratio (RR) and $95\%$ confidence intervals (CI) were used to analyze the incidence of ACS, PAOD, and IS patients receiving conventional therapy with SGLT2i or oral hypoglycemic drugs. The Higgins and Thompson I2 statistic and the Cochrane Q test were used to analyze the degree of study heterogeneity. The level of statistical significance for the Q test was set at a P-value < 0.1. In principle, if I2 was greater than $50\%$, a random-effects model was used for meta-analysis, otherwise, a fixed effects model was used. In order to better account for any clinical background heterogeneity, we also adopted a random effects model for all analyses in the study. We ran subgroup analyses to look at the impact of individual types of SGLT2i treatment, because different SGLT2i treatments might produce different outcomes. In all analyses, P-values < 0.05 (two-sided) were considered statistically significant. All analyses and visualizations were generated with R version 4.1.2 and R packages meta and dmetar. Trial sequential analysis (TSA) was used to estimate the required sample size to reach $80\%$ study power based on the incidence rate in control group and the relative risk reduction rate obtained from each meta-analysis [17]. The parameters of the study were calculated to be an alpha level of 0.05 and a power of $80\%$. TSA was performed using TSA 0.9.5.10 Beta.
## Results
Figure 1 shows the final 43 RCTs with a total of 79,504 patients with type 2 diabetes; 48,568 patients received SGLT2i in combination with background treatment, whereas 30,936 patients used placebo or oral hypoglycemic drugs. Four of the 43 studies have not yet been published. In terms of the types of SGLT2i treatments, these were canagliflozin (10 studies), dapagliflozin (15 studies), empagliflozin (15 studies), and ertugliflozin (6 studies).
## Baseline characteristics of included studies
Table 1 shows the baseline characteristics of the included studies. All eligible RCTs were published between 2010 and 2020. Median follow-up time was 1.9 years, sample size ranged from 218 to 17,143 participants, and $21.4\%$ to $54.5\%$ were female. First-line medications most often used to treat diabetes were metformin ($58.1\%$), sulfonylurea ($20.9\%$), and insulin ($18.6\%$). The risk of bias in the 43 studies is shown in Additional file 2. Incomplete data on ClinicalTrials.gov means that there may be minor problems with certain studies, and around half of the studies had a low risk of bias (see Additional file 3). Five of the 43 studies contain evidence of atherosclerotic cardiovascular disease (ASCVD) at baseline, such as associated high cardiovascular risk [18–20], cerebrovascular disease or high blood pressure [21], and atherosclerosis in the coronary, cerebral, or peripheral vascular systems [22]. Subgroup analyses were used to examine the effect of ASCVD history or evidence on the overall results. Subgroup analyses show that the incidence rates of ACS, PAOD, and IS are similar across the two subgroups (5 vs. 38 studies), and that the risk ratios are consistent with the overall results of this study (see Additional file 4).Table 1Characteristics of the included randomized clinical trials in this studyStudyNumber of patients (M/F)Mean age (SD)InterventionsBackground therapyTreatmentControlTreatmentControlLavalle-González et al., 2013 [33]1101 ($\frac{705}{396}$)549 ($\frac{266}{283}$)55.4 (9.3)54.7 (9.7)Canagliflozin ($\frac{100}{300}$ mg)/Placebo and SitagliptinMetforminCefalu et al., 2013 [34]968 ($\frac{493}{475}$)482 ($\frac{238}{244}$)58.9 (9.4)56.3 (9.0)Canagliflozin ($\frac{100}{300}$ mg)/GlimepirideMetforminNCT01106690, 2013227 ($\frac{140}{87}$)115 ($\frac{76}{39}$)56.9 (10.3)58.3 (9.6)Canagliflozin ($\frac{100}{300}$ mg)/Placebo and SitagliptinMetformin and pioglitazoneYale et al., 2014 [35]179 ($\frac{106}{73}$)90 ($\frac{57}{33}$)68.7 (8.2)68.2 (8.4)Canagliflozin ($\frac{100}{300}$ mg)/PlaceboAccordance with local guidelinesNeal et al., 2015 [20]2886 ($\frac{1903}{983}$)1441 ($\frac{955}{486}$)62.2 (8.1)62.3 (7.9)Canagliflozin ($\frac{100}{300}$ mg)/PlaceboSulfonylureaBode et al., 2015 [36]477 ($\frac{253}{224}$)237 ($\frac{143}{94}$)64.3 (6.3)63.2 (6.2)Canagliflozin ($\frac{100}{300}$ mg)/PlaceboStable antihyperglycemic (AHA) regimenRosenstock et al., 2016 [37]949 ($\frac{453}{496}$)237 ($\frac{116}{121}$)54.9 (9.9)55.2 (9.8)Canagliflozin ($\frac{100}{300}$ mg)/MetforminMetforminNCT01989754, 20172904 ($\frac{1851}{1053}$)2903 ($\frac{1792}{1111}$)63.9 (8.4)64 (8.3)Canagliflozin (100 mg 13 weeks then 300 mg)/PlaceboAccordance with local guidelinesPerkovic et al., 2019 [23]2200 ($\frac{1438}{762}$)2197 ($\frac{1465}{732}$)62.9 (9.2)63.2 (9.2)Canagliflozin (100 mg)/ PlaceboAccordance with local guidelinesLingvay et al., 2019 [38]394 ($\frac{201}{193}$)392 ($\frac{221}{171}$)57.5 (10.7)55.7 (11.1)Canagliflozin (100 mg 13 weeks then 300 mg)/SemaglutideMetforminNauck et al., 2011 [39]406 ($\frac{227}{179}$)408 ($\frac{227}{181}$)58.1 (9.4)58.6 (9.8)Dapagliflozin (not mentioned)/GlipizideMetforminStrojek et al., 2011 [40]450 ($\frac{217}{233}$)146 ($\frac{72}{74}$)59.7 (9.4)60.3 (10.2)Dapagliflozin ($\frac{2.5}{5}$/12 mg)/PlaceboGlimepirideHenry et al., 2012 [41]827 ($\frac{657}{170}$)409 ($\frac{314}{95}$)51.5 (10.3)52.3 (10.1)Dapagliflozin ($\frac{5}{10}$ mg)/MetforminMetforminBailey et al., 2013 [42]409 ($\frac{216}{193}$)137 ($\frac{76}{61}$)54 (NA)53.7 (NA)Dapagliflozin ($\frac{2.5}{5}$/10 mg)/PlaceboMetforminLeiter et al., 2014 [43]482 ($\frac{323}{159}$)483 ($\frac{324}{159}$)63.9 (7.6)63.6 (7.0)Dapagliflozin (10 mg)/PlaceboUsual careNCT01137474, 2014633 ($\frac{358}{275}$)311 ($\frac{171}{140}$)NA (NA)NA (NA)Dapagliflozin ($\frac{2.5}{5}$/10 mg)/PlaceboOAD with or without insulinWilding et al., 2014 [44]610 ($\frac{290}{320}$)197 ($\frac{99}{98}$)59.8 (8.1)58.8 (8.6)Dapagliflozin ($\frac{2.5}{5}$/10 mg)/PlaceboInsulinCefalu et al., 2015 [21]460 ($\frac{314}{146}$)462 ($\frac{318}{144}$)62.8 (7.0)63 (7.7)Dapagliflozin (10 mg)/PlaceboStable background treatment except rosiglitazoneBailey et al., 2015 [45]410 ($\frac{198}{212}$)75 ($\frac{31}{44}$)NA (NA)52.7 (10.3)Dapagliflozin ($\frac{2.5}{5}$/11 mg)/PlaceboMetforminMatthaei et al., 2015 [46]109 ($\frac{47}{62}$)109 ($\frac{61}{48}$)61.1 (9.7)60.9 (9.2)Dapagliflozin (10 mg)/PlaceboMetformin and sulfonylureaMüller-Wieland et al., 2018 [47]313 ($\frac{201}{112}$)312 ($\frac{207}{105}$)57.4 (9.4)58.6 (8.4)Dapagliflozin (13 mg)/GlimepirideMetforminScott et al., 2018 [48]306 ($\frac{186}{120}$)307 ($\frac{169}{138}$)66.6 (8.6)67.7 (8.5)Dapagliflozin (5 mg titrated to 10 mg)/*Sitagliptin plus* Placebo DapagliflozinMetformin with or without sulfonylureaFioretto et al., 2018 [49]160 ($\frac{91}{69}$)161 ($\frac{91}{70}$)65.3 (6.2)66.2 (6.5)Dapagliflozin (12 mg)/PlaceboInsulin, metformin, sulfonylurea or TZDYang et al., 2018 [50]139 ($\frac{66}{73}$)133 ($\frac{64}{69}$)56.5 (8.4)58.6 (8.9)Dapagliflozin (10 mg)/PlaceboInsulinWiviott et al., 2019 [19]8574 ($\frac{5403}{3171}$)8569 ($\frac{5319}{3250}$)63.9 (6.8)64 (6.8)Dapagliflozin (10 mg)/PlaceboCurrent background therapyHäring et al., 2013 [51]1042 ($\frac{568}{474}$)431 ($\frac{227}{204}$)55.4 (9.9)56 (9.7)Empagliflozin ($\frac{10}{25}$ mg)/PlaceboMetformin or sulfonylureaFerrannini et al., 2013 [52]547 ($\frac{277}{270}$)112 ($\frac{57}{55}$)58.9 (8.6)57.6 (9.8)Empagliflozin (10 mg)/Sitagliptin and MetforminMetforminBarnett et al., 2014 [53]419 ($\frac{249}{170}$)319 ($\frac{181}{138}$)63.7 (8.9)64.1 (8.7)Empagliflozin ($\frac{10}{25}$ mg)/PlaceboMetformin, insulin or sulfonylureaRosenstock et al., 2014 [54]375 ($\frac{181}{194}$)188 ($\frac{75}{113}$)57.4 (9.1)55.3 (10.1)Empagliflozin ($\frac{10}{25}$ mg)/Placebo Empagliflozin 10 mg plus Placebo Empagliflozin 25 mgInsulin or metforminZinman et al., 2015 [18]4687 ($\frac{3336}{1351}$)2333 ($\frac{1680}{653}$)63 (8.6)63.2 (8.8)Empagliflozin ($\frac{10}{25}$ mg)/PlaceboCurrent background therapyRoden et al., 2015 [55]1325 ($\frac{765}{560}$)877 ($\frac{486}{391}$)56 (10.3)55.7 (10.0)Empagliflozin ($\frac{10}{25}$ mg)/Placebo and SitagliptinMetformin, sulfonylureasNCT01649297, 2015876 ($\frac{483}{393}$)107 ($\frac{55}{52}$)57.6 (10.2)57.9 (11.2)Empagliflozin ($\frac{5}{12.5}$ mg BID, $\frac{10}{25}$ mg QD)/ PlaceboMetforminRosenstock et al., 2015 [56]324 ($\frac{186}{138}$)170 ($\frac{90}{80}$)59.2 (10.2)58.1 (9.4)Empagliflozin ($\frac{10}{25}$ mg)/PlaceboInsulinAraki et al., 2015 [57]273 ($\frac{195}{78}$)63 ($\frac{47}{16}$)61.6 (9.8)60 (10.2)Empagliflozin ($\frac{10}{25}$ mg)/Metformin and SulfonylureaSulfonylurea, biguanide, TZD, AGI or DPP-4Hadjadj et al., 2016 [58]1019 ($\frac{593}{426}$)341 ($\frac{187}{154}$)52.6 (11.0)52.5 (10.9)Empagliflozin ($\frac{5}{12.5}$ mg BID, $\frac{10}{25}$ mg QD)/ MetforminMetforminRidderstråle et al., 2018 [59]765 ($\frac{432}{333}$)780 ($\frac{421}{359}$)56.2 (10.3)55.7 (10.4)Empagliflozin (25 mg)/*Glimepiride plus* Placebo EmpagliflozinMetforminRodbard et al., 2019 [60]410 ($\frac{209}{201}$)411 ($\frac{206}{205}$)58 (10.0)57 (10.0)Empagliflozin (25 mg)/SemaglutideAccordance with local guidelinesPratley et al., 2018 [61]498 ($\frac{261}{237}$)247 ($\frac{154}{93}$)55.1 (9.8)54.8 (10.7)Ertugliflozin ($\frac{5}{15}$ mg)/SitagliptinMetforminRosenstock et al., 2018 [62]412 ($\frac{190}{222}$)209 ($\frac{98}{111}$)56.7 (8.8)56.5 (8.7)Ertugliflozin ($\frac{5}{15}$ mg)/*Placebo plus* GlimepirideGlimepiride or insulinGrunberger et al., 2018 [63]313 ($\frac{159}{154}$)154 ($\frac{72}{82}$)67.1 (8.4)67.5 (8.9)Ertugliflozin ($\frac{5}{15}$ mg)/PlaceboException of metformin, rosiglitazone, and other SGLT2 inhibitorsDagogo-Jack et al., 2018 [64]309 ($\frac{163}{146}$)153 ($\frac{100}{53}$)59.4 (9.0)58.3 (9.2)Ertugliflozin ($\frac{5}{15}$ mg)/PlaceboMetformin or sitagliptinHollander et al., 2019 [65]880 ($\frac{409}{471}$)435 ($\frac{222}{213}$)58.4 (9.8)57.8 (9.2)Ertugliflozin ($\frac{5}{15}$ mg)/GlimepirideMetforminCannon et al., 2020 [22]5493 ($\frac{3860}{1633}$)2745 ($\frac{1901}{844}$)64.4 (8.1)64.4 (8.0)Ertugliflozin ($\frac{5}{15}$ mg)/PlaceboInsulin, metformin and sulfonylurea
## Effect of SGLT2i on acute coronary syndrome
A forest plot comparing the SGLT2i treatment for ACS to that of control is shown in Fig. 2a. Thirty-five studies, including canagliflozin (7 studies), dapagliflozin (12 studies), empagliflozin (10 studies), and ertugliflozin (6 studies) reported ACS as an adverse event, with 41,881 individuals in the SGLT2i group and 29,356 individuals in the control group, and an incidence of $3.10\%$ in the SGLT2i group compared to $3.45\%$ in the control group. There was no significant heterogeneity across the studies (I2 = $0\%$, $$P \leq 0.82$$ for the Q test) (Fig. 2a), and the overall risk ratio was not significant (RR = 0.97, $95\%$ CI: 0.89–1.05). ACS did not differ significantly between the four SGLT2i medication groups and the control group: canagliflozin (RR = 0.97, $95\%$ CI 0.79–1.19), dapagliflozin (RR = 1.05, $95\%$ CI 0.93–1.18), empagliflozin (RR = 0.88, $95\%$ CI 0.72–1.07), and ertugliflozin (RR = 0.85, $95\%$ CI 0.71–1.03). A TSA was conducted using 34 studies with a total of 70,612 patients, a control group incidence rate of $3.45\%$ and a relative risk reduction of $9.88\%$ (Fig. 2b). Sequential analysis of trials indicates a sample size of 86,159 is needed to achieve $80\%$ power, and there are not enough samples and effects for the cumulative Z-curve to pass the trial sequential monitoring boundaries. Fig. 2a Forest plot, and b Trial Sequential Analysis of effects of SGLT2i on acute coronary syndrome
## Effect of SGLT2i on peripheral arterial occlusive disease
A forest plot depicting the SGLT2i treatment for PAOD to that of control is shown in Fig. 3. Twenty studies, including canagliflozin (5 studies), dapagliflozin (4 studies), empagliflozin (9 studies), and ertugliflozin (2 studies) reported PAOD as an adverse event, with a total of 34,972 individuals in the SGLT2i group and 24,980 individuals in the control group, and an incidence of $0.55\%$ in the SGLT2i group compared to $0.51\%$ in the control group. There was no significant heterogeneity across the studies (I2 = $0\%$, $$P \leq 0.90$$ for the Q test) (Fig. 3) and the risk ratio was not significant (RR = 0.98, $95\%$ CI 0.78–1.24). Subgroup analysis revealed that PAOD did not differ between the four SGLT2i medication groups: canagliflozin (RR = 1.18, $95\%$ CI: 0.70–1.99), dapagliflozin (RR = 0.86, $95\%$ CI 0.58–1.27), empagliflozin (RR = 1.16, $95\%$ CI 0.75–1.79), and ertugliflozin (RR = 0.83, $95\%$ CI 0.49–1.40). A TSA was conducted using 20 studies with a total of 59,952 patients, a control group incidence of $0.51\%$, with a $7.27\%$ decrease in relative risk for those who took preventative measures. To make any conclusions from the sequential analysis of trials, the sample size must be far larger than 59,952 in order to detect the relative risk reduction rate of $7.27\%$ for peripheral arterial occlusive disease in the SGLT2i group compared with the control group with $80\%$ power. Analysis shows that there are not enough samples and effects for the cumulative Z-curve to approach the trial sequential monitoring boundaries if the setting is set at $80\%$ power. Fig. 3Forest plot of effects of SGLT2i on peripheral arterial occlusive disease
## Effect of SGLT2i on ischemic stroke
A forest plot is used to show how often SGLT2i causes IS compared to the controls (Fig. 4). IS was identified as an adverse event in 23 trials, including canagliflozin (6 studies), dapagliflozin (7 studies), empagliflozin (6 studies), and ertugliflozin (4 studies), with 36,417 individuals in the SGLT2i group and 26,123 individuals in the control group, and an incidence of $0.71\%$ in the SGLT2i group compared to $0.77\%$ in the control group. There was no significant heterogeneity across the studies (I2 = $0\%$, $$P \leq 0.96$$ for the Q test) (Fig. 4a). The overall risk ratio was not significant (RR = 0.95, $95\%$ CI 0.79–1.14). Subgroup analysis revealed that IS did not differ significantly between the four SGLT2i medication groups and the control group: canagliflozin (RR = 1.06, $95\%$ CI 0.67–1.67), dapagliflozin (RR = 1.04, $95\%$ CI 0.79–1.37), empagliflozin (RR = 0.86, $95\%$ CI 0.53–1.38), and ertugliflozin (RR = 0.80, $95\%$ CI 0.55–1.16). A TSA was conducted using 23 studies with a total of 62,540 patients, a control group incidence rate of $0.77\%$ and a relative risk reduction of $7.79\%$. To detect a $7.79\%$ decrease in the risk of IS in the SGLT2i group compared with the control group and achieve $80\%$ power, a sample size of more than 62,540 is needed. Analysis shows that there are not enough samples and effects for the cumulative Z-curve to approach the trial sequential monitoring boundaries if the setting is set at $80\%$ power. Fig. 4Forest plot of effects of SGLT2i on ischemic stroke
## Effect of SGLT2i on cardiovascular mortality and all-cause mortality
Cardiovascular mortality was reported as an adverse event in 23 studies, including canagliflozin (5 studies), dapagliflozin (8 studies), empagliflozin (5 studies), and ertugliflozin (5 study), with a total of 33,634 individuals in the SGLT2i group and 23,130 individuals in the control group, and an incidence of $2.61\%$ in the SGLT2i group compared to $3.10\%$ in the control group. The included studies exhibited no heterogeneity (I2 = $32\%$, $$P \leq 0.11$$ for the Q test) (Fig. 5a). The overall risk ratio was significant (RR = 0.85, $95\%$ CI 0.77–0.93). In the subgroup analysis, with the exception of canagliflozin (RR = 0.76, $95\%$ CI 0.60–0.97) and empagliflozin that had lower risk ratios (RR = 0.62, $95\%$ CI 0.50–0.78), dapagliflozin (RR = 0.98, $95\%$ CI 0.83–1.17) and ertugliflozin (RR = 0.92, $95\%$ CI 0.78–1.10) does not show a benefit for cardiovascular mortality (Fig. 5a). A TSA was conducted using 17 studies, a control group incidence rate of $3.1\%$ and a relative risk reduction of $15.81\%$ (Fig. 5b). Sequential analysis of trials indicates a sample size of 116,947 is required to reach $80\%$ power. Although the included samples size was 53,379, the cumulative Z-curve already surpassed trial sequential monitoring boundaries, providing statistical power for the considerable protective impact of SGLT2i on cardiovascular mortality. Fig. 5a Forest plot, and b Trial Sequential Analysis of effects of SGLT2i on cardiovascular mortality in 23 studies There was a total of 38 studies, including canagliflozin (9 studies), dapagliflozin (13 studies), empagliflozin (10 studies), and ertugliflozin (6 studies) that looked at all-cause mortality in adverse events, with a total of 42,665 individuals in the SGLT2i group and 29,472 individuals in the control group, and an incidence of $3.98\%$ in the SGLT2i group compared to $4.73\%$ in the control group. There was no significant heterogeneity across the studies (I2 = $0\%$, $$P \leq 0.61$$ for the Q test) (Fig. 6a). The overall risk ratio was significant (RR = 0.88; $95\%$ CI 0.82–0.94). Subgroup analysis also showed that, with the exception empagliflozin (RR = 0.69; $95\%$ CI: 0.58–0.83) had lower risk of all-cause mortality, canagliflozin (RR = 0.88; $95\%$ CI 0.76–1.01), dapagliflozin (RR = 0.93; $95\%$ CI 0.83–1.05), and ertugliflozin (RR = 0.94; $95\%$ CI 0.81–1.08) showed similar effect to control. A TSA was conducted using 30 studies with a total of 63,108 patients, a control group incidence rate of $4.73\%$ and a relative risk reduction of $15.86\%$ (Fig. 6b). Sequential analysis of trials indicates a sample size of 23,249 is required to reach $80\%$ power, which is satisfied by the included studies, thus the cumulative Z-curve reached the trial sequential monitoring boundaries, demonstrating the significant protective effect of SGLT2i on all-cause mortality. Fig. 6a Forest plot, and b Trial Sequential Analysis of effects of SGLT2i on all-cause mortality in 38 studies
## Publication bias
Egger tests showed no publication bias, and the distribution of publications on the funnel plots for each meta-analysis was symmetrical (Additional file 5), suggesting that publication bias in this study is unlikely. Furthermore, when only papers with low risk of bias were included in the analysis, the outcomes of this study remained unaffected.
## Literature quality assessment
Additional File 3 shows the quality assessment figure for the 43 studies in this meta-analysis. The potential for bias was broken down into its constituent parts, beginning with the risk of bias associated with the randomization procedure, followed by those associated with deviation from the intended intervention, missing outcomes, the way the outcome was measured, selective reporting, and the overall risk of bias. We found over half of the studies have a low risk of bias, while just a few having a high risk of bias.
## Discussion
In this meta-analysis of 43 RCTs to compare the effectiveness of SGLT2i treatment in reducing the risk of ACS, PAOD, and IS in a total of 79,502 patients with type 2 diabetes (48,568 used SGLT2i treatment and 30,936 used placebo or oral hypoglycemic drugs). For each drug, we covered large RCTs, such as CANVAS (NCT01032629, 4330 people) [20], CANVAS-R (NCT01989754, 5812 people), CREDENCE (NCT02065791, 4401 people) [23] for canagliflozin; DECLARE-TIMI58 (NCT01730534, 17,160 people) [19] for dapagliflozin; EMPA-REG (NCT01131676, 7088 people) [18] for empagliflozin; and MK-8835-004 (NCT01986881, 8246 people) [22] for ertugliflozin. There was no significant difference in the risk of these three diseases between the SGLT2i group and the control group. We examined cardiovascular mortality and all-cause mortality and found that they were lower in the group using SGLT2i, which is consistent with previous studies [24–27], lending support to this study’s validity and indicating that there was no bias in the included trials.
## Implications for SGLT2i in acute coronary syndrome
Our results showed that the use of SGLT2i did not significantly change the incidence of ACS. This is in contrast to a 2017 network meta-analysis by Lee et al. that found SGLT2i to significantly decrease the incidence of ACS compared with the placebo group ($$n = 6606$$, RR = 0.50, $95\%$ CI 0.29–0.86)[28], but no significant differences when compared to metformin ($$n = 1434$$, RR = 0.66, $95\%$ CI = 0.08–5.64) or sulfonylurea ($$n = 2264$$, RR = 0.58, $95\%$ CI 0.17–1.97). Nevertheless, our results vary from those of Lee et al., because four large-scale RCTs conducted after 2017 ($$n = 35$$,614) were left out of their analysis.
## Implications for SGLT2i on peripheral arterial occlusive disease
A 2021 meta-analysis by Lin et al. ( $$n = 65$$,131) found an increased risk of developing peripheral arterial disease (PAD) in patients using SGLT2i hypoglycemic medications (OR = 1.21, $95\%$ CI 1.03–1.42), particularly in patients with canagliflozin (OR = 1.53, $95\%$ CI 1.14–2.05)[29]. Our results showed that the use of SGLT2i did not significantly change the incidence of PAOD in diabetic patients. The key difference between our definition of PAOD and the Lin et al. study’s definition of PAD is that the latter includes 17 specific terms to better explain amputation and diabetic foot-related PAD. In addition, individuals with type 1 diabetes were included in the Lin et al. study, while our emphasis was on those with type 2. Consistent with our findings, another meta-analysis conducted in 2021 by Liao et al. ( $$n = 59$$,692) indicated that SGLT2i had no effect on PAOD (RR = 1.03, $95\%$ CI 0.75–1.25)[30]. However, sample sizes under 1000 people were not analyzed in Liao et al. study and they included patients other than type 2 diabetes.
## Implications for SGLT2i on ischemic stroke
Zhou et al. conducted a meta-analysis in 2021 ($$n = 38$$,723) and showed that the usage of SGLT2i did not substantially change the incidence of IS (RR = 1.04, $95\%$ CI 0.92–1.18) [31], which is in line with our results on IS. Tsai et al. ’s meta-analysis ($$n = 46$$,969) from 2021 also reported no significant difference for IS (RR = 0.99, $95\%$ CI 0.89–1.12) [32].
There are limitations to this study. Even if heterogeneity is absent (I2 < $50\%$), baseline variations in clinical settings, age, follow-up and disease duration may bias the results. We provide both the fixed and random effects model to allow for the clinical heterogeneity seen across several studies. The random effects model produces findings that are comparable to those obtained using the fixed effects model. Second, in order to determine the total occurrences connected to our outcomes, we assumed that studies would use the same definitions of adverse events and inclusion and exclusion criteria. Third, there may be discrepancies in the number of cases reported for the same disease on ClinicalTrials.gov across funding organizations. We may have under-estimated the actual incidence since we applied a strict disease definition to prevent multiple-counting of the same patient. Advantages of our meta-analysis include the fact that it is one of the few to include ACS, PAOD, and IS, as well as the fact that we included both small and unpublished studies found on ClinicalTrials.gov.
## Conclusion
Our meta-analysis of RCTs through November 2022 shows SGLT2i use was associated with a reduction in cardiovascular mortality and all-cause mortality that are consistent with previous research. However, contrary to notions about the cardiovascular effects of SGLT2i, people with diabetes who are treated with these drugs do not have a significantly decreased chance of developing ACS, PAOD, or IS compared to the controls. There is currently not enough data for their meta-analysis to be statistically significant. This may be because of the low incidence of disease in the control group and the modest relative risk reduction for SGLT2i treatment.
## Supplementary Information
Additional file 1. Database search algorithm. Additional file 2. Cochrane risk-of-bias tool (RoB2) used to assess the quality of a study. Additional file 3. Summary of overall risk of biases in a study. Additional file 4. Subgroup analysis of studies separated by the presence or absence of atherosclerotic cardiovascular disease at baseline. Additional file 5. Funnel plots of publication bias for (a) acute coronary syndrome, (b) peripheral arterial occlusive disease, (c) ischemic stroke, (d) cardiovascular mortality, and (e) all-cause mortality.
## References
1. **9. Pharmacologic approaches to glycemic treatment standards of medical care in diabetes-2022**. *Diabetes Care* (2022.0) **45** S125-S143. DOI: 10.2337/dc22-S009
2. DeFronzo RA, Norton L, Abdul-Ghani M. **Renal, metabolic and cardiovascular considerations of SGLT2 inhibition**. *Nat Rev Nephrol* (2017.0) **13** 11-26. DOI: 10.1038/nrneph.2016.170
3. Tentolouris A, Vlachakis P, Tzeravini E, Eleftheriadou I, Tentolouris N. **SGLT2 inhibitors: a review of their antidiabetic and cardioprotective effects**. *Int J Environ Res Public Health.* (2019.0) **16** 2965. DOI: 10.3390/ijerph16162965
4. Kaze AD, Zhuo M, Kim SC, Patorno E, Paik JM. **Association of SGLT2 inhibitors with cardiovascular, kidney, and safety outcomes among patients with diabetic kidney disease: a meta-analysis**. *Cardiovasc Diabetol* (2022.0) **21** 47. DOI: 10.1186/s12933-022-01476-x
5. Chan GC, Ng JK, Chow KM, Szeto CC. **SGLT2 inhibitors reduce adverse kidney and cardiovascular events in patients with advanced diabetic kidney disease: a population-based propensity score-matched cohort study**. *Diabetes Res Clin Pract* (2023.0) **195** 110200. DOI: 10.1016/j.diabres.2022.110200
6. Li M, Yi T, Fan F, Qiu L, Wang Z, Weng H, Ma W, Zhang Y, Huo Y. **Effect of sodium-glucose cotransporter-2 inhibitors on blood pressure in patients with heart failure: a systematic review and meta-analysis**. *Cardiovasc Diabetol* (2022.0) **21** 139. DOI: 10.1186/s12933-022-01574-w
7. Jhund PS. **SGLT2 inhibitors and heart failure with preserved ejection fraction**. *Heart Fail Clin* (2022.0) **18** 579-586. DOI: 10.1016/j.hfc.2022.03.010
8. Elrakaybi A, Laubner K, Zhou Q, Hug MJ, Seufert J. **Cardiovascular protection by SGLT2 inhibitors—Do anti-inflammatory mechanisms play a role?**. *Mol Metab* (2022.0) **64** 101549. DOI: 10.1016/j.molmet.2022.101549
9. Chen S, Coronel R, Hollmann MW, Weber NC, Zuurbier CJ. **Direct cardiac effects of SGLT2 inhibitors**. *Cardiovasc Diabetol* (2022.0) **21** 45. DOI: 10.1186/s12933-022-01480-1
10. Bancks MP, Ning H, Allen NB, Bertoni AG, Carnethon MR, Correa A, Echouffo-Tcheugui JB, Lange LA, Lloyd-Jones DM, Wilkins JT. **Long-term absolute risk for cardiovascular disease stratified by fasting glucose level**. *Diabetes Care* (2019.0) **42** 457-465. DOI: 10.2337/dc18-1773
11. Nelson AJ, Peterson ED, Pagidipati NJ. **Atherosclerotic cardiovascular disease and heart failure: determinants of risk and outcomes in patients with diabetes**. *Prog Cardiovasc Dis* (2019.0) **62** 306-314. DOI: 10.1016/j.pcad.2019.07.001
12. Liu Z, Ma X, Ilyas I, Zheng X, Luo S, Little PJ, Kamato D, Sahebkar A, Wu W, Weng J. **Impact of sodium glucose cotransporter 2 (SGLT2) inhibitors on atherosclerosis: from pharmacology to pre-clinical and clinical therapeutics**. *Theranostics* (2021.0) **11** 4502-4515. DOI: 10.7150/thno.54498
13. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, Shamseer L, Tetzlaff JM, Akl EA, Brennan SE. **The PRISMA 2020 statement: an updated guideline for reporting systematic reviews**. *Int J Surg* (2021.0) **88** 105906. DOI: 10.1016/j.ijsu.2021.105906
14. 14.Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ. Cochrane Handbook for Systematic Reviews of Interventions version 6.2; 2021.
15. Sterne JAC, Savovic J, Page MJ, Elbers RG, Blencowe NS, Boutron I, Cates CJ, Cheng HY, Corbett MS, Eldridge SM. **RoB 2: a revised tool for assessing risk of bias in randomised trials**. *BMJ* (2019.0) **366** l4898. DOI: 10.1136/bmj.l4898
16. Egger M, Davey Smith G, Schneider M, Minder C. **Bias in meta-analysis detected by a simple, graphical test**. *BMJ* (1997.0) **315** 629-634. DOI: 10.1136/bmj.315.7109.629
17. Wetterslev J, Jakobsen JC, Gluud C. **Trial Sequential Analysis in systematic reviews with meta-analysis**. *BMC Med Res Methodol* (2017.0) **17** 39. DOI: 10.1186/s12874-017-0315-7
18. Zinman B, Wanner C, Lachin JM, Fitchett D, Bluhmki E, Hantel S, Mattheus M, Devins T, Johansen OE, Woerle HJ. **Empagliflozin, cardiovascular outcomes, and mortality in type 2 diabetes**. *N Engl J Med* (2015.0) **373** 2117-2128. DOI: 10.1056/NEJMoa1504720
19. Wiviott SD, Raz I, Bonaca MP, Mosenzon O, Kato ET, Cahn A, Silverman MG, Zelniker TA, Kuder JF, Murphy SA. **Dapagliflozin and cardiovascular outcomes in type 2 diabetes**. *N Engl J Med* (2018.0) **380** 347-357. DOI: 10.1056/NEJMoa1812389
20. Neal B, Perkovic V, de Zeeuw D, Mahaffey KW, Fulcher G, Ways K, Desai M, Shaw W, Capuano G, Alba M. **Efficacy and safety of canagliflozin, an inhibitor of sodium-glucose cotransporter 2, when used in conjunction with insulin therapy in patients with type 2 diabetes**. *Diabetes Care* (2014.0) **38** 403-411. DOI: 10.2337/dc14-1237
21. Cefalu WT, Leiter LA, de Bruin TWA, Gause-Nilsson I, Sugg J, Parikh SJ. **Dapagliflozin's effects on glycemia and cardiovascular risk factors in high-risk patients with type 2 diabetes: a 24-week, multicenter, randomized, double-blind, placebo-controlled study with a 28-week extension**. *Diabetes Care* (2015.0) **38** 1218-1227. DOI: 10.2337/dc14-0315
22. Cannon CP, Pratley R, Dagogo-Jack S, Mancuso J, Huyck S, Masiukiewicz U, Charbonnel B, Frederich R, Gallo S, Cosentino F. **Cardiovascular outcomes with ertugliflozin in type 2 diabetes**. *N Engl J Med* (2020.0) **383** 1425-1435. DOI: 10.1056/NEJMoa2004967
23. Perkovic V, Jardine MJ, Neal B, Bompoint S, Heerspink HJL, Charytan DM, Edwards R, Agarwal R, Bakris G, Bull S. **Canagliflozin and renal outcomes in type 2 diabetes and nephropathy**. *N Engl J Med* (2019.0) **380** 2295-2306. DOI: 10.1056/NEJMoa1811744
24. Palmer SC, Tendal B, Mustafa RA, Vandvik PO, Li S, Hao Q, Tunnicliffe D, Ruospo M, Natale P, Saglimbene V. **Sodium-glucose cotransporter protein-2 (SGLT-2) inhibitors and glucagon-like peptide-1 (GLP-1) receptor agonists for type 2 diabetes: systematic review and network meta-analysis of randomised controlled trials**. *BMJ* (2021.0) **372** m4573-m4573. DOI: 10.1136/bmj.m4573
25. Butler J, Usman MS, Khan MS, Greene SJ, Friede T, Vaduganathan M, Filippatos G, Coats AJS, Anker SD. **Efficacy and safety of SGLT2 inhibitors in heart failure: systematic review and meta-analysis**. *ESC Heart Fail* (2020.0) **7** 3298-3309. DOI: 10.1002/ehf2.13169
26. Zannad F, Ferreira JP, Pocock SJ, Anker SD, Butler J, Filippatos G, Brueckmann M, Ofstad AP, Pfarr E, Jamal W. **SGLT2 inhibitors in patients with heart failure with reduced ejection fraction: a meta-analysis of the EMPEROR-reduced and DAPA-HF trials**. *Lancet* (2020.0) **396** 819-829. DOI: 10.1016/S0140-6736(20)31824-9
27. Zelniker TA, Wiviott SD, Raz I, Im K, Goodrich EL, Bonaca MP, Mosenzon O, Kato ET, Cahn A, Furtado RHM. **SGLT2 inhibitors for primary and secondary prevention of cardiovascular and renal outcomes in type 2 diabetes: a systematic review and meta-analysis of cardiovascular outcome trials**. *Lancet* (2019.0) **393** 31-39. DOI: 10.1016/S0140-6736(18)32590-X
28. Lee G, Oh SW, Hwang SS, Yoon JW, Kang S, Joh HK, Kwon H, Kim J, Park D. **Comparative effectiveness of oral antidiabetic drugs in preventing cardiovascular mortality and morbidity: a network meta-analysis**. *PLoS ONE* (2017.0) **12** e0177646. DOI: 10.1371/journal.pone.0177646
29. Lin C, Zhu X, Cai X, Yang W, Lv F, Nie L, Ji L. **SGLT2 inhibitors and lower limb complications: an updated meta-analysis**. *Cardiovasc Diabetol* (2021.0) **20** 91. DOI: 10.1186/s12933-021-01276-9
30. Liao XX, Li WQ, Peng ZK, Yu HB, Tan J. **Three new categories of hypoglycaemic agents and various cardiovascular diseases: a meta-analysis**. *J Clin Pharm Ther* (2022.0) **47** 636-642. DOI: 10.1111/jcpt.13588
31. Zhou Z, Jardine MJ, Li Q, Neuen BL, Cannon CP, de Zeeuw D, Edwards R, Levin A, Mahaffey KW, Perkovic V. **Effect of SGLT2 inhibitors on stroke and atrial fibrillation in diabetic kidney disease: results from the CREDENCE trial and meta-analysis**. *Stroke* (2021.0) **52** 1545-1556. DOI: 10.1161/STROKEAHA.120.031623
32. Tsai WH, Chuang SM, Liu SC, Lee CC, Chien MN, Leung CH, Liu SJ, Shih HM. **Effects of SGLT2 inhibitors on stroke and its subtypes in patients with type 2 diabetes: a systematic review and meta-analysis**. *Sci Rep* (2021.0) **11** 15364. DOI: 10.1038/s41598-021-94945-4
33. Lavalle-González FJ, Januszewicz A, Davidson J, Tong C, Qiu R, Canovatchel W, Meininger G. **Efficacy and safety of canagliflozin compared with placebo and sitagliptin in patients with type 2 diabetes on background metformin monotherapy: a randomised trial**. *Diabetologia* (2013.0) **56** 2582-2592. DOI: 10.1007/s00125-013-3039-1
34. Cefalu WT, Leiter LA, Yoon KH, Arias P, Niskanen L, Xie J, Balis DA, Canovatchel W, Meininger G. **Efficacy and safety of canagliflozin versus glimepiride in patients with type 2 diabetes inadequately controlled with metformin (CANTATA-SU): 52 week results from a randomised, double-blind, phase 3 non-inferiority trial**. *Lancet* (2013.0) **382** 941-950. DOI: 10.1016/S0140-6736(13)60683-2
35. Yale JF, Bakris G, Cariou B, Nieto J, David-Neto E, Yue D, Wajs E, Figueroa K, Jiang J, Law G. **Efficacy and safety of canagliflozin over 52 weeks in patients with type 2 diabetes mellitus and chronic kidney disease**. *Diabetes Obes Metab* (2014.0) **16** 1016-1027. DOI: 10.1111/dom.12348
36. Bode B, Stenlöf K, Harris S, Sullivan D, Fung A, Usiskin K, Meininger G. **Long-term efficacy and safety of canagliflozin over 104 weeks in patients aged 55–80 years with type 2 diabetes**. *Diabetes Obes Metab* (2015.0) **17** 294-303. DOI: 10.1111/dom.12428
37. Rosenstock J, Chuck L, González-Ortiz M, Merton K, Craig J, Capuano G, Qiu R. **Initial combination therapy with canagliflozin plus metformin versus each component as monotherapy for drug-naïve type 2 diabetes**. *Diabetes Care* (2016.0) **39** 353-362. DOI: 10.2337/dc15-1736
38. Lingvay I, Catarig AM, Frias JP, Kumar H, Lausvig NL, le Roux CW, Thielke D, Viljoen A, McCrimmon RJ. **Efficacy and safety of once-weekly semaglutide versus daily canagliflozin as add-on to metformin in patients with type 2 diabetes (SUSTAIN 8): a double-blind, phase 3b, randomised controlled trial**. *Lancet Diabetes Endocrinol* (2019.0) **7** 834-844. DOI: 10.1016/S2213-8587(19)30311-0
39. Nauck MA, Del Prato S, Meier JJ, Durán-García S, Rohwedder K, Elze M, Parikh SJ. **Dapagliflozin versus glipizide as add-on therapy in patients with type 2 diabetes who have inadequate glycemic control with metformin: a randomized, 52-week, double-blind, active-controlled noninferiority trial**. *Diabetes Care* (2011.0) **34** 2015-2022. DOI: 10.2337/dc11-0606
40. Strojek K, Yoon K, Hruba V, Elze M, Langkilde A, Parikh S. **Effect of dapagliflozin in patients with type 2 diabetes who have inadequate glycaemic control with glimepiride: a randomized, 24-week, double-blind, placebo-controlled trial**. *Diabetes Obes Metab* (2011.0) **13** 928-938. DOI: 10.1111/j.1463-1326.2011.01434.x
41. Henry R, Murray A, Marmolejo M, Hennicken D, Ptaszynska A, List J. **Dapagliflozin, metformin XR, or both: initial pharmacotherapy for type 2 diabetes, a randomised controlled trial**. *Int J Clin Pract* (2012.0) **66** 446-456. DOI: 10.1111/j.1742-1241.2012.02911.x
42. Bailey CJ, Gross JL, Hennicken D, Iqbal N, Mansfield TA, List JF. **Dapagliflozin add-on to metformin in type 2 diabetes inadequately controlled with metformin: a randomized, double-blind, placebo-controlled 102-week trial**. *BMC Med* (2013.0) **11** 43-43. DOI: 10.1186/1741-7015-11-43
43. Leiter LA, Cefalu WT, de Bruin TW, Gause-Nilsson I, Sugg J, Parikh SJ. **Dapagliflozin added to usual care in individuals with type 2 diabetes mellitus with preexisting cardiovascular disease: a 24-week, multicenter, randomized, double-blind, placebo-controlled study with a 28-week extension**. *J Am Geriatr Soc* (2014.0) **62** 1252-1262. DOI: 10.1111/jgs.12881
44. Wilding J, Woo V, Rohwedder K, Sugg J, Parikh S. **Dapagliflozin in patients with type 2 diabetes receiving high doses of insulin: efficacy and safety over 2 years**. *Diabetes Obesity Metab* (2014.0) **16** 124-136. DOI: 10.1111/dom.12187
45. Bailey CJ, Morales Villegas EC, Woo V, Tang W, Ptaszynska A, List JF. **Efficacy and safety of dapagliflozin monotherapy in people with type 2 diabetes: a randomized double-blind placebo-controlled 102-week trial**. *Diabet Med* (2015.0) **32** 531-541. DOI: 10.1111/dme.12624
46. Matthaei S, Bowering K, Rohwedder K, Grohl A, Parikh S. **Group ftS: dapagliflozin improves glycemic control and reduces body weight as add-on therapy to metformin plus sulfonylurea: a 24-week randomized double-blind clinical trial**. *Diabetes Care* (2015.0) **38** 365-372. DOI: 10.2337/dc14-0666
47. Müller-Wieland D, Kellerer M, Cypryk K, Skripova D, Rohwedder K, Johnsson E, Garcia-Sanchez R, Kurlyandskaya R, Sjöström CD, Jacob S. **Efficacy and safety of dapagliflozin or dapagliflozin plus saxagliptin versus glimepiride as add-on to metformin in patients with type 2 diabetes**. *Diabetes Obes Metab* (2018.0) **20** 2598-2607. DOI: 10.1111/dom.13437
48. Scott R, Morgan J, Zimmer Z, Lam RLH, O'Neill EA, Kaufman KD, Engel SS, Raji A. **A randomized clinical trial of the efficacy and safety of sitagliptin compared with dapagliflozin in patients with type 2 diabetes mellitus and mild renal insufficiency: the CompoSIT-R study**. *Diabetes Obes Metab* (2018.0) **20** 2876-2884. DOI: 10.1111/dom.13473
49. Fioretto P, Del Prato S, Buse JB, Goldenberg R, Giorgino F, Reyner D, Langkilde AM, Sjöström CD, Sartipy P, Investigators DS. **Efficacy and safety of dapagliflozin in patients with type 2 diabetes and moderate renal impairment (chronic kidney disease stage 3A): the DERIVE study**. *Diabetes Obes Metab* (2018.0) **20** 2532-2540. DOI: 10.1111/dom.13413
50. Yang W, Ma J, Li Y, Li Y, Zhou Z, Kim JH, Zhao J, Ptaszynska A. **Dapagliflozin as add-on therapy in Asian patients with type 2 diabetes inadequately controlled on insulin with or without oral antihyperglycemic drugs: a randomized controlled trial**. *J Diabetes* (2018.0) **10** 589-599. DOI: 10.1111/1753-0407.12634
51. Häring H-U, Merker L, Seewaldt-Becker E, Weimer M, Meinicke T, Woerle HJ, Broedl UC. **Investigators E-RMT: Empagliflozin as add-on to metformin plus sulfonylurea in patients with type 2 diabetes: a 24-week, randomized, double-blind, placebo-controlled trial**. *Diabetes Care* (2013.0) **36** 3396-3404. DOI: 10.2337/dc12-2673
52. Ferrannini E, Berk A, Hantel S, Pinnetti S, Hach T, Woerle HJ, Broedl UC. **Long-term safety and efficacy of empagliflozin, sitagliptin, and metformin: an active-controlled, parallel-group, randomized, 78-week open-label extension study in patients with type 2 diabetes**. *Diabetes Care* (2013.0) **36** 4015-4021. DOI: 10.2337/dc13-0663
53. Barnett AH, Mithal A, Manassie J, Jones R, Rattunde H, Woerle HJ, Broedl UC. **investigators E-RRt: efficacy and safety of empagliflozin added to existing antidiabetes treatment in patients with type 2 diabetes and chronic kidney disease: a randomised, double-blind, placebo-controlled trial**. *Lancet Diabetes Endocrinol* (2014.0) **2** 369-384. DOI: 10.1016/S2213-8587(13)70208-0
54. Rosenstock J, Jelaska A, Frappin G, Salsali A, Kim G, Woerle HJ, Broedl UC. **Investigators obotE-RMT: improved glucose control with weight loss, lower insulin doses, and no increased hypoglycemia with empagliflozin added to titrated multiple daily injections of insulin in obese inadequately controlled type 2 diabetes**. *Diabetes Care* (2014.0) **37** 1815-1823. DOI: 10.2337/dc13-3055
55. Roden M, Merker L, Christiansen AV, Roux F, Salsali A, Kim G, Stella P, Woerle HJ, Broedl UC. **investigators E-REGEM: safety, tolerability and effects on cardiometabolic risk factors of empagliflozin monotherapy in drug-naïve patients with type 2 diabetes: a double-blind extension of a Phase III randomized controlled trial**. *Cardiovasc Diabetol* (2015.0) **14** 154-154. DOI: 10.1186/s12933-015-0314-0
56. Rosenstock J, Jelaska A, Zeller C, Kim G, Broedl UC, Woerle HJ. **investigators E-RBt: impact of empagliflozin added on to basal insulin in type 2 diabetes inadequately controlled on basal insulin: a 78-week randomized, double-blind, placebo-controlled trial**. *Diabetes Obes Metab* (2015.0) **17** 936-948. DOI: 10.1111/dom.12503
57. Araki E, Tanizawa Y, Tanaka Y, Taniguchi A, Koiwai K, Kim G, Salsali A, Woerle HJ, Broedl UC. **Long-term treatment with empagliflozin as add-on to oral antidiabetes therapy in Japanese patients with type 2 diabetes mellitus**. *Diabetes Obes Metab* (2015.0) **17** 665-674. DOI: 10.1111/dom.12464
58. Hadjadj S, Rosenstock J, Meinicke T, Woerle HJ, Broedl UC. **Initial combination of empagliflozin and metformin in patients with type 2 diabetes**. *Diabetes Care* (2016.0) **39** 1718-1728. DOI: 10.2337/dc16-0522
59. Ridderstråle M, Rosenstock J, Andersen KR, Woerle HJ, Salsali A. **investigators ERHHSt: empagliflozin compared with glimepiride in metformin-treated patients with type 2 diabetes: 208-week data from a masked randomized controlled trial**. *Diabetes Obes Metab* (2018.0) **20** 2768-2777. DOI: 10.1111/dom.13457
60. Rodbard HW, Rosenstock J, Canani LH, Deerochanawong C, Gumprecht J, Lindberg SØ, Lingvay I, Søndergaard AL, Treppendahl MB, Montanya E. **Oral semaglutide versus empagliflozin in patients with type 2 diabetes uncontrolled on metformin: The PIONEER 2 trial**. *Diabetes Care* (2019.0) **42** 2272-2281. DOI: 10.2337/dc19-0883
61. Pratley RE, Eldor R, Raji A, Golm G, Huyck SB, Qiu Y, Sunga S, Johnson J, Terra SG, Mancuso JP. **Ertugliflozin plus sitagliptin versus either individual agent over 52 weeks in patients with type 2 diabetes mellitus inadequately controlled with metformin: the VERTIS FACTORIAL randomized trial**. *Diabetes Obes Metab* (2018.0) **20** 1111-1120. DOI: 10.1111/dom.13194
62. Rosenstock J, Frias J, Páll D, Charbonnel B, Pascu R, Saur D, Darekar A, Huyck S, Shi H, Lauring B. **Effect of ertugliflozin on glucose control, body weight, blood pressure and bone density in type 2 diabetes mellitus inadequately controlled on metformin monotherapy (VERTIS MET)**. *Diabetes Obes Metab* (2018.0) **20** 520-529. DOI: 10.1111/dom.13103
63. Grunberger G, Camp S, Johnson J, Huyck S, Terra SG, Mancuso JP, Jiang ZW, Golm G, Engel SS, Lauring B. **Ertugliflozin in patients with stage 3 chronic kidney disease and type 2 diabetes mellitus: the VERTIS RENAL randomized study**. *Diabetes Ther* (2018.0) **9** 49-66. DOI: 10.1007/s13300-017-0337-5
64. Dagogo-Jack S, Liu J, Eldor R, Amorin G, Johnson J, Hille D, Liao Y, Huyck S, Golm G, Terra SG. **Efficacy and safety of the addition of ertugliflozin in patients with type 2 diabetes mellitus inadequately controlled with metformin and sitagliptin: the VERTIS SITA2 placebo-controlled randomized study**. *Diabetes Obes Metab* (2018.0) **20** 530-540. DOI: 10.1111/dom.13116
65. Hollander P, Hill J, Johnson J, Jiang ZW, Golm G, Huyck S, Terra SG, Mancuso JP, Engel SS, Lauring B. **Results of VERTIS SU extension study: safety and efficacy of ertugliflozin treatment over 104 weeks compared to glimepiride in patients with type 2 diabetes mellitus inadequately controlled on metformin**. *Curr Med Res Opin* (2019.0). DOI: 10.1080/03007995.2019.1583450
|
---
title: 'Saturated fat and human health: a protocol for a methodologically innovative
systematic review and meta-analysis to inform public health nutrition guidelines'
authors:
- Bradley C. Johnston
- Dena Zeraatkar
- Jeremy Steen
- Diego Rada Fernandez de Jauregui
- Hongfei Zhu
- Mingyao Sun
- Matthew Cooper
- Malgorzata Maraj
- Anna Prokop-Dorner
- Boris Castro Reyes
- Claudia Valli
- Dawid Storman
- Giorgio Karam
- Joanna Zajac
- Long Ge
- Mateusz J. Swierz
- Nirjhar Ghosh
- Robin W. M. Vernooij
- Yaping Chang
- Yunli Zhao
- Lehana Thabane
- Gordon H. Guyatt
- Pablo Alonso-Coello
- Lee Hooper
- Malgorzata M. Bala
journal: Systematic Reviews
year: 2023
pmcid: PMC10012519
doi: 10.1186/s13643-023-02209-1
license: CC BY 4.0
---
# Saturated fat and human health: a protocol for a methodologically innovative systematic review and meta-analysis to inform public health nutrition guidelines
## Abstract
### Background
The health effects of dietary fats are a controversial issue on which experts and authoritative organizations have often disagreed. Care providers, guideline developers, policy-makers, and researchers use systematic reviews to advise patients and members of the public on optimal dietary habits, and to formulate public health recommendations and policies. Existing reviews, however, have serious limitations that impede optimal dietary fat recommendations, such as a lack of focus on outcomes important to people, substantial risk of bias (RoB) issues, ignoring absolute estimates of effects together with comprehensive assessments of the certainty of the estimates for all outcomes.
### Objective
We therefore propose a methodologically innovative systematic review using direct and indirect evidence on diet and food-based fats (i.e., reduction or replacement of saturated fat with monounsaturated or polyunsaturated fat, or carbohydrates or protein) and the risk of important health outcomes.
### Methods
We will collaborate with an experienced research librarian to search MEDLINE, EMBASE, CINAHL, and the Cochrane Database of Systematic Reviews (CDSR) for randomized clinical trials (RCTs) addressing saturated fat and our health outcomes of interest. In duplicate, we will screen, extract results from primary studies, assess their RoB, conduct de novo meta-analyses and/or network meta-analysis, assess the impact of missing outcome data on meta-analyses, present absolute effect estimates, and assess the certainty of evidence for each outcome using the GRADE contextualized approach. Our work will inform recommendations on saturated fat based on international standards for reporting systematic reviews and guidelines.
### Conclusion
Our systematic review and meta-analysis will provide the most comprehensive and rigorous summary of the evidence addressing the relationship between saturated fat modification for people-important health outcomes. The evidence from this review will be used to inform public health nutrition guidelines.
### Trial registration
PROSPERO Registration: CRD42023387377.
## Background
Non-communicable diseases, including cardiovascular disease (CVD), cancer, and diabetes are responsible for 4 of 5 deaths worldwide [1]. Modifying dietary habits may reduce the incidence of non-communicable diseases, though what constitutes an optimal fat intake and dietary pattern is highly debated.
The health effects of dietary fats are a controversial issue on which experts and authoritative organizations have often disagreed [2, 3]. While some guidelines, for example, have recommended restricting dietary fats to less than 30–$35\%$ of total energy intake, others have concluded that reduction of total dietary fats has little effect on improving health outcomes [4–7]. The relationship between saturated fats and cardiovascular disease is another case in point, about which authoritative organizations and experts continue to disagree [8, 9].
Care providers, guideline developers, policy-makers, and researchers use systematic reviews to advise patients on optimal dietary habits, formulate dietary recommendations and policies, and to plan future research [10–12]. While a plethora of systematic reviews on the health effects of dietary fats have been published to date [13], existing reviews have serious limitations. Most systematic reviews, for example, have addressed only one or a few health-related outcomes, whereas dietary recommendations and related policy implementation require consideration of all people-important outcomes (e.g., all-cause, cardiovascular, and cancer mortality, non-fatal stroke and myocardial infarction, cancer incidence, type 2 diabetes, dementia, and quality of life), as well as patient or public health-related values and preferences that bear on dietary recommendations [11, 12, 14]. Further, existing reviews often contain substantial deficiencies. For example, reviews often fail to present absolute effect estimates, which can lead to misinterpretation of findings [15–20]. Equally important, only a handful of reviews have formally and comprehensively evaluated the certainty of evidence for each outcome—a critical step in contextualizing review findings and generating dietary recommendations [15, 19–21], and aside from one dietary guideline on red and processed meat [22], none have used a contextualized approach, as recently recommended by the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) working group [23–25]. Finally, the extent to which risk of bias (RoB) associated with missing participant outcome data reduces the certainty in results [26, 27] also represents a key issue in contextualizing review findings and the generation of dietary recommendations [15]. For instance, evidence suggests that almost one in every three trials with statistically significant results lose significance when making plausible assumptions about the outcomes of participants with missing data [26].
We propose a methodologically innovative systematic review to summarize existing evidence on reducing or replacing dietary saturated fat with other nutrients (mono- or polyunsaturated fat, carbohydrates, protein) and important health outcomes. This work will improve upon the limitations of previous reviews and guidelines by addressing a comprehensive list of people-important single outcomes that patients and members of the public can better understand (rather than composite outcomes), analyze RoB using a nutrition specific assessment [14, 28] as well as tools for assessing missing data [27, 29], and calculating absolute effect estimates and evaluating the certainty of these estimates using the GRADE fully contextualized approach [15, 24].
## Research question
Among adult patients and members of the public with or without cardiometabolic conditions, what is the impact of modifying dietary fat intake (lowering saturated fat intake, or lowering saturated fat while increasing polyunsaturated fat and/or increasing monounsaturated fat and/or protein and/or complex carbohydrates) on the risk of critically important health outcomes (e.g., mortality, stroke, myocardial infarction, quality of life)?
## Scope
Our systematic review will summarize foods and diets addressing reduced saturated fats, and the replacement or modification of saturated fats with monounsaturated, polyunsaturated fats (e.g., omega-3 and omega-6), carbohydrates or protein and the relationship with all-cause mortality; cardiovascular mortality; cancer mortality; cardiovascular disease including coronary heart disease and myocardial infarctions; stroke; total cancer incidence; type 2 diabetes; dementia including Alzheimer’s disease; satisfaction with diet; and quality of life in patients with or without cardiometabolic conditions (e.g., previous history of cardiovascular events such as myocardial infarction or stroke, diabetes, hypertension) but without other chronic (e.g., cancer) or infectious conditions.
## Inclusion criteria
We will include randomized controlled trials (RCTs) of individuals or groups (six or more clusters). Randomized trials have to state an intention to reduce saturated fat (SFA) intake via appropriate food or nutrient-based aims, or trial reports have to provide a dietary aim in general, such as reducing total fat or improving heart health (reduced saturated fats and encouraged fruit and vegetables), while also achieving a statistically significant ($p \leq 0.05$) reduction in saturated fat reduction between the the intervention arm and control arm during the trial period. Eligible interventions have to be low fat dietary advice, supplementation with naturally occurring oils or fats (e.g., food based olive oil or fish), or provision of modified or low-fat foods, as compared to an intake of a higher saturated fat diet, placebo or a control diet higher in fat (e.g., usual diet). Our intended time-point of interest for the duration of the diet intervention will be 2 years (24 months) or more.
We will exclude trials that are formula based (e.g., weight-loss formulas such a NutriSystem), have a pharmacological intervention for weight-loss (e.g., Olestra), or the primary aim to assess weight-loss (experimental arm is calorie restricted while the control arm is ad libitum). If trials employ active interventions such cardiometabolic or smoking cessation medications (e.g., statins, Metformin, Chantix), the study will be eligible if both groups are provided drugs, active intervention. If a trial demonstrated a statistically significant between group reduction in SFA and encouraged active interventions such as physical exercise, or cardiovascular medications in one arm (intervention) with no exercise or medication in the alternative arm (control), we will exclude. We will also exclude observational studies given that over 50,000 patients have been randomized to SFA reduction and replacement interventions, and the available trials have captured all of our people-important outcomes. For the purpose of informing SFA dietary guidelines, we will use the most recent, high-quality systematic reviews of cohort studies (e.g., [30, 31]).
## Search strategy
We will collaborate with an experienced research librarian to search MEDLINE, EMBASE, CINAHL, and registers for reviews including PROSPERO and the Cochrane Database of Systematic Reviews (CDSR) for systematic reviews of RCTs. For reviews, we will run an updated search from the date of the last comprehensive systematic review (e.g., [32]) for new primary studies. Since questions addressed by older reviews are likely to have been also addressed in more recent reviews, we will restrict the search to reviews published from 2015 onward. We will also search the reference lists of included reviews, related publications on PubMed and Google Scholar, clinicaltrials.gov and contact content and research experts in this area to further augment our search. We will also search included studies from previous systematic reviews to ensure the subsequent studies have not been reported with longer follow-up data.
## Screening and study selection
Pairs of reviewers will complete calibration exercises, after which they will perform screening of search results independently and in duplicate. Reviewers will resolve discrepancies by discussion or by adjudication by an expert research methodologist. To start, we will search for systematic reviews that include one or more RCTs that address the association between dietary fats and important health outcomes (above) in adults with or without cardiometabolic conditions, but without other chronic or infectious conditions. In cases in which two or more systematic reviews reporting on the same exposure and outcome have overlapping searches within three months of one another, we will select the systematic review that is the most comprehensive (i.e., includes the most eligible studies). To identify RCTs published after the end of the search used in the most comprehensive systematic review identified, we will run an updated search of MEDLINE, EMBASE, CINAHL, and CDSR from the date of the last comprehensive systematic review.
## Data extraction
Following calibration exercises, reviewers, working independently, in duplicate and using a standardized, piloted tested data extraction form, will extract data from included reviews and primary studies. Reviewers will resolve discrepancies by discussion or by adjudication by an expert research methodologist.
From each original study, we will extract information on the study characteristics (population, intervention, comparator, outcomes) and results. For event (dichotomous) data we will extract numbers of events in each study arm at the last reported time point available during the intervention period, and the number of participants included in each arm. For continuous data, we will extract means, standard deviations (or other variance data), and numbers in each arm at the last reported time point during the intervention period. We will also extract planned dietary composition in both study arms, assessed dietary composition in both arms, longest duration of the intervention, whether the intervention consists of dietary advice, advice plus some food provision, or provision of most foods, cointerventions (pharmaceutical and non-pharmaceutical), data on compliance in both arms, and surrogate outcomes including total serum cholesterol and low density lipoprotein (LDL) cholesterol in both arms at the latest available date during the intervention (during the randomized period) [32].
## Risk of bias
Following calibration exercises, reviewers, working independently and in duplicate, will assess the RoB of eligible primary studies. Reviewers will resolve discrepancies by discussion or by adjudication by an expert research methodologist.
Systematic reviews often have important limitations related to their assessment of RoB, such as the application of tools that do not address all important sources of bias, tools that include constructs unrelated to RoB (e.g., generalizability), or apply RoB tools inconsistently [33]. Hence, in duplicate we will assess the RoB of primary studies de novo using a modification of the Cochrane risk of bias 1.0 instrument for RCTs [34, 35]. Based on our previous use of the tool, it has been modified in a standardized way in order to make the assessment more manageable and specific to nutrition studies [28]. For instance, our modified Cochrane RoB instrument for RCTs uses more comprehensive instructions for all RoB items with clear definitions and examples for low and high RoB, and uses four response options [36]. In particular, to evaluate sequence generation, allocation concealment and blinding, we will first assess if these items were adequately reported (i.e., clearly reported, mostly reported, mostly not reported, clearly not reported) and second, we will evaluate how serious the RoB is using an index of suspicion (i.e., 1 = definitely low risk, 2 = probably low risk, 3 = probably high risk, 4 = definitely high risk). For example, a study may report “double-blinded” without details as to which of the five possible study team members of an RCT where specifically blinded. For both data analysts and data adjudicators we would answer “clearly not reported” and “probably high risk of bias”. We will also modify the selective outcome reporting item to avoid confounding “outcome reporting bias” with “publication bias”, and use the more comprehensive and specific item for selective reporting used in the Cochrane RoB 2.0 tool [37]. Our team has previously successfully used similar modified instruments for the assessment of RoB for nutrition RCTs [28, 38].
## Risk of bias related to missing participant outcome data
Following calibration exercises, reviewers, working independently and in duplicate and using a standardized and piloted tested data extraction form will extract data on missing participant outcome data (MPOD) from included RCTs that address the association between dietary fats and all people-important outcomes. Reviewers will resolve discrepancies by discussion or by adjudication by an expert research methodologist. As needed, we will contact the trialists to ask for available but unreported MPOD in the primary study report.
We will define participant outcome data as ‘‘missing’’ if they are unavailable to the reviewers; that is, unavailable to investigators of the primary studies, or available to the primary study investigators but not included in published reports and not provided after inquiry.
For our primary analysis, we will use a complete-case analysis (sometimes referred to as an available-case analysis) where participants with missing participant outcome data are excluded from both the numerator and denominator when calculating relative and absolute risks. We will subsequently compare the complete-case analysis to a series of sensitivity analysis to explore the impact of missing data on our outcomes and assess the robustness of the effect estimates as suggested using GRADE 17 guidance [26, 27, 29, 39, 40]. To do so, we will assume that the event rate for those participants in the control group who had missing data was the same as the event rate for those participants in the control group who were successfully followed. For the intervention group, we will calculate effects using the following assumed ratios of event rates in those with missing data in comparison to those successfully followed: 1.5:1, 2:1, 3:1, and 5:1 [26].
## Data synthesis and analysis
For our review of RCTs, for each dietary fat and health outcome of interest, based on guidance from the Cochrane Handbook we will conduct de novo meta-analyses comparing lower versus higher intake of saturated fats, and saturated fats replaced or modified with polyunsaturated fatty acids (PUFA), monounsaturated fatty acids (MUFA), protein and complex carbohydrates. Replacement trials will be those wherein participants are asked to reduce their fat or SFA, and authors report evidence of significant decrease in SFA with a corresponding increase in other nutrients (e.g., SFA is reduced by ~ $6\%$ of daily calories [energy], while there is an increase in energy from PUFA (e.g., $4\%$) and/or MUFA (e.g., $2\%$). Replacement trials, while more robust if they provide known quantities of specific food interventions to participants, may or may not provide the intervention (e.g., nuts, olive oil). Depending the types of trials and intervention arms (reduction; replacement; modification of fat and macronutrients), we may conduct both standard pairwise comparisons and a network meta-analysis.
For each outcome reported in each review, we will present the intervention, comparator, number of studies and participants, the baseline risk, the absolute and relative effects and the corresponding certainty of evidence. We will use data from GLOBOCAN [41] and the Emerging Risk Factors Collaboration [42] to estimate the baseline and absolute risks for major cardiometabolic and cancer outcomes, respectively. Absolute risks for cardiometabolic outcomes will be estimated over 10.8 years, while cancer outcomes will be estimated over a lifetime [41, 42]. Using these baseline risks, we will calculate the absolute risk reductions for our respective outcomes [43]. Since our review will inform a dietary guideline, wherein decision-makers need to consider evidence from all people-important outcome data, we will use a fully contextualized approach and we will categorize the magnitude of effects as trivial, small but important, moderate or large using guidance from GRADE [24, 25] and the Cochrane Collaboration [44]. Using thresholds for the magnitude of importance developed in consultation with an international dietary guideline panel on red and processed meat [22], we will use the following categorization. For fatal outcomes, ≤ 10 events per 1000 will be considered a trivial (unimportant) effect size, 11–25 per 1000 will be considered a small but important effect, and 26–40 per 1000 will be considered moderate. For non-fatal outcomes, ≤ 20 per 1000 will be considered trivial, 21–40 per 1000 will be considered small but important, and 41–60 per 1000 will be considered moderate. For mixed fatal and non-fatal outcomes, ≤ 15 per 1000 will be considered trivial, 16–30 per 1000 will be considered small but important, and 31–45 per 1000 will be considered moderate in size. For continuous patient-reported quality of life measures, will search the literature for anchor-based minimal important difference (MID) estimates, and if no MID is identified we will use half the baseline standard deviation from normative data for the quality of life measure [45–47]. As per GRADE guidance, we will present our data in summary of findings tables [43].
We will conduct subgroup analyses or meta-regression, if appropriately powered, with a chi-square test of interaction to assess the following anticipated effect modifiers:i)We will consider the primary macronutrient replacing the dietary fat under investigation (e.g., replacement of saturated fatty acid (SFA) with PUFA, MUFA, protein or carbohydrates). We anticipate that, for instance, replacing SFA with PUFA or MUFA will reduce the risk of cardiovascular events [32, 48] more than SFA reduction alone, or replacement with CHO or protein.ii)We will conduct subgroups among RCTs that provide food (e.g., nuts, olive oil, fish) or fat supplementation versus those with dietary advice only, anticipating larger treatment effects in trials that provide food/supplementation interventions [32, 49].iii)We will explore co-interventions as an effect modifier (e.g., statins, blood pressure lowering agents, exercise, or behavioral support groups), anticipating larger treatment effects in trials that provide active co-interventions [50, 51].iv)We will use meta-regression to examine the association between the change in low density lipoprotein (LDL) and total cholesterol, surrogates for SFA reduction, and the log relative risk changes for each of our outcomes. We anticipate that participants with lower cholesterol levels will have a lower risk of stroke, myocardial infarction, coronary heart disease, and cardiovascular mortality.v)Based on a modified version of the Cochrane risk of bias 1.0 instrument [28], we explore if studies at lower RoB have estimates of effect that differ significantly from studies at higher RoB, anticipating that studies at higher risk of bias will have larger treatment effects [52].
We will conduct a number of sensitivity analysis including (a) for primary studies that report on measures of dietary fats from participant-reported dietary intake surveys (i.e., dietary records or recalls, food frequency questionnaires) versus those that report tissue biomarkers (adipose polyunsaturated linoleic acid levels, subcutaneous fat aspirate, plasma fatty acid concentration) with or without participant-reported dietary intake assessments. While the number of established nutritional biomarkers is small, biomarkers may be applied directly in disease association analyses as has been done successfully for dairy fats [53] and alpha linolenic acid [54] based on valid markers for omaga 3 fatty acids [55], and may be used to calibrate self-report assessments to reduce systematic and random measurement error [56]. We will also conduct sensitivity analysis for (b) studies that report results corresponding to intake of dietary fats in absolute quantities (i.e., g/day) and those that report results from energy density models (i.e., % energy).
We will conduct pairwise meta-analyses using Revman 5.0, the meta package [57] and R version 3.5.1 (R Foundation for Statistical Computing).
## Evaluation of the certainty of evidence
We will assess the certainty of the evidence using the GRADE fully contextualized approach [24, 25] approach and present results for reducing and replacing/modifying saturated fat separately in a summary of findings table [43].
## Discussion
Our systematic review will provide a comprehensive and rigorous summary of the evidence addressing the relationship between dietary saturated and unsaturated fats and important health outcomes, including an estimate of the magnitude of effect, and the certainty of evidence.
This work represents a novel and efficient approach to evidence synthesis in fields in which many evidence syntheses have been previously published on dietary fats and health outcomes. Our systematic reviews will begin by utilizing the search strategies and selection of relevant studies done by existing well conducted systematic literature reviews (e.g., [32]), and subsequently we will search for primary studies from the date of the last systematic review forward.
## Implications
Our findings will be used towards the development of dietary guidelines addressing saturated fats and the risk of cardiovascular and cancer outcomes, including recommendations for those at very low, low, and high risk of a cardiovascular event.
## Dissemination
We will disseminate our findings by publication in a peer reviewed journal and by presentation at national and international conferences. Based on the findings of our overview, we will make GRADE summary of findings tables, plain language summaries, and infographics in user-friendly and open-access outputs for clinicians and patients and community members.
## Strengths and limitations
The strengths of this systematic review will include an extensive search for relevant systematic reviews of RCTs and RCTs; duplicate screening and extraction of data; the assessment of the RoB of studies using rigorous criteria aligned with advances in the methodology of bias assessment [28, 35, 37] including the assessment of the impact of missing participant outcome data on the RoB [27], and the transparent assessment of the magnitude and certainty of estimates for each of our outcomes using GRADE criteria [24, 25]. The GRADE approach to assessing the certainty of estimates is based on comprehensive methodology that has been described in detail in a series of eight BMJ publications and over 30 publications in the Journal of Clinical Epidemiology and has been adopted by over 110 international organizations, including the Cochrane Collaboration, Joanna Briggs Institute and the World Health Organization, each of which regularly apply GRADE to nutritional questions [11, 21]. The application of the GRADE approach will facilitate the consideration of important criteria that bear on the certainty of evidence on the relationship between saturated fats and health outcomes, including RoB, inconsistency, indirectness, imprecision, and publication bias. All this will serve to improve public transparency in making and communicating judgments about the magnitude and certainty of evidence on dietary fat and health outcomes.
## Conclusions
Our review will provide a comprehensive and rigorous summary of the evidence addressing the relationship between saturated, polyunsaturated and monounsaturated fats, and important health outcomes, including absolute estimates of the magnitude of effect and the certainty of evidence for these effects. Our findings will be used to establish dietary recommendations on saturated fat consumption, including reduction versus replacement of SFA with PUFA, MUFA, CHO, and protein.
## References
1. 1.World Health Organization. Global status report on noncommunicable diseases 2014. World Health Organization; 2014. https://apps.who.int/iris/handle/10665/148114. Accessed 12 Mar 2023.
2. Teicholz N. **The scientific report guiding the US dietary guidelines: is it scientific?**. *BMJ* (2015.0) **351** h4962. DOI: 10.1136/bmj.h4962
3. Forouhi NG, Krauss RM, Taubes G, Willett W. **Dietary fat and cardiometabolic health: evidence, controversies, and consensus for guidance**. *BMJ* (2018.0) **361** k2139. DOI: 10.1136/bmj.k2139
4. 4.US Department of Agriculture, US Department of Health & Human Services. Dietary guidelines for Americans 2015–2020: Skyhorse Publishing Inc.; 2015.
5. 5.Department of Health and Human Services, Agriculturem USDo. Scientific Report of the 2015 Dietary Guidelines Advisory Committee. 2015.
6. Herforth A, Arimond M, Alvarez-Sanchez C, Coates J, Christianson K, Muehlhoff E. **A global review of food-based dietary guidelines**. *Adv Nutr* (2019.0) **10** 590-605. DOI: 10.1093/advances/nmy130
7. 7.National Health Services. The Eatwell Guide. UK; 2016. https://www.nhs.uk/live-well/eat-well/food-guidelines-and-food-labels/the-eatwell-guide/. Accessed 12 Mar 2023.
8. Demasi M. **US nutritionists call for dietary guideline limits on saturated fat intake to be lifted**. *BMJ* (2020.0) **371** m4226. DOI: 10.1136/bmj.m4226
9. Astrup A, Magkos F, Bier DM, Brenna JT, de Oliveira Otto MC, Hill JO. **Saturated fats and health: a reassessment and proposal for food-based recommendations: JACC State-of-the-Art Review**. *J Am Coll Cardiol* (2020.0) **76** 844-857. DOI: 10.1016/j.jacc.2020.05.077
10. Brannon PM, Taylor CL, Coates PM. **Use and applications of systematic reviews in public health nutrition**. *Annu Rev Nutr* (2014.0) **34** 401-419. DOI: 10.1146/annurev-nutr-080508-141240
11. Johnston BC, Alonso-Coello P, Bala MM, Zeraatkar D, Rabassa M, Valli C. **Methods for trustworthy nutritional recommendations NutriRECS (Nutritional Recommendations and accessible Evidence summaries Composed of Systematic reviews): a protocol**. *BMC Med Res Methodol* (2018.0) **18** 162. DOI: 10.1186/s12874-018-0621-8
12. Zeraatkar D, Johnston BC, Guyatt G. **Evidence collection and evaluation for the development of dietary guidelines and public policy on nutrition**. *Annu Rev Nutr* (2019.0) **39** 227-247. DOI: 10.1146/annurev-nutr-082018-124610
13. Schwingshackl L, Zähringer J, Beyerbach J, Werner SW, Heseker H, Koletzko B. **Total dietary fat intake, fat quality, and health outcomes: a scoping review of systematic reviews of prospective studies**. *Ann Nutr Metab* (2021.0) **77** 4-15. DOI: 10.1159/000515058
14. 14.Guyatt GH, Oxman AD, Kunz R, Atkins D, Brozek J, Vist G, et al. GRADE guidelines: 2. Framing the question and deciding on important outcomes. J Clin Epidemiol. 2011;64(4):395–400.
15. Johnston BC, Guyatt GH. **Causal inference, interpreting and communicating results on red and processed meat**. *Am J Clin Nutr* (2020.0) **111** 1107-1108. DOI: 10.1093/ajcn/nqaa043
16. Forrow L, Taylor WC, Arnold RM. **Absolutely relative: how research results are summarized can affect treatment decisions**. *Am J Med* (1992.0) **92** 121-124. DOI: 10.1016/0002-9343(92)90100-P
17. Naylor CD, Chen E, Strauss B. **Measured enthusiasm: does the method of reporting trial results alter perceptions of therapeutic effectiveness?**. *Ann Intern Med* (1992.0) **117** 916-921. DOI: 10.7326/0003-4819-117-11-916
18. Tucker G, Metcalfe A, Pearce C, Need AG, Dick IM, Prince RL. **The importance of calculating absolute rather than relative fracture risk**. *Bone* (2007.0) **41** 937-941. DOI: 10.1016/j.bone.2007.07.015
19. Zajac JF, Storman D, Swierz MJ, Koperny M, Weglarz P, Staskiewicz W. **Are systematic reviews addressing nutrition for cancer prevention trustworthy? A systematic survey of quality and risk of bias**. *Nutr Rev* (2022.0) **80** 1558-1567. DOI: 10.1093/nutrit/nuab093
20. Zeraatkar D, Bhasin A, Morassut RE, Churchill I, Gupta A, Lawson DO. **Characteristics and quality of systematic reviews and meta-analyses of observational nutritional epidemiology: a cross-sectional study**. *Am J Clin Nutr* (2021.0) **113** 1578-1592. DOI: 10.1093/ajcn/nqab002
21. Vernooij R, Guyatt GH, Zeraatkar D, Han MA, Valli C, El Dib R, Alonso-Coello P, Bala MM, Johnston BC. **Reconciling contrasting guideline recommendations on red and processed meat for health outcomes**. *J Clin Epidemiol* (2021.0) **138** 215-218. DOI: 10.1016/j.jclinepi.2021.07.008
22. Johnston BC, Zeraatkar D, Mi A, Vernooij RWM, Valli C, El Dib R. **Unprocessed red meat and processed meat consumption: dietary guideline recommendations from the nutritional recommendations (NutriRECS) consortium**. *Ann Intern Med* (2019.0) **171** 756-745. DOI: 10.7326/M19-1621
23. Guyatt GH, Oxman AD, Vist GE, Kunz R, Falck-Ytter Y, Alonso-Coello P. **GRADE: an emerging consensus on rating quality of evidence and strength of recommendations**. *BMJ* (2008.0) **336** 924-926. DOI: 10.1136/bmj.39489.470347.AD
24. Zeng L, Brignardello-Petersen R, Hultcrantz M, Siemieniuk RAC, Santesso N, Traversy G. **GRADE guidelines 32: GRADE offers guidance on choosing targets of GRADE certainty of evidence ratings**. *J Clin Epidemiol* (2021.0) **137** 163-175. DOI: 10.1016/j.jclinepi.2021.03.026
25. Zeng L, Brignardello-Petersen R, Guyatt G. **When applying GRADE, how do we decide the target of certainty of evidence rating?**. *Evid Based Ment Health* (2021.0) **24** 121-123. DOI: 10.1136/ebmental-2020-300170
26. 26.Akl EA, Briel M, You JJ, Sun X, Johnston BC, Busse JW, et al. Potential impact on estimated treatment effects of information lost to follow-up in randomised controlled trials (LOST-IT): systematic review. BMJ (Clinical research ed). 2012;344:e2809.
27. Guyatt GH, Ebrahim S, Alonso-Coello P, Johnston BC, Mathioudakis AG, Briel M. **GRADE guidelines 17: assessing the risk of bias associated with missing participant outcome data in a body of evidence**. *J Clin Epidemiol* (2017.0) **87** 14-22. DOI: 10.1016/j.jclinepi.2017.05.005
28. Zeraatkar D, Johnston BC, Bartoszko J, Cheung K, Bala MM, Valli C. **Effect of lower versus higher red meat intake on cardiometabolic and cancer outcomes: a systematic review of randomized trials**. *Ann Intern Med* (2019.0) **171** 721-731. DOI: 10.7326/M19-0622
29. Johnston BC, Guyatt GH. **Best (but oft-forgotten) practices: intention-to-treat, treatment adherence, and missing participant outcome data in the nutrition literature**. *Am J Clin Nutr* (2016.0) **104** 1197-1201. DOI: 10.3945/ajcn.115.123315
30. Schwab U, Reynolds AN, Sallinen T, Rivellese AA, Risérus U. **Dietary fat intakes and cardiovascular disease risk in adults with type 2 diabetes: a systematic review and meta-analysis**. *Eur J Nutr* (2021.0) **60** 3355-3363. DOI: 10.1007/s00394-021-02507-1
31. 31.De Souza RJ, Mente A, Maroleanu A, Cozma AI, Ha V, Kishibe T, et al. Intake of saturated and trans unsaturated fatty acids and risk of all cause mortality, cardiovascular disease, and type 2 diabetes: systematic review and meta-analysis of observational studies. BMJ. 2015;351.
32. 32.Hooper L, Martin N, Jimoh OF, Kirk C, Foster E, Abdelhamid AS. Reduction in saturated fat intake for cardiovascular disease. Cochrane Database Syst Rev. 2020;5(5):Cd011737.
33. Zeraatkar D, Kohut A, Bhasin A, Morassut RE, Churchill I, Gupta A. **Assessments of risk of bias in systematic reviews of observational nutritional epidemiologic studies are often not appropriate or comprehensive: a methodological study**. *BMJ Nutr Prev Health* (2021.0) **4** 487-500. DOI: 10.1136/bmjnph-2021-000248
34. Hartling L, Ospina M, Liang Y, Dryden DM, Hooton N, Krebs Seida J, Klassen TP. **Risk of bias versus quality assessment of randomised controlled trials: cross sectional study**. *BMJ* (2009.0) **19** b4012. DOI: 10.1136/bmj.b4012
35. da Costa BR, Beckett B, Diaz A, Resta NM, Johnston BC, Egger M, Jüni P, Armijo-Olivo S. **Effect of standardized training on the reliability of the Cochrane risk of bias assessment tool: a prospective study**. *Syst Rev* (2017.0) **6** 44. DOI: 10.1186/s13643-017-0441-7
36. Akl EA, Sun X, Busse JW, Johnston BC, Briel M, Mulla S. **Specific instructions for estimating unclearly reported blinding status in randomized trials were reliable and valid**. *J Clin Epidemiol* (2012.0) **65** 262-267. DOI: 10.1016/j.jclinepi.2011.04.015
37. Sterne JAC, Savovic J, Page MJ, Elbers RG, Blencowe NS, Boutron I. **RoB 2: a revised tool for assessing risk of bias in randomised trials**. *BMJ* (2019.0) **366** l4898. DOI: 10.1136/bmj.l4898
38. 38.Ge L, Sadeghirad B, Ball GDC, da Costa BR, Hitchcock CL, Svendrovski A, et al. Comparison of dietary macronutrient patterns of 14 popular named dietary programmes for weight and cardiovascular risk factor reduction in adults: systematic review and network meta-analysis of randomised trials. BMJ. 2020;369:m696. : 10.1136/bmj.m696.
39. Akl EA, Johnston BC, Alonso-Coello P, Neumann I, Ebrahim S, Briel M. **Addressing dichotomous data for participants excluded from trial analysis: a guide for systematic reviewers**. *PLoS ONE* (2013.0) **8** e57132. DOI: 10.1371/journal.pone.0057132
40. White IR, Horton NJ, Carpenter J, Pocock SJ. **Strategy for intention to treat analysis in randomised trials with missing outcome data**. *BMJ* (2011.0) **342** d40. DOI: 10.1136/bmj.d40
41. Ferlay J, Soerjomataram I, Dikshit R, Eser S, Mathers C, Rebelo M. **Cancer incidence and mortality worldwide: sources, methods and major patterns in GLOBOCAN 2012**. *Int J Cancer* (2015.0) **136** E359-E386. DOI: 10.1002/ijc.29210
42. 42.The Emerging Risk Factors C. Diabetes mellitus, fasting blood glucose concentration, and risk of vascular disease: a collaborative meta-analysis of 102 prospective studies. Lancet. 2010;375(9733):2215–22.
43. 43.Guyatt GH, Oxman AD, Santesso N, Helfand M, Vist G, Kunz R, et al. GRADE guidelines: 12. Preparing summary of findings tables-binary outcomes. J Clin Epidemiol. 2013;66(2):158–72.
44. 44.Schünemann HJ, Vist GE, Higgins JPT, Santesso N, Deeks JJ, Glasziou P, Akl EA, Guyatt GH. Chapter 15: Interpreting results and drawing conclusions. In: Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA (editors). Cochrane Handbook for Systematic Reviews of Interventions version 6.1 (updated September 2020). Cochrane, 2020. Available from www.training.cochrane.org/handbook.
45. Carrasco-Labra A, Devji T, Qasim A, Phillips MR, Wang Y, Johnston BC. **Minimal important difference estimates for patient-reported outcomes: a systematic survey**. *J Clin Epidemiol* (2021.0) **133** 61-71. DOI: 10.1016/j.jclinepi.2020.11.024
46. Johnston BC, Ebrahim S, Carrasco-Labra A, Furukawa TA, Patrick DL, Crawford MW. **Minimally important difference estimates and methods: a protocol**. *BMJ Open* (2015.0) **5** e007953. DOI: 10.1136/bmjopen-2015-007953
47. Norman GR, Sloan JA, Wyrwich KW. **Interpretation of changes in health-related quality of life: the remarkable universality of half a standard deviation**. *Med Care* (2003.0) **41** 582-592. DOI: 10.1097/01.MLR.0000062554.74615.4C
48. 48.Hooper L, Martin N, Abdelhamid A, Davey Smith G. Reduction in saturated fat intake for cardiovascular disease. Cochrane Database Syst Rev. 2015(6):Cd011737.
49. Estruch R, Ros E, Salas-Salvado J, Covas MI, Corella D, Aros F. **Primary prevention of cardiovascular disease with a mediterranean diet supplemented with extra-virgin olive oil or nuts**. *N Engl J Med* (2018.0) **378** e34. DOI: 10.1056/NEJMoa1800389
50. Karmali KN, Lloyd-Jones DM, Berendsen MA, Goff DC, Sanghavi DM, Brown NC, Korenovska L, Huffman MD. **Drugs for primary prevention of atherosclerotic cardiovascular disease: an overview of systematic reviews**. *JAMA Cardiol* (2016.0) **1** 341-349. DOI: 10.1001/jamacardio.2016.0218
51. Posadzki P, Pieper D, Bajpai R, Makaruk H, Könsgen N, Neuhaus AL, Semwal M. **Exercise/physical activity and health outcomes: an overview of Cochrane systematic reviews**. *BMC Public Health* (2020.0) **20** 1724. DOI: 10.1186/s12889-020-09855-3
52. Savovic J, Turner RM, Mawdsley D, Jones HE, Beynon R, Higgins JPT, Sterne JAC. **Association between risk-of-bias assessments and results of randomized trials in cochrane reviews: the ROBES Meta-Epidemiologic Study**. *Am J Epidemiol* (2018.0) **187** 1113-1122. DOI: 10.1093/aje/kwx344
53. Naghshi S, Aune D, Beyene J, Mobarak S, Asadi M, Sadeghi O. **Dietary intake and biomarkers of alpha linolenic acid and risk of all cause, cardiovascular, and cancer mortality: systematic review and dose-response meta-analysis of cohort studies**. *BMJ* (2021.0) **375** n2213. DOI: 10.1136/bmj.n2213
54. Trieu K, Bhat S, Dai Z, Leander K, Gigante B, Qian F. **Biomarkers of dairy fat intake, incident cardiovascular disease, and all-cause mortality: a cohort study, systematic review, and meta-analysis**. *PLoS Med* (2021.0) **18** e1003763. DOI: 10.1371/journal.pmed.1003763
55. Serra-Majem L, Nissensohn M, Øverby NC, Fekete K. **Dietary methods and biomarkers of omega 3 fatty acids: a systematic review**. *Br J Nutr* (2012.0) **107** S64-76. DOI: 10.1017/S000711451200147X
56. Prentice RL. **Dietary assessment and opportunities to enhance nutritional epidemiology evidence**. *Ann Intern Med* (2020.0) **172** 354-355. DOI: 10.7326/M19-3290
57. Balduzzi S, Rücker G, Schwarzer G. **How to perform a meta-analysis with R: a practical tutorial**. *Evid Based Ment Health* (2019.0) **22** 153-160. DOI: 10.1136/ebmental-2019-300117
|
---
title: 'Determinants and economic burden of HIV/AIDS in Iran: a prospective study'
authors:
- Tahmineh Reshadat-Hajiabad
- Alireza Khajavi
- Ali Mohammad Hosseinpour
- Amin Bojdy
- Amir Hashemi-Meshkini
- Mehdi Varmaghani
journal: BMC Health Services Research
year: 2023
pmcid: PMC10012526
doi: 10.1186/s12913-023-09229-6
license: CC BY 4.0
---
# Determinants and economic burden of HIV/AIDS in Iran: a prospective study
## Abstract
### Background
Since the start of the AIDS outbreak, the human immunodeficiency virus (HIV) has infected about 84.2 million people, and approximately 40.1 million people have died due to AIDS-related diseases. So, this study aims to provide a comprehensive population-based description of patient costs and the economic burden of HIV/AIDS in Iran.
### Methods
The study population of this cross-sectional cost-of-illness study consisted of HIV-infected patients who were receiving services in Mashhad and were under the supervision of BIDCC. There are four BIDCC centers in Mashhad, we considered all patients referred to these centers. Costs data were evaluated from a social perspective with a bottom-up approach and as a prevalence based. The data from 157 individuals were included in the study. For collecting data on direct and indirect costs belonging to patients and their families, a questionnaire was developed. Also, the Demographic characteristic of participants and the stage of the disease and Transmission category were analyzed.
### Results
In this study, 57.32 of the subjects were Male. The majority of participants in this study were in the age group 30–59 years ($$n = 124$$,$78.98\%$). Based on where the patients live, the majority of patients have lived in the urban region ($$n = 144$$, $91.72\%$). The most common way to transmit this disease is through unprotected sex ($30.57\%$) and then Infected spouse ($28.03\%$), and then injecting drugs ($21.02\%$). The highest cost of this disease is attributed to medicine (10339.32 $ for 6 months), after medicine, the cost of tests was 9101.22 $.
### Conclusion
It seems that to reduce costs for patients with disease HIV/AIDS, the focus should be on diagnostic tests and care. Early diagnosis and rapid initiation of antiviral treatments can be effective in preventing serious and debilitating diseases.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12913-023-09229-6.
## Introduction
Since the start of the AIDS outbreak, the human immunodeficiency virus (HIV) has infected about 84.2 million people, and approximately 40.1 million people have died due to AIDS-related diseases [1]. United Nations program on HIV/AIDS (UNAIDS) has estimated that about 38.4 million people were living with human immunodeficiency virus (HIV) around the world. Also, People newly infected with HIV in 2021 were 1.5 in 2021, and AIDS-related deaths in 2021 were 650,000 [2–4]. Consequently, HIV/AIDS is one of the main burdens of disease worldwide [5].
Regarding the newest estimates of the UNAIDS on HIV/AIDS, a total of 53,000 people ($95\%$ CI [38,000–140,000) people are living with HIV (all ages) in Iran until 2021, among them, women aged 15 and over, the men aged 15 and over, and the Children aged 0 to 14 were estimated to be about 17,000 [12,000–44,000], 35,000 [25,000–96,000], and 1,400 [1,000–3,200], respectively. Furthermore, AIDS-related deaths of all ages in Iran in 2019 were 2,500 [1,200–5,600], of which 2,000 deaths were recorded for men and they are between 16 and 40 years old [2, 6].
The long duration of this disease has created the need for patients for long and high-level treatments, which has caused the reduction or loss of savings and household income and increased the debts of the patients’ families and the patients themselves. The high costs of health care for patients and their families make them neglect the ability to prepare nutritious food, invest and work, and even affect their children’s education and many other things. has reduced the death rate, but still every year a large number of infected patients lose their lives due to this disease. [ 1] The most important economic effect of HIV/AIDS is death at the working age [7]. Losing working time, changing jobs, time spent caring for the sick, and days that cannot go to work due to disability [8]. All these pieces of evidence indicate severe economic pressure for patients with HIV/AIDS who have to pay the high costs of care and try to compensate for lost income, reduced household income and savings, loans, loans, and children being removed from school. The most important cases are the effects of HIV/AIDS on families [9, 10].
The studies on economic burden and cost-of-illness (COI) evaluate the direct and indirect costs of particular diseases in a defined period [11]. Estimating direct and indirect medical costs due to different disease stages is valuable for partial and full economic evaluation studies because it offers a comprehensive insight into affected costs on prevention or screening strategies in non-communicable diseases in society [12]. Precise knowledge of COI enables policymakers and planners to prioritize healthcare plans and policies, and interventions, and even better allocate the resources to healthcare according to budget limitations to achieve policy effectiveness [13].
According to our knowledge, HIV patient costing studies in Iran have not comprehensively evaluated the economic impact, direct and indirect costs of illness, and also productivity costs due to HIV/ AIDS. This study aims to provide a comprehensive population-based description of patient costs and the economic burden of HIV/AIDS in Iran.
## Study setting
The study was carried out in the northeastern province of Iran (Khorasan Razavi). Mashhad, the capital city of this province, had approximately 3,001,184 population in 2022, a population density of around 8550.38 people per square kilometer [14].
## Patients and methods
The study population of this cross-sectional study consisted of HIV-infected patients who were receiving services in Mashhad and were under the supervision of Behavioral and Infectious Diseases Counseling Centers in Mashhad (BIDCC), from March to September 2021. There are four BIDCC centers in Mashhad, we considered all patients referred to these centers. The number of patients identified in this partial economic evaluation and cost analysis in Mashhad at the beginning of the study was 695 patients, which according to the recommended staff of each center, a total of 295 active patients entered our study. Out of this number, non-Iranian patients and patients who were not available at this time, as well as patients who did not want to cooperate, were excluded. Finally, 157 patients were included in the study (Fig. 1). This study was descriptive-analytical, and the perspective of society was used for computing the direct and indirect costs. The calculated expenses contained out-of-pocket costs and government and insurance payments. The monetary unit considered in this study was the US dollar.
Fig. 1Participant recruitment and follow-up
## Inclusion and exclusion criteria for selecting of sample
The inclusion criteria included officially registered subjects, who regularly visited health centers to receive services and were willing to participate in the study. Exclusion criteria included those who died during the study period.
## Direct medical costs
The data relating to medical costs including direct medical costs and direct nonmedical costs were gathered prospectively and simultaneously from several paths for better accuracy: through considering medical and financial documents of the patients, face-to-face interviews with subjects, as well as interviews with employees and specialists, who cared for a patient. The leading items of direct medical costs comprised visits by a family physician and specialist (inpatient and outpatient), diagnostic services, laboratory tests, hospital stay, other prescribed drugs, and booster medicines. The average number of every service received for inpatients was taken out from medical records and for outpatients extracted by face-to-face interviews with patients. At the end of retrieved data, the mean number of gathered data for each patient was confirmed by specialists and checked by national tariffs. The mean direct medical costs per patient in HIV disease were defined as follows: a total of three costs per patient was equal (mean number of specialist and general physician visits per three months × visit fee) + (mean number of diagnostic received services per three months × cost for every diagnostic service) + (mean number of tests per three months × cost for every test) + (the cost of each unit of prescribed drugs × the number of medicines on a three month of treatment) + (mean hospital stay per six months × cost per each day) [15].
## Direct nonmedical costs
The information on direct nonmedical costs in patients with HIV diseases was achieved using patients’ self-declaration information by patients using face-to-face or telephone interviews through a checklist that was prepared in advance to calculate costs. Direct non-medical costs include transportation (inner-city and long-distance), food and accommodation for the patient and their companions, purchase of any medical supplies and aids in the treatment process (such as wheelchairs, walkers, home care beds), and changes in their home due to illness (for example, the installation of an elevator for a person who is paralyzed following HIV / AIDS). In this study, because there was no cost to patients in other areas, only the cost of the patient’s travel was considered [16].
## Indirect costs
Indirect costs retrieve from reduced productivity of patients or family members in response to illness, death, or treatment. Productivity loss cost is due to the absence from work of patients and their families, who care for them. In total, the following items were calculated as productivity lost in this study: Number of days because of disability of the patient and companions including time spent receiving outpatient and inpatient services such as travel time, days of hospitalization, number of days of recovery after discharge. Job loss as a result of illness. Number of days spent in the hospital and nursing at home, number of days of disability of family, relatives, and friends due to patient care. The rate (percentage) of decrease in patient income due to illness.
In the present study, the human capital approach based on the minimum wage has been used to calculate indirect costs [17]. The data required to calculate this part of the costs are also obtained based on patients’ self-report through face-to-face or telephone interviews with patients and patient companions.
Finally, the indirect cost of each person and the disease status was calculated using the following formula: Minimum daily wage * Total number of days of disability of patient and companions = indirect cost.
The average national wage for laborers, who worked in Iran was calculated as 885,165 Iranian Rial (USD 1 = IRR 279,199 [Iranian Rial]) per day, and this number was multiplied by the number of lost days. Minimum monthly wage (26554950 Rial = 95.11 $) Minimum daily wage (88,165 Rial = 3.17 $).
## Determinants of AIDS
In this study, a total of 157 AIDS patients were involved based on inclusion criteria. 57.32 of the subjects were Male. The majority of participants in this study were in the age group 30–59 years ($$n = 124$$,$78.98\%$). Based on where the patients live, the majority of patients have lived in the urban region ($$n = 144$$, $91.72\%$). more than $15\%$ of patients had to change their employment status. All the socioeconomic and descriptive statistics of the studied subjects were demonstrated in Table 1.
Table 1Frequency of some demographic characteristics (based on sex) of patients participating in the studyDemographic characteristicsMalenumber(percentage)Femalenumber (percentage)90(57.32)67 (42.68)AgeUnder 5 years02 (2.99)5 to 17 years5 (5.56)6 (8.96)18 to 29 years2 (2.22)6 (8.96)30 to 59 years75 (83.33)49 (73.13)upper 60 years8 (8.89)4 (5.97)Place of residence of the patient in terms of urban/ruralUrban85(94.44)59(88.06)Rural5(5.56)8(11.94)Marital statusmarried44(48.89)40(59.70)Divorced22(24.44)6(8.96)deceased wife4(4.44)13(19.44)Single20(20.22)8(11.94)Educationilliterate3(3.33)9(13.43)High school61(67.78)41(61.19)diploma23(25.56)15((22.39)College education3(3.33)2(2.99)Job change due to diseaseyes16(17.78)8(11.94)no74(82.22)59(88.06)Unemployment due to HIVyes14(15.56)8(11.94)no76(84.44)59(88.06)Having insuranceyes70(77.78)59(88.06)no20(22.22)8(11.94)Insurance typeno insurance20(22.22)8(11.94)Iranian health insurance48(53.33)41(61.19)social security insurance20(22.22)17(25.37)Armed forces insurance2(2.22)1(1.49)Supplementary insuranceyes8(8.89)4(5.97)no82(91.11)63(94.03)Head of Householdyes62(68.89)14(20.90)no28(31.11)53(79.10)Fixed place to liveyes87(96.67)64(95.52)no3(3.33)3(4.48) The majority of patients were in stage 1 HIV diseases ($$n = 135$$, $85.99\%$). This amount in men and women is equal to $$n = 76$$, $84.44\%$, and 59,$88.06\%$, respectively (Table 2). According to the results of the study, the most common way to transmit this disease is through unprotected sex ($30.57\%$) then an infected spouse ($28.03\%$), and then injecting drugs ($21.02\%$). So more than $40\%$ of the studied men contracted the disease through sexual intercourse. Among women, the most common cause of infection was through an infected spouse (Table 3).
Table 2Frequency of some demographic characteristics (based on gender) of patients participating in the study (stages of the disease)CharacteristicsMaleFemalenumber (percentage)number (percentage)90(57.32)67 (42.68)Stage of the disease176(84.44)59(88.06)26(6.67)4(5.97)36(6.67)4(5.97)42(2.22)0 Table 3Frequency of some demographic characteristics (based on gender) of patients participating in the study (transmission route)CharacteristicsMaleFemalenumber (percentage)number (percentage)90(57.32)67 (42.68)Transmission categoryInjecting drug use32(35.56)1(1.49)Unsafe sex40(44.44)8(11.94)Transmission through an infected spouse3(3.33)41(61.19)Transmission from Mother-to-child6(6.67)9(13.43)Transmitted through blood2(2.22)3(4.48)Transmission through dental visits1(1.11)0Transmission through tattoos01(1.49)Unknown6(6.67)4(5.97) The results show that $41.4\%$ of the patients in the study have an addiction history. Among the patients who had an addiction, about $26.75\%$ had a history of injecting drugs, and among those who had a history of injection, $19.75\%$ had a history of joint injection. Also, $36.94\%$ of the patients had a history of prison, $44.59\%$ of the patients had a history of unprotected sex, and $27.39\%$ had multiple sexual partners (Table 4).
Table 4Frequency of some demographic characteristics (based on gender) of patients participating in the study (specific records)Characteristicsmalefemalenumber (percentage)number (percentage)90(57.32)67 (42.68)*Having a* history of addictionyes60(66.67)5(7.46)no30(33.33)62(92.54)*Having a* history of injectionyes39(43.33)3(4.48)no51(56.67)64(95.52)*Having a* history of joint injectionyes29(32.22)2(2.99)no61(67.78)65(97.01)*Having a* prison recordyes54(60.00)4(5.97)no36(40.00)63(94.03)Having unsafe sexyes57(63.33)13(19.40)no33(36.67)54(80.60)History of sexual relations with non-spouseyes55(61.11)12(17.91)no35(38.89)55(82.09)Having multiple sexual partnersyes35(38.89)8(11.94)no55(61.11)59(88.06)History of venereal diseaseyes8(8.89)2(2.99)no82(91.11)65(97.01)
## Cost
Considering that the government covers almost all the necessary treatment needs of HIV/AIDS patients, we have divided the costs into two parts: the patient’s share and the government’s share, and calculated each one separately. *In* general, the highest cost of this disease is attributed to medicine (10339.32 $ for 6 months), and of this cost, 6988.67 $ was the government’s share, and 3350.65 $ was the patient’s share. After medicine, the cost of tests was 9101.22 $, of which 8464.61 $ was the government’s share and 636.6 $ was the patients’ share. Among the expenses of the patients, hospitalization and dental costs were 1485.67 $ and 770.52 $, respectively for all patients in 6 months (Table 5).
Table 5Direct medical costs* (During 6 months)Type of serviceAverage direct medical costs per patient)patient share(Average direct medical costs per patient)GovernmentShare(Total direct medical costs)Patient share(Total direct medical costs)Government share(Total direct medical costs)Patient and government costs(Specialist visit2.07 ± 3.583.02 ± 1.17303.95471.37775.33Medicine25.46 ± 53.2744.51 ± 15.153350.656988.6710339.32Psychologist09.18 ± 0.7401193.771193.77Midwife visit0.12 ± 0.431.07 ± 0.573.5830.2133.79Experiments5.41 ± 13.3460.89 ± 55.69636.608464.619101.22Dental62.52 ± 77.870770.520770.52Emergency services5.65 ± 4.750130.010130.01Radiology18.68 ± 22.480276.680276.68Sonography5.92 ± 6.26044.37044.37Physiotherapy132.88 ± 159.690398.640398.64Hospitalization116.48 ± 147.905.51 ± 19.861485.6771.631557.31Chemotherapy130.73 ± 00130.730130.73Condom1.59 ± 2.624.18 ± 4.07130.98355.66486.64Methadone8.43 ± 0.8246.41 ± 5.08210.601253.081463.68*The monetary unit considered in this study was the US dollar In Table 6, the average cost per patient and the cost of the total days lost due to HIV/AIDS are calculated separately for accompanying patients and caregivers, and the largest share of this cost is for the patient (1696.9 ± 848.28 and in general, the cost of lost days for all patients in these 6 months is 12794.56 $, the largest share of which is related to the patients themselves with a cost of about 11583.48 $.
Table 6The average cost* per patient and the cost of the total days lost due to HIV/AIDSType of serviceTotal number of days lostAverage per patient(day)Average cost lost during 6 months per patientTotal cost lost during 6 monthsDays lost due to illness (patient)498931.7848.28 ± 1696.9011583.48Days lost due to illness (accompanying the patient)480.38.40 ± 27.0757.06Days lost due to illness (Caring for the patient)3712.337.23 ± 310.341154.01Total540834.4745.4912794.56*The monetary unit considered in this study was the US dollar Table 7 also shows the direct and indirect medical costs as well as the indirect costs due to the lost productivity of the patients separately between the patients with addiction and the patients without addiction, as we can see the obvious difference between the patients with a history there is addiction and no history of addiction.
Table 7Direct medical costs and Indirect costs* by patients with a history of addiction and without a history of addictionType of serviceAddictionNo addictionDirect medical costs(Patient share)65.05 ± 115.8351.31 ± 106.03Direct medical costs(Government share)129.57 ± 70.6113.11 ± 55.44Direct nonmedical costs(Patient share)23.07 ± 67.5031.65 ± 44.30Direct nonmedical costs(Government share)00Days lost due to illness(patient)1313.33 ± 2039.29519.72 ± 1322.05Days lost due to illness (accompanying the patient)1.70 ± 7.0913.14 ± 34.14Days lost due to illness (Caring for the patient)41.65 ± 287.1834.11 ± 327.22The percentage of attracting income to HIV AIDS0.2 ± 0.330.19 ± 0.3The percentage of income reduction due to HIV AIDS0.44 ± 0.420.18 ± 0.33*The monetary unit considered in this study was the US dollar
## Discussion
This study was the first comprehensive study that has considered the economic burden and Determinants of HIV/AIDS on patients in Iran with a prospective approach from a societal perspective. In this study, the costs are separated into two parts: the patient’s share and the government’s share. The reason for this is that since HIV/AIDS is a special and contagious disease, the costs related to the medical services of patients, including tests, doctor’s visits, medicine, consultation, etc. are offered to patients for free, are all over the world. On the other hand, due to the specific conditions of the disease and the effect that this disease has on all the organs of the body, as well as the weakness of the immune system due to this disease and the need to strengthen the body, patients also pay expenses out of their pockets for getting the additional therapy. In this study, an attempt has been made to calculate all these costs. During these six months, it was estimated that 39844.7 $ of direct and indirect costs for HIV/AIDS disease for 157 patients participating in the research, that the government spent 18829.03 $ for participating patients with HIV/AIDS, and the patients also spent 8221.1 $ have paid from their own pockets for direct medical and non-medical expenses, which is 21015.66 $ for the patient, if we have included the days lost due to disability due to illness.
Average total productivity losses (Days lost due to illness of patients, accompanying the patient, and Caring for the patient) because of HIV/AIDS was to be high in a monthly period (5.73 days). The findings of this study are in line with the findings of the Nepal study so in their study Average total productivity losses were found 5.05 days [18]. In our study, the proportion of average total productivity losses to total costs (sum direct cost and indirect cost and productivity loss) of this study was $31.1\%$. while in the Nepal study average, total productivity losses to total costs were found $32.3\%$ [18].
According to the results, there was the cost per patient varied based on disease stage. Patients in stage one incurred the major cost for society and the health system and stage four incurred the minimum cost. Also, the highest cost is the direct cost related to antiviral drugs. Based on the current study findings, in general10339.32$ were spent on drugs by the government and patients, of which 6988.67$ were paid by the government. And 3350.65$ have been spent by the patients. The average cost of antiviral drugs per patient is estimated to be approximately 25.46 ± 53.27 and 44.51 ± 15.15 $, respectively, for the patient’s share and the government’s share. Even though no related study has been carried out on this disease in patients with HIV in Iran, the results of this study are consistent with the results of other studies in this field. A study conducted by Julien Kuhlmann and colleagues in Bogota, Colombia in 2017, showed that the cost of providing medicine is the largest for HIV/AIDS. In this study, the drug cost for each patient was estimated at an average of 8616 US dollars, which includes $75\%$ of the total costs [19]. Also, our study is consistent with the results of a study conducted by Sarah Mostardt et al. The results of this study showed that the cost of antiviral drugs is the most expensive in HIV + disease [20]. Generally, the results of other studies in North America and Europe are consistent with our results based on the percentage of spending on drugs [20–22]. Poudel et el. in their study found that the highest cost of the direct cost was related to the Cost of diagnosis or test, so their results were not in line with the results of our study [18]. Briefly, it seems that there are two reasons for the high cost of medicines in HIV/AIDS patients: First, the complaint of most patients with this disease is the breakdown of the immune system, which forces patients to use strengthening medicines that these medicines constitute a large part of patients’ expenses. Second, in addition to the HIV/AIDS disease itself, these patients gradually suffer from a variety of chronic diseases, which may not happen in healthy people or may occur in old age. For patients with HIV/AID, the same issue brings the cost of medicine for all kinds of diseases, which is also expensive, and in general, it is because the cost of medicine is higher than other services.
Because many patients with HIV, are obliged to visit health care centers far from their place of dwelling to have necessary treatment, the direct and indirect nonmedical costs imposed on the patient and the government are increased. Furthermore, the indirect costs of getting treatment for these patients are very high, because many patients are forced to request long-term leave or dismissal from their workplace due to social issues. So, in this study results demonstrated that per each patient, there were 31.7 days away from work. According to the results of this study, a large percentage of patients are of working age and are less than 60 years old, so, this disease imposes a large cost on society due to lost productivity.
We had some limitations in this study. First, in cost estimation, efforts have been made to calculate different types of costs such as direct treatment and non-treatment costs as well as indirect costs. But intangible costs, including pain, suffering, stress and anxiety, and social stigma of patients and those around them, especially in the severe stages of the disease, which exist to a significant extent, have not been included in the calculations. However, due to the lack of accurate and appropriate calculation methods, such costs are usually not included in economic burden studies. The second weakness is that we could not include a large number of patients in the study due to a lack of access and lack of cooperation.
Because a large number of patients participating in the study had a history of addiction, a comparison was made between patients with and without a history of addiction in this study, which showed that patients with a history of addiction incurred more costs. Also, the comparison of the cost of lost days due to illness showed that the number of days lost by patients who had an addiction and their companions and caregivers is far more than those who do not have a history of addiction.
The strength of the present study was using the bottom-up approach, in which the researchers could have collected accurate information on the direct and indirect costs of HIV patients. The fact that the cost information was collected natively and through the self-declaration of patients and specialists by the researcher from the centers under study was based on observation, and can be another positive point in the current research. In the end, despite all the difficulties of accessing these patients and building trust in them, this research happened for the first time in Iran. The limitations of the present study was the sample consisted of HIV-infected patients under the supervision of BIDCC centers. As a probable source of bias, the sample might underrepresent the newly-diagnosed cases and subjects of high socioeconomic status, due to the social stigma in a religious city such as Mashhad.
## Conclusion
The results of the present study showed that HIV/AIDS can be considered a disease that imposes a large economic burden on the government and patients from society’s point of view. It involves society, and in addition to the treatment issues, it can affect all the economic and social issues of society, therefore, as many measures as possible for the prevention and timely diagnosis of this disease should be put on the agenda of the government and the policymakers should prioritize the health system of HIV/AIDS. Furthermore, it seems that to reduce costs for patients with disease HIV/AIDS, the focus should be on diagnostic tests and care. Early diagnosis and rapid initiation of antiviral treatments can be effective in preventing serious and debilitating diseases. The costs of early diagnosis and then early treatment are far less than the costs of treatment and reduced productivity. Having special insurance for these patients can greatly reduce the out-of-pocket payments of these patients. To reduce costs, we should pay attention to treatment available to all patients, provide services at the regional level, and set up more comprehensive service centers for the care of HIV/AIDS patients. We can consider private subsidies for patients and public hospitals. prepare for low-cost diagnostic tests, especially for people who live on the outskirts of cities and pregnant women at the time of delivery. These policies not only reduce the economic burden of the disease but also encourage patients to test and start treatment. A complete policy for livelihood support of patients can be skill development and income generation programs.
## Electronic supplementary material
Below is the link to the electronic supplementary material.
Supplementary Material 1. The data on the Determinants and economic burden of HIV/AIDS in Iran in 2022
## References
1. 1.UNAIDS data. 2019. https://www.unaids.org/sites/default/files/media_asset/2019-UNAIDS-data_en.pdf. Accessed 15 Jun 2019.
2. 2.data U. 2020. https://www.unaids.org/sites/default/files/media_asset/2020_aids-data-book_en.pdf. Accessed 16 Mar 2022.
3. 3.Global AIDS Monitoring, Geneva. UNAIDS; 2019. https://www.unaids.org/sites/default/files/media_asset/global-aids-monitoring_en.pdf. Accessed 23 Jun 2020.
4. 4.Frank TD, of HIV et al. 1980–2017, and forecasts to 2030, for 195 countries and territories: a systematic analysis for the Global Burden of Diseases, Injuries, and Risk Factors Study 2017. The lancet HIV, 2019. 6(12): p. e831-e859.
5. 5.WHO. Global Health Observatory (GHO) data. https://www.who.int/gho/hiv/en/. Accessed 16 Mar 2022.
6. 6.Center for Communicable Diseases Control., Ministry of Health and Medical Education. Latest statistics on HIV infection in Islamic Republic of Iran, April to June 2017. 2018.
7. 7.Collins DL, Leibbrandt M. The financial impact of HIV/AIDS on poor households in South Africa. 2007, LWW. p. S75-S81.
8. Kumarasamy N. **Financial burden of health services for people with HIV/AIDS in India**. *Indian J Med Res* (2007.0) **126** 509. PMID: 18219077
9. 9.Cohen, D., Poverty and HIV/AIDS in sub-Saharan Africa. 2000: Social development & Poverty Elimination Division, Bureau for Developmnent.
10. Hecht R. **Putting it together: AIDS and the millennium development goals**. *PLoS Med* (2006.0) **3** e455. DOI: 10.1371/journal.pmed.0030455
11. 11.Segel JE. Cost of illness studies—A primer: RTI International, RTI-UNC Center of Excellence in Health Promotion Economics. North Carolina, US. 2006.
12. Roehrborn C. **First-year costs of treating prostate cancer: estimates from SEER-Medicare data**. *Prostate Cancer Prostatic Dis* (2009.0) **12** 355-60. DOI: 10.1038/pcan.2009.21
13. Jo C. **Cost-of-illness studies: concepts, scopes, and methods**. *Clin Mol Hepatol* (2014.0) **20** 327. DOI: 10.3350/cmh.2014.20.4.327
14. 14.Statistical Centre of Iran. https://www.amar.org.ir. Accessed 23 Nov 2022.
15. Mojahedian MM. **Estimating the cost of illness of prostate cancer in Iran**. *Clin Ther* (2019.0) **41** 50-8. DOI: 10.1016/j.clinthera.2018.11.001
16. Javanbakht M. **Cost-of-illness analysis of type 2 diabetes mellitus in Iran**. *PLoS ONE* (2011.0) **6** e26864. DOI: 10.1371/journal.pone.0026864
17. McDonald S, Roberts J. **AIDS and economic growth: a human capital approach**. *J Dev Econ* (2006.0) **80** 228-50. DOI: 10.1016/j.jdeveco.2005.01.004
18. Poudel AN, Newlands D, Simkhada P. **The economic burden of HIV/AIDS on individuals and households in Nepal: a quantitative study**. *BMC Health Serv Res* (2017.0) **17** 1-13. DOI: 10.1186/s12913-017-1976-y
19. 19.Kuhlmann J et al. A cost-of-illness study of patients with HIV/AIDS in Bogotá, Colombia. Value in health regional issues, 2017. 14: p.103–107.
20. Mostardt S. **Cost of HIV and determinants of health care costs in HIV-positive patients in Germany: results of the DAGNÄ K3A study**. *Eur J Health Econ* (2013.0) **14** 799-808. DOI: 10.1007/s10198-012-0425-4
21. 21.Beck EJ et al. The cost of treatment and care for people living with HIV infection: implications of published studies, 1999–2008. Current Opinion in HIV and AIDS, 2010. 5(3): p.215–224.
22. Lopez-Bastida J. **The economic costs and health-related quality of life of people with HIV/AIDS in the Canary Islands, Spain**. *BMC Health Serv Res* (2009.0) **9** 1-8. DOI: 10.1186/1472-6963-9-55
|
---
title: Differential expression of pyroptosis-related genes in the hippocampus of patients
with Alzheimer’s disease
authors:
- Pengcheng Xia
- Huijun Ma
- Jing Chen
- Yingchao Liu
- Xiaolin Cui
- Cuicui Wang
- Shuai Zong
- Le Wang
- Yun Liu
- Zhiming Lu
journal: BMC Medical Genomics
year: 2023
pmcid: PMC10012531
doi: 10.1186/s12920-023-01479-x
license: CC BY 4.0
---
# Differential expression of pyroptosis-related genes in the hippocampus of patients with Alzheimer’s disease
## Abstract
### Background
Alzheimer’s disease (AD) is a progressive, neurodegenerative disorder with insidious onset. Some scholars believe that there is a close relationship between pyroptosis and AD. However, studies with evidence supporting this relationship are lacking.
### Materials and methods
The microarray data of AD were retrieved from the Gene Expression Omnibus (GEO) database with the datasets merged using the R package inSilicoMerging. R software package Limma was used to perform the differential expression analysis to identify the differentially expressed genes (DEGs). We further performed the enrichment analyses of the DEGs based on Gene Ontology (GO) and the Kyoto Encyclopedia of Genes and Genomes (KEGG) databases to identify the metabolic pathways with a significant difference. The Gene Set Enrichment Analysis (GSEA) was applied to identify the significant pathways. The protein-protein interaction (PPI) network was constructed based on the STRING database with the hub genes identified. Quantitative real-time PCR (qRT-PCR) analyses based on HT22 cells were performed to validate the findings based on the microarray analysis. Gene expression correlation heatmaps were generated to evaluate the relationships among the genes.
### Results
A new dataset was derived by merging 4 microarray datasets in the hippocampus of AD patients in the GEO database. *Differential* gene expression analysis yielded a volcano plot of a total of 20 DEGs (14 up-regulated and 6 down-regulated). GO analysis revealed a group of GO terms with a significant difference, e.g., cytoplasmic vesicle membrane, vesicle membrane, and monocyte chemotaxis. KEGG analysis detected the metabolic pathways with a significant difference, e.g., *Rheumatoid arthritis* and Fluid shear stress and atherosclerosis. The results of the Gene Set Enrichment Analysis of the microarray data showed that gene set ALZHEIMER_DISEASE and the gene set PYROPTOSIS were both up-regulated. PPI network showed that pyroptosis-related genes were divided into two groups. In the Aβ-induced HT22 cell model, three genes (i.e., BAX, IL18, and CYCS) were revealed with significant differences. Gene expression correlation heatmaps revealed strong correlations between pyroptotic genes and AD-related genes.
### Conclusion
The pyroptosis-related genes BAX, IL18, and CYCS were significantly different between AD patients and normal controls.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12920-023-01479-x.
## Introduction
Alzheimer’s disease (AD) is the most common neurodegenerative disease of the elderly, and its incidence increases with age in populations around the world [1]. For example, in 2017, approximately 6.08 million Americans were diagnosed with clinical AD or mild cognitive impairment due to AD, and this number is expected to reach 15 million by 2060.[2]. The most typical clinical manifestation of AD is the gradual decline of cognitive ability in patients, which appears in the early stage of AD without the two pathological signs of senile plaques and neurofibrillary tangles [3, 4]. Although the molecular mechanisms of these two pathological changes have been well studied, therapeutic strategies targeting these changes have not been successful in the treatment of AD. So far, there are no effective drugs to prevent or treat cognitive decline in AD patients.
Pyroptosis, also known as inflammatory necrosis, is a type of programmed cell death characterized by the continuous expansion of cells until the cell membrane ruptures, resulting in the release of cellular contents and the activation of a strong inflammatory response [5]. As a new type of programmed cell death discovered and confirmed in recent years, pyroptosis is characterized by its dependence mainly on caspase-1, caspase-4, caspase-5, and caspase-11 and accompanied by the release of a large number of pro-inflammatory factors [6]. The morphological characteristics, occurrence, and regulatory mechanism of pyroptosis are different from those of other programmed cell death conditions such as apoptosis and necrosis [7]. Pyroptosis mainly relies on the activation of a group of proteins of the caspase family by the inflammasome, cleaving and activating the gasdermin protein, which is translocated to the membrane to form holes and to make cell swell, causing the cytoplasmic outflow and finally leading to cell membrane rupture and pyroptosis [8]. Studies have shown that pyroptosis is widely involved and plays important roles in the occurrence and development of infectious diseases, nervous system-related diseases, and atherosclerotic diseases [9–11]. Pyroptosis also plays important role in AD. For example, studies have shown that amyloid-β induces NOD-like receptor (NLR) family pyrin domain-containing 1 (NLRP1)-dependent neuronal pyroptosis in a mouse model of AD [12], Parkinson disease protein 7 (PARK7/DJ-1) affects oxidative stress and pyroptosis in hippocampal neurons of a mouse model of AD by regulating the nuclear factor-erythroid 2-related factor 2 (Nrf2) pathway [13], while schisandrin inhibits NLRP1 inflammasome-mediated neuronal pyroptosis in a mouse model of AD [14]. To date, the molecular mechanisms regulating the development of pyroptosis are still unclear. Therefore, it is important to identify and investigate the pyroptosis-related genes differentially expressed in AD, to help understand the occurrence of pyroptosis in related diseases, and to further explore their functions in the development, prognosis, and clinical prevention and treatment of these diseases.
In our study, in order to explore the roles of pyroptosis-related genes in AD, microarray datasets from the hippocampus of AD patients were first merged to identify the differentially expressed genes (DEGs) related to AD. The DEGs were further annotated and enriched based on Gene Ontology (GO; http://geneontology.org/) and Kyoto Encyclopedia of Genes and Genomes (KEGG; https://www.genome.jp/kegg/) databases. Through the Gene Set Enrichment Analysis (GSEA) database, the overall gene expression variations associated with AD and pyroptosis were detected. The protein-protein interaction (PPI) network was constructed based on the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING; https://string-db.org/) database to analyze the relationships among the pyroptosis-related genes. The findings were validated with both the dataset GSE48350 and the cellular models of AD, showing that the expressions of genes associated with pyroptosis were significantly altered in AD, providing novel insights into the pathogenesis and potential clinical treatment of AD.
## Data preparation
Gene expression profiles of AD were obtained from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/; accessed April 23, 2021). The sample informations involved in all GEO datasets in this study are in Supplementary File S2. The GSE36980 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA157435) dataset contains gray matter RNA samples from the frontal, temporal cortex, and hippocampus of 88 postmortem brains, 26 of which were pathologically diagnosed AD or AD-like disorder. High-quality RNA (RIN ≥ 6.9) samples were subjected to microarray analysis using the Affymetrix Human Gene 1.0 ST platform, and only results that passed the Human Gene 1.0 ST array quality control check were retrieved. In total, gene expression profiles were collected from three sets of samples: 33 frontal cortex samples (15 from AD patients), 29 temporal cortex samples (10 AD patients), and 17 hippocampal samples (7 AD patients). In particular, in the dataset GSE1297 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA90219), we have analyzed 9 controls and 22 different severities based on 31 independent microarrays of *Hippocampal* gene expression in AD subjects and correlation of these gene expressions with MiniMental Status Examination (MMSE) and neurofibrillary tangles (NFT) scores in all 31 subjects. In dataset GSE28146 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA139561), where the major white matter tracts have been excluded using laser capture microdissection, we extracted formalin from the same subjects’ CA1 hippocampal gray matter was selectively collected from fixed, paraffin-embedded (FFPE) hippocampal sections. The samples in GSE29378 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA140105) were based on total RNA from 60 μm frozen human hippocampal sections. Control and AD brains were well matched for all non-disease characteristics. Both CA1 and CA3 sections of the same individual were taken from the same section. Several regional and disease-related comparisons were made. Four datasets (GSE36980, GSE1297, GSE28146, and GSE29378) were combined for deg detection in the hippocampus. The dataset GSE 48,350 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA209800) contains cases from normal controls (NC; ages 20 to 99 years) and AD. The expression changes of synaptic and immune-related genes were analyzed, and the age-related changes, AD-related changes, and region-specific change patterns of gene expression were investigated. These AD cases were processed concurrently with controls (young and old) in dataset GSE11882, which only contains data from normal controls. The dataset GSE 48,350 was used to validate the DEGs identified in the hippocampus. The final data were obtained by combining multiple datasets using the R package from silicomerging [15] to generate a single data matrix (Table S1) and further processing the data matrix using the method of Johnson et al. [ 16] to remove the batch effects matrix. ( Table S2).
## Differential gene expression analysis
Limma [17] is a differential quality articulation screening technique in light of summed up straight models. We played out the differential investigation in light of the R programming bundle Limma (Form 3.40.6) to acquire the DEGs between various examination gatherings and control gatherings. In particular, we originally played out the Log2 change of the articulation range dataset and afterward utilized lmFit capability to play out the numerous straight relapse examination. We further utilized eBays capability to compute the directed t-insights, directed f-measurement, and log-chances of differential articulation by observational Bayes balance of the standard blunders towards a typical value, and lastly got the massive distinction of every quality. The changed P-esteem was broken down to address the misleading positive outcomes in the GEO datasets. The boundaries " Adjusted $P \leq 0.05$ and Log2 (Fold Change) > 0.6 or < − 0.6” were characterized as the edges for the screening of differential articulation of mRNAs. The crate plot and heatmap were produced by the capabilities ggplot2 and heatmap, separately, of the R programming bundle.
## GO annotation and KEGG pathway enrichment analysis
We played out the improvement investigations of the DEGs distinguished by Limma in light of KEGG [18] rest Programming interface (https://www.kegg.jp/kegg/rest/keggapi.html) to recognize the metabolic pathways enhanced with massive distinction. The GO comments of the DEGs were performed in view of the R bundle org.hs.eg.db (Version 3.1.0) as the foundation and the R programming bundle clusterProfiler (Version 3.14.3) to acquire the quality sets enhanced with tremendous contrast in light of $P \leq 0.05$ and false discovery rate (FDR) < 0.25. The base and most extreme qualities were set to 5 and 5000, separately.
To further explore the functions and relevant pathways of the potential target genes, the DEGs were analyzed by functional enrichment analyses and KEGG analysis (Fig. 2; Table S3). GO analysis revealed these DEGs mainly focused on the functions such as cytoplasmic vesicle membrane, vesicle membrane, monocyte chemotaxis, cytoplasmic vesicle part, whole membrane, cytoplasmic vesicle, intracellular vesicle, mononuclear cell migration, lateral plasma membrane, and cell chemotaxis. KEGG analysis revealed the following metabolic pathways with a significant difference: Rheumatoid arthritis, Fluid shear stress and atherosclerosis, Type I diabetes mellitus, Leishmaniasis, MAPK signaling pathway, Th1 and Th2 cell differentiation, IL-17 signaling pathway, Hematopoietic cell lineage, Chagas disease (American trypanosomiasis), Th17 cell differentiation, TNF signaling pathway, and Yersinia infection. These results may indicate that immunity, inflammation, and metabolic abnormalities could participate in the occurrence and progress of AD.
Fig. 2KEGG and GO analysis of differentially expressed genes (DEGs). ( A) KEGG analysis of DEGs. ( B) Biological Process (BP) terms of GO analysis of DEGs. ( C) Cellular Component (CC) terms of GO analysis of DEGs. ( D) Molecular Function (MF) terms of GO analysis of DEGs.
## Gene set enrichment analysis
This GSEA (https://www.gseamsigdb.org/gsea/index.jsp) is commonly used to determine statistically significant differences between two biological states (e.g., phenotypes) in an innately defined set of genes [19]. In our study, GSEA was applied to identify important pathways in the merged datasets. The Spearman correlation coefficient between genes and sample labels is defined as the weight of genes [20]. Statistical significance was assessed by comparing the enrichment scores to the enrichment results generated by random permutation of 1000 gene sets to obtain nominal P-values. The significance level of metabolic pathways was determined by normalized enrichment score (NES) ≥1.0, FDR ≤ 0.25, P ≤ 0.05.
## Protein-protein Interaction (PPI) analysis
A PPI network based on protein-protein interaction (PPI) analysis was established in the STRING database (Version 11.0; http://string-db.org/) [21].
## Validation of pyroptosis-related genes
Validation of the pyroptosis-related genes identified in the microarray datasets was performed based on dataset GSE48350. The pyroptosis-related genes in GSE48350 were compared using the Wilcoxon test. A total of five datasets (i.e., GSE1297, GSE28146, GSE29378, GSE36980, and GSE48350) were used to investigate the association between pyroptosis-related genes and AD-related genes. The two-gene and multiple-gene correlation maps were generated by the R software packages ggstatsplot heatmap, respectively. Spearman’s correlation analysis was performed to analyze the correlations between quantitative variables without normal distributions with the significant difference set to $P \leq 0.05.$ The quantitative real-time PCR (qRT-PCR) analysis was performed using the mouse hippocampal neuron cell line HT22 as the validated cell model induced by 10 µM Aβ1–42 (P9001, rPeptide, Beyotime, Beijing, China) to verify the expression patterns of pyroptosis-related genes revealed in the microarray analysis. The primer sequences were synthesized by RIBOBIO Corporation, Guangzhou, China (File S1). Trizol (Thermo Fisher Scientific Inc., MA, USA) method was employed to extract the total RNA from HT22 cells in each group according to the manufacturer’s protocol. The total RNA (1 µg) was reverse-transcribed to cDNA by use of PrimeScript RT Reagent Kit with gDNA Eraser (Accurate Biotechnology Co., Ltd., Hunan, China). *All* genes involved in the experiment were examined by a quantitative real-time PCR amplifier (Applied Biosystems QuantStudio 5, ABI Company, Oyster Bay, New York, USA) with SYBR® Premix Ex Taq (Accurate Biotechnology Co., Ltd., Hunan, China). PCR procedure: pre-denaturation at 95 °C for 30 s, 40 cycles of denaturation at 95 °C for 5 s, annealing at 60 °C for 30 s, extension at 72 °C for 30 s, and finally melting at 95 °C for 30 s.
Lastly, the expression levels of the pyroptosis-related genes identified in the dataset were further evaluated. Compared with the NC group, the expression of CASP5 (Fig. 6A) and IL18 (Fig. 6D) in the AD group was increased in GSE1297 and GSE48350 respectively, while the expression of CYCS (Fig. 6A), IL1B (Fig. 6B) and CASP1(Fig. 6E) was decreased in GSE1297, GSE28146, and GSE36980 respectively, and the expressions of CHMP7, CHMP2A, and CYCS were decreased in GSE48350 (Fig. 6D). In GSE29378, pyroptosis-related genes showed no statistical differences (Fig. 6C). In the Aβ-induced HT22 cell model of AD, a total of three genes (i.e., BAX, IL18, and CYCS) showed a significant difference in their expressions (Fig. 6F, Table S8). These results confirmed that pyroptosis-related genes may participate in the occurrence of AD.
Fig. 6Expression levels of pyroptosis-related differentially expressed genes (DEGs) in five datasets of GSE1297 (A), GSE28146 (B), GSE 29,378 (C), GSE48350 (D), and GSE36980 (E), and the HT22 cell models of Alzheimer’s disease (AD) (F)
## Statistical analysis
Statistical analysis was performed using GraphPad Prism (version 8.0.0). The data of the GEO dataset were tested for normality and homogeneity of variance. Data that passed these two tests were compared between the two groups using the t-test. P values less than 0.05 were considered statistically significant.
## Analysis of differentially expressed genes (DEGs) in the combined datasets
Firstly, we merged four gene sets (Table S1) based on the number of genes in the dataset (Fig. 1A). And then, we remove the batch effect between these gene sets. The Uniform Manifold Approximation and Projection (UMAP) plot showed these changes before and after removal (Fig. 1B, C). These are also displayed by the boxplot which showed that the data distributions between the datasets become much more consistent, i.e., the medians existed along the same line (Fig. 1D, E; Table S2). Next, the dataset after the batch effect removal was executed for the differential gene analysis. The volcano plot showed that a total of 20 DEGs existed, including 14 up-regulated and 6 down-regulated (Table 1; Fig. 1F). Their relative expression levels between samples are displayed in the cluster map and heatmap in Fig. 1G.
Fig. 1Analysis of differentially expressed genes (DEGs) based on combined datasets. ( A) Characteristics of combined datasets. ( B) The Uniform Manifold Approximation and Projection (UMAP) plot before dataset merging. ( C) UMAP plot after dataset merging. ( D) Relative expression levels of genes before dataset merging. ( E) Relative expression levels of genes after dataset merging. ( F) Volcano plot of DEGs after dataset merging. ( G) Heatmap of DEGs after dataset merging Table 1The 20 differentially expressed genes (DEGs) were identified in patients with Alzheimer’s disease (AD).GeneLog2(Fold Change)P-ValueAdjusted P-ValueRegulationReportedBy GAD1 –0.86732.26E-050.0033DownLi S, et al. [ 22] RGS4 –0.73497.44E-050.0062DownMuma N, et al. [ 23] MCTP1 –0.73283.19E-050.0041DownKim K, et al. [ 24] DDX3Y –0.73140.00020.0109DownVakilian H, et al. [ 25] NEFM –0.68170.00030.0145DownGeorge C, et al. [ 26] CACNA2D3 –0.62880.00380.0523DownHuang C, et al. [ 27] RAB13 0.58650.00010.0088UpZhang X, et al. [ 28] ABCA1 0.58940.00030.0159UpWahrle S, et al. [ 29] DUSP1 0.59970.00010.0090UpLeandro G, et al. [ 30] AEBP1 0.60720.00010.0097UpPiras I, et al. [ 31] CCL2 0.62490.00060.0212UpHartlage-Rübsamen M, et al. [ 32] ANXA1 0.64040.00010.0088UpMcArthur S, et al. [ 33] NUPR1 0.65923.34E-070.0002UpMontero-Calle A, et al. [ 34] SLC14A1 0.67039.07E-070.0004UpRecabarren D, et al. [ 35] SERPINA3 0.68125.21E-050.0051UpNorton E, et al. [ 36] RASL12 0.71377.37E-070.0003UpMirza Z, et al. [ 37] EMP1 0.76427.62E-089.73E-05UpGhani M, et al. [ 38] CD44 0.86214.77E-087.82E-05UpUberti D, et al. [ 39] HLA-DQA1 1.03640.00120.0300UpZhang X, et al. [ 40] FOS 1.07452.35E-050.0033UpChoi H, et al. [ 41]
## Gene Set Enrichment Analysis (GSEA)
To explore the merged hippocampal dataset more comprehensively, we conducted the GSEA. The results showed that the gene set ALZHEIMER_DISEASE was higher expressed in the AD group than in the NC group (Fig. 3A, B; Table S4). The relative expression levels of the representative genes between the samples were shown in Fig. 3A. Considering the DEGs enriched in inflammation-related pathways, we carried out GSEA analysis based on the gene set PYROPTOSIS. Interestingly, the gene set ALZHEIMER_DISEASE and PYROPTOSIS from the merged hippocampal dataset showed similar expression trends(Fig. 4A, B; Table S5), implying that pyroptosis-related genes may participate in the occurrence of AD.
Fig. 3GSEA analysis of merged datasets. ( A) Differentially expressed gene (DEG) sets between Alzheimer’s disease (AD) and NC groups. ( B) Heatmap of AD gene sets between AD and NC groups Fig. 4GSEA analysis and protein-protein interaction (PPI) network of the pyroptotic gene set. ( A) GSEA analysis of the pyroptotic gene set. ( B) Heatmap of the pyroptotic gene set. ( C) PPI network of the pyroptotic gene set
## PPI network construction
Subsequently, all the pyroptosis-related genes were further analyzed by the STRING database to construct the PPI network (Fig. 4C; Table. S6). A total of 12 nod genes (i.e., CASP5, BAX, CASP4, IRF2, IRF1, HMGB1, IL18, IL1A, IL1B, CASP1, CASP3, and CYCS) were revealed in the same collective, while another three nod genes (i.e., CHMP2A, CHMP6, and CHMP7) were defined in another collective.
## Correlation between pyroptosis-related genes and AD-related genes
Then we carried out the correlation exploration between pyroptosis-related genes from GSEA and AD-related genes from Disgenet in the datasets (Table S7). A large number of genes between them showed significant correlations. IRF1 was most positively correlated with ACE in GSE1297 (Fig. 5A), IRF2 was most negatively correlated with MAPT in GSE28146(Fig. 5B), IL18 was most negatively correlated with APP in GSE29378 (Fig. 5C), CASP4 was most positively correlated with PLAU in GSE48350(Fig. 5D), CYCS was most negatively correlated with ADAM10 in GSE36980 (Fig. 5E). Notably, GSE36980 contained only 6 of the top 10 genes associated with AD due to the analyses based on different platforms.
Fig. 5Heatmaps of the correlation between Alzheimer’s disease (AD)-related genes and pyroptosis-related genes of datasets GSE1297 (A), GSE28146 (B), GSE29378 (C), GSE48350 (D), and GSE36980 (E). The abscissa and ordinate represent genes. Different colors represent different correlation coefficients, i.e., red for positive correlation and blue for negative correlation, with the darker color representing the stronger correlation
## Discussion
Studies have hinted that there might be a possible relationship between pyroptosis and AD development [42–45], however, the experimental evidence supporting this correlation is sparse. In our study, we demonstrated that there was a strong relevance between pyroptosis-related and AD-related genes, and the pyroptosis-related genes were differentially expressed in the hippocampus of AD patients and models, which provided strong experimental evidence to support the involvement of pyroptosis in the development of AD.
The hippocampus is located between the thalamus and the medial temporal lobe of the brain. It is a part of the limbic system, mainly responsible for the storage, conversion, and orientation of short-term memory, and also confirmed to play an important role in the development of AD [46]. Thus, we explored four hippocampal sequencing datasets of AD patients and found 20 DEGs between healthy people and AD patients. The previous research proved that most of them were involved in nervous diseases, for example, GAD1 was involved in the neuropathology of schizophrenia [21], RGS4 showed decreased mRNA levels in the prefrontal cortex from AD patient autopsies [47], CD44 was increased in lymphocytes derived from AD patients [39], and the FOS exhibited the intensification of immunoreactivity in AD cases [48]. These DEGs played vital roles in nervous systems, hinting at their possible associations with AD.
Function analysis was performed based on these DEGs to delve into the exact pathway. KEGG models showed most of DEGs enriched in pathways relative to immunity and inflammation, like MAPK, IL-17 signaling pathway, and Th1, Th2, and Th17 cell differentiation, while GO models also showed in inflammation like monocyte chemotaxis, mononuclear cell migration, and so on [49, 50]. A study has shown that the main indications of pyroptosis include the formation of inflammasomes, the activation of caspase and gasdermin, and the release of a large number of pro-inflammatory factors [44]. The typical pathway involved in pyroptosis generally includes the caspase-1 pathway identifying the detrimental effects through the inflammasome, recruiting, activating, and cleaving caspase-1, activating inflammatory factors such as IL-18 and IL-1β, cleaving the N-terminal sequence of gasdermin D (GSDMD) to bind to membranes to create the membrane pores, ultimately leading to pyroptosis [51]. Therefore, we conducted a GSEA analysis of AD-related and pyroptosis-related genes based on the combined dataset. Notably, these two gene sets showed similar trends, indicating that lesions in the hippocampus were closely associated with AD, and the gene set of pyroptosis could participate in this disease. Subsequently, we conducted PPI network construction on a pyroptosis-related gene and correlation analysis between pyroptosis-related genes and AD-related genes. A large number of pyroptosis-related genes and AD-related genes in the datasets showed significant correlations, which confirmed our conjecture that pyroptosis could play a vital role in AD.
It is well-known that inflammasomes play important roles in the development of AD, especially NLRP3 inflammasomes. [ 52, 53]. And activation of the NLRP3 inflammasome could cause caspase-1-mediated production of interleukin (IL)-1β and IL-18 in microglia [54]. A study reported that fatal epilepsy in IL18 KO/APP/PS1 mice was completely reversed by the anticonvulsant levetiracetam, while the IL18-deficient AD mice with chemically induced seizures exhibited lower thresholds and increased gene expression associated with increased neuronal activity [55], which implied that IL18 might be involved in the development of AD. Our PCR analysis showed that the expressions of IL18 were increased in the AD model which confirmed this point. Meanwhile, the level of Bax was also raised. The research found that the localization of bax in senile plaques in the hippocampi of AD patients was correlated with the localization of the β-amyloid protein in the adjacent sections of the same brain, while bax was generally strongly stained in tau-positive tangles in the AD hippocampi, suggesting its vital role in tangle formation [56, 57]. Furthermore, the levels of bax were decreased in the dentate granule cells of the AD hippocampi, which was probably related to the survival of the neurons in AD [58]. To date, rare studies revealed the functions of CYCS in AD. One paper mentioned its possible value in AD diagnosis, however, the exploration of the mechanism is scarce. It has been reported that CYCS plays a role in apoptosis, while the inhibition of anti-apoptotic members of the BCL-2 family or the activation of pro-apoptotic members could lead to changes in the permeability of the mitochondrial membrane, thereby reducing the release of CYCS into the cytoplasm [59–61]. Our work found that the mRNA levels of CYCS decreased in the HT22 cell model of AD, which supported these points in depth.
Taken together, these results demonstrated that pyroptosis played an important role in AD. Further verifications are needed for the DEGs and the molecular mechanisms with the metabolic pathways involved in the development of AD revealed in this study.
## Conclusion
The pyroptosis-related genes BAX, IL18, and CYCS were significantly different between AD patients and normal controls. This proves that the mechanism of pyroptosis is very important for AD, and these significantly differentially expressed genes can be potential targets for the diagnosis and treatment of AD.
## Electronic supplementary material
Below is the link to the electronic supplementary material.
Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Supplementary Material 4
Supplementary Material 5 Supplementary Material 6 Supplementary Material 7 Table S8. Primers and their sequences used in the quantitative real-time PCR of the pyroptosis-related genes
Supplementary Material 9 Supplementary Material 10
## References
1. Hodson R. **Alzheimer’s disease**. *Nature* (2018.0) **559** 1. DOI: 10.1038/d41586-018-05717-6
2. Brookmeyer R, Abdalla N, Kawas C, Corrada M. **Forecasting the prevalence of preclinical and clinical Alzheimer’s disease in the United States**. *Alzheimer’s Dement J Alzheimer’s Assoc* (2018.0) **14** 121-9. DOI: 10.1016/j.jalz.2017.10.009
3. Dansson H, Stempfle L, Egilsdóttir H, Schliep A, Portelius E, Blennow K, Zetterberg H, Johansson F. **Predicting progression and cognitive decline in amyloid-positive patients with Alzheimer’s disease**. *Alzheimers Res Ther* (2021.0) **13** 151. DOI: 10.1186/s13195-021-00886-5
4. Moloney C, Lowe V, Murray M. **Visualization of neurofibrillary tangle maturity in Alzheimer’s disease: a clinicopathologic perspective for biomarker research**. *Alzheimer’s Dement J Alzheimer’s Assoc* (2021.0) **17** 1554-74. DOI: 10.1002/alz.12321
5. Han C, Yang Y, Guan Q, Zhang X, Shen H, Sheng Y, Wang J, Zhou X, Li W, Guo L, Jiao Q. **New mechanism of nerve injury in Alzheimer’s disease: β-amyloid-induced neuronal pyroptosis**. *J Cell Mol Med* (2020.0) **24** 8078-90. DOI: 10.1111/jcmm.15439
6. Man S, Karki R, Kanneganti T. **Molecular mechanisms and functions of pyroptosis, inflammatory caspases and inflammasomes in infectious diseases**. *Immunol Rev* (2017.0) **277** 61-75. DOI: 10.1111/imr.12534
7. Frank D, Vince J. **Pyroptosis versus necroptosis: similarities, differences, and crosstalk**. *Cell Death Differ* (2019.0) **26** 99-114. DOI: 10.1038/s41418-018-0212-6
8. Liu X, Zhang Z, Ruan J, Pan Y, Magupalli V, Wu H, Lieberman J. **Inflammasome-activated gasdermin D causes pyroptosis by forming membrane pores**. *Nature* (2016.0) **535** 153-8. DOI: 10.1038/nature18629
9. Zhaolin Z, Guohua L, Shiyuan W, Zuo W. **Role of pyroptosis in cardiovascular disease**. *Cell Prolif* (2019.0) **52** e12563. DOI: 10.1111/cpr.12563
10. Wang Q, Wu J, Zeng Y, Chen K, Wang C, Yang S, Sun N, Chen H, Duan K, Zeng G. **Pyroptosis: a pro-inflammatory type of cell death in cardiovascular disease**. *Clin Chim Acta* (2020.0) **510** 62-72. DOI: 10.1016/j.cca.2020.06.044
11. Chen Q, Zheng J, Wang D, Liu Q, Kang L, Gao X, Lin Z. **Nitrosonisoldipine is a selective inhibitor of inflammatory caspases and protects against pyroptosis and related septic shock**. *Eur J Immunol* (2021.0) **51** 1234-45. DOI: 10.1002/eji.202048937
12. Tan M, Tan L, Jiang T, Zhu X, Wang H, Jia C, Yu J. **Amyloid-β induces NLRP1-dependent neuronal pyroptosis in models of Alzheimer’s disease**. *Cell Death Dis* (2014.0) **5** e1382. DOI: 10.1038/cddis.2014.348
13. Cheng L, Zhang W. **DJ-1 affects oxidative stress and pyroptosis in hippocampal neurons of Alzheimer’s disease mouse model by regulating the Nrf2 pathway**. *Experimental and therapeutic medicine* (2021.0) **21** 557. DOI: 10.3892/etm.2021.9989
14. Li Q, Wang Q, Guan H, Zhou Y, Liu L. **Schisandrin inhibits NLRP1 inflammasome-mediated neuronal pyroptosis in mouse models of Alzheimer’s Disease**. *Neuropsychiatr Dis Treat* (2021.0) **17** 261-8. DOI: 10.2147/NDT.S279147
15. Taminau J, Meganck S, Lazar C, Steenhoff D, Coletta A, Molter C, Duque R, de Schaetzen V, Weiss Solís D, Bersini H, Nowé A. **Unlocking the potential of publicly available microarray data using inSilicoDb and inSilicoMerging R/Bioconductor packages**. *BMC Bioinformatics* (2012.0) **13** 335. DOI: 10.1186/1471-2105-13-335
16. Johnson W, Li C, Rabinovic A. **Adjusting batch effects in microarray expression data using empirical Bayes methods**. *Biostatistics (Oxford England)* (2007.0) **8** 118-27. DOI: 10.1093/biostatistics/kxj037
17. Ritchie M, Phipson B, Wu D, Hu Y, Law C, Shi W, Smyth G. **Limma powers differential expression analyses for RNA-sequencing and microarray studies**. *Nucleic Acids Res* (2015.0) **43** e47. DOI: 10.1093/nar/gkv007
18. **KEGG for taxonomy-based analysis of pathways and genomes**. *Nucleic Acids Res* (2023.0) **51** D587-92. DOI: 10.1093/nar/gkac963
19. Subramanian A, Tamayo P, Mootha V, Mukherjee S, Ebert B, Gillette M, Paulovich A, Pomeroy S, Golub T, Lander E, Mesirov J. **Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles**. *Proc Natl Acad Sci USA* (2005.0) **102** 15545-50. DOI: 10.1073/pnas.0506580102
20. Suárez-Fariñas M, Lowes M, Zaba L, Krueger J. **Evaluation of the psoriasis transcriptome across different studies by gene set enrichment analysis (GSEA)**. *PLoS ONE* (2010.0) **5** e10247. DOI: 10.1371/journal.pone.0010247
21. Szklarczyk D, Morris J, Cook H, Kuhn M, Wyder S, Simonovic M, Santos A, Doncheva N, Roth A, Bork P. **The STRING database in 2017: quality-controlled protein-protein association networks, made broadly accessible**. *Nucleic Acids Res* (2017.0) **45** D362-8. DOI: 10.1093/nar/gkw937
22. Li S, Xiao J, Huang C, Sun J. **Identification and validation of oxidative stress and immune-related hub genes in Alzheimer’s disease through bioinformatics analysis**. *Sci Rep* (2023.0) **13** 657. DOI: 10.1038/s41598-023-27977-7
23. Muma N, Mariyappa R, Williams K, Lee J. **Differences in regional and subcellular localization of G(q/11) and RGS4 protein levels in Alzheimer’s disease: correlation with muscarinic M1 receptor binding parameters**. *Synapse (New York NY)* (2003.0) **47** 58-65. DOI: 10.1002/syn.10153
24. Kim K, Kim Y, Choi K, Lee M, Lee S, Im W, Shin J, Kim J, Hong Y, Kim M. **Downregulated miR-18b-5p triggers apoptosis by inhibition of calcium signaling and neuronal cell differentiation in transgenic SOD1 (G93A) mice and SOD1 (G17S and G86S) ALS patients**. *Translational neurodegeneration* (2020.0) **9** 23. DOI: 10.1186/s40035-020-00203-4
25. Vakilian H, Mirzaei M, Sharifi Tabar M, Pooyan P, Habibi Rezaee L, Parker L, Haynes P, Gourabi H, Baharvand H, Salekdeh G. **DDX3Y, a male-specific region of Y chromosome gene, May modulate neuronal differentiation**. *J Proteome Res* (2015.0) **14** 3474-83. DOI: 10.1021/acs.jproteome.5b00512
26. 26.George C, Gontier G, Lacube P, François J, Holzenberger M, Aïd S. The Alzheimer’s disease transcriptome mimics the neuroprotective signature of IGF-1 receptor-deficient neurons. Brain: a journal of neurology 2017, 140:2012–2027.
27. Huang C, Luo J, Wen X, Li K. **Linking diabetes Mellitus with Alzheimer’s Disease: Bioinformatics Analysis for the potential pathways and characteristic genes**. *Biochem Genet* (2022.0) **60** 1049-75. DOI: 10.1007/s10528-021-10154-8
28. Zhang X, Huang T, Yancey J, Luo H, Zhang Y. **Role of Rab GTPases in Alzheimer’s Disease**. *ACS Chem Neurosci* (2019.0) **10** 828-38. DOI: 10.1021/acschemneuro.8b00387
29. Wahrle S, Jiang H, Parsadanian M, Kim J, Li A, Knoten A, Jain S, Hirsch-Reinshagen V, Wellington C, Bales K. **Overexpression of ABCA1 reduces amyloid deposition in the PDAPP mouse model of Alzheimer disease**. *J Clin Investig* (2008.0) **118** 671-82. PMID: 18202749
30. Leandro G, Evangelista A, Lobo R, Xavier D, Moriguti J, Sakamoto-Hojo E. **Changes in expression profiles revealed by Transcriptomic Analysis in Peripheral Blood mononuclear cells of Alzheimer’s Disease Patients**. *J Alzheimer’s disease: JAD* (2018.0) **66** 1483-95. DOI: 10.3233/JAD-170205
31. Piras I, Krate J, Delvaux E, Nolz J, De Both M, Mastroeni D, Serrano G, Sue L, Beach T, Coleman P, Huentelman M. **Association of AEBP1 and NRN1 RNA expression with Alzheimer’s disease and neurofibrillary tangle density in middle temporal gyrus**. *Brain Res* (2019.0) **1719** 217-24. DOI: 10.1016/j.brainres.2019.06.004
32. Hartlage-Rübsamen M, Waniek A, Meissner J, Morawski M, Schilling S, Jäger C, Kleinschmidt M, Cynis H, Kehlen A, Arendt T. **Isoglutaminyl cyclase contributes to CCL2-driven neuroinflammation in Alzheimer’s disease**. *Acta Neuropathol* (2015.0) **129** 565-83. DOI: 10.1007/s00401-015-1395-2
33. 33.McArthur S, Cristante E, Paterno M, Christian H, Roncaroli F, Gillies G, Solito E. Annexin A1: a central player in the anti-inflammatory and neuroprotective role of microglia. Journal of immunology (Baltimore, Md: 1950) 2010, 185:6317–6328.
34. Montero-Calle A, San Segundo-Acosta P, Garranzo-Asensio M, Rábano A, Barderas R. **The Molecular Misreading of APP and UBB induces a Humoral Immune response in Alzheimer’s Disease Patients with Diagnostic ability**. *Mol Neurobiol* (2020.0) **57** 1009-20. DOI: 10.1007/s12035-019-01809-0
35. Recabarren D, Alarcón M. **Gene networks in neurodegenerative disorders**. *Life Sci* (2017.0) **183** 83-97. DOI: 10.1016/j.lfs.2017.06.009
36. Norton E, Da Mesquita S, Guerrero-Cazares H. **SERPINA3 in glioblastoma and Alzheimer’s disease**. *Aging* (2021.0) **13** 21812-3. DOI: 10.18632/aging.203603
37. Mirza Z, Rajeh N. **Identification of electrophysiological changes in Alzheimer’s Disease: a microarray based Transcriptomics and Molecular Pathway Analysis Study**. *CNS Neurol Disord Drug Target* (2017.0) **16** 1027-38
38. Ghani M, Sato C, Lee J, Reitz C, Moreno D, Mayeux R, St George-Hyslop P, Rogaeva E. **Evidence of recessive Alzheimer disease loci in a caribbean hispanic data set: genome-wide survey of runs of homozygosity**. *JAMA Neurol* (2013.0) **70** 1261-7. PMID: 23978990
39. Uberti D, Cenini G, Bonini S, Barcikowska M, Styczynska M, Szybinska A, Memo M. **Increased CD44 gene expression in lymphocytes derived from Alzheimer disease patients**. *Neuro-degener Dis* (2010.0) **7** 143-7. DOI: 10.1159/000289225
40. Zhang X, Zou M, Wu Y, Jiang D, Wu T, Zhao Y, Wu D, Cui J, Li G. **HLA-DQA1/DRB1Regulation of the late onset alzheimer’s Disease Associated expression**. *Am J Alzheimer’s Dis Other dement* (2022.0) **37** 15333175221085066
41. Choi H, Kim C, Song H, Cha M, Cho H, Son S, Kim H, Mook-Jung I. **Amyloid β-induced elevation of O-GlcNAcylated c-Fos promotes neuronal cell death**. *Aging Cell* (2019.0) **18** e12872. DOI: 10.1111/acel.12872
42. Olsen I, Singhrao S. **Inflammasome involvement in Alzheimer’s Disease**. *J Alzheimer’s disease: JAD* (2016.0) **54** 45-53. DOI: 10.3233/JAD-160197
43. Rui W, Xiao H, Fan Y, Ma Z, Xiao M, Li S, Shi J. **Systemic inflammasome activation and pyroptosis associate with the progression of amnestic mild cognitive impairment and Alzheimer’s disease**. *J Neuroinflammation* (2021.0) **18** 280. DOI: 10.1186/s12974-021-02329-2
44. Xue W, Cui D, Qiu Y. **Research Progress of pyroptosis in Alzheimer’s Disease**. *Front Mol Neurosci* (2022.0) **15** 872471. DOI: 10.3389/fnmol.2022.872471
45. Moonen S, Koper MJ, Van Schoor E, Schaeverbeke JM, Vandenberghe R, von Arnim CAF, Tousseyn T, De Strooper B, Thal DR. **Pyroptosis in Alzheimer’s disease: cell type-specific activation in microglia, astrocytes and neurons**. *Acta Neuropathol* (2023.0) **145** 175-95. DOI: 10.1007/s00401-022-02528-y
46. van der Flier W, Scheltens P. **Alzheimer disease: hippocampal volume loss and Alzheimer disease progression**. *Nat reviews Neurol* (2009.0) **5** 361-2. DOI: 10.1038/nrneurol.2009.94
47. Emilsson L, Saetre P, Jazin E. **Low mRNA levels of RGS4 splice variants in Alzheimer’s disease: association between a rare haplotype and decreased mRNA expression**. *Synapse (New York NY)* (2006.0) **59** 173-6. DOI: 10.1002/syn.20226
48. Anderson A, Cummings B, Cotman C. **Increased immunoreactivity for Jun- and Fos-related proteins in Alzheimer’s disease: association with pathology**. *Exp Neurol* (1994.0) **125** 286-95. DOI: 10.1006/exnr.1994.1031
49. 49.Trzeciak P, Herbet M, Dudka J. Common Factors of Alzheimer’s Disease and Rheumatoid Arthritis-Pathomechanism and Treatment. Molecules (Basel, Switzerland) 2021, 26.
50. Chen Z, Zhong C. **Decoding Alzheimer’s disease from perturbed cerebral glucose metabolism: implications for diagnostic and therapeutic strategies**. *Prog Neurobiol* (2013.0) **108** 21-43. DOI: 10.1016/j.pneurobio.2013.06.004
51. Fricker M, Tolkovsky A, Borutaite V, Coleman M, Brown G. **Neuronal cell death**. *Physiol Rev* (2018.0) **98** 813-80. DOI: 10.1152/physrev.00011.2017
52. Pennisi M, Crupi R, Di Paola R, Ontario M, Bella R, Calabrese E, Crea R, Cuzzocrea S, Calabrese V. **Inflammasomes, hormesis, and antioxidants in neuroinflammation: role of NRLP3 in Alzheimer disease**. *J Neurosci Res* (2017.0) **95** 1360-72. DOI: 10.1002/jnr.23986
53. Jung E, Suh K, Han J, Kim H, Kang H, Choi W, Mook-Jung I. **Amyloid-β activates NLRP3 inflammasomes by affecting microglial immunometabolism through the Syk-AMPK pathway**. *Aging Cell* (2022.0) **21** e13623. DOI: 10.1111/acel.13623
54. Feng Y, Tan Z, Wu L, Dong F, Zhang F. **The involvement of NLRP3 inflammasome in the treatment of Alzheimer’s disease**. *Ageing Res Rev* (2020.0) **64** 101192. DOI: 10.1016/j.arr.2020.101192
55. Tzeng T, Hasegawa Y, Iguchi R, Cheung A, Caffrey D, Thatcher E, Mao W, Germain G, Tamburro N, Okabe S. **Inflammasome-derived cytokine IL18 suppresses amyloid-induced seizures in Alzheimer-prone mice**. *Proc Natl Acad Sci USA* (2018.0) **115** 9002-7. DOI: 10.1073/pnas.1801802115
56. Tortosa A, López E, Ferrer I. **Bcl-2 and bax protein expression in Alzheimer’s disease**. *Acta Neuropathol* (1998.0) **95** 407-12. DOI: 10.1007/s004010050817
57. MacGibbon G, Lawlor P, Sirimanne E, Walton M, Connor B, Young D, Williams C, Gluckman P, Faull R, Hughes P, Dragunow M. **Bax expression in mammalian neurons undergoing apoptosis, and in Alzheimer’s disease hippocampus**. *Brain Res* (1997.0) **750** 223-34. DOI: 10.1016/S0006-8993(96)01351-0
58. Deegan S, Saveljeva S, Logue S, Pakos-Zebrucka K, Gupta S, Vandenabeele P, Bertrand M, Samali A. **Deficiency in the mitochondrial apoptotic pathway reveals the toxic potential of autophagy under ER stress conditions**. *Autophagy* (2014.0) **10** 1921-36. DOI: 10.4161/15548627.2014.981790
59. 59.He Y, Zhao T, Chen F, Song C, Zhong C, Luo Z. Bcl-2Functional Analysis of the Promoter Regions of Two Apoptosis-Related Genes (and) and Their Regulation by Zn in Yellow Catfish. International journal of molecular sciences 2021, 22.
60. Qi P, Li J, Gao S, Yuan Y, Sun Y, Liu N, Li Y, Wang G, Chen L, Shi J. **Network Pharmacology-Based and experimental identification of the Effects of Quercetin on Alzheimer’s Disease**. *Front Aging Neurosci* (2020.0) **12** 589588. DOI: 10.3389/fnagi.2020.589588
61. Afzelius B. **Genetical and ultrastructural aspects of the immotile-cilia syndrome**. *Am J Hum Genet* (1981.0) **33** 852-64. PMID: 7034533
|
---
title: Constructing a novel mitochondrial-related gene signature for evaluating the
tumor immune microenvironment and predicting survival in stomach adenocarcinoma
authors:
- Jingjia Chang
- Hao Wu
- Jin Wu
- Ming Liu
- Wentao Zhang
- Yanfen Hu
- Xintong Zhang
- Jing Xu
- Li Li
- Pengfei Yu
- Jianjun Zhu
journal: Journal of Translational Medicine
year: 2023
pmcid: PMC10012538
doi: 10.1186/s12967-023-04033-6
license: CC BY 4.0
---
# Constructing a novel mitochondrial-related gene signature for evaluating the tumor immune microenvironment and predicting survival in stomach adenocarcinoma
## Abstract
### Background
The incidence and mortality of gastric cancer ranks fifth and fourth worldwide among all malignancies, respectively. Accumulating evidences have revealed the close relationship between mitochondrial dysfunction and the initiation and progression of stomach cancer. However, rare prognostic models for mitochondrial-related gene risk have been built up in stomach cancer.
### Methods
In current study, the expression and prognostic value of mitochondrial-related genes in stomach adenocarcinoma (STAD) patients were systematically analyzed to establish a mitochondrial-related risk model based on available TCGA and GEO databases. The tumor microenvironment (TME), immune cell infiltration, tumor mutation burden, and drug sensitivity of gastric adenocarcinoma patients were also investigated using R language, GraphPad Prism 8 and online databases.
### Results
We established a mitochondrial-related risk prognostic model including NOX4, ALDH3A2, FKBP10 and MAOA and validated its predictive power. This risk model indicated that the immune cell infiltration in high-risk group was significantly different from that in the low-risk group. Besides, the risk score was closely related to TME signature genes and immune checkpoint molecules, suggesting that the immunosuppressive tumor microenvironment might lead to poor prognosis in high-risk groups. Moreover, TIDE analysis demonstrated that combined analysis of risk score and immune score, or stromal score, or microsatellite status could more effectively predict the benefit of immunotherapy in STAD patients with different stratifications. Finally, rapamycin, PD-0325901 and dasatinib were found to be more effective for patients in the high-risk group, whereas AZD7762, CEP-701 and methotrexate were predicted to be more effective for patients in the low-risk group.
### Conclusions
Our results suggest that the mitochondrial-related risk model could be a reliable prognostic biomarker for personalized treatment of STAD patients.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12967-023-04033-6.
## Introduction
The incidence (accounting for $5.6\%$ of all cancer cases) and mortality (accounting for $7.7\%$ of all cancer deaths) of gastric cancer (GC) ranks fifth and fourth worldwide among all malignancies, which critically threatens human health. It was estimated that over one million new cases of GC were reported worldwide in 2020 [1]. Gastric adenocarcinomas derived from gastric glandular epithelial cells accounts for more than $90\%$ of all GCs [2]. GC is a multifactorial disease, which is contributed by both environmental and genetic factors [3], such as smoking, family history, Epstein–*Barr virus* (EBV) infection, alcohol consumption, and diet [4]. Most GCs are diagnosed at late stage of progression due to limited premalignant indications and symptoms [2]. Nowadays, the application of the endoscopic examination largely improved the survival rate of GC patients, and a $30\%$ reduction in GC mortality using endoscopic screening [3]. However, stomach cancer is still one of the most lethal malignant tumors, with a 5 year survival rate of around $20\%$ [5]. Therefore, it is necessary to identify more reliable biomarkers for predicting the prognosis and exploring more potential therapeutic targets in GC.
Accumulating evidences indicated that mitochondria plays essential roles in regulation of cell growth, cell death, and cell metabolism during the whole process of tumor progression [6]. Mitochondria are involved in bioenergetics metabolism, such as ATP production, reactive oxygen species (ROS) production, apoptosis, and calcium homeostasis [7]. Moreover, mitochondrial dysfunction may contribute to the chemoresistance [8]. Therefore, mitochondrial-targeting therapies may be applied for the treatment of GC, including ROS production and elimination, mitochondrial fission and fusion, ATP production, and apoptosis [6, 9, 10]. For instance, nanohybrid-induced oxidative stress triggered mitochondria-mediated autophagy, which inhibited cell growth in cancer cell [11].
Considering that mitochondrial dysfunction was a risk factor for the tumorigenesis of GC, identifying effective mitochondrial-related biomarkers for the prognosis of GC patients should be an encouraging direction of research. Several studies have constructed GC prognosis-related models to predict patient survival [12–14]. However, rare studies have been applied for the establishment of prognostic models for GC associated with mitochondria.
Tumor microenvironment (TME) was mainly composed of the stromal cells, immune cells and cytokines [15]. The components of TME affected the immune cell evasion or inhibition, and drug resistance in malignancies. For example, immune progenitors in the complex microenvironment of the TME were more likely to differentiate into M2 macrophages and Treg cells, but not to play their tumor-inhibiting functions as fully mature immune cells [16]. In addition, the response to immune checkpoint blockade (ICB) was closely related to the constitution of the TME. ICB revived an effective anti-tumor immune response [17]. It was reported that PD-1/PD-L1 inhibitors immunotherapy has an impact on the therapy of patients with advanced gastrointestinal malignancy. Moreover, a study reported that tumors with a higher tumor mutation burden (TMB), had a better immunotherapy response, especially with PD-1/PD-L1 blockade [18]. Thus, to figure out the correlation between the risk score and TME, we explored the TME signatures, the immune cell infiltration in TME, the expression level of immune checkpoints, and the response to immunotherapy.
In summary, a novel mitochondrial-related risk model was constructed using NOX4, FKBP10, ALDH3A2, and MAOA gene set, which could effectively predict the prognosis and immunotherapy responsiveness for patients with STAD. In addition, we estimated the drug sensitivity of STAD patients to 138 drugs, including chemotherapy drugs, immunotherapy drugs, and targeted drugs, et al., and found that patients in high-risk group was more sensitive to rapamycin, PD-0325901 and dasatinib, whereas patients in low-risk group was more sensitive to AZD7762, CEP-701 and methotrexate. Taken together, our mitochondrial-related risk model could be a reliable prognostic biomarker for personalized treatment of STAD patients.
## Data collection
RNA-seq data and microsatellite status information for 407 STAD samples were downloaded from the TCGA database (https://www.cancer.gov/tcga). The clinical information was extracted from the UCSC Xena (http://xena.ucsc.edu) [19]. 61 samples were excluded due to incomplete clinical information or survival less than 30 days. In total, 346 samples, comprising 317 tumor samples and 29 healthy samples, were analyzed in the present study. Two validation cohorts, GSE66229 and GSE15459, were applied in the present study (https://www.ncbi.nlm.nih.gov/geo/). In GSE15459, 10 samples were excluded due to survival less than 30 days. Altogether, 300 and 182 samples were analyzed in GSE66229 and GSE15459, respectively. The list of mitochondrial-related genes was collected from MitoCarta 3.0 database (https://www.broadinstitute.org/mitocarta/mitocarta30-inventory-mammalian-mitochondrial-proteins-and-pathways) [20] and the Gene set enrichment analyses (GSEA, http://www.gsea-msigdb.org/gsea/index.jsp) [21, 22] (Additional file 10: Table S1).
## Identification of differentially expressed genes (DEGs)
The “limma” package of R (version 3.5.1) was applied to produce DEGs between normal and tumor samples, or between high-risk and low-risk groups from the training set. |Log [2] fold change|> 2 and adjusted $P \leq 0.01$ were the criteria for defining DEGs. “ GdcVolcanoPlot” packages in R were employed to generate volcano map to visualize the DEGs, and a Venn plot was exploited to display the common DEGs in both DEGs groups and mitochondrial-related genes.
## Construction and validation of prognostic mitochondrial-related risk score signature
The mitochondrial-related genes were screened by univariate Cox regression, Lasso regression analysis and multivariable Cox regression analysis to construct a novel prognostic gene signature. Each sample’s risk score was calculated using the following formula:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{Risk score }} = \, \Sigma {\text{expgenei}} * {\beta i}$$\end{document}Risk score=Σexpgenei∗βiwhere expgene, i, and βi represent the expression level of gene, the number of signature genes, and the coefficient index, respectively. In all participated cohorts, the samples were divided into low-risk and high-risk groups based on the risk score (median cut-off value). To analyze the survival conditions for the prognosis signature, the optimized cutoff and the Kaplan–Meier (K–M) survival curve were conducted by R package “survival” and “survminer”. The predictive performance was presented by ROC curve, risk plot and concordance index (C-index). Detailed information for prognostic genes was obtained from The Human Protein Atlas (HPA, https://www.proteinatlas.org/) and National Center for Biotechnology Information (NCBI, https://www.ncbi.nlm.nih.gov/).
## Construction and valuation of nomogram
Risk score and clinical factors including age, gender, T stage, N stage, M stage, tumor stage, family history, H pylori infection, grade, reflux history, and disease types were analyzed using univariate Cox regression analysis to screen the factors significantly related to survival ($P \leq 0.1$). Then, multivariate COX regression analysis was applied to identify the candidate predictors significantly related to survival ($P \leq 0.05$). Based on this, nomograms were constructed using these predictors, and scores in nomogram model were assigned for these variables. By adding the scores of the predictors enrolled in nomogram model, the total score of each patient was obtained. Finally, the patient's survival outcome in 1, 3 and 5 years can be calculated using the total score and the probability of survival outcome. ROC curve, calibration curves and decision curve analysis (DCA) were applied to estimate the discrimination and accuracy of the nomogram model.
## Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses
In the present study, R “clusterProfiler”, “org.Hs.eg.db”, “enrichplot” and “ggplot2” package (R version: 3.5.1) were employed to analyze the function of mitochondrial-related DEGs, or the DEGs between high-risk and low-risk groups. Furthermore, adjusted $P \leq 0.05$ was used to filter the functional candidates.
## Gene set enrichment analyses (GSEA)
Curated sets v7.4 collections were obtained from the Molecular Signatures Database as the target sets with which GSEA was performed by using GSEA 4.2.1 software. The total transcriptome of tumor samples was used for the GSEA, and only gene sets with $P \leq 0.001$ and FDR, q < 0.001 were regarded to be statistically significant.
## Tumor microenvironment
Stromal scores were calculated using the ESTIMATE algorithm by R (version 3.5.1) package “estimate”. The list of TME-related biomarkers was extracted from the Gene set enrichment analyses (GSEA, http://www.gsea-msigdb.org/gsea/index.jsp) [21, 22] (Additional file 10: Table S2–S5).
## Calculation of relative abundance of 22 immune cell subtypes
The abundance of 22 tumor-infiltrating immune cells (TIICs) in STAD samples was calculated using the CIBERSORT algorithm by R package. CIBERSORT is a deconvolution algorithm that can infer 22 kinds of TIICs and harnesses the ability to predict the relative abundance of each immune cell population by calculating the expression of specific marker [23]. The relative abundance of the TIICs between high-risk and low-risk groups was compared using the Wilcox text. The list of immune check points was referenced from a published study [24]. Immune score and tumor purity were also calculated using the ESTIMATE algorithm by R (version 3.5.1) package “estimate”. The list of immune cell signatures was downloaded from TISIDB (http://cis.hku.hk/TISIDB/download.php) [25].
## Prediction of therapeutic sensitivity in patients with different risk scores
The capability of risk score in predicting the response to immunotherapy or 138 drugs for chemotherapies/targeted therapies was explored in the present study. The $50\%$ inhibiting concentration (IC50) values of the 138 drugs were calculated using the “pRRophetic” package of R (version 3.5.1) and the value was normally transformed. The detailed information of 138 drugs was acquired from Genomics of Drug Sensitivity in Cancer (GDSC, https://www.cancerrxgene.org/). The potential response to immunotherapy was inferred by the tumor immune dysfunction and exclusion (TIDE, http://tide.dfci.harvard.edu) score.
## Mutation analysis
The somatic mutation data were downloaded from cBioportal database (https://www.cbioportal.org/) [26, 27]. The R (version 3.5.1) package “maftools” was then used to draw a waterfall plot to illustrate the mutation landscape in STAD patients with the high- and low-risk group and calculate the TMB score for each sample.
## Cell culture and patient sample collection
The normal gastric epithelial cell line GES-1 and human gastric cancer cell lines SGC-7901 and HGC-27 were purchased from the American Type Culture Collection (ATCC, Manassas, VA, USA). Cells were cultured in DMEM medium (Gibco, Thermo Fisher Scientific, Inc., Waltham, MA, USA) supplemented with $10\%$ fetal bovine serum (FBS, Hyclone). Cells were routinely cultured in a humidified atmosphere containg $5\%$ CO2 at 37 ℃. A total of 10 fresh tumor and paired adjacent normal tissues from patients with STAD were collected in the First Affiliated Hospital of Shanxi Medical University (Taiyuan, Shanxi, China). All patients provided written informed consent, and this study was approved by the ethics committee of Shanxi Medical University.
## RNA extraction and qRT-PCR assays
RNAs in tissues and cell lines of STAD were extracted with a RNAiso Plus (Takara,Tokyo, Japan) and were reversely transcribed into cDNA usin PrimeScriptTM RT Master Mix (Takara). Quantitative real-time PCR (qRT-PCR) was performed by TB Green ®Premix Ex TaqTM II (Takara). β-actin was used as reference genes. The primers were listed in Additional file 10: Table S6.
## Statistical analysis
R (version 3.5.1) and the GraphPad Prism 8 software were applied for statistical analysis. A Student’s t-test was used to analyze the expression and the distribution of risk score, stromal score, immune score, tumor purity and TMB in different groups. Chi-square test is applied to evaluate the difference in immunotherapy response, status of top 5 mutant genes and clinical factors in different groups. The correlation was evaluated using the Spearman method. C-index was used to estimate the predictive power of age and risk score to OS. $P \leq 0.05$ was defined as statistically significant.
## Identification of DEGs related to mitochondrion and functional enrichment analysis in STAD
*The* general workflow of our current study was presented in Fig. 1. As shown in Additional file 10: Table S7, 2381 DEGs, including 2145 protein-coding genes, were screened and visualized via volcano maps between normal and tumor groups (Fig. 2A, Additional file 10: Table S8). Next, combined analysis for selected mitochondrial-related genes from the GSEA and 2145 DEGs from our study were performed to filter out 183 candidate mitochondrial-related DEGs in STAD (Fig. 2B, Additional file 10: Table S9).Fig. 1Workflow diagram. The flowchart graph of this studyFig. 2Identification of DEGs related to mitochondrion and construction of prognostic risk model using TCGA-STAD cohort. A Volcano plot of 2381 DEGs in STAD tumor and normal groups. B Venn diagram showed that the overlap of 2381 DEGs and 2030 mitochondrial genes led to 183 hub genes being identified. C Univariate Cox regression analysis revealed 19 genes were associated with prognosis of patients with STAD. D LASSO regression of the 19 OS-related genes. Cross-validation in the LASSO regression model to select the tuning parameter. The abscissa shows the log (λ) value, and the ordinate shows partial likelihood deviance. The red dots in the figure show partial likelihood deviations ± standard error for diverse tuning parameters. E Multivariable Cox regression analysis revealed 4 genes were associated with prognosis of patients with STAD. F Gene expressions of the 4 prognosis-related genes in TCGA-STAD. P values were showed as: ***$P \leq 0.001$ GO enrichment analysis were then carried out to uncover important roles of mitochondrial-related DEGs in STAD. These DEGs potentially participated in small molecule catabolic process, regulation of mitochondrial organization, et al. Regarding the cellular component, they were mainly related to mitochondrial matrix, mitochondrial outer membrane, et al. In terms of molecular function, these DEGs were involved in tubulin binding, and ubiquitin-like protein ligase binding, et al. ( Additional file 1: Fig. S1A, Additional file 10: Table S10). Moreover, KEGG pathway analysis was also applied to demonstrate important pathways being involved by these DEGs, such as lipid and atherosclerosis, Hepatitis B infection, Diabetic cardiomyopathy, and apoptosis, et al. ( Additional file 1: Fig. S1B, Additional file 10: Table S11).
## Construction and validation of a mitochondrial-related risk signature
Based on above 183 mitochondrial-related DEGs, 19 genes were further selected as potential risk factors for the prognosis of patients with STAD through univariate Cox regression analysis ($P \leq 0.05$, Fig. 2C). *The* gene number was further narrowed down to 9 according to LASSO regression analysis and to 4 by multivariable Cox regression analysis (Fig. 2D, E). Finally, 4 mitochondrial-related DEGs, including NOX4, FKBP10, ALDH3A2 and MAOA, were utilized to establish a prognostic model for patients with STAD (Table 1).Table 1The information of 4 prognosis-related genesGene symbolGene IDFull nameLocationFunction of the encoded proteinNOX450507NADPH oxidase 4MembraneThe ROS generated by NOX4 have been implicated in numerous biological functions including signal transduction, cell differentiation and tumor cell growthFKBP1060681FKBP prolyl isomerase 10MitochondriaFKBP10 localizes to the endoplasmic reticulum and acts as a molecular chaperone. Alternatively spliced variants encoding different isoforms have been reported, but their biological validity has not been determinedALDH3A2224Aldehyde dehydrogenase 3 family member A2Endoplasmic reticulumAldehyde dehydrogenase isozymes are thought to play a major role in the detoxification of aldehydes generated by alcohol metabolism and lipid peroxidationMAOA4128Monoamine oxidase AMitochondriaThis gene is one of two neighboring gene family members that encode mitochondrial enzymes which catalyze the oxidative deamination of amines, such as dopamine, norepinephrine, and serotonin As shown in Fig. 2F, higher expressions of NOX4 and FKBP10, and lower expressions of ALDH3A2 and MAOA were observed in tumor samples compared with the normal tissues, respectively. Moreover, the immunohistochemistry results from HPA database showed that FKBP10 was upregulated in gastric cancer tissues, while ALDH3A2 and MAOA were downregulated in gastric cancer tissues, when compared with corresponding non-cancerous tissues (Additional file 1: Fig. S1C). Further K-M analysis demonstrated that patients with higher expression of NOX4 ($$P \leq 0.030$$), FKBP10 ($$P \leq 0.040$$), and MAOA ($$P \leq 0.018$$) had a shorter OS than those with lower expression, respectively (Additional file 1: Fig. S1D). However, the patients with higher level of ALDH3A2 had a better OS than those with lower expression, even though it is a bit beyond statistically significant difference ($$P \leq 0.052$$, Additional file 1: Fig. S1D).
Then, the risk score for each patient with STAD in both training and validation cohorts was computed based on the following formula:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{Risk score}}\, = \,0.{157}*{\text{MAOA}} - 0.{198}*{\text{ALDH3A2}}\, + \,0.{133}*{\text{FKBP1}}0\, + \,0.{146}*{\text{NOX4}}{.}$$\end{document}Risk score=0.157∗MAOA-0.198∗ALDH3A2+0.133∗FKBP10+0.146∗NOX4.Patients were divided into high-risk and low-risk subgroups based on the median risk score. K–M curves showed that patients in high-risk group had worse OS ($$P \leq 0.0009$$, Fig. 3A). To assess the accuracy of prognostic risk models in predicting 1-, 3-, and 5-year OS, ROC curves were plotted with AUC values of 0.635, 0.640, and 0.793, respectively (Fig. 3B). The relationship between the risk score and the survival time, survival status, and risk ranking, and a heatmap of the expressions of the 4 genes were shown in Fig. 3C. Taken together, these results demonstrated the robustness of our risk model in predicting the prognosis of patients with STAD.Fig. 3Assessing the performance of the prognostic risk model in the training and validation cohort. A, D Kaplan–Meier curves of the OS of patients in the high- and low-risk groups in the TCGA-STAD training cohort (A), and GSE66229 cohort (D). B, E ROC curves for predicting 1-, 3-, and 5-year OS in the TCGA-STAD training cohort (B), and GSE66229 cohort (E). C, F Distribution of risk score, survival status (red dots indicate dead, blue dots indicate alive) and the gene expression of 4 model genes in the TCGA-STAD training cohort (C), and GSE66229 cohort (F) The robustness of the prognostic risk model was further validated in GSE66229 and GSE15459 datasets. In line with that of the training cohort (TCGA-STAD), patients in high-risk group also had worse prognosis in the validation cohorts (Fig. 3D, Additional file 2: Fig. S2A). In GSE66229 dataset, the AUC values of the ROC curve for 1-year, 3-year and 5-year survival were 0.620, 0.625, and 0.601, respectively (Fig. 3E). Corresponding AUC values of 0.620, 0.647, and 0.657 were observed in GSE15459 dataset (Additional file 2: Fig. S2B). The higher the risk score, the worse the survival (Fig. 3F, Additional file 2: Fig. S2C). The heatmaps of the expressions of the 4 genes were shown in Fig. 3F, Additional file 2: Fig. S2C. Consistent with the TCGA-STAD training cohort, the expressions of the NOX4 and FKBP10 were significantly up-regulated, while the expressions of the MAOA and ALDH3A2 were significantly down-regulated in STAD in GSE66229 validation cohort (Additional file 2: Fig. S2D). Next, we systematically analyzed the relationship between the risk score and clinical characteristics in STAD. The risk scores were remarkably higher in patients with H pylori infection, and cystic, mucinous and serous neoplasms (Additional file 3: Fig. S3). Nevertheless, no differences were observed in the mean of risk score among the groups of age, gender, T stage, N stage, M stage, tumor stage, family history of GC, grade, and reflux history (Additional file 3: Fig. S3). The clinical characteristics of the low-risk and high-risk subgroups were then compared, and the difference of Gender ($$P \leq 0.006$$), N stage ($$P \leq 0.043$$), H pylori infection ($$P \leq 0.035$$), disease type ($$P \leq 0.046$$) and survival status ($$P \leq 0.008$$) among the two risk subgroups reached statistical significance (Table 2).Table 2Clinical characteristics between low- and high-risk groupsVariablesLow riskNo. (%) High riskNo. (%) P valueAge (years)0.466 < 6773 ($45.91\%$)79 ($50.00\%$) ≥ 6786 ($54.09\%$)79 ($50.00\%$)Gender0.006 Male90 ($56.60\%$)113 ($71.52\%$) Female69 ($43.40\%$)45 ($28.48\%$)T stage0.385 T$\frac{1}{244}$ ($27.67\%$)37 ($23.42\%$) T$\frac{3}{4115}$ ($72.33\%$)121 ($76.58\%$)N stage0.043 N$\frac{0}{197}$ ($61.01\%$)84 ($53.16\%$) N$\frac{2}{362}$ ($38.99\%$)69 ($43.67\%$) NX0 ($0.00\%$)5 ($3.16\%$)M stage0.624 M0145 ($91.19\%$)141 ($89.24\%$) M18 ($5.03\%$)12 ($7.59\%$) MX6 ($3.77\%$)5 ($3.16\%$)Tumor stage0.958 I/II75 ($47.17\%$)75 ($47.47\%$) III/IV84 ($52.83\%$)83 ($52.53\%$)Family history of stomach cancer0.205 Yes5 ($3.14\%$)10 ($6.33\%$) No122 ($76.73\%$)121 ($76.58\%$) N/A32 ($20.13\%$)27 ($17.09\%$)H. pylori infection0.035 Yes6 ($3.77\%$)11 ($6.96\%$) No80 ($50.31\%$)49 ($31.01\%$) N/A73 ($45.91\%$)98 ($62.03\%$)Grade0.152 G13 ($1.89\%$)5 ($3.16\%$) G262 ($38.99\%$)46 ($29.11\%$) G392 ($57.86\%$)101 ($63.92\%$) GX2 ($1.26\%$)6 ($3.80\%$)Reflux history0.988 Yes19 ($11.95\%$)16 ($10.13\%$) No86 ($54.09\%$)72 ($45.57\%$) N/A54 ($33.96\%$)70 ($44.30\%$) Disease type0.046Adenomas and adenocarcinomas150 ($94.34\%$)139 ($87.97\%$) Cystic, mucinous and serous neoplasms9 ($5.66\%$)19 ($12.03\%$) Survival status0.008 Alive105 ($66.04\%$)81 ($51.27\%$) Dead54 ($33.96\%$)77 ($48.73\%$)Bold indicates P value ≤ 0.05 was considered statistically significant
## Construction of nomogram
The nomogram integrated the risk score and all important clinical features, which can be used to quantitatively predict the prognosis of patients and provide a reference for clinical decision making. In our study, risk score ($$P \leq 0.0005$$) and age ($$P \leq 0.020$$) were finally identified as prognostic indicators by using univariate and multivariate Cox regression analysis to construct nomogram (Table 3). As a result, a predictive nomogram integrating risk score (a score of 100) and age (a score of 67.5) for prognosis was constructed (Additional file 4: Fig. S4A). ROC curves showed that the AUC values of the nomogram were 0.651, 0.664, and 0.749 for 1-, 3-, and 5-years OS, respectively (Additional file 4: Fig. S4B). The calibration curve showed that the actual survival probabilities at 1-, 3- and 5-year were almost in accordance with the survival probabilities predicted by the nomogram model (Additional file 4: Fig. S4C). The decision curves showed that the nomogram model was better than other factors in predicting the prognosis in STAD (Additional file 4: Fig. S4D).Table 3Univariate and multivariate Cox regression analysis of various prognostic parameters in STAD patientsVariablesPatient($$n = 317$$)Univariate analysisMultivariate analysisHR [$95\%$ CI]P valueHR [$95\%$ CI]P valueAge < 6715211 ≥ 671652.093[1.412, 3.103] < 0.0011.514[1.068, 2.146]0.020Gender Male2031 Female1141.254 [0.851, 1.849]0.253T stage T1131 T2681.642[0.347, 7.763]0.532 T31511.861[0.338, 10.249]0.475 T4851.611[0.283, 9.160]0.591N stage N0981 N1831.453[0.729, 2.895]0.288 N2651.040[0.426, 2.537]0.932 N3661.658[0.691, 3.977]0.257 NX51.732[0.213, 14.072]0.607M stage M02861 M1201.095[0.441, 2.717]0.845 MX111.125[0.381, 3.323]0.832Tumor stage I441 II1060.988[0.364, 2.679]0.981 III1351.203[0.318, 4.550]0.785 IV322.785[0.652, 11.898]0.167Family history of stomach cancer Yes151 No2430.946[0.541, 1.655]0.845 N/A590.833[0.316, 2.196]0.712H. pylori infection Yes171 No1290.983[0.583, 1.657]0.949 N/A1710.576[0.214, 1.550]0.275Grade G181 G21082.529[0.322,19.847]0.377 G31933.226[0.418, 24.870]0.261 GX84.065[0.378, 43.697]0.247Reflux history Yes351 No1580.972[0.567, 1.665]0.917 N/A1240.561[0.231, 1.364]0.202Disease type Adenomas and adenocarcinomas2891 Cystic, mucinous and serous Neoplasms280.615[0.301, 1.256]0.182Risk High15811 Low1590.525[0.354, 0.780]0.0010.538[0.379, 0.765] < 0.001Bold indicates P value ≤ 0.05 was considered statistically significant
## Functional enrichment analysis of the DEGs in high-risk and low-risk groups
We further conducted functional enrichment analyses of the 298 DEGs in high-risk and low-risk groups (Additional file 10: Table S12). GO enrichment analysis indicated that the differential genes annotated to biological processes were involved in extracellular matrix (ECM) organization and extracellular structure organization. *Differential* genes annotated to cellular component categories were mainly enriched in collagen-containing ECM and collagen trimer. *Differential* genes annotated to molecular function categories were mainly enriched in ECM structural constituent and collagen binding (Fig. 4A, Additional file 10: Table S13). The top 10 pathways obtained by KEGG analysis were: protein digestion and absorption, proteoglycans in cancer, focal adhesion, human papillomavirus infection, PI3K-Akt signaling pathway, ECM-receptor interaction, cell adhesion molecules, axon guidance, cAMP signaling pathway, and vascular smooth muscle contraction (Fig. 4B, Additional file 10: Table S14). GSEA results showed that risk score was significantly associated with ECM glycoproteins, core matrisome, ECM organization in high-risk group (Fig. 4C). The detailed GSEA results for the high-risk and low-risk groups were presented in Additional file 5: Fig. S5.Fig. 4Enrichment analysis in high-risk group and the low-risk group. A Circle map. Bands with different colors in the right half circle symbolized 6 significant GO pathways, including biological process (BP), cellular component (CC), and molecular function (MF). The 6 pathways were enriched by genes listed in the left half circle. B Circle map. Bands with different colors in the right half circle symbolized top 10 significant KEGG pathways. The 10 top pathways were enriched by genes listed in the left half circle. C GSEA recognized different gene sets in the high-risk groups
## Mitochondrial-related risk score was associated with TME signatures in STAD
Given the TME-associated signal pathway was enriched through functional enrichment analyses, we explored the relationship between the risk score and the TME signatures. As shown in Fig. 5A, risk score was highly positively correlated with stromal score in STAD, and stromal score was higher in high-risk group compared to that in the low-risk group. We further investigated the relationship between risk score and TME components. Our results indicated that risk score had a significant and positive correlation with the expressions of the majority of carcinoma associated fibroblast (CAF) signatures (Fig. 5B), as well as ECM-collagen and matrisome signatures (Fig. 5C, D). Taken together, these results suggested a close relationship between mitochondrial-related risk score and TME signatures in STAD.Fig. 5Risk score was associated with TME signatures in STAD. A Association between stromal score and risk score and its distribution in the low- and high-risk groups. B *Correlation analysis* for risk score and the expressions of carcinoma associated fibroblast (CAF) up and down signatures. C *Correlation analysis* for risk score and the expressions of ECM and collagen signatures. D *Correlation analysis* for risk score and the expressions of matrisome signatures. P values were showed as: ***$P \leq 0.001$
## Mitochondrial-related risk score was associated with immune signatures and immunotherapy responses in STAD
The tumor immune microenvironment was closely related with the therapeutic effects and prognosis of patients with malignant tumor. It is reasonable to check the relationship between the risk score and the immune cell infiltration in STAD. The contents of naive B cells, regulatory T cell (Tregs), M0 macrophage, and M2 macrophage were remarkably higher in the high-risk group. In contrast, CD8+ T cells and resting CD4+ T cells were higher in low-risk group (Fig. 6A, Additional file 6: Fig. S6A, B). Consistent with the above results, risk score was positively correlated with the expressions of the majority signatures of M2 macrophage, while negatively correlated with the expressions of the majority signatures of activated CD8+ T cell (Fig. 6B). The correlations between the risk score and the signatures of other immune cells were presented in Additional file 8: Fig. S8. Consistent with the previous studies [28], the ESTIMATE results showed that the patients in high-risk had higher immune score, and significantly lower tumor purity ($$P \leq 0.370$$), than those in the low-risk groups (Fig. 6C, D). Due to the positive correlation between the risk score, and matrisome and CAF signatures, as well as the negative correlation between the risk score and activated CD8+ T cells signatures, we speculated activated CD8+ T cells signatures was negatively correlated with the matrisome and CAF signatures. Interestingly, our results showed that the expression of activated CD8+ T cell signatures were negatively correlated with both matrisome and CAF signatures (Additional file 6: Fig. S6C). Taken together, these results strongly suggested the tumor immunosuppressive microenvironment might contributed to the worse prognosis of the patients with STAD in high-risk group, which needs to be validated in further study. Fig. 6The different immune profiles between the low- and high-risk groups in the TCGA-STAD dataset. Two risk groups were divided based on the median risk score. A CIBERSORT analysis. B Correlation between risk score and the expressions of activated CD8+ T cell and M2 macrophages signatures. C, D ESTIMATE algorithm. E Expression variation of immune checkpoint. P values were showed as: ns not significant; *$P \leq 0.05$; **$P \leq 0.01$; ***$P \leq 0.001$ Nowadays, immune checkpoint inhibitors were studied and well applied in cancer immunotherapy. In the present study. Our results indicated that 43 immune checkpoints were considerably modulated in high-risk group (Fig. 6E). In addition, risk score was significantly positively correlated with the expression level of 7 immune checkpoints, including CD200, NRP1, TNFSF4, B7-H3, TNFSF18, LAIR1 and OX40 (r > 0.2, Additional file 6: Fig. S6D). Currently, the inhibitors for PD-1 and CTLA-4 are research hotspots in the treatment of advanced STAD. As shown in Fig. 6C, the expressions of PD-1, PD-L1 and CTLA-4 were significantly down-regulated in the high-risk group. Consistently, risk score was significantly negatively correlated with the expressions of PD1, PD-L1 and CTLA-4, respectively (Additional file 6: Fig. S6D).
Given the above results, we further used the TIDE algorithm to evaluate the ability of risk score in predicting the responses to immunotherapy in STAD. Our results showed that risk score had a significantly positive correlation with TIDE score (Additional file 7: Fig. S7), indicating that patients in low-risk group received better response to immunotherapy. The immunotherapy response rate in high-risk group ($32.28\%$) was significantly lower than that in low-risk group ($66.67\%$) (Fig. 7A).Fig. 7Risk score is a potential biomarker to predict benefits from immune therapies in STAD. A TIDE predicted the proportion of patients with response to immunotherapy in low-risk and high-risk groups. B The proportion of patients with response to immunotherapy in low-risk and high-risk groups in the PRJEB25780 immunotherapy cohort (45 patients with advanced gastric cancer who had received PD-L1 inhibitor treatment). C TIDE predicted the proportion of patients with response to immunotherapy in low-immune score and high-immune score groups. D TIDE predicted the proportion of patients of four groups based on the risk score and immune score with response to immunotherapy. E TIDE predicted the proportion of patients with response to immunotherapy in low-stromal score and high-stromal score groups. F TIDE predicted the proportion of patients of four groups based on the risk score and immune score with response to immunotherapy. G TIDE predicted the proportion of patients with response to immunotherapy in MSS, MSI-L and MSI-H groups. H TIDE predicted the proportion of patients of six groups based on the risk score and microsatellite status with response to immunotherapy. MSS, Microsatellite stability; MSI-L, Microsatellite Instability-Low; MSI-H, Microsatellite Instability-High. P values were showed as: ns not significant; ***$P \leq 0.001$ Next, PRJEB25780 cohort (PD-L1 inhibitor treatment for 45 patients with advanced gastric cancer [29]) was applied to validate whether the mitochondrial-related risk signature could accurately predict the responses to immunotherapy for patients with STAD. Consistent with the prediction results by TIDE, the immunotherapy respond rate in high-risk group ($13.04\%$) was significantly lower than that in low-risk group ($40.91\%$) in PRJEB25780 validation cohort (Fig. 7B).
As shown in Fig. 7C, the immunotherapy response rate in the low-immune subgroup ($34.59\%$) was remarkably lower than that in the high-immune subgroup ($64.56\%$). Interestingly, the immunotherapy response rate in low-risk group ($45.56\%$) was remarkably higher than that in high-risk group ($20.29\%$) in the subgroup with low-immune score. Moreover, the immunotherapy response rate in low-risk group ($94.20\%$) was significantly higher than that in high-risk group ($41.57\%$) in the subgroup with high-immune score, strongly suggesting that combined risk score and immune score was a robust indicator to predict the responses to immunotherapy in STAD (Fig. 7D). In addition, the immunotherapy response rate in the high stromal group ($37.97\%$) was significantly lower than that in the low stromal group ($61.01\%$) (Fig. 7E). In the low stromal subgroup, the immunotherapy response rates were $65.35\%$ and $53.45\%$ in low risk and high risk subgroup, respectively, which were similar to the low-stromal group ($61.01\%$), indicating combined risk score and stromal score was not better than stromal score alone in predicting the response to immunotherapy in STAD patients with low-stromal score. However, the immunotherapy response rate in the low-risk + high-stromal group ($68.97\%$) was significantly higher than that in high-stromal group ($37.97\%$), whereas the immunotherapy response rate in the high-risk + high-stromal group ($20.00\%$) was significantly lower than that in the high-stromal group ($37.97\%$), strongly suggesting that combined risk score and stromal score can more accurately predict response to immunotherapy in STAD patients with high-stromal score (Fig. 7F).
The phenotype for microsatellite instability–high (MSI-H) is a distinct tumor subclass that is highly susceptible to immunotherapy. Consistent with the previous studies [30], the immunotherapy response rate MSI-H subgroup ($72.73\%$) was remarkably higher than that in the MSS subgroup ($46.30\%$) and MSI-L subgroup ($36.96\%$) (all $$P \leq 0.0005$$, Fig. 7G). Furthermore, our results also showed that the immunotherapy response rate in low-risk group ($65.35\%$) was remarkably higher than that in high-risk group ($29.57\%$) in the subgroup with MSS. The immunotherapy response rate in low-risk group ($57.89\%$) was also significantly higher than that in high-risk group ($22.22\%$) in the subgroup with MSI-L. The above results strongly suggested that combination of risk-score and MSS/MSI-L can be used as a robust indicator to predict the response to immunotherapy in STAD (Fig. 7H).
## Mutation status of STAD patients in high-risk and low-risk groups
Progressive accumulation of mutations throughout life can lead to cancer. Genome sequencing has revolutionized our understanding of somatic mutation in cancer, providing a detailed view of the mutational processes and genes that drive cancer [31]. Therefore, we mapped the mutation landscape of STAD in both the high risk and low risk groups, and analyzed the relationship between risk score and mutation profile. As shown in Fig. 8A, the top 20 high-frequency mutated genes in high-risk and low-risk group were presented. TP53, TTN, MUC16, LRP1B, SYNE1, CSMD3, FAT4, OBSCN, ARID1A, FLG, CSMD1, DNAH5, SPTA1, PCLO and RYR2 were the common high-frequency mutation genes in both groups. In top 5 mutant genes, the mutation rates of TTN and MUC16 were significantly decreased in high-risk group (Fig. 8B).Fig. 8Mutation status in the high- and the low-risk groups in STAD. A The top 20 genes according to mutation frequency in low and high-risk groups, respectively. B Mutation rate of the top five mutant genes in high-risk and low-risk groups. C Relationship between the risk score and TMB. D Correlation between risk score and TMB score in STAD. E Kaplan–Meier curves of the OS of patients in the high- and low-TMB groups in the TCGA-STAD training cohort. P values were showed as: ns not significant; *$P \leq 0.05$; ***$P \leq 0.001$ Besides, accumulating evidences supported that TMB functioned as a potentially predictive biomarker for multiple applications, including the biomarker for response to immunotherapy in malignancies [32–36]. Our data showed that TMB in high-risk group was significantly lower than that in low-risk group, and the risk score had a significant negative correlation with TMB in STAD (Fig. 8C, D). Interestingly, the higher TMB tended to have a better OS compared with the lower TMB, but without a statistically significant difference ($$P \leq 0.153$$, Fig. 8E). Moreover, low-risk group tended to have a better OS compared with high-risk group in high TMB subgroup. The similar results were obtained in the low TMB subgroup, but without a statistically significant difference ($$P \leq 0.099$$, Fig. 8E). Taken together, these results strongly suggested that combination of risk score and TMB might be a valuable biomarker for predicting the prognosis for STAD patients (Fig. 8E).
## Risk score predicts therapeutic benefits in STAD
To find the potency of risk score as an index for predicting the response to drugs (including chemotherapy, targeted therapy, and immunotherapy) in STAD, we inferred the IC50 value of the 138 drugs in TCGA-STAD patients. We found that patients in high-risk group might be more sensitive to rapamycin, PD-0325901, dasatinib, et al., whereas patients in low-risk group might be more sensitive to AZD7762, CEP-701, methotrexate, et al., which could provide a reliable reference for clinical treatment (Fig. 9). Detailed information on the top 10 sensitive drugs for the high-risk and low-risk subgroup was illustrated in Tables 4 and 5.Fig. 9Risk score predicts drug therapeutic benefits in STAD. Proportion of normalized IC50 value of the 60 drugs between the low-risk and high-risk groups. P values were showed as: ns not significant; *$P \leq 0.05$; **$P \leq 0.01$; ***$P \leq 0.001$Table 4Detailed information of the top 10 sensitivity drugs in high-risk groupsDrug nameIntroductionDrug targetDrug target pathwayRapamycinDrugs that selectively target mTORC1 are expected to impair cancer metabolism and are considered promising anti-cancer therapiesMTORC1PI3K/MTOR signalingPD.0325901Mirdametinib (PD-0325901) is an oral, non-ATP-competitive, highly selective, and potent small-molecule inhibitor of MEK1 and MEK2MEK1, MEK2ERK MAPK signalingDasatinibDasatinib is an orally available short-acting dual ABL/SRC tyrosine kinase inhibitor (TKI). It potently inhibits BCR-ABL and SRC family kinases (SRC, LCK, YES, FYN), but also c-KIT, PDGFR-a and PDGFR-b, and ephrin receptor kinaseABL, SRC, Ephrins, PDGFR, KITRTK signaling, kinasesMG-132The peptide-aldehyde proteasome inhibitor MG132 (carbobenzoxyl-l-leucyl-l-leucyl-l-leucine) induces the apoptosis of cells by a different intermediary pathway. Although the pathway of induction of apoptosis is different, it plays a crucial role in anti-tumor treatmentProteasome, CAPN1Protein stability and degradationCytarabineCytarabine (molecular formula: C9H13N3O5) interferes with DNA synthesis, acting on DNA/RNA polymerase (and other nucleotide reductase enzymes), reducing cell ability to replicateAntimetaboliteOtherA-443654A-443654, a specific Akt inhibitor, interferes with mitotic progression and bipolar spindle formation. A-443654 induces G2/M accumulation, defects in centrosome separation, and formation of either monopolar arrays or disorganized spindlesAKT1, AKT2, AKT3PI3K/MTOR signalingAZ628AZ628 is a hydrophobic Raf-kinase inhibitor currently in clinical trial of various cancerBRAFERK MAPK signalingWH-4–023WH-4-023 is a LCK inhibitorsSRC, LCKOther, kinasesMitomycin. CMitomycin C (MMC) is an alkylating agent with extraordinary ability to crosslink DNA, preventing DNA synthesisDNA crosslinkerDNA replicationTW.37TW-37 is a novel, potent and non-peptide Bcl-2 small-molecule inhibitorBCL2, BCL-XL, MCL1Apoptosis regulationTable 5Detailed information of the top 10 sensitivity drugs in low-risk groupsDrug nameIntroductionDrug targetDrug target pathwayAZD7762AZD7762 is a checkpoint kinase 1 (Chk 1) inhibitor, which has been reported to sensitize many tumor cells to DNA damageCHEK1, CHEK2Cell cycleCEP-701CEP‐701 is an inhibitor of tyrosine kinase. Treatment with CCEP-701 modulates various signalling pathways and leads to cell growth arrest and programmed cell death in several tumour typesFLT3, JAK2, NTRK1, NTRK2, NTRK3Other, kinasesMethotrexateMethotrexate (MTX) is a commonly used antimetabolite, which inhibits folate and DNA synthesis to be effective in the treatment of various malignanciesAntimetaboliteDNA replicationMS-275MS-275, a selective class I inhibitor of histone deacetylase (HDAC), exerts anti-tumor activity in various cancer types, including multiple myeloma (MM)HDAC1, HDAC3Chromatin histone acetylationShikoninMany studies have demonstrated that shikonin exerts strong anticancer effects on various types of cancer by inhibiting cell proliferation and migration, inducing apoptosis, autophagy, and necroptosisNot definedOtherGefitinibGefitinib is an orally active, selective epidermal growth factor receptor-tyrosine kinase inhibitorEGFREGFR signalingBIBW2992BIBW2992 is an irreversible blocker of the ErbB family, acting at the tyrosine kinases of these proteins. Further investigations for the treatment of many other tumors with BIBW2992, e.g., HNSCC and breast cancer, are ongoingERBB2, EGFREGFR signalingSunitinibSunitinib is a tyrosine kinase inhibitor indicated for the treatment of gastrointestinal stromal tumor, advanced renal cell carcinoma, and pancreatic neuroendocrine tumorsPDGFR, KIT, VEGFR, FLT3, RET, CSF1RRTK signalingS-Trityl-L-cysteineS-Trityl-L-cysteine (STLC) is a well-recognized lead compound known for its anticancer activity owing to its potent inhibitory effect on human mitotic kinesin Eg5KIF11MitosisBortezomibBortezomib (BTZ) is the first proteasome inhibitor approved by the Food and Drug Administration. It can bind to the amino acid residues of the 26S proteasome, thereby causing the death of tumor cellsProteasomeProtein stability and degradation
## Experimental verification of hub gene expression in STAD
To verify the expressions of hub genes in STAD samples, qRT-PCR was conducted on 10 pairs of STAD tumor and paired adjacent normal tissues. As shown in Fig. 10A, the expressions of NOX4 and FKBP10 were significantly higher in STAD tumor tissues than those in the paired adjacent normal tissues, respectively (all $P \leq 0.05$). The expressions of ALDH3A2 and MAOA were significantly lower in STAD tumor tissues than those in the paired adjacent normal tissues, respectively (all $P \leq 0.05$, Fig. 10A). We also verified the expressions of the four hub genes in a human normal gastric epithelial cell line GES-1 and human gastric cancer cell lines through qRT-PCR. Our results showed that the expressions of NOX4 and FKBP10 were significantly higher than in the normal gastric epithelial cell line GSE-1 (all $P \leq 0.05$, Fig. 10B), while the expressions of ALDH3A2 and MAOA were significantly lower than in the normal gastric epithelial cell line GSE-1 (all $P \leq 0.05$, Fig. 10B). These results supported our hypothesis and provide solid evidence for the rationality of choosing these four genes for prognostic model construction (see Fig. 11).Fig. 10Experimental verification of 4 genes expression in STAD. A Expression of 4 genes in 10 paired STAD tissues and normal tissues was evaluated by qRT-PCR. B Expression of 4 genes in a human normal gastric epithelial cell line GSE-1 and human gastric cancer cell lines through qRT-PCRFig. 11Graph summarization. The work summary graph of this study
## Discussion
GC remains a frequent cancer worldwide with high incidence and mortality globally [1]. Effective biomarkers are still missing even though 3 biomarkers (HER2, MSI-H and PD-L1) have been proven to predict the responses of targeted therapy in GC [37]. Therefore, identifying more effective biomarkers for targeted therapy and prognosis prediction is highly demand. Mitochondria were important pharmacological targets due to their critical roles in cell proliferation and death. The mitochondrial energy metabolisms are now known to be reprogrammed to meet the challenges of high energy demand, with better use of available fuels for malignant cell growth and migration [38]. Thus, mitochondria play a vital and multifunctional role in tumor occurrence and development, and targeting mitochondria provides therapeutic opportunities [39]. A growing body of research showed that mitochondrial-related genes can be used as biomarkers for the diagnosis and treatment of malignancies.
NOX4 has been identified as a biomarker and therapeutic target for a variety of human cancers. NOX4 was upregulated in pancreatic cancer and was involved in the development of pancreatic cancer by promoting cell proliferation, regulating cell metabolism, and mediating angiogenesis, suggesting NOX4 was a potential therapeutic target for pancreatic cancer [40]. Up-regulation of NOX4 predicted worse prognosis and accelerated tumor growth in colorectal carcinoma [41]. Moreover, it has been shown that NOX4 recruited M2-macrophages via ROS/PI3K signaling pathway-dependent cytokines production, thus contributing to the cell division in NSCLC [42]. Consistent with the previous studies, NOX4 in our model was significantly up-regulated in STAD, and high level of NOX4 was associated with worse prognosis in patients with STAD.
It has been reported that a cancer-specific molecular mechanism for NSCLC was related with FKBP10-dependent protein translation. The expression of FKBP10 was positive in cancer lesions [43]. Li et al. reported that FKBP10 silencing decreased the expression of integrin αV and integrin α6, and P-AKT, suggesting that FKBP10 might promote metastasis [44]. FKBP10 was up-regulated in GC tissues and might be a reliable therapeutic target in GC [45]. Consistent with the above studies, our results showed that the high expression of FKBP10 was related with high risk, and predicated poor prognosis of STAD patients. Consistent with the previous studies, FKBP10 in our model was significantly up-regulated in STAD, and high level of FKBP10 was associated with worse prognosis in patients with STAD.
It has been shown that ALDH3A2 was overexpressed in low-grade GC compared with high-grade GC, and patients with low expression of ALDH3A2 had worse OS than those with high ALDH3A2 expression. ALDH3A2 was reported as a reliable biomarker for the immunotherapy, as well as an independent predictor for the prognosis of GC [46]. In renal clear cell carcinoma, low level of ALDH3A2 was related with shorter survival [47]. Consistently, ALDH3A2 was significantly down-regulated in STAD. However, ALDH3A2 couldn’t effectively predict the prognosis for patients with STAD, and this inconsistent effects of ALDH3A2 on prognosis may be caused by the different inclusion and exclusion criteria across different study, which needs more comprehensive investigation.
MAOA exerted different biological effects in different tumors. MAOA was found to be involved in mitochondrial dysfunction, and promoted malignant growth and metastasis in gastric cancer [48]. MAOA promoted prostate cancer progression by increasing cell growth and cancer stem cells, which suggested that MAOA might be a potential therapeutic target for the treatment of prostate cancer [49]. In the present study, MAOA was significantly down-regulated in STAD. However, the high level of MAOA was associated with worse prognosis in patients with STAD, which need to be validated in more samples in further study.
Currently, many biomarkers were applied for prognostic prediction of GC, such as NOX4, FKBP10 and ALDH3A2, but most of them are studied for a single biomarker [44–46, 50]. Increasing evidences indicated that prognostic model constructed by multi-genes as a prognostic index was more comprehensive and effective than single gene in kinds of malignancies. For instance, Nie et al. constructed a GC prognosis model based on metabolic signature, which were mainly related to the dysregulation of the metabolic microenvironment [51]. Wu et al. constructed a immune-related prognostic model [52]. As the dysfunction and dysregulation of mitochondria have been associated with cancer, we constructed a STAD prognostic model based on mitochondrial-related genes, which could effectively predict the prognosis for patients with STAD.
The DEGs between the high-risk and low-risk groups were mainly enriched to the extracellular matrix (ECM) and focal adhesion pathway. ECM accumulation was a classical characteristic feature of tumors, and a higher ECM content predicted a poorer prognosis in a broad range of cancer types [53]. The TME is a composition of cancer cells, non-cancerous stromal cells, soluble growth factors, cytokines, proteases, and ECM, which provides essential signals for tumor survival, growth, and acquisition of invasiveness, while hindering antitumor immunity [24, 54, 55]. Fibroblasts constitute one of the most vital cells in the stroma and turn into cancer-associated fibroblasts (CAFs) in TME. CAFs not only play active roles in tumorigenesis and progression both by soluble factors and direct cell-to-cell contact, but also sculpt TME by suppressing anti-tumor immune responses or by recruiting immunosuppressive cells [54]. Matrisome defined as the compendium of genes encoding core ECM proteins, or structural component of the ECM [53]. Consistently, our results indicated that risk score was positively correlated with the CAF signature, ECM signature, and Matrisome signatures. Moreover, risk score had a positive correlation with stromal score and a negative correlation with tumor purity, which could be on behalf of higher infiltration degrees of stromal cells in the TME of the high-risk group.
The immune cells play essential roles in TME. The success of cancer immunotherapy relies on the comprehensive understanding of the tumor microenvironment and immune evasion mechanisms in which the tumor, stroma, and infiltrating immune cells coordinated in a complex network. The main benefit of immunotherapy is to generate memory CD8+ T cells for sustained protection against metastasis and preventing recurrence of the disease [56]. Active immune cells could enter into the tumor parenchyma and perform their anti-tumor function [55]. Therefore, the ultimate goal of immunotherapy is to convert an immunodorminant TME into an immunostimulatory TME, which allows the immune system to clear tumor lesions [56]. Treg cells are involved in tumor progression by inhibiting antitumor immunity. High Treg cell infiltration in the TME was involved in unfavorable prognosis in patients with various types of cancer [53]. M2-polarized macrophages, commonly deemed tumor-associated macrophages (TAMs), were contributors to many pro-tumorigenic outcomes in cancer [57]. Macrophage type 2 (M2) cells, and Tregs cells could make immunologic barriers against CD8+ T cell‐mediated antitumor immune responses [58]. Our results indicated that naive B cells, regulatory T cell (Tregs), M0 macrophage and M2 macrophage were significantly enriched in high-risk group, whereas CD8+ T cells and resting CD4+ T cells were remarkably enriched in low-risk group. Taken together, these results suggested that high-risk group was suffused with immunosuppressive cells such as Tregs, M2 macrophages, producing the immunosuppressive microenvironment to hamper CD8+ T cells-mediated eradication for tumor cells.
Monoclonal antibodies against immunological checkpoint molecules provided a vast breakthrough in cancer therapeutics. For instance, PD-1/PD-L1 and CTLA-4 inhibitors showed promising therapeutic outcomes [59]. In our study, high-risk group had a considerably lower rate of immunotherapy response than that in the low-risk group, which was consistent with the expression levels of PD-1/PD-L1 and CTLA-4 in the high-risk and low-risk groups. Interestingly, combination of risk score and immune score, or stromal score can more accurately predict the responsiveness to immunotherapy of patients with STAD. These findings further demonstrated the effectiveness of the risk score as a biomarker in predicating the response to immunotherapy. The TMB was associated with the formation of neoantigens which activated antitumor immunity, which was a reliable biomarker to predict the response to PD-L1 therapy [60]. In our study, the TMB in high-risk group was lower than that in low-risk group, which strongly suggested that the lower response rate to immunotherapy in high-risk group might be due to the lower TMB.
In addition, risk score might be helpful in screening the therapeutics drugs for patients with STAD. For instance, an independent study showed that high-risk STAD patients showed higher sensitivity to the chemotherapy agents, including rapamycin [61]. Another study found that methotrexate is suitable to inhibit the function of Early B-cell factors (EBFs) in gastric cancer [62]. In the present study, rapamycin, PD-0325901 and dasatinib were found to be more effective for patients in the high-risk group, whereas AZD7762, CEP-701 and methotrexate were predicted to be more effective for patients in the low-risk group. However, the toxicities of the screened drugs was uncertain. For instance, due to the unpredictable cardiac toxicity, the development of AZD7762 was not going forward in patients with advanced solid tumors [63]. 5-fluorouracil, doxorubicin, high-dose methotrexate (FAMTX) schedule was reported to be active in advanced gastric cancer, and the main toxicity was myelosuppression [64]. Oxidative stress is a component of many diseases, including cancer. Although numerous small molecule drugs evaluated as antioxidants have exhibited potential therapeutic ability in preclinical studies, results from clinical trial was disappointed. A greater understanding of the pharmacological mechanisms through which anti oxidative drug act might provide a rational usage would lead to greater therapeutic success in malignancies [65, 66].
Our research has some unique advantages. First, the combination of multigene has robust predictive capability for cancer prognosis than single genes. An integrated mitochondria-related gene prognostic risk model would play more vital roles in the diagnosis and prognosis of STAD patients. Second, the results of the study provide us with a more accessible method to determine whether patients belong to the high- or low-risk group, which is simple and feasible. In addition, we evaluated its predictive value, chemotherapy efficacy, immunotherapy efficacy and immune cell infiltration for patients with STAD, which could provide individualized management and treatment for STAD patients.
## Limitations
Although our findings in this study have important clinical consequences, there are still some limitations. Firstly, this is a retrospective study, and an independent prospective cohort is needed to verify the risk model constructed in this study. Secondly, this study heavily relied on datasets and computational predictions while validation component is poor. More experimental studies are needed to validate in further studies. Finally, the carcinogenic effects of the prognostic genes in the model and the mechanisms of interaction between prognostic genes and mitochondrial dysfunction in STAD are mainly unknown and need to be further explored. Based on above information, our future direction will focus on three aspects: [1] Applying mouse model to verify our current hypothesis; [2] Collecting gastric cancer cases and clinical information to validate the risk score model, and compare it with the gold standard of clinical diagnosis, and ensure that the risk model constructed in the present study can be applicable for clinical practice; [3] To screen more novel genes associated with mitochondria in more datasets.
## Conclusions
We established a STAD patient risk score model including NOX4, FKBP10, ALDH3A2, and MAOA. Functionally, the risk score was highly correlated to the TME and immune cell infiltration of STAD patients. Combined analysis for risk score and stromal score, or immune score, or MSS/MSI can predict the response to immunotherapy more accurately than single index in STAD. Regarding drug sensitivity, patients in high-risk group was more sensitive to rapamycin, PD-0325901 and dasatinib, whereas patients in low-risk group was more sensitive to AZD7762, CEP-701 and methotrexate. Taken together, our mitochondrial-related risk model could be a reliable prognostic biomarker for personalized treatment of STAD patients.
## Supplementary Information
Additional file 1: Figure S1. Identification of DEGs related to mitochondrion and functional enrichment analysis in STAD. A The GO analysis of 183 mitochondrial-related DEGs, including biological process (BP), cellular component (CC), and molecular function (MF). B The KEGG analysis of 183 mitochondrial-related DEGs. C Protein expressions of the 3 prognosis-related genes in gastric cancer tissues and normal gastric tissues from HPA database. D Kaplan–Meier curves for OS of patients with high or low expression of each prognosis-related genes. Additional file 2: Figure S2. Assessing the performance of the prognostic risk model in the validation cohort. A Kaplan–Meier curves of the overall survival (OS) in the validation cohort GSE15459. B ROC curves for 1-, 3-, and 5-year OS of the prognostic risk model in the validation cohort GSE15459. C-D Distribution of risk score, survival status (red dots indicate dead, blue dots indicate alive) and the four genes expression heat map in the validation cohort GSE15459. E Gene expression of 4 prognosis-related genes in GSE66229. P values were showed as: ns not significant; *$p \leq 0.05$; **$p \leq 0.01$; ***$p \leq 0.001$Additional file 3: Figure S3. The relationships between the risk score and clinical characteristics of STAD patients. Age, Gender, T stage, N stage, M stage, Tumor stage, Family history of stomach cancer, H pylori infection, Grade, Reflux, Disease type. NX, lymph nodal status could not be determined; MX, metastatic status could not be determined; GX, tumor grade could not be determined; N/A, not available. P values were showed as: ns not significant; *$p \leq 0.05$; **$p \leq 0.01$; ***$p \leq 0.001$Additional file 4: Figure S4. Construction of nomogram using TCGA-STAD cohort. A A nomogram was constructed based on risk score and related clinical characteristics. B ROC curves and AUC for 1-, 3-, and 5-year OS of the nomogram. C Calibration curves of 1-, 3-, and 5-year OS in the nomogram and ideal model. D DCA results of risk score and clinical characteristics in the TCGA cohort. Additional file 5: Figure S5. Results of GSEA analysis in low-risk and high-risk groups in STAD. A The GSEA findings of the c2 reference gene sets for high-risk groups. B The GSEA findings of the c2 reference gene sets for low-risk groups. Additional file 6: Figure S6. The condition of Immune infiltration in low-risk and high-risk group. A The proportion of 22 immune cells quantified by CIBERSORT algorithm in low-risk group. B The proportion of 22 immune cells quantified by CIBERSORT algorithm in high-risk group. C *Correlation analysis* between the activated CD8 + T cell signatures and CAF signatures, as well as that between the activated CD8 + T cell signatures and matrisome signatures. P values were showed as: ns not significant; *$p \leq 0.05$; **$p \leq 0.01$; ***$p \leq 0.001$Additional file 7: Figure S7. Correlation analysis for risk score and TIDE score in STAD.Additional file 8: Figure S8. Correlation analysis for risk score and 14 immune cells signatures in STAD.Additional file 9: Figure S9. Risk score predicts drug therapeutic benefits in STAD. Proportion of normalized IC50 value of the 138 drugs between the low-risk and high-risk groups. P values were showed as: ns not significant; *$p \leq 0.05$; **$p \leq 0.01$; ***$p \leq 0.001$Additional file 10: Tables S1–S14. Table S1. The list of mitochondrial-related genes. Table S2. The list of carcinoma associated fibroblast up signatures ($$n = 24$$). Table S3. The list of carcinoma associated fibroblast down signatures ($$n = 24$$). Table S4. The list of ECM and Collagen signatures ($$n = 225$$). Table S5. The list of matrisome signatures [1026]. Table S6. The primer sequences. Table S7. The list of all DEGs in tumor and normal groups ($$n = 2381$$). Table S8. The list of protein-coding DEGs in tumor and normal groups ($$n = 2145$$). Table S9. The list of mitochondrial-related DEGs ($$n = 183$$). Table S10. GO terms in tumor and normal groups. Table S11. KEGG terms in tumor and normal groups. Table 12. The list of DEGs in high and low risk groups ($$n = 298$$). Table S13. GO terms in high and low risk groups. Table S14. KEGG terms in high and low risk groups.
## References
1. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, Bray F. **Global cancer statistics 2020 GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries**. *CA Cancer J Clin* (2021.0) **71** 209-49. DOI: 10.3322/caac.21660
2. Hoft SG, Noto CN, DiPaolo RJ. **Two distinct etiologies of gastric cancer: infection and autoimmunity**. *Front Cell Dev Biol* (2021.0) **9** 752346. DOI: 10.3389/fcell.2021.752346
3. Machlowska J, Baj J, Sitarz M, Maciejewski R, Sitarz R. **Gastric cancer: epidemiology, risk factors, classification, genomic characteristics and treatment strategies**. *Int J Mol Sci* (2020.0) **21** 4012. DOI: 10.3390/ijms21114012
4. Poorolajal J, Moradi L, Mohammadi Y, Cheraghi Z, Gohari-Ensaf F. **Risk factors for stomach cancer: a systematic review and meta-analysis**. *Epidemiol Health* (2020.0) **42** e2020004. DOI: 10.4178/epih.e2020004
5. Ilic M, Ilic I. **Epidemiology of stomach cancer**. *World J Gastroenterol* (2022.0) **28** 1187-1203. DOI: 10.3748/wjg.v28.i12.1187
6. Tanprasert P, Limpakan S, Chattipakorn SC, Chattipakorn N, Shinlapawittayatorn K. **Targeting mitochondria as a therapeutic anti-gastric cancer approach**. *Apoptosis* (2022.0) **27** 163-183. DOI: 10.1007/s10495-022-01709-0
7. Lee H. **Somatic alterations in mitochondrial DNA and mitochondrial dysfunction in gastric cancer progression**. *World J Gastroenterol* (2014.0) **20** 3950. DOI: 10.3748/wjg.v20.i14.3950
8. Rodrigues T, Ferraz LS. **Therapeutic potential of targeting mitochondrial dynamics in cancer**. *Biochem Pharmacol* (2020.0) **182** 114282. DOI: 10.1016/j.bcp.2020.114282
9. Ahmadian E, Babaei H, Mohajjel NA, Eftekhari A, Eghbal MA. **Venlafaxine-induced cytotoxicity towards isolated rat hepatocytes involves oxidative stress and mitochondrial/lysosomal dysfunction**. *Adv Pharm Bull* (2016.0) **6** 521-530. DOI: 10.15171/apb.2016.066
10. Chodari L, Dilsiz AM, Vahedi P, Alipour M, Vahed SZ, Khatibi S, Ahmadian E, Ardalan M, Eftekhari A. **Targeting mitochondrial biogenesis with polyphenol compounds**. *Oxid Med Cell Longev* (2021.0) **2021** 4946711. DOI: 10.1155/2021/4946711
11. Ma X, Gong N, Zhong L, Sun J, Liang XJ. **Future of nanotherapeutics: targeting the cellular sub-organelles**. *Biomaterials* (2016.0) **97** 10-21. DOI: 10.1016/j.biomaterials.2016.04.026
12. Li Z, Liu ZM, Xu BH. **A meta-analysis of the effect of microRNA-34a on the progression and prognosis of gastric cancer**. *Eur Rev Med Pharmacol Sci* (2018.0) **22** 8281-8287. PMID: 30556868
13. Chang W. **Inflammation-related factors predicting prognosis of gastric cancer**. *World J Gastroenterol* (2014.0) **20** 4586. DOI: 10.3748/wjg.v20.i16.4586
14. Shao W, Yang Z, Fu Y, Zheng L, Liu F, Chai L, Jia J. **The pyroptosis-related signature predicts prognosis and indicates immune microenvironment infiltration in gastric cancer**. *Front Cell Dev Biol* (2021.0) **9** 676485. DOI: 10.3389/fcell.2021.676485
15. Peng Y, Liu C, Li M, Li W, Zhang M, Jiang X, Chang Y, Liu L, Wang F, Zhao Q. **Identification of a prognostic and therapeutic immune signature associated with hepatocellular carcinoma**. *Cancer Cell Int* (2021.0) **21** 98. DOI: 10.1186/s12935-021-01792-4
16. Khalaf K, Hana D, Chou JT, Singh C, Mackiewicz A, Kaczmarek M. **Aspects of the tumor microenvironment involved in immune resistance and drug resistance**. *Front Immunol* (2021.0) **12** 656364. DOI: 10.3389/fimmu.2021.656364
17. Petitprez F, Meylan M, de Reyniès A, Sautès-Fridman C, Fridman WH. **The tumor microenvironment in the response to immune checkpoint blockade therapies**. *Front Immunol.* (2020.0) **11** 784. DOI: 10.3389/fimmu.2020.00784
18. Samstein RM, Lee CH, Shoushtari AN, Hellmann MD, Shen R, Janjigian YY, Barron DA, Zehir A, Jordan EJ, Omuro A. **Tumor mutational load predicts survival after immunotherapy across multiple cancer types**. *Nat Genet* (2019.0) **51** 202-206. DOI: 10.1038/s41588-018-0312-8
19. Goldman MJ, Craft B, Hastie M, Repečka K, McDade F, Kamath A, Banerjee A, Luo Y, Rogers D, Brooks AN. **Visualizing and interpreting cancer genomics data via the Xena platform**. *Nat Biotechnol* (2020.0) **38** 675-678. DOI: 10.1038/s41587-020-0546-8
20. Rath S, Sharma R, Gupta R, Ast T, Chan C, Durham TJ, Goodman RP, Grabarek Z, Haas ME, Hung W. **MitoCarta3.0: an updated mitochondrial proteome now with sub-organelle localization and pathway annotations**. *Nucleic Acids Res.* (2021.0) **49** D1541-7. DOI: 10.1093/nar/gkaa1011
21. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES. **Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles**. *Proc Natl Acad Sci U S A* (2005.0) **102** 15545-15550. DOI: 10.1073/pnas.0506580102
22. Mootha VK, Lindgren CM, Eriksson KF, Subramanian A, Sihag S, Lehar J, Puigserver P, Carlsson E, Ridderstrale M, Laurila E. **PGC-1alpha-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes**. *Nat Genet* (2003.0) **34** 267-273. DOI: 10.1038/ng1180
23. Zhao HB, Zeng YR, Han ZD, Zhuo YJ, Liang YK, Hon CT, Wan S, Wu S, Dahl D, Zhong WD. **Novel immune-related signature for risk stratification and prognosis in prostatic adenocarcinoma**. *Cancer Sci* (2021.0) **112** 4365-4376. DOI: 10.1111/cas.15062
24. Wu J, Li L, Zhang H, Zhao Y, Zhang H, Wu S, Xu B. **A risk model developed based on tumor microenvironment predicts overall survival and associates with tumor immunity of patients with lung adenocarcinoma**. *Oncogene* (2021.0) **40** 4413-4424. DOI: 10.1038/s41388-021-01853-y
25. Ru B, Wong CN, Tong Y, Zhong JY, Zhong S, Wu WC, Chu KC, Wong CY, Lau CY, Chen I. **TISIDB: an integrated repository portal for tumor-immune system interactions**. *Bioinformatics* (2019.0) **35** 4200-4202. DOI: 10.1093/bioinformatics/btz210
26. Cerami E, Gao J, Dogrusoz U, Gross BE, Sumer SO, Aksoy BA, Jacobsen A, Byrne CJ, Heuer ML, Larsson E. **The cBio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data**. *Cancer Discov* (2012.0) **2** 401-404. DOI: 10.1158/2159-8290.CD-12-0095
27. Gao J, Aksoy BA, Dogrusoz U, Dresdner G, Gross B, Sumer SO, Sun Y, Jacobsen A, Sinha R, Larsson E. **Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal**. *Sci Signal* (2013.0) **6** l1. DOI: 10.1126/scisignal.2004088
28. Gong Z, Zhang J, Guo W. **Tumor purity as a prognosis and immunotherapy relevant feature in gastric cancer**. *Cancer Med* (2020.0) **9** 9052-9063. DOI: 10.1002/cam4.3505
29. Lin Y, Pan X, Zhao L, Yang C, Zhang Z, Wang B, Gao Z, Jiang K, Ye Y, Wang S. **Immune cell infiltration signatures identified molecular subtypes and underlying mechanisms in gastric cancer**. *Npj Genom Med* (2021.0) **6** 83. DOI: 10.1038/s41525-021-00249-x
30. Chao J, Fuchs CS, Shitara K, Tabernero J, Muro K, Van Cutsem E, Bang YJ, De Vita F, Landers G, Yen CJ. **Assessment of pembrolizumab therapy for the treatment of microsatellite instability-high gastric or gastroesophageal junction cancer among patients in the KEYNOTE-059, KEYNOTE-061, and KEYNOTE-062 clinical trials**. *Jama Oncol* (2021.0) **7** 895-902. DOI: 10.1001/jamaoncol.2021.0275
31. Martincorena I, Campbell PJ. **Somatic mutation in cancer and normal cells**. *Science* (2015.0) **349** 1483-1489. DOI: 10.1126/science.aab4082
32. Marcus L, Fashoyin-Aje LA, Donoghue M, Yuan M, Rodriguez L, Gallagher PS, Philip R, Ghosh S, Theoret MR, Beaver JA. **FDA approval summary: pembrolizumab for the treatment of tumor mutational burden-high solid tumors**. *Clin Cancer Res* (2021.0) **27** 4685-4689. DOI: 10.1158/1078-0432.CCR-21-0327
33. 33.FDA approves first drug for cancers with a high tumor mutational burden. American Cancer Society. 2020.
34. Kim JY, Kronbichler A, Eisenhut M, Hong SH, van der Vliet HJ, Kang J, Shin JI, Gamerith G. **Tumor mutational burden and efficacy of immune checkpoint inhibitors: a systematic review and meta-analysis**. *Cancers* (2019.0) **11** 1798. DOI: 10.3390/cancers11111798
35. Owada-Ozaki Y, Muto S, Takagi H, Inoue T, Watanabe Y, Fukuhara M, Yamaura T, Okabe N, Matsumura Y, Hasegawa T. **Prognostic impact of tumor mutation burden in patients with completely resected non-small cell lung cancer: brief report**. *J Thorac Oncol* (2018.0) **13** 1217-1221. DOI: 10.1016/j.jtho.2018.04.003
36. Riviere P, Goodman AM, Okamura R, Barkauskas DA, Whitchurch TJ, Lee S, Khalid N, Collier R, Mareboina M, Frampton GM. **High tumor mutational burden correlates with longer survival in immunotherapy-naïve patients with diverse cancers**. *Mol Cancer Ther* (2020.0) **19** 2139-2145. DOI: 10.1158/1535-7163.MCT-20-0161
37. Choi S, Park S, Kim H, Kang SY, Ahn S, Kim K. **Gastric cancer: mechanisms, biomarkers, and therapeutic approaches**. *Biomedicines* (2022.0) **10** 543. DOI: 10.3390/biomedicines10030543
38. Roth KG, Mambetsariev I, Kulkarni P, Salgia R. **The mitochondrion as an emerging therapeutic target in cancer**. *Trends Mol Med* (2020.0) **26** 119-134. DOI: 10.1016/j.molmed.2019.06.009
39. Zong WX, Rabinowitz JD, White E. **Mitochondria and cancer**. *Mol Cell* (2016.0) **61** 667-676. DOI: 10.1016/j.molcel.2016.02.011
40. Bi Y, Lei X, Chai N, Linghu E. **NOX4: a potential therapeutic target for pancreatic cancer and its mechanism**. *J Transl Med.* (2021.0) **19** 1. DOI: 10.1186/s12967-021-03182-w
41. Lin X, Yang L, Fu S, Lin W, Gao Y, Chen H, Ge Z. **Overexpression of NOX4 predicts poor prognosis and promotes tumor progression in human colorectal cancer**. *Oncotarget* (2017.0) **8** 33586-33600. DOI: 10.18632/oncotarget.16829
42. Zhang J, Li H, Wu Q, Chen Y, Deng Y, Yang Z, Zhang L, Liu B. **Tumoral NOX4 recruits M2 tumor-associated macrophages via ROS/PI3K signaling-dependent various cytokine production to promote NSCLC growth**. *Redox Biol* (2019.0) **22** 101116. DOI: 10.1016/j.redox.2019.101116
43. Ramadori G, Ioris RM, Villanyi Z, Firnkes R, Panasenko OO, Allen G, Konstantinidou G, Aras E, Brenachot X, Biscotti T. **FKBP10 regulates protein translation to sustain lung cancer growth**. *Cell Rep* (2020.0) **30** 3851-3863. DOI: 10.1016/j.celrep.2020.02.082
44. Gong LB, Zhang C, Yu RX, Li C, Fan YB, Liu YP, Qu XJ. **FKBP10 acts as a new biomarker for prognosis and lymph node metastasis of gastric cancer by bioinformatics analysis and in vitro experiments**. *Onco Targets Ther* (2020.0) **13** 7399-7409. DOI: 10.2147/OTT.S253154
45. Liang L, Zhao K, Zhu JH, Chen G, Qin XG, Chen JQ. **Comprehensive evaluation of FKBP10 expression and its prognostic potential in gastric cancer**. *Oncol Rep* (2019.0) **42** 615-628. PMID: 31233188
46. Yin Z, Wu D, Shi J, Wei X, Jin N, Lu X, Ren X. **Identification of ALDH3A2 as a novel prognostic biomarker in gastric adenocarcinoma using integrated bioinformatics analysis**. *Bmc Cancer.* (2020.0) **20** 1
47. Zhao Y, Tao Z, Chen X. **A three-metabolic-genes risk score model predicts overall survival in clear cell renal cell carcinoma patients**. *Front Oncol* (2020.0) **10** 570281. DOI: 10.3389/fonc.2020.570281
48. Chen L, Guo L, Sun Z, Yang G, Guo J, Chen K, Xiao R, Yang X, Sheng L. **Monoamine oxidase A is a major mediator of mitochondrial homeostasis and glycolysis in gastric cancer progression**. *Cancer Manag Res* (2020.0) **12** 8023-8035. DOI: 10.2147/CMAR.S257848
49. Liao C, Lin T, Li P, Geary LA, Chen K, Vaikari VP, Wu JB, Lin C, Gross ME, Shih JC. **Loss of MAOA in epithelia inhibits adenocarcinoma development, cell proliferation and cancer stem cells in prostate**. *Oncogene* (2018.0) **37** 5175-5190. DOI: 10.1038/s41388-018-0325-x
50. Tang CT, Gao YJ, Ge ZZ. **NOX4, a new genetic target for anti-cancer therapy in digestive system cancer**. *J Dig Dis* (2018.0) **19** 578-585. DOI: 10.1111/1751-2980.12651
51. Nie Y, Liu L, Liu Q, Zhu X. **Identification of a metabolic-related gene signature predicting the overall survival for patients with stomach adenocarcinoma**. *PeerJ* (2021.0) **9** e10908. DOI: 10.7717/peerj.10908
52. Wu M, Xia Y, Wang Y, Fan F, Li X, Song J, Ding J. **Development and validation of an immune-related gene prognostic model for stomach adenocarcinoma**. *Bioscience Rep.* (2020.0) **40** BSR20201012. DOI: 10.1042/BSR20201012
53. Izzi V, Davis MN, Naba A. **Pan-cancer analysis of the genomic alterations and mutations of the matrisome**. *Cancers* (2020.0) **12** 2046. DOI: 10.3390/cancers12082046
54. Gok Yavuz B, Gunaydin G, Kosemehmetoglu K, Karakoc D, Ozgur F, Guc D. **The effects of cancer-associated fibroblasts obtained from atypical ductal hyperplasia on anti-tumor immune responses**. *Breast J* (2018.0) **24** 1099-1101. DOI: 10.1111/tbj.13139
55. Labani-Motlagh A, Ashja-Mahdavi M, Loskog A. **The tumor microenvironment: a milieu hindering and obstructing antitumor immune responses**. *Front Immunol.* (2020.0) **11** 940. DOI: 10.3389/fimmu.2020.00940
56. Locy H, de Mey S, de Mey W, De Ridder M, Thielemans K, Maenhout SK. **Immunomodulation of the tumor microenvironment: turn foe into friend**. *Front Immunol.* (2018.0) **9** 2909. DOI: 10.3389/fimmu.2018.02909
57. Boutilier AJ, Elsawa SF. **Macrophage polarization states in the tumor microenvironment**. *Int J Mol Sci* (2021.0) **22** 6995. DOI: 10.3390/ijms22136995
58. Farhood B, Najafi M, Mortezaee K. **CD8(+) cytotoxic T lymphocytes in cancer immunotherapy: a review**. *J Cell Physiol* (2019.0) **234** 8509-8521. DOI: 10.1002/jcp.27782
59. Darvin P, Toor SM, Sasidharan Nair V, Elkord E. **Immune checkpoint inhibitors: recent progress and potential biomarkers**. *Exp Mol Med* (2018.0) **50** 1-11. DOI: 10.1038/s12276-018-0191-1
60. Xu F, Huang X, Li Y, Chen Y, Lin L. **m6A-related lncRNAs are potential biomarkers for predicting prognoses and immune responses in patients with LUAD**. *Mol Ther Nucleic Acids* (2021.0) **24** 780-791. DOI: 10.1016/j.omtn.2021.04.003
61. Tong X, Yang X, Tong X, Zhai D, Liu Y. **Complement system-related genes in stomach adenocarcinoma: Prognostic signature, immune landscape, and drug resistance**. *Front Genet* (2022.0) **13** 903421. DOI: 10.3389/fgene.2022.903421
62. Wang Q, Liang J, Hu X, Gu S, Xu Q, Yan J. **Early B-cell factors involve in the tumorigenesis and predict the overall survival of gastric cancer**. *Biosci Rep.* (2021.0) **41** 20210055. DOI: 10.1042/BSR20210055
63. Sausville E, Lorusso P, Carducci M, Carter J, Quinn MF, Malburg L, Azad N, Cosgrove D, Knight R, Barker P. **Phase I dose-escalation study of AZD7762, a checkpoint kinase inhibitor, in combination with gemcitabine in US patients with advanced solid tumors**. *Cancer Chemother Pharmacol* (2014.0) **73** 539-549. DOI: 10.1007/s00280-014-2380-5
64. Bar SG, Tsalic M, Gaitini D, Steiner M, Haim N. **Etoposide, doxorubicin and cisplatin alternating with 5-fluorouracil, doxorubicin and high-dose methotrexate in patients with advanced adenocarcinoma of the stomach or the gastroesophageal junction**. *J Chemother* (2002.0) **14** 623-626. DOI: 10.1179/joc.2002.14.6.623
65. Forman HJ, Zhang H. **Author correction: targeting oxidative stress in disease: promise and limitations of antioxidant therapy**. *Nat Rev Drug Discov* (2021.0) **20** 652. DOI: 10.1038/s41573-021-00267-5
66. Ahmadian E, Eftekhari A, Kavetskyy T, Khosroushahi AY, Turksoy VA, Khalilov R. **Effects of quercetin loaded nanostructured lipid carriers on the paraquat-induced toxicity in human lymphocytes**. *Pestic Biochem Physiol* (2020.0) **167** 104586. DOI: 10.1016/j.pestbp.2020.104586
|
---
title: 'Volume-accumulated reflectivity of the outer retina (integral) on spectral
domain optical coherence tomography as a predictor of cone cell density: a pilot
study'
authors:
- Wenting Li
- Wenwen Chen
- Xiyue Zhou
- Tingting Jiang
- Juan Zhang
- Min Wang
- Jihong Wu
- Junxiang Gu
- Qing Chang
journal: BMC Ophthalmology
year: 2023
pmcid: PMC10012552
doi: 10.1186/s12886-023-02827-2
license: CC BY 4.0
---
# Volume-accumulated reflectivity of the outer retina (integral) on spectral domain optical coherence tomography as a predictor of cone cell density: a pilot study
## Abstract
### Background
The study aims to investigate the relationship between the volume-accumulated reflectivity (termed “integral”) on spectral domain optical coherence tomography (SD-OCT) and cone density on adaptive optics (AO) imaging.
### Methods
In this cross-sectional study, both eyes of 32 healthy subjects and 5 patients with inherited retinal diseases (IRD) were studied. The parameter, integral, was defined as the volume-accumulated reflectivity values in a selected region on OCT images; integrals of the ellipsoid zone (EZ) and interdigitation zone (IZ) were measured at 2°, 3°, 4°, 5°and 6° eccentricity along the four meridians on fovea-centered OCT B-scans. Cone density in the same region was measured using a flood illumination adaptive optics camera RTX1.
### Results
Integrals of EZ, IZ and cone density shared similar distribution patterns. Integral of the IZ was better correlated with cone density in both healthy people ($r = 0.968$, $p \leq 0.001$) and those with IRD ($r = 0.823$, $p \leq 0.001$) than direct measurements of reflectivity on OCT images. A strong correlation was found between best corrected visual acuity (BCVA) and cone density at 2° eccentricity (r = -0.857, $$p \leq 0.002$$). BCVA was also correlated with the integral of the IZ at the foveola (r = -0.746, $$p \leq 0.013$$) and fovea (r = -0.822, $$p \leq 0.004$$).
### Conclusions
The new parameter “integral” of the photoreceptor outer segment measured from SD-OCT was noted to correlate with cone density and visual function in this pilot study.
## Introduction
Optical coherence tomography (OCT) is a non-invasive imaging modality providing the morphological features of the retina at all levels with high scanning speed and axial resolution. Since its introduction to ophthalmology, spectral domain OCT (SD-OCT) has been widely used for disease diagnosis, treatment monitoring, and prognosis assessment [1, 2]. The four hyperreflective bands presented in the outer layer of the retina on OCT images from the inner layer to the outer layer were named external limiting membrane (ELM), ellipsoid zone (EZ), interdigitation zone (IZ), and retinal pigment epithelium/Bruch’s complex (RPE), respectively [3]. The segment from EZ to IZ represents the outer segment of the photoreceptor and is found to be initially disrupted in many inherited retinal diseases (IRD) [4].
Several methods for quantifying the outer layers of the retina on OCT have been found to achieve a refined clinical analysis, including the individual band or central macula thickness [5–7], volume [5, 8, 9], disruption length [10, 11] and reflectance [10, 12]. Several studies have been devoted to assessing the correlation between these quantitative values with the visual function, such as best corrected visual acuity (BCVA) and ERG (Electroretinogram) findings [5, 6, 10–14]. Besides, adaptive optics (AO) imaging is a new imaging modality. Several aberrations such as media opacity and prominent vessel shadowing [15] affecting OCT imaging quality were corrected in AO. With a lateral resolution of fewer than 2 microns, AO allows in-vivo visualization of retinal cells and enables the monitoring of a single photoreceptor cell [16, 17]. However, few investigators underlined the association between quantitative assessment on OCT and cone density on AO imaging [18–20]. Although AO provides direct information about cone cells, it is relatively difficult to operate and requires better coordination and fixation. OCT is more widely available, so there is still a need to find a closely related metric to AO in OCT for a quick and accurate assessment of cone density.
Our previous study introduced a new parameter, “integral”, as a quantitative method for the photoreceptor outer segment on OCT [21]. In this study, we further assessed the distribution characteristics of the integral and evaluated its association with cone density measured by adaptive optics imaging.
## Methods
The single-centered cross-sectional study was conducted in Eye and ENT Hospital of Fudan University (Shanghai, China) conformed to the Declaration of Helsinki. The protocol was approved by the Institutional Review Board of the Eye and ENT Hospital of Fudan University. Informed consent was obtained from all participants.
## Participants
Healthy volunteers and patients diagnosed with IRD presenting with outer retinal impairment on OCT B-scans were studied. Healthy subjects from 20 to 39 years were included if the best corrected visual acuity (BCVA) was 0.00LogMAR or better, the refractive error was between -6D and + 3D, and axial length ranged from 22 to 26 mm. All were divided into two age groups, the younger group (< 30 years) and the elder group (≥ 30 years). Subjects with other ocular conditions, media opacities, posterior scleral staphyloma, history of eye surgery or trauma were excluded. Patients who visited the Eye and ENT Hospital of Fudan University between January 2019 and September 2020 and were diagnosed with IRD featuring disruption of the outer retinal layer on OCT cross-sectional B-scans were enrolled. The diagnosis was based on the inheritance pattern, fundus appearance, characteristic electroretinograms or genetic analysis. Exclusion criteria included poor fixation for a fovea-centered OCT or AO scan, poor image quality, media opacities, macular edema, posterior sclera staphyloma, and uncontrolled eye movement.
All subjects underwent a complete eye examination, including slit-lamp examination, fundus imaging, measurement of axial length, spherical error, and BCVA. Ocular axial length was measured by IOL Master 500 (Carl Zeiss Meditec, Dublin, CA, USA). OCT images were collected using SD-OCT (Spectralis HRA + OCT, Heidelberg Engineering, Heidelberg, Germany). Besides, all the patients were performed electrophysiological (ERG) studies and the molecular testing previously reported [22], which included targeted exon sequencing followed by sanger sequencing and segregation analysis.
## Adaptive optics imaging and analysis
AO images were obtained in all the included eyes using a flood illumination (FIO) adaptive optics camera (RTX1, Imagine Eyes, Orsay, France) without pupil dilation. The system was based on a central wavelength of 850 nm. Before image acquisition, the participant’s axial length and the refractive error should be entered to correct for spherical ametropia through an inbuilt formula. A built-in fixation target displayed as a yellow cross was set for participants, first in the central macular, then to a predetermined location in the periphery of the retinal coordinate. Imaging depth was adjusted from 0-90 μm to achieve the sharpest photoreceptor cells. Cone cells within 2 degrees eccentricity could not be accurately identified due to the limitation of device resolution and the effect of macular cell bulging [23]. Thus, images were acquired at 2°, 3°, 4°, 5°and 6°eccentricity along four meridians (superior, inferior, temporal and nasal). Eccentricity was defined as the distance between the foveal center and the captured image center. Each captured image was 4°\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times$$\end{document}× 4° (1200 μm \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times$$\end{document}× 1200 μm) and the final output image was the average of 40 high-resolution raw frames. Good fixation allowed the AO to capture images with the imaging center right at the pre-set location to maximize repeatability. A wide-field AO montage was then created using the montage tool I2K Retina software (Dual Align, Clifton Park, NY, USA) (Fig. 1).Fig. 1The AO and OCT Images of the right eye in a healthy subject. A: Correspondence of the areas captured in the horizontal meridian on the montaged near-infrared retinal fundus image, OCT B-scan and AO montage. Scale bar, 200 μm. B-F: AO images were acquired at 2°, 3°, 4°, 5°and 6° eccentricity in the horizontal meridian from left to right. Scale bar, 20 μm. G: The wide-field AO montage of the AO images in the horizontal meridian created using the montage tool I2K Retina software. H: The grid diagram showed all the sampling areas in the four meridians. Each grid on the diagram represented the range of 1°\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times$$\end{document}× 1°. The sampling areas include multiple anatomical regions of the retina (foveola, fovea, parafovea and perifovea). I: For each sampling area, cone density and integrals of the hyperreflective layers were measured. The four peaks on the grayscale curve represented ELM, EZ, IZ and RPE-Bruch’s complex from left to right. Abbreviation: AO, adaptive optics; OCT, optical coherence tomography; ELM, external limiting membrane; EZ, ellipsoid zone; IZ, interdigitation zone; RPE, retinal pigment epithelium For AO analysis, a 0.3°\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times$$\end{document}× 0.3°(90 μm \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times$$\end{document}× 90 μm) region of interest (ROI) was placed at each eccentricity along the four meridians. ROI was manually shifted slightly away to get a clear cone mosaic image when the ROI fell right at the blood vessels or in the shadows. Cones were identified by an automated counting software (AO detect, Imagine Eyes, France) with reflectance values higher than the surrounding background value. Output results included cone density, cone spacing, and percentage of cones with six neighbors (as determined from the Voronoi diagram). The manual adjustment was made for the cells that were not identified or incorrectly identified in the automatic count (Fig. 2). Cone cell identification was verified by two independent investigators (WL, XZ), and the final presented results were the average of the two measurements. Fig. 2The AO cone recognition in a healthy subject at 2 o eccentricity. The white square was a 0.3 o \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times$$\end{document}× 0.3 o(90 μm \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times$$\end{document}× 90 μm) ROI. Left: Red dots represented the cone cells identified by the automated counting software AO detect. Green dots represented the cone cells missed by the automatic recognition. Right: Voronoi tessellation after manual adjustment. Scale bar, 50 μm Abbreviation AO, Adaptive optics
## Quantitative analysis of photoreceptor outer segment
OCT scan was localized to a 30-degree range centered on the fovea, with horizontal and vertical line scans right across the fovea and a signal-to-noise ratio of no less than 30 dB. A hundred B-scans were averaged for each image by ART (Automatic Real-Time) software. Regions of interest were set at the fovea center and 2°, 3°, 4°, 5°and 6° eccentricity along the four meridians corresponding to the AO sampling areas. Each ROI was 200 μm in width along the scanning line and contained the four hyperreflective bands in the outer retina (Fig. 1).
The grayscale images were imported to ImageJ (http://imageJ.nih.gov/ij/; National Institutes of Health, Bethesda, MD, USA). Average grayscale values of each row of pixels at each sampling area were calculated and the grayscale curve was plotted (Fig. 3). The calculation of “integral” has been introduced in our previous article [21]. Briefly, the integral of each layer corresponding to the curve peak was accumulated and adjusted using an integration algorithm. The derivation and the second order derivatives of curves could be used to differentiate between adjacent layers in exceptional cases (Fig. 3).Fig. 3Calculation of integrals through OCT B-scan of the outer retina. Images are of the right eye of a 52-year-old man with retinitis pigmentosa. A: The B-scan image of the horizontal line through the fovea. B: The sampling area was outlined in white in (A). The high-reflection bands from superficial to deep represent ELM, EZ, IZ and RPE, respectively. The AUC of each peak divided was recorded as the original integral of each band. The integral value was calculated as the percentage ratio of each original integral over the total integral of the four bands at the same sampling position. Abbreviation: OCT, optical coherence tomography; ELM, external limiting membrane; EZ, ellipsoid zone; IZ, interdigitation zone; RPE, retinal pigment epithelium; AUC, area under the curve Another currently-used measurement of the outer retina in SD-OCT images, reflectivity, was also acquired by directly measuring the peak greyscale value of each band in the OCT images [18]. The reflectivity value of each band was also adjusted for its percentage over the total value of the four bands.
## Statistical analysis
Statistical analyses were performed using SPSS version 21.0 (SPSS Inc., Chicago, IL, USA).
All the normally-distributed variables were expressed as mean ± standard deviation. BCVA was represented as the logarithm of the minimum angle of resolution (LogMAR). Shapiro–Wilk tests were performed for variable normality. The interocular variability of cone density and integral values was calculated using Mann–Whitney U test. T-tests were performed to compare the integral of each layer and AO results between sex groups, age groups, and the four meridians. After adjustment of age and gender, the correlation of the EZ and IZ integral values in healthy subjects with cone density was assessed respectively using Spearman correlation. A generalized estimating equation (GEE) model was enrolled to adjust the weight of each eye in the statistics considering the potential correlation between the two eyes. Stepwise multiple linear regression was used to analyze the correlation among cone density, BCVA, and integral values of EZ and IZ respectively in IRD patients after adjustment of age and gender. The correlations of integral and reflectivity with cone density were compared using the Z test. Correlation curves were plotted by the commercially available software GraphPad Prism 8 (GraphPad Software, San Diego, CA, USA). P values less than 0.05 were considered statistically significant.
## Results
Sixty-four eyes of 32 healthy subjects (9 males and 23 females aged 29.06 ± 4.51 years, range: 23–39 years) and 10 eyes of 5 IRD patients (1 male and 4 females) were included. Each patient came from a different family. Among the healthy subjects, 18 belonged to the younger group (age < 30y) and 14 belonged to the elder group (age ≥ 30y). All the IRD patients were bilaterally involved, of which one was diagnosed with cone-rod dystrophy (CORD) and four were diagnosed with retinitis pigmentosa (RP). Characteristics of the healthy subjects and IRD patients are shown in Tables 1 and 2, respectively. Table 1Demographic data of the healthy subjectsAbbreviation: BCVA best corrected visual acuity, SD standard deviationTable 2Clinical and genetic characteristics of the 5 IRD patients**The diagnosis* was based on family history, typical symptoms, OCT and visual field findings. Abbreviation: AL axial length, RE rarefactive error, BCVA best corrected visual acuity, CORD cone-rod dystrophy, RP retinitis pigmentosa
## Distribution characteristics of adaptive optics result in healthy people
Cone density was symmetrical bilateral (Mann–Whitney U test, $$p \leq 0.84$$) and normally distributed at each eccentricity Table 3. No significant difference was found between males and females in each eccentricity. The inter-individual coefficient of variation was on average $17.4\%$. In both age groups, cone cells showed the same pattern of decline with increasing eccentricity. Fovea density was higher in younger than elder at the fovea (2°) and 6° (T-test, $$p \leq 0.014$$ and $$p \leq 0.006$$, respectively) but had no difference in other regions (T-test, $$p \leq 0.146$$, $$p \leq 0.432$$ and $$p \leq 0.095$$, from 3° to 5°, respectively) (Fig. 4).Table 3Cone density (/mm2) of the four meridians after manual adjustment and normality test in 64 healthy eyes from 2 to 6 degrees eccentricityAbbreviation: SD Standard deviationFig. 4Distribution characteristics of cone density in healthy subjects. A: Cone density as a function of eccentricity, decreasing from the fovea to the peripheral retina. B: Cone density comparison between younger subjects and elder subjects (group 1, younger group: 23–29 years; group 2, elder group: 30–39 years). Fovea density was higher in the younger subjects at 2° and 6 o eccentricity (T test, $$p \leq 0.014$$ and $$p \leq 0.006$$, respectively). C: Box plots presenting cone density distribution in the four meridians. Lower cone density was observed in the vertical (superior + inferior) meridian ($p \leq 0.05$) To evaluate the distribution of the cones in different directions, the cone density on the four meridians in both eyes was accumulated and compared using paired T-test. No statistically significant difference in cone density was observed between the nasal and temporal meridians ($$p \leq 0.78$$) or the superior and inferior meridians ($$p \leq 0.99$$). Further comparison between the vertical and horizontal directions indicated an overall lower cone density on the vertical meridian than on the horizontal meridian ($p \leq 0.05$) (Fig. 4).
## Macular cone density in patients with inherited retinal diseases
Figure 5 shows the cone cell morphology and corresponding OCT images in the right eye of the 5 IRD patients. All the patients had ELM preserved with EZ and IZ interruption in the patients with RP, IZ loss and blurred EZ in the patient with CORD. ERG cone responses showed varying degrees of significantly diminished amplitudes and delayed implicit time in all the patients with poor visual acuity. Case 1 and case 2 had significant bilateral visual impairment and loss of typical cone cell mosaic, presenting a decreased density from the foveal to the perifoveal. Case 3 and case 4 had normal BCVA bilaterally with continuous EZ and IZ in the central fovea on OCT images; the AO images showed a typical decreasing cone cell density pattern from the fovea to the peripheral retina. Cone cell density in case 3 was within the normal range at 2° and 3° eccentricity but lower than the normal lower limit at 4° to 6° eccentricity, as shown in Table 3. In case 4, cone density was within the $95\%$ confidential intervals ($95\%$ Cis) of normal subjects at all eccentricities closer to the lower limit of the normal range as it progressed to the periphery. All the RP patients showed the progression of photoreceptor degeneration from peripheral to central macula in AO images. The fovea-centered decline of cone density was observed in the patient with CORD (case 5) while the cell mosaic was close to normal in the peripheral macula. Fig. 5Cone morphology in the temporal meridian in IRD subjects. A: Cone morphology and OCT images in IRD subjects and healthy control. Scale bar, AO image: 20 μm; OCT image: 400 μm. B/C: Comparison of cone density at each eccentricity between cases and healthy control. Healthy control was set as the healthy subjects of the corresponding age group. $95\%$ limits of agreement (LOA) were calculated for comparisons, shown as the dashed lines paralleled with the control group
## Distribution characteristics of integral values in healthy people
Integrals in healthy people had mostly consistent distribution characteristics with those of cone density. The average integrals of EZ and IZ were normally distributed (Table 4). Integrals were not significantly different bilaterally (Mann–Whitney U test, $$p \leq 0.97$$ and $$p \leq 0.053$$, respectively). Gender did not lead to distinctive differences in integral values at the parafoveal and perifoveal (T test, EZ: $$p \leq 0.347$$, $$p \leq 0.092$$, $$p \leq 0.205$$, $$p \leq 0.643$$, IZ: $$p \leq 0.074$$, $$p \leq 0.310$$, $$p \leq 0.091$$ and $$p \leq 0.121$$, from 3 to 6 o respectively). However, the IZ integral showed gender difference at 2 degrees (T test, $$p \leq 0.023$$), where the EZ integral was of no difference (T test, $$p \leq 0.141$$). The EZ and IZ integrals between the two age groups had almost no difference, and the difference was evident only in the EZ integral at the fovea (2°) (T test, $$p \leq 0.033$$). For the comparison of integrals among meridians, the difference was observed only for the EZ integral between superior and inferior (T test, $$p \leq 0.014$$) Table 5.Table 4EZ and IZ integrals and normality test in healthy subjects from 2 to 6 degrees eccentricitya Shapiro-Wilk test. Abbreviation: EZ Ellipsoid zone, IZ Interdigitation zone, SD Standard deviationTable 5Comparison of cone density (/mm2) for different parameters (laterality, sex, age and meridian)*$p \leq 0.05.$ Abbreviation: EZ Ellipsoid zone, IZ Interdigitation zone
## Relationships among integral values, cone density and BCVA
After correcting for meridian effects with multiple linear regression, cone cell density was highly corrected with IZ integral ($r = 0.968$, $p \leq 0.001$) and IZ reflectivity ($r = 0.960$, $p \leq 0.001$). The correlation between cone cell density and IZ integral was significantly higher than that with IZ reflectivity ($z = 7.5763$, $p \leq 0.001$). Cone cell density was negatively correlated with EZ integral (r = -0.616, $p \leq 0.05$) but was not statistically correlated with EZ reflectivity. ( Fig. 6).Fig. 6Relationship between the integral, reflectivity and cone density in patients and healthy subjects. Relationship between the IZ integral (A), IZ reflectivity (B) and cone density(/mm2) in eyes with inherited retinal diseases ($r = 0.823$, $p \leq 0.001$; $r = 0.789$, $p \leq 0.001$, respectively), and between the EZ integral (C), IZ integral (D), IZ reflectivity (E) and cone density(/mm2) in healthy subjects (r = -0.616, $p \leq 0.05$; $r = 0.968$, $p \leq 0.001$; $r = 0.960$, $p \leq 0.001$, respectively) Abbreviation EZ, ellipsoid zone; IZ, interdigitation zone.
Thirty-seven regions of interest that could be counted were selected from images of IRD patients for correlation analysis. After correction using the GEE model for correlation between the two eyes, cone density was found to be significantly correlated with the IZ integral ($r = 0.823$, $p \leq 0.001$) and IZ reflectivity ($r = 0.789$, $p \leq 0.001$) (Fig. 6). No significant difference was found between the two correlations ($z = 0.8770$, $$p \leq 0.3805$$). A strong correlation was found between BCVA and cone density at 2° eccentricity (r = -0.857, $$p \leq 0.002$$), and between BCVA and the IZ integral at the foveola (r = -0.746, $$p \leq 0.013$$) and the fovea (2°) (r = -0.822, $$p \leq 0.004$$) (Fig. 7). No significant correlation was found between BCVA and reflectivity, either in EZ or IZ.Fig. 7Relationships among integral values, cone density and BCVA. A: Relationship between BCVA(LogMAR) and cone density(/mm2) (r = -0.857, $$p \leq 0.002$$). B: Relationship between BCVA(LogMAR) and IZ integrals at the foveola (left) (r = -0.746, $$p \leq 0.013$$) and at the fovea (right) (r = -0.822, $$p \leq 0.004$$). Abbreviation: BCVA, best corrected visual acuity
## Discussion
In this article, we investigated the characteristics of a newly proposed parameter “integral” and cone density distribution in healthy subjects and several patients with IRD. Similar distribution characteristics were observed between the EZ and IZ integrals and cone packing density. The integrals of EZ and IZ both had close relationships with cone density, especially for the correlation between IZ integral and cone density, which indicated that the IZ integral had the potential to reflect cone cell density. In addition, we found that loss of cone cells emerged before the decline of BCVA in IRD patients.
We observed a significant variation in cone density between different subjects, and the average coefficient of variation reached $17\%$. Previous works of literature reported inter-individual variation between 11 and $20\%$ with the variation highest at the fovea [24–27]. The main difference in cone packing density compared to previous studies was in the cone density at 2° eccentricity, of which the possible explanation might lie in the differences between ethnicity, age composition, subject number, ocular dominance, AO devices, and measurement method (e.g., algorithms, manual adjustment, or the ROI size). Besides, we observed a denser cone density in the horizontal meridian in healthy subjects, which corresponded to the elliptical isodensity contours in histological studies, referred to as horizontal cone streaks [27]. The horizontal cone streaks were observed in several studies using AO, whether based on scanning laser ophthalmology (SLO) or FIO [28, 29]. The effect of age on cone cell density was found mainly at the fovea. The result was similar to Song et al. measuring cone density within 0.5 mm from the center of the fovea using AOSLO [30] and Legras et al. at the same 2° eccentricity using RTX1 [24].
The EZ and IZ integrals shared similar distribution characteristics with cone packing density. The EZ and IZ integral differences were observed between the two age groups, indicating that integral also tends to change with age as cone density. The main difference in our results lay in the horizontal meridian and the vertical meridian. Unlike cone density, neither EZ nor IZ integral difference was observed between the horizontal and the vertical meridian. In OCT images, the reflections of EZ and IZ together depend on cone cells’ structure, of which only a single analysis of EZ or IZ cannot fully reflect. We will further synthetically analyze the combined effects of EZ and IZ in future studies.
The comparability was based on a similar imaging principle: the light reflection from a specific structure. Cone recognition in the flood illumination adaptive optics applied in our study is based on an axial boundary of the cone outer segment using coherent interference of two reflections [31], while the main light scattering organelle is mitochondria in the cone cell inner segment on OCT [32]. Previous articles have reported the correlation between OCT reflectivity and cone density [18, 19]. Based on the existing literature, we further included IRD patients, expanded the imaging range, and improved the reflectivity measurement using the integration algorithm. Results showed that the outer retinal lesion in IRD begins with photoreceptor cell degeneration, so the suggestive role of integral is more meaningful early in the lesion. Besides, AO-FIO allows imaging of all healthy cone cells in healthy subjects. In patients, however, shortened cone cells with IZ disruption may not be detected by AO-FIO, resulting in fewer detected cone cells than the actual cone number. Therefore, we measured the correlation between integrals and cone density in healthy individuals and patients separately. We also had an extensive measurement range with 6 o eccentricity in each meridian.
Several inherited retinal diseases, including retinitis pigmentosa, cone or cone-rod dystrophy, and Stargardt disease, are characterized by progressive disruption of EZ and IZ on OCT. OCT changes can occur even in patients with normal fundus photography or undamaged visual acuity. The IZ’s visible disruption precedes the EZ’s disruption on OCT images [33]. Therefore, early quantification of the outer retina on OCT, especially of the IZ layer in IRD patients, is of great value for early diagnosis and prognostic prediction. Previous OCT quantification methods focused mainly on the thickness or reflectivity of a particular layer. With the introduction of adaptive optics, a cone density decrease was found where the EZ and IZ bands remained continuous on OCT [34, 35]. A previous study has found a better connection between cone density and reflectivity than between cone density and EZ thickness [18]. Thus, calculation concerning reflectivity was a more accurate method than thickness measurement with the additional advantage of reduced errors in manual or automatic segmentation. Our parameter “integral” differed from reflectivity as it calculated the whole hyperreflective substances in a certain layer, taking into account the entire thickness of the layer rather than the peak reflectivity in the layer as in previous studies. In this article limiting patients to those with IRD, we found that the IZ integral was more suggestive of cone cell density than the IZ reflectivity in healthy subjects. In our results, the IZ integral was also significantly correlated with BCVA and cone density at the fovea. All the results suggested that the parameter “integral” measured the cumulative spatial effect of reflectivity and thus served as an improved indicator of the outer retinal structure, which could be a biomarker of cone disruption in IRD patients at early stages.
There were several limitations in our study. We included a healthy population aged 20 to 40 years, which did not match the age of the majority of our patients. The number of patients we included was small and further experiments were needed for more genotype-specific patients to analyze the phenotype-genotype correlation. Furthermore, cone cells with IZ disruption could not be detected by AO-FIO, which might further require split-detection adaptive optics, a non-confocal AOSLO, to detect these cone cells by simultaneously recording signals from the photoreceptor inner segment [35, 36]. Besides, good fixation was challenging for patients with poor visual conditions. Despite the good cooperativeness of our included subjects to reduce motion artifacts, the repeatability of AO imaging was still affected by the narrow imaging range and the asthenopia associated with prolonged gazing at the light spot. In addition, the slope of the peripheral retina affected the Stiles-Crawford properties of the photoreceptors in AO imaging and also affected the integral calculation in OCT [37]. We calculated the average of multiple regions to minimize its effect. Finally, the study was a preliminary research and prospective studies were needed to verify the feasibility of our method.
In conclusion, the parameter “integral” measured in OCT images could be a feasible estimator of cone density if the AO devices were unavailable. EZ and IZ Integrals have the potential to be applied to early detection, function prediction, and longitudinal follow-up of more photoreceptor-involved diseases.
## References
1. Chung H, Hwang JU, Kim JG, Yoon YH. **Optical coherence tomography in the diagnosis and monitoring of cystoid macular edema in patients with retinitis pigmentosa**. *Retina* (2006.0) **26** 922-927. DOI: 10.1097/01.iae.0000250008.83779.23
2. Lima LH, Sallum JM, Spaide RF. **Outer retina analysis by optical coherence tomography in cone-rod dystrophy patients**. *Retina* (2013.0) **33** 1877-1880. DOI: 10.1097/IAE.0b013e31829234e6
3. Staurenghi G, Sadda S, Chakravarthy U, Spaide RF. **International Nomenclature for Optical Coherence Tomography P. Proposed lexicon for anatomic landmarks in normal posterior segment spectral-domain optical coherence tomography: the IN*OCT consensus**. *Ophthalmology* (2014.0) **121** 1572-8. DOI: 10.1016/j.ophtha.2014.02.023
4. Spaide RF, Curcio CA. **ANATOMICAL CORRELATES TO THE BANDS SEEN IN THE OUTER RETINA BY OPTICAL COHERENCE TOMOGRAPHY Literature Review and Model**. *Retina-J Ret Vit Dis* (2011.0) **31** 1609-1619
5. Arepalli S, Traboulsi EI, Ehlers JP. **Ellipsoid Zone Mapping and Outer Retinal Assessment in Stargardt Disease**. *Retina* (2018.0) **38** 1427-1431. DOI: 10.1097/IAE.0000000000001716
6. 6.Augstburger E, Ores R, Mohand-Said S, Mrejen S, Keilani C, Antonio A, et al. Outer Retinal Alterations Associated With Visual Outcomes in Best Vitelliform Macular Dystrophy. Am J Ophthalmol. 2019;208:429–37.
7. 7.Kong XR, Ho A, Munoz B, West S, Strauss RW, Jha A, et al. Reproducibility of Measurements of Retinal Structural Parameters Using Optical Coherence Tomography in Stargardt Disease. Transl Vis Sci Techn. 2019;8(3):46.
8. Scarinci F, Fawzi AA, Shaarawy A, Simonett JM, Jampol LM. **Longitudinal Quantitative Evaluation of Outer Retinal Lesions in Acute Posterior Multifocal Placoid Pigment Epitheliopathy Using Optical Coherence Tomography**. *Retina-J Ret Vit Dis* (2017.0) **37** 851-857
9. Narala R, Scarinci F, Shaarawy A, Simonett JM, Flaxel CJ, Fawzi AA. **Longitudinal Quantitative Evaluation of Photoreceptor Volume Following Repair of Macula-Off Retinal Detachment**. *Retina* (2016.0) **36** 1432-1438. DOI: 10.1097/IAE.0000000000000971
10. Oh J, Smiddy WE, Flynn HW, Gregori G, Lujan B. **Photoreceptor inner/outer segment defect imaging by spectral domain OCT and visual prognosis after macular hole surgery**. *Invest Ophthalmol Vis Sci* (2010.0) **51** 1651-1658. DOI: 10.1167/iovs.09-4420
11. Maheshwary AS, Oster SF, Yuson RMS, Cheng LY, Mojana F, Freeman WR. **The Association Between Percent Disruption of the Photoreceptor Inner Segment-Outer Segment Junction and Visual Acuity in Diabetic Macular Edema**. *Am J Ophthalmol* (2010.0) **150** 63-67. DOI: 10.1016/j.ajo.2010.01.039
12. 12.Gong YT, Chen LJ, Pang CP, Chen HY. Ellipsoid zone optical intensity reduction as an early biomarker for retinitis pigmentosa. Acta Ophthalmol. 2021;99(2):e215–21.
13. 13.Arrigo A, Grazioli A, Romano F, Aragona E, Bordato A, di Nunzio C, et al. Choroidal Patterns in Stargardt Disease: Correlations with Visual Acuity and Disease Progression. J Clin Med. 2019;8(9):1388.
14. Parodi MB, Cicinelli MV, Iacono P, Bolognesi G, Bandello F. **Multimodal imaging of foveal cavitation in retinal dystrophies**. *Graef Arch Clin Exp* (2017.0) **255** 271-279. DOI: 10.1007/s00417-016-3450-7
15. Liang JZ, Williams DR, Miller DT. **Supernormal vision and high-resolution retinal imaging through adaptive optics**. *J Opt Soc Am A* (1997.0) **14** 2884-2892. DOI: 10.1364/JOSAA.14.002884
16. Litts KM, Cooper RF, Duncan JL, Carroll J. **Photoreceptor-Based Biomarkers in AOSLO Retinal Imaging**. *Invest Ophth Vis Sci* (2017.0) **58** 255-267. DOI: 10.1167/iovs.17-21868
17. Song H, Rossi EA, Latchney L, Bessette A, Stone E, Hunter JJ. **Cone and rod loss in Stargardt disease revealed by adaptive optics scanning light ophthalmoscopy**. *JAMA Ophthalmol* (2015.0) **133** 1198-1203. DOI: 10.1001/jamaophthalmol.2015.2443
18. Flores M, Debellemaniere G, Bully A, Meillat M, Tumahai P, Delbosc B. **Reflectivity of the Outer Retina on Spectral-Domain Optical Coherence Tomography as a Predictor of Photoreceptor Cone Density**. *Am J Ophthalmol* (2015.0) **160** 588-595. DOI: 10.1016/j.ajo.2015.06.008
19. Saleh M, Flores M, Gauthier AS, Elphege E, Delbosc B. **Quantitative analysis of photoreceptor layer reflectivity on en-face optical coherence tomography as an estimator of cone density**. *Graef Arch Clin Exp* (2017.0) **255** 2119-2126. DOI: 10.1007/s00417-017-3761-3
20. Guyon B, Elphege E, Flores M, Gauthier AS, Delbosc B, Saleh M. **Retinal Reflectivity Measurement for Cone Impairment Estimation and Visual Assessment After Diabetic Macular Edema Resolution (RECOVER-DME)**. *Invest Ophthalmol Vis Sci* (2017.0) **58** 6241-6247. DOI: 10.1167/iovs.17-22380
21. 21.Gu J, Jiang T, Yu M, Yu J, Li W, Liu S, et al. A Novel Approach to Quantitative Evaluation of Outer Retinal Lesions Via a New Parameter “Integral” in Spectral Domain Optical Coherence Tomography. Transl Vis Sci Technol. 2020;9(12):8.
22. Gao FJ, Li JK, Chen H, Hu FY, Zhang SH, Qi YH. **Genetic and Clinical Findings in a Large Cohort of Chinese Patients with Suspected Retinitis Pigmentosa**. *Ophthalmology* (2019.0) **126** 1549-1556. DOI: 10.1016/j.ophtha.2019.04.038
23. Muthiah MN, Gias C, Chen FK, Zhong J, McClelland Z, Sallo FB. **Cone photoreceptor definition on adaptive optics retinal imaging**. *Br J Ophthalmol* (2014.0) **98** 1073-1079. DOI: 10.1136/bjophthalmol-2013-304615
24. Legras R, Gaudric A, Woog K. **Distribution of cone density, spacing and arrangement in adult healthy retinas with adaptive optics flood illumination**. *PLoS ONE* (2018.0) **13** e0191141. DOI: 10.1371/journal.pone.0191141
25. Lombardo M, Serrao S, Ducoli P, Lombardo G. **Eccentricity dependent changes of density, spacing and packing arrangement of parafoveal cones**. *Ophthalmic Physiol Opt* (2013.0) **33** 516-526. DOI: 10.1111/opo.12053
26. Lombardo M, Lombardo G, Schiano Lomoriello D, Ducoli P, Stirpe M, Serrao S. **Interocular symmetry of parafoveal photoreceptor cone density distribution**. *Retina* (2013.0) **33** 1640-1649. DOI: 10.1097/IAE.0b013e3182807642
27. Curcio CA, Sloan KR, Kalina RE, Hendrickson AE. **Human photoreceptor topography**. *J Comp Neurol* (1990.0) **292** 497-523. DOI: 10.1002/cne.902920402
28. Chui TY, Song H, Burns SA. **Individual variations in human cone photoreceptor packing density: variations with refractive error**. *Invest Ophthalmol Vis Sci* (2008.0) **49** 4679-4687. DOI: 10.1167/iovs.08-2135
29. Dabir S, Mangalesh S, Kumar KA, Kummelil MK, Sinha Roy A, Shetty R. **Variations in the cone packing density with eccentricity in emmetropes**. *Eye (Lond)* (2014.0) **28** 1488-1493. DOI: 10.1038/eye.2014.229
30. Song H, Chui TY, Zhong Z, Elsner AE, Burns SA. **Variation of cone photoreceptor packing density with retinal eccentricity and age**. *Invest Ophthalmol Vis Sci* (2011.0) **52** 7376-7384. DOI: 10.1167/iovs.11-7199
31. Jonnal RS, Rha J, Zhang Y, Cense B, Gao W, Miller DT. **In vivo functional imaging of human cone photoreceptors**. *Opt Express* (2007.0) **15** 16141-16160. DOI: 10.1364/OE.15.016141
32. Litts KM, Zhang YH, Freund KB, Curcio CA. **Optical Coherence Tomography and Histology of Age-Related Macular Degeneration Support Mitochondria as Reflectivity Sources**. *Retina-J Ret Vit Dis* (2018.0) **38** 445-461
33. 33.Mitamura Y, Mitamura-Aizawa S, Katome T, Naito T, Hagiwara A, Kumagai K, et al. Photoreceptor Impairment and Restoration on Optical Coherence Tomographic Image. J Ophthalmol. 2013;2013:518170.
34. Nakatake S, Murakami Y, Funatsu J, Koyanagi Y, Akiyama M, Momozawa Y. **Early detection of cone photoreceptor cell loss in retinitis pigmentosa using adaptive optics scanning laser ophthalmoscopy**. *Graef Arch Clin Exp* (2019.0) **257** 1169-1181. DOI: 10.1007/s00417-019-04307-0
35. Sun LW, Johnson RD, Langlo CS, Cooper RF, Razeen MM, Russillo MC. **Assessing Photoreceptor Structure in Retinitis Pigmentosa and Usher Syndrome**. *Invest Ophthalmol Vis Sci* (2016.0) **57** 2428-2442. DOI: 10.1167/iovs.15-18246
36. Razeen MM, Cooper RF, Langlo CS, Goldberg MR, Wilk MA, Han DP. **Correlating Photoreceptor Mosaic Structure to Clinical Findings in Stargardt Disease**. *Transl Vis Sci Technol* (2016.0) **5** 6. DOI: 10.1167/tvst.5.2.6
37. Mariotti L, Devaney N, Lombardo G, Lombardo M. **Understanding the changes of cone reflectance in adaptive optics flood illumination retinal images over three years**. *Biomed Opt Express* (2016.0) **7** 2807-2822. DOI: 10.1364/BOE.7.002807
|
---
title: The association of asthma duration with body mass index and Weight-Adjusted-Waist
index in a nationwide study of the U.S. adults
authors:
- Xiaoxiao Han
- Xiaofang He
- Gui Hao
- Lifang Cao
- Yinliang Qi
- Kexing Han
journal: European Journal of Medical Research
year: 2023
pmcid: PMC10012562
doi: 10.1186/s40001-023-01089-4
license: CC BY 4.0
---
# The association of asthma duration with body mass index and Weight-Adjusted-Waist index in a nationwide study of the U.S. adults
## Abstract
### Backgrounds
The association between obesity and asthma has been of interest, but whether the duration of asthma has an effect on obesity is still limitedly studied.
### Aim
The purpose of this study was to investigate the association between asthma duration and obesity-related indexes, where obesity-related indexes include Body mass index (BMI) and Weight-adjusted-waist index (WWI).
### Methods
Data from National Health and Nutrition Examination Survey (NHANES) 2009–2018 were obtained to conduct this cross-sectional study. Duration of asthma was used as the independent variable and obesity-related indexes as the response variables. Multiple linear regression was used to assess the association between the independent variable and the response variables, and subsequently smoothed curve fitting and threshold effect analysis were performed to clarify whether there was a nonlinear correlation between the independent variable and the response variables. Finally, subgroup analysis was conducted to find sensitive populations.
### Results
A total of 9170 participants were included in the analysis. Asthma duration was statistically different between the two groups when all participants were grouped by median WWI (Q1 < 11.65, Q2 ≥ 11.65) ($P \leq 0.001$), but not by median BMI (Q1 < 31.8, Q2 ≥ 31.8) ($$P \leq 0.130$$). There was a positive association between asthma duration and WWI [β = 0.016, $95\%$ CI (0.016, 0.017)], but a negative one with BMI [β = − 0.098, $95\%$ CI (− 0.112, − 0.085)], and the correlations between the independent and response variables became more pronounced with increasing asthma duration (P for trend < 0.01). In addition, there were nonlinear relationships between asthma duration with BMI and WWI (log likelihood ratio < 0.001), with the best valid inflection points for asthma duration being 2 years (with WWI as the response variable) and 3 years (with BMI as the response variable), respectively. In the subgroup analysis, the positive association between asthma duration and WWI was more pronounced in the participants who were male, aged less than 40 years, and had asthma onset before 12 years of age. In contrast, when BMI was used as the response variable, the negative association between it and asthma duration was more pronounced among participants of female, aged 60 years or older, and with asthma onset less than 12 years of age.
### Conclusions
In US adults, asthma duration might cause changes in obesity-related indexes. Longer asthma duration might cause weight loss, but might increase the risk of abdominal obesity.
## Introduction
Asthma is a common chronic disease in the respiratory system, but the prevention and control of asthma is complicated by its numerous causes and the absence of a single biomarker that can be identified in time [1]. As the understanding of asthma has gradually increased, the association between obesity and asthma has attracted the interest of researchers [2, 3]. Although a study showed no significant positive association between obesity and the risk of readmission in asthma patients [4], most researchers have concluded that obesity increases the risk of suffering from asthma [5–8], especially since previous studies recommended 10 traits (including obesity) which could predict future asthma attacks and demonstrated a $13\%$ increase in the risk of asthma attacks with each additional trait [9, 10]. In addition, some studies also indicated that obesity was strongly associated with more severe asthma and poorer lung function [11, 12].
In contrast to previous views, researchers have suggested a bidirectional correlation between obesity and asthma [13]. In addition to the presence or absence of asthma, the duration of asthma is a notable presence in the management of asthma. Previous studies demonstrated that although asthma duration would not affect the severity of asthma [14], it could influence eventual pulmonary function, such as higher diffusing capacity and lung hyperinflation [15]. In addition, it has been suggested that asthma duration of less than 10 years might be a beneficial factor in the therapeutic effect of patients to IL-5 biologics [16]. It could be shown that, in addition to the presence or absence of asthma, the duration of asthma has a subtle effect on the organism. However, whether asthma duration could increase the burden of obesity remains rarely reported. In a limited number of small sample cross-sectional studies, Holguin F reported a positive association between asthma duration and BMI in patients with early-onset asthma (< 12 years) [17], but Ahmadiafshar A did not reach similar conclusions in their study [18]. Therefore, whether asthma duration would cause changes in obesity-related indexes is still not precisely described.
Previous studies have used body mass index (BMI) as a measure of obesity [19, 20], but the accuracy of BMI has been called into question in recent years [21, 22]. More researchers believe that BMI is more appropriate as a crude estimate of obesity [23]. Asthma is characterized by the presence of persistent chronic airway inflammation, the vast majority of asthmatic patients experience the use of glucocorticoids [1], and the most typical side effect of steroid medications is the development of central obesity [24]. Previous study demonstrated that body fat had a negative effect on human pulmonary function and that this negative effect was inevitable for non-obese populations [25]. As researchers have gained a better understanding of obesity, recent studies have shown that central obesity responds to a more realistic body fat situation [26] and that central (or androgenic) obesity, which is primarily a response to visceral fat distribution, was associated with asthma and impaired pulmonary function in adolescents and adults compared to peripheral obesity [27]. To better reflect the true situation of obesity, some researchers first proposed a new obesity index and named it the weight-adjusted waist index (WWI) [28]. Due to the adjustment for body weight, in the adult population this index mainly reflects weight-independent central obesity [28] and has been shown by many researchers to have better accuracy compared to BMI [29–32]. However, the association between asthma duration and WWI has still not been reported.
Therefore, the aim of this study was to investigate the association between asthma duration and obesity-related indexes with data from the National Health and Nutrition Examination Survey (NHANES), where obesity-related indexes include BMI, an indicator of traditional evaluation of obesity, and WWI, an indicator of central obesity.
## Data source
The National Health and Nutrition Examination Survey (NHANES), established in the 1990s, is a public service survey of the entire United States population in which demographic, dietary, physical, lifestyle, medical, and laboratory information is regularly collected to assess the health and nutritional status of the nation's population. The data in NHANES are currently updated every 2 years, and additional information may be added each time the data is updated. The NHANES survey protocol was approved by the National Center for Health Statistics (NCHS) Ethics Review Board, and written informed consent was provided to all participants. Because the NHANES data are open to the public, ethical review of this study was exempt.
## Participants
NHANES 2009–2018 included complete information on questionnaires related to asthma and information on covariates that needed to be adjusted for in the follow-up models was also complete. A total of 99,093 participants enrolled in NHANES 2009–2018, and we first excluded 77,090 participants who did not have asthma. Subsequently, we removed 2580 participants without clear information on age of asthma onset. Among the 19,423 participants with clear information on the age of asthma onset, we removed 5642 participants who no longer had asthma. At the same time, we removed 651 participants with missing weight information and 1081 participants with missing waist circumference information. Finally, we removed 2879 participants who were younger than 20 years of age. Finally, a total of 9170 participants were included in this study (Fig. 1).Fig. 1Flow chart for participants
## Definition of asthma duration
Information about the participants’ asthma status was obtained from a medical conditions questionnaire. Firstly, participants who responded positively to the question “Ever been told you have asthma?” were retained. Subsequently, participants who responded positively to the question “Still have asthma?” was reserved. Finally, participants that were able to give an exact answer to the question “Age when first had asthma?” were included in this study. Overall, asthma duration = age (years)−age at onset of asthma (years) and all enrolled participants in the study were still currently suffering from asthma.
## BMI and WWI
BMI and WWI were obtained from body measurement information in the physical examination module, and body measurements were collected by trained health technicians at mobile examination centers (MECs). BMI was calculated as weight in kilograms divided by the square of height in meters, and WWI was calculated as waist circumference in centimeters divided by the square root of weight in kilograms [33]. Body measurement information from NHANES 2009–2018 were subjected to review and the measurement methods remained consistent over this time period.
## Covariates
To better estimate the association between asthma duration and obesity-related indices, we included a number of factors with potential effects on asthma duration and/or obesity-related indices as covariates in the subsequent models, based on previous studies [17, 18, 34, 35]. Demographic information included participants' age, race (black, white, and other races), education level (less than high school, high school, and more than high school), and ratio of family income to poverty (PIR). Dietary information was extracted from a 24-h dietary questionnaire, which included participant self-reported information on total nutrients, and the mean of the sum of the participant-specific nutrient intakes for the 1st and 2nd 24-h periods was ultimately included in this study. Dietary information included in this study included energy (kcal), sugar (gm), fat (gm), cholesterol (mg), and moisture (gm). According to NHANES official instructions, the “moisture” included in this study refers not only to drinking water, but to all moisture in the diet, including food and beverages. Participants who answered yes to the question “Smoked at least 100 cigarettes in life?” were identified as smokers and were classified according to their current smoking status (now, ever, never) with their responses to the question “Do you now smoke cigarettes?”. Alcohol consumption was defined as “more than 12 drinks in the past 1 year”. Diabetes and hypertension were defined by the participants’ responses to the questions “Doctor told you have diabetes?” and “Ever told you had high blood pressure?”. Due to limitations of the NHANES data, we were unable to capture the specific types of medications used by participants, we obtained information on the participants’ prescription medication use, and a positive response to the question “Taken prescription medicine past month?” was identified as prescription medication use. A “prescription medication” in this context would be any prescription medication used by the participant within the past month.
## Statistical analysis
Analysis of all data was completed in R (http://www.R-project.org) and EmpowerStats (http://www.empowerstats.com). Because both independent and response variables were continuous variables in this study, we grouped participants based on median WWI and BMI as WWIQ1 (< 11.65), WWIQ2 (≥ 11.65), BMIQ1 (< 31.8), and BMIQ2 (≥ 31.8), respectively. For comparisons between groups of information on general characteristics of participants, continuous variables were expressed as mean ± standard deviation, and rates or percentages were used to describe categorical variables. Since covariates might have missing values, the missing values of continuous variables were interpolated using the mean of the variable if the percentage of missing values was less than $10\%$ of the total sample size, otherwise we grouped the continuous variables according to the interquartile distribution and set the missing values as “Unclear”. The missing values of the categorical variables were grouped separately and named “Unclear”. The association between asthma duration with WWI and BMI was assessed separately using multiple linear regression with asthma duration as the independent variable. We used β to respond the effect values between the independent variable and the respondent variable, which means that for each unit increase in the independent variable the respondent variable will increase by β unit values, and we also calculated the $95\%$ confidence interval ($95\%$ CI) for each effect value. Three models were generated by adjusting for different covariates, Model 1 (no covariates adjusted), Model 2 (age, gender, race, and asthma onset age were adjusted), and Model 3 (all covariates were adjusted). Participants were divided into four groups based on the quartile distribution of asthma duration to assess whether there was a trend change (P for trend) in the correlations between the independent variable and obesity-related indexes (WWI and BMI) as the asthma duration increased. Smoothing curves were used to assess whether there were non-linear relationships between the effects of asthma duration and obesity indexes, and threshold effect analysis was used to determine the best valid inflection points for the independent variable. A log-likelihood ratio (LLR) of less than 0.05 was used in the threshold effects analysis as the basis for adopting the nonlinear model. Subsequently, a subgroup analysis of the association between the independent variable and the response variables was performed to find sensitive individuals.
## Characteristics of participants
A total of 9170 participants were eventually enrolled in this study, of whom 3,121 were male and 6049 were female. The mean age of all participants was 56.43 ± 15.51 years with a mean duration of asthma was 24.84 ± 18.78 years. The mean WWI was 11.61 ± 0.83 and BMI was 33.19 ± 8.78 kg/m2.
When grouped by median WWI (11.65), asthma duration was statistically different between the two groups ($P \leq 0.001$). However, when grouped by median BMI (31.8 kg/m2), no significant statistical difference in asthma duration was seen between the two groups ($$P \leq 0.130$$). In addition, there were statistically significant differences in age, gender, and asthma onset age of participants between groups regardless of the median obesity index used for grouping ($P \leq 0.001$). The results of the baseline characteristics of the participants were displayed in Tables 1 and 2.Table 1Characteristics of participants grouped by median WWICharacteristicsWWIQ1($$n = 4589$$)WWIQ2($$n = 4581$$)P-valueGender (%) < 0.001 Male40.8427.22 Female59.1672.78Age (years)52.62 ± 16.2660.24 ± 13.69 < 0.001Stratified by age (years) (%) < 0.001 20–3924.438.12 40–5937.9632.90 60–8037.6158.98Asthma onset age (years)23.89 ± 18.0025.79 ± 19.48 < 0.001Stratified by asthma onset age (years) (%) < 0.001 < 1229.9221.48 ≥ 1270.0878.52Asthma duration (years)23.89 ± 18.0025.79 ± 19.48 < 0.001Quartiles of asthma duration (years) (%) < 0.001 Q1 (0–9)25.0224.10 Q2 (10–19)22.7922.75 Q3 (20–38)28.5025.63 Q4 (39–79)23.6927.53Race (%) < 0.001 White46.5548.66 Black28.9419.91 Other races24.5231.43Education level (%) < 0.001 < High school22.9929.71 High school24.0623.34 > High school52.9146.82 Unclear0.040.13PIR2.19 ± 1.531.87 ± 1.35 < 0.001Stratified by PIR (%) < 0.001 < 1.3540.5544.97 1.35–3.4536.3339.75 > 3.4523.1215.28Smoking (%)0.532 Now13.8813.67 Ever25.0425.15 Never40.0138.94 Unclear21.0722.24Alcohol (%) < 0.001 Yes63.0456.01 No29.8137.81 Unclear7.156.18Physical activity intensity (%)0.187 Light26.3227.31 Moderate22.7523.99 Vigorous33.3431.59 Unclear17.5917.11Hypertension (%) < 0.001 Yes56.3574.50 No43.6525.13 Unclear0.000.37Diabetes (%) < 0.001 Yes21.8649.79 Borderline5.953.21 No72.1146.82 Unclear0.090.17Prescription medications (%)0.830 Yes71.3071.47 No28.6828.49 Unclear0.020.04Energy (kcal) (%) < 0.001 3.11–1565.425.2331.26 1565.5–2043.426.2830.30 2043.5–7878.531.4925.13 Unclear17.0013.32Sugars (gm) (%) < 0.001 3.79–69.6425.5631.02 69.65–113.9826.5929.95 113.99–688.7930.8625.71 Unclear17.0013.32Fat (gm) (%) < 0.001 3.88–54.2625.7430.84 54.40–80.7727.8528.51 80.85–279.4029.4227.33 Unclear17.0013.32Cholesterol (mg) < 0.001 0–164.528.4427.94 165.0–310.524.8031.89 311.0–1692.029.7726.85 Unclear17.0013.32Moisture (gm) (%) < 0.001 418.80–2127.3128.5028.03 2127.32–3130.0528.0528.33 3130.06–11,897.2026.4530.32 Unclear17.0013.32Mean ± SD for continuous variables: P-value was calculated by weighted linear regression model% for Categorical variables: P-value as calculated by weighted chi-square testTable 2Characteristics of participants grouped by median BMI (kg/m2)CharacteristicsBMIQ1($$n = 4560$$)BMIQ2($$n = 4610$$)P-valueGender (%) < 0.001 Male39.9128.22 Female60.0971.78Age (years)57.53 ± 16.6755.34 ± 14.19 < 0.001Stratified by age (years) (%) < 0.001 20–3917.5715.01 40–5930.0940.72 60–8052.3544.27Asthma onset age (years)32.83 ± 23.0430.36 ± 20.69 < 0.001Stratified by asthma onset age (years) (%)0.705 < 1225.8825.53 ≥ 1274.1274.47Asthma duration (years)24.70 ± 19.1224.97 ± 18.430.130Quartiles of asthma duration (years) (%)0.211 Q1 (0–9)24.7624.36 Q2 (10–19)23.2522.30 Q3 (20–38)26.1028.03 Q4 (39–79)25.9025.31Race (%) < 0.001 White50.8644.38 Black20.5528.26 Other races28.6027.35Education level (%)0.003 < High school28.0524.66 High school22.9824.40 > High school48.9050.82 Unclear0.070.11PIR2.19 ± 1.511.87 ± 1.37 < 0.001Stratified by PIR (%) < 0.001 < 1.3538.9746.51 1.35–3.4538.3337.74 > 3.4522.7015.75Smoking (%) < 0.001 Now12.8314.71 Ever27.3022.91 Never40.3738.59 Unclear19.5023.80Alcohol (%) < 0.001 Yes62.0457.05 No30.7036.88 Unclear7.266.07Physical activity intensity (%)0.107 Light27.4626.18 Moderate22.5024.23 Vigorous33.0731.87 Unclear16.9717.72Hypertension (%) < 0.001 Yes56.7873.97 No43.0325.86 Unclear0.200.17Diabetes (%) < 0.001 Yes23.4048.09 Borderline72.4846.62 No4.125.03 Unclear0.000.26Prescription medications (%)0.026 Yes72.5270.26 No27.4329.72 Unclear0.040.02Energy (kcal) (%) < 0.001 3.11–1565.428.7727.72 1565.5–2043.426.6229.93 2043.5–7878.526.7529.85 Unclear17.8512.49Sugars (gm) (%) < 0.001 3.79–69.6429.5027.09 69.65–113.9827.0629.46 113.99–688.7925.5930.95 Unclear17.8512.49Fat (gm) (%) < 0.001 3.88–54.2628.0528.52 54.40–80.7728.0728.29 80.85–279.4026.0330.69 Unclear17.8512.49Cholesterol (mg) (%) < 0.001 0–164.529.4726.92 165.0–310.526.1030.56 311.0–1692.026.5830.02 Unclear17.8512.49Moisture (gm) (%) < 0.001 418.80–2127.3131.1025.47 2127.32–3130.0526.7129.65 3130.06–11,897.2024.3432.39 Unclear17.8512.49Mean ± SD for continuous variables: P-value was calculated by weighted linear regression model% for Categorical variables: P-value as calculated by weighted chi-square test
## The association between the asthma duration and obesity indexes
In the all-adjusted model, there was a positive association between asthma duration and WWI [β = 0.016, $95\%$ CI (0.015, 0.017)]. Interestingly, the relationship between asthma duration and BMI was negative [β = − 0.098, $95\%$ CI (− 0.112, − 0.085)]. Furthermore, the correlations between asthma duration and both obesity indexes became more significant with increasing asthma duration (P for trend < 0.01) (Table 3, Fig. 2). A smoothed curve fitting analysis for the association between asthma duration and these two obesity-related indexes revealed a significant nonlinear correlation between the independent and the responding variables (Fig. 3). Subsequently, threshold effect analysis suggested that the positive correlation between asthma duration and WWI switched from [β = 0.09, $95\%$ CI (0.05, 0.14)] to [β = 0.02, $95\%$ CI (0.01, 0.02)] when asthma duration exceeded 2 years. And when the response variable was BMI, threshold effect analysis indicated a positive association between asthma duration and BMI when asthma duration did not exceed 3 years [β = 0.41, $95\%$ CI (0.11, 0.71)]. However, once the duration of asthma exceeded 3 years, the relationship between asthma duration and BMI turned negative [β = − 0.10, $95\%$ CI (− 0.12, − 0.09)] (Table 4).Table 3The association of asthma duration with WWI and BMI (kg/m2)OutcomesModel 1β, ($95\%$ CI)Model 2β, ($95\%$ CI)Model 3β, ($95\%$ CI)WWI0.003 (0.002, 0.004)0.021 (0.019, 0.022)0.016 (0.015, 0.017)Quartiles of asthma duration (years) Q1 (0–9)ReferenceReferenceReference Q2 (10–19)− 0.048 (− 0.097, 0.001)0.157 (0.110, 0.204)0.122 (0.078, 0.166) Q3 (20–38)− 0.126 (− 0.173, − 0.079)0.337 (0.286, 0.388)0.270 (0.222, 0.319) Q4 (39–79)0.125 (0.077, 0.172)0.868 (0.808, 0.928)0.641 (0.580, 0.702)P for trend < 0.01 < 0.01 < 0.01BMI (kg/m2)− 0.006 (− 0.016, 0.003)− 0.031 (− 0.045, − 0.018)− 0.098 (− 0.112, − 0.085)Quartiles of asthma duration (years) Q1 (0–9)ReferenceReferenceReference Q2 (10–19)− 0.994 (− 1.516, − 0.472)− 1.252 (− 1.786, − 0.719)− 1.626 (− 2.121, − 1.130) Q3 (20–38)− 0.434 (− 0.934, 0.067) 0.08929− 1.315 (− 1.896, − 0.735)− 2.160 (− 2.706, − 1.613) Q4 (39–79)− 0.501 (− 1.008, 0.006)− 1.794 (− 2.482, − 1.106)− 4.705 (− 5.393, − 4.017)P for trend < 0.01 < 0.01 < 0.01Model 1 = no covariates were adjusted. Model 2 = Model 1 + gender, race were adjusted. Model 3 = Model 2 + asthma onset age, education level, PIR, smoking, alcohol, physical activity intensity, hypertension, diabetes, prescription medications, energy, sugars, fat, cholesterol, moisture were adjustedFig. 2Trend test for the association between asthma duration and obesity-related indexes. a Trend test for the association between asthma duration and WWI. b Trend test for the association between asthma duration and BMI. * The green squares represents the effect value. The blue line represents the $95\%$ confidence interval of the effect value. * All the covariates were adjustedFig. 3The association between the asthma duration and obesity indexes. a The association between the asthma duration and WWI. b The association between the asthma duration and BMI. * Solid rad line represents the smooth curve fit between variables. Blue bands represent the $95\%$ of confidence interval from the fit. * All the covariates were adjustedTable 4Threshold effect analysis for the association of asthma duration with BMI (kg/m2) and WWIOutcomesBMI (kg/m2)WWILinear effect model β, ($95\%$CI)− 0.10 (− 0.11, − 0.08)0.02 (0.01, 0.02)Non-linear model Inflection point (K)32 β, ($95\%$CI) (< K)0.41 (0.11, 0.71)0.09 (0.05, 0.14) β, ($95\%$CI) (≥ K)− 0.10 (− 0.12, − 0.09)0.02 (0.01, 0.02) LLR < 0.001 < 0.001*Gender, race, asthma onset age, education level, PIR, smoking, alcohol, physical activity intensity, hypertension, diabetes, prescription medications, energy, sugars, fat, cholesterol, moisture were adjusted in the models. LLR: Log-likelihood ratio; K: Inflection point.
## Results of subgroup analysis
In order to identify the stability of the association between asthma duration and the two obesity indexes in this study and to find sensitive cohorts, we performed a subgroup analysis. When WWI was the dependent variable, the positive association between asthma duration and WWI was more pronounced among male [β = 0.022, $95\%$ CI (0.020, 0.024)], aged less than 40 years [β = 0.021, $95\%$ CI (0.013, 0.029)], and asthma onset at age less than 12 years [β = 0.017, $95\%$ CI (0.015, 0.019)] (Table 5). When BMI was taken as the response variable, the negative association between asthma duration and BMI was more pronounced in participants who were female [β = − 0.106, $95\%$ CI (− 0.124, − 0.087)], aged 60 years or older [β = − 0.268, $95\%$ CI (− 0.305, − 0.232)], and with asthma onset younger than 12 years [β = − 0.070, $95\%$ CI (− 0.092, − 0.048)] (Table 6).Table 5A subgroup analysis of the association between asthma duration and WWICharacteristicsModel 1β, ($95\%$ CI)Model 2β, ($95\%$ CI)Model 3β, ($95\%$ CI)Gender Male0.002 (0.001, 0.004)0.026 (0.024, 0.028)0.022 (0.020, 0.024) Female0.004 (0.003, 0.005)0.018 (0.016, 0.019)0.014 (0.012, 0.015)Stratified by age (years) 20–390.002 (− 0.003, 0.007)0.037 (0.030, 0.045)0.021 (0.013, 0.029) 40–59− 0.004 (− 0.006, − 0.002)0.012 (0.008, 0.017)− 0.001 (− 0.006, 0.003) 60–800.002 (0.001, 0.003)0.014 (0.011, 0.017)0.016 (0.013, 0.019)Race White0.005 (0.004, 0.006)0.023 (0.021, 0.025)0.018 (0.016, 0.019) Black0.000 (− 0.002, 0.002)0.012 (0.010, 0.015)0.003 (0.000, 0.006) Other races0.005 (0.003, 0.006)0.024 (0.022, 0.026)0.020 (0.018, 0.022)Stratified by asthma onset age (years) < 120.018 (0.010, 0.027)0.018 (0.010, 0.026)0.021 (0.013, 0.029) 12–39− 0.002 (− 0.005, 0.002)− 0.001 (− 0.004, 0.002)− 0.001 (− 0.004, 0.002) ≥ 400.005 (0.002, 0.009)0.007 (0.004, 0.010)0.009 (0.005, 0.012)Table 6A subgroup analysis of the association between asthma duration and BMI (kg/m2)CharacteristicsModel 1β, ($95\%$ CI)Model 2β, ($95\%$ CI)Model 3β, ($95\%$ CI)Gender Male− 0.001 (− 0.014, 0.012)− 0.012 (− 0.032, 0.008)− 0.077 (− 0.097, − 0.058) Female− 0.006 (− 0.019, 0.007)− 0.043 (− 0.061, − 0.025)− 0.106 (− 0.124, − 0.087)Stratified by age (years) 20–390.064 (0.012, 0.117)0.313 (0.225, 0.400)0.136 (0.054, 0.219) 40–59− 0.055 (− 0.073, − 0.036)− 0.069 (− 0.125, − 0.014)− 0.202 (− 0.255, − 0.150) 60–800.014 (0.003, 0.025)− 0.289 (− 0.326, − 0.251)− 0.268 (− 0.305, − 0.232)Race White− 0.017 (− 0.031, − 0.004)− 0.031 (− 0.050, − 0.011)− 0.108 (− 0.127, − 0.089) Black− 0.010 (− 0.030, 0.009)− 0.029 (− 0.058, − 0.001)− 0.120 (− 0.151, − 0.090) Other races0.010 (− 0.008, 0.028)− 0.043 (− 0.070, − 0.017)− 0.118 (− 0.146, − 0.090)Stratified by asthma onset age (years) < 12− 0.096 (− 0.134, − 0.058)− 0.076 (− 0.114, − 0.038)− 0.071 (− 0.106, − 0.035) 12–390.024 (− 0.011, 0.059)0.018 (− 0.016, 0.052)0.026 (− 0.007, 0.059) ≥ 400.046 (− 0.055, 0.148)0.034 (− 0.066, 0.133)0.016 (− 0.076, 0.108)Model 1 = no covariates were adjusted. Model 2 = Model 1 + gender, race were adjusted. Model 3 = Model 2 + asthma onset age, education level, PIR, smoking, alcohol, physical activity intensity, hypertension, diabetes, prescription medications, energy, sugars, fat, cholesterol, moisture were adjusted*In the subgroup analysis stratified by each covariate, the model is not adjusted for the stratification variable itself. PIR: Ratio of family income to poverty.
## Discussion
The “obesity paradox” has attracted the interest of researchers in the development of many diseases [36–38]. Although chronic diseases of the respiratory system are also affected by the “obesity paradox”, they are mainly lung cancer and chronic obstructive pulmonary disease [39, 40]. The “obesity paradox” is rarely mentioned in the development of asthma because there was a general consensus that obesity could cause a higher prevalence of asthma and the risk of associated adverse events [2]. However, the association between asthma duration and obesity is still not well defined. In the present study, we first demonstrated that an increase with the duration of asthma caused a change in obesity-related indexes. Surprisingly, the correlation between asthma duration and them showed different trends when BMI and WWI were used as response variables, respectively. Because all participants in this study were adults, our results indicated a decrease in weight but an increase in the risk of central obesity with increasing asthma duration in these participants. In adults, BMI basically only responds to changes in body weight. The ability of weight change alone to reflect the true status of fat accumulation and obesity has been questioned by researchers in recent years [21, 22], especially when the concept of muscle-fat-liver axis was introduced in which researchers recognized that weight loss is also very likely to be caused by loss of muscle mass, while visceral fat can be a more true reflection of obesity [41]. Consequently, what we could believe with the present study was that with longer asthma duration, asthmatic patients were at risk of developing more central obesity, even if they were losing weight. In spite of the unadjusted model showing a negative association between asthma duration and WWI, such a finding was validated by trend tests in the full adjusted model after adjusting for all covariates. And the seemingly contradictory correlation between asthma duration and the two different obesity-related indexes in this study might be closely related to muscle steatosis [42], disease-induced reduction in exercise [43], and glucocorticoid use [44], among other reasons. Moreover, we also found that the association between asthma duration and BMI was not always negative, and presented a positive correlation between the two in participants with asthma duration not exceeding 3 years. It is known that muscle has a higher density compared to fat [45], and this phenomenon might be due to the fact that the rate of fat accumulation is greater than the rate of muscle loss early in the onset of asthma, while the disadvantages of muscle loss gradually manifest themselves as the duration of asthma increases. However, there is a positive association between asthma duration and WWI from the beginning to the end, which also indicated that WWI could better reflect the real situation of fat accumulation.
In a subgroup analysis, we found that the association between asthma duration and the two obesity indexes in this study was more significant in participants with early-onset asthma. Although there was no significant difference between early-onset and late-onset asthma in terms of asthma severity [46–48], it was undeniable that early-onset asthma might have a higher risk of steroid use. All of these factors mentioned above are strongly associated with a reduction in muscle mass and the development of central obesity. For example, it has been shown that airway obstruction symptoms were positively associated with decreased muscle mass and activity levels [49], and it has been accepted by all researchers that one of the side effects of glucocorticoids is the development of central obesity [24]. In addition, we found that although female had higher BMI and WWI (Tables 1, 2), the positive correlation between asthma duration and WWI was more pronounced in male participants, which might be related to the fact that the prolonged duration of asthma altered the changes in testosterone in male participants and thus affected the metabolism of visceral fat [50].
As far as we know, this was the first cross-sectional study to explore the relationship between asthma duration and obesity-related indexes in a large sample. However, our study still had several limitations. First, cross-sectional studies cannot explain causality, and follow-up prospective studies are necessary. Second, previous studies have shown that obese patients tend to have worse pulmonary function [27], which may lead to misdiagnosis of asthma in a proportion of obese patients who have not perfected the relevant tests. Therefore, since the diagnosis of asthma in this study was determined by a patient self-reported questionnaire, this also implied that some participants might have been misdiagnosed with asthma. However, based on the current sample size, we concluded that this situation would not ultimately lead to a substantial change in our current results. In addition, the potential influences of asthma duration and obesity-related indices are numerous, and although we included relevant covariates in the models for adjustment based on previous studies, there was no guarantee of bias from other potential covariates. Finally, this was a study based on a cohort of US adults, and the applicability of the current results to populations of other age groups and countries requires follow-up studies.
## Conclusions
There was a negative association between asthma duration and BMI, but a positive association with WWI. Males with asthma onset younger than 12 years of age and aged less than 40 years should be more cautious about the risk of higher WWI. Also, females with asthma onset less than 12 years of age and older than 60 years of age should be aware of a higher risk of weight loss.
## References
1. Levy ML, Bacharier LB, Bateman E, Boulet LP, Brightling C, Buhl R. **Key recommendations for primary care from the 2022 Global Initiative for Asthma (GINA) update**. *NPJ Prim Care Respir Med* (2023) **33** 7. DOI: 10.1038/s41533-023-00330-1
2. Miethe S, Karsonova A, Karaulov A, Renz H. **Obesity and asthma**. *J Allergy Clin Immunol* (2020) **146** 685-693. DOI: 10.1016/j.jaci.2020.08.011
3. Garcia-Rio F, Alvarez-Puebla MJ, Esteban-Gorgojo I, Barranco P, Olaguibel JM. **Obesity and asthma: key clinical questions**. *J Investig Allergol Clin Immunol* (2019) **29** 262-271. DOI: 10.18176/jiaci.0316
4. Gonzalez-Barcala FJ, Nieto-Fontarigo JJ, Lourido-Cebreiro T, Rodríguez-García C, San-Jose ME, Carreira JM. **Obesity does not increase the risk of asthma readmissions**. *J Clin Med* (2020). DOI: 10.3390/jcm9010221
5. Henkin S, Brugge D, Bermudez OI, Gao X. **A case-control study of body mass index and asthma in Asian children**. *Ann Allergy Asthma Immunol* (2008) **100** 447-451. DOI: 10.1016/S1081-1206(10)60469-3
6. Leung TF, Kong AP, Chan IH, Choi KC, Ho CS, Chan MH. **Association between obesity and atopy in Chinese schoolchildren**. *Int Arch Allergy Immunol* (2009) **149** 133-140. DOI: 10.1159/000189196
7. Deesomchok A, Fisher T, Webb KA, Ora J, Lam YM, Lougheed MD. **Effects of obesity on perceptual and mechanical responses to bronchoconstriction in asthma**. *Am J Respir Crit Care Med* (2010) **181** 125-133. DOI: 10.1164/rccm.200906-0934OC
8. Vàzquez-Nava F, Morales Romero J, Crodova Fernandez JA, Saldívar-González AH, Vàzquez-RodriguezBarrientos Gomez Mdel FC. **Association between obesity and asthma in preschool Mexican children**. *ScientificWorldJournal* (2010) **10** 1339-1346. DOI: 10.1100/tsw.2010.134
9. McDonald VM, Gibson PG. **Treatable traits in asthma and COPD**. *Arch Bronconeumol* (2022) **58** 583-585. DOI: 10.1016/j.arbres.2021.07.003
10. McDonald VM, Hiles SA, Godbout K, Harvey ES, Marks GB, Hew M. **Treatable traits can be identified in a severe asthma registry and predict future exacerbations**. *Respirology* (2019) **24** 37-47. DOI: 10.1111/resp.13389
11. Villeneuve T, Guilleminault L. **Asthma and obesity in adults**. *Rev Mal Respir* (2020) **37** 60-74. DOI: 10.1016/j.rmr.2019.03.016
12. Dixon AE, Que LG. **Obesity and asthma**. *Semin Respir Crit Care Med* (2022) **43** 662-674. DOI: 10.1055/s-0042-1742384
13. Zhu Z, Guo Y, Shi H, Liu CL, Panganiban RA, Chung W. **Shared genetic and experimental links between obesity-related traits and asthma subtypes in UK Biobank**. *J Allergy Clin Immunol* (2020) **145** 537-549. DOI: 10.1016/j.jaci.2019.09.035
14. Bourdin A, Molinari N, Vachier I, Varrin M, Marin G, Gamez AS. **Prognostic value of cluster analysis of severe asthma phenotypes**. *J Allergy Clin Immunol* (2014) **134** 1043-1050. DOI: 10.1016/j.jaci.2014.04.038
15. Schultz C, Tupper OD, Ulrik CS. **Static lung volumes and diffusion capacity in adults 30 years after being diagnosed with asthma**. *Asthma Res Pract* (2022) **8** 4. DOI: 10.1186/s40733-022-00086-4
16. Eger K, Kroes JA, Ten Brinke A, Bel EH. **Long-term therapy response to anti-IL-5 biologics in severe asthma-a real-life evaluation**. *J Allergy Clin Immunol Pract* (2021) **9** 1194-1200. DOI: 10.1016/j.jaip.2020.10.010
17. Holguin F, Bleecker ER, Busse WW, Calhoun WJ, Castro M, Erzurum SC. **Obesity and asthma: an association modified by age of asthma onset**. *J Allergy Clin Immunol* (2011) **127** 1486-93.e2. DOI: 10.1016/j.jaci.2011.03.036
18. Ahmadiafshar A, Tabbekhha S, Mousavinasab N, Khoshnevis P. **Relation between asthma and body mass index in 6–15 years old children**. *Acta Med Iran* (2013) **51** 615-619. PMID: 24338192
19. Glass AR, Swerdloff RS, Bray GA, Dahms WT, Atkinson RL. **Low serum testosterone and sex-hormone-binding-globulin in massively obese men**. *J Clin Endocrinol Metab* (1977) **45** 1211-1219. DOI: 10.1210/jcem-45-6-1211
20. Allan CA, McLachlan RI. **Androgens and obesity**. *Curr Opin Endocrinol Diabetes Obes* (2010) **17** 224-232. DOI: 10.1097/MED.0b013e3283398ee2
21. Flegal KM, Kit BK, Orpana H, Graubard BI. **Association of all-cause mortality with overweight and obesity using standard body mass index categories: a systematic review and meta-analysis**. *JAMA* (2013) **309** 71-82. DOI: 10.1001/jama.2012.113905
22. Heymsfield SB, Cefalu WT. **Does body mass index adequately convey a patient's mortality risk**. *JAMA* (2013) **309** 87-88. DOI: 10.1001/jama.2012.185445
23. Fui MN, Dupuis P, Grossmann M. **Lowered testosterone in male obesity: mechanisms, morbidity and management**. *Asian J Androl* (2014) **16** 223-231. DOI: 10.4103/1008-682X.122365
24. Lengton R, Iyer AM, van der Valk ES, Hoogeveen EK, Meijer OC, van der Voorn B. **Variation in glucocorticoid sensitivity and the relation with obesity**. *Obes Rev* (2022) **23** e13401. DOI: 10.1111/obr.13401
25. Gonzalez-Barcala FJ, Takkouche B, Valdes L, Leis R, Alvarez-Calderon P, Cabanas R. **Body composition and respiratory function in healthy non-obese children**. *Pediatr Int* (2007) **49** 553-557. DOI: 10.1111/j.1442-200X.2007.02420.x
26. Thomas EL, Frost G, Taylor-Robinson SD, Bell JD. **Excess body fat in obese and normal-weight subjects**. *Nutr Res Rev* (2012) **25** 150-161. DOI: 10.1017/S0954422412000054
27. Dixon AE, Peters U. **The effect of obesity on lung function**. *Expert Rev Respir Med* (2018) **12** 755-767. DOI: 10.1080/17476348.2018.1506331
28. Park Y, Kim NH, Kwon TY, Kim SG. **A novel adiposity index as an integrated predictor of cardiometabolic disease morbidity and mortality**. *Sci Rep* (2018) **8** 16753. DOI: 10.1038/s41598-018-35073-4
29. Li Q, Qie R, Qin P, Zhang D, Guo C, Zhou Q. **Association of weight-adjusted-waist index with incident hypertension: the Rural Chinese Cohort Study**. *Nutr Metab Cardiovasc Dis* (2020) **30** 1732-1741. DOI: 10.1016/j.numecd.2020.05.033
30. Kim JY, Choi J, Vella CA, Criqui MH, Allison MA, Kim NH. **Associations between Weight-Adjusted Waist Index and Abdominal Fat and Muscle Mass: multi-ethnic study of atherosclerosis**. *Diabetes Metab J* (2022) **46** 747-755. DOI: 10.4093/dmj.2021.0294
31. Kim NH, Park Y, Kim NH, Kim SG. **Weight-adjusted waist index reflects fat and muscle mass in the opposite direction in older adults**. *Age Ageing* (2021) **50** 780-786. DOI: 10.1093/ageing/afaa208
32. Cai S, Zhou L, Zhang Y, Cheng B, Zhang A, Sun J. **Association of the weight-adjusted-waist index with risk of all-cause mortality: a 10-year follow-up study**. *Front Nutr* (2022) **9** 894686. DOI: 10.3389/fnut.2022.894686
33. Qin Z, Chang K, Yang Q, Yu Q, Liao R, Su B. **The association between weight-adjusted-waist index and increased urinary albumin excretion in adults: a population-based study**. *Front Nutr* (2022) **9** 941926. DOI: 10.3389/fnut.2022.941926
34. Tiwari A, Rahman K, Abejie B, Jain VV, Vempilly JJ. **Longer duration of asthma is significantly associated with increased RV/TLC ratio**. *Respir Med* (2017) **124** 44-48. DOI: 10.1016/j.rmed.2017.01.011
35. Chen Y, Yang J, Han K, Wang Y, Zhuang C, Zhu L. **An elevated METS-IR index is associated with higher asthma morbidity and earlier age of first asthma in US adults: results based on a cross-sectional study**. *Front Endocrinol (Lausanne).* (2022) **13** 920322. DOI: 10.3389/fendo.2022.920322
36. Koifman E, Arow Z. **Obesity paradox—truth or misconception**. *Cardiovasc Revasc Med* (2022) **38** 9-10. DOI: 10.1016/j.carrev.2022.02.017
37. Tutor AW, Lavie CJ, Kachur S, Milani RV, Ventura HO. **Updates on obesity and the obesity paradox in cardiovascular diseases**. *Prog Cardiovasc Dis* (2022). DOI: 10.1016/j.pcad.2022.11.013
38. Gravina G, Ferrari F, Nebbiai G. **The obesity paradox and diabetes**. *Eat Weight Disord* (2021) **26** 1057-1068. DOI: 10.1007/s40519-020-01015-1
39. Yohannes AM. **The paradox of obesity in patients with chronic obstructive pulmonary disease**. *Ann Am Thorac Soc* (2022) **19** 1638-1639. DOI: 10.1513/AnnalsATS.202206-525ED
40. Barbi J, Patnaik SK, Pabla S, Zollo R, Smith RJ, Sass SN. **Visceral obesity promotes lung cancer progression-toward resolution of the obesity paradox in lung cancer**. *J Thorac Oncol* (2021) **16** 1333-1348. DOI: 10.1016/j.jtho.2021.04.020
41. ZambonAzevedo V, Silaghi CA, Maurel T, Silaghi H, Ratziu V, Pais R. **Impact of sarcopenia on the severity of the liver damage in patients with non-alcoholic fatty liver disease**. *Front Nutr* (2021) **8** 774030. DOI: 10.3389/fnut.2021.774030
42. Karakousis ND, Kotsiou OS, Gourgoulianis KI. **Bronchial asthma and sarcopenia: an upcoming potential interaction**. *J Pers Med.* (2022). DOI: 10.3390/jpm12101556
43. Cassim R, Koplin JJ, Dharmage SC, Senaratna BC, Lodge CJ, Lowe AJ. **The difference in amount of physical activity performed by children with and without asthma: a systematic review and meta-analysis**. *J Asthma* (2016) **53** 882-892. DOI: 10.1080/02770903.2016.1175474
44. Trzil JE. **Feline asthma: diagnostic and treatment update**. *Vet Clin North Am Small Anim Pract* (2020) **50** 375-391. DOI: 10.1016/j.cvsm.2019.10.002
45. Ilich JZ, Kelly OJ, Inglis JE, Panton LB, Duque G, Ormsbee MJ. **Interrelationship among muscle, fat, and bone: connecting the dots on cellular, hormonal, and whole body levels**. *Ageing Res Rev* (2014) **15** 51-60. DOI: 10.1016/j.arr.2014.02.007
46. Liu QH, Kan X, Wang YB, Liu KX, Zeng D. **Differences in the clinical characteristics of early- and late-onset asthma in elderly patients**. *Biomed Res Int* (2020) **2020** 2940296. DOI: 10.1155/2020/2940296
47. Heffler E, Blasi F, Latorre M, Menzella F, Paggiaro P, Pelaia G. **The severe asthma network in Italy: findings and perspectives**. *J Allergy Clin Immunol Pract* (2019) **7** 1462-1468. DOI: 10.1016/j.jaip.2018.10.016
48. Tan DJ, Walters EH, Perret JL, Burgess JA, Johns DP, Lowe AJ. **Clinical and functional differences between early-onset and late-onset adult asthma: a population-based Tasmanian Longitudinal Health Study**. *Thorax* (2016) **71** 981-987. DOI: 10.1136/thoraxjnl-2015-208183
49. Won HK, Kang Y, An J, Lee JH, Song WJ, Kwon HS. **Relationship between asthma and sarcopenia in the elderly: a nationwide study from the KNHANES**. *J Asthma* (2022). DOI: 10.1080/02770903.2022.2047716
50. Han YY, Yan Q, Yang G, Chen W, Forno E, Celedon JC. **Serum free testosterone and asthma, asthma hospitalisations and lung function in British adults**. *Thorax* (2020) **75** 849-854. DOI: 10.1136/thoraxjnl-2020-214875
|
---
title: Apigenin reduces the suppressive effect of exosomes derived from irritable
bowel syndrome patients on the autophagy of human colon epithelial cells by promoting
ATG14
authors:
- Rui Fu
- Saiyue Liu
- Mingjin Zhu
- Jiajie Zhu
- Mingxian Chen
journal: World Journal of Surgical Oncology
year: 2023
pmcid: PMC10012571
doi: 10.1186/s12957-023-02963-5
license: CC BY 4.0
---
# Apigenin reduces the suppressive effect of exosomes derived from irritable bowel syndrome patients on the autophagy of human colon epithelial cells by promoting ATG14
## Abstract
### Background
Inflammatory bowel disease (IBS) is a chronic disorder of the gastrointestinal tract. Exosomes have been involved in various pathological processes including IBS. Apigenin has been reported to suppress inflammatory bowel disease (IBS). However, the regulatory roles of exosomes derived from IBS patients (IBS-exos) on human colon epithelial cells are still unclear.
### Methods
Exosomes were collected from IBS patients (IBS-exos) and co-cultured with CACO-2 cells. Apigenin was used to treat IBS-exos-treated CACO-2 cells. By exploring the public data bank, we figured out the regulators control the autophagy of CACO-2 cells.
### Results
Administration of apigenin dose-dependently abolished the inhibitory effect of IBS-exo on the autophagy of CACO-2 cells. A mechanistic study showed that miR-148b-3p bound to 3′UTR to suppress ATG14 and decrease autophagy. Moreover, results suggested that ATG14 overexpression promoted the autophagy of CACO-2 cells in the presence of miR-148b-3p mimic.
### Conclusion
The current study showed that apigenin dose-dependently abolished the inhibitory effect of IBS-exo on CACO-2 cell autophagy by regulating miR-148b-3p/ATG14 signaling.
## Introduction
Irritable bowel syndrome (IBS) is characterized by abdominal pain and alterations in bowel habits [1]. The prevalence rates of IBS in the USA and Canada are reported as about $12\%$ [2]. Studies have indicated that IBS patients have a poor life quality and heavily use the health care system [3, 4].
Apigenin, one of the most studied phenolics, is present principally as glycosylated in significant amount in vegetables, fruits, and herbs [5]. Apigenin has been used as an antioxidant and anti-inflammatory [6, 7]. Apigenin was also reported to suppress various human cancers [8]. A recent study also demonstrated that apigenin is involved in anti-inflammation and autophagy and suppresses IBS [9, 10]. However, the role of apigenin in the autophagy of human colon epithelial cells is unclear.
Exosomes, 30–120-nm membrane-derived vesicles containing DNAs, mRNAs, microRNAs, or proteins, participate in cell communication and protein/RNA delivery [11, 12]. Exosomes have been shown to be secreted by a broad spectrum of cells and regulate the pathological development of numerous diseases [13, 14]. For example, exosomes secreted by mesenchymal stem cell (MSC) reduce myocardial ischemia/reperfusion injury [15]. Gallet et al. have shown that cardiosphere-derived cell-secreted exosomes reduce scarring and alleviate myocardial infarction [16]. Irritable bowel syndrome (IBS) is a chronic disorder of the intestines [1]. However, the regulatory relationship between exosomes derived from IBS patients (IBS-exos) and human colon epithelial cells is still unclear.
Autophagy is a cell survival mechanism which adapts cells to metabolic stresses [17]. Autophagy plays an important role in different cellular processes [18, 19]. Studies also demonstrated that autophagy plays a protective role against some human diseases, and autophagy dysfunction has previously been associated with a variety of diseases including cancer, neurodegeneration, and IBS [18, 20–22]. Studies also indicated that intestinal epithelial cells constitute the first physical barrier to protect the intestinal mucosa from injury, and the activation of intestinal epithelial cell autophagy is essential to maintain intestine function [23, 24]. However, the relationship between IBS-exos and the autophagy of human colon epithelial cells remains to be elucidated.
miR-148b-3p involves in different biological processes including autophagy and apoptosis. For instance, overexpressing miR-148b-3p down-regulated the viability, but increased the apoptosis of hypoxia/reoxygenation-treated cardiomyocytes [25]. A study also indicated that miR-148b-3p regulated pancreatic autophagy via suppression of autophagy elated 12 (ATG12) [26]. miR-148a has been shown to regulate autophagy by down-regulating IL-6/STAT3 signaling [27]. However, the function of miR-148b-3p in the autophagy of human colon epithelial cells is largely unknown.
Autophagy-related 14 (ATG14) play a very important role in autophagy by directing Complex I to function in autophagy by regulating its localization [28]. Xiong et al. have shown that ATG14 plays a critical role in hepatic autophagy and lipid metabolism [29]. Diao et al. indicated that ATG14 enhances membrane tethering and fusion of autophagosomes to endolysosomes [30]. But the role of ATG14 in the autophagy of human colon epithelial cells is rarely studied.
This study aims to explore the relationship between IBS-exos and autophagy of human colon epithelial cells, and the effect of apigenin on autophagy in human colon epithelial cells, therefore providing data for a better understanding of the role of apigenin in autophagy and IBS.
## Human blood
The study was approved by the Ethics Committee of Tongde Hospital of Zhejiang Province. Fifteen blood samples of IBS patients diagnosed according to Rome III criteria or controls were used to isolate exosomes. Written informed consent was received.
## Cell culture
CACO-2 cells were purchased from Shanghai Biology Institute and maintained in DMEM (Gibco, Carlsbad, CA, USA) with $10\%$ FBS (Gibco) in an incubator at 37°C plus $5\%$ CO2 atmosphere.
## Isolation and characterization of exosomes
Exosome in serum was collected as described previously [31]. Briefly, the serums were initially centrifuged at 3000 g for 15 min, to remove cells and other debris, and then the supernatants were span at 10000 g for 20 min to remove shedding vesicles and other vesicles that were larger than exosomes. Finally, the supernatants were span at 100,000 g for 1h at 4°C. Pellets were re-suspended in PBS and characterized by transmission electron microscopy (TEM) and immuno-staining.
## Exosome uptake analysis
Exosomes were stained by green fluorescent linker PKH67 (UR52303, Umibio, Shanghai, China). One milliliter of exosomes (1 μg/mL) was incubated with 2 μL PKH67 for 25 min at room temperature. In order to bind excess dye, 2 mL of $0.5\%$ BSA/PBS was added. The labeled exosomes were washed at 100,000 g for 1 h, and the exosome pellet was suspended with PBS and used for uptake experiments. CACO-2 cells were seeded (50,000/well) and treated by medium with/without PKH67-labeled exosomes for 24 h. DAPI was used to stain the nucleus. Uptaking was observed under a fluorescence microscope (Leica Microsystems, Wetzlar, Germany).
## qRT-PCR
RNA was isolated and reverse transcribed into cDNA (Invitrogen, Waltham, MA, USA). Q-PCR was done using the SYBR Green qPCR Master Mixes (Thermo Fisher, Rockford, IL, USA) as follows: 95°C for 10 min followed by 40 cycles of 95°C for 15 s and 60°C for 45 s. U6 or β-actin was used as control. *The* gene relative expression was calculated by the 2−ΔΔCt formula. The primers were as follows (5′-3′):has-miR-148b-3p, F: CGCGTCAGTGCATCACAGAA, R: AGTGCAGGGTCCGAGGTATT;U6, F: CTCGCTTCGGCAGCACA, R: AACGCTTCACGAATTTGCGT.ATG14, F: TCATTATGAGCGTCTGGC, R: ATGCTGGTGTCTCCGTTG;β-actin, F: AATGCCTTCACGATGTTC, R: AGCCTGCTGTAATATTGC.
## Immunoblotting
Protein was isolated using RIPA lysis buffer (JRDUN, Shanghai, P.R. China), concentration-measured by an enhanced BCA protein assay kit (Thermo Fisher Scientific), separated by $10\%$ SDS-PAGE, and immunoblotted to PVDF membranes (Millipore, Billerica, MA, USA), blocked with $5\%$ nonfat dry milk for 1 h at room temperature, and probed with primary antibodies at 4°C overnight. After washing with PBST, the bolts were incubated with a second antibody for 1 h at 37°C. An enhanced chemiluminescence system (Tanon, Shanghai, P.R. China) was used to visualize protein. Primary antibodies’ information was provided as follows: CD9 (Ab92726, Abcam, St. Louis, MO, USA), CD81 (Ab109201, Abcam), ATG14 (Ab227849, Abcam), and GAPDH (60004-1-1G, Proteintech, UK).
## Overexpression of ATG14
The pLVX-puro containing ATG14 (ovTAG14) or vector (ovNC) alone were purchased from Genechem company (Shanghai, China). CACO-2 cells were transfected with the plasmids using Lipo2000, and the cells were analyzed 48 h after transfection.
## Immunofluorescent staining
To evaluate the expression of autophagy markers LC3, cells were fixed, blocked, and probed with anti-LC3 antibody overnight at 4°C. Cells were washed and incubated in fluorochrome-conjugated secondary antibody for 1 h in the dark. Nuclei were counterstained by 4′,6-diamidino-2-phenylindole, dihydrochloride (DAPI), and cells were observed under a fluorescent microscope.
## Dual-luciferase reporter gene assay
Binding sites of miR-148b-3p and ATG14 were predicted by TargetScan. According to the prediction, wild type and mut sequences were synthesized respectively and cloned to luciferase reporter vectors (pGL3-Basic). Then, WT 3′ UTR or Mut 3′UTR plasmid was co-transfected with miR-148b-3p inhibitor or mimics into CACO-2 cells. After 48 h of transfection, a dual-luciferase reporter gene kit (Beijing Yuanpinghao Biotechnology Co., Ltd.) was used to determine the luciferase activity of cells in each group.
## Statistical analysis
Prism7.0 (La Jolla, CA) was used to analyze the data. Data was expressed as mean ± SD. Comparisons were performed by T-test or one-way ANOVA with Tukey’s post hoc test. P values less than 0.05 were considered as significant.
## Isolation and identification of exosomes
We collected serum samples from IBS patients and controls to extract exosomes. Exosomes from IBS patients (IBS-exo) or controls (control-exo) are shown in Fig. 1A. Immunoblotting further confirmed the expression of markers CD9 and CD81 (Fig. 1B). Co-culture assay showed that exosomes were up-taken by CACO-2 cells (Fig. 1C). This laid the foundation of this study. Fig. 1Isolation and characterization of exosomes. A Exosome morphology. B Western blotting detection of exosome markers, CD9 and CD81. *** $p \leq 0.001$ vs vehicle;!!!$p \leq 0.001$ vs oeNC + IBS-exo. C Uptake of exosomes by human colon epithelial Caco-2 cells was determined using PKH-67 dye
## Exosome from IBS patients decreased the autophagy in CACO-2 cells
We next examined the effect of exosomes on autophagy of CACO-2, using LC3 as a maker of autophagy. Data revealed that co-culture with IBS-exos reduced autophagy in CACO-2 cells, compared to controls (Fig. 2).Fig. 2The autophagy of CACO-2 cells was decreased after co-cultured with IBS-exo. IMF staining examination of the effect of IBS-exo on CACO-2 autophagy
## Apigenin dose-dependently abolished the inhibitory effect of IBS-exo on CACO-2 autophagy
In order to know whether apigenin affects CACO-2 autophagy, IMF staining was performed. Results suggested that apigenin dose-dependently abolished the inhibitory effect of IBS-exo CACO-2 autophagy (Fig. 3A, B). Western blotting showed that apigenin dose-dependently diminished IBS-exo-caused decrease of ATG14 protein in CACO-2 cells (Fig. 3C). This result indicates that apigenin abolished the suppression of autophagy by IBS-exos through regulating ATG14.Fig. 3Apigenin dose-dependently promoted autophagy of CACO-2 cells that were co-cultured with IBS-exos. A, B IMF staining of LC3 was used to examine the autophagy of CACO-2 cells. *** $p \leq 0.01$ vs control; ###$p \leq 0.001$ vs IBS-exo. C ATG14 expression levels
## miR-148b-3p bound 3′UTR of ATG14 to suppress its expression
We further investigated the potential mechanisms by which apigenin promote autophagy of CACO-2 cells. By searching available data bank, we speculated ATG14 was a target of miR-148b-3p. So, we transfected CACO-2 cells with miR-148b-3p miNC, inhibitor, or mimic. QRT-PCR results indicated that transfection of miR-148b-3p inhibitor increased ATG14, while miR-148b-3p mimic decreased ATG14 (Fig. 4A). Transfection of miR-148b-3p mimic also decreased ATG14, while transfection of miR-148b-3p inhibitor increased ATG14 at protein level (Fig. 4B). Dual-luciferase reporter assay also confirmed the binding of has-miR-148b-3p and ATG14. Together, these findings indicated that miR-148b-3p suppressed ATG14 transcription through the binding on its 3′UTR.Fig. 4miR-148b-3p suppressed ATG14 via binding to 3′UTR. A Levels of miR-148b-3p and ATG14. B Protein levels of ATG14. C has-miR-148b-3p binding sites and corresponding mutation. D Dual-luciferase reporter gene verification of the binding of has-miR-148b-3p and ATG14. *** $p \leq 0.001$ vs miNC
## Overexpressing ATG14 promoted CACO-2 autophagy in the presence of miR-148b-3p mimic
To study the role of ATG14/miR-148b-3p in autophagy, ATG14 was successfully overexpressed in CACO-2 cells (Fig. 5A, B). IMF staining results indicated that overexpression of ATG14 abolished miR-148b-3p mimic caused autophagy of CACO-2 cells (Fig. 5C). Then, we examined the relative protein levels of ATG14. Western blots showed that overexpression of ATG14 reversed miR-148b-3p mimic caused a decrease of ATG14 (Fig. 5D). These results indicated that ATG14 overexpression promoted CACO-2 autophagy in the presence of miR-148b-3p mimic. Fig. 5ATG14 overexpression promoted CACO-2 autophagy in the presence of miR-148b-3p mimic. A, B Relative mRNA and protein levels of ATG14 in CACO-2 cells after overexpressing ATG14. *** $p \leq 0.001$ vs oeNC. C Overexpression of ATG14 abolished miR-148b-3p mimic caused autophagy of CACO-2 cells. *** $p \leq 0.001$ vs miNC; ###$p \leq 0.001$ vs mimic. D Protein levels of ATG14 in CACO-2 cells after transfecting with oeNC or oeATG14 in the presence of miR-148b-3p mimic
## ATG14 overexpression promoted the autophagy of CACO-2 cells in the presence of miR-148b-3p mimic through increasing ATG14
To study the underlying mechanism by which ATG14 overexpression promoted CACO-2 autophagy, CACO-2 cells were treated by miR-148-3p mimic in the presence of apigenin. Results showed that apigenin promoted the autophagy of CACO-2 after co-cultured with miR-148-3p mimic (Fig. 6A, B). Q-PCR results indicated that apigenin did not affect miR-148b-3p in CACO-2 cells after co-cultured with miR-148b-3p mimic (Fig. 6C). However, apigenin treatment enhanced the protein level of ATG14 in CACO-2 cells in the presence of miR-148-3p mimic (Fig. 6D). Together, the data suggested that ATG14 overexpression promoted CACO-2 autophagy in the presence of miR-148b-3p mimic. Fig. 6Overexpressing ATG14 promoted CACO-2 autophagy in the presence of miR-148b-3p mimic through increasing ATG14. A, B Apigenin promoted the autophagy of CACO-2 after co-cultured with miR-148-3p mimic. ** $p \leq 0.01$ vs miNC; ###$p \leq 0.001$ vs mimic. C Apigenin did not affect miR-148b-3p in CACO-2 cells after co-cultured with miR-148b-3p mimic. D Apigenin treatment enhanced the protein level of ATG14 in CACO-2 cells in the presence of miR-148-3p mimic
## Discussion
In this study, IBS-exos were successfully isolated and used to treat CACO-2 cells. We demonstrated that IBS-exos decreased the autophagy in CACO-2 cells. Administration of apigenin dose-dependently abolished the inhibitory effect of IBS-exo on CACO-2 autophagy. A mechanistic study indicated that miR-148b-3p suppressed ATG14 to suppress autophagy through the binding to its 3′UTR. In contrast, ATG14 overexpression promoted the autophagy of CACO-2 cells in the presence of miR-148b-3p mimic. These results identified a novel role of miR-148b-3p/ATG14 in CACO-2 autophagy and may facilitate the development of new drugs for IBS.
Apigenin plays a role in various diseases. For example, Malik et al. have demonstrated that apigenin ameliorated STZ-induced diabetic nephropathy [32]. Anusha et al. have reported that apigenin has a protective Parkinson’s disease via suppression of ROS-mediated apoptosis [33]. Apigenin has also been shown to suppress lupus [34]. The results indicate a key role of apigenin in the regulation of CACO-2 cell autophagy, showing for the first time that apigenin dose-dependently abolished the inhibitory effect of IBS-exos on the autophagy of CACO-2 cells.
miRNAs are small, noncoding RNA (21–25 nucleotides) that regulate gene expression [35]. Studies have shown that miRNA dysregulation regulates different biological processes [36, 37]. Kim et al. indicated that miR-148b-3p regulates angiogenesis and is a therapeutic candidate for overcoming endothelial cell dysfunction and angiogenic disorders [38]. Arambula-Meraz et al. have observed a correlation between miR-148b-3p with two established biomarkers of prostate cancer, PSA and PCA3, suggesting its potential as a biomarker of prostate cancer [39]. MiR-148b-3p has also been shown to inhibit the pro-angiogenic phenotype of endothelial cells [40]. This study further explored its biological function. We showed miR-148b-3p suppressed the transcription of ATG14 to suppress autophagy of CACO-2 cells through binding to its 3′UTR. These findings indicated a new function of miR-148b-3p in IBS, showing miR-148b-3p inhibited the autophagy of CACO-2 cells by suppressing ATG14 transcription through binding to its 3′UTR.
Our results indicated that miR-148b-3p bound directly to the 3′UTR of ATG14 promoter to negatively regulated ATG14 expression, which was verified by the facts that overexpressing ATG14 abolished miR-148b-3p-caused suppression of CACO-2 autophagy. The findings demonstrate for the first time that miR-148b-3p targets ATG14, and highlight the importance of miR-148b-3p/ATG14 signaling axis in human colon epithelial cells and IBS. Future studies in animals will provide more relevant data. Although shortcomings exist, this study demonstrated a new role of apigenin in the autophagy of CACO-2 cells.
## Conclusion
This study demonstrated a new function of apigenin in the autophagy of human colon epithelial cells, showing that apigenin dose-dependently abolished the inhibitory effects of IBS-exo on CACO-2 autophagy by regulating miR-148b-3p/ATG14 signaling.
## Code availability
Not applicable
## References
1. Canavan C, West J, Card T. **The epidemiology of irritable bowel syndrome**. *Clin Epidemiol.* (2014.0) **6** 71-80. PMID: 24523597
2. Lovell RM, Ford AC. **Global prevalence of and risk factors for irritable bowel syndrome: a meta-analysis**. *Clin Gastroenterol Hepatol.* (2012.0) **10** 712-721 e714. DOI: 10.1016/j.cgh.2012.02.029
3. Whitehead WE, Burnett CK, Cook EW, Taub E. **Impact of irritable bowel syndrome on quality of life**. *Dig Dis Sci.* (1996.0) **41** 2248-2253. DOI: 10.1007/BF02071408
4. Occhipinti K, Smith JW. **Irritable bowel syndrome: a review and update**. *Clin Colon Rectal Surg.* (2012.0) **25** 46-52. DOI: 10.1055/s-0032-1301759
5. Hostetler GL, Ralston RA, Schwartz SJ. **Flavones: food sources bioavailability, metabolism, and bioactivity**. *Adv Nutr.* (2017.0) **8** 423-435. DOI: 10.3945/an.116.012948
6. Papay ZE, Kosa A, Boddi B, Merchant Z, Saleem IY, Zariwala MG. **Study on the pulmonary delivery system of apigenin-loaded albumin nanocarriers with antioxidant activity**. *J Aerosol Med Pulm Drug Deliv.* (2017.0) **30** 274-288. DOI: 10.1089/jamp.2016.1316
7. Wang YC, Huang KM. **In vitro anti-inflammatory effect of apigenin in the Helicobacter pylori-infected gastric adenocarcinoma cells**. *Food Chem Toxicol.* (2013.0) **53** 376-383. DOI: 10.1016/j.fct.2012.12.018
8. Yan X, Qi M, Li P, Zhan Y, Shao H. **Apigenin in cancer therapy: anti-cancer effects and mechanisms of action**. *Cell Biosci.* (2017.0) **7** 50. DOI: 10.1186/s13578-017-0179-x
9. Zhang X, Bu H, Jiang Y, Sun G, Jiang R, Huang X. **The antidepressant effects of apigenin are associated with the promotion of autophagy via the mTOR/AMPK/ULK1 pathway**. *Mol Med Rep.* (2019.0) **20** 2867-2874. PMID: 31322238
10. Mascaraque C, Gonzalez R, Suarez MD, Zarzuelo A, Sanchez de Medina F, Martinez-Augustin O. **Intestinal anti-inflammatory activity of apigenin K in two rat colitis models induced by trinitrobenzenesulfonic acid and dextran sulphate sodium**. *Br J Nutr.* (2015.0) **113** 618-626. DOI: 10.1017/S0007114514004292
11. Lin J, Li J, Huang B, Liu J, Chen X, Chen XM. **Exosomes: novel biomarkers for clinical diagnosis**. *TheScientificWorldJournal.* (2015.0) **2015** 657086. DOI: 10.1155/2015/657086
12. 12.Kalluri R, LeBleu VS. The biology, function, and biomedical applications of exosomes. Science. 2020;367(6478):eaau6977. 10.1126/science.aau6977.
13. Lin Y, Anderson JD, Rahnama LMA, Gu SV, Knowlton AA. **Exosomes in disease and regeneration: biological functions, diagnostics, and beneficial effects**. *Am J Physiol Heart Circ Physiol.* (2020.0) **319** H1162-H1180. DOI: 10.1152/ajpheart.00075.2020
14. Isola AL, Chen S. **Exosomes: the messengers of health and disease**. *Curr Neuropharmacol.* (2017.0) **15** 157-165. DOI: 10.2174/1570159X14666160825160421
15. Lai RC, Arslan F, Lee MM, Sze NS, Choo A, Chen TS. **Exosome secreted by MSC reduces myocardial ischemia/reperfusion injury**. *Stem Cell Res.* (2010.0) **4** 214-222. DOI: 10.1016/j.scr.2009.12.003
16. Gallet R, Dawkins J, Valle J, Simsolo E, de Couto G, Middleton R. **Exosomes secreted by cardiosphere-derived cells reduce scarring, attenuate adverse remodelling, and improve function in acute and chronic porcine myocardial infarction**. *Eur Heart J.* (2017.0) **38** 201-211. PMID: 28158410
17. Chang NC. **Autophagy and stem cells: self-eating for self-renewal**. *Front Cell Dev Biol.* (2020.0) **8** 138. DOI: 10.3389/fcell.2020.00138
18. Levine B, Klionsky DJ. **Development by self-digestion: molecular mechanisms and biological functions of autophagy**. *Dev Cell.* (2004.0) **6** 463-477. DOI: 10.1016/S1534-5807(04)00099-1
19. Yorimitsu T, Klionsky DJ. **Autophagy: molecular machinery for self-eating**. *Cell Death Differ.* (2005.0) **12** 1542-1552. DOI: 10.1038/sj.cdd.4401765
20. Cuervo AM. **Autophagy: in sickness and in health**. *Trends Cell Biol.* (2004.0) **14** 70-77. DOI: 10.1016/j.tcb.2003.12.002
21. Shintani T, Klionsky DJ. **Autophagy in health and disease: a double-edged sword**. *Science.* (2004.0) **306** 990-995. DOI: 10.1126/science.1099993
22. Iida T, Onodera K, Nakase H. **Role of autophagy in the pathogenesis of inflammatory bowel disease**. *World J Gastroenterol.* (2017.0) **23** 1944-1953. DOI: 10.3748/wjg.v23.i11.1944
23. Hu CA, Hou Y, Yi D, Qiu Y, Wu G, Kong X, Yin Y. **Autophagy and tight junction proteins in the intestine and intestinal diseases**. *Anim Nutr.* (2015.0) **1** 123-127. DOI: 10.1016/j.aninu.2015.08.014
24. Lin R, Jiang Y, Zhao XY, Guan Y, Qian W, Fu XC. **Four types of Bifidobacteria trigger autophagy response in intestinal epithelial cells**. *J Dig Dis.* (2014.0) **15** 597-605. DOI: 10.1111/1751-2980.12179
25. Sun M, Zhai M, Zhang N, Wang R, Liang H, Han Q, Jia Y, Jiao L. **MicroRNA-148b-3p is involved in regulating hypoxia/reoxygenation-induced injury of cardiomyocytes in vitro through modulating SIRT7/p53 signaling**. *Chem Biol Interact.* (2018.0) **296** 211-219. DOI: 10.1016/j.cbi.2018.10.003
26. Gao B, Wang D, Sun W, Meng X, Zhang W, Xue D. **Differentially expressed microRNA identification and target gene function analysis in starvation-induced autophagy of AR42J pancreatic acinar cells**. *Mol Med Rep.* (2016.0) **14** 590-598. DOI: 10.3892/mmr.2016.5240
27. Miao B, Qi WJ, Zhang SW, Wang H, Wang C, Hu L. **miR-148a suppresses autophagy by down-regulation of IL-6/STAT3 signaling in cerulein-induced acute pancreatitis**. *Pancreatology.* (2019.0) **19** 557-565. DOI: 10.1016/j.pan.2019.04.014
28. Obara K, Sekito T, Ohsumi Y. **Assortment of phosphatidylinositol 3-kinase complexes--Atg14p directs association of complex I to the pre-autophagosomal structure in Saccharomyces cerevisiae**. *Mol Biol Cell.* (2006.0) **17** 1527-1539. DOI: 10.1091/mbc.e05-09-0841
29. Xiong X, Tao R, DePinho RA, Dong XC. **The autophagy-related gene 14 (Atg14) is regulated by forkhead box O transcription factors and circadian rhythms and plays a critical role in hepatic autophagy and lipid metabolism**. *J Biol Chem.* (2012.0) **287** 39107-39114. DOI: 10.1074/jbc.M112.412569
30. Diao J, Liu R, Rong Y, Zhao M, Zhang J, Lai Y. **ATG14 promotes membrane tethering and fusion of autophagosomes to endolysosomes**. *Nature.* (2015.0) **520** 563-566. DOI: 10.1038/nature14147
31. Zhu W, Huang L, Li Y, Zhang X, Gu J, Yan Y. **Exosomes derived from human bone marrow mesenchymal stem cells promote tumor growth in vivo**. *Cancer Lett.* (2012.0) **315** 28-37. DOI: 10.1016/j.canlet.2011.10.002
32. Malik S, Suchal K, Khan SI. **Apigenin ameliorates streptozotocin-induced diabetic nephropathy in rats via MAPK-NF-kappaB-TNF-alpha and TGF-beta1-MAPK-fibronectin pathways**. *Am J Physiol Renal Physiol.* (2017.0) **313** F414-F422. DOI: 10.1152/ajprenal.00393.2016
33. Anusha C, Sumathi T, Joseph LD. **Protective role of apigenin on rotenone induced rat model of Parkinson’s disease: suppression of neuroinflammation and oxidative stress mediated apoptosis**. *Chem Biol Interact.* (2017.0) **269** 67-79. DOI: 10.1016/j.cbi.2017.03.016
34. Kang HK, Ecklund D, Liu M, Datta SK. **Apigenin, a non-mutagenic dietary flavonoid, suppresses lupus by inhibiting autoantigen presentation for expansion of autoreactive Th1 and Th17 cells**. *Arthritis Res Ther.* (2009.0) **11** R59. DOI: 10.1186/ar2682
35. Li G, Sun L, Mu Z. **MicroRNA-1298-5p inhibits cell proliferation and the invasiveness of bladder cancer cells via down-regulation of connexin 43**. *Biochem Cell Biol.* (2020.0) **98** 227-237. DOI: 10.1139/bcb-2019-0137
36. Wang CM, Cheng BH, Xue QJ, Chen J, Bai B. **MiR-1298 affects cell proliferation and apoptosis in C6 cells by targeting SET domain containing 7**. *Int J Immunopathol Pharmacol.* (2017.0) **30** 264-271. DOI: 10.1177/0394632017720546
37. Cai G, Qiao S, Chen K. **Suppression of miR-221 inhibits glioma cells proliferation and invasion via targeting SEMA3B**. *Biol Res.* (2015.0) **48** 37. DOI: 10.1186/s40659-015-0030-y
38. Kim H, Ko Y, Park H. **MicroRNA-148a/b-3p regulates angiogenesis by targeting neuropilin-1 in endothelial cells**. *Exp Mol Med.* (2019.0) **51** 1-11. PMID: 31827074
39. Arambula-Meraz E, Bergez-Hernandez F, Leal-Leon E. **Expression of miR-148b-3p is correlated with overexpression of biomarkers in prostate cancer**. *Genet Mol Biol.* (2020.0) **43** e20180330. DOI: 10.1590/1678-4685-gmb-2018-0330
40. Zhang H, Ye Q, Du Z, Huang M, Zhang M, Tan H. **MiR-148b-3p inhibits renal carcinoma cell growth and pro-angiogenic phenotype of endothelial cell potentially by modulating FGF2**. *Biomed Pharmacother.* (2018.0) **107** 359-367. DOI: 10.1016/j.biopha.2018.07.054
|
---
title: Conservation and divergence of canonical and non-canonical imprinting in murids
authors:
- Julien Richard Albert
- Toshihiro Kobayashi
- Azusa Inoue
- Ana Monteagudo-Sánchez
- Soichiro Kumamoto
- Tomoya Takashima
- Asuka Miura
- Mami Oikawa
- Fumihito Miura
- Shuji Takada
- Masumi Hirabayashi
- Keegan Korthauer
- Kazuki Kurimoto
- Maxim V. C. Greenberg
- Matthew Lorincz
- Hisato Kobayashi
journal: Genome Biology
year: 2023
pmcid: PMC10012579
doi: 10.1186/s13059-023-02869-1
license: CC BY 4.0
---
# Conservation and divergence of canonical and non-canonical imprinting in murids
## Abstract
### Background
Genomic imprinting affects gene expression in a parent-of-origin manner and has a profound impact on complex traits including growth and behavior. While the rat is widely used to model human pathophysiology, few imprinted genes have been identified in this murid. To systematically identify imprinted genes and genomic imprints in the rat, we use low input methods for genome-wide analyses of gene expression and DNA methylation to profile embryonic and extraembryonic tissues at allele-specific resolution.
### Results
We identify 14 and 26 imprinted genes in these tissues, respectively, with 10 of these genes imprinted in both tissues. Comparative analyses with mouse reveal that orthologous imprinted gene expression and associated canonical DNA methylation imprints are conserved in the embryo proper of the Muridae family. However, only 3 paternally expressed imprinted genes are conserved in the extraembryonic tissue of murids, all of which are associated with non-canonical H3K27me3 imprints. The discovery of 8 novel non-canonical imprinted genes unique to the rat is consistent with more rapid evolution of extraembryonic imprinting. Meta-analysis of novel imprinted genes reveals multiple mechanisms by which species-specific imprinted expression may be established, including H3K27me3 deposition in the oocyte, the appearance of ZFP57 binding motifs, and the insertion of endogenous retroviral promoters.
### Conclusions
In summary, we provide an expanded list of imprinted loci in the rat, reveal the extent of conservation of imprinted gene expression, and identify potential mechanisms responsible for the evolution of species-specific imprinting.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s13059-023-02869-1.
## Background
The brown Norway rat is an important model for human pathophysiology [1]. To facilitate pharmacogenomic studies and the identification of disease-associated variants, efforts have been made to assemble the rat genome [2] and measure the extent of genetic variability between 40 distinct lab strains [3, 4]. However, despite the utility of the rat in modeling human disease and recent advances in rat genomics, the mouse has been and continues to be the predominant model species for foundational studies of mammalian genetics. The discovery of genomic imprinting, for example, was enabled by nuclear transfer and genetic technologies developed in the mouse [5–8] and numerous follow-up studies of the molecular basis of imprinting have been carried out using mouse models.
This process of genomic imprinting results in monoallelic gene expression in a parent-of-origin manner and is essential for mammalian growth and development. Recent studies in the mouse reveal that oocyte and sperm chromatin is “imprinted” by differential epigenetic modifications that can be maintained in the embryo and adult [9]. Canonical imprinted genes are regulated by parent-specific DNA methylation (DNAme) deposited in spermatozoa or oocytes, resulting in differentially methylated regions (DMRs). Imprinted DMRs are maintained on both alleles of the embryo, conferring parent-of-origin monoallelic expression. While roughly 260 imprinted genes have been identified in mouse, 63 of which are also imprinted in human [10], only 13 imprinted genes have been reported in the rat to date [11]. Of note, 12 of these genes, including H19 and Igf2, are regulated by DMRs deposited in the gametes, consistent with canonical imprinting. Furthermore, as all 13 rat imprinted genes were characterized based on homology with their mouse or human orthologs, no rat-specific imprinted genes have been reported. Thus, the extent of genomic imprinting in rats, and the conservation of imprinting between mammals, remains unclear. The lack of a comprehensive list of imprinted genes in rats also hinders the application of comparative genomics to identify novel conserved genomic features that may contribute to imprinting.
Recent studies in the mouse have shown that an alternative mechanism of genomic imprinting, so-called non-canonical imprinting, mediates paternal-specific gene expression in extraembryonic tissues [12, 13]. Non-canonical imprints are distinguished from canonical imprints by the lack of a DNAme imprint and the enrichment in oocytes of histone 3 lysine 27 trimethylation (H3K27me3), a mark deposited by the Polycomb repressive complex PRC2. Such maternal-specific H3K27me3 is replaced by DNAme in extraembryonic tissues, resulting in a DMR which restricts expression from the maternal allele [12–15]. While the molecular basis of this switch remains unknown, recent studies in the mouse implicate GLP and G9A, which deposit H3K9me2 [16], in maintaining maternal DNAme and imprinted gene expression in extraembryonic tissues [17, 18]. Of the ~7 known non-canonical imprinted genes identified in mice, only Sfmbt2 has been confirmed as an imprinted gene in rat [11, 19]. Whether the other non-canonical imprints, such as that controlling expression of the essential growth factor gene Gab1 [12, 13, 15, 20], are conserved, remains an open question [11, 21, 22].
Parent-of-origin-specific control of gene dosage is hypothesized to be the ultimate driving factor for the evolution of genomic imprinting [23]. At the interface between embryo and mother, the placenta is responsible for nutrient transport to the embryo, and gene dosage in this tissue is critical for modulating resource allocation between the mother and embryo [21]. Indeed, the placenta shows the highest degree of imprinted gene expression in human and mouse [24], and non-canonical imprinting is restricted to the placenta [12, 21]. In an extreme case, in lieu of random X chromosome inactivation, female mice undergo inactivation of the paternal X chromosome in the placenta [25]. However, the prevalence of non-canonical imprinting and imprinted X chromosome inactivation in species other than the house mouse (Mus musculus) has not been explored.
While the catalogs of human and mouse imprinted genes have been generated over four decades of clinical, cytogenetic, and complementation research [10], recent advances in genomics have enabled the comprehensive identification of candidate imprinted genes. By crossing genetically distinct individuals, F1 hybrids are generated with homologous chromosomes that can be differentiated at the genic level in silico using known genetic variants. Bioinformatic pipelines, such as our allele-aware tool for processing epigenomic data (MEA) [26] and others such as WASP [27] and Allelome. PRO [28], input data derived from F1 hybrids and generate maternal- and paternal-genome-specific profiles. Importantly, parent-of-origin effects, including genomic imprinting, can be delineated from genetic or strain-specific effects by generating F1s from reciprocal crosses [29]. Together, allele-specific analysis of genic expression by RNA sequencing (RNAseq) and DNAme levels by whole genome bisulphite sequencing (WGBS) data derived from reciprocal F1 hybrids represent the gold standard for identifying candidate imprinted genes genome-wide in a given tissue or cell type [30, 31].
In this study, we performed RNAseq and WGBS on reciprocal F1 hybrid rat extraembryonic and embryonic tissues from early post-implantation conceptuses. Allele-specific profiling of these datasets yielded extensive maps of parent-specific gene expression and DNA methylation. In parallel, we generated similar datasets from matching mouse extraembryonic and embryonic tissues. Comparisons between rat and mouse revealed conserved canonical genomic imprints in the embryo proper but divergent non-canonical imprinting in the extraembryonic tissue. Detailed inspection of species-specific non-canonical imprinted loci reveal multiple potential mechanisms by which imprinted expression may evolve, including recent insertions of endogenous retroviral promoters that are subject to maternal-specific methylation following fertilization, as well as novel ZFP57 binding motifs that arise through single-nucleotide substitutions. Finally, analysis of H3K27me3 in rat oocytes reveals that species-specific deposition of this mark likely explains the divergent non-canonical imprinting status of some genes in extraembryonic tissues. Altogether, these results provide an atlas of genomic imprinting, including genome-wide maps of parent-of-origin gene expression and parent-specific DNAme levels in the embryo and extraembryonic tissue, and reveal relatively recent evolutionary divergence of non-canonical imprinting in murids.
## Measuring imprinted gene expression in the rat
Genetic variation between parental genomes is a prerequisite for the characterization of the transcriptome and methylome with allele-specific resolution, and in turn the systematic identification of putative imprinted genes. To produce embryos from which genomic imprinting in the rat could be assessed, we conducted reciprocal crosses of genetically distinct rat strains BN/NCrlCrlj, WKY/NCrlCrlj, and F344/NSlc (hereafter referred to as “B,” “W,” and “F,” respectively) for which whole genome sequences are available [3, 4] (Fig. 1a). Post-implantation F1 embryos were dissected at Carnegie stage 7, corresponding to embryonic day 8.5 (E8.5) in rat. To agnostically measure embryonic and extraembryonic gene expression, we performed strand-specific RNAseq on epiblast and ectoplacental cone (EPC) cells, respectively. Subsequently, to identify imprinted gene expression, we analyzed the reciprocal F1 RNAseq datasets using MEA [26], which discriminates transcripts from each allele based on informative parental SNVs as well as INDELs. Expressed autosomal transcripts (RPKM ≥1) with sufficient allele-specific read coverage (allelic RPM ≥0.5 on either allele in at least $\frac{6}{11}$ replicates, $$n = 13$$,164 and 16,642 transcripts for epiblast and EPC samples, respectively) were categorized as paternally or maternally expressed if they showed statistically significant bias in monoallelic expression (Bonferroni-adjusted p-value <0.05, Student’s t test). *Several* genes previously reported as imprinted in distantly related mammalian species including mouse, rat, human, and cow such as Igf2 and Peg10 [32–35] showed paternal allele-specific transcription in both epiblast and EPC, while H19 showed maternal-specific expression (Fig. 1b,c & Additional file 1: Fig. S1a), validating our RNAseq-based method and parameters for identifying imprinted genes. In total, 18 paternally (Snrpn, Magel2, Peg10, Sgce, Igf2, Zrsr1, Zdbf2, Sfmbt2, Sall1, Gab1, Gsto1, Rpl39l, Zfp516, Zfp64, Slc38a1, LOC108350526, Syt16-AS, Gadl1-3′UTR) and 11 maternally (Tssc4, Cd81, Ascl2, Trpm5, Mir675, H19, Grb10, Maged2, Meg3, Rtl1-AS, Gnas) expressed imprinted genes were identified in rat epiblast and EPCs using stringent cutoffs defined above. To identify genes that show a bias in parental expression levels (in addition to monoallelic expression), we performed linear modeling of allele-specific data using Limma. An additional 8 paternally (Aig1, Wasf3, Pomt1, Tmem67, Tmem30b, Slc38a4, Arhgap32, and Ube3d) and 8 maternally expressed (Wdr27, Fam196b, Tbx3, Spp1, Igfbp3, LOC103691708, Itga1, and Cdc42ep1) genes were identified using conservative cutoffs (allelic RPM ≥0.5 on either allele in at least 2 replicates per cross in either tissue, $$n = 26$$,356 transcripts, Benjamini-Hochberg adjusted p-value <0.05, eBayes F-statistic, ≥4-fold change in expression between alleles) and are included in Fig. 1d (see Additional file 2: Table S1 for a full list of imprinted genes). *While* genes identified using Limma include the known mouse imprinted gene Slc38a4 and show clear parent-of-origin expression patterns, only Slc38a4 and LOC103691708 are associated with a DMR. We therefore focused subsequent analyses on genes that met our stringent statistical cutoffs, as defined above. Fig. 1Imprinted gene expression in rat embryonic and extraembryonic cells. a Experimental design. Two distinct reciprocal crosses of rat strains (BN/CrlCrlj, “B” and WKY/NCrlCrlj, “W” and F344/NSlc, “F”) were conducted and cells from the E8.5 epiblast (Epi) and the ectoplacental cone (EPC) were collected. RNAseq was performed on all samples (in duplicate or triplicate), and WGBS was performed in duplicate on BW/WB matings. The maternal (red) strain is listed first in cross names. b Scatterplot of paternal expression ratios in rat EPCs. The paternal expression ratio was averaged over 11 samples, and expressed transcripts (RPKM ≥1) with sufficient allelic coverage (RPM ≥0.5) in at least 6 samples are shown ($$n = 16$$,642). Transcripts showing parent-of-origin imprinted gene expression (Student’s t test, Bonferroni-adjusted p-val <0.05) are colored red (maternally expressed) or blue (paternally expressed). c Rat genome browser screenshots of the maternally expressed imprinted gene H19 and paternally expressed imprinted gene Igf2. For each cross, duplicates or triplicates were merged and the mean expression level is displayed in reads per million (RPM). A subset of all read alignments (gray) are highlighted if they originated from a maternal (red) or paternal (blue) alleles. The genomic position of known *Refseq* genes and CpG islands (CGIs) are included. d Ideogram karyotype summary of imprinted gene expression identified using a combination of T-test and Limma in rat Epi and EPC cells. Genes that show imprinted gene expression in human or mouse are indicated with an asterisk and tilde, respectively. LOC103691708, Itga1, and genes showing maternal imprinted expression uniquely in EPC cells and normally expressed in adult rat blood (RPKM ≥1) are not shown due to space limitations (see Additional file 2: Table S1 for a full list of imprinted genes) Consistent with previous findings in mouse and human [36], putative rat canonical imprinted genes show parent-specific expression in epiblast and EPCs and are positioned in clusters such as H19/Igf2, Trpm5/Tssc4/Alsc2/Cd81, and Peg10/Sgce (Fig. 1d and Table 1). In contrast, putative non-canonical imprinted genes such as Sfmbt2, Gab1, and Sall1 show paternal-specific expression exclusively in EPCs. Thus, allele-specific expression analysis of epiblast and EPCs confirmed the classification of known canonical and non-canonical imprinted genes. Furthermore, we identified 8 novel imprinted genes in the rat, including Zfp516, Slc38a1, Zfp64, Gsto1, Rpl39l, Syt16AS, Gadl1-3’UTR, and LOC108350526. As these genes have not been reported to be imprinted in other mammals, we chose to characterize them in greater detail, as described below. Table 1List of canonically imprinted genes identified in the rat and mouseGenomic lociChr.gDMREpi-DMREPC-DMRIdentified imprinted genesDMR type in miceImprinting in non-rodent mammalscH19-Igf21PatPatPatIgf2, H19PatHuman, cow, pig, opossum, wallabyDlk1-Meg36PatN.I.N.I.Rtl1AS, Meg3, miRNA clusterPatHuman, cow, pigRasgrf18PatPatPatRasgrf1bPatPigPeg31MatMatMatPeg3bMatHuman, cow, pigSnrpn1MatMatMatSnrpn, Snurf, Magel2MatHuman, cow, pigInpp5f1MatN.I.N.I.MatHuman, cow, pigKcnq1ot1a1MatN.I.N.I.Kcnq1ot1, Trpm5, Tssc4, Ascl2, Cd81MatHuman, cow, pigAirn-Igf2r1MatN.I.N.I.Igf2rbMatHuman, cow, pig, opossum, wallabyPlagl11MatMatMatMatHuman, cow, pigMcts23MatN.I.N.I.MatHumanNnat3MatN.I.N.I.MatHuman, cow, pigGnas (Nespas-GnasXL)3MatN.I.N.I.GnasMatHuman, cow, pigGnas (exon 1A)3MatMatMatGnasMatHumanSgce-Peg104MatMatMatSgce, Peg10MatHuman, cow, pig, wallabyMest4MatMatMatMatHuman, cow, opossum, wallabyNap1l54MatN.I.N.I.MatHuman, cow, pigTrappc97MatMatMatMatHumanSlc38a47MatN.I.N.I.Slc38a4MatPigZdbf29MatN.I.N.I.Zdbf2Mat (transient)HumanFkbp612MatN.I.N.I.MatGrb1014MatMatMatGrb10MatHumanZrsr114MatMatMatZrsr1MatImpact18MatMatMatImpactbMatCowCdh1519MatN.I.MatMat (transient)Pat Paternally methylated, Mat Maternally methylated, N.I. Not informativeaNot annotated in ratbThese genes are reported to be imprinted (expressed) in rats in previous studies [11]cOne or more genes in the orthologous imprinted cluster have been reported to be imprinted in human, cow (and pig), or marsupials [11, 35, 37–44]
## Identification of canonical differentially methylated regions in the rat
Having identified known and novel imprinted genes in rat based on allele-specific expression patterns, we next sought to locate candidate regulatory regions responsible for their parent-of-origin transcriptional regulation. Towards this end, we conducted WGBS on the same cell types and used MEA to generate parent-of-origin-specific methylomes. To identify the complement of DMRs, including imprinted regions, we used DSS [45, 46] and stringent statistical significance (p-adj <0.001) but lenient DNAme difference (delta >$10\%$) filters and identified 45,119 and 9165 DMRs in EPC and epiblast samples, respectively (Additional file 3: Table S2). Notably, while the vast majority of DMRs found in EPCs are maternally methylated ($$n = 40$$,427, $90\%$), a roughly equal number of paternal and maternal DMRs were identified in epiblast samples (maternally methylated $$n = 4386$$, $48\%$), which may reflect potential maternal somatic cell contamination or incomplete resetting of parental DNAme levels in EPC samples. To identify imprinted DMRs, we subsequently analyzed our previously generated rat oocyte and sperm WGBS datasets [47] using stringent parameters [47] (see Additional file 4: Table S3 for the complete list of data generated and analyzed in this study). Integrated analysis of DNAme levels revealed a total of 5876 regions that show dramatic DNAme differences (delta ≥$50\%$) between gametes and the parental alleles of epiblast or EPCs (Additional file 1: Fig. S1b & Additional file 3: Table S2). While $56\%$ ($\frac{270}{484}$) of DMRs in epiblasts were apparently established in gametes, $50\%$ (2,$\frac{749}{5}$,462) of DMRs in EPCs were hypermethylated in both gametes. Importantly, 45 DMRs are shared between all samples (Additional file 1: Fig. S1c), a subset of which overlap CpG islands (CGIs) located near known imprinted genes, such as the Peg3:CGI-promoter, Commd1/Zrsr1:CGI-promoter (Additional file 1: Fig. S1d), Grb10:CGI-promoter, Impact:CGI-intragenic, Mest:CGI-promoter, Snrpn:promoter, H19:promoter, and Kcnq1:intragenic. Notably, DMRs were also identified near the promoters of novel candidate rat-specific imprinted genes such as Zfp516, Zfp64, and Syt16-AS (Additional file 1: Fig. S1e). These results validate our agnostic approach for discovering DMRs between parental alleles of F1 hybrid rats, and reaffirm that canonical imprinting of the previously described imprinted genes listed above likely arose in a common ancestor over 90 million years ago.
## Evolutionary conservation of imprinting in rat and mouse
We next wished to determine whether any of the candidate novel imprinted genes identified in the rat are imprinted in the mouse, using the same genome-wide approach used for the rat. *We* generated RNAseq and WGBS data from reciprocal crosses of the well characterized mouse strains C57BL/6N and JF1/Ms, hereafter referred to as “C” and “J”, respectively (Additional file 1: Fig. S2a). Embryos were again dissected at Carnegie stage 7, corresponding to E7.25 in mouse. Rat embryos at this stage were larger and more elongated relative to mouse (Additional file 1: Fig. S2b-c), as previously observed [48, 49]. Analysis of allele-agnostic levels of transcript expression revealed a high concordance between replicates and reciprocal crosses (Spearman correlation >0.93) in both rat and mouse (Additional file 1: Fig. S2d-e), indicating these data are highly reproducible. Using the Ensembl (Biomart) homologous gene annotation and hierarchical clustering of gene expression levels in rat and mouse showed greater variation between epiblast and EPC cell types in each species than inter-species variation of the same cell type (Additional file 1: Fig. S2f). These observations reveal that epiblast and EPC transcriptional programs are distinct and strongly conserved between rat and mouse and confirm that rat and mouse samples were indeed collected at roughly matching developmental stages. Importantly, known imprinted genes such as H19 and Igf2 showed allele-specific expression and nearby DMRs in mouse epiblast and EPCs, as expected (Additional file 1: Fig. S3a-c). Analysis of mouse F1 hybrid methylomes also uncovered an overrepresentation of maternally methylated DMRs in EPC samples (Additional file 1: Fig. S3d-e & Additional file 3: Table S2). Integrated analysis with mouse gamete WGBS [50, 51] uncovered 452 DMRs shared between all samples (Additional file 1: Fig. S3e), 18 of which overlap the 20 known gametic DMRs (Additional file 3: Table S2). Together, these data provide a rich resource to investigate canonical and non-canonical genomic imprinting conservation between two species that diverged only ~13 million years ago [52].
Integrated analysis of matching expression and DNAme data enabled the unambiguous classification of rat genes into two categories: canonical and non-canonical imprinted genes, the latter category being defined by a lack of a germline DMR and imprinted expression restricted to the paternal allele in extraembryonic cells [12]. To measure conservation of imprinted gene expression, as well as allele-specific methylation over nearby DMRs, we compared parent-of-origin gene expression and DNAme levels between rat and mouse homologous genes. Twenty-one known imprinted genes showed imprinted expression in both species, including 18 canonical genes (Snrpn, Peg10, Igf2, Zrsr1, Sgce, Magel2, Zdbf2, Tssc4, Cd81, Ascl2, Trpm5, Mir675, H19, Grb10, Maged2, Meg3, Rtl1AS, and Gnas) (Fig. 2a), many of which are adjacent to gametic DMRs in both species (Fig. 2b), and 3 non-canonical genes (Sfmbt2, Sall1 and Gab1), each with nearby EPC-specific DMRs. The conserved imprinting status of Snrpn, Magel2, Peg10, Sgce, Igf2, Zdbf2, Mir675, H19, Grb10, Meg3, Rtl1AS, and Gnas was expected, as they are also imprinted in humans, consistent with an origin of imprinting in an ancient common ancestor [53–55].Fig. 2Conservation of genomic imprinting in rat and mouse. a Heatmap of parental expression ratios in rat and mouse Epi and EPC cells. Genes previously identified as imprinted in human are indicated with an asterisk. b Heatmap of allele-specific DNAme levels over DMRs associated with imprinted genes in a. The relative position of DMRs is included. DMRs that also show parent-specific methylation in human are indicated with an asterisk. c–d Rat and mouse genome browser screenshots of the Kcnq1 locus. Figure legend as in Fig. 1c, with the addition of DNA methylation information. The average DNAme level over each CpG is shown as individual bar plots, and CpGs covered by at least 1 allele-specific read (Epi, EPC) or 5 reads (gametes) are shown. Pink indicates methylation of both alleles. Rat DMRs are included. In rat, a paternally expressed unannotated antisense ncRNA is expressed from the gene body of Kcnq1. The position of the putative Kcnq1ot1 imprinted CGI promoter is indicated by a dashed box. d The mouse Kcnq1 locus is shown for comparison Among the canonically imprinted genes identified in rat, the growth factor receptor binding gene Grb10 shows maternal-biased expression in the epiblast and EPC and is associated with a maternally methylated DMR established in the oocyte (Fig. 2a-b & Additional file 1: Fig. S4a). The evolutionary conservation of maternal expression of Grb10 is of particular interest as this gene shows maternal expression in mouse [56], confirmed here (Grb10:CGI promoter), and isoform- and tissue-specific imprinting in human [57]. While there is only one annotated isoform of Grb10 in rat, we find that conservation of oocyte-specific DNAme at the CGI promoter is maintained in embryonic and extraembryonic tissues associated with maternal-specific expression in both cell types. De novo transcriptome assembly, described in detail below, clearly revealed two major Grb10 isoforms in the rat, both of which are transcribed from the maternal allele in epiblast and EPCs (Additional file 1: Fig. S4a), consistent with what is observed in mouse.
Another canonically imprinted gene, the CCCH zinc finger gene Zrsr1, is paternally expressed in association with a gametic maternally methylated DMR in mouse [58], confirmed here (Zrsr1:CGI promoter) (Fig. 2a,b, Additional file 1: Fig. S4b). The *Zrsr1* gene is intronic and antisense to Commd1, a maternally expressed gene, and the maternally methylated Zrsr1:CGI promoter DMR is established in oocytes via Commd1 transcription in mouse [59]. Of note, mouse, rat, and human all express Commd1 at high (RPKM>15) levels in oocytes (Additional file 1: Fig. S4c). However, Commd1 is biallelically expressed in human and is distal to Zrsr1 (Additional file 1: Fig. S4d). Here, we show that Zrsr1 is paternally expressed in association with a maternally methylated Zrsr1:CGI promoter DMR in both mouse and rat (Fig. 2a,b). Thus, transcription of Commd1 in oocytes, coupled with the proximal positioning in murids of Zrsr1 and its CGI promoter, likely potentiated the co-evolution of Zrsr1 and Commd1 imprinting in this lineage.
Surprisingly, another canonically imprinted transcript in mouse and human, Kcnq1ot1, was not identified in our allele-specific analysis. Kcnq1ot1 is a paternally expressed non-coding gene that is associated with silencing of adjacent genes, including Kcnq1, Ascl2, and Cd81 [53–55], resulting in their maternal-specific expression. Despite the apparent lack of Kcnq1ot1 in the rat, we confirmed maternal-specific expression of four orthologous imprinted genes: Ascl2, Cd81, Trpm5, and Tssc4 (Fig. 2a), prompting us to explore this genomic region in greater detail. Manual inspection revealed that Kcnq1ot1 is not annotated in the rat reference genome “rn6” and thus was not assayed by our pipeline. However, there was a clear paternal-specific RNAseq signal in the gene body of Kcnq1, as in mouse (Fig. 2c,d). To define Kcnq1ot1 and potentially other unannotated imprinted genes in the rat, we performed de novo transcriptome assembly of rat epiblast and EPC RNAseq data, agnostic to allelic assignment. This analysis uncovered an antisense transcript within the gene body of rat Kcnq1 (Fig. 2c). Importantly, this transcript, which shares $87\%$ homology with mouse, is transcribed exclusively from the paternal allele in rat epiblast and EPCs (Fig. 2a,c). Furthermore, mirroring the mouse Kcnq1ot1:CGI promoter DMR (Fig. 2d), the CGI promoter of the putative rat Kcnq1ot1 transcript is hypomethylated in sperm and hypermethylated in oocytes, and maintenance of parental DNAme levels is observed in epiblast and EPCs (Fig 2b,c). Indeed, agnostic identification of all DMRs between parental alleles identified the rat Kcnq1ot1 CGI promoter in both epiblast and EPCs (Fig. 2c). The recently released “rn7” rat reference genome includes a novel 32-kb ncRNA identified by the automated NCBI Eukaryotic Genome Annotation Pipeline as LOC120099961 and manual inspection confirms this novel transcript is likely the orthologue of mouse Kcnq1ot1. Thus, in addition to in silico prediction approaches, allele-specific RNAseq and WGBS can be leveraged to identify imprinted ncRNAs that are not annotated in the reference genome.
We next wished to determine whether non-canonical imprinted gene expression is evolutionarily conserved [22]. Focusing on EPC-specific imprints, we found that three genes previously identified as imprinted in mouse: Sfmbt2, Gab1, and Sall1 [12, 15, 17, 18], show paternal expression (Fig. 2a) as well as an associated maternally methylated DMR (Fig. 2b) in rat and mouse EPCs. A fourth imprinted gene that is regulated by both maternal H3K27me3 and DNAme in mouse, Slc38a4 [18], shows $81\%$ paternal allele expression in rat but did not pass our stringent statistical cutoffs due to variability between replicates (Additional file 2: Table S1). Notably, the bobby sox homolog gene Bbx, one of the transiently imprinted genes associated with maternal H3K27me3 in mice [12, 13, 60], is clearly biallelically expressed in rat (Additional file 2: Table S1). Other non-canonical imprinted genes in mouse could not be assessed for allele-specific expression in rat due to a lack of parental genetic variation (Jade1 and Smoc1) or the absence of an annotated ortholog (Platr4 and Platr20). Interestingly, none of the three non-canonical imprinted genes identified in rat are imprinted in human or macaques [61]. Together, these data indicate that the establishment of non-canonical imprints at Sfmbt2, Gab1, and Sall1 likely originated in the rodent lineage. Furthermore, the maintenance of their imprinting status in both mouse and rat indicates that their dosage likely plays an important role in extraembryonic development, at least in murids.
## De novo identification of genomic imprinting in rat
While the application of cross-species comparisons of allele-specific expression and DNAme data led to the identification of 21 orthologous genes imprinted in rat and mouse, two factors constrain the comprehensive identification of imprinted genes using this approach. Firstly, the *Ensembl* gene annotation, which we relied on to analyze orthologous genes, includes many genes in the mouse with no apparent ortholog in the rat, including known imprinted genes, such as Kcnq1ot1 (Fig. 2c). If such orthologs are actually present in the rat genome, our reliance on Ensembl annotations precludes the comprehensive characterization of imprinting in the rat. Secondly, due to natural divergence, the rat may truly lack genes that are orthologs to those present in the mouse (such as Platr4), and vice versa. If so, novel imprinted genes in the latter category would not be identified using the approach described above.
To circumvent the shortcomings associated with identifying novel imprinted genes using homologous gene annotations ($$n = 16$$,770 genes), we chose to use NCBI RefSeq transcript ($$n = 69$$,157 transcripts) annotations to calculate total and parental genome-specific transcript expression levels, as reported above (Fig. 1). Additionally, since imprinted gene expression may arise from unannotated promoters [15], we supplemented the NCBI *Refseq* gene annotations with de novo transcriptome assembly using our epiblast and EPC RNAseq samples ($$n = 3242$$ additional transcripts). Using these new rat transcript annotations and employing the same statistical tests and filtering criteria described above, we identified 8 novel putative imprinted genes in the rat that do not show imprinting in the mouse in our data or in mouse imprinted gene databases (Fig. 3a & Additional file 1: Fig. S5a-b). While an additional set of 33 genes were scored as maternally expressed only in rat EPCs, we chose not to study these in greater detail because all were found to be highly expressed (RPKM>1 from either allele) in adult blood and thus may reflect an artifact of maternal decidua expression (Additional file 1: Fig. S5c-d, see Additional file 2: Table S1). Thus, all putative rat-specific imprinted genes identified here are imprinted in EPCs, in line with the observations in mouse and human that relative to embryonic or somatic tissues, the placenta exhibits a greater level of parent-of-origin-specific gene expression. Furthermore, as none of these 8 genes are imprinted in human, their imprinting likely arose after the rat and mouse lineages diverged. Fig. 3De novo genomic imprinting in rat. a,b Heatmaps of parental expression ratios and allele-specific DNAme levels in rat and mouse Epi and EPC cells as in Fig. 2a,b. *Imprinted* genes in rat that are not imprinted in mouse are shown. c,d Rat and mouse genome browser screenshots of the Zfp64 and LOC108350526 loci, two rat-specific imprinted genes. Browser tracks are as shown in Fig. 2c,d. The rat-specific transcript LOC108350526 and syntenic mouse region is highlighted in green. The location of LTR retrotransposons (RepeatMasker) is included Consistent with non-canonical imprinting, post-fertilization DMRs are evident at $\frac{7}{8}$ paternally expressed genes in rat EPCs, including LOC108350526, Zfp64, Zfp516, Slc38a1, Gadl1-3′UTR, Syt16AS, and Rpl39l. Another chromatin mark, such as maternal H3K27me3, may therefore be functioning as the imprint at these genes. As an example, the promoter of the novel rat gene LOC108350526 shows rat-specific hypomethylation in oocytes and sperm (Fig. 3b,c). Consistent with non-canonical imprinting, DNAme on the maternal allele is acquired post-fertilization specifically in rat EPCs, resulting in asymmetric parental DNAme levels in association with paternal-biased expression (Fig. 3b,c). In mouse, the syntenic region of the rat LOC108350526 promoter is hypermethylated in the gametes and remains hypermethylated on both alleles following fertilization (Fig. 3b,d). This transcript has no homolog in mouse or human, and its putative open reading frame does not encode any known protein domain, suggesting it is a novel non-canonical imprinted ncRNA that arose in the rat lineage. Interestingly, the nearby gene Zfp64 shows clear paternal expression exclusively in rat EPCs (Fig. 3a,c). The CGI promoter of Zfp64 is maternally methylated in gametes in both rat and mouse, yet this parental asymmetry is maintained only in rat EPCs, as the locus becomes hypomethylated in the rat epiblast and mouse embryo (Fig. 3b,d). Consistent with the loss of DNAme, TET1, a methyl-cytosine dioxygenase [62], is enriched at the Zfp64 CGI promoter in mouse ESCs (Additional file 1: Fig. S5e), perhaps explaining the loss of DNAme at this region in the mouse. The fact that LOC108350526 is located adjacent to Zfp64 makes this locus worthy of future investigations into the relationship between canonical and non-canonical imprinting (see “Discussion”). Taken together, these results indicate that canonical genomic imprinting generally shows a greater degree of conservation in mammals than does non-canonical imprinting, raising the question, what drives the establishment of species-specific non-canonical imprinting?
## Epigenetic profiling of rat oocytes
A common feature of the rat-specific imprinted genes identified here is paternal-specific transcription in extraembryonic tissue in the absence of a canonical DNA methylation imprint in oocytes. These are hallmark features of non-canonical imprinted genes in the mouse, which were recently shown to depend upon H3K27me3 deposited by PRC2 in oocytes [12, 13]. As noted previously, rat oocytes show relatively fewer hypermethylated loci than mouse oocytes (Additional file 1: Fig. S6a) [47], perhaps enabling more widespread H3K27me3 deposition. We therefore hypothesized that rat-specific non-canonical imprinted genes are associated with rat-specific H3K27me3 domains in oocytes. To determine whether H3K27me3 is indeed enriched at rat-specific non-canonical imprinted loci, we profiled fully grown rat oocytes (FGOs) by CUT&RUN [63]. Additionally, we profiled H3K4me3 and H3K36me3, which generally mark the promoter regions and gene bodies, respectively, of actively transcribed genes. Genome-wide analysis revealed strong correlation between biological replicates, as well as between H3K36me3 and DNAme levels, as shown previously in mouse oocytes (Additional file 1: Fig. S6b) [47]. Additionally, we observed an anticorrelation between H3K27me3 and H3K36me3 in rat oocytes (Fig. 4a), consistent with the observation that H3K36me3 inhibits PRC2 activity in vitro [64, 65] and is anti-correlated with H3K27me3 in vivo [66]. Chromatin state analysis using ChromHMM [67] revealed that H3K27me3 indeed marks a larger fraction of the genome in rat than mouse oocytes (Additional file 1: Fig. S6c). Conversely, H3K36me3 domains are more widespread in the mouse, likely reflecting a greater prevalence of non-genic transcripts in the mouse [47]. Analysis of recently published H3K27me3 CUT&RUN data from rat oocytes [68] reveals a strong correlation with our data (Spearman rank correlation 0.95, Additional file 1: Fig. S6d), confirming the reproducibility of this method. Additionally, rat oocyte and early embryo H3K27me3 levels are generally correlated (Spearman rank correlation >0.67, Additional file 1: Fig. S6d), indicating that H3K27me3 levels are largely maintained in the rat embryo. Fig. 4Species-specific H3K27me3 in oocytes is associated with species-specific non-canonical imprinting Zfp64 and Zfp516, and transient imprinting of Bbx. a 2D scatterplot showing genome-wide H3K36me3 and H3K27me3 levels over 1-kb bins in rat oocytes. Datapoints are colored by average oocyte DNAme levels. A random subset of 10,000 bins is shown. The density of all data points ($$n = 1$$,164,268) is summarized by a contour plot. b Cross-species Spearman correlation values between H3K4me3, H3K27me3, and H3K36me3, and DNAme levels over syntenic 1-kb bins in rat and mouse oocytes ($$n = 823$$,926). c Scatterplot showing differences in rat and mouse oocyte DNAme, H3K27me3, and H3K36me3 levels over syntenic 1-kb bins. A random set of 10,000 bins are shown, and all bins ($$n = 823$$,926) are summarized by a contour plot. The number of bins with a delta H3K27me3 >0.5 RPKM and DNAme >$75\%$ is indicated by a dashed box. d–f Rat and mouse genome browser screenshots of the Zfp64-LOC108350526, Zfp516, and Bbx loci. Replicates of CUT&RUN data were merged and mean levels are shown as counts per million aligned reads (CPM). Promoters are indicated by a dashed line To directly compare rat and mouse oocyte epigenomes, we first assessed the enrichment of H3K27me3, H3K36me3, H3K4me3, and DNAme levels across syntenic 1-kb regions, which revealed globally similar distributions of each of these marks (Fig. 4b). Indeed, genes expressed in rat and mouse oocytes show similar patterns of H3K4me3 and H3K36me3 over their promoters and gene bodies, respectively (Additional file 1: Fig. S7a). However, in mouse, the gene bodies of silent genes were also enriched for H3K4me3 (Additional file 1: Fig. S7a). Such “broad” H3K4me3 domains were previously identified in mouse oocytes [69–71] but were not observed in human oocytes [72], raising the question of the evolutionary origins and biological relevance of such domains. Rat and human H3K4me3 domains are on average 10kb ($37\%$) smaller than those in mouse (Additional file 1: Fig. S7b), including at the previously analyzed Kdm4c promoter (Additional file 1: Fig. S7c). These results indicate that, with respect to H3K4me3, rat oocytes may be more similar to human than mouse. Consistent with the ChromHMM results and the relatively restricted distribution of H3K4me3, species-specific H3K27me3-enriched regions were >3× more abundant in rat than mouse (as measured in syntenic 1-kb bins, Fig. 4c). Why the distribution of these chromatin marks differs so dramatically between these closely related species is an intriguing question that awaits further investigation.
We next focused on the epigenetic state of Gab1, Sall1, and Sfmbt2, the three non-canonical imprinted genes showing paternal allele-specific expression in both the rat and mouse (Fig. 2a). Interestingly, all maternally methylated DMRs associated with these loci overlap with an H3K27me3 domain in oocytes of both murids (Additional file 1: Fig. S8a), confirming that non-canonical imprinting of these genes is conserved. In line with our findings, H3K27me3 is clearly maintained at these loci to the blastocyst stage in mouse and rat (Additional file 1: Fig. S8a). As recent reports have indicated that G9A and GLP play an important role in establishment [17] and maintenance [17, 18] of non-canonical imprinting, we analyzed our recent mouse oocyte H3K9me2 ChIP-seq data [73] and found that Gab1, Sfmbt2, and Sall1 also overlap an H3K9me2 domain (Additional file 1: Fig. S8b). Taken together, these data are consistent with the model that PRC2 and/or G9A/GLP play an important role in the establishment of non-canonical imprints.
Unlike Gab1, Sall1, and Sfmbt2, the 8 genes mentioned above (Fig. 3a,b) are not imprinted in mouse, indicating that whatever the mechanism of imprinting establishment at these loci, it occurred relatively recently (<13 million years ago) in the rat lineage. Intriguingly, only LOC108350526 showed relatively higher levels of H3K27me3 and concomitant lower levels of H3K36me3 and DNAme in rat compared to mouse oocytes (Additional file 1: Fig. S9a). Indeed, a rat-specific H3K27me3 domain that overlaps with the promoter of LOC108350526 is clearly present in rat but absent over syntenic regions in mouse oocytes (Fig. 4d). Furthermore, while a maternal H3K27me3 domain exists at the Zfp516 locus in both rat and mouse oocytes, the mouse H3K27me3 domain is adjacent to the Zfp516 promoter and covers a region 2–3× smaller than in the rat (Fig. 4e). More specifically, the mouse H3K27me3 domain terminates at the gene body of Zfp516, perhaps as a consequence of transcription-coupled H3K36me3 deposition in the mouse oocyte. Indeed, Zfp516 is expressed in mouse (RPKM = 3.07) but not rat (RPKM < 0.00) oocytes (Additional file 1: Fig. S9b). Thus, differential expression of Zfp516 in rat versus mouse oocytes correlates with a rat-specific H3K27me3 domain, which in turn may regulate imprinted expression in rat EPCs. Indeed, both LOC108350526 and Zfp516 maintain H3K27me3 levels to the blastocyst stage in rat, demonstrating that these are bona fide non-canonical imprinted genes (Additional file 1: Fig. S9c). Conversely, *Bbx is* a transiently imprinted gene associated with maternal H3K27me3 in mouse [12, 13, 60] that is biallelically expressed in rat (Additional file 2: Table S1). Comparison of the rat and mouse *Bbx locus* clearly reveals a mouse-specific H3K27me3 domain that overlaps with the CGI promoter in oocytes (Fig. 4f). Taken together, these observations indicate that the species-specific imprinting of a subset of genes in the mouse or rat may be explained by the species-specific targeting of PRC2 in oocytes. Analysis of the remaining rat-specific non-canonical imprinted genes did not reveal a difference in H3K27me3 levels in oocytes, indicating that alternative epigenetic marks, and/or timing of H3K27me3 removal, is likely at play. For example, differences in H3K27me3 maintenance in early rat and mouse embryos could give rise to species-specific imprinting. However, comparing rat and mouse preimplantation embryo H3K27me3 data did not uncover any clear species-specific maintenance over the 8 rat-specific non-canonical imprinted genes identified here. Only one such gene, Gsto1, displayed partial loss of H3K27me3 levels at the 2C stage in mouse compared to robust maintenance to the blastocyst in rat (Additional file 1: Fig. S9c). Additional investigations into the dynamics of H3K27me3, H3K9me2, and other repressive marks in mammalian embryos, combined with functional studies, will hopefully provide insights into the relative importance of such repressive covalent histone marks in the establishment and maintenance of extraembryonic non-canonical imprinting.
Genetic differences can also contribute to species-specific imprinting. The Slc38a1 gene, which encodes an amino acid transporter, is another putative rat-specific non-canonical imprinted gene (Fig. 3a). As an H3K27me3 domain overlaps the Slc38a1 promoter in both rat and mouse oocytes and early embryos (Additional file 1: Fig. S9c), H3K27me3 per se is apparently not sufficient to confer non-canonical imprinting. Analysis of DNAme levels on the other hand reveals a rat-specific maternally methylated DMR (Slc38a1:intragenic ERVL/K) in EPCs (Fig. 3b) that overlaps with an MTD retroelement (Fig. 5a,b). Analysis of sequence synteny uncovered two key aspects that differentiate the rat Slc38a1 locus from that in the mouse. First, while the underlying intronic DMR sequence is present in mouse, only the rat gene harbors a ZFP57 binding motif in the differentially methylated MTD retroelement (Fig. 5b). Analysis of the MTD consensus sequence [74] reveals that the rat MTD element gained two substitutions, resulting in the genesis of a ZFP57 motif (GCAGCG --> GCGGCA). Interestingly, $\frac{7}{39}$ distinct rat strains have a substitution (GCGACA) “away” from the MTD consensus that abrogates the ZFP57 motif [75]. While the homologous sequence in mouse (GCAGCA) shares one substitution with the rat, this sequence does not conform to a ZFP57 motif. No further substitutions occurred over those 6 bases in any of the 36 sequenced mouse strains [76], including JF1 [77]. Secondly, the canonical CGI promoter of Slc38a1 contains a rat-specific ERVK promoter insertion (Fig. 5b). Whether either of these features are causal for the establishment of non-canonical imprinting remains to be tested, perhaps by analyzing rat strains that subsequently lost the ZFP57 motif. As maintenance of differential gametic DNAme at loci showing canonical imprinting is dependent upon ZFP57/ZFP445 binding, it is tempting to speculate that the novel ZFP57 motif in the intron of Slc38a1 plays an essential role in its non-canonical imprinting. Since rat gametes are unmethylated at the novel ZFP57 motif, an allele-specific gain of maternal DNAme must occur post-fertilization, perhaps in conjunction with loss of maternal H3K27me3 [14]. Further studies aimed at understanding how maternal H3K27me3 established in oocytes is replaced by maternal-specific methylation in extraembryonic tissues will deepen our understanding of the relationship between these repressive epigenetic marks and their roles in non-canonical imprinting. Fig. 5Rat-specific non-canonical imprinted gene Slc38a1. a Rat and mouse genome browser screenshots of the Slc38a1 locus. Slc38a1 is paternally expressed in rat and biallelically expressed in mouse EPCs. Tracks are presented as in Fig. 2d. b Ensembl Region Comparison screenshot of the Slc38a1 locus in rat (rn6), mouse (mm10), and human (hg19). Syntenic regions are shown in green. The locations *Refseq* genes, LTRs, CGIs (purple), and ZFP57 binding motifs (orange) are shown. Rat EPC DMRs are included. The CGI promoter and intronic DMR are highlighted in yellow and magnified (right panel) for clarity
## Alternate promoters of murid-specific non-canonical imprinted genes
Interestingly, maternally methylated DMRs that arise following fertilization frequently overlap alternative promoters derived from ancient insertions of endogenous retroviruses [15, 17], which harbor strong promoters in their long terminal repeats (LTRs). To assess the potential evolutionary conservation of LTRs in promoting imprinted gene expression, we analyzed syntenic regions between rat, mouse, and human. For example, the non-canonical imprinted gene Gab1 is expressed from the paternal allele in rat and mouse EPCs, in association with a maternal H3K27me3 imprint and an EPC-specific DMR (Additional file 1: Fig. S10a). Importantly, the Gab1 DMR overlaps an alternate ERVK LTR promoter that drives paternal-specific expression of Gab1 in both species (Additional file 1: Fig. S10b). Notably, this ERVK LTR is absent from the human genome, implicating the insertion of this ERVK element in the rodent lineage in the provenance of imprinted expression of Gab1. Similar to Gab1, a maternal H3K27me3 domain also overlaps the Sall1 promoter in rat and mouse oocytes (Additional file 1: Fig. S10c), and an upstream ERVK promoter present in rat and mouse but absent in the orthologous human locus overlaps an extraembryonic DMR and drives paternal-specific expression of Sall1 in EPCs (Additional file 1: Fig. S10d). Thus, in Muridae, both genes are likely governed by an LTR-dependent non-canonical imprinting mechanism. While Sall1 was predicted to be imprinted in human using in silico approaches [78], no empirical data has yet been produced to support this conjecture. In contrast, while Sfmbt2 shows imprinted gene expression in mouse and rat EPCs and overlaps a maternally deposited H3K27me3 domain (Additional file 1: Fig. S10e), the promoter of this gene does not overlap an annotated LTR element (Additional file 1: Fig. S10f). As five gaps exist in the rat reference genome between the Sfmbt2 CGI promoter and the putative RLTR11B ERVK alternate promoter identified in mouse [15], we cannot rule out the possibility that an alternate promoter embedded in this region contributes to the imprinted expression of Sfmbt2 in the rat. Taken together, these results are consistent with the model that alternate promoters, particularly those provided by LTRs, play an important role in promoting non-canonical imprinted expression of genes showing murid-specific imprints, and that maternally deposited H3K27me3 likely serves as the gametically inherited genomic imprint at many of these loci.
## Imprinted X chromosome inactivation in rat
X chromosome inactivation (XCI) involves the random and widespread silencing of genes on one X chromosome in adult female cells. During mouse embryogenesis however, XCI is not always random, with the fidelity of XCI influenced by genetic background. Indeed strain-specific XCI is observed across most adult mouse tissues [79]. Furthermore, placental cells preferentially inactivate the paternally inherited X chromosome in mice [25, 80], likely due to non-canonical imprinting of Xist [12, 13, 81]. Interestingly, a recent study showed that H3K27me3 is enriched at the *Xist* gene in rat oocytes and early embryos [68], consistent with cytological evidence of non-random XCI in yolk sacs [82] and our CUT&RUN data (Fig. 6a). To determine whether either mode of XCI skewing occur in rat, we calculated the mean paternal expression ratio of all autosomal and X-linked transcripts (Fig. 6b). Rat epiblast cells do not show XCI skewing, suggesting that rat lab strains have not undergone sufficient sequence divergence to favor allelic bias in XCI. However, in line with findings in mouse, the paternally inherited X chromosome is preferentially inactivated in EPCs. Thus, imprinted XCI in the trophoblast lineage is conserved in rat and mouse. Fig. 6X chromosome inactivation (XCI) in rat and mouse. a Rat and mouse genome browser screenshots of the Xist locus. *Refseq* genes and CGIs are included. CUT&Run (rat) and ChIP-seq (mouse) data are shown in counts per million (CPM). Xist is highlighted in green. b Violin plots showing the distribution of paternal expression ratios of autosomal (top) and X chromosome (bottom) transcripts in rat and mouse epiblast and EPC samples. Replicate 1 of mouse cross “CJ” was omitted due to an XO genotype. Differences in XCI skewing between CJ and JC samples (bottom right) are likely due to differences in genome quality between the parental strains, with reads preferentially aligning to the reference “C” genome
## Discussion
The rat is an increasingly utilized model organism. However, despite the relevance of imprinted gene regulation in development, behavior, and physiology, few studies have focused on genomic imprinting in this species. Indeed, only 13 imprinted genes have been described in the rat thus far, all of which were previously identified in mouse or human [11, 34, 68, 83]. Here, employing a genome-wide approach, we expand this list to 19 canonical (Table 1, including the unannotated Kcnq1ot1) and 11 non-canonical (Table 2) imprinted genes. We further characterize the epigenetic marks present at these loci in oocytes as well as early embryonic tissues, and their association with parent-of-origin regulated gene expression, thereby providing an unprecedented view of the rat imprintome and the potential mechanisms by which imprinting is established. However, comprehensive assessment of allele-specific gene expression and DNAme levels depends on naturally occurring genetic variation between parental genomes, which are absent at some regions in the three distinct rat strains used here. Therefore, the full catalog imprinting in the rat will require additional studies. Further, while we investigated the embryonic precursor cells to the placenta and the adult soma, some genes are only imprinted in specific adult tissues. In addition, extraembryonic samples are frequently contaminated with maternal decidua, making it difficult to discriminate between maternally expressed imprinted genes in our EPC samples from genes that are normally expressed in adult blood. Finally, the rat reference genome has many incomplete chromosome assemblies and gene model annotations. All of these limitations can be rectified in future studies that profile additional tissues generated from distinct rat strain crosses using an updated reference genome. Table 2Non-canonical imprinted genes and associated DMRs in the ratGeneFull nameChr. EPC-DMR (Overlapping LTR-LTR family)Putative mechanism underlying specie-specific imprintingGene ontology (process)NoteGsto1Glutathione S-Transferase Omega 1chr1N.I.UnknownGlutathione metabolic processZfp64Zinc finger protein 64chr3MatOocyte epigenome (DNAme vs H3K27me3)Mesenchymal cell differentiationLOC108350526chr3Mat (RMER13A/RLTR20B3-ERVK)Oocyte epigenome (DNAme vs H3K27me3)UnknownSyt16AS (LOC102550910)Synaptotagmin 16 - antisensechr6Mat (MTC-ERVL)UnknownExocytosis (Syt16)Gadl1-3'UTRGlutamate Decarboxylase Like 1chr8Mat (RMER3B-ERVK)UnknownCarboxylic acid metabolic process (Gadl1)Slc38a1Solute Carrier Family 38 Member 1chr7Mat (RMER6B/ORR1A4/MTD-ERVK/ERVL)LTR insertion and ZFP57 motifsAmino acid transmembrane transportRpl39lRibosomal Protein L39 Likechr10MatUnknownUnknownSfmbt2Scm like with four mbt domains 2chr17MatCommon imprinting between rats and miceNegative regulation of gene expressionSfmbt2 is essential for trophoblast maintenance and placenta development in mice [84, 85].Zfp516Zinc finger protein 516chr18MatOocyte epigenome (DNAme vs H3K27me3)Brown fat cell differentiationSall1Spalt Like Transcription Factor 1chr19Mat (RLTR20B3-ERVK)Common imprinting between rats and miceHeart/kidney/limb developmentGab1GRB2 Associated Binding Protein 1chr19Mat (RLTR15-ERVK)Common imprinting between rats and miceMAPK cascade/AngiogenesisGab1 is essential for development of placenta (syncytiotrophoblast) in mice [20].
Comparative analyses of imprinting in mice and humans indicate that the majority of canonically imprinted genes are conserved between rodents and primates, which diverged ~90 million years ago [86]. In contrast, we report the non-canonical imprinting of several genes in Muridae, including Sfmbt2, Gab1, and Sall1, that were not identified in recent screens of imprinting in humans [87] or macaques [61]. While we identified H3K27me3 deposited in oocytes as a putative imprint of these three genes, further studies are required to determine whether H3K27me3 is essential for the establishment of their imprinting in rat. In addition to PRC2, G9A and GLP were recently implicated in the establishment [17] and maintenance [17, 18] of non-canonical imprinting in mouse. Indeed, imprinted expression of Gab1, Sfmbt2, and Sall1 was shown to depend on both H3K9me2 and H3K27me3 pathways and all three genes overlap an H3K9me2 domain in oocytes [73]. While G9A has low H3K27 methyltransferase activity in vitro [88], and ancestral SET domain proteins in Paramecium show both H3K9 and H3K27 methyltransferase activity in vivo [89], experiments on the molecular basis of non-canonical imprinting revealed that imprinted expression is disrupted in Eed maternal knockout embryos, coincident with loss of maternal H3K27me3 [13]. Taken together, these observations indicate that H3K9 and H3K27 methylation are likely deposited by independent pathways and function in a non-redundant manner. Whether establishment and maintenance of the 8 non-canonically imprinted genes unique to the rat are subject to the same imprinting pathways remains to be determined. Interestingly, recent studies profiling H3K27me3 levels in human, macaque, cow, pig, rat, and mouse embryos demonstrated that this mark is maintained following fertilization only in rat and mouse [61, 68]. Whether an alternate repressive pathway, such as G9A and GLP, compensates for the maintained repression of the maternal allele of non-canonically imprinted genes in human, cow, and pig remains to be determined. Furthermore, whether the imprinted expression of rat-specific imprinted genes contribute to differences in rat and mouse physiology and/or behavior remains to be established. Regardless, as the placenta is a particularly rapidly evolving tissue [90, 91], we speculate that the rapid evolution of non-canonical imprinting may contribute to this process. Indeed, the idea that imprinting drove placental evolution and viviparity was the subject of a recent review [92]. Furthermore, ERVK elements, which frequently overlap non-canonical imprints [15, 17], are a class of retroviruses that have been implicated in the rapid evolution of the placenta [93].
Species-specific imprinted gene expression specifically in extraembryonic tissue supports the model that the developing placenta is at the nexus of ongoing “parental conflict” in gene dosage [23]. The most compelling evidence for the parental conflict hypothesis is based on functional analyses of imprinted genes in the mouse, which reveal that many such genes play a role in regulating embryo and offspring size [94]. Among the novel imprinted genes identified in the rat, Zfp516 (79 and $51\%$ paternal allele expression in EPC and epiblast cells, respectively) is of particular interest. The mouse orthologue is biallelically expressed ($51\%$ paternal allele expression) in epiblast and EPCs (Additional file 2: Table S1) and is essential for embryogenesis and brown adipose tissue (BAT) production [95]. Zfp516 binds and induces transcription of BAT genes, including Uncoupling Protein 1 (Ucp1), which in turn regulates a host of BAT-enriched genes [95] and is used as a biomarker for BAT in mouse and human [96]. This is consistent with other imprinted genes linked to adipogenesis and fat metabolism, including Mest, Klf14, Dlk1, H19, Peg3, Peg10, Grb10, and Cdkn1c [97]. Thus, the recent evolution of Zfp516 imprinted regulation in the rat lineage, wherein expression from the paternal allele promotes transcription and in turn brown fat production, including in the developing placenta, is consistent with the parental conflict theory.
The rat-specific imprinting status of Slc38a1 is also noteworthy as four other genes within the same gene family, Slc38a4, Slc22a18, Slc22a2, and Slc22a3, are imprinted in the mouse. All these genes encode related solute carriers, and Slc38a4 is highly expressed in the placenta, where nutrient transport is critical. Indeed, Slc38a4 knockout mice have a placental phenotype affecting embryonic growth [98]. Canonical imprinting of Slc38a4 in mice is dependent upon transcription in oocytes of a murid-specific retroelement that inserted upstream of the Slc38a4 CGI promoter [99], leading to de novo DNAme of the canonical DMR. Notably, Slc38a4 is paternally expressed in the mouse placenta [15, 100] and G9A-depleted embryos fail to maintain the maternally methylated DMR [101], suggesting that both canonical and non-canonical imprinting mechanisms govern Slc38a4 expression. Thus, in line with the observation that imprinted genes converge on shared pathways [10], Slc38a1 represents a strong candidate rat-specific non-canonical imprinted gene. Given the observed sequence-specific differences at the Slc38a1 CGI promoter and intronic DMR, investigation of the relative roles of retroelement transcription in EPCs and ZFP57 binding motifs in maintenance of post-fertilization gain of maternal DNAme will likely yield important insights into the evolution of non-canonical imprinting.
Slc38a1 and Zfp64, another rat-specific non-canonically imprinted gene, were recently identified as paternally expressed in association with maternal H3K27me3 in mouse blastocysts [60]. However, the asymmetry in parental expression was lost following implantation in mouse, which may explain why these genes were not identified in previous studies focusing on post-implantation tissues such as EPCs. Nevertheless, transient imprinting of mouse Slc38a1 and Zfp64 is in line with the observation that global levels of maternal H3K27me3 are maintained to the blastocyst stage in murids [68]. In other words, Slc38a1 and Zfp64 are seemingly “primed” for imprinting in rat and mouse, yet the maintenance of these imprints is specific to the rat. Further analysis of such transient (to the blastocyst stage) and non-canonical imprinted genes will serve as important models for studying the provenance of imprinting maintenance. Indeed, our meta-analysis of the Slc38a1 locus implicates a gain of a putative ZFP57 binding site and/or an insertion of an alternate ERVK promoter in the rat genome.
Zfp64 is a zinc finger protein that regulates mesenchymal differentiation via Notch signaling in mouse [102]. Additionally, ZFP64 directly promotes the expression of MLL in human leukemia [103]. While Zfp64 was found to be imprinted in the mouse placenta [104], a recent extensive search for imprinted genes revealed only modest (<$70\%$) paternal-biased expression in an array of mouse tissues [105]. Here, we confirm these findings and show that Zfp64 is indeed biallelically expressed in embryonic ($50\%$ paternal expression) and extraembryonic ($63\%$ paternal expression) mouse cells. Consistent with this observation, its CGI promoter is hypomethylated in both cell types. In contrast, the rat Zfp64:CGI promoter is maternally methylated in association with imprinted expression ($95\%$ paternal expression) in ectoplacental cone cells. The genomic imprint of the Zfp64:CGI promoter is conserved between rat and mouse, in association with active transcription of a murid-specific MTC LTR element in oocytes [47]. Notably, Zfp64 has not been reported as imprinted in human. However, the Zfp64 CGI promoter lacks ZFP57 binding motifs in both rat and mouse, and ZFP57 in mouse ESCs is not enriched at this region (data not shown), as determined by ChIP-seq [106]. Thus, rat-specific imprinting of Zfp64 cannot be explained simply by these sequence differences. However, species-specific demethylation of the Zfp64 CGI promoter could be due to differential activity of the TET enzymes at this locus. Consistent with this possibility, TET1 is enriched at the Zfp64 CGI promoter in mouse ESCs. TET1 profiling in mouse and rat embryos would confirm whether differential active demethylation can result in species-specific imprinting, and yield insights into ZFP57-independent modes of imprint maintenance.
In contrast to the genetic differences between rat and mouse Zfp64 and Slc38a1, rat-specific imprinting of Zfp516 and LOC108350526 is associated with domains of H3K27me3 in oocytes unique to the rat. While the molecular basis of PRC2 targeting in oocytes remains to be determined, H2AK119 ubiquitination is apparently not required for maintenance of H3K27me3 or imprinted expression of non-canonical imprinted genes after fertilization [107, 108]. Furthermore, although the mechanism underlying the switch from maternally inherited levels of H3K27me3 to maternal-specific dense DNA methylation in the extraembryonic lineage remains unknown [14, 15, 22], the paternally expressed gene Zinc Finger DBF-Type Containing 2 (Zdbf2) may offer insights. *This* gene is regulated by a maternally methylated somatic DMR that overlaps with a domain of H3K27me3 in oocytes and a H3K27me3-to-DNAme switch was shown to be induced by active transcription across the locus which, in turn, is promoted by an upstream paternally methylated gametic DMR [109]. Indeed, genetic ablation of both DNMT1 and EZH2 results in biallelic Zdbf2 expression in mouse zygotes [18]. Whether a similar phenomenon is at play at all non-canonical imprints that undergo H3K27me3-to-DNAme switching remains to be tested.
## Conclusions
In summary, our results provide the first atlas of genomic imprinting in the rat, including 22 conserved and 8 novel candidate rat-specific imprinted genes. Notably, several of the novel genes identified are imprinted specifically in placental precursor cells and have been previously shown in other species to play an important role in metabolism. Determining whether disruption of such non-canonical imprinting impacts fetal and/or placental growth is an important avenue for future investigation. Clearly, measuring the extent and conservation of imprinted gene expression in additional mammals, such as pig, cow, human, and macaques, complemented by interrogation of genetic and epigenetic differences at species-specific imprinted loci, will deepen our understanding of the evolution of genomic imprinting. Finally, there is an increasing body of evidence implicating canonical imprinted genes in complex behavioral traits [10, 110–114]. Our observation that canonical imprinted gene expression is highly conserved in the rat opens the door to future mechanistic studies on the role of these genes in behavioral and other traits in this increasingly tractable model.
## Rat and mouse embryo collection
E8.5 F1 hybrid rat embryos were isolated from reciprocally crossing BN/CrlCrlj (BN rat, Charles River) and WKY/NCrlCrlj (WKY rat, Charles River) strains as well as between strains BN/CrlCrlj and F344/NSlc (F344/N rat, Japan SLC). E7.25 F1 hybrid mouse embryos were isolated from reciprocally crossing C57BL/6N (Clea Japan) and JF1/Ms (National Institute of Genetics) strains. For both rat and mouse embryos, the ectoplacental cone (EPC) was dissected from the rest of the embryo before removing the visceral endoderm (VE). Finally, the epiblast (EB) was dissected from the extraembryonic ectoderm (ExE). Each ExE lysate was used for rapid genetic sex determination by direct PCR amplification using primers for mouse Zfy (5′-CCTATTGCATGGACTGCAGCTTATG-3′ and 5′-GACTAGACATGTCTTAACATCTGTCC-3′) and rat Sry (5′-CATCGAAGGGTTAAAGTGCCA-3′ and 5′-ATAGTGTGTAGGTTGTTGTCC-3′). Only female epiblasts and EPCs were subjected to strand-specific RNA sequencing and whole genome bisulphite sequencing (in the case of BN-F344/N hybrids, only RNA sequencing was performed).
## Rat oocyte collection
GV stage rat oocytes were collected from the ovaries of 3–4-week-old BrlHan:WIST (Clea Japan) rats 48–50 h after administration of PMSG. Briefly, cumulus-oocyte complexes were collected in PB1 with 200 μM IBMX (Sigma), and subsequently incubated in the presence of 5–10 μg/ml cytochalasin B for 5–10 min to loosen the connection between cumulus cells and oocytes. Then, the cumulus cells were mechanically removed by repeated pipetting using a glass capillary, and the zona pellucida were removed by brief incubation in acidic Tyrode’s solution.
## Strand-specific RNA library preparation and sequencing
One nanogram of embryonic total RNA was reverse transcribed using SMARTer Stranded Total RNASeq Kit v2 - Pico Input Mammalian (Takara Bio) according to the manufacturer’s protocols, in which Read 2 corresponds to the sense strand by the template-switching reactions. The RNAseq libraries were quantified by qPCR using KAPA Library Quantification Kit (Kapa Biosystems). All libraries were mixed and subjected to paired-end 100 bp sequencing (paired-end 101 nt reads in which the first 1 nt of Read 1 and the last 1 nt of Read 2 were trimmed) on HiSeq 2500 system (Illumina). For each library, reads were trimmed using Trimmomatic (v0.32) [115] to remove 2 nucleotides of the 5′ end of read 2.
## Whole genome bisulphite sequencing (WGBS)
One hundred nanograms of genomic DNA from JF1 (dam) and C57BL/6 (sire) mouse hybrids was prepared for Methyl-seq (Illumina) library construction. Twenty nanograms of genomic DNA from the other mouse and rat hybrid crosses was prepared for rPBAT or tPBAT library construction. Each WGBS library was quantified using KAPA Library Quantification Kit (Kapa Biosystems). The tPBAT libraries from BN (dam) and WKY (sire) hybrids and Methyl-seq libraries were subjected to paired-end 150 bp sequencing on HiSeq-X-Ten system (Illumina). The other tPBAT and rPBAT libraries were subjected to paired-end 100 nt sequencing (paired-end 101 nt reads in which the first 1 nt of Read 1 and the last 1 nt of Read 2 were trimmed) on the HiSeq 2500 system.
## CUT and RUN library preparation and sequencing
CUT&RUN libraries were prepared as previously described [13] with some modifications. Briefly, oocytes were incubated with rabbit anti-H3K27me3 antibody ($\frac{1}{100}$, Diagenode, #C15410069), rabbit anti-H3K4me3 antibody ($\frac{1}{100}$, Active Motif, #39159), or rabbit anti-H3K36me3 ($\frac{1}{100}$, Abcam, #9050) in a wash buffer (20 mM HEPES (pH7.5), 150 mM NaCl, 0.5 mM Spermidine, complete EDTA-free protease inhibitor cocktail, $0.02\%$ Digitonin containing 2 mM EDTA at 4°C on a plastic dish overnight. After the oocytes were washed twice by the wash buffer on a plastic dish, they were incubated with Protein A-MNase (500 ng/ml diluted in the wash buffer) provided by the Steven Henikoff Lab [63] at 4°C for 3 h on a plastic dish. After washing twice, the oocytes were transferred into a 1.5-ml DNA LoBind tube (Eppendorf) containing pre-activated BioMag Plus Concanavalin A (ConA) beads (Bang Laboratories, Inc.). The sequential step of the antibody incubation followed by the ConA binding prevents loss of oocytes during wash steps. After the oocytes were bound to ConA beads and the buffer was replaced to the wash buffer, CaCl2 was added at a final concentration of 2 mM to activate the MNase. After 20 min of digestion on ice, the digestion reaction was stopped, DNA fragments were recovered, and DNA libraries were prepared as described previously [13]. PCR amplification was performed for 11–15 cycles. Libraries were sequenced on a NextSeq500 with single-end 75 bp reads (Illumina).
## RNAseq read alignment and allele-agnostic quantification using a haploid reference genome
Sequencing reads were trimmed using Trimmomatic (v0.32) and the following parameters: SLIDINGWINDOW:3:10 MINLEN:36 ILLUMINACLIP:TruSeq2-3-PE.fa:2:30:10. Read pairs that survived trimming were aligned to genome build mm10 or rn6 using STAR (v2.4.0.i) and PCR duplicate reads were flagged using Picard MarkDuplicates (v1.92). Library quality was assessed using samtools flagstat (v1.1) and Picard CollectRNASeqMetrics. Uniquely aligned, non-PCR-duplicate reads were kept for downstream analysis using samtools parameters: samtools view -bh -q 255 -F 1540. NCBI RefSeq mouse transcript annotations ($$n = 106$$,520 transcripts, 35,977 genes) were downloaded from the UCSC Table Browser (last updated 2017-11-16). NCBI RefSeq rat transcript annotations ($$n = 69$$,194 isoforms, 30,871 transcripts) were downloaded from the UCSC Table Browser (last updated 2018-03-09). Transcript expression values were calculated by averaging read coverage over exons using VisRseq (v0.9.12) and normalized to the total number of aligned reads and transcript length in kilobases (RPKM). Genome browser compatible normalized bigWigs were generated using bedtools genomecov (v2.22.1).
## Comparative analysis of mouse and rat homologous gene expression
Mouse and rat homologous gene annotations were downloaded from Ensembl Biomart ($$n = 14$$,297). Mouse and rat epiblast and EPC gene expression similarity matrices and hierarchical clustering was subsequently generated using the Morpheus (Broad Institute) Spearman correlation tool on adjusted (log2(RPKM+1)) RPKM values. Mouse-rat syntenic 1-kb bins ($$n = 1$$,899,360) were generated previously [47], and syntenic regions were illustrated using Emsembl Comparative Genomics and Region Comparison tools.
## Allele-specific RNAseq analysis using a diploid pseudogenome
Strain-specific SNVs and INDELs were downloaded from [3, 4, 76, 77], and diploid pseudogenomes “C57BL/6NJ x JF1” and “BN/Mcwi x WKY/NCrl” and “BN/Mcwi x F344/N” were generated using MEA. Allele-specific read alignment was performed using MEA with default parameters. Allele-specific transcripts expression over exons was calculated using VisRseq (v0.9.12), and the contribution of paternal allele transcript expression was calculated using the formula:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textrm{Paternal}\ \textrm{bias}=\textrm{paternal}\ \textrm{coverage}/\left(\textrm{paternal}+\textrm{maternal}\ \textrm{coverage}\right)$$\end{document}Paternalbias=paternalcoverage/paternal+maternalcoverage Strand- and allele-specific bigWigs were organized into UCSC Track Hubs along with reference-aligned bigWigs, as previously described.
## Identifying rat imprinted genes
NCBI RefSeq rat transcript annotations ($$n = 69$$,194 transcripts, 30,871 genes) were downloaded from the UCSC Table Browser (last updated 2018-03-09), and total and allele-specific transcript expression was quantified as described above. An additional set of 3242 transcripts were identified by de novo transcriptome assembly using Stringtie (v2.1.4) [116] and default parameters followed by merging with the NCBI Refseq transcript annotations using Stringtie --merge and default parameters. Transcripts were filtered based on total (allele-agnostic) expression (RPKM ≥1) and allele-specific coverage (RPM ≥0.5). Paternal expression ratios were calculated as described above. Since transcripts are not necessarily expressed in all cell types analyzed in this study, or do not necessarily overlap a parental genetic variant, a paternal bias score of “−2” or “−1” was specified to transcripts not expressed (RPKM < 1) or did not meet allelic thresholds (allelic RPM <0.5), respectively. Student’s t tests were performed on the paternal expression ratio value for each transcript that passed filtering criteria in at least 1 sample, and Bonferroni correction was applied to the resulting P-values. Transcripts with an adjusted p-value <0.05 and for which sufficient allele-specific and total coverage was reported in at least 6 samples (rat) or 4 samples (mouse) were categorized as imprinted. It should be noted that since extraembryonic tissues are located between maternal and embryonic tissues, extraembryonic cell isolation can be contaminated with maternal decidua, which can in turn confound and overestimate the number and extent of maternally expressed imprinted genes [117]. An additional filtering step was performed for rat EPC samples, whereby maternally expressed imprinted transcripts that are expressed in blood (RPKM >1 from PRJEB23955 [118], 9 replicates) were discarded. To avoid the artifactual scoring of maternally expressed transcripts caused by possible residual maternal decidua in rat EPC samples, we further categorized $\frac{35}{42}$ maternally expressed imprinted transcripts as being normally expressed (allele-agnostic RPKM >1) in adult blood. Known imprinted genes in mouse such as Ascl2, Trpm5, and Tssc4 are included in the list of 35 genes, reflecting stringent annotations. Given their shared imprinted status in mouse, Ascl2, Trpm5, and Tssc4 are reported in Fig. 1d and Fig. 2a,b.
An additional test for imprinted expression in rat was carried out using linear modeling using Limma v3.50.1 [119] in R v4.1.2 [120]. Transcripts with allelic RPM ≥0.5 on either allele in at least 2 replicates in both crosses (BWxWB and BFxFB) in either tissue (epiblast or EPC) type were kept. An eBayes statistic was calculated following linear model fitting using “lmFit”, and transcripts with an adjusted p-value <0.05 and a ≥4-fold change in expression between parental alleles were kept. The same filtering step for maternally expressed imprinted genes in EPC samples described above was applied. A set of 37 imprinted genes were identified, 16 of which were not identified in the T-test described above, and were included in Fig. 1d. *Two* genes identified by Limma, Itga1 and LOC103691708, are encoded on chromosome 2 and are not included in Fig. 1d due to space limitations. See Additional file 2: Table S1 for the full list of imprinted genes identified by T-tests and Limma.
## Identifying mouse imprinted genes
NCBI RefSeq mouse transcript annotations ($$n = 106$$,520 transcripts, 35,977 genes) were downloaded from the UCSC Table Browser (last updated 2017-11-16) and transcript expression was quantified as described above in rat.
## Agnostic identification of parent-specific DNA methylation
Sequencing reads were trimmed using Trimmomatic (v0.32) and the following parameters: SLIDINGWINDOW:3:10 MINLEN:36 ILLUMINACLIP:TruSeq2-3-PE.fa:2:30:10. Read pairs that survived trimming were aligned to diploid rat or mouse pseudogenomes (as described above) using MEA and Bismark (v0.16.3). Duplicate reads were removed, and the number of methylated and unmethylated reads overlapping each CpG of the rat or mouse genome was reported. This resulted in total (allele-agnostic) and maternal- and paternal-specific CpG report files. Replicates and reciprocal crosses were merged for visualization and analyses. CpGs covered by at least 5 (total) or 1 (maternal or paternal) sequencing reads were assessed for methylation levels and genome-wide tracks were subsequently generated for visualization. The R package DSS v2.14.0 was used to identify DMRs from replicate-merged allele-specific epiblast and EPC CpG report files with smoothing enabled and default parameters. CpGs that showed greater than $10\%$ differential methylation levels between parental alleles with an associated p-value <0.001 were grouped into DMRs (Additional file 3: Table S2). DMR calls from epiblast and EPC samples were merged using bedtools merge. DNAme levels were then re-calculated from oocytes, sperm, and epiblast and EPC samples using bedops bedmap v2.4.39 (Neph et al., 2012). A final list of DMRs wherein all samples are covered by at least 2 CpGs and showed a parental difference of ≥$50\%$ DNAme was created for rat and mouse (Additional file 1: Fig. S1b & Additional file 3: Table S2).
For each imprinted gene identified in the allele-specific RNAseq analysis, bedtools Closest v2.27.0 was used to determine the nearest DMR to genic transcription start site. DMRs located within 5kb upstream of an imprinted gene’s TSS were called as promoter DMRs. The Sall1 DMR was determined by manual curation of an upstream ERVK promoter element. The Magel2 DMR, which is shared with the *Snrpn* gene, was determined based on conservation with mouse and human DMRs as Magel2 is incorrectly assembled in the rn6 reference.
## CUT&RUN sequencing analysis
Sequencing reads were trimmed using Trimmomatic (v0.32) and the following parameters: SLIDINGWINDOW:3:10 MINLEN:20 ILLUMINACLIP:TruSeq2_SE.fa:2:30:10. Reads that survived trimming were aligned to genome build mm10 or rn6 using Bowtie2 (v2.2.3) and PCR duplicate reads were flagged using Picard MarkDuplicates (v1.92). Library quality was assessed using samtools flagstat (v1.1). Uniquely aligned, non-PCR-duplicate reads were used for bigwig creation using deeptools bamCoverage (v3.3.0) and parameters: --binSize 1,50,200 --smoothLength 0,100,2000 minMappingQuality 10 --normalizeUsing CPM –ignoreDuplicates --ignoreForNormalization chrX chrM chrY --outFileFormat bigwig.
## Comparison of mouse and rat oocyte CUT&RUN data
Rat oocyte H3K4me3, H3K27me3, and H3K36me3 CUT&RUN Spearman correlation metrics were calculated using RPKM levels over 10-kb genomics bins and the Morpheus Similarity Matrix tool. Bins with at least 10 CpGs covered by at least 5 sequencing read alignments separated by over 1 read length ($$n = 246$$,$\frac{622}{285}$,920, $86\%$) were considered. Conservation of epigenetic modification profiles in oocytes was measured by Spearman correlation of RPKM and DNAme levels over 1-kb rat-mouse syntenic regions generated previously [47] using the Morpheus Similarity Matrix function. Differential enrichment of H3K27me and H3K36me3 levels as well as differential DNAme levels were calculated by subtracting the mouse signal from the rat and plotted as a 2D scatterplot using matplotlib pyplot. A contour plot illustrating the density of data points was generated using seaborn kdeplot.
Rat GVO H3K4me3, H3K27me3, H3K36me3, DNAme [47], and RNAseq [47] data was compared to mouse MII oocyte H3K4me3 [70], H3K27me3 [121], H3K36me3 [66], DNAme [50], and RNAseq [70] over transcribed and repressed genes in oocytes. Gene expression levels in the oocyte were calculated using the rat-mouse homologous gene annotation described above. For each species, genes were categorized as expressed (RPKM ≥1) or not (RPKM<1), and heatmaps were generated over both groups using deeptools ComputeMatrix and PlotHeatmap functions for both species.
To directly compare rat, mouse, and human oocyte H3K4me3 peak sizes, human and mouse oocyte H3K4me3 CUT&RUN data were downloaded from [72] as well as rat [68] and reprocessed using the same parameters. SICER2 [122] was used with default parameters to call peaks, and peaks within 3kb were grouped together. The distribution of H3K4me3 peak sizes was plotted using VisRseq, and peak call files were converted to bigBed for visualization in the UCSC Genome Browser [123].
## Software used
Bedops (v2.4.39) [124], Bedtools (v2.22.1) [125], Bismark (v0.16.3) [126], Bowtie2 (2.2.3) [127], Deeptools2 (v3.3.0) [128], DSS (v2.14.0) [45, 46], Limma (v3.50.1) [119], MEA (v1.0) [26], Morpheus (Broad Institute), PicardTools(v1.92) [129], R base (v4.1.2) [120], Samtools (v1.1) [130], Seaborn [131], SICER2 [122], STAR (v2.4.0.i) [132], Stringtie (v2.1.4) [116], Trimmomatic (v0.32) [115], UCSC Genome Browser [123], UCSC Track Hubs [133], VisRseq (v0.9.12) [134]
## Supplementary Information
Additional file 1: Supplemental figures 1-10.Additional file 2: Table S1. Data displayed in figure panels. Additional file 3: Table S2. Rat and mouse differentially methylated regions. Additional file 4: Table S3. *Datasets* generated and mined in this study. Additional file 5. Allele specific RNAseq. Additional file 6. Allele agnostic RNAseq. Additional file 7. Syntenic 1kb regions oocyte epigenetic profiling. Additional file 8. Rat and mouse oocyte 1kb bin DNAme levels. Additional file 9. Rat 10kb bins epigenetic profiling. Additional file 10. Review history.
## Review history
The review history is available as Additional file 10.
## Peer review information
Tim Sands and Wenjing She were the primary editors of this article and managed its editorial process and peer review in collaboration with the rest of the editorial team.
## Authors’ Twitter handles
Twitter handles: @jrichardalbert (Julien Richard Albert); @a_z_usa__ (Azusa Inoue); @AnaMonteagudo4 (Ana Monteagudo-Sánchez); @keegankorthauer (Keegan Korthauer); @maxvcg (Maxim Greenberg); @LorinczMatthew (Matthew Lorincz)
## References
1. Lindblad-Toh K. **Genome sequencing: three’s company**. *Nature.* (2004.0) **428** 475-476. DOI: 10.1038/428475a
2. Gibbs RA, Weinstock GM, Metzker ML. **Genome sequence of the Brown Norway rat yields insights into mammalian evolution**. *Nature.* (2004.0) **428** 493-521. DOI: 10.1038/nature02426
3. Atanur SS, Diaz AG, Maratou K. **Genome sequencing reveals loci under artificial selection that underlie disease phenotypes in the laboratory rat**. *Cell.* (2013.0) **154** 691-703. DOI: 10.1016/j.cell.2013.06.040
4. Hermsen R, de Ligt J, Spee W. **Genomic landscape of rat strain and substrain variation**. *BMC Genomics* (2015.0) **16** 357. DOI: 10.1186/s12864-015-1594-1
5. Surani MA, Barton SC. **Development of gynogenetic eggs in the mouse: implications for parthenogenetic embryos**. *Science.* (1983.0) **222** 1034-1036. DOI: 10.1126/science.6648518
6. Surani MA, Barton SC, Norris ML. **Development of reconstituted mouse eggs suggests imprinting of the genome during gametogenesis**. *Nature.* (1984.0) **308** 548-550. DOI: 10.1038/308548a0
7. McGrath J, Solter D. **Completion of mouse embryogenesis requires both the maternal and paternal genomes**. *Cell.* (1984.0) **37** 179-183. DOI: 10.1016/0092-8674(84)90313-1
8. Cattanach BM, Kirk M. **Differential activity of maternally and paternally derived chromosome regions in mice**. *Nature.* (1985.0) **315** 496-498. DOI: 10.1038/315496a0
9. Chen Z, Zhang Y. **Maternal H3K27me3-dependent autosomal and X chromosome imprinting**. *Nat Rev Genet* (2020.0) **3** 1-17
10. Tucci V, Isles AR, Kelsey G. **Genomic imprinting and physiological processes in mammals**. *Cell.* (2019.0) **176** 952-965. DOI: 10.1016/j.cell.2019.01.043
11. 11.Kobayashi H. Canonical and non-canonical genomic imprinting in rodents. Front Cell Dev Biol. 2021;9:713878.
12. Inoue A, Jiang L, Lu F, Suzuki T, Zhang Y. **Maternal H3K27me3 controls DNA methylation-independent imprinting**. *Nature.* (2017.0) **547** 419-424. DOI: 10.1038/nature23262
13. Inoue A, Chen Z, Yin Q, Zhang Y. **Maternal Eed knockout causes loss of H3K27me3 imprinting and random X inactivation in the extraembryonic cells**. *Genes Dev* (2018.0) **32** 1525-1536. DOI: 10.1101/gad.318675.118
14. Chen Z, Yin Q, Inoue A, Zhang C, Zhang Y. **Allelic H3K27me3 to allelic DNA methylation switch maintains noncanonical imprinting in extraembryonic cells**. *Sci Adv* (2019.0) **5** eaay7246. DOI: 10.1126/sciadv.aay7246
15. Hanna CW, Pérez-Palacios R, Gahurova L. **Endogenous retroviral insertions drive non- canonical imprinting in extra-embryonic tissues**. *Genome Biol* (2019.0) **20** 1-17. DOI: 10.1186/s13059-019-1833-x
16. Tachibana M, Ueda J, Fukuda M. **Histone methyltransferases G9a and GLP form heteromeric complexes and are both crucial for methylation of euchromatin at H3-K9**. *Genes Dev* (2005.0) **19** 815-826. DOI: 10.1101/gad.1284005
17. Zeng TB, Pierce N, Liao J, Szabó PE. **H3K9 methyltransferase EHMT2/G9a controls ERVK-driven noncanonical imprinted genes**. *Epigenomics.* (2021.0) **13** 1299-1314. DOI: 10.2217/epi-2021-0168
18. Andergassen D, Smith ZD, Kretzmer H, Rinn JL, Meissner A. **Diverse epigenetic mechanisms maintain parental imprints within the embryonic and extraembryonic lineages**. *Dev Cell* (2021.0) **56** 2995-3005.e4. DOI: 10.1016/j.devcel.2021.10.010
19. Wang Q, Chow J, Hong J. **Recent acquisition of imprinting at the rodent Sfmbt2 locus correlates with insertion of a large block of miRNAs**. *BMC Genomics* (2011.0) **12** 204. DOI: 10.1186/1471-2164-12-204
20. Itoh M, Yoshida Y, Nishida K, Narimatsu M, Hibi M, Hirano T. **Role of Gab1 in heart, placenta, and skin development and growth factor- and cytokine-induced extracellular signal-regulated kinase mitogen-activated protein kinase activation**. *Mol Cell Biol* (2000.0) **20** 3695-3704. DOI: 10.1128/MCB.20.10.3695-3704.2000
21. Hanna CW. **Placental imprinting: Emerging mechanisms and functions**. *PLoS Genet* (2020.0) **16** e1008709-e1008716. DOI: 10.1371/journal.pgen.1008709
22. Hanna CW, Kelsey G. **Features and mechanisms of canonical and noncanonical genomic imprinting**. *Genes Dev* (2021.0) **35** 821-834. DOI: 10.1101/gad.348422.121
23. Wilkins JF, Haig D. **What good is genomic imprinting: the function of parent-specific gene expression**. *Nat Rev Genet* (2003.0) **4** 359-368. DOI: 10.1038/nrg1062
24. Babak T, DeVeale B, Tsang EK. **Genetic conflict reflected in tissue-specific maps of genomic imprinting in human and mouse**. *Nat Genet* (2015.0) **47** 544-549. DOI: 10.1038/ng.3274
25. Takagi N, Sasaki M. **Preferential inactivation of the paternally derived X chromosome in the extraembryonic membranes of the mouse**. *Nature.* (1975.0) **256** 640-642. DOI: 10.1038/256640a0
26. Richard Albert J, Koike T, Younesy H. **Development and application of an integrated allele-specific pipeline for methylomic and epigenomic analysis (MEA)**. *BMC Genomics* (2018.0) **19** 463. DOI: 10.1186/s12864-018-4835-2
27. van de Geijn B, McVicker G, Gilad Y, Pritchard JK. **WASP: allele-specific software for robust molecular quantitative trait locus discovery**. *Nat Chem Biol* (2015.0) **12** 1061-1063
28. Andergassen D, Dotter CP, Kulinski TM. **Allelome.PRO, a pipeline to define allele-specific genomic features from high-throughput sequencing data**. *Nucleic Acids Res* (2015.0) **43** e146. PMID: 26202974
29. Wang X, Clark AG. **Using next-generation RNA sequencing to identify imprinted genes**. *Heredity.* (2014.0) **113** 156-166. DOI: 10.1038/hdy.2014.18
30. Li Y, Li J. **Technical advances contribute to the study of genomic imprinting**. *PLoS Genet* (2019.0) **15** e1008151. DOI: 10.1371/journal.pgen.1008151
31. 31.Hubert J-N, Demars J. Genomic imprinting in the new omics era: a model for systems-level approaches. Front Genet. 2022;13:838534.
32. Barlow DP, Stöger R, Herrmann BG, Saito K, Schweifer N. **The mouse insulin-like growth factor type-2 receptor is imprinted and closely linked to the Tme locus**. *Nature.* (1991.0) **349** 84-87. DOI: 10.1038/349084a0
33. Giannoukakis N, Deal C, Paquette J, Goodyer CG, Polychronakos C. **Parental genomic imprinting of the human IGF2 gene**. *Nat Genet* (1993.0) **4** 98-101. DOI: 10.1038/ng0593-98
34. Overall M, Bakker M, Spencer J, Parker N, Smith P, Dziadek M. **Genomic imprinting in the rat: linkage of Igf2 and H19 genes and opposite parental allele-specific expression during embryogenesis**. *Genomics* (1997.0) **45** 416-20. DOI: 10.1006/geno.1997.4933
35. Chen Z, Hagen DE, Wang J. **Global assessment of imprinted gene expression in the bovine conceptus by next generation sequencing**. *Epigenetics.* (2016.0) **11** 501-516. DOI: 10.1080/15592294.2016.1184805
36. Barlow DP, Bartolomei MS. **Genomic imprinting in mammals**. *Cold Spring Harb Perspect Biol* (2014.0) **6** a018382. DOI: 10.1101/cshperspect.a018382
37. Dong Y, Jin L, Liu X. **IMPACT and OSBPL1A are two isoform-specific imprinted genes in bovines**. *Theriogenology.* (2022.0) **184** 100-109. DOI: 10.1016/j.theriogenology.2022.02.023
38. Bischoff SR, Tsai S, Hardison N. **Characterization of conserved and nonconserved imprinted genes in swine**. *Biol Reprod* (2009.0) **81** 906-920. DOI: 10.1095/biolreprod.109.078139
39. Ding YY, Liu LY, Zhou J. **Detection of a novel single nucleotide polymorphism and imprinted status analysis of the Ras protein-specific guanine nucleotide-releasing factor 1 gene in domestic pigs**. *Genet Mol Res* (2014.0) **13** 10574-10581. DOI: 10.4238/2014.December.12.20
40. Oczkowicz M, Szmatoła T, Piórkowska K, Ropka-Molik K. **Variant calling from RNA-seq data of the brain transcriptome of pigs and its application for allele-specific expression and imprinting analysis**. *Gene.* (2018.0) **641** 367-375. DOI: 10.1016/j.gene.2017.10.076
41. Ahn J, Wu H, Lee J. **Identification of a Novel Imprinted Transcript in the Porcine GNAS Complex Locus Using Methylome and Transcriptome of Parthenogenetic Fetuses**. *Genes (Basel)* (2020.0) **11** E96. DOI: 10.3390/genes11010096
42. Suzuki S, Renfree MB, Pask AJ. **Genomic imprinting of IGF2, p57(KIP2) and PEG1/MEST in a marsupial, the tammar wallaby**. *Mech Dev* (2005.0) **122** 213-222. DOI: 10.1016/j.mod.2004.10.003
43. Suzuki S, Ono R, Narita T. **Retrotransposon silencing by DNA methylation can drive mammalian genomic imprinting**. *PLoS Genet* (2007.0) **3** e55. DOI: 10.1371/journal.pgen.0030055
44. Ishihara T, Griffith OW, Suzuki S, Renfree MB. **Placental imprinting of SLC22A3 in the IGF2R imprinted domain is conserved in therian mammals**. *Epigenetics Chromatin* (2022.0) **15** 32. DOI: 10.1186/s13072-022-00465-4
45. Feng H, Conneely KN, Wu H. **A Bayesian hierarchical model to detect differentially methylated loci from single nucleotide resolution sequencing data**. *Nucleic Acids Res* (2014.0) **42** e69. DOI: 10.1093/nar/gku154
46. Park Y, Wu H. **Differential methylation analysis for BS-seq data under general experimental design**. *Bioinformatics.* (2016.0) **32** 1446-1453. DOI: 10.1093/bioinformatics/btw026
47. Brind’Amour J, Kobayashi H, Richard Albert J. **LTR retrotransposons transcribed in oocytes drive species-specific and heritable changes in DNA methylation. Nature**. *Communications.* (2018.0) **9** 3331
48. Huber GC. **The development of the albino rat, Mus norvegicus albinus**. *J Morphol* (1915.0) **26** 247-358. DOI: 10.1002/jmor.1050260205
49. Brons IG, Smithers LE, Trotter MW. **Derivation of pluripotent epiblast stem cells from mammalian embryos**. *Nature.* (2007.0) **448** 191-195. DOI: 10.1038/nature05950
50. Shirane K, Toh H, Kobayashi H. **Mouse oocyte methylomes at base resolution reveal genome-wide accumulation of non-CpG methylation and role of DNA methyltransferases**. *PLoS Genet* (2013.0) **9** e1003439. DOI: 10.1371/journal.pgen.1003439
51. Kubo N, Toh H, Shirane K. **DNA methylation and gene expression dynamics during spermatogonial stem cell differentiation in the early postnatal mouse testis**. *BMC Genomics* (2015.0) **16** 624. DOI: 10.1186/s12864-015-1833-5
52. Kimura Y, Hawkins MT, McDonough MM, Jacobs LL, Flynn LJ. **Corrected placement of Mus-Rattus fossil calibration forces precision in the molecular tree of rodents**. *Sci Rep* (2015.0) **5** 14444. DOI: 10.1038/srep14444
53. Umlauf D, Goto Y, Cao R. **Imprinting along the Kcnq1 domain on mouse chromosome 7 involves repressive histone methylation and recruitment of Polycomb group complexes**. *Nat Genet* (2004.0) **36** 1296-1300. DOI: 10.1038/ng1467
54. Schertzer MD, Braceros KCA, Starmer J. **lncRNA-Induced Spread of Polycomb Controlled by Genome Architecture, RNA Abundance, and CpG Island DNA**. *Mol Cell* (2019.0) **75** 523-537.e10. DOI: 10.1016/j.molcel.2019.05.028
55. Chiesa N, De Crescenzo A, Mishra K. **The KCNQ1OT1 imprinting control region and non-coding RNA: new properties derived from the study of Beckwith–Wiedemann syndrome and Silver–Russell syndrome cases**. *Hum Mol Genet* (2012.0) **21** 10-25. DOI: 10.1093/hmg/ddr419
56. Miyoshi N, Kuroiwa Y, Kohda T. **Identification of the Meg1/Grb10 imprinted gene on mouse proximal chromosome 11, a candidate for the Silver–Russell syndrome gene**. *Proc Natl Acad Sci* (1998.0) **95** 1102-1107. DOI: 10.1073/pnas.95.3.1102
57. Blagitko N, Mergenthaler S, Schulz U. **Human GRB10 is imprinted and expressed from the paternal and maternal allele in a highly tissue-and isoform-specific fashion**. *Hum Mol Genet* (2000.0) **9** 1587-1595. DOI: 10.1093/hmg/9.11.1587
58. Hatada I, Kitagawa K, Yamaoka T. **Allele-specific methylation and expression of an imprinted U2af1-rs1 (SP2) gene**. *Nucleic Acids Res* (1995.0) **23** 36-41. DOI: 10.1093/nar/23.1.36
59. Joh K, Matsuhisa F, Kitajima S. **Growing oocyte-specific transcription-dependent de novo DNA methylation at the imprinted Zrsr1-DMR**. *Epigenetics Chromatin* (2018.0) **11** 28. DOI: 10.1186/s13072-018-0200-6
60. Santini L, Halbritter F, Titz-Teixeira F. **Genomic imprinting in mouse blastocysts is predominantly associated with H3K27me3**. *Nat Commun* (2021.0) **12** 3804. DOI: 10.1038/s41467-021-23510-4
61. 61.Chu C, Zhang W, Kang Y, et al. Analysis of developmental imprinting dynamics in primates using SNP-free methods to identify imprinting defects in cloned placenta. Dev Cell. 2021;56:2826–2840.e7.
62. Tahiliani M, Koh KP, Shen Y. **Conversion of 5-methylcytosine to 5-hydroxymethylcytosine in mammalian DNA by MLL partner TET1**. *Science.* (2009.0) **324** 930-935. DOI: 10.1126/science.1170116
63. 63.Skene PJ, Henikoff S. An efficient targeted nuclease strategy for high-resolution mapping of DNA binding sites. Elife. 2017;6:e21856.
64. Schmitges FW, Prusty AB, Faty M. **Histone methylation by PRC2 is inhibited by active chromatin marks**. *Mol Cell* (2011.0) **42** 330-341. DOI: 10.1016/j.molcel.2011.03.025
65. Finogenova K, Bonnet J, Poepsel S. **Structural basis for PRC2 decoding of active histone methylation marks H3K36me2/3**. *Elife.* (2020.0) **9** e61964. DOI: 10.7554/eLife.61964
66. Xu Q, Xiang Y, Wang Q. **SETD2 regulates the maternal epigenome, genomic imprinting and embryonic development**. *Nat Genet* (2019.0) **51** 844-856. DOI: 10.1038/s41588-019-0398-7
67. Ernst J, Kellis M. **ChromHMM: automating chromatin-state discovery and characterization**. *Nat Chem Biol* (2012.0) **9** 215-216
68. Lu X, Zhang Y, Wang L. **Evolutionary epigenomic analyses in mammalian early embryos reveal species-specific innovations and conserved principles of imprinting**. *Sci Adv* (2021.0) **7** eabi6178. DOI: 10.1126/sciadv.abi6178
69. Dahl JA, Jung I, Aanes H. **Broad histone H3K4me3 domains in mouse oocytes modulate maternal-to-zygotic transition**. *Nature.* (2016.0) **537** 548. DOI: 10.1038/nature19360
70. Zhang B, Zheng H, Huang B. **Allelic reprogramming of the histone modification H3K4me3 in early mammalian development**. *Nature.* (2016.0) **537** 553-557. DOI: 10.1038/nature19361
71. Hanna CW, Taudt A, Huang J. **MLL2 conveys transcription-independent H3K4 trimethylation in oocytes**. *Nat Struct Mol Biol* (2018.0) **25** 73-82. DOI: 10.1038/s41594-017-0013-5
72. Xia W, Xu J, Yu G. **Resetting histone modifications during human parental-to-zygotic transition**. *Science.* (2019.0) **365** 353-360. DOI: 10.1126/science.aaw5118
73. Au Yeung WK, Brind’Amour J, Hatano Y. **Histone H3K9 methyltransferase G9a in oocytes is essential for preimplantation development but dispensable for CG methylation protection**. *Cell Rep* (2019.0) **27** 282-293. DOI: 10.1016/j.celrep.2019.03.002
74. Bao W, Kojima KK, Kohany O. **Repbase Update, a database of repetitive elements in eukaryotic genomes**. *Mob DNA* (2015.0) **6** 11. DOI: 10.1186/s13100-015-0041-9
75. Smith JR, Hayman GT, Wang SJ. **The Year of the Rat: The Rat Genome Database at 20: a multi-species knowledgebase and analysis platform**. *Nucleic Acids Res* (2020.0) **48** D731-D742. PMID: 31713623
76. Keane TM, Goodstadt L, Danecek P, White MA. **Mouse genomic variation and its effect on phenotypes and gene regulation**. *Nature.* (2011.0) **477** 289-294. DOI: 10.1038/nature10413
77. Takada T, Ebata T, Noguchi H. **The ancestor of extant Japanese fancy mice contributed to the mosaic genomes of classical inbred strains**. *Genome Res* (2013.0) **23** 1329-1338. DOI: 10.1101/gr.156497.113
78. Luedi PP, Dietrich FS, Weidman JR, Bosko JM, Jirtle RL, Hartemink AJ. **Computational and experimental identification of novel human imprinted genes**. *Genome Res* (2007.0) **17** 1723-1730. DOI: 10.1101/gr.6584707
79. Calaway JD, Lenarcic AB, Didion JP. **Genetic architecture of skewed X inactivation in the laboratory mouse**. *PLoS Genet* (2013.0) **9** e1003853. DOI: 10.1371/journal.pgen.1003853
80. Borensztein M, Syx L, Ancelin K. **Xist-dependent imprinted X inactivation and the early developmental consequences of its failure**. *Nat Publ Group* (2017.0) **24** 226-233
81. Collombet S, Ranisavljevic N, Nagano T. **Parental-to-embryo switch of chromosome organization in early embryogenesis**. *Nature.* (2020.0) **15** 723-722
82. Wake N, Takagi N, Sasaki M. **Non-random inactivation of X chromosome in the rat yolk sac**. *Nature.* (1976.0) **262** 580-581. DOI: 10.1038/262580a0
83. Pedone PV, Cosma MP, Ungaro P. **Parental imprinting of rat insulin-like growth factor II gene promoters is coordinately regulated**. *J Biol Chem* (1994.0) **269** 23970-23975. DOI: 10.1016/S0021-9258(19)51033-X
84. Miri K, Latham K, Panning B, Zhong Z, Andersen A, Varmuza S. **The imprinted polycomb group gene Sfmbt2 is required for trophoblast maintenance and placenta development**. *Development.* (2013.0) **140** 4480-4489. DOI: 10.1242/dev.096511
85. Matoba S, Kozuka C, Miura K. **Noncanonical imprinting sustains embryonic development and restrains placental overgrowth**. *Genes Dev* (2022.0) **36** 483-94. PMID: 35483741
86. Kumar S, Stecher G, Suleski M, Hedges SB. **TimeTree: A resource for timelines, timetrees, and divergence times**. *Mol Biol Evol* (2017.0) **34** 1812-1819. DOI: 10.1093/molbev/msx116
87. Zhang W, Chen Z, Yin Q, Zhang D, Racowsky C, Zhang Y. **Maternal-biased H3K27me3 correlates with paternal-specific gene expression in the human morula**. *Genes Dev* (2019.0) **33** 382-387. DOI: 10.1101/gad.323105.118
88. Tachibana M, Sugimoto K, Fukushima T, Shinkai Y. **Set domain-containing protein, G9a, is a novel lysine-preferring mammalian histone methyltransferase with hyperactivity and specific selectivity to lysines 9 and 27 of histone H3**. *J Biol Chem* (2001.0) **276** 25309-25317. DOI: 10.1074/jbc.M101914200
89. Frapporti A, Miró Pina C, Arnaiz O. **The Polycomb protein Ezl1 mediates H3K9 and H3K27 methylation to repress transposable elements in Paramecium**. *Nat Commun* (2019.0) **10** 2710. DOI: 10.1038/s41467-019-10648-5
90. Carter AM, Enders AC. **Comparative aspects of trophoblast development and placentation**. *Reprod Biol Endocrinol* (2004.0) **2** 46. DOI: 10.1186/1477-7827-2-46
91. Chuong EB. **Retroviruses facilitate the rapid evolution of the mammalian placenta**. *Bioessays.* (2013.0) **35** 853-861. PMID: 23873343
92. Kaneko-Ishino T, Ishino F. **Evolution of viviparity in mammals: what genomic imprinting tells us about mammalian placental evolution**. *Reprod Fertil Dev* (2019.0) **31** 1219. DOI: 10.1071/RD18127
93. Chuong EB, Rumi MAK, Soares MJ, Baker JC. **Endogenous retroviruses function as species-specific enhancer elements in the placenta**. *Nat Genet* (2013.0) **45** 325-329. DOI: 10.1038/ng.2553
94. Bartolomei MS, Ferguson-Smith AC. **Mammalian Genomic Imprinting**. *Cold Spring Harb Perspect Biol* (2011.0) **3** a002592. DOI: 10.1101/cshperspect.a002592
95. Dempersmier J, Sambeat A, Gulyaeva O. **Cold-inducible Zfp516 activates UCP1 transcription to promote browning of white fat and development of brown fat**. *Mol Cell* (2015.0) **57** 235-246. DOI: 10.1016/j.molcel.2014.12.005
96. Cypess AM, Lehman S, Williams G. **Identification and importance of brown adipose tissue in adult humans**. *N Engl J Med* (2009.0) **360** 1509-1517. DOI: 10.1056/NEJMoa0810780
97. Small KS, Todorčević M, Civelek M. **Regulatory variants at KLF14 influence type 2 diabetes risk via a female-specific effect on adipocyte size and body composition**. *Nat Genet* (2018.0) **50** 572-580. DOI: 10.1038/s41588-018-0088-x
98. Matoba S, Nakamuta S, Miura K. **Paternal knockout of Slc38a4/SNAT4 causes placental hypoplasia associated with intrauterine growth restriction in mice**. *Proc Natl Acad Sci U S A* (2019.0) **116** 21047-21053. DOI: 10.1073/pnas.1907884116
99. Bogutz AB, Brind’Amour J, Kobayashi H. **Evolution of imprinting via lineage-specific insertion of retroviral promoters**. *Nat Commun* (2019.0) **10** 1-14. DOI: 10.1038/s41467-019-13662-9
100. Okae H, Hiura H, Nishida Y. **Re-investigation and RNA sequencing-based identification of genes with placenta-specific imprinted expression**. *Hum Mol Genet* (2012.0) **21** 548-558. DOI: 10.1093/hmg/ddr488
101. Auclair G, Borgel J, Sanz LA. **EHMT2 directs DNA methylation for efficient gene silencing in mouse embryos**. *Genome Res* (2016.0) **26** 192-202. DOI: 10.1101/gr.198291.115
102. Sakamoto K, Tamamura Y, Katsube K, Yamaguchi A. **Zfp64 participates in Notch signaling and regulates differentiation in mesenchymal cells**. *J Cell Sci* (2008.0) **121** 1613-1623. DOI: 10.1242/jcs.023119
103. Lu B, Klingbeil O, Tarumoto Y. **A Transcription Factor Addiction in Leukemia Imposed by the MLL Promoter Sequence**. *Cancer Cell* (2018.0) **34** 970-981.e8. DOI: 10.1016/j.ccell.2018.10.015
104. Wang X, Soloway PD, Clark AG. **A survey for novel imprinted genes in the mouse placenta by mRNA-seq**. *Genetics.* (2011.0) **189** 109-122. DOI: 10.1534/genetics.111.130088
105. Andergassen D, Dotter CP, Wenzel D. **Mapping the mouse Allelome reveals tissue-specific regulation of allelic expression**. *eLife.* (2017.0) **6** e146. DOI: 10.7554/eLife.25125
106. Strogantsev R, Krueger F, Yamazawa K. **Allele-specific binding of ZFP57 in the epigenetic regulation of imprinted and non-imprinted monoallelic expression**. *Genome Biol* (2015.0) **16** 112. DOI: 10.1186/s13059-015-0672-7
107. Mei H, Kozuka C, Hayashi R, Kumon M, Koseki H, Inoue A. **H2AK119ub1 guides maternal inheritance and zygotic installation of H3K27me3 in mouse embryos**. *Nat Genet* (2021.0) **53** 539-50. DOI: 10.1038/s41588-021-00820-3
108. Chen Z, Djekidel MN, Zhang Y. **Distinct dynamics and functions of H2AK119ub1 and H3K27me3 in mouse preimplantation embryos**. *Nat Genet* (2021.0) **53** 551-563. DOI: 10.1038/s41588-021-00821-2
109. Greenberg MVC, Glaser J, Borsos M. **Transient transcription in the early embryo sets an epigenetic state that programs postnatal growth**. *Nat Genet* (2016.0) **49** 110-118. DOI: 10.1038/ng.3718
110. Lefebvre L, Viville S, Barton SC, Ishino F, Keverne EB, Surani MA. **Abnormal maternal behaviour and growth retardation associated with loss of the imprinted gene Mest**. *Nat Genet* (1998.0) **20** 163-169. DOI: 10.1038/2464
111. Li L, Keverne EB, Aparicio SA, Ishino F, Barton SC, Surani MA. **Regulation of maternal behavior and offspring growth by paternally expressed Peg3**. *Science.* (1999.0) **284** 330-333. DOI: 10.1126/science.284.5412.330
112. Horsthemke B, Wagstaff J. **Mechanisms of imprinting of the Prader-Willi/Angelman region**. *Am J Med Genet A* (2008.0) **146A** 2041-2052. DOI: 10.1002/ajmg.a.32364
113. 113.Glaser J, Iranzo J, Borensztein M, et al. The imprinted Zdbf2 gene finely tunes control of feeding and growth in neonates. eLife. 2022;11:e65641.
114. Bonthuis PJ, Steinwand S, Stacher Hörndli CN. **Noncanonical genomic imprinting in the monoamine system determines naturalistic foraging and brain-adrenal axis functions**. *Cell Rep* (2022.0) **38** 110500. DOI: 10.1016/j.celrep.2022.110500
115. Bolger AM, Lohse M, Usadel B. **Trimmomatic: a flexible trimmer for Illumina sequence data**. *Bioinformatics.* (2014.0) **30** 2114-2120. DOI: 10.1093/bioinformatics/btu170
116. Pertea M, Pertea GM, Antonescu CM, Chang T-C, Mendell JT, Salzberg SL. **StringTie enables improved reconstruction of a transcriptome from RNA-seq reads**. *Nat Biotechnol* (2015.0) **33** 290-295. DOI: 10.1038/nbt.3122
117. Goncalves A, Leigh-Brown S, Thybert D. **Extensive compensatory cis-trans regulation in the evolution of mouse gene expression**. *Genome Res* (2012.0) **22** 2376-2384. DOI: 10.1101/gr.142281.112
118. Sahm A, Bens M, Henning Y. **Higher gene expression stability during aging in long-lived giant mole-rats than in short-lived rats**. *Aging (Albany NY)* (2018.0) **10** 3938-3956. DOI: 10.18632/aging.101683
119. Ritchie ME, Phipson B, Wu D. **limma powers differential expression analyses for RNA-sequencing and microarray studies**. *Nucleic Acids Res* (2015.0) **43** e47. DOI: 10.1093/nar/gkv007
120. 120.R-Core-TeamR: A language and environment for statistical computing2021ViennaR Foundation for Statistical Computing. *R: A language and environment for statistical computing* (2021.0)
121. Zheng H, Huang B, Zhang B. **Resetting epigenetic memory by reprogramming of histone modifications in mammals**. *Mol Cell* (2016.0) **63** 1066-1079. DOI: 10.1016/j.molcel.2016.08.032
122. Zang C, Schones DE, Zeng C, Cui K, Zhao K, Peng W. **A clustering approach for identification of enriched domains from histone modification ChIP-Seq data**. *Bioinformatics.* (2009.0) **25** 1952-1958. DOI: 10.1093/bioinformatics/btp340
123. Kent WJ, Sugnet CW, Furey TS. **The human genome browser at UCSC**. *Genome Res* (2002.0) **12** 996-1006. DOI: 10.1101/gr.229102
124. Neph S, Kuehn MS, Reynolds AP. **BEDOPS: high-performance genomic feature operations**. *Bioinformatics.* (2012.0) **28** 1919-1920. DOI: 10.1093/bioinformatics/bts277
125. Quinlan AR, Hall IM. **BEDTools: a flexible suite of utilities for comparing genomic features**. *Bioinformatics.* (2010.0) **26** 841-842. DOI: 10.1093/bioinformatics/btq033
126. Krueger F, Andrews SR. **Bismark: a flexible aligner and methylation caller for Bisulfite-Seq applications**. *Bioinformatics.* (2011.0) **27** 1571-1572. DOI: 10.1093/bioinformatics/btr167
127. Langmead B, Wilks C, Antonescu V, Charles R. **Scaling read aligners to hundreds of threads on general-purpose processors**. *Bioinformatics.* (2019.0) **35** 421-432. DOI: 10.1093/bioinformatics/bty648
128. Ramírez F, Ryan DP, Grüning B. **deepTools2: a next generation web server for deep-sequencing data analysis**. *Nucleic Acids Res* (2016.0) **44** W160-W165. DOI: 10.1093/nar/gkw257
129. 129.Broad-Institute. Picard toolkit. GitHub repository 2019.
130. Li H, Handsaker B, Wysoker A. **The Sequence Alignment/Map format and SAMtools**. *Bioinformatics.* (2009.0) **25** 2078-2079. DOI: 10.1093/bioinformatics/btp352
131. Waskom M. **seaborn: statistical data visualization**. *J Open Source Software* (2021.0) **6** 3021. DOI: 10.21105/joss.03021
132. Dobin A, Davis CA, Schlesinger F. **STAR: ultrafast universal RNA-seq aligner**. *Bioinformatics.* (2013.0) **29** 15-21. DOI: 10.1093/bioinformatics/bts635
133. Raney BJ, Dreszer TR, Barber GP. **Track data hubs enable visualization of user-defined genome-wide annotations on the UCSC Genome Browser**. *Bioinformatics.* (2014.0) **30** 1003-1005. DOI: 10.1093/bioinformatics/btt637
134. Younesy H, Möller T, Lorincz MC, Karimi MM, Jones SJM. **VisRseq: R-based visual framework for analysis of sequencing data**. *BMC Bioinformatics* (2015.0) **16** S2. DOI: 10.1186/1471-2105-16-S11-S2
135. 135.Kobayashi H, Inoue A, Kobayashi T, Richard Albert J. Conservation and divergence of canonical and non-canonical imprinting in murids. Oocyte datasets. Gene Expression Omnibus. 2023. Available from: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE185573.
136. 136.Kobayashi H, Kobayashi T, Richard Albert J. Conservation and divergence of canonical and non-canonical imprinting in murids. RNAseq Datasets. Gene Expression Omnibus. 2023. Available from: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE185574.
137. 137.Kobayashi H, Kobayashi T, Richard Albert J. Conservation and divergence of canonical and non-canonical imprinting in murids. WGBS Datasets. Gene Expression Omnibus. 2023. Available from: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE186492.
138. 138.Richard Albert J. Code to reproduce ad hoc data filtering and figure plotting for Conservation and divergence of canonical and non-canonical imprinting in murids. GitHub repository. 2023. Available from: https://github.com/julienrichardalbert/the_rat_imprintome/releases/tag/v0.2.
139. 139.Korthauer K. Code to reproduce allele-specific expression analysis for Conservation and divergence of canonical and non-canonical imprinting in murids. GitHub repository. 2023. Available from: https://github.com/kdkorthauer/rat_imprintome/releases/tag/accepted-submission.
140. 140.Korthauer K, Richard Albert J. Stable code repository for Conservation and divergence of canonical and non-canonical imprinting in murids. Zenodo. 2023. Available from: 10.5281/zenodo.7551841.
|
---
title: Plasma exosome-derived connexin43 as a promising biomarker for melanoma patients
authors:
- Yue Shen
- Ming Li
- Li Liao
- Suyue Gao
- Yongzhen Wang
journal: BMC Cancer
year: 2023
pmcid: PMC10012581
doi: 10.1186/s12885-023-10705-9
license: CC BY 4.0
---
# Plasma exosome-derived connexin43 as a promising biomarker for melanoma patients
## Abstract
### Background
To examine the levels of exosome-derived connexin 43 (Cx43) in plasma and estimate its forecast value in patients with melanoma.
### Methods
We measured the plasma exosome-derived Cx43 levels in the plasma of 112 melanoma patients and 50 healthy controls.
### Results
The plasma exosome-derived Cx43 levels in patients with melanoma were substantially downregulated as opposed to the levels in healthy controls ($P \leq 0.001$). Kaplan–*Meier analysis* indicated that overall survival (OS) and disease-free survival (DFS) were poorer in patients with melanoma who exhibited lower levels of plasma exosome-derived Cx43 (both $P \leq 0.001$). The levels of plasma exosome-derived Cx43 were considerably elevated in patients with melanoma whose tumor was situated in the skin, tumor size < 10 cm, with Clark level I–III, TNM stages IIb–IV, and had no lymph node metastasis as opposed to patients whose tumor was situated in the viscera or mucosa, tumor size ≥ 10 cm, Clark level IV–V, TNM stages IIb–IV and had lymph node metastasis (all $P \leq 0.05$). The receiver operating characteristic (ROC) of plasma exosome-derived Cx43 for forecasting 5-year DFS in patients with melanoma was 0.78 ($95\%$ confidence interval (CI): 0.70–0.86), with a specificity of $77.78\%$ and a sensitivity of $81.55\%$. The ROC of plasma exosome-derived Cx43 for forecasting 5-year OS of patients with melanoma was 0.77 ($95\%$ CI: 0.68–0.84), with a specificity of $80.0\%$ and sensitivity of $65.98\%$.
### Conclusion
The overall findings indicated that the levels of plasma exosome-derived Cx43 in patients with melanoma were considerably downregulated. It can therefore be inferred that the levels of plasma exosome-derived Cx43 might be a prospective prognostic indicator for 5 5-year OS and 5-year DFS of patients with melanoma.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12885-023-10705-9.
## Introduction
Melanoma has been extensively recognized as a highly malignant tumor originating from the melanocytes in the basal layer of the epidermis [1, 2]. Most melanomas occur in the skin, but also in the mucosa and viscera, and account for about $3\%$ of all tumors. Malignant melanoma has been found to be the third commonest type of skin malignancy (6.8–$20\%$), and is more common in adults, especially in fair-skinned Caucasians [3, 4]. Over the past few years, the prevalence and fatality of malignant melanoma have been increasing progressively, with an annual growth rate of 3–$5\%$ [5]. It has been reported that skin malignant melanoma has gene mutations, such as BRAF, CKIT, and NRAS [6]. Clinical studies have shown that molecular targeted therapy is the main strategy for metastatic melanoma, although only a limited proportion of patients can enjoy its benefits [7]. Consequently, there is no specific treatment for most patients with metastatic melanoma. Early surgical resection is the preferred treatment for malignant melanoma, but the prognosis of patients is poor and the postoperative survival is generally not optimistic [8].
Connexin (Cx) is a member of the transmembrane protein family. More than 20 kinds of connexins have been found in humans, among which, Cx43 is the most common [9, 10]. Cx43 is an abundant and widely distributed junction protein, which is mainly distributed on the surface of cell membranes in normal tissues. It is mainly involved in cell gap junction communication, providing a necessary channel for information exchange between cells, and also providing a connection pathway for tumor cells and vascular endothelial cells [11, 12]. In addition, the function of gap junction channels is regulated by the expression of CX43, and both quantitative and qualitative changes of gap junction proteins can influence the occurrence, differentiation, invasion as well as metastasis of tumors [13]. In recent years, CX43 has been found to be strongly linked to the occurrence as well as the development of many malignant tumors. Bišćanin et al. found that Cx43 was more highly expressed in normal tissues compared with colorectal cancer tissues, suggesting that Cx43 has antitumor properties [14]. Teleki et al. found that gap junction proteins Cx26, Cx30, Cx32, Cx43, and Cx46 were differentially expressed in breast cancer progression and prognosis, and significant correlations were also found at the mRNA level. Elevated levels of Cx43 significantly improve the prognosis of breast cancer [15]. Compared to reliance on vascular invasion or necrosis as an evaluation index, Cx43 protein detection is more valuable as an independent prognostic indicator. High expression of Cx43 is linked to reduced overall survival (OS) in patients with oral squamous cell carcinoma [16]. Cx43’s function in the production of vascular endothelial growth factor (VEGF) and tumor angiogenesis in mice suggests that reduced Cx43 expression leads to increased expression of VEGF in B16F10 melanoma cells and 4T1breast cancer cells, which promotes tumor metastasis [17]. However, overexpression of Cx43 can reduce angiogenesis and inhibit tumor growth.
Exosomes are membrane-covered vesicles secreted by a variety of living cells, which contain proteins, lipids, nucleic acids, and a variety of other biological macromolecules [18–20]. Exosomes perform a vital function in many pathophysiological processes, including antigen presentation in the immune response, repair of tissue damage, and tumor growth and migration [21, 22]. Tumor-derived or tumor-related exosomes have significant implications for the regulation of tumorigenesis and development. Analysis and detection of tumor exosomes can assist in the early diagnosis of tumors and provide new treatment methods for patients [23–27].
A growing body of research shows that exosomes comprise of numerous essential proteins that might be utilized for early tumor diagnosis, patient prognosis analysis, and tumor-targeting treatment [28–30]. In this research, we intended to examine the plasma exosome-derived Cx43 level and determine the prognostic value of patients with melanoma.
## Patients
We collected blood specimens from 112 patients with melanoma patients from the Suzhou Municipal Hospital, from October 2011 to March 2016. All patients were diagnosed with malignant melanoma by surgical resection and pathology; the course of the disease is 1 ~ 2 years, and there is no treatment. Exclusion criteria: patients treated with surgery, radiotherapy, and chemotherapy; Patients with other malignant tumor diseases; Insanity and contraindications to the study drugs [31]. The patients’ follow-up occurred for a median duration of 65.5 months (range: 13.0–119.0 months) ending in March 2021. The statistics for survival were collected from follow-up records, and DFS and OS were computed.
During the same period, specimens from 50 healthy subjects (with a median age of 64 years, ranging from 48 to 72 years) were taken from the Suzhou Municipal Hospital as healthy controls. After receiving the informed consent of patients or their families and the subjects provided formal informed consent, all specimens were included in the experiment. Approval of the research protocol was gotten from the Ethics Committees of the Suzhou Municipal Hospital, (identification nos. HMU [Ethics] 2022 − 120).
## Plasma exosome isolation
The plasma samples were taken out at -80℃, centrifuged at 4℃ for 2000 g/min for 10 min in order to remove cell fragments, and then centrifugated at 4℃ for 10,000 g/min for 30 min so as to remove macromolecular impurities. The exosome was purified with Beckmann over speed centrifuge at 4℃ at 110,000×g for 70 min, and then the final concentration of 8 mol/L urea was added. A protease inhibitor was added (lysate: protease inhibitor: 50:1). Samples were mixed, ultrasonic cracked, centrifuged at 14,000×g at 4℃ for 20 min. We then performed protein quantification and SDS-PAGE quality control.
## Transmission electron microscopy (TEM)
The exosomes were diluted and filtered. The sample of 5uL was taken and dropped on the copper mesh followed by incubation at an ambient temperature for 5 min. After incubation, we used absorbent paper to drain excess liquid on one side, and then added a drop of $2\%$ uranyl acetate to the copper mesh ensued by incubation at an ambient temperature for 1 min. After incubation, we drained excess liquid on one side with absorbent paper and dried it at room temperature for about 20 min. Finally, we observed the appearance of exosomes utilizing an electron microscope, captured images, and recorded.
## Nanoparticle tracking analysis (NTA)
The specimens were frozen and then thawed in a water bath at a temperature of 25 °C before being placed on ice. Exosome specimens were diluted in 1x PBS before being utilized to detect NTA (ZetaVIEW S/N 17–310). To examine particle mobility and determine the number of exosomes, NTA Software (ZetaView 8.04.02) was utilized.
## Western blotting (WB)
Use a 1.5 mm glass plate and a 15-well sample comb to prepare a $12\%$ separated gel and $5\%$ concentrated gel. Perform the electrophoresis at 80 V of the stabilized voltage until Loading Buffer enters the separating gel, and then continue the electrophoresis at 120 V until the Loading Buffer reaches the bottom of the gel. Select the PVDF membrane with a constant current of 200 mA and membrane transfer time of 90 min. Block the PVDF membrane with PBST diluted $5\%$ skim milk powder for 1 h, wash three times using PBST for 10 min for each washing. Put the membrane into the hybridization box, add the primary antibody, and put it in the decolorizing shaker at 4℃ overnight, including CD63, CD9, and TSG101(Santa Cruz Biotechnology, TX, USA). Remove the primary antibody and wash the membrane with PBST 3 times for 10 min for each wash. Next, add the secondary antibody, place the hybridization box on the shaker, shake slowly and incubate at room temperature for 1 h. Afterward, withdraw the secondary antibody and wash the membrane 3 times with PBST for 10 min during each wash. Add the appropriate amount of ECL luminescent solution and use a digital imaging system to take continuous photos of the membrane.
## ELISA
Diluted the plasma samples with 1×PBS buffer (1:500 dilution). The exosomes should be precipitated with 100 mL RIPA lysate on ice for a duration of 30 min. After shaking and mixing, dilution of the specimens was done using the PBS buffer at a ratio of 1:3. In the Cx43 antibody-coated microtiter plate, the blank and standard control were added. Incubated the 100 µl of the diluted exosomal sample at a temperature of 37 °C for a duration of 1 h, removed the fluid from the microplate by shaking, pat dry, added the solution, incubated again at a temperature of 37 °C for 1 h, then washed 3 times, added solution B, incubated yet again at a temperature of 37 °C for a duration of 30 min, washed 5 times, added 90 µl substrate, incubated at a temperature of 37℃ for a period of 15 min, added 50 µl of termination solution, and immediately measure the absorbance at 450 nm wavelength.
## Statistical analysis
The SPSS 24.0 Software (IBM) was utilized to conduct statistical analyses. Continuous information was displayed as counts (%ages), medians (ranges), or means ± standard deviations. Wilcoxon’s rank-sum tests or Chi-squared tests were conducted for correlation analysis. The log-rank was employed to determine the variations in OS and DFS between two cohorts. The prognostic significance of plasma exosome-derived Cx43 levels was evaluated using a ROC curve analysis. Statistically significant differences were defined as those with a P value of less than 0.05.
## Baseline characteristics of enrolled melanoma patients
Table 1 displays the baseline features of 112 patients with melanoma. There were 59 men ($52.68\%$) and 53 women ($47.32\%$); 64 ($57.14\%$) were ≥ 60 years old while 48 ($42.86\%$) were < 60 years old. In 98 instances ($87.50\%$), the tumor size was < 10 cm, and in 14 cases, it was ≥ 10 cm ($12.50\%$). With regards to tumor location, eighty-six melanomas ($76.79\%$) were found in the skin, whereas twenty-six ($23.21\%$) were found in the viscera or mucosa. Fifty-two melanomas ($46.43\%$) were Clark level I–III and 60 ($53.57\%$) were Clark level IV–V, and 50 ($44.64\%$) were stage 0–IIa, and 62 ($55.36\%$) were stage IIb–IV. There were 63 cases ($56.25\%$) with lymph node metastasis.
Table 1Baseline characteristics of enrolled melanoma patientsCharacteristicMelanoma patients ($$n = 112$$) Gender Male59(52.68)Female53(47.32) Age(years) < 6048(42.86)≥ 6064(57.14) Tumor diameter (cm) < 1098(87.50)≥ 1014(12.50) Tumor location Skin86(76.79)Mucous membrane, viscera26(23.21) Depth of tumor invasion(Clark level) I-III52(46.43)IV-V60(53.57) Tumor stage 0-IIa50(44.64)IIb-IV62(55.36) Lymph node metastasis YES63(56.25)NO49(43.75)
## Characterization of plasma exosomes in melanoma patients
The observation results under transmission electron microscopy showed that the background was clear, exosomes aggregated and distributed in the field of vision, with relatively uniform and relatively full size. The diameter ranged from 100 to 200 nanometers. The shape of exosome was a double disc -like vesicle structure with a full lipid envelop. The particle size of exosomes was detected by NTA, and the overall particle median size was around 100 nm, with the majority of the particles distributed between 50 and 200 nm (Fig. 1A). A limited number of particles with diameters ranging from 0 to 50 nm were found, and none had a diameter larger than 300 nm (Fig. 1B). The expression levels of exosomes marker proteins CD63, CD9, and TSG101 were detected by Western blot. The result suggested that CD63, CD9, and TSG101 were positive in the Exo cohort (Fig. 1C). The results indicated that exosomes can be successfully isolated and used in the subsequent experiments of this study.
Fig. 1Exosome characterization
## The levels of plasma exosome-derived Cx43 in melanoma patients
As opposed to the plasma exosome-derived Cx43 levels in healthy controls, the levels of plasma exosome-derived Cx43 in patients with melanoma were substantially downregulated (0.66 ± 0.05 mmol/L vs.0.71 ± 0.03 mmol/L, $P \leq 0.001$; Fig. 2A). Kaplan–Meier (KM) analysis illustrated that OS and DFS were poorer in patients with melanoma who exhibited reduced levels of plasma exosome-derived Cx43 as opposed to the patients who had elevated levels (both $P \leq 0.001$; Fig. 2B, C).
Fig. 2The levels of plasma exosome-derived Cx43 in melanoma patients
## Relationship between plasma exosome-derived Cx43 and pathological features in patients with melanoma
There was no significant difference in sex or age in the levels of plasma exosome-derived Cx43 between patients with melanoma and healthy control subjects (both $P \leq 0.05$; Fig. 3A, B). Nevertheless, the levels of plasma exosome-derived Cx43 in patients with melanoma whose tumor was situated in the skin, with a size of < 10 cm, Clark level I–III, TNM stage IIb–IV, and had no lymph node metastasis were considerably elevated as opposed to patients with a tumor located in the viscera or mucosa, with a size of ≥ 10 cm, Clark level IV–V, TNM stage IIb–IV and had lymph node metastasis (all $P \leq 0.05$, Fig. 3C–G).
Fig. 3Relationship between plasma exosome-derived Cx43 levels and pathological features in patients with melanoma
## Prognostic value of plasma exosome-derived Cx43 levels for predicting 5-year DFS and OS of melanoma patients
The ROC of plasma exosome-derived Cx43 for forecasting the 5-year DFS of patients with melanoma was 0.78 ($95\%$ CI: 0.70–0.86). ( Fig. 4A; Table 2). The positive predictive value and positive likelihood ratio were 97.70 ($95\%$ CI: 91.90–99.70) and 3.67 ($95\%$ CI: 1.10–12.50), sequentially, with a threshold value of 0.69 mmol/L with a specificity of $77.78\%$ ($95\%$ CI: 40.0–$97.20\%$) and sensitivity of $81.55\%$ ($95\%$ CI: 72.70–$88.50\%$), the negative predictive value and negative likelihood ratio were 26.90 ($95\%$ CI: 11.60–47.80) and 0.24 ($95\%$ CI: 0.10–0.40), respectively.
Fig. 4Prognostic value of plasma exosome-derived Cx43 levels for predicting 5-year DFS and OS of melanoma patients Table 2The prognostic value of plasma exosome-derived Cx43 levels in melanoma patientsVariableprognostic value ($$n = 112$$) 5-year DFS AUROC0.78($95\%$ CI: 0.70–0.86)Cutoff value ($95\%$CI)0.69Sensitivity, %81.55($95\%$ CI: 72.70–88.50)Specificity, %77.78($95\%$ CI: 40.00-97.20)Positive predictive value, %97.70($95\%$ CI: 91.90–99.70)Negative predictive value, %26.90($95\%$ CI: 11.60–47.80)Positive likelihood ratio3.67($95\%$ CI: 1.10–12.50)Negative likelihood ratio0.24($95\%$ CI: 0.10–0.40) 5-year OS AUROC0.77($95\%$ CI: 0.68–0.84)Cutoff value ($95\%$CI)0.67Sensitivity, %65.98($95\%$ CI: 55.70–75.30)Specificity, %80.00($95\%$ CI: 51.90–95.70)Positive predictive value, %95.50($95\%$ CI: 87.50–99.10)Negative predictive value, %26.70($95\%$ CI: 14.60–41.90)Positive likelihood ratio3.30($95\%$ CI: 1.20–9.20)Negative likelihood ratio0.43($95\%$ CI: 0.30–0.60) The ROC of plasma exosome-derived Cx43 for forecasting melanoma patients’ 5-year OS was found to be 0.77 ($95\%$ CI: 0.68–0.84) (Fig. 4B; Table 2). The positive predictive value and positive likelihood ratio were 95.50 ($95\%$ CI: 87.50–99.10) and 3.30 ($95\%$ CI: 1.20–9.20), correspondingly, with a threshold of 0.67 mmol/L with a specificity of $80.0\%$ ($95\%$ CI: 51.90–$95.70\%$) and sensitivity of $65.98\%$ ($95\%$ CI: 55.70–$75.30\%$), the negative predictive value and negative likelihood ratio were 26.70 ($95\%$ CI: 14.60–41.90) and 0.43 ($95\%$ CI: 0.30–0.60), correspondingly.
## Discussion
Cx43 protein performs a vital function in the genesis, invasion, and metastasis of tumors. Cx43 is synthesized in ribosomes attached to the endoplasmic reticulum and is transported via the Golgi to the cell membrane to perform its function as a linker [32]. Phosphorylation of Cx43 confers specific functions. The abnormal expression or localization of Cx43 is related to the alteration of homogeneous or heterogeneous gap junction intercellular communication in malignant cells. Studies have confirmed that the mechanism of the decrease of intercellular junction protein in tumor cells may be related to gene silencing, gene mutation, and post-translational modification [33]. Tang et al. found that the expression of Cx43 was reduced in $78.3\%$ of patients with primary gastric cancer, and the decreased expression of Cx43 was linked to advanced lymph node metastasis and clinical stage. Low expression of Cx43 was proved to be beneficial to the progression of primary gastric cancer [34]. Dominguez et al. confirmed that the aberrations of Cx43 expression are associated with thyroid papillary carcinoma [35]. Caltabiano et al. also suggested that the loss of functional Cx43 would lead to more malignant phenotypes in astrocytic brain tumors [36].
Wang et al. showed that Cx43 expression was lower in melanoma than in human epidermal melanocytes [37]. The overexpression of Cx43 significantly inhibits the proliferation as well as colony formation of melanoma cells in vitro. However, the effect of Cx43 knockout on cell proliferation and colony formation is reversed. Bioinformatics prediction and luciferase-reported gene assay have shown that miR-106a targets the 3′ untranslated regions of Cx43 and regulates its mRNA and protein expression levels in melanoma cells. The expression level of miR-106a is upregulated in melanoma cells, and its overexpression attenuates the effect of upregulated expression of Cx43. Ableser et al. found that Cx43 plays an anticancer role in the development of melanoma [38]. Numerous research reports have confirmed that exosomes are vesicle-like bodies secreted by cells that can help the immune escape of tumor cells and promote tumor cell metastasis [39–41]. Cx43 can be used as a marker for differential diagnosis of tumors. Exosomes can activate the immune system to inhibit tumor development, and can also be used as a potential natural carrier to deliver miRNA and chemotherapeutic drugs to tumor cells. To date, the expression, as well as prognostic value of exosomes in melanoma, are yet to be elucidated. The current research is thought to be the first to investigate about the prospective use of plasma exosome-derived CX43 in the diagnosis of melanoma.
In this study, we isolated plasma exosome-derived Cx43 from patients with melanoma and healthy control subjects and identified them with TEM, NTA, and WB. TEM illustrated that the background was clear, exosomes aggregated and distributed in the field of vision, with relatively uniform and relatively full size. The diameter ranged from 100 to 200 nm. The shape of exosome was a double disc-like vesicle structure with a full lipid envelope. The overall particle median size was around 100 nm, with the majority of the particles falling between 50 and 200 nm. NTA showed that only a few particles were distributed within the range of 0 to 50 nm in diameter, and no diameter exceeded 300 nm. Western blot suggested that CD63, CD9, and TSG101 were positive in the exosomal cohort. The results indicated that exosomes can be successfully isolated and used in the subsequent experiments of this study.
The levels of plasma exosome-derived Cx43 were subsequently contrasted among patients with melanoma and healthy control subjects, which demonstrated a substantial downregulation in plasma exosome Cx43 in melanoma patients. KM analysis illustrated that melanoma patients who exhibited lower levels of plasma exosome-derived Cx43 had poorer OS and DFS. We also examined the associations between levels of plasma exosome-derived Cx43 and pathological features in individuals with melanoma. There was no significant difference in sex or age in the plasma exosome-derived Cx43 levels between the two cohorts. Nevertheless, the levels of plasma exosome-derived Cx43 in patients with melanoma whose tumor was situated in the skin, with a size of < 10 cm, Clark level I–III, TNM stage IIb–IV, and had no lymph node metastasis were considerably elevated as opposed to patients with a tumor located in the viscera or mucosa, with a size of ≥ 10 cm, Clark level IV–V, TNM stage IIb–IV and had lymph node metastasis.
In melanoma patients, we examined more precisely the prognostic significance of the plasma exosome-derived Cx43. The ROCs of plasma-exosome-derived Cx43 were 0.78 ($95\%$ CI: 0.70–0.86) and 0.77 ($95\%$ CI: 0.66–0.84), correspondingly, to forecast 5-year OS and DFS for patients with melanoma. All the preceding data demonstrate that it has success in forecasting five-year DFS and five-year OS of melanoma patients.
## Conclusion
Our study demonstrated that the levels of plasma exosome-derived Cx43 in patients with melanoma were considerably downregulated. The patients with melanoma exhibiting lower levels of plasma exosome-derived Cx43 were found to have poorer OS and DFS. This implies that plasma exosome-derived Cx43 levels could function as a prospective prognostic indicator for 5-year OS and 5-year DFS of patients with melanoma.
## Electronic supplementary material
Below is the link to the electronic supplementary material.
Supplementary Material 1
## References
1. Chattopadhyay C, Kim DW, Gombos DS, Oba J, Qin Y, Williams MD, Esmaeli B, Grimm EA, Wargo JA, Woodman SE. **Uveal melanoma: from diagnosis to treatment and the science in between**. *Cancer* (2016.0) **122** 2299-312. DOI: 10.1002/cncr.29727
2. Rastrelli M, Tropea S, Rossi CR, Alaibac M. **Melanoma: epidemiology, risk factors, pathogenesis, diagnosis and classification**. *In vivo* (2014.0) **28** 1005-11. PMID: 25398793
3. Eddy K, Chen S. **Overcoming immune evasion in melanoma**. *Int J Mol Sci* (2020.0) **21** 8984. DOI: 10.3390/ijms21238984
4. 4.Spencer KR, Mehnert JM. Mucosal melanoma: epidemiology, biology and treatment.Melanoma2016:295–320.
5. Quintanilla-Dieck MJ, Bichakjian CK. **Management of early-stage melanoma**. *Facial Plast Surg Clin* (2019.0) **27** 35-42. DOI: 10.1016/j.fsc.2018.08.003
6. Lee AY, Brady MS. **Neoadjuvant immunotherapy for melanoma**. *J Surg Oncol* (2021.0) **123** 782-8. DOI: 10.1002/jso.26229
7. 7.Quaresmini D, Guida M. Neoangiogenesis in melanoma: an issue in biology and systemic treatment.Front Immunol2020:2836.
8. Pham F, Dalle S. **Safety of pembrolizumab for resected stage III melanoma**. *Exp Opin Drug Saf* (2020.0) **19** 1221-7. DOI: 10.1080/14740338.2020.1811228
9. Swartzendruber JA, Nicholson BJ, Murthy AK. **The role of Connexin 43 in lung disease**. *Life* (2020.0) **10** 363. DOI: 10.3390/life10120363
10. Acosta ML, Nor MNM, Guo CX, Mugisho OO, Coutinho FP, Rupenthal ID, Green CR. **Connexin therapeutics: blocking connexin hemichannel pores is distinct from blocking pannexin channels or gap junctions**. *Neural regeneration research* (2021.0) **16** 482. DOI: 10.4103/1673-5374.290097
11. 11.Kameritsch P, Pogoda K. The role of connexin 43 and pannexin 1 during acute inflammation.Frontiers in Physiology2020:1397.
12. Subedi YP, Altenberg GA, Chang C-WT. **Advances in the development of connexin hemichannel inhibitors selective toward Cx43**. *Future Med Chem* (2021.0) **13** 379-92. DOI: 10.4155/fmc-2020-0291
13. Zefferino R, Piccoli C, Di Gioia S, Capitanio N, Conese M. **How cells communicate with each other in the tumor microenvironment: suggestions to design novel therapeutic strategies in cancer disease**. *Int J Mol Sci* (2021.0) **22** 2550. DOI: 10.3390/ijms22052550
14. Bišćanin A, Ljubičić N, Boban M, Baličević D, Pavić I, Bišćanin MM, Budimir I, Dorosulic Z, Duvnjak M. **CX43 expression in colonic adenomas and surrounding mucosa is a marker of malignant potential**. *Anticancer Res* (2016.0) **36** 5437-42. DOI: 10.21873/anticanres.11122
15. Teleki I, Szasz AM, Maros ME, Gyorffy B, Kulka J, Meggyeshazi N, Kiszner G, Balla P, Samu A, Krenacs T. **Correlations of differentially expressed gap junction connexins Cx26, Cx30, Cx32, Cx43 and Cx46 with breast cancer progression and prognosis**. *PLoS ONE* (2014.0) **9** e112541. DOI: 10.1371/journal.pone.0112541
16. Brockmeyer P, Jung K, Perske C, Schliephake H, Hemmerlein B. **Membrane connexin 43 acts as an independent prognostic marker in oral squamous cell carcinoma**. *Int J Oncol* (2014.0) **45** 273-81. DOI: 10.3892/ijo.2014.2394
17. 17.Braeuer RR, Zigler M, Villares GJ, Dobroff AS, Bar-Eli M. Transcriptional control of melanoma metastasis: the importance of the tumor microenvironment. In: Seminars in cancer biology: 2011:Elsevier; 2011:83–88.
18. Zhu L, Sun H-T, Wang S, Huang S-L, Zheng Y, Wang C-Q, Hu B-Y, Qin W, Zou T-T, Fu Y. **Isolation and characterization of exosomes for cancer research**. *J Hematol Oncol* (2020.0) **13** 1-24. DOI: 10.1186/s13045-020-00987-y
19. Nie H, Liao Z, Wang Y, Zhou J, He X, Ou C. **Exosomal long non-coding RNAs: emerging players in cancer metastasis and potential diagnostic biomarkers for personalized oncology**. *Genes & Diseases* (2021.0) **8** 769-80. DOI: 10.1016/j.gendis.2020.12.004
20. He X, Kuang G, Wu Y, Ou C. **Emerging roles of exosomal miRNAs in diabetes mellitus**. *Clin Transl Med* (2021.0) **11** e468. DOI: 10.1002/ctm2.468
21. Kok VC, Yu C-C. **Cancer-derived exosomes: their role in cancer biology and biomarker development**. *Int J Nanomed* (2020.0) **15** 8019. DOI: 10.2147/IJN.S272378
22. Xiang Z, Jiang B, Li W, Zhai G, Zhou H, Wang Y, Wu J. **The diagnostic and prognostic value of serum exosome-derived carbamoyl phosphate synthase 1 in HEV‐related acute liver failure patients**. *J Med Virol* (2022.0) **94** 5015-25. DOI: 10.1002/jmv.27961
23. Xu K, Zhang C, Du T, Gabriel ANA, Wang X, Li X, Sun L, Wang N, Jiang X, Zhang Y. **Progress of exosomes in the diagnosis and treatment of lung cancer**. *Biomed Pharmacother* (2021.0) **134** 111111. DOI: 10.1016/j.biopha.2020.111111
24. Wu J, Cui LL, Yuan J, Wang Y, Song S. **Clinical significance of the phosphorylation of MAPK and protein expression of cyclin D1 in human osteosarcoma tissues**. *Mol Med Rep* (2017.0) **15** 2303-7. DOI: 10.3892/mmr.2017.6224
25. Wu J, Zhang J-R, Qin J. **Clinical significance of methylation of E-cadherin and p14ARF gene promoters in skin squamous cell carcinoma tissues**. *Int J Clin Exp Med* (2014.0) **7** 1808. PMID: 25126184
26. 26.Wolfe W, Xiang Z, Yu X, Li P, Chen H, Yao M, Fei Y, Huang Y, Yin Y, Xiao H. The Challenge of Applications of Probiotics in Gastrointestinal Diseases. Advanced Gut & Microbiome Research 2023, 2023.
27. Lu H, Li Z, Liu L, Tao Y, Zhou Y, Mao X, Zhu A, Wu H, Zheng X. **A Pan-Cancer analysis of the oncogenic roles of RAD51 in human tumors**. *Adv Gut Microbiome Res* (2022.0) **2022** 1591377. DOI: 10.1155/2022/1591377
28. Zhou W, Zhou Y, Chen X, Ning T, Chen H, Guo Q, Zhang Y, Liu P, Zhang Y, Li C. **Pancreatic cancer-targeting exosomes for enhancing immunotherapy and reprogramming tumor microenvironment**. *Biomaterials* (2021.0) **268** 120546. DOI: 10.1016/j.biomaterials.2020.120546
29. Zhang J, Fu B, Li M, Mi S. **Secretome of activated fibroblasts Induced by Exosomes for the Discovery of biomarkers in Non-Small Cell Lung Cancer**. *Small* (2021.0) **17** 2004750. DOI: 10.1002/smll.202004750
30. Zhao X, Ren Y, Lu Z. **Potential diagnostic and therapeutic roles of exosomes in pancreatic cancer**. *Biochim et Biophys Acta (BBA)-Reviews Cancer* (2020.0) **1874** 188414. DOI: 10.1016/j.bbcan.2020.188414
31. 31.Keilholz U, Ascierto P, Dummer R, Robert C, Lorigan P, Van Akkooi A, Arance A, Blank C, Sileni VC, Donia M. ESMO consensus conference recommendations on the management of metastatic melanoma: under the auspices of the ESMO Guidelines Committee. Annals of Oncology 2020, 31(11):1435–1448.
32. Lim WY, Madden LE, Becker DL. **Pulpal upregulation of connexin 43 during pulpitis**. *Clin Oral Invest* (2021.0) **25** 1327-35. DOI: 10.1007/s00784-020-03439-6
33. Katturajan R, Prince SE. **A role of connexin 43 on the drug-induced liver, kidney, and gastrointestinal tract toxicity with associated signaling pathways**. *Life Sci* (2021.0) **280** 119629. DOI: 10.1016/j.lfs.2021.119629
34. Tang B, Peng Z-H, Yu P-W, Yu G, Qian F. **Expression and significance of Cx43 and E-cadherin in gastric cancer and metastatic lymph nodes**. *Med Oncol* (2011.0) **28** 502-8. DOI: 10.1007/s12032-010-9492-5
35. Dominguez C, Karayan-Tapon L, Desurmont T, Gibelin H, Crespin S, Fromont G, Levillain P, Bouche G, Cantereau A, Mesnil M. **Altered expression of the gap junction protein connexin43 is associated with papillary thyroid carcinomas when compared with other noncancer pathologies of the thyroid**. *Thyroid* (2011.0) **21** 1057-66. DOI: 10.1089/thy.2011.0041
36. Caltabiano R, Torrisi A, Condorelli D, Albanese V, Lanzafame S. **High levels of connexin 43 mRNA in high grade astrocytomas. Study of 32 cases with in situ hybridization**. *Acta Histochem* (2010.0) **112** 529-35. DOI: 10.1016/j.acthis.2009.05.008
37. Wang J, Li H, Zhang J, Zhang C, Hou X. **Suppression of connexin 43 expression by miR-106a promotes melanoma cell proliferation**. *Eur Rev Med Pharmacol Sci* (2019.0) **23** 965-71. PMID: 30779062
38. Ableser MJ, Penuela S, Lee J, Shao Q, Laird DW. **Connexin43 reduces melanoma growth within a keratinocyte microenvironment and during tumorigenesis in vivo**. *J Biol Chem* (2014.0) **289** 1592-603. DOI: 10.1074/jbc.M113.507228
39. Tittarelli A, Navarrete M, Lizana M, Hofmann-Vega F, Salazar-Onfray F. **Hypoxic melanoma cells deliver microRNAs to dendritic cells and cytotoxic T lymphocytes through connexin-43 channels**. *Int J Mol Sci* (2020.0) **21** 7567. DOI: 10.3390/ijms21207567
40. 40.Meng S, Fan X, Zhang J, An R, Li S. GJA1 expression and its prognostic value in cervical cancer. Biomed Res Int 2020, 2020.
41. 41.Ling Z, Xiao H, Chen W. Gut Microbiome: The Cornerstone of Life and Health. In., vol. 2022: Hindawi; 2022.
|
---
title: 'Risk factors for thrombolysis-related intracranial hemorrhage: a systematic
review and meta-analysis'
authors:
- Jiana Chen
- Zhiwei Zeng
- Zongwei Fang
- Fuxin Ma
- Meina Lv
- Jinhua Zhang
journal: Thrombosis Journal
year: 2023
pmcid: PMC10012586
doi: 10.1186/s12959-023-00467-6
license: CC BY 4.0
---
# Risk factors for thrombolysis-related intracranial hemorrhage: a systematic review and meta-analysis
## Abstract
### Background
Thrombolysis-related intracranial hemorrhage has a high mortality rate, and many factors can cause intracranial hemorrhage. Until now, systematic reviews and assessments of the certainty of the evidence have not been updated.
### Aim
We conducted a systematic review to identify risk factors for thrombolysis-related intracranial hemorrhage.
### Method
The protocol for this systematic review was prospectively registered with PROSPERO (CRD42022316160). All English studies that met the inclusion criteria published before January 2022 were obtained from PubMed, EMBASE, Web of Science, and Cochrane Library. Two researchers independently screened articles, extracted data, and evaluated the quality and evidence of the included studies. Risk factors for intracranial hemorrhage were used as the outcome index of this review. Random or fixed-effect models were used in statistical methods.
### Results
Of 6083 citations, we included 105 studies in our analysis. For intracranial hemorrhage, moderate-certainty evidence showed a probable association with age, National Institutes of Health stroke scale, leukoaraiosis, hypertension, atrial fibrillation, diabetes, total cholesterol, proteinuria, fibrinogen levels, creatinine, homocysteine, early infarct signs, antiplatelet therapy and anticoagulant therapy; In addition, we found low-certainty evidence that there may be little to no association between risk of intracranial hemorrhage and weight, sex, platelet count, uric acid, albumin and white matter hyperintensity. Leukoaraiosis, cardiovascular disease, total cholesterol, white blood cell count, proteinuria, fibrinogen levels, creatinine, homocysteine and early CT hypodensities are not included in most intracranial hemorrhage risk assessment models.
### Conclusion
This study informs risk prediction for thrombolysis-related intracranial hemorrhage, it also informs guidelines for intracranial hemorrhage prevention and future research.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12959-023-00467-6.
## Introduction
Thrombolytic drugs, especially rt-PA (e.g., alteplase), are the most effective pharmacological therapy for acute ischemic stroke (AIS), to increase survival and reduce morbidity [1, 2]. However, the risk of severe hemorrhagic transformation in patients treated with rt-PA also increased [3]. Intracranial hemorrhage (ICH), also called cerebral hemorrhage, is the most serious complication of stroke thrombolysis and an important obstacle to generalized thrombolytic therapy [4]. ICH has been reported in $1.7\%$ to $8.8\%$ of patients with acute ischemic stroke treated with iv thrombolysis, mortality and morbidity rates increase in patients with symptomatic ICH [5–7]. Therefore, it is necessary to accurately predict the bleeding risk of patients, which will help physicians weigh the benefits and risks of thrombolytic therapy and reduce the occurrence of intracranial hemorrhage.
The risk assessment model (RAM) for thrombolysis-related intracranial hemorrhage consists of a combination of multiple predictors. Risk for specified endpoints can be obtained based on the relevant predictors to inform recommendations for strata of patients [8]. In clinical treatment or medication decisions, we can apply relevant models for risk prediction to reduce the occurrence of intracranial hemorrhage. Therefore, establishing and using a thrombolysis-related intracranial hemorrhage model is crucial.
RAMs are currently available for patients on thrombolytic therapy, which can be scored and stratified according to risk factors. Although these models can prevent intracranial hemorrhage to some extent, most of them were developed using existing data that were not based on a systematic review of all potential risk factors [9]. However, model development requires a systematic review to determine the importance of risk factors [9]. Predictors included in existing models were not comprehensive, and effect sizes of the risk factors were not subjected to meta-analysis, which may reduce the predictive power of the model.
Therefore, this review included studies of thrombolysis-related intracerebral hemorrhage models and risk factors to conduct a systematic review and meta-analysis of risk factors for intracerebral bleeding that may inform treatment, future guideline recommendations, and the development of RAMs.
## Search strategy
The protocol for this systematic review was prospectively registered with PROSPERO (CRD42022316160). Data were reviewed from four databases: PubMed, EMBASE, Web of Science, and the Cochrane Library. Studies in English published before January 2022 were included. The groups of search keywords included were: [1] thrombolysis OR thrombolytic drug OR thrombolytic agent OR fibrinolytic agent OR fibrinolytic drug OR thrombolytic therapy; [2] intracerebral hemorrhage OR ICH OR cerebral hemorrhage OR hemorrhagic infarction OR subarachnoid hemorrhage OR subdural hemorrhage OR epidural hemorrhage; and [3] prediction model OR predict* OR risk prediction OR risk factor. A detailed search strategy is presented in Supplemental Material 1.
## Study selection
Studies were selected independently by two researchers and checked to prevent potential errors. A third independent researcher resolved disputes arising in the process of study selection. Studies that met the following criteria were included: [1] use of thrombolytic drugs [e.g., Tissue plasminogen activator (tPA), urokinase]; [2] comparison between the ICH group and Non-ICH group; and [3] the outcome index was risk factors or predictors. Studies that met the following criteria were excluded: [1] patients with ICH treated with non-thrombolytic drugs; [2] no access to data (including no data related to the risk factors in the study, the study was in the design or recruitment stage, no permission to use the data had been granted, contacted the corresponding author but no reply had been received).
## Data extraction
Data were extracted independently by two researchers and checked to prevent potential errors. A third independent researcher resolved disputes arising in the process of data extraction. The data extracted included the name of the first author, year of publication, time frame, population and their demographics (e.g., sample size, number of centers, age, and sex), study design (e.g., cohort or case–control), type of prediction model study (development, validation, and impact), outcomes and measures of association [e.g., odds ratio (OR) or risk ratio (RR) or hazard ratio (HR), $95\%$ confidence interval (CI) and P-value].
## Risk of bias assessment
We assessed the risk of bias in the included studies by using the Prediction Study Risk of Bias Assessment Tool (PROBAST) for RAM studies [10] and the Quality in Prognosis Studies tool (QUIPS) for prognostic factor studies [11–13].
The risk of bias was serious across all identified studies, each presenting risk of bias in at least 1 domain or item (Supplemental table 4). Among the 105 included studies, 72 were retrospective, which may have introduced classification bias [S2-3, S5-8, S11-14, S16, S18-20, S22, S24-28, S32-34, S36-38, S40-42, S44-46, S48, S50-51, S53, S55, S58, S60-68, S70-75, S78, S80-83, S86-93, S96, S99, S101-103, S105]. Certainty in evidence was downgraded for imprecision, given that the confidence interval suggests that there may be no association. 9 of the 22 prediction model studies and 4 of the 83 risk factors studies did not have a clear description of appropriate outcome measurement [S5-7, S9-10, S14-16, S21, S34, S41, S75, S105]. Supplemental tables 2 and 3 provide the detailed judgements for each of the risk of bias domain criteria.
## Certainty of evidence assessment
We assessed the certainty of the evidence for each of the risk factors per outcome, based on the GRADE approach [14]. The approach considers the following domains: risk of bias, indirectness, inconsistency, imprecision, and publication bias. We developed evidence profiles and rated the overall certainty of evidence as high, moderate, and low or very low, depending on the grading of the individual domains [14]. We narratively described the strength of the association using the terms “there is,” “there probably is,” or “there may be,” depending on whether the quality of the evidence was “high,” “moderate,” or “low/very low,” respectively.
## Statistical analysis
We standardized each risk factor by log-transformation and unifying the direction of the predictors [15]. In studies that reported the measure of association as hazard ratio or risk ratio, we converted them to OR using the baseline risk reported in the studies [16, 17]. We conducted a meta-analysis of associations using the generic inverse variance-based method to produce an overall measure of association. We used the Review Manager 5.3 software for meta-analysis. The statistical indicators were odds ratio (OR) and $95\%$ confidence interval (CI). The Chi-Square test (χ2) test was used to test the heterogeneity of results. If P ≥ 0.1 and I2 ≤ $50\%$, the fixed-effect model was used for meta-analysis. The random-effect model was used when $P \leq 0.1$ and I2 > $50\%$. To explore the stability of the results, we conducted a sensitivity analysis by eliminating studies one by one.
## The characteristics of included studies
A total of 6083 articles were retrieved based on the search criteria. After screening, 105 articles met the inclusion criteria and were analyzed [S1-105]. The flow chart and results of the screening are shown in Fig. 1. Supplemental table 1 describes the characteristics of the included studies reporting on the outcomes of ICH. 22 studies were prediction model development studies [S1-22], and 83 were risk factor studies [S23-105]. 95 studies were cohorts [S1, S4-5, S7-49, S51-65, S67-69, S72-78, S80-84, S86-88, S90-105], 43 of which were multicenter [S1, S4-5, S7-15, S19, S21-23, S25, S31, S34-36, S40, S49, S55-58, S61-64, S67, S74-75, S91-94, S100, S102-105]; 10 were case–control studies [S2, S3, S6, S50, S66, S70-71, S79, S85, S89], 6 of which were multicenter [S3, S6, S70, S79, S85, S89]. Most of the patients were between 50 and 80 years old, and most of them were male. Among the 105 studies, the populations of 96 studies were stroke patients [S1-6, S8-20, S22-41, S43-55, S57-66, S68-74, S76-78, S80-84, S86-104], 6 were in patients with myocardial infarction [S21, S56, S67, S75, S79, S85], one was in patients with pulmonary embolism [S7], one was in patients with major artery occlusion [S42], and one was in patients with deep venous thrombosis [S105].Fig. 1Flow chart and results of literature screening
## Analysis of risk factors of thrombolysis-related ICH
Investigated were 110 candidate risk factors for ICH from 105 studies. Supplemental table 2 provides the evidence profile for risk factors of thrombolysis-related ICH. Supplemental figure (sFigure) 1–110 provides the forest plots of the meta-analysis of each of the risk factors.
## Demographic factors
We found moderate-certainty evidence that there is probably an association between risk of ICH and age (OR, 1.77; $95\%$CI, 1.52–2.07) [S5, S7, S9-10, S12-14, S17, S19, S21, S23, S56, S65, S67, S73-75, S77, S79, S83, S91, S105] and race (OR, 1.86; $95\%$CI; 1.40–2.45) [S5, S38, S75]. We found low-certainty evidence that there may be little to no association between risk of ICH and body weight < 70 kg (OR, 1.26; $95\%$CI, 0.96–1.67) [S9-10, S21, S72, S75, S79, S85] and sex (OR, 0.96; $95\%$CI, 0.63–1.47) [S5, S56, S65, S75, S105]. See sFigure 1–4 for details.
## Functional factors
We found moderate-certainty evidence that there is probably an association between risk of ICH and the Alberta Stroke Programme Early CT Score (ASPECTS) ≤ 7 (OR, 1.97; $95\%$CI, 1.25–3.12) [S1-3, S47, S64], National Institutes of Health stroke scale (NIHSS) (OR, 1.27; $95\%$CI, 1.22–1.33) [S4-6, S8-10, S12-20, S22, S24, S27-30, S36-38, S47, S50, S64-65, S69, S73-74, S77-78, S81, S83-84, S89, S91, S96-97, S99], modified Rankin scale (mRS) > 2 (OR, 1.65; $95\%$CI, 1.19–2.27) [S15, S53], Thrombolysis in Cerebral Infarction (TICI) score (3, 2, 1, 0; Each one decrease) (OR, 1.82; $95\%$CI, 1.04–3.18) [S34], CHADS2 score > 2 (OR, 14.00; $95\%$CI, 1.59–123.28) [S47], low ejection fraction (EF) (OR, 16.22; $95\%$CI, 2.89–91.03) [S47], higher SEDAN score (OR, 9.25; $95\%$CI, 2.37–36.10) [S47], arterial stiffness index (ASI) (OR, 1.90; $95\%$CI, 1.09–3.31) [S54] and K trans (The contrast volume transfer coefficient) (OR, 5.04; $95\%$CI, 2.01–12.64) [S102]. We found low-certainty evidence that there may be little to no association between risk of ICH and apparent diffusion coefficient (ADC) (OR, 2.72; $95\%$CI, 0.43–17.14) [S40, S42]. See sFigure 5–14 for details.
## Medical illness and patient history factors
We found moderate-certainty evidence that there is probably an association between risk of ICH and peripheral vascular disease (PVD) (OR, 1.59; $95\%$CI, 1.12–2.26) [S7], cerebral small vascular diseases (CSVD) (OR, 2.69; $95\%$CI, 1.98–3.66) [S20, S38, S70-71, S100], cerebral microbleeds (OR, 2.72; $95\%$CI, 1.45–5.10) [S38], leukoaraiosis (OR, 2.61; $95\%$CI, 1.74–3.91) [S70-71, S100], poor collaterals (OR, 4.36; $95\%$CI, 1.82–10.41) [S46, S90], recent facial or head trauma (2 weeks) (OR, 13.00; $95\%$CI, 3.40–49.70) [S85], cerebral artery occlusion (OR, 8.52; $95\%$CI, 3.20–22.64) [S37] and decreased levels of consciousness (OR, 2.36; $95\%$CI, 1.51–3.68) [S33, S37, S101]. We found low-certainty evidence that there may be little to no association between risk of ICH and stroke (OR, 4.68; $95\%$CI, 1.49–14.70) [S7, S10, S38, S47, S56, S58, S61, S75, S100, S105]. See sFigure 15–23 for details.
We found moderate-certainty evidence that there is probably an association between risk of ICH and cardiovascular disease (OR, 2.09; $95\%$CI, 1.75–2.49) [S4, S7-9, S13-14, S16, S19-23, S30, S37, S47, S58, S61, S77, S83, S91], prior myocardial infarction (OR, 1.80; $95\%$CI, 1.33–2.44) [S7], valvular heart diseases (OR, 2.09; $95\%$CI, 1.07–4.08) [S8], hypertension (OR, 1.42; $95\%$CI, 1.21–1.67) [S9, S13-14, S19-21, S58], atrial fibrillation (AF) (OR, 2.62; $95\%$CI, 1.92–3.59) [S4, S13-14, S16, S19-20, S22, S30, S37, S47, S61, S77, S83], congestive heart failure (OR, 2.57; $95\%$CI, 1.16–5.69) [S23] and diabetes (OR, 1.84; $95\%$CI, 1.34–2.51) [S13-14, S20, S90-91]. We found low-certainty evidence that there may be little to no association between risk of ICH and dyslipidemia (OR, 1.18; $95\%$CI, 0.57–2.47) [S19, S74, S87] and visual field deficits (OR, 1.07; $95\%$CI, 0.29–3.91) [S33]. In addition, we found very low-certainty evidence that there may be little to no association between risk of ICH and smoke (OR, 0.47; $95\%$CI, 0.02–14.61) [S77, S86]. See sFigure 24–33 for details.
## Laboratory and physical examination factors
We found high-certainty evidence that there is an association between risk of ICH and thrombin-activated fibrinolysis inhibitor (TAFI) (OR, 12.90; $95\%$CI, 1.41–118.01) [S39] and plasminogen activator inhibitor (PAI)-1 (OR, 12.75; $95\%$CI, 1.17–138.95) [S39]. We found moderate-certainty evidence that there is probably an association between risk of ICH and blood sugar (OR, 1.14; $95\%$CI, 1.10–1.20) [S1, S5-6, S9-10, S12, S16-19, S22, S30, S32, S34, S47, S57, S65, S83-84, S91, S97], Platelet derived growth factor-CC (PDGF-CC) (OR, 1.03; $95\%$CI, 1.00–1.06) [S98], blood pressure (OR, 2.59; $95\%$CI, 1.07–6.27) [S99], Systolic blood pressure (SBP) (OR, 1.15; $95\%$CI, 1.10–1.20) [S5, S8-10, S12, S17, S19, S23, S28, S36, S56, S75-77, S97, S99], pulse pressure (OR, 2.37; $95\%$CI, 1.01–5.57) [S19, S67], international normalized ratio (INR) (OR, 2.47; $95\%$CI, 1.34–4.55) [S30, S75] and activated partial thromboplastin time (APTT) (OR, 2.13; $95\%$CI, 1.02–4.45) [S86]. We found low-certainty evidence that there may be little to no association between risk of ICH and mean platelet volume (MPV) (OR, 1.02; $95\%$CI, 1.00–1.04) [S50] and prothrombin time activity percentage (PTA) (OR, 1.02; $95\%$CI, 1.00–1.03) [S78]. See sFigure 34–44 for details.
We found moderate-certainty evidence that there is probably an association between risk of ICH and total cholesterol (TC) (OR, 0.91; $95\%$CI, 0.83–0.99) [S31], low density lipoprotein cholesterol (LDL-C) (OR, 0.87; $95\%$CI, 0.82–0.92) [S31, S50], high density lipoprotein cholesterol (HDL-C) (OR, 1.09; $95\%$CI, 1.01–1.18) [S31], triglyceride (TG) (OR, 0.84; $95\%$CI, 0.72–0.97) [S31], TC/HDL-C (OR, 1.73; $95\%$CI, 1.01–2.96) [S51], TG/HDL-C (OR, 2.06; $95\%$CI, 1.24–3.43) [S51], LDL-C/HDL-C (OR, 1.93; $95\%$CI, 1.07–3.50) [S51], white blood cell count (OR, 1.10; $95\%$CI, 1.01–1.19) [S78], absolute eosinophil count (AEC) (OR, 0.22; $95\%$CI, 0.07–0.72) [S81] and low serum-free triiodothyronine (fT3) (OR, 0.24; $95\%$CI, 0.11–0.51) [S26, S53]. We found low-certainty evidence that there may be an association between risk of ICH and neutrophil to lymphocyte ratio (NLR) (OR, 1.09; $95\%$CI, 1.00–1.18) [S22, S104]. See sFigure 45–55 for details.
We found high-certainty evidence that there is an association between risk of ICH and activated protein C (APC) (OR, 25.19; $95\%$CI, 4.76–133.3) [S59]. We found moderate-certainty evidence that there is probably an association between risk of ICH and albuminuria (OR, 1.66; $95\%$CI, 1.20–2.28) [S44, S49], fibrinogen (FIB) (OR, 6.64; $95\%$CI, 3.40–12.97) [S47, S73, S86, S88-89], fibrinogen degradation products (FDP) (OR, 7.50; $95\%$CI, 1.26–44.64) [S88], globulin (OR, 1.18; $95\%$CI, 1.09–1.29) [S78], caveolin (OR, 2.35; $95\%$CI, 1.71–3.24) [S52, S77], matrix metalloproteinase-9 (MMP9) / tissue inhibitor of metalloproteinases (TIMP) (OR, 1.72; $95\%$CI, 1.31–2.28) [S63], S100B (OR, 2.80; $95\%$CI, 1.40–5.60) [S92], cellular fibronectin (c-Fn) (OR, 2.10; $95\%$CI, 1.30–3.39) [S94], glomerular filtration rate (GFR) (OR, 1.83; $95\%$CI, 1.38–2.43) [S80, S105], creatinine (OR, 5.50; $95\%$CI, 1.08–28.01) [S93], homocysteine (OR, 13.65; $95\%$CI, 3.58–52.05) [S96], apelin (OR, 0.24; $95\%$CI, 0.09–0.68) [S103], Interleukin-1β (IL-1β) (OR, 1.06; $95\%$CI, 1.03–1.09) [S103], Interleukin-6 (IL-6) (OR, 1.53; $95\%$CI, 1.12–2.11) [S103], malondialdehyde (MDA) (OR, 2.49; $95\%$CI, 1.32–4.70) [S103] and superoxide dismutase (SOD) (OR, 0.20; $95\%$CI, 0.05–0.75) [S103]. See sFigure 56–72 for details.
We found low-certainty evidence that there may be little to no association between risk of ICH and platelet count (OR, 1.00; $95\%$CI, 0.98–1.01) [S6, S8, S15, S18, S28-29, S36, S86, S90], uric acid (UA) (OR, 1.00; $95\%$CI, 0.99–1.00) [S50], diastolic blood pressure (DBP) (OR, 1.43; $95\%$CI, 0.89–2.31) [S4, S85] and albumin (OR, 2.30; $95\%$CI, 0.89–6.00) [S50, S62, S95]. We found very low-certainty evidence that there may be little to no association between risk of ICH and glycated hemoglobin A1c (HbA1c) (OR, 3.36; $95\%$CI, 0.50–22.56) [S27, S62] and mean artery pressure (MAP) (OR, 3.68; $95\%$CI, 0.61–22.13) [S32, S47, S73]. See sFigure 73–78 for details.
We found moderate-certainty evidence that there is probably an association between risk of ICH and early computed tomography (CT) hypodensities (OR, 2.47; $95\%$CI, 1.54–3.95) [S23, S33], hyperdense middle cerebral artery (HDMCA) sign (OR, 1.57; $95\%$CI, 1.09–2.25) [S12, S33, S48, S76, S83], early infarct signs (OR, 2.86; $95\%$CI, 1.30–6.33) [S12, S32, S83], fluid-attenuated inversion recovery (FLAIR) hyperintensity (OR, 13.58; $95\%$CI, 3.72–49.60) [S25, S60], early CT signs of cerebral ischaemia (OR, 3.40; $95\%$CI, 2.19–5.27) [S6, S46, S55], brain infarction volume (OR, 1.81; $95\%$CI, 1.25–2.62) [S24, S35, S69], high-permeability region size on PCT (HPrs-PCT) (OR, 1.00; $95\%$CI, 1.00–1.00) [S29], supratentorial territory of the posterior cerebral artery (PCA) (OR, 4.31; $95\%$CI, 1.20–15.49) [S37], cerebral blood volume (CBV) (OR, 100.00; $95\%$CI, 3.20–3125.31) [S72] and calcification volume on the lesion side (CV-L) (OR, 1.50; $95\%$CI, 1.14–1.98) [S66]. We found low-certainty evidence that there may be little to no association between risk of ICH and white matter hyperintensity (WMH) (OR, 2.45; $95\%$CI, 0.95–6.32) [S45, S68]. See sFigure 79–89 for details.
We found moderate-certainty evidence that there is probably an association between risk of ICH and MMP-9-1562C/T polymorphism genotypes (OR, 13.08; $95\%$CI, 1.04–164.51) [S41], PAI-1 5G/5G genotype (OR, 4.75; $95\%$CI, 1.18–19.12) [S87], rs1801020, C allele (OR, 2.04; $95\%$CI, 1.38–3.01) [S4] and rs669, A allele (OR, 2.19; $95\%$CI, 1.57–3.06) [S4]. See sFigure 90–93 for details.
## Medication factors
We found moderate-certainty evidence that there is probably an association between risk of ICH and antithrombotic therapy (OR, 2.28; $95\%$CI, 1.81–2.87) [S2, S8-10, S12, S19, S21, S23, S32, S42, S46-47, S64, S75, S79, S84]. We found moderate-certainty evidence that there is probably an association between risk of ICH and thrombolytic therapy (OR, 2.00; $95\%$CI, 1.36–2.94) [S2, S21, S23, S42, S46, S75, S84]. Subgroup analysis showed that t-PA (OR, 2.33; $95\%$CI, 1.54–3.50) [S2, S21, S23, S42, S75, S84] and urokinase (OR, 1.06; $95\%$CI, 1.01–1.11) [S46] were statistically significant. We found moderate-certainty evidence that there is probably an association between risk of ICH and antiplatelet therapy (OR, 2.15; $95\%$CI, 1.70–2.72) [S8-10, S12, S32, S47, S64]. Subgroup analysis showed that single antiplatelet therapy (SAPT) (OR, 1.72; $95\%$CI, 1.52–1.93) [S8-10, S12, S32, S64] and dual antiplatelet therapy (DAPT) (OR, 3.54; $95\%$CI, 1.86–6.76) [S9, S12, S64] were statistically significant. We found moderate-certainty evidence that there is probably an association between risk of ICH and anticoagulant therapy (OR, 4.40; $95\%$CI, 1.38–14.01) [S12, S47, S79]. See sFigure 94–97 for details.
We found moderate-certainty evidence that there is probably an association between risk of ICH and antihypertensive drugs (OR, 1.63; $95\%$CI, 1.24–2.16) [S8, S36], lipid-lowering drugs (OR, 3.23; $95\%$CI, 2.33–4.48) [S17, S82, S99], microcatheter injection (MCI) (OR, 3.60; $95\%$CI, 1.12–11.57) [S34], additional endovascular therapy (OR, 8.71; $95\%$CI, 2.54–29.89) [S37], deviation from the protocol (OR, 11.10; $95\%$CI, 2.40–51.34) [S55], periventricular transit time to the peak (TTP) (OR, 4.74; $95\%$CI, 1.62–13.83) [S69] and vaspin (OR, 0.26; $95\%$CI, 0.12–0.58) [S103]. We found low-certainty evidence that there may be little to no association between risk of ICH and time from onset to treatment (OTT) (OR, 1.06; $95\%$CI, 0.99–1.15) [S4, S9, S12, S20, S90]. In addition, we found very low-certainty evidence that there may be little to no association between risk of ICH and time to recanalization (OR, 3.19; $95\%$CI, 0.18–55.75) [S18, S43]. See sFigure 98–106 for details.
## Other factors
We found moderate-certainty evidence that there is probably an association between risk of ICH and age & NIHSS (OR, 4.08; $95\%$CI, 2.69–6.18) [S11, S19], age & hypertension (OR, 2.10; $95\%$CI, 1.02–4.31) [S19] and age & DBP (OR, 6.10; $95\%$CI, 2.30–16.18) [S19]. We found low-certainty evidence that there may be little to no association between risk of ICH and age & weight (OR, 2.40; $95\%$CI, 0.90–6.40) [S19]. See sFigure 107–110 for details.
## Sensitivity analysis
Sensitivity analysis was conducted by eliminating studies one by one. There were no significant changes in the outcome except for body weight, ADC, dyslipidemia, smoke, HbA1c, DBP, MAP, albumin, early CT hypodensities, HDMCA sign, early infarct signs, brain infarction volume, WMH, anticoagulant therapy and time to recanalization, indicating that most of the results were stable.
## Summary of findings
We evaluated 110 risk factors for thrombolysis-related ICH. We found high-certainty evidence that there is an association between the risk of ICH and TAFI, PAI-1 and APC. We also identified several statistically significant predictors, such as age, age & NIHSS, ASPECT, NIHSS, cerebral small vascular diseases (CSVD), leukoaraiosis, cardiovascular disease, hypertension, AF, diabetes, blood sugar, SBP, INR, TC, LDL-C, HDL-C, TG, NLR, white blood cell count, AEC, low fT3, albuminuria, FIB, GFR, creatinine, homocysteine, early CT hypodensities, HDMCA sign, early infarct signs, FLAIR hyperintensity, early CT signs of cerebral ischemia, thrombolytic therapy, antiplatelet therapy, anticoagulant therapy, antihypertensive drugs and lipid-lowering drugs, which supported by moderate certainty of the evidence. And low-certainty evidence suggests that body weight, sex, sex & body weight, dyslipidemia, visual field deficits, platelet count, UA, DBP, albumin, WMH and time from onset to treatment (OTT) were not statistically significant. We found very low-certainty evidence that there may be little to no association between risk of ICH and ADC, smoke, HbA1c, MAP and time to recanalization. Therefore, in addition to thrombolytic therapy can affect ICH, other risk factors such as blood sugar, SBP, INR, TC, LDL-C, HDL-C, TG, fT3, albuminuria, GFR, creatinine, homocysteine, antiplatelet therapy, anticoagulant therapy, antihypertensive drugs and lipid-lowering drugs should also be paid attention to during treatment as a way to reduce the occurrence of ICH. We summarize and group (treatable vs. non-treatable) the different certainty-risk factors into a new table to permit easy reading. Please see Supplementary Table 5 for details.
## Implications for practice
Our study identified candidate risk factors for ICH, such as age, body weight, NIHSS, age & NIHSS, diabetes, hypertension, AF, blood sugar, platelet count, SBP, time from onset to treatment (OTT) and antiplatelet therapy that have been considered in the analysis of some developed and widely used RAMs in daily practice, such as the GRASPS, SICH, SITs, SITs-MOST, SPAN-100, STARTING-SICH, THRIVE-C and RICH models [S5, S8-11, S12, S14, S17]. However, some factors that we identified as having a probable association with ICH, based on our meta-analysis results, were not included or considered in the development of most of the RAMs, such as stroke, decreased levels of consciousness, cerebral artery occlusion, poor collaterals, leukoaraiosis, early CT hypodensities, HDMCA sign, FLAIR hyperintensity, LDL-C, INR, brain infarction volume, low fT3, albuminuria, FIB, GFR and anticoagulant therapy. Researchers can add the above risk factors to the data collection process to create a complete clinical prediction model.
We found that congestive heart failure was associated with an increased risk of ICH. However, it should be noted that congestive heart failure has not been previously considered a risk factor for ICH. Patients with congestive heart failure are at high risk of stroke [S106]. A likely mechanism of stroke in these patients is cerebral embolism from a ventricular thrombus, which leads to ICH [S107]. In addition, the relationship between abnormal lipid metabolism and ICH remains controversial. Some studies [S108-109] reported no association with a higher risk of ICH, while others [S110] observed lower LDL-C levels to portend higher rates of ICH. This may be because low levels of lipids will adversely affect the integrity of small vessels in the brain, leading to blood extravasation due to the compromised integrity of the microvascular endothelial cells [S111-112].
We found that smoking may not be related to ICH. This is mainly because Wang et al. found that smoking benefits ICH [S86]. This contrasts with another article that considers smoking harmful to ICH [S77]. Wang et al. 's study about smoking has limitations. The smoking rate in the study population was low, and the distribution was uneven among the patients, which may result in smoking as a protective factor. Given the harmful effects of smoking on health, smoking cessation should still be strongly recommended to prevent stroke. First, long-term smoking can make the adrenal gland release adrenaline, with increases blood pressure and lead to ICH.
Additionally, nicotine in tobacco could directly destroy vascular endothelial function and arterial elasticity, promoting the destruction of the blood–brain barrier and increasing the possibility of ICH. The correlative study demonstrated a strong dose–response between the number of cigarettes smoked daily and ischemic stroke among young men [S113]. Likewise, there is evidence for a dose–response between cigarette smoking and the risk of stroke in middle-aged and older adults [S114].
Although this study did not find a correlation between UA and ICH, Song et al. reported a dose–response relationship between UA and ICH in a cohort of 1230 patients. Higher serum UA was independently related to a lower risk of ICH [S115]. This is an interesting discovery. Only a few studies have investigated the relationship between UA and ICH. It may be related to UA's ability to scavenging of free radicals, inhibit inflammatory cascade reactions, prevent mitochondrial damage, suppress lipid peroxidation, and reduce BBB permeability [S116-117].
In addition, no significant gender difference in ICH was found in this study. In some studies, male sex is a risk factor for developing ICH [S118], but the exact mechanism is unknown. However, other studies have reported that women have a higher risk of ICH [S119]. It may be caused by an inherent biological vulnerability of women for ICH. What’s more, weight and gender had similar results. Some studies suggest that higher body weight may be associated with ICH. Kim et al. found that lower body weight was associated with ICH and explained it as a "paradoxical effect of obesity" [S120]. Some scholars also found a potential explanation: when calculating the rt-PA dose, doctors overestimated their weight in the hyperacute phase, and patients' "overdose" led to ICH. I think we should focus on body mass index (BMI) rather than weight. BMI can avoid the diagnostic error caused by abnormal height (too high or too short) and objectively evaluate the treatment effect.
The mechanism of thrombolytic therapy-induced hemorrhage for cerebral infarction may differ from that of thrombolytic therapy-induced hemorrhage for other disorders. Compared with the mechanism of thrombolysis for bleeding in other diseases, the causes of bleeding in thrombolytic therapy-induced hemorrhage may also include destruction of the blood–brain barrier and reperfusion of injured brain tissue [18]. Therefore, because the bleeding mechanism is not identical, the risk of intracranial hemorrhage may change and is worth continuing to explore in future studies.
## Strengths
Our study followed rigorous methods, conducted extensive searches, duplicate and independent screening and data extraction, and assessed the certainty of evidence based on a structured framework. Also, we conducted sensitivity analyses to determine the stability of the results. The greatest advantage is the comprehensiveness of the study results, which may have some clinical significance in preventing the occurrence of thrombolysis-related ICH.
## Limitations and challenges
Since most of the studies included in this review were retrospective studies, classification and recall bias may lead to potential limitations. And this is not individual patient data meta-analysis and that any associations identified are univariate, with their inherent limitations. In addition, potential limitations of the included studies related to the inconsistency and variability across eligibility criteria in the original studies and variability in study design, study type, sample size, and definitions of the risk factors. Therefore, more rigorous and large-scale studies are needed to confirm our findings, and further analysis is necessary to provide a more reliable basis for clinical work.
## Conclusion
In this systematic review, we identified all reported risk factors for ICH associated with thrombolysis therapy. Some of these factors are not included in current ICH risk prediction models. Our findings will help inform experts in developing population-based guidelines and accurate, user-friendly RAMs to better guide individual patient prophylactic management.
## Supplementary Information
Additional file 1: Supplemental material 1. Search strategy
## References
1. Lansberg MG, O’Donnell MJ, Khatri P. **Antithrombotic and thrombolytic therapy for ischemic stroke: Antithrombotic Therapy and Prevention of Thrombosis, 9th ed: American College of Chest Physicians Evidence-Based Clinical Practice Guidelines**. *Chest* (2012.0) **141** e601S-e636. DOI: 10.1378/chest.11-2302
2. Lees KR, Bluhmki E, von Kummer R. **Time to treatment with intravenous alteplase and outcome in stroke: an updated pooled analysis of ECASS, ATLANTIS, NINDS, and EPITHET trials**. *Lancet* (2010.0) **375** 1695-1703. DOI: 10.1016/S0140-6736(10)60491-6
3. 3.National Institute of Neurological Disorders and Stroke rt-PA Stroke Study Group. Tissue plasminogen activator for acute ischemic stroke. N Engl J Med. 1995;333:1581–1587.
4. Derex L, Nighoghossian N. **Intracerebral haemorrhage after thrombolysis for acute ischemic stroke: an update**. *J Neurol Neurosurg Psychiatry* (2008.0) **79** 1093-1099. DOI: 10.1136/jnnp.2007.133371
5. 5.The National Institute of Neurological Disorders and Stroke t-PA Stroke Study Group. Intracerebral hemorrhage after intravenous t-PA therapy for ischemic stroke. Stroke 1997;28:2109–18.
6. 6.Hacke W, Kaste M, Fieschi C, et al. Randomised double-blind placebo-controlled trial of thrombolytic therapy with intravenous alteplase in acute ischemic stroke (ECASS II). Second European–Australasian Acute Stroke Study Investigators. Lancet 1998;352:1245–51.
7. 7.Hacke W, Kaste M, Bluhmki E, et al. Thrombolysis with alteplase 3 to 4.5 hours after acute ischemic stroke. N Engl J Med 2008;359:1317–29.
8. Darzi AJ, Karam SG, Charide R. **Prognostic factors for VTE and bleeding in hospitalized medical patients: a systematic review and meta-analysis**. *Blood* (2020.0) **135** 1788-1810. DOI: 10.1182/blood.2019003603
9. 9.Riley RD, Moons KG, Snell KI, Ensor J, Hooft L, Altman DG, et al. A guide to systematic review and meta-analysis of prognostic factor studies. BMJ. 2019;364:k4597.
10. 10.Wolff RWP, Whiting P, Mallett S, et al, eds. PROBAST: a “risk of bias” tool for prediction modelling studies [abstract]. Cochrane Colloquium Abstracts. 2015
11. 11.Tang SW, Zhang Y, Tao BL, Yang ZR, Sun F, Zhan SY. Risk of bias assessment: (7) Assessing Bias in Studies of Prognostic Factors [in Chinese]. Zhong hua Liu Xing Bing Xue Za Zhi. 2018;39(7):1003-1008.
12. 12.Hayden JA, van der Windt DA, Cartwright JL, Cˆoté P, Bombardier C. Assessing bias in studies of prognostic factors. Ann Intern Med. 2013;158(4):280–286.
13. 13.Reeves BC, Deeks JJ, Higgins JPT, Wells GA. Including non-randomized studies. In: Cochrane Handbook for Systematic Reviews of Interventions. 2008:391.
14. Iorio A, Spencer FA, Falavigna M. **Use of GRADE for assessment of evidence about prognosis: rating confidence in estimates of event rates in broad categories of patients**. *BMJ* (2015.0) **350** h870. DOI: 10.1136/bmj.h870
15. 15.Ahn S, Fessler JA. Standard errors of mean, variance, and standard deviation estimators. Ann Arbor: Electrical Engineering and Computer Science Department, The University of Michigan; 2003:1–2. https://web.eecs.umich.edu/~fessler/papers/files/tr/stderr.pdf.
16. 16.Grant RL. Converting an odds ratio to a range of plausible relative risks for better communication of research findings [published correction appears in BMJ. 2014;348:g2124]. BMJ. 2014;348(1):f7450.
17. Wang Z. **Converting odds ratio to relative risk in cohort studies with partial data information**. *J Stat Softw* (2013.0) **55** 1-11. DOI: 10.18637/jss.v055.i05
18. Yaghi S, Willey JZ, Cucchiara B. **Treatment and Outcome of Hemorrhagic Transformation After Intravenous Alteplase in Acute Ischemic Stroke: A Scientific Statement for Healthcare Professionals From the American Heart Association/American Stroke Association**. *Stroke* (2017.0) **48** e343-e361. DOI: 10.1161/STR.0000000000000152
|
---
title: 'Effects of dietary diversity on frailty in Chinese older adults: a 3-year
cohort study'
authors:
- Ying Duan
- Qi Qi
- Yan Cui
- Ling Yang
- Min Zhang
- Huaqing Liu
journal: BMC Geriatrics
year: 2023
pmcid: PMC10012609
doi: 10.1186/s12877-023-03875-5
license: CC BY 4.0
---
# Effects of dietary diversity on frailty in Chinese older adults: a 3-year cohort study
## Abstract
### Background
Frailty has emerged as a global health burden with increased population aging. A diverse diet is essential for an adequate and balanced supply of nutrients. However, limited evidence supports the relationship between dietary diversity and frailty. We therefore assessed the associations of dietary diversity with the risk of frailty.
### Methods
We used the Chinese Longitudinal Healthy Longevity Survey to analyze a prospective cohort of Chinese older adults. A total of 1948 non-frail older adults were included in the final sample. Participants were categorized into groups with high or low dietary diversity scores (DDSs) using a food frequency questionnaire. A Generalized Estimating Equation were used to estimate risk ratios (RRs) and $95\%$ confidence intervals (CIs) for determining frailty incidence.
### Results
Among 1,948 participants, 381 had frailty with the prevalence of $19.56\%$ during the 3-year follow-up period. Compared with the low DDS group, the high DDS group exhibited a lower risk of frailty (RR, 0.72; $95\%$ CI: 0.57–0.91). Compared with those with a consistently low DDS, the RR of participants with a consistently high DDS for frailty was 0.56 ($95\%$ CI: 0.42–0.74). Moreover, meat, beans, fish, nuts, fresh fruits, and fresh vegetables were inversely associated with frailty. In stratified analysis, a consistently high DDS, compared with a consistently low DDS, reduced the risk of frailty for people aged 65-79 years and those living in town and rural areas.
### Conclusion
This study found a prospective association between dietary diversity and frailty among Chinese older adults. These findings stressed that it is important to improve dietary diversity for older adults to promote healthy ageing, particularly for young older adults and in town and rural areas.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12877-023-03875-5.
## Introduction
The number of older adults aged more than 65 years has increased from 461 million in 2004 to an estimated 2 billion in 2050 [1]. Frailty has emerged as a global health burden, and the incidence of frailty is likely to increase under population aging [2]. The prevalence of frailty in the elderly individuals ranges from 12 to $24\%$ [3]. Frailty is associated with a range of adverse outcomes, including morbidity, mortality, and increased health care costs [4, 5], and it has major implications for clinical practice and public health.
Frailty is defined as increased vulnerability to stressors across multiple bodily systems [6], including cognitive, psychosocial, and physical components [7]. Frailty is dynamic process that deteriorates or improves over time [8]. Strategies for preventing and delaying the progression of frailty are crucial [9].
Nutrition, a modifiable factor for frailty [6], plays a critical role in causing, mediating, and reversing frailty in older adults [10]. Various foods and nutrients have been reported to assist in preventing frailty. Cross-sectional studies in China [11], the United States [12], and the United Kingdom [13] have revealed that fruit and vegetable consumption is associated with a reduced risk of frailty. A prospective study [14] of elderly Japanese reported that higher baseline dairy and milk consumption was associated with a lower risk of frailty. Another prospective cohort study [15] in Spain also revealed that an increased intake of yogurt milk and low-fat was associated with a lower incidence of frailty. A meta-analysis of 10 studies reported a lower frailty prevalence among older adults with high protein intake than among older adults with low protein intake [16]. A systematic review of longitudinal data on vitamin D and frailty indicated an association between lower vitamin D intake and a higher risk of frailty [17].
Nutrition is a key factor for the prevention and treatment of frailty. However, a single nutrient or food cannot reflect the nutritional status of an individual in real life. Few studies have explored the effects of individual nutrition on frailty from a holistic perspective. Dietary diversity is defined as the number of different food groups or foods consumed in a given period, and it ensures a rich provision of macronutrients and micronutrients [18]. A diverse diet is essential for an adequate and balanced supply of nutrients. Dietary diversity score (DDS), as an indicator of dietary diversity, can be applied to all age groups [19]. Two cross-sectional studies [20, 21] in Japan have reported that DDS may be associated with frailty in older adults. The present 3-year cohort study explored the association between DDS and frailty among Chinese older adults.
## Study population
The Chinese Longitudinal Healthy Longevity Survey (CLHLS) is a nationwide, prospective cohort study of community-dwelling older adults in China. The survey has been conducted in 23 counties and cities randomly selected from 31 provinces, and the population of these areas covers $85\%$ of China's population. The survey was carried out in 1998, and follow-up surveys are conducted every 2 to 3 years. The trained staff interviewed the elders face to face and systematically collected their information. [ 22].
Our study used data for the period from 2011 to 2014. 9765 individuals participated in baseline interviews between 2011 and 2012. We excluded participants aged less than 65 years ($$n = 86$$) in 2011. Furthermore, from this sample, we excluded participants who presented with frailty in 2011 ($$n = 1570$$), those with missing information relating to frailty ($$n = 4522$$) and those with missing values related to DDS ($$n = 16$$). We also excluded participants who were lost or dead. Finally, we analyzed the data of 1948 individuals from 2011 to 2014 to determine the relationship between DDS and frailty. Figure 1 depicts the flowchart of the patient selection process in this study. The missing participants were more likely to be female, aged 80 years or above, financially dependent, of informal education, of other marital status, of underweight, and to live in town and rural areas and to not smoke, not drink, not exercise (Table S1).Fig. 1Flowchart of the participant selection process
## Assessment of frailty
The frailty index (FI) counts deficits in health to assessment of frailty [23]. According to a study [24] measuring frailty, the FI contains items related to 44 health deficits, including activities of daily living (basic and instrumental), chronic diseases, and psychological function. The FI is calculated by adding all deficits and dividing the sum by the total number of deficits. Detailed information on the calculation procedure is presented in the Table S2. In the present study, the FI reflected cumulative health deficits, and it is comparable to the indices used in other CLHLS-based studies [25–27] and those used in studies conducted in the United States [28], Canada [29], and Hong Kong [30]. The FI is a continuous variable and ranges from 0 to 1; a higher value indicates a higher degree of frailty. According to the FI, participants were divided into two groups: non-frailty group (FI ≤ 0.21) and frailty group (FI > 0.21) [31].
## Assessment of dietary diversity
The participants completed a food frequency questionnaire [32] on 11 major food groups or items: meat, fish, eggs, beans, mushrooms or algae, tea, garlic, milk products, nuts, fresh vegetables, and fresh fruits; the questionnaire was used to determine their DDS during a face-to-face interview. Because almost all Chinese people consume cereals and oil daily, we did not include these two food groups in the DDS questionnaire [33]. The DDS questionnaire we used was primarily composed of items on healthy food. It was created to assess the adequacy of food consumption and the healthiness of diets, and its scientific validity has been demonstrated [34].
DDS was calculated according to the frequency of intake of the 11 food groups. The scoring criteria and intake frequencies are detailed in the Table S3. The total DDS was calculated as the sum of the scores of the 11 food groups, with the highest and lowest scores being 11 and 0, respectively. The higher was the score, the greater was the dietary diversity. We divided the participants into two groups according to the median DDS, namely the low and high DDS groups.
We also classified changes in dietary diversity (CDD) from 2011 to 2014 into the following four categories: declining dietary diversity, improving dietary diversity, consistently low dietary diversity, and consistently high dietary diversity (Table S4).
## Assessment of covariates
According to previous research [35], frailty is influenced by older people’s individual heterogeneity, socioeconomic level, and health status. Therefore, to account for crucial differences, we evaluated various baseline characteristics. Covariates included age group (65–79 or ≥ 80 years), sex (male or female), body mass index (BMI; underweight, normal, overweight, or obese), residential location (urban or town and rural), and marital status (currently married and living with spouse or other). Education status was classified into formal education [≥ 1 year(s) of education] and informal education (< 1 year of education). Drinking, smoking, and exercise status was defined according two questions (“drink/smoke/exercise or not at present?” and “drink/smoke/exercise or not in the past?”). It was defined as no if participants answer no to both questions; otherwise, was defined as yes. The participants were regarded as having a history of chronic disease if they self-reported hypertension, heart disease, bronchitis, asthma, emphysema, pneumonia, or diabetes. We distinguished financial support into financial independence and dependence. We regarded financial independence as receiving a work or retirement wage, and financial dependence as financially relying on other family members.
## Statistical analysis
Independent chi-square tests were used to examine the initial basic characteristics of the groups (sex, age, residential location, education status, BMI, drinking status, smoking status, exercise status, marital status, financial support, and history of chronic disease). A Generalized Estimating Equation (GEE) were used to determine the relationship between dietary diversity and frailty incidence. In addition, we conducted a stratified analysis based on age and residential location.
For all outcomes, we constructed models without any adjusted covariates (Model 1), models adjusted only for sex and age (Model 2), and models further adjusted for residential location, education status, BMI, drinking status, smoking status, exercise status, marital status, financial support, and history of chronic disease (Model 3). The sampling weight variables in CLHLS are calculated based on the age-sex-residence-specific distribution of the population. Our analysis results were weighted to ensure its representativeness.
Given that large samples losses occur due to incomplete frailty data, we conducted a sensitivity analysis. If the missing value in the 44 health deficits of participants is less than or equal to 5, these participants will not be excluded. Given pre-frailty older adults may be change dietary behaviors, we conducted another sensitivity analysis and examined the relationship between frailty and dietary diversity after excluding subjects with pre-frailty at baseline.
Statistical analysis was conducted using R studio, version 4.1.2 (R Foundation for Statistical Computing). Statistical significance was defined as a two-sided P value threshold of 0.05.
## Baseline characteristics
Table 1 provides the baseline characteristics of the study participants. The sample was composed of 1948 participants, with 1040 men ($53.4\%$) and 908 women ($46.6\%$). Among all participants, $43.0\%$ were 80 years and older, $79.6\%$ lived in town and rural areas, $48.2\%$ were currently married and living with their spouse, $44.3\%$ had informal education, $38.9\%$ were financially independent, and $57.3\%$ had no chronic illnesses. Table 1Baseline characteristics of participants without frailtyCharacteristics N (%) Low DDS (weighted %)High DDS (weighted %)X2 Total 194845.754.3 Age(years) 21.76* 65–791111(57.0)44.355.7 ≥ 80837(43.0)55.744.3 Sex 32.78* Male1040(53.4)41.358.7 Female908(46.6)50.549.5 Residential location 61.64* Town and rural1550(79.6)48.351.7 Urban398(20.4)30.369.7 Education 132.62* Informal education863(44.3)58.441.6 Formal education1083(55.7)38.961.1 Financial support 217.51* Financial dependence1190(61.1)55.944.1 Financial independence757(38.9)31.968.1 *Marital status* 23.67* Currently married and living with spouse937(48.2)55.644.4 Other1008(51.8)40.659.4 *Smoking status* 0.36 No1106(56.9)44.755.3 Yes837(43.1)45.754.3 *Drinking status* 10.39* No1185(61.2)47.952.1 Yes750(38.8)42.757.3 *Exercise status* 71.20* No720(37.0)53.746.3 Yes1222(63.0)39.560.5 Body mass index (kg/m2) 197.00* Underweight (< 18.5)330(17.1)70.829.2 Normal (18.5–23.99)1075(55.6)45.854.2 Overweight (24–27.99)399(20.6)32.068.0 Obese (≥ 28)131(6.8)43.256.8 Chronic disease 7.60* No1116(57.3)43.956.1 Yes832(42.7)48.451.6 DDS Dietary diversity score*$P \leq 0.05$ The mean score of the DDS was 5.7 with the standard deviation of 1.8. From 2011 to 2014, 608 ($21.2\%$) participants maintained low DDS, 279 ($14.3\%$) participants changed from high DDS to low DDS, 328 ($16.8\%$) participants changed from low DDS to high DDS, and 733 ($37.6\%$) participants maintained high DDS.
The participants who are aged 65–79, female, live in urban areas, were overweight, have formal education, have another marital status, be financially independent, be physically active drink, do not smoke and have no history of chronic diseases were more likely to have high DDS.
## Association of DDS with frailty prevalence
Among 1,948 participants, 381 had frailty with the prevalence of $19.56\%$ during the 3-year follow-up period. These cases involved 208 participants with a low DDS and 173 participants with a high DDS. Overall, the crude rate of frailty events was higher in the low DDS group than in the high DDS group (Table 2). In the unadjusted analysis, the risk ratio (RR) of the participants with a high DDS for frailty was 0.61 [$95\%$ confidence interval (CI): 0.50–0.76] compared with that of the participants with a low DDS. Following adjustment for age, sex, residential location, education status, BMI, drinking status, smoking status, exercise status, marital status, financial support, and history of chronic disease (Model 3), the inverse association was still significant (RR, 0.72; $95\%$ CI: 0.57–0.91).Table 2Association between DDS and frailty Characteristics Model 1 RR ($95\%$ CI)Model 2 RR ($95\%$ CI)Model3 RR ($95\%$ CI) DDS as continuous variable 0.83(0.78,0.88)*0.86(0.81,0.92)*0.88(0.82,0.94)* DDS as categorical variable (ref. = Low DDS) High DDS0.61(0.50,0.76)*0.69(0.55,0.85)*0.72(0.57,0.91)* Model 1: no adjustment; Model 2: adjusted for age and sex; Model 3: adjusted for age, sex, residential location, education status, body mass index, drinking status, smoking status, exercise status, marital status, financial support, and history of chronic disease DDS Dietary diversity score, CI Confidence interval, RR Risk ratio*$P \leq 0.05$ When DDS was adopted as a continuous variable, this association did not change (RR, 0. 83; $95\%$ CI: 0.78–0.88 in Model 1; RR, 0.88; $95\%$ CI: 0.82–0.94 in Model 3).
## Association of CDD with frailty prevalence
The association between CDD and frailty is presented in Table 3. Compared with those with a consistently low DDS, the RR of the participants with a consistently high DDS for frailty was 0.46 ($95\%$ CI: 0.35–0.59) in the crude model. Following adjustment for all the covariates, the inverse association between a consistently high DDS and frailty remained significant ($P \leq 0.05$).Table 3Association between CDD and frailtyRR ($95\%$CI)Consistently Low Dietary DiversityDeclining Dietary DiversityImproving Dietary DiversityConsistently High Dietary DiversityModel 11 (reference)0.80(0.58,1.09)0.68(0.49,0.92)*0.46(0.35,0.59)*Model 21 (reference)0.86(0.62,1.18)0.73(0.53,1.01)0.53(0.41,0.70)*Model 31 (reference)0.94(0.67,1.33)0.79(0.56,1.11)0.56(0.42,0.74)* Model 1: no adjustment; Model 2: adjusted for age and sex; Model 3: adjusted for age, sex, residential location, education status, body mass index, drinking status, smoking status, exercise status, marital status, financial support, and history of chronic disease CDD Changes in dietary diversity, CI Confidence interval, RR Risk ratio*$P \leq 0.05$ In stratified analysis (Table 4), a consistently high DDS, compared with a consistently low DDS, reduced the risk of frailty for people aged 65–79 years (RR, 0.46; $95\%$ CI: 0.33–0.64) and those living in town and rural areas (RR, 0.46; $95\%$ CI: 0.33–0.64) after adjusting for all covariates, but not for people aged 80 years and older (RR, 1.00; $95\%$ CI: 0.56–1.80) and those living in urban areas (RR, 1.38; $95\%$ CI: 0.48–3.93).Table 4Associations between CDD and frailty, stratified by age and residential locationCharacteristicsFinal model RR ($95\%$ CI)Consistently Low Dietary DiversityDeclining Dietary DiversityImproving Dietary DiversityConsistently High Dietary DiversityStratified by agea 65–79Reference0.84(0.57,1.25)0.80(0.55,1.17)0.46(0.33,0.64)* ≥ 80Reference1.50(0.73,3.06)0.68(0.33,1.44)1.00(0.56,1.80)Stratified by residential locationb Town and ruralReference0.94(0.65,1.35)0.73(0.50,1.05)0.46(0.33,0.64)* UrbanReference1.62(0.45,5.92)2.34(0.71,7.70)1.38(0.48,3.93) CDD Changes in dietary diversity, CI Confidence interval, RR Risk ratioaAdjusted for sex, residential location, education status, body mass index, drinking status, smoking status, exercise status, marital status, financial support, and history of chronic diseasebAdjusted for age, sex, education status, body mass index, drinking status, smoking status, exercise status, marital status, financial support, and history of chronic disease*$P \leq 0.05$ In the two sensitivity analyses, this association between dietary diversity and frailty remains unchanged (Table S5 and Table S6).
## Association of single specific foods with frailty prevalence
We adopted the occurrence of frailty as the dependent variable and the intake of the 11 major food groups in 2011 as the independent variable in GEE. Meat (RR, 0.47; $95\%$ CI: 0.32–0.68), fish (RR, 0.56; $95\%$ CI: 0.39–0.80), beans (RR, 0.68; $95\%$ CI: 0.47–0.97), nuts (RR, 0.59; $95\%$ CI: 0.40–0.87), fresh fruits (RR, 0.69; $95\%$ CI: 0.49–0.97) and fresh vegetables (RR, 0.25; $95\%$ CI: 0.14–0.45) was inversely associated with frailty following adjustment for all covariates. Most of the associations between other single food groups and frailty were in the expected direction (Fig. 2).Fig. 2RRs and $95\%$ CIs indicating the associations between single food groups and frailty. CI: confidence interval, RR: risk ratio. Adjusted for age, sex, residential location, education status, body mass index, drinking status, smoking status, exercise status, marital status, financial support, and history of chronic disease
## Discussion
From a holistic nutritional perspective, this study evaluated a prospective association of dietary diversity with frailty in a Chinese older adult population. Our study revealed that a high DDS reduced the incidence of frailty. In addition, the participants with a consistently high DDS had a lower risk of frailty than those with a consistently low DDS during the 3-year follow-up period. Whether the participants with pre-frailty at baseline were excluded or the participants with some missing values (≤ 5) of the health deficit were remained, dietary diversity showed a protective effect on frailty of older adults, which to some extent showed the stability of our results. This study further verified the relationship between dietary diversity and frailty and provided new evidence that can be applied for preventing or delaying frailty development in older adults.
A high DDS is associated with an adequate intake of nutrients and a favorable nutritional status. A low DDS is related to the risk of undernutrition [36], which is characterized by inadequate nutrient intake and reduced energy reserves. Obesity may be one of the risk factors for frailty [37]. However, this association remains contradictory, and another research showed that obesity is related to reducing the risk of frailty in multivariate analysis [38]. Interestingly, in our study, overweight subjects are more likely to have high DDS. These findings indicated that the relationship between obesity, frailty and DDS should be further investigated in future study. Energy and nutrient deficiencies may affect mitochondrial function and induce muscle-related symptoms, including frailty [39]. Malnourished older adults have a high prevalence of frailty [40]. Additionally, the participants with a high DDS had a high intake of protein, vitamins, and antioxidant nutrients [41, 42], all of which have been reported to the prevention of frailty. Loss of muscle mass and strength is regarded as a key pathology leading to frailty [43]. Adequate protein intake helps to maintain muscle function in older adults [44]. A high DDS can reduce inflammation and oxidative stress [45], both of which accelerate the loss of muscle and bone mass and the deterioration of central nervous system function [46–48], leading to frailty. The effects of many nutrients depend on the presence of other nutrients in different food groups, and only through a high dietary diversity can nutritionally balance and disease prevention be achieved [49]. For example, food protein sources are crucial, and vitamins and minerals in fruits and vegetables are also essential for the synthesis of muscle protein [20]. If a diet lacks diversity, nutrients that contribute to frailty prevention are ineffective. Nutrient interactions (i.e., their balance) are more critical in health and aging than nutrients acting alone [50]. In addition, eating a variety of foods throughout the day requires health awareness in the performance of activities such as shopping, cooking, and meal planning. These intentional instrumental activities may effectively assist individuals in maintaining functional ability [51] and physical performance [52]. Moreover, studies have indicated that a diverse diet can promote a healthier gut microbiome [53], which plays a role in the anabolic resistance of skeletal muscle to dietary proteins [54] and may play a role in the prevention of frailty.
To understand the relationship between dietary diversity and frailty, the effects of long-term dietary behaviors on frailty were explored in this study. Our study revealed that a consistently high DDS can reduce the risk of frailty among Chinese older adults following adjustment for confounding factors. Older adults may experience a decline in their chewing ability as a result of aging [55], and chewing ability is associated with dietary diversity [56]. Therefore, the dietary diversity of older adults may change in the future. Perhaps only the long-term rather than the short-term maintenance of high dietary diversity has a beneficial effect on frailty. Consistent with our findings, our previous study [57] demonstrated the protective effect of long-term tea consumption on frailty in older adults.
The town and rural population differ from urban dwellers in eating habits and conditions [58]. In urban areas, residents with low dietary diversity may indirectly indicate a reduced ability to go out, leading to fewer opportunities to replenish the pantry; however, in town and rural areas, dietary diversity may reflect different dietary choices. Urban areas have more complete food supply systems and better infrastructure, and urban residents have higher food diversity and availability than rural residents [59, 60]. In our study, consistently high dietary diversity was associated with a lower risk of frailty for town and rural residents, but not for older adults living in urban areas. With the increase in age, the prevalence of frailty increases [61] and the ability to chew and digest decreases. Decreased chewing and digestion due to aging will lead to lower DDS [62]. This may explain the protective effect of consistently high DDS against frailty in older adults aged 65–79, but not in those aged 80 or older.
We further explored the associations between specific food groups and frailty, and the results revealed the positive effect of meat, beans, fish, nuts, fresh fruits, and fresh vegetables on the prevention of frailty. Fruits and vegetables are rich sources of antioxidants such as carotene and vitamin C [63]. Meat and fish are rich in protein, which increases muscle synthesis [64]. Nuts and legumes are good sources of vegetable protein, which prevents muscle mass loss [65]. These are all related to preventing frailty [43, 48, 66]. Milk products are rich in calcium, which may be related to the prevention of osteoporosis and then to reduce frailty. Some research results also show that calcium intake is related to frailty [37, 67]. In our study, milk products have slight trend on the reduction of the risk of frailty, although this reduced effect did not reach statistical significance.
These findings stressed that public health worker should take actions or interventions on diet to reduce the incidence of frailty in older adults. Dietary diversity should be recommended. It is important to strengthen healthy dietary behaviors education for older adults and caregivers to increase the awareness of dietary diversity. Moreover, the community can strengthen diverse foods supplies for older adults to meet their needs of dietary diversity, particularly for young older adults and in town and rural areas.
To the best of our knowledge, this study is the first to examine the association between DDS and frailty in Chinese older adults using nationally representative cohort data. An advantage of the study is its exploration of the role of ongoing dietary diversity. Our study has some limitations. First, self-reported information collected using the food frequency questionnaire is prone to recall bias. In addition, this questionnaire only collected information on the frequency of food intake and not the specific amount of food intake. Finally, the included participants were more likely to be male, aged 65—79 years, currently married and living with spouse, financially dependent, of formal education, of normal BMI, and to live in town and rural areas and to not smoke, not drink, exercise. Differences in baseline characteristics between the lost and included samples showed that our study may have selection bias. And these may affect the robustness of our results to some extent. For example, the prevalence of frailty increases with the increase of age [61], and dietary patterns may also change with aging. Studies have also shown that frailty occurs more often in women than in men [68].
## Conclusion
This study found a prospective association between dietary diversity and frailty among Chinese older adults.
## Supplementary Information
Additional file 1: Table S1. Baseline characteristics of lost sample and final sample. Table S2. Scoring criteria for frailty. Table S3. Scoringcriteria for dietary diversity. Table S4. Changes in dietary diversity. Table S5. Association between dietary diversity and frailty after retaining participants with some health defect absence value (≤5). Table S6. Association between dietary diversity and frailty after excluding participants with pre-frailty in 2011.
## References
1. 1.Brivio P, Paladini MS, Racagni G, Riva MA, Calabrese F, Molteni R. From Healthy Aging to Frailty: In Search of the Underlying Mechanisms. Curr Med Chem. 2019;26(20):3685–3701.: 10.2174/0929867326666190717152739.
2. Blodgett JM, Theou O, Howlett SE, Rockwood K. **A frailty index from common clinical and laboratory tests predicts increased risk of death across the life course**. *Geroscience* (2017.0) **39** 447-455. DOI: 10.1007/s11357-017-9993-7
3. 3.O’Caoimh R, Sezgin D, O’Donovan MR, Molloy DW, Clegg A, Rockwood K, et al. Prevalence of frailty in 62 countries across the world: a systematic review and meta-analysis of population-level studies. Age Ageing. 2021;50(1):96–104. 10.1093/ageing/afaa219.
4. Muscedere J, Waters B, Varambally A, Bagshaw SM, Boyd JG, Maslove D. **The impact of frailty on intensive care unit outcomes: a systematic review and meta-analysis**. *Intensive Care Med* (2017.0) **43** 1105-1122. DOI: 10.1007/s00134-017-4867-0
5. Walters K, Frost R, Kharicha K, Avgerinou C, Gardner B, Ricciardi F. **Home-based health promotion for older people with mild frailty: the HomeHealth intervention development and feasibility RCT**. *Health Technol Assess* (2017.0) **21** 1-128. DOI: 10.3310/hta21730
6. Clegg A, Young J, Iliffe S, Rikkert MO, Rockwood K. **Frailty in elderly people**. *Lancet* (2013.0) **381** 752-762. DOI: 10.1016/S0140-6736(12)62167-9
7. Pickard S, Cluley V, Danely J, Laceulle H, Leon-Salas J, Vanhoutte B. **New horizons in frailty: the contingent, the existential and the clinical**. *Age Ageing* (2019.0) **48** 466-471. DOI: 10.1093/ageing/afz032
8. Kojima G, Taniguchi Y, Iliffe S, Jivraj S, Walters K. **Transitions between frailty states among community-dwelling older people: A systematic review and meta-analysis**. *Ageing Res Rev* (2019.0) **50** 81-88. DOI: 10.1016/j.arr.2019.01.010
9. Puts MTE, Toubasi S, Andrew MK, Ashe MC, Ploeg J, Atkinson E. **Interventions to prevent or reduce the level of frailty in community-dwelling older adults: a scoping review of the literature and international policies**. *Age Ageing* (2017.0) **46** 383-392. DOI: 10.1093/ageing/afw247
10. Hoogendijk EO, Afilalo J, Ensrud KE, Kowal P, Onder G, Fried LP. **Frailty: implications for clinical practice and public health**. *Lancet* (2019.0) **394** 1365-1375. DOI: 10.1016/S0140-6736(19)31786-6
11. Gao J, Jia Y, Dai J, Fu H, Wang Y, Yan H. **Association of Fruit and Vegetable Intake and Frailty among Chinese Elders: A Cross-Sectional Study in Three Cities**. *J Nutr Health Aging* (2019.0) **23** 890-895. DOI: 10.1007/s12603-019-1236-4
12. Fung TT, Struijk EA, Rodriguez-Artalejo F, Willett WC, Lopez-Garcia E. **Fruit and vegetable intake and risk of frailty in women 60 years old or older**. *Am J Clin Nutr* (2020.0) **112** 1540-1546. DOI: 10.1093/ajcn/nqaa256
13. Kojima G, Iliffe S, Jivraj S, Walters K. **Fruit and Vegetable Consumption and Incident Prefrailty and Frailty in Community-Dwelling Older People: The English Longitudinal Study of Ageing**. *Nutrients* (2020.0) **12** 3882. DOI: 10.3390/nu12123882
14. Otsuka R, Tange C, Tomida M, Nishita Y, Kato Y, Yuki A. **Dietary factors associated with the development of physical frailty in community-dwelling older adults**. *J Nutr Health Aging* (2019.0) **23** 89-95. DOI: 10.1007/s12603-018-1124-3
15. Lana A, Rodriguez-Artalejo F, Lopez-Garcia E. **Dairy consumption and risk of frailty in older adults: A prospective cohort study**. *J Am Geriatr Soc* (2015.0) **63** 1852-1860. DOI: 10.1111/jgs.13626
16. Coelho-Júnior HJ, Rodrigues B, Uchida M, Marzetti E. **Low Protein Intake Is Associated with Frailty in Older Adults: A Systematic Review and Meta-Analysis of Observational Studies**. *Nutrients* (2018.0) **10** 1334. DOI: 10.3390/nu10091334
17. Zhou J, Huang P, Liu P, Hao Q, Chen S, Dong B. **Association of vitamin D deficiency and frailty: A systematic review and meta-analysis**. *Maturitas* (2016.0) **94** 70-76. DOI: 10.1016/j.maturitas.2016.09.003
18. 18.Food and Agriculture Organization (FAO) Guidelines for Measuring Household and Individual Dietary Diversity. Food and Agriculture Organization of the United Nations; Rome, Italy: 2011. Available online: http://www.fao.org/3/a-i1983e.pdf. (Accessed on 24 Jan 2023).
19. Zhang J, Zhao A. **Dietary Diversity and Healthy Aging: A Prospective Study**. *Nutrients* (2021.0) **13** 1787. DOI: 10.3390/nu13061787
20. Motokawa K, Watanabe Y, Edahiro A, Shirobe M, Murakami M, Kera T. **Frailty Severity and Dietary Variety in Japanese Older Persons: A Cross-Sectional Study**. *J Nutr Health Aging* (2018.0) **22** 451-456. DOI: 10.1007/s12603-018-1000-1
21. Hayakawa M, Motokawa K, Mikami Y, Yamamoto K, Shirobe M, Edahiro A. **Low Dietary Variety and Diabetes Mellitus Are Associated with Frailty among Community-Dwelling Older Japanese Adults: A Cross-Sectional Study**. *Nutrients* (2021.0) **13** 641. DOI: 10.3390/nu13020641
22. Zeng Y. **Towards Deeper Research and Better Policy for Healthy Aging –Using the Unique Data of Chinese Longitudinal Healthy Longevity Survey**. *China Economic J* (2012.0) **5** 131-149. DOI: 10.1080/17538963.2013.764677
23. Searle SD, Mitnitski A, Gahbauer EA, Gill TM, Rockwood K. **A standard procedure for creating a frailty index**. *BMC Geriatr* (2008.0) **8** 24. DOI: 10.1186/1471-2318-8-24
24. Liu ZY, Wei YZ, Wei LQ, Jiang XY, Wang XF, Shi Y. **Frailty transitions and types of death in Chinese older adults: a population-based cohort study**. *Clin Interv Aging* (2018.0) **13** 947-956. DOI: 10.2147/CIA.S157089
25. Dupre ME, Gu D, Warner DF, Zeng Y. **Frailty and type of death among older adults in China: prospective cohort study**. *BMJ* (2009.0) **338** b1175. DOI: 10.1136/bmj.b1175
26. Gu D, Dupre ME, Sautter J, Zhu H, Liu Y, Yi Z. **Frailty and mortality among Chinese at advanced ages**. *J Gerontol B Psychol Sci Soc Sci* (2009.0) **64** 279-289. DOI: 10.1093/geronb/gbn009
27. Bennett S, Song X, Mitnitski A, Rockwood K. **A limit to frailty in very old, community-dwelling people: a secondary analysis of the Chinese longitudinal health and longevity study**. *Age Ageing* (2013.0) **42** 372-377. DOI: 10.1093/ageing/afs180
28. Kulminski A, Yashin A, Arbeev K, Akushevich I, Ukraintseva S, Land K. **Cumulative index of health disorders as an indicator of aging-associated processes in the elderly: results from analyses of the National Long Term Care Survey**. *Mech Ageing Dev* (2007.0) **128** 250-258. DOI: 10.1016/j.mad.2006.12.004
29. Mitnitski A, Song X, Skoog I, Broe GA, Cox JL, Grunfeld E. **Relative fitness and frailty of elderly men and women in developed countries and their relationship with mortality**. *J Am Geriatr Soc* (2005.0) **53** 2184-2189. DOI: 10.1111/j.1532-5415.2005.00506.x
30. Goggins WB, Woo J, Sham A, Ho SC. **Frailty index as a measure of biological age in a Chinese population**. *J Gerontol A Biol Sci Med Sci* (2005.0) **60** 1046-1051. DOI: 10.1093/gerona/60.8.1046
31. Hoover M, Rotermann M, Sanmartin C, Bernier J. **Validation of an index to estimate the prevalence of frailty among community-dwelling seniors**. *Health Rep* (2013.0) **24** 10-17. PMID: 24258362
32. Qiu L, Sautter J, Gu D. **Associations between frequency of tea consumption and health and mortality: evidence from old Chinese**. *Br J Nutr* (2012.0) **108** 1686-1697. DOI: 10.1017/S0007114511007173
33. 33.Yin Z, Fei Z, Qiu C, Brasher MS, Kraus VB, Zhao W, et al. Dietary diversity and cognitive function among elderly people: a population-based study. J Nutr Health Aging. 2017;21(10):1089e94.: 10.1007/s12603-017-0912-5.
34. Aihemaitijiang S, Zhang L, Ye C, Halimulati M, Huang X, Wang R, Zhang Z. **Long-Term High Dietary Diversity Maintains Good Physical Function in Chinese Elderly: A Cohort Study Based on CLHLS from 2011 to 2018**. *Nutrients* (2022.0) **14** 1730. DOI: 10.3390/nu14091730
35. Sirven N, Dumontet M, Rapp T. **The dynamics of frailty and change in socio-economic conditions: evidence for the 65+ in Europe**. *Eur J Public Health* (2020.0) **30** 715-719. DOI: 10.1093/eurpub/ckaa068
36. Cano-Ibáñez N, Gea A, Martínez-González MA, Salas-Salvadó J, Corella D, Zomeño MD. **Dietary Diversity and Nutritional Adequacy among an Older Spanish Population with Metabolic Syndrome in the PREDIMED-Plus Study: A Cross-Sectional Analysis**. *Nutrients* (2019.0) **11** 958. DOI: 10.3390/nu11050958
37. Güngör Başaran AY, Akal YE. **Nutrition Status, Muscle Mass, and Frailty in Older People: A Cross-Sectional Study Conducted in Cyprus**. *J Am Nutr Assoc* (2022.0) **41** 318-324. DOI: 10.1080/07315724.2021.1884142
38. García-Esquinas E, Graciani A, Guallar-Castillón P, López-García E, Rodríguez-Mañas L, Rodríguez-Artalejo F. **Diabetes and risk of frailty and its potential mechanisms: A prospective cohort study of older adults**. *J Am Med Dir Assoc* (2015.0) **16** 748-754. DOI: 10.1016/j.jamda.2015.04.008
39. Ferrucci L, Bandinelli S, Benvenuti E, Di Iorio A, Macchi C, Harris TB. **Subsystems contributing to the decline in ability to walk: bridging the gap between epidemiology and geriatric practice in the InCHIANTI study**. *J Am Geriatr Soc* (2000.0) **48** 1618-1625. DOI: 10.1111/j.1532-5415.2000.tb03873.x
40. 40.Verlaan S, Ligthart-Melis GC, Wijers SLJ, Cederholm T, Maier AB, de van der Schueren MAE. High Prevalence of Physical Frailty Among Community-Dwelling Malnourished Older Adults-A Systematic Review and Meta-Analysis. J Am Med Dir Assoc. 2017;18(5):374–382.:10.1016/j.jamda.2016.12.074.
41. Meng L, Wang Y, Li T, Loo-Bouwman CAV, Zhang Y, Man-Yau SI. **Dietary Diversity and Food Variety in Chinese Children Aged 3-17 Years: Are They Negatively Associated with Dietary Micronutrient Inadequacy?**. *Nutrients* (2018.0) **10** 1674. DOI: 10.3390/nu10111674
42. Azadbakht L, Mirmiran P, Azizi F. **Variety scores of food groups contribute to the specific nutrient adequacy in Tehranian men**. *Eur J Clin Nutr* (2005.0) **59** 1233-1240. DOI: 10.1038/sj.ejcn.1602234
43. Cruz-Jentoft AJ, Michel JP. **Sarcopenia: A useful paradigm for physical frailty**. *Eur Geriatr Med* (2013.0) **4** 102-105. DOI: 10.1016/j.eurger.2013.02.009
44. Houston DK, Nicklas BJ, Ding J, Harris TB, Tylavsky FA, Newman AB. **Dietary protein intake is associated with lean mass change in older, community-dwelling adults: The Health, Aging, and Body Composition (Health ABC) Study**. *Am J Clin Nutr* (2008.0) **87** 150-155. DOI: 10.1093/ajcn/87.1.150
45. Welch AA. **Nutritional influences on age-related skeletal muscle loss**. *Proc Nutr Soc* (2014.0) **73** 16-33. DOI: 10.1017/S0029665113003698
46. Meng SJ, Yu LJ. **Oxidative Stress, Molecular Inflammation and Sarcopenia**. *Int J Mol Sci* (2010.0) **11** 1509-1526. DOI: 10.3390/ijms11041509
47. Domazetovic V. **Oxidative stress in bone remodeling: Role of antioxidants**. *Clin Cases Miner Bone Metab* (2017.0) **14** 209-216. DOI: 10.11138/ccmbm/2017.14.1.209
48. Mulero J, Zafrilla P, Martinez-Cacha A. **Oxidative stress, frailty and cognitive decline**. *J Nutr Health Aging* (2011.0) **15** 756-760. DOI: 10.1007/s12603-011-0130-5
49. Mirmiran P, Azadbakht L, Esmaillzadeh A, Azizi F. **Dietary diversity score in adolescents-a good indicator of the nutritional adequacy of diets: Tehran lipid and glucose study**. *Asia Pac J Clin Nutr* (2004.0) **13** 56-60. PMID: 15003915
50. Solon-Biet SM, McMahon AC, Ballard JW, Ruohonen K, Wu LE, Cogger VC. **The ratio of macronutrients, not caloric intake, dictates cardiometabolic health, aging, and longevity in ad libitum-fed mice**. *Cell Metab* (2014.0) **19** 418-430. DOI: 10.1016/j.cmet.2014.02.009
51. Otsuka R, Kato Y, Nishita Y, Tange C, Nakamoto M, Tomida M. **Dietary diversity and 14-year decline in higher-level functional capacity among middle-aged and elderly Japanese**. *Nutrition* (2016.0) **32** 784-789. DOI: 10.1016/j.nut.2016.01.022
52. Yokoyama Y, Nishi M, Murayama H, Amano H, Taniguchi Y, Nofuji Y. **Dietary Variety and Decline in Lean Mass and Physical Performance in Community-Dwelling Older Japanese: A 4-year Follow-Up Study**. *J Nutr Health Aging* (2017.0) **21** 11-16. DOI: 10.1007/s12603-016-0726-x
53. Heiman ML, Greenway FL. **A healthy gastrointestinal microbiome is dependent on dietary diversity**. *Mol Metab* (2016.0) **5** 317-320. DOI: 10.1016/j.molmet.2016.02.005
54. Ni Lochlainn M, Bowyer RCE, Steves C. **Dietary Protein and Muscle in Aging People: The Potential Role of the Gut Microbiome**. *Nutrients* (2018.0) **10** 929. DOI: 10.3390/nu10070929
55. Kurosu A, Osman F, Daggett S, Peña-Chávez R, Thompson A, Myers SM. **Factors Associated with Self-Reported Dysphagia in Older Adults Receiving Meal Support**. *J Nutr Health Aging* (2021.0) **25** 1145-1153. DOI: 10.1007/s12603-021-1700-9
56. Lee MS, Huang YC, Wahlqvist ML. **Chewing ability in conjunction with food intake and energy status in later life affects survival in Taiwanese with the metabolic syndrome**. *J Am Geriatr Soc* (2010.0) **58** 1072-1080. DOI: 10.1111/j.1532-5415.2010.02870.x
57. 57.Gao T, Han S, Mo G, Sun Q, Zhang M, Liu H. Long-term tea consumption reduces the risk of frailty in older Chinese people: Result from a 6-year longitudinal study. Front Nutr. 2022;9:916791.: 10.3389/fnut.2022.916791.
58. 58.National Bureau of Statistics of China. China Statistical Yearbook 2006. In: Zheng JP, Jin ZF, Zhu WS, editors. China Statistics Press; 2006.
59. Wang H, Liu C, Fan H, Tian X. **Rising food accessibility contributed to the increasing dietary diversity in rural and urban China**. *Asia Pac J Clin Nutr* (2017.0) **26** 738-747. DOI: 10.6133/apjcn.052016.03
60. Park S, Kim HJ, Kim K. **Do Where The Elderly Live Matter? Factors Associated with Diet Quality among Korean Elderly Population Living in Urban Versus Rural Areas**. *Nutrients* (2020.0) **12** 1314. DOI: 10.3390/nu12051314
61. Tembo MC, Holloway-Kew KL, Sui SX, Dunning T, Low ACH, Yong SJ. **Prevalence of Frailty in Older Men and Women: Cross-Sectional Data from the Geelong Osteoporosis Study**. *Calcif Tissue Int* (2020.0) **107** 220-229. DOI: 10.1007/s00223-020-00713-3
62. 62.Ma ZZ, He JB, Sun SJ, Lu T. Patterns and stability of food preferences among a national representative sample of young, middle-aged, and elderly adults in China: a latent transition analysis. Food Qual Prefer. 2021;94:e104322.:10.1016/j.foodqual.2021.104322.
63. Bonnefoy M, Berrut G, Lesourd B, Ferry M, Gilbert T, Guérin O. **Frailty and nutrition: searching for evidence**. *J Nutr Health Aging* (2015.0) **19** 250-257. DOI: 10.1007/s12603-014-0568-3
64. Wolfe RR, Miller SL, Miller KB. **Optimal protein intake in the elderly**. *Clin Nutr* (2008.0) **27** 675-684. DOI: 10.1016/j.clnu.2008.06.008
65. Huang RY, Yang KC, Chang HH, Lee LT, Lu CW, Huang KC. **The Association between Total Protein and Vegetable Protein Intake and Low Muscle Mass among the Community-Dwelling Elderly Population in Northern Taiwan**. *Nutrients* (2016.0) **8** 373. DOI: 10.3390/nu8060373
66. Ni Lochlainn M, Cox NJ, Wilson T, Hayhoe RPG, Ramsay SE, Granic A. **Nutrition and Frailty: Opportunities for Prevention and Treatment**. *Nutrients* (2021.0) **13** 2349. DOI: 10.3390/nu13072349
67. Zainuddin NS, Husin MH, Ahmad NH, Yun Hua W, Wan Chien H, Shahar S. **Association between nutritional status, food insecurity and frailty among elderly with low income**. *JSKM* (2017.0) **15** 51-60. DOI: 10.17576/JSKM-2017-1501-07
68. Bustamante-Ara N, Villarroel L, Paredes F, Huidobro A, Ferreccio C. **Frailty and health risks in an agricultural population, Chile 2014–2017**. *Arch Gerontol Geriatr* (2019.0) **82** 114-119. DOI: 10.1016/j.archger.2019.01.012
|
---
title: Perceived factors that influence adoption, implementation and sustainability
of an evidence-based intervention promoting healthful eating and physical activity
in childcare centers in an urban area in the United States serving children from
low-income, racially/ethnically diverse families
authors:
- Leilah Siegel
- Yuka Asada
- Shuhao Lin
- Marian L. Fitzgibbon
- Angela Kong
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012626
doi: 10.3389/frhs.2022.980827
license: CC BY 4.0
---
# Perceived factors that influence adoption, implementation and sustainability of an evidence-based intervention promoting healthful eating and physical activity in childcare centers in an urban area in the United States serving children from low-income, racially/ethnically diverse families
## Abstract
### Introduction
Early childcare centers offer optimal settings to provide healthy built environments where preschool age children spend a majority of their week. Many evidence-based interventions (EBIs) promoting healthful eating and physical activity for early childcare settings exist, but there is a limited understanding of how best to support adoption, implementation and sustainability in community settings. This study examined how early childcare teachers and administrators from Chicago-area childcare centers serving children from low-income, racially/ethnically diverse communities viewed an EBI called Hip to Health (H3), and the factors they perceived as relevant for EBI adoption, implementation, and sustainability.
### Methods
A multiple methods study including key informant interviews and a brief survey was conducted. Key informant interviews with teachers and administrators from childcare centers located in Chicago, IL were completed between December 2020 and May 2021. An interview guide and coding guide based on the Consolidated Framework for Implementation Research (CFIR) was developed. Interview transcripts were team coded in MAXQDA Qualitative Data Analysis software. Thematic analysis was used to identify findings specific to adoption, implementation, and sustainability. Participants were also asked to respond to survey measures about the acceptability, feasibility, and appropriateness of H3.
### Results
Overall, teachers ($$n = 20$$) and administrators ($$n = 16$$) agreed that H3 was acceptable, appropriate, and feasible. Low start-up costs, ease-of-use, adaptability, trialability, compatibility, and leadership engagement were important to EBI adoption. Timely and flexible training was critical to implementation. Participants noted sustainability was tied to low ongoing costs, access to ongoing support, and positive observable benefits for children and positive feedback from parents.
### Conclusions
These findings suggest that EBIs suitable for adoption, implementation, and sustainment in childcare centers serving racially/ethnically diverse, low-income families should be adaptable, easy to use, and low-cost (initial and ongoing). There is also some evidence from these findings of the heterogeneity that exists among childcare centers serving low-income families in that smaller, less resourced centers are often less aware of EBIs, and the preparation needed to implement EBIs. Future research should examine how to better support EBI dissemination and implementation to these settings.
## Introduction
Physical inactivity and poor diet quality are major drivers of chronic disease. Intervening early on in life to encourage children's engagement in physical activity and eating healthful foods may protect against non-communicable disease development (1–3). This is particularly important for low-income and racial/ethnic minoritized groups who are disproportionately impacted by chronic disease [4, 5]. To address these health disparities and promote greater health equity, prevention efforts that promote healthful eating and physical activity need to have sufficient reach and be disseminated equitably.
Evidence-based interventions (EBIs) promoting physical activity and healthful eating that target preschool children in childcare settings have been shown to be effective in the United States and other high-income countries [6, 7]. Embedding EBIs as part of standard programming in existing childcare settings has the potential to expand the reach of EBIs and improve children's health on a population basis. However, in practice, EBIs are not always readily adopted [8] and even if adopted, challenges to implementation and sustainability exist (8–10). A recent systematic review suggests that strategies to support EBI implementation are usually needed, but selection of strategies are largely dependent on the local context [9]. Improving EBI translation to real world practice settings requires a greater understanding of why and how childcare centers implement these types of programs. It also requires a greater understanding of the factors needed for such EBIs to be successfully sustained.
This study assessed how teachers and administrators in an urban area in the United States viewed a specific EBI. Hip Hop to Health (H3) is an EBI that was developed for and previously tested in Head Start classrooms for African American and Latinx preschool children to be delivered by preschool teachers [11, 12]. This study examined the factors that teachers and administrators perceived as relevant to this EBI's adoption, implementation, and sustainability in childcare settings serving children from low-income and/or racial/ethnic minoritized families.
## Hip hop to health (H3) EBI
H3 is an EBI that was developed to be delivered by teachers in childcare settings serving African American and Latinx children. The randomized effectiveness trial testing H3 found significant between-group differences in physical activity, screen time, and diet quality that favored the intervention group [11, 12]. H3 consists of eight lessons which feature activities on topics such as “Go and Grow Foods vs. Slow Foods,” “Grains,” “Vegetables,” and “Drinking Water and Moving Your Body.” “ Go and Grow Foods” are healthy foods that should be eaten often, whereas “Slow” foods are foods that should be eaten in moderation. Each lesson is 35–40 minutes and consists of 20 mins of physical activity that can be done along with an accompanying musical soundtrack. Lessons also include additional activities such as reading stories, sampling foods, and using puppets [11].
## Study design
This is a multiple methods study including a survey and key informant interviews. Both qualitative and quantitative methods were used to obtain a more complete picture of the implementation, adoption, and sustainability factors. Data were gathered from early childcare administrators ($$n = 16$$) and teachers ($$n = 20$$) between December 2020 and May 2021. The study was reviewed and approved by the Institutional Review Board at the University of Illinois Chicago (protocol # 2020-0139) which reviewed the ethics and protection of the rights and welfare of the individuals involved in the proposed research.
## Study approach
Participants were verbally consented and then asked to watch a brief video describing the H3 curriculum. Following the video, participants were shown a sample lesson from H3 and completed a brief survey collecting quantitative data. Qualitative data was then collected via a semi-structured interview by a trained staff member.
## Sampling, setting, and key informants
Chicago is the third largest city in the United States and has a population of over 3 million people [13]. Demographically, Chicago's population is nearly evenly split between white residents ($33\%$), African American residents ($29\%$), and Latinx residents ($29\%$) [13]. Over $20\%$ of city inhabitants live below the national poverty line, with this rate varying based on race and ethnicity: $32\%$ of African American residents and $22\%$ of Latinx residents live below the poverty line as compared with $10\%$ of white residents [13].
Forty-five childcare centers were initially identified based on purposive sampling of those [1] located in Chicago or surrounding suburbs; [2] serving a population of 3–5 years of age; and [3] serving a largely low-income or African American or Latinx population. Head Start program participation was noted but not required.
The childcare centers were purposively sampled from (a) a list of Head Start centers that the senior author had generated from a previous study, and (b) online searches of early childhood centers in Chicago or surrounding suburbs that met additional criteria described above. Some sites were no longer active or were not able to be reached.
Teachers and administrators from these centers were identified via contact information online, or from the list of contacts of Head Start centers. An email describing the study was sent to these teachers and administrators inviting them to participate in the study. The email contained an attached flyer describing the study in more detail as well as the informed consent document. Teachers and administrators from 22 different centers indicated interest and were scheduled for interviews. Interview materials were only available in English so self-reported comfort with speaking and reading English was a requirement for participation.
Teacher interviews were conducted with early childhood education staff whose job titles were teacher, teacher's aide, or instructional coach. Teachers were interviewed because of their key role in EBI implementation; their perceptions and receptivity to the EBI are critical components to implementation. Administrator interviews were conducted with site directors, coordinators, nutritionists, and education managers at early childcare centers. They represent center leadership and are often involved in decision making; their insights are particularly important for learning about organizational support and capacity of the center.
## Interview procedures
All interviews were conducted via Zoom by a trained staff member in English. The staff member conducting the interviews identifies as African American, with the majority of the research team identifying as persons of color (Asian/Asian American). Previous studies have cited that mistrust, implicit bias, and lack of cultural competence could serve as barriers to individuals from racially minoritized backgrounds participating in research; these individuals are more likely to participate when they perceive that the researcher is similar in background to themselves (14–17).
A semi-structured interview guide was developed for the key informant interviews, guided by the Consolidated Framework for Implementation Research (CFIR) [18]. CFIR is a meta-theoretical framework that can be used to identify barriers and facilitators related to EBI adoption, implementation, and sustainability. The interview guide included select CFIR constructs within the following domains: [1] intervention characteristics, [2] outer setting, [3] inner setting, and [4] characteristics of the individual. A summary of constructs included for each domain is summarized in the Supplementary Table. Key informants were asked to provide their input about H3. The questions also captured demographic information and key informants' previous experiences with adopting, implementing, and sustaining similar programs. Before finalizing the interview guide, pilot interviews were conducted with five teachers and administrators from the target population to check for clarity, correct terminology, and flow. The interview guide was revised several times following pilot interviews. Slight variations in the questions used in the interview guides were also included to make the questions more relevant for a teacher or an administrator. Verbal informed consent was obtained from each participant before beginning the interviews. All interviews were audio recorded and professionally transcribed.
## Survey procedures
Acceptability, feasibility, and appropriateness of the EBI (i.e., H3) were assessed using the Acceptability of Intervention Measure (AIM), Feasibility of Intervention Measure (FIM), and Intervention Appropriateness Measure (IAM) developed by Weiner et al. [ 19]. Each measure has four items assessed on a five-point Likert scale with responses ranging from completely disagree to completely agree; higher scores reflect better acceptability, feasibility, and appropriateness. Participants were asked to respond to survey questions after watching the brief video and before beginning the qualitative interview. All participants completed both the qualitative interview and the quantitative survey.
## Qualitative analysis
An a priori draft codebook was created following the CFIR-informed interview guide and revised during several rounds of coding. All transcripts were uploaded into MAXQDA Qualitative Analysis software [20]. To begin, coders (LS, YA, SL, AK) independently coded a subset of the transcripts to discuss discrepancies in coding and revise coding definitions as needed. Coders then met weekly to discuss coding progress, further refine the coding guide, and to identify patterns. Once coders reached approximately >$85\%$ inter-rater agreement and no additional revisions were required of the coding guide, each remaining transcript was double coded. Coders used MAXQDA functions such as code matrices and summary grids to visualize data and look for cross-cutting patterns. Based on weekly team discussions and iterative revisions to data displays, themes specific to adoption, implementation, and sustainability were developed and documented [21].
## Quantitative analysis
Descriptive statistics, presented as means or percentages as appropriate, were calculated to describe the study sample and to summarize FIM, AIM, and IAM scores.
## Demographic characteristics
Table 1 describes characteristics of key informants and the childcare centers where they are employed. Twenty early childcare teachers and 16 administrators were interviewed; most were affiliated with Head Start programs ($95\%$ of teachers, $62\%$ of administrators), with the majority of teachers and administrators holding their positions for more than 6 years. Interview respondents were all female; $56\%$ of administrators and $40\%$ of teachers self-identified as African American, and $6\%$ of administrators and $21\%$ of teachers self-identified as Hispanic. The highest level of education obtained for most administrators was a master's degree ($50\%$); the highest level of education for the majority of teachers was a bachelor's degree ($55\%$). The average age of all respondents was 47.5 years. Fifty percent of administrators and $70\%$ of teachers practiced in large centers, defined as centers with multiple locations serving more than 100 children. The remaining practiced in single site settings. Mid-size was defined as single site centers with more than 50 children and small was defined as single site centers with <50 children.
**Table 1**
| Demographics | Administrators (n = 16) | Teachers (n = 20) |
| --- | --- | --- |
| Sex (female) | 100% | 100% |
| Mean age in years | 48 | 47 |
| Ethnicity | | |
| Hispanic | 6% | 21% |
| Non-Hispanic | 94% | 73% |
| Race | | |
| African American | 56% | 40% |
| Native American | 0% | 5% |
| Two or more races | 6% | 5% |
| White | 38% | 50% |
| # of years in position | | |
| 0–5 | 19% | 20% |
| 6–10 | 31% | 40% |
| 11–20 | 31% | 20% |
| 21+ | 19% | 20% |
| Highest level of education completed | | |
| High school diploma | 0% | 10% |
| Associate's degree | 12% | 15% |
| Bachelor's degree | 38% | 55% |
| Master's degree | 50% | 20% |
| Center location | | |
| Suburban | 69% | 50% |
| Urban | 31% | 50% |
| Head start? | | |
| N | 38% | 5% |
| Y | 62% | 95% |
| Center size | | |
| Small (1 site < 50 students) | 44% | 5% |
| Mid-size (1 site > 50 students) | 6% | 25% |
| Large (Multiple sites < 100 students) | 50% | 70% |
## Quantitative data results: Acceptability, feasibility, and appropriateness of the EBI
Participants provided their impressions of H3's acceptability, feasibility, and appropriateness by responding to AIM, FIM, and IAM survey items. Table 2 reports AIM, IAM, and FIM mean scores by individual item and category totals. Most teachers and administrators agreed that H3 was acceptable (mean: 4.24, SD: 0.50), feasible (mean: 4.31, SD: 0.46), and appropriate (mean: 4.21, SD: 0.47).
**Table 2**
| Item | Description | Mean (SD) | Range |
| --- | --- | --- | --- |
| AIM: Approval | Hip Hop to Health meets my approval | 4.08 (0.84) | 1–5 |
| AIM: Appeal | Hip Hop to Health is appealing to me | 4.39 (0.60) | 3–5 |
| AIM: Welcome | I welcome Hip Hop to Health | 4.31 (0.58) | 3–5 |
| AIM: Like | I like Hip Hop to Health | 4.19 (0.58) | 3–5 |
| | Total | 4.24 (0.50) | |
| FIM: Implement | Hip Hop to Health seems implementable | 4.11 (0.54) | 3–5 |
| FIM: Possible | Hip Hop to Health seems possible | 4.36 (0.54) | 3–5 |
| FIM: Doable | Hip Hop to Health seems doable | 4.39 (0.49) | 4–5 |
| FIM: Easy | Hip Hop to Health seems easy to use | 4.28 (0.61) | |
| | Total | 4.31 (0.46) | |
| IAM: Fitting | Hip Hop to Health seems fitting | 4.28 (0.57) | 3–5 |
| IAM: Suitable | Hip Hop to Health seems suitable | 4.19 (0.47) | 3–5 |
| IAM: Applicable | Hip Hop to Health seems applicable | 4.19 (0.58) | 3–5 |
| IAM: Match | Hip Hop to Health seems like a good match | 4.17 (0.61) | 3–5 |
| | Total | 4.21 (0.47) | |
## Qualitative data results: Factors influencing adoption, implementation, and sustainability
Table 3 summarizes main themes that emerged from the key informant interviews and are organized by CFIR domains and constructs along with accompanying quotes. Most themes were related to CFIR constructs within the domains of “intervention characteristics” (i.e., cost, adaptability, trialability, complexity) and “inner setting” (i.e., compatibility, available resources, leadership engagement) [12]. Themes cut across adoption, implementation, and sustainability as shown in Table 3 and are described further in the next sections.
**Table 3**
| CFIR Domains: | A | I | S | Example quotes |
| --- | --- | --- | --- | --- |
| Constructs | | | | |
| Themes | | | | |
| Intervention Characteristics: Costs | x | x | x | “I mean, definitely cost because we are funded by grants, so whatever dollar amount is allocated for health and nutrition, education would definitely play a role in implementing the curriculum.” (Teacher ID 2010) |
| Initial and ongoing costs are reasonable | | | | “Cost. That's the big one.” (Administrator ID 1005) |
| Intervention Characteristics: Adaptation | x | x | x | “That's the part that I was wondering about is with the flexibility and I feel like if the 20 mins are divided, that it could be done...break up program into smaller chunks.” (Adminstrator ID 1004) |
| Can be adapted to fit into current practices and routines | | | | “Maybe having two kits per classrooms in case you want to do it in smaller groups. Instead of a classroom of 20 trying to do it, you know, maybe two teachers are doing it at different times.” (Administrator ID 1002) |
| | | | | “I mean, because I don't know it thoroughly, I guess I would say maybe the language part, maybe just adding different languages in there. Maybe not for kids so much but for families. Cause we do flyers home and, you know, I think they would appreciate it a lot more if they were able to read it and understand it. So maybe that would be a good to change.” (Teacher ID 2011) |
| Intervention Characteristics: Trialability | x | | | “… obviously learning about it and then maybe having the opportunity to try it, like pilot it in a couple of classrooms and then you know, see how it goes and then make the decision as to purchase it for the entire program.” (Administrator ID 1009) |
| Ability to pilot the program and to gather feedback from teachers | | | | “She would definitely want the teachers to try it out and then get feedback because she does trust teacher input, you know, after her own evaluation, see if it, she thinks it would be successful in the classroom and then give it that test run and then ask for feedback from the teachers on whether or not it was successful or what could make it successful.” (Teacher ID 2012) |
| Intervention Characteristics: Complexity | x | x | x | “Again, it just really boils down to it being easy, not something that's a burden on the teachers feeling like it's another task to do in the classroom.” (Administrator ID 1008) |
| Curriculum must be easy to use | | | | “And then I would look at how difficult, or how simple it is to, to put together or to, to implement the curriculum. And then I would look at how much work, how much additional work and how many additional supplies would be needed. That's normally what has either caused me to use a curriculum or to let it fall off.” (Administrator ID 1016) |
| Inner Setting: Compatibility | x | x | x | “…we already use food experiences and exercise and stuff like that. I think it would just be ongoing because it'll be a part of our curriculum. That's something we already do anyway.” (Teacher ID 2005) |
| Aligns with priorities, standards, and current practices | | | | “Well, I think it should make sure it fits into the Head Start standards for sure. And other, the PI or the PFA standards. And to just make sure that there's no, I mean, I don't think there's anything that is culturally inappropriate. That it can fit into the program day. I mean, I kind of don't know, but those are things that I imagine are important.” (Administrator ID 1004) |
| Inner Setting: Available Resources: Training is critical to implementation and maintenance. Training needs to be timely, flexible, and ongoing. | x | x | x | “Would definitely require constant training to keep it going. Cause we, you know, often in programs we start things, we stop, we start, stop.” (Administrator ID 1002) |
| | | | | “The long-term cost availability continued trainings and support…” (Administrator ID 1010) |
| | | | | “It would be accepted if it's presented and implemented in a timely fashion. If this were to be something that we were to pilot, it would need to start in August, educate the staff so they can learn it and then bring it in. When the children come in, if it were to be something that was brought in in November or December, it would be more difficult because since we're grant funded, we have several deadlines. So to be most effective, it seems that it's, that a timeline would be implemented in the August early August, teaching the teachers and then them full speed ahead at end of August when the children were to come on, would be the most effective way. It's crucial the timeline, the one you would present it to teachers to be very honest.” (Administrator 1005) |
| Inner Setting: Leadership Engagement | x | x | x | “it's a matter of getting them [administrators] on board and then also have a creating that time for, for training the teachers and getting the resources. Important to leadership on board and providing training and resources to support teachers in implementation” (Teacher ID 2012) |
| Commitment from administrators and others with decision making capabilities | | | | |
| | | | | “Yes, the parents I mean, in the office of Head Start there's a parent board. So they have also with budgets, they have to get it approved by the parent board.” (Teacher ID 2004) |
| Positive benefits to children | | | x | “Well, because if it proves to be beneficial, it benefits the program. I guess there's some more success stories about, you know, the overall wellbeing of families and children” (Teacher ID 2009) |
| Maintenance of the EBI is strengthened if there is evidence that children are benefiting from the program. | | | | “Long-Term? I think again, I think the child's outcomes, so if children are, retaining the information and the curriculum is successful in getting that, I think that we would not have a reason to change.” (Administrator ID 1009) |
## Adoption
Understanding factors that lead to the adoption of an EBI helps researchers to both adapt and design future interventions. Constructs particularly relevant to EBI adoption were cost, adaptability, trialability, complexity (i.e., ease of use), compatibility, and leadership engagement.
## Costs: Initial costs are reasonable
CFIR defines the construct of “cost” as “costs of the intervention and costs associated with implementing the intervention, including investment, supply, and opportunity costs” [18]. Initial or start-up costs were of particular importance in considering whether to adopt a curriculum. Interview participants were given a sample one-time curriculum price of $65 and asked if they thought that cost was “reasonable.” Many participants from large centers stated that the cost was reasonable since they'd “paid far more” for other curricula materials. Many participants from small centers reported that the amount quoted was reasonable because it was lower than the amount they had in mind, even though they did not have other curriculum to compare it to.
## Adaptability: Can be adapted to fit into current practices and routines
Adaptability is defined as, “the degree to which an intervention can be adapted, tailored, refined, or reinvented to meet local needs” [18]. Both teachers and administrators described intervention adaptability as important to its adoption. Specifically, they mentioned several characteristics of adaptability, such as [1] having the intervention translated into multiple languages; [2] being able to modify the length or use it as a series of separate modules; [3] adapting it for slightly younger or older children; and [4] being able to modify it to for virtual use (Table 3).
## Trialability: Ability to pilot the program and to gather feedback from teachers
Trialability is a CFIR construct that is defined as: “The ability to test the intervention on a small scale in the organization, and to be able to reverse course (undo implementation) if warranted.” [ 18]. Teachers and administrators mentioned that being able to pilot the curriculum before deciding whether or not to adopt it was very important. Additionally, it would be important to obtain positive feedback from teachers before committing to a program (Table 3).
## Complexity: Curriculum must be easy to use
Complexity is defined as “perceived difficulty of the intervention, reflected by duration, scope, radicalness, disruptiveness, centrality, and intricacy and number of steps required to implement” [18]. Many administrators and teachers expressed the need for the EBI to be easy to use, which considered multiple dimensions. One administrator commented that adopting the curriculum “should not [be] something that's a burden on the teachers…like it's another task to do in the classroom.” Curriculum that is disruptive to current workflow and practices or had too many steps would be barriers to adoption (Table 3).
## Compatibility: Aligns with priorities, standards, and current practices
Compatibility is defined as “the degree of tangible fit between meaning and values attached to the intervention by involved individuals, how those align with individuals' own norms, values, and perceived risks and needs, and how the intervention fits with existing workflows and systems” [18]. Teachers and administrators responded that it was important that an intervention fit their organization's values, norms, and policies to be considered. Many participants reported that their organizations placed a priority on student health and that they already implemented activities to promote nutrition and/or physical activity. For teachers and administrators affiliated with Head Start, an EBI that aligned with Head Start standards was of great importance for adoption.
## Leadership engagement: Commitment from administrators and those with decision making capabilities
CFIR describes leadership engagement as “commitment, involvement, and accountability of leaders and managers with the implementation” [18]. Engagement from leadership is of particular importance in larger centers such as the Head Start affiliated centers. Buy-in from administrators is necessary to support organizational capacity at all phases, but it is of particular importance when deciding to adopt a program. One teacher mentioned “it's a matter of getting them [administrators] on board and then also creating that time for training the teachers and getting the resources.” Centers affiliated with Head Start had to get approval from different boards before programs were adopted, including an advisory board consisting of parents (Table 3).
## Implementation
Many of the constructs relevant to adoption are also relevant to EBI implementation as summarized in Table 3. However, there was a particular emphasis on training (CFIR construct: Available resources) to support EBI implementation.
## Available resources: Training needs to be timely and flexible
The construct “Available Resources” is defined as: “the level of resources dedicated for implementation and on-going operations, including money, training, education, physical space, and time” [18]. Both teachers and administrators described training as being critical for implementation; specifically, two main considerations included: (a) timing; (b) delivery/format. First, trainings should be offered when teachers were onboarded for the school year and received trainings for other curriculum/procedures (e.g., in August). This was particularly relevant for Head Start centers.
Second, when teachers were asked about optimal training delivery, many saw advantages to both online offerings and in person. Many expressed a preference for hands-on learning, but they also liked the convenience and permanence of online trainings.
## Sustainability
Interview respondents were asked if they could see their center using the H3 curriculum in the long term, and what factors would influence their center's ability to use the curriculum in the long term. The most common themes related to EBI sustainability were ongoing costs, training support, and evidence that children are positively benefiting from the program.
## Cost: Ongoing costs need to be manageable and to support ongoing training
Both teachers and administrators stated that low on-going costs were extremely important in sustaining EBI implementation. Specific costs mentioned were for printing, fuel, food, and replacing program components.
Ongoing support or continued training was also mentioned as a crucial factor for EBI sustainability. In large centers in particular, there is a consistent need to offer training to current staff in the form of booster sessions and to train new teachers since staff turnover is common.
## Positive benefits to children: Need to see evidence that the EBI is working
Both teachers and administrators reported the importance of seeing children's positive reactions and benefits from the program in observable ways. In addition, many teachers would consider the program successful if the benefits also extended to parents.
## Differences by center size and type
*In* general, larger or multi-site centers, such as Head Start centers, had adopted EBI programs in the past. Participants from these centers reported more familiarity and readiness, as well as cited existing regulations and policies that support nutrition and physical activity curricula. One administrator stated: “Some months we have a focus, it could be portion sizing, it could be a healthy eating activity. And we do this program once a year; they come in and they teach the kids. It's not the teacher's doing it. It's this organization doing it. And then the children get to take something home with them, for example, like the plates, little dividing of the plate for the serving sizes. So that's currently what we do.” In contrast, none of the participants from small centers in this study had implemented a formal EBI previously. Many expressed “creating” their own program by pulling together resources or using those that were given to them. One teacher said: “Our director, every week she sends recipes about healthy nutrition so we can show [them] to the kids and share [them] with the parents. Every week we do that. And for the physical activity, other than going to the playground in the shade or simple activities in the classroom, like dancing, that's it. That's all.”
## Discussion
These findings highlight factors related to EBI adoption, implementation, and sustainability in childcare centers within an urban area in the United States serving low-income, racially/ethnically diverse families. Successfully adopting, implementing, and sustaining an EBI promoting positive health behaviors in early childhood can be one strategy to promote greater health equity in these populations. In this study, teachers and administrators responded favorably to the EBI (i.e., H3) presented to them and agreed that it was acceptable, appropriate, and feasible. In considering EBI adoption, implementation, and sustainability, respondents stressed the need for the EBI to fit into what they were already doing. It also needed to be low-cost (start-up, ongoing), easy to use, and have training supports that were flexible to the needs of the center and would be ongoing. However, there were notable differences between small and large centers in their readiness and capacity for EBI adoption that warrant further attention.
Teachers and administrators interviewed were largely in favor of the EBI proposed as reflected in both qualitative and quantitative findings (e.g., AIM, FIM, IAM scores) [19]. In most cases, centers were already promoting physical activity and healthful eating in some form; therefore, many viewed the EBI as compatible with existing practices and could reinforce what they were already doing. Compatibility has been recognized as a facilitator to program adoption based on previous studies of physical activity and nutrition interventions delivered in childcare settings (22–30). For example, EBIs that “fit well within existing curricula”), “enhanced the classroom,” or were aligned with existing “preschool and government health objectives” were considered facilitators to implementation [23, 24, 27].
There was also consensus among teachers and administrators that EBIs needed to be easy to use and could be easily adapted to a center's routine or practices. The adaptations most often mentioned by key informants in the current study included breaking up sessions into shorter lessons to accommodate daily routines and adapting lesson plans to accommodate varying class sizes, age groups, and language needs (e.g., translation of parent handouts). Similar adaptations have been identified in previous studies. These studies included settings with predominantly white populations (e.g., Sweden and Scotland); however, the income status of families with children enrolled in the centers was not reported [23, 27, 31, 32] as is often the case in many of these studies. The theme of adaptability was also found in studies conducted in Head Start centers which serve low-income families [26, 33, 34], which is more similar to the target population in our study. Implementation also occurs more smoothly when interventions are perceived as easy to use, require little to no preparation (e.g., ready to use), and are not overly burdensome. This facilitator to implementation (ease of use) has been largely reported in centers with predominantly white populations (income status not reported) [23, 24, 31, 35, 36]. When this theme (ease of use) was reported in racially/ethnically diverse settings and/or Head Start centers [22, 25, 37], it was more common to perceive interventions in terms of its complexity rather than ease of use. For instance, interventions viewed as too complex and therefore difficult to implement were those with too many activities, had excessive paperwork, or required too much planning. In contrast to our findings, these themes were gathered after an intervention was implemented and therefore, centers could speak better to the challenges they encountered with implementation.
Cost was important to both EBI adoption and sustainability. Both administrators and teachers mentioned that the cost of the intervention was an essential factor for deciding to use the curriculum and being able to continue to use it over time. Specific cost-factors that were mentioned were the initial cost of the curriculum, ongoing costs such as replacing materials that became lost or worn out, and food costs. This highlighted the importance of considering the cost to maintain the curriculum over time (beyond startup costs). This finding was addressed by Eismann et al. [ 38], who reported that organizations often fail to successfully implement EBIs in part because they do not realize up front what costs will be needed to sustain the intervention. Burton et al. [ 39] noted that participants perceived the cost of their EBI (including “investment, supply, and opportunity cost”) was prohibitive and a barrier to implementation. Consideration of cost and cost effectiveness is not often reported in studies examining healthy eating and physical activity practices or programs in childcare settings as noted by previous systematic reviews on physical activity and healthy eating interventions in childcare settings [40, 41].
Another key finding was the importance that both administrators and teachers placed on being well-trained to implement the intervention, which is critical for the success of any EBI. This was consistently found in studies across contexts including centers serving populations that were predominantly white, racially/ethnically diverse, and low-income (22–25, 27, 30, 32–34, 36, 42). As emphasized in this current study, trainings should be planned with the partner organizations to adequately consider their needs and preferences. Specifically, the type of training, whether online, in-person or a hybrid, as well as the timeline of training were mentioned as being critically important. Due to the calendar of the school year, having trainings begin shortly after teachers arrive back at school in August was mentioned as a key factor to implementation success. Also, the availability of resources, ongoing trainings and support from the university, and a designated contact person that teachers and administrators can contact for help or with questions were listed as being extremely important for sustaining an EBI. A 2021 paper by Combs et al. reported that training is a key part of EBI implementation, but that training must be conducted in a manner that is most useful to the center in terms of scheduling and mode [43]. Teachers in the current study reported both pros and cons to online training—while it offered additional convenience it lacked a hands-on component that many early childhood teachers reported was very helpful when learning a new curriculum. Combs et al. also found that while online training did not lead to lower levels of adherence to the curriculum or dosage, it was associated with lower reports of quality of delivery [43]. Their recommendation was to include some experiential component to online training; the findings of this paper support that recommendation.
Finally, it is important to note some differences by center type/size that could have implications for EBI adoption and dissemination. *In* general, centers affiliated with Head Start were larger and/or part of multi-site centers. When interviewing administrators and teachers from Head Start centers, most had implemented EBIs or similar programs in the past so there were mechanisms in place and organizational capacity to support EBI adoption and implementation. In contrast, key informants from small, single site centers were not at all familiar with EBIs; however, they too, prioritized promoting physical activity and healthful eating in their centers. This suggests that EBI dissemination may favor centers that are larger, have greater organizational capacity (e.g., leadership, available resources, etc.), and are more likely to be networked with external organizations (e.g., academic institutions). Researchers have a role in perpetuating this bias since the development and testing of EBIs usually originate from grant funded studies carried out by academic institutions. A gap in the current research is the equitable dissemination of EBIs [44]. Addressing this gap means better dissemination of EBIs through potential systems and policy changes, as well as developing implementation strategies to support EBI adoption and implementation in smaller, less resourced and networked centers.
## Limitations
These findings have some limitations. This study used purposive sampling methods that recruited teachers and administrators from childcare centers based on characteristics (e.g., centers serving low-income, racially/ethnically diverse families) that were representative of the target population of this study and could speak to the phenomenon under investigation. One limitation of purposive sampling is that it can be prone to researcher bias, since the researcher is making a decision about who to sample [45]. Another possible limitation is that these results may not be representative of EBI facilitators and barriers outside of the studied population [46]. However, this approach was still used as it provided the most time and resource-effective means of recruiting the targeted population due to challenges during the COVID-19 pandemic. A purposive sampling approach also provided additional insight into EBI implementation within this specific population.
This was also a cross sectional study that captured perceived views, thoughts, and insights from key informants at one point in time, prior to EBI implementation. Cross sectional designs are limited in their ability to deduce a causal relationship between the variables being studied and to describe a phenomenon over a period of time [47]. However this design allowed for a relatively timely and straightforward study. A cross sectional design also allowed for the study of multiple possible implementation factors concurrently [47]. Lastly, a third possible limitation is the relatively prospective nature of these findings; however, assessing the EBI prior to implementation may save valuable time and resources when H3 is fully implemented, and ultimately lead to a more impactful intervention.
## Conclusions
Overall, the study findings indicate that EBIs should be easy to implement, low-cost (initial and ongoing), have proper training supports, and be compatible with the practices and policies of early childcare centers to be successfully adopted, implemented, and sustained. Further attention should also be given to more equitable dissemination of EBIs and understanding how to support the adoption, implementation, and sustainability of EBIs in smaller, less-resourced centers.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by Institutional Review Board of the University of Illinois Chicago (Protocol # 2020-0139). Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.
## Author contributions
AK, LS, MF, and YA contributed to the design of the study. LS collected the data and wrote the first draft of the manuscript. YA directed the qualitative analysis. LS, YA, AK, and SL analyzed and interpreted data. LS and AK developed the figures and tables. All authors reviewed the manuscript, provided critical feedback, and approved the final manuscript.
## Funding
Research reported in this publication was supported in part by the University of Illinois Cancer Center Pilot Project Program funds.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frhs.2022.980827/full#supplementary-material
## References
1. Loprinzi PD, Lee IM, Andersen RE, Crespo CJ, Smit E. **Association of concurrent healthy eating and regular physical activity with cardiovascular disease risk factors in U. S**. *Youth Am J Health Promot.* (2015) **30** 2-8. DOI: 10.4278/ajhp.140213-QUAN-71
2. Raitakari O, Pahkala K, Magnussen CG. **Prevention of atherosclerosis from childhood**. *Nat Rev Cardiol.* (2022) **19** 543-54. DOI: 10.1530/ey.19.14.17
3. Kerr J, Anderson C, Lippman SM. **Physical activity, sedentary behaviour, diet, and cancer: an update and emerging new evidence**. *Lancet Oncol.* (2017) **18** e457-71. DOI: 10.1016/S1470-2045(17)30411-4
4. Cuenca-García M, Ortega FB, Ruiz JR, González-Gross M, Labayen I, Jago R. **Combined influence of healthy diet and active lifestyle on cardiovascular disease risk factors in adolescents**. *Scand J Med Sci Sports.* (2014) **24** 553-62. DOI: 10.1111/sms.12022
5. Massetti GM, Thomas CC, Ragan KR. **Disparities in the context of opportunities for cancer prevention in early life**. *Pediatrics.* (2016) **138** S65-77. DOI: 10.1542/peds.2015-4268J
6. Taveras EM, Gillman MW, Kleinman KP, Rich-Edwards JW, Rifas-Shiman SL. **Reducing racial/ethnic disparities in childhood obesity: the role of early life risk factors**. *JAMA Pediatr.* (2013) **167** 731-8. DOI: 10.1001/jamapediatrics.2013.85
7. Scott-Sheldon LAJ, Hedges LV, Cyr C, Young-Hyman D, Khan LK, Magnus M. **Childhood obesity evidence base project: a systematic review and meta-analysis of a new taxonomy of intervention components to improve weight status in children 2–5 years of age, 2005–2019**. *Child Obes.* (2020) **16** S221-s48. DOI: 10.1089/chi.2020.0139
8. Kenney EL, Wintner S, Lee RM, Austin SB. **Obesity prevention interventions in US public schools: are schools using programs that promote weight stigma?**. *Prev Chronic Dis.* (2017) **14** E142. DOI: 10.5888/pcd14.160605
9. Cassar S, Salmon J, Timperio A, Naylor PJ, van Nassau F, Contardo Ayala AM. **Adoption, implementation and sustainability of school-based physical activity and sedentary behaviour interventions in real-world settings: a systematic review**. *Int J Behav Nutr Phys Act.* (2019) **16** 120. DOI: 10.1186/s12966-019-0876-4
10. Totura CM, Figueroa HL, Wharton C, Marsiglia FF. **Assessing implementation of evidence-based childhood obesity prevention strategies in schools**. *Prev Med Rep.* (2015) **2** 347-54. DOI: 10.1016/j.pmedr.2015.04.008
11. Fitzgibbon ML, Stolley MR, Schiffer LA, Braunschweig CL, Gomez SL, Van Horn L. **Hip-Hop to Health Jr. Obesity prevention effectiveness trial: postintervention results**. *Obesity (Silver Spring).* (2011) **19** 994-1003. DOI: 10.1038/oby.2010.314
12. Kong A, Buscemi J, Stolley MR, Schiffer LA, Kim Y, Braunschweig CL. **Hip-Hop to Health Jr. Randomized effectiveness trial: 1-year follow-up results**. *Am J Prev Med.* (2016) **50** 136-44. DOI: 10.1016/j.amepre.2015.07.008
13. 13.United States Census Bureau. QuickFacts: Chicago City, Illinois. (2021). Available online at: https://www.census.gov/quickfacts/chicagocityillinois (accessed September 27, 2022).. *QuickFacts: Chicago City, Illinois* (2021)
14. Otado J, Kwagyan J, Edwards D, Ukaegbu A, Rockcliffe F, Osafo N. **Culturally competent strategies for recruitment and retention of African American populations into clinical trials**. *Clin Transl Sci.* (2015) **8** 460-6. DOI: 10.1111/cts.12285
15. Ashing-Giwa KT, Padilla GV, Tejero JS, Kim J. **Breast cancer survivorship in a multiethnic sample: challenges in recruitment and measurement**. *Cancer.* (2004) **101** 450-65. DOI: 10.1002/cncr.20370
16. Bell LS, Butler TL, Herring RP, Yancey AK, Fraser GE. **Recruiting blacks to the Adventist health study: do follow-up phone calls increase response rates?**. *Ann Epidemiol.* (2005) **15** 667-72. DOI: 10.1016/j.annepidem.2005.02.003
17. Yancey AK, Ortega AN, Kumanyika SK. **Effective recruitment and retention of minority research participants**. *Annu Rev Public Health.* (2006) **27** 1-28. DOI: 10.1146/annurev.publhealth.27.021405.102113
18. Damschroder LJ, Aron DC, Keith RE, Kirsh SR, Alexander JA, Lowery JC. **Fostering implementation of health services research findings into practice: a consolidated framework for advancing implementation science**. *Implement Sci.* (2009) **4** 50. DOI: 10.1186/1748-5908-4-50
19. Weiner BJ, Lewis CC, Stanick C, Powell BJ, Dorsey CN, Clary AS. **Psychometric assessment of three newly developed implementation outcome measures**. *Implement Sci.* (2017) **12** 108. DOI: 10.1186/s13012-017-0635-3
20. 20.VERBI Software. MAXQDA. (2020). Berlin, Germany: VERBI Software. Available online at: maxqda.com (accessed August 9, 2021).. *MAXQDA.* (2020)
21. Braun V, Clarke V. **Using thematic analysis in psychology**. *Qual Res Psych.* (2006) **3** 77-101. DOI: 10.1191/1478088706qp063oa
22. Bellows L, Anderson J, Gould SM, Auld G. **Formative research and strategic development of a physical activity component to a social marketing campaign for obesity prevention in preschoolers**. *J Community Health.* (2008) **33** 169-78. DOI: 10.1007/s10900-007-9079-z
23. Malden S, Reilly JJ, Hughes A, Bardid F, Summerbell C, De Craemer M. **Assessing the acceptability of an adapted preschool obesity prevention programme: ToyBox-Scotland**. *Child Care Health Dev.* (2020) **46** 213-22. DOI: 10.1111/cch.12736
24. Soltero EG, Parker NH, Mama Dr PS, Ledoux TA, Lee RE. **Lessons learned from implementing of garden education program in early child care**. *Health Promot Pract.* (2021) **22** 266-74. DOI: 10.1177/1524839919868215
25. Murtha K, Thompson K, Cleland P, Gallegos D. **Adaptation and evaluation of a nutrition and physical activity program for early childhood education settings in Aboriginal and Torres Strait Islander communities in remote Far North Queensland**. *Health Promot J Austr.* (2021) **32** 163-71. DOI: 10.1002/hpja.352
26. Hoffman JA, Schmidt EM, Castaneda-Sceppa C, Hillman CH. **The theoretical foundation, fidelity, feasibility, and acceptability of a teacher training to promote physical activity among preschoolers in child care: a pilot study**. *Prev Med Rep.* (2019) **13** 214-7. DOI: 10.1016/j.pmedr.2019.01.003
27. Lebron CN, Ofori A, Sardinas K, Luaces M, Natale R, Messiah SE. **Barriers and facilitators to obesity prevention dissemination and implementation efforts in the childcare centre setting from the provider perspective**. *Child Care Health Dev.* (2020) **46** 352-9. DOI: 10.1111/cch.12752
28. Hassani K, Buckler EJ, McConnell-Nzunga J, Fakih S, Scarr J, Mâsse LC. **Implementing appetite to play at scale in British Columbia: evaluation of a capacity-building intervention to promote physical activity in the early years**. *Int J Environ Res Public Health.* (2020) **17** 1132. DOI: 10.3390/ijerph17041132
29. Ledoux T, Thompson D, O'Connor T, Avery D, Kochi C, O'Connor DP. **Cross-site process evaluation results for the early childhood education center setting: CORD study**. *Child Obes.* (2020) **16** 350-7. DOI: 10.1089/chi.2019.0314
30. Swindle T, Johnson SL, Davenport K, Whiteside-Mansell L, Thirunavukarasu T, Sadasavin G. **A mixed-methods exploration of barriers and facilitators to evidence-based practices for obesity prevention in head start**. *J Nutr Educ Behav* (2019) **51** 1067-79 e1061. DOI: 10.1016/j.jneb.2019.06.019
31. Bergstrom H, Haggard U, Norman A, Sundblom E, Schafer Elinder L, Nyberg G. **Factors influencing the implementation of a school-based parental support programme to promote health-related behaviours—interviews with teachers and parents**. *BMC Public Health.* (2015) **15** 541. DOI: 10.1186/s12889-015-1896-x
32. Driediger M, Vanderloo LM, Burke SM, Irwin JD, Gaston A, Timmons BW. **The implementation and feasibility of the supporting physical activity in the childcare environment (SPACE) intervention: a process evaluation**. *Health Educ Behav.* (2018) **45** 935-44. DOI: 10.1177/1090198118775489
33. Allar I, Jones E, Elliott E, Kristjansson A, Taliaferro A, Mann M. **The perceived impact of I am moving, I am learning on physical activity and family involvement: a preliminary investigation**. *Am J Health Behav.* (2017) **41** 683-92. DOI: 10.5993/AJHB.41.6.2
34. Sussman AL, Davis S. **Integrating formative assessment and participatory research: Building healthier communities in the CHILE Project**. *Am J Health Educ.* (2010) **41** 244-9. DOI: 10.1080/19325037.2010.10599150
35. Howie EK, Brewer A, Brown WH, Pfeiffer KA, Saunders RP, Pate RR. **The 3-year evolution of a preschool physical activity intervention through a collaborative partnership between research interventionists and preschool teachers**. *Health Educ Res.* (2014) **29** 491-502. DOI: 10.1093/her/cyu014
36. Petrunoff N, Lloyd B, Watson N, Morrisey D. **Suitability of a structured fundamental movement skills program for long day care centres: a process evaluation**. *Health Promot J Aust.* (2009) **20** 65-8. DOI: 10.1071/HE09065
37. Alhassan S, Whitt-Glover MC. **Intervention fidelity in a teacher-led program to promote physical activity in preschool-age children**. *Prev Med.* (2014) **1** S34-36. DOI: 10.1016/j.ypmed.2014.07.024
38. Eisman AB, Quanbeck A, Bounthavong M, Panattoni L, Glasgow RE. **Implementation science issues in understanding, collecting, and using cost estimates: a multi-stakeholder perspective**. *Implement Sci.* (2021) **16** 75. DOI: 10.1186/s13012-021-01143-x
39. Burton W, Twiddy M, Sahota P, Brown J, Bryant M. **Participant engagement with a UK community-based preschool childhood obesity prevention programme: a focused ethnography study**. *BMC Public Health.* (2019) **19** 1074. DOI: 10.1186/s12889-019-7410-0
40. Wolfenden L, Barnes C, Jones J, Finch M, Wyse RJ, Kingsland M. **Strategies to improve the implementation of healthy eating, physical activity and obesity prevention policies, practices or programmes within childcare services**. *Cochrane Database Syst Rev.* (2020) **2** CD011779. DOI: 10.1002/14651858.CD011779.pub3
41. Sanchez-Flack JC, Herman A, Buscemi J, Kong A, Bains A, Fitzgibbon ML. **Systematic review of the implementation of obesity prevention interventions in early childcare and education settings using the RE-AIM framework**. *Transl Behav Med.* (2020) **10** 1168-76. DOI: 10.1093/tbm/ibz179
42. Davis SM, Sanders SG, FitzGerald CA, Keane PC, Canaca GF, Volker-Rector R. **An evidence-based preschool intervention for obesity prevention in Head Start**. *J Sch Health.* (2013) **83** 223-9. DOI: 10.1111/josh.12018
43. Combs KM, Drewelow KM, Habesland MS, Lain MA, Buckley PR. **Does training modality predict fidelity of an evidence-based intervention delivered in schools?**. *Prev Sci.* (2021) **22** 928-38. DOI: 10.1007/s11121-021-01227-6
44. Brownson RC, Kumanyika SK, Kreuter MW, Haire-Joshu D. **Implementation science should give higher priority to health equity**. *Implement Sci.* (2021) **16** 28. DOI: 10.1186/s13012-021-01097-0
45. Guarte JM, Barrios EB. **Estimation under purposive sampling**. *Commun Stat Simul Comput.* (2006) **35** 277-84. DOI: 10.1080/03610910600591610
46. Andrade C. **The inconvenient truth about convenience and purposive samples**. *Indian J Psychol Med.* (2021) **43** 86-8. DOI: 10.1177/0253717620977000
47. Wang X, Cheng Z. **Cross-sectional studies: strengths, weaknesses, and recommendations**. *Chest.* (2020) **158** S65-S71. DOI: 10.1016/j.chest.2020.03.012
|
---
title: 'A Partnership Among Local Public Health Agencies, Elementary Schools, and
a University to Address Childhood Obesity: A Scalable Model of the Assess, Identify,
Make It Happen Process'
authors:
- Benjamin C. Ingman
- Carla Loecke
- Elaine S. Belansky
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012628
doi: 10.3389/frhs.2022.816536
license: CC BY 4.0
---
# A Partnership Among Local Public Health Agencies, Elementary Schools, and a University to Address Childhood Obesity: A Scalable Model of the Assess, Identify, Make It Happen Process
## Abstract
### Background
One pathway to addressing childhood obesity is through implementing evidence-based practices (EBPs) shown to promote nutrition and physical activity in K-12 school settings. Assess, Identify, Make it happen (AIM) is a strategic planning process to engage stakeholders in implementing EBPs in their K-12 schools. Local Public Health Agencies (LPHAs) are a potential partner to facilitate this process to a broader audience of rural school communities.
### Methods
A process and outcome evaluation design was applied in this study to examine the extent to which LPHAs effectively implemented AIM with rural/frontier schools in comparison to university staff. Data collection included post-meeting surveys completed by facilitators, a post-intervention interview with facilitators, a survey of school task force members at the end of the AIM process, and systematic documentation of the intervention.
### Results
Reach—Among the 26 eligible elementary schools, 18 ($69\%$) agreed to participate.
Effect—In total, schools facilitated by LPHAs fully implemented an average of 4.0 changes per school, while schools facilitated by the university staff fully implemented an average of 3.7 changes.
Adoption—Among the five LPHAs in the target region, all five agreed to partner on the initiative, but some agencies were unable to identify sufficient personnel to facilitate all schools in their catchment area.
Implementation—[1] In total, 89 of 94 ($95\%$) meetings scheduled by LPHA facilitators occurred. 47 of 48 ($98\%$) meetings scheduled by the university staff occurred. [ 2] The university staff self-reported $93\%$ of agenda items in the AIM process as “completely” followed while LPHA facilitators reported $41\%$ of agenda items as “completely” followed. [ 3] Task force satisfaction with the AIM process and facilitator showed limited variance across LPHAs and university-facilitated schools.
Maintenance—Of the 16 school districts that agreed to participate in the school-based version of AIM, 9 ($56\%$) also participated in a district-wide version of AIM 2 years later.
### Conclusion
AIM is an effective process for implementing EBPs in elementary schools when facilitated by LPHAs. Effective partnerships, a nuanced approach to fidelity, scalability considerations, and the role of technical assistance and training all contributed to the successful implementation of this LPHA-Elementary school partnership.
## Childhood Obesity, Schools, and the Role of Research Intermediaries
Childhood obesity rates have continued to climb over the last several decades across the United States, with higher rates of obesity among rural youth [1], Latinx youth [2], and youth living in poverty [3]. Schools are situated to address these systemic inequities by promoting nutrition and physical activity [4]. This is especially the case in rural communities, where schools are often considered the hubs of social and cultural activities [5].
The evidence for school-based practices and policies that promote students' physical activity [6], nutrition [7], and mental and behavioral health [8] continues to grow. Despite ongoing concerns about the efficacy of childhood obesity prevention programs [9], there are many practices reflected in the literature that have been shown to increase student opportunities for physical activity and nutrition in schools [10, 11]. Evidence-based practices (EBPs) in K-12 schools that promote nutrition include cafeteria-based practices [e.g., offering healthy beverages and foods [12], placing fruits and vegetables earlier in the line [13], scheduling recess before lunch [14], using an “offer” rather than “serve” system [15]]; as well as practices outside the cafeteria [e.g., healthy food for class parties and rewards [16], regular access to water [17], and school store policies that promote healthy food and drinks [18]]. Increased physical activity in schools is linked to practices for physical education [e.g., using an evidence-based curriculum and equipment [19, 20]], environment [e.g., adequate indoor and outdoor facilities [21]], recess [e.g., not withholding recess as punishment, providing equipment and organized activities during recess [22]] classrooms [e.g., classroom activity breaks [23], standing desks [24]], and extracurricular activities [e.g., providing intramural or interscholastic sports [25]]. However, many schools have not implemented those practices or recommendations [26]. This disconnect between research and practice, routinely documented in the fields of public health and healthcare [27, 28] are also reflected in the implementation status of practices and policies in K-12 schools [29, 30].
Research intermediaries, or organizations that help community-based entities learn about and implement EBPs (among other functions) [31], have made progress in facilitating the connection between research and practice in K-12 schools. In particular, leveraging practices of community engagement to facilitate the translation of EBPs to school environments has shown promise [32, 33]. However, additional strategies are necessary to reach schools in rural, high-poverty settings where resources and research tend to be scarce [34]. One pathway to address these gaps in knowledge and translation is through engaging school stakeholders in a process to implement EBPs in their schools. Such a process can reach more schools if additional organizations and agencies are identified and mobilized as research intermediaries.
## AIM (Assess, Identify, Make It Happen)
Assess, Identify, Make it Happen (AIM) is a strategic planning process to promote healthy nutrition and physical activity in K-12 schools. In this process, a task force of community stakeholders convenes to Assess the current status of evidence-based practices (EBPs) shown to promote healthy nutrition and physical activity, Identify EBPs to put in place, and Make it happen by implementing those EBPs. The 12-month process is facilitated by a trained and certified facilitator.
AIM was tested in rural, elementary schools using a pair-randomized control trial and demonstrated to be an effective strategy for promoting the implementation of effective school-based environment, policy, and practice features previously shown to increase students' physical activity and healthy nutrition [29]. AIM schools made an average of 4.4 evidence-based changes per school with $90\%$ still in place a year later compared to schools that used the CDC's School Health Index which made an average of 0.6 effective changes with $66\%$ in place a year later. This first study demonstrated that AIM is an effective method of promoting the implementation of EBPs when facilitated by university staff working directly with rural communities. While these results bode well for the process itself, relying on university staff to implement AIM poses a challenge to scalability (i.e., relies on university-based personnel and considerable travel expenses). A delivery model in which individuals from rural communities facilitate the process in their own communities would greatly improve the scalability of AIM.
## Local Public Health Agencies
Local Public Health Agencies (LPHAs) were identified as entities well positioned to promote the scalability of AIM. Among the ten essential services of LPHAs are to: Communicate effectively to inform and educate people about health, factors that influence it, and how to improve it; Strengthen, support, and mobilize communities and partnerships to improve health; Create, champion, and implement policies, plans, and laws that impact health [35]. These functions closely align with the purposes of the AIM process. Additionally, LPHAs are physically proximate to target populations, have considerable knowledge of the community, and prioritize addressing childhood obesity. Although LPHAs in rural/frontier regions may face challenges such as a lack of qualified staff, and limited access to training, information, and resources [36, 37], they are also well positioned to leverage local cultural assets and flexible structures for developing new productive partnerships and networks [38]. Further, half of the 2,400 Local Health Departments/Agencies in the USA serve rural populations [39]. This confluence of factors positions LPHAs as a promising pathway to scalability for school- and community-based initiatives. Others have been successful in partnering with LPHAs to implement school-based initiatives [40], although concrete assessments of implementation characteristics in applying such an approach are scant in the literature.
Partnering with LPHAs to facilitate the AIM process required important changes to several elements of the AIM process, facilitator training, and technical assistance [41]. Specifically, this change in implementation model was coupled with the development of an AIM website, the revision of AIM meeting guides and materials, streamlining and automating labor-intensive aspects of the process (e.g., creating an automated survey and report generating system). For these reasons, an implementation science framework was adopted to evaluate not only the outcomes of the intervention, but also to describe key dimensions of implementation across the RE-AIM framework [42]. This work contributes to discourse of implementation science that seeks to understand the effectiveness of interventions when implemented in real-world settings and provides additional perspectives on the factors that influence successful implementation [43].
The purpose of this study was to examine the extent to which LPHAs could effectively facilitate AIM with rural/frontier schools in comparison to university staff. The RE-AIM framework was used for this inquiry because it provides a systematic and comprehensive structure to evaluate interventions as implemented in complex, real-world settings.
## Program Description: AIM Process
The goal of the AIM process is to implement evidence-based practices (EBPs) for promoting student nutrition and physical activity in school settings. For each school participating in AIM, a task force of school stakeholders (including the school principal, classroom teachers, physical education teachers, school staff, food service directors, nurses, and parents) convenes for a series of meetings led by a facilitator trained and certified in the process. The AIM facilitator is provided a facilitator guide, which includes detailed agendas, activities, and talking points for each meeting, as well as tasks to complete between meetings. Before the AIM process begins, baseline data is collected via a three-module survey based on the School Environment and Policy Survey (SEPS) [29]. This survey is completed by the principal, food service director, and physical education teacher and generates a Best Practice Report that provides the status (fully implemented, partially implemented, not implemented) of EBPs for nutrition and physical activity.
After the task force has been recruited and oriented to the process, they discuss strengths and challenges related to student health behaviors and school practices to promote student health. The task force also reviews the Best Practice Report to understand the current implementation status of nutrition and physical activity EBPs in the school and generate a list of potential changes to make to the school. This list of potential changes is later revised and clarified before final selections are made based on the importance of a change for student health and the feasibility of implementing it.
The task force engages in several planning activities to promote the successful implementation of the selected changes. This includes planning to get approval and buy-in for changes, identifying individuals to champion changes, creating a task-oriented timeline for implementing changes, and planning for sustainability. The task force convenes for a final meeting to review progress in implementation, and plan any next steps for the group, such as checking in on implementation or transitioning to a wellness team.
The AIM process was implemented with two separate cohorts and revised between cohort 1 (eight schools) and cohort 2 (10 schools) based on feedback from facilitators and task force members. The most significant revision was the amount of time dedicated to AIM meetings and activities; the number of meetings was reduced from 9 to 7 meetings, and the length of meetings was reduced from 120 to 60–75 min (see Table 1).
**Table 1**
| AIM Process for Cohort 1 (2014–2015) 9 meetings, 120 min each 9 (6) schools | AIM Process for Cohort 1 (2014–2015) 9 meetings, 120 min each 9 (6) schools.1 | AIM Process for Cohort 2 (2015–2016) 7 meetings, 60–75 min each 10 (7) schools | AIM Process for Cohort 2 (2015–2016) 7 meetings, 60–75 min each 10 (7) schools.1 |
| --- | --- | --- | --- |
| Meeting title | Meeting description | Meeting title | Meeting description |
| ASSESS | | ASSESS | |
| 1. Getting started | Introduction to AIM, school snapshot Pt 1: strengths | 1. Looking for opportunities | Identify strengths and opportunities related to healthy eating and physical activity in different parts of the school (e.g., cafeteria., classroom, before/after school) |
| 2. Looking for opportunities | School snapshot Pt 2: opportunities, best practice report, list of possible changes | 2. Investigating best practices | Review best practice report, make a list of possible changes |
| IDENTIFY | | IDENTIFY | |
| 3. Evaluating change possibilities | Rating importance and feasibility | 3. Identifying changes | Rate importance and feasibility, select changes |
| 4. Selecting changes | Review importance and feasibility, select changes | | |
| MAKE IT HAPPEN | | MAKE IT HAPPEN | |
| 5. Planning for approval and buy-in | Create action plans: Focus tasks on getting approval to make changes and building buy-in among stakeholders | 4. Building support for changes | Action Planning: Tasks to get approval and build buy-in for changes |
| 6. Planning for implementation | Create action plans: Focus tasks on nuts and bolts of implementing practices | 5. Planning for implementation | Action Planning: tasks to put changes in place and sustain them over the long term |
| 7. Planning for sustainability | Create action plans: Focus on tasks to sustain changes over time; create timeline for implementing practices | 6. Wrapping up | Create timeline for implementing changes and assign tasks, plan for summer |
| 8. Checking our progress | Assign remaining tasks, plan for summer | 7. Checking in | Check in to document progress and keep things on track |
| 9. Moving forward | Check in the following fall to document progress and keep things on track | | |
## Program Setting
This study took place from 2014 to 2016 in a rural/frontier plains region in Colorado encompassing seven counties and 15,962 square miles (larger than the state of Maryland) that includes the lowest county health rankings and highest childhood poverty rates in the state [44].
## Program Recruitment
Project staff recruited LPHAs and schools through in-person visits at each site during the academic school year preceding the intervention. School recruitment meetings were typically attended by the school principal and physical education teacher. Schools received $4,000 to complete the AIM process. LPHA recruitment meetings were attended by agency directors and staff identified as potential AIM facilitators, who were in most cases nurses. Informational flyers explaining the AIM process and Memorandums of Understanding were key artifacts used during recruitment efforts. LPHAs were remunerated at a rate of $10\%$ FTE of the facilitator per each school facilitated (e.g., one school facilitated through AIM by an LPHA employee earning $50,000 resulted in a $5,000 payment to the LPHA).
Local Public Health Agencies staff also participated in a readiness assessment interview during the recruitment phase, which provided an opportunity to discuss their motivations and reservations to participating. LPHAs noted the shared priority of addressing obesity (all five included obesity in their most recent Health Assessment Plans) and the potential benefits of closely collaborating with schools in their service area.
## Training and Technical Assistance for LPHAs
Local Public Health Agencies directors designated staff to facilitate the AIM process. LPHA staff were trained through a 5-day training in August and a 1-day booster training midway through the school year. Two facilitators who worked with both cohort 1 and cohort 2 attended a 1-day training focused on revisions from the previous year in lieu of attending the 5-day training a second time. Ongoing support to discuss progress and answer questions consisted of monthly conference calls among facilitators and university staff, and individualized ad hoc technical assistance [see [45]].
## Process and Outcome Evaluation Design
This study used a process and outcome evaluation approach to monitor and evaluate the implementation of the AIM process [46]. Process evaluation efforts, which were guided by the RE-AIM framework [42], began with the recruitment of LPHAs and schools and ended 6 months after all participating schools had completed the AIM process. Outcome evaluation was focused on the implementation of evidence-based practices in participating schools and general satisfaction with the AIM process and facilitators. The RE-AIM framework was selected to guide data collection because it attends to various factors of implementing real-world public health interventions (Reach, Effect, Adoption, Implementation, Maintenance; see Table 2). This study was approved by the Colorado Multiple Institution Review Board.
**Table 2**
| RE-AIM Dimensions (42) | Evaluation metrics in this work |
| --- | --- |
| Reach. Proportion of the target population that participated in the intervention | • Number and demographic characteristics of participating school districts in the target region |
| Effect (or Efficacy). Success rate if implemented as in guidelines/protocol | • Number of physical activity and nutrition evidence-based practices fully implemented, partially implemented, planned for implementation, and not implemented |
| Adoption. Proportion of settings that adopt the intervention | • Number and characteristics of LPHAs in the target region implementing AIM |
| Implementation. Extent to which the intervention was implemented as intended | • Number and length of meetings facilitated • Facilitator time spent preparing and feelings of preparedness • Facilitator fidelity to facilitator guide • Extent of idea sharing and tension noted during meetings • Taskforce satisfaction with AIM process and facilitators |
| Maintenance. Extent to which a program is sustained over time | • School district participation in a subsequent version of AIM • Anecdotal continuation of wellness teams |
## Post-meeting Surveys (AIM Facilitators)
All AIM facilitators (LPHA staff and university staff) completed a post-meeting survey at the end of each AIM meeting. These surveys included attention to logistical aspects of the meeting (date, time, and length of the meeting); facilitator preparation; fidelity to the meeting guide; task force dynamics (member participation and tension during the meeting); and feedback about the meeting agenda and process. There was an average of 33 items per post-meeting survey. Implementation status of changes was included in the final AIM meeting survey. These surveys were completed with a $100\%$ response rate.
## Post-intervention Interviews (AIM Facilitators)
All AIM facilitators participated in a semi-structured interview at the end of the intervention. These interviews were held in person at the health agency office or in a community setting and focused on LPHA facilitator perspectives on four topics: [1] facilitation of the AIM process at the school, [2] partnership with the university team, [3] impacts on the agency or its personnel, and [4] suggested improvements to the AIM process.
## Post-process Survey (AIM Task Force Members)
Those participating in the AIM process as members of school task forces completed a 53-item survey at the end of the AIM process. Key topics included in this survey were perceptions of the facilitator and overall satisfaction with the AIM process. In total, 80 task force surveys were completed, representing a $100\%$ response rate for task force members in attendance at the final AIM meetings.
## Process Documentation
Other data, correspondence, meeting notes, and artifacts that document the process were collected throughout the intervention to inform and contextualize dimensions of the intervention as guided by the RE-AIM framework.
## Data Analysis
Evidence-based practices were coded as nutrition or physical activity by the task forces proposing the changes. These practices were then coded to the sub-areas of changes by two researchers. Discrepancies in coding were identified and discussed by raters to determine the final coding. Interviews with LPHAs were transcribed and analyzed using structural, open, and axial coding [47, 48]. Two researchers completed the analysis, with regular meetings to identify inconsistencies and discrepancies in coding and to discuss emergent themes [49]. Project documents and records were analyzed by researchers to ensure the accurate and complete depiction of the intervention as it unfolded.
## Reach
The target region for recruitment included 26 elementary schools. These schools served 4,323 students ($48\%$ Hispanic, $66\%$ Free/reduced lunch). Among these schools, 18 ($69\%$) agreed to participate and LPHA staff facilitated 12. A local individual was hired as university staff to facilitate the remaining six schools (see Figure 1). Schools that participated in the intervention as facilitated by LPHAs had a slightly higher Hispanic population ($49\%$) and slightly lower free and reduced lunch rate ($64\%$) than the target population (see Figure 2).
**Figure 1:** *Demographics of participating and non-participating elementary schools.* **Figure 2:** *Scatterplot of student Free/Reduced Lunch rate and % Hispanic for participating and non-participating schools.*
## Effect
The AIM process is designed to expedite the implementation of evidence-based practices that promote nutrition and physical activity for students at participating schools. The implementation status of identified practices was documented at the final meeting of the AIM process using the following options: fully implemented, partially implemented, planned for implementation, and not implemented.
LPHA cohort 1 had an average of 5.20 changes implemented per school; LPHA cohort 2 had an average of 3.29 changes per school. The university-facilitated schools had an average of 3.67 changes fully implemented per school in both cohorts 1 and 2. In total, schools facilitated by LPHAs saw an average of 4.00 changes fully implemented per school, while schools facilitated by university staff had an average of 3.67 changes fully implemented per school. The results of the type of changes implemented are further delineated in Figure 3.
**Figure 3:** *Number and implementation status of EBPs.*
## Adoption
We attempted to recruit five LPHAs for partnership, and successfully recruited $100\%$ of these LPHAs. In total, these five LPHAs serviced a population of 71,162 across seven counties. While all five LPHAs agreed to partner and implement AIM, two agencies were unable to identify personnel to facilitate all schools in their catchment area. Namely, one agency was able to facilitate just one of the six schools in their region, and another agency was able to facilitate one of the two schools in their region. Both LPHAs cited lack of available qualified personnel as the primary factor that limited their capacity to facilitate AIM in all schools in their regions. Among the 18 schools successfully recruited for participation in the process, the five LPHAs were able to facilitate 12 ($67\%$) of those schools.
## Number and Length of Meetings
In total, 94 meetings were scheduled with the 12 schools facilitated by LPHAs. Among these, 89 ($95\%$) meetings took place. The six schools facilitated by university staff were scheduled for a total of 48 meetings, and 47 ($98\%$) took place.
Meeting lengths varied between cohorts 1 and 2 due to revisions made to the meeting guide based on feedback from cohort 1. There was no difference in mode for the meeting length between LPHA and university facilitators for either cohort (Cohort 1 mode = 1:46–2:00 h; Cohort 2 mode = 1:01–1:15 h). There was, however, a tendency for the university facilitator meetings to run longer than the LPHA facilitators across both cohorts. This was most pronounced during cohort 2 where the university facilitator meetings skewed longer (right) and the LPHA facilitator meetings skewed shorter (left; see Figure 4).
**Figure 4:** *Length of meetings for cohorts 1 and 2.*
### Feeling Prepared
Facilitators indicated how much time they spent preparing for each meeting. The university facilitator reported spending more than 60 min preparing for $77\%$ of meetings while LPHA facilitators reported spending more than 60 min preparing for $50\%$ of meetings (see Figure 5.1). Relatedly, the university facilitator strongly agreed with the statement “I felt very prepared to facilitate this meeting” for $94\%$ of meetings while the LPHA facilitators strongly agreed with that statement for $39\%$ of meetings (see Figure 5.2).
**Figure 5:** *Facilitator ratings. 5.1: How much time did you spend reviewing materials in preparation for this meeting? 5.2: I felt very prepared to facilitate this meeting. 5.3: Indicate how closely you followed the facilitator guide (each agenda item rated). 5.4: Most taskforce members shared their ideas during the meeting. 5.5: There was tension among some of the taskforce members during the meeting.*
## Fidelity to Facilitator Guide
Assess, Identify, Make it happen facilitators rated how closely they followed the facilitator guide for each agenda item of each meeting using the following scale: Not at all (0–$24\%$, did not do this part of the meeting); Some (addressed $25\%$−$49\%$ of the items); Mostly (addressed $50\%$−$74\%$ of the items); Completely (addressed $75\%$−$100\%$ of the items). The university facilitator reported $93\%$ of agenda items as “completely” while LPHA facilitators reported $41\%$ of agenda items as “completely” (see Figure 5.3).
## Idea Sharing and Tension During AIM Meetings
Facilitators also rated the extent to which they agreed or disagreed with two statements: “Most task force members shared their ideas during the meeting” and “There was tension among some of the task force members during the meeting.” The university facilitator strongly agreed that most task force members shared their ideas during the meeting $94\%$ of the time, while the LPHA facilitators strongly agreed with this statement $44\%$ of the time (see Figure 5.4). The university facilitator also strongly disagreed with the statement of tension among task force members $98\%$ of the time, while LPHA facilitators strongly disagreed with this statement $79\%$ of the time (see Figure 5.5).
## Taskforce Satisfaction With the Process and Facilitators
At the end of the AIM process, task force members were invited to participate in a task force survey which included items focused on their satisfaction and interpretations of the AIM process (Figure 6.1) and facilitator (Figures 6.2, 6.3). These results show limited difference between satisfaction with the facilitator, although the LPHA-facilitated schools show slightly higher overall satisfaction with the process.
**Figure 6:** *Taskforce member ratings. 6.1: How satisfied are you with the AIM process? 6.2: How satisfied are you with the facilitator? 6.3: Our AIM Facilitator was from our community.*
## Maintenance
The AIM process and partnerships with LPHAs resulted in several new connections and enduring practices amongst schools and LPHAs. At the close of the initiative, we offered an AIM Do-It-Yourself training and disseminated manuals for applying AIM without the support of a university facilitator. We did not systematically evaluate the uptake of such an approach at schools, however. Other outcomes from the initiative include school districts successfully transitioning AIM task forces into functional wellness teams, and LPHA staff continuing to meet with school district personnel to support them in their wellness efforts. Post-intervention interviews with LPHA staff also expressed optimism on the long-term outcomes for LPHA-school partnerships resulting from this initiative.
Relatedly, a subsequent iteration of AIM was offered 2 years after this initiative was completed in the same region. This version of AIM was altered in focus (from nutrition and physical activity to all components of the Whole School, Whole Community, Whole Child model) [50], scope (from school to district level), and implementation model (from nine, 60–75 min meetings, to three, 6 h meetings facilitated by University staff). Of the 16 school districts that agreed to participate in the initial version of AIM discussed in this study, nine ($56\%$) also participated in this subsequent, extended version of AIM. Further, of the six districts that declined to participate in the initial version of AIM, 4 ($67\%$) agreed to participate in the subsequent, extended version of AIM.
## Discussion
Implementing the AIM process in partnership with LPHAs allowed for a more scalable model of the AIM process to be implemented across a large, rural/frontier geographic region with outcomes comparable to previous iterations of AIM. This study raises a few points of ongoing consideration for those engaged in implementing interventions in partnership with local organizations as research intermediaries.
## Comparisons Between University and LPHA Facilitators of AIM
This study demonstrates that LPHAs succeeded in facilitating schools through the AIM process and that schools were successful in implementing EBPs. This positions AIM as a promising model for broader implementation to make schools in rural/frontier communities healthier places for students. There were, however, differences between LPHA and university facilitators in their facilitation of AIM in this initiative. The LPHA facilitators averaged lower marks than the university facilitator on [1] fidelity to the process, [2] the percentage of meetings that took place vs. those that were planned, and [3] the length and completion rate of meetings. Meetings facilitated by LPHAs also reported greater tension and lower incidence of all task force members sharing their opinions during the meetings when compared to meetings guided by the university facilitator. These differences are at odds with the outcome measures, which showed an average of slightly more evidence-based practices implemented with LPHAs (4.00 EBPs per school) than with the university facilitator (3.67 EBPs per school). These results support previous research that suggests intermediaries may be effective in facilitating the uptake of EBPs through community-engaged approaches [31, 32].
## Considerations of Fidelity
While fidelity is typically positioned as a key determinant to maintaining desirable outcomes of interventions, this study revealed that higher fidelity to the process as prescribed was not associated with an increased prevalence of desired outcomes [51]. From a training and technical assistance perspective, our approach to fidelity was aligned with suggestions that an adaptive approach to fidelity is essential when scaling up programming [52]. In this initiative, facilitators were encouraged to waver from the facilitator guide when they considered it in the best interest of the process and task force. In some instances, facilitators were supported in making more significant alterations to the process as long as critical activities of AIM were retained. Anecdotal evidence from this initiative supports the effectiveness of this adaptive approach to process fidelity. For example, there were instances in this implementation of AIM in which facilitators' high fidelity to the process was viewed as inflexibility to the local context and considered a detriment to quality by task force members. Conversely, approaching the AIM process with flexibility to the needs and contexts of LPHA and school partners was viewed as critical to ensuring the success of the initiative. These findings inspire a continued consideration of fidelity in the context of health-based interventions in partnership with community organizations in school settings [53].
## Importance of Effective Partnerships, Scalability Considerations, Training, and Technical Assistance
This study also emphasizes the benefits of adopting a flexible and supportive approach to partnering with community-based research intermediaries. In retrospect, we view approaches to [1] adapting to local capacity, [2] scalability, and [3] training and technical assistance, as worthy of emphasis.
## Adapting to Local Capacity
Adapting the intervention plan based on the capacity of LPHAs was critical to ensuring success and promoting the greatest reach possible. For instance, although it was not the intended implementation model, we hired a community affiliate to operate as facilitator to account for the lack of available personnel in two LPHAs. Flexibility in implementation with this agency allowed us to still reach the target audience of schools in this region despite a lack of capacity at the LPHA.
## Scalability
The effort to create a scalable model was executed with consideration of key dimensions of scalability [see [41]]. Revisions to the process that better positioned it for success in this scalable model include developing a new training and support model, revamping materials (meeting guide, website, supportive materials), amending the method of implementation (meeting evaluations, school surveys to generate automated reports), and, perhaps most importantly, reducing the amount of time required to complete the process. In the context of rural LPHAs and schools, it is important that initiatives that add to the existing workload honor the time constraints and responsibilities of existing partners and take efforts to promote the greatest efficiency possible. This approach was also more cost-efficient than previous versions of the process [29].
## Training and Technical Assistance
Finally, many LPHA staff reported that the training and technical assistance they received throughout this intervention was both critical in aiding their successful facilitation and dissimilar to much of the training and support they had received in the past. This underscores the importance of attending to training and technical assistance when seeking to expand the reach of a model or intervention. In this case, a training and technical assistance approach that draws on various theories of education, training, and professional development was found to develop the necessary knowledge and skills in facilitators. This contributes to discourse concerning the importance of technical assistance in implementing new interventions and programs [54, 55].
## Conclusions
Implementing AIM with rural LPHAs as facilitators was an effective method of implementing evidence-based practices for physical activity and nutrition in rural elementary schools. The results outlined above support the continued exploration of partnerships with LPHAs as research intermediaries and the promise of further applications of AIM as a catalyst of expediting the research to practice delay.
Future studies may further engage in the question of fidelity in implementation science. Namely, the findings of this study support the importance of discourse that interrogates the notion of fidelity to interventions alongside responsiveness to the context and locality in which an intervention is implemented [51]. Other research may address how partnerships with LPHAs can be leveraged and best structured to address areas of need in rural contexts (e.g., professional development needs, lack of funding, resources, or personnel) and promote positive outcomes to address a compendium of health behaviors and conditions.
## Data Availability Statement
The datasets generated and analyzed during this study are not publicly available due to considerations of confidentiality. De-identified selections of data may be made available from the corresponding author on reasonable request and in compliance with COMIRB.
## Ethics Statement
This study was reviewed and approved by the Colorado Multiple Institution Review Board (COMIRB). Participants provided their written informed consent to participate int his study.
## Author Contributions
BI, CL, and EB all had central roles in designing and implementing the intervention, collecting and analyzing data, and drafting the manuscript. All authors read and approved the final manuscript.
## Funding
This work was made possible with funding from the Colorado Health Foundation (PI: EB, Grant ID: 5733).
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Probst JC, Barker JC, Enders A, Gardiner P. **Current state of child health in rural america: how context shapes children's health**. *J Rural Health.* (2018) **34** s3-12. DOI: 10.1111/jrh.12222
2. Flórez KR, Katic BJ, López-Cevallos DF, Murillo R, Cancel-Tirado D, Aponte-Soto L. **The double burden of food insecurity and obesity among Latino youth: Understanding the role of generational status**. *Pediatr Obes.* (2019) **14** e12525. DOI: 10.1111/ijpo.12525
3. Drewnowski A. **The economics of food choice behavior: why poverty and obesity are linked**. *Nestle Nutr Inst Workshop Ser.* (2012) **73** 95-112. DOI: 10.1159/000341303
4. Story M, Nanney MS, Schwartz MB. **Schools and obesity prevention: creating school environments and policies to promote healthy eating and physical activity**. *Milbank Q.* (2009) **87** 71-100. DOI: 10.1111/j.1468-0009.2009.00548.x
5. Tieken MC. *Why Rural Schools Matter.* (2014)
6. Demetriou Y, Höner O. **Physical activity interventions in the school setting: a systematic review**. *Psychol Sport Exerc.* (2012) **13** 186-96. DOI: 10.1016/j.psychsport.2011.11.006
7. Scherr RE, Linnell JD, Dharmar M, Beccarelli LM, Bergman JJ, Briggs M. **A multicomponent, school-based intervention, the shaping healthy choices program, improves nutrition-related outcomes**. *J Nutr Educ Behav* (2017) **49** 368-79.e1. DOI: 10.1016/j.jneb.2016.12.007
8. Inman DD, van Bakergem KM, LaRosa AC, Garr DR. **Evidence-based health promotion programs for schools and communities**. *Am J Prev Med.* (2011) **40** 207-19. DOI: 10.1016/j.amepre.2010.10.031
9. Kropski JA, Keckley PH, Jensen GL. **School-based obesity prevention programs: an evidence-based review**. *Obesity.* (2008) **16** 1009-18. DOI: 10.1038/oby.2008.29
10. Brennan LK, Brownson RC, Orleans CT. **Childhood obesity policy research and practice: evidence for policy and environmental strategies**. *Am J Prev Med.* (2014) **46** e1-16. DOI: 10.1016/j.amepre.2013.08.022
11. Brown T, Summerbell C. **Systematic review of school-based interventions that focus on changing dietary intake and physical activity levels to prevent childhood obesity: an update to the obesity guidance produced by the National Institute for Health and Clinical Excellence**. *Obes Rev.* (2009) **10** 110-41. DOI: 10.1111/j.1467-789X.2008.00515.x
12. Condon EM, Crepinsek MK, Fox MK. **School meals: Types of foods offered to and consumed by children at lunch and breakfast**. *J Am Diet Assoc* (2009) **109** S67-78. DOI: 10.1016/j.jada.2008.10.062
13. Wansink B, Just D. **Healthy foods first: students take the first lunchroom food 11% more often than the third**. *J Nutr Educ Behav.* (2011) **43** S8. DOI: 10.1016/j.jneb.2011.03.032
14. Getlinger MJ, Laughlin VT, Bell E, Akre C, Arjmandi BH. **Food waste is reduced when elementary-school children have recess before lunch**. *J Am Diet Assoc.* (1996) **96** 906-8. DOI: 10.1016/S0002-8223(96)00245-3
15. Hakim SM, Meissen G. **Increasing consumption of fruits and vegetables in the school cafeteria: the influence of active choice**. *J Health Care Poor Underserved* (2013) **24** 145-57. DOI: 10.1353/hpu.2013.0109
16. Turner L, Chriqui JF, Chaloupka FJ. **Classroom parties in US elementary schools: the potential for policies to reduce student exposure to sugary foods and beverages**. *J Nutr Educ Behav.* (2013) **45** 611-9. DOI: 10.1016/j.jneb.2013.04.261
17. Patel AI. **A trial of the efficacy and cost of water delivery systems in San Francisco Bay Area Middle Schools, 2013**. *Prev Chronic Dis.* (2016) **13** E88. DOI: 10.5888/pcd13.160108
18. Briefel RR, Crepinsek MK, Cabili C, Wilson A, Gleason PM. **School food environments and practices affect dietary behaviors of US public school children**. *J Am Diet Assoc* (2009) **109** S91-107. DOI: 10.1016/j.jada.2008.10.059
19. Barroso CS, McCullum-Gomez C, Hoelscher DM, Kelder SH, Murray NG. **Self-reported barriers to quality physical education by physical education specialists in Texas**. *J Sch Health.* (2005) **75** 313-9. DOI: 10.1111/j.1746-1561.2005.tb07348.x
20. Sallis JF, McKenzie TL, Alcaraz JE, Kolody B, Faucette N, Hovell MF. **The effects of a 2-year physical education program (SPARK) on physical activity and fitness in elementary school students. Sports, play and active recreation for kids**. *Am J Public Health.* (1997) **87** 1328-34. DOI: 10.2105/AJPH.87.8.1328
21. Wechsler H, Devereaux RS, Davis M, Collins J. **Using the school environment to promote physical activity and healthy eating**. *Prev Med.* (2000) **31** S121-37. DOI: 10.1006/pmed.2000.0649
22. Ramstetter CL, Murray R, Garner AS. **The crucial role of recess in schools**. *J Sch Health.* (2010) **80** 517-26. DOI: 10.1111/j.1746-1561.2010.00537.x
23. Kibbe DL, Hackett J, Hurley M, McFarland A, Schubert KG, Schultz A. **Ten Years of TAKE 10!(**. *Prev Med* (2011) **52** S43-50. DOI: 10.1016/j.ypmed.2011.01.025
24. Benden ME, Zhao H, Jeffrey CE, Wendel ML, Blake JJ. **The evaluation of the impact of a stand-biased desk on energy expenditure and physical activity for elementary school students**. *Int J Environ Res Public Health.* (2014) **11** 9361-75. DOI: 10.3390/ijerph110909361
25. 25.Committee Committee on Physical Activity Physical Education in the School Environment Food Nutrition Board and Institute of Medicine. Educating the Student Body: Taking Physical Activity and Physical Education to School (Kohl HW, Cook HD, editors). Washington, DC: National Academies Press (2013). Available online at: http://www.ncbi.nlm.nih.gov/books/NBK201500/ (accessed November 1, 2021).. *Educating the Student Body: Taking Physical Activity and Physical Education to School* (2013)
26. Barnes C, McCrabb S, Stacey F, Nathan N, Yoong SL, Grady A. **Improving implementation of school-based healthy eating and physical activity policies, practices, and programs: a systematic review**. *Transl Behav Med.* (2021) **11** 1365-410. DOI: 10.1093/tbm/ibab037
27. Green LW, Ottoson JM, García C, Hiatt RA, Roditis ML. **Diffusion theory and knowledge dissemination, utilization and integration**. *Front Public Health Ser Syst Res.* (2014) **3** 3. DOI: 10.1146/annurev.publhealth.031308.100049
28. Kessler R, Glasgow RE. **A proposal to speed translation of healthcare research into practice: dramatic change is needed**. *Am J Prev Med.* (2011) **40** 637-44. DOI: 10.1016/j.amepre.2011.02.023
29. Belansky ES, Cutforth N, Chavez R, Crane LA, Waters E, Marshall JA. **Adapted intervention mapping: a strategic planning process for increasing physical activity and healthy eating opportunities in schools via environment and policy change**. *J Sch Health.* (2013) **83** 194-205. PMID: 23343320
30. Wolfenden L, Nathan NK, Sutherland R, Yoong SL, Hodder RK, Wyse RJ. **Strategies for enhancing the implementation of school-based policies or practices targeting risk factors for chronic disease**. *Cochrane Database Syst Rev* (2017) **2017** CD011677. DOI: 10.1002/14651858.CD011677.pub2
31. Franks RP, Bory CT. **Strategies for developing intermediary organizations: considerations for practice**. *Fam Soc.* (2017) **98** 27-34. DOI: 10.1606/1044-3894.2017.6
32. D'Angelo G, Pullmann MD, Lyon AR. **Community engagement strategies for implementation of a policy supporting evidence-based practices: a case study of washington state**. *Adm Policy Ment Health Ment Health Ser Res.* (2017) **44** 6-15. DOI: 10.1007/s10488-015-0664-7
33. Sprague Martinez L, Rapkin BD, Young A, Freisthler B, Glasgow L, Hunt T. **Community engagement to implement evidence-based practices in the HEALing communities study**. *Drug Alcohol Depend.* (2020) **217** 108326. DOI: 10.1016/j.drugalcdep.2020.108326
34. Bettenhausen JL, Winterer CM, Colvin JD. **Health and poverty of rural children: an under-researched and under-resourced vulnerable population**. *Acad Pediatr* (2021) **21** S126-33. DOI: 10.1016/j.acap.2021.08.001
35. 35.CDC – 10 Essential Public Health Services—CSTLTS. (2021). Available online at: https://www.cdc.gov/publichealthgateway/publichealthservices/essentialhealthservices.html (accessed November 1, 2021).. (2021)
36. Rosenblatt RA, Casey S, Richardson M. **Rural-urban differences in the public health workforce: local health departments in 3 rural western states**. *Am J Public Health.* (2002) **92** 1102-5. DOI: 10.2105/AJPH.92.7.1102
37. Turner AM, Stavri Z, Revere D, Altamore R. **From the ground up: Information needs of nurses in a rural public health department in Oregon**. *J Med Libr Assoc.* (2008) **96** 335-42. DOI: 10.3163/1536-5050.96.4.008
38. Rosenstock L, Hernandez LM. **Public health agencies: their roles in educating public health professionals**. *Who Will Keep the Public Healthy? Educating Public Health Professionals For the 21st Century* (2003)
39. Leider JP, Meit M, McCullough JM, Resnick B, Dekker D, Alfonso YN. **The state of rural public health: enduring needs in a new decade**. *Am J Public Health.* (2020) **110** 1283-90. DOI: 10.2105/AJPH.2020.305728
40. Cottrell L, Northrup K, Wittberg R. **Challenges and lessons learned from a prevention research center partnership**. *Am J Prev Med.* (2017) **52** S255-7. DOI: 10.1016/j.amepre.2016.08.016
41. Milat AJ, King L, Bauman AE, Redman S. **The concept of scalability: Increasing the scale and potential adoption of health promotion interventions into policy and practice**. *Health Promot Int.* (2013) **28** 285-98. DOI: 10.1093/heapro/dar097
42. Glasgow RE, Vogt TM, Boles SM. **Evaluating the public health impact of health promotion interventions: the RE-AIM framework**. *Am J Public Health.* (1999) **89** 1322-7. DOI: 10.2105/AJPH.89.9.1322
43. Damschroder L, Hall C, Gillon L, Reardon C, Kelley C, Sparks J. **The consolidated framework for implementation research (CFIR): progress to date, tools and resources, and plans for the future**. *Implement Sci* (2015) **10** A12. DOI: 10.1186/1748-5908-10-S1-A12
44. 44.Robert Wood Johnson Foundation. County health rankings. County Health Rankings Roadmaps (2020). Available online at: https://www.countyhealthrankings.org/reports/state-reports (accessed November 1, 2021).. *County health rankings.* (2020)
45. Meyers DC, Durlak JA, Wandersman A. **The quality implementation framework: A synthesis of critical steps in the implementation process**. *Am J Community Psychol.* (2012) **50** 462-80. DOI: 10.1007/s10464-012-9522-x
46. Handley MA, Gorukanti A, Cattamanchi A. **Strategies for implementing implementation science: a methodological overview**. *Emerg Med J.* (2016) **33** 660-4. DOI: 10.1136/emermed-2015-205461
47. Glaser BG, Strauss AL. *The Discovery of Grounded Theory: Strategies for Qualitative Research* (2009)
48. Tolley EE, Ulin PR, Mack N, Robinson ET, Succop SM. *Qualitative Methods in Public Health: A Field Guide For Applied Research* (2016)
49. Houston SD, Hyndman J, McLean J, Jamal A. **The methods and meanings of collaborative team research**. *Qual Inq* (2010) **16** 285-97. DOI: 10.1177/1077800409346411
50. Lewallen TC, Hunt H, Potts-Datema W, Zaza S, Giles W. **The whole school, whole community, whole child model: a new approach for improving educational attainment and healthy development for students**. *J Sch Health.* (2015) **85** 729-39. DOI: 10.1111/josh.12310
51. Lewis CC, Lyon AR, McBain SA, Landes SJ. **Testing and exploring the limits of traditional notions of fidelity and adaptation in implementation of preventive interventions**. *J Prim Prev.* (2019) **40** 137-41. DOI: 10.1007/s10935-019-00539-2
52. Pérez D, Van der Stuyft P, Zabala MC, Castro M, Lefèvre P. **A modified theoretical framework to assess implementation fidelity of adaptive public health interventions**. *Implement Sci* (2016) **11** 91. DOI: 10.1186/s13012-016-0457-8
53. Carroll C, Patterson M, Wood S, Booth A, Rick J, Balain S. **A conceptual framework for implementation fidelity**. *Implement Sci.* (2007) **2** 40. DOI: 10.1186/1748-5908-2-40
54. Butterfoss FD. **The coalition technical assistance and training framework: Helping community coalitions help themselves**. *Health Promot Pract.* (2004) **5** 118-26. DOI: 10.1177/1524839903257262
55. Chaple M, Sacks S. **The impact of technical assistance and implementation support on program capacity to deliver integrated services**. *J Behav Health Serv Res.* (2016) **43** 3-17. DOI: 10.1007/s11414-014-9419-6
|
---
title: Multi-Criteria Decision Analysis to Prioritize People for COVID-19 Vaccination
When Vaccines Are in Short Supply
authors:
- Hend Chaker Masmoudi
- Amal Rhili
- Imen Zamali
- Ahlem Ben Hmid
- Melika Ben Ahmed
- Myriam Razgallah Khrouf
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012629
doi: 10.3389/frhs.2022.760626
license: CC BY 4.0
---
# Multi-Criteria Decision Analysis to Prioritize People for COVID-19 Vaccination When Vaccines Are in Short Supply
## Abstract
COVID-19 pandemic underscored the need for a rapid tool supporting decision-makers in prioritizing patients in the immediate and overwhelming context of pandemics, where shortages in different healthcare resources are faced. We have proposed Multi-Criteria Decision Analysis (MCDA) to create a system of criteria and weights to prioritize uses of COVID-19 vaccines in groups of people at significantly higher risk of severe COVID-19 disease or death, when vaccines are in short supply, for use in Tunisia. The prioritization criteria and the levels within each criterion were identified based on available COVID-19 evidence with a focus on the criteria selected by Tunisian scientific committees. To determine the weights for the criteria and levels, reflecting their relative importance, a panel of frontline physicians treating COVID-19 were invited to participate in an online survey using 1,000 minds MCDA software (www.1000minds.com) which implements the PAPRIKA (Potentially All Pairwise RanKings of all possible Alternatives) method. Ten criteria and twenty-three levels have been selected for prioritizing the uses of COVID-19 vaccines in groups of people at significantly higher risk of severe disease or death. Among the invited physicians, sixty have completed the survey. The obtained scores were, in decreasing order of importance (mean weights in parentheses, summing to $100\%$). Obesity ($16.2\%$), Age ($12.7\%$), Chronic pulmonary diseases ($10.8\%$), Chronic cardiovascular conditions ($10.3\%$), Bone marrow or organ transplantation ($10.1\%$), Immunodeficiency or Immunosuppression ($9.6\%$), Diabetes ($9\%$), Renal failure ($8.4\%$), evolutive cancer ($6.9\%$), and high blood pressure ($6\%$). MCDA-based prioritization scoring system comprising explicit criteria and weights provides an adaptable and multicriteria approach that can assist policy-makers to prioritize uses of COVID-19 vaccines.
## Introduction
MCDA approach is a rapid and innovative tool to create a “scoring” or “points” system for prioritizing patients for elective health service [1]. In public health systems, an optimal prioritization to serve the most urgent patients first can be needed for different applications. This aims for a transparent, equitable, and accountable allocation of limited resources. For example, MCDA has been used for coronary artery bypass graft in New Zealand [1], or solid organ transplantation among patients waitlisted for transplantation [2].
In the context of pandemics, the increase of demand for different health services and resources underscore the need for a rapid tool supporting decision-makers in planning public health strategies and targeting priority groups. During the COVID-19 pandemic, MCDA has been applied to prioritize COVID-19 patients for hospital [3] and intensive care admissions [4].
In MENA countries, few pilot studies used the MCDA model in the region for healthcare applications, to create a value-based system to assess innovative/biology drugs in Egypt [5] or to purchase generic medicines, in Kuwait [6], and in the UAE [7]. A broader utilization of the MCDA model in the region is considered to increase the consistency and transparency of policy decisions [8].
Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) or COVID-19, originated at Wuhan city of China in early December 2019, has spread across the globe with a profound impact on health and the economy. The high burden of Coronavirus disease 19 (COVID-19) morbidity and mortality has led to a large global effort to develop safe and effective vaccines along with public health measures to contain the pandemic. On December 11, 2020, the FDA issued the authorization for emergency use of the first COVID-19 vaccine [9], followed by the authorization for emergency use of several COVID-19 vaccines in various countries.
The global production capacity of pharmaceutical industries, constraints related to technology licensing, and the high demand for the vaccine, have limited SARS-CoV-2 vaccine supplies worldwide, especially impacting the access of low- and middle-income countries (LMICs) to vaccines. Hence, a huge effort was required to optimize resource deployment in the context of a vaccine shortage.
The WHO Strategic Advisory Group of Experts on Immunization (WHO SAGE) provided a values framework for Allocation and Prioritization of COVID-19 Vaccination and a prioritization roadmap to support countries in planning public health strategies [10]. Groups with comorbidities or health states determined to be at significantly higher risk of severe disease or death are among groups to prioritize for COVID-19 vaccination. Each country needed to adopt and further adapt SAGE guidelines depending on the local context, the size of the target groups, vaccine supply, and the evolving knowledge about COVID-19 and vaccines.
Tunisia, in alignment with SAGE values and the suggested prioritization roadmap, addressed a national strategic vaccination plan. Plan's first stage aimed to protect healthcare professionals as essential workers and, second, to reduce the mortality and morbidity burden by prioritizing the elderly and people with comorbidities. Hence, the first supplies of vaccines received were allocated to healthcare professionals, then to the elderly and people with comorbidities to ensure a prioritization based on utilitarian and egalitarian principles, respectively.
Age and specific pre-existing conditions have been proven to be prominent risk factors for COVID-19 morbidity and mortality (11–14). Authorities and health regulatory agencies, as in France [15], the UK [16, 17], and in the USA [18] enumerated the comorbidities that should be considered in their relative vaccination plans. The relative risk of pre-existing comorbidities to COVID-19 morbidity and mortality is variable and the co-existence of more than one condition increases this risk [19]. The size of the group of people vulnerable to COVID-19 and to prioritize for vaccination may vary significantly between countries, depending on the whole size of the population, the size of the elder population, and the prevalence rate of comorbidities.
Tunisian National Scientific Committees have considered a scoring system providing an adaptable and multicriteria approach of prioritization of higher risk of morbidity and mortality groups. The criteria of prioritization were decided by the National Committees and were set in e-vax, a dedicated national platform for the registration for the whole population willing to be vaccinated (https://www.evax.tn/).
Our research work aims to use the MCDA model to support decision makers in creating a scoring-based system to prioritize vulnerable people for COVID-19 vaccination, and to reduce therefore COVID-19 morbidity and mortality when the vaccine is in shortage.
## Materials and Methods
To create the MCDA system for prioritizing people for COVID-19 vaccination, we followed the guidelines of the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) Task Force for MCDA application in Prioritizing patients' access to healthcare [20, 21].
First, to identify the prioritization criteria and the levels within each criterion, we reviewed the literature, consulted Tunisian decision-makers, and considered also the criteria included in the Evax platform, the national Tunisian platform for COVID-19 vaccination registration. Second, to determine the weights for the criteria and their levels, reflecting their relative importance, a panel of 100 experts and frontline physicians treating COVID-19 in Tunisia were invited to participate in an online survey using 1,000 minds MCDA software (www.1000minds.com) which implements the PAPRIKA (Potentially All Pairwise RanKings of all possible Alternatives) method [22]. This software and method have been used in previous studies to prioritize patients for elective surgery [1], non-critical COVID-19 patients for hospital admission [3, 4], critical patients for intensive care [3], and to also create the WHO's priority list for antibiotic-resistant bacteria to help develop new drugs [23]. The software shows each participant a series of pairs of combinations of the levels on two criteria at the time, representing two hypothetical candidates for vaccination, and asks: “Which individual would you prioritize for vaccination?” ( Figure 1). Each combination involves a trade-off between the two criteria, and the participant's choices reveal how they feel about the relative importance, or ‘weight’, of the two criteria. Each time a participant answers a question, based on all their preceding answers, PAPRIKA adapts with respect to choosing the next question to ask by applying the logical property of ‘transitivity’ – until all possible combinations of the levels on two criteria at the time have been pairwise ranked, either explicitly or implicitly by the participant [For technical details on using PAPRIKA for scoring additive Multi-attribute Value Models please refer to Hansen and Ombler [22]]. Finally, from the participant's explicit pairwise rankings (i.e., answers to the questions) the software uses quantitative methods to derive weights for the levels on each criterion. Obtained Weights for each criterion's level were averaged across all participants to produce mean weights for the group as a whole.
**Figure 1:** *An example of a trade-off question presenting a combination of two levels of two criteria. Age in years, BMI, Body Mass Index.*
Additional questions were included to collect information on the participants' medical specialties, and their affiliation (public or private institution). The software recorded for each participant the number of questions answered and the time spent to complete the survey.
The experts invited to participate in the survey were deliberately selected to be from different regions in Tunisia and from the various medical specialties working in healthcare settings admitting COVID-19 patients, including intensive care anesthetists, emergency physicians, pulmonologists, infectious disease specialists, internists, endocrinologists, cardiologists, and oncologists, with a focus on the first five cited specialties, as more in charge of COVID-19 severe cases. These experts are all frontline physicians treating COVID-19 patients in Tunisia working mainly in the public sector in national institutions for COVID-19 care and are therefore familiar with dealing with many COVID patients from their various fields of expertise.
Stata 16.1 was used to undertake a one-way analysis of variance for normally distributed variables and the Kruskal-Wallis rank test was run for variables not normally distributed, to test the significance of differences in the criteria's mean weights ($p \leq 0.05$). We tested the robustness of our model by assessing the Heterogeneity in preferences (weights) among subgroups or sub-specialties of the participants taking part in the present study [20].
## Results
The board of experts from the COVID-19-Tunisian scientific committees identified 10 criteria for prioritizing vulnerable populations for vaccination, as part of the first goal of the national vaccination program. The criteria, with their levels in parentheses, are: [1] Age (<50, 50–64, 65–75, >75 years); [2] Body Mass Index (BMI <30, BMI 30–40, BMI >40); [3] Diabetes (No, Yes); [4] Chronic pulmonary diseases (No, Yes); [5] Chronic cardiovascular diseases (No, Yes); 6. Renal failure (No, Yes); [7] Bone marrow or Organ Transplantation (No, Yes); [8] Immunodeficiency or Immunosuppression related to treatment or condition (No, Yes); [9] Evolutive Cancer (No, Yes); [10] High Blood pressure (No, Yes).
Sixty physicians completed the survey out of 100 invited participants. The mean number of pairwise-ranking questions answered by each participant was 31, taking 10 min 22 s in total on average. The characteristics of the survey participants are summarized in Table 1.
**Table 1**
| Participants | Number = 60 |
| --- | --- |
| Gender | |
| Male | 25 (41.6%) |
| Female | 35 (58.4%) |
| Sector | |
| Public | 50 (83.3%) |
| Private | 10 (16.7%) |
| Specialties | |
| Infectious disease specialists and internists | 21 (35%) |
| Intensive care anesthesists | 13 (21.7%) |
| Pulmonologists | 9 (15%) |
| Emergency physicians | 4 (6.7%) |
| Endocrinologists | 5 (8.3%) |
| Oncologists | 4 (6.7%) |
| Cardiologists | 2 (3.3%) |
| Nerurologists | 2 (3.3%) |
Mean weights of the criteria and their levels are reported in Figure 2. The scores of Age and comorbidities in prioritizing COVID-19 candidates to vaccination, as revealed from the experts' answers and capture of preferences were, Obesity ($16.2\%$), Age ($12.7\%$), Chronic pulmonary diseases ($10.8\%$), Chronic cardiovascular conditions ($10.3\%$), Bone marrow or organ transplantation ($10.1\%$), Immunodeficiency or Immunosuppression ($9.6\%$), Diabetes ($9\%$), Renal failure ($8.4\%$), evolutive cancer ($6.9\%$), and high blood pressure ($6\%$). The weight for the highest-ranked level on a criterion represents the criterion's overall weight (relative to the other criteria, with these weights summing to $100\%$). Each criterion's lowest level has a value of zero. For the two criteria with more than two levels – age and obesity – the weight for their middle levels is relative to the lowest- and highest-ranked levels, respectively. The relative importance of any pair of criteria, as illustrated in Figure 3, was obtained by dividing the preference value of the left criterion by the preference value of the top criterion. For example, “obesity” was 1.8 times more important than “diabetes” (16.2 vs. $9\%$), whereas “chronic cardiovascular diseases,” “chronic pulmonary diseases” and “bone marrow or organ transplantation” were almost equally important (their overall weights were very close). As determined by each criterion's overall weight, Figure 3 highlights the high importance placed by the panel on “obesity” and “age,” as, compared to any other criteria, their relative-importance ratio is constantly >1, reaching 1.8–2.7 when compared to “evolutive cancer” and “hypertension,” the least important prioritization criteria, according to the experts, on average. Analysis of variance showed statistically significant differences in the mean scores between several criteria ($p \leq 0.05$).
**Figure 2:** *Criteria weights (means) for prioritizing people for COVID-19 vaccination. The bolded values sum to 1 (100%), where the preference values are the criterion weights multiplied by the scores and the bar graph shows the relative importance of every level of each criterion. BMI, Body Mass Index.* **Figure 3:** *Relative importance of the criteria. Based on the mean preference values, each number in the table is a ratio corresponding to the importance of the criterion on the left relative to the criterion at the top. The ratios are obtained by dividing the left preference values by the top preference values.*
We simulated a ranking of 11 hypothetical candidates for vaccination, distinguished by their ratings on the criteria. The ranking is based on the total score obtained by summing their weights (Table 2).
**Table 2**
| Age | Diabetes | Obesity | Immune deficiency or Immunosuppression | Renal Failure | Chronic pulmonary diseases | Chronic cardiovascular diseases | High Blood pressure (hypertension) | Evolutive cancer | Bone marrow or Organ Transplantation |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| >75 | Yes | BMI 30–40 | Yes | No | No | No | No | No | No |
| >75 | No | BMI 30–40 | No | No | No | Yes | No | No | No |
| 50–64 | No | BMI 30–40 | No | No | Yes | No | No | Yes | No |
| 50–64 | No | BMI >40 | No | No | No | No | Yes | No | No |
| 50–64 | No | BMI >40 | No | No | No | No | Yes | No | No |
| 65–74 | Yes | BMI 30–40 | No | No | No | No | No | No | No |
| 65–74 | Yes | BMI <30 | No | Yes | No | No | No | No | No |
| 65–74 | No | BMI 30–40 | No | No | No | No | Yes | No | No |
| 18–49 | No | BMI >40 | No | No | No | No | Yes | No | No |
| 65–74 | No | BMI 30–40 | No | No | No | No | No | No | No |
| 18–49 | No | BMI 30–40 | No | No | No | No | No | Yes | No |
We tested the robustness of our model by assessing the Heterogeneity in preferences (weights) among subgroups of the participants taking part in the present study [20]. Results are summarized in Table 3.
**Table 3**
| Unnamed: 0 | Obesity: BMI >40 | Age: >75 ys | Chronic cardiovascular disease | Diabetes | Chronic pulmoray disease |
| --- | --- | --- | --- | --- | --- |
| Infectious disease specialists and internists | 14.69% | 12.52% | 10.14% | 9.18% | 11.77% |
| Pulmonologists | 16.71% | 12.69% | 9.80% | 9.04% | 8.64% |
| Intensive care anesthesists | 18.88% | 14.68% | 9.29% | 8.52% | 10.73% |
| Emergency physicians | 16.06% | 10.14% | 9.29% | 11.17% | 9.26% |
| Endocrinologists | 16.32% | 14.52% | 14.59% | 7.84% | 15.26% |
| Oncologists | 12.00% | 16.25% | 10.37% | 7.62% | 7.80% |
| Mean /SD | 15.78/2.3 | 13.47/2.14 | 10.55/2.01 | 8.89/1.27 | 10.58/2.7 |
## Discussion
COVID-19 pandemic underscores the need for a rapid tool to establish a scoring system for the optimal allocation of available resources. In the context of the limited supply of COVID-19 vaccines, prioritizing candidates that the most vulnerable people receive the vaccine first was one of the main goals of the COVID-19 vaccination program in Tunisia. A scoring system is an adaptable approach for phased reception of anti-COVID-19 vaccines and provides a multi-criteria approach to prioritize candidates for vaccination and ensure high-risk individuals get immunized first.
Besides a person's age, pre-existing conditions predispose people infected with COVID-19 to an unfavorable clinical course and increased risk of intubation and death. Of Tunisia's population of 11.94 million, 2.8 million have at least one underlying condition and 710,000 have at least two comorbidities (http://ghdx.healthdata.org/). Decision-makers in Tunisia declared the main criteria to consider for prioritization: age, diabetes, cardiovascular diseases, chronic pulmonary diseases, kidney failure, immunodeficiency or immunosuppression, transplantation, obesity, evolutive or recent cancer, and hypertension.
These criteria are inclusive of the main factors and comorbidities predisposing to COVID-19 morbidity and mortality that different studies, organizations, and authorities reported as factor risks for severe COVID-19 or mortality [15, 17, 18]. As selected by a dedicated Tunisian committee of scientists and decision-makers, these criteria reflect the epidemiological context of Tunisia, and also using them in our study allows us to create a scoring system consistent with the national strategy.
In our study, “obesity” and “age” were found to be the most important criteria for determining people's priority for vaccination. Surprisingly, obesity (BMI >40) has been found as the highest weighted criteria in our score-system, higher the weight associated with age (age >75 years old). It is worth noticing that an Italian panel of 100 physicians weighted obesity (BMI >40) higher than age, and other comorbidities factors when weighting criteria for prioritizing hospital admission of patients affected by COVID-19 in the context of a shortage of hospital beds [3].
In the past, obesity has been strongly correlated with mortality from viral infections such as H1N1 influenza and the previous SARS and MERS coronaviruses causing widespread infections [24]. Different studies have reported obesity as a strong predictor of COVID-19 severity. A case-control study in Mexico including 32,583 patients (12,304 cases and 20,279 controls), presenting only one co-morbidity, to determine the independent effect of each comorbidity on Covid-19 susceptibility, found obesity as the strongest predictor factor [25]. Interestingly, a study found a significant positive linear association between increasing BMI and admission to intensive care units (ICU) due to COVID-19 with a significantly higher risk for every BMI-unit increase [26]. Another study from New York showed that age >65 years and obesity are on two most important predisposing factors leading to hospital admission and critical COVID-19 illness – more than hypertension, diabetes, or cardiovascular diseases [27]. Recently, the Center for Disease Control and Prevention (CDC) has issued an updated list of underlying medical conditions associated with higher risk for severe COVID-19 [18]. Besides elder age, considered a key factor in the proposed clinical severity risk score, obesity and diabetes with complication had the highest COVID-19 death risk ratio of 1.3 among comorbid conditions, followed by chronic kidney disease, chronic obstructive pulmonary disease, and neurocognitive disorders (Death risk ratio of 1, 2). The existence of multiple comorbidities was reported as increasing the risk of COVID-19 mortality. Noteworthy that obesity was not mentioned in preliminary reports in China among the most common comorbidities predisposing for COVID-19 infection and COVID-19 disease severity, which was later attributed to the lower rates of obesity seen in far-east cultures [28]. This underscores the need for consideration of the epidemiological state of one country for effective planning and the establishment of scoring systems.
Considering the criteria included in the prioritization system developed in this study, “heart condition/insufficiency,” “respiratory insufficiency,” “bone marrow or organ transplantation”, had scores between 10.8 and 10.1 %, followed by “immunodeficiency,” “diabetes,” and “renal failure” (9.6, 9, and $8.4\%$, respectively). “ High blood pressure” and “evolutive cancer” were the least important criteria for the experts on average. When comparing these found scores to the risk ratio relative to each comorbidity in COVID-19 severity or fatality as reported in cohort studies, we encountered two limiting factors. First, the comorbidity to assess is not preponderant in the corresponding population (as obesity in China). Second, patients may present multiple comorbidities that may be correlated (exp: diabetes and obesity) resulting in multicollinearity in regression analysis. Within QCOVID, a coronavirus risk prediction model used to support the NHS coronavirus response in England, the risk ratios of hospital admission associated with the following factors, chronic kidney stage 4, solid organ transplant, and type 2 diabetes, were, respectively, 1.34, 1.79, 1.57, and 2.64 (Adjusted Hazard ratios for body mass index (BMI) and age). There was a high variability of the ratios found for different stages of kidney failures or with the type of diabetes [29].
Including more levels and more precision in the severity or types of disease in our study, as adding different stages of renal failure, types of diabetes or complicated diabetes, the severity of hypertension, would have resulted in increased granularity in the scoring of these comorbidities and could be considered as a limitation of our study.
To the best of our knowledge, this is the first time MCDA was applied to prioritize candidates for COVID-19 vaccination. This approach, by capturing and representing the “experts” preferences, is a rapid and innovative way to support decision-making protocols especially in the context of a lack of guidelines and limited available resources.
## Data Availability Statement
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author/s.
## Ethics Statement
Ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.
## Author Contributions
HC, MA, and MK conceived the study. HC and AR worked on the design of the survey, processed MCDA model steps and used 1,000 minds software and STATA. HC, MA, IZ, AH, and MK reviewed the literature and the national committees to decide the final criteria to include and score in the study. HC, AR, and IZ drafted the manuscript. MA and MK supervised the research steps. All authors have read and agreed to the published version of the manuscript.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Hansen P, Hendry A, Naden R, Ombler F, Stewart R. **A new process for creating points systems for prioritising patients for elective health services**. *Clin Govern: An Int J.* (2012.0) **17** 200-9. DOI: 10.1108/14777271211251318
2. Kamath PS, Kim WR. **The model for end-stage liver disease (MELD)**. *Hepatology.* (2007.0) **45** 797-805. DOI: 10.1002/hep.21563
3. De Nardo P, Gentilotti E, Mazzaferri F, Cremonini E, Hansen P, Goossens H. **Multi-criteria decision analysis to prioritize hospital admission of patients affected by COVID-19 in low-resource settings with hospital-bed shortage**. *Int J Infect Dis.* (2020.0) **98** 494-500. DOI: 10.1016/j.ijid.2020.06.082
4. Roy M, Hansen P, Sullivan T, Ombler F, Kiore M, Stapleton A. **Rapid development of a tool for prioritizing patients with coronavirus disease 2019 for intensive care**. *Crit Care Explor* (2021.0) **3** 368. DOI: 10.1097/CCE.0000000000000368
5. Fouad S, ElMordy B. **Developing a multi-criteria framework for innovation and biologic drugs assessment decision making in Egypt**. *Value in Health.* (2017.0) **20** A702. DOI: 10.1016/j.jval.2017.08.1826
6. Abdullah AH, Holtorf AP, Al-Hussaini M, Lemay J, Alowayesh M, Kalo Z. **Stakeholder driven development of a multi-criteria decision analysis tool for purchasing off-patent pharmaceuticals in Kuwait**. *J Pharm Policy Pract.* (2019.0) **12** 9. DOI: 10.1186/s40545-019-0171-4
7. Farghaly MN, Al Dallal SAM, Fasseeh AN, Monsef NA, Suliman EAMA, Tahoun MA. **Recommendation for a pilot MCDA tool to support the value-based purchasing of generic medicines in the UAE**. *Front Pharmacol* (2021.0) **12** 737. DOI: 10.3389/fphar.2021.680737
8. Fasseeh A, Karam R, Jameleddine M, George M, Kristensen FB, Al-Rabayah AA. **Implementation of health technology assessment in the Middle East and North Africa: comparison between the current and preferred status**. *Front Pharmacol.* (2020.0) **11** 15. DOI: 10.3389/fphar.2020.00015
9. **ACIP Evidence to Recommendations for Use of Pfizer-BioNTech COVID-19 Vaccine 2021**
10. **WHO SAGE Roadmap For Prioritizing Uses of COVID-19 Vaccines In The Context of Limited Supply 2021**
11. Guan WJ, Liang WH, Zhao Y, Liang HR, Chen ZS, Li YM. **Comorbidity and its impact on 1590 patients with COVID-19 in China: a nationwide analysis**. *Eur Respir J* (2020.0) **55** 2000547. DOI: 10.1183/13993003.00547-2020
12. Gallo Marin B, Aghagoli G, Lavine K, Yang L, Siff EJ, Chiang SS. **Predictors of COVID−19 severity: a literature review**. *Rev Med Virol.* (2021.0) **31** 1-10. DOI: 10.1002/rmv.2146
13. Gold MS, Sehayek D, Gabrielli S, Zhang X, McCusker C, Ben-Shoshan M. **COVID-19 and comorbidities: a systematic review and meta-analysis**. *Postgrad Med.* (2020.0) **132** 749-55. DOI: 10.1080/00325481.2020.1786964
14. Ssentongo P, Ssentongo AE, Heilbrunn ES, Ba DM, Chinchilli VM. **Association of cardiovascular disease and 10 other pre-existing comorbidities with COVID-19 mortality: a systematic review and meta-analysis**. *PLoS ONE.* (2020.0) **15** e0238215. DOI: 10.1371/journal.pone.0238215
15. **Covid-19: quels leviers pour vacciner plus vite les personnes les plus vulnérables?**
16. **Briefing – Vaccine Prioritisation Feb 10, 2021**
17. **Guidance on shielding and protecting people who are clinically extremely vulnerable from COVID-19 2021, April 1**
18. **Underlying medical conditions associated with high risk for severe COVID-19: information for healthcare providers 2021, March 29**
19. Yang J, Zheng W, Shi H, Yan X, Dong K, You Q. **Who should be prioritized for COVID-19 vaccination in China? A descriptive study**. *BMC Med.* (2021.0) **19** 45. DOI: 10.1186/s12916-021-01923-8
20. Thokala P, Devlin N, Marsh K, Baltussen R, Boysen M, Kalo Z. **Multiple criteria decision analysis for health care decision making–an introduction: report 1 of the ISPOR MCDA emerging good practices task force**. *Value Health.* (2016.0) **19** 1-13. DOI: 10.1016/j.jval.2015.12.003
21. Marsh K. M IJ, Thokala P, Baltussen R, Boysen M, Kalo Z. **Multiple criteria decision analysis for health care decision making–emerging good practices: report 2 of the ISPOR MCDA emerging good practices task force**. *Value Health.* (2016.0) **19** 125-37. DOI: 10.1016/j.jval.2015.12.016
22. Hansen P, Ombler F. **A new method for scoring additive multi-attribute value models using pairwise rankings of alternatives**. *J Multi-Cri Dec Anal.* (2008.0) **15** 87-107. DOI: 10.1002/mcda.428
23. Rello J, Kalwaje Eshwara V, Lagunes L, Alves J, Wunderink RG, Conway-Morris A. **A global priority list of the TOp TEn resistant Microorganisms (TOTEM) study at intensive care: a prioritization exercise based on multi-criteria decision analysis**. *Europ J Clin Microbiol Inf Dis.* (2019.0) **38** 319-23. DOI: 10.1007/s10096-018-3428-y
24. Petrakis D, Margina D, Tsarouhas K, Tekos F, Stan M, Nikitovic D. **Obesity a risk factor for increased COVID19 prevalence, severity and lethality (Review)**. *Mol Med Rep.* (2020.0) **22** 9-19. DOI: 10.3892/mmr.2020.11127
25. Hernandez-Garduno E. **Obesity is the comorbidity more strongly associated for Covid-19 in Mexico. A case-control study**. *Obes Res Clin Pract.* (2020.0) **14** 375-9. DOI: 10.1016/j.orcp.2020.06.001
26. Gao M, Piernas C, Astbury NM, Hippisley-Cox J, O'Rahilly S, Aveyard P. **Associations between body-mass index and COVID-19 severity in 6·9 million people in England: a prospective, community-based, cohort study**. *Lancet Diab Endocrinol.* (2021.0) **9** 350-9. DOI: 10.1016/S2213-8587(21)00089-9
27. Petrilli CM, Jones SA, Yang J, Rajagopalan H, O'Donnell L, Chernyak Y. **Factors associated with hospitalization and critical illness among 4,103 patients with Covid-19 disease in New York City**. (2020.0). DOI: 10.1101/2020.04.08.20057794
28. Akbas F, Atmaca HU. **Obesity and COVID-19: Time to take action**. *Obes Facts* (2020.0) **13** 608-10. DOI: 10.1159/000511446
29. Clift AK, Coupland CAC, Keogh RH, Diaz-Ordaz K, Williamson E, Harrison EM. **Living risk prediction algorithm (QCOVID) for risk of hospital admission and mortality from coronavirus 19 in adults: national derivation and validation cohort study**. *BMJ* (2020.0) **371** m3731. DOI: 10.1136/bmj.m3731
|
---
title: 'Evaluation of a Large-Scale School Wellness Intervention Through the Consolidated
Framework for Implementation Research (CFIR): Implications for Dissemination and
Sustainability'
authors:
- Gabriella M. McLoughlin
- Rachel Sweeney
- Laura Liechty
- Joey A. Lee
- Richard R. Rosenkranz
- Gregory J. Welk
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012642
doi: 10.3389/frhs.2022.881639
license: CC BY 4.0
---
# Evaluation of a Large-Scale School Wellness Intervention Through the Consolidated Framework for Implementation Research (CFIR): Implications for Dissemination and Sustainability
## Abstract
### Background
Numerous studies have tested school-based interventions promoting healthy behaviors in youth, but few have integrated dissemination and implementation (D&I) frameworks. Using D&I frameworks can inform if and how an evidence-based intervention is implemented and maintained and provide strategies to address contextual barriers. Such application is necessary to understand how and why interventions are sustained over time. We evaluated a school wellness initiative called SWITCH® (School Wellness Integration Targeting Child Health) to [1] assess implementation outcomes of adoption, fidelity, and penetration, [2] discern implementation determinants through the Consolidated Framework for Implementation Research (CFIR), and [3] examine differences among inexperienced and experienced schools and influential factors to sustainment.
### Methods
A total of 52 schools from Iowa, United States enrolled in the 2019–2020 iteration of SWITCH (22 inexperienced; 30 experienced). The CFIR guided the adaptation of mixed methods data collection and analysis protocols for school settings. Specific attention was focused on [1] fidelity to core elements; [2] adoption of best practices; and [3] penetration of behavior change practices. Determinants were investigated through in-depth qualitative interviews and readiness surveys with implementation leaders. A systematic process was used to score CFIR domains (between −2 and +2) indicating positive or negative influence. Independent t-tests were conducted to capture differences between samples, followed by a cross-case analysis to compare determinants data. Inductive coding yielded themes related to sustainment of SWITCH beyond formal implementation support.
### Results
Experienced schools had higher scores on fidelity/compliance (t = −1.86 $$p \leq 0.07$$) and adoption (t = −2.03 $$p \leq 0.04$$). CFIR determinants of innovation source, culture, relative priority, and leadership engagement were positive implementation determinants, whereas tension for change and networks and communications were negative determinants. Distinguishing factors between experienced and inexperienced schools were Readiness for Implementation and Self-efficacy (experienced significantly higher; $p \leq 0.05$). Strategies to enhance sustainability were increasing student awareness/advocacy, keeping it simple, and integrating into school culture.
### Conclusions
Findings provide specific insights related to SWITCH implementation and sustainability but more generalized insights about the type of support needed to help schools implement and sustain school wellness programming. Tailoring implementation support to both inexperienced and experienced settings will ultimately enhance dissemination and sustainability of evidence-based interventions.
## Introduction
School-based health promotion interventions have been shown to have a positive impact on promoting student physical activity and nutrition behaviors (1–5); however, systematic application of dissemination and implementation science (D&I) frameworks are needed to advance the gap between research and practice [6, 7]. Furthermore, despite the promise of comprehensive programs, limited research exists to illustrate steps to sustain programs over time [8, 9]. Particular emphasis is needed to evaluate strategies aimed at building capacity in school systems since programming is a shared responsibility. Without guidance on how to sustain interventions, school leaders are likely to abandon programming over time, leading to diminished impacts on children's health and well-being.
The present paper reports on the capacity-building process employed in a school wellness initiative called SWITCH® (School Wellness Integration Targeting Child Health). The initiative was built on the foundation of an evidence-based obesity prevention program called Switch that worked through schools to help students “switch what they do, view, and chew” (10–12). Through a United States Department of Agriculture (USDA) grant, emphasis shifted to building capacity in schools to independently coordinate and sustain school wellness programming based on Switch. Formalized D&I strategies were critical in facilitating the transition from an evidence-based program (i.e., Switch) to an evidence-based process (i.e., SWITCH) for sustaining health promotion in schools. Schools self-enroll in a cyclical training (Fall) and implementation (Spring) process which prepares them to develop a comprehensive approach to student health promotion (physical activity, screen time, nutrition behaviors). The process is aimed at helping schools to meet mandates such as the USDA final rule, which tasks schools with developing and evaluating school wellness programs and policies [13, 14].
Foundational research by our team documented the feasibility of training school leaders [15], the acceptability of educational modules for classroom, physical education, and lunchroom settings (16–18) and the validity of school readiness and wellness environment assessment tools [19, 20]. Subsequent studies evaluated alternative implementation strategies [21], the levels of engagement by 4-H leaders (county-level Extension officers who facilitate local-level implementation) assisting in programming [22] and the factors that influenced implementation and scale-up [13]. This most recent evaluation focused on capacity-building and highlighted changes in organizational readiness, reflecting prior literature and warranting its inclusion in subsequent evaluation (23–26). Guided by D&I principles, SWITCH programming has transitioned to be fully managed and coordinated by leaders within the state 4-H network who lead local-level programs and initiatives (https://www.iowaswitch.org/). The established infrastructure provides an ideal model to understand the factors influencing implementation and sustainability of school wellness programming.
The Consolidated Framework for Implementation Research (CFIR) [27, 28], referred to as a determinants framework in the D&I literature, offers specific advantages for a more comprehensive analyses of factors influencing implementation of SWITCH. Specifically, CFIR comprises 39 constructs housed within six key domains: Intervention Characteristics (factors within the intervention itself such as cost and complexity); Outer Setting (factors external to the implementation setting such as policy); Inner Setting (factors within organization such as networks, culture); Readiness for Implementation (organizational and individual capacity for implementation); Characteristics of Individuals (implementation leaders' confidence and motivations to implement); and Implementation Process (practices that facilitate implementation such as planning and executing). Such framework has been used predominantly in healthcare settings to investigate determinants of implementation (28–31), with growing application to school and community settings [32, 33]. The CFIR website (www.CFIRguide.org) provides comprehensive resources for researchers conducting qualitative and mixed methods evaluation to ground their analysis through systematic coding of interview/qualitative data to facilitate interpretation [31]. The CFIR constructs guided several recent mixed method studies on the 2018–2019 iteration of SWITCH [13]; however, it was not possible to fully integrate the interview and implementation outcome data and this hindered our ability to understand determinants that linked to specific implementation outcomes.
The present study on the 2019–2020 iteration of SWITCH employs an integrated mixed methods analysis, based on CFIR coding methods [23], to better understand the factors that influence implementation and sustainability of school wellness programming. The CFIR methodology has documented utility for clinical research [23, 28, 31], but this is one of the first systematic applications of CFIR mixed methods analysis methods for evaluating programming in community / school settings. The study builds directly on our past work [13] by seeking to understand the factors that explain variability in implementation effectiveness between experienced and inexperienced schools. Readiness for implementation has been identified as a barrier to sustaining evidence-based interventions in schools [9]; but few studies have directly examined the relationships between implementation determinants (such as readiness) and outcomes in school-based health promotion research [34, 35]. Addressing this gap was the main goal of the 2019–2020 iteration of SWITCH. Accordingly, this study had three primary aims: Results of this study will provide critical information which may help inform implementation strategies for scale-up and sustainability in school-based interventions.
## Materials and Methods
A mixed methods implementation study grounded in the CFIR was conducted to evaluate key outcomes, determinants, and nuanced relationships between these factors among new and experienced schools in the 2019–2020 cycle of SWITCH. Evaluation approaches followed recommended data collection and analytic methodologies of CFIR, developed by Damschroder and colleagues [27, 31]. To our knowledge, this is one of the first documented adaptations of the CFIR mixed methods protocols with the goal of understanding relationships between implementation determinants and outcomes within a school health promotion context.
## Participants and Procedures
A total of 52 schools enrolled in the 2019–2020 iteration of SWITCH (30 had prior experience and 22 had no previous exposure). Demographic information for these schools is shown in Table 1. The cyclical training (fall) and implementation (spring) process of SWITCH across the academic year facilitates a continuous quality improvement process [36], whereby feedback from schools and implementation outcome data drive modifications to the program each year. More information about the training process can be found in Additional File 1, our previously published article [13], and the program website (https://www.iowaswitch.org/). Briefly, schools were asked to form a wellness team which comprised three members of staff across different school settings (e.g., classroom teachers, physical education, food service, other teachers, administration, counselors, nurses, etc.) and to register prior to the beginning of the academic year. Following registration, schools were asked to attend a total of four webinars and an in-person conference during the fall semester, as well as complete several pre-program audit tools. The implementation phase spanned a 12-week period from January–April of 2020, but due to the coronavirus (COVID-19) outbreak, schools were forced to close in Iowa on March 13th thus forcing a transition to virtual communications/implementation after week 8 of the program. It was not possible to capture final outcome data, but schools completed the midpoint evaluation of school implementation. Below we outline data sources for implementation outcomes and determinants, and the steps taken to rigorously analyze these data.
**Table 1**
| Unnamed: 0 | Free/reduced meals (%) | Free/reduced meals (%).1 | Racial/ethnic minority (%) | Racial/ethnic minority (%).1 | Enrollment | Enrollment.1 | Experience (years) | Experience (years).1 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| | Mean | SD | Mean | SD | Mean | SD | Mean | SD |
| Total (n = 52) | 49.1 | 19.0 | 15.4 | 18.6 | 226.3 | 180.1 | 1.8 | 0.8 |
| Inexperienced (n = 22) | 50.7 | 21.9 | 20.0 | 21.5 | 224.6 | 200.5 | | |
| Experienced (n = 30) | 48.0 | 17.2 | 12.1 | 15.9 | 227.5 | 168.0 | 2.47 | 0.57 |
## Measurement of Implementation Outcomes: Adoption, Fidelity, and Penetration
The field of D&I offers many frameworks and theories to help researchers and practitioners discern why evidence-based practices are or are not implemented in routine care. Regarding implementation outcomes frameworks, the framework by Proctor and colleagues [37] conceptualized several distinctive outcomes that are important to include within implementation evaluations: [1] acceptability (the degree to which an innovation is a perceived good fit); [2] adoption (intent to implement); [3] appropriateness (degree of compatibility within setting); [4] cost (to implement, value for money); [5] feasibility (possibility of successful implementation); [6] fidelity/compliance (executed as intended); [7] penetration (reach within setting); and 8) sustainability (long-term impact). For the purpose of this study, we chose to examine the determinants of adoption, fidelity, and penetration among schools enrolled in SWITCH due to the heavily integrated implementation practices needed to create systems change in the school setting.
Adoption is operationalized by Proctor and colleagues [37] as “intention, initial decision, or action to try or employ an innovation or evidence-based practice” (p. 69). Thus, we measured adoption through implementation surveys at the 6-week mark, examining uptake of best practices in various settings (use of curricular modules, posters, reinforced themes through discussion and tracking). Each best practice was scored as 0 (not at all implemented), 2 (somewhat implemented), and 3 (fully implemented) and a summed score was generated based on the average of each component, to give possible range of 0–9.
Fidelity relates to “the degree to which an intervention was implemented as it was prescribed in the original protocol or as it was intended by the program developers” (p. 69) [37]. The quality elements of SWITCH comprise; wellness team meeting (ideally at least once per week); using SWITCH website to promote student behavior tracking; engaging parents and other stakeholders; and integration of SWITCH modules/posters across the school setting. Fidelity therefore was calculated by using a summed score of quality elements which were scored the same way as best practices, giving a possible range of 0–12.
Finally, penetration is defined as the “integration of a practice within a service setting and its subsystems” (p.70) [37]. This was calculated by determining the number of participants who used or interacted with an evidence-based practice, divided by the total number of participants eligible or within the sample. Since the behavioral tracking and goal setting interface is an integral component for students [38], it provides a good indicator of how many students are actively engaged in SWITCH within each school, thus providing data on penetration. We used data from SWITCH behavior tracking across weeks 1–8 (to account for COVID-19-related school closures). These data are presented as a decimal score (range 0–1.0, translated to 0–$100\%$).
## Organizational Readiness
The School Wellness Readiness Assessment (SWRA) tool [20] was used to assess baseline readiness for implementation. Developed in line with the theory of organizational readiness for change [26, 39] and community capacity-building frameworks [40], the SWRA captures the unique, complex structure and specific settings within schools that impact student health, including classrooms, physical education, and lunchroom settings, and the broader school leadership and cultural context.
The SWRA includes questions across four subscales designed to assess setting-specific and school-wide wellness readiness: classroom readiness, physical education (PE) readiness, food services readiness, and school readiness. The SWRA items were assessed using a 5-point scale (strongly disagree, disagree, neither agree nor disagree, agree, and strongly agree scale, coded as 0, 1, 2, 3, and 4, respectively). A copy of the SWRA is provided in Additional File 2. Wellness teams completed the 40-item SWRA through the program website. Scores for each of the subscales were calculated by averaging together the item responses in each section with higher scores representing higher states of readiness in specific settings and schools.
## Qualitative Interviews Grounded in CFIR
Following procedures developed by Damschroder and colleagues [28, 31, 41], an interview guide was developed which aimed to understand the influence of each CFIR domain on implementation of SWITCH (see Additional File 3). Each school's wellness team was invited to participate and we asked as many people as possible to attend the interviews (usually 3 per team). Questions were open-ended; examples included, “*What is* your perception of the quality of the modules, posters, and other SWITCH materials that were provided?” ( Innovation Characteristics – Design Quality and Packaging) and “How do you think your school culture affected the implementation of SWITCH programming?” ( Inner Setting – Implementation Climate). Interviews were conducted by a qualitative and survey methodologist to ensure impartiality in responses from school wellness teams. To address issues of sustainability, interviewers asked “Think of the changes you have made in your school setting. To what degree do you think these changes are sustainable?” and then prompted participants to expand on their responses with examples. The goal was to encourage candid responses so time limits were not imposed on these conversations. This ensured in-depth understanding of each context and implementation climate.
Of the 52 schools enrolled in SWITCH, 45 ($87\%$ of sample) completed interviews. Of these 45, 17 were new and 28 were experienced. Each school that participated had between 1 and 3 members of their school wellness team present. Table 2 shows representation of the various school staff positions within school wellness teams and those who were present in interviews; classroom and physical education teachers were included in most wellness teams and were most present on interviews, followed by food service and principals. Interviews lasted between 31 and 63 min, were conducted through video conferencing software (i.e., Zoom), and transcribed verbatim.
**Table 2**
| School staff role | # Represented in total sample N = 52 schools | # Represented in total sample N = 52 schools.1 | # Represented in inexperienced schools n = 22 | # Represented in inexperienced schools n = 22.1 | # Represented in experienced schools n = 30 | # Represented in experienced schools n = 30.1 |
| --- | --- | --- | --- | --- | --- | --- |
| | Wellness teams | Interview n = 45 | Wellness teams | Interview n = 17 | Wellness teams | Interview n = 28 |
| Classroom teacher | 59 | 21 | 22 | 7 | 37 | 14 |
| Counselor | 5 | 2 | 5 | 2 | 0 | 0 |
| Food service/nutrition | 17 | 3 | 7 | 2 | 10 | 1 |
| Instructional coach | 5 | 3 | 2 | 1 | 3 | 2 |
| Nurse | 15 | 4 | 3 | 0 | 12 | 4 |
| Paraprofessional | 1 | 0 | 0 | 0 | 1 | 0 |
| Physical education | 32 | 13 | 13 | 3 | 19 | 10 |
| Principal | 19 | 7 | 10 | 4 | 9 | 3 |
| Superintendent | 1 | 0 | 0 | 0 | 1 | 0 |
| Other | 9 | 2 | 4 | 1 | 5 | 1 |
## Qualitative Data Coding and Case Memos
The structure of the interview guide facilitated a predominantly deductive data analysis approach, in that each of the questions corresponded to a construct within each of the framework domains [31]. However, we remained open, such that any themes that emerged through inductive approaches were included in our analyses; such combination of deductive and inductive coding integrates data-driven codes with theory-driven ones [42]. For example, for the interview questions that addresses sustainability (additional files), we coded data from these responses deductively where they aligned with relevant constructs of CFIR but also inductively to provide critical information to the research team on what factors influence sustainment.
First, the lead and second author met to develop a coding consensus document (Additional File 4), which described each CFIR construct and anticipated potential responses and themes that would emerge through the data. Applying the CFIR systematic coding approach facilitated the assignment of numerical scoring to the qualitative data, such that if a particular construct was deemed to have a positive influence on implementation based on interview responses, a score of +1 or +2 was assigned for that construct. Conversely, if a construct was deemed to be a negative influence, a score of −1 or −2 was given. If it was not clear whether a positive/negative influence manifested, a score of 0 was given; a score of “X” was used for mixed results (see Additional File 5 for details on CFIR rating rules) [31].
Second, to establish inter-rater reliability, the two coders selected five transcripts and created independent case memos using the CFIR memo templates [41]. Scores were compared and a percent agreement score was calculated; if the overall agreement score was <$80\%$, the coders met to ensure consensus before coding another set of five transcripts. Once ≥$80\%$ agreement was met, the second author coded the remaining transcripts, before a randomly selected set of five transcripts was reviewed by the lead author. All coding was completed in memo documents (see Additional File 6). Finally, to facilitate content analysis and interpretation of trends in interview data, all memos were entered in to NVivo qualitative analysis software and coded into respective nodes, following the CFIR codebook template [41]. To prepare the quantified CFIR data for merging into the larger dataset, each school ID was aligned with the scores for each construct and domain of the model. Any X scores (implying a mixed/uncertain rating) were converted to 0 for the purpose of analysis. Any scores without a score remained blank so as not to misguide subsequent analyses.
## Aims 1 and 2: Evaluate Outcomes and Determinants
All school demographic, implementation outcome, and quantified implementation determinant data from the coded CFIR interviews were merged using SAS software (Version 9.4, Cary NC) to facilitate descriptive and inferential analyses. First, descriptive tests were conducted to obtain means (and SD) for all implementation outcome and determinant data, then split by experience level (0 = inexperienced; 1 = experienced). Following recommendations from Damschroder et al. [ 23] Pearson bivariate correlations were run to establish correlations between implementation outcomes and determinants to examine associations and to understand potential influences of implementation for schools that experienced greater success. All tests were run in SAS software (Cary, NC), and α significance was assumed as $p \leq 0.05$; correlations with $p \leq 0.10$ were also highlighted due to the novel nature of this work. Based on such associations, salient quotes from interview transcripts were extracted to provide rich contextual details on determinants.
## Aim 3: Investigate Nuanced Determinants for New and Experience Schools
To investigate distinguishing factors among inexperienced and experienced schools, an in-depth cross-case analysis was conducted based on prior evaluations through the CFIR in other settings [30, 31, 43, 44]. Cross-case analysis provides a broad scope for researchers to systematically compare multiple “cases” (i.e., schools) and is a derivative of Qualitative Comparative Analysis (QCA) [45, 46]. We pursued a combination of exploratory analysis and cross-case analysis to investigate the distinguishing factors between experienced and inexperienced schools. Given that our sample size afforded exploratory inferential testing, we first conducted independent t-tests to examine differences in mean scores for each CFIR construct between the two samples (α significance was assumed as $p \leq 0.05$). In addition, we sought constructs which had >0.5 difference in mean score between the two samples, to highlight other distinguishing factors which may influence implementation [31]. Subsequently, the research team explored qualitative extracts using NVivo as a means to contextualize findings from correlation analyses. Such an approach allowed for deeper contextual understanding of implementation practices which triangulate implementation determinants and outcomes [13].
To establish credibility, dependability, and trustworthiness, three key steps were taken in the analyses [47, 48]. First, although the coding methods applied a deductive process, the lead researcher regularly conducted peer debriefing with other members of the research team to minimize potential bias and assumptive coding. Second, the mixed methods design facilitated methods triangulation throughout analysis procedures which ensured that distinguishing factors gleaned through cross-case analysis were properly contextualized and refuted if not enough substantive evidence existed [49]. Finally, the use of coding memos provided the researchers with a method of maintaining an audit trail while coding, in which they took rigorous notes. This was exceptionally useful when establishing inter-rater reliability.
## Aim 1: Implementation Outcomes
Schools reported strong fidelity (mean score 7.6 ± 2.91); however, this varied by schools and by item. Experienced schools reported better fidelity overall except for using the SWITCH website (see Figure 1). Parent outreach was the lowest implemented practice; outreach activities mostly entailed sending newsletters that were provided by the SWITCH team ($70\%$); experienced schools reported more parent outreach practices than inexperienced schools. The most common method of school-wide integration was sending emails to the staff to inform them of the program and activities ($88\%$) followed by using posters to promote SWITCH themes in different settings ($73\%$). For adoption (5.53 ± 2.17), experienced schools reported significantly higher rates according to independent samples t-tests (t = −2.03, $$p \leq 0.04$$). This difference was consistent across use of modules, posters, and tracking/reinforcing themes. The highest implemented practice was classroom tracking followed by tracking in physical education setting (see Additional File 7).
**Figure 1:** *Fidelity to SWITCH quality elements (Mean, SD) by experience level. Checkpoint surveys conducted at week 6; Implementation fidelity scores 0, not implemented at all, 2, somewhat implemented, 3, implemented fully; meeting, school wellness team meeting; website, setting up classrooms and student tracking in the website; parents, parent outreach activity; integration, implementing educational modules/resources across each of the SWITCH settings.*
Regarding penetration, behavioral tracking data demonstrate that inexperienced and experienced schools were approximately equal in terms of tracking rates between week 1 and week 7; 43 ± $29\%$ of students in inexperienced schools and 46 ± $32\%$ of experienced schools tracked each week (mean score 0.448, or $45\%$). Tracking naturally dropped due to COVID-19-related school closures but it is noteworthy that rates were essentially $0\%$ for inexperienced schools but $25\%$ for experienced schools. This indicates that the experienced schools were more likely to retain tracking rates to a greater extent than inexperienced schools. Only data from the first 8 weeks are used for the related correlation analyses.
## Aim 2: Implementation Determinants
The process of converting qualitative interview data to numerical scores through CFIR protocols facilitated our ability to detect factors that were influential to SWITCH implementation outcomes. However, analysis of Cronbach's alpha revealed that none of the CFIR domains had acceptable internal consistency (all < 0.40). We therefore felt it important to show variability in the data as opposed to means and SD of global domains. Figure 2 displays scores from each domain as dual-sided histograms to facilitate examination of variability, separated by experience level (discussed below). From examination it appears that for all schools, factors within the Outer Setting and Implementation Process domains were most positively ranked, but high variability must be noted.
**Figure 2:** *Dual-sided histogram of CFIR domain scores, by experience level. Graph shows percentage of schools falling in specific ranges for global domain score (Y axis): −1 to −0.5; −0.5 to 0; 0 to 0.2; 0.21 to 0.4; 0.41 to 0.6; 0.61 to 0.8; 0.81 to 1.0; 1.01 to 1.2; 1.21 to 1.4; 1.41 to 1.*
Table 3 displays all means ± SD for CFIR construct data. In terms of positive influential factors, data reveal that the most positive scores from coding of interview data were Readiness for Implementation – Leadership Engagement (i.e., building administration involvement/support; mean = 1.22 ± 1.02), Individual Characteristics – Knowledge and Beliefs about the Intervention (i.e., school wellness teams' perceptions of SWITCH; 1.51 ± 0.51), and Implementation Process – External Stakeholders (i.e., county 4-H Extension officer support; 1.42 ± 1.12). Regarding negative influences, lowest scores were assigned to Inner Setting – Relative Priority (i.e., priority given to SWITCH over other programs; −0.31 ± 1.26), Readiness for Implementation – Available Resources (i.e., time, personnel, equipment; −0.96 ± 0.88), and challenges in Implementation Process - Key Stakeholders (i.e., engaging parents; −0.22 ± 1.31).
**Table 3**
| CFIR Domain | Construct | Inexperienced | Inexperienced.1 | Experienced | Experienced.1 | Total | Total.1 | Unnamed: 8 | t | Cohen's d | p |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| | | Mean | SD | Mean | SD | Mean | SD | >0.5 difference | | | |
| Intervention characteristics | Innovation source | 0.41 | 0.71 | 0.21 | 0.57 | 0.29 | 0.63 | | | | |
| | Evidence strength and quality | 1.06 | 0.90 | 0.79 | 0.88 | 0.89 | 0.88 | | | | |
| | Relative advantage | 0.12 | 0.49 | 0.11 | 0.42 | 0.11 | 0.44 | | | | |
| | Adaptability | 0.71 | 0.69 | 0.96 | 0.88 | 0.87 | 0.81 | | | | |
| | Trialability | 0.00 | 0.00 | 0.07 | 0.66 | 0.04 | 0.52 | | | | |
| | Complexity (reverse) | 0.06 | 1.03 | 0.43 | 0.92 | 0.29 | 0.97 | | | | |
| | Design quality | 0.71 | 0.85 | 0.25 | 1.27 | 0.42 | 1.14 | | | | |
| | Cost (reverse) | 0.41 | 0.51 | 0.64 | 0.87 | 0.56 | 0.76 | | | | |
| Outer setting | Student needs and resources | 0.41 | 0.71 | 0.75 | 0.84 | 0.62 | 0.81 | | | | |
| | Cosmopolitanism | 0.53 | 1.23 | 0.93 | 1.18 | 0.78 | 1.20 | * | | | |
| | Peer pressure | 0.71 | 1.05 | 0.25 | 0.70 | 0.42 | 0.87 | * | | | |
| | External policy and incentives | 1.06 | 1.14 | 0.68 | 0.94 | 0.82 | 1.03 | | | | |
| Inner setting | Structural characteristics | 0.00 | 1.00 | 0.39 | 1.23 | 0.24 | 1.15 | | | | |
| | Networks communications | 0.53 | 1.37 | 0.43 | 1.26 | 0.47 | 1.29 | | | | |
| | Tension for change | 0.29 | 0.59 | 0.32 | 0.61 | 0.31 | 0.60 | | | | |
| | Relative priority | −0.41 | 0.94 | −0.25 | 1.43 | −0.31 | 1.26 | | | | |
| | Culture | 0.88 | 0.86 | 0.82 | 0.94 | 0.84 | 0.90 | | | | |
| | Compatibility | 0.76 | 1.15 | 1.32 | 0.67 | 1.11 | 0.91 | * | | | |
| | Organizational incentives and rewards | 0.00 | 0.00 | 0.07 | 0.26 | 0.04 | 0.21 | | | | |
| | Goals and feedback | 0.53 | 0.80 | 0.43 | 0.57 | 0.47 | 0.66 | | | | |
| Readiness for implementation | Readiness for implementation | −0.12 | 1.22 | 0.89 | 0.74 | 0.51 | 1.06 | ** | −3.09 | 1.003 | 0.005 |
| | Learning climate | −0.12 | 0.33 | 0.14 | 0.52 | 0.04 | 0.47 | | | | |
| | Leadership engagement | 1.35 | 1.06 | 1.14 | 1.01 | 1.22 | 1.02 | | | | |
| | Available resources | −1.24 | 0.83 | −0.79 | 0.88 | −0.96 | 0.88 | | | | |
| | Access to Knowledge and Information | 0.65 | 1.46 | 0.36 | 1.16 | 0.47 | 1.27 | | | | |
| Individual characteristics | Knowledge and beliefs about intervention | 1.35 | 0.49 | 1.61 | 0.50 | 1.51 | 0.51 | | | | |
| | Self-efficacy | −0.24 | 1.35 | 0.61 | 0.92 | 0.29 | 1.16 | ** | −2.50 | 0.731 | 0.01 |
| | Individual stage of change | −0.06 | 0.24 | −0.07 | 0.54 | −0.07 | 0.45 | | | | |
| | Individual identification with organization | 0.00 | 0.00 | −0.07 | 0.66 | −0.04 | 0.52 | | | | |
| | Other personal attributes | 0.41 | 1.12 | 0.25 | 1.08 | 0.31 | 1.08 | | | | |
| Implementation process | Planning | 0.29 | 1.05 | 0.21 | 1.42 | 0.24 | 1.28 | | | | |
| | Implementation leaders | 0.88 | 1.11 | 0.79 | 0.74 | 0.82 | 0.89 | | | | |
| | Engaging | 0.47 | 1.50 | 1.00 | 1.15 | 0.80 | 1.31 | * | | | |
| | Opinion leader | 0.24 | 1.39 | 0.54 | 1.07 | 0.42 | 1.20 | | | | |
| | Champions | 0.94 | 0.90 | 1.07 | 1.02 | 1.02 | 0.97 | | | | |
| | External change agents | 1.24 | 1.15 | 1.54 | 1.10 | 1.42 | 1.12 | | | | |
| | Key stakeholders | −0.47 | 1.18 | −0.07 | 1.39 | −0.22 | 1.31 | | | | |
| | Innovation participants | 1.24 | 1.09 | 1.75 | 0.44 | 1.56 | 0.78 | * | | | |
| | Executing | 0.59 | 1.12 | 0.86 | 1.01 | 0.76 | 1.05 | | | | |
| | Reflecting and evaluating | 0.12 | 0.33 | 0.25 | 0.65 | 0.20 | 0.55 | | | | |
Table 4 illustrates the results from exploratory Pearson bivariate correlation analyses for the whole sample. Almost all associations were positive, except Inner Setting-Networks and Communications (r = −0.28; $$p \leq 0.07$$) and Tension for Change ($r = 0.27$; $$p \leq 0.09$$), both negatively correlated with Adoption. Tension for Change was also negatively associated with Penetration (r = −0.33; $$p \leq 0.02$$). Salient interview extracts which relate to implementation outcomes are available in Additional File 8 and provide context for the whole sample. Figure 3 displays findings from the SWRA tool to assess baseline readiness/capacity. For the overall sample, a significant correlation was found between classroom readiness and adoption ($r = 0.366$, $$p \leq 0.02$$). This illustrates that schools that reported greater classroom capacity were also using modules, tracking, and using posters more often than schools with lower classroom capacity. For the inexperienced schools, overall school capacity was positively correlated to adoption, indicating that organization-level readiness was associated with use of best practices across the school ($r = 0.513$, $$p \leq 0.04$$). The lack of relationship between other capacity indicators and implementation outcomes is potentially due to the lack of variability in the capacity means. School and Class capacity had the largest range in scores (2 or 2.25 to 5) compared to PE and lunch capacity (between 3 and 5).
## Aim 3: Differences in Outcomes and Determinants Among New and Experienced Schools, and Influential Factors to Sustainment of SWITCH
Cross-case analyses facilitated understanding of distinguishing implementation determinants between inexperienced and experienced schools. For inexperienced schools, the highest ranked positive determinant was Leadership Engagement (Inner Setting-Readiness for Implementation), suggesting that school administration support was an important contributing factor. For experienced schools, the highest ranked positive determinant was Engaging-Innovation Participants (Implementation Process), indicating that student involvement and advocacy was helpful for success. Table 3 highlights the differences in mean scores between the two samples and distinguishing factors according to independent samples t-tests and large differences in means not detected through inferential testing. The two constructs which were statistically different were Readiness for Implementation (Inner Setting) and Self-efficacy (Characteristics of Individuals); experienced schools had positive and significantly higher means in these constructs indicating they were positive determinants to implementation. Other distinguishing constructs which had large score differences were Cosmopolitanism (Outer Setting), Peer Pressure (Outer Setting), Compatibility (Inner Setting), Engaging (Implementation Process), and Innovation Participants (Implementation Process). Experienced schools had higher means except for Peer Pressure, which was higher for inexperienced schools.
Table 5 highlights the distinguishing constructs which separated the two samples based on deductive coding, with salient interview extracts from school participants. Extracts were chosen to represent some of the diverging quotations from the two distinct subsamples and reflect the ways in which they experienced implementation facilitators and barriers according to each CFIR construct. An example from the Readiness for Implementation highlights a difference for inexperienced schools, “From the first meeting it sounded like it was just maybe teaching a couple of lessons, and [team member] was going to be doing most of it, but we quickly found out that, that really wasn't the case” and experienced schools, “I think our core team does really well at keeping these things planned, and sticking together, and letting administration know what we're doing, and getting the okay.” Another example from the Engaging construct highlights one perspective: “I don't feel we communicated well-enough to allow, or to educate the teachers on the importance of this program” (inexperienced). This is contrasted with an experienced school team member who said, “We had more success getting kids connected to their parents this year, compared to last year.” These differences highlight nuanced barriers/facilitators among the two samples.
**Table 5**
| Construct | Inexperienced | Experienced |
| --- | --- | --- |
| Readiness for implementation** | “From the first meeting it sounded like it was just maybe teaching a couple of lessons, and [team member] was going to be doing most of it, but we quickly found out that, that really wasn't the case, but it worked out well, though. After we got over that initial shock of, ‘oh my gosh, it's a lot more work' but it did go well” | “I think our core team does really well at keeping these things planned, and sticking together, and letting administration know what we're doing, and getting the okay. But then going about and implementing it and getting the help we need to go it from the parents' community, just doing it that way” |
| Self-efficacy** | “It was mostly just me and [other teacher]. They were on board, but yeah, again, just, it was brand new to us, so we didn't know how to incorporate everyone else into it just fully yet” | “I had 100% confidence in my teachers, because we sat down the year before and chose to do it again. Like I said, I feel that they did the best that they could with the amount of time that they had to be able to implement additional curriculum into their already busy curriculum” |
| Cosmopolitanism | “I didn't do a good job of reaching out to the community to see if there was anyone interested in helping us” | “We did the Iowa Farm-to-School local food day this past school year. And we were able to get apples and cider from [local orchard], and then we got fresh leaf lettuce and vegetables from our own greenhouse here. And so, we were able to explain that to the kids, and [4-H officer] actually came in and helped during that” |
| Peer pressure | “So knowing kind of the ins and outs and how [SWITCH] should look from another previous school that had success with it, really helped us just kind of get going and get it running at our school” | “We used to put a lot of things of what our school did to share our ideas, and we didn't [in the community of practice] but we did on our school Facebook page and shared a lot in that way. So, this year I didn't feel like I knew what a lot of schools had done” |
| Compatibility | “Our biggest hurdle was finding time for sixth, seventh, and eighth grade classroom activities just because our schedule just didn't work out very well. We ended up having all six, seventh and eighth graders on Mondays for Switch. We're a really tiny school, but that's still about 50 kids, which is a large group in a gym trying to teach” | “I guess I just keep going back to our kickoff that and with the teachers came up with on that and how it directly coincided with SWITCH and they were phenomenal. I think that they had the opportunity to do it something within their classroom I think they would do it” |
| Engaging | “I don't feel we communicated well enough to allow, or to educate the teachers on the importance of this program. In that regard, I need to do a better job next year along with whoever's helping in this” | “We had more success getting kids connected to their parents this year, compared to last year. We only had two connected last year, and I was one of them. And I think we ended up with about 25 parents connected to kids, which doesn't sound like a lot… that's still something that we want to improve upon so they know what the kids are doing so they can then support it at home” |
Finally, results from inductive coding with regard to sustainability revealed three overarching themes: [1] The importance of student awareness; [2] Keeping it simple; and [3] Integrating within school culture. Additional File 9 shows salient quotes from interviews related to these themes, with quotes separated by experience level. For the first theme, when wellness team members were asked if they felt their changes were sustainable, many pointed to the impact SWITCH has had on students as a key reason why the program would be maintained in their setting. One inexperienced school member said, “the kids have now become aware of [how] they can change what they do, view, and chew… And maybe next year when they see us in the hallway, it'll click and [they will] remember that kind of stuff.” Many school wellness team members emphasized that while they could not implement all parts of SWITCH as much as they wanted to, they mentioned specific practices that seemed simple and granular which could be sustained. For example, one experienced school member said, “We've tried to do one thing at a time, to see if it was going to work. Changing the milk, we can do that. We do that all the time, now. And the brain breaks in the classroom, that's sustainable.” This indicates that the wellness teams are thinking more about the discrete practices/policies they have in place as opposed to the comprehensive nature of the program, which may be too overwhelming. Finally, participants discussed how they “really see this as it's just part of our culture” (inexperienced school) when discussing this question. One experienced school member explicitly discussed how they are planning to keep SWITCH going despite common challenges of staff turnover which are pervasive in schools: This quote emphasizes the work that wellness teams have carried out to fully embed SWITCH within their systems so that it is compatible for their schools.
## Discussion
The aims of this study were to assess implementation outcomes of adoption, fidelity, and penetration of SWITCH to identify the factors that may influence sustainability. Grounded by CFIR, we discerned implementation determinants through a deductive approach and specifically examined the differences in outcomes and determinants among new and experienced schools. The use of CFIR as a guiding framework is novel in the school wellness setting, specifically the use of the framework systematic data analysis procedures, which facilitated a deep contextual understanding of relationships between implementation determinants and outcomes. Thus, a key innovation is the adaptation of a framework predominantly intended for healthcare settings (i.e., CFIR) to the school setting, marking an important advancement in the field of implementation science.
## Aim 1: Assess Implementation Outcomes
The SWITCH program represents a capacity-building process which allows school wellness teams to develop and sustain comprehensive programs of their own which in turn are more sustainable over time. The moderate-high rates of Penetration also correspond with self-reported Adoption of program best practices across the school setting. Implementation data from adoption, fidelity, and penetration measures highlight the differences between experienced and experienced schools, a result that aligns with preliminary findings from prior evaluations [13]. However, the finding that all schools struggled to engage parents despite increased efforts in the 2020 academic year reflects a wealth of prior research documenting this lack of engagement problem [8, 50, 51]. Outreach practices of sending communications (emails/newsletters) and holding events for parental engagement were the most frequently reported, reflecting similar trends with school nutrition program promotion [52]. Such findings stress the need to view implementation outcomes as incrementally changing constructs that must be studied over time. This finding is consistent with generalized recommendations for continuous quality improvement models [37, 53].
## Aim 2: Assess Determinants of Implementation
The finding that Cosmopolitanism was higher in experienced schools, but Peer Pressure was lower than inexperienced schools, provides valuable information for how to support implementation efforts. Having links to other schools and organizations was viewed as a positive determinant of fidelity; interview data yielded some reasons for this, such as implementation support for delivering lessons and additional program materials and equipment, which may have further pushed a culture of health in school buildings. Although some initial research has demonstrated the positive role of external networks and support [54, 55], very little is known about the effectiveness of implementation strategies which provide targeted support from this domain. Accordingly, a potential implementation strategy for future work with schools may be to provide a local network of support, bringing together other sectors such as food retail and community centers, ultimately enhancing the culture of health in the community [56, 57].
## Aim 3: Differences Between New and Experienced Schools
For inexperienced schools, Leadership Engagement was the highest rated positive determinant of implementation. This is noteworthy since lack of support or involvement from school administration is a frequently reported barrier in school-based interventions (58–61). In SWITCH, administrators were able to be a part of the wellness team and attend conferences and trainings which likely enhanced their exposure to—and awareness of—school wellness programming. For all schools, Available Resources was the most negatively ranked determinant, indicating this was the biggest challenge for implementation. Examples from interviews highlighted the role of personnel time, equipment availability, and funding as supports for implementation. Therefore, an implementation strategy for inexperienced schools may be a cost-matching initiative through local county 4-H extension or through collaborating with community stakeholders, as described above and recommended through findings of Waltz et al. [ 62]. County extension offices have been encouraged to support SWITCH programming in their county, so this finding supports the importance of this practice. Engagement of Extension in this way also enhances cross-sector collaborations to build more sustainable school and community health programming [57].
As expected, the Readiness for Implementation domain and findings from the SRWA assessment highlight the importance of capacity-building programs for systems change [9]. Both Readiness for Implementation and Self-Efficacy were significantly higher for experienced schools than inexperienced, bolstering findings from the SWRA. This is not surprising, as items from the SWRA relate to Self-Efficacy in the individual and organizational psychological domains, such as “staff members at all levels share a belief that they can implement school wellness programs effectively.” Thus, implementation strategies to bolster capacity for implementation may be most appropriate. Within the D&I literature, the Expert Recommendations for Implementing Change (ERIC) research provides groundwork for selecting implementation strategies based on reported implementation challenges through models such as CFIR, facilitating tailored implementation support (62–64). For example, a CFIR-ERIC matching protocol conducted by Waltz et al. [ 62] and adapted by Cook et al. for school settings [65] highlighted that for Readiness for Implementation barriers, experts recommended “Assess for readiness and identify barriers and facilitators” as potential implementation strategies. In SWITCH, a core wellness team of at least three school staff members are trained over the course of a semester and complete the SWRA tool and School Wellness Environment Profile assessment, thus these strategies are already key components of the intervention model. Input from school stakeholders is often absent from the literature on implementation strategies, and a next step may be to include them in mapping procedures to advance the field.
The Implementation Process domain revealed that experienced schools ranked Engaging-general and Engaging-Innovation Participants distinctively higher than inexperienced schools, indicating these were more positively related to implementation. Related to innovation participants, youth advocacy in school wellness and health promotion has been demonstrated as an effective strategy for implementation and student health outcomes (66–68) and some studies are emerging regarding how student advocacy groups can be studied through a D&I lens [69]. Engaging – Key Stakeholders was seen as a negative implementation determinant for all schools. Parents have been reported as the most difficult stakeholder group to engage in school-wide initiatives, and in previous cycles of SWITCH [13, 32, 50]; however, some schools reported that when they did hold an event at school or at another academic-related event (i.e., parent-teacher conference), parents showed great interest. Thus, more research is needed to identify effective ways for engaging parents in school wellness, ideally with parents as the primary participants, to identify potential implementation strategies.
Finally, the inductive coding pertaining to sustainability revealed three primary themes which illustrate the strategies schools sought to maintain elements of SWITCH. A recent review highlights that most articles reporting facilitators/barriers to sustainment of interventions in schools cite factors from the Inner Setting as key determinants [9]. Findings from the current study provide potential strategies that could be applied to mitigate barriers to sustainability, specifically [1] promoting student awareness and engagement, [2] focusing on a small number of key elements, and [3] integrating programming within school culture. These strategies were mentioned by participants as next steps for their wellness environment as formal implementation concluded, and all relate to potential barriers within the Inner Setting domain. However, it must be acknowledged that we were not able to test formalized strategies to enhance sustainability. Thus, a logical next step in this area may be to operationalize “sustainment” and to test the relative effectiveness of different strategies to enhance the sustainability of capacity-building interventions such as SWITCH. The present study provides insights into this development by identifying barriers and facilitators of adoption, fidelity and penetration.
## Limitations
There are some limitations that could influence interpretations from this type of evaluation. First, and most important, the COVID-19 pandemic led to school closures which prevented completion of the 12-week implementation cycle. Thus, it is not clear whether the documented differences between inexperienced and experienced schools would have persisted or varied. Further, to prevent overburdening school staff, we refrained from collecting checkpoint survey data once schools closed and began remote learning, which may have limited understanding of fidelity and adoption within schools. Finally, we acknowledge potential limitations of applying CFIR constructs and coding methods non-healthcare settings. Our study was one of the first to employ CFIR in school settings using a fully integrated mixed methods procedure. Therefore, the CFIR constructs/methods and their applications to school and community-based settings may need to evolve over time as replication of these methods occur. Ongoing work with SWITCH has utilized these findings, but the results provide generalizable insights about factors that influence the scale up and sustainment of interventions in other community-based settings [70]. The process and systematic approach to the use of CFIR in the analyses also provide a guide for other school-based researchers seeking to utilize D&I methods to evaluate programming.
## Conclusions
The present study highlighted various determinants that influenced implementation and sustainability of SWITCH. The study added novel insights which can be tested and applied in other studies in school and community settings. Specifically, we documented that inexperienced schools face greater challenges and need tailored support, findings which indirectly document the gains in capacity built through previous iterations of SWITCH. The mixed methods approach used in the study was particularly important in understanding the factors influencing implementation and the greater challenges faced by inexperienced schools.
An advantage of CFIR in the project is that it provides a systematic method for enhancing the rigor and quality of implementation evaluations. Replication of the methods in other school-based projects would enable more effective comparisons. The adoption of “common measures” for implementation determinants and outcomes is already evident in other lines of research (70–73). Similar standardization efforts in school-based research would enhance generalizability and transferability of qualitative findings to other contexts and geographic locations. It is clear that what gets measured often is what gets achieved. By standardizing methods and measures, there is greater potential for enhancing implementation and sustainability of school-based interventions through incremental evaluation.
## Data Availability Statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics Statement
The studies involving human participants were reviewed and approved by the Institutional Review Board (#14–651) at Iowa State University. The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
GM and GW led the mixed methods design and evaluation components. RS and GM led the qualitative analysis procedures. RS and LL facilitated survey and interview data collection procedures. GM analyzed survey data and developed measures for school capacity. RR and JL provided feedback on analysis and interpretation of qualitative data. All authors contributed to the development of the research study and provided ongoing feedback throughout the implementation evaluation process and read and approved the final manuscript.
## Funding
USDA NIFA grant: 2015–68001-23242. The USDA was not involved in the design of the study and collection, analysis, and interpretation of data or writing of the manuscript.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary Material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frhs.2022.881639/full#supplementary-material
## References
1. Mansfield JL, Savaiano DA. **Effect of school wellness policies and the healthy, hunger-free kids act on food-consumption behaviors of students, 2006–2016: a systematic review**. *Nutr Rev.* (2017.0) **75** 533-52. DOI: 10.1093/nutrit/nux020
2. Watson A, Timperio A, Brown H, Best K, Hesketh KD. **Effect of classroom-based physical activity interventions on academic and physical activity outcomes: a systematic review and meta-analysis**. *Int J Behav Nutr Physical Act.* (2017.0) **14** 114. DOI: 10.1186/s12966-017-0569-9
3. Sisnowski J, Street JM, Merlin T. **Improving food environments and tackling obesity: a realist systematic review of the policy success of regulatory interventions targeting population nutrition**. *PLoS ONE.* (2017.0) **12** e0182581. DOI: 10.1371/journal.pone.0182581
4. Russ LB, Webster CA, Beets MW, Phillips DS. **Systematic review and meta-analysis of multi-component interventions through schools to increase physical activity**. *Journal of Physical Activity and Health.* (2015.0) **12** 1436-46. DOI: 10.1123/jpah.2014-0244
5. Amini M, Djazayery A, Majdzadeh R, Taghdisi M-H, Jazayeri S. **Effect of school-based interventions to control childhood obesity: a review of reviews**. *Int J Prev Med.* (2015.0) **6** 1-15. DOI: 10.4103/2008-7802.162059
6. Brownson RC CG, Proctor EK. **Future Issues in Dissemination and Implementation Research**. *In: Brownson RC CG, Proctor EK, editor Dissemination and Implementation Research in Health: Translating Science to Practice Second Edition.* (2018.0). DOI: 10.1093/oso/9780190683214.003.0029
7. Brownson RC, Colditz GA, Proctor EK. *Dissemination and Implementation Research in Health: Translating Science to Practice, Second Edition* (2018.0). DOI: 10.1093/oso/9780190683214.001.0001
8. Herlitz L, MacIntyre H, Osborn T, Bonell C. **The sustainability of public health interventions in schools: a systematic review**. *Implement Sci* (2020.0) **15** 4. DOI: 10.1186/s13012-019-0961-8
9. Shoesmith A, Hall A, Wolfenden L, Shelton RC, Powell BJ, Brown H. **Barriers and facilitators influencing the sustainment of health behaviour interventions in schools and childcare services: a systematic review**. *Implement Science* (2021.0) **16** 1134. DOI: 10.1186/s13012-021-01134-y
10. Welk GJ, Chen S, Nam YH, Weber TE. **A formative evaluation of the SWITCH® obesity prevention program: print vs. online programming**. *BMC Obesity* (2015.0) **2** 20. DOI: 10.1186/s40608-015-0049-1
11. Gentile DA, Welk GJ, Eisenmann JC, Reimer RA, Walsh DA, Rusell DW. **Evaluation of a multiple ecological level child obesity prevention program: Switch® what you Do, View, and Chew**. *BMC Med* (2009.0) **7** 49. DOI: 10.1186/1741-7015-7-49
12. Eisenmann JC, Gentile DA, Welk GJ, Callahan R, Strickland S, Walsh M. **SWITCH: rationale, design, and implementation of a community, school, and family-based intervention to modify behaviors related to childhood obesity**. *BMC Public Health* (2008.0) **8** 223. DOI: 10.1186/1471-2458-8-223
13. McLoughlin GM, Candal P, Vazou S, Lee JA, Dzewaltowski DA, Rosenkranz RR. **Evaluating the implementation of the SWITCH® school wellness intervention and capacity-building process through multiple methods**. *Int J Behav Nutr Phys Act.* (2020.0) **17** 162. DOI: 10.1186/s12966-020-01070-y
14. **Local school wellness policy implementation under the Healthy, Hunger-Free Kids Act of 2010, final rule**. (2016.0) **81** 50151-70
15. Chen S, Dzewaltowski DA, Rosenkranz RR, Lanningham-Foster L, Vazou S, Gentile DA. **Feasibility study of the SWITCH implementation process for enhancing school wellness**. *BMC Public Health* (2018.0) **18** 1119. DOI: 10.1186/s12889-018-6024-2
16. Lou Y, Wu H, Welk GJ, Vazou S, Chen S, Gentile DA. **The SWITCH implementation process on school lunch consumption patterns and plate waste**. *J Nut Edu Behav.* (2018.0) **50** S167-S8. DOI: 10.1016/j.jneb.2018.04.216
17. Welk G, Chen S, Vazou S, Lanningham-Foster L, Gentile D, Rosenkranz R. **Implementation feasibility of school modules designed to enhance the evidence-based switch obesity prevention program**. *J Nutr Edu Behav.* (2016.0) **48** S133. DOI: 10.1016/j.jneb.2016.04.384
18. Chen S, Liu Y, Welk G. **Using a hybrid design to analyze effectiveness and implementation of a refined energy-balance education module for upper elementary physical education**. *Ejournal de la recherche sur l'intervention en éducation physique et sport -eJRIEPS* (2019.0) **3** 108-24. DOI: 10.4000/ejrieps.3648
19. Lee JA, McLoughlin GM, Welk GJ. **School wellness environments: perceptions vs. realities**. *J School Nurs* (2020.0) **58** 1059840520924453. DOI: 10.1177/1059840520924453
20. Lee JA, Welk GJ, Vazou S, Ellingson LD, Lanningham-Foster L, Dixon P. **Development and application of tools to assess elementary school wellness environments and readiness for wellness change (doctoral dissertation)**. (2018.0)
21. Rosenkranz R, Dixon P, Dzewaltowski D, McLoughlin GM, Lee JA, Chen S. **A cluster-randomized trial comparing two SWITCH implementation support strategies for school wellness intervention effectiveness**. *J Sport Health Sci*. DOI: 10.1016/j.jshs.2021.12.001
22. McLoughlin GM, Vazou S, Liechty L, Torbert A, Lanningham-Foster L, Rosenkranz RR. **Transdisciplinary approaches for the dissemination of the switch school wellness initiative through a distributed 4-H/extension network**. *Child Youth Care Forum.* (2020.0) **50** 99-120. DOI: 10.1007/s10566-020-09556-3
23. Damschroder LJ, Reardon CM, Sperber N, Robinson CH, Fickel JJ, Oddone EZ. **Implementation evaluation of the Telephone Lifestyle Coaching (TLC) program: organizational factors associated with successful implementation**. *Transl Behav Med.* (2017.0) **7** 233-41. DOI: 10.1007/s13142-016-0424-6
24. Bice MR, Brown SL, Parry T. **Retrospective evaluation of factors that influence the implementation of CATCH in Southern Illinois schools**. *Health Promot Pract.* (2014.0) **15** 706-13. DOI: 10.1177/1524839914526206
25. Spoth R, Greenberg M. **Impact challenges in community science-with-practice: lessons from PROSPER on transformative practitioner-scientist partnerships and prevention infrastructure development**. *Am J Community Psychol.* (2011.0) **48** 106-19. DOI: 10.1007/s10464-010-9417-7
26. Weiner BJ. **A theory of organizational readiness for change**. *Implement Sci* (2009.0) **4** 67. DOI: 10.1186/1748-5908-4-67
27. Kirk MA, Kelley C, Yankey N, Birken SA, Abadie B, Damschroder L. **A systematic review of the use of the consolidated framework for implementation research**. *Implement Sci* (2016.0) **11** 72. DOI: 10.1186/s13012-016-0437-z
28. Damschroder LJ, Aron DC, Keith RE, Kirsh SR, Alexander JA, Lowery JC. **Fostering implementation of health services research findings into practice: a consolidated framework for advancing implementation science**. *Implement Sci.* (2009.0) **4** 50. DOI: 10.1186/1748-5908-4-50
29. Fernandez ME, Walker TJ, Weiner BJ, Calo WA, Liang S, Risendal B. **Developing measures to assess constructs from the inner setting domain of the consolidated framework for implementation research**. *Implement Sci.* (2018.0) **13** 52. DOI: 10.1186/s13012-018-0736-7
30. Soi C, Gimbel S, Chilundo B, Muchanga V, Matsinhe L, Sherr K. **Human papillomavirus vaccine delivery in Mozambique: identification of implementation performance drivers using the consolidated framework for implementation research (CFIR)**. *Implement Sci.* (2018.0) **13** 151. DOI: 10.1186/s13012-018-0846-2
31. Damschroder LJ, Lowery JC. **Evaluation of a large-scale weight management program using the consolidated framework for implementation research (CFIR)**. *Implement Sci.* (2013.0) **8** 51. DOI: 10.1186/1748-5908-8-51
32. Leeman J, Wiecha JL, Vu M, Blitstein JL, Allgood S, Lee S. **School health implementation tools: a mixed methods evaluation of factors influencing their use**. *Implement Sci* (2018.0) **13** 48. DOI: 10.1186/s13012-018-0738-5
33. Bozsik F, Berman M, Shook R, Summar S, DeWit E, Carlson J. **Implementation contextual factors related to youth advocacy for healthy eating and active living**. *Transl Behav Med.* (2018.0) **8** 696-705. DOI: 10.1093/tbm/ibx006
34. Cassar S, Salmon J, Timperio A, Naylor P-J, van Nassau F, Contardo Ayala AM. **Adoption, implementation and sustainability of school-based physical activity and sedentary behaviour interventions in real-world settings: a systematic review**. (2019.0) **16** 120. DOI: 10.1186/s12966-019-0876-4
35. Koester M, Bejarano CM, Davis AM, Brownson RC, Kerner J, Sallis JF. **Implementation contextual factors related to community-based active travel to school interventions: a mixed methods interview study**. *Implement Sci Commun.* (2021.0) **2** 94. DOI: 10.1186/s43058-021-00198-7
36. Roberts S, Keane J, Ward C, Restrick L. **Plan, do, study, act**. *Physiotherapy* (2002.0) **88** 769. DOI: 10.1016/S0031-9406(05)60736-5
37. Proctor E, Silmere H, Raghavan R, Hovmand P, Aarons G, Bunger A. **Outcomes for implementation research: conceptual distinctions, measurement challenges, and research agenda**. *Adm Policy Ment Health.* (2011.0) **38** 65-76. DOI: 10.1007/s10488-010-0319-7
38. McLoughlin GM, Rosenkranz RR, Lee JA, Wolff MM, Chen S, Dzewaltowski DA. **The importance of self-monitoring for behavior change in youth: Findings from the SWITCH® school wellness feasibility study**. *Int J Environ Res Public Health* (2019.0) **16** 3806. DOI: 10.3390/ijerph16203806
39. Holt DT, Helfrich CD, Hall CG, Weiner BJ. **Are you ready? How health professionals can comprehensively conceptualize readiness for change**. *J General Internal Med* (2010.0) **25** 50-5. DOI: 10.1007/s11606-009-1112-8
40. Hawe P, Shiell A, Riley T. **Theorising interventions as events in systems**. *Am J Community Psychol.* (2009.0) **43** 267-76. DOI: 10.1007/s10464-009-9229-9
41. **Consolidated Framework for Implementation Research- Qualitative Analysis**
42. Fereday J, Muir-Cochrane E. **Demonstrating rigor using thematic analysis: a hybrid approach of inductive and deductive coding and theme development**. *Int J Qualit Methods.* (2006.0) **5** 80-92. DOI: 10.1177/160940690600500107
43. Wilhelm AK, Schwedhelm M, Bigelow M, Bates N, Hang M, Ortega L. **Evaluation of a school-based participatory intervention to improve school environments using the consolidated framework for implementation research**. *BMC Public Health.* (2021.0) **21** 1615. DOI: 10.1186/s12889-021-11644-5
44. Keith RE, Crosson JC, O'Malley AS, Cromp D, Taylor EF. **Using the Consolidated Framework for Implementation Research (CFIR) to produce actionable findings: a rapid-cycle evaluation approach to improving implementation**. *Implement Sci.* (2017.0) **12** 15. DOI: 10.1186/s13012-017-0550-7
45. Berg-Schlosser D, De Meur G, Rihoux B, Ragin CC. **Qualitative comparative analysis (QCA) as an approach**. *Conf Comp Methods: Qualit Comp Anal (QCA) Related Tech.* (2009.0) **1** 18. DOI: 10.4135/9781452226569.n1
46. Rihoux B, Lobe B. **The case for qualitative comparative analysis (QCA): Adding leverage for thick cross-case comparison**. *Sage Handbook Case-Based Methods* (2009.0). DOI: 10.4135/9781446249413.n13
47. Lincoln YS, Guba EG. *Naturalistic Inquiry* (1985.0). DOI: 10.1016/0147-1767(85)90062-8
48. Patton MQ. *Qualitative Research and Evaluation Methods* (2015.0)
49. Whittemore R, Chase SK, Mandle CL. **Validity in qualitative research**. *Qual Health Res.* (2001.0) **11** 522-37. DOI: 10.1177/104973201129119299
50. McDowall PS, Schaughency E. **Elementary school parent engagement efforts: relations with educator perceptions and school characteristics**. *J Educational Res* (2017.0) **110** 348. DOI: 10.1080/00220671.2015.1103687
51. Clarke JL, Griffin TL, Lancashire ER, Adab P, Parry JM, Pallan MJ. **Parent and child perceptions of school-based obesity prevention in England: a qualitative study**. *BMC Public Health* (2015.0) **15** 2567. DOI: 10.1186/s12889-015-2567-7
52. McLoughlin GM, Turner L, Leider J, Piekarz-Porter E, Chriqui JF. **Assessing the relationship between district and state policies and school nutrition promotion-related practices in the United States**. *Nutrients.* (2020.0) **12** 2356. DOI: 10.3390/nu12082356
53. Brownson RC, Colditz GA, Proctor EK. *Dissemination and Implementation Research in Health: Translating Science to Practice* (2012.0). DOI: 10.1093/acprof:oso/9780199751877.001.0001
54. Watson DP, Adams EL, Shue S, Coates H, McGuire A, Chesher J. **Defining the external implementation context: an integrative systematic literature review**. *BMC Health Serv Res.* (2018.0) **18** 209. DOI: 10.1186/s12913-018-3046-5
55. Weatherson KA, Gainforth HL, Jung ME. **A theoretical analysis of the barriers and facilitators to the implementation of school-based physical activity policies in Canada: a mixed methods scoping review**. *Implement Sci.* (2017.0) **12** 41. DOI: 10.1186/s13012-017-0570-3
56. Mazzucca S, Arredondo EM, Hoelscher DM, Haire-Joshu D, Tabak RG, Kumanyika SK. **Expanding implementation research to prevent chronic diseases in community settings**. *Annu Rev Public Health.* (2021.0) **42** 135-58. DOI: 10.1146/annurev-publhealth-090419-102547
57. Kumanyika SK. **A Framework for Increasing Equity Impact in Obesity Prevention**. *Am J Public Health.* (2019.0) **109** 1350-7. DOI: 10.2105/AJPH.2019.305221
58. McLoughlin GM, Graber KC, Woods AM, Templin T, Metzler M, Khan NA. **The status of physical education within a nationally recognized school health and wellness program**. *J Teach Phys Educ.* (2020.0) **39** 274-83. DOI: 10.1123/jtpe.2019-0052
59. Levay AV, Chapman GE, Seed B, Wittman H. **Examining school-level implementation of British Columbia, Canada's school food and beverage sales policy: a realist evaluation**. *Public Health Nutr.* (2020.0) **23** 1460-71. DOI: 10.1017/S1368980019003987
60. Carlson JA, Engelberg JK, Cain KL, Conway TL, Geremia C, Bonilla E. **Contextual factors related to implementation of classroom physical activity breaks**. *Transl Behav Med.* (2017.0) **7** 581-92. DOI: 10.1007/s13142-017-0509-x
61. Allison KR, Vu-Nguyen K, Ng B, Schoueri-Mychasiw N, Dwyer JJM, Manson H. **Evaluation of daily physical activity (DPA) policy implementation in Ontario: surveys of elementary school administrators and teachers**. *BMC Public Health.* (2016.0) **16** 1-16. DOI: 10.1186/s12889-016-3423-0
62. Waltz TJ, Powell BJ, Fernández ME, Abadie B, Damschroder LJ. **Choosing implementation strategies to address contextual barriers: diversity in recommendations and future directions**. *Implement Sci.* (2019.0) **14** 42. DOI: 10.1186/s13012-019-0892-4
63. Powell B, Beidas R, Lewis C, Aarons G, McMillen J, Proctor E. **Methods to improve the selection and tailoring of implementation strategies**. *J Behav Health Serv Res.* (2017.0) **44** 177-94. DOI: 10.1007/s11414-015-9475-6
64. Powell BJ, Waltz TJ, Chinman MJ, Damschroder LJ, Smith JL, Matthieu MM. **A refined compilation of implementation strategies: results from the Expert Recommendations for Implementing Change (ERIC) project**. *Implement Sci.* (2015.0) **10** 21. DOI: 10.1186/s13012-015-0209-1
65. Cook CR, Lyon AR, Locke J, Waltz T, Powell BJ. **Adapting a compilation of implementation strategies to advance school-based implementation research and practice**. *Prev Sci.* (2019.0) **20** 914-35. DOI: 10.1007/s11121-019-01017-1
66. Millstein RA, Woodruff SI, Linton LS, Edwards CC, Sallis JF. **Development of measures to evaluate youth advocacy for obesity prevention**. *Int J Behav Nutr Phys Act.* (2016.0) **13** 84. DOI: 10.1186/s12966-016-0410-x
67. Millstein RA, Woodruff SI, Linton LS, Edwards CC, Sallis JF. **A pilot study evaluating the effects of a youth advocacy program on youth readiness to advocate for environment and policy changes for obesity prevention**. *Transl Behav Med.* (2016.0) **6** 648-58. DOI: 10.1007/s13142-016-0408-6
68. Morse LL, Allensworth DD. **Placing students at the center: the whole school, whole community, whole child model**. *Journal of School Health.* (2015.0) **85** 785-94. DOI: 10.1111/josh.12313
69. Lane HG, Deitch R, Wang Y, Black MM, Dunton GF, Aldoory L. **“Wellness champions for change,” a multi-level intervention to improve school-level implementation of local wellness policies: study protocol for a cluster randomized trial**. *Contemp Clin Trials.* (2018.0) **75** 29-39. DOI: 10.1016/j.cct.2018.10.008
70. McKay H, Naylor P-J, Lau E, Gray SM, Wolfenden L, Milat A. **Implementation and scale-up of physical activity and behavioural nutrition interventions: an evaluation roadmap**. *Int J Behav Nutr Physical Activity* (2019.0) **16** 102. DOI: 10.1186/s12966-019-0868-4
71. Locke J, Lee K, Cook CR, Frederick L, Vázquez-Colón C, Ehrhart MG. **Understanding the organizational implementation context of schools: a qualitative study of school district administrators, principals, and teachers**. *School Ment Health.* (2019.0) **11** 379-99. DOI: 10.1007/s12310-018-9292-1
72. Lyon AR, Cook CR, Brown EC, Locke J, Davis C, Ehrhart M. **Assessing organizational implementation context in the education sector: confirmatory factor analysis of measures of implementation leadership, climate, and citizenship**. *Implement Sci.* (2018.0) **13** 5. DOI: 10.1186/s13012-017-0705-6
73. Lyon A, Frazier S, Mehta T, Atkins M, Weisbach J. **Easier said than done: intervention sustainability in an urban after-school program**. *Administ Policy Mental Health Mental Health Services Res.* (2011.0) **38** 504-17. DOI: 10.1007/s10488-011-0339-y
|
---
title: Predictors of sustainment of two distinct nutrition and physical activity programs
in early care and education
authors:
- Taren Swindle
- Laura L. Bellows
- Virginia Mitchell
- Susan L. Johnson
- Samjhana Shakya
- Dong Zhang
- James P. Selig
- Leanne Whiteside-Mansell
- Geoffrey M. Curran
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012648
doi: 10.3389/frhs.2022.1010305
license: CC BY 4.0
---
# Predictors of sustainment of two distinct nutrition and physical activity programs in early care and education
## Abstract
### Introduction
The goal of the present study was to investigate factors associated with sustainment of two evidence-based programs for nutrition promotion in early care and education (ECE) settings – Food Friends (FF) and Together, We Inspire Smart Eating (WISE).
### Materials and methods
In a cross-sectional study design, ECE directors ($$n = 55$$) from centers that had previously been trained in WISE or FF completed a survey. Program-specific measures included Steckler's Perception of Innovations, the Program Sustainability Assessment Tool (PSAT), and the Organizational Readiness for Change Assessment (ORCA). For our primary outcomes, two measures of sustainment were examined: Nutrition Continued Practice (i.e., the use of or general focus on nutrition programs) and Program Fidelity (i.e., how well centers used specific evidence-based practices of WISE or FF). Multiple regression was used to determine the association of these outcomes with program, years since last implementation, and overall scores on predictors. Follow-up correlation analyses were used to investigate outcome relationships with context submeasures due to high intercorrelations between predictor submeasures.
### Results
Nutrition Continued Practice was significantly predicted by program and overall PSAT score. WISE programs had significantly higher Nutrition Continued Practice scores than FF program ($$p \leq 0.03$$). All subscales of the PSAT (e.g., environmental support, funding stability, organizational capacity, program adaptation, communications, and strategic planning) were significantly correlated with Nutrition Continued Practice (all rs > 0.30, all ps < 0.03). Program Fidelity was significantly predicted by PSAT and Steckler Perception of Innovation scores. All subscales of the PSAT were strongly positively correlated with Program Fidelity (all rs > 0.48, all ps < 0.001); relative advantage ($r = 0.54$, $p \leq 0.001$) and level of institutionalization ($r = 0.61$, $p \leq 0.001$) were positively correlated with Program Fidelity.
### Conclusion
This study suggests that factors associated with the continued practice of program principles are partially distinct from those that are associated with the sustainment of specific practices driving program fidelity. Results suggest capacity building strategies may be important for both continued attention to nutrition and physical activity as well as sustaining fidelity to specific evidence-based practices.
## Introduction
Healthy eating (1–3), regular physical activity (3–5), and maintaining a healthy body weight [3, 6, 7] are established preventive measures to curb risk for a range of diseases including cardiovascular diseases, non-alcoholic liver diseases, metabolic syndrome, diabetes, and several cancers. However, most children do not meet recommendations for healthy diet and physical activity (PA) (8–14). Establishing early nutrition and PA habits are important for lifelong health and healthy weight [3, 15]. Early care and education (ECE) environments are promising settings for promoting nutrition and PA for children. In the United States (U.S.), 12.5 million of children under 5 years spend approximately 30 hours in ECE centers per week [16, 17]. In other high-income countries, usage rates are similarly high; $45\%$ of children under 5 years of age in Australia are in childcare [18], and over $80\%$ of children in the European Union receive formal childcare before attending compulsory school [19]. Establishing and sustaining effective programs in ECE settings may have a significant, positive effect on child health.
Sustainability is the endurance of a program after a defined program period and after the ending of external implementation support, which is characterized by (a) the integration of the program in an existing institutional or community system [20, 21] (b) the continuation of the intervention [14, 20], and (c) progress in target behavior, yielding continued gains to the target population [20]. Sustaining programs for promoting child health has proved more challenging than establishing initial implementation of such programs [22, 23]. Specifically, there have been many public health efforts implemented to prevent and control childhood obesity, but lack of sustainment of program/intervention efforts is a major translational issue in public health (23–25). In fact, 40 to $60\%$ of interventions are not sustained after external funding ends (22, 25–29). Implementation science recognizes that closing such gaps in sustainment of programs is crucial to achieve continued benefits for the target population [20, 30] and to maintain community engagement [25, 30].
Reflecting the growing emphasis on sustainability in implementation, there are several theories, models, and frameworks dedicated to understanding this topic [31]. One of the most prominent models, the Dynamic Sustainability Framework (DSF), posits that characteristics that influence program sustainment include internal context (e.g., staff availability, program budget), external context (e.g., political support for a program or for the needs a program serves), and program-specific components (e.g., how fun or engaging a program is perceived to be), and the interaction among these [32]. Recent systematic reviews [23, 25, 33], although not informed by the DSF in their framing and design, have supported the framework by identifying factors that align with key DSF constructs for predicting sustainment in educational settings. Internal contextual barriers to sustainment included lack of staff and staff turnover, time, training, and general financial resources; external contextual barriers included community, political engagement, and parental involvement; program-specific barriers included teacher perceptions of how interesting or fun the program was and how adaptable the program was to individual center needs.
Across these reviews, only two studies were identified that examined sustainment of obesity prevention or nutrition promotion programs in ECE. Whether the general pattern of key factors for sustaining programs holds in the ECE setting is unknown. Ward and colleagues used a mixed methods approach to assess factors related to sustainment of the Healthy Start-Départ Santé intervention program after 2 years of the initial training in 140 ECEs in Canada [34]. Qualitative interviews suggested lack of time, resistance among childcare staff, and low parental involvement as barriers while facilitators included support from policy to implement the program, budget-friendly menu, and staff engagement. In Illinois, U.S., Allar et al. [ 35] investigated the use of a physical activity program (I am Moving, I am Learning) approximately 10 years after initial implementation in Head Start, a government-funded program that serves children from families with low incomes. These authors identified that low equipment requirements, and the fun, flexible nature of the movement program were perceived as contributors to the sustainment of this program by teachers and parents [35]. Additionally, the integration of this program into the regular classroom routine was also identified as being important for sustainment. Despite the importance of sustaining childhood obesity intervention programs and the potential for ECE as a target setting for sustaining such programs, there are limited studies that examine the sustainability of childhood obesity prevention programs in ECE. Investigating factors associated with sustainment in the context of ECE offers opportunity to test empirical theories such as the DSF.
The current study addresses this research gap by identifying barriers and facilitators of the sustainability of two intervention programs in ECEs in the United States: [1] Food Friends® (FF), which includes Fun with New Foods and Get Movin' with Mighty Moves and [2] Together, We Inspire Smart Eating (WISE)®. Both programs have a focus on nutrition; FF also has a PA component. Specifically, the purpose of the present study was to understand sustainment factors associated with continued use of FF and WISE over time, as well as any factors that might be unique to the sustainment of each program. To that end, directors of centers that had or were currently implementing FF and WISE completed a survey that assessed (a) continued attention to nutrition and physical activity support at their center and (b) current FF and WISE fidelity, (c) internal and external contextual factors related to sustainment and program-specific components (e.g., how successful they perceived the program to be at their center, how the program compared to alternative options, how often the program was used at their center).
## Interventions
Food *Friends is* a preschool program implemented mainly in Colorado, U.S. that was designed to address healthful eating behaviors and PA patterns in preschoolers (i.e., children ages 3 to 5). FF includes offering new foods and taste tests over 18 weeks; teachers are trained to role model trying the new foods. There are 8 FFs mascots that introduce children to each food group. FF has a companion program, Mighty Moves, focused on supporting development of motor skills through structured activities, music, and classroom enhancements (e.g., scarves). It was implemented successfully for over 20 years and has been shown to both increase children's willingness to try and consume novel foods (food preference) and improve gross motor performance in the short-term [36, 37] and longitudinally [38, 39].
WISE was similarly designed to increase healthy eating habits in early childhood in children aged three to eight years old across a 9-month school year, although it does not include a physical activity component. WISE includes weekly food experiences and supporting activities that align with ECE educational standards and has been shown to create positive changes in both child and family eating behaviors. These include incorporation of more fruit and vegetables into the diet after experiencing WISE and decreased intake of nutrient-poor foods (e.g., chips, cookies, candies) compared to children not exposed to WISE [40, 41]. WISE has been disseminated since 2012 and continues to be disseminated primarily in Arkansas, US.
## Participant recruitment
Both FF and WISE maintain databases of previously trained ECE centers, which provided the sampling pool for the survey. In total, the WISE database included 209 centers, and the FF database included 212 centers. All centers in the training databases were eligible for survey participation. Directors from each center were invited to complete the survey via email invitation first; these invitations were followed with phone invitations if the email did not receive a response. Our target sample size was 112 (n WISE = 49 and n FF = 63) to provide $80\%$ power to detect medium sized effects and reflect the imbalance of trained centers in each state to date [17]. However, due to recruitment challenges experienced during the COVID-19 pandemic, actual recruitment numbers differed.
Prior to sending email invitations, study staff confirmed email contact information for the site director via website or phone call. Each center director received an initial email invitation to the survey. Centers that did not respond to the initial email invitation or two reminder emails were contacted by phone by trained study staff. Data collection took place between January and September 2021.
The survey was divided into sections that assessed general use of nutrition practices at the center (i.e., first portion) and a section that assessed specific use of either FF or WISE (i.e., second portion). Participants had the option to continue to the FF and WISE specific portion of the survey. Only participants who completed the second portion of the survey were included in the following analysis.
## Survey
The survey was divided into 5 sections: [1] Your Role at the Center [2] Nutrition and Physical Activity at the Center, [3] FF/WISE Programming at the Center, [4] Factors Influencing the Use of FF/WISE at the Center, and [5] What It Is Like at the Center. These sections reflected adaptations of three key measures: Steckler's Perception of Innovations [42], the Organizational Readiness for Change Assessment (ORCA) [43], and the Program Sustainability Assessment Tool (PSAT) [44]. The Steckler measure, consistent with the DSF construct of Intervention, was chosen to measure attitudes toward the innovations broadly (i.e., nutrition and physical activity) and adapted for each program to measure attitudes about FF/WISE specifically. The ORCA measure captured issues relevant to the DSF construct of Practice Setting (e.g., culture, leadership), and the PSAT captured constructs relevant to both the Practice Setting (e.g., organizational capacity) and the Ecological System (e.g., external environmental support). The complete survey is included in Supplementary materials; the survey was estimated to take 30 to 45 minutes to complete. Participants were asked to think about their center when it was operating normally (before COVID-19). A summary of the survey content is provided in Table 1 including the constructs measured in each section of the survey, the number of items per construct, and relevant reliability and validity information. Correlations between measured variables, mean scores and standard deviations can be found in Table 2.
## Director role at the center
In this section, items assessed characteristics of the center and the person completing the survey including: [1] level of involvement in decisions about nutrition and physical activity at the center, [2] years of experience in ECE and at the center, [3] role at the center and years in the role, [4] other roles at the center, [5] whether the program was a Head Start, [6] the center's total capacity and hours of operation, [7] tax status of the center, and [8] school district (if applicable). These items were used to describe survey participants and to screen for eligibility for completing the survey. Individuals with no role in making decisions about nutrition and physical activity at the center were asked to provide an alternate email for the person involved in those decision. At the end of this section, respondents were asked if they wanted to continue the survey.
## Nutrition and physical activity at the center
Items in this section focused on Continued Attention to Nutrition/PA at the center, Concern about Nutrition/PA, Program Component Usage, and Nutrition/PA Training. Continued *Attention is* sustaining attention to the issue (i.e., nutrition/PA through policy and/or resource allocation), even if specific programs/interventions are not sustained per se (e.g., Rate the level of focus for your program for providing children opportunity to try new or unfamiliar foods). Continued Attention items were self-developed based on salient aspects of the training and evidence-base of FF and WISE (e.g., intentional exposures to new foods). Items on Concern about Nutrition/PA were adapted from the Steckler and colleagues measure of Awareness Concern about Prevention scale [42]. Nutrition/PA Training items assessed how frequently in the last 5 years that staff had received training in specific nutrition/PA topics (none to a lot). Nutrition/PA Training items were akin to a checklist, which would preclude internal consistency as an appropriate assessment of reliability. To note, self-developed responses for FF and WISE were developed to capture similar aspects of the programs, and thus the same questions were asked for each program, allowing for data to be aggregated across programs. Sum scores were created for these constructs with higher scores reflecting greater use and training levels.
At the end of this section, participants indicated if they had used FF and WISE in the prior 7 years. Survey items also branched to ask for the number of years the program was used, the most recent year of use of each program, and their role with the program. We also include an open-ended response on reason for discontinuing use. Based on participants' response to the question about use of FF/WISE, the remainder of the survey was specific to their experience with either FF and WISE (i.e., branching logic replaced program names as applicable throughout the remainder of the survey). If the program indicated no use of FF and WISE in the past 7 years, the survey ended.
## Food Friends/WISE programming at the center
In this section, participants provided responses to items about Program Fidelity, as well as Level of Use, Level of Success, Relative Advantage, and Level of Institutionalization which were adapted items from the Steckler Perception of Innovations Measure [37]. Program Fidelity items were designed to mirror that of the published WISE fidelity measure [45] and a corresponding and adapted item set for FF. Items were averaged to get an overall fidelity score. All remaining scales were based on Steckler measures on Perceptions of the Innovation [42]. Level of Use included yes/no questions about integration of the programs into routine and standing curriculum. Level of Success included items rated on a sliding scale from Not at All [0] to Completely (100; e.g., The program met your goals). Relative advantage items ask about perceived effectiveness and quality of the programs and were averaged to create a scale score. Finally, the Level of Institutionalization scale assessed factors associated with integrating the programs into center activities (e.g., weekly classroom schedules, overall curriculum) with ratings on a 4-point scale (Strongly Disagree to Strongly Agree). Scale scores were created by averaging across items; for subscale means and standard deviations, see Table 2.
## Factors influencing the use of Food Friends/WISE at the center
This section included items adapted from the Program Sustainability Assessment Tool (PSAT) [44]. Specifically, items were selected and adapted from the constructs of Environmental Support [e.g., FF/WISE had champions or advocates who garnered additional resources (e.g., food, community, donations)], Funding and Resource Stability [e.g., FF/WISE had sustained funding at your center (e.g., food costs, replacement materials)], Organizational Capacity (e.g., Our center had adequate staff to complete FF/WISE goals.), Program Adaptation [e.g., Our center adapted to changes in the environment for FF/WISE (e.g., turnover, leadership change)], Communications (e.g., Our center promoted FF/WISE in a way that generated interest [e.g., wall displays, parent communications)], and Strategic Planning (e.g., Our center had a long-term sustainability plan for FF/WISE beyond our initial year of implementation). Each of these constructs was captured with 3 items each on a 1 (To Little or No Extent) to 7 (*To a* Very Great Extent) scale.
## What it is like at the center
The final section of the survey included items from the Organizational Readiness for Change Assessment (ORCA) including items on Staff Culture, Opinion Leaders, and General Resources [43]. These questions were rated on a 1 (Strongly Disagree) to 5 (Strongly Agree) scale.
## Cognitive interviewing
The research team conducted 5 cognitive interviews to refine and adapt survey items, three with prior participants in the FF program and two prior participants in the WISE program. For the interviews, one study Principal Investigator (PI) and one research assistant held a video conference with each participant. The participant opened the survey on their personal computer and shared their screen as they completed the survey. The research team invited the participants to talk aloud as the completed the survey, explain the rational for their responses, ask questions about the items and/or instructions, and note any aspects that were confusing or unclear. The researchers documented the items on which participants had comments and questions and asked the participants to suggest improved wording. In addition, the researchers prompted the participants at the end of each page to give feedback about the format, item response options, and instructions. The researchers also monitored for signs of confusion (e.g., excess scrolling, mouse movements) to prompt participants to explain their thought processes. Finally, the researchers asked participants to review the initial survey instructions both before and after completing the survey to improve clarity about the survey's purpose and contents. Improvements were made to the survey iteratively to test changes in wording with subsequent interview participants. The PSAT Partnership and Program Evaluation sub-scales were excluded from the full survey because of confusion and poor performance during cognitive interviews.
## Careless responding
Four measures were used to investigate levels of careless responding to identify problem cases in the data: Mahalanobis Distance, long-string analysis, survey duration, and even-odd consistency [46]. If a participant response was flagged under at least two of the above conditions, their responses were investigated for concordant responding (e.g., their responses to conceptually similar items were checked for consistent responses). Mahalanobis *Distance is* a measure of multivariate normality. Participant response sets (i.e., their pattern of responses to every survey question) were compared to the average response set using a Mahalanobis Distance value, and p-values were generated identifying participants whose response sets were multivariate outliers. Long-string analysis looks for consistent identical responding within surveys (e.g., selecting “Slightly Agree” for ten items in a row), and acceptable cut-off values are determined based on survey design. An even-odd consistency correlation can assess the extent to which participants chose similar answers to even and odd questions within a given survey, with inconsistent responses indicated by low correlation scores (see Supplementary materials for more information on this process).
## Program sustainment
The key outcomes of program sustainability were conceptualized in two ways: Continued Practice (i.e., the use of or general focus on nutrition programs or PA programs at the center) and Program Fidelity (i.e., how well centers used specific evidence-based practices of FF or WISE). Continued Practice was calculated by summing up scores from four measures, described in the Nutrition and Physical Activity at the Center portion of the survey, that capture the extent to which program elements were being used at centers. These included a measure of continued attention to nutrition and PA at the center (e.g., “Rate the level of focus for your program: teaching children about nutrition”), concern about nutrition and physical activity at the center (e.g., “How true are the following statements at your center?: I am concerned with the level of activity children get”), the use of nutrition program components (e.g., “How often do children at your site engage in the following activities: Teacher/adult-led physical activities during outdoor play (like recess)”), and nutrition and PA training (e.g., “How much training content have staff at your center received in the following: portion sizes for children; creating positive mealtimes”). From these items, composite scores for Nutrition Continued Practice and Physical Activity Continued Practice were calculated separately. FF consisted of two programs that were targeted at changing nutrition and physical activity practices in ECE contexts, whereas WISE is only targeting nutrition in ECE. Therefore, central comparisons of continued practice are made on the continued practice of nutrition, and physical activity continued practice is a secondary variable/outcome measured only for FF.
## Program Fidelity
Program Fidelity was calculated by a set of seven items that measure the extent to which each center was following key components of FF or WISE. Participants indicated the extent to which their centers were using key elements of FF or WISE in the last year each program was implemented (e.g., “Used the Food Friends puppets and characters with the lessons” or “Used the Windy Wise mascot with WISE lessons). Responses to the seven items are summed to create the Program Fidelity score.
The regression model predicting Program Fidelity indicated that program, lag, and overall ORCA, Steckler Perception of Innovation, and PSAT scores accounted for a significant amount of variance in Program Fidelity scores (F[5, 45] = 13.31, $p \leq 0.001$, R2 = 0.55). Both the overall PSAT score (β = 0.626, $t = 6.00$, $p \leq 0.001$) and overall Steckler Perception of Innovation (β = 0.219, $t = 2.10$, $$p \leq 0.041$$) were significant, positive predictors of Program Fidelity scores. Program Fidelity scores were significantly and positively correlated with all PSAT subscales: organizational capacity ($r = 0.73$, $p \leq 0.001$), program adaptation ($r = 0.66$, $p \leq 0.001$), communications ($r = 0.66$, $p \leq 0.001$), strategic planning ($r = 0.58$, $p \leq 0.001$), environmental support ($r = 0.57$, $p \leq 0.001$), and funding stability ($r = 0.46$, $p \leq 0.001$). Program Fidelity was significantly correlated with only two of the four Steckler Perceptions of Innovation measures: level of institutionalization ($r = 0.61$, $p \leq 0.001$) and relative advantage ($r = 0.54$, $p \leq 0.001$).
## Analysis plan
The original analysis plan for this survey data indicated that measures of sustainability would be determined based on continued practice and attention to best practices in nutrition education and adherence to specific program elements. Responses to the PSAT, Steckler Perceptions of Innovation, and the ORCA would be used as predictors of these measures of sustainability, as well as the interaction between the subscales of the PSAT measure and lag. The previous analysis plan [17] was altered due to two main factors that emerged from the current data set: lower than anticipated sample size and high intercorrelation between theorized predictors of sustainment. Due to our final sample of $$n = 55$$, regression analysis with the initial number of predictors (all subscales of each measure, lag, program, and interaction between lag and PSAT subscales) were no longer adequately powered. Additionally, both the adapted PSAT (α = 0.97) and Steckler Perception of Innovation (α = 0.89) measures had high internal consistency across items regardless of subscale. When correlations between subscales were investigated, the intercorrelations among subscales within these scales caused substantial multicollinearity issues (i.e., VIF values > 5 and reversal of the direction of bivariate correlations directions vs. beta-weights, see Table 2 for correlations of all variables used in current analysis). For instance, the six subscales of the PSAT measure had intercorrelations ranging from $r = 0.51$ to $r = 0.87.$
Therefore, it was determined multiple regression models with sustainment variables as outcomes, and the overall average scores of the PSAT, Steckler Perceptions of Innovations, and ORCA subscales, program type, and lag entered as predictors would be used to determine which overall measures were predictive of sustainment outcomes. Following these regressions, any overall scale that was predictive of a sustainment outcome would be investigated further by looking at the bivariate correlation between corresponding subscales and the sustainment outcome. Distribution of scores for all subscales and overall measures were investigated to determine if there were significant outlier scores or issues with normality. There were no individual averages for the PSAT, Steckler, or ORCA that were greater than three standard deviations away from the mean, and the Mahalanobis *Distance analysis* described above to investigate careless responding did not identify multivariate outliers among participant response sets. Program (FF or WISE) differences in PSAT, Steckler Perceptions of Innovations, and ORCA subscales were also assessed using MANOVAs in order to determine if there were program-specific differences in these variables. All analyses were conducted in SPSS 27 (Windows, Version 27.0. Armonk, NY: IBM Corp).
## Sample demographics
A total of 105 participants ($$n = 51$$ WISE participants, $$n = 54$$ FF participants) began the survey. Of the 105 individuals that began the survey, 82 ($78\%$) completed the first portion of the study, 58 ($55\%$) proceeded to the end, and three were later removed from the sample due to careless responding. Thus, there were a total of 55 participants (nWISE = 26, nFF= 29) whose responses about their centers were included in the final analysis. Most participants were female ($$n = 52$$, $94.5\%$), White ($$n = 48$$, $87.3\%$; Black = 4, $7.3\%$; American Indian or Alaskan Native = 1, $1.8\%$; missing = 2, $3.6\%$) and non-Hispanic ($$n = 51$$, $92.3\%$; Hispanic = 3, $5.5\%$; missing = 1, $1.8\%$). The average participant age was 49.3 years (SD = 8.9, minimum (min) = 32 years, maximum (max) = 65 plus years; missing = 19). Participants had worked an average of 21.0 years in ECE (SD = 9.4, min = 32 years, max = 45 years; missing = 2) and had worked at their current center for an average of 14.0 years (SD = 8.2, min = 4 years, max = 37 years; missing = 2). Most participants had been in their current role for 0-5 years ($$n = 16$$, $29\%$). Most of the participating centers were not Head Starts $$n = 40$$ ($72.7\%$); served fewer than 100 children; were open 4 or 5 days a week; and were mainly metropolitan (> 50,000 population) and micropolitan (10,000-50,000 population), as determined by U.S. Department of Agriculture Rural-Urban Community Area codes [47]. See Table 3 for a breakdown of center-level demographics by state.
**Table 3**
| Unnamed: 0 | Overall (N, %) | Food Friends (n, %) | WISE (n, %) |
| --- | --- | --- | --- |
| Type of program | Type of program | Type of program | Type of program |
| Head Start | 13 (23.6%) | 4 (13.8%) | 9 (34.6%) |
| Non–Head Start | 40 (72.7%) | 25 (86.2%) | 15 (57.7%) |
| Missing | 2 (3.6%) | 0 (0%) | 2 (7.7%) |
| Number of children served | Number of children served | Number of children served | Number of children served |
| 1–25 children | 12 (21.8%) | 6 (20.7%) | 6 (23.1%) |
| 26–50 children | 10 (18.2%) | 5 (17.2%) | 5 (19.2%) |
| 51–100 children | 17 (30.9) | 12 (41.4%) | 5 (19.2%) |
| 101–200 children | 9 (16.4%) | 4 (13.8%) | 5 (19.2%) |
| Over 200 children | 5 (9.1%) | 2 (6.9%) | 3 (11.5%) |
| Missing | 2 (3.6%) | 0 (0%) | 2 (7.7%) |
| Number of hours open | | | |
| 6 h or less per day | 3 (5.5%) | 1 (3.4%) | 2 (7.7%) |
| 7 to 12 h per day | 49 (89.1%) | 27 (93.1%) | 22 (84.6%) |
| 13 to 18 h per day | 1 (1.8%) | 1 (3.4%) | 0 (0%) |
| Missing | 2 (3.6%) | 0 (0%) | 2 (7.7%) |
| Days per week center is open | Days per week center is open | Days per week center is open | Days per week center is open |
| 2 days | 1 (1.8%) | 1 (3.4%) | 0 (0%) |
| 3 days | 0 (0%) | 0 (0%) | 0 (0%) |
| 4 days | 11 (20.0%) | 9 (31.0%) | 2 (7.7%) |
| 5 days | 41 (74.5%) | 19 (65.5%) | 22 (92.3%) |
| Missing | 2 (3.6%) | 0 (0%) | 2 (7.7%) |
| Tax status | | | |
| Non–profit | 39 (70.9%) | 21 (72.4%) | 18 (69.2%) |
| For–profit | 7 (12.7%) | 5 (17.2%) | 2 (7.7%) |
| Don't Know | 7 (12.7%) | 3 (10.3%) | 4 (15.4%) |
| Missing | 2 (3.6%) | 0 (05) | 2 (7.7%) |
| USDA rural–urban community area classification | USDA rural–urban community area classification | USDA rural–urban community area classification | USDA rural–urban community area classification |
| Metropolitan | 16 (29.1%) | 8 (27.6%) | 8 (30.8%) |
| Micropolitan | 16 (29.1%) | 6 (20.7%) | 10 (38.5%) |
| Small town to rural | 9 (16.4%) | 7 (24.1%) | 2 (7.7%) |
| Missing | 14 (25.5%) | 8 (27.6%) | 6 (23.1%) |
Lag was determined as the number of years since FF or WISE has been implemented at a center. For WISE centers, the mean number of years it had been since implementation was 1.4 years (SD = 1.3 years, min: 0 years, max: 6 years). For FF centers, the mean number of years since last implementation was 3.3 years (SD = 2.4 years, min = 0 years, max = 8 years). The mean difference in lag between FF and WISE was significant (t[53]= 3.51, $p \leq 0.001$). Chi-square tests did not indicate that center demographics or director demographics were significantly associated with survey completion.
## Program differences in predictors of sustainment
Two-way MANOVAs were used to investigate if PSAT, Steckler Perceptions of Innovations, and ORCA subscale measures differed by program type (FF/WISE) after controlling for lag. There were no differences in PSAT scores by program (F[6, 33] = 2.05, $$p \leq 0.087$$, ηp2 = 0.27). Neither the Steckler Perceptions of Innovations (F[4, 28] = 1.33, $$p \leq 0.285$$, ηp2 = 0.16) or ORCA (F[3, 43] = 0.195, $$p \leq 0.899$$, ηp2 = 0.013) subscales differed significantly by program.
## Predictors of sustainment continued practice
The regression model with program, lag, overall ORCA, Steckler Perception of Innovation, and PSAT scores accounted for a significant proportion of variance in Nutrition Continued Practice scores (F[5, 45] = 4.13, $$p \leq 0.004$$, R2 = 0.24; see Table 4). Program was a significant predictor of Nutrition Continued Practice scores (β = −0.32, t = −2.28, $$p \leq 0.028$$). WISE programs reported higher Nutrition Continued Practice ($M = 11.47$, SD = 1.83) compared to FF programs ($M = 10.27$, SD = 2.13). Overall PSAT score was also a significant predictor of Nutrition Continued Practice (β =0.423, $t = 3.11$, $$p \leq 0.003$$). Because of issues with multicollinearity among PSAT subscales, follow-up analyses looking at the relationship between PSAT subscales and Nutrition Continued Practice were performed with simple bivariate correlations. Nutrition Continued Practice was significantly positively correlated with all PSAT subscales: communications ($r = 0.51$, $p \leq 0.001$), funding stability ($r = 0.49$, $p \leq 0.001$), strategic planning ($r = 0.45$, $p \leq 0.001$), organizational capacity ($r = 0.43$, $$p \leq 0.001$$), environmental support ($r = 0.39$, $$p \leq 0.004$$), and program adaptation ($r = 0.34$, $$p \leq 0.01$$). The regression model with program, lag, and overall ORCA, Steckler Perception of Innovation, and PSAT scores did not predict a significant portion of variance in FF-only Physical Activity Continued Practice scores (F[4, 22] = 0.28, $$p \leq 0.89$$, R2 = 0.05).
**Table 4**
| Sustainment outcome | t | p | β | F | df | p.1 | adj. R2 |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Nutrition Continued Practice | Nutrition Continued Practice | Nutrition Continued Practice | Nutrition Continued Practice | Nutrition Continued Practice | Nutrition Continued Practice | Nutrition Continued Practice | Nutrition Continued Practice |
| Overall Model | | | | 4.13 | 5, 45 | 0.004 | 0.24 |
| Program | −2.28 | 0.03 | −0.32 | | | | |
| Lag | 0.21 | 0.16 | 0.21 | | | | |
| PSAT | 3.11 | 0.003 | 0.42 | | | | |
| Steckler Perception of Innovations | 0.51 | 0.61 | 0.07 | | | | |
| ORCA | 0.84 | 0.40 | 0.11 | | | | |
| Physical activity continued practice (CO Only) | Physical activity continued practice (CO Only) | Physical activity continued practice (CO Only) | Physical activity continued practice (CO Only) | Physical activity continued practice (CO Only) | Physical activity continued practice (CO Only) | Physical activity continued practice (CO Only) | Physical activity continued practice (CO Only) |
| Overall model | | | | 0.28 | 4, 22 | 0.89 | 0.05 |
| Lag | −0.68 | 0.50 | −0.15 | | | | |
| PSAT | 0.55 | 0.59 | 0.13 | | | | |
| Steckler Perception of Innovations | 0.07 | 0.95 | 0.014 | | | | |
| ORCA | −0.40 | 0.69 | −0.09 | | | | |
| Program Fidelity | Program Fidelity | Program Fidelity | Program Fidelity | Program Fidelity | Program Fidelity | Program Fidelity | Program Fidelity |
| Overall model | | | | | | | |
| Program | 0.63 | 0.53 | 0.07 | 13.31 | 545 | < 0.001 | 0.55 |
| Lag | −1.21 | 0.23 | −0.14 | | | | |
| PSAT | 6.00 | < 0.001 | 0.63 | | | | |
| Steckler Perception of Innovations | 2.10 | 0.04 | 0.22 | | | | |
| ORCA | 0.61 | 0.54 | 0.06 | | | | |
## Discussion
This study contributes to the limited literature on sustainment of nutrition/PA programs in ECE [25] by examining predictors of sustainment across two nutrition/PA programs in two U.S. locations. Specifically, we examined how indicators of the Dynamic Sustainability Framework constructs were associated with sustainment in the presence of other DSF constructs, answering recent calls to use theory to evaluate the sustainment of interventions [30]. Specifically, our study was able to identify evidence to support the importance of each DSF construct in understanding sustainment, both for sustaining attention to nutrition/PA broadly and to sustaining the programs as designed. Overall, our data suggest that contextual and system factors may be more important for sustainment than characteristics of the intervention.
For the construct of Intervention, perceptions of the innovation were a significant predictor of sustained Program Fidelity but not Continued Attention (either nutrition on PA), providing evidence that program-specific attitudes influence program-specific outcomes. The Steckler constructs of Institutionalization and Relative Advantage were most highly associated with sustained Program Fidelity. That is, perceiving FF/WISE as better than alternative program options and integrating FF/WISE into center schedules, routines, and norms was correlated with programs' continued use of specific program elements (i.e., Program Fidelity). This finding is consistent with a recent review finding perceived benefits and program integration as key factors for sustained implementation of health behavior programs in schools and ECEs [25]. It is also consistent with qualitative research on sustaining IMIL in ECE settings, which identified integration into the curriculum and routine as key for sustainment [35].
We also examined program differences in outcomes to further examine the association of Intervention characteristics with sustainment outcomes. Only one difference between FF/WISE programs was observed; Nutrition Continued Practice was significantly higher for WISE compared to FF after controlling for lag and other predictors. This may be because of the singular focus of WISE on nutrition compared to the dual focus of FF on nutrition and PA. For example, Ward and colleagues found that ECE centers were more likely to maintain healthy eating than physical activity components of their intervention, stating that focusing on both may be a challenge for centers [35]. Overall, the similarities in findings for FF/WISE suggest either true overlap in sustainment related outcomes and predictors despite the program type, lack of power to detect differences, or similarities due to measurement characteristics. Future in-depth qualitative research will explore these possibilities.
Beyond the DSF construct of Intervention, some findings support the association of the Practice Setting and Ecological System with sustainment outcomes. In fact, the overall PSAT score was the most important predictor in the presence of other predictors for both outcomes. Specifically, both Nutrition Continued Practice and Program Fidelity were significantly predicted by overall PSAT scores with high correlations will all PSAT sub-scores. Indicators of the importance of the Practice Setting included moderate to strong correlations between sustainment outcomes and communication, strategic planning, the center's adaptation of programs, and organizational capacity. While communications and planning are potentially malleable targets for supporting sustainment, organizational capacity may be less so. Consistent with a 2020 review by Herlitz et al. of sustainment of public health programming in schools [34], our study suggests that some organizations may be disadvantaged from the outset for achieving sustainment. Specifically, program capacity was an important predicator of sustainability across both programs and both targeted outcomes, consistent with the importance of capacity in prior reviews of sustainment of community-based public health interventions [48] and of health behavior interventions in schools and ECE settings [25]. Prior research has also suggested that adaptation to the local context is key for sustainment of a program as well as sustained impact if fidelity to components are maintained [48]. The self-report nature of our study did not allow us to determine if adaptations were fidelity consistent or inconsistent. In-depth observations at study sites in subsequent research will shed light on this issue. Despite these indicators of the importance of the practice setting, organizational readiness (as measured by the ORCA) was not related to either sustainment outcome in the presence of other predictors in our sample. This is counter to a recent review of health behavior interventions in schools and ECE settings [25], which found organizational readiness to be among the most frequently identified inner context factors important for sustainment.
The importance of the Ecological System was supported with a strong correlation between PSAT Environmental Support and Program Fidelity, a moderate association between PSAT Environmental Support and Nutrition Continued Practice, and moderate associations between PSAT Funding Stability and Program Fidelity and Nutrition Continued Practice. Our findings on the importance of funding are consistent with a review of studies on sustainment of obesity prevention programs in community settings, which identified resources as the most frequently identified factor for sustainment [24]. Shoesmith et al. also identified funding availability as the most frequently cited outer context barrier to sustainment in their review of school and ECE-based health behavior interventions [25]. Funding stability for an ECE program may have direct impact on use of a nutrition/PA program (e.g., purchase of supplies) or indirect impact (e.g., under-staffed, under-resourced work climates). Future research should explore these potential mechanisms. Our data suggest that support beyond funding is also needed. Although our study did not examine nuance in types of environmental support, prior research has identified parent engagement as key to sustainment in the ECE setting [35]. Center leadership and teachers may benefit from an external “pull” from parents to provide this type of programming. Sustainment strategies targeting the ECE Ecological System are limited in the literature and may have value.
Taken together, these results support the importance of all levels of the DSF in understanding sustainment. Specifically, intervention characteristics (e.g., program type, perceptions of innovation), practice setting traits (e.g., organizational capacity, communications), and the ecological system (e.g., environmental support) were important predictors in our study. Although not tested in our study, elements identified by the DSF may be interlinked in complex manners. For example, evidence-based practice integration and continued training over time have been identified as important predictors of sustainment [25, 33], but these activities are more difficult to implement for institutions where financial stability and staffing constraints are more prominent, perhaps linking certain sustainment predictors together via institutional revenue and monetary resources. We were not able to test interactions as expected because of challenges with measuring factors related to sustainment.
## Challenges, limitations, and strengths
The primary challenge we faced in measurement were high intercorrelations between sub-scales of the PSAT in our sample. Specifically, all sub-scales were correlated at or beyond $r = 0.52$, contributing to high variance inflation factors in the proposed analyses and a need for a revised analysis approach. This was a somewhat unexpected finding because original confirmatory factor analyses of the PSAT in over 250 public health programs (e.g., tobacco control, diabetes prevention) supported a factor structure with 8 distinct domains [44]. However, a recent examination of the PSAT in school settings demonstrated an overarching Cronbach's alpha for internal consistency of 0.95 [33], suggesting high overlap between scales much like our sample. Together with the findings of our study, data suggest that the PSAT may need further revision and testing to have appropriate discriminant validity between sub-scales for educational settings. Further, the lack of association between the ORCA constructs and outcomes in our study may suggest need for further measure development/adaptation around organizational readiness for the ECE setting. In future work, a sufficiently powered sample could be used to perform confirmatory factor analyses (CFA) and invariance testing to establish similar performance over various samples for these measures.
The study has additional limitations and strengths. A key limitation is that study recruitment and data collection was conducted during the COVID-19 pandemic. Programs that were able to be reached and participate during this time may differ in systematic ways from programs that were non-responsive. Specifically, it is possible that only more resourced and/or engaged centers were able to respond, which may have truncated the range of variables in our study. This concern is somewhat attenuated by the findings on program capacity's influence on sustainment outcomes, which indicates useful variability was present in the sample. A related limitation is that our sample size did not reach desired numbers for the previously proposed moderation analysis. Based on initial recruitment predictions, it was estimated that approximately $40\%$ of the potential recruitment pool would respond to the director survey ($$n = 150$$; WISE programs = 45, FF programs = 105). We did not reach these numbers, and many programs that started the survey did not complete it in its entirety ($45\%$). Thus, our study was slightly under-powered compared to our original design. Several strengths offset these limitations. First, we were able to collect information about two distinct programs across two U.S. locations. This increases the generalizability of our findings about the key factors associated with sustaining nutrition/PA programs in ECE. We were also able to model wide variation in lag since implementation, despite surprising null findings regarding its predictive power. Finally, our study was able to simultaneously examine multiple domains theorized by the DSF to be associated with sustainment outcomes in an ECE setting. This approach revealed that, for the present sample, contextual and systems characteristics were the most predictive of continued attention to nutrition/PA and specific program practices.
## Implications for future research and practice
Similar to prior systematic reviews [49], our results indicated that organizational capacity and centers' adaptation of programs were strongly correlated with Program Fidelity. Targeted capacity building and intentional local adaptation during the pre-implementation phase may better prepare programs to self-sustain evidence-based practices over time. Partnered approaches to building local capacity are emerging as examples to inform further research in this area [33, 50, 51]. Future research could explore the value of sustainment strategies targeting contextual factors in the pre-implementation and implementation phases for long-term outcomes. Implementation practitioners may see more benefit from advocating for systems changes and addressing contextual challenges than working directly with implementers and the innovation. Additionally, intentional efforts to support centers as they adapt programs may support long-term fidelity and sustainment.
In the presence of a supportive system and stable context or adjacent to addressing these factors, our data particularly support the importance of local perceptions of innovation as an area for future research and practice. In our study, perceiving FF or WISE as being better or more advantageous than other alternatives was related to higher Program Fidelity. Future research could explore the unique value of sustainment strategies that target adopter perceptions of innovations as well as technical assistance or facilitation approaches that provide structured support for ECE centers to integrate innovations into their program goals and schedule, both at the outset and as an ongoing effort. Practitioners may support implementers by directly addressing their thoughts, attitudes, and motivations related to the targeted innovation. These factors should be considered from the outset of program development and initial training.
## Conclusions
Our study supports the importance of each DSF construct in understanding sustainment, both for sustaining attention to nutrition/PA broadly and to sustaining the programs as designed. Further, our data demonstrate that contextual and system factors may be more important for sustainment than characteristics of the intervention. This study also suggests that factors associated with the continued practice of program principles are partially distinct from those that are associated with the sustainment of specific practices driving program fidelity. Thus, capacity building strategies may be important for both continued attention to nutrition and PA as well as sustaining fidelity to specific evidence-based practices.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by UAMS Institutional Review Board, University of Arkansas for Medical Sciences. Written informed consent for participation was not required for this study in accordance with the National Legislation and the Institutional Requirements.
## Author contributions
TS led the drafting and revision of this manuscript. TS and LB led the conception and design of this study in addition to obtaining funding. GC conducted the analyses for the study, contributed to drafting the manuscript, and coordinated the initial submission of this manuscript. DZ contributed to the drafting and editing of this manuscript. SJ contributed to the design of this study and editing of this manuscript. GC and JS contributed to the conception and design of the study and editing of this manuscript. All authors approved the manuscript before submission.
## Funding
The project was funded by NIH R21CA237984 (TS and LB). TS was also supported by the National Institute of General Medical Sciences of the National Institutes of Health under Award Number 5P20GM109096. TS and GC have salary support from NIH UL1 TR003107. TS, GC, DZ, VM, SJ, and JS have salary support from R37CA252113.
## Conflict of interest
Authors LW-M and TS have a financial interest in the technology (WISE) discussed in this presentation/publication. These financial interests have been reviewed and approved in accordance with the UAMS conflict of interest policies. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Author disclaimer
The content is solely the responsibility of the authors and does not necessarily represent the official views of the funding agencies.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frhs.2022.1010305/full#supplementary-material
## References
1. Lohse T, Faeh D, Bopp M, Rohrmann S. **Adherence to the cancer prevention recommendations of the World Cancer Research Fund/American Institute for Cancer Research and mortality: a census-linked cohort**. *Am J Clin Nutr.* (2016.0) **104** 678-85. DOI: 10.3945/ajcn.116.135020
2. Norat T, Aune D, Chan D, Romaguera D. **Fruits and vegetables: updating the epidemiologic evidence for the WCRF/AICR lifestyle recommendations for cancer prevention**. *Cancer Treat Res.* (2014.0) **159** 35-50. DOI: 10.1007/978-3-642-38007-5_3
3. Clinton SK, Giovannucci EL, Hursting SD. *The World Cancer Research Fund/American Institute for Cancer Research Third Expert Report on Diet, Nutrition, Physical Activity, and Cancer: Impact and Future Directions* (2020.0) 663-71
4. Brenner DR, Poirier AE, Grundy A, Khandwala F, McFadden A, Friedenreich CM. **Cancer incidence attributable to inadequate physical activity in Alberta in 2012**. *CMAJ Open.* (2017.0) **5** E338-44. DOI: 10.9778/cmajo.20160044
5. Thomson CA, McCullough ML, Wertheim BC, Chlebowski RT, Martinez ME, Stefanick ML. **Nutrition and physical activity cancer prevention guidelines, cancer risk, and mortality in the women's health initiative**. *Cancer Prevent Res.* (2014.0) **7** 42-53. DOI: 10.1158/1940-6207.CAPR-13-0258
6. Bray GA, Heisel WE, Afshin A, Jensen MD, Dietz WH, Long M. **The science of obesity management: an endocrine society scientific statement**. *Endocr Rev.* (2018.0) **39** 79-132. DOI: 10.1210/er.2017-00253
7. Lauby-Secretan B, Scoccianti C, Loomis D, Grosse Y, Bianchini F, Straif K. **Body Fatness and Cancer — Viewpoint of the IARC Working Group**. *Nw Eng J Med.* (2016.0) **375** 794-8. DOI: 10.1056/NEJMsr1606602
8. Belanger K, Barnes JD, Longmuir PE, Anderson KD, Bruner B, Copeland JL. **The relationship between physical literacy scores and adherence to Canadian physical activity and sedentary behaviour guidelines 11 Medical and Health Sciences 1117 Public Health and Health Services**. *BMC Public Health.* (2018.0) **18** 1-9. DOI: 10.1186/s12889-018-5897-4
9. O'Brien KT, Vanderloo LM, Bruijns BA, Truelove S, Tucker P. **Physical activity and sedentary time among preschoolers in centre-based childcare: a systematic review 11 medical and health sciences 1117 public health and health services**. *Int J Behav Nutr Phys Act.* (2018.0) **15** 1-16. DOI: 10.1186/s12966-018-0745-6
10. Spence AC, Campbell KJ, Lioret S, McNaughton SA. **Early childhood vegetable, fruit, and discretionary food intakes do not meet dietary guidelines, but do show socioeconomic differences and tracking over time**. *J Acad Nutr Diet* (2018.0) **118** 1634-43. DOI: 10.1016/j.jand.2017.12.009
11. Banfield EC, Liu Y, Davis JS, Chang S, Frazier-Wood AC. **Poor adherence to US dietary guidelines for children and adolescents in the national health and nutrition examination survey population**. *J Acad Nutr Diet.* (2016.0) **116** 21-7. DOI: 10.1016/j.jand.2015.08.010
12. Bailey ADL, Fulgoni VL, Shah N, Patterson AC, Gutierrez-Orozco F, Mathews RS. **Nutrient intake adequacy from food and beverage intake of us children aged 1–6 years from nhanes 2001–2016**. *Nutrients.* (2021.0) **13** 1-13. DOI: 10.3390/nu13030827
13. Gu X, Tucker KL. **Dietary quality of the US child and adolescent population: Trends from 1999 to 2012 and associations with the use of federal nutrition assistance programs**. *Am J Clin Nutr.* (2017.0) **105** 194-202. DOI: 10.3945/ajcn.116.135095
14. Cooper BR, Bumbarger BK, Moore JE. **Sustaining evidence-based prevention programs: correlates in a large-scale dissemination initiative**. *Prevent Sci.* (2015.0) **16** 145-57. DOI: 10.1007/s11121-013-0427-1
15. Jackson JK, Jones J, Nguyen H, Davies I, Lum M, Grady A. **Obesity prevention within the early childhood education and care setting: A systematic review of dietary behavior and physical activity policies and guidelines in high income countries**. *Int J Environ Res Public Health* (2021.0) **8** 1-21. DOI: 10.3390/ijerph18020838
16. Laughlin L. *Who's Minding the Kids? Child Care Arrangements* (2011.0)
17. Swindle T, Zhang D, Johnson SL, Whiteside-Mansell L, Curran GM, Martin J. **A mixed-methods protocol for identifying successful sustainability strategies for nutrition and physical activity interventions in childcare**. *Implement Sci Commun.* (2021.0) **12** 1-10. DOI: 10.1186/s43058-021-00108-x
18. **Child Child Care in Australia report March quarter 2020 - Department of Education Skills and Employment Australian Government**
19. **Living conditions in Europe - childcare arrangements - Statistics Explained**
20. Moore JE, Mascarenhas A, Bain J, Straus SE. **Developing a comprehensive definition of sustainability**. *Implementation Sci* (2017.0) **12** 637. DOI: 10.1186/s13012-017-0637-1
21. Johnson K, Hays C, Center H, Daley C. **Building capacity and sustainable prevention innovations: a sustainability planning model**. *Eval Program Plann.* (2004.0) **27** 135-49. DOI: 10.1016/j.evalprogplan.2004.01.002
22. Scheirer MA. **Is sustainability possible? A review and commentary on empirical studies of program sustainability**. *Am J Evaluation.* (2005.0) **26** 320-47. DOI: 10.1177/1098214005278752
23. Mok WKH, Sharif R, Poh BK, Wee LH, Reilly JJ, Ruzita AT. **Sustainability of Childhood Obesity interventions: a systematic review**. *Pakistan J Nutr.* (2019.0) **18** 603-14. DOI: 10.3923/pjn.2019.603.614
24. Shelton RC, Cooper BR, Stirman SW. **The sustainability of evidence-based interventions and practices in public health and health care**. *Annu Rev Public Health.* (2018.0) **39** 55-76. DOI: 10.1146/annurev-publhealth-040617-014731
25. Shoesmith A, Hall A, Wolfenden L, Shelton RC, Powell BJ, Brown H. **Barriers and facilitators influencing the sustainment of health behaviour interventions in schools and childcare services: a systematic review implementation science**. *BioMed Central Ltd.* (2021.0) **18** 34. DOI: 10.1186/s13012-021-01134-y
26. Tibbits MK, Bumbarger BK, Kyler SJ, Perkins DF. **Sustaining evidence-based interventions under real-world conditions: results from a large-scale diffusion project**. *Prevent Sci.* (2010.0) **11** 252-62. DOI: 10.1007/s11121-010-0170-9
27. Walker B, Wolford B, Sasser D, Verbois C, Bell L. **Launching a comprehensive SNAP-Ed social marketing campaign utilizing the cooperative extension model**. *J Nutr Educ Behav.* (2016.0) **48** S84. DOI: 10.1016/j.jneb.2016.04.224
28. Bond GR, Drake RE, McHugo GJ, Peterson AE, Jones AM, Williams J. **Long-term sustainability of evidence-based practices in community mental health agencies**. *Adm Policy Ment Health.* (2014.0) **41** 228-36. DOI: 10.1007/s10488-012-0461-5
29. Savaya R, Spiro S, Elran-Barak R. **Sustainability of social programs: a comparative case study analysis**. *Am J Evaluation.* (2008.0) **29** 478-93. DOI: 10.1177/1098214008325126
30. Walugembe DR, Sibbald S, le Ber MJ, Kothari A. **Sustainability of public health interventions: Where are the gaps?**. *In: Health Research Policy and Systems. BioMed Central Ltd.* (2019.0) **17** 405. DOI: 10.1186/s12961-018-0405-y
31. Nilsen P, Pratkanis A, Leippe M, Baumgardner M, Hardeman W, Jonston M. **Making sense of implementation theories, models and frameworks**. *Implementation Sci.* (2015.0) **10** 53. DOI: 10.1186/s13012-015-0242-0
32. Chambers DA, Glasgow RE, Stange KC. **The dynamic sustainability framework: Addressing the paradox of sustainment amid ongoing change**. *Implementation Sci.* (2013.0) **8** 1-11. DOI: 10.1186/1748-5908-8-117
33. Whelan J, Love P, Millar L, Allender S, Bell C. **Sustaining obesity prevention in communities: a systematic narrative synthesis review**. *Obesity Rev.* (2018.0) **19** 839-51. DOI: 10.1111/obr.12675
34. Ward S, Chow AF, Humbert ML, Bélanger M, Muhajarine N, Vatanparast H. **Promoting physical activity, healthy eating and gross motor skills development among preschoolers attending childcare centers: process evaluation of the Healthy Start-Départ Santé intervention using the RE-AIM framework**. *Eval Program Plann.* (2018.0) **68** 90-8. DOI: 10.1016/j.evalprogplan.2018.02.005
35. Allar I, Jones E, Elliott E, Kristjansson A, Taliaferro A, Mann M. **The perceived impact of i am moving, i am learning on physical activity and family involvement: a preliminary investigation**. *Am J Health Behav* (2017.0) **41** 683-92. DOI: 10.5993/AJHB.41.6.2
36. Bellows LL, Davies PL, Anderson J, Kennedy C. **Effectiveness of a physical activity intervention for Head Start preschoolers: a randomized intervention study**. *Am J Occup Ther [Internet].* (2013.0) **67** 28-36. DOI: 10.5014/ajot.2013.005777
37. Johnson SL, Bellows L, Beckstrom L, Anderson J. **Evaluation of a social marketing campaign targeting preschool children**. *Am J Health Behav.* (2007.0) **31** 44-55. DOI: 10.5993/AJHB.31.1.5
38. Johnson SL, Ryan SM, Kroehl M, Moding KJ, Boles RE, Bellows LL. **A longitudinal intervention to improve young children's liking and consumption of new foods: findings from the Colorado LEAP study**. *Int J Behav Nutr Phys Act*. DOI: 10.1186/s12966-019-0808-3
39. Bellows LL, Davies PL, Courtney JB, Gavin WJ, Johnson SL, Boles RE. **Motor skill development in low-income, at-risk preschoolers: a community-based longitudinal intervention study**. *J Sci Med Sport [Internet].* (2017.0) **20** 997-1002. DOI: 10.1016/j.jsams.2017.04.003
40. Whiteside-Mansell L, Swindle TM. **Evaluation of together we inspire smart eating: pre-school fruit and vegetable consumption**. *Health Educ Res* (2019.0) **34** 62-71. DOI: 10.1093/her/cyy048
41. Whiteside-Mansell L, Swindle T, Davenport K. **Evaluation of “Together, We Inspire Smart Eating” (WISE) nutrition intervention for young children: assessment of fruit and vegetable consumption with parent reports and measurements of skin carotenoids as biomarkers**. *J Hunger Environ Nutr* (2019.0) 521271-11. DOI: 10.1080/19320248.2019.1652127
42. Steckler A, Goodman RM, McLeroy KR, Davis S, Koch G. **Measuring the diffusion of innovative health promotion programs**. *Am J Health Promot.* (1992.0) **6** 214-24. DOI: 10.4278/0890-1171-6.3.214
43. Helfrich CD Li YF, Sharp ND, Sales AE. **Organizational readiness to change assessment (ORCA): Development of an instrument based on the promoting action on research in health services (PARIHS) framework**. *Implementation Sci.* (2009.0) **4** 1-13. DOI: 10.1186/1748-5908-4-38
44. Luke DA, Calhoun A, Robichaux CB, Elliott MB, Moreland-Russell S. **The program sustainability assessment tool: a new instrument for public health programs**. *Prev Chronic Dis [Internet].* (2014.0) **11** 130184. DOI: 10.5888/pcd11.130184
45. Swindle T, Selig J, Rutledge J, Whiteside-Mansell L, Curran G. **Fidelity monitoring in complex interventions: a case study of the WISE intervention**. *Arch Public Health* (2018.0) **76** 53. DOI: 10.1186/s13690-018-0292-2
46. Niessen ASM, Meijer RR, Tendeiro JN. **Detecting careless respondents in web-based questionnaires: Which method to use?**. *J Res Pers.* (2016.0) **63** 1-11. DOI: 10.1016/j.jrp.2016.04.010
47. **USDA ERS -**
48. Herlitz L, MacIntyre H, Osborn T, Bonell C. **The sustainability of public health interventions in schools: a systematic review Implementation Science**. *BioMed Central Ltd.* (2020.0) **17** 8. DOI: 10.1186/s13012-019-0961-8
49. McLoughlin GM, Candal P, Vazou S, Lee JA, Dzewaltowski DA, Rosenkranz RR. **Evaluating the implementation of the SWITCH**. *Int J Behav Nutr Physical Activity.* (2020.0) **17** 1-18. DOI: 10.1186/s12966-020-01070-y
50. Bodkin A, Hakimi S. *Sustainable by Design: A Systematic Review of Factors for Health Promotion Program Sustainability*. DOI: 10.1186/s12889-020-09091-9
51. Brock DJP, Estabrooks PA, Hill JL, Barlow ML, Alexander RC, Price BE. **Building and sustaining community capacity to address childhood obesity: a three-year mixed-methods case study of a community-academic advisory board**. *Fam Community Health.* (2019.0) **42** 62. DOI: 10.1097/FCH.0000000000000212
|
---
title: 'Health Care Expenditures Among Individuals With Chronic Psychotic Disorders
in Ontario: An Analysis Over Time'
authors:
- Claire de Oliveira
- Tomisin Iwajomo
- Paul Kurdyak
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012663
doi: 10.3389/frhs.2022.848072
license: CC BY 4.0
---
# Health Care Expenditures Among Individuals With Chronic Psychotic Disorders in Ontario: An Analysis Over Time
## Abstract
Chronic psychotic disorders are severe and disabling mental disorders associated with poor psychiatric and medical outcomes, and among the most costly mental disorders to treat. Understanding trends in aggregate health care expenditures over time, and respective drivers, can provide relevant insights for decision makers, namely around appropriate allocation of scarce resources within the health care sector. Using administrative health care times series data from Ontario, this analysis examined trends in aggregate public health care expenditures and activity from 2012 to 2019 among all individuals with a diagnosis of a chronic psychotic disorder. Total aggregate health care expenditures for individuals with a chronic psychotic disorder in Ontario increased at a moderate rate over this time period, in line with the growth of the number of people diagnosed, and thus not likely driven by unit costs or resource use. Psychiatric hospitalizations made up the largest share of health care expenditures (~$30\%$). Nonetheless, among all health services, expenditures of acute medical hospitalizations, outpatient prescription drugs and home care saw the largest growth over time. Mean/per capita health care expenditures were greater for females, and increased with age as well as with the presence of comorbidities/chronic conditions. In particular, mean/per capita health care expenditures increased steadily with the number of comorbidities and were highest for individuals with 5 or more comorbidities and those with congestive heart failure, highlighting the ever-increasing importance of addressing physical health conditions among this patient population. These findings will have important implications for decision makers, namely around the appropriate allocation of health care resources for patients with chronic psychotic disorders. Future research should continue to monitor health care expenditures for individuals with chronic psychotic disorders as well as extend this analysis beyond 2019 to understand how the COVID-19 pandemic, and resulting lockdowns, has impacted aggregate health care expenditures and outcomes for patients living with chronic psychotic disorders.
## Introduction
Chronic psychotic disorders are severe and disabling mental disorders associated with poor psychiatric and medical outcomes. Despite affecting only 1–$1.2\%$ of the population, these disorders are associated with high health care costs due to the young age at onset and the need for intensive care over the life course [1]. Understanding trends in health care expenditures over time, and respective drivers, can provide relevant insights for decision makers, namely around appropriate allocation of scarce resources within the health care sector. Unfortunately, little work has examined changes in aggregate health care expenditures for chronic psychotic disorders-related care over time.
In Ontario, Canada, the economic burden of chronic psychotic disorders to the public health care system was estimated to be roughly $2.1 billion CAD in 2012, the equivalent to $3\%$ of the provincial health care budget for that year [1]. It was also found that costs varied greatly by age groups and, in some instances, sex—while psychiatric hospitalizations accounted for most of the cost at younger ages, in particular for males (likely due to the earlier onset of the illness), long-term care and acute medical hospitalizations accounted for most of the cost for older age groups, due to increasing morbidity with age. Ultimately, these findings highlighted the need for health care systems to address both physical and mental illness simultaneously, especially for older patients with chronic psychotic disorders, and to understand how the interplay between mental and physical health contribute to increased costs among patients with chronic psychotic disorders. However, this work only examined data for 1 year and did not examine how/whether health service expenditures changed over time or whether they differed by patient profile, in particular morbidity. This analysis sought to examine trends in total aggregate public health care expenditures and activity in Ontario from 2012 to 2019 for individuals with chronic psychotic disorders to understand how expenditures reflect trends in activity and changes in patient profiles and morbidity.
## Data
This analysis employed health care records housed at ICES (formerly known as the Institute for Clinical Evaluative Sciences) in Toronto, Ontario, and collected through the administration of Ontario's public health care insurance plan. These data include individual-level linkable and longitudinal data on most publicly funded health care services for all legal residents of Ontario, Canada's most populous province. Data on institution-based care are recorded in the Discharge Abstract Database (acute medical hospitalizations and psychiatric hospitalizations in non-psychiatric designated beds), the Ontario Mental Health Reporting System (psychiatric hospitalizations in psychiatric designated beds), the Continuing Care Reporting System (continuing and long-term care), and the National Rehabilitation Reporting System (rehabilitation); data on ambulatory care, such as emergency department visits, are included in the National Ambulatory Care Reporting System. The Ontario Health Insurance Plan claims database captures data on physician visits, including fee-for-service visits and shadow-billed services, as well as laboratory and diagnostic claims. The Ontario Drug Benefit Program claims database includes information on all outpatient prescription drugs dispensed to individuals covered under the provincial public drug plan (i.e., individuals aged 65 years and older, as well as those under the age of 65 years living in a long-term care home, a home for special care or a Community Home for Opportunity, receiving professional home and community care services, enrolled in the Trillium Drug Program, or on social assistance). The Home Care Database records all unique visits provided by home care professionals. The Registered Persons Database, a population-based registry, provides basic demographic data, such as age and sex, on all legal residents of Ontario and their eligibility for public health care insurance. A full description of these databases can be found in Supplementary Table 1. All databases were linked using unique encoded identifiers and analyzed at ICES, whose legal status under Ontario's health information privacy law allows it to collect and analyse demographic and health care data, without consent, for health system evaluation and improvement. The use of health care data in this project was authorized under section 45 of Ontario's Personal Health Information Protection Act.
To construct the times series, all individuals over the age of 15 eligible for public health care insurance ever diagnosed with chronic psychotic disorder were selected on January 1st of each year of the analysis, using a validated algorithm [2]. More specifically, this included anyone who had a hospitalization with a diagnosis of schizophrenia, schizoaffective disorder, and psychosis not otherwise specified in the Discharge Abstract Database (using ICD-10 codes F20, excluding F20.4, F25, and F29) and the Ontario Mental Health Reporting System (using DSM-IV codes 295.* and 298.*) since 1988, and/or 3 physician visits for psychosis-related care within a 3 year window in the Ontario Health Insurance Plan claims database since 1991. Patients with a diagnosis of psychotic disorder not otherwise specified were included in the analysis as evidence suggests that these patients are ultimately diagnosed with schizophrenia or schizoaffective disorder [3]. Once all individuals who were ineligible for public health care insurance were dropped (~$46\%$ of all individuals ever diagnosed with psychosis), those not residing in the province over the analysis period as well as those under the age of 16 (as psychosis is quite rare before this age) and over the age of 105 (as any age >105 is likely an error) were further excluded (where these last 3 exclusions made up <$1\%$ of the remaining sample).
The Johns Hopkins Adjusted Clinical Groups (ACG)® System Version 10 software [4] was used to determine comorbidities, which were estimated through the use of proprietary software and hospitalization and physician billings data. Comorbidities were defined as the more limiting ACG® System Aggregated Diagnosis Group (ADG) categories: ADG 3—Time limited: Major, ADG 4—Time limited: Major—Primary Infections, ADG 9—Likely to Recur: Progressive, ADG 11—Chronic Medical: Unstable, ADG 16—Chronic Specialty: Unstable-Orthopedic, ADG 22—Injuries/Adverse Effects: Major, and ADG 32—Malignancy. The ADGs were calculated at the start of each calendar year using a 2-year look-back period. Chronic conditions were ascertained through existing ICES-derived cohorts and acquired registries, and included asthma [Asthma Database [5]], cancer [Ontario Cancer Registry [6]], chronic obstructive pulmonary disorder [COPD Database [7]], congestive heart failure [Congestive Heart Failure Database [8]], Crohn's/colitis [Ontario Crohn's and Colitis Cohort Database [9]], diabetes [Ontario Diabetes Database [10]], HIV [Ontario HIV Database [11]], hypertension [Ontario Hypertension Database [12]], and rheumatoid arthritis [Ontario Rheumatoid Arthritis Database [13]]. An overview of the patient profile in 2012 can be found in Supplementary Table 2.
## Estimation of Health Care Expenditures
A validated cost algorithm, available at ICES, was employed to estimate all direct patient-level health care costs from the third-party public payer perspective (i.e., the Ontario Ministries of Health and Long-term Care) [14]. The costing methodology defined in the algorithm uses a bottom-up/micro-costing approach to cost services at the individual patient level, which identifies individual episodes of care or utilization in the health care system and attaches costs or amounts paid (or where lacking, prices) to each one. Given Ontario's public health insurance system, prices are rarely set by providers in a private marketplace; therefore, costs/amounts paid by the Ministry of Health were used. In cases where individual unit costs were not available (e.g., long-term care), a top-down approach, which allocates corporate aggregate (i.e., institutional) costs to individual visits or cases/episodes of care, was employed. Further details on the costing methodology can be found elsewhere [14]. The costs captured by the algorithm account for over $90\%$ of costs of all government paid health care [15]. Given the close correspondence between costs and expenditures, the measure of costs of episodes used is indicative of health care expenditures. Health care costs were categorized into the follow health care expenditure categories: acute medical hospitalizations, psychiatric hospitalizations, and other institution-based care (i.e., inpatient rehabilitation, complex continuing care, and long-term care), hospital outpatient clinic visits, emergency department (ED) visits, other ambulatory care (i.e., same-day surgery, cancer clinic visits, and dialysis clinic visits), physician services, outpatient prescription drugs, and home care. All costs were expressed in 2020 Canadian constant dollars.
## Analysis of Annual Health Care Expenditures
One of the main concerns for health policy makers is to understand the drivers in the growth of aggregate health care expenditures. For example, it is important to understand whether expenditures change due to changes in the number of individuals diagnosed with a given condition and accessing health care services or whether the average person is sicker than before and thus using health care more intensively (either through the use of more services in general or using them for longer periods of time, e.g., longer hospitalizations) or whether care has become more costly (i.e., the price of health care has increased over time due to technological advancements).
To ascertain this, health care expenditure profiles (i.e., total and mean/per capita expenditures) were estimated annually from 2012 to 2019 and examined by sex and health service. In addition, total and mean/per capita health care expenditures were compared to health care activity profiles each year, namely the number of individuals receiving treatment and the mean number of episodes of care for the most costly health care encounters, acute medical and psychiatric hospitalizations, as well as the respective mean length of stay. Finally, annual health care expenditures were examined by age and morbidity profiles (i.e., mean/per capita expenditures per number of co-morbidities and per chronic condition). Linear regression models with number of patients/cost as the dependent variables and year as the independent variable were estimated to determine the significance of trends over time. For each slope estimated, the respective p-value was estimated.
## Health Care Expenditure Profiles by Year
Despite some variation between 2012 and 2014, total aggregate health care expenditures for individuals with chronic psychotic disorders increased at a steady rate over time (Figure 1). On average, total health care expenditures increased $19.7\%$ from 2012 to 2019, from $2.7 billion in 2012 to about $3.3 billion in 2019 (in 2020 Canadian dollars) ($$p \leq 0.0002$$). Health care expenditures increased over time for both sexes (males, $$p \leq 0.0008$$; females, $$p \leq 0.0012$$), but were slightly higher for males for all years, though the difference between sexes increased slightly in more recent years. The number of individuals living with chronic psychotic disorders rose steadily from 160,195 individuals in 2012 to 194,335 individuals in 2019 (all, $$p \leq 0.0064$$; males, $p \leq 0.0001$; females, $p \leq 0.0001$) (Figure 2). There were slightly more males than females with chronic psychotic disorders ($p \leq 0001$). Mean/per capita expenditures were relatively constant over time (varying between $16,000 and $17,000), with a slight dip in 2014 (Supplementary Figure 1) and were greater for females compared to males. Broken by type of health service, total health care expenditures were highest for psychiatric hospitalizations, which were about $878 million in 2012, decreasing to $785 million in 2014, and growing slowly until 2019, reaching about $964 million. However, the largest increases in health care expenditures over time were observed for acute medical hospitalizations, outpatient prescription drugs and home care, which increased 51, 42, and $42\%$ ($$p \leq 0.0011$$; $p \leq 0.0001$; $p \leq 0.0001$), respectively, from 2012 to 2019 (Supplementary Figure 2). These trends also held for mean/per capita expenditures (Figure 3), where the majority of health care expenditures were due to psychiatric hospitalizations (~$30\%$), followed by other institution-based care (~$17\%$), acute medical hospitalizations (~12–$15\%$), physician services (~13–$14\%$), and outpatient prescription drugs (~11–$14\%$).
**Figure 1:** *Total health care expenditures for individuals diagnosed with chronic psychotic disorders, 2012-2019, all and by sex. Source: administrative health care data from ICES. The p-values of the trend analysis for all expenditures and expenditures for females and males are as follows: p = 0.0002, p = 0.0012, and p = 0.0008, respectively).* **Figure 2:** *Number of individuals with a chronic psychotic disorder in Ontario, 2012–2019, all and by sex. Source: administrative health care data from ICES. The p-values of the trend analysis for all expenditures and expenditures for females and males are as follows: p = 0.0064, p < 0.0001, and p < 0.0001, respectively); the p-value for the differences between females and males is p < 0.0001.* **Figure 3:** *Mean/per capita health care expenditures for individuals diagnosed with chronic psychotic disorders, 2012–2019, by health service. Source: administrative health care data from ICES. The p-values from the trend analysis for mean/per capita expenditures are as follows: medical hospitalizations, p = 0.0011; psychiatric hospitalizations, p = 0.214; ED visits, p = 0.0093; physician care, p = 0.0003; hospital outpatient clinic visits, p = 0.0059; outpatient prescription drugs, p < 0.001; home care, p < 0.001; other institution-based care, p = 0.0747; other ambulatory care, p = 0.0013.*
## Health Care Activity Profiles by Year
To understand the evolution of mean/per capita expenditures over time, it is important to jointly examine total health expenditures and total number of individuals with chronic psychotic disorders as well as the changes in health activity, where applicable. While total aggregate health care expenditures increased by $19.7\%$ ($$p \leq 0.0002$$), the total number of prevalent cases increased at a similar rate, $21.3\%$ ($$p \leq 0.0064$$) (Figure 4). The mean number of psychiatric hospitalizations from 2021 to 2019 remained relatively stable over time (at about 1.7 episodes per patient) (Supplementary Figure 3); these were slightly higher among females than males. The average length of stay of these hospitalizations also remained relatively constant over time (roughly 12–13 days) and, although similar between males and females, was slightly higher for females (Supplementary Figure 4). The number of acute medical hospitalizations was also relatively stable over time (1.5–1.6 episodes per patient) but was slightly higher for males than females (Supplementary Figure 5), while the average length of stay increased slightly over time (from 15 to 17 days) and was higher for males (Supplementary Figure 6).
**Figure 4:** *Total health care expenditures and number of individuals with a chronic psychotic disorder in Ontario, 2012–2019. Source: administrative health care data from ICES. The p-values from the trend analysis for total expenditures and number of individuals with a chronic psychotic disorder are as follows: p = 0.0002 and p = 0.0064.*
## Health Care Expenditures by Age
Figure 5 provides the mean/per capita health care expenditures by age in 2019, the last year of the analysis. Overall, health care expenditures followed an s-shaped/sigmoidal curve from 16 to 105 years old. Mean/per capita expenditures were roughly $18,000 at age 16 (typically the age of first episode psychosis) and then decreased to just under $10,500 at age 43. Not surprising, mean/per capita expenditures then increased with age, due to higher expenditures of acute medical hospitalizations and other institution-based care (not shown), peaking at just under $42,000 around age 83 (Figure 5) (expenditures at older ages varied quite a bit due to smaller sample sizes). On average, males had higher costs than females across ages, in particular at older ages.
**Figure 5:** *Mean health care expenditures for individuals diagnosed with chronic psychotic disorders, 2012–2019, by age and sex. Source: administrative health care data from ICES.*
## Morbidity Profiles by Year
From 2012 to 2019, on average, each individual had at least one comorbidity/chronic condition, with females having slightly more than males (Supplementary Figures 7, 8). Although the changes over time were quite small, the number of comorbidities and chronic conditions has been increasing slightly over time, which is in line with the increases seen in expenditures of acute medical hospitalizations. Total health care expenditures were higher for individuals with no comorbidities, due to the larger number of healthier individuals living with a chronic psychotic disorder (Supplementary Figure 9). Nonetheless, expenditures generally increased over time for all groups (though expenditures were somewhat constant over time for individuals without comorbidities). Figure 6 depicts the mean/per capita expenditures from 2012 to 2019 by number of comorbidities, where mean/per capita health care expenditures increased with the number of comorbidities. However, mean/per capital health care expenditures remained fairly constant over time (p-values for trends for 0 to 5+ comorbidities were as follows: $$p \leq 0.03$$, $$p \leq 0.0537$$, $$p \leq 0.897$$, $$p \leq 0.7734$$, $$p \leq 0.1959$$, $$p \leq 0.7856$$, respectively). Individuals without comorbidities had mean/per capita expenditures of about $11,000–$12,000, while those with five or more comorbidities had expenditures between $62,000 and $65,000. Total health care expenditures were also higher for individuals with hypertension, diabetes, COPD, and asthma, all chronic conditions, which affect large numbers of individuals; furthermore, total expenditures for these chronic conditions saw the largest increases over the analysis period (Supplementary Figure 10). Mean/per capita expenditures also varied by chronic condition, but mostly remained constant over time, as shown in Figure 7 (where p-values varies between 0.03 and 0.9866). However, individuals with chronic psychotic disorders and congestive heart failure had the largest mean/per capita health care expenditures at $42,000–$45,000 and witnessed the largest increase over time.
**Figure 6:** *Mean/per capita health care expenditures for individuals diagnosed with chronic psychotic disorders, 2012–2019, by number of comorbidities (measured using the Johns Hopkins Aggregated Diagnosis Groups software). Source: administrative health care data from ICES. The p-values of the trends analysis for mean/per capita expenditures are as follows: 0 ADGs, p = 0.03; 1 ADG, p = 0.0537; 2 ADGs, p = 0.897; 3 ADGs, p = 0.7734; 4 ADGs, p = 0.1959; 5+ ADGs, p = 0.7856.* **Figure 7:** *Mean/per capita health care expenditures for individuals diagnosed with chronic psychotic disorders, 2012–2019, by chronic condition. Source: administrative health care data from ICES. The p-values of the trends analysis for mean/per capita expenditures are as follows: asthma, p = 0.4995; cancer, p = 0.1721; COPD, p = 0.6726; congestive heart failure, p = 0.023; Crohn's/colitis, p = 0.1878; diabetes, p = 0.9866; HIV, p = 0.2584; hypertension, p = 0.0928; rheumatoid arthritis, p = 7,172.*
## Discussion
The level of and changes in aggregate health care expenditures is a key concern for policy makers as health care expenditures make up a sizeable portion of government budgets. From 2012 to 2019, total aggregate health care expenditures of individuals with chronic psychotic disorders in Ontario increased at a moderate rate, in line with the number of patients living with a chronic psychotic disorder. However, health care expenditures by type of health service grew at different rates. While total expenditures of psychiatric hospitalizations increased by $10\%$ from 2012 to 2019, total expenditures for acute medical hospitalizations, outpatient prescription drugs and home care increased by 51, 42, and $42\%$ respectively, for the same time period. Moreover, there was some redistribution of costs from 2012 to 2019, where mean expenditures of psychiatric hospitalizations decreased slightly from 32.1 to $29.5\%$ within total mean health care expenditures, while mean expenditures of acute medical hospitalizations and outpatient prescription drug increased from 12.4 to $14.7\%$ and from 10.8 to $13.6\%$, respectively. Females typically had higher mean expenditures, as well as older individuals (as expenditures increased with age), individuals with multimorbidity (in particular, those with more than five comorbidities) and individuals with congestive heart failure.
Previous research from Australia found little change in public health care expenditures for individuals with psychosis over a 10-year period [16]. Instead, the authors found a significant redistribution of expenditures within the public health care sector over time. In particular, they found that inpatient expenditures decreased from 78 to $42\%$ of total health sector costs from 2000 to 2010, while expenditures with ambulatory care (including outpatient and community mental health care) and pharmaceutical treatment almost tripled over the same time period, this last driven by an increase in expenditures of atypical antipsychotic drugs [16]. The rise over time in health care expenditures on anti-psychotic drugs has also been noted in the UK [17]. These findings are generally in line with results of this analysis. For example, the decrease in expenditures for psychiatric hospitalizations over time, alongside the increase in expenditures of outpatient prescription drugs during the same time period, is likely due to the growing trend in Ontario of treating these individuals outside hospital settings.
In tandem, the increase in acute medical hospitalizations expenditures over time can be explained by the increasing comorbidity individuals experience as they age, as found elsewhere. Other work has shown that individuals with more comorbidities have higher expenditures [18]. Research from Ontario has found that the presence of multiple chronic conditions was quite common among high-cost patients with severe mental illness, highlighting the need for quality of care interventions directed at managing psychosis and multimorbidity [19]. The results of this analysis are in line with these findings, highlighting the need to address multimorbidity among patients with chronic psychotic disorders, as chronic physical health conditions are important drivers of health care spending. These findings show that individuals with cardiovascular disease, such as congestive heart failure, have higher mean health care expenditures compared to individuals with other chronic conditions, such as asthma. Patients with severe mental illness, such as psychosis, are less likely to receive standard levels of care for physical diseases, such as diabetes [20], as well as cardiovascular screening and prevention, when compared to a non-psychiatric population (21–23). Moreover, prior work has shown that the leading cause of death for individuals with chronic psychotic disorders is circulatory conditions, whereas for the general population it is cancer [24], suggesting that individuals with chronic psychotic disorders have not benefited from reductions in cardiovascular deaths to the same extent as the general population [25, 26].
These results have important implications for health policy makers as it shows that expenditures have largely increased as a function of the number of patients treated and less so due to increases in unit prices or resources (as evidenced, for example, by the relatively stable number of mean hospitalizations and length of stays over time). This work also provides some insight regarding the appropriate allocation of resources within the health care system for patients living with chronic psychotic disorders. Although not directly examined, on one hand, with further moves toward de-institutionalization (as suggested by the decrease in psychiatric hospitalization expenditures), more resources should likely be allocated to the provision of outpatient care, in particular psychotherapy, which is currently not covered under the Ontario Health Insurance Plan. On the other hand, given the high economic burden of chronic conditions and comorbidities, as evidenced in this analysis, more resources should also be invested in preventive health interventions to reduce the prevalence of obesity, diabetes, and smoking, which tend to be prevalent among these individuals and are predictive of the development of cardiovascular disease, such as congestive heart failure. Prior work has shown that patients with physical and mental co-morbidities are often ill served with potentially severe consequences [1]. Finally, this work also provides important insights regarding cost differences by sex and the need for tailored approaches for females and males.
## Strengths and Limitations
The main contribution to the literature is the analysis of a long time series from a large Canadian province to examine trends in aggregate public health care expenditures among individuals with chronic psychotic disorders, overall and by sex, age, and morbidity. Most studies tend to examine hospital-based samples only [27]; by including individuals who were diagnosed in outpatient settings in addition to hospital settings, this study included a sample representative of the entire population of individuals with a diagnosis of a chronic psychotic disorder. This analysis examined prevalent cohorts over multiple years and made use of the same databases for all years, thus enabling a direct comparison of expenditures over time. Finally, this work examined expenditures for several multimorbidity and chronic condition sub-groups, thus shedding light on the interaction between physical and mental health and its impact on health care expenditures.
The health care expenditures examined in this analysis account for most health care costs paid under a universal health care system. However, these expenditures do not include those of community care, which includes addiction services, and thus are an underestimate of the true health care spending for this patient population. Although psychiatric inpatient care is the mainstay for individuals with severe forms of psychosis, many patients make use of community care, the use of which has likely increased over the last decade in line with research done elsewhere [16]. In addition, although in line with the perspective of this analysis, this analysis only examined outpatient drug expenditures for patients covered under the provincial public insurance plan. Finally, this analysis was done using data up to 2019 and thus did not examine the impact of the COVID-19 pandemic, and related lockdowns, on health care utilization and expenditures for this patient population. It will be important to understand how care has changed due to the pandemic and in particular how the uptake of telepsychiatry has changed health care utilization patterns and outcomes. Unfortunately, the impact of the pandemic on changes in mental health care delivery and utilization in this patient population, and related expenditures, has yet to be examined [28].
## Conclusion
Total aggregate health care expenditures for individuals with a chronic psychotic disorder in Ontario have been increasing at a moderate rate over time, in line with the growth of the number of people living with a chronic psychotic disorder. Although expenditures of psychiatric hospitalizations made up most of total expenditures, expenditures of acute medical hospitalizations, outpatient prescription drugs and home care have grown the most over the last 8 years. Mean/per capita health care expenditures were greater for females, and increased with age as well as with the presence of comorbidities/chronic conditions. In particular, mean/per capita health care expenditures were highest for individuals with five or more comorbidities and those with congestive heart failure, highlighting the ever-increasing importance of addressing physical health conditions among this patient population.
Future research should continue to monitor aggregate health care expenditures for individuals with chronic psychotic disorders but also explore the use of decomposition techniques to better understand changes in expenditures over time. Moreover, future work should seek to extend this analysis beyond 2019 to understand how the COVID-19 pandemic, and resulting lockdowns, has impacted health care utilization, and consequently health care expenditures, as well as outcomes for patients living with chronic psychotic disorders.
## Data Availability Statement
The data analyzed in this study is subject to the following licenses/restrictions: the dataset from this study is held securely in coded form at ICES. While legal data sharing agreements between ICES and data providers (e.g., healthcare organizations and government) prohibit ICES from making the dataset publicly available, access may be granted to those who meet pre-specified criteria for confidential access, available at www.ices.on.ca/DAS (email: [email protected]). The full dataset creation plan and underlying analytic code are available from the authors upon request, understanding that the computer programs may rely upon coding templates or macros that are unique to ICES and are therefore either inaccessible or may require modification. Requests to access these datasets should be directed to [email protected].
## Author Contributions
CdO and PK conceptualized and designed the analysis. TI had access to the data and carried out the analysis, supervised by CdO. CdO drafted the initial manuscript and all authors critically reviewed the manuscript for important intellectual content. All authors interpreted the results and approved the final manuscript as submitted and agree to be accountable for all aspects of the work.
## Funding
This study was funded by the Canadian Institutes of Health Research and supported by ICES, which is funded by an annual grant from the Ontario Ministry of Health and Ministry of Long-term Care. The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. The opinions, results, and conclusions reported in this article are also independent from the other sources that provided data and funding. No endorsement by ICES, the Ontario Ministry of Health and Ministry of Long-term *Care is* intended or should be inferred. Furthermore, parts of this material are based on data and/or information compiled and provided by the Canadian Institute for Health Information (CIHI) and the Ministry of Health. However, the analyses, conclusions, opinions, and statements expressed in the material are those of the authors and not necessarily those of CIHI or the Ministry of Health. In addition, parts of this material are based on data and information provided by Ontario Health (Cancer Care Ontario) (OH [CCO]). The opinions, results, view, and conclusions reported in this article were those of the authors and do not necessarily reflect those of OH (CCO). No endorsement by OH (CCO) was intended or should be inferred.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary Material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frhs.2022.848072/full#supplementary-material
## References
1. de Oliveira C, Cheng J, Rehm J, Kurdyak P. **The economic burden of chronic psychotic disorders in Ontario**. *J Ment Health Policy Econ.* (2016) **19** 181-92. PMID: 27991417
2. Kurdyak P, Lin E, Gnam W, Green D, Vigod S. **Validation of a population-based algorithm to detect chronic psychotic illness**. *Can J Psychiatry.* (2015) **60** 362-8. DOI: 10.1177/070674371506000805
3. Pope MA, Joober R, Malla AK. **Diagnostic stability of first-episode psychotic disorders and persistence of comorbid psychiatric disorders over 1 year**. *Can J Psychiatry.* (2013) **58** 588-94. DOI: 10.1177/070674371305801008
4. 4.The Johns Hopkins ACG Case Mix System Version 9.0 Release Notes (Systems Documentation). Baltimore, MD: Johns Hopkins University Bloomberg School of Public Health (2009).. (2009)
5. Gershon AS, Wang C, Guan J, Vasilevska-Ristovska J, Cicutto L, To T. **Identifying patients with physician diagnosed asthma in health administrative databases**. *Can Respir J.* (2009) **16** 183-8. DOI: 10.1155/2009/963098
6. Prodhan S, King MJ, De P, Gilbert J. **Health services data: the Ontario Cancer Registry (a unique, linked, and automated population-based registry). In: Levy A, Goring S, Editors**. *Data and Measures in Health Services Research. Health Services Research.* (2016) 1-27
7. Gershon AS, Wang C, Guan J, Vasilevska-Ristovska J, Cicutto L, To T. **Identifying individuals with physician diagnosed COPD in health administrative databases**. *COPD.* (2009) **6** 388-94. DOI: 10.1080/15412550903140865
8. Schultz SE, Rothwell DM, Chen Z, Tu K. **Identifying cases of congestive heart failure from administrative data: a validation study using primary care patient records**. *Chronic Dis Inj Can.* (2013) **33** 160-6. DOI: 10.24095/hpcdp.33.3.06
9. Benchimol EI, Guttmann A, Mack DR, Nguyen GC, Marshall JK, Gregor JC. **Validation of international algorithms to identify adults with inflammatory bowel disease in health administrative data from Ontario, Canada**. *J Clin Epidemiol.* (2014) **67** 887-96. DOI: 10.1016/j.jclinepi.2014.02.019
10. Hux JE, Ivis F, Flintoft V, Bica A. **Diabetes in Ontario: determination of prevalence and incidence using a validated administrative data algorithm**. *Diabetes Care.* (2002) **25** 512-6. DOI: 10.2337/diacare.25.3.512
11. Antoniou T, Zagorski B, Loutfy MR, Strike C, Glazier RH. **Validation of case-finding algorithms derived from administrative data for identifying adults living with human immunodeficiency virus infection**. *PLoS One.* (2011) **6** e21748. DOI: 10.1371/journal.pone.0021748
12. Tu K, Chen Z, Lipscombe LL; Canadian Hypertension Education Program Outcomes Research Taskforce. **Prevalence and incidence of hypertension from 1995 to 2005: a population-based study**. *CMAJ* (2008) **178** 1429-35. DOI: 10.1503/cmaj.071283
13. Widdifield J, Bernatsky S, Paterson JM, Tu K, Ng R, Thorne C. **Accuracy of Canadian Health administrative databases in identifying patients with rheumatoid arthritis: a validation study using the medical records of rheumatologists**. *Arthritis Care Res (Hoboken).* (2013) **65** 1582-91. DOI: 10.1002/acr.22031
14. Wodchis WP, Bushmeneva K, Nikitovic M, McKillop I. **Guidelines on person-level costing using administrative databases in Ontario**. *Working Paper Series* (2013)
15. Wodchis WP, Austin PC, Henry DA. **A 3-year study of high-cost users of health care**. *CMAJ.* (2016) **188** 182-8. DOI: 10.1503/cmaj.150064
16. Neil AL, Carr VJ, Mihalopoulos C, Mackinnon A, Lewin TJ, Morgan VA. **What difference a decade? The costs of psychosis in Australia in 2000 and 2010: comparative results from the first and second Australian national surveys of psychosis**. *Aust N Z J Psychiatry.* (2014) **48** 237-48. DOI: 10.1177/0004867413508453
17. Ilyas S, Moncrieff J. **Trends in prescriptions and costs of drugs for mental disorders in England, 1998-2010**. *Br J Psychiatry.* (2012) **200** 393-8. DOI: 10.1192/bjp.bp.111.104257
18. Aragón MJ, Chalkley M, Rice N. **Medical spending and hospital inpatient care in England: an analysis over time**. *Fisc Stud.* (2016) **37** 405-32. DOI: 10.1111/j.1475-5890.2016.12102
19. de Oliveira C, Mason J, Kurdyak P. **Characteristics of patients with mental illness and persistent high-cost status: a population-based analysis**. *CMAJ.* (2020) **192** E1793-E801. DOI: 10.1503/cmaj.200274
20. Kurdyak P, Vigod S, Duchen R, Jacob B, Stukel T, Kiran T. **Diabetes quality of care and outcomes: Comparison of individuals with and without schizophrenia**. *Gen Hosp Psychiatry.* (2017) **46** 7-13. DOI: 10.1016/j.genhosppsych.2017.02.001
21. De Hert M, Correll CU, Bobes J, Cetkovich-Bakmas M, Cohen D, Asai I. **Physical illness in patients with severe mental disorders. I. Prevalence, impact of medications and disparities in health care**. *World Psychiatry.* (2011) **10** 52-77. DOI: 10.1002/j.2051-5545.2011.tb00014.x
22. De Hert M, Dekker JM, Wood D, Kahl KG, Holt RI, Möller HJ. **Cardiovascular disease and diabetes in people with severe mental illness position statement from the European Psychiatric Association (EPA), supported by the European Association for the Study of Diabetes (EASD) and the European Society of Cardiology (ESC)**. *Eur Psychiatry.* (2009) **24** 412-24. DOI: 10.1016/j.eurpsy.2009.01.005
23. Osborn DP, Baio G, Walters K, Petersen I, Limburg H, Raine R. **Inequalities in the provision of cardiovascular screening to people with severe mental illnesses in primary care: cohort study in the United Kingdom THIN Primary Care Database 2000-2007**. *Schizophr Res.* (2011) **129** 104-10. DOI: 10.1016/j.schres.2011.04.003
24. Gatov E, Rosella L, Chiu M, Kurdyak PA. **Trends in standardized mortality among individuals with schizophrenia, 1993-2012: a population-based, repeated cross-sectional study**. *CMAJ.* (2017) **189** E1177-E87. DOI: 10.1503/cmaj.161351
25. Compton MT, Daumit GL, Druss BG. **Cigarette smoking and overweight/obesity among individuals with serious mental illnesses: a preventive perspective**. *Harv Rev Psychiatry.* (2006) **14** 212-22. DOI: 10.1080/10673220600889256
26. Montejo AL. **The need for routine physical health care in schizophrenia**. *Eur Psychiatry.* (2010) **25** S3-S5. DOI: 10.1016/S0924-9338(10)71699-0
27. Pihlajamaa J, Suvisaari J, Henriksson M, Heilä H, Karjalainen E, Koskela J. **The validity of schizophrenia diagnosis in the Finnish Hospital Discharge Register: findings from a 10-year birth cohort sample**. *Nord J Psychiatry.* (2008) **62** 198-203. DOI: 10.1080/08039480801983596
28. Kozloff N, Mulsant BH, Stergiopoulos V, Voineskos AN. **The COVID-19 global pandemic: implications for people with schizophrenia and related disorders**. *Schizophr Bull.* (2020) **46** 752-7. DOI: 10.1093/schbul/sbaa051
|
---
title: 'Initial adaptation of the OnTrack coordinated specialty care model in Chile:
An application of the Dynamic Adaptation Process'
authors:
- PhuongThao D. Le
- Karen Choe
- María Soledad Burrone
- Iruma Bello
- Paola Velasco
- Tamara Arratia
- Danielle Tal
- Franco Mascayano
- María José Jorquera
- Sara Schilling
- Jorge Ramírez
- Diego Arancibia
- Kim Fader
- Sarah Conover
- Ezra Susser
- Lisa Dixon
- Rubén Alvarado
- Lawrence H. Yang
- Leopoldo J. Cabassa
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012675
doi: 10.3389/frhs.2022.958743
license: CC BY 4.0
---
# Initial adaptation of the OnTrack coordinated specialty care model in Chile: An application of the Dynamic Adaptation Process
## Abstract
### Background
In 2005, Chile became the first country in Latin America to guarantee universal free access for the diagnosis and treatment of schizophrenia. A cluster randomized control trial utilizing the Dynamic Adaptation Process framework is underway to adapt and test the OnTrack coordinated specialty care model to provide recovery-oriented, person-centered care by a multidisciplinary team for individuals with first episode psychosis (FEP) in Chile.
### Methods
A qualitative formative research study was conducted to inform the initial adaptation of the OnTrack Chile (OTCH) program. We conducted key informant interviews ($$n = 17$$) with various stakeholders (policymakers; directors/managers of community mental health centers; mental health professionals) and focus group discussions ($$n = 6$$) with individuals with FEP and caregivers ($$n = 35$$ focus group participants total). Data was analyzed using thematic analysis, organized by participants' perspectives on the benefits, barriers, and recommendations for the key principles, multidisciplinary team, psychosocial components, and the training and supervision model of OnTrack.
### Results
Participants expressed enthusiasm and support for OnTrack's recovery-oriented and person-centered principles of care. While many participants lauded the emphasis on shared decision-making and family involvement, some reported reticence, citing that it is culturally normative for patients and families to adopt a passive role in treatment. Peer specialists, and the family psychoeducation and support and supported education and employment components were perceived as aspects that could encourage the promotion of personhood and autonomy development. However, implementation challenges, including the prevailing biomedical approach, professional hierarchy, and the lack of infrastructure, human, and financial resources necessitate some modifications to these aspects. Some mental health professionals further conveyed reservations regarding the perceived hierarchical structure of the supervision model.
### Conclusion
OnTrack represents a shift from a biomedical model to a valued, aspirational, person-centered and culturally responsive model that focuses on recovery, shared decision-making and psychosocial care. With the appropriate governmental and agency-level provision of resources and modifications to some of the program components, particularly regarding the shared decision-making framework, peer specialist, family engagement, and the training supervision model, OTCH could be a transformative program for a more comprehensive, evidence-based care for individuals with FEP in Chile.
## Introduction
Psychotic disorders such as schizophrenia are among the leading causes of disability globally [1, 2]. Despite striking personal and societal costs, recovery rates are low and have not significantly improved in the last five decades [3]. Thus, the implementation of effective treatments is critical to achieving optimal outcomes for individuals with these serious mental illness (SMI) worldwide. In particular, early interventions for first episode psychosis (FEP), the time a person first begins experiencing psychotic symptoms, have proven to yield substantial benefits in clinical and functional recovery [4, 5].
Among various evidence-based treatments for FEP, coordinated specialty care (CSC) is particularly promising [6]. CSC is a team-based, multi-element, recovery-oriented treatment program that provides evidence-based services to adolescents and young adults as soon as possible after FEP onset [7]. Services include pharmacotherapy, individual and group psychotherapy, family psychoeducation and support, supported employment and education, and case management [6]. A team of specialists works with the service user and involves family members to tailor the treatment. This approach has been implemented with notable success in high-income countries such as the U.S. (i.e., NAVIGATE, the Connection Program, OnTrack) and in other countries such as Australia, Canada, United Kingdom, and in Scandinavian countries [7].
## The OnTrack model
OnTrack is an evidence-based CSC intervention that has been successfully implemented across New York (NY) state and nationally since 2013 [8]. The OnTrackNY model consists of a range of evidence-based practices for psychosis delivered by a multidisciplinary team with specialized training, with the primary goal of helping young people experiencing early psychosis achieve their school, work, and relationship goals. In accordance with CSC programs, none of the services are mandatory; rather, the team works with the individual and the family to understand which services will help them achieve their goals, and the model is delivered in a flexible way both in the office and in the community to meet people's needs and preferences. The OnTrackNY team places the person and family at the center of treatment decisions and delivers interventions that are person-centered, recovery-oriented, and culturally resonant, using a shared decision-making (SDM) framework. Evidence-based interventions offered include medication management, primary care coordination, individual and group psychotherapy based on cognitive behavioral interventions, family psychoeducation and support, supported employment and education services, case management, and peer support [8]. Mechanisms for team functioning promote team collaboration, coordination and communication, including time set aside for a weekly team meeting and the ability for team members to deliver joint sessions. Supplementary Box 1 describes the core principles, multidisciplinary team, and psychosocial components of the OnTrackNY model. Teams throughout the U.S. serving individuals from diverse cultural backgrounds have implemented the OnTrackNY model. Adaptations to the team structure, functioning, services offered, and training received have facilitated effective implementation of the model that is responsive to the local contexts and needs. Furthermore, recognizing the OnTrackNY teams' needs for more detailed guidance navigating cultural considerations more effectively when delivering the interventions, the OnTrackNY training team worked collaboratively to develop a training guide, Delivering Culturally Competent Care in FEP, which focused on how culture affects the care of individuals experiencing a FEP and providing best practices [9].
## FEP care in Chile
Chile is one of the first countries in the Global South to provide universal access to FEP services [10]. Historically, the Chilean healthcare system has consisted of public and private financing, insurance, and delivery, with the wealthiest of the population concentrated in the private sector [11]. Consequently, the publicly insured often have inadequate access to and quality of care; namely, considerable proportions of people with FEP and schizophrenia were left untreated due to minimal coverage and high treatment costs [11]. In 2005, Chile underwent a comprehensive public healthcare system reform in which the Garantías Explícitas en Salud (GES) program was implemented, guaranteeing universal free access for the diagnosis and treatment of schizophrenia, including FEP [12, 13].
Although current GES guidelines include the psychosocial approach, such as supportive employment and community reintegration activities, current FEP services in Chile remain predominantly focused on the biomedical approach of providing medications for symptom management and brief visits to the psychiatrist [10]. Furthermore, prior studies have noted important cultural and contextual factors that should be evaluated in the implementation of recovery-oriented, community-based interventions for individuals with SMI in Chile. For example, the hierarchical nature in Chilean social structures can create conflicts between mental healthcare providers with different levels of training and professional status, such as between psychiatrists and non-specialist providers such as community mental health workers and peer support workers [14, 15]. Another prevailing factor is the emphasis on dedication and loyalty to family (“familismo” or family ties), and that family support and acceptance are significant sources of meaning for individuals with FEP as they navigate their recovery [16].
## OTCH and the DAP
A large cluster randomized controlled trial (cRCT) of OnTrack Chile (OTCH) is being implemented to adapt and test the effectiveness of the OnTrack model in this Latin American context (ClinicalTrials.gov #NC T04247711). OnTrack is uniquely positioned among CSC programs due its well-established training infrastructure, high rates of patient engagement, improvements in patient symptom severity and functional outcomes, and track record of scaling up in urban settings [10].
The cRCT is based on the Dynamic Adaptation Process (DAP) [17]. The DAP is derived from a well-known, widely used framework in dissemination and implementation (D&I) science called Exploration, Preparation, Implementation, Sustainment (EPIS), as a way to thoroughly identify and incorporate adaptations at multiple levels, and to facilitate implementation across each phase of EPIS. In contrast to most D&I models, within the DAP, modifications and adaptations are made by a team exclusively devoted to this task known as “Research Adaptation Team,” who is composed of multiple stakeholders and aimed to reflect what was learned about: (a) understanding contextual conditions, and how context might be modified; and (b) how these conditions might modify the nature of the content of the intervention curriculum. In the OTCH trial, the Research Adaptation Team includes trainers from OnTrackNY, local trainers, and the research team. Clinic directors, site staff, and study consultants (e.g., policy makers) are also invited to participate in regular meetings as needed. This team uses a participatory group discussion approach that capitalizes on both researchers' and community stakeholders' knowledge (captured via in-depth interviews and focus groups) to improve the fit between the intervention and the new context, and facilitate the translation of research into practice.
This paper presents findings from the formative research conducted as part of the Preparation stage of the OTCH trial, to understand stakeholders' perspectives on the fit of the OnTrackNY model within the current Chilean mental health care system and FEP services, and to inform the initial adaptations of OTCH. Specifically, we aim to understand stakeholders' perceptions regarding four areas of the OnTrack model: [1] key principles of care (recovery-oriented, person-centered, and culturally competent care, including the SDM framework); [2] multidisciplinary team, including peer specialists; [3] program components promoting community integration (i.e., family involvement, supported education and employment); and [4] training and supervision model.
## Study design and setting
The current study is a content analysis of qualitative research conducted between 2019 and 2020, during the project's Preparation phase. We conducted key informant interviews (KIIs) with providers, administrators, and policymakers, and focus groups (FGs) with patients and caregivers.
Study sites included three of the 20 community mental health centers (CMHC) in Santiago, Chile, that were participating in the cRCT. The 20 participating CMHCs were first divided into two groups based on poverty levels of the catchment area (10 below and 10 above median poverty level). Fieldwork was conducted in two different regions of Chile. The percentage of individuals living below the nationally defined poverty line varies across these municipalities—from $11.6\%$ to $42.4\%$. We included CMHCs from the different areas, which include the poorest populations. Of the first five CMHCs to be included in the trial, two CMHCs were excluded from this formative research component because the local IRB required an evaluation fee. Thus, the qualitative research was conducted in the remaining three CMHCs.
## Participant selection
Recruitment for KII participants aimed at gathering opinions from stakeholders at different levels of decision-making: policymakers (policy level), CMHC managers and directors (organizational level), and mental health professionals (provider level). Potential participants who met the defined inclusion criteria were identified. Of the potential participants, we employed a purposive sampling approach to identify and invite key informants representing each participant group. A total of $$n = 17$$ individuals participated in KIIs (five policymakers; four CMHC directors/managers; eight mental health professionals). At each of the same selected CMHCs we conducted a focus group with individuals with psychosis and a separate focus group with family members. Eligibility criteria for users were: 16–30 years of age and diagnosed with psychosis (without symptoms or with attenuated symptoms). We conducted three FGs with individuals with FEP ($$n = 19$$ participants) and three FGs with caregivers ($$n = 16$$ participants).
## Data collection
We developed semi-structured interview guides based on the research objectives for this phase, as was determined by the research steering committee. Interviews with policymakers, CMHC healthcare managers, and mental health professionals focused on how OTCH could be adapted to follow national legislation; the conditions for implementation and sustainability of OTCH, including questions on pragmatic concerns (e.g., staffing, resources, training) and organizational (e.g., leadership, culture) factors; and perspectives about the OTCH training and supervision model. FGs queried how FEP patients and caregivers view the services provided in Chile, including their strengths and weaknesses, and their perceptions of the OnTrack model and components. Sample topics and questions for the KIIs and FGs are included in Supplemental material 1.
Local researchers (PV, TA) conducted KIIs and FGs. KIIs were held at the participants' or researchers' offices and lasted 64 min on average. FGs were held at the CMHCs and lasted 60 min on average. Interviews were conducted in Spanish, audiotaped, and transcribed verbatim. In addition, all interviews were summarized by the local researchers, and bilingual master's level research assistants translated the summaries into English.
## Data analysis
Data analysis for this study utilized mainly the translated English summaries, although we referred back to the original Spanish transcripts for clarification of codes and text when appropriate. We employed an inductive thematic analysis approach [18], starting with open coding to iteratively develop a codebook, which was then applied and refined through several rounds of consensus coding. Through collaboration and discussion, identified themes and codes were organized into a formal codebook on Microsoft Excel, with separate sheets for the four assessed areas of the OnTrack model (key principles of care, multidisciplinary team, community-based program components, training and supervision approach).
Once the initial codebook was established, pairs of coders were trained prior to coding independently. Groups of at least four U.S. masters-level research assistants met to discuss coding and resolve disagreements by consensus, and if necessary, discussed any remaining coding questions. Online collaborative documents (e.g., Google Docs, Google Sheets) were employed to apply codes to the text (using the “Comment” function) and to keep track of examples of illustrative quotes associated with the codes. Spreadsheet cells were color coded per theme and categorized by the benefits, barriers, and recommendations/adaptations according to participants' perceptions. The U.S. team met weekly over 19 months to conduct consensus coding, and analysis was supervised by experienced qualitative researchers (PTL, LHY).
We used several analytical strategies to ensure the trustworthiness and rigor of our analysis, including developing an audit trail, using multiple coders, and conducting frequent team debriefing meetings [19]. We also presented preliminary analyses to the *Chilean analysis* team and larger OTCH research team to discuss the codebook and the emergent themes. Chilean researchers provided background information and their own analyses to help contextualize the findings. The final round of analysis focused on participants' perceptions of the benefits and barriers of the OnTrack model, specifically in four thematic areas: [1] foundational principles on OnTrack; [2] multidisciplinary team; [3] psychosocial program components; and the [4] training and supervision approach.
## Results
Characteristics of KII and FG participants are included in Table 1. Most of the participants were from Chile, with the exception of one user who was of African descent and one user from Korea. Of the $$n = 19$$ users, all of them were living with a caregiver and all were single or divorced; 10 had a pension from the government, and 6 received economic support from their families.
**Table 1**
| Unnamed: 0 | Policymaker | Policymaker.1 | CMHC managers/ | CMHC managers/.1 | Mental health | Mental health.1 | Service users | Service users.1 | Family members | Family members.1 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| | (n = 5) | (n = 5) | directors (n = 4) | directors (n = 4) | professionals (n = 8) | professionals (n = 8) | (n = 19) | (n = 19) | (n = 16) | (n = 16) |
| Age (mean ± SD) | 39.4 ± 2.5 | 39.4 ± 2.5 | 48 ± 15.1 | 48 ± 15.1 | 38.6 ± 5.8 | 38.6 ± 5.8 | 23.7 ± 3.6 | 23.7 ± 3.6 | 51.1 ± 6.5 | 51.1 ± 6.5 |
| Gender ( n ; %) | Gender ( n ; %) | Gender ( n ; %) | Gender ( n ; %) | Gender ( n ; %) | Gender ( n ; %) | Gender ( n ; %) | Gender ( n ; %) | Gender ( n ; %) | Gender ( n ; %) | Gender ( n ; %) |
| Female | 3 | 60 | 1 | 25 | 3 | 37.5 | 6 | 31.6 | 14 | 87.5 |
| Male | 2 | 30 | 3 | 75 | 5 | 62.5 | 13 | 68.4 | 2 | 12.5 |
| Race/ethnicity ( n ; %) | Race/ethnicity ( n ; %) | Race/ethnicity ( n ; %) | Race/ethnicity ( n ; %) | Race/ethnicity ( n ; %) | Race/ethnicity ( n ; %) | Race/ethnicity ( n ; %) | Race/ethnicity ( n ; %) | Race/ethnicity ( n ; %) | Race/ethnicity ( n ; %) | Race/ethnicity ( n ; %) |
| Black/ | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 5.6 | 0 | 0 |
| Afro-descendent | | | | | | | | | | |
| Hispanic or Latino | 5 | 100 | 4 | 100 | 8 | 100 | 17 | 88.8 | 16 | 0 |
| Asian | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 5.6 | 0 | 0 |
Most participants perceived OnTrackNY as an “ideal” program for patients with FEP, noting that not only does OTCH align with current GES guidelines, but it also offers a comprehensive, multi-faceted approach to FEP care that could facilitate a cultural shift in the way Chileans understand recovery. Many participants, including patients and families, also lauded the various recovery-oriented program components, family involvement and psychoeducation, peer specialists, and supported education and employment. Table 2 presents the summary of the benefits, barriers, and recommendations/adaptations according to participants' perceptions regarding the OnTrack foundational principles, multidisciplinary team approach, psychosocial program components, and the training and supervision model.
**Table 2**
| Unnamed: 0 | Benefits | Barriers | Recommendations |
| --- | --- | --- | --- |
| (1) Foundational principles | (1) Foundational principles | (1) Foundational principles | (1) Foundational principles |
| Recovery-oriented, person-centered, and culturally competent care | • Recovery-oriented and person-centered care can guide FEP individuals towards independence and autonomy• Considering culture of the patient is essential for recovery; incorporating patient-oriented care activities can promote help-seeking behaviors and coping skills | None reported | None reported |
| Shared decision-making | • Engaging patients and families facilitates more tailored and comprehensive treatments | • Mental health professionals may feel that their authority will be undermined• Users and families are used to delegate responsibility to mental health professionals, uncomfortable in taking a more active role in making treatment decisions | • Ensure that patients are given comprehensive overview of the treatment process (e.g., scope of treatment, psychoeducation on medication, shared decision-making approach)• Educate family members on importance of engagement |
| (2) Multidisciplinary team | (2) Multidisciplinary team | (2) Multidisciplinary team | (2) Multidisciplinary team |
| Multidisciplinary team | • Multidisciplinary team members deliver different aspects of treatment (symptom management, psychoeducation, supportive education and employment) | • Some team roles (e.g., employment and education specialists, occupational therapist) currently do not exist• Structural constraints in material resources and time, lack of ability to hire additional staff | • Expand OTCH to cover other mental health conditions, given limited resources and lower prevalence of FEP compared to other mental health conditions (e.g., adapt training curriculum and team model)• Increase support from headquarters to (1) lower level of benefit requirements; 2) hire additional staff |
| Peer support | • Peers' lived experiences can help service users better understand their condition and create deeper connections• Peers can support intervention team implement tailored treatment for users | • Mental health professionals may feel challenged by peers | • Existing infrastructure and resources are not supportive of employing peers |
| (3) Psychosocial program components | (3) Psychosocial program components | (3) Psychosocial program components | (3) Psychosocial program components |
| • Family involvement and family psychoeducation | • Involving family and providing psychoeducation can help improve family's understanding of condition and support for treatment• Engaging families in early stages can help ensure continuity of care and prevent relapse | • Some families may resist due to deference for mental health professionals• Family members often lack financial means to pay for transportation to clinic | • To increase users' receptivity to home visits, involve family members in a discussion, as soon as treatment initiates, on relevance and importance of treatment and home visits• Increase psychoeducation on medication, such as possible side effects• Invest financial resources for patients to visit clinic and for clinical team to make home visits |
| Supported education and employment | • Community-based support can help patients better adhere to treatment, as well as build resilience and self-autonomy | • Some clinicians subscribe to biomedical model, which prioritizes symptom management, and thus, may reject community program• Stigma of mental health at structural- and community-level is barrier for users' social reintegration, leaving patients socially isolated | • Increase support from headquarters• Invest financial and infrastructural resources (i.e., physical space, transportation fund)• Reduce structural and public stigma around mental health (provide community psychoeducation) |
| (4) Training and supervision | (4) Training and supervision | (4) Training and supervision | (4) Training and supervision |
| Overall training and supervision | • Training can equip mental health professionals with knowledge and skills to improve patient care, and grow as professionals• Team performance evaluations can improve theory-practice gap and overall FEP care | • Providers are already overworked, and may not be able to attend training and supervision• Lack of clarity about required time commitment• Providers need to work extra hours to make up lost wages during training hours• High turnover rate, in part due to clinicians being hired with no formal clinical experience | • Raise level of qualification required of clinicians when onboarding• Enhance training and preparation of clinic staff (e.g., in cognitive behavioral therapy)• Incentivize training by offering compensation• Provide greater clarity on (1) case management and delegation; (2) staff roles• Adjust the work plan such that all team members can participate |
| Supervision model | • Supervision sessions can provide feedback to improve implementation• Supervision provides guidance on care process, address challenging cases• Weekly consultation meetings benefits both provider and patient | • Mental health professionals such as psychiatrists are resistant to “supervision” as it undermines their authority | • Adopt a more “horizontal” approach (e.g., peer supervision, train-the-trainer model) |
## Recovery-oriented, person-centered, and culturally responsive care
While acknowledging the challenges shifting from the current biomedical model in usual care, most participants welcomed the key care processes of recovery-oriented, person-centered, and culturally responsive care proposed in OnTrack, perceiving these principles to be novel and integral to achieving recovery for individuals with FEP in Chile: In particular, many mental health professionals particularly appreciated OnTrack's emphasis on tailoring and contextualizing treatment plans according to FEP individuals' unique sociocultural backgrounds, recognizing that this approach will facilitate recovery and community integration: Patients and families in FGs also expressed support for culturally responsive and patient-centered care, especially in having program activities that encourage help-seeking behaviors and help patients develop coping skills. Recovery skills such as self-acceptance was emphasized: “Self-acceptance is key, with that [patients] come alive. I think when they're younger, they have a hard time accepting the illness.” ( Focus group #2.2).
## Shared decision-making
The shared decision-making process, a central tenet of the person-centered approach, was met with mixed opinions. Some participants expressed support as this process could help engage the users and their family members, and thus facilitate the development of more comprehensive treatment plans: However, CMHC managers and directors noted the reality that patients in Chile typically have a passive role in treatment. Thus, the shared decision-making framework was perceived by some mental health professionals, CMHC managers and directors as potentially undermining their authority. Moreover, they shared that some patients may also be uncomfortable playing a more active role in their treatment:
## Multidisciplinary team
Collectively, participants believed the OnTrack multi-component services provided by the interdisciplinary team could facilitate recovery: However, a few policymakers expressed concerns over the diverse competencies required of team members necessary to provide FEP services: Additionally, given structural constraints in program resources and time, some mental health professionals stated that a greater number of trained staff would need to be hired to alleviate the current excessive workload. Fulfilling each position of the team in OTCH would require additional time and training, both of which may not be feasible: Similarly, some FG respondents expressed hesitancy toward including some types of practitioners. For example, while some participants agreed with the inclusion of a general medical doctor, others questioned this team member, stating that there are already doctors in the primary care system.
## Peer specialist
The majority of participants held positive perceptions of the support by peers, recognizing it as an integral component that allowed for culturally responsive care and tailoring to each service user, and even facilitating community integration: Patients and families similarly believed that service users would benefit from talking to peers who could relate to their lived experiences, and that peer support could enhance communications and connections among patients, caregivers, and staff. As a result, treatment adherence would increase, while unhealthy behaviors such as substance use would decrease.
Still, some participants expressed concerns about the inclusion of peers. Some policymakers noted that due to the biomedical model currently in place, professionals in the clinic could feel discredited or challenged by the peers, increasing the risk of prejudice against peers specialists. Some policymakers and mental health professionals also expressed concerns regarding the expenses associated with the recruitment and maintenance of peer support services: Peer support is an important component of the OnTrackNY model, and is consistent with recent efforts to meaningfully engage service users in mental health care. Peer support work can improve clinical as well as psychosocial outcomes (27–30). In Chile, there are also promising evidence regarding the positive aspects of the incorporation of peer support workers in mental health services [14, 15].
Nevertheless, although study participants recognized the value of having peers as part of the multidisciplinary team, many voiced hesitations regarding the feasibility of implementing this aspect in the current context in Chile. Given the lack of readily available peer workforce within the community mental health centers, and the current administrative and legal barriers to hire or include peers, it was decided that it was not currently feasible to include this part of the model in OTCH. To highlight the peer experience, the OTCH team will develop video recordings of individuals with lived experience sharing their recovery stories to use when they are training the OTCH teams. Aspirationally, OTCH teams would start advocating and garnering systems-level support to create a paid workforce that could start working within the team as peer specialists.
## Family psychoeducation and support
Most participants, including policymakers, highlighted the benefits of family involvement in providing more comprehensive care and facilitating sustained engagement throughout the recovery process.
In FGs, patients and families pointed out that the lack of family involvement in usual care often complicated their treatment and relationships with current providers, and expressed their need for support in their own mental health, psychoeducation, and crisis intervention skills to aid their loved ones in the recovery process. CMHC managers and directors also highlighted the involvement of families in a more prominent and stable role in patients' recovery as a significant challenge but essential to improve treatment adherence and reduce relapse.
However, some respondents cited potential resistance to family involvement, given that families typically delegate full responsibility and care to mental health professionals. Furthermore, once in treatment, patients reportedly tend to reduce participation in the program as soon as their symptoms are alleviated. Therefore, mental health providers and administrators expressed that incorporating family members in the early stages of treatment progress could help ensure continuity of care and prevent relapse.
Additionally, CMHC managers and directors also highlighted the tendency for family members to misinterpret symptoms as a reason for delayed FEP treatment. As a result, family psychoeducation was perceived as a particularly proactive component for the adolescent population:
## Supported education and employment
Patients and families often discussed the lack of support in education and employment services as a major deficit in the current mental health care system in Chile. Mental health professionals reported that patients currently do not receive this level of support and face community isolation: Given the difficulties patients often face securing a stable occupation, respondents considered it especially beneficial for patients and families to receive supported education and employment services such as job training, resumé writing, and mock interviews, all of which could better facilitate patients' reintegration into society.
Given the high stigmatization reported both within the clinical setting and in the communities, many mental health professionals perceived community integration as pivotal factors in treatment engagement. Respondents also highlighted the importance of systematically identifying and connecting patients to community-based supports (e.g., community workshops and services), to help patients build resilience and self-autonomy, as well as improve treatment adherence and thereby long-term mental health outcomes. However, some policymakers stated that providers may resist a more community-based approach, given the traditional approach of focusing on symptom reduction: Mental health professionals further described administrative barriers, citing that national guidelines enforced at the regional hospitals were too rigid and that symptom management was prioritized over community work:
## Overall training and supervision approach
Respondents often reported ongoing challenges with inadequate, expensive, or lack of training opportunities in the current system. Given this, many mental health professionals valued the future potential of the OTCH training and supervision program for how it could equip them with the necessary skills and knowledge to improve patient care, develop as professionals, and create broader positive change for FEP care: Some respondents added that team performance evaluations may address gaps in theoretical understanding and clinical practice, improving overall FEP care: However, some policymakers and mental health professionals expressed concerns about implementing the training and supervision program due to financial, infrastructural, personnel, and time limitations. For example, psychiatrists and psychologists may not be available for training and supervision due to their existing responsibilities: “*There is* no time for training and supervision of this program. This health center receives money per hour attended to the patient” (Mental health professional #1). Another respondent added that this could add a new level of stress to already overworked teams: Moreover, participants from all stakeholder groups shared that given the novelty of the OTCH model in Chile, the lack of specificity in time allocation may pose a barrier to its implementation: “*There is* concern in the destination of time and agenda for the organization, and subsequent monitoring of the [training and supervision] structure” (Policymaker #3). Mental health professionals were unclear about the expected time commitment, such as the number of weekly hours required of them, and suggested adding a training mandate and clarifying work hours: “training should be mandatory, and the only way to do it is during work hours” (Mental health professional #1). Participants also recommended to decrease or adjust the workload in the training plan to accommodate their overburdened staff: “It should be ensured that the training strategy is no longer a workload for a team that can often be worn out” (Policymaker #1). One policymaker suggested conducting training satisfaction assessments to ensure the appropriateness of the program's curriculum.
## Supervision
Several participants viewed supervision as an aspect that could support the broader implementation of OTCH: However, many participants noted that in Chile the clinical teams are more accustomed to meeting in teams to collaboratively discuss cases rather than with a “supervisor.” Thus, a hierarchical supervision model created discomfort among those who may feel their performances are in question: “The supervision, reports, would be absolutely rejected by professionals, especially for more experienced psychiatrists, as supervision is not a practice used in Chile” (Policymaker #4). A few mental health professionals also expressed discomfort and fears around being evaluated, especially by an outside entity, and suggested a reframing of the supervision relationship: Mental health professionals explained that they may be more willing to participate in supervision if they feel they are engaging with other team members as equal counterparts and believe their expertise is sought out and respected. A few participants even suggested not using the term “supervision.” One policymaker suggested an alternative format of supervision: “Supervision-related instruction, like existing trainings, could be provided in person through the healthcare system or online.” ( Policymaker #3).
## Discussion
This formative qualitative research study, conducted as part of cRCT employing the DAP framework, uniquely contributes to literature as the trial is one of the first systematic efforts to apply the DAP framework in the Latin American context, and provided perspectives from a variety of stakeholders at different levels of decision-making, including policymakers, directors/managers of community mental health centers, mental health professionals, and individuals with FEP and caregivers. As summarized in Table 2, the first round of stakeholder interviews and discussions yielded extremely useful feedback about the initial perceptions regarding the fit of the OnTrack model in Chile, and some recommendations for its ongoing implementation.
In line with the significant amount of positive outlook that the OnTrack model is receiving throughout the field [20], participants from all stakeholder groups generally perceived that the OnTrack model introduces a novel and aspirational framework of FEP care that has the potential to link patients and their families to early treatment to facilitate recovery. The multi-faceted approach of OnTrack, including its focus on recovery-oriented and patient-centered care, was considered crucial to treatment for users with FEP. From offering a range of evidence-based treatment options from a multidisciplinary team with specialized training, to facilitating family engagement and community reintegration, OnTrack could help empower patients to develop and reach personalized goals, thereby improving treatment adherence and relapse prevention in a culturally responsive manner.
Despite these benefits, specific recommendations and considerations regarding the implementation of OnTrack in the Chile context are proposed (see “Recommendations” column in Table 2). We highlight and discuss four specific areas of adaptations: [1] shared decision-making framework; [2] peer specialist; [3] family engagement; and [4] training and supervision.
## SDM framework
The shared decision-making (SDM) paradigm depends on the treatment team's ability to help confer agency, allowing the client to make treatment choices independently [21]. Clinicians who can show 'partnership' with service users can alleviate fear, empower, increase treatment engagement, and reduce relapse following onset of FEP (22–25). Yet, many mental health professionals and healthcare workers in Chile are already struggling to meet the rigid standards of care, and have not received appropriate training to implement such activities.
In addition, structural barriers (economic, human and infrastructure) inhibit the full acceptance of the recovery-oriented, psychosocial approach of OTCH. Prior studies have found that programs tend to favor traditional medical care components and resist funding for psychosocial activities such as recovery skills workshops, family psychoeducation and support. [ 26] But in order to advance evidence-based care for FEP, substantial investments must be made — particularly, leadership buy-in and infrastructural and financial resources are instrumental. Financial assistance such as providing transportation funds will allow providers to travel to the patients' homes or neighborhoods, which can support community-based reintegration activities.
Furthermore, adaptations specific to the SDM framework have to consider the cultural overlay that impacts how people relate to making decisions about their treatment and the ways in which they have been socialized to be passive recipients of care. Thus, the adaptation team recognized that rather than presenting SDM as an empowering strategy that places the young person in charge of treatment decisions, providers in OTCH teams will have to assess how individual and family preferences impact decision-making and what feels most acceptable. This might mean that for each individual, SDM is used for certain treatment decisions more readily than others. Another adaptation will be to modify the language that is used to describe SDM, to shift from one where the young person is encouraged to be independent from the family (which is a very Western concept), to one that resonates more with people's preferences and their situations and respects the family dynamics as they pertain to decision-making and power structures. At the team level, because of the existing power differential between psychiatrists and other team members, training strategies to help with the implementation of SDM will be modified to initially focus on providers other than psychiatrists. It is possible that non-psychiatric providers might be more open to the concept of SDM and might be early adopters to working under this framework. Furthermore, training materials for OTCH teams would be developed that feature the Chilean team and would be more culturally resonant to the providers.
## Family engagement
Discussions with focus groups alluded to the negative and isolating experiences patients with FEP and their families often face. Patients experienced struggles with confronting stigma, feeling misunderstood, uncertainty about the future, unemployment, and social withdrawal — which can lead to cumulative disabilities. Family members expressed confusion when negotiating their roles in the treatment process, which could delay treatment-seeking among patients. In addition, several mental health professionals cited treatment initiation under GES as a negative experience for families, often marked by hopelessness. This is consistent with studies documenting that entry to care is often delayed and only catalyzed by the emergence of positive symptoms; people commonly experience psychotic symptoms for over a year before initiating treatment [6]. Initial care may occur in the context of crisis (e.g., hospitalization), which can lead to heightened internalized stigma among patients [31], as well as traumatization and diminished hope among caregivers [32, 33].
The psychosocial treatment components of OnTrack, such as individualized goal-setting, psychoeducation, and family involvement, can reduce mental health stigma and delays in initiating care, and increase treatment engagement through a culturally responsive lens [23]. And importantly, engagement of family members is critical to maintaining social connectedness, promoting recovery (e.g., providing emotional and treatment support) as users regain independence, and attaining a normal life after developing psychosis.
Indeed, the OnTrack model promotes and prioritizes family involvement as it is associated with better outcomes. Team members are encouraged to involve families in all treatment decisions and during all phases of care. Although families are central in Chilean culture, there is also a deference to authority including mental health providers; this cultural value places family members in more passive roles. Accordingly, the framework of family empowerment promoted in the OnTrack model may be dissonant with expectations that families have for relating to the team. Several strategies to overcome this have been proposed. For instance, modifications to the content of the family psychoeducation materials are needed, such as including information to educate family members about psychosis using language, concepts and examples that are culturally resonant. There is also a need for the teams to increase their capacity to provide more concrete support and case management for families so that they can more effectively participate in the patient's treatment. This can be achieved by helping the mental health centers and teams develop individualized plans for creating time and space in their workloads and identifying resources that would facilitate the delivery of these types of services in the clinics and in the community.
## Training and supervision model
The OnTrack model recommends a supervision structure that places the Team Leader as the primary clinical supervisor responsible for promoting team collaboration and ensuring that services are delivered in a model-consistent manner. This team-based approach with a centralized supervisory structure is typical of team-based interventions delivered in the U.S. and other countries. Yet this structure seems culturally incongruent to the ways in which mental health professionals are accustomed to functioning in Chile. As such, there is a need to clearly communicate the benefits of supervision, and adapt the supervisory structure so that it becomes more acceptable within the Chilean context. This could be done using a peer supervision or train-the-trainer model that moves away from a hierarchical framework and rather supports mutual accountability and peer discourse for professional development and synergies, and thus ensuring accountability throughout the team.
Furthermore, the implementation of OnTrack in New York State has been overseen by a centralized training team that resides in an intermediary organization. Accordingly, when agencies agree to start an OnTrack program, part of the contractual agreement includes the team's participation in training and technical assistance activities to ensure that fidelity to the model is upheld. Because the OTCH teams do not have protected time to deliver this model and rather it is being retrofitted into an already existing work environment, the barriers and resistance to participating in training and technical assistance activities are often substantial. Mental health professionals report feeling overworked and adding additional meetings for training feels unrealistic. When training the OTCH teams, it will be important to assess the formative training that each provider has (e.g., Occupational Therapists vs. Psychologists) to develop a training approach that meets providers where they are, leverages their strengths and fills knowledge and practice gaps to help ensure that all providers are equipped to deliver the services offered within the model. A training program that provides a professional certificate of completion could serve as a mechanism for meeting continuing educational requirements useful for professional promotion and advancement and therefore increase motivation to participate in the training provided. Additionally, the supervision and training strategy may need to be tailored and individualized at the site level to account for the level of organizational support and resources available. Implementing a fidelity process could also help the team as a whole develop an awareness of how well they are functioning across the defined roles and responsibilities.
## Limitations
Our findings should be considered in light of the study limitations. First, given relatively small sample sizes, the study participants' perspectives may not be representative of all stakeholders. The users included in the study also came from socioeconomically disadvantaged backgrounds and were in the care of family members; thus their ability to fully voice their opinions might have been limited. Second, because the principles and approaches of OnTrack are novel to the Chilean context, participants' perceptions, positive or negative, are anticipated, and not yet derived from actual experiences of implementing the model. Third, data analysis was based on English summaries of the transcripts, which may limit the thoroughness of analytical insights and may have missed cultural nuances during the translation process. However, this method enabled rapid and timely analysis of data to propel the study forward to the following phases.
## Conclusion
The cRCT trial of OTCH represents one of the first systematic efforts to apply the DAP in the Latin American context. This formative research study, conducted in the Preparation phase, assessed stakeholders' perspectives on the acceptability and feasibility of OnTrack's key principles, multidisciplinary team, psychosocial components, and training and supervision model. Our findings indicate that OnTrack Chile signifies a shift from a biomedical model to a person-centered and culturally responsive approach that focuses on recovery, shared decision-making, and psychosocial care. However, we identified potential cultural conflicts that may arise in the implementation of the DSM framework, having peer specialists, family engagement strategies, and the training and supervision model. Proposed initial adaptations regarding these three elements of the OnTrack model have been noted, and many are underway. We will continue to seek and document stakeholders' perspectives as OTCH is being implemented and continuously adapted in the following phases. The study underscores the valuable and essential process of engaging multiple local stakeholders, including the service users, to better understand the contextual and cultural context, and to identify the potential adaptations needed.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors upon request and approval.
## Ethics statement
The studies involving human participants were reviewed and approved by the Instituto de Ciencias de la Salud, Universidad de O'Higgins, Rancagua, Chile. The Ethics Committee waived the requirement of written informed consent for participation.
## Author contributions
MB, IB, PV, TA, FM, MJ, JR, DA, KF, SC, ES, LD, RA, LY, and LC designed, collected data, and provided feedback on analyses. PL, KC, MB, PV, TA, DT, LY, and LC analyzed data and drafted sections of the manuscript. LY and LC reviewed versions of the manuscript. All authors contributed to the article and approved the submitted version.
## Funding
OnTrack *Chile is* funded by the U.S. National Institute of Mental Health (R01MH115502).
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frhs.2022.958743/full#supplementary-material
## References
1. Nielsen RE, Uggerby AS, Jensen SOW, McGrath JJ. **Increasing mortality gap for patients diagnosed with schizophrenia over the last three decades — A Danish nationwide study from 1980 to 2010**. *Schizophrenia Res.* (2013) **146** 22-7. DOI: 10.1016/j.schres.2013.02.025
2. Vos T, Flaxman AD, Naghavi M, Lozano R, Michaud C, Ezzati M, AlMazroa MA. *Years lived with disability (YLDs) for 1160 sequelae of 289 diseases and injuries 1990–2010: a systematic analysis for the Global Burden of Disease Study 2010 Lancet 380.* (2012) 2163-96. DOI: 10.1016/S0140-6736(12)61729-2
3. Jaaskelainen E, Juola P, Hirvonen N, McGrath JJ, Saha S, Isohanni M. **Systematic review and meta-analysis of recovery in Schizophrenia**. *Schizophrenia Bullet.* (2013) **39** 1296-306. DOI: 10.1093/schbul/sbs130
4. Fusar-Poli P, McGorry PD, Kane JM. **Improving outcomes of first-episode psychosis: an overview**. *World Psychiatry.* (2017) **16** 251-65. DOI: 10.1002/wps.20446
5. Correll CU, Galling B, Pawar A, Krivko A, Bonetto C, Ruggeri M. **Comparison of early intervention services vs treatment as usual for early-phase psychosis: a systematic review, meta-analysis, and meta-regression**. *JAMA Psychiatry.* (2018) **75** 555. DOI: 10.1001/jamapsychiatry.2018.0623
6. **First Episode Psychosis**. (2015)
7. **NIMH What is Coordinated Specialty Care (CSC)?**. (2020)
8. Bello I, Lee R, Malinovsky I, Watkins L, Nossel I, Smith T. **OnTrackNY: the development of a coordinated specialty care program for individuals experiencing early psychosis**. *PS.* (2017) **68** 318-20. DOI: 10.1176/appi.ps.201600512
9. Bello I, Malinovsky I, Nossel I, Ngo H, Piscitelli S, Lee R. **Delivering Culturally Competent Care in FEP**. (2018)
10. Alvarado R. *On Track Chile For First Episode Psychosis. clinicaltrialsgov.* (2020)
11. Minoletti A, Soto-Brandt G, Toro O, Irarrázaval M, Zanga R, Alvarado R. **Schizophrenia treatment coverage provided by the public and private health systems of Chile**. *PS.* (2021) **72** 478-81. DOI: 10.1176/appi.ps.202000018
12. Gaspar PA, Castillo RI, Maturana A, Villar MJ, Ulloa K, González G. **Early psychosis detection program in Chile: A first step for the South American challenge in psychosis research**. *Early Interv Psychiatry.* (2019) **13** 328-34. DOI: 10.1111/eip.12766
13. **Guía Clínica: Para el tratamiento de personas desde el primer episodio de Esquizofrenia**. (2016)
14. Agrest M, Le PD, Yang LH, Mascayano F, Alves-Nishioka S, Dev S. **Implementing a community-based task-shifting psychosocial intervention for individuals with psychosis in Chile: Perspectives from users**. *Int J Soc Psychiatry.* (2019) **65** 38-45. DOI: 10.1177/0020764018815204
15. Le PD, Agrest M, Mascayano F, Dev S, Kankan T, Dishy G. **Understanding users' perspectives of psychosocial mechanisms underpinning peer support work in Chile**. *Commun Ment Health J.* (2022) **58** 111-20. DOI: 10.1007/s10597-021-00800-1
16. Becker TD, Blasco D, Burrone MS, Dishy G, Velasco P, Reginatto G. **Stigma towards psychosis in urban Chile: engaging “What Matters Most” to resist stigma through recovery-oriented services**. *Psychiatric Rehab J in press.* (2022)
17. Aarons GA, Green AE, Palinkas LA, Self-Brown S, Whitaker DJ, Lutzker JR. **Dynamic adaptation process to implement an evidence-based child maltreatment intervention**. *Implementation Sci.* (2012) **7** 32. DOI: 10.1186/1748-5908-7-32
18. Braun V, Clarke V. *Thematic Analysis: A Practical Guide to Understanding and Doing* (2021)
19. Creswell JW. *Research Design: Qualitative, Quantitative, and Mixed Methods Approaches* (2009)
20. Mascayano F, Bello I, Andrews H, Arancibia D, Arratia T, Burrone MS. **OnTrack Chile for people with early psychosis: a study protocol for a hybrid type 1**. *Trial.* (2022) **23** 1-17. DOI: 10.21203/rs.3.rs-350678/v1
21. Elwyn G, Frosch D, Thomson R, Joseph-Williams N, Lloyd A, Kinnersley P. **Shared decision making: a model for clinical practice**. *J Gen Intern Med.* (2012) **27** 1361-7. DOI: 10.1007/s11606-012-2077-6
22. Allard J, Lancaster S, Clayton S, Amos T, Birchwood M. **Carers' and service users' experiences of early intervention in psychosis services: Implications for care partnerships**. *Early Int Psychiatry.* (2018) **12** 410-6. DOI: 10.1111/eip.12309
23. O'Toole MS, Ohlsen RI, Taylor TM, Purvis R, Walters J, Pilowsky LS. **Treating first episode psychosis–the service users' perspective: a focus group evaluation**. *J Psychiatr Ment Health Nurs.* (2004) **11** 319-26. DOI: 10.1111/j.1365-2850.2004.00730.x
24. Tindall R, Francey S, Hamilton B. **Factors influencing engagement with case managers: Perspectives of young people with a diagnosis of first episode psychosis**. *In J Mental Health Nurs.* (2015) **24** 295-303. DOI: 10.1111/inm.12133
25. Van Schalkwyk GI, Davidson L, Srihari V. **Too late and too little: narratives of treatment disconnect in early psychosis**. *Psychiatr Q.* (2015) **86** 521-32. DOI: 10.1007/s11126-015-9348-4
26. Bartels S, Brunette M, Aschbrenner K, Daumit G. **Implementation of a system-wide health promotion intervention to reduce early mortality in high risk adults with serious mental illness and obesity**. *Imp Sci.* (2015) **10** A15. DOI: 10.1186/1748-5908-10-S1-A15
27. Cabassa LJ, Camacho D, Vélez-Grau CM, Stefancic A. **Peer-based health interventions for people with serious mental illness: a systematic literature review**. *J Psychiatr Res.* (2017) **84** 80-9. DOI: 10.1016/j.jpsychires.2016.09.021
28. Fuhr DC, Salisbury TT, De Silva MJ, Atif N, van Ginneken N, Rahman A. **Effectiveness of peer-delivered interventions for severe mental illness and depression on clinical and psychosocial outcomes: a systematic review and meta-analysis**. *Soc Psychiatry Psychiatr Epidemiol.* (2014) **49** 1691-702. DOI: 10.1007/s00127-014-0857-5
29. Fuhr DC, Weobong B, Lazarus A, Vanobberghen F, Weiss HA, Singla DR. **Delivering the thinking healthy programme for perinatal depression through peers: an individually randomised controlled trial in India**. *Lancet Psychiatry.* (2019) **6** 115-27. DOI: 10.1016/S2215-0366(18)30466-8
30. Sikander S, Ahmad I, Atif N, Zaidi A, Vanobberghen F, Weiss HA. **Delivering the thinking healthy programme for perinatal depression through volunteer peers: a cluster randomised controlled trial in Pakistan**. *Lancet Psychiatry.* (2019) **6** 128-39. DOI: 10.1016/S2215-0366(18)30467-X
31. Yang LH, Blasco D, Lieff SA, Le PD, Li Y, Broeker M. **Stigma of treatment stages for first-episode psychosis: a conceptual framework for early intervention services**. *Harv Rev Psychiatry.* (2021) **29** 131-41. DOI: 10.1097/HRP.0000000000000288
32. Corcoran C, Gerson R, Sills-Shahar R, Nickou C, McGlashan T, Malaspina D. **Trajectory to a first episode of psychosis: a qualitative research study with families**. *Early Interv Psychiatry.* (2007) **1** 308-15. DOI: 10.1111/j.1751-7893.2007.00041.x
33. Tindall RM, Simmons MB, Allott K, Hamilton BE. **Essential ingredients of engagement when working alongside people after their first episode of psychosis: a qualitative meta-synthesis**. *Early Interv Psychiatry.* (2018) **12** 784-95. DOI: 10.1111/eip.12566
|
---
title: Do the Expert Recommendations for Implementing Change (ERIC) strategies adequately
address sustainment?
authors:
- Nicole Nathan
- Byron J. Powell
- Rachel C. Shelton
- Celia V. Laur
- Luke Wolfenden
- Maji Hailemariam
- Sze Lin Yoong
- Rachel Sutherland
- Melanie Kingsland
- Thomas J. Waltz
- Alix Hall
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012683
doi: 10.3389/frhs.2022.905909
license: CC BY 4.0
---
# Do the Expert Recommendations for Implementing Change (ERIC) strategies adequately address sustainment?
## Abstract
### Background
Sustainability science is an emerging area within implementation science. There is limited evidence regarding strategies to best support the continued delivery and sustained impact of evidence-based interventions (EBIs). To build such evidence, clear definitions, and ways to operationalize strategies specific and/or relevant to sustainment are required. Taxonomies and compilations such as the Expert Recommendations for Implementing Change (ERIC) were developed to describe and organize implementation strategies. This study aimed to adapt, refine, and extend the ERIC compilation to incorporate an explicit focus on sustainment. We also sought to classify the specific phase(s) of implementation when the ERIC strategies could be considered and applied.
### Methods
We used a two-phase iterative approach to adapt the ERIC. This involved: [1] adapting through consensus (ERIC strategies were mapped against barriers to sustainment as identified via the literature to identify if existing implementation strategies were sufficient to address sustainment, needed wording changes, or if new strategies were required) and; [2] preliminary application of this sustainment-explicit ERIC glossary (strategies described in published sustainment interventions were coded against the glossary to identify if any further amendments were needed). All team members independently reviewed changes and provided feedback for subsequent iterations until consensus was reached. Following this, and utilizing the same consensus process, the Exploration, Preparation, Implementation and Sustainment (EPIS) Framework was applied to identify when each strategy may be best employed across phases.
### Results
Surface level changes were made to the definitions of 41 of the 73 ERIC strategies to explicitly address sustainment. Four additional strategies received deeper changes in their definitions. One new strategy was identified: Communicate with stakeholders the continued impact of the evidence-based practice. Application of the EPIS identified that at least three-quarters of strategies should be considered during preparation and implementation phases as they are likely to impact sustainment.
### Conclusion
A sustainment-explicit ERIC glossary is provided to help researchers and practitioners develop, test, or apply strategies to improve the sustainment of EBIs in real-world settings. Whilst most ERIC strategies only needed minor changes, their impact on sustainment needs to be tested empirically which may require significant refinement or additions in the future.
## Introduction
Over the last two decades, research investment in, and application of, implementation science theories, frameworks and methods has resulted in significant improvements in the initial implementation of evidence-based interventions (EBIs) in both clinical and community settings (1–3). Key to advancing the field has been the concerted efforts, particularly in the last few years, to identify effective implementation strategies (and the mechanisms through which they operate) (4–7). Implementation strategies are “methods or techniques used to improve the adoption, implementation, sustainment and scale-up of interventions.” [ 3, 8], Systematic reviews of implementation trials have assessed the impact implementation strategies have had on the adoption and implementation of EBIs in real world settings (2, 3, 9–11).
Poor and inconsistent reporting of implementation strategies has been a longstanding issue for the field [8]. Historically, the language used to define implementation strategies has been inconsistent and highly variable [12, 13], with different terms used to describe the same strategy or the same terms being used to define different strategies [13, 14]. Consequently, descriptions of implementation strategies have lacked the necessary detail required for an adequate understanding of the exact nature, function, and make-up of an implementation intervention (i.e., combination of one or more implementation strategies used to support the delivery of an evidence-based practice, program or intervention) (12, 14–16). Such information is essential for scientific advancement, as it allows for replication in advancing the science and improvements of previous research, as well as for scale-up and translation of effective strategies into practice beyond the initial site [14]. These inconsistencies make it difficult to identify core functions of the implementation intervention or the implementation strategies, to synthesize research findings, and ultimately identify the active components of a particular implementation intervention. This problem is especially true for complex, multicomponent implementation interventions such as those typically employed in clinical and public health [14].
The introduction and application of taxonomies or compilations of implementation strategies and behavior change techniques is one approach that has been used to address such issues (12, 13, 17–20). Compilations standardize the naming and definitions of implementation strategies, enabling implementation interventions to be described in a consistent manner. A number of implementation-specific taxonomies and compilations have been developed to standardize and clarify the classification and reporting of implementation strategies (8, 11, 13, 17–19). The Expert Recommendations for Implementing Change (ERIC) compilation [8, 13] has been widely used in health and public health and has provided much-needed common terminology for implementation strategies. Developed and refined by implementation experts, the compilation shows high face validity and consists of 73 strategies grouped into nine categories (see Table 1) [21].
**Table 1**
| Conceptual strategy category from original ERIC compilation (20) | Strategy number from original ERIC compilation (20) | Strategy name | Strategy definition (8, 13) | Specific phase(s) when the strategies could be considered and applied |
| --- | --- | --- | --- | --- |
| Use evaluative and iterative strategies | 4 | Assess for readiness and identify barriers and facilitators | Assess various aspects of an organization and the broader context to determine its degree of readiness to implement and sustain, barriers that may impede implementation and sustainment, and strengths that can be used in the implementation and sustainment effort | Preparation, implementation and Sustainment |
| | 5 | Audit and provide feedback | Collect and summarize clinical performance data over a specified time period and give it to clinicians and administrators to monitor, evaluate, and modify provider behavior | Preparation, Implementation and Sustainment |
| | New sustainment strategy | Communicate with stakeholders the continued impact of the EBP | Communicate data to external stakeholders, end-users and consumers to demonstrate the ongoing benefit, cost effectiveness or return on investment of the innovation with continued implementation. | Implementation, Sustainment |
| | 14 | Conduct cyclical small tests of change | Implement changes in a cyclical fashion using small tests of change before taking changes system-wide. Tests of change benefit from systematic measurement, and results of the tests of change are studied for insights on how to do better. This process continues serially over time, and refinement is added with each cycle | Implementation and Sustainment |
| | 18 | Conduct local needs assessment | Collect and analyze data related to the initial and ongoing need for and fit of the innovation | All phases |
| | 23 | Develop a formal implementation blueprint | Develop a formal implementation blueprint that includes all goals and strategies. The blueprint should include the following: (1) aim/purpose of the implementation; (2) scope of the change (e.g., what organizational units are affected); (3) timeframe and milestones; and (4) appropriate performance/progress measures; (5)plan for maintenance and sustainment of the EBI once it has been implemented. Use and update this plan to guide the implementation effort over time | Preparation, Implementation and Sustainment |
| | 61 | Stage implementation scale up | Phase implementation efforts by starting with small pilots or demonstration projects and gradually move to a system wide rollout while sustaining delivery of the EBP in the original sites | Implementation, Sustainment |
| | 26 | Develop and implement tools for quality monitoring | Develop, test, and introduce into quality-monitoring systems the right input—the appropriate language, protocols, algorithms, standards, and measures (of processes, patient/consumer outcomes, and implementation outcomes) that are often specific to the innovation being implemented and sustained | Preparation, Implementation and Sustainment |
| | 27 | Develop and organize quality monitoring systems | Develop and organize systems and procedures that monitor clinical processes and/or outcomes for the purpose of quality assurance and improvement | Preparation, implementation and sustainment |
| | 46 | Obtain and use patients/consumers and family feedback | Develop strategies to increase patient/consumer and family feedback on the implementation and sustainment effort | Preparation, implementation and Sustainment |
| | 56 | Purposely reexamine the implementation | Monitor progress and adjust clinical practices and implementation strategies to continuously improve the quality of care | Implementation and Sustainment |
| | 8 | Centralize technical assistance | Develop and use a centralized system to deliver technical assistance focused on implementation and sustainment issues | Preparation and Implementation and Sustainment |
| Provide interactive assistance | 33 | Facilitation | A process of interactive problem solving and support that occurs in a context of a recognized need for improvement and a supportive interpersonal relationship | All phases |
| | 53 | Provide clinical supervision | Provide clinicians with ongoing supervision focusing on the innovation. Provide training for clinical supervisors who will supervise clinicians who provide the innovation | Implementation and Sustainment |
| | 54 | Provide local technical assistance | Develop and use a system to deliver technical assistance focused on implementation and sustainment issues using local personnel | Preparation and Implementation and Sustainment |
| | 51 | Promote adaptability | Identify the ways a clinical innovation can be tailored to meet local needs and clarify which elements of the innovation must be maintained to preserve fidelity. Continue to assess and adapt the fit of the innovation to ensure that is appropriate and sustained if still relevant. | All phases |
| | 63 | Tailor strategies | Tailor the implementation or sustainment strategies to address barriers and leverage facilitators that were identified through ongoing data collection | Preparation, Implementation and Sustainment |
| | 67 | Use data experts | Involve, hire, and/or consult experts to inform management on the use of data generated by implementation and sustainment efforts | Preparation and Implementation and Sustainment |
| | 68 | Use data warehousing techniques | Integrate clinical records across facilities and organizations to facilitate implementation across systems, continually assess that they are still appropriate | Preparation, Implementation and Sustainment |
| | 6 | Build a coalition | Recruit, cultivate and maintain relationships with partners in the implementation and sustainment effort | All phases |
| Develop staekholder interrelationships | 7 | Capture and share local knowledge | Capture local knowledge from implementation sites on how implementers and clinicians made something work and continue to work in their setting and then share it with other sites | Implementation and Sustainment |
| | 17 | Conduct local consensus discussions | Include local providers and other stakeholders in discussions that address whether the chosen problem is important and whether the clinical innovation to address it is appropriate and continues to be appropriate | Exploration and Sustainment |
| | 40 | Involve executive boards | Involve existing governing structures (e.g., boards of directors, medical staff boards of governance) in the implementation and sustainment effort, including the review of data on implementation and sustainment processes | All phases |
| | 47 | Obtain formal commitments | Obtain written commitments from key partners that state what they will do to implement the innovation and how they will support sustainment if it has the intended beneficial effects | Preparation |
| | Extension of strategy #47 explicit to sustainment | Re-affirm formal commitments | Revisit the written commitments obtained from key partners that state what they will do to implement and sustain the innovation. Assess whether these commitments are being upheld and whether new commitments are required to help sustain the innovation | Sustainment |
| | 52 | Promote network weaving | Identify, build and maintain existing high-quality working relationships and networks within and outside the organization, organizational units, teams, etc. to promote information sharing, collaborative problem-solving, and a shared vision/goal related to implementing and sustaining the innovation | All phases |
| | 64 | Use advisory boards and workgroups | Create and engage a formal group of multiple kinds of stakeholders to provide input and advice on implementation and sustainment efforts and to elicit recommendations for improvements | All phases |
| | 24 | Develop academic partnerships | Partner with a university or academic unit for the purposes of shared and ongoing training and bringing relevant research skills to an implementation or sustainment project | All phases |
| | 25 | Develop an implementation glossary | Develop and distribute a list of terms describing the innovation, implementation, and stakeholders in the organizational change | Preparation and Implementation |
| | 36 | Identify early adopters | Identify early adopters at the local site to learn from their experiences with the practice innovation | Exploration, Preparation and Implementation |
| | Extension of strategy #36 explicit to sustainment | Identify successful sustainers | Identify successful sustainer at the local site to learn from their experiences with the practice innovation | Sustainment |
| | 38 | Inform local opinion leaders | Inform providers identified by colleagues as opinion leaders or “educationally influential” about the clinical innovation in the hopes that they will influence colleagues to adopt it | Preparation and Implementation |
| | Extension of strategy #38 explicit to sustainment | Re-engage with local opinion leaders | Periodically re-engage with providers identified by colleagues as opinion leaders or “educationally influential” about the importance of continuing to deliver the practice innovation in the hopes that they will influence colleagues to sustain its use | Sustainment |
| | 35 | Identify and prepare champions | Identify and prepare individuals who dedicate themselves to supporting, marketing, and driving through an implementation, overcoming indifference or resistance that the intervention may provoke in an organization and continue to support sustainment | Preparation and Implementation and Sustainment |
| | 45 | Model and simulate change | Model or simulate the change that will be implemented prior to implementation | Exploration and Preparation |
| | 48 | Organize clinician implementation team meetings | Develop and support teams of clinicians who are implementing the innovation and give them protected time to reflect on the implementation effort, share lessons learned, and support one another's learning | Preparation and Implementation and Sustainment |
| | 57 | Recruit, designate, and train for leadership | Recruit, designate, train and retrain as necessary, leaders for the change effort | Preparation and Implementation and Sustainment |
| | 65 | Use an implementation advisor | Seek guidance from experts in implementation and sustainability | All phases |
| | 72 | Visit other sites | Visit sites where a similar implementation or sustainment effort has been considered successful | Preparation, implementation and Sustainment |
| | 15 | Conduct educational meetings | Hold meetings targeted toward different stakeholder groups (e.g., providers, administrators, other organizational stakeholders, and community, patient/consumer, and family stakeholders) to teach them about the clinical innovation | Preparation and Implementation and Sustainment |
| Train and educate stakeholders | 16 | Conduct educational outreach visits | Have a trained person meet with providers in their practice settings to educate providers about the clinical innovation with the intent of changing the provider's practice | Implementation and Sustainment |
| | 29 | Develop educational materials | Develop and format manuals, toolkits, and other supporting materials in ways that make it easier for stakeholders to learn about the innovation and for clinicians to learn how to deliver the clinical innovation | Preparation |
| | Extension of strategy #29 explicit to sustainment | Review and update educational materials | Review manuals, toolkits, and other supporting materials on how to deliver the clinical innovation and ensure they continue to be appropriate. Update the resources based on changing scientific evidence as needed | Sustainment |
| | 60 | Shadow other experts | Provide ways for key individuals to directly observe experienced people engage with or use the targeted practice change/innovation | Implementation and Sustainment |
| | 19 | Conduct ongoing training | Plan for and conduct training in the clinical innovation in an ongoing way, including training of new staff and booster training for existing staff | Preparation and Implementation and Sustainment |
| | 20 | Create a learning collaborative | Facilitate the formation of groups of relevant stakeholders or organizations and foster a collaborative learning environment to improve implementation and sustainment of the clinical innovation | Preparation and Implementation and Sustainment |
| | 31 | Distribute educational materials | Distribute educational materials (including guidelines, manuals, and toolkits) in person, by mail, and/or electronically | Implementation and Sustainment |
| | 43 | Make training dynamic | Vary the information delivery methods to cater to different learning styles and work contexts, and shape the training in the innovation to be interactive | Preparation and Implementation and Sustainment |
| | 55 | Provide ongoing consultation | Provide ongoing consultation with one or more experts in the strategies used to support implementing and sustaining the innovation | Preparation and Implementation and Sustainment |
| | 71 | Use train-the-trainer strategies | Train designated personnel or organizations to train others in the clinical innovation | Implementation and Sustainment |
| | 73 | Work with educational institutions | Encourage educational institutions to train clinicians in the innovation | Preparation and Implementation and Sustainment |
| | 21 | Create new clinical teams | Change who serves on the clinical team, adding different disciplines and different skills to make it more likely that the clinical innovation is delivered (or is more successfully delivered) in an ongoing way | Preparation and Implementation and Sustainment |
| Support clinicians | 30 | Develop resource sharing agreements | Develop partnerships with organizations that have resources needed to implement and sustain the innovation | Preparation and Implementation and Sustainment |
| | 32 | Facilitate relay of clinical data to providers | Provide as close to real-time data as possible about key measures of process/outcomes using integrated modes/channels of communication in a way that promotes use of the targeted innovation | Implementation and Sustainment |
| | 58 | Remind clinicians | Develop, review and update reminder systems designed to help clinicians to recall information and/or prompt them to use the clinical innovation | Preparation and Implementation and Sustainment |
| | 59 | Revise professional roles | Shift and revise roles among professionals who provide care, and redesign job characteristics | Preparation and Implementation and Sustainment |
| | 37 | Increase demand | Attempt to influence the market for the clinical innovation to increase competition intensity and to increase the maturity of the market for the clinical innovation | Preparation and Implementation and Sustainment |
| Engage consumers | 39 | Intervene with patients/consumers to enhance uptake and adherence | Develop strategies with patients to encourage and problem solve around adherence | Preparation, implementation and Sustainment |
| | 41 | Involve patients/consumers and family members | Engage or include patients/consumers and families in the implementation and sustainment efforts | All phases |
| | 50 | Prepare patients/consumers to be active participants | Prepare patients/consumers to be active in their care, to ask questions, and specifically to inquire about care guidelines, the evidence behind clinical decisions, or about available evidence-supported treatments | All phases |
| | 69 | Use mass media | Use media to reach large numbers of people to spread the word about the clinical innovation | Implementation and Sustainment |
| | 1 | Access new funding | Access new or existing money to facilitate the implementation and/or sustainment | All phases |
| Utilize financial strategies | 2 | Alter incentive/allowance structures | Work to incentivize the adoption, implementation and sustainment of the clinical innovation | Preparation and Implementation and Sustainment |
| | 3 | Alter patient/consumer fees | Create fee structures where patients/consumers pay less for preferred treatments (the clinical innovation) and more for less-preferred treatments | Preparation, implementation and sustainment |
| | 28 | Develop disincentives | Provide financial or professional disincentives for failure to implement or use the clinical innovations | Preparation and Implementation and Sustainment |
| | 34 | Fund and contract for the clinical innovation | Governments and other payers of services issue requests for proposals to deliver the innovation, use contracting processes to motivate providers to deliver the clinical innovation, and develop new funding formulas that make it more likely that providers will deliver and sustain the innovation | Preparation and Implementation and Sustainment |
| | 42 | Make billing easier | Make it easier to bill for the clinical innovation | Preparation, implementation and sustainment |
| | 49 | Place innovation on fee for service lists/formularies | Work to place the clinical innovation on lists of actions for which providers can be reimbursed (e.g., a drug is placed on a formulary, a procedure is now reimbursable) | Preparation, implementation and sustainment |
| | 66 | Use capitated payments | Pay providers or care systems a set amount per patient/consumer for delivering clinical care | Preparation, implementation and sustainment |
| | 70 | Use other payment schemes | Introduce, review and update payment approaches (in a catch-all category) to support implementation and sustainment of the innovation | Preparation, implementation and sustainment |
| | 9 | Change accreditation or membership requirements | Strive to alter accreditation standards so that they require or encourage use of the clinical innovation. Work to alter membership organization requirements so that those who want to affiliate with the organization are encouraged or required to use the clinical innovation | Preparation, implementation and sustainment |
| Change infrastructure | 10 | Change liability laws | Participate in liability reform efforts that make clinicians more willing to deliver the clinical innovation | Preparation, implementation and sustainment |
| | 11 | Change physical structure and equipment | Evaluate periodically current configurations and adapt, as needed, the physical structure and/or equipment (e.g., changing the layout of a room, adding equipment) to best accommodate the targeted innovation | Preparation and Implementation and Sustainment |
| | 12 | Change record systems | Change records systems to allow better assessment of implementation or clinical outcomes | Preparation and Implementation and Sustainment |
| | 13 | Change service sites | Change the location of clinical service sites to increase access | Preparation and Implementation and Sustainment |
| | 22 | Create or change credentialing and/or licensure standards | Create an organization that certifies clinicians in the innovation or encourage an existing organization to do so. Change governmental professional certification or licensure requirements to include delivering the innovation. Work to alter continuing education requirements to shape professional practice toward the innovation | Preparation and Implementation and Sustainment |
| | 44 | Mandate change | Have leadership declare the priority of the innovation and their determination to have it implemented and sustained | Preparation and Implementation and Sustainment |
| | 62 | Start a dissemination organization | Identify or start a separate organization that is responsible for disseminating and supporting the ongoing delivery of the clinical innovation. It could be a for-profit or non-profit organization | Preparation and Implementation and Sustainment |
Sustainability research has been identified as a priority area within implementation science [8]. Sustainability has been defined as “[1] after a defined period of time, [2] the program, clinical intervention, and/or implementation strategies continue to be delivered and/or [3] individual behavior change (i.e., clinician, patient) is maintained; [4] the program and individual behavior change may evolve or adapt while [5] continuing to produce benefits for individuals/systems” [22]. A 2020 review by Moullin et al. [ 23] did however highlight that a number of other conceptual distinctions have been made in the field, particularly in relation to sustainment that is the “sustained use of an EBI” vs. sustainability the “sustained benefits of an EBI.” The sustainment of EBIs is critical as premature ceasing of EBIs may mean that the potential public health and clinical healthcare benefits cease or may not be achieved [24]. Additionally, if EBIs are not sustained there is a significant waste of public health and clinical resources utilized for initial implementation which may have implications for reducing trust of research/academic institutions (24–26).
Whilst there is growing research focused on sustainment as an outcome [27] including consideration of specific factors (24, 27–31) associated with sustainment that may be distinct from those that matter for implementation [32, 33] the field is bereft of evidence of the most effective strategies to support the sustainment of EBIs [24, 27]. A 2019 review of strategies used to sustain public health interventions identified only six studies that purposefully set out to sustain an EBI [27]. Overall only nine sustainment strategies were reported with “ongoing funding,” “booster training,” “supervision and feedback” being the most frequently reported. However, there was insufficient evidence to determine the effectiveness of any one strategy in impacting sustainment. The review reported that most strategies were inadequately described providing very little detail which would enable replication. Such vague and incomplete descriptions of strategies is a limitation of the current evidence base, and highlights the need for a compilation that adequately addresses strategies that support sustainment to ensure they are consistently defined and reported. The review also emphasized the importance of sustainment being considered from the outset of a project and the need for identifying sustainment-focused strategies during the planning of an EBI. Furthermore, strategies relevant to early phases of the initial implementation process are also likely to hold relevance and lay the foundation for longer-term sustainment. However, there is currently no guidance on which strategies should be enacted, and at which phases, to best sustain an EBI.
Given that there are existing compilations for implementation strategies, it is possible that they could be extended or clarified to specifically address sustainment. However key to designing future interventions is the selection of strategies which best addresses the contextual determinants i.e., the barriers and facilitators that impede or promote [4] the sustainment of EBIs [34]. While there may be some overlap with the barriers and facilitators to adoption, implementation, and sustainment of EBIs (e.g., organizational culture and resources), it is likely that there are also barriers and facilitators to sustainment of EBIs (e.g., changes in socio-political environment and funding structures) that may be distinct [35]. Existing compilations may therefore be lacking in identifying and describing strategies that are specific to and necessary for sustaining an EBI. It is however acknowledged that the sustainment of an EBI is inextricably impacted by strategies selected during the previous adoption and or implementation phases [36, 37]. For example, the sustainment of an EBI may be hindered if the adoption and implementation phase has relied on researchers to deliver the intervention, without consideration given to the infrastructure needed to deliver the EBI once research funding ends. Therefore, strategies for the sustainment of EBIs should be considered and planned for in unison with strategies for implementation for any progress to be made in this area. To do this compilations of implementation strategies could specifically incorporate issues relevant to sustainment. This may include updating existing implementation strategies to directly address sustainment or including new strategies that target sustainment-specific barriers and facilitators. Furthermore, whilst frameworks such as the Consolidated Framework for Implementation Research (CFIR) [38] are useful to identify what factors may influence sustainment they do not address how or when change needs to occur [39]. Therefore if we are to plan for sustainment at the beginning of implementation efforts, as has been recommended [36], direction on which strategies need to be employed during which phase of the implementation process is needed.
This research is still in its infancy, and there is an opportunity to establish the use of a compilation of sustainment strategies to allow for consistent reporting and, ultimately, empirical testing. As it is likely that sustainment strategies need to be considered during all phases of implementation, extending an existing compilation of implementation strategies that is already widely used, is likely to support the consideration of sustainment at appropriate phases of implementation and avoid unnecessary duplication. Thus, the aim of this study is to adapt, refine and extend an existing compilation of implementation strategies (ERIC) [13, 21] to explicitly incorporate sustainment, as well as specify the phases of implementation that such strategies are likely to be most salient according to the Exploration, Preparation, Implementation and Sustainment (EPIS) [40] framework.
## Adapting and extending the ERIC compilation to incorporate sustainment
A two-phase iterative approach to adapt the ERIC compilation to include sustainment was undertaken, based on procedures similar to those previously used in the development [41] or adaptation [42] of ERIC or other taxonomies. This involved:
## Adapting and extending through consensus
Consistent with other approaches to developing and extending the ERIC compilation [13, 21, 42], we convened a team of 11 researchers, policy-makers, and practitioners (co-authors of this paper) from Australia, Canada and The United States, who undertook an iterative process of reviewing and adapting the current compilation to incorporate strategies specific to sustainment. For the purpose of this study we defined sustainment as “the sustained use or delivery of an intervention in practice following cessation of external implementation support” [26, 36]. The team are experts in implementation and or sustainability science, and or health service delivery, and included two of the original authors of the ERIC compilation (BP and TW) an expert on the conceptual distinction of ERIC strategies [13, 21, 34]. Both BP and TW have adapted the ERIC for specific contexts [42, 43]. In order to adapt and extend the ERIC the following steps were undertaken.
## Step 1: Barriers to sustainment
We first identified barriers to sustainment from existing studies. These nine publications (27–29, 44–49) were found through snowballing for literature of “barriers to sustainment” which a research assistant extracted into an excel spreadsheet.
## Step 2: Mapping ERIC strategies to address key barriers
To help identify where wording changes may be needed or where additional strategies may need to be created two authors (AH and NN) independently mapped these barriers to existing ERIC strategies. Where the authors felt that a barrier could not be adequately linked to an existing ERIC strategy, they independently drafted proposed wording changes to an existing strategy or identified if a new strategy was needed. The two authors then met to discuss coding, suggested wording changes and or new strategies until they reached consensus. A third author (BP) then reviewed, provided feedback and then met with AH and NN to discuss revisions until consensus was reached.
## Step 3: Iterative consensus process
Following completion of Step 2 all team members were asked to independently review the suggested wording changes and the proposed new strategies developed by AH, NN and BP. They were specifically asked to review and document any edits they believe should be made, or any disagreements they had with the current suggestions, along with detail of their reasoning. After each iteration AH and NN reviewed all feedback. Where there were instances of disagreement between authors they met to develop a proposed amendment and circulated this to all authors for their review. This process of review and updating by the entire team continued for three rounds until consensus was reached.
## Preliminary application of the sustainment-explicit glossary
Following the above, the authors undertook a preliminary test of the application and logic of the sustainment-explicit ERIC glossary to determine its ease of application in the field of sustainment, and if any further adaptions or amendments were needed. As this is still an emerging field to identify potential trials which have employed sustainment strategies we reviewed the National Institutes of Health (NIH) database of trials funded in 2019. We also searched the table of contents of the leading implementation science journals, which included: Implementation Science, Implementation Science Communications, and Frontiers in Public Health for sustainment interventions published between 2018 and 2020. Overall, 12 trials or protocols were identified. As our goal was to check the logic of our proposed adaptation we randomly selected a small number of these studies ($$n = 6$$) to test the sustainment-explicit glossary. Two authors (AH and NN) independently coded the strategies described in those publications against those in the sustainment-explicit ERIC glossary. The authors then compared coding to identify areas of confusion, disagreement, or if any additional strategies emerged. This process was designed to identify where updates were needed to improve the content or wording of the glossary and ensure feasibility in its application. The final glossary was reviewed and agreed on by all authors involved.
## Implementation phase and strategy utility
To help researchers and practitioners identify when they might consider employing each strategy, we categorized each strategy against the phase(s) of implementation according to the Exploration, Preparation, Implementation and Sustainment (EPIS) Framework [37]. To complete this categorization, the same iterative process described above was followed. EPIS was selected as a guiding taxonomy, as it is a widely used and provides clear definitions for each phase. Definitions of the EPIS as defined by the developers [40] were provided to co-authors to help them code the ERIC strategy to the EPIS phase(s).
Table 1 shows that the majority of strategies ($$n = 44$$) were identified as being relevant for consideration during three of the four phases of the EPIS Framework, with 43 of the 44 likely to be needed during preparation, implementation and sustainment phases. Only five strategies were identified as being only relevant during the sustainment phase, which were the four that received deeper levels of adaptation to focus on sustainment (noted above) as well as the novel strategy (also noted above). Thus, majority of existing ERIC strategies were viewed as relevant for more than one EPIS phase, including sustainment.
## Results
The sustainment explicit ERIC glossary is presented in Table 1.
## Adapting ERIC definitions
Of the 73 ERIC strategies, the definitions of 45 were amended to make sustainment more explicit. For the majority ($$n = 41$$) this involved minor surface level changes to include the words “sustainment” or “sustainability.” For example, the definition of “Centralized Technical Assistance” was changed to “develop and use a centralized system to deliver technical assistance focused on implementation and sustainment issues.” Other surface level changes to definitions were more elaborative. For example, the definition of “Promote Adaptability” was changed to “Identify the ways a clinical innovation can be tailored to meet local needs and clarify which elements of the innovation must be maintained to preserve fidelity. Continue to assess and adapt the fit of the innovation to ensure that it is appropriate and sustained if still relevant.” The other four strategies where adaptations were made were identified as being in need of slightly deeper level adaptations. These deeper level adaptations were extensions of existing strategies and reflect changes made to the substance of the definition [42], to specifically encompass issues of sustainment, typically because the original definition more explicitly focused on the application of the strategy at an earlier phase of implementation. For example Obtain formal commitments (strategy 47) was defined as “Obtain written commitments from key partners that state what they will do to implement the innovation and how they will support sustainment if it has the intended beneficial effects” however it was acknowledged that this didn't accurately capture a key barrier to sustainment in regards to ongoing support or decisions around continuation. Accordingly Re-affirm formal commitments (an extension of strategy 47) was added which was defined as “Revisit the written commitments obtained from key partners that state what they will do to implement and sustain the innovation. Assess whether these commitments are being upheld and whether new commitments are required to help sustain the innovation.” The additional strategies are: Review and update educational materials (extension of strategy 29); Identify successful sustainers (extension of strategy 36); Re-engage with local opinion leaders (extension of strategy 38); Re-affirm formal commitments (extension of strategy 47). See Table 1 for the detailed definitions of these strategies.
## Novel sustainment strategies
One new sustainment focused strategy was identified: Communicate with stakeholders the continued impact of the EBP. This strategy takes the information obtained from Audit and provide feedback and/or Develop and organize quality monitoring systems strategies and communicates data to external stakeholders, end-users, and consumers to demonstrate the ongoing benefit, cost effectiveness, or return on investment of the innovation with continued implementation. Conceptually, this strategy seems to fit within the ERIC Use evaluative and iterative strategies cluster [21].
## Preliminary application of the sustainment-explicit ERIC glossary
Application of the sustainment-explicit ERIC identified wide variation in detail and language used to describe the specific strategies employed in the reviewed studies. Consequently, following the initial independent review by the two authors, a thorough discussion and joint application was undertaken to help identify any gaps or areas in need of improvement in the compilation. No new strategies were identified through the coding of published sustainment trials or manuscripts that needed to be considered for inclusion in the glossary. Minor wording changes were made to help clarify some of the strategies and how they relate to sustainment to ensure consistency in interpretation and application.
## Discussion
This is one the first of studies to systematically evaluate an existing compilation of implementation strategies for their relevance for supporting the sustainment of evidence-based programs. The two-phase iterative approach resulted in superficial wording changes to the definitions of 41 of the 73 existing ERIC strategies, slightly deeper wording changes to four ERIC strategies, and the addition of one new strategy. The study also provides guidance to researchers and implementation support practitioners looking to design implementation or sustainment interventions by identifying the phase, according to EPIS framework, when the strategy may need to be considered and employed. It is hoped that a sustainment-explicit glossary based on an existing compilation of implementation strategies will encourage and support those undertaking implementation research to explicitly consider sustainment from the outset and to use a common language when planning and describing their research and practice.
Whilst others have adapted or applied the ERIC compilation to be relevant to a particular setting [42] or class of interventions [50], or to advance understanding of a particular subset of strategies [51], our sustainment-explicit ERIC glossary required minimal changes. We were able to include sustainment concepts by making no changes to strategy names, minimal modifications to definitions and identified only one new strategy. Our extensive mapping exercise of the ERIC strategies to known barriers and facilitators of sustainment from a broad range of studies in clinical and community settings (27–29, 44–49) and sustainability frameworks [24, 36, 52], ensured that we were adequately capturing strategies specific to addressing the main barriers to sustainment.
The preliminary application of the glossary further highlighted the lack of standardized reporting that is already emerging within the sustainment literature. Of the studies reviewed ($$n = 6$$), many of the strategies utilized were not adequately described in enough detail, or were hard to disentangle from other strategies, which would make it difficult for any future studies wishing to synthesize the effects of these strategies. To avoid the challenges that this has caused historically in the field of implementation science, we implore those planning, or currently undertaking, sustainment research to use consistent terminology to describe their chosen strategies, particularly when multiple strategies are used. Furthermore, as recommended by Michie and Johnston [53] for implementation interventions, we encourage trialists to describe these strategies with sufficient detail in terms of “what,” “who,” “when,” “where” and “how,” so these components of each strategy can be sufficiently understood and replicated by others. Frameworks such as those developed by Proctor et al. [ 8] or Presseau et al. [ 54] provide useful guidance for specifying this behavior (in the context of implementation and sustainment interventions) [1]. If strategies addressing sustainment are consistently described in future research trials this will enable replication studies to be undertaken and study findings synthesized to identify effective strategies or combinations of strategies, and the optimal timing of their delivery, all of which will enhance the design of future sustainment interventions. Whilst the sustainment-explicit ERIC glossary captures all strategies previously identified [27], as evidence in the field continues to grow there may be a need for new strategies to be added. Therefore, this glossary will need to be continuously refined to maintain its utility in sustainment research.
Our application of the EPIS Framework found that a large majority of strategies should be considered during the design and earlier phases of implementation. This is consistent with others who have advocated that implementation and sustainment are interconnected and therefore need to be planned for in advance (55–58). This is also supported by more recent sustainability frameworks such as the Dynamic Sustainability Framework or the RE-AIM extension for sustainment which posits sustainability is not “static,” but rather dynamic, impacted by the changing context in which the intervention is being delivered, the evolving scientific evidence, and the dynamic needs of a population. In a recent study the original developers of the ERIC assessed which strategies experts perceived as being most essential for implementation of three high priority mental health care practices in the US Department of Veteran Affairs [43]. The authors found that experts consistently selected a similar set of ERIC strategies as essential for implementation success, regardless of type of EBI [43] or implementation phase. Again, this study highlights the interconnectedness of sustainment with the earlier phases of implementation, and how strategies can be perceived as relevant across the different implementation phases. Shelton et al. [ 36] suggests that in planning for sustainability, monitoring the reach, adoption, effectiveness and implementation of an EBI is essential to identify early on when challenges are arising and if and how strategies can be adapted, refined, or introduced to support the sustainment of the EBI and address health inequities that may be exacerbated over time.
Robust and valid frameworks or theories specific to sustainability such as the Dynamic Sustainability Framework [52] or the Integrated Sustainability Framework [24] should be employed alongside the sustainment-explicit ERIC glossary, when planning sustainment trials. These frameworks and theories will help identify issues specific to sustainment that should be addressed by any strategies being developed and evaluated [59]. Unfortunately, a large proportion of sustainability research is not based on relevant theories, frameworks, or models and for those studies that have, there is wide variation and limited validity in the theories and frameworks commonly applied [59]. There is significant need for sustainability research to evaluate the application of sustainability frameworks alongside a compilation such as ERIC [60]. This is important if we are to identify how or why strategies impacting sustainment exert their effects (i.e., the mechanisms through which they work) [6]. Once this is known we may improve the effectiveness and cost-effectiveness of future interventions by keeping, strengthening, adding or removing strategies that target (or don't) mediators which lead to improvements in sustained implementation [5, 61].
There are several limitations to this study. First, unlike the methods used to develop the original ERIC compilation, we only had a small number of implementation and sustainability experts ($$n = 11$$) convened to specifically work on this project. Whilst we represented community and clinical perspectives from various countries to gain a broader perspective on this issue, a larger, more diverse, group of experts should further review and revise this glossary for use in sustainment-focused work. Second, we only tested the application of the glossary with a small number of studies. This was undertaken as to test the logic of the amendments; it was not designed to be an extensive application of the sustainment-explicit ERIC or to identify what strategies are being used in sustainment trials. Accordingly, this glossary has not been extensively tested, further application and review of this glossary is needed and welcomed and through its use, it may be evident that further updates are required. Finally, ongoing work is needed to assess the extent to which the sustainment-explicit ERIC glossary is relevant to low- and middle-income countries [62], as this study did not explicitly address this question.
## Conclusions
The sustainment-explicit ERIC glossary addresses the need for explicit and clear definitions of strategies to be used in sustainment interventions. The application of relevant strategies during planning and implementation phases may subsequently enhance the evidence-base for the field, and ultimately the sustainment, spread and scale of interventions and improvements in our communities health [63]. Future work is needed to empirically test the effectiveness of these strategies in sustaining EBIs in clinical and community settings.
## Data availability statement
The original contributions presented in the study are included in the article. Further inquiries can be directed to the corresponding author.
## Author contributions
NN obtained funding for the study. NN, AH, BP, and LW conceived the study concept and developed the study design. NN and AH undertook initial adaptations of the ERIC and classification against EPIS. BP and TW provided expert advice as original developers of ERIC. RCS, CL, LW, MH, SY, RS, and MK advised on and undertook the adaption, extension, consensus process, and pilot testing of the tool. NN and AH developed the draft manuscript. All authors contributed to the article and approved the final version of the manuscript.
## Funding
This project is funded through the NHMRC as part of NN's Medical Research Future Fund (MRFF) Investigator Grant (GS2000053), BP is supported in part through the U.S. National Institutes of Health (K01MH113806, R01CA262325, P50CA19006, and P50MH126219) and the Agency for Healthcare Research and Quality (R13HS025632), RS is supported by an NHMRC Medical Research Future Fund (MRFF) Investigator Grant (G2000052), LW is supported by an NHMRC Investigator Grant (G1901360), and SY is supported by an ARC DECRA (G1600359). The funders had no role in the study design, conduct of the study, analysis, or dissemination of findings.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Wolfenden L, Foy R, Presseau J, Grimshaw JM, Ivers NM, Powell BJ. **Designing and undertaking randomised implementation trials: guide for researchers**. *BMJ.* (2021) **372** m3721. DOI: 10.1136/bmj.m3721
2. Barnes C, McCrabb S, Stacey F, Nathan N, Yoong SL, Grady A. **Improving implementation of school-based healthy eating and physical activity policies, practices, and programs: a systematic review**. *Transl Behav Med.* (2021) **11** 1365-410. DOI: 10.1093/tbm/ibab037
3. Powell BJ, Fernandez ME, Williams NJ, Aarons GA, Beidas RS, Lewis CC. **Enhancing the impact of implementation strategies in healthcare: a research agenda**. *Front Public Health* (2019) **7** 3. DOI: 10.3389/fpubh.2019.00003
4. Lewis CC, Boyd MR, Walsh-Bailey C, Lyon AR, Beidas R, Mittman B. **A systematic review of empirical studies examining mechanisms of implementation in health**. *Implement Sci.* (2020) **15** 21. DOI: 10.1186/s13012-020-00983-3
5. Lee H, Hall A, Nathan N, Reilly KL, Seward K, Williams CM. **Mechanisms of implementing public health interventions: a pooled causal mediation analysis of randomised trials**. *Implement Sci.* (2018) **13** 1-11. DOI: 10.1186/s13012-018-0734-9
6. Lewis CC, Powell BJ, Brewer SK, Nguyen AM, Schriger SH, Vejnoska SF. **Advancing mechanisms of implementation to accelerate sustainable evidence-based practice integration: protocol for generating a research agenda**. *BMJ Open.* (2021) **11** e053474. DOI: 10.1136/bmjopen-2021-053474
7. Geng EH, Baumann AA, Powell BJ. **Mechanism mapping to advance research on implementation strategies**. *PLoS Med.* (2022) **19** e1003918. DOI: 10.1371/journal.pmed.1003918
8. Proctor EK, Powell BJ, McMillen JC. **Implementation strategies: recommendations for specifying and reporting**. *Implement Sci.* (2013) **8** 139. DOI: 10.1186/1748-5908-8-139
9. Wolfenden L, Goldman S, Stacey FG, Grady A, Kingsland M, Williams CM. **Strategies to improve the implementation of workplace-based policies or practices targeting tobacco, alcohol, diet, physical activity and obesity**. *Cochrane Database of Syst Rev* (2018) **11** CD012439. DOI: 10.1002/14651858.CD012439.pub2
10. Wolfenden L, Jones J, Williams CM, Finch M, Wyse RJ, Kingsland M. **Strategies to improve the implementation of healthy eating, physical activity and obesity prevention policies, practices or programmes within childcare services**. *Cochrane Database Syst Rev.* (2016) **10** Cd011779. DOI: 10.1002/14651858.CD011779.pub2
11. 11.Effective Practice Organisation of Care (EPOC). EPOC Taxonomy. (2015). Available online at: http://www.epoc.cochrane.org/epoc-taxonomy (accessed December 15, 2021).. (2015)
12. Mazza D, Bairstow HP, Buchan Chakraborty SP, Van Hecke O, Grech C. **Refining a taxonomy for guideline implementation: results of an exercise in abstract classification**. *Implement Sci.* (2013) **8** 32. DOI: 10.1186/1748-5908-8-32
13. Powell BJ, Waltz TJ, Chinman MJ, Damschroder LJ, Smith JL, Matthieu MM. **A refined compilation of implementation strategies: results from the Expert Recommendations for Implementing Change (ERIC) project**. *Implement Sci.* (2015) **10** 21. DOI: 10.1186/s13012-015-0209-1
14. Michie S, Fixsen D, Grimshaw JM, Eccles MP. **Specifying and reporting complex behaviour change interventions: the need for a scientific method**. *Implement Sci.* (2009) **4** 40. DOI: 10.1186/1748-5908-4-40
15. Lengnick-Hall R, Stadnick NA, Dickson KS, Moullin JC, Aarons GA. **Forms and functions of bridging factors: specifying the dynamic links between outer and inner contexts during implementation and sustainment**. *Implement Sci.* (2021) **16** 34. DOI: 10.1186/s13012-021-01099-y
16. McIntyre SA, Francis JJ, Gould NJ, Lorencatto F. **The use of theory in process evaluations conducted alongside randomized trials of implementation interventions: a systematic review**. *Transl Behav Med.* (2020) **10** 168-78. DOI: 10.1093/tbm/iby110
17. Slaughter SE, Zimmermann GL, Nuspl M, Hanson HM, Albrecht L, Esmail R. **Classification schemes for knowledge translation interventions: a practical resource for researchers**. *BMC Med Res Methodol.* (2017) **17** 161. DOI: 10.1186/s12874-017-0441-2
18. Kok G, Gottlieb NH, Peters GJ, Mullen PD, Parcel GS, Ruiter RA. **A taxonomy of behaviour change methods: an Intervention Mapping approach**. *Health Psychol Rev.* (2016) **10** 297-312. DOI: 10.1080/17437199.2015.1077155
19. Michie S, Richardson M, Johnston M, Abraham C, Francis J, Hardeman W. **The behavior change technique taxonomy (v1) of 93 hierarchically clustered techniques: building an international consensus for the reporting of behavior change interventions**. *Ann Behav Med.* (2013) **46** 81-95. DOI: 10.1007/s12160-013-9486-6
20. McHugh S, Presseau J, Luecking CT, Powell BJ. **Examining the complementarity between the ERIC compilation of implementation strategies and the behaviour change technique taxonomy: a qualitative analysis**. *Implementat Sci.* (2022) **17** 56. DOI: 10.1186/s13012-022-01227-2
21. Waltz TJ, Powell BJ, Matthieu MM, Damschroder LJ, Chinman MJ, Smith JL. **Use of concept mapping to characterize relationships among implementation strategies and assess their feasibility and importance: results from the Expert Recommendations for Implementing Change (ERIC) study**. *Implement Sci.* (2015) **10** 109. DOI: 10.1186/s13012-015-0295-0
22. Moore JE, Mascarenhas A, Bain J, Straus SE. **Developing a comprehensive definition of sustainability**. *Implement Sci.* (2017) **12** 110. DOI: 10.1186/s13012-017-0637-1
23. Moullin JC, Sklar M, Green A, Dickson KS, Stadnick NA, Reeder K. **Advancing the pragmatic measurement of sustainment: a narrative review of measures**. *Implement Sci Commun.* (2020) **1** 76. DOI: 10.1186/s43058-020-00068-8
24. Shelton RC, Cooper BR, Stirman SW. **The sustainability of evidence-based interventions and practices in public health and health care**. *Annu Rev Public Health.* (2018) **39** 55-76. DOI: 10.1146/annurev-publhealth-040617-014731
25. Wiltsey Stirman S, Kimberly J, Cook N, Calloway A, Castro F, Charns M. **The sustainability of new programs and innovations: a review of the empirical literature and recommendations for future research**. *Implement Sci.* (2012) **7** 17. DOI: 10.1186/1748-5908-7-17
26. Scheirer MA, Dearing JW. **An agenda for research on the sustainability of public health programs**. *Am J Public Health.* (2011) **101** 2059-67. DOI: 10.2105/AJPH.2011.300193
27. Hailemariam M, Bustos T, Montgomery B, Barajas R, Evans LB, Drahota A. **Evidence-based intervention sustainability strategies: a systematic review**. *Implementat Sci.* (2019) **14** 57. DOI: 10.1186/s13012-019-0910-6
28. Shoesmith A, Hall A, Wolfenden L, Shelton RC, Powell BJ, Brown H. **Barriers and facilitators influencing the sustainment of health behaviour interventions in schools and childcare services: a systematic review**. *Implement Sci.* (2021) **16** 62. DOI: 10.1186/s13012-021-01134-y
29. Cowie J, Nicoll A, Dimova ED, Campbell P, Duncan EA. **The barriers and facilitators influencing the sustainability of hospital-based interventions: a systematic review**. *BMC Health Serv Res.* (2020) **20** 588. DOI: 10.1186/s12913-020-05434-9
30. Hodge LM, Turner KM. **Sustained implementation of evidence-based programs in disadvantaged communities: a conceptual framework of supporting factors**. *Am J Community Psychol.* (2016) **58** 192-210. DOI: 10.1002/ajcp.12082
31. Hall A, Shoesmith A, Shelton RC, Lane C, Wolfenden L, Nathan N. **Adaptation and Validation of the Program Sustainability Assessment Tool (PSAT) for Use in the Elementary School Setting**. *Int J Environ Res Public Health.* (2021) **18** 11414. DOI: 10.3390/ijerph182111414
32. McFadyen T, Wolfenden L, Kingsland M, Tindall J, Sherker S, Heaton R. **Sustaining the implementation of alcohol management practices by community sports clubs: a randomised control trial**. *BMC Public Health.* (2019) **19** 1660. DOI: 10.1186/s12889-019-7974-8
33. Nathan N, Hall A, McCarthy N, Sutherland R, Wiggers J, Bauman AE. **Multi-strategy intervention increases school implementation and maintenance of a mandatory physical activity policy: outcomes of a cluster randomised controlled trial**. *Br J Sports Med* (2022) **56** 385-93. DOI: 10.1136/bjsports-2020-103764
34. Waltz TJ, Powell BJ, Fernández ME, Abadie B, Damschroder LJ. **Choosing implementation strategies to address contextual barriers: diversity in recommendations and future directions**. *Implementat Sci.* (2019) **14** 42. DOI: 10.1186/s13012-019-0892-4
35. Shelton RC, Dunston SK, Leoce N, Jandorf L, Thompson HS, Crookes DM. **Predictors of activity level and retention among African American lay health advisors (LHAs) from The National Witness Project: Implications for the implementation and sustainability of community-based LHA programs from a longitudinal study**. *Implement Sci.* (2016) **11** 41. DOI: 10.1186/s13012-016-0403-9
36. Shelton RC, Chambers DA, Glasgow RE. **An Extension of RE-AIM to enhance sustainability: addressing dynamic context and promoting health equity over time**. *Front Public Health* (2020) **8** 134. DOI: 10.3389/fpubh.2020.00134
37. Aarons GA, Hurlburt M, Horwitz SM. **Advancing a Conceptual model of evidence-based practice implementation in public service sectors**. *Adm Policy Ment Health.* (2011) **38** 4-23. DOI: 10.1007/s10488-010-0327-7
38. Damschroder LJ, Aron DC, Keith RE, Kirsh SR, Alexander JA, Lowery JC. **Fostering implementation of health services research findings into practice: a consolidated framework for advancing implementation science**. *Implement Sci.* (2009) **4** 50. DOI: 10.1186/1748-5908-4-50
39. Nilsen P. **Making sense of implementation theories, models and frameworks**. *Implement Sci.* (2015) **10** 53. DOI: 10.1186/s13012-015-0242-0
40. 40.The EPIS Framework Website. (2016). Available online at: https://episframework.com/what-is-epis (accessed August 09, 2022).. (2016)
41. Michie S, Ashford S, Sniehotta FF, Dombrowski SU, Bishop A, French DP. **refined taxonomy of behaviour change techniques to help people change their physical activity and healthy eating behaviours: the CALO-RE taxonomy**. *Psychol Health.* (2011) **26** 1479-98. DOI: 10.1080/08870446.2010.540664
42. Cook CR, Lyon AR, Locke J, Waltz T, Powell BJ. **Adapting a Compilation of Implementation Strategies to Advance School-Based Implementation Research and Practice**. *Prev Sci.* (2019) **20** 914-35. DOI: 10.1007/s11121-019-01017-1
43. Waltz TJ, Powell BJ, Matthieu MM, Smith JL, Damschroder LJ, Chinman MJ. **Consensus on strategies for implementing high priority mental health care practices within the US department of veterans affairs**. *Implement Res Pract.* (2021) **2** 26334895211004607. DOI: 10.1177/26334895211004607
44. Cassar S, Salmon J, Timperio A, Naylor PJ, van Nassau F, Contardo Ayala AM. **Adoption, implementation and sustainability of school-based physical activity and sedentary behaviour interventions in real-world settings: a systematic review**. *Int J Behav Nutr Phys Act.* (2019) **16** 120. DOI: 10.1186/s12966-019-0876-4
45. Herlitz L, MacIntyre H, Osborn T, Bonell C. **The sustainability of public health interventions in schools: a systematic review**. *Implementat Sci.* (2020) **15** 4. DOI: 10.1186/s13012-019-0961-8
46. Bodkin A, Hakimi S. **Sustainable by design: a systematic review of factors for health promotion program sustainability**. *BMC Public Health.* (2020) **20** 964-964. DOI: 10.1186/s12889-020-09091-9
47. Acosta J, Chinman M, Ebener PA, Malone PS, Cannon JS, D'Amico EJ. **Sustaining an Evidence-Based Program Over Time: Moderators of Sustainability and the Role of the Getting to Outcomes**. *Prev Sci.* (2020) **21** 807-19. DOI: 10.1007/s11121-020-01118-2
48. Popowich AD, Mushquash AR, Pearson E, Schmidt F, Mushquash CJ. **Barriers and facilitators affecting the sustainability of dialectical behaviour therapy programmes: a qualitative study of clinician perspectives**. *Couns Psychother Res.* (2020) **20** 68-80. DOI: 10.1002/capr.12250
49. Hunter SB, Felician M, Dopp AR, Godley SH, Pham C, Bouskill K. **What influences evidence-based treatment sustainment after implementation support ends? A mixed method study of the adolescent-community reinforcement approach**. *J Substance Abuse Treat.* (2020) **113** 107999. DOI: 10.1016/j.jsat.2020.107999
50. Graham AK, Lattie EG, Powell BJ, Lyon AR, Smith JD, Schueller SM. **Implementation strategies for digital mental health interventions in health care settings**. *Am Psychol.* (2020) **75** 1080-92. DOI: 10.1037/amp0000686
51. Dopp AR, Narcisse MR, Mundey P, Silovsky JF, Smith AB, Mandell D. **A scoping review of strategies for financing the implementation of evidence-based practices in behavioral health systems: State of the literature and future directions**. *Implementat Res Pract.* (2020) **1** 2633489520939980. DOI: 10.1177/2633489520939980
52. Chambers DA, Glasgow RE, Stange KC. **The dynamic sustainability framework: addressing the paradox of sustainment amid ongoing change**. *Implement Sci.* (2013) **8** 117. DOI: 10.1186/1748-5908-8-117
53. Michie S, Johnston M. **Changing clinical behaviour by making guidelines specific**. *BMJ.* (2004) **328** 343-5. DOI: 10.1136/bmj.328.7435.343
54. Presseau J, McCleary N, Lorencatto F, Patey AM, Grimshaw JM, Francis JJ. **Action, actor, context, target, time (AACTT): a framework for specifying behaviour**. *Implement Sci.* (2019) **14** 102. DOI: 10.1186/s13012-019-0951-x
55. Pluye P, Potvin L, Denis JL. **Making public health programs last: conceptualizing sustainability**. *Eval Program Plann.* (2004) **27** 121-33. DOI: 10.1016/j.evalprogplan.2004.01.001
56. Walugembe DR, Sibbald S, Le Ber MJ, Kothari A. **Sustainability of public health interventions: where are the gaps?**. *Health Res Policy Syst.* (2019) **17** 8. DOI: 10.1186/s12961-018-0405-y
57. Alidina S, Zanial N, Meara JG, Barash D, Buberwa L, Chirangi B. **Applying the Exploration, Preparation, Implementation, Sustainment (EPIS) Framework to Safe Surgery 2020 Implementation in Tanzania's Lake Zone**. *J Am Coll Surg* (2021) **233** 177-191. DOI: 10.1016/j.jamcollsurg.2021.04.006
58. Becan JE, Bartkowski JP, Knight DK, Wiley TRA, DiClemente R, Ducharme L. **A model for rigorously applying the Exploration, Preparation, Implementation, Sustainment (EPIS) framework in the design and measurement of a large scale collaborative multi-site study**. *Health Justice.* (2018) **6** 9-9. DOI: 10.1186/s40352-018-0068-3
59. Birken SA, Haines ER, Hwang S, Chambers DA, Bunger AC, Nilsen P. **Advancing understanding and identifying strategies for sustaining evidence-based practices: a review of reviews**. *Implement Sci.* (2020) **15** 88. DOI: 10.1186/s13012-020-01040-9
60. Shelton RC, Lee M, Brotzman LE, Wolfenden L, Nathan N, Wainberg ML. **What Is Dissemination and Implementation Science?: An Introduction and Opportunities to Advance Behavioral Medicine and Public Health Globally**. *Int J Behav Med* (2020) **27** 3-20. DOI: 10.1007/s12529-020-09848-x
61. Wolfenden L, Bolsewicz K, Grady A, McCrabb S, Kingsland M, Wiggers J. **Optimisation: defining and exploring a concept to enhance the impact of public health initiatives**. *Health Res Policy Syst.* (2019) **17** 1-13. DOI: 10.1186/s12961-019-0502-6
62. Brownson RC, Kumanyika SK, Kreuter MW, Haire-Joshu D. **Implementation science should give higher priority to health equity**. *Implement Sci.* (2021) **16** 28. DOI: 10.1186/s13012-021-01097-0
63. Laur C, Corrado AM, Grimshaw JM, Ivers N. **Trialists perspectives on sustaining, spreading, and scaling-up of quality improvement interventions**. *Implementat Sci Commun.* (2021) **2** 35. DOI: 10.1186/s43058-021-00137-6
|
---
title: “Provider discretionary power practices to support implementation of patient-centered
HIV care in Lusaka, Zambia”
authors:
- Chanda Mwamba
- Njekwa Mukamba
- Anjali Sharma
- Kasapo Lumbo
- Marksman Foloko
- Herbert Nyirenda
- Sandra Simbeza
- Kombatende Sikombe
- Charles B. Holmes
- Izukanji Sikazwe
- Carolyn Bolton Moore
- Aaloke Mody
- Elvin Geng
- Laura K. Beres
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012689
doi: 10.3389/frhs.2022.918874
license: CC BY 4.0
---
# “Provider discretionary power practices to support implementation of patient-centered HIV care in Lusaka, Zambia”
## Abstract
### Introduction
Traditional patient-provider relationships privilege the providers, as they possess the formal authority and clinical knowledge applied to address illness, but providers also have discretion over how they exercise their power to influence patients' services, benefits, and sanctions. In this study, we assessed providers' exercise of discretionary power in implementing patient-centered care (PCC) practices in Lusaka, Zambia.
### Methods
HIV clinical encounters between patients on antiretroviral therapy (ART) and providers across 24 public health facilities in Lusaka Province were audio recorded and transcribed verbatim. Using qualitative content analysis, we identified practices of discretionary power (DP) employed in the implementation of PCC and instances of withholding DP. A codebook of DP practices was inductively and iteratively developed. We compared outcomes across provider cadres and within sites over time.
### Results
We captured 194 patient-provider interactions at 24 study sites involving 11 Medical Officers, 58 Clinical Officers and 10 Nurses between August 2019 to May 2021. Median interaction length was 7.5 min. In a hierarchy where providers dominate patients and interactions are rapid, some providers invited patients to ask questions and responded at length with information that could increase patient understanding and agency. Others used inclusive language, welcomed patients, conducted introductions, and apologized for delayed services, narrowing the hierarchical distance between patient and provider, and facilitating recognition of the patient as a partner in care. Although less common, providers shared their decision-making powers, allowing patients to choose appointment dates and influence regimens. They also facilitated resource access, including access to services and providers outside of scheduled appointment times. Application of DP was not universal and missed opportunities were identified.
### Conclusion
Supporting providers to recognize their power and intentionally share it is both inherent to the practice of PCC (e.g., making a patient a partner), and a way to implement improved patient support. More research is needed to understand the application of DP practices in improving the patient-centeredness of care in non-ART settings.
## Introduction
Health providers traditionally hold power over patients within a patient-provider interaction because they are elevated by specialized clinical knowledge, resource access, and health system policies and procedures that confer authority to providers (1–4). Leveraging gaps and inherent flexibility in health service delivery guidance, public service health providers have opportunities to use their discretion in how they exercise their power during patient interactions in service of either facilitating or frustrating patient experiences of healthcare (5–7). Lipsky coined the term ‘street-level bureaucracy' (SLB) to describe this phenomenon [7]. Lipsky's theory of SLB [7] postulates that frontline workers function as interpreters of policies and have “the power to exercise a degree of discretion over the services, benefits and sanctions” they provide [8]. To fill gaps in training, logistical or supervisory support, health providers rely on their own belief, practice, service delivery, professional, and social networks to operationalize and implement health policies (5–8). This policy interpretation and shared meaning enter routine practice within which health providers moderate services, benefits, and sanctions (7–11). This moderation embedded in informality and resource-constraints empowers providers' role as gatekeepers, who can use their discretion to restrict or enhance access [10], putting patients at a disadvantage; however, this does not always have to be the case as healthcare providers can respond to individual patient and contextual needs through innovations that improve both service delivery processes and outcomes [12, 13]. It is important to understand how providers utilize their power within a patient-provider interaction to facilitate patient-centeredness in health service delivery.
Health systems increasingly recognize the importance of patient-centered care (PCC) [14]. The World Health Organization recommendation for people-centered HIV care and treatment reiterates the importance of improving patients' experience by respecting patient autonomy and ability to choose the best course of action within their socio-cultural context [15]. Conceptualizations of PCC demonstrate that it operates through patient-provider micro-interactions, health system structures (meso-level), and the larger socio-cultural context (macro-levels) [13, 14].
People living with HIV experience the requirements of early HIV diagnosis, linkage to care, and lifetime commitment to ART differently depending on their age, gender, sexual identity, and health status, with each step in the HIV care cascade complicated by their personal, professional, and social circumstances (16–21). Traditionally, HIV care and treatment has been top-down and required patient compliance rather than ownership of life-long antiretroviral therapy (ART) [11, 22]. As a result, health providers have used verbal and non-verbal communication to direct, proscribe, control, and persuade patients to adopt what health providers consider acceptable and appropriate behavior for ‘healthy subjects' (10, 11, 22–24). In an exercise of power, health providers perceive patients who miss or delay visits as ‘bad' and subject them to punishing enhanced counseling irrespective of previous ART adherence or the circumstances which delayed them (11, 25–27). In other instances, health providers may change patient care practices due to resource constraints, for example, substituting ART regimens due to low or quickly expiring supplies, which may inadvertently increase the burden of treatment for the patient [28]. Additionally, standardized treatment protocols may not account for differential needs of women, adolescents, men, the differently abled, and the aged, putting them under undue stress as they navigate their social dependencies and roles to meet stringent health service requirements and multiple, required appointments at varied locations (16–21). Thus, health providers may exert their discretionary power using their best judgement under the prevailing circumstances, but nonetheless deliver sub-optimal services and leave patients dissatisfied with their HIV care [25]. In turn, dissatisfied patients exercise their agency to disengage from HIV care, silently transfer to another clinic, or seek alternate treatments, creating space for opportunistic infections [29, 30].
We created a multi-component intervention to improve patient-centered HIV care in Lusaka, Zambia, which promoted shared decision-making, good communication, and welcoming rather than punishing tones and procedures for those re-engaging in HIV care. While providing no definition during training, we introduced the term ‘discretionary power' simply stating that healthcare workers had the power to use their discretion per patients' specific situation without compromising on policies and guidelines. We provided and solicited examples of discretionary power used to better serve patients from trainees on the 1st day of training. On the 2nd day of training held a week later, we invited them to share experiences with their use of discretionary power after being introduced to the concept. We hypothesized that, when supported, health providers would use discretionary power to increase open, positive interactions and innovative and responsive HIV care [6, 11]. Nested within a stepped-wedge, hybrid implementation-effectiveness trial of the PCC intervention, this study utilized qualitative methods to assess the exercise of DP as it relates to the implementation of PCC through examining provider choices (actions and inactions) in the patient-provider clinical interaction. Insights into how providers communicate with patients to understand their socio-cultural context and how they leverage discretionary power to alleviate identified challenges can inform future interventions that improve people-centeredness of care.
## Study background
The Center for Infectious Disease Research in Zambia (CIDRZ) implemented the ‘Person-Centered Public Health for HIV Treatment in Zambia' (PCPH) study, across 24 Ministry of Health (MoH) facilities in Lusaka Province, Zambia, from August 2019 to November 2021, with the aim of improving health care workers' and patient experiences that would then lead to improved service delivery and clinical outcomes, including retention and viral suppression among patients living with HIV. The PCPH intervention included [1] training health care workers in patient-centered care principles and skills (PCC) including communication practices, [2] on-site mentoring of health care workers on application of PCC, [3] measuring the patient experience and feeding it back to health provider's quarterly through data review meetings, [4] in-kind incentives for improved facility-level performance. The PCPH stepped-wedge trial rolled out the intervention across four, 6-month periods with eight new intervention sites in Period 1, four additional intervention sites in Periods two and three, respectively, and eight additional intervention sites in Period 4.
To understand the implementation of PCC communication within the patient-provider interaction, individual-level consultations between HIV care providers and patients were audio recorded for quantitative analysis using the Roter Interaction Analysis System (RIAS) [31]. Nested within this, we transcribed the audio recordings to assess provider use of discretionary power in the implementation of patient-centered HIV care using thematic content analysis [32].
## Study population, recruitment, and sampling
HIV providers, including Medical Officers (MO), Clinical Officers (CO), and Nurses, were purposively sampled to achieve balance across cadres within each of the 24 study facilities. The study sought 5 providers per facility with 1–3 consultations recorded per provider per study period, following them from Period 1 through Period 4. HIV providers were made aware of the study opportunity through an announcement at a staff meeting including study sensitization. Interested providers gave voluntary, written consent to participate after the recruitment meeting. Research staff members then returned on an HIV clinic consultation day to record consults given by participating providers. Patients living with HIV who were present on the day when clinical consultation recordings were planned and seeing a consenting provider were informed about the study in the waiting room in ART clinics prior to seeing providers for their visit. PCPH Qualitative Research Assistants (QRAs) working at the public health facilities identified and recruited ART patients queuing to see any of the recruited HIV providers. Eligible HIV clients were those that were: (a) 18 years and above and (b) spoke one of the study languages: Nyanja, Bemba, Tonga, or English, and (c) voluntarily consented to participate in the study. Consenting patients were enrolled sequentially in the order of their consultation appointment until the sample size for their provider was reached.
To assess use of discretionary power in the implementation of PCC, we sampled intervention facilities during the main intervention phase and over time, including consultations recorded during the first study period in which the facility was an intervention site (Periods 1–3), and all intervention facilities in the final period, Period 4.
Three cadres of providers were included in this study in their role as ART providers: Clinical Officers (COs), Medical Officers (MOs) and Nurses. COs are primary health workers who have completed a 3-year post-secondary general medical education and provide the majority of Zambia's healthcare [33, 34]. MOs have completed a minimum of 6 years in medical training, are in short supply [33, 34], and the most senior provider cadre. Some Nurses provide ART consultations after The Zambian Ministry of Health introduced a nurse-centered antiretroviral treatment (ART) prescription initiative to train and support nurses in prescribing ART due to physician shortages [33, 34].
## Study procedures and data collection
Patient-provider interactions were audio-recorded using recorders which were positioned discreetly in the HIV clinical consultation rooms. The recorders were turned on by a QRA at the beginning of a consultation between a consenting provider and consenting patient then the QRA exited the consultation room. After the consenting patient exited the consultation room, the QRA turned off the recorder. For the discretionary power analysis, audio recording were transcribed verbatim and simultaneously translated into English, if applicable.
## Data analysis
We applied qualitative thematic content analysis to identify practices of discretionary power (DP) employed in the implementation of PCC and instances of withholding DP. To develop the set of DP practices for which we coded, two independent analysts (CM, LKB) reviewed literature on DP and discussed the conceptual meaning of both DP and PCC. The analysts then read a $15\%$ sample of transcripts to inductively develop a code book of DP practices, with ongoing dialogue and refinement through using coding memos to guide reflection. The final code book was then applied across all transcripts. Differences in coding were resolved through dialogue. The lead author summarized themes and categories, discussing results with the second analyst and study investigators.
## Results
We enrolled 79 health providers (11 MOs, 58 COs and 10 Nurses) between August 2019 to November 2021 from the 24 intervention facilities. During period 4 data collection, 6 enrolled health providers were not available for the study team to capture their follow-up consultation sessions. Transcripts from 194 consultations were included in our analysis; median recorded consultation time was 7 min 29 s (min: 1:54, interquartile range: 5:38–11:19, max: 23:59). Of the interactions, 23 involved MOs, 142 Cos, and 29 Nurses. We compared the use of discretionary power across provider cadres from the intervention facilities.
Our inductive coding of DP practices facilitating patient-centeredness of care showed three central themes, represented below. Themes and supporting quotes are summarized in Table 1. We contrast the use of DP to improve patient-centeredness under each theme by emphasizing both implemented and missed opportunities for clinicians to use DP to improve patient-centeredness that were evident in the data.
**Table 1**
| Themes | Sub-themes | Quotes |
| --- | --- | --- |
| Narrowing hierarchical distance | Welcoming patients: not universally done, opens dialogue prior to clinical consultation | CO: How are you? Patient: Am fine and how are you? CO: How is home? Patient: Home is fine CO: It's good that you have decided to come here today Patient: Thank you- [CO] |
| | Inviting introductions: not universally done, recognizes patient and improves accessibility of provider prior to clinical consultation | Nurse: My name is Mr. XXX, I will be the one attending to you today. I have seen that they have written XXX, is that your name? Patient: Yes. Nurse: I am a nurse so feel free. This is a men's clinic, a clinic that attends to men, so we respect you. [Nurse] |
| | Offered apology | CO: How are you? Sorry I have delayed you, I was checking on another patient. Patient: It is fine thanks, I understand. [CO] |
| | Language of team work | CO: Okay we will continue to help each other to better your health. [Nurse] |
| | Ceding power to the patient | CO: So, you want to stay on the old drugs Patient: Yes, but they are refusing me to stay on the old ones CO: Who is refusing you? Patient: The same people giving us the drugs, I told them I don't want to change but they refused CO: They can't refuse you, you have a right to choose it's your body, it's your health and it's your life Patient: I told them I did not want to, but they have refused CO: No, you won't change [CO] |
| Active engagement of patients as care partners | Patient invited to ask questions | Nurse: Is there anything else you want to ask? Patient: No, I have understood everything you have shared. I do not think there are any questions. Nurse: It is clear? Patient: Yes. Nurse: We are here for you, even as you come through next week, feel free to ask. [Nurse] |
| | Information sharing to empower patients' participation in decision making over care | CO: Where you told about your test results? Patient: Yes, they did inform me. That it is high. CO: What did they say is high, educate me so I know? Patient: They said blood is high, not too sure but it is high. CO: I appreciate you informing me on that. It means the HIV virus in your body has multiplied. The whole essence of this medication is to prevent the multiplication of the virus. There are other people who tend to experience a rise in the viral load. And I am not saying you do not take your medication, as you have informed me that you take your medication always. Patient: So, what causes the increase in the virus even when a person takes their medication regularly? |
| | | CO: It could be due to the fact that you get sick every now and then, for most people when they fall ill every now and then, their immune system is weakened. For others it is because at some point they stop taking their medication for some time. When this happens, the medication will not work as it is intended. Yet for others, their bodies just stop responding to a type of medication meaning we have to give them a different type so that the immune system can have a boost. Have you understood now? Patient: Yes. [CO] |
| | Appointment scheduling | CO: On which day of October would you like to come? Patient: The medication that I am getting today is for how many months? Clinical Officer: Three months. Patient: So, my next appointment is in October? CO: Yes. Patient: I need to choose a date? CO: Yes, a date on which you prefer to come. Patient: Let me just take a look at the calendar. [CO] |
| Exercising flexible work processes and resource distribution | Facilitating patient referrals | Clinical Officer: Since you are already here pass by MCH (maternal and child health), I will tell the person [counselor] to take you to MCH so she can listen to what they will say. Patient: Okay, thank you. [CO] |
| | Fee exemption | MO: Sometimes when someone has lost a lot of weight we worry because they may have tuberculosis and we do not want to treatment. So here the doctor had written that you will do an X-ray Patient: Is it free? MO: No, it is k20 (1 USD) Patient: I do not have any money with me. Provider: Okay. You will go for the X-ray today. I will ask them to exempt you so we can make sure that you do not have tuberculosis. If the chest is okay, I will put you on medication for preventing tuberculosis. [MO, 21st September 2021] |
| | Offered to give patients extra drugs | Nurse: So, what you do is when you do not know the time when you will be coming back, it is better you come through to the facility before leaving (travelling out of town) and let us know so that we can give you extra medication. Patient: Okay. Nurse: Then when you go somewhere far and the medication has finished, carry this same card with you, you can go anywhere and get medication [Nurse] |
| | Giving out personal contact details | Medical Officer: Apart from getting the medication, we will not ignore that complaint (patient complaint) so I am asking you to go to the lab, after that we will see if there is a problem. Since you want to tell your wife about this issue (HIV status), I will give you my number so that when she returns back home from her mother's house you can ask her to talk to me. I will invite her to come here. I will emphasize that she comes here, she should not be scared. HIV is now under control as long as you follow the instructions and lead a healthy life, you can just see how people are living now. Before, when someone is positive, they used to be defeated but these days medication is available [MO] |
## Narrowing hierarchical distance
Traditional power dynamics within the health care system preference the provider. Thus, intentional actions by providers to narrow the hierarchical distance between providers and patients is an act of power-sharing. This power-sharing lays the foundation for a positive patient-provider interaction, which is core to the practice of patient-centered care and taking a biopsychosocial perspective to care [35]. This power-sharing was the most frequent DP practice in the data practiced the most by nurses, manifesting in multiple applications of DP, highlighted below.
## Welcoming patients
Welcoming patients meant health providers spent some time chatting with patients before beginning the diagnostic process. Some health providers made kind gestures including inviting patients to take a seat, greeting them, and telling them that they were 'welcome' at the health facilities. The greetings were extended to inquire about their family's well-being, and when patients talked about a difficulty they, a spouse, or another family member was having, the health providers took time to listen and offered supportive words or advice to help them deal with the situation, even if it wasn't related to health.
The welcoming sessions sometimes ended with the health provider and patient sharing a joke and laughing together, or light banter about a general topic such as the weather.
Occasionally, patients were thanked, congratulated, and commended for visiting the health facilities and providers expressed happiness about their next appointment. Other providers sought patient feedback on the service they had received that day.
## Inviting introductions
To put patients at ease, some health providers introduced themselves at the outset of a consultation by revealing their names and positions in the health facility. Despite receiving the patient's name from the patient file, health providers extended introductions to patients by inviting them to introduce themselves as well. Some providers just confirmed the patient's name, while others inquired as to whether or not the patient knew them or had previously dealt with them.
To get patients comfortable chatting and answering personal questions about their health and behavior, some health providers advised patients to ‘feel free' and shared that they ‘respected' their opinions during introductions.
## Offered apology
Rarely, apologies were used to acknowledge and explain service shortcomings. Health providers expressed regret for service issues that resulted in long wait times and failure to process whole blood count and CD4 count testing for patients
## Language of teamwork
A few Clinical Officers and Nurses used language that emphasized the idea of an equal care partnership, instead of an expert provider / recipient patient relationship. Providers used phrases such as “will continue to help each other…” and “we can educate one another” to ‘recognize' patients and narrow the hierarchical gap.
Conversely, the analysis of patient-provider interactions revealed that some health providers mostly MOs maintained hierarchical discursive patterns, hurrying through the consult, neglecting to greet, welcome or introduce themselves and presumably sticking to prescribed HIV related inquiries.
In one instance health providers imposed a punishment on a patient for missing a previous appointment. The patient was told that because they were late for their appointment, they would be seen at the end of the queue of patients who were waiting to be seen that day, even though the patient arrived early in the day. When the patient informed their Clinical Officer, he agreed to the punishment.
## Active engagement of patients as care partners
Traditionally, the provider-patient dynamic would be that a provider made decisions and a patient was expected to do what the provider said, without discussion or questions. Changing this behavior includes offering the patient time and space to direct the dialogue within a consultation, equipping the patient to exercise agency over their own care, and ceding power in health system engagement or treatment plans directly to the patient. This demonstrates enhancing core dimensions of PCC including patient involvement in care, recognizing the patient as a unique person, patient empowerment, and patient-clinician communication [5].
## Patient invited to ask questions
One of the most commonly coded sub-themes, Nurses and COs encouraged patient participation in discussions and decisions about their treatment by inviting them to ask questions. They told patients to ‘feel free to ask,' stating ‘we are here for you.' One nurse reminded patients that they had the ‘right' to ask questions of their providers. They recommended individuals not just to rely on information from other sources, but to double-check by speaking with health care providers. Medical doctors only did this on rare occasions.
As a result of the invitations to ask questions, patients asked about care and treatment themes, for example, ‘what happens when you do not take your medication for like 4 days', and ‘If you are not feeling well-and you have a clinic card, do you come straight here (clinician office)'. Women frequently asked questions concerning reproductive health in the context of HIV as illustrated below: In some cases, particularly among the MOs the conventional dynamic of the provider as the ‘knowledge authority' and the expectation placed on patients to follow the provider's decisions was demonstrated in some circumstances. Some providers asked closed questions throughout the interaction to control the engagement.
## Information sharing to empower patients' participation in decision making over care
Some health providers, most commonly COs, often exercised their discretion over time and information to share knowledge with patients to help them gain a better understanding and ‘empower' them to take charge of their own care. This meant offering a level of detail about their health condition and assessments that went beyond short answers to equip the patient with information as a resource. This demonstrates enhancing core dimensions of PCC including patient involvement in care, patient empowerment, patient information, and patient-clinician communication [13, 35].
To keep patients informed, providers took the time to converse with them, volunteering information about conditions, care decisions, and processes. For example, providers explained in non-clinical terms how ART worked in the body, what a viral load measured, different ways ART could be accessed, when patients should expect to draw and receive lab results, what U=U means, and demonstrative of changes related to the global pandemic, details about the COVID-19 vaccine. This information was offered both based on provider judgement and in response to specific patient inquires. When patients inquired, health workers provided answers tailored to their unique needs and concerns. Some patients expressed an awareness of why certain procedures were crucial after hearing the material.
On the other hand, some health providers withheld information to orient patients and clarify treatment processes. They made referrals without first discussing the patient's condition or concerns or assessing the co-incident symptoms.
## Ceding power to the patient
Although uncommon in DP practices, some health providers attempted to share care and treatment decision-making 'powers' with patients through their discursive approaches. They also listened to what the patients had to say and respected their choices. Patients' concerns were conveyed through information exchange, and health providers educated them of their 'rights' to participate in care decision-making.
Demonstrating power transfer, a health provider educated a patient of their power, saying, ‘you have a right to choose, it's your body, it's your health and it's your life' in a case in which a patient was being ordered to switch from one ARV to another within their first-line regime and they were against it, as illustrated below:
## Appointment scheduling
The vast majority of ‘next appointment dates' were dictated by the providers. However, some COs and nurses explicitly invited patient engagement. Some practices included a provider suggesting a day and asking if patients were ‘comfortable' with the appointment day, while others asked for a suggested ‘suitable' or ‘preferred' day.
In addition, to limit the number of clinic visits, health staff coordinated some patients' appointment dates for drug pick-ups and laboratory tests such as viral load testing. Some providers advised patients that if something came up during their appointment or if they fell ill and need medical assistance, they may rearrange their appointment day.
Patient participation was not always sought with authoritative power remaining with the health providers on the other hand. They gave instructions, denying patients chance to participate in treatment decisions. This was especially apparent when it came to scheduling clinic appointments and placing emphasis on treatment adherence.
## Exercising flexible work processes and resource distribution
In a few instances, some health providers went above and beyond their routine duties to help patients in practical ways, such as facilitating patient referrals, offering consultations outside of regular appointments, telephone consultations, and providing more drugs to meet patients' needs.
When making referrals, some providers made personal referrals to other health providers or asked for favors by writing to them notes or following up with them to ensure a smooth process for the patient.
## Fee exemption
In one instance, an MO, concerned about a patient's weight loss, exempted her from the 20 Kwacha x-ray fee to ensure that results were not delayed. After the participant explained that she did not have any money with her, the Medical Officer said, “You will go for the X-ray today. I will ask them to exempt you so we can make sure that you do not have tuberculosis.” This same provider offered personalized services, engaging with the patient's family member at the patient's request:
## Offered to give patients extra drugs
Offers of extra drugs or alternative arrangements such as enabling drug pick-ups by a patient representative revealed discretionary power practices based on efforts to keep patients' adherent to treatment regardless of livelihood or personal circumstances.
## Giving out personal contact details
Some MOs advanced the notion of discretionary power by providing personalized service by giving personal phone numbers to patients who needed follow-up telephone consultations or guidance for their family members as shown below.
We found some missed opportunities for health providers to exercise flexibility when patients tried to negotiate for care services and support beyond the routine delivery of care services, but the health providers did not adapt or respond to meet those needs. Some patients requested for different appointment dates, increased drug refill quantities and collection of drugs by another person due to work commitments. However, the health providers declined these requests, reminding patients condescendingly ‘to be grateful for the free medication' as remarked by one Clinical Officer.
When comparing the types of DP practices used by the three health cadres, it was found that nurses consistently sought to reduce the hierarchical distance between them and their patients by welcoming patients and inviting introductions, followed by COs and MOs, who focused on the standard HIV treatment protocol. Both nurses and COs engaged patients as care partners inviting questions and information sharing while MOs facilitated resource distribution and provided personal contact information to support patients outside of usual working hours.
## Discussion
Through our analysis of patient-provider clinical interactions, we found that providers primarily utilized the power at their discretion to implement more patient-centered HIV care by narrowing the hierarchical distance between themselves and patients and engaging patients as partners through inviting patient questions and sharing information. While these are less traditional demonstrations of discretionary power [1, 5, 10] the time, provider manner, and information resources associated with these acts are very much at the discretion of the provider. Consistent with Lipsky's definition of SLB, providers can follow HIV guidelines while maintaining a hierarchy, but the choice to promote equality and patient empowerment has been associated with improved HIV outcomes [5, 7, 14]. Attention to these power dynamics in the implementation of PCC is particularly important because, while these were the most frequent, they were not universal, even within these facilities where an intervention to improve patient-provider interactions and PCC was ongoing.
Information sharing and inviting patients to ask questions may play a particularly prominent role in empowering patients to participate in decision making, engage as care partners with health providers and foster agency and self-efficacy over their own HIV care (35–37). Our findings suggest that inviting patients to ask questions is a specific use of the time and avenue to provide information at their discretion to improve an interaction and work toward establishing a relationship with a patient [24, 37]. These findings concur with extant literature [1, 23, 38] and a Nurse provider in our sample who directly educated patients on their right to ask questions of providers, that beyond providing training in PCC to providers, mechanisms to teach patients of their rights to participate in the patient-provider relationship may enhance PCC [23, 38]. Lipsky recommends supporting front line workers through “ongoing processes of supportive criticism and inquiry to take ownership” [7]. In this context, ownership of PCC principals and so exercise their discretionary power in pursuit of PCC goals and to include these power-sharing indicators into performance evaluations as health providers often change their behavior to reflect what is being measured [7].
We also noted interactions that had limited information with no opportunities for asking questions presented to patients [1]. Changes in the type of information provided, the amount of information provided, and the language used to give information are all required [39]. Patients require information that they can comprehend to exercise critical judgment in decisions about their health care [23, 39]. Therefore, providers need to be proactive in facilitating these patient empowerment processes. The first step in doing this is for providers to acknowledge patients as partners, through ceding some power to patients to give them ‘courage' in having a say in their HIV care and developing a positive relationship with patients [1, 23, 40, 41]. In a study among Nurses, they argued that they had no time to share information or answer questions [1]. However, patients may negotiate their relationship with health providers until a mutually satisfying relationship is reached [23, 39]. The Clinical care protocols provide a benchmark for health providers against which to deliver quality care with patients however, implementation in resource limited settings may shape the discretionary application of decisions and actions to manipulate the system to meet the needs of patients [7, 9, 12].
Although operating within policy frameworks and rules, findings highlight that there were few cases when providers exercised discretion to offer flexible services such as agreeing on dates when patients would attend a next appointment or providing patients with contact numbers to foster phycho-social and emotional support for patients and their family. However, their mere existence demonstrates the possibility of applying them to implement PCC and suggests that further work is needed to increase the application of these practices, or to reduce the provider-level discretion involved in their application [9]. Usually, policy-practice divergence occurs as a result of the flexibility with which implementers interpret policy and make decisions without their involvement in ‘higher level' policy formulation and its implementation [5, 7, 9]. For example, while choosing a differentiated service delivery (DSD) model that improves patient HIV care access likely requires provider time and openness to patient preferences, a policy that providers must ask patients if the reason they missed an appointment was related to an inflexible appointment schedule and, if so, offer DSD models may improve PCC without relying on provider discretion [42]. In Zambia, patients recognized that while health providers care for their well-being, ‘they care rudely' by shouting at patients, shutting down questions and conversations, and using their discretionary powers without considering patient needs, for example, exercising inflexible opening, break and closing hours and thereby prolonging wait times in overcrowded non-private conditions [25]. In another study in Zambia about patient preferences, patients demonstrated strong preferences for kind vs. rude health providers [43]. While there have been calls for health providers to provide ‘patient-centered' care by exercising ‘flexibility' to meet patients' varying needs and circumstances [25], evidence suggests that current HIV care systems and models are not sufficiently client-centered to ensure agility and adaptability to client circumstances [44, 45]. In an analysis to re-orient the South African health system toward public health goals and patient centered care highlighted challenges related to dispersed accountability, complex rules and hierarchical procedures [12].
The power differentials between health professionals are determined by the hierarchy of health professionals, which influences the use of discretionary power practices in patient engagements (46–48) and could facilitate the delivery of inflexible care practices. In the hierarchy of the study health providers, MOs with the traditional senior positions have the most power and autonomy followed by the COs, and then nurses who have relatively less power in organized health care [33, 34, 49]. These power dynamics may affect the strategic choices made by each health professional about whether to narrow the hierarchical distance with patients and engage patients as care partners, and these decisions directly influence patient experiences (46–48). For example, we found that power-sharing procedures were largely practiced by nurses in this study, but MOs rarely attempted to share decision-making power with patients or facilitate direct patient participation in care decisions. This suggests that work culture change based on front-line provider choices should be nurtured through training, sensitization, and support for providers in adopting patterns of practice, routines, and policy interpretations to deal with service dynamics (50–53). These changes could, for example be facilitated by new forms of leadership that enable sensemaking in support of building patient-provider relationships and inter-professional collaboration across organizational boundaries (12, 50–53). Leadership teams have a vital role to play in aligning values, fostering employee relationships, and supporting shared understanding of work culture transformation and mutual accountability (50–53). The impact that such effects can have on the patient care experience are well-known [54, 55] including improved patient education, treatment adherence, and self-care on the part of patients, all of which lead to improved health outcomes (50–55).
## Limitations
The estimated patient-provider interaction time is based on the length of the interaction recording. This may over-estimate the actual interaction time, as it may include interruptions, such as a provider being called out of the consultation room, that were not accounted for in the record of the interaction length. With regards to the providers, not all of them included were trained in PCC during the study implementation as the training took a facility approach involving different teams to improving PCC. Another limitation is that the analysis did not compare DP practice in the control facilities and focused on provider behavior in intervention facilities. Further, patient experiences were not explored to understand whether they were satisfied with the provider's discretionary use of power. However, the findings are useful in that they can be used to inform about the use of discretionary power among providers in health settings and how this affects delivery of PCC.
## Future research
To better understand the dynamics of DP practice, more research into the relationship of providers to the practice of different DP techniques would be beneficial. For example, why were nurses able to be more welcoming than MOs and COs. This information could help to unravel health care system complexities that could influence provider policy implementation, as well as serve as a foundation for improving cadre in needed areas, changing the structure or organization of work to improve information sharing and discretionary use of time, and taking into account power dynamics, all with the goal of improving patient care.
## Conclusion
To promote the implementation of patient-centered care, it is critical to understand how health providers exercise DP to support or impede a patient-centered care experience. Health systems can support patient-centered care within patient-provider interactions by training and mentoring providers to share power with patients by narrowing the hierarchical distance between them and patients, sharing information and engaging patients as partners, and offering flexible services, all of which are inherent in the practice of patient-centered care.
## Data availability statement
The datasets presented in this article are not readily available because the Government of Zambia allows data sharing when applicable local conditions are satisfied. To request data access, contact the Secretary to the CIDRZ Ethics and Compliance Committee/Head of Research Operations, Hope Mwanyungwi, mentioning the intended use for the data. Requests to access the datasets should be directed to [email protected].
## Ethics statement
The studies involving human participants were reviewed and approved by the University of Zambia Biomedical Research Ethics Committee (UNZABREC), the University of Alabama at Birmingham (UAB), and the National Health Research Authority and the Zambian Ministry of Health. The patients/participants provided their written informed consent to participate in this study.
## Author contributions
CBM, LB, EG, AS, and AM supported conceptualization. CM, NM, KL, MF, HN, SS, and KS supported data collection. CM and LB conducted formal analysis. CM, LB, AS, NM, EG, AM, IS, and CBM contributed to data interpretation. IS, CBM, CH, and EG acquired study funding. AS, KS, CH, IS, CBM, AM, EG, and LB were study investigators. CM, LB, AS, NM, AM, and EG designed the methodology. SS, KS, NM, and CM conducted project administration. CM, NM, AS, and LB wrote the original manuscript draft. All co-authors reviewed and edited the final draft. All authors contributed to the article and approved the submitted version.
## Funding
This research was supported by the Bill and Melinda Gates Foundation grant number INV-010563.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Henderson S. **Power imbalance between nurses and patients: a potential inhibitor of partnership in care**. *J Clin Nurs.* (2003) **12** 501-8. DOI: 10.1046/j.1365-2702.2003.00757.x
2. McQueen A. **Nurse–patient relationships and partnership in hospital care**. *J Clin Nurs.* (2000) **9** 723-31. DOI: 10.1046/j.1365-2702.2000.00424.x
3. Beisecker AE. **Patient power in doctor-patient communication: What do we know?**. *Health Commun* (1990) **2** 105-22. DOI: 10.1207/s15327027hc0202_4
4. Haug MR, Lavin B. **Practitioner or patient-who's in charge?**. *J Health Soc Behav.* (1981) **1** 212-29. DOI: 10.2307/2136517
5. Gilson L, Schneider H, Orgill M. **Practice and power: a review and interpretive synthesis focused on the exercise of discretionary power in policy implementation by front-line providers and managers**. *Health Policy Plann* (2014). DOI: 10.1093/heapol/czu098
6. Evans T, Harris J. **Street-level bureaucracy, social work and the (exaggerated) death of discretion**. *Br J Soc Work* (2004) **34** 871-95. DOI: 10.1093/bjsw/bch106
7. Lipsky M. *Street-Level Bureaucracy: Dilemmas of the Individual in Public Service* (1980)
8. Erasmus E. **The use of street-level bureaucracy theory in health policy analysis in low- and middle-income countries: a meta-ethnographic synthesis.**. *Health Policy Plann* (2014) **29** 70-8. DOI: 10.1093/heapol/czu112
9. Atinga RA, Agyepong IA, Esena RK. **Ghana's community-based primary health care: why women and children are ‘disadvantaged'by its implementation**. *Soc Sci Med* (2018) **201** 27-34. DOI: 10.1016/j.socscimed.2018.02.001
10. Nunes J, Lotta G. **Discretion, power and the reproduction of inequality in health policy implementation: practices, discursive styles, and classifications of Brazil's community health workers**. *Soc Sci Med* (2019) **242** 112551. DOI: 10.1016/j.socscimed.2019.112551
11. Campbell C, Scott K, Skovdal M, Madanhire C, Nyamukapa C, Gregson S. **good patient? How notions of ‘a good patient'affect patient-nurse relationships and ART adherence in Zimbabwe**. *BMC Infect Dis.* (2015) **15** 1-1. DOI: 10.1186/s12879-015-1139-x
12. Gilson L, Elloker S, Olckers P, Lehmann U. **Advancing the application of systems thinking in health: South African examples of a leadership of sensemaking for primary health care**. *Health Res Policy Sys.* (2014) **12** 1-3. DOI: 10.1186/1478-4505-12-30
13. Scholl I, Zill JM, Härter M, Dirmaier J. **An integrative model of patient-centeredness–a systematic review and concept analysis**. *PLoS ONE.* (2014) **9** e107828. DOI: 10.1371/journal.pone.0107828
14. de Man J, Mayega RW, Sarkar N, Waweru E, Leys M, van Olmen J. **Patient-centered care and people-centered health systems in sub-saharan africa: why so little of something so badly needed?**. *Int J Person Center Med.* (2016) **6** 162-73. DOI: 10.5750/ijpcm.v6i3.591
15. 15.World Health Organization. Updated Recommendations On Service Delivery For The Treatment and Care of People Living With HIV. Geneva: World Health Organization (2021).. *Updated Recommendations On Service Delivery For The Treatment and Care of People Living With HIV* (2021)
16. Eshun-Wilson I, Rohwer A, Hendricks L, Oliver S, Garner P. **Being HIV positive and staying on antiretroviral therapy in Africa: a qualitative systematic review and theoretical model**. *PLoS ONE.* (2019) **14** e0210408. DOI: 10.1371/journal.pone.0210408
17. Herce ME, Chi BH, Liao RC, Hoffmann CJ. **Re-thinking linkage to care in the era of universal test and treat: insights from implementation and behavioral science for achieving the second 90**. *AIDS Behav.* (2019) **23** 120-8. DOI: 10.1007/s10461-019-02541-5
18. Shukla S, McCoyd JL. **HIV Health care providers as street-level bureaucrats: unreflective discourses and implications for women's health and well-being**. *Ethics Soc Welfare* (2019) **13** 133-49. DOI: 10.1080/17496535.2018.1539507
19. Colvin CJ, Konopka S, Chalker JC, Jonas E, Albertini J, Amzel A. **systematic review of health system barriers and enablers for antiretroviral therapy (ART) for HIV-infected pregnant and postpartum women**. *PLoS ONE.* (2014) **9** e108150. DOI: 10.1371/journal.pone.0108150
20. Chikwari CD, Dringus S, Ferrand RA. **Barriers to, and emerging strategies for, HIV testing among adolescents in sub-Saharan Africa**. *Curr Opin HIV AIDS* (2018) **13** 257-64. DOI: 10.1097/COH.0000000000000452
21. Kiplagat J, Tran DN, Barber T, Njuguna B, Vedanthan R, Triant VA, Pastakia SD. **How health systems can adapt to a population ageing with HIV and comorbid disease**. *The Lancet HIV* (2022) **23** 9. DOI: 10.1016/S2352-3018(22)00009-1
22. Wilson BM. **Promoting compliance: the patient-provider partnership**. *Adv Renal Replac Therapy* (1995) **2** 199-206. DOI: 10.1016/S1073-4449(12)80053-7
23. Hewison A. **Nurses' power in interactions with patients**. *J Adv Nurs.* (1995) **21** 75-82. DOI: 10.1046/j.1365-2648.1995.21010075.x
24. Jaramillo J, Yadav R, Herrera R. **Why every word counts: towards patient-and people-centered tuberculosis care**. *Int J Tubercul Lung Dis* (2019) **23** 547-51. DOI: 10.5588/ijtld.18.0490
25. Mwamba C, Sharma A, Mukamba N, Beres L, Geng E, Holmes CB. **They care rudely!': resourcing and relational health system factors that influence retention in care for people living with HIV in Zambia**. *BMJ Global Health* (2018) **3** e001007. DOI: 10.1136/bmjgh-2018-001007
26. Jewkes R, Abrahams N, Mvo Z. **Why do nurses abuse patients? Reflections from South African obstetric services**. *Soc Sci Med* (1998) **47** 1781-95. DOI: 10.1016/S0277-9536(98)00240-8
27. Layer EH, Kennedy CE, Beckham SW, Mbwambo JK, Likindikoki S, Davis WW. **LTC Tanzania collaborative study team. Multi-level factors affecting entry into and engagement in the HIV continuum of care in Iringa, Tanzania**. *PLoS ONE.* (2014) **9** e104961. DOI: 10.1371/journal.pone.0104961
28. Zakumumpa H, Kiweewa FM, Khuluza F, Kitutu FE. **“The number of clients is increasing but the supplies are reducing”: provider strategies for responding to chronic antiretroviral (ARV) medicines stock-outs in resource-limited settings: a qualitative study from Uganda**. *BMC Health Serv Res.* (2019) **19** 1-1. DOI: 10.1186/s12913-019-4137-7
29. Sikazwe I, Eshun-Wilson I, Sikombe K, Beres LK, Somwe P, Mody A. **Patient-reported Reasons for stopping care or switching clinics in zambia: a multisite, regionally representative estimate using a multistage sampling-based Approach in Zambia**. *Clin Infect Dis* (2021) **73** e2294-302. DOI: 10.1093/cid/ciaa1501
30. Chammartin F, Zürcher K, Keiser O, Weigel R, Chu K, Kiragga AN. **Outcomes of patients lost to follow-up in African antiretroviral therapy programs: individual patient data meta-analysis**. *Clin Infect Dis.* (2018) **67** 1643-52. DOI: 10.1093/cid/ciy347
31. Roter D, Larson S. **The Roter interaction analysis system (RIAS): utility and flexibility for analysis of medical interactions**. (2002) **46** 243-51. DOI: 10.1016/S0738-3991(02)00012-5
32. Anderson R. **Thematic content analysis (TCA)**. (2007) **3** 1-4
33. Ferrinho P, Siziya S, Goma F, Dussault G. **The human resource for health situation in Zambia: deficit and maldistribution**. *Hum Resour Health.* (2011) **9** 1-1. DOI: 10.1186/1478-4491-9-30
34. Morris MB, Chapula BT, Chi BH, Mwango A, Chi HF, Mwanza J. **Use of task-shifting to rapidly scale-up HIV treatment services: experiences from Lusaka, Zambia**. *BMC Health Serv Res.* (2009) **9** 1-9. DOI: 10.1186/1472-6963-9-5
35. Epstein RM, Street RL. **The values and value of patient-centered care**. *Annals Fam Med* (2011) **9** 100-3. DOI: 10.1370/afm.1239
36. Sahlsten MJ, Larsson IE, Plos KA, Lindencrona CS. **Hindrance for patient participation in nursing care**. *Scand J Caring Sci.* (2005) **19** 223-9. DOI: 10.1111/j.1471-6712.2005.00336.x
37. Sahlsten MJ, Larsson IE, Sjöström B, Plos KA. **Nurse strategies for optimising patient participation in nursing care**. *Scand J Caring Sci.* (2009) **23** 490-7. DOI: 10.1111/j.1471-6712.2008.00649.x
38. Jarrett N, Payne S. **A selective review of the literature on nurse-patient communication: has the patient's contribution been neglected'**. *J Adv Nurs.* (1995) **22** 72-8. DOI: 10.1046/j.1365-2648.1995.22010072.x
39. Bellows M, Kovacs Burns K, Jackson K, Surgeoner B, Gallivan J. **Meaningful and effective patient engagement: what matters most to stakeholders**. *Pat Exp J.* (2015) **2** 18-28. DOI: 10.35680/2372-0247.1069
40. Johnsson A, Wagman P, Boman Å, Pennbrant S. **Striving to establish a care relationship—Mission possible or impossible?—Triad encounters between patients, relatives and nurses.**. *Health Expect.* (2019) **22** 1304-13. DOI: 10.1111/hex.12971
41. Morse JM. **Negotiating commitment and involvement in the nurse-patient relationship**. *J Adv Nurs.* (1991) **16** 455-68. DOI: 10.1111/j.1365-2648.1991.tb03436.x
42. Roy M, Bolton Moore C, Sikazwe I, Holmes CB. **A review of differentiated service delivery for HIV treatment: effectiveness, mechanisms, targeting, and scale**. *Curr HIV/AIDS Rep.* (2019) **16** 324-34. DOI: 10.1007/s11904-019-00454-5
43. Zanolini A, Sikombe K, Sikazwe I, Eshun-Wilson I, Somwe P, Bolton Moore C. **Understanding preferences for HIV care and treatment in Zambia: evidence from a discrete choice experiment among patients who have been lost to follow-up**. *PLoS Med.* (2018) **15** e1002636. DOI: 10.1371/journal.pmed.1002636
44. El-Sadr WM, Harripersaud K, Rabkin M. **Reaching global HIV/AIDS goals: what got us here, won't get us there**. *PLoS Med* (2017) **14** e1002421. DOI: 10.1371/journal.pmed.1002421
45. Genberg B, Wachira J, Kafu C, Wilson I, Koech B, Kamene R. **Health system factors constrain HIV care providers in delivering high-quality care: perceptions from a qualitative study of providers in Western Kenya**. *J Int Assoc Prov AIDS Care (JIAPAC).* (2019) **18** 2325958218823285. DOI: 10.1177/2325958218823285
46. Reeves S, Harris R. **Interprofessional teamwork in health and social care: key tensions and future possibilities**. *Power Dist Perspect.* (2016) **10** 173. DOI: 10.1515/9783110493351-011
47. McDonald J, Jayasuriya R, Harris MF. **The influence of power dynamics and trust on multidisciplinary collaboration: a qualitative case study of type 2 diabetes mellitus**. *BMC Health Serv Res.* (2012) **12** 1-0. DOI: 10.1186/1472-6963-12-63
48. Fox A, Reeves S. **Interprofessional collaborative patient-centred care: a critical exploration of two related discourses**. *J Interprof Care* (2015) **29** 113-8. DOI: 10.3109/13561820.2014.954284
49. Mbindyo P, Blaauw D, English M. **The role of Clinical Officers in the Kenyan health system: a question of perspective**. *Hum Resour Health.* (2013) **11** 1-1. DOI: 10.1186/1478-4491-11-32
50. Trinchero E, Brunetto Y, Borgonovi E. **Examining the antecedents of engaged nurses in I taly: Perceived Organisational Support (POS); satisfaction with training and development; discretionary power**. *J Nurs Manag.* (2013) **21** 805-16. DOI: 10.1111/jonm.12143
51. Mash R, Christodoulou M, De Sa A. **How to change organisational culture: Action research in a South African public sector primary care facility**. *Af J Prim Health Care Family Med* (2016) **8** 1-9. DOI: 10.4102/phcfm.v8i1.1184
52. Shaller D. *Patient-Centered Care: What Does It Take* (2007)
53. Ponte PR, Conlin G, Conway JB, Grant S, Medeiros C, Nies J. **Making patient-centered care come alive: achieving full integration of the patient's perspective**. *JONA: J Nurs Administ* (2003) **33** 82-90. DOI: 10.1097/00005110-200302000-00004
54. Stewart M, Brown JB, Donner A, McWhinney IR, Oates J, Weston WW. **The impact of patient-centered care on outcomes**. *J Fam Pract* (2000) **49** 796-804. PMID: 11032203
55. Stewart MA. **Effective physician-patient communication and health outcomes: a review**. *CMAJ: Can Med Assoc J* (1995) **152** 1423. PMID: 7728691
|
---
title: Strategies for effective implementation and scale-up of a multi-level co-designed
men's health initiative “Sheds for Life” in Irish Men's Sheds
authors:
- Aisling McGrath
- Noel Richardson
- Niamh Murphy
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012692
doi: 10.3389/frhs.2022.940031
license: CC BY 4.0
---
# Strategies for effective implementation and scale-up of a multi-level co-designed men's health initiative “Sheds for Life” in Irish Men's Sheds
## Abstract
Sheds for *Life is* a gender-specific tailored men's health initiative engaging “hard-to-reach” men in the Men's Shed setting in Ireland. It is implemented by multiple stakeholders at individual, provider, organization and systems level and thus multiple contextual factors influence its scalability. This research used established implementation science frameworks to guide participatory research approaches that captured the process and identified facilitators of and barriers to implementation and scale-up. Active recruitment, co-design processes, leadership and stakeholder engagement emerged as key facilitators of implementation. Prominent barriers were institutional capacity and funding. Acceptability, adoption and appropriateness of the initiative were high among stakeholders with sustainability largely contingent on funding and staff resources. Findings make a valuable contribution to knowledge by capturing the process involved in the implementation of a complex multi-level men's health intervention. It provides a “how to” guide of strategies to engage hard-to-reach men with health promotion, the operationalization and application of implementation frameworks in community-based health promotion, and the implementation of health promotion in Men's Sheds. Documented barriers and facilitators that impact implementation of a community-based men's health program are rare and provide a valuable blueprint for practitioners, researchers and policy makers in the field.
## Introduction
The burden of ill health in men is caused by multiple complex factors that are particularly exacerbated for vulnerable groups of socially disadvantaged men [1, 2]. While it may be perceived that traditional masculine ideals which impede positive men's health behaviors are typical of mainly older men, evidence suggests these barriers remain a systemic issue that continue to pervade through generations [3]. Understanding the complexities of masculinities within health systems and how men engage with, and are impacted by them has highlighted a need for tailored men's health programs underpinned by gender-specific approaches [4, 5]. This fact is further compounded by the disparity in mortality for men during COVID-19 which was likely a consequence of failure to invest in men's health [6, 7]. This need is particularly pertinent for men who are at risk of being more isolated from, or reticent about, accessing formal health services or social supports due to geography, experiences of mental health issues, social disadvantage, unemployment, low educational attainment or significant changes in life course (e.g., retirement)—groups that are considered “hard-to-reach” (HTR) in health endeavors [8]. Moreover, designing models of care that are accessible to men and that address changing masculinities across the life course, can be instrumental in reaching out to HTR men while simultaneously acknowledging their diversity [9].
The Men's Sheds (“Sheds”) are autonomous, grassroots organizations that originated in Australia in the 1980s and grew exponentially in Ireland from 2011 following the economic recession. Founded and sustained by Shed members (“Shedders”), membership within Sheds attracts diverse representations of men from different socioeconomic backgrounds, and importantly, are effective in attracting cohorts of HTR men (10–12). The proliferation of Sheds across Ireland was testament to a growing need for men to identify with a space that facilitated meaning, social support, safety and belonging [10, 13]. By virtue of their grassroots, member focused approach, Sheds are variable spaces that differ in size, range of activities (e.g., woodwork, music, gardening, art, and mechanics) and resources but have commonality in offering men a safe and familiar environment that fosters a sense of social support and belonging, through developing new skills, shared projects, team work and camaraderie [14, 15]. Not surprisingly, Sheds have therefore been identified as inherently health promoting spaces for men [13, 16]. Based upon their inherent health promoting qualities and ready access to men who may be reticent to engage with traditional health services, Sheds represent an attractive setting in which to build structured health initiatives. In light of this, Sheds have emerged as an exemplar for the promotion of men's health and wellbeing by health and social policy makers, earmarked as spaces that are capable of engaging HTR men in health endeavors [10, 17]. Notwithstanding the fact that Sheds potentially offer a strong foundation upon which to build structured health promotion, tension may arise from imposing formal healthcare upon the informal setting of the Sheds, where its informality is an integral element to its inherent health promotion and where formality may be the very convention men seek to resist [10, 18]. Nevertheless, Shed members (Shedders) have demonstrated an appetite for health promotion in Sheds [10], suggesting it is timely to capitalize on this opportunity. The critical consideration in the design, implementation and evaluation of health promotion programs in *Sheds is* that Shedders are at the center of all decision making and that the ethos of the Shed environment is preserved [10, 18].
Recognizing the utility of Sheds as a means to engage HTR men with health while also understanding the need to prioritize wellbeing for its membership in a tailored and respectful way, the Irish Men's Sheds Association (IMSA) first developed the concept of Sheds for Life (SFL) in 2016 [19]. Sheds for *Life is* a men's health initiative tailored to the Shed setting in Ireland. Through ongoing consultation with stakeholders, Sheds for Life was developed and refined into a 10-week program consisting of four core pillars of a health check, healthy eating, physical activity and mental health along with several option components focusing on life skills and disease prevention [19]. A detailed protocol is available which outlines the various components of SFL [19] and the development of this approach will also be further discussed in the context of this research (see Results). Prior to the implementation of a structured SFL program, the IMSA embarked on scoping work at various regional Shed meetings to engage Shedders to identify their health needs and preferences. The IMSA also began to develop partnerships with provider organizations who were actively seeking to engage HTR groups of men in their health promoting initiatives. This resulted in the piloting of discrete wellbeing workshops in Sheds [19]. Initial scoping work which sought to investigate how SFL piloting was experienced in practice determined that respecting the Shed environment was critical to the acceptability of SFL and strategic evaluation of the development of SFL would be required to facilitate effective implementation [10]. In June 2018 the current authors commenced the formal evaluation of SFL with a dual focus on both efficacy and implementation.
Findings from research show that in order to engage men, particularly those who are HTR, health promotion must include men in decision making and encourage a collaborative process involving all key stakeholders; researchers, practitioners, participants and policy makers [10, 20]. Community-based participatory research approaches also emphasize the importance of creating partnerships with the people for whom the research is ultimately meant to benefit [21]. Moreover, SFL scoping work highlighted the importance of strengthening ties with local providers and community organizations, an established strategy when seeking to scale-up programs nationally, especially under real world conditions [10, 22]. This led to a pragmatic study design using community-based participatory research approaches (CBPR) that were geared toward upholding autonomy and increasing the agency of participants [10]. Questions emerged as to the “what” and the “how” of SFL that ought to be evaluated, particularly with regard to reconciling gold standard evaluation methods with the high variability, autonomy and ethos of the Sheds as implementation settings. Moreover, beyond the environment of the Sheds there is also a need to understand the complex intervening variables that act as a backdrop to implementation of SFL (e.g., those at provider, organizational and systems levels) [23, 24]. The use of implementation science can be valuable in identifying barriers and facilitators to effectively implementing programs as well as promoting systematic uptake in real world settings from the outset [25]. Indeed, implementation science encompasses many of the principles of CBPR, with both approaches linked to improved knowledge translation. These include the engagement of key stakeholders to understand contextual factors, a focus on capacity building, partnership in the research process, and systems development through a cyclical and iterative process with a view to long-term sustainability, [21, 24, 26].
Sheds for Life operates within a complex system of shifting elements such as the diverse and variable contexts of the Sheds and the wider implementation environment, including the competing priorities of provider organizations and systems level funding and polices. As a result, there is a need to continually engage current and emerging stakeholders as well as inform key adaptations and processes that are necessary to implement SFL in multiple locations while executing appropriate implementation strategies to embed SFL in the routine environment of the Shed. Indeed, these dimensions continually evolve over time and require on-going monitoring. Thus, this research was guided by a combination of implementation and evaluation frameworks. While implementation science was used to address implementation issues, there is still a delay when following the traditional route of efficacy-effectiveness-implementation. The speed of moving research findings into routine adoption can be improved by considering hybrid designs that combine elements of effectiveness and implementation research [27, 28]. Hybrid designs focus on the dual testing of both effectiveness of the clinical intervention and its implementation. This type of trial design is not dictated by the type of hybrid, meaning that many types of randomized and non-randomized studies can utilize this approach [28]. Hybrid type 2 designs are ideal when there is momentum for implementation in terms of system or policy demands [28] - particularly relevant in the case of SFL where there have been calls to implement targeted health promotion in the Sheds supported by a rich landscape of men's health research and policy in Ireland [10].
Alongside the need to identify suitable programs to engage men with health, there is a lack of practical guidance on how to effectively implement and scale-up heath interventions [24]. In the context of SFL, scale-up is the deliberate effort to increase impact of SFL so as to benefit more Shedders while fostering more sustainable program development that may influence policy [29]. This involves assessing scalability through measuring feasibility, acceptability, costs, sustainability and adaptability [30]. The effectiveness-implementation design of this research aimed to engage all key stakeholders in the development, testing, implementation and scale-up of SFL. It aimed to investigate both the process and effectiveness of the SFL intervention with a focus on the key strategies involved in implementation and future scale-up to maximize reach to HTR men within the non-conventional settings of Sheds and the wider implementation environment. A detailed protocol which outlines the effectiveness-implementation design is available [19] as well as work which describes effectiveness outcomes [31]. This study discusses the implementation research of SFL in terms of the process of implementation, identification of barriers, facilitators and strategies that impact on implementation outcomes, guided by established implementation frameworks [24, 32, 33]. This paper addresses an important gap in the literature by applying an implementation lens to the evaluation of a community-based men's health promotion program using gender-specific approaches. Findings from this research can play a significant role in determining the implementation effectiveness, sustainability, and potential scale-up of the SFL initiative and, more broadly, in terms of the wider rollout of community-based men's health programs.
## Research design
A mixed methods process evaluation was used to guide the implementation of SFL guided by a combination of applicable implementation frameworks [24, 32, 33]. This consisted of a combination of focus groups, interviews, observations, questionnaires and administrative data (e.g., attendance records). In order to explain or understand implementation outcomes, the perspectives and experiences of a broad representation of stakeholders at the participant, provider, organization and wider systems level were sought. Purposive sampling was used to identify key stakeholders for interview who could inform implementation outcomes across the multi-level implementation environment. Mixed methods were used to inform implementation outcomes. A diverse range of Shed member views were sought from Shed settings based on Shed size (small/large), and geographical location (urban/rural). At the provider, organization and systems level a diverse range of views were sought based on their role within SFL e.g., funder, deliverer, partner, implementer. Semi-structured topic guides and interview schedules were developed for focus groups and interviews. These were designed using a hybrid deductive-inductive approach applying implementation frameworks to assess implementation outcomes, exploring topics areas such as adoption, acceptability, and appropriateness of SFL. These were also designed to allow room for exploring attitudes toward SFL as well as changes in knowledge and behaviors where applicable. The PRACTical planning for Implementation and Scale-up guide (PRACTIS) guide was used as part of an iterative process to characterize parameters of the implementation setting, engage key stakeholders, identity implementation barriers and facilitators, and address potential barriers to implementation within the evolving implementation climate [24]. Ongoing consultation with stakeholders was deemed appropriate to the implementation approach as contextual shifts can be unpredictable and assessment of the broader implementation environment required flexibility and iteration [34]. The first author was positioned within the organization (IMSA) for the duration of the research which informed ongoing monitoring of the implementation approach. Alongside this, semi-structured interviews ($$n = 19$$; Provider level $$n = 15$$, Organizational level $$n = 2$$, Systems level $$n = 2$$) were conducted at provider, organizational and systems level using interview schedules which were designed based on the Consolidated Framework for Implementation Research (CFIR) constructs [32] and used to inform a taxonomy of implementation outcomes [33]. Implementation monitoring consisted of ongoing engagement with service provider organizations through quarterly stakeholder meetings ($$n = 12$$). Meetings took place at least twice weekly between the health and wellbeing team responsible for coordinating SFL and the principal researcher from the period of January 2018 to January 2022. Approximately 50 meetings occurred with individual provider organizations and monthly report meetings took place with funding bodies, alongside quarterly financial reports.
The effectiveness evaluation involved following a cohort of SFL participants ($$n = 421$$) across $$n = 22$$ Shed settings for up to 12 months to assess impact of SFL on health and wellbeing outcomes. In terms of the assessment of implementation data, this data collection approach was also used to assess outcomes such as cost [35] while administrative data was gathered (Shed numbers and attendance rates) to inform penetration. Throughout this time the first author spent ~500 h among participants within the Sheds setting which facilitated direct observation of SFL in practice as well as observation of Shedders' experiences of SFL. Purposive sampling was also used to conduct focus groups ($$n = 8$$) with participating Sheds based on Shed size, location and level of attendance in SFL. This approach sought to gather a diverse representation of Shedders' experiences of SFL implementation. Informal short interviews ($$n = 16$$) were also conducted ad-hoc during Shed visits to further inform Shedders' experiences of implementation of SFL. This process was guided by CFIR constructs with a view to also informing the effectiveness of implementation strategies.
## Selection of implementation frameworks
The implementation and sustainment of an effective, evidence-based program in the real-world setting is complex and therefore multiple frameworks are increasingly being used and recommended in studies to address multiple facets of implementation (36–38). The use of theories, frameworks and models, which are often used interchangeably in implementation science can also cause further complexities for researchers [23, 36]. Nilsen [23] recommends selecting implementation frameworks based on three overarching aims: [1] describing or guiding the process of translating research into practice [2] understanding the determinants that influence implementation outcomes and [3] evaluating the implementation [23]. As the SFL research aimed to evaluate the implementation of the SFL initiative as well as understand the process and determinants of implementation, frameworks that suitably guided the process and evaluation of the research were selected. These frameworks consisted of a determinant framework to specify constructs that may influence the SFL process and predict implementation outcomes, a process framework to specify steps to execute for implementation phases and an evaluation framework to specify multiple levels of outcomes to assess [19].
The process framework applied to SFL implementation was the PRACTical planning for Implementation and Scale-up guide (PRACTIS) [24]. The PRACTIS was used in an iterative process to practically guide the implementation process and evaluation in collaboration with key stakeholders. This framework was selected as it incorporated the use of CBPR and is operational in real world contexts, considering the influence of the wider implementation climate [24]. In this study, it was used to promote successful implementation and scale-up of SFL. Sheds for Life implementation was guided by four key steps, and will provide structure to the presentation of research findings, namely; characterizing the parameters of the implementation setting; identifying and engaging key stakeholders; identifying implementation barriers and facilitators; and addressing potential barriers to implementation across individual, provider, organizational and systems levels.
The determinant framework used was The Consolidated Framework for Implementation Research (CFIR) [32]. This framework was selected to characterize and understand constructs across five domains (intervention characteristics, outer setting, inner setting, characteristics of the individuals involved, and the process of implementation) which interact in complex ways to influence implementation outcomes. The CFIR was used as a practical guide to systematically assess potential barriers and facilitators during SFL implementation as well as guide methods for data collection and analysis. The guide was used in the development of interview schedules as well as in data analysis via a deductive approach where key themes were mapped to CFIR constructs across the five CFIR domains (see Table 2).
The evaluation framework applied to SFL was the taxonomy for implementation outcomes [33]. This framework was chosen to inform outcomes pertaining to implementation i.e., acceptability, adoption, appropriateness, feasibility, fidelity, implementations costs, penetration and sustainability. These were assessed in the SFL evaluation using mixed methods to measure implementation effect (see Figure 1). This evaluation framework was selected as the constructs by Proctor et al. [ 33] have potential to capture participant and provider attitudes (acceptability), behaviors (penetration, adoption) as well as contextual factors (appropriateness, sustainability and implementation cost) [33]. Figure 1 depicts the process of SFL implementation and the application of stages of the PRACTIS with use of the CFIR and taxonomy for implementation outcomes.
**Figure 1:** *Sheds for Life implementation evaluation flowchart. CFIR, Consolidated Framework for Implementation Research; SFL, Sheds for Life; PRACTIS, Practical planning for Implementation and Scale-up guide; Sheds, Men's Sheds Shedders, Men's Shed members.*
Data pertaining to SFL participation (attendance records, self-reported attendance, numbers who participated vs. numbers eligible) were triangulated to assess penetration. Cost-effectiveness was determined by comparing the costs (direct and indirect) of SFL to its benefits which were captured as the impact on quality-adjusted life-years (QALYs) derived from the short form-6D algorithm [35]. Qualitative data were triangulated and analyzed using a framework-driven approach throughout implementation testing of SFL and refined using a constant comparison process applying the CFIR to identify barriers and facilitators. Focus groups and interviews were transcribed and, as per recommendations by the National Cancer Institute's White Paper on qualitative research in implementation science, a hybrid approach of thematic deductive and inductive analysis was used to identify barriers and facilitators and inform implementation strategies to address barriers and subsequent outcomes [37, 39]. Initial codes were identified and data were then discussed with stakeholders throughout implementation of SFL in line with CBPR approaches, in order to ensure accuracy and identify strategies to address barriers to effective implementation. Figure 2 captures a stakeholder map of those involved in SFL delivery.
**Figure 2:** *Sheds for Life (SFL) stakeholder map at systems, organization, provider and Shed level. IMSA, Irish Men's Sheds Association; SFL, Sheds for Life; HSE, Health Service Executive & Sláintecare (Funding SFL); Pobal (administration and management of Slaintecare SFL funding); IHF, Irish Heart Foundation (health check and CPR provider) ICS, Irish Cancer Society & MKF, Marie Keating Foundation (cancer awareness component); NOSP, National Office of Suicide Prevention (safeTALK component); DHF, Dental Health Foundation (oral health components) GIW, Get Ireland Walking & Siel Bleu (physical activity provider); HFME, Healthy Food Made Easy (HSE; healthy eating and cooking component); Age Action (digital literacy component); MHI, Mental Health Ireland (mental health component); Other (providers who may deliver new SFL content).*
## Implementation testing and scalability assessment of Sheds for life
A detailed description of the implementation plan is outlined in the SFL protocol [19]. In brief, the first implementation of the structured 10-week SFL implementation involved $$n = 22$$ Sheds and $$n = 421$$ Shedders across four counties in Ireland (two counties in March to May 2019 and two counties in September to December 2019) facilitated by $$n = 12$$ provider organizations and their subsequent regional deliverers (Figure 2 shows a conceptual map of SFL stakeholders). Participants ($$n = 421$$) were followed at baseline, 3, 6, and 12 months. These results are described in detail elsewhere and highlight both the efficacy of the SFL initiative in encouraging positive and sustained changes in health and wellbeing outcomes for Shedders [31], as well as supporting the case for scale-up [29]. Baseline characteristics of participants also highlight that SFL was effective in engaging a cohort of HTR men [12]. Implementation of SFL proceeded in the four counties outlined but due to the onset of COVID-19 Sheds remained closed for an extended period. Barriers and facilitators to further implementation within the changing implementation landscape were also monitored during this time. This process is described in detail below.
Insights into the determinants of implementation detailed below were then used to inform scalability assessment of SFL using the Intervention Scalability Assessment Tool (ISAT) [30]. The ISAT is designed to assist policy makers, practitioners and researchers to determine the scalability of discrete health interventions. The ISAT is scored by a series of readiness questions to assist in identifying strengths and weaknesses across the domains. Domains in part A provide background information on the public health problem, the context within which it is proposed that the intervention will be scaled up, and a description of the intervention. Domains in part B consider implementation and feasibility factors relating to all aspects, including fidelity and adaptations, reach and acceptability, delivery settings and agents, as well as implementation infrastructure and training. Each question is scored from 0–3, where the minimum score for each domain is 0 and the maximum score is 3. In order to derive a final score for the domain, the average score across the questions is taken (if there is more than one question).
## Results
Results presented describe the process of implementation, the identification of implementation determinants (barriers and facilitators) as guided by the CFIR, identification of subsequent strategies to address barriers and how these steps informed implementation outcomes. Qualitative data will be used to support findings. The CFIR refers to barriers and facilitators as implementation determinants, as these determinants often have dual capacity to act as either a barrier or facilitator [32]. Therefore, determinants in the context of this work mean contextual factors with potential to be either barrier or facilitator. The PRACTIS guide is used to structure presentation of results as per the four staged process of implementation [24]; Step 1 summarizes the process of characterizing the implementation setting of SFL; Step 2 summarizes the process of identifying and engaging key stakeholders; Step 3; summarizes the process of identifying implementation barriers and facilitators which include a detailed summary of those identified and; Step 4; summarizes the process of addressing (where possible), barriers to implementation with a detailed description of implementation strategies used to address same. Figure 1 provides a flowchart of the evaluation process which is described in detail below.
## Step 1: Characterization of the Sheds for life implementation setting
Early familiarization with characteristics of the real-world implementation context aids planning and accountability that may enhance implementation efforts [24]. Prior to the formal evaluation of SFL, members of the IMSA team consulted with Shedders at regional Shed “Cluster” meetings in 2017, which determined both an appetite for health and wellbeing in Sheds and signposted toward potential program content: The IMSA then began to identify potential partners that they deemed suitable to deliver various aspects of health and wellbeing in Sheds, some of which had previously expressed interest in working with Sheds under their individual remits. This allowed ad-hoc piloting of what would later become components of SFL.
Previously described scoping work [10] highlighted that a key requirement for service provider organizations to work with Sheds was that they understood the ethos and Shed environment. This led to the development of a “Guidance for Effective Engagement with Men's Sheds” (GEEMS) manual and workshop, which were designed to promote understanding of the Shed environment and ethos for provider organizations and which remain a key implementation strategy of SFL. This was augmented by ENGAGE training—national men's health training for service providers seeking to work more effectively with men [40]—which was delivered to service provider organizations seeking to participate in SFL delivery.
Following pilot testing of various SFL components, the IMSA expressed a desire to structure SFL into a suite of program offerings and the current research team then commenced the formal evaluation of SFL in collaboration with SFL stakeholders which began with characterizing the parameters of the implementation setting [24]. This commenced with an iterative consultation process with the IMSA and research team exploring intervention design, adoption, delivery, sustainability and potential scalability as well as important multi-level contextual characteristics [24]. Consideration was also given to evaluation design in terms of both effectiveness and implementation. This consultation process contributed to describing the Five P's for effective implementation as outlined by the PRACTIS guide [24]. Table 1 outlines the output from characterization of the implementation setting.
**Table 1**
| The Five P's | Definition | Description |
| --- | --- | --- |
| People | The type and number of people that the intervention will reach, and the individuals that will be involved/required for implementation and scale-up | Considering capacity of the IMSA, research team and prospective provider organizations, consultation determined that a feasible approach would be to deliver SFL across four counties on a phased basis (two counties per phase) with the aim of engaging upwards of n = 350 Shedders via a clustered approach of circa n = 15 Sheds. The selection of counties was based on seeking a diverse representation of Shedders and Sheds in terms of size and geographical location (urban/rural). There was an overarching focus on engaging HTR men through a whole Shed approach. Shed support volunteers acted as a conduit on the ground to relay important information about SFL to Shedders during program delivery in conjunction with IMSA staff and the principal researcher. Sheds for Life was delivered by allied provider organizations whose ethos and goals aligned with the goals of the IMSA and who were deemed to be able to effectively respond to the needs of Shedders. This involved organizations who had participated in the GEEMS training and understood and respected the ethos and environment of the Sheds. This process was overseen by the IMSA in collaboration with academic partners. |
| Place | The setting/organizations that will be involved/required for implementation and scale-up | Sheds for Life consisted of a targeted intervention with the aim of delivery directly in the Shed settings. As Sheds are highly variable in terms of size and resources, alternative venues such as local community centers were sourced for those elements of SFL that could not be delivered in the Shed. |
| Process | The intervention or implementation process that will occur in practice | Sheds for Life sought to use gender-specific approaches to engage HTR Shedders with SFL. Recruitment involved an expression of interest process whereby Shedders retained a degree of autonomy and control by self-selecting into SFL. The principal researcher and health and wellbeing manager of the IMSA visited prospective participating Sheds to discuss the process of the SFL program and evaluation. Sheds for Life consisted of a 10-week, gender-specific intervention that commenced with a health check, weekly physical activity, healthy eating and mental health workshops, as well as optional components (e.g., suicide prevention, digital literacy, CPR, cancer, oral health and diabetes awareness) that allowed Sheds to tailor SFL to suit their individual needs. |
| Provisions | The resources that will be necessary to achieve intervention implementation and scale-up | •IMSA staff supported SFL recruitment and oversaw implementation (administration etc.). •Service provider organization staff delivered components of SFL in participating Sheds. •Recruitment materials were used to provide clarity (SFL expression of interest forms for Sheds). •Training workshop and GEEMS manuals were provided for providers of SFL. •SFL Handbook and component resources (leaflets, booklets, signposting etc.) were provided for participants. •Attendance records were given to providers to track attendance and attendance certificates were provided to participants. •Text-based reminder services were used and program calendars were supplied to participating Sheds. •Researcher gathered data one-to-one with Shedders and standardized protocols were used to measure outcomes at baseline, 3, 6, and 12 months. •Standardized protocols were also used to gather costs of implementation for economic evaluation. •Funding was provided by the Health Service Executive section 39 funding. Funding was also provided through individual grants and budgets of provider organizations with a view to securing alternative funding streams. The Irish Research Council's employment-based postgraduate scholarship funded the principal researcher's employment within the IMSA. |
| Principles | The underlying principles of the intervention (e.g., individual behavior change) and implementation process (e.g., building capacity for implementation) that will be used to scale-up in practice | Intervention: Capitalizing on the safe, familiar environment and social support within Sheds, gender-specific implementation strategies were used to engage “HTR” men with health and wellbeing. Using a co-design process, self-efficacy was enhanced through normalizing conversations about health and wellbeing in the Shed environment. Targeted outcomes included subjective wellbeing, diet, physical activity, mental health, social capital and help seeking. Implementation: Building on existing structures within Sheds, strengths-based approaches were used to maximize Shedders' involvement in the design and subsequent adaptations of SFL as it evolves. There was also an explicit focus on strengthening existing partnerships and identifying new partners who could potentially respond to evolving needs of Shedders. Identifying new funding opportunities to support SFL implementation was also a key target. |
In summary, SFL was designed to build upon the inherent health promoting qualities of Sheds (delivery setting) while using participatory research methods to identify gender-specific strategies that would further enhance the reach of the program to HTR men (intervention population). The aim of the SFL design was to enhance health and wellbeing outcomes for Shedders while normalizing conversations about health for HTR men in Sheds through informal delivery and strength-based approaches (intervention characteristics). There was a strong emphasis in the recruitment phase on increasing the acceptability of SFL through trust and rapport building at Shedder level (intervention context). Evaluation methods were refined during this time to identify ways to monitor implementation for what was a complex multi-level intervention. This involved the previously outlined hybrid type 2 effectiveness-implementation design which also incorporated analysis of cost effectiveness. The implementation process also involved a partnership approach with all key stakeholders (Shedders, providers, IMSA, funders).
## Step 2: Identification and engagement of Sheds for life key stakeholders
The PRACTIS guide highlights the importance of participatory research to facilitate implementation and sustainability of complex community-based interventions [24]. From the outset of the formal evaluation of SFL there was strong emphasis placed on identifying those aspects of the partnership between the multiple stakeholders that impacted most on SFL implementation and that would facilitate scale-up of the program. The IMSA also recognized the need for this stakeholder engagement as it was a critical success factor to ensure effective implementation of SFL: The structured format of SFL was designed to engage key stakeholders from the outset. At a top-down systems and national men's health policy level [41, 42], the need for community-based men's health programs such as SFL was clearly mandated. These priorities also aligned with the National Health Service Executive's (HSE) priority programs. Thus, core components of SFL aligned with the key pillars of the Healthy Ireland Framework including healthy eating, physical activity and mental health [42]. This was a key facilitator of stakeholder engagement at systems (HSE) level and helped leverage funding to support core staff at the IMSA to oversee delivery of SFL: The SFL advisory group was consulted quarterly and brought considerable experience in men's health policy, practice and community development work to help guide and shape the evaluation and implementation of SFL. This further guided the actions of what would be structured as the SFL stakeholder group.
At the organizational level the first author was positioned within the IMSA for the duration of the research and worked closely with the health and wellbeing manager to promote effective implementation and co-production of SFL in line with evidence on men's health practice, while also ensuring that the implementation strategy aligned with existing practices and infrastructure.
Acknowledging how critical provider organizations (POs) were to the delivery of SFL, the IMSA spent time building relationships with multiple POs prior to the formal evaluation (see Figure 2). The implementation process focused on strengthening these partnerships through the formation of a structured stakeholder group. Provider organizations were consulted throughout the implementation process about implementation strategies, assessment of the implementation environment and they participated in the evaluation process to promote pragmatic and context-driven research. New providers were invited to join the SFL team in response to identified Shedder needs prior to implementation of SFL. In the absence of large-scale funding for SFL, priority was placed on identifying partners that understood the need for SFL. These providers were sought with a view to adopting a sustainable delivery model under real-world conditions where providers could undertake delivery as part of their routine work plans - as opposed to seeking short-term (and often unsustainable) grant funding to get SFL established. This meant that a prudent approach was needed in matching Sheds' needs with SFL offerings. The participatory approach with providers was therefore critical to sustained engagement: While there were no financial incentives, stakeholders had an active role in the development of evaluation tools (questionnaires) to encourage adoption where evaluation of each POs component of SFL was a key engagement strategy: Moreover, the priorities of POs aligned with those of the IMSA and SFL in reaching HTR men which is a noted challenge in community-based work, and thus SFL provided opportunities to connect and foster long-term buy-in and support.
Shedders were viewed as key stakeholders throughout the evaluation process of SFL as both hosts of SFL in the Sheds setting and intervention users. While SFL had a top-down policy directive, it mostly evolved as a bottom-up initiative to address a particular need within Sheds. Considerable time was spent in the Sheds as outlined in the methods to capture Shedders' experiences of SFL in practice as well as to co-design the structure and delivery of SFL. Sheds for Life was promoted as a program “For Shedders by Shedders” with Shedders having a crucial role in the identification of barriers and facilitators at Shed level. This engagement and co-design process were critical to acceptability and appropriateness of SFL implementation (these strategies will be further described in subsequent sections (see Table 3): Table 1 provides details on the structure of SFL with a further detailed breakdown available in the SFL protocol [19]. Findings from scoping work [10] in consultation with key stakeholders guided the decision to structure SFL as 10-week program. This format was viewed acceptable by POs and the IMSA as it was long enough in duration in terms of the practicalities of delivery and encouraging positive and sustained behavior change. Crucially, from Shedders' perspective, it also respected the fluid nature of Sheds in which a longer program might conflict with Shed routine. Moreover, this structure was pragmatic enough to consider whether SFL was feasible in the real-world, capricious Shed environment while prioritizing future sustainability within existing funding structures. This structure and format were also informed by what worked in other programs in Ireland with similar cohorts of men within community settings [43]. In terms of its design, the flexibility of SFL such as the optional components provided Shedders with an opportunity to tailor SFL to suit their needs while also instilling a sense of autonomy and control: In summary, SFL emerged from an invested process of engagement, consultation, relationship building and pilot testing. These efforts seeded partnership networks that understood the processes and recognized the value in engaging men with health. This was an important consideration at a time when Sheds had been earmarked as settings that facilitated access to HTR men and where expectations placed on Sheds to expand into formal healthcare delivery may have caused tensions within Sheds [10]. While it was recognized that the implementation evaluation would lead to refinement of SFL, meaning its structure could ultimately evolve, it was understood that this process of delivery and vested partnerships were the crux of its sustainability:
## Step 3: Identification of implementation determinants (barriers and facilitators)
The purpose of identifying contextual barriers and facilitators to SFL implementation was to enhance implementation effectiveness through integration of research findings into practice [24]. Barriers and facilitators were identified throughout SFL implementation via the multiple data collection techniques outlined at Shedder, Shed, PO, organization and systems level. The CFIR was used as a guide to group determinants at each level of implementation- some of which influenced all ecological levels. Table 2 describes the determinants to SFL implementation as guided by the CFIR with adaptations that were also context-specific. Figure 3 also conceptualizes the most prominent determinants in an ecological model of SFL implementation.
## Determinants at Shedder and Shed level
Table 2 provides a description of the implementation determinants at Shed and Shedder level. Table 3 outlines implementation outcomes, their influencing determinants as well as strategies to address barriers toward implementation at Shed and Shedder level. Alongside the scoping work [10] which highlighted the importance of respecting the Shed environment as a core determinant to Shedders' acceptability of SFL, SFL was built upon an evidence base of men's health research and practice work that employed gender-specific strategies to engage men with health while utilizing the Shed as a foundation for SFL (40, 41, 43–45). This helped to engage HTR Shedders in a familiar and safe way and to overcome barriers at the individual level such as previous adverse experiences with engaging in health and Shedders perceptions of socially acceptable ways that men should behave in relation to discussing and engaging with health issues:
**Table 3**
| Implementation outcome definition (33) | Measurement | Level (33) | Influencing determinant(s) | Strategies to address barriers to implementation and enhance outcomes |
| --- | --- | --- | --- | --- |
| Acceptability Acceptability is the perception among implementation stakeholders that a given treatment, service, practice, or innovation is agreeable, palatable, or satisfactory. | Stakeholder consultations & Interviews | Provider | Shared vision Relative advantage Compatibility & Complexity | •Allied partnership approach: SFL was delivered and designed in collaboration with POs who clearly perceived the advantage of implementing SFL through a shared vision, aligning with their organization in accessing a HTR group of men. SFL responded to the increasing calls by national policies to implement gender-specific strategies that engage HTR men with health which were applicable to PO's. •Stakeholder engagement: POs were continually engaged to promote shared decision making in the implementation of SFL to limit perceived complexity. |
| | Focus groups, Interviews & Ethnography | Shedder | Personal attributes Knowledge & beliefs about SFL Previous experience Trust Perceived complexity Relative advantage Identification with organization Ownership Compatibility & structural characteristics | •The intervention was designed and refined with underlying gender-specific approaches that enhanced the organic health promotion in Sheds. •Targeted intervention: delivered in a targeted way by bringing SFL to the Sheds and delivering the majority of its components directly in the Sheds natural environment or other local community setting, which were viewed as familiar, safe and non-clinical, environments for Shedders. This removed barriers toward participation and made participation convenient. •Expression of interest and Active Recruitment: Sheds were encouraged through shared decision making to opt into SFL participation—it was not foisted on Shedders. When Sheds expressed interest the researcher and health and wellbeing team in the IMSA visited each individual Shed and discussed the process of SFL in an informal way, reducing perceived complexity, building trust and actively recruiting individual Shedders and addressing their concerns. This strategy also aimed to enhance the relationship and sense of trust between the IMSA and Sheds. •Co design process: SFL was described to prospective participants as a program “for Shedders by Shedders”. Prospective participants were encouraged to see themselves as pioneers, actively shaping the program through their participation and paving the way for future delivery and scale-up. Reinforcing Shedder' sense of ownership was designed to build safety and trust, and to reassure participants that SFL was not being implemented to undermine the routine environment and ethos of the Sheds. Involving Shedders in the implementation process also facilitated access to local knowledge and resources for SFL implementation while building relationships enhanced the sense of social capital that positively influenced implementation. |
| Adoption Adoption is defined as the intention, initial decision, or action to try or employ an innovation or evidence-based practice. | Stakeholder consultations, interviews & observation | Provider | Shared vision Understanding of men's health Opinion leaders Stakeholder participation | •POs who understood the value of implementing SFL in Sheds and understood the need for gender-specific approaches were engaged in the stakeholder process. •Opinion leaders within the POs were valuable in building momentum to join the partnership network. •The Participatory Research Approach where all key stakeholders acted as decision makers in SFL design and implementation that is built upon evidence-based practice was a key facilitator in adoption at PO level. |
| | Consultation & Observation | Organization | External mandates & funding Understanding of Shedders Relative priority Leadership | •The implementing organization responded to both top down (policy and funding incentives) and bottom up (Shedder needs) calls to deliver health promotion in Sheds. •Sheds for Life was viewed a priority program in the organization. •Leadership from key implementers (health and wellbeing manager and researcher) who worked in partnership to strengthen implementation enhanced the perceived importance of SFL among other competing priorities. |
| | Administrative data, Focus groups, Interviews & Ethnography | Shed setting | Trust Social support Self-efficacy Leadership Shared decision making Autonomy Knowledge & beliefs about SFL | •Trust and relationship building through time spent in the Shed setting at recruitment phases was a key enabler of adoption within the Sheds. The co-design process facilitated reassurance among Shedders that SFL would remain respectful of the Shed environment and the autonomy of Shedders. •Shed support volunteers or champions played a key role in encouraging Sheds to try SFL. Designated contact points in each Shed act as a conduit between Shedders and program delivery. •Leaders within Sheds were also pivotal to adoption and engagement at Shed level and time was spent with identified leaders during Shed visits and national Shed volunteer coordinator events to ensure that key influencers understood the value of SFL for Sheds. In person visits by the recruitment team to Sheds were also a critical facilitator to adoption as it ensured that messages about SFL were disseminated to all Shedders (rather than one influencer who may not intend to adopt) and this encouraged shared decision making among Sheds. •SFL capitalized on the organic health promotion that occurs through the already existing social support between Shed members in Sheds. More reticent Shedders were encouraged to participate by Shedders with a higher sense of self-efficacy. •Use of “Hooks”: A free comprehensive health check at the beginning of SFL is a critical incentive to engage men in the SFL program alongside other life-skill components such as CPR. |
| Appropriateness Appropriateness is the perceived fit, relevance, or compatibility of the innovation or evidence based practice for a given practice setting, provider, or consumer; and/or perceived fit of the innovation to address a particular issue or problem. | Focus groups, Interviews & Ethnography, participatory research | Shedder & Shed setting | Compatibility Ownership Autonomy Perceived complexity Structural characteristics | •Male specific: An underlying principle of SFL was to deliver in the male-only environment of the Shed in the company of like-minded men which promotes a sense of safety and motivation through friendly competition. •SFL was co-designed as a tailored intervention with core components but allows autonomous decision making over adaptable or supplementary elements which the Sheds can “self-select” into. It is continually refined in collaboration with Shedders to respond to their needs. •Respecting the Shed environment: The co-design process and early testing of SFL determined characteristics of Sheds to be key determinants of implementation (see Table 2). •Timing: Shedders are also recommended to designate a specific day of the week to dedicate to SFL so that it does not encroach on the typical routine of the Shed. A readiness assessment also informs whether SFL is suitable for a Shed at that time in terms of competing priorities, resources or maturity (e.g., newer Sheds may see SFL as an opportunity to build relationships whereas Sheds heavily established in workshop based activities may view SFL as detracting from primary Shed aims). During assessment by implementers at recruitment phase, Shedders with few resources or members may use nearby community resources or join with another Shed to participate in SFL. As determined via co-design, SFL also aims to be implemented during times that are conducive with the Shed environment such as spring or autumn avoiding busier project periods for the Sheds such as Christmas or summer. |
| | | | | •Sheds for life was delivered free of charge to eliminate cost barriers for Shedders. •Autonomous Participation: Alongside the expression of interest process, individual Shedders are asked to participate in as much of SFL as possible while recognizing and respecting that other life commitments happen. The central goal of SFL is to enrich, not undermine the Sheds already health enhancing environment and so alongside ongoing collaboration with Shedders, participants of SFL are also guided not to overburden themselves by committing to too many SFL components. •Structure, Clarity & Supportive Resources: As perceived complexity was a noted determinant, participants receive supportive resources during SFL such as dedicated SFL and Healthy Food Made Easy handbooks as well as material on mental health and other various components. Participants are visited by the recruitment team to explain the process of SFL and also receive text reminders and prompts during SFL delivery along with program calendars and screening appointment cards. |
| | Stakeholder consultations, interviews & observation | Provider | Complexity Delivery style Relationship with Sheds Networks & communication Adaptability | •Both formal and informal meetings with stakeholders were used to limit complexity for POs and the IMSA coordinated and oversaw delivery of individual SFL components. •Credibility and capacity building: POs were seen as part of an allied partnership network bringing expertise from a variety of credible and informed sources thus enhancing perceived quality of SFL in Sheds. POs also participated in GEEMS and ENGAGE training for effectively working with men. •Adaptability: POs through stakeholder engagement were encouraged to tailor their components to suit both the cohort of men and the Shed environment. •Informality of Sheds: SFL was refined to be delivered in an informal, interactive and relaxed way with a conversational tone. Through iterative feedback POs of SFL were encouraged to spend time building rapport and trust with participants prior to delivery of SFL components. Informal delivery respects the ethos of the Sheds and facilitates comfort and active participation. •Strengths-Based Approach: SFL aims to be delivered using a strengths based approach where facilitators utilize the capacity, skills and knowledge of the men while demonstrating empathy and respect and using positive, non-stigmatizing or non-judgemental language and tone. |
| | | Organization | External mandates & funding Understanding of Shedders Engaging | •Men's health policy was an enabler to leverage support for SFL. Involving Shedders in the decision making process meant the organization was best positioned to understand and prioritize Sheds and Shedder needs. •The sustained engagement of appropriate stakeholders maintained momentum for implementation. |
| Implementation Cost Cost (incremental or implementation cost) is defined as the cost impact of an implementation effort. | | Provider | Available resources Complexity Adaptability Relative advantage | •While delivering SFL incurred additional time and monetary cost in terms of adaptations and delivery—POs that were able to incorporate SFL into part of their routine delivery could facilitate implementation with the advantage of accessing a group of HTR men for their own organization. |
| | | Organization | Funding Available resources Cosmopolitanism Engaging | •Sustainable funding would be a key determinant of SFL implementation and maintenance of partnerships. The capacity of the organization to network and engage key stakeholders who could support SFL delivery was a key enabler of supporting implementation costs. The evaluation of SFL was a key facilitator in highlighting the impact and cost-effectiveness (35) of SFL which gave the organization leverage to engage funders for substantial funding for SFL (e.g., Slaintecare). |
| Feasibility Feasibility is defined as the extent to which a new treatment, or an innovation, can be successfully used or carried out within a given agency or setting | Stakeholder consultations, interviews & observation | Provider | Compatibility Adaptability Shedders needs & resources Available resources Complexity Leadership Engaging | •The participatory research approach, pilot testing and partnership building were key facilitators in ensuring feasibility at provider level. •Feasibility has been demonstrated through measurement of impact on health and wellbeing outcomes of participants up to 12 months (31). |
| | Stakeholder consultations, interviews & observation | Organization | Available resources Understanding of Shedders Relative priority Leadership Cosmopolitanism Engaging | •The partnership approach to SFL alongside the leadership at organizational level and the refined research approaches were key facilitators to feasibility of SFL at organizational level. |
| | Focus groups, Interviews & Ethnography | Shed setting | Compatibility Structural characteristics Intervention source Leadership | •The co-design process where SFL was viewed as “internally” developed was critical to ensure that SFL was compatible and appropriate for Sheds. The initiative was also based upon evidence-based practice that engages men at community level, previous piloting of SFL informed the current strategy. •Leadership was also a key facilitator at Shed level to ensure successful implementation of SFL. •The implementation team endeavored to deliver SFL directly in the Shed setting, where resources were lacking in Sheds, kits including portable ovens and kitchen supplies were sourced to facilitate delivery of HFME within the Shed. |
| Fidelity Fidelity is defined as the degree to which an intervention was implemented as it was prescribed in the original protocol or as it was intended by the program developers | Stakeholder consultations, interviews & observation | Provider | Self-efficacy Knowledge and beliefs about the intervention Available Resources Adaptability Access to information & knowledge | •Fidelity was viewed as an important outcome for SFL as it moved across Shed settings. Fidelity was facilitated by consistent use of POs. Stakeholder engagement was used to ensure deliverers at ground level understood the underlying principle of SFL and GEEMS and ENGAGE training was made available. •Iterative feedback though the participatory research approach was used to address any identified issues with fidelity. It was recognized through the process evaluation that adaptations at local level were necessary for fidelity of SFL and there were facilitated through a consultation process. |
| Penetration Penetration is defined as the integration of a practice within a service setting and its subsystems. | Administrative data & observation | Shed setting | Knowledge & beliefs about the intervention Perceived complexity Leadership Ownership Compatibility | •Penetration of SFL at Shed level was encouraged through multiple implementation and gender-specific strategies outlined. Penetration in phase 1 delivery was captured by assessing the number of Shedders in the participating Sheds who eligible to attend vs. the number of Shedders who enrolled in SFL. •Assessment of the baseline profiles of Shedders also assessed whether SFL was reaching the HTR cohort within Sheds (46) |
| | Consultation and observation | Organization | External mandates & funding Knowledge and beliefs about the intervention Relative priority Leadership | •Penetration at the organizational level was facilitated by the evaluation of SFL which demonstrated the efficacy and cost-effectiveness of the approach. Sheds for Life was recognized by the organization as a priority program which is capable of leveraging support for Sheds at a systems level. Leadership of key implementers was an important enabler to champion SFL at organizational level. |
| Sustainability Sustainability is defined as the extent to which a newly implemented treatment is maintained or institutionalized within a service setting's ongoing, stable operations | Consultation and observation | Organization | External mandates & funding Available resources Relative priority Organizational incentives and rewards Leadership Networks and communication Cosmopolitanism Engaging | •Sustainability of SFL is facilitated by leadership at organizational level and the necessary resources needed to maintain momentum among stakeholders across implementation levels. The ability of the organization to retain key implementers as well as the support and funding at systems level are key determinants of sustainability. |
The recruitment phase of SFL was a critical facilitator to implementation as this period allowed trust and relationship building which was key to acceptability and adoption of SFL by Shedders: The time spent in the Sheds by the researcher and health and wellbeing team was also critical at this point in terms of identifying the local contextual factors and structural characteristics within Sheds that needed to be considered in molding SFL to suit individual Sheds. This also facilitated an understanding of the intricacies of the different operational systems of individual Sheds which determined that SFL should be seasonal (autumn & spring) and that SFL would not be appropriate to Sheds currently engaged in demanding project work. This was an important finding in terms of respecting the environments of Sheds and generating positive perceptions of SFL among Shedders rather than it being seen as an innovation foisted upon them. Moreover, this was a critical time to identify formal and informal opinion leaders in Sheds that would facilitate buy-in, to ensure whole Sheds received adequate communication about SFL and to dispel misconceptions about SFL. The relationships within Sheds were also key determinants to implementation of SFL. In particular the social support and informal peer mentoring among Shedders was key to supporting and engaging more HTR Shedders. Moreover, Shedders recognized the value of SFL in enriching the social support within Sheds by bringing Shedders together: The co-design process, targeted (delivered directly in Sheds) delivery and modular format of SFL instilled a sense ownership, autonomy and control over SFL within Sheds which was key to acceptability and adoption. Shedders recognized the value of SFL being implemented directly in Sheds which was key for engagement of HTR Shedders: Overall at Shed and Shedder level, the implementation of SFL demonstrated feasibility and impact in terms of positive and sustained health and wellbeing outcomes among participants as outlined in a SFL outcomes paper [31]. Moreover, SFL successfully transferred across Shed settings demonstrating its transferability and feasibility for scale-up in this regard. In terms of penetration the design of SFL demonstrated that it was capable of reaching the target cohort of HTR men within Sheds. Penetration has been highlighted elsewhere [46] but was assessed via administrative data and attendance records. This determined that of the $$n = 565$$ Shedders eligible to participate in SFL, $$n = 421$$ enrolled, a reach rate of $75\%$. The adoption of SFL at Shedder level was facilitated by the gender-specific strategies and co-design process where Shedders worked in partnership with the researcher and IMSA team to identify best practice at Shed level:
The informal delivery approach was a key facilitator to sustained engagement of Shedders. Overall the approach was appropriate to the Shed environments, which are highly variable informal settings, and implementation requires careful consideration of the multiple determinants outlined. It is also important to note that these variables do not remain fixed and evolve with Shedder needs. Therefore, in order for further implementation of SFL to remain impactful and appropriate to the Shed setting, the determinants and strategies outlined are critical to its sustained success most notably investment in relationships and partnerships with Shedders.
## Determinants at provider level
Partnerships are key to the successful implementation of SFL in terms of both delivery of SFL content but also in terms of championing the wider SFL movement and providing valuable insights to address facilitators of, and barriers to SFL within the stakeholder engagement process. Fostering partnerships with those who shared the vision and recognized the relative advantage in accessing a group of HTR men in their health promotion endeavors was key to acceptability and initial adoption of SFL at PO level. Moreover, the administrative assistance by the IMSA in terms of coordination and delivery of SFL limited complexity for POs thus enhancing acceptability. The stakeholder engagement instilled a sense of ownership among POs of SFL and, alongside the enjoyment and sense of reward offered from working in Sheds, adoption of SFL remained high for POs throughout implementation of SFL, which is demonstrated by their continued and sustained engagement: The stakeholder engagement, real-time feedback and discussion facilitated by the research team and the IMSA was a key strategy to overcoming barriers in relation to fidelity and adaptations needed to strengthen delivery such as ensuring an informal delivery style, suitable deliverers for Sheds and encouraging relationship building among POs and Shedders. Indeed, the informal nature of Sheds can present challenges to implementation (e.g., sporadic attendance) and was a key discussion point throughout stakeholder meetings. However, SFL was refined to be delivered in an informal, interactive and relaxed way with a conversational tone. Through iterative feedback POs of SFL were encouraged to spend time building rapport and trust with participants prior to delivery of SFL components. Informal delivery respects the ethos of the Sheds and facilitates comfort and active participation. Moreover, trust facilitates a sense of safety and a positive dynamic where participants can be open and honest. This was also an important facilitator in promoting understanding of Shedders and Sheds for all stakeholders alongside the capacity building focus of the GEEMS and ENGAGE training.
Feasibility and cost for the POs must be viewed in the context of continually shifting variables within the wider implementation climate. For instance, while adoption and POs' commitment to SFL remain high, these organizations are predominately NGOs meaning that sustained funding can be precarious. Therefore, commitment is largely contingent on determinants such as staff capacity and funding as well as key implementers and leaders within the individual POs who maintain support and momentum for SFL. This must also be considered in terms of the capacity of POs for scale-up of SFL. While POs may be committed to scaling up, funding structures are needed to support this: These determinants therefore require ongoing monitoring through continued engagement with the POs. Furthermore, in relation to appropriateness, although currently structured as a 10-week intervention with both core and optional components, SFL was designed as a flexible, dynamic program, subject to ongoing adaptation to meet evolving needs. This means that the SFL implementation strategy also needs to remain flexible to accommodate new POs over time in response to new or evolving requirements and preferences from Shedders. Thus, the structure and partnership network of SFL will inevitably evolve and grow over time. Whilst this presents certain challenges, it can also be seen as a strength of SFL, not least in terms of its potential to remain fresh and contemporary, but also its embedment in real-world conditions where determinants are understood and can be managed. It is heavily invested in a partnership network that recognizes the value of SFL and respects the ethos of Sheds.
## Determinants at organizational level
At the organizational level, there was general acceptability of the SFL initiative as the IMSA had an existing men's health remit which was supported by external funding of the National Health Service (HSE) and mandated by men's health policy [41]. While SFL took on significant momentum, this presented challenges for the organization in terms of the capacity of its small team of staff to manage the significant level of administration work required and the complexity of multiple stakeholders at Shedder, Shed, PO and systems level This also meant that there was pressure on the organization to fulfill other competing priorities and to secure funding to support general operations and work systems. This brought potential to conflict with the ethos of SFL and Sheds themselves and meant that leadership by SFL implementers was critical to ensure implementation effectiveness of SFL. Advocacy was required in terms of highlighting the importance of the foundational work required to implement SFL, ensuring that Shedders needs remained prioritized and the Shed environment respected. This also meant careful selection of POs (as opposed to seeking partnerships or funding from organizations that didn't have consistent ideals): Capacity was therefore a core determinant of SFL feasibility and scale-up both in terms of coordination and planning of SFL as well as maintaining important networks and communication at multiple levels, particularly at ground level with Sheds. The implementation of the first phase of SFL at organizational level was largely the responsibility of the health and wellbeing manager and researcher until further funding was secured for a health administration role: While it was important at this time for key implementers within the host organization to gain insights into the implementation of SFL across multiple levels, sustaining this momentum with limited capacity could ultimately be a barrier to the sustainability of SFL. For instance, the capacity demands required at ground level meant little attention could be awarded for advocacy at a systems level:
Moreover, the researcher's contribution to implementation efforts ended once the evaluation was complete. Alongside this, staff turnover is an inevitable feature of NGOs because of more limited prospects of promotion, job security and salary increments. This meant that there was limited capacity to retain staff who understood the intricacies of SFL, as well as a loss of leadership at organizational level which was also a consequence of contextual shifts due to the COVID-19 pandemic. Therefore, persistent barriers to sustainability and subsequent scale-up of SFL at organizational level are leadership and staff resources: Nevertheless, the evaluation of SFL which demonstrated that the program is cost effective [35], reaches HTR groups [46] and provides important benefits in terms of health and wellbeing for Shedders [31], meant that it was possible to leverage financial support for SFL at a systems level. This meant that the Irish Men's Sheds Association was awarded ongoing funding for delivery costs of SFL under “Sláintecare”- a framework for health service reform in Ireland which focuses on preventative strategies within the community setting [47] which was integrated into a sustainable funding model under the public health framework, Healthy Ireland [42]. While this funding increases the sustainability of SFL, in terms of scalability, the organization will likely need further funding support to increase capacity of staff to oversee delivery of SFL in multiple locations. While there are capacity issues that may impact scalability of SFL, the initiative has demonstrated it is an effective, transferable model that is scalable with the right leadership and support at organizational level: As with all NGOs, there was significant disruption to the organization during the COVID-19 pandemic [48]. Alongside staff turnover mentioned above, this impact was felt across multiple levels in terms of Shed closures and the direct impact on Shedders [49], funding disruptions and pressure to fulfill previous mandates agreed pre-pandemic. This ultimately rendered it unfeasible to deliver SFL throughout the pandemic due to multiple contextual factors beyond safety concerns, such as Shed readiness and capacity of POs to deliver. However, with the arrival of a sustainable funding stream and the evidence to support the efficacy of SFL with envisioned adaptations and leadership—the demand for SFL is likely to be high at Shedder level. While POs remain committed to SFL it will be important for the organization to continue its engagement of key stakeholders involved in SFL delivery to regain momentum and renew vigor that may have been lost during COVID-19 as well as establish new relationships required to respond to Shedders' needs post-pandemic, where the pandemic may have elevated wellbeing as a priority:
## Determinants at systems level
Operations at systems level have an important influence on the sustainability and scalability of SFL. Local communities are supportive of Sheds which is an important facilitator to implementation of SFL in terms of accessing resources at community level. While Sheds are viewed as important spaces at local level and recognized as an effective way of reaching men, there are issues with local services seeking to implement health initiatives in Sheds while operating in silo from the national organization. This could be a potential barrier to the wider acceptability of SFL if it becomes associated at Shed level with other initiatives that did not give the same level of due consideration to the need to adopt gendered approaches to program delivery, relationship building, and respecting the ethos of Sheds: The funding of NGOs is also an important systems level determinant of sustainability of SFL. While NGOs are important contributors to preventative services, funding is a prevailing issue which has a significant impact in their capacity to deliver as well as recruit and retain important staff members that are often overworked and under rewarded [50]. This was amplified during the COVID-19 pandemic and is an important variable in the implementation of SFL in terms of both the overseeing body and the POs capacity to deliver.
The strength and quality of evidence gathered was a key determinant of acceptability and adoption of SFL at a systems level. Policy, research and practice work also supported the need for men's health initiatives at community level [41] which were further incorporated into strategic frameworks at policy level [42]. Furthermore, as previously highlighted, the evidence from the SFL evaluation helped in securing funding under Slaintecare –[47]. This was fortuitous for SFL as the program fit the remit of Slaintecare reform and also the new “Healthy Communities” health service structures which focus on addressing health inequalities through a geographical (area-based) population profiling and segmentation approach [47]. This approach has the potential to place SFL on a more solid footing within the implementation system without betraying the essence or integrity of the program.
## Scale up of Sheds for life
Finally, when scoring the readiness of SFL scalability using the Intervention Scalability Assessment Tool (ISAT) [30], SFL is an initiative that merits scale-up, providing careful attention is paid to fidelity, workforce capacity and leadership (see Figure 3). Assessment of scalability has determined a horizontal scale-up approach as most suitable within the SFL context [30]. This is defined as the introduction of SFL across different Shed settings in a phased manner following the pilot through a stepwise expansion, learning lessons along the way to help refine further expansion [30]. The SFL assessment highlights several domains (particularly across part A) that are high scoring while other domains scored lower as outlined in Figure 4. For further insights into the scoring of SFL scalability see Supplementary File 1.
**Figure 4:** *Sheds for life scalability assessment using ISAT (30). ISAT, intervention scalability assessment tool.*
## Discussion
This research describes the process and determinants of SFL implementation both of which inform implementation effect. The careful selection of implementation frameworks was an important facilitator toward guiding this work which helped to limit further complexities of an already complex implementation climate [36, 51]. For the SFL evaluation, three frameworks were applied to guide the process, identify determinants and capture outcomes [24, 32, 33] which proved important for the research team when applying a new and innovative science to evaluation.
This work has highlighted the value of implementation research in monitoring the complexities of a multi-level co-developed intervention. While participatory research approaches are critical to the success of complex real-world innovations such as SFL they require a long-term process with commitment to sustainability [26]. This can present challenges when attempting to reconcile limited community-based resources with what is needed to capture the complexities of implementation within a system of continually shifting variables [52]. Indeed, a limitation to this research is the capacity of a small research team to monitor all levels of implementation and therefore it is possible that important determinants at either Shedder, PO, organizational or systems levels may have been overlooked. In the case of SFL however, it is important to remember that complexity is not just a property of the intervention but of the context or system into which it is placed, which includes multiple and dynamic interacting parts that generate nonlinear relationships [52]. While this research may not provide a definitive list, it plays an important role in capturing the process of implementation for scale-up of SFL as well as providing a blueprint for other community-based health initiatives in general, and men's health initiatives in particular, that may stand to benefit from this process. The messiness of implementation requires strong leadership and advocacy which was a core determinant of SFL's successful implementation. Implementation science requires strong partnerships between the implementers and researchers involved in the intervention [53]. For SFL this working partnership provided valuable momentum to implementation efforts. However, when the research becomes part of the implementation process, there is a risk that when this active ingredient is removed from further implementation that the effect may be impacted - a potential unintended consequence of implementation approaches [54]. For instance, in the case of SFL the researcher spent hundreds of hours within Sheds discussing SFL and engaging with and building relationships with Shedders. Indeed, Shedders may not have distinguished the evaluation from the intervention. The researcher was placed at the epicenter of a small, albeit national organization, which oversaw the implementation and assumed multiple roles within the implementation efforts, particularly during COVID-19. This means that the researcher becomes a core part of implementation efforts. In this case the researcher was not solely viewed as external consultant but rather a key advisor within the SFL team [55]. Understandably, this can raise questions about objectivity and impartiality which required the researcher to navigate ethical implications of an implementer/researcher role. Indeed this work mirrored many of the first-hand experiences captured by Cheetham, Wiseman [56] of how researchers can be subject to political pushes, pressures and sense of accountability. However, the assistance of the research team, SFL advisory, consultation with international academics and local researchers, combined with an open and transparent process of knowledge co-production with SFL stakeholders along with assertive boundary negotiations, were important in facilitating the embedded researcher to remain independent and impartial. Embedding a researcher has advantages too [55, 56], particularly in the case of public health and community-based organizations which may not have the resources to conduct rigorous evaluation, where funding is short-term and staff are heavily involved with hands-on activities. Indeed, Wolfenden et al. [ 55] argue that the challenges and costs of evaluating intervention trials, particularly those assessing the impact of implementation strategies, means that trials testing the impact of health interventions or implementation strategies represent 11 and $2\%$ of research output, respectively. This research therefore provides a valuable contribution to translational research and, in terms of the sustainability of SFL, the dissemination of the findings is proving valuable in leveraging further resources. Nevertheless, understanding the role of researchers at the intersection of academia and community-based practice is an important consideration for implementation science efforts.
While this research has highlighted multiple determinants that impacted and continue to impact SFL implementation, effective strategies outlined such as the gender-specific approaches at Shed level have increased the potential for, and demonstrated the utility of, the Shed setting as a suitable environment for SFL implementation. It has demonstrated that the model is transferrable despite the variability of Sheds when determinants such as the importance of relationship building, active recruitment and co-design processes are considered. An important question for SFL is ultimately what fidelity of the initiative looks like, particularly post pandemic. Indeed, while SFL is currently structured as 10-week intervention with multiple program offerings, this implementation science study highlights that while there should be fidelity to core components of SFL in terms of content to retain effect [31], the process of implementation and key implementation strategies are perhaps more critical to SFL fidelity than strict adherence to program content. In fact the inherent nature of Sheds means a constantly changing practice environment which is a key challenge for implementation research [51]. A critical juncture for SFL scalability, to potentially 450 Sheds in Ireland, will be its ability to maintain the co-designed nature of SFL, and the time spent investing in relationships with Sheds. In fact, without Shedders' acceptability and perceived appropriateness of SFL, there will ultimately be no implementation as the Sheds rightfully own SFL. The importance of these approaches is highlighted in the wider context of men's health research where the focus on addressing gender inequality in health programming has become more clearly conceptualized as a gender-transformative approach [57].
Considering the Milat et al. [ 29] guide to scaling up health interventions, the Sheds for Life evaluation has met the criteria of assessing effectiveness, reach, adoption, its ability to align with the strategic context, and acceptability and feasibility have been demonstrated. Moreover, scoring of the ISAT tool has showcased the initiative to meet the criteria for scale-up with careful attention required for fidelity, leadership and capacity [30]. Furthermore, Indig et al. [ 58], discuss how interventions found effective in a controlled setting should be scaled-up and an added strength to SFL is its implementation testing in what was certainly an uncontrolled and unpredictable environment. However, scale-up is a complex process and applying a multi-level perspective on transition to scale is required [59]. Moreover, while SFL has had a demonstrable impact on the health and wellbeing outcomes of Shedders, dilution of this impact should be avoided and often in the process of scaling-up health interventions, the effectiveness is reduced due to difficulties in maintaining the dose and fidelity of the original implementation [29].
This research has determined that currently SFL is an appropriate and acceptable model that has been widely adopted at Shed and PO level, while also establishing itself as a leading priority program for its host organization. The hybrid-effectiveness design of the SFL evaluation has demonstrated that SFL has emerged as the most appropriate model to reach the target cohort of HTR men [12]. Moreover, it has captured the implementation process and identified important facilitators and barriers to enhance implementation efforts. It is also efficacious [31] and cost-effective [35]. It is a scalable model that has also now established itself within the systems environment. The future of SFL and its potential to continue to engage Shedders and enhance their health and wellbeing outcomes is bright. Its scalability largely relies on leadership, financial and human resources and increased capacity for staff to oversee its delivery. Scaling up using a horizontal scale-up approach which introduces SFL to Sheds in a phased manner is feasible and yet requires continued refinement during further expansion [30]. This approach by its very definition means it is important that research efforts remain to monitor the scalability of SFL in order for the initiative to retain fidelity to its ethos and integrity as it begins to scale up nationally. Indeed, real-word implementation means that, even if it were possible to ensure that all implementation barriers to scalability were identified and subsequently addressed, additional threats to the implementation and scale-up process that are not anticipated will likely emerge [60]. Milat et al. [ 29], in a guide to scaling up interventions, place emphasis on subsequent evaluation and monitoring efforts during scale-up that focus on measuring effectiveness over time as well as other important implementation outcomes such as levels of penetration, adoption and acceptability. Nevertheless, our identification of implementation strategies (Table 3) provides tangible examples for researchers and practitioners that can act as a “how to” guide for successful implementation of community-based interventions. The key determinants highlighted in this work demonstrate that understanding the influence of the process is as important as the outcome. While effectively guiding the process can be complex, this can be made more manageable by using the right implementation approach. The implementation process must recognize the value of investing time in relationships and capacity building through working in partnership. This is the very essence of community-based work and can mean the “how” of implementation is as health enhancing as the “what”'.
## Conclusion
This research has captured the process and determinants of effective implementation of a community-based men's health promotion programme. Guided by implementation science, it has informed the scalability of SFL as well as identifying a “how to” of implementation strategies that can act as a blue print for other men's health settings and programs and health promotion more broadly. The evaluation of SFL highlights the importance of knowledge co-production in men's health work as well as in translational and implementation research efforts. While the evaluation of real-world multi-level interventions is complex, this work highlights the value and utility of embedded research which facilitates iterative decision making and allows adaptions to implementation subsequently promoting translation of research and knowledge production into practice in real-time. The evaluation demonstrates the importance of gender-specific approaches to men's health promotion where co-designed processes can help to positively redefine what health engagement means to HTR men. This work highlights that the process of implementation is as critical as the content that is delivered, meaning fidelity to the process is fundamental to retain effectiveness in scale-up efforts. This is the first evaluation to capture an implementation process of health promotion in Sheds. Moreover, this work makes a valuable contribution to research where there exists a dearth of research outputs capturing implementation strategies. It offers practitioners and researchers an example of the operationalization of implementation frameworks in practice as well as identifying strategies to engage key stakeholders, the most important of which are those who will ultimately use, and should rightfully own, the intervention. Therefore, real-world interventions should be designed with this in mind through strengths based, grassroots approaches.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by Waterford Institute of Technology Research Ethics Committee (REF: WIT2018REC0010). The patients/participants provided their written informed consent to participate in this study. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.
## Author contributions
AM, NM, and NR: conceptualization, methodology, funding acquisition, project administration, visualization, and writing—review and editing. AM: investigation and writing—original draft. NM and NR: supervision. All authors have read and agreed to the published version of the manuscript.
## Funding
AM was supported through an Irish Research Council Doctoral Award (Project ID EBPPG/$\frac{2018}{256}$). The funders were not involved in the design of the study, manuscript writing or collection of data, nor were the funders involved in data analysis or in manuscript writing.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frhs.2022.940031/full#supplementary-material
## References
1. Robertson S, Baker P. **Men and health promotion in the United Kingdom: 20 years further forward?**. *Health Educ J.* (2016) **76** 102-13. DOI: 10.1177/0017896916645558
2. Layte R, Banks J. **Socioeconomic differentials in mortality by cause of death in the Republic of Ireland, 1984-2008**. *Eur J Public Health.* (2016) **26** 451-8. DOI: 10.1093/eurpub/ckw038
3. McGraw J, White KM, Russell-Bennett R. **Masculinity and men's health service use across four social generations: findings from Australia's Ten to Men study**. *SSM Popul Health.* (2021) **15** 100838. DOI: 10.1016/j.ssmph.2021.100838
4. 4.Baker. From the Margins to the Mainstream: Advocating the Inclusion of Men's Health in Policy. A Scoping Study. London: Global Action on Men's Health (2020).. *From the Margins to the Mainstream: Advocating the Inclusion of Men's Health in Policy. A Scoping Study* (2020)
5. Lefkowich M, Richardson N, Robertson S. **“If we want to get men in, then we need to ask men what they want”: pathways to effective health programing for men**. *Am J Mens Health.* (2015) **11** 1512-24. DOI: 10.1177/1557988315617825
6. Baker P, White A, Morgan R. **Men's health: COVID-19 pandemic highlights need for overdue policy action**. *Lancet.* (2020) **395** 1886-8. DOI: 10.1016/S0140-6736(20)31303-9
7. Tharakan T, Khoo CC, Giwercman A, Jayasena CN, Sofikitis N, Salonia A. **Are sex disparities in COVID-19 a predictable outcome of failing men's health provision?**. *Nat Rev Urol.* (2022) **19** 47-63. DOI: 10.1038/s41585-021-00535-4
8. 8.WHO. The Health and Well-being of Men in the WHO European Region: Better Health Through a Gender Approach. Copenhagen: World Health Organisatio (2018).. *The Health and Well-being of Men in the WHO European Region: Better Health Through a Gender Approach* (2018)
9. 9.Pan American Health Organisation. Masculinities and Health in the Region of the Americas. Washington, DC (2019).. *Masculinities and Health in the Region of the Americas* (2019)
10. Bergin N, Richardson N. **'Sheds for Life': getting the balance right in delivering health promotion through Sheds in Ireland**. *Health Promot Int.* (2020) **36** 605-15. DOI: 10.1093/heapro/daaa082
11. Misan GMH, Oosterbroek C, Wilson NJ. **Informing health promotion in rural men's sheds by examination of participant health status, concerns, interests, knowledge and behaviours**. *Health Promot J Austr* (2017) **28** 207-216. DOI: 10.1071/HE16081
12. McGrath, A, Murphy, N, Egan, T, Ormond, G, Richardson, N. **Understanding shedders: Which socio-demographic, health and wellbeing characteristics best inform appropriate health promotion action in men's sheds and a ‘Shed for Life'?**. *Health Promot J Austral.* (2022) **3** 1-13. DOI: 10.1002/hpja.649
13. Wilson NJ, Cordier R. **narrative review of Men's Sheds literature: reducing social isolation and promoting men's health and well-being**. *Health Soc Care Commun.* (2013) **21** 451-63. DOI: 10.1111/hsc.12019
14. Golding B. *Shoulder to Shoulder: Broadening the Men's Shed Movement: Common Ground Research Networks* (2021)
15. Lefkowich M, Richardson N. **Men's health in alternative spaces: exploring men's sheds in Ireland**. *Health Promot Int.* (2016) **33** 525-35. DOI: 10.1093/heapro/daw091
16. Kelly D, Steiner A, Mason H, Teasdale S. **Men's Sheds: a conceptual exploration of the causal pathways for health and well-being**. *Health Soc Care Commun.* (2019) **27** 1147-57. DOI: 10.1111/hsc.12765
17. Kelly D, Teasdale S, Steiner A, Mason H. **Men's sheds in Scotland: the potential for improving the health of men**. *J Public Health Policy.* (2021) **42** 258-70. DOI: 10.1057/s41271-020-00268-5
18. Kelly D, Steiner A, Mason H, Teasdale S. **Men's sheds as an alternative healthcare route? A qualitative study of the impact of Men's sheds on user's health improvement behaviours**. *BMC Public Health.* (2021) **21** 553. DOI: 10.1186/s12889-021-10585-3
19. McGrath A, Murphy N, Richardson N. **Study protocol: evaluation of sheds for life (SFL): a community-based men's health initiative designed “for shedders by shedders” in Irish Men's sheds using a hybrid effectiveness-implementation design**. *BMC Public Health.* (2021) **21** 801. DOI: 10.1186/s12889-021-10823-8
20. Thorpe RJ, Halkitis PN. **Biopsychosocial determinants of the health of boys and men across the lifespan**. *Behav Med.* (2016) **42** 129-31. DOI: 10.1080/08964289.2016.1191231
21. Jull J, Giles A, Graham ID. **Community-based participatory research and integrated knowledge translation: advancing the co-creation of knowledge**. *Implement Sci.* (2017) **12** 150. DOI: 10.1186/s13012-017-0696-3
22. Bauman A, Nutbeam D. **Planning and evaluating population interventions to reduce noncommunicable disease risk – reconciling complexity and scientific rigour?**. *Public Health Res Pract* (2014) **25** e2511402. DOI: 10.17061/phrp2511402
23. Nilsen P. **Making sense of implementation theories, models and frameworks**. *Implement Sci.* (2015) **10** 53. DOI: 10.1186/s13012-015-0242-0
24. Koorts H, Eakin E, Estabrooks P, Timperio A, Salmon J, Bauman A. **Implementation and scale up of population physical activity interventions for clinical and community settings: the PRACTIS guide**. *Int J Behav Nutr Phys Act.* (2018) **15** 51. DOI: 10.1186/s12966-018-0678-0
25. Rapport F, Clay-Williams R, Churruca K, Shih P, Hogden A, Braithwaite J. **The struggle of translating science into action: Foundational concepts of implementation science**. *J Eval Clin Pract.* (2018) **24** 117-26. DOI: 10.1111/jep.12741
26. Israel BA, Eng E, Schulz AJ, Parker EA, Satcher D. *Methods in Community-Based Participatory Research for Health* (2005)
27. Curran GM, Bauer M, Mittman B, Pyne JM, Stetler C. **Effectiveness-implementation hybrid designs: combining elements of clinical effectiveness and implementation research to enhance public health impact**. *Med Care.* (2012) **50** 217-26. DOI: 10.1097/MLR.0b013e3182408812
28. Landes SJ, McBain SA, Curran GM. **An introduction to effectiveness-implementation hybrid designs**. *Psychiatry Res.* (2019) **280** 112513. DOI: 10.1016/j.psychres.2019.112513
29. Milat AJ, Newson R, King L, Rissel C, Wolfenden L, Bauman A. **A guide to scaling up population health interventions**. *Public Health Res Pract.* (2016) **26** e2611604. DOI: 10.17061/phrp2611604
30. Milat A, Lee K, Conte K, Grunseit A, Wolfenden L, van Nassau F. **Intervention Scalability Assessment Tool: a decision support tool for health policy makers and implementers**. *Health Res Policy Syst.* (2020) **18** 1. DOI: 10.1186/s12961-019-0494-2
31. McGrath A, Murphy N, Egan T, Richardson N. **Sheds for life: health and wellbeing outcomes of a tailored community-based health promotion initiative for men's sheds in Ireland**. *BMC Public Health* (2022) **22** 1590-1609. DOI: 10.1186/s12889-022-13964-6
32. Damschroder LJ, Aron DC, Keith RE, Kirsh SR, Alexander JA, Lowery JC. **Fostering implementation of health services research findings into practice: a consolidated framework for advancing implementation science**. *Implement Sci.* (2009) **4** 50. DOI: 10.1186/1748-5908-4-50
33. Proctor E, Silmere H, Raghavan R, Hovmand P, Aarons G, Bunger A. **Outcomes for implementation research: conceptual distinctions, measurement challenges, and research agenda**. *Adm Policy Ment Health.* (2011) **38** 65-76. DOI: 10.1007/s10488-010-0319-7
34. Hamilton AB, Brunner J, Cain C, Chuang E, Luger TM, Canelo I. **Engaging multilevel stakeholders in an implementation trial of evidence-based quality improvement in VA women's health primary care**. *Transl Behav Med.* (2017) **7** 478-85. DOI: 10.1007/s13142-017-0501-5
35. McGrath A, Murphy N, Egan T, Ormond G, Richardson N. **An economic evaluation of 'sheds for life: a community-based men's health initiative for men's sheds in Ireland**. *Int J Environ Res Public Health.* (2022) **19** 2204. DOI: 10.3390/ijerph19042204
36. Damschroder LJ. **Clarity out of chaos: use of theory in implementation research**. *Psychiatry Res.* (2020) **283** 112461. DOI: 10.1016/j.psychres.2019.06.036
37. Moullin JC, Dickson KS, Stadnick NA, Albers B, Nilsen P, Broder-Fingert S. **Ten recommendations for using implementation frameworks in research and practice**. *Implement Sci Commun.* (2020) **1** 42. DOI: 10.1186/s43058-020-00023-7
38. Nilsen P, Albers B, Shlonsky A, Mildon R. **Making sense of implementation theories, models, and frameworks**. (2020) 53-79. PMID: 25895742
39. Hamilton AB, Finley EP. **Qualitative methods in implementation research: an introduction**. *Psychiatry Res.* (2019) **280** 112516. DOI: 10.1016/j.psychres.2019.112516
40. Osborne A, Carroll P, Richardson N, Doheny M, Brennan L, Lambe B. **From training to practice: the impact of ENGAGE, Ireland's national men's health training programme**. *Health Promot Int.* (2016) **33** 458-67. DOI: 10.1093/heapro/daw100
41. 41.Department of Health and Children. National Men's Health Policy: Working with Men in Ireland to Achieve Opitmun Health & Wellbeing. Dublin (2008).. *National Men's Health Policy: Working with Men in Ireland to Achieve Opitmun Health & Wellbeing* (2008)
42. 42.Health Service Executive. Healthy Ireland- Men 2017-2021: Working With Men in Irelannd to Achieve Optimum Health and Wellbeing. Dublin (2017).. *Healthy Ireland- Men 2017-2021: Working With Men in Irelannd to Achieve Optimum Health and Wellbeing* (2017)
43. Kelly L, Harrison M, Richardson N, Carroll P, Robertson S, Keohane A. **Reaching beyond the 'worried well': pre-adoption characteristics of participants in 'Men on the Move', a community-based physical activity programme**. *J Public Health.* (2019) **41** e192-202. DOI: 10.1093/pubmed/fdy134
44. Wyke S, Hunt K, Gray CM, Fenwick E, Bunn C, Donnan PT. **Football Fans in Training (FFIT): a randomised controlled trial of a gender-sensitised weight loss and healthy living programme for men - end of study report**. *Public Health Res Southampton.* (2015) 3. DOI: 10.3310/phr03020
45. Sharp P, Bottorff JL, Hunt K, Oliffe JL, Johnson ST, Dudley L. **Men's perspectives of a gender-sensitized health promotion program targeting healthy eating, active living, and social connectedness**. *Am J Mens Health.* (2018) **12** 2157-66. DOI: 10.1177/1557988318799159
46. McGrath A, Egan T, Ormond G, Richardson N. **Understanding Shedders: how can need and socioemographic charactertistics inform appropriate health promotion action and a 'Shed for Life'?**. *Health Promot J Aust* (2022) **3** 1-13. DOI: 10.20944/preprints202203.0128.v1
47. **Slaintecare. Right Care**. *Right Place. Right Time. Sláintecare Implementation Strategy & Action Plan 2021-2023* (2021)
48. Tierney A, Boodoosingh R. **Challenges to NGOs' ability to bid for funding due to the repatriation of volunteers: the case of Samoa**. *World Dev* (2020) **136** 105113. DOI: 10.1016/j.worlddev.2020.105113
49. McGrath A, Murphy N, Richardson N. **The impact of the COVID-19 pandemic on the wellbeing of Irish Men's Shed members**. *Health Promot Int.* (2020) **36** 1007-19. DOI: 10.1093/heapro/daaa113
50. Asogwa IE, Varua ME, Datt R, Humphreys P. **The impact of COVID-19 on the operations and management of NGOs: resilience and recommendations**. *Int J Org Anal* (2022). DOI: 10.1108/IJOA-12-2021-3090
51. Nilsen P, Bernhardsson S. **Context matters in implementation science: a scoping review of determinant frameworks that describe contextual determinants for implementation outcomes**. *BMC Health Serv Res.* (2019) **19** 189. DOI: 10.1186/s12913-019-4015-3
52. Hawe P. **Lessons from complex interventions to improve health**. *Annu Rev Public Health.* (2015) **36** 307-23. DOI: 10.1146/annurev-publhealth-031912-114421
53. Leeman J, Birken SA, Powell BJ, Rohweder C, Shea CM. **Beyond “implementation strategies”: classifying the full range of strategies used in implementation science and practice**. *Implement Sci.* (2017) **12** 125. DOI: 10.1186/s13012-017-0657-x
54. Gruß I, Bunce A, Davis J, Gold R. **Unintended consequences: a qualitative study exploring the impact of collecting implementation process data with phone interviews on implementation activities**. *Implement Sci Commun.* (2020) **1** 101. DOI: 10.1186/s43058-020-00093-7
55. Wolfenden L, Yoong SL, Williams CM, Grimshaw J, Durrheim DN, Gillham K. **Embedding researchers in health service organizations improves research translation and health service performance: the Australian Hunter New England Population Health example**. *J Clin Epidemiol.* (2017) **85** 3-11. DOI: 10.1016/j.jclinepi.2017.03.007
56. Cheetham M, Wiseman A, Khazaeli B, Gibson E, Gray P, Van der Graaf P. **Embedded research: a promising way to create evidence-informed impact in public health?**. *J Public Health* (2018) **40** i64-70. DOI: 10.1093/pubmed/fdx125
57. Ruane-McAteer E, Amin A, Hanratty J, Lynn F. **Corbijn van Willenswaard K, Reid E, et al. Interventions addressing men, masculinities and gender equality in sexual and reproductive health and rights: an evidence and gap map and systematic review of reviews**. *BMJ Global Health.* (2019) **4** e001634. DOI: 10.1136/bmjgh-2019-001634
58. Indig D, Lee K, Grunseit A, Milat A, Bauman A. **Pathways for scaling up public health interventions**. *BMC Public Health.* (2017) **18** 68. DOI: 10.1186/s12889-017-4572-5
59. Bulthuis SE, Kok MC, Raven J, Dieleman MA. **Factors influencing the scale-up of public health interventions in low- and middle-income countries: a qualitative systematic literature review**. *Health Policy Plan.* (2020) **35** 219-34. DOI: 10.1093/heapol/czz140
60. McAlearney AS, Walker DM, Livaudais-Toman J, Parides M, Bickell NA. **Challenges of implementation and implementation research: Learning from an intervention study designed to improve tumor registry reporting**. *SAGE Open Med* (2016) **4** 2050312116666215. DOI: 10.1177/2050312116666215
|
---
title: Identification of ferroptosis related biomarkers and immune infiltration in
Parkinson’s disease by integrated bioinformatic analysis
authors:
- Na Xing
- Ziye Dong
- Qiaoli Wu
- Yufeng Zhang
- Pengcheng Kan
- Yuan Han
- Xiuli Cheng
- Yaru Wang
- Biao Zhang
journal: BMC Medical Genomics
year: 2023
pmcid: PMC10012699
doi: 10.1186/s12920-023-01481-3
license: CC BY 4.0
---
# Identification of ferroptosis related biomarkers and immune infiltration in Parkinson’s disease by integrated bioinformatic analysis
## Abstract
### Background
Increasing evidence has indicated that ferroptosis engages in the progression of Parkinson’s disease (PD). This study aimed to explore the role of ferroptosis-related genes (FRGs), immune infiltration and immune checkpoint genes (ICGs) in the pathogenesis and development of PD.
### Methods
The microarray data of PD patients and healthy controls (HC) from the Gene Expression Omnibus (GEO) database was downloaded. *Weighted* gene co-expression network analysis (WGCNA) was processed to identify the significant modules related to PD in the GSE18838 dataset. Machine learning algorithms were used to screen the candidate biomarkers based on the intersect between WGCNA, FRGs and differentially expressed genes. Enrichment analysis of GSVA, GSEA, GO, KEGG, and immune infiltration, group comparison of ICGs were also performed. Next, candidate biomarkers were validated in clinical samples by ELISA and receiver operating characteristic curve (ROC) was used to assess diagnose ability.
### Results
In this study, FRGs had correlations with ICGs, immune infiltration. Then, plasma levels of LPIN1 in PD was significantly lower than that in healthy controls, while the expression of TNFAIP3 was higher in PD in comparison with HC. ROC curves showed that the area under curve (AUC) of the LPIN1 and TNFAIP3 combination was 0.833 ($95\%$ CI: 0.750–0.916). Moreover, each biomarker alone could discriminate the PD from HC (LPIN1: AUC = 0.754, $95\%$ CI: 0.659–0.849; TNFAIP3: AUC = 0.754, $95\%$ CI: 0.660–0.849). For detection of early PD from HC, the model of combination maintained diagnostic accuracy with an AUC of 0.831 ($95\%$ CI: 0.734–0.927), LPIN1 also performed well in distinguishing the early PD from HC (AUC = 0.817, $95\%$ CI: 0.717–0.917). However, the diagnostic efficacy was relatively poor in distinguishing the early from middle-advanced PD patients.
### Conclusion
The combination model composed of LPIN1 and TNFAIP3, and each biomarker may serve as an efficient tool for distinguishing PD from HC.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12920-023-01481-3.
## Background
Parkinson’s disease (PD) is a common neurodegenerative disorder which involves in classic motor features of Parkinsonism including tremor, akinesia and bradykinesia, as well as nonmotor symptoms such as constipation, sleep disturbance and cognitive impairment and so on [1, 2]. The typical pathologic characteristics of PD are pathologic accumulation of cytoplasmic misfolded α-synuclein, in form of Lewy bodies and progressive loss of dopaminergic neurons in the substantia nigra pars compacta (SNpc). The incidence increases with age while relatively little is known of the exact neurodegenerative pathogenesis, which relates to multiple factors including genetics, oxidative stress, immune activation, mitochondrial dysfunction or lipid dyshomeostasis [3].
Ferroptosis, an iron-dependent non-apoptotic regulated programmed cell death, was firstly proposed in 2012 [4], which is mainly driven by iron dyshomeostasis and lipid peroxidation, leading to oxidative stress in cells and affecting metabolic processes of protein, nucleic acid, carbohydrates and lipids, ultimately leads to cell death [4, 5]. However, ferroptosis distinguishes from apoptosis, necrosis, autophagy and other forms of cell death in morphologically, biochemically and genetically [4]. Previous studies on ferroptosis mainly focused on cancer, and iron metabolism has become a hot spot in tumorigenesis, progression and treatment prognosis. To data, ferroptosis-related genes (FRGs) have been recognized as diagnostic biomarkers for multiple cancers [6, 7]. A rat organotypic hippocampal slice culture model showed that erastin induced ferroptosis can promote neuronal death by creating a void in the antioxidant defenses of cell, but Fer-1 prevents glutamate-induced neurotoxicity [4]. Iron is an oxidant and excess free iron can induce oxidative stress, inflammation and excitotoxicity, causing cellular damage and neurodegeneration [8, 9]. The dyshomeostasis and intracellular retention of iron are associated with senescence of mutiple types of cells, including neurons, which accelerates aging by inducing DNA damage and blocking genomic repair systems [10].
A recent discovery that α-synuclein oligomers can bind to the plasma membrane and drive cell ferroptosis via altered membrane conductance, abnormal calcium influx and lipid peroxides production, which provides the direct evidence that ferroptosis is referred as an essential pathogenic mechanism in synucleinopathies [11]. Ever increasing evidence linking α-synuclein to the metabolism of iron and lipids, suggesting a possible role of α-synuclein in ferroptosis [11]. Previous research has found that selective iron deposition pattern in substantia nigra is greatly influenced by the age of PD onset [12]. Activated glia promote dysregulation of iron homeostasis, thereby aggravating microglial activation, which plays a pivotal role in ferroptosis and subsequent neurodegeneration [13]. Characteristics of ferroptosis, such as iron accumulation, glutathione (GSH) depletion, lipid peroxidation and elevated reactive oxygen species (ROS), may be observed in PD patients [14]. Moreover, ferric ammonium citrate (FAC)-induced ferroptosis in dopaminergic cells is related to the phosphorylation of p53 signaling pathway not MAPK pathway [15]. However, the controversial results in erastin-treated Lund human mesencephalic cells indicate that whether erastin-induced ferroptosis is RAS-dependent needs further investigation [16, 17]. Conservative iron chelation modality (avoiding changes in systemic iron levels) established in mammalian models and clinical trials that offers a new therapeutic strategy based on iron scavenging and redeployment for neuroprotection [18].
In the present study, we investigated the biological pathways of pathophysiology from the perspective of ferroptosis in PD based on bioinformatics analysis and identified gene co-expression modules by WGCNA, further examine the relationship of FRGs with immune infiltration and immune checkpoint genes (ICGs). Moreover, the expression profiles of candidate genes were detected in clinical blood samples. The possible role and function of core genes in regulating ferroptosis and immune infiltration in PD were also explored.
## Methods
The work flow of this study is shown in Fig. 1.
Fig. 1 The work flow of this study
## Data acquisition and preprocessing
We applied “Parkinson’s disease”, “human beings”, “peripheral blood”, “expression profiling by array” as key words and ensured that each group has more than 10 subjects, the gene expression matrix of GSE18838 dataset [19] was obtained from the NCBI Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) (Accessed: 1 May 2022). The GSE18838 dataset included 17 PD and 11 healthy control (HC) whole blood samples, which was performed on GPL5175 platform ([HuEx-1_0-st] Affymetrix Human Exon 1.0 ST Array [transcript (gene) version]). The clinical characteristics of participates in GSE18838 are detailed in (Additional file 1: Table S1).
FerrDb (http://www.zhounan.org/ferrdb/) (Accessed: 2 May 2022) collected 259 ferroptosis-related genes (FRGs) including driver, suppressor and marker [20]. The confidence level of recorded genes involved in ferroptosis was assigned to 4 degrees including validated, screened, predicted and deduced.
## WGCNA analysis and intersect between DEGs and interesting module
In this study, we utilized a gene expression profile of GSE18838 to construct a weighted gene co-expression network between PD and HC using the “WGCNA” package in R software [21] and analyzed the relationships between gene modules and clinical phenotype of PD. Briefly, cluster analysis was used to explore whether there was outlier samples in the GSE18838 dataset to ensure the accuracy of further exploration. According to the scale-free topology criterion, we used the function pickSoftThreshold to select soft powers β = 12 and the soft thresholding parameter showed strong relations between genes while penalized the weak correlation. Then, the adjacency matrix was transformed into a topological overlap matrix (TOM) to measure the network connectivity of genes as well as the corresponding dissimilarity (1-TOM). A hierarchical clustering tree diagram of the 1-TOM matrix was constructed to classify genes showing similar expression profiles with gene co-expression modules. Then dynamic tree cut method was performed to separate different modules of all genes and merged the similar models using MEDissThres = 0.25. The different branches represented a different module. Subsequently, module-trait relationships were estimated via pearson analysis and the module with high correlation coefficients was considered as interesting module. *The* genes in the module were selected for following research.
Differentially expressed genes (DEGs) between PD and HC were identified utilizing the “limma” package in R software based on the following threshold: p-value < 0.05 and |log2FC|>0.5 in the GSE18838 dataset. The p-value was adjusted by Benjamini–Hochberg method to control the false discovery rate (FDR). The DEGs were visualized as volcano plot by using " ggplot2” package in R software.
Then, the intersect between DEGs, FRGs and co-expression genes that were extracted from interesting module was visualized as Venn diagram.
## Enrichment analysis of GSVA and GSEA
Gene set variation analysis (GSVA) was performed on the expression profile of GSE18838 using “GSVA” package in R software and the reference gene sets were hallmark gene sets, GO-BP, GO-CC, GO-MF, KEGG and C7: immunologic signatures, which were downloaded from the MSigDB database (https://www.gsea-msigdb.org/gsea/msigdb) [22] (Accessed: 2 May 2022). Gene Set Enrichment Analysis (GSEA) was operated using “GSEA” R package to investigate relate pathways of the candidate diagnostic genes and the reference gene set were KEGG. The number of random sample permutations was set at 1000, $p \leq 0.05$ was considered as significant enrichment.
## Machine learning algorithm for candidate genes
After identifying DEGs, we performed three machine learning algorithms as least absolute shrinkage and selection operator (LASSO) logistic regression, random forest (RF) and support vector machine-recursive feature elimination (SVM-RFE) to screen candidate genes for PD using “glmnet”, “randomforest”, and “e1071” package in R software, respectively. Then, we combined the genes from LASSO, RF and SVM-RFE algorithms for further analysis. The expression of the candidate gene was firstly validated in GSE18838 dataset and a two-sided $p \leq 0.05$ was considered statistically significant. Ultimately, the area under the receiver operating characteristic (ROC) curve analysis (AUC) was calculated to evaluate the accuracy of selected genes for diagnosing PD patients. The transcription factor (TF)-miRNA coregulatory network was constructed on NetworkAnalyst (https://www.networkanalyst.ca) (Accessed: 2 June 2022).
## GO and KEGG analysis
To explore the potential molecular mechanism of key genes associated with PD, Gene Ontology (GO) including biological process (BP), cellular component (CC) and molecular function (MF), and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses were operated using “clusterProfiler” R package [23]. The metascape database (http://metascape.org/) (Accessed: 10 June 2022) is an online database used for gene annotation, functional enrichment, interactome and membership analysis, used for KEGG and Reactome pathway analysis in the present study. p value < 0.05 as the screening threshold.
GO analysis was performed to illustrate the functional annotations of 179 WGCNA-DEGs. The result of cell composition for GO analysis was shown in Fig. 7A. The most enriched GO terms in the biological process category were mitochondrial respiratory complex I assembly, positive regulation of autophagy, response to reactive oxygen species and T cell activation/differentiation, and in the molecular function category were NADH dehydrogenase (ubiquinone) activity, MHC protein binding, immune receptor activity, and ATP metabolism process, Ras protein signal transduction, response to reactive oxygen species and positive regulation of I-kappaB kinase/NF-kappaB signaling and so on (Fig. 7B, C). KEGG and *Reactome analysis* was conducted to investigate the related signaling pathways. Among the Reactome pathways, macroautophagy, MHC class II antigen presentation, metabolism of lipids, toll-like receptor cascades and cellular responses to stress were involved in PD (Fig. 7D). In addition, KEGG pathway analysis also revealed that lysosome, FoxO signaling pathway, diabetic cardiomyopathy and PD-L1 expression and PD-1 checkpoint pathway in cancer may related to PD (Fig. 7E).
Fig. 7 GO and KEGG pathway enrichment results of WGCNA-DEGs. A, B, C *The analysis* of GO_CC, GO_MF and GO_BP. D, E The Reactome and KEGG pathway For 15 ferroptosis-WGCNA genes, the significantly enriched GO terms indicated that cellular response to TOR signaling, signaling transduction by p53 class mediator, selective autophagy, response to active oxygen species or metal ion or oxidative stress, fatty acid metabolic process and neuron death were associated with ferroptosis and PD (Table 2). The KEGG results suggest that mTOR signaling pathway, cellular senescence, neutrophil extracellular trap formation, pathways of neurodegeneration-multiple diseases, NF-kappa B signaling pathway and so on, which may play an important role in PD (Table 3).
Table 2The enriched terms of Gene ontology for ferroptosis-WGCNA genesTermIDDescriptionp.adjustGene IDBPGO:0007568aging7.92E-04MTOR/ATG7/MAPK8/ATM/MAPK3BPGO:0000422autophagy of mitochondrion2.35E-03HIF1A/ATG7/ATG13BPGO:0008366axon ensheathment3.99E-02MTOR/LPIN1BPGO:0007569cell aging6.02E-03MTOR/MAPK8/ATMBPGO:0036473cell death in response to oxidative stress2.63E-02HIF1A/ATG7BPGO:0071347cellular response to Interleukin-13.18E-02HIF1A/MAPK3BPGO:0090398cellular senescence2.60E-02MAPK8/ATMBPGO:0006631fatty acid metabolic process3.06E-02MTOR/LPIN1/MAPK3BPGO:0007612learning4.27E-02MTOR/HIF1ABPGO:0002260lymphocyte homeostasis1.69E-02HIF1A/TNFAIP3BPGO:0016236macroautophagy2.90E-05MTOR/HIF1A/ATG7/ATG13/MAPK8/MAPK3BPGO:0070997neuron death2.74E-02MTOR/HIF1A/ATG7BPGO:0034250positive regulation of cellular amide metabolic process4.97E-02MTOR/MAPK3BPGO:0010506regulation of autophagy3.60E-05MTOR/HIF1A/ATG7/MAPK8/ATM/MAPK3BPGO:0001959regulation of cytokine-mediated signaling pathway4.58E-02HIF1A/TNFAIP3BPGO:0035303regulation of dephosphorylation3.81E-02MTOR/LPIN1BPGO:0051090regulation of DNA-binding transcription factor activity3.85E-02MAPK8/TNFAIP3/MAPK3BPGO:0010821regulation of mitochondrion organization4.27E-02HIF1A/MAPK8BPGO:0031644regulation of nervous system process4.27E-02MTOR/LPIN1BPGO:1,903,203regulation of oxidative stress-induced neuron death7.98E-03HIF1A/ATG7BPGO:0031396regulation of protein ubiquitination1.18E-02MTOR/TNFAIP3/HERPUD1BPGO:0010038response to metal ion2.86E-02HIF1A/MAPK8/MAPK3BPGO:0006979response to oxidative stress2.15E-04HIF1A/ATG7/MAPK8/ABCC1/TNFAIP3/MAPK3BPGO:0000302response to reactive oxygen species2.11E-03HIF1A/MAPK8/TNFAIP3/MAPK3BPGO:0061912selective autophagy1.95E-02ATG13/MAPK3BPGO:0072331signal transduction by p53 class mediator9.24E-03MTOR/CD44/ATMBPGO:0031929TOR signaling5.64E-03MTOR/HIF1A/ATMCCGO:0000407phagophore assembly site2.29E-02ATG7/ATG13MFGO:0004707MAP kinase activity5.86E-03MAPK8/MAPK3MFGO:0106310protein serine kinase activity7.56E-03MTOR/MAPK8/ATM/MAPK3MFGO:0004674protein serine/threonine kinase activity7.56E-03MTOR/MAPK8/ATM/MAPK3 Table 3List of top enriched KEGG pathways of ferroptosis-WGCNA genesIDDescriptionGeneRatioBgRatiop.adjustGene IDhsa04140Autophagy - animal$\frac{6}{14141}$/81491.05E-$\frac{052475}{3091}$/10,$\frac{533}{9776}$/$\frac{5599}{5595}$hsa04930Type II diabetes mellitus$\frac{3}{1446}$/81491.58E-$\frac{032475}{5599}$/5595hsa04012ErbB signaling pathway$\frac{3}{1485}$/81494.40E-$\frac{032475}{5599}$/5595hsa05235PD-L1 expression and PD-1 checkpoint pathway in cancer$\frac{3}{1489}$/81494.54E-$\frac{032475}{3091}$/5595hsa04657IL-17 signaling pathway$\frac{3}{1494}$/81495.00E-$\frac{035599}{7128}$/5595hsa04066HIF-1 signaling pathway$\frac{3}{14109}$/81496.85E-$\frac{032475}{3091}$/5595hsa04668TNF signaling pathway$\frac{3}{14112}$/81497.02E-$\frac{035599}{7128}$/5595hsa04071Sphingolipid signaling pathway$\frac{3}{14119}$/81497.58E-$\frac{035599}{4363}$/5595hsa04068FoxO signaling pathway$\frac{3}{14131}$/81499.14E-$\frac{035599}{472}$/5595hsa04210Apoptosis$\frac{3}{14136}$/81499.57E-$\frac{035599}{472}$/5595hsa04910Insulin signaling pathway$\frac{3}{14137}$/81499.57E-$\frac{032475}{5599}$/5595hsa04150mTOR signaling pathway$\frac{3}{14156}$/81491.24E-$\frac{022475}{23}$,$\frac{175}{5595}$hsa04218Cellular senescence$\frac{3}{14156}$/81491.24E-$\frac{022475}{472}$/5595hsa05010Alzheimer disease$\frac{4}{14384}$/81491.85E-$\frac{022475}{9776}$/$\frac{5599}{5595}$hsa04621NOD-like receptor signaling pathway$\frac{3}{14184}$/81491.85E-$\frac{025599}{7128}$/5595hsa04613Neutrophil extracellular trap formation$\frac{3}{14190}$/81491.96E-$\frac{022475}{10}$,$\frac{533}{5595}$hsa05169Epstein-*Barr virus* infection$\frac{3}{14202}$/81492.26E-$\frac{02960}{5599}$/7128hsa05208Chemical carcinogenesis - reactive oxygen species$\frac{3}{14223}$/81492.57E-$\frac{023091}{5599}$/5595hsa04920Adipocytokine signaling pathway$\frac{2}{1469}$/81492.57E-$\frac{022475}{5599}$hsa04137Mitophagy - animal$\frac{2}{1472}$/81492.59E-$\frac{023091}{5599}$hsa05022Pathways of neurodegeneration - multiple diseases$\frac{4}{14476}$/81492.62E-$\frac{022475}{9776}$/$\frac{5599}{5595}$hsa04658Th1 and Th2 cell differentiation$\frac{2}{1492}$/81493.55E-$\frac{025599}{5595}$hsa04933AGE-RAGE signaling pathway in diabetic complications$\frac{2}{14100}$/81493.78E-$\frac{025599}{5595}$hsa04064NF-kappa B signaling pathway$\frac{2}{14104}$/81493.78E-$\frac{027128}{472}$hsa04620Toll-like receptor signaling pathway$\frac{2}{14104}$/81493.78E-$\frac{025599}{5595}$hsa04660T cell receptor signaling pathway$\frac{2}{14104}$/81493.78E-$\frac{025599}{5595}$hsa05016Huntington disease$\frac{3}{14306}$/81493.92E-$\frac{022475}{9776}$/5599hsa04931Insulin resistance$\frac{2}{14108}$/81493.92E-$\frac{022475}{5599}$hsa04722Neurotrophin signaling pathway$\frac{2}{14119}$/81494.53E-$\frac{025599}{5595}$
## Infiltration of immune cells and correlation analysis
The “CIBERSORT” algorithm was applied to calculate the ratios of immune infiltrating cells in PD and HC samples [24]. The number of permutations of default signature matrix was set to 1000 and the standard immune cell expression file (LM22.txt) was obtained from official website (https://cibersort.stanford.edu/) (Accessed: 7 June 2022). The different proportion of immune cells and expression of immune checkpoint genes associated with T cells (Additional file 2: Table S2) between two groups were detected by Wilcoxon rank sum test [25] and spearman correlation analysis was performed on candidate genes and infiltrating immune cells, ICGs.
## Patient enrollment and blood acquisition
70 PD patients and 39 healthy controls were recruited in this study at the Tianjin Huanhu Hospital. The Ethics Committee of Huanhu Hospital approved this study and written informed consent was obtained from all study participants. Disease severity was evaluated using the modified Hoehn and Yahr (H&Y) scale. PD patients were divided into early stage (early) and middle-advanced stages (mid-advanced) groups according to their HY scale. Early stage contained 30 patients (H&Y sca1e 1-2.5) and middle-advanced stage included 40 patients (H&Y scale 3–5). The scale of MDS-UPDRS III (MDS Unified-Parkinson Disease Rating Scale) was used to examine movement function of PD patients. All patients were diagnosed by at least 3 professional and fellowship-trained movement disorders neurologists according to the UK Society Brain Bank Criteria for the diagnosis of PD. Healthy control subjects had no personal or family history of neurodegenerative diseases. Exclusion criteria were as follows: a history of deep brain stimulation and anticancer therapy; major depression; dementia; hepatorenal disease; stroke or other cerebrovascular disease.
2ml EDTA-K2 anticoagulant whole blood was collected in the morning after the subjects fasted for 10 h. Blood was centrifuged at 1000 g for 15 min at room temperature to obtain plasma then stored at -80℃ for further analysis.
## Enzyme-linked immunosorbent assay
Plasmic concentrations of LPIN1 and TNFAIP3 in PD and HC were determined by commercially available enzyme-linked immunosorbent assay (ELISA) kits obtained from Herbal Source (Nanjing, China) and CUSABIO (Wuhan, China), respectively. The assay was performed according to the manufacturer’s instructions and the results were detected using SpectraMax iD5 multifunctional microplate reader at 450 nm (Molecular Devices, the USA).
## Statistical analysis
All data were analyzed using SPSS statistical software (version 26.0), GraphPad Prism software (Version 8.0) and R software (version 4.3.1; including “GEOquery”, “limma”, “WGCNA”, “FactoMineR”, “clusterProfiler”, “GSVA”, “GSEA”, “glmnet”, “randomforest”, “e1071”, “CIBERSORT”, “pROC”, “ggplot2” and “stats” package). For all analysis, p value < 0.05 was considered statistically significant. Data normality was first evaluated using Shapiro–Wilk test, then t test was used to compare data with normal distribution between two groups, and Mann-Whitney U test was used to compare data of non-normal distribution between two groups. One-way ANOVA analysis or Kruskal-Wallis test was used to compare data among three groups. Data were presented as mean ± standard deviation (SD) or median (quantile). Chi-square test was used for comparing sex ratios between PD patients and healthy controls. Receiver operating characteristic (ROC) curves were generated to evaluate their sensitivities and specificities in distinguishing PD from the healthy controls.
## Identification of key WGCNA module and DEGs
After the cluster analysis, no samples were removed (Additional file 3: Figure S3). The WGCNA network was constructed based on the GSE18838 dataset to identify the meaningful gene modules related with PD. A soft threshold power of 12 was selected, the scale-free topology fit index R^2 reached 0.84, and mean connectivity is 18.10, indicating that a scale-free network was established (Fig. 2A, B). Co-expression gene modules were identified through the dynamic tree cut method, after merging similar modules, the key modules were further screened based on MEDissThres = 0.25 (Fig. 2C, D). Then we analyzed the relationship between the key modules and clinical phenotype, and the heatmap of all genes in the key modules was displayed (Fig. 2E, F). Among the 10 modules analyzed, the greenyellow module was significantly associated with the clinical traits of PD and was chosen as a key module (cor = 0.49, $$p \leq 0.008$$, Fig. 2E). We selected 884 genes for following research according to the criterion of q.weighted < 0.05. Besides, a high correlation was observed between PD and the greenyellow module (cor = 0.492) while the correlation between module memberships (MM) and gene significance (GS) in the greenyellow module is 0.28 (cor = 0.28, $$p \leq 2.3$$e-07, Fig. 2G).
Fig. 2 WGCNA network and module detection. A Selection of the soft-thresholding powers. The left panel displays the scale-free index versus soft-thresholding power. The right panel shows the mean connectivity versus soft-thresholding power. The x-axis represents weighting parameters (power). The y-axis represents the scale-free fit index and connectivity for each power. B Histogram of the number of node connections and validation that the network conforms to a scale-free distribution at a given threshold. C Module division. D Module merge. Each color represents a module in the co-expression network by WGCNA. E Heatmap of the correlation between module and PD samples traits. F The heatmap visualizing the gene network. G Scatterplot showing the correlation between gene significance and module membership in the greenyellow module Additionally, 399 DEGs between PD and HC samples were obtained through the PCA and different expression analysis (Fig. 3A, B). The intersect between DEGs, FRGs and co-expression genes in interesting module was visualized by Venn plot (Fig. 3C), thus we screened 15 ferroptosis-related-WGCNA genes and 179 WGCNA-DEGs.
Fig. 3 PCA plot of gene chip and volcano plot of different expression genes. A PCA analysis plot of GSE18838 gene chip. B Volcano plot of differential expressed genes between PD and HC samples in GSE18838 dataset. C Venn diagram displaying the overlap between DEGs, FRGs and PD-related genes identified by WGCNA.
## GSEA and GSVA
We performed GSEA and GSVA analysis to screen biological differences between PD and HC. The enrichment analysis results of GO-BP, GO-CC and GO-MF were displayed (Fig. 4A, B, C). When KEGG and hallmark gene sets as the reference sets, the GSVA enrichment analysis revealed that PI3K-AKT-mTOR signaling, reactive oxygen species pathway, P53 signaling pathway and regulation of autophagy were involved in the pathogenesis of PD (Fig. 4D, E). We also found some related immunologic pathways significantly enriched between PD and HC (Fig. 4F). In addition, GSEA analysis of KEGG pathway uncovered some underlying pathways in PD (Table 1), such as autophagy, apoptosis, necroptosis, NOD-like receptor signaling pathway, TNF signaling pathway, ubiquitin mediated proteolysis, cellular senescence, mitophagy, Parkinson disease, alcoholic liver disease and neutrophil extracellular trap formation.
Fig. 4 The results of different reference gene sets of GSVA. A GO-BP gene sets. B GO-CC gene sets. C GO-MF gene sets. D KEGG gene sets. E *Hallmarker* gene sets. F Immunologic signatures gene sets Table 1The KEGG pathway of GSEA analysisIDDescriptionSet sizeEnrichment coreNESRankhsa04140Autophagy - animal127-0.488-1.8403669hsa04145Phagosome127-0.423-1.5942572hsa05169Epstein-*Barr virus* infection187-0.524-2.0502794hsa04062Chemokine signaling pathway173-0.454-1.7662868hsa04210Apoptosis130-0.432-1.6343326hsa04621NOD-like receptor signaling pathway162-0.423-1.6382793hsa04910Insulin signaling pathway130-0.443-1.6773804hsa04514Cell adhesion molecules128-0.464-1.7491477hsa04218Cellular senescence144-0.459-1.7563332hsa04142Lysosome125-0.515-1.9293342hsa05206MicroRNAs in cancer159-0.454-1.7513033hsa04936Alcoholic liver disease134-0.424-1.6082460hsa05152Tuberculosis165-0.494-1.9162480hsa05161Hepatitis B155-0.510-1.9623092hsa05162Measles134-0.520-1.9702403hsa04613Neutrophil extracellular trap formation102-0.556-2.0393033hsa04071Sphingolipid signaling pathway111-0.498-1.8402794hsa04650Natural killer cell mediated cytotoxicity110-0.554-2.0413065hsa04668TNF signaling pathway110-0.534-1.9673122hsa04931Insulin resistance104-0.515-1.8863804hsa04144Endocytosis222-0.361-1.4292580hsa04120Ubiquitin mediated proteolysis130-0.398-1.5053719hsa04620Toll-like receptor signaling pathway94-0.435-1.5773316hsa04080Neuroactive ligand-receptor interaction3420.3581.6773930hsa05012Parkinson disease2200.3171.4161331hsa04217Necroptosis122-0.391-1.4653016hsa01200Carbon metabolism105-0.408-1.4951901hsa04137Mitophagy - animal65-0.444-1.5013894hsa04730Long-term depression56-0.465-1.5333459
## Candidate genes selected by machine learning methods
We used LASSO logistic regression algorithm to identify 8 genes from 15 ferroptosis-related-WGCNA genes as key biomarkers for PD (Fig. 5A), RF and SVM-RFE algorithm were also used to screen candidate genes (Fig. 5B, C). *Overlapped* genes obtained via three algorithms were considered as candidate biomarkers, and finally two genes, LPINI and TNFAIP3 were attained as the biomarkers (Fig. 5D). The KEGG pathway of GSEA analysis on two characteristic genes were shown (Fig. 6A, B). LPINI involved in alcoholic liver disease and TNFAIP3 mainly related to Epstein-*Barr virus* infection, measles, necroptosis, NOD-like receptor signaling pathway and TNF signaling pathway. In order to further test the diagnostic efficacy of LPINI and TNFAIP3 for PD, we analyzed the expression levels and validated with the GSE18838 microarray expression matrix. Then we found the two genes were downregulated in PD whole blood and ROC curve indicated that they had better diagnostic potential, the AUC is 0.872 ($95\%$ CI: 0.723-1.000) and 0.818 ($95\%$ CI: 0.647–0.989) for LPINI and TNFAIP3, respectively (Fig. 6C, D). Moreover, GSE72267 was treated as a validation data set including 40 PD patients and 20 healthy controls (Additional file 4: Figure S4). The TF-miRNA coregulatory network of LPIN1 and TNFAIP3 was established on NetworkAnalyst (Fig. 6E).
Fig. 5 Identification of candidate genes associated with diagnosis using the machine learning method. A LASSO regression analysis. B Random Forest. C Support Vector Machine. D Venn diagram for screened candidate genes between LASSO, RF and SVM. RF: random forest; SVM: support vector machine; LASSO: least absolute shrinkage and selection operator analysis Fig. 6 The KEGG pathway and the validation in GSE18838 dataset of candidate genes. A The plot showing the KEGG pathways enriched by LPINI. B The plot showing the KEGG pathways enriched by TNFAIP3. C The expression levels of LPINI and TNFAIP3 in GSE18838. D The ROC curve of two candidate genes. E The TF-miRNA coregulatory network of LPINI and TNFAIP3. Circle represents protein, diamond represent transcription factor (TF), and arrow represent miRNA. ** $p \leq 0.01$, ***$p \leq 0.001$
## Estimation of infiltrating immune cells and correlation analysis
Firstly, we estimated the proportion of 22 infiltrating immune cells using the gene matrix of 28 samples by “CIBERSORT” algorithm. Compared to the results for HC, the proportions of naïve B cells, plasma cells, naïve CD4 T cells, regulatory T cells, macrophages M0, and macrophages M1 were significantly lower in the PD samples, while the proportions of memory B cells, gamma delta T cells, and resting dendritic cells were significantly higher (Fig. 8A). Positive and negative relationships between candidate genes and infiltrating immune cells were all discovered via spearman analysis. LPINI had positive correlation with naïve B cells, plasma cells and naïve CD4 T cells, while had negative correlation with memory B cells, gamma delta T cells and resting dendritic cells. TNFAIP3 had positive correlation with naïve B cells, naïve CD4 T cells, regulatory T cells, macrophages M0 and macrophages M1, while had negative correlation with gamma delta T cells and resting dendritic cells (Fig. 8B).
Fig. 8 *The status* of immune cell infiltration and expression of immune checkpoint genes. A Boxplots comparing the proportions of 22 major immune cell subsets between PD and HC samples. B Correlation between LPINI, TNFAIP3, and infiltrating immune cells by CIBERSORT. C The expression of immune checkpoint genes between PD and HC samples. D Correlation between LPINI, TNFAIP3, and ICGs.
In addition, for immune checkpoint genes expressed on T cells, TNFRSF18, TNFRSF25, CD28, CTLA-4, ICOS, BTLA, MYLK, CD27, CD226, ADORA2A and CD40L were different significantly between two groups (Fig. 8C). For correlation analysis between candidate genes and immune checkpoint genes, which were displayed on Fig. 8D. LPIN1 had significant correlations with all the above different ICGs, however, TNFAIP3 only was correlated with TNFRSF18, TNFRSF25, CD28, ICOS, MYLK, CD226, ADORA2A and CD40L.
## Demographic and clinical characteristics of the PD patients and healthy controls
Demographic characteristics of participants are summarized in Table 4, the clinical characteristics of early and middle-advanced PD patients are shown in Table 5. Between the healthy controls and PD patients, RBC, Hb, Hct, the ratio of monocyte and lymphocyte were significantly different ($$p \leq 0.000$$, $$p \leq 0.000$$, $$p \leq 0.000$$, $$p \leq 0.031$$). Furthermore, the difference in WBC, RBC, Hb, Hct were also statistically significant between the HC and early-stage PD patients ($$p \leq 0.031$$, $$p \leq 0.000$$, $$p \leq 0.000$$, $$p \leq 0.000$$) (Table 4). For the comparison of early and middle-advanced stage PD patients, age, disease duration (years), the score of MDS-UPDRS III “off”, WBC, neutrophils (%), lymphocyte (%), the ratio of neutrophils and lymphocyte, the ratio of monocyte and lymphocyte also had statistical difference ($$p \leq 0.002$$, $$p \leq 0.000$$, $$p \leq 0.000$$, $$p \leq 0.026$$, $$p \leq 0.003$$, $$p \leq 0.001$$, $$p \leq 0.002$$, $$p \leq 0.031$$) (Table 5). There was a significant difference in the UPDRS score between the early and middle-advanced stage of PD, which was in line with the disease degree of two stages.
Table 4Demographic and clinical characteristics of the PD patients and healthy controls used in this studyVariablePD ($$n = 70$$)HC ($$n = 39$$)p valueAge, years65.42(8.85)63.38(9.03)0.254Male/female ratio$\frac{41}{2925}$/140.571Disease duration, years7.16(3.94)H&Y stage, off (1|1.5|2|2.5|3|4|5|)5|7|10|8|28|10|2MDS-UPDRS III “off” [0-132]48.94(18.64)WBC, 10^9/L6.09(2.09)6.29(1.15)0.064Neutrophils (%)62.28(10.90)59.03(6.63)0.202Lymphocyte (%)29.14(9.66)32.18(6.77)0.084Monocytes (%)6.19(2.15)6.02(1.35)0.972 N/L2.87(2.94)1.97(0.68)0.121M/L0.24(0.15)0.19(0.07)0.031RBC, 1012/L4.37(0.48)4.78(0.40)0.000Hb, g/L133.97(13.98)145.38(11.76)0.000Hct, L/L0.402(0.04)0.440(0.033)0.000Values are means ± SD unless otherwise stated. PD, Parkinson’s disease; HC, healthy controls; H&Y stage, Hoehn and Yahr scale; MDS-UPDRS III, Movement Disorders Society-Unified Parkinson Disease Rating Scale, motor part; N/L, neutrophils/lymphocyte; M/L, monocytes/ lymphocyte Table 5Demographic and clinical characteristics of early and middle-advanced PD patientsVariableEarly($$n = 30$$)Mid-advanced ($$n = 40$$)p valueAge, years61.60(9.58)68.30(7.12)0.002Male/female ratio$\frac{16}{1425}$/150.441Disease duration, years5.28(3.46)8.58(3.72)0.000MDS-UPDRS III “off” [0-132]38.97(15.70)56.43(17.24)0.000WBC, 10^9/L5.51(1.79)6.52(2.21)0.026Neutrophils (%)57.89(10.40)65.57(10.20)0.003Lymphocyte (%)33.34(9.51)25.99(8.60)0.001Monocytes (%)6.71(2.68)5.81(1.57)0.146 N/L2.41(3.47)3.21(2.46)0.002M/L0.24(0.20)0.24(0.08)0.031RBC, 10^12/L4.38(0.48)4.37(0.49)0.927Hb, g/L133.27(14.68)134.5(13.59)0.718Hct, L/L0.400(0.041)0.403(0.039)0.739Values are means ± SD unless otherwise stated. PD, Parkinson’s disease; HC, healthy controls; H&Y stage, Hoehn and Yahr scale; MDS-UPDRS III, Movement Disorders Society-Unified Parkinson Disease Rating Scale, motor part; N/L, neutrophils/lymphocyte; M/L, monocytes/ lymphocyte
## Plasmic levels of LPIN1 and TNFAIP3 in PD patients and healthy controls
The plasmic concentration of LPIN1 in patients with PD (105.7 ng/mL [range 56.98 to 161.3 ng/mL]) was significantly lower than that in HC (121.0 ng/mL [range 87.03 to 773.4 ng/mL]) ($p \leq 0.0001$) (Fig. 9A). While there was a significant increase of TNFAIP3 plasma concentration in PD patients (45.91 pg/ml [range 4.61 to 193.9 pg/ml]) compared with HC (20.50 pg/ml [range 5.84 to 159.5 pg/ml]) ($p \leq 0.0001$) (Fig. 9B). When the PD patients were divided into early stage and middle-advanced stage, the plasma level of LPIN1in early stage PD (101.7 ng/mL [range 77.96 to 137.7 ng/mL]) was significantly lower than that in HC ($p \leq 0.0001$), while there was no statistically significant difference between early and middle-advanced stage PD patients (110.0 ng/mL [range 56.98 to 161.3 ng/mL]) ($$p \leq 0.2806$$) (Fig. 9C). A significant elevation of TNFAIP3 level in early stage PD patients (35.06 pg/mL [range 4.61 to 135.2 pg/mL]) compared with HC was found ($$p \leq 0.0407$$), as well as there was also significant difference between early stage and middle-advanced stage PD patients (50.63 pg/mL [range 7.75 to 193.9 pg/mL]) ($$p \leq 0.0459$$) (Fig. 9D). Furthermore, a correlation plot between the expression levels of two molecules and clinical parameter was shown in Additional file 5: Figure S5, TNFAIP3 had a weak correlation with age, basophil, Hoehn and Yahr scale, disease stage.
Fig. 9 The ELISA verification of two biomarkers. A, B *The plasma* level of LPIN1 and TNFAIP3 in HC and PD. C, D *The plasma* level of LPIN1 and TNFAIP3 in HC, early and middle-advanced PD patients. HC: healthy controls; PD: Parkinson’s disease; early: early stage; mid-advanced: middle and advanced stage. * $p \leq 0.05$, ****$p \leq 0.0001.$ ns, no significance
## Diagnostic value of plasmic LPIN1 and TNFAIP3 in PD
Receiver operating characteristic (ROC) curves were applied to evaluate the potential diagnostic value of LPIN1 and TNFAIP3 in PD. The area under ROC curve (AUC) of LPIN1 and TNFAIP3 for PD were 0.754 ($95\%$ CI: 0.659–0.849, $p \leq 0.0001$, sensitivity = 0.771, specificity = 0.692) and 0.754 ($95\%$ CI: 0.660–0.849, $p \leq 0.0001$, sensitivity = 0.686, specificity = 0.821) (Fig. 10A) (Additional file 6: Table S6). In distinguishing the early stage PD from HC, the AUC of LPIN1 and TNFAIP3 were 0.817 ($95\%$ CI: 0.717–0.917, $p \leq 0.0001$, sensitivity = 0.867, specificity = 0.692) and 0.650 ($95\%$ CI: 0.507–0.794, $$p \leq 0.040$$, sensitivity = 0.667, specificity = 0.718) (Fig. 10B) (Additional file 7: Table S7). However, LPIN1 and TNFAIP3 don’t performed well in distinguishing the early stage from middle-advanced stage PD patients (LPIN1: AUC = 0.599, $95\%$ CI: 0.465–0.733, $$p \leq 0.146$$; TNFAIP3: AUC = 0.647, $95\%$ CI: 0.510–0.783, $$p \leq 0.035$$) (Fig. 10C) (Additional file 8: Table S8). Then, we used logistic regression analysis and the results indicated that LPIN1 and TNFAIP3 performed better in combination for prediction (HC vs. PD, AUC = 0.833, $95\%$ CI: 0.750–0.916, $p \leq 0.0001$; HC vs. early PD, AUC = 0.831, $95\%$ CI: 0.734–0.927, $p \leq 0.0001$) (Fig. 10D, E), while the diagnostic efficacy was relatively poor in discriminating early and middle-advanced PD (AUC = 0.637, $95\%$ CI: 0.505–0.768, $$p \leq 0.041$$) (Fig. 10F).
Fig. 10 The ROC of two biomarkers. A, B, C Each biomarker plot one ROC (HC vs. PD, HC vs. early PD, early vs. middle-advanced PD). D, E, F Two biomarkers combined using binary logistic regression model (HC vs. PD, HC vs. early PD, early vs. middle-advanced PD). sensitivity (true positive rate) and 1-specificity (false positive rate); AUC: area under curve; CI: $95\%$ confidence interval
## Discussion
Herein, we performed WGCNA analysis, intersected between DEGs, FRGs and interesting module, then identified 15 ferroptosis-related WGCNA genes and 179 WGCNA-DEGs genes. Enrichment analysis including GSVA, GSEA, GO and KEGG were operated. LPINI and TNFAIP3, as candidate genes, were determined by machine learning method (LASSO, SVM and RF). Moreover, LPINI and TNFAIP3 were differently expressed in the plasma of PD patients and healthy controls detected by ELISA. With the estimation of infiltrating immune cells and correlation analysis, we found the FRGs was associated with ICGs, immune regulation. In addition, ROC curve indicated that LPINI and TNFAIP3 may provide a novel diagnostic biomarker for PD. These results demonstrated that candidate genes might participate in the processes of regulating immune cell infiltration and immune checkpoint genes expression in PD.
Aging is a major risk factor for various neurodegenerative disorders and accompany with gently accumulation of iron in the brain that relates with lipid peroxidation and reactive oxygen species production that represents the state of oxidative stress [5]. Iron can upregulate the levels of α-synuclein, amyloid precursor protein (APP) and amyloid β-peptide (Aβ) [5]. Selective deposition of iron in SN is one of the essential pathogenic factors [12], glutathione (GSH) loss in SN and oxidative stress are predispositions to PD [5]. In addition, recent emerging evidence suggests that ferroptosis is a prevalent cell death pathway for dopaminergic neurons [16]. For example, iron accumulation in aging glial cells could impair neurons by increasing proinflammatory factors to establish neuroinflammation [26]. Ferroptosis is defined as Fe (II)-dependent regulated necrosis accompanied lipid peroxidation [27], a mitochondria-dependent type of cell death [28], which was an important cell death pathway in Lund human mesencephalic cells, these had been confirmed ex vivo (in organotypic slice cultures) and in vivo (in the MPTP mouse model of PD) [16]. A study found that when SH-SY5Y human neuroblastoma cells were treated with PQ (paraquat dichloride) and Fer-1 (a specific inhibitor of ferroptosis) together, Fer-1 could inhibit the production of lipid reactive oxygen species and ameliorate ferroptosis by upregulating the expression of GPX4 (glutathione peroxidase 4) and SLC7A11 (cystine/glutamate antiporter). Fer-1 also inhibited the accumulation of ferrous iron in mitochondria, protected against PQ-induced damage, and maintained mitochondrial integrity [29]. Moreover, mounting studies have shown that potential physiological roles of ferroptosis in cancer, ischemia/reperfusion injuries, neurodegeneration and other pathological conditions, nevertheless the exact contribution of ferroptosis to these pathologies is unclear [5].
Lipin1 is a Mg2+-dependent phosphatidic acid phosphatase (PAP) enzyme closely related to glycolipid metabolism, produced by the expression of LPIN1 [30], referred as a member of the lipin family, which converts phosphatidic acid (PA) to diacylglycerol (DAG), a precursor of triacylglycerol and phospholipids [30, 31]. Additionally, LPIN1 functions as a transcriptional coregulator via directly interacting with nuclear peroxisome proliferator-activated receptor α (PPARα) and PPARα co-stimulatory factor 1 α (PGC1α) to regulate the genes involved in fatty acid oxidation [32]. It is reported that LPIN1 can promote several processes, including cell differentiation, inflammation and autophagy [31]. The human lipin1 has three isoforms (lipin1α, lipin1β, lipin1γ) derived from alternative mRNA splicing. Lipin1α and lipin1β are lowly expressed in the brain, conversely, lipin1γ is highly expressed in normal human brain, indicating that lipin1γ may be a specialized regulatory protein in brain lipid metabolism [32, 33]. Latest study confirmed the presence of cognitive impairment in the mice with hippocampus of Lipin1-deficient, including the worsen spatial learning and memory ability, decreased synapse number, reduced protein levels of BDNF, SYP and PSD95. Shang et al. reported that lipin1 impaired synaptic plasticity, disturbed lipid homeostasis, and damaged spatial learning and memory by inhibiting DAG-PKD-ERK signaling pathway in Fld mice (a mutation in the *Lpin1* gene) [34]. In another research, authors considered that neuroprotection of LPIN1 was associated with inhibition of the PKD/Limk1/Cofilin signaling pathway, and LPIN1 might ameliorate the cognitive impairments in Diabetic encephalopathy (DE) animal models [35]. The loss of Lipin1 decreases DAG expression, which may lead to lipid metabolism disorders, induce autophagy overaction and promote Diabetic Peripheral Neuropathy (DPN). In contrast, overexpression of Lipin1 can reduce autophagy disorders and alleviate DPN [36]. Autophagy plays an important role in neurodegenerative diseases and nerve tissue injury [37].
A20, also known as TNF-α-induced protein 3 (TNFAIP3), is a ubiquitin editing enzyme with both E3 ubiquitin ligase activity and deubiquitinating enzyme (DUB) activity [38], also functions as a key negative regulator of NF-κB transcription factors and an anti-inflammatory molecule that plays an important part in both immune responses and cell death [39], which can suppress NF-κB signaling downstream from T cell receptor (TCR), B cell receptor (BCR), tumor necrosis factor receptor (TNFR), interleukin 1 receptor (IL-1R), Toll-like receptors (TLRs), NOD-like receptors (NLRs) and so on [40]. NF-κB signaling pathway can activate the innate and adaptive immune system, yet its improper activation indicates the development of chronic inflammation and cell death [41]. Moreover, NF-κB has been implicated in the pathogenesis of a variety of neurodegenerative diseases [42]. TNFAIP3, as a central negative regulator of NF-κB transcription factors by multiple mechanisms, which probably has functions in the regulation of NF-κB signaling in astrocytes and in neurons within the CNS [42]. Microglia A20 deficiency exacerbated multiple sclerosis (MS)-like disease, due to hyperactivation of the NLRP3 inflammasome leading to increased interleukin-1β secretion in mice, suggesting that A20 critically controls microglia activation and inhibits inflammasome-dependent neuroinflammation [43]. After deleting A20 in microglia, CD8 + T cells spontaneously infiltrate the CNS and acquire a viral response signature, also upregulate genes associated with the antiviral response and neurodegenerative diseases [44].
As a regulator of cell death, on the hand, A20 can inhibit TNFα-induced apoptosis through the inhibition of phospholipase A2 and caspase 8 activation, reduce production of reactive oxygen species, diminish collapse of mitochondrial membrane potential, suppress the c-Jun N-terminal kinase and pro-inflammatory cytokines [42]. A20 can also restrict necroptosis in T cells and macrophages via its deubiquitinating motif [38]. On the other hand, A20 may have a proapoptotic function and restrict cell survival, probably due to upregulation of NF-κB-dependent antiapoptotic proteins Bcl-2 and Bcl-x [38]. A20 has been shown to promote survival of CD4 + T cells by restricting the ubiquitylation-dependent activation of mTOR and promoting autophagy [45]. Gradually, depending on the cell type and activated signaling pathway, more evidence indicates that A20 can indirectly counteract inflammatory response by protecting cells from death, which largely dependents on its ubiquitin-binding properties [38, 46].
In the present study, when KEGG and hallmark gene sets as the reference sets, the GSVA enrichment analysis revealed that reactive oxygen species pathway, p53 pathway and regulation of autophagy were involved in the pathogenesis of PD (Fig. 4D, E). For the KEGG analysis of GSEA, we found some pathways including autophagy–animal, apoptosis, NOD-like receptor signaling pathway, cellular senescence, lysosome, Parkinson disease, necroptosis and so on (Table 1). Furthermore, LPIN1 and TNFAIP3 were also involved in the regulation of mentioned signal pathway.
Lastly, we performed immune infiltration analysis on the peripheral blood microarray expression matrix of PD and compared the expression of immune checkpoint genes related to T cells, then revealed that the proportion of immune cells and expression of ICGs were significantly different between two groups. Previous work has started to elucidate the complex effects of ferroptosis on different aspects of the immune function [47]. on the one hand, ferroptosis affects the number and function of immune cells. On the other hand, ferroptotic cells can be recognized by immune cells and then trigger a series of specific inflammatory responses. Furthermore, ferroptosis of immune cells may destroy immune response, and ferroptosis of non-immune cells may cause the release of DAMPs (danger-associated molecular patterns) that induces immune activation [47]. As a programmed necroptosis, ferroptosis is inherently more immunogenic than apoptosis and results in the release of inflammatory cytokines, leading to necro-inflammatory response, which can drive the pro-inflammatory state in certain biological contexts [48]. Because of its high metabolic activity, brain tissue is particularly susceptible to oxidative stress that is a hallmark of various neurodegenerative disorders [49]. Cells under oxidative stress may release immunogenic molecules that triggers a systemic immune response, ultimately leading to cell necrosis [48]. In line with the above mentioned, the specific necrotic signaling pathway of ferroptosis may produce pathogenic cytokines peroxides that impairs the immune response via activating immune cells [48].
In addition, our experiment showed that LPIN1 was under-expressed and TNFAIP3 was upregulated in the plasma of PD patients that was consistent to the validation in GSE72267 (Additional file 4: Figure S4). A previous real-time PCR assay also showed decreased TNFAIP3 expression in PD whole blood samples [50], while in the GSE18838 microarray expression matrix, LPIN1 and TNFAIP3 both were downregulated in PD whole blood. Each biomarker alone could discriminate the PD and HC (AUC > 0.75), however, TNFAIP3 didn’t performed well in distinguishing the early PD from healthy controls (LPIN1: AUC = 0.817, false positive rate = 0.308, false negative rate = 0.133; TNFAIP3: AUC = 0.650, false positive rate = 0.282, false negative rate = 0.333). The diagnostic model formed by the combination of two biomarkers had an AUC of 0.833 (sensitivity = 0.671, specificity = 0.923) in distinguishing PD from HC and an AUC of 0.831 (sensitivity = 0.900, specificity = 0.692) in distinguishing the early PD from HC.
In this study, there are still some limitations. Firstly, The TNFAIP3 levels are inconsistent in different population samples, numerous variables can lead to the inconsistent results, such as choices of assays, methods of sample acquisition, drug treatment, disease severity. Besides, existing clinical information remains incomplete, and validation is required at the genetic level of clinical samples by multiple methods. Therefore, to objectively evaluate the diagnostic effects of LPIN1 and TNFAIP3, it is necessary to strictly control the inclusion and exclusion criteria of PD subjects and collect more complete, accurate clinical data to regulate the influence of other miscellaneous variables on experiment.
## Conclusion
In summary, our results confirmed abnormally under-expression or upregulation of LPINI and TNFAIP3 in the PD plasma, ferroptotic cells and circulating immune system responses are implicated in the pathogenesis of PD. Furthermore, ferroptosis-related genes have correlations with immune checkpoint genes, immune infiltration. Thus, this study further improved the understanding of the effect mechanism of ferroptosis on peripheral blood mononuclear cells (mainly including lymphocyte and monocyte). However, the specific mechanism of LPINI and TNFAIP3 regulate ferroptosis and immunity in PD is not clear. More research is needed to explore the biological effects of LPINI and TNFAIP3 on peripheral immune cells and provide reliably clinical diagnostic markers for PD.
## Electronic supplementary material
Below is the link to the electronic supplementary material.
Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Supplementary Material 4
Supplementary Material 5 Supplementary Material 6 Supplementary Material 7 Supplementary Material 8
## References
1. Bloem BR, Okun MS, Klein C. **Parkinson’s disease**. *Lancet* (2021.0) **397** 2284-303. DOI: 10.1016/S0140-6736(21)00218-X
2. Hayes MT. **Parkinson’s disease and parkinsonism**. *Am J Med* (2019.0) **132** 802-7. DOI: 10.1016/j.amjmed.2019.03.001
3. Jankovic J, Tan EK. **Parkinson’s disease: etiopathogenesis and treatment**. *J Neurol Neurosurg Psychiatry* (2020.0) **91** 795-808. DOI: 10.1136/jnnp-2019-322338
4. Dixon SJ, Lemberg KM, Lamprecht MR, Skouta R, Zaitsev EM, Gleason CE. **Ferroptosis: an iron-dependent form of nonapoptotic cell death**. *Cell* (2012.0) **149** 1060-72. DOI: 10.1016/j.cell.2012.03.042
5. Yan HF, Zou T, Tuo QZ, Xu S, Li H, Belaidi AA. **Ferroptosis: mechanisms and links with diseases**. *Signal Transduct Target Ther* (2021.0) **6** 49. DOI: 10.1038/s41392-020-00428-9
6. Yu H, Guo P, Xie X, Wang Y, Chen G. **Ferroptosis, a new form of cell death, and its relationships with tumourous diseases**. *J Cell Mol Med* (2017.0) **21** 648-57. DOI: 10.1111/jcmm.13008
7. Wang H, Cheng Y, Mao C, Liu S, Xiao D, Huang J. **Emerging mechanisms and targeted therapy of ferroptosis in cancer**. *Mol Ther* (2021.0) **29** 2185-208. DOI: 10.1016/j.ymthe.2021.03.022
8. Ward RJ, Zucca FA, Duyn JH, Crichton RR, Zecca L. **The role of iron in brain ageing and neurodegenerative disorders**. *Lancet Neurol* (2014.0) **13** 1045-60. DOI: 10.1016/S1474-4422(14)70117-6
9. Tang D, Kroemer G. *Curr Biol* (2020.0) **30** R1292-R7. DOI: 10.1016/j.cub.2020.09.068
10. Sfera A, Bullock K, Price A, Inderias L, Osorio C. **Ferrosenescence: the iron age of neurodegeneration?**. *Mech Ageing Dev* (2018.0) **174** 63-75. DOI: 10.1016/j.mad.2017.11.012
11. Angelova PR, Choi ML, Berezhnov AV, Horrocks MH, Hughes CD, De S. **Alpha synuclein aggregation drives ferroptosis: an interplay of iron, calcium and lipid peroxidation**. *Cell Death Differ* (2020.0) **27** 2781-96. DOI: 10.1038/s41418-020-0542-z
12. Xuan M, Guan X, Gu Q, Shen Z, Yu X, Qiu T. **Different iron deposition patterns in early- and middle-late-onset Parkinson’s disease**. *Parkinsonism Relat Disord* (2017.0) **44** 23-7. DOI: 10.1016/j.parkreldis.2017.08.013
13. Wang ZL, Yuan L, Li W, Li JY. **Ferroptosis in Parkinson’s disease: glia-neuron crosstalk**. *Trends Mol Med* (2022.0) **28** 258-69. DOI: 10.1016/j.molmed.2022.02.003
14. Mahoney-Sanchez L, Bouchaoui H, Ayton S, Devos D, Duce JA, Devedjian JC. **Ferroptosis and its potential role in the physiopathology of Parkinson’s Disease**. *Prog Neurobiol* (2021.0) **196** 101890. DOI: 10.1016/j.pneurobio.2020.101890
15. Zhang P, Chen L, Zhao Q, Du X, Bi M, Li Y. **Ferroptosis was more initial in cell death caused by iron overload and its underlying mechanism in Parkinson’s disease**. *Free Radic Biol Med* (2020.0) **152** 227-34. DOI: 10.1016/j.freeradbiomed.2020.03.015
16. Do Van B, Gouel F, Jonneaux A, Timmerman K, Gele P, Petrault M. **Ferroptosis, a newly characterized form of cell death in Parkinson’s disease that is regulated by PKC**. *Neurobiol Dis* (2016.0) **94** 169-78. DOI: 10.1016/j.nbd.2016.05.011
17. Gouel F, Do Van B, Chou ML, Jonneaux A, Moreau C, Bordet R. **The protective effect of human platelet lysate in models of neurodegenerative disease: involvement of the akt and MEK pathways**. *J Tissue Eng Regen Med* (2017.0) **11** 3236-40. DOI: 10.1002/term.2222
18. Devos D, Cabantchik ZI, Moreau C, Danel V, Mahoney-Sanchez L, Bouchaoui H. **Conservative iron chelation for neurodegenerative diseases such as Parkinson’s disease and amyotrophic lateral sclerosis**. *J Neural Transm (Vienna)* (2020.0) **127** 189-203. DOI: 10.1007/s00702-019-02138-1
19. Shehadeh LA, Yu K, Wang L, Guevara A, Singer C, Vance J. **SRRM2, a potential blood biomarker revealing high alternative splicing in Parkinson’s disease**. *PLoS ONE* (2010.0) **5** e9104. DOI: 10.1371/journal.pone.0009104
20. 20.Zhou N, Bao J. FerrDb: a manually curated resource for regulators and markers of ferroptosis and ferroptosis-disease associations. Database (Oxford). 2020;2020.
21. Langfelder P, Horvath S. **WGCNA: an R package for weighted correlation network analysis**. *BMC Bioinformatics* (2008.0) **9** 559. DOI: 10.1186/1471-2105-9-559
22. Liberzon A, Birger C, Thorvaldsdottir H, Ghandi M, Mesirov JP, Tamayo P. **The Molecular Signatures database (MSigDB) hallmark gene set collection**. *Cell Syst* (2015.0) **1** 417-25. DOI: 10.1016/j.cels.2015.12.004
23. Yu G, Wang LG, Han Y, He QY. **clusterProfiler: an R package for comparing biological themes among gene clusters**. *OMICS* (2012.0) **16** 284-7. DOI: 10.1089/omi.2011.0118
24. Chen B, Khodadoust MS, Liu CL, Newman AM, Alizadeh AA. **Profiling Tumor infiltrating Immune cells with CIBERSORT**. *Methods Mol Biol* (2018.0) **1711** 243-59. DOI: 10.1007/978-1-4939-7493-1_12
25. 25.Hu FF, Liu CJ, Liu LL, Zhang Q, Guo AY. Expression profile of immune checkpoint genes and their roles in predicting immunotherapy response. Brief Bioinform. 2021;22(3).
26. 26.Ward RJ, Dexter DT, Crichton RR. Iron, Neuroinflammation and Neurodegeneration. Int J Mol Sci. 2022;23(13).
27. Toyokuni S, Yanatori I, Kong Y, Zheng H, Motooka Y, Jiang L. **Ferroptosis at the crossroads of infection, aging and cancer**. *Cancer Sci* (2020.0) **111** 2665-71. DOI: 10.1111/cas.14496
28. Wang H, Liu C, Zhao Y, Gao G. **Mitochondria regulation in ferroptosis**. *Eur J Cell Biol* (2020.0) **99** 151058. DOI: 10.1016/j.ejcb.2019.151058
29. Zuo Y, Xie J, Li X, Li Y, Thirupathi A, Zhang J. **Ferritinophagy-mediated ferroptosis involved in Paraquat-Induced neurotoxicity of dopaminergic neurons: implication for neurotoxicity in PD**. *Oxid Med Cell Longev* (2021.0) **2021** 9961628. DOI: 10.1155/2021/9961628
30. Harris TE, Finck BN. **Dual function lipin proteins and glycerolipid metabolism**. *Trends Endocrinol Metab* (2011.0) **22** 226-33. DOI: 10.1016/j.tem.2011.02.006
31. Reue K, Wang H. **Mammalian lipin phosphatidic acid phosphatases in lipid synthesis and beyond: metabolic and inflammatory disorders**. *J Lipid Res* (2019.0) **60** 728-33. DOI: 10.1194/jlr.S091769
32. Csaki LS, Dwyer JR, Fong LG, Tontonoz P, Young SG, Reue K. **Lipins, lipinopathies, and the modulation of cellular lipid storage and signaling**. *Prog Lipid Res* (2013.0) **52** 305-16. DOI: 10.1016/j.plipres.2013.04.001
33. Han GS, Carman GM. **Characterization of the human LPIN1-encoded phosphatidate phosphatase isoforms**. *J Biol Chem* (2010.0) **285** 14628-38. DOI: 10.1074/jbc.M110.117747
34. Shang P, Zheng F, Han F, Song Y, Pan Z, Yu S. **Lipin1 mediates cognitive impairment in fld mice via PKD-ERK pathway**. *Biochem Biophys Res Commun* (2020.0) **525** 286-91. DOI: 10.1016/j.bbrc.2020.02.070
35. Xie M, Wang M, Liu W, Xu M, Shang P, Jiang D. **Lipin1 is involved in the Pathogenesis of Diabetic Encephalopathy through the PKD/Limk/Cofilin signaling pathway**. *Oxid Med Cell Longev* (2020.0) **2020** 1723423. DOI: 10.1155/2020/1723423
36. Wang M, Xie M, Yu S, Shang P, Zhang C, Han X. **Lipin1 alleviates autophagy disorder in sciatic nerve and improves Diabetic Peripheral Neuropathy**. *Mol Neurobiol* (2021.0) **58** 6049-61. DOI: 10.1007/s12035-021-02540-5
37. Chu CT. **Mechanisms of selective autophagy and mitophagy: implications for neurodegenerative diseases**. *Neurobiol Dis* (2019.0) **122** 23-34. DOI: 10.1016/j.nbd.2018.07.015
38. Priem D, van Loo G, Bertrand MJM. **A20 and cell death-driven inflammation**. *Trends Immunol* (2020.0) **41** 421-35. DOI: 10.1016/j.it.2020.03.001
39. 39.Martens A, van Loo G. A20 at the Crossroads of Cell Death, Inflammation, and Autoimmunity. Cold Spring Harb Perspect Biol. 2020;12(1).
40. Mooney EC, Sahingur SE. **The Ubiquitin System and A20: implications in Health and Disease**. *J Dent Res* (2021.0) **100** 10-20. DOI: 10.1177/0022034520949486
41. Zhang Q, Lenardo MJ, Baltimore D. **30 years of NF-kappaB: a blossoming of relevance to human pathobiology**. *Cell* (2017.0) **168** 37-57. DOI: 10.1016/j.cell.2016.12.012
42. Abbasi A, Forsberg K, Bischof F. **The role of the ubiquitin-editing enzyme A20 in diseases of the central nervous system and other pathological processes**. *Front Mol Neurosci* (2015.0) **8** 21. DOI: 10.3389/fnmol.2015.00021
43. Voet S, Mc Guire C, Hagemeyer N, Martens A, Schroeder A, Wieghofer P. **A20 critically controls microglia activation and inhibits inflammasome-dependent neuroinflammation**. *Nat Commun* (2018.0) **9** 2036. DOI: 10.1038/s41467-018-04376-5
44. Mohebiany AN, Ramphal NS, Karram K, Di Liberto G, Novkovic T, Klein M. **Microglial A20 protects the brain from CD8 T-Cell-mediated immunopathology**. *Cell Rep* (2020.0) **30** 1585-97. DOI: 10.1016/j.celrep.2019.12.097
45. Matsuzawa Y, Oshima S, Takahara M, Maeyashiki C, Nemoto Y, Kobayashi M. **TNFAIP3 promotes survival of CD4 T cells by restricting MTOR and promoting autophagy**. *Autophagy* (2015.0) **11** 1052-62. DOI: 10.1080/15548627.2015.1055439
46. Martens A, Priem D, Hoste E, Vetters J, Rennen S, Catrysse L. **Two distinct ubiquitin-binding motifs in A20 mediate its anti-inflammatory and cell-protective activities**. *Nat Immunol* (2020.0) **21** 381-7. DOI: 10.1038/s41590-020-0621-9
47. 47.Chen X, Kang R, Kroemer G, Tang D. Ferroptosis in infection, inflammation, and immunity. J Exp Med. 2021;218(6).
48. Li JY, Yao YM, Tian YP. **A trigger of Proinflammatory State Progression to Immunogenicity in Necroinflammatory Disease**. *Front Immunol* (2021.0) **12** 701163. DOI: 10.3389/fimmu.2021.701163
49. Reimao S, Ferreira S, Nunes RG, Pita Lobo P, Neutel D, Abreu D. **Magnetic resonance correlation of iron content with neuromelanin in the substantia nigra of early-stage Parkinson’s disease**. *Eur J Neurol* (2016.0) **23** 368-74. DOI: 10.1111/ene.12838
50. Perga S, Martire S, Montarolo F, Navone ND, Calvo A, Fuda G. **A20 in multiple sclerosis and Parkinson’s Disease: Clue to a common dysregulation of anti-inflammatory pathways?**. *Neurotox Res* (2017.0) **32** 1-7. DOI: 10.1007/s12640-017-9724-y
|
---
title: The association between malnutrition status and hemorrhagic transformation
in patients with acute ischemic stroke receiving intravenous thrombolysis
authors:
- Yerim Kim
- Minwoo Lee
- Hee Jung Mo
- Chulho Kim
- Jong-Hee Sohn
- Kyung-Ho Yu
- Sang-Hwa Lee
journal: BMC Neurology
year: 2023
pmcid: PMC10012700
doi: 10.1186/s12883-023-03152-3
license: CC BY 4.0
---
# The association between malnutrition status and hemorrhagic transformation in patients with acute ischemic stroke receiving intravenous thrombolysis
## Abstract
### Objectives
We evaluated the impact of malnutrition as estimated by the controlling nutritional status (CONUT) score and prognostic nutritional index (PNI) on hemorrhagic transformation (HT) and stroke outcomes after intravenous thrombolysis (IVT).
### Materials and methods
Using a multicenter registry database, we enrolled 808 patients with acute ischemic stroke who received IVT between August 2013 and May 2021. We defined malnutrition as a CONUT score ≥ 2 and low PNI. The primary outcome measure was the occurrence of symptomatic HT contributing to early neurologic deterioration (END-SHT) after IVT. Multivariable analysis was performed to analyze the association between CONUT score, PNI, and END-SHT after IVT.
### Results
The rate of END-SHT was higher with increasing CONUT scores and PNI values. In the multivariable analysis, CONUT score ≥ 5 and low PNI were significantly associated with END-SHT (odds ratio [$95\%$ confidence interval], CONUT score ≥ 5: 12.23 [2.41–62.07], $$p \leq 0.003$$; low PNI: 4.98 [1.76–14.09], $$p \leq 0.003$$). The receiver operating characteristic curve showed that both the CONUT score and PNI had good predictive ability. The cutoff values for CONUT and PNI were 5 and 42.3, respectively, for END-SHT.
### Conclusion
Malnutrition, as denoted by a higher CONUT score and lower PNI, was associated with END-SHT. The joint application of both nutritional markers could be useful in predicting END-SHT after IVT.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12883-023-03152-3.
## Introduction
The prevalence of malnutrition in acute stroke has been estimated to range from 12.2–$62\%$ [1, 2]. It is generally accepted that malnutrition affecting ischemic stroke patients have been negatively associated with several clinical outcomes [3–9]. However, the prevalence of malnutrition after stroke varies widely, due to heterogenous study subjects, stroke subtypes, varying nutritional parameters, and differing definitions of malnutrition. Nonetheless, since early nutritional assessment could improve clinical outcomes in acute ischemic stroke,[10] its importance particularly increases during the early stages of stroke.
Nutritional interventions for acute ischemic stroke have received little attention in real-world practice. The difficulties in assessing nutritional status in stroke patients could explain this situation. Weight and dietary history may not be captured from patients and their kin; simple assessments such as body mass index (BMI) were not always available in immobile stroke patients [2]. Notably, objective nutritional markers, such as the controlling nutritional status (CONUT) score and the prognostic nutritional index (PNI), were easy to calculate using serum albumin, cholesterol levels, and lymphocyte count in peripheral blood and could be feasible in clinical practice [11].
Given that the CONUT score and PNI represent nutrition and immune response, we assumed that the scores at hospitalization could predict the outcomes of stroke patients receiving intravenous thrombolysis (IVT). Since the CONUT score and PNI at hospitalization were less likely to be affected by neurological disability, the score could represent pre-thrombolysis nutritional status [12, 13]. However, only a few studies have evaluated the impact of nutritional status on stroke outcomes after IVT [13]. Using a multicenter database, we aimed to investigate the impact of the CONUT score and PNI on hemorrhagic transformation (HT) and stroke outcomes in patients receiving IVT.
## Subjects
We consecutively registered all patients with acute ischemic stroke between August 2013 and May 2021 at four university-affiliated institutions (Hallym Sacred Heart Hospital, Chuncheon Sacred Heart Hospital, Kangdong Sacred Heart Hospital, and Dongtan Sacred Heart Hospital). In this study, we identified patients with acute ischemic stroke who were treated with IVT. All included patients received intravenous recombinant tissue plasminogen activator (tPA) according to the current guidelines for acute stroke management. According to the imaging protocols of each institution, we routinely performed follow-up brain computed tomography (CT) or magnetic resonance imaging (MRI) within 24 h after IVT. Additionally, these imaging tools were used when patients experienced neurological deterioration. We excluded the following patients: [1] patients with unavailable CONUT score and PNI; [2] patients without follow-up brain CT or MRI within 24 h of stroke onset; [3] patients who were not available for assessment of early neurological deterioration (END), HT after IVT, and the modified Rankin Scale (mRS) at 3 months; [4] patients with a pre-stroke mRS score ≥ 2; [5] patients who had additional endovascular interventional therapy after IVT and [6] patients who had active malignancy, infection/inflammation, hematologic disease, or infusion of blood products within 24 h after IVT.
## Clinical Data collection and definition of parameters
The following data were directly obtained from the registry database: [1] demographics, including age and sex; [2] medical history, including stroke risk factors, prior stroke, hypertension, diabetes mellitus (DM), coronary artery disease, hyperlipidemia, atrial fibrillation, current smoking status, pre-stroke status, and prior use of statins and antithrombotic drugs; [3] stroke characteristics, acute stroke treatment, initial NIHSS score, and ischemic stroke mechanism according to the Trial of Org 10,172 in Acute Stroke Treatment (TOAST) classification with some modifications,[14] tPA dose, and reperfusion therapy (IVT and intra-arterial thrombectomy [IAT]); and [4] laboratory data sampled within 12 h of stroke onset (median time of interval from stroke onset to sample acquisition 2.9 h [interquartile range 1.3-6.0]).
## Malnutrition tools
The CONUT score was estimated using the serum albumin level, lymphocyte count, and total cholesterol level. The detailed parameters and scoring system for the CONUT score are described in Supplementary Table 1. According to the traditional stratification of the CONUT scoring system, a CONUT score of 2–4 was mild malnutrition, 5–8 was moderate malnutrition, and 9–12 was severe malnutrition [11]. The PNI score was estimated using the formula: 5x lymphocyte count (109/L) + 10 × serum albumin (g/dL).
## Outcome measures
END was defined as an increment of at least 1 point in motor power or a total NIHSS score deterioration of ≥ 2 points within 7 days of admission compared to the initial NIHSS score [15]. We categorized the etiology of END as follows: [1] symptomatic hemorrhagic transformation (END-SHT) and [2] stroke progression (END-prog) after IVT [16]. With respect to the definition of END, the primary outcome measures were the occurrence of symptomatic hemorrhagic transformation contributing to early neurologic deterioration (END-SHT) and any HT after IVT. END-SHT was defined as the presence of HT, which was defined using the European Cooperative Acute Stroke *Study criteria* based on follow-up brain CT or MR images [17]. Any HT was defined as a new bleeding lesion on follow-up images after IVT. Secondary outcome measures were stroke progression (END-prog) and good functional outcome at 3 months (mRS 0 to 2) [18]. END-prog was defined as END caused by the progression of the initial ischemic lesion, which occurred due to the enlargement of infarct size or the presence of significant perilesional edema on follow-up images [15]. Two vascular neurologists (M Lee and S-H Lee) reviewed the END data to confirm END-SHT and END-prog in a double-blinded manner (interclass correlation coefficient, 0.88; $p \leq 0.001$ in END-SHT and 0.89; $p \leq 0.001$ in END-prog).
## Statistical analysis
We hypothesized that malnutrition (high CONUT score and low PNI) would increase the risk of END-SHT and poor functional outcomes after IVT. With respect to the primary and secondary outcome measures, the categorized CONUT score (normal, mild, and moderate to severe) and PNI were compared using Pearson’s chi-squared test for categorical variables and Student’s t-test or Mann-Whitney U test for continuous variables. We dichotomized PNI as low and high groups based on a cut-off value calculated by receiver operating characteristic (ROC) curve analysis. We performed a binary logistic regression analysis to evaluate the independent effects of the categorized CONUT score on stroke outcomes using clinically plausible variables or univariate $p \leq 0.10.$ Crude and adjusted odds ratios (ORs) and $95\%$ confidence intervals (CIs) were calculated.
We performed a ROC curve to determine the predictive ability and optimal cutoff value of the CONUT score and PNI on END-SHT, using the ‘pROC’ package of R. the Youden index estimated the cut-off values of the CONUT score and PNI for END-SHT. Statistical analyses were performed using IBM SPSS version 21.0 software (IBM Corporation, Armonk, NY, USA) and R version 4.0.3 (R core Team 2020; R Foundation for Statistical Computing, Vienna, Austria).
## Results
Among 10,808 consecutive patients with acute ischemic stroke, 1,259 underwent IVT and IAT. Of the 1,259 patients, 808 who received IVT alone were included in our study. The mean age of these patients was 66.7 (± 13.3) years, 513 ($63.5\%$) were male, the median CONUT score was 1 (0–2), and the median PNI was 52.5 (47.1–57.5). With the respect to the CONUT scoring system, the proportion of malnutrition in acute ischemic stroke patients was $39.6\%$ (mild, $34.2\%$; moderate to severe, $5.4\%$). We established a PNI value higher than 42.3 as the optimal cut-off point for END-SHT. Based on this criteria, patients were divided into low and high PNI groups. According to the CONUT scoring system, patients with moderate-to-severe malnutrition were likely to be older and had higher stroke severity, previous stroke, and DM history. Interestingly, the groups with moderate-to-severe malnutrition, whether through the CONUT or PNI scoring system, had more large artery atherosclerosis (LAA) and cardioembolism stroke subtypes and fewer small vessel occlusion (SVO) stroke subtypes in our study. The baseline demographic and clinical characteristics are described in Tables 1 and 2, respectively.
Table 1Baseline characteristics according to CONUT scoring systemNormal(CONUT 0–1)($$n = 488$$)Mild(CONUT 2–4)($$n = 276$$)Moderate to severe(CONUT 5–12)($$n = 44$$)p-valueAge (SD)65.2 (12.7)68.6 (13.8)71.3 (13.8)< 0.001Male, n (%)311 (63.7)178 (64.5)24 (54.5)0.44NIHSS (IQR)7 (5–11)8 (5–14)8 (6–13)0.02BMI, kg/m2 (SD)24.0 (22.0-25.7)23.7 (21.2–25.6)22.3 (20.8–24.6)0.01Stroke mechanism, n (%)0.04SVO89 (18.2)29 (10.5)3 (6.8)LAA146 (29.9)81 (29.3)14 (31.8)CE132 (27.0)95 (34.4)126 (36.4)UD106 (21.7)63 (22.8)11 (25.0)OD15 (3.1)8 (2.9)0 (0.0)Interval from arrival to IVT, min (IQR)39 (28–54)40 (31–55)44 (36-55.5)0.22Previous stroke, n (%)73 (15.0)52 (18.8)14 (31.8)0.01HTN, n (%)292 (59.8)163 (59.1)25 (56.8)0.92DM, n (%)112 (23.0)101 (36.6)15 (34.1)< 0.001HL, n (%)92 (18.9)53 (19.2)10 (22.7)0.83CAD, n (%)28 (5.3)44 (15.9)5 (11.4)< 0.001Current smoking, n (%)113 (23.2)52 (18.8)8 (18.2)0.33Atrial fibrillation, n (%)132 (27.0)94 (34.1)18 (40.9)0.04Previous use of antithrombotics, n (%)116 (23.8)93 (33.7)11 (25.0)0.01tPA dose, n (%)0.0020.6 mg/kg73 (15.0)17 (6.2)4 (9.1)0.9 mg/kg415 (85.0)259 (93.8)40 (90.9)Total cholesterol, mg/dL (SD)190.8 (35.5)134.9 (28.5)112.3 (14.4)< 0.001Albumin, mg/dL (SD)4.2 (0.4)4.1 (0.4)2.8 (0.8)< 0.001Hemoglobin, g/dL (SD)13.9 (2.0)13.5 (1.9)13.2 (2.1)0.001Creatinine, mg/dL (SD)0.93 (0.32)1.11 (0.79)1.10 (0.49)< 0.001Platelet, x1000/µL (SD)235.7 (68.9)215.9 (70.6)221.1 (77.0)0.001LDL, mg/dL (SD)118.1 (30.2)85.4 (29.8)97.8 (41.8)< 0.001HbA1c, % (SD)6.1(1.2)6.1 (1.0)6.0 (1.2)0.37Prothrombin time, INR (SD)1.01 (0.11)1.04 (0.14)1.04 (0.17)0.01CRP, mg/dL (SD)6.8 (14.1)8.3 (15.6)12.8 (39.7)0.054Initial random glucose, mg/dL (SD)142.4 (53.0)147.9 (64.1)136.6 (61.5)0.30SBP, mmHg (SD)155.1 (28.1)148.6 (25.5)156.4 (25.5)0.01PNI, (IQR)54.6 (50.6–59.5)48.8 (42.2–54.0)31.6 (29.9–37.4)< 0.001Abbreviation: CONUT, controlling nutritional status; SD, standard deviation; NIHSS, National Institute Health of Stroke Scale; BMI, body mass index; IQR, interquartile range; SVO, small vessel occlusion; LAA, large artery atherosclerosis; CE, cardioembolism; UD, undetermined; OD, other determined; IVT, intravenous thrombolysis; HTN, hypertension; DM, diabetes mellitus; HL, hyperlipidemia; CAD, coronary artery disease; tPA, tissue plasminogen activator; LDL, low density lipoprotein; HbA1c, glycated hemoglobin; INR, international normalized ratio; CRP, C reactive protein; SBP, systolic blood pressure; PNI, prognostic nutritional index Table 2Baseline characteristics according to PNI.low PNI ≤ 42.3($$n = 116$$)high PNI > 42.3($$n = 692$$)p-valueMale, n (%)68 (58.6)445 (64.3)0.25NIHSS (IQR)10 (6–16)7 (5–12)< 0.001BMI, kg/m2 (SD)23.4 (21.4–24.8)23.9 (21.8–25.7)0.04SVO6 (5.2)115 (16.6)LAA41 (35.3)200 (28.9)CE38 (32.8)205 (29.6)UD28 (24.1)152 (22.0)OD3 (2.8)20 (2.9)Interval from arrival to IVT, min (IQR)42 (35–56)39 (29–54)0.09Previous stroke, n (%)23 (19.8)116 (16.8)0.43HTN, n (%)68 (58.6)412 (59.5)0.92CAD, n (%)14 (12.1)61 (8.8)0.30Current smoking, n (%)22 (19.0)151 (21.8)0.54Atrial fibrillation, n (%)44 (37.9)200 (28.9)0.06Previous use of antithrombotics, n (%)25 (21.6)195 (28.2)0.15tPA dose, n (%)0.060.6 mg/kg7 (6.0)87 (12.6)0.9 mg/kg109 (94.0)605 (87.4)Total cholesterol, mg/dL (SD)151.0 (44.9)173.1 (42.3)0.21Albumin, mg/dL (SD)3.0 (0.7)4.2 (0.4)< 0.001Hemoglobin, g/dL (SD)13.5 (2.0)13.8 (2.0)0.16Creatinine, mg/dL (SD)1.16 (0.88)0.97 (0.46)0.002Platelet, x1000/µL (SD)222.3 (73.1)229.1 (70.1)0.32LDL, mg/dL (SD)103.5 (38.4)106.2 (33.7)0.03HbA1c, % (SD)6.0 (2.0)6.1 (1.1)0.27Prothrombin time, INR (SD)1.04 (0.16)1.02 (0.12)0.35CRP, mg/dL (SD)11.3 (27.8)7.0 (14.4)0.001Initial random glucose, mg/dL (SD)151.5 (80.9)142.7 (52.6)0.004SBP, mmHg (SD)153.3 (24.7)152.9 (27.7)0.16Abbreviation: PNI, prognostic nutritional index; SD, standard deviation; NIHSS, National Institute Health of Stroke Scale; BMI, body mass index; IQR, interquartile range; SVO, small vessel occlusion; LAA, large artery atherosclerosis; CE, cardioembolism; UD, undetermined; OD, other determined; IVT, intravenous thrombolysis; HTN, hypertension; DM, diabetes mellitus; HL, hyperlipidemia; CAD, coronary artery disease; tPA, tissue plasminogen activator; LDL, low density lipoprotein; HbA1c, glycated hemoglobin; INR, international normalized ratio; CRP, C reactive protein; SBP, systolic blood pressure; For the main outcome, the rates of END-SHT and HT were higher with increasing CONUT scores (END-SHT: normal; 21 ($4.3\%$), mild; 9 ($3.3\%$) and moderate to severe; 32 ($72.7\%$), any HT: normal; 46 ($9.4\%$), mild; 28 ($10.1\%$) and moderate to severe; 32 ($72.7\%$), p for trend < 0.001 Fig. 1). The low PNI group also had higher rates of END-SHT and HT than the high PNI group (END-SHT: 33 ($28.4\%$) versus 29 ($4.2\%$); any HT: 33 ($28.4\%$) versus 73 ($10.5\%$); $p \leq 0.001$; Fig. 2). For the secondary outcomes, the rates of END-prog [normal; 28 ($5.7\%$), mild; 6 ($2.2\%$) and moderate to severe; 5 ($11.4\%$)] and poor functional outcome at 3 months also increased with higher CONUT scores [normal; 166 ($34.0\%$), mild; 123 ($44.6\%$) and moderate to severe; 27 ($61.4\%$)] (Fig. 1). Similarly, the low PNI group had a higher rate of poor functional outcomes at 3 months than the high PNI group [70 ($60.3\%$) versus 246 ($35.5\%$), $p \leq 0.001$, Fig. 2].
Fig. 1The distribution of stroke outcomes according to CONUS scoreAbbreviations: CONUT, controlling nutritional status; END-SHT, symptomatic hemorrhagic transformation; HT, hemorrhagic transformation; END-prog, stroke progression; mRS, modified Rankin Scale.
Fig. 2The distribution of stroke outcomes according to PNI scoreAbbreviations: PNI, prognostic nutritional index; END-SHT, symptomatic hemorrhagic transformation; HT, hemorrhagic transformation; END-prog, stroke progression; mRS, modified Rankin Scale.
In multivariable analysis, moderate to severe nutritional status based on the CONUT scoring system was significantly associated with END-SHT and HT (OR [$95\%$ CI], END-SHT: 14.52 [3.15–66.95], $$p \leq 0.001$$; HT: 9.40 [2.74–32.20], $p \leq 0.001$). A low PNI was also significantly associated with END-SHT and HT (OR [$95\%$ CI], END-SHT: 4.90 [1.72–13.92], $$p \leq 0.003$$; HT: 3.40 [1.46–7.89], $$p \leq 0.004$$). For secondary outcomes, moderate-to-severe malnutrition and low PNI increased the risk of END-prog and poor functional outcomes at 3 months (Tables 3 and 4). In sensitivity analysis, both CONUT and PNI were associated with 3-month mortality (Supplementary Tables 2 and 3).
Table 3Multivariate analysis showing association between CONUT scoring system and outcomesEND-SHTAny HTEND-prog3-month mRS 3–6OR$95\%$ CIp-valueOR$95\%$ CIp-valueOR$95\%$ CIp-valueOR$95\%$ CIp-valueCONUT scoring systemnormalrefrefrefrefmild0.570.20–1.620.300.850.43–1.680.640.340.12–1.030.0571.270.78–2.050.34moderate to severe12.232.41–62.070.0037.902.20–28.400.0023.011.51–17.790.033.101.12–8.550.03Age1.020.98–1.050.341.010.99–1.030.431.030.99–1.060.191.051.03–1.07< 0.001Male0.870.39–1.980.751.220.69–2.170.491.280.57–2.840.550.770.51–1.120.16NIHSS1.020.95–1.090.661.010.97–1.060.571.050.99–1.120.081.161.12–1.20< 0.001Stroke mechanismSVOrefrefrefrefLAA0.980.26–3.640.981.170.46–2.990.741.740.53–5.790.361.670.94–2.990.08CE0.700.14–3.570.671.830.58–5.750.31.190.18–8.140.860.950.40–2.230.91UD0.980.25–3.890.971.550.60–4.020.370.350.07–1.870.221.130.60–2.130.70OD1.270.11–14.260.850.790.09–7.050.831.120.14–16.240.681.840.62–5.450.27Previous stroke0.680.24–1.960.480.980.50–1.910.940.370.11–1.240.111.450.89–2.380.14DM0.410.16–1.050.060.790.45–1.390.412.431.17–5.030.021.671313 − 2.480.01CAD1.790.54–5.910.341.470.60–3.570.400.670.13–3.510.631.60.83–3.060.16Atrial fibrillation2.040.57–7.280.271.010.54–1.870.981.550.31–7.710.601.530.75–3.110.24Previous use of antithrombotics1.220.49–3.060.671.010.54–1.870.980.620.24–1.580.310.460.29–0.740.001tPA dose,1.360.37–4.980.653.301.06–10.260.042.100.36–5.750.48-0.950.55–1.640.84Total cholesterol0.980.96–0.9960.010.980.97–0.99< 0.0011.000.99–1.020.701.000.99–1.010.91Albumin0.460.19–1.110.080.900.49–1.650.731.570.66–3.730.311.120.73–1.730.6Hemoglobin1.230.98–1.550.081.181.01–1.390.040.820.66–1.020.071.040.94–1.160.44Platelet1.000.995–1.010.921.000.996-1.000.801.000.995–1.010.981.000.998-1.000.86Creatinine1.370.73–2.580.331.040.65–1.660.890.730.30–1.800.491.611.12–2.310.01LDL1.021.01–1.040.011.021.01–1.04< 0.0011.000.98–1.020.971.010.998–1.020.14PT3.540.49–25.120.210.980016-6.060.981.220.10-15.320.880.750.19–2.870.67CRP1.010.996–1.030.151.011.00-1.030.0451.000.99–1.020.661.010.998–1.020.11Abbreviation: CONUT, controlling nutritional status; END-SHT, symptomatic hemorrhagic transformation; HT, hemorrhagic transformation; END-prog; stroke progression; mRS, modified Rankin Scale; OR, odds ratio; CI, confidence interval; NIHSS, National Institute Health of Stroke Scale; SVO, small vessel occlusion; LAA, large artery atherosclerosis; CE, cardioembolism; UD, undetermined; OD, other determined; DM, diabetes mellitus; HL, hyperlipidemia; CAD, coronary artery disease; tPA, tissue plasminogen activator; LDL, low density lipoprotein; PT, prothrombin time; CRP, C-reactive protein Table 4Multivariate analysis showing association between PNI system and outcomesEND-SHTAny HTEND-prog3-month mRS 3–6OR$95\%$ CIp-valueOR$95\%$ CIp-valueOR$95\%$ CIp-valueOR$95\%$ CIp-valueLow PNI4.981.76–14.090.0033.481.50–9.090.0043.521.02–12.190.0472.951.35–6.460.01Age0.9950.97–1.030.920.9990.98–1.020.961.030.99–1.060.111.051.03–1.06< 0.001Male1.140.57–2.290.721.480.89–2.450.130.850.41–1.750.650.790.55–1.130.19NIHSS0.980.92–1.040.430.990.95–1.030.641.040.98–1.100.171.141.10–1.09< 0.001Stroke mechanismSVOrefrefrefrefLAA0.970.31–3.020.961.310.58–3.100.531.440.44–4.740.551.620.92–2.880.1CE0.680.15–3.040.621.890.64–5.600.250.740.12–4.430.740.890.39–2.060.79UD1.160.36–3.720.801.730.72–4.150.220.320.07–1.580.161.180.64–2.170.6OD0.390.04–4.170.440.360.04–3.360.370.470.11–5.210.471.540.52–4.600.44Interval from arrival to IVT1.010.998–1.010.161.0010.995–1.010.701.000.995–1.010.391.000.997–1.010.52Atrial fibrillation2.190.68–7.020.191.520.65–3.550.341.830.43–7.870.421.330.68–2.630.41tPA dose1.340.42–4.340.622.71.02–7.160.0461.120.87–8.260.510.930.54–1.580.78Albumin0.230.11–0.51< 0.0010.450.25–0.780.011.580.72–3.460.261.210.79–1.840.39Creatinine0.990.60–1.640.970.880.59–1.330.560.710.30–1.650.421.491.05–2.120.03LDL1.010.996–1.010.311.000.996–1.010.431.010.995–1.020.321.000.999–1.010.15CRP1.010.99–1.020.291.010.997–1.020.151.000.99–1.020.641.010.996–1.020.28IRG1.000.996–1.010.681.000.998–1.010.311.000.99–1.010.731.001.00-1.010.10Abbreviation: PNI, prognostic nutritional index; END-SHT, symptomatic hemorrhagic transformation; HT, hemorrhagic transformation; END-prog; stroke progression; mRS, modified Rankin Scale; OR, odds ratio; CI, confidence interval; NIHSS, National Institute Health of Stroke Scale; SVO, small vessel occlusion; LAA, large artery atherosclerosis; CE, cardioembolism; UD, undetermined; OD, other determined; IVT, intravenous thrombolysis; tPA, tissue plasminogen activator; LDL, low density lipoprotein; CRP, C-reactive protein; IRG, initial random glucose The ROC curve showed that the predictive ability of CONUT score and PNI for END-SHT was close to good (AUC of CONUT: 0.74, $95\%$ CI [0.66–0.82]. $p \leq 0.001$; AUC of PNI: 0.74, $95\%$ CI [0.66–0.81], $p \leq 0.001$). There were no significant differences in the prediction of END-SHT between CONUT and PNI levels. The cutoff values of the CONUT score and PNI were 5 and 42.3, respectively, for END-SHT (Fig. 3).
Fig. 3ROC curve showing the predictive ability of CONUT score and PNI for END-SHTAbbreviations: ROC, receiver operating characteristics; CONUT, controlling nutritional status; PNI, prognostic nutritional index; END-SHT, symptomatic hemorrhagic transformation
## Discussion
The main findings of this study are as follows: [1] Patients with worse malnutrition status, as estimated by both CONUT score and PNI, had a higher occurrence of END-SHT and worse stroke outcomes in acute ischemic stroke patients after IVT; [2] In multivariable analysis, malnutrition was associated with increasing the risk of END-SHT and poor stroke outcomes after IVT; and [3] the predictive ability of CONUT and PNI scores for END-SHT was reliable, and we suggest the cutoff values (CONUT score: 5; PNI: 42.3) of both nutritional markers for predicting END-SHT. CONUT score and PNI were initially reported to be effective predictors of nutritional status in malignancy, cardiovascular disease, and heart failure [19–23]. Recently, both markers have been suggested to be useful for predicting outcomes after acute ischemic stroke [24].
The frequency of malnutrition was $39.6\%$ in this study, consistent with that of previous studies [13, 25, 26]. Although malnutrition is not a rare condition in acute ischemic stroke, clinicians may lack knowledge about the importance of nutritional status in stroke management [27]. The recent guidelines for acute ischemic stroke recommended that all stroke patients should be evaluated for individual baseline nutritional status, and malnutrition should be corrected as soon as possible [28]. The increased workload of evaluating nutritional markers in clinical practice could lead to this phenomenon. Hence, the easy estimation of nutritional markers using the CONUT score and the PNI could be suitable and feasible options in acute stroke settings. Several previous studies on this issue revealed that malnutrition, as estimated by the CONUT score and PNI, was associated with short- and long-term stroke prognosis,[7, 24, 29–31] but the prognosis after IVT remains unclear. Based on the results of a previous study that malnutrition estimated by PNI could increase poor stroke outcomes after IVT,[13] our main novel finding is that both CONUT score and PNI could be used to predict bleeding risk after IVT.
Inflammatory reactions and oxidative stress after ischemic stroke play a major role in stroke pathophysiology. Especially in IVT-treated stroke patients, these pathophysiologic reactions could increase bleeding tendency and infarct volume through blood-brain barrier breakdown and aggravating microangiopathy [32]. Serum albumin, as a multifunctional protein, plays neuroprotective roles in ischemic stroke by reducing erythrocyte aggregation and exhibiting an antioxidant effect [33]. In the inflammatory reaction, lymphocyte infiltration in the ischemic area could release pro-inflammatory cytokines and cytotoxic substances. Lower lymphocyte levels have been associated with poor functional outcomes after acute ischemic stroke [24, 34]. Although the association between total cholesterol and stroke outcomes remains unclear, low total cholesterol generally plays a role in promoting endothelial injury via arterial medial layer smooth muscle cell necrosis [35]. Therefore, we suggest that the detrimental pathophysiologic process in malnutrition could enhance the bleeding risk and poor stroke outcomes after IVT. Among the several reported nutritional markers, CONUT score and PNI reflecting nutrition and inflammation in subjects could be more reasonable nutritional markers for predicting HT after IVT.
Notably, the ROC curve showed that both CONUT score and PNI were useful for predicting HT after IVT and revealed the suggested cutoff values for both markers in our study. The optimal cut-off value of the CONUT score for predicting HT after IVT was 5. Previous studies showed that a CONUT score ≥ 5 (moderate to severe malnutrition state) could be associated with a poor 3-month functional outcome and mortality [24, 29]. Meanwhile, the optimal cutoff value of PNI for predicting HT after IVT was 42.3, which is reasonable compared to a previous study with an optimal cutoff value of 44.2 for 3-month outcomes after IVT [13]. Our novel findings showed that the different predictive values of CONUT score and PNI highlight the need for constantly updating the IVT strategy and guidelines.
The secondary outcomes were partly consistent with those of prior studies on CONUT score and PNI. Interestingly, malnutrition was associated with early stroke progression after IVT in the present study. The higher frequency of comorbidities, higher stroke severity, and unfavorable laboratory findings in the malnourished subjects in our study could explain this phenomenon. Notably, malnourished subjects had more LAA and CE and fewer SVO stroke subtypes, which was also found in a previous study [13]. Stroke progression contributing to END after IVT in LAA and CE stroke subtypes may be related to failed reperfusion, thrombus migration, re-occlusion, and recurrent embolism [36, 37]. Hence, the discrepancy of stroke subtypes according to the nutritional state could also affect stroke progression. Previous studies with general stroke populations and IVT-treated subjects using a single parameter, such as BMI, showed contradictory results between obesity and stroke outcomes (obesity paradox) [38, 39]. However, previous studies using nutritional parameters addressed this issue and revealed relatively consistent results, especially in IVT-treated subjects. We suggest that parameters reflecting multifactorial pathophysiologic processes could be more reliable compared to single parameters in assessing stroke outcomes in stroke patients. Nonetheless, further studies are warranted to explore the association between malnutrition and stroke outcomes.
Despite its consecutive multicenter registry-based nature, this study has several limitations. First, we retrospectively collected clinical data from the registry database despite the relatively large sample size. Second, the CONUT score and PNI were not evaluated at discharge or the delayed stroke stage. In addition, nutritional information such as dietary intake, weight change during hospitalization, and weight change after rehabilitation were not available in our study. However, since the main aim of this study was to evaluate acute stroke outcomes at hospitalization, serial changes in nutritional status were inevitable in our study. Third, the relatively small sample size of subjects with moderate-to-severe malnutrition could hinder the generalization of our results. However, our crude data suggest an epidemiological prevalence of severely malnourished subjects treated with IVT in future studies. Fourth, most of the variables were badly biased toward moderate-to-severe malnutrition. Although we adjusted these variables in the multivariable model, balancing the variables using propensity score matching may be needed in our analysis. However, with our worse clinical findings in malnourished subjects, clinicians may be more interested in establishing stroke interventions for those populations in clinical practice. Lastly, although we adjusted for several variables that may affect the outcomes, unmeasured confounding factors (e.g., muscle mass, nitrogen balance, and body composition) could hinder the applicability of our main findings.
## Conclusion
We suggest that nutritional status, as assessed by the CONUT score and PNI, could be associated with HT and poor stroke outcomes after IVT. The joint use of both markers, which are easily tested in acute stroke settings, could reasonably predict HT after IVT. Further prospective studies with larger sample sizes are needed to address the practical application of the CONUT score and PNI in clinical settings.
## Electronic supplementary material
Below is the link to the electronic supplementary material.
Supplementary Material 1
## References
1. Foley NC, Salter KL, Robertson J, Teasell RW, Woodbury MG. **Which reported estimate of the prevalence of malnutrition after stroke is valid?**. *Stroke* (2009.0) **40** e66-74. DOI: 10.1161/STROKEAHA.108.518910
2. Yoo SH, Kim JS, Kwon SU, Yun SC, Koh JY, Kang DW. **Undernutrition as a predictor of poor clinical outcomes in acute ischemic stroke patients**. *Arch Neurol* (2008.0) **65** 39-43. DOI: 10.1001/archneurol.2007.12
3. Collaboration FT. **Poor nutritional status on admission predicts poor outcomes after stroke: observational data from the FOOD trial**. *Stroke* (2003.0) **34** 1450-6. DOI: 10.1161/01.STR.0000074037.49197.8C
4. Davalos A, Ricart W, Gonzalez-Huix F, Soler S, Marrugat J, Molins A. **Effect of malnutrition after acute stroke on clinical outcome**. *Stroke* (1996.0) **27** 1028-32. DOI: 10.1161/01.str.27.6.1028
5. Davis JP, Wong AA, Schluter PJ, Henderson RD, O’Sullivan JD, Read SJ. **Impact of premorbid undernutrition on outcome in stroke patients**. *Stroke* (2004.0) **35** 1930-4. DOI: 10.1161/01.STR.0000135227.10451.c9
6. Martineau J, Bauer JD, Isenring E, Cohen S. **Malnutrition determined by the patient-generated subjective global assessment is associated with poor outcomes in acute stroke patients**. *Clin Nutr* (2005.0) **24** 1073-7. DOI: 10.1016/j.clnu.2005.08.010
7. Naito H, Hosomi N, Nezu T, Kuzume D, Aoki S, Morimoto Y. **Prognostic role of the controlling nutritional status score in acute ischemic stroke among stroke subtypes**. *J Neurol Sci* (2020.0) **416** 116984. DOI: 10.1016/j.jns.2020.116984
8. Kang MK, Kim TJ, Kim Y, Nam KW, Jeong HY, Kim SK. **Geriatric nutritional risk index predicts poor outcomes in patients with acute ischemic stroke - automated undernutrition screen tool**. *PLoS ONE* (2020.0) **15** e0228738. DOI: 10.1371/journal.pone.0228738
9. Maruyama K, Nakagawa N, Koyama S, Maruyama JI, Hasebe N. **Malnutrition increases the incidence of Death, Cardiovascular events, and infections in patients with stroke after Rehabilitation**. *J Stroke Cerebrovasc Dis* (2018.0) **27** 716-23. DOI: 10.1016/j.jstrokecerebrovasdis.2017.10.002
10. Nishioka S, Wakabayashi H, Nishioka E, Yoshida T, Mori N, Watanabe R. **Nutritional improvement correlates with recovery of activities of Daily Living among malnourished Elderly Stroke Patients in the Convalescent Stage: a cross-sectional study**. *J Acad Nutr Diet* (2016.0) **116** 837-43. DOI: 10.1016/j.jand.2015.09.014
11. Ignacio de Ulibarri J, Gonzalez-Madrono A, de Villar NG, Gonzalez P, Gonzalez B, Mancha A. **CONUT: a tool for controlling nutritional status. First validation in a hospital population**. *Nutr Hosp* (2005.0) **20** 38-45. PMID: 15762418
12. Tang H, Gong F, Guo H, Dai Z, Wang J, Liu B. **Malnutrition and risk of mortality in ischemic stroke patients treated with intravenous thrombolysis**. *Front Aging Neurosci* (2022.0) **14** 834973. DOI: 10.3389/fnagi.2022.834973
13. Xiang W, Chen X, Ye W, Li J, Zhang X, Xie D. **Prognostic Nutritional Index for Predicting 3-Month Outcomes in ischemic stroke patients undergoing thrombolysis**. *Front Neurol* (2020.0) **11** 599. DOI: 10.3389/fneur.2020.00599
14. Ko Y, Lee S, Chung JW, Han MK, Park JM, Kang K. **MRI-based Algorithm for Acute ischemic stroke subtype classification**. *J Stroke* (2014.0) **16** 161-72. DOI: 10.5853/jos.2014.16.3.161
15. Kim JT, Lee JS, Kim BJ, Park JM, Kang K, Lee SJ. **Frequency, management, and outcomes of early neurologic deterioration due to stroke progression or recurrence**. *J Stroke Cerebrovasc Dis* (2023.0) **32** 106940. DOI: 10.1016/j.jstrokecerebrovasdis.2022.106940
16. Park TH, Lee JK, Park MS, Park SS, Hong KS, Ryu WS. **Neurologic deterioration in patients with acute ischemic stroke or transient ischemic attack**. *Neurology* (2020.0) **95** e2178-e91. DOI: 10.1212/WNL.0000000000010603
17. Yaghi S, Willey JZ, Cucchiara B, Goldstein JN, Gonzales NR, Khatri P. **Treatment and outcome of Hemorrhagic Transformation after Intravenous Alteplase in Acute Ischemic Stroke: A Scientific Statement for Healthcare Professionals from the American Heart Association/American Stroke Association**. *Stroke* (2017.0) **48** e343-e61. DOI: 10.1161/STR.0000000000000152
18. Jeong HG, Kim BJ, Yang MH, Han MK, Bae HJ. **Neuroimaging markers for early neurologic deterioration in single small subcortical infarction**. *Stroke* (2015.0) **46** 687-91. DOI: 10.1161/STROKEAHA.114.007466
19. Iwakami N, Nagai T, Furukawa TA, Sugano Y, Honda S, Okada A. **Prognostic value of malnutrition assessed by Controlling Nutritional Status score for long-term mortality in patients with acute heart failure**. *Int J Cardiol* (2017.0) **230** 529-36. DOI: 10.1016/j.ijcard.2016.12.064
20. Iseki Y, Shibutani M, Maeda K, Nagahara H, Ohtani H, Sugano K. **Impact of the Preoperative Controlling Nutritional Status (CONUT) score on the survival after curative surgery for Colorectal Cancer**. *PLoS ONE* (2015.0) **10** e0132488. DOI: 10.1371/journal.pone.0132488
21. Mohri T, Mohri Y, Shigemori T, Takeuchi K, Itoh Y, Kato T. **Impact of prognostic nutritional index on long-term outcomes in patients with breast cancer**. *World J Surg Oncol* (2016.0) **14** 170. DOI: 10.1186/s12957-016-0920-7
22. 22.Cheng YL, Sung SH, Cheng HM, Hsu PF, Guo CY, Yu WC, et al. Prognostic Nutritional Index and the risk of mortality in patients with Acute Heart failure. J Am Heart Assoc. 2017;6(6). 10.1161/JAHA.116.004876.
23. Chen QJ, Qu HJ, Li DZ, Li XM, Zhu JJ, Xiang Y. **Prognostic nutritional index predicts clinical outcome in patients with acute ST-segment elevation myocardial infarction undergoing primary percutaneous coronary intervention**. *Sci Rep* (2017.0) **7** 3285. DOI: 10.1038/s41598-017-03364-x
24. Naito H, Nezu T, Hosomi N, Aoki S, Kinoshita N, Kuga J. **Controlling nutritional status score for predicting 3-mo functional outcome in acute ischemic stroke**. *Nutrition* (2018.0) **55–56** 1-6. DOI: 10.1016/j.nut.2018.02.025
25. Unosson M, Ek AC, Bjurulf P, von Schenck H, Larsson J. **Feeding dependence and nutritional status after acute stroke**. *Stroke* (1994.0) **25** 366-71. DOI: 10.1161/01.str.25.2.366
26. Mann G, Hankey GJ, Cameron D. **Swallowing function after stroke: prognosis and prognostic factors at 6 months**. *Stroke* (1999.0) **30** 744-8. DOI: 10.1161/01.str.30.4.744
27. Zheng T, Zhu X, Liang H, Huang H, Yang J, Wang S. **Impact of early enteral nutrition on short term prognosis after acute stroke**. *J Clin Neurosci* (2015.0) **22** 1473-6. DOI: 10.1016/j.jocn.2015.03.028
28. Powers WJ, Rabinstein AA, Ackerson T, Adeoye OM, Bambakidis NC, Becker K. **2018 guidelines for the early management of patients with Acute ischemic stroke: a Guideline for Healthcare Professionals from the American Heart Association/American Stroke Association**. *Stroke* (2018.0) **49** e46-e110. DOI: 10.1161/STR.0000000000000158
29. Lopez Espuela F, Roncero-Martin R, Zamorano JDP, Rey-Sanchez P, Aliaga-Vera I, Portilla Cuenca JC. **Controlling Nutritional Status (CONUT) score as a predictor of all-cause mortality at 3 months in stroke patients**. *Biol Res Nurs* (2019.0) **21** 564-70. DOI: 10.1177/1099800419860253
30. Cai ZM, Wu YZ, Chen HM, Feng RQ, Liao CW, Ye SL. **Being at risk of malnutrition predicts poor outcomes at 3 months in acute ischemic stroke patients**. *Eur J Clin Nutr* (2020.0) **74** 796-805. DOI: 10.1038/s41430-020-0605-8
31. 31.Han X, Cai J, Li Y, Rong X, Li Y, He L, et al. Baseline objective Malnutritional Indices as Immune-Nutritional Predictors of Long-Term recurrence in patients with Acute ischemic stroke. Nutrients. 2022;14(7). 10.3390/nu14071337.
32. Gasche Y, Copin JC, Sugawara T, Fujimura M, Chan PH. **Matrix metalloproteinase inhibition prevents oxidative stress-associated blood-brain barrier disruption after transient focal cerebral ischemia**. *J Cereb Blood Flow Metab* (2001.0) **21** 1393-400. DOI: 10.1097/00004647-200112000-00003
33. Belayev L, Liu Y, Zhao W, Busto R, Ginsberg MD. **Human albumin therapy of acute ischemic stroke: marked neuroprotective efficacy at moderate doses and with a broad therapeutic window**. *Stroke* (2001.0) **32** 553-60. DOI: 10.1161/01.str.32.2.553
34. Kim J, Song TJ, Park JH, Lee HS, Nam CM, Nam HS. **Different prognostic value of white blood cell subtypes in patients with acute cerebral infarction**. *Atherosclerosis* (2012.0) **222** 464-7. DOI: 10.1016/j.atherosclerosis.2012.02.042
35. Tirschwell DL, Smith NL, Heckbert SR, Lemaitre RN, Longstreth WT, Psaty BM. **Association of cholesterol with stroke risk varies in stroke subtypes and patient subgroups**. *Neurology* (2004.0) **63** 1868-75. DOI: 10.1212/01.wnl.0000144282.42222.da
36. Yu WM, Abdul-Rahim AH, Cameron AC, Korv J, Sevcik P, Toni D. **The incidence and Associated factors of early neurological deterioration after Thrombolysis: results from SITS Registry**. *Stroke* (2020.0) **51** 2705-14. DOI: 10.1161/STROKEAHA.119.028287
37. Sanak D, Herzig R, Kral M, Bartkova A, Zapletalova J, Hutyra M. **Is atrial fibrillation associated with poor outcome after thrombolysis?**. *J Neurol* (2010.0) **257** 999-1003. DOI: 10.1007/s00415-010-5452-4
38. Doehner W, Schenkel J, Anker SD, Springer J, Audebert HJ. **Overweight and obesity are associated with improved survival, functional outcome, and stroke recurrence after acute stroke or transient ischaemic attack: observations from the TEMPiS trial**. *Eur Heart J* (2013.0) **34** 268-77. DOI: 10.1093/eurheartj/ehs340
39. Gensicke H, Wicht A, Bill O, Zini A, Costa P, Kagi G. **Impact of body mass index on outcome in stroke patients treated with intravenous thrombolysis**. *Eur J Neurol* (2016.0) **23** 1705-12. DOI: 10.1111/ene.13071
|
---
title: Context-specific adaptation of an eHealth-facilitated, integrated care model
and tailoring its implementation strategies—A mixed-methods study as a part of the
SMILe implementation science project
authors:
- Sabine Valenta
- Janette Ribaut
- Lynn Leppla
- Juliane Mielke
- Alexandra Teynor
- Katharina Koehly
- Sabine Gerull
- Florian Grossmann
- Verena Witzig-Brändli
- Sabina De Geest
journal: Frontiers in Health Services
year: 2023
pmcid: PMC10012712
doi: 10.3389/frhs.2022.977564
license: CC BY 4.0
---
# Context-specific adaptation of an eHealth-facilitated, integrated care model and tailoring its implementation strategies—A mixed-methods study as a part of the SMILe implementation science project
## Abstract
### Background
Contextually adapting complex interventions and tailoring their implementation strategies is key to a successful and sustainable implementation. While reporting guidelines for adaptations and tailoring exist, less is known about how to conduct context-specific adaptations of complex health care interventions.
### Aims
To describe in methodological terms how the merging of contextual analysis results (step 1) with stakeholder involvement, and considering overarching regulations (step 2) informed our adaptation of an Integrated Care Model (ICM) for SteM cell transplantatIon faciLitated by eHealth (SMILe) and the tailoring of its implementation strategies (step 3).
### Methods
Step 1: We used a mixed-methods design at University Hospital Basel, guided by the Basel Approach for coNtextual ANAlysis (BANANA). Step 2: Adaptations of the SMILe-ICM and tailoring of implementation strategies were discussed with an interdisciplinary team ($$n = 28$$) by considering setting specific and higher-level regulatory scenarios. Usability tests were conducted with patients ($$n = 5$$) and clinicians ($$n = 4$$). Step 3: Adaptations were conducted by merging our results from steps 1 and 2 using the Framework for Reporting Adaptations and Modifications–Enhanced (FRAME). We tailored implementation strategies according to the Expert Recommendations for Implementing Change (ERIC) compilation.
### Results
Step 1: Current clinical practice was mostly acute-care-driven. Patients and clinicians valued eHealth-facilitated ICMs to support trustful patient-clinician relationships and the fitting of eHealth components to context-specific needs. Step 2: Based on information from project group meetings, adaptations were necessary on the organizational level (e.g., delivery of self-management information). Regulations informed the tailoring of SMILe-ICM`s visit timepoints and content; data protection management was adapted following Swiss regulations; and steering group meetings supported infrastructure access. The usability tests informed further adaptation of technology components. Step 3: Following FRAME and ERIC, SMILe-ICM and its implementation strategies were contextually adapted and tailored to setting-specific needs.
### Discussion
This study provides a context-driven methodological approach on how to conduct intervention adaptation including the tailoring of its implementation strategies. The revealed meso-, and macro-level differences of the contextual analysis suggest a more targeted approach to enable an in-depth adaptation process. A theory-guided adaptation phase is an important first step and should be sufficiently incorporated and budgeted in implementation science projects.
## Introduction
In recent decades, there has been a growing interest in adapting health care interventions. As implementing adapted interventions is often more efficient than developing new ones for each setting, this saves human, time and financial resources [1, 2]. While the current concept of adaption follows one introduced by Rogers in 1995 [3], it is defined diversely in the literature [4]. However, all agree that adaption processes are conducted to match the needs of the target population and to improve an intervention's fit, acceptability and effectiveness in the target context (4–7). Based on these adaptations' targets, e.g., content, method of delivery, the surrounding context (including congruity with the target population's culture), they can effectively redefine an intervention (8–10). Adaptations can include deletions, additions, or modifications [5] and can occur proactively (planned) or reactively [11].
Within a systematic review of adaptations of evidence-based public health interventions, Escoffery et al. [ 5] identified 42 distinct program adaptations. Among the most common reasons for adaptation were cultural changes ($64.3\%$), followed by new target populations ($59\%$) or settings ($57\%$). Few interventions were adapted to improve their feasibility or acceptability; and only $36\%$ of the adaption studies applied existing frameworks to guide their adaption processes. In a more recent scoping review of current adaption practices [8], $84\%$ of identified studies focused on micro- (i.e., individual-) level intervention adaptions; and the majority ($73\%$) did not report using guidelines or frameworks. Numerous frameworks to provide general guidance for planning and evaluating adaptions have recently emerged in the fields of HIV prevention, sexually transmitted infections, pregnancy, and substance abuse prevention (5–7). A 2019 scoping review [6] identified 13 adaptation frameworks with eight common steps: [1] conducting a needs assessment of the target community; [2] searching for and understanding available interventions with similar aims; [3] selecting a specific intervention; [4] deciding which parts require adaptions; [5] making the appropriate adaptions; [6] testing the adapted intervention; [7] implementing the adapted intervention; and [8] evaluating the adapted intervention. However, as the majority of published frameworks lack a theoretical basis and a multilevel contextual focus regarding the adaptation process, their guidance includes important gaps (5–8).
While context is defined as “a set of characteristics and circumstances that consist of active and unique factors and interacts, influences, modifies and facilitates or constrains the intervention and its implementation” [12], most studies focused on adaptation as a stand-alone process, ignoring the interactions between context, implementation, intervention design, and the adaption process itself [7, 8]. Complexity arises not only from the characteristics of the chosen intervention (e.g., the number of intervention components and the interactions between them) and of the target context, but also from the interaction between the two (13–16). This is especially true when complex health care interventions are facilitated via the “use of information and communication technology for health,” [17] i.e., electronic Health (eHealth) technology [18, 19]. Evidence indicates that $44\%$–$67\%$ of patients discontinue their use of offered eHealth tools due to mismatches between the technology and their context, particularly their needs (20–22); and only $0.01\%$ of all available eHealth applications make it into common use [23], with, however, a rising trend within the COVID-19 pandemic [24]. More recently published frameworks, such as the revised Medical Research Council (MRC) framework [13] or guidelines including the recently published ADAPT guidance [25] consider context a core element, upon which all four defined intervention research phases (i.e., development or adaptation of an identified intervention, feasibility, evaluation, and implementation of the intervention) depend. Within the updated MRC framework, considering that, because of reciprocal interactions, a complex intervention's effects are often highly dependent on the surrounding context, Skivington et al. [ 13] describe context as a core component underlying all phases of intervention research (i.e., development or adaptation of an intervention, its feasibility, evaluation, and implementation). As both effectiveness and implementation often depend heavily on context, the ADAPT guidance also describes context as a key component [25] that demands consideration in all steps of an adaptation process. However, while there is growing evidence on conceptual guidance and frameworks for adaptation processes, there is still an existing lack of empirical insights what methodological approach might work to operationalize such guidance [8]. And even if guidelines exist on how to report adaptions and modifications, e.g., the Framework for Reporting Adaptations and Modifications-Expanded (FRAME) [9], few, if any have yet specified how to conduct the adaptation process and how to methodologically merge context-specific information into the adaptation and implementation processes [13].
Implementation science provides a specific methodology to explore aspects of the context as the first step towards implementation [26]. This contextual analysis allows the interventionists to map information relevant to later steps (27–30) and to select or the most effective implementation strategies, which offer proven pathways to support successful adoption, implementation, sustainability and scaling up of interventions, programmes or practices in clinical practice [30]. Further, it involves specific methodological considerations, e.g., stakeholder involvement —an interactive relationship-building process between researchers and stakeholders. This is intended to facilitate a shared understanding and informed decision-making (31–33). Stakeholder involvement in implementation research is currently gaining increasing attention, as it provides a foundation to upon which both to build the intervention's acceptability and to ensure its sustainability in the target context [8, 34]. Still, even while the importance of context has been emphasized and methodology for its analysis has emerged [7], to date, the understanding of complex interventions’ adaptations is understudied in implementation science [6]. More detail and guidance are needed on how to conduct contextual adaptations of health care interventions and how to tailor their implementation strategies [7].
This study aims to fill this gap by reporting on the methods used to adapt an eHealth-facilitated Integrated Care Model (ICM) and to tailor its implementation strategies to the targeted setting based on a case example [35]. This will involve combining a contextual analysis with an in-depth adaptation process. More specifically, we applied a three-step approach with the following specific aims for each: Aim Step 1: to conduct a contextual analysis focusing on (a) current, context-specific practice patterns and patients' needs; (b) patients’ and clinicians' technology openness; and (c) patients' and clinicians' views regarding the challenges, benefits and requirements for implementing an eHealth-facilitated ICM in their setting. Aim Step 2: to inform the adaptation process by involving key stakeholders and end-users, and by consulting standard operating procedures (SOPs), overarching Swiss and medical device regulations. Aims Step 3: to merge the results of aim 1 (contextual analysis) and aim 2 (project group meeting results) toinform the final adaptation of the eHealth-facilitated ICM following the Framework for Reporting Adaptations and Modifications–Enhanced (FRAME) [9]; and to tailor our implementation strategies based on the Expert Recommendations for Implementing Change (ERIC) compilation [36].
## Overall design of this study and description of case example
This multi-level mixed-methods study [37, 38] combined quantitative and qualitative research methods in an equal-status concurrent approach to gain knowledge about the context-specific adaptation of a complex eHealth-facilitated integrated care model. Figure 1 provides an overview of the applied study designs, data collection content and timing, and the analyses for each step. To maximize our understanding of a context-driven adaptation process, we followed a three-step approach. We analyzed first quantitative, then qualitative data, then examined the two merged within a multi-level approach [35].
**Figure 1:** *Overview of the study design, data collection timepoints and analysis for steps 1 and 2 to inform the adaptation and tailoring of implementation strategies (step 3). ERIC, Expert Recommendations for Implementing Change; FRAME, Framework for Reporting Adaptations and Modifications–Expanded; MDR, Medical Device Regulation; QUAN, quantitative data; QUAL, qualitative data; SOPs, standard operating procedures.*
A nested case study approach was used to generate an in-depth, multi-faceted understanding of a complex issue—intervention adaptation—in a real-life context [39, 40]. This approach allowed us to use a case example from an international multicenter implementation science project to develop (41–44), adapt (Phase A), implement and evaluate (Phase B) [45] an Integrated Care Model (ICM) in allogeneic SteM cell transplantatIon faciLitated by eHealth (SMILe-ICM). Based on an in-depth contextual analysis [41], theory [12, 36, 46, 47] and stakeholder as well as end-user input [42, 43], our study group developed SMILe–ICM for our first participating center in Germany [described in detail elsewhere (41–44)]. For use in the Swiss target setting, this version required various adaptations, including the tailoring of its implementation strategies. In brief, as originally developed, the SMILe-ICM is based on the five building blocks of the eHealth enhanced Chronic Care Model (eCCM) [46]. Table 1 describes the SMILe-ICM's original core components, its delivery methods and its delivery timepoints.
**Table 1**
| Four core modules |
| --- |
| Monitoring & follow-up of vital signs, symptoms and health behaviour (42, 43) |
| → targets alloSCT patients’ insecurities regarding recognizing and reacting to new symptoms; monitors 17 items |
| Infection prevention (42, 43) |
| → targets patients’ challenges regarding infection prevention measures by including (1) adequate hand hygiene; (2) airborne pathogen-related risk reduction; and (3) safe food handling, preparation, and consumption. |
| Medication adherence (44) |
| → targets patients’ immunosuppressant intake (41–43) |
| Physical activity (42, 43) |
| → targets patients’ physical activity alongside their energy levels |
| Two delivery modes and detailed intervention description (42, 43) |
| (1) SMILe-ICM consists of a technology component, i.e., a mobile app for patients (SMILeApp, initially Android only) and a monitoring interface for care professionals (SMILeCare). In the initial German version (43), patients could insert 17 relevant parameters (i.e., vital signs and symptoms to be checked daily) to the SMILeApp. All data entered to the SMILeApp are transferred to the alloSCT center. With each patient's approval, their input can be overseen by APNs via the SMILeCare monitoring interface. This data transfer allows the APNs to monitor, identify and act upon critical values, symptom-related issues or trends based on pre-established cut-offs and risk-adjusted care protocols. Care protocols also specify when other members of the alloSCT team (e.g., responsible physicians, nurses) will be involved. Patients can also read up on important symptoms in the SMILeApp lexicon and receive a step counter to assess daily physical activity.(2) SMILe-ICM is delivered via a human part, i.e., APNs. In the original German version, the APNs conducted 12 personal consultations (Visits 1–3 during inpatient stay, visits 4–12 as from outpatient stay) at pre-defined timepoints starting 14 days prior to the patient's alloSCT and extending to one year after. The post-transplant nursing visits are planned in conjunction with the routine outpatient clinic follow-up schedule: While most inpatients attend a 3-week rehabilitation program directly after discharge, outpatient intervention sessions start intensifying 3 weeks after discharge (42, 43)—first weekly, then monthly for stable patients. During these visits, the APN team provides intervention modules on symptom recognition and assessment, infection prevention, physical activity and medication adherence. |
## Step 1: Materials and methods for the contextual analysis
As depicted in Figure 1, quantitative and qualitative data for step 1 (contextual analysis) were collected from April 2019 to January 2020. The contextual analysis has been approved by the ethics committee [Ethics Committee Northwest and Central Switzerland (EKNZ); BASEC 2019-00307] and is based on previous work by Leppla et al. [ 41]: An in-depth contextual analysis was conducted to inform the development and implementation of the SMILe-ICM to the first participating study's center (Freiburg, Germany) [41].
## Theoretical frameworks to guide step 1
Step 1. The Basel Approach for Contextual ANAlysis (BANANA) [48] guided the SMILe contextual analysis in our first participating center [41], as well as for this study, which is theoretically based on the Context and Implementation of Complex Interventions (CICI) framework [12]. BANANA [48] was developed to provide a step approach to conducting contextual analyses in implementation science projects as follows: choosing a theory, model or framework; using empirical evidence; involving stakeholders; designing a study specifically for the contextual analysis; and determining the relevance of contextual factors for implementation strategies/outcomes and intervention co-design [48]. In accordance with the overarching SMILe project [42, 43], the Swiss contextual analysis and adaptation phase was also theoretically based on the eCCM [46], which supports operationalization of all necessary chronic illness management dimensions.
## Setting and sample
The contextual analysis was conducted at the Department of Hematology, University Hospital Basel (USB, Switzerland). From April 2019—January 2020, a convenience sampling procedure to survey allogeneic stem cell transplanted (alloSCT) patients from the USB outpatient clinic had been conducted. Eligible patients were: [1] transplanted and followed up at the USB; [2] aged ≥18 years; [3] between six weeks and three years post-alloSCT; [4] able to read and understand German. Those with cognitive or physical impairment that prevented adequate communication were excluded. Clinicians had to meet the following inclusion criteria: [1] >6 months' employment in the transplant center; [2] ≥$50\%$ in direct clinical practice; and [3] familiarity with post-transplant care. For individual interviews with patients, the same eligibility criteria applied. Purposive sampling was used to ensure approximately equal variation regarding age, gender, education, and living situation. For the clinicians' survey and focus group sample, clinicians had to meet the following inclusion criteria: [1] >6 months' employment in the transplant center; [2] ≥$50\%$ in direct clinical practice; and [3] familiarity with post-transplant care.
## Surveys
Based on the initial contextual analysis [41], patients, clinicians and the transplant director filled in a questionnaire to assess the alloSCT center's structural characteristics, practice patterns regarding chronic illness management, technology openness and perceived importance of eHealth for healthcare applications. Building on previous work by the BRIGHT study team [49, 50] and the PICASSO-TX study [51, 52], Leppla et al. [ 41] adjusted the questionnaires to the alloSCT setting. Supplementary Table S1 provides an overview of all assessed variables, assessment tools, and their psychometric characteristics, highlighting variables that were adapted and/or added for this study.
## Interviews and focus groups
Individual interviews with patients took place from May until June 2019 to capture a rich understanding of patients` experiences with follow-up care after their alloSCT, their self-management tasks, and their determinants and preferences regarding eHealth application use in their daily lives. Since the aim was to obtain in-depth individual information and understand the personal experience, group discussions (e.g., focus group discussions) were considered as less appropriate in this case. Therefore, semi-structured interviews were conducted by the second author (JR), who is a specially educated nurse in the field of hematology and trained interviewer. Based on previous work [41] and following an interview guide, which has been developed based on eCCM dimensions [46] and CICI framework [12], open-ended questions were asked. These interviews, which took place during the patients' appointments at the USB, were audio-recorded and transcribed verbatim.
Focus group interviews with clinicians were conducted in June 2019 to explore their experience with follow-up care and their view of eHealth applications in clinical practice. Since a focus group with amutual exchange of perceptions and expert knowledge can lead to deeper insights into the needs and thoughts of a target group [53] and can generate additional ideas, we have chosen this qualitative approach instead of conducting individual interviews. The focus groups were moderated by the first author (SV), while the second author (JR) mind-mapped key themes on a flip chart to help memorize previous thoughts and summarize all of the focus groups' input [54, 55]. During the focus group interviews, the participants could see the emerging maps and could therefore add or change keywords at the end of the focus group session. Both interviewers are experienced qualitative researcher and were trained in conducting focus group sessions. In accordance with Leppla et al. 's earlier contextual analysis [41], clinicians were asked open-ended, semi-structured questions to explore this study's main areas of interest, i.e., we aimed to explore adaptation and implementation requirements for this specific Swiss setting: Therefore, the developed SMILe-ICM for our first participating center (Freiburg, Germany) was briefly illustrated and explained to the clinicians. Afterwards, we then used the group discussion to illuminate what clinicians believe needs to be adapted or further developed for their clinical setting and what is needed to tailor and implement such an eHealth-facilitated ICM for the setting-specific needs.
## Data analysis
Descriptive statistics were computed as appropriate for the measurement levels and data distributions (frequencies, means, standard deviations, ranges, medians, interquartile ranges). Secondly, the correlations between the main variable of interest—patients' willingness to use self-management devices in future—and the independent variables—age, gender and education—were analyzed with the Spearman's rho test (age and education) and the Mann-Whitney test (gender). The significance level was set at $p \leq 0.05.$ Significant correlations were tested using logistic regression with a forward approach. Statistical analysis was performed using R Version 3.6.2 [56]. Semi-structured interviews with patients were thematically analyzed following Braun et al. 's [57] six-phase procedure. The ATLAS.ti 8 software package was used for data management [58]. To analyze the focus groups with clinicians, a mind-mapping technique [54, 55] was applied. As shown in Figure 1, the contextual analysis's QUAN and QUAL results have been merged and synthesized according to the eCCM dimensions [46].
## Step 2 and 3: Material and methods to inform and to conduct the adaptation process
As shown in Figure 1, based on integration of the previously-gathered contextual analysis information, multi-stakeholder input and a user-centered design (UCD) approach [59], intervention adaptation and tailoring of implementation strategies were conducted from March 2020—January 2021.
## Theoretical frameworks to guide steps 2 and 3
Step 2. To inform the context-specific adaptation of the technology components, we applied UCD techniques [59] by building upon previous work from the iterative software development process [42]. Intervention designers place end users' (e.g., patients/caregivers/clinicians) preferences, needs and feedback at the center of each phase of the design process, with the goal of developing or adapting highly usable and accessible products via various research and design techniques (60–62). One way of achieving this goal is usability testing very early in the design process, e.g., using interface mock-ups or—later in the process—live applications.
Combining UCD with agile software development enhances its positive aspects. Agile software development offers an iterative, incremental system of software construction [63, 64]. While helping researchers focus strongly on creating high-priority functionality, it also acknowledges the value of stakeholder groups, encouraging regular presentations of current product increments to them. Typically leading to particularly safe, effective products, UCD also ensures that the intended users find those products useful and manageable, thereby enhancing their acceptability [65].
Step 3. To theoretically describe the SMILe-ICM's adaptations and why they were necessary, we followed the Framework for Reporting Adaptations and Modifications–Enhanced (FRAME) [9], which provides a coding structure to document types of intervention modifications. Based on literature review, qualitative interviews and stakeholder involvement, Stirman et al. [ 9] developed a coding system [66] and added additional considerations such as reason for adaptation (e.g., cultural/religious norms, time constraints, access to resources), goal of the adaptation (e.g., increase reach, improve fit) and whether the adaptation was proactive or reactive. The updated FRAME includes eight key components: [1] when and how in the implementation process the modification was made (i.e., timing); [2] whether the modification was proactively planned or reactively unplanned; [3] who participated in adaptation-related decisions; [4] what is modified (i.e., intervention's content, contextual type of delivery, staff training, implementation strategies); [5] at what level of delivery (e.g., individual or unit level) the modification is made; [6] the type or nature of context or content-level modifications (e.g., adding or skipping elements); [7] the extent to which the modification is fidelity-consistent; and [8] the reasons for the modification, including (a) the modification's intent or goal (e.g., to increase reach or fidelity) and (b) contextual factors that influenced the decision (e.g., socio-political factors such as laws or organizational reasons such as staff shortages) [9].
To finally choose implementation strategies for each phase of the SMILe project and to contextually adapt and tailor them for the USB's routine clinical practice, we followed the Expert Recommendations for Implementing Change (ERIC) taxonomy [30], which defines a set of 73 implementation strategies [67]. These can be grouped into nine categories: use evaluative and iterative strategies; provide interactive assistance; adapt and tailor to the target context; develop stakeholder relationships; train and educate stakeholders; support clinicians; engage consumers; utilize financial strategies; change infrastructure. In a first step, determinants to implement SMILe-ICM into the Swiss setting has been identified according to micro-, meso-, and macro-level by the interdisciplinary clinical and scientific steering group meetings and categorized in line with CICI framework [12] [overview about determinants are published elsewhere [45]]. We also followed implementation stages according to Exploration, Preparation, Implementation, Sustainment (EPIS) framework [72] to classify the chosen implementation strategies according to the SMILe project's pre-phase, Phases A (development/adaptation) and Phase B (implement and evaluate) and sustainment.
## Setting, sample and materials for the adaptation process
Regular project group meetings with key stakeholders, which were led by the first author (SV), were conducted to adapt the SMILe-ICM and to tailor its implementation strategies to the target setting. Therefore, stakeholders were identified following a Stakeholder Analysis Matrix [68]: in clinical and research team discussions and brainstorming rounds, we analyzed which internal team members and which external USB staff would be affected by implementing the two main new components (human and technology) of the SMILe-ICM, as well as when (before or after the inpatient stay) and how (directly or indirectly) they would be affected. For each identified stakeholder, we analyzed their impact/influence (low, middle, high) on our project and identified the necessary resources for their engagement [31]. For setting-specific adaptations, in addition to stakeholder involvement, we consulted standard operating procedures (SOPs) of the USB's hematology department [69] and higher-level regulatory scenarios [e.g., *Swiss data* protection regulations [70], medical device regulations [71]].
## Quantitative assessment of users' satisfaction
Based on our previous described agile software development process [42] and structured collaboration between nursing scientists and software specialists, a purposive sample of alloSCT patients ($$n = 4$$), which is described as a sufficient number of participants for end-user usability tests for technology components [72], was formed in $\frac{01}{2021.}$ To select the sample, SV and JR screened electronic health records to guarantee that different educational levels, genders, and ages were represented, and that the members would be likely to sign a written informed consent form before participation. The four Advanced Practice Nurses (APNs), who would deliver the intervention in test phase B, were approached and recruited by SV and JR.
## Synthesis of findings
Results from project group meetings and end-user tests were merged with setting-specific SOPs and overarching Swiss regulations (see Figure 1); then, all were fitted into a meta-matrix [73]. This synthesizes contextual analysis results (step 1) based on the eCCM [46] and the results of step 2 to inform the SMILe-ICM adaptation following FRAME [9] and to tailor its implementation strategies according to ERIC guidelines [36].
## Results
According to our three-step approach, our developed methodology will be applied to the results section and described based on the SMILe case example described above (41–44).
## Step 1: Results of the contextual analysis
The merged quantitative and qualitative contextual analysis results are briefly described in the following and synthesized in Table 4 in line with eCCM dimensions [46]. Supplementary Tables S2, S3 and Figures S1, S2 provide detailed information.
## Sample characteristics
As shown in Figure 2 (flow chart), a convenience sample of 64 eligible patients was invited to participate in the survey (response rate $94\%$). Of those who accepted, ten were further invited to participate in the individual interviews, which lasted between 40 and 64 min (mean = 48 min); all accepted and participated. For the clinician survey, fifteen HCPs were invited to participate in focus group meetings, all of whom accepted. A random sample of five also agreed to participate in the interviews. Table 2 shows the participating patients' and clinicians' demographic information.
**Figure 2:** *Flow chart of included patients and clinicians. (A) Participants within the contextual analysis (step 1). (B) Participating patients, stakeholders and additional experts to inform the adaptation process (step 2).* TABLE_PLACEHOLDER:Table 2
## Structural characteristics of alloSCT center
Located in Northwestern Switzerland, the Department of Hematology of University Hospital Basel (USB) is one of the country's three SCT centers performing not only autologous, but also alloSCTs. The other two are attached to University Hospitals Zurich and Geneva. In Switzerland, about 250 alloSCTs are performed annually [74], of which roughly 100 take place at USB. Several thousand alloSCT patients are currently in follow-up care [75]. According to the Swiss Federal Law on Health Insurance, health insurance covers all allowable costs of medical treatment and hospitalization [76]. Patients are hospitalized around 10 days prior to alloSCT until a mean of 30 days (±5 days) post-transplantation. After engraftment, patients are discharged once they show stable blood values and health condition. After discharge, they return for follow-up 1–2 times per week for the first 3 months (depending on their health status), then once per week until 6 months post-SCT. Follow-up intervals gradually increase to once yearly. Uniquely for the Swiss setting and as shown in Supplementary Table S2, $42\%$ of patients are additionally followed up in external hematological centers closer to their homes.
## Patients' and clinicians' perspectives on current practice patterns
According to synthesized and merged results from patients' and clinicians' surveys, as well as individual patient interviews, three major themes occurred regarding the Basel alloSCT center's practice patterns.
## Transition to home as most complex phase
Results revealed that follow-up was currently acute-care driven. The most complex phase was seen as the transition to home. While a large majority of patients ($93\%$) generally very satisfied with the provided care, the majority ($72\%$) denied having been contacted by their responsible clinicians after an appointment to ask about their general progress (Supplementary Figure S1A); and $45\%$ indicated that they did not understand the written information they received some or most of the time (Supplementary Figure S1B). While the majority affirmed that they had been advised to adhere to recommendations (Supplementary Figure S1C), $35\%$ did not adhere to checking their cheek temperature on a regular basis (Supplementary Table S2). The mean overall patient-perceived chronic illness management rating was 30.6 (±7.8, range: 11–55). Clinicians' chronic illness management scores (Supplementary Table S2) revealed high variability (mean CIMI-BRIGHT: 2.92 (± 0.58, range: 2.49–3.87). Critical deficits were apparent in 25 items, i.e., <$50\%$ positive responses regarding self-management support (10 items), delivery system design (7 items), clinical decision support (5 items) and use of clinical information systems (3 items).
While being generally very satisfied with the care and discharge support they received, patients described the first weeks after discharge from hospital as the most complex, marked by insecurity how to handle self-management recommendations at home. “ Yes, I have about 30 tablets a day. When I was still an inpatient, it wasn't so much; but then it increased and another training would have been helpful.” ( female, 42 years).
## Wish for continuous self-management support across the entire patient pathway
A trustful, continuous relationship with the health staff was described by patients as very important. As some experienced frequent changes of assistant doctors, continuity of care was suboptimal. “ Communication and openness, that is the most important thing. But the doctors have changed a bit. (…) And now I'm being cared for by different doctors and that has made it more complicated.” Further, patients wished they had received more intensive self-management support between the time their alloSCT was approved and the time the actual alloSCT was conducted (which can be months). “ After the patient actually knows that he is going to be transplanted, months can still pass and that might also help to get support there already” (female, 62 years).
## Caregivers' support and burden
Patients reported that, especially in the initial period after discharge, family members provided support with medication management, household chores and transport to and from the hospital. When family members also supported them emotionally, especially when symptoms or side effects of medication occurred, patients also recognized their burden. “ The awareness that things are different now, the family did not realize that right away (…) and all that was of course not easy for them either” (male, 64 years).
## Patients' and clinicians' technology openness
Contextual analysis results revealed that patients' eHealth openness was high: $81\%$ would be open to try new technologies and $80\%$ would quickly get used to it, while the majority would prefer to receive new applications on their own smartphone ($87\%$) or tablet ($46\%$) (Supplementary Figures S2A,B). As presented in Supplementary Table S2, $50\%$ would be willing to use an App for their medication plan; but $54\%$ would not feel confident entering or updating their medication plans on their own. As shown in Table 3, patients considered the idea of developing new technologies that would allow physicians and nurses to monitor vital signs and symptoms as most important. Higher-educated patients perceived it as more important to develop new technologies that support physical activity ($p \leq 0.05$); and compared to patients aged over 60 years, those under 60 scored higher on perceived importance of new technologies that provide information ($p \leq 0.05$, Supplementary Table S3). While all surveyed clinicians ($100\%$, Supplementary Table S2) indicated that written guidelines for care were easily available, electronic medical records were not yet used and no information systems were available to monitor patients at home.
**Table 3**
| To what extent do you think it's important to develop new technologies that … | Results | Results.1 | Results.2 |
| --- | --- | --- | --- |
| To what extent do you think it's important to develop new technologies that … | N | Median (25th–75th percentile) | Range |
| support physical activity? | 60 | 5.5 (3–8) | 0–10 |
| measure physical activity? | 60 | 5 (3–7.25) | 0–10 |
| give information about healthy eating? | 60 | 7 (5.75–8) | 0–10 |
| give information about infection prevention? | 60 | 7.5 (6–8) | 0–10 |
| give information about correct hand hygiene? | 60 | 8 (6.75–8) | 0–10 |
| regularly request your vital signs? | 60 | 8 (6–9) | 0–10 |
| regularly request your physical symptoms? | 59 | 8 (6–9) | 0–10 |
| allow doctors and nurses to monitor your vital signs, symptoms and medication intake? | 59 | 9 (7–9) | 0–10 |
| remind you of your appointments at the transplantation center? | 60 | 6.5 (3.75–9) | 0–10 |
## Clinicians' views on the challenges, benefits and requirements for implementing the SMILe-ICM
During the focus groups with clinicians, the current SMILe-ICM version was demonstrated. Afterwards, participants were asked about their perceptions regarding the challenges, benefits and requirements for adapting and implementing the original SMILe-ICM for their setting. Four main topics arose from the focus group interviews: [1]Implementation must be based on context-specific requirements. Clinicians agreed that both human and technology intervention delivery modes should be implemented. However, the entire alloSCT team needs to support this new eHealth-facilitated ICM in daily clinical practice. That will require sufficient resources (staff), as well as APNs who are very well-trained regarding technology and self-management support. Furthermore, the division of tasks should be clearly regulated and open exchanges between participating centers ensured. Clinicians highlighted the advantage of starting the intervention as early as day −14 before alloSCT.[2]Human role should be an experienced, trained nurse with competencies in alloSCT care support. Clinicians stated that the proposed eHealth-facilitated ICM should be provided by APNs, who are trained in self-management support, education and care coordination, and are educated beyond an experienced ward nurse: with extended competencies over the whole patient pathway, they only require support from senior physicians.[3]SMILeApp/technology requirements and potential barriers to use. After having received an overview of the SMILeApp modules' content, all clinicians agreed that, for patients and caregivers, medication management as well as psychosocial support should be included in the SMILe-ICM, which they estimated would be most helpful in the first 3 months after alloSCT. Additionally, barriers to patients' accessibility (e.g., because they are too sick, less educated) to App use should be considered and included in the intervention sessions.[4]Costs must be covered to guarantee sustainability. According to clinicians, such new care models should be fully covered by health insurance to ensure sustainability.
## Sample
As shown in Figure 2, an interdisciplinary clinical team from USB, i.e., nurses, physicians, management ($$n = 11$$), IT specialists and computer scientists ($$n = 7$$), a clinical and research steering group ($$n = 10$$), had an in-depth discussion on adapting the SMILe-ICM. Their objective was to coordinate specific processes in two- to four-weekly project group meetings and provide written feedback. Additional clinical experts, i.e., a psycho-oncologist, a nutritionist, a pharmacist, a lawyer originating from the setting and disposition team members were asked to clarify specific questions arising from the project group meetings. Hematology department SOPs [69] and higher-level regulatory scenarios were also consulted [e.g., *Swiss data* protection regulations [70], medical device regulations [71]] for setting-specific adaptations.
## Stakeholder, end-user and overarching regulation involvement to inform differences and needs for adaptation
The combination of project group meetings’ information and setting-specific SOPs informed the adaptation of the SMILe-ICM. Table 4 summarizes the results of the stakeholder involvement and consultation of overarching regulations. This will be described in relation to the adaptation project group meetings' time frames and how micro-, meso- and macro-level information was merged.
**Table 4**
| Dimensions of the eHealth enhanced Chronic Care Model | Synthesis of findings from step 1: contextual analysis | Synthesis of findings from step 2 to inform adaption and implementation of SMILe-ICM for the USB setting | Synthesis of findings from step 2 to inform adaption and implementation of SMILe-ICM for the USB setting.1 | Nature of adaptation according to FRAME | Nature of adaptation according to FRAME.1 | Nature of adaptation according to FRAME.2 | Implementation strategies according to ERIC guidelines—in bold adapted/added for the Swiss setting |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Dimensions of the eHealth enhanced Chronic Care Model | Synthesis of findings from step 1: contextual analysis | Source of information | Result | WHAT is modified and at what LEVEL OF DELIVERY | Nature and description of modification | Goal | |
| Organizational level | • Gaps in chronic illness management as no interdisciplinarity• Low- to mid-level chronic illness management | Project group meetings with clinicians and comparison with USB SOPs (69) | In contrast to the first participating center [for which SMILe-ICM was originally developed (41–44)], shorter inpatient stays: Patients at USB hospitalized as from day Tx -10. | – Contextual adaptation—Timepoint of delivery | – Condensing: Visit 1 starts from day Tx minus10; Session 2 limited to one visit due to short timeframe within inpatient setting | – Improve fit with existing practice patterns (care processes) at USB; Improve feasibility– Improve effectiveness/outcomes | Pre-Phase – Access new funding– Develop academic/clinical/technical partnerships– Inform local opinion leaders Phase A– Identify early adopters– Visit other sites (e.g., first participating center Freiburg, Germany)– Conduct local needs assessment and consensus discussion– Adapt and tailor to Swiss context– Adapt educational material– Organize clinical implementation teams |
| Self-manage-ment support | • Follow up current acute care driven care model (with lack of structured and continued self-management support in follow-up care)• Patients and clinicians wish to add electronic monitoring module for psychosocial and medication management• An eHealth-facilitated ICM is perceived as supportive to help patients overcome their insecurities in recognizing and judging new symptoms, especially after discharge to home | Contextual analysis, project group meetings with clinicians combined with information contained gathered from USB SOPs (69). | In contrast to the first participating center (41–44), no post-discharge rehabilitation program, but direct follow-up appointments after discharge in outpatient setting. | – Contextual adaptation—Timepoint of delivery | – Spreading: breaking up first intervention session after discharge: intervention session 4 (immediately after discharge) was intensified (to cover visits 4a and 4b) due to regular follow-up visits directly after discharge at USB | – Improve fit with existing practice patterns (care processes) at USB– Improve feasibility– Improve effectiveness outcomes | |
| Self-manage-ment support | • Follow up current acute care driven care model (with lack of structured and continued self-management support in follow-up care)• Patients and clinicians wish to add electronic monitoring module for psychosocial and medication management• An eHealth-facilitated ICM is perceived as supportive to help patients overcome their insecurities in recognizing and judging new symptoms, especially after discharge to home | Project group meetings with clinicians combined with information contained gathered from SOPs and expert consultations (i.e., dietician, psycho-oncologist) | In contrast to the first participating center (41–44) and in comparison with USB-SOPs (69): patients receive pre-discharge nutrition and medication management counselling from dietician/pharmacist. | – Contextual adaptions: Dosing of intervention | – Tailoring/condensing: Intervention session content on SMILe-core “medication management” and ”infection prevention” modules tailored to existing discharge planning at USB: shortening elements that are already delivered by standard care; if required, detailed version given in pre-discharge visit (3rd visit). | – Improve fit with existing practice patterns (care processes) at USB– Improve feasibility– Improve effectiveness outcomes– Address cultural factors | |
| Self-manage-ment support | | Consultation with clinicians/experts: Psycho-oncologist, Senior Physician and dietician | Revealed information to adapt lexicon text and contact details in SMILeApp according to USB SOPs (69) | – Contextual adaptions: Content adaption | – Tailoring: Refine lexicon entries to Swiss context according to SOPs by keeping core information / elements as original ones, replacing contact details in App that apply to USB-setting | – Improve fit with end-user needs and preferences | |
| Self-manage-ment support | | Usability tests and project group meetings with software developers and usability experts (Augsburg Team) | As in line with first participating center (41–44) and based on USB user test results: need to insert visualization of inserted values for patients’ SMILeApp | – Content: Adding elements | – Integrating: Technological functionality added = The evolution of entered values (vital signs, symptoms) can now also be viewed by patients on their own devices (mobile phones or tablets). | – Improve fit with end user needs and preferences– Improve feasibility– Improve effectiveness/ outcomes– Address cultural factors | |
| Self-manage-ment support | | Usability tests and project group meetings with software developers/usability experts and nursing scientists/APNs | Similar to first participating center (41–44) and based on contextual analysis and user tests results at USB: adding electronic monitoring module to assess adherence to immunosuppressants and other medication | – Content: Adding technology functionality | – Integrating: Technological functionality added (electronic monitoring) = patients are additionally asked whether they have taken their immunosuppressive medication and all others as prescribed (yes/no/skip). Patients receive additional electronic monitoring system (128) | – Improve effectiveness/ outcomes– Address cultural factors | |
| Self-manage-ment support | | Usability tests and project group meetings with software developers/usability experts and nursing scientists/APNs | In contrast to the first participating center (41–44) and based on user tests: Revealed information for changing the salutation in the SMILeApp | – Content: Tailoring technology components to patients’ preferences | – Tailoring: Changed from informal greeting (“Du”) to formal greeting (“Sie”) in SMILeApp | – Address cultural factors– Increase reach or engagement | |
| Delivery System Design | • eHealth-facilitated ICM has been perceived as valuable by patients and clinicians for their setting to improve care and symptom management provided by specialized trained APNs• APNs must be embedded in alloSCT team and ongoing exchange/clear division of work is needed | Project group meetings with IT team, software developers/usability experts and nursing scientists/APNs | In contrast to the first participating center (41–44): IT team members highlighted the need for technical adaptions, i.e., need for an iOS-version instead of providing tablets to patients, who do not have compatible cell phones. | – Content: Adding technological functionality | – Integrating: Technological functionality added = App is now also compatible with iOS devices. | – Improve effectiveness outcomes– Address cultural factors– Improve fit with recipients– Improve feasibility | |
| Clinical Decision Support System | • eHealth-facilitated ICM has been perceived as valuable by patients and clinicians for their setting to improve care and symptom management provided by specialized trained APNs• APNs must be embedded in alloSCT team and ongoing exchange/clear division of work is needed | Project group meetings with IT team, software developers/usability experts and nursing scientists/APNs | In contrast to the first participating center (41–44): IT team members highlighted the need for technical adaptions, i.e., need for an iOS-version instead of providing tablets to patients, who do not have compatible cell phones. | – Content: Adding technological functionality | – Integrating: Technological functionality added = App is now also compatible with iOS devices. | – Improve effectiveness outcomes– Address cultural factors– Improve fit with recipients– Improve feasibility | |
| Clinical Information System&eHealth education | • Patients’ high openness and willingness to use new applications on own smartphone or tablet• Technology needs to be connected to an APN, which is trained in providing eHealth, facilitated self-management support.• Supervision is needed (human and technology) | Swiss data protection regulations (68), and consultation with IT team and USB lawyers. | IT team members and USB lawyers highlighted that the data protection concept formulated for SMILe technology would have to be reformulated and adapted to the “Basel-Stadt Information and Data Protection Law” (78). | – Contextual adaptions: Content adaption | – Tailoring: Adaptation of the data protection concept for SMILe Technology to the Swiss setting on the basis of the law on information and data protection. | – Address cultural factors | Phase A: – Revise professional roles– Create new clinical teams– Conduct educational meetingsPhase B and sustainment:– Ongoing consensus discussion and information of local opinion leaders– Provide clinical supervision– Provide local technical assistance– Remind clinicians– Promote spread of clinical innovation |
| Clinical Information System&eHealth education | • Patients’ high openness and willingness to use new applications on own smartphone or tablet• Technology needs to be connected to an APN, which is trained in providing eHealth, facilitated self-management support.• Supervision is needed (human and technology) | Project group meetings with data protection officer, software developers/usability experts and nursing scientists/APNs and consultation with higher-level regulatory scenarios: Swiss data protection regulations (70), medical device regulations (71) | Similar with first participating center (41–44) and according to IT team, data protection officer and overarching regulations [Medical Device Regulation (71)], Swiss data protection regulations (70), planned SMILeApp functionality as medical device was not yet possible due to high costs and organizational efforts | – Content: Skipping elements | – Put on hold: Programmed feedback loops in the SMILeApp and monitoring interface were not yet implemented. | – Improve feasibility– Reduce cost | |
## Findings from clinical expert group meetings merged with setting-specific SOPs
From June to December 2020, the identified clinical expert group—six APNs working in the USB hematological department's in- and outpatient settings—met every 4 weeks to work out the adaptation of the intervention materials to the Swiss setting. Within these meetings, the team compared all written materials for each SMILe intervention session (visit 1–12) with existing setting-specific SOPs [69].
For this purpose, 16 SOPs (e.g., post-alloSCT nutritional, infection prevention recommendations) were consulted. Based on the project groups' meeting exchanges, merged with written feedback from the APNs and considering setting-specific SOPs, SMILe-ICM was adapted and tailored to the Swiss setting as shown in Table 4. In contrast to the first participating center, structured pre-discharge information packages on medication management and dietary recommendations were already included in usual-care discharge planning at the USB. In addition, intervention materials had to be adapted (e.g., the recommendation on wearing an FFP3 mask vs. FFP2 mask) based on SOPs and additional consultation with clinical experts (i.e., dietician, pharmacists).
By comparing the developed intervention sessions and with USB's clinical expert knowledge and SOPs, it also became apparent that, compared to the first center, the wealth of general information on the alloSCT process (e.g., details of the transplantation procedure, possible side effects of chemotherapy) within visit 1 is not possible due to the differences between the two hospitals' usual alloSCT clinical care processes: at the first center, visit 1 takes place as early as d-14 pre-alloSCT. This is not possible in the USB setting, where hospitalization only starts 10 days prior to alloSCT, with numerous examinations taking place on the first two inpatient days. At USB, then, visit 1 is only feasible from d-7, a full week closer to alloSCT.
## Interdisciplinary meetings with software developers, nursing scientists/clinical nurse specialists and the IT Team
As shown in Figure 2, an interdisciplinary team consisting of three setting-specific IT-specialists, four SMILe-team software engineering developers and two nursing scientists/clinical nurse specialists met every 4–8 weeks (June to October 2021) to discuss adaptations to the SMILe Software components and obtain access to the relevant setting- specific IT infrastructure (i.e., installation of backend components). These meetings revealed a need for additional technical adaptations to fit the Swiss setting, i.e., to compile an iOS-version of the SMILeApp in addition to the original Android version. In terms of implementation, providing tablets to patients who had no Android-enabled cell phones would have been impractical compared to generating an iOS-compatible version. In light of the fact that $47\%$ of the Swiss population use Apple (iOS) smartphones, it would also impact the intervention's sustainability [77].
## Consultation of overarching regulations
Following *Swiss data* protection regulations [70], “Swiss Information and Data Protection Law” [78], and in consultation with USB lawyers, the developed data protection concept for the German setting was adapted to the Swiss setting (see Table 4). Although automated feedback algorithms for each parameter (e.g., temperature >38.5: contact the center immediately) had been developed [43], due to Switzerland's strict medical device regulations (enacted in May 2020), these could not be implemented in the Swiss version of the SMILeApp [71]. Specifically, the App's feedback loops—both those already developed and those that were planned—would have classified the SMILeApp as a class IIb medical device, which would require extensive and costly certification, additional certification via Swissmedic and ongoing quality management [71, 79]. The research budget of the project did not allow the investment at that time.
## Usability tests: end-users' satisfaction
In January 2021, we conducted end-user tests with five patients (mean age 65; $80\%$ male; $100\%$ living in partnerships; education levels ranging from vocational school to master's degree) and four APNs (see Figure 2), who evaluated the SMILe monitoring interface. The tests lasted between 12 and 20 min (mean = 15 min). After completing the tasks, patients and APNs filled in SUS questionnaires. For the APNs, a mean score of 89.5 indicates a very high level of user satisfaction (i.e., scores above 80). Because circumstances of the COVID-19 pandemic made it necessary to conduct our usability tests with patients in a virtual environment, the conditions for the test were sub-optimal, particularly regarding their ability to ask the testers about the questions. As a result, patients' quantitative scores could not be used due to a high rate of missing responses ($60\%$). However, qualitative information from the think-aloud part and subsequent consultations with patients and APNs showed that they perceived both the SMILeApp and the monitoring interface as very intuitive and easy to use. Suggestions for improvement included adding more symptoms (e.g., itching), a medication intake reminder and a change of color, as well as to use the formal forms of German address (Sie/Ihr/etc.) in the SMILeApp.
## Interdisciplinary clinical and scientific steering group meetings
From June to December 2020, interdisciplinary clinical steering group meetings were held every 4 to 8 weeks. The participants included the clinical expert group (6 APNs), the head nurse of hematology, higher-level nursing management from the Department of Internal Medicine and two senior Hematology physicians (see Figure 3). These meetings had two aims: based on presentations and discussions of the progress and information gathered from project group meetings, to make joint interdisciplinary decisions on modification and setting-specific adjustments; and to support access to infrastructure, such as allocating telephones and workstations for the planned SMILe-APN function. Further, implementation strategies were discussed and tailored to the Swiss setting based on ERIC guidelines.
**Figure 3:** *The adapted SMILe-ICM for the Swiss setting. Note. Elements of the original SMILe-ICM highlighted in green have been adapted to the Swiss setting. Elements of SMILe-ICM highlighted in red has been added as new functions. alloSCT, allogeneic stem cell transplantation; APN, Advanced Practice Nurse; eCCM,eHealth-enhanced Chronic Care Model.*
## Interdisciplinary scientific steering group meetings
Throughout the phases of the SMILe project, the USB clinical project group leaders (the co-PI SV and JR, who hold joint appointments as APNs in the USB and respectively as a postdoctoral and a PhD student at the university) participated in regular project meetings with their SMILe research team (the PI SDG and other co-PIs, i.e., the developers LL and AT) and scientific team members, every two to four weeks. Within these meetings, project adaptation progress was presented and discussed, joint decisions on adaptations made, and contextually adapted implementation strategies chosen. These meetings also focused on strategic project-related decisions regarding study planning (e.g., third-party funding, preparation of study materials, ethical approval, study registration).
## Adapting SMILe-ICM according to FRAME
The integration of the above-described contextual analysis results, the decisions of the project group meetings, the setting-specific SOPs, the higher-level regulatory scenarios (e.g., Medical Device Regulation) and the usability test results all informed the adaptation of the SMILe-ICM as recommended by FRAME [9]. A summary of these processes is available in a Meta-matrix (Table 4). The following paragraphs describe the elements adapted or added to the original SMILe-ICM for the Swiss setting, which are summarized in Figure 3.
## Timepoint and planning of adaptations
Following FRAME guidelines, all modifications were proactively planned and executed before implementing the SMILe-ICM in the new setting.
## Involved decision makers and aims regarding SMILe-ICM adaptations
As outlined in Table 3, the interdisciplinary clinical and scientific steering group members, as well as key stakeholders (IT team members, lawyers) made joint decisions as to which SMILe intervention components had to be adapted to fit the USB setting. Core components were kept, but their fit improved regarding the needs of both end user groups (i.e., patients and APNs), existing practice patterns (i.e., care processes) and overall feasibility.
## Contextual adaptations for the human component of intervention delivery
As summarized in Table 4, the SMILe-ICM has been contextually adapted due to meso-level (i.e., organizational) and macro-level (i.e., Swiss legal principles) differences between the centers while maintaining its core components. Due to different contact points both pre-alloSCT (d-14 at Freiburg vs. d-10 at USB) and post-discharge, the timepoints of delivery had to be adapted to the Swiss setting.
## Contextual adaptations to the technology part
SMILeApp lexicon information and contact details have been adapted based on USB Hematology Department SOPs [69] and in consultation with senior physicians and clinical experts (i.e., psycho-oncologist and dietician). According to *Swiss data* protection regulations [70], data protection management also required adaptations. Additionally, based on the end-user test results, the salutation pronouns used in the SMILeApp have been changed from informal (“Du”) to formal (“Sie”); and based on the usability test results, in consultation with technology developers, the technology components have been partly adjusted (e.g., the color of the SMILeApp background).
## Context-specific extension for the technology part
Because of the need for SMILeApp to be usable on iPhones [almost half of Swiss mobile telephones [77]], an iOS version was added. Further, based on the results of our contextual analysis and end-user tests, and in line with the first participating center [41, 43], which revealed the wish to add an electronic medication management module as well as an overview of values entered into the app, these technological functionalities have been added to the SMILeApp and its monitoring interface.
## Skipping planned elements
According to Swiss and international regulations [71, 79], certification of the SMILeApp as a medical device was not yet possible for the first participating center [41, 43] and also not yet feasible for the Swiss setting due to limited financial resources. Therefore, neither an already-developed [41, 43] automated feedback system for user-entered values, an automatically updated medication plan for patients, nor color-highlighting of conspicuous values based on predefined cut-offs in the monitoring interface could be realized in this iteration of the app.
## Tailoring implementation strategies to the Swiss setting
Based on our synthesis of the key contextual findings and the integration of project group adaptation process results, implementation strategies were tailored to the Swiss setting and classified congruently with the categories used within the overall SMILe project's pre-phase, Phase A (development & adaptation) and Phase B (implementation & evaluation) and sustainment.
As Table 4 indicates, we have chosen seventeen of the 73 ERIC implementation strategies [30]. Specifically for Phase A, an initial local needs assessment, as well as creation of partnerships and involvement of local opinion leaders have been recognized as essential (see Table 4 for all chosen implementation strategies). Especially regarding adaptation, it became evident that visiting other sites, adaptation and tailoring to the Swiss context and the organization of clinical implementation teams are all crucial to a context-specific implementation. In combination with this context's low level of chronic illness management, the perceived requirements for new clinical roles (i.e., context-specific APN training) postulated the creation of new clinical teams and conduction of educational meetings.
Further implementation strategies were formulated to support the project's Phase B [i.e., implementation, evaluation [45]]. In addition to ongoing consensus discussion and informing of local opinion leaders, provision of clinical supervision, and provision of local technical assistance, these included provision of reminders for clinicians and dissemination of clinical innovation. Merging contextual analysis results with stakeholder involvement discussion outcomes revealed that ongoing exchanges, supervisions and support are all needed to ensure the SMILe-ICM's implementation and long-term sustainability.
## Discussion
With this mixed-methods study, we aimed to contribute to the understanding on how to contextually adapt complex interventions and tailor implementation strategies, what is so far understudied in the field of implementation science [6]. Our elaborated, step-by-step methodological approach combines a theory-driven contextual analysis [48] with the in-depth, theoretically framed [9] adaptation process of a complex intervention, explaining how to tailor its implementation strategies based on recommendations from implementation science methodology [30] for any context in real-world settings.
## Reflections and implications from our step-wise approach
In recent years, context has become a central concept in adapting health care interventions [80, 81]. In our first step, a methodologically grounded contextual analysis following BANANA [48] supported our understanding of current clinical practice patterns, as well as of clinicians' and patients' views on their needs and technology openness. BANANA had already proved useful for the contextual analysis of our first participating center (Freiburg, Germany) [41]. It also helped us to conduct a profound evaluation of contextual aspects in the current study (in Basel) [48]. While our contextual analysis confirmed the need, wish and openness to implement the first SMILe–ICM [42, 43] into the Swiss setting, its results revealed predominantly similar findings to those of the Freiburg setting [41]: current clinical practice was rather acute-care-driven, with a similar PACIC rating [Basel = 30.6, Freiburg = 32.6 [41]] and CIMI BRIGHT scores [Basel = 2.92, Freiburg = 2.74 [41]]. As in the Freiburg setting [41], patients highlighted a wish to have continuous self-management support across the entire patient pathway. This is in line with previous evidence calling for patient-centered, continued care coordination, especially for the complex posttransplant transition phase—the period during which patients are transferred from full in-patient support to full individual responsibility for self-management tasks at home (82–84). Concerning eHealth openness, this study's contextual analysis results were quite similar to those obtained for Freiburg [41], and are congruent with evidence (85–87) supporting cancer patients' openness to eHealth-applications. I.e., patients support eHealth as long as personal contact is maintained. Considering that most existing technology's efficacy depends largely on its link to timely and personal health care provider responses, evidence supports the planned integration of eHealth and human support (18, 19, 85–87). In turn, this may lead to closer patient involvement in shared decision-making processes, as well as to faster communication and therefore to increased overall satisfaction (86, 88–91).
However, within our second step—the adaption phase, i.e., the in-depth exchange with key stakeholders merged with information from the meso- [i.e., setting specific SOPs [69]] and macro-levels [i.e., Medical Device Regulation [71], *Swiss data* protection regulations [70] and cantonal data protection law [78]]—we discovered considerable differences between the centers' practice patterns and organizational structures. Such variability in practice patterns at alloSCT centers is also described in the literature—although the related information is still limited and based primarily on evidence from the U.S. (92–96). Still, with our first step, i.e., a context-specific needs assessment and the exploration of clinicians' and patients' perspectives regarding current practice patterns in the target setting, we obviously missed a crucial component: information about the institutional procedures and how these differ between settings. Details of these procedures were important to inform the adaptation of our eHealth-facilitated ICM. Based on our results, then, we would suggest that even in the first steps, differences between the contexts should be mapped in terms of meso-level information (i.e., characterization of usual practice patterns) and macro level specifications (i.e., legal requirements). Ideally, detailed information on the selected intervention, its essential components and important delivery modes should be obtained in advance. This could then be combined with an exploration of targeted meso- and macro-level information by involving diverse key stakeholders, e.g., clinical experts, policy stakeholders and potential end-users, as early as possible [25] in focus group or individual interviews. This step could lead into a tightly-defined contextual analysis with targeted research questions chosen to inform the open needs of the adaptation and implementation phases, i.e., to shorten the investigations on the individual patient and clinician level by collecting better-targeted information and focusing more on meso- and macro-level aspects as it is also suggested by the recently published ADAPT guidance [76]: Even in their first step, in order to minimize the necessary time and personnel expenditure for a context-specific adaptation, the authors suggest mapping similarities and differences between original and new contexts [76].
In terms of providing a foundation to facilitate the acceptability and sustainability of developed or adapted intervention in new contexts, stakeholder involvement is rapidly gaining acceptance as an indispensable tool [8, 34]. Our in-depth stakeholder involvement was also central to our intervention adaptation (step 2): Throughout the adaptation phase, our stakeholder group meetings brought diverse perspectives on how to adapt our eHealth-facilitated ICM. To meet the needs of clinicians, patients, and researchers, every available perspective was necessary.
Indeed, our adaptation process, including its in-depth initial contextual analysis and broad stakeholder involvement, took us twenty-one months to complete. While previous research suggests allowing 6–9 months to adapt an intervention [4, 8, 25, 81, 97], no systematic overview and comparison of the time and effort it takes to adapt complex health care interventions yet exists. Such an overview would allow a clear record of the adaptation process's duration and effort. However, to allow cross-study comparison of these variables, it would be necessary to consider the investment not only of time, but also of personnel (e.g., percentage of staff, number of employees). Such details are even less available in the literature than those regarding time investment. To lower the cost, the literature also discusses “rapid methods” to adapt and optimize an intervention in a fast-changing context [98, 99]. To inform quick adaptation and optimization of behavioral interventions in evolving public health contexts, within a short timeframe of 1–2 months, Morton et al. [ 100] used rapid methods to modify the online “Germ Defence” intervention [101]. However, it cannot be assumed that every complex intervention can be rapidly adapted and implemented in every new context: every new context's norms, resources, and delivery structures differ from those of the original [99]. Especially regarding the adaptation and implementation of eHealth-facilitated ICMs, evidence suggests that, to understand the complex adaptation processes used in implementation science projects, health information technology adaptation research should apply multilevel and multidimensional evaluation instruments [99]. Consistent with our study design, mixed-methods approaches that involve key stakeholders to explore the dynamic relationships between technology and social factors are needed [102, 103].
While stakeholder inclusion is an adaptation process that helps ensure that the approach is appropriate, feasible, and acceptable [104], it is also time-, resource- and effort-intensive [31, 105]. For interventions in chaotic, resource-competitive clinical settings—particularly compared to the controlled environment of a research setting—including stakeholders is especially difficult [99, 106]: numerous environmental factors (especially limited time and funding) can act as barriers to participation in clinical research projects [107, 108]. For example, for many stakeholders, involvement requires skills and competences both to present their own opinions and interests and to argue for or against those of others [31, 105]. However, within adaptation guidance papers or studies, reflections on how to deal with differences of opinion are scarce [8]. For our approach, a combination of favorable factors—including strong joint leadership engagement, both from the clinical nursing management and from the research infrastructure staff, a well-established academic-practice partnership between the clinical setting and the research institute [109] and the fact that implementers worked partly in clinical settings and partly in academic institutions—have all been identified as strong facilitators for our adaptation and implementation process [45]. Moreover, this dynamic combination of people and competencies supported us in realizing a shared vision and commitment among interprofessional stakeholders. And while clinical stakeholders bring in-depth clinical expertise and setting-specific knowledge to the table, academic partners often provide access to funding sources and to expert researchers [110, 111].
However, such resource-intensive multiple-methods study designs with in-depth involvement of key stakeholders commonly suffer from limited funding possibilities [105], which can impede widespread adoption for even the most well-positioned innovations (112–115). Therefore, specific funder commitments must be secured to adequately support such projects—not only to the stage of clinical trials, but all through the adaptation and implementation phases [31, 116] to ensure sustainability [99, 117]. In terms of lowering costs and efforts, another emerging opportunity to support complex eHealth adaptation and implementation studies is the application of big routine data sets [99, 118]. Automated processing of such data (e.g., hospitals' electronic health records) or socio-economic figures (e.g., national-level statistics on age, education, occupation, income) could quickly distill macro-level information from regularly-updated data on large, diverse populations at low cost [119]. As part of our third step, FRAME provided guidance on how to report adaptations and modifications, including several cursory indications of which aspects should be considered in the adaptation process (e.g., who participates in the decision to modify, what should be modified) [9]. Consistent with previous studies applying FRAME's (120–122) coding structure, FRAME [9] was particularly useful for the structured, step-by-step tracking of adaptations to our eHealth-facilitated ICM. And from the earliest stages of our adaptation and implementation process, ERIC recommendations [30] supported us both to choose and to describe context-specific implementation strategies. Since this project's launch, a new framework for documenting adaptations to implementation strategies has been released—the 2021 Framework for Reporting Adaptations and Modifications to Evidence-based Implementation Strategies (FRAME-IS) [123]. Developed on the basis of the existing FRAME [9], elements of FRAME-IS [123] closely mirror those of the original [9], but with language specifying, for example, that modifications are made to implementation strategies, not to the overall intervention. While the authors suggest using the ERIC compilation [30] within the FRAME-IS, this tool provides guidance not only to choose implementation strategies from ERIC's offerings [30], but also to document and justify modifications to implementation strategies. These steps could be supportive of any further adaptation and implementation science projects. In line with recently published recommendations [124] suggesting to identify multiple barriers that can support a structured approach to choose a wide range of implementation strategies supported also the tailoring of our implementation strategies in addition to the stakeholder involvement and the application of theoretical frameworks for categorization (i.e., EPIS [125] and CICI [12] framework).
## Strengths and limitations
This study's most notable strength is its comprehensive multilevel mixed-methods approach [37, 38], which facilitated the gathering, merging and interpretation of context-specific qualitative and quantitative information. Further, our in-depth adaptation process not only informed the theoretically framed adaptations (following FRAME) [9], but also enabled us to tailor implementation strategies to the new context following ERIC compilations [30]. This innovative approach bridges important gaps in terms both of clinical innovation for complex, eHealth-facilitated care delivery and of intervention methodology.
Despite such promising elements, this study has certain limitations. First, realities of clinical practice, e.g., changes in personnel and/or leadership and especially the unforeseen crisis of the COVID-19 pandemic can affect the success of any implementation. Therefore, an ongoing analysis of practice patterns and change, which was not systematically planned within our approach, will be needed. However, within our implementation and evaluation phase [45], we have continued to conduct regular stakeholder group meetings with the involved clinical management, the research steering group and the APNs: this belongs to our implementation strategies for Phase B (i.e., ongoing consensus discussions and information from local opinion leaders, provide clinical supervision, provide local technical assistance, remind clinicians, spread clinical innovation).
Second, due to COVID-19 pandemic regulations, we were forced to conduct the usability tests (step 2) in a virtual environment. Due to this adverse condition, the quantitative results have to be considered with caution. Nevertheless, the qualitative information from the think-aloud part and subsequent consultations with patients and APNs gave us insightful and repetitive information on adapting the technological component of the SMILe-ICM.
Limitations of our human and time resources prevented the completion of the digitalization process as originally planned. According to the Medical Device Regulation introduced in May 2020 [71], the SMILeApp could not yet be classified as a Medical Device [126]. However, while we are currently evaluating the eHealth-facilitated ICM in our two participating centers [45], we are confident that the methods used will increase the probability of sustainable implementation and acceptance in real-world clinical practice.
## Implications for research, practice, and policy
Adapting a health care intervention that integrates eHealth-facilitated components is a complex undertaking. All tailoring must focus specifically on the target context and a local implementation [99]. For this purpose, we have developed a step-wise mixed-methods adaptation approach that features strong stakeholder involvement that is in line with the recently-published ADAPT guidance [25],. Oriented towards a comprehensive knowledge of key contextual factors, this approach will help to operationalize and structure adaptation processes, facilitating shared understanding and informed decision-making (31–33). For example, as this study's results indicate that critical differences were predominantly at the macro level rather than at that of patients and clinicians, directing our first step at that level—mapping contextual differences between the two test centers with a focus on macro-level factors (e.g., differences in existing care processes along the patient pathway)—would have saved time and resources.
Additional research is warranted to explore changes in contextual factors using mixed-methods evaluations across multiple data collection periods. For example, in addition to quantitative data monitoring, regular focus group meetings with key stakeholders could identify critical contextual developments—particularly issues arising in the adapted intervention's implementation or execution—at an early stage [102, 103].
In light of practical and policy implications, establishing academic-service partnerships and involving key stakeholders throughout the adaptation process are key implementation strategies for an implementation science adaptation project; therefore, these should be considered and planned from the beginning [8, 34]. Convincing policymakers and funders to adequately support in-depth adaptation processes and the contextual analyses that guide them, more studies will need to report transparently on the related processes, their timelines and their long-term value [31, 116]. Overcoming barriers to the adaption and adoption of eHealth-facilitated care, particularly the currently prohibitive Medical Device Regulation`s requirements and reimbursement policies [24], will require key changes in policy priorities, beginning with a system-level definition of innovation, a clear overarching mission and efficiently-aligned funding structures [99, 117].
## Conclusion
This study describes our development of step-wise implementation-science-based methodological approach on how to conduct a context-driven, theoretically framed intervention adaptation, including the tailoring of its implementation strategies to the target setting. As this is an example of implementation science methodology, its underlying principles—particularly its end-user focus, its extensive use of stakeholder involvement, and the value it places on contextual knowledge—can be applied directly both to study environments and to real-life settings. Further, it is the product of a multidisciplinary effort led by both clinicians and researchers. To adapt and implement a complex intervention into daily clinical practice, the study project employed the perspectives of clinical health care professionals, IT team members and patients.
Our experience emphasizes that a contextual analysis is essential to understand current needs, practice patterns and openness towards an intervention's implementation. However, in light of our discovery that certain meso- and macro level differences were actually critical to adaptation, we strongly recommend that the contextual analysis include information on both levels (i.e., exploring contextual differences) from the beginning. This can be done by discussing relevant issues with key stakeholders and clinical experts, then using their input to map out targeted meso-and macro-level differences or similarities from the beginning.
In spite of having to correct for this omission, our approach finally supported a smooth implementation (i.e., with high acceptability and adoption) of the adapted eHealth-facilitated ICM into practice, thereby increasing the likelihood of its sustainability. Therefore, we conclude that a well-planned, theory-guided, contextually tailored adaptation phase provides a vital step towards an intervention's successful, sustainable implementation. As this relies strongly on a comprehensive knowledge of the target context, an in-depth contextual analysis should be incorporated and budgeted in every implementation science project.
## Data availability statement
The original contributions presented in the study are included in the article/supplementary materials. Further inquiries can be directed to the corresponding author.
## Ethics statement
The studies involving human participants were reviewed and approved by Ethics Committee Northwest and Central Switzerland (EKNZ; BASEC 2019-00307). The patients/participants provided their written informed consent to participate in this study.
## Author contributions
SDG, SV, LL and AT conceived, designed, supervised and obtained the funding for this study. SV, LL, JR, JM, AT, SDG contributed to its conceptual, theoretical-methodological elaboration. JR, KK, VWB, AT and SV collected the data and conducted the quantitative and qualitative data analyses: while KK and VWB mainly conducted the quantitative data collection at University Hospital Basel and University Hospital Zurich (will be published elsewhere), JR and SV collected and analyzed the qualitative data. AT and her SMILe technology team members from the University of Applied Science Augsburg (DE) collected and analyzed the end-user test data. This step was supported by SV and JR. Working with the involved clinicians and stakeholders, SV, JR, AT, KK, SG and FG planned and conducted the SMILe-ICM's contextual adaptation and tailored its implementation strategies as necessary. SV, JR, LL, JM and AT wrote the first draft of the manuscript, which was initially supervised by SDG. All authors critically revised the manuscript for important intellectual content, contributed to its revision, and read and approved the submitted version. The corresponding senior author, SDG, had final responsibility for the decision to submit this paper for publication. All authors contributed to the article and approved the submitted version.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frhs.2022.977564/full#supplementary-material.
## References
1. Evans RE, Craig P, Hoddinott P, Littlecott H, Moore L, Murphy S. **When and how do “effective” interventions need to be adapted and/or re-evaluated in new contexts? The need for guidance**. *J Epidemiol Community Health* (2019.0) **73** 481-2. DOI: 10.1136/jech-2018-210840
2. Craig P, Di Ruggiero E, Frohlich KL, Mykhalovskiy E, White M. (2018.0)
3. Rogers EM. *Diffusion of innovations* (1995.0)
4. Evans RE, Moore G, Movsisyan A, Rehfuess E. **How can we adapt complex population health interventions for new contexts? Progressing debates and research priorities**. *J Epidemiol Community Health* (2021.0) **75** 40-5. DOI: 10.1136/jech-2020-214468
5. Escoffery C, Lebow-Skelley E, Haardoerfer R, Boing E, Udelson H, Wood R. **A systematic review of adaptations of evidence-based public health interventions globally**. *Implement Sci* (2018.0) **13** 125. DOI: 10.1186/s13012-018-0815-9
6. Escoffery C, Lebow-Skelley E, Udelson H, Böing EA, Wood R, Fernandez ME. **A scoping study of frameworks for adapting public health evidence-based interventions**. *Transl Behav Med* (2019.0) **9** 1-10. DOI: 10.1093/tbm/ibx067
7. Movsisyan A, Arnold L, Evans R, Hallingberg B, Moore G, O’Cathain A. **Adapting evidence-informed complex population health interventions for new contexts: a systematic review of guidance**. *Implement Sci* (2019.0) **14** 1-20. DOI: 10.1186/s13012-019-0956-5
8. Movsisyan A, Arnold L, Copeland L, Evans R, Littlecott H, Moore G. **Adapting evidence-informed population health interventions for new contexts: a scoping review of current practice**. *Health Res Policy Syst* (2021.0) **19** 13. DOI: 10.1186/s12961-020-00668-9
9. Stirman SW, Baumann AA, Miller CJ. **The FRAME: an expanded framework for reporting adaptations and modifications to evidence-based interventions**. *Implement Sci* (2019.0) **14** 58. DOI: 10.1186/s13012-019-0898-y
10. McKleroy VS, Galbraith JS, Cummings B, Jones P, Harshbarger C, Collins C. **Adapting evidence-based behavioral interventions for new settings and target populations**. *AIDS Educ Prev* (2006.0) **18** 59-73. DOI: 10.1521/aeap.2006.18.supp.59
11. Moore JE, Bumbarger BK, Cooper BR. **Examining adaptations of evidence-based programs in natural contexts**. *J Prim Prev* (2013.0) **34** 147-61. DOI: 10.1007/s10935-013-0303-6
12. Pfadenhauer LM, Gerhardus A, Mozygemba K, Lysdahl KB, Booth A, Hofmann B. **Making sense of complexity in context and implementation: the context and implementation of Complex interventions (CICI) framework**. *Implement Sci* (2017.0) **12** 21. DOI: 10.1186/s13012-017-0552-5
13. Skivington K, Matthews L, Simpson SA, Craig P, Baird J, Blazeby JM. **A new framework for developing and evaluating complex interventions: update of medical research council guidance**. *Br Med J* (2021.0) **374** n2061. DOI: 10.1136/bmj.n2061
14. Hawe P. **Lessons from complex interventions to improve health**. *Annu Rev Public Health* (2015.0) **36** 307-23. DOI: 10.1146/annurev-publhealth-031912-114421
15. Lobb R, Colditz GA. **Implementation science and its application to population health**. *Annu Rev Public Health* (2013.0) **34** 235-51. DOI: 10.1146/annurev-publhealth-031912-114444
16. Cambon L, Terral P, Alla F. **From intervention to interventional system: towards greater theorization in population health intervention research**. *BMC Public Health* (2019.0) **19** 339. DOI: 10.1186/s12889-019-6663-y
17. 17.World Health Organization. Ehealth at WHO Geneva. Switzerland: eHealth Unit (2020). Available at:
https://www.who.int/ehealth/about/en/
(Accessed March 11, 2020).. *Ehealth at WHO Geneva* (2020.0)
18. Aapro M, Bossi P, Dasari A, Fallowfield L, Gascón P, Geller M. **Digital health for optimal supportive care in oncology: benefits, limits, and future perspectives**. *Support Care Cancer*. DOI: 10.1007/s00520-020-05539-1
19. Warrington L, Absolom K, Conner M, Kellar I, Clayton B, Ayres M. **Electronic systems for patients to report and manage side effects of cancer treatment: systematic review**. *J Med Internet Res* (2019.0) **21** e10875. DOI: 10.2196/10875
20. Jeffs E, Vollam S, Young JD, Horsington L, Lynch B, Watkinson PJ. **Wearable monitors for patients following discharge from an intensive care unit: practical lessons learnt from an observational study**. *J Adv Nurs* (2016.0) **72** 1851-62. DOI: 10.1111/jan.12959
21. Simblett S, Greer B, Matcham F, Curtis H, Polhemus A, Ferrão J. **Barriers to and facilitators of engagement with remote measurement technology for managing health: systematic review and content analysis of findings**. *J Med Internet Res* (2018.0) **20** e10480. DOI: 10.2196/10480
22. Thies K, Anderson D, Cramer B. **Lack of adoption of a Mobile app to support patient self-management of diabetes and hypertension in a federally qualified health center: interview analysis of staff and patients in a failed randomized trial**. *JMIR Human Factors* (2017.0) **4** e24. DOI: 10.2196/humanfactors.7709
23. Aitken M, Clancy B, Nass D. *The growing value of digital health: Evidence and impact on human health and the healthcare system* (2017.0)
24. Aitken M, Ness D. *Digital health trends 2021* (2021.0)
25. Moore G, Campbell M, Copeland L, Craig P, Movsisyan A, Hoddinott P. **Adapting interventions to new contexts-the ADAPT guidance**. *Br Med J* (2021.0) **374** n1679. DOI: 10.1136/bmj.n1679
26. Eccles MP, Mittman BS. **Welcome to implementation science**. *Implement Sci* (2006.0) **1** 1. DOI: 10.1186/1748-5908-1-1
27. Damschroder LJ, Waltz TJ, Abadie B, Powell BJ. *Choosing implementation strategies to address local contextual barriers* (2017.0)
28. Watson DP, Adams EL, Shue S, Coates H, McGuire A, Chesher J. **Defining the external implementation context: an integrative systematic literature review**. *BMC Health Serv Res* (2018.0) **18** 209. DOI: 10.1186/s12913-018-3046-5
29. Rogers L, De Brún A, McAuliffe E. **Defining and assessing context in healthcare implementation studies: a systematic review**. *BMC Health Serv Res* (2020.0) **20** 591. DOI: 10.1186/s12913-020-05212-7
30. Powell BJ, Waltz TJ, Chinman MJ, Damschroder LJ, Smith JL, Matthieu MM. **A refined compilation of implementation strategies: results from the expert recommendations for implementing change (ERIC) project**. *Implement Sci* (2015.0) **10** 21. DOI: 10.1186/s13012-015-0209-1
31. Boaz A, Hanney S, Borst R, O'Shea A, Kok M. **How to engage stakeholders in research: design principles to support improvement**. *Health Res Policy Syst* (2018.0) **16** 60. DOI: 10.1186/s12961-018-0337-6
32. Hudson JL, Moon Z, Hughes LD, Moss-Morris R, Hamilton K, Cameron LD, Hagger MS, Hankonen N, Lintunen T. **Engagement of stakeholders in the design, evaluation, and implementation of Complex interventions**. *The handbook of behavior change* (2020.0) 349-60
33. Martínez J, Piersol CV, Lucas K, Leland NE. **Operationalizing stakeholder engagement through the stakeholder-centric engagement charter (SCEC)**. *J Gen Intern Med* (2022.0) **37** 105-8. DOI: 10.1007/s11606-021-07029-4
34. Arwal SH, Aulakh BK, Bumba A, Siddula A. **Learning by doing in practice: a roundtable discussion about stakeholder engagement in implementation research**. *Health Res Policy Syst* (2017.0) **15** 105. DOI: 10.1186/s12961-017-0275-8
35. Wendler MC. **Triangulation using a meta-matrix**. *J Adv Nurs* (2001.0) **35** 521-5. DOI: 10.1046/j.1365-2648.2001.01869.x
36. Waltz TJ, Powell BJ, Chinman MJ, Smith JL, Matthieu MM, Proctor EK. **Expert recommendations for implementing change (ERIC): protocol for a mixed methods study**. *Implement Sci* (2014.0) **9** 39. DOI: 10.1186/1748-5908-9-39
37. Creswell JW. *A concise introduction to mixed methods research* (2021.0)
38. Guetterman TC, Fetters MD. **Two methodological approaches to the integration of mixed methods and case study designs: a systematic review**. *Am Behav Sci* (2018.0) **62** 900-18. DOI: 10.1177/0002764218772641
39. Yin RK. *Case study research and applications: design and methods* (2018.0)
40. Crowe S, Cresswell K, Robertson A, Huby G, Avery A, Sheikh A. **The case study approach**. *BMC Med Res Methodol* (2011.0) **11** 100. DOI: 10.1186/1471-2288-11-100
41. Leppla L, Mielke J, Kunze M, Mauthner O, Teynor A, Valenta S. **Clinicians and patients perspectives on follow-up care and eHealth support after allogeneic hematopoietic stem cell transplantation: a mixed-methods contextual analysis as part of the SMILe study**. *Eur J Oncol Nurs* (2020.0) **45** 101723. DOI: 10.1016/j.ejon.2020.101723
42. Leppla L, Hobelsberger S, Rockstein D, Werlitz V, Pschenitza S, Heidegger P. **Implementation science meets software development to create eHealth components for an integrated care model for allogeneic stem cell transplantation facilitated by eHealth: the SMILe study as an example**. *J Nurs Scholarsh* (2021.0) **53** 35-45. DOI: 10.1111/jnu.12621
43. Leppla L, Schmid A, Valenta S, Mielke J, Beckmann S, Ribaut J. **Development of an integrated model of care for allogeneic stem cell transplantation facilitated by eHealth-the SMILe study**. *Support Care Cancer* (2021.0) **29** 8045-57. DOI: 10.1007/s00520-021-06328-0
44. Ribaut J, Leppla L, Teynor A, Valenta S, Dobbels F, Zullig LL. **Theory-driven development of a medication adherence intervention delivered by eHealth and transplant team in allogeneic stem cell transplantation: the SMILe implementation science project**. *BMC Health Serv Res* (2020.0) **20** 827. DOI: 10.1186/s12913-020-05636-1
45. De Geest S, Valenta S*, Ribaut J, Gerull S, Mielke J, Simon M. **The SMILE integrated care model in allogeneic SteM cell transplantatIon faciLitated by eHealth: a protocol for a hybrid effectiveness-implementation randomised controlled trial**. *BMC Health Serv Res* (2022.0) **22**. DOI: 10.1186/s12913-022-08293-8
46. Gee PM, Greenwood DA, Paterniti DA, Ward D, Miller LMS. **The eHealth enhanced chronic care model: a theory derivation approach**. *J Med Internet Res* (2015.0) **17** e86. DOI: 10.2196/jmir.4067
47. Michie S, van Stralen MM, West R. **The behaviour change wheel: a new method for characterising and designing behaviour change interventions**. *Implement Sci* (2011.0) **6**. DOI: 10.1186/1748-5908-6-42
48. Mielke J, Leppla L, Valenta S, Zullig L, Zúñiga F, Staudacher S. **Unravelling implementation context: the Basel approach for coNtextual ANAlysis (BANANA) in implementation science and its application in the SMILe project**. *Implement Sci Commun* (2022.0) **3** 102. DOI: 10.1186/s43058-022-00354-7
49. Berben L, Denhaerynck K, Dobbels F, Engberg S, Vanhaecke J, Crespo-Leiro MG. **Building research initiative group: chronic illness management and adherence in transplantation (BRIGHT) study: study protocol**. *J Adv Nurs* (2015.0) **71** 642-54. DOI: 10.1111/jan.12519
50. Denhaerynck K, Berben L, Dobbels F, Russell CL, Crespo-Leiro MG, Poncelet AJ. **Multilevel factors are associated with immunosuppressant nonadherence in heart transplant recipients: the international BRIGHT study**. *Am J Transplant*. DOI: 10.1111/ajt.14611
51. Vanhoof JM, Vandenberghe B, Geerts D, Philippaerts P, De Mazière P, DeVito Dabbs A. **Shedding light on an unknown reality in solid organ transplant patients’ self-management: a contextual inquiry study**. *Clin Transplant*. DOI: 10.1111/ctr.13314
52. Vanhoof JM, Vandenberghe B, Geerts D, Philippaerts P, Mazière P, DeVito Dabbs A. **Technology experience of solid organ transplant patients and their overall willingness to use interactive health technology**. *J Nurs Scholarsh* (2017.0) **50** 1-12. DOI: 10.1111/jnu.12362
53. Ochieng Nyumba T, Wilson K, Derrick CJ, Mukherjee N. **The use of focus group discussion methodology: insights from two decades of application in conservation**. *Methods Ecol Evol* (2018.0) **9** 20-32. DOI: 10.1111/2041-210X.12860
54. Ali SH, Merdjanoff AA, Parekh N, DiClemente RJ. **Development of an integrated approach to virtual mind-mapping: methodology and applied experiences to enhance qualitative health research**. *Qual Health Res* (2022.0) **32** 571-80. DOI: 10.1177/10497323211058161
55. Burgess-Allen J, Owen-Smith V. **Using mind mapping techniques for rapid qualitative data analysis in public participation processes**. *Health Expect* (2010.0) **13** 406-15. DOI: 10.1111/j.1369-7625.2010.00594.x
56. Urbanek S, Plummer M. *The R project for statistical computing Vienna university of economics and business* (2022.0)
57. Braun V, Clarke V, Hayfield N, Terry G, Liamputtong P. **Thematic analysis**. *Handbook of research methods in health social sciences* (2019.0) 843-60
58. 58.ATLAS.ti Scientific Software Development GmbH. ATLAS.Ti—discover insights for your research projects. Berlin, Germany (2019). Available at:
https://atlasti.com/product/
(Accessed, March 11, 2020).. *ATLAS.Ti—discover insights for your research projects* (2019.0)
59. 59.Abras C, Maloney-Krichmar D, Preece J. User-centered design. In: Bainbridge W, editor. Encyclopedia of Human-Computer Interaction. Massachusetts, US: Berkshire Publishing Group (2004). p. 445–56.
60. 60.Cockton G, Lárusdóttir M, Gregory P, Cajander Å. Integrating user-centred design in agile development. Switzerland: Springer (2016).
61. Farao J, Malila B, Conrad N, Mutsvangwa T, Rangaka MX, Douglas TS. **A user-centred design framework for mHealth**. *PloS one* (2020.0) **15** e0237910. DOI: 10.1371/journal.pone.0237910
62. da Silva TS, Martin A, Maurer F, Silveira M. (2011.0)
63. Beck K, Andres C. *Extreme programming explained: embrace change* (2004.0)
64. Schwaber K. *Agile project management with scrum* (2004.0)
65. De Vito Dabbs A, Myers BA, Mc Curry KR, Dunbar-Jacob J, Hawkins RP, Begey A. **User-centered design and interactive health technologies for patients**. *Comput Inform Nurs* (2009.0) **27** 175-83. DOI: 10.1097/NCN.0b013e31819f7c7c
66. Stirman SW, Miller CJ, Toder K, Calloway A. **Development of a framework and coding system for modifications and adaptations of evidence-based interventions**. *Implement Sci* (2013.0) **8** 65. DOI: 10.1186/1748-5908-8-65
67. Waltz TJ, Powell BJ, Matthieu MM, Damschroder LJ, Chinman MJ, Smith JL. **Use of concept mapping to characterize relationships among implementation strategies and assess their feasibility and importance: results from the expert recommendations for implementing change (ERIC) study**. *Implement Sci* (2015.0) **10** 109. DOI: 10.1186/s13012-015-0295-0
68. Franco-Trigo L, Fernandez-Llimos F, Martínez-Martínez F, Benrimoj SI, Sabater-Hernández D. **Stakeholder analysis in health innovation planning processes: a systematic scoping review**. *Health Policy* (2020.0) **124** 1083-99. DOI: 10.1016/j.healthpol.2020.06.012
69. 69.University Hospital Basel. Hematology. Basel, Switzerland: (2022). Available at:
https://www.unispital-basel.ch/en/ueber-uns/departemente/theragnostik/kliniken/hematology/
(Accessed January 07, 2022).. *Hematology* (2022.0)
70. 70.Bösler B. DIN Standards Committee Medicine. Health informatics—information security management in health using. Berlin, Germany (2016). Available at:
https://www.din.de/en/getting-involved/standards-committees/named/publications/wdc-beuth:din21:259215227?destinationLanguage=&sourceLanguage=
(Accessed May 05, 2020).. *Health informatics—information security management in health using* (2016.0)
71. 71.European Commission. Medical Devices—Sector (2021). Available at:
https://ec.europa.eu/health/md_sector/new_regulations_en
(Accessed August 30, 2021).
72. Krug S. *Don't make me think, revisited: a common sense approach to web usability* (2014.0)
73. O’Cathain A, Murphy E, Nicholl J. **Three techniques for integrating data in mixed methods studies**. *Br Med J* (2010.0) **341**. DOI: 10.1136/bmj.c4587
74. Passweg JR, Baldomero H, Chabannon C, Basak GW, de la Cámara R, Corbacioglu S. **Hematopoietic cell transplantation and cellular therapy survey of the EBMT: monitoring of activities and trends over 30 years**. *Bone Marrow Transplant* (2021.0) **56** 1651-64. DOI: 10.1038/s41409-021-01227-8
75. Gerull S
76. Schmid CPR, Beck K, Kauer L, McGuire TG, van Kleef RC. **Health plan payment in Switzerland**. *Risk adjustment, risk sharing and premium regulation in health insurance markets* (2018.0) 453-89
77. Turulski A-S
78. 78.Kanton Basel-Stadt. SG 153.260—Gesetz über die Information und den Datenschutz (Informations- und Datenschutzgesetz, IDG) [Law on Information and Data Protection] Basel, Switzerland (2018). Available at:
https://www.gesetzessammlung.bs.ch/app/de/texts_of_law/153.260
(Accessed April 01, 2020).
79. Eidgenossenschaft S. *Verordnung des Schweizerischen Heilmittelinstituts über seine Gebühren (GebV-Swissmedic)* (2018.0)
80. Copeland L, Littlecott H, Couturiaux D, Hoddinott P, Segrott J, Murphy S. **The what, why and when of adapting interventions for new contexts: a qualitative study of researchers, funders, journal editors and practitioners’ understandings**. *PloS One* (2021.0) **16** e0254020. DOI: 10.1371/journal.pone.0254020
81. Haley AD, Powell BJ, Walsh-Bailey C, Krancari M, Gruß I, Shea CM. **Strengthening methods for tracking adaptations and modifications to implementation strategies**. *BMC Med Res Methodol* (2021.0) **21** 133. DOI: 10.1186/s12874-021-01326-6
82. Giaccone L, Felicetti F, Butera S, Faraci D, Cerrano M, Vici MD. **Optimal delivery of follow-up care after allogeneic hematopoietic stem-cell transplant: improving patient outcomes with a multidisciplinary approach**. *J Blood Med* (2020.0) **11** 141. DOI: 10.2147/JBM.S206027
83. Khera N, Martin P, Edsall K, Bonagura A, Burns LJ, Juckett M. **Patient-centered care coordination in hematopoietic cell transplantation**. *Blood Adv* (2017.0) **1** 1617-27. DOI: 10.1182/bloodadvances.2017008789
84. 84.National Academies of Sciences E, and Medicine. Exploring the state of the science of stem cell transplantation and posttransplant disability: proceedings of a workshop. Washington, DC: The National Academies Press (2022).. *Exploring the state of the science of stem cell transplantation and posttransplant disability: proceedings of a workshop* (2022.0)
85. Cox A, Lucas G, Marcu A, Piano M, Grosvenor W, Mold F. **Cancer survivors’ experience with telehealth: a systematic review and thematic synthesis**. *J Med Internet Res* (2017.0) **19** e11. DOI: 10.2196/jmir.6575
86. Mooney KH, Beck SL, Wong B, Dunson W, Wujcik D, Whisenant M. **Automated home monitoring and management of patient-reported symptoms during chemotherapy: results of the symptom care at home RCT**. *Cancer Med* (2017.0) **6** 537-46. DOI: 10.1002/cam4.1002
87. Basch E, Schrag D, Henson S, Jansen J, Ginos B, Stover AM. **Effect of electronic symptom monitoring on patient-reported outcomes among patients with metastatic cancer: a randomized clinical trial**. *JAMA*. DOI: 10.1001/jama.2022.9265
88. Linn AJ, Vervloet M, van Dijk L, Smit EG, Van Weert JC. **Effects of eHealth interventions on medication adherence: a systematic review of the literature**. *J Med Internet Res* (2011.0) **13** e103. DOI: 10.2196/jmir.1738
89. Basch E, Deal AM, Dueck AC, Scher HI, Kris MG, Hudis C. **Overall survival results of a trial assessing patient-reported outcomes for symptom monitoring during routine cancer treatment**. *JAMA* (2017.0) **318** 197-8. DOI: 10.1001/jama.2017.7156
90. Elbert NJ, van Os-Medendorp H, van Renselaar W, Ekeland AG, Hakkaart-van Roijen L, Raat H. **Effectiveness and cost-effectiveness of ehealth interventions in somatic diseases: a systematic review of systematic reviews and meta-analyses**. *J Med Internet Res* (2014.0) **16** e110. DOI: 10.2196/jmir.2790
91. Kuijpers W, Groen WG, Aaronson NK, van Harten WH. **A systematic review of web-based interventions for patient empowerment and physical activity in chronic diseases: relevance for cancer survivors**. *J Med Internet Res* (2013.0) **15** e37. DOI: 10.2196/jmir.2281
92. Majhail NS, Mau L-W, Chitphakdithai P, Denzen EM, Joffe S, Lee SJ. **Transplant center characteristics and survival after allogeneic hematopoietic cell transplantation in adults**. *Bone Marrow Transplant* (2020.0) **55** 906-17. DOI: 10.1038/s41409-019-0748-1
93. Majhail NS, Mau LW, Chitphakdithai P, Payton T, Eckrich M, Joffe S. **National survey of hematopoietic cell transplantation center personnel, infrastructure, and models of care delivery**. *Biol Blood Marrow Transplant* (2015.0) **21** 1308-14. DOI: 10.1016/j.bbmt.2015.03.020
94. Lee SJ, Astigarraga CC, Eapen M, Artz AS, Davies SM, Champlin R. **Variation in supportive care practices in hematopoietic cell transplantation**. *Biol Blood Marrow Transplant* (2008.0) **14** 1231-8. DOI: 10.1016/j.bbmt.2008.08.008
95. Baumgartner A, Bargetzi M, Bargetzi A, Zueger N, Medinger M, Passweg J. **Nutritional support practices in hematopoietic stem cell transplantation centers: a nationwide comparison**. *Nutrition* (2017.0) **35** 43-50. DOI: 10.1016/j.nut.2016.10.007
96. Wood WA, McGinn MK, Wilson D, Deal AM, Khera N, Shea TC. **Practice patterns and preferences among hematopoietic cell transplantation clinicians**. *Biol Blood Marrow Transplant* (2016.0) **22** 2092-9. DOI: 10.1016/j.bbmt.2016.07.014
97. Miller CJ, Wiltsey-Stirman S, Baumann AA. **Iterative decision-making for evaluation of adaptations (IDEA): a decision tree for balancing adaptation, fidelity, and intervention impact**. *J Community Psychol* (2020.0) **48** 1163-77. DOI: 10.1002/jcop.22279
98. Glasgow RE, Chambers D. **Developing robust, sustainable, implementation systems using rigorous, rapid and relevant science**. *Clin Transl Sci* (2012.0) **5** 48-55. DOI: 10.1111/j.1752-8062.2011.00383.x
99. Glasgow RE, Phillips SM, Sanchez MA. **Implementation science approaches for integrating eHealth research into practice and policy**. *Int J Med Inf* (2014.0) **83** e1-e11. DOI: 10.1016/j.ijmedinf.2013.07.002
100. Morton K, Ainsworth B, Miller S, Rice C, Bostock J, Denison-Day J. **Adapting behavioral interventions for a changing public health context: a worked example of implementing a digital intervention during a global pandemic using rapid optimisation methods**. *Front Public Health* (2021.0) **9** 668197. DOI: 10.3389/fpubh.2021.668197
101. Ainsworth B, Miller S, Denison-Day J, Stuart B, Groot J, Rice C. **Infection control behavior at home during the COVID-19 pandemic: observational study of a web-based behavioral intervention (germ defence)**. *J Med Internet Res* (2021.0) **23** e22197. DOI: 10.2196/22197
102. Cresswell KM, Sheikh A. **Undertaking sociotechnical evaluations of health information technologies**. *Inform Prim Care* (2014.0) **21** 78-83. DOI: 10.14236/jhi.v21i2.54
103. Yen P-Y, McAlearney AS, Sieck CJ, Hefner JL, Huerta TR. **Health information technology (HIT) adaptation: refocusing on the journey to successful HIT implementation**. *JMIR Med Inform* (2017.0) **5** e28. DOI: 10.2196/medinform.7476
104. 104.Movsisyan A, Arnold L, Copeland L, Evans R, Littlecott H, Moore G, et al. Adapting evidence-informed population health interventions for new contexts: a scoping review of current practice. Health Res Policy Syst. (2021) 19(1):13. 10.1186/s12961-020-00668-9. DOI: 10.1186/s12961-020-00668-9
105. Oliver K, Kothari A, Mays N. **The dark side of coproduction: do the costs outweigh the benefits for health research?**. *Health Res Policy Syst* (2019.0) **17** 33. DOI: 10.1186/s12961-019-0432-3
106. Chambers DA, Glasgow RE, Stange KC. **The dynamic sustainability framework: addressing the paradox of sustainment amid ongoing change**. *Implement Sci* (2013.0) **8** 117. DOI: 10.1186/1748-5908-8-117
107. Goldstein KM, Gierisch JM, Tucker M, Williams JW, Dolor RJ, Henderson W. **Options for meaningful engagement in clinical research for busy frontline clinicians**. *J Gen Intern Med* (2021.0) **36** 2100-4. DOI: 10.1007/s11606-020-06587-3
108. Wenke R, Noble C, Weir KA, Mickan S. **What influences allied health clinician participation in research in the public hospital setting: a qualitative theory-informed approach**. *BMJ Open* (2020.0) **10** e036183. DOI: 10.1136/bmjopen-2019-036183
109. 109.Pflegewissenschaft—Nursing Science. Academy-Practice Partnerships (APP) (2022). Available at:
https://nursing.unibas.ch/de/ins/partnerschaften/
(Accessed March 01, 2022).
110. Albert NM, Chipps E, Falkenberg Olson AC, Hand LL, Harmon M, Heitschmidt MG. **Fostering academic-clinical research partnerships**. *J Nurs Adm* (2019.0) **49** 234-41. DOI: 10.1097/nna.0000000000000744
111. Rycroft-Malone J, Wilkinson JE, Burton CR, Andrews G, Ariss S, Baker R. **Implementing health research through academic and clinical partnerships: a realistic evaluation of the collaborations for leadership in applied health research and care (CLAHRC)**. *Implement Sci* (2011.0) **6** 74. DOI: 10.1186/1748-5908-6-74
112. Franc S, Daoudi A, Mounier S, Boucherie B, Laroye H, Peschard C. **Telemedicine: what more is needed for its integration in everyday life?**. *Diabetes Metab* (2011.0) **37** S71-7. DOI: 10.1016/s1262-3636(11)70969-7
113. Kaplan B. **Revisiting health information technology ethical, legal, and social issues and evaluation: telehealth/telemedicine and COVID-19**. *Int J Med Inf* (2020.0) **143** 104239. DOI: 10.1016/j.ijmedinf.2020.104239
114. Keasberry J, Scott IA, Sullivan C, Staib A, Ashby R. **Going digital: a narrative overview of the clinical and organisational impacts of eHealth technologies in hospital practice**. *Aust Health Rev* (2017.0) **41** 646-64. DOI: 10.1071/ah16233
115. Schreiweis B, Pobiruchin M, Strotbaum V, Suleder J, Wiesner M, Bergh B. **Barriers and facilitators to the implementation of eHealth services: systematic literature analysis**. *J Med Internet Res* (2019.0) **21** e14197. DOI: 10.2196/14197
116. Raine R, Fitzpatrick R, Barratt H, Bevan G, Black N, Boaden R. *Challenges, solutions and future directions in the evaluation of service innovations in health care and public health* (2016.0)
117. Desveaux L, Soobiah C, Bhatia RS, Shaw J. **Identifying and overcoming policy-level barriers to the implementation of digital health innovation: qualitative study**. *J Med Internet Res* (2019.0) **21** e14994. DOI: 10.2196/14994
118. Murdoch TB, Detsky AS. **The inevitable application of big data to health care**. *JAMA* (2013.0) **309** 1351-2. DOI: 10.1001/jama.2013.393
119. Todd OM, Burton JK, Dodds RM, Hollinghurst J, Lyons RA, Quinn TJ. **New horizons in the use of routine data for ageing research**. *Age Ageing* (2020.0) **49** 716-22. DOI: 10.1093/ageing/afaa018
120. Piller A, Juckett LA, Hunter EG. **Adapting interventions for occupational therapy practice: application of the FRAME coding structure**. *OTJR* (2021.0) **41** 206-15. DOI: 10.1177/15394492211011609
121. Chlebowski C, Hurwich-Reiss E, Wright B, Brookman-Frazee L. **Using stakeholder perspectives to guide systematic adaptation of an autism mental health intervention for latinx families: a qualitative study**. *J Community Psychol* (2020.0) **48** 1194-214. DOI: 10.1002/jcop.22296
122. La Bash H, Galovski T, Stirman SW. **Adapting evidence-based psychotherapies while maintaining fidelity**. *Curr Treat Options Psychiatry* (2019.0) **6** 198-209. DOI: 10.1007/s40501-019-00177-9
123. Miller CJ, Barnett ML, Baumann AA, Gutner CA, Wiltsey-Stirman S. **The FRAME-IS: a framework for documenting modifications to implementation strategies in healthcare**. *Implement Sci* (2021.0) **16** 1-12. DOI: 10.1186/s13012-021-01105-3
124. Waltz TJ, Powell BJ, Fernández ME, Abadie B, Damschroder LJ. **Choosing implementation strategies to address contextual barriers: diversity in recommendations and future directions**. *Implement Sci* (2019.0) **14**. DOI: 10.1186/s13012-019-0892-4
125. Moullin JC, Dickson KS, Stadnick NA, Rabin B, Aarons GA. **Systematic review of the exploration, preparation, implementation, sustainment (EPIS) framework**. *Implement Sci* (2019.0) **14** 1. DOI: 10.1186/s13012-018-0842-6
126. 126.Der Schweizerische Bundesrat. [The Swiss Federal Council]. Medizinprodukteverordnung [Medical Device Regulation] Bern, Switzerland (2020). Available at:
https://www.admin.ch/opc/de/classified-compilation/19995459/index.html#a1
(Accessed August 22, 2020).
127. Schmid E, Gonon P
128. Ribaut J, De Geest S, Leppla L, Gerull S, Teynor A, Valenta S. **Exploring stem cell transplanted Patients’ perspectives on medication self-management and electronic monitoring devices measuring medication adherence: a qualitative sub-study of the Swiss SMILe implementation science project**. *Patient Prefer Adherence* (2022.0) **16** 11-22. DOI: 10.2147/ppa.S337117
|
---
title: 'Tailoring implementation strategies for scale-up: Preparing to take the Med-South
Lifestyle program to scale statewide'
authors:
- Jennifer Leeman
- Lindy B. Draeger
- Kiira Lyons
- Lisa Pham
- Carmen Samuel-Hodge
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012719
doi: 10.3389/frhs.2022.934479
license: CC BY 4.0
---
# Tailoring implementation strategies for scale-up: Preparing to take the Med-South Lifestyle program to scale statewide
## Abstract
### Background
Tailoring implementation strategies for scale-up involves engaging stakeholders, identifying implementation determinants, and designing implementation strategies to target those determinants. The purpose of this paper is to describe the multiphase process used to engage stakeholders in tailoring strategies to scale-up the Med-South Lifestyle Program, a research-supported lifestyle behavior change intervention that translates the Mediterranean dietary pattern for the southeastern US.
### Methods
Guided by Barker et al. framework, we tailored scale-up strategies over four-phases. In Phase 1, we engaged stakeholders from delivery systems that implement lifestyle interventions and from support systems that provide training and other support for statewide scale-up. In Phase 2, we partnered with delivery systems (community health centers and health departments) to design and pilot test implementation strategies (2014–2019). In Phase 3, we partnered with both delivery and support systems to tailor Phase 2 strategies for scale-up (2019–2021) and are now testing those tailored strategies in a type 3 hybrid study (2021–2023). This paper reports on the Phase 3 methods used to tailor implementation strategies for scale-up. To identify determinants of scale-up, we surveyed North Carolina delivery systems ($$n = 114$$ community health centers and health departments) and elicited input from delivery and support system stakeholders. We tailored strategies to address identified determinants by adapting the form of Phase 2 strategies while retaining their functions. We pilot tested strategies in three sites and collected data on intermediate, implementation, and effectiveness outcomes.
### Findings
Determinants of scale-up included limited staffing, competing priorities, and safety concerns during COVID-19, among others. Tailoring yielded two levels of implementation strategies. At the level of the delivery system, strategies included implementation teams, an implementation blueprint, and cyclical small tests of change. At the level of the support system, strategies included training, educational materials, quality monitoring, and technical assistance. Findings from the pilot study provide evidence for the implementation strategies' reach, acceptability, and feasibility, with mixed findings on fidelity. Strategies were only moderately successful at building delivery system capacity to implement Med-South.
### Conclusions
This paper describes the multiphase approach used to plan for Med-South scale-up, including the methods used to tailor two-levels of implementation strategies by identifying and targeting multilevel determinants.
## Introduction
Evidence suggests that implementation strategies are most effective when tailored to address the multilevel factors that determine when an intervention is successfully integrated into practice [1]. Multiple researchers have described methods for tailoring implementation strategies to promote and support the implementation of new interventions within one or more settings [2, 3]. Fewer have described methods for tailoring implementation strategies to scale-up interventions at the regional or national levels. The purpose of this paper is to describe the process we used to tailor implementation strategies to prepare for statewide scale-up of the Med-South Lifestyle Program.
The Med-South Lifestyle Program (Med-South) is a research-supported intervention with demonstrated effectiveness at improving dietary intake, physical activity, and blood pressure control (4–6). Med-South involves four structured, one-to-one, monthly sessions during which a counselor (health educator, nurse, nutritionist, or community health worker) promotes healthy lifestyle change through education, goal setting, action planning, and referrals to community resources (e.g., places to be physically active). Monthly sessions typically last 45–60 min. Counselors provide shorter 10–15 min booster calls between sessions. In previous studies, counselors have delivered sessions in-person, either in a healthcare setting or the home. In response to COVID-19, counselors in this study delivered the sessions both in-person and via phone or videoconference. Formerly called Heart-to-Health, the program has been re-named Med-South to highlight its promotion of a Mediterranean dietary pattern that has been adapted for a southern US population.
We define scale-up as a systematic approach to “rolling out a successful local program to regional, national, or international levels” [7].” *In this* study, our goal was to move from local to statewide roll out of Med-South. Tailoring implementation strategies for scale-up is different from tailoring strategies to implement an intervention at the local level [8]. One reason for this is the need to tailor strategies for each of the two levels of systems involved in scale-up: the delivery system and the support system [9, 10]. Delivery systems include the clinical, public health, or other community-based settings that are intended to adopt and implement an intervention into practice. Examples of strategies at the delivery-system level include implementation teams and cyclical, small tests of change [11]. Scale-up typically also involves one or more support systems, also referred to as intermediary and purveyor organizations [12, 13], that provide training, technical assistance and other implementation strategies to promote and support delivery systems to adopt and implement an intervention. In addition to requiring strategies at two levels of systems, scale-up often requires tailoring to address determinants beyond those considered during local implementation. For example, strategies will need to be tailored to determinants at the level of the support system (e.g., staffing, resources, mission). Tailoring strategies for scale-up also may require attention to policy, regulatory, budgetary, and other factors that may impede or support uptake at the regional or national level [14].
In this paper, we describe how we applied an adapted version of Barker et al. [ 15] framework for scaling up health interventions to prepare for the statewide scale-up of Med-South. We chose Barker's framework because it describes a systematic approach to moving implementation from the local to the regional or national levels. The Barker framework describes scale-up as a four-phase process: [1] set-up entrée, [2] develop the scalable unit, [3] test scale-up, and [4] go to full scale (Figure 1) [15]. The purpose of Phase 1 is to engage stakeholders who will provide entrée to two types of organizations: those that will implement the intervention (i.e., delivery systems) and those who will support intervention scale-up (i.e., support systems). Phase 2 involves building the scalable unit or “change package,” which includes the intervention and the implementation strategies needed to put it into practice. Phase 3 involves tailoring implementation strategies to support scale-up and then testing them across multiple settings. In Phase 4, the intervention is taken to scale at the regional or national level. At each phase in the framework, decisions about engaging stakeholders and tailoring strategies are influenced by multilevel barriers and facilitators (i.e., implementation determinants).
**Figure 1:** *Framework for scaling up interventions, adapted from Barker et al. (15).*
The U.S. Centers for Disease Control and Prevention funded this research through two five-year grants to the University of North Carolina's Prevention Research Center. We briefly summarize the first of these studies, during which we completed Phases 1 and 2 of Barker's framework (2014–2019). We then present methods and findings from Phase 3, which we completed during the first 2 years of the second study (2019–2024).
## Phase 1: Set up entrée
To plan for Med-South scale-up, we consulted the Prevention Research Center's community advisory board, which includes representatives from underserved communities, delivery systems, and state-level support systems in North Carolina. With support from the advisory board, we engaged representatives from community health centers (CHC), health departments (HD), and other community organizations (e.g., hospital, wellness center, community college, and agricultural extension) in each of two counties to participate in engaged research/practice workgroups.
## Phase 2: Build scalable unit
The engaged research/practice workgroups adapted the intervention, tailored implementation strategies, and pilot tested both the intervention and implementation strategies in two counties' CHCs and HDs [5, 6, 16]. After a 1-year planning period, we conducted two successive, one-arm, type 3 effectiveness-implementation trials; the first trial in the original county and then we replicated Med-South implementation in a second county. Both trials demonstrated broad reach to the intended population, fidelity to intervention protocols, and improvements in participant outcomes (dietary intake, physical activity levels, and blood pressure control) [5, 6]. The research team packaged the intervention and implementation strategies into a web-based change package that includes intervention and implementation protocols, a participant handbook, workflows for identifying and referring eligible participants, and metrics for monitoring implementation (https://hpdp.unc.edu/med-south-lifestyle-program/). Table 1 provides an overview of the implementation strategies developed in Phase 2, which are named using terminology from the Expert Recommendations for Implementing Change (ERIC) project [11]. For each strategy, the table specifies who enacted the strategy (i.e., research team and/or delivery system). Most strategies were enacted collaboratively by members of both the research team and delivery system. Finally, Table 1 describes each strategy's function (central purpose) and form (the specific activities or formats used to carry out the strategy's central function) [17].
**Table 1**
| Name (ERIC)a | Actor | Function | Form |
| --- | --- | --- | --- |
| Use advisory boards and workgroups | Research team and delivery sysm | Engage users in tailoring, enacting, and improving implementation | Monthly engaged research/practice workgroup meetings to oversee planning, implement a communication plan, review quality monitoring data, and improve implementation |
| Develop a formal implementation blueprint | | Standardize Med-South implementation process | Create workflow diagrams |
| Develop and distribute educational materials | | Provide resources to support intervention delivery | Adapt participant manual, create community resource inventory, and distribute |
| Conduct ongoing training | | Increase delivery system capacity to deliver Med-South per protocols | Deliver five-day, on-site training to counselors and supervisors |
| Centralize technical assistance | Research team | Monitor and support Med-South delivery per protocols | Make monthly phone calls with counselors to review cases and answer questions |
| Develop and implement tools for quality monitoring | | Monitor Med-South implementation per protocols | Establish tools and protocols to track data on intervention reach, fidelity, and effectiveness |
## Phase 3: Testing scale-up
A central product of Phase 2 was the creation of a change package that includes both the intervention and the strategies needed to implement the intervention within CHCs or HDs. The goal of Phase 3 was to develop the strategies needed to take the change package to scale across CHCs and HDs statewide. This required further tailoring of Phase 2 strategies to reduce the high level of research team involvement, which was neither feasible for scale-up nor sustainable over time. Phase 3 also involved the selection and tailoring of new strategies to overcome barriers and leverage facilitators to statewide scale-up.
## Design
In Phase 3, we engaged a stakeholder workgroup that included representatives from delivery systems (i.e., CHCs and HDs) and from three of North Carolina's state-level support systems (Institute of Public Health, Area Health Education Centers, and Community Health Center Association). In contrast to Phase 2's highly collaborative research/practice workgroup, the Phase 3 workgroup served in a consultative role to the research team [18]. With guidance from our stakeholder workgroup, we conducted formative work to identify multi-level determinants of scale-up and tailor strategies to target those determinants (2019–2020). We then pilot tested strategies using a one-arm pretest/posttest design (2020–2021). The University's Non-Biomedical Institutional Review Board (IRB) approved and monitored the study (#19-2079).
## Setting and sample
Formative data were collected via surveys of staff who make decisions about lifestyle programs in all CHCs and HDs in North Carolina and conversations with representatives from state-level support systems. CHC and HD staff were provided a $30 e-gift card for completing surveys. Based on formative findings, implementation strategies were tailored for scale-up strategies and then pilot tested with a convenience sample of three sites (1 CHC, 1 HD, and 1 CHC/HD partnership) that we recruited with input from the stakeholder workgroup. Each site signed a Memorandum of Understanding, in which they committed to identify staff to deliver and implement Med-South, release staff to participate in training and technical assistance, and deliver Med-South to at least 15 patients or clients. Each site was paid $5,000 to reimburse for time spent on study-related activities and $50 for each hour of Med-South delivery. Sites recruited clients/patients to participate in Med-South and then referred them to the research team, who screened for eligibility, obtained informed consent, and collected baseline and follow-up survey data. Eligibility criteria were broad and included anyone over 18 who did not have a health condition requiring them to follow a prescriptive diet (e.g., kidney disease). Med-South participants were reimbursed $40 for each of two data collection phone calls.
## Measures
Table 2 provides an overview of Phase 3 formative and pilot study measures. Further detail on these measures is provided below.
**Table 2**
| Construct | Measure | Data source | Timing |
| --- | --- | --- | --- |
| Formative | Formative | Formative | Formative |
| Determinants of scale-up [CFIR; (19 )] | Survey | Decision makers at CHCs and HDs | Spring 2020 |
| | Conversations | State-wide support systems | Spring 2020 |
| Pilot Study | Pilot Study | Pilot Study | Pilot Study |
| Capacity to deliver Med-South | Survey | Staff who implemented Med-South | Completion of training Fall 2020 |
| Capacity to implement Med-South | Survey | | Completion of training Fall 2020 |
| Reach (20) | REDCap-based system for tracking enrollment | | During recruitment and implementation |
| | | | Fall 2020–Spring 2021 |
| Acceptability | WEVAL (21) | | Completion of training Fall 2020 |
| | | | End of study |
| Feasibility | Focus group interview | | Spring 2021 |
| Fidelity (implementation) | Tracking logs | | During training |
| | Structured questions during technical assistance calls | | Monthly during implementation |
| Effectiveness | Survey (dietary intake, physical activity) (22–24) | Clients/patients who participated in Med-South | Baseline and end of study |
| | Weight | | |
| | Blood Pressure | | |
## Formative surveys and discussions
The research team developed an online survey to assess determinants of Med-South scale-up, with input from our stakeholder workgroup. The survey was guided by the Consolidated Framework for Implementation Research [CFIR; [19]] and included a 5-point Likert scale response ranging from strongly disagree to strongly agree. The survey was designed to assess factors relevant to organizational decision-making about Med-South adoption. For this reason, survey items focused on CFIR constructs related to barriers and facilitators at the level of the intervention and the inner and outer settings of the delivery systems where the intervention would be implemented. We administered the survey via email to all CHCs and HDs in NC.
To assess the resources available to support scale-up, members of the research team had conversations with leadership from statewide support systems. The research team conducted one-to-one phone or Zoom discussions with leaders of North Carolina's Institute of Public Health, Area Health Education Center, and Association of Community Health Centers. Discussions explored the support each organization was able and willing to provide to recruit CHCs and HDs and provide training or technical assistance to their staff. We also asked our Prevention Research Center advisory board members for input on how we might align Med-South scale-up with other state-level initiatives.
## Pilot test: Measures of intermediate, implementation, and effectiveness outcomes
Figure 2 depicts the framework used to evaluate the pilot test of Med-South scale-up. The figure describes how implementation strategies at the level of the support system were intended to build delivery system capacity to deliver and implement Med-South (intermediate outcomes). The figure further describes how implementation strategies at the level of the delivery system were intended to impact implementation and effectiveness outcomes. Below we describe the measures used to assess each type of outcome.
**Figure 2:** *Framework for evaluating Med-South Scale-up.*
## Intermediate outcomes: Capacity to deliver and implement Med-South
Delivery system capacity was operationalized as CHC and HD staff confidence in their ability to deliver and implement Med-South. To assess capacity, we administered two surveys. The first survey assessed confidence to deliver Med-South and included 8 items with a 10-point analog scale (1 = lowest and 10 = highest level of confidence). Items addressed respondents' ability to use Med-South materials (e.g., participant handbook), work with clients to change health behaviors, and share knowledge of nutrition and physical activity guidelines. The second survey included 10 items with a 5-point Likert scale that assessed staff confidence in their ability to work with others to complete implementation strategies (e.g., convene an implementation team). Both surveys were administered at completion of training.
At least one counselor at each site ($$n = 5$$) completed a survey assessing confidence in their ability to deliver and implement Med-South. Means for confidence to deliver Med-South ranged from 7.6 to 8.8 on a 10-point scale, with 10 being highly confident (Table 5). The three items with the lowest mean scores addressed confidence related to nutrition knowledge, physical activity knowledge, and motivational interviewing. Means for confidence to implement Med-South ranged from 4 to 4.6 on a 5-point scale (Table 6). The four items with the lowest mean score addressed confidence related to process flow diagrams and conducting cyclical small tests of change (i.e., Plan-Do-Study-Act cycles).
## Implementation outcomes: Reach, acceptability, feasibility, and fidelity
Reach was operationalized as the number and demographics of clients/patients enrolled and retained [20]. Reach data were extracted from REDCap, a secure online system that both the research team and CHC and HD staff used to track data on participant enrollment and participation. We assessed staff perceptions of the acceptability of support system-level strategies (e.g., training) via an adapted version of the Workshop Evaluation (WEVAL) survey [21], which was administered at completion of training and assessed perceptions of educational materials and trainings using a five-point Likert response scale. WEVAL is a validated measure administered immediately following training to assess participant perceptions of quality, relevance, and support provided. To further assess acceptability and feasibility, a member of the team with expertise in qualitative methods (JL) conducted three, end-of-study, semi-structured, video-conferenced focus group interviews with the implementation team and counselors at each site. Using an interview guide, staff were asked to reflect on their experience engaging with support system-level strategies (e.g., trainings), applying delivery system-level strategies (e.g., convening an implementation team), and delivering Med-South to their clients/patients. The study project manager monitored fidelity to support-system strategies via tracking logs. Fidelity to delivery-system strategies was assessed via structured questions during technical assistance calls during which implementation teams were asked to report on progress toward implementing Med-South and revisions made to address those barriers.
## Effectiveness outcomes
Data were collected to assess blood pressure, weight, and self-reported dietary and physical activity behaviors at baseline and 4-months. Each site's counselors collected blood pressure and weight data at the first and last Med-South counseling sessions. Members of the study team collected data on self-reported dietary behaviors by phone, using a validated brief dietary screener for fruits, vegetables, and fiber [22] and a single item about nuts and nut butter intake from a validated fat quality survey [23]. A single item (adapted from the 2 items used in Behavioral Risk Factor Surveillance System [BRFSS]) was used to assess usual daily consumption of sugar-sweetened beverages [24].
Sample sizes were not sufficient to test for statistical significance. On average sites observed clinically meaningful improvements in blood pressure levels, and mean improvements in dietary behaviors and weight changes were similar to those observed in trials of the Med-South program [5, 6]; (Table 8).
**Table 8**
| Site (Program completers) | Daily servings of fruits and vegetables | Weight (lb.) | Systolic blood pressure (mm Hg) | Diastolic blood pressure (mm Hg) |
| --- | --- | --- | --- | --- |
| HD (n = 13) | 0.4 | −2.4 | −3.5 | −0.6 |
| CHC (n = 8) | 0.9 | +0.07 | −4.6 | −3.2 |
| HD/CHC (n = 4) | 1.3 | +0.1 | −8.3 | −5.2 |
| Average Pre-/Post-Change | 0.9 | −2.2 | −5.5 | −3.0 |
## Analysis
Quantitative data were summarized using descriptive statistics. Interviews were audiotaped and transcribed. Two members of the research team applied framework analysis to review interview transcripts to assess perceptions of the acceptability and feasibility of the intervention and implementation strategies, reports of fidelity to implementation strategies, and recommendations for improvement [25]. The two team members reviewed transcripts, developed a code book, coded transcripts, and used a matrix to chart key findings across both codes and cases. They met to compare and reconcile coding and matrices. Final analyses were shared with the full research team and stakeholder workgroup.
## Formative findings: Determinants of scale-up
Survey respondents ($$n = 114$$) included 58 HDs and 56 CHCs (67 and $64\%$ response rates, respectively). Table 3 provides an overview of survey responses. A majority of respondents agreed that Med-South had potential to improve lifestyle counseling ($82\%$) and that it would be feasible for staff to deliver the intervention ($67\%$) and attend required trainings ($60\%$). Respondents also agreed that Med South aligned with their organization's mission ($85\%$) and priorities ($93\%$). However, only a minority agreed that Med-South would be feasible for their organization to deliver ($40\%$) or that their organization would have sufficient time ($40\%$) or staff ($38\%$) to implement. Furthermore, only a minority reported that patients/clients would have time ($31\%$) or transportation ($13\%$) to attend sessions or have access to heart healthy foods ($18\%$). Mean scores were not significantly different across geographic region or setting type with one exception: respondents from western North Carolina gave lower ratings to outer-setting level items than respondents in other regions of the state. The COVID-19 pandemic started just prior to survey administration, and in open-ended comments, multiple respondents noted the distinct challenges created by the pandemic.
**Table 3**
| Survey Item | n (%) Agreea | Mean (SD) |
| --- | --- | --- |
| Characteristics of the intervention | | |
| Med-South has potential to improve lifestyle counseling in this organization | 89 (82) | 4.26 (0.80) |
| Med-South would be difficult for our staff to deliver | 33 (67)b | 2.97 (0.99) |
| It would be difficult for our staff to attend training on Med-South | 40 (60)b | 3.09 (1.16) |
| Inner setting | | |
| Promoting healthy eating and physical activity is a priority for our organization | 101 (93) | 4.61 (0.74) |
| Med-South fits the mission of our organization | 93 (85) | 4.39 (0.85) |
| We have the physical space to implement Med-South | 60 (56) | 3.46 (1.34) |
| Implementing the Med-South Lifestyle Program is feasible for our organization | 58 (53) | 3.57 (1.17) |
| We have the time to implement Med-South | 43 (40) | 3.05 (1.17) |
| We have the staff to implement Med-South | 41 (38) | 2.94 (1.24) |
| Outer setting | | |
| Our patient/clients want more support to improve lifestyle behaviors | 70 (65) | 3.80 (0.80) |
| Our patients/clients would have time to attend Med-South sessions | 33 (31) | 3.09 (0.79) |
| Our patients/clients have access to fresh fruits, vegetables, and other heart-healthy foods | 19 (18) | 2.56 (0.86) |
| Our patients/clients would have transportation to attend Med-South sessions | 14 (13) | 2.54 (0.83) |
Discussions with support system stakeholders also occurred shortly after the start of the COVID-19 pandemic, and stakeholders told us they were overwhelmed with the work required to respond to the pandemic. Concurrently, their resources were heavily invested in preparing CHCs for North Carolina's initiative to transform Medicaid into a medical home model effective July 2020. Nonetheless, two support systems were interested and had some capacity to support scale-up. The North Carolina Area Health Education Centers offered the support of their statewide learning management system, which includes staff and infrastructure to host trainings and process continuing education credits. The North Carolina Institute of Public Health had capacity to support the transition of some training content to an online format. The Prevention Research Center advisory board identified a potential facilitator at the outer setting level: In the near future, the state would be revising its public insurance program (Medicaid) to provide incentives to sites that improve specific quality measures (e.g., control of hemoglobin A1c and high blood pressure).
## Tailoring implementation strategies for scale-up
As summarized in Table 4, tailoring for scale-up was designed to retain the function of Phase 2 strategies and modify their forms. This included transitioning implementation strategies from the research team to either the delivery system (i.e., CHCs and HDs) or an established support system. For example, responsibility for the function “engaging users in tailoring, enacting, and improving implementation strategies” was transitioned from an engaged research/practice workgroup to a site-based implementation team. Based on formative findings, strategies also were tailored to [1] enhance feasibility by reducing burden to CHCs and HDs, [2] limit potential exposure to COVID-19, and [3] leverage support system resources. To reduce burden to CHCs and HDs, we reduced the overall amount of training time and shifted from an in-person to virtual format. To transition trainings to virtual format, we leveraged support system resources to support creation and delivery of both self-directed, online modules and synchronous web-conferences. Converting trainings to a virtual format also served to reduce safety risks during the COVID-19 pandemic. In addition to creating new forms for the functions performed by Phase 2 strategies, we also identified the need to “increase delivery system capacity to implement Med-South per protocols”. This new function was needed to build delivery system capacity to take responsibility for implementation (e.g., create a workflow diagram and implementation plan). The research team also began to plan for ways to leverage the state's new Medicaid transformation initiative to promote and support Med-South as a means of improving quality measures related to control of hemoglobin A1c and high blood pressure.
**Table 4**
| Phase 2 Strategy (ERIC)a | Function (new function italicized) | Phase 2 Form | Phase 3 Strategy (ERIC)a | Phase 3 Form |
| --- | --- | --- | --- | --- |
| Strategies transitioned to delivery system | Strategies transitioned to delivery system | Strategies transitioned to delivery system | Strategies transitioned to delivery system | Strategies transitioned to delivery system |
| Use advisory boards and workgroups | Engage users in tailoring, enacting, and improving implementation | Monthly engaged research/practice workgroup meetings | Organize clinician implementation team meetings | Delivery system leadership designate team members, endorse a team charter and a plan for monthly meetings |
| Develop a formal implementation blueprint | Standardize Med-South implementation process | Engaged research/practice workgroup creates workflow diagrams | Develop a formal implementation blueprint | Implementation team completes readiness assessment, workflow diagrams, and implementation plan |
| Centralize technical assistance | Monitor and support Med-South implementation per protocols | Research team makes monthly phone calls to counselors to review cases and answer questions | Conduct cyclical small tests of change | Implementation team conducts Plan-Do-Study-Act (PDSA) cycles to iteratively improve implementation |
| Strategies transitioning to support system | Strategies transitioning to support system | Strategies transitioning to support system | Strategies transitioning to support system | Strategies transitioning to support system |
| Conduct ongoing training | Increase delivery system capacity to deliver Med-South per protocols | Research team and members of delivery system deliver five-day, on-site training to counselors and supervisors | Conduct ongoing training (on intervention) | Two 1-h, self-guided online modules on current nutrition and physical activity guidelines & Four 2-h web-conference sessions, with 4 h on Med-South delivery |
| | Increase delivery system capacity to implement Med-South | | Provide ongoing training (on implementation) | Four 2-hour web-conference sessions, with 4 hours on Med-South implementation |
| Strategies delivered by research team | Strategies delivered by research team | Strategies delivered by research team | Strategies delivered by research team | Strategies delivered by research team |
| Develop and distribute education materials | Provide resources to support intervention delivery | Adapt participant manual, create community resource inventory, and distribute | Distribute educational materials | Distribute participant manuals and create and distribute community resource inventories |
| Develop and implement tools for quality monitory | Monitor and support Med-South implementation and delivery per protocols | Establish tools and protocols to track data on intervention reach, fidelity, and effectiveness | Develop and implement tools for quality monitoring | REDCap system used to track Med-South delivery and effectiveness. |
| Centralize technical assistance | | Research team makes monthly phone calls to counselors to review cases and answer questions | Centralize technical assistance | Monthly technical assistance phone calls with implementation teams to review REDCap data and address questions/barriers |
## Pilot testing scale-up
Figure 2 depicts the framework used to evaluate the pilot test of Med-South scale-up. The figure describes the implementation strategies support systems used to build delivery system capacity to deliver and implement Med-South (intermediate outcomes). The figure further describes the implementation strategies delivery systems used to implement Med-South. Finally, the figure depicts how both support and delivery system strategies impact implementation outcomes, which in turn impact effectiveness outcomes. Below we summarize findings from the pilot test of Med-South scale-up.
## Reach
The sites enrolled 39 participants of whom 25 completed the 4-month intervention period (Table 7). Completion rates were $100\%$ at Site A, $61.5\%$ at Site B, and $30.8\%$ at Site C (the CHC/HD partnership site), for an overall rate of $64\%$. Enrollees were predominantly female ($79.5\%$) and Black ($60.0\%$). Of enrolled participants, 28 ($72\%$) attended the first intervention session. For those who attended the first session, $89\%$ ($\frac{25}{28}$) completed the program and provided follow-up survey data.
**Table 7**
| Unnamed: 0 | Site A | Site B | Site C | Overall |
| --- | --- | --- | --- | --- |
| Enrolled | 13 | 13 | 13 | 39 |
| Female | 12 | 10 | 9 | 31 |
| Male | 1 | 3 | 4 | 8 |
| Black | 5 | 8 | 10 | 23 |
| White | 8 | 4 | 2 | 14 |
| Other race/ethnicity | 0 | 1 | 1 | 2 |
| Completed | 13 | 8 | 4 | 25 |
| Female | 12 | 7 | 2 | 21 |
| Black | 5 | 5 | 2 | 12 |
| White | 8 | 2 | 1 | 11 |
| Other | 0 | 1 | 1 | 2 |
## Acceptability
Seven staff completed the WEVAL survey, and on a five-point Likert scale, all either agreed or strongly agreed that they were satisfied with intervention materials, comfortable using materials, expected to use what they learned in the trainings, and perceived the intervention to be compatible with the needs of their patients/clients. Interview findings provide further support for the acceptability of support system-level implementation strategies. Staff reported that trainings were thorough and the educational materials (i.e., participant handbook) were beautiful. They appreciated the monthly technical assistance calls and the research team's responsiveness to questions during calls and via email. Staff also identified concerns with the support system strategies, including the gap between completion of training and the first counseling session and difficulties with the electronic REDCap system used to capture data on intervention delivery. Staff were highly satisfied with the Med-South intervention which they viewed as an opportunity to improve patients' health and to try something new.
## Implementation fidelity
Fidelity to support system strategies was high. Trainings and technical assistance were delivered as intended, with high levels of participation from CHC and HD staff. Fidelity to delivery system implementation strategies was mixed. All three sites held monthly implementation meetings, used process flow diagrams, and communicated with key stakeholders. Only two of three sites used implementation plans, and none of the sites used readiness assessments or PDSA cycles as intended.
## Feasibility
Staff identified several barriers to implementing and delivering Med-South. Sites had limited staff to deliver Med-South, particularly during the COVID-19 pandemic, and staff had multiple competing demands on their time. Patients and clients also had limited time available to schedule 1-h counseling sessions, especially during the workday. At the one site where a CHC and HD partnered on implementation, HD staff had difficulty engaging participants who were referred by the CHC, and therefore unfamiliar with the HD.
## Discussion
In this paper, we illustrate the use of a multiphase process to tailor implementation strategies in preparation for scale-up. The process followed Barker and colleagues' four-phase framework which, similar to other scale-up frameworks, describes multiple phases that begin with formative work and progress from small pilot studies to full scale implementation [26]. We contribute to these prior frameworks by distinguishing two levels of implementation strategies and by describing the process used to tailor strategies for scale-up across phases. In describing the tailoring process, we specify how we largely retained the strategies' functions across phases and tailored their forms to (a) transition strategies from the research team to delivery or support systems and (b) address barriers to implementation and scale-up.
Study findings provide initial support for the feasibility, acceptability, and impact of the strategies used to scale-up Med-South. At the height of the COVID-19 pandemic, the three pilot sites reached 39 patients/clients and retained 25, almost half of whom were African Americans. Most drop out occurred after completing the baseline survey and prior to participating in the first intervention session. For patients/clients who attended the first session, $89\%$ completed the program. Of note, completion rates varied across the three sites with the CHC/HD partnership site retaining only 4 of the 13 patients enrolled. CHC and HD staff reported moderate to high levels of confidence in their ability to deliver and implement Med-South, viewed implementation strategies as acceptable, and used some of the implementation strategies with fidelity to protocols. These findings are based on small sample sizes with the goal of contributing to the progressive refinement of strategies over iterative cycles of testing.
Our ability to fully transition implementation strategies to established support systems was limited by the COVID-19 pandemic. During the study period, both delivery and support systems were overwhelmed by the demands of vaccination and testing. Following an initial delay in recruitment, we were able to identify HDs and CHCs whose staff were ready to focus their attention beyond COVID-19. This was less true for support systems who experienced a continued need to focus their resources on supporting delivery system response to the pandemic. Nevertheless, we identified two support systems that were willing to assume some responsibility for training and plan to explore opportunities to further transition responsibility to support systems as we move forward.
The multiphase process used in this study is applicable to scale-up initiatives involving a range of delivery systems, for example, hospitals, clinics, health departments, and schools, among others. This study's process also is applicable to a range of support systems, including any government, academic, for-profit, or non-profit organizations that provides support to delivery systems [8]. In this study, support system-level strategies focused on horizontal scale-up or the spread of Med-South across settings. Future research is needed to develop and test support system strategies that focus on vertical scale-up, in other words, on strategies that target change at the “policy, political, legal, regulatory, budgetary or other health systems changes” needed to support an intervention's scale-up at the regional or national level [[14], p. 21].
The multiphase process used in this study is particularly relevant when researchers function as the primary support system for the initial implementation and scale-up of new interventions. The extent of researcher engagement varies. In this study and many others, researchers initially developed a highly engaged, collaborative relationship with community and practice partners to co-create implementation strategies. This initial high-level of engagement provided an in-depth understanding of implementation determinants in the local setting and the strategies needed to address them [18]. This high-level of researcher engagement is difficult to scale-up and sustain across many practice settings [26, 27]. We describe how we addressed this challenge by reducing the research team's engagement in implementation and transitioning responsibility for implementation to either the delivery system or an established support system. One of the central goals of this transition was to build delivery-system capacity to assume responsibility for the functions previously performed by the highly-engaged research/practice workgroup. Specifically, we created site-based implementation teams and trained them to tailor, enact, and iteratively improve implementation strategies. As detailed below, we were only partially successful in achieving this objective.
As depicted in this study's evaluation framework (Figure 1), the success of Med-South scale-up was contingent on building delivery system capacity to both deliver and implement Med-South. Findings from the pilot study indicate that staff were confident in their ability to deliver Med-South. Findings on staff capacity to implement Med-South were mixed, including only moderate levels of confidence and low fidelity related to creating process follow diagrams and conducting cyclical, small tests of change (i.e., PDSA cycles), two methods delivery systems can use to tailor implementation strategies. Delivery system capacity to tailor implementation to local needs is critical to implementing and sustaining intervention. Delivery systems need to have the capacity to map processes, identify barriers, and monitor and address gaps in implementation so they can tailor implementation to the needs of their community and context (e.g., staffing, population served, funding) [28]. The challenges this study experienced are not unique. Multiple researchers have reported on delivery systems' low levels of adherence to PDSA cycle protocols [29]. This study is novel in providing the data needed to advance understanding of where gaps occurred, including data on implementation outcomes at the levels of both support and delivery systems as well as intermediate outcomes to assess whether strategies had the intended effects on implementation determinants (e.g., delivery system capacity). Using a multiphase approach to scale-up has allowed us to tailor strategies to address gaps and iteratively prepare for scale-up. The gap in delivery system capacity may have resulted, in part, from how training was tailored to a shorter, virtual format. To address this gap, we have further tailored protocols for technical assistance to reinforce training content related to implementation and to require sites to report back on the completion of specific implementation activities (e.g., completing PDSA cycles). We are testing the revised strategies in our current study of Med-South scale-up across 20 HDs and CHCs (2021–2023).
## Conclusions
This paper illustrates how a multiphase approach was used to prepare for the statewide scale-up of a health intervention, with a specific focus on tailoring two-levels of implementation strategies. The approach might be applied to plan for the scale-up of interventions across a range of delivery and support systems.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by the University of North Carolina at Chapel Hill's Non-biomedical Institutional Review Board. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.
## Author contributions
JL and CS-H led the research study. JL drafted the manuscript. All authors participated in data collection and analysis, contributed to the interpretation of findings and tailoring of implementation strategies, and provided feedback on the manuscript.
## Funding
This publication was funded by the Centers for Disease Control and Prevention of the U.S. Department of Health and Human Services (HHS) as part of a financial assistance award (cooperative agreement numbers: U48 DP006400) with 100 percent funded by CDC/HHS. The contents are those of the author(s) and do not necessarily represent the official views of, nor an endorsement, by CDC/HHS, or the U.S. Government.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Baker R, Camosso-Stefinovic J, Gillies C, Shaw EJ, Cheater F, Flottorp S. **Tailored interventions to address determinants of practice**. *Cochrane Database Syst Rev* (2015) **5** CD005470. DOI: 10.1002/14651858.CD005470.pub3
2. Powell BJ, Beidas RS, Lewis CC, Aarons GA, McMillen JC, Proctor EK. **Methods to improve the selection and tailoring of implementation strategies**. *J Behav Health Serv Res.* (2017) **44** 177-94. DOI: 10.1007/s11414-015-9475-6
3. Fernandez M, Ten Hoor GA, van Lieshout S, Rodrigueq SA, Beidas RS, Parcel G. **Implementation mapping: using intervention mapping to develop implementation strategies**. *Frontiers in public health.* (2019) **18** 158. DOI: 10.3389/fpubh.2019.00158
4. Keyserling TC, Samuel-Hodge CD, Pitts SJ, Garcia BA, Johnston LF, Gizlice Z. **A community-based lifestyle and weight loss intervention promoting a Mediterranean-style diet pattern evaluated in the stroke belt of North Carolina: the heart healthy lenoir project**. *BMC Public Health.* (2016) **16** 732. DOI: 10.1186/s12889-016-3370-9
5. Samuel-Hodge CD, Gizlice Z, Allgood SD, Bunton AJ, Erskine A, Leeman J. **Strengthening community-clinical linkages to reduce cardiovascular disease risk in rural NC: feasibility phase of the CHANGE study**. *BMC Public Health.* (2020) **20** 264. DOI: 10.1186/s12889-020-8223-x
6. Samuel-Hodge CD, Ziya G, Allgood SD, Bunton AJ, Erskine A, Leeman J. **A hybrid implementation-effectiveness study of a community health worker-delivered intervention to reduce cardiovascular disease risk in a rural, underserved non-hispanic black population: the CHANGE study**. *Am J Health Promot.* (2022) **36** 948-58. DOI: 10.1177/08901171221078272
7. Ben Charif A, Zomahoun HT, LeBlanc A, Langlois L, Wolfenden L, Yoong SL. **Effective strategies for scaling up evidence-based practices in primary care: a systematic review**. *Implement Sci.* (2017) **12** 139. DOI: 10.1186/s13012-017-0672-y
8. Leeman J, Boisson A, Go V. **Scaling up public health interventions: engaging partners across multiple levels**. *Annu Rev Public Health.* (2022) **43** 155-71. DOI: 10.1146/annurev-publhealth-052020-113438
9. Leeman J, Toles M. **What does it take to scale-up a complex intervention? Lessons learned from the Connect-Home transitional care intervention**. *J Adv Nurs* (2020) **76** 387-97. DOI: 10.1111/jan.14239
10. Wandersman A, Duffy J, Flaspohler P, Noonan R, Lubell K, Stillman L. **Bridging the gap between prevention research and practice: the interactive systems framework for dissemination and implementation**. *Am J Community Psychol.* (2008) **41** 171-81. DOI: 10.1007/s10464-008-9174-z
11. Powell BJ, Waltz TJ, Chinman MJ, Damschroder LJ, Smith JL, Matthieu MM. **A refined compilation of implementation strategies: results from the expert recommendations for implementing change (ERIC) project**. *Implement Sci.* (2015) **10** 21. DOI: 10.1186/s13012-015-0209-1
12. Franks RP, Bory CT. **Who supports the successful implementation and sustainability of evidence-based practices? Defining and understanding the roles of intermediary and purveyor organizations**. *New Direct Child Adolescent Develop.* (2015) **2015** 41-56. DOI: 10.1002/cad.20112
13. Proctor E, Hooley C, Morse A, McCrary S, Kim H, Kohl PL. **Intermediary/purveyor organizations for evidence-based interventions in the US child mental health: characteristics and implementation strategies**. *Implement Sci.* (2019) **14** 3. DOI: 10.1186/s13012-018-0845-3
14. ExpandNet WHO. *Nine Steps for Developing a Scaling-Up Strategy* (2010)
15. Barker PM, Reid A, Schall MW. **A framework for scaling up health interventions: lessons from large-scale improvement initiatives in Africa**. *Implement Sci.* (2016) **11** 12. DOI: 10.1186/s13012-016-0374-x
16. Allgood SD, Leeman J, Rosemond CA, Ammerman AS, Samuel-Hodge C, Cykert S. **Reducing cardiovascular disease in a rural community**. *Public Health Nurs.* (2019) **36** 676-82. DOI: 10.1111/phn.12649
17. Perez Jolles M, Lengnick-Hall R, Mittman BS. **Core functions and forms of complex health interventions: a patient-centered medical home illustration**. *J Gen Intern Med.* (2019) **34** 1032-8. DOI: 10.1007/s11606-018-4818-7
18. Ramanadhan S, Davis MM, Armstrong R, Baquero B, Ko LK, Leng JC. **Participatory implementation science to increase the impact of evidence-based cancer prevention and control**. *Cancer Causes Control CCC.* (2018) **29** 363-9. DOI: 10.1007/s10552-018-1008-1
19. Damschroder LJ, Aron DC, Keith RE, Kirsh SR, Alexander JA, Lowery JC. **Fostering implementation of health services research findings into practice: a consolidated framework for advancing implementation science**. *Implement Sci.* (2009) **4** 50. DOI: 10.1186/1748-5908-4-50
20. Kessler RS, Purcell EP, Glasgow RE, Klesges LM, Benkeser RM, Peek CJ. **What does it mean to “employ” the RE-AIM model?**. *Eval Health Prof.* (2013) **36** 44-66. DOI: 10.1177/0163278712446066
21. Bartholomew NG, Joe GW, Rowan-Szal GA, Simpson DD. **Counselor assessments of training and adoption barriers**. *J Subst Abuse Treat.* (2007) **33** 193-9. DOI: 10.1016/j.jsat.2007.01.005
22. Block G, Gillespie C, Rosenbaum EH, Jenson C. **A rapid food screener to assess fat and fruit and vegetable intake**. *Am J Prev Med.* (2000) **18** 284-8. DOI: 10.1016/S0749-3797(00)00119-7
23. Kraschnewski JL, Gold AD, Gizlice Z, Johnston LF, Garcia BA, Samuel-Hodge CD. **Development and evaluation of a brief questionnaire to assess dietary fat quality in low-income overweight women in the southern United States**. *J Nutr Educ Behav.* (2013) **45** 355-61. DOI: 10.1016/j.jneb.2012.10.008
24. 24.Centers for Disease Control and Prevention. Behavioral Risk Factor Surveillance System Questionnaire. (2017). Available online at: https://www.cdc.gov/brfss/questionnaires/pdf-ques/2017_BRFSS_Pub_Ques_508_tagged.pdf (accessed June 10, 2021).. *Behavioral Risk Factor Surveillance System Questionnaire* (2017)
25. Gale NK, Heath G, Cameron E, Rashid S, Redwood S. **Using the framework method for the analysis of qualitative data in multi-disciplinary health research**. *BMC Med Res Methodol.* (2013) **13** 117. DOI: 10.1186/1471-2288-13-117
26. Milat AJ, Bauman A, Redman S. **Narrative review of models and success factors for scaling up public health interventions**. *Implement Sci.* (2015) **10** 113. DOI: 10.1186/s13012-015-0301-6
27. Zomahoun HTV, Ben Charif A, Freitas A, Garvelink MM, Menear M. **The pitfalls of scaling up evidence-based interventions in health**. *Global Health Action.* (2019) **12** 1670449. DOI: 10.1080/16549716.2019.1670449
28. Chambers DA, Glasgow RE, Stange KC. **The dynamic sustainability framework: addressing the paradox of sustainment amid ongoing change**. *Implement Sci.* (2013) **8** 117. DOI: 10.1186/1748-5908-8-117
29. Taylor MJ, McNicholas C, Nicolay C, Darzi A, Bell D, Reed JE. **Systematic review of the application of the plan-do-study-act method to improve quality in healthcare**. *BMJ Qual Saf.* (2014) **23** 290-8. DOI: 10.1136/bmjqs-2013-001862
|
---
title: Applicability of Exercise and Education Programmes for Knee Osteoarthritis
Management to Switzerland
authors:
- Lea Ettlin
- Anne-Kathrin Rausch Osthoff
- Irina Nast
- Karin Niedermann
journal: Frontiers in Health Services
year: 2021
pmcid: PMC10012725
doi: 10.3389/frhs.2021.760814
license: CC BY 4.0
---
# Applicability of Exercise and Education Programmes for Knee Osteoarthritis Management to Switzerland
## Abstract
Objectives: The aim of this study was to assess the applicability of six OARSI (Osteoarthritis Research Society International) approved exercise and education programmes for the conservative management of knee osteoarthritis to the Swiss health care system.
Methods: The RE-AIM framework was used in this cross-sectional survey study to analyse the characteristics of the six exercise and education programmes. A survey was developed based on the RE-AIM dimensions, “Reach, Effectiveness, Adoption, Implementation, and Maintenance,” for rating the applicability of the programmes (on a scale of 1 = “least applicable” to 10 = “most applicable”). Programme scores of ≥7 indicated applicability to the Swiss health care system. Nine selected physiotherapy experts for knee OA management in Switzerland were invited for the rating.
Results: The six programmes were rated by six of the nine invited research experts with mean scores of between 5.9 and 9.45. Four programmes scored 7 or more. These four programmes all included supervised exercise sessions and education with the goal that the participants understand the diagnosis and the management of OA. The two lower rated programmes focused on exercise counselling or weight reduction.
Conclusion: The programme with the highest scores consists of exercise and education and scored higher than 7 in all RE-AIM dimensions. Therefore, this programme is most applicable to the Swiss health care system as only a few adaptations would be needed for its successful implementation.
## Introduction
The international clinical guidelines for the management of knee osteoarthritis (OA), which were developed by the OA Research Society International (OARSI), American College of Rheumatology (ACR), and European Alliance of Associations for Rheumatology (EULAR), recommend exercise, education, and weight control, when appropriate, as first-line interventions (1–3). Pharmacological therapy or passive treatment might support first-line interventions (1–3). There is strong evidence that regularly performed exercises for knee OA have a positive effect on pain and joint function (1–5). The long-term goal of these interventions is the enhancement of self-management in people with knee OA. However, clinical practise is often at variance with guidelines (6–10). Because of this, the OA guideline recommendations have already been translated into exercise and education programmes, or models of care, in several countries to make their application feasible and simplify implementation into clinical practise. There are six well-established, OARSI-endorsed programmes that meet the specific needs of their national health care systems [9, 10]. These programmes have been successfully applied in real-world settings and have been proven effective [11].
According to national data, knee OA is the most common diagnosis in Swiss hospitals [12]. However, both research and observation of clinical practise have shown that the guidelines recommendations are not being systematically applied in knee OA management and exercise and education appear to be underexploited in Switzerland, which results in an evidence-performance gap [13]. In accordance with experience from other countries, the implementation of a national exercise and education programme in Switzerland would contribute to overcoming this evidence-performance gap [9, 10]. Consequently, a network of physiotherapy experts for knee OA management was formed in 2019 to promote the implementation of an exercise and education programme, with a preference for one that was already established and approved by OARSI. The network included six researchers from three Universities of Applied Sciences in the three national language areas of Switzerland, i.e., German, French, and Italian, two clinical practitioners representing each of two different specialist physiotherapy societies, and one patient representative of the Swiss League Against Rheumatism (SLAR).
The implementation of a structured programme in a health care system is challenging and time-consuming [9, 10, 14, 15]. For the successful implementation of a structured programme for knee OA, it must be aligned to the characteristics of the Swiss health care system and meet local needs. Switzerland's health care system is organised at the cantonal level and is decentralised, complex, and has a high level of local autonomy. This structure, along with the fee for service reimbursement [16], represent barriers to the implementation of new care structures and innovations [17]. Many aspects (e.g., programme characteristics, provider characteristics, community factors, and health care system structures) influence the implementation process and need to be considered carefully [18]. This has proven to be the case for other disease management programmes, such as diabetes [17]. An implementation into the Swiss health system requires not only an analysis of the clinical effectiveness of the programme, but also an assessment of the necessary national adaptations. However, Rodriguez and Bernal [19] have stated that in the implementation of evidence-based treatment, the more cultural adaptations of an innovation are made, the higher the risk of compromising the effectiveness of the treatment, i.e., essential components linked to positive outcomes are changed and may compromise the effectiveness of the treatment [20, 21]. Therefore, the process for selecting a programme applicable to Switzerland should be: 1. Identification of the programme that best meets the needs and requirements of the Swiss health care system, and 2. Application of the rule “The less adaptations, the better.” The aim of this study was to analyse the six OARSI-approved exercise and education programmes, using the RE-AIM framework, to assess their applicability for implementation in the Swiss health care system.
## Design
This cross-sectional survey study was based on a secondary analysis of the six OARSI-approved exercise and education programmes, using the RE-AIM framework [15] to rate the applicability of the programmes for implementation in Switzerland.
## Participants
All members of the above-mentioned network of experts in knee OA management were invited to participate in the survey and to rate the applicability of the programmes. They were personally informed of the survey and invited to participate by email, with the survey (Word-document) attached. A follow-up reminder was sent 2 weeks later.
For transparency and consistency of reporting in implementation studies, this study follows the “Standards for Reporting Implementation studies” (StaRI) statement.
## Characteristics of the OARSI-Approved Programmes
The six OARSI-approved programmes are: 1. “ Osteoarthritis Chronic Care Program (OACCP) Australia,” 2. “ Better management of patients with osteoarthritis (BOA) Sweden,” 3. “ Good Life with Osteoarthritis in Denmark (GLA:D®),” 4. “ Osteoarthritis Healthy Weight For Life (OA HWFL) Australia,” 5. “ *Amsterdam osteoarthritis* cohort (AMSOA) of The Netherlands,” and 6. “ Joint Implementation of Osteoarthritis guidelines in the West Midlands” (JIGSAW). The characteristics of the six programmes, evaluated using the RE-AIM dimensions, regarding target population, interventions, duration, primary outcomes, number of patients enrolled, and reimbursement, were extracted from published studies, reports, and the websites of the respective programmes. Table 1 displays a short overview of the programmes, showing the aim of the programme, how many people were enrolled, and how the local reimbursement system was organised.
**Table 1**
| Programme | Country of origin | Aim of programme (published on their website, if available) | Number of patients enrolled | Reimbursement policies |
| --- | --- | --- | --- | --- |
| BOA (8, 10, 22–24) | Sweden | All patients with OA in Sweden should be offered adequate information and exercise according to the guidelines | Approx. 10,941 patients/year (2008–2020) | Coverage by universal healthcare: 100% coverage with a limited out-of-pocket spending (maximum fee of ~120 USD for outpatient visits during a 12-month period) |
| OACCP (25–28) | Australia | A pathway to improved care is offered for people who have one of the most common, debilitating, costly and rapidly growing chronic conditions—OA | Approx. 1,250 patients/year (2012–2020) | Coverage by universal healthcare (Medicare): Subsidies of supervised exercise sessions and consultations (assessments) according to management plan with a possible rebate from the private health insurers. Providers set the cost of the program |
| GLA:D® (7, 29, 30) | Denmark | Implementation of current clinical guidelines for OA into clinical care. Evidence-based treatment plan for OA, consisting of patient education and neuromuscular exercise | Approx. 10,000 patients/year in Denmark (2013–2020) | Coverage by universal healthcare: 100% reimbursement when referred by an orthopaedic surgeon; 40% coverage when referred by a general practitioner; 0% when self-referred (patient must pay all costs) |
| OA HWFL (31–34) | Australia | Improvement of daily living and associated quality of life by reducing knee and hip pain and stiffness and improving mobility. Furthermore, improvement of preparation (pre-hab) for knee or hip replacement surgery (if relevant). | Approx. 1,125 patients/year (2008–2016) | Coverage by universal healthcare: 100% of programme costs if participants eligible (i.e., over 548 USD worth of products, service and support, provided for free) |
| AMSOA (35–38) | The Netherlands | Collection of data (AMS-OA cohort): Provision of insight into the relation between clinical characteristics and functional outcome of patients with hip or knee OA. | Approx. 82 patients/year (2009–2017) | Coverage by universal healthcare: Full, partial, or no coverage (according to the health care policy and supplementary insurance) |
| JIGSAW (39–41) | United Kingdom (EU) | Primary care support: - addressing the unmet needs of adults with OA, through the provision of innovations - systematic implementation of International Guidelines and Quality Standards for OA at practise level across European sites | Primary Care in Clinical Commissioning Groups (CCGs) UK. 3 groups, ~100 practises | Coverage by universal healthcare: 100% by the National Health Service (NHS). European Institute of Innovation & Technology (EIT)-Health funded Joint Implementation of osteoarthritis Guidelines in Western Europe (JIGSAW-E). Implementation of this funded model was being tested in the Netherlands, Norway, Denmark and Portugal. |
## Use of Framework for Programme Analysis
The RE-AIM framework was used to analyse the Reach, Effectiveness, Adoption, Implementation, and Maintenance of the six programmes [42]. Briefly described, “the reach dimension of the framework refers to the percentage and characteristics of individuals receiving the intervention; effectiveness refers to the impact of the intervention, including anticipated as well as unanticipated outcomes; adoption concerns the percentage and representativeness of settings that adopt the intervention; implementation refers to the consistency and cost of delivering the intervention; and maintenance refers to long-term sustainability at both the setting and individual levels” [43].
The dimensions manifest themselves at various implementation levels (individual, organisation, community) and can facilitate the investigation of the impact of a programme on public health in a specific health care system [44]. The RE-AIM dimensions were used to develop a survey to rate the applicability of each programme to the Swiss health care system. Glasgow et al. [ 15] stated that, “The use of the RE-AIM metrics might not result in a clear-cut decision, but it will facilitate the more informed and comprehensive consideration of all the relevant factors and make explicit the values and priorities” [42]. The RE-AIM website (www.re-aim.org) defines a score of 9–10 as “excellent,” 7–8 as “good, but needs a little work,” 5–6 as “fair, needs additional attention,” and <5 as “poor, needs serious attention.”
## Development of Survey
The questions in the survey were based on the key pragmatic planning questions for RE-AIM formulated by Glasgow and Estabrooks [44]. They were further developed for the survey to rate the applicability of the six chosen programmes for implementation in Switzerland [44]. The original RE-AIM questions were used as headings for each dimension. The context of the programmes in each dimension was explained in a short text and the developed questions posed to rate the applicability. These questions are displayed in Table 2.
**Table 2**
| BOA (8, 22, 23), Sweden | OACCP (25–27), Australia | GLA:D® (7, 29), Denmark | OA HWFL (web-based program) (31–33), Australia | AMSOA (35–37), The Netherlands | JIGSAW (39), United Kingdom |
| --- | --- | --- | --- | --- | --- |
| “Reach”—Who is intended to benefit and who actually participates in the intervention? | “Reach”—Who is intended to benefit and who actually participates in the intervention? | “Reach”—Who is intended to benefit and who actually participates in the intervention? | “Reach”—Who is intended to benefit and who actually participates in the intervention? | “Reach”—Who is intended to benefit and who actually participates in the intervention? | “Reach”—Who is intended to benefit and who actually participates in the intervention? |
| Symptoms of hip, knee, or hand OA (referred or self-referred) | Diagnosed knee or hip OA with pain ≥4/10 (referred) | Symptoms of, or diagnosed hip or knee OA (referred or self-referred) | Overweight (BMI ≥ 28) patients with diagnosed (supported by X-Ray) hip and knee OA, for pre-rehab | Diagnosed hip or knee OA; ≥18 years; non-traumatic pain | GP consultation because of hip, knee, hand, foot OA, and joint pain; ≥45 years |
| “Effectiveness”—What are the most important benefits (primary outcomes) you are trying to achieve and what is the likelihood of negative outcomes? | “Effectiveness”—What are the most important benefits (primary outcomes) you are trying to achieve and what is the likelihood of negative outcomes? | “Effectiveness”—What are the most important benefits (primary outcomes) you are trying to achieve and what is the likelihood of negative outcomes? | “Effectiveness”—What are the most important benefits (primary outcomes) you are trying to achieve and what is the likelihood of negative outcomes? | “Effectiveness”—What are the most important benefits (primary outcomes) you are trying to achieve and what is the likelihood of negative outcomes? | “Effectiveness”—What are the most important benefits (primary outcomes) you are trying to achieve and what is the likelihood of negative outcomes? |
| Pain, physical function, PA, QoL, and self-efficacy. Fear-avoidance behaviour and reported sick leave. | Pain, physical function, willingness for surgery, length of stay after surgery. | Pain, physical function, PA, QoL. Physical performance, self-efficacy. Use of painkillers, reported sick leave. | Weight loss, improved nutrition, QoL. | Pain, physical function, activity limitations (knee muscle strength), physical performance, and knee instability. | Uptake of Quality Indicators of NICE guidelines. |
| “Adoption”—Where (in outpatient setting) will the programme be applied and who (provider/health care professionals) will apply it? | “Adoption”—Where (in outpatient setting) will the programme be applied and who (provider/health care professionals) will apply it? | “Adoption”—Where (in outpatient setting) will the programme be applied and who (provider/health care professionals) will apply it? | “Adoption”—Where (in outpatient setting) will the programme be applied and who (provider/health care professionals) will apply it? | “Adoption”—Where (in outpatient setting) will the programme be applied and who (provider/health care professionals) will apply it? | “Adoption”—Where (in outpatient setting) will the programme be applied and who (provider/health care professionals) will apply it? |
| Where: Care centres | Where: Public hospitals, outpatient clinics | Where: Private practises and municipal rehabilitation centres | Where: Online and phone-based | Where: Rehabilitation centre | Where: Primary care practises |
| Who: PTs, OTs, expert patients | Who: Coordinated multidisciplinary team | Who: PTs and expert patients | Who: Coordinated care support team (pharmacists, dietitians, OTs, nurses) | Who: Coordinated multidisciplinary team | Who: GPs and practise nurses or PTs (depending on the country's primary care system) |
| “Implementation”- How consistently can the programme be delivered, are adaptations needed (regarding content and duration)? | “Implementation”- How consistently can the programme be delivered, are adaptations needed (regarding content and duration)? | “Implementation”- How consistently can the programme be delivered, are adaptations needed (regarding content and duration)? | “Implementation”- How consistently can the programme be delivered, are adaptations needed (regarding content and duration)? | “Implementation”- How consistently can the programme be delivered, are adaptations needed (regarding content and duration)? | “Implementation”- How consistently can the programme be delivered, are adaptations needed (regarding content and duration)? |
| Duration: 12 or more weeks | Duration: 6–8 weeks | Duration: 8 weeks | Duration: 18 or more weeks (up to 2 years) | Duration: 12 weeks | Duration: Consultation model and additional offer of 4 sessions |
| Content: Consultation at multidisciplinary clinic. Individualised exercises sessions (twice a week) in groups or home-based: strength, cardiovascular training. Face-to-face progress reassessments. | Content: OA education session, 6-8 weeks group or home-based exercise sessions. | Content: 2 OA education sessions, 12 Individualised exercise sessions (60 min/twice a week) in groups or home-based: NEMEX. | Content: (telehealth) 4 dietetic consultations and lifestyle education. PA plan and PT-developed exercises (strength, balance and mobility exercises). 3 phases of minimum 6 weeks: Motivate, Consolidate, Maintain. | Content: 12 weeks of group exercise sessions (60 min): knee joint stabilisation, muscle strength, performance of daily activity; and home exercises (5 days/week). Supplementary: Psychological support and medical management. | Content: OA education, including OA guidebook. Advice on exercise and PA. Additional offer: Analgesia and referral to the practise nurse for 4 sessions to support self-management. |
| “Maintenance”—When will the initiative become fully operational (system level), how long do results last (patient level), how long will the initiative be sustained? | “Maintenance”—When will the initiative become fully operational (system level), how long do results last (patient level), how long will the initiative be sustained? | “Maintenance”—When will the initiative become fully operational (system level), how long do results last (patient level), how long will the initiative be sustained? | “Maintenance”—When will the initiative become fully operational (system level), how long do results last (patient level), how long will the initiative be sustained? | “Maintenance”—When will the initiative become fully operational (system level), how long do results last (patient level), how long will the initiative be sustained? | “Maintenance”—When will the initiative become fully operational (system level), how long do results last (patient level), how long will the initiative be sustained? |
| Patient level: Follow-ups at 3 and 12 months | Patient level: Follow-ups at 12, 26, and 52 weeks | Patient level: Follow-ups at 3 and 12 months | Patient level: Follow-ups at 6 and 18 weeks, if longer duration: at least every 6 months | Patient level: Follow-ups at 6, 12, and 38 weeks | Patient level: No follow-up (consultation model) |
| System level: Operating nationwide. No referral needed for patients to participate. | System level: Operating nationwide. Included in different pathways, i.e., EMOS, a clinical pathway defined by orthopaedic surgeons. | System level: Operating nationwide. | System level: Included in national (and international) treatment guidelines for weight loss in OA. | System level: Operating in the rehabilitation centre where it was originally developed. | System level: Approximately 100 practises across 6 collaborating sites in 6 countries. |
For each of these programme characteristics, the RE-AIM dimensions were rated on a Likert scale from 0 (not applicable at all) to 10 (very applicable). The programmes were anonymized, and their characteristics presented in a random sequence to prevent programme identification.
## Analysis of the Ratings
The overall mean scores with standard deviations (SD) of each programme and all dimensions were calculated and represented in a bar chart to provide a visual display. The programmes scoring >7 for the various dimensions were considered to be (highly) applicable for implementation in Switzerland.
## Analysis of the Programmes Using RE-AIM
Table 2 shows the programme characteristics, based on the RE-AIM dimensions, rated by the survey experts.
Dimension “Reach”: all the programmes included patients with knee pain and/or knee OA and provided an intervention based on exercise and patient education. OA HWFL only included patients with knee OA and/or knee pain and a BMI > 28, with the focus mainly on support for weight management. JIGSAW was the only programme that had no exercise sessions included, giving only advice on exercise and physical activity.
Dimension “Effectiveness”: the impact of individualised exercise and education on pain, function and quality of life was assessed for all programmes with follow-up and showed a significant effect in all those outcomes. JIGSAW, which had no follow-up, improved the patient pathway by applying the National Institute for Health and Care Excellence (NICE) guideline recommendations.
Dimension “Adoption”: all programmes were well-established in their country of origin, and some in other countries also. BOA, OACCP, GLA:D®, AMSOA, and JIGSAW were offered in outpatient settings, such as clinics, care or rehabilitation centres, and clinical practises. OACCP was provided in hospitals and outpatient clinics, AMSOA in a rehabilitation centre, and OA HWFL in an online and phone-based setting. Both AMSOA and OACCP combined numerous health care professions in multidisciplinary teams, whilst the other programmes only involved two to three health care professions.
Dimension “Implementation”: the programmes were similar in their content (e.g., support for self-management, patient education, and exercise) and in the minimal duration of the programmes, but their exercise programmes differed in intensity and the opportunity to prolong. The duration of the programmes was normally a minimum of 6 weeks and maximum of 3 months. The OACCP and OA-HWFL programmes from Australia offered additional longer-term support when necessary. The programmes varied in their approaches, but, at a minimum, all programmes included consultation or educational sessions, together with recommendations on exercises or an exercise programme. The structure of the JIGSAW programme was different and focused on the first consultation in primary care and on patient education. JIGSAW, a consultation model, included an additional four sessions of patient education, but with no follow-up measures for patients. The supervised and home-based exercise programmes accorded with international guidelines but were applied with different concepts.
Dimension “Maintenance”: The follow-ups to measure the outcomes of the patients were assessed after completion of the exercise sessions and a few months after the end of the programme. The operationalisation of the programmes was analysed on the system level. All programmes showed a nationwide roll-out and, therefore, good “Adoption” by the target population (which influences “Maintenance”). AMSOA was the only exception, since the programme remained well-established only in the rehabilitation centre where it was originally developed. The six programmes were initiated between 2008 and 2013 and are all ongoing.
## Rating of Applicability
Six of the nine network members invited to participate ($66.7\%$) returned their ratings. The clinical practitioners representing two different specialist physiotherapy societies and a representative of a German universality of applied Sciences did not respond. Figure 1 displays the mean scores of the ratings on the RE-AIM dimensions for each programme. The means scores across the six programmes ranged between 5.9 and 9.4. The overall mean scores of each programme and the range of the mean scores of dimensions were: GLAD 9.4 (8.2–9.7), AMSOA 7.9 (7.0–8.7), BOA 7.8 (6.3–9.3), OACCP 7.0 (5.8–7.5), JIGSAW 6.3 (4.8–6.5), and OA HWFL 5.9 (2.7–8.2).
**Figure 1:** *Ratings of the RE-AIM dimensions for each programme.*
## Discussion
The comparative analysis and rating of the six programmes, following the RE-AIM framework, found that the GLA:D® programme would be the most applicable to the Swiss health care system. The result suggests that GLA:D® could be implemented successfully in Switzerland with only few adaptations.
## Implementation Considerations
The ratings provide guidance for planning both the implementation strategy and the activities on the different dimensions. Through consideration of the respective levels, the results of this rating assist in deciding which dimensions need more attention during the development of the implementation strategy and activities. For example, combining the locations (where) and the health care professionals involved (who) resulted in the lowest scores in “Adoption” for all programmes. In the complex and decentralised Swiss health care system, together with the high autonomy of the health care professions, an innovation would have a better chance of successful implementation when only a few key stakeholders were involved in the programme. Furthermore, an evidence-performance gap is present in the conservative management of knee OA, thus the programme is intended to be offered in the outpatient setting. GLA:D® is offered mainly in outpatient settings and only involves a few stakeholders. Therefore, the programme reached a relatively high score of 8.2, meaning “good, but needs a little work.” So, the team implementing GLA:D® would need to pay extra attention to “Adoption” during the implementation process. “ Adoption” is associated with the leading indicators of acceptability, appropriateness, and feasibility [45]. The implementation activities should, therefore, focus on those leading indicators and involve important stakeholders to maximise scores in “Adoption.” The standardisation of the outcome measures and reporting could improve interdisciplinary work between referring doctors and physiotherapists. However, OACCP stated that it was essential that all the health care professionals involved were convinced that the programme enhances conservative management [10].
In the other four dimensions, i.e., Reach, Effectiveness, Implementation, and Maintenance, GLA:D® showed a mean score of 9.5 or higher, which is defined as “excellent.” These dimensions would require fewer adaptations and need less attention during the implementation process. This means that targeting patients with knee OA or knee pain, with no further inclusion or exclusion criteria, seems to be a good option, according to the rating. Other programmes reached lower scores because they set an age limit, or only include referred patients, or overweight patients with a BMI ≥ 28. High scores for GLA:D® also resulted from its main focus being on the symptoms of knee OA, physical function, and the behaviour of the patients (in terms of fear-avoidance behaviour or sick leave). The additional factors in other programmes (e.g., willingness for surgery, weight control, nutrition), which are not intended to be included in the primary outcomes of a Swiss programme, might have resulted in lower programme scores. Regarding the content, the four programmes BOA, AMSOA, GLAD, and OACCP provide exercise sessions with supervision. Individualised and supervised exercise is applied to ensure sufficient load and progression and to support quality of performance for long-term self-management [46]. Supervised exercise may improve adherence and may also lead to better outcomes [22]. GLA:D® integrates neuromuscular exercises (NEMEX), i.e., functional knee stability, during the exercises. NEMEX have been proven to be safe and effective [32, 46]. Sensomotor control and functional stability should be supervised by a physiotherapist, because the quality of performance, i.e., in NEMEX, is crucial [46]. The four programmes combine supervised with home-based exercises for best results [46]. The high score (9.5) of GLA:D® for “Implementation,” with a difference of almost 2 compared to the other programme scores, indicates that the duration of 8 weeks, or 12 sessions, and the integration of NEMEX might be well-accepted by all parties, i.e., patients, providers, and insurers. In contrast, the costs of online treatments, or a longer duration, e.g., OA HWFL, might not be fully covered by Swiss insurers. High scores for “Maintenance” were reached in programmes with follow-ups at 3 and 12 months, together with the strategy for a nationwide rollout with no referral needed for participation. This seems to be the “Maintenance” strategy most compatible with the intended goals of a Swiss programme.
## Additional Offers for Improving Conservative Management
This analysis and the ratings show that also the lower scoring programmes, i.e., all the programmes except GLA:D®, have some important aspects that impact the implementation goals in Switzerland.
## Online Programme
After the initial implementation of a programme, some other aspects need to be considered for potential inclusion, e.g., in times of a global pandemic, an online programme has the advantage that patients can perform the exercises whenever, or wherever, they want. BOA provides an additional web-based version, named “Joint Academy” [47]. Although direct support from a physiotherapist showing how to perform the exercises would be lacking, some patients might prefer an online programme. Joint Academy offers online support from a physiotherapist for 6 weeks duration. Personalised exercises are delivered by email and supported by videos and take-home messages. Although an online exercise programme might not be supported financially by Swiss healthcare insurers, patients may choose to pay themselves for this type of offer.
## Weight Management
OA HWFL, which focused on weight reduction and nutrition, achieved lower scores for applicability for implementation in the Swiss health care system in most RE-Aim-dimensions. However, the working collaboration of referring medical doctors, physiotherapists, and nutritionists helped to improve knee OA management [48]. Overweight is a risk factor in developing knee OA and has a negative impact on the course of this chronic disease. Weight reduction improves knee pain and function [49, 50]. Weight management could, therefore, be beneficial and should be considered when the population shows many cases of knee OA correlating to overweight or obesity. This additional intervention, ideally complementary to an exercise and education programme, also has the potential to improve knee OA management [31].
## Written Information
The JIGSAW programme had the lowest mean score over all dimensions for applicability for implementation in the Swiss health care system, compared to the other programmes. Although patient education is included in the exercise and education programmes of, e.g., GLA:D®, BOA, or AMSOA, a written booklet elaborating the most important information on OA could further support patient self-management or motivate them to join an exercise and education programme.
## Strengths and Limitations
A strength of this study is the use of the RE-AIM dimensions to analyse and rate the chosen programmes to ascertain which would be most applicable for implementation in Switzerland, and to indicate potential challenges to the implementation process. A limitation of this analysis could be selection bias, since the information provided in the survey was the only information on the programmes that was either published in English or available on the respective websites. Another limitation may be that the non-responders included the two different specialist physiotherapy societies could therefore lead to underrepresentation of clinical practitioners, even though there was still one participant who is a researcher but also working in clinical practise. The underrepresentation could result in a lack of the opinions or perspectives of people who are specialised in the practical application of a programme, especially, on the feasibility of the contents, (i.e., dimension implementation) or the practicability of the programme when there are many different health care professionals included (i.e., dimension adoption). However, this study focused on the representatives of the main providers of such a programme in Switzerland, i.e., physiotherapists, who are important stakeholders in the future implementation of an exercise and education programme for the management of knee OA.
In Switzerland, there is a need for a structured and systematic programme for patients with knee OA, due to the high prevalence of the disease and the lack of knowledge of the beneficial effects of exercise and education. A renowned and established programme might be favourably accepted and contribute to closing the existing evidence-performance gap in clinical practise. Furthermore, such a programme would represent an improvement in non-surgical and non-pharmacological management and follow-up. In conclusion, the GLA:D® programme with the highest scores has already been implemented in other countries. The programme consists of exercise and education and scored higher than 7 in all RE-AIM dimensions. Therefore, this programme is most applicable to the Swiss health care system as only few adaptations would be needed for its successful implementation in Switzerland.
## Data Availability Statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics Statement
This study did not fall within the scope of the Swiss Human Research Act and authorisation from an Ethics Committee was not required. Raters were informed that by participation in the survey they automatically provided their informed consent and that their ratings were used in an anonymised manner.
## Author Contributions
LE and KN were contributing to conception and design of the study. LE collected and analysed the programme information and data of the ratings. KN contributed to the drafting and revision of the manuscript. All authors have read and approved the manuscript.
## Conflict of Interest
KN is a member of the network of experts in physiotherapy for knee OA management. However, KN did not participate in the survey and the ratings. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. McAlindon TE, Bannuru RR, Sullivan MC, Arden NK, Berenbaum F, Bierma-Zeinstra SM. **OARSI guidelines for the non-surgical management of knee osteoarthritis**. *Osteoarthritis Cartilage.* (2014.0) **22** 363-88. DOI: 10.1016/j.joca.2014.01.003
2. Fernandes L, Hagen KB, Bijlsma JWJ, Andreassen O, Christensen P, Conaghan PG. **EULAR recommendations for the non-pharmacological core management of hip and knee osteoarthritis**. *Ann Rheum Dis.* (2013.0) **72** 1125-35. DOI: 10.1136/annrheumdis-2012-202745
3. Hochberg MC, Altman RD, April KT, Benkhalti M, Guyatt G, McGowan J. **American College of Rheumatology 2012 recommendations for the use of nonpharmacologic and pharmacologic therapies in osteoarthritis of the hand, hip, and knee**. *Arthritis Care Res.* (2012.0) **64** 465-74. DOI: 10.1002/acr.21596
4. Imoto AM, Pardo JP, Brosseau L, Taki J, Desjardins B, Thevenot O. **Evidence synthesis of types and intensity of therapeutic land-based exercises to reduce pain in individuals with knee osteoarthritis**. *Rheumatol Int.* (2019.0) **39** 1159-79. DOI: 10.1007/s00296-019-04289-6
5. Fransen M, McConnell S, Harmer AR, Van der Esch M, Simic M, Bennell KL. **Exercise for osteoarthritis of the knee**. *Cochrane Database Syst Rev.* (2015.0) **1** CD004376. DOI: 10.1002/14651858.CD004376.pub3
6. Li CS, Karlsson J, Winemaker M, Sancheti P, Bhandari M. **Orthopedic surgeons feel that there is a treatment gap in management of early OA: international survey**. *Knee Surg Sports Traumatol Arthrosc.* (2014.0) **22** 363-78. DOI: 10.1007/s00167-013-2529-5
7. Skou ST, Roos EM. **Good Life with osteoArthritis in Denmark (GLA:D**. *BMC Musculoskelet Disord.* (2017.0) **18** 72. DOI: 10.1186/s12891-017-1439-y
8. Thorstensson CA, Garellick G, Rystedt H, Dahlberg LE. **Better management of patients with osteoarthritis: development and nationwide implementation of an evidence-based supported osteoarthritis self-management programme**. *Musculoskeletal Care.* (2015.0) **13** 67-75. DOI: 10.1002/msc.1085
9. Dziedzic KS, Allen KD. **Challenges and controversies of complex interventions in osteoarthritis management: recognizing inappropriate and discordant care**. *Rheumatology.* (2018.0) **57** iv88-98. DOI: 10.1093/rheumatology/key062
10. Allen KD, Choong PF, Davis AM, Dowsey MM, Dziedzic KS, Emery C. **Osteoarthritis: models for appropriate care across the disease continuum**. *Best Pract Res Clin Rheumatol.* (2016.0) **30** 503-35. DOI: 10.1016/j.berh.2016.09.003
11. Eyles JP, Hunter DJ, Bennell KL, Dziedzic KS, Hinman RS, van der Esch M. **Priorities for the effective implementation of osteoarthritis management programs: an OARSI international consensus exercise**. *Osteoarthritis Cartilage.* (2019.0) **27** 1270-9. DOI: 10.1016/j.joca.2019.02.704
12. 12.Swiss Federal Statistical Office. Gesundheitsstatistik (2014). Available online at: https://www.bfs.admin.ch/bfs/fr/home/statistiques/sante.assetdetail.349483.html (accessed February 1, 2021).. *Gesundheitsstatistik* (2014.0)
13. Ettlin L, Nast I, Huber EO, Niedermann K. **Does the conservative non-pharmacological management of knee osteoarthritis in Switzerland reflect the clinical guidelines? A survey among general practitioners, rheumatologists, and orthopaedic surgeons**. *Front Rehabilit Sci.* (2021.0) **2** 658831. DOI: 10.3389/fresc.2021.658831
14. Kirchner JE, Smith JL, Powell BJ, Waltz TJ, Proctor EK. **Getting a clinical innovation into practice: an introduction to implementation strategies**. *Psychiatry Res.* (2020.0) **283** 112467. DOI: 10.1016/j.psychres.2019.06.042
15. Glasgow RE, Harden SM, Gaglio B, Rabin B, Smith ML, Porter GC. **RE-AIM planning and evaluation framework: adapting to new science and practice with a 20-year review**. *Front Public Health.* (2019.0) **7** 64. DOI: 10.3389/fpubh.2019.00064
16. 16.OECD. Health at a Glance 2017: OECD Indicators. Paris: OECD Publishing (2017). Available online at: https://www.oecd-ilibrary.org/social-issues-migration-health/health-at-a-glance-2017_health_glance-2017-en (accessed May 25, 2021).. *Health at a Glance 2017: OECD Indicators* (2017.0)
17. Lauvergeon S, Mettler D, Burnand B, Peytremann-Bridevaux I. **Convergences and divergences of diabetic patients' and healthcare professionals' opinions of care: a qualitative study**. *Health Expect.* (2015.0) **18** 111-23. DOI: 10.1111/hex.12013
18. Durlak JA, DuPre EP. **Implementation matters: a review of research on the Influence of implementation on program outcomes and the factors affecting implementation**. *Am J Community Psychol.* (2008.0) **24** 327-50. DOI: 10.1007/s10464-008-9165-0
19. Domenech Rodriguez MM, Bernal G. *Cultural Adaptations: Tools for Evidence-Based Practice with Diverse Populations* (2012.0)
20. Domenech Rodríguez MM, Bernal G, Bernal G, Domenech Rodríguez MM. **Bridging the gap between research and practice in a multicultural world**. *Cultural Adaptations: Tools for Evidence-Based Practice With Diverse Populations* (2012.0) 265-87. DOI: 10.1037/13752-013
21. Bernal G, Domenech Rodríguez MM, Bernal G, Domenech Rodríguez MM. **Cultural adaptation in context: psychotherapy as a historical account of adaptations**. *Cultural Adaptations: Tools for Evidence-Based Practice with Diverse Populations* (2012.0) 3-22. DOI: 10.1037/13752-001
22. Dell'Isola A, Jönsson T, ranstam J, Dahlberg LE, Hansson EE. **Education, home exercise, and supervised exercise for people with hip and knee osteoarthritis as part of a nationwide implementation program: data from the better management of patients with osteoarthritis registry**. *Arthritis Care Res* (2020.0) **72** 201-7. DOI: 10.1002/acr.24033
23. Jönsson T, Ekvall Hansson E, Thorstensson CA, Eek F, Bergman P, Dahlberg LE. **The effect of education and supervised exercise on physical activity, pain, quality of life and self-efficacy - an intervention study with a reference group**. *BMC Musculoskelet Disord.* (2018.0) **19** 198. DOI: 10.1186/s12891-018-2098-3
24. 24.BOA-registret (2021). Available online at: https://boa.registercentrum.se/boa-in-english/better-management-of-patients-with-osteoarthritis-boa/p/By_o8GxVg (accessed June 3, 2021).. (2021.0)
25. Teoh LSG, Eyles JP, Makovey J, Williams M, Kwoh CK, Hunter DJ. **Observational study of the impact of an individualized multidisciplinary chronic care program for hip and knee osteoarthritis treatment on willingness for surgery**. *Int J Rheum Dis.* (2017.0) **20** 1383-92. DOI: 10.1111/1756-185X.12950
26. Eyles JP, Mills K, Lucas BR, Williams MJ, Makovey J, Teoh L. **Can we predict those with osteoarthritis who will worsen following a chronic disease management program?**. *Arthritis Care Res.* (2016.0) **68** 1268-77. DOI: 10.1002/acr.22836
27. Claes B-EA, Leung HWC, Matters K, Williams MJ, Hunter DJ. **Interim analysis: an interdisciplinary team approach in facilitating weight reduction and improving function for people with knee or hip osteoarthritis. the osteoarthritis chronic care program at royal north shore hospital**. *Nutr Dietet.* (2015.0) **72** 232-9. DOI: 10.1111/1747-0080.12166
28. 28.Agency for Clinical Innovation (ACI). Osteoarthritis Chronic Care Program Model of Care. (2020). Available online at: https://aci.health.nsw.gov.au/resources/musculoskeletal/osteoarthritis_chronic_care_program/osteoarthritis-chronic-care-program (accessed June 3, 2021).. *Osteoarthritis Chronic Care Program Model of Care* (2020.0)
29. Skou ST, Pedersen BK, Abbott JH, Patterson B, Barton C. **Physical activity and exercise therapy benefit more than just symptoms and impairments in people with hip and knee osteoarthritis**. *J Orthop Sports Phys Ther.* (2018.0) **48** 439-47. DOI: 10.2519/jospt.2018.7877
30. 30.GLA:D International (2021). Available online at: https://gladinternational.org/ (accessed June 3, 2021).. (2021.0)
31. Atukorala I, Makovey J, Lawler L, Messier SP, Bennell K, Hunter DJ. **Is there a dose-response relationship between weight loss and symptom improvement in persons with knee osteoarthritis?**. *Arthritis Care Res.* (2016.0) **68** 1106-14. DOI: 10.1002/acr.22805
32. McGill B, O'Hara BJ, Grunseit AC, Bauman A, Lawler L, Phongsavan P. **Healthy weight for life programme: evaluating the practice and effectiveness of a weight loss maintenance programme in the private health insurance setting**. *SAGE Open Med* (2019.0) **7** 2050312119873814. DOI: 10.1177/2050312119873814
33. Makovey J, Lawler L, Bennell KL, Hunter DJ. **Dose response relationship between weight loss and improvement in quality of life in persons with symptomatic knee osteoarthritis**. *Osteoarthritis Cartilage.* (2015.0) **23** A386. DOI: 10.1016/j.joca.2015.02.713
34. 34.OAHWFL (2021). Available online at: https://healthyweightforlife.com.au/osteoarthritis-hwfl/ (accessed June 3, 2021).. (2021.0)
35. Knoop J, Leeden M van der, Esch M van der, Thorstensson CA, Gerritsen M, Voorneman RE. **Association of lower muscle strength with self-reported knee instability in osteoarthritis of the knee: results from the Amsterdam Osteoarthritis Cohort**. *Arthritis Care Res.* (2012.0) **64** 38-45. DOI: 10.1002/acr.20597
36. Knoop J, Dekker J, van der leeden M, Van der Esch M, Thorstensson CA, Gerritsen M. **Knee joint stabilization therapy in patients with osteoarthritis of the knee: a randomized, controlled trial**. *Osteoarthrit Cartil* (2013.0) **21** 1025-34. DOI: 10.1016/j.joca.2013.02.310
37. Sanchez-Ramirez DC, van der leeden, van der esch. **Increased knee muscle strength is associated with decreased activity limitations in established knee osteoarthritis: two-year follow-up study in the Amsterdam osteoarthritis cohort**. *J Rehabil Med* (2015.0) **47** 647-54. DOI: 10.2340/16501977-1973
38. **ResearchGate**
39. Schiphof D, Vlieland TPV, van Ingen R, Peter WF, Meesters JJ, de Wit MP. **Joint implementation of guidelines for osteoarthritis in Western Europe: JIGSAW-E in progress in The Netherlands**. *Osteoarthritis Cartilage.* (2017.0) **25** 414. DOI: 10.1016/j.joca.2017.02.714
40. Hay E, Dziedzic K, Foster N, Peat G. *Optimal Primary Care Management of Clinical Osteoarthritis and Joint Pain in Older People: A Mixed-Methods Programme of Systematic Reviews, Observational and Qualitative Studies, and Randomised Controlled Trials* (2018.0) 294. DOI: 10.3310/pgfar06040
41. 41.WMAHSN (2020). Available online at: https://www.wmahsn.org/programmes/view/joint-implementation-of-guidelines-for-osteoarthritis-oa-in-western-europe-jigsaw-e (accessed June 3, 2021).. (2020.0)
42. Glasgow RE, Klesges L M, Dzewaltowski DA, Estabrooks PE, Vogt TM. **Evaluating the impact of health promotion programs: using the RE-AIM framework to form summary measures for decision making involving complex issues**. *Health Educ Res.* (2006.0) **21** 688-94. DOI: 10.1093/her/cyl081
43. King DK, Glasgow RE, Leeman-Castillo B. **Reaiming RE-AIM: using the model to plan, implement, and evaluate the effects of environmental change approaches to enhancing population health**. *Am J Public Health.* (2010.0) **100** 2076-84. DOI: 10.2105/AJPH.2009.190959
44. Glasgow RE, Estabrooks PE. **Pragmatic applications of RE-AIM for health care initiatives in community and clinical settings**. *Prev Chronic Dis.* (2018.0) **15** E02. DOI: 10.5888/pcd15.170271
45. Weiner BJ, Lewis CC, Stanick C, Powell BJ, Dorsey CN, Clary AS. **Psychometric assessment of three newly developed implementation outcome measures**. *Implement Sci.* (2017.0) **12** 108. DOI: 10.1186/s13012-017-0635-3
46. Ageberg E, Roos EM. **Neuromuscular exercise as treatment of degenerative knee disease**. *Exerc Sport Sci Rev.* (2015.0) **43** 14-22. DOI: 10.1249/JES.0000000000000030
47. Dahlberg LE, Grahn D, Dahlberg JE, Thorstensson CA. **A web-based platform for patients with osteoarthritis of the hip and knee: a pilot study**. *JMIR Res Protoc.* (2016.0) **5** e115. DOI: 10.2196/resprot.5665
48. Hunter DJ, Hinman RS, Bowden JL, Egerton T, Briggs AM, Bunker SJ. **Effectiveness of a new model of primary care management on knee pain and function in patients with knee osteoarthritis: Protocol for THE PARTNER STUDY**. *BMC Musculoskelet Disord.* (2018.0) **19** 132. DOI: 10.1186/s12891-018-2048-0
49. Gandhi R, Wasserstein D, Razak F, Davey JR, Mahomed NN. **BMI independently predicts younger age at hip and knee replacement**. *Obesity.* (2010.0) **18** 2362-6. DOI: 10.1038/oby.2010.72
50. Midgley J. **An evidence review of osteoarthritis, with focus on the obese – considering conservative management, multi-morbidity, surgery and the implications of restricted access to knee or hip replacement because of weight**. *Int J Orthop Trauma Nurs.* (2020.0) **40** 100840. DOI: 10.1016/j.ijotn.2020.100840
|
---
title: Safety and tolerability of Sacubitril/Valsartan in heart failure patient with
reduced ejection fraction
authors:
- Muhammad Nauman Khan
- Najia Aslam Soomro
- Khalid Naseeb
- Usman Hanif Bhatti
- Rubina Rauf
- Iram Jehan Balouch
- Ali Moazzam
- Sonia Bashir
- Tariq Ashraf
- Musa Karim
journal: BMC Cardiovascular Disorders
year: 2023
pmcid: PMC10012729
doi: 10.1186/s12872-023-03070-9
license: CC BY 4.0
---
# Safety and tolerability of Sacubitril/Valsartan in heart failure patient with reduced ejection fraction
## Abstract
### Background
Angiotensin receptor blocker and a neprilysin inhibitor (ARNI) has emerged as an innovative therapy for patients of heart failure with reduced ejection fraction (HFrEF). The purpose of this study was to assess the safety and tolerability of Sacubitril/Valsartan in patient with HFrEF in Pakistani population.
### Methods
This proof-of-concept, open label non-randomized clinical trial was conducted at a tertiary care cardiac center of Karachi, Pakistan. Patients with HFrEF were prescribed with Sacubitril/Valsartan and followed for 12 weeks for the assessment of safety and tolerability. Safety measures included incidence of hypotension, renal dysfunction, hyperkalemia, and angioedema.
### Results
Among the 120 HFrEF patients, majority were male ($79.2\%$) with means age of 52.73 ± 12.23 years. At the end of 12 weeks, four ($3.3\%$) patients died and eight ($6.7\%$) dropped out of the study. In the remaining 108 patients, $80.6\%$ [87] of the patients were tolerant to the prescribed dose. Functional class improved gradually with $75.0\%$ [81] in class I and $24.1\%$ [26] in class II, and only one ($0.9\%$) patient in class III at the end of 12 weeks. Hyperkalemia remains the main safety concern with incidence rate of $21.3\%$ [23] followed by hypotension in $19.4\%$ [21], and renal dysfunction in $3.7\%$ [4] of the patients.
### Conclusions
Sacubitril/Valsartan therapy in HFrEF patients is safe and moderately tolerated among the Pakistani population. It can be used as first line of treatment for these patients.
### Trial registration
NCT05387967. Registered 24 May 2022—Retrospectively registered, https://clinicaltrials.gov/ct2/show/NCT05387967
## Background
Heart failure (HF) is a prominent cause of morbidity and death, as well as a major financial burden on healthcare systems worldwide [1–3]. The pathogenesis of HF is based on over-activation of neurohumoral mechanisms. Angiotensin-converting enzyme inhibitors (ACEIs) have been found to lower overall HF mortality by 16–$40\%$ since the release of the CONSENSUS study in 1987 and the SOLVD-Treatment trial in 1991 [4, 5]. In 2011, the EMPHASIS-HF study [6] verified and expanded the use of MRA eplerenone in patients with mild symptomatic HF. The foundations of contemporary HF treatment are these neurohumoral antagonists.
HF remains a significant source of morbidity and mortality despite various therapies aimed at neurohumoral blocking. Sacubitril/*Valsartan is* a novel combination drug containing an angiotensin receptor blocker (ARB) (Valsartan) and a neprilysin inhibitor (Sacubitril) (ARNI) approved by the US food and Drug Administration (FDA) and the European Medicines Agency (EMA) for the treatment of patients with HF with reduced ejection fraction (HFrEF) [7–9]. Following the release of the PARADIGM-HF study in 2014 [10], another paradigm change in HF therapy occurred. When compared to Enalapril, use of ARNI had lowered all-cause mortality by $16\%$ and cardiovascular mortality by $20\%$.
The American College of Cardiology (ACC)/American Heart Association (AHA) and the European Society of Cardiology (ESC), according to this study, recently revised evidence-based recommendations for the management of HF [7–9]. Both clinical practice guidelines recommended Sacubitril/Valsartan as a class I indication in patients with persistent symptomatic HF with reduced ejection fraction (HFrEF) despite optimum therapy.
Since there is a scarcity of local evidence regarding safety and tolerability of using ARNI in HFrEF patients. Therefore, this study was conducted with an aim of to assess the safety and tolerability of Sacubitril/Valsartan, over a period of 12 weeks, in HFrEF patients coming to a tertiary care public sector hospital located in Karachi.
## Methods
This study was designed as an open label non-randomized clinical trial being conducted in the Adult Cardiology Department of National Institute of Cardiovascular Diseases (NICVD), Karachi, Pakistan during January 2021 to June 2021. The sample for this study was calculated to be $$n = 121$$, keeping the confidence level at $95\%$ and a $7\%$ margin of error with anticipated treatment success of $81.1\%$ [11].
In clinical trial was performed in accordance with the Declaration of Helsinki, study was started after obtaining approval from ethical review committee of NICVD (ERC-$\frac{05}{2020}$) and written informed consent was obtained from all the patients regarding their participation in the study and publication of data while maintaining confidentiality and anonymity. This clinical trial was registered at ClinicalTrial.gov with NCT05387967. The required number of consecutive patients meeting the inclusion criteria were recruited for this study. After obtaining written informed consent, patient’s demographic and baseline clinical characteristics were obtained. Inclusion criteria for the study are either gender, between 18 and 80 years of age, diagnosed with HFrEF with New York Heart Association (NYHA) class II-IV, and left ventricular ejection fraction (LVEF) ≤ $40\%$. Pre-inclusion safety parameter were patients who were stable on any dose of beta blockers, ACEI or ARB prior to enrolment in the study. Patients who refused to participate in the study or patients with hyperkalemia (baseline potassium > 5.2 mmol/L), hypotension (baseline systolic blood pressure (SBP) < 90 mmHg), renal dysfunction (baseline estimated glomerular filtration rate (eGFR) < 30 mL/min), anemia (hemoglobin level: < 13.5 g/dL in men and < 12.0 g/dL in women), and history of hypersensitivity to the active substances, Sacubitril/Valsartan, or to any of the excipients or drugs of similar chemical classes were excluded from the study (Fig. 1). In addition to the study specific exclusion criteria (mentioned above), the pre-recruitment initial screen criteria were patients with the history of prior heart failure hospitalization, on anti-coagulants, receiving Sodium/glucose cotransporter-2 inhibitors (SGLT2i) therapy, or with either CRT (Cardiac Resynchronization Therapy) or ICD (implantable cardioverter-defibrillator) device. These patients were screened out from the study considering the potential confounding role. Fig. 1Study flow chart. HFrEF heart failure with reduced ejection fraction All the recruited patients were prescribed with Sacubitril/Valsartan at a starting dose of 50 ($\frac{24}{26}$) mg BID which was up-titrated, over the period of initial 6 weeks, to the maximum tolerated dose up to 200 ($\frac{97}{103}$) mg BID and further followed for a total of 12 weeks. A weekly telephonic follow-up was made to assess the patient’s medication adherence level and any adverse events. All the patients were kept under a close follow-up for the period of 12 weeks, the safety and tolerability outcomes were assessed.
Safety parameters included incidence of any of the following during 12 weeks of follow-up period; hypotension (SBP < 90 mmHg), renal dysfunction (eGFR < 30 mL/min), hyperkalemia (potassium > 5.2 mmol/L), and angioedema (rapid edema, or swelling, of the area beneath the skin or mucosa). Tolerability was defined as the dose tolerated by the patients which did not require down titration or discontinuation of prescribed dose during follow-up with the frequency of follow-up after 1st week, 2nd week, 4th week, 8th week, and 12th week. Functional status of all the patients was also assessed as NYHA functional class on every follow-up week. Echocardiography was performed, at baseline as well as at 12th week follow-up, and improvement in left ventricular ejection fraction, systolic, and diastolic dimension were evaluated.
All the collected information was recorded using a predefined structural proforma. Collected data were analyzed using SPSS version-21 (IBM Corp). Mean ± standard deviation (SD) were computed for the quantitative (continuous) variables, such as hemodynamic parameters, laboratory parameters, and echocardiographic parameters, and paired samples t-test was conducted to compare baseline with 12th week assessments. NYHA classification before and at the end of 12-weeks therapy was compared with the help of Chi-square test. Frequency and percentages were calculated for categorical variables such as tolerability and safety measures and Chi-square test/ Fisher's exact test were applied for the compare the results for various baseline characteristics of the patients.
## Baseline demographics
A total of 120 enrolled patients meeting eligibility criteria with successful 12 weeks follow-up were enrolled in the study and out of which four ($3.3\%$) died during the study duration and eight ($6.7\%$) patients discontinued medication on their own, hence excluded from the final analysis. One of these patient moved out of town and refused to visit for follow-up after first week of treatment, and remaining seven patients (5 after 1st week and 2 after 2nd week of treatment) discontinued the study medication and self-withdrawn from the study with no reported complications or symptoms. Out of the remaining 108 patients, $78.7\%$ [85] were male and the mean age of study patients was 53.04 ± 11.89 years. Whereas, $64.8\%$ [70] of the study population had coronary artery disease, $56.5\%$ [61] had diabetes mellitus, and $41.7\%$ [45] had hypertension as the comorbidity (Table 1). A total of $65.8\%$ ($\frac{79}{120}$) were on beta blockers, $24.2\%$ ($\frac{29}{120}$) were on diuretics, and $4.2\%$ ($\frac{5}{120}$) patients were on digoxin. Table 1Distribution of baseline demographic characteristics of the study patientsTotalTotal (N)120Gender Male$79.2\%$ [95] Female$20.8\%$ [25]Age (years)52.73 ± 12.23 ≤ 50 years$38.3\%$ [46] > 50 years$61.7\%$ [74]Weight (kg)70.08 ± 15.79Risk profile Hypertension$44.2\%$ [53] Diabetes mellitus$55.8\%$ [67] Smoking$35\%$ [42]Coronary artery disease$65.8\%$ [79]Cerebrovascular accident (CVA)/stroke$0.0\%$ [0]Atrial fibrillation$9.2\%$ [11]
## Clinical characteristics
A significant improvement ($$p \leq 0.010$$) in NYHA function class was observed after 12 weeks of therapy with a majority of patients, $75\%$ [81], in class I at 12 weeks as against majority in class II, $71.3\%$ [77], at baseline. Similarly, a significant improvement in left ventricular EF was also observed from 26.71 ± $5.35\%$ at baseline to 33.36 ± $12.06\%$ after 12-week therapy along with significant improvement in systolic dimensions (Table 2).Table 2Comparison of functional class, hemodynamics, laboratory, and echocardiographic parameters at baseline and after 12 weeksCharacteristicsBaselineAt 12th weekp valueTotal (N)108108–Hemodynamics Systolic blood pressure (mmHg)119.68 ± 25.23112.15 ± 19.890.002* Diastolic blood pressure (mmHg)70.48 ± 16.2766.83 ± 13.820.019*Laboratory parameters Creatinine (mg/dL)1.05 ± 0.321.05 ± 0.270.967 Potassium (mg/dL)4.39 ± 0.544.55 ± 0.40.005* eGFR (mL/min)77.42 ± 27.7175.89 ± 26.520.582Echocardiography Ejection fraction (%)26.71 ± 5.3533.36 ± 12.06< 0.001* Systolic dimensions (mm)45.83 ± 9.9442.52 ± 11.440.004* Diastolic dimensions (mm)57.55 ± 8.756.71 ± 9.630.228NYHA functional class I$0\%$ [0]$75\%$ [81]0.010* II$71.3\%$ [77]$24.1\%$ [26] III$25\%$ [27]$0.9\%$ [1] IV$3.7\%$ [4]$0\%$ [0]eGFR estimated glomerular filtration rate, NYHA New York Heart Association*Significant at $5\%$ The distribution of NYHA classification on every follow-up week is presented in Fig. 2A. Functional class improved by at least one NYHA class in $88\%$ of the patients, remained the same in $12\%$, and did not deteriorated in any of the patient. Functional class improvement in male and female patients is provided in Fig. 2B.Fig. 2Distribution of the New York Heart Association Functional Classification on every follow-up week (A) and functional class improvement status by gender (B). NYHA New York Heart Association Functional Classification
## Safety
When measuring the safety of the drug in the participants, four clinical parameters were taken into consideration i.e., the occurrence of either hypotension, hyperkalemia, renal dysfunction, or angioedema among the participants. During the study duration of 12 weeks, hyperkalemia remains the main safety concern with incidence rate of $21.3\%$ [23] followed by hypotension in $19.4\%$ [21], and renal dysfunction in $3.7\%$ [4] of the patients. However, none of the patient developed angioedema during this study. When assessed for various baseline characteristics, the incidence rate of hypotension was significantly higher ($$p \leq 0.017$$) among non-hypertensive patients with an incidence rate of $29.8\%$ ($\frac{14}{37}$) as against $11.5\%$ ($\frac{7}{50}$) among hypertensive patients. Female patients had higher incidence of renal dysfunction compared to male patients with incidence rate of $13\%$ ($\frac{3}{19}$) vs. $1.2\%$ ($\frac{1}{68}$), $$p \leq 0.030$$, respectively. Hyperkalemia was found to be associated with the presence of diabetes ($$p \leq 0.010$$) with the incidence rate of $33.3\%$ ($\frac{15}{36}$) compared to $12.7\%$ ($\frac{8}{51}$) for non-diabetic patients.
## Tolerability
A total of $80.6\%$ [87] of the patients were tolerant to the prescribed dose with the distribution of $8\%$ ($\frac{7}{87}$) on 50 mg BID, $28.7\%$ ($\frac{25}{87}$) on 100 mg BID, and remaining $63.2\%$ ($\frac{55}{87}$) on 200 mg BID. A total of 21 ($19.4\%$) patients required either down titration or temporary discontinuing of the prescribed medication, 10 patients due to increase complained of vertigo, 5 patients got symptomatic, 3 patients due to increase in creatinine level, and 3 patients due to increase in potassium level. Tolerance was observed to be unaffected by the demographic and baseline characteristics of the patients, shown in Table 3.Table 3Comparison of tolerability and safety measures during 12-weeks of therapy by various baseline and demographic characteristicsCharacteristicsTolerabilitySafety measurementsHypotensionRenal dysfunctionHyperkalemiaGender Male$80\%$ [68]$22.4\%$ [19]$1.2\%$ [1]$18.8\%$ [16] Female$82.6\%$ [19]$8.7\%$ [2]$13\%$ [3]$30.4\%$ [7]p value> 0.9990.2340.030*0.256Age ≤ 50 years$82.5\%$ [33]$22.5\%$ [9]$0\%$ [0]$17.5\%$ [7] > 50 years$79.4\%$ [54]$17.6\%$ [12]$5.9\%$ [4]$23.5\%$ [16]p value0.6950.5380.2940.460Hypertension No$78.7\%$ [37]$29.8\%$ [14]$2.1\%$ [1]$21.3\%$ [10] Yes$82\%$ [50]$11.5\%$ [7]$4.9\%$ [3]$21.3\%$ [13]p value0.6730.017*0.6310.996Diabetes mellitus No$81\%$ [51]$19\%$ [12]$3.2\%$ [2]$12.7\%$ [8] Yes$80\%$ [36]$20\%$ [9]$4.4\%$ [2]$33.3\%$ [15]p value0.9020.902> 0.9990.010*Smoking No$80\%$ [56]$17.1\%$ [12]$5.7\%$ [4]$24.3\%$ [17] Yes$81.6\%$ [31]$23.7\%$ [9]$0\%$ [0]$15.8\%$ [6]p value0.8430.4120.2950.303*Significant at $5\%$
## Discussion
Despite a significant improvement in the management of HFrEF in recent years, it remains the major cause of morbidity and mortality in cardiac patients. With the favorable results from various clinical trials [10–12], Sacubitril/*Valsartan is* a class I indication in patients with HFrEF [7–9]. Nonetheless, due to its dual mode of action, its safety and tolerability along with optimal dosage and up-titration remained a major concern among the physicians. Additionally, effectiveness of Sacubitril/Valsartan in Asian HF patients is still a point of discussion as its benefits in the Asian sub-group of PARADIGM-HF trial fell short of reaching the statistical significance [13]. Therefore, in the current study our aim was to implement a systematic up-titration regime along with the assessment of safety and tolerability of Sacubitril/Valsartan for HFrEF patients in Pakistani population. After 12 weeks of observation, a majority $80.6\%$ of the patients well tolerated the prescribed therapy. A significantly improved functional class was also witnessed with $75.0\%$ in class I and $24.1\%$ in class II, and only one ($0.9\%$) patient in class III at the end of 12 weeks. Although, hyperkalemia ($21.3\%$) remains the main safety concern followed by hypotension ($19.4\%$), and renal dysfunction ($3.7\%$) in our population.
In the PARADIGM-HF study hypotension was the leading safety concern among South Asian HF patients received Sacubitril/Valsartan with the incidence rate of $10.5\%$ followed by hyperkalaemia in $8.9\%$, and angioedema in $0.3\%$ [13]. Corresponding incidence rates in our study are higher than PARADIGM-HF possibly because of differences in population physical characteristics such as shorter structure and lower body weight, which is why comparatively lower doses of drugs are advocated for the Asians [14, 15]. Additionally, the incidence of hypotension was found to be more common among normotensive patients, female patients had higher risk of renal dysfunction, and hyperkalemia was found to be more common in diabetic patients. This finding could be possibly due to some extent of sub-clinical renal dysfunction. The PARALLEL-HF study reported safety and well tolerance of Sacubitril/Valsartan among Japanese patients with HFrEF compared to enalapril [16].
Findings of our study reading tolerability of Sacubitril/Valsartan in our population are reassuring with more than $80\%$ tolerability similar to the TITRATION study where tolerability success was achieved in $85.2\%$ of the patients treated with Sacubitril/Valsartan [11]. A maximum dose of 200 mg bid was selected in this study as this level of dose was well tolerated in both TITRATION and PARADIGM-HF studies [11, 13]. The TITRATION study evaluated gradual initiation/up-titration from 50 to 200 mg bid over three or six weeks [17]. Gradual up-titration regime over 6 weeks, compared to 3 weeks, was more effective in achieving and maintaining the target dose of Sacubitril/Valsartan. Therefore, in our study we adopted stepwise up-titration regime from initial does of 50 to the target dose of 200 mg bid over a 6 weeks period. The up-titration regime was in accordance with safety and tolerability of the patient with the dosage rather than forced up-titration. Among the non-tolerating patients we observed increased complains of vertigo, symptomatic, increased creatinine and potassium level, hence, a systematic titration based on patients physical as well as laboratory examination with close monitoring can improve the tolerability of Sacubitril/Valsartan in these patients. In real-life clinical practice, several patient related (such as race, age, or co-morbid conditions) as well as system related (availability or accessibility to the health care system) factors can influence the tolerability of Sacubitril/Valsartan among these patients. A study by Hsu et al. [ 18] reported the PREDICT-HF model to be a useful clinical model for risk stratification of patients with HFrEF. The permanent discontinuation of Sacubitril/Valsartan therapy as found to be 8.3 per 100 patient-years for high risk patients as compared to 2.5 per100 patient-years for patients with standard-risk. The commonest reported reason for discontinuation of Sacubitril/Valsartan therapy was hypotension ($37.9\%$) followed by hyperkalemia or renal impairment ($19.7\%$), and allergic reaction or adverse effects ($13.6\%$) [18].
Some of the recent studies have demonstrated safety and efficacy Sacubitril/Valsartan in real-life study of patients with HFrEF. A study conducted by Armentaro et al. [ 19], the Sacubitril/Valsartan therapy in HFrEF patients was observed to improve NYHA functional class with the improvement of renal function, reduction of NT-proBNP levels, and improvement in several hemodynamic, clinical, and echocardiographic parameters during 2-year follow-up of 60 patients [19]. In another study Armentaro et al. [ 20] reported a potential therapeutical role of Sacubitril/Valsartan therapy in HFrEF patients with metabolic co-morbid conditions. In addition to functional and echocardiographic improvements, a persistent metabolic improvement has been observed over the follow-up period of 12 months [20].
Even though, this is the first prospective clinical study in Pakistani population, single center experience, lack of control group, and limited sample size remained the main limitation of our study. The short duration of follow-up remained another important limitation considering the chronic nature of the diseases. Additionally, in real-life clinical practice HFrEF patients generally present with multiple co-morbid conditions, hence, exclusion of low eGFR, low BP, low hemoglobin, and CRT/ICD treatment may limit the generalizability of study findings. Further large scale multicenter randomized studies with longer follow-up duration are needed to elaborate the role of Sacubitril/Valsartan in HFrEF patients of Pakistani population.
## Conclusion
In conclusion, Sacubitril/Valsartan therapy in HFrEF patients is safe and moderately tolerated among the Pakistani population with stable hemodynamic parameters and significant improvement in left ventricular function and function class. Hence, it can be used as first line of treatment for the patients with HFrEF with a gradual up-titration regime over a 6 weeks period.
## References
1. Mosterd A, Hoes AW. **Clinical epidemiology of heart failure**. *Heart* (2007) **93** 1137-1146. DOI: 10.1136/hrt.2003.025270
2. Bonneux L, Barendregt JJ, Meeter K, Bonsel GJ, van der Maas PJ. **Estimating clinical morbidity due to ischemic heart disease and congestive heart failure: the future rise of heart failure**. *Am J Public Health* (1994) **84** 20-28. DOI: 10.2105/AJPH.84.1.20
3. Okura Y, Ramadan MM, Ohno Y, Mitsuma W, Tanaka K, Ito M, Suzuki K, Tanabe N, Kodama M, Aizawa Y. **Impending epidemic: future projection of heart failure in Japan to the year 2055**. *Circ J* (2008) **72** 489-491. DOI: 10.1253/circj.72.489
4. **Effects of enalapril on mortality in severe congestive heart failure. Results of the Cooperative North Scandinavian Enalapril Survival Study (CONSENSUS)**. *N Engl J Med.* (1987) **316** 1429-35. DOI: 10.1056/NEJM198706043162301
5. **Effect of enalapril on survival in patients with reduced left ventricular ejection fractions and congestive heart failure**. *N Engl J Med* (1991) **325** 293-302. DOI: 10.1056/NEJM199108013250501
6. Zannad F, McMurray JJ, Krum H, van Veldhuisen DJ, Swedberg K, Shi H, Vincent J, Pocock SJ, Pitt B. **Eplerenone in patients with systolic heart failure and mild symptoms**. *N Engl J Med* (2011) **364** 11-21. DOI: 10.1056/NEJMoa1009492
7. Ponikowski P, Voors AA, Anker SD, Bueno H, Cleland JGF, Coats AJS, Falk V, González-Juanatey JR, Harjola VP, Jankowska EA, Jessup M, Linde C, Nihoyannopoulos P, Parissis JT, Pieske B, Riley JP, Rosano GMC, Ruilope LM, Ruschitzka F, Rutten FH, Meer PVD. **2016 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure: The Task Force for the diagnosis and treatment of acute and chronic heart failure of the European Society of Cardiology (ESC). Developed with the special contribution of the Heart Failure Association (HFA) of the ESC**. *Eur J Heart Fail.* (2016) **18** 891-975. DOI: 10.1002/ejhf.592
8. Ponikowski P, Voors AA, Anker SD, Bueno H, Cleland JGF, Coats AJS, Falk V, González-Juanatey JR, Harjola VP, Jankowska EA, Jessup M, Linde C, Nihoyannopoulos P, Parissis JT, Pieske B, Riley JP, Rosano GMC, Ruilope LM, Ruschitzka F, Rutten FH, van der Meer P. **2016 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure: the Task Force for the diagnosis and treatment of acute and chronic heart failure of the European Society of Cardiology (ESC). Developed with the special contribution of the Heart Failure Association (HFA) of the ESC**. *Eur Heart J.* (2016) **37** 2129-200. DOI: 10.1093/eurheartj/ehw128
9. Yancy CW, Jessup M, Bozkurt B, Butler J, Casey DE, Colvin MM, Drazner MH, Filippatos G, Fonarow GC, Givertz MM, Hollenberg SM, Lindenfeld J, Masoudi FA, McBride PE, Peterson PN, Stevenson LW, Westlake C. **2016 ACC/AHA/HFSA focused update on new pharmacological therapy for heart failure: an update of the 2013 ACCF/AHA guideline for the management of heart failure: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines and the Heart Failure Society of America**. *J Am Coll Cardiol* (2016) **68** 1476-1488. DOI: 10.1016/j.jacc.2016.05.011
10. McMurray JJ, Packer M, Desai AS, Gong J, Lefkowitz MP, Rizkala AR, Rouleau JL, Shi VC, Solomon SD, Swedberg K, Zile MR. **Angiotensin–neprilysin inhibition versus enalapril in heart failure**. *N Engl J Med* (2014) **371** 993-1004. DOI: 10.1056/NEJMoa1409077
11. Senni M, McMurray JJ, Wachter R, McIntyre HF, Anand IS, Duino V, Sarkar A, Shi V, Charney A. **Impact of systolic blood pressure on the safety and tolerability of initiating and up-titrating Sacubitril/Valsartan in patients with heart failure and reduced ejection fraction: insights from the TITRATION study**. *Eur J Heart Fail* (2018) **20** 491-500. DOI: 10.1002/ejhf.1054
12. Tsutsui H, Momomura S, Saito Y, Ito H, Yamamoto K, Ohishi T, Okino N, Guo W. **Efficacy and safety of Sacubitril/Valsartan (LCZ696) in Japanese patients with chronic heart failure and reduced ejection fraction: Rationale for and design of the randomized, double-blind PARALLEL-HF study**. *J Cardiol* (2017) **70** 225-231. DOI: 10.1016/j.jjcc.2016.11.011
13. Kristensen SL, Martinez F, Jhund PS, Arango JL, Bĕlohlávek J, Boytsov S, Cabrera W, Gomez E, Hagège AA, Huang J, Kiatchoosakun S. **Geographic variations in the PARADIGM-HF heart failure trial**. *Eur Heart J* (2016) **37** 3167-3174. DOI: 10.1093/eurheartj/ehw226
14. Dewan P, Docherty KF, McMurray JJ. **Sacubitril/Valsartan in Asian patients with heart failure with reduced ejection fraction**. *Korean Circ J* (2019) **49** 469-484. DOI: 10.4070/kcj.2019.0136
15. Teng TK, Tromp J, Tay WT, Anand I, Ouwerkerk W, Chopra V, Wander GS, Yap JJ, MacDonald MR, Xu CF, Chia YM, Shimizu W, Richards AM, Voors A, Lam CS. **Prescribing patterns of evidence-based heart failure pharmacotherapy and outcomes in the ASIAN-HF registry: a cohort study**. *Lancet Glob Health* (2018) **6** e1008-e1018. DOI: 10.1016/S2214-109X(18)30306-1
16. Tsutsui H, Momomura SI, Saito Y, Ito H, Yamamoto K, Sakata Y, Desai AS, Ohishi T, Iimori T, Kitamura T, Guo W. **Efficacy and safety of Sacubitril/Valsartan in Japanese patients with chronic heart failure and reduced ejection fraction-results from the PARALLEL-HF study**. *Circ J* (2021) **85** 584-594. DOI: 10.1253/circj.CJ-20-0854
17. Senni M, McMurray JJ, Wachter R, McIntyre HF, Reyes A, Majercak I, Andreka P, Shehova-Yankova N, Anand I, Yilmaz MB, Gogia H. **Initiating Sacubitril/Valsartan (LCZ696) in heart failure: results of TITRATION, a double-blind, randomized comparison of two uptitration regimens**. *Eur J Heart Fail* (2016) **18** 1193-1202. DOI: 10.1002/ejhf.548
18. Hsu CY, Chang HY, Chao CJ, Chiou WR, Lin PL, Chung FP, Lin WY, Huang JL, Liang HW, Liao CT, Lee YH. **Utility of PREDICT-HF score in high-risk Asian heart failure patients receiving Sacubitril/Valsartan**. *Front Cardiovasc Med* (2022) **9** 950389. DOI: 10.3389/fcvm.2022.950389
19. Armentaro G, D’Arrigo G, Magurno M, Toscani AF, Condoleo V, Miceli S, Cassano V, Maio R, Arturi F, Tripepi G, Sesti G. **Impact of Sacubitril/Valsartan on clinical and echocardiographic parameters in heart failure patients with reduced ejection fraction: data from a real life 2-year follow-up study**. *Front Pharmacol* (2021) **12** 733475. DOI: 10.3389/fphar.2021.733475
20. Armentaro G, D’Arrigo G, Miceli S, Cassano V, Perticone M, Maio R, Marra AM, Arturi F, Cittadini A, Tripepi G, Sesti G. **Long term metabolic effects of Sacubitril/Valsartan in non-diabetic and diabetic patients with heart failure reduced ejection fraction: a real life study**. *Front Physiol* (2022) **13** 89710. DOI: 10.3389/fphys.2022.897109
|
---
title: 'Adherence to Mediterranean dietary quality index and risk of breast cancer
in adult women: a case-control study'
authors:
- Farhang Djafari
- Parivash Ghorbaninejad
- Fatemeh Dehghani Firouzabadi
- Fatemeh Sheikhhossein
- Hossein Shahinfar
- Maryam Safabakhsh
- Hossein Imani
- Sakineh Shab-Bidar
journal: BMC Women's Health
year: 2023
pmcid: PMC10012732
doi: 10.1186/s12905-023-02247-w
license: CC BY 4.0
---
# Adherence to Mediterranean dietary quality index and risk of breast cancer in adult women: a case-control study
## Abstract
### Background
Breast cancer (BC) is the fifth most prevalent cause of cancer-related deaths in Iran. Given that the role of whole-diet on cancer risk is important, this study aimed to assess the association of MedDQI and breast cancer risk.
### Methods
This hospital-based case-control study was performed on 150 women with pathologically confirmed breast cancer within the period of less than 3 months. Controls were 150 apparently healthy that were matched by age. Dietary data was collected using a validated questionnaire. To examine participants’ adherence to MedDQI, the MedDQI was created according to foods and nutrients highlighted or minimized in the MedDQI construction.
### Results
After adjusting for possible confounders, participants in the highest quartile of the MedDQI score had $55\%$ lower odds of breast cancer than women in the bottom quartile (OR: 0.45, $95\%$ CI: 0.21, 0.94, P trend: 0.02). Stratified analysis by menopausal status showed such association in postmenopausal women (OR: 0.24, $95\%$ CI: 0.07, 0.8, P trend: 0.055) after controlling for age and energy intake.
### Conclusion
The results showed an inverse association between adherence to the MedDQI and risk of breast cancer among Iranian women. More prospective studies are needed to confirm our results.
## Introduction
Breast cancer (BC), the second-high prevalent cancer, is spreading around the world, especially in developing countries. Based on GLOBOCAN statistics for 2018, about $11.6\%$ of women were recognized with breast cancer that year, with 626,679 associated deaths [1]. In Iran, BC is the fifth most prevalent cause of cancer-related deaths, including $24.4\%$ of all cancers with a standard age rate (ASR) of 23.1 per 100,000 [2]. Several risk factors counting family history of BC, reproductive factors, and environmental factors contribute to the development of BC [3, 4]. The management and prognosis of metastatic breast cancer involve several metabolic aspects, including metabolic syndrome and obesity, glucose metabolism, and microRNA modulation [5]. Among the environmental factors, diet as a modifiable factor can play an important role in the development of BC [6], however, studies in this term are limited that according to some recent studies, adherence to the Mediterranean diet has been related to reducing risk of diseases such as cancers, hypertension, cardiovascular diseases, obesity, and diabetes [7–11]. Particularly, numerous studies found a negative association between the consumption of Mediterranean diet and risk of breast cancer [12–14]. Further, in a case-control study, authors tested the association between non-adherence to Mediterranean diet with lifestyle habits in the incidence.
of breast cancer. They found a clear positive relationship and synergism between non-adherence to Mediterranean diet with current smoker, physical inactivity, and alcohol consumption [15]. Previous studies have mostly assessed the associations between Mediterranean dietary pattern and the risk of diseases that do not consider categorizing sources of fat, protein, or carbohydrate, separately [16]. In this regard, focusing on different sources of fat, protein, or carbohydrate, considering their effects on the quality of the diet, can be a better approach to show the influence of the diet on the risk of diseases.
The Mediterranean dietary quality index (MedDQI), a beneficial tool for assessing the quality of the diet, was developed by Gerber et al. [ 17]. This index evaluates diet quality with an emphasis on different sources of fat (saturated fatty acids and cholesterol versus olive oil) and two different sources of protein (meat and fish) with the contrary scores, both on the poor and good sides, respectively. Given the effective role of different food sources in the development or prevention of cancer [18], focusing on the MedDQI can give a better picture of the association of the diet and BC risk. To the best of our knowledge, no study has investigated the association of the MedDQI and the risk of breast cancer. We, therefore, aimed to examine the potential association between the MedDQI with the risk of breast cancer in Iranian women.
## Study design and participants
This hospital-based case-control study was conducted between September 23, 2017 and June 21, 2018 among Iranian women (46.6 ± 10.7 years old) who referred to cancer research center, Imam Khomeini Hospital in Tehran. To calculate the sample size calculation, the type I error of $5\%$ and the study power of $95\%$ was used. We hypothesized $5\%$ of the difference in mean and SD of dietary grains between cases and controls and reached almost 150 patients with breast cancer and 150 healthy controls [19]. Cases ($$n = 150$$) who were suggested to participate in our study by a pathologist, were pathologically diagnosed with BC. While, controls ($$n = 150$$) were apparently healthy women among relatives of patients in other wards of cited Hospital, like dermatology, urology, orthopedic, etc., by poster installation. There was no relationship between these two groups and they were matched just by age. We included patients with a diagnosis period of lower than 3 months to minimize the effect of awareness of BC on patient’s dietary reports. Furthermore, subjects who had any other cancers’ history and long-term dietary restrictions and also controls with BC history were excluded (Fig. 1). The skilled interviewer recorded information on age (year), energy intake (kcal/d), education (university graduated, n (%)), urban-residency (yes, n (%)), family history of breast cancer (yes, n (%)), physical activity (Met/min/week), marital status (married, n (%)), smoking (never smoked, n (%)), alcohol consumption (never used, n (%)), dietary supplement use (yes, n (%)), length of breast-feeding (year), menopausal status (yes, n (%)), history of hormone replacement therapy (yes, n (%)) and BMI (kg/m2), throughout a 45-min structured face‐to‐face interview by a standard questionnaire.
Fig. 1The flow diagram for participant selection
## Dietary intake assessment
Usual dietary intake of women was evaluated using a valid and reliable 147-item Food Frequency Questionnaire (FFQ) [20] which included a list of groceries and a standard size of each food item. The trained dieticians asked the participants to recall their consumption frequency of each item on a daily, weekly, monthly, and annual basis. When the participants’ reports were not adaptive with the given portion sizes, they were asked to consider their own portion sizes. To estimate energy and nutrient intakes, the household measures and the USDA food composition database which modified for Iranian foods [21, 22], were used to convert the consumed food portion sizes to grams. The calculations were also performed by a modified version of NUTRITIONIST IV software for Iranian foods (version 7.0; N-Squared Computing, Salem, OR, USA).
## Construction of Med-DQI
We calculated the diet score based on the Mediterranean diet quality index (MedDQI) (Table 1). This dietary index includes 7 food components which were given a score of 0, 1 or 2 according to the daily intake of each item. Finally, a total score was obtained by summing up the scores of these food items and ranged from 0 to 14. To obtain lower score on this index shows a higher nutrition quality and following the Mediterranean dietary pattern [17]. To estimate the MedDQI score, we categorized participants based on quartile groups of the above-mentioned component’s intakes to minimize misclassification.
Table 1Score formation of the Mediterranean Dietary Quality IndexScoring012Saturated fatty acids (% energy)< 1010–13> 13Cholesterol (milligram)< 300300–400> 400Meats (gram)< 2525–125> 125Olive oil (milliliter)> 1515–5< 5Fish (gram)> 6060–30< 30Cereals (gram)> 300300–100< 100Vegetables + fruits (gram)> 700700–400< 400
## Assessment of other variables
Weight was measured with light clothing and without shoes, by using a digital weighing scale (Seca725 GmbH & Co. Hamburg, Germany) to the nearest 100 g and the height was assessed while standing and keeping the shoulders and hips against the wall without shoes, using a stadiometer (Seca, Germany) with an accuracy of 0.1 cm. Body mass index (BMI) was calculated as weight divided by squared height and presented as kg/m2. A validated short from International Physical Activity Questionnaire [23] was used to assess subject’s physical activity levels. Recorded amounts were presented based on Metabolic Equivalents (METs)[23]. Then, the duration and frequency of physical activity days were multiplied by the MET value of the activity and sum of them was calculated as the total exercise minute per week.
## Statistical analyses
All individuals were categorized according to the quartiles of MedDQI score. We analysed the study participants’ characteristics and dietary intakes according to MedDQI score quartiles, using one-way analysis of variance (ANOVA) and χ2 tests for continuous and categorical variables, respectively. Data were shown as the mean ± SD for continuous variables and percent (%) for categorical ones. Odds ratio and $95\%$ confidence intervals were obtained using logistic regression to determine the relationship of adherence to the MedDQI score with risk of breast cancer. The risk was reported in crude and 3 adjusted models including confounders such as age (year), energy intake (kcal/d), education (university graduated, n (%)), urban-residency (yes, n (%)), family history of breast cancer (yes, n (%)), physical activity (Met/min/week), marital status (married, n (%)), smoking (never smoked, n (%)), alcohol consumption (never used, n (%)), dietary supplement use (yes, n (%)), length of breast-feeding (year), menopausal status (yes, n (%)), history of hormone replacement therapy (yes, n (%)) and BMI (kg/m2). In this analysis, the first quartile of exposure was considered as the reference category. All statistical analyses were done using the Statistical Package for the Social Sciences (SPSS version 22; SPSS Inc.). We considered $p \leq 0.05$ as the significance level.
## Results
The mean age of participants was 46.6 ± 10.7 year in both case and control groups. Moreover, the BMI of the participants in case and control groups were 28.1 ± 4.6 and 28.2 ± 5.2 kg/m2 respectively. General characteristics of the study subjects with and without BC are indicated in Table 2. Also, we showed this information for participants in this table. Women with BC were more likely to be older and they had a longer breastfeeding period than women without BC. Additionally, compared with the participants in the first quartile of the MedDQI, those in the top quartile had lower BMI. No significant differences were observed in other variables.
Table 2Characteristics of the study subject across patients with and without breast cancer and also across the quartile categorize of the Mediterranean Dietary Quality Index (MedDQI).Breast cancerQuartiles of Mediterranean Dietary Quality IndexYes ($$n = 150$$)No ($$n = 150$$)P*Q1 ($$n = 103$$)MedDQI score range: 3–5Q2 ($$n = 91$$)Q3 ($$n = 49$$)Q4 ($$n = 57$$)MedDQI score range:8–11P*Age (y)46.6 ± 10.746.6 ± 10.7 < 0.001 47.0 ± 10.547.1 ± 10.044.5 ± 10.546.6 ± 12.30.4BMI (kg/m2)28.1 ± 4.628.2 ± 5.20.828.70 ± 5.3228.73 ± 4.6627.84 ± 4.2526.65 ± 4.97 0.01 Physical activity (Met/min/week)475.8 ± 1043.2590.0 ± 843.80.2644.4 ± 970.2406.8 ± 708388.8 ± 561655.8 ± 1389.60.8Family history of breast cancer yes, (%)30 [20]39 [26]0.221 (20.4)20 [22]12 (24.5)16 [28]0.8University graduated n, (%)21 [14]28 (18.7)0.619 (18.5)13 (14.3)8 (16.3)9 (15.9)0.3Urban-resided n, (%)140 (93.3)139 (92.7)0.899 (96.1)85 (93.4)46 (93.9)49 [86]0.1Married, n (%)129 [86]136 (90.7)0.286 (83.5)83 (91.2)46 (93.9)50 (87.7)0.5Menopause status yes, n (%)62 (41.3)54 [36]0.341 (39.8)37 (40.7)15 (30.6)23 (40.4)0.6Age at first menarche (year)14.4 ± 1.713.4 ± 1.60.214.7 (1.7)13.6 (1.2)13.2 (1.5)13.6 (1.7)0.2Number of child, n2.6 ± 2.12.4 ± 2.10.32.5 (1.9)2.5 (2.4)2.3 (1.6)2.6 (2.3)0.9Length of breastfeeding (year)4.2 ± 3.53.5 ± 2.7 0.059 4.07 (3.3)3.8 (2.8)3.3 (3.1)4.2 (3.2)0.8History of HRT, n (%)12 [8]8 (5.3)0.36 (5.8)2 (2.2)6 (12.2)2 (10.5)0.07Smoking, never smoked, n (%)147 [98]144 [96]0.3101 (98.1)87 (95.6)47 (95.9)56 (98.2)0.7Alcohol, never used, n (%)149 [99]148 [98]0.5102 [99]90 (98.9)48 [98]57 [100]0.5Dietary supplement use, yes, n (%)80 (53.3)84 [56]0.0961 (59.2)49 (53.8)24 [49]30 (52.6)0.6Medication use §,yes, n (%)64 [42]62 [41]0.845 (43.7)37 (40.7)18 (36.7)26 (45.6)0.3Comorbidities †, n (%)38 [25]31 [20]0.346 (44.7)42 (46.2)26 (53.1)29 (50.9)0.3BC, Breast CancerBMI, body mass indexHRT, hormone replacement therapykg/m2, kilogram/meter2MET/min/wk, metabolic equivalent minute per week*ANOVA for continuous variables and Chi-square test for categorical variables§Lipid lowering and anti-hypertensive medications†Diabetes, Hypertension and Hyperlipidemia Dietary intakes of the patients across the case and control groups as well as across the quartiles of MedDQI are provided in Table 3. Compared to controls, women with BC consumed higher amounts of saturated fatty acids and total energy. In addition, participants in the highest quartile of MedDQI had higher intakes of saturated fatty acids, cholesterol, meat, and lower intakes of olive oils and total fruits and vegetables.
Table 3Dietary intakes of study participants across case and control groups as well as across quartile categories of the Mediterranean Dietary Quality IndexBreast cancerQuartiles of Mediterranean Dietary Quality IndexYes ($$n = 150$$)No ($$n = 150$$)P*Q1 ($$n = 103$$)MedDQI score range: 3–5Q2($$n = 91$$)Q3($$n = 49$$)Q4($$n = 57$$)MedDQI score range: 8–11P*Saturated fatty acids (% energy)9.1 ± 48.1 ± 3 0.02 6.9 ± 1.97.7 ± 2.38.9 ± 2.712.7 ± 4.8 < 0.001 Cholesterol (mg/d)255 ± 117229 ± 1260.06194 ± 63208 ± 74269 ± 112359 ± 179 < 0.001 Meats (gr/d)51.9 ± 41.946.6 ± 31.50.2637.3 ± 22.244.5 ± 30.457.3 ± 47.471.6 ± 64.2 < 0.001 Olive oil (ml/d)1.9 ± 3.21.8 ± 30.823 ± 3.91.6 ± 2.70.8 ± 1.31 ± 2.4 < 0.001 Fish (gr/d)8.9 ± 11.77.8 ± 90.399.4 ± 13.17.5 ± 8.46.6 ± 89.2 ± 9.70.63Cereals (gr/d)343 ± 288495 ± 23900.44633 ± 2875299 ± 314314 ± 225314 ± 2520.23Vegetables + fruits (gr/d)999 ± 413974 ± 3430.571,098,326958 ± 331929 ± 433881 ± 447 < 0.001 Energy intake(kcal/d)2914.1 ± 1159.02660.3 ± 799.6 0.02 3173.0 ± 1945.72888.0 ± 921.32980.4 ± 1040.92986.1 ± 995.70.06 Odds ratios and $95\%$CI of the BC across quartiles of the MedDQI are presented in Table 4. In the crude model, there was a significant inverse association between adherence to MedDQI and odds of BC (OR fourth vs. first quartile: 0.47, $95\%$ CI: 0.23, 0.92, P trend: 0.01). After controlling for confounders, participants in the top quartile of Med-DQI score had $55\%$ less likely to have BC compared with those in the bottom quartile (OR fourth vs. first quartile: 0.45, $95\%$ CI: 0.21, 0.94, P trend: 0.022). These findings remained significant even after adjustment for BMI (OR fourth vs. first quartile: 0.45, $95\%$ CI: 0.21, 0.94, P trend: 0.02).
Table 4Risk for breast cancer according to quartiles of the Mediterranean Dietary Quality Index with stratification by menopausal statusOR ($95\%$ CI)Q1MedDQI score range: 3–5Q2Q3Q4MedDQI score range: 8–11P for trendTotalNo. of cases/controls$\frac{52}{5137}$/$\frac{5422}{2739}$/18Crude11.48 (0.84–2.63)1.25 (0.63–2.47)0.47 (0.23–0.92) 0.012 Model 111.49 (0.84–2.65)1.25 (0.63–2.5)0.47 (0.24–0.93) 0.012 Model 211.45 (0.79–2.65)1.19 (0.58–2.45)0.45 (0.21–0.94) 0.022 Model 311.45 (0.79–2.65)1.45 (0.79–2.65)0.45 (0.21–0.94) 0.02 PremenopauseNo. of cases/controls$\frac{32}{3021}$/$\frac{3314}{2021}$/13Crude11.67 (0.8–3.51)1.52 (0.65–3.54)0.66 (0.28–1.54)0.15Model 111.69 (0.8–3.56)1.54 (0.66–3.62)0.67 (0.28–1.59)0.16Model 212.07 (0.94–4.58)1.57 (0.63–3.91)0.72 (0.28–1.8)0.11Model 312.06 (0.93–4.55)1.57 (0.63–3.93)0.73 (0.28–1.54)0.12PostmenopauseNo. of cases/controls$\frac{20}{2116}$/$\frac{218}{718}$/5Crude11.69 (0.51–3.56)0.83 (0.25–2.72)0.26 (0.08–0.84)0.07Model 111.33 (0.53–3.34)0.9 (0.27–2.99)0.24 (0.07–0.8) 0.055 Model 211.22 (0.44–3.39)1.05 (0.27–4.04)0.23 (0.05–0.91)0.1Model 311.23 (0.44–3.46)0.9 (0.25–3.88)0.18 (0.04–0.77)0.06Model 1: Adjusted for age and energy intakeModel 2: Further adjusted for education, residency, family history of breast cancer, physical activity, marital status, smoking, alcohol consumption, supplement use, length of breast-feeding, menopausal status and history of hormone replacement therapyModel 3: Further adjusted for BMI.
Stratified analysis by menopausal status expressed that after adjusting for age and energy intake, postmenopausal women with the highest adherence to the MedDQI had $76\%$ lower odds for having BC than those with the lowest adherence (OR fourth vs. first quartile: 0.24, $95\%$ CI: 0.07, 0.8, P trend: 0.055). There was no significant association between MedDQI and odds of BC in premenopausal women in crude or adjusted models.
(Model 1: Adjusted for age and energy intake; Model 2: Further adjusted for education, residency, family history of breast cancer, physical activity, marital status, smoking, alcohol consumption, supplement use, length of breastfeeding, menopausal status, and history of hormone replacement therapy; Model 3: Further adjusted for BMI)
## Discussion
In this hospital-based case-control study, we found a significant inverse association between MedDQI scores and odds of BC among Iranian women. This inverse association was also seen in postmenopausal women after controlling for energy intake and age. No significant relation was found between MedDQI scores and BC in premenopausal women. This is the first study looking at the link between the MedDQI scores and the risk of BC in Iranian women. BC is the most common malignancy in women [1] Statistics demonstrate a significant increase in the incidence of BC over the past 25 years worldwide [24]. Diet plays a considerable role in the primary prevention of BC. In the present study, adherence to a diet with low MedDQI scores was inversely associated with odds of BC. The effects of a non-Mediterranean diet in the incidence of breast cancer is also well established [25]. Consumption of fresh fruits and vegetables increase the consumer polyphenols that help in fighting against tumorigenesis. In the other hand increased intake of fiber and carbohydrate may be associated with the breast cancer prognosis. Soybean proteins consumption are involved in breast cancer risk reduction and lowering the chances of breast cancer reoccurrence. Ethanol consumption exert their carcinogenic effect on breast tissues and mediate breast cancer development. Processed meat releases the carcinogens compound likes heterocyclic amines, which mediate the onset of breast cancer. High saturated fat diet increases receptor-positive cancer, particularly ER + risk of breast cancer [25]. This relationship was found in postmenopausal women after taking potential cofounders into account. Studies looking at the link between healthy eating habits and the BC have produced conflicting results [26–29]. Similar to our research, several studies have found an inverse relationship between the risk of BC and good eating habits, or diets with low MedDQI scores [26]. Similar to our study, a recent meta-analysis revealed that following a Mediterranean-style diet may help lower the risk of breast cancer, albeit this link was not significant in premenopausal women [30]. Moreover, in an updated meta-analysis conducted by Morze et al. the highest adherence to MedDiet was inversely associated with cancer mortality and BC risk [31]. An increased adherence to the prudent or similar dietary patterns, those rich in fruit, vegetables, fish, whole grains, and low-fat dairy products, were strongly associated with a reduced risk of BC, according to a recent meta-analysis of 32 observational studies looking at the associations between various dietary patterns and odds of BC. These connections were similarly significant in premenopausal women, despite our study, though. Additionally, a Turati study revealed that following a Mediterranean diet was linked to a lower incidence of BC in both pre- and post-menopausal women [12]. The small sample size ($$n = 300$$) in this subgroup could be the reason why no correlation between MedDQI scores and risk of BC in premenopausal women was found. We did not look at the associations between BC subtypes. The relationship between following a diet with low Med-DQI scores and lowered risk of a specific subtype of BC has been discussed in several papers [12, 32, 33] *In a* prospective cohort study, a Mediterranean diet was associated with decreased risk of estrogen and progesterone receptor-negative (PR- ER-) BC [12]. According to another cohort study, following a Mediterranean diet reduced women’s likelihood of developing (ER-)-type BC [32]. A cohort study that involved 49,258 generally healthy women showed that adopting the Mediterranean diet did not significantly lower the risk of BC, in contrast to the current study [28]. Some research in this area has not discovered any conclusive links between healthy eating habits or other similar dietary patterns and the risk of BC [34–36]. The lack of several confounders being controlled for in some research may account for this disparity. Additionally, various dietary evaluation methods and components used in different research may contribute to the contradictory results. Based on the case-control nature of this study, the patients might have reported their current dietary intakes rather than their usual diet. This would result in a biased association between the MedDQI scores and risk of BC compared to those reported in prospective cohort studies, in which the exposure has been measured prior to disease incidence. Second, we ascertained the MedDQI scores based on dietary intakes of both cases and controls, while cases might had changed their dietary intakes after disease manifestation. This would lead to a higher adherence to MedDQI scores in cases than controls and would eventually result in a biased relationship. Finally, no information about subtypes of BC is collected in this study. It has been indicated that diet with low MedDQI scores may have an impact on some subtypes of BC and no effect on others. The underlying mechanisms for the possible favorable of the diet with high MedDQI scores and odds of BC are not completely known. The MedDQI scores is a valuable tool to predict dietary quality and has been already verified using nutritional biomarkers [17]. This index was founded on the recommendations by the National Research Council (NRC) and American Heart Association (AHA) about the diet and health [37]. The intake of $30\%$ or less of the daily total energy from fat, $10\%$ or less of the total energy derived from saturated fat, 30 mg/d or less from cholesterol, $55\%$ of energy from complex carbohydrates and 5 servings or more from fruits and vegetables have been taken into account by NRC and AHA. Fruit, vegetables and whole grains are good sources of antioxidants, dietary fiber and polyphenols, which can mediate the inverse association between the Dietary Approaches to Stop Hypertension (DASH) diet in relation to BC [38]. The consumption of olive oil, due to its monounsaturated fatty acids, have shown to be beneficial associated with some types of BC prevention and survival [39]. In spite of the fact that this study has demonstrated that dietary habits and lifestyle have an impact on the incidence of breast cancer, there is no doubt that hormones are the main cause of the disease. Efforts must also be made to encourage a healthy lifestyle, with special emphasis on the importance of eating a diet consisting primarily of fruits, vegetables, and whole grains with a low intake of red meat and saturated fats. To build the foundation for future advances in evidence-based public health efforts in this region, continued and expanded research on diet, lifestyle and breast cancer risk is urgently necessary.
Suitable sample size, careful assessment of confounding elements and their controlling in the analyses and being the first investigation in Middle East women could be considered as the strengths of the present study. However, some limitations should be considered. First, the case control design of the study does not permit us inferring causality. In addition, these kinds of studies are subject to selection and recall bias. In case-control studies, cases may report their past diet correctly due to their cancer diagnosis. This can attenuate this association. Another concern for case-control studies is that cases may have changed their diet before diagnosis due to early symptoms of the disease. To reduce this error, we recruited newly-diagnosed cases in the study. Furthermore, we used FFQ for assessing dietary intakes which can result in misclassification in our study participants. In addition, an energy adjusted MedDQI scores was applied which could reduce the possibility of subject misclassification. However, we used a qualified questionnaire to evaluate dietary intakes. Finally, we did not collect information about estrogen or progesterone receptor status of study patients.
## Conclusion
In conclusion, our findings showed an inverse association between diet with high MedDQI scores and odds of BC among Iranian women. More cohort studies are needed to approve our findings.
## References
1. Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal A. **Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries**. *Cancer J Clin* (2018.0) **68** 394-424. DOI: 10.3322/caac.21492
2. Farhood B, Geraily G, Alizadeh A. **Incidence and mortality of various cancers in Iran and compare to other countries: a review article**. *Iran J public health* (2018.0) **47** 309-16. PMID: 29845017
3. Pharoah PD, Day NE, Duffy S, Easton DF, Ponder BA. **Family history and the risk of breast cancer: a systematic review and meta-analysis**. *Int J Cancer* (1997.0) **71** 800-9. DOI: 10.1002/(SICI)1097-0215(19970529)71:5<800::AID-IJC18>3.0.CO;2-B
4. Nelson HD, Zakher B, Cantor A, Fu R, Griffin J, O’Meara ES. **Risk factors for breast cancer for women aged 40 to 49 years: a systematic review and meta-analysis**. *Ann Intern Med* (2012.0) **156** 635-48. DOI: 10.7326/0003-4819-156-9-201205010-00006
5. Amabile MI, Frusone F, De Luca A, Tripodi D, Imbimbo G, Lai S. **Locoregional surgery in metastatic breast cancer: do concomitant metabolic aspects have a role on the management and prognosis in this setting?**. *J Personalized Med* (2020.0) **10** 227. DOI: 10.3390/jpm10040227
6. Miller AB, Kelly A, Choi NW, Matthews V, Morgan RW, Munan L. **A study of diet and breast cancer**. *Am J Epidemiol* (1978.0) **107** 499-509. DOI: 10.1093/oxfordjournals.aje.a112569
7. Giacosa A, Barale R, Bavaresco L, Gatenby P, Gerbi V, Janssens J. **Cancer prevention in Europe: the Mediterranean diet as a protective choice**. *Eur J cancer prevention: official J Eur Cancer Prev Organisation (ECP)* (2013.0) **22** 90-5. DOI: 10.1097/CEJ.0b013e328354d2d7
8. 8.Magriplis E, Panagiotakos D. Presence of Hypertension Is Reduced by Mediterranean Diet Adherence in All Individuals with a More Pronounced Effect in the Obese: The Hellenic National Nutrition and Health Survey (HNNHS). 2020;12(3).
9. 9.Becerra-Tomás N, Blanco Mejía S, Viguiliouk E, Khan T, Kendall CWC, Kahleova H et al. Mediterranean diet, cardiovascular disease and mortality in diabetes: A systematic review and meta-analysis of prospective cohort studies and randomized clinical trials. 2020;60(7):1207–27.
10. 10.Aprano S, Framondi L, Di Matteo R, Laudisio D, Savastano S.
11. Schröder H. **Protective mechanisms of the Mediterranean diet in obesity and type 2 diabetes**. *J Nutr Biochem* (2007.0) **18** 149-60. DOI: 10.1016/j.jnutbio.2006.05.006
12. Buckland G, Travier N, Cottet V, González CA, Luján-Barroso L, Agudo A. **Adherence to the mediterranean diet and risk of breast cancer in the european prospective investigation into cancer and nutrition cohort study**. *Int J Cancer* (2013.0) **132** 2918-27. DOI: 10.1002/ijc.27958
13. Trichopoulou A, Bamia C, Lagiou P, Trichopoulos D. **Conformity to traditional Mediterranean diet and breast cancer risk in the greek EPIC (european prospective investigation into Cancer and Nutrition) cohort**. *Am J Clin Nutr* (2010.0) **92** 620-5. DOI: 10.3945/ajcn.2010.29619
14. van den Brandt PA, Schulpen M. **Mediterranean diet adherence and risk of postmenopausal breast cancer: results of a cohort study and meta-analysis**. *Int J Cancer* (2017.0) **140** 2220-31. DOI: 10.1002/ijc.30654
15. La Torre G, De Carlo I, Sestili C, Cocchiara R, Lia L, Di Bella O. **Non-adherence to Mediterranean diet and synergy with lifestyle habits in the occurrence of breast cancer: a case-control study in Italy**. *Eur Rev Med Pharmacol Sci* (2021.0) **25** 4535-9. PMID: 34286496
16. D’Alessandro A, De Pergola G. **Mediterranean Diet and Cardiovascular Disease: a critical evaluation of a Priori Dietary indexes**. *Nutrients* (2015.0) **7** 7863-88. DOI: 10.3390/nu7095367
17. Gerber M. **The comprehensive approach to diet: a critical review**. *J Nutr* (2001.0) **131** 3051s-5s. DOI: 10.1093/jn/131.11.3051S
18. Trichopoulou A, Katsouyanni K, Stuver S, Tzala L, Gnardellis C, Rimm E. **Consumption of olive oil and specific food groups in relation to breast cancer risk in Greece**. *J Natl Cancer Inst* (1995.0) **87** 110-6. DOI: 10.1093/jnci/87.2.110
19. 19.Ataollahi M, Sedighi S, Masoumi SZ. Nutritional and unhealthy behaviors in women with and without breast cancer.Iranian Red Crescent Medical Journal. 2014;16(9).
20. Esmaillzadeh A, Mirmiran P, Azizi F. **Whole-grain intake and the prevalence of hypertriglyceridemic waist phenotype in tehranian adults**. *Am J Clin Nutr* (2005.0) **81** 55-63. DOI: 10.1093/ajcn/81.1.55
21. 21.Azar M, Sarkisian E. Food composition table of Iran. Tehran: National Nutrition and Food Research Institute, Shaheed Beheshti University. 1980;65.
22. David H, Linda L, Pamela P. *USDA National Nutrient Database for Standard Reference, Release 24* (2016.0)
23. 23.forms. CIGfdpaaotIPAQI-sal.
24. Azamjah N, Soltan-Zadeh Y, Zayeri F. **Global Trend of breast Cancer mortality rate: a 25-Year study**. *Asian Pac J cancer prevention: APJCP* (2019.0) **20** 2015-20. DOI: 10.31557/APJCP.2019.20.7.2015
25. 25.Buja A, Pierbon M, Lago L, Grotto G, Baldo V.Breast Cancer Primary Prevention and Diet: An Umbrella Review. 2020;17(13).
26. Xiao Y, Xia J, Li L, Ke Y, Cheng J, Xie Y. **Associations between dietary patterns and the risk of breast cancer: a systematic review and meta-analysis of observational studies**. *Breast Cancer Res* (2019.0) **21** 16. DOI: 10.1186/s13058-019-1096-1
27. 27.Jacobs I, Taljaard-Krugell C, Wicks M, Cubasch H, Joffe M, Laubscher R et al. Dietary Patterns and Breast Cancer Risk in Black Urban South African Women: The SABC Study.Nutrients. 2021;13(11).
28. Couto E, Sandin S, Löf M, Ursin G, Adami HO, Weiderpass E. **Mediterranean dietary pattern and risk of breast cancer**. *PLoS ONE* (2013.0) **8** e55374. DOI: 10.1371/journal.pone.0055374
29. Heidari Z, Jalali S, Sedaghat F, Ehteshami M, Rashidkhani B. **Dietary patterns and breast cancer risk among iranian women: a case-control study**. *Eur J Obstet Gynecol Reprod Biol* (2018.0) **230** 73-8. DOI: 10.1016/j.ejogrb.2018.09.018
30. Li Y, Hu B-Q, Wu X-J, Qi X-W, Jiang J, Cui X. **Adherence to Mediterranean diet and the risk of breast cancer: a meta-analysis**. *Transl Cancer Res* (2018.0) **7** 1290-7. DOI: 10.21037/tcr.2018.10.13
31. Morze J, Danielewicz A, Przybyłowicz K, Zeng H, Hoffmann G, Schwingshackl L. **An updated systematic review and meta-analysis on adherence to mediterranean diet and risk of cancer**. *Eur J Nutr* (2021.0) **60** 1561-86. DOI: 10.1007/s00394-020-02346-6
32. Hirko KA, Willett WC, Hankinson SE, Rosner BA, Beck AH, Tamimi RM. **Healthy dietary patterns and risk of breast cancer by molecular subtype**. *Breast Cancer Res Treat* (2016.0) **155** 579-88. DOI: 10.1007/s10549-016-3706-2
33. Soltani S, Benisi-Kohansal S, Azadbakht L, Esmaillzadeh A. **Association between Adherence to “Dietary approaches to Stop Hypertension” eating plan and breast Cancer**. *Nutr Cancer* (2021.0) **73** 433-41. DOI: 10.1080/01635581.2020.1756354
34. Cui X, Dai Q, Tseng M, Shu XO, Gao YT, Zheng W. **Dietary patterns and breast cancer risk in the shanghai breast cancer study. Cancer epidemiology, biomarkers & prevention: a publication of the American Association for Cancer Research**. *cosponsored by the American Society of Preventive Oncology* (2007.0) **16** 1443-8
35. Männistö S, Dixon LB, Balder HF, Virtanen MJ, Krogh V, Khani BR. **Dietary patterns and breast cancer risk: results from three cohort studies in the DIETSCAN project**. *Cancer causes & control: CCC* (2005.0) **16** 725-33. DOI: 10.1007/s10552-005-1763-7
36. Adebamowo CA, Hu FB, Cho E, Spiegelman D, Holmes MD, Willett WC. **Dietary patterns and the risk of breast cancer**. *Ann Epidemiol* (2005.0) **15** 789-95. DOI: 10.1016/j.annepidem.2005.01.008
37. Gerber MJ, Scali JD, Michaud A, Durand MD, Astre CM, Dallongeville J. **Profiles of a healthful diet and its relationship to biomarkers in a population sample from Mediterranean southern France**. *J Am Diet Assoc* (2000.0) **100** 1164-71. DOI: 10.1016/S0002-8223(00)00340-0
38. Steinmetz KA, Potter JD. **Vegetables, fruit, and cancer prevention: a review**. *J Am Diet Assoc* (1996.0) **96** 1027-39. DOI: 10.1016/S0002-8223(96)00273-8
39. Buckland G, Travier N, Agudo A, Fonseca-Nunes A, Navarro C, Lagiou P. **Olive oil intake and breast cancer risk in the Mediterranean countries of the european prospective investigation into Cancer and Nutrition study**. *Int J Cancer* (2012.0) **131** 2465-9. DOI: 10.1002/ijc.27516
|
---
title: Evaluating the implementation of a multi-level mHealth study to improve hydroxyurea
utilization in sickle cell disease
authors:
- J. S Hankins
- M. B Potter
- M. E Fernandez
- C Melvin
- L DiMartino
- S. R Jacobs
- H. B Bosworth
- A. A King
- J Simon
- J. A Glassberg
- A Kutlar
- V. R Gordeuk
- N Shah
- A. A Baumann
- L. M Klesges
journal: Frontiers in Health Services
year: 2023
pmcid: PMC10012741
doi: 10.3389/frhs.2022.1024541
license: CC BY 4.0
---
# Evaluating the implementation of a multi-level mHealth study to improve hydroxyurea utilization in sickle cell disease
## Abstract
### Background
Sickle Cell Disease (SCD) is a progressive genetic disease that causes organ damage and reduces longevity. Hydroxyurea is an underutilized evidence-based medication that reduces complications and improves survival in SCD. In a multi-site clinical trial, part of the NIH-funded Sickle Cell Disease Implementation Consortium (SCDIC), we evaluate the implementation of a multi-level and multi-component mobile health (mHealth) patient and provider intervention to target the determinants and context of low hydroxyurea use. Given the complexity of the intervention and contextual variability in its implementation, we combined different behavioral and implementation theories, models, and frameworks to facilitate the evaluation of the intervention implementation. In this report, we describe engagement with stakeholders, planning of the implementation process, and final analytical plan to evaluate the implementation outcomes.
### Methods
During 19 meetings, a 16-member multidisciplinary SCDIC implementation team created, conceived, and implemented a project that utilized Intervention Mapping to guide designing an intervention and its evaluation plan. The process included five steps: [1] needs assessment of low hydroxyurea utilization, [2] conceptual framework development, [3] intervention design process, [4] selection of models and frameworks, and [5] designing evaluation of the intervention implementation.
### Results
Behavioral theories guided the needs assessment and the design of the multi-level mHealth intervention. In designing the evaluation approach, we combined two implementation frameworks to best account for the contextual complexity at the organizational, provider, and patient levels: [1] the Consolidated Framework for Implementation Research (CFIR) that details barriers and facilitators to implementing the mHealth intervention at multiple levels (users, organization, intervention characteristics, broader community), and [2] the Technology Acceptance Model (TAM), a conceptual model specific for explaining the intent to use new information technology (including mHealth). The Reach Effectiveness Adoption Implementation and Maintenance (RE-AIM) framework was used to measure the outcomes.
### Discussion
Our research project can serve as a case study of a potential approach to combining different models/frameworks to help organize and plan the evaluation of interventions to increase medication adherence. The description of our process may serve as a blueprint for future studies developing and testing new strategies to foster evidence-based treatments for individuals living with SCD.
## Introduction
Sickle cell disease (SCD) is a devastating genetic disease where progressive organ damage leads to premature death [1]. Only a few evidence-based treatments exist for SCD, such as the disease-modifying agent hydroxyurea, and all are vastly underutilized. Uptake of disease-modifying therapies by patients with SCD is impacted by socio-economic barriers that create health inequities for this population (2–4). To accelerate the equitable translation of evidence-based treatments into clinical care for individuals with SCD, the Sickle Cell Disease Implementation Consortium (SCDIC) was created in 2016 [5]. The SCDIC is a cooperative research program funded by the NHLBI and composed of eight academic and clinical sites and one data coordinating center. SCDIC members include clinicians, epidemiologists, implementation scientists, government representatives, behavioral scientists, and patient partners. A primary goal of the SCDIC is to develop and test the effectiveness of interventions aimed at increasing the uptake of evidence-based therapies for SCD while seeking input from key stakeholders during the entire implementation strategy development process [5].
As the first FDA-approved evidence-based drug for SCD, daily oral hydroxyurea is a medication recommended by guidelines [6] and with robust evidence for reducing acute and chronic disease complications, lowering acute-care utilization, and improving survival among individuals with SCD (7–11). In the U.S., <$50\%$ of patients with SCD appropriately utilize hydroxyurea (12–15), severely limiting its population impact. SCDIC consortium members thus chose to focus on strategies to increase hydroxyurea uptake. The consortia members and stakeholder partners worked together to develop a multi-level/multi-component mobile health (mHealth) intervention to increase provider prescribing practices and patient medication adherence to hydroxyurea. The implementation of this novel mHealth intervention is currently being tested in a multicenter study [16, 17].
During the development of the multi-level/multi-component mHealth intervention, a major challenge within the SCDIC was to construct an evaluation approach to identify how each contextual level (i.e., at the patient, provider, and clinical setting) and intervention component (i.e., the different mHealth features) could influence implementation and effectiveness outcomes, both as individual factors and interactively. For instance, patients, providers, and institutional characteristics can influence the implementation and distal effectiveness outcomes. Still, implementation outcomes can also interact across multiple levels (e.g., provider adoption might impact patient reach), modifying the effectiveness outcomes. To address this challenge, we used the intervention mapping methodology [18], which allows for strong collaboration among researchers and patient partners to develop the multi-level/multi-component mHealth intervention and to guide the design of the evaluation plan.
In this paper, we will report the process development of an intervention evaluation plan that considers what and how contextual factors influence implementation outcomes while anticipating possible interrelationships (e.g., synergy) among the intervention components. This paper will discuss different theories, models, and frameworks and how they are optimally combined to evaluate the implementation of a multi-level/multi-component mHealth intervention to improve hydroxyurea utilization in SCD (clinicaltrials.gov NCT03344900) [17]. We describe the rationale, process, and application of the implementation process and the engagement with stakeholders to develop the analytical plan of the intervention and define implementation outcomes. The description of our design process may serve as a blueprint for future studies developing and testing new strategies to foster the use of evidence-based treatments for individuals living with SCD.
## Methods
This descriptive narrative of the process selects and blends theories, models, and frameworks to design the clinical trial evaluation that tests a new multi-level/multi-component mHealth intervention to improve hydroxyurea adherence among patients with SCD [17].
## Settings and population
This multi-center study included seven SCDIC clinical sites, all academic institutions in urban, suburban, and rural areas, with some degree of organizational and population variability. All sites have experienced SCD providers (adult and pediatric hematologists and advanced care practitioners, all trained in SCD management) and trainees (e.g., residents and fellows). The prevalence of eligible patient participants (i.e., patients with SCD treated with hydroxyurea who were not receiving monthly blood transfusions and not using mobile apps to improve adherence) ranged from 40 to $60\%$ of the patient population at each site. Among eligible participants, approximately $70\%$ were covered under government health plans (primarily Medicaid). Most patients were aged >25 years, although, in two sites, about half of eligible patients were adolescents. Among eligible patients, approximately $70\%$ had a severe SCD genotype (HbSS and HbSβ0-thalassemia).
## Research team
Our SCDIC team comprises 16 members, including implementation scientists, hematologists, health science researchers, behavioral scientists, research coordinators, biostatisticians, patient partners, clinicians, and epidemiologists who collaborated on the study throughout 19 meetings from December 2017 to October 2019. In addition to the research team, the SCDIC steering committee (including principal site investigators, NHLBI representatives, patient partners, data coordinating center staff, and study coordinators) and the SCDIC Implementation Research Committee (composed of implementation scientists) offered an additional layer of scientific input, suggested changes to the study, and voted to approve the study in its final form.
## Summary of activities
Our activities began by providing level-setting knowledge about SCD and dissemination and implementation research to all members of the SCDIC. As the research team was diverse and background knowledge was often non-overlapping, much of the initial work centered on the generation of a shared vision, i.e., establishing agreement on the goals of the program and the process whereby a research question was developed and addressed. The design of the intervention components followed a comprehensive needs assessment phase examining the barriers and enablers to hydroxyurea utilization in the SCD population at multiple levels (Activity 1). Next, a conceptual model evolved using Intervention Mapping (Activity 2) to guide the development of the interventions and the planning of the implementation strategies (Activity 3), the selection of the models and frameworks (Activity 4), and the evaluation plan (Activity 5). We used an inclusive consensus approach throughout all activity phases and defined and selected priorities, designed interventions, and measured outcomes. Our study is fully accrued and is on track to complete data analysis by March 2023. Next, we will present the details of each activity.
## Activity 1: Needs assessment of hydroxyurea utilization
Our needs assessment focused on care redesign to improve hydroxyurea uptake and considered intervention targets, modalities, and strategy mechanisms. Barriers and enablers to hydroxyurea utilization in the SCD population were conducted through literature synthesis (19–22), population-level claims analysis [13, 23], patient and provider surveys [24, 25], semi-structured interviews, and focus groups [16, 26]. To synthesize and organize the needs assessment findings at the patient level, the Health Belief Model (HBM) [27] was used, and the Social Cognitive Theory (SCT) was used as the behavioral change model [28]. The HBM has been broadly used to evaluate the acceptance of health services among people with SCD, including attendance to clinic visits [29], sepsis [30], and stroke screening [31]. The constructs within the HBM align well with the disease characteristics and demographics (e.g., intermittent acute exacerbations align with perceived susceptibility), therefore, was selected for this project. In alignment with our behavioral models, our needs assessment findings identified the main determinants of poor hydroxyurea adherence among patients with SCD: perceived high susceptibility, high disease severity, low motivation to take medications, memory deficit (leading to poor habituation), low understanding of hydroxyurea benefit (i.e., medication knowledge), and low self-efficacy with taking hydroxyurea [16, 26]. Among providers of patients with SCD, the main determinants of low rates of prescribing hydroxyurea are: limited knowledge of the drug, lack knowledge of the national care guidelines, and low self-efficacy in dosing hydroxyurea [24, 32]. Following this formative evaluation, our team identified two priority target levels for intervention: 1) patient hydroxyurea adherence and 2) provider prescribing of hydroxyurea.
## Activity 2: Development of the conceptual model
Informed by the needs assessment, we developed a comprehensive conceptual model centered on Intervention Mapping. To organize the components of the intervention at both the patient and the provider levels, we used Intervention Mapping to identify specific methods and practical applications to create change in determinants, performance objectives, and behavioral change outcomes. Intervention *Mapping is* a protocol based on using theory and evidence for developing effective behavior change interventions [18]. A primary aim of the consortia was to develop a research protocol that included a consensus framework. Drawing on the team's expertise and experience in using different implementation frameworks, we selected frameworks based on overarching aims [33]. For instance, to understand what influences implementation, we selected a determinant framework, and to evaluate the implementation we chose an evaluation framework. We thus combined different models and frameworks to 1) identify determinants of poor hydroxyurea utilization at multiple levels, 2) guide the design of the interventions and strategies to be used at different intervention levels, and 3) plan the implementation evaluation. This comprehensive process allowed us to identify the salient targets to increase the use of the evidence-based treatment for SCD, hydroxyurea, while identifying and prioritizing important influencers of the implementation at the contextual level.
## Activity 3: Intervention design process
SCDIC investigators noted the widespread use of technology among patients with SCD (>$90\%$ own smartphones, $91\%$ use them regularly for communication, and $87\%$ rate the highest possible comfort levels [34, 35] and the desire of providers to receive SCD guideline information in a mobile platform [36]. The research team arrived at the consensus that to best “package” all necessary functions of the intervention and deliver it to the two targets (patients and providers), mHealth was the ideal conduit. mHealth intervention refers to the use of mobile technology for medical and public health practices [37] and here it serves as the channel for effecting behavioral modifications.
Using user-centered design principles, we co-created (with patients and providers) a two-level intervention: [1] InCharge Health app, which incorporated features that acted on the modifiable determinants of patients' poor hydroxyurea adherence (e.g., low cue to action, low self-efficacy in taking medication) (Figure 1), and [2] HU Toolbox app for providers, which incorporated features that acted on modifiable determinants of clinician failure to prescribe hydroxyurea (e.g., low disease knowledge, low self-efficacy in prescribing hydroxyurea) (Figure 1). User-centered design is an approach to designing and developing products that grounds its process on the information about the people who will ultimately use the products to improve usability and user experience [38, 39]. Patients used InCharge Health within their own lived environment, while providers used HU Toolbox in the clinic/office. Because in clinical practice, providers prescribe and counsel patients on the benefits of hydroxyurea during regular visits, providers were both the actors and targets of our multi-level intervention (Figure 1) [40].
**Figure 1:** *Multi-level/multi-component intervention mapping and implementation outcomes. The intervention components (mHealth activities) address the determinants of hydroxyurea utilization at the patient and provider levels. Performance outcomes depict the actions taken by the targets of the intervention. Provider-appropriate prescribing influences patients’ hydroxyurea adherence, promoting increased hydroxyurea utilization, reduced organ damage, and improved quality of life.*
A menu of implementation strategies was used to increase the implementation outcomes of both InCharge Health and HU Toolbox. The “train and educate stakeholders strategy” [41, 42] included ongoing consultation and training of patients and providers on how to use both interventions, both as one-on-one activities (patient or provider individual meetings while in clinic) and group activities (provider educational meetings during staff and faculty meetings). “ Support of the clinicians”” strategy [41, 42] was utilized during the study as reminders to clinicians to use the HU Toolbox and the facilitation of relay of clinical data as a function of the HU Toolbox app (i.e., guidance on correct prescribing of hydroxyurea app function). Other implementation strategies included understanding barriers and facilitators to digital health interventions, a thorough understanding of the implementation context, and patient feedback. These strategies were prospectively tracked and documented by each participating site.
## Activity 4: Selection of implementation models and frameworks
Our challenge in organizing patient, provider, and clinic-level contextual factors within SCDIC included the selection of implementation frameworks that would appropriately represent the multiple dimensions of the intervention (two target levels and several mHealth features) while accounting for the complex context where the population received care and where we sought to enhance the impact of our intervention. For instance, unmet social needs can lead to health disparities [43] and may significantly influence how patients with SCD engage with the intervention. Recruitment efforts that lead to equitable reach across the patient population and equity in the adoption and delivery of the intervention among providers are of paramount importance in the SCD population, given that the majority belong to underserved groups in the United States. Additionally, SCD characteristics (e.g., the disease severity) and co-morbidities might also affect response to the intervention (i.e., response heterogeneity) and lead to a lack of robust effects. Given the variability in patient characteristics across the participating sites, this was of particular concern.
SCDIC investigators considered several models and frameworks to define theory-based domains associated with contextual variables and the overall robustness of the mHealth innovation. Our goal was to examine the influences of patient, provider, and clinical setting level characteristics on study outcomes and the qualitative examination of barriers and enablers of the mHealth innovation. Given that different models and frameworks identify other metrics for evaluating implementation success, the group decided to combine the Consolidated Framework for Implementation Research (CFIR) [44] and the Reach Effectiveness Adoption Implementation and Maintenance (RE-AIM) [45] frameworks. CFIR is a determinant framework with five domains (inner setting, outer setting, intervention, process, and individuals involved) that systematically assesses potential barriers and facilitators (the determinants) of an implementation. RE-AIM is a versatile planning and evaluation framework, in which dimensions (reach effectiveness, maintenance, adoption, and implementation) systematically capture the outcomes of the implementation while assessing the equitability of the implementation [46]. Because we planned qualitative data collection as part of our formative evaluation of the implementation process, we used RE-AIM Qualitative Evaluation for Systematic Translation (RE-AIM QuEST) to develop the interview questions [47].
In addition to CFIR and RE-AIM/RE-AIM QuEST, we also used the Technology Acceptance Model (TAM) [48, 49] to design and measure the outcomes specifically associated with mHealth use. Users’ acceptance of new technology, including new mHealth innovations, impacts its successful adoption. TAM is a conceptual model that explains the intent to use new information technology (including mHealth) or information science among users, including medical providers. TAM has five constructs: perceived usefulness, perceived ease of use, compatibility, mobile health care systems self-efficacy, and technical support and training. However, perceived ease of use and perceived usefulness are the two dominant determinants of technology use. mHealth care systems self-efficacy is the health care professional's perception of their ability to use health care systems to accomplish the health care task and must be accounted for when new technology is implemented. The combination of implementation frameworks RE-AIM/RE-AIM QuEST, CFIR, and TAM informed the planning of the study while complementing each other in evaluating the complexity of the influential multi-level factors on implementation (Figure 2). TAM was used to measure the specific constructs related to the intervention/m-Health tool. CFIR was used in specific quantitative measures of intervention characteristics, characteristics of individuals, inner settings, and processes. The five RE-AIM domains (Reach, Effectiveness, Adoption, Implementation, Maintenance) were used to evaluate implementation outcomes with quantitative and qualitative assessments.
**Figure 2:** *Interrelation of theories, models, and frameworks. The Health Belief Model (HBM) and Social Cognitive Theory (SCT) informed the design of the InCharge Health app. Technology Acceptance Model (TAM) informed the creation of the HU Toolbox. The Consolidated Framework for Implementation Research (CFIR) constructs will inform the evaluation of the multiple-level influential factors, including the inner setting (organization), the users (patients and providers), intervention characteristics (InCharge Health and HU Toolbox apps), and app implementation process. The Reach Effectiveness Adoption implementation and Maintenance (RE-AIM) framework domains will be used to evaluate the implementation process. Dashed arrows represent the possible influential effects across CFIR constructs and RE-AIM domains.*
## Activity 5: Evaluation plan
To visualize the potential interrelatedness of the multiple influential factors of the context, we created a matrix that grouped CFIR and TAM domains and mapped them to all 5 RE-AIM domains for both the patient-level (Supplementary Table S1) and provider-level (Supplementary Table S2) strategies. We designed specific plans to examine how influential factors of app utilization potentially moderate the relationship between [1] how the level of InCharge Health implementation might correlate with hydroxyurea adherence and [2] how the level of HU Toolbox implementation might correlate to increases in providers' knowledge and self-efficacy in prescribing hydroxyurea. To assess outcomes within each RE-AIM domain, quantitative measures were selected with possible moderation or mediation by CFIR constructs (Supplementary Tables S1, S2). In mixed methods evaluation, qualitative data complements and expands, within key themes, the “why” and “how” barriers and facilitators to recruitment, implementation, and sustainability affect implementation. The description of all measures utilized in this study and their frequency have been previously described [17]. In brief, quantitative surveys are given at baseline and study exit, while qualitative data (semi-structured interviews) are conducted at the end of study participants at each site.
## Discussion
Dissemination and Implementation research has a growing number of models and frameworks, and parsimoniously using them is essential, but this is not always possible when there is high complexity in contextual factors and intervention components. While there might exist a need to combine aspects of various theories, models, and frameworks in designing interventions and complex evaluations, the methods for approaching this blending process are not widely available. Our study serves as a case study demonstrating the process of co-developing a multi-level/multi-component intervention and designing its evaluation. In our study, we combined HBM, SCT, RE-AIM, CFIR, and TAM while incorporating the broad perspectives of the diverse team members to plan and operationalize the frameworks, measurements, and evaluation. This manuscript outlines a rationale, process, and application that may be useful as an example for others to consider in designing an implementation evaluation, particularly around electronic health interventions to improve medication adherence in chronic diseases.
When clinical guidelines are implemented, theories, models, and frameworks are not always used to guide intervention and strategy development or their evaluation [50, 51]. When theories are not used to plan, undertake, or evaluate implementation, the correct diagnosis of the underlying reasons for the success or failure of implementation (i.e., the determinants or the barriers and facilitators of low guideline adoption) may not be identified, therefore reducing the likelihood of effective interventions. Our study exemplifies how theories, models, and frameworks can be used to guide the entire process of planning, execution, and evaluation of guideline implementation, in our case, the utilization of hydroxyurea in SCD.
The process evaluation of a multi-level implementation study is complex and includes considerations of potential interactions between the intervention elements and the targeted levels of the intervention. Additionally, in chronic diseases, where there is substantial variability among the clinical characteristics of the patient, patient and organizational care settings, stakeholder perspectives, and the health care provider's expertise level, attention to contextual factors is particularly relevant when interpreting the effects of the intervention on outcomes. The careful development of an evaluation process that accounts for the different intervention and contextual components is thus paramount but potentially very complex. The description of the development of the process evaluation of a multi-faceted intervention can advance our understanding of their effects by illustrating the evaluation process design.
## Lessons learned
Over the course of 19 meetings, a 16-member multidisciplinary implementation team within the SCDIC conceived and implemented a project that utilized Intervention Mapping to guide the process, which leveraged and combined different models and frameworks to plan the evaluation of the implementation of a multi-level intervention to increase adherence to the evidence-based therapy for individuals with SCD, hydroxyurea. Our team tackled the difficult task of addressing a vital problem in the SCD field (the low hydroxyurea uptake) by overcoming the challenges of forming a functional and integrated diverse team that had first to learn how to work together, share knowledge, and develop the conceptual models, to create the interventions, select, and combine implementation frameworks, and design the evaluation plans that are appropriate for use in complex multilevel interventions. Our diverse community of scientists, government representatives, clinicians, and patient partners recognized that a proper understanding of the implementation process would require careful consideration of the multi-level factors that can influence the implementation in the SCD population. Embedded in this concept was the notion that for the group to properly function and advance the research question, developing trust, respect, and having a shared vision was essential. The research team integration followed collaboration and team science principles, which set clear expectations for sharing credit, authorship, and maintaining self-awareness [52]. Strong communication was at the core of our team's functioning. The group consensus was initially slow to develop in trying to choose the design of the intervention (i.e., targets, function) and which model or framework for evaluation. Implementation researchers came with knowledge but often loyalty to a particular model, while clinical investigators were new to implementation methods and evaluation. The process creating a functional partnership between implementation science researchers and clinical investigators underscored the need to develop a “common language” between both teams while investing time to build knowledge of the respective fields across all team members. Organizing our thinking around a logic model and focusing on the “diagnosis of the implementation gap” (i.e., the determinants of the hydroxyurea utilization gap (illustrated in Figure 1) and choosing the frameworks second was instrumental in creating efficiency. Fortunately, there was sufficient lead time and support from the funding agency to examine the strengths and drawbacks associated with various models and frameworks and the possibility of combining aspects of each to reach an agreement. Accounting for the lead time (and possible delays) in creating multidisciplinary research teams and reaching team integration is important and should not be overlooked.
Our selection of theories, models, and frameworks followed our needs assessment phase. For this project, no models were adapted to fit the study intervention or evaluations. However, for some frameworks, not all constructs were applicable. For instance, the “Mobile health care systems self-efficacy” driver within TAM was less pertinent to developing the provider-level intervention. That's because, among SCD providers, the perception that mHealth could help with the task of prescribing hydroxyurea was less of a factor when the dominant perceived ease of use and usefulness of the intervention were used (i.e., judging if mHealth was helpful was less important than how much this tool could increase their productivity). Additionally, the outer setting construct within CFIR was not examined, as implementing mHealth in clinics is mainly controlled by the local leadership and policies and generally not regulated by the existing health systems policies. We did not encounter problems aligning our outcomes and influential contextual factors to the existing domains and constructs of the frameworks used. However, the issue of problems aligning with existing domains may occur depending on the research question, and population studied. Therefore, the careful selection of models and frameworks deserves extra attention and effort from the investigators, as adaptations to existing models and frameworks may be required.
Our work describing the process development of how to combine implementation models and frameworks to study how multi-level contextual factors optimally and comprehensively affect implementation outcomes adds to this growing literature. When models and frameworks are combined to evaluate the implementation of an intervention or practice, essential factors that caused the organization to reject or accept the intervention can be uncovered. For instance, formally appointing key stakeholders to ensure the fidelity of the intervention or gaining visible support from the system and local leaders are contextual factors that are not necessarily known as pre-conditions for optimal implementation [53, 54]. While CFIR has been used to evaluate the implementation of the transition to adult care activities in SCD [55], rare examples of the combination of different models and frameworks in SCD exist [56], and none focused on medication utilization. Whereas RE-AIM provides a practical framework for planning and evaluating mHealth interventions, other models such as TAM and CFIR could explain why implementation might succeed or fail if used proactively and help to identify relevant modifiable factors affecting adoption, implementation, and maintenance. By understanding the why, we hope to identify mediators and moderators of the intervention and narrow down the components of the intervention that should be modified, removed, or new ones that need to be created in further adaptation while shedding light on the mechanism of interventions and identification of new implementation strategies. Finally, adding qualitative data to CFIR will allow us to map the level of influence on the CFIR constructs and domains. A qualitative design can expand quantitative data and provide new hypotheses to explain why implementation succeeds. Therefore, adding a qualitative design when using CFIR can be useful.
## Limitations
Our process to select the theories, models, and frameworks was iterative and followed matching theories and strategies to the type of intervention we sought to implement to best fit the quality gap we were trying to address, namely, low hydroxyurea utilization in SCD. Our process ensured diverse perspectives and consensus. It is possible, however, that this selection was not optimal. For example, our choice of models and frameworks preceded the publication of the RE-AIM expansion, which recognizably could represent an alternative to our approach. RE-AIM and PRISM can also be combined to investigate contextual predictors of the implementation outcomes, and an expanded version of RE-AIM has recently been published [57]. Although the process development of the intervention and its evaluation were carefully developed and reported, not all implementation strategies might have been pre-identified for tracking. This may limit our ability to evaluate our implementation in the future fully. A pre-specification and comprehensive identification of all implementation strategies need to be done, and appropriate time should be allotted to this activity during study planning.
## Conclusions
In conclusion, we report the processes of developing a multi-level, multi-component intervention to foster greater use of hydroxyurea therapy among patients with SCD and detail this multi-level intervention and our comprehensive plan for its evaluation. Careful consideration of how the multiple components of the intervention can interact with the various targets and contextual factors will facilitate the description of the implementation's how, when, what, where, and who of the implementation and the why. The results of our mHealth adherence-enhancing study and future research will refine the evaluation approach to create new knowledge in developing an evaluation model for multi-level interventions to increase hydroxyurea uptake among adolescents and adults with SCD.
## Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s.
## Author contributions
JSH and LMK: study concept, wrote the manuscript. MBP, MEF, LD, AAB, HBB, VRG, NS: study concept, edited manuscript. CM, SRJ, AAK, JS, JAG: revised manuscript. All authors contributed to the article and approved the submitted version.
## Conflict of interest
JSH receives consultancy fees from Forma Therapeutics, CVS Health, and Global Blood Therapeutics.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, the funding agencies or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frhs.2022.1024541/full#supplementary-material.
## References
1. Kato GJ, Piel FB, Reid CD, Gaston MH, Ohene-Frempong K, Krishnamurti L. **Sickle cell disease**. *Nat Rev Dis Primers* (2018) **4** 18010. DOI: 10.1038/nrdp.2018.10
2. Desai P, Little J, Kanter J, Bridges K, Andemariam B, Lanzkron S. **Kneeling was the first step for sickle cell disease**. *Ann Intern Med* (2021) **174** 1004-5. DOI: 10.7326/M20-7418
3. Power-Hays A, McGann PT. **When actions speak louder than words - racism and sickle cell disease**. *N Engl J Med* (2020) **383** 1902-3. DOI: 10.1056/NEJMp2022125
4. Lee L, Smith-Whitley K, Banks S, Puckrein G. **Reducing health care disparities in sickle cell disease: a review**. *Public Health Rep* (2019) **134** 599-607. DOI: 10.1177/0033354919881438
5. DiMartino LD, Baumann AA, Hsu LL, Kanter J, Gordeuk VR, Glassberg J. **The sickle cell disease implementation consortium: translating evidence-based guidelines into practice for sickle cell disease**. *Am J Hematol* (2018) **93** E391-E5. DOI: 10.1002/ajh.25282
6. Yawn BP, Buchanan GR, Afenyi-Annan AN, Ballas SK, Hassell KL, James AH. **Management of sickle cell disease: summary of the 2014 evidence-based report by expert panel members**. *JAMA* (2014) **312** 1033-48. DOI: 10.1001/jama.2014.10517
7. Lobo CL, Pinto JF, Nascimento EM, Moura PG, Cardoso GP, Hankins JS. **The effect of hydroxcarbamide therapy on survival of children with sickle cell disease**. *Br J Haematol* (2013) **161** 852-60. DOI: 10.1111/bjh.12323
8. Le PQ, Gulbis B, Dedeken L, Dupont S, Vanderfaeillie A, Heijmans C. **Survival among children and adults with sickle cell disease in Belgium: benefit from hydroxyurea treatment**. *Pediatr Blood Cancer* (2015) **62** 1956-61. DOI: 10.1002/pbc.25608
9. de Araujo OM, Ivo ML, Ferreira Junior MA, Pontes ER, Bispo IM, de Oliveira EC. **Survival and mortality among users and non-users of hydroxyurea with sickle cell disease**. *Rev Lat Am Enfermagem* (2015) **23** 67-73. DOI: 10.1590/0104-1169.3385.2526
10. Voskaridou E, Christoulas D, Bilalis A, Plata E, Varvagiannis K, Stamatopoulos G. **The effect of prolonged administration of hydroxyurea on morbidity and mortality in adult patients with sickle cell syndromes: results of a 17-year, single-center trial (LaSHS)**. *Blood* (2010) **115** 2354-63. DOI: 10.1182/blood-2009-05-221333
11. Steinberg MH, Barton F, Castro O, Pegelow CH, Ballas SK, Kutlar A. **Effect of hydroxyurea on mortality and morbidity in adult sickle cell anemia: risks and benefits up to 9 years of treatment**. *JAMA* (2003) **289** 1645-51. DOI: 10.1001/jama.289.13.1645
12. Kang HA, Barner JC. **The association between hydroxyurea adherence and opioid utilization among Texas medicaid enrollees with sickle cell disease**. *J Manag Care Spec Pharm* (2020) **26** 1412-22. DOI: 10.18553/jmcp.2020.26.11.1412
13. Mathias JG, Nolan VG, Klesges LM, Badawy SM, Cooper WO, Hankins JS. **Hydroxyurea use after transitions of care among young adults with sickle cell disease and Tennessee medicaid insurance**. *JAMA Netw Open* (2021) **4** e2128971. DOI: 10.1001/jamanetworkopen.2021.28971
14. Ritho J, Liu H, Hartzema AG, Lottenberg R. **Hydroxyurea use in patients with sickle cell disease in a medicaid population**. *Am J Hematol* (2011) **86** 888-90. DOI: 10.1002/ajh.22134
15. Kayle M, Valle J, Paulukonis S, Holl JL, Tanabe P, French DD. **Impact of medicaid expansion on access and healthcare among individuals with sickle cell disease**. *Pediatr Blood Cancer* (2020) **67** e28152. DOI: 10.1002/pbc.28152
16. Alberts NM, Badawy SM, Hodges J, Estepp JH, Nwosu C, Khan H. **Development of the InCharge health Mobile app to improve adherence to hydroxyurea in patients with sickle cell disease: user-centered design approach**. *JMIR Mhealth Uhealth* (2020) **8** e14884. DOI: 10.2196/14884
17. Hankins JS, Shah N, DiMartino L, Brambilla D, Fernandez ME, Gibson RW. **Integration of mobile health into sickle cell disease care to increase hydroxyurea utilization: protocol for an efficacy and implementation study**. *JMIR Res Protoc* (2020) **9** e16319. DOI: 10.2196/16319
18. Fernandez ME, Ruiter RAC, Markham CM, Kok G. **Intervention mapping: theory- and evidence-based health promotion program planning: perspective and examples**. *Front Public Health* (2019) **7** 209. DOI: 10.3389/fpubh.2019.00209
19. Loiselle K, Lee JL, Szulczewski L, Drake S, Crosby LE, Pai AL. **Systematic and meta-analytic review: medication adherence among pediatric patients with sickle cell disease**. *J Pediatr Psychol* (2016) **41** 406-18. DOI: 10.1093/jpepsy/jsv084
20. Badawy SM, Thompson AA, Penedo FJ, Lai JS, Rychlik K, Liem RI. **Barriers to hydroxyurea adherence and health-related quality of life in adolescents and young adults with sickle cell disease**. *Eur J Haematol* (2017) **98** 608-14. DOI: 10.1111/ejh.12878
21. Walsh KE, Cutrona SL, Kavanagh PL, Crosby LE, Malone C, Lobner K. **Medication adherence among pediatric patients with sickle cell disease: a systematic review**. *Pediatrics* (2014) **134** 1175-83. DOI: 10.1542/peds.2014-0177
22. Lanzkron S, Strouse JJ, Wilson R, Beach MC, Haywood C, Park H. **Systematic review: hydroxyurea for the treatment of adults with sickle cell disease**. *Ann Intern Med* (2008) **148** 939-55. DOI: 10.7326/0003-4819-148-12-200806170-00221
23. Brousseau DC, Richardson T, Hall M, Ellison AM, Shah SS, Raphael JL. **Hydroxyurea use for sickle cell disease among medicaid-enrolled children**. *Pediatrics* (2019) **144** e20183285. DOI: 10.1542/peds.2018-3285
24. Smeltzer MP, Howell KE, Treadwell M, Preiss L, King AA, Glassberg JA. **Identifying barriers to evidence-based care for sickle cell disease: results from the sickle cell disease implementation consortium cross-sectional survey of healthcare providers in the USA**. *BMJ Open* (2021) **11** e050880. DOI: 10.1136/bmjopen-2021-050880
25. Kanter J, Gibson R, Lawrence RH, Smeltzer MP, Pugh NL, Glassberg J. **Perceptions of US adolescents and adults with sickle cell disease on their quality of care**. *JAMA Netw Open* (2020) **3** e206016. DOI: 10.1001/jamanetworkopen.2020.6016
26. Hodges JR, Phillips SM, Norell S, Nwosu C, Khan H, Luo L. **Intentional and unintentional nonadherence to hydroxyurea among people with sickle cell disease: a qualitative study**. *Blood Adv* (2020) **4** 4463-73. DOI: 10.1182/bloodadvances.2020001701
27. Stretcher V, Rosenstock IM. **The health belief model**. *Health Behavior and Health Education: Theory, Research, and Practice* (1997)
28. Bandura A. *Social cognitive theory. Handbook of theories of social psychology, vol 1* (2012) 349-73
29. Cronin RM, Hankins JS, Byrd J, Pernell BM, Kassim A, Adams-Graves P. **Modifying factors of the health belief model associated with missed clinic appointments among individuals with sickle cell disease**. *Hematology* (2018) **23** 683-91. DOI: 10.1080/10245332.2018.1457200
30. Schultz CL, Tchume-Johnson T, Schapira MM, Bellamy S, Smith-Whitley K, Ellison A. **Adherence to prompt fever evaluation in children with sickle cell disease and the health belief model**. *Pediatr Blood Cancer* (2015) **62** 1968-73. DOI: 10.1002/pbc.25634
31. Bollinger LM, Nire KG, Rhodes MM, Chisolm DJ, O'Brien SH. **Caregivers' perspectives on barriers to transcranial doppler screening in children with sickle-cell disease**. *Pediatr Blood Cancer* (2011) **56** 99-102. DOI: 10.1002/pbc.22780
32. Okocha EC, Gyamfi J, Ryan N, Babalola O, Etuk EA, Chianumba R. **Barriers to therapeutic use of hydroxyurea for sickle cell disease in Nigeria: a cross-sectional survey**. *Front Genet* (2021) **12** 765958. DOI: 10.3389/fgene.2021.765958
33. Nilsen P. **Making sense of implementation theories, models and frameworks**. *Implement Sci* (2015) **10** 53. DOI: 10.1186/s13012-015-0242-0
34. Jonassaint CR, Shah N, Jonassaint J, De Castro L. **Usability and feasibility of an mHealth intervention for monitoring and managing pain symptoms in sickle cell disease: the sickle cell disease Mobile application to record symptoms via technology (SMART)**. *Hemoglobin* (2015) **39** 162-8. DOI: 10.3109/03630269.2015.1025141
35. Shah N, Jonassaint J, De Castro L. **Patients welcome the sickle cell disease Mobile application to record symptoms via technology (SMART)**. *Hemoglobin* (2014) **38** 99-103. DOI: 10.3109/03630269.2014.880716
36. Lunyera J, Jonassaint C, Jonassaint J, Shah N. **Attitudes of primary care physicians toward sickle cell disease care, guidelines, and comanaging hydroxyurea with a specialist**. *J Prim Care Community Health* (2017) **8** 37-40. DOI: 10.1177/2150131916662969
37. Palmer M, Barnard S, Perel P, Free C. **Mobile phone-based interventions for improving adherence to medication prescribed for the primary prevention of cardiovascular disease in adults**. *Cochrane Database Syst Rev* (2018) **6** CD012675. PMID: 29932455
38. Schnall R, Rojas M, Bakken S, Brown W, Carballo-Dieguez A, Carry M. **A user-centered model for designing consumer mobile health (mHealth) applications (apps)**. *J Biomed Inform* (2016) **60** 243-51. DOI: 10.1016/j.jbi.2016.02.002
39. Lyon AR, Brewer SK, Arean PA. **Leveraging human-centered design to implement modern psychological science: return on an early investment**. *Am Psychol* (2020) **75** 1067-79. DOI: 10.1037/amp0000652
40. Badawy SM, DiMartino L, Brambilla D, Klesges L, Baumann A, Burns E. **Impact of the COVID-19 pandemic on the implementation of Mobile health to improve the uptake of hydroxyurea in patients with sickle cell disease: mixed methods study**. *JMIR Form Res* (2022) **6** e41415. DOI: 10.2196/41415
41. Waltz TJ, Powell BJ, Matthieu MM, Damschroder LJ, Chinman MJ, Smith JL. **Use of concept mapping to characterize relationships among implementation strategies and assess their feasibility and importance: results from the expert recommendations for implementing change (ERIC) study**. *Implement Sci* (2015) **10** 109. DOI: 10.1186/s13012-015-0295-0
42. Powell BJ, Waltz TJ, Chinman MJ, Damschroder LJ, Smith JL, Matthieu MM. **A refined compilation of implementation strategies: results from the expert recommendations for implementing change (ERIC) project**. *Implement Sci* (2015) **10** 21. DOI: 10.1186/s13012-015-0209-1
43. Cullen MR, Lemeshow AR, Russo LJ, Barnes DM, Ababio Y, Habtezion A. **Disease-specific health disparities: a targeted review focusing on race and ethnicity**. *Healthcare* (2022) **10** 603. DOI: 10.3390/healthcare10040603
44. Damschroder LJ, Aron DC, Keith RE, Kirsh SR, Alexander JA, Lowery JC. **Fostering implementation of health services research findings into practice: a consolidated framework for advancing implementation science**. *Implement Sci* (2009) **4** 50. DOI: 10.1186/1748-5908-4-50
45. Glasgow RE, Vogt TM, Boles SM. **Evaluating the public health impact of health promotion interventions: the RE-AIM framework**. *Am J Public Health* (1999) **89** 1322-7. DOI: 10.2105/AJPH.89.9.1322
46. Shelton RC, Chambers DA, Glasgow RE. **An extension of RE-AIM to enhance sustainability: addressing dynamic context and promoting health equity over time**. *Front Public Health* (2020) **8** 134. DOI: 10.3389/fpubh.2020.00134
47. Forman J, Heisler M, Damschroder LJ, Kaselitz E, Kerr EA. **Development and application of the RE-AIM QuEST mixed methods framework for program evaluation**. *Prev Med Rep* (2017) **6** 322-8. DOI: 10.1016/j.pmedr.2017.04.002
48. Davis F. **Perceived usefulness, perceived ease of use, and user acceptance of information technology**. *MIS Q* (1989) **13** 319-40. DOI: 10.2307/249008
49. Venkatesh V, Davis FD. **A model of the antecedents of perceived ease of use: development and test***. *Decision Sciences* (1996) **27** 451-81. DOI: 10.1111/j.1540-5915.1996.tb01822.x
50. Liang L, Bernhardsson S, Vernooij RW, Armstrong MJ, Bussieres A, Brouwers MC. **Use of theory to plan or evaluate guideline implementation among physicians: a scoping review**. *Implement Sci* (2017) **12** 26. DOI: 10.1186/s13012-017-0557-0
51. Davies P, Walker AE, Grimshaw JM. **A systematic review of the use of theory in the design of guideline dissemination and implementation strategies and interpretation of the results of rigorous evaluations**. *Implement Sci* (2010) **5** 14. DOI: 10.1186/1748-5908-5-14
52. Bennett LM, Gadlin H. **Collaboration and team science: from theory to practice**. *J Investig Med* (2012) **60** 768-75. DOI: 10.2310/JIM.0b013e318250871d
53. King DK, Shoup JA, Raebel MA, Anderson CB, Wagner NM, Ritzwoller DP. **Planning for implementation success using RE-AIM and CFIR frameworks: a qualitative study**. *Front Public Health* (2020) **8** 59. DOI: 10.3389/fpubh.2020.00059
54. Damschroder LJ, Reardon CM, AuYoung M, Moin T, Datta SK, Sparks JB. **Implementation findings from a hybrid III implementation-effectiveness trial of the diabetes prevention program (DPP) in the veterans health administration (VHA)**. *Implement Sci* (2017) **12** 94. DOI: 10.1186/s13012-017-0619-3
55. Calhoun C, Luo L, Baumann AA, Bauer A, Shen E, McKay V. **Transition for adolescents and young adults with sickle cell disease in a US midwest urban center: a multilevel perspective on barriers, facilitators, and future directions**. *J Pediatr Hematol Oncol* (2022) **44** e872-e80. DOI: 10.1097/MPH.0000000000002322
56. Bello-Manga H, Haliru L, Ahmed KA, Tabari AM, Farouk BU, Bahago GY. **Primary prevention of stroke in children with sickle cell Anemia in Nigeria: protocol for a mixed methods implementation study in a community hospital**. *JMIR Res Protoc* (2022) **11** e37927. DOI: 10.2196/37927
57. Holtrop JS, Estabrooks PA, Gaglio B, Harden SM, Kessler RS, King DK. **Understanding and applying the RE-AIM framework: clarifications and resources**. *J Clin Transl Sci* (2021) **5** e126. DOI: 10.1017/cts.2021.789
|
---
title: 'Prerequisites for implementing physical activity on prescription for children
with obesity in paediatric health care: A cross-sectional survey'
authors:
- Charlotte Boman
- Susanne Bernhardsson
- Katarina Lauruschkus
- Stefan Lundqvist
- Karin Melin
journal: Frontiers in Health Services
year: 2023
pmcid: PMC10012761
doi: 10.3389/frhs.2022.1102328
license: CC BY 4.0
---
# Prerequisites for implementing physical activity on prescription for children with obesity in paediatric health care: A cross-sectional survey
## Abstract
### Background
Physical inactivity is a main driver of childhood obesity that tracks into adulthood, making it crucial to address early in life. Swedish physical activity on prescription (PAP) is an effective intervention for increasing physical activity levels in adults and is being implemented in primary care in Sweden. Before implementing PAP for children, both intervention effectiveness and implementation prerequisites need to be examined. Framed by the Normalization Process Theory (NPT) domains, this study aimed to investigate perceptions of PAP amongst paediatric staff and managers working with children with obesity, as well as acceptability, appropriateness, feasibility, and barriers and facilitators for implementing PAP in paediatric health care.
### Methods
Staff and managers in 28 paediatric outpatient clinics in western Sweden were surveyed using validated implementation instruments and open-ended questions. Data were analysed using Mann–Whitney U tests and Kruskal–Wallis tests. Qualitative data were categorised into NPT domains.
### Results
The survey response rate was $54\%$ ($\frac{125}{229}$). Most respondents ($82\%$) reported PAP to be familiar and many ($56\%$) perceived it as a normal part of work; nurses and physiotherapists to a greater extent ($p \leq 0.001$). This was anticipated to increase in the future ($82\%$), especially amongst those with the longest work experience ($$p \leq 0.012$$). Respondents reported seeing the potential value in their work with PAP ($77\%$), being open to working in new ways to use PAP ($94\%$), and having confidence in their colleagues' ability to use PAP ($77\%$). Barriers and facilitators were found in all the NPT domains, mainly collective action and reflexive monitoring, where, for example, inadequacies of education, resources, and research on PAP for children were reported as barriers. Most respondents agreed that PAP was acceptable, appropriate, and feasible ($71\%$ to $88\%$).
### Conclusions
PAP is familiar and perceived as an acceptable, appropriate, and feasible intervention, and by many viewed as a normal part of clinical routines in paediatric outpatient clinics in western Sweden, especially by physiotherapists and nurses. Barriers and faciliators are mainly related to collective action and reflexive monitoring. The wide acceptance demonstrates receptiveness to PAP as an intervention to promote an active lifestyle for children with obesity.
## Introduction
Childhood obesity has increased dramatically in recent decades and prevalence remains high in many countries [1, 2], making it an urgent public health concern. The prevalence of obesity in European children aged 5–9 years was $11.4\%$ in 2016 [3]. In Sweden, $6\%$ of children aged 6–9 had obesity in 2019, an increase by $4\%$ since 2016 [4]. Additionally, the COVID-19 pandemic has driven weight gain amongst children (5–9), caused for example by decreased physical activity, increased screen time, and increased dietary intake [10]. Obesity is considered a complex multifactorial condition [11], which tracks into adulthood and is associated with cardiometabolic and psychosocial comorbidity as well as premature mortality (12–15). One of the main behavioural drivers and an important risk factor, is physical inactivity [16, 17], making it critical to address this issue early in life.
For children who are overweight or obese, studies have shown positive effects of physical activity on weight-related outcomes, e.g., body fat and insulin resistance [18, 19], while evidence for interventions to increase children's overall physical activity levels remains inconsistent [20, 21]. However, research highlights that although there is evidence for physical activity interventions, implementation strategies to translate evidence-based results into practice are lacking [22, 23]. In paediatric health care, behaviour-changing interventions are commonly used with the aim to improve dietary intake, increase physical activity, and reduce sedentary time [16]. Physical activity on prescription (PAP) is one such intervention that is being implemented in many countries, including Sweden, to promote lifestyle change in the form of increasing physical activity [24] and decreasing sedentary time [25]. The Swedish PAP intervention comprises three core components: a person-centred dialogue, individually tailored activity recommendation with a written prescription, and a structured follow-up [26].
Studies of PAP have shown effectiveness in adults, including patients with overweight or obesity, measured as increased physical activity levels [27], but for children there is a paucity of studies. One study [28] showed PAP to be both feasible and increasing physical activity levels amongst children with cerebral palsy, and one study [29] showed effects on BMI scores in children with obesity after a web-based intervention of which PAP was one component. The Swedish National Board of Health and Welfare's National Guidelines for Methods of Preventing Disease posit PAP as an evidence-based practice targeting adults [30]. Because an inactive lifestyle amongst Swedish children and youth is a common health concern [31], several regions in Sweden have started to use PAP for physically inactive children. As part of a combined lifestyle treatment, PAP might be a potentially behaviour-changing and structured intervention for children with obesity, in accordance with the description of requirements and needs in the national guidelines for treatment of childhood obesity in Sweden [32].
Several barriers and facilitators for implementation of health promoting interventions have been identified. A recent review showed that implementation support strategies, such as educational materials and meetings, opinion leaders, small incentives or grants, and tailored interventions may improve implementation of programmes to prevent obesity and promote physical activity for young children [23]. For adults, identified barriers for implementing PAP include practitioners' lack of knowledge about the intervention and lack of organisational support (33–36). Reports including paediatric contexts also identified lack of time, lack of evidence for PAP for children, and limited collaboration with activity organisers as barriers [37, 38]. Facilitators include affirmative attitudes amongst practitioners' and central and local supporting structures. However, no study has investigated the prerequisites amongst staff and managers for implementing PAP for children with obesity in paediatric health care.
Identifying implementation determinants before implementing a new intervention is crucial for implementation success. Especially in the highly complex healthcare context [39], implementing new interventions can be challenging. It is also important to understand what works and does not work in the implementation process, for which using a theory is recommended [40, 41]. The Normalization Process Theory (NPT), especially developed for use in health care, was designed to help us understand how complex interventions become implemented in routine healthcare practice [39]. This theory is concerned with explaining the work people do during the implementation process, and comprises four core constructs, or domains [42]. The constructs/domains can be described as a set of mechanisms that energise and shape implementation processes, with a focus on how an intervention can become part of everyday practice [43], making them relevant to assess before implementing a new intervention.
Other implementation determinants that are important to assess before implementing a new intervention are the acceptability, appropriateness, and feasibility of the intervention; three determinants often used during early-stage implementation and seen as leading indicators of implementation success [44]. There is a lack of knowledge about whether PAP is perceived as a suitable intervention for children with obesity. To address this knowledge gap, it is important to investigate the prerequisites, barriers and facilitators amongst staff and managers for implementing PAP for childhood obesity in paediatric health care. This knowledge is highly warranted before the intervention is implemented more widely.
The aims of this study were to examine [1] how staff and managers perceive PAP for children with obesity in terms of the NPT domains coherence, cognitive participation, collective action, and reflexive monitoring; [2] what barriers and facilitators they report for working with PAP for children with obesity; and [3] how they perceive acceptability, appropriateness, and feasibility of PAP for children with obesity.
## Study design and setting
The study design was a cross-sectional survey, guided by the NPT and collecting quantitative and qualitative data using a web-based questionnaire. Findings are reported, when applicable, according to the STROBE checklist [45].
The study was conducted in the paediatric healthcare organisations in Region Västra Götaland, Sweden, comprising 26 clinics, and four rehabilitation clinics providing healthcare services for children with obesity. The organisations all cater to children with obesity and offer specialist health services. Region Västra Götaland is Sweden's second largest county council, providing healthcare services to approximately 1.7 million residents in western Sweden. One major city, Gothenburg, is located in the region, while the rest of the region comprises three smaller cities, medium-sized towns, and rural areas located in four regional areas (Table 1). In Gothenburg, PAP has already been introduced amongst healthcare professionals, through for example education, tutoring, networking, and PAP clinics supporting families whose children have been prescribed physical activity.
**Table 1**
| Gothenburg | Regional area | Regional area.1 | Regional area.2 | Regional area.3 |
| --- | --- | --- | --- | --- |
| | Södra Bohuslän | Fyrbodal | Skaraborg | Södra Älvsborg |
| POC Frölunda | POC Kungälv | POC Dalsland | POC Lidköping | POC Alingsås |
| POC Hisingen | POC Mölndal | POC Lysekil | POC Mariestad | POC Lerum |
| POC Kungshöjd | POC Mölnlycke | POC NÄL | POC SkaS | POC Skene |
| POC Öckerö | POC Partille | POC Strömstad | POC Skövde | POC Ulricehamn |
| Obesity centre at Queen Silvia's Children's Hospital | POC Stenungssund | | | POC Viskan |
| Primary care rehabilitation clinic, Angered | POC Tjörn | | | Primary care rehabilitation clinic, Lerum |
| Primary care rehabilitation clinic, Gamlestaden | | | | Primary care rehabilitation clinic, Sörhaga |
| Specialist centre for children and youth, Angered | | | | |
| Specialist centre for children and youth, Gamlestaden | | | | |
## Participants
The inclusion criterium for participating in the survey was to be either staff or manager at a paediatric healthcare clinic or a rehabilitation clinic providing outsourced rehabilitation services for children with obesity, in Region Västra Götaland. No prior experience of working with PAP was required. Approximately 240 eligible participants were identified with the assistance of managers and administrative staff. The heads of departments approved the clinics' participation in the study; all 30 clinics accepted the invitation to participate.
## Data collection and outcomes
All eligible participants were invited to answer a web-based questionnaire comprising four validated instruments measuring implementation outcomes. The questionnaire was distributed via e-mail during a three-week period in February and March 2021. To increase response rate and reduce the risk of non-response bias, three reminders at one-week intervals were sent.
In this study, the NPT was used to investigate and understand the collective work with PAP for children with obesity at the paediatric healthcare clinics. To assess the implementation process from the perspective of staff directly involved in the work of implementing PAP and their managers, the Normalization MeAsure Development tool (NoMAD) [42] was used. This instrument was specifically developed for implementation in healthcare contexts [42] and can be applied at any stage of an implementation process [46]. It is adaptable to different interventions and settings, and can be combined with other measurements focusing on other dimensions of implementation [42].
The NoMAD instrument consists of 23 items, of which three general questions are indicators of normalisation answered on 11-point Likert-type scales ranging from “still feels very new” to “feels completely familiar” for item 1 and from “not at all” to “completely” for items 2 and 3 [46]. Twenty items target the four core NPT domains: [1] coherence, that is the “sense-making” work people do to initiate a new intervention or practice; [2] cognitive participation, described as the relational work around the practice; [3] collective action, the work to perform/operationalise the practice; and [4] reflexive monitoring, the appraisal work to understand the new practice [42]. Each item has two options, A and B. Option A is answered on 5-point Likert-type scales with response options “strongly agree”, “agree”, “neither agree nor disagree”, “disagree”, and “strongly disagree”. Option B is applicable only to those who found no relevance in answering option A [46].
In this study the validated Swedish version S-NoMAD [47] was used. The word “intervention” was replaced by “PAP for children with obesity” or just “PAP”. Although the instrument was developed for healthcare professionals directly involved in the intervention, we wanted to address the perspectives of both staff and managers. Some items were slightly modified by adding wording addressing managers, e.g., “Do you feel PAP is currently a normal part of your work/area of responsibility?” and “Management supports/I as a manager support PAP for children with obesity”.
To supplement NoMAD, the implementation determinants acceptability, appropriateness, and feasibility of implementing PAP in paediatric health care were assessed, adding the staffs’ and managers' perceptions and attitudes towards PAP. Acceptability is defined as the perception amongst stakeholders that a given treatment, service, practice, or innovation is agreeable, palatable, or satisfactory [44]. Appropriateness is the perceived fit, relevance, or compatibility of the innovation or evidence-based practice for a given practice setting, provider, or consumer; and/or perceived fit of the innovation to address a particular issue or problem. Feasibility is the extent to which a new treatment, or an innovation, can be successfully used or carried out within a given agency or setting [44]. These outcomes were measured with the Acceptability of Intervention Measure (AIM), Intervention Appropriateness Measure (IAM), and Feasibility of Intervention Measure (FIM) [48]. All three are validated instruments with the purpose of assessing the fit and match of a practice or intervention to a given context, targeting different criteria [48]. The measures comprise four items each, answered on 5-point ordinal scales with response options “strongly agree”, “agree”, “neither agree nor disagree”, “disagree”, and “strongly disagree”. The instruments were translated and cross-culturally adapted into Swedish, adapted to children with obesity, and validated [49].
In addition, we collected demographic data and data on PAP experience and prescribing frequency. Two open-ended questions explored barriers and facilitators, in which the respondents were given the opportunity to describe their own experiences and thoughts regarding determinants for implementing PAP.
## Data analysis
The quantitative variables are presented descriptively using frequencies and percentages and medians and interquartile ranges. To facilitate future comparisons with other studies, means and standard deviations (SD) are also presented. The respondents' practice location was categorised into Gothenburg and other regional areas of the Region Västra Götaland (Table 1). Work experience in the organisation was categorised into [1], <2 years; [2], 2–5 years; [3], 6–10 years; and [4], >10 years. Professions were categorised into six groups: [1], nurse; [2], physician; [3], dietician; [4], physiotherapist; [5], manager and [6], other. Number of years working with/taking decisions about PAP was categorised into [1], <3 years; [2], 3–5 years; and [3], >5 years or longer. Missing data analyses were performed using chi-square tests to examine proportion of managers and practice location amongst non-responders vs. responders.
To facilitate presentation and interpretation, response categories were merged into fewer categories. Responses to the three general items were coded as: 0–4 = not familiar and 5–10 = familiar for item 1; 0–4 = not a normal part of work and 5–10 = a normal part of work for item 2; and 0–4 = it will not become a normal part of work and 5–10 = it will become a normal part of work for item 3. For the NoMAD items, the disagree/strongly disagree response categories were merged into disagree, and the strongly agree/agree response categories were merged into agree. Item 3.2 was reverse-scored due to its negative wording. One item (2.2) was not analysed since it was accidentally removed from the questionnaire. For the AIM/IAM/FIM suite of instruments, the response categories completely disagree/disagree were merged into disagree and agree/completely agree were merged into agree.
Comparative analyses of participants from the Gothenburg clinics in which PAP has already been introduced vs. clinics in the rest of the region were performed using chi-square tests for the dichotomised general questions. Because the assumptions of the chi-square tests were not met for the NOMAD and AIM/IAM/FIM items, we performed Mann–Whitney U tests using the original 5-point scales. Differences between years of work experience in the organisation and between professions in all variables were performed using Kruskal–Wallis tests with pairwise comparisons, applying Bonferroni correction. For variables where there were significant differences in the main Kruskal-Wallis test, we only report significant differences in the pairwise comparisons. Because age and work experience correlated, no comparisons were made between age groups. A p-value of ≤0.05 was considered statistically significant. All cases for which all items in at least one instrument were completed, were included in the analyses.
Internal consistency of the NoMAD items was acceptable for coherence (Cronbach's α = 0.748), and questionable for cognitive participation (α = 0.600), collective action (α = 0.638), and reflexive monitoring (α = 0.687). For the the AIM/IAM/FIM measures, internal consistency was excellent for acceptability and appropriateness (Cronbach's α = 0.924 and 0.943, respectively) and good for feasibility (α = 0.892). Quantitative data were analysed using IBM SPSS, version 28 (IBM Corp, Armonk, NY).
The barriers and facilitators described in free text answers to the open-ended questions were coded and sorted into categories corresponding to the NPT domains. This was done in an iterative process by the first author together with two physiotherapist colleagues with experience of working with PAP for children.
## Results
A total of 229 healthcare professionals (of whom 30 managers) were invited to participate in the survey, and 125 responded (response rate $54.5\%$). Of the 104 non-responders, 18 were managers. Missing data analysis showed no significant differences between responders and non-responders related to the proportion of managers or practice location in the *Gothenburg area* vs. other regional areas. Item-level missing values ranged from 7 to 12 ($5.6\%$–$9.6\%$) for S-NoMAD and from 0 to 8 ($0.0\%$–$6.4\%$) for AIM, IAM, and FIM. Mean age of the respondents was 48.2 years (SD 9.6). Respondent characteristics are presented in Table 2.
**Table 2**
| Characteristic | n (%) |
| --- | --- |
| Age (years) | Age (years) |
| <30 | 4 (3.2) |
| 30–39 | 21 (16.8) |
| 40–49 | 44 (35.2) |
| 50–59 | 37 (29.6) |
| >59 | 19 (15.2) |
| Work experience in the organisation (years) | Work experience in the organisation (years) |
| <2 | 26 (20.8) |
| 2–5 | 42 (33.6) |
| 6–10 | 27 (21.6) |
| >10 | 30 (24.0) |
| Profession | Profession |
| Nurse, including paediatric nurse | 43 (34.4) |
| Physician, including paediatrician | 32 (25.6) |
| Dietician | 13 (10.4) |
| Physiotherapist | 9 (7.2) |
| Manager | 12 (9.6) |
| Othera | 16 (12.8) |
| Role in relation to PAP | Role in relation to PAP |
| Works with PAP | 68 (54.4) |
| Is aware of PAP but does not work with it | 56 (44.8) |
| Is not aware of PAP | 1 (0.8) |
| Experience of working with PAP (years) | Experience of working with PAP (years) |
| <3 | 62 (49.6) |
| 3–5 | 31 (24.8) |
| >5 | 32 (25.6) |
| Frequency of prescribing PAP | Frequency of prescribing PAP |
| Prescribers | 64 (51.2) |
| Daily | 2 (1.6) |
| Once per week | 10 (8.0) |
| Once per month | 27 (21.6) |
| Once per year | 25 (20.0) |
| Non-prescribers | 61 (48.8) |
## General questions about PAP
A majority of the respondents ($81.1\%$) reported being familiar with PAP (Table 3). A higher proportion of respondents in the *Gothenburg area* reported being familiar with PAP than those in regional areas ($90.0\%$ vs. $70.6\%$, χ2 = 6.772, $$p \leq 0.009$$). Physiotherapists reported familiarity with PAP to a greater extent than “other” professions (Mdn 9 vs. Mdn 5, $U = 55$, $$p \leq 0.025$$). Fifty-six percent described PAP as currently being a normal part of their work; a higher proportion of respondents from the *Gothenburg area* reported this than those in the regional areas ($70.0\%$ vs. $40.7\%$, χ2 = 9.882, $$p \leq 0.002$$). Nurses reported feeling PAP was a normal part of their work to a greater extent than “other” professions (Mdn 6.5 vs. Mdn 0.5, $U = 38$, $p \leq 0.001$), as did physiotherapists (Mdn 9 vs. Mdn 0.5, $U = 53$, $p \leq 0.001$). A majority ($82.0\%$) reported believing that PAP will become a normal part of their work. Respondents with >10 years of work experience in the organisation reported this belief to a greater extent than those with 2–5 years' experience (Mdn 9 vs. Mdn 6, $U = 23$, $$p \leq 0.012$$).
**Table 3**
| Items | N (missing) | 0–41. Not familiar2. Currently not a normal part of work3. Will not become a normal part of workn (%) | 5–101. Familiar2. Currently a normal part of work3. Will become a normal part of workn (%) | Median (IQR)a | Mean (SD)a |
| --- | --- | --- | --- | --- | --- |
| 1. When you use PAP, how familiar does it feel? | 111 (14) | 21 (18.9) | 90 (81.1) | 7 (5–9) | 6.53 (2.61) |
| 2. Do you feel PAP is currently a normal part of your work/area of responsibility? | 114 (11) | 50 (43.9) | 64 (56.1) | 5 (2–8) | 4.99 (3.40) |
| 3. Do you feel PAP will become a normal part of your work/area of responsibility? | 89 (36) | 16 (18.0) | 73 (82.0) | 7 (6–9) | 7.08 (2.62) |
## Coherence
Most respondents ($67.9\%$) agreed that they could distinguish between PAP and their usual ways of working, and $56.1\%$ reported that they have a shared understanding of its purpose (Table 4). Respondents in the *Gothenburg area* agreed to a greater extent than those in regional areas to having a shared understanding of PAP (Mdn 4 vs. Mdn 3, $U = 1170$, $$p \leq 0.005$$) and of how the intervention affects the nature of their work (Mdn 4 vs. Mdn 3, $U = 1104$, $$p \leq 0.017$$) (Table 5). About three quarters of the respondents ($76.6\%$) agreed on the potential value of PAP. No differences were seen related to work experience in the organisation or profession in this domain. Option B responses were selected by 3–8 respondents ($2.6\%$ to $6.8\%$).
A barrier for using PAP described in the open-ended questions was the respondents' experiences of not knowing the PAP intervention well enough and working with single components alone, particularly the written prescription for physical activity. The opposite, a comprehension of the PAP intervention and considering and including all of its components, was described as a facilitator. Statements like “I consider it important that PAP is well supported by a good assessment so it will be at the right level, for example goal setting, activity, duration, and that the patient is motivated. If not, then it might just be ‘another piece of paper’ for the individual.” were typical.
## Cognitive participation
Almost half ($47.2\%$) agreed that there are key people who drive PAP forward and get others involved. Respondents in Gothenburg agreed to a greater extent than those in regional areas that there are key people driving PAP forward and involving others (Mdn 4 vs. Mdn 3, $U = 1000$, $$p \leq 0.003$$). Most reported being open to working with colleagues in new ways to use PAP ($94.5\%$) and agreed to continuing to support PAP ($85.7\%$). No differences were seen related to work experience or profession in this domain. Option B responses were selected by 1–8 respondents ($0.9\%$ to $6.8\%$).
A reported barrier in this domain for using PAP was the absence of physiotherapists at the clinics and the perceived uncoordinated pathways to healthcare units offering PAP support. Facilitators for using PAP were colleagues being supportive of PAP and successful healthcare collaboration. Statements like “In my clinic we have divided the tasks between us a little. However, I could prescribe PAP more often, but mostly it's done by my colleague who is a nurse.” were reported.
## Collective action
Over half of the respondents ($57.8\%$) agreed they can easily integrate PAP into their existing work and only $1.8\%$ agreed that PAP disrupts working relationships. A majority ($77.2\%$) reported having confidence in their colleagues’ ability to use PAP. Over half ($56\%$) agreed that work is assigned to those with skills appropriate to PAP. One fourth ($26\%$) agreed that sufficient training is provided to enable staff and managers to implement PAP. Respondents in Gothenburg agreed to a greater extent than those in regional areas that work is assigned to those with skills appropriate to PAP (Mdn 4 vs. Mdn 3, $U = 1116$, $$p \leq 0.032$$), that sufficient training to implement PAP is provided (Mdn 3 vs. Mdn 2, $U = 704$, $p \leq 0.001$), and that sufficient resources to support PAP are available (Mdn 3 vs. Mdn 3, $U = 1121$, $$p \leq 0.029$$). No differences were seen related to work experience or profession. One fourth ($26.9\%$) reported that sufficient resources are available to support PAP and half ($51.1\%$) agreed that management adequately supports PAP. Option B responses were selected by 1–22 respondents ($0.9\%$ to $19.5\%$).
Barriers from the open-ended questions were inadequate education and insufficient time to use PAP. Statements like “I would like to learn more about PAP, but I have too many duties to have time to plunge into it. It's not my most prioritised task, instead it's something I do on the side, a few times a month” were typical. Facilitators were staff taking on the role of using PAP and having more time with patients when delivering PAP.
## Reflexive monitoring
Thirty-seven percent reported being aware of reports about the effects of PAP. Managers agreed to a higher extent than “other” professions that they were aware of reports (Mdn 4 vs. Mdn 3, $U = 45$, $$p \leq 0.013$$) and respondents with more than 10 years of work experience in the organisation agreed to a higher extent than those with 6–10 years of experience that they were aware of reports (Mdn 4 vs. Mdn 2, $U = 30$, $$p \leq 0.007$$). Sixty percent agreed that PAP is worthwhile and $48.5\%$ valued the effects PAP has had on their work. Respondents in the *Gothenburg area* agreed to a greater extent than those in regional areas that they valued the effects (Mdn 4 vs. Mdn 3, $U = 890$, $$p \leq 0.011$$). The respondents agreed that feedback about PAP can be used to improve it in the future ($81.2\%$). No differences were seen related to work experience. Option B responses were selected by 2–22 respondents ($1.7\%$ to $12.1\%$).
A reported barrier for using PAP was the lack of research on PAP for children. Statements like “I’d like to see randomised studies that are large enough to show the effectiveness of PAP if I am to become positive about the intervention” are illustrative. The opportunity to provide discounted activities was reported as an important facilitator.
## Acceptability
Most respondents stated that PAP meets with their approval ($85.6\%$), is appealing ($85.6\%$), and that they like ($84.0\%$) and welcome ($83.2\%$) PAP (Table 6). Respondents in the *Gothenburg area* agreed to a greater extent than those in regional areas that PAP meets their approval (Mdn 5 vs. Mdn 4, $U = 1528$, $$p \leq 0.022$$). Respondents with more than 10 years of work experience agreed to a higher extent than those with 2–5 years of experience that they welcome working with PAP (Mdn 5 vs. Mdn 4, $U = 23$, $$p \leq 0.019$$). No differences were found by profession.
**Table 6**
| Statement | N (missing) | Agree n (%) | Neutral n (%) | Disagree n (%) | Mediana (IQR) | Mean (SD)a |
| --- | --- | --- | --- | --- | --- | --- |
| Acceptability | Acceptability | Acceptability | Acceptability | Acceptability | Acceptability | Acceptability |
| PAP meets my approval | 125 | 107 (85.6) | 17 (13.6) | 1 (0.8) | 5 (5–5) | 4.36 (0.75) |
| PAP is appealing to me | 124 (1) | 107 (85.6) | 16 (12.8) | 1 (0.8) | 4.5 (4.5–5) | 4.35 (0.74) |
| I like PAP | 125 | 105 (84.0) | 18 (14.4) | 2 (1.6) | 4 (4–5) | 4,31 (0.78) |
| I welcome PAP | 124 (1) | 104 (83.2) | 18 (14.4) | 2 (1.6) | 5 (5–5) | 4.35 (0.79) |
| Appropriateness | Appropriateness | Appropriateness | Appropriateness | Appropriateness | Appropriateness | Appropriateness |
| PAP seems fitting | 123 (2) | 102 (81.6) | 16 (12.8) | 5 (4.0) | 5 (5–5) | 4.24 (0.90) |
| PAP seems suitable | 123 (2) | 104 (83.2) | 15 (12.0) | 4 (3.2) | 4 (4–5) | 4.32 (0.81) |
| PAP seems applicable | 121 (4) | 100 (80.0) | 19 (15.2) | 2 (1.6) | 5 (5–5) | 4.25 (0.78) |
| PAP seems like a good match | 117 (8) | 98 (78.4) | 12 (9.6) | 7 (5.6) | 4 (4–5) | 4.24 (0.87) |
| Feasibility | Feasibility | Feasibility | Feasibility | Feasibility | Feasibility | Feasibility |
| PAP seems implementable | 118 (7) | 98 (78.4) | 16 (12.8) | 4 (3.2) | 4 (4–5) | 4.23 (0.81) |
| PAP seems possible | 122 (3) | 110 (88.0) | 12 (9.6) | 0 (0.0) | 5 (5–5) | 4.43 (0.67) |
| PAP seems doable | 122 (3) | 103 (82.4) | 16 (12.8) | 3 (2.4) | 4 (4–5) | 4.25 (0.78) |
| PAP seems easy to use | 117 (8) | 89 (71.2) | 26 (20.8) | 2 (1.6) | 4 (4–5) | 3.99 (0.77) |
## Appropriateness
Most agreed that PAP seems fitting ($81.6\%$), suitable ($83.2\%$), applicable ($80\%$), and like a good match ($78.4\%$) for children with obesity (Table 6). No differences were seen by practice location, profession, or years of work experience.
## Feasibility
Most respondents reported PAP being implementable ($78.4\%$), possible ($88\%$), doable ($82.4\%$), and easy to use ($71.2\%$) (Table 6). No differences were found by practice location, profession, or years of work experience.
## Discussion
This study reports prerequisites and determinants for implementing the PAP intervention for children with obesity amongst healthcare professionals at paediatric clinics in western Sweden. Our findings suggest that those prerequisites are good, and that, in fact, implementation is underway to various extents. Main findings are that most respondents perceive PAP as familiar and many, in particular nurses and physiotherapists, as a normalised part of their work. Barriers and facilitators for working with PAP were identified across all NPT domains, especially related to collective action and reflexive monitoring. The respondents perceived PAP as highly acceptable, appropriate, and feasible, regardless of profession and experience of working in the organisation.
Respondents from the *Gothenburg area* perceived PAP as more normalised than those in regional areas; a geographical difference seen in all the NPT domains as well as regarding acceptability of the intervention. Identified facilitators for PAP use were comprehension of the PAP intervention, taking on the role of using PAP, and the interventions's ease of use. Barriers were inadequate education, insufficient time, uncoordinated pathways to other healthcare units, poor collaboration with activity organisers, and the lack of research on PAP for children.
The geographical differences are likely attributed to the PAP support structure that has been in place in Gothenburg for several years. Gothenburg represents a unique context in Sweden, with a PAP support structure in the form of education, networking, and PAP clinics to which patients are referred for extra support in changing their physical activity patterns. None of these support structures are established elsewhere in the region or in Sweden, and there are considerable regional variations across Sweden in the support for work with PAP [37].
Nurses and physiotherapists perceived PAP as normalised to a great extent. Both professions have worked with PAP for many years in Sweden, particularly for adults. Studies in adult populations have also shown nurses' engagement in PAP and other types of physical activity referrals [33, 35, 36]. In paediatric health care, nurses have a central role in the work with children and families, including counselling about physical activity and following up intervention effects.
Most respondents perceived PAP as acceptable, appropriate and feasible for children with obesity. Feasibility of PAP as part of an internet-based intervention for children with obesity was recently reported in another Swedish study [29]. However, as PAP was one of three intervention components, it is not possible to attribute the results to PAP alone. Amongst adults, feasibility and effects of PAP have recently been shown in two studies, of which one showed sustained results five years after the intervention [50, 51]. Although not yet evaluated as a stand-alone intervention in children with obesity, the high acceptability, appropriateness and feasibility of PAP found in our and other studies are important prerequisites for future studies on effectiveness in this population.
Both staff and managers perceived PAP as a possible intervention, implying an understanding of the feasibility of using it in routine clinical practice and the possibility of implementing it in paediatric health care. The high acceptability of PAP by managers is an important prerequisite to the normalisation of PAP. This finding is in contrast to previous studies on PAP, which have identified lack of supportive management [35] and organisational support [33, 35, 36, 38] as problematic.
One reason for the high scores on appropriateness of PAP may be the intervention's person-centredness and individually tailored components, which correspond well with a respectful and structured obesity management according to Swedish national guidelines [32]. Another reason might be the discounts offered for many of the prescribed physical activities, which can enable the child's participation in an activity. Families with obese children are often socio-economically disadvantaged [52], so this financial incentive could be an important facilitator.
The collective and individual understanding of an intervention and how it differs from usual ways of working is important for clinical practice [42]. In the domain coherence, almost two thirds of the respondents reported they could “make sense” of PAP and understand how it affected their work. These findings were nuanced by qualitative data where respondents expressed insufficient knowledge of PAP and uncertainty about its clinical use. Similar findings have been shown in previous research on PAP for adults, where lack of information and knowledge about PAP and its application was found amongst practitioners [34, 35].
Patients have described not receiving sufficient information about PAP during an intervention period [53]. Our findings show a variation in the respondents' perceptions of PAP and its usability in paediatric health care. It is natural for healthcare professionals to experience uncertainty regarding the rationale and clinical use of PAP, particularly in a context for which the intervention has not primarily been developed. This variation in perceptions might reflect that the work with PAP has been transferred from an adult context to the paediatric context without having been fully developed and adapted for children with obesity, which may contribute to uncertainty about its application.
For successful integration into practice, the collective contribution to enact and sustain the work with a new intervention is important. Regardless of profession and years of working in the organisation, most items in the cognitive participation domain were scored high amongst the respondents in our study. In the open-ended questions, respondents described how PAP work was organised in their own clinic and amongst other clinics with licensed practitioners. In the Gothenburg area, key people were driving the PAP work forward and could share good experiences with new colleagues.
The lack of physiotherapists in the paediatric healthcare organisation was described as a barrier, implying that physiotherapists are viewed as one of the most legitimate professions for working with PAP. Physiotherapists' familiarity with PAP and their perception that PAP is already a normal part of their work also corroborate this view. Physiotherapist is a profession with skills for working with physical activity [54], but that is largely missing in paediatric health care. To access these skills and competency, some staff referred patients onward to physiotherapists in PAP clinics or rehabilitation clinics. This uncoordinated referral system between prescribers and physiotherapists was seen as a barrier for working with PAP. Nevertheless, another study described a similar referral setup in primary and secondary care for adults, in which patients perceived PAP to be both feasible and increasing physical activity [50]. Hence, the need for formal and coordinated referral pathways between clinics may be greater for children and their families than for adults. The lack of coordination between clinics has been identified earlier as a considerable barrier for families [38].
To improve work with PAP, many respondents called for more training. In the collective action domain, lack of training, structure, and time was described as barriers to efficiently delivering PAP. Similar barriers have also been reported for adult populations managed in primary care [33, 35, 36], as well as for children with intellectual developmental disorders [38]. A recent systematic review of implementation of obesity prevention interventions for children also identified lack of knowledge, e.g., concerning physical activity recommendations, as a barrier amongst primary care nurses and physicians [55].
Only half of our respondents, including managers, agreed that management adequately supports PAP. However, in view of the high acceptability and feasibility of PAP reported by both staff and managers, the perceived lack of management support may imply poor communication between staff and managers rather than an actual lack of support. Improved communication and collaboration amongst staff and managers would likely improve chances for an intervention to become normalised in routine practice. Insufficient training, managerial support, and resources were reported as important barriers for implementing physical activity prevention interventions for children with obesity also in primary care [55].
In the reflexive monitoring domain almost $40\%$ of the respondents agreed they were aware of reports about the effects of PAP. This finding is difficult to interpret since research is mostly lacking on PAP for children, but communal or individual evaluations may have been undertaken in clinical practice. Managers reported being aware of effects to a greater extent than other professions, possibly implying they might be better informed by policy documents and national guideline recommendations [30, 56] about the health benefits of physical activity for children. Although lack of research on PAP for children was reported as a barrier, staff might also recognise PAP as an evidence-based intervention for adults and could have gained knowledge through networking, education, and information material for both adults and children.
## Strengths and limitations
A main strength of the study is the use of a theory-based framework and instrument to assess and categorise the factors that might influence implementation of PAP in the paediatric context. A particular strength in using the NPT is its focus on the implementation work healthcare professionals actually do, rather than their cognitions, e.g., beliefs and attitudes. Another strength is our use of validated instruments, which are also pragmatic and easy to use. The NoMAD was particularly helpful in pointing out problems that can be addressed when implementing PAP for children with obesity, enabling improvements related to collective action and reflexive monitoring. Assessing the dual perspective of practitioners and managers also strengthens the findings. Supplementing NoMAD with the AIM, IAM, and FIM instruments to assess important implementation determinants provided a comprehensive overview of aspects necessary to address in a future implementation of PAP for children with obesity. Several efforts were made to reduce bias. Sampling bias was minimised since the survey was distributed to all staff and managers at all paediatric clinics in the study population. We attempted to reduce non-response bias by sending several reminders to answer the questionnaire.
There were some limitations to the study. The intention to capture multiple perspectives meant that not all participants had practical experience of PAP, making several questions irrelevant for some respondents and likely contributing to both unit-level and item-level missing data. The use of self-reported data entails a risk for both self-selection bias and social desirability bias. We did not perform sensitivity analyses, but believe our analyses are robust enough with the used tests. We did not investigate gender, because a vast majority of both practitioners and managers in the population studied are women. The low alpha values for some of the NoMAD items indicate low internal consistency, which might have affected the results of the statistical analyses.
There is an obvious need for research on effectiveness of physical activity promoting interventions for childhood obesity, as well as implementation process and outcome evaluations of such interventions. To improve the understanding of barriers and facilitators for using PAP, further research is needed from the perspective of staff and managers, as well as that of the children and their parents.
Our study can provide helpful information to develop support structures for PAP work, streamline the use of the intervention, and inform future implementation strategies. The broad inclusion criteria of the study, including all professions and managers involved in paediatric health care, and the study setting – Region Västra Götaland which is Sweden's second largest county council – enhances generalisability of our findings to other paediatric populations and to other regions in Sweden, and possibly also to other countries with similar paediatric healthcare systems.
## Conclusions
The prerequisites for implementing PAP for children with obesity in paediatric health care in western Sweden can be considered good. The intervention is familiar and perceived as acceptable, appropriate, and feasible by paediatric healthcare practitioners and managers, constituting important facilitators for implementing PAP. For many participants, PAP was already perceived as a normal part of their work, and a majority believed it would become a normal part of their work in the future. The wide acceptance demonstrates receptiveness to PAP as an intervention to promote an active lifestyle for children with obesity. Barriers and facilitators for working with PAP exist in all NPT domains, particularly in the domains collective action and reflexive monitoring where main barriers are the lack of education, resources, and research on PAP for children.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The Swedish Ethical Review Authority issued an advisory statement that the authority has no ethical objections to the study (Reference no. 2020-05693). Because no intervention was involved and no sensitive personal data were collected, ethics review was not required for this study (SFS 2003:460). Participants provided their informed consent by checking a box within the survey.
## Author contributions
CB contributed to study design, led data collection, data analysis, and drafted the manuscript. SB, KL, SL and KM contributed to study design, data collection, data analysis and revised the manuscript. All authors contributed to the article and approved the submitted version.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Garrido-Miguel M, Cavero-Redondo I, Alvarez-Bueno C, Rodriguez-Artalejo F, Moreno LA, Ruiz JR. **Prevalence and trends of overweight and obesity in European children from 1999 to 2016: a systematic review and meta-analysis**. *JAMA Pediatr* (2019.0) **173** e192430. DOI: 10.1001/jamapediatrics.2019.2430
2. **Worldwide trends in body-mass index, underweight, overweight, and obesity from 1975 to 2016: a pooled analysis of 2416 population-based measurement studies in 128.9 million children, adolescents, and adults**. *Lancet* (2017.0) **390** 2627-42. DOI: 10.1016/S0140-6736(17)32129-3
3. 3.World Health Organization. Global Health Observatory (2020). . Available at:
https://www.who.int/data/gho/data/themes/theme-details/GHO/body-mass-index-(bmi)
(Accessed October 24, 2022).
4. 4.Public Health Agency of Sweden. Overweight and obesity are common and increase with age in 6–9 year olds. Report no. 20100. Available at:
https://www.folkhalsomyndigheten.se/publicerat-material/publikationsarkiv/oe/overvikt-och-fetma-ar-vanligt-och-okar-med-alder-hos-6-9-aringar/
(Accessed October 24, 2022).
5. Jenssen BP, Kelly MK, Powell M, Bouchelle Z, Mayne SL, Fiks AG. **COVID-19 and changes in child obesity**. *Pediatrics* (2021.0) **147** 1-3. DOI: 10.1542/peds.2021-050123
6. Jia P, Zhang L, Yu W, Yu B, Liu M, Zhang D. **Impact of COVID-19 lockdown on activity patterns and weight status among youths in China: the COVID-19 impact on lifestyle change survey (COINLICS)**. *Int J Obes (Lond)* (2021.0) **45** 695-9. DOI: 10.1038/s41366-020-00710-4
7. Lange SJ, Kompaniyets L, Freedman DS, Kraus EM, Porter R, Blanck HM. **Longitudinal trends in body mass index before and during the COVID-19 pandemic among persons aged 2–19 years—United States, 2018–2020**. *MMWR Morb Mortal Wkly Rep.* (2021.0) **70** 1278-83. DOI: 10.15585/mmwr.mm7037a3
8. Vogel M, Geserick M, Gausche R, Beger C, Poulain T, Meigen C. **Age- and weight group-specific weight gain patterns in children and adolescents during the 15 years before and during the COVID-19 pandemic**. *Int J Obes (Lond)* (2022.0) **46** 144-52. DOI: 10.1038/s41366-021-00968-2
9. Woolford SJ, Sidell M, Li X, Else V, Young DR, Resnicow K. **Changes in body mass index among children and adolescents during the COVID-19 pandemic**. *JAMA* (2021.0) **326** 1434-6. DOI: 10.1001/jama.2021.15036
10. Pietrobelli A, Pecoraro L, Ferruzzi A, Heo M, Faith M, Zoller T. **Effects of COVID-19 lockdown on lifestyle behaviors in children with obesity living in verona, Italy: a longitudinal study**. *Obesity (Silver Spring)* (2020.0) **28** 1382-5. DOI: 10.1002/oby.22861
11. Kopelman P, Jebb SA, Butland B. **Executive summary: foresight “tackling obesities: future choices” project**. *Obes Rev* (2007.0) **8** vi-ix. DOI: 10.1111/j.1467-789X.2007.00344.x
12. Horesh A, Tsur AM, Bardugo A, Twig G. **Adolescent and childhood obesity and excess morbidity and mortality in young adulthood-a systematic review**. *Curr Obes Rep* (2021.0) **10** 301-10. DOI: 10.1007/s13679-021-00439-9
13. Jebeile H, Cardel MI, Kyle TK, Jastreboff AM. **Addressing psychosocial health in the treatment and care of adolescents with obesity**. *Obesity (Silver Spring)* (2021.0) **29** 1413-22. DOI: 10.1002/oby.23194
14. Pulgaron ER. **Childhood obesity: a review of increased risk for physical and psychological comorbidities**. *Clin Ther* (2013.0) **35** A18-32. DOI: 10.1016/j.clinthera.2012.12.014
15. Simmonds M, Llewellyn A, Owen CG, Woolacott N. **Predicting adult obesity from childhood obesity: a systematic review and meta-analysis**. *Obes Rev* (2016.0) **17** 95-107. DOI: 10.1111/obr.12334
16. Luttikhuis H O, Baur L, Jansen H, Shrewsbury VA, O'Malley C, Stolk RP. **Interventions for treating obesity in children**. *Cochrane Database Syst Rev* (2009.0) CD001872. DOI: 10.1002/14651858.CD001872.pub2
17. LeCroy MN, Kim RS, Stevens J, Hanna DB, Isasi CR. **Identifying key determinants of childhood obesity: a narrative review of machine learning studies**. *Child Obes* (2021.0) **17** 153-9. DOI: 10.1089/chi.2020.0324
18. Headid RJ, Park SY. **The impacts of exercise on pediatric obesity**. *Clin Exp Pediatr* (2021.0) **64** 196-207. DOI: 10.3345/cep.2020.00997
19. Whooten R, Kerem L, Stanley T. **Physical activity in adolescents and children and relationship to metabolic health**. *Curr Opin Endocrinol Diabetes Obes* (2019.0) **26** 25-31. DOI: 10.1097/MED.0000000000000455
20. Metcalf B, Henley W, Wilkin T. **Effectiveness of intervention on physical activity of children: systematic review and meta-analysis of controlled trials with objectively measured outcomes (EarlyBird 54)**. *Br Med J* (2012.0) **345** e5888. DOI: 10.1136/bmj.e5888
21. Nooijen CF, Galanti MR, Engstrom K, Moller J, Forsell Y. **Effectiveness of interventions on physical activity in overweight or obese children: a systematic review and meta-analysis including studies with objectively measured outcomes**. *Obes Rev* (2017.0) **18** 195-213. DOI: 10.1111/obr.12487
22. McGoey T, Root Z, Bruner MW, Law B. **Evaluation of physical activity interventions in children via the reach, efficacy/effectiveness, adoption, implementation, and maintenance (RE-AIM) framework: a systematic review of randomized and non-randomized trials**. *Prev Med* (2016.0) **82** 8-19. DOI: 10.1016/j.ypmed.2015.11.004
23. Wolfenden L, Barnes C, Jones J, Finch M, Wyse RJ, Kingsland M. **Strategies to improve the implementation of healthy eating, physical activity and obesity prevention policies, practices or programmes within childcare services**. *Cochrane Database Syst Rev* (2020.0) **2** Cd011779. DOI: 10.1002/14651858.CD011779.pub3
24. Arsenijevic J, Groot W. **Physical activity on prescription schemes (PARS): do programme characteristics influence effectiveness? Results of a systematic review and meta-analyses**. *BMJ Open* (2017.0) **7** e012156. DOI: 10.1136/bmjopen-2016-012156
25. 25.Swedish agency for health technology assessment and assessment of social services. Metods of promoting physical activity: a systematic review. Stockholm: Swedish agency for health technology assessment and assessment of social services (SBU). (2006), SBU-rapport:181.
26. Kallings LV. **The Swedish approach on physical activity on prescription. (“Implementation of physical activity in health care—facilitators and barriers”)**. *Clinical Health Promotion* (2016.0) **2016** 31-3
27. Onerup A, Arvidsson D, Blomqvist A, Daxberg EL, Jivegard L, Jonsdottir IH. **Physical activity on prescription in accordance with the Swedish model increases physical activity: a systematic review**. *Br J Sports Med* (2019.0) **53** 383-8. DOI: 10.1136/bjsports-2018-099598
28. Lauruschkus K, Hallstrom I, Westbom L, Tornberg A, Nordmark E. **Participation in physical activities for children with cerebral palsy: feasibility and effectiveness of physical activity on prescription**. *Arch Physiother* (2017.0) **7** 13. DOI: 10.1186/s40945-017-0041-9
29. Thorén A, Janson A, Englund E, Silfverdal SA. **Development, implementation and early results of a 12-week web-based intervention targeting 51 children age 5–13 years and their families**. *Obes Sci Pract* (2020.0) **6** 516-23. DOI: 10.1002/osp4.440
30. 30.The Swedish National Board of Health and Welfare. National guidelines for the prevention and treatment of unhealthy lifestyle habits. Stockholm: The Swedish National Board of Health and Welfare (2021). 21099.. *National guidelines for the prevention and treatment of unhealthy lifestyle habits* (2021.0) 21099
31. Delisle Nyström C, Larsson C, Alexandrou C, Ehrenblad B, Eriksson U, Friberg M. **Results from Sweden's 2018 Report card on physical activity for children and youth**. *J Phys Act Health* (2018.0) **15** S413-s4. DOI: 10.1123/jpah.2018-0519
32. 32.The National Board of Health and Welfare, Sweden. National guidelines for obesity care; support for governance and management. Stockholm: The national board of Health and Welfare (2022). 2022-4-7822.. *National guidelines for obesity care; support for governance and management* (2022.0)
33. Bohman DM, Mattsson L, Borglin G. **Primary healthcare nurses’ experiences of physical activity referrals: an interview study**. *Prim Health Care Res Dev* (2015.0) **16** 270-80. DOI: 10.1017/S1463423614000267
34. Curbach J, Apfelbacher C, Knoll A, Herrmann S, Szagun B, Loss J. **Physicians’ perspectives on implementing the prevention scheme “physical activity on prescription": results of a survey in bavaria**. *Z Evid Fortbild Qual Gesundhwes* (2018.0) **131–132** 66-72. DOI: 10.1016/j.zefq.2018.02.001
35. Gustavsson C, Nordqvist M, Broms K, Jerden L, Kallings LV, Wallin L. **What is required to facilitate implementation of Swedish physical activity on prescription?—interview study with primary healthcare staff and management**. *BMC Health Serv Res* (2018.0) **18** 196. DOI: 10.1186/s12913-018-3021-1
36. Persson G, Brorsson A, Ekvall Hansson E, Troein M, Strandberg EL. **Physical activity on prescription (PAP) from the general practitioner's Perspective—a qualitative study**. *BMC Fam Pract* (2013.0) **14** 128. DOI: 10.1186/1471-2296-14-128
37. 37.The Public Health Authority. PAP In Sweden—a description of the regional work with physical activity on prescription (PAP) as a method of treatment. Sweden: The Public Health Authority (2022). 21232.. *PAP In Sweden—a description of the regional work with physical activity on prescription (PAP) as a method of treatment* (2022.0) 21232
38. Boman C, Bernhardsson S. **Exploring needs, barriers, and facilitators for promoting physical activity for children with intellectual developmental disorders: a qualitative focus group study**. *J Intellect Disabil* (2022.0) 17446295211064368. DOI: 10.1177/17446295211064368
39. May C, Finch T. **Implementing, embedding, and integration practices: an outline of normalization process theory**. *Sociology* (2009.0) **43** 535-54. DOI: 10.1177/0038038509103208
40. Nilsen P. **Making sense of implementation theories, models and frameworks**. *Implement Sci* (2015.0) **10** 53. DOI: 10.1186/s13012-015-0242-0
41. Rycroft-Malone J, Bucknall T. **Using theory and frameworks to facilitate the implementation of evidence into practice**. *Worldviews Evid Based Nurs* (2010.0) **7** 57-8. DOI: 10.1111/j.1741-6787.2010.00194.x
42. Rapley T, Girling M, Mair FS, Murray E, Treweek S, McColl E. **Improving the normalization of complex interventions: part 1—development of the NoMAD instrument for assessing implementation work based on normalization process theory (NPT)**. *BMC Med Res Methodol* (2018.0) **18** 133. DOI: 10.1186/s12874-018-0590-y
43. May C. **Towards a general theory of implementation**. *Implement Sci* (2013.0) **8** 18. DOI: 10.1186/1748-5908-8-18
44. Proctor E, Silmere H, Raghavan R, Hovmand P, Aarons G, Bunger A. **Outcomes for implementation research: conceptual distinctions, measurement challenges, and research agenda**. *Adm Policy Ment Health* (2011.0) **38** 65-76. DOI: 10.1007/s10488-010-0319-7
45. Egger M, von Elm E, Altman DG, Pocock SJ, Gøtzsche PC, Vandenbroucke JP. **The strengthening the reporting of observational studies in epidemiology (STROBE) statement: guidelines for reporting observational studies**. *J Clin Epidemiol* (2008.0) **61** 344e9. DOI: 10.1016/j.jclinepi.2007.11.008
46. Finch TL, Girling M, May CR, Mair FS, Murray E, Treweek S. **Improving the normalization of complex interventions: part 2—validation of the NoMAD instrument for assessing implementation work based on normalization process theory (NPT)**. *BMC Med Res Methodol* (2018.0) **18** 135. DOI: 10.1186/s12874-018-0591-x
47. Elf M, Nordmark S, Lyhagen J, Lindberg I, Finch T, Aberg AC. **The Swedish version of the normalization process theory measure S-NoMAD: translation, adaptation, and pilot testing**. *Implement Sci* (2018.0) **13** 146. DOI: 10.1186/s13012-018-0835-5
48. Weiner BJ, Lewis CC, Stanick C, Powell BJ, Dorsey CN, Clary AS. **Psychometric assessment of three newly developed implementation outcome measures**. *Implement Sci* (2017.0) **12** 108. DOI: 10.1186/s13012-017-0635-3
49. Bernhardsson S, Boman C, Lundqvist S, Arvidsson D, Borjesson M, Larsson MEH. **Implementation of physical activity on prescription for children with obesity in paediatric health care (IMPA): protocol for a feasibility and evaluation study using quantitative and qualitative methods**. *Pilot Feasibility Stud* (2022.0) **8** 117. DOI: 10.1186/s40814-022-01075-3
50. Andersen P, Holmberg S, Årestedt K, Lendahls L, Nilsen P. **Factors associated with increased physical activity among patients prescribed physical activity in Swedish routine health care including an offer of counselor support: a 1-year follow-up**. *BMC Public Health* (2022.0) **22** 509. DOI: 10.1186/s12889-022-12940-4
51. Lundqvist S, Cider A, Larsson MEH, Hagberg L, Bjork MP, Borjesson M. **The effects of a 5-year physical activity on prescription (PAP) intervention in patients with metabolic risk factors**. *PLoS One* (2022.0) **17** e0276868. DOI: 10.1371/journal.pone.0276868
52. Olstad DL, Teychenne M, Minaker LM, Taber DR, Raine KD, Nykiforuk CI. **Can policy ameliorate socioeconomic inequities in obesity and obesity-related behaviours? A systematic review of the impact of universal policies on adults and children**. *Obes Rev* (2016.0) **17** 1198-217. DOI: 10.1111/obr.12457
53. Joelsson M, Bernhardsson S, Larsson ME. **Patients with chronic pain may need extra support when prescribed physical activity in primary care: a qualitative study**. *Scand J Prim Health Care* (2017.0) **35** 64-74. DOI: 10.1080/02813432.2017.1288815
54. Albert FA, Crowe MJ, Malau-Aduli AEO, Malau-Aduli BS. **Functionality of physical activity referral schemes (PARS): a systematic review**. *Front Public Health* (2020.0) **8** 257. DOI: 10.3389/fpubh.2020.00257
55. Ray D, Sniehotta F, McColl E, Ells L. **Barriers and facilitators to implementing practices for prevention of childhood obesity in primary care: a mixed methods systematic review**. *Obes Rev* (2022.0) **23** e13417. DOI: 10.1111/obr.13417
56. 56.The Public Health Authority. Guidelines for physical activity and sedentary life: Knowledge support for the promotion of physical activity and reduced sedentary behavior. Sweden: The Public Health Authority (2021). 21099.. *Guidelines for physical activity and sedentary life: Knowledge support for the promotion of physical activity and reduced sedentary behavior* (2021.0) 21099
|
---
title: 'Factors contributing to the sustained implementation of an early childhood
obesity prevention intervention: The INFANT Program'
authors:
- Penelope Love
- Rachel Laws
- Sarah Taki
- Madeline West
- Kylie D. Hesketh
- Karen J. Campbell
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012774
doi: 10.3389/frhs.2022.1031628
license: CC BY 4.0
---
# Factors contributing to the sustained implementation of an early childhood obesity prevention intervention: The INFANT Program
## Abstract
### Background
The INFANT *Program is* an efficacious, group-based program for first-time parents, delivered at three-monthly intervals when INFANT are aged 3–18 months through an existing universal care service in Victoria, Australia. Many lessons have been learnt from its origins as a cluster randomized control trial to its small-scale, community-level implementation. This study aimed to describe factors contributing to its sustained implementation to inform large-scale implementation across Australia.
### Methods
This study used a multi-site qualitative exploratory approach. INFANT facilitators trained between 2013 and 2017 were sent an online survey, with optional telephone interviews. The Consolidated Framework for Implementation Research (CFIR) was selected as the underpinning theoretical framework as it offered the opportunity to explore a breadth of possible barriers and enablers across patterns of implementation (never, discontinued, ongoing).
### Results
All participants were female ($$n = 31$$), the majority were Maternal and Child Health Nurses ($48\%$), representing five regional and nine metro local government areas (LGAs), across all patterns of implementation (never implemented $$n = 4$$; discontinued implementation $$n = 5$$; ongoing implementation $$n = 5$$). All consenting participants were interviewed ($$n = 11$$) representing four regional and seven metro LGAs, across all patterns of implementation (never implemented $$n = 3$$; discontinued implementation $$n = 4$$; ongoing implementation $$n = 4$$). The main reason for attending INFANT Program training was to become skilled to implement the program. Mapping identified barriers and enablers to the CFIR revealed the inner and outer settings and implementation process to be of greatest influence. Main differences between LGAs with ongoing and discontinued implementation related to funding availability, organizational management support and endorsement, organizational resourcing and capacity, integration into routine practice and establishing role clarity with partner organizations, and planning for sustained implementation from the start.
### Conclusion
This study provides important insights into the barriers and enablers to the sustained implementation of an evidence-based intervention (the INFANT Program) during small scale community-level implementation. The authors therefore contend that the pre-requisite for scale-up of a population health intervention is not just proof of effectiveness but also proof of sustained implementation at the local/organizational level. Study findings have broad transferability given their similarity to those identified for health promotion interventions implemented globally, in healthcare, education and community settings.
## Introduction
The first 1,000 days (conception to 24 months) are acknowledged as a crucial period for growth and development in early childhood, laying the foundation for life-long health behaviors and the prevention of chronic disease [1, 2]. The early establishment of healthy behaviors [3], such as prolonged breastfeeding [4], reduced consumption of energy-dense, nutrient-poor foods/beverages [5], limited screen time and sedentary behavior [6], and prevention of rapid infant weight gain [7], is considered critical for the prevention of childhood obesity and overweight which affects an estimated 38.3 million children under the age of 5 years globally [8]. In Australia, $25\%$ of children aged 2–4 years are already experiencing overweight/obesity [9], with a minority meeting the recommended dietary and movement guidelines [10], and those living in lower socioeconomic or regional areas most affected [11]. It is predicted that if current rates of childhood weight gain continue, prevalence of these conditions among Australian children will reach $33\%$ by 2025 [12].
Research indicates that early intervention at or within a few months of birth can benefit obesity prevention in the first 1,000 days (2, 13–15). The World Health Organization's Commission on Ending Childhood Obesity [16] describes a continuum of care for the prevention, management and treatment of obesity among infants and children using a multi-strategy approach targeting the individual, family, community and public policy. Recent reviews support this approach, suggesting the use of interventions that include multicomponent (healthy eating, sleep, sedentary or screen-time, and physical activity or active play) guidance and support [17], and targeting system-level determinants of a child's diet and movement behaviors, such as caregiver behaviors, household and external environments, and food supply chains [18]. The main influence of health behaviors in the early years is the family and home environment [19], therefore family-focused health services are well placed to provide this multicomponent support. In Victoria, Australia, this opportunity is available through the universal free Maternal and Child Health (MCH) service which provides 10 consultations between birth and age 3.5 years, with an uptake of 83–$97\%$ in the first 12 months [20].
While still an understudied area, expert consensus is emerging regarding the conceptualization of sustained implementation, especially clarifying the definition, developing an underpinning framework, and advancing measurement/assessment criteria (21–30). Sustained implementation is considered to have occurred when, “after a defined period of time, the program/intervention/strategies continue to be delivered and/or individual behavior change (i.e., clinician, patient) is maintained, either as originally planned or with some degree of adaptation, while continuing to produce benefits for individuals/systems” [27, 31]. Sustained implementation, originally described as “institutionalization” [32] and more recently as “routinization” [33], “maintenance” [30] and “continuation beyond financial security” [28] is less frequently investigated in comparison to adoption and initial implementation, often due to budgetary and timeframe constraints [34].
Barker et al. [ 35] propose sustained implementation be a consideration during program development, small-scale replication, and real-world “at scale” implementation. This proposition appears justified given the number of programs/interventions/strategies implemented “at scale” that fail to be sustained long-term. In one of the earliest publications to examine the sustained implementation of public health programs, Scheirer [36] reported 40–$60\%$ were being implemented to some extent 1–6 years post program adoption. A multi-stage international literature search by Indig et al. [ 37] identified 40 public health interventions in high income countries (USA, Australia, Netherlands, Canada, UK, New Zealand, Finland) that showed reliable evidence of being implemented “at scale” between 1990 and 2014, of which $80\%$ were still being implemented “to some extent” largely through institutionalization ($55\%$) or commercialization ($20\%$). A recent realist review [38] of nutrition and/or physical activity interventions implemented “at scale” (at a State or National level) within Australia since 2010, found four of the identified seven interventions ($57\%$) were still being implemented 8 years post program adoption (one national and three in New South Wales).
As highlighted by Glasgow et al. [ 39] more than 20 years ago, numerous evaluated interventions are “lost in translation” because implementation is not sustained in real-world settings. Further, the “Determinants of Diet and Physical Activity” (DEDIPAC) Knowledge Hub study [40], informed by two umbrella reviews, reported a lack of research providing detail of implementation processes from the perspective of the health professional, practitioner or policy-maker, especially after completion of research projects. Understanding what factors impact sustained implementation is therefore essential to inform the, often significant, investments made by public health and government entities in developing and implementing programs “at scale” in real-world settings [41]. While there is a sense of urgency to implement programs “at scale” in order to maximize their reach [42], it would appear that selection of programs is often based on availability and opportunity rather than proven efficacy or ability for sustained implementation [43].
The present study explored barriers and enablers influencing sustained implementation of the INFANT Program following the cessation of the state-wide prevention initiative, Healthy Together Victoria. Sustained implementation was defined as delivery of the INFANT Program (six three-monthly program sessions with first time parents of infants aged 3–18 months using a group-based format) between 2016 and 2017. Perspectives were obtained from trained INFANT facilitators, providing important insights into implementation processes, barriers and enablers experienced by health practitioners tasked with program implementation in “real-world” settings. Ethical approval for this study was obtained through Deakin University (HEAG-H 183_2014).
## Program context
The INFANT *Program is* believed to be the first of its kind to address obesity risk behaviors in the first 1,000 days of life using a universally delivered service. Delivered in Australia, this is an efficacious, group-based program for first-time parents, comprising six 1.5-hour sessions delivered at three-monthly intervals when their infant is aged approximately 3, 6, 9, 12, 15 and 18 months [44] with positive health outcomes evident for mother and infant [45]. The evolution of the INFANT Program from randomized controlled trial to small-scale community-level implementation [46] and the varying models of program implementation used [47] have been reported elsewhere. In 2014, the INFANT Program was included as a strategy within the state-wide prevention initiative, Healthy Together Victoria (HTV) [48]. HTV operated across Victoria (2011–2016) to deliver a package of programs and strategies using a systems approach, with specific health promotion workforce funding and support provided to 14 local government areas (LGAs),1 based on socio-demographic indices and chronic disease risk factor prevalence. Due to national governance changes, funding ceased in 2015 (ahead of its scheduled 2018 end date) with the resultant cessation of HTV. Despite this early withdrawal of funding, a few LGAs continued to implement some of their activities, including the delivery of the INFANT Program. This provided an opportunity to investigate factors influencing the uptake and sustained implementation of the INFANT Program by LGAs, especially those previously receiving HTV funding.
## Study design
This study used a multi-site qualitative exploratory approach to facilitate an in-depth understanding of barriers and enablers to the sustained implementation of the INFANT Program within Victoria, Australia [49]. This was considered a pragmatic and appropriate approach given the intent was to explore constructs to inform future examinations of the area. The researchers followed the *Consolidated criteria* for reporting qualitative studies (COREQ) checklist [50].
## Theoretical framework
The Consolidated Framework for Implementation Research (CFIR) [51] was selected as the underpinning theoretical framework as it offered the opportunity to explore a breadth of possible barriers and enablers across patterns of implementation (never, discontinued, ongoing). The CFIR comprises 37 constructs across 5 domains, each considered important for the adoption, implementation and embedding of interventions into routine practice [51] (Table 1). At the time of this study the CFIR was considered the most contemporary model available, underpinned by implementation research with practical application across diverse settings. Since this study was concluded, specific sustainability models have emerged, such as Integrated Sustainability Framework (ISF) [29]. The use of the CFIR model to reflect elements of sustainability is however still considered relevant given the strong alignment between the constructs of the CFIR and ISF. The CFIR Guide Tool (CFIR Booklet (cfirguide.org) was used to develop survey and interview questions. ( Table 1, Supplementary material 1). While the CFIR can be applied using a quantitative approach [52, 53], this study applied a qualitative approach as commonly used by others [54, 55].
**Table 1**
| Domain | Construct | Example questions used for surveys (S) /interviews (I) (see Supplementary material 1 for detail) |
| --- | --- | --- |
| Intervention characteristics | • Intervention source development and implementation decision-making process • Strength and quality of evidence to support choice of intervention • Relative advantage of implementing intervention versus an alternative • Adaptability of intervention to meet local needs • Trialability of intervention prior to implementation • Complexity and difficulty of implementation • Design quality and packaging of intervention • Costs associated with implementation | • What were the reasons why you attended the facilitator training? (S; I) • What did you know about the INFANT Program (if anything) before you attended the facilitator training? (I) • In your opinion, why do you think it was decided that the INFANT Program should/not be implemented in your area? (S; I) |
| Outer setting | • Patient needs and resources met in relation to implementation barriers/enablers • Cosmopolitanism (organization networks with other external organizations) • Peer pressure to implement intervention • External policy and incentives (mandates, strategies) to spread intervention uptake | • In your opinion, why do you think it was decided that the INFANT Program should/not be implemented in your area? (S; I) • What factors do you think helped /hindered the implementation of the INFANT Program in your area? (S; I) • How was the decision made that the program would/not be implemented in your organization? (I) |
| Inner setting | • Structural characteristics of the organization, such as maturity, age and size • Networks and communications (informal or formal) within organization • Culture, norms, values and basic assumptions of the organization • Implementation climate (receptivity, compatibility, relative priority, incentives) • Readiness for implementation (leadership engagement and commitment, available resources, access to knowledge, information incorporated into work tasks) | • Was the decision influenced by any other organizations implementing the INFANT Program, and if so, how? (I) • How does the INFANT Program fit within existing services within your organization? (I) |
| Implementer characteristics | • Knowledge and beliefs about the intervention and value placed on intervention • Self-efficacy/belief in own capabilities to implement intervention to achieve goals • Individual stage of change (level of preparedness to implement intervention) • Individual identification with the organization (commitment to organization) • Other personal attributes (learning styles, capacity, competency, motivation, etc.) | • What did you know about the INFANT Program (if anything) before you attended the facilitator training? (I) • How well did the training prepare you to implement the INFANT Program in your area? (I) |
| Implementation process | • Planning processes for implementation • Engagement strategies (with opinion leaders, champions, key stakeholders) • Executing according to implementation plan • Reflecting and evaluating (qualitative and quantitative feedback on progress) | • How was the INFANT Program planned and implemented in your area? (I) • How have you gone about evaluating the INFANT Program in your organization? (I) |
## Data instrumentation
Open-ended questions within the surveys were used to explore barriers and enablers to the sustained implementation of the INFANT Program following facilitator training, with follow-up interviews to explore findings in more depth (Supplementary material 1). Participants completed a 15-minute online survey regarding their perspectives of the INFANT Program training, reasons for attending training, intentions of program delivery after training, and tailored questions depending on the pattern of program implementation (never, discontinued, ongoing). The survey was structured according to pattern of implementation, with tailored questions framed by the CFIR domains [51] to identify enablers and barriers to ongoing (sustained) implementation. Questions comprised open-ended and 7-point Likert scale (completely disagree-completely agree) responses. Follow-up 30–45 min audio-recorded telephone interviews were conducted with consenting survey participants to explore survey responses further. Interview questions asked participants to reflect on organizational decision-making about the planning process, resourcing and support for the implementation of the INFANT Program after completing the face-to-face training.
## Data collection
All Victorian-based staff who had completed the INFANT Program facilitator training between 2013 and 2017 ($$n = 88$$) were contacted, using email contact details provided during training registration. Those contacted were invited to complete an online survey and an optional telephone interview. Those consenting to an interview were contacted directly by PL to schedule a convenient date and time for the interview. All interviews were conducted by PL using a semi-structured interview guide, ranging in duration from 21–47 min. Audio-recorded interviews were transcribed verbatim by an external agency. No incentives were offered to participate in the study.
Of the 88 Victorian-based INFANT Program trainees, two were not contactable, four were on leave and 16 had moved to other positions, resulting in a final sample size of 63 participants, representing 16 LGAs (six regional and 10 metro) at various stages of implementation (never implemented $$n = 6$$; discontinued implementation $$n = 5$$; ongoing implementation $$n = 5$$). Thirty-one participants completed the online survey, with 11 consenting to follow-up interviews, representing 14 LGAs across all patterns of implementation (never implemented $$n = 4$$; discontinued implementation $$n = 5$$; ongoing implementation $$n = 5$$).
## Data analysis
Qualitative data analysis was underpinned by a contextualist epistemology, where knowledge emerges from and is situated within the context of the data [56]. As the interpretation of qualitative data can be influenced by the roles and backgrounds of the researchers, these are made explicit. All researchers have a health qualification and work within a research context. At the time of the study MW was a research assistant with nutrition experience, and PL, RL, and ST were postdoctoral researchers with experience in the implementation of public health nutrition interventions at a community level. MW, ST, PL, and RL had no involvement in the development of the INFANT Program. RL had specific involvement in evaluating the small-scale community implementation of the INFANT Program. KDH and KJC are chief investigators of the INFANT Program, responsible for its development, randomized control trial, small-scale community implementation, and ongoing evaluation.
A reflexive thematic analysis approach, as described by Clarke et al. [ 57], was undertaken using open-ended survey responses and interview transcripts to determine shared meaning underpinned by the CFIR domains [51]. Data were coded deductively (informed by the CFIR framework) and inductively (to identify other codes) using NVIVO v12 (QSR International, Melbourne, Australia [58]. A sub-sample of interviews was coded independently by three co-authors (PL, ST, and MW), followed by discussion regarding interpretation and application of the coding framework. All coding was completed by MW. NVIVO coding summaries were used for case comparison analysis to identify similarities and differences between barriers and enablers for different patterns of implementation across the LGAs, namely, never, discontinued, and ongoing (sustained) implementation. Consensus on final theming was developed in agreement between PL, RL, KDH, and KJC. As an exploratory study with a small sample size, data saturation was not a consideration.
## Description of participants
Thirty-one participants completed the online survey, with 11 consenting to follow-up interviews. All participants were female, mainly between the ages of 40–59 years ($71\%$). Most participants were Maternal and Child Health Nurses ($48\%$), followed by dietitians ($2.5\%$), and in part-time roles ($68\%$). Across all LGAs, the main reason for attending INFANT Program training was to become skilled to implement the program. The majority of participants attended training to take on the role of program facilitator ($88\%$), and “mostly” and “completely” agreed that training provided the necessary knowledge ($81\%$) and confidence ($74\%$) to implement the INFANT Program (Table 2).
**Table 2**
| Study participant descriptor | Unnamed: 1 | Pattern of implementation (n = 31 survey responses) | Pattern of implementation (n = 31 survey responses).1 | Pattern of implementation (n = 31 survey responses).2 | Pattern of implementation (n = 31 survey responses).3 |
| --- | --- | --- | --- | --- | --- |
| | | Total | Ongoing | Discontinued | Never |
| Gender | Female | 31 | 15 | 11 | 5 |
| Age | 20–29 years | 3 | 3 | 0 | 0 |
| | 30–39 years | 4 | 3 | 1 | 0 |
| | 40–49 years | 12 | 5 | 6 | 1 |
| | 50–59 years | 10 | 4 | 3 | 3 |
| | 60+ years | 2 | 0 | 1 | 1 |
| Profession | Maternal child health nurse | 15 | 8 | 5 | 2 |
| | Dietitian | 8 | 6 | 0 | 2 |
| | Health promotion officer | 2 | 1 | 1 | 0 |
| | Early childhood professional | 2 | 0 | 1 | 1 |
| | Social worker | 1 | 0 | 1 | 0 |
| | Early intervention worker | 1 | 0 | 1 | 0 |
| | Children and family resource officer | 1 | 0 | 1 | 0 |
| | Bicultural families and children officer | 1 | 0 | 1 | 0 |
| Full/Part-time | Full time | 10 | 5 | 3 | 2 |
| | Part-time | 21 | 10 | 8 | 3 |
| Years in role | < 5 years | 7 | 5 | 2 | 0 |
| | 5–10 years | 9 | 6 | 3 | 0 |
| | 11–15 years | 4 | 2 | 1 | 1 |
| | >15 years | 11 | 2 | 5 | 4 |
| Reason for attending training (multiple options) | Intention to deliver | 16 | 7 | 8 | 1 |
| | Gain additional knowledge | 11 | 6 | 3 | 2 |
| | Learn about the program | 19 | 10 | 4 | 5 |
| | Personal professional development | 10 | 5 | 3 | 2 |
| | Organization already delivering | 10 | 8 | 2 | 0 |
| Program LGAs | HTV-funded (7 LGAs) | 21 | 11 (2 LGAs) | 9 (4 LGAs) | 1 (1 LGA) |
| | Non-HTV funded (7 LGAs) | 10 | 4 (3 LGAs) | 2 (1 LGA) | 4 (3 LGAs) |
## Patterns of implementation
Online survey participants represented 14 LGAs across all patterns of implementation (never implemented $$n = 4$$; discontinued implementation $$n = 5$$; ongoing implementation $$n = 5$$). Of these LGAs, 11 participants consented to interviews representative of all patterns of implementation (never implemented $$n = 3$$; discontinued implementation $$n = 4$$; ongoing implementation $$n = 4$$). All patterns of implementation were evident across regional and metro LGA locations. Regional LGAs ($$n = 5$$) reported $$n = 1$$ as never implemented; $$n = 2$$ with discontinued implementation; and $$n = 3$$ with ongoing implementation. Metro LGAs ($$n = 9$$) reported $$n = 3$$ as never implemented; $$n = 3$$ with discontinued implementation; and $$n = 2$$ with ongoing implementation. Of the 14 LGAs, seven (3 regional; 4 metro) had received specific health promotion workforce funding through the HTV initiative (never implemented $$n = 1$$; discontinued implementation $$n = 4$$; ongoing implementation $$n = 2$$), and seven were non-HTV funded (never implemented $$n = 3$$; discontinued implementation $$n = 1$$; ongoing implementation $$n = 3$$) (Table 2).
## Barriers and enablers to sustained implementation
Mapping identified barriers and enablers to the CFIR (Supplementary material 2) revealed the inner and outer settings and implementation process to be of greatest influence.
## Inner setting
Organizational implementation climate and readiness for implementation were most frequently described by participants. LGAs that had never implemented the INFANT Program felt that the “timing was not right”, with a lack of agreement between organizations regarding the implementation approach. These LGAs also reflected on limited leadership engagement and the lack of a program “champion”. A lack of management support was the main barrier cited by the HTV-funded LGA that had never implemented the INFANT Program whilst a lack of funding and availability of staff to coordinate and deliver the program were main barriers cited by non-HTV funded LGAs with no implementation—“I'm sure it can be done it was just too hard for us without resources at our disposal” [Never implemented, metro LGA].
LGAs with discontinued or ongoing implementation felt the INFANT Program was highly compatible with existing services and a priority. LGAs with discontinued implementation reflected that the program was competing with other priorities, and in some cases, other programs. The main barrier cited by all four HTV-funded LGAs that had discontinued implementation was the cessation of funding—“(HTV) funding ceased, and management deemed it [the INFANT Program] was no longer needed” [Discontinued implementation, metro LGA]. The non-HTV funded LGA that had discontinued implementation cited a lack of management support and poor program attendance as the main barriers.
Only LGAs with ongoing implementation described consideration of sustained implementation at the start—“We made the decision at the start that it [INFANT Program implementation] was going to keep going beyond the funding time… we needed to embed it into services that we have already” [Ongoing implementation, regional LGA]. LGAs with ongoing implementation also mentioned the importance of establishing organizational connections prior to undertaking the training to achieve early buy-in. For both HTV-funded and non-HTV funded LGAs, management support was cited as the main enabler to implementation.
Available implementation capacity and resources was described as a limiting factor by LGAs across all patterns of implementation, especially when attendance rates were low, and even if program delivery was incorporated into staff roles as “once (HTV) funding stopped, the positions stopped” [Discontinued implementation, regional LGA]. LGAs with ongoing implementation described how staff capacity had been created through the allocation of health promotion hours within existing staff roles, clarifying role responsibilities between partner organizations (such as referrals by maternal child health, and scheduling by community health), and establishing designated administration support to streamline enrolment, reminder notifications, and securing venues. LGAs with discontinued or ongoing implementation described strong organizational engagement, especially between dietetic and maternal child health services.
## Outer setting
Across all patterns of implementation, LGAs described the INFANT Program as meeting a community need, complementing and strengthening the universal Maternal and Child Health (MCH) Service offered across Victoria. LGAs that had never implemented suggested that the program be promoted more as “people haven't any idea of what it is or the benefits” [Never implemented, metro LGA] and commented on the need to consider more contemporary approaches to program delivery in line with current technology -“introducing the electronic form of it… because most people have smartphones” [Never implemented, metro LGA].
All LGAs expressed a desire to be connected with local organizations to assist with program recruitment, implementation and to provide “positive feedback from another organization already running the program” [Ongoing implementation, regional LGA]. Access to the INFANT Program research team for implementation guidance was a valued support by LGAs with discontinued or ongoing implementation.
LGAs with discontinued or ongoing implementation suggested better alignment between funding and policy directives, with recurrent funding, resourcing and monitoring to enable sustained implementation—“It would be lovely to just be able to do it in a fully funded, dedicated way… through state or federal funding… in the same way that other services are provided then you can dedicate staff to it” [Discontinued implementation, regional LGA].
Across all LGAs, two main models of program implementation were apparent, one led by the MCH team (based within local government) and the other a partnership between the MCH team and dietitians (based within community health). All LGAs with ongoing implementation had a partnership model in place.
## Implementation process
While LGAs across all patterns of implementation described the INFANT Program as aligning to existing services and having the potential to replace ad hoc group information sessions, only LGAs with ongoing implementation spoke about integration of the program into service provision. Examples included delivery of the first INFANT Program session as part of existing New Parent Groups, enrolling participants into all sessions with automated reminder notifications and opt-out consent (rather than individual session enrolment) and offering “open” groups so participants could attend any missed sessions. LGAs with discontinued and ongoing implementation both made adaptations to program delivery, predominantly delivering four of the six sessions (3, 6, 9, and 12 months) given the high attrition rates at the 15 and 18 month sessions. LGAs all described undertaking some form of program evaluation, expressing concerns about unrealistic targets, what data to collect, and participant burden.
The importance of engagement and involvement of key partner organizations and stakeholders was evident across all patterns of implementation. LGAs that had never implemented the INFANT Program echoed their feedback regarding a lack of consensus by partner organizations about the appropriate implementation approach, with no opinion leaders or program champions. LGAs with discontinued implementation spoke of the need for a designated implementation team so that implementation was not in addition to existing workloads—“… to do it properly… get all the admin done… all that really needs a designated team. We were a bit caught between what we were already doing…” [Discontinued implementation, regional LGA]. LGAs with ongoing implementation described the partnership between dietetic (community health) and maternal child health (local council) as ideal for the implementation of the INFANT Program, but that this required a shared understanding and clarity regarding implementation roles and responsibilities. Engagement with “external change agents” was suggested across all patterns of implementation and included the promotion and/or extension of the INFANT Program into childcare centers, playgroups, and ante-natal groups.
## Intervention characteristics
Across all patterns of implementation, LGAs were aware of the INFANT Program prior to attending training, through professional conferences, colleagues and management, and as an endorsed HTV program. All LGAs considered the program to be evidence-based with strong research outcomes, offering a relative advantage to the organization in terms of alignment with current MCH services, and providing consistency of information to parents. LGAs with ongoing implementation regarded the program as “value-adding” by providing a more structured approach, replacing ad hoc group information sessions “to allow time to deliver INFANT Program which covers these topics plus more” [Ongoing implementation, metro LGA]. LGAs with discontinued or ongoing implementation described similar complexities in relation to scheduling of sessions with similar aged infants, and venue availability and costs. Regional LGAs in particular were challenged by small birth rates and large geographical distances which limited attendance rates and group size. Costs associated with the INFANT Program were described in terms of implementation capacity, and not in relation to accessing training or program resources.
All LGAs described the need to consider flexibility with implementation of the INFANT Program to meet community needs, such as providing more visual images or using an interpreter for different cultural groups, tailoring delivery for groups with mixed age groups, and most commonly, providing fewer sessions given the low attendance rates at the 15 and 18 month sessions. One non-HTV funded LGA that had never implemented the program was concerned about how much flexibility and adaptation could be applied before impacting on program fidelity—“[I'm} concerned that adapting the program by combining sessions or offering them in other formats does not have the evidence base” [Never implemented, metro LGA].
LGAs across all patterns of implementation considered the training and website resources to be of high quality. LGAs with ongoing implementation described the training as enhancing group facilitation skills. LGAs with discontinued implementation felt the training had reinforced existing knowledge, enhancing levels of confidence to deliver program sessions. LGAs that had never implemented expressed a need for specific implementation guidance and examples of how LGAs had implemented the program, especially where this had occurred without additional funding.
## Implementer characteristics
Across all patterns of implementation, study participants considered themselves to possess the appropriate knowledge and beliefs to implement the INFANT Program, describing the delivery of infant feeding and active play information to parents as “our bread and butter” and “part of our core work” [Ongoing implementation, metro LGA]. LGAs with both discontinued and ongoing implementation described the training as enhancing levels of confidence to present the content and facilitate the group discussion in a different way, with “a greater focus on active listening” [Ongoing implementation, regional LGA].
## Discussion
This study explored barriers and enablers to sustained implementation of an early childhood health behavior program for parents, the INFANT Program, during small scale implementation in Victoria, Australia, from the perspective of trained INFANT facilitators. Challenges regarding complexities of program implementation were apparent across all patterns of implementation, with requests for specific implementation guidance and connections with other LGAs achieving successful implementation. The main differences between LGAs with ongoing and discontinued implementation related to the “inner and outer setting” and “implementation process”, specifically, funding availability, organizational resourcing and capacity, organizational management support and endorsement, integrating implementation into routine practice, establishing early buy-in and role clarity with partner organizations, and planning for sustained implementation from the start.
The enablers and barriers identified in this study are similar to those reported in the literature and can therefore be considered to have relevance to other health promotion interventions. Muellmann et al. [ 40] describe five main enablers, relevant to multi-level interventions and policies promoting healthy eating and physical activity, namely, stakeholder networks, structures in settings, continued funding and political support, standardized training of staff with detailed implementation protocols, and socio-cultural tailoring of content to fit the needs and context of the targeted population. In addition to these enablers, Mikkeslen et al. [ 59] report the need for capacity building of health professionals across health, education and community settings, including pre-service and in-service training, so that implementation activities continue after any research support concludes. Similarly, systematic reviews of health promotion [28], community-based obesity prevention [23], healthcare [22], schools and childcare services [60] and public health [61] interventions highlight the importance of several recurring enablers, namely: strategic planning, program alignment, integration into existing programs and policies, accessing new/existing money to facilitate sustainment, leadership prioritization and support to mobilize implementation, adequate human resourcing, workforce development and capacity building regarding implementation planning and evaluation, systematic adaptation to enhance compatibility of the intervention with the organization, monitoring progress and demonstrating effectiveness, and establishing organizational partnerships.
The provision of external funding through the HTV initiative was a key catalyst for INFANT Program implementation, however four of the seven HTV-funded LGAs discontinued implementation once HTV funding ceased. LGAs with ongoing implementation of the INFANT Program utilized strategies that were not reliant on external funding support, in particular, creating staff capacity through the allocation of health promotion hours within existing staff roles, and establishing a partnership model for implementation between community health dietetic and maternal child health services. Cross-disciplinary and cross-organizational partnerships, with a shared agenda, can frequently add tangible resources to the implementation process [62]. Investment in organizational capacity and infrastructure creates a foundation for the intervention activities to continue if/when external resources, such as a research team or government agency, are discontinued [59].
For both HTV-funded and non-HTV funded LGAs, management support was cited as the main enabler to ongoing implementation. The role of leaders and transformational leadership in supporting sustained implementation is well documented [23, 52, 63, 64] in the form of policy and reward systems, organizational decision-makers, and community champions. Effective leaders can mobilize capacity and collaboration, frequently overcoming organizational indifference or opposition to a new intervention. Leaders are also instrumental in generating program awareness and securing ongoing investment. The uptake of the INFANT Program by HTV-funded LGAs is indicative of the leadership endorsement of the program as part of the HTV initiative. With the cessation of the HTV initiative, the loss of this endorsement and the removal of funding resulted in many HTV-funded LGAs discontinuing their implementation of the INFANT Program.
Early consideration of sustained implementation was identified as a common strategy by LGAs with ongoing implementation of the INFANT Program. This view complements the growing consensus that sustained implementation should be considered from the beginning of the implementation process, with dedicated planning to define program components and determinants to inform appropriate implementation strategies [28]. An integral part of this early planning phase includes dedicated exploration of an organization's readiness in terms of commitment and capability for implementation, which has now been incorporated as a consideration during INFANT facilitator training. When organizational readiness for change is high, organizations display greater initiation, persistence and cooperation to achieve successful implementation [53, 65]. In a recently updated systematic review, Miake-Lye et al. [ 66] mapped organizational readiness assessment instruments to the CFIR, and identified “readiness to implementation” as the most commonly reported construct.
The challenge of fidelity and adaptation was identified as a barrier by LGAs that had never implemented the INFANT Program post facilitator training. These LGAs described being unsure what degree of adaptation would be possible without impacting program fidelity and were seeking specific implementation guidance. The assumption that intervention effects lessen if implemented “at scale” without careful adherence to research protocols has been challenged by Chambers et al. [ 34] who suggest that this constrains the intervention “fit” (compatibility) within the given context and positions sustained implementation as “the endgame”. They propose that sustained implementation be a consideration throughout the implementation process to accommodate adaptation so that the intervention becomes integrated into the local context [34]. To facilitate a more precise understanding of adaptations made to the INFANT Program when implemented in real-world settings, comprehensive documentation using the Framework for Reporting Adaptations and Modifications-Enhanced (FRAME) [67] has subsequently been incorporated into the INFANT Effectiveness-Implementation Trial [68] to inform the timing, context and process for adaptation to facilitate sustained implementation. Capturing intervention adaptation is a key inclusion to establish the degree to which intervention components are modified for organizational compatibility without jeopardizing intervention outcomes.
De-implementation strategies are likely to become an important consideration for the sustained implementation of the INFANT Program, as LGAs all commented about competing organizational priorities and the need for flexibility to meet community needs, such as tailoring session content and/or delivery mode. Acknowledging that intervention adaptation, whether organic or planned, occurs and is beneficial to sustained implementation, elicits an additional consideration of de-implementation of intervention strategies /components considered no longer compatible or effective [30]. De-implementing detrimental or redundant practices is distinct from implementing evidence-based practices, and is considered more difficult, requiring more intense strategies. Norton and Chambers [69] propose four types of de-implementation actions—removing, replacing, reducing or restricting the use of a specific intervention strategy /component. They suggest that future research identify and map specific sustainment barriers to appropriate de-implementation strategies, as is done for implementation strategy development, as well as optimal timeframes and pace at which de-implementation should occur, to mitigate potential harm or unintended negative consequences.
## Implications for the INFANT Program
This study has provided the opportunity to investigate sustained implementation of the INFANT Program during small scale community-level implementation. The factors influencing sustained implementation of the INFANT Program highlight a number of organizational (inner) and system-level (outer) barriers and enablers that are interconnected around prioritization and endorsement, leadership and management support, human and financial resourcing, and capacity building.
These study findings have contributed important insights in preparation for large scale implementation across Victoria and its associated effectiveness-implementation trial [68]. Findings have informed the refinement of intervention characteristics, namely, online facilitator training and refresher training and a community of practice (collaborative online forum), and delivery as four group sessions (3–12 months) supplemented with app-based messages (birth to 18 month). Post COVID-19 and the emergence of telehealth, virtual (online) group delivery has also become a consideration for future exploration. Findings have also informed the selection of specific implementation strategies to support adoption and sustained implementation of the INFANT Program across local government areas. Using the Expert Recommendations for Implementing Change (ERIC) compilation [70], key strategies have been selected to address barriers to sustainability, namely: Furthermore, the INFANT effectiveness-implementation trial [68] will include an evaluation timepoint at 24-months post facilitator training which will assess program sustainability using the Program Sustainability Assessment Tool (self-administered surveys) [73] with follow-up in-depth interviews.
## Implications for research
Since completion of this study, Shelton et al. have developed the Integrated Sustainability Framework (ISF) [29] which proposes key multilevel factors needed for sustained implementation of interventions across settings and contexts, namely, outer contextual characteristics (policy environment and funding, organizational partnerships), inner contextual characteristics (organizational infrastructure and support, leadership and program champions, funding), implementation processes (e.g., recruitment, training, strategic planning and communication, evaluation), characteristics of interventionists (role commitment and motivation, self-efficacy, payment), and intervention characteristics (perceived benefit/need for program, program fit and adaptability). The ISF factors are similar to those identified in the CFIR used in this study [51]. Future research utilizing the ISF would be useful to advance the application of a specific framework to guide implementation sustainability research.
## Strengths and limitations
This study used a recognized theoretical framework within the field of implementation science, the Consolidated Framework for Implementation Research (CFIR) [51]. Described as a determinant framework, mapping identified barriers and enablers to the CFIR advances understanding of how and why sustained implementation occurs across multiple levels of influence using a systems approach [74]. While a relatively small sample size, the response rate was high ($49.2\%$) for this type of research with almost equal representation of LGAs across all patterns of implementation (never, discontinued and ongoing), reducing the risk of social bias. The online survey questions were informed by the literature. Closed response options may have limited participant responses, however, open-ended response fields were also provided to elaborate on survey responses, and participants were offered an optional interview opportunity to expand on responses. This study collected data in 2017 from INFANT Program facilitators who completed their training between 2013 and 2017 with training completion dates evenly distributed across patterns of implementation, therefore any effects of potential participant recall bias would have been similarly distributed.
## Conclusion
This study provides important insights into the barriers and enablers to the sustained implementation of an evidence-based intervention (the INFANT Program) during small scale community-level implementation. The opportunity to gain insights on real-world implementation prior to delivery at-scale is rare, with decisions to scale-up interventions frequently occurring without adequate evidence of effectiveness and/or sustainment [43]. The authors therefore contend that the pre-requisite for scale-up of a population health intervention is not just proof of effectiveness [75] but also proof of sustained implementation at the local/organizational level. In addition, assessment of implementation readiness should occur beyond the stages of adoption and early implementation to inform strategies that support sustained implementation. The use of hybrid type 2 effectiveness-implementation trials is therefore strongly recommended to achieve such concurrent evaluation [68, 76].
The factors influencing sustained implementation of the INFANT Program, predominantly organizational and system-level barriers and enablers, have broad transferability given their remarkable similarity to those identified for health promotion interventions implemented across the world, in healthcare, education and community settings. This study is a reminder that sustained implementation requires investment, effective governance, partnerships and supportive systems. These should be fundamental inclusions when planning “at scale” intervention delivery to optimize opportunities to integrate intervention components into routine practices and policies thereby sustaining implementation.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by Deakin University Human Ethics Advisory Group-Health (HEAG-H 183_2014). The patients/participants provided their written informed consent to participate in this study.
## Author contributions
PL, RL, and KJC conceived and designed the study. PL obtained study ethics and undertook recruitment and study interviewees. ST and MW analyzed the data with cross-checking by PL. PL led the manuscript writing with input from all authors. All authors approved the manuscript for submission.
## Funding
The INFANT RCT trial was funded by a NHMRC grant (APP425801). Additional funds were supplied by the Heart Foundation Victoria and Deakin University. The INFANT Follow-up Study of outcomes at 3.5 and 5 years was funded by a NHMRC grant (APP1008879). RL held an NHMRC ECR Fellowship (#1089415) when conducting the small-scale implementation study. During this current study, PL and ST were funded through the NHMRC CRE for Early Prevention of Obesity in Childhood (EPOCH). KDH is supported by a Heart Foundation Future Leader Fellowship (#105929). The INFANT Effectiveness-Implementation *Trial is* funded by a NHMRC Partnership Grant (APP1161223) and VicHealth.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frhs.2022.1031628/full#supplementary-material
## References
1. Moore TG, Arefadib N, Deery A, Keyes M, West S. *The First Thousand Days: An Evidence Paper—Summary* (2017.0)
2. Blake-Lamb TL, Locks LM, Perkins ME, Woo Baidal JA, Cheng ER, Taveras EM. **Interventions for childhood obesity in the first 1,000 days a systematic review**. *Am J Prev Med.* (2016.0) **50** 780-9. DOI: 10.1016/j.amepre.2015.11.010
3. Deal BJ, Huffman MD, Binns H, Stone NJ. **Perspective: childhood obesity requires new strategies for prevention**. *Adv Nutr.* (2020.0) **11** 1071-8. DOI: 10.1093/advances/nmaa040
4. Yan J, Liu L, Zhu Y, Huang G, Wang P. **The association between breastfeeding and childhood obesity: a meta-analysis**. *BMC Public Health.* (2014.0) **14** 1-11. DOI: 10.1186/1471-2458-14-1267
5. Monasta L, Batty GD, Cattaneo A, Lutje V, Ronfani L, van Lenthe FJ. **Early-life determinants of overweight and obesity: a review of systematic reviews**. *Obes Rev.* (2010.0) **11** 695-708. DOI: 10.1111/j.1467-789X.2010.00735.x
6. Hnatiuk J, Salmon J, Campbell KJ, Ridgers ND, Hesketh KD. **Early childhood predictors of toddlers' physical activity: longitudinal findings from the Melbourne InFANT program**. *Int J Behav Nutr Phys Act* (2013.0) **10** 123. DOI: 10.1186/1479-5868-10-123
7. Zheng M, Lamb KE, Grimes C, Laws R, Bolton K, Ong KK. **Rapid weight gain during infancy and subsequent adiposity: a systematic review and meta-analysis of evidence**. *Obes Rev.* (2017.0) **19** 321-32. DOI: 10.1111/obr.12632
8. 8.UNICEF WHO World Bank Group. Levels and Trends in Child Malnutrition: Key Findings of the 2020 Edition of the Joint Child Malnutrition Estimates. Geneva: World Health Organization (2020).. *Levels and Trends in Child Malnutrition: Key Findings of the 2020 Edition of the Joint Child Malnutrition Estimates* (2020.0)
9. 9.AIHW. Overweight and Obesity Among Australian Children and Adolescents. Canberra: Australian Institute of Health and Welfare (2020).. *Overweight and Obesity Among Australian Children and Adolescents* (2020.0)
10. 10.ABS. National Health Survey First Results 2017–2018. Canberra, AC: Australian Institute of Statistics (2019).. *National Health Survey First Results 2017–2018* (2019.0)
11. 11.AIHW. A Picture of Overweight and Obesity in Australia. Canberra: Australian Institute of Health and Welfare (2017).. *A Picture of Overweight and Obesity in Australia* (2017.0)
12. Haby M, Marwick A, Peeters A, Shaw J, Vos T. **Future predictions of body mass index and overweight prevalence in Australia, 2005–2025**. *Health Promot Int.* (2012.0) **27** 250-60. DOI: 10.1093/heapro/dar036
13. Hennessy M, Heary C, Laws R, van Rhoon L, Toomey E, Wolstenholme H. **The effectiveness of health professional-delivered interventions during the first 1000 days to prevent overweight/obesity in children: a systematic review**. *Obes Rev.* (2019.0) **20** 1691-707. DOI: 10.1111/obr.12924
14. Ash T, Agaronov A, Young T, Aftosmes-Tobio A, Davison K. **Family-based childhood obesity prevention interventions: a systematic review and quantitative content analysis**. *Int J Behav Nutr Phys Act.* (2017.0) **14** 113. DOI: 10.1186/s12966-017-0571-2
15. Askie LM, Espinoza D, Martin A, Daniels LA, Mihrshahi S, Taylor R. **Interventions commenced by early infancy to prevent childhood obesity-The EPOCH collaboration: an individual participant data prospective meta-analysis of four randomized controlled trials**. *Pediatr Obes.* (2020.0) **15** e12618. DOI: 10.1111/ijpo.12618
16. 16.WHO. Report of the Commission on Ending Childhood Obesity. Geneva: World Health Organisation (2016).. *Report of the Commission on Ending Childhood Obesity* (2016.0)
17. Brown T, Moore TH, Hooper L, Gao Y, Zayegh A, Ijaz S. **Interventions for preventing obesity in children**. *Cochrane Database Syst Rev.* (2019.0) **7** CD001871. DOI: 10.1002/14651858.CD001871.pub4
18. Laws R, Adam M, Esdaile E, Love P, Campbell K. **What works to improve nutrition and food sustainability across the first 2000 days of life: a rapid review**. *Nutrients* (2022.0) 14731. DOI: 10.3390/nu14040731
19. Love P, Laws R, Litterbach E, Campbell KJ. **Factors influencing parental engagement in an early childhood obesity prevention program implemented at scale: the infant program**. *Nutrients* (2018.0) **10** 509. DOI: 10.3390/nu10040509
20. 20.DHHS. Victoria statewide Maternal and Child Health Service Annual Report 2017–2018. Victoria: Department of Health and Human Services, Victoria State Government (2019).. *Victoria statewide Maternal and Child Health Service Annual Report 2017–2018* (2019.0)
21. Proctor E, Luke D, Calhoun A, McMillen C, Brownson R, McCrary S. **Sustainability of evidence-based healthcare: research agenda, methodological advances, and infrastructure support**. *Implement Sci* (2015.0) **10** 88. DOI: 10.1186/s13012-015-0274-5
22. Lennox L, Maher L, Reed J. **Navigating the sustainability landscape: a systematic review of sustainability approaches in healthcare**. *Implement Sci.* (2018.0) **13** 27. DOI: 10.1186/s13012-017-0707-4
23. Whelan J, Love P, Millar L, Allender S, Bell C. **Sustaining obesity prevention in communities: a systematic narrative synthesis review**. *Obes Rev Off J Int Assoc Study Obes* (2018.0) **19** 839-51. DOI: 10.1111/obr.12675
24. Wiltsey Stirman S, Kimberly J, Cook N, Calloway A, Castro F, Charns M. **The sustainability of new programs and innovations: a review of the empirical literature and recommendations for future research**. *Implement Sci.* (2012.0) **7** 17. DOI: 10.1186/1748-5908-7-17
25. Schell SF, Luke DA, Schooley MW, Elliott MB, Herbers SH, Mueller NB. **Public health program capacity for sustainability: a new framework**. *Implement Sci.* (2013.0) **8** 15. DOI: 10.1186/1748-5908-8-15
26. Sarma H, D'Este C, Ahmed T, Bossert T, Banwell C. **Developing a conceptual framework for Implement Sci to evaluate a nutrition intervention scaled-up in a real-world setting**. *Public Health Nutr.* (2020.0) **24** s7-s22. DOI: 10.1017/S1368980019004415
27. Moore J, Mascarenhas A, Bain J, Straus S. **Developing a comprehensive definition of sustainability**. *Implement Sci* (2017.0) **12** 10. DOI: 10.1186/s13012-017-0637-1
28. Bodkin A, Hakimi S. **Sustainable by design: a systematic review of factors for health promotion program sustainability**. *BMC Public Health* (2020.0) **20** 964. DOI: 10.1186/s12889-020-09091-9
29. Shelton RC, Rhoades Cooper B, Wiltsey Stirman S. **The sustainability of evidence-based interventions and practices in public health and health care**. *Ann Rev Public Health.* (2018.0) **39** 55-76. DOI: 10.1146/annurev-publhealth-040617-014731
30. Shelton RC, Chambers DA, Glasgow RE. **An extension of RE-AIM to enhance sustainability: addressing dynamic context and promoting health equity over time**. *Front Public Health* (2020.0) **8** 134. DOI: 10.3389/fpubh.2020.00134
31. Palinkas L, Chou C, Spear S, Mendon S, Villamar J, CH B. **Measurement of sustainment of prevention programs and initiatives: the sustainment measurement system scale**. *Implement Sci* (2020.0) **15** 71. DOI: 10.1186/s13012-020-01030-x
32. Goodman RM, Steckler A. **model for institutionalization of health promotion programs**. *Family Commun Health.* (1989.0) **11** 63-78. DOI: 10.1097/00003727-198902000-00009
33. Greenhalgh T, Macfarlane F, Barton Sweeney C. **Woodard F**. *Milbank Quart Multidisc J Popul Health Health Policy.* (2012.0) **90** 516-47. DOI: 10.1111/j.1468-0009.2012.00673.x
34. Chambers DA, Glasgow RE, Stange KC. **The dynamic sustainability framework: addressing the paradox of sustainment amid ongoing change**. *Implement Sci* (2013.0) **8** 117. DOI: 10.1186/1748-5908-8-117
35. Barker PM, Reid A, Schall MW. **A framework for scaling up health interventions: lessons from large-scale improvement initiatives in Africa**. *Implement Sci.* (2016.0) **11** 12. DOI: 10.1186/s13012-016-0374-x
36. Scheirer M. **Is sustainability possible? A review and commentary on empirical studies of program sustainability**. *Am J Eval.* (2005.0) **26** 320-47. DOI: 10.1177/1098214005278752
37. Indig D, Lee K, Grunseit A, Milat A, Bauman A. **Pathways for scaling up public health interventions**. *BMC Public Health.* (2017.0) **18** 68. DOI: 10.1186/s12889-017-4572-5
38. Koorts H, Cassar S, Salmon J, Lawrence M, Salmon P, Dorling H. **Mechanisms of scaling up: combining a realist perspective and systems analysis to understand successfully scaled interventions**. *Int J Behav Nutr Phys Act* (2021.0) **18** 42. DOI: 10.1186/s12966-021-01103-0
39. Glasgow RE, Vogt TM, Boles SM. **Evaluating the public health impact of health promotion interventions: the RE-AIM framework**. *Am J Public Health.* (1999.0) **89** 1322-7. DOI: 10.2105/AJPH.89.9.1322
40. Muellmann S, Steenbock B, De Cocker K, De Craemer M, Hayes C, O'Shea MP. **Views of policy makers and health promotion professionals on factors facilitating implementation and maintenance of interventions and policies promoting physical activity and healthy eating: results of the DEDIPAC project**. *BMC Public Health* (2017.0) **17** 932. DOI: 10.1186/s12889-017-4929-9
41. Shelton RC, Charles T, King Dunstan S, Jandorf L, Erwin DO. **Advancing understanding of the sustainability of lay health advisor (LHA) programs for African-American women in community settings**. *TBM.* (2017.0) **7** 415-26. DOI: 10.1007/s13142-017-0491-3
42. Aarons G, Sklar M, Mustanski B, Benbow N, Brown C. **“Scaling-out” evidence-based interventions to new populations or new health care delivery systems**. *Implement Sci* (2017.0) **12** 96-132. DOI: 10.1186/s13012-017-0640-6
43. Lee K, van Nassau F, Grunseit A, Conte K, Milat A, Wolfenden L. **Scaling up population health interventions from decision to sustainability—a window of opportunity? A qualitative view from policy-makers**. *Health Res Policy Syst.* (2020.0) **18** 118. DOI: 10.1186/s12961-020-00636-3
44. Campbell K, Lioret S, McNaughton SA, Crawford D, Salmon J, Ball K. **A parent-focused intervention to reduce infant obesity risk behaviors: a randomized trial**. *Paediatrics* (2013.0) **131** 652-60. DOI: 10.1542/peds.2012-2576
45. Hesketh KD, Salmon J, McNaughton SA, Crawford D, Abbott G, Cameron AJ. **Long-term outcomes (2 and 3.5 years post-intervention) of the INFANT early childhood intervention to improve health behaviors and reduce obesity: cluster randomised controlled trial follow-up**. *Int J Behav Nutr Phys Act* (2020.0) **17** 95. DOI: 10.1186/s12966-020-00994-9
46. Love P, Laws R, Hesketh KD, Campbell KJ. **Lessons on early childhood obesity prevention interventions from the Victorian INFANT program**. *Public Health Res Pract* (2019.0) **29** 2911904. DOI: 10.17061/phrp2911904
47. Laws R, Hesketh KD, Ball K, Cooper C, Vrljic K, Campbell KJ. **Translating an early childhood obesity prevention program for local community implementation: a case study of the Melbourne InFANT Program**. *BMC Public Health* (2016.0) **16** 748. DOI: 10.1186/s12889-016-3361-x
48. 48.State_Government_of_Victoria. What is Healthy Together Victoria: Creating a Healthier Victoria. (2015). Available online at: https://www2.health.vic.gov.au/about/publications/policiesandguidelines/What-is-Healthy-Together-Victoria (accessed July 15, 2019). (2015.0)
49. Creswell JW, Creswell JD. *Research Design: Qualitative, Quantitative, and Mixed Methods Approaches* (2018.0)
50. Tong A, Sainsbury P, Craig J. **Consolidated criteria for reporting qualitative research (COREQ): a 32-item checklist for interviews and focus groups**. *Int J Qual Health Care.* (2007.0) **19** 349-57. DOI: 10.1093/intqhc/mzm042
51. Damschroder LJ, Aron DC, Keith RE, Kirsh SR, Alexander JA, Lowery JC. **Fostering implementation of health services research findings into practice: a consolidated framework for advancing Implement Sci**. *Implement Sci.* (2009.0) **4** 50. DOI: 10.1186/1748-5908-4-50
52. Vidgen HA, Love PV, Wutzke SE, Daniels LA, Rissel CE, Innes-Hughes C. **A description of health care system factors in the implementation of universal weight management services for children with overweight or obesity: case studies from Queensland and New South Wales, Australia**. *Implement Sci.* (2018.0) **13** 109. DOI: 10.1186/s13012-018-0801-2
53. McLoughlin G, Sweeney R, Liechty L, Lee J, Rosenkranz RGW. **Evaluation of a large-scale school wellness intervention through the consolidated framework for implementation research (CFIR): implications for dissemination and sustainability**. *Front Health Serv* (2022.0) 2. DOI: 10.3389/frhs.2022.881639
54. Damschroder LJ, Reardon CM, Sperber N, Robinson CH, Fickel JJ EZ O. **Implementation evaluation of the Telephone Lifestyle Coaching (TLC) program: organizational factors associated with successful implementation**. *Transl Behav Med.* (2017.0) **2** 233-41. DOI: 10.1007/s13142-016-0424-6
55. Kirk MA, Kelley C, Yankey N, Birken SA, Abadie B, Damschroder L. **Systematic review of the use of the consolidated framework for implementation research**. *Implement Sci.* (2016.0) **11** 72. DOI: 10.1186/s13012-016-0437-z
56. Braun V, Clarke V. *Successful Qualitative Research* (2013.0) 28-31
57. Clarke V, Braun V, Terry G, Hayfiend N. **Thematic analysis**. *Handbook of Research Methods in Health and Social Sciences* (2019.0) 843. DOI: 10.1007/978-981-10-5251-4_103
58. 58.>QSR, International,. NVIVO 11.3.2. Available online at: http://www.qsrinternational.com/what-is-nvivo (accessed January 11, 2019).
59. Mikkelsen B, Novotny R, Gittelson J. **Multi-level, multi-component approaches to community based interventions for healthy living—a three case comparison**. *International journal of environmental research and public health* (2016.0) **13** 1023. DOI: 10.3390/ijerph13101023
60. Shoesmith A, Hall A, Wolfenden L, Shelton R, Powell B, Brown H. **Barriers and facilitators influencing the sustainment of health behaviour interventions in schools and childcare services: a systematic review**. *Implement Sci.* (2021.0) **16** 62. DOI: 10.1186/s13012-021-01134-y
61. Hailemariam M, Bustos T, Montgomery B, Barajas R, Evans LB, Drahota A. **Evidence-based intervention sustainability strategies: a systematic review**. *Implement Sci.* (2019.0) **14** 57. DOI: 10.1186/s13012-019-0910-6
62. Whelan J, Love P, Millar L, Allender S, Morley C, Bell C. **rural community moves closer to sustainable obesity prevention - an exploration of community readiness pre and post a community-based participatory intervention**. *BMC Public Health.* (2019.0) **19** 1420. DOI: 10.1186/s12889-019-7644-x
63. Watt N, Sigfrid L, Legido-Quigley H, Hogarth S, Maimaris W, Otero-Garcia L. **Health systems facilitators and barriers to the integration of HIV and chronic disease services: a systematic review**. *Health Policy Plan.* (2017.0) **32** 13-26. DOI: 10.1093/heapol/czw149
64. Aarons GA, Green AE, Trott E, Willging CE, Torres EM, Ehrhart MG. **The roles of system and organizational leadership in system-wide evidence-based intervention sustainment: a mixed-method study**. *Adm Policy Ment Health.* (2016.0) **43** 991-1008. DOI: 10.1007/s10488-016-0751-4
65. Weiner BJ. **A theory of organizational readiness for change**. *Implement Sci.* (2009.0) **4** 67. DOI: 10.1186/1748-5908-4-67
66. Miake-Lye IM, Delevan DM, Ganz DA, Mittman BS, Finley EP. **Unpacking organizational readiness for change: an updated systematic review and content analysis of assessments**. *BMC Health Serv Res* (2020.0) **20** 106. DOI: 10.1186/s12913-020-4926-z
67. Wiltsey Stirman S, Baumann AA, Miller CJ. **The FRAME: an expanded framework for reporting adaptations and modifications to evidence-based interventions**. *Implement Sci.* (2019.0) **14** 58. DOI: 10.1186/s13012-019-0898-y
68. Laws R, Love P, Hesketh KD, Koorts H, Denney-Wilson E, Moodie M. **Protocol for an effectiveness-implementation hybrid trial to evaluate scale up of an evidence-Based Intervention addressing lifestyle behaviours from the start of life: INFANT**. *Front Endocrinol* (2021.0) **12** 196. DOI: 10.3389/fendo.2021.717468
69. Norton WE, Chambers DA. **Unpacking the complexities of deimplementing inappropriate health interventions**. *Implement Sci.* (2020.0) **15** 2. DOI: 10.1186/s13012-019-0960-9
70. Powell BJ, Waltz TJ, Chinman MJ, Damschroder LJ, Smith JL, Matthieu MM. **A refined compilation of implementation strategies: results from the expert recommendations for implementing change (ERIC) PROJECT**. *Implement Sci.* (2015.0) **10** 21. DOI: 10.1186/s13012-015-0209-1
71. Victorian Government. *Healthy Kids, Healthy Futures* (2021.0)
72. Victorian Government. *Victorian Public Health and Wellbeing Plan 2019–2023.* (2019.0)
73. Luke DA, Calhoun A, Robichaux CB, Elliott MB, Moreland-Russell S. **Peer reviewed: the program sustainability assessment tool: a new instrument for public health programs**. *Prev Chronic Dis* (2014.0) **11** 130184. DOI: 10.5888/pcd11.130184
74. Nilsen P. **Making sense of implementation theories, models and frameworks**. *Implement Sci.* (2015.0) **10** 1-13. DOI: 10.1186/s13012-015-0242-0
75. Milat AJ, Newson R, King L, Rissel C, Wolfenden L, Bauman A. **A guide to scaling up population health interventions**. *Public Health Res Pract.* (2016.0) **26** e2611604. DOI: 10.17061/phrp2611604
76. Hesketh KD, Downing KL, Galland BC, Nicholson JM, Taylor R, Orellana L. **Protocol for the Let's Grow randomised controlled trial: examining efficacy, cost-effectiveness and scalability of a m-Health intervention for movement behaviours in toddlers**. *BMJ Open* (2022.0) **12**. DOI: 10.1136/bmjopen-2021-057521
|
---
title: 'Mobile phone-based lifestyle support for families with young children in primary
health care (MINISTOP 2.0): Exploring behavioral change determinants for implementation
using the COM-B model'
authors:
- Kristin Thomas
- Margit Neher
- Christina Alexandrou
- Ulrika Müssener
- Hanna Henriksson
- Marie Löf
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012784
doi: 10.3389/frhs.2022.951879
license: CC BY 4.0
---
# Mobile phone-based lifestyle support for families with young children in primary health care (MINISTOP 2.0): Exploring behavioral change determinants for implementation using the COM-B model
## Abstract
### Background
Obesity in childhood is a public health concern worldwide and mobile phone-based interventions (mHealth) has shown to facilitate obesity prevention. However, more research is needed on the implementation of digital tools in routine primary care. This study explored behavior change determinants for implementing a health promotion mHealth intervention (MINISTOP 2.0 app) targeting parents of 4-year-olds.
### Methods
Secondary data from telephone interviews ($$n = 15$$) with child health care nurses working within primary child healthcare in Sweden was analyzed using directed content analysis and the COM-B model.
### Results
Barriers for implementation included: limited knowledge about using technology and reservations about how and to what extent parents would use mHealth. Potential facilitators included nurses' openness to learn and try new tools, confidence in their role and engagement in reaching parents as well as beliefs that the app could improve practice by prompting dialogue and being a shared platform. Nurses expressed a strong professional identity and shared understanding of their practice, mechanisms that could potentially inhibit or facilitate implementation.
### Conclusions
Findings suggest cautious optimism regarding implementing mobile phone-based tools in child primary healthcare in terms of capability, opportunity and motivation among stakeholders. Implementation strategies such as educational outreach visits and making the intervention testable among stakeholders could further facilitate implementation in this clinical context. However, more research is needed on behavior change determinants in different stages of real-world implementation.
## Introduction
Obesity in childhood is a public health concern worldwide. According to recent figures, around 38 million children under the age of five are overweight or obese [1, 2]. Sweden also shows relatively high prevalence with ~10–$15\%$ of 4-year-olds being classified as having overweight and obesity [3]. Child primary health care is a key setting for obesity prevention through its reach among diverse populations and regular health visits throughout childhood from birth to school-age [4, 5]. However, there are studies showing the complexity in implementing obesity prevention in routine child health care, primarily due to difficulties in getting parents on board (6–8), but also due to limited resources in health care organizations [9]. Mobile phone-based interventions (so called mHealth) could facilitate obesity prevention in routine care through for example mobile applications aiming to promote healthy lifestyles among children and their families. mHealth in the area of health promotion have shown promising results on body weight and body mass index [10] as well as physical activity, smoking cessation and eating habits (10–15) and quality of care [16].
Although mHealth interventions could facilitate obesity prevention, it is unclear how these relatively novel tools can be implemented in daily routines and which determinants influence the adoption of mHealth tools in healthcare long-term [17]. A systematic review showed that the most common barriers to implementing digital tools in routine healthcare were poor compatibility between the new tool and current workflow, unclear evidence of the technology, and poor organizational readiness to implement digital tools [18]. A recent review on why health care professionals implement mHealth tools noted both technical aspects as well as social and organizational factors, such as the importance of ease of use, trustworthiness of the content and technical support, leadership support, peer influence and costs [17]. However, the review had its focus on the implementation of mHealth targeting health services and care professionals, rather than patients per se.
Implementing mHealth in routine care can be understood as clinical behavior change e.g., child health care nurses recommend or use a mobile application during health visits. The Theoretical Domains Framework (TDF) and the COM-B model [19, 20] has been widely used in research to understand determinants of implementation in terms of behavior change. For example, the COM-B model [20] has been used to explore determinants of health behavior change among patient populations [21, 22] as well as to investigate barriers and facilitators for evidence based practice in healthcare [23]. The TDF synthesize theories and constructs from 33 behavior change theories into 14 domains, argued to generate behavior. The COM-B model further consolidates these 14 domains into three overarching domains representing aspects that are argued to have to be present for a behavior to take place: “capability,” that is, an individual's capacity and competency to engage in a behavior; “opportunity,” which includes environmental factors that influence behaviors such as social support; and “motivation,” which refers to the willingness to engage in a behavior including both conscious and subconscious processes (Table 1 for a complete list of COM-B domains and TDF constructs). In summary, although mHealth can be promising tools to facilitate obesity prevention and promote healthy behaviors among families, more knowledge is needed on how mHealth can be incorporated in routine practice. Investigating behavioral determinants among nurses for using mHealth is a critical first step in understanding mHealth implementation of family-facing mHealth technology.
**Table 1**
| Capability | Unnamed: 1 |
| --- | --- |
| Psychological capability | Knowledge• Knowledge related to research evidence e.g., nutrition and health• Practice-based knowledge e.g., national guidelines• Tacit knowledge e.g., understanding of target group needs• Expressed need for training on specific mHealth tools before use in practiceCognitive skills• Competencies used when meeting patients• Communication skills e.g., using non-dramatic terminology• Inter-personal skills e.g., validating parents to create a safe space• Perceived opportunities in using mHealth tools to facilitate communication and access health data |
| Physical capabilities | Not apparent in data |
| Opportunity | |
| Social opportunity | Social influences• Social norms, social support and professional identity were highly prevalent• Shared understanding among nurses regarding practice routines• Shared understanding among nurses regarding using a biopsychosocial approach |
| Physical opportunity | Factors in the environmental context and resources• “Eco-system of support” around at-risk children e.g., dieticians, physicians and teachers• Expressed need for increased collaboration among stakeholders and access to specialized care• Perceived opportunities of mHealth tools e.g., reach diverse populations through tailored mHealth resources |
| Motivation | |
| Reflective motivation | Social and professional roles• Professional identity and responsibility to support and guide parents.• Roles and responsibilities of health care professionals and target group (parents)• Perceived need for mHealth tools targeting whole families including parents and childrenBeliefs about capabilities• Confidence to engage in health promotion work and use mHealth tools• Perceived need for induction to use specific mHealth toolsOptimism• Confidence that health care professionals and families are able to use mHealth tools |
| | • Perceived opportunities of mHealth tools to offer a shared platform with families, accessing material in several languages and opportunities for continuous support, monitoring and follow-up of the children and their families.• Perceived misgivings about using mHealth e.g., ensuring long term use among familiesEmotions• Both negative and positive emotions regarding using mHealth tools in practice e.g., frustration and curiosity• Expressed emotional engagement among nurses in families' efforts to adopt healthy routines |
The mHealth intervention “MINISTOP” is a mobile application that was initially developed targeting parents of 4-year-olds. The app ultimately aimed to reduce the prevalence of overweight and obesity by giving support to improve diet and physical activity (MINISTOP 1.0 app). The app showed promising results on dietary and activity behaviors in a randomized controlled trial (OR: 2.0; $95\%$ CI 1.2–3.1; $$p \leq 0.008$$) [24, 25]. MINISTOP 1.0 has thereafter been refined and modified to be used within Swedish primary child healthcare targeting parents of 2–3-year-olds [MINISTOP 2.0 app [26]].
## Aim
To explore behavior change determinants for implementing a mHealth intervention (MINISTOP 2.0 app) for family nursing practice in primary healthcare.
## Study design and setting
This study is a qualitative interview study with registered nurses working in child primary health care *Secondary data* [27, 28] from semi-structured interviews was analyzed using directed content analysis [29] and the COM-B model [20]. The study is part of a larger research project aiming to develop and evaluate the effect of a mHealth intervention on health behavior change among parents of children aged 2–3 years [MINISTOP 2.0 trial [26]]. As part of the development of the app, interviews with nurses and parents were conducted and secondary data from interviews with nurses were used in this study. The app provides a 6-month behavior change program and includes the following features: information and practical tips provided in 13 themes with one theme released every 2 weeks (e.g., healthy snacks, fruits and vegetables, physical activity and screen time, food as rewards), a registration feature where parents can report their child's intake of fruits and vegetables, sweet drinks, physical activity and screen time and a library of healthy recipes and tips for physical activity indoor and outdoor. The parents also receive feedback in graphs and messages once a week based on their registrations. The app has also been translated and adapted for Somali- and Arabic speaking families including a large series of audio files in Somali and Arabic. As it is a web-based app, it also has a user-interface where the nurses can register new users (parents) and through that interface they are also able to follow the parental dietary and physical activity registrations mentioned above. The research was performed according to the *Consolidated criteria* for Reporting Qualitative Research (COREQ) checklist [30] (see Supplementary material 1).
Swedish primary child healthcare is commissioned to work with health promotion and disease prevention in a structured way. As part of this work, routine health visits follow a national health program including regular consultations with a registered nurse. In addition to height, weight, cognitive and social development of the child, these regular visits are used as a platform for the promotion of health behaviors such as physical activity and a healthy diet.
## Participants and procedure
In the original study, informants were recruited using convenience sampling with inclusion criteria: [1] currently employed at one of the participating centers and [2] willing to participate. An invite to take part in a telephone interview was sent out via e-mail by CA (co-author) during September 2019, and nurses registered their interest by replying to the e-mail. Invitation letters were sent to nurses ($$n = 35$$) at health care centers that had agreed to participate in the MINISTOP 2.0 trial [26]. Recruitment was conducted from 24 primary care centers. A total of 15 nurses registered an interest to take part in interviews and were interviewed. The invitation letter consisted of information on the study including study aims, that participation was voluntary and that they could leave the study at any time.
## Data collection
Secondary data [27, 28] from interviews with child health care nurses conducted within the research project described above was used. The aim of the original study was to explore nurses' perceptions of parents' needs and concerns regarding diet and physical activity and nurses' perceptions about how the MINISTOP 1.0 app could be refined to meet the needs of the target group. In the original study, data was collected through semi-structured telephone interviews using an interview guide (Supplementary material 2) developed by the research group that has expertise in nutrition, physical activity, behavior change, and qualitative methodology. The interview guide was generated to explore nurses' perceptions about the need of the target group and preferences for using a digital tool in routines. For example, the interview guide explored nurses' perceptions regarding current work routines, needs and concerns among target group and perceptions of the MINISTOP 1.0 app. Thus, the secondary analysis used an implementation perspective, which was not done in the original study. CA (female and PhD student) conducted all the interviews which lasted on average 1 h (range 37–90 min). Field notes were taken after interviews. Informed verbal consent was obtained and recorded at the beginning of each interview. Informants were told that the interviewer was a PhD student however no relationship was established between participants and the interviewer prior to study commencement. Only participants and interviewer were present during interviews.
## Data analysis
Secondary data analysis was carried out using raw data from previously collected material. Directed content analyses [29] and the COM-B model [20] was used in data analysis. MN and KT conducted all the secondary data analysis including generating the codebook. A codebook based on the COM-B domains were used in data analysis (Supplementary material 3). Initially, key parts of COM-B were translated from English to Swedish i.e., domains (capability, opportunity, motivation and behavior). In Michie et al. [ 31], an extensive explanation is given of the connection between the TDF and COM-B [[31], p 87–93]. The authors point out that to identify what needs to change or when a more detailed understanding of the behavior is required, the TDF can be used to expand on COM-B domains identified in the behavioral analysis. To gain a richer understanding for the domains and how they could be operationalised for this particular dataset, constructs from the TDF were therefore also used when generating the codebook [31] (e.g., cognitive skills, beliefs about capabilities social influence etc.). Then, KT and MN individually generated codebooks based on the translated domains and constructs. The two codebook drafts were discussed and consensus about one final codebook was reached (see Supplementary material 3).
KT and MN performed the secondary data analysis separately using the codebook to ensure consistency. However, inter-coder reliability was strived for through regular meetings between KT and MN throughout the analysis process. Firstly, all the transcripts were read through to gain an understanding and impression of the data as a whole. Then, all data was reviewed for content and coded according to the pre-defined categories in the codebook. This coding phase involved identifying data that corresponded with, or exemplified, COM-B domains by using the codebook. Only data that corresponded to pre-defined categories were coded. Preliminary findings (sorting and coding of data) were discussed between MN and KT in an iterative process until agreement was reached. In a final step, KT and MN together drafted text that described each category including selecting citations that could illustrate the content. KT drafted the first version of the results section for this manuscript.
## Results
The study aimed to explore determinants for implementing a parent-oriented mHealth intervention in health promotion practice in primary child healthcare. In total, 15 nurses from nine primary child healthcare centers took part in telephone interviews. The participants were on average 47 years of age (between 37 and 55) and had on average worked in their profession for 7 years (between three and 11 years). Implementation referred to nurses introducing the MINISTOP 2.0 app to parents of 4-year-olds within family nursing practice during routine health visits. The analysis explored how nurses perceived their current health promotion practice and used the COM-B model [20] to systematically map determinants in data. Results are presented below for each COM-B domain.
## Behavior
In the interviews, the nurses described and reflected on what their current work routines entailed. Nurses expressed health promotion work to be continuous, preventative, and comprehensive, aiming to support families from infancy to school-age. The work included monitoring children's social, psychological, and physical health mainly through meeting families during scheduled visits and referring to specialists when needed. In their conversations with parents, they provided information about health risks related to overweight and obesity in childhood and guided parents through healthier living. This is the professional work in which the MINISTOP 2.0 app would be introduced. “ Within child health care all health visits are preventative … so we talk about growth curves and health behaviours in all visits at the clinic” (Informant 7).
## Capability
In the COM-B model, capability refers to psychological and physical capability to engage in a behavior. The data mainly included nurses exhibiting knowledge and cognitive skills to do health promotion work and to use mHealth in this practice. Thus, aspects of physical capabilities were not apparent in the data.
Nurses expressed extensive knowledge on the responsibility, approach, and routines of the primary child healthcare organization. Their knowledge encompassed both research evidence about for instance nutrition and health, practice-based knowledge about e.g., national guidelines and pedagogical tools as well as tacit knowledge about parents' everyday life and concerns. Furthermore, the nurses stressed the need to keep themselves updated to ensure quality of care. Nurses kept updated through their contact with parents, learning from cultural bridge-builders, other colleagues and searching the Internet. ” I have gained so much from them [bridge builders] …a lot…a culture competency which I didn't know…yeah didn't realise existed before I started here” (Informant 11). In addition, nurses highlighted the importance of staying curious and open-minded about the meaning of food in different cultural contexts to be able to support parents. “ *It is* quite difficult to know I think … if I don't come from the same food culture as the person I meet…then I don't know exactly what they eat” (Informant 14). Regarding using mHealth in practice, the nurses highlighted the need for training about specific tools to increase capability, including perhaps testing MINISTOP themselves before disseminating it among parents. ”[ sigh] Yeah… you need training as well of course… and… guidelines on how … yes how to use it [mobile application] and that we get a united way of working is important” (Informant 8).
Cognitive skills referred to the competencies the nurses used when meeting parents. These skills included for instance adopting a light-hearted approach, using non-dramatic terminology, and validating parents' concerns to create a safe environment for parents to share information. Using one's cognitive skills also involved to always assess the situation and the individual in front of you and changing communication techniques accordingly.” what experience does the parent have?//…the approach becomes who are you?…and what do you need from me?…these things I need to explore before I give advice” (Informant 10). Thus, health promotion work was described as a two-way process with shared responsibility between professional and parent. Although nurses expressed confidence in their role, child obesity was described as a potentially sensitive subject that can provoke strong emotions among parents such as pride, obstinance, guilt and shame. ” Others blame themselves and believe that I have actually done wrong as a mother, yeah, and so there is some shame and guilt in this” (Informant 9). Regarding their cognitive skills, the data suggested that there was a capability among nurses to use mHealth in current health promotion work. For example, nurses expressed that mHealth could be used together with parents and could facilitate communication by making the topic of obesity less dramatic. Also, the potential of accessing data on families continuously through the mobile phone was thought to enable monitoring long-term and ultimately improve the communication during visits.
## Opportunity
Within COM-B, opportunity refers to social and physical opportunities to engage in a behavior. The data included aspects on social influences and factors in the environmental context relating to health promotion work and using mHealth in practice.
Data on social influences, conveyed that social norms, social support and group-identity were highly prevalent in nurses' health promotion work. This was illustrated by nurses' shared understanding and acceptance of practice routines and understanding of health whereby social, psychological, and physical health concerns were continuously monitored and addressed, from infancy to school-age. “ In child primary health care we work preventative at all visits…so we talk about this with children and growth curves and lifestyle…at all visits” (Informant 7). Furthermore, nurses expressed that health promotion work is more than promoting healthy behaviors: it is also about parenthood and inducing confidence in parents to be able to follow through with health behavior change.
Regarding, environmental context and resources, nurses described a network of actors around at-risk children that could be potential resources in promoting health such as nurses, dieticians, physicians, specialists, interpreters and bridge-builders and teachers in nursery. ” What happens at home…what happens there…because of course we call them back after six months but what happens…in the family at nursery at the grandparents?” ( Informant 7). Nurses expressed that collaboration within this “ecosystem of support” can be challenging but also rewarding for instance through working with bridge-builders to learn about different food cultures. Nurses talked about needing more resources such as increased access to specialized care as well as hands-on tools and materials that could facilitate communication and dissemination of information. For example, nurses talked about opportunities with future mHealth interventions to be available in several languages including pictures which are valid across cultures. “ And especially if there is something we use these pictures…or if you have difficulties with the language or…then it is very good to have a picture” (Informant 7). In some cases, nurses perceived that the pedagogic and information materials that they currently had access to were not up to date and that their methods of counseling were not attractive to the families they served. “ But the fact that there is no…that there…if we talk about balanced diet examples…everything is in Swedish…everything is adapted to how a Swedish plate looks like…not how it looks for Somali families or a family from China//here we have so much to learn…so much to learn” (Informant 13).
## Motivation
Within the COM-B model, motivation refers to both reflective (conscious) and automatic (subconscious) processes to engage in a behavior. The data mainly included motivation in terms of social and professional roles, beliefs about capabilities, optimism, intention and emotions associated with health promotion work and using mHealth in practice. Thus, other dimensions of motivation that is described in COM-B such as reinforcement, goals and beliefs about consequences were not apparent in data.
Regarding social and professional roles, nurses conveyed a strong professional identity and responsibility to support and guide parents. Nurses also described boundaries for their responsibility, or ability to help, with families that despite several efforts, were difficult to reach. “ You know [sighing] sometimes you don't get there…sometimes there is no interest…sometimes you can't do it” (Informant 14). Nurses highlighted that obesity often is a problem in the family as a unit whereby both professionals and parents have a responsibility. Relating to this, the nurses sought for mHealth interventions that targeted whole families rather than parents per se by for instance engaging children in the material. Furthermore, the nurses expressed that recruiting parents and supporting parents' long-term use of MINISTOP were important and part of their professional role.
Data on beliefs about capabilities included professional confidence to engage in health promotion work but also about using mHealth. Nurses were confident in performing health promotion work also stating that they perceived that their parent-and -child interaction skills developed over time. When asked to give their initial reaction on implementing a mHealth in this practice, nurses expressed that they would like to try out using the tool but that there could be a need for staff introduction and training to be able to integrate the tool fully in practice. ” *What is* needed is that everybody works with the app the same way…because sometimes we meet each other's children and…so it is important that everybody has the same training so we work the same way…refer similar cases” (Informant 4).
Nurses expressed both optimism and challenges toward working with mHealth in routine practice. Apart from an optimism that both themselves and parents are technically savvy, nurses could identify additional benefits for example, the opportunity for parents to constantly access information and support as opposed to only when visiting the clinic. Nurses expressed that the mHealth intervention could facilitate the work with hard-to-reach families, especially parents of high-risk children and parents with poor reading and writing skills through alternative channels such as audio and video. ” *It is* actually those families…yes…if we look in general…if we look at our families here now…so issues around diet…around teeth …around overweight then it is problem in this group…that's where it is most difficult to reach…and of course you can influence…we can see…but can you reach…can you…can we use a tool that we use together then it would be easier…I believe” (Informant 13). Other benefits mentioned were the potential of having a shared platform with parents, accessing material in several languages and opportunities for continuous support, monitoring and follow-up of the children and their families. Nurses also expressed misgivings about using mHealth, such as the added distraction for parents,” That…they use the app…exactly…when do I use the app…that's the point…as a parent…do I look at that instead of my child?” ( Informant 10). Other challenges described were achieving good communication in technical solutions, and difficulties in achieving long-term use among parents. ” Then it is this with the …the in-person meeting…to like still…whatever app you have…to be able to refer to the in-person meeting” (Informant 10). As their advice on screen activity was usually about limiting the time spent with media, they worried about introducing another screen-dependent activity in the lives of the families they served. Although they indicated that modern families were very cognizant about using mobile phones in general, they also reflected on the risk of MINISTOP “disappearing” in the crowd of mobile applications that parents use every day. *In* general, the nurses expressed an optimism that MINISTOP would fit with current routines and feasibility to recruit families. ” It fits very well in our like when we talk about…diet and sleeping and screen time and activity and so on…this is exactly what we discuss at every visit” (Informant 7).
The nurses described that the goal of monitoring health behaviors in families in their practice led them to take intentional charge of the conversations with parents in different ways. These intentions also led them to provide suggestions concerning the mHealth and the desirable functions for staff. Although they were used to using mobile phone applications in general, they were not very familiar with the MINISTOP application, but they stated that they welcomed new ways to manage their health promotion task. Nurses expressed that parents' intention to work with mHealth would depend on the characteristics of the intervention itself, but also on the health behavior interest of the family. They believed that mHealth would be useful for health promotion, but also that outside support was necessary to keep the issue at the top of the family's agenda.
The nurses expressed negative emotions like frustration, resignation and worry but also positive emotions such as excitement and curiosity regarding their work and using mHealth in practice. ” Yes…no but there is nothing that I feel at the moment…that no but…when I see this it feels really exciting I think…we can hope that the parents also think that…or will think [laughter]” (Informant 15). The nurses were emotionally engaged in the success and failure of their efforts to involve parents in the health promotion conversations, and disappointed when the child's health data indicated that results were lacking. As conversations about child obesity could lead to parental emotions such as shame and blame, nurses expressed that they experienced negative emotional stress in these situations. When asked to reflect on the use of MINISTOP by parents, the participants expressed feeling hesitation, but also excitement and curiosity, and they expressed that they looked forward to working with the app.
In summary, potential barriers for behavior change were limited knowledge and reservations among nurses regarding the use of the intervention among the target group. Potential facilitators for behavior change were nurses' openness, confidence and engagement in their professional role and beliefs that digital tools could improve practice.
## Discussion
This study explored behavior change determinants for implementing an mHealth intervention (MINISTOP 2.0 app) in current health promotion practice in primary child healthcare. Determinants in terms of both barriers and facilitators for implementation were identified. Limited knowledge about MINISTOP and reservations about how and to what extent parents would use the intervention were identified as the main possible hinders. Potential facilitators were nurses' openness to learn and try innovations, nurses' confidence and engagement in their professional role and beliefs that mHealth could improve practice by prompting dialogue and being a shared platform. Finally, nurses expressed a strong professional identity and shared understanding of their practice, mechanisms that could potentially hinder or facilitate implementation of mHealth.
One of the potential barriers for implementing MINISTOP 2.0 was the capability among nurses to use the intervention. The COM-B model posits that the capability and motivation to engage in a behavior are interrelated, and would together with behavioral opportunities, contribute to adoption of mHealth [20]. Although nurses expressed these hinders, our data indicated that nurses also exhibited capability (in terms of cognitive skills), opportunity and motivation to use and implement the MINISTOP in their daily routines. Indeed, well-known facilitators of practice change were observed such as nurses' openness to change and beliefs that the mHealth intervention would improve practice. In contrast, previous research in pediatric care has shown that poor buy-in and engagement among adopters together with limited time and information are typical barriers to implement mHealth [31]. Innovations that are compatible with existing norms, values and ways of working have shown to easier engage adopters which could partly explain our findings. Indeed, the child primary healthcare context was found to be characterized by a strong professional identity, engagement, and long-term relationship with families. All these mechanisms have the potential to facilitate implementation of any innovation that is compatible with these notions and ethos [32]. The nurses expressed that MINISTOP was compatible with their work for example by taking a preventative and holistic approach to health and offering information to parents.
Research on implementation has shown that an important determinant is how end-users perceive different attributes of the intervention so called innovation characteristics. Indeed, the central tenet of the Diffusion of Innovations theory [33] is that how attributes of a technology are perceived by stakeholders will influence implementation. Facilitating attributes include perceived relative advantage (the intervention is perceived to be superior to existing routines), complexity (an intervention is simple to use and understand), compatibility (an intervention matches established routines and norms), observability (potential benefits of an intervention is visible), and trialability (an intervention can be tested prior full-scale implementation) of an innovation [33]. The possible barriers can be understood as the complexity and trialability of MINISTOP 2.0, that is, nurses expressed wanting more knowledge about the intervention and testing it before full-scale implementation. Effective implementation strategies to facilitate implementation could thus be educational outreach visits and making MINISTOP testable among nurses to promote familiarity [34]. However, to fully facilitate the implementation of digital interventions in routine practice, implementation strategies need to target all barriers for change. Thus, strategies need to be systematically developed based on a thorough investigation of determinants specific for this particular context and intervention [31, 35].
Our findings echo existing literature on mHealth implementation. Indeed, a systematic review on mHealth adoption in healthcare highlighted perceived usefulness and familiarity, training and access to resources to be key determinants of successful implementation [17]. Usefulness in the mHealth implementation literature typically refers to that mHealth interventions will not only fit current routines, but also make valuable additions to these. Our findings can add to the understanding of the usefulness of mHealth tools in healthcare, knowledge that is central for intervention development and implementation. In our data, nurses spoke about the usefulness of MINISTOP 2.0 in terms of the opportunities to monitor behavior over time and to create a shared platform incorporating multiple stakeholders. Nurses described how obesity prevention engaged a network of actors and that mHealth hold great potential in providing a shared platform in this work. A qualitative study on the use of mHealth in general practice similarly characterized usefulness as creating shared platforms for patients and healthcare providers [36]. However, it can be a challenge for future mHealth designs to, on one hand, include multiple stakeholders and, on the other hand, tailor content to specific target groups. mHealth interventions can be designed to be a platform where information is disseminated to patients, alternatively, mHealth can be designed as a dynamic tool, a place, where different stakeholders can communicate and share experiences continuously. In addition, the nurses voiced that they would like mHealth interventions to be a tool shared between themselves and parents. This could for example be done by parents registering their health behaviors, data that nurses would have access to, and used in consultations.
## Methodological considerations
A strength of the study is the use of a theory-based codebook in data analysis which supported neutrality and consistency in the analysis process [37]. Trustworthiness and scientific rigor were strived for in several ways [37]. Researchers with varied research backgrounds and competencies were involved during study design, data collection and analysis which could have increased credibility. In addition, investor triangulation was used in data analysis to ensure dependability of the interpretation of data. The theory-based codebook enabled systematic data analysis which together with investigator triangulation could have increased dependability and confirmability. Furthermore, we have strived to increase transferability by providing detailed description of the procedure and thick descriptions of the results, illustrated by quotes from data. Transferability may have also been increased by the fact that healthcare centers, in which informants work, were located in diverse socioeconomic and geographical areas.
A potential challenge with using secondary data is that the interview material and the original interview guide may not adequately answer the study aim. Therefore, before starting data analysis, we read through the material to assess whether it included sufficient scope and depth to capture our study aim. We deemed that the data were sufficient to investigate conditions for mHealth implementation among nurses and offered valuable insights in this regard. However, including interviews with practice managers and regional managers could have strengthened the study. The study offers the perspective of registered nurses and knowledge on implementation in the healthcare visit setting. Future studies could include other groups of informants to enrich our understanding of implementation on other levels of the primary care organization.
The study explored future, potential, determinants for implementing an innovation in current practice, rather than actual experienced determinants. Perceived determinants prior an implementation is not necessarily the same as the ones that are later experienced during actual implementation. However, the nurses were experienced in health promotion work and showed extensive knowledge about obesity prevention and preconditions increasing the validity of the findings. Future research needs to investigate readiness to change and determinants among healthcare professionals with hands-on experience with mHealth implementation. This study adopted a point of departure in current practice routines to understand determinants for future implementation.
## Conclusions
This study indicates cautious optimism regarding the preconditions for implementing mHealth in child primary healthcare in terms of capability, opportunity and motivation among stakeholders. Implementation strategies such as educational outreach visits and making the intervention testable among stakeholders could further facilitate implementation in this clinical context. However, more research is needed on the impact of behavior change determinants in different stages of real-world implementation.
## Data availability statement
The raw data supporting the conclusion of this article can be made available by the authors on reasonable request.
## Ethics statement
The studies involving human participants were reviewed and approved by the regional Ethical Review Board of Uppsala, Sweden (ref no 2019-02747; 2020-01526). The patients/participants provided their written informed consent to participate in this study.
## Author contributions
ML is the principle investigator for the MINISTOP 2.0 trial, including this qualitative study. CA and HH performed the data collection in the original study. KT and MN conducted the secondary data analysis. KT drafted the first version of the manuscript with significant contributions from ML, CA, UM, MN, and HH. All authors read and approved the final version of the manuscript.
## Funding
This research was conducted within the MINISTOP 2.0 trial and funded by several grants: Swedish Research Council for Health, Working Life and Welfare (FORTE, 2018-01410; PI ML), Region Östergötland (LIO-893101; PI ML), Region Östergötland (LIO-920441; HH), and Lions Research Fund (PI HH). None of the funders have had input on study design, data collection, data analysis, or preparation of the manuscript.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frhs.2022.951879/full#supplementary-material
## References
1. **World health Organization**
2. Bentham J, Di Cesare D, Bilano V, Bixby H, Zhou B, Stevens GA. **Worldwide trends in body-mass index, underweight, overweight, and obesity from 1975 to 2016: a pooled analysis of 2416 population-based measurement studies in 128·9 million children, adolescents, and adults**. *Lancet.* (2017.0) **390** 2627-42. DOI: 10.1016/S0140-6736(17)32129-3
3. Ahrens W, Pigeot I, Pohlabeln H, De Henauw S, Lissner L, Molnár D. **Prevalence of overweight and obesity in European children below the age of 10**. *Int J Obes.* (2014.0) **38** S99-S107. DOI: 10.1038/ijo.2014.140
4. Lissner L, Wijnhoven TMA, Mehlig K, Sjöberg A, Kunesova M, Yngve A. **Socioeconomic inequalities in childhood overweight: heterogeneity across five countries in the WHO European Childhood Obesity Surveillance Initiative (COSI-2008)**. *Int J Obes.* (2016.0) **40** 796-802. DOI: 10.1038/ijo.2016.12
5. Nishtar S, Gluckman P, Armstrong T. **Ending childhood obesity: a time for action**. *Lancet* (2016.0) **387** 825-7. DOI: 10.1016/S0140-6736(16)00140-9
6. Thompson SR, Watson MC, Tilford S. **The ottawa charter 30 years on: still an important standard for health promotion**. *Int J Heal Promot Educ.* (2018.0) **56** 73-84. DOI: 10.1080/14635240.2017.1415765
7. Mateo GF, Granado-Font E, Ferré-Grau C, Montaña-Carreras X. **Mobile phone apps to promote weight loss and increase physical activity: a systematic review and meta-analysis**. *J Med Internet Res.* (2015.0) **17** 1-11. DOI: 10.2196/jmir.4836
8. 8.WHO. Overview of mHealth, mHealth new horizons for health through mobile technologies: a second global survey on eHealth. World Health Organization Global Observatory for eHealth (2011).. *Overview of mHealth, mHealth new horizons for health through mobile technologies: a second global survey on eHealth* (2011.0)
9. Free C, Phillips G, Galli L, Watson L, Felix L, Edwards P. **Effectiveness of mobile-health technology-based health behaviour change or disease management interventions for healthcare consumers: a systematic review**. *PLoS Med* (2013.0) **10** e1001362. DOI: 10.1371/journal.pmed.1001362
10. Adam PCG, Smolenski DJ, Wong HTH, de Wit JBF. **A meta-analysis of overall effects of weight loss interventions delivered via mobile phones and effect size differences according to delivery mode, personal contact, and intervention intensity and duration**. *Obes Rev.* (2017.0) **18** 450-9. DOI: 10.1111/obr.12492
11. Ludwig K, Arthur R, Sculthorpe N, Fountain H, Buchan DS. **Text messaging interventions for improvement in physical activity and sedentary behavior in youth: Systematic review**. *JMIR MHealth UHealth.* (2018.0) **6** 1-18. DOI: 10.2196/preprints.10799
12. O'Rourke L, Humphris G, Baldacchino A. **Electronic communication based interventions for hazardous young drinkers: a systematic review**. *Neurosci Biobehav Rev.* (2016.0) **68** 880-90. DOI: 10.1016/j.neubiorev.2016.07.021
13. Smedslund G, Wollcheid S, Fang L, Nilsen W, Steorio A, Larun L. **Effect of early, brief computerized interventions on risky alcohol and cannabis use among young people**. *Campbell Syst Rev* (2017.0) **2017** 6. DOI: 10.4073/csr.2017.6
14. Whittaker R, McRobbie H, Bullen C, Rodgers A. **Mobile phone-based interventions for smoking cessation (review)**. *Cochrane Libr.* (2016.0) **4** 1-22. DOI: 10.1002/14651858.CD006611.pub4
15. Scott-Sheldon LAJ, Lantini R, Jennings EG, Thind H, Rosen RK, Salmoirago-Blotcher E. **Text messaging-based interventions for smoking cessation: a systematic review and meta-analysis**. *JMIR MHealth UHealth.* (2016.0) **4** e49. DOI: 10.2196/mhealth.5436
16. Slater H, Campbell JM, Stinson JN, Burley MM, Briggs AM. **End user and implementer experiences of mhealth technologies for noncommunicable chronic disease management in young adults: systematic review**. *J Med Internet Res* (2017.0) **19** e406. DOI: 10.2196/jmir.8888
17. Gagnon MP, Ngangue P, Payne-Gagnon J, Desmartis M. **m-Health adoption by healthcare professionals: a systematic review**. *J Am Med Informatics Assoc.* (2016.0) **23** 212-20. DOI: 10.1093/jamia/ocv052
18. Weiner BJ. **A theory of organizational readiness for change**. *Implement Sci.* (2009.0) **4** 67. DOI: 10.1186/1748-5908-4-67
19. Atkins L, Francis J, Islam R, O'Connor D, Patey A, Ivers N. **A guide to using the theoretical domains framework of behaviour change to investigate implementation problems**. *Implement Sci.* (2017.0) **12** 1-18. DOI: 10.1186/s13012-017-0605-9
20. Michie S, van Stralen MM, West R. **The behaviour change wheel: a new method for characterising and designing behaviour change interventions**. *Implement Sci.* (2011.0) **6** 42. DOI: 10.1186/1748-5908-6-42
21. Khalilollahi A, Kasraian D, Kemperman ASAM, van Wesemael P. **Application of the COM-B model to the correlates of children's outdoor playing and the potential role of digital interventions: a systematic literature review**. *Children's Geographies* (2022.0). DOI: 10.1080/14733285.2022.2075692
22. Timlin D, McCormack JM, Simpson EEA. **Using the COM-B to identify barriers and facilitators towards adoption of a diet associated with cognitive function (MIND diet)**. *Public Health Nutr* (2021.0) **24** 1657-70. DOI: 10.1017/S1368980020001445
23. De Leo A, Bayes S, Bloxsome D, Butt J. **Exploring the usability of the COM-B model and theoretical domains framework (TDF) to define the helpers of and hindrances to evidence-based practice in midwifery**. *Implement Sci Commun.* (2021.0) **2** 7. DOI: 10.1186/s43058-020-00100-x
24. Henriksson P, Leppänen MH, Henriksson H, Delisle Nyström C, Cadenas-Sanchez C, Ek A et al. **Physical fitness in relation to later body composition in pre-school children**. *J Sci Med Sport.* (2019.0) **22** 574-9. DOI: 10.1016/j.jsams.2018.11.024
25. Nyström CD, Sandin S, Henriksson P, Henriksson H, Trolle-Lagerros Y, Larsson C et al. **Mobile-based intervention intended to stop obesity in preschool-aged children: the MINISTOP randomized controlled trial**. *Am J Clin Nutr.* (2017.0) **105** 1327-35. DOI: 10.3945/ajcn.116.150995
26. Henriksson H, Alexandrou C, Henriksson P, Henström M, Bendtsen M, Thomas K. **MINISTOP 2**. *BMC Public Health* (2020.0) **20** 1-11. DOI: 10.1186/s12889-020-09808-w
27. Heaton J. **Secondary analysis of qualitative data: an overview**. *Hist Soc Res* (2008.0) **33** 33-45
28. Alexandrou C, Müssener U, Thomas K, Henriksson H, Löf M. **Adapting a parental support app to promote healthy diet and physical activity behaviors (MINISTOP) for a multi-ethnic setting: a qualitative study on the needs and preferences of parents and nurses within Swedish child health care**. *Nutrients* (2021.0) **13** 7. DOI: 10.3390/nu13072190
29. Hsieh HF, Shannon SE. **Three approaches to qualitative content analysis**. *Qual Health Res.* (2005.0) **15** 1277-88. DOI: 10.1177/1049732305276687
30. Tong A, Sainsbury P, Craig J. **Consolidated criteria for reporting qualitative research (COREQ): a 32-item checklist for interviews and focus groups**. *Int J Qual Heal Care.* (2007.0) **19** 349-57. DOI: 10.1093/intqhc/mzm042
31. Michie S, Atkins L, West R. *The Behaviour Change Wheel: A Guide To Designing Interventions* (2014.0)
32. Kirchner JAE, Smith JL, Powell BJ, Waltz TJ, Proctor EK. **Getting a clinical innovation into practice: an introduction to implementation strategies**. *Psychiatry Res* (2020.0) **283** 112467. DOI: 10.1016/j.psychres.2019.06.042
33. Rogers EM. **Diffusion of preventive innovations**. *Addict Behav.* (2002.0) **27** 989-93. DOI: 10.1016/S0306-4603(02)00300-3
34. Powell BJ, Waltz TJ, Chinman MJ, Damschroder LJ, Smith JL, Matthieu MM. **A refined compilation of implementation strategies: results from the expert recommendations for implementing change (ERIC) project**. *Implement Sci.* (2015.0) **10** 1-14. DOI: 10.1186/s13012-015-0209-1
35. Fernandez ME, Gill A, van Lieshout S, Rodriguez SA, Beidas RS, Parcel G. **Implementation mapping: using intervention mapping to develop implementation strategies**. *Front Public Health.* (2019.0) **7** 158. DOI: 10.3389/fpubh.2019.00158
36. Nguyen AD, Frensham LJ, Baysari MT, Carland JE, Day RO. **Patients' use of mobile health applications: what general practitioners think**. *Fam Pract.* (2018.0) **36** 214-8. DOI: 10.1093/fampra/cmy052
37. Guba E. **Criteria for assessing the trustworthiness of naturalistic inquiries**. *Educ Commun Technol.* (1981.0) **29** 75-91. DOI: 10.1007/BF02766777
|
---
title: Locus of Control and Self-Esteem as Predictors of Maternal and Child Healthcare
Services Utilization in Nigeria
authors:
- Josephine Aikpitanyi
- Friday Okonofua
- Lorretta Ntoimo
- Sandy Tubeuf
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012786
doi: 10.3389/frhs.2022.847721
license: CC BY 4.0
---
# Locus of Control and Self-Esteem as Predictors of Maternal and Child Healthcare Services Utilization in Nigeria
## Abstract
This study investigated the influence of locus of control and self-esteem on the utilization of maternal and child healthcare services in Nigeria. Specifically, it explored the differences in utilization of antenatal care, skilled birth care, postnatal care, and child vaccination by women having internal and external locus of control and women having high and low self-esteem. It also examined the association between utilization of maternal and child healthcare on other sociodemographic characteristics. We collected information on non-cognitive traits of 1,411 randomly selected women along with information on utilization of various indicators of maternal and child healthcare services. We estimated logistic regression models for various components of maternal and child healthcare services utilization and found that women's internal locus of control was a significant predictor of utilization of antenatal care, skilled birth care and completion of child vaccination. We also found that having a high self-esteem was a significant predictor of utilization of antenatal care, postnatal care and completion of child vaccination after adjusting for other control variables. By improving our understanding of non-cognitive traits as possible barriers to maternal and child healthcare utilization, our findings offer important insights for enhancing participants' engagement in intervention programs that are initiated to improve maternal and child health outcomes in lower-middle-income countries.
## Introduction
While motherhood is expected to be a positive and joyful experience, for too many women it is correlated with suffering, disabilities and even death. Every day, 800 women die from conditions associated with pregnancy and childbirth resulting in an estimated 300,000 maternal deaths worldwide per year [1]. Over 99 percent of all cases of maternal deaths occur in lower-middle-income countries, with more than half of them occurring in Sub-Saharan Africa [2]. Women of childbearing age in Sub-Saharan Africa are challenged by significant medical and socio-demographic factors that translate to poor health outcomes including high maternal morbidity and deaths [3]. Nigeria is the most populous nation in Africa and as a result, bears a high burden of worsening maternal and child health outcomes with a maternal mortality rate of 917 per 100,000 live births and child mortality rate of 117 per 1,000 live births [4]. Major indicators of maternal and child healthcare services utilization such as antenatal care attendance, skilled assistance during childbirth, postnatal check, and child vaccination show disparities across the country. These within-country disparities cut across states, geographical regions and socioeconomic status with current statistics indicating that $77\%$ of women in rural areas do not utilize skilled birth care during childbirth when compared with $37\%$ of women in urban areas [5].
Findings from empirical research in several lower-middle-income countries show that avoiding maternal deaths and complications is possible when women have unhindered access to antenatal care, skilled assistance during childbirth, and postnatal care after delivery [5]. Findings also show that a myriad of childhood diseases and deaths could be avoided by completing the required child vaccinations [6]. Several factors have been identified as barriers to utilization of maternal and child healthcare services in lower-middle-income countries with a large proportion of studies citing lack of finances as obstacles [7, 8]. This is evident in the lack of adequate healthcare financing mechanisms, resulting in high out-of-pocket costs of care. Other barriers include lack of transportation to healthcare facilities, low quality of care, lack of trust in healthcare providers, and non-availability of healthcare facilities [9].
There have been various programmes initiated to improve maternal and child health outcomes in Nigeria. These comprise components from both supply and demand of healthcare. The supply-side is focused on expanding access to quality maternal healthcare services and improving maternal and child health outcomes through the deployment and training of skilled midwives and community health workers, increasing supplies and medicines to healthcare centers, improving infrastructural development, and supporting the creation of ward development committees in rural communities [10]. The demand-side is focused on increasing the utilization of healthcare services during pregnancy and childbirth by providing conditional cash transfers to pregnant women at public primary healthcare facilities [11]. While these initiatives have recorded significant successes, there have been instances where the desired effects were not achieved as studies have reported that a share of women who received intervention packages, including conditional cash transfers in low resource settings continued not utilizing skilled maternity care. For instance, Baba-Ari et al. in a study to identify factors that influenced the uptake of maternal healthcare interventions found that provisions of cash transfers were not sufficient to increase uptake of maternal healthcare services in Northern Nigeria [12]. McConnell et al. also found that interventions to reduce financial barriers were not enough to encourage uptake of skilled birth care in Kenya [13]. Similarly, Cohen et al. found that even with the provision of cash transfers, many women in Kenya continued using poor-quality healthcare facilities for childbirth [14].
Several studies have sought to explain this phenomenon. For example, Stoop et al. reported that low levels of trust in local and national authorities were important contributors to vaccine hesitancy and hindered the uptake of child vaccinations in lower-middle-income countries [15]. Smith et al. in a study to investigate the causes and covariates of antenatal care access identified late recognition of pregnancy as one of the major reasons for delayed antenatal care attendance among pregnant women in South Africa [16]. Other studies have cited lack of decision-making power on the part of women, perceived low quality of care, women's preference for cultural, traditional, or religious forms of care, and fear of cesarean sections as reasons for non-utilization of maternal healthcare services in low resource settings [17, 18]. Extending these past studies, we postulate that the non-utilization of maternal and child healthcare services in low resource settings could be related to intrinsic behavioral and non-cognitive traits that hinder women's ability to make rational decisions to improve their health and welfare.
While the study of cognitive traits on health-promoting behaviors has assumed increasing importance over the past years, very little is known about the influence of non-cognitive traits on health-promoting behaviors of reproductive age women. Empirical research from psychology and economics shows that non-cognitive traits are important predictors of individuals' economic and social outcomes [19]. Experts are of the opinion that while the role of cognition in information processing, learning, and decision-making is essential, other non-cognitive traits also matter for attaining better life outcomes [19]. The origin of the study of non-cognitive traits lies in the earlier works of sociologists Samuel Bowles and Herbert Gintis [20] in their research on the determinants of education in America. They used the phrase to distinguish factors other than those measured by cognitive tests such as the ability to read, write and apply simple numerical concepts [21]. These included a wide range of traits such as empathy, resilience, locus of control, self-esteem, and the Big Five personality traits [21].
In this paper, we focused on locus of control and self-esteem because of their reference in empirical studies as important predictors of health outcomes. Economists have studied the effects of locus of control and self-esteem on socioeconomic and health-related outcomes such as dieting and engaging in physical exercises and have reported significant associations [22, 23]. Locus of control is described as one of the most researched psychological concepts that exert a great impact on the decision-making ability of an individual. This could be reflected in the utilization of healthcare services or in any other field of life. Developed by the psychologist Julian Rotter in his social learning theory of personality, locus of control examines how a person perceives the relationship between his or her behavior and the expected reward [24]. People who believe that they are in control of their destinies have an internal locus of control and are referred to as “internals” while people with external locus of control believe that their fates are determined by luck, chance, or powerful others and are referred to as “externals” [24].
Self-esteem involves self-evaluation followed by an emotional reaction toward oneself. Researchers describe self-esteem as a combination of a person's self-assessment and their self-concept of characteristics and abilities [25]. In the last two decades, researchers have studied self-esteem in relation to problems such as drug abuse, unemployment, crime, and violence. Recent studies, have related the concept to other aspects, such as health and well-being, academic success, and learning abilities [26]. In psychology, self-esteem has gained increased attention, because several studies indicate its role in important life outcomes, such as physical health, interpersonal relationships, and psychopathology [27].
Given the high rate of maternal and child morbidity and mortality in lower-middle-income countries and the large disparities in the utilization of healthcare services, examining the influence of locus of control and self-esteem on utilization of maternal and child healthcare services becomes relevant. With renewed interest in improving health outcomes exacerbated by the emergence of more infectious diseases, empirical evidence accumulates regarding the impact of psychological behaviors on health and the growing need to understand the predictors of health-promoting behaviors in low resource settings. More knowledge is needed to understand the relationship between non-cognitive variables such as locus of control and self-esteem in the utilization of maternal and child healthcare services among reproductive age women in low resource settings. The objective of this paper is to contribute to the emerging field of behavioral research and empirically evaluate whether and how locus of control and self-esteem act as predictors of the utilization of maternal and child healthcare services in Nigeria.
The rest of the paper proceeds as follows. Section 2 provides a summary of previous literature on locus of control and self-esteem. Section 3 describes our study design, empirical strategy, and key features of the chosen outcome, explanatory and control variables, along with their measurements. Section 4 presents our results. In the following section, we utilize regression models to find possible associations between locus of control and self-esteem and utilization of antenatal care, skilled birth care, postnatal care, and completion of child vaccination and we comment on the associated results. Section 5 discusses findings from the study and concludes with some final remarks.
## Internal and External Locus of Control
Locus of control is described as a psychological concept that captures “a generalized attitude, belief or expectancy regarding the nature of the causal relationship between one's own behaviour and its consequences” [24]. Individuals hold beliefs on whether life outcomes are a result of their efforts or the result of fate, chance, luck, or the intervention of others that they regard as more powerful than they are. Individuals who believe that outcomes are due to their efforts have an “internal” locus of control while individuals who believe that outcomes are due to luck have an “external” locus of control [24]. Locus of control is perceived as a stable trait and is described as one of the most enduring ideas in behavioral research and theory [28]. People who have external locus of control attribute life outcomes to external forces such as chance, luck, fate, destiny, or powerful others, while those with internal locus of control take responsibility for their successes and failures [24].
Rotter hypothesized that individuals develop a sense of control when they perceive reinforcement as based on their behavior, with behaviors that result in reinforcement serving to strengthen their perception of control [24]. However, when there is a failure of reinforcement, a reverse effect occurs that weakens or diminishes an individual's perception of control. Carton and Nowicki reviewed the research on the antecedents of individual differences in locus of control and reported that parents influenced the development of their children's locus of control [29]. They found that consistent use of reward and punishment as well as encouragement of autonomy by parents were associated with the development of internal locus of control. They also found that children who had less supportive parents and those who had experienced stressful and disruptive life events had external locus of control [29]. Specifically, they reported that individuals who were internally oriented in their locus of control tended to exhibit greater productivity, motivation and initiative, and were more successful in life [29].
Empirical studies have described locus of control as an important predictor of different life outcomes, including healthy living and wellbeing, life satisfaction, educational attainment, employment and wages, and could act as a buffer against many negative life events that people may experience [23, 28, 29]. For example, Heckman et al. found that locus of control played a significant role in explaining risky behaviors of adolescents and young adults, including imbibing habits such as daily smoking, use of drugs, crime participation, and incarceration [30]. Lassi et al. in their study on the association between locus of control and tobacco and alcohol consumption among young adults in South West England also found that individuals with more external locus of control had higher odds of consuming more tobacco and alcohol when compared with individuals with more internal locus of control [31].
Cobb-Clark et al. in their study to find the connection between individuals' healthy habits and their locus of control in Australia reported that people with internal locus of control tended to invest more in their health by imbibing healthier lifestyles [22]. Gale et al. also found that individuals with internal locus of control had reduced risk of obesity and reported higher satisfaction with their health [32]. Kesavayuth et al. in their study to find the impact of locus of control on healthcare utilization in Australia also found that people with internal locus of control were healthier than those with external locus of control were and relied less on both preventive and curative medical care [23].
## High and Low Self-Esteem
William James first covered the concept of self-esteem in 1890 [33]. He defined it as “a positive self-consideration obtained by people when they can consistently meet or exceed the important goals of their lives” [33]. Morris Rosenberg, the author of one of the oldest and most widely used self-esteem evaluation scales defined self-esteem as, “a positive or negative overall attitude towards oneself ” [34]. Murphy et al. related self-esteem to individuals' personal beliefs about social relationships, skills and abilities; they defined self-esteem as “a global barometer of self-evaluation involving cognitive appraisals about general self-worth and affective experiences of the self that are linked to these global appraisals” [35]. Scholars have adopted a two-way approach in the evaluation of self-esteem. This two-way approach divides self-esteem into the categories of high self-esteem, defined as “when a person feels that they have value for self ” and low self-esteem, characterized as “when a person believes that they have no self-value and, as a result, suffer from self-pity and self-contempt” [36].
An important consequence of the self-esteem situational variability is that individuals with different levels of self-esteem tend to apply opposite behavioral strategies in front of at-risk situations. While individuals with high self-esteem exhibit self-enhancing tendencies, individuals with low self-esteem tend to use self-protective strategies to avoid the attention of other people, and to conceal their inadequacies. Moreover, low self-esteem individuals are highly sensitive to defeats and interpersonal refusals, as they are often less able to use their mental resources as a defensive instrument to avoid dramatic fluctuations [37]. Empirical and theoretical studies over the last two decades have identified self-esteem as a powerful and significant psychological factor in individuals' overall quality of life, and in determining healthy living and well-being.
Studies have found that feeling worthy and empowered are related to having a high self-esteem, which can result in positive lifestyle changes and improve the tendency to have better health outcomes [38]. For example, Das and Pattanaik found that individuals with high self-esteem saw themselves as being more active and capable to influence their lives through personal efforts and the setting of higher life goals [39]. Similarly, Fuscaldi et al. reported that diabetic patients with higher self-esteem easily adopted healthier behaviors concerning the diagnosis and treatment of the disease when compared with patients with lower self-esteem [40]. Baumeister et al. also identified low self-esteem as a risk factor for aggression, depression, felony and lower educational outcomes [26]. These findings showed that individuals who regarded themselves highly and had high self-esteem were more likely to be involved in health-promoting behaviors, and that individuals who regarded themselves lowly and had low self-esteem were less likely to engage in health-promoting behaviors.
## Locus of Control and Self-Esteem
Several researches and theories suggest the existence of a strong relationship between locus of control and self-esteem. Empirical research on the association between locus of control and self-esteem have identified that locus of control was significantly related to self-esteem on the dimension of individuals' control ideology, self-blame and system blame, suggesting that having an internal locus of control was associated with having high self-esteem [37, 38]. For instance, Giblin et al. found that individuals that had internal locus of control also had high self-esteem [41]. Dielman et al. reported a similar relationship between attribution for outcome and self-esteem. They reported that individuals with high self-esteem attributed successful outcomes to internal causes, while individuals with low self-esteem attributed successful outcomes to external causes [42]. Alizadeh also found positive significant correlations between internal locus of control and self-esteem. They found that individuals with internal locus of control also had high self-esteem and had better physical and mental health, while individuals with external locus of control also had low self-esteem and were psychologically distressed and perhaps even depressed [43].
Recognizing the potential influence of locus of control and self-esteem on health and well-being raises important questions about the variability of both measures in any given population. From a theoretical angle, researchers have found that both traits were shaped within the context of the social environment and were likely to be formed by an individual's position in the social structure [38, 39]. While several of these past studies have been based on finding the influence between locus of control and self-esteem on health outcomes and other aspects of healthcare, we did not find any study on the influence of these traits on utilization of skilled maternal and child healthcare services. Our study becomes very relevant in being one of the first to find the relationship between these traits and utilization of maternal and child healthcare services with recommendations for designing effective interventions, especially in low resource settings.
## Study Design and Population
We used data from an intervention project aiming to increase women's access to skilled maternal healthcare services that was conducted in two rural Local Government Areas (Esan South East and Etsako East) in Edo State, southern Nigeria between August 2017 and June 2020. The two Local Government Areas have a total population of 455,432 persons, with Esan South East accounting for 241,492 and Etsako East LGA accounting for 213,940 [44]. Both LGAs are located in the rural and riverine areas of the state, adjacent to River Niger, with Estako East in the northern part of the Edo State part of the river, while Esan South *East is* in the southern part. Administratively, each LGA comprises of 10 political/health wards and there are several communities in each ward. Nigeria operates a three-tier health care system with primary healthcare as the entry into the health system [45]. The principal sources of maternal and child healthcare in the two local government areas are primary healthcare centers. The sampling technique used to select communities and respondents in each local government area is presented in detail in a previous publication [46]. A sample of 20 communities was targeted and a sample of 1411 women of reproductive age (15 – 45 years) that gave birth in two years prior to the survey were selected to participate in the project. Trained research assistants administered questionnaires through face-to-face interviewing of respondents.1 The questions were fielded in English or Pidgin English as appropriate.
The intervention consisted in providing financial assistance to pregnant women in form of a community-based health insurance scheme that subsidized the cost of delivery by $80\%$. Additionally, women in the programme were provided with mobile phones with dedicated SIM cards linked to health care providers and transporters to ease access to health care facilities. Advocacy visits were made to relevant authorities to ensure that health care providers (doctors and nurses) were posted to the various health care centers and arrangements were made with local transporters in collaboration with the health care providers to transport women to health care facilities freely and when they called. Although the intervention activities were available to all women of reproductive age within the community, financial assistance for childbirth and mobile phones with dedicated SIM cards were given to only women that were registered in the community-based health insurance scheme.
We utilized two data collection points. The baseline questionnaire carried out before the intervention in 2017 consisted of pre-validated questions adapted from the Nigeria Demographic and Health Survey [47]. It included questions on women's socio-demographic characteristics, partners' and other family characteristics, reproductive history, and antenatal, intrapartum, and postnatal care experience for current pregnancy and births in the preceding years. The baseline questioned women on reasons for use, non-use of primary health centers for maternal, and child healthcare services and identified barriers to utilization of maternal healthcare services from a list of factors. These included financial constraints, lack of transportation to healthcare facilities, inadequate number of trained health care providers, and lack of drugs and other needed consumables at healthcare facilities [48]. In addition to the repeated baseline survey questionnaire conducted at follow-up in July 2020, we added a set of standardized questions on locus of control and self-esteem to investigate whether and how women's non-cognitive traits mattered for the utilization of maternal and child healthcare services. In this paper, we utilized data from the follow-up survey, which contained information on women's locus of control and self-esteem.
## Empirical Strategy
To understand the theoretical aspects of individuals' healthcare-seeking behavior, we looked at the Grossman model of demand for health [49]. This model was constructed within a human capital framework and categorizes health as a durable capital stock that depreciates over an individual's life cycle and could be increased by investing in health inputs such as consuming medical services and imbibing a healthy lifestyle (exercise, recreation, diet). Grossman opined that individuals demanded “good health” for two reasons. First, to increase the number of healthy days that allowed for market and non-market activities (here, good health is considered as an investment good). Second, to improve welfare or utility (here, good health is seen as a consumption good).
In the Grossman model, demand for healthcare services is derived from the demand for health. This implies that factors, which influence demand for good health also influence demand for medical services and healthy lifestyles. Thus, individuals do not demand medical services for their own sake, but as a means to attain good health [49]. Individuals consume healthcare services in order to improve their health status, so the cost of gaining good health is a reduction in the consumption of other goods and they could choose the type of healthcare to consume at any point in time using available information, by weighing the costs of utilization to the perceived benefits of utilizing healthcare services. In our study, we proposed that a pregnant mother might consume healthcare services, not just to improve her own health but also to improve that of her unborn child. This could be achieved through early registration for antenatal care and attending at least four antenatal care visits as recommended by the World Health Organization [50], utilizing the services of a skilled birth attendant during childbirth, receiving postnatal care within 48 h of childbirth and completing the required child vaccinations. Thus, not consuming these healthcare services might result in worsened health outcomes for both mother and child.
In the Grossman model, the role of cognitive skills is specified through its analysis of the effect of education on health outcomes. In this study, we argued that the claims made about education levels were applicable to non-cognitive traits as well. This is in line with the findings of Chiteji in a study conducted in the USA on the influence of non-cognitive skills in encouraging healthy behavior. The study suggested that non-cognitive traits might raise the efficiency of household production just as education does in the standard Grossman model [51]. We based our assumptions that in a country setting such as that of Nigeria, where women occupy marginal positions in society, their healthcare-seeking behavior might be affected by a myriad of psychological, physical and behavioral factors. Thus, we analyzed the demand-side factors, which are expected to influence healthcare service utilization, and provide a general perspective based on which, we might formulate hypotheses regarding which variables affect maternal and child healthcare-seeking behavior. In this study, our intent was to establish the influence of locus of control and self-esteem as demand-side predictors of maternal and child healthcare services utilization. The outcome variables, yi, thus took a binary form, where; When dealing with binary responses, it is not appropriate to use ordinary least squares models because the standard assumptions are not satisfied. Instead, non-linear binary choice models such as logit and probit models are more suitable for analysis [52]. Logit and probit models give very similar results in empirical work, but since most previous research of healthcare utilization uses the logit model, we followed the same line in this study. The probability that a given individual utilized healthcare services, i.e., that the outcome variable took the value 1, given a set of explanatory variables, could be written as This probability could be transformed into; which indicated how often an event (yi = 1; the individual utilized health care) occurred, relative to how often it did not occur (yi = 0; the individual did not utilize health care).
Following the theoretical framework, we employed logit regression models to assess the association between non-cognitive traits and utilization of maternal and child healthcare. For each of the four indicators of healthcare utilization yi, we estimated models of the form: Where i indicated the individual, T indicated traits (locus of control and self-esteem) and X indicated all other control variables, as described in Table 1. We presented results as beta coefficients, with standard errors. With four outcomes and two non-cognitive traits, we adjusted our results for multivariate regressions to enable us identify as many significant associations as possible.
**Table 1**
| Variables | Descriptions |
| --- | --- |
| Outcome variables | Outcome variables |
| Antenatal care | This was derived from response to the question “Did you see anyone for antenatal care when you were pregnant with your last child?” It was computed as “0” = 'No' and “1” = “Yes”. |
| Skilled birth care | This was derived from response to the question “Who assisted with the delivery of your last child?” It was computed as “0” = No skilled birth attendant (doctor, trained nurse or midwife) during last childbirth and “1” = Skilled birth attendant during last childbirth. |
| Postnatal check | This was derived from the question “Did anyone check on your health within 48 h after the delivery of your last child?” It was computed as “0” = “No” and 1 = “Yes”. |
| Child vaccination | This was derived from the question “Did you complete the vaccination for your last child?” It was computed as “0” = “No” and “1” = “Yes”. |
| Explanatory variables | Explanatory variables |
| Locus of control | Locus of control was derived from the responses to seven questions. Answers were reported on a 5-point scale that ranged from 1 = strongly disagree to 5 = strongly agree |
| Self-esteem | Self-esteem was derived from the responses to 10 questions. Answers were reported on a 5-point scale that ranged from 1 = strongly disagree to 5 = strongly agree |
| Sociodemographic characteristics | Sociodemographic characteristics |
| Age | Grouped in categories of 15–24, 25–34, 35–44 and >44 |
| Education | Grouped as no education, primary education, secondary education and higher education |
| Marital status | Grouped as married and not married |
| Occupation | Grouped as working and not working |
| Literacy | Grouped as cannot read, can read a bit and can read well |
| Religion | Grouped as Christianity, Islam and Others |
| Household characteristics | Household characteristics |
| Family type | Grouped as monogamy and polygamy |
| Partner's age | Grouped in categories of 18–37, 38–57, 58–77 and >77 |
| Partner's education | Grouped as no education, primary education, secondary education and higher education |
| Partner's occupation | Grouped as working and not working |
| Decision on healthcare | Grouped as Respondent herself, Partner, Both and Others |
| Payment for healthcare | Grouped as Respondent herself, Partner, Both and Others |
| Previous pregnancy complications | Yes = If respondent had complications in previous pregnancies and No = If respondent had no previous pregnancy complications |
We followed Kesavayuth et al. and Buddelmeyer and Powdthavee that recommended regressing non-cognitive traits on sociodemographic variables. We regressed non-cognitive traits T (locus of control and self-esteem) on women's sociodemographic variables that were included as control variables in the model in order to determine the sociodemographic factors that influenced locus of control and self-esteem using a linear regression model [23, 53].
Where X1 was a vector of women's sociodemographic variables and e the random error term.
## Measures of Healthcare Services Utilization
Our outcome variables included various indicators of maternal and child healthcare utilization. In total, we examined four indicators (antenatal care, skilled birth care, postnatal care, and completion of child vaccination). Our measure of antenatal care differed from the World Health Organisation's recommendation of having “at least four antenatal care visits” [50]. This was because a large number of women in our study population were unable to provide information on the number of times that they received antenatal checks. As a result, we collected data on antenatal care as a binary variable in response to the general question “Did you see anyone for antenatal care during your last pregnancy?” the variable was coded as “0” if the response was “No” and “1” if the response was “Yes”.
Utilization of skilled birth care was derived as a response to the question “Who assisted with the delivery of your last child?” It was coded as “0” if no skilled birth attendant (doctor, trained nurse or midwife) was present and “1” if a skilled birth attendant was present. Utilization of postnatal care was derived from the question “Did anyone check on your health within 48 hours after the delivery of your last child?” It was computed as “0” if the response was “No” and “1” if the response was “Yes”. Completion of child vaccination was derived from the question “Did you complete the vaccination of your last child?” It was computed as “0” if the response was “No” and “1” if the response was “Yes”. Table 1 gives a summary description of all outcome, explanatory, and control variables.
## Measures of Non-cognitive Traits
The explanatory variables of interest were “locus of control” and “self-esteem”. We asked participants all seven of the original items from the Psychological Coping Resources component of the Mastery Module developed by Pearlin and Schooler. Mastery refers to beliefs about the extent to which life's outcomes are under one's own control using the following statements “(a) I have little control over the things that happen to me; (b) *There is* really no way I can solve some of the problems I have; (c) *There is* little I can do to change many of the important things in my life; (d) I often feel helpless in dealing with the problems of life; (e) Sometimes, I feel that I am being pushed around in life; (f) What happens to me in the future mostly depends on me; (g) I can do just about anything I really set my mind to do;” along with answers on a Likert scale from 1 to 5 [54].
Specifically, respondents were asked the extent to which they agree with the seven statements. Possible responses ranged from 1 (strongly disagree) to 5 (strongly agree). In line with Cobb-Clark et al. and Kesavayuth et al., which recommended generating an index variable for locus of control, we generated an index variable for locus of control using a factor analysis method. Factor analysis is a statistical technique that linearly transforms an original set of variables into a substantially smaller set of uncorrelated variables that represent most of the information in the original set of variables. An internal test of consistency yields a Cronbach's reliability statistic of 0.78 indicating that the seven items were highly reliable [55]. *We* generated a single index for the locus of control variable by reversing the scores of the responses to questions 1 through 5, and then adding the scores of questions 6 and 7. The total score thus ranged from 5 to 35 with higher values indicating a more internal locus of control.
We measured self-esteem with the scale developed by Morris Rosenberg in 1965 [33]. It consists of 10 items including the following statements “(a) I feel that I am a person of worth; (b) I feel that I have a number of good qualities; (c) I am inclined to feel that I am a failure; (d) I am able to do things as well as most people; (e) I feel that I do not have much to be proud of; (f) I take a positive attitude toward myself; (g) On the whole, I am satisfied with myself; (h) I certainly feel useless at times; (i) I wish I could have more respect for myself; (j) At times, I think that I am no good at all.” Respondents were asked the extent to which they agreed with the ten statements with possible answers on a 5-point scale from 1 (strongly disagree) to 5 (strongly agree). Five statements were positively scored whereas the rest five statements were negatively scored. Following Kaplan and Pokorny, and Das and Pattanaik we generated an index score for self-esteem using a factor analysis method [39, 56]. We reversed the negative scores and added them to the positive scores to generate a self-esteem index with higher scores indicating higher self-esteem. A *Cronbach alpha* score of 0.70 indicates reliability of the scale. The Rosenberg self-esteem scale is considered reliable and a valid tool for quantitative self-esteem assessment [26, 35, 37].
## Control Variables
In analyzing the influence of non-cognitive traits on maternal and child healthcare utilization, it is important to control for other differences in observable characteristics that may influence the use of healthcare services. The Anderson framework, which highlights the importance of predisposing, enabling and health need variables in explaining healthcare utilization [57], was useful for this purpose. We therefore adopted the Anderson framework in choosing appropriate control variables; predisposing factors include indicators for women's age, education level, marital status, level of literacy, religion and parity. Enabling characteristics included household characteristics such as whether a woman was in a monogamous or polygamous union, partners' age, partners' education level, partners' occupation, ability to make decisions on healthcare use and ability to pay for healthcare. The indicator of “health need” included having had a previous pregnancy complication. We checked for correlations between indicators before including them in the models.
## Data Analysis
We coded the obtained data and entered them into Stata 17.0 for Windows. We calculated means and standard deviations for continuous variables (locus of control and self-esteem). We also calculated frequencies and percentages for categorical variables. We performed standard descriptive analyses on the main outcome and explanatory variables. Next, we explored the correlations between the locus of control, self-esteem, antenatal care, skilled birth care, postnatal checks, and child vaccination and examined whether multicollinearity was an issue using variance inflation factors. Computed factors ranged between 1.05 and 2.20, which was sufficiently low to assume that it would not significantly affect coefficient estimates. We examined the association between non-cognitive traits (locus of control and self-esteem) and the four outcomes of the utilization of maternal and child healthcare using Chi-square tests. To begin our analysis, first, we estimated two linear regression models to determine the sociodemographic characteristics of women that were associated with locus of control and self-esteem. We then fitted four logit regression models to examine the associations between locus of control and self-esteem and the measures of maternal and child healthcare services utilization.
## Results
In Table 2, we presented the descriptive statistics (number of observations, proportion and range of sum score). Our findings revealed that $13.7\%$ of women in the study sample did not see a skilled healthcare provider for antenatal care visits, $11.9\%$ did not utilize skilled birth care, 7.4 % did not have postnatal check after childbirth and 5.0 % did not complete vaccinations for their children. However, utilization of various indicators of skilled maternity and child healthcare were higher in the study population than that of the national average.2 The correlation matrices for all these measures of maternal and child healthcare utilization, locus of control, and self-esteem are shown in Table 3. We noted a positive but medium correlation between locus of control and self-esteem (0.23). Correlations between locus of control and the measures of maternal and child healthcare utilization were negative and tended to be low (0.05–0.17). Similar negative and low correlations were found between self-esteem and the measures of maternal and child healthcare services utilization (0.08–0.02). We found medium to high correlations within the measures of maternal and child healthcare services utilization (0.34–0.85). The highest correlation was between utilization of antenatal care and skilled birth care services (0.85).
In Table 4, we presented the descriptive characteristics of women based on their locus of control and self-esteem. We found that women with external locus of control were less likely to utilize the various components of maternal and child healthcare (13.2, 11.6, 6.8, and 4.6 %) when compared to women with internal locus of control (0.6, 0.7, 0.7, and 0.4 %). We also found similar results for self-esteem. Women with low self-esteem had lower likelihoods of utilizing antenatal care, skilled birth care, postnatal care, and completing their children's vaccinations (11.0, 9.6, 6.3, and 9.6 %) when compared with women with high self-esteem (2.7, 2,6, 1.3, and 0.8 %).
**Table 4**
| Variables | Locus of control | Locus of control.1 | Locus of control.2 | Self-esteem | Self-esteem.1 | Self-esteem.2 |
| --- | --- | --- | --- | --- | --- | --- |
| | Internal | External | p-value | High | Low | p-value |
| | N (%) | N (%) | | N (%) | N (%) | |
| | 157 (11.13) | 1,254 (88.87) | | 333 (23.60) | 1,078 (76.40) | |
| Antenatal | | | 0.001 | | | 0.153 |
| No | 8 (0.59) | 178 (13.18) | | 37 (2.74) | 149 (11.03) | |
| Yes | 144 (10.66) | 1021 (75.57) | | 288 (21.32) | 877 (64.91) | |
| Skilled birth | | | 0.010 | | | 0.338 |
| No | 9 (0.66) | 157 (11.55) | | 35 (2.58) | 131 (9.64) | |
| Yes | 145 (10.67) | 1048 (77.12) | | 292 (21.49) | 901 (66.30) | |
| Postnatal | | | 0.611 | | | 0.069 |
| No | 10 (0.74) | 92 (6.77) | | 17 (1.25) | 85 (6.26) | |
| Yes | 144 (10.60) | 1112 (81.89) | | 310 (22.83) | 946 (69.66) | |
| Vaccination | | | 0.287 | | | 0.118 |
| No | 5 (0.37) | 63 (4.64) | | 11 (0.81) | 57 (4.20) | |
| Yes | 149 (10.97) | 1141 (84.02) | | 316 (23.27) | 974 (71.72) | |
| Age | | | 0.000 | | | 0.005 |
| 15–24 | 15 (1.06) | 251 (17.79) | | 43 (3.05) | 223 (15.80) | |
| 25–34 | 52 (3.69) | 524 (37.14) | | 134 (9.50) | 442 (31.33) | |
| 35–44 | 50 (3.54) | 354 (25.04) | | 109 (7.73) | 295 (20.91) | |
| >44 | 40 (2.83) | 125 (8.86) | | 47 (3.33) | 118 (8.36) | |
| Education | | | 0.000 | | | 0.254 |
| | 12 (0.85) | 165 (11.69) | | 43 (3..05) | 134 (9.50) | |
| Primary | 83 (5.88) | 503 (35.65) | | 132 (9.36) | 454 (32.18) | |
| Secondary | 42 (2.98) | 544 (38.55) | | 137 (9.71) | 449 (31.82) | |
| Higher | 20 (1.42) | 42 (2.98) | | 21 (1.49) | 41 (2.91) | |
| Marital status | | | 0.000 | | | 0.095 |
| Married | 60 (4.25) | 708 (50.18) | | 168 (11.91) | 600 (42.52) | |
| Not married | 97 (6.87) | 546 (38.70) | | 165 (11.69) | 478 (33.88) | |
| Occupation | | | 0.008 | | | 0.000 |
| Not working | 23 (1.63) | 303 (21.47) | | 110 (7.80) | 216 (15.31) | |
| Working | 134 (9.50) | 951 (67.40) | | 223 (15.80) | 862 (61.09) | |
| Literacy | | | 0.466 | | | 0.910 |
| Cannot read | 62 (4.39) | 534 (37.85) | | 144 (10.21) | 452 (32.03) | |
| Can read a bit | 62 (4.39) | 505 (35.79) | | 131 (9.28) | 436 (30.90) | |
| Can read well | 33 (2.34) | 215 (15.24) | | 58 (4.11) | 190 (13.47) | |
| Religion | | | 0.066 | | | 0562 |
| Christianity | 155 (10.99) | 1195 (84.69) | | 319 (22.61) | 1031 (73.07) | |
| Islam | 0 (0.00) | 42 (2.98) | | 8 (0.57) | 34 (2.41) | |
| Others | 2 (0.14) | 17 (1.20) | | 6 (0.43) | 13 (0.92) | |
| Parity | | | 0.483 | | | 0.013 |
| 1 | 117 (8.29) | 962 (68.18) | | 254 (18.00) | 825 (58.47) | |
| 2 | 38 (2.69) | 263 (18.64) | | 65 (4.61) | 236 (16.73) | |
| 3 | 2 (0.14) | 29 (2.06) | | 14 (0.99) | 17 (1.20) | |
| Family type | | | 0.000 | | | 0.684 |
| Monogamy | 129 (9.14) | 756 (53.58) | | 212 (15.02) | 673 (47.70) | |
| Polygamy | 28 (1.98) | 498 (35.29) | | 121 (8.58) | 405 (28.70) | |
| Partner's age | | | 0.016 | | | 0.019 |
| 18–37 | 49 (3.47) | 455 (32.25) | | 113 (8.01) | 391 (27.71) | |
| 38–57 | 94 (6.66) | 718 (50.89) | | 188 (13.32) | 624 (44.22) | |
| 58–77 | 14 (0.99) | 55 (3.90) | | 27 (1.91) | 42 (2.98) | |
| >77 | 0 (0.00) | 26 (1.84) | | 5 (0.35) | 21 (1.49) | |
| Partner's education | | | 0.000 | | | 0.015 |
| | 20 (1.42) | 194 (13.75) | | 62 (4.39) | 152 (10.77) | |
| Primary | 67 (4.75) | 312 (22.11) | | 100 (7.09) | 279 (19.77) | |
| Secondary | 47 (3.33) | 595 (42.17) | | 127 (9.00) | 515 (36.50) | |
| Higher | 23 (1.63) | 153 (10.84) | | 44 (3.12) | 132 (9.36) | |
| Partner's occupation | | | 0.569 | | | 0.000 |
| Not working | 23 (1.63) | 206 (14.60) | | 89 (6.31) | 140 (9.92) | |
| Working | 134 (9.50) | 1048 (74.27) | | 244 (17.29) | 938 (66.48) | |
| Decision on healthcare | | | 0.000 | | | 0.000 |
| Self | 32 (2.27) | 160 (11.34) | | 68 (4.82) | 124 (8.79) | |
| Partner | 44 (3.12) | 694 (49.18) | | 123 (8.72) | 615 (43.59) | |
| Both | 81 (5.5.74) | 393 (27.85) | | 140 (9.92) | 334 (23.67) | |
| Others | 0 (0.00) | 7 (0.50) | | 2 (0.14) | 5 (0.35) | |
| Payment for healthcare | | | 0.000 | | | 0.000 |
| Self | 35 (2.48) | 191 (13.54) | | 73 (5.17) | 153 (10.84) | |
| Partner | 57 (4.04) | 834 (59.11) | | 180 (12.76) | 711 (50.39) | |
| Both | 65 (4.61) | 217 (15.38) | | 73 (5.17) | 209 (14.81) | |
| Others | 0 (0.00) | 12 (0.85) | | 7 (0.50) | 5 (0.35) | |
| Previous preg. comp | | | 0.000 | | | 0.322 |
| Yes | 39 (2.76) | 219 (15.52) | | 67 (4.75) | 191 (13.54) | |
| No | 118 (8.36) | 1035 (73.35) | | 266 (18.85) | 887 (62.86) | |
In Table 5, we used linear regression to analyse the sociodemographic variables that predict locus of control and self-esteem. We found that locus of control was associated with age indicating that older women were more internal in their locus of control. Furthermore, we found that women who had internal locus of control were more educated, having at least secondary education, were employed and lived with a partner. Similarly, women with high self-esteem had more years of education, were employed, and lived with a partner. In the next step, we performed the regression analyses of the associations of the outcome variables with locus of control and self-esteem as described in equation [3].
**Table 5**
| Variables | Locus of control | Self-esteem |
| --- | --- | --- |
| | Coeff (S.E) | Coeff (S.E) |
| Age | Age | Age |
| 15–24 | Ref | Ref |
| 25–34 | −0.200 (0.076)** | −0.105 (0.033)*** |
| 35–44 | −0.349 (0.084)*** | −0.174 (0.036)*** |
| >44 | −0.635 (0.103)*** | −0.199 (0.044)*** |
| Education | Education | Education |
| | Ref | Ref |
| Primary | −0.081 (0.087) | −0.015 (0.037) |
| Secondary | 0.114 (0.102) | −0.065 (0.044) |
| Higher | −0.444 (0.169)*** | −0.181 (0.072)** |
| Marital status | Marital status | Marital status |
| Married | Ref | Ref |
| Not married | −0.243 (0.053)*** | −0.046 (0.022)** |
| Occupation | Occupation | Occupation |
| Not working | Ref | Ref |
| Working | 0.315 (0.064)*** | 0.165 (0.027)*** |
| Literacy | Literacy | Literacy |
| Cannot read | Ref | Ref |
| Can read a bit | −0.207 (0.067)*** | 0.011 (0.029) |
| Can read well | −0.198 (0.096)* | 0.040 (0.041) |
| Religion | Religion | Religion |
| Christianity | Ref | Ref |
| Islam | 0.094 (0.153) | 0.015 (0.065) |
| Others | −0.219 (0.225) | −0.074 (0.096) |
| Parity | Parity | Parity |
| 1 | Ref | Ref |
| 2 | −0.085 (0.066) | 0.012 (0.028) |
| 3 | 0.040 (0.180) | −0.192 (0.077) |
| Family type | Family type | Family type |
| Monogamy | Ref | Ref |
| Polygamy | 0.083 (0.054) | 0.015 (0.023) |
| No of obs. | 1411 | 1411 |
| R-Squared | 0.0711 | 0.0489 |
| Adjusted R-Squared | 0.0612 | 0.0387 |
We used four logit regression models to examine the associations between locus of control and self-esteem on the utilization of antenatal care, skilled birth care, postnatal care, and child vaccination, and presented our results in Tables 6, 7. In Table 6, we found negative coefficients of locus of control for all outcome variables indicating an inverse relationship between locus of control and utilization of various components of maternal and child healthcare services. This meant that women with more external locus of control had lower likelihood of utilizing maternal and child healthcare services when compared with women with more internal locus of control. The estimated coefficients on locus of control were statistically significant in the utilization of antenatal care, skilled birth care, and child vaccination. We found the coefficient of locus of control not statistically significant in the utilization of postnatal care. In Table 7, we also found an inverse relationship between self-esteem and utilization of various components of maternal and child healthcare services as shown by the negative coefficients indicating that women with low self-esteem were less likely to utilize maternal and child healthcare services when compared with women with high self-esteem. We found this reflected in the estimated coefficients of self-esteem on utilization of postnatal care and completion of child vaccination, as they showed statistically significant coefficients. We however found the estimated coefficients of self-esteem not statistically significant on utilization of antenatal care and skilled birth care after adjusting for other control variables.
In Table 8, we showed the marginal effects of locus of control and self-esteem on the probability of utilizing maternal and child healthcare services, computed at the means of the main explanatory variables. We found that utilization of antenatal care, skilled birth care, postnatal care and completion of child vaccination were negatively associated with locus of control and the utilization of antenatal care and skilled birth care showed significant associations. This confirmed our earlier findings and showed that women with external locus of control were 9.5 percent points less likely to utilize antenatal care and 7.0 percent points less likely to utilize skilled birth care. We also found negative associations between antenatal care, skilled birth care, postnatal care and completion of child vaccination and self-esteem. These associations were significant for utilization of antenatal care, skilled birth care and postnatal care. In comparison with women with high self-esteem, women with low self-esteem were 2.7 percent points less likely to utilize antenatal care, 1.6 percent points less likely to utilize skilled birth care and 3.0 percent points less likely to utilize postnatal care.
**Table 8**
| Unnamed: 0 | Antenatal care | Skilled birth care | Postnatal care | Child vaccination |
| --- | --- | --- | --- | --- |
| | Coeff (std. err) | Coeff (std. err) | Coeff (std. err) | Coeff (std. err) |
| Locus of control | −0.095*** (0.021) | −0.070*** (0.021) | −0.009 (0.021) | −0.018 (0.016) |
| Self-esteem | −0.027** (0.021) | −0.016** (0.203) | −0.030*** (0.015) | −0.020 (0.012) |
## Discussion
Our study sought to investigate the influence of locus of control and self-esteem on the utilization of maternal and child healthcare services in two rural Local Government Areas in Nigeria. Investigation of the relationship between health-related outcomes and non-cognitive traits such as locus of control and self-esteem is one aspect of improving health behavior that has gained increasing acceptance in recent years, especially in high-income countries [19, 27]. Previous studies, which have examined the influence of locus of control and self-esteem on health-related outcomes and lifestyles, have proved their usefulness in the establishment of priorities and allowed intervention programs to focus on potentially modifiable factors in health-seeking behavior.
Our study found significant associations between locus of control and self-esteem on utilization of maternal and child healthcare services in the study population. We found statistically significant associations between locus of control and utilization of antenatal care, skilled birth care, and completion of child vaccination. We found that women with external locus of control were less likely to utilize maternal and child healthcare services when compared with women with internal locus of control. This was consistent with the findings of Kesavayuth et al. in a study on the impact of locus of control in healthcare utilization in Australia, that as locus of control tended toward externality, utilization of healthcare services tended to decline [21]. Our study also found statistically significant associations between self-esteem and utilization of postnatal care and completion of child vaccination. We found that women who had low self-esteem were less likely to utilize the various components of maternal and child healthcare services. This finding was, however, not statistically significant in the utilization of antenatal care and skilled birth care.
With the low utilization of maternal and child healthcare services, much still needs to be done to improve maternal and child health outcomes in Nigeria. It is important to give special attention to utilization of healthcare services in underserved regions of the country such as rural communities. Since there was a significant association between non-cognitive traits and utilization of maternal and child healthcare services in our findings, intervention programs should take cognisance of the fact that this dimension of intrinsic barriers exists and channel effort into identifying the level and orientations of non-cognitive traits in the study participants. This could be done through the establishment of counseling clinics, where the psychological needs of women are identified and addressed confidentially, as issues of psychology and mental health are rarely discussed in many lower-middle-income countries because of the fear of stigmatization [58].
The role of professional counseling in the design of interventions aimed at improving maternal and child healthcare outcomes cannot be overemphasized, especially in low resource settings. For instance, Zeligman et al. in a study of the impact of locus of control on trauma survivors in a university in South Eastern USA found that professional counselors were able to consider individuals' perception of control, their ability to take responsibilities for their actions as well as how they perceived problems and obstacles [59]. Another aspect that could be useful in the design of health interventions is the fact that locus of control and self-esteem could be impacted by cultural, religious and environmental factors. For women with external locus of control, interventions to improve healthcare utilization could be channeled through people that have been identified in the various communities as having strong influences and acting as “powerful others” to women. These could be partners/husbands, mothers-in-law, and religious, community, or traditional leaders.
Similar to other studies, we found level of education significantly associated with both locus of control and self-esteem and significantly associated with the utilization of antenatal care and skilled birth care. In this regard, governments at all levels could intensify efforts on increasing the proportion of educated women in various communities, regardless of their socioeconomic or sociodemographic status. This is possible through putting initiatives in place to encourage the enrolment of young girls in school and committing more resources to enhancing adult education in order to give uneducated women the opportunity of receiving formal education. In addition, regular health education sessions on the importance of utilizing skilled maternity care during pregnancy and childbirth should be incorporated. Women should be educated on the necessity of early commencement of antenatal checks, knowing the required number of antenatal care visits and on the importance of completing the required vaccinations for their children. There is also need for stakeholders in the Federal and State Ministries of Health to engage in health promotion and awareness campaigns before and during health intervention programmes and to embark on regular and systematic evaluation and review of national programmes initiated to improve maternal and child healthcare utilization in lower-middle-income countries. Such reviews should be done for assessing not only the effectiveness of the programme as a whole but also effectiveness of the various components.
Although our study did not involve experiments based on affecting behavioral change, understanding human behavior is central to enacting effective policies. The results of this study make significant contributions for reforming the demand-side components of health systems and improving maternal and child health outcomes in Nigeria as it provides new insights into the influence of locus of control and self-esteem as predictors of utilization of maternal and child healthcare services. Low utilization of maternal and child healthcare services, especially in rural communities, is one of the most important challenges facing the country in its efforts to reduce the high rate of maternal and child deaths. Clearly, policies and programmes based on the initiation of psychological counseling units for women and provision of need-tailored interventions through good developmental planning and adequate budgetary allocations that target these specific barriers are critical for the country in its bid to improve women's access to skilled maternity care and reduce the number of maternal and child morbidities and mortalities.
The strength of our study lies on being one of the first to contribute to the behavioral economics literature in lower-middle-income countries by examining the influence of non-cognitive traits such as locus of control and self-esteem on utilization of maternal and child healthcare services in a study population of Nigerian women in selected rural communities. Our main results have important implications for creating effective policy actions. Specifically, women with a strong internal locus of control are more likely to utilize skilled maternal and child healthcare services, while those with an external locus of control might need extra incentives to reach similar uptake of healthcare services utilization. While we cannot claim causality from our estimates, locus of control and self-esteem are nevertheless interesting predictors of maternal and child healthcare utilization in low resource settings. We therefore recommend the inclusion of standardized questions on non-cognitive traits into national surveys in lower-middle-income countries to facilitate more research in this field of knowledge.
Our study has some limitations. One of such limitations was that some of the healthcare outcomes were binary. More knowledge of the relationship between maternal and child healthcare utilization and non-cognitive traits would be possible if more information on the outcome variables were available in the dataset. For example, knowing the frequency and timing of antenatal care visits of women in our study population would have allowed for a more precise understanding of the relationship between the outcome variable and non-cognitive traits. In addition, data on locus of control and self-esteem were collected only at the follow-up survey. While it would be better to have collected data on the variables at both baseline and follow-up periods to enable us determine whether there was a change in utilization of maternal and child healthcare services with regards to these traits, our results remain valid due to the assumption of stability in non-cognitive traits as found in previous studies [19, 28]. In our study, we aimed to know why women had not utilized skilled maternal and child healthcare services, despite it being available. Hence, the inclusion of standardized questions on women's non-cognitive traits in the follow-up questionnaire. However, taking into account the stability of locus of control over time as found in the study of Cobb-Clark and Schurer [28] and that of Mendolia and Walker [19] on the stability of non-cognitive traits, we might infer that our participants' locus of control and self-esteem remained constant at both data collection points and this justifies our results.
The main objective of our study was to draw awareness to locus of control and self-esteem as significant predictors of maternal and child healthcare services utilization in low resource settings. Future research should build on the findings and recommendations from our study and devote effort to implementing and translating the knowledge from experimental interventions, aimed at changing health behavior into a broader understanding of the political and policy implications of behavioral economics, particularly in lower-middle-income countries.
## Data Availability Statement
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.
## Ethics Statement
The studies involving human participants were reviewed and approved by National Health Research Ethics Committee (NHREC) of Nigeria—Protocol Number NHREC/$\frac{01}{01}$/2007–$\frac{10}{04}$/2017. Community leaders and household heads in the study settings granted the researchers permission to conduct the study. Participation was voluntary, and all study participants signed written informed consent. Rights to privacy and anonymity were respected throughout the study. The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
All authors listed have made a substantial, direct, and intellectual contribution to the work and approved it for publication.
## Funding
This study is supported by the Cooperation for Development PhD Scholarship, Université catholique de Louvain, and the Institute of Health Research and Society (IRSS), Université catholique de Louvain.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. 1.Organization WH. Trends in maternal mortality: 2000 to 2017: Estimates by WHO, UNICEF, UNFPA, World Bank Group and the United Nations Population Division.
Geneva: World Health Organisation. (2019).. *Trends in maternal mortality: 2000 to 2017: Estimates by WHO, UNICEF, UNFPA, World Bank Group and the United Nations Population Division.* (2019.0)
2. Say L, Chou D, Gemmill A, Tunçalp Ö, Moller A-B, Daniels J. **Global causes of maternal death: a WHO systematic analysis**. *Lancet Glob Health.* (2014.0) **2** e323-33. DOI: 10.1016/S2214-109X(14)70227-X
3. 3.Organization WH. Maternal Health Care: Policies, Technical Standards and Service Accessibility in Eight Countries in the Western Pacific Region. Geneva: World Health Organization (2018).. *Maternal Health Care: Policies, Technical Standards and Service Accessibility in Eight Countries in the Western Pacific Region* (2018.0)
4. 4.Worldbank. World Development Indicators. Washington, DC: World Bank (2018). Available online at: http://databank.worldbank.org/data/source/healthnutrition-and-population-statistics. (2018.0)
5. Izugbara CO, Wekesah FM, Adedini SA. *Maternal health in Nigeria: A Situation Update* (2016.0)
6. Tao W, Petzold M, Forsberg BC. **Routine vaccination coverage in low- and middle-income countries: further arguments for accelerating support to child vaccination services**. *Glob Health Action.* (2013.0) **6** 20343. DOI: 10.3402/gha.v6i0.20343
7. Banke-Thomas A, Abejirinde IO, Banke-Thomas O, Maikano A, Ameh CA. **Cost of maternal health services in low and middle-income countries: protocol for a systematic review**. *BMJ Open.* (2019.0) **9** e027822. DOI: 10.1136/bmjopen-2018-027822
8. Ekpenyong MS, Bond C, Matheson D. **Challenges of maternal and prenatal care in Nigeria**. *J Intensive Crit Care* (2019.0) **5** 1-7. DOI: 10.21767/2471-8505.100125
9. Ovikuomagbe O. **Determinants of maternal healthcare utilization in Nigeria**. *Afr Res Rev* (2017.0) **11** 283-94. DOI: 10.4314/afrrev.v11i2.21
10. Okonofua FE, Ogu RN, Ntoimo LF, Gana M, Okike ON, Durodola A. **Where do delays occur when women receive antenatal care? A client flow multi-site study in four health facilities in Nigeria**. *Ghana Med J.* (2018.0) **52** 8-14. DOI: 10.4314/gmj.v52i1.3
11. Ezumah N, Manzano A, Ezenwaka U, Obi U, Ensor T, Etiaba E. **Role of trust in sustaining provision and uptake of maternal and child healthcare: Evidence from a national programme in Nigeria**. *Soc Sci Med.* (2022.0) **293** 114644. DOI: 10.1016/j.socscimed.2021.114644
12. Baba-Ari F, Eboreime EA, Hossain M. **Conditional cash transfers for maternal health interventions: factors influencing uptake in north-central Nigeria**. *Int J Health Policy Manag.* (2018.0) **7** 934-42. DOI: 10.15171/ijhpm.2018.56
13. McConnell M, Rothschild CW, Ettenger A, Muigai F, Cohen J. **Free contraception and behavioural nudges in the postpartum period: evidence from a randomised control trial in Nairobi, Kenya**. *BMJ Glob Health.* (2018.0) **3** e000888. DOI: 10.1136/bmjgh-2018-000888
14. Cohen J, Rothschild C, Golub G, Omondi GN, Kruk ME, McConnell M. **Measuring the impact of cash transfers and behavioural 'nudges' on maternity care in Nairobi, Kenya**. *Health Aff.* (2017.0) **36** 1956-64. DOI: 10.1377/hlthaff.2017.0537
15. Stoop N, Hirvonen K, Maystadt JF. **Institutional mistrust and child vaccination coverage in Africa**. *BMJ Glob Health.* (2021.0) **6** e004595. DOI: 10.1136/bmjgh-2020-004595
16. Smith A, Burger R, Black V. **Demand-side causes and covariates of late antenatal care access in Cape town, South Africa**. *Matern Child Health J.* (2019.0) **23** 512-21. DOI: 10.1007/s10995-018-2663-2
17. Ntoimo LFC, Okonofua FE, Aikpitanyi J, Yaya S, Johnson E, Sombie I. **Influence of women's empowerment indices on the utilization of skilled maternity care: evidence from rural Nigeria**. *J Biosoc Sci* (2020.0) **54** 77-93. DOI: 10.1017/S0021932020000681
18. Adedokun ST, Uthman OA. **Women who have not utilized health service for delivery in Nigeria: who are they and where do they live?**. *BMC Pregnancy Childbirth.* (2019.0) **19** 93. DOI: 10.1186/s12884-019-2242-6
19. Mendolia S, Walker I. **The effect of noncognitive traits on health behaviours in adolescence**. *Health Econ.* (2014.0) **23** 1146-58. DOI: 10.1002/hec.3043
20. Bowles S, Gintis H. *Schooling in Capitalist America: Educational Reform and the Contradictions of Economics Life* (1976.0). DOI: 10.1017/S002205070009714X
21. Bowles S, Gintis H. *Schooling in Capitalist America: Educational Reform and the Contradictions of Economic Life* (1976.0) p. 340
22. Cobb-Clark DA, Kassenboehmer SC, Schurer S. **Healthy habits: The connection between diet, exercise, and locus of control**. *J Econ Behav Organ.* (2014.0) **98** 1-28. DOI: 10.1016/j.jebo.2013.10.011
23. Kesavayuth D, Poyago-Theotoky J, Tran DB, Zikos V. **Locus of control, health and healthcare utilization**. *Econ Model.* (2020.0) **86** 227-38. DOI: 10.1016/j.econmod.2019.06.014
24. Rotter J. **Generalized expectancies for internal versus external control of reinforcement**. *Psychol Monogr.* (1966.0) **80** 1-28. DOI: 10.1037/h0092976
25. Goldenberg JL, Shackelford TI. **Is It Me or Is It Mine? Body-self integration as a function of self-esteem, body-esteem, and mortality salience**. *Self Identity.* (2005.0) **4** 227-41. DOI: 10.1080/13576500444000254
26. Baumeister RF, Bushman BJ, Campbell WK. **Self-esteem, narcissism, and aggression: does violence result from low self-esteem or from threatened egotism?**. *Curr Direct Psychol Sci.* (2000.0) **9** 26-9. DOI: 10.1111/1467-8721.00053
27. Cox J, Loughran MJ, Adams EM, Navarro RL. **Life satisfaction and health related quality of life among low-income medical patients: the mediating influence of self-esteem**. *Psychol Health Med.* (2012.0) **17** 652-8. DOI: 10.1080/13548506.2012.656659
28. Cobb-Clark DA, Schurer S. **Two economists musings on the stability of locus of control**. *The Economic Journal.* (2013.0) **123** F358-400. DOI: 10.1111/ecoj.12069
29. Carton JS, Nowicki S. **Antecedents of individual differences in locus of control of reinforcement: a critical review**. *Genet Soc Gen Psychol Monogram.* (1994.0) **120** 33
30. Heckman J, Stixrud J, Urzua S. **The effects of cognitive and non-cognitive abilities on labor market outcomes and social behaviour**. *J Labor Economic.* (2006.0) **24** 411-82. DOI: 10.1086/504455
31. Lassi G, Taylor AE, Mahedy L, Heron J, Eisen T, Munafo MR. **Locus of control is associated with tobacco and alcohol consumption in young adults of the avon longitudinal study of parents and children**. *R Soc Open Sci.* (2019.0) **6** 181133. DOI: 10.1098/rsos.181133
32. Gale C, Batty G, Deary I. **Locus of control at age 10 years and health outcomes and behaviors at age 30 years: the 1970 British cohort study**. *Psychosomatic Med.* (2008.0) **70** 397-403. DOI: 10.1097/PSY.0b013e31816a719e
33. James W. *Psychology: The Briefer Course* (1892.0). DOI: 10.1037/11630-000
34. Rosenberg M. *Society and Adolescent Self-Image* (1965.0). DOI: 10.1515/9781400876136
35. Murphy LMB, Thompson JR, Morris MA. **Adherence behavior among adolescents with type I insulin-dependent diabetes mellitus: the role of cognitive appraisal processes**. *J Pediatr Psychol.* (1997.0) **22** 811-925. DOI: 10.1093/jpepsy/22.6.811
36. Pazzaglia F, Moè A, Cipolletta S, Chia M, Galozzi P, Masiero S. **Multiple dimensions of self-esteem and their relationship with health in adolescence**. *Int J Environ Res Public Health* (2020.0) **17**. DOI: 10.3390/ijerph17082616
37. Hunter KI, Linn MW, Harris R. **Characteristics of high and low self-esteem in the elderly**. *Int J Aging Hum Dev.* (1982.0) **14** 117-26. DOI: 10.2190/BGVM-2X68-PPFQ-EYV8
38. Doswell WM, Millor GK, Thompson H, Braxter B. **Self-image and self-esteem in African-American preteen girls: implications for mental health**. *Issues Ment Health Nurs.* (1998.0) **19** 71-94. DOI: 10.1080/016128498249222
39. Das PPP, Pattanaik P. **Self-esteem, locus of control and academic achievement among adolescents**. *Int J Sci Res Multidiscip Stud.* (2016.0) **2** 1-5
40. Fuscaldi FS, Balsanelli AC, Grossi SA. **Locus of control in health and self-esteem in type-2 diabetic patients**. *Rev Esc Enferm USP.* (2011.0) **45** 855-61. DOI: 10.1590/S0080-62342011000400009
41. Giblin PT, Poland ML, Ager JW. **Clinical applications of self-esteem and locus of control to adolescent health**. *J Adolesc Health Care.* (1988.0) **9** 1-14. DOI: 10.1016/0197-0070(88)90012-5
42. Dielman TE, Leech SL, Lorenger AT, Horvath WJ. **Health locus of control and self-esteem as related to adolescent health behavior and intentions**. *Adolescence.* (1984.0) **19** 935-50. PMID: 6516937
43. Alizadeh T, Farahani MN, Shahraray M, Alizadegan S. **The relationship between self-esteem and locus of control with infertility related stress of no related infertile men and women**. *J Reprod Infertili.* (2005.0)
44. 44.Central Office of Research and. Edo State Statistical Yearbook (2014-2020). Available online at: http://mda.edostate.gov.ng/budget/edo-state-statistical-yearbook-2014-2020/ (accessed September 13, 2021).. *Edo State Statistical Yearbook (2014-2020)*
45. 45.NPHCDA. National Primary Health Care Development Agency: Minimum Standards for Primary Health Care in Nigeria. Abuja; Department of Planning, Research and Statistics, National Primary Health Care Development Agency (2012).. *National Primary Health Care Development Agency: Minimum Standards for Primary Health Care in Nigeria* (2012.0)
46. Okonofua FE, Ntoimo LFC, Ogungbangbe J, Anjorin S, Imongan W, Yaya S. **Predictors of Women's utilization of primary health care for skilled pregnancy care in rural Nigeria**. *BMC Pregnancy Childbirth.* (2018.0) **18** 106. DOI: 10.1186/s12884-018-1730-4
47. 47.National Population Commission (NPC) [Nigeria] and ICF. Nigeria Demographic and Health Survey 2018 Key Indicators Report. Abuja; Maryland, MD: NPC and ICF (2019).. *Nigeria Demographic and Health Survey 2018 Key Indicators Report* (2019.0)
48. Ntoimo LFC, Okonofua FE, Igboin B, Ekwo C, Imongan W, Yaya S. **Why rural women do not use primary health centres for pregnancy care: evidence from a qualitative study in Nigeria**. *BMC Pregnancy Childbirth.* (2019.0) **19** 277. DOI: 10.1186/s12884-019-2433-1
49. Grossman M. **On the concept of health capital and the demand for health**. *J Polit Econ* (1972.0) **8** 223-55. DOI: 10.1086/259880
50. 50.Organization WH. Integrated Management of Pregnancy and Childbirth: Standards for Maternal and Neonatal Care. Geneva: WHO: Department of making pregnancy safer. (2007).. *Integrated Management of Pregnancy and Childbirth: Standards for Maternal and Neonatal Care* (2007.0)
51. Chiteji N. **Time preference, noncognitive skills and well-being across the life course: do noncognitive skills encourage healthy behaviors?**. *Am Econ Rev.* (2010.0) **100** 200-4. DOI: 10.1257/aer.100.2.200
52. Verbeek M. *A guide to modern econometrics* (2008.0)
53. Buddelmeyer H, Powdthavee N. **Can having internal locus of control insure against negative shocks? Psychological evidence from panel data**. *J Econ Behav Organ.* (2016.0) **122** 88-109. DOI: 10.1016/j.jebo.2015.11.014
54. Pearlin L, Schooler C. **The structure of coping**. (1978.0) **19** 2-21. DOI: 10.2307/2136319
55. Cronbach LJ. **Coefficient alpha and the internal structure of tests**. *Psychometrika* (1951.0) **16** 297-334. DOI: 10.1007/BF02310555
56. Kaplan BK, Cassel JC, Gore S. **Social support and health**. *Med Care.* (1977.0) **15** 47-58. DOI: 10.1097/00005650-197705001-00006
57. Andersen RM. **Revisiting the behavioral model and access to medical care: does it matter?**. *J Health Soc Behav.* (1995.0) **36** 1-10. DOI: 10.2307/2137284
58. Ganasen K, Parker S, Hugo C, Stein D, Emsley R, Seedat S. **Mental health literacy: focus on developing countries**. *Afr J Psychiatry.* (2008.0) **11** 23-8. DOI: 10.4314/ajpsy.v11i1.30251
59. Zeligman M, Grossman L, Tanzosh A. **Posttraumatic growth in trauma survivors: meaning making and locus of control**. *J Couns Pract.* (2019.0) **10** 1-21. DOI: 10.22229/ptg1022019
|
---
title: Prevalence of comorbid depression and associated factors among hospitalized
patients with type 2 diabetes mellitus in Hunan, China
authors:
- Rehanguli Maimaitituerxun
- Wenhang Chen
- Jingsha Xiang
- Atipatsa C. Kaminga
- Xin Yin Wu
- Letao Chen
- Jianzhou Yang
- Aizhong Liu
- Wenjie Dai
journal: BMC Psychiatry
year: 2023
pmcid: PMC10012793
doi: 10.1186/s12888-023-04657-4
license: CC BY 4.0
---
# Prevalence of comorbid depression and associated factors among hospitalized patients with type 2 diabetes mellitus in Hunan, China
## Abstract
### Background
Depression and diabetes are major health challenges, with heavy economic social burden, and comorbid depression in diabetes could lead to a wide range of poor health outcomes. Although many descriptive studies have highlighted the prevalence of comorbid depression and its associated factors, the situation in Hunan, China, remains unclear. Therefore, this study aimed to identify the prevalence of comorbid depression and associated factors among hospitalized type 2 diabetes mellitus (T2DM) patients in Hunan, China.
### Methods
This cross-sectional study involved 496 patients with T2DM who were referred to the endocrinology inpatient department of Xiangya Hospital affiliated to Central South University, Hunan. Participants’ data on socio-demographic status, lifestyle factors, T2DM-related characteristics, and social support were collected. Depression was evaluated using the Hospital Anxiety and Depression Scale-depression subscale. All statistical analyses were conducted using the R software version 4.2.1.
### Results
The prevalence of comorbid depression among hospitalized T2DM patients in Hunan was $27.22\%$ ($95\%$ Confidence Interval [CI]: 23.3–$31.1\%$). Individuals with depression differed significantly from those without depression in age, educational level, per capita monthly household income, current work status, current smoking status, current drinking status, regular physical activity, duration of diabetes, hypertension, chronic kidney disease, stroke, fatty liver, diabetic nephropathy, diabetic retinopathy, insulin use, HbA1c, and social support. A multivariable logistic regression model showed that insulin users (adjusted OR = 1.86, $95\%$ CI: 1.02–3.42) had a higher risk of depression, while those with regular physical activity (adjusted OR = 0.48, $95\%$ CI: 0.30–0.77) or greater social support (adjusted OR = 0.20, $95\%$ CI: 0.11–0.34) had a lower risk of depression. The area under the curve of the receiver operator characteristic based on this model was 0.741 with a sensitivity of 0.785 and specificity of 0.615.
### Conclusions
Depression was moderately prevalent among hospitalized T2DM patients in Hunan, China. Insulin treatment strategies, regular physical activity, and social support were significantly independently associated with depression, and the multivariable model based on these three factors demonstrated good predictivity, which could be applied in clinical practice.
## Background
Diabetes is one of the fastest growing chronic diseases across the globe, causing microvascular and macrovascular complications and reduced life expectancy [1]. According to the latest data released by the International Diabetes Federation, the global prevalence of diabetes was $10.5\%$ in 2021 and is expected to rise to $12.2\%$ in 2045; the number of people affected by the disease is estimated to increase to 783.2 million in 2045 from 536.6 million in 2021 [2]. China was ranked first by the number of adults with diabetes in 2021 (140.9 million) and is expected to retain its position in 2045 (174.4 million) [2]. According to a nationally-representative cross-sectional study conducted by the Chinese Center for Disease Control and Prevention, the overall standardized prevalence estimation of diabetes increased significantly from $10.9\%$ ($95\%$ Confidence Interval [CI]: 10.4–$11.5\%$) in 2013 to $12.4\%$ ($95\%$ CI: 11.8–$13.0\%$) in 2018 [3]. Type 2 diabetes mellitus (T2DM) accounts for the majority of diabetes cases, with its prevalence increasing from $3.7\%$ ($95\%$ CI: 3.6–$3.8\%$) in 2008 to $6.6\%$ ($95\%$ CI: 6.4–$6.7\%$) in 2017, in Beijing. However, the annual rate of increase slowed from $18.1\%$ ($95\%$ CI: 14.4–$22.0\%$) to $1.5\%$ ($95\%$ CI: 0.8–$2.2\%$) before and after 2011, respectively [4].
Depression, characterized by persistent sadness and a lack of interest or pleasure in previously rewarding or enjoyable activities [5], is one of the most common mental disorders that coexist with T2DM patients [6–8]. Comorbid depression in T2DM patients significantly worsens prognosis and raises mortality rates by increasing the incidence of microvascular and macrovascular complications, as well as reducing treatment compliance and self-care ability. This could ultimately lead to impaired glycemic control and poor quality of life [9–11]. Additionally, compared with those with only diabetes, comorbid depression combined with diabetes could lead to higher healthcare costs and a greater socioeconomic burden [12, 13]. Therefore, understanding the prevalence of comorbid depression and associated factors among T2DM patients is critical not only for appropriate allocation of psychological intervention resources for healthcare providers but also for the facilitation of early identification of those with a heightened risk of depression.
Globally, numerous studies have addressed the issue of comorbid depression in T2DM patients. An updated systematic review and meta-analysis showed that the pooled prevalence of comorbid depression among T2DM patients was $28\%$ globally, $24\%$ in Europe, $27\%$ in Africa, $29\%$ in Australia, and $32\%$ in Asia [14]. Liu et al. [ 15] found that the pooled prevalence of comorbid depression among T2DM patients in China was $25.9\%$ ($95\%$ CI: 20.6–$31.6\%$), with the figure being higher in females, participants aged ≥60 years, those with a primary school or lower education, individuals with a duration of T2DM ≥10 years, participants with diabetic complications, insulin users and participants living alone, and being lower in those with current alcohol use. However, no study has reported the prevalence of comorbid depression and associated factors among hospitalized patients with T2DM in Hunan, which is located in central China and middle reaches of the Yangtze River. The province covers an area of 211.8 thousand square kilometers, with a registered population of 73 million and a permanent population of 69.18 million in 2020. The present study aimed to identify the prevalence of comorbid depression and associated factors among hospitalized T2DM patients in Hunan, using a cross-sectional study design to comprehensively assess the role of socio-demographic status, lifestyle factors, T2DM-related characteristics, and social support in comorbid depression among T2DM patients.
## Study design and participants
This cross-sectional study was conducted in Hunan, China. All individuals referred to the endocrinology inpatient department of Xiangya Hospital affiliated to Central South University—a top-level general hospital located in the capital city of Hunan Province (Changsha), China—between March and December 2021, with T2DM confirmed by the endocrinologist, and aged ≥40 years were consecutively invited to participate in this study. Those with dementia or who could not speak Mandarin were excluded. The sample size of this study was 496, which was more than the minimum sample required of 385 as determined based on the sample size formula for categorical outcome (proportion) in cross-sectional studies (N = Z2p(1-p)/d2) [16] using these assumptions: $Z = 1.96$, $$p \leq 20.6$$% (the lower limit of the $95\%$ CI of the pooled prevalence of comorbid depression among T2DM patients in China reported by a previous meta-analysis) [15], and $d = 0.2$p.
## Procedures
Data on socio-demographic status, lifestyle factors, social support, and depression were collected by investigators who underwent unified training and had at least a bachelor’s degree in medicine. Information such as body mass index (BMI), duration of diabetes, family history of diabetes, diabetic complications and comorbidities, and insulin treatment strategies was obtained from electronic medical records. HbA1c was measured as part of routine inpatient visits using an ARKRAY automatic glycohemoglobin analyzer (ARKRAY Factory, Shanghai, China) on the ADAMS A1c HA-8180 system.
## Outcome of interest
The primary outcome of this study was the prevalence of comorbid depression. The Hospital Anxiety and Depression Scale-depression subscale (HADS-D) was used to evaluate depressive symptoms. The HADS-D, developed by Zigmond and Snaith, is a 7-item self-report scale with 4 response alternatives from 0 to 3 [17]. Its total score ranges between 0 and 21 with a score of ≥8 recommended as the cut off level for depression [18, 19]. Specifically, those with a total score of ≥8 were categorized into depression group, and those with a total score of < 8 were categorized into non-depression group. The HADS-D had satisfactory reliability and validity in the Chinese population. The concurrent validity of the HADS-D compared to the mental component summary of the Chinese version of Medical Outcomes Study 12-item Short Form (C-SF-12) (version 2) was described between − 0.53 and − 0.51 [20], and the Cronbach’s α coefficients ranged from 0.81 to 0.87 [20–22]. In the current study, the Cronbach’s α coefficient of HADS-D was 0.837.
## Independent variables
The independent variables of this study were socio-demographic status (age, sex, ethnicity, marital status, educational level, household income, living status, and current work status), lifestyle factors (current smoking and drinking status, and regular physical activity), T2DM related characteristics (BMI, duration of diabetes, family history of diabetes, diabetic complications and comorbidities, HbA1c, and insulin treatment strategies), and social support.
Current smoking was defined as smoking greater than or equal to 1 cigarette in the past 30 days, current drinking was defined as having at least one drink of any alcoholic beverage in the past 30 days, and regular physical activity was defined as the performance of at least one activity, such as walking, square dancing, and cycling, for at least 30 minutes per day in the past 30 days. T2DM related characteristics were collected from the electronic medical records, and BMI was categorized into < 24 and ≥ 24 based on Chinese Guidelines for Medical Nutrition Treatment of Overweight/Obesity (2021 edition) [23]. The duration of diabetes was categorized into < 10 years and ≥ 10 years based on the findings of previous studies [24–26], and HbA1c was categorized into ≤7 and > $7\%$ according to the Clinical Guidelines for Prevention and Treatment of Type 2 Diabetes Mellitus in older adults in China (2022 edition) [27].
Social support was measured using the Social Support Rating Scale (SSRS) [28], which is a 10-item scale with three dimensions—subjective support (4 items), objective support (3 items), and support utilization (3 items). The total score ranges from 12 to 66, with higher scores suggesting greater social support. A total score of > 44 (≤44) was regarded as high (low) social support. The SSRS has been widely used and is well validated in Chinese populations [29, 30], and in the current study, the Cronbach’s α coefficient of SSRS was 0.788.
## Statistical analyses
Continuous variables distributed normally or abnormally were described by mean ± standard deviation (SD) or median and interquartile range (IQR). Categorical variables were described by frequency (n) and proportion (%).
The difference between the depression and non-depression groups by each independent variable was examined using the chi-square test. The contribution of each independent variable to the outcome variable (depression) was quantified by crude OR and its corresponding $95\%$ CI using univariable logistic regression analyses. Independent variables that differed significantly between the depression and non-depression groups were entered into the multivariable logistic regression model, from which the contribution of each independent variable was quantified by adjusted OR (aOR) and its corresponding $95\%$ CI. Finally, the receiver operator characteristic (ROC) curve based on the multivariable logistic regression model was drawn to evaluate the predictive value of the model. All statistical analyses were two-sided at the $5\%$ significant level and were conducted in the R software version 4.2.1 (https://www.r-project.org/).
## Characteristics of the study participants
The mean age of the study population was 59.57 ± 9.92 with a range of 40 to 96. Among the 496 participants, 284 ($57.26\%$) were men, 469 ($94.56\%$) were Han Ethnicity, 447 ($90.12\%$) were married, and 217 ($43.75\%$) attended high school or above. Regarding lifestyle, 83 ($16.73\%$) were current smokers, 56 ($11.29\%$) were current drinkers, and 320 ($64.52\%$) performed regular physical activity. The mean duration of diabetes was 11.21 ± 7.75 years, with the majority ($54.84\%$) having duration of diabetes ≥10 years. Furthermore, 314 ($63.31\%$), 138 ($27.82\%$), 92 ($18.55\%$), 165 ($33.27\%$), 71 ($14.32\%$), and 117 ($23.60\%$) of the participants had hypertension, hyperlipidemia, coronary heart disease, chronic kidney disease, stroke, and fatty liver, respectively, whereas 262 ($52.82\%$), 226 ($45.57\%$), 49 ($9.88\%$), 379 ($76.41\%$), and 235 ($47.38\%$) participants had diabetic nephropathy, diabetic retinopathy, diabetic foot, diabetic peripheral neuropathy, and diabetic peripheral vascular disease, respectively. The mean total score of SSRS was 42.14 ± 7.55, and based on the cutoff value of 44, 289 ($58.27\%$) and 207 ($41.73\%$) were categorized as having low and high social support, respectively (Tables 1 and 2).Table 1Socio-demographic and lifestyle factors of the study participants ($$n = 496$$)Independent variableCategoryFrequency (n)Proportion (%)Age40–59 years27154.63≥60 years22545.36SexMale28457.26Female21242.74EthnicityHan46994.56Minority275.44Marital statusMarried44790.12Unmarried499.88Educational levelMiddle school or below27956.25High school or above21743.75Per capita monthly household income≤5000 yuan34168.75> 5000 yuan15531.25Living aloneYes306.05No46693.95Current work statusEmployed14529.23Not employed35170.77Current smoking statusNo41383.27Yes8316.73Current drinking statusNo44088.71Yes5611.29Regular physical activityNo17635.48Yes32064.52Table 2T2DM-related characteristics of the study participants ($$n = 496$$)Independent variableCategoryFrequency (n)Proportion (%)BMI< 2426954.23≥2422745.77Duration of diabetes< 10 years22445.16≥10 years27254.84Family history of diabetesNo27154.64Yes22545.36HypertensionNo18236.69Yes31463.31HyperlipidemiaNo35872.18Yes13827.82Coronary heart diseaseNo40481.45Yes9218.55Chronic kidney diseaseNo33166.73Yes16533.27StrokeNo42585.69Yes7114.32Fatty liverNo37976.41Yes11723.60Diabetic nephropathyNo23447.18Yes26252.82Diabetic retinopathyNo27054.44Yes22645.57Diabetic footNo44790.12Yes499.88Diabetic peripheral neuropathyNo11723.59Yes37976.41Diabetic peripheral vascular diseaseNo26152.62Yes23547.38Insulin useNo12224.60Yes37475.40HbA1c≤$7\%$11322.78> $7\%$38377.22Social supportLow28958.27High20741.73BMI, body mass index
## Prevalence of comorbid depression
The median (IQR) of the total score of HADS-D was 5.00 (6.00), and according to the cutoff value of 8, 135 and 361 participants were categorized into depression group and non-depression group, respectively. The prevalence of comorbid depression among hospitalized T2DM patients in Hunan was $27.22\%$ ($95\%$ CI: 23.3–$31.1\%$).
## Univariable analyses of factors associated with comorbid depression
Tables 3 and 4 show the results of univariable associations between the independent variables and comorbid depression. The depression and non-depression groups differed significantly in relation to age, educational level, per capita monthly household income, current work status, current smoking status, current drinking status, regular physical activity, duration of diabetes, hypertension, chronic kidney disease, stroke, fatty liver, diabetic nephropathy, diabetic retinopathy, insulin use, HbA1c, and social support ($P \leq 0.05$).Table 3Univariable analyses for the associations of socio-demographic and lifestyle factors with comorbid depressionIndependent variableCategoryDepression group ($$n = 135$$, %)Non-depression group ($$n = 361$$, %)Crude OR ($95\%$ CI)P valueAge40–59 years58 (42.96)213 (59.00)10.001**≥60 years77 (57.04)148 (41.00)1.91 (1.28–2.85)SexMale68 (50.37)216 (59.83)10.058Female67 (49.63)145 (40.17)1.47 (0.99–2.18)EthnicityHan130 (96.30)339 (93.91)10.296Minority5 (3.70)22 (6.09)0.59 (0.22–1.60)Marital statusMarried116 (85.9)331 (91.69)10.056Unmarried19 (14.07)30 (8.31)1.81 (0.98–3.33)Educational levelMiddle school or below96 (71.11)183 (50.69)1< 0.001***High school or above39 (28.89)178 (49.31)0.42 (0.27–0.64)*Per capita* monthly household income≤5000 yuan114 (84.44)227 (62.88)1< 0.001***> 5000 yuan21 (15.56)134 (37.12)0.31 (0.19–0.52)Living aloneYes12 (8.89)18 (4.99)10.105No123 (91.11)343 (95.01)0.54 (0.25–1.25)Current work statusEmployed18 (13.33)127 (35.18)1< 0.001***Not employed117 (86.67)234 (64.82)3.53 (2.05–6.06)Current smoking statusNo120 (88.89)293 (81.16)10.040*Yes15 (11.11)68 (18.84)0.54 (0.30–0.98)Current drinking statusNo126 (93.33)314 (86.98)10.047*Yes9 (6.67)47 (13.02)0.48 (0.23–0.99)Regular physical activityNo68 (50.37)108 (29.92)1< 0.001***Yes67 (49.63)253 (70.09)0.43 (0.23–0.63)* $P \leq 0.05$; ** $P \leq 0.01$; *** $P \leq 0.001$Table 4Univariable analyses for the association of T2DM related characteristics with comorbid depressionIndependent variableCategoryDepression group ($$n = 135$$, %)Non-depression group ($$n = 361$$, %)Crude OR ($95\%$ CI)P valueBMI< 2474 (54.81)195 (54.02)10.874≥2461 (45.19)166 (45.98)0.97 (0.65–1.44)Duration of diabetes< 10 years50 (37.04)174 (48.20)10.002**≥10 years85 (62.96)187 (51.80)1.96 (1.27–3.04)Family history of diabetesNo76 (56.30)195 (54.02)10.650Yes59 (43.70)166 (45.98)0.91 (0.61–1.36)HypertensionNo100 (74.07)214 (59.28)10.002**Yes35 (25.93)147 (40.72)1.96 (1.27–3.04)HyperlipidemiaNo34 (25.19)104 (28.81)10.423Yes101 (74.81)257 (71.19)0.83 (0.53–1.31)Coronary heart diseaseNo32 (23.70)60 (16.62)10.071Yes103 (76.30)301 (83.38)1.56 (0.96–2.53)Chronic kidney diseaseNo59 (43.70)106 (29.36)10.003**Yes76 (56.30)255 (70.64)1.87 (1.24–2.81)StrokeNo29 (21.48)42 (11.63)10.005**Yes106 (78.52)319 (88.37)2.08 (1.23–3.50)Fatty liverNo22 (16.30)95 (26.32)10.019*Yes113 (83.70)266 (73.68)0.55 (0.33–0.91)Diabetic nephropathyNo90 (66.67)172 (47.65)1< 0.001***Yes45 (33.33)189 (52.35)2.20 (1.45–3.32)Diabetic retinopathyNo77 (57.04)149 (41.27)10.002**Yes58 (42.96)212 (58.73)1.89 (1.27–2.82)Diabetic footNo16 (11.85)33 (9.14)10.368Yes119 (88.15)328 (90.86)1.34 (0.71–2.52)Diabetic peripheral neuropathyNo28 (20.74)89 (24.65)10.361Yes107 (79.26)272 (75.35)1.25 (0.71–2.52)Diabetic peripheral vascular diseaseNo71 (52.59)190 (52.63)10.994Yes64 (47.41)171 (47.37)1.00 (0.67–1.49)Insulin useNo21 (15.56)101 (27.98)10.004**Yes114 (84.44)260 (72.02)2.11 (1.25–3.54)HbA1c≤$7\%$64 (47.41)131 (36.29)10.014*> $7\%$71 (52.59)230 (63.71)0.24 (0.05–0.44)Social supportLow114 (84.44)175 (48.48)1< 0.001***High21 (15.56)186 (51.52)0.17 (0.10–0.29)* $P \leq 0.05$; ** $P \leq 0.01$; *** $P \leq 0.001$; BMI, body mass index
## Multivariable analyses of factors associated with comorbid depression
Table 5 shows the results of the multivariable logistic regression model on factors associated with comorbid depression among hospitalized T2DM patients in Hunan. Insulin users (aOR = 1.86, $95\%$ CI: 1.02–3.42, $$P \leq 0.044$$) were at an increased risk of comorbid depression. Those performing regular physical activity (aOR = 0.48, $95\%$ CI: 0.30–0.77, $$P \leq 0.002$$) or having higher social support (aOR = 0.20, $95\%$ CI: 0.11–0.34, $P \leq 0.001$) had a reduced risk of comorbid depression. The ROC curve based on the multivariable logistic regression model including the factors of insulin treatment strategies, regular physical activity, and social support is shown in Fig. 1. The area under the curve (AUC) was 0.741 with a sensitivity of 0.785 and a specificity of 0.615.Table 5Multivariable logistic regression analysis on factors associated with comorbid depressionIndependent variableDescriptionBSEAdjusted OR ($95\%$ CI)P valueAge40–59 years Vs ≥60 years−0.5180.2821.27 (0.74–2.18)0.379Educational levelMiddle school or below Vs High school or above−0.5180.2820.59 (0.34–1.03)0.064Per capita monthly household income≤5000 yuan Vs > 5000 yuan−0.5140.3310.60 (0.31–1.14)0.120Current work statusEmployed Vs Not employed0.4980.3451.63 (0.83–3.21)0.155Current smoking statusNo Vs Yes−0.5010.3840.61 (0.29–1.30)0.204Current drinking statusNo Vs Yes0.2680.491.31 (0.50–3.41)0.585Regular physical activityNo Vs Yes−0.7200.2390.48 (0.30–0.77)0.002**Duration of diabetes< 10 years Vs ≥10 years0.0890.1561.24 (0.59–2.59)0.565HypertensionNo Vs Yes0.2220.2791.24 (0.72–2.15)0.437Chronic kidney diseaseNo Vs Yes0.2870.311.33 (0.72–2.44)0.364StrokeNo Vs Yes0.5700.3191.75 (0.94–3.28)0.079Fatty liverNo Vs Yes−0.2460.3110.79 (0.43–1.45)0.447Diabetic nephropathyNo Vs Yes0.1080.3161.11 (0.6–2.07)0.734Diabetic retinopathyNo Vs Yes0.4260.2551.54 (0.93–2.54)0.091Insulin useNo Vs Yes0.6130.3081.86 (1.02–3.42)0.044*HbA1c≤$7\%$ Vs > $7\%$−0.4270.2670.65 (0.38–1.10)0.108Social supportLow Vs High−1.6240.280.20 (0.11–0.34)< 0.001**** $P \leq 0.05$; ** $P \leq 0.01$; *** $P \leq 0.001$Fig. 1The ROC curve based on the multivariable logistic regression model
## Discussion
Compared with healthy controls, T2DM patients had a two-fold risk of developing depression [6], with comorbid depression in T2DM potentially leading to a wide range of poor health outcomes [31–33]. The present study was aimed at identifying the prevalence of comorbid depression and associated factors among hospitalized T2DM patients in Hunan. Few studies have evaluated the prevalence of comorbid depression among hospitalized patients exclusively. These limited studies showed that the prevalence was $49.2\%$ ($\frac{70}{142}$) in Pakistan [34], $33.1\%$ ($\frac{47}{142}$) in Morocco [35], $53.8\%$ ($\frac{85}{160}$) in Saudi Arabia [36], and $23.2\%$ ($\frac{50}{216}$) in Vietnam [37]. The prevalence of comorbid depression among hospitalized T2DM patients in Hunan, found in the present study ($27.22\%$, $\frac{135}{496}$) was lower than that in Pakistan and Saudi Arabia and almost comparable with that in Vietnam and Morocco. The differences in the economic development level, cultural background, instruments used to assess depression, and sample characteristics in diabetes severity may account for the prevalence differences observed for different countries. Given the negative effects brought by the coexistence of depression and T2DM, it is suggested that in addition to regular blood glucose monitoring, routine screening of depression should be conducted among hospitalized T2DM patients by well-trained healthcare professionals in clinical practice in Hunan. Furthermore, timely and effective support such as psychosocial intervention and cognitive behavioral therapy should be implemented for those with comorbid depression [38, 39].
Socio-demographic characteristics such as sex, age, educational level, and income were found to be associated with comorbid depression among T2DM patients in many previous studies [34, 37, 40–46]. In a recent systematic review and meta-analysis by Liu et al. [ 15], the pooled prevalence of comorbid depression among population-based T2DM patients in China was found to be higher in females, participants aged ≥60, those with a primary school or lower education, and individuals living alone. However, some studies showed different results. For example, Tran et al. [ 37] found that sex was not associated with comorbid depression among hospitalized T2DM patients, whereas Huang et al. [ 42] found that sex was independently associated with the prevalence and incidence of comorbid depression among diabetes patients in Taiwan. Similarly, Khan et al. [ 34] found that higher scores of depression were significantly associated with sex and age. This study showed that the prevalence of comorbid depression among T2DM patients was univariately but not independently associated with age, educational level, per capita monthly household income, and current work status. Therefore, it is imperative to conduct more studies with a larger sample of hospitalized T2DM patients to identify the association between socio-demographic characteristics and comorbid depression among hospitalized T2DM patients in China.
Regular physical activity is an important component of treatment strategies for diabetes patients, and its association with depression has been well-established among the general population [47, 48]. In diabetic populations, Mendes et al. [ 49] found that for older outpatients with diabetes, those with non-adherence to physical activity showed more depressive symptoms. Additionally, Ahola et al. [ 50] found that, after adjustments, more leisure-time physical activity was associated with more depressive symptoms in adult individuals with type 1 diabetes mellitus. Furthermore, a study using data from the Korea National Health and Nutritional Examination Survey [51] found that moderate intensity physical activity at work and during leisure time affected depression. Similarly, the present study found that regular physical activity in hospitalized T2DM patients was associated with a lower risk of depression after adjustments. Therefore, it is highly recommended for healthcare professionals in China to promote healthy lifestyles, including regular physical activity, among hospitalized T2DM patients to help manage depressive symptoms. Additionally, as family support is crucial for adherence to physical activity, efforts should be made by family members to encourage increased physical activity among hospitalized T2DM patients.
Diabetic neuropathy affects up to $50\%$ of diabetes patients and is a major cause of morbidity and increased mortality. Its clinical manifestations include painful neuropathic symptoms and insensitivity, which increases the risk of burns, injuries, and foot ulceration [52]. The presence of diabetic neuropathy could worsen quality of life and induce changes in social and family roles, thus increasing the risk of depression [53]. A previous meta-analysis with 13 eligible studies involving 3898 individuals confirmed the association between diabetic neuropathy and depression among T2DM patients [54]. However, the present study found that diabetic retinopathy was associated with depression in the univariable analyses but not in the fully adjusted multivariable model. It is worth noting here that in the multivariable model, the contribution of diabetic retinopathy to depression has reached the margin of statistical significance (aOR = 1.54, $95\%$CI: 0.93–2.54, $$P \leq 0.091$$). Therefore, caution should be taken when interpreting this association.
For T2DM patients, insulin is the cornerstone of treatment for lowering glucose and HbA1c concentrations [55]. Although the optimal timing and indications for insulin therapy remain controversial, most of the patients inevitably require insulin therapy to attain adequate glycemic control in the natural history of T2DM [56, 57]. In the present study, $75.40\%$ of the participants were treated with insulin, and this group had a 1.86-fold risk of developing depression compared with their counterparts. This is consistent with a previous meta-analysis that included 12 eligible studies [58]. Compared with those who were not on insulin treatment, those on insulin treatment were more likely to have advanced T2DM with less endogenous insulin, increasing their susceptibility to metabolic dysregulation; hence, they were more vulnerable to develop depression [59]. Therefore, those on insulin treatment may need more regular check-ups for depression in clinical practice.
Social support is a psychosocial element that influences people by providing them with emotional, informational, companionship, and financial support to increase their adherence to diabetes treatment and management guidelines [60]. The protective role of social support against depression has been identified in different populations in secondary data analyses [61–63]. Accordingly, Azmiardi et al. [ 64] conducted a systematic review and meta-analysis of 11 eligible studies involving a total of 3151 individuals and found that T2DM patients with lower social support had a two-fold risk of depression than those with greater social support. Similarly, this study found that higher social support was associated with a lower risk of depression after adjustments. These findings, therefore, underscore the importance of social support in the control and management of depression among hospitalized T2DM patients in China. Additionally, to build a systematic approach to reduce the burden of comorbid depression in T2DM, further studies on the mechanisms underlying this relationship are required.
This study found that nearly one fourth of hospitalized T2DM patients suffered from depression in Hunan, China. Considering the linkage between comorbid depression and subsequent poor health outcomes, routine screening for depression among hospitalized T2DM patients in this area is highly recommended. Insulin treatment strategies, regular physical activity, and social support were independently associated with depression. Therefore, the risk of comorbid depression among hospitalized T2DM patients might be reduced through enhancing physical activity and offering more social support in clinical practice, and those on insulin treatment should be paid special attention for preventing comorbid depression.
Some limitations should be considered when interpreting the findings of the present study. This was a cross-sectional study, implying that the associations of insulin treatment strategies, regular physical activity, and social support with comorbid depression might be bidirectional. Based on the considerations of recall bias, this study collected information on current smoking and drinking instead of the amount and frequency of smoking and drinking, which might play an important role in the presence of depression. Therefore, future prospective studies are needed to elucidate the causal relationships. Additionally, this study was hospital-based with a single-centre of participants aged ≥40 in Hunan, and the mean duration of diabetes of the participants was 11.21 ± 7.75 years with the majority having duration of diabetes ≥10 years. This may preclude the possibility of identifying the association between newly-diagnosed diabetes and depression, and whether the findings can be generalized into all T2DM patients in Hunan remains unclear. Therefore, future studies with larger, more representative samples of T2DM patients are needed.
## Conclusions
Depression is moderately prevalent with $27.22\%$ of hospitalized T2DM patients suffering from depression in Hunan. Routine screening for depression among hospitalized T2DM patients in this area is highly recommended. Participants undergoing insulin treatment have a higher risk of comorbid depression, and those with regular physical activity or higher social support were at a lower risk of comorbid depression. The multivariable model based on the foregoing factors showed good predictivity, suggesting that it may be useful in clinical practice.
## References
1. Heald AH, Stedman M, Davies M, Livingston M, Alshames R, Lunt M. **Estimating life years lost to diabetes: outcomes from analysis of National Diabetes Audit and Office of National Statistics data**. *Cardiovasc Endocrinol Metab* (2020.0) **9** 183-185. PMID: 33225235
2. Sun H, Saeedi P, Karuranga S, Pinkepank M, Ogurtsova K, Duncan BB. **IDF diabetes atlas: global, regional and country-level diabetes prevalence estimates for 2021 and projections for 2045**. *Diabetes Res Clin Pract* (2022.0) **183** 109119. PMID: 34879977
3. Wang L, Peng W, Zhao Z, Zhang M, Shi Z, Song Z. **Prevalence and treatment of diabetes in China, 2013-2018**. *JAMA* (2021.0) **326** 2498-2506. PMID: 34962526
4. Wang Z, Wu Y, Wu J, Wang M, Wang X, Wang J. **Trends in prevalence and incidence of type 2 diabetes among adults in Beijing, China, from 2008 to 2017**. *Diabet Med* (2021.0) **38** e14487. PMID: 33278034
5. 5.World Health OrganizationDepression2022. *Depression* (2022.0)
6. Roy T, Lloyd CE. **Epidemiology of depression and diabetes: a systematic review**. *J Affect Disord* (2012.0) **142** S8-21. PMID: 23062861
7. Semenkovich K, Brown ME, Svrakic DM, Lustman PJ. **Depression in type 2 diabetes mellitus: prevalence, impact, and treatment**. *Drugs* (2015.0) **75** 577-587. PMID: 25851098
8. Mukherjee N, Chaturvedi SK. **Depressive symptoms and disorders in type 2 diabetes mellitus**. *Curr Opin Psychiatry* (2019.0) **32** 416-421. PMID: 31135489
9. Lin EH, Rutter CM, Katon W, Heckbert SR, Ciechanowski P, Oliver MM. **Depression and advanced complications of diabetes: a prospective cohort study**. *Diabetes Care* (2010.0) **33** 264-269. PMID: 19933989
10. Zou Y, You W, Wang J, Wang F, Tian Z, Lu J. **Depression and retinopathy in patients with type 2 diabetes mellitus: a meta-analysis**. *Psychosom Med* (2021.0) **83** 239-246. PMID: 33657086
11. Gonzalez JS, Peyrot M, McCarl LA, Collins EM, Serpa L, Mimiaga MJ. **Depression and diabetes treatment nonadherence: a meta-analysis**. *Diabetes Care* (2008.0) **31** 2398-2403. PMID: 19033420
12. Egede LE, Walker RJ, Bishu K, Dismuke CE. **Trends in costs of depression in adults with diabetes in the United States: medical expenditure panel survey, 2004-2011**. *J Gen Intern Med* (2016.0) **31** 615-622. PMID: 26969312
13. Tardif I, Guénette L, Zongo A, Demers É, Lunghi C. **Depression and the risk of hospitalization in type 2 diabetes patients: a nested case-control study accounting for non-persistence to antidiabetic treatment**. *Diabetes Metab* (2022.0) **48** 101334. PMID: 35231612
14. Khaledi M, Haghighatdoost F, Feizi A, Aminorroaya A. **The prevalence of comorbid depression in patients with type 2 diabetes: an updated systematic review and meta-analysis on huge number of observational studies**. *Acta Diabetol* (2019.0) **56** 631-650. PMID: 30903433
15. Liu X, Li Y, Guan L, He X, Zhang H, Zhang J. **A systematic review and Meta-analysis of the prevalence and risk factors of depression in type 2 diabetes patients in China**. *Front Med (Lausanne)* (2022.0) **9** 759499. PMID: 35620713
16. Bolarinwa OA. **Sample size estimation for health and social science researchers: the principles and considerations for different study designs**. *Niger Postgrad Med J* (2020.0) **27** 67-75. PMID: 32295935
17. Zigmond AS, Snaith RP. **The hospital anxiety and depression scale**. *Acta Psychiatr Scand* (1983.0) **67** 361-370. PMID: 6880820
18. Melin EO, Thunander M, Svensson R, Landin-Olsson M, Thulesius HO. **Depression, obesity, and smoking were independently associated with inadequate glycemic control in patients with type 1 diabetes**. *Eur J Endocrinol* (2013.0) **168** 861-869. PMID: 23536618
19. Melin EO, Thunander M, Landin-Olsson M, Hillman M, Thulesius HO. **Depression, smoking, physical inactivity and season independently associated with midnight salivary cortisol in type 1 diabetes**. *BMC Endocr Disord* (2014.0) **14** 75. PMID: 25224993
20. Li Q, Lin Y, Hu C, Xu Y, Zhou H, Yang L. **The Chinese version of hospital anxiety and depression scale: psychometric properties in Chinese cancer patients and their family caregivers**. *Eur J Oncol Nurs* (2016.0) **25** 16-23. PMID: 27865248
21. Olssøn I, Mykletun A, Dahl AA. **The hospital anxiety and depression rating scale: a cross-sectional study of psychometrics and case finding abilities in general practice**. *BMC Psychiatry* (2005.0) **5** 46. PMID: 16351733
22. Yang Z, Huang X, Liu X, Hou J, Wu W, Song A. **Psychometric properties and factor structure of the Chinese version of the hospital anxiety and depression scale in people living with HIV**. *Front Psychiatry* (2019.0) **10** 346. PMID: 31156484
23. 23.Guidelines for medical nutrition treatment of overweight/obesity in China (2021 edition). Chinese J Front Med Sci (Electronic Version). 2021;13(11):1–55.
24. Naskar S, Victor R, Nath K. **Depression in diabetes mellitus-a comprehensive systematic review of literature from an Indian perspective**. *Asian J Psychiatr* (2017.0) **27** 85-100. PMID: 28558904
25. Majdan M, Krajcovicova L, Pekarcikova J, Chereches R, O'Mullane M. **Predictors of depression symptoms in patients with diabetes in Slovakia**. *Int J Psychiatry Med* (2012.0) **44** 351-366. PMID: 23885517
26. Ebrahim M, Tamiru D, Hawulte B, Misgana T. **Prevalence and associated factors of depression among diabetic outpatients attending diabetic clinic at public hospitals in eastern Ethiopia: a cross-sectional study**. *SAGE Open Med* (2021.0) **9** 20503121211066244. PMID: 34992781
27. 27.Clinical guidelines for prevention and treatment of type 2 diabetes mellitus in the elderly in China (2022 edition). Chin J Internal Med 2022;61(1):12–50.
28. Xiao S. **The theoretical basis and research application of social support rating scale**. *J Clin Psychol Med* (1994.0) **02** 98-100
29. Guoyu Y, Zhengzhi F, Benli X, Tiejun Z, Yunbo L, Tao W. **The reliability, validity and National Norm of social support scale in servicemen in Chinese**. *Chin Ment Health J* (2006.0) **05** 309-312
30. 30.Changfei L, Cunxian J, Jiyu Z, Xinxin W, Dongfang W, Liyuan L. Reliability and validity of social support rating scale in rural suicides in Chinese. Chin Ment Health J. 2011;25(3):218–22.
31. van Dooren FE, Nefs G, Schram MT, Verhey FR, Denollet J, Pouwer F. **Depression and risk of mortality in people with diabetes mellitus: a systematic review and meta-analysis**. *PLoS One* (2013.0) **8** e57058. PMID: 23472075
32. Hofmann M, Köhler B, Leichsenring F, Kruse J. **Depression as a risk factor for mortality in individuals with diabetes: a meta-analysis of prospective studies**. *PLoS One* (2013.0) **8** e79809. PMID: 24278183
33. Farooqi A, Khunti K, Abner S, Gillies C, Morriss R, Seidu S. **Comorbid depression and risk of cardiac events and cardiac mortality in people with diabetes: a systematic review and meta-analysis**. *Diabetes Res Clin Pract* (2019.0) **156** 107816. PMID: 31421139
34. Khan P, Qayyum N, Malik F, Khan T, Khan M, Tahir A. **Incidence of anxiety and depression among patients with type 2 diabetes and the predicting factors**. *Cureus* (2019.0) **11** e4254. PMID: 31131177
35. 35.Benmaamar S, Lazar N, El Harch I, Maiouak M, Qarmiche N, Otmani N, et al. Depression and anxiety in patients with diabetes in a Moroccan region. Encephale. 2022;48(6):601–6.
36. Alzahrani A, Alghamdi A, Alqarni T, Alshareef R, Alzahrani A. **Prevalence and predictors of depression, anxiety, and stress symptoms among patients with type II diabetes attending primary healthcare centers in the western region of Saudi Arabia: a cross-sectional study**. *Int J Ment Health Syst* (2019.0) **13** 48. PMID: 31341512
37. Tran NMH, Nguyen QNL, Vo TH, Le TTA, Ngo NH. **Depression among patients with type 2 diabetes mellitus: prevalence and associated factors in Hue City, Vietnam**. *Diabetes Metab Syndr Obes* (2021.0) **14** 505-513. PMID: 33568927
38. 38.Wang ZD, Xia YF, Zhao Y, Chen LM. Cognitive behavioural therapy on improving the depression symptoms in patients with diabetes: a meta-analysis of randomized control trials. Biosci Rep. 2017;37(2):BSR20160557.
39. Xie J, Deng W. **Psychosocial intervention for patients with type 2 diabetes mellitus and comorbid depression: a meta-analysis of randomized controlled trials**. *Neuropsychiatr Dis Treat* (2017.0) **13** 2681-2690. PMID: 29123401
40. Fugger G, Dold M, Bartova L, Kautzky A, Souery D, Mendlewicz J. **Major depression and comorbid diabetes - findings from the European Group for the Study of resistant depression**. *Prog Neuro-Psychopharmacol Biol Psychiatry* (2019.0) **94** 109638
41. AlBekairy A, AbuRuz S, Alsabani B, Alshehri A, Aldebasi T, Alkatheri A. **Exploring factors associated with depression and anxiety among hospitalized patients with type 2 diabetes mellitus**. *Med Princ Pract* (2017.0) **26** 547-553. PMID: 29131123
42. Huang CJ, Lin CH, Lee MH, Chang KP, Chiu HC. **Prevalence and incidence of diagnosed depression disorders in patients with diabetes: a national population-based cohort study**. *Gen Hosp Psychiatry* (2012.0) **34** 242-248. PMID: 22325626
43. Lopez-de-Andrés A, Jiménez-Trujillo MI, Hernández-Barrera V, de Miguel-Yanes JM, Méndez-Bailón M, Perez-Farinos N. **Trends in the prevalence of depression in hospitalized patients with type 2 diabetes in Spain: analysis of hospital discharge data from 2001 to 2011**. *PLoS One* (2015.0) **10** e0117346. PMID: 25706646
44. Arambewela MH, Somasundaram NP, Jayasekara H, Kumbukage MP. **Prevalence of depression and associated factors among patients with type 2 diabetes attending the diabetic Clinic at a Tertiary Care Hospital in Sri Lanka: a descriptive study**. *Psychiatry J* (2019.0) **2019** 7468363. PMID: 30863775
45. Dehesh T, Dehesh P, Shojaei S. **Prevalence and associated factors of anxiety and depression among patients with type 2 diabetes in Kerman, Southern Iran**. *Diabetes Metab Syndr Obes* (2020.0) **13** 1509-1517. PMID: 32440180
46. Ding X, Rong S, Wang Y, Li D, Wen L, Zou B. **The Association of the Prevalence of depression in type 2 diabetes mellitus with visual-related quality of life and social support**. *Diabetes Metab Syndr Obes* (2022.0) **15** 535-544. PMID: 35237054
47. Weyerer S. **Physical inactivity and depression in the community. Evidence from the upper Bavarian field study**. *Int J Sports Med* (1992.0) **13** 492-496. PMID: 1428382
48. Ströhle A, Höfler M, Pfister H, Müller AG, Hoyer J, Wittchen HU. **Physical activity and prevalence and incidence of mental disorders in adolescents and young adults**. *Psychol Med* (2007.0) **37** 1657-1666. PMID: 17579930
49. Mendes R, Martins S, Fernandes L. **Adherence to medication, physical activity and diet in older adults with diabetes: its association with cognition, anxiety and depression**. *J Clin Med Res* (2019.0) **11** 583-592. PMID: 31413770
50. Ahola AJ, Tikkanen-Dolenc H, Forsblom C, Harjutsalo V, Groop PH. **Symptoms of depression are associated with reduced leisure-time physical activity in adult individuals with type 1 diabetes**. *Acta Diabetol* (2021.0) **58** 1373-1380. PMID: 34009436
51. Kim DJ. **Effects of physical activity on depression in adults with diabetes**. *Osong Public Health Res Perspect* (2018.0) **9** 143-149. PMID: 30159219
52. Tesfaye S, Selvarajah D. **Advances in the epidemiology, pathogenesis and management of diabetic peripheral neuropathy**. *Diabetes Metab Res Rev* (2012.0) **28** 8-14. PMID: 22271716
53. Yoshida S, Hirai M, Suzuki S, Awata S, Oka Y. **Neuropathy is associated with depression independently of health-related quality of life in Japanese patients with diabetes**. *Psychiatry Clin Neurosci* (2009.0) **63** 65-72. PMID: 19067994
54. Bartoli F, Carrà G, Crocamo C, Carretta D, La Tegola D, Tabacchi T. **Association between depression and neuropathy in people with type 2 diabetes: a meta-analysis**. *Int J Geriatr Psychiatry* (2016.0) **31** 829-836. PMID: 26729627
55. Ohn JH, Kwak SH, Cho YM, Lim S, Jang HC, Park KS. **10-year trajectory of β-cell function and insulin sensitivity in the development of type 2 diabetes: a community-based prospective cohort study**. *Lancet Diabetes Endocrinol* (2016.0) **4** 27-34. PMID: 26577716
56. Home P, Riddle M, Cefalu WT, Bailey CJ, Bretzel RG, Del Prato S. **Insulin therapy in people with type 2 diabetes: opportunities and challenges?**. *Diabetes Care* (2014.0) **37** 1499-1508. PMID: 24855154
57. Kramer CK, Zinman B, Retnakaran R. **Short-term intensive insulin therapy in type 2 diabetes mellitus: a systematic review and meta-analysis**. *Lancet Diabetes Endocrinol* (2013.0) **1** 28-34. PMID: 24622264
58. Bai X, Liu Z, Li Z, Yan D. **The association between insulin therapy and depression in patients with type 2 diabetes mellitus: a meta-analysis**. *BMJ Open* (2018.0) **8** e020062. PMID: 30498035
59. Surwit RS, van Tilburg MA, Parekh PI, Lane JD, Feinglos MN. **Treatment regimen determines the relationship between depression and glycemic control**. *Diabetes Res Clin Pract* (2005.0) **69** 78-80. PMID: 15955389
60. Strom JL, Egede LE. **The impact of social support on outcomes in adult patients with type 2 diabetes: a systematic review**. *Curr Diab Rep* (2012.0) **12** 769-781. PMID: 22949135
61. Weldesenbet AB, Kebede SA, Tusa BS. **The effect of poor social support on depression among HIV/AIDS patients in Ethiopia: a systematic review and Meta-analysis**. *Depress Res Treat* (2020.0) **2020** 6633686. PMID: 33489371
62. Schiller VF, Dorstyn DS, Taylor AM. **The protective role of social support sources and types against depression in caregivers: a Meta-analysis**. *J Autism Dev Disord* (2021.0) **51** 1304-1315. PMID: 32683544
63. Rueger SY, Malecki CK, Pyun Y, Aycock C, Coyle S. **A meta-analytic review of the association between perceived social support and depression in childhood and adolescence**. *Psychol Bull* (2016.0) **142** 1017-1067. PMID: 27504934
64. Azmiardi A, Murti B, Febrinasari RP, Tamtomo DG. **Low social support and risk for depression in people with type 2 diabetes mellitus: a systematic review and Meta-analysis**. *J Prev Med Public Health* (2022.0) **55** 37-48. PMID: 35135047
|
---
title: 'Implementing clinical decision support for reducing women Veterans'' cardiovascular
risk in VA: A mixed-method, longitudinal study of context, adaptation, and uptake'
authors:
- Julian Brunner
- Melissa M. Farmer
- Bevanne Bean-Mayberry
- Catherine Chanfreau-Coffinier
- Claire T. Than
- Alison B. Hamilton
- Erin P. Finley
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012802
doi: 10.3389/frhs.2022.946802
license: CC BY 4.0
---
# Implementing clinical decision support for reducing women Veterans' cardiovascular risk in VA: A mixed-method, longitudinal study of context, adaptation, and uptake
## Abstract
Evaluations of clinical decision support (CDS) implementation often struggle to measure and explain heterogeneity in uptake over time and across settings, and to account for the impact of context and adaptation on implementation success. In 2017–2020, the EMPOWER QUERI implemented a cardiovascular toolkit using a computerized template aimed at reducing women Veterans' cardiovascular risk across five Veterans Healthcare Administration (VA) sites, using an enhanced Replicating Effective Programs (REP) implementation approach. In this study, we used longitudinal joint displays of qualitative and quantitative findings to explore [1] how contextual factors emerged across sites, [2] how the template and implementation strategies were adapted in response to contextual factors, and [3] how contextual factors and adaptations coincided with template uptake across sites and over time. We identified site structure, staffing changes, relational authority of champions, and external leadership as important contextual factors. These factors gave rise to adaptations such as splitting the template into multiple parts, pairing the template with a computerized reminder, conducting academic detailing, creating cheat sheets, and using small-scale pilot testing. All five sites exhibited variability in utilization over the months of implementation, though later sites exhibited higher template utilization immediately post-launch, possibly reflecting a “preloading” of adaptations from previous sites. These findings underscore the importance of adaptive approaches to implementation, with intentional shifts in intervention and strategy to meet the needs of individual sites, as well as the value of integrating mixed-method data sources in conducting longitudinal evaluation of implementation efforts.
## Introduction
Computerized clinical decision support (CDS) interventions—tools that combine patient information with medical knowledge to guide clinical decisions [1]—have a well-documented track record of shaping practice and patient outcomes [1, 2]. Computerized templates, which are a type of CDS, have been deployed to make evidence-based approaches to care more accessible and convenient, for example by facilitating assessment of risk factors for falls [3, 4], or referral to psychotherapy [5]. However, the mere availability of a template doesn't ensure that practitioners will use it [6]. One of the few studies to report uptake of a computerized template found that it was utilized $5\%$ of the time [7]. For templates to be useful, they must be used.
Users must be made aware of the template and its value. It must be accessible and convenient to use. It must be tailored to reflect local clinical context, and its use must be supported by the local clinical and organizational culture [8].
Implementation scientists have understood this for years, which is why so much scholarship in implementation science is devoted to (a) adequately capturing contextual factors in a given implementation [9, 10], (b) enumerating and evaluating implementation strategies to prevent useful innovations from being ignored [11], and (c) characterizing the nature of adaptations made to interventions [12, 13].
Although context is diversely defined, it generally refers to social and organizational factors occurring both narrowly within a site and broadly in the site's ecological setting, and is widely recognized for its potential role in impacting intervention effectiveness [9, 14, 15]. As Nilsen and Bernhardsson have written, “Accounting for the influence of context is necessary to explain how or why certain implementation outcomes are achieved, and failure to do so may limit the generalizability of study findings to different settings or circumstances” [9]. Contextual factors may include the culture, climate, policy, resources, and readiness for implementation of the practice setting and/or external environment [15].
Implementation strategies, or the techniques used to encourage adoption or implementation of a desired intervention, are likewise a critical element of implementation, comprising the “how to” of efforts to achieve practice change [11, 16]. Description and evaluation of implementation strategies is one of the core tasks of implementation science, supporting both replication of effective implementation efforts and progress toward a more generalizable science of implementation [16]. Meanwhile, adaptations to evidence-based interventions, and to the implementation strategies used in their delivery, are increasingly recognized as occurring frequently (if not inevitably) in scale-up and spread (17–19). Adaptations pose a provocative challenge for diffusion efforts, as they may be associated with improved or reduced intervention effectiveness, and may similarly increase or decrease likelihood of adoption and sustainment; systematic identification and evaluation of adaptations is therefore a critical undertaking [13, 17, 20].
Studies on computerized templates have often acknowledged the importance of each of these aspects of implementation (contextual factors, implementation strategies, and adaptations) [21, 22], but have rarely examined them directly. This omission is often a byproduct of the methods used to evaluate computerized templates. Implementations of templates and other CDS, when evaluated, are most frequently assessed on the basis of quantitative data alone [23, 24]. If qualitative data are collected as part of an evaluation, they are typically limited to reports from users of the tool, with the perspectives and insights of implementers not systematically documented or reported. Finally, when qualitative data are collected about EHR-based interventions, they are normatively gathered at one or two timepoints (e.g., at baseline and post-implementation), and are therefore insufficient in their ability to capture longitudinal changes in implementation strategies, intervention adaptations, and contextual factors [25].
To address these gaps, we used a convergent mixed-methods design to explore the implementation and uptake of a computerized template for cardiovascular (CV) risk reduction, with the following research questions: [1] what contextual factors emerged in implementation of the CV template across sites?; [ 2] how were implementation strategies and aspects of the CV template adapted in response to contextual factors?; and [3] how did context factors, use of implementation strategies, and adaptations coincide with differences in template use across sites and across time?
## Evidence-based intervention: The CV template
The CV template was developed in response to evidence of provider-level barriers to reducing CV risk [26]. These barriers included time constraints, a lack of awareness of current CV disease prevention guidelines, difficulty interpreting guidelines, difficulty accessing relevant patient data at point of care, and low self-efficacy to counsel patients in behavioral change (26–30). The computerized template was intended to aggregate data relevant to CV risk reduction from multiple places in the EHR, and add patient-reported information collected before the visit to enable more comprehensive screening and facilitate provider-patient discussion about each patient's CV risks and possible action steps. The template was made available for use by any provider at a participating site, and all providers were introduced to the template during a local team meeting.
This work was conducted as part of a multi-component trial in Department of Veterans' Affairs (VA) health care facilities funded by VA's Quality Enhancement Research Initiative (QUERI). The trial, called Enhancing Mental and Physical Health of Women through Engagement and Retention (EMPOWER) QUERI, focused on expanding access to important health services for women Veterans [31].
Our EMPOWER QUERI team implemented the CV template as part of a larger “CV toolkit” to identify and document cardiovascular risk screening across women Veterans and engage women in health behavior change. In addition to the CV template described above, which is the focus of this analysis, the toolkit involved two other components, each of which are described at greater length elsewhere [26]: [1] a single-page paper-based self-screener completed by patients while waiting for a primary care or women's health visit; and [2] a facilitated group for CV goal-setting adapted and gender-tailored from a program (“Gateway to Healthy Living”) developed by the VA's national Center for Health Promotion and Disease Prevention. The template and other components of the toolkit were implemented in the context of a non-randomized stepped-wedge trial aimed at engaging and retaining women Veterans in evidence-based care [31]. To maximize the applicability of findings across settings, the trial (EMPOWER QUERI) purposively recruited sites with heterogeneous size and structure, particularly with different models for delivering women's health care [31].
## Baseline implementation approach: Replicating effective programs
Replicating Effective *Programs is* an implementation framework aimed at tailoring evidence-based interventions for delivery in novel settings and/or to novel populations [32, 33]. REP was selected for this project because of its well-established evidence base and its track record of constructive application in VA implementation studies [31, 34]. REP follows a phased process in which the existing intervention is packaged for delivery in a new setting (pre-conditions phase), tailored in response to feedback from multi-level stakeholders (pre-implementation phase), implemented using a combination of training, engaging champions, and technical assistance (implementation phase), then further customized and examined for sustainability and potential spread (maintenance and evolution phase) [31, 35]. In this study we drew upon REP several times in sequence, with all but the initial “pre-conditions” phase repeated at each site.
## Data collection
Our convergent mixed-method implementation evaluation included two longitudinal data sources, periodic reflections (qualitative) and assessment of template uptake using VA administrative data (quantitative). Qualitative and quantitative data were collected in parallel over the course of the study, then analyzed and integrated as described below.
## Periodic reflections
Periodic reflections are a form of guided discussion with implementation stakeholders frequently used to document the dynamic conditions of implementation, including team activities, interactions with site and other partners, key challenges and events, and adaptations to the intervention and/or implementation strategies [25]. We conducted 39 reflections as telephone discussions with the CV template implementation team (the single, central team that initiated the overall project, including the co-PIs and project director). Reflections were conducted approximately monthly over the period before, during, and after implementation of a computerized template for cardiovascular risk at five VA facilities (Oct 2016–May 2020). Because each reflection focused on developments since the prior reflection, with alternating periods of activity and inactivity, duration of the discussions varied with the pace of the project developments (20–60 min). Reflections were facilitated by a PhD-level anthropologist, who documented discussion content in detailed, near-verbatim notes. We linked qualitative analyses with descriptive data on template use across the implementation period at all five facilities.
## Template uptake
We measured template uptake at each site by extracting data from the VA's electronic health record. Template uptake was defined as a percentage: the number of patients for whom a template was initiated by participating providers at each site, divided by the number of patients who were eligible to receive a template in that month (i.e., women Veterans who were seen and who had not had a template previously completed).
## Analysis
Our analytic process is summarized in Figure 1. As formal implementation efforts were ending, one investigator (JB) conducted initial review of reflections data to categorize text relevant to identified research questions (e.g., contextual factors, adaptations to intervention, adaptations to implementation strategies); two coders (JB, EF) then reviewed categorized text using a hybrid inductive-deductive content analysis approach. Given the relative dearth of literature identifying high-priority contextual factors in implementation of CDS, we took an inductive approach to contextual factors, independently identifying key themes emerging in the relevant data, then meeting to discuss potential themes and illustrative examples until we achieved consensus for each section of coded text. All text relevant to use of implementation strategies was first coded deductively in accordance with the Expert Recommendations for Implementation Change (ERIC) taxonomy of implementation strategies [11]; subsequently, all text descriptive of adaptations to the CV template intervention or implementation plan was coded in accordance with the Framework for Reporting Adaptations and Modifications—Expanded Version (FRAME) [20] or Framework for Reporting Adaptations and Modifications to Evidence-based Implementation Strategies (FRAME-IS) [13], respectively. Following coding, data were reviewed again to create written site summaries identifying: (i) contextual factors, (ii) adaptations to the CV template; and (iii) adaptations to implementation strategies, with approximate dates identified for discrete events. Using these site summaries, two investigators (JB, EF) independently created longitudinal displays of the factors and events most relevant to adoption of the template, i.e., “timeline maps.” The format of these maps, which include a chronological depiction of events and factors grouped into “swim lanes,” builds upon previous applications of systems thinking to program implementation [36]. The investigators then met to discuss and reconcile their timeline maps (“initial reconciliation”). The timeline maps were then reviewed by our interdisciplinary team (“member checking”) to verify the accuracy of the maps and identify additional factors viewed as salient by implementation team members, including those who participated in periodic reflections. Once initial reconciliation and member checking were complete and the team reached consensus on the timeline maps for each site, quantitative data on template uptake by month were added to each map.
**Figure 1:** *Summary of Data Analysis and Integration.*
## Results
The CV template was implemented in three waves across five sites during the period June 2017–March 2020. In sections below, we: [1] describe contextual factors emerging across sites during pre-implementation and implementation phases at each site; [2] identify adaptations to the CV template and implementation strategies, and; [3] examine template uptake and its convergence with contextual factors, use of implementation strategies, and adaptations at each clinic over time.
## Contextual factors
Four key types of contextual factors emerged inductively from our analyses: [1] the pre-existing structure of each site including the model of women's health (WH) care delivery; [2] staffing changes the occurred during implementation; [3] the relative authority of local champions; and [4] leadership external to the clinic.
## Site structure
Because the intervention was targeted at women Veterans, each site's model for delivering women's health care was a meaningful factor. Three of the five sites were stand-alone comprehensive women's health (WH) clinics, and the other two were general primary care clinics with designated women's health providers (Table 1). Within the three stand-alone women's health clinics, the implementation team aimed to engage the entire clinical staff. At the general primary care clinics, only a designated WH provider and their team nurses and medical/clerical support staff were involved with template use.
**Table 1**
| Unnamed: 0 | Model of women's health care | Local project champion(s) | Template users | Relationships among sites |
| --- | --- | --- | --- | --- |
| Site A | Stand-alone women's health clinic | Women's health clinic medical director; women's health site clinical lead | All PC teams in the women's health clinic | Shared VA health care system with Site B |
| Site B | Stand-alone women's health clinic | Women's health clinic medical director; PC team RN | All PC teams in the women's health clinic | Shared VA health care system with Site A |
| Site C | Stand-alone women's health clinic | Women's health clinic medical director; women's health program manager | All PC teams in the women's health clinic | Sole participating site within their VA health care system |
| Site D | Women's health embedded in primary care | PC deputy director; designated women's health provider; PC team RN; PC team clerk | One designated women's health PC team | Shared VA health care system with Site E |
| Site E | Women's health embedded in primary care | PC deputy director; designated women's health provider; PC team RN | One designated women's health PC team | Shared VA health care system with Site D |
## Staffing changes
In several clinics, substantial changes in clinic staffing occurred over the course of implementation. At one site (D), the person who had been designated as the sole nurse who would use the template took a leave of absence shortly after implementation. Later, the sole provider designated to use the template left the facility, and then the clinic was shut down amidst the COVID-19 pandemic. Additionally, at site B, extensive staffing changes occurred shortly before implementation, which was noted as a potential impediment: “(Site B) has had some major turnover. Thinking about adding anything to a primary care list under those conditions is not ideal.”
## Relative authority of local champions
Consistent with REP, the implementation team sought to engage local champions at each of the five sites, but the organizational position and disposition of the champions differed in important ways. At one site (Site C), the key champion had broad authority over the women's health clinic, practiced in the clinic herself, and was unusually supportive and engaged in the implementation of the template.
As the implementation team noted during reflections, “(the champion is) the women's health medical director who said yes (to implementing the template) a year ago. She said, “you're a gift.” *She is* the person who designed the women's health clinic, including the flow, and hired around that.” The site champion's strong support was reflected in a positive response from clinic members overall at that site. “ The reception was overwhelmingly good. They all came right in—when I say all, it was everybody (in the women's health clinic): the front office, the nurses, the providers, the entire team came in and met with us and watched the slide presentation and talked about it. They gave us changes to the wording on the template. They were very engaged and very excited.” By contrast, while the other four sites each had supportive champions, none of those champions had the same level of local authority (e.g., direct supervisory relationships) or such close working relationships (e.g., long-term co-location) with the clinic staff for whom the template was intended.
## External leadership
Leadership external to the clinic itself also played a key role, in some cases facilitating rapid change and in others seeming to slow desired progress. In one site, clinic staff requested that the template be accompanied by a clinical reminder to make the template easier to access and prompt its use, only to face continued opposition from a key facility-level leader who objected to a new clinical reminder that was not for all providers. Over the course of 5 months, the implementation team and local providers together made the case that a reminder would be beneficial, and ultimately persuaded the facility-level leader by arguing that the reminder would support progress on high priority performance measures tracked by the facility. Although ultimately successful, resistance from leadership resulted in significant delay in CV template modifications.
In another site, the involvement of a (high-level facility leader) was instrumental in engaging clinical application coordinators (CACs) to execute technical changes to the template. “( The CACs told us) “we're part of (the leader's) group over here,” … So she has a leadership role there…and in the (research unit)…and the school of medicine because she's a provider. She's—besides being incredibly smart—very powerful there, so we're very lucky that she's backing us. And she's been backing us from the first, 5 years ago, but I didn't understand that support until everybody in the clinic mentioned (her)– there's a power there…that's going to help get things done.”
## Adaptations to intervention and implementation strategies
All sites received the phased REP implementation approach, including the strategies of pre-implementation tailoring of the CV template, identifying and engaging champions, and providing ongoing technical assistance during the implementation phase. Over the course of implementation, adaptations were made to both the CV template and to the use of implementation strategies at sites, including both planned changes and changes that were unplanned but emerged as a result of local events and factors occurring at the sites (“responsive”). Table 2 provides a summary of adaptations and the sites where they occurred.
**Table 2**
| Adaptations | Adaptations.1 | When the modification was made | Planned vs. Responsive | Who determined the modification should be made | What is modified | Nature of modification | Goal of the modification |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Adaptations to CV Template | Tailor to local resources | Site A: Pre-imp Site B: Pre-imp Site C: Pre-imp Site D: Pre-imp Site E: Pre-imp | Planned | Implementation team + users | Content | Tailoring | Improve fit |
| | Re-customization | Site A: Imp Site B: Imp Site C: Pre-imp Site D: Pre-imp Site E: Pre-imp | Planned | Implementation team + users | Content | Shortening; Reordering; Refining | Improve fit, increase satisfaction |
| | Split template into two (nurse component + provider component) | Site A: Imp Site B: Imp Site C: Pre-imp Site D: Pre-imp Site E: Pre-imp | Responsive | Site lead | Context | Setting and Personnel | Improve fit |
| | Clinical Reminder | Site A: Imp Site B: Imp Site C: Pre-imp Site D: Pre-imp Site E: Pre-imp | Responsive | Individual practitioners | Implementation | – | Provide prompt |
| Adaptations to the REP Implementation Approach | Academic detailing | Site A: Imp Site B: Imp Site C: N/A Site D: N/A Site E: N/A | Responsive | Implementation team | Content | Integration of another strategy | Increase provider motivation/self-efficacy |
| | Creation of cheat sheets | Site A: Imp Site B: Imp Site C: N/A Site D: N/A Site E: N/A | Responsive | Implementation team | Content | Integration of another strategy | Increase provider self-efficacy |
| | Small scale pilot testing | Site A: N/A Site B: N/A Site C: N/A Site D: Pre-imp Site E: Pre-imp | Responsive | Site lead | Content | Integration of another strategy | Staged implementation |
Planned adaptations of the CV template began with tailoring to local resources. Because each VA facility offers a different array of programs for CV risk management, the template was tailored to accurately reflect those resources, allowing providers to make patient referrals appropriate to the local setting. A second planned adaptation of the template focused on re-customizing to meet sites' local workflows. The implementation team solicited input from local champions and other template users about the usability of the template and ways to better match the template to local workflows; resulting changes included a reduction in the number of template fields that were mandatory, consolidation of potentially redundant fields describing patients, and other modifications intended to streamline the template.
Interestingly, two unplanned, responsive adaptations of the template emerged from discussions around tailoring and customization. The first of these adaptations involved splitting the template into two separate components. A local program champion, in preparation for implementation, noted that the template could be adapted to better reflect the team-based care delivered at her facility. She suggested that the work of entering information from the written screener into the EHR and answering patient questions about the screener could be done by a nurse before the provider arrived to help patients set goals and make referrals to relevant programs. The template was therefore divided into two components to reflect local workflow patterns: [1] a nurse-facing template that mirrored the patient screener, allowing the nurse to enter patient data and document CV risks; and [2] a provider-focused template that encouraged the provider to communicate with the patient about prioritizing CV risks, identify action steps for reducing risks (e.g., smoking cessation), and offer potential referrals to support health behavior change.
A second responsive adaptation occurred following a request that template completion be facilitated by the prompt of an electronic clinical reminder. In pre-trial pilot work to develop the template, clinical stakeholders had specifically noted that they were overburdened by clinical reminders and did not want another added [26]. As a result, the implementation team was surprised when front-line clinicians at multiple sites requested that the template be facilitated by an electronic reminder. “ I think the biggest surprise was that the nurse who does the front end, the one who does the vitals and everything, she looked up and said, “is there any way you could make this a reminder? Because it's easier on us if you just make it a reminder.”” After the reminder was implemented and positively received at one site, it effectively became a site-level “menu option” for the others, all of whom eventually elected to incorporate the reminder. This was accomplished by working with site-level EHR administrators who were able to target the reminder at the site's designated women's health providers.
Finally, adaptations were also made to the planned use of implementation strategies, particularly in the first two sites, where CV template use was slow to get off the ground after launch. At two of the sites, the implementation team conducted academic detailing: attending regular clinical meetings and encouraging the use of the template, soliciting feedback about it, and offering strategies for its use. “( Implementation lead) goes to the monthly meetings, so she did that for (site) last month, really pushing to get the trainees to use the template….” At the same two sites, the implementation lead also worked with clinical champions to develop brief cheat sheets, or written instructions that were affixed to clinic computer monitors, to remind and assist providers and staff in using the template.
Finally, at a later site, the implementation team adopted small-scale pilot testing in response to a site's concern about expanding template use across the clinic prior to conducting a small trial first. “ Their main concern was for the nurses' time in putting the part 1 screener into the template … We decided at the end of the call that we would only have (a nurse) do the template for (a single provider's) patients, and pilot with them first, and then discuss with the other nurses.”
## Template uptake: Site-level implementation
Descriptions below provide a brief summary of overall site-level template uptake, examining the longitudinal course of contextual factors, implementation strategies, adaptations, and implementation progress over time at each site.
## Site A
Site A (Figure 2) had relatively low overall uptake of the template (Mean $3\%$, SD $2\%$). After a ten-month initial period following template launch where uptake remained close to zero, two changes were made: a clinical reminder was introduced and the template was split into a nurse-facing template focused on assessing CV risk, and a provider-facing template focused on goal-setting and referrals. A modest increase in template use was observed immediately following these changes. This site was the first to implement the template and had the longest cumulative exposure to the template.
**Figure 2:** *Site A Timeline Map.*
## Site B
Site B (Figure 3) also had low overall template uptake (Mean $3\%$, SD $4\%$). Similar to Site A, template uptake at site B was very low until a reminder was introduced and the template was split into two parts, but the modest increase in uptake was temporary. Though staffing in the women's health clinic was relatively stable during the implementation period, substantial turnover had occurred shortly before implementation: “…three providers have changed, three (clerks) have changed, the nurse has changed, a new LVN has changed, two psychiatrists have gone, the others are there but are part-time. ( The clinics) have been waylaid by mental health issues from the get-go.” Site B, while geographically distinct from Site A, belongs to the same VA health care system, with shared organizational leadership.
**Figure 3:** *Site B Timeline Map.*
## Site C
Template uptake at Site C (Figure 4) (mean $18\%$, SD $7\%$) was consistently higher than at sites A and B, and increased slowly but substantially after technical assistance began and a reminder for the second portion of the template was implemented. “ The first screener went on as a clinical reminder immediately, and then this last time they said it would be nice if the provider part came up as a clinical reminder too.” A year after the second reminder was implemented, utilization returned to its pre-reminder level. Of note, splitting the template into two parts and supporting implementation with clinical reminders were innovations/adaptations that emerged first at Site C and later spread to all other sites.
**Figure 4:** *Site C Timeline Map.*
## Site D
Template uptake at Site D (Figure 5) was relatively low (mean $8\%$, SD $6\%$). At sites D and E, facility leadership was concerned about the potential time burden that the template would impose and elected to limit the initial implementation of the template to a single primary care team as a small-scale pilot. Template uptake was moderate and highly variable. One of two nurses who had been designated to use the template took a leave of absence shortly after implementation, and her absence was accompanied by a marked decrease in template use. Later, the sole provider designated to use the template left the facility, and the clinic was shut down amidst the COVID-19 pandemic.
**Figure 5:** *Site D Timeline Map.*
## Site E
The (Figure 6) overall level of template uptake at this site (mean $28\%$, SD $13\%$) was substantially higher than at other facilities, and early changes in template use (e.g., a brief spike in uptake above and beyond already-high uptake) did not appear to correspond to events or activities known to the project team. Site E belongs to the same health care system as Site D, and as such shares organizational leadership. Accordingly, the organizational leaders' decision to use small scale pilot testing (with only one care team exposed to the intervention) applied to site E as well as site D. Implementation of the CV template closed ahead of schedule in March 2020 due to COVID-19.
**Figure 6:** *Site E Timeline Map.*
## Template uptake: Cross-site comparison
Notably, there was meaningful heterogeneity of CV template utilization (Figure 7) even among sites within a single organization (VA) and targeting a single population (women Veterans). Heterogeneity occurred across sites in rate of initial uptake, timing and reach of peak uptake, and trajectory of uptake over time.
**Figure 7:** *Cross-Site Comparison of Template Utilization by Month.*
Implementation across sites occurred in three waves, with one initial site followed by two sites beginning ~10 months later, followed by two additional sites a year later. The timing of waves does not appear to have had significant cross-site effects, as each of the latter waves saw both comparatively high and low performers.
That said, later sites exhibited greater template uptake immediately post-launch, which may reflect incorporation from the beginning of adaptations developed during implementation at earlier sites. All of the sites exhibited variability in utilization over the months of implementation, with apparent convergence between level of utilization and disrupted staffing (reduced template use), overall clinic and leadership buy-in (reduced or enhanced template use), and the onset of COVID-19 and countermeasures (reduced or halted template use).
## Discussion
The current analysis integrated convergent, longitudinal, mixed-method data to examine contextual factors and adaptations associated with implementation of a clinical decision support tool (the CV template) for reducing cardiovascular risk among women Veterans. Our use of timeline maps as site-specific longitudinal qualitative/quantitative displays, along with the use of periodic reflections [25] to capture ongoing insights from implementers, provides a novel approach for assessing implementation of evidence-based interventions and both planned and emergent adaptations. These findings offer a number of insights with implications for design of future CDS implementation and evaluation.
Perhaps unsurprisingly, the contextual factors that emerged as most influential in these findings were related to each site's resources for change (e.g., staffing) and leadership buy-in. Three sites (A, B, D) experienced significant staffing challenges, either immediately prior to implementation launch or during the implementation period, and all saw disappointing template uptake in the months following the shortage. This is consistent with prior studies identifying availability of adequate staff as an important factor shaping capacity for novel change efforts (37–40), particularly given that adoption of new techniques and technologies typically requires additional time and cognitive demand [what Reed et al. [ 41] refer to as “headroom”] in the period until changes are fully integrated and become routine.
Although champions are widely recognized as a critical component of implementation success [42, 43], these data illustrate the importance of ensuring that site champions have adequate organizational and/or relational authority to support change efforts. The broader importance of leadership buy-in was illustrated in both positive (Site C) and negative (Site A) directions, with leadership support in Site C helping to facilitate adaptation, in the form of implementing clinical reminders to support uptake of the split template, and leadership reluctance in Sites A and B resulting in an extended period of delay before that same adaptation could be put in place. The late-breaking crisis of COVID-19 emerged, too, as an illustration of how acute system shocks can fully derail routine practice, let alone practice change efforts.
These data identified several adaptations to the CV template, taking both planned and responsive forms. Planned adaptations, based in the REP implementation framework, included tailoring and re-customization at each site in dialogue with site-level partners. In exploring the more emergent adaptations we identified, we adapted FRAME language to describe these adaptations as responsive (in place of the original FRAME term, “reactive”) to better reflect the intentional and engaged nature of adaptations made in dialogue with sites. These responsive adaptations included both splitting the intervention into two components to allow for a better fit with clinic workflow and integrating computerized reminders to use the template. Both of these adaptations occurred initially in one site but were later offered to and adopted by all four other sites. This provides an excellent example of how adaptations to an evidence-based intervention can be positive and can improve acceptability and feasibility in implementation [17, 18, 44], and may be seen as arguing for the value of formative evaluation in collaboration with implementing sites, particularly during periods of early spread [41, 45, 46]. The fact that all sites saw increased use of the CV template following introduction of the clinical reminder underscores the potential value of a “prompt” in achieving consistent behavior change (47–49). The finding that some sites saw a significantly smaller increase than others in template use following introduction of the reminder is consistent with a prior Cochrane review [50], and suggests that even effective implementation strategies and adaptations may be less impactful in settings where context is less supportive of practice change, whether due to inadequate staffing or other challenges [44, 51].
REP as an implementation framework can be viewed as a bundled set of implementation strategies, and in prior work we have noted that REP-specified activities comprise at least 19 distinct implementation strategies [35]. Even so, examination of these data allowed for identification of three additional implementation strategies introduced by the implementation team in response to site-level challenges. These included academic detailing and creation of “cheat sheets” for providers in two sites, in an effort to bolster providers' motivation and self-efficacy for utilization of the template, and use of a small-scale piloting approach in another site, where concern was expressed regarding the feasibility of template adoption in a busy clinic. It is worth noting that these strategies emerged in response to local challenges, and were not, in this small sample, typically spread to other sites; moreover, these strategy adaptations were not always successful in achieving a significant increase in template uptake. For both adaptations to the intervention and to implementation strategies, the FRAME and FRAME-IS frameworks provided a thoughtful structure for considering the form and intended function of adaptations, once more demonstrating their analytic utility in implementation evaluation. Use of these frameworks as part of the timeline mapping analysis was particularly valuable in highlighting when adaptations occurred at each site, and whether observable changes in template uptake occurred in subsequent months. Recent contributions to the literature on adaptation in implementation science acknowledge the methodological challenges of assessing adaptations' impact [19, 44, 52], which remain a roadblock to more generalizable understanding of adaptation in the context of implementation [18]. These findings and the timeline mapping method provide an example of how innovative use of integrative methods can facilitate evaluation of site-level impact of adaptations over the life course of implementation.
Strengths of this analysis include integration of convergent mixed-method data on template uptake with regular, longitudinal reflections by the implementation team on ongoing events, contextual factors, implementation activities, and adaptations occurring at each site. The timeline mapping approach offers a pragmatic method for examining the longitudinal trajectory of implementation at site and cross-site levels, providing a multi-level perspective on what is happening in implementation, and avoiding the weaknesses of implementation evaluations that rely solely on outcomes gathered at isolated moments in time and may inadvertently obscure key events. In doing so, the use of timeline mapping also answers the call to “embrace a richer and more diverse methodological repertoire when researching complex systems,” [53] by directing attention to learning across sites and the interrelationships among contextual factors and adaptations. Limitations of this approach include the reliance on implementation team perspectives, which may overly bias site-level factors rather than individual provider behavior. Future research should examine integration of individual interviews with providers and clinic staff in order to further assess the accuracy of implementation teams' sensemaking around implementation progress, and to consider the relationships between provider and staff perspectives, implementation team perspectives, and the longitudinal course of implementation uptake as demonstrated by quantitative data [54].
## Conclusions
Heterogeneity in uptake of CDS across sites is widespread but poorly understood. Our analysis used longitudinal joint displays of quantitative and qualitative data to identify key contributors to variable uptake across sites and over time, including contextual factors, active adaptation of the CV template and implementation strategies, and activities and events temporally associated with increases or decreases in template utilization at the site level. These findings underscore the importance of adaptive approaches to implementation, allowing for iterative, intentional shifts in intervention and strategy to meet the needs of individual sites, as well as the value of integrating mixed-method data sources in conducting longitudinal evaluation of implementation efforts.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by VA Central IRB. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.
## Author contributions
JB, BB-M, MF, AH, and EF contributed to the conception and design of the study. CC-C, CT, MF, BB-M, and JB performed statistical analyses. JB and EF performed qualitative analyses. JB wrote the first draft of the manuscript. All authors contributed to manuscript revision, read, and approved the submitted version.
## Funding
EMPOWER was funded by the VA Quality Enhancement Research Initiative (QUERI; grant number 15-272). AH is supported by a VA HSR&D Research Career Scientist Award (RCS 21-135).
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Sutton RT, Pincock D, Baumgart DC, Sadowski DC, Fedorak RN, Kroeker KI. **An overview of clinical decision support systems: benefits, risks, and strategies for success**. *Npj Digit Med.* (2020) **3** 17. DOI: 10.1038/s41746-020-0221-y
2. Kawamoto K, Houlihan CA, Balas EA, Lobach DF. **Improving clinical practice using clinical decision support systems: a systematic review of trials to identify features critical to success**. *BMJ.* (2005) **330** 765. DOI: 10.1136/bmj.38398.500764.8F
3. Phelan EA, Aerts S, Dowler D, Eckstrom E, Casey CM. **Adoption of evidence-based fall prevention practices in primary care for older adults with a history of falls**. *Front Public Health* (2016) 4. DOI: 10.3389/fpubh.2016.00190
4. Fung CH. **Computerized condition-specific templates for improving care of geriatric syndromes in a primary care setting**. *J Gen Intern Med.* (2006) **21** 989-94. DOI: 10.1111/j.1525-1497.2006.00522.x
5. Sripada RK, Pfeiffer PN, Rauch SAM, Ganoczy D, Bohnert KM. **Factors associated with the receipt of documented evidence-based psychotherapy for PTSD in VA**. *Gen Hosp Psychiatry.* (2018) **54** 12-7. DOI: 10.1016/j.genhosppsych.2018.07.001
6. Greenes RA, Bates DW, Kawamoto K, Middleton B, Osheroff J, Shahar Y. **Clinical decision support models and frameworks: Seeking to address research issues underlying implementation successes and failures**. *J Biomed Inform.* (2018) **78** 134-43. DOI: 10.1016/j.jbi.2017.12.005
7. Fielstein EM, Brown SH, McBrine CS, Clark TK, Hardenbrook SP, Speroff T. **The effect of standardized, computer-guided templates on quality of VA disability exams**. *AMIA Annu Symp Proc.* (2006) **2006** 249-53. PMID: 17238341
8. Van de Velde S, Heselmans A, Delvaux N, Brandt L, Marco-Ruiz L, Spitaels D. **A systematic review of trials evaluating success factors of interventions with computerised clinical decision support**. *Implementation Sci.* (2018) **13** 114. DOI: 10.1186/s13012-018-0790-1
9. Nilsen P, Bernhardsson S. **Context matters in implementation science: a scoping review of determinant frameworks that describe contextual determinants for implementation outcomes**. *BMC Health Serv Res.* (2019) **19** 189. DOI: 10.1186/s12913-019-4015-3
10. Balasubramanian BA, Heurtin-Roberts S, Krasny S, Rohweder CL, Fair K, Olmos-Ochoa TT. **Factors related to implementation and reach of a pragmatic multisite trial: the my own health report (MOHR) study**. *J Am Board Fam Med.* (2017) **30** 337-49. DOI: 10.3122/jabfm.2017.03.160151
11. Powell BJ, Waltz TJ, Chinman MJ, Damschroder LJ, Smith JL, Matthieu MM. **A refined compilation of implementation strategies: results from the expert recommendations for implementing change (ERIC) project**. *Impl Sci* (2015) 10. DOI: 10.1186/s13012-015-0209-1
12. Stirman SW, Miller CJ, Toder K, Calloway A. **Development of a framework and coding system for modifications and adaptations of evidence-based interventions**. *Impl Sci* (2013) 8. DOI: 10.1186/1748-5908-8-65
13. Miller CJ, Barnett ML, Baumann AA, Gutner CA, Wiltsey-Stirman S. **The FRAME-IS: a framework for documenting modifications to implementation strategies in healthcare**. *Implementation Sci.* (2021) **16** 36. DOI: 10.1186/s13012-021-01105-3
14. Tomoaia-Cotisel A, Scammon DL, Waitzman NJ, Cronholm PF, Halladay JR, Driscoll DL. **Context matters: the experience of 14 research teams in systematically reporting contextual factors important for practice change**. *Ann Family Med* (2013) **11** 115-23. DOI: 10.1370/afm.1549
15. Squires JE, Graham ID, Hutchinson AM, Michie S, Francies J, Sales A. **Identifying the domains of context important to implementation science: a study protocol**. *Impl Sci* (2015) 10. DOI: 10.1186/s13012-015-0325-y
16. Proctor EK, Powell BJ, McMillen JC. **Implementation strategies: recommendations for specifying and reporting**. *Impl Sci* (2013) 8. DOI: 10.1186/1748-5908-8-139
17. Chambers DA, Glasgow RE, Stange KC. **The dynamic sustainability framework: addressing the paradox of sustainment amid ongoing change**. *Impl Sci* (2013) 8. DOI: 10.1186/1748-5908-8-117
18. Chambers DA, Norton WE. **The Adaptome**. *Am J Prev Med.* (2016) **51** S124-31. DOI: 10.1016/j.amepre.2016.05.011
19. Holtrop JS, Gurfinkel D, Nederveld A, Phimphasone-Brady P, Hosokawa P, Rubinson C. **Methods for capturing and analyzing adaptations: implications for implementation research**. *Implementation Sci.* (2022) **17** 51. DOI: 10.1186/s13012-022-01218-3
20. Wiltsey Stirman S, Baumann AA, Miller CJ. **The FRAME: an expanded framework for reporting adaptations and modifications to evidence-based interventions**. *Implementation Sci.* (2019) **14** 58. DOI: 10.1186/s13012-019-0898-y
21. Thaker VV, Lee F, Bottino CJ, Perry CL, Holm IA, Hirschhorn JN. **Impact of an electronic template on documentation of obesity in a primary care clinic**. *Clin Pediatr.* (2016) **55** 1152-9. DOI: 10.1177/0009922815621331
22. Harry ML, Truitt AR, Saman DM, Henzler-Buckingham HA, Allen CI, Walton KM. **Barriers and facilitators to implementing cancer prevention clinical decision support in primary care: a qualitative study**. *BMC Health Serv Res.* (2019) **19** 534. DOI: 10.1186/s12913-019-4326-4
23. Bright TJ, Wong A, Dhurjati R, Bristow E, Bastian L, Coeytaux RR. **Effect of clinical decision-support systems: a systematic review**. *Ann Intern Med.* (2012) **157** 29. DOI: 10.7326/0003-4819-157-1-201207030-00450
24. Pawloski PA, Brooks GA, Nielsen ME, Olson-Bullis BA. **A systematic review of clinical decision support systems for clinical oncology practice**. *J Nat Comp Cancer Network.* (2019) **17** 331-8. DOI: 10.6004/jnccn.2018.7104
25. Finley EP, Huynh AK, Farmer MM, Bean-Mayberry B, Moin T, Oishi SM. **Periodic reflections: a method of guided discussions for documenting implementation phenomena**. *BMC Med Res Methodol* (2018) **18** 1-5. DOI: 10.1186/s12874-018-0610-y
26. Bean-Mayberry B, Moreau J, Hamilton AB, Yosef J, Joseph NT, Batuman F. **Cardiovascular risk screening among women veterans: identifying provider and patient barriers and facilitators to develop a clinical toolkit**. *Women's Health Issues.* (2022) **32** 284-92. DOI: 10.1016/j.whi.2021.12.003
27. Cabana MD, Kim C. **Physician adherence to preventive cardiology guidelines for women**. *Women's Health Issues.* (2003) **13** 142-9. DOI: 10.1016/S1049-3867(03)00034-3
28. Crosson JC, Heisler M, Subramanian U, Swain B, Davis GJ, Lasser N. **Physicians' perceptions of barriers to cardiovascular disease risk factor control among patients with diabetes: results from the translating research into action for diabetes (TRIAD) study**. *J Am Board Family Med.* (2010) **23** 171-8. DOI: 10.3122/jabfm.2010.02.090125
29. Kedward J, Dakin L. **A qualitative study of barriers to the use of statins and the implementation of coronary heart disease prevention in primary care**. *Br J Gen Pract* (2003) **53** 684-9. PMID: 15103875
30. Tsui J, Dodson K, Jacobson T. **Cardiovascular disease prevention counseling in residency: resident and attending physician attitudes and practices**. *J Natl Med Assoc.* (2004) **96** 1080-3. PMID: 15303414
31. Hamilton AB, Farmer MM, Moin T, Finley EP, Lang AJ, Oishi SM. **Enhancing mental and physical health of women through engagement and retention (EMPOWER): a protocol for a program of research**. *Impl Sci* (2017) 12. DOI: 10.1186/s13012-017-0658-9
32. Kilbourne AM, Neumann MS, Pincus HA, Bauer MS, Stall R. **Implementing evidence-based interventions in health care: application of the replicating effective programs framework**. *Impl Sci* (2007) 2. DOI: 10.1186/1748-5908-2-42
33. Kilbourne AM, Glasgow RE, Chambers DA. **What can implementation science do for you? Key success stories from the field**. *J Gen Intern Med.* (2020) **35** 783-7. DOI: 10.1007/s11606-020-06174-6
34. Kilbourne AM, Abraham KM, Goodrich DE, Bowersox NW, Almirall D, Lai Z. **Cluster randomized adaptive implementation trial comparing a standard versus enhanced implementation intervention to improve uptake of an effective re-engagement program for patients with serious mental illness**. *Impl Sci* (2013) 8. DOI: 10.1186/1748-5908-8-136
35. Huynh AK, Hamilton AB, Farmer MM, Bean-Mayberry B, Stirman SW, Moin T. **A pragmatic approach to guide implementation evaluation research: strategy mapping for complex interventions**. *Front Public Health* (2018) 6. DOI: 10.3389/fpubh.2018.00134
36. 36.Guide to Timeline Mapping. FSG. (2021). Available online at: https://www.fsg.org/wp-content/uploads/2021/08/Guide-to-Timeline-Mapping.pdf (accessed July 05, 2022).. (2021)
37. Kane JC, Adaku A, Nakku J, Odokonyero R, Okello J, Musisi S. **Challenges for the implementation of world health organization guidelines for acute stress, PTSD, and bereavement: a qualitative study in Uganda**. *Implementation Sci.* (2015) **11** 36. DOI: 10.1186/s13012-016-0400-z
38. McAteer J, Stone S, Fuller C, Michie S. **Using psychological theory to understand the challenges facing staff delivering a ward-led intervention to increase hand hygiene behavior: a qualitative study**. *Am J Infect Control.* (2014) **42** 495-9. DOI: 10.1016/j.ajic.2013.12.022
39. Geerligs L, Rankin NM, Shepherd HL, Butow P. **Hospital-based interventions: a systematic review of staff-reported barriers and facilitators to implementation processes**. *Implementation Sci.* (2018) **13** 36. DOI: 10.1186/s13012-018-0726-9
40. Jones A. **Implementation of hospital care pathways for patients with schizophrenia**. *J Nurs Manag.* (2000) **8** 215-25. DOI: 10.1046/j.1365-2834.2000.00175.x
41. Reed JE, Howe C, Doyle C, Bell D. **Simple rules for evidence translation in complex systems: a qualitative study**. *BMC Med.* (2018) **16** 92. DOI: 10.1186/s12916-018-1076-9
42. Bonawitz K, Wetmore M, Heisler M, Dalton VK, Damschroder LJ, Forman J. **Champions in context: which attributes matter for change efforts in healthcare?**. *Implementation Sci.* (2020) **15** 62. DOI: 10.1186/s13012-020-01024-9
43. Demes JAE, Nickerson N, Farand L, Montekio VB, Torres P, Dube JG. **What are the characteristics of the champion that influence the implementation of quality improvement programs?**. *Eval Program Plann.* (2020) **80** 101795. DOI: 10.1016/j.evalprogplan.2020.101795
44. Rabin BA, McCreight M, Battaglia C, Ayele R, Burke RE, Hess PL. **Systematic, multimethod assessment of adaptations across four diverse health systems interventions**. *Front Public Health.* (2018) **6** 102. DOI: 10.3389/fpubh.2018.00102
45. Stetler C, Damschroder LJ, Helfrich CD, Hagedorn HJ. **Guide for applying a revised version of the PARIHS framework for implementation**. *Implementation Science.* (2011) **2011** 6. DOI: 10.1186/1748-5908-6-99
46. Pestka DL, White KM, DeRoche KK, Benson BJ, Beebe TJ. **Trying to fly the plane while we were building it. Applying a learning health systems approach to evaluate early-stage barriers and facilitators to implementing primary care transformation: a qualitative study**. *BMJ Open.* (2022) **12** e053209. DOI: 10.1136/bmjopen-2021-053209
47. Agha S, Tollefson D, Paul S, Green D, Babigumira JB. **Use of the fogg behavior model to assess the impact of a social marketing campaign on condom use in Pakistan**. *J Health Commun.* (2019) **24** 284-92. DOI: 10.1080/10810730.2019.1597952
48. Fogg BJ. *Tiny Habits: The Small Changes That Change Everything* (2019)
49. West R, Michie S. **A brief introduction to the COM-B Model of behaviour and the PRIME theory of motivation**. *Qeios* (2020). DOI: 10.32388/WW04E6
50. Shojania K, Jennings A, Mayhem A, Ramsay C, Eccles M, Grimshaw J. **The effects of on-screen, point of care computer reminders on process and outcomes of care**. *Cochrane Database Syst Rev.* (2011) **8** CD001096. DOI: 10.1002/14651858.CD001096.pub2
51. Waltz TJ, Powell BJ, Fernández ME, Abadie B, Damschroder LJ. **Choosing implementation strategies to address contextual barriers: diversity in recommendations and future directions**. *Implementation Sci.* (2019) **14** 42. DOI: 10.1186/s13012-019-0892-4
52. Kirk MA, Moore JE, Wiltsey Stirman S, Birken SA. **Towards a comprehensive model for understanding adaptations' impact: the model for adaptation design and impact (MADI)**. *Implementation Sci.* (2020) **15** 56. DOI: 10.1186/s13012-020-01021-y
53. Greenhalgh T, Papoutsi C. **Studying complexity in health services research: desperately seeking an overdue paradigm shift**. *BMC Med.* (2018) **16** 95. DOI: 10.1186/s12916-018-1089-4
54. Mills T, Lawton R, Sheard L. **Advancing complexity science in healthcare research: the logic of logic models**. *BMC Med Res Methodol.* (2019) **19** 55. DOI: 10.1186/s12874-019-0701-4
|
---
title: Adapting in-person diabetes group visits to a virtual setting across federally
qualified health centers
authors:
- Daisy Nuñez
- Diana Marino-Nuñez
- Erin M. Staab
- Tracy Dinh
- Mengqi Zhu
- Wen Wan
- Cynthia T. Schaefer
- Amanda Campbell
- Michael T. Quinn
- Arshiya A. Baig
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012803
doi: 10.3389/frhs.2022.961073
license: CC BY 4.0
---
# Adapting in-person diabetes group visits to a virtual setting across federally qualified health centers
## Abstract
Diabetes group visits (GVs) have been shown to improve glycemic control, enrich patient self-care, and decrease healthcare utilization among patients with type 2 diabetes mellitus (T2DM). While telehealth has become routine, virtual GVs remain understudied, especially in federally qualified health centers (FQHCs). We conducted a 5-year cluster randomized trial with a waitlist control group to test the impact of diabetes GVs on patients' outcomes in Midwestern FQHCs. Due to COVID-19, the 6 waitlisted FQHCs adapted to virtual GVs. FQHC staff were provided training and support to implement virtual GVs. The GV intervention included 6 monthly 1–1.5-h long education sessions and appointments with a primary care provider. We measured staff perspectives and satisfaction via GV session logs, monthly webinars, and staff surveys and interviews. Adaptations for implementation of virtual GV included: additional staff training, video conferencing platform use, decreased session length and group size, and adjusting study materials, activities, and provider appointments. Sites enrolled a total of 48 adults with T2DM for virtual GVs. Most FQHCs were urban and all FQHCs predominantly had patients on public insurance. Patients attended 2.1 ± 2.2 GVs across sites on average. Thirty-four patients ($71\%$) attended one or more virtual GVs. The average GV lasted 79.4 min. Barriers to virtual GVs included patient technology issues and access, patient recruitment and enrollment, and limited staff availability. Virtual GV facilitators included providing tablets, internet access from the clinic, and technical support. Staff reported spending on average 4.9 h/week planning and implementing GVs (SD = 5.9). On average, 6 staff from each FQHC participated in GV training and 1.2 staff reported past GV experience. All staff had worked at least 1 year at their FQHC and most reported multiple years of experience caring for patients with T2DM. Staff-perceived virtual GV benefits included: empowered patients to manage their diabetes, provided patients with social support and frequent contact with providers, improved relationships with patients, increased team collaboration, and better patient engagement and care-coordination. Future studies and health centers can incorporate these findings to implement virtual diabetes GVs and promote accessible diabetes care.
## Introduction
Diabetes mellitus (DM) affects 30 million people in the U.S. [1]. Type 2 DM (T2DM) accounts for 90–$95\%$ of cases of diabetes in adults [1]. Adults with T2DM often face co-morbid chronic diseases [2, 3]. The prevalence of diabetes is disproportionately higher among Hispanics ($12.5\%$) and African-Americans ($11.7\%$) compared to non-Hispanic whites ($7.5\%$) [1]. Hispanics and African-Americans have higher rates of diabetes-related complications, including amputations and CKD (4–6).
Federally qualified health centers (FQHCs) treat a larger proportion of patients with diabetes than other primary care physician offices [7]. FQHCs also serve a high number of vulnerable patient populations, including patients of low socio-economic status (SES) and racial minorities [8], which have been disproportionately impacted by the pandemic. Research has shown that around $70\%$ of patients in FQHCs have uncontrolled hemoglobin A1c values [9]. Given this, FQHCs must optimize diabetes care to address population health needs.
The complex nature of diabetes care requires patients to sustain healthy lifestyle practices, manage their medications, and attend multiple provider visits. Diabetes group visits (GVs) provide an alternative form of diabetes care that consists of shared appointments with a diabetes educator in a group setting and an individual visit with a primary care provider [10]. In this way, GVs add to the education and social support common to Diabetes Self-Management Education and Support (DSME) by incorporating a comprehensive medical visit to promote diabetes self-management. Diabetes GVs have been shown to effectively reduce hemoglobin A1c, improve self-management, and promote preventative care among patients (11–13). Despite the efficacy of diabetes GVs in improving patient outcomes and high staff satisfaction with GVs [13, 14], widespread integration of GVs into standard diabetes care in FQHCs remains limited.
The pandemic has required significant workflow modifications across FQHCs, such as increased telehealth visits to prevent the spread of this communicable disease [15]. Telehealth visits play a critical role in the continuum of care for patients with multi-morbid chronic conditions, including diabetes [16]. FQHCs utilized the opportunity to implement virtual diabetes GVs to adapt an effective care model to the trends of telehealth as well as increase the accessibility of diabetes care. Virtual GVs encountered barriers to implementation similar to individual telehealth visits, including technological access, resistance to change in clinical practice and cost challenges [17].
There is limited research that has systematically implemented and evaluated virtual GVs for adults with DM in the primary care setting. The aim of this research study was to adapt the diabetes GV research model to a virtual setting and to understand staff perspectives around the benefits, barriers, and facilitators to implementing virtual diabetes GVs across FQHCs.
## Design
We conducted a cluster randomized trial with a waitlist control arm to test the impact of diabetes GVs on patients' outcomes in Midwestern FQHCs. The intervention framework is motivated by observed needs across four components in diabetes care: individual medical assessment, patient education, social support, and self-management. The University of Chicago research team partnered with the Midwest Clinicians' Network (MWCN), a non-profit corporation with membership consisting of FQHCs in ten Midwestern states, to conduct this trial. After an 18-month trial comparing GVs to usual care, FQHCs in the waitlist control arm received the intervention. Due to the COVID-19 pandemic, this intervention was modified to a virtual format. In this paper, we report only on the waitlist control arm's experience implementing virtual GVs. Results of the initial trial showed improved diabetes distress, social support, care knowledge, self-care, care self-efficacy, and quality of life among patients highly engaged in GVs and a text-messaging program across an in-person and virtual cohort. Further results from the initial trial will be reported separately.
## FQHC recruitment and training
FQHCs were recruited through the MWCN and filled out an application form to be included in the study. Applications were reviewed for FQHC characteristics, such as patient population, prevalence of T2DM among their patient panel, and form of patient insurance.
Sixteen FQHCs were randomized, 8 were assigned to the intervention and the remaining 8 were assigned to the waitlist control arm. Of the 8 FQHCs in the waitlist control arm, 3 withdrew, leaving 5 FQHCs in the control arm. FQHC 4 had two separate sites (sites 4a and 4b) participate in the study for a total of 6 sites. Each FQHC site needed to assemble an organizing team of three to four staff with at least one medical provider (e.g., physician, advanced practice nurse, or physician assistant). Originally, sites 4a and 4b had separate teams for in-person GVs, but for virtual GV implementation the same staff conducted GVs for both sites.
After 18 months, FQHCs in the waitlist control arm received training through a one-and-a-half day in-person training session in Chicago on how to conduct in-person group visits. At the session in early March 2020, staff from the University of Chicago and MWCN educated FQHC staff on GV structure and implementation, patient and staff recruitment, and potential barriers to GV implementation and success. However, prior to recruiting patients, due to the COVID-19 pandemic, the waitlist control arm from our trial had to quickly adapt to a virtual format. FQHC staff received 6 additional training webinars. There were 19 training and technical assistance webinars that lasted 1–1.5 h over the course of 15 months. We invited a clinical psychologist with experience leading virtual group therapy to present on effective utilization of telehealth services for groups. We also invited a pediatric endocrinologist and her research team to present on virtual type 1 diabetes group sessions (18–20). The research study team also reviewed research literature on benefits of virtual GVs, compiled tips for onboarding patients, created virtual GV planning worksheets, and shared ideas to inform staff training on implementing virtual GVs. FQHC staff were also trained on accessing REDCap, a secure web platform for building and managing online databases and surveys, to enter data and distribute surveys and enrollment forms. Most sites had a readily available telehealth platform which they were using for clinical visits, which they planned to use for the virtual group visits.
## Patient recruitment and enrollment
Upon consulting with experts in telehealth, our MWCN partners, and FQHC staff, it was decided that sites would enroll up to 12 patients for the virtual GVs, instead of up to 15 patients as we had done for in-person sessions, to facilitate virtual group discussion. Having a 12 patient limit was recommended by a licensed psychologist to promote social support in the virtual space and to accommodate for a shorter GV time of 1.5 h. Recruitment materials such as phone scripts and invitation letters were revised to inform patients that the GVs would be in a virtual format. As patients were being recruited, FQHC staff included additional questions such as what devices the patients would be joining from, if they had headphones, etc. to best help them set up for the virtual GVs. We recommended FQHC staff provide an orientation session with patients individually or as a group before the first GV session to introduce them to the video visit platform and to review the consent form and baseline survey. Consent forms, confidentiality agreements, and surveys were revised and converted to online formats. The consent forms were reviewed via phone or video with patients. Patients were given the options to complete forms in-person, over the phone, or returned via email or mail.
## Virtual group visit intervention
The FQHCs were asked to conduct 6 monthly 1–1.5 h long virtual GVs with up to 12 patients with uncontrolled T2DM (A1C ≥ $8\%$). Each visit was led by trained FQHC staff on a video conferencing platform. Additional guest speakers from various health professions provided group education at virtual GVs. Patients participated in facilitator-led group discussions that enabled material review and peer support. Patients were recommended to make a medical visit with a trained primary care provider within 2 weeks of each virtual GV.
To document the basic purposes that motivated the GV intervention, a Core Functions and Forms matrix [21] was used (Table 1). The motivating needs included access to comprehensive diabetes care, patient education, social support, and self-management. The core function column elaborates on the intended structural and procedural goal for each system need. Moreover, in the forms column we list the specific action items necessary to deliver each core function. The motivating needs, core functions and forms were all deduced by DN and DM and reviewed by AB. We also engaged in monthly webinars and conversations with FQHC staff to inform this adaptation framework. These core functions and forms were considered in the development of the virtual intervention.
**Table 1**
| Motivating need | Core function | Forms |
| --- | --- | --- |
| 1. Access to comprehensive diabetes medical care | 1. Implement use of diabetes group visits and individual medical assessment in health center setting | • In-person or virtual learning sessions to train health center staff on implementation of group visit intervention |
| Need for improvement in quality of diabetes care via effective interventions | | |
| | | • Adapt to video conference call using a HIPAA-compliant telehealth platform for alternative access as necessary |
| 2. Patient education | 1. Improve patient knowledge about diabetes, nutrition, exercise, medication, and self-management | • Group education led by trained diabetes educator at appropriate health literacy levels |
| Limited understanding around diabetes disease process and care | 2. Use of text messaging in diabetes care for diabetes education and self-management | • Use of CareMessage, a 25-week texting program that educates patients on diabetes, nutrition, exercise, stress management and medication |
| 3. Social support | 1. Create a space where patients with diabetes can connect and support each other in their care process | • Facilitate group conversations around diabetes care and coping skills |
| Need for support in the disease management | | • Allow patients to have a family member or support person attend the group visit sessions with them if they choose |
| 4. Self-management | 1. Empower patients to take control of their diabetes, improve self-management, and make healthy lifestyle changes | • Identify needs and goals to help measure personal health progress |
| Facilitate care and goal setting | | • Set aside individual goal setting sessions as needed |
## Session logs
Following each monthly virtual GV, FQHC teams completed session logs to record data about attendance, visit format, topics covered during visit, length of visit, presence of support people, patient location during visit, and additional education materials, services, or incentives provided to patients. Session logs also allowed for teams to reflect on what did or did not work well during the session. We used session logs to understand virtual group visit content and the ways in which the intervention was implemented at each FQHC.
## Staff surveys
FQHC staff completed an enrollment team survey and a pre-training survey prior to the training session in Chicago measuring their attitudes about and confidence in implementing the GV model. As previously stated, the initial training session was in-person and following the onset of the COVID-19 pandemic, staff completed training for virtual GVs. All surveys after the initial training represent staff views on virtual GVs. They completed a post-survey after 6 months of virtual GVs evaluating the perceived impact of GVs on patients, clinicians, and the FQHC. Staff rated their agreement with survey items on a five-point Likert scale of “Strongly disagree,” “Disagree,” “Neither disagree or agree,” “Agree,” and “Strongly agree.”
## Staff interviews
Post-intervention, trained research team members conducted 20–45-min telephone interviews with FQHC staff from June to September 2021. The interview questions were based on an interview guide designed to assess staff characteristics and involvement; barriers and facilitators to implementing and maintaining a virtual diabetes GVs intervention; characteristics of the virtual GV intervention as implemented and adapted to each site; desire and ability to sustain the GV intervention; and evaluation of the training. Interviews were audio recorded then transcribed by a professional transcription company for analysis.
## Study documentation
Process data for the present study was retrieved from institutional review board (IRB) documents, progress reports, and training recordings. AURA IRB is an electronic research administrative system which facilitates research administration activities. To assess adaptations needed for research implementation of virtual GVs, we analyzed AURA IRB protocol amendments and any accompanying materials (e.g., surveys, confidentiality forms, consent forms, and planning worksheets). The IRB documents, surveys, training materials, and enrollment forms were updated by the co-authors and principal investigator to reflect necessary changes for virtual diabetes GV sessions. Study progress reports provided updates on project progress and project management for research funders. We also reviewed recorded training and technical assistance webinars to assess what additions the research team made to staff training for virtual diabetes GV implementation. To assess strategies FQHC staff incorporated to engage patients, we reviewed webinars and session logs where FQHCs reflected on their experiences with GV sessions. We also reviewed yearly continuing applications, where these experiences were summarized by co-author ES, and staff interviews where FQHC staff elaborated further on some of these experiences. We then compared the activities from the sample curriculum provided to the engagement strategies FQHCs shared to see what adaptations they made for virtual settings.
## Statistical analysis
Descriptive statistics were assessed for survey data and linear mixed effect models were used to evaluate changes in attitudes before and after GV implementation.
## Qualitative analysis of staff interviews
Four investigators used a modified template approach to text analysis using the interview guide to create an initial codebook [22]. The transcripts were assigned to coder pairs using all possible combinations. Each member coded the assigned transcript independently then met with their partner to discuss to agreement. Further coding was done to identify subthemes and expand the codebook accordingly. NVivo 12 was used to code and organize the interview data.
## FQHC and staff characteristics
From the initial cluster randomized trial, 8 FQHCs with 9 clinic sites were assigned to the waitlist control group. One FQHC withdrew because they could not obtain institutional approval, another because of staff changes, and a third due to time and resource concerns. In the end, 5 FQHCs with 6 clinic sites were enrolled for implementation of virtual GVs. The 5 FQHCs were from Missouri, Illinois, Wisconsin, Indiana, and Iowa. Table 2 highlights FQHC characteristics including information about patient population and staff experience. Two FQHCs were urban, two suburban, and one rural. All FQHCs had previously held GVs for at least one health condition and at the time of enrollment, $83\%$ ($$n = 5$$/6) of FQHCs were having GVs for diabetes, heart disease, prenatal, or other conditions.
**Table 2**
| FQHC# | Setting | Number of patients | Public Insurance | Self-pay | Private Insurance | Number of staff on GV team | Staff with prior GV experience | Years caring for patients with T2DM mean (range) | Years working at FQHC site mean (range) | Staff that attended training |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 1 | Urban | 5472 | 60% | 17% | 14% | 4 | 2 | 25.3 (10–45) | 6.3 (5–7) | 3 |
| 2 | Suburban | 1704 | 89% | 5% | 6% | 5 | 0 | 13.5 (12–15) | 1.3 (0.6–2) | 5 |
| 3 | Rural | 11605 | 56% | 34% | 10% | 6 | 0 | 0.3 (0–1) | 1.2 (0.2–1) | 5 |
| 4 | Urban | 21264 | 63% | 13% | 21% | 11 | 1 | 8.2 (3–19) | 3.2 (0.5–7) | 10 |
| 5 | Urban | 17389 | 63% | 14% | 22% | 9 | 3 | 11.4 (2–20) | 7.4 (0–19) | 7 |
There were 35 FQHC staff members enrolled throughout the 6 clinic sites. Twenty-two staff members attended the in-person training session in March 2020 and 30 staff attended at least one training and/or technical assistance webinar. All 5 FQHCs were represented by at least one staff member at all training sessions. Thirty-one ($89\%$) completed the pre-training survey in February 2020. The mean age was 42.0 (SD = 11.1), $90\%$ female, $61\%$ non-Hispanic white, $16\%$ African American, $16\%$ more than one race, $3\%$ Hispanic/Latino, and $3\%$ Pacific Islander. The mean number of years in practice was 11.5 (SD = 9.0) and years providing diabetes care was 11.1 (SD = 10.8). One-third ($$n = 6$$/18) of staff had previous experience with GVs.
## Adaptations and implementation of virtual GVs
Table 1 denotes the adaptation model used for virtual GVs. Access to comprehensive diabetes medical care, patient education, social support and goal setting served as motivating factors for the interventions. Table 3 describes the adaptations made for the implementation of virtual group visits. There were adjustments to staff training, GV location, GV session time allotted, group size, patient recruitment and enrollment materials, survey administration, clinical measures, individual medical assessment, and education and interactive learning activities. All sites implemented virtual GVs. FQHC 2 held GVs from October 2020 to March 2021, FQHCs 4 and 5 from November 2022 to April 2021, FQHC 3 from December 2022 to May 2021, and FQHC 1 from March 2021 to August 2021. A total of 29 GVs were completed, and the average session duration was 82.1 (SD = 22.8) min. Seventeen of 35 ($49\%$) staff members completed a post-GV survey 1 month after the 6th GV at their HC (from April to June 2021 and September 2021). Staff reported spending on average 4.9 h each week planning and implementing group visits (SD = 5.9). Majority, $65\%$ ($$n = 11$$/17) of staff members were interested in continuing virtual GVs, and all were interested in participating in in-person groups. Staff attitudes toward GVs were compared from pre-training, when FQHCs were expecting to implement in-person GVs, to post-implementation of virtual GVs. Staff had improved awareness of barriers to GVs [$\frac{3.8}{5}$ (SD = 0.8) to $\frac{4.3}{5}$ (SD = 0.5), $$p \leq 0.03$$] but were less confident in their FQHCs ability to sustain GVs [$\frac{4.2}{5}$ (SD = 0.6) to $\frac{3.7}{5}$ (SD = 0.6), $$p \leq 0.01$$]. There was no significant change in staff's perception of the team's preparedness, motivation, or knowledge to implement or continue GVs. Measure of self-efficacy or awareness of what is needed to successfully implement GVs improved [$\frac{3.3}{5}$ (SD = 1.1) to $\frac{4.2}{5}$ (SD = 0.5), $$p \leq 0.003$$].
**Table 3**
| Unnamed: 0 | In-person | Virtual adaptations |
| --- | --- | --- |
| Staff training | In-person learning session with UChicago research staff in Chicago | · Learning sessions held via webinar |
| | | · Additional training on virtual group visits (GV): |
| | | o Explain benefits to virtual GVs |
| | | o Share literature review of previous studies on virtual GVs |
| | | o Host guest speakers to discuss facilitating virtual GVs |
| | | o Consider mock virtual GV sessions |
| Location | · Private conference room, private clinic room, or other space available at the site | · Video conference call using a HIPAA-compliant telehealth platform |
| Time allocation | · Suggested time between 1.5 and 2 h | · Suggested time between 1 and 1.5 h to avoid teleconference fatigue |
| Patient recruitment | · Enroll up to 15 patients per group | · Enroll up to 12 patients per group |
| | | · Revise recruitment phone scripts and letter invitations to reflect the virtual format of the intervention |
| | | o Participating sites request patient email address to send REDCap forms |
| | | o Assess patient capacity for virtual sessions (ask what device they will be joining from, if they have headphones, etc. to help them set up) |
| Confidentiality | ·Patients sign confidentiality form at the first GV session | · Patients sign confidentiality form via REDCap and participating sites collect emails of any accompanying support person participating in GV sessions for the online REDCap form (emails are not accessible by study team) |
| Consent forms | ·Staff/providers: Review the consent form and obtain written informed consent at the first learning session | · Staff/providers: Review the consent form via webinar then ask participants to print and sign the consent forms and return to the study team via mail. |
| | · Patients: Review consent form with patients before the first group visit session and obtain written consent from each intervention patient | · Patients: Contact patient (phone or video) to review consent, then email a personalized link to complete form via REDCap, or email, mail or pick up a copy of the consent form. The patient can return the signed consent form in person, by mail, or they can scan or take a photo of the signed consent form and email it to participating staff. |
| Surveys | · Staff surveys administered in-person after learning sessions | · Staff surveys administered online via REDCap |
| | · Patient surveys administered in-person prior to beginning the first group visit and after completing the sixth GV | |
| | | · Patient surveys administered via email invitations to online REDCap surveys, verbally over the phone or via video call, mailed or emailed survey pdf version, or physical copy received and returned to participating sites by mail, scanned, or in-person. |
| | | · Revise surveys to include virtual aspect and identify virtual-specific barriers and/or benefits to GVs |
| Clinical Measures | · Point of care testing | · In primary care visit |
| | · Patients check into the clinic for their GV appointment and have their vitals checked | · Drive up services for lab draws |
| | · Lab work if available at site | |
| Individual Medical Assessment | · Privately during group visits | · Recommended within 2 weeks before or after the group portion via phone, video, or in clinic as determined by each participating site |
| Education | · In-person activities such as cooking and physical activity demonstrations | · Activities adapted to virtual platforms |
| | | · Use of innovative virtual games |
From the 5 FQHCs recruited a total of 251 patients were spoken to about the study and 91 agreed to participate. Out of 160 patients who did not agree to participate, 85 were unable to participate mostly due to other scheduled responsibilities and 7 due to having no access to internet or devices; 50 were not interested because they did not think they needed more diabetes education or they were already going to other diabetic groups or specialists; 11 for unknown reasons; 5 lost to follow-up; and 9 were ineligible due to not having a cell phone/texting, hemoglobin A1Cs below 8 or no diabetes, and for being out of town. Of the 91 that agreed to participate, 42 were not enrolled mostly due to loss to follow-up, for being unable to participate, or were ineligible. In the end, a total of 49 patients were enrolled in the study. One patient was withdrawn prior to the first GV and is not included in the analyses.
Sites enrolled a total of 48 adults with T2DM for the virtual GVs, with baseline hemoglobin A1C 9.8 ± $1.8\%$, mean age 55 ± 12, $67\%$ female, $67\%$ African American, $27\%$ non-Hispanic white, and $6.2\%$ Hispanic. Table 4 encompasses information about GV eligibility, enrollment, and attendance by FQHC site. All FQHCs implemented GVs. Attendance ranged from 0 to 9 patients at GV sessions, and an average of 4 (3.8) patients attended each session across all FQHCs. Each patient attended a mean of 2.1 ± 2.2 GV sessions across sites. Thirty-four patients ($71\%$) completed one or more virtual GVs and 14 patients attended no virtual GVs. Of the 34 patients that attended, 20 ($59\%$) attended with video from home, 4 ($11\%$) with phone only from home, 3 ($9\%$) with video from clinic room, 3 ($9\%$) with video from home and other/unknown location, 2 ($6\%$) with video from home and clinic room, and 2 ($6\%$) with video and phone only from home. For patient surveys at baseline 38 were completed and at 6 months 22 were completed for a total of 60. Of the 60 patient surveys, 42 ($70\%$) were completed or returned in person, 6 ($10\%$) by phone, 1 ($2\%$) by mail, and 11 ($18\%$) were unspecified. Those that were unspecified were reported as either majority being paper copies or mostly over the phone.
**Table 4**
| Unnamed: 0 | Eligible | Enrolled | GV1 N | GV2 N | GV3 N | GV4 N | GV5 N | GV6 N | Average GV duration (min) | Time staff spent planning and implementing GV (hours/week) |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Site 1 | 53 | 12 | 4 | 3 | 1 | 3 | 0 | 2 | 118.0 | M = 5.5, SD = 3.54 |
| Site 2 | 35 | 6 | 5 | 5 | 3 | 2 | 4 | 3 | 90.0 | M = 1.17, SD = 0.76 |
| Site 3 | 19 | 5 | 3 | 3 | 2 | 3 | 3 | 3 | 75.8 | * |
| Site 4a | 137 | 7 | 4 | 2 | 1 | 0 | 0 | 0 | 60.0 | M = 13.0, SD = 9.89** |
| Site 4b | 278 | 6 | 1 | 2 | 2 | 0 | 0 | 0 | 60.0 | |
| Site 5 | 308 | 12 | 7 | 9 | 7 | 6 | 5 | 6 | 72.5 | M = 3.6, SD = 4.72 |
## Barriers to implementing virtual GVs
The COVID-19 pandemic presented barriers to virtual GV implementation. As a result of COVID-19, FQHC staff had modified work environments (e.g., spacing, remote work), additional clinical tasks (e.g., administering vaccines) and less availability. FQHC 1 delayed GV implementation by about 4 months due to substantial staff turnover. In the post-GV staff survey, most staff cited other COVID-19 related priorities at the FQHCs as the biggest barriers for implementation. Additionally, during webinar check-ins, FQHC staff reported some patients felt restricted and isolated because of the pandemic.
Other patient-related barriers to implementation included recruitment and retention, patient attendance, internet and device access, and technology navigation. Common reasons for patients not participating were mostly due to scheduling conflicts or not being interested. Even after enrollment, some patients did not attend GVs (Table 5). Some patients did not have access to internet or devices. FQHC location in a rural area was an additional challenge for internet access and connectivity. Some patients also had difficulty navigating and logging into the video conference platform (e.g., patients continuously forgetting login credentials).
**Table 5**
| Theme | Subtheme | Selected staff quotes |
| --- | --- | --- |
| Challenges of virtual group visits | a. Technology | “the biggest challenge would probably be just the access of technology to our patients … making sure that everybody who was wanting to be involved had internet access, that they had something to join with, whether it was a phone or a tablet, computer.” |
| | b. Patient attendance | “the challenges of the virtual were getting the participants in to actually educate them on how to do the login … me not being a technical person myself. I had to get that education as well.” |
| | c. Adaptation to virtual contact | “Eventually we started to get patients that would confirm and not show up during the meeting. It went from tons of participation, everyone being excited to people were just too busy to come or they would again, register and all the information would be verified but then at the time of the Zoom meeting, nobody comes in.” |
| | | “interpersonal benefit that's lost a little bit with that. But we were able to get a group that had a pretty good rapport and was quite engaged throughout. So that, was good. |
| | | “cadence and the timing and how to make sure that no one person was monopolizing the conversation or speaking too much and how to make sure that one of the patients who was on her cell phone quite a bit, how to manage that. So it wasn't distracting to anyone else, how to get the quiet ones to speak. “ |
| Health center recommendations for recruitment and retention | a. Provider recommendation | “our letters came from the physicians and then during their provider visits they were referring direct referrals over, and that was very helpful. Patients have a good relationship with their provider and there's a lot of trust there…the patients took that pretty seriously. So we thought that really helped with recruitment.” |
| | b. Honest description of expectations | “definitely explaining the program thoroughly when you were recruiting people [...] letting them also know that it's optional. Because I think sometimes people feel pressure to be in it, and they don't necessarily have the time. So definitely, starting off from the base to make sure that you have people that know what they're in, what they're expected of for the group, and then what's expected of us too. And then, just making sure that they can make that time commitment.” |
| | c. Provide incentives | “I do think incentives helped the retention of keeping those people that started. I think it helped keep them coming back each month to see, kind of what they'd learn and then what they might receive in the mail for participating.” |
| | d. Building rapport | “For retention, we had a lot of hands-on, we do each month kind of connecting with the persons each month to make sure that they had what they needed. So, I think that's kind of essential for keeping people going, even if it's even just once a month.” |
| | | “I'd already built up a rapport with some of the patients that I had called and reached out to. So they kind of knew me already” |
| Benefits of virtual group visits | a. Health center | “I think for our organization, that's a benefit because we know tele-health has a benefit to our patients and if it's going to be an effective program in our system, then we need to, as the system, we need to be comfortable with it and sell it as a positive thing to our patients too.” |
| | b. Patients | “they would have the opportunity to speak with others that were going through some of the same things that they were going through … to be able to share how they overcame or how they were working through or dealing with some of their issues with diabetes” |
| | | “it was nice to see was the support and the morale with the group. We had patients that were sharing their contact information with each other and were showing all the different ways that they have” |
| | c. Staff | “they just came together so well as a team and support each other and shared information and work together to provide good information to the patients. It was just wonderful to watch. I was just so excited and so happy about it.” |
Staff also experienced difficulties adjusting to technology, allotted time, and to virtual contact. As noted in Table 5, staff needed additional technical support. During interviews, staff mentioned adjusting to the virtual format during cooking and physical activity demonstrations was more challenging because of camera and sound manipulation (Table 6). From webinars, FQHC staff reported that it was difficult getting patients engaged with the time allotted and amount of material to cover.
**Table 6**
| Topic | Virtual activities examples |
| --- | --- |
| Physical Activity | Cardio drum session |
| | Purpose: demonstrate accessible physical activity (chair and low impact) to motivate patients to think outside the box for exercise |
| | Adaptations: use breakout rooms on telehealth platforms to have patients work in smaller groups or in partners after exercise demonstrations to encourage them to attempt the routines |
| | Challenges: patients and staff need their own exercise equipment; need to adjust camera positioning |
| | Suggestions: provide the necessary exercise equipment (exercise balls, sticks, and buckets) and deliver them well ahead of time; have one or several staff members use a handheld device while streaming on the telehealth platform to show different angles |
| Nutrition | Recipe presentation |
| | Purpose: learn about nutritional value of foods to encourage healthier food choices |
| | Adaptations: may supplement or replace a traditional cooking demonstration; have a volunteer prepare a recipe from the American Diabetes Associated Food Hut website or another reliable source and present its nutritional value (e.g., carbs, serving size, calories, taste, etc.); offer a grocery store gift card as an incentive; have patients recreate recipe at home so they can taste it as well |
| | Challenges: not everyone may have necessary ingredients available; allergies and dietary restrictions |
| | Suggestions: find a recipe with common ingredients and provide a list of substitutes well in advance; if within budget, deliver ingredients to patients; have a nutritionist or a registered dietician guest speaker present; plan for a mix of cultural foods |
| Education | Emoji game |
| | Purpose: identify and brainstorm how to treat symptoms of hyperglycemia and hypoglycemia |
| | Adaptations: designate an emoji for each symptom (ex. water drops for extreme thirst) then show each emoji on the screen and ask patients how they would treat the symptom the emoji represents |
| | Challenges: emojis differ across devices |
| | Suggestions: instead of sending the emoji through chat, share images of emojis on the screen so everyone sees the same emoji; use basic emojis available in most devices |
| Incentives | Healthy gift basket, grocery store gift card, cookbook, tablets, coloring books, diabetic socks, self-care kits, kitchen supplies, portion plates |
| | Purpose: to help maintain patients engaged and make them feel supported |
| | Adaptations: incentives may be delivered via mail for physical items or email for gift cards or other e-resources accessible through links, staff may also coordinate a time for patients to pick up from clinic |
| | Challenges: health centers may not have the funds to sponsor incentives |
| | Suggestions: pitch idea to stakeholders; apply for grants; find free resources for patients like activities to de-stress; motivational songs; hint at incentives when sending invitations; coordinate incentives to match learning topic |
Additional barriers to virtual GV implementation included reimbursement and incorporating the provider visit into sessions. FQHCs expressed they were not billing for the diabetes education portion of the virtual GVs. During webinars, staff expressed interest in learning more about billing, referrals, and insurance coverage. Other barriers included the provider not being present during all sessions and patient confusion about the team's provider role. Some FQHCs also reported experiencing difficulty incorporating provider visits with the GV session.
## Facilitators to implementing virtual GVs
The FQHCs developed various strategies for overcoming patient barriers to participation. Virtual GV facilitators included inviting patients who did not have devices or internet access at home to go to the FQHC and join virtually from individual clinic rooms. Access to Wi-Fi or internet connection was provided in $38\%$ ($$n = 11$$/29) of virtual GVs. Some FQHCs also provided transportation for those patients who needed to go to the clinic site for internet access in $21\%$ ($$n = 6$$/29) of sessions. Other facilitators included providing devices for patients (e.g., tablets, hotspots); allowing patients to call in without video if necessary; and mailing copies of materials ahead of time or having patients pick them up from the clinic. In $48\%$ ($$n = 14$$/29) of the virtual GVs, patients were provided a tablet or device to participate in session. Patients also received incentives (e.g., gift cards, gift baskets, fresh produce delivery) and educational materials for 54 ($$n = 15$$/29) and $86\%$ ($$n = 25$$/29) of sessions, respectively.
Some FQHCs provided a pre-session for technical support and training for both patients and staff prior to the first group visit; make-up sessions; 10-min breakout room sessions to get to know providers; and a 30-min “open house” before official GV start time to revisit guidelines, play games to review previous lessons and provide additional technological assistance. As noted in Table 6, the FQHCs thought of many creative ways to keep virtual sessions engaging and interactive, such as playing a game using emojis to identify symptoms of hyperglycemia and hypoglycemia and leading accessible physical activities like chair cardio drumming.
During post-intervention staff interviews, staff suggested recommendations that could improve recruitment and retention (Table 5). They suggested having providers recommend the program to patients, giving detailed descriptions of the virtual GV intervention, providing incentives, and building rapport with patients for better outcomes.
## Staff perceived benefits of virtual GVs
Figure 1 highlights staff perceptions of virtual GV benefits at the patient, staff, and FQHC level based on staff surveys.
**Figure 1:** *Staff perceived virtual GVs for patients, staff, and health center. Staff perceived virtual group visit benefits across three categories: patient, staff, and health center.*
## Patient benefits
In terms of benefits of virtual GVs for the patient, all staff agreed that they empowered patients to manage their diabetes and provided patients with social support, connection, and more frequent contact with medical providers. Staff were least confident in the ability of virtual GVs to improve clinical outcomes and lower cost of care for patients with only 65 ($$n = 11$$/17) and $53\%$ ($$n = 9$$/17) respondents agreeing that they do so respectively.
## Staff benefits
Staff largely agreed with all proposed benefits to providers and staff. These included improved communication, trust, and understanding with patients, increased opportunity for teamwork, collaboration, and creativity, and more variety in their work. The least agreed upon statement was that virtual GVs allowed providers and staff to get to know each other with $71\%$ ($$n = 12$$/17) agreeing.
## FQHC benefits
There was greater variety in perceived benefits to the FQHC. Most staff agreed that virtual GVs lead to better patient engagement and care coordination as well as higher patient satisfaction. However, staff were less confident that virtual GVs increased provider productivity or led to higher reimbursements with only 29 ($$n = 5$$/17) and $18\%$ ($$n = 3$$/17) staff members agreeing respectively.
## Discussion
Given the unpredictability of the COVID-19 pandemic, we modified the approach from in-person diabetes group visits to a virtual format across Midwestern FQHCs. Virtual GVs were implemented in all FQHC sites and staff found them beneficial. While the intervention's inclusion criteria and core components remained the same, additional consideration was needed for staff training, group size, recruitment and enrollment forms, and survey administration. Main challenges included technological barriers for both patients and staff, and patient recruitment and retention. Facilitators for virtual GVs included providing patients with tablets, orienting patients to the virtual platform, and incorporating creative activities for patient engagement. Successful outcomes included representation of all 5 FQHCs at training sessions and majority of staff interest in continuing virtual GVs.
All FQHCs implemented virtual GVs and staff found the intervention beneficial for patients, staff and the health center. Other studies on virtual visits or telehealth reported staff-perceived or patient-reported benefits such as improved self-efficacy [23] and peer support [23, 24] as general GV benefits. In addition, virtual specific GV benefits included time saving [24], scheduling and location flexibility [25, 26], and ease of participation due to reduced transportation barriers [25, 26]. Our study is in agreement with these findings and adds additional perceived benefits. In our study, the most common staff-perceived benefits for patients included self-empowerment, improved quality of life, social support and connection. Staff felt virtual diabetes GVs improved trust and communication with patients, teamwork and collaboration, and better understanding with patients. While staff showed significant improvement in awareness of barriers and of what is needed to successfully implement GVs, as previously mentioned in the results, their confidence in their ability to sustain/implement GVs decreased. A possible explanation for this finding is their increased knowledge and awareness of challenges and barriers in virtual GVs led them to feel less confident about their ability to sustain the intervention. Specifically, the continuous outreach from staff in contacting patients and providing additional facilitators (e.g., devices, internet access, transportation, etc.) to improve retention yet having low attendance may have discouraged some staff members. Additionally, it is important to note that FQHC staff were expecting in-person GVs at the time of enrollment. Although $65\%$ of staff were interested in continuing virtual GVs, all FQHC staff remained interested in participating in in-person GVs. A strong preference for in-person GVs and low acceptability of virtual GVs may lead to variation in sustainability confidence [27, 28].
Nonetheless, majority of staff agreed that virtual GVs benefited the FQHC's improvements in care coordination and offered an opportunity to implement an alternative model of care. Other studies report less staffing and overhead costs as additional network benefits [26]. However, only a few staff in our study agreed with higher reimbursement/revenue as a perceived benefit for the FQHC. This may be because FQHCs billed for individual provider visits alone, but did not account for diabetes education. Overall, implementation and reported benefits of our intervention and that of other studies suggest virtual GVs are feasible and beneficial for patients, staff, and FQHCs across different health conditions. Our adaptation model is not limited to diabetes and may be of use to other health education programs interested in implementing virtual GVs.
Programmatic changes had to be made to adapt in-person diabetes GVs to a virtual format. First, staff training was modified to include education on virtual program implementation, barriers, and facilitators. Second, group size was modified to facilitate group interaction in a virtual setting and reduce risk of “Zoom fatigue” [29]. Moreover, staff supported patient participation in virtual group visits to ensure evenly distributed conversation and engagement across patients. Third, enrollment forms and survey administration were made more accessible by providing various options for completion and return (e.g., by mail, email, over the phone, etc.). Of those that responded, majority returned the surveys in person or completed over the phone. No participants completed surveys electronically. As noted in staff interview results, this may be because existing rapport and repeated contact between FQHC and patients may encourage more engaged research participation. Future programs implementing virtual GVs may offer options by mail, email, over the phone, etc. to optimize patient response and later assess which format works best for them.
While adaptation to virtual GVs was accomplished, it was not without challenges. Other studies on transitioning to telehealth reported internet connectivity [25], access to technology [30], and participant login issues [25] as challenges. Interview and survey results from this study found similar challenges including technology access, technical concerns, and adaptation to virtual contact. Although internet and technology access remain an issue especially among minorities [31], virtual GV implementation sites may reduce these barriers by providing devices, Wi-Fi, and pre-sessions for technical support as the FQHCs in our study did. It is important to mention that the implementation of this study occurred earlier in the pandemic when not all FQHCs had telehealth platforms set up. This may explain why FQHC staff reported some difficulty getting accustomed to interacting with patients virtually. Considering telehealth services are now more widespread [32], situating patients and staff to telehealth may present a lesser challenge thereby making implementation more feasible. Nevertheless, FQHC staff were able to build rapport and maintain patients engaged despite these barriers.
When orienting patients with technology for virtual GVs, staff need to be comfortable navigating it as well. Other studies reported retraining staff and patients [30] and limited staff experience with software [25] as additional technology related challenges. As previously noted in barriers to implementation, some FQHC staff did not feel confident and needed additional technical support. While there was additional training on virtual group facilitation, telehealth services, and REDCap usage, there was no specific training on a given virtual platform (i.e., Zoom or Microsoft Teams). Instead, each FQHC used their own preferred virtual platform. This was done purposefully so FQHCs could use what was already available to them to facilitate rapid virtual GV implementation. With the rise of virtual care, telehealth is now more centralized with additional training and technology implemented to accommodate the shift [32]. Even so, staff experience levels with technology should be assessed to provide additional technology support as necessary.
Another challenge FQHC staff faced was patient recruitment and retention. Challenges to patient recruitment and retention are seen across various lifestyle modification programs (33–35). In our study, additional challenges included the COVID-19 pandemic and the rapid transition to a virtual format. Even though poor patient recruitment and retention is common, building rapport and trust with patients, getting providers to recommend virtual GVs, providing incentives, and describing challenges and benefits of virtual GVs as FQHC staff did may help.
## Limitations
The present study has limitations that are important to consider in future application of this research. Given the rapid onset of the COVID-19 pandemic, the FQHCs in this study were asked to transition from in-person use of diabetes GVs to virtual ones. Clinical demands were higher with COVID-19 related services, therefore limited staff time to implement virtual GVs. Moreover, this rapid transition led FQHCs to implement a video platform that was familiar to them but was not consistent across sites. We recruited FQHCs from the Midwest Clinicians Network clinics, which while diverse, may not be generalizable across other regions and clinic networks.
## Future directions
Future programs seeking to implement virtual GVs should take into account various factors. FQHCs may need to budget for or apply for grants to fund any technological, software, or hardware support. Moreover, implementation timelines should incorporate time to address technological challenges and support for patients. Additionally, future programs may consider using a standardized virtual platform, ideally one that is familiar and with features that facilitate group discussion such as breakout rooms, screen sharing, chat boxes, and raise hand option. It is also important for staff to consider creative activities and modifications to timing and group size to lower risk of virtual fatigue. Holding a mock GV session or conducting all staff training on said platform may help orient staff to the virtual platform and address any challenges that may arise. Future programs may also consider providing staff with additional information on insurance coverage and billing and reimbursement for virtual GVs.
## Conclusion
In summary, FQHCs adapted diabetes GVs from in-person to a virtual platform during the COVID-19 pandemic. Modifications included changes in patient recruitment and enrollment, staff training, and learning to facilitate virtual sessions in a creative way to keep patients engaged. Challenges to implementation of virtual GVs included limited access to technologic support and lower staff availability due to pandemic demands. Facilitators of virtual GVs included providing technical assistance to patients, such as access to tablet devices, internet access from the clinic, technical support prior to GVs, and incorporating creative activities to engage patients in a virtual setting. Overall, FQHC staff reported overall satisfaction and support of future implementation of virtual GVs. Future studies should consider staff and patient support with technology and training modifications to facilitate the implementation of virtual diabetes GVs. Moreover, additional research should consider the ways to improve provider interaction with patients during GVs and include a control arm to assess the impact of virtual group visits on clinical outcomes.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by the University of Chicago Biological Sciences Division Institutional Review Board. The patients/participants provided their written informed consent to participate in this study.
## Author contributions
Study concept and design by AB, ES, DN, DM-N, MZ, WW, MQ, AC, and CS. Data acquisition was performed by DN and DM-N. Data analysis and interpretation were performed by AB, ES, DN, DM-N, MZ, WW, and TD. DN and DM-N wrote the initial manuscript draft. Critical revision of the manuscript for intellectual content was performed by all authors.
## Funding
This research was supported by the National Institute of Diabetes and Digestive and Kidney Diseases Chicago Center for Diabetes Translation Research (P30 DK092949) and the U.S. Department of Health and Human Services Office of Minority Health (1 CPIMP171145-01-00). Additional funding was received from the Dean's Office of the Biological Sciences Division of the University of Chicago. Study data were collected using REDCap, hosted by the University of Chicago Center for Research Informatics (NIH CTSA UL1 TR000430). AB was supported by a NIDDK Mentored Patient-Oriented Career Development Award (K23 DK087903-01A1).
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Author disclaimer
The contents of this manuscript are solely the responsibility of the authors and do not necessarily represent the official views of OMH.
## References
1. Atlanta GA. *Centers for Disease Control and Prevention, US Department of Health and Human Services. Control CfD, Prevention* (2020) 12-15
2. Iglay K, Hannachi H, Joseph Howie P, Xu J, Li X, Engel SS. **Prevalence and co-prevalence of comorbidities among patients with type 2 diabetes mellitus**. *Curr Med Res Opin.* (2016) **32** 1243-52. DOI: 10.1185/03007995.2016.1168291
3. Li C, Ford ES, Strine TW, Mokdad AH. **Prevalence of depression among U**. *Diabetes Care.* (2008) **31** 105-7. DOI: 10.2337/dc07-1154
4. Chiou T, Tsugawa Y, Goldman D, Myerson R, Kahn M, Romley JA. **Trends in racial and ethnic disparities in diabetes-related complications, 1997–2017**. *J Gen Internal Med.* (2020) **35** 950-1. DOI: 10.1007/s11606-019-05308-9
5. Karter AJ FA, Liu JY, Moffet HH, Ackerson LM, Selby JV. **Ethnic disparities in diabetic complications in an insured population**. *JAMA* (2002) **287** 2519-27. DOI: 10.1001/jama.287.19.2519
6. Traven SA, Synovec JD, Walton ZJ, Leddy LR, Suleiman LI, Gross CE. **Notable racial and ethnic disparities persist in lower extremity amputations for critical limb ischemia and infection**. *JAAOS J Am Acad Orthop Surg.* (2020) **28** 885-8922. DOI: 10.5435/JAAOS-D-19-00630
7. Bitar H, Alismail S. **The role of eHealth, telehealth, and telemedicine for chronic disease patients during COVID-19 pandemic: a rapid systematic review**. *Dig Health.* (2021) **7** 20552076211009396. DOI: 10.1177/20552076211009396
8. Shi L, Lebrun LA, Tsai J, Zhu J. **Characteristics of ambulatory care patients and services: a comparison of community health centers and physicians' offices**. *J Health Care Poor Underserved.* (2010) **21** 1169-83. PMID: 21099069
9. Forrest CB, Whelan E-M. **Primary care safety-net delivery sites in the United States: a comparison of community health centers, hospital outpatient departments, and physicians' offices**. *JAMA.* (2000) **284** 2077-83. DOI: 10.1001/jama.284.16.2077
10. Chin MH, Auerbach SB, Cook S, Harrison JF, Koppert J, Jin L. **Quality of diabetes care in community health centers**. *Am J Public Health.* (2000) **90** 431-4. DOI: 10.2105/AJPH.90.3.431
11. Burke RE, O'Grady ET. **Group visits hold great potential for improving diabetes care and outcomes, but best practices must be developed**. *Health Aff Millwood.* (2012) **31** 103-9. DOI: 10.1377/hlthaff.2011.0913
12. Housden L, Wong ST, Dawes M. **Effectiveness of group medical visits for improving diabetes care: a systematic review and meta-analysis**. *CMAJ Can Med Assoc J Journal de l'Association Medicale Canadienne.* (2013) **185** E635-44. DOI: 10.1503/cmaj.130053
13. Baig AA, Staab EM, Benitez A, Hermans SP, Ham SA, Wan W. **Impact of diabetes group visits on patient clinical and self-reported outcomes in community health centers**. *BMC Endocr Disord.* (2022) **22** 60. DOI: 10.1186/s12902-022-00972-1
14. Mannion SE, Staab EM, Li J, Benitez A, Wan W, Campbell A. **Patient and clinician satisfaction with diabetes group visits in community health centers**. *J Patient Exp.* (2021) **8** 23743735211056467. DOI: 10.1177/23743735211056467
15. Edelman D, Gierisch JM, McDuffie JR, Oddone E, Williams JW. **Shared medical appointments for patients with diabetes mellitus: a systematic review**. *J Gen Intern Med.* (2015) **30** 99-106. DOI: 10.1007/s11606-014-2978-7
16. Gajarawala SN, Pelkowski JN. **Telehealth Benefits and Barriers**. *J Nurse Pract JNP.* (2021) **17** 218-21. DOI: 10.1016/j.nurpra.2020.09.013
17. Scott Kruse C, Karem P, Shifflett K, Vegi L, Ravi K, Brooks M. **Evaluating barriers to adopting telemedicine worldwide: a systematic review**. *J Telemed Telecare.* (2018) **24** 4-12. DOI: 10.1177/1357633X16674087
18. Reid MW, Krishnan S, Berget C, Cain C, Thomas JF, Klingensmith GJ. **CoYoT1 clinic: home telemedicine increases young adult engagement in diabetes care**. *Diabetes Technol Therap.* (2018) **20** 370-9. DOI: 10.1089/dia.2017.0450
19. Bakhach M, Reid MW, Pyatak EA, Berget C, Cain C, Thomas JF. **Home telemedicine (CoYoT1 clinic): a novel approach to improve psychosocial outcomes in young adults with diabetes**. *Diabetes Educ.* (2019) **45** 420-30. DOI: 10.1177/0145721719858080
20. Bisno DI, Reid MW, Fogel JL, Pyatak EA, Majidi S, Raymond JK. **Virtual group appointments reduce distress and improve care management in young adults with type 1 diabetes**. *J Diabetes Sci Technol.* (2021) **30** 19322968211035768. DOI: 10.1177/19322968211035768
21. Perez Jolles M, Lengnick-Hall R, Mittman BS. **Core functions and forms of complex health interventions: a patient-centered medical home illustration**. *J Gen Internal Med.* (2019) **34** 1032-8. DOI: 10.1007/s11606-018-4818-7
22. Crabtree BF, Miller WL, BF Crabtree, WL Miller. **A template approach to text analysis: developing and using codebooks**. *Doing Qualitative Research* (1992)
23. Tokuda L, Lorenzo L, Theriault A, Taveira TH, Marquis L, Head H. **The utilization of video-conference shared medical appointments in rural diabetes care**. *Int J Med Inform.* (2016) **93** 34-31. DOI: 10.1016/j.ijmedinf.2016.05.007
24. Raymond JK, Berget CL, Driscoll KA, Ketchum K, Cain C, Thomas JF. **CoYoT1 clinic: innovative telemedicine care model for young adults with type 1 diabetes**. *Diabetes Technol Therap.* (2016) **18** 385-390. DOI: 10.1089/dia.2015.0425
25. Shah MK, Gibbs AC, Ali MK, Narayan KMV, Islam N. **Overcoming the digital divide in the post–COVID-19 “reset”: enhancing group virtual visits with community health workers**. *J Med Internet Res.* (2021) **23** e27682. DOI: 10.2196/27682
26. Lacagnina S, Tips J, Pauly K, Cara K, Karlsen M. **Lifestyle medicine shared medical appointments**. *Am J Lifestyle Med.* (2021) **15** 23-7. DOI: 10.1177/1559827620943819
27. Wade VA, Eliott JA, Hiller JE. **Clinician acceptance is the key factor for sustainable telehealth services**. *Qual Health Res.* (2014) **24** 682-94. DOI: 10.1177/1049732314528809
28. Van Citters AD, Dieni O, Scalia P, Dowd C, Sabadosa KA, Fliege JD. **Barriers and facilitators to implementing telehealth services during the COVID-19 pandemic: a qualitative analysis of interviews with cystic fibrosis care team members**. *J Cystic Fibrosis* (2021) **20** 23-28. DOI: 10.1016/j.jcf.2021.09.004
29. Bailenson JN. **Nonverbal overload: a theoretical argument for the causes of zoom fatigue**. *Technol Mind Behav.* (2021) **2** 7079-105. DOI: 10.1037/tmb0000030
30. Hills HA, Lengnick-Hall R, Johnson KA, Vermeer W, Hendricks Brown C, McGovern M. **Rapid adaptation during the COVID crisis: challenges experienced in delivering service to those with opioid use disorders**. *Implement Res Pract.* (2022) **3** 26334895221096295. DOI: 10.1177/26334895221096295
31. Jain V, Al Rifai M, Lee MT, Kalra A, Petersen LA, Vaughan EM. **Racial and geographic disparities in internet use in the US among patients with hypertension or diabetes: implications for telehealth in the era of COVID-19**. *Diabetes Care.* (2021) **44** e15-7. DOI: 10.2337/dc20-2016
32. Wosik J, Fudim M, Cameron B, Gellad ZF, Cho A, Phinney D. **Telehealth transformation: COVID-19 and the rise of virtual care**. *J Am Med Inform Assoc.* (2020) **27** 957-62. DOI: 10.1093/jamia/ocaa067
33. Gucciardi E, Demelo M, Offenheim A, Stewart DE. **Factors contributing to attrition behavior in diabetes self-management programs: a mixed method approach**. *BMC Health Serv Res.* (2008) **8** 33. DOI: 10.1186/1472-6963-8-33
34. Vincent D, McEwen MM, Hepworth JT, Stump CS. **Challenges and success of recruiting and retention for a culturally tailored diabetes prevention program for adults of Mexican descent**. *Diabetes Educ.* (2013) **39** 222-30. DOI: 10.1177/0145721713475842
35. Blumenthal DS, Sung J, Coates R, Williams J, Liff J. **Recruitment and retention of subjects for a longitudinal cancer prevention study in an inner-city black community**. *Health Serv Res.* (1995) **30** 197-205. PMID: 7721592
|
---
title: 'Understanding the Role of Clinical Champions and Their Impact on Clinician
Behavior Change: The Need for Causal Pathway Mechanisms'
authors:
- Alexandra L. Morena
- Larissa M. Gaias
- Celine Larkin
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012807
doi: 10.3389/frhs.2022.896885
license: CC BY 4.0
---
# Understanding the Role of Clinical Champions and Their Impact on Clinician Behavior Change: The Need for Causal Pathway Mechanisms
## Abstract
### Background
The clinical champion approach is a highly utilized implementation strategy used to mitigate barriers and improve outcomes of implementation efforts. Clinical champions are particularly effective at addressing provider-level barriers and promoting provider-behavior change. Yet, the specific causal pathways that explain how clinical champions impact provider behavior change have not been well-explicated. The current paper applies behavior change models to develop potential causal pathway mechanisms.
### Methods
The proposed mechanisms are informed by previous literature involving clinical champions and empirically supported behavior change models. These models are applied to link specific attributes to different stages of behavior change and barriers for providers.
### Results
Two unique pathway mechanisms were developed, one that explicates how providers develop intention to use EBPs, while the other explicates how providers transition to EBP use and sustainment. Clinical champions may promote intention development through behavioral modeling and peer buy-in. In contrast, champions promote behavioral enactment through skill building and peer mentorship.
### Conclusion
Clinical champions likely play a critical role in reducing provider implementation barriers for providers across various phases of behavior change. The proposed pathways provide potential explanations for how clinical champions promote provider behavior change. Future research should prioritize empirically testing causal pathway mechanisms.
## Introduction
Evidence-based practices (EBPs) can help reduce suffering and prevent premature death in those experiencing mental health conditions [1, 2]. Yet, EBPs are underutilized among mental health clinicians and organizations [3, 4]. Many individuals seeking mental health care do not receive empirically supported treatments [4] and there is a substantial gap between establishing empirical support and integrating an EBP into routine care [5, 6]. Implementation science research aims to bridge this gap by investigating factors that promote and impede uptake of best practices while also developing and testing strategies to promote successful implementation [4]. Implementation strategies are techniques or methods used to enhance the uptake, implementation, and sustainment of a clinical program or practice [7]. While the field has focused on developing and testing strategies, mechanisms explaining how strategies operate need to be more thoroughly explored [8]. Causal mechanisms explain the processes through which implementation strategies generate a desired effect or outcome and will allow researchers to better select implementation strategies based on their project goals [9, 10].
In the current paper, we examine causal pathways through which one such implementation strategy, the identification and preparation of clinical champions address and reduce provider-level barriers to uptake of EBPs. The “identifying and preparing clinical champions” strategy was selected due to its described impact on provider behavior change across implementation efforts [11]. While this strategy is regularly utilized in various implementation efforts, less is known about the underlying mechanisms that may be contributing to its success. To inform the development of the proposed causal pathway mechanisms, the current paper provides a theoretically informed review of the current clinical champion literature to outline relevant attributes and describe common responsibilities and processes champions engage in.
## What (AND WHO) are Clinical Champions?
Clinical champions are individuals who are dedicated to supporting, advocating for, and spearheading an implementation initiative, and who overcome resistance that may occur at the organizational level [12]. They have an intrinsic interest to implement change and use their position to motivate others (13–15). Previous research has referenced clinical champions of a specific topic (e.g., hand-washing champions), discipline (e.g., nurse champion), or broadly within an organization or implementation effort (e.g., executive champion) [15].
Research involving clinical champions have begun to identify attributes that may impact an implementation effort [14, 16, 17]. Clinical champions have been described as having strong communication and mentorship skills. Strong communication and mentorship skills involves processes including collaborating with others, advocating for change, ability to negotiate as well as educate and facilitate learning [15, 17]. Strong communication and mentorship skills can facilitate buy-in by conveying their conviction and positive perceptions about the initiative to their peers [14, 17, 18]. Champions can also effectively tailor messages to different audiences to maximize engagement and buy-in [15, 16, 18].
Previous research has also emphasized clinical champion's EBP knowledge and competency [14, 17, 19]. Clinical champions often emerge due to their knowledge and previous experience which, in part, is how clinical champions promote EBP adoption within their clinical environment [13, 17]. Additionally, as clinical champions serve as a resource for providers to develop EBP competency, effective champions must be knowledgeable, experienced, and have strong self-efficacy to effectively educate others [15, 19]. Through their knowledge and expertise, clinical champions may also engage in skill-sharing, and promote benefits of integrating the EBP into clinical practice [13, 14, 17].
Clinical champions have also been described as being deeply embedded in their clinical setting. Embeddedness in the clinical setting means that a clinical champion has frequent face-to-face to contact with their peers as well as leadership and are regularly present on the frontlines (i.e., the point of change) [16, 20]. This embeddedness results in a robust understanding of their setting's culture and workflow [16, 20]. This embeddedness may allow the clinical champion to model integration of an EBP into their daily workflow as well as providing education and support to peers [16, 17]. A clinical champion's presence in the clinical setting can have substantial impact on implementation; results from Rycroft-Malone et al. [ 18] found clinical champions' embeddedness allowed for a more grass-roots approach to implementation, which yielded better EBP uptake. This embeddedness also relates to clinical champions' institutional savvy [16], which allows them to effectively navigate the complex social hierarchies and culture that exists within their setting/organization. This allows clinical champions to identify points of potential resistance and leverage their relationships/influence to overcome resistance [16]. Not only are clinical champions dedicated to their clinical practice (i.e., frequently on the frontlines), but they are also dedicated to their role and readily embrace change [15, 17, 19]. This dedication to both the innovation and overall implementation effort may provide clinical champions with the drive needed to overcome resistance in their clinical setting, which may be key to their impact on both the implementation effort and provider behavior change [16, 17, 19].
Arguably the most impactful attribute of clinical champions is their standing as informal leaders in their clinical organizations [16, 19, 20]. Informal leaders are regarded as highly influential individuals who do not hold positional authority but are highly respected due to their expertise, trust, and relationship-building capabilities (21–23). The power of informal leaders is defined as “one's ability to initiate action and ensure the desired outcomes are produced” [21]. Informal leaders take time to emerge, as a provider needs to develop both the technical expertise and trust from others before being deemed influential [22]. Clinical champions may be perceived as implementation experts by their peers and, due to being highly respected, are then able to actively engage them in implementation efforts [16, 19, 20]. Although clinical champions may hold formal leadership positions [17], effective clinical champions are typically informal leaders [16, 17]. The power being referenced throughout this paper is in reference to this “subtle” power informal leaders possess. Clinical champions' influence, power, and relationship-building capabilities have been considered crucial to implementation success [14, 16, 19]. Additionally, clinical champions have been described as highly respected and valued individuals within their clinical organization who routinely establish meaningful relationships with their peers [14, 16, 17]. It is through these meaningful relationships that clinical champions can effectively engage their peers in implementation efforts. Thus, the current paper defines informal leadership/influence as the following: an individual who is highly respected by their peers who establish strong and meaningful relationships with them and are viewed as a credible and reliable source of information and skill-sharing [14, 16, 19, 24, 25]. It is important to note that clinical champions may overlap substantially with another group of influential individuals within healthcare organizations, local opinion leaders. Local opinion leaders are individuals who are respected informational sources who have influence over others' decisions and behaviors [24, 25]. Both clinical champions and local opinion leaders are viewed a fellow peers who have an in-depth understanding to a provider's day-to-day experience while also being viewed as credible and reliable source of information [24, 25]. Clinical champions also overlap with early adopters, individuals that readily adopt new ideas, are typically solution-oriented, and embrace innovation [26, 27]. Yet, there are also areas where clinical champions may diverge from these groups. Clinical champions are unique to local opinion leaders, as champions act through charismatic leadership, as opposed to organizational norms and structure [24]. Instead, clinical champions may take an active role toward shifting organizational climate to be more amenable to change [16]. Additionally, clinical champions are unique to early adopters, as early adopters do not necessarily take the responsibility for promoting and spearheading change initiatives in their settings [28].
## Clinician-Level Implementation Barriers
Clinical champions may be particularly relevant to an implementation effort due to their potential to influence the behavior of other frontline providers within their clinical setting. Frontline providers are essential to an EBP implementation effort because they have autonomy to make clinical decisions and can also utilize their power and influence to persuade others to adopt new practices [29]. Each provider has their own unique set of values, interests, and ways of enacting the organizational culture [29], which may facilitate or impede their use of EBPs. The Consolidated Framework for Implementation Research [CFIR, [29]] model has identified provider-level determinants that impact an implementation initiative, including knowledge and attitudes about the implementation initiative (i.e., a specific intervention), self-efficacy (e.g., belief in one's own ability to carry out implementation goals), individual stage of change (e.g., progressing toward becoming a skilled and enthusiastic implementer), and identification with their organization.
Provider-level barriers have been identified across multiple studies and are generally defined as attitudinal or behavioral and involve cognitive or psychological processes that impede or prevent a target behavior from occurring [4]. Commonly cited examples include knowledge, attitudes, and self-efficacy [3, 10]. Barriers surrounding EBP knowledge, competency, or utility in a particular treatment context can impact providers' intentions to enact an EBP and/or their ability to deploy or sustain the EBP [4]. Such barriers can persist and limit a provider's ability to deploy or sustain an EBP, even when they intend to do so [4, 30].
In addition to provider-level barriers, the CFIR model describes implementation determinants at other levels including the inner and outer setting [29]. The outer setting refers to how external policy, economic, or social context impact implementation [29]. In contrast, the inner setting refers to the setting-specific culture, norms, and general characteristics (i.e., geographic location, size) [29]. Together, barriers that exist at different levels of the implementation setting may interact and in turn impact the implementation effort's outcome and success [29, 31]. For example, results from Mosson et al. [ 32] found that managers in smaller organizations or those in rural areas had difficulty with implementation due to barriers that impacted them at the individual level, such as lack of training or preparedness. This occurred, in part, due to the organization's geographical location, causing training opportunities to be infrequent [32].
While it is important to acknowledge barriers at other levels (e.g. organizational, cultural), provider-level barriers may be important to target due to their amenability to change [33]. Targeting an individual's beliefs, attitudes, or behaviors may be more feasible than addressing barriers at the system level. Previous research suggests that when EBP attitudes and beliefs are targeted via social persuasion tactics, EBP utilization and fidelity increased [33]. Previous research has also emphasized a key factor to successful EBP implementation is most successful when line-level clinicians and organizational members are willing and ready to change [34]. Thus, focusing on individual-level attitudes, beliefs, behaviors, or practices may allow for improved design, implementation, and generalizability of strategies to explicitly target such barriers [35, 36].
## Implementation Strategies and the Need For Mechanism Research
To reduce implementation barriers and promote uptake of EBPs in healthcare and other settings, implementation scientists have developed and tested implementation strategies [5, 10]. Implementation strategies can target different populations including patients, clinical providers, stakeholders, and policymakers, as well as different phases (e.g., preparation, delivery, sustainment) and levels (e.g., individual, inner setting, outer setting) of the implementation effort [10]. Given the range of implementation strategies that were being used and reported in the literature, Powell and colleagues [12] compiled a taxonomy of 73 common and discrete implementation strategies, the Expert Recommendations for Implementing Change (ERIC). The ERIC taxonomy was created using an expert panel and aimed to unify and standardize how implementation strategies are referenced and defined in the literature. Although the ERIC taxonomy has effectively allowed for standardized and well-defined implementation strategies, our understanding of how such strategies operate in real-world implementation efforts is limited [9, 10]. Per the ERIC taxonomy [12], the implementation strategy “identify and prepare champions” is defined as, “preparing individuals who are dedicated to supporting, marketing, and driving through an implementation, overcoming indifferences or resistance that the intervention may provoke within an organization.” This strategy is a multi-faceted approach that prioritizes fostering stakeholder relationships, providing mentorship to line-level providers, and overseeing the implementation process (i.e., creating implementation plans, addressing barriers, engaging stakeholders) [17, 37, 38].
An area of the implementation strategy literature that needs to be more thoroughly developed is causal pathway mechanisms [9, 10]. Causal pathway mechanisms outline the processes through which the implementation strategy of interest operate to impact one or more implementation outcomes [8]. By better understanding how strategies operate to yield the desired outcome, researchers will be able to better select and modify implementation strategies to enhance both clinical and implementation outcomes [9, 10]. Understanding processes of implementation strategies will help explain why a certain strategy is effective or ineffective and in what contexts they operate best [8]. Most importantly, understanding processes of implementation strategies will allow for better strategy selection which in turn will better address barriers in the specific implementation context [8]. Per Lewis et al. [ 8], if there is no intentionality behind strategy selection, a suboptimal or less potent strategy may be selected and applied to implementation efforts.
Existing frameworks, such as the Behavior Change Wheel [39], have been developed to examine how behavior change is most likely to occur given intervention components/functions and under certain policy conditions. However, linkages between certain implementation strategies and specific behavioral changes have not been as well explicated. Better understanding mechanisms will require implementation scientists to move beyond simply describing whether a strategy was effective toward developing testable theories that explain relationships between implementation strategies and outcomes as well as allow for outcome prediction [9, 10].
The current paper proposes two causal pathway models for one specific implementation strategy: identifying and preparing clinical champions [5]. This implementation strategy was intentionally selected due to its potential for addressing provider-level barriers. When implementation researchers and practitioners were asked to rank the implementation strategies that would best address each of the CFIR barrier domains, identifying and preparing clinical champions was the most consistently rated strategy across all provider-level barriers [11]. This implementation strategy leverages the influence and respect clinical champions hold within their clinical setting and the interpersonal relationships they form with other providers, which are particularly salient in the context of provider behavior change [40]. By building and engaging in interpersonal relationships within their organizational setting, clinical champions can effectively mitigate provider-level barriers that impact a provider's intention and motivation as well as their ability to deploy an EBP.
## Clinical Champions and Provider Behavior Change
While previous literature emphasizes clinical champions' power and influence as a crucial factor in this strategy's effectiveness [14, 16, 17], exactly how clinical champions change provider behavior has yet to be empirically examined. A conceptual model outlined by Shea [19] posits that a clinical champion's commitment to an implementation effort, as well as their experience, self-efficacy, and performance, promote peer engagement with a clinical champion, which directly influences a clinical champion's impact. A key component of this model is the need for clinical champions to facilitate buy-in from their peers [19]. Clinical champions can promote buy-in by being knowledgeable, trustworthy, and reputable within their organization [16, 19]. Clinical champions can also influence their peers through various forms of power, including expert power (i.e., ability to influence behavior due to skills, knowledge, and abilities), referent power (i.e., impact individual behavior through being well-liked and admired), and informational power (i.e., promote behavior change via exchange of knowledge) [41, 42]. Expert and referent power are considered power that leads to social dependent change, as they are cultivated via how other individuals view the person. Expert power develops through being perceived as credible and trustworthy, while referent power is cultivated through being admired and well-liked [42]. Informational power is considered a power that leads to socially independent change, as disseminating information between the power source and target requires acceptance of knowledge, which the target continues to absorb and apply on their own following knowledge exchange [43].
To outline how peer engagement with a clinical champion can reduce provider-level barriers to implementing EBPs and facilitate behavior-change, we apply two social-cognitive models, the Theory of Planned Behavior [TPB; [44]] and the Health Action Process Approach [HAPA; [45]]. The Theory of Planned Behavior [TPB; [44]] describes how an individual develops intentions to perform a specific behavior and posits that their motivation or intention to engage in a behavior impacts the likelihood of them doing so [44]. Intention development is influenced by an individual's attitudes toward the behavior (i.e., individual's evaluation of the behavior), subjective norms (i.e., perceived social expectations or pressures to engage in a specific behavior), and their perceived behavioral control (i.e., perceived ability to adequately perform the behavior) [40, 46]. Extending beyond intent and motivation, the HAPA model is a social-cognitive dual-phase model that describes the processes needed to transition from intent to action [47]. The HAPA model contains two phases to explain executing a particular behavior: a motivational (intention formation) and volitional phase (planning for behavior enactment and action) [45, 48]. A key component of the HAPA model is the distinction between individuals who are engaging in pre-motivational processes that lead to developing intent and those who are engaging in post-intentional volitional processes or behavioral enactment [45]. During the motivational phase, an individual is considered a “pre-intender” as they are developing intent to adopt a new behavior [45, 49], similar to the TPB. Once intent is established, an individual enters the volitional phase, which is broken into two separate classifications, “intenders” (i.e., individuals intending to perform the behavior in question) and “actors” (i.e., individuals already engaging in desired behavior) [45]. These unique stages of behavior change align well with the types of individual-level barriers providers face, which also can relate to developing intention to use EBPs (i.e., attitudes and knowledge) and deploying EBPs (i.e., planning) [4, 30].
These models were selected to inform the proposed causal pathways due to their relevance to understanding adult behavior change and their routine application in implementation science to address the research-to-practice gap (50–52). These theories clearly explicate individual-level factors that impact behavior change and can also account for the role of others within the implementation context, such as clinical champions, in influencing both implementation intentions and action (e.g., through social norms). These models also have been supported empirically (53–55). The TPB has been routinely used in previous implementation science research aiming to investigate clinical provider's intentions to adopt EBPs [2, 40]. Additionally, the HAPA model provides a unique contribution beyond the TBP by also accounting for behavioral engagement and other factors that address the intention-behavior gap [56]. Whereas other adult-focused behavioral change theories describe constructs that predict overall behavior intention, the HAPA model can also outline potential causal pathway mechanisms relevant to providers who actively transition from intention to behavioral enactment and then need to sustain that behavior enactment over time [51, 56].
Establishing hypothesized causal pathway mechanisms will advance our current understanding of how this specific implementation strategy, identifying and preparing clinical champions, influences provider delivery of EBPs. Understanding these mechanisms is crucial to advancing the scientific literature and will allow for better understanding of existing implementation strategies, strategy development, and appropriate strategy selection for use in specific implementation contexts [9, 10]. The proposed causal pathways explain how clinical champions impact providers in various stages of the behavior change process: pre-intenders (i.e., those developing intent to perform the desired behavior), intenders (i.e., individuals intending to perform the behavior), and actors (i.e., individuals who have engaged in the desired behavior) [45, 49].
## Pathway 1: Pre-intenders
The provider-level barriers most relevant to pre-intenders include attitudes and beliefs about an EBP, perceived utility of the EBP in clinical practice, and lacking confidence in EBP utilization [4, 29, 30]. Strong intent must be formed prior to behavioral enactment and is considered a necessary precursor to behavioral action, with intent being regarded as a bridge between key motivational processes, specifically setting a specific goal (i.e., identifying which novel behavior to adopt) and pursuing said goal (i.e., behavioral enactment) [45]. Strong behavioral intentions are formed when an individual's attitudes toward said behavior are positive, they perceive that their peers endorse said behavior (i.e., establishing subjective norms), and that they can adequately execute the behavior [40]. Please see Figure 1 for visual representation of the proposed causal pathway discussed in detail below.
**Figure 1:** *Causal pathway for development of intention to use EBPs.*
Subjective norms are crucial to target, as subjective norms have been identified as the strongest predictor of intent to utilize EBPs [40]. Subjective norms are not changed through one provider alone, they must become the collective values, beliefs, and behaviors of the clinical setting. Through cultivating multiple one-on-one communication channels advocating for EBP utilization, it is possible that clinical champions promote a shift in norms at the setting-level, which can also impact behavioral or attitudinal shifts at the individual-level. Clinical champions may promote intention development for pre-intenders through shifting a provider's subjective norms due to their informal leadership/influence (i.e., being respected, viewed as credible/reliable source of information, build strong relationships) [2, 24, 25, 40] and strong communication and mentorship skills (i.e., collaborating, advocating, and facilitating learning) [15, 17].
Through their informal leadership/influence, clinical champions are perceived as a credible source and engage in social comparison. Being perceived as a credible source, providers are likely to absorb and apply information about the EBP as said information is coming from a trusted “expert” source [16, 17]. As a credible source, clinical champions also engage in social reinforcement, a crucial process that routinely occurs in the clinical setting, as providers want to determine whether their clinical practices are like others [24, 57]. Providers report higher intent to utilize an EBP when a respected provider within their organization approves of them using or reinforcing the behavior [2, 40]. Other findings suggest that intent to engage in a specific behavior is higher when messages approving the behavior came from a respected colleague as opposed to supervisors [2]. Such findings suggest that in this context, messaging from influential informal leaders impacts behavior more profoundly than when such messages come from formal leadership figures, who may not have direct experience delivering the practice. Through informal leadership/influence, champions exert referent (i.e., being well-liked and admired) and expert power (i.e., ability to influence behavior due to skills, knowledge, and abilities [41, 42], due to being highly respected and reliable sources of information.
Clinical champions may also promote shifts in subjective norms through their strong communication and mentorship skills. Strong communication and mentorship skills involves negotiating and collaborating with others as well as advocating for change and the ability to educate and present information to others [15, 17]. As previously mentioned, communicating about the EBP (e.g., its benefits to patient care) can help facilitate peer buy-in to utilize the EBP [14, 17, 18]. Clinical champions promote buy-in not only by providing accurate information about the EBP, but by also tailoring their message to specific provider groups [15, 16, 18]. Tailoring messaging may allow for the clinical champion to enhance peer buy-in because they are able to address concerns about utilizing the EBP and address other barriers (e.g., knowledge) in a way that can resonate strongly with multiple and varied providers. Thus, the clinical champion may aid in establishing an overall consensus regarding the EBP's importance which ultimately can shift behavior of a specific individual [15, 16]. Through these strong communication and mentorship skills, it is possible that clinical champions exert expert power (i.e., ability to influence behavior due to skills, knowledge, or abilities) and informational power (i.e., promote behavior change via knowledge exchange) [41, 42] as they utilize their communication skills to promote knowledge and skill-building in their peers [14, 16].
Clinical champions also promote intention development through changing pre-intenders' attitudes toward the EBP itself. In addition to changing subjective norms, a clinical champion's informal leadership/influence can also facilitate change in provider's attitudes. Again, by being a credible source, highly respected, and having strong interpersonal relationships with peers, when clinical champions advocate for and promote use of an EBP in clinical practice, their opinions regarding how to deliver high-quality care are likely to be taken seriously by their peers [14, 16]. Additionally, clinical champions EBP knowledge and competency may also be highly influential on provider attitudes toward the EBP. Clinical champions must have high levels of knowledge and competency about the EBP they're trying to implement to effectively prepare their fellow providers to integrate the EBP into their own practice [14, 17]. Through disseminating this knowledge, clinical champions may be able to effectively address misconceptions about the EBP, address questions providers may have about the EBP, all of which may facilitate development of positive outcome expectancies, an essential component of the motivational phase of the HAPA model [45]. Clinical champions with high levels of EBP knowledge and competency are likely to regularly promote and demonstrate the EBP's utility in the clinical context, which may also promote changes in provider's attitudes toward the EBP [14, 16, 17, 58]. A clinical champion's knowledge and competency allows them to exert both expert and informational power (i.e., promote behavior change via knowledge change) as they utilize their own knowledge to persuade peers to adopt the EBP [41, 42].
Clinical champion's may also impact provider's attitudes toward the EBP through their strong communication and mentorship skills. By being exposed to accurate information from their peers, it is plausible that providers can develop more positive attitudes about the EBP itself, especially when said information is coming from a well-liked and respected source. Per the Diffusion of Innovation theory [DOI, [59]], an individual's decision to adopt a specific practice is, in part, influenced by acquiring knowledge about the innovation and being persuaded that the practice will benefit them [59, 60]. Clinical champion's strong communication and mentorship may allow for an individual obtaining the necessary knowledge about the EBP to deem it beneficial while also being persuaded to adopt the innovation from a highly respected peer within their social network. Through these processes, clinical champions are again exerting their expert and informational power.
Lastly, clinical champions may also enhance pre-intenders' perceived behavioral control through their strong communication and mentorship skills. Strong communication skills utilized by clinical champions may enhance pre-intender's perceived behavior control, as a common practice of clinical champions includes individualized training and providing tailored feedback (14–16, 58). Such behaviors require a clinical champion to be able to effectively collaborate with peers, advocate for EBP use, and provide individualized feedback. Similarly, strong mentorship skills have been operationalized as willingness to facilitate learning, collaboration, and to provide education [61]. Strong mentorship skills may allow for clinical champions to enhance pre-intender's perceived behavioral control because they received the necessary education and support needed to develop more confidence in their ability to apply the skills needed to utilize the EBP effectively. It is important to note a clinical champion's mentorship may be most impactful when they are committed to the implementation effort and training [15, 19].
## Pathway 2: Intenders and Actors
In contrast to pre-intenders, intenders have developed the necessary intent needed to enact the behavior but have not transitioned from intent to action (i.e., enacting the desired behavior) [45, 49]. Provider-level barriers commonly experienced by intenders are those related to EBP deployment (i.e., lack of planning and skills) [4]. Intenders have successfully transitioned into the volitional phase (i.e., the phase in which someone develops plans to act or has transitioned to action) but have yet to become an actor. The HAPA model suggests that action and coping planning are critical for translating intention into action for intenders [45]. Previous research suggests that action planning aids individuals to effectively identify cues for behavioral engagement as well as develop actionable steps needed to effectively execute the behavior of interest [47, 62]. In contrast, coping planning entails identifying potential barriers that may impact behavior execution and developing plans to mitigate them [47].
## Intenders
For intenders, clinical champions' promote behavior change through aiding in action and coping planning, due to their embeddedness in the clinical setting (i.e., being present on the front lines, frequent face-to-face contact with peers) [16, 45]. Frequent presence on the frontlines allows clinical champions to regularly and readily engage with clinical providers to support their integration of EBP into their routine practice, which in turn promotes skill building and planning [16, 17]. On a similar note, frequent face-to-face contact with fellow providers also facilitates skill building and planning, as doing so allows for providers to easily access support and mentorship from clinical champions when needed [16, 17]. Face-to-face engagement when providing education and skill-building is known to be more impactful than passive education strategies, like treatment guides or websites [12, 17]]. Through their embeddedness, clinical champions may also be exerting their referent [i.e., being well-liked and admired [41]] and expert power, as they leverage their relationships and ability to cultivate strong social bonds to facilitate EBP adoption.
Clinical champion's informal leadership/influence also promotes action and coping planning. To develop skills and outline actionable steps to integrate EBPs into clinical practice, providers need an individual whom they respect and view as a credible source of information whom they can turn to for advice, education, and support [14, 16, 19, 23]. As clinical champions wield influence and are viewed as informal leaders, it is possible that intenders trust and rely on the clinical champion as a resource to aid in planning and skill development [16, 17, 58]. Additionally, any mentorship, training, and education provided by the clinical champion is taken seriously by their peers due to their informal leadership/influence [14, 16].
Lastly, development and enhancement of provider's action and coping planning skills can be impacted by clinical champion's strong communication and mentorship skills (i.e., negotiating, collaborating, advocating, and educating/facilitating learning). Through their strong communication skills, clinical champions may increase clinicians' self-efficacy and over time can lead to sustained used of the EBP during clinical care [17, 63]. Strong communication and mentorship skills also aids providers in engaging in action planning. Action planning involves breaking down the EBP or intervention into easy to execute steps, meaning it is crucial for a clinical champion to be an effective communicator [56], so these steps are easily understood by the intender. Previous research in both medical and education contexts have investigated the benefits of peer-to-peer coaching and skill sharing, with findings suggesting that such teaching approaches promote the use of evidence-based practices, reflection on current workplace practices, and collaborative discussion [64, 65]. Additionally, medical settings that applied peer coaching practices have found peer coaching improves educators' skill transferability, confidence, and overall satisfaction [64].
A clinical champion's EBP knowledge and competency may also promote intender's development of action planning, coping planning, and achieving behavioral enactment [19]. Skills in these domains relate to both engaging in the behavior of interest as well as performing the clinical champion role itself [19]. Thus, clinical champions will be most impactful on intender behavior change when they themselves are knowledgeable about the EBP as well as have high self-efficacy and confidence in their ability to use the EBP as well as lead and mentor other providers [19, 66]. Through their previous experience and high self-efficacy, clinical champions themselves have already engaged in all action planning components, including defining EBP steps, outlining how to execute each step of the EBP during clinical care, and identifying potential barriers [56]. It is possible that clinical champions themselves have also engaged in coping planning by identifying barriers that occurred in their own clinical practice and developed steps to mitigate them [56].
## Actors
Lastly, clinical champions can also aid a third group of providers, actors (i.e., individuals who have begun actively enacting the desired behavior), in maintaining their behavioral enactment and continuing to use the EBP. Although actors have successfully enacted the desired behavior (i.e., EBP use), they still experience barriers and can benefit from support. A common barrier cited by providers is that implementing EBPs consistently and with fidelity is challenging [4]. Relevant volitional phase processes for actors include maintenance self-efficacy (i.e., perceived ability to maintain desired behavior) and action control (i.e., self-monitoring, awareness of standards, effort) [45, 49]. Through clinical champions strong communication and mentorship skills, actors can receive continuous mentorship and training/education regarding EBP utilization, which in turn allows for development of maintenance self-efficacy and consistent EBP use [56, 63].
Clinical champions' can also impact actor's action control through their embeddedness in the clinical setting (i.e., presence on the frontlines and frequent face-to-face contact with peers). By being embedded in the clinical setting, clinical champions hold actors accountable to consistently utilize the EBP when relevant. Previous research also suggests that due to their embeddedness, clinical champions often engage regularly in compliance monitoring [16, 58]. Holding actors accountable can promote development of action control, as actors will begin to self-monitor and be more self-aware of EBP use in their clinical practice. See Figure 2 for visual representation of this causal pathway.
**Figure 2:** *Causal pathway for development of behavioral enactment and maintenance.*
## Pathway Summary
In summary, clinical champions largely operate through social influence. Social influence involves one individual to evaluate a respected figure's own perceptions of the specific innovation [59, 67]. Social influence is interwoven with diffusion, a social process in which members of a specific social system communicate an adoption decision for a particular practice [59, 67]. Clinical champions and similar roles may impact diffusion of an innovation through their adoption and advocating of the EBP, as doing so may influence the social system to shift their normative practices (i.e., accepting the innovation or practice) [67]. Such roles may also aid in accelerating one's decision to adopt the EBP [60, 67]. Adoption decisions are often accelerated when an influential person within a social system adopts the innovation themselves and communications that decision across the social network [67].
Secondly, clinical champions wield social influence due to their resemblance to their peers because they too are often fellow line-level providers. Potential adopters of a specific innovation often seek judgment from a trusted and respected “expert” peer [67], and this interpersonal communication is known to be more impactful when there are professional similarities between communicators [60]. Per the Diffusion of Innovation theory [DOI, [59]] there are five factors involved in an individual's process in deciding to adopt an innovation: acquiring knowledge about the innovation, being persuaded the innovation is beneficial, engaging in activities that may impact a choice, incorporating the innovation into daily workflow, and seeking reinforcement about their decision [59, 60]. Clinical champions may be tapping into all these factors, as they often provide education to peers about the EBP, attempt to persuade individuals to adopt the EBP, validate one's utilization of the EBP and decision to adopt, and provide the support needed for the individual to incorporate the EBP into their routine patient care [2, 14, 16].
## How Causal Pathways Inform Clinical Champion Selection
These causal pathways outline how a clinical champion, once identified, can influence provider behavior change. However, these pathways do not include the selection or identification of clinical champions themselves. Despite research that has described common characteristics of clinical champions, less is known regarding how clinical champions are selected, identified, and trained [17, 68]. A review conducted by Wood et al. [ 17] suggests that clinical champions are typically selected/recruited or emergent/self-designated. Selected champions are providers who typically had prior experience utilize the EBP or intervention, whereas emergent champions take on this role due to an inherent interest, expertise, or conviction for the EBP or intervention [17]. In contrast, a review of clinical champion utilization in nursing homes by Woo et al. [ 68] found that clinical champions were either selected by clinical leadership or selection was not adequately described. Similar trends have also been observed regarding clinical champion training, with many studies not adequately describing how clinical champions are trained or stating whether they were trained at all [17, 68]. The few studies that have described clinical champion training report common training modalities including in-person workshops about the intervention, online training modules, and education about implementation and leadership strategies [37, 38, 69].
The pathways outlined above can provide suggestions for the recruitment, identification, and training of clinical champions. When attempting to identify potential clinical champions, it is important to first observe the clinical setting and take note of the interpersonal relationships that exist [21]. Seek the provider who is not only able to form strong interpersonal connections with their peers, but also the one to whom others gravitate toward for guidance and support, as these providers have influence over others and may be strong informal leaders [21, 23]. Previous research suggests that for behavioral modeling and knowledge transfer to be most impactful, the individual must identify and relate to the individual modeling the desired behavior [70, 71]. If the clinical champion selected is not representative of most providers being targeted, this implementation strategy may not generate the desired effect (i.e., provider behavior change). Thus, when selecting clinical champions, it is important to seek providers who display these informal leadership attributes, such as expertise, trust, and influence [16, 21, 23]. Although not much research has been focused on the selection of clinical champions, research examining the selection of key opinion leaders can provide some insight into best practices. For example, [72] suggest that local opinion leaders can be identified via self-selection (i.e., individuals volunteer due to personal reasons or strong desire to serve), staff selection (i.e., project staff select leaders based on observations), “judge ratings”, (i.e., fellow line-level providers select local opinion leaders rather than formal leadership staff). This final identification method may be well-suited for selecting clinical champions, as it relies on fellow providers and is easy to implement in larger-sized settings [72].
## Additional Considerations for Future Research
In the current paper, we have hypothesized causal pathways rationalized with well-established scientific behavior change models to explore how and why clinical champions facilitate provider behavior change. These pathways explain how clinical champions can mitigate provider-level barriers commonly experienced by providers who are in various stages of the behavior change process (i.e., pre-intenders, intenders, and actors). Explicating potential causal pathways is one step toward better understanding implementation strategies and enhancing their impact as they connect implementation strategies to behavior change theories and go beyond describing general effectiveness [9, 10]. While these causal pathways outline how clinical champions prompt provider behavior change, the next step is to empirically test them; this can inform our understanding of which implementation strategies operate best in certain contexts [9, 10]. A review by Lewis et al. [ 8] found that mediation models were the most common approach for examining how a specific implementation strategy impacted the relationships between implementation outcomes, with implementation determinants acting as a mediator. A mediation model design could examine whether the presence of a clinical champion and their embodiment of the characteristics described above impacts the outlined mechanisms (e.g., subjective norms, action planning) and whether those mechanisms in turn impact EBP use and fidelity. Moderated-mediation models could also examine how clinical champions could interact with features of the implementation context at the organizational or community level, as discussed further below. Future research should test the proposed causal pathway models in different contexts and settings to observe how these proposed mechanisms operate across different healthcare settings.
## Contextual Considerations
It is important to discuss the importance of contextual factors, specifically the inner (i.e., organization's size, location, culture, climate) and outer (i.e., political, economic, social factors) setting, as clinical champions are impacted greatly by the environments in which they are engaging and trying to transform [29, 32]. As implementation efforts occur in complex systems that have various levels of influence ranging from the system- to patient-level, it is important to understand how these factors interact with implementation strategies and impact implementation [73]. Barriers at both the inner and outer setting levels can impede leadership's impact on achieving successful implementation efforts [32, 73]. Understanding how organizational context interacts with implementation strategies is needed to further the field's understanding of how implementation strategies operate and yield desired outcomes. It is possible that implementation strategies can optimize uptake and application of EBPs by either making the inner and outer setting more amenable to implementation or by adapting the EBP to better fit within the organization [73].
Common outer setting barriers include organization and policy makers interest and willingness to support the implementation initiative and other policy-related barriers (e.g., financial disincentives) [16, 32]. Implementation efforts occurring within organizations who supported the implementation initiative via funding or trainings (i.e., providing trainings for how to utilize the EBP) typically had more impactful implementation outcomes and success [32]. Supportive environments allow for implementation leaders to provide staff with ample high-quality training opportunities and the financial support needed to invest in the implementation process [32]. Results from [16] observed that clinical champions working in an environment with substantial outer context barriers, such as financial disincentives, resulted in unsuccessful implementation and even de-implementation due to the inability to overcome such barriers. In environments where outer context barriers are prevalent and/or outside the sphere of a clinical champion's reach, this implementation strategy may not be the most impactful [16].
Inner setting factors are described as structural (e.g., size, geographic location) and modifiable (e.g., culture, climate, readiness for implementation, communication) [29, 32]. Inner setting barriers that are relevant to clinical champion's impact include implementation climate [32, 73, 74], staff resistance to change [16], and relationships between clinical champions and organization leadership [16, 32]. As clinical champion's leverage their influence and ability to form strong interpersonal relationships [16, 17], for settings with suboptimal or low-quality relationships between line-level staff and formal leadership or administrators, clinical champions may experience substantial difficulties achieving success due to an inability to establish buy-in or overcome resistance from leadership [16, 32]. Strong positive relationships between clinical champions and implementation leadership are crucial, as leadership is critical to a successful implementation effort [75]. As leadership provides support, feedback, and guidance to implementation [75], clinical champions with poor relationships with leaders may not be able to overcome certain barriers.
Implementation climate [i.e., extent to which an organization's policy, practice, and culture is amenable to implementing innovative and new practices [66]] is highly relevant and impactful regarding clinical champions and their impact on both the implementation effort and provider behavior change. Implementation climate has been identified as a key indicator of both staffs' prolonged use of a specific innovation as well as the quality in which the innovation is delivered [74]. A supportive implementation climate in which the organization is open to new ideas [66] and both leadership and line-level staff prioritize using EBPs creates optimal conditions for clinical champions (and other implementation leaders) to successfully facilitate implementation [32, 66]. When implementation climate is poor, clinical champions may experience challenges related to communicating and educating line-level providers about the EBP, which will hinder their overall impact and implementation success [16].
## Equity
Future research should also prioritize understanding clinical champions in the context of health equity and disparity research. To move toward equitable health outcomes, implementation science needs to proactively tailor and modify implementation strategies and EBPs to address health disparities [76, 77]. This includes how implementation strategies can facilitate equitable implementation across healthcare services [76, 77]. To our knowledge, previous research has yet to explicitly investigate how clinical champions impact health disparity implementation efforts and should be explored in future work. For example, it will be important to identify whether certain characteristics of clinical champions are particularly critical (and whether there might be additional necessary characteristics) for implementing practices that are explicitly aimed at reducing biases and inequities within the healthcare system.
In addition, related to the discussion above regarding context, it is important to consider the heterogeneity of resources that exist across the healthcare system and how such differences impact clinical champions. Populations experiencing health inequities often receive services in low-resource settings, which can be impacted by staff shortages, higher staff turnover, and lack of funding [77, 78]. Thus, these settings require additional implementation resources to be successful, including both financial resources and leadership supports [77, 79]. Low-resource settings often experience barriers at the inner setting level (i.e., staff turnover, resource shortages, competing demands), which may reduce the impact clinical champions have. When testing implementation strategy mechanisms, an explicit focus should be given to the likelihood for the strategy to have its intended impact across varied settings, without implicitly assuming that equitable outcomes will be facilitated through the use of the implementation strategy [80].
It is also critical to consider the potential for inequities to arise throughout the process of recruiting, selecting, and training clinical champions. Enacting implementation strategies without a focus on equity could allow disparities to emerge in both implementation and clinical outcomes [80]. Although it is important for clinical champions to be well-networked and hold influence within their organizations, power within organizations can replicate hierarchies or systems of power and privilege reflected in society more broadly [63, 81]. Perceptions of core clinical champion characteristics, such as trust/respect, communication, and leadership are influenced by socio-cultural positionality, including, but not limited to, race/racism, gender/sexism, and ability/ableism [82]. Biases can emerge when identifying certain providers as leaders for example [e.g., [83, 84]] and therefore the role of culture and positionality should not be ignored when identifying clinical champions. Power imbalance and hierarchies present in both healthcare teams and organizations can have negative impacts on implementation. Power imbalances that plague healthcare organizations include those related to communication (i.e., receptivity and responsiveness from leadership), trust and respect, as well as role allocation (i.e., lack of recognition or delineation of duties) [81]. For clinical champions, imbalances related to communication may impede collaboration and educating peers about the EBP, which may reduce their overall effectiveness. Power imbalances in the context of trust and respect may be the most detrimental to clinical champions, as without trust and respect clinical champions yield no influence as informal leadership [16, 21, 34]. Additionally, if there is a lack of trust and respect between a clinical champion and formal leadership, clinical champions may experience irreconcilable barriers to implementing change.
## Limitations of Clinical Champions
It is also important to address potential weaknesses of the clinical champion implementation strategy. While the current paper focuses on clinical champion attributes and the mechanisms through which these attributes exert their effects, not all clinical champions display the described characteristics, embrace their role, or achieve success. A common challenge experienced by clinical champions involved limited bandwidth and competing demands that prevent them from fulfilling their clinical champion role duties [14, 16, 58]. Clinical champions are often also providing patient care, which is their primary responsibility as a healthcare provider. Such competing demands could impact a clinical champion's ability to provide adequate training/education about the EBP and monitor usage and fidelity [14, 58]. Another weakness of clinical champions noted in previous literature is lack of ownership of the initiative or role [14, 16]. Not all clinical champions prioritize the EBP being implemented in their organization or may have other responsibilities they are more dedicated to [14, 16]. As embracing the initiative and using their drive and commitment to motivate others to adopt the EBP, clinical champions lacking this dedication may not be as successful. Clinical champions may experience social barriers or weaknesses that impede them from fulfilling their duties, such as lacking influence [16] and navigating boundaries with their fellow peers [58]. Part of the clinical champion's role is to observe their peers and provide them with feedback or point out ways in which they can improve their utilization of an EBP, some clinical champion's may feel conflicted or uncomfortable navigating boundaries with their peers or hierarchies that exist within their clinical setting [58]. Lastly, it is important to acknowledge that clinical champions alone may not be sufficient in achieving system-level change [15, 16]. Clinical champions may be particularly effective in environments with engaged and supportive leadership, training supports, and a positive implementation climate.
It is also possible that clinical champions may lack certain attributes that can be cultivated with further training and support. For example, a clinical champion may be identified who, although is a strong informal leader who wields influence, may not have EBP-specific expertise. In such cases, EBP-specific trainings should be provided to allow for expertise to be established [38]. In contrast, some clinical champions may not possess strong informal leadership skills, in which case leadership trainings should be provided [37]. As the field continues to develop an understanding of this implementation strategy, further research is needed to explore how we can better support and prepare clinical champions so they can be most effective.
## Conclusions
Clinical champions likely play a critical role in reducing provider implementation barriers for clinicians across various phases of behavior change (e.g., pre-intenders, intenders, and actors). Clinical champions promote the development of intention by displaying significant knowledge about the EBP, behavioral modeling of the EBP's utility, and establishing peer buy-in. In contrast, clinical champions aid intenders to transition into behavioral enactment through skill building which can promote sustained usage of the EBP. This paper contributes to the gap in scholarship outlining potential mechanisms of implementation strategies and can lead to future research testing such mechanisms.
## Author Contributions
AM devised the main conceptual ideas and outline, developed the pathway mechanisms, guided and supervised by LG, and drafted mechanisms figures with guidance from CL. AM and LG wrote the manuscript, with consultation and input from CL. All authors contributed to the article and approved the submitted version.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict ofinterest.
## Publisher's Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Lewis CC, Boyd M, Puspitasari A, Navarro E, Howard J, Kassab H. **Implementing measurement-based care in behavioral health: A review**. *JAMA Psychiatr.* (2019) **76** 324-35. DOI: 10.1001/jamapsychiatry.2018.3329
2. Lushin V, Becker-Haimes EM, Mandell D, Conrad J, Kaploun V, Bailey S. **What motivates mental health clinicians-in-training to implement evidence-based assessment? A survey of social work trainees**. *Adm Policy Mental Health Mental Health Serv Res.* (2019) **46** 411-24. DOI: 10.1007/s10488-019-00923-4
3. Stewart RE, Williams N, Byeon YV, Buttenheim A, Sridharan S, Zentgraf K. **The clinician crowdsourcing challenge: using participatory design to seed implementation strategies**. *Implement Sci.* (2019) **14** 63. DOI: 10.1186/s13012-019-0914-2
4. Stewart RE, Beidas RS, Last BS, Hoskins K, Byeon YV, Williams NJ. **Applying NUDGE to inform design of EBP implementation strategies in community mental health settings**. *Adm Policy Mental Health Mental Health Serv Res.* (2021) **48** 131-42. DOI: 10.1007/s10488-020-01052-z
5. Powell BJ, McMillen JC, Proctor EK, Carpenter CR, Griffey RT, Bunger AC. **A Compilation of strategies for implementing clinical innovations in health and mental health**. *Med Care Res Rev.* (2012) **69** 123-57. DOI: 10.1177/1077558711430690
6. Proctor EK, Landsverk J, Aarons G, Chambers D, Glisson C, Mittman B. **Implementation research in mental health services: an emerging science with conceptual, methodological, and training challenges**. *Adm Policy Mental Health Mental Health Serv Res.* (2009) **36** 24-34. DOI: 10.1007/s10488-008-0197-4
7. Proctor EK, Powell BJ, McMillen JC. **Implementation strategies: Recommendations for specifying and reporting**. *Implement Sci.* (2013) **8** 139. DOI: 10.1186/1748-5908-8-139
8. Lewis CC, Boyd MR, Walsh-Bailey C, Lyon AR, Beidas R, Mittman B. **A systematic review of empirical studies examining mechanisms of implementation in health**. *Implement Sci.* (2020) **15** 21. DOI: 10.1186/s13012-020-00983-3
9. Lewis CC, Klasnja P, Powell BJ, Lyon AR, Tuzzio L, Jones S. **From classification to causality: advancing understanding of mechanisms of change in implementation science**. *Front Public Health.* (2018) **6** 136. DOI: 10.3389/fpubh.2018.00136
10. Powell BJ, Fernandez ME, Williams NJ, Aarons GA, Beidas RS, Lewis CC. **Enhancing the impact of implementation strategies in healthcare: a research agenda**. *Front Public Health.* (2019) **7** 3. DOI: 10.3389/fpubh.2019.00003
11. Waltz TJ, Powell BJ, Fernández ME, Abadie B, Damschroder LJ. **Choosing implementation strategies to address contextual barriers: diversity in recommendations and future directions**. *Implem Sci.* (2019) **14** 42. DOI: 10.1186/s13012-019-0892-4
12. Powell BJ, Waltz TJ, Chinman MJ, Damschroder LJ, Smith JL, Matthieu MM. **A refined compilation of implementation strategies: results from the Expert Recommendations for Implementing Change (ERIC) project**. *Implement Sci.* (2015) **10** 21. DOI: 10.1186/s13012-015-0209-1
13. Cranley Cranley LA, Cummings GG, Profetto-McGrath J, Toth F, Estabrooks CA. **Facilitation roles and characteristics associated with research use by healthcare professionals: a scoping review**. *BMJ Open* (2017). DOI: 10.1136/bmjopen-2016-014384
14. Flanagan ME, Plue L, Miller KK, Schmid AA, Myers L, Graham G. **A qualitative study of clinical champions in context: clinical champions across three levels of acute care**. *SAGE Open Med* (2018). DOI: 10.1177/2050312118792426
15. Miech EJ, Rattray NA, Flanagan ME, Damschroder L, Schmid AA, Damush TM. **Inside help: An integrative review of champions in healthcare-related implementation**. *SAGE Open Med.* (2018) **6** 205031211877326. DOI: 10.1177/2050312118773261
16. Bonawitz K, Wetmore M, Heisler M, Dalton VK, Damschroder LJ, Forman J. **Champions in context: Which attributes matter for change efforts in healthcare?**. *Implement Sci.* (2020) **15** 62. DOI: 10.1186/s13012-020-01024-9
17. Wood K, Giannopoulos V, Louie E, Baillie A, Uribe G, Lee KS. **The role of clinical champions in facilitating the use of evidence-based practice in drug and alcohol and mental health settings: a systematic review**. *Implement Res Pract.* (2020) **1** 263348952095907. DOI: 10.1177/2633489520959072
18. Rycroft-Malone J, Gradinger F, Owen Griffiths H, Anderson R, Crane RS, Gibson A. **‘Mind the gaps': The accessibility and implementation of an effective depression relapse prevention programme in UK NHS services: learning from mindfulness-based cognitive therapy through a mixed-methods study**. *BMJ Open.* (2019) **9** e026244. DOI: 10.1136/bmjopen-2018-026244
19. Shea CM. **A conceptual model to guide research on the activities and effects of innovation champions**. *Implement Res Pract.* (2021) **2** 2633489521990443. DOI: 10.1177/2633489521990443
20. Forchuk C, Martin ML, Jensen E, Ouseley S, Sealy P, Beal G. **Integrating an evidence-based intervention into clinical practice: ‘Transitional relationship model': integrating an evidence-based intervention into practice**. *J Psychiatric Mental Health Nurs* (2012) **20** 584-94. DOI: 10.1111/j.1365-2850.2012.01956.x
21. Downey M, Parslow S, Smart M. **The hidden treasure in nursing leadership: Informal leaders**. *J Nurs Manag.* (2011) **19** 517-21. DOI: 10.1111/j.1365-2834.2011.01253.x
22. Lawson D, Fleshman JW. **Informal leadership in health care**. *Clin Colon Rectal Surg.* (2020) **33** 225-7. DOI: 10.1055/s-0040-1709439
23. Smart M. *Informal Leadership* (2010)
24. Locock L, Dopson S, Chambers D, Gabbay J. **Understanding the role of opinion leaders in improving clinical effectiveness**. *Soc Sci Med.* (2001) **53** 745-57. DOI: 10.1016/S0277-9536(00)00387-7
25. Curran GM, Thrush CR, Smith JL, Owen RR, Ritchie M, Chadwick D. **Implementing research findings into practice using clinical opinion leaders: Barriers and lessons learned**. *Joint Commis J Qual Patient Safe.* (2005) **31** 700-7. DOI: 10.1016/S1553-7250(05)31091-9
26. Borbas C, Morris N, McLaughlin B, Asinger R, Gobel F. **The role of clinical opinion leaders in guideline implementation and quality improvement**. *Chest.* (2000) **118** 24S-32S. DOI: 10.1378/chest.118.2_suppl.24S
27. Haun MW, Stephan I, Wensing M, Hartmann M, Hoffmann M, Friederich H-C. **Intent to adopt video-based integrated mental health care and the characteristics of its supporters: Mixed methods study among general practitioners applying diffusion of innovations theory**. *JMIR Ment Health.* (2020) **7** e23660. DOI: 10.2196/23660
28. Rogers EM, Shoemaker FF. **Diffusion of innovations**. *Communication of Innovations: A Cross-Cultural Approach. 2nd ed.* (1971)
29. Damschroder LJ, Aron DC, Keith RE, Kirsh SR, Alexander JA, Lowery JC. **Fostering implementation of health services research findings into practice: a consolidated framework for advancing implementation science**. *Implement Sci.* (2009) **4** 50. DOI: 10.1186/1748-5908-4-50
30. Wakida EK, Talib ZM, Akena D, Okello ES, Kinengyere A, Mindra A. **Barriers and facilitators to the integration of mental health services into primary health care: a systematic review**. *Syst Rev.* (2018) **7** 211. DOI: 10.1186/s13643-018-0882-7
31. Lukas CV, Holmes SK, Cohen AB, Restuccia J, Cramer IE, Shwartz M. **Transformational change in health care systems: An organizational model**. *Health Care Manag Rev.* (2007) **32** 309-20. DOI: 10.1097/01.HMR.0000296785.29718.5d
32. **The impact of inner and outer context on line managers' implementation leadership**. *Br J Social Work* (2018) **48** 1447-68. DOI: 10.1093/bjsw/bcx077
33. Cook CR, Lyon AR, Kubergovic D, Browning Wright D, Zhang Y. **A supportive beliefs intervention to facilitate the implementation of evidence-based practices within a multi-tiered system of supports**. *School Ment Health.* (2015) **7** 49-60. DOI: 10.1007/s12310-014-9139-3
34. Smith BD. **Substance use treatment counselors' attitudes toward evidence-based practice: the importance of organizational context**. *Subst Use Misuse.* (2013) **48** 379-90. DOI: 10.3109/10826084.2013.765480
35. Davis M, Siegel J, Becker-Haimes EM, Jager-Hyman S, Beidas RS, Young JF. **Identifying common and unique barriers and facilitators to implementing evidence-based practices for suicide prevention across primary care and specialty mental health settings**. *Arch Suicide Res.* (2021). DOI: 10.1080/13811118.2021.1982094.
36. Nakamura BJ, Higa-McMillan CK, Okamura KH, Shimabukuro S. **Knowledge of and attitudes towards evidence-based practices in community child mental health practitioners**. *Adminis Policy Mental Health Mental Health Serv Res.* (2011) **38** 287-300. DOI: 10.1007/s10488-011-0351-2
37. Kipping SM, De Souza JL, Marshall LA. **Co-creation of the safewards model in a forensic mental health care facility**. *Issues Ment Health Nurs.* (2019) **40** 2-7. DOI: 10.1080/01612840.2018.1481472
38. O'Sullivan PS, Yuan P, Satre DD, Wamsley M, Satterfield J. **A sequential implementation model for workforce development: a case study of medical residency training for substance use concerns**. *Teach Learn Med.* (2018) **30** 84-94. DOI: 10.1080/10401334.2017.1314216
39. Michie S, Atkins L, West R. *The behaviour change wheel. A guide to designing interventions. 1st ed.* (2014) 1003-10
40. Kelly PJ, Deane FP, Lovett MJ. **Using the theory of planned behavior to examine residential substance abuse workers intention to use evidence-based practices**. *Psychol Addict Behav.* (2012) **26** 661-4. DOI: 10.1037/a0027887
41. French JR, Raven B, Cartwright D. **The bases of social power**. *Classics Organ Theory.* (1959) **7** 311-20
42. Lunenburg FC. **Power and leadership: an influence process**. *Int J Manage Bus Adm.* (2012) **15** 1-9
43. Raven BH. **The bases of power and the power/interaction model of interpersonal influence**. *Analyses Soc Issues Public Policy (ASAP).* (2008) **8** 1-22. DOI: 10.1111/j.1530-2415.2008.00159.x
44. Ajzen I. **The theory of planned behavior**. *Organ Behav Human Decis Process.* (1991) **50** 179-211. DOI: 10.1016/0749-5978(91)90020-T
45. Schwarzer R. **Modeling health behavior change: how to predict and modify the adoption and maintenance of health behaviors**. *Appl Psychol.* (2008) **57** 1-29. DOI: 10.1111/j.1464-0597.2007.00325.x
46. Burgess AM, Chang J, Nakamura BJ, Izmirian S, Okamura KH. **Evidence-based practice implementation within a theory of planned behavior framework**. *J Behav Health Serv Res.* (2017) **44** 647-65. DOI: 10.1007/s11414-016-9523-x
47. Zhang C-Q, Zhang R, Schwarzer R, Hagger MS. **A meta-analysis of the health action process approach**. *Health Psychol* (2019) **38** 623-637. DOI: 10.1037/hea0000728
48. MacPhail M, Mullan B, Sharpe L, MacCann C, Todd J. **Using the health action process approach to predict and improve health outcomes in individuals with type 2 diabetes mellitus**. *Diabetes Metab Syndr Obesity Targets Ther.* (2014) **7** 469-79. DOI: 10.2147/DMSO.S68428
49. Sniehotta FF, Scholz U, Schwarzer R. **Bridging the intention-behaviour gap: Planning, self-efficacy, and action control in the adoption and maintenance of physical exercise**. *Psychol Health.* (2005) **20** 143-60. DOI: 10.1080/08870440512331317670
50. Lyon AR, Cook CR, Duong MT, Nicodimos S, Pullmann MD, Brewer SK. **The influence of a blended, theoretically-informed pre-implementation strategy on school-based clinician implementation of an evidence-based trauma intervention**. *Implement Sci.* (2019) **14** 54. DOI: 10.1186/s13012-019-0905-3
51. Lyon AR, Pullmann MD, Dorsey S, Levin C, Gaias LM, Brewer SK. **Protocol for a hybrid type 2 cluster randomized trial of trauma-focused cognitive behavioral therapy and a pragmatic individual-level implementation strategy**. *Implement Sci.* (2021) **16** 3. DOI: 10.1186/s13012-020-01064-1
52. Zhang Y, Cook C, Larson M, Azad G, Merle J, Thayer J. **A theory-informed pre-implementation enhancement strategy targeting mechanisms and outcomes of implementation: A triple-blind randomized controlled trial**. *Res Squ. [Preprint]* (2022). DOI: 10.21203/rs.3.rs-1359752/v1
53. **Changing health-related behaviours: An approach based on the theory of planned behaviour**. *The Scope of Social Psychology* (2007) 55-76
54. Steinmetz H, Knappstein M, Ajzen I, Schmidt P, Kabst R. **How effective are behavior change interventions based on the theory of planned behavior?**. *Zeitschrift Für Psychologie.* (2016) **224** 216-33. DOI: 10.1027/2151-2604/a000255
55. Potthoff S, Presseau J, Sniehotta FF, Johnston M, Elovainio M, Avery L. **Planning to be routine: Habit as a mediator of the planning-behaviour relationship in healthcare professionals**. *Implement Sci.* (2017) **12** 24. DOI: 10.1186/s13012-017-0551-6
56. Sanetti LMH, Kratochwill TR, Long ACJ. **Applying adult behavior change theory to support mediator-based intervention implementation**. *School Psychol Q.* (2013) **28** 47-62. DOI: 10.1037/spq0000007
57. West E, Barron DN, Dowsett J, Newton JN. **Hierarchies and cliques in the social networks of health care professionals: Implications for the design of dissemination strategies**. *Soc Sci Med* (1999) **48** 633-46. DOI: 10.1016/S0277-9536(98)00361-X
58. Goedken CC, Livorsi DJ, Sauder M, Vander Weg MW, Chasco EE, Chang N-C. **“The role as a champion is to not only monitor but to speak out and to educate”: the contradictory roles of hand hygiene champions**. *Implement Sci* (2019) **14** 110. DOI: 10.1186/s13012-019-0943-x
59. Rogers EM. *Diffusion of Innovations. 3rd ed.* (1983)
60. Sanson-Fisher RW. **Diffusion of innovation theory for clinical change**. *Med J Aust.* (2004) **180** S55-6. DOI: 10.5694/j.1326-5377.2004.tb05947.x
61. Wright F. *Mentoring Clinical Champions: Teaching Evidence-Based Practice in Nursing* (2012)
62. Heckhausen H, Gollwitzer PM. **Thought contents and cognitive functioning in motivational versus volitional states of mind**. *Motivat Emot.* (1987) **11** 101-20. DOI: 10.1007/BF00992338
63. Green B, Oeppen RS, Smith DW, Brennan PA. **Challenging hierarchy in healthcare teams—ways to flatten gradients to improve teamwork and patient care**. *Br J Oral Maxillofac Surg.* (2017) **55** 449-53. DOI: 10.1016/j.bjoms.2017.02.010
64. Bell AE, Meyer HS, Maggio LA. **Getting better together: a website review of peer coaching initiatives for medical educators**. *Teach Learn Med.* (2020) **32** 53-60. DOI: 10.1080/10401334.2019.1614448
65. Devine M, Meyers R, Houssemand C. **How can coaching make a positive impact within educational settings?**. *Procedia Soc Behav Sci.* (2013) **93** 1382-9. DOI: 10.1016/j.sbspro.2013.10.048
66. Lyon AR, Cook CR, Brown EC, Locke J, Davis C, Ehrhart M. **Assessing organizational implementation context in the education sector: Confirmatory factor analysis of measures of implementation leadership, climate, and citizenship**. *Implement Sci.* (2018) **13** 5. DOI: 10.1186/s13012-017-0705-6
67. Dearing JW, Cox JG. **Diffusion of innovations theory, principles, and practice**. *Health Aff.* (2018) **37** 183-90. DOI: 10.1377/hlthaff.2017.1104
68. Woo K, Milworm G, Dowding D. **Characteristics of quality improvement champions in nursing homes: a systematic review with implications for evidence-based practice**. *Worldviews Evid Based Nurs.* (2017) **14** 440-6. DOI: 10.1111/wvn.12262
69. Hagedorn HJ, Wisdom JP, Gerould H, Pinsker E, Brown R, Dawes M. **Implementing alcohol use disorder pharmacotherapy in primary care settings: a qualitative analysis of provider-identified barriers and impact on implementation outcomes**. *Addict Sci Clin Pract.* (2019) **14** 24. DOI: 10.1186/s13722-019-0151-7
70. Hwang N, Fitzpatrick B. **Student–Teacher Gender Matching and Academic Achievement**. *AERA Open.* (2021) **7** 1-11. DOI: 10.1177/23328584211040058
71. Yarnell LM, Bohrnstedt GW. **Student-teacher racial match and its association with black student achievement: An exploration using multilevel structural equation modeling**. *Am Educ Res J.* (2017) **55** 287-324. DOI: 10.3102/0002831217734804
72. Valente TW, Pumpuang P. **Identifying opinion leaders to promote behavior change**. *Health Educ Behav.* (2007) **34** 881-96. DOI: 10.1177/1090198106297855
73. von Thiele Schwarz U, Aarons GA, Hasson H. **The Value Equation: Three complementary propositions for reconciling fidelity and adaptation in evidence-based practice implementation**. *BMC Health Serv Res.* (2019) **19** 868. DOI: 10.1186/s12913-019-4668-y
74. Jacobs SR, Weiner BJ, Bunger AC. **Context matters: measuring implementation climate among individuals and groups**. *Implement Sci.* (2014) **9** 46. DOI: 10.1186/1748-5908-9-46
75. Richter A, Schwarz UT, Lornudd C, Lundmark R, Mosson R, Hasson H. **iLead-A transformational leadership intervention to train healthcare managers' implementation leadership**. *Implement Sci.* (2016) **11** 108. DOI: 10.1186/s13012-016-0475-6
76. Eslava-Schmalbach J, Garzón-Orjuela N, Elias V, Reveiz L, Tran N, Langlois EV. **Conceptual framework of equity-focused implementation research for health programs (EquIR)**. *Int J Equity Health* (2019) **18** 80. DOI: 10.1186/s12939-019-0984-4
77. Baumann AA, Cabassa LJ. **Reframing implementation science to address inequities in healthcare delivery**. *BMC Health Serv Res.* (2020) **20** 1-9. DOI: 10.1186/s12913-020-4975-3
78. Murray LK, Skavenski S, Bass J, Wilcox H, Bolton P, Imasiku M. **Implementing evidence-based mental health care in low-resource settings: a focus on safety planning procedures**. *J Cogn Psychother.* (2014) **28** 168-85. DOI: 10.1891/0889-8391.28.3.168
79. Yapa HM, Bärnighausen T. **Implementation science in resource-poor countries and communities**. *Implement Sci.* (2018) **13** 154. DOI: 10.1186/s13012-018-0847-1
80. Gaias LM, Arnold KT, Liu FF, Pullmann MD, Duong MT, Lyon AR. **Adapting strategies to promote implementation reach and equity (ASPIRE) in school mental health services**. *Psychol Schools.* (2021). DOI: 10.1002/pits.22515
81. Okpala P. **Addressing power dynamics in interprofessional health care teams**. *Int J Healthcare Manage.* (2021) **14** 1326-32. DOI: 10.1080/20479700.2020.1758894
82. Hall ET. *Beyond Culture* (1976)
83. Garrison C. **The Lonely Only: Physician reflections on race, bias, and residency program leadership**. *Fam Med.* (2019) **51** 59-60. DOI: 10.22454/FamMed.2019.339526
84. Scott KA, Brown DJ. **Female first, leader second? Gender bias in the encoding of leadership behavior**. *Organ Behav Hum Decis Process.* (2006) **101** 230-42. DOI: 10.1016/j.obhdp.2006.06.002
|
---
title: 'The quick pivot: Capturing real world modifications for the re-implementation
of an early psychosis program transitioning to virtual delivery'
authors:
- Wanda Tempelaar
- Nicole Kozloff
- Allison Crawford
- Aristotle Voineskos
- Don Addington
- Tallan Alexander
- Crystal Baluyut
- Sarah Bromley
- Sandy Brooks
- Lauren de Freitas
- Seharish Jindani
- Anne Kirvan
- Andrea Morizio
- Alexia Polillo
- Rachel Roby
- Alexandra Sosnowski
- Victoria Villanueva
- Janet Durbin
- Melanie Barwick
journal: Frontiers in Health Services
year: 2023
pmcid: PMC10012808
doi: 10.3389/frhs.2022.995392
license: CC BY 4.0
---
# The quick pivot: Capturing real world modifications for the re-implementation of an early psychosis program transitioning to virtual delivery
## Abstract
### Background
Team-based Early Psychosis Intervention (EPI) services is standard of care for youth with psychosis. The COVID-19 pandemic required most EPI services to mount an unplanned, rapid pivot to virtual delivery, with limited guidance on how to deliver virtual clinical services or whether quality of re-implementation and treatment outcomes would be impacted. We used a structured approach to identify essential modifications for the delivery of core components and explored facilitators and barriers for re-implementation and fidelity of a virtually delivered EPI intervention.
### Materials and methods
NAVIGATE is a structured approach to team-based EPI. It provides detailed modules to guide delivery of core components including medication management, psychoeducation and psychotherapies, supported employment/education, and family education. Having initially implemented NAVIGATE at the Centre for Addiction and Mental Health (CAMH) in 2017, the EPI service transitioned to virtual delivery amid the COVID pandemic. Using a practice profile developed to support implementation, we detailed how core components of NAVIGATE were rapidly modified for virtual delivery as reported in structured group meetings with clinicians. The Framework for Reporting Adaptations and Modifications for Evidence-Based Interventions (FRAME) was used to describe modifications. Fidelity to the EPI standards of care was assessed by the First Episode Psychosis Fidelity Scale (FEPS-FS). Re-implementation barriers and facilitators and subsequent mitigation strategies were explored using structured clinician interviews guided by the Consolidated Framework for Implementation Research (CFIR).
### Results
Identified modifications related to the intervention process, context, and training. We identified contextual factors affecting the re-implementation of virtually delivered NAVIGATE and then documented mitigating strategies that addressed these barriers. Findings can inform the implementation of virtual EPI services elsewhere, including guidance on processes, training and technology, and approaches to providing care virtually.
### Discussion
This study identified modifications, impacts and mitigations to barriers emerging from rapid, unplanned virtual delivery of EPI services. These findings can support delivery of high-quality virtual services to youth with psychosis when virtual care is indicated.
## Introduction
Early psychosis intervention (EPI) is an evidence-based treatment that has become the standard of care for youth with psychosis [1]. EPI care is provided by a multidisciplinary team who provide comprehensive treatment including psychoeducation and psychotherapy for psychosis (most commonly, cognitive behavioural therapy), case management, individual psychopharmacological intervention, family education and support, and support for education and employment [1, 2]. Previous EPI effectiveness studies demonstrated superior outcomes including reduced mortality, decreased risk of relapse, fewer hospital readmissions, and increased employment rates relative to care as usual (3–7). Furthermore, evidence shows that a manualized package of EPI services called NAVIGATE results in improved functional outcomes compared to care as usual. Clients receiving NAVIGATE showed greater improvement in quality of life and psychopathology, greater involvement in work and school, and remained in treatment longer compared to clients receiving community care [2].
EPI models of care, such as NAVIGATE, are designed for in-person delivery, emphasizing frequent contacts and community outreach. The COVID-19 pandemic prompted an abrupt shift to virtual delivery of EPI care to ensure continuity in the face of public health restrictions (8–10). However, little was known about the modifications required to provide EPI care virtually or their impacts. The abrupt need for virtual care delivery without suspending service meant there was no time for planning or training to prepare for this shift. Clinicians and clients had to quickly adapt to a new delivery method with ongoing adjustments occurring over time.
The impacts of these modifications and whether virtual delivery of EPI care would achieve the same benefits as the in-person intervention were unknown. With the shift to virtual delivery, it is important to better understand the nature of the modifications that are made and their impact on treatment delivery and outcomes. Modifications, especially if unplanned, may or may not align with the core components required to ensure the intervention is effective [11]. For instance, modifications that alter or remove core components of the EPI model, or fail to align with population needs may reduce the effectiveness of virtual EPI compared to the original, in-person intervention [11, 12].
Previous work on investigating modifications of evidence-based interventions led to the development of frameworks that can be used to systematically describe and evaluate modifications to evidence-based interventions, including the Framework for Modification and Adaptations [11, 13]. The FRAME captures characteristics of modifications and was recently updated to include broader aspects of the implementation process, such as reasons for the modifications (e.g., to improve feasibility, engagement, outcome), level of the modifications (client, clinician, program), timing of the modifications (prior, during, for scale up), and fidelity to the original intervention (consistent or inconsistent) [11]. This detailed framework facilitates understanding of the relationships between the modification and key outcomes that can be tested in implementation studies [14, 15]. This is important because modifications that remove or alter core components of an intervention may be less effective. Despite significant developments to identify and classify modifications and their impact on outcomes using structured frameworks, there is little guidance on how to systematically document (ad hoc) modifications in a dynamic setting, how to assess the impacts of these modifications over time, and how contextual factors relate to modifications and outcomes.
The Centre for Addiction and Mental Health (CAMH) in Toronto, Ontario, is home to the largest EPI program in Canada, providing assessment and ongoing services to people aged 14–29 years who present with early psychosis. CAMH implemented the NAVIGATE model for EPI service delivery in 2017 for all clients attending the EPI outpatient clinic, and is currently leading a multisite implementation effectiveness study of NAVIGATE across EPI programs in the province of Ontario [16]. NAVIGATE is expected to increase consistency of delivery and improve program fidelity to EPI practice standards [16]. CAMH has a dedicated Virtual Mental Health and Outreach program that provides telepsychiatry to clients in remote and rural areas. During the COVID-19 pandemic, this program expanded to support other CAMH programs in their delivery of virtual care.
Soon after the onset of the pandemic, CAMH's Slaight Centre Early Intervention Service (SCEIS) was awarded COVID-19-related research funding to investigate the re-implementation of NAVIGATE from in-person to virtual delivery. The aims of this study are [1] to identify the modifications required to re-implement and deliver the NAVIGATE model virtually, [2] to assess whether these modifications affected fidelity to the EPI practice standards, [3] to explore implementation facilitators and barriers related to re-implementation, a term coined here to reflect a second implementation effort following the earlier, full implementation of an intervention, [4] to examine satisfaction with virtual delivery of NAVIGATE among clients, family members and clinicians, and [5] to investigate service engagement with virtual delivery of NAVIGATE. To address these aims, we conducted a mixed methods study using a convergent study design to investigate the unplanned shift to virtual delivery of EPI [17]. The current manuscript addresses aims 1, 2 and 3, and illustrates the application and utility of a practice profile [18], the FRAME framework for identifying and documenting model adaptations and unanticipated impacts [11, 13], and the Consolidated Framework for Implementation Research (CFIR) for identifying barriers to re-implementation [19]. Objectives 4 and 5 related to outcomes will be reported separately.
## Design
The study used a mixed methods, pragmatic, implementation and evaluation design described in more detail elsewhere [20]. Youth and family members with lived experience, front-line clinicians, and clinical administrators were engaged in a structured, stepwise approach to track adaptations needed to provide NAVIGATE care virtually. Structured approaches were used to evaluate re-implementation outcomes as measured by fidelity, and to explore implementation facilitators and barriers. Throughout this manuscript we refer to “virtual” delivery of care when care is provided via phone or tele/videoconference.
## Study setting and population
This study was conducted at SCEIS, the outpatient EPI program at the Centre for Addiction and Mental Health (CAMH) in Toronto, Canada. SCEIS serves people aged 14–29 years old who present with early psychosis (schizophrenia, schizoaffective disorder, schizophreniform disorder, bipolar I disorder or major depressive disorder with psychotic features, substance-induced psychotic disorder, unspecified psychotic disorder). Located in downtown Toronto, Canada, this EPI service is staffed by approximately 40 clinicians who assess approximately 600 new clients annually.
The Ontario Ministry of Health provides coverage for all medically necessary services including EPI to residents through the Ontario Health Insurance Plan (OHIP), and this coverage was maintained in the transition to virtual care.
SCEIS provides EPI services according to the NAVIGATE model, a highly structured program of coordinated specialty care with clearly defined roles for staff [21]. Initially implemented at CAMH in late 2017, the model consists of four core clinical roles: Individual Resiliency Training (IRT), Supported Employment and Education (SEE), Family Education Program (FEP), and individualized medication management [21]. Additional core components that are fundamental to the NAVIGATE program include: Team Lead who facilitates monitoring; Practice Feedback and Training; and Caseloads small enough to allow for the intensity and frequency of required contact. Manualized protocols are used to operationalize current EPI standards, and all clients are systematically offered all treatment components with regular team meetings to review client progress, fidelity, and need for adjustments. All clients receive substance use support as part of the IRT manual. Where there is additional need for substance use support beyond the general manual, clients can receive specialized support from a clinical psychologist at SCEIS or from additional programs at the substance use disorder services at CAMH.
## Stakeholders
This study engaged youth and family members with lived experience, front-line clinicians and administrators according to current best practices [22]. Stakeholders contributed meaningfully to the study design, data collection, integration of findings and knowledge dissemination. We held monthly meetings with the principal investigators, operational research staff, youth and family members with lived experience, front-line clinicians and clinical leads (“steering committee”) to review the progress of re-implementation and data collection, and to plan for knowledge dissemination. Monthly “knowledge user meetings” were held during the first phase of the study with front-line clinicians and clinical leads to discuss program modifications and their impacts, barriers to virtual care delivery, clinician resources and training.
## Context: the COVID-19 pandemic
The shift to virtual delivery of care occurred abruptly in March 2020 due to COVID-19-related public health directives to stay at home during the first COVID-19 lockdown in Toronto. The first COVID-19 lockdown lasted from March to June 2020 (with ongoing restrictions persisting to varying degrees until the time of submission) and prompted a hospital-wide transition to virtual delivery for most outpatient services. Exceptions were made to allow in-person appointments for a small number of clients for whom virtual assessment and treatment was not feasible (e.g., clients in crisis and/or requiring a hospital admission, those receiving intramuscular injections, and/or those lacking access to virtual care).
The abrupt shift in the modality of care delivery pre-empted any preparation and planning for this transition. Fortuitously, several facilitating events occurred. Prior to March 2020, CAMH had taken steps towards integrating a digital platform to enhance capability for virtual meetings and enable the use of virtual care throughout the organization. After an extensive process, a digital platform (Cisco Webex) was chosen that met the Ministry of Health's privacy and confidentiality requirements including safeguarding Personal Health Information of clients. Proof-of-concepts in clinical and non-clinical settings had been conducted with this digital platform prior to the COVID-19 pandemic [23]. Other enabling factors at CAMH that predated the pandemic included exclusive use of electronic medical records, and the transition to using laptops instead of desktop computers in order to facilitate remote and mobile work.
Once the pandemic triggered the shift to virtual care, CAMH rapidly scaled the deployment of the Cisco communications platform and initiated organization-wide training for clinicians in the use of Cisco Webex and the Ontario Telemedicine Network (OTN), two provincially approved digital platforms for providing virtual care. This training was provided to over 400 CAMH clinicians.
CAMH developed and implemented a virtual care policy and protocol that covered procedures for providing care in a virtual setting such as privacy, confidentiality, documentation practices, and practical instructions for providing virtual care. Subsequently, the Virtual Mental Health and Outreach team developed digital mental health training for clinicians on delivering virtual care in clinical settings. Training content included the context and evidence base for virtual care; clinical experiences; individual and group settings; safety and confidentiality procedures; technology; and the therapeutic relationship in a virtual setting [24]. Other tools for facilitating virtual care delivery were made available across CAMH including a communications application allowing for instant messaging and phone calls with other team members and clients (Cisco Jabber); a secure file transfer platform to share files; and applications for faxing and scanning documents remotely. CAMH EPI clinicians were provided with mobile phones to facilitate voice communications and text reminders with clients.
Virtual care was enabled across Ontario by a shift in the Ontario Ministry of Health billing codes and requirements to enable remuneration of virtual care (via videoconference or phone) provided by physicians.
## Objective 1: modifications
Our approach to documenting modifications included the use of the NAVIGATE practice profile [18, 25] and the FRAME framework [11]. A practice profile is a tool for describing the core components of an innovation or model of care, including the principles that underlie the model. Core components are prescribed by the innovation developer but how each core component is executed and by whom is determined by the implementing organization to guide implementation and delivery. Core components are the features of a model or intervention that must be present to ensure that it is delivered as intended to achieve expected outcomes. The profile provides a structure for documenting variations to the innovation as well as implementation outcomes. Once an innovation is described in sufficient detail, effective implementation methods can be applied to explore the organizational functions needed, develop staff competencies, monitor data for continuous improvement and sustainment, and ensure that leadership and administrative practices remain facilitative.
Prior to the pandemic, research team members developed a NAVIGATE practice profile [26]. This development took place through an iterative process that included a review of key NAVIGATE manuals and other model documents, published articles from the RAISE-ETP study that developed and first implemented NAVIGATE, as well as feedback from clinicians and implementation specialists familiar with the model [21]. A penultimate draft was reviewed by model originators, further revised and finalized. The final practice profile identified seven core components: Individual Resiliency Training (IRT), Supported Employment and Education (SEE), Family Education Program (FEP), Individualized Medication Management, Team Leadership, Practice Feedback and Training, and Caseload (Figure 1). We used this NAVIGATE practice profile to describe and document modifications for each core component in the current study. We adjusted the descriptions of how the components were delivered virtually and added information on mitigation strategies that were taken to facilitate the change or to reduce potential negative impacts and amplify positive impacts of the modifications. Structured reflection sessions were conducted remotely with clinicians in each NAVIGATE role (IRT, SEE, FEP, prescribers, team lead) during the re-implementation process to document modifications and impacts. At each discussion, we monitored challenges, contextual factors, and impact and tracked subsequent modifications or mitigating strategies. From these discussions we were able to identify the reasons modifications were made and at what level they occurred. This method of tracking modifications in structured reflection sessions has previously shown potential as a straightforward and low-burden approach for documenting events across a dynamic implementation setting [27]. Sessions with the clinicians and the clinical manager occurred at 2–3 and then again at 12 months into the study. Interim updates by clinician representatives were provided as part of monthly meetings throughout the first year of the study and clinicians representing different NAVIGATE roles reviewed and finalized the modifications described in the practice profile. Barriers identified during the initial group sessions were reviewed by the research team to inform new adaptations for enhancing the re-implementation process.
**Figure 1:** *NAVIGATE core components.*
Modifications to the practice profile were then coded using the FRAME to document underlying process, rationale and purpose [11]. For our context of re-implementation, we added an additional factor to capture the “effects” of modifications. We identified potential and realized positive, neutral and/or negative intended/unintended effects of modifications and described mitigating strategies that were undertaken to lessen negative impacts, if applicable. To document the “reasons” underlying each modification, we added the specifier “COVID-19 pandemic” as the “outer setting context” to indicate why the modification was made. Documenting modifications in response to culture was not applicable to our context, as modifications were not related to the implementation of the intervention in cultures different from where the intervention was first implemented.
## Objective 2: fidelity
Implementation fidelity refers to the extent to which an intervention is delivered as intended by the program developers and in line with the program model [28]. In the present study, we used the First Episode Psychosis Services Fidelity Scale (FEPS-FS) to assess fidelity to evidence-based practices for EPI delivery [29].
## Fidelity to EPI practices
The FEPS-FS is a validated measure of fidelity to the standards of EPI care [29]. Scale development was based on a review of evidence combined with an expert consensus process and is not tied to any specific model of EPI delivery. Thirty-three items are rated on a 5-point scale from “not implemented” to “fully implemented.” A rating of 4 is considered satisfactory adherence. The scale is designed such that the items measure delivery in relation to the core components of the EPI model (adherence); quality of delivery using strategies such as clinician observation is not assessed [30].
The FEPS-FS items assess team structure (integrated approach), client continuity of care (early intervention, retention), and client receipt of medical and psychosocial treatments (comprehensive care). In Ontario, a community of practice for EPI programs, the Early Psychosis Intervention Ontario Network (EPION), developed a process to assess fidelity with this scale using a site visit methodology [26, 31]. Fidelity ratings are based on interviews with staff, client chart review and administrative data and are usually made after a 1-to-2-day site visit by independent assessors. In this study, COVID-19 related restrictions required us to assess fidelity remotely via phone/video staff interviews and virtually trained on-site health record abstractors [32].
Fidelity assessments were done twice; retrospectively to capture practice prior to pandemic restrictions, when care was provided in-person (January–December 2019), and after the shift to virtual care delivery (July 2020–June 2021). For each assessment period, 10 client charts were randomly selected for clients enrolled in the program for at least one year during that period. These charts were abstracted by remotely trained on-site staff. Two independent fidelity assessors conducted phone/video interviews with program informants about NAVIGATE delivery during each of these periods. Both at the beginning and throughout each interview, the assessors reminded the participant about the practice period in question. For each period, interviews were held with the team lead, prescriber and 4 clinicians in different NAVIGATE roles. The assessors then reviewed the chart, interview and program administrative data to develop preliminary ratings that were discussed in a consensus meeting with a fidelity expert and then finalized.
Individual item ratings and the total mean score were reported for each period. Item ratings were grouped into one of five domains that pertain to: team structure, access and continuity, comprehensive assessment, medical treatments and psychosocial treatments.
## Objective 3: implementation facilitators and barriers
Facilitators and barriers were captured with a CFIR informed semi-structured interview. The CFIR is a determinant framework of 39 factors known to influence implementation, categorized into five major domains: intervention characteristics; outer setting; inner setting; staff characteristics; and implementation process [30]. Since CAMH clinicians had previously implemented NAVIGATE, the CFIR-informed interview focused specifically on the re-implementation of virtual delivery. We included 38 CFIR constructs, omitting cost as the delivery was part of standard care. We interviewed 8 clinicians (IRT, SEE, FEP, prescriber, team lead) by videoconference. Interviews were administered and coded deductively using a variation of the Rapid Analysis (RA) method, an alternative to in-depth analysis of interview data that allows for faster analysis and dissemination of implementation findings while using fewer resources [19, 33]. Coding identified facilitators and barriers as well as the direction (valence) and strength of the association between factors and implementation success. For the first analytic step of the RA method, the analysts captured interview comments on a templated summary table in real time. The summary table aligned with the CFIR interview guide (domain and factors). The second analytic step involved assigning a valence rating to each factor to denote a positive or negative influence on implementation (+, neutral, −). Strength of the association was then rated (−2, −1, 0, mixed, +1, +2) and determined by a number of factors, including level of agreement among participants, strength of conviction, and use of concrete examples. In the last analytic step, memos were written to summarize the findings for each factor.
## Cross-cutting modifications
Group meetings with clinicians revealed that three types of modifications needed for the virtual delivery of NAVIGATE were cross-cutting and independent of NAVIGATE core components, while others were unique to a core component. Cross-cutting modifications related to technology, procedures, and training.
Technological modifications included providing hardware and software to clinicians to facilitate remote work (including laptops and mobile phones), and the organization-wide roll-out of Cisco Webex, a digital platform for providing virtual care.
Procedural modifications related to privacy, safety and confidentiality guidelines for virtual care delivery which included obtaining client consent for virtual appointments, Mental Health Act certification procedures (i.e., for involuntary commitment), and changes to physician remuneration for virtual care.
Training modifications involved clinician orientation to new software applications including the digital platform used for virtual care, clinician training on building engagement with clients in the context of virtual care, provided by a youth with lived experience, risk assessment and addressing crisis management with clients in crisis, suggestions for providing trauma-informed care in a virtual setting, and considering health equity in virtual care delivery. Several of the cross-cutting modifications stemmed from decisions made at the organizational level and impacted the whole organization. For instance, changes made to the remuneration for provision of virtual care, a particularly relevant decision, was made at the provincial governmental level (Ministry of Health).
## Core component related modifications
We documented 26 modifications related to the four NAVIGATE core clinical roles: 8 modifications for IRT, 5 for SEE, 4 for FEP and 9 for the prescriber role (Tables 1a–1f). Most of these modifications occurred during the onset of the shift to virtual care delivery. About two-thirds of the modifications were unplanned or reactive modifications. Most modifications were made by clinicians and/or the clinical manager and occurred at the clinic/unit level ($59\%$), although one-third occurred across the organization ($31\%$) (Table 1a). Most modifications were unrelated to the content of the intervention ($69\%$) and were consistent to the provincial standards for EPI care ($63\%$). Overall, modifications served to increase or maintain client engagement ($34\%$) and to increase and maintain client retention ($28\%$) and improve feasibility of delivery ($19\%$). Little changes were noted for the three NAVIGATE core components that were not directly related to a clinician role. The team leadership role continued as before the shift to virtual care, though all meetings were held virtually, including supervision and training. One of the functions for the team leadership role captured in the practice profile is community outreach, which includes providing targeted education to health, social service, or community groups. There were few community outreach activities, even before the COVID-19 pandemic, and this did not increase with the switch to virtual care delivery. Regarding training and practice feedback, no significant changes were noted to the onboarding process, other than a modality switch to virtual meetings and adding training on virtual delivery of care. The team meetings continued without changes in a virtual setting, and clinicians met virtually with the clinical lead or substitute weekly. With the switch to virtual care delivery there was an increased demand for training on how to use the virtual applications. Caseload size did not change, though workload increased, and mitigating strategies for the increased workload were captured in the Practice Profile.
## Modifications for individual resiliency training
Modifications to individual resiliency training (IRT) components occurred early during the shift to virtual care delivery and were largely unplanned and reactive to the shift to virtual delivery (Table 1b). Most decisions about modifications were made by the treatment team and the clinical manager, were fidelity consistent, and served to maintain client engagement or retention. For example, clients were offered shorter but more frequent appointments, appointment reminders were sent more often, and hardcopy worksheets and handouts from the NAVIGATE modules were replaced with fillable PDF files that could be shared with clients on the video screen during appointments. Modifications were intended to maintain delivery of the IRT core components despite restrictions to in-person practice. One advantage mentioned by the IRT clinicians was that they were able to gain insights into client's living situations when they attended via videoconference from home. Disadvantages of providing IRT care via videoconference or phone were a less fulsome assessment of nonverbal cues, and client and clinician challenges with technology, connectivity, and engagement during appointments. An increased workload for IRT clinicians occurred, partly due to training demands, but also related to increased communication with clients and clinicians (e.g., via email).
**Table 1b**
| FRAME Elements | Brief report from virtual NAVIGATE IRT modifications |
| --- | --- |
| Process | Process |
| When did the modification occur? | IRT modifications resulted from the transition to virtual care due to the COVID-19 pandemic in March/April 2020. These changes occurred during the maintenance/sustainment phase of NAVIGATE delivery in order to continue to deliver care to clients throughout the pandemic by updating the mode of delivery (i.e., via phone/videoconferencing, and eventually adjusting back to in-person appointments as the provincial mandates permitted). |
| Were adaptations planned? | Modifications were primarily unplanned and reactive, resulting from the sudden onset of the pandemic. For instance, clients were offered extra appointments that were shorter in duration as well as more appointment reminders via email or SMS text messaging if needed. Planned/reactive iterative adaptations involved updating NAVIGATE materials and modules into PDF fillable files to share synchronously virtually, and training sessions provided to clinicians. |
| Who participated in the decision to modify? | The SCEIS program leader/clinical manager made most of the decisions on a clinic/unit level. Many partners contributed to decisions and were involved in making modifications relevant to the IRT role including members of the “virtual-NAVIGATE study team” SCEIS staff such as individual practitioners. Certain decisions around virtual care provision were taken on a hospital-wide or provincial level, involving administrators, and CAMH management. |
| Adaptations | Adaptations |
| What was modified? | The process of delivering NAVIGATE was modified so that the continuity of care could be maintained safely in a virtual context in response to provincial mandates. This included providing staff with work cell phones to text and/or call clients, sharing materials via screen sharing instead of face-to-face, and involving the family member in the IRT session to improve access and activation. Training was offered to SCEIS clinicians on a wide range of virtual care topics (e.g., the technical aspects of using the virtual platform as well as addressing building engagement and ensuring privacy in a virtual setting). |
| At what level of delivery were modifications made? | The majority of modifications were made within the clinic/unit level at SCEIS. Some modifications made for the target intervention group included modifying material so that it could be shared with clients across EPI sites electronically. |
| What was the type or nature of context or content-level modifications? | Format changes pertained to transitioning from in-person appointments to delivering care virtually, making adjustments to virtual appointments that warranted in-person care, and altering the amount and length of appointments (i.e., extra appointments that were shorter in duration). Contextual changes included alterations in setting which changed from in-person (at SCEIS) to clients’ homes. Process changes involved sending more appointment reminders via email and text, with the ability to attach documents to Webex invites. Content modifications centred on modifying materials (e.g., creating fillable PDF files) as well as creating web-based resources to support the virtual delivery of NAVIGATE. |
| What was the relationship to fidelity? | The majority of modifications were fidelity consistent, as efforts were made to critically consider and preserve the core elements of the IRT role while making the necessary adjustments to continue delivery of care. |
| Rationale | Rationale |
| a. What was the goal? b. What were the reasons? | a. Modifications to the IRT role in order to deliver NAVIGATE virtually aimed to increase/ maintain client engagement, retention, and satisfaction as well as to improve feasibility.b. Reasons for modifying NAVIGATE to be delivered virtually largely centred around the outer setting context, namely, the pandemic. There were no specific organizational/setting, provider, or recipient reasons for this transition. |
| Outcome | Outcome |
| a. What were the positive outcomes? b. What were the negative outcomes? | a. Continuity of care could be maintained via phone (including texting) for those who do not have access to devices and/or with connectivity issues; less perceived stigma for clients by not having to come on-site (which can increase attendance); greater insight into client's living situations; less formal appointments which can enhance engagement; more joint appointments/”warm handovers” with other care providers; reduced length of appointments increased attention compared to longer virtual appointments and facilitated brief check-ins of clients’ symptoms while improving time-management for clinicians; improved fit to a virtual context and for the target population at SCEIS; increased collaboration between clients and clinical providers.b. Less fulsome assessments of mental health status/nonverbal cues and safety (especially when connecting via phone); unable to support clients going to the emergency department for crisis services; challenges with building/maintaining the therapeutic relationship; COVID-exposure risks for staff and clients who needed to come on-site; less boundaries and appropriate behaviour when meeting virtually; issues with technology and connectivity (which could lead to less time to connect); privacy issues; client mistrust of technology; increased clinician workload and appointments; less time for IRT and more focus on case management tasks. |
To mitigate challenges introduced by modifications to IRT, the research team gathered web-based resources (websites, brief videos, mobile phone apps) related to the content of the IRT NAVIGATE modules to enhance client engagement in the virtual IRT sessions. These resources were selected by IRT clinicians and youth with lived experience and shared among IRT clinicians. To mitigate technological challenges, IRT clinicians connected with clients via phone to guide them on how to use the digital platform or encouraged clients to seek digital support from family members. To lessen the burden of training demands on clinicians, the team disseminated weekly, bite-sized information by email with practical tips on technology and procedures related to virtual care delivery and clinician wellness.
## Modifications for support for education and employment
Modifications made to the Support for Education and Employment (SEE) component were all unplanned (Table 1c). Modification decisions were mostly taken by the treatment team with some input from the clinical manager and individual clinicians. Most modifications were consistent with fidelity, with the exception of a reduction in clinician visits to community-based education and employment settings. SEE modifications included changes to how SEE clinicians were introduced to clients during IRT sessions, fewer opportunities for community outreach visits due to COVID-19 related restrictions, countered by more opportunities to organize and attend virtual meetings with specialized and local supports at educational institutions (e.g., joint meetings with school counsellors). There was also a shift to focus on skills for participating in remote job interviews and learning strategies for remote schooling.
**Table 1c**
| FRAME Elements | Brief report from virtual NAVIGATE SEE modifications |
| --- | --- |
| Process | Process |
| When did the modification occur? | Similar to IRT, the SEE modifications resulted from the transition to virtual care due to the COVID-19 pandemic, see above. |
| Were adaptations planned? | All modifications were unplanned and reactive, resulting from the sudden onset of the pandemic. For instance, SEE workers could no longer introduce themselves during in-person appointments with the IRT clinician, which instead transitioned to IRT clinicians offering SEE support during IRT sessions and following up with SEE clinicians if the client was interested. |
| Who participated in the decision to modify? | The SCEIS treatment/intervention team made most of the decisions on a clinic/unit level. The SCEIS program leader/clinical manager and individual SCEIS practitioners also participated in the decision to add additional appointments to get to know clients and establish a therapeutic relationship. |
| Adaptations | Adaptations |
| What was modified? | The context and process of providing supportive employment and education was modified. This included conducting fewer outreach community visits and using phone calls as a reminder when clients did not show for an appointment. These phone calls typically resulted in phone appointments. |
| At what level of delivery were modifications made? | All of the modifications to the SEE role were made within the clinic/unit level at SCEIS. |
| What was the type or nature of context or content-level modifications? | Contextual process changes reflected less outreach community visits compared to in-person care resulting from provincial mandates for lockdown and closures. |
| What was the relationship to fidelity? | Most modifications were fidelity consistent. SEE clinicians were not able to do community visits due to COVID-19 restrictions, which is inconsistent with fidelity. |
| Rationale | Rationale |
| a. What was the goal? b. What were the reasons? | a. Modifications to the SEE role aimed to increase/maintain client engagement and retention as well as to improve feasibility.b. Reasons largely centred around the outer setting context, namely, the pandemic. Provider reasons for using additional appointments centred on clinical judgement. There were no specific organizational/setting or recipient reasons for this transition. |
| Outcome | Outcome |
| a. What were the positive outcomes? b. What were the negative outcomes? | a. Meeting virtually allowed for more opportunities to conduct joint appointments (i.e., related to school, employment, counselling) which reduced barriers/increased access for clients to attend SEE sessions (e.g., less travel time). Continuity of care was maintained, especially via phone appointments which were sometimes particularly convenient, and an increase in appointment attendance was observed.b. SEE workers were less able to facilitate connections with employers and counsellors as well as conduct in-person outreach visits or casually drop in, requiring more planning and effort from the client (which may be problematic for job development). There were less opportunities for competitive jobs during pandemic, resulting in more work identifying which jobs were not currently experiencing a hiring freeze. |
## Modifications for family education and support
As with the IRT component, the majority of modifications affecting provision of family education and support (FEP) were mostly unplanned (Table 1d). Modification decisions were mostly made by the clinical team. Planned adaptations included the creation of additional material (e.g., PowerPoint presentation) to support virtual delivery of psychoeducation groups. As with SEE, the process by which FEP clinicians connected with families was adjusted. Advantages of virtual FEP delivery included increased access to care meetings for caregivers and family members. Some family members and caregivers experienced barriers to using the digital platform and/or internet. Also challenging was how best to facilitate effective communication in a virtual group meeting. To address this, FEP clinicians developed and shared a structure for group meetings with all attendees and offered individual appointments as needed.
**Table 1d**
| FRAME Elements | Brief report from virtual NAVIGATE FEP modifications |
| --- | --- |
| Process | Process |
| When did the modification occur? | Similar to IRT, the FEP modifications resulted from the transition to virtual care due to the COVID-19 pandemic, see above. |
| Were adaptations planned? | The majority of modifications were unplanned and reactive. For instance, family clinicians were no longer able to join the initial or other in-person appointments to introduce themselves, and similar to the SEE role, had to instead connect with the IRT clinician to determine if they have the client's consent to connect with their family members. Planned/ reactive iterative adaptations reflected additional material developed to support virtual psychoeducation groups (i.e., creating a PowerPoint presentation to share on-screen synchronously, and then sent to family members at the end of the meeting). |
| Who participated in the decision to modify? | The SCEIS treatment/intervention team along with the SCEIS program leader/clinical manager made most of the decisions on a clinic/unit level. Individual SCEIS practitioners also participated in the decision to create material to support care being delivered virtually. |
| Adaptations | Adaptations |
| What was modified? | The context and process of providing family support was modified. This included delivering more NAVIGATE content via phone and offering more videoconferencing groups compared to in-person groups, resulting in more loved ones attending virtually compared to in-person. Structure was also added to virtual groups to facilitate organized communication (using the chat function and “raise hand” function to structure comments and questions). |
| At what level of delivery were modifications made? | All of the modifications to the family clinician role were made within the clinic/unit level at SCEIS. |
| What was the type or nature of context or content-level modifications? | Contextual format changes reflected added virtual groups and the development of virtual material. Contextual process modifications included how the family clinician would connect with the client and their family members during initial and subsequent visits compared to in-person care. |
| What was the relationship to fidelity? | Most modifications were fidelity consistent. |
| Rationale | Rationale |
| a. What was the goal? b. What were the reasons? | a. Modifications to the family clinician role aimed to increase/maintain client engagement and retention.b. Reasons largely centred on the outer setting context, namely, the pandemic. There were no specific organizational/setting, provider or recipient reasons for this transition. |
| Outcome | Outcome |
| a. What were the positive outcomes b. What were the negative outcomes? | a. Meeting virtually allowed for a reduction of barriers (time, commuting) and flexibility in attending psychoeducation groups and facilitated balancing other commitments such as working remotely for family members. This led to an increase in group attendance. Phone appointments were particularly convenient for one-on-one sessions.b. For family clinicians, it is harder to connect with all family members virtually in a group session compared to in-person. Other group session challenges included communication procedures (i.e., asking questions, time allotted for each person to speak, managing interruptions, etc.). Some older family members experienced a technology learning curve which was a barrier at the time. Family members also expressed reduced abilities to speak candidly virtually, especially when their loved one (the client) was at home. |
## Modifications for prescriber
Prescribers were unable to conduct certain activities in a virtual setting as compared to in-person care (Table 1e). This included physical assessments which were postponed early in the pandemic, e.g., monitoring of weight and blood pressure, and assessment of antipsychotic-related movement side effects. To mitigate these challenges and maintain adherence to clinical guidelines, prescribers leveraged community-based resources more often (using local laboratories for bloodwork and community nursing clinics for medication injections).
**Table 1e**
| FRAME Elements | Brief report from virtual NAVIGATE Prescriber modifications |
| --- | --- |
| Process | Process |
| When did the modification occur? | Similar to IRT, the prescriber modifications resulted from the transition to virtual care due to the COVID-19 pandemic, see above. |
| Were adaptations planned? | The majority of modifications were unplanned and reactive. For instance, more time was needed for administrative work (e.g., faxing/calling-in prescriptions and ordering bloodwork), which limited time spent with the client and typically resulted in additional appointments. Planned/reactive iterative adaptations reflected updates to Mental Health Act (MHA) assessment procedures (i.e., the process of filling out and sending original documentation) and not making significant medication changes (in particular, clozapine) to avoid admissions and intensive follow-up during the first year of the pandemic before vaccines were available. |
| Who participated in the decision to modify? | CAMH leadership made most of the decisions on an organizational level, including to where clients could do their bloodwork, which changed from on-site at CAMH prior to the pandemic, to clients’ local labs post-March 2020. This often resulted in delayed and decreased compliancy to standardized bloodwork follow-up. Prescribers were also no longer able to conduct a fulsome physical assessment virtually. Individual SCEIS practitioners and the treatment team also participated in the decision to use additional appointments to get to know clients and build rapport virtually and to leverage community resources more often to administer injections. |
| Adaptations | Adaptations |
| What was modified? | The context and process of the prescriber role was modified. This included increasing the frequency of appointments initially during the start of the pandemic and using additional appointments to develop fulsome impressions. Clients were no longer able to complete bloodwork at CAMH at the time of their appointment, in-person self-report questionnaires and physical assessment of side-effects were conducted less frequently. |
| At what level of delivery were modifications made? | Modifications to the prescriber role were made primarily across the CAMH organization as a whole. Some modifications also were made at the clinic/unit level and the individual practitioner level at SCEIS. |
| What was the type or nature of context or content-level modifications? | The contextual process was modified for conducting fulsome physical assessments including bloodwork on-site, assessment of side-effects, and administering self-report questionnaires, which all could no longer continue as a result of the onset of the pandemic. The process for conducting MHA assessments was also modified to a virtual context. Format changes pertained to how appointments were conducted (i.e., videoconference or phone rather than in-person), the use of additional appointments, and allotting added time for increased administrative work. |
| What was the relationship to fidelity? | Roughly half of the prescriber modifications were fidelity consistent. Core elements of the prescriber role that were impacted included fulsome physical assessments, medication changes, and conducting bloodwork on-site at CAMH, which are modifications that are inconsistent with fidelity. |
| Rationale | Rationale |
| a. What was the goal? b. What were the reasons? | a. Modifications made to the prescriber role aimed to improve feasibility as well as increase/maintain client engagement and retention.b. Reasons largely centred on the outer setting context, namely, the pandemic. There were no specific organizational/setting, provider or recipient reasons for this transition. |
| Outcome | Outcome |
| a. What were the positive outcomes b. What were the negative outcomes? | a. Meeting virtually allowed for continuity of care with reduced barriers to attending appointments virtually (i.e., reduced travel time and associated costs, decreased stigma/trauma). Prescribers could also check on medication adherence when calling in prescriptions to the pharmacy. Clients often received their injections locally (i.e., at home or at a clinic close by to them).b. Virtual appointments impede physical examinations with clients and missing important clinical presentations by not being able to read non-verbal cues as accurately. This often led to difficulties in building rapport. Challenges with client attention and boundaries arose virtually (i.e., clients engaging in distracting or less appropriate behaviour such as attending appointments while driving or intoxicated, and having others in the household who may be able to listen). There was an increase in last-minute reschedule requests and no-shows during the pandemic as well as adding additional appointments, often resulting in more time spent connecting with clients. Some clients also experienced connectivity issues. |
## Modifications for the caseload size, team leadership, and training and practice feedback components
Few changes were noted for the three NAVIGATE core components that are not directly related to a clinician role (Table 1f). The team leadership role continued without significant changes, though all meetings were held virtually including supervision and training. Targeted community education decreased, likely related to fewer opportunities for community education as many community events were cancelled/postponed due to the COVID-19 restrictions and educational institutions were busy with the COVID-19 related practicalities including the shift to remote learning with less opportunities for psychoeducation. Training and practice feedback required several changes to the content of the training and practice feedback, e.g., training on virtual care was provided, and practice feedback focussed more on the shift to virtual care delivery and the challenges related to this new method of care delivery. Caseload size did not change, though workload increased significantly for the clinicians and clinical manager due to the added complexity introduced by technology, more frequent appointments, and, anecdotally, improved appointment attendance facilitated by virtual care.
**Table 1f**
| FRAME Elements | Brief report from e-NAVIGATE Caseload Size, Team Leadership, and Training and Practice Feedback components modifications |
| --- | --- |
| Process | Process |
| When did the modification occur? | In general, little modifications occurred to these core components. Similar to the clinician roles, modifications to the Team Leadership, and Training and Practice Feedback resulted from the transition to virtual care due to the COVID-19 pandemic, see above. |
| Were adaptations planned? | The majority of modifications were planned and reactive, such as the additional trainings, e.g., clinician training for improving building engagement with clients in a virtual setting or crisis management. |
| Who participated in the decision to modify? | Most decisions were made with the clinical manager and clinicians. |
| Adaptations | Adaptations |
| What was modified? | Caseload was not modified and continued to be high. Despite little increase in caseload, workload increased. The leadership role was not modified. Training and Practice Feedback noted increase in training early during the pandemic. |
| At what level of delivery were modifications made? | Most modifications were made at the clinic/unit level and the individual practitioner level at SCEIS. |
| What was the type or nature of context or content-level modifications? | There were no changes to the content of the program. |
| What was the relationship to fidelity? | Mostly fidelity consistent. Core elements that were impacted included fulsome physical assessments, medication changes, and conducting bloodwork on-site at CAMH, which are modifications that are inconsistent with fidelity. |
| Rationale | Rationale |
| a. What was the goal? b. What were the reasons? | a. Modifications made to the prescriber role aimed to improve feasibility as well as increase/maintain client engagement and retention.b. Reasons largely centred on the outer setting context, namely, the pandemic. There were no specific organizational/setting, provider or recipient reasons for this transition. |
| Outcome | Outcome |
| a. What were the positive outcomes b. What were the negative outcomes? | a. Meeting virtually allowed for continuity of meetings with reduced barriers to attending appointments virtually (i.e., reduced travel time)b. Virtual appointments impede physical examinations with clients and missing important clinical presentations by not being able to read non-verbal cues as accurately. This often led to difficulties in building rapport. Challenges with client attention and boundaries arose virtually (i.e., clients engaging in distracting or less appropriate behaviour such as attending appointments while driving or intoxicated, and having others in the household who may be able to listen). There was an increase in last-minute reschedule requests and no-shows during the pandemic as well as adding additional appointments, often resulting in more time spent connecting with clients. Some clients also experienced connectivity issues. |
## Fidelity to EPI standards
Table 2 reports item, domain and total fidelity ratings based on the FEPS-FS for two time periods: during 2019, prior to the onset of COVID restrictions and the switch to virtual care delivery, and during 2021, after modifications had been implemented. Of the 33 items in the scale, 4 could not be rated due to lack of data and/or relevance to the Ontario context. For the remaining 29 items, the total mean rating exceeded 4.00 for both time periods, although there were some item level rating differences.
**Table 2**
| Domain | Item | In-person NAVIGATE (Pre- COVID) | Virtual NAVIGATE (during COVID-19 pandemic) |
| --- | --- | --- | --- |
| Structure | Structure | Structure | Structure |
| 2 | Participant/provider ratio | 5.00 | 5.00 |
| 3 | Multidisciplinary team | 5.00 | 5.00 |
| 4 | Assignment of case manager | 5.00 | 5.00 |
| 5 | Psychiatrist caseload | 5.00 | 5.00 |
| 6 | Psychiatrist role on team | 5.00 | 5.00 |
| 7 | Weekly multi-disciplinary team meetings | 5.00 | 5.00 |
| 8 | Explicit diagnostic admission criteria | 5.00 | 5.00 |
| 10 | Duration of FEP program | 4.00 | 4.00 |
| 1 | Practicing team leader | 3.00 | 3.00 |
| | Mean domain score | 4.67 | 4.67 |
| Access and continuity (engagement and retention) | Access and continuity (engagement and retention) | Access and continuity (engagement and retention) | Access and continuity (engagement and retention) |
| 31 | Communication between SCEIS and inpatient services | 5.00 | 5.00 |
| 32 | Timely Contact After Discharge from Hospital | 5.00 | 5.00 |
| 12 | Early Intervention (Inpatient care prior to admission) | 3.00 | 1.00 |
| 13 | Timely contact with referred individual | 3.00 | 5.00 |
| 11 | Targeted community education | 2.00 | 1.00 |
| 28 | Active engagement (community visits) | 1.00 | 1.00 |
| | Mean domain score | 3.17 | 3.00 |
| Assessments | Assessments | Assessments | Assessments |
| 14 | Family involvement in initial assessment | 4.00 | 4.00 |
| 15 | Comprehensive clinical assessment (initial) | 5.00 | 5.00 |
| 16 | Comprehensive psychosocial assessment (initial) | 5.00 | 3.00 |
| 17 | Treatment / Care Plan after initial assessment | 4.00 | 5.00 |
| 25 | Annual formal comprehensive assessment | 5.00 | 5.00 |
| | Mean domain score | 4.60 | 4.40 |
| Medical | Medical | Medical | Medical |
| 18 | Antipsychotic medication prescription | 5.00 | 5.00 |
| 19 | Antipsychotic dosing within recommendations | 5.00 | 5.00 |
| 24 | Supporting Health Management | 5.00 | 5.00 |
| | Mean domain score | 5.00 | 5.00 |
| Psychosocial Treatment | Psychosocial Treatment | Psychosocial Treatment | Psychosocial Treatment |
| 21 | Client psychoeducation | 5.00 | 4.00 |
| 23 | Cognitive behavior therapy (CBT) | 5.00 | 5.00 |
| 26 | Services for Substance Use Disorders | 5.00 | 5.00 |
| 27b | Supported education | 5.00 | 5.00 |
| 30 | Crisis intervention services | 5.00 | 5.00 |
| 27a | Supported Employment | 3.00 | 3.00 |
| | Mean domain score | 4.67 | 4.50 |
| | Mean overall score | 4.38 | 4.28 |
The program structure domain mean score did not change between the traditional in-person and virtual NAVIGATE care delivery and it remained high, at 4.67. The access and continuity domain mean score declined slightly from 3.17 to 3.00. Within this domain, the early intervention item rating decreased from 3.00 to 1.00, indicating an increase in the percentage of clients that were hospitalized prior to entry in the EPI program. The targeted community education item rating also decreased from 2.00 to 1.00, indicating fewer community education sessions were being conducted. The rating for timely contact with referred individual increased with virtual delivery of NAVIGATE care, indicating more clients were seen within 2 weeks of referral. The assessment domain mean score remained high, declining slightly from 4.60 to 4.40 due to lower rating for the initial comprehensive psychosocial assessment item with virtual delivery of NAVIGATE i.e., fewer clients had all components of the comprehensive assessment documented in their consultation note. The medical treatment domain mean and item scores did not change over time and remained high, at 5.00. The psychosocial treatment domain mean score declined slightly but remained high, at 4.50.
## Objective 3: facilitators and barriers
Factors (italicized) affecting virtual EPI delivery are described in Table 3 including their strength and valence. Note that factors were overwhelmingly facilitative, with 10 ($27\%$) showing as mixed. No factors emerged as barriers to re-implementation in this setting and context. Table 3 provides ratings and summaries for each factor.
**Table 3**
| CFIR Domain/Construct | Rating | Summary Statement |
| --- | --- | --- |
| Intervention Characteristics | Intervention Characteristics | Intervention Characteristics |
| Intervention Source | +1 | Clinicians understood that NAVIGATE was developed in the U.S. and that it is intended to provide evidence-based EPI care that is more formalized, standardized and consistent. Some clinicians stated that according to research, standardized care improves outcomes. NAVIGATE was seen as being implemented due to the desire for more organized and coordinated EPI care. |
| Evidence Strength and Quality | Mixed | Clinicians felt that NAVIGATE is effective for clients, largely based on their experiences and observations from other clinicians and clients as well as their overall understanding of NAVIGATE. A few mentioned their knowledge about the research behind NAVIGATE. Regarding their initial perceptions of whether NAVIGATE would work virtually, most clinicians admitted that they were doubtful that it would be as effective as in-person. However, over time, they found that it worked equally as well, with some exceptions such as monitoring side effects which requires face to face interaction. |
| Relative Advantage | +1 | Clinicians saw NAVIGATE as augmenting EPI services to a better alternative to how services were previously delivered. With NAVIGATE there is consistency, standardization, and the entire team is involved in client and family care (previously team was disjointed). The virtual delivery of NAVIGATE provided advantages in several ways including: accessibility (clients able to meet more often), flexibility (particularly around school and work), and cost savings (e.g., transportation). Some disadvantages that clinicians identified included not having a platform for clients to complete scales before meeting with the psychiatrist, inequity issues for clients who did not have access to technology and challenges for clinicians in reading body language for assessment purposes. |
| Adaptability | +2 | Adaptations to ensure that the virtual delivery of NAVIGATE worked included implementing and learning how to use WebEx, providing staff with laptops and phones, and converting the manual into PDF fillable forms. Clinicians felt these adaptations were very effective and “working great”. An issue that has not been resolved is the transfer of clients’ self-rated side-effects that they completed on an iPad while waiting to see the psychiatrist. |
| Trialability | 0 | Clinicians acknowledged that there was no opportunity to try out the adaptations because there was no time. There was no indication of this being problematic or advantageous. |
| Complexity | Mixed | For some clinicians, implementing virtual NAVIGATE was regarded as complex, particularly at the beginning because it had to be done quickly with many details to be worked out (e.g., ensuring confidentiality, privacy) and technology was challenging for some people (e.g., family members). However, for others, it was not “terribly difficult” or much extra work because they could rely on others “to figure it out”. |
| Design Quality and Packaging | +1 | Although a few clinicians felt that the materials and supports were not enough at the beginning (e.g., virtual version of the manual, tip sheets) or too much (lots of documents to read and videos to watch), most clinicians felt that they received helpful guidance, information and support from IT as well as from reflective practice meetings. |
| Cost | Missing | Clinicians could not comment because they were not aware of the costs involved. |
| Outer Setting | Outer Setting | Outer Setting |
| Client Needs and Resources | Mixed | Clinicians’ perceptions of the extent that NAVIGATE meets the needs of clients were mixed. Most clinicians perceived NAVIGATE as being valuable to clients and families based on positive feedback they received, particularly the team approach to care. However, they also noted that for some clients the material was daunting and long, whereas others appreciated the structured approach to their care. Clients with co-morbidities, cultural and language differences and issues accessing the technology were also perceived as barriers to participating in NAVIGATE. To clinicians’ knowledge clients were not consulted on prior to the re-implementation of virtual NAVIGATE. |
| Peer Pressure | 0 | Clinicians were not aware of any other sites implementing NAVIGATE prior to SCEIS. |
| Cosmopolitanism | +1 | Clinicians spoke of networking and collaborating with other EPI clinics via ECHO sessions, which informed their NAVIGATE practice. Affiliations with EPION and connections with other mental health agencies and former places of work also influenced clinicians’ NAVIGATE work. |
| External Policies and Incentives | +2 | Provincial best practices and standards for EPI was seen as a major incentive for the implementation of NAVIGATE. |
| Process | Process | Process |
| Planning | Mixed | General consensus among clinicians was that there was a lack of planning in the move to virtual delivery, which they recognized as unavoidable due to the sudden need to pivot (i.e., pandemic). Hence at the start, the pivot to virtual delivery was overwhelming. However, clinicians felt that the implementation leaders were the appropriate people and that they did their best to make it as easy and smooth as possible. One participant felt that the SEE role did not receive a lot of guidance. At the time of the interview, most clinicians felt that virtual NAVIGATE was fully implemented. |
| Engaging | Engaging | Engaging |
| Opinion Leaders | +1 | Clinicians felt that the key people who were instrumental in pivoting to the virtual delivery of virtual NAVIGATE worked hard and were collaborative in their approach. They noted several strategies leaders used to encourage and inform staff to move to virtual that entailed numerous emails, links to training, meetings and providing opportunities to ask questions as well as encouraging flexibility in the delivery of NAVIGATE. Clients were informed about the changes through email discussions. Clinicians noted that there was no choice but to move to virtual but made concessions for in person appointments when it was possible. |
| Formally Appointed Implementation Leaders | +2 | Although there was not a lot of planning, the leadership was viewed as collaborative and helpful. |
| Champions | +1 | Identified champions included the implementation leader as well as team members and younger staff who helped others who were not as technically advanced. There was little resistance because everyone knew that it was necessary to pivot to virtual delivery. |
| External Change Agents | 0 | Most clinicians could not identify people outside of SCEIS that helped with pivoting to virtual delivery other than the IT department. |
| Executing | Mixed | Clinicians held mixed opinions about the collaborative execution of the implementation. They spoke of the changes as being a “tsunami”. Some felt that their perspective was sought via team “huddles” and problem-solving discussions as well as opportunities to pose questions to the implementation leaders. Others felt that they were “told” about the programs and systems to use and thus it was more instructive than collaborative. |
| Reflecting and Evaluating | Mixed | Some clinicians spoke of receiving feedback about what was working and what was not working, as well as statistics about engagement (no shows, who they were seeing) that included discussions and reflections on the information. Others received informal feedback (i.e., no statistics) and others did not recall receiving any specific feedback about virtual delivery. |
| Inner Setting | Inner Setting | Inner Setting |
| Structural Characteristics | Mixed | CAMH was seen as a large, resource-intense setting and hence staff were provided with laptops, phones and rooms for private meetings with clients (virtually as well as in-person) that positively impacted the move to virtual delivery. Areas that still needed changes included finding a way for clients to input their information (without compromising confidentiality) to use the modules effectively as well as improvements to the charting system (electronic). |
| Networks and Communications | +2 | Clinicians mentioned that there were multiple and continuous channels of communication via emails, online team meetings, sharing links to resources, updated policies and problem-solving including communication outside of SCEIS with other EPI sites via ECHO. Although the volume of new information and communications was perceived as overwhelming, it was generally recognized that it was necessary in order to support the transition to virtual delivery of NAVIGATE within days. |
| Culture | +2 | Clinicians regarded the culture of SCEIS as highly positive, collaborative, warm, healthy, supportive, client-centered, and acknowledged that it impacted positively the transition to the online delivery of NAVIGATE. Working together as a team and being focused on delivery the highest quality care possible were perceived as key contributors to the success of transition to virtual NAVIGATE. |
| Implementation Climate | Implementation Climate | Implementation Climate |
| Tension for Change | +2 | Clinicians unanimously noted that there was high tension for change for NAVIGATE – in other words, a program like NAVIGATE was highly needed because of it imposed consistency in delivering care, a holistic and standardized approach, multiple roles with clear scope of practice that benefitted various areas of need for clients. |
| Compatibility | Mixed | The extent to which virtual NAVIGATE fitted with the existing structures and workflows was perceived as mixed; overall, the virtual delivery of NAVIGATE was compatible with the existing flows but certain roles noted limitations such as poor linkage between virtual NAVIGATE and the charting system, the function of conducting and including assessments virtually, doing injections and benefitting from administrative support. |
| Relative Priority | +2 | The transition to virtual delivery of NAVIGATE was unambiguously perceived as the main priority by all clinicians. There were no other competing priorities and all clinicians fully dedicated their time and attention to the virtual delivery of NAVIGATE, which contributed to its success. |
| Organizational Incentives and Rewards | +1 | There were several incentives noted for both clients and clinicians; for clients, these included the convenience of accessing care, which increased participation, reduced costs related to parking, transportation and time, increased flexibility. For clinicians, Covid and the urgent need to find a way to deliver care to clients was noted as the main incentive. Many clinicians also mentioned that their efforts were recognized by their manager. |
| Goals and Feedback | 0 | Most clinicians were not aware of any targets set for the virtual delivery of NAVIGATE and this did not appear to influence their performance. Clinicians were aware of the research component of NAVIGATE and participated in focus groups to share their experiences. Some talked about internal team meetings as an opportunity to share feedback on NAVIGATE or its virtual delivery. |
| Learning Climate | +1 | Overall, clinicians perceived SCEIS’ learning climate positively and acknowledged that it was encouraging of learning and taking on new initiatives. Clinicians valued the availability of multiple learning opportunities, both internally and externally and the support for participation in these opportunities. |
| Readiness for Implementation | Readiness for Implementation | Readiness for Implementation |
| Leadership Engagement | +2 | Clinicians unanimously believed that there was support from leadership for the virtual delivery of NAVIGATE and multiple discussions regarding what was needed, special considerations for virtual delivery of care (e.g., privacy; when in-person was essential, role-specific tasks such as who monitors side effects) and that leadership was on board and engaged in the transition process. |
| Available Resources | +1 | Clinicians recognized the availability of many sources of information and supports (e.g., WebEx support, tele-mental health, PSSP, educational services, internal team, etc.) and overall having the resources needed to perform their role successfully. Some clinicians did not receive the original NAVIGATE training and perceived this as a limitation. They valued getting laptops early in the process, which was essential to the virtual transition, but mentioned that access to cell phones was delayed. |
| Access to Knowledge and Information | Mixed | With respect to access to knowledge and information related to e-NAVIGATE, clinicians described mixed feelings and experiences: there was no formal training, time did not allow for this, but there were multiple resources available to support the transition via links, training videos and emails. The amount of information to be accessed, absorbed and implemented in a very short period of time made the initial experience overwhelming for many clinicians. This improved with time. |
| Characteristics of Individual Clinicians | Characteristics of Individual Clinicians | Characteristics of Individual Clinicians |
| Knowledge and Beliefs about the Intervention | +1 | Clinicians regarded the NAVIGATE model positively and valued the evidence base and the holistic approach. With respect to its virtual delivery, clinicians believed that it had great advantages and met the needs of a large number of clients but it could not be the only way to deliver care. For instance, some roles (e.g., psychiatrists) noted the need to have in-person assessments periodically to have a more accurate sense of the clients’ status. |
| Self-Efficacy | +1 | Overall, clinicians reported a sense of self-efficacy in delivering NAVIGATE virtually. For many, this confidence stemmed from feeling effective in the delivery of NAVIGATE in person, which provided a solid basis for the transition to the virtual delivery. |
| Individual Stage of Change | +1 | Clinicians talked about feeling prepared to deliver NAVIGATE virtually but feeling slightly hesitant and overwhelmed at the start given the abrupt transition. With time, there was an increased sense of preparedness with practice and continuous refinement of the online resources to support staff. |
| Individual Identification with the Organization | +2 | There was a general consensus among clinicians that their commitment to SCEIS strongly and positively influenced their interest in learning, taking on new initiatives, adapting to change, and providing the best care for clients. It was noted that the transition to the virtual delivery of NAVIGATE ultimately was an exercise in change management and was closely tied to how the employer was perceived. |
| Other Personal Attributes | Mixed | Clinicians discussed mixed thoughts and experiences related to the transition to virtual delivery of NAVIGATE and alignment with their preferred learning style. Some appreciated the convenience of accessing materials online and learning at their own pace; in contrast, others found it distracting and ineffective to be trained online. Overall, clinicians reported high levels of motivation to make virtual delivery of NAVIGATE work. |
| Characteristics of Clients | Characteristics of Clients | Characteristics of Clients |
| Beliefs and Experience | +1 | Based on the feedback received and their own observations, clinicians believed that the virtual NAVIGATE experience was positive for both clients and their families. Overall, the virtual delivery of NAVIGATE brought great advantages stemming from the convenience of accessing care. Clinicians believed that virtual NAVIGATE facilitated fewer no-shows and increased access to care and client engagement. A period of adjustment was needed at the start of the transition as clients and their families, similar to the healthcare providers, had to learn the details of the online system. |
| Success | Success | Success |
| Success | +2 | The transition to the virtual delivery of NAVIGATE was perceived as successful, with the team being able to adapt smoothly to the new demands of virtual delivery of NAVIGATE and to learn and work together as a team. Clinicians unanimously recommended continuing the virtual delivery of NAVIGATE while recognizing that in an ideal scenario the clients would have a choice for in-person or virtual NAVIGATE, to fit their needs. Having a virtual delivery option was perceived as a way to increase access to care across the country. |
## Intervention characteristics
Adaptability (+2) of NAVIGATE to the virtual context was most strongly associated with its re-implementation (see Table 3). Adaptations to ensure that the virtual delivery of NAVIGATE was appropriate and effective included implementing and learning how to use the Cisco Webex platform, providing clinicians with laptops and phones, and converting the intervention manual into PDF fillable forms. Clinicians felt these modifications were very effective and “working great”. One issue that remained unresolved was the transfer of client-rated side-effects completed on an iPad while waiting to see the psychiatrist.
NAVIGATE was originally implemented due to the desire for more organized and coordinated EPI care throughout Ontario (Intervention Source +1). Virtual delivery of NAVIGATE provided advantages in several ways including accessibility (clients able to meet more often), flexibility (scheduling around school and work), and cost savings (e.g., no need for transportation). Some disadvantages included not having a platform for clients to complete a questionnaire before meeting with the psychiatrist, inequity issues for clients who did not have access to virtual care, and challenges for clinicians in reading body language for assessment purposes (Relative Advantage +1).
Although a few clinicians felt the materials and supports were either not supportive enough at the beginning of the re-implementation (e.g., fillable PDF version of the manual, tip sheets) or provided too much information to absorb (lots of documents to read and videos to watch), most felt that they received helpful guidance, information and support from IT personnel as well as from the reflective practice meetings (Design Quality and Packaging +1).
Two intervention factors had mixed ratings. Clinicians felt NAVIGATE was effective for clients, largely based on their experiences and observations shared from other clinicians and clients, as well as their overall understanding of intervention (Evidence Strength and Quality, mixed). A few clinicians mentioned they were knowledgeable about the research evidence underlying the intervention. Yet, most clinicians initially felt doubtful that NAVIGATE would be as effective virtually as in-person. With time, however, they found that it worked equally well with the exception of monitoring side effects, which required face to face interaction.
With respect to Complexity (mixed), some clinicians found re-implementing NAVIGATE for virtual delivery to be difficult, particularly at the beginning, because it had to be done quickly with many details to be worked out (e.g., ensuring confidentiality, privacy). As well, the technology was challenging for some users (e.g., family members). Other clinicians reported that it was “not terribly difficult” or not much extra work to re-implement because they could rely on others “to figure it out”.
## Outer setting factors
Provincial best practices and EPI standards were seen as providing a major incentive for the implementation of NAVIGATE (External Policies and Incentives, +2). Somewhat less facilitative was the experience of networking and collaborating with other EPI services via EPI-SET ECHO training sessions, which are intended to inform NAVIGATE practice [16]. The ECHO (Extension for Community Healthcare Outcomes) model connects geographically dispersed healthcare providers in online communities of practice with the aim of increasing healthcare access [35]. Affiliations with EPION and with other mental health agencies and former places of work also influenced clinicians' work (Cosmopolitanism, +1).
The extent to which NAVIGATE met Client Needs was mixed among respondent clinicians. Most perceived NAVIGATE as valuable to clients and families, based on the positive feedback they received, particularly the structured and team approach to care. However, they also noted that for some clients, the material was daunting and lengthy. Cultural and language differences, clients having comorbidities, and issues accessing the technology were also perceived to be barriers to participating in NAVIGATE. The rapid pivot to virtual delivery also meant there was no time to consult clients about the change. Peer Pressure [0] was perceived as neither a barrier nor a facilitator since no other provider organizations were delivering NAVIGATE at that time of this study.
## Process factors
The strongest facilitator for re-implementation was the presence of Formally Appointed Implementation Leaders (+2). Although there was not a lot of pre-pandemic planning, the leaders were viewed as collaborative and helpful. The presence of Champions (+1) and Opinion Leaders (+1) was also facilitative. Clinicians felt that key people who were instrumental in pivoting to the virtual delivery of NAVIGATE worked hard and were collaborative in their approach. They noted several strategies leaders used to encourage and inform clinicians to move to virtual delivery of care including numerous emails, links to training, meetings, and providing opportunities to ask questions as well as encouraging flexibility in the delivery of NAVIGATE. Clients were informed about changes through email discussions. Clinicians further noted that there was no choice but to move to virtual care delivery but made concessions for in-person appointments when it was possible.
Clinicians held mixed opinions about the Executing of the re-implementation. They spoke of the changes as being a “tsunami”. Some clinicians mentioned they were consulted via team “huddles”, problem-solving discussions and opportunities to pose questions to the implementation leaders. Others felt that they were “told” about the changes and that execution was more instructive than collaborative.
The consensus among clinicians was that there was a lack of Planning (mixed) in the move to virtual delivery, which they recognized as unavoidable due to the sudden need to maintain service in the pandemic. Initially, the pivot to virtual delivery was overwhelming. However, clinicians felt that the implementation leaders were the appropriate people to lead the way and that they did their best to make it as easy and smooth as possible. One clinician felt that the SEE role did not receive a lot of guidance. At the time of the interview, most clinicians felt that virtual NAVIGATE had been fully re-implemented.
Opportunities for Reflecting and Evaluating were also mixed. Some clinicians spoke of receiving feedback about what was working and what was not working, as well as statistics about engagement (clients who did not attend their appointment, who they were seeing) that included discussions and reflections on the information shared. Others received informal feedback (i.e., no statistics) and others did not recall receiving any specific feedback about how virtual delivery was going.
## Inner setting factors
Structural Characteristics (mixed) of the organization were noted as having both positive and negative influences on re-implementation. A strength was that CAMH is a large, resource-intensive setting where staff were provided with laptops, mobile phones, and rooms for private meetings with clients (virtually as well as in-person). Barriers were that clients were unable to input personal information when using the virtual modules without compromising confidentiality, and improvements are needed to the electronic health record.
CAMH as a setting was also highly facilitative for re-implementation due to its Culture (+2) and Networks and Communications (+2). Clinicians regarded the workplace culture as highly positive, collaborative, warm, healthy, supportive, client-centered, and acknowledged that it impacted positively on the transition to virtual delivery of NAVIGATE. Working together as a team and focusing on delivering the highest quality care possible were perceived as key contributors to the success of the re-implementation. The multiple and continuous channels of communication via emails, virtual team meetings, sharing links to resources, updated policies and problem-solving including communication outside of CAMH with other EPI sites via ECHO were perceived as very supportive. Although the volume of new information and communications was overwhelming, it was generally recognized as necessary to support the transition to virtual delivery within a matter of days.
Within the Implementation Climate, specifically Tension for Change (+2) and Relative Priority (+2) were the strongest facilitators in this domain. Clinicians unanimously noted a high tension for change for NAVIGATE because it provided consistency in delivering care, a holistic and standardized approach, and its multiple roles had a clear scope of practice that benefitted various client needs. The transition to virtual NAVIGATE was unambiguously perceived as the main organizational priority by all clinicians. Competing priorities fell to the wayside and all clinicians fully dedicated their time and attention to the virtual delivery, which contributed to its success.
Organizational Incentives and Reward (+1) were also facilitative with several incentives noted for both clients and clinicians. Client incentives included the convenience of accessing care which increased participation, reduced time, parking and transportation costs, and increased flexibility. Clinicians were strongly motivated by the urgent need to find a way to maintain care delivery in the face of pandemic restrictions. Many also mentioned that their efforts to re-implement were recognized by their clinical manager.
The Learning Climate (+1) at SCEIS was perceived positively and as encouraging of learning and taking on new initiatives. Clinicians valued the availability of multiple learning opportunities, both internally and externally, and the supports provided for participating in these opportunities.
Leadership Engagement (+2) was the strongest readiness facilitator. Clinicians unanimously believed there was support from leadership for the virtual delivery of NAVIGATE. Multiple discussions were held regarding what was needed, special considerations for virtual delivery of care were put in place (e.g., privacy; when in-person was essential, role-specific tasks such as who monitors side effects) and leadership were on board and engaged in the re-implementation process.
Re-implementation was supported by Available Resources (+1) including many sources of information and supports to ensure clinicians had the resources needed to perform their role successfully (e.g., Webex support, Virtual Mental Health and Outreach program, educational services, internal team, etc.). Some clinicians had not received the original NAVIGATE training in the initial implementation and perceived this as a limitation. Clinicians valued getting laptops early in the process, which was essential to the virtual transition, but noted that access to cell phones was delayed.
Experience with Access to Knowledge and Information was mixed as re-implementation did not include formal training due to the rapidity of the pivot. There were, however, multiple resources available to support the transition via links, training videos and emails. The amount of information to be accessed, absorbed and implemented in a very short period of time made the initial experience overwhelming for many clinicians but this improved with time.
## Characteristics of clinicians
The most facilitative factor related to the clinicians was their Individual Identification with the Organization (+2). There was a general consensus among clinicians we interviewed that their commitment to CAMH strongly and positively influenced their interest in learning, taking on new initiatives, adapting to change, and providing the best care for clients. It was noted that the transition to the virtual delivery of NAVIGATE ultimately was an exercise in change management and was closely tied to how the employer was perceived.
Clinicians’ Knowledge and Beliefs about the Intervention (+1) was also supportive of re-implementation. Clinicians regarded the NAVIGATE model positively and valued the evidence base and the holistic approach. They viewed virtual delivery as advantageous but some clinicians (i.e., psychiatrists) noted the need to have in-person assessments periodically to have a more accurate sense of the clients' status.
Clinicians reported a sense of Self-Efficacy (+1) in delivering NAVIGATE virtually. For many, this confidence stemmed from feeling effective in the delivery of NAVIGATE in person, which provided a solid basis for the transition to virtual delivery. They felt prepared to deliver NAVIGATE virtually (Individual Stage of Change +1), but also slightly hesitant and overwhelmed at the start given the abrupt transition. With time, there was an increased sense of preparedness with practice and continuous refinement of the online resources to support clinicians. Participants discussed mixed thoughts and experiences related to the transition to virtual delivery of NAVIGATE and alignment with their preferred learning style. Some appreciated the convenience of accessing materials online and learning at their own pace; in contrast, others found it distracting and ineffective to be trained online. Overall, clinicians reported high levels of motivation to make virtual delivery of NAVIGATE work.
## Client characteristics
Clinicians believed virtual navigate provided a positive experience for both clients and their families. The virtual delivery of navigate was very advantageous for continuing to access care when in-person care could not be delivered. There were fewer no-shows, increased access to care and better client engagement. A period of adjustment was needed at the start of the transition as clients and their families had to become familiar with the digital platform, as did the clinicians.
The transition to the virtual delivery of NAVIGATE was Perceived as Successful (+2), with the team being able to adapt smoothly to the new demands and to learn and work together as a team. Clinicians unanimously recommended continuing with virtual delivery of NAVIGATE while recognizing that in an ideal scenario, clients would have a choice of in-person or virtual delivery to fit their needs and preferences. Having a virtual delivery option was perceived as a way to increase access to care across the country.
## Stakeholder engagement
The stakeholders, including youth and family with lived experiences, front-line clinicians, and clinical leads, participated consistently and meaningfully throughout the course of this study.
In the initial phases, all stakeholders participated in the grant application and development of the practice profile with front-line clinicians [22, 36]. For objective 1, Modifications, front-line clinicians, clinical leads and youth and family with lived experiences participated in monthly meetings to explore and review modifications that occurred during the shift to virtual care delivery. Following these meetings, trainings were organized in collaboration with clinical staff, leadership and youth and family members with lived experience. Youth with lived experiences also contributed to the development of the web-based resources to enhance engagement. Regarding objective 2, Fidelity, feedback from front-line clinicians and clinical leads informed the fidelity ratings. Regarding objective 3, Implementation Facilitators and Barriers, front-line clinicians and clinical leads participated in the interviews.
Furthermore, youth and family with lived experience, front-line clinicians and clinical leads contributed to team discussions on data interpretation and development of a knowledge translation plan and products.
## Discussion
In this mixed methods study investigating the unplanned shift to virtual delivery of EPI care, we identified several modifications required to deliver the NAVIGATE program virtually by using the NAVIGATE practice profile and the FRAME framework. We discussed the potential impact of these modifications on fidelity and outcomes during structured meetings with clinicians, revised the practice profile, and captured modifications using the FRAME. We then formally evaluated impacts on fidelity to the provincial EPI-standards with a validated assessment tool (FEPS-FS) prior to and after the modifications were made. We investigated implementation facilitators and barriers for the virtual delivery of NAVIGATE with clinicians and identified several contextual factors that were critical to re-implementation of NAVIGATE. To our knowledge, this is the first study to describe a re-implementation process this comprehensively. We summarize overall results and experiences with this re-implementation process, strengths and limitations of the approaches we used, and opportunities and needs for future research.
## Modifications
Regarding the first aim of the study, the identification of modifications needed for virtual EPI care, we identified several cross-cutting and role-specific modifications. Most of these modifications were adaptable, though some challenges were identified that could not be mitigated in a virtual setting (e.g., conducting physical assessment).
Our assessment of the modifications needed to support virtual care delivery is largely similar to two recent studies that also describe the shift to virtual care in early psychosis coordinated specialty care programs [37, 38]. McCormick and colleagues investigated the pandemic-driven shift to continue care delivery via videoconference and phone at 23 sites across Texas, US [37]. Their results show many sites lacked training, resources, policies and procedures to shift to virtual care, and the challenges that were identified included limited capacity to deliver community-based outreach, family engagement, and vocational support, and difficulties with access and connectivity for clients. Similar to the modifications we identified in our study, organizations provided training to support staff early on in the pandemic, leveraged virtual tools e.g., e-mailing clinical worksheets, sharing mobile apps, and sharing other resources such as videos, and reimbursed virtual care - an important facilitator for virtual care delivery [37]. Similarly, Meyer-Kalos and colleagues explored challenges and solutions in the shift to virtual care delivery across several EPI services in the United States, Israel, and China [38]. These authors also highlighted the importance of implementing procedures to provide care virtually, adapting appointments times and duration, and adapting materials for digital use. They describe specific challenges and mitigating strategies at the clinician-level per NAVIGATE role, such as challenges for SEE clinicians associated with the COVID-19 constricted labor market and unavailability of outreach visits [38, 39]. They described modifications similar to ours, such as shifting focus to practicing skills for remote learning and working and conducting job interviews remotely. Regarding the prescriber role, both our study and Meyer-Kalos’ reported challenges with follow-up for medication benefits and side-effects, and a reluctance by prescribers to make changes to medication, particularly switching to clozapine because it requires monitoring with blood tests that were challenging to obtain during the pandemic [38]. Mitigating strategies also overlap across our studies, with increased frequency of appointments and involvement of family members to improve monitoring of medication [38].
There are similarities between the modifications and mitigating strategies we identified in similar studies in the child and youth health mental services sector in Ontario [9]. Common strategies included provision of software and hardware, clinician training in software to provide virtual care, adapting materials, offering phone sessions and adding text message-based support to address accessibility issues, development of safety protocols, and breaking sessions into smaller segments to increase client engagement. Clinicians were encouraged to engage in self-care activities and some clinics installed flexible hours of service to accommodate clients' and clinicians' other responsibilities [9]. The similarities across settings surfaced several cross-cutting modifications needed for delivery of virtual care as well as specific adjustments related to NAVIGATE role-based core components.
Coding of modifications in the FRAME [11] highlighted that modifications were mainly initiated by the COVID-19 pandemic. Modifications occurred at different levels, ranging from the SCEIS team to the CAMH organization to the provincial government. Decisions underlying the modifications were also made at these levels, by individual clinicians, clinic manager, and organizational leadership. Of note is that, the COVID-19 pandemic did not only trigger this pivot to virtual care but the COVID-19 related effects were wide-ranging, from impacts on the health systems organization, e.g., reduced access to primary care, but also impacting clients in the reduced opportunity for finding work or attending school remotely, which is reflected in several modifications.
## Fidelity
Fidelity assessment with the FEPS-FS revealed that the majority of EPI items ($\frac{23}{29}$) were rated as satisfactorily or fully implemented, and that the core structure of the NAVIGATE program was strongly preserved despite modifications for virtual delivery. These positive results may be related to the extra training and support clinicians received to facilitate re-implementation from the onset of the pandemic.
Compared to the fidelity assessment of in-person NAVIGATE care, the level of program delivery was maintained for many of the assessed items and improved in several areas in the virtual context. The results for the domain access and continuity were mixed, e.g., item scores on timely contact with the referred individual improved, but more new clients had experienced inpatient psychiatric admissions prior to entering the EPI program, and delivery of targeted community education events decreased. The faster connection to a clinician after referral could reflect improved access to care virtually (fewer missed appointments), reduced clinic waitlist, and greater client flexibility to meet during the daytime (individuals were less constrained by work or school hours). On the other hand, most clients experienced an inpatient admission before their admission to NAVIGATE, and this proportion increased compared to in-person care before the COVID-19 pandemic.
Increased inpatient admission could be related to the COVID-19 pandemic. Worsening mental health symptoms and/or increased substance use during the pandemic [40, 41] could have led to more hospitalizations for psychosis [42] or a decline in visits to the primary care providers who could have otherwise referred for outpatient early intervention care [43]. As well, there may have been fewer opportunities for youth to connect with their wider support system, such as teacher or coaches, who might otherwise have detected mental health issues and supported them with finding appropriate supports/early treatments.
Additionally, targeted supports for community-based education and employment also decreased. This was a challenge for the CAMH EPI program before the pandemic because of how hospital-based care is organized. The decrease in educational supports also stemmed from the cancellation or postponement of community education due to COVID-19 restrictions, and educational institutions prioritized COVID-19 related practicalities including the shift to remote learning.
Despite reservations voiced by staff about the virtual delivery of medical care in the FRAME discussions, fidelity ratings for health management in the medical care domain remained high. Fidelity feedback for health management suggested that mitigation strategies were identified such as leveraging alternatives to physical assessment (e.g., measure weight at home or blood pressure at pharmacy or primary care practice). Also, while it is possible that physicians were more cautious about medication management, prescribing remained within recommended guidelines which is what the fidelity review assesses.
Implementation remained high for delivery of psychosocial treatments, which aligns with efforts to sustain client retention by offering different options for connecting, shifting to shorter, more frequent meetings, and synchronously sharing fillable PDFs. As captured in the FRAME, most modifications were described as fidelity-consistent, which is reflected by the “fully implemented” fidelity scores. To our knowledge, there are no other published studies investigating fidelity for a virtual comprehensive EPI care program compared to in-person care. There are, however, several studies that report on treatment fidelity for virtual delivery compared to in-person delivery of a structured psychosocial intervention in other populations. In these publications, there was no evidence that virtual delivery achieved worse fidelity compared to in-person delivery (44–46).
## Facilitators and barriers
CFIR interviews surfaced several factors that facilitated the re-implementation of virtual care. The most salient facilitators were adaptability of NAVIGATE, external policies and incentives, and the tension for change brought on by the COVID-19 pandemic. Implementation leaders were also highly facilitative, despite the abrupt shift and limited time for planning. Workplace culture, clinicians' identification with the organization, and the transition to virtual NAVIGATE becoming a strong relative priority in the organization.
Few barriers were mentioned, but clinicians noted that virtual delivery did not always align with client needs and resources. Some clients found the intervention related material challenging to get through, while others were challenged by cultural and language differences, co-morbidities, issues accessing technology and challenges with adequately monitoring side-effects in a virtual setting.
These results are largely in line with a recent study exploring the pandemic-related transition to virtual care across child and youth mental services in Ontario [9]. Using a multi-level mixed method design and CFIR interviews, Danseco identified several facilitators including staff engagement and motivation, provision of enabling software and hardware, leadership support, and training activities [9]. Clinicians also mentioned the positive impact of collaboration and having a champion or community of colleagues for learning virtual care delivery together. Barriers in the Danseco study included internet connection issues, lack of resources, and privacy concerns [9]. Clinicians also noted fatigue from engaging in online sessions and a feeling of isolation from their colleagues. The authors concluded that overall, many service providers had similar experiences implementing virtual care. With the appropriate support, infrastructure, and resources, many clinicians and clients found virtual delivery of care acceptable and would like to continue using it or having it as an option [9].
Our findings also align with factors associated with implementation success across a diverse array of settings and interventions, including weight management in a large integrated U.S. healthcare system, an e-health application in Norway, and a Canadian study of a maternal and child health intervention undertaken in Mali and Ethiopia [47].
## Re-implementation process
Use of the NAVIGATE practice profile and the FRAME to identify modifications facilitated a structured, explicit and comprehensive assessment of modifications in a dynamic context that could have negatively impacted care delivery [11, 48]. Taking stock of modifications to core intervention components is crucial for understanding fidelity and effectiveness outcomes [18]. The addition of clinician-reported barriers, mitigating strategies and impacts to our practice profile enabled us to track what strategies were used to reduce potentially negative impacts. This approach tracking and using data “along the way” to inform subsequent adaptations (e.g., updates to training, material) contrasts with more linearly designed studies that conduct fulsome impact assessments prior to refining and evaluating an adapted version of an intervention that is hypothesized to fit better [49]. Rapid and iterative assessments of modifications and impacts provided a great advantage to optimizing re-implementation, especially when unplanned modifications could negatively impact outcomes [11, 48]. A similar stepwise process of revising/developing policies and workflows, providing training, reflecting/evaluating, and taking steps for further improvement during the abrupt shift to virtual care in the pandemic was also observed in other health care agencies that implemented virtual delivery of care in Ontario [9]. Another advantage of using the practice profile was that clarity on the intervention components made it was easy for clinicians to identify where and what modifications were needed and/or had occurred.
A disadvantage of our approach was that some of the FRAME domains overlap with the determinant domains of the CFIR, which is less efficient compared to using one instrument only. Other studies also described an overlap between the FRAME and CFIR and decided to reduce certain items of the FRAME for efficiency [50]. Furthermore, several of the components of the Process domain of the FRAME were similar between the modifications and were summarized to lessen redundancy. Additionally, the original FRAME framework does not capture the impact of a modification. We added a category of impact and mitigating strategy to the FRAME constructs because systematic consideration of all potential impacts on a range of implementation and intervention outcomes is critical for further optimization of the intervention [51].
Regarding the fidelity assessment, we measured fidelity to the provincial EPI standards with a validated measure, the FEPS-FS. We intend to measure fidelity to the core components of NAVIGATE by reviewing delivery metrics from randomly selected charts, and report the results of thisin a future paper.
## Strengths and limitations
To our knowledge this is the first study of the re-implementation of a comprehensive early psychosis intervention for virtual care delivery. We investigated modifications, fidelity to EPI standards, and determinant factors which are the first 3 objectives of our larger, mixed-methods study. Previous studies have investigated satisfaction, and facilitators and barriers to virtual care delivery based mostly on interviews with health care providers [52], though some included client experiences [53, 54]. Our study presents a more rigorous approach to investigating re-implementation of a comprehensive intervention during an abrupt shift to virtual care initiated by the demands of the COVID-19 pandemic. Here, we report on the first objectives of the study. In a later paper we will describe client's and clinician's experiences and measures of engagement later to provide a fulsome description of the impact of virtual care delivery of NAVIGATE.
The success of our re-implementation may be unique to the COVID-19 pandemic context. COVID-19 related restrictions to social contacts likely triggered a strong motivation to continue care while adhering to these restrictions, leading to a quick pivot to virtual care delivery. Other key facilitating factors may also be unique to this context, including the support provided by CAMH and the Virtual Mental Health and Outreach program specifically, the adaptability of the NAVIGATE program, and other availability of resources such as materials and funds [30].
Furthermore, the switch to virtual care delivery may have unintentionally created disparities in the mental health care system for people with limited or no access to technology or to the private space needed to attend virtual appointments [55]. Relatedly, social isolation may be an unavoidable outcome of virtual care delivery that will require further examination to address. Ongoing remuneration for virtual service delivery remains uncertain and will undoubtably be an important consideration to monitor moving forward.
## Future steps
Virtual EPI care has the potential to complement traditional in-person EPI care and improve access in specific contexts, e.g., in remote geographic areas. Improving access to specialty health care is particularly relevant for individuals living in rural and remote communities as they tend to experience poorer health, greater disability, and higher mortality [56]. To facilitate equitable care, it will be important to investigate client experiences with virtual care, and to address related barriers stemming from sociodemographic factors that lead to health disparities [55, 57]. Following on the results from this work, more research is needed to assess the efficacy and generalizability of virtual EPI care and patient's preferences towards virtual or hybrid care, beyond the COVID-19 pandemic context.
## Conclusion
In conclusion, re-implementation of NAVIGATE for virtual delivery during the COVID-19 pandemic was rapid, unplanned, and complex. Understanding how re-implementation transpired, involved an exploration of barriers, strategies, and impacts across levels of the organization. This study suggests that a comprehensive EPI program can be re-implemented for virtually delivery while maintaining high EPI standards with the appropriate support, infrastructure, and resources. Virtual delivery of NAVIGATE holds promise for increasing access to effective care for youth with psychosis. Going forward, it will be important to ensure future pivots to virtual delivery for NAVIGATE and other interventions maintain equitable care.
## Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author.
## Ethics statement
The studies involving human participants were reviewed and approved by Centre for Addiction and Mental Health's Research Ethics Board. The patients/participants provided their written informed consent to participate in this study.
## Author contributions
WT, NK, AC, AV, DA, SB, LDF, AK, AP, RR, JD, and MB drafted and revised sections of the paper and approved the final version. TA, CB, SB, SJ, AM, AS, and VV made critical revisions to the paper and approved the final version. MB developed the implementation evaluation. DA consulted on the fidelity assessment plan. JD developed the fidelity assessment plan. AC consulted on virtual health care delivery. AV consulted on the development of the project. TA, AM, and VV consulted on the evaluation of engagement of people with lived experience. NK was the principal investigator. MB and JD contributed equally to the paper. All authors contributed to the article and approved the submitted version.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Correll CU, Galling B, Pawar A, Krivko A, Bonetto C, Ruggeri M. **Comparison of early intervention services vs. treatment as usual for early-phase psychosis: a systematic review, meta-analysis, and meta-regression**. *JAMA Psychiatry* (2018.0) **75** 555-65. DOI: 10.1001/jamapsychiatry.2018.0623
2. Kane JM, Robinson DG, Schooler NR, Mueser KT, Penn DL, Rosenheck RA. **Comprehensive versus usual community care for first-episode psychosis: 2-year outcomes from the NIMH RAISE early treatment program**. *Am J Psychiatry* (2016.0) **173** 362-72. DOI: 10.1176/appi.ajp.2015.15050632
3. Anderson KK, Norman R, MacDougall A, Edwards J, Palaniyappan L, Lau C. **Effectiveness of early psychosis intervention: comparison of service users and nonusers in population-based health administrative data**. *Am J Psychiatry* (2018.0) **175** 443-52. DOI: 10.1176/appi.ajp.2017.17050480
4. Bond GR, Drake RE, Luciano A. **Employment and educational outcomes in early intervention programmes for early psychosis: a systematic review**. *Epidemiol Psychiatr Sci* (2015.0) **24** 446-57. DOI: 10.1017/S2045796014000419
5. Craig TKJ, Garety P, Power P, Rahaman N, Colbert S, Fornells-Ambrojo M. **The lambeth early onset (LEO) team: randomised controlled trial of the effectiveness of specialised care for early psychosis**. *Br Med J* (2004.0) **329** 1067. DOI: 10.1136/bmj.38246.594873.7C
6. Malla A, Schmitz N, Norman R, Archie S, Windell D, Roy P. **A multisite Canadian study of outcome of first-episode psychosis treated in publicly funded early intervention services**. *Can J Psychiatry* (2007.0) **52** 563-71. DOI: 10.1177/070674370705200904
7. Petersen L, Jeppesen P, Thorup A, Abel MB, Øhlenschlaeger J, Christensen TØ. **A randomised multicentre trial of integrated versus standard treatment for patients with a first episode of psychotic illness**. *Br Med J* (2005.0) **331** 602. DOI: 10.1136/bmj.38565.415000.E01
8. Brown E, Gray R, Lo Monaco S, O’Donoghue B, Nelson B, Thompson A. **The potential impact of COVID-19 on psychosis: a rapid review of contemporary epidemic and pandemic research**. *Schizophr Res* (2020.0) **222** 79-87. DOI: 10.1016/j.schres.2020.05.005
9. Danseco E, Kurzawa J, Sundar P, Brown J, Huang C. **Evaluating the sector-wide implementation of virtual child and youth mental health services in response to the COVID-19 pandemic: perspectives from service providers, agency leaders and clients**. *Implement Res Pract* (2021.0) **2** 263348952110456. DOI: 10.1177/26334895211045690
10. Inchausti F, MacBeth A, Hasson-Ohayon I, Dimaggio G. **Telepsychotherapy in the age of COVID-19: A commentary**. *J Psychother Integr* (2020.0) **30** 394-405. DOI: 10.1037/int0000222
11. Wiltsey Stirman S, Baumann AA, Miller CJ. **The FRAME: an expanded framework for reporting adaptations and modifications to evidence-based interventions**. *Implement Sci* (2019.0) **14** 58. DOI: 10.1186/s13012-019-0898-y
12. Mackie TI, Ramella L, Schaefer AJ, Sridhar M, Carter AS, Eisenhower A. **Multi-method process maps: an interdisciplinary approach to investigate ad hoc modifications in protocol-driven interventions**. *J Clin Transl Sci* (2020.0) **4** 260-9. DOI: 10.1017/cts.2020.14
13. Stirman SW, Miller CJ, Toder K, Calloway A. **Development of a framework and coding system for modifications and adaptations of evidence-based interventions**. *Implement Sci* (2013.0) **8** 65-65. DOI: 10.1186/1748-5908-8-65
14. Quimby KR, Murphy MM, Harewood H, Howitt C, Hambleton I, Jeyaseelan SM. **Adaptation of a community-based type-2 diabetes mellitus remission intervention during COVID-19: empowering persons living with diabetes to take control**. *Implement Sci Commun* (2022.0) **3** 5. DOI: 10.1186/s43058-022-00255-9
15. Wong S, Hassett L, Koorts H, Grunseit A, Tong A, Tiedemann A. **Planning implementation and scale-up of physical activity interventions for people with walking difficulties: study protocol for the process evaluation of the ComeBACK trial**. *Trials* (2022.0) **23** 40. DOI: 10.1186/s13063-021-05990-3
16. Kozloff N, Foussias G, Durbin J, Sockalingam S, Addington J, Addington D. **Early psychosis intervention-spreading evidence-based treatment (EPI-SET): protocol for an effectiveness-implementation study of a structured model of care for psychosis in youth and emerging adults**. *BMJ Open* (2020.0) **10** e034280. DOI: 10.1136/bmjopen-2019-034280
17. Fetters MD, Curry LA, Creswell JW. **Achieving integration in mixed methods designs—principles and practices**. *Health Serv Res* (2013.0) **48** 2134-56. DOI: 10.1111/1475-6773.12117
18. Metz A. (2016.0)
19. Nevedal AL, Reardon CM, Opra Widerquist MA, Jackson GL, Cutrona SL, White BS. **Rapid versus traditional qualitative analysis using the consolidated framework for implementation research (CFIR)**. *Implement Sci* (2021.0) **16** 67. DOI: 10.1186/s13012-021-01111-5
20. Tempelaar W, Barwick M, Crawford A, Voineskos A, Addington D, Addington J. **Adapting evidence-based early psychosis intervention services for virtual delivery: protocol for a pragmatic mixed methods implementation and evaluation study**. *JMIR Res Protoc* (2021.0) **10** e34591. DOI: 10.2196/34591
21. Mueser KT, Penn DL, Addington J, Brunette MF, Gingerich S, Glynn SM. **The NAVIGATE program for first-episode psychosis: rationale, overview, and description of psychosocial components**. *Psychiatr Serv* (2015.0) **66** 680-90. DOI: 10.1176/appi.ps.201400413
22. Hawke LD, Relihan J, Miller J, McCann E, Rong J, Darnay K. **Engaging youth in research planning, design and execution: practical recommendations for researchers**. *Health Expect* (2018.0) **21** 944-9. DOI: 10.1111/hex.12795
23. 23.
https://www.cisco.com/c/en/us/products/collateral/unified-communications/camh-case-study.html
24. 24.
https://digital.camhx.ca/
25. Blase K, Fixsen D. *Core intervention components: identifying and operationalizing what makes programs work. ASPE research brief* (2013.0)
26. Durbin J, Selick A, Langill G, Cheng C, Archie S, Butt S. **Using fidelity measurement to assess quality of early psychosis intervention services in Ontario**. *Psychiatr Serv* (2019.0) **70** 840-4. DOI: 10.1176/appi.ps.201800581
27. Finley EP, Huynh AK, Farmer MM, Bean-Mayberry B, Moin T, Oishi SM. **Periodic reflections: a method of guided discussions for documenting implementation phenomena**. *BMC Med Res Methodol* (2018.0) **18** 153. DOI: 10.1186/s12874-018-0610-y
28. Breitenstein SM, Gross D, Garvey CA, Hill C, Fogg L, Resnick B. **Implementation fidelity in community-based interventions**. *Res Nurs Health* (2010.0) **33** 164-73. DOI: 10.1002/nur.20373
29. Addington DE, Norman R, Bond GR, Sale T, Melton R, McKenzie E. **Development and testing of the first-episode psychosis services fidelity scale**. *Psychiatr Serv* (2016.0) **67** 1023-5. DOI: 10.1176/appi.ps.201500398
30. Damschroder LJ, Aron DC, Keith RE, Kirsh SR, Alexander JA, Lowery JC. **Fostering implementation of health services research findings into practice: a consolidated framework for advancing implementation science**. *Implement Sci* (2009.0) **4**. DOI: 10.1186/1748-5908-4-50
31. Selick A, Langill G, Cheng C, Addington D, Archie S, Butt S. **Feasibility and acceptability of a volunteer peer fidelity assessment model in early psychosis intervention programmes in Ontario: results from a pilot study**. *Early Interv Psychiatry* (2021.0) **15** 480-5. DOI: 10.1111/eip.12961
32. Addington D, Noel V, Landers M, Bond GR. **Reliability and feasibility of the first-episode psychosis services fidelity scale-revised for remote assessment**. *Psychiatr Serv* (2020.0) **71** 1245-51. DOI: 10.1176/appi.ps.202000072
33. Gale RC, Wu J, Erhardt T, Bounthavong M, Reardon CM, Damschroder LJ. **Comparison of rapid vs in-depth qualitative analytic methods from a process evaluation of academic detailing in the veterans health administration**. *Implement Sci* (2019.0) **14** 11. DOI: 10.1186/s13012-019-0853-y
34. Marshall S, Taki S, Love P, Laird Y, Kearney M, Tam N. **The process of culturally adapting the healthy beginnings early obesity prevention program for arabic and Chinese mothers in Australia**. *BMC Public Health* (2021.0) **21** 284. DOI: 10.1186/s12889-021-10270-5
35. Arora S, Thornton K, Murata G, Deming P, Kalishman S, Dion D. **Outcomes of treatment for hepatitis C virus infection by primary care providers**. *N Engl J Med* (2011.0) **364** 2199-207. DOI: 10.1056/NEJMoa1009370
36. Henderson J, Sword W, Niccols A, Dobbins M. **The connections research team. Implementing stakeholder-informed research in the substance abuse treatment sector: strategies used by connections, a Canadian knowledge translation and exchange project**. *Subst Abuse Treat Prev Policy* (2014.0) **9** 21. DOI: 10.1186/1747-597X-9-21
37. McCormick KA, Sevillano L, Klodnick VV, Lopez MA, Cohen DA. **Virtual technology’s critical role in sustaining coordinated specialty care in Texas during the COVID-19 pandemic**. *Psychiatr Serv* (2022.0) appips202100429. DOI: 10.1176/appi.ps.202100429
38. Meyer-Kalos PS, Roe D, Gingerich S, Hardy K, Bello I, Hrouda D. **The impact of COVID-19 on coordinated specialty care (CSC) for people with first episode psychosis (FEP): preliminary observations, and recommendations, from the United States**. *Israel and China Null* (2021.0) **34** 387-410
39. Szmulewicz A, Öngür D, Shinn AK, Carol EE, Dow J, Yilmaz N. **Impact of the COVID-19 pandemic on the employment and educational outcomes of individuals in a first-episode psychosis clinic**. *Psychiatr Serv* (2022.0) **73** 165-71. DOI: 10.1176/appi.ps.202100114
40. Horigian VE, Schmidt RD, Feaster DJ. **Loneliness, mental health, and substance use among US young adults during COVID-19**. *J Psychoactive Drugs* (2021.0) **53** 1-9. DOI: 10.1080/02791072.2020.1836435
41. Imtiaz S, Wells S, Rehm J, Hamilton HA, Nigatu YT, Wickens CM. **Cannabis use during the COVID-19 pandemic in Canada: a repeated cross-sectional study**. *J Addict Med* (2021.0) **15** 484-90. DOI: 10.1097/ADM.0000000000000798
42. Chong WWY, Acar ZI, West ML, Wong F. **A scoping review on the medical and recreational use of Cannabis during the COVID-19 pandemic**. *Cannabis Cannabinoid Res* (2022.0) **7** 591-602. DOI: 10.1089/can.2021.0054
43. Stephenson E, O’Neill B, Gronsbell J, Butt DA, Crampton N, Ji C. **Changes in family medicine visits across sociodemographic groups after the onset of the COVID-19 pandemic in Ontario: a retrospective cohort study**. *CMAJ Open* (2021.0) **9** E651-8. DOI: 10.9778/cmajo.20210005
44. Corona LL, Stainbrook JA, Simcoe K, Wagner L, Fowler B, Weitlauf AS. **Utilization of telemedicine to support caregivers of young children with ASD and their part C service providers: a comparison of intervention outcomes across three models of service delivery**. *J Neurodev Disord* (2021.0) **13** 38. DOI: 10.1186/s11689-021-09387-w
45. Ruble LA, McGrew JH, Toland MD, Dalrymple NJ, Jung LA. **A randomized controlled trial of COMPASS web-based and face-to-face teacher coaching in autism**. *J Consult Clin Psychol* (2013.0) **81** 566-72. DOI: 10.1037/a0032003
46. Woolf C, Caute A, Haigh Z, Galliers J, Wilson S, Kessie A. **A comparison of remote therapy, face to face therapy and an attention control intervention for people with aphasia: a quasi-randomised controlled feasibility study**. *Clin Rehabil* (2016.0) **30** 359-73. DOI: 10.1177/0269215515582074
47. Barwick M, Barac R, Damschroder L. *Factors associated with effective implementation: research and practical implications. Implementation science 3.0* (2020.0) 81-100
48. Moore JE, Bumbarger BK, Cooper BR. **Examining adaptations of evidence-based programs in natural contexts**. *J Prim Prev* (2013.0) **34** 147-61. DOI: 10.1007/s10935-013-0303-6
49. Chambers DA, Norton WE. **The adaptome: advancing the science of intervention adaptation**. *Am J Prev Med* (2016.0) **51** S124-131. DOI: 10.1016/j.amepre.2016.05.011
50. Haley AD, Powell BJ, Walsh-Bailey C, Krancari M, Gruß I, Shea CM. **Strengthening methods for tracking adaptations and modifications to implementation strategies**. *BMC Med Res Methodol* (2021.0) **21** 133. DOI: 10.1186/s12874-021-01326-6
51. Kirk MA, Moore JE, Wiltsey Stirman S, Birken SA. **Towards a comprehensive model for understanding adaptations’ impact: the model for adaptation design and impact (MADI)**. *Implement Sci* (2020.0) **15** 56. DOI: 10.1186/s13012-020-01021-y
52. Mohammed HT, Hyseni L, Bui V, Gerritsen B, Fuller K, Sung J. **Exploring the use and challenges of implementing virtual visits during COVID-19 in primary care and lessons for sustained use**. *PLoS One* (2021.0) **16** e0253665. DOI: 10.1371/journal.pone.0253665
53. Donelan K, Barreto EA, Sossong S, Michael C, Estrada JJ, Cohen AB. **Patient and clinician experiences with telehealth for patient follow-up care**. *Am J Manag Care* (2019.0) **25** 40-4. PMID: 30667610
54. Nanda M, Sharma R. **A review of patient satisfaction and experience with telemedicine: a virtual solution during and beyond COVID-19 pandemic**. *Telemed J E Health* (2021.0) **27** 1325-31. DOI: 10.1089/tmj.2020.0570
55. Crawford A, Serhal E. **Digital health equity and COVID-19: the innovation curve cannot reinforce the social gradient of health**. *J Med Internet Res* (2020.0) **22** e19361. DOI: 10.2196/19361
56. Leider JP, Meit M, McCullough JM, Resnick B, Dekker D, Alfonso YN. **The state of rural public health: enduring needs in a new decade**. *Am J Public Health* (2020.0) **110** 1283-90. DOI: 10.2105/AJPH.2020.305728
57. De Vera K, Challa P, Liu RH, Fuller K, Feroz AS, Gamble A. **Virtual primary care implementation during COVID-19 in high-income countries: a scoping review**. *Telemed J E Health* (2022.0) **28** 920-31. DOI: 10.1089/tmj.2021.0377
|
---
title: 'Better safe than sorry: Evaluating the implementation process of a home-visitation
intervention aimed at preventing unintentional childhood injuries in the hospital
setting'
authors:
- Ligat Shalev
- Mary C. J. Rudolf
- Sivan Spitzer
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012825
doi: 10.3389/frhs.2022.944367
license: CC BY 4.0
---
# Better safe than sorry: Evaluating the implementation process of a home-visitation intervention aimed at preventing unintentional childhood injuries in the hospital setting
## Abstract
### Background
Child home injuries prevention interventions have rarely been implemented in hospitals. The SHABI program (“Keeping our Children Safe”; in Hebrew: “SHomrim Al BetIchut Yeladenu”) recruits at-risk families arriving with child injury to the Emergency Department. Medical/nursing students conduct two home visits four months apart, providing safety equipment and guidance. One hundred thirty-five families had a first visit and 98 completed the second. Fifty percentage of families were ultra-Orthodox Jews, $11\%$ Arab, and $28\%$ had ≥3 preschool children. We investigated SHABI's implementation using the Consolidated Framework for Implementation Research (CFIR).
### Methods
Between May 2018 and March 2021 SHABI was implemented in the Emergency Department of a hospital in Israel's northern periphery, an area with high child injury rates. The Implementation process was examined through Emergency Department medical records and tracking registries, hospital management, nurses', and home visitors' meetings notes ($$n = 9$$), and a research diary. Hospital's inner setting and SHABI's characteristics were evaluated through interviews with hospital management, nurses, and home visitors 8 months after baseline ($$n = 18$$). Home visitors' characteristics were evaluated through interviews, post-visit questionnaire on challenges encountered ($$n = 233$$), families' perceptions of SHABI and home visitors' skills through telephone interviews ($$n = 212$$); and home visitors awareness of dangers at home ($$n = 8$$) baseline and 8 months later. Qualitative data were analyzed through explanatory content analysis according to CFIR constructs. Quantitative data were analyzed using X2 and Wilcoxon test for dependent subgroups.
### Results
Despite alignment between SHABI and the hospital's mission, structural hospital-community disconnect prevented implementation as planned, requiring adaptation and collaboration with the medical school to overcome this barrier. Recruitment was included in the initial patient triage process but was only partially successful. Medical/nursing students were recruited as home visitors, and following training proved competent. Children were a distraction during the visits, but home visitors developed strategies to overcome this.
### Conclusions
Injury prevention programs in hospitals have significant benefits. Identifying implementation barriers and facilitators allowed implementers to make adaptations and cope with the innovative implementation setting. Models of cooperation between hospital, community and other clinical settings should be further examined.
## Introduction
Unintentional childhood injuries are a major worldwide health and healthcare concern (1–3). In the United States, almost two million children <5 years old are admitted annually to the Emergency Department following an unintentional injury [4]. A sibling's previous admission due to an injury poses additional risk for a child's arrival at the Emergency Department for an injury [5]. Yet, many of the injuries occurring in the home environment could have been prevented by improving home safety and increasing parental supervision [6, 7].
Over the years, a leading strategy for unintentional injury prevention employs parental guidance through home-visitation [8]. Such interventions have been implemented mainly in community settings, such as primary care clinics [9, 10] or early childhood centers [11, 12]. Interestingly, despite Emergency Department admittance rates and hospitals being central stakeholders for reducing child injuries, their role in injury prevention has been minimal and remains unclear.
Hospitals' perceptions on recurrent visits due to disease differ. Traditionally, interaction between the patient and hospital starts with seeking care for an illness, continues with treatment provided by the hospital, and results in recovery and discharge; once a patient is discharged, hospital responsibilities cease [13]. In recent years, there have been efforts to reduce recurrent hospital visits for both adults [14] and children [15]. This includes an expansion in hospital care models involving the community setting through staff home visits or follow-up phone calls after discharge [16]. Although recurrent visits due to child injuries remain a pressing matter, little has changed regarding hospital outreach to prevent avoidable hospital visits due to child injury.
Evidence regarding hospital leadership in designing and implementing home-visitation interventions for reducing child injury is particularly lacking. A literature search reveals only one study reporting a hospital-led intervention where families were approached 3 days post-hospital discharge following a child's injury [17]. One thousand one hundred and seventy-two families received two home visits 1 year apart and two follow-up phone calls in the interim by a home visitor whose professional qualifications were not reported. While the control group received only a general safety pamphlet, the intervention group received an information pack on injury prevention; instructions by a home visitor on how to correct unsafe practices observed in the home, e.g., child's reaching small objects or lack of a smoke detector; instructions on how to prevent similar injuries to what was reported; and coupons to purchase safety devices including installation information. Findings showed no change in child injury rates between the control and intervention groups, nor significant change in parents' awareness and knowledge about child injury. Parents succeeded in improving, on average, only two unsafe practices out of the 11 measured [17]. Another study recruited families to a home-visitation program from a hospital pediatric continuity clinic and focused on parental guidance on child injury prevention [18]. However, recruitment from the clinic's logs included arrivals for any reason-an illness or an injury. Two further studies reported recruiting families to a home-visitation intervention via hospital medical records [19, 20], however their focus was improving child development and parenting practices, and home safety and child injury reduction were only secondary outcome measures.
Interestingly, a common thread in all the studies reviewed is that while the hospitals provided contact details of families via electronic medical records, the extent of their responsibility and involvement remained vague. Moreover, these studies are limited in their reporting of the design and implementation process of hospital-led interventions, and none to date have evaluated the possible reasons for success or failure in achieving the desired outcome in injury reduction. The implementation process of such interventions remains a “black box”. There is a need for understanding processual levers and barriers that can assist in successful implementation in a variety of contexts and settings, and which in turn could contribute to reducing recurrent hospital visits due to child injury.
In the past decade, Implementation Science has emerged as a new field of inquiry to better understand the complexities of translating evidence-based interventions into every-day practice in real-world settings [21, 22]. Complexities manifest also when implementing an intervention in different contexts and settings [22]. Implementation Science helps in scaling-up successful interventions, and in choosing the best approach by understanding the factors that influence the implementation process [23, 24]. Further, when interventions are implemented two potentially conflicting forces may act simultaneously-fidelity vs. adaptability. Fidelity is the degree to which an intervention is implemented according to the original design, and adaptability is the extent to which an intervention may need adjustment according to setting, context, or facing barriers [25].
To date, few published studies have used the lens of implementation science to examine implementation efforts focused on reducing child injury through home visits [26, 27]. Nicks et al. [ 26] examined the implementation process of altering a computer-based intervention into home-visitation design. The software identified home injury dangers according to the data inserted by families. In their study they evaluated the facilitators and barriers encountered, but their findings were limited to the process of altering a computer-based program into a home visit design, and not on the levers and barriers in conducting the home visits. Smithson et al. [ 27] conducted a systematic review for identifying facilitators and barriers for injury prevention from the perspective of community leaders, counselors, implementers, and families. While their study contributed to the identification of levers and barriers affecting the implementation process, this study did not specifically examine home-based interventions, and therefore its insights are limited.
The present study aimed to understand the barriers and facilitators to implementing a novel hospital-led intervention for reducing child injury through home visits.
## The SHABI program
SHABI (“Keeping our Children Safe”; in Hebrew: “SHomrim Al BetIchut Yeladenu”) is a program delivered in a hospital setting. Families are recruited by the pediatric Emergency Department nursing team when attending with an injured pre-school child. They are then assigned to a home visitor-a nursing or medical student, for two home visits-the first immediately following the hospital visit and the second 4 months later. The visits include a tour through the home accompanied by the parents, joint discussion on child safety in each area of the home with a checklist developed from “Beterem-Safe Kids Israel” [28], and installation of provided safety equipment. Two months later, the home visitor calls the family and offers further injury prevention guidance. The second home visit includes an additional home tour and guidance.
The students are trained in five sessions led by various experts conducted over 11 months, involving an injury prevention expert, a local ultra-Orthodox Jewish Rabbi, and the head of social services in a local Arab village. The training includes topics such as child injury epidemiology and prevention, relationship-building, cultural competence skills, and guidance on adapting the visit to the family's culture.
## The conceptual framework used in this study
To evaluate the factors affecting the implementation process, such as organizational factors and the effect implementers had, we used the Consolidated Framework for Implementation Research (CFIR) [29]. CFIR was chosen as it is one of the foremost conceptual frameworks in the field of Implementation Science due to its integration of relevant theories into one unified model [29]. CFIR was contextualized to assist in exploring the factors that influenced SHABI's implementation in the hospital setting, namely: [1] Implementation process- Assessing the intervention's planning and execution, followed by feedback and evaluation process (e.g., pre-implementation meetings); [2] Inner setting- Identifying the organizational factors that affect the intervention implemented (e.g., the organizational vision); [3] Intervention characteristics-Understanding the implementers' perceptions about the intervention (e.g., advantages or difficulties in execution); [4] Individual characteristics-Implementers' knowledge, opinions and skills; [5] outer setting-Examining the contextual factors such as regulations or policies (e.g., federal or national policies) [29]. This last domain was not investigated as it was outside the study's scope.
## Study design and setting
The study was conducted from May 2018 to March 2021 in the Pediatric Emergency Department of a hospital with 330 beds, located in Israel's northern social-geographic periphery. The hospital's surrounding towns and villages rank low in socio-economic status (SES), with 170,000 residents from diverse Jewish and Arab communities, of whom $10\%$ are 0–4 years old [30]. The area is characterized by higher rates of admissions, mortality, and attendance for unintentional childhood injuries compared with the national average [31]. Intervention design and pre-implementation meetings were conducted from May 2018 to April 2019, and SHABI was delivered from May 2019 to June 2020. A significant improvement in home-safety items was observed 4 months after the first visit [14 (IQR 12–16)] vs. [17 (IQR 15–19); $p \leq 0.001$], accompanied by an overall increase in home safety (Mean ± SD 71.9 ± $9.5\%$ vs. 87.1 ± $8.6\%$; $p \leq 0.001$) [32]. We have reported SHABI's impact on home safety previously [32].
## Participants and procedures
The study involved the following participants: Helsinki approval was obtained through the Hospital Ethics Committee (0029-19-ZIV).
## Data collection
Data collection included analysis of documents, questionnaires developed for this study since aside from one existing relevant questionnaire no relevant tools were found in the literature, in-person semi-structured interviews adapted from CFIR's interview guide tool (https://cfirguide.org/) with both hospital and home visitor teams, and through brief telephone interviews with the participating families:
## Data analysis
Semi-structured interviews along with families' post-visit telephone interviews were recorded and transcribed. All data were analyzed through explanatory content analysis [33] based on CFIR constructs [29]. To achieve interrater reliability, two researchers validated the analysis (LS and SS) to ensure the trustworthiness of the results. In case of disagreement, further discussions were held until agreement was reached.
Potential dangers at home were categorized into injury categories and counted for potential dangers reported. Descriptive statistics were used to describe Emergency Department attendance and participation in SHABI. Comparisons of percentages between different groups were analyzed using X2. Non-normally distributed data were analyzed using Wilcoxon test for dependent subgroups (using SPSS version 27.0).
## Results
Analysis of the data showed a variety of factors affecting SHABI's implementation through the prism of CFIR: the implementation process, the hospital's inner setting, SHABI's characteristics and nurses and home visitors' perceptions and skills. Data is presented in Table 1 according to the themes that emerged and exemplified through relative quotes from hospital management, nurses, home visitors, and families.
**Table 1**
| CIFR domain and themes | Barriers | Facilitators | Quotes ([+]=facilitator, [-]=barrier) |
| --- | --- | --- | --- |
| Implementation process | Implementation process | Implementation process | Implementation process |
| Families and home visitors' recruitment process and adherence to the program | 773 eligible families arrived at the Emergency Department due to child injury; only 63% were approached by nurses to participate in SHABI Families often failed to agree to participate or be contacted as they felt they had no need for the intervention Less Arab families completed both visits (7 of 15 Arab families completed both visits vs. 91 of 120 Jewish families; p = 0.02) | Separation between families' recruitment (done in hospital by nurses) and the home visit components (coordinated by the medical school) | [-]“The mother said there is no safer home than her own and no need for a visit” (Home visitors' post-discharge recruitment phone call to a Jewish mother of two preschoolers from a low SES city) |
| Inner setting | Inner setting | Inner setting | Inner setting |
| Compatibility of the hospital's vision with SHABI's mission The hospital's barriers in operating in the community | Despite its mission statement and the hospital director's views on responsibility to the community, in reality, hospital management encountered difficulties in extending its role to the community and operating outside of the hospital setting While recruitment was partially successful in the hospital, concerns about staff insurance outside the hospital precluded hospital nurses conducting home visits as originally intended | SHABI's mission in promoting health in communities located in the hospital catchment area aligned with the hospital's declared mission In the light of hospital barriers, the medical school stepped in and took responsibility for delivery of the home visitation service, and recruited medical and nursing students as home visitors | [+]“I look at the hospital as a community hospital... As a worldview, I would not reduce my responsibility only to what happens within the hospital. I see a broader responsibility within the community as well” (Hospital director) [-]“We work in the hospital, and cannot provide family medicine, community care. It is two different worlds... Hospital is one thing and community is another” (Head of nursing) |
| The hospital's top-down decision-making process | Nurses perceived that the head nurse daily reports on recruitment was a form of criticism, and that the SHABI coordinator was hardly involved | The top-down decision-making process obligated the nurses to recruit ensuring that it was part of their job | [+]“We received an explanation at the staff meeting with all the managers. We were given an explanation about the program- what was required of us. It is clear to me that this is not democracy, I do not choose what to do at my workplace, it is part of the job” (Nurse #4). |
| Strategy and available resources | The lack of time in a busy Emergency Department and burdensome nursing tasks affected nurses' ability to recruit | Including recruitment as an additional task in the initial patient triage process facilitated recruitment | [-] “The problem is that SHABI takes time-this is another form that needs to be filled out, and there are many other things that need to be done. There are more people waiting” (Nurse #1) |
| Intervention characteristics | Intervention characteristics | Intervention characteristics | Intervention characteristics |
| Recruitment following an injury | Nurses perceived some families were too agitated about their child injury to be approached | Recruitment immediately after a child's injury was perceived to be a definite motivator for parents to consent to SHABI and to actively make changes to their homes | [+]“The parents were very happy that I arrived and wanted to schedule the visit. Both parents were present. They encountered a serious incident [injury] with their daughter, and now are dedicated to prevent similar incidents in the future” (Home visitor #5) |
| SHABI as a bridge between the hospital and the community | | In Israel, the hospital and community interfaces operate independently. Hospital management perceived SHABI as an appropriate bridge between hospital care and preventative community efforts | [+] “I think it is the connection, the connection point, between what we do in the hospital when a child arrives after a home injury, and what happens in the community” (Head of nursing) |
| Home-visitation intervention design | Home visitors perceived that at times the home tour was felt to be invasive by families | Hospital management, home visitors and families generally perceived that the visit was effective in improving home safety. This drove the home visitors to invest in the intervention The checklist helped to guide the visit and home visitors to discuss safety in each home area | [-] “I felt it (house tour) was an invasion of their privacy. I mean, if the bedroom is messy and the parent does not feel comfortable with it, then it hurts his/her ability to open up to me or listen to the things I want to say. A tour through the home has disadvantages… It can create antagonism” (Home visitor 4#) |
| Implementers' characteristics | Implementers' characteristics | Implementers' characteristics | Implementers' characteristics |
| Perception of SHABI's importance | Nurses prioritized their efforts in recruiting families arriving with fall injuries (345 of 508 families with fall injuries were recruited vs. 163 who were not recruited; χ2 = 15.3, p < 0.001) in comparison to a foreign body (58 of 119 families with foreign body were recruited vs. 61 who were not recruited; χ2 = 12.2, p < 0.001) or due to animal injuries (e.g., dog bite; 23 of 67 families with animal injuries were recruited vs. 44 who were not recruited; χ2 = 25.8, p < 0.001) | Nurses perceived SHABI as important which was a significant driver to recruiting families in the Emergency Department | [+]“The [nursing] team members need to build the passion for it [recruitment]... and it also depends on the team member. If they are passionate, it will be more successful” (Nurse #1) |
| Communication skills | Paucity of Arab speaking home visitors may have influenced communication with Arab families | While nurses and home visitors worried that SHABI's visits might be perceived as judgmental and critical, their sensitivity and explanations that injuries are common allowed constructive engagement | [+]“I explain again and again that it is not a matter of you being a bad parent. There is not a single child that goes through childhood without something happening to him. And it is good to avoid next time” (Nurse # 4) [+]“[The home visitor] was very pleasant, gave a good feeling and did not give a critical and judging feeling, but a sense of sharing and togetherness” (A Jewish mother of three preschoolers from a low SES city) |
| Home visitors' training | Home visitors had difficulty in encounters with culturally diverse families | Through training and encountering families, home visitors increased their understanding about cultural and religious groups with whom they had little familiarity | [+]“The program completely changed my stereotype. I came from the center of the country to a city like Safed, a low socio-economic city and people with a different background than mine… and it changes something. You suddenly see the person. You do not see he is ultra-Orthodox” (Home visitor #3) |
| Home visitors' awareness of potential dangers in the home | | Increasing awareness influenced home visitors to recognize potential dangers in the home from baseline and 8 months later [6 (IQR 5–7)] vs. [8 (IQR 7–8); p < 0.05] | |
| Home visitors' self-confidence in conducting the visits | | The improvement in home-visitors' self-confidence, which was low at the beginning but improved with experience, influenced the visits' effectiveness | [+] “It took a while until I learned how to conduct the conversation [first family phone call] and gain confidence. At first, I would only do it in front of the computer, with the text in front of me and only when there is no noise around me. Now I do it on the go... I initially had the challenge of my insecurity” (Home visitor #1) |
| Home visitors' skills in conducting the visits | The presence of children distracted from home visitors' ability to conduct the visit | Focusing on building relationships, rather than immediately discussing home safety, enhanced parents' engagement Involving the children in the visit kept them occupied and secured parents' attention | [+]“I kept trying to involve the kids in the visit. I say to the kids: who knows what a door stopper is' [door slamming prevention accessory]? And put it on their noses. 'Who can guess what this product does?' ” (Home visitor #3) |
SHABI was designed as a hospital-led program, and its implementation faced several barriers and likewise, facilitators. Analysis indicated that despite the compatibility between SHABI's mission in preventing child injuries and the hospital mission in increasing community health, the hospital found it difficult to operate SHABI outside of its own setting as planned as well as hiring Emergency Department nurses as home visitors. As a result, the medical school took over SHABI's operational aspects and recruited medical/nursing students as home visitors. This collaboration between the hospital and the medical school helped bridge the gap.
SHABI's implementation was facilitated by the top-down decision-making process and nurses perceived SHABI's importance in preventing child injury. Despite the inclusion of recruitment to SHABI in the initial patient triage process, it was still only partially successful. Nurses approached only $63\%$ of eligible families and failed to recruit foreign body or animal injury cases.
Medical and nursing students were recruited as home visitors. Both medical and nursing student cohorts had very few Arabic speakers and none applied for the position. This lack of Arabic speakers may have influenced attrition of Arab families, who were more likely to drop out after the first home visit than Jewish families (7 of 15 Arab families completed both visits vs. 91 of 120 Jewish families; $$p \leq 0.02$$). During SHABI's operation and following training sessions, home visitors increased their awareness of dangers at home from baseline and 8 months later [6 (IQR 5–7)] vs. [8 (IQR 7–8); $p \leq 0.05$]. They also improved their confidence in conducting home visits and enhanced their understanding of cultural and religious groups with whom they had little familiarity. Finally, children's presence in the visits often drew parents' attention, and home visitors involving them in the visit helped reduce distractions.
## Discussion
SHABI is a home-visitation program that aims to prevent unintentional childhood injuries through delivery of a hospital-based service. This study's goal was to evaluate the barriers and facilitators of implementing SHABI using the theoretical and conceptual framework of CFIR [29], exploring different stakeholders' experiences-families and implementers, to better understand the implementation process and outcomes.
Hospitals are an important setting for child injury prevention considering the high arrival and admission rates. Review of hospital-led interventions revealed only two home-visitation studies focused on home safety and injury rate (17–20), however the hospitals' responsibility and involvement remained unclear. This case study contributes to the literature by demonstrating and evaluating the ambiguity regarding the hospital's role and responsibility in implementing SHABI. In the early implementation stages the hospital expressed structural difficulties in operating SHABI outside of its setting as well as in hiring nurses as home visitors. Unlike health systems in other countries, in Israel, hospital and community care settings operate separately using different computerized documentation systems and lacking the mechanisms to mediate between the two [34, 35]. To mediate this in SHABI, the collaboration between the hospital and medical school served as a bypass for that structural barrier between hospital and community.
The use of this bypass to overcome the disconnect between the hospital and community was implemented in the ETGAR program [36] aimed at reducing recurrent admissions following discharge from hospital. ETGAR, also developed by the medical school, uses medical students to visit patients and provide guidance following discharge [36]. As demonstrated by ETGAR, there is a need for improved coordination and collaboration between hospital and community. Literature suggests that there is specific value for bridging hospital-community silos to the field of child injury prevention. Towner and Dowswell [37] reviewed child injury prevention interventions and found that collaboration between organizations can create an environment in which multiple players, such as municipalities or voluntary agencies, contribute their resources, namely knowledge, experience, or ability, and assist each other when encountering a barrier [37]. Despite the benefits of collaborations, as demonstrated in the SHABI program, the bypass created by the hospital and medical school provides only a temporary solution. The structural difficulties of hospitals' involvement in community-hospital prevention programs emphasize the need for designing a sustainable solution that will enable hospitals to become major actors actively contributing to various prevention fields.
Albeit SHABI's recruitment being successful to an extent, one of the organizational catalysts for its implementation was the hospital's top-down decision-making process. Top-down decision-making characterizes hierarchical and clinical implementation settings such as hospitals and Emergency Department [38]. Decision-making of this kind can be an influential element in implementing new programs and was found as a motivator for implementers, yet it can also lead to resistance [39, 40]. Implementers' beliefs about an intervention serve as an additional significant facilitator for implementation success [41], including staff attitudes regarding hospital-based interventions [42]. For example, Garbutt et al. [ 43] evaluated implementers' beliefs regarding a US national program for papilloma virus vaccines among at-risk girls. They found that implementers who achieved high vaccination rates were those who held a strong belief on the vaccine's importance, who felt self-efficacy and confidence in the vaccine contribution, and were personally committed to the mission. Efforts must therefore be invested in educating implementers about a program's importance in order to create a sense of ownership and achieve sustainable change. Despite some difficulties in accepting the hierarchical decision process, our nurses perceived SHABI as a valuable program, and made efforts to persuade parents to participate.
SHABI's implementers included Emergency Department nurses and home visitors comprised of medical and nursing students rather than only Emergency Department nurses as originally planned. We found that home visitors increased their self confidence in conducting home visits, as well as their awareness toward dangers at home. Along with the significant improvement found in home safety, it seems that professional qualification is not an essential component, and home visitors with adequate training do not harm the program's outcome measures. In the child injury field, several home-visitation studies have used both professional [10, 18] and non-professionals [17, 44, 45] as home visitors. Conflicting findings were found as to home safety increase and/or injury rates decrease, but the literature is unclear as to the necessity for professional qualifications. Further research is needed regarding implementers' required qualification, characteristics, and skills.
Arab families have relatively high levels of injuries in the home [31, 46] and were therefore key targets. Recruitment was lower and there was greater drop out after the first SHABI home visit. This might have been mitigated if the home visitors had among them Arab speaking students. Smithson et al. [ 27] found that a major barrier to preventing child home injuries is messages that are often not culturally adapted. However, home-visitation interventions where locals were employed as home visitors failed to show significant improvement in home safety and/or injury rate [44, 45]. Further research is needed to understand the distinctive skills and characteristics child injury prevention implementers require.
Two opposing forces act simultaneously in the implementation field [21, 22]. Fidelity is the need to maintain uniformity according to the original research protocol, compared to adaptability which is the need for protocol adaptation in new settings and contexts to increase implementation success [25, 47, 48]. In SHABI, for example, adaptive mechanisms were applied on several occasions. During SHABI's Emergency Department implementation, families' recruitment component was included as part of the patient triage process to ensure that eligible families are included. Additionally, the home visitors developed a strategy for including the children to keep parents' attention during the home visit. Yet the process changes described had no major structural implications to SHABI's core program, did not affect the programs' aspired outcome measures, but maybe increased SHABI's implementation success. Adaptive mechanisms are important as by applying them failure of the implementation process may be prevented [48].
Another aspect in child injury is the vast and diverse existing data on injury prevention. This variability is expressed in several ways, such as dangers in different home areas (kitchen vs. the bedroom), or different injury mechanisms (poisoning vs. burns). This also leads to differences in safety guidelines provided, such as improving the physical environment vs. changing parental behavior or recommending safety devices vs. moving objects out of the child's reach [28]. The variance creates diversity in research tools that evaluate effectiveness [8, 49, 50]. The lack of uniformity of injury prevention messages, measurement and evaluation tools creates difficulty in developing standards and quality indicators. This difficulty is particularly evident in the attempt to scale-up successful interventions to other settings or larger population groups [24].
Home safety checklists are a common research tool used in child injury prevention, but they generally have not undergone a validation process [17, 44, 45]. In SHABI we used a checklist developed by “Beterem” [28], which although based on the literature, has not been formally validated. The HOME inventory (The Home Observation for Measurement of the Environment) [51] appears to be the only validated tool, however only eight out of the 219 items assess home safety, while the rest examine topics such as child physical and emotional development or parent-child attachment. In SHABI we chose the “Beterem” checklist since it has been used widely in Israel. There is no doubt that there is a need to develop validated research tools and standards of quality indicators in the field of child injury.
There are several limitations to this study. The insights gained result from study of a specific clinical setting in the Israeli health system. Further studies are needed in other hospitals in Israel and beyond using various methods and theoretical frameworks in order to extend the conclusions. There was a disparity between the numbers of Arab and Jewish families included in the research population, although the figures reflect the sociodemographic of the hospital's catchment area where the population is $20\%$ Arab. Nonetheless, the lack of Arabic speakers among the home visitors may have reduced SHABI's accessibility to Arab families. It would have been of interest to explore home visitors' attitudes toward local population groups prior to the intervention particularly as the focus on cultural sensitivity was a strength and home visitors claimed that their cultural competence had increased. Lastly, due to the lack of suitable validated research tools, we developed the tools for the current research. This limitation was mitigated by triangulation of the findings from the hospital management's, implementers', and families' perspectives.
## Conclusions
This is the first time that the Implementation Science lens has been used to explore a hospital-led home-visitation intervention aimed at preventing child injury. The conceptual CFIR theoretical framework focused on the entire implementation process, hospital inner setting, SHABI's characteristic, and nurses and home visitors' characteristics. We found that a sustainable solution is needed to bridge the disconnect between the hospital and the community, so that hospitals can become a key player in preventing child injuries. Nurses and home visitors applied adaptive means to increase SHABI's implementation success in the recruitment process at the hospital and during the home visits. Finally, our work further highlights the need to further explore settings for implementing interventions using home visits to prevent child injury.
## Data availability statement
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.
## Ethics statement
Helsinki approval was obtained through the Ziv Medical Center Ethics Committee (0029-19-ZIV; 22 October 2017). Written informed consent to participate in this study was provided by the participants' legal guardian/next of kin.
## Author contributions
LS was involved in study planning and intervention design, collecting, analyzing data, and drafting the manuscript. MR was the initiator of the study, involved in study planning and intervention design, and reviewing the manuscript. SS was involved in study planning and intervention design, analysis, and reviewing the manuscript. All authors read and approved the final manuscript.
## Funding
This work was supported by Pratt Foundation, Australia. Pratt Foundation–Israel, PO Box 37052, Jerusalem, Israel.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Peden M, Oyegbite K, Ozanne-Smith J, Hyder AA, Branche C, Fazlur Rahman AKM. *World Report on Child Injury Prevention* (2008.0)
2. 2.World Health Organization. Causes of Deaths Among Children Under 5 Years. (2016). Available online at:https://www.who.int/data/gho/data/indicators/indicator-details/GHO/number-of-deaths (accessed September 30, 2021).. *Causes of Deaths Among Children Under 5 Years* (2016.0)
3. Yin X, Dai W, Du Y, Li D. **The injury mechanisms and injury pyramids among children and adolescents in Zhuhai City, China**. *BMC Public Health* (2021.0) **21** 436. DOI: 10.1186/s12889-021-10425-4
4. 4.Centers for Disease Control Prevention NC for IP C. Nonfatal unintentional injury admission rate to the Emergency Department per 100,000. (2021). Available online at: https://webappa.cdc.gov/sasweb/ncipc/nfirates.html (accessed March 2, 2021).. *Nonfatal unintentional injury admission rate to the Emergency Department per 100,000* (2021.0)
5. Johnston BD, Grossman DC, Connell FA, Koepsell TD. **High-risk periods for childhood injury among siblings**. *Pediatrics.* (2000.0) **105** 562-8. DOI: 10.1542/peds.105.3.562
6. Phelan KJ, Morrongiello BA, Khoury JC, Xu Y, Liddy S, Lanphear B. **Maternal supervision of children during their first 3 years of life: the influence of maternal depression and child gender**. *J Pediatr Psychol.* (2014.0) **39** 349-57. DOI: 10.1093/jpepsy/jst090
7. Rimsza ME, Schackner RA, Bowen KA, Marshall W. **Can child deaths be prevented? the Arizona child fatality review program experience**. *Pediatrics* (2002.0) **110** e11. DOI: 10.1542/peds.110.1.e11
8. Kendrick D, Mulvaney CA, Ye L, Stevens T, Mytton JA, Stewart-Brown S. **Parenting interventions for the prevention of unintentional injuries in childhood (Review)**. *Cochrane Database Syst Rev* (2013.0) **2013** CD006020. DOI: 10.1002/14651858.CD006020.pub3
9. Johnston BD, Huebner CE, Tyll LT, Barlow WE, Thompson RS. **Expanding developmental and behavioral services for newborns in primary care: effects on parental well-being, practice, and satisfaction**. *Am J Prev Med.* (2004.0) **26** 356-66. DOI: 10.1016/j.amepre.2003.12.018
10. Kendrick D, Marsh P, Fielding K, Miller P. **General practice-preventing injuries in children: cluster randomized controlled trail in primary care**. *BMJ.* (1999.0) **318** 980-3. DOI: 10.1136/bmj.318.7189.980
11. Llewellyn G, McConnell D, Honey A, Mayes R, Russo D. **Promoting health and home safety for children of parents with intellectual disability: a randomized controlled trial**. *Res Dev Disabil.* (2003.0) **24** 405-31. DOI: 10.1016/j.ridd.2003.06.001
12. Beckett K, Goodenough T, Deave T, Jaeckle S, McDaid L, Benford P. **Implementing an injury prevention briefing to aid delivery of key fire safety messages in UK children's centres: qualitative study nested within a multi-centre randomised controlled trial**. *BMC Public Health.* (2014.0) **14** 1256. DOI: 10.1186/1471-2458-14-1256
13. Sijm-Eeken M, Zheng J, Peute L. **Towards a lean process for patient journey mapping-a case study in a large academic setting**. *Stud Health Technol Inform.* (2020.0) **270** 1071-5. DOI: 10.3233/SHTI200326
14. Hommos MS, Kuperman EF, Kamath A, Kreiter CD. **The development and evaluation of a novel instrument assessing residents' discharge summaries**. *Acad Med.* (2017.0) **92** 550-5. DOI: 10.1097/ACM.0000000000001450
15. Vo D, Zurakowski D, Faraoni D. **Incidence and predictors of 30-day postoperative readmission in children**. *Paediatr Anaesth.* (2018.0) **28** 63-70. DOI: 10.1111/pan.13290
16. Prieto-Centurion V, Markos MA, Ramey NI, Gussin HA, Nyenhuis SM, Joo MJ. **Interventions to reduce rehospitalizations after chronic obstructive pulmonary disease exacerbations: a systematic review**. *Ann Am Thorac Soc.* (2014.0) **11** 417-24. DOI: 10.1513/AnnalsATS.201308-254OC
17. King WJ, Klassen TP, LeBlanc J, Bernard-Bonnin AC, Robitaille Y, Pham B. **The effectiveness of a home visit to prevent childhood injury**. *Pediatrics.* (2001.0) **108** 382-8. DOI: 10.1542/peds.108.2.382
18. Gielen AC, McDonald EM, Wilson MEH, Hwang WT, Serwint JR, Andrews JS. **Effects of improved access to safety counseling, products, and home visits on parents' safety practices: results of a randomized trial**. *Arch Pediatr Adolesc Med.* (2002.0) **156** 33-40. DOI: 10.1001/archpedi.156.1.33
19. Armstrong KL, Fraser JA, Dadds MR, Morris J. **Promoting secure attachment, maternal mood and child health in a vulnerable population: a randomized controlled trial**. *J Paediatr Child Health.* (2000.0) **36** 555-62. DOI: 10.1046/j.1440-1754.2000.00591.x
20. Kemp L, Harris E, McMahon C, Matthey S, Vimpani G, Anderson T. **Child and family outcomes of a long-term nurse home visitation programme: a randomised controlled trial**. *Arch Dis Child.* (2011.0) **96** 533-40. DOI: 10.1136/adc.2010.196279
21. Colditz GA, Emmons KM. **The promise and challenges of dissemination and implementation research. In: Brownson RC, Colditz GA, Proctor EK, editors**. *Dissemination Research in Health: Translating Science to Practice Research in Health* (2018.0) 1-18
22. Owen N, Goode A, Sugiyama T, Koohsari MJ, Healy G, Fjeldsoe B. **Designing for dissemination in chronic disease prevention and management. In: Brownson RC, Colditz GA, Proctor EK, editors**. *Dissemination and Implementation Research in Health: Translating Science to Practice* (2018.0) 107-120
23. 23.World Health Organization. A Guide to Implementation Research in the Prevention and Control of Noncommunicable Diseases (2016).. *A Guide to Implementation Research in the Prevention and Control of Noncommunicable Diseases* (2016.0)
24. 24.World Health Organization. Practical Guidance for Scaling Up Health Service Innovations. Geneva: World Health Organization (2009).. *Practical Guidance for Scaling Up Health Service Innovations* (2009.0)
25. Pérez D, van der Stuyft P, Zabala MC, Castro M, Lefèvre P. **A modified theoretical framework to assess implementation fidelity of adaptive public health interventions**. *Implement Sci.* (2016.0) **11** 1-11. DOI: 10.1186/s13012-016-0457-8
26. Nicks SE, Weaver NL, Recktenwald A, Jupka KA, Elkana M, Tompkins R. **Translating an evidence-based injury prevention program for implementation in a home visitation setting**. *Health Promot Pract* (2016.0) **17** 578-85. DOI: 10.1177/1524839915622196
27. Smithson J, Garside R, Pearson M. **Barriers to, and facilitators of, the prevention of unintentional injury in children in the home: a systematic review and synthesis of qualitative research**. *Inj Prev.* (2010.0) **17** 119-26. DOI: 10.1136/ip.2010.026989
28. 28.Obtained by “Beterem-Safe Kids Israel” (2021). Available online at: https://www.beterem.org/home/. (2021.0)
29. Damschroder LJ, Aron DC, Keith RE, Kirsh SR, Alexander JA, Lowery JC. **Fostering implementation of health services research findings into practice: a consolidated framework for advancing implementation science**. *Implement Sci.* (2009.0) **4** 1-15. DOI: 10.1186/1748-5908-4-50
30. **In Hebrew: Municipalities in Israel: Data files for processing in 2018**
31. Falek A, Orr D, Kalif E. *In Hebrew: Child Injuries in Israel: Beterem National Report 2020* (2020.0)
32. Shalev L, Luder A, Spitzer S, Krupik D, Essa-Hadad J, Rudolf MCJ. **Keeping our children safe: piloting a hospital-based home-visitation program in Israel**. *Isr J Health Policy Res.* (2022.0) **11** 21. DOI: 10.1186/s13584-022-00525-w
33. Ivankova NV, Creswell JW, Stick SL. **Using mixed-methods- sequential explanatory design: From theory to practice**. *Field Methods.* (2006.0) **18** 3-20. DOI: 10.1177/1525822X05282260
34. Rudolf MCJ, Bord S, Hasson R, Sahar Y, Rubin L, Manor N. **Between-country analysis of implementing an obesity prevention intervention using RE-AIM: HENRY in Israel and UK**. *Health Promot Int* **2021** 1-43. DOI: 10.1093/heapro/daab119
35. Admi H, Muller E, Ungar L, Reis S, Kaffman M, Naveh N. **Hospital-community interface: a qualitative study on patients with cancer and health care providers' experiences**. *Eur J Oncol Nurs.* (2013.0) **17** 528-35. DOI: 10.1016/j.ejon.2013.02.005
36. Sagi D, Spitzer-Shohat S, Schuster M, Daudi L, Rudolf MCJ. **Teaching plain language to medical students: Improving communication with disadvantaged patients**. *BMC Med Educ* (2021.0) **21** 407. DOI: 10.1186/s12909-021-02842-1
37. Towner E, Dowswell T. **Community-based childhood injury prevention interventions: what works?**. *Health Promot Int.* (2002.0) **17** 273-84. DOI: 10.1093/heapro/17.3.273
38. Klein R. **Performance evaluation and the NHS: a case study in conceptual perplexity and organizational complexity**. *Public Adm.* (1982.0) **60** 385-407. DOI: 10.1111/j.1467-9299.1982.tb00490.x
39. Flynn R, Rotter T, Hartfield D, Newton AS, Scott SD. **A realist evaluation to identify contexts and mechanisms that enabled and hindered implementation and had an effect on sustainability of a lean intervention in pediatric healthcare**. *BMC Health Serv Res* (2019.0) **19** 912. DOI: 10.1186/s12913-019-4744-3
40. Mwendera CA, De Jager C, Longwe H, Kumwenda S, Hongoro C, Phiri K. **Challenges to the implementation of malaria policies in Malawi**. *BMC Health Serv Res* (2019.0) **19** 194. DOI: 10.1186/s12913-019-4032-2
41. Rogers EM. *Diffusion of Innovations 5th Edition. 5th ed.* (2003.0)
42. Geerligs L, Rankin NM, Shepherd HL, Butow P. **Hospital-based interventions: a systematic review of staff-reported barriers and facilitators to implementation processes**. *Implement Sci* (2018.0) **13** 36. DOI: 10.1186/s13012-018-0726-9
43. Garbutt JM, Dodd S, Walling E, Lee AA, Kulka K, Lobb R. **Theory-based development of an implementation intervention to increase HPV vaccination in pediatric primary care practices**. *Implement Sci* (2018.0) **13** 45. DOI: 10.1186/s13012-018-0729-6
44. Swart L, van Niekerk A, Seedat M, Jordaan E. **Paraprofessional home visitation program to prevent childhood unintentional injuries in low-income communities: a cluster randomized controlled trial**. *Inj Prev.* (2008.0) **14** 164-9. DOI: 10.1136/ip.2007.016832
45. Odendaal W, van Niekerk A, Jordaan E, Seedat M. **The impact of a home visitation programme on household hazards associated with unintentional childhood injuries: a randomised controlled trial**. *Accid Anal Prev.* (2009.0) **41** 183-90. DOI: 10.1016/j.aap.2008.10.009
46. Ittai S, Gad BJ, Naim S, David F, Vardit J, Moshe R. **Hospitalizations due to falls in Jewish and Arab children in northern Israel**. *Eur J Epidemiol.* (2000.0) **16** 47-52. DOI: 10.1023/A:1007683020026
47. Nahum-Shani I, Qian M, Almirall D, Pelham WE, Gnagy B, Fabiano GA. **Experimental design and primary data analysis methods for comparing adaptive interventions**. *Psychol Methods.* (2012.0) **17** 457-77. DOI: 10.1037/a0029372
48. Anyon Y, Roscoe J, Bender K, Kennedy H, Dechants J, Begun S. **Reconciling adaptation and fidelity: implications for scaling up high quality youth programs**. *J Prim Prev.* (2019.0) **40** 35-49. DOI: 10.1007/s10935-019-00535-6
49. Hubbard S, Cooper N, Kendrick D, Young B, Wynn PM, He Z. **Network meta-analysis to evaluate the effectiveness of interventions to prevent falls in children under age 5 years**. *Inj Prev.* (2015.0) **21** 98-108. DOI: 10.1136/injuryprev-2013-041135
50. Wynn PM, Zou K, Young B, Majsak-Newman G, Hawkins A, Kay B. **Prevention of childhood poisoning in the home: overview of systematic reviews and a systematic review of primary studies**. *Int J Inj Contr Saf Promot.* (2016.0) **23** 3-28. DOI: 10.1080/17457300.2015.1032978
51. Linver MR, Brooks-Gunn J, Cabrera N. **The home observation for measurement of the environment (HOME) inventory: the derivation of conceptually designed subscales**. *Parenting.* (2004.0) **4** 99-114. DOI: 10.1080/15295192.2004.9681266
|
---
title: 'Socioeconomic Inequity in the Screening and Treatment of Hypertension in Kenya:
Evidence From a National Survey'
authors:
- Robinson Oyando
- Edwine Barasa
- John E. Ataguba
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012826
doi: 10.3389/frhs.2022.786098
license: CC BY 4.0
---
# Socioeconomic Inequity in the Screening and Treatment of Hypertension in Kenya: Evidence From a National Survey
## Abstract
### Background
Non-communicable diseases (NCDs) account for $50\%$ of hospitalisations and $55\%$ of inpatient deaths in Kenya. Hypertension is one of the major NCDs in Kenya. Equitable access and utilisation of screening and treatment interventions are critical for reducing the burden of hypertension. This study assessed horizontal equity (equal treatment for equal need) in the screening and treatment for hypertension. It also decomposed socioeconomic inequalities in care use in Kenya.
### Methods
Cross-sectional data from the 2015 NCDs risk factors STEPwise survey, covering 4,500 adults aged 18–69 years were analysed. Socioeconomic inequality was assessed using concentration curves and concentration indices (CI), and inequity by the horizontal inequity (HI) index. A positive (negative) CI or HI value suggests a pro-rich (pro-poor) inequality or inequity. Socioeconomic inequality in screening and treatment for hypertension was decomposed into contributions of need [age, sex, and body mass index (BMI)] and non-need (wealth status, education, exposure to media, employment, and area of residence) factors using a standard decomposition method.
### Results
The need for hypertension screening was higher among poorer than wealthier socioeconomic groups (CI = −0.077; $p \leq 0.05$). However, wealthier groups needed hypertension treatment more than poorer groups (CI = 0.293; $p \leq 0.001$). Inequity in the use of hypertension screening (HI = 0.185; $p \leq 0.001$) and treatment (HI = 0.095; $p \leq 0.001$) were significantly pro-rich. Need factors such as sex and BMI were the largest contributors to inequalities in the use of screening services. By contrast, non-need factors like the area of residence, wealth, and employment status mainly contributed to inequalities in the utilisation of treatment services.
### Conclusion
Among other things, the use of hypertension screening and treatment services in Kenya should be according to need to realise the Sustainable Development Goals for NCDs. Specifically, efforts to attain equity in healthcare use for hypertension services should be multi-sectoral and focused on crucial inequity drivers such as regional disparities in care use, poverty and educational attainment. Also, concerted awareness campaigns are needed to increase the uptake of screening services for hypertension.
## Introduction
Non-communicable disease-related morbidity and mortality pose an increasing challenge globally, especially in low-and middle-income countries (LMICs), where most of the world's population live [1]. In 2016, for instance, ~40 million deaths globally were due to non-communicable diseases (NCDs) with LMICs accounting for $80\%$ of the deaths [2]. LMICs also continue to struggle in containing the relatively high disease burden from maternal and child ill-health and infectious diseases such as HIV/AIDS, tuberculosis, leading to a “double burden” of communicable and NCDs [3, 4]. This not only poses further resource constraints to the already overstretched healthcare resources in LMICs but is also a threat to the attainment of equity in health and healthcare between and within countries (5–7).
The major NCDs—cardiovascular diseases (CVDs), cancers, chronic respiratory diseases and diabetes—present a unique challenge to the global health agenda of attaining universal health coverage (UHC)1 by 2030 [1]. Furthermore, the detrimental health, economic, and developmental consequences of NCDs have seen their inclusion in the 2030 Sustainable Development Goals (SDGs) [8]. SDG 3.4 explicitly aims to reduce by one-third premature mortality due to NCDs through prevention and treatment. Prioritising the reduction of shared NCDs risk factors such as physical inactivity, unhealthy diets, use of tobacco, and harmful use of alcohol is imperative to achieve the SDGs [9]. Similarly, for hypertension which is a major risk factor for CVDs [1], evidence shows that increasing access to preventive interventions such as timely screening among those at risk and providing treatment to those diagnosed are cost-effective measures of attaining the NCD pre-mature mortality target (10–12).
A well-functioning health system should ensure equity in the utilisation of health services, that is, based on need and not the ability to pay [13, 14]. Yet, there is convincing evidence that the poor (who bear the greatest NCDs burden and are most in need of screening and treatment) relative to the rich, utilise NCDs healthcare services the least (15–19). This phenomenon is termed the inverse care law [20]. Demand and supply-side factors such as high levels of poverty, the substantial economic burden associated with the long-term care of NCDs, and insufficient health system capacity to handle NCDs (chiefly at the primary care level) are some of the reasons that contribute to socioeconomic inequalities in NCDs (1, 3, 15, 17, 21–23).
Empirical evidence from previous studies that assessed inequity and socioeconomic inequality in hypertension converge to the same conclusion: that the poor, relative to the wealthy, utilise fewer hypertension services [15, 17, 18, 24, 25]. Elwell-Sutton et al. [ 15], for instance, showed marked pro-rich inequality in the utilisation of treatment services for hypertension and dyslipidaemia in China. In addition, pro-rich horizontal inequity in the utilisation of hypertension, hyperglycaemia and dyslipidaemia treatment were reported in the same study [15]. Of interest, income and other non-need factors (i.e., health insurance, education and longest-held occupation) mainly explained the observed inequality in NCDs treatment. These findings compare well with studies from other LMICs, which generally show that income, area of residence, level of education, occupational class, increasing age and lifestyle risk factors are significant contributors to the socioeconomic inequality in the prevalence or utilisation of NCD services [18, 25, 26].
In Kenya, NCDs account for $50\%$ of hospitalisations and $55\%$ of inpatient deaths, with estimates indicating that mortality due to NCDs is likely to increase by over $50\%$ in the next decade [27]. Besides, there are stark disparities in screening and treatment services utilisation for hypertension, mainly to the disadvantage of poorer socioeconomic groups. For example, $73\%$ of the poorest quintile population have never been screened for hypertension compared to $38\%$ in the richest quintile [27]. Furthermore, a study in Kenya that estimated socioeconomic inequalities in hypertension prevalence found that the poor bore the highest burden, with body mass index (BMI), wealth status and education level mainly explaining the observed inequalities [24].
Although there is evidence suggesting inequalities in NCDs in Kenya, there is still a gap in knowledge, especially in assessing horizontal equity (i.e., equal treatment for equal need) in utilising screening and treatment for NCDs based on need. Also, to our knowledge, no study has assessed the factors contributing to socioeconomic inequality in using both interventions for hypertension in the Kenyan context. Therefore, using a nationally representative NCDs risk factors survey data set, this study aims to assess horizontal inequity in the screening and treatment of hypertension and decompose socioeconomic inequalities in the screening and treatment of hypertension in Kenya.
## Data
This paper used the most recent and nationally representative cross-sectional STEPwise survey (STEPs) conducted by the Kenya National Bureau of Statistics (KNBS) between April and June 2015 in the country's 47 counties [27]. The survey used the fifth National Sample Survey and Evaluation Programme (NASSEP V) master sample frame developed by the KNBS. The sample frame was developed using the Enumeration Areas (EAs) generated from the 2009 Kenya Population and Housing Census to form 5,360 clusters split into four equal sub-samples. A three-stage cluster sample design was used to collect the STEPs data. In the first stage, a total of 200 clusters (100 rural and 100 urban) were selected systematically from the NASSEP V sampling frame using the equal probability selection method to ensure the resulting sample retained the properties of probability proportional to size as was used in the creation of the frame. The second sampling stage involved a uniform selection of 30 households from the listed households in each cluster. An eligible participant was randomly selected from listed household members in the third sampling stage [27].
A total of 6,000 households were identified, but 4,754 consented (i.e. $79.2\%$ response rate) and participated in the study. A total of 4,500 households were retained after data cleaning. A more detailed description of the STEPs data collection methodology is contained elsewhere [27]. Sample weights were included in the statistical analyses to ensure that estimates were nationally representative. De-identified data set from the STEPs survey (which is available upon request from KNBS) was used in this study. Additionally, ethics clearance was obtained from the Human Research Ethics Committee of the University of Cape Town (Ref: $\frac{186}{2020}$).
## Measuring Socioeconomic Status
Socioeconomic status (SES) can be measured using several approaches classified as “direct measures,” that is, expenditure, income, consumption; and “proxy measures,” including education, occupation or social class, but mainly asset indices [28]. It is important to note that there are debates on the right choice of SES measure regarding health inequality assessment. Some argue that the choice of welfare measure may not overly affect inequality findings [29, 30] while others maintain that the computed health inequality results could be sensitive, in some contexts, to the choice of welfare measure [28, 31]. Following similar studies [17, 32] and based on data availability, principal component analysis (PCA) [33] was used in this paper to generate an index of SES.
Briefly, the multivariable statistical approach (PCA) reduces the number of variables in a data set into smaller dimensions [34]. Put another way, beginning with an initial set of correlated variables, PCA generates uncorrelated components, in which case each component or index is a linear weighted combination of the original variables [33]. The first principal component provides what is needed to construct a household welfare index–if it explains a substantial proportion of the variance, with larger weights assigned to assets that vary most across households [29, 34]. Data on 15 selected variables (e.g. source of drinking water, type of sanitary facility, roof, floor and wall material, source of energy for cooking and lighting, and ownership of TV, radio, refrigerator, washing machine, bicycle, motorcycle, landline, and cell phone) were used to generate standardised weighted scores. These variables were used to create a dummy of each variable, signifying the presence of each item given that categorical variables are converted into a meaningless scale in PCA [35]. The composite weighted index was used to rank the sample into five wealth quintiles (1—poorest, 5—richest).
## Defining Hypertension
Having hypertension was defined in this paper based on any or all the three criteria: [1] previous hypertension diagnosis by a health worker, [2] use of prescribed anti-hypertensive medication or [3] having a systolic and/or diastolic blood pressure of ≥ 140/≥ 90 mmHg [36].
## Measuring Need and Use of Hypertension Screening and Treatment
The need for hypertension screening was defined as individuals who smoke, are obese (≥ 30 kg/m2) and are 30 years and above (for both men and women), as stipulated in Kenya's cardiovascular treatment guidelines [36]. The need for hypertension treatment was defined as those diagnosed with hypertension in the survey (i.e. a third systolic and/or diastolic blood pressure of ≥ 140/≥ 90 mmHg, respectively) [36].
The utilisation of screening services was assessed as having ever received a screening service for hypertension from a formal health provider (i.e. doctor or other health workers) before the survey. Similarly, utilisation of treatment was assessed as taking prescribed hypertension treatment two weeks before the survey. For a granular presentation of inequality findings, the share of need and use of screening and treatment interventions were compared across the SES groups and regional divides in Kenya. Table 1 further summarises the definitions of variables used in the analysis.
**Table 1**
| Intervention | NCD | Need | Use |
| --- | --- | --- | --- |
| Screening | Hypertension | Respondents who are obese (≥30 kg/m2), smoke and are aged at least 30 years and have not been screened in the past | Respondents reporting ever screened by a health worker |
| Treatment | Hypertension | Respondents diagnosed with hypertension in the survey (i.e. systolic and/or diastolic blood pressure reading ≥ 140 mmHg or ≥ 90 mmHg) | Respondents reporting the use of prescribed anti-hypertensive treatment at least two weeks before the survey |
## Measuring Inequality in Care Utilisation
Inequality in screening and treatment can be assessed using various methodological approaches, as discussed by Wagstaff et al. [ 37]. This paper used the concentration curves and concentration indices to assess inequality in the screening and treatment of hypertension. The rationale for using these measures is their consistency in ranking individuals according to their SES; sensitivity to changes in population distribution across SES, and ability to assess relative vs. absolute inequality (37–39). The concentration curve (CC) plots the cumulative share of the use of screening or treatment services (y-axis) against the cumulative share of households, ranked from poorest to richest (x-axis). So, if everyone uses screening or treatment services irrespective of their SES rank, the CC will consistently lie on the equality (45-degree) line. If, by contrast, there is a pro-poor (pro-rich) distribution in the use of screening or treatment services, the CC will lie above (below) the line of equality, with the gap between the CC and equality line depicting the extent of inequality [40].
The concentration index (CIH) was computed as twice the covariance between screening or treatment for hypertension and an individual's socioeconomic rank divided by the mean of the health variable. Theoretically, the CIH lies between −1 (i.e. when the use of screening/treatment is concentrated on the poorest individual) and +1 (i.e. when the use of screening/treatment is concentrated on the richest individual). Overall, a positive (negative) CIH corresponds to a pro-rich (pro-poor) distribution. For a binary variable, the concentration index does not lie within the usual bounds but rather between (μH - 1) and (1- μH) and thus requires normalisation [41]. Although there is a debate between Wagstaff (41–43) and Erreygers [44, 45] regarding the appropriate normalisation approach, this paper used the Wagstaff's [41] normalisation primarily because the health variable of interest was binary (i.e. 1 = use of screening/treatment; 0 = otherwise).
## Decomposing the Concentration Index of Screening and Treatment
While the CC and the CIH are relevant in examining the existence of socioeconomic inequalities in screening/treatment; they do not explain the factors contributing to observed inequality. Consequently, to understand the factors contributing to relative inequality, the CIH was further decomposed following the methodology suggested by Wagstaff et al. [ 46]. Identifying these factors is critical for policy decisions around addressing the “underlying causes of inequality.” Thus, CIH can be decomposed as: where Cj (Z¯j) is the concentration index (mean) of the jth contributing factor, GCε is the generalised concentration index of the error term (ε) and βj is obtained from the linearly additive equation related to the contributing factors (z) to the screening or treatment variable (h) shown in Equation 2.
where α and βj are the coefficients to be estimated and εi is the error term. The deterministic portion of the concentration index in Equation 1 can be interpreted as the contribution of each contributing factor (z) to the concentration index (CIH), which consists of two parts. It is a product of the concentration index of each contributing factor (Cj) and the elasticity of hi with respect to zj(i.e. NGj= βjZ¯j/μH). The unexplained portion is computed as the difference between CIH and the deterministic portion. The residual cannot be systematically explained by variations in the contributing factors across socioeconomic groups [46]. *The* generalised linear model (with binomial family and identity link) was applied in the decomposition analysis [47]. Guided by variable availability in the dataset and well-established literature in the field (15, 24–26, 48, 49), determinants of care utilisation were separated into “need” (i.e. body mass index (BMI), age and sex for screening; age and sex only for treatment) and “non-need” (i.e. education level, exposure to media, employment status, rural or urban residence, and quintiles of SES) factors for both screening and treatment. A negative (positive) contribution suggests a given determinant contributes to inequality in the pro-poor (pro-rich) direction.
Given the challenge in computing analytical standard errors (SEs) for the components in the decomposition (i.e. elasticities and each contributing factor's contribution to the concentration index) in Equation 1, the bootstrap method [50, 51] was used to obtain such SEs in the analysis. The sampling structure of the data was taken into account as applied by Ataguba et al. [ 52] to avoid inconsistent estimates of bootstrap SEs. A total of 1,000 replications were used to estimate the SEs for each estimate.
## Measuring Horizontal Inequity in Care Utilisation
Horizontal equity analysis assesses inequity in care utilisation by standardising health service utilisation based on need [40]. Inequity in care use estimated through the horizontal inequity (HI) index embodies the principle that healthcare should be utilised according to need (i.e. “equal treatment for equal need”). The HI was computed as the difference between the concentration index for actual (observed) care utilisation and need-expected utilisation. An indirect standardisation approach was used to predict the need-expected utilisation of screening and treatment [40, 53]. HI lies within the range of −1 to +1, with a negative (positive) value indicating a pro-poor (pro-rich) inequity. Theoretically, a zero value for HI means there is no inequity. To estimate how much care each individual would receive if they were treated equally to everyone in the sample with equal needs, we fitted a regression model [40]. All statistical analyses were conducted in Stata (version 15.1).
## Descriptive Analysis
Most respondents were female ($60\%$). About $30\%$ aged 20–39 years and $47\%$ had attained primary education (Table 2). Only $19\%$ of respondents were not employed, and more than half ($54\%$) resided in a rural area. Hypertension prevalence was $30\%$, with a higher prevalence among wealthier individuals (Table 2). In addition, the prevalence of hypertension was higher among obese individuals ($52\%$) compared to other BMI categories (underweight $19\%$, normal $26\%$, and overweight $39\%$) (data not shown). Although there was no significant difference, the prevalence of hypertension was slightly higher among non-smokers ($30\%$) compared to non-smokers ($29\%$) (data not shown).
**Table 2**
| Unnamed: 0 | All | Household socioeconomic groups | Household socioeconomic groups.1 | Household socioeconomic groups.2 | Household socioeconomic groups.3 | Household socioeconomic groups.4 | p- value* |
| --- | --- | --- | --- | --- | --- | --- | --- |
| | Col % | Poorest quintile | 2nd quintile | 3rd quintile | 4th quintile | Richest quintile | |
| N | 4500 | 918 | 891 | 899 | 909 | 883 | |
| Sex | Sex | Sex | Sex | Sex | Sex | Sex | Sex |
| Female | 60.0 | 65.2 | 60.4 | 58.4 | 61.1 | 54.8 | <0.01 |
| Male | 40.0 | 34.8 | 39.6 | 41.6 | 38.9 | 45.2 | |
| Age (years) | Age (years) | Age (years) | Age (years) | Age (years) | Age (years) | Age (years) | Age (years) |
| <19 | 4.5 | 5.0 | 5.5 | 4.4 | 4.2 | 3.5 | <0.01 |
| 20–29 | 28.6 | 24.3 | 25.7 | 26.2 | 30.1 | 36.8 | |
| 30–39 | 28.0 | 28.7 | 29.4 | 27.9 | 24.6 | 29.2 | |
| 40–49 | 17.7 | 16.8 | 15.9 | 20.0 | 19.0 | 16.6 | |
| 50–59 | 12.1 | 13.5 | 13.1 | 12.0 | 12.8 | 8.9 | |
| 60+ | 9.2 | 11.6 | 10.5 | 9.5 | 9.3 | 5.0 | |
| Education level | Education level | Education level | Education level | Education level | Education level | Education level | Education level |
| No formal schooling | 16.8 | 50.9 | 11.0 | 9.1 | 8.5 | 3.4 | <0.01 |
| Primary | 46.5 | 40.0 | 63.3 | 59.1 | 45.3 | 24.9 | |
| Secondary | 25.5 | 8.2 | 22.0 | 25.5 | 33.0 | 39.1 | |
| Tertiary | 11.2 | 0.9 | 3.7 | 6.3 | 13.2 | 32.6 | |
| Marital status | Marital status | Marital status | Marital status | Marital status | Marital status | Marital status | Marital status |
| Married/Cohabiting | 67.9 | 72.6 | 65.2 | 68.2 | 70.0 | 63.1 | <0.01 |
| Not married | 32.1 | 27.4 | 34.8 | 31.8 | 30.0 | 36.9 | |
| Exposure to media | Exposure to media | Exposure to media | Exposure to media | Exposure to media | Exposure to media | Exposure to media | Exposure to media |
| Has TV/Radio | 70.3 | 30.7 | 65.6 | 78.5 | 70.3 | 95.4 | |
| No TV/Radio | 29.7 | 69.3 | 34.4 | 21.5 | 29.7 | 4.6 | <0.01 |
| Employment status | Employment status | Employment status | Employment status | Employment status | Employment status | Employment status | Employment status |
| Unemployed | 18.5 | 3.6 | 9.2 | 15.5 | 24.8 | 40.1 | <0.01 |
| Informal employment | 39.7 | 33.9 | 43.0 | 43.9 | 39.4 | 38.4 | |
| Formal employment | 41.8 | 62.5 | 47.8 | 40.6 | 35.8 | 21.5 | |
| Residence | Residence | Residence | Residence | Residence | Residence | Residence | Residence |
| Rural | 53.7 | 77.9 | 76.5 | 56.3 | 42.5 | 14.3 | <0.01 |
| Urban | 46.3 | 22.1 | 23.5 | 43.7 | 57.5 | 85.7 | |
| BMI (Kg/m 2 ) | BMI (Kg/m 2 ) | BMI (Kg/m 2 ) | BMI (Kg/m 2 ) | BMI (Kg/m 2 ) | BMI (Kg/m 2 ) | BMI (Kg/m 2 ) | BMI (Kg/m 2 ) |
| <18.5 (underweight) | 11.7 | 27.3 | 10.4 | 9.0 | 6.8 | 4.8 | <0.01 |
| 18.5–24.9 (normal) | 56.8 | 60.8 | 66.0 | 59.4 | 54.8 | 42.7 | |
| 25.0–29.9 (overweight) | 21.0 | 9.5 | 17.6 | 22.6 | 22.8 | 32.6 | |
| ≥30.0 (obese) | 10.5 | 2.4 | 6.0 | 9.0 | 15.6 | 19.9 | |
| Presence of NCD | Presence of NCD | Presence of NCD | Presence of NCD | Presence of NCD | Presence of NCD | Presence of NCD | Presence of NCD |
| Has hypertension | 30.2 | 24.7 | 28.4 | 30.9 | 32.7 | 34.1 | |
## Inequality in Need and Use of Hypertension Screening and Treatment Services
Poorer individuals had a higher need for hypertension screening (concentration curves lie above the equality line), while hypertension screening favoured the rich (concentration curves lie below the equality line) (Figure 1). Although not significant, the need for hypertension screening was pro-poor (CI = −0.036; $p \leq 0.05$). On the other hand, hypertension screening (CI = 0.293; $p \leq 0.01$) was significantly pro-rich (Table 3).
**Figure 1:** *Concentration curves showing need and use of hypertension (A) screening and (B) treatment in Kenya.* TABLE_PLACEHOLDER:Table 3 A further breakdown of who benefits from hypertension screening revealed that the wealthier quintiles benefited disproportionately more than they should given their share of need (Figure 2A). For example, while only $17\%$ of those needing hypertension screening are in the wealthiest quintile, $27\%$ of individuals using screening interventions are in the wealthiest quintile (Figure 2A). There were disparities in the need and use of hypertension screening in all the regions (Figure 2B). For instance, the disparity between the share of need and use for hypertension screening was highest in the Rift Valley region ($27\%$ vs. $31\%$) (Figure 2B).
**Figure 2:** *Distribution of share of need and use of hypertension screening services by (A) wealth quintile and (B) regions in Kenya (STEPs 2015).*
Figure 1B shows a pro-rich distribution of the need for hypertension treatment, a finding confirmed by the concentration indices in Table 3. The use of hypertension treatment was pro-rich (CI = 0.030; $p \leq 0.05$) (Table 3). However, none of the pro-rich or the pro-poor inequality findings for hypertension treatment was statistically significant at conventional levels. Individuals in the poorest quintile were disadvantaged in using hypertension treatment compared to their population share of need. For instance, while $17\%$ of respondents needing hypertension treatment are in the poorest quintile, only $8\%$ of those using hypertension treatment are in the poorest quintile (Figure 3A). Overall, a disproportionate share exists in using hypertension treatment in the Kenyan regions, given the population share of need. Notably, the disparity in the need and use of hypertension ($28\%$ vs. $20\%$) treatment was highest in the Rift Valley region (Figure 3B).
**Figure 3:** *Distribution of share of need and use of hypertension treatment services by (A) wealth quintile and (B) regions in Kenya (STEPs 2015).*
## Inequity in Using Screening and Treatment Services for Hypertension in Kenya
The use of hypertension screening was significantly pro-rich after controlling for need (HI = 0.185; $p \leq 0.001$). Also, the use of hypertension treatment services was significantly pro-rich at conventional levels (Table 4).
**Table 4**
| Intervention | Horizontal equity index |
| --- | --- |
| Hypertension screening | 0.185*** (0.024) |
| Hypertension treatment | 0.095*** (0.074) |
## Decomposition of Inequality in Care Use
Summary results of the decomposition analysis of inequality in screening and treatment are presented in Figures 4, 5, showing an aggregate contribution of need and non-need factors. *In* general, non-need factors contributed most to the pro-rich inequality in screening and treatment. Specifically, wealth status, exposure to media, education, and area of residence contributed most to inequality in screening among the non-need factors for hypertension. For the need factors, BMI explained inequality in the pro-rich direction for hypertension screening. However, sex explained inequality in the pro-poor direction for hypertension screening (Figure 4).
**Figure 4:** *Summary contributions to SES inequality in the utilisation of screening services for hypertension in Kenya (STEPs 2015).* **Figure 5:** *Summary contributions to SES inequality in the utilisation of treatment services for hypertension in Kenya (STEPs 2015).*
Non-need factors like wealth and employment status were the largest contributors (in the pro-rich direction) to the inequality in hypertension treatment (Figure 5). Additionally, age and education status also contributed to inequality in hypertension treatment in the pro-poor direction (Figure 5).
As shown in Table 5, sex was the main statistically significant contributor to inequality in hypertension screening among the need factors. Only a few categories are significant for some need factors like age and BMI. Among non-need factors, exposure to media was a statistically significant contributor (in the pro-rich direction) to inequality in hypertension screening. For hypertension treatment, age explained the inequality in the pro-poor direction. Among the non-need factors, wealth status explained, to a greater extent, inequality in hypertension (in the pro-rich direction) treatment (Table 5). However, none of the need and non-need contributors to inequality hypertension treatment was statistically significant at conventional levels.
**Table 5**
| Unnamed: 0 | Hypertension screening | Hypertension screening.1 | Hypertension screening.2 | Hypertension screening.3 | Hypertension screening.4 | Hypertension treatment | Hypertension treatment.1 | Hypertension treatment.2 | Hypertension treatment.3 | Hypertension treatment.4 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Determinants | Elasticity | CI | Contribution | Total contribution | % contribution | Elasticity | CI | Contribution | Total contribution | % Contribution |
| BMI | BMI | BMI | BMI | BMI | BMI | BMI | BMI | BMI | BMI | BMI |
| Underweight (Ref) | | | | | | | | | | |
| Normal | −0.004 | −0.155*** (0.014) | 0.001 (7.976) | 0.018 | 6.068 | | | | | |
| Overweight | 0.026 | 0.255*** (0.032) | 0.007 (0.004) | | | | | | | |
| Obese | 0.025 | 0.398*** (0.036) | 0.010** (0.002) | | | | | | | |
| Sex | Sex | Sex | Sex | Sex | Sex | Sex | Sex | Sex | Sex | Sex |
| Male (Ref) | | | | | | | | | | |
| Female | 0.400 | −0.078** (0.017) | −0.031** (0.006) | −0.031 | −11.023 | 0.203 | −0.078** (0.044) | −0.016 (1.110) | −0.016 | −19.231 |
| Age | Age | Age | Age | Age | Age | Age | Age | Age | Age | Age |
| <19 (Ref) | | | | | | | | | | |
| 20–29 | 0.073 | 0.115*** (0.025) | 0.008 (4.332) | −0.006 | −1.666 | −0.854 | 0.115(0.109) | −0.098(2.944) | −0.073 | −89.802 |
| 30–39 | 0.095 | −0.019** (0.026) | −0.002 (0.267) | | | −0.511 | −0.019(0.084) | 0.001(1.264) | | |
| 40–49 | 0.085 | 0.024 (0.028) | 0.002 (0.003) | | | −0.220 | 0.024(0.062) | −0.005(0.016) | | |
| 50–59 | 0.058 | −0.083 (0.044) | −0.005 (0.010) | | | −0.148 | −0.083(0.062) | 0.012(1.000) | | |
| 60+ | 0.063 | −0.135** (0.058) | −0.009** (0.004) | | | −0.060 | −0.135(0.088) | 0.008(0.013) | | |
| Exposure to media | Exposure to media | Exposure to media | Exposure to media | Exposure to media | Exposure to media | Exposure to media | Exposure to media | Exposure to media | Exposure to media | Exposure to media |
| No TV/Radio (Ref) | | | | | | | | | | |
| TV/Radio | 0.050 | 0.531*** (0.023) | 0.027** (0.004) | 0.027 | 10.891 | −0.021 | 0.384*** (0.000) | −0.008 (0.997) | −0.008 | −26.667 |
| Marital status | Marital status | Marital status | Marital status | Marital status | Marital status | Marital status | Marital status | Marital status | Marital status | Marital status |
| Married (Ref) | | | | | | | | | | |
| Not married | −0.060 | 0.058** (0.026) | −0.004 (37.064) | −0.004 | −1.226 | 0.124 | 0.058 (0.079) | 0.007 (0.151) | 0.007 | 18.342 |
| Residence | Residence | Residence | Residence | Residence | Residence | Residence | Residence | Residence | Residence | Residence |
| Urban (Ref) | | | | | | | | | | |
| Rural | −0.032 | −0.539*** (0.043) | 0.017 (3.243) | 0.017 | 6.014 | −0.055 | 0.539*** (0.064) | 0.029 (3.851) | 0.029 | 35.877 |
| Employment | Employment | Employment | Employment | Employment | Employment | Employment | Employment | Employment | Employment | Employment |
| Not employed. (Ref) | | | | | | | | | | |
| Informal employment | −0.011 | 0.021 (0.027) | 0.000 (0.476) | 0.011 | 3.548 | −0.173 | −0.081 (0.0978) | −0.014 (0.455) | 0.021 | 25.520 |
| Formal emp. | 0.029 | 0.392*** (0.035) | 0.011 (0.017) | | | −0.082 | 0.434*** (0.044) | 0.035 (8.105) | | |
| Education | Education | Education | Education | Education | Education | Education | Education | Education | Education | Education |
| No school (Ref) | | | | | | | | | | |
| Primary school | 0.074 | −0.149*** (0.033) | −0.011 (7.858) | 0.056 | 19.905 | −0.251 | −0.149*** (0.061) | 0.019 (0.425) | 0.014 | 46.667 |
| Secondary school | 0.079 | 0.311*** (0.029) | 0.025 (298.5) | | | −0.221 | 0.320 (0.084) | −0.019 (0.587) | | |
| Tertiary | 0.067 | 0.621*** (0.042) | 0.042*** (0.007) | | | −0.069 | 0.621*** (0.050) | 0.014 (0.896) | | |
| Wealth quintile | Wealth quintile | Wealth quintile | Wealth quintile | Wealth quintile | Wealth quintile | Wealth quintile | Wealth quintile | Wealth quintile | Wealth quintile | Wealth quintile |
| Quintile 1 (Ref) | | | | | | | | | | |
| Quintile 2 | 0.026 | −0.491*** (0.055) | −0.013 (153.8) | 0.098 | 34.731 | 0.007 | 0.491*** (0.059) | −0.004 (0.305) | 0.186 | 227.125 |
| Quintile 3 | 0.037 | 0.005 (0.069) | 0.000 (0.096) | | | 0.118 | 0.005** (0.083) | 0.001 (0.148) | | |
| Quintile 4 | 0.052 | 0.508*** (0.073) | 0.026 (0.247) | | | 0.196 | 0.508 (0.098) | 0.010 (0.055) | | |
| Quintile 5 | 0.084 | 1.000*** (0.038) | 0.084*** (0.017) | | | 0.090 | 1.000*** (0.053) | 0.090 (11.462) | | |
| Residual | | | 0.057 | 0.057 | 35.758 | | | | −0.296 | −117.831 |
## Discussion
This study demonstrated the existence of socioeconomic inequality and horizontal inequity in the use of screening and treatment interventions for hypertension in Kenya. These findings can serve as a baseline for future progress assessments towards attaining SDG 3.4 targeted at NCDs and UHC goals in Kenya. *In* general, the results confirm that need does not match the use of screening and treatment services for hypertension across the SES groups and the Kenyan regions.
This paper's findings add to the evidence that the Kenyan health system is unequal and inequitable (54–56). It suggests that policy interventions geared towards attaining equity in the Kenyan health system should pay special attention to NCDs like hypertension. Among other policy options, it has been established that timely screening among those at risk and treatment among those diagnosed are cost-effective strategies for combating the burden of hypertension [10, 11]. However, our findings reveal considerable gaps in meeting the population need for both interventions in Kenya.
Poorer socioeconomic groups need more hypertension screening than their wealthier counterparts, but wealthier socioeconomic groups benefit more from screening services than their share of need. This finding could, in part, be explained by broader access barriers such as availability (i.e. biased availability of health facilities in urban locations), acceptability (i.e. providers and patients attitudes and expectations of each other) and affordability of screening services [57]. For instance, transport costs have been shown to not only lead to catastrophic expenditures for hypertension treatment in public health facilities in Kenya but also contribute to about $40\%$ of total out-of-pocket costs [58]. Furthermore, the unaffordability of NCDs screening services in Kenya has been documented, with healthcare costs being disproportionately higher in the private relative to the public sector [59]. Sex was the primary “need” factor contributing to socioeconomic inequality in hypertension screening, suggesting that females are more likely to seek screening services for hypertension. A similar finding has been reported in a study that assessed socioeconomic inequality for diabetes and hypertension screening in South Africa [25]. Of interest, being obese was the other significant contributor (in the pro-rich direction) to socioeconomic inequality in hypertension screening (Table 5). Given that being obese is a risk factor for cardiovascular diseases [27], this finding suggests that those who are wealthy and obese are more likely to utilise hypertension screening services compared to other BMI categories. Also, given that “non-need” determinants such as exposure to media and education contributed to inequality in the screening for hypertension in the pro-rich direction, unawareness of the importance of timely or early screening may provide some insights into the possible reasons for the underutilisation of screening services among poorer socioeconomic groups.
Whereas the geographic spread of health facilities in Kenya has increased over time [60], other supply-side factors such as the physical inaccessibility of health facilities remain a barrier for using screening services as the area of residence contributed to pro-rich inequality. Also, the weak health system capacity to offer care for NCDs, particularly at the primary care level, could explain the inequality in hypertension screening [61, 62]. A fragmented health service delivery structure biased towards offering curative rather than preventive healthcare services is among examples of health system weaknesses (63–65). The inequality and inequity in the screening for hypertension compare well with the findings of a South African study that has shown marked pro-rich inequality and inequity in the screening for hypertension, diabetes and cholesterol, with non-need factors (i.e. wealth status, health insurance, and education) mainly contributing to the inequality [25].
This study also found that hypertension treatment needs do not match how different SES groups use the service. The wealthier quintiles relative to the poorer ones benefited more than their treatment needs, and this disparity existed in the Kenyan regions. Among other things, this finding can be explained by the overall unaffordability of NCDs services in Kenya [59] and the low levels of health insurance coverage [66], with lower socioeconomic groups being disproportionately affected. Similar patterns have been reported in previous studies in Kenya that have assessed inequality in health and healthcare use or access at the sub-national level (24, 54–56). While inequality in the use of hypertension treatment was not statistically significant, the horizontal inequity finding suggests that hypertension treatment is not distributed according to need. Besides, compared to the inequality observed in hypertension screening, the “relative equality” in hypertension treatment could be partly explained by the difference in distribution of the burden of hypertension in the Kenyan population (i.e. $42\%$ of rural dwellers have hypertension while $30\%$ of urban dwellers have hypertension). Therefore, it can be argued that the pro-rich distribution in the use of hypertension treatment services could be as a result of the pro-urban bias in the distribution of healthcare facilities in Kenya and thus the rural dwellers are disadvantaged at utilising hypertension treatment services while they are most in need [60]. Of note, the findings compare well with previous multi-country studies that found significant pro-rich inequality in the use of secondary cardiovascular medicines in LMICs [17] and hypertension treatment [32].
In the decomposition analysis of inequality in hypertension treatment, non-need factors primarily contributed to the observed inequality. For instance, area of residence, wealth, employment and education status contributed to inequality in hypertension treatment. Similar findings have been reported in China, where non-need factors such as income, area of residence, longest-held occupation, and level of education were significant contributors to the socioeconomic inequality in the utilisation of hypertension, hyperglycaemia and dyslipidaemia treatment [15]. Unlike findings from other settings [18, 26], in this study, hypertension prevalence was higher among wealthier socioeconomic groups than their poorer counterparts. This disproportionate hypertension burden may lead to differences in healthcare demand between the rich and the poor in Kenya.
Several policy recommendations are imperative from the findings of this study. First, since service delivery falls within the docket of county governments in Kenya, there is an urgent need to enhance the capacity of primary care facilities to implement cost-effective strategies such as timely screening so that need can match service use for this critical intervention. Second, for demand in the utilisation of screening services to be increased, national and county governments, including other relevant actors, should implement strategic awareness-raising campaigns targeting at-risk populations as age and sex contributed to the SES inequality in the screening of hypertension. This can be through targeted health education messages in the mass media and other appropriate channels. Third, while recent efforts by the government of Kenya to attain UHC by 2022 are timely and commendable, more needs to be done to ensure the realisation of equity in the use of NCD services. Given the interplay of factors beyond the health sector that affect health, as was seen in the role of non-need factors in contributing to inequality in screening and treatment, there is a need for multi-sectoral approaches at various levels (i.e. local, national and regional) to address drivers of poverty and social inequity with a critical focus in marginalised areas. Some of the sectors that could collaborate with the health sector in addressing inequities/inequalities in NCDs include education and media. For instance, for increased health education on NCDs, the education sector can include NCDs in the curriculum. Also, various media channels can be used to raise awareness on the benefits of early NCDs screening as exposure to media was shown to contribute to SES inequality in hypertension screening.
## Study Strengths and Limitations
This study had strengths and limitations. One key strength was the national representativeness of the data set used, which gave the national picture of socioeconomic inequality and inequity in the screening and treatment for hypertension in Kenya. Also, while previous studies have mainly assessed inequalities in the prevalence of NCDs (and in most cases using self-reported data), this study examines inequity and socioeconomic inequality in key interventions using objective measures of need for screening and treatment. This study also used a novel methodological approach: the decomposition analysis, to uncover factors contributing to socioeconomic inequality in screening and treatment for hypertension in Kenya.
This study also had limitations. The first limitation was data-driven. As is common in studies on care utilisation, we relied on self-reported data in defining the use of both screening and treatment. This could potentially bias our inequality findings, especially if there were cases of misreporting. Likewise, although previous studies [67, 68] have reported no association between under-reporting of care utilisation and demographic characteristics, except for age, we cannot rule out under-reporting of care utilisation in the low SES groups. Second, the STEPs data set was cross-sectional in design and thus limiting the establishment of temporal trends in inequality and inequity in the use of screening and treatment. Furthermore, it is essential to note that causality is not implied for the factors explaining observed inequality in screening and treatment.
## Conclusion
Kenya faces a rising disease burden from non-communicable diseases, as expected in many low-and middle-income countries. This paper provides the first empirical evidence on socioeconomic inequality and inequity in screening and treatment interventions for NCDs based on need in Kenya. These findings provide a benchmark for future equity and equality assessments for NCDs in Kenya. In keeping with the global UHC agenda and other key NCDs targets, there is an urgent need for concerted efforts to ensure equity in providing NCDs healthcare services in Kenya. Indeed, given the ongoing policy reforms to attain UHC in Kenya, a window of opportunity exists to avert inequity in NCDs, with this paper highlighting some of the critical issues for consideration.
## Data Availability Statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics Statement
The studies involving human participants were reviewed and approved by Human Research Ethics Committee of the University of Cape Town (Ref: $\frac{186}{2020}$). The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
RO, JA, and EB conceived the study. RO conducted the data analysis and wrote the first draft of the manuscript. All authors have read and agreed to the published version of the manuscript.
## Funding
The authors acknowledge funding support from the Wellcome Trust International Masters Fellowship (Grant Award Number: 214622) awarded to RO. EB is funded by a Wellcome Trust core grant (#092654).
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. 1.World Health Organization. Global Status Report on Noncommunicable Diseases 2014. World Health Organization (2014).. *Global Status Report on Noncommunicable Diseases 2014* (2014.0)
2. Naghavi M, Abajobir AA, Abbafati C, Abbas KM, Abd-Allah F, Abera SF. **Global, regional, and national age-sex specific mortality for 264 causes of death, 1980–2016: a systematic analysis for the Global Burden of Disease Study 2016**. *Lancet.* (2017.0) **390** 1151-210. DOI: 10.1016/S0140-6736(17)32152-9
3. Aikins Ad-G, Agyemang C. *Chronic Non-communicable Diseases in Low and Middle-income Countries* (2015.0)
4. Kengne AP. **Mayosi BM. Readiness of the primary care system for non-communicable diseases in sub-Saharan Africa**. *Lancet Glob Health.* (2014.0) **2** e247-e8. DOI: 10.1016/S2214-109X(14)70212-8
5. Aikins Ad-G, Kushitor M, Koram K, Gyamfi S, Ogedegbe G. **Chronic non-communicable diseases and the challenge of universal health coverage: insights from community-based cardiovascular disease research in urban poor communities in Accra, Ghana**. *BMC Public Health* (2014.0) **14** S3. DOI: 10.1186/1471-2458-14-S2-S3
6. Bonfrer I, van de Poel E, Grimm M, Van Doorslaer E. **Does the distribution of healthcare utilization match needs in Africa?**. *Health Policy Plan.* (2013.0) **29** 921-37. DOI: 10.1093/heapol/czt074
7. Di Cesare M, Khang Y-H, Asaria P, Blakely T, Cowan MJ, Farzadfar F. **Inequalities in non-communicable diseases and effective responses**. *Lancet.* (2013.0) **381** 585-97. DOI: 10.1016/S0140-6736(12)61851-0
8. 8.United Nations. Transforming Our World - the 2030 Agenda for Sustainable Development. (2015). Available online at: https://sustainabledevelopment.un.org/topics/sustainabledevelopmentgoals (accessed November 14, 2019).. *Transforming Our World - the 2030 Agenda for Sustainable Development* (2015.0)
9. 9.World Health Organization. From Burden to “Best Buys”: Reducing the Economic Impact of Non-Communicable Diseases in Low-and Middle-Income Countries. Geneva: WHO (2011).. *From Burden to “Best Buys”: Reducing the Economic Impact of Non-Communicable Diseases in Low-and Middle-Income Countries* (2011.0)
10. Baker J, Mitchell R, Lawson K, Pell J. **Ethnic differences in the cost-effectiveness of targeted and mass screening for high cardiovascular risk in the UK: cross-sectional study**. *Heart.* (2013.0) **99** 1766-71. DOI: 10.1136/heartjnl-2013-304625
11. Howard K, White S, Salkeld G, McDonald S, Craig JC, Chadban S. **Cost-effectiveness of screening and optimal management for diabetes, hypertension, and chronic kidney disease: a modeled analysis**. *Value in health.* (2010.0) **13** 196-208. DOI: 10.1111/j.1524-4733.2009.00668.x
12. Schmidt H. **Barnhill A. Equity and noncommunicable disease reduction under the sustainable development goals**. *PLos Med.* (2015.0) **12** e1001872. DOI: 10.1371/journal.pmed.1001872
13. Carrin G, Mathauer I, Xu K, Evans DB. **Universal coverage of health services: tailoring its implementation**. *Bull World Health Organ.* (2008.0) **86** 857-63. DOI: 10.2471/BLT.07.049387
14. 14.World Health Organization. Health Systems Financing: The Path to Universal Coverage. Geneva: World Health Organization (2012).. *Health Systems Financing: The Path to Universal Coverage* (2012.0)
15. Elwell-Sutton TM, Jiang CQ, Zhang WS, Cheng KK, Lam TH, Leung GM. **Inequality and inequity in access to health care and treatment for chronic conditions in China: the Guangzhou Biobank Cohort Study**. *Health Policy Plan.* (2012.0) **28** 467-79. DOI: 10.1093/heapol/czs077
16. Mayosi BM, Flisher AJ, Lalloo UG, Sitas F, Tollman SM, Bradshaw D. **The burden of non-communicable diseases in South Africa**. *Lancet.* (2009.0) **374** 934-47. DOI: 10.1016/S0140-6736(09)61087-4
17. Murphy A, Palafox B, O'Donnell O, Stuckler D, Perel P, AlHabib KF. **Inequalities in the use of secondary prevention of cardiovascular disease by socioeconomic status: evidence from the PURE observational study**. *Lancet Global Health.* (2018.0) **6** e292-301. DOI: 10.1016/S2214-109X(18)30031-7
18. Sozmen K, Belgin U. **Socioeconomic inequalities in non-communicable diseases and self assessed health in Turkey**. *Iran J Public Health.* (2014.0) **43** 736. PMID: 26110144
19. Vellakkal S, Millett C, Basu S, Khan Z, Aitsi-Selmi A, Stuckler D. **Are estimates of socioeconomic inequalities in chronic disease artefactually narrowed by self-reported measures of prevalence in low-income and middle-income countries? Findings from the WHO-SAGE survey**. *J Epidemiol Commun Health.* (2015.0) **69** 218-25. DOI: 10.1136/jech-2014-204621
20. Hart JT. **The inverse care law**. *Lancet.* (1971.0) **297** 405-12. DOI: 10.1016/S0140-6736(71)92410-X
21. Kane J, Landes M, Carroll C, Nolen A. **Sodhi S. A systematic review of primary care models for non-communicable disease interventions in sub-Saharan Africa**. *BMC Fam Pract.* (2017.0) **18** 46. DOI: 10.1186/s12875-017-0613-5
22. McIntyre D, Obse AG, Barasa EW, Ataguba JE. *Challenges in Financing Universal Health Coverage in Sub-Saharan Africa* (2018.0). DOI: 10.1093/acrefore/9780190625979.013.28
23. 23.World Health Organization. Assessing National Capacity for the Prevention and Control of Noncommunicable Diseases: Report of the 2019 Global Survey. WHO (2020).. *Assessing National Capacity for the Prevention and Control of Noncommunicable Diseases: Report of the 2019 Global Survey* (2020.0)
24. Gatimu SM, John TW. **Socioeconomic inequalities in hypertension in Kenya: a decomposition analysis of 2015 Kenya STEPwise survey on non-communicable diseases risk factors**. *Int J Equity Health.* (2020.0) **19** 1-11. DOI: 10.1186/s12939-020-01321-1
25. Zulu T. *Socioeconomic Inequalities in Non-Communicable Diseases in South Africa* (2019.0)
26. Su M, Si Y, Zhou Z, Shen C, Dong W, Fan X. **Comparing the income-related inequity of tested prevalence and self-reported prevalence of hypertension in China**. *Int J Equity Health.* (2018.0) **17** 82. DOI: 10.1186/s12939-018-0796-y
27. 27.Ministry of Health. Kenya STEPwise Survey for Non-Communicable Diseases Risk Factors 2015 Report. (2015). Available online at: http://aphrc.org/wp-content/uploads/2016/04/Steps-Report-NCD-2015.pdf (accessed August 8, 2019).. *Kenya STEPwise Survey for Non-Communicable Diseases Risk Factors 2015 Report* (2015.0)
28. Chuma J, Molyneux C. **Estimating inequalities in ownership of insecticide treated nets: does the choice of socio-economic status measure matter?**. *Health Policy Plan.* (2009.0) **24** 83-93. DOI: 10.1093/heapol/czn050
29. Filmer D, Pritchett LH. **Estimating wealth effects without expenditure data—or tears: an application to educational enrollments in states of India**. *Demography.* (2001.0) **38** 115-32. DOI: 10.1353/dem.2001.0003
30. Wagstaff A, Watanabe N. **What difference does the choice of SES make in health inequality measurement?**. *Health Econ.* (2003.0) **12** 885-90. DOI: 10.1002/hec.805
31. Lindelow MJHe. **Sometimes more equal than others: how health inequalities depend on the choice of welfare indicator**. *Health Econ.* (2006.0) **15** 263-79. DOI: 10.1002/hec.1058
32. Palafox B, McKee M, Balabanova D, AlHabib KF, Bahonar A, Ismail N. **Wealth and cardiovascular health: a cross-sectional study of wealth-related inequalities in the awareness, treatment and control of hypertension in high-, middle-and low-income countries**. *Int J Equity Health.* (2016.0) **15** 1-14. DOI: 10.1186/s12939-016-0478-6
33. Vyas S, Kumaranayake L. **Constructing socio-economic status indices: how to use principal components analysis**. *Health Policy Plan.* (2006.0) **21** 459-68. DOI: 10.1093/heapol/czl029
34. Howe LD, Hargreaves JR, Huttly SR. **Issues in the construction of wealth indices for the measurement of socio-economic position in low-income countries**. *Emerg Themes Epidemiol.* (2008.0) **5** 3. DOI: 10.1186/1742-7622-5-3
35. McKenzie DJ. **Measuring inequality with asset indicators**. *J Popul Econ* (2005.0) **18** 229-60. DOI: 10.1007/s00148-005-0224-7
36. 36.Ministry of Health. Kenya National Guidelines for Cardiovascular Diseases Management. Nairobi: Ministry of Health. 2018.. *Kenya National Guidelines for Cardiovascular Diseases Management*
37. Wagstaff A, Paci P, Van Doorslaer E. **On the measurement of inequalities in health**. *Soc Sci Med.* (1991.0) **33** 545-57. DOI: 10.1016/0277-9536(91)90212-U
38. Ataguba JE, Akazili J, McIntyre D. **Socioeconomic-related health inequality in South Africa: evidence from general household surveys**. *Int J Equity Health.* (2011.0) **10** 48. DOI: 10.1186/1475-9276-10-48
39. Kakwani N, Wagstaff A, Van Doorslaer E. **Socioeconomic inequalities in health: measurement, computation, and statistical inference**. *J Econom.* (1997.0) **77** 87-103. DOI: 10.1016/S0304-4076(96)01807-6
40. O'Donnell O, Van Doorslaer E, Wagstaff A, Lindelow M. *Analyzing Health Equity Using Household Survey Data: A Guide to Techniques and Their Implementation* (2008.0). DOI: 10.1596/978-0-8213-6933-3
41. Wagstaff A. **The bounds of the concentration index when the variable of interest is binary, with an application to immunization inequality**. *Health Econ.* (2005.0) **14** 429-32. DOI: 10.1002/hec.953
42. Wagstaff A. **Correcting the concentration index: a comment**. *J Health Econ.* (2009.0) **28** 516-20. DOI: 10.1016/j.jhealeco.2008.12.003
43. Wagstaff A. **The concentration index of a binary outcome revisited**. *Health Econ.* (2011.0) **20** 1155-60. DOI: 10.1002/hec.1752
44. Erreygers G. **Correcting the concentration index**. *J Health Econ.* (2009.0) **28** 504-15. DOI: 10.1016/j.jhealeco.2008.02.003
45. Erreygers G, Van Ourti T. **Putting the cart before the horse: a reply to Wagstaff on inequality measurement in the presence of binary variables**. *Health Econ.* (2011.0) **20** 1161. DOI: 10.1002/hec.1754
46. Wagstaff A, Doorslaer vE, Watanabe N. **On decomposing the causes of health sector inequalities with an application to malnutrition inequalities in Vietnam:**. (2003.0) **112** 207-23. DOI: 10.1016/S0304-4076(02)00161-6
47. Yiengprugsawan V, Lim LL, Carmichael GA, Dear KB, Sleigh AC. **Decomposing socioeconomic inequality for binary health outcomes: an improved estimation that does not vary by choice of reference group**. *BMC Res Notes.* (2010.0) **3** 57. DOI: 10.1186/1756-0500-3-57
48. Niessen LW, Mohan D, Akuoku JK, Mirelman AJ, Ahmed S, Koehlmoos TP. **Tackling socioeconomic inequalities and non-communicable diseases in low-income and middle-income countries under the sustainable development agenda**. *Lancet.* (2018.0) **391** 2036-46. DOI: 10.1016/S0140-6736(18)30482-3
49. Williams J, Allen L, Wickramasinghe K, Mikkelsen B, Roberts N, Townsend N. **A systematic review of associations between non-communicable diseases and socioeconomic status within low-and lower-middle-income countries**. *J Glob Health* (2018.0) **8** 020409. DOI: 10.7189/jogh.08.020409
50. Efron B. **Better bootstrap confidence intervals**. *J Am Stat Assoc.* (1987.0) **82** 171-85. DOI: 10.1080/01621459.1987.10478410
51. Efron B, Tibshirani R. **Bootstrap methods for standard errors, confidence intervals, and other measures of statistical accuracy: rejoinder**. *Stat Sci.* (1986.0) **1** 77. DOI: 10.1214/ss/1177013817
52. Ataguba JE, Ojo KO, Ichoku HE. **Explaining socio-economic inequalities in immunization coverage in Nigeria**. *Health Policy Plan.* (2016.0) **31** 1212-24. DOI: 10.1093/heapol/czw053
53. Wagstaff A, Van Doorslaer E. **Measuring and testing for inequity in the delivery of health care**. *J Hum Resour.* (2000.0) **35** 716-33. DOI: 10.2307/146369
54. Barasa E, Nguhiu P, McIntyre D. **Measuring progress towards sustainable development goal 3.8 on universal health coverage in Kenya**. *BMJ Glob Health.* (2018.0) **3** e000904. DOI: 10.1136/bmjgh-2018-000904
55. Ilinca S, Di Giorgio L, Salari P, Chuma J. **Socio-economic inequality and inequity in use of health care services in Kenya: evidence from the fourth Kenya household health expenditure and utilization survey**. *Int J Equity Health.* (2019.0) **18** 196. DOI: 10.1186/s12939-019-1106-z
56. Nguhiu PK, Barasa EW, Chuma J. **Determining the effective coverage of maternal and child health services in Kenya, using demographic and health survey data sets: tracking progress towards universal health coverage**. *Trop Med Int Health.* (2017.0) **22** 442-53. DOI: 10.1111/tmi.12841
57. McIntyre D, Thiede M, Birch S. **Access as a policy-relevant concept in low-and middle-income countries**. *Health Econ Policy Law.* (2009.0) **4** 179-93. DOI: 10.1017/S1744133109004836
58. Oyando R, Njoroge M, Nguhiu P, Kirui F, Mbui J, Sigilai A. **Patient costs of hypertension care in public health care facilities in Kenya**. *Int J Health Plann Manage.* (2019.0) **34** e1166-e78. DOI: 10.1002/hpm.2752
59. Subramanian S, Gakunga R, Kibachio J, Gathecha G, Edwards P, Ogola E. **Cost and affordability of non-communicable disease screening, diagnosis and treatment in Kenya: patient payments in the private and public sectors**. *PLoS One.* (2018.0) **13** e0190113. DOI: 10.1371/journal.pone.0190113
60. Noor AM, Alegana VA, Gething PW, Snow RW. **A spatial national health facility database for public health sector planning in Kenya in 2008**. *Int J Health Geogr.* (2009.0) **8** 13. DOI: 10.1186/1476-072X-8-13
61. Ashigbie PG, Rockers PC, Laing RO, Cabral HJ, Onyango MA, Buleti JPL. **Availability and prices of medicines for non-communicable diseases at health facilities and retail drug outlets in Kenya: a cross-sectional survey in eight counties**. *BMJ Open.* (2020.0) **10** e035132. DOI: 10.1136/bmjopen-2019-035132
62. 62.Government of Kenya. Kenya Service Availablity and Readiness Assessment Mapping (SARAM) Report. Nairobi: Ministry of Health (2014).. *Kenya Service Availablity and Readiness Assessment Mapping (SARAM) Report* (2014.0)
63. Atun R, Jaffar S, Nishtar S, Knaul FM, Barreto ML, Nyirenda M. **Improving responsiveness of health systems to non-communicable diseases**. *Lancet.* (2013.0) **381** 690-7. DOI: 10.1016/S0140-6736(13)60063-X
64. Ali MK, Rabadán-Diehl C, Flanigan J, Blanchard C, Narayan KV, Engelgau M. **Systems and capacity to address noncommunicable diseases in low-and middle-income countries**. *Sci Transl Med* (2013.0) **5** 181cm4-cm4. DOI: 10.1126/scitranslmed.3005121
65. Beaglehole R, Epping-Jordan J, Patel V, Chopra M, Ebrahim S, Kidd M. **Improving the prevention and management of chronic disease in low-income and middle-income countries: a priority for primary health care**. *Lancet.* (2008.0) **372** 940-9. DOI: 10.1016/S0140-6736(08)61404-X
66. Kazungu JS, Barasa EW. **Examining levels, distribution and correlates of health insurance coverage in Kenya**. *Trop Med Int Health.* (2017.0) **22** 1175-85. DOI: 10.1111/tmi.12912
67. Bhandari A, Wagner T. **Self-reported utilization of health care services: improving measurement and accuracy**. *Med Care Res Rev.* (2006.0) **63** 217-35. DOI: 10.1177/1077558705285298
68. Tsui EL, Leung GM, Woo PP, Choi S, Lo S-V. **Under-reporting of inpatient services utilisation in household surveys–a population-based study in Hong Kong**. *BMC Health Serv Res.* (2005.0) **5** 31. DOI: 10.1186/1472-6963-5-31
|
---
title: 'Sustaining capacity building and evidence-based NCD intervention implementation:
Perspectives from the GRIT consortium'
authors:
- Ashlin Rakhra
- Shivani Mishra
- Angela Aifah
- Calvin Colvin
- Joyce Gyamfi
- Gbenga Ogedegbe
- Juliet Iwelunmor
journal: Frontiers in Health Services
year: 2022
pmcid: PMC10012828
doi: 10.3389/frhs.2022.891522
license: CC BY 4.0
---
# Sustaining capacity building and evidence-based NCD intervention implementation: Perspectives from the GRIT consortium
## Abstract
### Background
Implementation science has been primarily focused on adoption of evidence-based interventions, and less so on sustainability, creating a gap in the field. The Global Research on Implementation and Translation Science (GRIT) *Consortium is* funded by the National Heart Lung and Blood Institute (NHBLI) to support the planning, implementation, and sustainability of Late-Stage Phase 4 Translational Research (T4TR) and capacity building for NCD prevention and control in eight low-and middle-income countries (LMICs). This paper highlights perspectives, including barriers, facilitators, opportunities, and motivators for sustaining capacity building and evidence-based hypertension interventions within LMICs.
### Methods
Guided by the Capacity, Opportunity, Motivation, Behavior (COM-B) Model, this study surveyed GRIT consortium members on the barriers, facilitators, key motivators, and opportunities for sustaining capacity building and evidence-based hypertension interventions in LMICs. Thematic analysis was used to identify themes and patterns across responses.
### Results
Twenty-five consortium members across all eight sites and from various research levels responded to the survey. Overarching themes identifying facilitators, key motivators and opportunities for sustainability included: [1] access to structured and continuous training and mentorship; [2] project integration with existing systems (i.e., political systems and health systems); [3] adaption to the local context of studies (i.e., accounting for policies, resources, and utilizing stakeholder engagement); and [4] development of interventions with decision makers and implementers. Barriers to sustainability included local policies and lack of infrastructure, unreliable access to hypertension medications, and lack of sufficient staff, time, and funding.
### Conclusion
Sustainability is an important implementation outcome to address in public health interventions, particularly as it pertains to the success of these initiatives. This study provides perspectives on the sustainability of NCD interventions with a focus on mitigating their NCD burden in LMICs. Addressing multilevel factors that influence the sustainability of capacity building and interventions will have notable implications for other global NCD efforts going forward. Current and future studies, as well as consortium networks, should account for sustainability barriers outlined as it will strengthen program implementation, and long-term outcomes.
## Background
The burden of non–communicable diseases (NCDs) continues to rise globally with a disproportionate impact in low and middle-income countries (LMIC) [1]. Deaths due to NCDs in LMICs are expected to increase from 30.8 million in 2015 to 41.8 million by 2030 [2]. To address this growing disease burden, continued evidence-based interventions (EBI) addressing NCDs and capacity building for NCD investigators in LMICs is needed. Moreover, comprehensive sustainability efforts addressing barriers and facilitators to NCD EBIs and capacity building uptake are crucial to maximize the impact of these efforts to ensure long-term health outcomes are maintained [3, 4].
While program sustainability is not a new concept, the field of implementation science has focused more so on understanding factors and strategies that influence the adoption and implementation of EBIs and less so on the factors and strategies impacting sustainability [5]. While studies have discussed multi-level factors influencing sustainability that relate to context (i.e., outer context, policies, legislation, funding and inner context, culture, structure), innovation or the intervention itself (i.e., fit, effectiveness), process (i.e., fidelity, monitoring, evaluation), political support, funding partnerships research on sustainability factors still needs to be more widely adopted [6, 7].
The Global Research on Implementation and Translational Science (GRIT) Consortium was convened in 2018 by the National Heart Lung and Blood Institute (NHLBI) to support the planning, implementation, and sustainability of Late-Stage Phase 4 Translational Research (T4TR) and capacity building initiatives for NCD prevention and control in LMICs. The overarching goal of the GRIT *Consortium is* to define and establish a strategy that connects consortium members to capacity-building initiatives that will enhance the sustainable uptake of evidence-based interventions for NCD prevention and control in LMICs [2, 8]. The network comprises investigators funded by the Hypertension Outcomes for T4 Research in LMICs (HyTREC) and the Translation Research Capacity Building Initiative in LMICs (TREIN) programs. The consortium consists of research teams from eight countries, five of which (Guatemala, Ghana, Kenya, India, and Vietnam) test implementation strategies to deliver evidence-based interventions within these countries for the prevention, treatment, and control of hypertension (HyTREC sites) and three of which (Malawi, Nepal, and Rwanda) provide capacity building in NCD and D&I research needed to close the gap between research and practice (TREIN sites). Additional details of each site in the consortium are published elsewhere and can be found in Table 1 (9–16).
**Table 1**
| Site & Project Title | Brief Summary of project |
| --- | --- |
| HyTREC Sites | HyTREC Sites |
| Ghana (10) Uptake of Task-Strengthening Strategy for Hypertension Control within Community Health Planning Services in Ghana: A Mixed Method Study | The goal of this study is to evaluate, in a hybrid clinical effectiveness- implementation cluster design, the effect of practice facilitation (PF) on the uptake of an evidence-based Task Strengthening Strategy for Hypertension control (TASSH), among 700 adults who present to 70 Community- Based Health Planning Services (CHPs) zones with uncontrolled hypertension. |
| Guatemala (14) Implementing a Multicomponent Intervention to Improve Hypertension Control in Central America. A Cluster Randomized Trial in Guatemala | A cluster randomized clinical trial to test the co-primary objectives: The effect of a multilevel and multicomponent intervention program on blood pressure (BP) control among Guatemalan hypertensive patients over an 18-month period The acceptability, adoption, feasibility, fidelity, reach, and sustainability of implementing the intervention in patients, providers, and health districts. |
| India (15) Integrated Tracking, Referral, and Electronic Decision Support, and Care Coordination (I-TREC) | The overall goal of this 5-year project is to adapt, implement, and evaluate an IT- enabled platform for integrated tracking, referral, electronic decision support, and care coordination (I-TREC) to treat hypertension and diabetes in rural communities that rely on public health care system using mixed methods approach (Quasi-experimental design). |
| Kenya (13) Strengthening Referral Networks for Management of Hypertension Across the Health System (STRENGTHS) in western Kenya: a study protocol of a cluster randomized trial | A cluster randomized control trial evaluating the effectiveness and cost- effectiveness of a combined health information technology (HIT) and peer support intervention on referral completion, BP improvement, and CVD risk reduction in Kenya. |
| Vietnam (12) Conquering Hypertension in Vietnam: Solutions at Grassroots level | A cluster randomized controlled trial to evaluate the implementation and effectiveness of two multi-faceted community and clinic-based strategies for the control of hypertension among adults residing in the rural Red River Delta region of Vietnam with uncontrolled hypertension. |
| TREIN Sites | TREIN Sites |
| Malawi (16) NCD BRITE- Building Research Capacity, Implementation and Translation Expertise for non-communicable diseases | The proposed program will build long-term, sustainable heart, lung, blood and sleeping diseases and disorders (HLBS) focused late-stage translation phase 4 research (T4TR) capacity in Malawi and will utilize this capacity together with research infrastructure and diseases burden needs assessments, to design a Malawi specific HLBS T4TR research plan. The trans-disciplinary consortium is purposefully designed to build capacity within the University of Malawi-College of Medicine (COM), the only public medical school in the country, and the Malawi Ministry of Health (MoH), to ensure sustainability. |
| Nepal (9) Translational Research Capacity Building Initiative to Address Cardiovascular Diseases in Nepal | Dhulikhel Hospital Kathmandu University Hospital, Nepal, will lead work to create a multi-sectoral, multidisciplinary collaborative team to develop Translational research capacity Building initiatives to prevent and manage CVD in Nepal. By the end of the project, we will have developed a critical mass of human Resources in Nepal, collaborating with national and International partners, to conduct Translational Research in CVD. We will have defined clearly identified prioritized needs and a well- defined Translational Research plan to address one or more major CVD Risk Factors and outcomes. |
| Rwanda (11) Developing T4 translational research capacity for control of hypertension in Rwanda | This project will create a collaborative team of academics, clinicians, community healthcare providers, and public health experts to engage in T4TR by building the competencies required to enhance uptake of proven interventions for control of hypertension in Rwanda. |
The GRIT Consortium's contribution to hypertension and other NCD knowledge and services is unique due to the collaborative stakeholder and implementer perspectives of multiple LMICs. The GRIT Consortium sites have identified common determinants and adoptable strategies for NCD interventions and capacity building in LMICs [2, 8]. The consortium not only addresses the knowledge gap between program implementation and sustainability, but also lays a groundwork for discussing other potential gaps in dissemination and implementation (D&I) practice in LMICs.
As the TREIN and HyTREC projects are in their final phases, the consortium has been focused on sustaining both the capacity building and intervention implementation efforts. For this study, we adapted the Michie and colleagues Behavior Change Wheel framework and the COM-B Model as the model uses three factors- capabilities, opportunities, motivations for identifying changes to ensure behavior change interventions are effective [17, 18]. The COM-B Model has been used in other studies addressing the implementation of hypertension interventions in LMICs, as well as in other contexts to develop effective interventions [19]. The goal of this study was to examine the capabilities, opportunities, and motivators for sustaining hypertension and other NCD intervention implementation and capacity building in LMICs [17, 18]. This study describes barriers, facilitators, motivators and opportunities identified by GRIT Consortium researchers to enhance future NCD sustainability efforts.
## Study design and procedure
This was a qualitative open-ended descriptive online survey conducted across the GRIT Consortium in March and April of 2021. This study used purposive sampling to recruit researchers across study roles and from all eight GRIT Consortium sites (Ghana, Guatemala, India, Kenya, Malawi, Nepal, Rwanda, and Vietnam). The survey remained open until saturation was reached.
## Conceptual framework
Implementing and sustaining behavior changes (i.e., NCD control and capacity building) may occur as a result of an interaction between three components: capability, opportunity, and motivation [6]. As such, the survey tool was guided by the Capabilities, Opportunities, Motivations, and Behavior (COM-B) Model [17, 18]. Capability is defined as one's psychological capacity (i.e., knowledge) and physical capacity (i.e., skills) to engage in a behavior; Opportunity represents external factors that affect one's capacity to perform (i.e., physical environment, social influences and cultural norms); and Motivation represents internal factors that allow one to employ capability and opportunity to perform a behavior (i.e., wants, needs, beliefs, intentions) (see Figure 1) [17, 18].
**Figure 1:** *The COM-B model.*
## Survey development
Guided by the COM-B model, a qualitative open-ended descriptive online survey was developed and administered to the GRIT consortium project sites. The initial survey was piloted among a sub-group of GRIT members and was subsequently revised and refined based on feedback to ensure clarity of wording and usability. In addition to demographic questions regarding the researchers' role on their study, research team members were asked about their experience as program implementers across the eight countries in the consortium. The survey assessed the three main domains of the COM-B through open-ended questions: [1] what would you say makes it easy or difficult to implement and sustain capacity building and/ or evidence-based HTN intervention implementation? ( capability); [2] what would you say motivates researchers/ and or other key stakeholders to implement and sustain capacity building and/ or evidence-based hypertension intervention implementation? ( motivation); and [3] what opportunities exist to continuously support researchers/ community members to implement and sustain capacity building and/ or evidence-based hypertension intervention implementation? ( opportunities).
## Data analysis
The survey tool was administered in English which was the common language among participants. The data relevant to each construct of the COM-B Model was documented by two authors using a data extraction sheet. The information was summarized and reported descriptively using content analysis to the COM-B Model. Discrepancies between the two authors were resolved by open discussions and consultation sessions among the research team. The consolidated criteria for reporting qualitative studies (COREQ) was followed [20].
## Participant characteristics
Twenty-five consortium members completed the questionnaire. Table 2 outlines the country site and study team roles of respondents within the consortium. 56 percent of respondents were principles investigators (PIs) or co-investigators (Co-Is), $24\%$ were coordinators, and the remaining respondents included statistician(s), data manager(s), researcher(s), program manager(s), and an international liaison. All eight sites in the consortium were represented in the responses. Table 2 outlines additional respondent demographics.
**Table 2**
| Country | Country.1 |
| --- | --- |
| Ghana | 12% (3) |
| Guatemala | 12% (3) |
| India | 12% (3) |
| Kenya | 8% (2) |
| Malawi | 12% (3) |
| Nepal | 8% (2) |
| Rwanda | 24% (6) |
| Vietnam | 12% (3) |
| Study Team Role | Study Team Role |
| PI/ Co-PI | 20% (5) |
| Investigator/ Co-Investigator | 36% (9) |
| Statistician | 4% (1) |
| Data Manager | 4% (1) |
| Coordinator (research, project, implementation) | 24% (6) |
| Researcher | 4% (1) |
| Program Manager | 4% (1) |
| Lead International Liaison | 4% (1) |
## Barriers and facilitators
Table 3 outlines the respondent-identified barriers and facilitators to sustaining capacity building and/ or evidence-based hypertension or NCD intervention implementation. Barriers identified by respondents included: [1] lack of hypertension medications ($17\%$); [2] lack of time (during implementation and post-intervention) ($19\%$); [3] lack of funding ($11\%$); [4] lack of staff ($17\%$); [5] low education or understanding of intervention/ disease among population/ patient and provider ($17\%$); [6] context (local policies, lack of infrastructure, context specific social and cultural beliefs.) ( $11\%$); [7] lack of hypertension diagnosis ($3\%$); [8] lack of epidemiology data ($3\%$); and [9] insufficient or lack of internet access at work ($3\%$). Facilitators included: [1] training opportunities ($22\%$); [2] mentorship and leadership support ($11\%$); [3] community/stakeholder engagement ($17\%$); [4] working in multi-disciplinary teams ($8\%$); [5] local context (adoption to and capacity of local systems) ($17\%$); [6] political support ($6\%$); [7] motivation of staff ($3\%$); [8] quarterly workshops to review challenges in EBI hypertension interventions ($3\%$); and 9) acceptance of hypertension (less stigma, not infectious, modifiable risk factor) ($8\%$).
**Table 3**
| Barriers (n = 36) | Barriers (n = 36).1 |
| --- | --- |
| Lack of HTN medication | 17% (6) |
| Lack of time | 19% (7) |
| Lack of funding | 11% (4) |
| Lack of staff | 17% (6) |
| Low education (population/ patient and provider) | 17% (6) |
| Context (policies, infrastructure, etc.) | 11% (4) |
| Late diagnosis of HTN | 3% (1) |
| Lack of epidemiology data | 3% (1) |
| Insufficient or lack of internet access at work | 3% (1) |
| Facilitators (n=36) | Facilitators (n=36) |
| Training | 22% (8) |
| Mentorship/ Leadership Support | 11% (4) |
| Community/ Stakeholder Engagement | 17% (6) |
| Multi- disciplinary teams | 8% (3) |
| Local Context (adoption to and capacity of systems) | 17% (6) |
| Political Support | 6% (2) |
| Motivation of staff | 3% (1) |
| Quarterly workshops to review challenges in evidence-based hypertension interventions | 3% (1) |
| Hypertension (less stigma, not infectious, modifiable risk factors) | 8% (3) |
## Key motivators
Figure 2 highlights the most-common motivators for sustaining capacity building and/ or evidence- based NCD intervention implementation. 31 percent of the respondents suggested visibility of positive impacts and receiving validation from beneficiaries; and $29\%$ of respondents suggested professional opportunities for long term research involvement (i.e., salary support, pathways to promotion, sharing new opportunities, etc.) were key motivators driving sustainable interventions. Additional motivators included delivery of clear feedback and expectations ($11\%$), strong collaborations from authorities (i.e., local government officials, local researchers, stakeholders) ($14\%$), and availability of basic resources to carry out the intervention (i.e., minimal funding, administrative and research software, logistics) ($11\%$).
**Figure 2:** *Motivators to Sustaining Capacity-Building and/or Evidence-Based NCD Intervention Implementation.*
## Opportunities
Opportunities to support researchers and community members implementing and sustaining capacity building and evidence-based NCD intervention implementation are outlined in Figure 3. $42\%$ of responses included training, mentorship, and funding for junior researchers, followed by $21\%$ of involvement of key stakeholders (i.e., community-based partnerships, Ministry of Health), followed by $17\%$ identifying political commitment and support. Funding (i.e., public and private funding, financial analysis & incentives) ($12\%$), effective monitoring ($4\%$), and adherences and perceived benefits of the intervention were also identified as areas of opportunity ($4\%$).
**Figure 3:** *Opportunities to Continuously Support Researchers/Community Members to Implement and Sustain Capacity-Building and/or Evidence-Based NCD Interventions.*
## Discussion
This study examined the capabilities, opportunities and motivations for sustaining capacity building and evidence-based NCD intervention implementation across eight LMICs. Our study is in accordance with other research findings that discuss multi-level factors that impact sustainability such as political support, funding stability, partnerships, and program evaluation and adaptation [3, 6, 7]. Overall, these findings highlight the need for commitment from the various stakeholders including research funding agencies, national and local governments, national and global philanthropy and multilateral organizations to make progress in LMIC research capacity for NCDs [3, 21].
While there was high diversity in respondents, with over $50\%$ being PIs or Co-Is, the need for training, mentorship and funding for junior or early researchers was a prominent theme among all respondents. Training and mentorship are proven strategies that lead to scientific success for junior researchers [22]. The lack of support for early-stage investigators in LMICs interested in the global NCD field has resulted in numerous barriers [22], many of which were reported in the findings of this study, including lack of sufficient staff, lack of knowledge among providers and researchers on the research process as well as addressing the interplay between local contextual setting factors. Studies have consistently shown that LMIC investigators are best positioned to address health challenges given their understanding of context, such as the cultural and political climate and health system readiness, in their home countries [5]. While there is growing effort for access to training and mentorship to be a long-term goal of projects and institutions [16, 22, 23], including the uptake from TREIN sites within the GRIT Consortium [22, 24, 25], increased access to mentorship still need to be adopted more widely to continue building local research capacity.
An additional finding of this study highlighted in the Integrated Sustainability *Framework is* the visibility of positive impacts and receiving validation from beneficiaries [5]. This included seeing improved health and well–being, building capacity of healthcare workers, and appreciation as motivators for sustaining the work they are doing [22]. Visibility of the program impact can be addressed through comprehensive evaluation with a focus on process measures [26]. An additional way in which implementers can see positive impacts and receive validation of their efforts is to openly connect with the communities they work with [22]. Nearly a quarter of responses identified the involvement of key stakeholders such as community-based partnerships, and Ministry of Health (MOH) as an opportunity to continuously support researchers and community members. Stakeholder involvement in intervention implementation not only encourages community support and creates a program that is more likely to be sustained due to changed community social norms and increased usage [2, 27], but would allow researchers to engage with stakeholders on the program impact.
In addition to stakeholder engagement, partnering with policy makers and financing institutions in the planning and implementation of NCD research and capacity building is crucial for securing funding or other resources needed for the continuation of sustainability efforts post intervention [6]. Involving policy makers and funding agencies when developing implementation programs and research to consider sustainability allows for more appropriate planning and allocation of funds potentially resulting in a much better understanding of why and how some interventions and programs last and others do not [4]. Lastly, engaging with policy makers and funding agencies could address limited national funding and financial barriers that reduce access to hypertension medications in this study as well as in others [28, 29].
## Implications and recommendations
Sustaining EBIs remains challenging, especially in LMICs where resources may be scarce. Based in eight countries across three continents, the current study adds renewed perspectives on how sustainability can be planned for, and considered in implementation research, which has received limited scientific attention– particularly in LMIC contexts. Findings from this study may serve as a springboard to identify specifically where implementation gaps exist and where targeted strategies are necessary. Findings also points to the need for equitable participation and stakeholder engagement with implementation practitioners and research funders to exchange knowledge on what influences sustainability throughout the life cycle of an EBI and to understand the values of the organization/health system that supports the sustainability of EBIs. Future research consortia may consider supplements or non-competitive funding opportunities to advance both knowledge and action related to the sustainability of evidence-based NCD interventions in LMICs.
## Strengthens and limitations
This study has a number of strengths including the use of data and implementer/ researcher perspectives from eight LMICs, making the findings more generalizable. Second, the study was guided by the COM-B Model. Limitation of this study include the small sample size of survey responses. Additionally, the results were self-reported by respondents thus needing to be validated in a study of long-term project sustainability. Lastly, the structure of the survey grouped both capacity building and intervention implementation in the same questions. While these could have been surveyed as separate concepts, the structure and sustainability of the GRIT Consortium addresses both capacity building and intervention implementation as integrated approaches.
## Conclusion
This study describes the perspectives from key implementers of capacity building and NCD intervention implementation efforts across eight low-and-middle income countries. This study addresses a gap in literature by examining the sustainability of evidence-based NCD implementation. Addressing multilevel factors that influence the sustainability of capacity building and interventions will have notable implications for other global NCD efforts going forward. Current and future studies, as well as consortium networks, should account for sustainability barriers and facilitators outlined as it will strengthen program implementation and long-term outcomes.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Author contributions
JI developed the survey using the COM-B Model. AR coordinated survey distribution and data collection. AR and SM conducted the content analysis and drafted the manuscript. SM, AA, and CC all provided feedback on manuscript sections. All authors reviewed and approved the final manuscript. All authors contributed to the article and approved the submitted version.
## Funding
This study was supported by a grant from the National Heart, Lung and Blood Institute (NHLBI), (1-U01 HL 138638).
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Islam SM, Purnat TD, Phuong NT, Mwingira U, Schacht K, Froschl G. **Non-communicable diseases (NCDs) in developing countries: a symposium report**. *Global Health.* (2014) **10** 81. DOI: 10.1186/s12992-014-0081-9
2. Aifah A, Iwelunmor J, Akwanalo C, Allison J, Amberbir A, Asante KP. **The kathmandu declaration on global CVD/hypertension research and implementation science: a framework to advance implementation research for cardiovascular and other noncommunicable diseases in low- and middle-income countries**. *Glob Heart.* (2019) **14** 103-7. DOI: 10.1016/j.gheart.2019.05.006
3. Iwelunmor J, Blackstone S, Veira D, Nwaozuru U, Airhihenbuwa C, Munodawafa D. **Toward the sustainability of health interventions implemented in sub-Saharan Africa: a systematic review and conceptual framework**. *Implement Sci.* (2016) **11** 43. DOI: 10.1186/s13012-016-0392-8
4. Wiltsey Stirman S, Kimberly J, Cook N, Calloway A, Castro F, Charns M. **The sustainability of new programs and innovations: a review of the empirical literature and recommendations for future research**. *Implement Sci.* (2012) **7** 17. DOI: 10.1186/1748-5908-7-17
5. Shelton RC, Cooper BR, Stirman SW. **The sustainability of evidence-based interventions and practices in public health and health care**. *Annu Rev Public Health.* (2018) **39** 55-76. DOI: 10.1146/annurev-publhealth-040617-014731
6. Scheirer MA. **Is sustainability possible? A review and commentary on empirical studies of program sustainability**. *Am J Eval.* (2005) **26** 320-47. DOI: 10.1177/1098214005278752
7. Luke DA, Calhoun A, Robichaux CB, Elliott MB, Moreland-Russell S. **The program sustainability assessment tool: a new instrument for public health programs**. *Prev Chronic Dis.* (2014) **11** 130184. DOI: 10.5888/pcd11.130184
8. Peprah E, Iwelunmor J, Price L. **Assessing stakeholder engagement for translation research and implementation science in low- and middle-income countries: lessons from Ghana, Guatemala, India, Kenya, Malawi, Nepal, Rwanda, and Vietnam**. *Glob Heart.* (2019) **14** 99-101. DOI: 10.1016/j.gheart.2019.05.009
9. Archana S, Karmacharya BM, Rashmi M, Abhinav V, Meghnath D, Natalia O. **Stakeholder engagement in planning the design of a national needs assessment for cardiovascular disease prevention and management in Nepal**. *Glob Heart.* (2019) **14** 181-9. DOI: 10.1016/j.gheart.2019.05.002
10. Asante KP, Iwelunmor J, Apusiga K, Gyamfi J, Nyame S, Adjei KGA. **Uptake of task-strengthening strategy for hypertension (TASSH) control within community-based health planning services in Ghana: study protocol for a cluster randomized controlled trial**. *Trials.* (2020) **21** 825. DOI: 10.1186/s13063-020-04667-7
11. Baumann AA, Mutabazi V, Brown AL, Hooley C, Reeds D, Ingabire C. **Dissemination and implementation program in hypertension in Rwanda: report on initial training and evaluation**. *Glob Heart.* (2019) **14** 135-41. DOI: 10.1016/j.gheart.2019.06.001
12. Ha DA, Tran OT, Nguyen HL, Chiriboga G, Goldberg RJ, Phan VH. **Conquering hypertension in Vietnam-solutions at grassroots level: study protocol of a cluster randomized controlled trial**. *Trials.* (2020) **21** 985. DOI: 10.1186/s13063-020-04917-8
13. Mercer T, Njuguna B, Bloomfield GS, Dick J, Finkelstein E, Kamano J. **Strengthening referral networks for management of hypertension across the health system (STRENGTHS) in western Kenya: a study protocol of a cluster randomized trial**. *Trials.* (2019) **20** 554. DOI: 10.1186/s13063-019-3661-4
14. Paniagua-Avila A, Fort MP, Glasgow RE, Gulayin P, Hernandez-Galdamez D, Mansilla K. **Evaluating a multicomponent program to improve hypertension control in Guatemala: study protocol for an effectiveness-implementation cluster randomized trial**. *Trials.* (2020) **21** 509. DOI: 10.1186/s13063-020-04345-8
15. Patel SA, Sharma H, Mohan S, Weber MB, Jindal D, Jarhyan P. **The integrated tracking, referral, and electronic decision support, and care coordination (I-TREC) program: scalable strategies for the management of hypertension and diabetes within the government healthcare system of India**. *BMC Health Serv Res.* (2020) **20** 1022. DOI: 10.1186/s12913-020-05851-w
16. Muula AS, Hosseinipour MC, Makwero M, Kumwenda J, Lutala P, Mbeba M. **Mentoring upcoming researchers for non-communicable diseases' research and practice in Malawi**. *Trials.* (2021) **22** 65. DOI: 10.1186/s13063-020-05006-6
17. Michie S, West R, Campbell R, Brown J, Gainforth H. *ABC of Behaviour Change Theories* (2014)
18. Michie S, van Stralen MM, West R. **The behaviour change wheel: a new method for characterising and designing behaviour change interventions**. *Implement Sci.* (2011) **6** 42. DOI: 10.1186/1748-5908-6-42
19. Iwelunmor J, Ezechi O, Obiezu-Umeh C, Gbajabiamila T, Musa AZ, Oladele D. **Capabilities, opportunities and motivations for integrating evidence-based strategy for hypertension control into HIV clinics in Southwest Nigeria**. *PLoS ONE.* (2019) **14** e0217703. DOI: 10.1371/journal.pone.0217703
20. Tong AS, Craig J. **Consolidated criteria for reporting qualitative research (COREQ): a 32-item checklist for interviews and focus groups**. *Int J Qual Health Care.* (2007) **19** 349-57. DOI: 10.1093/intqhc/mzm042
21. Malekzadeh A, Michels K, Wolfman C, Anand N, Sturke R. **Strengthening research capacity in LMICs to address the global NCD burden**. *Glob Health Action.* (2020) **13** 1846904. DOI: 10.1080/16549716.2020.1846904
22. Sharma A, Chiliade P, Michael Reyes E, Thomas KK, Collens SR, Rafael Morales J. **Building sustainable organizational capacity to deliver HIV programs in resource-constrained settings: stakeholder perspectives**. *Glob Health Action.* (2013) **6** 22571. DOI: 10.3402/gha.v6i0.22571
23. Hooley C, Baumann AA, Mutabazi V, Brown A, Reeds D, Cade WT. **The TDR MOOC training in implementation research: evaluation of feasibility and lessons learned in Rwanda**. *Pilot Feasibility Stud.* (2020) **6** 66. DOI: 10.1186/s40814-020-00607-z
24. van Oosterhout JJ, Hosseinipour M, Muula AS, Amberbir A, Wroe E, Berman J. **The Malawi NCD BRITE Consortium: building research capacity, implementation, and translation expertise for noncommunicable diseases**. *Glob Heart.* (2019) **14** 149-54. DOI: 10.1016/j.gheart.2019.05.004
25. Shrestha A, Maharjan R, Karmacharya BM, Bajracharya S, Jha N, Shrestha S. **Health system gaps in cardiovascular disease prevention and management in Nepal**. *BMC Health Serv Res.* (2021) **21** 655. DOI: 10.1186/s12913-021-06681-0
26. Kalolo A, Radermacher R, Stoermer M, Meshack M, De Allegri M. **Factors affecting adoption, implementation fidelity, and sustainability of the redesigned community health fund in Tanzania: a mixed methods protocol for process evaluation in the Dodoma region**. *Glob Health Action.* (2015) **8** 29648. DOI: 10.3402/gha.v8.29648
27. McGuire C, Calhoun LM, Mumuni T, Maytan-Joneydi A, Odeku M, Speizer IS. **Government stakeholders' perspectives on the family planning environment in three Nigerian cities: qualitative findings from the nigerian urban reproductive health initiative (NURHI) sustainability study**. *Glob Health Action.* (2020) **13** 1847821. DOI: 10.1080/16549716.2020.1847821
28. Sarfo FS, Mobula LM, Burnham G, Ansong D, Plange-Rhule J, Sarfo-Kantanka O. **Factors associated with uncontrolled blood pressure among Ghanaians: evidence from a multicenter hospital-based study**. *PLoS ONE* (2018) **13** e0193494-e. DOI: 10.1371/journal.pone.0193494
29. Tran DN, Manji I, Njuguna B, Kamano J, Laktabai J, Tonui E. **Solving the problem of access to cardiovascular medicines: revolving fund pharmacy models in rural western Kenya**. *BMJ Glob Health.* (2020) **5** e003116. DOI: 10.1136/bmjgh-2020-003116
|
---
title: 'Age Moderates the Effect of Obesity on Mortality Risk in Critically Ill Patients
With COVID-19: A Nationwide Observational Cohort Study*'
authors:
- Corstiaan A. den Uil
- Fabian Termorshuizen
- Wim J. R. Rietdijk
- Roos S. G. Sablerolles
- Hugo P. M. van der Kuy
- Lenneke E. M. Haas
- Peter H. J. van der Voort
- Dylan W. de Lange
- Peter Pickkers
- Nicolette F. de Keizer
journal: Critical Care Medicine
year: 2023
pmcid: PMC10012838
doi: 10.1097/CCM.0000000000005788
license: CC BY 4.0
---
# Age Moderates the Effect of Obesity on Mortality Risk in Critically Ill Patients With COVID-19: A Nationwide Observational Cohort Study*
## Body
The COVID-19 pandemic caused a surge of patients admitted to ICUs worldwide. In turn, the pandemic initiated research efforts focused on finding determinants, including body mass index (BMI), of outcome. Patients with higher BMI are more likely to experience a more severe course of COVID-19 [1]. However, once admitted to the ICU, (severe) obesity does not clearly drive mortality as there is conflicting literature (2–4). Age and BMI in critically ill patients with COVID-19 are inversely correlated [5]. Although there is much attention on BMI as a risk factor for mortality, the influence of this factor has not been examined across different age strata. Previous studies may have been underpowered as younger patients have been underrepresented in clinical studies [6, 7]. We, therefore, aimed to describe and explore the association between obesity and hospital mortality of patients admitted to the ICU with COVID-19 across different age strata using updated Netherlands Intensive Care Evaluation (NICE) registry data.
## OBJECTIVES:
A high body mass index (BMI) is associated with an unfavorable disease course in COVID-19, but not among those who require admission to the ICU. This has not been examined across different age groups. We examined whether age modifies the association between BMI and mortality among critically ill COVID-19 patients.
### DESIGN:
An observational cohort study.
### SETTING:
A nationwide registry analysis of critically ill patients with COVID-19 registered in the National Intensive Care Evaluation registry.
### PATIENTS:
We included 15,701 critically ill patients with COVID-19 (10,768 males [$68.6\%$] with median [interquartile range] age 64 yr [55–71 yr]), of whom 1,402 ($8.9\%$) patients were less than 45 years.
### INTERVENTIONS:
None.
### MEASUREMENTS AND MAIN RESULTS:
In the total sample and after adjustment for age, gender, Acute Physiology and Chronic Health Evaluation IV, mechanical ventilation, and use of vasoactive drugs, we found that a BMI greater than or equal to 30 kg/m2 does not affect hospital mortality (adjusted odds ratio [ORadj] = 0.98; $95\%$ CI, 0.90–1.06; $$p \leq 0.62$$). For patients less than 45 years old, but not for those greater than or equal to 45 years old, a BMI greater than or equal to 30 kg/m2 was associated with a lower hospital mortality (ORadj = 0.59; $95\%$ CI, 0.36–0.96; $$p \leq 0.03$$).
### CONCLUSIONS:
A higher BMI may be favorably associated with a lower mortality among those less than 45 years old. This is in line with the so-called “obesity paradox” that was established for other groups of critically ill patients in broad age ranges. Further research is needed to understand this favorable association in young critically ill patients with COVID-19.
## Data Collection and Study Population
We used patient data included in the national NICE quality registry, in which demographics, physiological and diagnostic data, ICU characteristics, and patient outcomes from all ICUs are registered [8]. The data are prospectively collected. We included all adult patients (age >18 yr) who were admitted to the ICU between March 1, 2020, and January 1, 2022, with a confirmed COVID-19 infection. Compared with a previous study [2], we used the same registry data over the first half year of the pandemic, but now were able to extend the inclusion period up to 22 months. The Scientific Board of the NICE foundation (number 2021-01) a priori approved this study and its exploratory design, and the study was approved by the medical ethics committee of the Erasmus MC (MEC 2021-0646, August 26, 2021, title: “Does Obesity Interact With Age in Explaining Hospital Mortality in Critically Ill COVID-19 Patients? A Nationwide Registry Analysis”) that waived the need for informed consent. Procedures were followed in accordance with the ethical standards of the responsible MEC on human experimentation and with the Helsinki Declaration of 1975.
## Study Variables
For baseline characteristics, we included patient characteristics, comorbidities, admission characteristics including Acute Physiology and Chronic Health Evaluation (APACHE)-IV probability, complications during first 24 hours after ICU admission, and clinical outcomes. For patient characteristics, we included age, sex, and BMI and presented these as continuous variables (median [interquartile range (IQR)]) or as number (percentage), where appropriate. Length and weight were preferably measured but could be estimated. BMI was subdivided in several categories, and obesity was defined as a BMI greater than or equal to 30 kg/m2. Comorbidities are defined as chronic obstructive pulmonary disease/respiratory insufficiency, renal insufficiency (creatinine >177 μmol/L [2.0 mg/dL], or renal insufficiency in the medical history), liver cirrhosis, severe heart failure (NYHA class IV), malignancy including hematological, immune deficiency, and diabetes mellitus. In addition, we examined the number of comorbidities (i.e., 0, 1, and ≥2). We noted whether a patient was intubated and thus mechanically ventilated prior to or immediately after ICU admission. Events occurring the first 24 hours of ICU admission, like the start of mechanical ventilation in the first 24 hours, acute renal failure, and administration of vasoactive medication, were collected. Clinical outcomes are defined as inhospital mortality, ICU length of stay, and hospital length of stay.
## Study Endpoint
The endpoint is all-cause hospital mortality.
## Statistical Analysis
We described the study population in three age strata (i.e., <45, 45–65, and >65 yr), pragmatically chosen based on strata sizes and on previous studies [9]. We analyzed the data in each age stratum according to the presence of obesity, and we compared survivors and nonsurvivors. Comparison of groups was done using a χ2 or Fisher exact test and with a Mann-Whitney U test or Kruskal Wallis test, when appropriate. We performed binary logistic regression models with hospital mortality as the outcome variable. We analyzed the associations between obesity (BMI ≥30 kg/m2) and hospital mortality stratified by age in three categories. This stratification was done by inclusion of terms for interaction of age × BMI. We built a multivariate model, where we included these terms for interaction and adjusted for APACHE-IV mortality probability in quintiles, age as continuous variable, gender, mechanical ventilation upon ICU admission, lowest Pao2/Fio2 ratio in quintiles, and the use of vasoactive drugs in the first 24 hours following ICU admission [2].
During the initial analysis, we found that the association between BMI and hospital mortality may be only present in patients under 45 years. For this reason, we decided to perform a post hoc analysis in this younger patient stratum. In this post hoc analysis, we explored whether one of the other study variables as confounders may explain the association between BMI and mortality. We examined the associations between obesity and hospital mortality controlled for several study variables in separate bivariate logistic regression models. ( We used the following factors in these post-hoc regression models: gender, immuno-insufficiency, renal insufficiency, respiratory insufficiency, malignancy, cardiovascular disease, liver cirrhosis, at least one comorbidity, APACHE IV probability, diabetes mellitus, acute renal failure, mechanical ventilation [upon admission and in the first 24 hours], and the administration of vasoactive drugs.) For these regressions, we estimated the odds ratio (OR) and $95\%$ CI. We examined statistical significance ($p \leq 0.05$) using a postestimation Wald test.
We performed sensitivity analyses for the univariate association between obesity and hospital mortality with differing BMI thresholds and age cutoffs. To check the assumption of linearity in our main multivariate model, we built an alternative multivariate model by entering age using refined categories (<45, 45–55, 55–60, 60–65, 65–70, 70–75, 75–80, and ≥80 yr) instead of a continuous variable in addition to age in broad categories. To check the initial results, we also built an alternative multivariable model where we included comorbidities and the terms for interaction of comorbidities × age and subdivided BMI in three categories (<25, 25–30, and >30 kg/m2). Regression diagnostics were performed to assess model fit (by eyeballing the calibration plot of $10\%$-categories of predicted versus observed mortality and by using the Hosmer-Lemeshow test and the deviance statistic) and to examine the potential for collinearity (through calculation of variance inflation factors [VIFs]).
To assess the impact of missing data on the results, we performed a sensitivity analysis using multiple imputation by chained equations of missing data for APACHE-IV probability, Pao2/Fio2, weight, length, and BMI. To assess the factor of time, we adjusted our main model by entering a categorized time variable, representing the series of COVID-19 waves and the periods in between. We also assessed the multivariate dose-response association between BMI and the risk for mortality from COVID-19 using BMI cutoffs of 25 and 30. Finally, to assess the impact of comorbidities on the association between BMI and mortality, we added the number of comorbidities to the main model and included the terms for interaction of age × BMI and age × number of comorbidities.
## RESULTS
We included 15,701 critically ill patients with COVID-19 (10,768 males, $68.6\%$, median age 64 [IQR, 55–71]). Supplementary Table 1 (http://links.lww.com/CCM/H283) presents the characteristics of the sample and for each age stratum separately. The median APACHE-IV mortality probability at admission was 0.22 (IQR, 0.14–0.34). As for the younger patients ($$n = 1$$,402; $8.9\%$), the APACHE-IV probability was 0.10 (IQR, 0.07–0.16). The median BMI in the youngest patients was 30.5 (26.6–35.6), and this was significantly higher compared with the 45–65 years (29.4 [IQR, 26.3–33.3]) and greater than 65 year subgroups (27.8 [IQR, 25.1–31.1]). *In* general, the younger patients had less comorbidities compared with older patients. In Supplementary Table 2 (http://links.lww.com/CCM/H283), we present the clinical characteristics for each age stratum according to BMI.
## Hospital Mortality Stratified by Age
Hospital mortality was $5.5\%$, $16.7\%$, and $42.1\%$ ($p \leq 0.001$) for patients less than 45, 45–65, and greater than 65 years, respectively. The association of different levels of BMI with hospital mortality across the three age strata is listed in Supplementary Table 3 (http://links.lww.com/CCM/H283). As for younger patients (<45 yr), we found differences in the number of comorbidities between survivors and nonsurvivors. Among the survivors, $81.2\%$ had no comorbidities, whereas in the nonsurvivors, $61.0\%$ had no comorbidities ($p \leq 0.05$).
## Hospital Mortality According to Obesity and Its Moderation by Age
Table 1 presents the logistic regression analysis (including 15,321 out of the 15,701 = $97.6\%$ of the total study sample) for the association between obesity and hospital mortality. In the total sample and without adjustment, we found that obesity is associated with a lower hospital mortality risk (OR, 0.74; $95\%$ CI, 0.69–0.80; $p \leq 0.001$). After age stratification, we found a significant association between BMI greater than or equal to 30 kg/m2 and hospital mortality in patients less than 45 years (OR, 0.58; $95\%$ CI, 0.36–0.93; $$p \leq 0.02$$). This association was not present in patients 45–65 years old and in those greater than 65 years old. The terms for interaction, however, did not reach the level of statistical significance ($$p \leq 0.12$$). The multivariate regression results showed that, in the total sample, the association disappeared, but the significant association between BMI greater than or equal to 30 kg/m2 and hospital mortality in patients less than 45 years (OR, 0.59; $95\%$ CI, 0.36–0.96; $$p \leq 0.03$$) remained. This association was not found in patients 45–65 years old (OR, 1.05; $95\%$ CI, 0.91–1.20) and in those greater than 65 years old (OR, 0.97; $95\%$ CI, 0.87–1.08). In the multivariate model, the terms for interaction of age × BMI were borderline significant ($$p \leq 0.08$$) and, thus, became stronger. When the regression analysis was performed in those patients admitted to the ICU with a primary diagnosis of viral pneumonia ($$n = 14$$,$\frac{425}{15}$,321 [$94.2\%$]), the OR for the association in patients less than 45 years remained similar in magnitude (OR, 0.61; $95\%$ CI, 0.36–1.04), though the significance disappeared ($$p \leq 0.07$$, terms for interaction $$p \leq 0.10$$).
**TABLE 1.**
| Model | Tested Category | Reference Category | Odds (Ref) | Probability (Ref) | OR | p | p for Interaction |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Univariate model | Obesity (BMI ≥ 30) | BMI < 30, all ages | 0.32 (0.43) | 0.24 (0.30) | 0.74 (0.69–0.80) | < 0.001 | 0.12 |
| Univariate model | Obesity (BMI ≥ 30; < 45 yr) | BMI < 30, <45 yr | 0.04 (0.07) | 0.04 (0.07) | 0.58 (0.36–0.93) | 0.02 | 0.12 |
| Univariate model | Obesity (BMI ≥ 30; 45–65 yr) | BMI < 30, 45–65 yr | 0.19 (0.20) | 0.16 (0.17) | 0.97 (0.85–1.10) | 0.6 | 0.12 |
| Univariate model | Obesity (BMI ≥ 30; > 65 yr) | BMI < 30, > 65 yr | 0.68 (0.75) | 0.41 (0.43) | 0.91 (0.82–1.01) | 0.08 | 0.12 |
| Multivariate modela | Obesity (BMI ≥ 30) | BMI < 30, all ages | | | 0.98 (0.90–1.06) | 0.62 | 0.08 |
| Multivariate modela | Obesity (BMI ≥ 30; < 45 yr) | BMI < 30, < 45 yr | | | 0.59 (0.36–0.96) | 0.03 | 0.08 |
| Multivariate modela | Obesity (BMI ≥ 30; 45–65 yr) | BMI < 30, 45–65 yr | | | 1.05 (0.91–1.20) | 0.52 | 0.08 |
| Multivariate modela | Obesity (BMI ≥ 30; > 65 yr) | BMI < 30, > 65 yr | | | 0.97 (0.87–1.08) | 0.55 | 0.08 |
## Post Hoc Analysis
Figure 1 presents the results of the post hoc analysis presenting the ORs for the association between obesity and hospital mortality when adjusting for several important clinical characteristics. The ORs remained similar in magnitude (OR between 0.50 and 0.65), though the significance disappeared when adjusting for APACHE-IV probability ($$p \leq 0.10$$). Supplementary Table 4 (http://links.lww.com/CCM/H283) presents the full results of the post hoc regression analysis.
**Figure 1.:** *Post hoc analysis results: Odds ratios (ORs) and 95% CIs of the association between obesity and hospital mortality adjusted for several important clinical characteristics in younger (<45 yr) patients. APACHE = Acute Physiology and Chronic Health Evaluation, BMI = body mass index.*
## Sensitivity Analyses
Supplementary Tables 5 and 6 (http://links.lww.com/CCM/H283) demonstrate the univariate regression analyses with threshold at BMI = 25 kg/m2 and BMI = 35 kg/m2, respectively. These analyses confirm the lowest mortality risk associated with being overweight (OR = 0.50) or morbidly obese (OR = 0.77) in the youngest group. Although the effect seemed to weaken at a BMI threshold greater than or equal to 35 kg/m2, the interaction terms were borderline significant in both analyses. Supplementary Tables 7 and 8 (http://links.lww.com/CCM/H283) show the univariate regression analyses using two different age cutoff values less than 40, 40–60, and greater than 60 years, and less than 55, 55–75, and greater than 75 years, respectively. This first analysis confirmed the lowest mortality risk, associated with obesity, in the youngest group. This effect disappeared in the second analysis where the youngest group was defined as less than 55 years. Entering age using refined categories into the model instead of the continuous variable for patients less than 45 years did not significantly change the results (data not shown). An alternative multivariable model where we included comorbidities and the terms for interaction of comorbidities × age, and subdivided BMI in three categories (<25, 25-30, and >30) showed again that the highest BMI category was associated with the lowest mortality for patients less than 45 years ($$p \leq 0.03$$, data not shown). Eyeballing our main model reflected very reasonable fit; however, the Hosmer-Lemeshow p value was significant, probably due to large patient numbers. Changing the main model by including age in refined categories did not improve model fit (H-L p value < 0.001). Assessing the model fit using the deviance statistic (in model with refined categories of age) and after entering age in refined categories (without age as continuous variable) experienced similar results (H-L $$p \leq 0.008$$ and $p \leq 0.001$, respectively). The fact that the model fit was not optimal when evaluated in terms of statistical significance is, therefore, not explained by the assumed linear relationship between log (odds) and age in the main analysis. There was no high correlation among the predictor variables (all VIFs <1.5), indicating no multicollinearity. There was no significant impact of missing data on the results (Supplementary Table 9, http://links.lww.com/CCM/H283). The results were similar when the main model was adjusted for the subsequent COVID-19 waves (data not shown). The dose-response association between BMI and the risk for mortality from COVID-19 using BMI cutoffs of 25 and 30 kg/m2 is demonstrated in Supplementary Table 10 (http://links.lww.com/CCM/H283), where we demonstrate that for patients less than 45 years, “being obese” was associated with a lower hospital mortality than “being overweight,” and patients with BMI less than 25 kg/m2 had the highest mortality. Supplementary Table 11 (http://links.lww.com/CCM/H283) demonstrates the regression analysis for the multivariate association between obesity and hospital mortality, after having added the number of comorbidities to the model, showing the ORs for all included variables.
## DISCUSSION
Although the literature is conflicting [3, 4], it was previously reported that the obesity paradox, that is a lower mortality in patients with a higher BMI, observed in various cohorts of critically ill patients, is not present in critically ill COVID-19 patients [2]. Using updated NICE data, we confirmed the lack of association between obesity and hospital mortality in the total sample. However, after stratification by age, we found for the younger patient group (<45 yr) a favorable effect of a higher BMI on survival, indicating that, in this subgroup, the obesity paradox is indeed apparent. This paradox is not explained by age-dependent differences in APACHE-IV scores, gender, respiratory, or other parameters. The effect was both significant and clinically relevant: a $40\%$ reduction in the odds of death. As the terms for interaction for age × BMI were borderline significant, our results suggest that obesity is an explanatory factor for lower hospital mortality among younger patients. This finding was further explored and confirmed using multiple post hoc and sensitivity analyses.
The question remains why a high BMI in younger patients results in a lower mortality. This may first be due to an unexplained biological mechanism, including a higher metabolic reserve in obese patients and differences in pulmonary mechanics and immunological aspects between obese and nonobese patients, especially in the young [10]. Second, unmeasured confounders may have resulted in confounding bias [11]. Confounding factors of the obesity-mortality relationship include unintended weight loss in the period preceding data collection, as well as data on premorbid physical wellness such as exercise tolerance, detailed preexisting heart disease, smoking, use of alcohol or drugs, socioeconomic status, and ethnicity (12–14). One may argue that particularly collider stratification bias may have partially explained our observations [15]. Collider stratification bias may arise when one investigates a patient sample within a specific stratum, that is, for our study young patients with COVID-19 who required admission to the ICU [11, 16, 17]. Younger obese patients have probably been less healthy or fit than in the hypothetical situation they would not have been obese. On the other hand, obese patients less than 45 years may just have been obese but may have suffered less from comorbidities. This is illustrated by the fact that young obese patients were less likely to have immune-insufficiency, and the highest tertile of APACHE-IV probability was less frequent in obese patients. We, therefore, performed the post hoc analysis and demonstrated that the association between BMI and mortality was consistent after adjusting for several important characteristics. It should be noted that the statistical significance disappeared when adjusting for the composite variable “APACHE-IV probability”; however, the consistent and large effect size (OR = 0.66) in addition to the multivariate regression results suggests that the effect or paradox is still present. Surge capacity issues may also have resulted in collider bias. Due to limited ICU capacity, one may expect a selection of patients with a higher BMI with COVID-19 but with less comorbidities and lower age to be admitted to the ICU, and this would plausibly translate into a better prognosis of these patients compared with patients with lower BMI but other prognostically less favorable reasons for ICU admission [13, 18, 19]. We found an inverse relation between age and BMI; however, a higher BMI was not associated with the number of comorbidities in young patients. Finally, the same difference in percentages that are closer to $50\%$ (as is the case with respect to mortality at higher ages) will lead to less extreme ORs, necessitating a larger sample size to reach the same statistical power. Thus, other factors such as high age and comorbidities may mask (“buffer”) the effect of high BMI at older age. Still, at young age, the largest difference in death rates between BMI greater than versus less than 30 was found. Thus, effect estimates both at a multiplicative and an additive scale suggest a survival benefit associated with a high BMI especially at young age.
Although we included all subsequent national ICU admissions from 22 months since the start of the pandemic, we acknowledge several methodological and other limitations. First, both the number of younger patients and death events in the young age group were relatively small and too limited to perform multiple subgroup analyses. Second, we mentioned methodological limitations including collider stratification bias above. Third, we assessed all-cause hospital mortality. It would have been informative to examine the cause of death between the subgroups [20], but this information is not available in the NICE registry, and there is unfortunately no record linkage possible with Statistics Netherlands. Fourth, our analyses might have suffered from some BMI group misallocation due to estimating rather than actually measuring height and weight. However, previous research demonstrated that either measurement or estimation of height and weight may not influence the association between BMI and mortality [21]. Fifth, although we assessed the factor of time, treatment changes over time as well as (varying) shortage of ICU beds could have influenced the outcome in different age groups and BMI categories. These data were not available or could not be analyzed in detail. Future research should focus on a better understanding of the obesity paradox in critically ill patients with COVID-19 in an even larger database, particularly in younger patients. It is for future research relevant to examine different statistical methods, including machine learning approaches, more comprehensively [11]. Further studies are also needed to investigate whether the obesity paradox observed for other critically ill patient groups is driven by the young.
## CONCLUSIONS
Overall, the obesity paradox is not present in critically ill patients with COVID-19, but we now report that it may emerge in younger (<45 yr) patients.
## References
1. Gao M, Piernas C, Astbury NM. **Associations between body-mass index and COVID-19 severity in 6.9 million people in England: A prospective, community-based, cohort study.**. *Lancet Diabetes Endocrinol* (2021) **9** 350-359. PMID: 33932335
2. Kooistra EJ, Brinkman S, van der Voort PHJ. **Body mass index and mortality in coronavirus disease 2019 and other diseases: A cohort study in 35,506 ICU patients.**. *Crit Care Med* (2022) **50** e1-e10. PMID: 34374504
3. Dana R, Bannay A, Bourst P. **Obesity and mortality in critically ill COVID-19 patients with respiratory failure.**. *Int J Obes* (2021) **1**
4. **Clinical characteristics and day-90 outcomes of 4244 critically ill adults with COVID-19: A prospective cohort study.**. *Intensive Care Med* (2021) **47** 60-73. PMID: 33211135
5. Kass DA, Duggal P, Cingolani O. **Obesity could shift severe COVID-19 disease to younger ages.**. *Lancet* (2020) **395** 1544-1545. PMID: 32380044
6. Bhasin A, Nam H, Yeh C. **Is BMI higher in younger patients with COVID-19? Association between BMI and COVID-19 hospitalization by age.**. *Obesity* (2020) **28** 1811-1814. PMID: 32610371
7. Liu Y, Mao B, Liang S. **Association between age and clinical characteristics and outcomes of COVID-19.**. *Eur Respir J* (2020) **55** 2001112. PMID: 32312864
8. Groenland CNL, Termorshuizen F, Rietdijk WJR. **Emergency department to ICU time is associated with hospital mortality: A registry analysis of 14,788 patients from six university hospitals in The Netherlands.**. *Crit Care Med* (2019) **47** 1564-1571. PMID: 31393321
9. Todorov A, Kaufmann F, Arslani K. **Gender differences in the provision of intensive care: A Bayesian approach.**. *Intensive Care Med* (2021) **47** 577-587. PMID: 33884452
10. De Jong A, Wrigge H, Hedenstierna G. **How to ventilate obese patients in the ICU.**. *Intensive Care Med* (2020) **1**
11. Decruyenaere A, Steen J, Colpaert K. **The obesity paradox in critically ill patients: A causal learning approach to a casual finding.**. *Crit Care* (2020) **24** 485. PMID: 32758295
12. Ades PA, Savage PD. **The obesity paradox: Perception vs knowledge.**. *Mayo Clin Proc* (2010) **85** 112-114. PMID: 20118385
13. **Clinical presentation, disease course, and outcome of COVID-19 in hospitalized patients with and without pre-existing cardiac disease: A cohort study across 18 countries.**. *Eur Heart J* (2022) **43** 1104-1120. PMID: 34734634
14. Banack HR, Stokes A. **The “obesity paradox” may not be a paradox at all.**. *Int J Obes (Lond)* (2017) **41** 1162-1163. PMID: 28584251
15. Sperrin M, Candlish J, Badrick E. **Collider bias is only a partial explanation for the obesity paradox.**. *Epidemiology* (2016) **27** 525-530. PMID: 27075676
16. Emerson P, Brooks D, Quasim T. **Factors influencing intensive care admission: A mixed methods study of EM and ICU.**. *Eur J Emerg Med* (2017) **24** 29-35. PMID: 27984369
17. Banack HR, Kaufman JS. **Does selection bias explain the obesity paradox among individuals with cardiovascular disease?**. *Ann Epidemiol* (2015) **25** 342-349. PMID: 25867852
18. Henkens M, Raafs AG, Verdonschot JAJ. **Age is the main determinant of COVID-19 related in-hospital mortality with minimal impact of pre-existing comorbidities, a retrospective cohort study.**. *BMC Geriatr* (2022) **22** 184. PMID: 35247983
19. Harwood R, Yan H, Talawila Da Camara N. **Which children and young people are at higher risk of severe disease and death after hospitalisation with SARS-CoV-2 infection in children and young people: A systematic review and individual patient meta-analysis.**. *EClinicalMedicine* (2022) **44** 101287. PMID: 35169689
20. Contou D, Cally R, Sarfati F. **Causes and timing of death in critically ill COVID-19 patients.**. *Crit Care* (2021) **25** 79. PMID: 33622349
21. Toft-Petersen AP, Wulff J, Harrison DA. **Exploring the impact of using measured or estimated values for height and weight on the relationship between BMI and acute hospital mortality.**. *J Crit Care* (2018) **44** 196-202. PMID: 29156253
|
---
title: 'Diagnostic performance of different anthropometric indices among Iranian adolescents
for intima media thickness in early adulthood: A prospective study and literature
review'
authors:
- Golaleh Asghari
- Ali Nikparast
- Maryam Mahdavi
- Pooneh Dehghan
- Majid Valizadeh
- Farhad Hosseinpanah
- Fereidoun Azizi
- Farzad Hadaegh
journal: Frontiers in Nutrition
year: 2023
pmcid: PMC10012864
doi: 10.3389/fnut.2023.1098010
license: CC BY 4.0
---
# Diagnostic performance of different anthropometric indices among Iranian adolescents for intima media thickness in early adulthood: A prospective study and literature review
## Abstract
### Background
There is debate regarding which anthropometric indices is the most appropriate predictor of cardiovascular disease (CVD) among adolescents. The purpose of this study was to investigate the association of body mass index (BMI), waist circumference (WC), waist-to-hip ratio (WHR), and waist-to-height ratio (WHtR) in adolescents with high carotid intima-media thickness (cIMT) in early adulthood, as the surrogate marker of CVD in a cohort study.
### Methods
A total of 875 Iranian adolescents (female = 421) aged 10–17 years old were entered the study. The cIMT was measured in early adulthood (20–38 years old) after 18.2 (median) years of follow-up and defined as > 90th percentile for sex and age groups. The gender specific association between a 1-SD increase in each anthropometric measures with high cIMT was examined using multivariate logistic regression analysis adjusted for age, smoking, family history of CVD, systolic blood pressure, total cholesterol, and fasting blood sugar. In the multivariable analysis, the interaction between sex and age groups with anthropometric measures were significant (all p-values < 0.05).
### Results
Among males, all anthropometric measures including BMI, WC, WHR, and WHtR were associated with high cIMT; the corresponding odds ratios were 1.43 (1.05–1.94), 1.63 (1.22–2.19), 1.33 (1.03–1.71), and 1.41 (1.07–1.87), respectively. However, after considering the related adulthood anthropometric measurements, the association remained significant for WC 1.48 (1.04–2.10) and WHR [1.28 (0.99–1.66), $$P \leq 0.06$$]. Moreover, among early adolescent boys aged 10–14 years, all of the anthropometric measures were significantly associated with high cIMT in the multivariate analysis that included the related adulthood anthropometric measures. The area under the curve (AUC) for the anthropometric measurements among males ranged from 0.576 for WHtR up to 0.632 for WC, without any superiority between them. Among females, only in linear regression analysis, a significant association were found between the higher value of WC and WHtR with cIMT measurement in adulthood; however, the risk reached to null after considering adult anthropometric measures.
### Conclusion
General and central obesity measures were significantly associated with high cIMT only among Iranian male adolescents, the relationship that were more prominent among pre-pubertal males.
## Introduction
Cardiovascular disease (CVD) is the leading cause of morbidity and mortality, accounting for nearly 17.9 million deaths annually and approximately $32\%$ of all deaths worldwide [1]. Despite the decline in CVD mortality in many industrialized countries, at least three out of four premature deaths were occur in low- and middle-income countries due to CVD and other non-communicable diseases [1]. Atherosclerosis, as a lifelong disease, is known as one of the major causes of CVD. Although the clinical manifestations of atherosclerosis usually appear after midlife, the disease process has a long silent stage that is initiated as early as childhood [2]. Therefore, early identification of individuals with subclinical atherosclerosis is crucial to implementing primary preventative strategies to reduce CVD risk in adulthood [3].
Several studies have shown a link between childhood cardiometabolic risk factors, including elevated blood pressure, glucose intolerance, abnormal blood lipids, and obesity, with the development of subclinical and clinical CVD in adulthood (4–7). Carotid intima-media thickness (cIMT), as measured by ultrasonography, is a validated, sensitive, and reproducible technique for detecting and quantifying subclinical atherosclerosis in asymptomatic individuals that corresponds with the development of coronary artery disease and has been identified as a predictor of future cardiovascular events as well as all-cause mortality [8]. According to the meta-analysis conducted in 2013, one standard deviation increase in cIMT increased the risk of myocardial infarction and stroke by 26 and $31\%$, respectively [9]. Furthermore, increased cIMT was found to be more strongly associated with mortality and cardiovascular endpoints in individuals younger than 50 years [10].
In spite of the efforts of the World Health Organization to achieve the goal of “halting the increase in obesity prevalence by 2025,” none of the countries have met this objective, and it is predicted that one out of eight children and adolescents in the world will be obese by 2030 [11]. According to a systematic review investigating childhood risk factors and adulthood CVDs risk, 19 of the 20 studies have reported a significant association between childhood adiposity and thicker cIMT in adulthood [4]. In a study published by Hosseinpanah et al. childhood body mass index (BMI) was found to be a strong predictor of cIMT in early adulthood [12]. Furthermore, another study indicated that one standard deviation increase in adolescent BMI corresponded to a 2.3 μm increase in cIMT in ages between 27 and 30 years [13]. Although BMI is the most commonly used index for the evaluation of general obesity, it does not consider body fat distribution and abdominal fat mass, which is particularly associated with increased CVD risk [14].
According to our knowledge, this study represents the first of its kind to examine the association between childhood central obesity measurements, including waist circumference (WC), waist-to-hip ratios (WHR), and waist-to-height ratio (WHtR), in addition to body mass index (BMI) and risk of high cIMT in early adulthood. Therefore, in this prospective study, we investigate the association between childhood adolescent anthropometric indices, including BMI, WC, WHR, and WHtR, with high cIMT in early adulthood during about two decades of follow-up.
## Study population
The current longitudinal study was performed within the framework of the Tehran Lipid and Glucose Study (TLGS). The TLGS is an ongoing large scale population-based prospective study aiming to investigate and prevent non-communicable diseases (NCDs) risk factors and promote a healthy lifestyle to reduce these risk factors [15]. The study participants were followed up every 3 years according to standard methods to monitor their demographics, lifestyle, biochemical profile, clinical information, and anthropometric indices. The baseline survey was a cross-sectional study including 15005 participants aged ≥ 3 years old who were selected using a multistage random sampling method in district 13 of Tehran, conducted from 1999 to 2001. Survey 2 (2002–2005), survey 3 (2006–2008), survey 4 (2009–2011), survey 5 (2012–2015), and survey 6 (2016–2019) were prospective follow-up surveys; this cohort is still being followed up.
At the recruitment, we collected data on 2,660 participants aged between 10 and 17 years old. Exclusions included those with prevalent cancer ($$n = 1$$), corticosteroid users ($$n = 8$$), and those without follow up till survey 5 or 6 ($$n = 981$$), leaving 1,670 participants. Out of these individuals, following phone contact with them, 711 individuals did not attend the medical center or express their willingness to participate in the cIMT measurement. Of the remaining individuals who measured the cIMT ($$n = 959$$), (missing data of cIMT, $$n = 5$$) the data of valid IMT were available on 954 cases. Finally, after excluding those with prevalent CVD ($$n = 8$$), and those with missing data on covariates at the end of study ($$n = 71$$), 875 individuals (Male = 454) were entered into our data analysis. The study protocol has been approved by the institutional ethics committee of the Research Institute for Endocrine Sciences, affiliated to the Shahid Beheshti University of Medical Sciences, and was conducted in accordance with the principles of the Declaration of Helsinki. At the beginning of this study, written informed consent was obtained from all parents or legal guardians of participants under 18. As well, all participants ≥ 18 years signed a written informed consent form before participating in the investigations.
## Anthropometric and laboratory assessment
A detailed description of the TLGS protocol and laboratory procedures has been provided elsewhere [15]. Briefly, obtaining demographic information and taking anthropometric measurements were performed by trained examiners in accordance with standardized protocols. Weight was measured while the participants were minimally dressed and without shoes, using a digital scale (Seca 707, Hanover, MD, USA) and recorded to the nearest 100 g. Height was measured while standing, without shoes, with shoulders held in a normal position, using a tape meter. Waist circumference (WC) was measured at the midpoint level (umbilicus level) while the subjects were in a standing position using an un-stretched tape meter, without any pressure on the body surface, and hip circumference (HC) was measured over light clothing at the widest girth of the hip using a tape meter. These measurements were recorded to the nearest 0.1 cm. Body mass index [BMI = weight (kg)/square of height (m2)], waist-to-hip ratio [WHR = waist circumference (cm)/hip circumference (cm)], and waist-to-height ratio [WHtR = waist circumference (cm)/height circumference (cm)] were calculated. Systolic and diastolic blood pressures were measured by a qualified physician in a seated position following the participant’s rest for 15 min, using a standard mercury sphygmomanometer (calibrated by the Iranian Institute of Standards and Industrial Researches) with the cuff placed on the right brachial at the heart level. Blood pressure was measured twice, at least a 30-s interval, and then the average of two measurements was recorded as the participant’s blood pressure. After 12–14 h of overnight fasting, blood samples were taken from all subjects and centrifuged within 30–45 min of collection. All blood analyses were conducted at the TLGS research laboratory on the day of blood collection using commercially available laboratory kits (Pars Azmoon Inc., Tehran, Iran) adapted to a Selectra 2 auto analyzer (Vital Scientific, Spankeren, Netherlands). Fasting plasma glucose (FPG) was assayed using the glucose oxidase based on the enzymatic colorimetric technique. Inter- and intra-assay coefficients of variation were both $2.2\%$ for serum glucose. Serum total cholesterol (TC) and triglycerides (TGs) were also measured by enzymatic calorimetric methods with cholesterol esterase and cholesterol oxidase, and glycerol phosphate oxidase, respectively. Inter- and intra-assay coefficients of variation were 2 and $0.5\%$ for TC and 1.6 and $0.6\%$ for TGs, respectively. High-density lipoprotein cholesterol (HDL-C) was assayed following the precipitation of the apolipoprotein B-containing lipoproteins with phosphotungstic acid. The low-density lipoprotein Cholesterol (LDL-C) was calculated by the *Friedewald formula* based on the serum TC, TG, and HDL-C concentrations expressed in mg/dl if the serum TG concentrations were less than 400 mg/dL [16].
## Carotid intima-media thickness assessment
Two qualified radiologists, who were blinded to the study participant’s details, measured the intima-media thickness of extra carinal carotid arteries based on B-mode ultrasound imaging techniques using a linear 7.5–10 MHz probe (Samsung Medison SonoAce R3 ultrasound machine). The measurement was done on both right and left carotids in the supine position, with the neck extended and slightly rotated to the opposite side as a first step, transverse carotid scan was conducted to evaluate the subject’s anatomy, locate atherosclerotic plaques (if present), and determine the site of maximal wall thickening in the near or far wall. Afterward, longitudinal scans with different angles were obtained from the artery. In order to obtain optimal gray scale imaging of the carotid artery, measurements on plaque-free arterial segments were implemented according to optimal B-mode imaging criteria, which is defined as Clear visualization of far wall arterial interfaces with completely anechoic luminal content. A scan depth adjustment was attempted in order to bring the arterial lumen to the center of the image while setting the focal zone at the level of the arterial lumen. Generally, IMT was regarded as a hypo-echoic band between the arterial wall’s echogenic intimal and adventitial surfaces. The cIMT was calculated by measuring three locations on either side of the distal segment of the common carotid artery between the leading edge of the first and second echogenic lines. The average measurements were taken as the final measurement on each side. IMT measurements were sporadically conducted on the distal segments of both sides of the internal carotid artery and carotid bulb in subjects who met the optimal technique and image criteria. In the present study, as measurements taken along the left common carotid artery (LCCA) displayed less inter- and intra-observer variation and were more in line with laboratory test results, we used left common carotid artery far wall measurements (LCCA) for defining high cIMT. In order to test the rate of reliability agreement, cIMT was measured by both radiologists in a subsample of 30 participants, consisting of $66.7\%$ females with the mean age and BMI of 41.7 ± 10.7 years and 24.4 ± 5.5 kg/m2, respectively. The degree of agreement between the two radiologists with regard to the cIMT measurements was evaluated using the inter-class correlation coefficient (ICC) and their $95\%$ confidence intervals based on the two-way mixed-effects model. According to the ICC analysis, the ICC values and $95\%$ CI were 0.79 and 0.55–0.90, respectively. *In* general, the ICC ranges between 0 and 1, where values between 0.75 and 0.9 indicate good reliability [17].
## Definitions
Family history of premature cardiovascular diseases was defined as a prior history of myocardial infarction, stroke, or sudden cardiac death in a male first-degree relative or father grandparent < 55 years old and in a female first-degree relative or mother grandparents < 65 years old. A high cIMT (as a surrogate for subclinical atherosclerosis) was defined as CIMT greater than the 90th percentile values specific for sex and age group [18]. Smoking habits were classified into two groups: (a) current smokers, (b) past/never smokers.
## Statistical analysis
Baseline characteristics of subjects were shown as a mean ± standard deviation (SD) and median and inter-quartile range (IQR) 25–75 for normally distributed continuous variables and skewed-distributed continuous variables, respectively. Categorical variables of baseline characteristics were presented as frequency (percentages). Characteristics of participants at baseline and the end of follow-up between sex as well as respondent and non-respondent (those with missing data on covariates, those who did not have any follow up, and those who did not participate in the cIMT measurement) were compared using an independent sample t-test, Mann–Whitney U test, and Chi-square test as appropriate. In order to examine the effect modification of sex and age groups on the impact of each anthropometric measures for high cIMT, logistic regression analyses were conducted. In the multivariable models, the interaction between sex and age groups with BMI, WC, and WHtR were significant (all p-values < 0.05). Therefore, all analysis were conducted separately for boys and girls in two age groups.
The association between different adolescent anthropometric indices and high cIMT in early adulthood were evaluated by calculating the multiple-adjusted odds ratios (ORs) using logistic regression analysis. ORs and $95\%$ confidence intervals (CIs) were evaluated per 1-SD of BMI, WC, WHR, and WHtR as continuous variables for the total boys and girls as well as in age-stratified groups in each gender. Multiple linear regression was also used to investigate the effect of each adolescent anthropometric indices in cIMT in early adulthood. Accordingly, three adjusted models were constructed: Model 1 was adjusted for age, family history of CVD, and smoking; Model 2: further adjusted for systolic blood pressure, total cholesterol, and fasting plasma glucose; Model 3: Model 2 + adulthood relevant anthropometric measurements (i.e., for waist circumference, adulthood waist circumference was included as another covariate). The area under the receiver-operating characteristic curve (AUC) and $95\%$ confidence interval was used to evaluate the predictive ability of each adolescent anthropometric indices, followed by a comparison of all adolescent anthropometric indices using the Delong test [19]. In addition, Youden’s index (sensitivity + specificity-1) was used to obtain the best cut-off from the ROC curve. The values for the maximum of the Youden’s index were considered the optimal cut-off points [20]. All statistical analyses were conducted using SPSS version 20 (SPSS, Chicago, IL, USA) regarding a two-tailed P-value of < 0.05 as a significant.
## Results
At the recruitment time, the study population included 454 boys and 421 girls with mean ages of 13.3, and 13.5, years old respectively; the corresponding values for BMI were 19.6 and 20.1 (Kg/m2), respectively. The characteristics and cardiometabolic profile of the participants according to gender are shown in Table 1. At baseline, girls had significantly higher WC, HC, WHtR, TC, TG, and LDL-C than boys. However, boys were taller compared to girls and had higher values of WHR, SBP, and HLD-C. No significant differences were seen regarding age, weight, BMI, DBP, and FPG between boys and girls.
**TABLE 1**
| Variable | Male (n = 454) | Female (n = 421) | p-value |
| --- | --- | --- | --- |
| Baseline | Baseline | Baseline | Baseline |
| Age (years) | 13.3 ± 2.1 | 13.5 ± 2.2 | 0.13 |
| Weight (kg) | 49.8 ± 16.2 | 48.4 ± 11.9 | 0.14 |
| Height (cm) | 157.5 ± 14.4 | 154.5 ± 8.7 | <0.001 |
| BMI (kg/m2) | 19.6 ± 4.1 | 20.1 ± 3.9 | 0.09 |
| WC (cm) | 67.5 ± 11.1 | 69.2 ± 9.1 | 0.02 |
| HC (cm) | 82 ± 11.3 | 88.8 ± 10.5 | <0.001 |
| WHR | 0.82 ± 0.06 | 0.78 ± 0.06 | <0.001 |
| WHtR | 0.43 ± 0.05 | 0.44 ± 0.05 | <0.001 |
| SBP (mmHg) | 106.1 ± 11.7 | 102.8 ± 11.2 | <0.001 |
| DBP (mmHg) | 70.1 ± 9.4 | 70.5 ± 9.1 | 0.57 |
| FPG (mg/dl) | 89.1 ± 7.5 | 88.1 ± 7.9 | 0.08 |
| Total cholesterol (mg/dl) | 166.1 ± 34.3 | 171.4 ± 30.8 | 0.02 |
| Triglycerides (mg/dl) | 91.5 (67–127.7) | 101 (77–139) | <0.01 |
| HDL-C (mg/dl) | 44 ± 10.6 | 42.4 ± 10.4 | 0.02 |
| LDL-C (mg/dl) | 100.3 ± 30.1 | 105.8 ± 27.7 | <0.01 |
| End of follow-up | End of follow-up | End of follow-up | End of follow-up |
| Age of cIMT measured (years) | 31.9 ± 2.3 | 32.4 ± 2.4 | <0.01 |
| Weight (kg) | 84.9 ± 15.9 | 66 ± 12.3 | <0.001 |
| Height (cm) | 176.4 ± 6.6 | 160.4 ± 5.7 | <0.001 |
| BMI (kg/m2) | 27.2 ± 4.6 | 25.7 ± 4.8 | <0.001 |
| WC (cm) | 94.4 ± 11.6 | 84.8 ± 10.6 | <0.001 |
| HC (cm) | 99.8 ± 8.1 | 100.9 ± 8.9 | 0.06 |
| WHR | 0.94 ± 0.05 | 0.84 ± 0.07 | <0.001 |
| WHtR | 0.53 ± 0.06 | 0.52 ± 0.06 | 0.16 |
| cIMT (mm) | 0.54 ± 0.10 | 0.57 ± 0.09 | <0.001 |
| Smoking no. (%) | 157 (34.6) | 25 (5.9) | <0.001 |
| Family history CVD no. (%) | 12 (2.6) | 11 (2.6) | 0.98 |
After the median (IQR) follow-up duration 18.2 (17.8–18.8) years, cIMT was measured in early adulthood (the mean age was 32.1 ± 2.3 years), and the mean value was 0.55 ± 0.09 mm. As shown in Table 1, at the end of the follow-up, the mean age of cIMT measurement was significantly higher among females than males. Also, females had significantly higher mean values of cIMT than males, however, the values of weight, BMI, WC, and WHR were higher in the latter. The males were more likely to be smoker than females. Although, no significant differences in WHtR and prevalence of family history of CVDs between genders were found.
The characteristics of respondents and non-respondents are provided in Supplementary Table 1. The respondents were younger, constituted a greater proportion of males, and had lower values of WC, HC, and DBP, however, no significant differences were found for other baseline characteristics between them.
Table 2 demonstrates the odds ratio and $95\%$ confidence intervals of each adolescent anthropometric indices for high cIMT (> 90th percentile) in early adulthood using multivariable logistic regression. In females, no significant association were found between each adolescent anthropometric indices and risk of high cIMT in early adulthood even in model 1. In male, all adolescent anthropometric indices were significantly associated with high cIMT in early adulthood in model 1 with ORs ranging from 1.35 for WHR to 1.60 for WC; after controlling multiple cardiometabolic risk factors (Model 2), all adolescent anthropometric indices were significantly associated with high cIMT in early adulthood, with ORs ranging from 1.33 for WHR to 1.63 for WC (all p-values < 0.05 for both models). After considering adulthood relevant anthropometric measurements as a confounder (model 3), all of the associations significantly attenuated excluding WC [1.48 ($95\%$ CI: 1.04–2.10)] and WHR [1.28 ($95\%$ CI: 0.99–1.66), p-value = 0.06]. According to age- and sex-stratified analysis, no significant association was found between each anthropometric indices and risk of high cIMT in early adulthood except in boys aged 15–17 years old. A 1-SD increase in each adolescent anthropometric indices in boys aged 10–14 years old, were significantly increased the risk of high cIMT in model 3, the corresponding values for BMI, WC, WHR, and WHtR were 1.81 (1.13–2.92), 2.04 (1.31–3.18), 1.36 (1.02–1.83), and 1.85 (1.22–2.80), respectively.
**TABLE 2**
| Unnamed: 0 | Male | Male.1 | Female | Female.1 |
| --- | --- | --- | --- | --- |
| | Odds ratio (95% CI) | P-value | Odds ratio (95% CI) | P-value |
| Total | Total | Total | Total | Total |
| Body mass index | Body mass index | Body mass index | Body mass index | Body mass index |
| Model 1 | 1.43 (1.09–1.87) | <0.01 | 1.21 (0.90–1.64) | 0.20 |
| Model 2 | 1.43 (1.05–1.94) | 0.02 | 1.25 (0.90–1.72) | 0.18 |
| Model 3 | 1.22 (0.83–1.79) | 0.31 | 0.84 (0.55–1.28) | 0.42 |
| Waist circumference | Waist circumference | Waist circumference | Waist circumference | Waist circumference |
| Model 1 | 1.60 (1.24–2.07) | <0.001 | 1.20 (0.87–1.65) | 0.27 |
| Model 2 | 1.63 (1.22–2.19) | <0.01 | 1.19 (0.85–1.68) | 0.31 |
| Model 3 | 1.48 (1.04–2.10) | 0.03 | 0.95 (0.65–1.39) | 0.78 |
| Waist-to-hip ratio | Waist-to-hip ratio | Waist-to-hip ratio | Waist-to-hip ratio | Waist-to-hip ratio |
| Model 1 | 1.35 (1.05–1.74) | 0.02 | 1.20 (0.89–1.61) | 0.22 |
| Model 2 | 1.33 (1.03–1.71) | 0.03 | 1.17 (0.86–1.58) | 0.32 |
| Model 3 | 1.28 (0.99–1.66) | 0.06 | 1.15 (0.84–1.56) | 0.38 |
| Waist-to-height ratio | Waist-to-height ratio | Waist-to-height ratio | Waist-to-height ratio | Waist-to-height ratio |
| Model 1 | 1.42 (1.11–1.82) | <0.01 | 1.18 (0.89–1.55) | 0.25 |
| Model 2 | 1.41 (1.07–1.87) | 0.01 | 1.18 (0.88–1.59) | 0.26 |
| Model 3 | 1.33 (0.95–1.85) | 0.10 | 0.97 (0.69–1.35) | 0.85 |
| 10–14 years old | 10–14 years old | 10–14 years old | 10–14 years old | 10–14 years old |
| Body mass index | Body mass index | Body mass index | Body mass index | Body mass index |
| Model 1 | 1.81 (1.27–2.59) | <0.01 | 1.16 (0.78–1.75) | 0.46 |
| Model 2 | 1.89 (1.25–2.84) | <0.01 | 1.21 (0.79–1.86) | 0.39 |
| Model 3 | 1.81 (1.13–2.92) | 0.01 | 0.69 (0.38–1.25) | 0.21 |
| Waist circumference | Waist circumference | Waist circumference | Waist circumference | Waist circumference |
| Model 1 | 1.95 (1.38–2.76) | <0.001 | 1.20 (0.80–1.80) | 0.38 |
| Model 2 | 2.00 (1.36–2.95) | <0.001 | 1.18 (0.76–1.83) | 0.45 |
| Model 3 | 2.04 (1.31–3.18) | <0.01 | 0.89 (0.54–1.46) | 0.65 |
| Waist-to-hip ratio | Waist-to-hip ratio | Waist-to-hip ratio | Waist-to-hip ratio | Waist-to-hip ratio |
| Model 1 | 1.39 (1.05–1.86) | 0.02 | 1.24 (0.87–1.77) | 0.23 |
| Model 2 | 1.37 (1.03–1.84) | 0.03 | 1.17 (0.81–1.70) | 0.39 |
| Model 3 | 1.36 (1.02–1.83) | 0.04 | 1.16 (0.80–1.68) | 0.44 |
| Waist-to-height ratio | Waist-to-height ratio | Waist-to-height ratio | Waist-to-height ratio | Waist-to-height ratio |
| Model 1 | 1.76 (1.27–2.46) | <0.01 | 1.16 (0.82–1.63) | 0.41 |
| Model 2 | 1.76 (1.22–2.54) | <0.01 | 1.17 (0.80–1.69) | 0.42 |
| Model 3 | 1.85 (1.22–2.80) | <0.01 | 0.88 (0.57–1.36) | 0.58 |
| 15–17 years old | 15–17 years old | 15–17 years old | 15–17 years old | 15–17 years old |
| Body mass index | Body mass index | Body mass index | Body mass index | Body mass index |
| Model 1 | 1.05 (0.68–1.62) | 0.83 | 1.28 (0.82–2.00) | 0.29 |
| Model 2 | 1.06 (0.64–1.76) | 0.81 | 1.29 (0.79–2.12) | 0.30 |
| Model 3 | 0.73 (0.36–1.49) | 0.38 | 1.03 (0.54–1.95) | 0.93 |
| Waist circumference | Waist circumference | Waist circumference | Waist circumference | Waist circumference |
| Model 1 | 1.21 (0.80–1.82) | 0.37 | 1.21 (0.72–2.04) | 0.48 |
| Model 2 | 1.27 (0.79–2.05) | 0.32 | 1.21 (0.68–0.2.13) | 0.52 |
| Model 3 | 0.91 (0.49–1.69) | 0.77 | 1.03 (0.54–1.97) | 0.93 |
| Waist-to-hip ratio | Waist-to-hip ratio | Waist-to-hip ratio | Waist-to-hip ratio | Waist-to-hip ratio |
| Model 1 | 1.36 (0.79–2.33) | 0.26 | 1.14 (0.65–2.02) | 0.65 |
| Model 2 | 1.47 (0.81–2.67) | 0.20 | 1.13 (0.63–2.02) | 0.68 |
| Model 3 | 1.24 (0.63–2.43) | 0.54 | 1.14 (0.63–2.06) | 0.67 |
| Waist-to-height ratio | Waist-to-height ratio | Waist-to-height ratio | Waist-to-height ratio | Waist-to-height ratio |
| Model 1 | 1.07 (0.72–1.60) | 0.73 | 1.21 (0.75–1.93) | 0.44 |
| Model 2 | 1.10 (0.69–1.73) | 0.70 | 1.20 (0.72–2.01) | 0.47 |
| Model 3 | 0.83 (0.46–1.50) | 0.54 | 1.07 (0.60–1.90) | 0.81 |
The early adolescent males were generally had better cardiometabolic health status both at the baseline and at the end of the follow up compared to late adolescent ones, however, no differences were found in term of cIMT between these groups (Supplementary Table 2). Among male participants, in the early adolescent group, regardless of baseline anthropometric measures, no significant associations were found for other confounder/mediators, however for late adolescent group the adulthood cholesterol level was significantly associated with high cIMT. Moreover, among female, in the early adolescent group, baseline FPG and the adulthood general and central adiposity measures were remained as significant predictors of high cIMT, however, no significant association were found for confounder/mediators in later adolescent ones (data not shown).
Sex and age group stratified association of adolescent anthropometric indices with cIMT (in mm) in early adulthood is also shown in Supplementary Table 3 using linear regression analysis. Accordingly, the result among males were generally in the line with the main analysis conducted using multivariate logistic regression. Importantly among females, however, in contrast to logistic regression analysis we found a significant association between WC, WHR, and WHtR and higher value of high cIMT in model 1, the relationship that remained significant only for WHtR after further adjustment for cardiometabolic risk factors (model 2).
Figure 1 illustrates the area under the receiver operating curve (AUC) and $95\%$ confidence intervals, sensitivity, specificity, and cut-off values of each adolescent anthropometric indices in boys for predicting high cIMT in early adulthood. The AUC levels ranged from 0.632 ($95\%$ CI: 0.585–0.676) for WC to 0.576 ($95\%$ CI: 0.529–0.622) for WHtR, the corresponding cut-off points were 72 (cm), and 0.47, respectively. Supplementary Table 4 shows the comparison of AUCs of adolescent anthropometric indices in boys. No significant differences were observed between the AUCs of adolescent anthropometric indices (all p-values > 0.05). However, we found a signal for higher discriminatory power of WC compared to BMI and WHtR in boys (p-value = 0.07).
**FIGURE 1:** *The area under the curves (AUCs*100), sensitivity (%), specificity (%), and cut-off values of adolescent anthropometric indices for predicting high cIMT in boys.*
## Discussion
This community-based study of Iranian adolescents with a two-decade follow-up period showed that anthropometric indices, including BMI, WC, WHR, and WHtR, can predict the risk of high cIMT in early adulthood among boys; however, these associations were not found in girls. Considering adjustment for a large set of covariates (including obesity mediators, and corresponding adulthood anthropometric measurements) among boys WC and WHR were significant predictors, although these relationships were tended to be significant for the latter. Importantly, among early but not late adolescent males, all of the anthropometric measurements were significantly associated with high cIMT. Among Iranian females, we found generally in linear regression analysis, a significant association between the higher value of WC and WHtR with cIMT measurement in adulthood, however, these risks reach to null after considering adult anthropometric measures. Regarding the discriminatory power of anthropometric measurement as assessed by AUC, no differences were found between the predictive performance of anthropometric measures in boys for high cIMT, however, a signal for higher discriminatory power of WC compared with BMI and WHtR were found (p-value = 0.07).
A number of prospective studies have demonstrated a link between pediatric obesity by measuring BMI, which represents general obesity, or triceps skinfold thickness (TSF), and the risk of thicker cIMT in adulthood (Table 3). In accordance with our findings, the international Childhood Cardiovascular Cohort (i3C) Consortium has shown that childhood BMI significantly predicted thicker cIMT in adulthood [21]. Furthermore, the pooled analysis of four longitudinal studies, including the Bogalusa Heart Study, the Insulin Study, Childhood Determinants of Adult Health, and the Cardiovascular Risk in Young Finns Study, have demonstrated that childhood BMI was associated with high cIMT (> 90th percentile) in adulthood after two decades of follow up [22].
**TABLE 3**
| References | Study cohort, country | Participants (n) | Age at baseline (year) | Adiposity index | Follow-up duration (year) | Age at follow-up (years) | Adulthood cIMT (mm) | Main finding(s) | Adjustment variables |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Davis et al. (34) | Muscatine Heart Study, United States | 725 (346 males) | 8–11 | BMI TSF | – | 33–42 | 0.79 ± 0.12 for men 0.72 ± 0.10 for women | 1-SD increase in childhood BMI was associated with 47% increases risk of high cIMT (> 75th percentile) in women | Childhood age |
| Oren et al. (13) | Atherosclerosis Risk in Young Adults (ARYA), Netherlands | 750 (352 males) | 12–16 | BMI | – | 27–30 | 0.49 ± 0.05 for men 0.48 ± 0.05 for women | 1-SD increase in adolescent BMI was associated with 2.3 μm increase in cIMT in young adults, but this association became non-significant when adult BMI adjusted | cIMT reader, childhood gender, lumen diameter, age, BP, pubertal stage, and adult BP, lipid levels, and BMI |
| Raitakari et al. (32) | Cardiovascular Risk in Young Finns Study, Finland | 2,229 (1,005 males) | 3–18 | BMI | 21 | 24–39 | 0.64 ± 0.11 for men 0.61 ± 0.09 for women | Greater BMI in childhood was associated with thicker cIMT in men (β = 0.013) and women (β = 0.04). In sub-group analysis, the association was only seen in the 12–18 years old These associations became non-significant when adult BMI adjusted | Childhood and adulthood age, gender, LDL-C, SBP, and smoking |
| Li et al. (35) | Bogalusa Heart Study, United States | 486 (191 males) | 4–17 | BMI | 22.2 | 25–37 | 0.757 ± 0.073 for men 0.719 ± 0.091 for women | 1-SD increase in childhood BMI was associated with 67% increases risk of high cIMT (highest quartile of cIMT z-score) | |
| Freedman et al. (36) | Bogalusa Heart Study, United States | 513 (203 males) | Six measure-ments between 4–35 years | BMI TSF | | 23–40 | 0.736 in total | BMI and TSF were significantly associated with high cIMT (≥ 90th percentile) in adulthood. Associations were stronger in women than men. This association attenuated when adult BMI was adjusted. The association was also significant among overweight children who remained obese in adulthood. | Race, age, gender, adult adiposity indices |
| Ferreira et al. (37) | Amsterdam Growth and Health Longitudinal Study (AGAHLS), Netherlands | 159 (75 males) | 13–16 | BMI TSF DXA | 24 | 36 | 0.62 ± 0.010 for men 0.63 ± 0.10 for women | Total body fatness and truncal subcutaneous fat accumulation (the latter in boys only) during childhood were positively associated with cIMT in adulthood | Gender, TC/HDL-C ratio, TG, HbA1C, resting heart rate and cardiopulmonary fitness in childhood, height and pulse pressure in adulthood |
| Juonala et al. (24) | Cardiovascular Risk in Young Finns Study, Finland | 2260 | 3–18 | BMI | | 29–34 | 0.622 in total | BMI in childhood correlated with adult cIMT, this association became non-significant when adjusted for adult BMI. The association was significant among those who gained weight and among overweight children who remained obese in adulthood. BMI measured at 12, 15, and 18 years old was only associated with cIMT in adulthood | Adult BMI |
| Li et al. (38) | Bogalusa Heart Study, United States | 868 (365 males) | 4–17 | BMI | 26.4 | 25–44 | 0.85 ± 0.016 for men 0.77 ± 0.13 for women | BMI only in black women was significant predictor of cIMT in young adults | Gender, race, SBP, HDL-C, LDL-C, TG |
| Freedman et al. (23) | Bogalusa Heart Study, United States | 1,142 (492 males) | 7–13 | BMI | | 36 | 0.87 ± 0.20 for men 0.77 ± 0.14 for women | Childhood BMI levels were related to cIMT in adulthood. This association attenuated when adult BMI was adjusted. | Gender, age, race, adult BMI |
| Juonala et al. (21) | International Childhood Cardiovascular Cohort (i3C) Consortium | 4,380 (2,002 males) | 3–18 | BMI | 22.4 | 20–45 | | Childhood BMI at or after 9 years of age was associated with higher cIMT in adulthood | |
| Juonala et al. (39) | Cardiovascular Risk in Young Finns Study, Finland | 1,809 (794 males) | 3–18 | BMI | 27 | 24–39 | – | Obesity was associated with 6-year change in adulthood cIMT. This association remained significant when adjusted for adulthood risk score and genotype score. | Childhood and adulthood HDL-C, physical activity, fruit consumption, and genotype score |
| Juonala et al. (18) | International Childhood Cardiovascular Cohort (i3C) Consortium | 6,328 (2,961 males) | 3–18 | BMI | 23.1 | 23–46 | 0.65 ± 0.12 in total | Overweight children who remained obese in adulthood had an increased risk of high cIMT (≥ 90th percentile). In sub-group analysis, the association was seen only in boys. The risks of high cIMT among overweight or obese children who became non-obese by adulthood were similar to those among persons who were never obese. | Gender, age, height, follow-up length, and cohort. |
| Magnussen et al. (40) | Bogalusa Heart Study and the Cardiovascular Risk in Young Finns Study | 1,781 (784 males) | 9–18 | BMI | 14–27 | 24–41 | | BMI alone was as good as and in some cases superior to dichotomous pediatric MetS definitions in predicting high cIMT in adulthood | Length of follow-up, cohort, and all other MetS components |
| Huynh et al. (41) | Childhood Determinants of Adult Health (CDAH), Australia | 2,328 (1,150 males) | 7–15 | BMI TSF | 20 | 26–36 | 0.58 ± 0.089 in total | Childhood body size or adiposity was associated with adulthood cIMT | Childhood and adulthood weight, height, childhood BP and age of menarche |
| Su et al. (42) | Young Taiwanese Cohort (YOTA) Study, Taiwan | 789 (313 males) | 6–18 | BMI | 8.5 | 21.32 | 0.438 ± 0.048 in total | Overweight/obese children and adolescents who remain obese or overweight in adulthood had higher risk of high cIMT (> 95th percentile) in adulthood | Gender, age, fasting glucose, TC, smoking, alcohol habit, household income |
| Koskinen et al. (43) | Cardiovascular Risk in Young Finns Study, Finland | 1,617 (766 males) | 9–24 | BMI TSF | 21–25 | 30–45 | | Children who were either overweight or metabolically abnormal had similar increased risk of elevated cIMT in adulthood | FBS, TG, HDL-C, LDL-C cholesterol, SBP in childhood, and BMI in adulthood as well as family history of coronary artery disease |
| Johnson et al. (44) | MRC National Survey of Health and Development Study, United Kingdom | 1,273 (604 males) | 2–20 | BMI Height | | 60–64 | 0.667 mm | Higher BMI associated with thicker cIMT in men only, and only at exposure ages 4 and 20 | Fathers’ education levels, BMI in childhood and adulthood, SBP, pulse pressure, HbA1C in adulthood as well as total energy intake in childhood |
| Ceponiene et al. (27) | Kaunas Cardiovascular Risk Cohort study, Lithuania | 380 (168 males) | 12–13 | BMI | 35 | 48–49 | 0.66 ± 0.11 for men 0.61 ± 0.08 for women | Higher BMI in childhood is associated with thicker cIMT in adulthood only in women. This association became non-significant when adult mediators adjusted | BMI, SBP, and sexual maturity score in childhood as well as BMI, SBP, HDL-C, LDL-C, smoking, and educational levels in adulthood |
| Yan et al. (25) | Beijing blood pressure cohort, China | 1,252 (692 males) | 6–18 | BMI | 22.9 | 27–42 | 0.54 ± 0.05 for men 0.5 ± 0.04 for women | Childhood BMI predicted high cIMT (> 75th percentile) in adulthood, this association became non-significant when adult BMI and SBP were adjusted. In addition, incremental BMI from childhood to adulthood predicted high cIMT in adulthood. | Follow-up length, TG, HDL-C, LDL-C, smoking, drinking, physical inactivity, and family history of stroke and coronary heart disease in adulthood |
| Du et al. (45) | Bogalusa Heart Study, United States | 1,052 (455 males) | 4–19 | BMI | 26.5 | 19–52 | 0.87 ± 0.19 for men 0.77 ± 0.13 for women | The association between childhood obesity and cIMT was only seen in individuals with low adiponectin levels. | Gender, race, age, LDL-C, HDL-C, TG, SBP in childhood as well as smoking in adulthood |
| Hao et al. (46) | Georgia Stress and Heart study, United States | 626 | 5–18 | BMI | | 24 | | Childhood trajectory of BMI was associated with cIMT in adulthood even after adjustment for adulthood BMI, but the associations for IMT were not significant after adjustment for BMI at baseline | Gender, race, age, father’s education level, SBP and DBP, BMI in adulthood, BMI in childhood |
| Koskinen et al. (22) | International Childhood Cardiovascular Cohort (i3C) Consortium | 2,893 (1,331 males) | 12–18 | BMI | 23.4 | | | Childhood BMI associated with high cIMT (> 90th percentile) even after lipid biomarkers adjustments | Gender, age, blood pressure, smoking, LDL-c, HDL-c, TG |
| Buscot et al. (26) | Cardiovascular Risk in Young Finns Study, Finland | 2,631 (1,208 males) | 6–18 | BMI | 30 | 34–49 | | Childhood adiposity was associated with high cIMT (> 90th percentile). The effect of youth obesity on the risk of high cIMT may not be reversible even with the normalization of high BMI in later life | Gender, year of birth, family history of CVD, socio-economic status, and physical activity level in adulthood |
| Hosseinpanah et al. (12) | Tehran lipid and glucose study (TLGS), Iran | 1,295 (670 males) | 3–18 | BMI | 18 | 29.7 ± 4 | 0.55 ± 0.1 | Childhood BMI associated with thicker cIMT in adulthood | Gender, age, family history of CVD, smoking, adulthood BMI |
| Tasdighi et al. (6) | Tehran lipid and glucose study (TLGS), Iran | 1,220 (631 males) | 10.9 ± 4 | BMI | 18 | 30 ± 3.8 | 0.55 ± 0.1 | Children with metabolically unhealthy obese phenotype was the only ones that had an increased risk of high CIMT incidence in early adulthood | Gender, age, family history CVD, smoking, adulthood BMI, and LDL |
The result of our study has shown that the predictive power of all anthropometric indices for the prediction of high cIMT in early adulthood was attenuated and only remained significant in WC among males after adjustment of adulthood relevant anthropometric measurements. Summarizing the results of a majority of longitudinal cohort studies, the positive associations between childhood BMI and thicker cIMT in adulthood were generally reduced after adjustment for adulthood BMI [23] or became non-significant [13, 24, 25]. These findings may provide some clues that the patterns of adiposity anthropometric indices trajectories from childhood to adulthood also plays an important role in predicting the thicker cIMT in adulthood along with childhood adiposity. Juonala et al. have demonstrated that childhood adiposity was associated with thicker cIMT in adulthood, but this association can be attributed to significant changes in BMI from childhood to adulthood [24]. They also reported that the corresponding adulthood cIMT values in individuals who were obesity gainers from childhood to adulthood compared to individuals who had been persistently obese were comparable. As well, the cIMT values tended to be lower, yet comparable, in individuals who had been consistently non-obese and those who had been obesity reducers from childhood to adulthood. However, The Cardiovascular Risk in Young Finns Study has indicated that the effect of childhood obesity on the risk of high cIMT, even with the normalization of high BMI in later life, has not been completely eliminated [26]. To our knowledge, no study compare the association between central adiposity measures with high cIMT in early adulthood.
We also found that the association between adolescent anthropometric indices and risk of high cIMT in early adulthood was observed only among boys. In line with our findings, in an I3C consortium of 6,380 participants, Juonala et al. have demonstrated that the mentioned association was seen in boys [18]. In contrast, another I3C Consortium has shown that no significant differences between gender were observed for the predictive power of childhood BMI for the risk of high cIMT in adulthood [21]. Additionally, Ceponiene et al. have indicated that higher childhood BMI was associated with thicker cIMT in women adults; that this association was not independent of adult cardiometabolic risk factors [27]. In the current study when we examined cIMT as a continuous rather than categorical variable as the outcome we found that Tehranian girls with higher WC had significantly higher values of cIMT in the presence of cardiometabolic risk factors. Considering the observational nature of our study, it is not possible for us to explain the sex difference in the association between anthropometric measures and subclinical atherosclerosis in early adulthood; however, previously we reported the prominent role of general and central adiposity measures in the development of CVD among adult men rather than women [28].
Studies that have evaluated childhood obesity and the risk of high cIMT in adulthood have mainly measured BMI as an obesity indicator (as shown in Table 3), which may fail to reflect central obesity [29]. The results from NHANES III, including the 15,184 participants aged 18–90 years old, showed that central adiposity increased the risk of CV mortality among individuals with normal BMI [30]. Wang et al. in the cohort of Chinese children, demonstrated that compared with persistently normal WC, gaining in abdominal obesity, whether as incident or persistent, were significantly associated with the short-term risk of high cIMT during a 2-year follow-up in childhood [31].
As an important finding, in our subgroup analysis by age group, we found that the unfavorable impact of increasing value for general and central adiposity indices on cIMT thickening were more prominent among early adolescent boys. According to the findings of the I3C consortium, Jounala et al. have shown that childhood BMI after age 9 years was significantly associated with high cIMT in adulthood [21]. In addition, Raitakari et al. have demonstrated that greater BMI in adolescents aged 12–18 years old was significantly associated with thicker cIMT in adulthood [32]. Importantly, other cardiometabolic risk factors besides anthropometric measurements specially baseline fasting glucose among early adolescent girl and adulthood cholesterol level among late adolescent boys had significant association with high cIMT in our study (data not shown); the role of these obesity mediators in the development of atherosclerosis were addressed in other studies as well [21, 33].
In light of these findings, it is reasonable to conclude that childhood adiposity beginning in early puberty, may cause progressive damage to the common carotid artery in adulthood.
## Strengths and limitations
Our study has several strengths. It is important to point out that not only childhood cardiometabolic confounders but also adulthood relevant anthropometric indices were considered in our analysis. However, there are also some limitations to be considered. First, although we adjusted major confounding variables in our analyses, residual or unmeasured factors, such as physical activity, dietary intakes, adipokines and cytokines, pubertal status, and genetic background were not considered in our data analysis. Second, we should point out that the respondents were generally healthier than non-respondents, however, the magnitude of the difference between these groups were not clinically important. Third, we conducted this study among Tehranian adolescents, so we cannot generalize our findings to other parts of the country, especially the rural zones.
## Conclusion
The present study was the first to evaluate the predictive power of general and central childhood adiposity indices and the risk of high cIMT in early adulthood not only in the Eastern Mediterranean Region but also worldwide. *Childhood* general and central anthropometric indices significantly predicted the high cIMT as the surrogate marker of CVD in early adulthood only among male adolescents without any differences in predictive power between the anthropometric indices. Following further adjustment for adulthood relevant anthropometric indices, generally, all of the male anthropometric indices except WC and WHR were no longer significant predictors of high cIMT in early adulthood. No superiority in the discriminatory ability for high cIMT among anthropometric indices was found in adolescent males. The predictive ability of high cIMT in early adulthood were prominent among pre-pubertal boys.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by the Ethics Committee of the Research Institute for Endocrine Sciences (RIES) of Shahid Beheshti University of Medical Sciences. Written informed consent to participate in this study was provided by the participants’ legal guardian/next of kin.
## Author contributions
FHa contributed to the conception and design of the study. PD contributed to the acquisition of the ultrasound data. MM analyzed the data. GA and AN took lead in the literature review and writing the manuscript in consultation with FHa. All authors reviewed and approved the final draft of the manuscript.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnut.2023.1098010/full#supplementary-material
## References
1. 1.World Health Organization [WHO].
Cardiovascular diseases: fact sheets.
Geneva: World Health Organization (2021).. (2021)
2. McGill H, McMahan C, Herderick E, Malcom G, Tracy R, Strong J. **Origin of atherosclerosis in childhood and adolescence.**. (2000) **72** 1307S-15S. DOI: 10.1093/ajcn/72.5.1307s
3. Alsheikh-Ali A, Kitsios G, Balk E, Lau J, Ip S. **The vulnerable atherosclerotic plaque: scope of the literature.**. (2010) **153** 387-95. DOI: 10.7326/0003-4819-153-6-201009210-00272
4. Pool L, Aguayo L, Brzezinski M, Perak A, Davis M, Greenland P. **Childhood risk factors and adulthood cardiovascular disease: a systematic review.**. (2021) **232** 118-26.e23. DOI: 10.1016/j.jpeds.2021.01.053
5. Koskinen J, Kytö V, Juonala M, Viikari J, Nevalainen J, Kähönen M. **Childhood risk factors and carotid atherosclerotic plaque in adulthood: the cardiovascular risk in young finns study.**. (2020) **293** 18-25. DOI: 10.1016/j.atherosclerosis.2019.11.029
6. Tasdighi E, Barzin M, Mahdavi M, Valizadeh M, Dehghan P, Moghaddam A. **Association of childhood obesity phenotypes with early adulthood carotid intima-media thickness (CIMT): tehran lipid and glucose study.**. (2022) **32** 249-57. DOI: 10.1016/j.numecd.2021.09.020
7. Valizadeh M, Tasdighi E, Barzin M, Hariri R, Mahdavi M, Dehghan P. **Association of childhood metabolic syndrome and metabolic phenotypes with the carotid intima-media thickness (CIMT) in early adulthood: tehran lipid and glucose study.**. (2022) **348** 128-33. DOI: 10.1016/j.ijcard.2021.12.010
8. Oren A, Vos L, Uiterwaal C, Grobbee D, Bots M. **Cardiovascular risk factors and increased carotid intima-media thickness in healthy young adults: the atherosclerosis risk in young adults (ARYA) study.**. (2003) **163** 1787-92. DOI: 10.1001/archinte.163.15.1787
9. van den Oord S, Sijbrands E, ten Kate G, van Klaveren D, van Domburg R, van der Steen A. **Carotid intima-media thickness for cardiovascular risk assessment: systematic review and meta-analysis.**. (2013) **228** 1-11. DOI: 10.1016/j.atherosclerosis.2013.01.025
10. Lorenz M, von Kegler S, Steinmetz H, Markus H, Sitzer M. **Carotid intima-media thickening indicates a higher vascular risk across a wide age range: prospective data from the carotid atherosclerosis progression study (CAPS).**. (2006) **37** 87-92. DOI: 10.1161/01.STR.0000196964.24024.ea
11. 11.World Obesity Atlas 2022.
One billion people globally estimated to be living with obesity by 2030.
London: World Obesity Federation (2022).. (2022)
12. Hosseinpanah F, Seyedhoseinpour A, Barzin M, Mahdavi M, Tasdighi E, Dehghan P. **Comparison analysis of childhood body mass index cut-offs in predicting adulthood carotid intima media thickness: tehran lipid and glucose study.**. (2021) **21**. DOI: 10.1186/s12887-021-02963-y
13. Oren A, Vos L, Uiterwaal C, Gorissen W, Grobbee D, Bots M. **Change in body mass index from adolescence to young adulthood and increased carotid intima-media thickness at 28 years of age: the atherosclerosis risk in young adults study.**. (2003) **27** 1383-90. DOI: 10.1038/sj.ijo.0802404
14. Menke A, Muntner P, Wildman R, Reynolds K, He J. **Measures of adiposity and cardiovascular disease risk factors.**. (2007) **15** 785-95. DOI: 10.1038/oby.2007.593
15. Azizi F, Ghanbarian A, Momenan A, Hadaegh F, Mirmiran P, Hedayati M. **Prevention of non-communicable disease in a population in nutrition transition: tehran lipid and glucose study phase II.**. (2009) **10**. DOI: 10.1186/1745-6215-10-5
16. Friedewald W, Levy R, Fredrickson D. **Estimation of the concentration of low-density lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge.**. (1972) **18** 499-502. DOI: 10.1093/clinchem/18.6.499
17. Koo T, Li MYA. **Guideline of selecting and reporting intraclass correlation coefficients for reliability research.**. (2016) **15** 155-63. DOI: 10.1016/j.jcm.2016.02.012
18. Juonala M, Magnussen C, Berenson G, Venn A, Burns T, Sabin M. **Childhood adiposity, adult adiposity, and cardiovascular risk factors.**. (2011) **365** 1876-85. DOI: 10.1056/NEJMoa1010112
19. DeLong E, DeLong D, Clarke-Pearson D. **Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach.**. (1988) **44** 837-45. DOI: 10.2307/2531595
20. Youden W. **Index for rating diagnostic tests.**. (1950) **3** 32-5. DOI: 10.1002/1097-0142(1950)3:1<32::AID-CNCR2820030106>3.0.CO;2-3
21. Juonala M, Magnussen C, Venn A, Dwyer T, Burns T, Davis P. **Influence of age on associations between childhood risk factors and carotid intima-media thickness in adulthood: the cardiovascular risk in young finns study, the childhood determinants of adult health study, the bogalusa heart study, and the muscatine study for the international childhood cardiovascular cohort (i3C) consortium.**. (2010) **122** 2514-20. DOI: 10.1161/CIRCULATIONAHA.110.966465
22. Koskinen J, Juonala M, Dwyer T, Venn A, Thomson R, Bazzano L. **Impact of lipid measurements in youth in addition to conventional clinic-based risk factors on predicting preclinical atherosclerosis in adulthood: international childhood cardiovascular cohort consortium.**. (2018) **137** 1246-55. DOI: 10.1161/CIRCULATIONAHA.117.029726
23. Freedman D, Patel D, Srinivasan S, Chen W, Tang R, Bond M. **The contribution of childhood obesity to adult carotid intima-media thickness: the Bogalusa heart study.**. (2008) **32** 749-56. DOI: 10.1038/sj.ijo.0803798
24. Juonala M, Raitakari M, Viikari SA, Raitakari O. **Obesity in youth is not an independent predictor of carotid IMT in adulthood. The cardiovascular risk in young finns study.**. (2006) **185** 388-93. DOI: 10.1016/j.atherosclerosis.2005.06.016
25. Yan Y, Hou D, Liu J, Zhao X, Cheng H, Xi B. **Childhood body mass index and blood pressure in prediction of subclinical vascular damage in adulthood: Beijing blood pressure cohort.**. (2017) **35** 47-54. DOI: 10.1097/HJH.0000000000001118
26. Buscot M, Thomson R, Juonala M, Sabin M, Burgner D, Lehtimäki T. **Distinct child-to-adult body mass index trajectories are associated with different levels of adult cardiometabolic risk.**. (2018) **39** 2263-70. DOI: 10.1093/eurheartj/ehy161
27. Ceponiene I, Klumbiene J, Tamuleviciute-Prasciene E, Motiejunaite J, Sakyte E, Ceponis J. **Associations between risk factors in childhood (12-13 years) and adulthood (48-49 years) and subclinical atherosclerosis: the Kaunas cardiovascular risk cohort study.**. (2015) **15**. DOI: 10.1186/s12872-015-0087-0
28. Hadaegh F, Zabetian A, Sarbakhsh P, Khalili D, James W, Azizi F. **Appropriate cutoff values of anthropometric variables to predict cardiovascular outcomes: 7.6 years follow-up in an Iranian population.**. (2009) **33** 1437-45. DOI: 10.1038/ijo.2009.180
29. Ross R, Neeland I, Yamashita S, Shai I, Seidell J, Magni P. **Waist circumference as a vital sign in clinical practice: a consensus statement from the IAS and ICCR working group on visceral obesity.**. (2020) **16** 177-89. DOI: 10.1038/s41574-019-0310-7
30. Sahakyan K, Somers V, Rodriguez-Escudero J, Hodge D, Carter R, Sochor O. **Normal-weight central obesity: implications for total and cardiovascular mortality.**. (2015) **163** 827-35. DOI: 10.7326/M14-2525
31. Wang H, Sun J, Zhang Z, Yang L, Zhao M, Bovet P. **Waist circumference change and risk of high carotid intima-media thickness in a cohort of Chinese children.**. (2021) **39** 1901-7. DOI: 10.1097/HJH.0000000000002881
32. Raitakari O, Juonala M, Kähönen M, Taittonen L, Laitinen T, Mäki-Torkko N. **Cardiovascular risk factors in childhood and carotid artery intima-media thickness in adulthood: the cardiovascular risk in young finns study.**. (2003) **290** 2277-83. DOI: 10.1001/jama.290.17.2277
33. Yajnik C, Katre P, Joshi S, Kumaran K, Bhat D, Lubree H. **Higher glucose, insulin and insulin resistance (HOMA-IR) in childhood predict adverse cardiovascular risk in early adulthood: the Pune children’s study.**. (2015) **58** 1626-36. DOI: 10.1007/s00125-015-3602-z
34. Davis P, Dawson J, Riley W, Lauer R. **Carotid intimal-medial thickness is related to cardiovascular risk factors measured from childhood through middle age: the Muscatine study.**. (2001) **104** 2815-9. DOI: 10.1161/hc4601.099486
35. Li S, Chen W, Srinivasan S, Bond M, Tang R, Urbina E. **Childhood cardiovascular risk factors and carotid vascular changes in adulthood: the Bogalusa heart study.**. (2003) **290** 2271-6. DOI: 10.1001/jama.290.17.2271
36. Freedman D, Dietz W, Tang R, Mensah G, Bond M, Urbina E. **The relation of obesity throughout life to carotid intima-media thickness in adulthood: the Bogalusa heart study.**. (2004) **28** 159-66. DOI: 10.1038/sj.ijo.0802515
37. Ferreira I, Twisk J, van Mechelen W, Kemper H, Seidell J, Stehouwer C. **Current and adolescent body fatness and fat distribution: relationships with carotid intima-media thickness and large artery stiffness at the age of 36 years.**. (2004) **22** 145-55. DOI: 10.1097/00004872-200401000-00024
38. Li S, Chen W, Srinivasan S, Tang R, Bond M, Berenson G. **Race (black-white) and gender divergences in the relationship of childhood cardiovascular risk factors to carotid artery intima-media thickness in adulthood: the Bogalusa heart study.**. (2007) **194** 421-5. DOI: 10.1016/j.atherosclerosis.2006.08.026
39. Juonala M, Viikari J, Kähönen M, Taittonen L, Laitinen T, Hutri-Kähönen N. **Life-time risk factors and progression of carotid atherosclerosis in young adults: the cardiovascular risk in young finns study.**. (2010) **31** 1745-51. DOI: 10.1093/eurheartj/ehq141
40. Magnussen C, Koskinen J, Chen W, Thomson R, Schmidt M, Srinivasan S. **Pediatric metabolic syndrome predicts adulthood metabolic syndrome, subclinical atherosclerosis, and type 2 diabetes mellitus but is no better than body mass index alone: the Bogalusa heart study and the cardiovascular risk in young finns study.**. (2010) **122** 1604-11. DOI: 10.1161/CIRCULATIONAHA.110.940809
41. Huynh Q, Blizzard L, Sharman J, Magnussen C, Schmidt M, Dwyer T. **Relative contributions of adiposity in childhood and adulthood to vascular health of young adults.**. (2013) **228** 259-64. DOI: 10.1016/j.atherosclerosis.2013.02.022
42. Su T, Liao C, Chien K, Hsu S, Sung F. **An overweight or obese status in childhood predicts subclinical atherosclerosis and prehypertension/hypertension in young adults.**. (2014) **21** 1170-82. DOI: 10.5551/jat.25536
43. Koskinen J, Magnussen C, Sabin M, Kähönen M, Hutri-Kähönen N, Laitinen T. **Youth overweight and metabolic disturbances in predicting carotid intima-media thickness, type 2 diabetes, and metabolic syndrome in adulthood: the cardiovascular risk in young finns study.**. (2014) **37** 1870-7. DOI: 10.2337/dc14-0008
44. Johnson W, Kuh D, Tikhonoff V, Charakida M, Woodside J, Whincup P. **Body mass index and height from infancy to adulthood and carotid intima-media thickness at 60 to 64 years in the 1946 British birth cohort study.**. (2014) **34** 654-60. DOI: 10.1161/ATVBAHA.113.302572
45. Du Y, Zhang T, Sun D, Li C, Bazzano L, Qi L. **Effect of serum adiponectin levels on the association between childhood body mass index and adulthood carotid intima-media thickness.**. (2018) **121** 579-83. DOI: 10.1016/j.amjcard.2017.11.029
46. Hao G, Wang X, Treiber F, Harshfield G, Kapuku G, Su S. **Body mass index trajectories in childhood is predictive of cardiovascular risk: results from the 23-year longitudinal Georgia stress and heart study.**. (2018) **42** 923-5. DOI: 10.1038/ijo.2017.244
|
---
title: Age-related ceRNA networks in adult Drosophila ageing
authors:
- Deying Yang
- Feng Xiao
- Jiamei Li
- Siqi Wang
- Xiaolan Fan
- Qingyong Ni
- Yan Li
- Mingwang Zhang
- Taiming Yan
- Mingyao Yang
- Zhi He
journal: Frontiers in Genetics
year: 2023
pmcid: PMC10012872
doi: 10.3389/fgene.2023.1096902
license: CC BY 4.0
---
# Age-related ceRNA networks in adult Drosophila ageing
## Abstract
As *Drosophila is* an extensively used genetic model system, understanding of its regulatory networks has great significance in revealing the genetic mechanisms of ageing and human diseases. Competing endogenous RNA (ceRNA)-mediated regulation is an important mechanism by which circular RNAs (circRNAs) and long non-coding RNAs (lncRNAs) regulate ageing and age-related diseases. However, extensive analyses of the multiomics (circRNA/miRNA/mRNA and lncRNA/miRNA/mRNA) characteristics of adult Drosophila during ageing have not been reported. Here, differentially expressed circRNAs and microRNAs (miRNAs) between 7 and 42-day-old flies were screened and identified. Then, the differentially expressed mRNAs, circRNAs, miRNAs, and lncRNAs between the 7- and 42-day old flies were analysed to identify age-related circRNA/miRNA/mRNA and lncRNA/miRNA/mRNA networks in ageing Drosophila. Several key ceRNA networks were identified, such as the dme_circ_0009500/dme_miR-289-5p/CG31064, dme_circ_0009500/dme_miR-289-5p/frizzled, dme_circ_0009500/dme_miR-985-3p/Abl, and XLOC_027736/dme_miR-985-3p/Abl XLOC_189909/dme_miR-985-3p/Abl networks. Furthermore, real-time quantitative PCR (qPCR) was used to verify the expression level of those genes. Those results suggest that the discovery of these ceRNA networks in ageing adult Drosophila provide new information for research on human ageing and age-related diseases.
## 1 Introduction
Drosophila melanogaster is an extensively used genetic model system that has been used for more than 100 years to study various aspects of the life sciences. In particular, fruit flies have been widely utilized to study ageing (Gubina et al., 2019) and human diseases, such as cancer (Enomoto et al., 2018), neurodegenerative disease (Cha et al., 2019), obesity and diabetes (Musselman et al., 2019), sterile inflammation (Nainu et al., 2019), and regeneration (Fox et al., 2020). D. melanogaster has also been utilized to study complex behavioural and developmental biology topics, including exercise (Watanabe and Riddle, 2019), courtship (Liu et al., 2019), and foraging (Khodaei and Long, 2019). Recently, mounting evidence has suggested that *Drosophila is* an outstanding model for studying ageing and age-related diseases (Surguchov et al., 2019; Brenman-Suttner et al., 2020). Ageing is a physiologic/pathologic process featuring declines in normal physiological functions and progressive impairment of cellular functions (Stallone et al., 2019). The ageing phenomenon has been conserved during biological evolution; even yeast other single-celled eukaryotes experience ageing (He et al., 2019). Thus, in-depth study of the regulatory mechanism of ageing in Drosophila can inform the study of human ageing and disease.
Current studies suggest that non-coding RNAs (ncRNAs) are involved in organismal ageing (Kim and Lee, 2019; Tower, 2019). With the development of sequencing technology, dynamic changes in the transcriptome [including changes in mRNA, long non-coding RNA (lncRNA), microRNA (miRNA), and circular RNA (circRNA) (Yang et al., 2016; Perry et al., 2017; Barter et al., 2019; Kinser and Pincus, 2020); the proteome (Brown et al., 2018); and the metabolome (Song et al., 2017)] have been described in the context of Drosophila ageing. Most miRNAs, mRNAs, and the proteins in fruit flies are evolutionarily conserved up to humans and regulate similar signalling pathways across organisms, such as the NF-κB, AMPK, mTOR, P53, PGC1α, and FoxO pathways (Zha et al., 2019; Kinser and Pincus, 2020; Wang et al., 2020). Large studies have demonstrated that the circRNA/miRNA/mRNA and axis the lncRNA/miRNA/mRNA axis play vital roles in ageing and age-related disease (Ruan et al., 2020; Wang et al., 2020).
As *Drosophila is* a workhorse model organism, thoroughly studying the characteristics of adult Drosophila from a multiomics perspective is necessary. Understanding how these conserved protein genes regulate ageing through ceRNA mechanisms in *Drosophila is* important. However, extensive analyses of the multiomics (circRNA/miRNA/mRNA lncRNA/miRNA/mRNA) characteristics of adult Drosophila ageing have not been reported. In the present study, we investigated the circRNA/miRNA/mRNA the lncRNA/miRNA/mRNA axis in adult Drosophila at two age points (day 7 and day 42) and determined the regulatory network of key differentially expressed (DE) genes in Drosophila ageing. The results provide knowledge on the gene regulation network of adult Drosophila ageing and a solid foundation for understanding the mechanisms of human ageing and age-related diseases.
## 2.1 Overview of multiomics data
The DE genes and proteins in Drosophila between day 7 and day 42 were identified, and their networks were analysed (Table 1). A total of 537 DE mRNAs and 43 DE lncRNAs were obtained from a previous study in our laboratory (Supplementary Data Sheet S1). A total of 6,003 circRNAs and 226 miRNAs were identified at day 7 and day 42 (Supplementary Data Sheet S1). Ultimately, 29 DE circRNAs and 24 DE miRNAs were found (Supplementary Data Sheet S1). The merged sequences of novel circRNAs (Supplementary Data Presentation S1–S3) and lncRNAs (Supplementary Data Sheet S3) are shown in supplementary files.
**TABLE 1**
| Unnamed: 0 | mRNA | LncRNAs | CircRNAs | miRNAs |
| --- | --- | --- | --- | --- |
| Total | 537 | 43 | 29 | 24 |
| Upregulated | 194 | 15 | 21 | 11 |
| Downregulated | 343 | 28 | 8 | 13 |
## 2.2 DE circRNAs and miRNAs in Drosophila between day 7 day 42
The DE circRNAs and miRNAs between 7 and 42-day old flies were analysed. Between day 7 and day 42, 29 DE circRNAs in Drosophila were identified, including 21 upregulated and 8 downregulated circRNAs at day 42 (Table 2). The circRNAs were derived from different source genes. These source genes were found to be involved in multiple molecular functions (Table 2). Evidently, the biological processes of the short lifespan-related source genes arm and pan were involved in the Wnt signalling pathway and had similar molecular functions, such as binding, protein binding, and transcription factor binding, transcription regulator activity. In addition, different circRNAs were observed to originate from the same mRNA transcript. For example, and dme_circ_0008175 and dme_circ_0008173 originated from the *Nlg1* gene, and dme_circ_0009514 dme_circ_0009500 were derived from the pan gene.
**TABLE 2**
| ID | CircBase ID | Fold change | p-value | Source gene | Portion of biological process term(s) from flybase database |
| --- | --- | --- | --- | --- | --- |
| Dme_circ_0009372 | Dme_circ_0005033 | 4.5034↑ | 0.007635 | Asator | Protein serine/threonine kinase activity; ATP binding |
| Dme_circ_0006708 | Dme_circ_0005241 | 4.2791↑ | 0.012415 | Dad | TGF-beta signalling pathway; negative regulation of BMP signalling pathway |
| Dme_circ_0004259 | Dme_circ_0002098 | 4.2083↑ | 0.014401 | shot | Cytoplasmic microtubule organization; wound healing; branching involved in open tracheal system development; cilium organization |
| Dme_circ_0002070 | Dme_circ_0002195 | 3.9032↑ | 0.026301 | Scp1 | - |
| Dme_circ_0010408 | Dme_circ_0003710 | 3.8118↑ | 0.031385 | CoRest | Negative regulation of transcription by RNA polymerase II; positive regulation of DNA methylation-dependent heterochromatin assembly; negative regulation of histone H4-K16 acetylation; negative regulation of histone H3-K27 methylation |
| Dme_circ_0003904 | Dme_circ_0003738 | 3.5948↑ | 0.045349 | CG33144 | Ubiquitin-dependent protein catabolic process |
| Dme_circ_0008175 | Dme_circ_0001709 | 3.5948↑ | 0.045412 | Nlg1 | Neuromuscular junction development; cellular process |
| Dme_circ_0005030 | Dme_circ_0002884 | 3.5765↑ | 0.046714 | Ccn | Negative regulation of cell death; signal transduction; cell adhesion |
| Dme_circ_0006667 | Dme_circ_0001321 | 3.5765↑ | 0.046714 | gish | Positive regulation of Wnt-TCF hedgehog signalling pathways; negative regulation of Hippo signalling pathway |
| Dme_circ_0010536 | Dme_circ_0000629 | 3.559↑ | 0.048096 | slgA | Arginine proline metabolism |
| Dme_circ_0008383 | Dme_circ_0004519 | 2.6054↑ | 0.043863 | mura | Protein ubiquitination; long-term memory |
| Dme_circ_0010134 | Dme_circ_0002087 | 1.1067↑ | 0.039466 | Stim | Developmental process; cellular homeostasis |
| Dme_circ_0009358 | - | 0.74476↑ | 0.002481 | CaMKI | Protein phosphorylation |
| Dme_circ_0000626 | - | 3.7656↑ | 0.03384 | CG17646 | Triglyceride metabolic process; transmembrane transport |
| Dme_circ_0011075 | - | 3.7618↑ | 0.03415 | Trf2 | Post-embryonic development; response to organic cyclic compound; programmed cell death; respiratory system development; response to oxygen-containing compound |
| Dme_circ_0006956 | - | 3.5884↑ | 0.045804 | GluClalpha | Cellular process; transport; localization; establishment of localization; biological regulation |
| Dme_circ_0009667 | - | 3.559↑ | 0.048096 | dlg1 | Hippo signalling pathway-fly |
| Dme_circ_0009514 | - | 4.0161↑ | 0.021197 | pan | Canonical Wnt signalling pathway |
| Dme_circ_0006334 | - | 3.8925↑ | 0.026799 | CG42402 | - |
| Dme_circ_0008173 | - | 3.881↑ | 0.027405 | Nlg1 | Neuromuscular junction development; cellular process |
| Dme_circ_0004404 | - | 0.61768↑ | 0.020721 | Dbp80 | Poly(A)+ mRNA export from nucleus |
| Dme_circ_0003891 | Dme_circ_0001623 | −3.8065↓ | 0.033572 | psq | Anterior/posterior axis specification, embryo; DNA binding |
| Dme_circ_0006619 | Dme_circ_0003501 | −4.3746↓ | 0.010954 | srp | Autophagy; cell fate commitment; midgut development |
| Dme_circ_0004913 | Dme_circ_0004913 | −0.52833↓ | 0.047568 | CG34347 | Actomyosin structure organization |
| Dme_circ_0004843 | - | −0.48662↓ | 0.02228 | CG15715 | |
| Dme_circ_0010498 | - | −0.73454↓ | 0.04614 | CG1304 | Proteolysis |
| Dme_circ_0006913 | - | −2.7169↓ | 0.042802 | Cyp12a5 | Oxidation-reduction process |
| Dme_circ_0010310 | - | −3.8065↓ | 0.033572 | arm | Wnt signalling pathway |
| Dme_circ_0009500 | - | −4.0543↓ | 0.021183 | pan | Canonical Wnt signalling pathway |
Furthermore, 24 DE miRNAs (11 upregulated and 13 downregulated) were identified at day 42 compared to day 7 (Supplementary Data Sheet S1). Dme_miR-9a-3p and dme_miR-985-3p were identified as canonical specific fruit fly miRNAs. Then, dme_miR-956-3p, dme_miR-284-3p, and dme_miR-289-5p were identified as non-canonical miRNAs. The remaining 19 DE miRNAs were canonically conserved miRNAs. Then, functional annotation was carried out for the DE miRNAs. The Gene Ontology (GO) annotations of age-related DE miRNAs based on their targets were investigated. 11 upregulated and 13 downregulated miRNAs targeted to 3,292 mRNAs and 2,951 mRNAs, respectively (Supplementary Data Sheet S1), which mainly enriched in biological process (1,314 terms and 1,233 terms, respectively) (Supplementary Data Sheet S1). The bar plot shows the top ten enrichment score value of the significant enrichment terms (Figures 1A, B), such as multicellular organism development, nervous system development, neurogenesis, development process, and cell differentiation. Specifically, 20 DE miRNAs were related to Drosophila aging based on previous reports, including 11 upregulated and 9 downregulated miRNAs in 42 days when compared to 7 days (Figure 1C). 13 DE miRNAs involved in the age-signalling pathways by target genes on post-transcriptional level (Figure 1C). Their tagets involved in regulation of ROS detoxification, autopage, circadian rhythm, apoptosis, and immunity biological process. Otherwise, three out of 24 total miRNAs were conserved in Drosophila, human and mouse, containing dme-miR-8-5p, dme-miR-133-3p, and dme-miR-10-5p.
**FIGURE 1:** *GO annotations age-related pathways of DE miRNAs in Drosophila between day 7 day 42. Twenty-four DE miRNAs were identified between 7 and 42-day old flies. (A), GO annotations of 11 upregulated miRNAs. (B), GO annotations of 13 downregulated miRNAs. (C), 20 DE miRNA affected the Drosophila agineg; rust colored frame, upregulated in 42 days when compared to 7 days (p < 0.05); green colored frame, downregulated in 42 days when compared to 7 days (p < 0.05). The references were followed as miR-289 (Chen et al., 2014; Nesler et al., 2016), miR-5 (Leaman et al., 2005; Chen et al., 2014), miR-8 (Soler Beatty et al., 2021; Chen et al., 2022), miR-12 (Yang et al., 2009), miR-125 (Bushey et al., 2009; Gendron and Pletcher, 2017; Luhur et al., 2017), miR-9a (Chen et al., 2014; Suh et al., 2015), miR-275 (Chen et al., 2014; Ji et al., 2019), miR-310 (Robins et al., 2005; Chen et al., 2014), miR-6 (Leaman et al., 2005; Chen et al., 2014), miR-14 (Xu et al., 2003; Varghese and Cohen, 2007), miR-276a (Chen et al., 2014; Zhang et al., 2021a), miR-276b (Chen et al., 2014; Ulgherait et al., 2020; Zhang et al., 2021b), miR-996 (Sun et al., 2015; Duan et al., 2018). BP, biological process; CC, celluar component; MF, molecular function.*
## 2.3 DE circRNA/miRNA/mRNA networks in Drosophila ageing
In this section, the DE circRNA/miRNA/mRNA networks are analysed. Through the ceRNA mechanism, miRNAs can negatively regulate mRNA expression. Overall, 12 DE circRNAs, 21 DE miRNAs, and 30 DE mRNAs had interactions (Figure 2A, Supplementary Data Sheet S4). The binding sites between circRNA vs. miRNA (Supplementary Data Sheet S5) and miRNA vs. mRNA (Supplementary Data Sheet S6) are shown. According to the trends in the expression quantity changes, five DE circRNAs (dme_circ_0006913, dme_circ_0008173, dme_circ_0009500, dme_circ_0009667, and dme_circ_0010536) targeted the DE miRNAs with opposite expression trends (Figure 2B). Based on qPCR results, the expression patterns of four circRNAs (dme_circ_0008173, dme_circ_0009500, dme_circ_0009667, and dme_circ_0010536), and four miRNAs (dme_miR-289-5p, dme_miR-985-3p, dme_miR-286-3p, dme_miR-14-5p), four mRNAs (frizzled, CG31064, Abl and SERCA) were consistent with the RNA-seq data (Figure 2C). Importantly, the expression trends of dme_circ_0009500/dme_miR-289-5p/CG31064, dme_circ_0009500/dme_miR-289-5p/frizzled, and dme_circ_0009500/dme_miR-985-3p/Abl conformed to the ceRNA mechanism.
**FIGURE 2:** *DE circRNA/miRNA/mRNA networks of Drosophila at day 7 day 42. (A), circRNA/miRNA/mRNA interaction networks; yellow, circRNAs; red, mRNAs; green, miRNAs. (B), Specific circRNA/miRNA/mRNA networks; “↑” “↓” indicate upregulation downregulation on day 42 compared to day 7 in RNA-seq data, respectively. (C), Data on the relative expression of circRNA, miRNA, mRNA detected by qPCR analysis. The relative gene expression levels were calculated by the cycle threshold values that were identified as 2−ΔΔCT. The ribosomal protein L32 (rp 49) gene was used as the reference gene to calculate the relative mRNA, miRNA, circRNA levels. “*” above the bars indicates a significant difference at the 0.05 level, “**” indicates a significant difference at the 0.01 level. (+) represents the qPCR results were consistent with the RNA-seq data.*
## 2.4 DE lncRNA/miRNA/mRNA networks in Drosophila ageing
According to the functional patterns of lncRNAs competing with mRNAs for binding to miRNAs, the interaction networks of DE lncRNAs/miRNAs/mRNAs were identified (Supplementary Data Sheet S7). Then, the binding sites between lncRNAs and miRNAs were included in Supplementary Data Sheet S8. In addition, DE miRNAs targeted DE mRNAs with opposite expression trends. Based on the DE genes in our database, 15 lncRNAs, 15 miRNAs, and 32 mRNAs had interactions (Figure 3A). Based on qPCR results, the expression patterns of two lncRNAs (XLOC_027736 and XLOC_189909), three miRNAs (dme_miR-289-5p, dme_miR-985-3p, and dme_miR-14-5p), and three mRNAs (frizzled, CG31064, and Abl) were consistent with the RNA-seq data (Figure 3A). Several specific lncRNA/miRNA/mRNA networks, including XLOC_027736/dme_miR-985-3p/Abl, XLOC_073604/dme_miR-994-3p-3p/mbl, XLOC_189909/dme_miR-985-3p/Abl (Figure 3B), were found. In the XLOC_027736/dme_miR-985-3p/Abl and XLOC_189909/dme_miR-985-3p/Abl networks, the expression trend of these genes was determined by qPCR analysis to be consistent with the RNA-seq data from our study (Figure 3C).
**FIGURE 3:** *DE lncRNA/miRNA/mRNA networks in Drosophila at day 7 day 42. (A), lncRNA/miRNA/mRNA interaction networks; yellow, lncRNAs; red, mRNAs; green, miRNAs. (B), Specific lncRNA/miRNA/mRNA networks; “↑” “↓” indicate upregulation downregulation on day 42 compared to day 7, respectively. (C), Data on the relative expression of lncRNAs, miRNAs, mRNAs by qPCR analysis. 7 days, 7 days; 42 days, 42 days. The relative gene expression levels were calculated using the cycle threshold values that were identified as 2−ΔΔCT. The ribosomal protein L32 (rp 49) gene was used as the reference gene to calculate the relative mRNA, miRNA, lncRNA levels. “*” above the bars indicates a significant difference at the 0.05 level, “**” indicates a significant difference at the 0.01 level. (+) represents the qPCR results were consistent with the RNA-seq data.*
## 2.5 Functional annotation of DE circRNA/lncRNA-associated networks
GO functional annotation of DE circRNAs/lncRNAs/mRNAs was carried out based on 74 target mRNA genes (Supplementary Data Sheet S9). The first 30 GO terms based on the lowest p values are listed (Figure 4). In circRNA/mRNA GO terms, there were just two major GO categories in the circRNA-associated networks, including biological processes (29 GO terms) and cellular components (1 GO term, perinuclear region of cytoplasm). Similarly, the GO annotations of the lncRNA-associated networks consisted of 25 GO terms in biological processes and 5 GO terms in the cellular component.
**FIGURE 4:** *GO annotations of the DE circRNA-/lncRNA-associated networks. (A), GO annotations of the DE circRNA/miRNA/mRNA networks. (B), GO annotations of the DE lncRNA/miRNA/mRNA networks. LogP values indicate the enrichment degree of targets in the corresponding GO term, a smaller value represents a higher enrichment degree. The number of genes represents the number of target genes enriched in GO terms.*
In the GO annotations of the circRNA-associated networks, there were several ageing-related biological processes in GO terms, such as homeostatic process (10 DE genes), cellular homeostasis (7 DE genes), cation homeostasis (5 DE genes), ion homeostasis (5 DE genes), cell-cell signalling (9 DE genes), cell fate determination (5 DE genes), developmental growth (7 DE genes) (Figure 5A). Furthermore, 13 out of the first 30 GO terms of the lncRNA-associated networks were related to homeostatic processs, including cellular homeostasis, retinal homeostasis, ion homeostasis, calcium ion homeostasis, and cellular cation homeostasis (Figure 5B). In the ageing-related ceRNA networks dme_circ_0009500/dme_miR-985-3p/Abl, XLOC_027736/dme_miR-985-3p/Abl and XLOC_189909/dme_miR-985-3p/Abl in Drosophila, Abl was involved in multiple GO terms, such as regulation of cell morphogenesis, developmental growth, regulation of cell differentiation, and regulation of neuron differentiation. Furthermore, frizzled was involved in the positive regulation of developmental growth, developmental growth involved in morphogenesis, cell-cell signalling, cell fate determination, and the Wnt signalling pathway in the dme_circ_0009500/dme_miR-289-5p/frizzled network.
**FIGURE 5:** *Tissue-specific expression patterns of the specific DE lncRNAs, circRNAs, miRNAs, mRNAs in the head, ovary, gut, fat body of flies. The expression levels of the DE genes in the circRNA/miRNA/mRNA lncRNA/miRNA/mRNA networks, as determined through qPCR, were consistent with the RNA-seq data. (A–D), The relative expression level of four DE circRNAs; (E–H), the relative expression level of four DE miRNAs; (I–L), the relative expression level of four DE mRNAs; (M,N), the relative expression level of two DE lncRNAs. The relative gene expression levels were calculated by the cycle threshold values that were identified as 2−ΔΔCT. The ribosomal protein L32 (rp 49) gene was used as the reference gene to calculate the relative mRNA, miRNA, lncRNA levels. The different letters above the bars indicate significant differences at the 0.05 level. (+) represents the qPCR results were consistent with the RNA-seq data.*
## 2.6 Tissue expression pattern analysis
The tissue-specific expression patterns of the DE lncRNAs, circRNAs, miRNAs, and mRNAs were analysed in the head, ovary, gut, and fat body, the qPCR results of which were consistent with the RNA-seq data in the DE circRNA/miRNA/mRNA networks and lncRNA/miRNA/mRNA networks (Figure 5). The results showed that the tissue-specific expression patterns of dme_circ_0009500/dme_miR-289-5p/CG31064, dme_circ_0009500/dme_miR-289-5p/frizzled, and dme_circ_0009500/dme_miR-985-3p/Abl were mainly expressed in the head. Specifically, dme_circ_0009667 was mainly located in the ovary, and dme_miR-14-5p was mainly expressed in the gut. Similar results were found in the lncRNA/miRNA/mRNA networks. The tissue patterns of the XLOC_027736/dme_miR-985-3p/Abl and XLOC_189909/dme_miR-985-3p/Abl networks were also mainly expressed in the head.
## 2.7 Binding sites of specific ceRNAs
The binding sites of dme_circ_0009500/dme_miR-289-5p/CG31064, dme_circ_0009500/dme_miR-289-5p/frizzled, dme_circ_0009500/dme_miR-985-3p/Abl, XLOC_027736/dme_miR-985-3p/Abl, and XLOC_189909/dme_miR-985-3p/Abl were analyzed (Figure 6). Specificly, there were five binding sites between dme_circ_0009500 and dme_miR-289-5p with the higher binding free energy from -15.23 to -20.30 (Figure 6A; and Supplementary Data Sheet S5). Furthermore, 3′UTR sequence of frizzled had the two binding site with miR-289-5p, and also had the higher binding free energy -23.31 (Figure 6E and Supplementary Data Sheet S6).
**FIGURE 6:** *The binding sites of specific ceRNAs. (A), the binding sites between circ_0009500/Dme_miR-289-5p; (B), the binding sites between XLOC_027736 dme_miR-985-3p; (C), the binding sites between XLOC_189909 dme_miR-985-3p; (D), the binding sites between dme_miR-289-5p CG31064; (E), the binding sites between dme_miR-289-5p frizzled; (F), the binding sites between dme_miR-985-3p Abl.*
## 3 Discussion
Drosophila is an ideal model for genetics, and the multiple age-related researches were carried out by Dahomey strain (Mołoń et al., 2020; De Groef et al., 2021). Previous study has reported the trend of wild-type female lifespan (Dahomey, Canton S, Oregon R) changes were similar (Sanz et al., 2010). Furthermore, several studies just used one wildtype strain in multiple-omics research (Shi et al., 2020; Wang et al., 2022). Thus, the wild-type female of Dahomey strain was utilized in our manuscript. Increasing evidence has suggested that ceRNA networks play key roles in a variety of biological processes, such as cancer (Wang et al., 2018; Abdollahzadeh et al., 2019), Alzheimer’s disease (AD) (Zhang et al., 2019), skeletal muscle myogenesis (Yue et al., 2019), and ageing (Zhao et al., 2019; Chen et al., 2020). miRNAs are the core molecules of the ceRNA regulatory system (Yue et al., 2019; Chen et al., 2020). Thus, screening DE miRNAs is essential for research on ceRNA networks related to Drosophila ageing. miRNAs generally induce mRNA degradation or repress translation of target transcripts through sequence-specific binding to the transcript 3′UTR (Bushati and Cohen, 2007; Chen et al., 2019). Each transcript can be targeted by multiple miRNAs, and each miRNA can target hundreds of different transcripts (mRNA, circRNA, and lncRNA transcripts) (Dori and Bicciato, 2019; Zhou et al., 2019; Kinser and Pincus, 2020). Thus, the miRNA regulatory network is far-reaching (Kinser and Pincus, 2020). Previous studies have verified that miRNAs are important small regulatory ncRNA molecules that control a fairly large number of biological processes; their important functions have generated interest in their use as biomarkers and their roles as regulators of ageing (Lai et al., 2019; Kinser and Pincus, 2020) and a number of cancer types (Liang et al., 2020; Mishan et al., 2020; Wang et al., 2020). Recent studies have also suggested that miRNAs are involved in the regulation of age-associated processes and pathologies in multiple mammalian tissues, including the brain, heart, bones, and muscles (Chen et al., 2020; John et al., 2020; Kinser and Pincus, 2020; Ullah et al., 2020).
In our study, more than $80\%$ (20 out of 24 total miRNAs) of DE miRNAs between day 7 and day 42 could affect fruit fly ageing. Among these miRNAs, miR-14 as a cell death suppressor, regulates fat metabolism, insulin production and metabolism through its targets (Xu et al., 2003; Nelson et al., 2014). In addition, knockout of five DE miRNAs (miR-133, miR-284, miR-286, miR-318, miR-956, and miR-988) decreased the Drosophila lifespan, and miR-286 KO increased female lifespan (Chen et al., 2014). The consreved miR-8 was the homolog of vertebrate miR-200 family. It is worth noting that miR-8 acted through U-shaped to activate PI3K and thereby promoted fat cell growth cell-autonomously and the Insulin-like Receptor signaling pathway (Soler Beatty et al., 2021; Chen et al., 2022). Then, two targets of miR-200 (dme-miR-8 homolog) in human were JAGGED1 (JAG1) (Xue et al., 2021) and epidermal growth factor receptor (EGFR) (Xue et al., 2019). In a lung cancer metastasis, overexpression of the miR-200 decreased JAG1 protein levels and impeded cell growth (Xue et al., 2021). Furthermore, Dmel\ EGFR is orthologous to human gene ERBB2, which has also been implicated in multiple cancers (Xue et al., 2019). The results indicate that these miRNAs may act as sponges through the circRNA/miRNA/mRNA and lncRNA/miRNA/mRNA networks involved in the regulation of Drosophila ageing. In a previous study, global identification of functional miRNA-mRNA interactions in Drosophila was performed (Wessels et al., 2019). However, circRNA/miRNA/mRNA and lncRNA/miRNA/mRNA interactions in fly ageing have remained unclear.
Most of the DE circRNAs accumulated in 42-day-old Drosophila, which has been reported in a previous study (Westholm et al., 2014) in flies and other model organisms (Knupp and Miura, 2018; Kim and Lee, 2019). circRNAs are highly stable molecules that play important roles in ageing (Weigelt et al., 2020) and age-related diseases (Ren et al., 2020; Wang P. et al., 2020). Additionally, six source genes of circRNAs were related to fruit fly ageing, including asator (short-lived, Asator GD7323) (Neely et al., 2010), dad (short-lived, Dad GD1335) (Kamiya et al., 2008), gish (short-lived, gish GD10588) (Neely et al., 2010; Fulford et al., 2019), pan (short-lived, panΔN.UAS) (Buchon et al., 2013; Franz et al., 2017), psq (long-lived, psq BG01031) (Magwire et al., 2010; Bonchuk et al., 2011), and arm (short-lived, arm S10. UAS. Tag:MYC) (Dupont et al., 2012). In the present study, multiple networks of DE circRNAs/miRNAs/mRNAs and DE lncRNAs/miRNAs/mRNAs in Drosophila between day 7 and day 42 were identified. GO annotations of the DE circRNA/miRNA/mRNA and DE lncRNA/miRNA/mRNA networks demonstrated that the DE circRNAs and lncRNAs may take part in Drosophila ageing via various biological processes. Multiple GO terms enriched in the DE circRNA/miRNA/mRNA networks, such as cellular homeostasis (Hartl, 2016) and developmental growth (Clarke et al., 2020), are clearly related to ageing. Specifically, multiple GO terms, including inorganic ion homeostasis, cation homeostasis, and cellular calcium ion homeostasis, are involved in regulating the sequestration of calcium ions. Ca2+ dyshomeostasis is associated with several ageing-related neurodegenerative diseases, such as AD, Huntington’s disease (HD), Parkinson’s disease (PD), and amyotrophic lateral sclerosis (ALS), with an altered Ca2+ buffering capacity, an altered regulation of Ca2+ channels and pumps, and an altered neuronal excitability (Bezprozvanny, 2010). A previous study demonstrated that ageing was closely linked to the dysregulation of Ca2+ homeostasis, resulting in a chronically elevated level of cytosolic Ca2+ in experimental models of neuronal ageing (Thibault et al., 2007; Duncan et al., 2010). The results suggest that the DE circRNA/miRNA/mRNA and DE lncRNA/miRNA/mRNA networks may play a significant role in Drosophila ageing.
Based on the same expression trend between the RNA-seq and qPCR detection results in our study, some specific ceRNA networks, including the dme_circ_0009500/dme_miR-289-5p/CG31064, dme_circ_0009500/dme_miR-289-5p/frizzled, dme_circ_0009500/dme_miR-985-3p/Abl, XLOC_027736/dme_miR-985-3p/Abl and XLOC_189909/dme_miR-985-3p/Abl networks, were identified. Thus, we determined that these networks merited further investigation. In our study, miR-289-5p was mainly expressed in the Drosophila head and upregulated in 42-day-old flies compared to 7-day-old flies. In previous studies, miR-289-5p, which may be responsible for silencing the expression of candidate genes during the diapause of Sarcophaga bullata, was overexpressed in diapausing pupae (Reynolds et al., 2017). Then, miR-289 participates in the control of a diverse array of pleiotropic cellular processes during Drosophila development (Nesler et al., 2013). It has been reported that miR-289 is downregulated in adult-onset AD Drosophila brains (Kong et al., 2014). Neuronal misexpression of miR-289 suppresses activity-dependent synaptic growth (Nesler et al., 2013). The target CG31064 of miR-289-5p was located in the adult head in our study in previous research (Aradska et al., 2015), which enabled a small GTPase binding activity (Gillingham et al., 2014). Another target, frizzled of miR-289-5p, has been reported to be involved in the regulation of the mTOR and signalling pathway the Wnt signalling pathway (Zeng et al., 2018). The GO frizzled terms were related to positive regulation of developmental growth, developmental growth involved in morphogenesis, cell-cell signalling, cell fate determination, and the Wnt signalling pathway. Previous study has reported that frizzled in involved in regulation of pro-survival processes in human PD through Wnt1/Fzd-1/β-catenin astrocyte-dopamine autoprotective loop (L'Episcopo et al., 2011). Therefore, dme_circ_0009500/dme_miR-289-5p/CG31064 and dme_circ_0009500/dme_miR-289-5p/frizzled may play roles in brain ageing in fruit flies.
Furthermore, the lncRNAs XLOC_027736 and XLOC_189909, the dme_circ_0009500 and the mRNA Abl were predicted to competitively bind miR-985-3p, which was mainly expressed in the Drosophila head. Abl phosphorylates cell adhesion and cytoskeletal proteins and acts as a scaffold in a signalling complex to regulate both epithelial and nervous system morphogenesis (Zhu and Bhat, 2011; Liu and Wu, 2014). The age-related phenotype associated with the Abl mutant led to a shorter lifespan for flies with three specific alleles (Abl l2, Abl l3, and Abl GD1344) (Belote et al., 1990; Huang et al., 2007; Neely et al., 2010). Furthermore, human ABL1 (Drosophila Abl homolog) protein kinases play many important roles in neuron development, maintenance, and signalling (Manley et al., 2022). In future studies, it will be necessary to determine the biological role of the lncRNAs XLOC_027736 XLOC_189909, the circRNA dme_circ_0009500, and miR-985-3p in the fly ageing process.
In our study, DESeq2 and EdgeR were used to analyse the DE circRNAs and miRNAs, respectively. DESeq2 and EdgeR are efficient tools for differential analysis of RNA-seq data with the more than $80\%$ overlapping, both of which use the negative binomial distribution (Liu et al., 2021). Then, DESeq2 can more accurately identify DE genes for small samples to reduce false positives (Love et al., 2014). Then, two biological replicates were utilized to analysis DE circRNAs. Thus, DESeq2 was chosen to identify the DE circRNAs. The above results suggest that our data are also available. To date, DE circRNAs and miRNAs cannot be reanalysed only by DESeq2 or EdgeR in the present study. This would bring some questions to our subsequent analysis. For example, some DE circRNAs and miRNAs may be lost, which may result that part ceRNA networks could not be recognized. More experiments will be needed to verify these networks in our future study.
In this study, several DE miRNAs were identified between day 7 and day 42, such as the dme_miR-289, dme_miRNA-14, and the conserved miR-8. Then, the potential ceRNA networks may play a role in Drosophila aging, for example, dme_circ_0009500/dme_miR-289-5p/frizzled and XLOC_027736/dme_miR-985-3p/Abl networks. Therefore, the result of DE miRNAs, circRNA/miRNA/mRNA, and lncRNA/miRNA/mRNA interactions provides an important foundation to parse the genetic process of Drosophila ageing.
## 4.1 Sample collection and preparation
Female flies (DahomeyWT) that had mated with male flies for 48 h after hatching were bred under a 12 h on/off light cycle at 25°C in $50\%$ humidity. Sample collection of adult flies at day 7 and day 42 was performed as described by Yang et al. [ 2016]. RNA-seq (mRNA, lncRNA, and circRNA) was conducted on two biological replicates, while miRNA sequencing was conducted on five biological replicates. All samples were stored at −80°C until use.
## 4.2 LncRNA, mRNA, circRNA data from Drosophila at days 7 and 42
RNA-seq data for the lncRNAs and mRNAs of Drosophila at day 7 and day 42 were obtained by Yang et al. [ 2016]. The circRNA analysis was based on the RNA-seq data for Drosophila at day 7 and day 42 from Yang et al. [ 2016]. Overall, $95.79\%$ clean reads were obtained from 26.7 GB of raw sequence data (SRP073695) then aligned to the D. melanogaster genome from FlyBase (Dmel_Release_6, http://FlyBase.org/). The find_circ (Memczak et al., 2013) and CIRI2 (Gao et al., 2015) software tools were utilized to identify circRNAs. Then, the overlapping Drosophila circRNAs identified by both software programs were selected. The input data for the circRNA differential expression analysis were readCount data obtained from the circRNA expression level analysis. Then, paired differential expression analysis of circRNAs between day 7 and day 42 was conducted with DESeq2 (Varet et al., 2016) based on a negative binomial distribution. The p-value was adjusted using Hochberg and Benjamini’s methods (Hinkley et al., 2018) to control the error discovery rate. A p-value <0.05 was considered to indicate a DE circRNA. These original data were from our laboratory.
## 4.3 MiRNAs in Drosophila at day 7 and day 42
TRIzol Reagent (Invitrogen, CA) was used to extract the total RNA from fruit flies at day 7 and day 42. Agarose gel electrophoresis was performed to verify the integrity of the total RNA samples. A NanoDrop ND-1000 instrument was used to accurately measure the concentrations and protein contamination of the total RNA samples. miRNA sequencing libraries were generated using rRNA-depleted RNA with a NEBNext® Ultra™ Multiplex Small RNA Library Prep Set Kit for Illumina® (New EnglBiolabs, United States) following the manufacturer’s recommendations. Subsequently, an Agilent 2,100 Bioanalyser and an Agilent DNA 1000 chip kit (Agilent, part #5067-1504) were utilized to accurately assess the quality and concentration of the sequencing libraries. The libraries were sequenced using an Illumina NextSeq 500. MiRNA fragment sequencing was performed by the Aksomics company.
Clean reads were generated from the raw sequence data from the Illumina NextSeq instrument through real-time base calling and quality filtering. The clean reads were recorded in FASTQ format and contained read information, sequences and quality encoding. Subsequently, the 5′- and 3′-adapter sequences were trimmed from the clean reads by Cutadapt, and reads with lengths shorter than 14 nt or longer than 40 nt were discarded. The trimmed reads were collapsed into FASTA format. The raw data has been uploaded to NCBI database (PRJNA716466). The trimmed reads that did not map to mature or precursor tRNA sequences were aligned with an allowance of only one mismatch to miRNA reference sequences with miRDeep2 (Friedlander et al., 2012). The expression profiles of miRNAs were determined based on the counts of the reads mapped. The DE miRNAs were identified with the R package EdgeR based on the count values (Robinson et al., 2010). A fold change cut-off of 1.5 and a p-value cut-off of 0.05 were applied only when replicates were used for screening DE miRNAs. *The* gene prediction of DE miRNA integrates two algorithms, miRanda (Enright et al., 2003) and TargetScan (Garcia et al., 2011). GO enrichment analysis of the targets of DE miRNAs was implemented with the GOseq R package (Young et al., 2010).
## 4.4 CeRNA analysis of lncRNA/circRNA-miRNA-mRNA
The lncRNAs, circRNAs, and miRNAs showed significantly different expression levels between day 7 and day 42 and were thus analysed. The potential ceRNAs were searched based on the sequences of the lncRNAs, circRNAs, and mRNAs. The offline software MiRanda (Enright et al., 2003) was utilized to predict miRNA binding seed sequence sites, and overlap of the same miRNA binding sites on lncRNAs/circRNAs-miRNAs and miRNAs-mRNAs was taken to indicate a lncRNA/circRNA-miRNA-mRNA interaction. Then, the “clusterProfiler” R package was utilized to perform Gene Ontology (GO) enrichment of ceRNA networks based on the mRNAs.
The DE mRNAs associated with ageing and the corresponding ceRNA networks (including the circRNAs, lncRNAs, miRNAs, and mRNAs) were selected to detect the expression level by qPCR. In total, 9 miRNAs, 10 mRNAs, 5 lncRNAs, and 5 circRNAs were selected to qualify the expression level. Samples (three biological duplicates) from 7- and 42-day-old fruit flies were used to isolate the total RNA using TRIzol™ LS reagent (Thermo Fisher) according to the manufacturer’s instructions. The total RNA (1 μg) from each sample was reverse transcribed with random primers using a RevertAid First-strand cDNA Synthesis Kit (Thermo Fisher) according to the manufacturer’s protocol, which was utilized to detect the expression levels of mRNA, lncRNA, and circRNA. In addition, the total RNA (1 μg) used for miRNA expression detection in each sample was reverse transcribed using a TaqMan™ MicroRNA Reverse Transcription Kit (Thermo Fisher) according to the kit instructions. *All* gene primers (Supplementary Data Sheet S10) were designed using the Primer 5.0 software and purchased from Sangon Biotech. The SYBR green method was used for qRT-PCR with TransStart® Green qPCR SuperMix (TransGen Biotech) following the manufacturer’s instructions. All tested samples were three biological replicates. The relative gene expression levels were calculated using cycle threshold values and the 2-ΔΔCT method. Ribosomal protein L32 (rp 49) was used as the reference gene to calculate the relative mRNA, miRNA, lncRNA, and circRNA levels. Differential expression levels were compared by independent-samples t-tests between groups.
## 4.5 Tissue-specific expression pattern
Based on the same expression pattern between day 7- and day 42-day-old flies, as determined by the RNA-seq data and qPCR tests on 7-day and 42-day-old flies, the tissue-specific expression model of four circRNAs (dme_circ_0008173, dme_circ_000950, dme_circ_0009667, and dme_circ_0010536), four miRNAs (dme_miR-289P, dme_miR-985-3P, dme_miR-286-3P, and dme_miR-14-5P), four mRNA (SERCA, frizzled, Abl, and CG31064), and two lncRNAs (XLOC_027736 and XLOC_189909) were selected to analyse the tissue specificity of the head, ovary, gut and fat body of female fruit flies at 7 days. Each sample included three biological replicates. The expression levels of these genes were detected by qRT-PCR as described above. Differential expression levels were compared by independent-samples t-tests between groups.
## Data availability statement
The original contributions presented in the study are publicly available. This data can be found here: https://www.ncbi.nlm.nih.gov/ (Accession number PRJNA716466).
## Author contributions
DY and FX: writing data analysis; JL: data analysis; XF: quality control data verification; QN, YL, MZ, and TY: data visualization and quality control; DY: revision and typesetting; ZH and MY: design of the study revision.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fgene.2023.1096902/full#supplementary-material
## References
1. Abdollahzadeh R., Daraei A., Mansoori Y., Sepahvand M., Amoli M. M., Tavakkoly-Bazzaz J.. **Competing endogenous RNA (ceRNA) cross talk language in ceRNA regulatory networks: A new look at hallmarks of breast cancer**. *J. Cell Physiol.* (2019) **234** 10080-10100. DOI: 10.1002/jcp.27941
2. Aradska J., Bulat T., Sialana F. J., Birner-Gruenberger R., Erich B., Lubec G.. **Gel-free mass spectrometry analysis of**. *Proteomics* (2015) **15** 3356-3360. DOI: 10.1002/pmic.201500092
3. Barter T. T., Greenspan Z. S., Phillips M. A., Mueller L. D., Rose M. R., Ranz J. M.. **Drosophila transcriptomics with and without ageing**. *Biogerontology* (2019) **20** 699-710. DOI: 10.1007/s10522-019-09823-4
4. Belote J. M., Hoffmann F. M., McKeown M., Chorsky R. L., andBaker B. S.. **Cytogenetic analysis of chromosome region 73AD of**. *Genetics* (1990) **125** 783-793. DOI: 10.1093/genetics/125.4.783
5. Bezprozvanny I. B.. **Calcium signaling and neurodegeneration**. *Acta Naturae* (2010) **2** 72-80. DOI: 10.32607/actanaturae.10761
6. Bonchuk A., Denisov S., Georgiev P., Maksimenko O.. **Drosophila BTB/POZ domains of "ttk group" can form multimers and selectively interact with each other**. *J. Mol. Biol.* (2011) **412** 423-436. DOI: 10.1016/j.jmb.2011.07.052
7. Brenman-Suttner D. B., Yost R. T., Frame A. K., Robinson J. W., Moehring A. J., Simon A. F.. **Social behavior and aging: A fly model**. *Genes Brain Behav.* (2020) **19** e12598. DOI: 10.1111/gbb.12598
8. Brown C. J., Kaufman T., Trinidad J. C., Clemmer D. E.. **Proteome changes in the aging**. *Int. J. Mass Spectrom.* (2018) **425** 36-46. DOI: 10.1016/j.ijms.2018.01.003
9. Buchon N., Osman D., David F. P., Fang H. Y., Boquete J. P., Deplancke B.. **Morphological and molecular characterization of adult midgut compartmentalization in Drosophila**. *Cell Rep.* (2013) **3** 1725-1738. DOI: 10.1016/j.celrep.2013.04.001
10. Bushati N., Cohen S. M.. **microRNA functions**. *Annu. Rev. Cell Dev. Biol.* (2007) **23** 175-205. DOI: 10.1146/annurev.cellbio.23.090506.123406
11. Bushey D., Tononi G., Cirelli C.. **The Drosophila fragile X mental retardation gene regulates sleep need**. *J. Neurosci.* (2009) **29** 1948-1961. DOI: 10.1523/JNEUROSCI.4830-08.2009
12. Cha S. J., Do H. A., Choi H. J., Lee M., Kim K.. **The Drosophila model: Exploring novel therapeutic compounds against neurodegenerative diseases**. *Antioxidants (Basel)* (2019) **8** 623. DOI: 10.3390/antiox8120623
13. Chen D., Liu J., Meng J., Li D., Zhao P., Duan Y.. **Integrative analysis of long non-coding RNAs (lncRNAs), miRNAs, and mRNA-associated ceRNA network in lung tissue of aging mice and changes after treatment with codonopsis pilosula**. *Med. Sci. Monit.* (2020) **26** e921580. DOI: 10.12659/MSM.921580
14. Chen J., Huang Y., Qi G.. **LncRNA-IRAR-mediated regulation of insulin receptor transcripts in**. *Insect Mol. Biol.* (2022) **31** 261-272. DOI: 10.1111/imb.12756
15. Chen P., Chen F., Lei J., Li Q., Zhou B.. **Activation of the miR-34a-mediated SIRT1/mTOR signaling pathway by urolithin A attenuates D-galactose-induced brain aging in mice**. *Neurotherapeutics* (2019) **16** 1269-1282. DOI: 10.1007/s13311-019-00753-0
16. Chen Y. W., Song S., Weng R., Verma P., Kugler J. M., Buescher M.. **Systematic study of Drosophila microRNA functions using a collection of targeted knockout mutations**. *Dev. Cell* (2014) **31** 784-800. DOI: 10.1016/j.devcel.2014.11.029
17. Clarke G. D., Li J., Kuo A. H., Moody A. J., Nathanielsz P. W.. **Cardiac magnetic resonance imaging: Insights into developmental programming its consequences for aging**. *J. Dev. Orig. Health Dis.* (2020) **12** 203. DOI: 10.1017/S2040174420001233
18. De Groef S., Wilms T., Balmand S., Calevro F., Callaerts P.. **Sexual dimorphism in metabolic responses to western diet in**. *Biomolecules* (2021) **12** 33. DOI: 10.3390/biom12010033
19. Dori M., Bicciato S.. **Integration of bioinformatic predictions and experimental data to identify circRNA-miRNA associations**. *Genes (Basel)* (2019) **10** 642. DOI: 10.3390/genes10090642
20. Duan H., de Navas L. F., Hu F., Sun K., Mavromatakis Y. E., Viets K.. **The mir-279/996 cluster represses receptor tyrosine kinase signaling to determine cell fates in the Drosophila eye**. *Development* (2018) **145** dev159053. DOI: 10.1242/dev.159053
21. Duncan R. S., Goad D. L., Grillo M. A., Kaja S., Payne A. J., Koulen P.. **Control of intracellular calcium signaling as a neuroprotective strategy**. *Molecules* (2010) **15** 1168-1195. DOI: 10.3390/molecules15031168
22. Dupont P., Besson M. T., Devaux J., Lievens J. C.. **Reducing canonical Wingless/Wnt signaling pathway confers protection against mutant Huntingtin toxicity in Drosophila**. *Neurobiol. Dis.* (2012) **47** 237-247. DOI: 10.1016/j.nbd.2012.04.007
23. Enomoto M., Siow C., Igaki T.. **Drosophila as a cancer model**. *Adv. Exp. Med. Biol.* (2018) **1076** 173-194. DOI: 10.1007/978-981-13-0529-0_10
24. Enright A. J., John B., Gaul U., Tuschl T., Sander C., Marks D. S.. **MicroRNA targets in Drosophila**. *Genome Biol.* (2003) **5** R1. DOI: 10.1186/gb-2003-5-1-r1
25. Fox D. T., Cohen E., Smith-Bolton R.. **Model systems for regeneration: Drosophila**. *Development* (2020) **147** dev173781. DOI: 10.1242/dev.173781
26. Franz A., Shlyueva D., Brunner E., Stark A., Basler K.. **Probing the canonicity of the Wnt/Wingless signaling pathway**. *PLoS Genet.* (2017) **13** e1006700. DOI: 10.1371/journal.pgen.1006700
27. Friedlander M. R., Mackowiak S. D., Li N., Chen W., Rajewsky N.. **miRDeep2 accurately identifies known and hundreds of novel microRNA genes in seven animal clades**. *Nucleic Acids Res.* (2012) **40** 37-52. DOI: 10.1093/nar/gkr688
28. Fulford A. D., Holder M. V., Frith D., Snijders A. P., Tapon N., Ribeiro P. S.. **Casein kinase 1 family proteins promote Slimb-dependent Expanded degradation**. *Elife* (2019) **8** e46592. DOI: 10.7554/eLife.46592
29. Gao Y., Wang J., Zhao F.. **Ciri: An efficient and unbiased algorithm for de novo circular RNA identification**. *Genome Biol.* (2015) **16** 4. DOI: 10.1186/s13059-014-0571-3
30. Garcia D. M., Baek D., Shin C., Bell G. W., Grimson A., Bartel D. P.. **Weak seed-pairing stability and high target-site abundance decrease the proficiency of lsy-6 and other microRNAs**. *Nat. Struct. Mol. Biol.* (2011) **18** 1139-1146. DOI: 10.1038/nsmb.2115
31. Gendron C. M., Pletcher S. D.. **MicroRNAs mir-184 and let-7 alter Drosophila metabolism and longevity**. *Aging Cell* (2017) **16** 1434-1438. DOI: 10.1111/acel.12673
32. Gillingham A. K., Sinka R., Torres I. L., Lilley K. S., Munro S.. **Toward a comprehensive map of the effectors of rab GTPases**. *Dev. Cell* (2014) **31** 358-373. DOI: 10.1016/j.devcel.2014.10.007
33. Gubina N., Naudi A., Stefanatos R., Jove M., Scialo F., Fernandez-Ayala D. J.. **Essential physiological differences characterize short- and long-lived strains of**. *J. Gerontol. A Biol. Sci. Med. Sci.* (2019) **74** 1835-1843. DOI: 10.1093/gerona/gly143
34. Hartl F. U.. **Cellular homeostasis and aging**. *Annu. Rev. Biochem.* (2016) **85** 1-4. DOI: 10.1146/annurev-biochem-011116-110806
35. He Q. Y., Liu G. H., Bao Z. J., Kong Q. P., Tao J., Yu Y. Q.. **Chinese aging anti-aging experts agree**. *Geriatr. Health Care* (2019) **25** 551-553
36. Hinkley T., Brown H., Carson V., Teychenne M.. **Cross sectional associations of screen time and outdoor play with social skills in preschool children**. *PLoS One* (2018) **13** e0193700. DOI: 10.1371/journal.pone.0193700
37. Huang C. H., Lin T. Y., Pan R. L., Juang J. L.. **The involvement of Abl and PTP61F in the regulation of Abi protein localization and stability and lamella formation in Drosophila S2 cells**. *J. Biol. Chem.* (2007) **282** 32442-32452. DOI: 10.1074/jbc.M702583200
38. Ji S., Luo Y., Cai Q., Cao Z., Zhao Y., Mei J.. **LC domain-mediated coalescence is essential for otu enzymatic activity to extend Drosophila lifespan**. *Mol. Cell* (2019) **74** 363-377.e5. DOI: 10.1016/j.molcel.2019.02.004
39. John A., Kubosumi A., Reddy P. H.. **Mitochondrial MicroRNAs in aging and neurodegenerative diseases**. *Cells* (2020) **9** 1345. DOI: 10.3390/cells9061345
40. Kamiya Y., Miyazono K., Miyazawa K.. **Specificity of the inhibitory effects of dad on TGF-beta family type I receptors, thickveins, saxophone, and baboon in Drosophila**. *FEBS Lett.* (2008) **582** 2496-2500. DOI: 10.1016/j.febslet.2008.05.052
41. Khodaei L., Long T. A. F.. **Kin recognition and co-operative foraging in**. *J. Evol. Biol.* (2019) **32** 1352-1361. DOI: 10.1111/jeb.13531
42. Kim S. S., Lee S. V.. **Non-Coding RNAs in**. *Mol. Cells* (2019) **42** 379-385. DOI: 10.14348/molcells.2019.0077
43. Kinser H. E., Pincus Z.. **MicroRNAs as modulators of longevity and the aging process**. *Hum. Genet.* (2020) **139** 291-308. DOI: 10.1007/s00439-019-02046-0
44. Knupp D., Miura P.. **CircRNA accumulation: A new hallmark of aging?**. *Mech. Ageing Dev.* (2018) **173** 71-79. DOI: 10.1016/j.mad.2018.05.001
45. Kong Y., Wu J., Yuan L.. **MicroRNA expression analysis of adult-onset Drosophila Alzheimer's disease model**. *Curr. Alzheimer Res.* (2014) **11** 1-891. DOI: 10.2174/1567205011666141001121416
46. L'Episcopo F., Serapide M. F., Tirolo C., Testa N., Caniglia S., Morale M. C.. **A Wnt1 regulated Frizzled-1/β-Catenin signaling pathway as a candidate regulatory circuit controlling mesencephalic dopaminergic neuron-astrocyte crosstalk: Therapeutical relevance for neuron survival and neuroprotection**. *Mol. Neurodegener.* (2011) **6** 49. DOI: 10.1186/1750-1326-6-49
47. Lai W. F., Lin M., Wong W. T.. **Tackling aging by using miRNA as a target and a tool**. *Trends Mol. Med.* (2019) **25** 673-684. DOI: 10.1016/j.molmed.2019.04.007
48. Leaman D., Chen P. Y., Fak J., Yalcin A., Pearce M., Unnerstall U.. **Antisense-mediated depletion reveals essential and specific functions of microRNAs in Drosophila development**. *Cell* (2005) **121** 1097-1108. DOI: 10.1016/j.cell.2005.04.016
49. Liang Y., Liang Q., Qiao L., Xiao F.. **MicroRNAs modulate drug resistance-related mechanisms in hepatocellular carcinoma**. *Front. Oncol.* (2020) **10** 920. DOI: 10.3389/fonc.2020.00920
50. Liu L., Wu C. F.. **Distinct effects of abelson kinase mutations on myocytes and neurons in dissociated Drosophila embryonic cultures: Mimicking of high temperature**. *PLoS One* (2014) **9** e86438. DOI: 10.1371/journal.pone.0086438
51. Liu S., Wang Z., Zhu R., Wang F., Cheng Y., Liu Y.. **Three differential expression analysis methods for RNA sequencing: Limma, EdgeR, DESeq2**. *J. Vis. Exp.* (2021) **175**. DOI: 10.3791/62528
52. Liu W., Ganguly A., Huang J., Wang Y., Ni J. D., Gurav A. S.. **Neuropeptide F regulates courtship in Drosophila through a male-specific neuronal circuit**. *Elife* (2019) **8** e49574. DOI: 10.7554/eLife.49574
53. Love M. I., Huber W., Anders S.. **Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2**. *Genome Biol.* (2014) **15** 550. DOI: 10.1186/s13059-014-0550-8
54. Luhur A., Buddika K., Ariyapala I. S., Chen S., Sokol N. S.. **Opposing post-transcriptional control of InR by FMRP and LIN-28 adjusts stem cell-based tissue growth**. *Cell Rep.* (2017) **21** 2671-2677. DOI: 10.1016/j.celrep.2017.11.039
55. Magwire M. M., Yamamoto A., Carbone M. A., Roshina N. V., Symonenko A. V., Pasyukova E. G.. **Quantitative and molecular genetic analyses of mutations increasing Drosophila life span**. *PLoS Genet.* (2010) **6** e1001037. DOI: 10.1371/journal.pgen.1001037
56. Manley P. W., Huth F., Moussaoui S., Schoepfer J.. **A kinase inhibitor which specifically targets the ABL myristate pocket (STAMP), but unlike asciminib crosses the blood–brain barrier**. *Bioorg. Med. Chem. Lett.* (2022) **59** 128577. DOI: 10.1016/j.bmcl.2022.128577
57. Memczak S., Jens M., Elefsinioti A., Torti F., Krueger J., Rybak A.. **Circular RNAs are a large class of animal RNAs with regulatory potency**. *Nature* (2013) **495** 333-338. DOI: 10.1038/nature11928
58. Mishan M. A., Khazeei Tabari M. A., Parnian J., Fallahi J., Mahrooz A., Bagheri A.. **Functional mechanisms of miR-192 family in cancer**. *Genes Chromosom. Cancer* (2020) **12** 722-735. DOI: 10.1002/gcc.22889
59. Mołoń M., Dampc J., Kula-Maximenko M., Zebrowski J., Mołoń A., Dobler R.. **Effects of temperature on lifespan of**. *Insects* (2020) **11** 470. DOI: 10.3390/insects11080470
60. Musselman L. P., Fink J. L., Baranski T. J.. **Similar effects of high-fructose and high-glucose feeding in a Drosophila model of obesity and diabetes**. *PLoS One* (2019) **14** e0217096. DOI: 10.1371/journal.pone.0217096
61. Nainu F., Salim E., Asri R. M., Hori A., Kuraishi T.. **Neurodegenerative disorders and sterile inflammation: Lessons from a Drosophila model**. *J. Biochem.* (2019) **166** 213-221. DOI: 10.1093/jb/mvz053
62. Neely G. G., Kuba K., Cammarato A., Isobe K., Amann S., Zhang L.. **A global**. *Cell* (2010) **141** 142-153. DOI: 10.1016/j.cell.2010.02.023
63. Nelson C., Ambros V., Baehrecke E. H.. **miR-14 regulates autophagy during developmental cell death by targeting ip3-kinase**. *Mol. Cell* (2014) **256** 376-388. DOI: 10.1016/j.molcel.2014.09.011
64. Nesler K. R., Sand R. I., Symmes B. A., Pradhan S. J., Boin N. G., Laun A. E.. **The miRNA pathway controls rapid changes in activity-dependent synaptic structure at the**. *PLoS One* (2013) **8** e68385. DOI: 10.1371/journal.pone.0068385
65. Nesler K. R., Starke E. L., Boin N. G., Ritz M., Barbee S. A.. **Presynaptic CamKII regulates activity-dependent axon terminal growth**. *Mol. Cell Neurosci.* (2016) **76** 33-41. DOI: 10.1016/j.mcn.2016.08.007
66. Perry N., Volin M., Toledano H.. **microRNAs in Drosophila regulate cell fate by repressing single mRNA targets**. *Int. J. Dev. Biol.* (2017) **61** 165-170. DOI: 10.1387/ijdb.160271ht
67. Ren S., Lin P., Wang J., Yu H., Lv T., Sun L.. **Circular RNAs: Promising molecular biomarkers of human aging-related diseases via functioning as an miRNA sponge**. *Mol. Ther. Methods Clin. Dev.* (2020) **18** 215-229. DOI: 10.1016/j.omtm.2020.05.027
68. Reynolds J. A., Peyton J. T., Denlinger D. L.. **Changes in microRNA abundance may regulate diapause in the flesh fly,**. *Insect Biochem. Mol. Biol.* (2017) **84** 1-14. DOI: 10.1016/j.ibmb.2017.03.002
69. Robins H., Li Y., Padgett R. W.. **Incorporating structure to predict microRNA targets**. *Proc. Natl. Acad. Sci. U.S.A.* (2005) **102** 4006-4009. DOI: 10.1073/pnas.0500775102
70. Robinson M. D., McCarthy D. J., Smyth G. K.. **edgeR: a Bioconductor package for differential expression analysis of digital gene expression data**. *Bioinformatics* (2010) **26** 139-140. DOI: 10.1093/bioinformatics/btp616
71. Ruan Y., Li Z., Shen Y., Li T., Zhang H., Guo J.. **Functions of circular RNAs and their potential applications in gastric cancer**. *Expert Rev. Gastroenterol. Hepatol.* (2020) **14** 85-92. DOI: 10.1080/17474124.2020.1715211
72. Sanz A., Fernández-Ayala D. J., Stefanatos R. K., Jacobs H. T.. **Mitochondrial ROS production correlates with, but does not directly regulate lifespan in Drosophila**. *Aging (Albany NY)* (2010) **2** 200-223. DOI: 10.18632/aging.100137
73. Shi Z., Lim C., Tran V., Cui K., Zhao K., Chen X.. **Single-cyst transcriptome analysis of Drosophila male germline stem cell lineage**. *Development* (2020) **147** dev184259. DOI: 10.1242/dev.184259
74. Soler Beatty J., Molnar C., Luque C. M., de Celis J. F., Martín-Bermudo M. D.. **EGFRAP encodes a new negative regulator of the EGFR acting in both normal and oncogenic EGFR/Ras-driven tissue morphogenesis**. *PLoS Genet.* (2021) **17** e1009738. DOI: 10.1371/journal.pgen.1009738
75. Song C., Zhu C., Wu Q., Qi J., Gao Y., Zhang Z.. **Metabolome analysis of effect of aspirin on Drosophila lifespan extension**. *Exp. Gerontol.* (2017) **95** 54-62. DOI: 10.1016/j.exger.2017.04.010
76. Stallone G., Infante B., Prisciandaro C., Grandaliano G.. **mTOR and aging: An old fashioned dress**. *Int. J. Mol. Sci.* (2019) **20** 2774. DOI: 10.3390/ijms20112774
77. Suh Y. S., Bhat S., Hong S. H., Shin M., Bahk S., Cho K. S.. **Genome-wide microRNA screening reveals that the evolutionary conserved miR-9a regulates body growth by targeting sNPFR1/NPYR**. *Nat. Commun.* (2015) **6** 7693. DOI: 10.1038/ncomms8693
78. Sun K., Jee D., de Navas L. F., Duan H., Lai E. C.. **Multiple**. *PLoS Genet.* (2015) **11** e1005245. DOI: 10.1371/journal.pgen.1005245
79. Surguchov A., Emamzadeh F. N., Surguchev A. A.. **Amyloidosis and longevity: A lesson from plants**. *Biol. (Basel)* (2019) **8** 43. DOI: 10.3390/biology8020043
80. Thibault O., Gant J. C., Landfield P. W.. **Expansion of the calcium hypothesis of brain aging and Alzheimer's disease: Minding the store**. *Aging Cell* (2007) **6** 307-317. DOI: 10.1111/j.1474-9726.2007.00295.x
81. Tower J.. **Drosophila flies in the face of aging**. *J. Gerontol. A Biol. Sci. Med. Sci.* (2019) **74** 1539-1541. DOI: 10.1093/gerona/glz159
82. Ulgherait M., Chen A., McAllister S. F., Kim H. X., Delventhal R., Wayne C. R.. **Circadian regulation of mitochondrial uncoupling and lifespan**. *Nat. Commun.* (2020) **11** 1927. DOI: 10.1038/s41467-020-15617-x
83. Ullah M., Ng N. N., Concepcion W., Thakor A. S.. **Emerging role of stem cell-derived extracellular microRNAs in age-associated human diseases and in different therapies of longevity**. *Ageing Res. Rev.* (2020) **57** 100979. DOI: 10.1016/j.arr.2019.100979
84. Varet H., Brillet-Gueguen L., Coppee J. Y., Dillies M. A.. **SARTools: A DESeq2-and EdgeR-based R pipeline for comprehensive differential analysis of RNA-seq data**. *PLoS One* (2016) **11** e0157022. DOI: 10.1371/journal.pone.0157022
85. Varghese J., Cohen S. M.. **microRNA miR-14 acts to modulate a positive autoregulatory loop controlling steroid hormone signaling in Drosophila**. *Genes Dev.* (2007) **21** 2277-2282. DOI: 10.1101/gad.439807
86. Wang J. Y., Yang Y., Ma Y., Wang F., Xue A., Zhu J.. **Potential regulatory role of lncRNA-miRNA-mRNA axis in osteosarcoma**. *Biomed. Pharmacother.* (2020) **121** 109627. DOI: 10.1016/j.biopha.2019.109627
87. Wang L., Chen S., Liu Y., Zhang H., Ren N., Ma R.. **The biological and diagnostic roles of MicroRNAs in meningiomas**. *Rev. Neurosci.* (2020) **31** 771-778. DOI: 10.1515/revneuro-2020-0023
88. Wang M., Hu Q., Lv T., Wang Y., Lan Q., Xiang R.. **High-resolution 3D spatiotemporal transcriptomic maps of developing Drosophila embryos and larvae**. *Dev. Cell* (2022) **57** 1271-1283.e4. DOI: 10.1016/j.devcel.2022.04.006
89. Wang P., Xu L. L., Zheng X. B., Hu Y. T., Zhang J. F., Ren S. S.. **Correlation between the expressions of circular RNAs in peripheral venous blood and clinicopathological features in hepatocellular carcinoma**. *Ann. Transl. Med.* (2020) **8** 338. DOI: 10.21037/atm.2020.02.134
90. Wang Q., Cai J., Fang C., Yang C., Zhou J., Tan Y.. **Mesenchymal glioblastoma constitutes a major ceRNA signature in the TGF-beta pathway**. *Theranostics* (2018) **8** 4733-4749. DOI: 10.7150/thno.26550
91. Watanabe L. P., Riddle N. C.. **New opportunities: Drosophila as a model system for exercise research**. *J. Appl. Physiol.* (2019) **127** 482-490. DOI: 10.1152/japplphysiol.00394.2019
92. Weigelt C. M., Sehgal R., Tain L. S., Cheng J., Esser J., Pahl A.. **An insulin-sensitive circular RNA that regulates lifespan in Drosophila**. *Mol. Cell* (2020) **79** 268-279.e5. DOI: 10.1016/j.molcel.2020.06.011
93. Wessels H. H., Lebedeva S., Hirsekorn A., Wurmus R., Akalin A., Mukherjee N.. **Global identification of functional microRNA-mRNA interactions in Drosophila**. *Nat. Commun.* (2019) **10** 1626. DOI: 10.1038/s41467-019-09586-z
94. Westholm J. O., Miura P., Olson S., Shenker S., Joseph B., Sanfilippo P.. **Genome-wide analysis of drosophila circular RNAs reveals their structural and sequence properties and age-dependent neural accumulation**. *Cell Rep.* (2014) **9** 1966-1980. DOI: 10.1016/j.celrep.2014.10.062
95. Xu P., Vernooy S. Y., Guo M., Hay B. A.. **The Drosophila microRNA Mir-14 suppresses cell death and is required for normal fat metabolism**. *Curr. Biol.* (2003) **13** 790-795. DOI: 10.1016/s0960-9822(03)00250-1
96. Xue B., Chuang C. H., Prosser H. M., Fuziwara C. S., Chan C., Sahasrabudhe N.. **miR-200 deficiency promotes lung cancer metastasis by activating Notch signaling in cancer-associated fibroblasts**. *Genes Dev.* (2021) **35** 1109-1122. DOI: 10.1101/gad.347344.120
97. Xue L., Su D., Li D., Gao W., Yuan R., Pang W.. **Retraction note to: miR-200 regulates epithelial-mesenchymal transition in anaplastic thyroid cancer via EGF/EGFR signaling**. *Cell Biochem. Biophys.* (2019) **77** 277. DOI: 10.1007/s12013-019-00879-0
98. Yang D., Lian T., Tu J., Gaur U., Mao X., Fan X.. **LncRNA mediated regulation of aging pathways in**. *Aging (Albany NY)* (2016) **8** 2182-2203. DOI: 10.18632/aging.101062
99. Yang Y., Xu S., Xia L., Wang J., Wen S., Jin P.. **The bantam microRNA is associated with drosophila fragile X mental retardation protein and regulates the fate of germline stem cells**. *PLoS Genet.* (2009) **5** e1000444. DOI: 10.1371/journal.pgen.1000444
100. Young M. D., Wakefield M. J., Smyth G. K., Oshlack A.. **Gene ontology analysis for RNA-seq: Accounting for selection bias**. *Genome Biol.* (2010) **11** R14. DOI: 10.1186/gb-2010-11-2-r14
101. Yue B., Wang J., Song C., Wu J., Cao X., Huang Y.. **Biogenesis and ceRNA role of circular RNAs in skeletal muscle myogenesis**. *Int. J. Biochem. Cell Biol.* (2019) **117** 105621. DOI: 10.1016/j.biocel.2019.105621
102. Zeng H., Lu B., Zamponi R., Yang Z., Wetzel K., Loureiro J.. **mTORC1 signaling suppresses Wnt/β-catenin signaling through DVL-dependent regulation of Wnt receptor FZD level**. *Proc. Natl. Acad. Sci. U. S. A.* (2018) **115** E10362-E10369. DOI: 10.1073/pnas.1808575115
103. Zha F., Qu X., Tang B., Li J., Wang Y., Zheng P.. **Long non-coding RNA MEG3 promotes fibrosis and inflammatory response in diabetic nephropathy via miR-181a/Egr-1/TLR4 axis**. *Aging (Albany NY)* (2019) **11** 3716-3730. DOI: 10.18632/aging.102011
104. Zhang R., Du J., Zhao X., Wei L., Zhao Z.. **Regulation of circadian behavioural output via clock-responsive miR-276b**. *Insect Mol. Biol.* (2021a) **30** 81-89. DOI: 10.1111/imb.12679
105. Zhang R., Zhao X., Du J., Wei L., Zhao Z.. **Regulatory mechanism of daily sleep by miR-276a**. *FASEB J.* (2021b) **35** e21222. DOI: 10.1096/fj.202001220R
106. Zhang Y., Yu F., Bao S., Sun J.. **Systematic characterization of circular RNA-associated CeRNA network identified novel circRNA biomarkers in Alzheimer's disease**. *Front. Bioeng. Biotechnol.* (2019) **7** 222. DOI: 10.3389/fbioe.2019.00222
107. Zhao L., Hu K., Cao J., Wang P., Li J., Zeng K.. **lncRNA miat functions as a ceRNA to upregulate sirt1 by sponging miR-22-3p in HCC cellular senescence**. *Aging (Albany NY)* (2019) **11** 7098-7122. DOI: 10.18632/aging.102240
108. Zhou R. S., Zhang E. X., Sun Q. F., Ye Z. J., Liu J. W., Zhou D. H.. **Integrated analysis of lncRNA-miRNA-mRNA ceRNA network in squamous cell carcinoma of tongue**. *BMC Cancer* (2019) **19** 779. DOI: 10.1186/s12885-019-5983-8
109. Zhu Z., Bhat K. M.. **The Hem protein mediates neuronal migration by inhibiting WAVE degradation and functions opposite of Abelson tyrosine kinase**. *Dev. Biol.* (2011) **357** 283-294. DOI: 10.1016/j.ydbio.2011.06.025
|
---
title: Reprogramming of VEGF-mediated extracellular matrix changes through autocrine
signaling
authors:
- Eibhlin Goggins
- Yelena Mironchik
- Samata Kakkad
- Desmond Jacob
- Flonne Wildes
- Zaver M. Bhujwalla
- Balaji Krishnamachary
journal: Cancer Biology & Therapy
year: 2023
pmcid: PMC10012930
doi: 10.1080/15384047.2023.2184145
license: CC BY 4.0
---
# Reprogramming of VEGF-mediated extracellular matrix changes through autocrine signaling
## ABSTRACT
Vascular endothelial growth factor (VEGF) plays key roles in angiogenesis, vasculogenesis, and wound healing. In cancers, including triple negative breast cancer (TNBC), VEGF has been associated with increased invasion and metastasis, processes that require cancer cells to traverse through the extracellular matrix (ECM) and establish angiogenesis at distant sites. To further understand the role of VEGF in modifying the ECM, we characterized VEGF-mediated changes in the ECM of tumors derived from TNBC MDA-MB-231 cells engineered to overexpress VEGF. We established that increased VEGF expression by these cells resulted in tumors with reduced collagen 1 (Col1) fibers, fibronectin, and hyaluronan. Molecular characterization of tumors identified an increase of MMP1, uPAR, and LOX, and a decrease of MMP2, and ADAMTS1. α-SMA, a marker of cancer associated fibroblasts (CAFs), increased, and FAP-α, a marker of a subset of CAFs associated with immune suppression, decreased with VEGF overexpression. Analysis of human data from The Cancer Genome Atlas Program confirmed mRNA differences for several molecules when comparing TNBC with high and low VEGF expression. We additionally characterized enzymatic changes induced by VEGF overexpression in three different cancer cell lines that clearly identified autocrine-mediated changes, specifically uPAR, in these enzymes. Unlike the increase of Col1 fibers and fibronectin mediated by VEGF during wound healing, in the TNBC model, VEGF significantly reduced key protein components of the ECM. These results further expand our understanding of the role of VEGF in cancer progression and identify potential ECM-related targets to disrupt this progression.
## Introduction
Vascular endothelial growth factor-A (VEGF-A or VEGF) is a versatile dimeric glycoprotein that plays important roles in normal tissue function such as in wound-healing and embryonic development, and in pathologies such as diabetic and hypersensitive retinopathy, rheumatoid arthritis, age-related macular degeneration, and cancer.1 Factors that regulate VEGF include hypoxia inducible factor-1α (HIF1-α), nuclear factor kappa-B (NF-kB), transforming growth factor (TFG-β), endothelin-1 and mechanical stress.2,3 VEGF is a potent angiogenic and vascular permeability factor and its expression is tightly coupled to oxygenation due to the presence of several hypoxia response elements in its promoter region.4 In cancers, VEGF plays an important role in tumor angiogenesis, vascular permeability, tumor growth, and metastasis.1 VEGF has six main isoforms, VEGF111, VEGF121, VEGF145, VEGF165, VEGF189, and VEGF2065–7 that have distinct effects on tumor growth and progression.8 Of these, VEGF165 has been extensively studied as it is the most frequently expressed isoform in tissues.1 The role of VEGF signaling in cancer, beyond its role in angiogenesis, is rapidly evolving. VEGF promotes cancer cell proliferation, migration and invasiveness,9 promotes stemness,10,11 and promotes immune suppression.12 Increased VEGF expression has been identified in several cancers,13 and it is associated with poor prognosis and increased metastasis in multiple cancers including triple negative breast cancer (TNBC).14–18 The VEGF targeted monoclonal antibody, bevacizumab, is approved in a range of solid tumor indications.19 Because of the role of VEGF in tumor angiogenesis as well as in invasion and metastasis, there is significant interest in understanding the role of VEGF in modifying the tumor extracellular matrix (ECM). Cancer cells have to navigate through the ECM on their metastatic journey20 and the establishment of neovasculature requires ECM remodeling.20,21 We previously identified increased matrigel degradation by VEGF overexpressing MCF-7 human breast cancer cells in an intact-cell perfusion system.9 In vivo magnetic resonance imaging (MRI) studies with VEGF overexpressing MCF-7 and MDA-MB-231 tumors, revealed a significant increase of vascular volume and permeability, changes in macromolecular transport through the ECM, and increased metastasis.9 Studies investigating the effects of VEGF on the tumor ECM have primarily focused on characterizing changes in the matrix metalloproteinases (MMPs) and other degradative enzymes.22,23 Effects of increased VEGF expression or VEGF targeting with bevacizumab on bone cartilage and osteoarthritis have also been previously described.24 Here, we have directly characterized changes in key ECM components such as collagen 1 (Col1), fibronectin (FN1), and hyaluronan (HA), together with alterations in degradative enzymes and cancer associated fibroblast (CAF) markers, to understand how VEGF alters the ECM in a human TNBC xenograft overexpressing VEGF165. We independently confirmed degradative enzyme changes in human TNBC with high and low VEGF mRNA levels by analyzing data from The Cancer Genome Atlas Program. We additionally characterized enzymatic changes in VEGF overexpressing MDA-MB-231 cells, as well as in human prostate cancer PC-3 cells, and in estrogen receptor (ER) positive MCF-7 breast cancer cells engineered to overexpress VEGF. Several of the changes in MDA-MB-231 VEGF overexpressing tumors were also observed in the cells suggesting that autocrine signaling mediated these changes. A significant increase of uPAR was observed with VEGF overexpression in both MDA-MB-231 and PC-3 cells, but not in MCF-7 cells. These results expand our understanding of the role of VEGF in ECM remodeling and provide new insights into the role of VEGF in TNBC.
## Cells and tumors
MDA-MB-231, PC-3 and MCF-7 cancer cells were obtained from ATCC (Manassas, VA). Establishment of VEGF165 overexpressing cancer cells was done as previously described9 and validated for VEGF expression by ELISA following manufacturer’s instruction (R&D, Minneapolis, MN)9,25 and by RT-PCR.
Two million MDA-MB-231 wild type (231_WT) or VEGF overexpressing (231_VEGF) cells were inoculated in the mammary fat pad of 4–6 weeks old female severe combined immunodeficient (SCID) mice. Tumors were excised once they reached a volume of ~300-500 mm3. Studies were performed with 5–10 tumors from each group. One half of each tumor was fixed in formalin for immunohistochemistry and the other half freeze-clamped for molecular analysis. 231_VEGF tumors were validated for VEGF expression by ELISA following manufacturer’s instructions (R&D, Minneapolis, MN).9,25 Animal handling was conducted in accordance with the regulations outlined by the Institutional Animal Care and Use Committee of Johns Hopkins University.
## Second harmonic generation (SHG) microscopy
SHG microscopy of hematoxylin and eosin (H&E) stained sections was performed as previously described.26 Briefly, tumors were paraffin-embedded and 5 μm thick H&E sections were used for SHG microscopy. Slides were analyzed using an Olympus Laser Scanning FV1000 MPE multiphoton microscope (Olympus Corp., US headquarters–Center Valley, PA) with a 25Xw/ 1.05XLPLN MP lens. Excitation was achieved at 860 nm and the second harmonic signal was detected at a wavelength of 430 nm. Col1 fiber parameters of percent fiber volume and inter-fiber distance were quantified, and Haralick texture features such as contrast and homogeneity were analyzed, using an in-house fiber analysis software written in MATLAB 7.4.0 (The MathWorks, Natick, MA, USA) as previously described.26
## Immunohistochemistry
Formalin-fixed, paraffin-embedded sections of tumors were deparaffinized followed by antigen retrieval. Antibodies used for immunohistochemistry (IHC) of targets-of-interest were: rat monoclonal CD31 antibody (Dianova, Hamburg, Germany) at 1:30 dilution, rabbit anti-Col1A1 antibody cross-reactive with mouse and human Col1A1 (OriGene, Rockville, MD, USA) at 1:70 dilution, mouse-monoclonal anti-FN1 antibody cross-reactive with mouse and human FN1 (Immunogen- Fusion proteinAg8016, Proteintech, Rosemont, IL, USA) at 1:100 dilution, and bovine nasal cartilage HABP (Millipore Sigma, Merck KGaA, Darmstadt, Germany) at 1:750 dilution. Slides were incubated overnight at 4°C. Following this, sections were incubated with horseradish peroxidase conjugated with anti-mouse or anti-rabbit IgG. For HABP, the VECTASTAIN ABC-AP Kit procedure was used. Finally, slides were stained with 3,3′-diaminobenzidine (DAB) and counterstained with hematoxylin.
High-resolution digital scans of the stained sections (five tumors per group, 1 section analyzed per tumor) were obtained using ScanScope (Aperio, Vista, CA). Quantification was done with the ImageScope software using the Positive Pixel Count V9 algorithm supplied by the manufacturer. The number of strongly positive or positive pixels normalized to the total number of pixels was obtained. Analyses were performed using the entire histological section, with entire viable and necrotic regions in the section mapped from adjacent H&E stained slides.
## RT-PCR and immunoblotting
RNA isolation was done using Qiagen kit (Qiagen, Valencia, CA, USA). To obtain RNA, tissues were homogenized with RLT buffer and passed through a QIAshredder. cDNA was synthesized using an iScript cDNA synthesis kit (Bio-Rad, Hercules, CA, USA).
For quantitative real-time PCR (qRT-PCR), 1 μl of 1:10 diluted cDNA was used. IQ SYBR Green Supermix and gene-specific primers in the iCycler RT-PCR detection system (Bio-Rad, Hercules, CA, USA) were used. For the ECM proteins, Col1A1, Col1A2, and FN1, and for fibroblast activation protein alpha (FAP-α), mouse ECM specific primers were designed. The house-keeping genes, hypoxanthine phosphoribosyltransferase-1 (HPRT-1) and 18s ribosomal RNA (18s rRNA), were used as controls. The threshold cycle (ct) from these house-keeping genes was used to calculate the expression of human and mouse-specific genes. The change in threshold cycle (Δct) values between HPRT-1 for targets of human origin and 18s for mouse related targets, and the gene of interest was calculated for 231_WT and 231_VEGF samples. To obtain ΔΔct and the fold mRNA expression, the average Δct of the 231_WT samples was subtracted from the Δct values of the 231_VEGF sample. Using the formula 2−ΔΔct, fold mRNA expression of individual samples was determined and plotted using GraphPad prism.
Protein isolation and immunoblotting was performed as previously described.27 Antibodies cross-reactive with mouse/human ECM proteins and specific for human enzymes of interest included rabbit-polyclonal anti-Col1A1 antibody (1:1000; OriGene, Rockville, MD, USA), mouse monoclonal anti-FN1 antibody (1:2000 dilution; Proteintech, Rosemont, IL, USA), rabbit polyclonal anti-MMP-1 antibody (1:1000 dilution; Neo BioLab, Woburn, MA, USA), rabbit polyclonal anti-MMP2 (1:1000 dilution; GeneTex, Inc., Irvine, CA, USA), rabbit polyclonal anti-MMP-14 antibody (1:1000 dilution; Neo BioLab, Woburn, MA, USA), mouse monoclonal anti-lysyl oxidase (LOX) antibody (1:1000 dilution; GeneTex, Inc., Irvine, CA, USA), mouse monoclonal anti-ADAMTS1 antibody (1:500 dilution; OriGene, Rockville, MD, USA), rabbit polyclonal anti-uPAR (1:1000 dilution; GeneTex, Inc., Irvine, CA, USA), mouse monoclonal anti-α-SMA antibody (1:2000; Novus Biologicals, Littleton, CO, USA), rabbit monoclonal antibody against neuropilin-1 (NRP-1) (1:1000, clone D62C6, Cell Signaling, Danvers, MA, USA, rabbit polyclone anti-FLT1 (VEGFR1) antibody (1:1000, MyBioSource, San Diego, CA) and rabbit polyclonal anti-FAP-α antibody (1:1000, Ab207178, Abcam, Cambridge, UK). Horseradish peroxidase-conjugated secondary antibodies were used at 1:2000 dilution. Blots were visualized using the SuperSignal West Pico Chemiluminescent substrate kit (Thermo Scientific, Rockford, IL, USA). The reference band from the molecular weight marker was used to determine the location of the protein of interest. Autoradiographs were scanned, and densitometry of the band intensities of various proteins of interest were obtained using ImageJ software. The band intensity for each protein was normalized to the intensity of GAPDH protein used as a loading control. Values are represented as Mean ± Standard Error of the Mean (SEM) from five individual tumor samples for the in vivo studies and at least three biological replicates for the cell studies.
## Human breast cancer analysis
Publicly available TCGA data sets for breast cancer were retrieved from the TCGA Data portal (https://tcga-data.nci.nih.gov/tcga)28 using the open-access, open-source, web-based platform cBioPortal for Cancer genomics (cbioportal.org).29 Clinical identifiers were applied to select treatment naïve patient samples that were triple negative. Three studies, the Korean breast cancer cohort study (SMC-2018),30 the breast invasive carcinoma TCGA study (TCGA-2015),31 and the breast invasive carcinoma TCGA Firehose Legacy study (Firehose Legacy), with source data from the repository at Broad Institute Genomic Data Analysis Center (GDAC), met our criteria. Based on curated RNA sequencing data for a given study, we next applied genomic filters to group patient samples with high and low VEGFA mRNA expression based on the z-score of samples (log RNA Seq V2 RSEM). Z-score values greater than 1.2-fold were grouped as VEGFA-high and values less than 1.2-fold were grouped as VEGFA-low. In the case of the TCGA study,30 RNA sequencing data were provided as transcripts per million (TPM). For this study, TPM values for VEGA expression were divided into four quartiles using the tool available in cBioPortal web portal. Data from the highest and lowest quartile were analyzed as high and low VEGFA expressing tumors. Comparison analyses for MMP1, uPAR, LOX and α-SMA were performed for VEGF-high and VEGF-low data sets following the instructions in cBioportal.32 The cBioportal-derived expression data presented in this study are based on a frequently employed analysis technique called the RNA-seq by Expectation Maximization (RSEM). RSEM takes into account the transcript length and provides acceptable and accurate results.33
## Statistical analysis
Statistical analysis was performed using GraphPad Prism (San Diego, CA). P values ≤.05 were considered significant. P values were based on a two-tailed t-test for the ELISA and mRNA analysis, and a one-tailed t-test for the IHC studies, based on the sample size. For the TCGA data sets, a Mann-Whitney test was performed, as the non-parametric Mann-Whitney test is most appropriate for large-scale RNA seq data. Additionally, the TCGA data sets were also evaluated with a two-tailed t-test.
## Validation of VEGF overexpression in cells and tumors and its effects on tumor vasculature
ELISA performed on supernatant derived from cells and protein isolated from tumor-derived samples showed a statistically significant increase of VEGF in 231_VEGF cells (Figure 1a) and 231_VEGF tumors (Figure 1b) compared to wild-type cells and tumors. We used CD31 as a marker of endothelial cells to confirm the functional effects of VEGF on increasing tumor vasculature. As shown in the representative IHC images in Figure 1c,a higher number of CD31 immunostained pixels was detected in 231_VEGF tumors (right) compared to 231_WT tumors (left). These results summarized in Figure 1d identified a trend (P ≤.07) of increased vessel density in VEGF overexpressing tumors. Growth curves for 231_WT and 231_VEGF tumors, fitted to a Gompertzian curve, are shown in Figure 1e. Tumor growth was significantly higher in 231_VEGF tumors compared to 231_WT tumors. A significant difference between the average doubling time for 231_WT and 231_VEGF tumors was observed (P ≤.05). The tumor doubling time (Td) was approximately 9 ± 1.32 d for 231_WT tumors compared to 5.5 ± 0.28 d for 231_VEGF tumors, estimated for tumor volumes from 110 mm3 to 300 mm3 (values represent Mean ± S.E.M.). We additionally characterized VEGF mRNA as well as VEGF levels in cell supernatant and lysate by ELISA for PC-3 and MCF-7 cells overexpressing VEGF to confirm increased levels of VEGF in these cells as shown in Supplementary Figures 1a,b. Figure 1.Validation of VEGF Overexpression. ( a) Validation of the overexpression of VEGF in the supernatant of 231_VEGF cells ($$n = 5$$) compared to 231_WT cells ($$n = 5$$). ( b) Validation of the overexpression of VEGF in tumors derived from 231_VEGF cells ($$n = 5$$) compared to tumors from 231_WT cells ($$n = 5$$). ( c) Representative CD31 immunostained images from 231_WT (left) and 231_VEGF (right) tumors. ( d) Quantification of strongly positive pixel (NSP) area normalized to the total pixel area in 231_WT ($$n = 5$$) and 231_VEGF ($$n = 5$$) tumors. 1 section was analyzed from each tumor. ( e) Growth curves of 231_WT and 231_VEGF tumors ($$n = 10$$ per group); each point in the graph represents a single tumor volume. Values represent Mean ± S.E.M. *P ≤.05,****P ≤.00005.Effects of VEGF overexpression on VEGF levels in MDA-MB-231 cells and tumors, vessel density as detected by CD31 immunostaining, and tumor volume doubling time.
## VEGF overexpression reduced Col1, fibronectin, and hyaluronan
We next characterized Col1 fibers in tumor sections using SHG microscopy. Representative images of Col1 fibers obtained using SHG microscopy in Figure 2a from 231_VEGF (right) and 231_WT tumors (left) show the significant decrease of Col1 fibers in VEGF overexpressing tumors. Compared to 231_WT tumors, percent fiber volume in 231_VEGF tumors significantly decreased (Figure 2b) and interfiber distance significantly increased (Figure 2c). Haralick feature analysis identified a significant decrease in contrast (Figure 2d) and a significant increase in homogeneity (Figure 2e) with VEGF overexpression. Figure 2.Col1 fiber changes detected with SHG microscopy. ( a) Representative Col1 fiber images acquired using SHG microscopy from 231_WT (left) and 231_VEGF (right) tumors. The images were acquired at a pixel resolution of 0.83 μm x 0.83 μm with a FOV of 425 × 425 μm2.(b) Quantification of percent fiber volume, (c) Interfiber distance, (d) contrast and (e) homogeneity from SHG image analysis of 231_WT ($$n = 5$$) and 231_VEGF ($$n = 5$$) tumors. Values represent Mean ± S.E.M. * P ≤.05, ** P ≤.005.Effects of VEGF overexpression on MDA-MB-231 tumor collagen 1 fiber decrease as detected by second harmonic generation microscopy.
The significant decrease in Col1 fibers identified with SHG microscopy was further confirmed with IHC of Col1A1. Representative viable and necrotic areas of immunostained sections from 231_WT tumors in Figure 3a and 231_VEGF tumors in Figure 3b show decreased Col1A1 immunostaining in viable tumor regions of VEGF overexpressing tumors, but not in necrotic tumor regions. These data, summarized in Figure 3c,d, demonstrate the significant decrease of Col1A1 with VEGF overexpression in viable tumor regions (Figure 3c), but not in necrotic tumor regions (Figure 3d). Along with Col1A1, a significant decrease of FN1 in viable tumor regions was detected with VEGF overexpression. Representative viable and necrotic areas of immunostained sections from 231_WT tumors in Figure 4a and 231_VEGF tumors in Figure 4b show the decreased FN1 immunostaining in viable tumor regions of VEGF overexpressing tumors, but not in necrotic tumor regions. These data, summarized in Figure 4c,d, demonstrate the significant decrease of FN1 with VEGF overexpression in viable tumor regions (Figure 4c), but not in necrotic tumor regions with VEGF overexpression (Figure 4d). Figure 3.Col1A1 immunostaining. ( a) Representative images of Col1A1 immunostained tumor sections from viable (left) and necrotic (right) regions of 231_WT tumors. ( b) Representative images of Col1A1 immunostained tumor sections from viable (left) and necrotic (right) regions of 231_VEGF tumors. Quantification of Col1A1 positive pixels normalized to the total pixel area of (c) viable tumor regions and (d) necrotic tumor regions from 231_WT ($$n = 5$$) and 231_VEGF ($$n = 5$$) tumors. Values represent Mean ± S.E.M. * P ≤.05.Effects of VEGF overexpression on MDA-MB-231 tumor collagen 1A1 fiber decrease as detected by immunostaining of tumor sections. Figure 4.Fibronectin immunostaining. ( a) Representative images of fibronectin immunostained tumor sections from viable (left) and necrotic (right) regions of 231_WT tumors. ( b) Representative images of fibronectin immunostained tumor sections from viable (left) and necrotic (right) regions of 231_VEGF tumors. Quantification of fibronectin positive pixels normalized to the total pixel area of (c) viable tumor regions and (d) necrotic tumor regions from 231_WT ($$n = 5$$) and 231_VEGF ($$n = 5$$) tumors. Values represent Mean ± S.E.M. * P ≤.05.Effects of VEGF overexpression on MDA-MB-231 tumor fibronectin-1 decrease as detected by immunostaining of tumor sections.
Immunostaining of HA binding protein (HABP) was used to characterize changes in HA in VEGF overexpressing tumors. Representative viable and necrotic areas of HABP immunostained sections from 231_WT tumors in Figure 5a and 231_VEGF tumors in Figure 5b show decreased HABP immunostaining in viable tumor regions of VEGF overexpressing tumors, and in necrotic tumor regions. These data, summarized in Figure 5c,d, identified a trend (P ≤.08) toward decreased HABP with VEGF overexpression in viable tumor regions (Figure 5c), and a significant decrease with VEGF overexpression in necrotic tumor regions (Figure 5d). Figure 5.HABP Immunostaining. ( a) Representative images of HABP immunostained tumor sections from viable (left) and necrotic (right) regions of 231_WT tumors. ( b) Representative images of HABP immunostained tumor sections from viable (left) and necrotic (right) regions of 231_VEGF tumors. Quantification of HABP positive pixels normalized to the total pixel area of (c) viable tumor regions and (d) necrotic tumor regions from 231_WT ($$n = 5$$) and 231_VEGF ($$n = 5$$) tumors, 1 section analyzed per tumor. Values represent Mean ± S.E.M. *P ≤.05.Changes in hyaluronic acid binding protein with VEGF overexpression in MDA-MB-231 tumors as detected by immunostaining of tumor sections.
## Immunoblot analyses of tumor tissue and cells
Immunoblot analysis of tumors further confirmed a clear reduction of Col1A1 and FN1 protein with VEGF overexpression, as shown in Figure 6a and summarized in Supplementary Figure 2a. Figure 6.*Immunoblot analysis* of ECM proteins and enzymes in tumors. ( a) Decrease of Col1A1 and fibronectin identified in 231_VEGF compared with 231_WT tumor samples. ( b) Increase of MMP-1 in 231_VEGF compared with 231_WT tumor samples. ( c) Decrease of MMP-2, and increase of uPAR in 231_VEGF compared with 231_WT tumor samples. ( d) Decrease of ADAMST1 and (e) increase of LOX in 231_VEGF compared with 231_WT tumor samples. ( f) Increase of α-SMA and decrease of FAP-α in 231_VEGF compared with 231_WT tumor samples. GAPDH was used as a loading control. $$n = 5$$ for 231_WT and $$n = 5$$ for 231_VEGF in all immunoblots. Changes in MDA-MB-231 tumor collagen 1A1 and fibronectin-1, enzymes, and fibroblast markers with VEGF overexpression as detected by immunoblotting of tumor tissue.
Because ECM remodeling is frequently achieved by the action of various enzymes, we interrogated changes in MMP1, MMP2, MMP14, urokinase-type plasminogen activator receptor (uPAR), ADAM Metallopeptidase with Thrombospondin Type 1 Motif 1 (ADAMTS1), and LOX in VEGF overexpressing tumors. MMP1, uPAR and LOX clearly increased with VEGF overexpression as shown in Figures 6b-e and summarized in Supplemental Figures 2B-E. MMP2 and ADAMTS1 decreased with VEGF overexpression as shown in Figure 6c,d, and summarized in Supplementary Figures 2C and D. Unlike the other ECM degrading enzymes, we did not identify a clear change in MMP14 with VEGF overexpression (data not shown). Since CAFs play a major role in ECM synthesis, we also evaluated two well-established markers of CAFs, α-SMA and FAP-α. We identified a trend (P ≤.08) of an increase of α-SMA, and a decrease of FAP-α in 231_VEGF tumors compared with 231_WT tumors (figure 6f, and Supplementary Figure 2 F). Multiple bands appearing in some immunoblots were either due to non-specific binding or due to protein phosphorylation.
We also characterized the same ECM remodeling enzymes in MDA-MB-231, PC-3 and MCF-7 VEGF overexpressing cells as shown in Figure 7. With the exception of ADAMTS1 that increased, MDA-MB-231 cells overexpressing VEGF showed changes similar to those observed in tumors for uPAR, MMP1, MMP2 and a trend toward increased LOX expression (P ≤.09)(Figure 7a, Supplementary Figure 3A). Similar to MDA-MB-231 cells, PC-3 cells overexpressing VEFG showed an increase of uPAR, but ADAMTS1, LOX and MMP1 decreased or remained unchanged; MCF-7 cells did not exhibit any change in these enzymes with VEGF overexpression (Figure 7b). To understand the autocrine mechanisms causing these changes, we characterized the VEGF binding receptors VEGFR1 and NRP-1 in these cells. The data presented in Figure 7c demonstrate that both MDA-MB-231, and PC-3, but not MCF-7, wild-type cells expressed high levels of NRP-1 supporting the possibility of autocrine signaling. Similar levels of NRP-1 were detected in MDA-MB-231 (Figure 7c, Supplementary Figure 3B) and PC-3 VEGF overexpressing cells (Figure 7c). NRP-1 was low in MCF-7 VEGF overexpressing cells. VEGFR1 was not detected in wild type or VEGF overexpressing cells in all three cell lines (data not shown). Figure 7.*Immunoblot analysis* of enzymes and receptors in cancer cells. ( a) Changes in ADAMTS1, uPAR, LOX, MMP1 and MMP2 in 231_VEGF cells compared to 231_WT cells. ( b) Changes in ADAMTS1, LOX, uPAR and MMP1 in wild type and VEGF overexpressing PC-3 and MCF-7 cells. ( c) Changes in NRP-1 in MDA-MB-231, PC-3 and MCF-7 wild type and VEGF overexpressing cells. Effect of VEGF overexpression on cellular enzymes, and VEGF receptor NRP1 in MDA-MB-231, PC-3 and MCF-7 cells.
## mRNA changes in human samples and xenografts
We mined the TCGA TARGET GTEx database to characterize mRNA changes in the degradative enzymes MMP1 (Figure 8a), uPAR (Figure 8b) and LOX (Figure 8c), and the CAF marker α-SMA (Figure 8d), in treatment naïve TNBC with high and low VEGF mRNA. Statistical analyses of the fold change were evaluated using the Mann-Whitney test and a 2-tailed t-test. Identical significant changes were identified with both tests. A significant increase of MMP1 mRNA in 2 of 3 studies, LOX mRNA in 1 of 3 studies with a trend toward increased expression in a second study (P ≤.08 in TCGA Firehose Legacy), uPAR mRNA in all 3 studies, and α-SMA mRNA in 1 of 3 studies with a trend toward increased expression in the TCGA Firehose Legacy study (P ≤.06) were identified in the high VEGF mRNA group compared to the low VEGF group, consistent with the changes observed in the tumor xenograft studies. Figure 8.mRNA expression in treatment naïve TNBC with high and low VEGF expression. Comparison of RNA sequencing data expressed as log2 based on RSEM method of analysis of (a) MMP1, (b) uPAR, (c) LOX and (d) α-SMA in TNBC patient samples with high or low VEGFA expression from three different studies. Data represent geometric mean at $95\%$ confidence interval from 5 low VEGFA (blue dots) and 34 high VEGFA (red dots) samples for the TCGA Firehose Legacy study, 8 samples each for low VEGFA and high VEGFA samples in the SMC 2018 Study. In TCGA_2015, there were only 2 samples in the low VEGFA (Blue) and 26 samples in the high VEGFA (red dots) group. Statistical analysis was performed with Mann Whitney test using GraphPad prism software and further confirmed with a two-tailed t-test. * P ≤.05, ** P ≤.005.Results from mining TCGA data to identify changes in enzymes and fibroblast marker mRNA in treatment naïve triple negative human breast cancers with high and low VEGF.
We characterized mRNA of the ECM proteins and degradative enzymes to understand transcriptional changes induced by VEGF overexpression in the xenografts. mRNA levels of Col1A1, Col1A2 and FN1 significantly decreased in 231_VEGF tumors (Supplementary Figures 4A-C) consistent with the reduction identified with IHC and in the immunoblots. Also, consistent with the immunoblots, mRNA of MMP1 and uPAR significantly increased (Supplementary Figures 4D-E), and mRNA of MMP2, ADAMTS1, and FAP-α significantly decreased (Supplemental Figures 4 F-H). We also confirmed a significant increase of VEGFA mRNA (Supplemental Figure 4I).
## Discussion
Our studies identified a clear reduction of key ECM components, Col1A1, FN1 and HA in MDA-MB-231 xenografts with VEGF overexpression. The patterns of Col1 fibers, identified by Haralick feature analysis, were also altered by VEGF overexpression. Increased expression of VEGF was confirmed directly in cells and tumors, as well as from the functional changes of increased vascularity detected by the endothelial cell marker, CD31, and increased tumor growth. Changes in enzymes such as an increase of MMP1, uPAR, and LOX, and a decrease of MMP2 and ADAMTS1, together with an increase of CAFs, most likely contributed to the ECM changes in the VEGF overexpressing tumors. Increases in mRNA expression for various matrix degrading enzymes together with the increase in α-SMA mRNA in the xenografts were confirmed in the TCGA analysis of treatment naïve TNBC with high and low VEGFA data sets. Our purpose with the TCGA analysis was to determine if the protein expression changes identified in our tumor models were reflected in publicly available human data, although the probed genes in the TCGA data may not necessarily predict protein expression.
The autocrine and intracrine roles of VEGF34 were evident from the changes in enzymes observed in VEGF overexpressing cancer cells. With the exception of ADAMTS1 that decreased in 231-VEGF tumors but increased in cells, the increase of uPAR, LOX, MMP1 and a decrease of MMP2 observed in the tumors was also observed in 231_VEGF cells that suggested that the enzymatic changes driving the ECM changes occurred directly within the cancer cells.
uPAR increased significantly in PC-3 prostate cancer cells overexpressing VEGF, and was consistently higher in the human TNBC data. MCF-7 breast cancer cells that are ER +ve and poorly invasive did not show any alterations of the degradative enzymes investigated.
The autocrine changes mediated by VEGF most likely occurred through binding of VEGF to NRP-1 that showed high expression in the MDA-MB-231 and PC-3 wild type and VEGF overexpressing cells, but not MCF-7 cells that was consistent with the absence of any alterations of the degradative enzymes investigated following VEGF overexpression. Previous observations of an increase of ECM degradation by VEGF overexpressing breast cancer cells9 are also consistent with enzyme changes occurring through autocrine signaling in the cells. Similar autocrine signaling mediated modulation of enzymes by VEGF was previously observed in A549 lung cancer cells,35 vascular smooth muscle cells,23 and chondrocytes.36 Both paracrine and autocrine signaling play a role in tumor microenvironment changes mediated by VEGF as demonstrated in a study where suppressing VEGF decreased metastasis via disrupting both the autocrine and paracrine signaling loops of VEGF.37 In MDA-MB-231 tumors, the three enzymes that increased with VEGF overexpression were MMP1, uPAR and LOX, while MMP2 and ADAMTS1 decreased. The increase of uPAR is consistent with the reduction of the ECM components observed here since uPAR plays a role in generating plasmin and activating the MMPs. MMP1, or collagenase 1, is a matrix metalloproteinase that specifically degrades collagen 1. The increase of MMP1 can explain the reduction of Col1 fibers observed with VEGF overexpression, despite the increase of LOX the enzyme that plays a role in cross-linking Col1 fibers.38 The reduction of MMP2 and ADAMTS1 may have indirectly contributed to ECM changes. MMP2 is a type IV collagenase that plays a role in releasing growth factors bound to the ECM and in degrading the basement membrane.39 ADAMTS1 is part of a family of extracellular proteolytic enzymes known to have diverse functions related to ECM remodeling, angiogenesis, cell migration and organogenesis. Dysregulation of these enzymes has been implicated in various diseases including multiple cancers. In breast cancer, overexpression of ADAMTS1 was shown to promote tumor progression and to be upregulated in metastatic TNBC (reviewed in40). Under physiological conditions, ADAMTS1 has been shown to sequester VEGF165 thereby acting as an angiogenesis inhibitor; binding of ADAMTS1 to VEGF165 disrupts the binding and phosphorylation of VEGFR2 leading to the suppression of endothelial cell proliferation.41,42 In wound-healing, VEGF-mediated increases of collagen and FN1 have been frequently documented.43 During wound-healing, the increased vascular permeability induced by VEGF results in extravasation of proteins such as fibrin and FN1 which provides a temporary matrix to nourish fibroblasts, facilitate their motility, and induce them to deposit a more structured collagen stroma.44 *Our data* revealed that, unlike during wound repair, in our TNBC xenograft model, VEGFA overexpression resulted in a decrease of Col1A1 and FN1, despite an increase of the CAF marker α-SMA, indicating that VEGF-mediated changes in the ECM are reprogrammed in cancer. Although, to the best of our knowledge, a direct evaluation of ECM changes with VEGF overexpression has not been performed, changes in the tumor ECM observed with anti-VEGF treatments provide useful insights in understanding the changes observed.
Three major ECM proteins, Col1A1, FN1 and HA, significantly decreased with VEGF overexpression. Collagen fibers are the most abundant structural protein in the ECM. TNBC, in particular, has a significantly increased deposition of collagen as well as increased matrix stiffness compared with luminal breast cancer subtypes.45 Previous studies have identified an association between increased Col1 fibers and metastasis in breast46 and prostate cancer.47 Here, SHG microscopy together with immunohistochemistry and molecular analysis clearly demonstrated that VEGF overexpression reduced Col1 fibers, protein and mRNA. Hypoxia is a potent transcriptional regulator of VEGF. The changes in Col1 fibers with VEGF overexpression are consistent with earlier observations that hypoxic tumor regions exhibit fewer Col1 fibers.26 Similar to Col1, FN1 was reduced by VEGF overexpression. FN1 is a glycoprotein that is present in dimeric and multimeric form in the ECM. It binds through an RGD sequence to endothelial cells, and is frequently found localized with endothelial cells.48 Here, despite an increase of the endothelial cell marker, CD31, with VEGF overexpression, FN1, in viable tumor regions, significantly decreased. FN1 plays important roles in cell migration, growth, and differentiation.49 FN1 interacts with Col1 in the tumor ECM during tumorigenesis.50 FN1 has been shown to contribute to tumor growth, progression, migration, and response to therapy (reviewed in51). In breast cancer, intracellular FN1 has been associated with increased metastasis.52 Treatment of a colon cancer xenograft with bevacizumab resulted in a significant increase of fibronectin,53 indirectly supporting our observation that increased VEGF reduces fibronectin.
Since Col1 and FN1 are associated with increased metastasis,46,52,54 their reduction with VEGF overexpression, which is also known to increase metastasis in these tumors,9 was surprising. It is possible that a less dense ECM may be more conducive to metastasis, or that although VEGF overexpression results in an ECM that is less permissive to metastasis, increased vascularity, and invasiveness override this with the net outcome of an increase of metastasis.
In addition to Col1 and FN1, we observed a reduction of HA, a high-molecular weight, unbranched, nonsulfated glycosaminoglycan that is an important structural component of various tissues including the tumor ECM. HA regulates adhesion, cell proliferation, EMT, gene expression, invasion, motility, signaling, metastasis, and morphogenesis55,56 by its ability to bind to various HA binding proteins collectively called hyaladherins.57,58 Additionally, breakdown products of HA can stimulate angiogenesis.59 Although an increase of HA is frequently associated with increased invasion,60 degradation of HA has also been observed in many pathologies including inflammation and cancer.61 Studies of colorectal cancer liver metastases treated with anti-VEGF therapy have identified an increase of HA and sulfated glycosaminoglycans and increased tumor stiffness after anti-VEGF treatment, although no significant changes in collagen were detected.62 The ECM changes observed here represent a net outcome of synthesis by CAFs and degradation by enzymes. CAFs are one of the most abundant stromal cell population in the tumor microenvironment, performing diverse functions such as synthesis and remodeling of the ECM, promoting migration, invasion, and metastasis, and affecting immune cell function.63,64 CAFs are a heterogeneous population of cells derived from multiple cellular sources.63 Here, α-SMA, which is expressed by most CAFs,65 increased with VEGF overexpression. These results are consistent with an earlier study in which α-SMA levels increased in VEGF overexpressing TNBC.8 However, FAP-α, a marker of a subset of CAFs associated with immune suppression decreased.66 *It is* possible that different CAF subsets may have different responses to VEGF, and that the ECM synthesis capabilities of subsets of CAFs may be different. In a murine intrahepatic cholangiocarcinoma model, blocking placental growth factor (PIGF)-α, a member of the VEGF family, impacted only one subset of CAFs that expressed low levels of Col1.67 Reduction of FAP-α expressing CAFs may have contributed to the changes in the ECM observed here.
Our data highlight the autocrine role of VEGF in increasing uPAR and other degradative enzymes and support the use of disrupting these autocrine or intracrine loops for treatment. A limitation of our study is that we used a single time point as a snap-shot to evaluate changes in the ECM, CAFs and enzymes in these tumors. The availability of noninvasive imaging to image the ECM, CAFs, and enzymes will allow evaluation of the spatial and temporal evolution of the reprogramming of the ECM and its causes that occurs with VEGF overexpression in tumors, compared to healing wounds.44 Such studies may provide additional insights into the role of VEGF in tumor progression and metastasis.
## Authors Contribution
Research design: EG, BK, ZMB; Data collection, Data and statistical analysis, interpretation: EG, YM, SK, DJ, FW, BK, ZMB; Quality control of data and algorithm: EG, SK, DJ, BK, ZMB; Manuscript writing, literature review: EG, BK, ZMB; Proof reading: EG, YM, DJ, FW, BK, ZMB
## Disclosure statement
No potential conflict of interest was reported by the author(s).
## Data availability statement
The data that support the findings of this study are available from the corresponding author, Zaver Bhujwalla, upon reasonable request.
## Institutional review board statement
This study does not involve any human subjects. Only curated data from the publicaly available source cBioportal (cBioportal.org/us) were used in the TCGA analysis.
## Supplementary material
Supplemental data for this article can be accessed online at https://doi.org/$\frac{10.1080}{15384047.2023.2184145}$
## References
1. Apte RS, Chen DS, Ferrara N.. **VEGF in Signaling and Disease: Beyond Discovery and Development**. *Cell* (2019) **176** 1248-15. DOI: 10.1016/j.cell.2019.01.021
2. Pagès G, Pouysségur J. **Transcriptional regulation of the Vascular Endothelial Growth Factor gene–a concert of activating factors***. *Cardiovasc Res* (2005) **65** 564-573. DOI: 10.1016/j.cardiores.2004.09.032
3. Wu MH, Huang CY, Lin JA, Wang SW, Peng CY, Cheng HC, Tang C-H. **Endothelin-1 promotes vascular endothelial growth factor-dependent angiogenesis in human chondrosarcoma cells**. *Oncogene* (2014) **33** 1725-1735. DOI: 10.1038/onc.2013.109
4. Forsythe JA, Jiang BH, Iyer NV, Agani F, Leung SW, Koos RD, Semenza GL. **Activation of vascular endothelial growth factor gene transcription by hypoxia-inducible factor 1**. *Mol Cell Biol* (1996) **16** 4604-4613. DOI: 10.1128/MCB.16.9.4604
5. Ferrara N. **Vascular endothelial growth factor: basic science and clinical progress**. *Endocr Rev* (2004) **25** 581-611. DOI: 10.1210/er.2003-0027
6. Peach CJ, Mignone VW, Arruda MA, Alcobia DC, Hill SJ, Kilpatrick LE, Robida PA, Moreno-Vinasco L, Schnaar RL, Roers A. **Molecular Pharmacology of VEGF-A Isoforms: Binding and Signalling at VEGFR2**. *Int J Mol Sci* (2018) 19. DOI: 10.3390/ijms20010019
7. Tischer E, Mitchell R, Hartman T, Silva M, Gospodarowicz D, Fiddes JC. **The human gene for vascular endothelial growth factor. Multiple protein forms are encoded through alternative exon splicing**. *J Biol Chem.* (1991) **266** 11947-11954. PMID: 1711045
8. Di Benedetto M, Toullec A, Buteau-Lozano H, Abdelkarim M, Vacher S, Velasco G, Christofari M, Pocard M, Bieche I, Perrot-Applanat M. **MDA-MB-231 breast cancer cells overexpressing single VEGF isoforms display distinct colonisation characteristics**. *Br J Cancer* (2015) **113** 773-785. DOI: 10.1038/bjc.2015.267
9. Pathak AP, McNutt S, Shah T, Wildes F, Raman V, Bhujwalla ZM, Singleton PA. **In vivo “MRI phenotyping” reveals changes in extracellular matrix transport and vascularization that mediate VEGF-driven increase in breast cancer metastasis**. *PLoS One* (2013) **8** e63146. DOI: 10.1371/journal.pone.0063146
10. Beck B, Driessens G, Goossens S, Youssef KK, Kuchnio A, Caauwe A, Sotiropoulou PA, Loges S, Lapouge G, Candi A. **A vascular niche and a VEGF–Nrp1 loop regulate the initiation and stemness of skin tumours**. *Nature* (2011) **478** 399-403. DOI: 10.1038/nature10525
11. Fantozzi A, Gruber DC, Pisarsky L, Heck C, Kunita A, Yilmaz M, Meyer-Schaller N, Cornille K, Hopfer U, Bentires-Alj M. **VEGF-mediated angiogenesis links EMT-induced cancer stemness to tumor initiation**. *Cancer Res* (2014) **74** 1566-1575. DOI: 10.1158/0008-5472.CAN-13-1641
12. Meder L, Schuldt P, Thelen M, Schmitt A, Dietlein F, Klein S, Borchmann S, Wennhold K, Vlasic I, Oberbeck S. **Combined VEGF and PD-L1 Blockade Displays Synergistic Treatment Effects in an Autochthonous Mouse Model of Small Cell Lung Cancer**. *Cancer Res* (2018) **78** 4270-4281. DOI: 10.1158/0008-5472.CAN-17-2176
13. Costache MI, Ioana M, Iordache S, Ene D, Costache CA, Saftoiu A. **VEGF Expression in Pancreatic Cancer and Other Malignancies: A Review of the Literature**. *Rom J Intern Med* (2015) **53** 199-208. DOI: 10.1515/rjim-2015-0027
14. Zhan P, Wang J, Lv XJ, Wang Q, Qiu LX, Lin XQ, Yu LK, Song Y. **Prognostic value of vascular endothelial growth factor expression in patients with lung cancer: a systematic review with meta-analysis**. *J Thorac Oncol* (2009) **4** 1094-1103. DOI: 10.1097/JTO.0b013e3181a97e31
15. Gasparini G, Toi M, Gion M, Verderio P, Dittadi R, Hanatani M, Gion M, Dittadi R, Verderio P, Boracchi P. **Prognostic significance of vascular endothelial growth factor protein in node-negative breast carcinoma**. *J Natl Cancer Inst* (1997) **89** 139-147. DOI: 10.1093/jnci/89.2.139
16. Linderholm B, Tavelin B, Grankvist K, Henriksson R. **Vascular endothelial growth factor is of high prognostic value in node-negative breast carcinoma**. *J Clin Oncol* (1998) **16** 3121-3128. DOI: 10.1200/JCO.1998.16.9.3121
17. Saponaro C, Malfettone A, Ranieri G, Danza K, Simone G, Paradiso A, Mangia A. **VEGF, HIF-1alpha expression and MVD as an angiogenic network in familial breast cancer**. *PLoS One* (2013) **8** e53070. DOI: 10.1371/journal.pone.0053070
18. Linderholm BK, Hellborg H, Johansson U, Elmberger G, Skoog L, Lehtio J, Lewensohn R. **Significantly higher levels of vascular endothelial growth factor (VEGF) and shorter survival times for patients with primary operable triple-negative breast cancer**. *Ann Oncol* (2009) **20** 1639-1646. DOI: 10.1093/annonc/mdp062
19. Garcia J, Hurwitz HI, Sandler AB, Miles D, Coleman RL, Deurloo R, Chinot OL. **Bevacizumab (Avastin(R)) in cancer treatment: A review of 15 years of clinical experience and future outlook**. *Cancer Treat Rev* (2020) **86** 102017. DOI: 10.1016/j.ctrv.2020.102017
20. Cox TR, Erler JT. **Remodeling and homeostasis of the extracellular matrix: implications for fibrotic diseases and cancer**. *Dis Model Mech* (2011) **4** 165-178. DOI: 10.1242/dmm.004077
21. Neve A, Cantatore FP, Maruotti N, Corrado A, Ribatti D. **Extracellular matrix modulates angiogenesis in physiological and pathological conditions**. *Biomed Res Int* (2014) **2014** 756078. DOI: 10.1155/2014/756078
22. Ito TK, Ishii G, Chiba H, Ochiai A. **The VEGF angiogenic switch of fibroblasts is regulated by MMP-7 from cancer cells**. *Oncogene* (2007) **26** 7194-7203. DOI: 10.1038/sj.onc.1210535
23. Wang H, Keiser JA. **Vascular endothelial growth factor upregulates the expression of matrix metalloproteinases in vascular smooth muscle cells: role of flt-1**. *Circ Res* (1998) **83** 832-840. DOI: 10.1161/01.RES.83.8.832
24. Nagai T, Sato M, Kobayashi M, Yokoyama M, Tani Y, Mochida J. **Bevacizumab, an anti-vascular endothelial growth factor antibody, inhibits osteoarthritis**. *Arthritis Res Ther* (2014) **16** 427. DOI: 10.1186/s13075-014-0427-y
25. Dore-Savard L, Lee E, Kakkad S, Popel AS, Bhujwalla ZM. **The Angiogenic Secretome in VEGF overexpressing Breast Cancer Xenografts**. *Sci Rep* (2016) **6** 39460. DOI: 10.1038/srep39460
26. Kakkad SM, Solaiyappan M, O’Rourke B, Stasinopoulos I, Ackerstaff E, Raman V, Bhujwalla ZM, Glunde K. **Hypoxic tumor microenvironments reduce collagen I fiber density**. *Neoplasia* (2010) **12** 608-617. DOI: 10.1593/neo.10344
27. Goggins E, Kakkad S, Mironchik Y, Jacob D, Wildes F, Krishnamachary B, Bhujwalla ZM. **Hypoxia Inducible Factors Modify Collagen I Fibers in MDA-MB-231 Triple Negative Breast Cancer Xenografts**. *Neoplasia* (2018) **20** 131-139. DOI: 10.1016/j.neo.2017.11.010
28. Haeussler M, Zweig AS, Tyner C, Speir ML, Rosenbloom KR, Raney BJ, Lee CM, Lee BT, Hinrichs AS, Gonzalez JN. **The UCSC Genome Browser database: 2019 update**. *Nucleic Acids Res* (2019) **47** D853-D8. DOI: 10.1093/nar/gky1095
29. Cerami E, Gao J, Dogrusoz U, Gross BE, Sumer SO, Aksoy BA, Jacobsen A, Byrne CJ, Heuer ML, Larsson E. **The cBio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data**. *Cancer Discov* (2012) **2** 401-404. DOI: 10.1158/2159-8290.CD-12-0095
30. Kan Z, Ding Y, Kim J, Jung HH, Chung W, Lal S, Cho S, Fernandez-Banet J, Lee SK, Kim SW, Kan Z, Ding Y, Kim J, HH Jung, Chung W, Lal S. **Multi-omics profiling of younger Asian breast cancers reveals distinctive molecular signatures**. *Nat Commun* (2018) **9** 1725. DOI: 10.1038/s41467-018-04129-4
31. Ciriello G, Gatza ML, Beck AH, Wilkerson MD, Rhie SK, Pastore A, Zhang H, McLellan M, Yau C, Kandoth C. **Comprehensive Molecular Portraits of Invasive Lobular Breast Cancer**. *Cell* (2015) **163** 506-519. DOI: 10.1016/j.cell.2015.09.033
32. Gao J, Aksoy BA, Dogrusoz U, Dresdner G, Gross B, Sumer SO, Sun Y, Jacobsen A, Sinha R, Larsson E. **Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal**. *Sci Signal* (2013) **6**. DOI: 10.1126/scisignal.2004088
33. Li B, Dewey CN. **RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome**. *BMC Bioinform* (2011) **12** 323. DOI: 10.1186/1471-2105-12-323
34. Wiszniak S, Schwarz Q, Broughton P, Bradshaw AD, DeLeon-Pennell KY. **Exploring the Intracrine Functions of VEGF-A**. *Biomolecules* (2021) 11. DOI: 10.3390/biom12010011
35. Desai S, Laskar S, Pandey BN. **Autocrine IL-8 and VEGF mediate epithelial-mesenchymal transition and invasiveness via p38/JNK-ATF-2 signalling in A549 lung cancer cells**. *Cell Signal* (2013) **25** 1780-1791. DOI: 10.1016/j.cellsig.2013.05.025
36. Pufe T, Harde V, Petersen W, Goldring MB, Tillmann B, Mentlein R. **Vascular endothelial growth factor (VEGF) induces matrix metalloproteinase expression in immortalized chondrocytes**. *J Pathol* (2004) **202** 367-374. DOI: 10.1002/path.1527
37. Su JC, Mar AC, Wu SH, Tai WT, Chu PY, Wu CY, Tseng L-M, Lee T-C, Chen K-F, Liu C-Y. **Disrupting VEGF-A paracrine and autocrine loops by targeting SHP-1 suppresses triple negative breast cancer metastasis**. *Sci Rep* (2016) **6** 28888. DOI: 10.1038/srep28888
38. Smith-Mungo LI, Kagan HM. **Lysyl oxidase: properties, regulation and multiple functions in biology**. *Matrix Biol* (1998) **16** 387-398. DOI: 10.1016/S0945-053X(98)90012-9
39. Monaco S, Sparano V, Gioia M, Sbardella D, Di Pierro D, Marini S, Coletta M, Monaco S, Sparano V, Gioia M, Sbardella D, Di Pierro D, Marini S. **Enzymatic processing of collagen IV by MMP-2 (gelatinase A) affects neutrophil migration and it is modulated by extracatalytic domains**. *Protein Sci* (2006) **15** 2805-2815. DOI: 10.1110/ps.062430706
40. Tan Ide A, Ricciardelli C, Russell DL. **The metalloproteinase ADAMTS1: a comprehensive review of its role in tumorigenic and metastatic pathways**. *Int J Cancer* (2013) **133** 2263-2276. DOI: 10.1002/ijc.28127
41. Luque A, Carpizo DR, Iruela-Arispe ML. **ADAMTS1/METH1 inhibits endothelial cell proliferation by direct binding and sequestration of VEGF165**. *J Biol Chem* (2003) **278** 23656-23665. DOI: 10.1074/jbc.M212964200
42. Freitas VM, do Amaral JB, Silva TA, Santos ES, Mangone FR, Pinheiro Jde J. **Decreased expression of ADAMTS-1 in human breast tumors stimulates migration and invasion**. *Mol Cancer* (2013) **12** 2. PMID: 23289900
43. Gourishetti K, Keni R, Nayak PG, Jitta SR, Bhaskaran NA, Kumar L, Kumar N, Nandakumar K, Shenoy R. **Sesamol-Loaded PLGA Nanosuspension for Accelerating Wound Healing in Diabetic Foot Ulcer in Rats**. *Int J Nanomedicine* (2020) **15** 9265-9282. DOI: 10.2147/IJN.S268941
44. Dvorak HF. **Tumors: Wounds That Do Not Heal-A Historical Perspective with a Focus on the Fundamental Roles of Increased Vascular Permeability and Clotting**. *Semin Thromb Hemost* (2019) **45** 576-592. DOI: 10.1055/s-0039-1687908
45. Acerbi I, Cassereau L, Dean I, Shi Q, Au A, Park C, Chen YY, Liphardt J, Hwang ES, Weaver VM. **Human breast cancer invasion and aggression correlates with ECM stiffening and immune cell infiltration**. *Integr Biol (Camb)* (2015) **7** 1120-1134. DOI: 10.1039/c5ib00040h
46. Kakkad SM, Solaiyappan M, Argani P, Sukumar S, Jacobs LK, Leibfritz D, Bhujwalla ZM, Glunde K. **Collagen I fiber density increases in lymph node positive breast cancers: pilot study**. *J Biomed Opt* (2012) **17** 116017. DOI: 10.1117/1.JBO.17.11.116017
47. Penet MF, Kakkad S, Pathak AP, Krishnamachary B, Mironchik Y, Raman V, Solaiyappan M, Bhujwalla ZM. **Structure and Function of a Prostate Cancer Dissemination-Permissive Extracellular Matrix**. *Clin Cancer Res* (2017) **23** 2245-2254. DOI: 10.1158/1078-0432.CCR-16-1516
48. Joshi P, Chung CY, Aukhil I, Erickson HP. **Endothelial cells adhere to the RGD domain and the fibrinogen-like terminal knob of tenascin**. *J Cell Sci* (1993) **106** 389-400. DOI: 10.1242/jcs.106.1.389
49. Missirlis D, Haraszti T, Kessler H, Spatz JP. **Fibronectin promotes directional persistence in fibroblast migration through interactions with both its cell-binding and heparin-binding domains**. *Sci Rep* (2017) **7** 3711. DOI: 10.1038/s41598-017-03701-0
50. Wang K, Wu F, Seo BR, Fischbach C, Chen W, Hsu L, Gourdon D. **Breast cancer cells alter the dynamics of stromal fibronectin-collagen interactions**. *Matrix Biol* (2017) **60-61** 86-95. PMID: 27503584
51. Wang JP, Hielscher A. **Fibronectin: How Its Aberrant Expression in Tumors May Improve Therapeutic Targeting**. *J Cancer* (2017) **8** 674-682. DOI: 10.7150/jca.16901
52. Fernandez-Garcia B, Eiro N, Marin L, Gonzalez-Reyes S, Gonzalez LO, Lamelas ML, Vizoso FJ. **Expression and prognostic significance of fibronectin and matrix metalloproteases in breast cancer metastasis**. *Histopathology* (2014) **64** 512-522. DOI: 10.1111/his.12300
53. Liang J, Cheng Q, Huang J, Ma M, Zhang D, Lei X, Xiao Z, Zhang D, Shi C, Luo L. **Monitoring tumour microenvironment changes during anti-angiogenesis therapy using functional MRI**. *Angiogenesis* (2019) **22** 457-470. DOI: 10.1007/s10456-019-09670-4
54. Reuter JA, Ortiz-Urda S, Kretz M, Garcia J, Scholl FA, Pasmooij AM, Cassarino D, Chang HY, Khavari PA. **Modeling inducible human tissue neoplasia identifies an extracellular matrix interaction network involved in cancer progression**. *Cancer Cell* (2009) **15** 477-488. DOI: 10.1016/j.ccr.2009.04.002
55. Laurent TC, Fraser JR. **Hyaluronan**. *FASEB J* (1992) **6** 2397-2404. DOI: 10.1096/fasebj.6.7.1563592
56. Underhill CB. **The interaction of hyaluronate with the cell surface: the hyaluronate receptor and the core protein**. *Ciba Found Symp* (1989) **143** 87-99. DOI: 10.1002/9780470513774.ch6
57. **Hyaluronan-binding proteins: tying up the giant**. *J Biol Chem* (2002) **277** 4585-4588. DOI: 10.1074/jbc.R100036200
58. Toole BP. **Hyaluronan: from extracellular glue to pericellular cue**. *Nat Rev Cancer* (2004) **4** 528-539. DOI: 10.1038/nrc1391
59. West DC, Hampson IN, Arnold F, Kumar S. **Angiogenesis induced by degradation products of hyaluronic acid**. *Science* (1985) **228** 1324-1326. DOI: 10.1126/science.2408340
60. Jadin L, Huang L, Wei G, Zhao Q, Gelb AB, Frost GI, Jiang P, Shepard HM. **Characterization of a novel recombinant hyaluronan binding protein for tissue hyaluronan detection**. *J Histochem Cytochem* (2014) **62** 672-683. DOI: 10.1369/0022155414540176
61. Monslow J, Govindaraju P, Pure E. **Hyaluronan - a functional and structural sweet spot in the tissue microenvironment**. *Front Immunol* (2015) **6** 231. DOI: 10.3389/fimmu.2015.00231
62. Rahbari NN, Kedrin D, Incio J, Liu H, Ho WW, Nia HT, Edrich CM, Jung K, Daubriac J, Chen I. **Anti-VEGF therapy induces ECM remodeling and mechanical barriers to therapy in colorectal cancer liver metastases**. *Sci Transl Med* (2016) **8** 360ra135. DOI: 10.1126/scitranslmed.aaf5219
63. Chen X, Song E. **Turning foes to friends: targeting cancer-associated fibroblasts**. *Nat Rev Drug Discov* (2019) **18** 99-115. DOI: 10.1038/s41573-018-0004-1
64. Houthuijzen JM, Jonkers J. **Cancer-associated fibroblasts as key regulators of the breast cancer tumor microenvironment**. *Cancer Metastasis Rev* (2018) **37** 577-597. DOI: 10.1007/s10555-018-9768-3
65. Gascard P, Tlsty TD. **Carcinoma-associated fibroblasts: orchestrating the composition of malignancy**. *Genes Dev* (2016) **30** 1002-1019. DOI: 10.1101/gad.279737.116
66. Nurmik M, Ullmann P, Rodriguez F, Haan S, Letellier E. **In search of definitions: Cancer-associated fibroblasts and their markers**. *Int J Cancer* (2020) **146** 895-905. DOI: 10.1002/ijc.32193
67. Aoki S, Inoue K, Klein S, Halvorsen S, Chen J, Matsui A, Nikmaneshi MR, Kitahara S, Hato T, Chen X. **Placental growth factor promotes tumour desmoplasia and treatment resistance in intrahepatic cholangiocarcinoma**. *Gut* (2021)
|
---
title: Association of the gut microbiome with kidney function and damage in the Hispanic
Community Health Study/Study of Latinos (HCHS/SOL)
authors:
- Brandilyn A. Peters
- Qibin Qi
- Mykhaylo Usyk
- Martha L. Daviglus
- Jianwen Cai
- Nora Franceschini
- James P. Lash
- Marc D. Gellman
- Bing Yu
- Eric Boerwinkle
- Rob Knight
- Robert D. Burk
- Robert C. Kaplan
journal: Gut Microbes
year: 2023
pmcid: PMC10012940
doi: 10.1080/19490976.2023.2186685
license: CC BY 4.0
---
# Association of the gut microbiome with kidney function and damage in the Hispanic Community Health Study/Study of Latinos (HCHS/SOL)
## ABSTRACT
### Background
The gut microbiome is altered in chronic kidney disease (CKD), potentially contributing to CKD progression and co-morbidities, but population-based studies of the gut microbiome across a wide range of kidney function and damage are lacking.
### Methods
In the Hispanic Community Health Study/Study of Latinos, gut microbiome was assessed by shotgun sequencing of stool ($$n = 2$$,438; 292 with suspected CKD). We examined cross-sectional associations of estimated glomerular filtration rate (eGFR), urinary albumin:creatinine (UAC) ratio, and CKD with gut microbiome features. Kidney trait-related microbiome features were interrogated for correlation with serum metabolites ($$n = 700$$), and associations of microbiome-related serum metabolites with kidney trait progression were examined in a prospective analysis ($$n = 3$$,635).
### Results
Higher eGFR was associated with overall gut microbiome composition, greater abundance of species from Prevotella, Faecalibacterium, Roseburia, and Eubacterium, and microbial functions related to synthesis of long-chain fatty acids and carbamoyl-phosphate. Higher UAC ratio and CKD were related to lower gut microbiome diversity and altered overall microbiome composition only in participants without diabetes. Microbiome features related to better kidney health were associated with many serum metabolites (e.g., higher indolepropionate, beta-cryptoxanthin; lower imidazole propionate, deoxycholic acids, p-cresol glucuronide). Imidazole propionate, deoxycholic acid metabolites, and p-cresol glucuronide were associated with prospective reductions in eGFR and/or increases in UAC ratio over ~6 y.
### Conclusions
Kidney function is a significant correlate of the gut microbiome, while the relationship of kidney damage with the gut microbiome depends on diabetes status. Gut microbiome metabolites may contribute to CKD progression.
## Introduction
Chronic kidney disease (CKD) afflicts approximately $15\%$ of the U.S. population1. CKD, defined as either kidney damage, generally indicated by urinary albumin, or decreased kidney function, measured by glomerular filtration rate (GFR), is associated with increased mortality2 and cardiovascular disease (CVD)3, and reduced health-related quality of life4. Prevention and management of CKD risk factors are vital to reduce CKD-related morbidity and mortality.
Known causes of CKD include hypertension, autoimmune diseases, diabetes, infections, or drug toxicity, while susceptibility to CKD is influenced by age, obesity, and genetics5. The gut microbiota, the community of microorganisms residing in the human gut, may also be important modulators of CKD risk. Many studies have shown an altered gut microbiome in patients with CKD compared to healthy controls6, perhaps attributed to CKD itself, though microbial metabolic activities, particularly the synthesis of uremic toxins, may contribute to CKD progression and CVD7–11.
Bi-directional relationships of kidney function and the gut microbiome are termed the “gut-kidney axis.” In health, gut microbiota maintain intestinal barrier integrity, preventing translocation of pathogens and inflammatory microbial products to the circulation12, and ferment polysaccharides into beneficial short-chain fatty acids (SCFAs), which may protect from CKD progression13. In CKD, there is increased translocation and retention of gut microbiome-derived products of protein fermentation, such as p-cresol sulfate, indoxyl sulfate, and phenylacetylglutamine, and these uremic toxins confer renal and cardiovascular toxicity9,12. Regarding the impact of CKD on the gut microbiota, increased gut urea secretion in CKD can lead to gut barrier degradation and overgrowth of urease-containing bacteria8,9, and patients with CKD may alter their diet and use medications which can also alter their microbiota. Gut microbiome dysbiosis caused by CKD is thought to further degrade the intestinal barrier, reduce SCFA production, and favor uremic toxin production9,12,14, implying a cyclical interaction wherein dysbiosis enhances CKD progression, increased CKD severity promotes further dysbiosis, and so on.
Previous investigations on the gut microbiome and CKD have been small (N < 300) case–control studies, many focused on end-stage renal disease (ESRD) rather than the entire range of kidney function and damage6. We sought to identify patterns of gut microbiome composition related to kidney function and damage, by assessing cross-sectional relationships of gut microbiome species and functions with estimated GFR (eGFR), urinary albumin:creatinine (UAC) ratio, and CKD in the large Hispanic Community Health Study/Study of Latinos (HCHS/SOL). High prevalence of diabetes in the study population allowed us to examine possible effect modification by diabetes status, important given known pathological differences in diabetic and non-diabetic kidney disease15. We additionally explored cross-sectional associations of kidney-related microbiome features with serum metabolites and investigated prospective associations of specific microbiome-related metabolites with change in eGFR and UAC ratio over time and incidence of CKD (Figure 1a). Figure 1.Estimated glomerular filtration rate (eGFR) is associated with overall gut microbiome composition in the Hispanic Community Health Study/Study of Latinos ($$n = 2$$,438). ( a) Overview of the present analyses: 1 – Cross-sectional analysis of the gut microbiome and kidney traits at HCHS/SOL Visit 2 ($$n = 2$$,438); 2 – Cross-sectional analysis of kidney-related gut microbiome features with serum metabolites at HCHS/SOL Visit 2 ($$n = 700$$); 3 – *Prospective analysis* of serum metabolites (specifically those strongly associated with kidney-related gut microbiome species) at HCHS/SOL Visit 1 with progression of kidney traits from Visit 1 to Visit 2 ($$n = 3$$,635). ( b) Barplots show R-squared (%) for kidney traits from PERMANOVA models of the Jensen-Shannon Divergence. Each kidney trait (predictor) was assessed in a separate model. Model 1 was adjusted for age, sex, field center, Hispanic/Latino background, U.S. nativity, antibiotics use, and Bristol stool type. Model 2 additionally adjusted for income, educational attainment, cigarette smoking, alcohol use, AHEI2010, predicted sodium intake, report of low-sodium diet, predicted protein intake, report of high protein/low carb diet, protein supplement use, and total physical activity. Model 3 additionally adjusted for BMI, waist-to-hip ratio, systolic blood pressure, diastolic blood pressure, triglycerides, HDL cholesterol, fasting glucose, hypertension medication, diabetes medication, and lipid-lowering medication. ( c) Spearman correlations of kidney traits with the first five principal coordinates of the Jensen-Shannon Divergence. ( d-e) Principal coordinate analysis plots of the first and third coordinates of the Jensen-Shannon Divergence, colored by eGFR in (d) or UAC ratio in (e); $75\%$ data ellipses for low (<60 ml/min/1.73 m2) and normal (≥60 ml/min/1.73 m2) eGFR in (d), or high (≥30) or normal (<30) UAC ratio in (e), are displayed on the plots. * $p \leq 0.05$; **$p \leq 0.01$; ***$p \leq 0.001$; ****$p \leq 0.0001.$
## Participant characteristics
Among 2,438 participants (Supplementary Figure S1), 292 ($12\%$) were classified as having CKD (134, 76, 73, 7, and 2 with stages 1–5 CKD, respectively). Participants with CKD were significantly older than participants without CKD, had higher levels of CKD risk factors including blood pressure and fasting glucose, and were more likely to be taking medication for hypertension, diabetes, and high cholesterol (all $p \leq 0.05$; Table 1). We also observed these differences comparing participants with low and normal eGFR (Supplementary Table S1), and with and without albuminuria (Supplementary Table S2). Participants with diabetes ($$n = 698$$) had significantly lower eGFR, higher UAC ratio, and higher prevalence of CKD than participants without diabetes ($$n = 1740$$) (Supplementary Table S3). Table 1.Characteristics of participants with and without chronic kidney disease (CKD)a in the HCHS/SOL Gut Origins of Latino Diabetes ancillary study. CKDNo CKDP-valuebN2922146 Age, years (mean ± SD)60.2 ± 10.555.6 ± 10.9<0.0001Male (%)39.733.20.03Field center (%) 0.37Bronx29.124.7 Chicago28.128.2 Miami17.820.6 San Diego2526.4 Hispanic/Latino background (%) 0.0002Dominican6.210.5 Central American7.29.4 Cuban12.313.4 Mexican41.842.4 Puerto Rican26.716 South American5.17.1 Mixed/missing0.71.3 U.S. nativity (%)13.7140.95Antibiotics in last 6 months (%)28.127.40.87Income (%) 0.006Less than $3000063.456.2 $30000 or more30.139.3 Missing6.54.5 Educational attainment (%) 0.0002Less than HS46.934.4 HS or equivalent20.221.8 More than HS31.542.7 Missing1.41.1 Cigarette smoking (%) 0.08Never57.263.7 Former26.423.1 Current16.413.2 Alcohol use (%) 0.04Nondrinker52.744.8 Low-level use44.551.8 High-level use2.73.4 Alternate Healthy Eating Index 2010 score (mean ± SD)50.7 ± 7.950.3 ± 7.50.63Predicted sodium intake, mg/d (mean ± SD)2928 ± 850.72985.3 ± 843.90.26Low sodium diet (%)27.423.50.16Predicted protein intake, g/d (mean ± SD)76.0 ± 18.976.0 ± 18.20.76High protein/low carb diet (%)17.115.00.38Protein supplements (%)0.32.80.02GPAQ total physical activity, MET-min/d (mean ± SD)461.9 ± 817.3573.1 ± 908.80.02BMI, kg/m2 (mean ± SD)31.1 ± 6.729.9 ± 5.70.01Waist to hip ratio (mean ± SD)1 ± 0.10.9 ± 0.1<0.0001Systolic blood pressure, mm Hg (mean ± SD)134.1 ± 21.4123.4 ± 17.3<0.0001Diastolic blood pressure, mm Hg (mean ± SD)75 ± 12.172.5 ± 10.20.007Triglycerides, mg/dL (mean ± SD)147.9 ± 119.3126 ± 87.5<0.0001High density lipoprotein cholesterol, mg/dL (mean ± SD)49.1 ± 1552.2 ± 150.0001Fasting glucose, mg/dL (mean ± SD)138.1 ± 72.1108.3 ± 34.5<0.0001Hypertension medication (%)54.831.8<0.0001Diabetes medication (%)42.816.8<0.0001Lipid-lowering medication (%)16.47.3<0.0001eGFR, ml/min/1.73 m2 (mean ± SD)83.6 ± 28.9102.1 ± 16.3<0.0001Urinary albumin:creatinine ratio (mean ± SD)283.4 ± 792.55.6 ± 5<0.0001aDefined as UAC ratio ≥30 and/or eGFR <60 ml/min/1.73 m2 based on CKD-EPI creatinine-cystatin C equation without race.bP-value from Wilcoxon rank-sum test for continuous variables or Chi-square test for categorical variables.
## Kidney traits and gut microbiome diversity
Kidney traits (eGFR [continuous, binary], UAC ratio [continuous, binary], and CKD) were not related to the Shannon diversity index in multivariable linear regression models adjusted for demographic and microbiome-related factors (Model 1), behavioral and socioeconomic factors (Model 2), and cardiometabolic factors (Model 3) (Supplementary Table 4). Continuous eGFR was significantly associated with overall gut microbiome composition, measured by the Jensen-Shannon Divergence (JSD) and generalized UniFrac distance, in multivariable permutational multivariate analysis of variance (PERMANOVA) models (from Model 3, JSD R-squared = $0.19\%$, $$p \leq 0.001$$; generalized UniFrac R-squared = $0.11\%$, $$p \leq 0.001$$) (Figure 1b; Supplementary Table 5). Other kidney traits (binary eGFR, continuous and binary UAC ratio, and CKD) were not associated with the JSD or generalized UniFrac distance upon full covariate adjustment (all $p \leq 0.15$) (Figure 1b; Supplementary Table 5). Consistently, only eGFR was significantly correlated with the first JSD principal coordinate (Figure 1c), with a visual shift along the first principal coordinate axis for participants with low eGFR (Figure 1d), but not for those with high UAC ratio (Figure 1e). In sensitivity analyses restricting to participants with none or little change in kidney function and damage over the past 6 y, continuous eGFR remained the only significant kidney trait predictor of the JSD (Supplementary Figure S2). Further, continuous eGFR remained a significant predictor of the JSD when excluding participants with low eGFR (<60 ml/min/1.73 m2) (R-squared = $0.14\%$, $$p \leq 0.006$$), suggesting that the finding is not solely driven by advanced CKD.
In stratified analysis of diabetes status, the UAC ratio (binary and continuous) and CKD were associated with significantly lower Shannon diversity index and altered overall gut microbiome composition measured by the JSD, only in participants without diabetes (all p-interaction <0.05) (Supplementary Table 6). While the association of continuous eGFR with overall microbiome composition was only significant in participants with diabetes, the interaction was not significant (p-interaction = 0.15) (Supplementary Table 6).
Based on the stronger relationships of continuous eGFR and UAC ratio with overall microbiome composition over the binary variables (Figure 1b), subsequent analyses focused on continuous eGFR and UAC ratio, as well as CKD.
## Kidney traits and gut microbiome species
Of 1,177 species tested in Analysis of Composition of Microbiomes (ANCOM2), 51, 7, and 13 species were associated with eGFR, UAC ratio, and CKD, respectively, at a detection level ≥0.7 with full covariate adjustment (Figure 2a). Higher eGFR was associated with enrichment of eight species from genus Prevotella, and many species from class Clostridia within genera Eubacterium, Clostridium, Roseburia, and Ruminococcus (Figure 2b; Supplementary Table 7). Higher eGFR was also related to depletion of species in classes Erysipelotrichia, Clostridia, Coriobacteriia, and Fusobacteriia (Figure 2b; Supplementary Table 7). Higher UAC ratio was associated with lower abundance of Clostridium sp. CAG:91, Ruminococcus sp. CAG:254 (both overlapping with eGFR), Haemophilus parainfluenzae, Bacteroides sp. CAG:98, and Phascolarctobacterium sp. CAG:207, and higher abundance of [Clostridium] spiroforme and *Firmicutes bacterium* CAG:94 (Figure 2b; Supplementary Table 8). The majority of species associated with CKD were also related to eGFR and/or UAC ratio (Figure 2b; Supplementary Table 9). Associations of species with kidney traits were similar in a sensitivity analysis excluding participants with low eGFR (Suplementary Figure S3). Figure 2.Gut microbiome species associated with kidney traits in the Hispanic Community Health Study/Study of Latinos ($$n = 2$$,438). ( a) Venn diagram of unique and overlapping species associated with eGFR, UAC ratio, and CKD in ANCOM2 models at a detection level of 0.7 or above. ANCOM2 models were adjusted for age, sex, field center, Hispanic/Latino background, U.S. nativity, antibiotics use, Bristol stool type, income, educational attainment, cigarette smoking, alcohol use, AHEI2010, predicted sodium intake, report of low-sodium diet, predicted protein intake, report of high protein/low carb diet, protein supplement use, total physical activity, BMI, waist-to-hip ratio, systolic blood pressure, diastolic blood pressure, triglycerides, HDL cholesterol, fasting glucose, hypertension medication, diabetes medication, and lipid-lowering medication. ( b) Phylogenetic tree of species associated with eGFR, UAC ratio, and CKD in ANCOM2 models described in (a). Node size reflects mean relative abundance, while node colors for species reflects taxonomic class. Effect size (beta) coefficients from multivariable linear regression of kidney traits on clr-transformed species abundance, adjusting for aforementioned covariates, are displayed in a circular heatmap around the tree. Beta coefficient legend key ranges from−0.015 to 0.015 for eGFR, −0.155 to 0.155 for log UAC ratio, and−0.81 to 0.81 for CKD. * Detection level ≥0.7 in ANCOM2 for a given kidney trait.
In stratified analysis of diabetes status, eGFR was associated with a greater number of species in participants with diabetes compared to those without diabetes, though effect estimates for eGFR-related species were similar for those with and without diabetes, with few significant interactions by diabetes status (Supplementary Figure S4a; Supplementary Table 10). In contrast, more species were associated with UAC ratio and CKD in participants without diabetes compared to those with diabetes, and effect estimates for UAC ratio-related species were not similar for participants with and without diabetes (Supplementary Figure S4a; Supplementary Table 10).
## Kidney traits and gut microbiome functions
Of 660 functional pathways and 1,067 enzymatic reactions tested in ANCOM2, 11 pathways and 20 reactions were associated with eGFR at a detection level ≥0.7 with full covariate adjustment (Figure 3a; Supplementary Tables 11–12). Higher eGFR was associated with higher abundance of GDP-sugar biosynthesis, fatty acid biosynthesis, nitrogen metabolism, glycosaminoglycan degradation, and TCA cycle functions (see Supplementary Table 13 for pathway/reaction ontology). The UAC ratio was associated with lower abundance of iso-bile acid synthesis and polysaccharide degradation functions (Figure 3a; Supplementary Tables 14–15). Lastly, CKD was associated with higher abundance of a D-erythronate degradation reaction (Figure 3a; Supplementary Tables 16–17). While different pathways and reactions were associated with kidney traits in participants with and without diabetes, few significant interactions were observed for these pathways and reactions by diabetes status (Supplementary Figure S4b-c; Supplementary Table 10). Figure 3.Gut microbiome functions associated with kidney traits in the Hispanic Community Health Study/Study of Latinos ($$n = 2$$,438). ( a) We used ANCOM2 to identify MetaCyc pathways and enzymatic reactions for which abundance was associated with eGFR, UAC ratio, or CKD. ANCOM2 models were adjusted for age, sex, field center, Hispanic/Latino background, U.S. nativity, antibiotics use, Bristol stool type, income, educational attainment, cigarette smoking, alcohol use, AHEI2010, predicted sodium intake, report of low-sodium diet, predicted protein intake, report of high protein/low carb diet, protein supplement use, total physical activity, BMI, waist-to-hip ratio, systolic blood pressure, diastolic blood pressure, triglycerides, HDL cholesterol, fasting glucose, hypertension medication, diabetes medication, and lipid-lowering medication. For each pathway/reaction associated with at least one kidney trait in ANCOM2 (detection level ≥0.7), we show the estimated effect size (beta) and $95\%$ confidence interval for eGFR (ml/min/1.73 m2), log UAC ratio, and CKD from multivariable linear regression models, with clr-transformed pathway/reaction abundance as outcomes, adjusting for same covariates as used in ANCOM2. Pathways/reaction names are annotated with colored circles according to their ontology in the MetaCyc database (only for categories with ≥2 constituents). ( b) Spearman correlations of kidney trait-related microbiome scores and clr-transformed pathway/reaction abundance. Scores were derived by Z-score standardizing the clr-transformed abundance of species associated with a given kidney trait, followed by summing/subtracting species that were positively/negatively related to the trait.
Kidney trait-related microbiome scores, derived from species associated with eGFR, UAC ratio, or CKD, were significantly correlated with most kidney-trait related microbiome functions (Figure 3b). The eGFR-related microbiome score was most strongly positively correlated with abundance of aconitate hydratase enzymatic reactions (Figure 3b). Additionally, microbiome functions positively related to eGFR tended to positively correlate with eGFR-enriched species (Supplementary Figure S5). For example, carbamoyl-phosphate synthase and aconitate hydratase had strong positive correlations with eGFR-enriched Prevotella species, and fatty acid biosynthesis reactions were positively correlated with [Eubacterium] rectale (Supplementary Figure S5).
## Kidney trait-related microbiome features and serum metabolites
Abundance of *Clostridium clostridioforme* CAG:132 was correlated with the greatest number of metabolites after FDR adjustment (243 out of 773 metabolites tested) (Figure 4a; Supplementary Table 18). When considering only the strongest correlations (Spearman |r| ≥ 0.3, q < 0.05), 18 metabolites were associated with at least 1 kidney trait-related species (Figure 4b); their peak areas according to binary eGFR, UAC ratio, and CKD status is shown in Supplementary Table 19. *In* general, gut microbiome species related to better kidney health were positively correlated with hydrocinnamate, cinnamoylglycine, indolepropionate, beta-cryptoxanthin, 4-ethylcatechol sulfate, 5alpha-androstan-3beta,17alpha-diol disulfate, 1 H-indole-7-acetic acid, lithocholate sulfate, hippurate, and branched chain 14:0 dicarboxylic acid (Figure 4b). Conversely, species related to worse kidney health were positively correlated with deoxycholic acid metabolites, imidazole propionate, and uremic toxins p-cresol sulfate, p-cresol glucuronide, and phenylacetylglutamine (Figure 4b). Correlations remained similar when further adjusting for age, sex, field center, eGFR, and UAC ratio (Supplementary Figure 6), and correlations also appeared similar in participants with and without diabetes (Supplementary Figure 7). Figure 4.Serum metabolites are associated with kidney-related gut microbiome species ($$n = 700$$). ( a) Number of serum metabolites (out of 773 named metabolites) associated with kidney trait-related species in Spearman correlation analysis at an FDR-adjusted p-value (q-value) <0.05. Species were clr-transformed and metabolites were inverse-normal transformed for analysis. Only species related to kidney traits in ANCOM2 analysis were included. ( b) Spearman correlations of species and kidney trait-related microbiome scores with serum metabolites. Only metabolites that were correlated with at least 1 of these species (q < 0.05) with |r| ≥ 0.3 were included in the heatmap. Species are annotated on the side with taxonomic class, variable(s) they were associated with in ANCOM2, and direction of association with kidney health. Metabolites are annotated on the top with their super-pathway classification, and direction of correlation with eGFR (only if q < 0.05 for correlation with eGFR). * q < 0.05.
Aconitate hydratase and carbamoyl-phosphate synthase enzymatic reaction abundance, related to higher eGFR, were inversely correlated with p-cresol sulfate and p-cresol glucuronide (Supplementary Figure 8a). Other kidney trait-related functional pathways and reactions did not strongly correlate (|r| ≥ 0.3, q < 0.05) with serum metabolites, but metabolites with |r| ≥ 0.2 were similar to those correlated with microbiome species (Supplementary Figure 8b; Supplementary Table 20). Among these, microbial iso-bile acid biosynthesis functions were inversely correlated with deoxycholic acid metabolites (Supplementary Figure 8b).
## Serum metabolites and prospective changes in kidney traits
For serum metabolites strongly correlated (|r| ≥ 0.3, q < 0.05) with kidney trait-related microbiome species (Figure 4b), we examined whether measures of these metabolites at HCHS/SOL visit 1 were associated with change in eGFR and UAC ratio, or with incidence of CKD, from visits 1 to 2 (over ~6 y). Metabolites associated with worsening kidney traits were lithocholate sulfate, 4-ethylcatechol sulfate, and p-cresol glucuronide (associated with reductions in eGFR), ursodeoxycholate and glycoursodeoxycholate (associated with increases in UAC ratio), and imidazole propionate (associated with reductions in eGFR, increases in UAC ratio, and incidence of CKD) (Figure 5). In contrast, hydrocinnamate, beta-cryptoxanthin, 1 H-indole-7-acetic acid, and cinnamoglycine were associated with reductions in UAC ratio over time (Figure 5). For some metabolites, associations differed in magnitude and/or statistical significance by diabetes status. For example, indolepropionate was associated with increases in eGFR only among participants with diabetes, and the association of imidazole propionate with worsening kidney traits was of greater magnitude in participants with diabetes (Figure 5). Findings were similar in the subset of participants randomly selected for metabolomics measurement (i.e., excluding participants selected based on kidney function decline), though some findings were attenuated, possibly due to reduced power (Supplementary Figure 9). Figure 5.Serum metabolites associated with kidney-related gut microbiome species are predictors of kidney trait progression ($$n = 3$$,635). Prospective association of serum metabolites (selected from Figure 4b) with eGFR and UAC ratio progression, and incident CKD. For the continuous outcomes of eGFR and UAC ratio, multivariable linear mixed-effects regression models were used to estimate the effect of inverse-normal transformed metabolites at HCHS/SOL visit 1 on eGFR and UAC ratio progression from HCHS/SOL visit 1 to visit 2, adjusting for the following visit 1 covariates: age, sex, field center, Hispanic/Latino background, U.S. nativity, income, educational attainment, cigarette smoking, alcohol use, AHEI2010, predicted sodium intake, predicted protein intake, protein supplement use, total physical activity, BMI, waist-to-hip ratio, systolic blood pressure, diastolic blood pressure, triglycerides, HDL cholesterol, fasting glucose, hypertension medication, diabetes medication, and lipid-lowering medication. Betas are from the interaction of time x metabolite. For the binary outcome of incident CKD, multivariable logistic regression was used to estimate the effect of inverse-normal transformed metabolites at HCHS/SOL visit 1 on incidence of CKD at visit 2, adjusting for the visit 1 covariates listed above. Analyses were performed in all available participants ($$n = 3$$,635), non-diabetics ($$n = 2$$,979), and diabetics ($$n = 656$$).
We also explored whether change in these serum metabolites over time (from visits 1 to 2) was associated with change in eGFR or UAC ratio over the same time period. We observed that increases in imidazole propionate and phenylacetylglutamine over time were associated with significant reductions in eGFR over time ($p \leq 0.05$), while changes in metabolites were not significantly related to change in UAC ratio over time (Supplementary Table 21).
## Discussion
In this large study of Hispanic/Latino adults, kidney function (eGFR) was significantly associated with overall gut microbiome composition and a wide range of species, while kidney damage (UAC ratio) was associated with lower gut microbiome diversity and altered composition only in participants without diabetes, suggesting that people with and without diabetes may have differences in gut microbiota involvement in kidney damage. Serum metabolites associated with kidney-related gut microbiome features (e.g., p-cresol glucuronide, imidazole propionate) were prospectively linked to changes in kidney function and damage over ~6 y, indicating that gut microbiota may play a role in CKD progression.
A number of previous studies observed depletion of taxa from Prevotella, Faecalibacterium, Roseburia, Coprococcus, and Eubacterium in patients with CKD compared to healthy controls16–26, similar to our findings of these taxa being related to higher eGFR. Many of these taxa are known SCFA producers, though we did not have measures of serum or stool SCFAs in this study to confirm their role. We also observed that microbial polysaccharide degradation functions (e.g., mannobiose and cellobiose 2-epimerase) were inversely related to the UAC ratio. These findings agree with evidence from other studies that SCFAs are depleted in stool and serum of CKD patients compared to healthy controls27, and that SCFAs ameliorate kidney injury28. SCFAs may protect the kidneys via several mechanisms, including reducing inflammation, oxidative stress, and pro-fibrotic factors in kidney cells13. Microbial functions related to synthesis of long-chain fatty acids, such as palmitate and oleate, were associated with higher eGFR. This was unexpected, since long-chain fatty acids accumulate in CKD and may induce renal injury29. We also found microbial ammonia-dependent carbamoyl-phosphate synthase was associated with higher eGFR, suggesting that ammonia utilization by gut bacteria30 may be beneficial for kidney health.
Our results of serum metabolites associated with kidney-trait related species suggest other microbial products which may preserve kidney function and/or reflect better kidney health. Diet-derived compounds, including hydrocinnamate and cinnamoglycine (from cinnamon) and beta-cryptoxanthin (from fruit), were positively correlated with species related to better kidney health, which may reflect the protective effect of healthy diet on incidence and progression of CKD31,32. In support of this, dietary fiber intake assessed at HCHS/SOL visit 1 was positively correlated with such diet-derived metabolites at visit 2, as well as with eGFR and species related to better kidney health (Supplementary Table 22). These metabolites were prospectively associated with reductions in the UAC ratio over time, suggesting they may prevent kidney damage or promote kidney structural repair. Indolepropionate, a bacterial product of tryptophan metabolism, was also positively correlated with species related to better kidney health. Indolepropionate was related to lower incidence of diabetes in a large multi-cohort analysis33 and also found to be lower in CKD patients vs. controls and associated with slower declines in kidney function34. We observed that indolepropionate was associated with increases in eGFR over time only among those with diabetes, indicating it may be important in diabetic kidney disease prevention.
In contrast, some microbial products may further CKD progression. Serum imidazole propionate, which was positively associated with species related to lower eGFR, is a microbial metabolite elevated in diabetes, shown to impair insulin signaling35. Consistently, we found that imidazole propionate was associated with prospective declines in eGFR and increases in UAC ratio, most prominently among participants with diabetes. Serum levels of microbially produced secondary bile acid deoxycholic acid have been associated with prospective CKD progression36 and coronary artery calcification in CKD37. We observed that species related to lower eGFR, including *Flavonifractor plautii* and Eggerthella lenta, were positively correlated with serum deoxycholic acid metabolites, while iso-bile acid biosynthesis functions (related to lower UAC ratio) were inversely correlated with deoxycholic acid metabolites. The iso-bile acid pathway is an oxidative pathway for bile acids that can detoxify deoxycholic acid38. Further, we found that deoxycholic acid metabolites were prospectively associated with increases in UAC ratio over time. Taken together, these results suggest that gut bacteria may promote kidney damage through synthesis of deoxycholic acid and/or insufficient conversion of deoxycholic acid to iso-bile acids. In agreement with these observations, other studies have observed enrichment of *Flavonifractor plautii* and *Eggerthella lenta* in CKD23, as well as enrichment of microbial secondary bile acid pathways in CKD23,24.
Microbially produced protein-bound uremic toxins are hypothesized as the primary mechanism for microbial involvement in CKD progression and morbidity. These include indoles (e.g., indoxyl sulfate), phenols (e.g., p-cresol sulfate, p-cresol glucuronide, phenylacetylglutamine), and hippurate39,40. Some studies found greater abundance of p-cresol and indoxyl sulfate-producing gut bacteria in CKD patients11,23, though one study suggests that the rate of bacterial generation of these toxins is not influenced by kidney function41. In our study, serum p-cresol sulfate, p-cresol glucuronide, and phenylacetylglutamine were positively correlated with some species related to worse kidney health, such as Eisenenbergiella tayi and Sellimonas intestinalis. Similar correlations, albeit slightly weaker, were observed for indoxyl sulfate (Supplementary Table 18), but serum hippurate was positively correlated with species related to better kidney health. In our prospective analysis, p-cresol glucuronide was associated with reductions in eGFR over time, confirming potential involvement of microbially produced uremic toxins in kidney disease.
Interestingly, despite the hypothesized contribution of uremia to gut microbiome dysbiosis in CKD8,9, in our study serum urea was only weakly associated with a few kidney trait-related species (Supplementary Table 18), suggesting that urea is not a primary driver of our findings. This may be because cases of CKD in our study were mostly mild CKD, thus serum urea is unlikely to be elevated enough to indicate uremia (though we did not have absolute measures of blood urea nitrogen in this study).
Associations of kidney traits with the gut microbiome differed by diabetes status. In participants without diabetes, higher UAC ratio was associated with reduced gut microbiome diversity and altered overall composition, while eGFR was not. In participants with diabetes, eGFR was a significant predictor of gut microbiome overall composition, while UAC ratio was not. While eGFR-microbiome relationships did not have statistically significant heterogeneity by diabetes status, the interaction of UAC ratio–microbiome relationships was significant. Diabetic and non-diabetic kidney disease develops with different pathological mechanisms. In diabetic kidney disease, or diabetic nephropathy, diabetes is the sole cause, with hyperglycemia being the central upstream driver15. Non-diabetic kidney disease can arise from hereditary or acquired causes, including poor nephron endowment, obesity, pregnancy, and injury- or aging-related nephron loss15. Without renal biopsy, etiology of CKD in people with diabetes (i.e., diabetic nephropathy or CKD coincident with diabetes) cannot be determined, as is the case in our study. Nevertheless, hyperglycemia is likely a contributing factor to CKD in people with diabetes regardless of etiology15. A number of studies have examined differences in the gut microbiome for people with diabetic kidney disease, diabetes without kidney disease, and healthy controls42, but little is known regarding gut microbiome associations with diabetic vs. non-diabetic kidney disease. Since we observed kidney damage was only related to the gut microbiome in participants without diabetes, this may suggest kidney damage in people with diabetes is attributed primarily to hyperglycemia and less to other factors (i.e., in a “sufficient causes” framework43, hyperglycemia may be a sufficient cause of kidney damage, precluding observation of an association with the gut microbiome). In accordance, fasting glucose was significantly correlated with the UAC ratio in participants with diabetes (Spearman $r = 0.24$, $p \leq 0.0001$), but not without diabetes ($r = 0.00$, $$p \leq 0.87$$). Alternatively, since sample size was lower in participants with diabetes, we may have had insufficient power to observe a relationship of the gut microbiome with kidney damage. Additional studies are needed to validate gut microbiome and kidney relationships in people with and without diabetes.
Our study was strengthened by large sample size, thorough control for potential confounders, analysis of continuous eGFR and UAC ratio across a wide continuum, and prospective analysis of microbiome-related metabolites with changes in kidney health. Our study also faced several limitations. CKD is usually defined with sustained low eGFR or albuminuria over a period of 3 months, but we relied on one-time measures. Our analysis of the gut microbiome and kidney traits was cross-sectional, limiting temporal inferences, though we did conduct a prospective analysis of microbiome-related metabolites with changes in kidney traits. We did not have measures of SCFAs to correlate with kidney-related species and functions. Finally, our study was performed solely in U.S. Hispanics/Latinos, which may limit generalizability.
In summary, our results suggest that the gut microbiome associated with better kidney health is characterized by higher abundance of Prevotella and Clostridia species known to produce SCFAs, while gut microbiota associated with poor kidney health may be involved in production of imidazole propionate, deoxycholic acid metabolites, and uremic toxins. Imidazole propionate, deoxycholic acid metabolites, and p-cresol glucuronide were associated with prospective reductions in eGFR and/or increases in the UAC ratio over ~6 y, suggesting their involvement in CKD progression and morbidity. Large prospective studies of the gut microbiome, microbial metabolites, and CKD progression are warranted to validate these results, followed by experimental models to support a causal relationship of the gut microbiome with CKD. Given the modifiable nature of the gut microbiome, there is promising therapeutic potential in altering the gut microbiome to prevent CKD progression40.
## Study cohort
The Hispanic Community Health Study/Study of Latinos (HCHS/SOL) is a prospective, population-based cohort study of 16,415 Hispanic/Latino adults (ages 18–74 y at the time of recruitment during visit 1 [2008–2011]) who were selected using a multi-stage probability sampling design from randomly sampled census block areas within four U.S. communities (Chicago, IL; Miami, FL; Bronx, NY; San Diego, CA)44,45. The HCHS/SOL Gut Origins of Latino Diabetes ancillary study46 was conducted to examine the role of gut microbiome composition on diabetes and other outcomes, enrolling ~3,000 participants from the HCHS/SOL approximately concurrent with the second in-person HCHS/SOL visit cycle (visit 2, 2014–2017). For the cross-sectional analyses at visit 2 (Figure 1a), we excluded participants with prevalent cancer or CVD at visit 2, currently on kidney dialysis at visit 2, missing measures of serum creatinine, cystatin C, or UAC ratio at visit 2, or with <100K sequence reads in their microbiome sample (Supplementary Figure S1). For the prospective analysis from visits 1 to 2 (Figure 1a), we excluded participants with prevalent cancer, CVD, or CKD at visit 1, currently on kidney dialysis at visit 1, or missing measures of serum creatinine, cystatin C, or UAC ratio at visits 1 or 2 (Supplementary Figure S1). The study was conducted with the approval of the Institutional Review Boards (IRBs) of the five participating universities in HCHS/SOL. Written informed consent was provided by all study participants.
## Kidney trait definitions
Estimated glomerular filtration rate (eGFR) was calculated from serum creatinine and cystatin C using the new CKD-EPI creatinine-cystatin C equation without race47, as recommended by the National Kidney Foundation (NKF) and the American Society of Nephrology48. eGFR was considered as a continuous variable and a binary variable, with eGFR <60 ml/min/1.73 m2 indicating low eGFR. Kidney damage was assessed using the urinary albumin:creatinine (UAC) ratio, which was also considered as a continuous variable and a binary variable, with UAC ratio ≥30 indicating albuminuria. Chronic kidney disease (CKD) was defined as UAC ratio ≥30 and/or eGFR <60 ml/min/1.73 m2. Stages of CKD were defined using the standard NKF definition: Stage 1, eGFR ≥90 ml/min/1.73 m2 and albuminuria; Stage 2, eGFR 60–89 ml/min/1.73 m2 and albuminuria; Stage 3, eGFR 30–59 ml/min/1.73 m2; Stage 4, eGFR 15–29 ml/min/1.73 m2; Stage 5, eGFR <15 ml/min/1.73 m2.
## Assessment of cardiometabolic traits
Using an automatic sphygmomanometer, three seated blood pressure measures were obtained for each participant after a 5-minute rest period and means of the second and third measurement were used to derive systolic blood pressure and diastolic blood pressure49. Centralized laboratory tests included blood glucose, insulin, hemoglobin A1c (HbA1c), triglycerides, and total, high-density lipoprotein (HDL), and low-density lipoprotein (LDL) cholesterol, all measured after overnight fast50. Diabetes was defined based on meeting the American Diabetes Association lab criteria for diabetes (fasting glucose, glucose post-oral glucose tolerance test, or HbA1c) or self-report of anti-diabetic medication.
## Covariate data
Participant characteristics were included for statistical adjustment in our analysis, based on known or suspected relationships with CKD and/or the gut microbiome. These variables were age (continuous), sex (male, female), field center (Chicago, Miami, Bronx, San Diego), Hispanic/Latino background (Dominican, Central American, South American, Cuban, Mexican, Puerto Rican, more than one heritage/other/missing), U.S. nativity (born in 50 U.S. states/DC or a U.S. territory, foreign born), antibiotic use in last 6 months (yes, no), Bristol stool type (8 categories), income (<$30,000, ≥30,000, missing), educational attainment (less than high school, high school or equivalent, greater than high school, missing), cigarette smoking (never, former, current), alcohol use (nondrinker, low-level use, high-level use), the Alternative Healthy Eating Index 2010 (AHEI2010; continuous), predicted sodium intake based on the NCI method (continuous), report of low-sodium diet (yes, no), predicted protein intake based on the NCI method (continuous), report of high protein/low carbohydrate diet (yes, no), protein supplement use (yes, no), total physical activity based on the Global Physical Activity Questionnaire (GPAQ; continuous), BMI (continuous), waist-to-hip ratio (continuous), systolic blood pressure (continuous), diastolic blood pressure (continuous), triglycerides (continuous), HDL cholesterol (continuous), fasting glucose (continuous), anti-hypertensive medication (yes, no), anti-diabetic medication (yes, no), lipid-lowering medication (yes, no). For the cross-sectional analyses at visit 2, all covariates were based on visit 2 data, except AHEI10, predicted sodium intake, and predicted protein intake which were based on visit 1 data (since dietary recalls were only collected at visit 1). For the prospective analysis from visits 1 to 2, all covariates were based on visit 1 data. Missing covariate data were imputed at the median and mode for continuous and categorical variables, respectively, with the exception of categorical variables with >$1\%$ missing, for which a missing category was created.
## Microbiome measurement
Stool samples were collected by participants at home using stool collection kits, as described previously46. Shallow shotgun sequencing was conducted in the Knight laboratory at the University of California San Diego51, as previously described in HCHS/SOL52. Briefly, DNA was extracted from fecal samples following the Earth Microbiome Project protocol53. Adapters and barcode indices were added following the iTru adapter protocol54, and the resulting libraries were purified, quantified, and normalized for sequencing on Illumina NovaSeq.
## Microbiome bioinformatics processing
FASTQ sequence reads were processed using the standard shotgun sequencing pipeline in Qiita. Briefly, per sample sequence adapters were removed via fastp, and sequence reads mapping to the human genome were filtered via minimap2. The sequence reads were then aligned against the WolR1 reference database of bacterial and archaeal genomes using Woltka with the Bowtie2 aligner55, to generate an operational genomic unit (OGU) table and a gene table. The sequence alignments were also classified at the species taxonomic rank, while functional profiles were obtained by collapsing the gene table into MetaCyc enzymatic reactions and functional pathways. Indices of α-diversity (Shannon diversity index) and β-diversity (Jensen-Shannon Divergence, generalized UniFrac) were calculated from the OGU table using “vegan,” “phyloseq,” and “GUniFrac” packages in R56–58.
## Metabolomics measurement
A subset of 825 participants had available metabolomics profiling of visit 2 fasting serum samples; these participants were selected for metabolomics profiling based on availability of gut microbiome samples collected within 30 d of visit 2. After restricting by our exclusion criteria (Suplementary Figure S1), 700 participants remained for cross-sectional metabolomics analysis at visit 2. Additionally, a subset of 6,180 participants had available metabolomics profiling of visit 1 fasting serum samples; 3,978 were a random subsample, while the remainder were selected based on participation in the echocardiographic ancillary study of HCHS/SOL or based on decline in eGFR from visits 1 to 2 (the random subsample and pre-selected samples were measured in different batches, “batch 1” and “batch 2,” which are pooled here). After restricting by our exclusion criteria (Suplementary Figure S1), 3,635 participants remained for prospective analysis of visit 1 metabolomics data. Using the discoveryHD4 platform at Metabolon Inc., quantification of serum metabolites was achieved by using an untargeted LC-MS-based metabolomic quantification protocol, as previously described59. We imputed values below detection as half the minimum value per metabolite.
## General principles
An overview of the analyses is provided in Figure 1a. The kidney traits considered in the statistical analyses were eGFR (continuous or binary), UAC ratio (continuous or binary), and CKD (binary). The continuous UAC ratio was log-transformed for all analyses due to a heavy right skew. All analyses were conducted in R version 3.6.3.
## Cross-sectional analyses of the gut microbiome and kidney traits
We considered nested models to serially adjust for potential confounders, based on known factors related to CKD, and to the gut microbiome in this cohort46. Model 1 (demographic and microbiome-related factor model) adjusted for age, sex, field center, Hispanic/Latino background, U.S. nativity, antibiotics use, and Bristol stool type. Model 2 (socioeconomic and behavioral model) adjusted for Model 1 covariates plus income, educational attainment, cigarette smoking, alcohol use, AHEI2010, predicted sodium intake, report of low-sodium diet, predicted protein intake, report of high protein/low carbohydrate diet, protein supplement use, and total physical activity. Model 3 (cardiometabolic model) adjusted for Model 2 covariates plus BMI, waist-to-hip ratio, systolic blood pressure, diastolic blood pressure, triglycerides, HDL cholesterol, fasting glucose, hypertension medication, diabetes medication, and lipid-lowering medication. Within-subject (α-) and between-subject (β-) diversity. Multivariable linear regression was used to examine the association of kidney traits (predictors) with the Shannon diversity index (outcome), adjusting for covariates. Permutational multivariate analysis of variance (PERMANOVA) was used to assess the association of kidney traits with overall microbiome composition, as measured by the Jensen-Shannon Divergence and generalized UniFrac distance, adjusting for covariates. A p-value <0.05 was considered significant in diversity analyses. Species and metagenomic pathways/enzymatic reactions. Microbial species and MetaCyc functional pathways and enzymatic reactions were analyzed in two stages: first using the Analysis of Composition of Microbiomes (ANCOM2) method60, followed by confirmatory multivariable linear regression, described below. ANCOM2 was used to detect species, pathways, and reactions for which abundance was related to kidney traits, adjusting for covariates. We controlled the false discovery rate (FDR) at $5\%$ and excluded species, pathways, or reactions from testing if they were present in <$20\%$ of the participants. An ANCOM2 detection level ≥0.7 was considered significant – this level indicates that the ratios of the species, pathway, or reaction to at least $70\%$ of other species, pathways, or reactions were detected to be significantly associated (FDR q < 0.05) with a kidney trait. To assess the direction and magnitude of the associations, we constructed multivariable linear regression models, with centered log ratio (clr)-transformed species/pathway/reaction abundance as outcomes, and kidney traits as the main predictors, adjusting for covariates. Kidney trait-related microbiome scores. We developed kidney trait-related microbiome scores based on species associated with eGFR, UAC ratio, or CKD, to relate with gut microbiome functional pathways/enzymatic reactions and serum metabolites. First, clr-transformed abundance of species associated with a given kidney trait in ANCOM2 (detection level ≥0.7) was Z-score standardized to give equal weight to each species; then, those species positively related to the kidney trait were summed while species negatively related to the kidney trait were subtracted within each participant to derive the score.
## Cross-sectional analysis of serum metabolites and kidney-related microbiome features
Metabolite concentrations were inverse-normal transformed for analysis. Only named metabolites with <$20\%$ missing were considered, totaling 773 metabolites. We examined unadjusted and partial Spearman correlations (adjusting for age, sex, field center, eGFR, and UAC ratio) of metabolites with kidney-related gut microbiome species, functional pathways, and enzymatic reaction clr-transformed abundance. We similarly examined correlations of serum metabolites with kidney-trait related microbiome scores. For each microbiome feature, we considered a metabolite with FDR-adjusted p-value (q-value) <0.05 as a significantly correlated metabolite.
## Prospective analysis of serum metabolites and kidney trait progression
For specific metabolites strongly correlated (Spearman |r| ≥ 0.3) with kidney trait-related microbiome species, we examined associations with eGFR and UAC ratio progression and incident CKD. For the continuous outcomes of eGFR and UAC ratio, multivariable linear mixed-effects regression models with a random intercept were used to estimate the effect of inverse-normal transformed metabolites at visit 1 on eGFR and UAC ratio progression from visit 1 to visit 2, adjusting for the following visit 1 covariates: age, sex, field center, Hispanic/Latino background, U.S. nativity, income, educational attainment, cigarette smoking, alcohol use, AHEI2010, predicted sodium intake, predicted protein intake, protein supplement use, total physical activity, BMI, waist-to-hip ratio, systolic blood pressure, diastolic blood pressure, triglycerides, HDL cholesterol, fasting glucose, hypertension medication, diabetes medication, and lipid-lowering medication. The effect of interest was the interaction of time × metabolite. For the binary outcome of incident CKD, multivariable logistic regression was used to estimate the effect of inverse-normal transformed metabolites at visit 1 on incidence of CKD at visit 2, adjusting for the visit 1 covariates listed above. For this analysis based on pre-specified predictors and outcomes, $p \leq 0.05$ was considered significant.
## Sensitivity and stratified analyses
We conducted several sensitivity analyses to confirm our findings. For the cross-sectional analysis of kidney traits and gut microbiome β-diversity, we performed a sensitivity analysis restricting to participants with either no change in binary eGFR, UAC ratio, and CKD status ($$n = 2$$,042), or <$10\%$ change in eGFR ($$n = 1$$,614), from HCHS/SOL visits 1 to 2 (approximately 6 y apart) – this was to confirm that findings remained similar in participants who had consistent kidney function over the past 6 y. Additionally, we performed a sensitivity analysis excluding participants with low eGFR (<60 ml/min/1.73 m2), to confirm that findings were not solely driven by those with advanced stages of CKD. Lastly, we performed stratified analyses to determine whether associations of the gut microbiome and kidney traits differ by diabetes status and tested for interaction of diabetes and kidney traits on the gut microbiome using cross-product terms.
## List of abbreviations
AHEI2010, Alternative Health Eating Index 2010; ANCOM, Analysis of Composition of Microbiomes; CKD, chronic kidney disease; CLR, centered log-ratio; CVD, cardiovascular disease; eGFR, estimated glomerular filtration rate; FDR, false discovery rate; HCHS/SOL, Hispanic Community Health Study/Study of Latinos; HDL, high-density lipoprotein; JSD, Jensen-Shannon Divergence; PERMANOVA, permutational multivariate analysis of variance; SCFA, short-chain fatty acid; UAC ratio, urinary albumin-to-creatinine ratio.
## Disclosure statement
No potential conflict of interest was reported by the author(s).
## Data availability statement
HCHS/SOL data are archived at the National Institutes of Health repositories dbGap and BIOLINCC. Sequence data from the samples described in this study is deposited in QIITA (study ID 11666). HCHS/SOL has established a process for the scientific community to apply for access to participant data and materials, including the metabolomics data used herein, with such requests reviewed by the project’s Steering Committee. These policies are described at https://sites.cscc.unc.edu/hchs/.
## Supplementary material
Supplemental data for this article can be accessed online at https://doi.org/$\frac{10.1080}{19490976.2023.2186685}$
## References
1. **Chapter 1: cKD in the general population**. *Am J Kidney Dis* (2019.0) **73** S1-18. DOI: 10.1053/j.ajkd.2018.12.001
2. **Chapter 3: morbidity and mortality in patients with CKD**. *Am J Kidney Dis* (2019.0) **73** S45-78. DOI: 10.1053/j.ajkd.2018.12.003
3. **Chapter 4: cardiovascular Disease in Patients with CKD**. *Am J Kidney Dis* (2019.0) **73** S79-98. DOI: 10.1053/j.ajkd.2018.12.004
4. Wyld MLR, Chadban SJ, Morton RL.. **Improving our understanding of quality of life in CKD**. *Am J Kidney Dis* (2016.0) **67** 820-821. DOI: 10.1053/j.ajkd.2016.03.412
5. Levey AS, Coresh J, Balk E, Kausz AT, Levin A, Steffes MW, Hogg RJ, Perrone RD, Lau J, Eknoyan G. **National kidney foundation practice guidelines for chronic kidney disease: evaluation, classification, and stratification**. *Ann Intern Med* (2003.0) **139** 137-147. DOI: 10.7326/0003-4819-139-2-200307150-00013
6. Zhao J, Ning X, Liu B, Dong R, Bai M, Sun S. **Specific alterations in gut microbiota in patients with chronic kidney disease: an updated systematic review**. *Ren Fail* (2021.0) **43** 102-112. DOI: 10.1080/0886022X.2020.1864404
7. Cosola C, Rocchetti MT, Sabatino A, Fiaccadori E, Di Iorio BR, Gesualdo L. **Microbiota issue in CKD: how promising are gut-targeted approaches?**. *J Nephrol* (2019.0) **32** 27-37. DOI: 10.1007/s40620-018-0516-0
8. Ramezani A, Raj DS. **The gut microbiome, kidney disease, and targeted interventions**. *J Am Soc Nephrol* (2014.0) **25** 657-670. DOI: 10.1681/ASN.2013080905
9. Hobby GP, Karaduta O, Dusio GF, Singh M, Zybailov BL, Arthur JM. **Chronic kidney disease and the gut microbiome**. *Am J Physiol Renal Physiol* (2019.0) **316** F1211-7. DOI: 10.1152/ajprenal.00298.2018
10. Kikuchi K, Saigusa D, Kanemitsu Y, Matsumoto Y, Thanai P, Suzuki N, Mise K, Yamaguchi H, Nakamura T, Asaji K. **Gut microbiome-derived phenyl sulfate contributes to albuminuria in diabetic kidney disease**. *Nat Commun* (2019.0) **10** 1835. DOI: 10.1038/s41467-019-09735-4
11. Wong J, Piceno YM, DeSantis TZ, Pahl M, Andersen GL, Vaziri ND. **Expansion of urease- and uricase-containing, indole- and p-cresol-forming and contraction of short-chain fatty acid-producing intestinal microbiota in ESRD**. *Am J Nephrol* (2014.0) **39** 230-237. DOI: 10.1159/000360010
12. Meijers B, Evenepoel P, Anders HJ. **Intestinal microbiome and fitness in kidney disease**. *Nat Rev Nephrol* (2019.0) **15** 531-545. DOI: 10.1038/s41581-019-0172-1
13. Heaney LM, Davies OG, Selby NM. **Gut microbial metabolites as mediators of renal disease: do short-chain fatty acids offer some hope?**. *Future Sci OA* (2019.0) **5** FSO384-FSO. DOI: 10.4155/fsoa-2019-0013
14. Caggiano G, Cosola C, Di Leo V, Gesualdo M, Gesualdo L. **Microbiome modulation to correct uremic toxins and to preserve kidney functions**. *Curr Opin Nephrol Hypertens* (2020.0) **29** 49-56. DOI: 10.1097/MNH.0000000000000565
15. Anders H-J, Huber TB, Isermann B, Schiffer M. **CKD in diabetes: diabetic kidney disease versus nondiabetic kidney disease**. *Nat Rev Nephrol* (2018.0) **14** 361-377. DOI: 10.1038/s41581-018-0001-y
16. Gryp T, Huys GRB, Joossens M, Van Biesen W, Glorieux G, Vaneechoutte M. **Isolation and quantification of uremic toxin precursor-generating gut bacteria in chronic kidney disease patients**. *Int J Mol Sci* (2020.0) **21** 21. DOI: 10.3390/ijms21061986
17. Hu Q, Wu K, Pan W, Zeng Y, Hu K, Chen D, Huang X, Zhang Q. **Intestinal flora alterations in patients with early chronic kidney disease: a case-control study among the Han population in southwestern China**. *J Int Med Res* (2020.0) **48** 300060520926033. DOI: 10.1177/0300060520926033
18. Jiang S, Xie S, Lv D, Wang P, He H, Zhang T, Zhou Y, Lin Q, Zhou H, Jiang J. **Alteration of the gut microbiota in Chinese population with chronic kidney disease**. *Sci Rep* (2017.0) **7** 2870. DOI: 10.1038/s41598-017-02989-2
19. Jiang S, Xie S, Lv D, Zhang Y, Deng J, Zeng L, Chen Y. **A reduction in the butyrate producing species Roseburia spp. and faecalibacterium prausnitzii is associated with chronic kidney disease progression**. *Antonie Van Leeuwenhoek* (2016.0) **109** 1389-1396. DOI: 10.1007/s10482-016-0737-y
20. Lun H, Yang W, Zhao S, Jiang M, Xu M, Liu F, Wang Y. **Altered gut microbiota and microbial biomarkers associated with chronic kidney disease**. *Microbiologyopen* (2019.0) **8** e00678. DOI: 10.1002/mbo3.678
21. Ren Z, Fan Y, Li A, Shen Q, Wu J, Ren L. *Alterations of the human gut microbiome in chronic kidney disease* (2020.0) **7** 2001936
22. Sato N, Kakuta M, Hasegawa T, Yamaguchi R, Uchino E, Murashita K, Nakaji S, Imoto S, Yanagita M, Okuno Y. **Metagenomic profiling of gut microbiome in early chronic kidney disease**. *Nephrol Dial Transplant* (2021.0) **36** 1675-1684. DOI: 10.1093/ndt/gfaa122
23. Wang X, Yang S, Li S, Zhao L, Hao Y, Qin J, Zhang L, Zhang C, Bian W, Zuo L. **Aberrant gut microbiota alters host metabolome and impacts renal failure in humans and rodents**. *Gut* (2020.0) **69** 2131-2142. DOI: 10.1136/gutjnl-2019-319766
24. Wu IW, Gao SS, Chou HC, Yang HY, Chang LC, Kuo YL, Dinh MCV, Chung W-H, Yang C-W, Lai H-C. **Integrative metagenomic and metabolomic analyses reveal severity-specific signatures of gut microbiota in chronic kidney disease**. *Theranostics* (2020.0) **10** 5398-5411. DOI: 10.7150/thno.41725
25. K-Y X, Xia G-H, J-Q L, Chen M-X, Zhen X, Wang S, You C, Nie J, Zhou H-W, Yin J. **Impaired renal function and dysbiosis of gut microbiota contribute to increased trimethylamine-N-oxide in chronic kidney disease patients**. *Sci Rep* (2017.0) **7** 1445. DOI: 10.1038/s41598-017-01387-y
26. Li Y, Su X, Zhang L, Liu Y, Shi M, Lv C, Gao Y, Xu D, Wang Z. **Dysbiosis of the gut microbiome is associated with CKD5 and correlated with clinical indices of the disease: a case–controlled study**. *J Transl Med* (2019.0) **17** 228. DOI: 10.1186/s12967-019-1969-1
27. Wang S, Lv D, Jiang S, Jiang J, Liang M, Hou F, Chen Y. **Quantitative reduction in short-chain fatty acids, especially butyrate, contributes to the progression of chronic kidney disease**. *Clin Sci (Lond)* (2019.0) **133** 1857-1870. DOI: 10.1042/CS20190171
28. Andrade-Oliveira V, Amano MT, Correa-Costa M, Castoldi A, Felizardo RJ, de Almeida DC, Bassi EJ, Moraes-Vieira PM, Hiyane MI, Rodas ACD. **Gut bacteria products prevent AKI induced by ischemia-reperfusion**. *J Am Soc Nephrol* (2015.0) **26** 1877-1888. DOI: 10.1681/ASN.2014030288
29. Gai Z, Wang T, Visentin M, Kullak-Ublick GA, Fu X, Wang Z. **Lipid accumulation and chronic kidney disease**. *Nutrients* (2019.0) **11** 11. DOI: 10.3390/nu11040722
30. Shi D, Caldovic L, Tuchman M. **Sources and fates of carbamyl phosphate: a labile energy-rich molecule with multiple facets**. *Biology* (2018.0) **7** 34. DOI: 10.3390/biology7020034
31. Cai Q, Dekker LH, Vinke PC, Corpeleijn E, Bakker SJL, de Borst MH, Navis GJ. **Diet quality and incident chronic kidney disease in the general population: the lifelines cohort study**. *Clin Nutr* (2021.0) **40** 5099-5105. DOI: 10.1016/j.clnu.2021.07.033
32. Hu EA, Coresh J, Anderson CAM, Appel LJ, Grams ME, Crews DC, Mills KT, He J, Scialla J, Rahman M. **Adherence to healthy dietary patterns and risk of CKD progression and all-cause mortality: findings from the CRIC (Chronic Renal Insufficiency Cohort) study**. *Am J Kidney Dis* (2021.0) **77** 235-244. DOI: 10.1053/j.ajkd.2020.04.019
33. Qi Q, Li J, Yu B, Moon JY, Chai JC, Merino J, Hu J, Ruiz-Canela M, Rebholz C, Wang Z. **Host and gut microbial tryptophan metabolism and type 2 diabetes: an integrative analysis of host genetics, diet, gut microbiome and circulating metabolites in cohort studies**. *Gut* (2022.0) **71** 1095-1105. DOI: 10.1136/gutjnl-2021-324053
34. Sun CY, Lin CJ, Pan HC, Lee CC, Lu SC, Hsieh YT, Huang S-Y, Huang H-Y. **Clinical association between the metabolite of healthy gut microbiota, 3-indolepropionic acid and chronic kidney disease**. *Clin Nutr* (2019.0) **38** 2945-2948. DOI: 10.1016/j.clnu.2018.11.029
35. Koh A, Molinaro A, Ståhlman M, Khan MT, Schmidt C, Mannerås-Holm L, Wu H, Carreras A, Jeong H, Olofsson LE. **Microbially produced imidazole propionate impairs insulin signaling through mTORC1**. *Cell* (2018.0) **175** 947-61.e17. DOI: 10.1016/j.cell.2018.09.055
36. Frazier R, Cai X, Lee J, Bundy JD, Jovanovich A, Chen J, Deo R, Lash JP, Anderson AH, Go AS. **Deoxycholic acid and risks of cardiovascular events, ESKD, and mortality in CKD: the CRIC study**. *Kidney Medicine* (2022.0) **4** 100387. DOI: 10.1016/j.xkme.2021.09.004
37. Jovanovich A, Isakova T, Block G, Stubbs J, Smits G, Chonchol M, Miyazaki M. **Deoxycholic acid, a metabolite of circulating bile acids, and coronary artery vascular calcification in CKD**. *Am J Kidney Dis* (2018.0) **71** 27-34. DOI: 10.1053/j.ajkd.2017.06.017
38. Devlin AS, Fischbach MA. **A biosynthetic pathway for a prominent class of microbiota-derived bile acids**. *Nat Chem Biol* (2015.0) **11** 685-690. DOI: 10.1038/nchembio.1864
39. Graboski AL, Redinbo MR. **Gut-derived protein-bound uremic toxins**. *toxins (Basel)* (2020.0) **12** 590. DOI: 10.3390/toxins12090590
40. ramezani a, massy za, meijers b, evenepoel p, vanholder r, raj ds. **role of the gut microbiome in uremia: a potential therapeutic target**. *Am J Kidney Dis* (2016.0) **67** 483-498. DOI: 10.1053/j.ajkd.2015.09.027
41. Gryp T, De Paepe K, Vanholder R, Kerckhof FM, Van Biesen W, Van de Wiele T, Verbeke F, Speeckaert M, Joossens M, Couttenye MM. **Gut microbiota generation of protein-bound uremic toxins and related metabolites is not altered at different stages of chronic kidney disease**. *Kidney Int* (2020.0) **97** 1230-1242. DOI: 10.1016/j.kint.2020.01.028
42. Wang Y, Zhao J, Qin Y, Yu Z, Zhang Y, Ning X, Sun S. **The specific alteration of gut microbiota in diabetic kidney diseases—a systematic review and meta-analysis**. *Front Immunol* (2022.0) **13** 908219. DOI: 10.3389/fimmu.2022.908219
43. Rothman KJ. **Causes**. *Am J Epidemiol* (1976.0) **104** 587-592. DOI: 10.1093/oxfordjournals.aje.a112335
44. Lavange LM, Kalsbeek WD, Sorlie PD, Avilés-Santa LM, Kaplan RC, Barnhart J, Liu K, Giachello A, Lee DJ, Ryan J. **Sample design and cohort selection in the Hispanic community health study/study of latinos**. *Ann Epidemiol* (2010.0) **20** 642-649. DOI: 10.1016/j.annepidem.2010.05.006
45. Sorlie PD, Avilés-Santa LM, Wassertheil-Smoller S, Kaplan RC, Daviglus ML, Giachello AL, Schneiderman N, Raij L, Talavera G, Allison M. **Design and implementation of the Hispanic community health study/study of latinos**. *Ann Epidemiol* (2010.0) **20** 629-641. DOI: 10.1016/j.annepidem.2010.03.015
46. Kaplan RC, Wang Z, Usyk M, Sotres-Alvarez D, Daviglus ML, Schneiderman N, Talavera GA, Gellman MD, Thyagarajan B, Moon J-Y. **Gut microbiome composition in the Hispanic community health study/study of latinos is shaped by geographic relocation, environmental factors, and obesity**. *Genome Biol* (2019.0) **20** 219. DOI: 10.1186/s13059-019-1831-z
47. Inker LA, Eneanya ND, Coresh J, Tighiouart H, Wang D, Sang Y, Crews DC, Doria A, Estrella MM, Froissart M. **New creatinine- and cystatin c–based equations to estimate GFR without race**. (2021.0) **385** 1737-1749. DOI: 10.1056/NEJMoa2102953
48. Delgado C, Baweja M, Crews DC, Eneanya ND, Gadegbeku CA, Inker LA, Mendu ML, Miller WG, Moxey-Mims MM, Roberts GV. **A unifying approach for GFR estimation: recommendations of the NKF-ASN task force on reassessing the inclusion of race in diagnosing kidney disease**. *Am J Kidney Dis* (2022.0) **79** 268-88.e1. DOI: 10.1053/j.ajkd.2021.08.003
49. Sorlie PD, Allison MA, Aviles-Santa ML, Cai J, Daviglus ML, Howard AG, Kaplan R, LaVange LM, Raij L, Schneiderman N. **Prevalence of hypertension, awareness, treatment, and control in the Hispanic community health study/study of latinos**. *Am J Hypertens* (2014.0) **27** 793-800. DOI: 10.1093/ajh/hpu003
50. Qi Q, Strizich G, Merchant G, Sotres-Alvarez D, Buelna C, Castaneda SF, Gallo LC, Cai J, Gellman MD, Isasi CR. **Objectively measured sedentary time and cardiometabolic biomarkers in US Hispanic/Latino adults: the Hispanic Community Health Study/Study of Latinos (HCHS/SOL)**. *Circulation* (2015.0) **132** 1560-1569. DOI: 10.1161/CIRCULATIONAHA.115.016938
51. Hillmann B, Al-Ghalith GA, Shields-Cutler RR, Zhu Q, Gohl DM, Beckman KB, Knight R, Knights D. **Evaluating the information content of shallow shotgun metagenomics**. *mSystems* (2018.0) **3** e00069-18. DOI: 10.1128/mSystems.00069-18
52. Peters BA, Lin J, Qi Q, Usyk M, Isasi CR, Mossavar-Rahmani Y, Derby CA, Santoro N, Perreira KM, Daviglus ML. **Menopause is associated with an altered gut microbiome and estrobolome with implications for adverse cardiometabolic risk in the Hispanic Community Health Study/Study of Latinos (HCHS/SOL)**. *mSystems 2022* **7**. DOI: 10.1128/msystems.00273-22
53. Thompson LR, Sanders JG, McDonald D, Amir A, Ladau J, Locey KJ, Prill RJ, Tripathi A, Gibbons SM, Ackermann G. **A communal catalogue reveals Earth’s multiscale microbial diversity**. *Nature* (2017.0) **551** 457-463. DOI: 10.1038/nature24621
54. Glenn TC, Nilsen RA, Kieran TJ, Sanders JG, Bayona-Vasquez NJ, Finger JW, Pierson TW, Bentley KE, Hoffberg SL, Louha S. **Adapterama I: universal stubs and primers for 384 unique dual-indexed or 147,456 combinatorially-indexed Illumina libraries (iTru & iNext)**. *PeerJ* (2019.0) **7** e7755. DOI: 10.7717/peerj.7755
55. Langmead B, Salzberg SL. **Fast gapped-read alignment with bowtie 2**. *Nat Methods* (2012.0) **9** 357-359. DOI: 10.1038/nmeth.1923
56. Pj M, Holmes S, Watson M. **Phyloseq: an R package for reproducible interactive analysis and graphics of microbiome census data**. *PLoS One* (2013.0) **8** e61217. DOI: 10.1371/journal.pone.0061217
57. Oksanen J, Blanchet F, Kindt R, Legendre P, O’hara R, Simpson G. **Multivariate analysis of ecological communities in R: vegan tutorial**. *R Package Version 2.6-4* (2013.0)
58. Chen J, Bittinger K, Charlson ES, Hoffmann C, Lewis J, Wu GD. *Associating microbiome composition with environmental covariates using generalized UniFrac distances* (2012.0) **28** 2106-2113
59. Chen G-C, Chai JC, Yu B, Michelotti GA, Grove ML, Fretts AM, Daviglus ML, Garcia-Bedoya OL, Thyagarajan B, Schneiderman N. **Serum sphingolipids and incident diabetes in a US population with high diabetes burden: the Hispanic Community Health Study/Study of Latinos (HCHS/SOL)**. *Am J Clin Nutr* (2020.0) **112** 57-65. DOI: 10.1093/ajcn/nqaa114
60. Mandal S, Van Treuren W, White RA, Eggesbo M, Knight R, Peddada SD. **Analysis of composition of microbiomes: a novel method for studying microbial composition**. *Microb Ecol Health Dis* (2015.0) **26** 27663. DOI: 10.3402/mehd.v26.27663
|
---
title: Optical Ultrastructure of Large Mammalian Hearts Recovers Discordant Alternans
by In Silico Data Assimilation
authors:
- Alessandro Loppini
- Julia Erhardt
- Flavio H. Fenton
- Simonetta Filippi
- Marcel Hörning
- Alessio Gizzi
journal: Frontiers in Network Physiology
year: 2022
pmcid: PMC10012998
doi: 10.3389/fnetp.2022.866101
license: CC BY 4.0
---
# Optical Ultrastructure of Large Mammalian Hearts Recovers Discordant Alternans by In Silico Data Assimilation
## Abstract
Understanding and predicting the mechanisms promoting the onset and sustainability of cardiac arrhythmias represent a primary concern in the scientific and medical communities still today. Despite the long-lasting effort in clinical and physico-mathematical research, a critical aspect to be fully characterized and unveiled is represented by spatiotemporal alternans patterns of cardiac excitation. The identification of discordant alternans and higher-order alternating rhythms by advanced data analyses as well as their prediction by reliable mathematical models represents a major avenue of research for a broad and multidisciplinary scientific community. Current limitations concern two primary aspects: 1) robust and general-purpose feature extraction techniques and 2) in silico data assimilation within reliable and predictive mathematical models. Here, we address both aspects. At first, we extend our previous works on Fourier transformation imaging (FFI), applying the technique to whole-ventricle fluorescence optical mapping. Overall, we identify complex spatial patterns of voltage alternans and characterize higher-order rhythms by a frequency-series analysis. Then, we integrate the optical ultrastructure obtained by FFI analysis within a fine-tuned electrophysiological mathematical model of the cardiac action potential. We build up a novel data assimilation procedure demonstrating its reliability in reproducing complex alternans patterns in two-dimensional computational domains. Finally, we prove that the FFI approach applied to both experimental and simulated signals recovers the same information, thus closing the loop between the experiment, data analysis, and numerical simulations.
## 1 Introduction
In nature, a broad variety of pattern formations can be found on very different length scales and functions, such as collective behavior of fish swarms (Jakobsen and Johnsen, 1988), the animal skin patterning (Murray, 2003; Miyazawa et al., 2010), the cell dynamics during embryogenesis (Ju et al., 2017), the formation of neuronal networks in brains (van den Heuvel and Hulshoff Pol, 2010), and the electromechanical function of the cardiovascular system (Christoph et al., 2018). The latter is crucial to maintain life as we know but is susceptible to malfunction due to its complex morphology such as the vascular system (Luther et al., 2011), cellular orientation (Papadacci et al., 2017), and mechano-electrophysiological wave patterning (Hörning et al., 2012). Slight variations in the organization of those patterns can have fatal consequences, and thus, cardiovascular diseases are the primary cause of death in industrial countries.
One of the complex and not fully understood heart behaviors, possibly inducing cardiac disease, is alternans. It describes a phase-dependent alternation on either a single-cell or tissue level and can be described as a beat-to-beat alternation of short and long heartbeats (membrane potential, intracellular calcium) or myocyte contractions. Alternans is known to be involved in a series of cardiovascular conditions as either cause or symptom. These include, among others, ventricular fibrillation, arrhythmias, and sudden cardiac death (Adam et al., 1984; Konta et al., 1990; Pastore et al., 1999), especially in patients after myocardial infarction (Ikeda et al., 2000). Other triggers for alternans are ischemia of the myocardium, ectopic heartbeats, and coronary occlusion (Green, 1935; Dilly and Lab, 1988; Taggart et al., 1996; Ren et al., 2011). In early studies, alternans was observed in terms of myocardial contractility, aortic pressure, and stroke volume (Mitchell et al., 1963). In medical applications, this phenomenon has therefore been widely employed as a predictive tool for determining risks for fibrillation, venous thromboembolism, arrhythmia, and sudden cardiac death (Dilly and Lab, 1988; Kim et al., 2009). Besides, it is used to assess the necessity and urgency of certain surgical operations, such as implantation of cardioverter defibrillators (Merchant et al., 2012).
Several mechanisms have been revealed during three decades of intensive research that can induce alternans. Early studies stated the critical role of calcium cycling and electrical restitution of the action potential in the generation of alternans (Badeer et al., 1967; Dilly and Lab, 1988; Konta et al., 1990). Repolarization gradients at the tissue level have further been shown to lead to complex macroscopic voltage alternans patterns (Pastore et al., 1999). Later, it was shown that fluctuations in the cyclic release of Ca2+ from the sarcoplasmic reticulum could lead to Ca2+ alternans tightly coupled with voltage repolarization alternans (Lab and Lee, 1990; Qu et al., 1999; Walker et al., 2003; Diaz et al., 2004). Similar to that, a fine-scaled initiation of alternans was linked to the subsequent formation of larger alternating regions (Jia et al., 2010). Additionally, alternans can be promoted by low temperature or application of drugs (Xie and Weiss, 2009; Gizzi et al., 2017; Loppini et al., 2021).
While alternans can be observed at single cells for the action potential duration (APD) and the calcium transient amplitude (CTA), in tissue, those oscillations can synchronize and lead to spatial concordant alternans (CA) or discordant alternans (DA) (Uzelac et al., 2017). CA is observed when the entire tissue alternates in phase, while DA is classified with at least two out-of-phase oscillating regions (Xie and Weiss, 2009; Gizzi et al., 2013; Gizzi et al., 2017) spatially separated by nodal lines, i.e., non-alternating domains (Hörning et al., 2017). The conduction velocity (CV) plays a crucial role in developing alternans. Usually, concordant or discordant APD and CTA depend on CV restitution (Karagueuzian et al., 2013). A slowing of the CV, caused by the incomplete recovery of the fast sodium current, concurs to promote large gradients of repolarization, thus sustaining DA patterns. Furthermore, alternans can be studied in terms of electromechanically out-of-phase regions. In this case, larger CTAs are triggered by shorter APDs and vice versa (Sato et al., 2006).
Based on the experimental finding, numerous computational models have been developed that can show the onset and transition of alternans in both single cells and tissue (Karma, 1993, 1994; Qu et al., 1999; Watanabe et al., 2001; Cherry and Fenton, 2004; Tao et al., 2008; Shiferaw et al., 2003; Huertas et al., 2010). However, the striking limitation of the current modeling approaches consists in the capability of reproducing complex DA patterns in anatomically realistic computational domains. In practice, the appearance of CA and DA in numerical simulations requires, up to now, an ad hoc tuning of physiological parameters, usually deviating from the optimal set obtained from experimental CV and restitution curves (Cairns et al., 2017). Innovative multiscale and multiphysics formulations of cell–cell couplings aim at filling this gap. Non-linear, stress-assisted, and fractional diffusion (Lin and Keener, 2010; Hurtado et al., 2016; Cherubini et al., 2017; Cusimano et al., 2020; Cusimano et al., 2021), ephaptic and gap junction–mediated couplings (Lenarda et al., 2018; Weinberg, 2017), cellular automata, and coarse-grained homogenized gap junction approaches (Treml et al., 2021; Irakoze and Jacquemet, 2021) represent the state-of-the-art in this direction. Furthermore, within the specific context of cardiac electrophysiology, recent studies are proposing novel methods of data estimation, data assimilation, and uncertainty quantification (Barone et al., 2020a; Barone et al., 2020b; Pathmanathan et al., 2020; Marcotte et al., 2021) to reproduce complex cardiac dynamics with a reduced computational cost.
On such a ground, we propose an innovative data assimilation technique using the optical ultrastructure obtained from a frequency analysis of voltage fluorescence activations on intact canine ventricles demonstrating its potential role in recovering complex alternans patterns in silico. The results presented in this study fundamentally advance the understanding of alternans. Furthermore, the proposed observation strategy may enable possible applications to personalized medicine, such as quantifying alternans and higher-other rhythms without heavy computational resources or massive experimental campaigns. As the ultrastructure of the heart is unique for every subject, it may be used as the base for studying possible diseased states and treatments. Thus, this study lays the promising foundation for such approaches in the near future.
This manuscript is organized as follows. Section 2 introduces the Fourier-based method and the experimental data assimilation technique in electrophysiological mathematical models. Section 3 demonstrates the ability of our frequency technique to recover alternans in cardiac tissue at high-frequency pacing rates, and we identify the optical ultrastructure to assimilate in silico. Besides, the optimal combination of data assimilation heterogeneities matches complex experimental alternans patterns at best. Section 4 closes the work with a discussion of limitations and perspectives of the current approach.
## 2.1 Experimental Setup
Right ventricle wedges from canine were prepared according to the experimental protocols approved by the Institutional Animal Care and Use Committee of the Center for Animal Resources and Education at Cornell University. Fluorescence optical mapping recordings of the membrane potential were recorded at a spatial resolution of 600 × 600 μm2 per pixel for a grid size of 7.7 × 7.7 cm2 and a temporal resolution of 2 ms at physiological conditions. For details of the experimental setup information, we refer to the previous studies (Fenton et al., 2009; Luther et al., 2011; Gizzi et al., 2013; Gizzi et al., 2017).
## 2.2.1 Fourier Transformation Imaging
Fourier transformation imaging was applied to the fluorescence optical mapping recordings, as introduced before (Hörning et al., 2017). The optical recordings were pixel-wise decomposed and transformed to the mathematically complex Fourier space, F x,y (f), as a function of the frequency f, i.e., Ix,yt→Fx,yf, [1] where I x,y (t) is the intensity at the spatial position (x, y) and t is the time. From that, the amplitude |F x,y (f)| and the phase arg(F x,y (f)) are calculated and spatially recomposed to two respective Fourier frequency-series.
## 2.2.2 APD Alternans Maps
Alternans maps were pixel-wise calculated on pre-analyzed signals. Pre-analysis involves detrending, nearest-neighbor averaging in time with a rectangular window (7 frames width), and space filtering with Gaussian kernel (4 pixels radius). The APD is the extracted by threshold crossing at $20\%$ maxIx,y. The temporal difference of subsequent action potentials ΔAPD is then quantified as ΔAPDn=APDn+1−APDn, [2] where n denotes the beat number. ΔAPD maps were recomposed, and a functional color scheme was applied that indicates non-alternating tissue and nodal lines when ΔAPD = 0 ± 2 ms, which is defined by the temporal resolution of the recordings. A larger or smaller ΔAPD shows phase-dependent alternans, as introduced before (Gizzi et al., 2013).
## 2.3 Mathematical Model
We based our numerical simulations on the four-variable minimal model for ventricular action potentials (Bueno-Orovio et al., 2008), solved on two-dimensional anisotropic heterogeneous spatial domains according to the fine-tuning performed by Fenton et al. [ 2013]. The model includes a phenomenological description of main transmembrane ion currents and is properly generalized with a heterogeneous diffusion contribution to account for spatial effects. The model’s equations are ∂tu=∇⋅Dij∇u−Jfi+Jso+Jsi, (3a) dtv=1−Θu−θvv∞−v/τv−−Θu−θvv/τv+, (3b) dtw=1−Θu−θww∞−w/τw−−Θu−θww/τw+, (3c) dts=1+tanhksu−us/2−s/τs, (3d) where u is the dimensionless cell membrane potential, v, w, and s are gating variables regulating ion current activation, and Θ(x) denotes the Heaviside step function. J fi, J so, J si represent fast-inward, slow-outward, and slow-inward transmembrane currents: Jfi=−Θu−θvu−θvuu−uv/τfi, (4a) Jso=1−Θu−θwu−uo/τo+Θu−θw/τso, (4b) Jsi=−Θu−θwws/τsi. ( 4c) Time constants and asymptotic values for gating variables depend on the membrane voltage: τv−$u = 1$−Θu−θv−τv1−+Θu−θv−τv2−, (5a) τw+u=τw1++τw2+−τw1+tanhkw+u−uw++$\frac{1}{2}$, (5b) τw−u=τw1−+τw2−−τw1−tanhkw−u−uw−+$\frac{1}{2}$, (5c) τsou=τso1+τso2−τso1tanhksou−uso+$\frac{1}{2}$, (5d) τsu=1−Θu−θwτs1+Θu−θwτs2, (5e) τou=1−Θu−θoτo1+Θu−θoτo2, (5f) v∞=1−Θu−θv−, (5g) w∞=1−Θu−θo1−u/τw∞+Θu−θow∞*. ( 5h) D ij is the two-dimensional diffusion tensor, defined as Dij≡σijSoCm=D11D12D21D22, [6] where σ ij is the conductivity tensor, S o represents the cell surface-to-volume ratio, and C m is the membrane capacitance. The tensor elements are defined in the two-dimensional Cartesian domain as D11=D∥cos2αx,y+D⊥sin2αx,y, (7a) D22=D∥sin2αx,y+D⊥cos2αx,y, (7b) D12=D21=D∥−D⊥cosαx,ysinαx,y. (7c) Here, the function α(x, y) represents the local fiber orientation and D ∥ and D ⊥ denote diffusivities along the directions parallel and perpendicular to the fibers. We used the same anisotropy settings as in previous computational studies on cardiac activation maps (Loppini et al., 2019). Model parameters are reported in Table 1. The parameter set is consistent with the one originally fine-tuned by Fenton et al. [ 2013], related to endocardial tissue at 37° Celsius. Specifically, parameter values are set to reproduce AP features, conduction velocity, and restitution curves as observed in canine cardiac tissues, the same considered in this study. Furthermore, we assume the anisotropy ratio as 1:3 in line with previous modeling analyses (Loppini et al., 2019). For a more detailed description of model parameters and for a comparison between the four-variable model results and experiments, we refer the reader to the two abovementioned studies.
**TABLE 1**
| Unnamed: 0 | Unnamed: 1 | Unnamed: 2 | Unnamed: 3 | Unnamed: 4 |
| --- | --- | --- | --- | --- |
| u o = 0 | kw+ =8 | τv+ = 1.4506 ms | τ fi = 0.10 ms | D ∥ = 0.010 cm2/ms |
| u u = 1.56 | kw− = 20 | τv1− = 55 ms | τ si = 2.9013 ms | D ⊥ = 0.003 cm2/ms |
| u s = 0.9087 | k s = 2.0994 | τv2− = 40 ms | τ s1 = 2.7342 ms | — |
| u so = 0.65 | k so = 2 | τw1+ = 175 ms | τ s2 = 2 ms | — |
| uw− = 0.00615 | θ v = 0.3 | τw1− = 40 ms | τ so1 = 40 ms | — |
| uw+ = 0.0005 | θv− = 0.2 | τw2+ = 230 ms | τ so2 = 1.2 ms | — |
| w∞∗ = 0.78 | θ w = 0.13 | τw2− = 115 ms | τ o1 = 470 ms | — |
| w∞∗ = 0.78 | θ o = 0.006 | τw∞ = 0.0273 ms | τ o2 = 6 ms | — |
As detailed in Section 3.4, the generalization to heterogeneous modeling by data assimilation is obtained by imposing a spatial variation of selected parameters, opportunely sorted around their reference values (i.e., Table 1) based on experimentally informed profiles. On this basis, the heterogeneous model is obtained by perturbing parameters of the homogeneous model so that global features in the evoked electrical activity are still correctly reproduced. We numerically integrated the model with an explicit Euler scheme implemented in Fortran, discretizing the spatial operators to account for heterogeneous diffusion and phase-field boundary conditions. We solved the model in both 1D and 2D domains, assuming zero-flux boundary conditions. Space and time discretization is Δx = 0.025 cm, Δt = 0.01 ms. Stability and conduction velocity convergence was verified upon mesh refinement and testing higher-order discretization schemes in time (second- and fourth-order Runge–Kutta), achieving non-significant variations in the computed results.
## 3 Results
Alternans in cardiac tissues is observed at high-frequency entrained (Gizzi et al., 2013; Loppini et al., 2021) and self-sustained freely rotating and heterogeneity-bound spiral waves (Hörning et al., 2017). In the past, those dynamics were difficult to visualize without the use of heavy spatial–temporal filters and thus hindering the fine-scale visualization and study of nodal lines that are observed in discordant alternans. Here, we apply the spatial-filter–free FFI analysis method that was recently introduced by Hörning et al. [ 2017]. It is worth mentioning that unique alternative methods assessing cardiac alternans are still required due to the critical differences in electrophysiological signals. The action potential amplitude (Chen et al., 2017) and calcium transient duration and amplitude (Clusin, 2008; Visweswaran et al., 2013) represent, in fact, different approaches that require a meticulous comparison.
## 3.1 Alternans in Intact Canine Ventricles
Simultaneous recordings of the epicardium and endocardium in RV canine preparations were observed (Gizzi et al., 2013). Physiological alternans-free wave conduction and alternans states could be observed depending on the pacing site position and pacing frequency. At lower pacing frequencies, no alternans is observed, as the APD is sufficiently short to prevent the interaction of subsequent waves. Figure 1A shows such an example observed at the entrainment frequency f $$p \leq 3.2$$ Hz on the epicardium. The phase and amplitude of the pixel-wise FFI-analyzed recordings show a continuously evolving phase and amplitude in the entire tissue. No spatially correlated phase or amplitude is observed at f $\frac{1}{2}$ = f p/2 = 1.6 Hz that would indicate alternans. Also, closer inspection of the local signaling does not indicate alternans (Figure 1A, right panels). The normalized APs at P1 (pink square, top) and P2 (cyan square, bottom) show no alteration in peak height or APD, as confirmed by the respective amplitude in the Fourier space. Only a single peak at the entrainment frequency f p is observed. The lower amplitude peak at 2f p shows the second mode and does not carry relevant information. Contrarily, alternans is observed when the epicardium is entrained with a higher frequency. Figure 1B shows the same analysis and local signaling recordings at f $$p \leq 8.0$$ Hz. In this case, the pacing frequency is sufficiently high so that subsequent waves influence each other. The phase and amplitude information at f $\frac{1}{2}$ = 4 Hz shows a typical pattern of discordant 2:2 alternans. That means that two subsequent waves lead to two different APDs in time. In space, the APD of each wave can transiently switch between the two APDs that are spatially confined by nodal lines. The latter can be identified by the spatial phase jump of about π, and the amplitude valley. The normalized APs at P1 (cyan) and P2 (pink) show APD alternation between shorter and longer APDs. Every second AP is shown in either black or red to facilitate visualization. For those time series, a second peak at f $\frac{1}{2}$ = 4 *Hz is* visible in the amplitude spectrum, since a second underlying frequency is present that correlates with two times the wave period (2T = f $\frac{1}{2}$).
**FIGURE 1:** *Fourier analysis in a high-frequency entrained canine heart. (A,B) show the epicardium of a canine heart that is paced from the top (RV anterior) entrained with f
p = 3.2 Hz (no alternans) and f
p = 8.0 Hz (2:2 DA), respectively. Shown is the Fourier space (phase and amplitude) of frequencies f and f
1/2. The white arrows indicate the direction of the propagation wave. Two AP rhythms measured at two independent locations (P1 and P2, 6 × 6 pixel FOV) and their respective Fourier spectra are shown exemplarily. (B) shows a typical example where nodal lines are visible at f
1/2. Every second AP time course is shown in red to facilitate visualization of the 2:2 AP rhythm. The second peak 2f
p in the Fourier space of the upper AP rhythms is a typical higher-order frequency mode. The positions P1 and P2 are highlighted by pink and cyan squares in (A) and (B). Three waves are marked by the wave numbers, as n − 1, n, and n + 1.*
## 3.2 Visualization of Higher-Order Discordant Alternans
Although 2:2 alternans is the most commonly observed AP rhythm, other higher-order AP rhythms exist (see, e.g., Figure 6 in Gizzi et al. [ 2013]). The epicardium that is shown in Figure 1 was additionally paced from the bottom (base) of the heart at f $$p \leq 8.5$$ Hz, which led to a spatially mixed mode of 2:2 and 4:4 alternans (Figure 2A). While the 2:2 AP rhythm shows a single amplitude peak at f $\frac{1}{2}$ = 4.25 Hz, two additional amplitude peaks are observed in the Fourier space for the 4:4 AP rhythm: one very close to f $\frac{1}{2}$ and one at f $\frac{3}{4.}$ As the two peaks at around f $\frac{1}{2}$ are very close to each other but implicate different information, they are from here on defined as f$\frac{1}{21}$ and f$\frac{1}{22.}$ The well-defined spatial distribution of the two different AP rhythms is visualized at the phase and amplitude reconstructions in Figure 2B. Here, f$\frac{1}{22}$ and f $\frac{3}{4}$ show additional spatial information to the frequencies at f p and f$\frac{1}{21}$ in the Fourier space. The 2:2 (left side, RV anterior) and 4:4 (right side, RV anterior) alternans regimes are spatially separated at f $\frac{3}{4.}$ The right side shows synchronized phase and correspondingly elevated amplitude signaling that is present on the left side. At f$\frac{1}{21}$ and f$\frac{1}{22}$, more complex phase dynamics are observed in addition to the nodal lines that separate different entrained alternating regions in the tissue. Figure 2C shows the corresponding phase wave directions. Numbers 1 and 2 indicate the temporal order of occurring phase waves. In this framework, nodal lines and nodal areas carry more information compared to the classically analyzed temporal differences in APD (Pastore et al., 1999). At the phase of f$\frac{1}{21}$ (top, right side, Figure 2B), the phase wave accelerates at the nodal line from 1 to 2 (Figure 2C), while contrarily at f$\frac{1}{22}$, the phase waves propagate in the opposite direction and jump by 2π (top, right side, Figure 2B). The latter indicate the typical phase waves, as observed in the Fourier space (f$\frac{1}{22}$) at 2:2 AP rhythms (see also Figure 1). For the sake of completeness, Figure 2 shows additionally f $\frac{1}{4}$ that corresponds to the duration of four AP rhythms. However, f $\frac{1}{4}$ is a rather non-significant frequency and appears only because of the intrinsic noise of the system, as the peak at f p describes already the main oscillation of the system, and the three other amplitude peaks at f$\frac{1}{21}$, f$\frac{1}{22}$, and f $\frac{3}{4}$ describe the periodic temporal variations.
**FIGURE 2:** *Simultaneous DA of different AP rhythms in a high-frequency entrained epicardium of a canine heart. (A) shows two AP rhythms (P1 and P2, 6 × 6 pixel FOV) of 2:2 and 4:4 alternans and their Fourier spectra, respectively. The stimulation site is on the bottom (base) of the heart with f
p = 8.50 Hz. f indicates the entrainment frequency, and f
3/4 indicates the presence of a 4:4 AP rhythm with its two corresponding peaks,
f1/21
and
f1/22
around f
p/2. The peak at f
1/4 corresponds to a wavelet of four APs that is fully expressed by f
p. Every second or fourth AP is shown in red to facilitate visualization of the 2:2 or 4:4 AP rhythms. (B) shows the corresponding spatial phase and amplitude information of the five frequency peaks. The Fourier data shown at f
3/4 indicate a 4:4 AP rhythm on the right side (RV anterior) of the heart only. The respective counter phase is illustrated at
f1/21
.
f1/22
shows the typical 2:2 phase of a 2:2 AP rhythm, similarly to that illustrated in Figure 1B. The positions P1 and P2 are highlighted by pink and cyan squares. (C) shows the direction of the respective phase information indicated by red arrows that are shown in (B). Nodal lines are outlined by black dotted lines. Numbers 1 and 2 shown in
f1/21
and
f1/22
indicate the phase waves that are shifted by π.*
## 3.3 Spatial Synchronization of Alternans Patterns
The frequency response observed at a single recorded pixel is useful to get an overview of the local alternans offset (in analogy to the well-known restitution curves). Figure 3A shows frequency maps with the normalized amplitudes depending on the entrainment f p for top (base, left panels) and left (RV posterior, right panels) paced canine ventricles. The top and bottom panels show data recorded at the epicardium (EPI) and endocardium (ENDO). The main peaks (bright yellow peaks) indicate f p. Above f p appears a second peak from about 5 Hz that indicates alternans (f$\frac{1}{22}$). Additionally, only for very high frequencies, here f $$p \leq 9.2$$ Hz, a peak at f $\frac{3}{4}$ is visible that indicates 4:4 alternans. Figure 3B shows a generalized scheme of the frequency maps for comparisons. The offset of 2:2 alternans differs between the EPI and the ENDO, while the peak f $\frac{3}{4}$ is observed for all four frequency maps at the same entrainment frequency f $$p \leq 9.2$$ Hz. At frequencies higher than 10.0 Hz, fibrillation is observed, as additionally shown in the right (f $$p \leq 10.7$$ Hz, RV posterior) paced EPI and ENDO frequency maps. The comparison between f $$p \leq 9.2$$ Hz and 10.7 *Hz is* shown in Figure 3C. Fibrillation shows an elevated baseline and a broad spectrum for frequencies above f p.
**FIGURE 3:** *Pacing-site–dependent frequency maps. (A) shows frequency maps obtained from the top (base, left panels) and left (RV posterior, right panels) paced canine ventricles, respectively. The top and bottom panels show data recorded at the epicardium (EPI) and endocardium (ENDO). 2:2 AP rhythms (f
1/2) are observed from about 4.5 Hz. 4:4 AP rhythms (f
3/4) are observed only in base paced canine recordings at a pacing of about 9.2 Hz. (B) shows a guide of the eye for (A) with the main frequencies (solid lines), secondary peaks (dashed lines), and higher-ordered peaks (dotted lines). (C) shows a comparison of the normalized amplitudes for 4:4 alternans (9.2 Hz, red line) and fibrillation (10.7 Hz, black line) that is observed at the ENDO paced from the RV posterior.*
While a critical frequency induces fibrillation, the complex spatiotemporal alternans patterns stabilize with the increasing entrainment frequency (Gizzi et al., 2013) (Figure 4A). Figure 4B shows selected snapshots of the EPI and ENDO from two different pacing sites. Initially, no alternans is observed at the EPI at a lower f p ≃ 4 Hz, but the initiation of alternans at the ENDO can be seen (endocardium base paced, Figure 4). Interestingly, this occurs in larger speckled patches rather than in defined areas, which indicates the alternans-offset difference among individual cells. This speckled-like early fine-scale initiation of alternans was suggested previously by Jia et al. [ 2010]. With increasing f p, those patterns synchronize spatially and lead to distinct phase areas that are spatially separated by nodal lines, as best visible at the EPI. Although the ENDO shows comparable stabilization of alternans in the phase, the amplitude shows more spatial variations. This is most likely caused by the influence of the Purkinje fibers that are confined to the subendocardial layer and believed to be responsible for the initiation of ventricular fibrillation (Fox et al., 2002; Muñoz et al., 2018).
**FIGURE 4:** *Stabilization of nodal line formation at higher entrainment frequencies. (A) shows the evolution of alternans from lower to higher pacing frequencies. The Fourier space, phase and amplitude at f
1/2 and ΔAPD, is shown from f
p = 2 Hz to 7.2 Hz. (B) shows the concordant alternans evolution of the frequency maps shown in Figures 3A,B. The top (base, left panels) and left (RV posterior, right panels) paced canine hearts are shown on the left and right sides, and the respective epicardium (EPI) and endocardium (ENDO) are shown at the bottom and top. The regimes of no alternans, concordant alternans (CA), and discordant alternans (DA) are indicated on the top and bottom of the figures. Red arrows indicate the position of the electrode.*
## 3.4 Data Assimilation From Optical Ultrastructure
As the differences of the electrophysiological properties of individual cells also lead to differences in the alternans-offset and restitution characteristics, it is useful to take pixel-based differences into account when modeling alternans dynamics in silico. The advantage of the *Fourier analysis* of heart tissues is that the optical ultrastructure can be revealed in the low-frequency regime, as shown in Figure 5. Especially, the amplitude information at $f = 0.5$ *Hz is* a stable indicator for morphologically restricted differences that are independent of the pacing location (Figure 5A) and pacing frequency (Figure 5B). Here, we utilize the low-frequency regime, as it is also an indirect measure of the signal height, i.e., the observed baseline of the AP rhythms. Therefore, we assume that the strength of the emitted signal depends on the local tissue properties and thus relates to the heart ultrastructure.
**FIGURE 5:** *Optical ultrastructure extracted from the low-frequency regime of the epicardium (EPI). (A) shows the Fourier space—phase and amplitude—at f = 0.5 Hz for f
p = 3 Hz frequency entrained tissues that are paced on the endocardium from four different directions, as indicated by the white arrows. (B) shows the amplitudes of the Fourier space at f = 0.5 Hz for different entrainment frequencies f
p that are stimulated at the base. The white arrows in (A,B) indicate the respective direction of wave propagation. Below the amplitude images are indicated the regimes of no alternans and discordant 2:2 alternans (DA).*
In order to validate this hypothesis, we propose a novel data assimilation approach using the ultrastructure observed at $f = 0.5$ Hz assuming the influence in the diffusive term, Eq. 2.3 (D ∥, D ⊥), and the time constants that shape the AP, Eq. 5 (τw+, τ so, and τ si). The tissue heterogeneities are applied on a pixel-based scale, so that the local variations of conduction velocity and action potential are accounted through specific parameters having a strong impact on the resulting APD. In detail, a mask encoding the actual tissue of the experimental samples was extracted by analyzing the signal-to-noise ratio, and the spatial map of the Fourier amplitude spectrum at $f = 0.5$ Hz was computed. The resulting ultrastructure profile was smoothed by using a Gaussian kernel, restricted to a radius of 6 pixels with a variance equal to 5. The heterogeneity field was evaluated by normalizing the ultrastructure mask as Hx,y=δrmax|r|+1, [8] where r=s−s¯, s = |F x,y (0.5 Hz)|, and δ is the parameter denoting the desired maximal variation. We set this value in the range [0,1] by assuming heterogeneity variations of the local parameters up to $100\%$. Eventually, the heterogeneity map is combined with the tissue map to include also information on tissue boundaries (i.e., phase-field). We refer to this heterogeneity field as an H 1(x, y) map. Furthermore, we considered a second heterogeneity field that enhances the data assimilation procedure from low-amplitude areas in the Fourier spatial map. We evaluated the reciprocal of the H 1 (x, y) map and normalized the result according to Eq. 8, still constraining the parameter δ ∈ [0, 1]. This second spatial scalar field is denoted as the H 2(x, y) map with average 1 and variation δ. It is worth noting that the proposed procedure represents a generalization of the phase-field method (Fenton et al., 2005) that permits both the inclusion of tissue boundary and specific modulations of model parameters. The resulting heterogeneity maps are used as a multiplication factor on selected model parameters to achieve a constitutive heterogeneity shaped on a tissue ultrastructure. In our analyses, we focused on heterogeneities related to diffusivity and APD parameters (D ∥, D ⊥, τw+, τ so, τ si), by using the following constitutive law: px,y=p¯Hix,y. [9] Here, p (x, y) denotes a spatial dependent parameter, p¯ is the original model parameter, and H i (x, y) is the heterogeneity field (with $i = 1$, 2). A visual representation of the adopted data assimilation technique is thoroughly provided in the next sections. Naming P 1 as the set of diffusivity parameters and P 2 as that of APD-regulating time constants, we investigated all possible combinations of H 1 and H 2 maps on P 1 and P 2, i.e., a specific heterogeneity profile was applied to diffusivities and APD-related time constants. It is worth noting that, with this setting, we are assuming a correlation between diffusivity and APD, with such parameters non-linearly coupled through complex electronic effects in cardiac tissue.
## 3.5 In Silico Data Assimilation and Alternans Model Prediction
We performed an extensive in silico study on both one-dimensional cables and two-dimensional tissues to test the heterogeneity effects on alternans onset and severity. In particular, we computed H 1 (x, y) and H 2 (x, y) maps from a selected experimental tissue to shape model parameters’ heterogeneity in 1D and 2D domains, investigating all possible combinations of the heterogeneity fields on diffusivity and APD-regulating time constants, at δ = 0.25, 0.5, 1. For 1D simulations, we extracted one-dimensional cuts of H 1 (x, y) and H 2 (x, y) maps along the experimental propagating wavefronts (not shown). This preliminary set of numerical simulations was used as a first benchmark of the data assimilation procedure. We observed that the heterogeneous model is able to 1) recover the expected average CV and AP features and 2) emphasize alternans onset and severity, also inducing conduction block phenomena not observed in the homogeneous case. We then tested data assimilation within 2D computational domains observing notable differences with respect to the homogeneous case. In the following, we show two representative examples comparing the overall results for the same ventricle stimulated with a pacing-down protocol both at the ventricle base and in the right anterior ventricular region. The pacing-down protocol consists in stimulating the tissue starting from a low frequency and progressively reducing the pacing frequency. In particular, at each frequency, we delivered a stimulation train of 10 beats to ensure the tissue reached a stationary regime. This protocol reproduces the experimental one, and in our analyses, we computed alternans patterns on the last two beats to avoid transient effects.
## 3.5.1 Base Ventricle Stimulation
The phase-field ultrastructure and heterogeneity maps are shown in Figures 6A,B. In this case, we tested the model with 1) spatially homogeneous parameters, 2) H 1 maps applied on diffusivity (H1 model), 3) H 2 maps applied on APD-regulating parameters (H2 model), and 4) H 1 and H 2 maps applied simultaneously (H3 model). Figure 6C shows simulated alternans maps for a selected frequency f $$p \leq 6.2$$ ± 0.4 Hz. On the left, the homogeneous model could not reproduce complex and discordant alternans maps during the pacing-down protocol. Both H1 and H2 models (center and right columns) succeeded in reproducing transition into the discordant alternans regime, though showing regular spatial boundaries. Interestingly, the H2 model produced multiple transitions between concordant and discordant alternans during pacing-down (Supplementary Figure S1). However, such a high number of transitions are not observed in experimental activation maps, suggesting that the H2 model is not the optimal choice. The optimal match was finally obtained with the H3 model, capable of recovering a consistent number of CA-DA transitions and complex alternans patterns, i.e., irregular nodal line shape (Figure 6D). The accuracy of the model was also checked by comparing the FFI phase maps computed at f = f p /2 (f $\frac{1}{2}$). In particular, the π-out-of-phase regions of the simulated tissue recovered the shapes obtained with the standard ΔAPD analysis. As detailed in previous paragraphs, such a phase shift is typical of 2:2 alternans. This result shows the applicability of the FFI phase maps on in silico data as well to reveal DA spatial patterns and also confirms the accuracy of the data assimilation model in reproducing experimental activation maps.
**FIGURE 6:** *Data assimilation procedure and cardiac alternans maps for stimulation at the base of the ventricle. (A) Spatial map of the Fourier spectrum at f = 0.5 Hz and tissue boundary. (B) Computed heterogeneity maps from the tissue ultrastructure (see the text) for both diffusivity, H
1 (x, y), and time constants regulating the APD, H
2 (x, y). Black dashed lines represent one-dimensional cuts of the heterogeneity maps (top panels). (C) Modeled alternans maps: homogeneous case, heterogeneity in diffusivity (H1 model), and heterogeneity in APD (H2 model). (D) Comparison between the modeled alternans map, with combined heterogeneities in APD-regulating time constants and diffusivity (H3 model), and experimental alternans. Top row: ΔAPD maps. Bottom row: FFI phase maps at f = f
p
/2 (f
1/2). Alternans maps are obtained at a pacing frequency f
p
= 6.2 ± 0.4 Hz.*
## 3.5.2 Anterior Right Ventricle Stimulation
We further investigated the H3 model behavior in response to anterior right ventricle stimulation. The adopted heterogeneity maps for this case are shown in Figure 7A. During pacing-down, the model reproduced both CA and DA alternans patterns as well as multiple transitions between the two regimes. Figure 7B shows simulation results at two pacing frequencies, f $$p \leq 4.0$$ Hz and f $$p \leq 5.6$$ Hz, corresponding to two representative cases of CA and DA maps characterized by complex alternans patterns. Also in this case, FFI phase maps at f $\frac{1}{2}$ extracted from simulated data are in agreement with the ΔAPD maps and further verify the accuracy of the method in grasping both CA and DA patterns. In particular, CA FFI phase maps show a less severe phase shift compared to the DA case (less than π). Indeed, in Figure 7, a change in phase around the blue–yellow transition denotes a minimal phase shift, given the 2π-periodicity. In contrast, a phase shift of ≃ π arises in the case of DA patterns. As shown in Figure 7C, simulated maps are in close agreement with the experimental ones in terms of both ΔAPD and FFI phase, and similar spatial alternans shapes are recovered both for CA and for DA.
**FIGURE 7:** *Data assimilation and cardiac alternans maps for right ventricle anterior stimulation. (A) Heterogeneity maps for both diffusivity, H
1 (x, y), and time constants regulating the APD, H
2 (x, y). (B) Modeled alternans maps at pacing frequencies f
p
= 4.0 Hz and 5.6 Hz and corresponding FFI phase maps at f = f
p
/2 (f
1/2). (C) Experimental alternans corresponding to modeled maps shown in panel (B) and corresponding experimental FFI phase maps at f
1/2.*
## 4 Conclusion
We have shown that single-pixel Fourier imaging of high-frequency entrained intact canine RV preparations is a valuable tool to visualize action potential alternans. Besides 2:2 DA, as observed in stable spiral waves in vitro (Hörning et al., 2017), we have also shown that higher-order DA, e.g., 4:4, can be observed and analyzed in an ex vivo heart preparation (Gizzi et al., 2013, 2017). This indicates the possibility of fast and reliable full heart analysis in vivo to detect electrical instabilities in cardiac tissues and thus enables the application to the medical field. The unnecessity of spatial filtering of the recorded signals further opens the possibility of detecting ultra-fine structured early alternans that is only restricted by the optical recording device. Contrarily, Fourier imaging needs a specific time window of periodic oscillations to fully take advantage of the Fourier analysis. Subsequent action potentials cannot be compared and visualized as for the established analysis of action potential duration difference, i.e., ΔAPD (see Figure 4B). So, depending on the purpose, Fourier imaging is a powerful alternative to detect and visualize alternans.
A second useful application is the use of the optical ultrastructure that can be extracted in the low-frequency regime in the Fourier space (see Figure 3). As the ultrastructure is related to the morphological properties of the tissue, we assimilated this frequency and pacing site–independent structure to recover alternans in silico. Using a phenomenological model tuned on CV and restitution curves (Fenton et al., 2013), we were able to reproduce strikingly similar CA and DA patterns as we have observed experimentally. In this context, experimental tissue heterogeneities included in the model could induce CA–DA transitions and complex shapes of alternating tissue areas and nodal boundary lines, not recovered in the fine-tuned homogeneous model. Furthermore, our analysis proved the FFI method to be a practical approach to uncover alternans on in silico data, showing phase maps in close agreement with ΔAPD dispersion.
Pros and cons of the present study shall be mentioned. As for the data assimilation, alternative methods can be used for parameter inference. Genetic algorithms or variational approaches aim at fitting recorded spatiotemporal cardiac activity targeting diffusive properties encoded in the conductivity tensor (Cairns et al., 2017; Barone et al., 2020b; Irakoze and Jacquemet, 2021). If these methods mostly work in the time domain, the data assimilation technique here proposed focuses on the frequency domain instead. It allows, in fact, to account for changes in cardiac tissue properties not considered in other parameter fitting techniques. We believe that our method, combined with other procedures, can enrich data assimilation toward customized models with high predictive power. The present numerical model, in fact, was limited to two-dimensional computational domains (though based on ventricular geometries). An additional level of predictability is expected to appear once the whole ventricular thickness is considered. In such a scenario, the mathematical characterization of intramural rotational anisotropy, combined with a surface-based FFI data assimilation, may open the path toward a multiscale assessment and control of alternans, as well as to scale-transitioning information theories (Garzón et al., 2009; Ashikaga and James, 2018).
We remark that various approaches could be used to derive heterogeneity fields from the FFI spectrum. Indeed, slight variations in the selected frequency or alternative transformation laws can lead to different H (x, y) maps. In the present study, we performed a specific choice considering the invariance of the emergent FFI spatial structure and the non-random organization of the amplitude dispersion. Besides, the adopted scaling can be interpreted as a “perturbed” version of the homogeneous model allowing investigating heterogeneity effects without additional parameter optimization. Although different interpretations of FFI peaks and valleys can be pursued to derive optimized heterogeneity maps and maximize data assimilation, we remark that the present method is generally applicable to multiple cardiac surfaces (endocardium–epicardium, atria–ventricles) and integrated with both biophysical and phenomenological models. Furthermore, one can sort parameters other than diffusivity and APD-regulating time constants based on heterogeneity fields and pursue different assumptions on their correlation. In this context, we assumed that diffusivity and APD-regulating time constants followed correlated spatial heterogeneity profiles. Accordingly, we developed our investigation on this hypothesis as a first explorative study on the effect of a frequency-based data assimilation procedure on cardiac modeling. We hope our study could be further tested and validated in future numerical analyses.
Tissues undergoing fluorescence optical mapping are inherently wet, and they must be kept without drying out to retain physiologically realistic activity. The wet tissue reflects directional light into the camera, causing bright patches in the image known as specular reflection. Regions with specular reflection do not contain information on the tissue texture. Furthermore, these bright spots could produce unrealistic distortion due to the change in angle between the surface and the light source during small residual deformations. On the contrary, diffusion only contains the wavelengths that were not absorbed by the tissue and therefore carry texture information. In such a perspective, including specialized lighting setups would concur to reduce specular reflection. In particular, a cross-polarized lighting setup may provide the best quality images with the least specular reflection and most detailed textures (Lentle and Hulls, 2018). The appearance of optical ultrastructure further connects the present study with a major and multidisciplinary research effort in high-resolution imaging of large biological tissues (Kuruppu et al., 2021).
To conclude, the FFI method outlined in this contribution represents a new and effective method to investigate alternans onset and development in whole-ventricle optical experiments. Accordingly, it can be potentially applied to both calcium and voltage data and does not require excessive pre-analysis, such as the APD-based approaches. Moreover, we have shown that spectral analysis of experimental data at low frequencies can be used to uncover invariant and coherent spatial structures associated with the underlying cardiac tissue properties—ultrastructure—which can serve as input for data assimilation in numerical simulations.
## Data Availability Statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics Statement
The animal study was reviewed and approved by the Institutional Animal Care and Use Committee of the Center for Animal Resources and Education at Cornell University.
## Author Contributions
AL, MH, and AG conceived the study. FF and AG conducted the experiments. JE and MH conceived and conducted the data analysis. AL conceived and conducted the numerical study. SF provided facilities and infrastructure. AL, JE, MH, and AG drafted the original manuscript. All authors contributed to the article and approved the submitted version.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors, and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary Material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnetp.2022.866101/full#supplementary-material
## References
1. Adam D. R., Smith J. M., Akselrod S., Nyberg S., Powell A. O., Cohen R. J.. **Fluctuations in T-Wave Morphology and Susceptibility to Ventricular Fibrillation**. *J. Electrocardiol.* (1984) **17** 209-218. DOI: 10.1016/s0022-0736(84)80057-6
2. Ashikaga H., James R. G.. **Inter-Scale Information Flow as a Surrogate for Downward Causation that Maintains Spiral Waves**. *Chaos* (2018) **28** 075306. DOI: 10.1063/1.5017534
3. Badeer H., Ryo U., Gassner W., Kass E., Cavaluzzi J., Gilbert J.. **Factors Affecting Pulsus Alternans in the Rapidly Driven Heart and Papillary Muscle**. *Am. J. Physiology-Legacy Content* (1967) **213** 1095-1101. DOI: 10.1152/ajplegacy.1967.213.5.1095
4. Barone A., Carlino M. G., Gizzi A., Perotto S., Veneziani A.. **Efficient Estimation of Cardiac Conductivities: A Proper Generalized Decomposition Approach**. *J. Comput. Phys.* (2020a) **423** 109810. DOI: 10.1016/j.jcp.2020.109810
5. Barone A., Gizzi A., Fenton F., Filippi S., Veneziani A.. **Experimental Validation of a Variational Data Assimilation Procedure for Estimating Space-Dependent Cardiac Conductivities**. *Comput. Methods Appl. Mech. Eng.* (2020b) **358** 112615. DOI: 10.1016/j.cma.2019.112615
6. Bueno-Orovio A., Cherry E. M., Fenton F. H.. **Minimal Model for Human Ventricular Action Potentials in Tissue**. *J. Theor. Biol.* (2008) **253** 544-560. DOI: 10.1016/j.jtbi.2008.03.029
7. Cairns D. I., Fenton F. H., Cherry E. M.. **Efficient Parameterization of Cardiac Action Potential Models Using a Genetic Algorithm**. *Chaos* (2017) **27** 093922. DOI: 10.1063/1.5000354
8. Chen D. D., Gray R. A., Uzelac I., Herndon C., Fenton F. H.. **Mechanism for Amplitude Alternans in Electrocardiograms and the Initiation of Spatiotemporal Chaos**. *Phys. Rev. Lett.* (2017) **118** 168101. DOI: 10.1103/PhysRevLett.118.168101
9. Cherry E. M., Fenton F. H.. **Suppression of Alternans and Conduction Blocks Despite Steep Apd Restitution: Electrotonic, Memory, and Conduction Velocity Restitution Effects**. *Am. J. physiology-Heart circulatory Physiol.* (2004) **286** H2332-H2341. DOI: 10.1152/ajpheart.00747.2003
10. Cherubini C., Filippi S., Gizzi A., Ruiz-Baier R.. **A Note on Stress-Driven Anisotropic Diffusion and its Role in Active Deformable Media**. *J. Theor. Biol.* (2017) **430** 221-228. DOI: 10.1016/j.jtbi.2017.07.013
11. Christoph J., Chebbok M., Richter C., Schröder-Schetelig J., Bittihn P., Stein S.. **Electromechanical Vortex Filaments during Cardiac Fibrillation**. *Nature* (2018) **555** 667-672. DOI: 10.1038/nature26001
12. Clusin W. T.. **Mechanisms of Calcium Transient and Action Potential Alternans in Cardiac Cells and Tissues**. *Am. J. Physiology-Heart Circulatory Physiol.* (2008) **294** H1-H10. DOI: 10.1152/ajpheart.00802.2007
13. Cusimano N., Gerardo-Giorda L., Gizzi A.. **A Space-Fractional Bidomain Framework for Cardiac Electrophysiology: 1d Alternans Dynamics**. *Chaos* (2021) **31** 073123. DOI: 10.1063/5.0050897
14. Cusimano N., Gizzi A., Fenton F. H., Filippi S., Gerardo-Giorda L.. **Key Aspects for Effective Mathematical Modelling of Fractional-Diffusion in Cardiac Electrophysiology: A Quantitative Study**. *Commun. Nonlinear Sci. Numer. Simulation* (2020) **84** 105152. DOI: 10.1016/j.cnsns.2019.105152
15. Diaz M. E., O’Neill S. C., Eisner D. A.. **Sarcoplasmic Reticulum Calcium Content Fluctuation Is the Key to Cardiac Alternans**. *Circ. Res.* (2004) **94** 650-656. DOI: 10.1161/01.RES.0000119923.64774.72
16. Dilly S. G., Lab M. J.. **Electrophysiological Alternans and Restitution during Acute Regional Ischaemia in Myocardium of Anaesthetized Pig**. *J. Physiol.* (1988) **402** 315-333. DOI: 10.1113/jphysiol.1988.sp017206
17. Fenton F. H., Cherry M. E., Karma A., Rappel W.-J.. **Modeling Wave Propagation in Realistic Heart Geometries Using the Phase-Field Method**. *Chaos* (2005) **15** 013502. DOI: 10.1063/1.1840311
18. Fenton F. H., Gizzi A., Cherubini C., Pomella N., Filippi S.. **Role of Temperature on Nonlinear Cardiac Dynamics**. *Phys. Rev. E* (2013) **87** 042717. DOI: 10.1103/PhysRevE.87.042717
19. Fenton F. H., Luther S., Cherry E. M., Otani N. F., Krinsky V., Pumir A.. **Termination of Atrial Fibrillation Using Pulsed Low-Energy Far-Field Stimulation**. *Circulation* (2009) **120** 467-476. DOI: 10.1161/CIRCULATIONAHA.108.825091
20. Fox J. J., Riccio M. L., Hua F., Bodenschatz E., Gilmour R. F.. **Spatiotemporal Transition to Conduction Block in Canine Ventricle**. *Circ. Res.* (2002) **90** 289-296. DOI: 10.1161/hh0302.104723
21. Garzón A., Grigoriev R. O., Fenton F. H.. **Model-Based Control of Cardiac Alternans on a Ring**. *Phys. Rev. E* (2009) **80** 021932. DOI: 10.1103/PhysRevE.80.021932
22. Gizzi A., Cherry E. M., Gilmour R. F., Luther S., Filippi S., Fenton F. H.. **Effects of Pacing Site and Stimulation History on Alternans Dynamics and the Development of Complex Spatiotemporal Patterns in Cardiac Tissue**. *Front. Physiol.* (2013) **4** 71. DOI: 10.3389/fphys.2013.00071
23. Gizzi A., Loppini A., Cherry E. M., Cherubini C., Fenton F. H., Filippi S.. **Multi-Band Decomposition Analysis: Application to Cardiac Alternans as a Function of Temperature**. *Physiol. Meas.* (2017) **38** 833-847. DOI: 10.1088/1361-6579/aa64af
24. Green H. D.. **The Nature of Ventricular Alternation Resulting from Reduced Coronary Blood Flow**. *Am. J. Physiology-Legacy Content* (1935) **114** 407-413. DOI: 10.1152/ajplegacy.1935.114.2.407
25. Hörning M., Blanchard F., Isomura A., Yoshikawa K.. **Dynamics of Spatiotemporal Line Defects and Chaos Control in Complex Excitable Systems**. *Sci. Rep.* (2017) **7** 7757. DOI: 10.1038/s41598-017-08011-z
26. Hörning M., Kidoaki S., Kawano T., Yoshikawa K.. **Rigidity Matching between Cells and the Extracellular Matrix Leads to the Stabilization of Cardiac Conduction**. *Biophysical J.* (2012) **102** 379-387. DOI: 10.1016/j.bpj.2011.12.018
27. Huertas M. A., Smith G. D., Györke S.. **Ca2+ Alternans in a Cardiac Myocyte Model that Uses Moment Equations to Represent Heterogeneous Junctional SR Ca2+**. *Biophysical J.* (2010) **99** 377-387. DOI: 10.1016/j.bpj.2010.04.032
28. Hurtado D. E., Castro S., Gizzi A.. **Computational Modeling of Non-linear Diffusion in Cardiac Electrophysiology: A Novel Porous-Medium Approach**. *Comput. Methods Appl. Mech. Eng.* (2016) **300** 70-83. DOI: 10.1016/j.cma.2015.11.014
29. Ikeda T., Sakata T., Takami M., Kondo N., Tezuka N., Nakae T.. **Combined Assessment of T-Wave Alternans and Late Potentials Used to Predict Arrhythmic Events after Myocardial Infarction. A Prospective Study**. *J. Am. Coll. Cardiol.* (2000) **35** 722-730. DOI: 10.1016/s0735-1097(99)00590-2
30. Irakoze É., Jacquemet V.. **Multiparameter Optimization of Nonuniform Passive Diffusion Properties for Creating Coarse-Grained Equivalent Models of Cardiac Propagation**. *Comput. Biol. Med.* (2021) **138** 104863. DOI: 10.1016/j.compbiomed.2021.104863
31. Jakobsen P. J., Johnsen G. H.. **Size-Specific Protection against Predation by Fish in Swarming Waterfleas, Bosmina Longispina**. *Anim. Behav.* (1988) **36** 986-990. DOI: 10.1016/S0003-3472(88)80057-5
32. Jia Z., Bien H., Entcheva E.. **Detecting Space-Time Alternating Biological Signals Close to the Bifurcation point**. *IEEE Trans. Biomed. Eng.* (2010) **57** 316-324. DOI: 10.1109/TBME.2009.2028652
33. Ju Y. S., Martincorena I., Gerstung M., Petljak M., Alexandrov L. B., Rahbari R.. **Somatic Mutations Reveal Asymmetric Cellular Dynamics in the Early Human Embryo**. *Nature* (2017) **543** 714-718. DOI: 10.1038/nature21703
34. Karagueuzian H. S., Stepanyan H., Mandel W. J.. **Bifurcation Theory and Cardiac Arrhythmias**. *Am. J. Cardiovasc. Dis.* (2013) **3** 1-16. PMID: 23459417
35. Karma A.. **Electrical Alternans and Spiral Wave Breakup in Cardiac Tissue**. *Chaos* (1994) **4** 461-472. DOI: 10.1063/1.166024
36. Karma A.. **Spiral Breakup in Model Equations of Action Potential Propagation in Cardiac Tissue**. *Phys. Rev. Lett.* (1993) **71** 1103-1106. DOI: 10.1103/PhysRevLett.71.1103
37. Kim J. W., Pak H.-N., Park J. H., Nam G.-B., Kim S. K., Lee H. S.. **Defibillator Electrogram T Wave Alternans as a Predictor of Spontaneous Ventricular Tachyarrhythmias in Defibrillator Recipients**. *Circ. J.* (2009) **73** 55-62. DOI: 10.1253/circj.cj-08-0311
38. Konta T., Ikeda K., Yamaki M., Nakamura K., Honma K., Kubota I.. **Significance of Discordant ST Alternans in Ventricular Fibrillation**. *Circulation* (1990) **82** 2185-2189. DOI: 10.1161/01.CIR.82.6.2185
39. Kuruppu S., Cheng L. K., Nielsen P. F., Gamage T. B., Avci R., Angeli T. R.. **High-Resolution Spatiotemporal Quantification of Intestinal Motility with Free-form Deformation**. *IEEE Trans. Biomed. Eng.* (2021) 1. DOI: 10.1109/TBME.2021.3135855
40. Lab M. J., Lee J. A.. **Changes in Intracellular Calcium during Mechanical Alternans in Isolated Ferret Ventricular Muscle**. *Circ. Res.* (1990) **66** 585-595. DOI: 10.1161/01.RES.66.3.585
41. Lenarda P., Gizzi A., Paggi M.. **A Modeling Framework for Electro-Mechanical Interaction between Excitable Deformable Cells**. *Eur. J. Mech. - A/Solids* (2018) **72** 374-392. DOI: 10.1016/j.euromechsol.2018.06.001
42. Lentle R. G., Hulls C. M.. **Quantifying Patterns of Smooth Muscle Motility in the Gut and Other Organs with New Techniques of Video Spatiotemporal Mapping**. *Front. Physiol.* (2018) **9** 00338. DOI: 10.3389/fphys.2018.00338
43. Lin J., Keener J. P.. **Modeling Electrical Activity of Myocardial Cells Incorporating the Effects of Ephaptic Coupling**. *Proc. Natl. Acad. Sci. U.S.A.* (2010) **107** 20935-20940. DOI: 10.1073/pnas.1010154107
44. Loppini A., Barone A., Gizzi A., Cherubini C., Fenton F. H., Filippi S.. **Thermal Effects on Cardiac Alternans Onset and Development: A Spatiotemporal Correlation Analysis**. *Phys. Rev. E* (2021) **103** L040201. DOI: 10.1103/PhysRevE.103.L040201
45. Loppini A., Gizzi A., Cherubini C., Cherry E. M., Fenton F. H., Filippi S.. **Spatiotemporal Correlation Uncovers Characteristic Lengths in Cardiac Tissue**. *Phys. Rev. E* (2019) **100** 020201. DOI: 10.1103/PhysRevE.100.020201
46. Luther S., Fenton F. H., Kornreich B. G., Squires A., Bittihn P., Hornung D.. **Low-Energy Control of Electrical Turbulence in the Heart**. *Nature* (2011) **475** 235-239. DOI: 10.1038/nature10216
47. Marcotte C. D., Fenton F. H., Hoffman M. J., Cherry E. M.. **Robust Data Assimilation with Noise: Applications to Cardiac Dynamics**. *Chaos* (2021) **31** 013118. DOI: 10.1063/5.0033539
48. Merchant F. M., Ikeda T., Pedretti R. F. E., Salerno-Uriarte J. A., Chow T., Chan P. S.. **Clinical Utility of Microvolt T-Wave Alternans Testing in Identifying Patients at High or Low Risk of Sudden Cardiac Death**. *Heart Rhythm* (2012) **9** 1256-1264. DOI: 10.1016/j.hrthm.2012.03.014
49. Mitchell J. H., Sarnoff S. J., Sonnenblick E. H.. **The Dynamics of Pulsus Alternans: Alternating End-Diastolic Fiber Length as a Causative Factor**. *J. Clin. Invest.* (1963) **42** 55-63. DOI: 10.1172/JCI104696
50. Miyazawa S., Okamoto M., Kondo S.. **Blending of Animal Colour Patterns by Hybridization**. *Nat. Commun.* (2010) **1** 66. DOI: 10.1038/ncomms1071
51. Muñoz L. M., Gelzer A. R. M., Fenton F. H., Qian W., Lin W., Gilmour R. F.. **Discordant Alternans as a Mechanism for Initiation of Ventricular Fibrillation**. *J. Am. Heart Assoc.* (2018) **7** e007898. DOI: 10.1161/JAHA.117.007898
52. Murray J. D.. **Mathematical Biology II: Spatial Models and Biomedical Applications**. *Interdisciplinary Applied Mathematics* (2003) **18**. DOI: 10.1007/b98869
53. Papadacci C., Finel V., Provost J., Villemain O., Bruneval P., Gennisson J.-L.. **Imaging the Dynamics of Cardiac Fiber Orientation**. *Sci. Rep.* (2017) **7** 830. DOI: 10.1038/s41598-017-00946-7
54. Pastore J. M., Girouard S. D., Laurita K. R., Akar F. G., Rosenbaum D. S.. **Mechanism Linking T-Wave Alternans to the Genesis of Cardiac Fibrillation**. *Circulation* (1999) **99** 1385-1394. DOI: 10.1161/01.CIR.99.10.1385
55. Pathmanathan P., Galappaththige S. K., Cordeiro J. M., Kaboudian A., Fenton F. H., Gray R. A.. **Data-Driven Uncertainty Quantification for Cardiac Electrophysiological Models: Impact of Physiological Variability on Action Potential and Spiral Wave Dynamics**. *Front. Physiol.* (2020) **11** 585400. DOI: 10.3389/fphys.2020.585400
56. Qu Z., Weiss J. N., Garfinkel A.. **Cardiac Electrical Restitution Properties and Stability of Reentrant Spiral Waves: A Simulation Study**. *Am. J. Physiology-Heart Circulatory Physiol.* (1999) **276** H269-H283. DOI: 10.1152/ajpheart.1999.276.1.H269
57. Ren L., Fang X., Wang Y., Qi G.. **T-Wave Alternans and Heart Rate Variability: A Comparison in Patients with Myocardial Infarction with or without Diabetes Mellitus**. *Ann. Noninvasive Electrocardiol.* (2011) **16** 232-238. DOI: 10.1111/j.1542-474X.2011.00437.x
58. Sato D., Shiferaw Y., Garfinkel A., Weiss J. N., Qu Z., Karma A.. **Spatially Discordant Alternans in Cardiac Tissue**. *Circ. Res.* (2006) **99** 520-527. DOI: 10.1161/01.RES.0000240542.03986.e7
59. Shiferaw Y., Watanabe M. A., Garfinkel A., Weiss J. N., Karma A.. **Model of Intracellular Calcium Cycling in Ventricular Myocytes**. *Biophysical J.* (2003) **85** 3666-3686. DOI: 10.1016/S0006-3495(03)74784-5
60. Taggart P., Sutton P. M. I., Boyett M. R., Lab M., Swanton H.. **Human Ventricular Action Potential Duration during Short and Long Cycles**. *Circulation* (1996) **94** 2526-2534. DOI: 10.1161/01.CIR.94.10.2526
61. Tao T., O'Neill S. C., Diaz M. E., Li Y. T., Eisner D. A., Zhang H.. **Alternans of Cardiac Calcium Cycling in a Cluster of Ryanodine Receptors: A Simulation Study**. *Am. J. Physiology-Heart Circulatory Physiol.* (2008) **295** H598-H609. DOI: 10.1152/ajpheart.01086.2007
62. Treml L. M., Bartocci E., Gizzi A.. **Modeling and Analysis of Cardiac Hybrid Cellular Automata via Gpu-Accelerated Monte Carlo Simulation**. *Mathematics* (2021) **9** 164. DOI: 10.3390/math9020164
63. Uzelac I., Ji Y. C., Hornung D., Schröder-Scheteling J., Luther S., Gray R. A.. **Simultaneous Quantification of Spatially Discordant Alternans in Voltage and Intracellular Calcium in Langendorff-Perfused Rabbit Hearts and Inconsistencies with Models of Cardiac Action Potentials and Ca Transients**. *Front. Physiol.* (2017) **8** 819. DOI: 10.3389/fphys.2017.00819
64. van den Heuvel M. P., Hulshoff Pol H. E.. **Exploring the Brain Network: A Review on Resting-State fMRI Functional Connectivity**. *Eur. Neuropsychopharmacol.* (2010) **20** 519-534. DOI: 10.1016/j.euroneuro.2010.03.008
65. Visweswaran R., McIntyre S. D., Ramkrishnan K., Zhao X., Tolkacheva E. G.. **Mechanisms of Calcium Transient and Action Potential Alternans in Cardiac Cells and Tissues**. *J. Cardiovasc. Electrophysiol.* (2013) **24** 1287-1295. DOI: 10.1111/jce.12200
66. Walker M. L., Wan X., Kirsch G. E., Rosenbaum D. S.. **Hysteresis Effect Implicates Calcium Cycling as a Mechanism of Repolarization Alternans**. *Circulation* (2003) **108** 2704-2709. DOI: 10.1161/01.CIR.0000093276.10885.5B
67. Watanabe M. A., Fenton F. H., Evans S. J., Hastings H. M., Karma A.. **Mechanisms for Discordant Alternans**. *J. Cardiovasc. Electrophysiol.* (2001) **12** 196-206. DOI: 10.1046/j.1540-8167.2001.00196.x
68. Weinberg S. H.. **Ephaptic Coupling Rescues Conduction Failure in Weakly Coupled Cardiac Tissue with Voltage-Gated gap Junctions**. *Chaos* (2017) **27** 093908. DOI: 10.1063/1.4999602
69. Xie L.-H., Weiss J. N.. **Arrhythmogenic Consequences of Intracellular Calcium Waves**. *Am. J. Physiology-Heart Circulatory Physiol.* (2009) **297** H997-H1002. DOI: 10.1152/ajpheart.00390.2009
|
---
title: Comparison of Gait Speed Reserve, Usual Gait Speed, and Maximum Gait Speed
of Adults Aged 50+ in Ireland Using Explainable Machine Learning
authors:
- James R. C Davis
- Silvin P. Knight
- Orna A. Donoghue
- Belinda Hernández
- Rossella Rizzo
- Rose Anne Kenny
- Roman Romero-Ortuno
journal: Frontiers in Network Physiology
year: 2021
pmcid: PMC10013005
doi: 10.3389/fnetp.2021.754477
license: CC BY 4.0
---
# Comparison of Gait Speed Reserve, Usual Gait Speed, and Maximum Gait Speed of Adults Aged 50+ in Ireland Using Explainable Machine Learning
## Abstract
Gait speed is a measure of general fitness. Changing from usual (UGS) to maximum (MGS) gait speed requires coordinated action of many body systems. Gait speed reserve (GSR) is defined as MGS–UGS. From a shortlist of 88 features across five categories including sociodemographic, cognitive, and physiological, we aimed to find and compare the sets of predictors that best describe UGS, MGS, and GSR. For this, we leveraged data from 3,925 adults aged 50+ from Wave 3 of The Irish Longitudinal Study on Ageing (TILDA). Features were selected by a histogram gradient boosting regression-based stepwise feature selection pipeline. Each model’s feature importance and input–output relationships were explored using TreeExplainer from the Shapely Additive Explanations explainable machine learning package. The mean Radj2 (SD) from fivefold cross-validation on training data and the Radj2 score on test data were 0.38 (0.04) and 0.41 for UGS, 0.45 (0.04) and 0.46 for MGS, and 0.19 (0.02) and 0.21 for GSR. Each model selected features across all categories. Features common to all models were age, grip strength, chair stands time, mean motor reaction time, and height. Exclusive to UGS and MGS were educational attainment, fear of falling, Montreal cognitive assessment errors, and orthostatic intolerance. Exclusive to MGS and GSR were body mass index (BMI), and number of medications. No features were selected exclusively for UGS and GSR. Features unique to UGS were resting-state pulse interval, Center for Epidemiologic Studies Depression Scale (CESD) depression, sit-to-stand difference in diastolic blood pressure, and left visual acuity. Unique to MGS were standard deviation in sustained attention to response task times, resting-state heart rate, smoking status, total heartbeat power during paced breathing, and visual acuity. Unique to GSR were accuracy proportion in a sound-induced flash illusion test, Mini-mental State Examination errors, and number of cardiovascular conditions. No interactions were present in the GSR model. The four features that overall gave the most impactful interactions in the UGS and MGS models were age, chair stands time, grip strength, and BMI. These findings may help provide new insights into the multisystem predictors of gait speed and gait speed reserve in older adults and support a network physiology approach to their study.
## Introduction
Gait speed is a measure of general fitness (Wu and Zhao, 2021); faster gait speed is associated with the ability to meet occupational demands in younger adults (Aldridge et al., 2020), whilst slower gait speed is associated with functional decline and morbidity in older adults (Bohannon, 1992; Kawajiri et al., 2019). Even though usual (or comfortable) gait speed (UGS) and maximum gait speed (MGS) are significantly intercorrelated (Kollen et al., 2006), changing from comfortable to maximum speed requires a general effort across many body systems. The difference between these two gait speeds has been referred to as walking speed reserve or gait speed reserve (GSR) (Noguerón García et al., 2020).
UGS is a commonly measured gait characteristic in clinical practice and has well-established associations with age (Samson et al., 2001), physical function (James et al., 2020), and frailty (O’Donoghue et al., 2021). On the other hand, MGS has been associated with physical and cognitive function (Umegaki et al., 2018; Aldridge et al., 2020). Gait speed reserve (GSR) may be a useful proxy measure of physiological reserve in humans. For example, some studies have suggested that in community-dwelling older adults, the simultaneous consideration of both usual and maximum gait speed could increase the specificity of the identification of frailty (Noguerón García et al., 2020; do Carmo Correia de Lima et al., 2019). The health associations of these three modalities of gait speed (UGS, MGS, and GSR) are somewhat different, but to our knowledge, there have been no systematic attempts to model predictors of GSR in a large representative sample of community-dwelling older adults where many demographic, anthropometric, and clinical features are measured across multiple physiological systems.
In this study, we aimed to use machine learning to, first, identify the set of features that, from a shortlist of features, best describe UGS, MGS, and GSR. The shortlist of features was theory-driven and not purely exploratory; that is, we selected features that might have physiological plausibility. Then, using explainable machine learning methods, we investigated the selected models for UGS, MGS, and GSR to observe how each feature in the model was associated with the output in a non-parametric manner. With the selected features and visualizations of the input–output relationships, we then discussed the clinical interpretations with respect to the cohort used and the hypothesis that UGS, MGS, and GSR are multisystem phenomena. To touch on the relevance of the gait speed variables with respect to clinical associations, a brief exploration of differences in gait speeds between fallers/non-fallers and fainters/non-fainters was included.
The paper is structured as follows: first, we describe the TILDA study and the methods used to collect data on the shortlisted features; next, the methods used to compare fallers/non-fallers and fainter/non-fainters are described; then, we describe the feature selection pipeline and provide an overview of the histogram gradient boosting regression machine learning model employed. The Materials and Methods section ends with a brief description of the Shapley additive explanations (SHAP) package used to explain the models. The results from feature selection, and SHAP interpretation are then presented separately for each of the three models: UGS, MGS, and GSR. Finally, the discussion and conclusions compare the results of each model, and comment on the potential clinical relevance.
## Design and Setting
We analysed data from adults aged 50+ from Wave 3 of TILDA, a population-based longitudinal study of ageing (https://tilda.tcd.ie/). TILDA study design, and the full cohort profile, have been previously described in detail (Kearney et al., 2011a; Donoghue et al., 2018). Wave 3 data collection took place in 2014 and 2015 and included a computer-assisted personal interview conducted by social interviewers in the participants’ home, a self-completion questionnaire completed in the participants’ own time, and a detailed suite of technology-aided health assessments conducted by trained research nurses at a dedicated health assessment centre. Ethical approval was obtained from the Faculty of Health Sciences Research Ethics Committee at Trinity College Dublin, Ireland (Reference: Main Wave 3 Tilda Study; approval date: June 9, 2014). All participants provided written informed consent, and all data collection procedures adhered to the World Medical Association Declaration of Helsinki on ethical principles for medical research involving human subjects.
## Analytical Sample
The primary analytical sample consisted of participants from TILDA Wave 3 aged 50 years or more who had data for both UGS and MGS.
## Gait Speed Measures
At Wave 3 of TILDA, gait speed was measured as part of a health centre assessment. Measurements in units of cm/s were made using a 4.88-m computerized walkway (GAITRite, CIR Systems, NY, United States). A 2-m space before and after the walkway was used for acceleration and deceleration. Participants were first asked to walk at their normal (usual) pace, UGS, and then as fast as they safely could, MGS. Two walking trials were obtained in each condition, and the mean value for each was used in this analysis. GSR was defined as MGS–UGS.
## Falls and Faints
To put UGS, MGS, and GSR into clinical context, we assessed their correlation with both historical and future falls and syncope.
Historical fallers/fainters were defined as participants who reported at least one fall/faint in the year prior to Wave 3.
Future fallers/fainters were defined as those who reported at least one fall/faint between Wave 3 and Wave 5 (approximately 4 years later).
Each of those four variables were binary categorical with occurrence of falls/faints coded as “1” and absence coded as “0.”
## Shortlisted Features
A shortlist of features from the TILDA Wave 3 dataset was manually curated by the lead author (JD, trained in a STEM discipline), in consultation with co-authors representing both STEM (BH) and health/medical (SK, OD, and RRO) fields. The features were chosen based on known or plausible associations with the three gait speed modalities under investigation. The feature curation considered features from the following five categories: socio-demographics/anthropometrics/medical history, cardiovascular system, physical strength, senses, and cognitive/psychological.
## Socio-Demographics/Anthropometrics/Medical History
Demographic information included Age in years, Sex (male = 0; female = 1), and level of educational attainment (Edu3): either primary/none (Edu3 = 1), secondary (Edu3 = 2), or tertiary/higher (Edu3 = 3).
Anthropometrics comprised Weight (kg), Height (cm), body mass index (BMI, kg/m2), and waist-to-hip ratio (WaistHipRatio, waist circumference/hip circumference) (Nolan et al., 2016).
Medical history: number of cardiovascular diseases (nCVD, from the following list: hypertension, angina, heart attack, congestive heart failure, diabetes, stroke, transient ischaemic attack, high cholesterol, heart murmur, abnormal heart rhythm, atrial fibrillation), if taking any Antidepressant medications (binary), if taking any Antihypertensive medications (binary), and the total number of medications being taken excluding supplements (nMeds) (TILDA, 2016).
Smoking status (Smoker) was categorized as: never (smoker = 0), past (smoker = 1), and current (smoker = 2). Alcohol intake was scored with the CAGE scale (Bush et al., 1987).
The number of reported difficulties with activities of daily living were also assessed. The six basic activities (ADLs) were dressing, including putting on shoes and socks; walking across a room; bathing or showering; eating, such as cutting up food; getting in or out of bed; and using the toilet, including getting up or down. The six independent activities (IADLs) were preparing a hot meal, doing household chores (laundry, cleaning), shopping for groceries, making telephone calls, taking medications, and managing money such as paying bills and keeping track of expenses (Romero-Ortuno et al., 2019).
## Cardiovascular System
During the TILDA Wave 3 health assessment, resting-state (RS) cardiovascular measurements were made during an approximately 10-min window in which the participant was lying supine in a comfortably lit room at an ambient temperature of between 21°C and 23°C. The full TILDA active stand protocol in which the resting-state window takes place has been detailed elsewhere (Finucane et al., 2014; Knight et al., 2020; Donoghue et al., 2021). Throughout the RS, participants underwent non-invasive continuous haemodynamic monitoring using a Finometer MIDI device (Finapres Medical Systems BV, Amsterdam, Netherlands). All RS parameters selected for the shortlist are mean values from the last minute of supine rest (Knight et al., 2020). Haemodynamic parameters were systolic blood pressure (sBP_RS), diastolic blood pressure (dBP_RS), mean arterial pressure (MAP_RS) all in units of mmHg, heart rate (HR_RS) in bpm, stroke volume (StrokeVolume_RS) in ml, left ventricular ejection time (LVET_RS) in ms, pulse interval (PulseInterval_RS) in ms, maximum slope (Maxslope_RS) in mmHg/s, cardiac output (CardiacOutput_RS) in L/min, and total peripheral resistance (TPR_RS) in dyn⋅s⋅cm−5. A near-infrared spectroscopy (NIRS) device, attached over the participants’ left frontal lobe area, was also employed during the RS, and the following cerebral oxygenation features were extracted, again as the mean values from the final minute of rest: oxygenated haemoglobin concentration (O2Hb_RS) and deoxygenated haemoglobin concentration (HHb_RS) both in units of μmol/L, and tissue saturation index (TSI_RS) as a percentage (Knight et al., 2020). Previously derived sample entropy values for resting sBP (sBP_RS_SampEn), dBP (dBP_RS_SampEn), MAP (MAP_RS_SampEn), heart rate (HR_RS_SampEn), O2Hb (O2Hb_RS_SampEn), HHb (HHb_RS_SampEn), and tissue saturation index (TSI_RS_SampEn) were also shortlisted (Knight et al., 2020). In addition, participants were asked if they experienced dizziness upon standing (PhasicDizziness, yes or no), and this feature was also included in the shortlist.
Resting heart rate variability measures were also shortlisted; these were obtained in two 5-min blocks as detailed elsewhere (Frewen et al., 2013). In short, for each block, participants were lying supine. In the first block, participants were asked to breath spontaneously (free breathing), and in the second block, they were asked to breath according to a pre-recorded set of auditory instructions (paced breathing at a frequency of 0.2 Hz). Measurements were obtained using three-lead electrocardiograms (Medilog Darwin, Oxford Instruments Medical Ltd., United Kingdom). The data were subject to a 0.01–1,000 Hz band-pass filtering before R peak detection was performed with a proprietary software (Pardey and Jouravleva, 2004). The data collection and processing are described in detail elsewhere (Frewen et al., 2013). Time domain features were mean heart rate in bpm, root mean square of successive differences between RR intervals in ms, standard deviation of NN intervals in ms, and difference between maximum and minimum heart rate in bpm, derived for both free (HR_Mean_Free, HR_rMSSD_Free, HR_SDNN_Free, HR_Span_Free) and paced breathing (HR_Mean_Paced, HR_rMSSD_Paced, HR_SDNN_Paced, HR_Span_Paced). The difference between free and paced breathing values was calculated for rMSSD (HR_rMSSD_PacedFreeDiff). In the frequency domain, total spectral power in the 0–0.4 Hz frequency band was measured for both free (HR_TotalPower_Free) and paced breathing (HR_TotalPower_Paced) in units of milliseconds squared (ms2).
sBP, dBP, and HR were also determined in a more conventional manner using a sphygmomanometer in seated (sBP_Seated, dBP_Seated, and HR_Seated) and standing (sBP_Standing, dBP_Standing, and HR_Standing) positions, all with units of mmHg. The difference between seated and standing values were calculated for each of the measures (sBP_SeatStandDiff, dBP_SeatStandDiff, and HR_SeatStandDiff).
Pulse wave velocity (PulseWaveVelocity), a non-invasive measure of arterial stiffness with units of m/s, was also included as a cardiovascular feature. In TILDA, the average of two measurements between the carotid and femoral arteries (in m/s) was obtained using a Vicorder® (SMT medical GmbH & Co., Wuerzburg, Germany). Full details have been described elsewhere (Nolan et al., 2016; Donoghue et al., 2018).
## Physical Strength
Upper and lower body strengths were assessed via grip strength and chair stands time. Grip strength was measured in kg using a hydraulic hand dynamometer (Baseline®, Fabrication Enterprises, Inc., White Plains, NY, United States). The value for grip strength referred to henceforth (GripStrength) is the maximum value from a total of eight measurements with four made on each hand. Lower body strength was assessed using the chair stands test in which the time (in seconds) was recorded for the participants to complete five chair stands as quickly as possible, keeping the arms folded across their chest (ChairStandsTime). Chair height was 46 cm.
## Cognitive and Psychological
Global cognition was assessed using two paper-based assessments: the Montreal Cognitive Assessment (MOCA) (Nasreddine et al., 2005) and the Mini-Mental State Examination (MMSE) (Arevalo-Rodriguez et al., 2015); from these, the number of errors (MOCA_errors and MMSE_errors) were extracted for the feature shortlist. Concentration, cognitive processing, and motor response were assessed using two computer-assisted tasks: the choice reaction task (Chintapalli and Romero-Ortuno, 2021) and the sustained attention to response task (SART) (O'Halloran et al., 2014). The choice reaction task required participants to hold down a central button until an on-screen stimulus (either the word “YES” or “NO”) appeared, at which time they had to press the corresponding button on a keyboard. After pressing either button, participants were then required to return to the central button to continue. This was repeated approximately 100 times. In the SART test, participants watched a screen that displayed the numbers 1–9 sequentially a total of 23 times. A number appeared for 300 ms with an interval of 800 ms between numbers: the entire trial lasts approximately 4 min. Participants were instructed to press a button at the appearance of every number except for a specific number (i.e., 3). We extracted the following features from the choice reaction task: mean and standard deviation of cognitive reaction time (CRT_mean and CRT_SD) and motor response time (MRT_mean and MRT_SD) and the number of correct CRT presses (CRT_correct). CRT is the time taken to release the central button in response to the stimulus; MRT is the time between releasing the central button and pressing the required button. From the SART, we extracted the following: mean and standard deviation of reaction time (SART_mean and SART_SD) and the number of trials in which the participant pressed the button when the number 3 appeared (SART_errors). CRT, MRT, and SART times are all measured in milliseconds.
The psychological domains of depression, anxiety, and loneliness were assessed using the Center for Epidemiologic Studies Depression Scale (CESD), the Hospital Anxiety and Depression Scale—Anxiety subscale (HADSA), and the UCLA Loneliness Scale (UCLA), respectively. Fear of falling (FOF) was determined with a yes or no question (Donoghue et al., 2018).
## Sensory
Visual acuity (VA) was measured using a LogMar chart. VA in the left eye (VisualAcuityLeft), right eye (VisualAcuityRight), and best VA (VisualAcuityBest) were included in this work. VA left and right were in logarithmic units. Best VA was defined as 100−(min([VAleft, VAright])×50. Contrast sensitivity (CS) was measured at five spatial frequencies; in cycles per degree (cpd), they were 1.5 cpd (cs_score_a), 3 cpd (cs_score_b), 6 cpd (cs_score_c), 12 cpd (cs_score_d), and 18 cpd (cs_score_e). The procedures for visual acuity and contrast sensitivity measurements are described in detail elsewhere (Duggan et al., 2017). Self-reported hearing (Hearing_SR) was ascertained by the question: “Is your hearing (with or without a hearing aid): 1. Excellent, 2. Very good, 3. Good, 4. Fair, or, 5. Poor?” Multisensory integration was measured using the Shams sound-induced flash illusion (SIFI) test (Shams et al., 2002). The procedure used in TILDA is described in more detail elsewhere (Hirst et al., 2021), but in short, participants were subjected to a set of beeps and flashes and asked to report how many flashes they perceived. *Five* general flash–beep combinations were presented to the participants: two beeps + two flashes; one beep + one flash; zero beeps + one flash; zero beeps + two flashes; and two beeps + one flash. The flash–beep configurations used in this analysis are the so-called “illusory” two-beep one-flash (2B1F) trials. In 2B1F trials, the flash is synchronous with one of the beeps; the other beep occurred either 70, 150, or 230 ms before (SIFI_2B1F_70, SIFI_2B1F_150, SIFI_2B1F_230) or after (SIFI_2B1F_m70, SIFI_2B1F_m150, and SIFI_2B1F_m230) the flash–beep pair. SIFI susceptibility represented accuracy for judging how many flashes were presented when one flash was presented with two beeps (2B1F). Lower accuracy, judging one flash as two, thus indicates higher SIFI susceptibility and stronger integration. SIFI susceptibility was expressed as proportion correct. As there were two trials per condition, these variables were considered discrete (i.e., participants scored 0, 0.5, or 1 proportion correct) (Hirst et al., 2020).
## Statistical Association Between Gait Speed Modalities and Faller/Fainter Status
The normality of the distribution of the three gait speed variables was determined using the one-sample Kolmogorov–Smirnoff test. All three gait speed variables resulted to be non-normally distributed. Hence, to examine the bivariate associations between UGS, MGS, and GSR and historical and future occurrence of falls and faints, we utilized the non-parametric two-sided independent samples Mann–Whitney U-test.
## Overview of Machine Learning Steps
A general overview of the machine earning steps are as follows. The machine learning regression model employed is called histogram gradient boosting regression. In a stepwise fashion, features are tried out one by one in this model, and the best one is selected; this step is repeated over and over until the model does not get any better. The final model is trained on the set of features that give the best performance. This final model is then passed through an explainable machine learning step whereby a method from the SHAP package called TreeExplainer is used to observe the relationships between each of the features in the model and the output of the model.
## Histogram Gradient Boosting Regression
The regression model employed for this analysis was the histogram gradient boosting regressor (HGBR) from Scikit-learn (Pedregosa et al., 2012) version 0.24. The Scikit-learn implementation is based on Microsoft’s light gradient boosting machines (Ke et al., 2017).
Gradient boosting (Friedman, 2001) is a machine learning technique that builds decision trees sequentially where each one is constructed such that it predicts the residuals from the previous tree. Gradient boosting is a powerful tool that has become the model of choice in many fields and applications (Chen and Guestrin, 2016) and have been shown to outperform deep-learning models where the data are tabular and the features themselves have individual meanings as opposed to data structured in a temporal and/or spatial manner as is the case for problems in image and audio domains (Lundberg et al., 2020). Light gradient boosting machines and histogram gradient boosting is an adaptation of gradient-boosted trees that places feature values into histogram-like bins, which allow for tree split points to be located more efficiently.
The HGBR model inherently supports missing values and categorical data. The support for missing data helps to avoid the need for data imputation or removal of features. The categorical data support avoids the need for dummy variables and one-hot encoding, which can drastically increase the dimensionality of the input feature space.
## Feature Selection Algorithm
All operations were performed using Python 3. The feature selection was executed on the Tinney High Performance Computing Cluster at Trinity College Dublin (https://www.tchpc.tcd.ie/node/1353).
Prior to any feature selection or training of any kind, the data were divided according to an $\frac{80}{20}$ train/test split. From the shortlisted set of 88 features across 5 domains, features were chosen for the final models using an automated stepwise feature selection algorithm. In this algorithm, each feature is individually added to a temporary model that contains all features previously selected for the final model: for the initial round, each temporary model contains a single feature. For each individual temporary model, a hyperparameter tuning is performed in which a fivefold cross-validation (CV) is used on the training data for each set of hyperparameters. The hyperparameter tuning is in the form of a 100-iteration randomized search of a set of predefined hyperparameter distributions:{‘max_iter’: [2000],‘loss’: [‘least_squares’],‘random_state’: [42],‘early_stopping’: [True],‘learning_rate’: loguniform(0.005, 0.1),‘max_leaf_nodes’: randint[2, 10],‘min_samples_leaf’: randint[100,200]} The evaluation metric employed was adjusted R2 (Radj2). This metric is used to avoid the continual increase in R2 that occurs with the addition of new features regardless of whether they significantly increase the variance explained by the model. For each temporary model, the best parameters are chosen based on the mean Radj2 from CV (Radj2¯). From these temporary models, the one that provides the biggest increase in Radj2¯ is chosen to continue with, i.e., the new feature upon which that temporary model is based is added to the final model. Before moving to the selection of the next feature, each feature in the model is removed one by one to check if any of them have become redundant in light of the addition of the newest feature; if the score improves on removing a feature, then that feature is removed from the model.
For the purpose of performance monitoring, on each iteration of the loop, the current best model is fit to the entire training dataset and evaluated on both the training and test sets to give training and test Radj2 scores. These scores do not influence the feature selection.
## SHapley Additive exPlanations Values
In this work, SHapley Additive exPlanations (SHAP) values (Lundberg and Lee, 2017) were employed to assess feature importance and investigate the impact of features on the model output. Specifically, the TreeExplainer method from the SHAP package is used. TreeExplainer is designed for use with tree-based machine learning models and builds interpretations that are theoretically guaranteed to be faithful at the local and global levels (i.e., the level of individual samples and the level of features as a whole) (Lundberg et al., 2020).
Shapley values, from which the SHAP package derives, were presented in the field of cooperative game theory (Shapley et al., 1953). They guarantee a fair distribution of contributions from each feature in a model. However, it is generally NP hard (i.e., complexity and computation time scales exponentially with number of features) to compute them, and as such, they have not been widely utilized. The SHAP package first developed a model agnostic heuristic that allowed for their use. A more recent development allows for exact Shapley values to be computed for tree-based models in a practical, low-order polynomial time. TreeExplainer is designed such that it does not need to compute Shapley values for the entire feature set but instead uses the tree structure to perform the exact computation on smaller feature sets made possible by the tree. Detailed derivations of Shapley values and of the TreeExplainer algorithm (Lundberg et al., 2020) can be found elsewhere, but briefly, Shapley derived these values as a method of attributing worth to each player in a game in a fair way. In coalitional game theory, n players form a grand coalition, S, that has a total worth, ΔS. Each player is representative of an input feature. Each smaller coalition, Q; Q⊂S, has worth ΔQ. A Shapley value is a unique solution that satisfies the following four axioms developed to ensure a fair distribution of worth:1. The sum of contributions from each player equals the total worth of the game.2. If a coalition W not containing player i has a worth equal to that of coalition W in union with player i, then the worth of player i is zero; i.e., the player i did not increase the worth of coalition W.3. If the worth of coalition W in union with player i is equal to the worth of W in union with j, the worth of player j is equal to the worth of player j.4. For players w and x, the contribution of a single feature for the sum of values from players w and x is equal to the sum of the contributions from that feature having values of w and x, where the same is true for any subset Q of features and instances w and x. Said in another way, for a feature, f, and values for f of x and w, the contribution ΔQ(f=x+w) is equal to the contribution of ΔQ((f=x)+(f=w)).
SHAP values are computed for each sample of each feature. This allows for global feature explanations to be constructed from the sample level either visually in the form of SHAP summary plots or as a single value such as mean absolute SHAP value or maximum absolute SHAP value. The nature of SHAP values being true to local impacts of features means that low-frequency, high-impact effects do not go unnoticed. For example, a particular feature might, for most samples, have a low impact; however, for some small subset of samples, the feature might have a very large impact. SHAP interaction values are also readily available that explain the impact of interactions between two features. SHAP values are presented as having a positive or negative impact on the output of the model with respect to the expected model output, i.e., the mean output of the model. Thus, for an individual sample, the SHAP value for a particular feature might be, for example, −2.5; this should be interpreted as the value of that feature for that sample is associated with a model output that is −2.5 units less than the model’s mean output.
All SHAP values shown in the results are for the test data.
## Results
Note on presentation of results: the method for feature selection describes a situation whereby features can be removed from the model if they are made redundant by the addition of new features; this did not occur in any of the models, and as such, all features named henceforth with regard to feature selection are to be understood as features added to the model.
## Analytical Cohort
In TILDA Wave 3, 4,309 participants completed the health centre assessment (Donoghue et al., 2018), where the gait speed tests were conducted. After exclusion of participants aged <50 years or with missing data for either UGS or MGS, there were $$n = 3$$,925 participants, with 2,156 ($55\%$) being female. An analytical sample inclusion flowchart can be seen in Figure 1. The educational attainment breakdown was as follows: third/higher, 1,685 ($43\%$); secondary, 1,571 ($40\%$); and primary/none, 669 ($17\%$). The analytical cohort had a mean (SD) age of 64.5 (7.8) years, UGS of 136.7 (19.2) cm/s, MGS of 171.0 (26.9) cm/s, and GSR of 34.3 (16.6) cm/s.
**FIGURE 1:** *Analytical sample inclusion flowchart.*
## Group Differences in Faller and Fainter Status
Of the Wave 3 participants, $21.3\%$ were historical fallers, $3.8\%$ historical fainters, $31.9\%$ future fallers, and $5.4\%$ future fainters. Table 1 shows the results of the association between UGS, MGS, and these clinical variables. Differences between historical fallers were all statistically significant, with a largest median difference for MGS. Statistical significance of $p \leq 0.05$ was demonstrated in historical fainters for UGS and MGS only, with the largest difference also for MGS. A similar pattern emerged for future fallers and fainters.
**TABLE 1**
| Historical falls and faints | Historical falls and faints.1 | Historical falls and faints.2 | Historical falls and faints.3 | Historical falls and faints.4 | Historical falls and faints.5 | Historical falls and faints.6 | Historical falls and faints.7 | Historical falls and faints.8 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| | Non-fallers [median (IQR)] | Fallers [median (IQR)] | Difference in group median | Mann–Whitney p-value | Non-fainters [median (IQR)] | Fainters [median (IQR)] | Difference in group median | Mann–Whitney p-value |
| UGS (cm/s) | 139.3 (24) | 133.1 (26) | 6.2 | <0.001 | 138.3 (25) | 133.0 (19) | 5.3 | 0.005 |
| MGS (cm/s) | 174.5 (33) | 166.2 (36) | 8.3 | <0.001 | 172.55 (34) | 165.95 (34) | 6.6 | 0.010 |
| GSR (cm/s) | 33.0 (20) | 30.4 (22) | 2.6 | <0.001 | 32.3 (20) | 28.95 (23) | 3.35 | 0.118 |
| Future falls and faints | Future falls and faints | Future falls and faints | Future falls and faints | Future falls and faints | Future falls and faints | Future falls and faints | Future falls and faints | Future falls and faints |
| | Non-fallers [median (IQR)] | Fallers [median (IQR)] | Difference in group median | Mann–Whitney p-value | Non-fainters [median (IQR)] | Fainters [median (IQR)] | Difference in group median | Mann–Whitney p-value |
| UGS (cm/s) | 139.7 (23) | 134.8 (27) | 4.95 | <0.001 | 138.4 (24) | 134.1 (30) | 3.7 | <0.001 |
| MGS (cm/s) | 175.1 (33) | 168.0 (36) | 7.05 | <0.001 | 173.2 (33) | 165.7 (39) | 6.0 | <0.001 |
| GSR (cm/s) | 33.20 (21) | 31.4 (21) | 1.85 | <0.001 | 32.7 (21) | 31.1 (21) | 1.6 | 0.209 |
## Usual Gait Speed
The peak Radj2¯(SD) achieved for the UGS model was 0.38 (0.04) with training and test scores of 0.43 and 0.41, respectively. The expected model output was 136.6 cm/s. The features chosen for the model, in order of selection as per Figure 2, were age, chair stands time, BMI, grip strength, number of medications, resting-state pulse interval, mean motor reaction time, height, depression score, sit-to-stand difference in diastolic blood pressure, and left visual acuity.
**FIGURE 2:** *Visualization of the feature selection process for the usual gait speed model. From left to right on the x-axis, the features are in order of addition to the model. The y-axis shows the dimensionless
Radj2
metric. Mean fivefold cross-validation scores with error bars showing
±
SD are shown in black, train scores in dashed blue, and test scores in dotted red.*
A SHAP summary plot is shown in Figure 3; each point on the x-coordinate represents a samples SHAP value, and its colour signifies the value of the feature for that sample, with light brown being high, black being low, and nan (missing) values appearing grey. On the y-axis, features are arranged from top to bottom in order of decreasing mean absolute SHAP value: chair stands time, age, body mass index, number of medications, grip strength, resting-state pulse interval, height, mean motor reaction time, CESD depression score, difference in seated and standing diastolic blood pressure, and visual acuity in the left eye. The figure suggests that upper limits (light brown) of certain variables (e.g., chair stands time, age, body mass index, and number of medications) are more negatively impactful than their lower limits, which are positively impactful. The opposite is the case for upper limits of grip strength, for example.
**FIGURE 3:** *SHAP summary plot for the final usual gait speed model. Features are ordered from top to bottom by decreasing mean absolute SHAP value. For each feature, each point represents a single sample in the test data. A sample’s x-coordinate displays the SHAP value for that sample with respect to a given feature. The colour of a sample indicates the value of the feature, with light brown being high, black low, and grey missing.*
Scatter plots of SHAP value vs. feature can be seen for all features in Figure 4. SHAP values (left y-axis) vs. input feature value (x- axis) with underlaid histogram (right y-axis shows histogram counts) are shown for each feature in the UGS model. Features are arranged top to bottom and left to right in order of decreasing mean absolute SHAP value. At the zero point on the left y-axis (SHAP value = 0), the corresponding x-coordinate values for that feature are associated with having no impact on the model (i.e., they are associated with the mean model output). The vertical spread observed in the SHAP values vs. input feature plots indicates the presence of interaction effects. Although not chosen for the model, the data points are coloured by sex.
**FIGURE 4:** *SHAP values (left y-axis) vs. input feature value (x- axis) with underlaid histogram (right y-axis showing histogram counts) for each feature in the usual gait speed model. Features are arranged top to bottom and left to right in order of decreasing mean absolute SHAP value. Points are coloured per sex: male is black “+,” and female is orange “x.”*
To further investigate the interaction effects suggested by vertical spreading in Figure 4, a plot (Figure 5) of features ordered by decreasing mean absolute SHAP interaction value was produced; in it, features are ranked from left to right in order of decreasing mean absolute SHAP interaction values (orange dotted line). Also shown in dashed blue are the mean maximum absolute SHAP interaction values, which can highlight the effects of outliers.
**FIGURE 5:** *Features ranked from left to right in order of decreasing mean absolute SHAP interaction values (orange dotted line) for the usual gait speed model. Also shown by blue dashed line are the mean maximum absolute SHAP interaction values, which can highlight the effects of outliers.*
The scatter plots of the top 4 interaction effects in the model (i.e., age, chair stands time, body mass index, and grip strength) are shown in Supplementary Appendix A. In the scatter plots, the points are coloured according to the value of the main interaction feature. The interactions are computed for the features in whatever numerical form they exist in, but for ease of visualization, continuous features are coloured according to what quartile a particular samples value falls in; black indicates that the value is in the lowest quartile and light brown the highest quartile. In each figure, the subplots are ordered from top–left to bottom–right by decreasing mean absolute SHAP interaction value.
## Maximum Gait Speed
The peak Radj2¯(SD) achieved for the MGS model was 0.45 (0.04), with training and test scores of 0.54 and 0.46, respectively. The expected model output was 170.9 cm/s. Features chosen for the model, in order of selection were, age, grip strength, chair stands time, body mass index, education, mean motor reaction time in the choice reaction time test, number of medications, height, the standard deviation of the mean reaction time in the sustained attention to response task, resting-state heart rate, fear of falling, MOCA errors, orthostatic intolerance during active stand, smoking status, total power of the heart rate during paced breathing, the root mean square of successive differences between heartbeats during paced breathing, and best visual acuity. Figure 6 shows the visualization of the feature selection process for this model.
**FIGURE 6:** *Visualization of the feature selection process for the maximum gait speed model. From left to right on the x-axis, the features are in order of addition to the model. The y-axis shows the dimensionless
Radj2
metric. Mean fivefold cross-validation scores with error bars showing
±
SD are shown in black, train scores in dashed blue, and test scores in dotted red.*
In the SHAP summary plot for the MGS model shown in Figure 7, the feature importance ranked in order of decreasing mean absolute SHAP values was age, chair stands time, grip strength, body mass index, height, number of medications, mean motor reaction time in the choice reaction time test, orthostatic intolerance during active stand, education, the standard deviation of the mean reaction time in the sustained attention to response task, fear of falling, MOCA errors, smoking, mean heart rate pre-active stand, the root mean square of successive differences between heartbeats during paced breathing, visual acuity, and total power of the heart rate during paced breathing.
**FIGURE 7:** *SHAP summary plot for the maximum gait speed model. Features are ordered from top to bottom by decreasing mean absolute SHAP value. For each feature, each point represents a single sample in the test data. A sample’s x-coordinate displays the SHAP value for that sample with respect to the given feature. The colour of a sample indicates the value of the feature, with light brown being high, black low, and grey missing.*
Figure 8 shows the SHAP values versus input feature values with underlaid histogram for each feature in the MGS model. Figure 9 shows a plot of features ordered by decreasing mean absolute SHAP interaction value, and Supplementary Appendix B contains the scatter plots of the top four interaction effects in the model.
**FIGURE 8:** *SHAP values (left y-axis) vs. input feature value (x-axis) with underlaid histogram (right y-axis shows histogram counts) for each feature in the maximum gait speed model. Features are arranged top to bottom and left to right in order of decreasing mean absolute SHAP value. Points are coloured per sex: male is black “+,” and female is orange “x.”* **FIGURE 9:** *Features ranked from left to right in order of decreasing mean absolute SHAP interaction values (orange dotted line) for the maximum gait speed model. Also shown by blue dashed line are the mean maximum absolute SHAP interaction values, which can highlight the effects of outliers.*
## Gait Speed Reserve
The peak Radj2¯(SD) achieved for the GSR model was 0.19 (0.02), with training and test scores of 0.22 and 0.21, respectively. The model expected output was 34.2 cm/s. Figure 10 shows the visualization of the feature selection process. In order of selection, the features chosen were mean motor reaction time in the choice reaction time test, grip strength, education, chair stands time, MOCA errors, accuracy proportion in the sound induced flash illusion (two beeps and one flash with stimulus-onset asynchrony of +150 ms), fear of falling, height, age, sex (0 = male; 1 = female), orthostatic intolerance in the active stand test, MMSE errors, and number of cardiovascular conditions.
**FIGURE 10:** *Visualization of feature selection process for gait speed reserve. From left to right on the x-axis, the features are in order of addition to the model. The y-axis shows the dimensionless
Radj2
metric. Mean fivefold cross-validation scores with error bars showing
±
SD are shown in black, train scores in dashed blue, and test scores in dotted red.*
In the SHAP summary plot for the GSR model shown in Figure 11, the feature importance ranked in order of decreasing mean absolute SHAP values was level of educational attainment, grip strength, mean MRT, MOCA errors, age, chair stands time, height, sex, accuracy proportion in the sound induced flash illusion, fear of falling, orthostatic intolerance, MMSE errors, and number of cardiovascular conditions.
**FIGURE 11:** *SHAP summary plot for the final gait speed reserve model. Features are ordered from top to bottom by decreasing mean absolute SHAP value. For each feature, each point represents a single sample in the test data. A sample’s x-coordinate displays the SHAP value for that sample with respect to the given feature. The colour of a sample indicates the value of the feature, with light brown being high, black low, and grey missing.*
Figure 12 shows the SHAP values versus input feature values with underlaid histogram for each feature in the GSR model. The absence of vertical spread in the SHAP vs. feature scatter plots is due to the maximum leaf nodes hyperparameter being set equal to two for the histogram gradient boosting model. This results in there being no interaction terms, since the predictions made by each tree only considered features independently (i.e., a maximum leaf node limit of two means that for a given tree only a single split is made along a single feature).
**FIGURE 12:** *SHAP values (left y-axis) vs. input feature value (x- axis) with underlayed histogram (right y-axis shows histogram counts) for each feature in the gait speed reserve model. Points are coloured per sex: male is black “+,” and female is orange “x.”*
The group mean differences in SHAP values for each feature along with $95\%$ confidence intervals can be seen in Figure 13 for (A) sex, (B) third level education vs. all others, and (C) first/no education vs. all others. For sex, the grip strength feature produced a larger difference in means than sex itself with grip strength having a less positive impact for women. Height, mean MRT, fear of falling, and SIFI accuracy were all significant, and all exhibited a negative mean impact difference. On the other hand, for education, there was a positive group mean difference for women in comparison to men. When comparing third/higher educational attainment to the rest, education itself seemed to make the only significant difference. However, when comparing primary/no educational attainment to secondary and tertiary educational attainment in Figure 13C, we observed several other significant differences other than education: MOCA errors, age, mean MRT, MMSE errors, illusion accuracy, orthostatic intolerance, fear of falling, and number of cardiovascular diseases.
**FIGURE 13:** *Bar graphs showing the group mean differences in SHAP values between subgroups with 95% confidence intervals for each feature in the gait speed reserve model. Panel (A) shows the differences in sex, Panel (B) shows the differences between participants with third/higher level of educational attainment and all others, and Panel (C) shows the differences between participants with first/no level of educational attainment and all others.*
## Summary of Results
Supplementary Appendix C, Table 1 summarizes the scores and features selected for each model. The network between predictors and the three gait variables is visually summarized in Figure 14. The outcomes UGS, MGS, and GSR constitute the main nodes of the network and are represented by white circles. Smaller nodes of different colours represent distinct features and are spatially organised according the following: at the centre, there are the features that have been automatically selected in the models for all three outcomes and, therefore, are common to all the outcomes; externally to the central network, disposed on a peripheral imaginary circle, there are features that are common to just two of the three outcomes; and closely around each outcome, there are the unique feature for that particular outcome. Each link between a feature node and an outcome node represents the impact that feature has on the model output: the line thickness is proportional to the mean absolute SHAP value for that feature in the model scaled according to that model. The colour of the features and correspondent link depends on the subset of features: gradations of blue (from dark blue to turquoise) for socio-demographics/anthropometrics/medical history features, light green for cardiovascular features, yellow-green for physical strength features, gradations of orange and light red for cognitive and psychological domain, and dark red/brown for sensory features.
**FIGURE 14:** *Graphical summary of features selected for the three models. Features unique to a model are shown positioned around that model’s node. Features common to all models are positioned in the central ring of nodes; the legend for those five features is located below the UGS node. The remaining nodes are for features common to two of the models. The thickness of lines connecting feature nodes to model nodes express a normalized mean absolute SHAP value of that feature in that model. The normalization is performed per model to reflect the relative importance of a feature to that specific model. The colour of the features and correspondent link depends on the subset of features: gradations of blue (from dark blue to turquoise) for socio-demographics/anthropometrics/medical history features, light green for cardiovascular features, yellow-green for physical strength features, gradations of orange and light red for cognitive and psychological domain, and dark red/brown for sensory features.*
## Overall Summary of Findings
In the present study, using data from Wave 3 of TILDA, we employed a gradient boosted trees-based stepwise feature selection pipeline for the discovery of clinically relevant predictors of GSR, UGS, and MGS using a shortlist of 88 features across 5 domains. The features selected for the respective models explained MGS and UGS to a greater extent than GSR. As shown in Figure 14, there were common features but also some features unique to each of the three models.
## Model Prediction
Based on model Radj2 values, GSR ($19\%$) was less predictable than MGS ($45\%$) and UGS ($38\%$). Whilst we are not aware of previous published data for comparison with our GSR prediction, a previous study by Bohannon reported linear regression R 2 values of $13\%$ for UGS and $41\%$ for MGS (Bohannon, 1997). Our results agree in that the MGS model yielded a larger prediction score than that of UGS but are comparably superior, especially given the fact that our R 2 unit is adjusted.
## Common Features
Across the three models, there were five common selected features: age, grip strength, chair stands time, mean motor reaction time in the choice reaction time test, and height. The top 4 features with the most impactful interactions (by mean absolute SHAP interaction value) were the same for the UGS and MGS models: age, chair stands time, grip strength, and BMI.
Our results agree with Bohannon’s previous findings that UGS and MGS decline with increasing age (Bohannon, 1997). Other authors have also shown similar findings for UGS (Samson et al., 2001; Romero-Ortuno et al., 2010; Schimpl et al., 2011). As per SHAP value vs. feature plots, increasing age was negatively associated with UGS, MGS, and GSR at ≥68, ≥68, and ≥66 years, respectively. Height is also unsurprising as a common predictor; indeed, taller people have longer legs and can achieve longer strides and higher velocity in any gait modality. Consequently, gait speed is often normalized by height (Bohannon, 1997; Kenny et al., 2013; Kasović et al., 2021).
It is also clinically plausible that higher grip strength (as a marker of upper limb strength) and shorter chair stands time (more representative of lower limb strength) were common determinants of all three performance metrics. Indeed, sarcopenia (low muscle mass and/or strength), of which both grip strength and the five chair stands test are indicative measures (Cruz-Jentoft et al., 2019), has been associated with reduced gait speed and poor functional outcomes in older people (Nishimura et al., 2017; Moreira et al., 2019; Perez-Sousa et al., 2019). In our models, slower chair stands time was associated with a decline in UGS, MGS, and GSR once time increased beyond 14.2, 13, and 10.6 s, respectively; whilst increases in UGS, MGS, and GSR began at values of 13.4, 13, and 10.6 s, respectively. Grip strength of ≤26 kg was associated with slower UGS, MGS, and GSR, whilst grip strengths of ≥35, 27, and 27 kg, respectively, were associated with faster performance. These values for grip strength, whilst interesting from an absolute point of view, have a reduced clinical significance given the large differences in grip strength between men and women. Except for height, the other features relationships to the model output appear quite homogeneous with respect to sex.
Higher mean motor reaction time in the choice reaction time test was associated with lower speed in all three models. In previous research, a shorter CRT has been associated with faster gait speed after adjusting for potential confounders and suggests that, in older adults, engaging more frequently in cognitively stimulating activities may improve neuromotor performance and mobility (Cai et al., 2020). In addition, our results resonate with previous TILDA work utilizing traditional linear statistics showing that participants in the slower MRT group (<250 ms) at Wave 1 seemed to have faster mobility decline as assessed by the timed up and go at Wave 3, 4 years later (Chintapalli and Romero-Ortuno, 2021). Interestingly, in the latter study, the MRT cutoff was set arbitrarily, but in the present study, the negative/positive impact thresholds for UGS, MGS, and GSR were 299, 231, and 229 ms, respectively. Interestingly, the less physically demanding UGS model was only negatively influenced above a relatively slower MRT threshold.
The counter-intuitive results of higher grip strength, quicker chair stands time, and quicker MRT being associated with an increase in UGS when compared to MGS may be revealing underlying determining mechanisms of both acts; MGS may be a more physically determined act than UGS and easier to improve on than UGS.
Common between UGS and MGS models were BMI and number of medications, in the clinically expected directions, i.e., obesity and number of medications had a negative impact on gait speed. As regards obesity, research has suggested that obese adults may select their walking speed to minimize pendular energy transduction, energy cost, and perceived exertion during walking (Fernández-Menéndez et al., 2019). In our UGS and MGS models, a BMI ≥29 kg/m2 had negative impact association. Hypothetically, it is possible that in TILDA, obese individuals equally reduced their UGS and MGS, which could possibly explain why BMI was not a feature in the GSR model. As regards the number of medications, a similar mechanism could apply. In any case, our findings are in keeping with previous research showing that drug interactions may increase the likelihood of gait speed decline amongst older adults (Naples et al., 2016). In our UGS and MGS models, more than two medications had a negative impact association. This is below the usual polypharmacy definition of 5+ medication, and the negative impact association with medications could be related to the underlying health condition rather than due to the medications themselves. Of note, visual acuity featured in both UGS (left) and MGS (best), but not in GSR, which could have a similar underlying reason (i.e., both UGS and MGS equally limited).
There were no features exclusively shared by UGS and GSR, but there were four features in the intersection of MGS and GSR: education, MOCA errors, fear of falling, and orthostatic intolerance. As regards the former two, tertiary education was associated with increased gait speed and primary and secondary levels with a decrease. Greater than three MOCA errors negatively impacted both models. Interestingly, better MOCA performance is associated with higher education (Borda et al., 2019) and places greater emphasis on frontal executive function and attention tasks than the MMSE (Wong et al., 2013). Planning for the MGS task may require greater attention and executive function than performing the UGS task (Umegaki et al., 2018), and this may explain MOCA being related to GSR and MGS. Two or more MMSE errors were associated with GSR decrease.
Analogously, orthostatic intolerance and fear of falling may selectively limit the more demanding MGS task but not the more comfortable UGS task. Orthostatic intolerance can be caused by orthostatic hypotension, which in some studies has been associated with reduced gait speed (Briggs et al., 2020). In addition, orthostatic intolerance can be a feature of vestibular disorders such as benign paroxysmal positional vertigo (BPPV) (Jeon et al., 2013), and research has suggested that the gait characteristics of BPPV can be attributed to an inadequate, cautious gait control (Schniepp et al., 2014), which may preferentially manifest in the MGS task. Fear of falling can also become stronger when facing the MGS task, compared to walking at UGS (Bueno et al., 2019).
## Unique Features
Features exclusive to UGS were depression, diastolic blood pressure drop from sitting to standing, and resting-state pulse interval. Higher levels of depressive symptoms have been associated with worse performance in specific quantitative gait variables in community-residing older adults, including lower velocity (Brandler et al., 2012). In our model, CESD negatively impacted UGS when CESD >2 points.
Similarly, TILDA work showed that slower recovery of BP after standing (systolic and/or diastolic) was independently associated with poorer gait performance (Briggs et al., 2020). On the other hand, a higher pulse interval indicates a higher heart rate variability and a more parasympathetic-driven autonomic cardiac control, which has been associated with healthier states (Abad et al., 2014) and mirrors the fact that, for the UGS model, higher pulse intervals had positive influence. In our model, a baseline pulse interval of 799 ms or less had a negative impact on UGS (this is roughly 75.1 bpm: 60 s/min/0.799 s per beat).
Exclusive to the MGS model was the standard deviation of the mean reaction time in the sustained attention to response task, smoking, the mean heart rate pre-active stand, the total power of the heart rate during paced breathing, and the root mean square of successive differences between heartbeats during paced breathing. In a previous study, community-dwelling participants who displayed poorer sustained attention walked more slowly during both single and dual gait tasks (Killane et al., 2014). In our model, standard deviation of the mean SART reaction time <157.7 ms was associated with slower MGS. Interestingly, research has shown that, in habitual smokers, smoking acutely reduces baseline levels of vagal-cardiac nerve activity and completely resets vagally mediated arterial baroreceptor-cardiac reflex responses (Niedermaier et al., 1993), which could be in keeping with heart rate and heart rate variability features being selected in this model. A baseline heart rate of 67.9 bpm or more had a negative impact on MGS in our model. Comparing this to the pulse interval of 799 ms (equivalent to 75.1 bpm) associated with the beginning of negative impact association in the UGS model, we see that in terms of an increasing heart rate, MGS begins to decline earlier than UGS.
Finally, features exclusive to GSR were accuracy proportion in the sound-induced flash illusion (two beeps and one flash with stimulus-onset asynchrony of +150 ms), sex, MMSE errors, and number of cardiovascular diseases. Male sex was associated with increased GSR, potentially because men may comparatively accelerate more than women during the MGS task. Alternatively, this may also be because the variance explained by the GSR model was relatively low and the effect of sex might disappear when additional features are selected as in other models. One or more cardiovascular diseases was negatively associated with GSR, which is in keeping with the possibility that this type of disease may limit MGS more than UGS. As noted by a previous study (Clark et al., 2013), the difference between UGS and MGS is predominantly dictated by the latter. A notable exclusive associate of GSR was the proportion of accuracy in the sound-induced flash illusion. This can be interpreted in the context that worse visual–somatosensory integration is associated with worse balance in older people (Mahoney et al., 2019) and that an increase in susceptibility to the sound-induced flash illusion during standing relative to sitting was present in fall-prone older adults (Stapleton et al., 2014).
## Strengths of the Methodology and Study
A main strength of the methodology is the use of the histogram gradient boosting regressor machine learning model that bins values for faster computation; offers native support for categorical features without the need for one-hot encoding (dummy variables); has native support for missing values not requiring removal of features/samples or imputation procedures; obviates the need to scale features as it is based on decision trees; allows for non-linear relationships, making no assumptions about underlying structure; and is capable modelling feature interactions. The native support for both categorical features and missing data, together with not needing to perform scaling, reduces the time and effort required during data pre-processing. This is especially useful in the feature selection stage of a study where many features that do not end up in the model would otherwise still have to undergo those pre-processing steps.
The use of a tree-based machine learning model such as HGBR leads to another strength in that it allowed for exploration of the input–output relationships by way of the TreeExplainer explainable machine learning method from SHAP. So far, TreeExplainer is the only SHAP method that allows for exact computation of Shapely values, which, with theoretical grounding in game theory, are used to assess the contributions of features to the model output. SHAP values allow for visualizations of input–output relationships and of the contributions of feature interactions. They can also be used to derive feature importance metrics that are built up from the contributions from each individual sample in the test data.
With the SHAP value versus feature plots, one can recognize the presence of what could be considered as “floor” and “ceiling” effects in the features. This highlights the importance of using non-linear models in this type of research, as even if the relationship observed within the “active” region of the feature is indeed linear, a linear model cannot detect the plateau regions and would instead return a model coefficient that underestimates the effect size in the “active” region. Potential clinical cutoffs and regions of interest for certain features are identifiable, as we have detailed above, making the models highly interpretable for clinicians. Beyond the technical aspects, the visualizations made possible by the explainable machine learning methods are also a strength for the more clinical reader. Having run a complex machine learning model, not only the associations captured between features and model output can be observed but also the relationships between feature interactions and the output. Cutoffs, regions of interest, clusters, and trends are all on show, which can allow for better insight and hypothesis generation.
Another strength of the study is the comparison of UGS, MGS, and GSR in terms of features selected to describe them from a range of 88 features across multiple domains. The TILDA data leveraged allowed for a relatively large sample size of 3,925 participants.
## Limitations of Methodology and Study
However, whilst these cutoffs and regions of interest may be able to inform the clinician, it is possible that they may vary between populations. In terms of the analytical sample, this only included TILDA Wave 3 participants who underwent the health centre assessment where gait speed tests were conducted. Even though at Wave 1, TILDA was designed as a nationally representative cohort of people aged 50 or more years living in Ireland (Donoghue et al., 2018), our analytical sample at Wave 3 is not population representative, and therefore, our results are not necessarily generalizable to the Irish population. Indeed, TILDA work showed that participants attending the health assessment centre were generally fitter than those having a health assessment in their homes (Kearney et al., 2011b), which means that other features may have been selected in the models should frailer people have been included in the analytical sample.
Despite having many advantages, the machine learning methodology also has limitations. The features selected need to be considered in terms of the “package” of features chosen for the final model. Furthermore, it cannot be assumed that features not chosen for a model are not also predictive of the outcome variable.
Even though measures were put in place to help reduce overfitting (cross-validation on training data used in choosing features and hyperparameters, and models evaluated on a held-out test dataset), in the absence of an external validation sample, the risk of overfitting still exists. Despite using held-out test data, the absence of an external validation test means that the generalizability of the results is unknown. The confidence intervals of the effects and associations are also not known in this work; however, application of bootstrapping methods may be used in future work to address this. A rigorous time complexity analysis was not performed, but given its stepwise nature, the computation time of the feature-selection step scales with the square of the number of features considered. The number of hyperparameter iterations and the k-fold cross-validation in place also scale up the computation time. Parallelized code could help to reduce computation time. The computation time can be reduced by the early stopping function that halts the feature selection if there is no improvement or a decline for two consecutive attempts. However, when (or if) this criteria is met depends on the data.
Furthermore, the models are dependent on the predictors that were entered. Even though the “shortlist” of predictors was quite comprehensive (i.e., 88 features across 5 domains), we may not have considered potentially relevant predictors that were either not measured or not shortlisted. In view of GSR being less predictable than UGS and MGS, it is possible that including additional features in the GSR model (perhaps personality/social/lifestyle factors) would improve the model prediction. Height-normalized gait speed could have been considered in the models, but this is not something that we wanted to consider a priori given the data-driven approach.
Another limitation touched on in the discussion is regarding the sex differences in grip strength and height. Height is not too much of an issue, as it is non-modifiable and is a common choice for gait speed normalization, but the thresholds observed in grip strength with respect to positive or negative deviation from the mean in UGS, MGS, or GSR are heavily distorted by sex. A sex-stratified investigation of grip strength in this context may be of clinical benefit given its modifiable nature and its high importance in all three models.
Finally, it must be made clear that despite the use of word “impact” when explaining the relationship of input features to the output, all results are associations and causal relationships cannot be assumed.
## Potential Clinical Relevance
The five features selected for all three models (age, grip strength, chair stands time, mean motor response time, and height) show common factors effecting UGS, MGS, and GSR: age, upper and lower body strength, physical reaction ability, and height.
Whilst there are similarities between the three gait speed models, the differences in features chosen for each model suggest that there are physiological differences in the nature of the three gait variables. This was also suggested in the different clinical associations between the gait speeds and clinical outcomes such as falls and faints. In the domain of psychology and cognition, UGS and MGS differ the most, with UGS being associated with depression, whilst MGS is associated with cognitive performance in the SART and MOCA tests (MOCA was also associated with GSR). Education was associated with MGS and GSR but not with UGS. Fear of falling being present in MGS and GSR but not in UGS could suggest that the fear may not be in relation to usual day-to-day activity and walking but instead towards moving out of comfort zone. The unique presence of MMSE and a sound-induced flash illusion variable in the GSR model could suggest that GSR is related to a cognitive and sensory domain. The sound-induced flash illusion test assesses multisensory integration. It may be possible that UGS is more reflective of baseline health and perhaps is more sensitive to negative health outcomes, leaning more towards the frailty end of the frailty-fitness spectrum (Romero-Ortuno and O'Shea, 2013). MGS and GSR, on the other hand, may reflect more of the fitness end of the spectrum, the ability to go beyond baseline towards better fitness and more reserve but not necessarily less frailty. A potential clinical take away from this work is that modifiable associates could be targeted for a particular gait characteristic with a view to improving the higher-level aspects of health such as frailty or fitness that is more linked to that variable. Given that each of the gait speed variables was predictive of potentially different health outcomes, this work shows avenues for ultimately targeting modifiable predictors of clinically meaningful outcomes.
## Conclusion
The selected variables explained a greater proportion of variation in MGS and UGS than GSR. There were common features to all three models (i.e., age, grip strength, chair stands time, mean motor reaction time in the choice reaction time test, and height) but also some unique features to each of them. By SHAP feature importance, the top 4 features were chair stands time, age, BMI, and number of medications to the UGS model; age, chair stands time, grip strength, and height to the MGS model; and level of educational attainment, grip strength, mean motor response time, and MOCA errors to the GSR model. Overall, findings on all three models were clinically plausible and support a network physiology approach (Bartsch et al., 2015) to the understanding of predictors of performance-based tasks. Each model contains features from multiple physiological systems and thus support the hypothesis that GSR and UGS and MGS are multisystem phenomena. By employing an explainable machine learning model, our observations may help clinicians gain new insights into the possible determinants of physiological reserve in older adults. Of the features selected, some are non-modifiable, e.g., age, sex, and height. Others, however, may be directly modifiable through changes in lifestyle, engaging in physical exercise, or cognitive stimulation (e.g., BMI, weight, smoking, education, chair stands time, grip strength, MOCA, motor response time, and SART). For some variables, it may be useful to focus on ensuring that a patient avoids reaching threshold values that are associated with a rapid decline in gait speed. Conversely, if engaging in rehabilitation, those threshold values may be the targets so as to reach a more stable situation with respect to walking speed. Having explored the predictors of GSR and found multisystem associations, further work will investigate whether GSR is a useful measure in predicting adverse health outcomes and if it can contribute to informing on overall physiological reserve.
The machine learning approach allowed for the stepwise selection of the set features that best explained a target variable in a non-parametric manner that can also capture high-order interactions. Explainable machine learning allowed for the selected models to be visualized to observe the input–output relationships and the relationship between feature interactions and the model output. Using a tree-based machine learning model enabled the use of the TreeSHAP explainable machine learning package, which uses the tree structure to be able to compute exact Shapely values in low-order polynomial time. Bootstrapping will be implemented in future iterations of the method to allow for confidence intervals to be included in the model explanation visualisations.
Future work will use these methods to explore other available gait parameters and physical markers and may include new features from other domains as additional inputs.
## Data Availability Statement
The data analyzed in this study is subject to the following licenses/restrictions: The datasets generated during and/or analysed during the current study are not publicly available due to data protection regulations but are accessible at TILDA on reasonable request. Requests to access these datasets should be directed to https://tilda.tcd.ie/data/accessing-data/.
## Ethics Statement
The studies involving human participants were reviewed and approved by Faculty of Health Sciences Research Ethics Committee at Trinity College Dublin, Ireland. The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
Conceptualization: JD and RR-O. Methodology: JD and RR-O. Software: JD. Formal analysis: JD. Investigation: JD and RR-O. Resources: RK and RR-O. Data curation: JD, SK, OD, and RR-O. Writing—original draft preparation: JD and RR-O. Writing—review and editing: JD, SK, OD, BH, RR, and RR-O. Validation: JD and RR-O. Visualization: JD and RR. Supervision: RR-O. Project administration: RK and RR-O. Funding acquisition: RK and RR-O. All authors have read and agreed to the published version of the manuscript.
## Conflict of Interest
The handling editor is currently organizing a Research Topic with one of the authors RR-O.
The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary Material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnetp.2021.754477/full#supplementary-material
## References
1. Abad C., do Nascimento A., Gil S., Kobal R., Loturco I., Nakamura F.. **Cardiac Autonomic Control in High Level Brazilian Power and Endurance Track-And-Field Athletes**. *Int. J. Sports Med.* (2014) **35** 772-778. DOI: 10.1055/s-0033-1363268
2. Aldridge C., Tringali V., Rhodes R., Kershisnik K., Creditt D., Gonzalez‐Mejia J.. **Walking at Work: Maximum Gait Speed Is Related to Work Ability in Hospital Nursing Staff**. *Jrnl Occup. Health* (2020) **62** e12171. DOI: 10.1002/1348-9585.12171
3. Arevalo-Rodriguez I., Smailagic N., Roqué i Figuls M., Ciapponi A., Sanchez-Perez E., Giannakou A.. **Mini-Mental State Examination (MMSE) for the Detection of Alzheimer's Disease and Other Dementias in People with Mild Cognitive Impairment (MCI)**. *Cochrane Database Syst. Rev.* (2015) CD010783. DOI: 10.1002/14651858.CD010783.pub2
4. Bartsch R. P., Liu K. K. L., Bashan A., Ivanov P. C.. **Network Physiology: How Organ Systems Dynamically Interact**. *PLoS One* (2015) **10** e0142143. DOI: 10.1371/journal.pone.0142143
5. Bohannon R. W.. **Comfortable and Maximum Walking Speed of Adults Aged 20-79 Years: Reference Values and Determinants**. *Age Ageing* (1997) **26** 15-19. DOI: 10.1093/ageing/26.1.15
6. Bohannon R. W.. **Walking after Stroke**. *Int. J. Rehabil. Res.* (1992) **15** 246-248. DOI: 10.1097/00004356-199209000-00009
7. Borda M. G., Reyes-Ortiz C., Pérez-Zepeda M. U., Patino-Hernandez D., Gómez-Arteaga C., Cano-Gutiérrez C. A.. **Educational Level and its Association with the Domains of the Montreal Cognitive Assessment Test**. *Aging Ment. Health* (2019) **23** 1300-1306. DOI: 10.1080/13607863.2018.1488940
8. Brandler T. C., Wang C., Oh-Park M., Holtzer R., Verghese J.. **Depressive Symptoms and Gait Dysfunction in the Elderly**. *Am. J. Geriatr. Psychiatry* (2012) **20** 425-432. DOI: 10.1097/JGP.0b013e31821181c6
9. Briggs R., Donoghue O. A., Carey D., O'Connell M. D. L., Newman L., Kenny R. A.. **What Is the Relationship between Orthostatic Blood Pressure and Spatiotemporal Gait in Later Life**. *J. Am. Geriatr. Soc.* (2020) **68** 1286-1292. DOI: 10.1111/jgs.16379
10. Bueno G. A. S., Gervásio F. M., Ribeiro D. M., Martins A. C., Lemos T. V., de Menezes R. L.. **Fear of Falling Contributing to Cautious Gait Pattern in Women Exposed to a Fictional Disturbing Factor: A Non-randomized Clinical Trial**. *Front. Neurol.* (2019) **10** 283. DOI: 10.3389/fneur.2019.00283
11. Bush B., Shaw S., Cleary P., Delbanco T. L., Aronson M. D.. **Screening for Alcohol Abuse Using the CAGE Questionnaire**. *Am. J. Med.* (1987) **82** 231-235. DOI: 10.1016/0002-9343(87)90061-1
12. Cai Y., Hausdorff J. M., Bean J. F., Manor B., You T., Leveille S. G.. **Participation in Cognitive Activities Is Associated with Foot Reaction Time and Gait Speed in Older Adults**. *Aging Clin. Exp. Res.* (2020). DOI: 10.1007/s40520-020-01583-3
13. Chen T., Guestrin C.. **XGBoost: A Scalable Tree Boosting System**. (2016) 785-794
14. Chintapalli R., Romero-Ortuno R.. **Choice Reaction Time and Subsequent Mobility Decline: Prospective Observational Findings from the Irish Longitudinal Study on Ageing (TILDA)**. *EClinicalMedicine* (2021) **31** 100676. DOI: 10.1016/j.eclinm.2020.100676
15. Clark D. J., Manini T. M., Fielding R. A., Patten C.. **Neuromuscular Determinants of Maximum Walking Speed in Well-Functioning Older Adults**. *Exp. Gerontol.* (2013) **48** 358-363. DOI: 10.1016/j.exger.2013.01.010
16. Cruz-Jentoft A. J., Bahat G., Bauer J., Boirie Y., Bruyère O., Cederholm T.. **Sarcopenia: Revised European Consensus on Definition and Diagnosis**. *Age Ageing* (2019) **48** 16-31. DOI: 10.1093/ageing/afy169
17. Davis J. R. C., Knight S. P., Donoghue O. A., Hernández B., Kenny R. A., Romero-Ortuno R.. **Comparison of Gait Speed reserve, Usual Gait Speed, and Maximum Gait Speed of Adults Aged 50+ in Ireland Using Explainable Machine Learning**. *medRxiv* (2021) 2021. DOI: 10.1101/2021.07.23.21260911
18. do Carmo Correia de Lima M., Loffredo Bilton T., Jefferson de Sousa Soares W., Paccini Lustosa L., Ferriolli E., Rodrigues Perracini M.. **Maximum Walking Speed Can Improve the Diagnostic Value of Frailty Among Community-Dwelling Older Adults a Cross-Sectional Study**. *J. Frailty Aging* (2019) **8** 1-3. DOI: 10.14283/jfa.2018.44
19. Donoghue O. A., McGarrigle C. A., Foley M., Fagan A., Meaney J., Kenny R. A.. **Cohort Profile Update: The Irish Longitudinal Study on Ageing (TILDA)**. *Int. J. Epidemiol.* (2018) **47** 1398-1398l. DOI: 10.1093/ije/dyy163
20. Donoghue O. A., O’Connell M. D. L., Bourke R., Kenny R. A.. **Is Orthostatic Hypotension and Co-existing Supine and Seated Hypertension Associated with Future Falls in Community-Dwelling Older Adults? Results from the Irish Longitudinal Study on Ageing (TILDA)**. *PLoS One* (2021) **16** e0252212. DOI: 10.1371/journal.pone.0252212
21. Duggan E., Donoghue O., Kenny R. A., Cronin H., Loughman J., Finucane C.. **Time to Refocus Assessment of Vision in Older Adults? Contrast Sensitivity but Not Visual Acuity Is Associated with Gait in Older Adults**. *The Journals Gerontol. Ser. A.* (2017) **72** 1663-1668. DOI: 10.1093/gerona/glx021
22. Fernández-Menéndez A., Saubade M., Hans D., Millet G. P., Malatesta D.. **The Determinants of the Preferred Walking Speed in Individuals with Obesity**. *Obes. Facts* (2019) **12** 543-553. DOI: 10.1159/000501968
23. Finucane C., O’Connell M. D. L., Fan C. W., Savva G. M., Soraghan C. J., Nolan H.. **Age-Related Normative Changes in Phasic Orthostatic Blood Pressure in a Large Population Study**. *Circulation* (2014) **130** 1780-1789. DOI: 10.1161/CIRCULATIONAHA.114.009831
24. Frewen J., Finucane C., Savva G. M., Boyle G., Coen R. F., Kenny R. A.. **Cognitive Function Is Associated with Impaired Heart Rate Variability in Ageing Adults: the Irish Longitudinal Study on Ageing Wave One Results**. *Clin. Auton. Res.* (2013) **23** 313-323. DOI: 10.1007/s10286-013-0214-x
25. Friedman J. H.. **Greedy Function Approximation: A Gradient Boosting Machine**. *Ann. Statist.* (2001) **29** 1189-1232. DOI: 10.1214/aos/1013203451
26. Hirst R. J., Setti A., De Looze C., Akuffo K. O., Peto T., Kenny R. A.. **The Effect of Eye Disease, Cataract Surgery and Hearing Aid Use on Multisensory Integration in Ageing**. *Cortex* (2020) **133** 161-176. DOI: 10.1016/j.cortex.2020.08.030
27. Hirst R. J., Whelan R., Boyle R., Setti A., Knight S., O'Connor J.. **Gray Matter Volume in the Right Angular Gyrus Is Associated with Differential Patterns of Multisensory Integration with Aging**. *Neurobiol. Aging* (2021) **100** 83-90. DOI: 10.1016/j.neurobiolaging.2020.12.004
28. James E. G., Conatser P., Karabulut M., Leveille S. G., Hausdorff J. M., Travison T.. **Walking Speed Affects Gait Coordination and Variability Among Older Adults with and without Mobility Limitations**. *Arch. Phys. Med. Rehabil.* (2020) **101** 1377-1382. DOI: 10.1016/j.apmr.2020.04.009
29. Jeon E.-J., Park Y.-S., Park S.-N., Park K.-H., Kim D.-H., Nam I.-C.. **Clinical Significance of Orthostatic Dizziness in the Diagnosis of Benign Paroxysmal Positional Vertigo and Orthostatic Intolerance**. *Am. J. Otolaryngol.* (2013) **34** 471-476. DOI: 10.1016/j.amjoto.2013.04.005
30. Kasović M., Štefan L., Štefan A.. **Normative Data for Gait Speed and Height Norm Speed in ≥ 60-Year-Old Men and Women**. *Cia* (2021) **16** 225-230. DOI: 10.2147/CIA.S290071
31. Kawajiri H., Mishina H., Asano S., Kono Y., Hayashi H., Niwa J.-i.. **Maximum Walking Speed at Discharge Could Be a Prognostic Factor for Vascular Events in Patients with Mild Stroke: A Cohort Study**. *Arch. Phys. Med. Rehabil.* (2019) **100** 230-238. DOI: 10.1016/j.apmr.2018.05.025
32. Ke G., Meng Q., Finley T., Wang T., Chen W., Ma W.. **LightGBM: a Highly Efficient Gradient Boosting Decision Tree**. (2017) 3149-3157
33. Kearney P. M., Cronin H., O'Regan C., Kamiya Y., Savva G. M., Whelan B.. **Cohort Profile: the Irish Longitudinal Study on Ageing**. *Int. J. Epidemiol.* (2011) **40** 877-884. DOI: 10.1093/ije/dyr116
34. Kearney P. M., Cronin H., O'Regan C., Kamiya Y., Whelan B. J., Kenny R. A.. **Comparison of centre and home-based Health Assessments: Early Experience from the Irish Longitudinal Study on Ageing (TILDA)**. *Age Ageing* (2011) **40** 85-90. DOI: 10.1093/ageing/afq124
35. Kenny R. A., Coen R. F., Frewen J., Donoghue O. A., Cronin H., Savva G. M.. **Normative Values of Cognitive and Physical Function in Older Adults: Findings from the Irish Longitudinal Study on Ageing**. *J. Am. Geriatr. Soc.* (2013) **61** S279-S290. DOI: 10.1111/jgs.12195
36. Killane I., Donoghue O. A., Savva G. M., Cronin H., Kenny R. A., Reilly R. B.. **Relative Association of Processing Speed, Short-Term Memory and Sustained Attention with Task on Gait Speed: a Study of Community-Dwelling People 50 Years and Older**. *Journals Gerontol. Ser. A: Biol. Sci. Med. Sci.* (2014) **69** 1407-1414. DOI: 10.1093/gerona/glu140
37. Knight S. P., Newman L., O’Connor J. D., Davis J., Kenny R. A., Romero-Ortuno R.. **Associations between Neurocardiovascular Signal Entropy and Physical Frailty**. *Entropy* (2020) **23** 4. DOI: 10.3390/e23010004
38. Kollen B., Kwakkel G., Lindeman E.. **Hemiplegic Gait after Stroke: Is Measurement of Maximum Speed Required**. *Arch. Phys. Med. Rehabil.* (2006) **87** 358-363. DOI: 10.1016/j.apmr.2005.11.007
39. Lundberg S. M., Lee S-I.. **A Unified Approach to Interpreting Model Predictions**. (2017) 4768-4777
40. Lundberg S. M., Erion G., Chen H., Degrave A., Prutkin J. M., Nair B.. **From Local Explanations to Global Understanding with Explainable AI for Trees**. *Nat. Mach Intell.* (2020) **2** 56-67. DOI: 10.1038/s42256-019-0138-9
41. Mahoney J. R., Cotton K., Verghese J.. **Multisensory Integration Predicts Balance and Falls in Older Adults**. *J. Gerontol. A. Biol. Sci. Med. Sci.* (2019) **74** 1429-1435. DOI: 10.1093/gerona/gly245
42. Moreira V. G., Perez M., Lourenço R. A.. **Prevalence of Sarcopenia and its Associated Factors: the Impact of Muscle Mass, Gait Speed, and Handgrip Strength Reference Values on Reported Frequencies**. *Clinics* (2019) **74** e477. DOI: 10.6061/clinics/2019/e477
43. Naples J. G., Marcum Z. A., Perera S., Newman A. B., Greenspan S. L., Gray S. L.. **Impact of Drug-Drug and Drug-Disease Interactions on Gait Speed in Community-Dwelling Older Adults**. *Drugs Aging* (2016) **33** 411-418. DOI: 10.1007/s40266-016-0373-2
44. Nasreddine Z. S., Phillips N. A., Bã©dirian V. r., Charbonneau S., Whitehead V., Collin I.. **The Montreal Cognitive Assessment, MoCA: a Brief Screening Tool for Mild Cognitive Impairment**. *J. Am. Geriatr. Soc.* (2005) **53** 695-699. DOI: 10.1111/j.1532-5415.2005.53221.x
45. Niedermaier O. N., Smith M. L., Beightol L. A., Zukowska-Grojec Z., Goldstein D. S., Eckberg D. L.. **Influence of Cigarette Smoking on Human Autonomic Function**. *Circulation* (1993) **88** 562-571. DOI: 10.1161/01.cir.88.2.562
46. Nishimura T., Arima K., Okabe T., Mizukami S., Tomita Y., Kanagae M.. **Usefulness of Chair Stand Time as a Surrogate of Gait Speed in Diagnosing Sarcopenia**. *Geriatr. Gerontol. Int.* (2017) **17** 659-661. DOI: 10.1111/ggi.12766
47. Noguerón García A., Huedo Ródenas I., García Molina R., Ruiz Grao M. C., Avendaño Céspedes A., Esbrí Víctor M.. **Gait Plasticity Impairment as an Early Frailty Biomarker**. *Exp. Gerontol.* (2020) **142** 111137. DOI: 10.1016/j.exger.2020.111137
48. Nolan H., Newman L., Donoghue O.. **Chapter 5: Objective Indicators of Health and Function (TILDA Wave 3 Report)**. (2016)
49. O'Halloran A. M., Finucane C., Savva G. M., Robertson I. H., Kenny R. A.. **Sustained Attention and Frailty in the Older Adult Population**. *Journals Gerontol. Ser. B: Psychol. Sci. Soc. Sci.* (2014) **69** 147-156. DOI: 10.1093/geronb/gbt009
50. O’Donoghue P. J., Claffey P., Rice C., Byrne L., Cunningham C., Kenny R. A.. **Association between Gait Speed and the SHARE Frailty Instrument in a Falls and Syncope Clinic**. *Eur. Geriatr. Med.* (2021) **12** 1101-1105. DOI: 10.1007/s41999-021-00509-0
51. Pardey J., Jouravleva S.. *The Next-Generation Holter Revolution: From Analyse-Edit-Print to Analyse-Print* (2004)
52. Pedregosa F., Varoquaux G., Gramfort A., Michel V., Thirion B., Grisel O.. **Scikit-learn: Machine Learning in Python**. *J. Machine Learn. Res.* (2012) **12** 12
53. Perez‐Sousa M. A., Venegas‐Sanabria L. C., Chavarro‐Carvajal D. A., Cano‐Gutierrez C. A., Izquierdo M., Correa‐Bautista J. E.. **Gait Speed as a Mediator of the Effect of Sarcopenia on Dependency in Activities of Daily Living**. *J. Cachexia, Sarcopenia Muscle* (2019) **10** 1009-1015. DOI: 10.1002/jcsm.12444
54. Romero-Ortuno R., Cogan L., Cunningham C. U., Kenny R. A.. **Do older Pedestrians Have Enough Time to Cross Roads in Dublin? A Critique of the Traffic Management Guidelines Based on Clinical Research Findings**. *Age and Ageing* (2010) **39** 80-86. DOI: 10.1093/ageing/afp206
55. Romero-Ortuno R., O'Shea D.. **Fitness and Frailty: Opposite Ends of a Challenging Continuum! Will the End of Age Discrimination Make Frailty Assessments an Imperative**. *Age and Ageing* (2013) **42** 279-280. DOI: 10.1093/ageing/afs189
56. Romero-Ortuno R., Scarlett S., O’Halloran A. M., Kenny R. A.. **Is Phenotypical Prefrailty All the Same? A Longitudinal Investigation of Two Prefrailty Subtypes in TILDA**. *Age Ageing* (2019) **49** 39-45. DOI: 10.1093/ageing/afz129
57. Samson M. M., Crowe A., de Vreede P. L., Dessens J. A. G., Duursma S. A., Verhaar H. J. J.. **Differences in Gait Parameters at a Preferred Walking Speed in Healthy Subjects Due to Age, Height and Body Weight**. *Aging Clin. Exp. Res.* (2001) **13** 16-21. DOI: 10.1007/BF03351489
58. Schimpl M., Moore C., Lederer C., Neuhaus A., Sambrook J., Danesh J.. **Association between Walking Speed and Age in Healthy, Free-Living Individuals Using Mobile Accelerometry-A Cross-Sectional Study**. *PLoS One* (2011) **6** e23299. DOI: 10.1371/journal.pone.0023299
59. Schniepp R., Wuehr M., Huth S., Pradhan C., Brandt T., Jahn K.. **Gait Characteristics of Patients with Phobic Postural Vertigo: Effects of Fear of Falling, Attention, and Visual Input**. *J. Neurol.* (2014) **261** 738-746. DOI: 10.1007/s00415-014-7259-1
60. Shams L., Kamitani Y., Shimojo S.. **Visual Illusion Induced by Sound**. *Cogn. Brain Res.* (2002) **14** 147-152. DOI: 10.1016/s0926-6410(02)00069-1
61. Shapley L. S., Harold William K., Albert William T.. *17 A Value for N-Person Games* (1953) 307-318. DOI: 10.1515/9781400881970-018
62. Stapleton J., Setti A., Doheny E. P., Kenny R. A., Newell F. N.. **A Standing Posture Is Associated with Increased Susceptibility to the Sound-Induced Flash Illusion in Fall-Prone Older Adults**. *Exp. Brain Res.* (2014) **232** 423-434. DOI: 10.1007/s00221-013-3750-7
63. **Wave 3 Documentation**. (2016)
64. Umegaki H., Makino T., Yanagawa M., Nakashima H., Kuzuya M., Sakurai T.. **Maximum Gait Speed Is Associated with a Wide Range of Cognitive Functions in Japanese Older Adults with a Clinical Dementia Rating of 0.5**. *Geriatr. Gerontol. Int.* (2018) **18** 1323-1329. DOI: 10.1111/ggi.13464
65. Wong G. K. C., Lam S. W., Wong A., Ngai K., Poon W. S., Mok V.. **Comparison of Montreal Cognitive Assessment and Mini-Mental State Examination in Evaluating Cognitive Domain Deficit Following Aneurysmal Subarachnoid Haemorrhage**. *PLoS One* (2013) **8** e59946. DOI: 10.1371/journal.pone.0059946
66. Wu T., Zhao Y.. **Associations between Functional Fitness and Walking Speed in Older Adults**. *Geriatr. Nurs.* (2021) **42** 540-543. DOI: 10.1016/j.gerinurse.2020.10.003
|
---
title: Linear and Nonlinear Directed Connectivity Analysis of the Cardio-Respiratory
System in Type 1 Diabetes
authors:
- Michele Sorelli
- T. Noah Hutson
- Leonidas Iasemidis
- Leonardo Bocchi
journal: Frontiers in Network Physiology
year: 2022
pmcid: PMC10013013
doi: 10.3389/fnetp.2022.840829
license: CC BY 4.0
---
# Linear and Nonlinear Directed Connectivity Analysis of the Cardio-Respiratory System in Type 1 Diabetes
## Abstract
In this study, we explored the possibility of developing non-invasive biomarkers for patients with type 1 diabetes (T1D) by quantifying the directional couplings between the cardiac, vascular, and respiratory systems, treating them as interconnected nodes in a network configuration. Towards this goal, we employed a linear directional connectivity measure, the directed transfer function (DTF), estimated by a linear multivariate autoregressive modelling of ECG, respiratory and skin perfusion signals, and a nonlinear method, the dynamical Bayesian inference (DBI) analysis of bivariate phase interactions. The physiological data were recorded concurrently for a relatively short time period (5 min) from 10 healthy control subjects and 10 T1D patients. We found that, in both control and T1D subjects, breathing had greater influence on the heart and perfusion with respect to the opposite coupling direction and that, by both employed methods of analysis, the causal influence of breathing on the heart was significantly decreased ($p \leq 0.05$) in T1D patients compared to the control group. These preliminary results, although obtained from a limited number of subjects, provide a strong indication for the usefulness of a network-based multi-modal analysis for the development of biomarkers of T1D-related complications from short-duration data, as well as their potential in the exploration of the pathophysiological mechanisms that underlie this devastating and very widespread disease.
## Introduction
Type 1 diabetes (T1D) is a chronic condition affecting roughly $5\%$ of the world’s diabetic population (Ogurtsova et al., 2017), which is estimated to reach 642 million ($95\%$ CI: 521–829 million) by 2040 (it was 151 million in 2000 (Wild et al., 2004)) with dramatic social and financial implications. T1D is associated with pathogenetic mechanisms that lead to the apoptosis of pancreatic beta cells and, thus, to an inadequate production of the insulin hormone. There is no currently available cure for T1D, and its clinical care is focused primarily on the normalization of blood glucose levels for averting the onset of long-term complications including cardiovascular disease and renal failure. The treatment of diabetic-related chronic complications accounts for a considerable percentage [about $80\%$ in the United Kingdom (Ogurtsova et al., 2017)] of the total medical costs of diabetes mellitus. Studies show that timing of medical intervention is key to reducing effects of comorbidities of T1D, with earlier interventions resulting in lower disease impact (Doria et al., 2012). Thus, there would be benefits to patients and healthcare systems alike from development of novel diagnostic techniques for early and non-invasive detection of T1D-related complications. Such diagnostic regimes could also have implications in outpatient monitoring and disease progression assessment.
The complex function of the cardiovascular system is realized by the synergistic activity of self-sustained cardiac, respiratory, and vascular oscillators (Ticcinelli et al., 2017), which is deemed to convey the necessary adaptability to sudden variations in the metabolic requirements of the organism or to changing environmental conditions (Penzel et al., 2017). There is a wide variety of clinically available devices for non-invasively monitoring the physiological systems that may be impacted by the progression of T1D. Such systems generate oscillatory modes that span a wide range of characteristic time scales, which can be isolated and separately characterized by means of established time-frequency representation (TFR) techniques (Clemson et al., 2016). In this regard, the wavelet transform (WT) analysis of laser Doppler flowmetry (LDF) signals of microvascular perfusion (Stefanovska et al., 1999) has contributed to the identification of myogenic (Aalkjaer et al., 2011), neurogenic (Söderström et al., 2003) and endothelial (Kvandal et al., 2006) frequency ranges in the microcirculatory vasomotion, in addition to the ones of the extrinsic cardiac and respiratory components (Stefanovska and Hozic, 2000) transmitted to the distal microvascular beds (Table 1). This, in turn, has enabled the non-invasive assessment of the underlying vasomotor mechanisms in pathological states.
**TABLE 1**
| Oscillation | Nominal range (Hz) |
| --- | --- |
| Cardiac | (0.6, 2.0) |
| Respiratory | (0.145, 0.6) |
| Myogenic | (0.052, 0.145) |
| Sympathetic | (0.021, 0.052) |
| Endothelial (NO-dependent) | (0.0095, 0.021) |
| Endothelial (NO-independent) | (0.005, 0.0095) |
Furthermore, the wavelet cross-spectrum (Clemson et al., 2016) and the phase coherence of bivariate data, (Sheppard et al., 2012; Tankanag et al., 2014; Perrella et al., 2018), along with statistical properties translated from information theory [e.g., Granger causality (Granger, 1969) and transfer entropy (Vejmelka and Palus, 2008; Sabesan et al., 2009)], have been used to gain insights into the presence of significant relations between oscillatory sources, and to determine the existence of a mutual physiological coordination, e.g., the well-known synchronous modulation of the heartbeat period by the breathing rhythm, produced at the respiratory centers located within the medulla oblongata and pons of the brainstem (Eckberg, 2003). However, beyond the effects manifested in the oscillators’ phase dynamics, the fundamental functional mechanisms underlying these interactions can be probed via more sophisticated techniques, able to provide information about the directional strength of the coupling and hence about the causality of the interaction (Rosenblum and Pikovsky, 2001; Palus and Stefanovska, 2003; Faes et al., 2004). Since the cardiovascular system must handle time-varying conditions, the employed methods should be capable of capturing non-stationary functional couplings. The dynamical Bayesian inference (DBI) technique, more recently introduced by Stankovski et al. [ 2012], seeks to account for such non-stationarities. In DBI, the cardiovascular system is modelled as a network of phase oscillators coupled by time-dependent functions, which are identified dynamically through a Bayesian estimation framework within subsequent time windows of the oscillators’ phase time series. Several researchers have employed DBI to investigate potential changes in the direct and indirect coupling between the cardiac, respiratory and vasomotor activities; their studies have detected a reduction in the respiratory sinus arrhythmia with ageing (Shiogai et al., 2010; Iatsenko et al., 2013; Stankovski et al., 2014; Ticcinelli et al., 2015; Ticcinelli et al., 2017), and a weakening of the coupling between the microvascular myogenic vasomotion and the central cardiac and respiratory oscillations in the elderly population and in primary hypertension (Ticcinelli et al., 2017). Since metabolic diseases, such as obesity and diabetes, have been recognized as models of accelerated ageing, the aforementioned alterations may also be present in subjects diagnosed with T1D.
Non-stationary metrics of time-frequency activity could elucidate stochastic coupling but require an adequate number of data points over stationary windows for inferences to be statistically significant. Linearly modelling the data may provide a valuable alternative. Multivariate autoregressive (MVAR) models have been used for describing interactions between time series originating from different nodes within a network (Baccalá et al., 2007; Vlachos et al., 2017). In detail, MVAR-based parametric techniques can be utilized to elucidate inter-node connections via coherence-based measures of implicit causality. One such measure is the directed transfer function (DTF), a frequency-domain descriptor of directed network connectivity with fundamental implications from Granger causality (Baccalá et al., 2007). DTF measures cascaded direct and indirect interactions, emphasizes source-based outflow and has been employed in several neuroscience applications (Kamiński et al., 2001; Blinowska et al., 2013; Kamiński and Blinowska, 2014; Vlachos et al., 2017; Adkinson et al., 2018; Hutson et al., 2018). DTF and other MVAR-based measures of directional connectivity may also be applied to the evaluation of the directional coupling between the cardiac, respiratory, and peripheral blood flow systems. The utility of these measures in neuro-cardio-respiratory network interactions has been shown lately in animal studies of sudden unexpected death in epilepsy (SUDEP), a condition that involves potential failure of central control units of cardiac and respiratory behavior (Hutson et al., 2020).
Employing directed connectivity measures to quantify the inter-modulation of the biological oscillations originating from separate but interconnected systems could have valuable diagnostic potential for assessing the deterioration of the cardiovascular and respiratory function in prevalent high-risk conditions such as T1D. According to the results of a recent review article (Klein et al., 2010), adult subjects diagnosed with type 2 diabetes are characterized by reduced respiratory parameters, which appear to be inversely related to blood glucose levels and the time since the initial diagnosis. This review has linked chronic hyperglycemia and inflammation, autonomic neuropathy, microangiopathy of the pulmonary arterioles, and stiffening of the lung parenchyma to the possible biological mechanisms underlying the lung function impairment. This may then result in a detrimental impact on the mutual physiological coupling between the breathing and heart function. In light of the above, in the present study we employed the DBI and DTF frameworks with the aim to non-invasively detect characteristics of the potential decline of connectivity in the cardio-respiratory oscillatory network in a preliminary, relatively small group of healthy controls and patients diagnosed with T1D.
## Experimental Setup and Subjects
10 healthy controls (age: 26.7 ± 1.5 years; M/F: $\frac{7}{3}$) and 10 T1D patients (age: 29.7 ± 13.3 years; M/F: $\frac{5}{5}$) were recruited for the present study. Research activities were carried out in accordance with the guidelines of the Declaration of Helsinki of the World Medical Association: the included subjects received detailed information on the research protocol and its purpose and signed an informed consent form. *The* general characteristics of the participants are summarized in Table 2; one control subject (i.e., $10\%$) and four T1D subjects ($40\%$) were smokers. ECG, breathing and microvascular perfusion signals were simultaneously recorded. Microvascular perfusion was measured on the distal phalanx of the right forefinger using a Periflux 5,000 laser Doppler flowmetry (LDF) system (Perimed AB, Sweden). The time constant of the output low-pass filter of the instrument was set to 0.03 s in order to preserve pulse waveforms. The heart and spontaneous respiratory activities were instead monitored by means of a BioHarness 3.0 wearable chest strap sensor (Zephyr Technology, United States) and transmitted to a PC via Bluetooth. A graphical illustration of the recording setup is shown in Figure 1.
The above three signals were concurrently recorded and digitized at a sampling frequency of 250 Hz (being synchronized through a dedicated data acquisition software). Each recording session lasted 5 min and took place in thermally stable conditions (T ≈ 23°C) following a preliminary acclimatization time interval of 10 min. During signal acquisition, subjects were seated in a chair with back support and leaned their right forearm on a table; furthermore, they were instructed to carefully avoid abrupt movements to prevent the displacement of the LDF probe and thus the introduction of motion-related artifacts in the recorded perfusion signals. An example of the raw signals acquired from a young control individual is shown in Figure 2.
**FIGURE 2:** *Example of (A) ECG, (B) breathing and (C) LDF perfusion signals recorded from a young control subject.*
The mean breathing rate of all subjects was inside the nominal physiological range, that is (0.145, 0.6) Hz (Table 1). Furthermore, LDF perfusion signals recorded from the pathological group did not exhibit a significantly different ($$p \leq 0.450$$) power within the nominal frequency range of the cardiac rhythm (0.6, 2) Hz. However, T1D subjects included a larger proportion of smokers and were on average older than the control subjects. Nevertheless, these differences did not reach statistical significance according to Pearson’s χ2 and Wilcoxon rank-sum tests, respectively.
## Dynamical Bayesian Inference
The functional physiological interaction between cardiac and respiratory processes was investigated by means of the dynamical Bayesian inference (DBI) technique (Duggento et al., 2012; Stankovski et al., 2012). This method regards the cardio-respiratory system as a network of coupled self-sustained nonlinear phase oscillators and uses a Bayesian inference scheme to dynamically estimate their time-evolving coupling strength and causality (i.e., the direction of interactions). Myogenic, sympathetic, and endothelial microvascular oscillations (Table 1) were not considered in the present study, due to the insufficient duration of the recorded signals. A comprehensive description of the approach can be found elsewhere (Duggento et al., 2012; Stankovski et al., 2012; Iatsenko et al., 2013; Clemson et al., 2016; Ticcinelli et al., 2017). Briefly, in DBI, the phase dynamics of two interacting oscillatory processes p1 and p2 is modelled as follows: φ˙1(t)=ω1(t)+d1(φ2,t)+k1(φ1,φ2,t)+ε1(t) [1] where ω1(⋅) is the natural frequency of the first oscillator, d1(⋅) and k1(⋅) are the coupling functions that describe the direct and indirect driving of the second oscillator (with the acceleration/deceleration of the first oscillator’s phase φ1 depending on the second’s φ2), whereas the stochastic term, ε(⋅), represents the noise (usually assumed to be Gaussian and white (Stankovski et al., 2012)). Since the above coupling functions are hypothesized to be 2π-periodic, the right-hand side of Eq. 1 can be decomposed into a linear combination of *Fourier basis* functions Φn=exp[i(n1φ1+n2φ2)]: φ˙1(t)=∑n=−NNc1,n·Φ1,n(φ1,φ2)+ε1(t) [2] where N is the order of the expansion and Φi,0=1 (where $i = 1$, 2). *In* general, the DBI technique sequentially applies the Bayesian theorem to adjacent time windows of the oscillators’ instantaneous phases, φi(t), in order to infer the bank of time-varying parameters ci,n characterizing the functional interaction between the underlying physiological processes, and the noise term, εi. The inferred ci,n values are then used to estimate a dynamic index of directional coupling strength and directionality of influence. In the present study, DBI analysis was based on the related Matlab toolbox developed by the research group on Nonlinear and Biomedical Physics at Lancaster University (http://www.physics.lancs.ac.uk/research/nbmphysics/diats/tfr/).
In detail, DBI analysis usually requires the extraction of the instantaneous frequency of the oscillations of interest, in order to track their characteristic time-dependent phase φi(t). In this regard, an adaptive parametric ridge reconstruction scheme (Iatsenko et al., 2016) was applied to the time-frequency representation (TFR) of the acquired signals in order to isolate the breathing and cardiac oscillatory components. In the present study, the cardiac component was isolated from both the ECG and the LDF signals of cutaneous perfusion. The adjustable parameters of the algorithm, which respectively tune the tolerance to deviations from the component’s mean rate of frequency change and mean frequency, were set to their default value of 1. The wavelet transform (WT) was adopted as TFR technique because of its logarithmic frequency resolution (Stefanovska et al., 1999); specifically, a Morlet wavelet with central frequency f0=1 was chosen as the mother function: γm(t)=12π(ei2πt−e−(2π)22)e−t$\frac{2}{2}$ [3] Prior to the application of the WT, signals were downsampled to 50 Hz, detrended by means of a third order polynomial fit, and band-passed inside the cardiac and respiratory frequency intervals listed in Table 1, to remove the influence of components lying outside the physiological range of interest. The discretization of the frequency domain was performed with a density of 128 voices/octave, which enabled the extraction of smooth ridge curves. DBI was then applied to consecutive overlapping windows of the original time series, with an overlap factor of $50\%$. The window width was set so as to include approximately five cycles of the slowest oscillatory component for inference of the coupling parameters ci,n, as reported in (Iatsenko et al., 2015; Clemson et al., 2016). For the analysis of cardio-respiratory interactions, this resulted in the adoption of 23 overlapping windows of 25 s. Thus, the lowest frequency we could theoretically observe was $\frac{1}{25}$ $s = 0.04$ Hz. The characteristic time-frequency ridges extracted from the signals in Figure 2 are shown in Figure 3.
**FIGURE 3:** *Time-frequency ridges of the (A) ECG, (B) breathing and (C) LDF perfusion signals shown in Figure 2. Ridges were estimated by means of the adaptive parametric approach developed in (Iatsenko et al., 2016).*
As done by Iatsenko et al. [ 2013] and Ticcinelli et al. [ 2017], a Fourier decomposition up to the second order (i.e., $$n = 2$$) was chosen for the phase dynamics model expressed in Eq. 1. Moreover, the propagation constant pw, that weights the diffusion of information between consecutive data windows w (Stankovski et al., 2012), was set to an arbitrary value of 0.2. Iatsenko et al. [ 2013] have nonetheless reported that this internal parameter of the DBI algorithm does not significantly affect the outcome of the Bayesian inference. The Euclidean norm of the coupling parameters ci,n estimated within each data window w was finally used to quantify the overall influence (including direct and indirect couplings) of the phase of the second oscillator on the first one’s, and vice versa, yielding the following directional coupling strength signals: s1→2(w)=∑n=−NN(c1,n(w))2 [4] s2→1(w)=∑n=−NN(c2,n(w))2 [5] where w indicates the dependence of the coupling coefficients on the particular time window.
Furthermore, a directionality index d1,2 (d1,2∈[−1,+1]) was estimated from each window w, in order to quantify the dynamic asymmetry of the bi-directional interaction: d1,2(w)=s1→2(w)−s2→1(w)s1→2(w)+s2→1(w) [6] This index, proposed by Rosenblum and Pikovsky [2001] has been used in the recent literature for detecting the predominant direction of influence between the cardiac and respiratory oscillators (Stankovski et al., 2012; Iatsenko, et al., 2013; Ticcinelli, et al., 2017). Namely, if d1,2∈(0,+1], then the first oscillator drives the second more than the other way around; conversely, if d1,2∈[−1,0), the second drives the first one. However, as reported in (Duggento et al., 2012), directional coupling strengths si→j(w) obtained via DBI represent an overall estimate of the combined phase relationships between the analyzed time series. Thus, spurious non-zero values can be inferred even when no functional interaction exists between the underlying oscillatory processes. This is why the reliability of si→j(w) should be ascertained by surrogate testing, i.e., rejecting directional coupling strengths below a specified acceptance threshold estimated from an adequately large set of surrogate interactions. In this regard, we adopted the inter-subject surrogate approach followed by Toledo et al. [ 2002] and Ticcinelli et al. [ 2017] validating our coupling strength estimates against the median value obtained from 100 unique combinations of randomly selected inter-group signals and subjects. Each of the 100 surrogate datasets was composed of mutually independent time series recorded from different individuals (e.g., ECG from control subject A, breathing from T1D patient B, LDF perfusion from control C). This technique allowed us to exclude from further consideration any directional couplings whose strength was equivalent to the one which might have arisen from chance or bias.
Figure 4 shows sample coupling strength signals of the time-evolving pairwise interactions among peripheral pulse, respiratory and ECG signals, estimated using DBI in a control subject and a T1D patient. Directional coupling strength estimates below the corresponding median values of the surrogates reported in Table 3 (100 surrogate subjects, for a total of Nw = 2,300 windows), were rejected. Also, only those windows for which both directional coupling strengths per paired interaction were found to be statistically significant according to the above rule were further considered in the statistical analysis of the directionality index, di,j. The overall results of the DBI analysis of the control and pathological groups including their statistical comparison (p-values) are shown in Figure 5 and summarized in Table 4. Statistically significant differences between the two groups were detected by means of one-tailed Wilcoxon rank-sum tests for independent samples.
**FIGURE 4:** *Cardio-respiratory directional coupling strength parameters estimated via DBI (Eqs 4, 5; two per panel), and respective directionality indices (Eq. 6; one per panel) obtained from a control subject (left) and a T1D patient (right).* TABLE_PLACEHOLDER:TABLE 3 **FIGURE 5:** *Box plots of DBI measures of connectivity (Eqs 4–6). Top panel: comparison of all (six) cardio-respiratory coupling parameters estimated with DBI for the control (blue) and T1D (red) groups. Bottom panel: comparison of the (three) directionality indices per pair interaction for the control (blue) and T1D (red) groups. Statistically significant decreases, identified by one-tailed Wilcoxon rank-sum tests, are denoted by (*) above the respective boxes.* TABLE_PLACEHOLDER:TABLE 4 Lungs–Heart interaction. Compared to controls, T1D patients exhibited a significant reduction in the directionality index dbre, ECG ($p \leq 0.001$), which reflects a lowered asymmetry of the cardio-respiratory interaction in the pathological group. This was due to a weakened influence of the breathing activity on the cardiac rhythm, as expressed by the statistically significant decrease in the directional coupling strength sbre→ECG ($p \leq 0.001$; Table 4, row 4). Conversely, the directional coupling from the heart to the lungs was not significantly different between the two groups (Table 4, row 5).
Lungs–Pulse interaction. T1D patients also exhibited a significant decrease in the dbre,pulse index ($$p \leq 0.011$$), which indicates a higher symmetry of the interaction between the breathing activity and the cardiac oscillatory mode of the LDF signals. However, in this case, none of the corresponding directional coupling strengths was significantly different between the compared subjects (Table 4, rows 1 and 2).
Heart–Pulse interaction. With respect to the healthy group, T1D patients were characterized by significantly lowered directional coupling strengths, spulse→ECG ($p \leq 0.001$; Table 4, row 7) and sECG→pulse ($p \leq 0.001$; Table 4, row 8). However, no statistically significant difference emerged in the overall directionality of influence, as expressed by the dECG,pulse index across the control and pathological groups (Table 4, row 9).
## Directed Transfer Function
Multivariate autoregressive (MVAR) modelling of the data within short-time segments, each data window aligned in time with concurrent ones from more than one time series, is recommended for network connectivity analysis assuming that these signals are recorded from different parts of a multi-dimensional, linear and wide-sense stationary system. For each window, the estimated array of MVAR model coefficients can then be further analyzed in the frequency domain and, depending on different types of normalization utilized, provides frequency-specific measures of directional functional connectivity between the nodes of the assumed network configuration of the system (Baccalá et al., 2007). We have successfully employed such measures in network analyses of intracranial EEG (iEEG) (Vlachos et al., 2017; Adkinson et al., 2018), and magnetoencephalographic (MEG) recordings (Krishnan et al., 2015) from patients with focal epilepsy for localization of their epileptogenic focus, as well as the assessment of the dynamics of brain’s network connections en route to a life-threatening neurological event, status epilepticus (T. N. Hutson et al., 2018). In the current study we fitted a MVAR model to each of 60-s consecutive non-overlapping data segments from the three recorded signals (ECG, breathing, perfusion) over 5 min. By using a 60-s time window, our frequency resolution is $\frac{1}{60}$ $s = 0.017$ Hz = 0.05 Hz/3, and thus the lowest frequency we can deal with moving the analysis in the frequency domain is three times less than the 0.05 Hz, the lowest frequency in the frequency band of (0.05, 2) Hz we are interested in here. Thus, the MVAR model was of dimension $D = 3$ [i.e., the data to be fitted were placed in three-dimensional column vectors X(t)], and of order $M = 7$ per subject. Also, the window length of 60 s (15,000 data points x three channels = 45,000 data points) is enough for a confident estimation of the 7 × 3 × 3 = 63 MVAR parameters as we are using more than 100 times as many data points as we have parameters to fit.
For each set of three 60-s running windows extracted at the same time from all three signals, the model linearly fits the data in the column vectors X(t) as follows: X(t)=∑τ=1MA(τ)X(t−τ)+E(t) [7] where the time index t is from 1 to N, with N being the number of data points per time series within a time window ($$n = 15$$,000), M is the order of the model ($M = 7$), and τ is increasing in steps of the time delay between samples (we used τ = 1, that is, in time units, equal to the sampling period 1/(250 Hz) = 4 ms). Matrices A(τ) contain the model’s coefficients, whereas the fitting error values are the components of the vector E (in the ideal MVAR model fit, E is multivariate Gaussian white noise). The coefficients of the MVAR model were estimated via the Vieira-Morf partial correlation method. Taking the discrete Fourier Transform of both sides of Eq. 7 and rearranging, we have: [I−∑τ=1pA(τ)e−i2πfτ]·X(f)=E(f), where I is the unitary matrix. Then, by defining: B¯(f)= {I−∑τ=1pAij(τ)e−i2πfτ,for i=j−∑τ=1pAij(τ)e−i2πfτ,for i≠j [8] where i=−1 in the exponents of Eq. 8, the directed transfer function (DTF) can be derived by utilizing the transfer matrix, H(f), defined as: H(f)=B¯−1(f) [9] Specifically, DTF is estimated via the following equation: DTFj→i(f)=|Hij(f)|2∑$k = 1$D|Hik(f)|2 [10] The statistical significance of the DTF values of each interaction derived from each 60-s window was determined. The statistical criteria for inferring the statistical significance and confidence interval of the derived frequency-domain Granger causality-based connectivity measures are recent and have been discussed by a small number of researchers. In this study, we have followed an asymptotic analysis for evaluation of the connectivity measures from the MVAR modelling of our data (Baccalá et al., 1997; Baccalá et al., 2016). In detail, the significance of the connectivity measures DTFj→i(f) at a specific frequency f between two nodes i and j was tested according to the following null hypothesis: H0:|DTFj→i(f)|2=0∀i,j∈{1,…,D} [11] Rejecting H0 at a specified significance level (typically α = 0.05) also required to reject non-statistically significant DTF values. Confidence intervals for the existing connections were estimated by determining the asymptotic distribution of DTF according to (Toppi et al., 2016). Only the thus identified statistically significant DTF values (ssDTF) were further analyzed in this study. Analogously to the DBI analysis, an index of directionality was finally obtained from the ssDTF estimates as follows: di,j(f)=ssDTFi→j(f)−ssDTFj→i(f)ssDTFi→j(f)+ssDTFj→i(f) [12] The statistically significant DTF values (ssDTF) of directional connectivity estimated per interaction from MVAR modelling (six directional interactions between the three recorded signals) were aggregated over all windows (60-s non-overlapping data segments) and subjects within the same group (control or T1D) and averaged over the physiologically relevant frequency band (0.05, 2) Hz. The median and quartiles of the ssDTF values obtained from the control and T1D groups are shown in Figure 6. It is relevant to highlight that MVAR modelling evaluate signals across identical frequencies over the entire physiological range of interest, in contrast to DBI which is based on the extraction of the specific time-varying frequency component of the cardiac pulsatility, within an effectively tighter range. Therefore, in this section interactions involving LDF signals are denoted as “perfusion”, rather than “pulse”. From Figure 6, we make the following statistically significant observations about the assessed directional interactions: “Perfusion→Breathing”, “Perfusion→ECG” and “ECG→Breathing” connectivity strengths are elevated in T1D subjects compared to controls. Conversely, the “Breathing→ECG” interaction in T1D is lower than the controls’. It is also noteworthy that “Breathing→ECG” is significantly higher in connectivity than “ECG→Breathing” for both T1D and controls. Also, the “Perfusion→ECG” coupling is higher than “ECG→Perfusion” in both groups.
**FIGURE 6:** *Box plots of DTF measures of connectivity in the (0.05, 2) Hz frequency band (Eqs 10, 12). Top panel: median ± first and third quartiles of ssDTF values across all directed interactions for control (blue) and T1D (red) groups. Bottom panel: directionality indices per pair interaction for the control (blue) and T1D (red) groups (*) above boxes denotes p-value < 0.05 estimated from non-parametric one-tailed Wilcoxon rank-sum tests comparing T1D and control groups.*
Statistically significant directional interactions ($p \leq 0.05$) between the network nodes for each pair of recorded signals are reported in Table 5 (columns 3 and 4) together with their directionality index di,j. Inter-group comparisons were conducted via one-tailed Wilcoxon rank-sum statistical tests, whose p-values are also included in Table 5 (last two columns).
**TABLE 5**
| Interaction | DTF connectivity | Median (controls) | Median (T1D) | p-value H1a | p-value H1b |
| --- | --- | --- | --- | --- | --- |
| Lungs ↔ Perfusion | ssDTFperf→bre | 0.027 | 0.074 | 0.982 | 0.019 a |
| Lungs ↔ Perfusion | ssDTFbre→perf | 0.310 | 0.129 | 0.031 a | 0.970 |
| | dbre, perf | 0.729 | 0.276 | 0.084 | 0.918 |
| Lungs ↔ Heart | ssDTFECG→bre | 0.002 | 0.011 | 1.000 | <0.001 a |
| Lungs ↔ Heart | ssDTFbre→ECG | 0.897 | 0.634 | <0.001 a | 1.000 |
| | dbre, ECG | 0.992 | 0.954 | <0.001 a | 1.000 |
| Heart ↔ Perfusion | ssDTFperf→ECG | 0.055 | 0.121 | 0.998 | 0.002 a |
| Heart ↔ Perfusion | ssDTFECG→perf | 0.002 | 0.006 | 0.763 | 0.258 |
| Heart ↔ Perfusion | dECG, perf | −0.940 | −0.865 | 0.646 | 0.379 |
Lungs–Heart interaction. In agreement with the results from the DBI method, DTF shows that T1D patients exhibit a statistically significant reduction ($p \leq 0.001$) in the directional coupling strength from the lungs to the heart, as well as in the directionality index dbre, ECG, compared to controls. The latter is due to a statistically significant increase ($p \leq 0.001$) in the directional strength ssDTFECG→bre observed in the pathological group, with a parallel significant decrease ($p \leq 0.001$) in ssDTFbre→ECG (Table 5, rows 4–6).
Lungs–Perfusion interaction. Similarly to the DBI analysis, the DTF measures of connectivity showed a decrease in the dbre,pulse index of T1D patients compared to the one estimated from the control group (0.729 vs. 0.276), implying a lowered asymmetry of the interaction between breathing and LDF signals. However, this decrease was not as significant ($$p \leq 0.084$$) as the one estimated via DBI ($$p \leq 0.011$$). This outcome is due to the mixed results (decrease with $$p \leq 0.031$$, and increase with $$p \leq 0.019$$) related to the directional coupling strengths ssDTFbre→perf and ssDTFperf→bre, respectively (Table 5, rows 1 and 2).
Heart–Perfusion interaction. T1D patients exhibited increased connectivity compared to the controls in both directions, with the difference in the “Perfusion→ECG” coupling reaching statistical significance ($$p \leq 0.002$$) (Table 5, rows 7 and 8). The above trends contributed to a diminished absolute value of the directionality index dECG,perf in the pathological group, which implies a more balanced interaction with respect to controls (Table 5, row 9).
Finally, within each group, a few interesting features were also observed from the DTF results in relation to the difference in the strength of the directional couplings per interaction. In particular, in both control subjects and T1D patients, the median coupling strengths from the lungs to the heart and from the lungs to the microcirculation were considerably higher than in the opposite direction (the same outcome of the DBI analysis). However, the inter-group differences in the directional strengths between heart and microcirculation were contradictory with respect to the DBI analysis, which associated a higher level of bidirectional connectivity to the control group (Table 5, columns 3 and 4).
## Discussion
Towards the goal of developing reliable and non-invasive biomarkers for T1D, we employed both nonlinear (bivariate) and linear (multivariate) measures to assess possible impairments in the coupling strength and directionality of influence between three representative nodes of the cardiovascular and respiratory systems (heart, lungs, microcirculation) in patients diagnosed with T1D compared to control subjects. The two adopted methods can capture equivalent or different features in the communication between the nodes of a physiological network because of their different capabilities, that is: linearity (DTF) vs. nonlinearity (DBI) in the data; multivariate (DTF) vs. bivariate (DBI) data analysis; measure of connectivity between signals at the same frequency (DTF) vs. different frequencies (DBI). Employing these two techniques, we did identify impairments (by both or one of the approaches) in the functional directional interactions between heart, lungs, and microcirculation in T1D patients. In detail, an impairment was defined as a statistically significant difference ($p \leq 0.05$) in the directional coupling strengths between the respective nodes, compared to the homologous estimate obtained from the control group (i.e., rejection of the null hypothesis H0).
Regarding the functional interactions between heart and lungs, DBI, the nonlinear framework, revealed a significantly reduced ($p \leq 0.001$) influence of the respiratory activity on the phase of the cardiac rhythm in the T1D group. A similar, statistically significant ($p \leq 0.001$) finding also emerged from the linear network analysis, using DTF. Moreover, the imbalance in the two communication channels from the lungs to the heart and vice versa, as captured by the directionality index, was also highly significantly different in both methods ($p \leq 0.001$). It is well known that the phase of the respiratory activity directly influences the action of the heart pump, as breathing-related changes in the intrathoracic volume alter the cardiac pre-load, thus affecting cardiac filling, post-load and other circulatory variables. Furthermore, respiration gates the timing of autonomic motoneuron firing (Eckberg, 2003), thus modulating the peripheral autonomic nervous system’s outflow to the heart, an indirect cardio-respiratory coupling occurring via neuronal control (Iatsenko et al., 2013; Kralemann et al., 2013). Therefore, our finding of a reduced driving relationship of the lungs to the heart in T1D patients could be related to autonomic neuropathy, vascular degeneration or lung tissue stiffening, common co-morbidities associated with diabetes mellitus (Klein et al., 2010).
An analogous decrease of the influence of respiration on the microvascular perfusion in the T1D group compared to controls was observed by DTF analysis ($$p \leq 0.031$$) but could not be verified by DBI ($p \leq 0.05$). However, like for the lungs-heart interaction, the imbalance in the directional coupling strengths between lungs and microcirculation, as reflected by the directionality index, was significantly less ($$p \leq 0.011$$) in T1D than in controls as shown by DBI as well as by DTF, though without reaching a statistical significance level ($$p \leq 0.084$$). Also, regarding the DBI analysis of phase interactions, it is notable that control subjects exhibited similar statistics with respect to the evaluation of breathing and ECG signals. This result would be in line with previous findings by Jamšek and Stefanovska on the coupling information among cardiac and respiratory processes which propagates to the distal microvascular beds (Jamšek and Stefanovska, 2007), and can be characterized through the analysis of LDF signals recorded non-invasively from the skin.
In T1D subjects, the DBI analysis highlighted a significantly decreased communication in both directions between the ECG and the microvascular pulse signal extracted from LDF signals. This finding, however, could not be validated by DTF too. It is noteworthy that these directional interactions were associated with significantly higher coupling strength values (Table 4). This could be due to the way DBI evaluates causal relationships and what it can capture. In this case, DBI basically assesses the phase coupling between ECG and pulse signals that, although recorded at different anatomical locations, originate from the same source, representing the electrical and mechanical activities of the heart, respectively (Kralemann et al., 2013).
Finally, the estimated directional couplings from the lungs to the heart and microvasculature, via either the DBI or DTF methods, were considerably higher than the ones from the heart and vasculature towards the lungs, in both control subjects and T1D patients. Since this outcome was common in both groups, it cannot be used as a biomarker for T1D. However, it agrees with the findings of Palus and Stefanovska [2003], which have shown that the respiratory process drives the heart activity at all breathing frequencies, whether paced or spontaneous, and may shed more light on the involved physiological mechanisms en route to a better understanding of the cardio-respiratory system.
A potential limitation of this study is the availability and analysis of signals from only a small number of nodes (lungs, heart, microcirculation) in the network under investigation. Both DTF and DBI measure the global (direct and indirect) interactions between two nodes A and B, the indirect interactions from A to B or from B to A occurring through other node(s) C that we may not have access to in the network (Kamiński et al., 2001; Baccalá et al., 2016). In this regard, it is established that each respiratory cycle is tightly controlled by four separate control centers in the pons and medulla (Smith et al., 1991; Hilaire and Pásaro, 2003; Dampney, 2017), which cannot operate without central intervention from the brain, and direct feedback from the heart. Furthermore, central autonomic neural control has a well-known role in the low- and high-frequency variability of the heart rate (Shaffer and Ginsberg, 2017). Thus, ignoring the brain (EEG) and investigating this complex neuro-cardio-respiratory network from only three nodes (lungs, heart, microcirculation) could have skewed the level of the estimated bivariate interactions in both T1D and control groups. However, the comparative statistical analysis of each measure across the two groups may take care of this skewness if it were in the same direction in both groups, per interaction.
In summary, we found that in both control and T1D subjects, breathing had greater influence on the heart and peripheral microvascular perfusion, compared to the opposite directional couplings and that, by both the employed methods of connectivity analysis, the causal influence of the respiratory activity on the heart was significantly decreased ($p \leq 0.05$) in T1D patients compared to the control group. These preliminary results can be linked to established comorbidities of T1D and, although obtained from a limited number of subjects, provide a strong indication for the usefulness of a network-based multi-modal analysis for the development of biomarkers from short-duration data, and for monitoring the disease and T1D-related complications over time, as well as its potential in the exploration of the pathophysiological mechanisms that underlie this devastating and very widespread disease.
## Data Availability Statement
The raw data supporting the conclusion of this article will be made available by the authors, without undue reservation.
## Ethics Statement
Ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements. Written informed consent to participate in this study was provided by the participants’ legal guardian/next of kin.
## Author Contributions
MS collected and shared data with United States site, conceptualized and performed the dynamical Bayesian inference, in addition to compiling and preparing results for publication. TH conceptualized and performed the directed transfer function analysis in addition to compiling and preparing results for publication. LI gave input and expertise on signal processing and analysis, and assisted with full preparation and review of the manuscript. LB provided expertise on data, disease state, and analytical decisions in addition to preparation and review of the manuscript. All authors contributed to review and revision of results and final manuscript.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Aalkjaer C., Boedtkjer D., Matchkov V.. **Vasomotion - What Is Currently Thought?**. *Acta Physiol.* (2011) **202** 253-269. DOI: 10.1111/j.1748-1716.2011.02320.x
2. Adkinson J. A., Karumuri B., Hutson T. N., Liu R., Alamoudi O., Vlachos I.. **Connectivity and Centrality Characteristics of the Epileptogenic Focus Using Directed Network Analysis**. *IEEE Trans. Neural Syst. Rehabil. Eng.* (2018) **27** 22-30. DOI: 10.1109/TNSRE.2018.2886211
3. Baccalá L. A., de Brito C. S., Takahashi D. Y., Sameshima K.. **Unified Asymptotic Theory for All Partial Directed Coherence Forms**. *Philos. Trans. A. Math. Phys. Eng. Sci.* (1997) **371** 20120158. DOI: 10.1098/rsta.2012.0158
4. Baccalá L. A., Sameshima K., Takahashi D.. **Generalized Partial Directed Coherence**. (2007) **3** 163-166. DOI: 10.1109/icdsp.2007.4288544
5. Baccalá L. A., Takahashi D. Y., Sameshima K.. **Directed Transfer Function: Unified Asymptotic Theory and Some of its Implications**. *IEEE Trans. Biomed. Eng.* (2016) **63** 2450-2460. DOI: 10.1109/tbme.2016.2550199
6. Blinowska K. J., Kamiński M., Brzezicka A., Kamiński J.. **Application of Directed Transfer Function and Network Formalism for the Assessment of Functional Connectivity in Working Memory Task**. *Philos. Trans. A. Math. Phys. Eng. Sci.* (2013) **371** 20110614. DOI: 10.1098/rsta.2011.0614
7. Clemson P., Lancaster G., Stefanovska A.. **Reconstructing Time-Dependent Dynamics**. *Proc. IEEE* (2016) **104** 223-241. DOI: 10.1109/jproc.2015.2491262
8. Dampney R.. **Central Mechanisms Regulating Coordinated Cardiovascular and Respiratory Function**. *Oxford Res. Encyclopedia Neurosci.* (2017) **1** 1-27. DOI: 10.1093/acrefore/9780190264086.013.64
9. Doria A., Niewczas M. A., Fiorina P.. **Can Existing Drugs Approved for Other Indications Retard Renal Function Decline in Patients with Type 1 Diabetes and Nephropathy?**. *Semin. Nephrol.* (2012) **32** 437-444. DOI: 10.1016/j.semnephrol.2012.07.006
10. Duggento A., Stankovski T., McClintock P. V., Stefanovska A.. **Dynamical Bayesian Inference of Time-Evolving Interactions: from a Pair of Coupled Oscillators to Networks of Oscillators**. *Phys. Rev. E Stat. Nonlin Soft Matter Phys.* (2012) **86** 061126. DOI: 10.1103/PhysRevE.86.061126
11. Eckberg D. L.. **The Human Respiratory Gate**. *J. Physiol.* (2003) **548** 339-352. DOI: 10.1113/jphysiol.2002.037192
12. Faes L., Porta A., Cucino R., Cerutti S., Antolini R., Nollo G.. **Causal Transfer Function Analysis to Describe Closed Loop Interactions between Cardiovascular and Cardiorespiratory Variability Signals**. *Biol. Cybern.* (2004) **90** 390-399. DOI: 10.1007/s00422-004-0488-0
13. Granger C. W. J.. **Investigating Causal Relations by Econometric Models and Cross-Spectral Methods**. *Econometrica* (1969) **37** 424-438. DOI: 10.2307/1912791
14. Hilaire G., Pásaro R.. **Genesis and Control of the Respiratory Rhythm in Adult Mammals**. *News Physiol. Sci.* (2003) **18** 23-28. DOI: 10.1152/nips.01406.2002
15. Hutson T., Pizarro D., Pati S., Iasemidis L. D.. **Predictability and Resetting in a Case of Convulsive Status Epilepticus**. *Front. Neurol.* (2018) **9** 172-178. DOI: 10.3389/fneur.2018.00172
16. Hutson T. N., Rezaei F., Gautier N. M., Indumathy J., Glasscock E., Iasemidis L.. **Directed Connectivity Analysis of the Neuro-Cardio- and Respiratory Systems Reveals Novel Biomarkers of Susceptibility to SUDEP**. *IEEE Open J. Eng. Med. Biol.* (2020) **1** 301-311. DOI: 10.1109/ojemb.2020.3036544
17. Iatsenko D., Bernjak A., Stankovski T., Shiogai Y., Owen-Lynch P. J., Clarkson P. B.. **Evolution of Cardiorespiratory Interactions with Age**. *Philos. Trans. A. Math. Phys. Eng. Sci.* (2013) **371** 20110622. DOI: 10.1098/rsta.2011.0622
18. Iatsenko D., McClintock P. V. E., Stefanovska A.. **Extraction of Instantaneous Frequencies from Ridges in Time-Frequency Representations of Signals**. *Signal. Process.* (2016) **125** 290-303. DOI: 10.1016/j.sigpro.2016.01.024
19. Iatsenko D., McClintock P. V. E., Stefanovska A.. **Linear and Synchrosqueezed Time-Frequency Representations Revisited: Overview, Standards of Use, Resolution, Reconstruction, Concentration, and Algorithms**. *Digital Signal. Process.* (2015) **42** 1-26. DOI: 10.1016/j.dsp.2015.03.004
20. Jamšek J., Stefanovska A.. **The Cardio-Respiratory Couplings Observed in the LDF Signal Using Wavelet Bispectrum**. (2007) **2007** 4072-4075. DOI: 10.1109/IEMBS.2007.4353228
21. Kamiński M., Ding M., Truccolo W. A., Bressler S. L.. **Evaluating Causal Relations in Neural Systems: Granger Causality, Directed Transfer Function and Statistical Assessment of Significance**. *Biol. Cybern* (2001) **85** 145-157. DOI: 10.1007/s004220000235
22. Kamiński M., Blinowska K. J.. **Directed Transfer Function Is Not Influenced by Volume Conductionâ€"inexpedient Pre-Processing Should Be Avoided**. *Front. Comput. Neurosci.* (2014) **8** 61. DOI: 10.3389/fncom.2014.00061
23. Klein O. L., Krishnan J. A., Glick S., Smith L. J.. **Systematic Review of the Association between Lung Function and Type 2 Diabetes Mellitus**. *Diabetic Med.* (2010) **27** 977-987. DOI: 10.1111/j.1464-5491.2010.03073.x
24. Kralemann B., Frühwirth M., Pikovsky A., Rosenblum M., Kenner T., Schaefer J.. *Nat. Commun.* (2013) **4** 2418. DOI: 10.1038/ncomms3418
25. Krishnan B., Vlachos I., Wang Z. I., Mosher J., Najm I., Burgess R.. **Epileptic Focus Localization Based on Resting State Interictal MEG Recordings Is Feasible Irrespective of the Presence or Absence of Spikes**. *Clin. Neurophysiol.* (2015) **126** 667-674. DOI: 10.1016/j.clinph.2014.07.014
26. Kvandal P., Landsverk S. A., Bernjak A., Stefanovska A., Kvernmo H. D., Kirkebøen K. A.. **Low-Frequency Oscillations of the Laser Doppler Perfusion Signal in Human Skin**. *Microvasc. Res.* (2006) **72** 120-127. DOI: 10.1016/j.mvr.2006.05.006
27. Ogurtsova K., da Rocha Fernandesda J. D. R., Huang Y., Linnenkamp U., Guariguata L., Cho N. H.. **IDF Diabetes Atlas: Global Estimates for the Prevalence of Diabetes for 2015 and 2040**. *Diabetes Res. Clin. Pract.* (2017) **128** 40-50. DOI: 10.1016/j.diabres.2017.03.024
28. Palus M., Stefanovska A.. **Direction of Coupling from Phases of Interacting Oscillators: An Information-Theoretic Approach**. *Phys. Rev. E Stat. Nonlin Soft Matter Phys.* (2003) **67** 055201. DOI: 10.1103/PhysRevE.67.055201
29. Penzel T., Porta A., Stefanovska A., Wessel N.. **Recent Advances in Physiological Oscillations**. *Physiol. Meas.* (2017) **38** E1-E7. DOI: 10.1088/1361-6579/aa6780
30. Perrella A., Sorelli M., Giardini F., Frassineti L., Francia P., Bocchi L.. **Wavelet Phase Coherence between the Microvascular Pulse Contour and the Respiratory Activity**. *World Congress Med. Phys. Biomed. Eng.* (2018) **68** 311-314. DOI: 10.1007/978-981-10-9038-7_58
31. Rosenblum M. G., Pikovsky A. S.. **Detecting Direction of Coupling in Interacting Oscillators**. *Phys. Rev. E* (2001) **64** 045202. DOI: 10.1103/PhysRevE.64.045202
32. Sabesan S., Good L. B., Tsakalis K. S., Spanias A., Treiman D. M., Iasemidis L. D.. **Information Flow and Application to Epileptogenic Focus Localization from Intracranial EEG**. *IEEE Trans. Neural Syst. Rehabil. Eng.* (2009) **17** 244-253. DOI: 10.1109/tnsre.2009.2023291
33. Shaffer F., Ginsberg J. P.. **An Overview of Heart Rate Variability Metrics and Norms**. *Front. Public Health* (2017) **5** 258. DOI: 10.3389/fpubh.2017.00258
34. Sheppard L. W., Stefanovska A., McClintock P. V.. **Testing for Time-Localized Coherence in Bivariate Data**. *Phys. Rev. E Stat. Nonlin Soft Matter Phys.* (2012) **85** 046205-046282. DOI: 10.1103/PhysRevE.85.046205
35. Shiogai Y., Stefanovska A., McClintock P. V.. **Nonlinear Dynamics of Cardiovascular Ageing**. *Phys. Rep.* (2010) **488** 51-110. DOI: 10.1016/j.physrep.2009.12.003
36. Smith J. C., Ellenberger H. H., Ballanyi K., Richter D. W., Feldman J. L.. **Pre-Bötzinger Complex: A Brainstem Region that May Generate Respiratory Rhythm in Mammals**. *Science* (1991) **254** 726-729. DOI: 10.1126/science.1683005
37. Söderström T., Stefanovska A., Veber M., Svensson H.. **Involvement of Sympathetic Nerve Activity in Skin Blood Flow Oscillations in Humans**. *Am. J. Physiology-Heart Circulatory Physiol.* (2003) **284** H1638-H1646. DOI: 10.1152/ajpheart.00826.2000
38. Stankovski T., Duggento A., McClintock P. V., Stefanovska A.. **Inference of Time-Evolving Coupled Dynamical Systems in the Presence of Noise**. *Phys. Rev. Lett.* (2012) **109** 024101-024105. DOI: 10.1103/PhysRevLett.109.024101
39. Stankovski T., McClintock P. V. E., Stefanovska A.. **Cardiorespiratory Coupling Functions, Synchronization and Ageing**. (2014). DOI: 10.1109/esgco.2014.6847579
40. Stefanovska A., Bracic M., Kvernmo H. D.. **Wavelet Analysis of Oscillations in the Peripheral Blood Circulation Measured by Laser Doppler Technique**. *IEEE Trans. Biomed. Eng.* (1999) **46** 1230-1239. DOI: 10.1109/10.790500
41. Stefanovska A., Hožič M.. **Spatial Synchronization in the Human Cardiovascular System**. *Prog. Theor. Phys. Suppl.* (2000) **139** 270-282. DOI: 10.1143/PTPS.139.270
42. Tankanag A. V., Grinevich A. A., Kirilina T. V., Krasnikov G. V., Piskunova G. M., Chemeris N. K.. **Wavelet Phase Coherence Analysis of the Skin Blood Flow Oscillations in Human**. *Microvasc. Res.* (2014) **95** 53-59. DOI: 10.1016/j.mvr.2014.07.003
43. Ticcinelli V., Stankovski T., Iatsenko D., Bernjak A., Bradbury A. E., Gallagher A. R.. **Coherence and Coupling Functions Reveal Microvascular Impairment in Treated Hypertension**. *Front. Physiol.* (2017) **8** 749. DOI: 10.3389/fphys.2017.00749
44. Ticcinelli V., Stankovski T., McClintock P. V. E., Stefanovska A.. **Ageing of the Couplings between Cardiac, Respiratory and Myogenic Activity in Humans**. (2015). DOI: 10.1109/embc.2015.7320093
45. Toledo E., Akselrod S., Pinhas I., Aravot D.. **Does Synchronization Reflect a True Interaction in the Cardiorespiratory System?**. *Med. Eng. Phys.* (2002) **24** 45-52. DOI: 10.1016/S1350-4533(01)00114-X
46. Toppi J., Mattia D., Risetti M., Formisano R., Babiloni F., Astolfi L.. **Testing the Significance of Connectivity Networks: Comparison of Different Assessing Procedures**. *IEEE Trans. Biomed. Eng.* (2016) **63** 2461-2473. DOI: 10.1109/TBME.2016.2621668
47. Vejmelka M., Palus M.. **Inferring the Directionality of Coupling with Conditional Mutual Information**. *Phys. Rev. E* (2008) **77** 026214. DOI: 10.1103/PhysRevE.77.026214
48. Vlachos I., Krishnan B., Treiman D. M., Tsakalis K., Kugiumtzis D., Iasemidis L. D.. **The Concept of Effective Inflow: Application to Interictal Localization of the Epileptogenic Focus from iEEG**. *IEEE Trans. Biomed. Eng.* (2017) **64** 2241-2252. DOI: 10.1109/tbme.2016.2633200
49. Wild S., Roglic G., Green A., Sicree R., King H.. **Global Prevalence of Diabetes: Estimates for the Year 2000 and Projections for 2030**. *Diabetes Care* (2004) **27** 1047-1053. DOI: 10.2337/diacare.27.5.1047
|
---
title: Criticality in the Healthy Brain
authors:
- Jifan Shi
- Kenji Kirihara
- Mariko Tada
- Mao Fujioka
- Kaori Usui
- Daisuke Koshiyama
- Tsuyoshi Araki
- Luonan Chen
- Kiyoto Kasai
- Kazuyuki Aihara
journal: Frontiers in Network Physiology
year: 2022
pmcid: PMC10013033
doi: 10.3389/fnetp.2021.755685
license: CC BY 4.0
---
# Criticality in the Healthy Brain
## Abstract
The excellence of the brain is its robustness under various types of noise and its flexibility under various environments. However, how the brain works is still a mystery. The critical brain hypothesis proposes a possible mechanism and states that criticality plays an important role in the healthy brain. Herein, using an electroencephalography dataset obtained from patients with psychotic disorders (PDs), ultra-high risk (UHR) individuals and healthy controls (HCs), and its dynamical network analysis, we show that the brain of HCs remains around a critical state, whereas that of patients with PD falls into more stable states. Meanwhile, the brain of UHR individuals is similar to that of PD in terms of entropy but is analogous to that of HCs in causality patterns. These results not only provide evidence for the criticality of the normal brain but also highlight the practicability of using an analytic biophysical tool to study the dynamical properties of mental diseases.
## Introduction
Studying the human brain is a large project, which involves investigating its organization, structure, function, and association with behavior. Detecting the physical mechanisms underlying the working of the brain is one of the most important topics. On the one hand, the brain should be well-structured to process information appropriately. On the other hand, it must be flexible enough to adapt to various environments and emergencies.
The critical brain hypothesis provides an intriguing explanation for the mechanism of the working of the brain, which assumes that in the normal brain, neural networks work near a critical state (Beggs, 2015; Massobrio et al., 2015; Beggs and Timme, 2012). Self-organized criticality (Cocchi et al., 2017; Lee et al., 2019) is one argument of the critical brain hypothesis, which has been supported by the neuronal avalanche phenomenon experimentally (Beggs and Plenz, 2003; Petermann et al., 2009): the size of the cortical activities exhibits power laws. Studies have also found a possibility that this critical mechanism in the brain could ensure maximized capacity and transmission of information (Haldeman and Beggs, 2005; Shew et al., 2009; Shew et al., 2011). However, although neuronal avalanches and power laws provide important statistical descriptions of the critical brain, an intrinsic dynamical interpretation is still missing.
The dynamical network analysis or dynamical network marker (DNM) theory has been studied in the recent decade; it provides a dynamics-based tool to detect the changes in a complex system under perturbation, especially near the critical point (Chen et al., 2012; Liu et al., 2012; Kuehn, 2011; Shi et al., 2016). The DNM theory generalizes the approach of detecting early warning signals of critical slowing down phenomena (Scheffer et al., 2001; Scheffer, 2009; Scheffer et al., 2009) to complex networks. According to the DNM theory, a core subnetwork called a dynamical network (DNMnet) can be found, which is the leading subnetwork of the system toward criticality. Its components (called the DNM group) exhibit large deviations in signals and strong correlations between them around the critical state. DNM groups can not only act as a marker for the criticality of complex systems but also provide an approach for predicting disease, economic crashes, etc (Liu et al., 2014a; Liu et al., 2015; Richard et al., 2016; Lesterhuis et al., 2017; Li et al., 2013; Liu et al., 2014b; Yang et al., 2018; Liu et al., 2019; Dakos and Bascompte, 2014).
Herein, we used electroencephalography (EEG) (Cooper et al., 2014) data recorded from healthy control subjects (HCs), ultra-high risk (UHR) individuals, and patients with psychotic disorder (PD) to explore states of the brain using the DNM theory. By analyzing the dynamical properties of the DNM group of the neural network in the brain, we found that brains of HCs were at a critical state, whereas those of patients with PD were stuck in more stable less-critical states. Brains of UHR individuals fell in a medial state with similar low entropy as those of patients with PD, but a weak causality pattern was observed between electrodes similar to HC. These results based on DNM modeling not only provide an evidence of the critical brain hypothesis but also show a biophysical framework to study the brain and mental diseases.
## Construction of the DNMnet
The 64-channel EEG data were recorded from subjects with the two-tone auditory oddball paradigm (see Methods). Mathematically, we suppose that the dynamics of brain signals under the standard stimulus evolves as follows: x(t+1)= f(x(t), λ), [1] where x=(x1,x2,…,xn)T∈*Rn is* a vector containing signals from all n electrodes, f is assumed to satisfy the existence and uniqueness condition, and λ is a bifurcation parameter. Because we assume the invariance of the brain dynamics, the signals y for the deviant stimulus (the deviant tone in two-tone auditory oddball paradigm, see Methods), which is a small perturbation from x, follow the same equation y(t+1)=f(y(t),λ). Thus, the evolution of the difference process z(t) = y(t)−x(t) can be approximated as follows: z(t+1)≜y(t+1)−x(t+1) ≈A(λ)(y(t)−x(t))+ ξ(t) =A(λ)z(t)+ξ(t), [2] where the matrix A(λ)∈Rn×n is the Jacobian matrix of f (⋅, λ), and ξ(⋅) is inevitable noise in the brain and assumed as a small additive noise term which is independent of A(λ). When the parameter λ→λ0, where λ0 is a codimension-one catastrophic bifurcation point of the system (Kéfi et al., 2013), the largest-modulus eigenvalue of A(λ) will influence system stability. According to the DNM theory (Liu et al., 2012), the network between nodes can be partitioned into a DNM group and a non-DNM group. The DNM group leads the criticality of the system, and deviations of signals and correlations between any nodes in the DNM group will increase sharply around a critical point. Utilizing these characteristics, we constructed the DNMnet and compared the dynamical differences between brains of HCs and those of patients with PD. The DNMnets under duration deviant (dD) experiments and frequency deviant (fD) experiments are shown in Figures 1A–D. Details can be found in the Methods.
**FIGURE 1:** *Auditory mismatch negativity (MMN) patterns, dynamical networks (DNMnet), and DNM indices of the three groups (HC, UHR, and PD). (A) is the duration MMN (dMMN) pattern and (C) is the frequency MMN (fMMN) pattern for the electrode FCz (channel number 4). (B) and (D) are DNMnets constructed in the duration deviation (dD) case and the frequency deviation (fD) case, respectively. (B) has 20 nodes and 12 edges which exhibited significant differences (the one-sided Wilcoxon rank-sum test,
p<10−3
for nodes and
p<0.005
for edges after FDR correction) between the HC and PD groups, and (D) has 17 nodes and 37 edges. For dD in (E) and fD in (F), DNM indices were computed, compared, and plotted using violin plots. The p-values were calculated by the one-sided Wilcoxon rank-sum test for “HC > UHR” and “HC > PD”, whereas for “UHR ≈ PD”, the two-sided Wilcoxon rank-sum test was used.*
## Criticality of the Brain: MMN Pattern and DNMIndex
Mismatch negativity (MMN) (Erickson et al., 2016) is one of the most well-known patterns in PD, whose amplitude of neural activities is reduced compared to the healthy brain. MMN is associated with cognitive impairments (Chung et al., 2017) and can serve as a biomarker for early interventions (Bodatsch et al., 2015) and development of novel treatments (Perez et al., 2017; Kantrowitz et al., 2018). However, mechanisms underlying altered connectivity remain unknown in the clinical field. Using the DNM theory, we found that the MMN amplitude is explained by the difference process z(t) of the brain signal. Furthermore, the nodes in the DNMnet with a large deviation in the signal are all located in the fronto-central area, which corresponds to the region with significant MMN patterns (Figures 1A–D). The DNM theory can provide a dynamical explanation of the MMN pattern, and in return, the MMN pattern provides an evidence for the criticality of the normal brain under the DNM framework.
A DNMIndex was designed to measure the criticality based on the characteristics extracted from the DNMnet as follows: DNMIndex=Π(σ(x))⋅Π(|ρ|), [3] where “Π (⋅) ” stands for the geometric mean of a vector, “ σ(⋅) ” is the component-wise standard deviation, “ x ” denotes the brain signals after preprocessing (corrected epochs) of nodes in the DNMnet, “ |⋅| ” is the component-wise absolute value function, and “ ρ ” denotes the correlations between nodes with edges in the DNMnet. The DNMIndex will be higher at the critical state than at non-critical states. Using the DNMIndex, we found the brain of HC is in a significantly more critical state than those of patients with PD and UHR individuals in both dD and fD cases (Figures 1E–F).
## Criticality of the Brain: Entropy
Mutual information between each pairs of nodes over time and its entropy are important metrics for isotropy and the criticality of a network. Based on the nodes in the DNMnet, we define the mutual information over time from node x to node y as follows: Ix→y=∬p(xt, yt+Δt)logp(xt,yt+Δt)p(xt)p(yt+Δt)dxdy [4] where {xt|$t = 1$, 2, …} and {yt|$t = 1$, 2, …} are the observed time series from node x and y; Δt is a time interval; p(xt, yt+Δt) is the joint probability density function; p(xt) and p(yt+Δt) are the respective marginal density functions; and the integral is over the entire (xt, yt+Δt) space. According to the DNM theory, if x and y belong to the DNM group, their correlation will become strong near the critical point; thus, Ix→y grows rapidly. However, if either x or y belongs to the non-DNM group, mutual information will remain stable and bounded. Thus, near the critical point, some rapidly growing mutual information will change their distribution over the entire DNMnet. We will use two indices to detect the anisotropy of mutual information in the network—the distribution entropy (DE) and network entropy (NE)—whose definitions are shown in the Methods. With the time interval Δt=20 ms, we calculated the DE and NE of the HC, UHR, and PD groups (Figures 2A–B for the dD case, and Supplementary Figures S7A–B for the fD case). The HC group always exhibited significantly higher DE and NE than the PD and UHR groups, while the difference between the PD and UHR groups was not significant. These results support the critical brain hypothesis that the HC group should be at a critical state, whose entropy is high possibly to ensure adaptivity. In addition, correlations between positive symptoms (clinical score) of PD and DE/NE were calculated (Supplementary Figure S11), which also implied that the worse the positive symptoms are, the more low-entropy and stable the brain is. In Supplementary Figure S12, we also showed that the orders of DE/NE for three groups were not sensitively influenced by the parameter Δt.
**FIGURE 2:** *Entropy and causality difference between the HC, UHR, and PD groups in the duration deviant (dD) case. (A) Violin plot for the distribution entropy of mutual information between each pair of nodes. (B) Network entropy. For “HC > UHR” and “HC > PD”, p-values were calculated by the one-sided Wilcoxon rank-sum test, while for “UHR ≈ PD”, the two-sided Wilcoxon rank-sum test was used. The HC group exhibited significantly higher entropy than the PD and UHR groups. The PD and UHR groups did not significant differences. (C–E) Heatmaps of the direct causality between 64 electrodes measured by the conditional transfer entropy (CTE). (F) shows probability density functions of the CTE. On average the PD has much stronger direct causalities between electrodes than the HC/UHR.*
In the biological sense, the high entropy at the critical state reflects a possibility that the brain can send information to different brain regions in time to deal with various stimuli.
## Criticality of the Brain: Causality Pattern
The causality pattern is another evidence for criticality, which can be detected by the conditional transfer entropy (CTE) between different nodes, which is given as CTEx→y=CMI(xt, yt+Δt|x¯t)=∫∫∫p(xt, yt+Δt,x¯t)logp(xt,yt+Δt|x¯t)p(xt|x¯t)p(yt+Δt|x¯t)dxdydx¯t, [5] where CMI denotes the condition mutual information, x and y are two variables to detect causality, x¯ includes all other variables except x, and Δt is the time interval. CTE is different from mutual information because it excludes indirect influences and is usually used to detect the direct causality. If the brain is at a stable state, causalities between different electrodes should be apparently present, while for critical states, random-like activities decrease the possibility of the appearance of regular causality patterns. With the time interval Δt=20 ms, we can obtain heatmaps of the mean CTE between every pair of 64 electrodes for the PD, UHR, and HC groups (Figures 2C–E for the dD case, and Supplementary Figures S7C–E for the fD case). The larger the value of CTE is, the stronger is the direct causality between variables. We found that the PD had a stronger causality network compared with the UHR/HC, which implies that psychosis makes the brain fall into a more stable state. In contrast, it is difficult for the normal brain to form general causality patterns at a critical state, as shown by the heatmap of the HC group. The probability density functions of CTE for each group are plotted in Figure 2F for a better intuition. In Supplementary Figure S12, we also showed that the orders of CTE for three groups were not sensitively influenced by the parameter Δt.
## ANCOVA for DNMIndex, DE, and NE
For EEG data, we performed analysis of covariance (ANCOVA) with age, premorbid IQ, and antipsychotic dose as covariates because these variables were significantly different among groups (Supplementary Table S1). For the duration deviant experiments, all these covariates showed no significant effects (F1,91 < 2.58, $p \leq 0.11$). Therefore, these factors did not affect our findings in the duration deviant experiments. For the frequency deviant experiments, premorbid IQ showed significant effects, whereas age and antipsychotic dose showed no significant effects (F1,91 < 0.75, $p \leq 0.39$). Therefore, we performed ANCOVA with premorbid IQ as a covariate in the frequency deviant experiments.
ANCOVA of the DNMIndex in the frequency deviant condition revealed significant effects of groups (F2,95 = 6.61, $$p \leq 0.01$$) and premorbid IQ (F1,95 = 11.08, $p \leq 0.001$). Post-hoc analyses revealed that the HC group had a significantly higher mean value than the PD ($p \leq 0.001$) and the UHR ($$p \leq 0.004$$) groups, while the difference between the PD and UHR groups was not significant ($$p \leq 1.00$$).
ANCOVA of DE in the frequency deviant condition revealed significant effects of groups (F2,95 = 5.72, $$p \leq 0.005$$) and premorbid IQ (F1,95 = 6.80, $$p \leq 0.01$$). Post-hoc analyses revealed that the HC group had a significantly higher mean value than the PD group ($$p \leq 0.003$$), while the difference between the HC and UHR groups ($$p \leq 0.33$$) and the difference between the UHR and PD groups ($$p \leq 0.48$$) were not significant. ANCOVA of NE in the frequency deviant condition revealed significant effects of groups (F2,95 = 5.44, $$p \leq 0.006$$) and premorbid IQ (F1,95 = 8.43, $$p \leq 0.005$$). Post-hoc analyses revealed that the HC group had a significantly higher mean value than the PD group ($$p \leq 0.004$$), while the difference between the HC and UHR groups ($$p \leq 0.58$$) and the difference between the UHR and PD groups ($$p \leq 0.31$$) were not significant.
These findings revealed the lower DNMIndex, lower distribution entropy, and lower network entropy of the frequency deviant experiments in the PD group even after controlling premorbid IQ. In UHR, the DNMIndex was lower, whereas DE and NE showed insignificant difference for the frequency deviant experiments after controlling premorbid IQ.
## Application of Criticality: Risk of Mental Disease
The level of the criticality of the brain contained in the DNMnet can provide a quantitative measure to weigh the risk of mental disease. From the EEG samples, we applied an ensemble classifier based on the DNMnet (see Methods). AUCs and cross validations were used to test the training accuracy and to ensure the reliability of the classifier (Supplementary Figure S8). With the final risk output, we found that the PD group exhibited the highest risk, and the HC group showed the lowest risk, whereas the UHR group evinced a medial risk (Figure 3A for the dD case and Figure 3B for the fD case). This risk analysis indicates that DNM results could help distinguish between the three groups and has a potential to guide prepsychotic diagnosis.
**FIGURE 3:** *Risk analysis of the HC, UHR, and PD groups for the duration deviant experiment (dD) dataset and the frequency deviant experiment (fD) dataset with DNM features. (A) and (B) Violin plots of risks for three groups in the dD and fD cases, respectively. The HC always has the lowest and the PD has the highest risk, whereas the UHR exhibits a medial risk. The one-sided Wilcoxon rank-sum test was used to determine the significance of difference between groups.*
## Discussion
In this study, we built a dynamical network model to explore the criticality of the brain. It not only provides a dynamics-based explanation of the traditional MMN patterns, but also uses features of the model (the DNMIndex, the entropy, and the causality pattern) to support the fact that the healthy brain is around the critical state. A risk index for PDs also indicates the practicality of the DNMnet, which can highlight the criticality of the neuronal network.
Criticality is an important concept for the activity of the brain. Around the critical state, the healthy brain can adapt to new situations and deal with various stimuli in a timely manner. However, if the DNMnet in the brain falls into a more stable state, as in the PD group, the stability may lead to cognitive inflexibility and functional impairments (Serrano-Guerrero et al., 2020). Previous studies reported altered connectivity underlying MMN in psychotic disorders (Dima et al., 2012; Ranlund et al., 2016; Braeutigam et al., 2018). However, MMN does not highlight the dynamical mechanism to clarify its reduction in psychotic disorders. Therefore, the DNMnet used in the current study reveals that MMN is only an external presentation of the changes in the criticality.
The UHR group in this study showed lower DNMIndex, DE, and NE compared to the HC group. These findings suggest that the DNMnet of the UHR has already fallen into the stable states before onset of psychosis. However, in the direct causality network constructed by the CTE, we found that the direct causality between electrodes for the UHR group was as weak as the HC. As the CTE detects the direct influence from electrode to electrode, while the DE and NE consider the long-range global communication, we speculate a possibility that the UHR group has healthy local functions similar to the HC group, but impaired global functions similar to the PD group. The lack of variation may eventually pull the UHR into a more stable state, similar to the PD group. It could be one way to explain why the brain of UHR individuals works normally as HC compared to those of PD, but eventually the slowing down variation makes it fall into the stable PD state.
In the application of risk analyses, we only used the information from the DNMnet, which is a subnetwork of the entire 64 channels, to estimate the risk of psychosis for each sample. We also compared it with the result of the risk estimated by the MMN amplitude (Supplementary Table S4; Supplementary Figures S9, S10). The DNMnet exhibited the same efficiency as the MMN risk which uses features from all channels. These findings indicate that the much simpler DNMnet can serve as not only a mechanistic biomarker (Pine and Leibenluft, 2015) but also as a pragmatic biomarker (Paulus, 2015) for diagnosing mental diseases.
There are several limitations in this study. First, potential medication effects may have influenced the findings because participants in UHR or PD took medication. Whether medication biases dynamic network markers needs to be clarified with medication-naïve participants in the future study. Second, this work is designed as a cross-sectional study. Therefore, we could not identify when the brain falls into stable states. Future longitudinal studies will clarify the trajectory of dynamic states in psychotic disorders.
In conclusion, we found that the criticality is a key feature of the healthy brain. Based on the DNM theory, the brain of HC was close to the critical states, whereas those of UHR individuals or patients with PD fell into more stable states. Our analysis not only offers a new viewpoint toward understanding the dynamic brain but also provides a possibility of a biophysical approach for researching other mental disorders.
## Subjects
A total of 49 HC subjects, 24 UHR individuals, and 29 patients with PD participated in this study, which contained participants of the Integrative Neuroimaging Studies for Schizophrenia Targeting Early Intervention and Prevention (IN-STEP) (Koike et al., 2013). Detailed participant information is provided in the Supplementary Notes and Supplementary Table S1. The Research Ethics Committee of the Faculty of Medicine, The University of Tokyo, approved this study (approval No. 629, 2226). We conducted this study in accordance with the Declaration of Helsinki. Written informed consent was obtained from all the participants.
## Electroencephalography Data
A 64-channel Geodesic EEG System (Electrical Geodesics Inc, Eugene, OR) was used to acquire EEG data. Electrodes were referenced to the vertex, and impedances were kept below 50 kΩ. The sampling rate was 500 Hz, and the analog filter bandpass was set at 0.1–100 Hz. The locations of the EEG electrodes are shown in Supplementary Figure S1A.
## Stimuli and Procedure
The two-tone auditory oddball paradigm with 2000 stimuli was performed for each subject when obtaining EEG data. For the duration deviant (dD) experiments, $90\%$ stimuli were standard tones (1,000 Hz, 50 ms) and $10\%$ were deviant tones (1,000 Hz, 100 ms). For the frequency deviant (fD) experiments, $90\%$ stimuli were standard tones (1,000 Hz, 50 ms) and $10\%$ were deviant tones (1,200 Hz, 50 ms). All stimuli were 80 dB SPL with a 1 ms rise/fall time. The stimulus onset asynchrony was 500 ms. The oddball paradigms were counter-balanced, and tones were presented binaurally through earphones while participants watched a silent cartoon.
## EEG Data Preprocessing
Original 64-channel signal files (Supplementary Figure S1B) were input into MATLAB (9.3.0) and were further preprocessed using the EEGLAB (v14_1_1b) package (Delorme and Makeig, 2004) (Supplementary Figures S1C–E). Detailed preprocessing procedures are provided in the Supplementary Material. Finally, corrected epochs (deviant epochs with the mean signal of standard epochs being subtracted for each sample) were utilized for further analyses (Supplementary Figure S1F).
## Auditory Mismatch Negativity
Corrected epochs were obtained by subtracting the event-related potential (ERP) waveforms in response to the standard stimuli from those in response to the deviant stimuli. We defined the peak latency as the most negative peak between 100 and 250 ms relative to the onset, and the MMN amplitude was calculated as the mean signal around the peak latency (we used the window of 135–205 ms for the duration MMN and the window of 100–200 ms for the frequency MMN) (Nagai et al., 2013). The MMN patterns for the 64 electrodes in the three groups (PD, UHR, and HC) under dD and fD are shown in Supplementary Figures S3 and S4, respectively. Using the MMN amplitude for each channel as the feature, we can compute p-values of the group difference under the Wilcoxon rank-sum test. After FDR correction, we identified 12 electrodes (Nos. 3, 4, 5, 9, 17, 18, 22, 30, 43, 54, 55, and 58) in the dD and 12 electrodes (Nos. 3, 4, 5, 8, 9, 17, 18, 30, 43, 54, 55, and 58) in the fD that exhibited significant difference (adjusted $p \leq 0.05$) between the HC and UHR/PD groups, albeit insignificant difference (adjusted $p \leq 0.05$) between the UHR and the PD groups. It should be noticed that all electrodes with significant MMN difference were located in the fronto-central area.
## Dynamical Network (DNMnet)
The neuronal signal under standard stimuli is supposed to obey the dynamics in Eq. 1, whereas the corrected epochs after preprocessing follow Eq. 2. Criticality is defined as when the dynamics is around a codimension-1 local bifurcation (Chen et al., 2012). The network of electrodes can be partitioned into a DNM group and a non-DNM group (Chen et al., 2012) by measuring the fluctuations of signals and their correlations between electrodes for each sample. The DNMnet is the leading network for complex systems that drives the system toward or away from critical states. Two conditions for constructing the DNMnet were used in this study:1. Any member of the nodes in DNMnet is highly fluctuating around a critical point.2. Any edge in the DNMnet becomes very strong around the critical state.
The one-sided paired-sample t-test and the one-sided Wilcoxon rank-sum test were used for selected nodes and edges, respectively. We selected nodes and edges with significant statistical difference between HC and PD groups. Significance was set at $p \leq 10$−3 for nodes and $p \leq 0.005$ for edges after FDR correction. Supplementary Tables S2, S3 and Supplementary Figures S5, S6 show the statistical properties of the nodes and edges in DNMnets. Specifically, 20 nodes (the one-sided paired-sample t-test with $p \leq 10$−3 after FDR correction) were extracted from the dD experiment dataset (Supplementary Figure S5). 12 edges with significantly increasing correlations (the one-sided Wilcoxon rank-sum test with $p \leq 0.005$ after FDR correction) were also detected (Supplementary Figure S6). No edge with significantly decreasing correlation (the one-sided Wilcoxon rank-sum test with significance level α=0.05) in HC was found. We denoted the 20 nodes and 12 edges as the DNMnet for the dD case (Figure 1B). The same analysis was also applied to the fD experiment dataset, from which 17 nodes (the one-sided paired-sample t-test with $p \leq 10$−3 after FDR correction, Supplementary Table S2) and 37 edges (the one-sided Wilcoxon rank-sum test with $p \leq 0.005$ after FDR correction, Supplementary Table S3) were selected to form the DNMnet for the fD case (Figure 1D). No significant decreasing correlation (the one-sided Wilcoxon rank-sum test with significance level α=0.05) in HC was found in the fD experiment dataset, either.
We remark that the Lyapunov *Exponent is* also a possible measure for detecting criticality. We note that the dynamical network we used here refers to the leading subnetwork (Chen et al., 2012). In the celebrated paper on synchronization (Arenas et al., 2008), the dynamical network is considered as the whole system of the interacting dynamical units. In this manuscript, most of the time we use the notation the DNMnet to avoid the confusion.
## Measures for the Criticality
When calculating the DNMIndex using Eq. 3, the corrected epochs were used as the signal. Nodes and edges in the DNMnet were considered.
We denote a network as G = {V, E}, where V = {x1, x2, …, xn} are n nodes and E = {ei,j|i, j∈ {1, 2, …, n} and i≠ j} are Ne = n(n−1) different directed edges. Using the mutual information in Eq. 4, the distribution entropy (DE) is defined as the normalized entropy of all mutual information on edges, as follows: Entd = −1logNe∑$i = 1$n∑$j = 1$j≠inpijlogpij, [6] where pij=Ixi→xj∑$i = 1$n∑$j = 1$, j≠inIxi→xj [7] is the normalized mutual information over the entire network. Mutual information Ixi→xj is calculated under Gaussian approximation as Ix→y= −$\frac{1}{2}$⋅log(1−ρxtyt+Δt2), where ρxy is the Pearson correlation coefficient between x and y. In contrast, the network entropy (NE) is defined as follows Entn= −1n∑$i = 1$n∑$j = 1$j≠inqijlogqij, [8] where qij=Ixi→xj∑$j = 1$, j≠inIxi→xj [9] is the normalized mutual information exiting each node. We use the full network between nodes in DNMnet to calculate the DE and NE. If the brain works around a critical state, some increasing mutual information will expand their distribution density on the network. Thus, DE and NE are larger at a critical state than at non-critical states. High entropy also reflects a possibility that the brain can send information to different brain regions in time to deal with various stimuli.
For the CTE defined in Eq. 5, every pair between 64 electrodes is considered to construct the causality patterns in (Figures 2C–E). CTE is also calculated under Gaussian approximation as CTEx→y= −$\frac{1}{2}$⋅log(1−ρxtyt+Δt|x¯t 2), where ρxy|z is the partial correlation between x and y conditional on z.
## Estimation of Psychotic Risk
Using the knowledge of the DNMnet after dynamical network analysis, we applied an ensemble classification on the EEG data to estimate the risk index for psychosis. A flowchart of the risk estimation is shown in Supplementary Figure S2.
We assume that there are nk corrected deviant epochs after preprocessing for sample k, which may belong to one of the PD, HC, or UHR groups. PD and HC epochs are used as the training set. Features for training comprise the elements in the covariance matrix for each sample. We assigned each epoch a label: 1 for PD and 0 for HC. Ensemble classification was then applied, and cross validation was performed. Using the classification tree, we can test all PD, HC, and UHR epochs. Finally, the risk for person k is defined as the mean output (the label value) of the epochs belonging to this sample, and this is defined as Rk =1nk∑$i = 1$nkOi, [10] where *Rk is* the risk for sample k, nk is the number of corrected epochs in sample k, and *Oi is* the classification output of the ith epoch. Rk is always between 0 and 1. To verify the accuracy and compare different classifiers, we used a risk accuracy index. If there are nHC samples in HC, nUHR samples in UHR, and nPD samples in PD, we sort the risks for all samples in an ascending order from smallest to largest. The risk accuracy for the HC group is set as the percent of HC samples in the first nHC smallest Rk s. The risk accuracy for the UHR group is set as the percent of UHR samples in the medial nUHR medium Rk s. The risk accuracy for the PD group is set as the percent of PD samples in the last nPD largest Rk s. Furthermore, the risk accuracy for the disease group is set as the percent of UHR and PD samples in the last nUHR+nPD as the largest Rk. The training accuracy of the classifier is defined as 1−loss. Thus, the DNM classifier has more than $80\%$ training accuracy (Supplementary Table S4). The risk accuracy indices for the test set were also calculated and are listed in Supplementary Table S4.
## Statistical Analyses
We used SPSS (IBM Corp., New York, United States) and MATLAB 9.3.0 (MathWorks Inc.) for statistical analyses. For demographic and clinical data, we performed a chi-squared test, independent t-tests, and analysis of variance (ANOVA) for comparison among groups. Bonferroni correction was performed in post-hoc analyses of ANOVA. For EEG data, we performed paired-sample t-tests and Wilcoxon rank-sum tests to compare difference in the MMN, DNM measures, and risks between different groups. When we quantified an alternative hypothesis of a one-side inequality, such as some index with HC > PD, we used the one-side Wilcoxon rank-sum tests. While we quantified an alternative hypothesis of a two-side inequality, such as some index with UHR≠PD (i.e., UHR < PD or UHR > PD), we used the two-side Wilcoxon rank-sum tests. False discovery rate (FDR) was also controlled for multiple comparisons (Benjamini and Hochberg, 1995). Analysis of covariance (ANCOVA) with age, premorbid IQ, and antipsychotic dose as covariates were also conducted, which showed that they did not affect the difference of measures for criticality between groups.
## Data Availability Statement
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding authors.
## Ethics Statement
The studies involving human participants were reviewed and approved by the Research Ethics Committee of the Faculty of Medicine, The University of Tokyo (approval No. 629, 2226). Written informed consent to participate in this study was provided by the participants.
## Author Contributions
JS, KIK, and KA designed the study. KEK, MT, MF, KU, DK, TA, and KIK performed the experiments and collected the data. JS, LC, and KA designed the model. JS and KEK analyzed the data, and performed the calculations. JS, KEK, and KA wrote the manuscript. All authors discussed the results and revised the manuscript.
## Conflict of Interest
KIK reports grants from Lily, grants from MSD, grants and personal fees from Astellas, grants and personal fees from Takeda, grants and personal fees from Dainippon-Sumitomo, grants from Novartis, grants from Tanabe-Mitsubishi, grants from Eisai, grants and personal fees from Otsuka, grants from Shionogi, grants from Ono Pharma, personal fees from Fuji-film-Wako, personal fees from Yoshitomi, personal fees from Kyowa, personal fees from Janssen, and personal fees from Meiji Seika Pharma, outside the submitted work during the past 36 months. KA reports grants and personal fees from KKE, grants and personal fees from NEC, grants from Sysmex, personal fees from Novo Nordisk Japan, and grants and personal fees from Toyota Central R&D Labs., outside the submitted work during the past 36 months.
The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary Material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnetp.2021.755685/full#supplementary-material
## References
1. Arenas A., Díaz-Guilera A., Kurths J., Moreno Y., Zhou C.. **Synchronization in Complex Networks**. *Phys. Rep.* (2008) **469** 93-153. DOI: 10.1016/j.physrep.2008.09.002
2. Beggs J.. **Editorial: Can There Be a Physics of the Brain**. *Phys. Rev. Lett.* (2015) **114** 220001. DOI: 10.1103/physrevlett.114.220001
3. Beggs J. M., Plenz D.. **Neuronal Avalanches in Neocortical Circuits**. *J. Neurosci.* (2003) **23** 11167-11177. DOI: 10.1523/jneurosci.23-35-11167.2003
4. Beggs J. M., Timme N.. **Being Critical of Criticality in the Brain**. *Front. Physio.* (2012) **3** 163. DOI: 10.3389/fphys.2012.00163
5. Benjamini Y., Hochberg Y.. **Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing**. *J. R. Stat. Soc. Ser. B (Methodological)* (1995) **57** 289-300. DOI: 10.1111/j.2517-6161.1995.tb02031.x
6. Bodatsch M., Brockhaus-Dumke A., Klosterkötter J., Ruhrmann S.. **Forecasting Psychosis by Event-Related Potentials-Systematic Review and Specific Meta-Analysis**. *Biol. Psychiatry* (2015) **77** 951-958. DOI: 10.1016/j.biopsych.2014.09.025
7. Braeutigam S., Dima D., Frangou S., James A.. **Dissociable Auditory Mismatch Response and Connectivity Patterns in Adolescents with Schizophrenia and Adolescents with Bipolar Disorder with Psychosis: A Magnetoencephalography Study**. *Schizophrenia Res.* (2018) **193** 313-318. DOI: 10.1016/j.schres.2017.07.048
8. Chen L., Liu R., Liu Z.-P., Li M., Aihara K.. **Detecting Early-Warning Signals for Sudden Deterioration of Complex Diseases by Dynamical Network Biomarkers**. *Sci. Rep.* (2012) **2** 342. DOI: 10.1038/srep00342
9. Chung D. T., Ryan C. J., Hadzi-Pavlovic D., Singh S. P., Stanton C., Large M. M.. **Suicide Rates after Discharge from Psychiatric Facilities**. *JAMA Psychiatry* (2017) **74** 694-702. DOI: 10.1001/jamapsychiatry.2017.1044
10. Cocchi L., Gollo L. L., Zalesky A., Breakspear M.. **Criticality in the Brain: A Synthesis of Neurobiology, Models and Cognition**. *Prog. Neurobiol.* (2017) **158** 132-152. DOI: 10.1016/j.pneurobio.2017.07.002
11. Cooper R., Osselton J. W., Shaw J. C.. *EEG Technology* (2014)
12. Dakos V., Bascompte J.. **Critical Slowing Down as Early Warning for the Onset of Collapse in Mutualistic Communities**. *Proc. Natl. Acad. Sci. USA* (2014) **111** 17546-17551. DOI: 10.1073/pnas.1406326111
13. Delorme A., Makeig S.. **EEGLAB: an Open Source Toolbox for Analysis of Single-Trial EEG Dynamics Including Independent Component Analysis**. *J. Neurosci. Methods* (2004) **134** 9-21. DOI: 10.1016/j.jneumeth.2003.10.009
14. Dima D., Frangou S., Burge L., Braeutigam S., James A. C.. **Abnormal Intrinsic and Extrinsic Connectivity within the Magnetic Mismatch Negativity Brain Network in Schizophrenia: a Preliminary Study**. *Schizophrenia Res.* (2012) **135** 23-27. DOI: 10.1016/j.schres.2011.12.024
15. Erickson M. A., Ruffle A., Gold J. M.. **A Meta-Analysis of Mismatch Negativity in Schizophrenia: from Clinical Risk to Disease Specificity and Progression**. *Biol. Psychiatry* (2016) **79** 980-987. DOI: 10.1016/j.biopsych.2015.08.025
16. Haldeman C., Beggs J. M.. **Critical Branching Captures Activity in Living Neural Networks and Maximizes the Number of Metastable States**. *Phys. Rev. Lett.* (2005) **94** 058101. DOI: 10.1103/PhysRevLett.94.058101
17. Kantrowitz J. T., Epstein M. L., Lee M., Lehrfeld N., Nolan K. A., Shope C.. **Improvement in Mismatch Negativity Generation during D-Serine Treatment in Schizophrenia: Correlation with Symptoms**. *Schizophrenia Res.* (2018) **191** 70-79. DOI: 10.1016/j.schres.2017.02.027
18. Kéfi S., Dakos V., Scheffer M., Van Nes E. H., Rietkerk M.. **Early Warning Signals Also Precede Non-catastrophic Transitions**. *Oikos* (2013) **122** 641-648. DOI: 10.1111/j.1600-0706.2012.20838.x
19. Koike S., Takano Y., Iwashiro N., Satomura Y., Suga M., Nagai T.. **A Multimodal Approach to Investigate Biomarkers for Psychosis in a Clinical Setting: the Integrative Neuroimaging Studies in Schizophrenia Targeting for Early Intervention and Prevention (IN-STEP) Project**. *Schizophrenia Res.* (2013) **143** 116-124. DOI: 10.1016/j.schres.2012.11.012
20. Kuehn C.. **A Mathematical Framework for Critical Transitions: Bifurcations, Fast-Slow Systems and Stochastic Dynamics**. *Physica D: Nonlinear Phenomena* (2011) **240** 1020-1035. DOI: 10.1016/j.physd.2011.02.012
21. Lee H., Golkowski D., Jordan D., Berger S., Ilg R., Lee J.. **Relationship of Critical Dynamics, Functional Connectivity, and States of Consciousness in Large-Scale Human Brain Networks**. *NeuroImage* (2019) **188** 228-238. DOI: 10.1016/j.neuroimage.2018.12.011
22. Lesterhuis W. J., Bosco A., Millward M. J., Small M., Nowak A. K., Lake R. A.. **Dynamic versus Static Biomarkers in Cancer Immune Checkpoint Blockade: Unravelling Complexity**. *Nat. Rev. Drug Discov.* (2017) **16** 264-272. DOI: 10.1038/nrd.2016.233
23. Li M., Zeng T., Liu R., Chen L.. **Detecting Tissue-specific Early Warning Signals for Complex Diseases Based on Dynamical Network Biomarkers: Study of Type 2 Diabetes by Cross-Tissue Analysis**. *Brief. Bioinform.* (2013) **15** 229-243. DOI: 10.1093/bib/bbt027
24. Liu R., Chen P., Aihara K., Chen L.. **Identifying Early-Warning Signals of Critical Transitions with strong Noise by Dynamical Network Markers**. *Sci. Rep.* (2015) **5** 17501. DOI: 10.1038/srep17501
25. Liu R., Li M., Liu Z.-P., Wu J., Chen L., Aihara K.. **Identifying Critical Transitions and Their Leading Biomolecular Networks in Complex Diseases**. *Sci. Rep.* (2012) **2** 813. DOI: 10.1038/srep00813
26. Liu R., Wang X., Aihara K., Chen L.. **Early Diagnosis of Complex Diseases by Molecular Biomarkers, Network Biomarkers, and Dynamical Network Biomarkers**. *Med. Res. Rev.* (2014b) **34** 455-478. DOI: 10.1002/med.21293
27. Liu R., Yu X., Liu X., Xu D., Aihara K., Chen L.. **Identifying Critical Transitions of Complex Diseases Based on a Single Sample**. *Bioinformatics* (2014a) **30** 1579-1586. DOI: 10.1093/bioinformatics/btu084
28. Liu X., Chang X., Leng S., Tang H., Aihara K., Chen L.. **Detection for Disease Tipping Points by Landscape Dynamic Network Biomarkers**. *Natl. Sci. Rev.* (2019) **6** 775-785. DOI: 10.1093/nsr/nwy162
29. Massobrio P., de Arcangelis L., Pasquale V., Jensen H. J., Plenz D.. **Criticality as a Signature of Healthy Neural Systems**. *Front. Syst. Neurosci.* (2015) **9** 22. DOI: 10.3389/fnsys.2015.00022
30. Nagai T., Tada M., Kirihara K., Yahata N., Hashimoto R., Araki T.. **Auditory Mismatch Negativity and P3a in Response to Duration and Frequency Changes in the Early Stages of Psychosis**. *Schizophr Res.* (2013) **150** 547-554. DOI: 10.1016/j.schres.2013.08.005
31. Paulus M. P.. **Pragmatism Instead of Mechanism**. *JAMA Psychiatry* (2015) **72** 631-632. DOI: 10.1001/jamapsychiatry.2015.0497
32. Perez V. B., Tarasenko M., Miyakoshi M., Pianka S. T., Makeig S. D., Braff D. L.. **Mismatch Negativity Is a Sensitive and Predictive Biomarker of Perceptual Learning during Auditory Cognitive Training in Schizophrenia**. *Neuropsychopharmacol* (2017) **42** 2206-2213. DOI: 10.1038/npp.2017.25
33. Petermann T., Thiagarajan T. C., Lebedev M. A., Nicolelis M. A., Chialvo D. R., Plenz D.. **Spontaneous Cortical Activity in Awake Monkeys Composed of Neuronal Avalanches**. *Proc. Natl. Acad. Sci. U S A.* (2009) **106** 15921-15926. DOI: 10.1073/pnas.0904089106
34. Pine D. S., Leibenluft E.. **Biomarkers with a Mechanistic Focus**. *JAMA Psychiatry* (2015) **72** 633-634. DOI: 10.1001/jamapsychiatry.2015.0498
35. Ranlund S., Adams R. A., Díez Á., Constante M., Dutt A., Hall M. H.. **Impaired Prefrontal Synaptic Gain in People with Psychosis and Their Relatives during the Mismatch Negativity**. *Hum. Brain Mapp.* (2016) **37** 351-365. DOI: 10.1002/hbm.23035
36. Richard A., Boullu L., Herbach U., Bonnafoux A., Morin V., Vallin E.. **Single-cell-based Analysis Highlights a Surge in Cell-To-Cell Molecular Variability Preceding Irreversible Commitment in a Differentiation Process**. *Plos Biol.* (2016) **14** e1002585. DOI: 10.1371/journal.pbio.1002585
37. Scheffer M., Bascompte J., Brock W. A., Brovkin V., Carpenter S. R., Dakos V.. **Early-warning Signals for Critical Transitions**. *Nature* (2009) **461** 53-59. DOI: 10.1038/nature08227
38. Scheffer M., Carpenter S., Foley J. A., Folke B.. **Catastrophic Shifts in Ecosystems**. *Nature* (2001) **413** 591-596. DOI: 10.1038/35098000
39. Scheffer M.. *Critical Transitions in Nature and Society* (2009)
40. Serrano-Guerrero E., Ruiz-Veguilla M., Martín-Rodríguez A., Rodríguez-Testal J. F.. **Inflexibility of Beliefs and Jumping to Conclusions in Active Schizophrenia**. *Psychiatry Res.* (2020) **284** 112776. DOI: 10.1016/j.psychres.2020.112776
41. Shew W. L., Yang H., Petermann T., Roy R., Plenz D.. **Neuronal Avalanches Imply Maximum Dynamic Range in Cortical Networks at Criticality**. *J. Neurosci.* (2009) **29** 15595-15600. DOI: 10.1523/JNEUROSCI.3864-09.2009
42. Shew W. L., Yang H., Yu S., Roy R., Plenz D.. **Information Capacity and Transmission Are Maximized in Balanced Cortical Networks with Neuronal Avalanches**. *J. Neurosci.* (2011) **31** 55-63. DOI: 10.1523/JNEUROSCI.4637-10.2011
43. Shi J., Li T., Chen L.. **Towards a Critical Transition Theory under Different Temporal Scales and Noise Strengths**. *Phys. Rev. E* (2016) **93** 032137. DOI: 10.1103/PhysRevE.93.032137
44. Yang B., Li M., Tang W., Liu W., Zhang S., Chen L.. **Dynamic Network Biomarker Indicates Pulmonary Metastasis at the Tipping point of Hepatocellular Carcinoma**. *Nat. Commun.* (2018) **9** 2. DOI: 10.1038/s41467-018-03024-2
|
---
title: Effects of Acute Partial Sleep Deprivation and High-Intensity Interval Exercise
on Postprandial Network Interactions
authors:
- Zacharias Papadakis
- Sergi Garcia-Retortillo
- Panagiotis Koutakis
journal: Frontiers in Network Physiology
year: 2022
pmcid: PMC10013041
doi: 10.3389/fnetp.2022.869787
license: CC BY 4.0
---
# Effects of Acute Partial Sleep Deprivation and High-Intensity Interval Exercise on Postprandial Network Interactions
## Abstract
Introduction: High-intensity interval exercise (HIIE) is deemed effective for cardiovascular and autonomic nervous system (ANS) health-related benefits, while ANS disturbance increases the risk for cardiovascular disease (CVD). Postprandial lipemia and acute-partial sleep deprivation (APSD) are considered as CVD risk factors due to their respective changes in ANS. Exercising in the morning hours after APSD and have a high-fat breakfast afterwards may alter the interactions of the cardiovascular, autonomic regulation, and postprandial lipemic systems threatening individuals’ health. This study examined postprandial network interactions between autonomic regulation through heart rate variability (HRV) and lipemia via low-density lipoprotein (LDL) cholesterol in response to APSD and HIIE.
Methods: Fifteen apparently healthy and habitually good sleepers (age 31 ± 5.2 SD yrs) completed an acute bout of an isocaloric HIIE (in form of 3:2 work-to-rest ratio at 90 and $40\%$ of VO2 reserve) after both a reference sleep (RSX) and 3–3.5 h of acute-partial sleep deprivation (SSX) conditions. HRV time and frequency domains and LDL were evaluated in six and seven time points surrounding sleep and exercise, respectively. To identify postprandial network interactions, we constructed one correlation analysis and one physiological network for each experimental condition. To quantify the interactions within the physiological networks, we also computed the number of links (i.e., number of significant correlations).
Results: We observed an irruption of negative links (i.e., negative correlations) between HRV and LDL in the SSX physiological network compared to RSX. Discussion: We recognize that a correlation analysis does not constitute a true network analysis due to the absence of analysis of a time series of the original examined physiological variables. Nonetheless, the presence of negative links in SSX reflected the impact of sleep deprivation on the autonomic regulation and lipemia and, thus, revealed the inability of HIIE to remain cardioprotective under APSD. These findings underlie the need to further investigate the effects of APSD and HIIE on the interactions among physiological systems.
## 1 Introduction
A plethora of data suggest the cardioprotective effects of exercise due to its impact either on reducing the associated to cardiovascular disease risk factors (e.g., hypertension, lipidemia, diabetes, and insulin resistance, obesity) or having a direct effect on processes and functions of different physiological systems (e.g., atherosclerotic process and cardiovascular function) (Powers et al., 2002; Phrommintikul et al., 2022). Moreover, exercise presents cardioprotective effects due to autonomic nervous system (ANS) adjustments on heart rate variability (HRV) and baroreflex sensitivity (Grant et al., 2012; Subramanian et al., 2019). High-intensity interval exercise (HIIE), exercise performed in brief, successive intervals consisting of a period of high-intensity (e.g., >$80\%$ of peak oxygen consumption) followed by lower-intensity recovery periods, is proposed as alternative time-efficient exercise method with wide health related benefits during both fasting and postprandial states (Gibala, 2007; Wisloff et al., 2007; Tjonna et al., 2008; Tyldum et al., 2009; Gibala et al., 2012; Weston et al., 2014; Bond et al., 2015; Sawyer et al., 2016; Ramírez-Vélez et al., 2018; Tucker et al., 2018). The protective cardiometabolic effect of HIIE is postulated due to greater induced antioxidant status both during and after HIIE (Harris et al., 2008; Di Francescomarino et al., 2009; Fisher-Wellman and Bloomer, 2009; Tyldum et al., 2009; Gabriel et al., 2012). On top of that, HIIE has a significantly greater impact on ANS (Gibala et al., 2012; Bhati and Moiz, 2017), with HIIE work to rest (W:R) ratio of 1:2 to be proposed as highly effective for cardiovascular and autonomic related health benefits (Heydari et al., 2013; Ramírez-Vélez et al., 2016; Ramirez-Velez et al., 2020). It has been shown that ANS disturbance due to exercise, presented as increased sympathetic tone and parasympathetic withdrawal leads to a decreased HRV, which in turn increases the risk for cardiovascular disease (CVD) (Besnier et al., 2017). HIIE affects simultaneously the interactions among physiological systems and organs where the strength of these interactions may represent different physiological states and pathological conditions. Such pathological conditions may be manifested due to failure of the system to perform various coupling and feedback interactions under an integrated physiological system with linear and non-linear characteristics (Bashan et al., 2012; Ivanov and Bartsch, 2014; Bartsch et al., 2015).
Postprandial lipemia is an independent risk factor for CVD (Hyson et al., 2003; Cromwell et al., 2007). Postprandial increased concentration of low-density (LDL) lipoprotein cholesterol induces a heightened inflammatory state in the vascular wall that is highly susceptible to oxidative changes (Littlefield and Grandjean, 2015) promoting vascular endothelial dysfunction (Wallace et al., 2010). In apparently healthy men, LDL cholesterol had inverse relation to HRV (Kupari et al., 1993; Christensen et al., 1999), which is a marker of ANS activation and linked to CVD (Thayer et al., 2010; Chung et al., 2020).
Short sleep duration has been linked to increased risks of morbidity and mortality (Hale et al., 2020; Krittanawong et al., 2020). Acute partial sleep deprivation (APSD), of less than 5 h of sleep, is associated to CVD (Tobaldini et al., 2017; Liew and Aung, 2021). This association to CVD is attributed to changes in autonomic nervous system (ANS) (Miller and Cappuccio, 2013; Tobaldini et al., 2014; Wright et al., 2015; Tobaldini et al., 2017; American Sleep Association, 2018; Seravalle et al., 2018; Liu and Chen, 2019; Oliver et al., 2020). Under APSD such detrimental health effects may represent the breakdown of dynamic network interactions among organs systems and metabolic processes, such as ANS and postprandial lipemia and their failure to ensure a healthy vital status (Bartsch et al., 2015; Ivanov et al., 2016; Balague et al., 2020; Lehnertz et al., 2020). It seems that after sleep deprivation the pronounced HR and HRV reduction reflect the inability of the cardiovascular system to respond and adapt to such a trigger (Zhong et al., 2005; Sauvet et al., 2010). When there is failure in the interaction and coordination between the parasympathetic activity (PA) withdrawal and decreased total HRV (i.e., decreased high frequency, increased low frequency, increased low frequency/high frequency ratio) then sleep deprivation may pose as a risk factor for CVD (Zhong et al., 2005; Tobaldini et al., 2014; Johnston et al., 2020).
It is possible that APSD prior to exercise to mask the intended health benefits of HIIE (Gibala, 2007; Wisloff et al., 2007; Harris et al., 2008; Tjonna et al., 2008; Di Francescomarino et al., 2009; Fisher-Wellman and Bloomer, 2009; Tyldum et al., 2009; Gabriel et al., 2012; Gibala et al., 2012; Heydari et al., 2013; Weston et al., 2014; Bond et al., 2015; Ramírez-Vélez et al., 2016; Sawyer et al., 2016; Ramírez-Vélez et al., 2018; Tucker et al., 2018; Ramirez-Velez et al., 2020), especially for those who are regular good sleepers particularly as it is related to postprandial HRV and cardiometabolic health (Christensen et al., 1999; Gielen and Hambrecht, 2005; Mestek et al., 2006; Mestek et al., 2008; Gielen et al., 2010; Gielen et al., 2011; Chung et al., 2020). Research questions related to the immediate HRV responses after acute HIIE following reference sleep and APSD are still under investigation. Previous work from our lab has examined such research questions (Papadakis et al., 2020; Papadakis et al., 2021a; Papadakis et al., 2021b) under the traditional Exercise Physiology approach, which focuses on a single physiological system by reducing complex multicomponent systems on their respective parts (Machamer et al., 2000; Bechtel and Richardson, 2010; Balague et al., 2020). We reported that the expected HRV disturbance as a response to an acute HIIE was not influenced by APSD (Papadakis et al., 2021a), HIIE after APSD was still cardioprotective for the postprandial endothelial function (Papadakis et al., 2020), and lastly that fasted HIIE and performance were not affected by sleep conditions (Papadakis et al., 2021b).
The human organism though, is comprised by multicomponent physiological systems that operate through non-linear feedback mechanisms at several spatio-temporal scales generating complex dynamics that continuously adapt to various intrinsic and extrinsic stimuli (Bashan et al., 2012; Ivanov and Bartsch, 2014; Bartsch et al., 2015). Accordingly, exercising in the morning hours after APSD may impact negatively interactions among physiological systems (e.g., cardiovascular and ANS) and, therefore, generate a differentiated network of physiological interactions compared to exercising after a full night sleep. It is possible, that many people after exercise, may consume a typical American high-fat convenience breakfast (Bourland and Vogt, 2009; Lang et al., 2014), a behavior that adds another level in the complex interactions between the aforementioned physiological systems. Such behavior, may disrupt the network interactions among the cardiovascular, autonomic regulation, and postprandial lipemic systems ultimately jeopardizing their health.
Therefore, this paper attempts to revisit our previous work (Papadakis et al., 2020; Papadakis et al., 2021a; Papadakis et al., 2021b), through the prism of Network Physiology of Exercise (NPE) (Balague et al., 2020), a new branch of the interdisciplinary field of Network Physiology (Ivanov et al., 2016; Ivanov, 2021; Ivanov et al., 2021). NPE addresses the fundamental question of how physiological systems coordinate and synchronize their dynamics as a network to optimize organism function, and how these network interactions change in response to exercise and training. NPE utilizes novel methods and approaches in Network Theory, Nonlinear Dynamics, Computational and Statistical Physics, and Biomedical Informatics to represent localized integrated organ systems and their interactions across various scales with the respective nodes (examined variables) and edges/links (respective interactions) in their dynamic network (Bartsch et al., 2014; Ivanov and Bartsch, 2014; Bartsch et al., 2015; Ivanov et al., 2016; Ivanov et al., 2017; Balague et al., 2020; Lehnertz et al., 2020; Meyer, 2020; Rizzo et al., 2020). Accordingly, we aimed to examine the postprandial network interactions between autonomic regulation through HRV and lipemia through low-density lipoprotein (LDL) cholesterol in response to APSD and HIIE.
## 2.1 Study Design and Participants
As stated, this paper is revisiting data collected as part of a bigger project that involved parameters related to sleep, exercise, and cardiovascular function and outcomes of these investigations presented in detail elsewhere (Papadakis et al., 2020; Papadakis et al., 2021a; Papadakis et al., 2021b). Briefly, a within-subject randomized crossover experimental design with three 3) experimental conditions (i.e., a reference sleep—no exercise “control condition” (RS) in which a standardized test meal was ingested in the morning after 9–9.5 h of time-in-bed in which at least 8 h of sleep was attained; a “reference sleep and high-intensity interval exercise condition” (RSX), similar to RS condition in terms of the meal and the obtained sleep time with the exception of a high-intensity interval exercise with 3:2 intervals at 90 and $40\%$ of VO2 reserve that averaged $70\%$ of VO2 reserve and expended 500 kcals of energy, and; a “short and disrupted sleep and high-intensity interval exercise—acute partial sleep deprivation condition” (SSX), similar to RSX in terms of the meal and the performed exercise with the exception of the sleep time that was regulated to 3–3.5 h of time-in-bed limited to no more than 3.5 h of sleep) was employed to answer the research questions as depicted earlier (Figure 1). All experimental conditions began after 48 h of controlling activities of daily living, medication use, standardized diet to what the individuals consumed during the first experimental condition, and supplementation of any kind with a minimum 72-h and maximum 2 weeks washout period between each condition. Thirty healthy males (25–55 years) with normal and overweight body mass index (BMI) met the following inclusion criteria: 1) being recreationally physically active, but not engaging in training for long-distance endurance events, 2) non-smokers, 3) not taking any medications known to alter blood pressure, lipidemic and glucose profile, and 4) not taking any medications known to alter sleep. Participants had to be “good” sleepers as indicated by a score of ≤5 on the Pittsburgh Sleep Quality Index (PSQI) (Buysse et al., 1989). Study was approved by the Institutional Review Board and performed in agreement to the Declaration of Helsinki with all participants having read and signed an informed consent form prior to participation.
**FIGURE 1:** *Original Experimental conditions.*
Experimental conditions completed on two consecutive days and began after 48 h of physical inactivity, no medication use, and the consumption of a diet standardized to what the individual consumed during the first intervention and free from supplementation of any kind. Conditions involved one pre-sleep standard meal consumed in the evening of the first experimental day, six hear rate variability (HRV) recordings, and a morning standard meal surrounding the sleep and exercise interventions. Conditions included: 1) a reference sleep—no exercise “control condition” (RS) in which a standardized test meal was ingested in the morning after 9–9.5 h of time-in-bed in which at least 8 h of sleep was attained) a “reference sleep and high-intensity interval exercise condition” (RSX) in which the test meal was ingested in the morning after reference sleep and after a session of high-intensity interval exercise (3:2 intervals at 90 and $40\%$ of VO2 reserve that average $70\%$ of VO2 reserve) to expend 500 kcals of energy, and; 3) a “short and disrupted sleep and high-intensity interval exercise condition—acute partial sleep deprivation” (SSX) in which an experimental test meal was ingested in the morning after 3–3.5 h of time-in-bed limited to no more than 3.5 h of sleep and after a session of high-intensity interval exercise to expend 500 kcals of energy. Participants arrived at 7p.m. at the laboratory and stayed around 8:30p.m. During this time blood was collect (indicated by the syringe) and an HRV measurement was taken before the evening meal. We discharged participants from the lab accounting for commuting time and bed-preparation time so at 9p.m. all to be in bed. Participants stayed in their homes until the awake time that it was set at 6:00a.m. During this time only data from Sensewear were collected to verify the sleep duration. Participants had to be at the lab at 7:00a.m. the next day. Between 7 and 9a.m., another blood draw was performed and a HRV measurement was taken around 7:00a.m., followed by the exercise condition and another blood draw immediate post-exercise. Around 8:30–9:00a.m. the morning meal was provided followed by another blood draw and a HRV measurement. Every 2 hours post-exercise a blood draw was performed and an HRV measurement was taken until round 3:00p.m.
## 2.2.1 Body Composition and Cardiovascular Fitness
Preliminary measurement of participants’ body composition via dual-energy X-ray absorptiometry (DXA) (Discovery DXA™, Hologic®, Bedford, MA) was performed. After that, an individualized maximal graded exercise test using a modified ramped treadmill protocol to determine participants’ cardiovascular fitness (VO2) via collection of respiratory gases (TrueOne 2400™, ParvoMedics®, Sandy, UT) was executed. Results of this test were used to calculate the experimental exercise intensities as illustrated in Figure 1 (i.e., 3:2 intervals at 90 and $40\%$ of VO2 reserve that average $70\%$ of VO2 reserve; VO2 reserve was calculated as (VO2 max—VO2 rest) X % intensity + VO2 rest; with VO2 rest to be 3.5 ml/kg/min) (American College of Sports Medicine, 2013), and also to familiarize participants with exercise intervals to ascertain their comfort during the experimental conditions.
## 2.2.2 Sleep and Physical Activity Monitor—Diet Records
On top of using the PSQI scale to identify the “regular good sleepers”, participants’ sleep was monitored by the Sense Wear armband (Sense Wear™, Body Media®, Pittsburgh, PA), which is a validated method to assess both sleep and physical activity parameters (Almeida et al., 2011; Van Wouwe et al., 2011; Sharif and Bahammam, 2013; Soric et al., 2013; Shin et al., 2015). Participants had to wear the monitor on their non-dominant arm for 23 h/day each day, for 1 week prior to experimental conditions and for 2 days leading up to experimental conditions. Information from the monitor was used to characterize participants’ sleep duration, sleep consistency, sedentary time, levels of physical activity and to ensure the homogeneity of study’s sample in terms of sleep and physical activity patterns. Moreover, participants’ diet was controlled as they were asked 2 days prior and during the experimental conditions to maintain their typical dietary habits and consume food that were easily reproducible. Dietary intake and macronutrient composition was analyzed using the ChooseMyPlate® (U.S. Department of Agriculture, Washington, DC). Ensuring a stable diet, sleep, and physical activity patterns-habits was paramount to reduce their respective influence on changes in the dependent variables.
## 2.3.1 Standard Evening Meal
The standardized consumed evening meal of day 1 (∼805 kcal) was turkey and cheese sandwich on whole grain bread, a medium banana, a 150 g cup of Greek yogurt, and a 24 oz Gatorade® drink. It was the last meal that all participants had from 7p.m. until 9p.m., before they went to bed and until the following morning. Participants remained at the lab until they returned to their residence to go to sleep.
## 2.3.2 Sleep
Participants completed all the sleep elements of the study at their residence in order to eliminate any disturbances that may occurred if they slept in an unfamiliar laboratory place. Per research design, the RS and RSX conditions allowed for 9.5 h of time-in-bed in the hopes that at least ≥8 h of sleep would have achieved. In the SSX, the research design allowed for 3.5 h of time-in-bed limiting the sleep to ≤3 h. Researchers instructed participants to return to their residence once they left the lab, as all the experimental conditions were calculated to allow enough time for commute and sleep preparation/hygiene routine. No other food was allowed, no watching television neither engaging in computer activities were allowed as participants were preparing for sleep. Participants had also to record both the time that they entered the bed and the time they woke up.
## 2.3.3 High-Intensity Interval Exercise
High-intensity interval exercise sessions were performed on Trackmaster® TMX 428CP treadmill. Following the research design, sessions began at least 10–12 h after the evening meal and completed 1 h before the test meal. After a 5-min warmup at 2 mph and $0\%$ grade, the HIIE sessions were completed in 3-min running intervals at $90\%$ of VO2 reserve separated by 2-min intervals of jogging/walking at $40\%$ of VO2 reserve until 500 kcal were expended. The average intensity of all HIIE sessions was equated to $70\%$ of VO2 reserve. The applied HIIE protocol of 3:2 min work to rest ratio was a modified one from previous studies (Kaikkonen et al., 2008; Matsuo et al., 2014; Osuka et al., 2017; Ito, 2019).
## 2.3.4 Standard Morning Meal
A standard commercially available meal was provided to participants 60 min after completing the HIIE sessions of day 2. The meal included a Jimmy Dean® sausage, egg, and cheese biscuit (∼410 kcals; 29 g fat; 26 g carbohydrate; 11.5 g protein); a Jimmy Dean® fully-cooked pork sausage patty (∼270 kcals; 24 g fat; 2 g carbohydrate; 10.5 g protein); a Little Debbie® honey bun (∼483 kcals; 27 g fat; 55 g carbohydrate; 5.5 g protein), and; a cup of whole milk (∼146 kcals; 8 g fat; 11 g carbohydrate; 7.5 g protein). The test meal had a total of 1,309 kcals (88 g fat; 94 g carbohydrate; 35 g protein). For the RS condition of day 2, the meal was provided 60 min after their arrival at the lab and matched their respective HIIE sessions.
## 2.3.5 Cardiac Autonomic Regulation—Heart Rate Variability
A standardized HRV protocol and methodology for circadian influence on cardiac autonomic assessment was followed as previously described (Malik, 1996; Sammito and Böckelmann, 2016; Riganello et al., 2019; Johnston et al., 2020). Participants after being in supine position for 10 min in a quiet and temperature-controlled environment (∼21–24°C and 40–$60\%$ relative humidity), heart rate (R-R intervals) using a Polar belt (FT1™, Polar Wearlink® Lake Success, NY) was recorded for 5 min at a sampling rate of 1,000 Hz. We used this heart rate recording to calculate the R-R interval and obtain the related heart rate variability indices as previously described (Achten and Jeukendrup, 2003; Engström et al., 2012; Buchheit, 2014). Cardiac autonomic modulation through HRV was assessed the night before (D1), the morning of the next day (D2), 0, 2, 4, and 6-h post-exercise (PE) (Figure 1). We followed previously published guidelines to reduce anxiety of measurement and control for recording errors (Malik, 1996; Sammito and Böckelmann, 2016; Johnston et al., 2020). CardioMood® smartphone application for iPhone was used to process the recorded data as described (Flatt and Esco, 2015; Markov et al., 2016; Baevsky and Chernikova, 2017; Perrotta et al., 2017). A total of five HRV parameters and time intervals were selected based on the literature (Malik, 1996; Trimmel et al., 2015; Abreu et al., 2019; Johnston et al., 2020). Specifically, 1) the standard deviation of RR interval (SDNN; i.e., marker of the sympathovagal balance influenced by the sympathetic activity) and 2) the root mean square of successive normal RR interval differences (RMSSD; i.e., marker of parasympathetic activity) were examined from the time domain of HRV indices. In addition, 3) the high frequency power (0.15–0.40 Hz) (HF; i.e., marker of parasympathetic activity), 4) total power (0–0.4 Hz) (TP; i.e., marker of the sympathovagal balance influenced by the sympathetic activity), and 5) low frequency power (0.04–0.15 Hz) (LF; i.e., that reflects sympathovagal balance, baroreceptor reflex activity or neither) (Akselrod et al., 1981; McCraty et al., 2001; Rahman et al., 2011; Martelli et al., 2014) were examined from the frequency domain of the HRV indices.
## 2.3.6 Blood Collection and Biochemical Analysis
Blood samples were drawn using universal procedures (WHO, 2010) as follows: on the evening before rest/sleep (D1); on the following morning just prior to exercise (D2); following exercise immediate post-exercise (IPE) and immediately prior to eating a standard test meal (0h-), and; again at two (2h-), four (4h-), and 6 h post-exercise (6h-PE) after eating the test meal (Figure 1). All assays were performed in duplicate on first thaw of the samples after being stored at -80°C. All blood samples were analyzed for lipemia related variables (e.g., triglycerides, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, total cholesterol) using commercial ELISA kits (Wako Pure Diagnostics® Richmond, VA). Standard curves for all assays were developed to determine the concentrations in the study samples. All blood variables were corrected for plasma volume shifts known to occur with exercise. All assays for each subject were run on the same day with the same reagent batch to minimize intra- and inter-variability and keep high the internal quality control of our laboratory analysis.
## 2.4 Data Analyses
As noted earlier, the dataset for this manuscript was based on previous work and all the related statistical analyses are described in detail elsewhere (Papadakis et al., 2020; Papadakis et al., 2021a; Papadakis et al., 2021b). We used LDL cholesterol, a marker of cardiovascular risk, due its important clinical significance in cardiovascular disease (Cromwell et al., 2007; Trejo-Gutierrez and Fletcher, 2007), exercise (Durstine et al., 2002), and HRV interactions (Kupari et al., 1993; Christensen et al., 1999; Thayer et al., 2010; Chung et al., 2020; Vijayabaskaran et al., 2022). Since we wanted to investigate the postprandial network interactions between autonomic regulation via HRV and lipemia via LDL under APSD and after HIIE, we constructed one correlation analysis and one physiological network only for the experimental conditions of RSX and SSX. The RSX network was defined as the healthy network. We used two groups of parameters: 1) the five HRV parameters (SDNN, RMSSD, HF, TP, and LF) assessed in six different occasions (D1, D2, 0, 2, 4 and 6-h post-exercise), and 2) LDL obtained in seven different moments (D1, D2, IPE, 0, 2, 4, and 6-post-exercise) (total of 37 parameters).
To obtain the correlation analysis (Figure 2A), the Pearson correlation coefficient was used to calculate the correlations between all possible pairs of the aforementioned HRV and LDL parameters, including inter-HRV/LDL (between HRV and LDL), intra-HRV (within HRV) and intra-LDL (within LD) parameters. To visualize the information provided by the correlation analysis, we next mapped the previously obtained correlation analysis into one physiological network (Figure 2B). This graphical approach is essential to identify patterns in the postprandial network structure and to track the differences in network characteristics for the different experimental conditions. The physiological network was constructed utilizing only the statistically significant correlations in the correlation analysis. The physiological network was comprised by two sub-networks: the HRV and the LDL sub-networks, where color nodes (30 for HRV and seven for LDL) represent the different HRV and LPL parameters, and the network links correspond to the correlation analysis elements reflecting the coupling strength between a given pair of parameters. Links strength is marked by line color and width and are divided into six types: strong positive links (Pearson coefficients >0.8), intermediate positive links (0.6 < Pearson coefficients <0.8), weak positive links (0.4 < Pearson coefficients <0.6); weak negative links (−0.4 > Pearson coefficients > −0.6); intermediate negative links (−0.6 > Pearson coefficients > −0.8), and strong negative links (Pearson coefficients < −0.8). With the aim of quantifying the interactions within the physiological network, we computed the number of links (i.e., number of significant correlations; Figure 3). Specifically, we calculated 1) the total number of links in the entire physiological network, 2) the number of links between the HRV and LDL sub-networks (inter-HRV/LDL); 3) the number of links within the HRV sub-network (intra-HRV); and 4) the number of links within the LDL sub-network (intra-LDL). Correlation matrices and physiological networks were processed and obtained by means of Matlab R2016b (Mathworks, Natik, MA, United States). The visualization framework used in our results is based on previous studies analyzing network interactions among physiological systems during different physiological states (Bashan et al., 2012; Bartsch et al., 2015; Lin et al., 2020; Prats-Puig et al., 2020).
**FIGURE 2:** *Correlation Matrices and Physiological Networks representing postprandial network interactions for Reference Sleep + Exercise (RSX) and Short Sleep + Exercise (SSX). (A) Matrix elements in the correlation matrix represent pairwise coupling strength between each possible pair of HRV and LDL parameters (Pearson correlation coefficient; see Methods). Non-significant correlations are represented in green. Color code is shown in vertical color bars. (B) Nodes in the physiological network represent the different HRV and LDL parameters, and the network links correspond to the correlation matrix elements, reflecting the coupling strength between HRV and LDL parameters. Links strength is marked by line color and width and are divided into six types: strong positive links (Pearson coefficients >0.8), intermediate positive links (0.6 < Pearson coefficients < 0.8), weak positive links (0.4 < Pearson coefficients < 0.6); weak negative links (−0.4 > Pearson coefficients > −0.6); intermediate negative links (−0.6 > Pearson coefficients > −0.8), and strong negative links (Pearson coefficients < −0.8). HRV, heart rate variability; LDL, Low-density lipoprotein; HF-1, High-frequency power at time point -1; SDNN-1, standard deviation of RR interval at time point-1; RMSSD-1, the root mean square of successive normal RR interval differences at time point-1; TP-1, total power at time point-1; LF-1, low-frequency at time point-1.* **FIGURE 3:** *Bar Charts Panel representing the number of links (i.e., significant correlations) within each physiological network for Reference Sleep + Exercise (RSX) and Short Sleep + Exercise (SSX). The heigh of the bars in (A) and (B) indicate 1) the total number of links in the entire physiological network, 2) the number of links between the HRV and LDL sub-networks (inter-HRV/LDL); (iii) the number of links within the HRV sub-network (intra-HRV); and (iv) the number of links within the LDL sub-network (intra-LDL). Panel (C) shows the details for the number of inter-HRV/LDL negative links. HRV, heart rate variability; LDL, Low-density lipoprotein; HF, High-frequency power; SDNN, standard deviation of RR interval; RMSSD, the root mean square of successive normal RR interval differences; TP, total power; LF: low-frequency.*
## 3 Results
From the 30 individuals who signed consent participation forms, only 15 participants were able to adhere to study’s requirements and/or completed the study. Baseline screening of anthropometric and physiological characteristics are presented on Table 1. No differences at $$p \leq 0.05$$ were observed between conditions for the pre-experimental collected data of sleep, diet, and physical activity.
**TABLE 1**
| Variable | Mean ± SE | Min | Max |
| --- | --- | --- | --- |
| Age (yrs.) | 31 ± 5 | 24.0 | 40.0 |
| Height (cm) | 179.3 ± 6.6 | 167.6 | 187.9 |
| Weight (kg) | 83.3 ± 10.9 | 70.7 | 105.7 |
| BMI (kg/m2) | 25.8 ± 2.7 | 21.1 | 29.9 |
| %BF | 21.0 ± 6.2 | 11.4 | 35.3 |
| Max VO2 (L/min) | 4.0 ± 0.7 | 3.2 | 5.6 |
| Max VO2 (ml/kg/min) | 49.1 ± 8.2 | 35.5 | 65.6 |
| Resting HR (bpm) | 55 ± 7 | 42.0 | 63.0 |
| Resting MAP (mmHg) | 85 ± 10 | 70.0 | 100.0 |
| PSQI | 4 ± 0.9 | 2.0 | 5.0 |
Experimental data for sleep and physical activity are presented in Table 2.
**TABLE 2**
| Variable | RSX | SSX |
| --- | --- | --- |
| SL | 8:11:04 ± 1:01:48 | 3:18:09 ± 0:52:02* |
| SD | 6:57:09 ± 0:47:38 | 2:39:30 ± 0:40:11* |
| SLE | 86 ± 8 | 81 ± 12 |
| EE | 2.8 ± 2.2 | 2.5 ± 2.0 |
| Sedentary | 0.8 ± 0.1 | 0.7 ± 0.1 |
| Light | 0.2 ± 0.1 | 0.2 ± 0.1 |
| Moderate | 0.1 ± 0.0 | 0.1 ± 0.0 |
Experimental data for exercise are presented in Table 3. Percentage of coefficients of variation (CV%) for LDL cholesterol were calculated for RSX and SSX for all seven timepoints. The CV% for RSX was $9.8\%$ and for the SSX was $10.5\%$, respectively.
**TABLE 3**
| Trials | Baseline | RSX | SSX |
| --- | --- | --- | --- |
| Weight (kg) | 83.3 ± 10.9 | 83 ± 11.5 | 83 ± 11.5 |
| Max Exercise VO2 (L/min) | - | 4 ± 0.6 | 4 ± 0.6 |
| Max Exercise VO2 (ml/kg/min) | - | 46 ± 8.1 | 45 ± 8.1 |
| Avg Exercise VO2 (L/min) | - | 3 ± 0.4 | 3 ± 0.5 |
| Avg Exercise VO2 (ml/kg/min) | - | 33 ± 5.1 | 33 ± 5.8 |
| Resting HR (bpm) | 55 ± 7.0 | 58 ± 13.0 | 58 ± 10.0 |
| Avg Exercise HR (bpm) | - | 153 ± 12.0 | 150 ± 12.0 |
| 90% VO2 reserve (ml/kg/min) | - | 42 ± 7.3 | 41 ± 7.3 |
| 40% VO2 reserve (ml/kg/min) | - | 20 ± 3.2 | 20 ± 3.2 |
| Resting MAP (mmHg) | 85 ± 10.0 | 85 ± 8.0 | 83 ± 7.0 |
| Exercise Duration (min) | - | 24 ± 2.6 | 24 ± 2.7 |
Figure 2 shows the correlation matrices and physiological networks representing postprandial network interactions for RSX and SSX conditions. Matrix elements in the correlation analysis represent pairwise coupling strength between each possible pair of HRV and LDL parameters. Nodes in the physiological network represent the different HRV and LDL parameters, and the network links correspond to the correlation analysis elements, reflecting the coupling strength between HRV and LDL parameters. We observed a clearly differentiated network of postprandial interactions between RSX and SSX due to the irruption of negative links in the SSX network.
As depicted in Figure 3, the total number of links in the physiological network increased by $86.9\%$ under SSX compared to RSX (356 vs 191 links) due to 1) the manifestation of weak and intermediate negative inter-HRV/LDL links, and 2) an increase of positive intra-HRV links.
No inter-HRV/LDL links were observed between the HRV and LDL sub-networks in RSX. However, a remarkable incursion of weak and intermediate negative links was observed between the HRV and LDL sub-networks in SSX. These negative association were present for all HRV parameters, with higher number of links between LDL and RMSSD, SDNN and HF (see Figure 2C).
Regarding the intra-HRV and intra-LDL links, the HRV sub-network was characterized by an increased number of intermediate positive links in SSX ($136\%$) compared to RSX. No remarkable differences were observed between RSX and SSX for intra-LDL links in the LDL sub-network.
## 4 Discussion
This study investigated postprandial network interactions between autonomic regulation through HRV, and lipemia through low-density lipoprotein (LDL) cholesterol in response to APSD and after HIIE. We reported an irruption of inter-HRV/LDL negative links in the physiological network of SSX compared to RSX, which a priori we defined as the healthy network. The presence of weak and intermediate negative links between the HRV and LDL sub-networks in SSX reflected the impact of sleep deprivation on the autonomic regulation and lipemia. Further, increased connectivity was noted within the HRV sub-network in SSX, with no differences documented for the LDL sub-network. These findings revealed the inability of HIIE to remain cardioprotective under APSD state, and underlie the need to further investigate the effects of APSD and HIIE on the interactions among physiological systems.
The presence of negative links between HRV and LDL for SSX is supported by evidence indicating that impaired balance of the ANS may be the mechanistic explanation linking APSD to CVD (Zhong et al., 2005; Tobaldini et al., 2014). Literature indicates that APSD increases the sympathoadrenal influence, has a greater impact on metabolic and cardiovascular functions that result to increased catecholamine levels, dampened glucose metabolism, increased heart rate (HR) and blood pressure (BP) (Gottlieb et al., 2006; Hall et al., 2008; Meerlo et al., 2008). In addition, it alters the hypothalamic-pituitary-adrenal (HPA) axis activity towards to a higher glucocorticoid release and subsequent systemic inflammatory and oxidative stress response (Meerlo et al., 2008). Therefore, it is apparent that physiologically speaking, APSD has detrimental health effects, which may be observed after very short-term exposure (Alvarez and Ayas, 2004; Gangwisch et al., 2006; Atkinson and Davenne, 2007; Knutson et al., 2007; Buxton and Marcelli, 2010; Centers for Disease and Prevention, 2011; Dettoni et al., 2012; Wu et al., 2012; Calvin et al., 2014). Moreover, acute exercise induces changes in autonomic tone and disturbs the ANS (Thompson et al., 2007; Fukuda et al., 2015), with the HIIE to disturb even more the ANS, and possible to contributing to an unhealthy cardiometabolic status, even after a single episode of HIIE (Besnier et al., 2017).
Considering RSX as the reference healthy and functional network, the HRV sub-network for SSX was characterized by an increment of intermediate positive links, reflecting an increased connectivity due to a reduced sleep time. As previously described (Balague et al., 2020) both underexpressed (weak) and overexpressed network connectivity could reflect unfunctional/pathological states. More specifically, overexpressed/excessive connectivity as observed within the HRV sub-network for SSX, could be associated with a transitory underexpression of coupling network connectivity (i.e., imbalance: some processes are overexpressed and others underexpressed). An example of such imbalance is the rigidity and reduction of diversity potential provoked by exercise-induced fatigue (Vazquez et al., 2016; Vazquez et al., 2020). Similarly, some pathological conditions (e.g., neuro-muscular disorders) could increase the density and/or strength of interactions among certain nodes, pushing the system toward a rigid order which, in turn, could reduce its adaptability to environmental constraints (Ivanov et al., 1998; Ivanov et al., 2001; Stergiou et al., 2006; Stergiou and Decker, 2011). The results of this study support the notion of characterizing a healthy network based on the number of network links.
Several sleep protocols have examined the sleep deprivation effects on healthy individuals in respect to cardiovascular changes and autonomic control via HRV (Tobaldini et al., 2014; Tobaldini et al., 2017). Some reported no differences in HRV with just 4 h sleep (Muenter et al., 2000), while other showed an increase in sympathetic activity (SA) as demonstrated in total decreased HRV, increased low-frequency (LF), and decreased high-frequency (HF) compared to control of 8 h sleep (Dettoni et al., 2012). Our results are in agreement with those that showed an impact of short sleep duration on autonomic regulation (Dettoni et al., 2012) as we reported negative associations for all the HRV parameters and lipemia.
It is important though to mention that studies that examine sleep, HRV, exercise and cardiometabolic health due to differences in the employed research designs, settings, examined variables, sleep durations, sample characteristics etc., yield heterogenous results that are difficult to compared and provide clear and comprehensive outcomes (Malik, 1996; Elsenbruch et al., 1999; Kaikkonen et al., 2008; Al Haddad et al., 2009; Dettoni et al., 2012; Myllymaki et al., 2012; Uchida et al., 2012; Heydari et al., 2013; Tobaldini et al., 2013; Oda and Shirakawa, 2014; Michael et al., 2017; Tobaldini et al., 2017; van Leeuwen et al., 2018; Abreu et al., 2019; Barroso et al., 2019; Costa et al., 2019; Schneider et al., 2019; Vitale et al., 2019). Therefore, research questions related to the immediate HRV responses after acute HIIE following reference sleep and APSD are still under investigation. Previous work from our lab has examined such research questions (Papadakis et al., 2020; Papadakis et al., 2021a; Papadakis et al., 2021b) using the traditional reductionistic approach focusing on a single physiological system and investigated the mechanistic interactions with other single systems by reducing complex multicomponent systems on their respective parts (Machamer et al., 2000; Bechtel and Richardson, 2010; Balague et al., 2020). We reported that HIIE was cardioprotective and APSD did not influence the HRV (Papadakis et al., 2021a), neither the postprandial endothelial function (Papadakis et al., 2020), nor the exercise performance (Papadakis et al., 2021b).
Investigating though the same research questions under the Network Physiology of Exercise perspective we showed that APSD and HIIE had an impact on the HRV and LDL. It seems though that our previous investigations were not able to capture the synchronization and integration among autonomic nervous and cardiometabolic systems. Recent work from Ivanov and Bartsch, [2014] and Ivanov et al., [ 2017] highlighted the fact that physiological states emerge due to specific network organization, topology, and their respective network of dynamic interactions. Moreover, such network of dynamic interactions is moving past the concepts of interconnectivity across of physiological systems and the statistical inference of static associations that govern physiological states (Sieck, 2017; Head, 2020). As such, our previous analyses (Papadakis et al., 2020; Papadakis et al., 2021a; Papadakis et al., 2021b), failed to provide a comprehensive understanding of the dynamic interaction of the involved physiological systems and their subsystems to generate dynamic integrated response at the organism level (Balague et al., 2020). The findings of this study reinforce previous works suggesting that the commonly utilized physiological parameters (e.g., VO2max) provide little information on the nature of the dynamic interactions among physiological systems and their common role in an integrated network. Coordinative variables, such as cardio-respiratory coordination or other psychophysiological parameters, can detect qualitative changes related to the coordinated activity among physiological systems, and their changes under exercise-related constraints (Balagué et al., 2013; Esquius et al., 2019; Garcia-Retortillo et al., 2019).
A major limitation of this study is that it cannot be considered as a true network analysis, as no time series of physiological variables were recorded and analyzed. Note that to capture interactions among physiological systems, time series analysis and the detection of coordinative variables would be the most appropriate strategy. This study was not initially conceived to investigate postprandial network interactions between autonomic regulation and lipemia, but to mimic real life settings between APSD and HIIE under the traditional framework of Exercise Physiology. The applied correlation analysis does not have the power to identify dynamic interactions between the investigated physiological systems. Therefore, this study has inherited all the limitations of the traditional Exercise Physiology framework, that is, the tacit assumption that results obtained by a sample can be generalized to a population level based on the representative observed changes of a “typical” (i.e., average) individual. This assumption though can be true only if the system is ergodic and its evolution in time is stationary and the structure of the interindividual multivariate dynamics is the same across all individuals (Balague et al., 2020). Moreover, since this is reanalysis of data collected for another purpose, it carries the limitations of our previous investigations (e.g., absence of APSD and no HIIE, only apparently healthy men who were good sleepers, environmental stress and factors outside of controlled laboratory settings, indirect method of measuring the cardiac autonomic activity, time of day and chronotype of our sample) (Papadakis et al., 2020; Papadakis et al., 2021a; Papadakis et al., 2021b). At the same time though, this study’s strength is the application of the NPE approach to examine a research question with controversial results when examined through the traditional Exercise Physiology framework. This study is providing preliminary evidence on the sensitivity of the NPE approach to capture interactions among different physiological systems. In this line, further research utilizing time series of physiological variables (Garcia-Retortillo et al., 2020) is needed to investigate the effects of APSD and HIIE on postprandial network interactions.
## 5 Conclusion
The human organism is composed of various integrated networks and sub-networks of interconnected organs, systems, and functions, a disruption or failure of one system can trigger a cascade of failures that can be manifested as a disease state (Ivanov and Bartsch, 2014; Goldman et al., 2015; Liu et al., 2015; Ivanov et al., 2017; Sieck, 2017; Liu and Chen, 2019; Balague et al., 2020; Barajas-Martinez et al., 2020; Corkey and Deeney, 2020). We investigated postprandial network interactions between autonomic regulation through HRV, and lipemia through LDL cholesterol in response to APSD and HIIE. We observed an increase of inter-HRV/LDL negative links in the SSX physiological network compared to RSX. These results reflected the impact of sleep deprivation on the autonomic regulation and lipemia and, revealed the inability of HIIE to remain cardioprotective under APSD.
## 5.1 Resource Identification Initiative
To take part in the Resource Identification Initiative, please use the corresponding catalog number and RRID in your current manuscript. For more information about the project and for steps on how to search for an RRID, please click here.
## Data Availability Statement
The data analyzed in this study is subject to the following licenses/restrictions No restrictions applied to the dataset. Requests to access these datasets should be directed to [email protected].
## Ethics Statement
The studies involving human participants were reviewed and approved by Baylor University. The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
All authors listed have made a substantial, direct, and intellectual contribution to the work and approved it for publication.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Abreu R. M. D., Rehder-Santos P., Simões R. P., Catai A. M.. **Can High-Intensity Interval Training Change Cardiac Autonomic Control? A Systematic Review**. *Braz. J. Phys. Ther.* (2019) **23** 279-289. DOI: 10.1016/j.bjpt.2018.09.010
2. Achten J., Jeukendrup A. E.. **Heart Rate Monitoring: Applications and Limitations**. *Sports Med.* (2003) **33** 517-538. DOI: 10.2165/00007256-200333070-00004
3. Akselrod S., Gordon D., Ubel F. A., Shannon D. C., Berger A. C., Cohen R. J.. **Power Spectrum Analysis of Heart Rate Fluctuation: a Quantitative Probe of Beat-To-Beat Cardiovascular Control**. *Science* (1981) **213** 220-222. DOI: 10.1126/science.6166045
4. Al Haddad H., Laursen P. B., Ahmaidi S., Buchheit M.. **Nocturnal Heart Rate Variability Following Supramaximal Intermittent Exercise**. *Int. J. Sports Physiol. Perform.* (2009) **4** 435-447. DOI: 10.1123/ijspp.4.4.435
5. Almeida G. J. M., Wasko M. C. M., Jeong K., Moore C. G., Piva S. R.. **Physical Activity Measured by the SenseWear Armband in Women with Rheumatoid Arthritis**. *Phys. Ther.* (2011) **91** 1367-1376. DOI: 10.2522/ptj.20100291
6. Alvarez G. G., Ayas N. T.. **The Impact of Daily Sleep Duration on Health: A Review of the Literature**. *Prog. Cardiovasc. Nurs.* (2004) **19** 56-59. DOI: 10.1111/j.0889-7204.2004.02422.x
7. American College of Sports Medicine (2013). ACSM's Guidelines for Exercise Testing and Prescription. Philadelphia, PA: Lippincott Williams & Wilkins.. *ACSM's Guidelines for Exercise Testing and Prescription* (2013)
8. **Sleep Deprivation - Latest Research & Treatments | American Sleep Assoc [Online]**. (2018)
9. Atkinson G., Davenne D.. **Relationships between Sleep, Physical Activity and Human Health**. *Physiol. Behav.* (2007) **90** 229-235. DOI: 10.1016/j.physbeh.2006.09.015
10. Baevsky R. М., Chernikova A. G.. **Heart Rate Variability Analysis: Physiological Foundations and Main Methods**. *Cardiometry* (2017) **10** 66-76. DOI: 10.12710/cardiometry.2017.10.6676
11. Balagué N., Aragonés D., Hristovski R., García-Retortillo S., Tenenbaum G.. **Attention Focus Emerges Spontaneously during Progressive and Maximal Exercise**. *Revista de Psicología Del. Deporte* (2013) **23** 57-63
12. Balagué N., Hristovski R., Almarcha M., Garcia-Retortillo S., Ivanov P. C.. **Network Physiology of Exercise: Vision and Perspectives**. *Front. Physiol.* (2020) **11** 611550. DOI: 10.3389/fphys.2020.611550
13. Barajas-Martínez A., Ibarra-Coronado E., Sierra-Vargas M. P., Cruz-Bautista I., Almeda-Valdes P., Aguilar-Salinas C. A.. **Physiological Network from Anthropometric and Blood Test Biomarkers**. *Front. Physiol.* (2020) **11** 612598. DOI: 10.3389/fphys.2020.612598
14. Barroso R., Silva-Filho A. C., Dias C. J., Soares N., Mostarda A., Azoubel L. A.. **Effect of Exercise Training in Heart Rate Variability, Anxiety, Depression, and Sleep Quality in Kidney Recipients: A Preliminary Study**. *J. Health Psychol.* (2019) **24** 299-308. DOI: 10.1177/1359105316676329
15. Bartsch R. P., Liu K. K., Ma Q. D., Ivanov P. C.. **Three Independent Forms of Cardio-Respiratory Coupling: Transitions across Sleep Stages**. *Comput. Cardiol. (2010)* (2014) **41** 781-784. PMID: 25664348
16. Bartsch R. P., Liu K. K. L., Bashan A., Ivanov P. C.. **Network Physiology: How Organ Systems Dynamically Interact**. *PLoS One* (2015) **10** e0142143. DOI: 10.1371/journal.pone.0142143
17. Bashan A., Bartsch R. P., Kantelhardt J. W., Havlin S., Ivanov P. C.. **Network Physiology Reveals Relations between Network Topology and Physiological Function**. *Nat. Commun.* (2012) **3** 702. DOI: 10.1038/ncomms1705
18. Bechtel W., Richardson R. C.. *Discovering Complexity: Decomposition and Localization as Strategies in Scientific Research* (2010)
19. Besnier F., Labrunée M., Pathak A., Pavy-Le Traon A., Galès C., Sénard J.-M.. **Exercise Training-Induced Modification in Autonomic Nervous System: An Update for Cardiac Patients**. *Ann. Phys. Rehabil. Med.* (2017) **60** 27-35. DOI: 10.1016/j.rehab.2016.07.002
20. Bhati P., Moiz J. A.. **High-Intensity Interval Training and Cardiac Autonomic Modulation**. *Saudi J. Sports Med.* (2017) **17** 129-134. DOI: 10.4103/sjsm.sjsm_2_17
21. Bond B., Gates P. E., Jackman S. R., Corless L. M., Williams C. A., Barker A. R.. **Exercise Intensity and the protection from Postprandial Vascular Dysfunction in Adolescents**. *Am. J. Physiology-Heart Circulatory Physiol.* (2015) **308** H1443-H1450. DOI: 10.1152/ajpheart.00074.2015
22. Bourland C. T., Vogt G. L.. **Breakfast Foods**. *The Astronaut's Cookbook* (2009) 28-36. DOI: 10.1007/978-1-4419-0624-3_2
23. Buchheit M.. **Monitoring Training Status with HR Measures: Do All Roads Lead to Rome?**. *Front. Physiol.* (2014) **5** 73. DOI: 10.3389/fphys.2014.00073
24. Buxton O. M., Marcelli E.. **Short and Long Sleep Are Positively Associated with Obesity, Diabetes, Hypertension, and Cardiovascular Disease Among Adults in the United States**. *Soc. Sci. Med.* (2010) **71** 1027-1036. DOI: 10.1016/j.socscimed.2010.05.041
25. Buysse D. J., Reynolds C. F., Monk T. H., Berman S. R., Kupfer D. J.. **The Pittsburgh Sleep Quality Index: A New Instrument for Psychiatric Practice and Research**. *Psychiatry Res.* (1989) **28** 193-213. DOI: 10.1016/0165-1781(89)90047-4
26. Calvin A. D., Covassin N., Kremers W. K., Adachi T., Macedo P., Albuquerque F. N.. **Experimental Sleep Restriction Causes Endothelial Dysfunction in Healthy Humans**. *J. Am. Heart Assoc.* (2014) **3** e001143. DOI: 10.1161/JAHA.114.001143
27. **Effect of Short Sleep Duration on Daily activities-United States, 2005-2008**. *MMWR Morb Mortal Wkly Rep.* (2011) **60** 239-242. DOI: 10.2307/23319838
28. Christensen J. H., Toft E., Christensen M. S., Schmidt E. B.. **Heart Rate Variability and Plasma Lipids in Men with and without Ischaemic Heart Disease**. *Atherosclerosis* (1999) **145** 181-186. DOI: 10.1016/s0021-9150(99)00052-0
29. Chung H., Oh Y. H., Moon J. H., Kim H. J., Kong M. H.. **Risk of Metabolic and Cardiovascular Risk Factors in Individuals with Autonomic Imbalance Measured by Heart Rate Variability**. *Korean J. Health Promot.* (2020) **20** 41-48. DOI: 10.15384/kjhp.2020.20.2.41
30. Corkey B. E., Deeney J. T.. **The Redox Communication Network as a Regulator of Metabolism**. *Front. Physiol.* (2020) **11** 567796. DOI: 10.3389/fphys.2020.567796
31. Costa J., Figueiredo P., Nakamura F., Rago V., Rebelo A., Brito J.. **Intra-Individual Variability of Sleep and Nocturnal Cardiac Autonomic Activity in Elite Female Soccer Players during an International Tournament**. *PLoS One* (2019) **14** e0218635. DOI: 10.1371/journal.pone.0218635
32. Cromwell W. C., Otvos J. D., Keyes M. J., Pencina M. J., Sullivan L., Vasan R. S.. **LDL Particle Number and Risk of Future Cardiovascular Disease in the Framingham Offspring Study-Implications for LDL Management**. *J. Clin. Lipidol.* (2007) **1** 583-592. DOI: 10.1016/j.jacl.2007.10.001
33. Dettoni J. L., Consolim-Colombo F. M., Drager L. F., Rubira M. C., Cavasin de Souza S. B. P., Irigoyen M. C.. **Cardiovascular Effects of Partial Sleep Deprivation in Healthy Volunteers**. *J. Appl. Physiol.* (2012) **113** 232-236. DOI: 10.1152/japplphysiol.01604.2011
34. Di Francescomarino S., Sciartilli A., Di Valerio V., Di Baldassarre A., Gallina S.. **The Effect of Physical Exercise on Endothelial Function**. *Sports Med.* (2009) **39** 797-812. DOI: 10.2165/11317750-000000000-00000
35. Durstine J. L., Grandjean P. W., Cox C. A., Thompson P. D.. **Lipids, Lipoproteins, and Exercise**. *J. Cardiopulmonary Rehabil.* (2002) **22** 385-398. DOI: 10.1097/00008483-200211000-00002
36. Elsenbruch S., Harnish M. J., Orr W. C.. **Heart Rate Variability during Waking and Sleep in Healthy Males and Females**. *Sleep* (1999) **22** 1067-1071. DOI: 10.1093/sleep/22.8.1067
37. Engström E., Ottosson E., Wohlfart B., Grundström N., Wisén A.. **Comparison of Heart Rate Measured by Polar RS400 and ECG, Validity and Repeatability**. *Adv. Physiother.* (2012) **14** 115-122. DOI: 10.3109/14038196.2012.694118
38. Esquius L., Garcia-Retortillo S., Balagué N., Hristovski R., Javierre C.. **Physiological- and Performance-Related Effects of Acute Olive Oil Supplementation at Moderate Exercise Intensity**. *J. Int. Soc. Sports Nutr.* (2019) **16** 12. DOI: 10.1186/s12970-019-0279-6
39. Fisher-Wellman K., Bloomer R. J.. **Acute Exercise and Oxidative Stress: A 30 Year History**. *Dyn. Med.* (2009) **8** 1. DOI: 10.1186/1476-5918-8-1
40. Flatt A. A., Esco M. R.. **Smartphone-Derived Heart-Rate Variability and Training Load in a Women's Soccer Team**. *Int. J. Sports Physiol. Perform.* (2015) **10** 994-1000. DOI: 10.1123/ijspp.2014-0556
41. Fukuda K., Kanazawa H., Aizawa Y., Ardell J. L., Shivkumar K.. **Cardiac Innervation and Sudden Cardiac Death**. *Circ. Res.* (2015) **116** 2005-2019. DOI: 10.1161/CIRCRESAHA.116.304679
42. Gabriel B., Ratkevicius A., Gray P., Frenneaux M. P., Gray S. R.. **High-Intensity Exercise Attenuates Postprandial Lipaemia and Markers of Oxidative Stress**. *Clin. Sci. (Lond)* (2012) **123** 313-321. DOI: 10.1042/CS20110600
43. Gangwisch J. E., Heymsfield S. B., Boden-Albala B., Buijs R. M., Kreier F., Pickering T. G.. **Short Sleep Duration as a Risk Factor for Hypertension: Analyses of the First National Health and Nutrition Examination Survey**. *Hypertension* (2006) **47** 833-839. DOI: 10.1161/01.HYP.0000217362.34748.e0
44. Garcia-Retortillo S., Javierre C., Hristovski R., Ventura J. L., Balagué N.. **Principal Component Analysis as a Novel Approach for Cardiorespiratory Exercise Testing Evaluation**. *Physiol. Meas.* (2019) **40** 084002. DOI: 10.1088/1361-6579/ab2ca0
45. Garcia-Retortillo S., Rizzo R., Wang J. W. J. L., Sitges C., Ivanov P. C.. **Universal Spectral Profile and Dynamic Evolution of Muscle Activation: A Hallmark of Muscle Type and Physiological State**. *J. Appl. Physiol.* (2020) **129** 419-441. DOI: 10.1152/japplphysiol.00385.2020
46. Gibala M. J.. **High-Intensity Interval Training: A Time-Efficient Strategy for Health Promotion?**. *Curr. Sports Med. Rep.* (2007) **6** 211-213. DOI: 10.1007/s11932-007-0033-8
47. Gibala M. J., Little J. P., Macdonald M. J., Hawley J. A.. **Physiological Adaptations to Low-Volume, High-Intensity Interval Training in Health and Disease**. *J. Physiol.* (2012) **590** 1077-1084. DOI: 10.1113/jphysiol.2011.224725
48. Gielen S., Hambrecht R.. **Treatment Strategies in Endothelial Dysfunction: Physical Exercise Versus Pharmacological Therapy**. *Eur. J. Cardiovasc. Prev. Rehabil.* (2005) **12** 318-320. DOI: 10.1097/01.hjr.0000174826.72022.c4
49. Gielen S., Sandri M., Erbs S., Adams V.. **Exercise-Induced Modulation of Endothelial Nitric Oxide Production**. *Curr. Pharm. Biotechnol.* (2011) **12** 1375-1384. DOI: 10.2174/138920111798281063
50. Gielen S., Schuler G., Adams V.. **Cardiovascular Effects of Exercise Training: Molecular Mechanisms**. *Circulation* (2010) **122** 1221-1238. DOI: 10.1161/CIRCULATIONAHA.110.939959
51. Goldman A. W., Burmeister Y., Cesnulevicius K., Herbert M., Kane M., Lescheid D.. **Bioregulatory Systems Medicine: An Innovative Approach to Integrating the Science of Molecular Networks, Inflammation, and Systems Biology with the Patient's Autoregulatory Capacity?**. *Front. Physiol.* (2015) **6** 225. DOI: 10.3389/fphys.2015.00225
52. Gottlieb D. J., Redline S., Nieto F. J., Baldwin C. M., Newman A. B., Resnick H. E.. **Association of Usual Sleep Duration with Hypertension: The Sleep Heart Health Study**. *Sleep* (2006) **29** 1009-1014. DOI: 10.1093/sleep/29.8.1009
53. Grant C. C., Viljoen M., Janse van Rensburg D. C., Wood P. S.. **Heart Rate Variability Assessment of the Effect of Physical Training on Autonomic Cardiac Control**. *Ann. Noninvasive Electrocardiol.* (2012) **17** 219-229. DOI: 10.1111/j.1542-474X.2012.00511.x
54. Hale L., Troxel W., Buysse D. J.. **Sleep Health: An Opportunity for Public Health to Address Health Equity**. *Annu. Rev. Public Health* (2020) **41** 81-99. DOI: 10.1146/annurev-publhealth-040119-094412
55. Hall M. H., Muldoon M. F., Jennings J. R., Buysse D. J., Flory J. D., Manuck S. B.. **Self-Reported Sleep Duration Is Associated with the Metabolic Syndrome in Midlife Adults**. *Sleep* (2008) **31** 635-643. DOI: 10.1093/sleep/31.5.635
56. Harris R. A., Padilla J., Hanlon K. P., Rink L. D., Wallace J. P.. **The Flow-Mediated Dilation Response to Acute Exercise in Overweight Active and Inactive Men**. *Obesity (Silver Spring)* (2008) **16** 578-584. DOI: 10.1038/oby.2007.87
57. Head G. A.. **Integrative Physiology: Update to the Grand Challenge 2020**. *Front. Physiol.* (2020) **11** 489. DOI: 10.3389/fphys.2020.00489
58. Heydari M., Boutcher Y. N., Boutcher S. H.. **High-Intensity Intermittent Exercise and Cardiovascular and Autonomic Function**. *Clin. Auton. Res.* (2013) **23** 57-65. DOI: 10.1007/s10286-012-0179-1
59. Hyson D., Rutledge J. C., Berglund L.. **Postprandial Lipemia and Cardiovascular Disease**. *Curr. Atheroscler. Rep.* (2003) **5** 437-444. DOI: 10.1007/s11883-003-0033-y
60. Ito S.. **High-Intensity Interval Training for Health Benefits and Care of Cardiac Diseases - the Key to an Efficient Exercise Protocol**. *World J. Cardiol.* (2019) **11** 171-188. DOI: 10.4330/wjc.v11.i7.171
61. Ivanov P. C., Bartsch R. P.. **Network Physiology: Mapping Interactions between Networks of Physiologic Networks**. *Networks of Networks: The Last Frontier of Complexity* (2014) 203-222. DOI: 10.1007/978-3-319-03518-5_10
62. Ivanov P. C., Liu K. K. L., Bartsch R. P.. **Focus on the Emerging New fields of Network Physiology and Network Medicine**. *New J. Phys.* (2016) **18** 100201. DOI: 10.1088/1367-2630/18/10/100201
63. Ivanov P. C., Liu K. K., Lin A., Bartsch R. P.. **Network Physiology: From Neural Plasticity to Organ Network Interactions**. *Emergent Complexity from Nonlinearity, in Physics, Engineering and the Life Sciences* (2017) 145-165. DOI: 10.1007/978-3-319-47810-4_12
64. Ivanov P. C., Nunes Amaral L. A., Goldberger A. L., Stanley H. E.. **Stochastic Feedback and the Regulation of Biological Rhythms**. *Europhys. Lett.* (1998) **43** 363-368. DOI: 10.1209/epl/i1998-00366-3
65. Ivanov P. C., Nunes Amaral L. A., Goldberger A. L., Havlin S., Rosenblum M. G., Stanley H. E.. **From 1/f Noise to Multifractal Cascades in Heartbeat Dynamics**. *Chaos* (2001) **11** 641-652. DOI: 10.1063/1.1395631
66. Ivanov P. C.. **The New Field of Network Physiology: Building the Human Physiolome**. *Front. Netw. Physiol.* (2021) **1** 711778. DOI: 10.3389/fnetp.2021.711778
67. Ivanov P. C., Wang J. W., Zhang X., Chen B.. **The New Frontier of Network Physiology: Emerging Physiologic States in Health and Disease from Integrated Organ Network Interactions**. *2019-20 MATRIX Annals* (2021) 237-254. DOI: 10.1007/978-3-030-62497-2_12
68. Johnston B. W., Barrett-Jolley R., Krige A., Welters I. D.. **Heart Rate Variability: Measurement and Emerging Use in Critical Care Medicine**. *J. Intensive Care Soc.* (2020) **21** 148-157. DOI: 10.1177/1751143719853744
69. Kaikkonen P., Rusko H., Martinmäki K.. **Post-Exercise Heart Rate Variability of Endurance Athletes after Different High-Intensity Exercise Interventions**. *Scand. J. Med. Sci. Sports* (2008) **18** 511-519. DOI: 10.1111/j.1600-0838.2007.00728.x
70. Knutson K. L., Spiegel K., Penev P., Van Cauter E.. **The Metabolic Consequences of Sleep Deprivation**. *Sleep Med. Rev.* (2007) **11** 163-178. DOI: 10.1016/j.smrv.2007.01.002
71. Krittanawong C., Kumar A., Wang Z., Jneid H., Baber U., Mehran R.. **Sleep Duration and Cardiovascular Health in a Representative Community Population (From NHANES, 2005 to 2016)**. *Am. J. Cardiol.* (2020) **127** 149-155. DOI: 10.1016/j.amjcard.2020.04.012
72. Kupari M., Virolainen J., Koskinen P., Tikkanen M. J.. **Short-Term Heart Rate Variability and Factors Modifying the Risk of Coronary Artery Disease in a Population Sample**. *Am. J. Cardiol.* (1993) **72** 897-903. DOI: 10.1016/0002-9149(93)91103-o
73. Lang J. M., Eisen J. A., Zivkovic A. M.. **The Microbes We Eat: Abundance and Taxonomy of Microbes Consumed in a Day's worth of Meals for Three Diet Types**. *PeerJ* (2014) **2** e659. DOI: 10.7717/peerj.659
74. Lehnertz K., Bröhl T., Rings T.. **The Human Organism as an Integrated Interaction Network: Recent Conceptual and Methodological Challenges**. *Front. Physiol.* (2020) **11** 598694. DOI: 10.3389/fphys.2020.598694
75. Liew S. C., Aung T.. **Sleep Deprivation and its Association with Diseases- a Review**. *Sleep Med.* (2021) **77** 192-204. DOI: 10.1016/j.sleep.2020.07.048
76. Lin A., Liu K. K. L., Bartsch R. P., Ivanov P. C.. **Dynamic Network Interactions Among Distinct Brain Rhythms as a Hallmark of Physiologic State and Function**. *Commun. Biol.* (2020) **3** 197. DOI: 10.1038/s42003-020-0878-4
77. Littlefield L. A., Grandjean P. W.. **Exercise Intensity and Postprandial Lipemia (La intensidad del ejercicio y la lipemia postprandial)**. *Retos* (2015) **27** 206-212. DOI: 10.47197/retos.v0i27.34380
78. Liu H., Chen A.. **Roles of Sleep Deprivation in Cardiovascular Dysfunctions**. *Life Sci.* (2019) **219** 231-237. DOI: 10.1016/j.lfs.2019.01.006
79. Liu K. K. L., Bartsch R. P., Ma Q. D. Y., Ivanov P. C.. **Major Component Analysis of Dynamic Networks of Physiologic Organ Interactions**. *J. Phys. Conf. Ser.* (2015) **640** 012013. DOI: 10.1088/1742-6596/640/1/012013
80. Machamer P., Darden L., Craver C. F.. **Thinking about Mechanisms**. *Philos. Sci.* (2000) **67** 1-25. DOI: 10.1086/392759
81. Malik M.. **Heart Rate Variability: Standards of Measurement, Physiological Interpretation, and Clinical Use: Task Force of the European Society of Cardiology and the North American Society for Pacing and Electrophysiology**. *Ann. Noninv Electrocard* (1996) **1** 151-181. DOI: 10.1111/j.1542-474x.1996.tb00275.x
82. Markov A., Solonin I., Bojko E.. **Heart Rate Variability in Workers of Various Professions in Contrasting Seasons of the Year**. *Int. J. Occup. Med. Environ. Health* (2016) **29** 793-800. DOI: 10.13075/ijomeh.1896.00276
83. Martelli D., Silvani A., McAllen R. M., May C. N., Ramchandra R.. **The Low Frequency Power of Heart Rate Variability Is Neither a Measure of Cardiac Sympathetic Tone Nor of Baroreflex Sensitivity**. *Am. J. Physiology-Heart Circulatory Physiol.* (2014) **307** H1005-H1012. DOI: 10.1152/ajpheart.00361.2014
84. Matsuo T., Saotome K., Seino S., Shimojo N., Matsushita A., Iemitsu M.. **Effects of a Low-Volume Aerobic-Type Interval Exercise on VO2max and Cardiac Mass**. *Med. Sci. Sports Exerc.* (2014) **46** 42-50. DOI: 10.1249/mss.0b013e3182a38da8
85. McCraty R., Atkinson M., Tomasino D., Stuppy W. P.. **Analysis of Twenty-Four Hour Heart Rate Variability in Patients with Panic Disorder**. *Biol. Psychol.* (2001) **56** 131-150. DOI: 10.1016/s0301-0511(01)00074-6
86. Meerlo P., Sgoifo A., Suchecki D.. **Restricted and Disrupted Sleep: Effects on Autonomic Function, Neuroendocrine Stress Systems and Stress Responsivity**. *Sleep Med. Rev.* (2008) **12** 197-210. DOI: 10.1016/j.smrv.2007.07.007
87. Mestek M. L., Garner J. C., Plaisance E. P., Taylor J. K., Alhassan S., Grandjean P. W.. **Blood Lipid Responses after Continuous and Accumulated Aerobic Exercise**. *Int. J. Sport Nutr. Exerc. Metab.* (2006) **16** 245-254. DOI: 10.1123/ijsnem.16.3.245
88. Mestek M. L., Plaisance E. P., Ratcliff L. A., Taylor J. K., Wee S.-O., Grandjean P. W.. **Aerobic Exercise and Postprandial Lipemia in Men with the Metabolic Syndrome**. *Med. Sci. Sports Exerc.* (2008) **40** 2105-2111. DOI: 10.1249/MSS.0b013e3181822ebd
89. Meyer R.. **The Non-Mechanistic Option: Defending Dynamical Explanations**. *Br. J. Philos. Sci.* (2020) **71** 959-985. DOI: 10.1093/bjps/axy034
90. Michael S., Graham K. S., Davis G. M.. **Cardiac Autonomic Responses during Exercise and Post-Exercise Recovery Using Heart Rate Variability and Systolic Time Intervals-A Review**. *Front. Physiol.* (2017) **8** 301. DOI: 10.3389/fphys.2017.00301
91. Miller M. A., Cappuccio F. P.. **Biomarkers of Cardiovascular Risk in Sleep-Deprived People**. *J. Hum. Hypertens.* (2013) **27** 583-588. DOI: 10.1038/jhh.2013.27
92. Muenter N. K., Watenpaugh D. E., Wasmund W. L., Wasmund S. L., Maxwell S. A., Smith M. L.. **Effect of Sleep Restriction on Orthostatic Cardiovascular Control in Humans**. *J. Appl. Physiol.* (2000) **88** 966-972. DOI: 10.1152/jappl.2000.88.3.966
93. Myllymaki T., Rusko H., Syvaoja H., Juuti T., Kinnunen M. L., Kyrolainen H.. **Effects of Exercise Intensity and Duration on Nocturnal Heart Rate Variability and Sleep Quality**. *Eur. J. Appl. Physiol.* (2012) **112** 801-809. DOI: 10.1007/s00421-011-2034-9
94. Oda S., Shirakawa K.. **Sleep Onset Is Disrupted Following Pre-Sleep Exercise that Causes Large Physiological Excitement at Bedtime**. *Eur. J. Appl. Physiol.* (2014) **114** 1789-1799. DOI: 10.1007/s00421-014-2873-2
95. Oliver M. D., Baldwin D. R., Datta S.. **The Relationship between Sleep and Autonomic Health**. *J. Am. Coll. Health* (2020) **68** 550-556. DOI: 10.1080/07448481.2019.1583652
96. Osuka Y., Matsubara M., Hamasaki A., Hiramatsu Y., Ohshima H., Tanaka K.. **Development of Low-Volume, High-Intensity, Aerobic-Type Interval Training for Elderly Japanese Men: A Feasibility Study**. *Eur. Rev. Aging Phys. Act* (2017) **14** 14. DOI: 10.1186/s11556-017-0184-4
97. Papadakis Z., Forsse J. S., Peterson M. N.. **Acute Partial Sleep Deprivation and High-Intensity Interval Exercise Effects on Postprandial Endothelial Function**. *Eur. J. Appl. Physiol.* (2020) **120** 2431-2444. DOI: 10.1007/s00421-020-04468-5
98. Papadakis Z., Forsse J. S., Peterson M. N.. **Effects of High-Intensity Interval Exercise and Acute Partial Sleep Deprivation on Cardiac Autonomic Modulation**. *Res. Q. Exerc. Sport* (2021a) **92** 824-842. DOI: 10.1080/02701367.2020.1788206
99. Papadakis Z., Forsse J. S., Stamatis A.. **High-Intensity Interval Exercise Performance and Short-Term Metabolic Responses to Overnight-Fasted Acute-Partial Sleep Deprivation**. *Int. J. Environ. Res. Public Health* (2021b) **18** 3655. DOI: 10.3390/ijerph18073655
100. Perrotta A. S., Jeklin A. T., Hives B. A., Meanwell L. E., Warburton D. E. R.. **Validity of the Elite HRV Smartphone Application for Examining Heart Rate Variability in a Field-Based Setting**. *J. Strength Cond Res.* (2017) **31** 2296-2302. DOI: 10.1519/JSC.0000000000001841
101. Phrommintikul A., Chattipakorn S. C., Chattipakorn N.. **Exercise and Cardioprotection: A "HIP" Side of HIPK2 in the Heart**. *EBioMedicine* (2022) **75** 103766. DOI: 10.1016/j.ebiom.2021.103766
102. Powers S. K., Lennon S. L., Quindry J., Mehta J. L.. **Exercise and Cardioprotection**. *Curr. Opin. Cardiol.* (2002) **17** 495-502. DOI: 10.1097/00001573-200209000-00009
103. Prats-Puig A., García-Retortillo S., Puig-Parnau M., Vasileva F., Font-Lladó R., Xargay-Torrent S.. **DNA Methylation Reorganization of Skeletal Muscle-Specific Genes in Response to Gestational Obesity**. *Front. Physiol.* (2020) **11** 938. DOI: 10.3389/fphys.2020.00938
104. Rahman F., Pechnik S., Gross D., Sewell L., Goldstein D. S.. **Low Frequency Power of Heart Rate Variability Reflects Baroreflex Function, Not Cardiac Sympathetic Innervation**. *Clin. Auton. Res.* (2011) **21** 133-141. DOI: 10.1007/s10286-010-0098-y
105. Ramírez-Vélez R., Hernandez A., Castro K., Tordecilla-Sanders A., González-Ruíz K., Correa-Bautista J. E.. **High Intensity Interval-Vs Resistance or Combined-Training for Improving Cardiometabolic Health in Overweight Adults (Cardiometabolic HIIT-RT Study): Study Protocol for a Randomised Controlled Trial**. *Trials* (2016) **17** 298. DOI: 10.1186/s13063-016-1422-1
106. Ramírez-Vélez R., Correa Rodriguez M., Tordecilla-Sanders A., Izquierdo M., Correa-Bautista J., Álvarez C.. **Exercise and Postprandial Lipaemia: Effects on Vascular Health in Inactive Adults**. *Lipids Health Dis.* (2018) **17** 69. DOI: 10.1186/s12944-018-0719-3
107. Ramírez-Vélez R., Tordecilla-Sanders A., Téllez-T L. A., Camelo-Prieto D., Hernández-Quiñonez P. A., Correa-Bautista J. E.. **Effect of Moderate- Versus High-Intensity Interval Exercise Training on Heart Rate Variability Parameters in Inactive Latin-American Adults: A Randomized Clinical Trial**. *J. Strength Cond Res.* (2020) **34** 3403-3415. DOI: 10.1519/JSC.0000000000001833
108. Riganello F., Prada V., Soddu A., di Perri C., Sannita W. G.. **Circadian Rhythms and Measures of CNS/Autonomic Interaction**. *Int. J. Environ. Res. Public Health* (2019) **16** 2336. DOI: 10.3390/ijerph16132336
109. Rizzo R., Zhang X., Wang J. W. J. L., Lombardi F., Ivanov P. C.. **Network Physiology of Cortico-Muscular Interactions**. *Front. Physiol.* (2020) **11** 558070. DOI: 10.3389/fphys.2020.558070
110. Sammito S., Böckelmann I.. **Factors Influencing Heart Rate Variability**. *Int. Cardiovasc. Forum J.* (2016) **6** 18-22. DOI: 10.17987/icfj.v6i0.242
111. Sauvet F., Leftheriotis G., Gomez-Merino D., Langrume C., Drogou C., Van Beers P.. **Effect of Acute Sleep Deprivation on Vascular Function in Healthy Subjects**. *J. Appl. Physiol.* (2010) **108** 68-75. DOI: 10.1152/japplphysiol.00851.2009
112. Sawyer B. J., Tucker W. J., Bhammar D. M., Ryder J. R., Sweazea K. L., Gaesser G. A.. **Effects of High-Intensity Interval Training and Moderate-Intensity Continuous Training on Endothelial Function and Cardiometabolic Risk Markers in Obese Adults**. *J. Appl. Physiol.* (2016) **121** 279-288. DOI: 10.1152/japplphysiol.00024.2016
113. Schneider C., Wiewelhove T., Raeder C., Flatt A. A., Hoos O., Hottenrott L.. **Heart Rate Variability Monitoring during Strength and High-Intensity Interval Training Overload Microcycles**. *Front. Physiol.* (2019) **10** 582. DOI: 10.3389/fphys.2019.00582
114. Seravalle G., Mancia G., Grassi G.. **Sympathetic Nervous System, Sleep, and Hypertension**. *Curr. Hypertens. Rep.* (2018) **20** 74. DOI: 10.1007/s11906-018-0874-y
115. Sharif M. M., Bahammam A. S.. **Sleep Estimation Using BodyMedia's SenseWear Armband in Patients with Obstructive Sleep Apnea**. *Ann. Thorac. Med.* (2013) **8** 53-57. DOI: 10.4103/1817-1737.105720
116. Shin M., Swan P., Chow C. M.. **The Validity of Actiwatch2 and SenseWear Armband Compared against Polysomnography at Different Ambient Temperature Conditions**. *Sleep Sci.* (2015) **8** 9-15. DOI: 10.1016/j.slsci.2015.02.003
117. Sieck G. C.. **Physiology in Perspective: The Importance of Integrative Physiology**. *Physiology* (2017) **32** 180-181. DOI: 10.1152/physiol.00009.2017
118. Soric M., Turkalj M., Kucic D., Marusic I., Plavec D., Misigoj-Durakovic M.. **Validation of a Multi-Sensor Activity Monitor for Assessing Sleep in Children and Adolescents**. *Sleep Med.* (2013) **14** 201-205. DOI: 10.1016/j.sleep.2012.11.003
119. Stergiou N., Decker L. M.. **Human Movement Variability, Nonlinear Dynamics, and Pathology: Is There a Connection?**. *Hum. Move. Sci.* (2011) **30** 869-888. DOI: 10.1016/j.humov.2011.06.002
120. Stergiou N., Harbourne R. T., Cavanaugh J. T.. **Optimal Movement Variability: A New Theoretical Perspective for Neurologic Physical Therapy**. *J. Neurol. Phys. Ther.* (2006) **30** 120-129. DOI: 10.1097/01.npt.0000281949.48193.d9
121. Subramanian S. K., Sharma V. K., Arunachalam V., Rajendran R., Gaur A.. **Comparison of Baroreflex Sensitivity and Cardiac Autonomic Function between Adolescent Athlete and Non-Athlete Boys - A Cross-Sectional Study**. *Front. Physiol.* (2019) **10** 1043. DOI: 10.3389/fphys.2019.01043
122. Thayer J. F., Yamamoto S. S., Brosschot J. F.. **The Relationship of Autonomic Imbalance, Heart Rate Variability and Cardiovascular Disease Risk Factors**. *Int. J. Cardiol.* (2010) **141** 122-131. DOI: 10.1016/j.ijcard.2009.09.543
123. Thompson P. D., Franklin B. A., Balady G. J., Blair S. N., Corrado D., Estes N. A.. **Exercise and Acute Cardiovascular Events Placing the Risks into Perspective: A Scientific Statement from the American Heart Association Council on Nutrition, Physical Activity, and Metabolism and the Council on Clinical Cardiology**. *Circulation* (2007) **115** 2358-2368. DOI: 10.1161/CIRCULATIONAHA.107.181485
124. Tjønna A. E., Lee S. J., Rognmo Ø., Stølen T. O., Bye A., Haram P. M.. **Aerobic Interval Training versus Continuous Moderate Exercise as a Treatment for the Metabolic Syndrome: A Pilot Study**. *Circulation* (2008) **118** 346-354. DOI: 10.1161/CIRCULATIONAHA.108.772822
125. Tobaldini E., Costantino G., Solbiati M., Cogliati C., Kara T., Nobili L.. **Sleep, Sleep Deprivation, Autonomic Nervous System and Cardiovascular Diseases**. *Neurosci. Biobehav Rev.* (2017) **74** 321-329. DOI: 10.1016/j.neubiorev.2016.07.004
126. Tobaldini E., Nobili L., Strada S., Casali K. R., Braghiroli A., Montano N.. **Heart Rate Variability in normal and Pathological Sleep**. *Front. Physiol.* (2013) **4** 294. DOI: 10.3389/fphys.2013.00294
127. Tobaldini E., Pecis M., Montano N.. **Effects of Acute and Chronic Sleep Deprivation on Cardiovascular Regulation**. *Arch. Ital. Biol.* (2014) **152** 103-110. DOI: 10.12871/000298292014235
128. Trejo-Gutierrez J. F., Fletcher G.. **Impact of Exercise on Blood Lipids and Lipoproteins**. *J. Clin. Lipidol.* (2007) **1** 175-181. DOI: 10.1016/j.jacl.2007.05.006
129. Trimmel K., Sacha J., Huikuri H. V.. *Heart Rate Variability: Clinical Applications and Interaction between HRV and Heart Rate* (2015)
130. Tucker W. J., Sawyer B. J., Jarrett C. L., Bhammar D. M., Ryder J. R., Angadi S. S.. **High-Intensity Interval Exercise Attenuates but Does Not Eliminate Endothelial Dysfunction after a Fast Food Meal**. *Am. J. Physiol. Heart Circ. Physiol.* (2018) **314** H188-H194. DOI: 10.1152/ajpheart.00384.2017
131. Tyldum G. A., Schjerve I. E., Tjønna A. E., Kirkeby-Garstad I., Stølen T. O., Richardson R. S.. **Endothelial Dysfunction Induced by Post-Prandial Lipemia: Complete protection Afforded by High-Intensity Aerobic Interval Exercise**. *J. Am. Coll. Cardiol.* (2009) **53** 200-206. DOI: 10.1016/j.jacc.2008.09.033
132. Uchida S., Shioda K., Morita Y., Kubota C., Ganeko M., Takeda N.. **Exercise and Sleep - Review and Future Directions**. *J. Phys. Fitness Sports Med.* (2012) **1** 317-324. DOI: 10.7600/jpfsm.1.317
133. van Leeuwen W. M. A., Sallinen M., Virkkala J., Lindholm H., Hirvonen A., Hublin C.. **Physiological and Autonomic Stress Responses after Prolonged Sleep Restriction and Subsequent Recovery Sleep in Healthy Young Men**. *Sleep Biol. Rhythms* (2018) **16** 45-54. DOI: 10.1007/s41105-017-0122-x
134. Van Wouwe N. C., Valk P. J. L., Veenstra B. J.. **Sleep Monitoring: A Comparison between Three Wearable Instruments**. *Mil. Med.* (2011) **176** 811-816. DOI: 10.7205/milmed-d-10-00389
135. Vázquez P., Hristovski R., Balagué N.. **The Path to Exhaustion: Time-Variability Properties of Coordinative Variables during Continuous Exercise**. *Front. Physiol.* (2016) **7** 37. DOI: 10.3389/fphys.2016.00037
136. Vázquez P., Petelczyc M., Hristovski R., Balagué N.. **Interlimb Coordination: A New Order Parameter and a Marker of Fatigue during Quasi-Isometric Exercise?**. *Front. Physiol.* (2020) **11** 612709. DOI: 10.3389/fphys.2020.612709
137. Vijayabaskaran S., B. D., Damodaran V., Nagarajan S.. **Association between Lipid Levels and Short-Term Heart Rate Variability (HRV) in Type 2 Diabetes**. *Egypt. Acad. J. Biol. Sci. C, Physiol. Mol. Biol.* (2022) **14** 21-30. DOI: 10.21608/eajbsc.2022.213921
138. Vitale J. A., Bonato M., La Torre A. L., Banfi G.. **Heart Rate Variability in Sport Performance: Do Time of Day and Chronotype Play A Role?**. *J. Clin. Med.* (2019) **8** 723. DOI: 10.3390/jcm8050723
139. Wallace J. P., Johnson B., Padilla J., Mather K.. **Postprandial Lipaemia, Oxidative Stress and Endothelial Function: A Review**. *Int. J. Clin. Pract.* (2010) **64** 389-403. DOI: 10.1111/j.1742-1241.2009.02146.x
140. Weston K. S., Wisløff U., Coombes J. S.. **High-intensity Interval Training in Patients with Lifestyle-Induced Cardiometabolic Disease: a Systematic Review and Meta-Analysis**. *Br. J. Sports Med.* (2014) **48** 1227-1234. DOI: 10.1136/bjsports-2013-092576
141. WHO (2010). WHO Guidelines on Drawing Blood: Best Practices in Phlebotomy. Geneva, Switzerland: World Health Organization.. *WHO Guidelines on Drawing Blood: Best Practices in Phlebotomy* (2010)
142. Wisløff U., Støylen A., Loennechen J. P., Bruvold M., Rognmo Ø., Haram P. M.. **Superior Cardiovascular Effect of Aerobic Interval Training Versus Moderate Continuous Training in Heart Failure Patients: A Randomized Study**. *Circulation* (2007) **115** 3086-3094. DOI: 10.1161/CIRCULATIONAHA.106.675041
143. Wright K. P., Drake A. L., Frey D. J., Fleshner M., Desouza C. A., Gronfier C.. **Influence of Sleep Deprivation and Circadian Misalignment on Cortisol, Inflammatory Markers, and Cytokine Balance**. *Brain Behav. Immun.* (2015) **47** 24-34. DOI: 10.1016/j.bbi.2015.01.004
144. Wu M.-C., Yang Y.-C., Wu J.-S., Wang R.-H., Lu F.-H., Chang C.-J.. **Short Sleep Duration Associated with a Higher Prevalence of Metabolic Syndrome in an Apparently Healthy Population**. *Prev. Med.* (2012) **55** 305-309. DOI: 10.1016/j.ypmed.2012.07.013
145. Zhong X., Hilton H. J., Gates G. J., Jelic S., Stern Y., Bartels M. N.. **Increased Sympathetic and Decreased Parasympathetic Cardiovascular Modulation in normal Humans with Acute Sleep Deprivation**. *J. Appl. Physiol.* (2005) **98** 2024-2032. DOI: 10.1152/japplphysiol.00620.2004
|
---
title: Breast cancer subtype and clinical characteristics in women from Peru
authors:
- Valentina A. Zavala
- Sandro Casavilca-Zambrano
- Jeannie Navarro-Vásquez
- Lizeth I. Tamayo
- Carlos A. Castañeda
- Guillermo Valencia
- Zaida Morante
- Mónica Calderón
- Julio E. Abugattas
- Henry L. Gómez
- Hugo A. Fuentes
- Ruddy Liendo-Picoaga
- Jose M. Cotrina
- Silvia P. Neciosup
- Katia Roque
- Jule Vásquez
- Luis Mas
- Marco Gálvez-Nino
- Laura Fejerman
- Tatiana Vidaurre
journal: Frontiers in Oncology
year: 2023
pmcid: PMC10013058
doi: 10.3389/fonc.2023.938042
license: CC BY 4.0
---
# Breast cancer subtype and clinical characteristics in women from Peru
## Abstract
### Introduction
Breast cancer is a heterogeneous disease, and the distribution of the different subtypes varies by race/ethnic category in the United States and by country. Established breast cancer-associated factors impact subtype-specific risk; however, these included limited or no representation of Latin American diversity. To address this gap in knowledge, we report a description of demographic, reproductive, and lifestyle breast cancer-associated factors by age at diagnosis and disease subtype for The Peruvian Genetics and Genomics of Breast Cancer (PEGEN-BC) study.
### Methods
The PEGEN-BC study is a hospital-based breast cancer cohort that includes 1943 patients diagnosed at the Instituto Nacional de Enfermedades Neoplásicas in Lima, Peru. Demographic and reproductive information, as well as lifestyle exposures, were collected with a questionnaire. Clinical data, including tumor Hormone Receptor (HR) status and Human Epidermal Growth Factor Receptor 2 (HER2) status, were abstracted from electronic medical records. Differences in proportions and mean values were tested using Chi-squared and one-way ANOVA tests, respectively. Multinomial logistic regression models were used for multivariate association analyses.
### Results
The distribution of subtypes was $52\%$ HR+HER2-, $19\%$ HR+HER2+, $16\%$ HR-HER2-, and $13\%$ HR-HER2+. Indigenous American (IA) genetic ancestry was higher, and height was lower among individuals with the HR-HER2+ subtype ($80\%$ IA vs. $76\%$ overall, $$p \leq 0.007$$; 152 cm vs. 153 cm overall, $$p \leq 0.032$$, respectively). In multivariate models, IA ancestry was associated with HR-HER2+ subtype (OR=1.38,$95\%$CI=1.06-1.79, $$p \leq 0.017$$) and parous women showed increased risk for HR-HER2+ (OR=2.7,$95\%$CI=1.5-4.8, $p \leq 0.001$) and HR-HER2- tumors (OR=2.4,$95\%$CI=1.5-4.0, $p \leq 0.001$) compared to nulliparous women. Multiple patient and tumor characteristics differed by age at diagnosis (<50 vs. >=50), including ancestry, region of residence, family history, height, BMI, breastfeeding, parity, and stage at diagnosis ($p \leq 0.02$ for all variables).
### Discussion
The characteristics of the PEGEN-BC study participants do not suggest heterogeneity by tumor subtype except for IA genetic ancestry proportion, which has been previously reported. Differences by age at diagnosis were apparent and concordant with what is known about pre- and post-menopausal-specific disease risk factors. Additional studies in Peru should be developed to further understand the main contributors to the specific age of onset and molecular disease subtypes in this population and develop population-appropriate predictive models for prevention.
## Introduction
Globally, breast cancer is the most commonly diagnosed cancer and the leading cause of cancer death in women [1, 2]. Breast cancer risk and mortality vary based on several risk factors. Age, race/ethnicity category, family history, genetics, lifestyle, anthropometric, reproductive, and hormonal factors have been associated with the risk of developing breast cancer (3–5). In addition, tumor subtype, socioeconomic status, education level, and access to care have been shown to impact mortality after diagnosis [6, 7]. Analyses stratified by race/ethnicity category have shown that despite sharing risk factors for developing breast cancer, disease risk, clinical characteristics, and risk of mortality differ between populations (6, 8–10). For example, U.S. Hispanics/Latinas (H/Ls) are less likely to develop breast cancer than non-Hispanic White (NHW) and African American women [11]. However, after diagnosis, H/L women are at higher risk of mortality compared with NHW women [12].
The use of gene expression profiles for molecular classification of breast cancer tumors (i.e., PAM50) has identified three main intrinsic subtypes: Luminal (A and B), HER2-enriched, and Basal-like [13, 14]. A combination of immunohistochemical markers for estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor 2 (HER2) are routinely used in clinic to classify tumors into these subtypes and to provide relevant information for individualized therapeutic decision making. Hormone receptor (HR) positive tumors, defined by ER and/or PR expression, are classified as HR+HER2− and HR+HER2+, based on the HER2 expression status, and are overrepresented among luminal intrinsic subtypes. HR−HER2+ and HR−HER2− are overrepresented among HER2-enriched and basal-like subtypes, respectively. Besides chemotherapy, patients with an HR+ disease diagnosis can benefit from endocrine therapy, such as tamoxifen or aromatase inhibitors [15], whereas patients with HER2+ tumors can be treated with anti-HER2 therapy (mainly trastuzumab and pertuzumab) [16]. For the HR−HER2− subtype, treatment options are limited. Currently, these patients receive systemic therapy, although targeted therapies, such as PARP and immune checkpoint inhibitors, are being evaluated in clinical trials and approved for BRCA1 and BRCA2 mutation carriers [17].
Multiple studies have suggested heterogeneity in the association between established breast cancer risk factors and tumor subtype. Family history of breast cancer in a first-degree relative is associated with increased breast cancer risk [3, 18, 19], and specific patterns of cancer family history increase the risk of particular tumor subtypes [20, 21]. For example, having one first-degree relative with a history of breast cancer was shown to be associated with increased risk of HR+ subtypes, whereas having two or more was associated with increased risk of HR− disease. [ 20, 21]. However, some studies have failed to confirm these findings (3, 22–24). Among reproductive factors, early menarche, and late menopause increase the risk of developing breast cancer (3, 20, 25–27) with no evidence of heterogeneity by tumor subtype [3, 20, 26, 27]. Parity is associated with reduced risk of HR+ disease (3, 19, 20, 27–33) and increased odds for developing HR− subtypes (3, 24, 27, 31, 33–35) in populations of European and African origins. Some studies have reported that older age at first full-term pregnancy was associated with increased risk of HR+ disease [27, 28, 30]. Longer breastfeeding history is associated with reduced breast cancer risk with lower odds of developing HR− tumors (19, 20, 25–28, 30–34, 36). Among African Americans, prolonged lactation is associated with reduced risk of HR−, but not HR+ disease, with an increased risk of HR− disease among parous women who have not breastfed [34, 37]. This observation has also been described among NHW women [32]. Reports on lifestyle factors, such as alcohol intake and smoking history, have shown heterogeneity by tumor subtype, with a stronger association with HR+HER2− subtypes [3, 38].
The effects of some of the abovementioned factors are different among pre- and post-menopausal women. Controversial evidence shows that high BMI (obesity) is protective against breast cancer in premenopausal women, and conversely, it suggests that obesity increases the risk in postmenopausal women [39, 40], especially for HR+ subtypes (41–43). Other factors known to affect breast cancer risk in both groups in the same direction can present different magnitudes of the effect by menopausal status, such as alcohol intake [44], physical activity [45, 46], and breastfeeding [47].
Previous studies have assessed the association of breast cancer risk with numerous structural, social, environmental, and genetic factors (4, 48–50); however, these studies are primarily composed of individuals of European origin. Few breast cancer studies describe patient characteristics in Latin America (26, 51–54), a region characterized by cultural and genetic heterogeneity (55–57). For example, Indigenous *American* genetic ancestry estimates vary across different Latin American countries, ranging between ~$5\%$ in Puerto Rico and ~$80\%$ in Peru and Bolivia (56–58). Previous studies have identified that the degree of Indigenous *American* genetic ancestry may modify the magnitude and direction of association with currently known breast cancer risk variants among H/L women [59] and is associated with differential lifestyle risk factors [60]. Latin American cohorts with high proportions of Indigenous American ancestry are underrepresented in breast cancer research [61].
The Peruvian Genetics and Genomics of Breast Cancer Study (PEGEN-BC) is a hospital-based cohort including patients from the Instituto Nacional de Enfermedades Neoplásicas (INEN) in Lima, Peru. We have previously described the distribution of demographic, anthropometric, reproductive, lifestyle, and clinical factors for 1,312 breast cancer participants, with an emphasis on the distribution by breast tumor subtypes [62]. Moreover, we reported that increasing Indigenous American ancestry is associated with higher odds of developing the HR−HER2+ subtype [62]. The current report aims to provide a more complete and updated description of these variables by tumor subtype and age at diagnosis, including a total of 1,943 breast cancer patients, highlighting potential heterogeneity in the latter categories.
## Study participants
The Peruvian Genetics and Genomics of Breast Cancer Study (PEGEN-BC) is a hospital-based cohort study. As of April 2022, we have recruited 1,943 participants from the INEN in Lima, Peru. Women were invited to participate if they had a diagnosis of invasive breast cancer in 2010 or later and were between 21 and 79 years of age when diagnosed. A blood sample was drawn by a certified phlebotomist at the INEN central laboratory. The present report includes analyses with a subset of 1,796 patients with available genetic ancestry estimates [63]. This study was approved by the INEN and the University of California Davis Institutional Review Boards. All individuals provided written informed consent to participate.
## Data collection
Each PEGEN-BC participant completed a standardized survey administered by a trained research coordinator at INEN. The survey includes questions regarding anthropometric (weight and height), demographic (place of birth and residence), lifestyle (alcohol intake and smoking history), and reproductive (menopause status, age at first pregnancy, number of full-term pregnancies, and breastfeeding history) variables, and family history of breast cancer. Weight and height were assessed by trained nurses/professionals at INEN at the time of diagnosis. Body mass index (BMI) was calculated as weight (kilograms) divided by height (meters) squared and categorized as underweight (BMI < 18.5 kg/m 2), normal (BMI ≥ 18.5 < 25 kg/m2), overweight (BMI ≥ 25 < 30 kg/m 2), and obese (BMI ≥ 30 kg/m2). Alcohol use was assessed as the self-reported frequency of glasses of alcohol consumed per day and categorized as < 1 glass/day, > 1 glass/day, and non-drinker (never). Smoking status was classified into “ever” (current and former) and “never.” If there was a history of familial breast cancer, the relative (i.e., mother, sister, and aunt) was indicated to determine cases with breast cancer family history in a first-degree relative. Clinical variables, including ER, PR, HER2, lymph node status, tumor grade, and clinical stage, were extracted from electronic records.
Genetic ancestry estimates for 1,796 PEGEN-BC participants were available from a previous study [63]. Briefly, genome-wide genotype data obtained with the Affymetrix Precision Medicine Array were pruned using PLINK v.1.9 [64] [window size = 50, number of variants = 5, variance inflation factor threshold = 2] and merged with data from four reference populations from the 1000 Genomes project [65]: Admixed Americans (Peru, Colombia, Mexico, Puerto Rico), Europeans (Americans with Northern and Western European Ancestry, Italy, Spain, Finland, Scotland), East Asians (China, Japan, Vietnam), and African populations (Nigeria, Kenya, Gambia, Sierra Leone). Individual continental, global genetic ancestry was estimated using ADMIXTURE [66] (unsupervised, $k = 4$), including 122,605 independent variants. The PEGEN-BC study includes a large proportion of patients with > $98\%$ Indigenous American ancestry, as previously reported [62], and therefore provides a source of non-admixed reference samples for this component.
Tumoral tissues were obtained from core biopsy or freshly resected invasive breast cancers pre-treatment that were formalin-fixed and paraffin-embedded following standard protocols at INEN. Tumor subtypes were defined using immunohistochemistry (IHC) markers by a certified pathologist at INEN. HR positivity was defined at $1\%$ or more cells showing ER and/or PR staining. HER2 positivity was defined as 3+ staining by IHC or by gene amplification detected by fluorescence in situ hybridization following a 2+ (borderline) IHC result. These markers were used to classify tumors as HR+HER2−, HR+HER2+, HR−HER2+, and HR−HER2−. Two independent pathologists from the University of California San Francisco reviewed the IHC slides at INEN for a subset of 52 patients. The concordance rate was $100\%$ for ER, $87\%$ for PR, and $85\%$ for HER2. Most of the discordant calls for HER2 were scored as “negative” or 1+ at INEN and 2+ by the independent pathologists. Immunohistochemical subtype classification was not available for 141 samples ($7\%$).
## Statistical analysis
We performed descriptive analyses of available demographic, anthropometric, reproductive, and clinical characteristics by breast cancer subtype. Differences in characteristics between tumor subtypes were tested by means of one-way ANOVA for normally distributed continuous variables and Chi-squared tests for categorical variables. Age at first full-term pregnancy presented a non-normal distribution; therefore, it was log2 transformed. The correlation between genetic ancestry and continuous and categorical variables was performed using Pearson’s correlation coefficient test and Point-Biserial Correlation Coefficient, respectively. Multinomial logistic regression models were used to calculate odds ratios (ORs) and $95\%$ confidence intervals (CI) for the association of multiple variables and subtype-specific breast cancer. East Asian and African ancestry proportions were not included in multivariable models due to the low contribution of these components and high correlation with the Indigenous American/*European axis* of ancestry variation. P-values (P) <= 0.05 were considered statistically significant. All analyses were conducted in R v.3.6.0 [67].
## Demographics, anthropometrics, and lifestyle factors in the PEGEN-BC study by tumor subtype
The most common breast cancer subtype among PEGEN-BC study participants was HR+HER2− ($52.4\%$), followed by HR+HER2+ ($18.7\%$), HR−HER2− ($16.0\%$), and HR−HER2+ ($12.9\%$) (Table 1). The average age at diagnosis was 49.8 years (SD = 11), and differences by tumor subtype were not statistically significant ($$p \leq 0.087$$). PEGEN-BC study patients included individuals born in the three main biogeographic regions of Peru (Figure 1): The Coastal ($55.5\%$), Mountainous ($36.4\%$), and Amazonian ($7.5\%$) regions. Less than $1\%$ of the patients were born in another country (mainly Venezuela). These groups did not show statistically significant differences in their distribution by tumor subtype (Table 1). Most patients resided in the Coastal region ($7\%$), and differences in the proportion of patients who resided in each biogeographic area by tumor subtype category were not statistically significant (Table 1).
Estimates of individual continental genetic ancestry were available for 1,796 patients. Average Indigenous American ancestry among patients was $76.5\%$, followed by $18.0\%$ European, $4.2\%$ African, and $1.4\%$ East Asian (Table 1). Furthermore, $92\%$ of PEGEN-BC study participants had > $50\%$ of Indigenous American ancestry, $25\%$ at least $90\%$, and $12\%$ at least $95\%$ of Indigenous American ancestry (Figure 2A). Seven patients ($0.4\%$) had more than $50\%$ of East Asian ancestry, and eight ($0.4\%$) had more than $50\%$ African ancestry. Principal components analysis showed that the PEGEN-BC patients defined the Indigenous American cluster along principal component (PC) 1 when compared against 1000 Genomes Project reference populations (Figures 2B, C), reflecting the high degree of Indigenous *American* genetic ancestry that characterizes this cohort.
**Figure 2:** *Population genetic structure of the PEGEN-BC study participants. (A) ADMIXTURE continental ancestry estimates obtained in unsupervised analysis, assuming K = 4. (B, C) Principal components analysis (PCA) including breast cancer patients and 1000 Genomes Project individuals. The first three principal components are shown.*
We found that the average Indigenous American ancestry proportion of participants was different across tumor subtypes. Individuals diagnosed with HR−HER2+ tumors showed the highest average proportion of Indigenous American ancestry ($79.5\%$, SD = 15) (Table 1).
The average height of patients was 153.3 cm (SD = 6.6), with lower average height among patients diagnosed with HR−HER2+ tumors compared with all other subtypes (152.1 vs. ~153.6 cm, $$p \leq 0.032$$). There were no statistically significant differences in weight or BMI by tumor subtype, with a large overall proportion of patients being overweight ($40.1\%$) (Table 1).
Most PEGEN-BC patients ($68.7\%$) reported low levels of alcohol consumption (< 1 glass/day), whereas $7.4\%$ reported consuming more than one glass per day. Moreover, $27.9\%$ of participants reported being a current or past smoker. There was no statistically significant association between alcohol consumption, smoking history, and tumor subtype (Table 1).
Demographic, anthropometric, and lifestyle variables that did not show statistically significant differences by tumor subtypes did not show significant differences by HR status either (Supplementary Table S1).
## Reproductive variables by tumor subtype
The average age at menarche among PEGEN-BC patients was 12.9 years (SD = 1.7), the average age at first full-term pregnancy was 23.2 years (SD = 5.7), and the average number of full-term pregnancies was 2.42 (SD = 1.8). Most study participants reported having had at least one child ($83.5\%$), and $80\%$ of parous women had at least two children (Table 2). The frequency of parous women and number of births differed by tumor subtype, being higher among HR− tumors ($p \leq 0.001$) (Table 2).
**Table 2**
| Variable | Overall | HR+HER2− | HR+HER2+ | HR-HER2+ | HR−HER2− | p-value |
| --- | --- | --- | --- | --- | --- | --- |
| Number of patients, N (%) | 1943 (100) | 945 (52.4) | 337 (18.7) | 232 (12.9) | 288 (16.0) | |
| Age at menarche in years, mean (SD) | 12.9 (1.7) | 12.9 (1.8) | 12.9 (1.7) | 13.1 (1.7) | 13.0 (1.7) | 0.364 |
| Missing, N (%) | 34 (1.8) | 17 (1.8) | 3 (0.9) | 7 (3.0) | 2 (0.7) | |
| Parous, yes, N (%) | 1623 (83.5) | 765 (81.0) | 273 (81.0) | 207 (89.2) | 263 (91.3) | < 0.001 |
| Missing, N (%) | 63 (3.2) | 32 (3.4) | 10 (3.0) | 9 (3.9) | 4 (1.4) | |
| Age at first full-term pregnancy in years, mean (SD) | 23.2 (5.7) | 23.5 (5.8) | 23.0 (5.3) | 22.9 (6.1) | 22.5 (5.4) | 0.095 |
| Missing*, N (%) | 72 (4.4) | 40 (5.2) | 7 (2.6) | 13 (6.3) | 6 (2.3) | |
| Parity, mean (SD) | 2.4 (1.8) | 2.3 (1.8) | 2.3 (1.9) | 2.7 (1.8) | 2.7 (1.7) | 0.002 |
| Missing*, N (%) | 6 (0.4) | 4 (0.5) | 1 (0.4) | 1 (0.5) | 0 (0.0) | |
| Parity categories, N (%) | Parity categories, N (%) | Parity categories, N (%) | Parity categories, N (%) | Parity categories, N (%) | Parity categories, N (%) | Parity categories, N (%) |
| No children | 275 (14.2) | 156 (16.5) | 57 (16.9) | 19 (8.2) | 22 (7.6) | < 0.001 |
| 1 child | 316 (16.3) | 162 (17.1) | 47 (13.9) | 38 (16.4) | 46 (16.0) | |
| 2–3 children | 888 (45.7) | 410 (43.4) | 161 (47.8) | 105 (45.3) | 148 (51.4) | |
| >3 children | 413 (21.3) | 189 (20.0) | 64 (19.0) | 63 (27.2) | 69 (24.0) | |
| Missing, N (%) | 51 (2.6) | 28 (3.0) | 8 (2.4) | 7 (3.0) | 3 (1.0) | |
| Breastfed*, yes, N (%) | 1563 (96.3) | 736 (96.2) | 264 (96.7) | 200 (96.6) | 255 (97.0) | 0.967 |
| Missing*, N (%) | 2 (0.1) | 1 (0.1) | 0 (0.0) | 0 (0.0) | 0 (0.0) | |
| Postmenopausal, N (%) | 1681 (86.5) | 839 (88.8) | 287 (85.2) | 198 (85.3) | 240 (83.3) | 0.016 |
| Missing, N (%) | 23 (1.2) | 12 (1.3) | 1 (0.3) | 5 (2.2) | 1 (0.3) | |
Breastfeeding was a common practice among parous women ($96.3\%$), and we did not observe the differences in the proportion of women who breastfed their children by tumor subtype category (Table 2).
More than $85\%$ of women reported being menopausal at recruitment. Patients with HR+HER2− tumors were more likely to report being menopausal than patients with other tumor subtypes ($$p \leq 0.016$$). However, since many of these patients had induced menopause due to treatment, we did not consider this variable in subsequent multivariate analyses and stratified by age at diagnosis instead.
All these variables remained significant in analyses stratified by HR status (Supplementary Table S2). In addition, age at first full-term pregnancy showed a higher average age among patients diagnosed with HR+ disease compared with HR− (23.4 vs. 22.7, $$p \leq 0.043$$, Supplementary Table S2).
## Clinical characteristics by tumor subtype
Overall, approximately $8\%$ of PEGEN-BC study patients reported a family history of breast cancer in a first-degree relative (Table 3). Differences in breast cancer family history by breast cancer subtype were not statistically significant.
**Table 3**
| Variable | Overall | HR+HER2− | HR+HER2+ | HR−HER2+ | HR−HER2− | p-value |
| --- | --- | --- | --- | --- | --- | --- |
| Number of patients, N (%) | 1943 (100)* | 945 (52.4) | 337 (18.7) | 232 (12.9) | 288 (16.0) | |
| Positive family history of breast cancer**, N (%) | 149 (7.7) | 84 (8.9) | 25 (7.4) | 9 (3.9) | 23 (8.0) | 0.091 |
| Missing | 61 (3.1) | 19 (2.0) | 5 (1.5) | 7 (3.0) | 2 (0.7) | |
| Grade, N (%) | Grade, N (%) | Grade, N (%) | Grade, N (%) | Grade, N (%) | Grade, N (%) | Grade, N (%) |
| 1 | 72 (3.7) | 58 (6.1) | 6 (1.8) | 0 (0.0) | 4 (1.4) | < 0.001 |
| 2 | 803 (41.3) | 550 (58.2) | 117 (34.7) | 36 (15.5) | 37 (12.8) | |
| 3 | 1005 (51.7) | 317 (33.5) | 209 (62.0) | 192 (82.8) | 239 (83.0) | |
| Missing | 63 (3.2) | 20 (2.1) | 5 (1.5) | 4 (1.7) | 8 (2.8) | |
| Stage, N (%) | Stage, N (%) | Stage, N (%) | Stage, N (%) | Stage, N (%) | Stage, N (%) | Stage, N (%) |
| I | 122 (6.3) | 67 (7.1) | 18 (5.3) | 7 (3.0) | 23 (8.0) | < 0.001 |
| II | 840 (43.2) | 480 (50.8) | 134 (39.8) | 70 (30.2) | 109 (37.8) | |
| III | 798 (41.1) | 332 (35.1) | 158 (46.9) | 137 (59.1) | 139 (48.3) | |
| IV | 105 (5.4) | 49 (5.2) | 19 (5.6) | 16 (6.9) | 12 (4.2) | |
| Missing | 78 (4.0) | 17 (1.8) | 8 (2.4) | 2 (0.9) | 5 (1.7) | |
| Positive lymph node status, N (%) | 1249 (64.3) | 585 (61.9) | 227 (67.4) | 176 (75.9) | 177 (61.5) | 0.002 |
| Missing | 90 (4.6) | 43 (4.6) | 9 (2.7) | 7 (3.0) | 21 (7.3) | |
More than $90\%$ of patients were diagnosed with Grades 2 and 3 tumors (Table 3). Patients with HR+HER2− tumors were more likely to be diagnosed with Grades 1 and 2 disease, whereas those with HR−HER2+ and HR−HER2− tumors were more likely to be high grade (Table 3). Most PEGEN-BC participants were diagnosed with stage II or III disease, with a larger number of stage I and II diagnoses among HR+HER2− patients than those with other subtypes (Table 3). Concordant with the distribution of tumor stage, we observed a high proportion of positive lymph node status among patients overall ($64.3\%$), with a statistically significantly higher proportion of lymph node positivity among patients with HR−HER2+ tumors compared with those with other disease subtypes ($78.2\%$ vs. ~$67\%$) (Table 3). Distribution of these variables by HR status is shown in Supplementary Table S2.
## Distribution of patient characteristics by age at diagnosis
We compared the distribution of anthropometric, demographic, reproductive, clinical, and lifestyle risk variables between patients diagnosed before the age of 50 years ($$n = 981$$) and at 50 years or older ($$n = 955$$). Compared with patients diagnosed at 50 years or older, younger patients had higher average Indigenous American ancestry (78.6 vs. 74.3, $p \leq 0.001$); they were more likely to reside in the Mountainous region ($17.3\%$ vs. $12.8\%$, $$p \leq 0.015$$), and they were 1.4 cm taller ($p \leq 0.001$) and had lower prevalence of obesity ($25.4\%$ vs. $30.0\%$, $$p \leq 0.036$$) (Table 4). Additionally, there was a higher proportion of older patients with more than three children compared with the younger group ($31\%$ vs. $13\%$, $p \leq 0.001$), and a larger proportion of younger patients reported breastfeeding their children ($98\%$ vs. $95\%$, $$p \leq 0.001$$) (Table 5). Regarding clinical characteristics, younger patients reported lower family history of breast cancer in a first-degree relative ($6.5\%$ vs. $9.5\%$, $$p \leq 0.02$$) and presented with more advanced disease ($44\%$ diagnosed at stage III compared with $42\%$, $$p \leq 0.017$$) (Table 5). We did not observe statistically significant differences in subtype distribution between both age categories.
Additional stratified analyses comparing demographic, anthropometric, reproductive, and clinical factors by tumor subtype in the two different age groups are included as Supplementary Materials (Supplementary Tables S3 and S4). As additional stratification reduced the number of observations per category, we suggest taking these results with caution.
## Correlation between Indigenous American genetic ancestry and other patient and tumor characteristics
We assessed the correlation between Indigenous American ancestry and patient and tumor characteristics to better understand the observed patterns in ancestry distribution and those factors by tumor subtype in the PEGEN-BC study. We observed an inverse correlation between Indigenous American ancestry and age at diagnosis (r = −0.15, $p \leq 0.001$), weight (r = −0.11, $p \leq 0.001$), height (r = −0.25, $p \leq 0.001$), age at first full-term pregnancy (r = −0.08, $$p \leq 0.002$$), family history of breast cancer in a first-degree relative (r = −0.12, $p \leq 0.001$), smoking history (r = −0.11, $p \leq 0.001$), HR+ status (r = −0.06, $$p \leq 0.012$$) and a positive correlation with age at menarche ($r = 0.06$, $$p \leq 0.017$$) and HER2+ status ($r = 0.053$, $$p \leq 0.029$$).
## Multivariable analyses testing the association between demographic, lifestyle factors, and breast cancer subtype
Variables that showed statistically significant associations at the $10\%$ level with tumor subtype in the univariate analyses (Tables 1 – 3) were included in a multivariate model, using HR+HER2− as reference (Table 6). Indigenous American ancestry remained associated with HR−HER2+ subtype (OR per $25\%$ increment in ancestry = 1.38, $95\%$ CI = 1.06–1.79, $$p \leq 0.02$$). Smoking history and height were no longer statistically significantly associated with subtype. Parous women were more likely to be diagnosed with HR−HER2+ (OR = 2.72, $95\%$ CI = 1.53–4.83, $p \leq 0.001$) and HR-HER2- (OR = 2.47, $95\%$ CI = 1.51–4.04, $p \leq 0.001$) disease compared with the HR+HER2− subtype. Family history of breast cancer in a first-degree relative was not included as a covariate in the multivariate model because the number of patients that reported family history of breast cancer in a first-degree relative was relatively small and rendered unstable estimates when included. We tested models excluding patients with a family history of breast cancer, and results were similar to those using the full dataset (Table 6).
**Table 6**
| Unnamed: 0 | Unnamed: 1 | All patients* | All patients*.1 | All patients*.2 | Patients without FamHist | Patients without FamHist.1 | Patients without FamHist.2 |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Subtype | Variable | OR | 95% CI | p-value | OR | 95% CI | p-value |
| HR+HER2+ | Indigenous American ancestry (Every 25% increment) | 1.09 | 0.89–1.34 | 0.402 | 1.14 | 0.91–1.41 | 0.255 |
| HR+HER2+ | Age at diagnosis (Every 5-year increment) | 0.99 | 0.98–1.00 | 0.062 | 0.99 | 0.98–1.00 | 0.188 |
| HR+HER2+ | Height (Every 1-cm increment) | 1.01 | 0.99–1.03 | 0.257 | 1.01 | 0.99–1.04 | 0.196 |
| HR+HER2+ | Smoking history (Ever vs. never [reference]) | 0.81 | 0.60–1.10 | 0.178 | 0.84 | 0.61–1.15 | 0.267 |
| HR+HER2+ | Parous (Reference: nulliparous) | 1.20 | 0.83–1.74 | 0.335 | 1.43 | 0.96–2.14 | 0.082 |
| HR−HER2+ | Indigenous American ancestry (Every 25% increment) | 1.38 | 1.06–1.79 | 0.017 | 1.37 | 1.05–1.80 | 0.022 |
| HR−HER2+ | Age at diagnosis (Every 5-year increment) | 0.99 | 0.98–1.01 | 0.455 | 1.00 | 0.98–1.01 | 0.717 |
| HR−HER2+ | Height (Every 1-cm increment) | 0.98 | 0.96–1.01 | 0.166 | 0.99 | 0.96–1.01 | 0.266 |
| HR−HER2+ | Smoking history (Ever vs. never [reference]) | 0.75 | 0.52–1.08 | 0.122 | 0.74 | 0.51–1.08 | 0.118 |
| HR−HER2+ | Parous (Reference: nulliparous) | 2.72 | 1.53–4.83 | < 0.001 | 2.60 | 1.46–4.64 | 0.001 |
| HR−HER2− | Indigenous American ancestry (Every 25% increment) | 1.17 | 0.93–1.46 | 0.177 | 1.25 | 0.99–1.59 | 0.065 |
| HR−HER2− | Age at diagnosis (Every 5-year increment) | 0.99 | 0.98–1.00 | 0.100 | 0.99 | 0.98–1.01 | 0.467 |
| HR−HER2− | Height (Every 1-cm increment) | 1.01 | 0.99–1.03 | 0.446 | 1.02 | 0.99–1.04 | 0.205 |
| HR−HER2− | Smoking history (Ever vs. never [reference]) | 0.78 | 0.56–1.08 | 0.133 | 0.72 | 0.51–1.01 | 0.061 |
| HR−HER2− | Parous (Reference: nulliparous) | 2.47 | 1.51–4.04 | < 0.001 | 2.40 | 1.44–3.99 | 0.001 |
Indigenous American ancestry, region of residence, height, BMI, breastfeeding history, number of full-term pregnancies, and family history of breast cancer in a first-degree relative showed statistically significant associations at the $10\%$ level with age at diagnosis categories. These variables were included in a multivariate model using age at diagnosis < 50 as reference (Table 7). We found that increasing Indigenous American ancestry and increasing height were associated with reduced odds of being diagnosed at 50 years or older (OR = 0.63, $95\%$ CI = 0.53–0.75, $p \leq 0.001$ and OR = 0.96, $95\%$ CI = 0.95–0.98, $p \leq 0.001$, respectively). Patients that resided in the Mountainous region had reduced odds of being diagnosed at 50 years of age or older compared with those in the Coastal region (OR = 0.63, $95\%$ CI = 056–0.9, $$p \leq 0.004$$). Breastfeeding was associated with lower odds of being diagnosed at 50 years of age or older (OR = 0.35, $95\%$ CI = 0.2–0.7, $$p \leq 0.001$$). Compared with nulliparous women, giving birth to at least one child increased the odds of being diagnosed at an older age (OR = 1.55, $95\%$ CI = 0.2–0.7, $p \leq 0.001$). Increasing BMI was no longer associated with age at diagnosis (Table 7).
**Table 7**
| Variable | OR* | 95% CI | p-value |
| --- | --- | --- | --- |
| Indigenous American ancestry (Every 25% increment) | 0.63 | 0.53–0.75 | < 0.001 |
| Height (Every 1-cm increment) | 0.96 | 0.95–0.98 | < 0.001 |
| Region of residence (Reference: Coastal region) Amazonian region Mountainous region | 0.680.63 | 0.43–1.070.46–0.86 | 0.1000.004 |
| BMI (Every 1 kg/m2 increment) | 1.02 | 1–1.05 | 0.080 |
| Parity (Per each additional child) | 1.55 | 1.43–1.69 | < 0.001 |
| Breastfed (Yes vs. no [reference]) | 0.35 | 0.20–0.70 | 0.001 |
| Family history of breast cancer** (Yes vs. no [reference]) | 1.20 | 0.78–1.84 | 0.410 |
## Discussion
In the present report, we aimed to provide a more complete description of the distribution of anthropometric, demographic, clinical, and known breast cancer–associated risk factors among Peruvian women that are part of The Peruvian Genetics and Genomics of Breast Cancer Study (PEGEN-BC). This work constitutes an update of a previously reported study, including a larger number of recruited patients and extending analyses to describe the distribution of patient characteristics not only by tumor subtype but also by age at diagnosis [62].
Being a hospital-based cohort, the PEGEN-BC study included a large proportion of women who resided in the Coastal region, where the INEN main hospital is located (Figure 1). Despite this bias in terms of residential representation, when looking at place of birth, the proportion of the cohort’s patients from the Coastal region followed closely that of the Peruvian population ($58.0\%$ Peru vs. $55.5\%$ of cohort patients). The study has an overrepresentation of patients born in the Mountainous region ($28.1\%$ Peru vs. $36.4\%$ of cohort patients) [68] and an underrepresentation of patients born in the Amazonian region ($13.9\%$ Peru vs. $7.5\%$ of cohort patients) [68]. The proportion of patients within each geographical region is consistent with what has been reported in two studies describing mortality of breast cancer [69] and incidence of triple-negative breast cancer tumors in Peruvian women [70].
A large proportion of patients were overweight/obese ($67\%$), and the prevalence of exposure to alcohol and tobacco was higher than what has been previously reported for Peruvian women [71, 72]. The average Indigenous American ancestry among the PEGEN-BC patients is $76.5\%$, which is higher than the average ancestry proportion of women in other breast cancer studies, including Latin America and U.S. Latinas (12, 51, 60, 73–89). In addition, the average height in our cohort was consistent with what has been reported in the literature for the Peruvian population [90] and with the known inverse correlation with Indigenous American ancestry [91]. Overall, some reproductive variables showed a similar trend to what has been reported, including a similar age at menarche [92] and a high breastfeeding rate [93]. The number of full-term pregnancies reported here (average of 2.8 children) was more closely related to what has been observed in rural areas of Peru (2.5) compared with urban areas (1.4) [94].
The distribution of tumor subtypes is similar to what has been previously described in other Latin American countries [95], with differences being partially explained by the inclusion of KI-67 expression and tumor grade for subtype classification [95], as indicated by the 2013 St. Gallen consensus [96]. This classification criterion was not used in this report since KI-67 was not available for more than $20\%$ of patients, and parameters for subtype determination based on this marker tend to be unstable across populations and studies [97]. A study describing patient and tumor characteristics from Peruvian breast cancer patients at INEN diagnosed between 2000 and 2002 [80] (PEGEN-BC patients were recruited if diagnosed in 2010 or later) reported a lower proportion of HR+ tumors compared with PEGEN-BC ($62.5\%$ vs. $71.1\%$). This difference is likely to be explained by the higher positivity percentage cutoff value for HR+ status used in the previous report ($10\%$, compared with $1\%$ in PEGEN-BC), increasing the proportion of HR+ tumors in our cohort. Other characteristics, such as age at diagnosis and stage, presented similar distribution to the PEGEN-BC study cohort.
We found statistically significant differences by tumor subtype for Indigenous *American* genetic ancestry and height. In addition, we observed suggestive associations for age at diagnosis, family history of breast cancer in a first-degree relative and tobacco exposure. Differences were mostly driven by the HR−HER2+ subtype. Among patients with HR−HER2+ disease, we observed that the average height was lower compared with patients diagnosed with other tumor subtypes and was less likely to report smoking or a positive family history of breast cancer in a first-degree relative. Even though subtype-specific associations have been reported for these variables in other populations (38, 98–101), results in the Peruvian cohort showed that of all the above variables Indigenous American ancestry proportion was the only one that was differentially distributed by tumor subtype in multivariable models.
We did not find statistically significant differences for age at menarche by tumor subtype. Some studies have shown consistent associations between age at menarche and reduced risk of HR+HER2− breast cancer [3, 19, 20]. One multicenter study did not find subtype-specific associations [27], consistent with our study. The PRECAMA Study, a Latin American population-based case-control study of premenopausal breast cancer, reported reduced odds for HR− tumors among women who were > 12 years old at menarche, compared with those younger at menarche [26, 51]. In the current study, we did not find a statistically significant difference in average age at menarche by tumor subtype despite the observed correlation between the former and Indigenous American ancestry proportion.
We observed a higher frequency of parous women diagnosed with HR− subtypes compared with HR+. Parity (ever vs. never) has been associated with a higher risk of HR−HER2− subtypes, especially among women of African origin (33–35). Higher number of full-term pregnancies has been associated with reduced breast cancer risk [19, 31], with lower odds of developing HR+ tumors (3, 19, 20, 24–27, 29–35). We found significant differences in number of births by subtype, being higher among HR− subtypes compared with HR+ (2.7 compared with 2.3, respectively). Results suggested a larger proportion of women with > 3 children among those with HR− disease subtypes. This observation was consistent with studies in African American women reporting a higher number of reported full-term pregnancies among women with HR− disease [33]. Studies that have tested the association between age at first full-term pregnancy and subtype-specific risk have shown a decreased risk of developing HR+HER2− tumors with unclear associations for other subtypes [25, 27, 31]. In African American cohorts, limited breastfeeding among parous women is associated with an increased risk for HR−HER2− subtypes [34]. The current study does not include detailed pregnancy and lactation history for the patients. As a result, we could not assess the association between time to breastfeeding cessation and cumulative time of breastfeeding and HR− subtypes.
There were statistically significant differences in the prevalence of demographic, anthropometric, and reproductive factors by age at diagnosis categories. The multivariate analysis showed that these variables are independently associated with age at diagnosis. Moreover, the differences in BMI by age at diagnosis were concordant with what is known about pre- and post-menopausal–specific disease risk factors (39–43). It must be considered that the observed differences in parity and height by age at diagnosis could be due to the correlation between age and the former (i.e., number of children and height are positively correlated with age) and not to an association between those variables and pre- versus post-menopausal disease.
The observed association between tumor subtype and Indigenous American ancestry could be due to a multiplicity of factors that we might not have collected information on in the PEGEN-BC study. For example, the study did not obtain information on the level of education or socioeconomic status of participants; both variables were previously shown to be associated with Indigenous American ancestry) among U.S. Latinas and Mexican women [76, 102, 103]. Socioeconomic status can also impact screening, which in turn can affect tumor subtype distribution and mortality rates. Reports showed that less than $20\%$ of Peruvian women 40–59 years of age have had a mammography, with vast differences according to socioeconomic status, educational level, health insurance, and region of residence [104, 105]. Plan Esperanza, launched in 2012, has aimed to provide universal cancer screening and decentralize oncological health care across Peru, focusing on underserved commuties [106].
The PEGEN-BC study had some additional limitations. First, since menopause can be induced by treatment, most of the PEGEN-BC participants were postmenopausal at the time of the interview ($86\%$). Therefore, we did not perform stratification by menopausal status and used age at diagnosis (< 50 vs. >= 50) instead to differentiate early onset versus late onset disease, as it has been widely used in epidemiological studies [107, 108]. Even though menopausal status and age at diagnosis are highly correlated, studies have shown that age at diagnosis is a driver for breast cancer heterogeneity, acting as a confounder in analyses stratified by menopausal status [109]. For this reason, the use of age as a proxy for menopausal status should be taken with caution. The second limitation concerns the relatively low variability of some of the assessed factors among PEGEN-BC study participants. For example, the assessment of the association between breastfeeding and the number of births and tumor subtype was hampered by the low prevalence of women without children and of women with children who did not breastfeed them. Additionally, we described the distribution of multiple factors across tumor subtypes, which provide evidence of heterogeneity; however, future case-control design studies should further explore subtype-specific breast cancer risk. Finally, average East Asian and *African* genetic ancestry components showed differences by subtypes in the univariate analyses. However, since ancestry estimates are correlated, and the proportions of East Asian and *African* genetic ancestries were relatively low as to provide reliable estimates, we focused the current description on the Indigenous American ancestry, which is the dominant component in Peruvians.
In summary, results confirmed the previously reported higher average Indigenous American ancestry among patients with HR−HER2+ breast cancer in this larger sample of PEGEN-BC study participants. Moreover, differences in tumor subtype by age at diagnosis were apparent and concordant with what is known about pre- and post-menopausal–specific disease associated risk factors. Larger studies are needed to understand the consistently observed association between ancestry, age of onset, and disease subtypes, considering the contribution of screening and treatment, to develop population-appropriate predictive models and targeted outreach and prevention campaigns.
## Data availability statement
All data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by University of California Davis Institutional Review Boards and the Instituto Nacional de Enfermedades Neoplásicas (INEN). The patients/participants provided their written informed consent to participate in this study. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.
## Author contributions
LF= Conceptualization, Methodology, Investigation, Formal Analysis, Writing- Review and editing, Supervision, Project administration, and Funding acquisition. VZ: Methodology, Investigation, Formal Analysis, Writing- Original Draft, Software, Data curation, and Visualization. TV= Conceptualization, Resources, Project administration at INEN. SC-Z= Resources, Project administration at INEN. JN-V= Investigation, Data curation. CC, GV, MC, JA, HG, HF, RL-P, JC, SN, KR, JV, LM, MG-N= Conducted patient recruitment investigation process. All authors contributed to the article and approved the submitted version.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc.2023.938042/full#supplementary-material
## References
1. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A. **Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries**. *CA Cancer J Clin* (2021.0) **71**. DOI: 10.3322/caac.21660
2. Siegel RL, Miller KD, Jemal A. **Cancer statistics, 2020**. *CA Cancer J Clin* (2020.0) **70** 7-30. DOI: 10.3322/caac.21590
3. Barnard ME, Boeke CE, Tamimi RM. **Established breast cancer risk factors and risk of intrinsic tumor subtypes**. *Biochim Biophys Acta* (2015.0) **1856** 73-85. DOI: 10.1016/J.BBCAN.2015.06.002
4. Holm J, Eriksson L, Ploner A, Eriksson M, Rantalainen M, Li J. **Assessment of breast cancer risk factors reveals subtype heterogeneity**. *Cancer Res* (2017.0) **77**. DOI: 10.1158/0008-5472.CAN-16-2574
5. Jones ME, Schoemaker MJ, Wright LB, Ashworth A, Swerdlow AJ. **Smoking and risk of breast cancer in the generations study cohort**. *Breast Cancer Res* (2017.0) **19**. DOI: 10.1186/S13058-017-0908-4
6. Daly B, Olopade OI. **A perfect storm: How tumor biology, genomics, and health care delivery patterns collide to create a racial survival disparity in breast cancer and proposed interventions for change**. *CA Cancer J Clin* (2015.0) **65**. DOI: 10.3322/caac.21271
7. Newman LA. **Breast cancer disparities: Socioeconomic factors versus biology**. *Ann Surg Oncol* (2017.0) **24**. DOI: 10.1245/s10434-017-5977-1
8. 8
National Cancer Institute. SEER cancer stat facts: Female breast cancer. Bethesda, MD: National Cancer Institute.. *SEER cancer stat facts: Female breast cancer*
9. Zavala VA, Bracci PM, Carethers JM, Carvajal-Carmona L, Coggins NB, Cruz-Correa MR. **Cancer health disparities in racial/ethnic minorities in the united states**. *Br J Cancer* (2021.0) **124**. DOI: 10.1038/s41416-020-01038-6
10. Thompson B, Hohl SD, Molina Y, Paskett ED, Fisher JL, Baltic RD. **Breast cancer disparities among women in underserved communities in the USA**. *Curr Breast Cancer Rep* (2018.0) **10**. DOI: 10.1007/s12609-018-0277-8
11. Miller KD, Goding Sauer A, Ortiz AP, Fedewa SA, Pinheiro PS, Tortolero-Luna G. **Cancer statistics for Hispanics/Latinos, 2018**. *CA Cancer J Clin* (2018.0) **68**. DOI: 10.3322/caac.21494
12. Fejerman L, Hu D, Huntsman S, John EM, Stern MC, Haiman CA. **Genetic ancestry and risk of mortality among U.S. latinas with breast cancer**. *Cancer Res* (2013.0) **73**. DOI: 10.1158/0008-5472.CAN-13-2014
13. Sørlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H. **Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications**. *Proc Natl Acad Sci U.S.A.* (2001.0) **98**. DOI: 10.1073/pnas.191367098
14. Perou CM, Sørlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA. **Molecular portraits of human breast tumours**. *Nature.* (2000.0) **406**. DOI: 10.1038/35021093
15. Shien T, Iwata H. **Adjuvant and neoadjuvant therapy for breast cancer**. *Jpn J Clin Oncol* (2020.0) **50**. DOI: 10.1093/JJCO/HYZ213
16. Wang J, Xu B. **Targeted therapeutic options and future perspectives for HER2-positive breast cancer**. *Signal Transduct Target Ther* (2019.0) **4**. DOI: 10.1038/S41392-019-0069-2
17. Yin L, Duan JJ, Bian XW, Yu SC. **Triple-negative breast cancer molecular subtyping and treatment progress**. *Breast Cancer Res* (2020.0) **22**. DOI: 10.1186/S13058-020-01296-5
18. Liu L, Hao X, Song Z, Zhi X, Zhang S, Zhang J. **Correlation between family history and characteristics of breast cancer**. *Sci Rep* (2021.0) **11**. DOI: 10.1038/S41598-021-85899-8
19. Xing P, Li J, Jin F. **A case-control study of reproductive factors associated with subtypes of breast cancer in northeast China**. *Med Oncol* (2010.0) **27**. DOI: 10.1007/S12032-009-9308-7
20. Tamimi RM, Colditz GA, Hazra A, Baer HJ, Hankinson SE, Rosner B. **Traditional breast cancer risk factors in relation to molecular subtypes of breast cancer**. *Breast Cancer Res Treat* (2012.0) **131**. DOI: 10.1007/S10549-011-1702-0
21. Zhou W, Ding Q, Pan H, Wu N, Liang M, Huang Y. **Risk of breast cancer and family history of other cancers in first-degree relatives in Chinese women: a case control study**. *BMC Cancer.* (2014.0) **14** 1-7. DOI: 10.1186/1471-2407-14-662
22. Phipps AI, Buist DSM, Malone KE, Barlow WE, Porter PL, Kerlikowske K. **Family history of breast cancer in first-degree relatives and triple-negative breast cancer risk**. *Breast Cancer Res Treat* (2011.0) **126**. DOI: 10.1007/S10549-010-1148-9
23. Mavaddat N, Pharoah PD, Blows F, Driver KE, Provenzano E, Thompson D. **Familial relative risks for breast cancer by pathological subtype: a population-based cohort study**. *Breast Cancer Res* (2010.0) **12**. DOI: 10.1186/BCR2476
24. Yang XR, Chang-Claude J, Goode EL, Couch FJ, Nevanlinna H, Milne RL. **Associations of breast cancer risk factors with tumor subtypes: a pooled analysis from the breast cancer association consortium studies**. *J Natl Cancer Inst* (2011.0) **103**. DOI: 10.1093/JNCI/DJQ526
25. Phipps AI, Malone KE, Porter PL, Daling JR, Li CI. **Reproductive and hormonal risk factors for postmenopausal luminal, HER-2-overexpressing, and triple-negative breast cancer**. *Cancer.* (2008.0) **113**. DOI: 10.1002/CNCR.23786
26. Romieu I, Biessy C, Carayol M, His M, Torres-Mejía G, Ángeles-Llerenas A. **Reproductive factors and molecular subtypes of breast cancer among premenopausal women in Latin America: the PRECAMA study**. *Sci Rep* (2018.0) **8**. DOI: 10.1038/S41598-018-31393-7
27. Brouckaert O, Rudolph A, Laenen A, Keeman R, Bolla MK, Wang Q. **Reproductive profiles and risk of breast cancer subtypes: a multi-center case-only study**. *Breast Cancer Res* (2017.0) **19**. DOI: 10.1186/S13058-017-0909-3
28. Lambertini M, Santoro L, Del Mastro L, Nguyen B, Livraghi L, Ugolini D. **Reproductive behaviors and risk of developing breast cancer according to tumor subtype: A systematic review and meta-analysis of epidemiological studies**. *Cancer Treat Rev* (2016.0) **49** 65-76. DOI: 10.1016/J.CTRV.2016.07.006
29. Ellingjord-Dale M, Vos L, Tretli S, Hofvind S, dos-Santos-Silva I, Ursin G. **Parity, hormones and breast cancer subtypes - results from a large nested case-control study in a national screening program**. *Breast Cancer Res* (2017.0) **19**. DOI: 10.1186/S13058-016-0798-X
30. Turkoz FP, Solak M, Petekkaya I, Keskin O, Kertmen N, Sarici F. **Association between common risk factors and molecular subtypes in breast cancer patients**. *Breast.* (2013.0) **22**. DOI: 10.1016/J.BREAST.2012.08.005
31. Sanderson M, Pal T, Beeghly-Fadiel A, Fadden MK, Dujon SA, Clinton C. **A pooled case-only analysis of reproductive risk factors and breast cancer subtype among black women in the southeastern united states**. *Cancer Epidemiol Biomarkers Prev* (2021.0) **30**. DOI: 10.1158/1055-9965.EPI-20-1784
32. Fortner RT, Sisti J, Chai B, Collins LC, Rosner B, Hankinson SE. **Parity, breastfeeding, and breast cancer risk by hormone receptor status and molecular phenotype: results from the nurses’ health studies**. *Breast Cancer Res* (2019.0) **21**. DOI: 10.1186/S13058-019-1119-Y
33. Ambrosone CB, Zirpoli G, Ruszczyk M, Shankar J, Hong CC, McIlwain D. **Parity and breastfeeding among African-American women: differential effects on breast cancer risk by estrogen receptor status in the women’s circle of health study**. *Cancer Causes Control.* (2014.0) **25**. DOI: 10.1007/S10552-013-0323-9
34. Palmer JR, Viscidi E, Troester MA, Hong CC, Schedin P, Bethea TN. **Parity, lactation, and breast cancer subtypes in African American women: results from the AMBER consortium**. *J Natl Cancer Inst* (2014.0) **106**. DOI: 10.1093/JNCI/DJU237
35. Millikan RC, Newman B, Tse CK, Moorman PG, Conway K, Smith LV. **Epidemiology of basal-like breast cancer**. *Breast Cancer Res Treat* (2008.0) **109**. DOI: 10.1007/S10549-007-9632-6
36. Gaudet MM, Press MF, Haile RW, Lynch CF, Glaser SL, Schildkraut J. **Risk factors by molecular subtypes of breast cancer across a population-based study of women 56 years or younger**. *Breast Cancer Res Treat* (2011.0) **130**. DOI: 10.1007/S10549-011-1616-X
37. Shinde SS, Forman MR, Kuerer HM, Yan K, Peintinger F, Hunt KK. **Higher parity and shorter breastfeeding duration: association with triple-negative phenotype of breast cancer**. *Cancer.* (2010.0) **116**. DOI: 10.1002/CNCR.25443
38. Ellingjord-Dale M, Vos L, Hjerkind KV, Hjartåker A, Russnes HG, Tretli S. **Alcohol, physical activity, smoking, and breast cancer subtypes in a Large, nested case-control study from the Norwegian breast cancer screening program**. *Cancer Epidemiol Biomarkers Prev* (2017.0) **26**. DOI: 10.1158/1055-9965.EPI-17-0611
39. García-Estévez L, Cortés J, Pérez S, Calvo I, Gallegos I, Moreno-Bueno G. **Obesity and breast cancer: A paradoxical and controversial relationship influenced by menopausal status**. *Front Oncol* (2021.0) **11**. DOI: 10.3389/fonc.2021.705911
40. Schoemaker MJ, Nichols HB, Wright LB, Brook MN, Jones ME, O’Brien KM. **Association of body mass index and age with subsequent breast cancer risk in premenopausal women**. *JAMA Oncol* (2018.0) **4**. DOI: 10.1001/JAMAONCOL.2018.1771
41. Bandera EV, Chandran U, Hong CC, Troester MA, Bethea TN, Adams-Campbell LL. **Obesity, body fat distribution, and risk of breast cancer subtypes in African American women participating in the AMBER consortium**. *Breast Cancer Res Treat* (2015.0) **150**. DOI: 10.1007/S10549-015-3353-Z
42. Gravena AAF, Lopes TCR, Demitto M de O, Borghesan DHP, Dell’ Agnolo CM, Brischiliari SCR. **The obesity and the risk of breast cancer among pre and postmenopausal women**. *Asian Pac J Cancer Prev* (2018.0) **19**. DOI: 10.22034/APJCP.2018.19.9.2429
43. Gathirua-Mwangi WG, Palmer JR, Champion V, Castro-Webb N, Stokes AC, Adams-Campbell L. **Maximum and time-dependent body mass index and breast cancer incidence among postmenopausal women in the black women’s health study**. *Am J Epidemiol.* (2022.0) **191**. DOI: 10.1093/AJE/KWAC004
44. Petri AL, Tjønneland A, Gamborg M, Johansen D, Høidrup S, Sørensen TIA. **Alcohol intake, type of beverage, and risk of breast cancer in pre- and postmenopausal women**. *Alcohol Clin Exp Res* (2004.0) **28**. DOI: 10.1097/01.ALC.0000130812.85638.E1
45. Ángeles-Llerenas A, Ortega-Olvera C, Pérez-Rodríguez E, Esparza-Cano JP, Lazcano-Ponce E, Romieu I. **Moderate physical activity and breast cancer risk: the effect of menopausal status**. *Cancer Causes Control.* (2010.0) **21**. DOI: 10.1007/S10552-009-9487-8
46. Si S, Boyle T, Heyworth J, Glass DC, Saunders C, Fritschi L. **Lifetime physical activity and risk of breast cancer in pre-and post-menopausal women**. *Breast Cancer Res Treat* (2015.0) **152**. DOI: 10.1007/S10549-015-3489-X
47. Unar-Munguía M, Torres-Mejía G, Colchero MA, González De Cosío T. **Breastfeeding mode and risk of breast cancer: A dose-response meta-analysis**. *J Hum Lact.* (2017.0) **33**. DOI: 10.1177/0890334416683676
48. Nickels S, Truong T, Hein R, Stevens K, Buck K, Behrens S. **Evidence of gene-environment interactions between common breast cancer susceptibility loci and established environmental risk factors**. *PloS Genet* (2013.0) **9**. DOI: 10.1371/JOURNAL.PGEN.1003284
49. Rudolph A, Chang-Claude J, Schmidt MK. **Gene-environment interaction and risk of breast cancer**. *Br J Cancer.* (2016.0) **114**. DOI: 10.1038/BJC.2015.439
50. Dierssen-Sotos T, Palazuelos-Calderón C, Jiménez-Moleón JJ, Aragonés N, Altzibar JM, Castaño-Vinyals G. **Reproductive risk factors in breast cancer and genetic hormonal pathways: a gene-environment interaction in the MCC-Spain project**. *BMC Cancer.* (2018.0) **18**. DOI: 10.1186/S12885-018-4182-3
51. Olivier M, Bouaoun L, Villar S, Robitaille A, Cahais V, Heguy A. **Molecular features of premenopausal breast cancers in Latin American women: Pilot results from the PRECAMA study**. *PloS One* (2019.0) **14**. DOI: 10.1371/JOURNAL.PONE.0210372
52. de Almeida LM, Cortés S, Vilensky M, Valenzuela O, Cortes-Sanabria L, de Souza M. **Socioeconomic, clinical, and molecular features of breast cancer influence overall survival of Latin American women**. *Front Oncol* (2022.0) **0**. DOI: 10.3389/fonc.2022.845527
53. Carvalho FM, Bacchi LM, Pincerato KM, Van de Rijn M, Bacchi CE. **Geographic differences in the distribution of molecular subtypes of breast cancer in Brazil**. *BMC Womens Health* (2014.0) **14**. DOI: 10.1186/1472-6874-14-102
54. Cazap E, Buzaid AC, Garbino C, de la Garza J, Orlandi FJ, Schwartsmann G. **Breast cancer in Latin America: results of the Latin American and Caribbean society of medical Oncology/Breast cancer research foundation expert survey**. *Cancer.* (2008.0) **113**. DOI: 10.1002/CNCR.23834
55. Homburger JR, Moreno-Estrada A, Gignoux CR, Nelson D, Sanchez E, Ortiz-Tello P. **Genomic insights into the ancestry and demographic history of south America**. *PloS Genet* (2015.0) **11**. DOI: 10.1371/journal.pgen.1005602
56. Adhikari K, Chacón-Duque JC, Mendoza-Revilla J, Fuentes-Guajardo M, Ruiz-Linares A. **The genetic diversity of the americas**. *Annu Rev Genomics Hum Genet* (2017.0) **18**. DOI: 10.1146/annurev-genom-083115-022331
57. Norris ET, Wang L, Conley AB, Rishishwar L, Mariño-Ramírez L, Valderrama-Aguirre A. **Genetic ancestry, admixture and health determinants in Latin America**. *BMC Genomics* (2018.0) **19** 861. DOI: 10.1186/s12864-018-5195-7
58. Heinz T, Álvarez-Iglesias V, Pardo-Seco J, Taboada-Echalar P, Gómez-Carballa A, Torres-Balanza A. **Ancestry analysis reveals a predominant native American component with moderate European admixture in bolivians**. *Forensic Sci Int Genet* (2013.0) **7**. DOI: 10.1016/j.fsigen.2013.05.012
59. Fejerman L, Stern MC, Ziv E, John EM, Torres-Mejia G, Hines LM. **Genetic ancestry modifies the association between genetic risk variants and breast cancer risk among Hispanic and non-Hispanic white women**. *Carcinogenesis.* (2013.0) **34**. DOI: 10.1093/carcin/bgt110
60. Fejerman L, Stern MC, John EM, Torres-Mejía G, Hines LM, Wolff RK. **Interaction between common breast cancer susceptibility variants, genetic ancestry, and nongenetic risk factors in Hispanic women**. *Cancer Epidemiol Biomarkers Prev* (2015.0) **24**. DOI: 10.1158/1055-9965.EPI-15-0392
61. Fejerman L, Ramirez AG, Nápoles AM, Gomez SL, Stern MC. **Cancer epidemiology in Hispanic populations: what have we learned and where do we need to make progress**. *Cancer Epidemiol Biomarkers Prev* (2022.0) **31**. DOI: 10.1158/1055-9965.EPI-21-1303
62. Marker KM, Zavala VA, Vidaurre T, Lott PC, Vásquez JN, Casavilca-Zambrano S. **Human epidermal growth factor receptor 2–positive breast cancer is associated with indigenous American ancestry in Latin American women**. *Cancer Res* (2020.0) **80**. DOI: 10.1158/0008-5472.CAN-19-3659
63. Zavala VA, Casavilca-Zambrano S, Navarro-Vásquez J, Castañeda CA, Valencia G, Morante; Z. **Association between ancestry-specific 6q25 variants and breast cancer subtypes in Peruvian women**. *Cancer Epidemiol Biomarkers Prev* (2022.0) **31**. DOI: 10.1158/1055-9965.EPI-22-0069
64. Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MAR, Bender D. **PLINK: A tool set for whole-genome association and population-based linkage analyses**. *Am J Hum Genet* (2007.0) **81**. DOI: 10.1086/519795
65. Campbell CL, Scheller C, Horn H, Kidd JM, Doddapaneni H. **A global reference for human genetic variation**. *Nature.* (2015.0) **526** 68-74. DOI: 10.1038/nature15393
66. Alexander DH, Novembre J, Lange K. **Fast model-based estimation of ancestry in unrelated individuals**. *Genome Res* (2009.0) **19**. DOI: 10.1101/gr.094052.109.vidual
67. Team RC. *R: A language and environment for statistical computing* (2018.0)
68. 68
Instituto Nacional de Estadística e Informática. Censos nacionales 2017: XII de población, VII de vivienda y III de comunidades indígenas. (2018).. *Censos nacionales 2017: XII de población, VII de vivienda y III de comunidades indígenas* (2018.0)
69. Torres-Roman JS, Martinez-Herrera JF, Carioli G, Ybaseta-Medina J, Valcarcel B, Pinto JA. **Breast cancer mortality trends in Peruvian women**. *BMC Cancer.* (2020.0) **20** 1-9. DOI: 10.1186/S12885-020-07671-X/FIGURES/4
70. De-La-Cruz-Ku G, Luyo M, Morante Z, Enriquez D, Möller MG, Chambergo-Michilot D. **Triple-negative breast cancer in Peru: 2000 patients and 15 years of experience**. *PloS One* (2020.0) **15**. DOI: 10.1371/JOURNAL.PONE.0237811
71. Machado MPA, Opaleye DC, Pereira TV, Padilla I, Noto AR, Prince M. **Alcohol and tobacco consumption concordance and its correlates in older couples in Latin America**. *Geriatr Gerontol Int* (2017.0) **17**. DOI: 10.1111/GGI.12974
72. Champagne BM, Sebrié EM, Schargrodsky H, Pramparo P, Boissonnet C, Wilson E. **Tobacco smoking in seven Latin American cities: the CARMELA study**. *Tob Control.* (2010.0) **19**. DOI: 10.1136/TC.2009.031666
73. Shieh Y, Fejerman L, Lott PC, Marker K, Sawyer SD, Hu D. **A polygenic risk score for breast cancer in U.S. latinas and Latin-American women**. *J Natl Cancer Inst* (2019.0) **112**. DOI: 10.1093/jnci/djz174
74. Slattery ML, Lundgreen A, Hines L, Wolff RK, Torres-Mejia G, Baumgartner KN. **Energy homeostasis genes and breast cancer risk: The influence of ancestry, body size, and menopausal status, the breast cancer health disparities study**. *Cancer Epidemiol.* (2015.0) **39**. DOI: 10.1016/j.canep.2015.08.012
75. Fejerman L, John EM, Huntsman S, Beckman K, Choudhry S, Perez-Stable E. **Genetic ancestry and risk of breast cancer among U.S. latinas**. *Cancer Res* (2008.0) **68**. DOI: 10.1158/0008-5472.CAN-08-2039
76. Engmann NJ, Ergas IJ, Yao S, Kwan ML, Roh JM, Ambrosone CB. **Genetic ancestry is not associated with breast cancer recurrence or survival in U.S. latina women enrolled in the kaiser permanente pathways study**. *Cancer Epidemiol Biomarkers Prev* (2017.0) **26**. DOI: 10.1158/1055-9965.EPI-17-0148
77. Sanchez SS, Tachachartvanich P, Stanczyk FZ, Gomez SL, John EM, Smith MT. **Estrogenic activity, race/ethnicity, and indigenous American ancestry among San Francisco bay area women**. *PloS One* (2019.0) **14**. DOI: 10.1371/JOURNAL.PONE.0213809
78. John EM, Sangaramoorthy M, Hines LM, Stern MC, Baumgartner KB, Giuliano AR. **Overall and abdominal adiposity and premenopausal breast cancer risk among hispanic women: the breast cancer health disparities study**. *Cancer Epidemiol Biomarkers Prev* (2015.0) **24**. DOI: 10.1158/1055-9965.EPI-13-1007-T
79. Slattery ML, Lundgreen A, John EM, Torres-Mejia G, Hines L, Giuliano AR. **MAPK genes interact with diet and lifestyle factors to alter risk of breast cancer: the breast cancer health disparities study**. *Nutr Cancer.* (2015.0) **67** 292-304. DOI: 10.1080/01635581.2015.990568
80. Vallejos C, Gómez H, Cruz W, Pinto J, Dyer R, Velarde R. **Breast cancer classification according to immunohistochemistry markers: Subtypes and association with clinicopathologic variables in a peruvian hospital database**. *Clin Breast Cancer.* (2010.0) **10** 294-300. DOI: 10.3816/CBC.2010.n.038
81. Macari A, Soberanis-Pina P, Varela-Santoyo E, Valle-Sanchez MA, Leal-Hidalgo JL, Torres-Guillen VM. **Prevalence and molecular profile of breast carcinoma using immunohistochemistry markers in Mexican women**. *World J Oncol* (2021.0) **12**. DOI: 10.14740/WJON1392
82. Gómez R, Ossa CA, Montoya ME, Echeverri C, Ángel G, Ascuntar J. **Impact of immunohistochemistry-based molecular subtype on chemosensitivity and survival in Hispanic breast cancer patients following neoadjuvant chemotherapy**. *Ecancermedicalscience.* (2015.0) **9**. DOI: 10.3332/ECANCER.2015.562
83. Romero-Cordoba SL, Salido-Guadarrama I, Rebollar-Vega R, Bautista-Piña V, Dominguez-Reyes C, Tenorio-Torres A. **Comprehensive omic characterization of breast cancer in Mexican-Hispanic women**. *Nat Commun* (2021.0) **12**. DOI: 10.1038/S41467-021-22478-5
84. Serrano-Gomez SJ, Sanabria-Salas MC, Hernández-Suarez G, García O, Silva C, Romero A. **High prevalence of luminal b breast cancer intrinsic subtype in Colombian women**. *Carcinogenesis.* (2016.0) **37**. DOI: 10.1093/CARCIN/BGW043
85. Fernandes GC, Michelli RA, Galvão HC, Paula AE, Pereira R, Andrade CE. **Prevalence of**. *Oncotarget.* (2016.0) **7**. DOI: 10.18632/oncotarget.12610
86. Fejerman L, Chen GK, Eng C, Huntsman S, Hu D, Williams A. **Admixture mapping identifies a locus on 6q25 associated with breast cancer risk in US latinas**. *Hum Mol Genet* (2012.0) **21**. DOI: 10.1093/hmg/ddr617
87. Hoffman J, Fejerman L, Hu D, Huntsman S, Li M, John EM. **Identification of novel common breast cancer risk variants at the 6q25 locus among latinas**. *Breast Cancer Res* (2019.0) **21**. DOI: 10.1186/s13058-018-1085-9
88. Fejerman L, Sanchez SS, Thomas R, Tachachartvanich P, Riby J, Gomez SL. **Association of lifestyle and demographic factors with estrogenic and glucocorticogenic activity in Mexican American women**. *Carcinogenesis.* (2016.0) **37**. DOI: 10.1093/CARCIN/BGW074
89. Bonilla C, Bertoni B, Hidalgo PC, Artagaveytia N, Ackermann E, Barreto I. **Breast cancer risk and genetic ancestry: a case-control study in Uruguay**. *BMC Womens Health* (2015.0) **15**. DOI: 10.1186/s12905-015-0171-8
90. **A century of trends in adult human height**. *Elife* (2016.0) **5**. DOI: 10.7554/ELIFE.13410
91. Ruiz-Linares A, Adhikari K, Acuña-Alonzo V, Quinto-Sanchez M, Jaramillo C, Arias W. **Admixture in Latin America: Geographic structure, phenotypic diversity and self-perception of ancestry based on 7,342 individuals**. *PloS Genet* (2014.0) **10**. DOI: 10.1371/journal.pgen.1004572
92. Barrios YV, Sanchez SE, Nicolaidis C, Garcia PJ, Gelaye B, Zhong Q. **Childhood abuse and early menarche among Peruvian women**. *J Adolesc Health* (2015.0) **56** 197. DOI: 10.1016/J.JADOHEALTH.2014.10.002
93. 93
UNICEF. The state of the world’s children. (2021).. *The state of the world’s children* (2021.0)
94. Huayanay-Espinoza CA, Quispe R, Poterico JA, Carrillo-Larco RM, Bazo-Alvarez JC, Miranda JJ. **Parity and Overweight/Obesity in Peruvian women**. *Prev Chronic Dis* (2017.0) **14**. DOI: 10.5888/PCD14.160282
95. Yábar A, Meléndez R, Muñoz S, Deneo H, Freire J, Domínguez V. **Effect of ki-67 assessment in the distribution of breast cancer subtypes: Evaluation in a cohort of Latin American patients**. *Mol Clin Oncol* (2017.0) **6**. DOI: 10.3892/MCO.2017.1185
96. Goldhirsch A, Winer EP, Coates AS, Gelber RD, Piccart-Gebhart M, Thürlimann B. **Personalizing the treatment of women with early breast cancer: highlights of the St gallen international expert consensus on the primary therapy of early breast cancer 2013**. *Ann Oncol Off J Eur Soc Med Oncol* (2013.0) **24**. DOI: 10.1093/annonc/mdt303
97. Guth AA, Chun Kim J, Schwartz S, Montes J, Snyder RA, Axelrod D. **The relationship of race, oncotype DX, and Ki67 in a population highly screened for breast cancer**. *Breast J* (2017.0) **23**. DOI: 10.1111/TBJ.12781
98. Cai S, Zuo W, Lu X, Gou Z, Zhou Y, Liu P. **The prognostic impact of age at diagnosis upon breast cancer of different immunohistochemical subtypes: A surveillance, epidemiology, and end results (SEER) population-based analysis**. *Front Oncol* (2020.0) **10**. DOI: 10.3389/FONC.2020.01729
99. Clarke CA, Keegan THM, Yang J, Press DJ, Kurian AW, Patel AH. **Age-specific incidence of breast cancer subtypes: understanding the black-white crossover**. *J Natl Cancer Inst* (2012.0) **104**. DOI: 10.1093/JNCI/DJS264
100. Butler EN, Tse CK, Bell ME, Conway K, Olshan AF, Troester MA. **Active smoking and risk of luminal and basal-like breast cancer subtypes in the Carolina breast cancer study**. *Cancer Causes Control.* (2016.0) **27** 775. DOI: 10.1007/S10552-016-0754-1
101. Kawai M, Malone KE, Tang MTC, Li CI. **Active smoking and the risk of estrogen receptor-positive and triple-negative breast cancer among women ages 20 to 44 years**. *Cancer.* (2014.0) **120**. DOI: 10.1002/CNCR.28402
102. Santiago-Torres M, De Dieu Tapsoba J, Kratz M, Lampe JW, Breymeyer KL, Levy L. **Genetic ancestry in relation to the metabolic response to a US versus traditional Mexican diet: a randomized crossover feeding trial among women of Mexican descent**. *Eur J Clin Nutr* (2017.0) **71** 395-401. DOI: 10.1038/EJCN.2016.211
103. Ziv E, John EM, Choudhry S, Kho J, Lorizio W, Perez-Stable EJ. **Genetic ancestry and risk factors for breast cancer among latinas in the San Francisco bay area**. *Cancer Epidemiol Biomarkers Prev* (2006.0) **15**. DOI: 10.1158/1055-9965.EPI-06-0092
104. Chang-Cabanillas S, Peñafiel-Sam J, Alarcón-Guevara S, Pereyra-Elías R. **Social determinants of mammography screening among women aged 50 to 59, Peru 2015**. *Health Care Women Int* (2021.0) **42** 1-15. DOI: 10.1080/07399332.2020.1786093
105. Hernández-Vásquez A, Chacón-Torrico H. **Use of mammography in Peruvian women: An analysis of the 2018 demographic and health survey**. *Medwave.* (2019.0) **19**. DOI: 10.5867/MEDWAVE.2019.09.7701
106. Vidaurre T, Santos C, Gómez H, Sarria G, Amorin E, López M. **The implementation of the plan esperanza and response to the imPACT review**. *Lancet Oncol* (2017.0) **18**. DOI: 10.1016/S1470-2045(17)30598-3
107. 107
Research on the menopause in the 1990s. Report of a WHO Scientific Group. World Health Organ Tech Rep Ser. (1996) 886:1–107.. *Research on the menopause in the 1990s* (1996.0) **886**
108. Phipps AI, Ichikawa L, Bowles EJA, Carney PA, Kerlikowske K, Miglioretti DL. **Defining menopausal status in epidemiologic studies: A comparison of multiple approaches and their effects on breast cancer rates**. *Maturitas.* (2010.0) **67**. DOI: 10.1016/J.MATURITAS.2010.04.015
109. Chollet-Hinton L, Anders CK, Tse CK, Bell MB, Yang YC, Carey LA. **Breast cancer biologic and etiologic heterogeneity by young age and menopausal status in the Carolina breast cancer study: A case-control study**. *Breast Cancer Res* (2016.0) **18** 1-10. DOI: 10.1186/S13058-016-0736-Y/TABLES/3
|
---
title: Computer-aided classification of indirect immunofluorescence patterns on esophagus
and split skin for the detection of autoimmune dermatoses
authors:
- Jens Hocke
- Jens Krauth
- Christopher Krause
- Stefan Gerlach
- Nicole Warnemünde
- Kai Affeldt
- Nina van Beek
- Enno Schmidt
- Jörn Voigt
journal: Frontiers in Immunology
year: 2023
pmcid: PMC10013071
doi: 10.3389/fimmu.2023.1111172
license: CC BY 4.0
---
# Computer-aided classification of indirect immunofluorescence patterns on esophagus and split skin for the detection of autoimmune dermatoses
## Abstract
Autoimmune bullous dermatoses (AIBD) are rare diseases that affect human skin and mucous membranes. Clinically, they are characterized by blister formation and/or erosions. Depending on the structures involved and the depth of blister formation, they are grouped into pemphigus diseases, pemphigoid diseases, and dermatitis herpetiformis. Classification of AIBD into their sub-entities is crucial to guide treatment decisions. One of the most sensitive screening methods for initial differentiation of AIBD is the indirect immunofluorescence (IIF) microscopy on tissue sections of monkey esophagus and primate salt-split skin, which are used to detect disease-specific autoantibodies. Interpretation of IIF patterns requires a detailed examination of the image by trained professionals automating this process is a challenging task with these highly complex tissue substrates, but offers the great advantage of an objective result. Here, we present computer-aided classification of esophagus and salt-split skin IIF images. We show how deep networks can be adapted to the specifics and challenges of IIF image analysis by incorporating segmentation of relevant regions into the prediction process, and demonstrate their high accuracy. Using this semi-automatic extension can reduce the workload of professionals when reading tissue sections in IIF testing. Furthermore, these results on highly complex tissue sections show that further integration of semi-automated workflows into the daily workflow of diagnostic laboratories is promising.
## Introduction
Autoimmune bullous diseases (AIBD) are a highly heterogenous group of autoantibody-driven diseases in which autoantibodies against various proteins of the desmosomes and basement membrane zone (BMZ) of the skin and surface epithelia (1–5). Depending on the antigens involved, blister formation either occurs intra-epidermally in pemphigus diseases or sub-epidermally in pemphigoid diseases. Deciphering the affected structures in this heterogeneous group of disorders is essential for both prognosis and treatment, as immunosuppressive therapy varies according to disease entity. Initial differentiation can be achieved by indirect immunofluorescence (IIF) microscopy on monkey esophagus for pemphigus diseases and primate salt-split skin for pemphigoid diseases as the most sensitive screening methods for autoimmune bullous diseases [6, 7]. IIF on monkey esophagus can detect circulating autoantibodies against the epithelial and endomysial autoantigens. IIF on primate salt-split skin discriminates autoantibodies against the BMZ. Additional testing with, for example, recombinant desmoglein 1 (DSG1), DSG3, BP180 (type XVII collagen), BP230, laminin 332, type VII collagen, and deamidated gliadin peptides by IIF, enzyme-linked immunosorbent assay (ELISA), or immunoblot analysis (2–4, 8) can be performed to identify target antigens.
Previously, BIOCHIP® mosaics have been developed that allow the simultaneous detection of different autoantibody specificities on a routine laboratory slide by placing multiple miniature so-called biochips in a single incubation field. In addition to recombinant proteins such as BP180 NC16A and cells recombinantly expressing DSG1, DSG3, BP230, type VII collagen, and laminin 332, tissue substrates such as monkey esophagus and primate salt-split skin can also be used (6, 9–11). The BIOCHIP® technology is increasingly used for routine diagnostics of autoimmune blistering diseases (AIBD) (12–22).
Circulating autoantibodies in pemphigus characteristically bind to the epithelium of esophagus in an intercellular pattern (Figure 1A). The pattern is seen as smooth linear fluorescence along the borders of the epithelial cells with a mesh-like appearance. The presented algorithm maps this pattern as ‘Intercellular’. Anti-BMZ reactivity in pemphigoid diseases, e.g. directed against BP180, BP230 or type VII collagen can also be visualized in this substrate (Figure 1B) revealing to a smooth linear fluorescence along the BMZ. This pattern is subsequently referred to as ‘BMZ’ pattern. Furthermore, in celiac disease and dermatitis herpetiformis, IgA autoantibodies label the endomysium in a characteristic pattern [11] which appears as a honeycombed pattern within the lamina muscularis mucosae. Here, we focused on ‘Intercellular’ and ‘BMZ’ pattern and as such endomysial pattern is not reported. In the diagnosis of AIBDs, other histopathological regions of the esophagus are not relevant.
**Figure 1:** *Exemplary immunofluorescence images of the incubated substrates esophagus and salt-split skin. The ‘intercellular’ pattern seen by intercellular labelling of monkey esophagus in pemphigus vulgaris/foliaceus is shown in (A). In (B), the basal membrane zone (‘BMZ’) pattern typical for pemphigoid diseases is indicated. The patterns found in salt-split skin are ‘epidermal’ (C) and ‘dermal’ (D). They arise by binding of autoantibodies in pemphigoid patients to the epidermal or dermal side of the artificial split.*
Separating dermis and epidermis of primate skin with 1M NaCl results in split formation within the lamina lucida, which allows differentiation between two linear IF patterns in pemphigoid diseases. Antibodies against BP180, BP230, and α6β4 integrin stain along the epidermal side of the artificial split (Figure 1C). For the computer-generated outputs, we refer to this binding as ‘epidermal’. In contrast, autoantibodies against the p200 antigen, laminin γ1, laminin 332, and type VII collagen bind along the dermal side [23], a pattern that is referred to as “dermal” in this manuscript (Figure 1D). In areas where the tissue is still connected, linear fluorescence is seen.
Reading IIF patterns is done by visual evaluation of microscope images and is not standardized. Hence, the results strongly depend on the practical experience of the professional and are subjective. The high variance in the appearance of tissue segments complicates the evaluation leading to an undesirably high variability of the results. Computer-aided evaluation of IIF became popular both because of the automated workflow and more standardized objective evaluation of IIF (24–26). Computer-aided diagnostics rapidly evolved as deep learning became a state-of-the-art method for computer vision tasks. Deep learning enables the solution of complex tasks such as image segmentation [27], object detection [28, 29], and image classification [30]. In the medical field, deep learning has been successfully used for the detection of skin cancer [31], lung nodules [32], and diabetic retinopathy [33].
Here, we show how deep learning can be used for computer-aided evaluation of IIF images of highly complex tissue substrates. The images are from slides containing millimeter-sized fragments of coated cover glasses with biological substrates called biochips (EUROIMMUN Medizinische Labordiagnostika AG, Lübeck, Germany). The primary esophagus and primate salt-split skin each cover large areas of the biochip, while only small structures on specific parts of the tissue are decisive for the classification result. Standard deep networks are not suitable for processing these images due to limitations in computer memory and the number of images available for training. We describe two architectures, both of which use segmentation to aid the training process of a classification network. We demonstrate their effectiveness on real data acquired with a EUROPattern® Microscope (EPM) 1.5 microscope (EUROIMMUN) by comparing their outputs with the results of visual evaluation of the same images by a professional.
In the present study, we would also commemorate Detlef Zillikens, director and chair of the Department of Dermatology, University of Lübeck, Lübeck, Germany. Detlef passed away in September 2022 after short and severe disease. He has been one of the leading clinician scientists in the field of AIBD who authored more than 600 articles on this subject. Detlef Zillikens has inspired, motivated, and mentored numerous colleagues including N.v. B. and E.S. E.S. was amongst his first students performing his medical thesis about the presence of inflammatory mediators in sera and blister fluids of bullous pemphigoid patients compared to sera and suction blisters of healthy volunteers in 1993 (34–37). This fruitful and passionate cooperation has been functional for 25 years bringing to light more than 250 articles and book chapters about AIBD. N.v. B. and E.S. are greatly indebted to Detlef Zillikens and thankful for his wise, winning, and kind mentoring and his friendship. All authors are sad and affected about the early death of this exceptional clinician and researcher.
Overall, we present classifiers for the detection and analysis of these IIF patterns. In addition to the classification of circulating antibodies as positive or negative, a brightness score is also returned for titer estimation.
## Human sera
Serum samples from patients were collected for routine diagnostic analysis by the Immunological Laboratory (Lübeck, Germany) and the Department of Dermatology, University of Lübeck. Samples were anonymized for analysis and stored at -20°C until use. Left-over patient samples were used for this study after completion of all diagnostic procedures. Serum samples from healthy blood donors served as controls. The study performed in accordance with the ethical guideline stated in the Declaration of Helsinki and positively evaluated by the ethics committee of the University of Lübeck (12–178).
## Salt-split skin and esophagus dataset
Data were acquired using the EUROPattern® Suite (EUROIMMUN), a system of hardware and software components for computer-aided IIF. Images were acquired using an EPM 1.5 microscope at 10x magnification. To picture an entire biochip, four images are combined. This results in images with slightly varying sizes of about 4,500 x 3,400 pixels.
*To* generate the training dataset, the biochips were incubated at various dilutions from 1:10 to 1:10,000 to ensure distinction of weak patterns for AIBD from negative samples. The salt-split skin training set consisted of 3,428 images, of which 1,040 images showed a positive reaction in epidermal tissue and 326 showed a positive reaction in dermal tissue. 2,076 of these images showed no positive reaction. The esophagus dataset consisted of 7,022 training images with 1,399 images with intercellular reactivity in the epithelium, 2,539 with anti-BMZ reactivity, and 3,166 images with no or unspecific reactions. The dataset also contained images with patterns caused by a variety of confounder antibodies (e.g., antinuclear antibodies), not associated with AIBD resulting in images with multiple patterns.
*To* generate the validation dataset, 110 patient samples were incubated at 1:10, 1:32, and 1:100 dilutions to test for correct assignment including weak specific patterns, resulting in three images of esophagus and salt-split skin per patient. For the 52 samples from healthy controls, only a 1:10 dilution was incubated, resulting in one image of esophagus and salt-split skin per control. For the validation dataset, 26 patients with serologically tested epidermal binding to salt-split skin and 8 patients with serologically tested dermal binding were used. 30 patients with serologically tested bullous pemphigoid showed linear binding at the BMZ on monkey esophagus, and 20 patients with serologically tested pemphigus vulgaris/foliaceus stained monkey epithelium with an intercellular pattern. One patient reacted positively to both epidermal BMZ and epithelial desmosomes. In addition, 52 substrates with no or with unspecific reactivity were used, incubated with sera from control subjects or patients with, for example, pemphigus on salt-split primate skin.
The collected training and validation data were read manually by IF professionals. The readers were medical technicians with at least 5 years of experience in reading IF tests including monkey esophagus and salt-split skin.
## Salt-split skin algorithm
The number of salt-split skin images for the training dataset and the memory required to process images of a given size with a deep network is limited. In addition, generalization to unseen data is difficult to achieve due to the large number of parameters needed in a suitable network. Therefore, we chose to first identify relevant subregions of the image and then analyze only those subregions. This approach addresses both memory requirements and generalization. The algorithm includes multiple steps of image processing (Figure 2). First, the image is segmented to identify the relevant subregions of the tissue. Based on the segmented regions, lines of attention are computed. Image patches are sampled along these lines and then processed individually by a deep neural network. The individual outputs are aggregated to obtain the final result.
**Figure 2:** *Overview of the steps for processing salt-split skin images. Segmentation is used to identify relevant parts of the image. Samples of these subregions are then analyzed via deep networks and the results are aggregated. This approach ensures low memory requirements and good generalization.*
Only the green channel is used for segmentation. The image is downscaled to a resolution of 512 x 512 pixels and then fed into a U-Net type deep network for segmentation. This type of network is known to work well in segmenting microscopy images [27]. Here, we used a Fusion-Net, which incorporates recent developments in deep learning and further improves performance [38]. The individual outputs are divided into roof (epidermal), floor (dermal), interspace, and background. Here, the interspace is the empty area between the roof and the floor.
The attention regions are then computed. The image content relevant for analysis is located along these attention lines. They are computed from the overlap of the segmented areas expanded by dilation (DIL) (Figure 2) by a disc of 3 pixels in diameter. The extended roof (R), floor (F) and interspace (I) subregions are interpreted as sets in the following. The lines L on the epidermal (LRoof) and dermal (LFloor) substrate are computed as *To focus* on the relevant image content, image patches of size 64 x 64 pixels are extracted along the attention lines. They are taken from a downscaled version of the original image with a size of 2,048 x 2,048 pixels. To achieve a uniform distribution of the patches along the lines, Poisson disc sampling is used [39]. For each line, 40 patches are extracted. The following inference steps are performed separately for roof and floor.
Each patch is then processed separately by a deep neural network to obtain local results (Figure 3). This is a very small network with only a few layers, so only a few parameters need to be optimized. The network has two outputs: a probability for a label and a brightness score for roof or floor. The label values are ‘positive’, ‘negative’, ‘background’, or ‘unclear’. The ‘unclear’ label is needed because not all sections along the attention lines provide the information needed for classification. The ‘background’ label was only used during training to indicate empty patches.
**Figure 3:** *For processing the samples of the subregions of the salt-split skin substrate, the displayed deep network architecture is used. After processing by convolution module shown below, the processing is divided into two branches of dense layers for the different outputs.*
Finally, the labels of the patches are aggregated to obtain a result for the entire attention line and thus for the entire image. Every patch xi is processed by a deep network f. For every index i∈N the network returns ‘negative’ as the label with the highest probability. Accordingly, i∈P if ‘positive’ is the label with highest probability. The other labels ‘background’ and ‘unclear’ are ignored during aggregation. Therefore, these regions do not influence the result. The result y∈[0,1] is given by The brightness b∈ℝ is obtained from the mean of the three samples P 3 with the highest probability value for the label ‘positive’: These brightness scores are then combined for images of different dilutions to estimate the titer.
## Training of the salt-split skin neural networks
Both the segmentation network and the network for processing patches were optimized using adaptive moment estimation (ADAM) [40]. For the segmentation network, a generalized dice loss [41] was used due to its robustness to imbalanced region sizes. The patch analysis network has two outputs, containing classification and regression output. Cross entropy was used to derive the classification output, and mean squared error was used to derive the regression output. Both were weighted equally. Separate analysis networks were trained for roof and floor to achieve a better adaption to the pattern features. In total, there are three networks: one for segmentation, one for the analysis of the roof, and one for the analysis of the floor.
Several augmentations were applied to improve the generalization of these networks. When training the segmentation network, flip and rotation were applied to the entire image. For the analysis networks, the used augmentations flip and Gaussian noise were applied only to the patches.
As mentioned earlier, when training the analysis networks, empty patches are labeled as ‘background’ for technical reasons. It is easier to train the networks with a fixed number of patches per image, but since the size of the salt-split skin segment varies, the number of extracted patches also varies. Therefore, a fixed, generous number of patches is assumed for each image, and missing patches are inserted as empty ones.
## Esophagus algorithms
For the diagnosis of AIBDs, the detection of circulating autoantibodies against epidermal/epithelial antigens is essential. Monkey esophagus is the most sensitive tissue substrate for serum autoantibodies in pemphigus vulgaris/foliaceus. In pemphigoid diseases, autoantibodies label the BMZ of monkey esophagus, but with a lower sensitivity compared with primate salt-split skin (15, 42–45). The algorithm for classification of the immunofluorescence (IF) patterns on esophagus tissue is divided into two subtasks.
The first task is to localize and segment the tissue segments in the image. An adapted version of the U-Net model [27] is trained on an esophagus dataset containing image pairs of IF images and segmentation ground truth images. The segmentation images consist of seven segments representing the esophagus tissue sections longitudinal muscle, circular muscle, muscular mucous membrane, BMZ, lamina propria, epithelium, and background. The trained neural network can process unseen IF images of the esophagus and assigns one of the aforementioned sections to each pixel of the image. The key information of the segmentation result is the spatial information of the epidermal BMZ that could be used in the following subtask.
The second task is the classification of the IF image into the classes ‘BMZ’, ‘intercellular’ and ‘negative’. One approach to achieve this goal is a whole-image classification neural network that is trained on a dataset of esophagus IF images and corresponding target labels, assigning a probability between 0 and 1 for each class to each image. However, the neural network had low accuracy in predicting the ‘BMZ’ class. Presumably, the sparse information of the very thin epidermal BMZ compared with the large number of other pixels occurring in the image and the high variance of tissue morphologies prevented the neural network from focusing on the crucial image regions and features. To bypass this disadvantage, the spatial information of the epidermal BMZ processed in the first step can be utilized. The classification neural network is designed to receive two input images: The first input is the original IF image, and the second is a post-processed mask based on the segmentation result. The post-processing drops all segmentation information except the epidermal BMZ and converts the remaining image to a binary image. The binary image is used as attention mechanism input and is intended to hint the neural network to pay special attention to the masked region (Figure 4). This approach dramatically increased the accuracy in our classification experiments.
**Figure 4:** *Steps for processing esophagus images. The first processing path segments the image with a convolutional neural network (CNN) into six tissue segments plus background and extracts the epidermal basement membrane as input to the classification path. The segmentation mask is post-processed to calculate the desmosome region. The spatial information is used for intensity estimation of found pattern.*
At last, not only the classification with a probability between 0 and 1 for each class, but also a quantification of pattern intensity, which represents the titer of the autoantibodies in patient serum, is valuable information for the evaluation of IIF images, in particular, if more than one pattern is detected. Hence, the algorithm extracts the intensity of the pattern in the regions where the pattern occurs and predicts its titer. For a positive ‘BMZ’ pattern, the intensity extraction is straightforward, when the segmentation information of the esophagus is given. However, for a positive ‘intercellular’ pattern, more post-processing is required to find the relevant region, since the desmosome region in the epithelium is in close proximity to the BMZ and the segmentation contains the epithelium as a whole region. Therefore, post-processing with convolution operations is calculated on the segmentation image to predict the region adjacent to the BMZ that extends into the epithelium (Figure 1A). In the case of a positive reaction, the ‘intercellular’ staining appears as the brightest part in that region and can be extracted using a previously defined quantile of the occurring pixel intensities.
## Training of the esophagus neural networks
The segmentation neural network for esophagus tissue images was trained on preprocessed and downscaled images of size 512 x 512 pixels. This ensures rapid and memory-optimized segmentation of the images, with resolution high enough to distinguish the different tissue regions that appear. In addition, the images were augmented by horizontal and vertical flipping, random rotation and zooming, resulting in a more generalized network. The segmentation network has seven outputs corresponding to the tissue segments longitudinal muscle, circular muscle, muscular mucous membrane, epidermal basement membrane, lamina propria, epithelium, and background. The network was optimized using the ADAM optimizer and categorical cross-entropy.
The classification neural network was trained on preprocessed images of size 2,048 x 2,048, preserving all the crucial information available in positive reactions. For each image, the previously trained segmentation network processes a segmentation map on the fly, extracting the BMZ, converting it to a binary image and upscales it to the size of 2,048 x 2,048 pixels, building an image input pair of original tissue image and binary map of the epidermal BMZ. The image pairs were also augmented by horizontal and vertical flipping, random rotation and zooming to increase the variance of the training dataset. The classification network also used the ADAM optimizer. The employed optimization function was binary cross-entropy.
## Titer estimation
The titer value is computed for the positive IF patterns of each patient to estimate the amount of serum autoantibodies. Titer values are taken from the dilution series(1:10,1:32,1:100,1:320,…). An image of dilution dk has the kth dilution from that series. For a single image, the brightness score s∈[1,5]of a detected pattern is converted to the titer by addition: dk + s -1. If multiple images are available, the lowest-dilution image with a negative pattern is defining the total dilution. If there is no image with a negative pattern, the highest-dilution image is taken, and the total dilution is computed in the same way as the single image.
## Results
The esophagus and salt-split skin classification algorithms were used to process the samples of the validation dataset. Therefore, each individual image was processed. For sera with several dilutions, each individual image was classified. The identified pattern and intensities were used to compile a patient-based result including the predicted titer. For sera with only one dilution, the result depended solely on the result of that image.
## Salt-split skin classification results
The trained salt-split skin models were evaluated using the validation dataset. One patient had to be discarded from this analysis due to a defect in the substrate image. Thus, the validation dataset consisted of 109 patient sera for analysis (Table 1). A pattern detection accuracy of $96\%$ was achieved for the salt-split skin ‘epidermal binding’ algorithm. Both positive percent agreement (PPA) and negative percent agreement (NPA) were $96\%$. Here, the results of one patient serum with a titer of 1:10 gave a barely positive IF result and were incorrectly classified as negative. The salt-split skin ‘dermal binding” algorithm achieved $97\%$ accuracy, with $100\%$ PPA and $97\%$ NPA. Of note, only eight patients showed dermal binding. The classification errors are exemplified in Figure 5. These errors occurred mostly in borderline cases where luminance was low and sometimes unspecific. When only ‘binding’ was evaluated, regardless of epidermal or dermal localization, binding was detected with $95\%$ accuracy, with $97\%$ PPA and $95\%$ NPA. Titer estimates were almost all in the +/-1 range for both patterns.
## Esophagus classification results
The algorithm detected 46 of 49 positive patterns. The algorithm classified no patient serum as false positive and 61 patient samples as true negatives. This resulted in an overall agreement of $97\%$, a PPA of $94\%$ and a $100\%$ NPA (Table 2). Of note, 3 sera that were classified as false negative by the algorithm were identified by visual classification as marginally positive with a titer of 1:10.
**Table 2**
| Unnamed: 0 | Unnamed: 1 | EUROPattern® visual mode | EUROPattern® visual mode.1 | EUROPattern® visual mode.2 | EUROPattern® visual mode.3 | EUROPattern® visual mode.4 | EUROPattern® visual mode.5 | EUROPattern® visual mode.6 | EUROPattern® visual mode.7 | EUROPattern® visual mode.8 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| | | Combined | Combined | Combined | ‘BMZ’ | ‘BMZ’ | ‘BMZ’ | ‘Intercellular’ | ‘Intercellular’ | ‘Intercellular’ |
| EPa Classifier | | positive | negative | total | found | not found | total | found | not found | total |
| EPa Classifier | positive | 46 | 0 | 46 | 27 | 2 | 29 | 20 | 0 | 23 |
| EPa Classifier | negative | 3 | 61 | 64 | 3 | 78 | 81 | 0 | 90 | 87 |
| EPa Classifier | total | 49 | 61 | 110 | 30 | 80 | 110 | 20 | 90 | 110 |
| | | | Accuracy: | 97.3% | | Accuracy: | 95.5% | | Accuracy: | 100.0% |
| | | | PPA: | 93.9% | | PPA: | 90.0% | | PPA: | 100.0% |
| | | | NPA: | 100.0% | | NPA: | 97.5% | | NPA: | 100.0% |
Regarding the pattern recognition for the ‘BMZ’ pattern, 27 of 30 patient sera were classified by the algorithm as true positive reactions, three patient sera were classified as false negative (the three marginally positive sera, example in Figure 6A), and two patient sera were classified with a positive pattern, but instead of a ‘BMZ’ the algorithm predicted an ‘intercellular’ pattern (an example is shown in Figure 6B). The algorithm predicted 78 patients as true negative samples to BMZ and two patients as false positive samples. The overall agreement for the pattern is $96\%$ compared to the manual reading with a PPA of $90\%$ and NPA of $98\%$. The agreement of titer prediction for BMZ, calculated only for the true positive samples, agreed in $70\%$ of patients. $30\%$ of the predictions differed by +/-1 titer step. There were no deviations greater than this (Figure 7).
**Figure 6:** *Classification errors on esophagus with faint staining. (A) ‘BMZ’ pattern incorrectly classified as ‘negative’. (B) ‘Intercellular’ pattern incorrectly classified by the algorithm as ‘BMZ’ in a patient with pemphigoid disease.* **Figure 7:** *Titer deviation of each pattern compared to the evaluation by an experienced reader of indirect immunofluorescence images. Titer deviation was only calculated in samples where the algorithm and conventional reading detected the same pattern. The reading by experienced personnel was used as a reference. A negative deviation indicates an underestimation of the titer steps, a positive deviation an overestimation by the algorithm.*
For the ‘intercellular’ pattern, all 20 patient sera were correctly identified by the algorithm and all 90 sera without this pattern were correctly classified as negative pattern. Thus, the overall agreement for the ‘intercellular’ pattern was $100\%$ with a PPA of $100\%$ and NPA of $100\%$. The titer prediction for the pattern had a $60\%$ match rate for the true positive patient samples. $40\%$ of the titer predictions had an error of +/- 1 titer step. As with the ‘BMZ’ pattern, no deviations greater than this were detected (Figure 7).
## Discussion
The diagnostics of AIBD is based on the detection of circulating and tissue-bound autoantibodies, with the latter remaining the gold standard. However, detection of circulating autoantibodies allows further differentiation of disease entities, which is relevant for both prognosis and choice of treatment. IIF testing on monkey esophagus and salt-split skin is commonly used as screening test and provides initial insight into likely disease entities [46]. Delayed testing for AIBD and delayed treatment can lead to adverse outcomes with permanent impairment or even fatality [47, 48]. While monkey esophagus epithelium has the highest sensitivity for autoantibodies against desmosomal proteins, autoantibodies against BMZ proteins can also be detected (5, 48–50). For the detection of BMZ structures, sensitivity is higher when salt-split skin is used as substrate, which has the advantage that epidermal-binding can also be distinguished from dermal-binding AIBD types (5, 50–52). Analysis of IIF on these substrates requires experienced readers. In this study, automated reading of IIF on primate esophagus and salt-split skin was evaluated. The presented algorithms for classification of both monkey esophagus and salt-split skin tissue sections for the detection of autoantibodies specific for AIBD showed a high accuracy with over $95\%$ agreement compared to the results of conventional reading by an IIF professional. The PPA was above $97\%$ for all positive IF patterns, on both the esophagus and salt-split skin. The NPA was at least $95\%$ for all patterns. The results showed a slightly higher agreement for blister floor staining on salt-split skin and for intercellular staining on monkey esophagus, whereas for blister roof staining on salt-split skin and BMZ staining on monkey esophagus, some samples with low staining intensity were not detected and some samples with unspecific staining were classified as positives.
The confounder autoantibodies described previously were present on both positive and negative samples of the training dataset. The deep learning algorithm was trained to assign a negative label to the patterns caused by the confounder antibodies on images if only confounder patterns are present. When both a cofounder pattern and an AIBD-relevant pattern, i.e. “BMZ”, “intercellular”, “epidermal”, or “dermal” were present, the confounder pattern was suppressed and only the AIBD-relevant pattern indicated. Therefore, the algorithm learned that the patterns caused by confounder antibodies are irrelevant for the desired outcome and can be ignored.
Both algorithms can assist laboratory staff in the challenging task of evaluating IF patterns on tissues. The classifiers are an excellent extension of the screening methods for AIBD. Automation of IIF evaluation has already been successfully used in rheumatology for the detection of antinuclear antibodies. Several commercially available systems can be used here [reviewed in [53]]. In the latter review, the authors conclude that higher standardization of results is achieved by less subjectivity and less influence by expertise and that automated workflows are more efficient due to higher throughput. We think standardization of reading IF patterns, especially tissues can be prone to have a high variance between the readings of different personnel. Computer-aided classification, therefore offers a second opinion in a deterministic manner. However, the limitation of our algorithm lies in the design of a pattern proposal system that only can give hints on learnt patterns. For rare patterns that were not contained in the training dataset, the algorithm will likely fail with its proposal and the IF professional must act independently. Also, our study revealed some discrepancies between the algorithm and the reading of the IF professionals. In these cases, the IF professional must take the final decision. Obviously, the benefit of having a high-throughput systems only takes effect in laboratories with a high number of sera subjected to IIF on esophagus and salt-split skin.
Software solutions for computer-aided diagnostics of IIF are already available for other substrates, using the EUROPattern® Microscope (EUROIMMUN), which was also used in the present study [54]. The EUROPattern® Suite software is capable of classifying images of HEp-2/HEp-2010 cells for the detection of anti-nuclear antibodies (ANA) [25, 55], anti-neutrophil cytoplasmatic antibodies (ANCA) [56], anti-mitochcondrial antibody (AMA), anti-Epstein- *Barr virus* (EBV) antibodies, anti-dsDNA antibodies using Crithidia luciliae-based IIF [26], recombinant cell-based assays for multiparametric serological testing in autoimmune encephalitis, e.g., with recombinant cells [54], and rat liver and kidney for detection of reactivity against liver-kidney microsomes (LKM). In the present study, the main challenge was that the small structures relevant for classification were present only on certain parts of the large area covered by the tissue on the biochip. This prevented the use of standard deep networks. We applied segmentation to focus the attention of the classification networks to the crucial regions. For salt-split skin substrate, unspecific and low luminance borderline cases led to the most difficulty because there is no clear decision boundary. The segmentation of esophagus tissue was particularly challenging due to its naturally high variance in tissue layout. Specifically, segmentation and thus detection of the BMZ, which is very thin compared to other sections of the tissue, is a challenging task. Furthermore, classification of the BMZ is complex because it lacks specific structural features. Besides the challenging algorithmic processing, collecting a sufficient amount of training data containing hand-labeled ground truth masks for the segmentation neural network is a tedious and time-consuming manual task.
Our results show that deep learning as a method for computer-aided diagnostics on microscopy images has evolved into a state-of-the-art method besides traditional computer vision. Segmentation and classification of neural networks showed good results in this work. Even very complex microscopy images with tissue layers that are difficult for professionals to evaluate can reliably be segmented and classified via computer-aided algorithms.
The present study has several limitations. Only a limited number of sera with salt-split skin dermal binding pattern were applied. Therefore, the accuracy for this pattern is of lower significance. A larger study focusing on the clinical application of the classifiers will certainly improve this aspect and is anticipated. Also, the algorithm only helps to find the expressing patterns for IgG autoantibodies in pemphigus and pemphigoid diseases. Patterns due to IgA autoantibodies are not expected to greatly differ between the ones by IgG reactivity but have not been formally employed in the present study. Currently, there is also no automated assignment of the identified patterns to a specific AIBD. Additional algorithms for advanced diagnostics of AIBD using BIOCHIP® mosaic-based IIF, including classifiers for recombinant BP180 as well as for cell-based IIF with recombinant BP230, DSG1, and DSG3 classifiers, are under development to detect autoantibodies specific for these disorders. The great advantage is that the results of IF image evaluation will be automatically proposed for verification and approval by laboratory personnel. In conclusion, the presented classifiers and algorithms allow the semi-automated assessment of autoantibody binding on monkey esophagus and primate salt-split skin in routine diagnostics of pemphigus and pemphigoid diseases. This innovation will further improve and facilitate the diagnosis of these rare autoimmune disorders.
## Data availability statement
The datasets for this article are not publicly available due to concerns regarding participant/patient anonymity. Requests to access the datasets should be directed to the corresponding author.
## Ethics statement
The studies involving human participants were reviewed and approved by Ethics committee of the University of Lübeck. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.
## Author contributions
JK and JH contributed equally to this work and share first authorship. All authors contributed to the article and approved the submitted version.
## Conflict of interest
JV, JH, SG, JK, CK, NW, and KA are employees of EUROIMMUN Medizinische Labordiagnostika AG. ES has a scientific cooperation with EUROIMMUN. JV, JH, SG, JK, CK, MH, ES, and NB are inventors of the patent application number 22199236 that has been filed by EUROIMMUN Medizinische Labordiagnostika AG on 30th Sept 2022. This patent application is pending and covers the algorithmic processing of the salt-split skin substrate.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Bieber K, Kridin K, Emtenani S, Boch K, Schmidt E, Ludwig RJ. **Milestones in personalized medicine in pemphigus and pemphigoid**. *Front Immunol* (2021) **11**. DOI: 10.3389/fimmu.2020.591971
2. Schmidt E, Zillikens D. **Pemphigoid diseases**. *Lancet* (2013) **381**. DOI: 10.1016/S0140-6736(12)61140-4
3. Schmidt E, Kasperkiewicz M, Joly P. **Pemphigus**. *Lancet* (2019) **394**. DOI: 10.1016/S0140-6736(19)31778-7
4. Kasperkiewicz M, Ellebrecht CT, Takahashi H, Yamagami J, Zillikens D, Payne AS. **Pemphigus**. *Nat Rev Dis Primers* (2017) **3** 1-18. DOI: 10.1038/nrdp.2017.26
5. van Beek N, Zillikens D, Schmidt E. **Autoimmune bullous dermatoses**. *Deutsches Ärzteblatt Int* (2021) **118**. DOI: 10.3238/arztebl.m2021.0136
6. Beek NV, Rentzsch K, Probst C, Komorowski L, Kasperkiewicz M, Fechner K. **Serological diagnosis of autoimmune bullous skin diseases: Prospective comparison of the biochip mosaic-based indirect immunofluorescence technique with the conventional multi-step single test strategy**. *Orphanet J Rare Dis* (2012) **7** 49. DOI: 10.1186/1750-1172-7-49
7. van Beek N, Zillikens D, Schmidt E. **Diagnosis of autoimmune bullous diseases**. *JDDG: J der Deutschen Dermatologischen Gesellschaft* (2018) **16**. DOI: 10.1111/ddg.13637
8. Amber KT, Murrell DF, Schmidt E, Joly P, Borradori L. **Autoimmune subepidermal bullous diseases of the skin and mucosae: Clinical features, diagnosis, and management**. *Clin Rev Allergy Immunol* (2018) **54** 26-51. DOI: 10.1007/s12016-017-8633-4
9. Goletz S, Probst C, Komorowski L, Schlumberger W, Fechner K, van Beek N. **A sensitive and specific assay for the serological diagnosis of antilaminin 332 mucous membrane pemphigoid**. *Br J Dermatol* (2019) **180**. DOI: 10.1111/bjd.17202
10. Komorowski L, Müller R, Vorobyev A, Probst C, Recke A, Jonkman MF. **Sensitive and specific assays for routine serological diagnosis of epidermolysis bullosa acquisita**. *J Am Acad Dermatol* (2013) **68**. DOI: 10.1016/j.jaad.2011.12.032
11. van Beek N, Krüger S, Fuhrmann T, Lemcke S, Goletz S, Probst C. **Multicenter prospective study on multivariant diagnostics of autoimmune bullous dermatoses using the biochip technology**. *J Am Acad Dermatol* (2020) **83**. DOI: 10.1016/j.jaad.2020.01.049
12. Goletz S, Giurdanella F, Holtsche MM, Nijenhuis M, Horvath B, Diercks GF. **Comparison of two diagnostic assays for anti-laminin 332 mucous membrane pemphigoid**. *Front Immunol* (2021) **12**. DOI: 10.3389/fimmu.2021.773720
13. Yilmaz K, Goletz S, Pas HH, Bos RRVD, Blauvelt A, White WL. **Clinical and serological characterization of orf-induced immunobullous disease**. *JAMA Dermatol* (2022) **158**. DOI: 10.1001/jamadermatol.2022.0290
14. Holtsche MM, Beek NV, Hashimoto T, Zenzo GD, Zillikens D, Prost-Squarcioni C. **Diagnosis of epidermolysis bullosa acquisita: Multicentre comparison of different assays for serum anti-type vii collagen reactivity**. *Acta Dermato-Venereologica* (2021) **101** adv00420-adv00420. DOI: 10.2340/00015555-3774
15. Schmidt E, Rashid H, Marzano A, Lamberts A, Zenzo GD, Diercks G. **European Guidelines (s3) on diagnosis and management of mucous membrane pemphigoid, initiated by the european academy of dermatology and venereology - part ii**. *J Eur Acad Dermatol Venereology* (2021) **35**. DOI: 10.1111/jdv.17395
16. Wang M, Li F, Wang X, Wang R, Chen T, Zhao J. **Biochip mosaic for the diagnosis of autoimmune bullous diseases in chinese patients**. *Eur J Dermatol* (2020) **30**. DOI: 10.1684/ejd.2020.3839
17. Tirumalae R, Kalegowda IY. **Role of biochip indirect immunofluorescence test in cutaneous vesiculobullous diseases**. *Am J Dermatopathol* (2020) **42**. DOI: 10.1097/DAD.0000000000001471
18. Yang A, Xuan R, Melbourne W, Tran K, Murrell D. **Validation of the biochip test for the diagnosis of bullous pemphigoid, pemphigus vulgaris and pemphigus foliaceus**. *J Eur Acad Dermatol Venereology* (2020) **34**. DOI: 10.1111/jdv.15770
19. Yang A, Xuan R, Melbourne W, Hashimoto T, Uzun S, Daneshpazhooh M. **Inter-rater reliability of the biochip indirect immunofluorescence dermatology mosaic in bullous pemphigoid and pemphigus patients**. *J Eur Acad Dermatol Venereology* (2019) **33**. DOI: 10.1111/jdv.15817
20. Adaszewska A, Kalińska-Bienias A, Jagielski P, Woźniak K, Kowalewski C. **The use of biochip mosaics in diagnostics of bullous pemphigoid: Evaluation and comparison to conventional multistep procedures**. *J Cutaneous Pathol* (2020) **47**. DOI: 10.1111/cup.13591
21. Simpson K, Scardamaglia L, Kok Y, Vu M, Kidd D, Yap T. **Comparison of the euroimmun dermatology profile elisa to the novel biochip mosaic 7 for the diagnosis of immunobullous skin disease**. *Australas J Dermatol* (2021) **62**. DOI: 10.1111/ajd.13611
22. Arunprasath P, Rai R, Venkataswamy C. **Comparative analysis of biochip mosaic-based indirect immunofluorescence with direct immunofluorescence in diagnosis of autoimmune bullous diseases: A cross-sectional study**. *Indian Dermatol Online J* (2020) **11** 915. DOI: 10.4103/idoj.idoj_156_20
23. Schmidt E, Zillikens D. **Modern diagnosis of autoimmune blistering skin diseases**. *Autoimmun Rev* (2010) **10**. DOI: 10.1016/j.autrev.2010.08.007
24. Bizzaro N, Antico A, Platzgummer S, Tonutti E, Bassetti D, Pesente F. **Automated antinuclear immunofluorescence antibody screening: A comparative study of six computer-aided diagnostic systems**. *Autoimmun Rev* (2014) **13**. DOI: 10.1016/j.autrev.2013.10.015
25. Voigt J, Krause C, Rohwäder E, Saschenbrecker S, Hahn M, Danckwardt M. **Automated indirect immunofluorescence evaluation of antinuclear autoantibodies on hep-2 cells**. *Clin Dev Immunol* (2012) **2012**. DOI: 10.1155/2012/651058
26. Hormann W, Hahn M, Gerlach S, Hochstrate N, Affeldt K, Giesen J. **Performance analysis of automated evaluation of crithidia luciliae-based indirect immunofluorescence tests in a routine setting - strengths and weaknesses**. *Clin Chem Lab Med* (2017) **56** 86-93. DOI: 10.1515/cclm-2017-0326
27. Ronneberger O, Fischer P, Brox T, Navab N., Hornegger J., Wells W., Frangi A.. **U-Net: Convolutional networks for biomedical image segmentation**. *Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015. MICCAI 2015* (2015) 9351. DOI: 10.1007/978-3-319-24574-4_28
28. Girshick R, Donahue J, Darrell T, Malik J. **Rich feature hierarchies for accurate object detection and semantic segmentation**. (2014). DOI: 10.1109/CVPR.2014.81
29. Redmon J, Divvala S, Girshick R, Farhadi A. **You only look once: Unified, real-time object detection**. (2016). DOI: 10.1109/CVPR.2016.91
30. LeCun Y, Bengio Y, Hinton G. **Deep learning**. *Nature* (2015) **521**. DOI: 10.1038/nature14539
31. Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM. **Dermatologist-level classification of skin cancer with deep neural networks**. *Nature* (2017) **542**. DOI: 10.1038/nature21056
32. Setio AAA, Ciompi F, Litjens G, Gerke P, Jacobs C, Riel SJV. **Pulmonary nodule detection in ct images: False positive reduction using multi-view convolutional networks**. *IEEE Trans Med Imaging* (2016) **35**. DOI: 10.1109/TMI.2016.2536809
33. Gulshan V, Peng L, Coram M, Stumpe MC, Wu D, Narayanaswamy A. **Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs**. *JAMA - J Am Med Assoc* (2016) **316**. DOI: 10.1001/jama.2016.17216
34. Schmidt E, Bröcker EB, Zillikens D. **High levels of soluble cd23 in blister fluid of patients with bullous pemphigoid**. *Arch Dermatol* (1995) **131**. DOI: 10.1001/archderm.1995.01690200106030
35. Schmidt E, Mittnacht A, Schömig H, Dummer R, Bröcker EB, Zillikens D. **Detection of il-1α, il-1β and il-1 receptor antagonist in blister fluid of bullous pemphigoid**. *J Dermatol Sci* (1996) **11**. DOI: 10.1016/0923-1811(95)00435-1
36. Schmidt E, Bastian B, Dummer R, Tony HP, Bröcker EB, Zillikens D. **Detection of elevated levels of il-4, il-6, and il-10 in blister fluid of bullous pemphigoid**. *Arch Dermatol Res* (1996) **288**. DOI: 10.1007/BF02507102
37. Schmidt E, Ambach A, Bastian B, Brocker EB, Zillikens D. **Elevated levels of interleukin-8 in blister fluid of bullous pemphigoid compared with suction blisters of healthy control subjects**. *J Am Acad Dermatol* (1996) **34**. DOI: 10.1016/S0190-9622(96)80146-0
38. Quan TM, Hildebrand DGC, Jeong WK. **Fusionnet: A deep fully residual convolutional neural network for image segmentation in connectomics**. *Front Comput Sci* (2021) **3**. DOI: 10.3389/fcomp.2021.613981
39. Cook RL. **Stochastic sampling in computer graphics**. *ACM Trans Graphics (TOG)* (1986) **5** 51-72. DOI: 10.1145/7529.8927
40. Kingma DP, Ba J. **Adam: A method for stochastic optimization**. *arXiv* (2014). DOI: 10.48550/arxiv.1412.6980
41. Sudre CH, Li W, Vercauteren T, Ourselin S, Cardoso MJ. **Generalised dice overlap as a deep learning loss function for highly unbalanced segmentations. In: Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support: Third International Workshop, DLMIA and 7th International Workshop, ML-CDS 2017, Held in Conjunction with MICCAI 2017**. (2017)
42. Joly P, Horvath B, Patsatsi A, Uzun S, Bech R, Beissert S. **Updated s2k guidelines on the management of pemphigus vulgaris and foliaceus initiated by the european academy of dermatology and venereology (eadv)**. *J Eur Acad Dermatol Venereology* (2020) **34**. DOI: 10.1111/jdv.16752
43. Prost-Squarcioni C, Caux F, Schmidt E, Jonkman M, Vassileva S, Kim S. **International bullous diseases group: consensus on diagnostic criteria for epidermolysis bullosa acquisita**. *Br J Dermatol* (2018) **179** 30-41. DOI: 10.1111/bjd.16138
44. Schmidt E, Goebeler M, Hertl M, Sardy M, Sitaru C, Eming R. **S2k guideline for the diagnosis of pemphigus vulgaris/foliaceus and bullous pemphigoid**. *JDDG - J German Soc Dermatol* (2015) **13**. DOI: 10.1111/ddg.12612
45. Borradori L, Beek NV, Feliciani C, Tedbirt B, Antiga E, Bergman R. **Updated s2 k guidelines for the management of bullous pemphigoid initiated by the european academy of dermatology and venereology (eadv)**. *J Eur Acad Dermatol Venereology* (2022) **36**. DOI: 10.1111/jdv.18220
46. Sárdy M, Kostaki D, Varga R, Peris K, Ruzicka T. **Comparative study of direct and indirect immunofluorescence and of bullous pemphigoid 180 and 230 enzyme-linked immunosorbent assays for diagnosis of bullous pemphigoid**. *J Am Acad Dermatol* (2013) **69**. DOI: 10.1016/j.jaad.2013.07.009
47. Syring FM, Zillikens D, Schmidt E, Kasperkiewicz M. **Anti-laminin 332 mucous membrane pemphigoid with irreversible ocular and tracheobronchial involvement: Delayed diagnosis of a severe autoimmune bullous disease**. *Hautarzt* (2019) **70**. DOI: 10.1007/s00105-019-4402-z
48. van Beek N, Schumacher N, Rose C, Schmidt E, Zillikens D. **Moderne diagnostik bullöser autoimmundermatosen**. *Der Pathologe* (2020) **41**. DOI: 10.1007/s00292-020-00795-8
49. Diercks GF, HH P, Jonkman MF. **Immunofluorescence of autoimmune bullous diseases**. *Surg Pathol Clinics* (2017) **10**. DOI: 10.1016/j.path.2017.01.011
50. Schmidt E, della Torre R, Borradori L. **Clinical features and practical diagnosis of bullous pemphigoid**. *Immunol Allergy Clinics North America* (2012) **32**. DOI: 10.1016/j.iac.2012.04.002
51. Lau I, Goletz S, Holtsche MM, Zillikens D, Fechner K, Schmidt E. **Anti-p200 pemphigoid is the most common pemphigoid disease with serum antibodies against the dermal side by indirect immunofluorescence microscopy on human salt-split skin**. *J Am Acad Dermatol* (2019) **81**. DOI: 10.1016/j.jaad.2019.03.077
52. Zenzo GD, della Torre R, Zambruno G, Borradori L. **Bullous pemphigoid: From the clinic to the bench**. *Clinics Dermatol* (2012) **30** 3-16. DOI: 10.1016/j.clindermatol.2011.03.005
53. Tozzoli R, Antico A, Porcelli B, Bassetti D. **Automation in indirect immunofluorescence testing: A new step in the evolution of the autoimmunology laboratory**. *Autoimmun Highlights* (2012) **3** 59-65. DOI: 10.1007/s13317-012-0035-2
54. Fraune J, Gerlach S, Rentzsch K, Teegen B, Lederer S, Affeldt K. **Multiparametric serological testing in autoimmune encephalitis using computer-aided immunofluorescence microscopy (caifm)**. *Autoimmun Rev* (2016) **15**. DOI: 10.1016/j.autrev.2016.07.024
55. Krause C, Ens K, Fechner K, Voigt J, Fraune J, Rohwäder E. **Europattern suite technology for computer-aided immunofluorescence microscopy in autoantibody diagnostics**. *Lupus* (2015) **24**. DOI: 10.1177/0961203314559635
56. Hochstrate N, Krause C, Ens K, Danckwardt M, Feirer C, Rohwaeder E. **Ana and anca diagnostics with computer-aided immunofluorescence microscopy (caifm)**. *Zeitschrift für Rheumatologie* (2013) **72** 24
|
---
title: 'Exploring Intergenerational Communication on Social Media Group Chats as a
Cancer Prevention Intervention Opportunity Among Vietnamese American Families: Qualitative
Study'
journal: JMIR Formative Research
year: 2023
pmcid: PMC10013128
doi: 10.2196/35601
license: CC BY 4.0
---
# Exploring Intergenerational Communication on Social Media Group Chats as a Cancer Prevention Intervention Opportunity Among Vietnamese American Families: Qualitative Study
## Abstract
### Background
Families use social media group chats to connect with each other about daily life and to share information. Although cancer is not a frequent topic of conversation in family settings, the adoption of mobile technology in the family context presents a novel opportunity to promote cancer prevention information. To the best of our knowledge, few studies have used private social media group chats to promote cancer prevention information to family members.
### Objective
In this formative study, we investigated how family group chat platforms can be leveraged to encourage colorectal cancer screening, human papillomavirus vaccination, and cervical cancer screening among intergenerational Vietnamese American families. This study aimed to cocreate a family-based communication intervention for introducing cancer screening information in family group chats. We sought to understand family members’ motivations for using group chats, family dynamics and conversation patterns, and group chat experiences and cultural norms for interacting with family members.
### Methods
Overall, 20 audio-recorded and semistructured interviews were conducted with young Vietnamese adults. The study was conducted between August and October 2018. Participants were Vietnamese Americans; aged between 18 and 44 years; living in Orange County, California; had an existing family group chat; and expressed an interest in becoming family health advocates. Data were analyzed using a framework analysis.
### Results
In total, 13 ($65\%$) of the 20 young adults reported having >1 group chat with their immediate and extended family. Preventive health was not a typical topic of family conversations, but food, family announcements, personal updates, humorous videos or photos, and current events were. Young adults expressed openness to initiating conversations with family members about cancer prevention; however, they also raised concerns that may influence family members’ receptivity to the messages. Themes that could potentially impact family members’ willingness to accept cancer prevention messages included family status and hierarchy, gender dynamics, relational closeness in the family, and source trust and credibility. These considerations may impact whether families will be open to receiving cancer screening information and acting on it. The participants also mentioned practical considerations for intervention and message design, which included the Vietnamese cultural conversation etiquette of hỏi thăm, respect for a physician’s recommendation, prevention versus symptom orientation, the family health advocate’s bilingual capacity, and the busy lives of family members. In response to exemplar messages, participants mentioned that they preferred to personalize template messages to accommodate conversational norms in their family group chats.
### Conclusions
The findings of this study inform the development of a social media intervention for increasing preventive cancer screening in Vietnamese American families.
## Background
The present communication landscape has inevitably shifted from in-person communication to technology-mediated contexts, which are now facilitated through social media messaging apps [1]. In the past decade, social media messaging platforms have been introduced into the family context, which has increased connectivity among family members [2]. This increased technology-mediated connectivity has social implications for family dynamics and communication [3,4].
Early forms of information and communication technologies focused on personal mobile communication, allowing a narrow reach (phone calls and SMS text messaging) between close friends or family on a one-on-one basis and a wide reach (using Twitter, Instagram, or Facebook) with distant acquaintances or strangers [4]. Now, group chat technology engages middle-reach audiences, which include immediate and extended family members [4]. Group chat apps have been studied to better understand family characteristics, personality, social support, frequency of use, and managing caregiving of family members [5-7]. To our knowledge, few studies have sought to understand how private group chats can be leveraged for disseminating health information to family members.
Cancer is the leading cause of death among Vietnamese Americans living in the United States [8,9]. Five-year age-adjusted human papillomavirus (HPV)-related cancers such as cervical cancer continue to be high in Vietnamese women in the United States (9.5 per 100,000) and colorectal cancer in both Vietnamese men and women (47.8 and 30.7 per 100,000, respectively) [8,10]. The high rates are due to a lack of early prevention behavior. Cervical cancer prevention measures include primary prevention measures (HPV vaccine) and secondary prevention measures (Papanicolaou test) [11,12]. Furthermore, for colorectal cancer, secondary prevention measures or early detection tools include colonoscopy and the fecal immunochemical test [11,13].
## Vietnamese Family Structure and Technology Use
Intergenerational communication between adult children and their older adult parents has increased the likelihood of colorectal cancer and hepatitis B screening in older adults [14,15]. However, existing studies lack the inclusion of communication with extended family networks and younger adults in mediated contexts. Vietnamese families value tight-knit structures, often including a complex network of extended family members such as aunts, uncles, cousins, and grandparents [16]. Vietnamese families experienced acculturation after resettling in the United States after the Vietnam war, causing discordance in relationships between the younger and older generations [17]. Despite this, family obligations, involvement, and values are still seen as important among young adult family members [18].
Families are increasingly using social media platforms to communicate with one another [19,20]. Minority groups, particularly Asian Americans, have been early adopters of smartphone technology to connect with family and friends [21]. The use of mediated platforms on smartphone technology presents an opportunity for prevention by introducing health topics into family conversations [22,23]. Smartphone apps such as WeChat, Viber, Facebook Messenger, and Kakao are popular apps used by many Asian American groups because they allow space for connecting on life events and health issues as well [4,19,24]. Research among caregivers of Vietnamese family members with dementia shows that they frequently use social media platforms, affording space for dementia education [19]. Particularly during the COVID-19 pandemic, the use of group chats to communicate about health became even more normalized, as this was a form of communication used to share information and provide social support to family members [25,26]. Using private group chat settings on smartphone-based apps allows smaller social networks such as families to foster more intimate conversations, which is an ideal setting for inserting cancer prevention information.
## Theories Guiding the Study
The Uses and Gratifications Theory (UGT) guided our research efforts to understand the motivations for family group chats [27]. Literature suggests that families tend to lean toward communication privacy in social media contexts; however, recent studies also show that family members use social media to share information with family members and cultivate greater openness [28,29]. UGT focuses on how people use media and their motivations for using specific channels [27]. UGT guided our understanding of why family members use their group chats, the types of preferred messaging platforms, and when they share information with their family members.
Given the family focus of these group chats, we also used the family communication pattern theory to guide our research efforts. The family communication patterns theory recognizes the importance of exploring how family dynamics and relationships impact communication patterns. According to this theory, intergenerational communication patterns among families are described as either conversation oriented or conformity oriented [30]. We sought to understand how family dynamics and communication norms may act as barriers and facilitators to conversations about health in the context of family group chats.
Finally, the principle of cultural grounding also directed our research. Recognizing the important role that culture plays in health, cultural grounding involves grounding the intervention development process and content in the experiences and expressions of the participants [31]. This entails having participants play an active role in cocreating culturally relevant material [31]. The principle of cultural grounding has been applied to intervention design in other contexts for school-based drug prevention programs, immigrant or rural health settings, and clinical trial participation promotion [32,33]. For this study, cultural grounding guided our intervention design by having Vietnamese young adults actively provide feedback on how, when, and which cancer prevention messages should be introduced into group chat conversations.
## Objectives
The purpose of this study was to understand [1] the topics that families talk about and share on their family group chats, [2] family members’ openness to cancer prevention conversations in the group chat context, and [3] how to introduce the topic of cancer prevention into family group chats to normalize the topic as a family conversation and increase its acceptability. We also sought to assess the feasibility of implementing an intervention to initiate conversations regarding colorectal and cervical cancer screening in group chat contexts among Vietnamese families.
## Sampling Method
We conducted 20 semistructured interviews with young adults who self-identified as Vietnamese; were aged between 18 and 44 years; lived in Orange County, California; had an existing family group chat; and expressed an interest in becoming family health advocates (FHAs). The interviews lasted between 30 and 45 minutes and were conducted in person between August and October 2018. Orange County presently houses the largest Vietnamese ethnic enclave in America with approximately 200,000 Vietnamese residents, which comprises >$33\%$ of the Asian community in the county [34]. We used convenience sampling to recruit participants from local churches, youth groups, and students from university departments. The participants completed an interest form and were then contacted to confirm their eligibility.
## Ethics Approval
Institutional Review Board approval (HS# 2018-4454) was obtained from the University of California Irvine Institutional Review Board before the start of the study. Participants who displayed an interest in participating in the study were given the study information sheet to review beforehand. In addition, before the start of each interview, the interviewer verbally reviewed the study details, risks, and benefits with the participants. The participants provided verbal consent to participate in the study. The interviews lasted for approximately 1 hour and were audio recorded for accuracy purposes. The names were replaced with pseudonyms to protect participant identity. Each participant received US $50 as compensation for their time.
## Interview Guide
The interview guide questions focused on [1] family members’ motivations to use group chats, [2] family dynamics and conversation patterns, and [3] group chat experiences and cultural norms for interacting with family members. The questions also explored typical topics of conversations on family group chats, who participated in the group chat, what prompted information sharing in the group chat, and whether health was ever discussed.
The questionnaire focused on the frequency of group chat conversations, openness between children and older family members, and family group chat dynamics. The participants described typical communication patterns among family members on the group chat, their sense of the older adults’ openness to receiving health information from young adults, and Vietnamese family structures or beliefs that may influence the communication dynamics in the chat.
Finally, the young adults provided feedback about existing evidence-based cancer prevention messages adapted from the Centers for Disease Control and Prevention; the American Cancer Society; and the Asian American Network for Cancer Awareness, Research, and Training. They were also asked to provide feedback on the effectiveness of 2 culturally tailored HPV vaccine videos for Vietnamese young adults that are part of a National Cancer Institute evidence-based cancer control program called HPV Vaccine Decision Narratives [35]. The participants ranked preferred messages for their group chats, provided feedback about the messages, and described how they might adapt the messages when sharing them in the group chats. This activity engaged participants in the process of co-designing culturally resonant cancer prevention messages.
## Data Analysis
Data from interviews with FHAs were analyzed using the framework analysis [36]. The data analysis process began with verbatim transcription, followed by data immersion, familiarization of the range of responses, and the development of a thematic framework. Familiarization began during the data collection phase as interviews were transcribed and interviewer memos were reviewed. Using an inductive approach, data were tagged, and descriptive labels were assigned using NVivo Pro 11 (QSR International). This step was followed by a priori deductive coding of sensitizing constructs [37], which described how the participants viewed themselves concerning their family members. Some of the examples included age, kid, close, trust, language, and lack of time. After the primary and secondary coding cycles, data were organized into higher-order themes for the thematic framework. The thematic framework was then developed and categorized into the following themes: current group chat characteristics, cultural or familial barriers, facilitators for introducing cancer prevention messages into the group chat, cultural considerations for message design, and responses to exemplar messages.
Two coders met weekly to discuss the coding process and identify common characteristics and differences between codes to ensure intercoder agreement [37]. The first author was Vietnamese American who offered her perspective on the interpretation of the data for meaningful themes. Through member checking, the second author strengthened the validity of the findings [38]. The purpose of including the first author was to interpret the themes that resonated with the intended audience: Vietnamese American families. Percent agreement (Cohen κ statistic) was not calculated during the coding process [39].
## Demographics
In total, 20 individuals participated in the interviews. The mean age of the participants was 21 (SD 1.2) years. Most ($\frac{17}{20}$, $85\%$) participants identified as women, were enrolled in college, and were US-born second-generation Vietnamese Americans (refer to Table 1 for demographics).
**Table 1**
| Demographics | Demographics.1 | Values |
| --- | --- | --- |
| Age (years), mean (SD) | Age (years), mean (SD) | 21.1 (1.2) |
| Gender, n (%) | Gender, n (%) | Gender, n (%) |
| | Woman | 17 (85) |
| | Man | 3 (15) |
| Immigration status, n (%) | Immigration status, n (%) | Immigration status, n (%) |
| | US born | 16 (80) |
| | Immigrant | 4 (20) |
| Generation, n (%) | Generation, n (%) | Generation, n (%) |
| | First | 2 (10) |
| | 1.5 | 2 (10) |
| | Second | 16 (80) |
| Level of education, n (%) | Level of education, n (%) | Level of education, n (%) |
| | Currently enrolled in college | 18 (90) |
| | College graduate | 2 (10) |
| Vietnamese language proficiency, n (%) | Vietnamese language proficiency, n (%) | Vietnamese language proficiency, n (%) |
| | Limited | 6 (30) |
| | Intermediate | 9 (45) |
| | Advanced | 5 (25) |
| Number of family group chats per person, n (%) | Number of family group chats per person, n (%) | Number of family group chats per person, n (%) |
| | 1 | 7 (35) |
| | 2 | 8 (40) |
| | 3 | 5 (25) |
## Social Media Platforms
Most ($\frac{13}{20}$, $65\%$) participants maintained several group chats with their extended family members. Among those who had multiple group chats, participants often had separate chats with just their cousins, with immediate family (parents and siblings), and with extended family (aunts, uncles, cousins, grandparents, and immediate family). Table 2 provides a list of different platforms used by young adults. The majority ($\frac{14}{20}$, $70\%$) of participants favored using Facebook Messenger because most family members had a Facebook account and because it was the easiest platform to communicate on.
**Table 2**
| Group chat platform | Participant use, n (%) |
| --- | --- |
| Facebook messenger | 14 (70) |
| iMessage | 13 (65) |
| SMS texting app | 4 (20) |
| Facebook group page | 3 (15) |
| Viber | 2 (10) |
| WhatsApp | 1 (5) |
## Conversation Topics
Families used group chats to share announcements; updates on family trips; graphic interchange formats or short, animated photos; and humorous videos related to common experiences (Table 3). Neither health nor cancer prevention was a typical topic of conversation in family group chats.
**Table 3**
| Conversation topic | Participants who mentioned the topic, n (%) |
| --- | --- |
| Family events and announcements (eg, planning family gatherings) | 18 (90) |
| Sharing news articles (eg, local, national, or world news) | 9 (45) |
| Sharing food information (eg, recipes, meals, and grocery sales) | 8 (40) |
| Personal updates (eg, health, whereabouts, school, and accomplishments) | 8 (40) |
| Common family experiences (eg, sharing jokes and sending vacation photos) | 7 (35) |
| Sharing humorous images (eg, memes and GIFsa) | 5 (25) |
| Encouraging and supportive messages (eg, studying for exams) | 2 (10) |
## Overview
Vietnamese American young adults anticipated some level of acceptance of cancer prevention messages among family members. Many recognized smartphone app affordances, such as convenience and maintaining social connections. Using such a platform to communicate information about cancer would be convenient for the family members. Furthermore, because the message would come from young adults embedded in the family group chat, their encouragement might influence family members’ acceptance of messages. Despite these affordances, they voiced concerns that family dynamics should be considered when introducing cancer prevention messages into a family group chat setting. Family member status, family hierarchy, gender dynamics, cultural norms, relational closeness, trust, and credibility were factors thought to influence the acceptance of cancer prevention messages. Table 4 shows a summary of the themes, including receptivity to cancer prevention messages and practical considerations for message design.
**Table 4**
| Theme | Theme.1 | Theme description |
| --- | --- | --- |
| Receptivity to cancer prevention messages | Receptivity to cancer prevention messages | Receptivity to cancer prevention messages |
| | Family member status and family hierarchy | Hierarchy and rank within the family structure could positively and negatively influence family members’ receptiveness to cancer screening information. |
| | Gender dynamics | Gender dynamics were also discussed as a barrier to discussing gender-specific cancers (eg, cervical cancer) if the opposite gender were present in the group chat. |
| | Cultural norms | Vietnamese family cultural norms play a role in how comfortable family members feel with engaging in conversations about cancer prevention. Some participants perceived discussing cancer prevention as taboo and not culturally acceptable within the family setting. Age was also mentioned as a concern of receptivity (eg, if a younger family member recommended screening to an older family member). |
| | Family relational closeness | FHAsa discussed how relational closeness influences openness to accepting cancer prevention messages in family group chats. Relational closeness was seen as both a potential facilitator and barrier, depending on their perceived closeness with family members. |
| | Source trust and credibility | FHAs mentioned that their family members would trust them as a source of credibility because they have a college education or are actively pursuing a career in the medical field. |
| Practical considerations for intervention and message design | Practical considerations for intervention and message design | Practical considerations for intervention and message design |
| | Cultural conversation etiquette | Several FHAs mentioned the cultural etiquette and conversational norm of hỏi thăm, which is asking generally about one’s overall well-being before any other conversation topic emerges. |
| | Respect for authority: a physician’s recommendation | FHAs acknowledged that the older generation has respect for their physician’s recommendation and opinion, which could both encourage and discourage screening. |
| | Prevention vs symptom orientation | Many participants stated that their family members tended to be more symptom oriented rather than prevention oriented, which presents another challenge for communicating prevention information to family members. Many FHAs mentioned that their families only take action when they feel “something is not right.” |
| | Vietnamese bilingual capacity | All participants (including intermediate and advanced speakers) described language as a barrier to communicating with older family members. They were prepared to use Google Translate and other workarounds (eg, involving parents and siblings) to translate for older family members. |
| | Busy lives and sustaining family cancer prevention conversations | Participants recognized time restrictions and busy schedules as barriers to engaging family members with cancer prevention information. FHAs mentioned that some family members may be more responsive than others given the time restrictions. |
## Family Member Status and Family Hierarchy
Several participants discussed how the family member status of the person introducing cancer prevention messages into family discussions played an important role. Hierarchy within the family structure could potentially influence family members’ receptiveness to cancer screening information. The typical Vietnamese family dynamic is patriarchal and embedded in values such as respect for older adults. Hieu, an 18-year-old young adult man, expressed apprehension about his family potentially being unwilling to listen to him: *For this* participant, not being heard and his opinions not mattering to the older adults was a major concern. Tina, a 23-year-old woman, expressed similar concerns. She said the following: Despite these concerns, young adults were still willing to be FHAs for their family members.
Although family hierarchy was perceived as a barrier to communication by some, it was perceived as a facilitator by others. Michael, a 22-year-old man, said in his response: In Michael’s case, he perceived his age as a determining factor in his relationship with older family members. Other participants expressed that their family structures were not traditional and that their families were open to discussing cancer prevention between the older and younger generations.
## Gender Dynamics
The participants expressed that the gender of the family member introducing cancer prevention was important when its incidence was sex specific, for example, in the case of cervical cancer prevention and Papanicolaou smear screening. Lily, a 21-year-old woman participant, voiced that: Gender dynamics were also discussed as a barrier because of the potential awkwardness of discussing cervical cancer and Papanicolaou screening if the FHA advocating cancer screening was a man. Michael, a young adult who was interviewed, said in his reply: Although it is important to educate both men and women about HPV-related cancers, the gendered nature of certain cancers and screening tests may be difficult to discuss in the context of a mixed-gender family group chat.
## Cultural Norms
Vietnamese family cultural norms play a role in how comfortable family members feel with engaging in conversations about cancer prevention. Some participants perceived discussing cancer prevention as taboo and culturally not acceptable within the family setting. Linh, a 21-year-old woman, said the following: Linh mentioned how cancer is a stigmatized topic of conversation and how age could present challenges for introducing cancer prevention in conversations. In addition to cultural norms, relational closeness in Vietnamese families also plays a role in the acceptance of cancer prevention messages.
## Family Relational Closeness
FHAs discussed how relational closeness may influence openness to accepting cancer prevention messages in family group chats. Relational closeness was perceived as both a potential facilitator and a barrier, depending on relationships with family members. Paula, a 21-year-old woman participant, said: Relational closeness affected whether the participants felt comfortable including their family members in a group chat setting and their hesitancy to initiate a conversation about cancer with these group chat members. Although some participants felt relational closeness was a difficult barrier to overcome, a select few mentioned that they felt confident that their family members would be more receptive to cancer prevention messages because of their perceived closeness within the relationship.
## Source Trust and Credibility
Trust and credibility surfaced as facilitators for the acceptance of cancer prevention messages. Participants mentioned that their family members may trust cancer screening information from them because they are family. For example, Karen, a 21-year-old woman, responded: Although participants mentioned that their role as a family member helps build trust, others also discussed the importance of credibility. Family members with medical or health science training were perceived as more credible. Allison, a 21-year-old participant, said: *In this* case, personal trust in family members and credibility because of subject matter knowledge was important to consider when introducing cancer prevention messages into the group chat.
## Practical Considerations for Intervention and Message Design
The practical considerations for a group chat intervention included considering cultural conversational norms, respect for authority (the validity of a physician’s recommendation), family members being symptom oriented rather than prevention oriented, the necessity of bilingual messages, and the timing of the messages. Table 4 shows the summary of themes found when we asked FHAs to share their thoughts on family members’ receptivity to cancer prevention messages.
## Cultural Conversation Etiquette
Some participants mentioned that it may be abrupt and awkward to introduce cancer prevention messages into family group chats without any pretext, as the topic is not typically discussed. FHAs suggested opening with messages expressing empathy and care for the health of family members instead. For example, Linh said: Other participants offered similar sentiments that starting with overall well-being opens the conversation to introduce cancer topics.
## Respect for Authority: a Physician’s Recommendation
Participants mentioned the importance of receiving cancer screening recommendations from medical authority figures. Respect for doctors’ opinions influences whether family members take suggestions for cancer screening from other family members seriously. One participant, Tammy, said: This theme indicates the need to build credibility for the advocated health behavior to effectively encourage family members to follow-up with cancer screening recommendations.
## Prevention Versus Symptom Orientation
Participants stated that their family members tended to be more symptom oriented rather than prevention oriented, which presents another challenge for communicating prevention information to family members. Kelly, a 19-year-old participant, said: Another participant, Tina, expressed how her family also tended to be symptom oriented. She said: FHAs echoed the sentiment of taking action when “feeling something” or “something isn’t right,” which seemed to be a normal phenomenon in most families.
## Vietnamese Bilingual Capacity
Participants described language as a barrier to communicating with older family members. Even intermediate and advanced speakers anticipated difficulties they might face when translating concepts from English to Vietnamese. Although 14 ($70\%$) out of 20 participants expressed that they could speak and write both English and Vietnamese, there were still concerns about whether they would be able to communicate in Vietnamese with older members. For example, Jessica, a 21-year-old advanced Vietnamese speaker, said: Consequently, several participants mentioned that when designing messages, they needed to have readily translated material available to them.
Participants suggested workarounds for the anticipated language barrier with first-generation family members who spoke less English. Including other family members in the translation of concepts was a strategy offered by young adults. For example, Michael said: FHAs were prepared to work around the problem with a variety of methods, which included using Google Translate, asking for a researcher’s help, or asking other Vietnamese-speaking family members to help facilitate conversation.
## Busy Lives and Sustaining Family Cancer Prevention Conversations
Participants recognized time restrictions and busy schedules as barriers to engaging family members with cancer prevention information. FHAs mentioned that some family members may be more responsive than others given the time restrictions. Recommendations included strategically disseminating social media group chat messages in the evenings. For example, Tracy, a 21-year-old woman participant, said: Other participants described similar predictions that since group chats are synchronous, participants could ignore or delay reading the messages. Long, a 21-year young adult man, said: The short time to have a productive conversation was a recurring theme to consider in the intervention design.
## Responses to Example Messages
The concluding section of the interview asked participants to review existing messages about colorectal cancer and HPV-related cancers from publicly available text or social media messages, infographics, websites, and video examples adapted from the Centers for Disease Control and Prevention; the American Cancer Society; and the Asian American Network for Cancer Awareness, Research, and Training. Participants wanted to be able to tailor messages “in their own voice” because the messages were “too academic” or sounded “too much like an advertisement.” Relevance to family members, such as specific demographics, age, and gender tailoring, were also essential elements to be considered. Participants also expressed their wish for messages to include symptoms, be actionable and shorter. Table 5 shows message ranks by popularity, positive feedback, and constructive criticism to exemplar messages.
**Table 5**
| Message example | Message example.1 | Message example.2 | Positive feedback | Constructive criticism |
| --- | --- | --- | --- | --- |
| Colorectal cancer messages | Colorectal cancer messages | Colorectal cancer messages | Colorectal cancer messages | Colorectal cancer messages |
| | “Did you know Asian Americans are at higher risk for colorectal cancer? Though the best test is the colonoscopy, you can get screened at home using the FITa test. Learn more here.” | Demographic tailoring for Asian AmericansOpen ended and poses a questionNeutral messageFeasible and specific actionsTargets susceptibilityOffers alternative option for screening | Demographic tailoring for Asian AmericansOpen ended and poses a questionNeutral messageFeasible and specific actionsTargets susceptibilityOffers alternative option for screening | Vague language and not as informativeNot a specific Asian subgroupDoes not mention age at screening“Advertising” languageLacking statistics |
| | “Don’t ignore symptoms of colon cancer! If you are experiencing pain in the abdomen, blood in stool, body fatigue, and weight loss, talk to your doctor right away. Click here for more info.” | Emphasizes urgencySymptoms are informative | Emphasizes urgencySymptoms are informative | Off-putting or offensive to older adultsDemanding tone and not genuine |
| | “If you are 50 or older, you need to be screened for colorectal cancer. Even if you feel healthy, make sure to talk to your doctor about getting screened. Click on this link for more information.” | Age tailoredTo the point | Age tailoredTo the point | Not relevant to people aged <50 yearsNot detailed enough |
| | “No matter how old you are, there are ways to prevent colorectal cancer through diet, exercise, and not smoking tobacco. Click on this link for more information.” | Feeling healthy may not mean you are healthyRelevant to both young and older adultsNeutral messageRelevant to smokers in the family | Feeling healthy may not mean you are healthyRelevant to both young and older adultsNeutral messageRelevant to smokers in the family | Not as relevant or targeted“Advertising” languageBroad symptoms that could be for any other disease (easy to ignore)Does not provide new information that is not already known |
| HPVb-related cancer messages | HPVb-related cancer messages | HPVb-related cancer messages | HPVb-related cancer messages | HPVb-related cancer messages |
| | “The rate of cervical cancer among Vietnamese American women is 40% higher than Whites. Cervical cancer can be prevented by getting an HPV vaccine, visiting your doctor for a Papanicolaou test when recommended, and not smoking. Click on this link for information.” | Demographic is relatableComparison statisticsActionable behaviors | Demographic is relatableComparison statisticsActionable behaviors | Needs to be less formal in languageLengthy information |
| | “The HPV vaccination is not only for women! HPV vaccination is recommended for young men and women through age 18-26. Talk to your doctor about getting vaccinated. Click for more info.” | Inclusive of both women and menAddresses misconceptions about the HPV vaccine for womenAge tailored | Inclusive of both women and menAddresses misconceptions about the HPV vaccine for womenAge tailored | Sounds like a PSAc or advertisement |
| | “Pap screening is necessary for cervical prevention even if you’ve already received the HPV vaccine. Pap screening is recommended for women 21 or older every three years. Click here for info.” | Emphasizes receiving screening even if vaccinated | Emphasizes receiving screening even if vaccinated | Not engaging or interestingNot relevant to menNot as personalNeeds more explanation between screening and vaccine |
| | “One of the most important things you can do to help prevent cervical cancer is to have regular screening tests starting at age 21 and repeat as recommended. Click on this link for more info!” | Highlights preventionAge relevance | Highlights preventionAge relevance | Not enough informationNo hookGeneral information and not as impactfulDoes not mention why it is important“Regular screening” does not imply urgency |
## Principal Findings
The purpose of this study was to understand how to effectively engage Vietnamese family members with cancer prevention messages in group chats. The results provide insights into the development of a social media cancer prevention intervention for family contexts. Although the social media literature focuses on how media affects people’s mood and well-being, our study explored family members’ motivations for using media to identify opportunities for entry points of influence for promoting cancer screening [27]. Vietnamese families use their family group chats to stay up-to-date with family members’ daily lives, family gatherings, and food or to provide moral support to others in the family group. Family group chat conversations around the planning of family celebrations potentially provide an entry point of influence to introduce cancer prevention discussions. The timing of discussions at Tết (Vietnamese New Year) may be beneficial for introducing the importance of cancer screenings. The pandemic has brought renewed attention to vaccination and discussion of health in family group chats.
Although prior research has recognized that people use many social media platforms such as Twitter, Facebook, and Instagram [40,41], FHAs mentioned using several family group chats on a single platform. Most FHAs mentioned that they had established several family group chats, segregated by the nuclear family (mom, dad, and siblings), peers (cousins and siblings), and intergenerational extended family (grandparents, aunts, uncles, parents, and cousins). Given this phenomenon, topics such as HPV vaccination should be disseminated in a peer family group chat, whereas topics such as colorectal cancer screening should be promoted in an intergenerational group chat context.
Several social influence strategies to encourage cancer screening in group chat contexts could be applied based on the findings of our study. Segmenting audiences by age and family relationships is one strategy mentioned; however, we may also consider how typical topics of group chat conversations such as food, humorous “memes,” family announcements, or even family health history can function as a social influence entrée to connect and introduce cancer prevention [42,43].
Another strategy may include building information requests off preexisting entry points. For example, using typical group chat activities such as sharing family memories through pictures and reminding family members of cancer screening recommendations to stay healthy may be a way to integrate cancer information. Family group chats also offer the potential to apply a foot-in-the-door social influence strategy [44]. This strategy initially involves making a small request to a family member that is likely to yield a positive response, followed by a cancer screening request (ie, a slightly more demanding request). Communication accommodation theory [45] also suggests the importance of adapting cancer screening messages to personalize and adapt to the cultural context of Vietnamese families’ conversational norms, which may involve the timing of introducing the message, who the message sender is, or how the message is introduced.
Our study lays out a range of innovative approaches to introducing and normalizing the topic of cancer screening within mediated family conversation contexts. Several key factors must be considered for message design including family dynamics, culture, language, and conversational norms for family receptivity to messages. Our results show the importance of aligning interventions to match cultural norms for increasing cancer screening acceptance among Vietnamese people [31,46]. This sheds light on the group chats as a novel strategy for introducing and reinforcing cancer screening among family members. For example, having an “inside family member” vouch for acting on screening recommendations from the doctor has the potential to reinforce cancer screenings in an informal web-based setting [22].
In addition, as part of cultural tailoring, women need to introduce “woman” cancer screening messages (eg, Papanicolaou screening) to other women family members. Ensuring that the young adult family member introducing cancer screening is the elder grandchild or a student in the health field can also increase the likelihood of message acceptance by family members. Young adults also expressed the desire to self-tailor cancer screening messages. Cocreating and adapting messages to familiar formats aligns with the principle of cultural grounding and is more likely to resonate with the target audience [31]. For example, one study found that self-tailored arguments by parents about HPV vaccination were more persuasive than motivational interviewing in a clinical parent intervention [47]. Allowing young adults to self-tailor screening messages for their families and insert their motivations as justification may increase acceptance of receiving messages about cancer prevention.
Finally, young adults expressed interest, openness, and motivation to participate and facilitate an intervention with their family members despite potential challenges. Although expectations by the family include that health information is typically delivered by medical professionals in medical settings, receiving reinforcement messages from trusted family members is equally important [48]. Therefore, receiving cancer prevention messages from trusted family members offers another strategy to reinforce credibility and prevent cancer. Reinforcement messages can be key to moving individuals toward actionable behaviors [49].
## Limitations
Vietnamese young adults from Orange County, California, were interviewed to identify key factors to consider when introducing cancer screening messages as part of family group chats. Data saturation for understanding the influence of family dynamics on group chat conversations may be incomplete because our data reflect the family dynamics of our informants [50]. The data generated in this study reflect the experiences and thoughts of Vietnamese women, who were the majority in this study. The perspectives of Vietnamese men were less represented. Furthermore, the findings represent the perspectives of the young adult FHAs and not the entire family group. Future research should consider the older family members’ thoughts on messaging, accessibility, and acceptability. Finally, this study was conducted before the COVID-19 pandemic, which changed how families discuss health. This study generated important considerations for effectively introducing culturally grounded cancer screening messages into Vietnamese family group chats.
## Conclusions
The results of this study help to understand [1] the feasibility of developing a social media intervention among Vietnamese families and [2] family communication norms and cultural considerations for effective intervention design. Given the increased popularity of social media use for family communication, it is important to continue this line of research to understand how social media platforms and intervention designs can be leveraged to encourage preventive screening. Not only is it important to understand how these platforms are used, but it is also crucial to understand what motivates families to actively participate in group chats. Furthermore, understanding family dynamics, family conversation, and the role of Vietnamese family culture will advance our understanding of how to effectively communicate cancer prevention with Vietnamese families, improve health outcomes, and reduce late-stage cervical and colorectal cancer incidences among Vietnamese Americans.
## References
1. Nouwens M, Griggio C, Mackay W. **"WhatsApp is for family; Messenger is for friends": communication places in app ecosystems**. *Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems* (2017.0). DOI: 10.1145/3025453.3025484
2. Kammerl R, Kramer M. **The changing media environment and its impact on socialization processes in families**. *Stud Commun Sci* (2016.0) **16** 21-7. DOI: 10.1016/j.scoms.2016.04.004
3. Ling R. **The social dynamics of mobile group messaging**. *Annals Int Commun Assoc* (2017.0) **41** 242-9. DOI: 10.1080/23808985.2017.1374199
4. Taipale S, Farinosi M. **The big meaning of small messages: the use of whatsapp in intergenerational family communication**. *Human Aspects of IT for the Aged Population. Acceptance, Communication and Participation* (2018.0)
5. Baldassar L. **De-demonizing distance in mobile family lives: co-presence, care circulation and polymedia as vibrant matter**. *Global Networks* (2016.0) **16** 145-63. DOI: 10.1111/glob.12109
6. Gazit T, Aharony N. **The importance of the Whatsapp family group: an exploratory analysis**. *Aslib J Inform Manag* (2016.0) **68** 174-92. DOI: 10.1108/AJIM-09-2015-0142
7. Meixner B, Lee M, Carter S. **Managing family healthcare with multimedia chat apps: a survey on what is missing**. *Proceedings of the 2nd International Workshop on Multimedia for Personal Health and Health Care* (2017.0). DOI: 10.1145/3132635.3132645
8. **Special section: cancer in Asian Americans, Native Hawaiians, and Pacific Islanders**. *Cancer Facts & Figures 2016* (2021.0)
9. Hastings KG, Jose PO, Kapphahn KI, Frank AT, Goldstein BA, Thompson CA, Eggleston K, Cullen MR, Palaniappan LP. **Leading causes of death among Asian American Subgroups (2003-2011)**. *PLoS One* (2015.0) **10** e0124341. DOI: 10.1371/journal.pone.0124341
10. Jin H, Pinheiro PS, Xu J, Amei A. **Cancer incidence among Asian American populations in the United States, 2009-2011**. *Int J Cancer* (2016.0) **138** 2136-45. DOI: 10.1002/ijc.29958
11. Kisling L, Das JM. **Prevention strategies**. *StatPearls* (2022.0)
12. **HPV vaccines**. *American Cancer Society* (2018.0)
13. **Recommended messages to reach Asian Americans**. *American Cancer Society* (2017.0)
14. Juon H, Rimal RN, Klassen A, Lee S. **Social norm, family communication, and HBV screening among Asian Americans**. *J Health Commun* (2017.0) **22** 981-9. DOI: 10.1080/10810730.2017.1388454
15. Kimura A, Sin M, Spigner C, Tran A, Tu S. **Barriers and facilitators to colorectal cancer screening in Vietnamese Americans: a qualitative analysis**. *J Cancer Educ* (2014.0) **29** 728-34. DOI: 10.1007/s13187-014-0646-6
16. Tingvold L, Middelthon A, Allen J, Hauff E. **Parents and children only? Acculturation and the influence of extended family members among Vietnamese refugees**. *Int J Intercult Relat* (2012.0) **36** 260-70. DOI: 10.1016/j.ijintrel.2011.03.005
17. Ho J, Birman D. **Acculturation gaps in Vietnamese immigrant families: impact on family relationships**. *Int J Intercult Relat* (2010.0) **34** 22-3. DOI: 10.1016/j.ijintrel.2009.10.002
18. Trieu MM. **Family obligation fulfillment among Southeast Asian American young adults**. *J Family Issues* (2014.0) **37** 1355-83. DOI: 10.1177/0192513x14551174
19. Lee J, Nguyen H, Park J, Tran L, Nguyen T, Huynh Y. **Usages of computers and smartphones to develop dementia care education program for Asian American family caregivers**. *Healthc Inform Res* (2017.0) **23** 338-42. DOI: 10.4258/hir.2017.23.4.338
20. O’Mara B, Gill GK, Babacan H, Donahoo D. **Digital technology, diabetes and culturally and linguistically diverse communities: a case study with elderly women from the Vietnamese community**. *Health Educ J* (2011.0) **71** 491-504. DOI: 10.1177/0017896911407054
21. Kratzke C, Cox C. **Smartphone technology and apps: rapidly changing health promotion**. *Int Electron J Health Educ* (2012.0) **15** 72-82
22. Kreps GL. **Online information and communication systems to enhance health outcomes through communication convergence**. *Hum Commun Res* (2017.0) **43** 518-30. DOI: 10.1111/hcre.12117
23. Kreps GL, Neuhauser L. **New directions in eHealth communication: opportunities and challenges**. *Patient Educ Couns* (2010.0) **78** 329-36. DOI: 10.1016/j.pec.2010.01.013
24. Pope Z, Lee JE, Zeng N, Lee HY, Gao Z. **Feasibility of smartphone application and social media intervention on breast cancer survivors' health outcomes**. *Transl Behav Med* (2019.0) **9** 11-22. DOI: 10.1093/tbm/iby002
25. Gong W, Wong BY, Ho S, Lai AY, Zhao S, Wang M, Lam T. **Family e-chat group use was associated with family wellbeing and personal happiness in Hong Kong adults amidst the covid-19 pandemic**. *Int J Environ Res Public Health* (2021.0) **18** 9139. DOI: 10.3390/ijerph18179139
26. Saud M, Mashud M, Ida R. **Usage of social media during the pandemic: seeking support and awareness about COVID-19 through social media platforms**. *J Public Affairs* (2020.0) e02417. DOI: 10.1002/pa.2417
27. Child J, Haridakis P. **Uses and gratifications theory considering media use in the context of family communication**. *Engaging Theories in Family Communication* (2017.0)
28. Child JT, Duck AR, Andrews LA, Butauski M, Petronio S. **Young adults’ management of privacy on Facebook with multiple generations of family members**. *J Family Commun* (2015.0) **15** 349-67. DOI: 10.1080/15267431.2015.1076425
29. Petronio S, Caughlin JP. **Communication privacy management theory: understanding families**. *Engaging Theories in Family Communication: Multiple Perspectives* (2018.0)
30. Koerner AF, Schrodt P. **An introduction to the special issue on family communication patterns theory**. *J Family Commun* (2014.0) **14** 1-15. DOI: 10.1080/15267431.2013.857328
31. Hecht ML, Krieger JL. **The principle of cultural grounding in school-based substance abuse prevention**. *J Language Social Psychol* (2016.0) **25** 301-19. DOI: 10.1177/0261927x06289476
32. Colby M, Hecht M, Miller-Day M, Krieger J, Syvertsen A, Graham J, Pettigrew J. **Adapting school-based substance use prevention curriculum through cultural grounding: a review and exemplar of adaptation processes for rural schools**. *Am J Community Psychol* (2013.0) **51** 190-205. DOI: 10.1007/s10464-012-9524-8
33. Palmer-Wackerly A, Krok J, Dailey P, Kight L, Krieger J. **Community engagement as a process and an outcome of developing culturally grounded health communication interventions: an example from the DECIDE project**. *Am J Community Psychol* (2014.0) **53** 261-74. DOI: 10.1007/s10464-013-9615-1
34. **American Community Survey (ACS)**. *United States Census Bureau*
35. **HPV vaccine decision narratives: encouraging informed HPV vaccine decision-making**. *NIH National Cancer Institute*
36. Gale NK, Heath G, Cameron E, Rashid S, Redwood S. **Using the framework method for the analysis of qualitative data in multi-disciplinary health research**. *BMC Med Res Methodol* (2013.0) **13** 117. DOI: 10.1186/1471-2288-13-117
37. Tracy S. *Qualitative Research Methods Collecting Evidence, Crafting Analysis, Communicating Impact* (2019.0)
38. Birt L, Scott S, Cavers D, Campbell C, Walter F. **Member checking: a tool to enhance trustworthiness or merely a nod to validation?**. *Qual Health Res* (2016.0) **26** 1802-11. DOI: 10.1177/1049732316654870
39. Cohen J. **A coefficient of agreement for nominal scales**. *Educ Psychol Measurement* (2016.0) **20** 37-46. DOI: 10.1177/001316446002000104
40. Perrin A, Anderson M. **Share of U.S. adults using social media, including Facebook, is mostly unchanged since 2018**. *Pew Research Center* (2019.0)
41. Primack BA, Shensa A, Escobar-Viera CG, Barrett EL, Sidani JE, Colditz JB, James AE. **Use of multiple social media platforms and symptoms of depression and anxiety: a nationally-representative study among U.S. young adults**. *Comput Human Behav* (2017.0) **69** 1-9. DOI: 10.1016/j.chb.2016.11.013
42. Lin J, Koehly L, Myers M, Marcum C. **Intergenerational communication about family health history: diagnoses, causes of death, and risk**. *Innov Aging* (2017.0) **1** 831. DOI: 10.1093/geroni/igx004.2995
43. Rodríguez VM, Corona R, Bodurtha JN, Quillin JM. **Family ties: the role of family context in family health history communication about cancer**. *J Health Commun* (2016.0) **21** 346-55. DOI: 10.1080/10810730.2015.1080328
44. Guéguen N. **Foot-in-the-door technique and computer-mediated communication**. *Comput Human Behav* (2002.0) **18** 11-5. DOI: 10.1016/S0747-5632(01)00033-4
45. Giles H, Baker S. **Communication accommodation theory**. *The International Encyclopedia of Communication* (2008.0)
46. Nguyen AB, Belgrave FZ, Sholley BK. **Development of a breast and cervical cancer screening intervention for Vietnamese American women: a community-based participatory approach**. *Health Promot Pract* (2011.0) **12** 876-86. DOI: 10.1177/1524839909355518
47. Tiro JA, Lee SC, Marks EG, Persaud D, Skinner CS, Street RL, Wiebe DJ, Farrell D, Bishop WP, Fuller S, Baldwin AS. **Developing a tablet-based self-persuasion intervention promoting adolescent HPV vaccination: protocol for a three-stage mixed-methods study**. *JMIR Res Protoc* (2016.0) **5** e19. DOI: 10.2196/resprot.5092
48. Smith D. **Health care consumer's use and trust of health information sources**. *J Commun Healthcare* (2013.0) **4** 200-10. DOI: 10.1179/1753807611Y.0000000010
49. Synder L, Hamilton M. **A meta-analysis of U.S. Health campaign effects on behavior: emphasize enforcement, exposure, and new information, and beware the secular trend**. *Public Health Communication* (2002.0)
50. Saunders B, Sim J, Kingstone T, Baker S, Waterfield J, Bartlam B, Burroughs H, Jinks C. **Saturation in qualitative research: exploring its conceptualization and operationalization**. *Qual Quant* (2018.0) **52** 1893-907. DOI: 10.1007/s11135-017-0574-8
|
---
title: Increased RTN3 phenocopies nonalcoholic fatty liver disease by inhibiting the
AMPK–IDH2 pathway
authors:
- Hao Huang
- Shuai Guo
- Ya‐Qin Chen
- Yu‐Xing Liu
- Jie‐Yuan Jin
- Yun Liang
- Liang‐Liang Fan
- Rong Xiang
journal: MedComm
year: 2023
pmcid: PMC10013133
doi: 10.1002/mco2.226
license: CC BY 4.0
---
# Increased RTN3 phenocopies nonalcoholic fatty liver disease by inhibiting the AMPK–IDH2 pathway
## Abstract
Reticulon 3 (RTN3), an endoplasmic reticulum protein, is crucial in neurodegenerative and kidney diseases. However, the role of RTN3 in liver tissues has not been described. Here, we employed public datasets, patients, and several animal models to explore the role of RTN3 in nonalcoholic fatty liver disease (NAFLD). The underlying mechanisms were studied in primary hepatocytes and L02 cells in vitro. We found an increased expression of RTN3 in NAFLD patients, high‐fat diet mice, and oxidized low‐density lipoprotein‐treated L02 cells. The RTN3 transgenic mice exhibited the phenotypes of fatty liver and lipid accumulation. Single‐cell RNA sequencing analysis indicated that increased RTN3 might induce mitochondrial dysfunction. We further showed this in primary hepatocytes, the L02 cell line, and the Caenorhabditis elegans strain. Mechanistically, RTN3 regulated these events through its interactions with glucose‐regulated protein 78 (GRP78), which further inhibited the adenosine 5 monophosphate‐activated protein kinase (AMPK)–isocitrate dehydrogenase 2 (IDH2) pathway. In the end, knockout of RTN3 relieved fatty liver and mitochondrial dysfunction. Our study indicated that RTN3 was important in NAFLD and lipid catabolism and that an increase in RTN3 in the liver might be a risk factor for nonalcoholic steatohepatitis and NAFLD.
This study suggested that RTN3 can interact with GRP78 on the endoplasmic reticulum. The increased expression of RTN3 can facilitate interactions between RTN3 and GRP78, thereby decreasing the activity of GRP78 in regulating AMPK phosphorylation, which further reduces the expression IDH2, finally resulting in mitochondrial dysfunction, increased reactive oxygen species, and NAFLD.
## INTRODUCTION
Nonalcoholic fatty liver disease (NAFLD) is one of the most common hepatic diseases and is defined as the accumulation of fat in the liver induced by nonalcoholic mechanisms. 1 It represents a clinical spectrum ranging from simple steatosis and nonalcoholic steatohepatitis (NASH) to cirrhosis and hepatocellular carcinoma. 2 Currently, the estimated worldwide prevalence of NAFLD is approximately $25\%$, which has become a crucial contributor to extrahepatic chronic diseases such as atherosclerosis and coronary heart disease. 3 As important organelles for lipid anabolism and catabolism, the endoplasmic reticulum (ER) and mitochondrion are crucial in the development and progression of NAFLD. 4 Many molecules and pathways involved in ER stress and mitochondrial dysfunction have been reported to promote the genesis and development of NAFLD. 5, 6 The discovery of new molecules that affect NAFLD will help us better understand the pathogenesis of NAFLD and aid in the treatment of this disease. 7 Reticulon 3 (RTN3) is an ER protein belonging to the RTN family with a signature C‐terminal RTN homolog domain that is important in shaping the tubule ER structure. 8 Functionally, RTN3 has been found to regulate the activity of Alzheimer's β‐secretase, mediate membrane contact between the ER and plasma membrane by interacting with the cytosolic region of epidermal growth factor receptor, promote lipid synthesis by activating sterol regulatory element‐binding protein 1c (SREBP1c) in adipose tissue, and induce chronic kidney disease via the insulin‐like growth factor 2 (IGF2) pathway. 9, 10, 11, 12 However, the role of RTN3 in liver tissues has not been described.
This study showed a strong correlation between increased RTN3 levels and NAFLD. Both transgenic mouse models overexpressing the wild‐type (WT) human RTN3 gene (Tg‐RTN3) and Caenorhabditis elegans strain overexpressing the RTN3 homology gene (Tg‐Ret1) presented with overt NAFLD and lipid accumulation. A mechanistic study revealed that the overexpression of RTN3 inhibited the adenosine 5 monophosphate‐activated protein kinase (AMPK)–isocitrate dehydrogenase 2 (IDH2) pathway by altering the interaction between RTN3 and glucose‐regulated protein 78 (GRP78), inducing mitochondrial dysfunction, which ultimately promotes the genesis and development of NAFLD. Finally, we found that reducing the expression of RTN3 can rescue the NAFLD and mitochondrial dysfunction caused by a high‐fat diet (HFD) by activating the AMPK–IDH2 pathway. Hence, our study indicated that overexpression of RTN3 in the liver might be a risk factor for NAFLD, and RTN3 may be a potential therapeutic target for NAFLD.
## Link between high expression of RTN3 and NAFLD
Because no previous study focused on RTN3 and liver diseases, we first analyzed the RNA levels of RTN3 in public datasets (GSE185051, GSE200409, and GSE200482). The results showed that the RNA levels of RTN3 were increased dramatically in the NAFLD patient group, HFD mouse group, and steatosis‐steatohepatitis diet (SSD) mouse group compared to healthy controls (Figure 1A–C). We then generated the NAFLD mouse model by HFD feeding. The immunohistochemistry (IHC) and western blot (WB) analysis revealed that the protein levels of RTN3 were greater in HFD‐WT mouse liver tissues ($$n = 6$$) than in the ad libitum diet (ALD‐WT) group ($$n = 6$$) (Figure 1D,E). According to previous study in NAFLD, 13 we selected normal human liver cell line L02 to perform the in vitro experiment. We treated L02 cells with oxidized low‐density lipoprotein (ox‐LDL). WB analysis showed that the longer the time treated with ox‐LDL, the higher the expression of RTN3 was (Figure 1F). Finally, we collected liver tissues from four NAFLD patients and two healthy controls (liver contusion patients) (Table S1). IHC analysis exhibited that the protein levels of RTN3 in NAFLD patients were greater than those in healthy controls (Figure 1G). These findings in public databases, patients, mice, and cell lines indicated a strong correlation between increased RTN3 levels and NAFLD.
**FIGURE 1:** *Link between high expression of reticulon 3 (RTN3) and nonalcoholic fatty liver disease (NAFLD). The mRNA levels of RTN3 in NAFLD patients and healthy control (A), in high‐fat diet (HFD) mice and wild‐type (WT) control (B), and in steatosis‐steatohepatitis diet (SSD) mice and WT control (C). The protein levels of RTN3 were detected in ad libitum diet (ALD)‐WT and HFD‐WT mice liver tissues by immunohistochemistry (IHC) (D) and western blotting (WB) (E). (F) WB showed the RTN3 levels in oxidized low‐density lipoprotein (ox‐LDL) (50 mg/L) treated L02 cell lines in different time. (G) IHC showed the RTN3 levels in liver tissues of healthy controls and NAFLD patients. *
p < 0.05, **
p < 0.01, and ***
p < 0.001. NASH: nonalcoholic steatohepatitis.*
## Overexpression of RTN3 exhibits NAFLD and fat accumulation
To reveal the relationship between RTN3 and NAFLD, we generated Tg‐RTN3 mice (Figure 2A). Hematoxylin–eosin (HE) staining and Oil Red O staining indicated that the Tg‐RTN3 mice ($$n = 6$$) presented with overt fat accumulation at 4 months of age with standard chow ($$n = 6$$) (Figure 2B,C). The liver weight and triglyceride (TAG) levels in liver tissues of Tg‐RTN3 mice were approximately $20.39\%$ and $48.02\%$ more than those in WT littermates (Figure 2D,E). Simultaneously, the liver function tests showed that the levels of serum alanine aminotransferase (ALT) and aspartate aminotransferase (AST) in Tg‐RTN3 mice were also much greater than those in their WT littermates (Figure 2F,G). In addition, the L02 cells transfected with pcDNA3.1‐RTN3 also presented more lipid accumulation than cells transfected with pcDNA3.1‐blank (Figure 2H). These observations suggested that overexpression of RTN3 may lead to NAFLD and fat accumulation.
**FIGURE 2:** *Overexpression of reticulon 3 (RTN3) exhibits nonalcoholic fatty liver disease (NAFLD) and fat accumulation. (A) Western blotting (WB) detected the RTN3 protein levels in wild‐type (WT) and Tg‐RTN3 mice liver tissues. (B) Hematoxylin–eosin (HE) staining and (C) Oil red O staining of WT and Tg‐RTN3 mice liver tissues. Liver weight (D) and liver triglyceride (TAG) levels (E) of WT and Tg‐RTN3 mice. The alanine aminotransferase (ALT) (F) and aspartate aminotransferase (AST) (G) levels in WT and Tg‐RTN3 mice serum. (H) Oil red O staining of L02 cells transfected with pcDNA3.1‐blank or pcDNA3.1‐RTN3. **
p < 0.01 and ***
p < 0.001.*
## Increased RTN3 is associated with lipid oxidation and mitochondrial respiration in NASH mice
Next, to further identify RTN3‐expressing hepatocyte subpopulations, a supervised analysis was performed on a previously published single‐cell RNA sequencing (scRNA‐seq) dataset isolated from the live tissues of chow diet, 15‐ and 30‐week high‐fat high‐fructose diet (HFHFD) mice, in which hepatocyte subpopulations unique to NASH were reported. 14 According to previous research, 11 hepatocyte subpopulations were identified in NASH model mice (Figure 3A). The results revealed the differential distribution of RTN3 among different NASH hepatocyte clusters (Figure 3B). RTN3 was predominantly highly expressed in the hepatocyte_1 cluster (Figure 3C). Analysis of Gene Ontology (GO) enrichment in cluster hepatocyte_1 suggested increased TAG and acylglycerol metabolic processes, as well as decreased adenosine triphosphate (ATP) metabolic processes, cellular respiration, and oxidative phosphorylation, in hepatocytes with high RTN3 expression (Figure 3D,E). Together, our results represent a comprehensive characterization of RTN3 in NASH, and the increased RTN3 showed potential functional importance in lipid oxidation and mitochondrial respiration.
**FIGURE 3:** *Increased reticulon 3 (RTN3) is associated with lipid oxidation and mitochondrial respiration from the nonalcoholic steatohepatitis (NASH) mice. (A) Composition and distribution of single cells from GSE166504. (B) The distribution profile of RTN3 for each cell by the t‐distributed stochastic neighbor embedding (t‐SNE) plot. (C) The violin plots of the expression profile of RTN3 in GSE166504. Dot plots summarizing the (D) upregulated and (E) downregulated enriched Gene Ontology (GO) in Rtn3 high expressed hepatocyte_1 cells. Dots are colored by the q‐value of significantly upregulated or downregulated genes within each gene set (<0.05 false discovery rate (FDR)).*
## Increased RTN3 can disrupt the morphology and function of mitochondria
Because scRNA‐seq indicated that increased RTN3 was related to mitochondrial respiration in NASH mouse models, we detected the morphology and function of mitochondria in the hepatocytes of Tg‐RTN3 mice. Transmission electron microscopy (TEM) exhibited that the size and density of hepatocyte mitochondria were significantly decreased in Tg‐RTN3 mice compared to controls (Figure 4A,B). Moreover, a loss of mitochondrial cristae structure was observed in the hepatocytes of Tg‐RTN3 mice (Figure 4A). Mitofusin 2 (MFN2) and fission mitochondrial 1 (FIS1) are responsible for regulating the fusion and fission of mitochondria, and optic atrophy 1 (OPA1) mitochondrial dynamin‐like GTPase (OPA1) plays a crucial role in maintaining the structure and function of mitochondrial cristae. 15 We then performed WB analysis of primary hepatocytes separated from Tg‐RTN3 and WT mice. The results showed reduced protein levels of MFN2 and OPA1 and increased protein levels of FIS1 in Tg‐RTN3 hepatocytes (Figure 4C), which indicated that overexpression of RTN3 may promote the fission of mitochondria and inhibit the fusion of mitochondria. This result was consistent with our TEM observations that in Tg‐RTN3 mice liver tissues, the size of mitochondria was reduced, and mitochondrial cristae structure was lost. Further detection of the levels of ATP and mitochondrial reactive oxygen species (ROS) showed that compared to those in WT hepatocytes, the ATP levels in Tg‐RTN3 hepatocytes were dramatically reduced (Figure 4D), and the ROS levels were increased (Figure 4E). We then transfected the pcDNA3.1‐RTN3 plasmid into L02 cells and detected a similar tendency in the levels of proteins that control mitochondrial morphology, as well as ATP and ROS levels (Figure 4F–H).
**FIGURE 4:** *Increased reticulon 3 (RTN3) can disrupt the morphology and function of mitochondria. (A) Transmission electron microscopy (TEM) described the mitochondria condition in liver tissues of wild‐type (WT) and Tg‐RTN3 mice. ER, endoplasmic reticulum; M, mitochondria. (B) The statistical results of the relative mitochondrial density. (C) The expression of OPA1, mitofusin 2 (MFN2), and fission mitochondrial 1 (FIS1) in primary hepatocytes from RTN3‐Tg mice and WT mice. The levels of adenosine triphosphate (ATP) (D) and reactive oxygen species (ROS) (E) in primary hepatocytes from RTN3‐Tg mice and WT mice. (F) The expression of OPA1, MFN2, and FIS01 in L02 cells transfected with pcDNA3.1‐blank or pcDNA3.1‐RTN3. The levels of ATP (G) and ROS (H) in L02 cells transfected with pcDNA3.1‐blank or pcDNA3.1‐RTN3. **
p < 0.01 and ***
p < 0.001.*
Ret1 is a homologous gene of RTN3 in C. elegans. In our generated Tg‐Ret1 C. elegans, Oil red O staining indicated that the number of large lipid drops in Tg‐Ret1 was much greater than that in the WT (Figure 5A). The TEM of Tg‐Ret1 C. elegans displayed fragmented mitochondria and increased lipid droplets compared to WT (Figure 5B). Real‐time polymerase chain reaction (RT‐PCR) also suggested that the mRNA levels in the C. elegans Mfn1,2 homologue (FZO1) (Mfn2) and the C. elegans Opa1 homologue (EAT3) (Opa1) were decreased in Tg‐Ret1 C. elegans compared to WT, while the expression of mRNA in Fis1 was increased (Figure 5C).
**FIGURE 5:** *Increased Ret1 can lead to lipid accumulation and mitochondrial dysfunction in Caenorhabditis elegans. (A) Oli red O staining of the wild‐type (WT) and Tg‐Ret1 C. elegans. The arrows indicated the large lipid droplets. (B) Transmission electron microscopy (TEM) describing the mitochondria and lipid droplets condition of the WT and Tg‐Ret1 C. elegans. LD, lipid droplet; M, mitochondria. (C) Real‐time polymerase chain reaction (RT‐PCR) revealed the mRNA levels of Ret1, EAT3, FZO1, and fission mitochondrial 1 (FIS1) in WT and Tg‐Ret1 C. elegans. **
p < 0.01 and ***
p < 0.001.*
These discoveries in mouse hepatocytes and C. elegans confirmed that overexpression of RTN3 can disrupt the morphology and function of mitochondria, which may further affect lipid catabolism in hepatocytes and result in NAFLD.
## Increased RTN3 can inhibit the AMPK–IDH2 pathway by interacting with GRP78
We then employed RNA‐seq to investigate the potential pathways between increased RTN3 and mitochondrial dysfunction. We found that a crucial gene, IDH2, was changed dramatically in the liver tissues of Tg‐RTN3 mice compared to WT littermates at 4 months of age fed standard chow (Figure 6A). IDH2, an enzyme located in the inner membrane of mitochondria, belongs to the family of isocitrate dehydrogenases (IDHs), which are significant in regulating intermediary metabolism and energy production. 16, 17 Former studies have indicated that a reduction in IDH2 can lead to mitochondrial dysfunction and NAFLD. 18, 19 RT‐PCR and WB also confirmed the reduction in IDH2 in Tg‐RTN3 primary hepatocytes or L02 cells transfected with pcDNA‐RTN3 compared to WT controls (Figure 6B–D).
**FIGURE 6:** *Increased reticulon 3 (RTN3) can inhibit adenosine 5 monophosphate‐activated protein kinase (AMPK)–isocitrate dehydrogenase 2 (IDH2) pathway via interacting with glucose‐regulated protein 78 (GRP78). (A) Significantly differentially expressed genes between wild‐type (WT) and Tg‐RTN3 mice liver tissues revealed by RNA‐seq data. (B) Real‐time polymerase chain reaction (RT‐PCR) showed the mRNA levels of RTN3 and IDH2 in WT and Tg‐RTN3 mice liver tissues. Western blotting (WB) analyzed the IDH2 levels in (C) WT and Tg‐RTN3 mice liver tissue group, and in (D) L02 cell lines transfected with pcDNA3.1‐blank and pcDNA3.1‐RTN3 group. (E) Coimmunoprecipitation (Co‐IP) confirmed that RTN3 can interact with GRP78 in mouse liver tissues. (F) WB and Co‐IP analysis showing the levels of GRP78, coimmunoprecipitated GRP78, AMPK, and p‐AMPK in primary hepatocytes from RTN3‐Tg mice and WT mice. (G) The statistical results of the RTN3, GRP78, and Co‐IPed GRP78. (H) The statistical results of the p‐AMPK. (I) Potential mechanism of how higher RTN3 expression induces mitochondrial dysfunction in liver. The figure was created with BioRender.com. *
p < 0.05 and ***
p < 0.001.*
However, it is unclear how RTN3 regulates the expression of IDH2 and ultimately leads to mitochondrial dysfunction and NAFLD. Mass spectrometry (MS) was used to detect the candidate RTN3‐interacting molecules in WT liver tissues. Interestingly, GRP78 was found to be a possible RTN3‐interacting protein (data not shown). Coimmunoprecipitation (Co‐IP) analysis further validated the interaction between RTN3 and GRP78 in primary hepatocytes (Figure 6E). GRP78, also known as binding immunoglobulin protein (BiP), is an ER protein that facilitates a wide range of protein folding processes. 20 Simultaneously, GRP78 has been proved to regulate the expression of AMPK, 21 a guardian of metabolism and mitochondrial homeostasis, 22 which has been proved can regulate the transcription of IDH2. 23, 24, 25 Further WB and Co‐IP studies also suggested that the protein levels of GRP78 were not changed by increasing the RTN3 levels. However, the amount of coimmunoprecipitated GRP78 was much higher in Tg‐RTN3 group than in WT group, and this increased interaction appeared to inhibit activated AMPK (p‐AMPK) (Figure 6F–H).
Collectively, our data suggested that RTN3 can interact with GRP78 on the ER. The overexpression of RTN3 can promote the interactions between RTN3 and GRP78, thereby reducing the ability of GRP78 in regulating AMPK phosphorylation, which further reduces the expression IDH2, finally resulting in mitochondrial dysfunction, increased ROS, and NAFLD (Figure 6I).
## Decreased RTN3 can rescue NAFLD and mitochondrial dysfunction caused by HFD by activating the AMPK–IDH2 pathway
RTN3‐null (RTN3 knockout [KO]) mice were also generated to further confirm the relationship between RTN3 and NAFLD (Figure 7A). We found that RTN3 KO mice ($$n = 6$$) exhibited less fat accumulation in liver tissues than WT mice ($$n = 6$$) after 3 months of HFD feeding (Figure 7B). The liver weight and TAG levels in the HFD‐RTN3 KO group were lower than those in the HFD‐WT group (Figure 7C,D). Liver function tests also suggested that the levels of ALT and AST were reduced in HFD‐RTN3 KO mice compared with HFD‐WT mice (Figure 7E,F). In addition, we knocked down the expression of Ret1 in obese worms (daf‐22) by RNAi and found that the number of larger lipid drops was reduced in the daf‐22; Ret1 RNAi group compared to the daf‐22 group (Figure 7G). All these observations suggest that reducing the expression of RTN3 can relieve the NAFLD and lipid accumulation caused by HFD. Referring to the aforementioned mechanism, we also detected the expression of OPA1, MFN2, and OPA1, and levels of ATP and ROS in the liver tissues of HFD‐RTN3 KO mice. The data exhibited that compared to HFD‐WT group, the protein levels of MFN2 and OPA1, and the ATP levels were increased, while the protein levels of FIS1 and ROS levels were decreased in HFD‐RTN3 KO group (Figure 8A–C). Similar results were also detected in daf‐22; Ret1 RNAi C. elegans compared to daf‐22 C. elegans by RT‐PCR (Figure 8D). Finally, an elevatory tendency of IDH2 and p‐AMPK was detected by WB in HFD‐RTN3 KO mice liver tissues (Figure 8E).
**FIGURE 7:** *Decreased reticulon 3 (RTN3) can rescue nonalcoholic fatty liver disease (NAFLD) caused by high‐fat diet (HFD). (A) Western blotting (WB) detected the RTN3 protein levels in wild‐type (WT) and RTN3 knockout (KO) mice liver tissues. (B) Hematoxylin–eosin (HE) staining of high‐fat diet (HFD)‐WT and HFD‐RTN3 KO mice liver tissues. Liver weight (C) and liver triglyceride (TAG) levels (D) of HFD‐WT and HFD‐RTN3 KO mice. The alanine aminotransferase (ALT) (E) and aspartate aminotransferase (AST) (F) levels in HFD‐WT and HFD‐RTN3 KO mice serum. (G) Oil red O staining of daf‐22 and daf‐22: RNAi Caenorhabditis elegans. The arrows indicated the large lipid droplets. *
p < 0.05 and **
p < 0.01.* **FIGURE 8:** *Decreased reticulon 3 (RTN3) can rescue mitochondrial dysfunction caused by high‐fat diet (HFD) via activating adenosine 5 monophosphate‐activated protein kinase (AMPK)–isocitrate dehydrogenase 2 (IDH2) pathway. (A) The expression of OPA1, mitofusin 2 (MFN2), and fission mitochondrial 1 (FIS1) in liver tissues from HFD‐WT mice and HFD‐RTN3 knockout (KO) mice. The levels of adenosine triphosphate (ATP) (B) and reactive oxygen species (ROS) (C) in liver tissues from HFD‐WT mice and HFD‐RTN3 KO mice. (D) Real‐time polymerase chain reaction (RT‐PCR) revealed the mRNA levels of Ret1, EAT3, FZO1, and FIS1 in dat‐22 and daf‐22: RNAi Caenorhabditis elegans. (E) The expression of IDH2 and p‐AMPK in liver tissues from HFD‐WT mice and HFD‐RTN3 KO mice. *
p < 0.05.*
In RTN3 KO mice and related C. elegans, studies have indicated that reducing the expression of RTN3 can rescue lipid accumulation, liver function, and mitochondrial dysfunction caused by HFD by activating the AMPK–IDH2 pathway. Reducing the expression of RTN3 in the liver may be a potential therapeutic strategy for treating NAFLD.
## DISCUSSION
As an ER membrane protein, RTN3 plays a significant role in the nervous system. 26, 27 Although richly expressed in neurons, this protein is expressed in many tissues and has eight spliced forms, and only a few studies have shown the role and function of RTN3 in peripheral tissues. 11, 12 In mammals, RTN3 and reticulon 4 (RTN4) are the only two RTN family members expressed in the liver. 28 Previous studies have demonstrated that RTN4 can facilitate hepatocyte proliferation and liver regeneration, suggesting that RTN4 is an important regulator of hepatic fibrosis. 29, 30 Moreover, the RTN4 receptor can activate the protein kinase alpha pathway and increase the nuclear translocation of liver X receptor α, promoting hepatic lipogenesis. 31 However, as the homolog of RTN4, the function of RTN3 in liver tissues was not previously investigated. Here, we provided the evidence that the increased expression of hepatic RTN3 leads to fat accumulation and NAFLD due to mitochondrial dysfunction and increased ROS. This conclusion was supported by experiments in mice, C. elegans, cell lines, and NAFLD patient samples, which help us understand how RTN3 and its interacting proteins contribute to mitochondrial dysfunction and NAFLD. Drug development for related targets such as RTN3 may be a potential therapeutic strategy for treating NAFLD. 7 In our study, we revealed that increased RTN3 expression may lead to mitochondrial dysfunction in hepatocytes. Previous studies reported that structural and functional alterations of mitochondria in hepatocytes contribute to the pathogenesis of NAFLD. 32, 33 Mitochondrial dysfunction impairs the ability to handle increased lipid flux, thereby disrupting lipid catabolism; respiratory oxidation may collapse with impairment of fat homeostasis, generation of lipid‐derived toxic metabolites, and overproduction of ROS, which further contribute to NAFLD. 34, 35 *In this* study, we first report that increased RTN3 may induce mitochondrial dysfunction and that the mitochondrial regulatory role of RTN3 is evolutionarily conserved.
By employing biochemical approaches, we identified overtly decreased expression of IDH2 in the liver tissues of Tg‐RTN3 mice. Previous studies have demonstrated that a reduction in IDH2, an enzyme located in mitochondria, may lead to mitochondrial dysfunction in hepatocytes and cardiomyocytes. 16, 17, 36 Simultaneously, the latest studies have indicated that IDH2‐null mice exhibit mitochondrial damage, increased ROS, and hepatic lipid dysregulation. 18 Previous studies have shown that increased ROS can directly promote lipid deposition in the liver, which is the pathogenesis of NAFLD. 34 IDH2 is an enzyme located at the inner mitochondrial membrane, while RTN3 is a tubular ER protein. How RTN3 impacts IDH2 expression is an important question. Here, we showed that RTN3 can interact with GRP78, which can further regulate the AMPK pathway, the upstream signaling pathway of IDH2. 21, 23, 24, 37 Previous studies have suggested that AMPK is the key molecule in regulating lipid metabolism by phosphorylation. GRP78 can activate AMPK to ameliorate dexamethasone‐induced fatty liver disease in C57BL/6 mice. 21 Vaspin can attenuate steatosis‐induced fibrosis via the GRP78 receptor by targeting the AMPK signaling pathway. As the upstream molecule of IDH2, AMPKα has been proven to regulate IDH2 transcription. 23, 25 AMPKα1 deficiency may attenuate IDH2 expression in mice, which is consistent with our studies in mice. Hence, increased RTN3 may bind more GRP78, which reduces the activity of GRP78 in regulating p‐AMPK, attenuates IDH2 expression, and induces mitochondrial dysfunction and NAFLD.
Recently, some studies have focused on the role of RTN3 in physiology and pathology. Mutations in RTN3 have been detected in early‐onset Alzheimer's disease. 38 One study found that RTN3 may mediate checkpoint kinase 2 activation and suppress hepatocellular carcinogenesis. 39 In lipid research, one study found that RTN3 can alter very low‐density lipoprotein secretion in HepG2 cells, 40 but in our Tg‐RTN3 mice, we only found that TAG levels were increased. 11 Formerly, we found that increased RTN3 may activate the SREBF chaperone (SCAP)–SREBP1 pathway by interacting with heat shock protein family A (Hsp70) member 5 in fat tissues. The expression of SREBP1 was dramatically increased in the fat tissues of Tg‐RTN3 mice, which proved that RTN3 is crucial in regulating lipid anabolism. 11 In the RTN3 KO mouse model, we also found that deletion of RTN3 can activate the IGF2–Janus kinase 2 (JAK2) pathway, which may further lead to chronic kidney disease and renal fibrosis. 12 Here, we found that increased RTN3 can also disrupt the structure and function of mitochondria. Since the disruption of mitochondria may promote the levels of ROS and have a crucial role in TAG catabolism in liver tissues, 41, 42 our studies further confirmed the function of RTN3 in TAG catabolism in liver tissues via the AMPK–IDH2 pathway by employing multiple mouse and C. elegans models.
## CONCLUSIONS
In summary, our study suggests that high expression of RTN3 is a possible lesion for lipid deposition in the liver by impairing mitochondria and increasing ROS levels, partly through competitive binding with GRP78 to attenuate the GRP78‐mediated AMPK–IDH2 pathway. Reducing the expression of RTN3 may be a potential therapeutic method for NAFLD. Our findings also imply that RTN3 is a pivotal upstream regulator of the AMPK–IDH2 pathway that functions in mitochondrial biological homeostasis.
## Human tissues
Human liver sections were collected from Xiangya Hospital and the Second Xiangya Hospital. This study was approved by the Institutional Review Board committee of Central South University in China (approval number: no. 2021‐1‐1 for human specimens, no. 2021‐2‐1 for animals, date: February 24, 2021).
## Mouse strains, C. elegans strains, cell lines, and key reagents
Tg‐RTN3 and RTN3 KO mice were described previously. 11 The WT mice (C57BL/6J) were purchased from Cyagen Company (SuZhou, China). Male mice at 7–8 weeks of age were selected to feed a HFD consisting of $60\%$ fat, $20\%$ protein, and $20\%$ carbohydrate for 3 months.
The WT C. elegans (N2 strain), Tg‐Ret1 C. elegans strain, and Daf‐22 C. elegans strain were purchased or generated as previously described. 11 Primary hepatocytes were isolated from mouse liver tissues as follows: fresh liver tissues were separated from newborn mice and cut up by scissors. The fragmented tissues were washed with phosphate‐buffered saline with 100 U/mL penicillin and 100 μg/mL streptomycin and digested with collagenase/hyaluronidase. After filtration and centrifugation, the separated cells were cultured with complete medium. The L02 cell line was purchased from the GuangZhou Jennio Biotech Co., Ltd. (GuangZhou, China) and cultured with complete medium. Both primary hepatocytes and the L02 cell line were maintained at 37°C in a humidified, $5\%$ CO2‐controlled atmosphere (Thermo Fisher Scientific).
The RTN3 antibody was generated in the Yan laboratory. 26 The other key reagents are presented in Table S2.
## Immunohistochemistry
Tissues were fixed with formalin and embedded in paraffin. Six‐micrometer sections were prepared for baking and dewaxing. After antigen retrieval and blocking, the RTN3 antibody was incubated for 8 h. Then, a broad spectrum IHC kit was used for the next steps including secondary antibody incubation, 3,3′‐diaminobenzidine staining, and hematoxylin staining. Finally, the slides were examined by routine light microscopy (DM6 M LIBS, Leica).
## Western blotting
Tissues or cells were dissociated on ice in radio‐immunoprecipitation assay lysis buffer with protease inhibitor cocktail for 1 h. The homogenates were centrifuged at 15,000 ×g for 120 min at 4°C to separate the supernatants. The protein concentrations were detected by bicinchoninic acid protein assays and analysis kits. A total of 30 μg protein lysates were separated by $4\%$–$12\%$ NuPAGE Bis‐Tris gel electrophoresis by standard methods with the primary and secondary antibodies mentioned above. Finally, the bands were detected in iBright Gel imaging system (Thermo Fisher Scientific).
## HE staining and Oil Red O staining
For HE staining, paraformaldehyde‐fixed liver tissue was embedded in paraffin and sliced into 6‐μm sections. The slides were baked at 60°C and dewaxed by dimethylbenzene. After alcohol treatment with an inverse concentration gradient, the slides were stained by HE with HE Stain Kit according to established protocols. Finally, the slides were examined by routine light microscopy (DM6 M LIBS, Leica).
For Oil red O staining, paraformaldehyde‐fixed tissues were embedded in optimal cutting temperature compound and sliced into 10‐μm sections. The cells were cultured in a Nunc Lab‐Tek Chamber Slide System (Thermo Fisher Scientific). The slides were treated with isopropanol for 5 min. Oil Red O staining, hematoxylin staining, and slide sealing were performed with an Oil Red O Stain Kit according to established protocols. Finally, the slides were examined by routine light microscopy (DM6 M LIBS, Leica).
## ALT assay, AST assay, and TAG detection
For ALT and AST assays, mouse serum was extracted at 1000 ×g for 5 min at 4°C, and the supernatants were separated and treated with ALT or AST assay kits according to established protocols. Finally, the levels of ALT or AST were detected with a colorimeter (Cary 60 UV‒Vis, Agilent) at a wavelength of 505 nm.
For TAG detection, the liver tissues were dissociated on ice in isopropanol. The homogenates were centrifuged at 15,000 ×g for 120 min at 4°C, and the supernatants were collected and treated with a Triglyceride Detection Kit according to established protocols. Levels of TAG were detected with a colorimeter (Cary 60 UV‒Vis) at a wavelength of 505 nm.
## Data collection, single‐cell RNA sequencing data processing, and GO enrichment
The RNA‐seq datasets of GSE185051 and GSE200482, as well as microarray dataset GSE200409, were obtained from the Gene Expression Omnibus (GEO) database with log2 transformation. The R (version 4.0.4) and RStudio (version 1.2.5033) were used to address all the data in this study. GSE185051 contained 52 NAFLD samples and five healthy liver samples, GSE200482 contained four normal chow diet C57BL6J mouse liver samples and five SSD C57BL6J mouse samples, and GSE200409 contained 12 mouse liver samples from mice fed a normal or HFD for different times.
scRNA‐seq data from the GSE166504 dataset were obtained from the GEO database. 14 The R package “Seurat” (version 4.0.2) was used to process the data. Six hepatocyte samples from mice fed a HFHFD for 15 weeks and four hepatocyte samples from mice fed a HFHFD for 30 weeks were selected for analysis. The scRNA‐seq data were analyzed as we previously described. 43 The t‐distributed stochastic neighbor embedding algorithm was used to explore and visualize cluster classifications across the cell samples. The cell clusters were annotated manually based on the related metadata of GSE166504.
The “clusterProfiler” (version 3.18.1) R package was used for GO enrichment analyses, which included biological process, molecular function, and cellular component. The p‐value was set at 0.05.
## Transmission electron microscopy
The samples were treated $4\%$ glutaraldehyde as we described previously. 12, 26 The dissection and staining were performed by the Advanced Research Center, Central South University, China. The thickness of the section was 70 nm, and the staining was performed using uranyl acetate and lead citrate. The pictures were collected by TEM (H‐7650; Hitachi).
## ATP assay and ROS assay
Liver tissues or cells were homogenized on ice with extraction solution. After centrifugation, the supernatants were collected for detection. The ATP levels were measured by phosphomolybdic acid colorimetry with an ATP array kit at a wavelength of 340 nm with a colorimeter (Cary 60 UV‒Vis). The ROS levels were measured by fluorometric analysis using 2,7‐dichlorofluorescein diacetate with an ROS array kit. The ROS levels were detected by colorimeter (Cary 60 UV‒Vis) with 488 nm excitation wavelength.
## RNAi screening and plasmid transfection
The HT115 bacteria which included the plasmids that can express the dsRNAs to target the ret‐1 and silence the expression of ret‐1 were prepared for adult C. elegans.
The WT RTN3 oding sequence (CDS) with a C‐terminal Flag‐tag in pcDNA3.1+ was designed and structured. The L02 cells were transiently transfected with pcDNA3.1‐blank and pcDNA3.1‐RTN3 using Lipofectamine™ 3000 CD Transfection Reagent following the manufacturer's instructions.
## RNA‐seq and RT‐PCR
Total RNA was isolated from tissues or cells with an RNA isolation kit. The RNA‐seq and bioinformatics analyses were performed by the BerryGenomics Biotech company (Beijing, China). cDNA was synthesized by RevertAid First Strand cDNA Synthesis Kit with 1 μg RNA and prepared with Maxima SYBR Green/ROX qPCR Master Mix (2×). Finally, the Fast 7500 RT‐PCR Systems (Applied Biosystems) and 2(−△△Ct) methods were used to compare the RNA levels of each group.
## MS analysis and Co‐IP
The protein supernatants of WT mouse liver tissues were extracted as described in the WB methods. A total of 500 μg lysates in 1 mL were used for MS and Co‐IP with primary antibody (IgG or anti‐RTN3) and Protein A + G beads overnight. After denaturation by heating and centrifugation, the supernatants were separated by $4\%$–$12\%$ NuPAGE Bis‐Tris gel electrophoresis. For MS analysis, the gel was stained with Coomassie blue staining solution R250 (ST1123, Beyotime Biotechnology), and the potential bands were cut and sent to Novogene Bioinformatics Institute (Beijing, China) for further MS analysis. For Co‐IP, the supernatants were detected by standard WB methods with antibodies against RTN3 and GRP78. The bands were detected in iBright Gel imaging system (Thermo Fisher Scientific).
## Statistical analysis
The data were subjected to statistical analysis with GraphPad Prism 6 (GraphPad Software). Five independent repetitions were conducted for each group in this study. Image J was used to perform the WB grayscale analysis. All data are presented as means ± standard deviation. Data were analyzed using paired Student's t‐test. $p \leq 0.05$ was considered significant.
## AUTHOR CONTRIBUTIONS
H.H. and S.G. wrote the draft of the manuscript and performed the bioinformatic analysis and cell and molecular experiments. Y.‐Q.C. enrolled the patient's samples. Y.‐X.L. and J.‐Y.J. performed hematoxylin–eosin and immunohistochemistry. Y.L. performed animal feeding. L.‐L.F. and R.X. revised the manuscript and designed and supported the project. All authors approved the final manuscript.
## CONFLICT OF INTEREST STATEMENT
The authors declare no conflicts of interest.
## ETHICS STATEMENT
All procedures followed were in accordance with the ethical standards of the Helsinki Declaration of 1975, as revised in 2000. This study was approved by the Institutional Review Board committee of the Central South University in China (approval number: No. 2021‐1‐1 for human specimen, No. 2021‐2‐1 for animal, date: February 24, 2021). All patients provided written informed consent.
## DATA AVAILABILITY STATEMENT
Experimental data related to the article are available from the corresponding author.
## References
1. Cotter TG, Rinella M. **Nonalcoholic fatty liver disease 2020: the state of the disease**. *Gastroenterology* (2020) **158** 1851-1864. PMID: 32061595
2. Tokushige K, Ikejima K, Ono M. **Evidence‐based clinical practice guidelines for nonalcoholic fatty liver disease/nonalcoholic steatohepatitis 2020**. *J Gastroenterol* (2021) **56** 951-963. PMID: 34533632
3. Wong WK, Chan WK. **Nonalcoholic fatty liver disease: a global perspective**. *Clin Ther* (2021) **43** 473-499. PMID: 33526312
4. Loomba R, Friedman SL, Shulman GI. **Mechanisms and disease consequences of nonalcoholic fatty liver disease**. *Cell* (2021) **184** 2537-2564. PMID: 33989548
5. Lebeaupin C, Vallee D, Hazari Y, Hetz C, Chevet E, Bailly‐Maitre B. **Endoplasmic reticulum stress signalling and the pathogenesis of non‐alcoholic fatty liver disease**. *J Hepatol* (2018) **69** 927-947. PMID: 29940269
6. Prasun P, Ginevic I, Oishi K. **Mitochondrial dysfunction in nonalcoholic fatty liver disease and alcohol related liver disease**. *Transl Gastroenterol Hepatol* (2021) **6** 4. PMID: 33437892
7. Negi CK, Babica P, Bajard L, Bienertova‐Vasku J, Tarantino G. **Insights into the molecular targets and emerging pharmacotherapeutic interventions for nonalcoholic fatty liver disease**. *Metabolism* (2022) **126**. PMID: 34740573
8. Parashar S, Chidambaram R, Chen S. **Endoplasmic reticulum tubules limit the size of misfolded protein condensates**. *Elife* (2021;10:e71642)
9. Sharoar MG, Yan R. **Effects of altered RTN3 expression on BACE1 activity and Alzheimer's neuritic plaques**. *Rev Neurosci* (2017) **28** 145-154. PMID: 27883331
10. Caldieri G, Barbieri E, Nappo G. **Reticulon 3‐dependent ER‐PM contact sites control EGFR nonclathrin endocytosis**. *Science* (2017) **356** 617-624. PMID: 28495747
11. Xiang R, Fan LL, Huang H. **Increased reticulon 3 (RTN3) leads to obesity and hypertriglyceridemia by interacting with heat shock protein family A (Hsp70) member 5 (HSPA5)**. *Circulation* (2018) **138** 1828-1838. PMID: 29716941
12. Fan LL, Du R, Liu JS. **Loss of RTN3 phenocopies chronic kidney disease and results in activation of the IGF2–JAK2 pathway in proximal tubular epithelial cells**. *Exp Mol Med* (2022) **54** 653-661. PMID: 35596061
13. Wan X, Xu C, Lin Y. **Uric acid regulates hepatic steatosis and insulin resistance through the NLRP3 inflammasome‐dependent mechanism**. *J Hepatol* (2016) **64** 925-932. PMID: 26639394
14. Su Q, Kim SY, Adewale F. **Single‐cell RNA transcriptome landscape of hepatocytes and non‐parenchymal cells in healthy and NAFLD mouse liver**. *iScience* (2021) **24**. PMID: 34755088
15. Rovira‐Llopis S, Banuls C, Diaz‐Morales N, Hernandez‐Mijares A, Rocha M, Victor VM. **Mitochondrial dynamics in type 2 diabetes: pathophysiological implications**. *Redox Biol* (2017) **11** 637-645. PMID: 28131082
16. Pan JH, Kim HS, Beane KE. **IDH2 deficiency aggravates fructose‐induced NAFLD by modulating hepatic fatty acid metabolism and activating inflammatory signaling in female mice**. *Nutrients* (2018) **10**
17. Han SJ, Choi HS, Kim JI, Park JW, Park KM. **IDH2 deficiency increases the liver susceptibility to ischemia‐reperfusion injury via increased mitochondrial oxidative injury**. *Redox Biol* (2018) **14** 142-153. PMID: 28938192
18. Lee SJ, Cha H, Kim H, Lee JH, Park JW. **Amelioration of late‐onset hepatic steatosis in IDH2‐deficient mice**. *Free Radic Res* (2017) **51** 368-374. PMID: 28415887
19. Gong F, Gao L, Ding T. **IDH2 protects against nonalcoholic steatohepatitis by alleviating dyslipidemia regulated by oxidative stress**. *Biochem Biophys Res Commun* (2019) **514** 593-600. PMID: 31064654
20. Sahin E, Bagci R, Bektur Aykanat NE, Kacar S, Sahinturk V. **Silymarin attenuated nonalcoholic fatty liver disease through the regulation of endoplasmic reticulum stress proteins GRP78 and XBP‐1 in mice**. *J Food Biochem* (2020) **44**. PMID: 32189355
21. Gupta AP, Singh P, Garg R. **Pancreastatin inhibitor activates AMPK pathway via GRP78 and ameliorates dexamethasone induced fatty liver disease in C57BL/6 mice**. *Biomed Pharmacother* (2019) **116**. PMID: 31108350
22. Herzig S, Shaw RJ. **AMPK: guardian of metabolism and mitochondrial homeostasis**. *Nat Rev Mol Cell Biol* (2018) **19** 121-135. PMID: 28974774
23. Cao Y, Liu X, Zhao J, Du M. **AMPKalpha1 regulates Idh2 transcription through H2B O‐GlcNAcylation during brown adipogenesis**. *Acta Biochim Biophys Sin* (2021) **53** 112-118. PMID: 33219380
24. Du D, Liu C, Qin M. **Metabolic dysregulation and emerging therapeutical targets for hepatocellular carcinoma**. *Acta Pharm Sin B* (2022) **12** 558-580. PMID: 35256934
25. Shao C, Lu W, Du Y. **Cytosolic ME1 integrated with mitochondrial IDH2 supports tumor growth and metastasis**. *Redox Biol* (2020) **36**. PMID: 32853879
26. Shi Q, Ge Y, Sharoar MG. **Impact of RTN3 deficiency on expression of BACE1 and amyloid deposition**. *J Neurosci* (2014) **34** 13954-13962. PMID: 25319692
27. Sharoar MG, Shi Q, Ge Y. **Dysfunctional tubular endoplasmic reticulum constitutes a pathological feature of Alzheimer's disease**. *Mol Psychiatry* (2016) **21** 1263-1271. PMID: 26619807
28. Cai Y, Saiyin H, Lin Q. **Identification of a new RTN3 transcript, RTN3‐A1, and its distribution in adult mouse brain**. *Brain Res Mol Brain Res* (2005) **138** 236-243. PMID: 15946766
29. Gao L, Utsumi T, Tashiro K. **Reticulon 4B (Nogo‐B) facilitates hepatocyte proliferation and liver regeneration in mice**. *Hepatology* (2013) **57** 1992-2003. PMID: 23299899
30. Zhang D, Utsumi T, Huang HC. **Reticulon 4B (Nogo‐B) is a novel regulator of hepatic fibrosis**. *Hepatology* (2011) **53** 1306-1315. PMID: 21480333
31. Hu W, Zhang W, Chen Y. **Nogo‐B receptor deficiency increases liver X receptor alpha nuclear translocation and hepatic lipogenesis through an adenosine monophosphate‐activated protein kinase alpha‐dependent pathway**. *Hepatology* (2016) **64** 1559-1576. PMID: 27480224
32. Buzzetti E, Pinzani M, Tsochatzis EA. **The multiple‐hit pathogenesis of non‐alcoholic fatty liver disease (NAFLD)**. *Metabolism* (2016) **65** 1038-1048. PMID: 26823198
33. Galloway CA, Yoon Y. **Mitochondrial morphology in metabolic diseases**. *Antioxid Redox Signal* (2013) **19** 415-430. PMID: 22793999
34. Chen Z, Tian R, She Z, Cai J, Li H. **Role of oxidative stress in the pathogenesis of nonalcoholic fatty liver disease**. *Free Radic Biol Med* (2020) **152** 116-141. PMID: 32156524
35. Zhao Q, Liu J, Deng H. **Targeting mitochondria‐located circRNA SCAR alleviates NASH via reducing mROS output**. *Cell* (2020) **183** 76-93. PMID: 32931733
36. Ku HJ, Park JW. **Downregulation of IDH2 exacerbates H2O2‐mediated cell death and hypertrophy**. *Redox Rep* (2017) **22** 35-41. PMID: 26865387
37. Abdolahi A, Vahabzadeh Z, Izadpanah E, Moloudi MR. **Vaspin attenuates steatosis‐induced fibrosis via GRP78 receptor by targeting AMPK signaling pathway**. *J Physiol Biochem* (2022) **78** 185-197. PMID: 35001345
38. Zou Y, He W, Wang K. **Identification of rare RTN3 variants in Alzheimer's disease in Han Chinese**. *Hum Genet* (2018) **137** 141-150. PMID: 29356939
39. Song S, Shi Y, Wu W. **Reticulon 3‐mediated Chk2/p53 activation suppresses hepatocellular carcinogenesis and is blocked by hepatitis B virus**. *Gut* (2021) **70** 2159-2171. PMID: 33303565
40. Siddiqi S, Zhelyabovska O, Siddiqi SA. **Reticulon 3 regulates very low density lipoprotein secretion by controlling very low density lipoprotein transport vesicle biogenesis**. *Can J Physiol Pharmacol* (2018) **96** 668-675. PMID: 29756473
41. Cheng X, Geng F, Pan M. **Targeting DGAT1 ameliorates glioblastoma by increasing fat catabolism and oxidative stress**. *Cell Metab* (2020) **32** 229-242. PMID: 32559414
42. Domondon M, Nikiforova AB, DeLeon‐Pennell KY, Ilatovskaya DV. **Regulation of mitochondria function by natriuretic peptides**. *Am J Physiol Renal Physiol* (2019) **317** F1164-F1168. PMID: 31509010
43. Huang H, Zhu L, Huang C. **Identification of hub genes associated with clear cell renal cell carcinoma by integrated bioinformatics analysis**. *Front Oncol* (2021) **11**. PMID: 34660292
|
---
title: Inhibition of ox‐LDL‐induced endothelial cell injury by LINC02381 knockdown
through the microRNA‐491‐5p/transcription factor 7 axis
authors:
- Xizheng Zhu
- Hui Xu
- Beijia Chen
journal: Immunity, Inflammation and Disease
year: 2023
pmcid: PMC10013137
doi: 10.1002/iid3.785
license: CC BY 4.0
---
# Inhibition of ox‐LDL‐induced endothelial cell injury by LINC02381 knockdown through the microRNA‐491‐5p/transcription factor 7 axis
## Abstract
Atherosclerosis (AS) is a complex multifactorial and chronic inflammatory vascular disease that contributes to the development of cardiovascular diseases. Abnormal cellular proliferation in human umbilical vein endothelial cells (HUVECs) is a crucial element in AS development. In this study, we investigated the potential role of the long noncoding RNA LINC02381/microRNA (miR)‐491‐5p/transcription factor 7 (TCF7) axis in regulating HUVEC injury in 30 participants suffering from AS and 30 healthy control participants. We established an in vitro model of AS in HUVECs using oxidized low‐density lipoprotein (ox‐LDL), and measured cellular mRNA and protein levels of LINC02381, miR‐491‐5p, and TCF7 in serum samples using reverse transcription‐quantitative polymerase chain reaction and Western blotting assays. We evaluated cell viability, apoptosis, and inflammation using Cell Counting Kit‐8, flow cytometry, and enzyme‐linked immunosorbent assays, respectively. Moreover, we analyzed apoptosis‐related protein expression using western blotting analysis and determined the association between miR‐491‐5p and LINC02381 or TCF7 using dual‐luciferase reporter assay, RNA pull‐down, and rescue experiments. We observed that LINC02381 was elevated, while miR‐491‐5p was downregulated in serum samples from participants with AS and in ox‐LDL‐treated HUVECs. LINC02381 knockdown was protective against HUVEC injury via miR‐491‐5p inhibition, which is its downstream target. Rescue experiments further demonstrated that miR‐491‐5p alleviated HUVEC injury by modulating TCF7. Thus, LINC02381 knockdown ameliorated HUVEC injury by regulating the miR‐491‐5p/TCF7 axis, which provides new insights into AS treatment strategies.
The present study aimed to unravel the role of lncRNA LINC02381/miR‐491‐5p/TCF7 axis in regulating HUVECs injury. Results indicated that silencing of LINC02381 ameliorated HUVECs injury by regulating miR‐491‐5p/TCF7 axis, shedding new insights for AS treatment.
## INTRODUCTION
According to the Global Burden of Disease institution, cardiovascular diseases have been the leading cause of mortality worldwide for more than 10 consecutive years. 1 Atherosclerosis (AS) is the primary causative pathology of cardiovascular associated diseases, 2 and is a systemic disease characterized by impaired lipid metabolism, intimal lipid deposition, atheromatous plaque formation, fibrous tissue proliferation, and vessel wall sclerosis in large and medium‐sized arteries. 3 The clinical manifestations of AS result in heart disease, 4 ischemic stroke, 5 and peripheral arterial disease. 6 Endothelial cell injury induced by oxidized low‐density lipoproteins (ox‐LDL) could contribute to the onset and development of AS. Human umbilical vein endothelial cells (HUVECs) have been widely utilized for in vitro studies of AS, with ox‐LDL‐induced HUVECs increasingly being used as an in vitro AS model. Previous studies have shown that the atherosclerotic process is closely associated with abnormal cellular proliferation and HUVEC apoptosis, which aggravate the progression of AS. 7, 8 Additional evidence suggests that inflammatory responses exert a crucial role in the pathogenesis of AS. 9 Inflammatory cytokines infiltrate the arterial wall, inducing foam cell formation and HUVEC apoptosis, leading to plaque growth, erosion, and rupture. 10 Research into the association between noncoding RNAs (ncRNAs) and cardiovascular disease has recently gained attention, 11 with several ncRNAs identified as biomarkers of different cardiovascular diseases. 12 ncRNAs, including miRNAs, long noncoding RNAs (lncRNAs), and circRNAs, have the common characteristic of having transcriptional but no protein translation activity and play biological roles at the level of RNAs. 13, 14 miRNAs are the major posttranscriptional gene regulators that regulate mRNA expression by inducing transcript degradation and translation repression. 15 lncRNAs can bind to the target genes regulated by miRNAs, thereby competing with and consequently inhibiting the regulatory effects of miRNAs. 16 Recent studies show that various types of ncRNAs modulate gene expression and play significant roles in the pathophysiology of AS. Li et al. 17 reported that the expression of lncRNA taurine‐upregulated gene 1 (TUG1) was increased, whereas miR‐21 levels were decreased in serum samples of patients with AS and in the atherosclerotic plaques of ApoE−/− mice. Yu et al. 18 observed increased expression of lncRNA kcnq1ot1 and decreased expression of miR‐452‐3p in the aorta of mice with AS and in lipid‐loaded macrophages. Previous studies have reported decreased expression of miR‐491‐5p in atherosclerotic plaque tissues and serum samples from patients with AS. 19, 20 miR‐491‐5p exerts a protective role in AS development by inhibiting the oxidative stress and inflammatory responses of THP‐1 macrophages via MMP‐9 and suppressing the cellular proliferation and migration of vascular smooth muscle cells. 19, 20, 21, 22 A recent study revealed that miR‐491‐5p is associated with the protective effect of circ_0003204 gene silencing on ox‐LDL‐induced HUVECs injury. 23 Thus, miR‐491‐5p may exert a protective role in AS development, however, the mechanisms underlying its actions remain to be identified. We used a bioinformatics approach to identify a potential binding site between LINC02381 and miR‐491‐5p. LINC02381 has been studied in several types of cancers and found to be involved in inflammatory diseases. In addition, LINC02381 has been suggested to act as a miRNA sponge for various miRNAs, including miR‐27b‐3p, miR‐1271‐5p, and miR‐21. 24, 25, 26 However, the role of LINC02381 and its relationship with miR‐491‐5p in the pathophysiology of AS remain unclear. In this study, we investigated whether LINC02381 exerts a role in ox‐LDL‐induced endothelial injury by modulating miR‐491‐5p expression using in vitro experiments with HUVECs.
## Patient recruitment and sample collection
We recruited 30 participants with AS and 30 healthy control participants from the Fifth Hospital in Wuhan, China. Patients with AS were diagnosed using coronary angiography. Exclusion criteria were as follows: severe heart valve disease, cardiac function grade III or higher within 2 weeks after acute myocardial infarction, malignant tumors, severe hematologic diseases, and severe liver and renal failure. The characteristics of AS patients are displayed in Table 1. The study was accredited by the Ethics Committee of the Fifth Hospital in Wuhan (approval number: 2020020601012315) and adhered to the tenets of the Declaration of Helsinki. A written informed consent was signed and obtained from each participant before the study.
**Table 1**
| Parameters | Healthy control | AS patients | p Value |
| --- | --- | --- | --- |
| Male/female | 15/15 | 15/15 | – |
| Age (year) | 51–68 | 52–71 | >.05 |
| Hypertension | 4 (13.3%) | 20 (66.7%) | <.05 |
| Diabetes mellitus (%) | 6 (20.0%) | 7 (23.3%) | >.05 |
| Current smoke (%) | 8 (26.7%) | 7 (23.3%) | >.05 |
| TC (mg/dL) | 190.35 ± 4.01 | 193.22 ± 3.45 | >.05 |
| TG (mg/dL) | 121.62 ± 12.28 | 123.51 ± 13.79 | >.05 |
| HDL (mg/dL) | 47.88 ± 4.33 | 46.32 ± 4.63 | >.05 |
| LDL (mg/dL) | 115.81 ± 7.06 | 116.77 ± 6.59 | >.05 |
| DBP (mm Hg) | 69.83 ± 4.06 | 81.83 ± 5.21 | <.05 |
| CRP (mg/L) | 2.81 ± 1.24 | 9.83 ± 2.21 | <.05 |
| CIMT (mm) | 0.55 ± 0.14 | 1.04 ± 0.11 | <.05 |
After recruitment and fasting for 8 h, 5 mL serum samples were extracted from each participant and stored at −80°C until analysis.
## Cell culture and ox‐LDL treatment
HUVECs were obtained from American Type Culture Collection and cultivated in Dulbecco's Modified Eagle Medium (Invitrogen; Thermo Fisher Scientific Inc.) at 37°C containing $5\%$ CO2, $10\%$ fetal bovine serum, and $1\%$ $1\%$ penicillin and streptomycin.
The cells were divided into control and ox‐LDL groups (0, 50, 100, and 150 μg/mL) and plated into 96‐well plates at a density of 4 × 104 cells/mL. 27 Cells in the ox‐LDL group were exposed to different concentrations of ox‐LDL for 24 h at 37°C with $5\%$ CO2.
## Cellular transfection
TCF7‐plasmid and its negative control (control‐plasmid, no. sc‐36617) were constructed and purchased from Santa Cruz Biotechnology. Control‐siRNAs, LINC02381‐siRNAs, miR‐491‐5p suppressor and its suppressor control, and miR‐491‐5p mimic and its mimic control were purchased from Guangzhou RiboBio Co. Ltd. Control‐siRNA, LINC02381‐siRNA, miR‐491‐5p inhibitor, inhibitor control, LINC02381‐siRNA + inhibitor control, LINC02381‐siRNA + miR‐491‐5p inhibitor, mimic control, miR‐491‐5p mimic, control‐plasmid, TCF7‐plasmid, miR‐491‐5p mimic + control‐plasmid, or miR‐491‐5p mimic + TCF7‐plasmid were transfected into HUVECs using Lipofectamine 2000 reagent (Invitrogen; Thermo Fisher Scientific Inc.), as per the manufacturer's protocol.
## Reverse transcription‐quantitative polymerase chain reaction assay
Serum samples and cells were collected for RNA extraction using the TRIzol Kit (Takara). DNase I (Thermo Fisher Scientific Inc.) was added to the extracted RNA to digest the genomic DNA, followed by cDNA reverse transcription using the TaqMan RNA Kit (Invitrogen; Thermo Fisher Scientific Inc.). The polymerase chain reaction (PCR) was conducted on the ABI 7500 System (Applied Biosystems) with the SYBR Premix Kit (Thermo Fisher Scientific Inc.). Thermal cycles included initial denaturation at 95°C for 5 min, followed by 40 cycles at 95°C for 10 s and at 64°C for 20 s. Results were processed using the 2‐∆∆CT method 28 for relative quantification of gene expression, with GAPDH and U6 genes as the internal references. 18, 29, 30 Primer sequences used were as follows: LINC02381 forward 5′‐CTGATGGCCACTCACGCTAT‐3′ reverse 5′‐GATCCGGAGGGAGAGCATTC‐3′ 29; GAPDH forward 5′‐TCCTGTGGCATCCACGAAACT‐3′;
reverse 5′‐GAAGCATTTGCGGTGGACGAT‐3′ 29; TCF7 forward 5′‐CTGGCTTCTACTCCCTGACCT‐3′; reverse 5′‐ACCAGAACCTAGCATCAAGGA‐3′; miR‐491‐5p forward 5′‐GGAGTGGGGAACCCTTCC‐3′;
reverse 5′‐GTGCAGGGTCCGAGGT‐3′ 30; U6 forward 5′‐CTCGCTTCGGCAGCACA‐3′; reverse 5′‐AACGCTTCACGAATTTGCGT‐3′. 30
## Enzyme‐linked immunosorbent assay assay
The expression levels of tumor necrosis factor‐alpha (TNF‐α), interleukin (IL)‐6, and IL‐1β in the cell supernatants were evaluated using their associated enzyme‐linked immunosorbent assay (ELISA) kits (TNF‐α, #7355; IL‐6, #8904; IL‐1β #8900; Cell Signaling Technology). 31
## Cell Counting Kit‐8 assay to evaluate cell viability
HUVECs in the logarithmic growth phase were harvested and counted after digestion with $0.25\%$ trypsin in 5 × 104 cells/mL cellular suspension samples. Approximately 100 μL of the cell suspension was plated in 96‐well plates and incubated at 37°C in a $5\%$ CO2 incubator. At 0, 24, 48, and 72 h, 10 μL of Cell Counting Kit‐8 solution was supplemented into each well and incubated for another 2 h at room temperature. Eventually, an enzyme marker was used to measure the sample absorbances. 32
## Flow cytometry assay for cell apoptosis
The transfected cells were collected by cellular digestion using $0.25\%$ trypsin without EDTA. Cells were removed into a centrifuge tube for centrifugation at 1000 × g for 5 min at 4°C. A binding buffer was used to resuspend the cells, which were diluted to a cellular concentration of 1 × 106 cells/mL. Subsequently, 5 μL of Annexin V‐FITC and 5 μL of propidium iodide at a concentration of 20 μg/mL were added to each well and incubated for 15 min at 25°C in the dark. Stained HUVECs were measured using flow cytometry (BD Bioscience) and Kaluza Analysis (version 2.1.1.20653; Beckman Coulter Inc.). 33
## Western blot assay
After cell treatments, 100 μL of radioimmunoprecipitation assay buffer containing 1 μmol/L phenylmethanesulfonyl fluoride was added to the treated cells and incubated on ice for 20 min. The cell lysate was transferred to a microcentrifuge tube and centrifuged at 12,000 rpm for 10 min, and the supernatant was removed for protein concentration calculation using the BCA method. SDS‐gel electrophoresis was performed, and the gel was subsequently placed in the electrotransfer solution and equilibrated for 15 min. Protein samples were then transferred to polyvinylidene fluoride membranes using the wet transfer method. Membranes were blocked in $5\%$ skim milk for 2 h and incubated with primary antibodies (cleaved‐caspase‐3, 1:1000, ab2302; caspase‐3, 1:1000, ab32351; GAPDH, 1:2000, ab9485; Abcam) overnight at 4°C. The membranes were then washed three times with TBST buffer at 25°C and incubated with secondary antibodies (goat anti‐rabbit IgG H&L [HRP] preadsorbed, 1:1000, ab7090; Abcam) for 2 h at 25°C. ECL solution was used for luminescence. The Western blot bands were scanned in grayscale using ImageJ software. 34
## Dual‐luciferase reporter assay
The whole length of LINC02381 and TCF7 was designed and obtained using PCR amplification and inserted in the psiCHECK‐2 vector to generate wild‐type LINC02381 (LINC02381‐WT) and wide‐type TCF7 (TCF7‐WT), whereas mutant LINC02381 (LINC02381‐MUT) and TCF7 (TCF7‐MUT) were engineered and purchased from GenePharma. The plasmids were incubated and transfected with miR‐491‐5p mimic or mimic control into HUVECs using Lipofectamine 2000 reagent (Invitrogen; Thermo Fisher Scientific Inc.), as per the manufacturer's instructions. After 24 h of transfection, Dual Luciferase Assay Kit was utilized to measure luciferase activity (Zeye Inc.). 35
## RNA pull‐down assay
Cells at a concentration of 1 × 107 cells were collected and lysed using an ultrasonic processor for 3 min. The LINC02381 probe was cotreated with magnetic beads at 25°C for 2 h and incubated with the cell lysate overnight at 4°C. After washing with elution buffer, the RNA complex bound to the magnetic beads was eluted, the RNA was extracted, and samples underwent qRT‐PCR to evaluate the miR‐491‐5p expression. 36
## Statistical analysis
Statistical analysis was conducted using SPSS version 19.0, and statistical graphs were generated using GraphPad Prism 5.0. Data were expressed using means ± standard deviation. Unpaired, two‐tailed Student's t‐test was used for intergroup comparisons, and one‐way analysis of variance followed by Tukey's test was used for the comparison of multiple samples. For all analyses, a $p \leq .05$ indicated statistical significance.
## Aberrant expressions of LINC02381 and miR‐491‐5p in serum samples from patients with AS
We determined the expression levels of LINC02381 and miR‐491‐5p in serum samples from 30 participants with AS and 30 corresponding healthy participants using RT‐qPCR. As displayed in Figure 1A, LINC02381 levels were significantly elevated in participants with AS compared to healthy controls. Contrarily, miR‐491‐5p was significantly decreased in patients with AS in comparison with healthy participants (Figure 1B).
**Figure 1:** *Expression of LINC02381 and miR‐491‐5p in serum samples of patients with atherosclerosis (AS) (n = 30). (A) LINC02381 levels detected by reverse transcription‐quantitative polymerase chain reaction (RT‐qPCR) in serum samples from 30 patients with AS and 30 healthy controls. (B) miR‐491‐5p levels detected by RT‐qPCR in serum samples from 30 patients with AS and 30 healthy controls. **p < .01 versus Healthy control.*
## Targeted association between LINC02381 and miR‐491‐5p
We confirmed the association between LINC02381 and miR‐491‐5p using dual‐luciferase reporter and RNA pull‐down assays. We used the starBase online platform to predict the putative binding sequences between LINC02381 and miR‐491‐5p (Figure 2A). The dual‐luciferase reporter assay verified that cotransfection with WT‐LINC02381 and miR‐491‐5p mimic prominently decreased luciferase activity compared to that with WT‐LINC02381 and mimic control (Figure 2B). Nevertheless, no prominent variations were found between the Mut‐LINC02381 transfection groups. Furthermore, the RNA pull‐down assay confirmed that miR‐491‐5p was abundantly enriched in the LINC02381 probe compared to the input group, suggesting that this probe can specifically bind with miR‐491‐5p (Figure 2C).
**Figure 2:** *Targeted relationship between miR‐491‐5p and LINC02381. (A) starBase tool predicted the binding sites between miR‐491‐5p and LINC02381. (B) Dual‐luciferase reporter assay (n = 3). (C) RNA pull‐down assay (n = 3). **p < 0.01 versus mimic control; ##
p < 0.01 versus Oligo probe.*
## Effects of ox‐LDL treatment on LINC02381 and miR‐491‐5p expressions in HUVECs
We investigated the effects of treatment with different doses of ox‐LDL for 24 h (0, 50, 100, and 150 μg/mL) on LINC02381 and miR‐491‐5p levels in HUVECs using RT‐qPCR analysis. We observed that LINC02381 levels gradually increased with increasing doses of ox‐LDL (Figure 3A), whereas miR‐491‐5p levels were downregulated with increasing doses of ox‐LDL in a dose‐independent manner (Figure 3B).
**Figure 3:** *Effects of aberrant LINC02381 levels on miR‐491‐5p expression in human umbilical vein endothelial cells (HUVECs). HUVECs were exposed to different concentrations (0, 50, 100, and 150 μg/mL) of oxidized low‐density lipoprotein (ox‐LDL) for 24 h. (A) Expression of LINC02381 was determined using reverse transcription‐quantitative polymerase chain reaction (RT‐qPCR) assay (n = 3). (B) Expression of miR‐491‐5p was determined using RT‐qPCR assay (n = 3). *p < .05 versus Control (0 μg/mL ox‐LDL); **p < .01 versus Control (0 μg/mL ox‐LDL).*
## Effects of LINC02381 on miR‐491‐5p in HUVECs
We transfected LINC02381 and miR‐491‐5p siRNA and inhibitor, respectively, in HUVECs to elucidate the relationship between them. As demonstrated in Figure 4A, LINC02381 expression was significantly downregulated after transfection in the LINC02381‐siRNA group, in comparison with the control‐siRNA group. Furthermore, the transfection efficacy experiment in Figure 4B illustrated that the miR‐491‐5p inhibitor successfully downregulated miR‐491‐5p expression compared with the inhibitor control group. Finally, LINC02381‐siRNA significantly upregulated miR‐491‐5p expression in HUVECs, whereas a partial enhancement was observed after cotransfection with the miR‐491‐5p inhibitor (Figure 4C).
**Figure 4:** *Regulatory mechanisms of LINC02381 on miR‐491‐5p expression in human umbilical vein endothelial cells (HUVECs). (A) Transfection efficiency of LINC02381‐siRNA in HUVECs (n = 3). (B) Transfection efficiency of miR‐491‐5p inhibitor in HUVECs (n = 3). (C) Expression of miR‐491‐5p in HUVECs after transfection with LINC02381‐siRNA and/or miR‐491‐5p inhibitor (n = 3). **p < .01 versus Control‐siRNA; ##
p < .01 versus inhibitor control; &&
p < .01 versus LINC02381‐siRNA + inhibitor control.*
## LINC02381 knockdown is protective against ox‐LDL‐induced HUVEC injury due to miR‐491‐5p upregulation
We used LINC02381‐siRNA to investigate the role of LINC02381 knockdown in ox‐LDL‐induced HUVEC injury. Ox‐LDL‐induced endothelial cell injury, evidenced by reduced cell proliferation and enhanced cell apoptosis and inflammatory response, contributes to the onset and development of AS. Caspase‐3 is the most important terminal cleavage enzyme in the process of apoptosis. Since caspase‐3 is activated by proteolytic cleavage, we performed a Western blot assay to detect total caspase‐3 and cleaved‐caspase‐3 levels. We detected the levels of the proinflammatory cytokines, TNF‐α, IL‐6, and IL‐1β, which are involved in accelerating AS progression, using ELISA. We observed that treatment with ox‐LDL significantly impaired HUVEC cell viability and increased apoptosis, cleaved‐caspase‐3 levels, and proinflammatory cytokine secretion, including TNF‐α, IL‐6, and IL‐1β (Figure 5A–F). Moreover, knockdown of LINC02381 also resulted in ox‐LDL‐induced cell viability reduction and apoptosis and inflammation response enhancement compared with the ox‐LDL + control‐siRNA group. Moreover, the therapeutic role of LINC02381 knockdown on ox‐LDL‐induced HUVEC injury could be partially restored with miR‐491‐5p inhibitor cotransfection.
**Figure 5:** *LINC02381 knockdown alleviated oxidized low‐density lipoprotein (ox‐LDL)‐induced human umbilical vein endothelial cell injury by regulating miR‐491‐5p. (A) Cell Counting Kit‐8 assay determined cell viability. (B, C) Cell apoptosis was quantified using flow cytometry assay. (D) Western blot assay detected cleaved‐caspase‐3 and caspase‐3 expressions. (E) Quantified results of cleaved‐caspase‐3/caspase‐3. (F) tumor necrosis factor‐α, interleukin‐6 (IL‐6), and IL‐1β levels were detected using enzyme‐linked immunosorbent assay assay. n = 3; **p < .01 versus Control; ##
p < .01 versus ox‐LDL + control‐siRNA; &&
p < .01 versus ox‐LDL + LINC02381‐siRNA + inhibitor control.*
## TCF7 is a downstream target and negative regulator of miR‐491‐5p
We used multiple prediction software including PITA, miRmap, microT, and TargetScan version 7.2 (https://www.targetscan.org/vert_80/) to predict the potential targets of miR‐491‐5p. According to TargetScan software prediction results, miR‐491‐5p has hundreds of target genes including TCF7, which is reported to be significantly upregulated in atherosclerotic patient serum as well as in ox‐LDL‐induced HUVECs and is involved in the regulation of ox‐LDL‐induced HUVEC injury. 36 Thus, we hypothesized that miR‐491‐5p may play a role in AS by regulating the expression of TCF7. Therefore, we selected TCF7 for further analysis. The putative binding sequences between miR‐491‐5p and TCF7 are presented in Figure 6A. The dual‐luciferase reporter assay was used to verify if cotransfection with WT‐TCF7 and miR‐491‐5p mimic reduced luciferase activity compared to that with WT‐TCF7 and the mimic control (Figure 6B). We did not observe any pronounced variation in the Mut‐TCF7 transfection groups. We further verified the regulatory mechanisms between miR‐491‐5p and TCF7 using rescue experiments. As shown in Figure 6C, miR‐491‐5p mimic significantly increased miR‐491‐5p expression levels in HUVECs. Meanwhile, transfection with TCF7‐plasmid increased TCF7 levels in HUVECs, suggesting a successful transfection (Figure 6D). In addition, RT‐qPCR and Western blotting assays revealed that the miR‐491‐5p mimic significantly suppressed mRNA and protein levels of TCF7, which was reversed with cotransfection of TCF7 plasmid (Figure 6E,F).
**Figure 6:** *Targeted relationship between TCF7 and miR‐491‐5p. (A) TargetScan tool predicted the binding sequences between TCF7 and miR‐491‐5p. (B) Dual‐luciferase reporter assay confirmed the targeted relationship between miR‐491‐5p and TCF7 (n = 3). (C) Transfection efficacy of miR‐491‐5p mimic in human umbilical vein endothelial cells (HUVECs). (D) Transfection efficacy of TCF7 overexpression in HUVECs (n = 3). (E) mRNA expression of TCF7 after cotransfection with TCF7 plasmid and miR‐491‐5p mimic (n = 3). (F) Protein expression of TCF7 after cotransfection with TCF7‐plasmid and miR‐491‐5p mimic (n = 3). **p < .01 versus mimic control; ##
p < .01 versus Control plasmid; &&
p < .01 versus miR‐491‐5p mimic + control‐plasmid.*
## Expression of TCF7 in serum samples from patients with AS and in ox‐LDL‐induced HUVECs
We determined the expression of TCF7 in serum samples from 30 participants with AS and in the ox‐LDL‐induced HUVECs model using RT‐qPCR. As displayed in Figure 7A, TCF7 mRNA levels were significantly upregulated in patients with AS compared to healthy controls. Moreover, the mRNA levels of TCF7 were also increased in ox‐LDL‐induced HUVECs in comparison with the control group (Figure 7B).
**Figure 7:** *Expression of TCF7 mRNA in serum samples of patients with atherosclerosis (AS) and in oxidized low‐density lipoprotein (ox‐LDL)‐induced human umbilical vein endothelial cells (HUVECs). (A) TCF7 mRNA levels detected by reverse transcription‐quantitative polymerase chain reaction (RT‐qPCR) in serum samples from 30 patients with AS and 30 healthy controls (n = 30). (B) TCF7 mRNA levels detected by RT‐qPCR in HUVECs exposed to different concentrations (0, 50, 100, and 150 μg/mL) of ox‐LDL for 24 h (n = 3). **p < .01 versus Healthy control; ##
p < .01 versus Control (0 μg/mL ox‐LDL).*
## miR‐491‐5p alleviates ox‐LDL‐triggered‐HUVEC injury by regulating TCF7 expression
We cotransfected ox‐LDL‐treated HUVECs with miR‐491‐5p mimic and TCF7‐plasmid to investigate the effects of miR‐491‐5p and TCF7 on HUVEC injury by measuring cellular viability, apoptosis, and inflammatory response. As displayed in Figure 8A, ox‐LDL‐triggered reduction in cell viability was suppressed with the miR‐491‐5p mimic; however, this protective effect was antagonized by TCF7 overexpression. Moreover, ox‐LDL‐induced cellular apoptosis and elevated cleaved‐caspase‐3 levels were decreased with miR‐491‐5p overexpression (Figure 8B–E), whereas the reduction of cell apoptosis by miR‐491‐5p mimic was counteracted with the TCF7‐plasmid (Figure 8B–E). Furthermore, although ox‐LDL‐induced inflammatory response (TNF‐α, IL‐6, and IL‐1β) was inhibited by the miR‐491‐5p mimic (Figure 8F), these effects were reversed by treatment with TCF7 plasmid.
**Figure 8:** *miR‐491‐5p protected against oxidized low‐density lipoprotein (ox‐LDL)‐induced human umbilical vein endothelial cell (HUVEC) injury by inhibiting TCF7 expression. HUVECs cells were exposed to 100 μg/mL ox‐LDL for 24 h and transfected with miR‐491‐5p mimic and/or TCF7 plasmid. (A) Cell viability was evaluated using Cell Counting Kit‐8 assay. (B, C) Cell apoptosis was determined using flow cytometry analysis. (D) Protein expressions of cleaved‐caspase‐3 and caspase‐3 were confirmed by Western blot assay. (E) Cleaved‐caspase‐3/caspase‐3 ratio was quantified. (F) Tumor necrosis factor‐α, interleukin‐6 (IL‐6), and IL‐1β levels were detected using enzyme‐linked immunosorbent assay assay. n = 3; **p < .01 versus Control; ##
p < .01 versus ox‐LDL + mimic control; &&
p < .01 versus ox‐LDL + miR‐491‐5p mimic + control‐plasmid.*
## DISCUSSION
AS is a major contributor to coronary artery disease, stroke, cerebral infarction, and peripheral vascular disease, making it one of the most prevalent signs of global morbidity and mortality. 37 Several factors have been reported to induce AS, and recent research has identified new causes of AS, including endothelial cell damage, 38 lipid metabolism, 39 intestinal microbiota dysbiosis, 40 and Chlamydia pneumoniae. 41 Treatment with ox‐LDL stimulates vascular cells to secrete inflammatory molecules and accumulate monocytes and T cells in the arterial wall. 42 As monocytes in the vessel wall are precursors to lipid macrophages, they establish fatty streaks, which are an anatomical feature of early AS. 43, 44 *In this* study, we successfully established an in vitro model of AS using HUVECs treated with ox‐LDL for 24 h, which resulted in significantly decreased cell viability and enhanced cellular apoptosis and inflammation, consistent with previous studies. 45, 46 Acting as regulatory noncoding RNAs, lncRNAs have transcript lengths >200 nt that have no protein translation functions. lncRNAs were initially considered to be genomic transcriptional junk without any biological functions. However, recent reports have demonstrated that lncRNAs are strongly correlated with cardiovascular diseases, including AS. A study reported significant downregulation of the lncRNA, FAF, in patients with coronary heart disease, revealing a negative correlation with independent risk factors of coronary heart disease. 47 Feng et al. 48 similarly reported that lncRNA DCRF knockdown promoted cardiac function, suppressed cellular autophagy in cardiomyocytes, and consequently alleviated diabetic cardiomyopathy. Wu et al. 49 identified six aberrantly expressed lncRNAs in AS using RNA sequencing. Furthermore, differentially expressed lncRNAs could identify underlying novel targets for the therapeutic diagnosis and treatment of AS. For instance, Bai et al. 50 utilized microarray profiling analysis and identified 236 differentially expressed lncRNAs in AS, contributing to the diagnosis and treatment of AS. Meng et al. 51 illustrated the clinical significance of the lncRNA, APPAT, in the diagnosis and progression of AS. Notably, LINC02381 was a newly identified tumor‐promoting lncRNA that was observed in various tumorigenesis processes, including osteosarcomas, 52 gliomas, 53 and cervical cancer. 29 In contrast, LINC02381 expression was reduced in gastric cancers 54 and colorectal carcinoma. 55 In addition, LINC02381 was significantly upregulated in chronic autoimmune inflammatory diseases, accelerating rheumatoid arthritis development. 56 In our study, we observed significantly increased levels of LINC02381 in AS serum samples and in ox‐LDL‐treated HUVECs. LINC02381 knockdown resulted in increased cell viability and decreased apoptosis and proinflammatory cytokines in ox‐LDL‐treated HUVECs. Thus, LINC02381 downregulation is protective against ox‐LDL‐induced injury in HUVECs, thereby alleviating AS progression.
miRNAs are a class of endogenous ncRNAs approximately 22 nt in length. They are highly evolutionarily conserved and regulate gene expression at the posttranscriptional level by binding to the 3′‐untranslated region (3′‐UTR) of target genes to degrade their mRNA or inhibit their translation. 57 Several reports have demonstrated the role and biological significance of miR‐491‐5p in diabetes and AS. Sidorkiewicz et al. 58 demonstrated that miR‐491‐5p was a putative diagnostic biomarker for type 2 diabetes mellitus, with an area under the curve of $94.0\%$. Another study illustrated that miR‐491‐5p was decreased in the tissues and plasma samples of participants with AS, indicating its protective role. 19 In our study, dual‐luciferase reporter, RNA pull‐down, and rescue experiments confirmed the targeted association between LINC02381 and miR‐491‐5p. Moreover, we found prominently decreased miR‐491‐5p levels in AS serum samples and in ox‐LDL‐induced HUVECs. Overexpression of miR‐491‐5p was protective against the ox‐LDL‐induced HUVEC injury model by improving cellular viability and suppressing cellular apoptosis and inflammation response. Moreover, our findings demonstrated that LINC02381 knockdown could mitigate ox‐LDL‐triggered HUVEC injury by upregulating miR‐491‐5p. This is further supported by the finding that lncRNAs act as sponges to regulate miRNAs. 59 Since miRNAs regulate cellular processes by binding to the 3′‐UTR of target genes, the TargetScan tool was utilized to predict the associated targets of miR‐491‐5p. Among them, TCF7 was screened as a candidate, and the targeted relationship between miR‐491‐5p and TCF7 was validated using the dual‐luciferase reporter assay. TCF7 has been reported to be involved in cardiovascular diseases, such as cardiac hypertrophy and acute coronary syndrome. 60, 61, 62 A recent study showed that TCF7 is highly expressed in immune cells on atherosclerotic plaques, and regulates inflammatory signaling via the NFκB/AKT/STAT1 pathway. 63 Moreover, in a previous study, TCF7 was demonstrated to be overexpressed in AS, in which TCF7 knockdown was protective against AS. 36 Similarly, in our study, we found that TCF7 mRNA levels were significantly upregulated in patients with AS, and the rescue experiments verified that the protective effects of miR‐491‐5p on AS progression could be partially suppressed by overexpressing TCF7.
There were some limitations in our experimental design and other aspects. For instance, the role of LINC02381 in ox‐LDL‐induced endothelial cell injury was not analyzed in AS animal models. Besides, whether the TCF7 gene is associated with the effect of LINC02381 on the ox‐LDL‐induced HUVECs injury remains to be clarified. Moreover, whether LINC02381 plays a role in predicting the prognosis of AS patients require further analysis. These issues will be addressed in our future studies.
## CONCLUSION
LINC02381 knockdown enhanced cellular proliferation and suppressed the apoptosis and inflammatory responses of ox‐LDL‐treated HUVECs by regulating the miR‐491‐5p/TCF7 axis, consequently mitigating AS progression. Therefore, LINC02381 could be developed as a novel clinical target for AS therapy.
## AUTHOR CONTRIBUTIONS
Xizheng Zhu contributed to the study design, data collection, statistical analysis, data interpretation, and manuscript preparation. Hui Xu contributed to data collection and statistical analysis. Beijia Chen contributed to data collection, statistical analysis, and manuscript preparation. All authors read and approved the final manuscript.
## CONFLICTS OF INTEREST STATEMENT
The authors declare no conflict of interest.
## DATA AVAILABILITY STATEMENT
Datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.
## References
1. Wong CX, Brown A, Lau DH. **Epidemiology of sudden cardiac death: global and regional perspectives**. *Heart Lung Circ* (2019) **28** 6-14. PMID: 30482683
2. Soliman GA. **Dietary fiber, atherosclerosis, and cardiovascular disease**. *Nutrients* (2019) **11** 1155. PMID: 31126110
3. Mitra R, O'Neil GL, Harding IC, Cheng MJ, Mensah SA, Ebong EE. **Glycocalyx in atherosclerosis‐relevant endothelium function and as a therapeutic target**. *Curr Atheroscler Rep* (2017) **19** 63. PMID: 29127504
4. Potz BA, Parulkar AB, Abid RM, Sodha NR, Sellke FW. **Novel molecular targets for coronary angiogenesis and ischemic heart disease**. *Coron Artery Dis* (2017) **28** 605-613. PMID: 28678145
5. Amarenco P, Denison H, Evans SR. **Ticagrelor added to aspirin in acute nonsevere ischemic stroke or transient ischemic attack of atherosclerotic origin**. *Stroke* (2020) **51** 3504-3513. PMID: 33198608
6. Firnhaber JM, Powell CS. **Lower extremity peripheral artery disease: diagnosis and treatment**. *Am Fam Physician* (2019) **99** 362-369. PMID: 30874413
7. Pan JX. **LncRNA H19 promotes atherosclerosis by regulating MAPK and NF‐kB signaling pathway**. *Eur Rev Med Pharmacol Sci* (2017) **21** 322-328. PMID: 28165553
8. Chen L, Zheng SY, Yang CQ, Ma BM, Jiang D. **MiR‐155‐5p inhibits the proliferation and migration of VSMCs and HUVECs in atherosclerosis by targeting AKT1**. *Eur Rev Med Pharmacol Sci* (2019) **23** 2223-2233. PMID: 30915770
9. Zhu Y, Xian X, Wang Z. **Research progress on the relationship between atherosclerosis and inflammation**. *Biomolecules* (2018) **8** 80. PMID: 30142970
10. Parma L, Baganha F, Quax PHA, de Vries MR. **Plaque angiogenesis and intraplaque hemorrhage in atherosclerosis**. *Eur J Pharmacol* (2017) **816** 107-115. PMID: 28435093
11. Lucas T, Bonauer A, Dimmeler S. **RNA therapeutics in cardiovascular disease**. *Circ Res* (2018) **123** 205-220. PMID: 29976688
12. Poller W, Dimmeler S, Heymans S. **Non‐coding RNAs in cardiovascular diseases: diagnostic and therapeutic perspectives**. *Eur Heart J* (2018) **39** 2704-2716. PMID: 28430919
13. Anastasiadou E, Jacob LS, Slack FJ. **Non‐coding RNA networks in cancer**. *Nat Rev Cancer* (2018) **18** 5-18. PMID: 29170536
14. Matsui M, Corey DR. **Non‐coding RNAs as drug targets**. *Nat Rev Drug Discovery* (2017) **16** 167-179. PMID: 27444227
15. Correia de Sousa M, Gjorgjieva M, Dolicka D, Sobolewski C, Foti M. **Deciphering miRNAs’ action through miRNA editing**. *Int J Mol Sci* (2019) **20** 6249. PMID: 31835747
16. Zhou RS, Zhang EX, Sun QF. **Integrated analysis of lncRNA‐miRNA‐mRNA ceRNA network in squamous cell carcinoma of tongue**. *BMC Cancer* (2019) **19** 779. PMID: 31391008
17. Li FP, Lin DQ, Gao LY. **LncRNA TUG1 promotes proliferation of vascular smooth muscle cell and atherosclerosis through regulating miRNA‐21/PTEN axis**. *Eur Rev Med Pharmacol Sci* (2018) **22** 7439-7447. PMID: 30468492
18. Yu XH, Deng WY, Chen JJ. **LncRNA kcnq1ot1 promotes lipid accumulation and accelerates atherosclerosis via functioning as a ceRNA through the miR‐452‐3p/HDAC3/ABCA1 axis**. *Cell Death Dis* (2020) **11** 1043. PMID: 33293505
19. He Z, Wang Y, He Q, Chen M. **microRNA‐491‐5p protects against atherosclerosis by targeting matrix metallopeptidase‐9**. *Open Med* (2020) **15** 492-500
20. Ding H, Pan Q, Qian L, Hu C. **Differentially expressed mRNAs and their upstream miR‐491‐5p in patients with coronary atherosclerosis as well as the function of miR‐491‐5p in vascular smooth muscle cells**. *Korean J Physiol Pharmacol* (2022) **26** 183-193. PMID: 35477546
21. Liao Y, Zhu E, Zhou W. **Ox‐LDL aggravates the oxidative stress and inflammatory responses of THP‐1 macrophages by reducing the inhibition effect of miR‐491‐5p on MMP‐9**. *Front Cardiovasc Med* (2021) **8**. PMID: 34660709
22. Peng H, Liu S, Li Y, Wang C, Zhong Y. **A novel circUBR4/miR‐491‐5p/NRP2 ceRNA network regulates oxidized low‐density lipoprotein‐induced proliferation and migration in vascular smooth muscle cells**. *J Cardiovasc Pharmacol* (2022) **79** 512-522. PMID: 34935701
23. Zhang D, Zhang G, Yu K, Zhang X, Jiang A. **Circ_0003204 knockdown protects endothelial cells against oxidized low‐density lipoprotein‐induced injuries by targeting the miR‐491‐5p‐ICAM1 pathway**. *J Thromb Thrombolysis* (2022) **53** 302-312. PMID: 34797473
24. Wang X, Wu P, Zeng C. **Long intergenic non‐protein coding RNA 02381 promotes the proliferation and invasion of ovarian endometrial stromal cells through the miR‐27b‐3p/CTNNB1 axis**. *Genes* (2022) **13** 433. PMID: 35327987
25. Huang S, Huang P, Wu H, Wang S, Liu G. **LINC02381 aggravates breast cancer through the miR‐1271‐5p/FN1 axis to activate PI3K/AKT pathway**. *Mol Carcinog* (2022) **61** 346-358. PMID: 34882856
26. Zhao G, Luo WD, Yuan Y. **LINC02381, a sponge of miR‐21, weakens osteogenic differentiation of hUC‐MSCs through KLF12‐mediated Wnt4 transcriptional repression**. *J Bone Miner Metab* (2022) **40** 66-80. PMID: 34778905
27. Zhu Z, Li J, Zhang X. **Salidroside protects against ox‐LDL‐induced endothelial injury by enhancing autophagy mediated by SIRT1‐FoxO1 pathway**. *BMC Complement Altern Med* (2019) **19** 111. PMID: 31146723
28. Livak KJ, Schmittgen TD. **Analysis of relative gene expression data using real‐time quantitative PCR and the**. *Methods* (2001) **25** 402-408. PMID: 11846609
29. Chen XH, Zhang ZX, Ma Y, Su HX, Xie P, Ran JT. **LINC02381 promoted cell viability and migration via targeting miR‐133b in cervical cancer cells**. *Cancer Manag Res* (2020) **12** 3971-3979. PMID: 32547232
30. Chen T, Li Y, Cao W, Liu Y. **miR‐491‐5p inhibits osteosarcoma cell proliferation by targeting PKM2**. *Oncol Lett* (2018) **16** 6472-6478. PMID: 30405785
31. Konstantinou GN, Lin J, Alcocer M. *Food Allergens: Methods in Molecular Biology* (2017) **1592** 79-94
32. Cai T, Zhou J, Zeng Y. **EVI5 is an oncogene that regulates the proliferation and metastasis of NSCLC cells**. *J Exp Clin Cancer Res* (2020) **39** 84. PMID: 32393392
33. Telford WG, Hawley T, Hawley R. *Flow Cytometry Protocols: Methods in Molecular Biology* (2018) **1678** 167-202
34. Kim B, Espina V. *Molecular Profiling: Methods in Molecular Biology* (2017) **1606** 133-139
35. Chen B, Wang H, Lv C, Mao C, Cui Y. **Long non‐coding RNA H19 protects against intracerebral hemorrhage injuries via regulating microRNA‐106b‐5p/acyl‐CoA synthetase long chain family member 4 axis**. *Bioengineered* (2021) **12** 4004-4015. PMID: 34288826
36. Guo JT, Wang L, Yu HB. **Knockdown of NEAT1 mitigates ox‐LDL‐induced injury in human umbilical vein endothelial cells via miR‐30c‐5p/TCF7 axis**. *Eur Rev Med Pharmacol Sci* (2020) **24** 9633-9644. PMID: 33015807
37. Frostegård J. **Immunity, atherosclerosis and cardiovascular disease**. *BMC Med* (2013) **11** 117. PMID: 23635324
38. Yamaoka‐Tojo M. **Endothelial glycocalyx damage as a systemic inflammatory microvascular endotheliopathy in COVID‐19**. *Biomed J* (2020) **43** 399-413. PMID: 33032965
39. Gupta M, Blumenthal C, Chatterjee S. **Novel emerging therapies in atherosclerosis targeting lipid metabolism**. *Expert Opin Invest Drugs* (2020) **29** 611-622
40. Sanchez‐Rodriguez E, Egea‐Zorrilla A, Plaza‐Díaz J. **The gut microbiota and its implication in the development of atherosclerosis and related cardiovascular diseases**. *Nutrients* (2020) **12** 605. PMID: 32110880
41. Filardo S, Di Pietro M, Farcomeni A, Schiavoni G, Sessa R. *Mediators Inflamm* (2015) **2015**. PMID: 26346892
42. Liu H, Wang S, Sun A. **Danhong inhibits oxidized low‐density lipoprotein‐induced immune maturation of dentritic cells via a peroxisome proliferator activated receptor**. *J Pharmacol Sci* (2012) **119** 1-9. PMID: 22739234
43. Kuznetsova T, Prange KHM, Glass CK, de Winther MPJ. **Transcriptional and epigenetic regulation of macrophages in atherosclerosis**. *Nat Rev Cardiol* (2020) **17** 216-228. PMID: 31578516
44. Tabas I, Lichtman AH. **Monocyte‐macrophages and T cells in atherosclerosis**. *Immunity* (2017) **47** 621-634. PMID: 29045897
45. Gong L, Lei Y, Liu Y. **Vaccarin prevents ox‐LDL‐induced HUVEC EndMT, inflammation and apoptosis by suppressing ROS/p38 MAPK signaling**. *Am J Transl Res* (2019) **11** 2140-2154. PMID: 31105824
46. Bian W, Jing X, Yang Z. **Downregulation of LncRNA NORAD promotes ox‐LDL‐induced vascular endothelial cell injury and atherosclerosis**. *Aging* (2020) **12** 6385-6400. PMID: 32267831
47. Xu H, Zhang X, Yu K, Zhang G, Shi Y, Jiang Y. **Analysis on the expression and prognostic value of LncRNA FAF in patients with coronary heart disease**. *BioMed Res Int* (2020) **2020**. PMID: 33204729
48. Feng Y, Xu W, Zhang W, Wang W, Liu T, Zhou X. **LncRNA DCRF regulates cardiomyocyte autophagy by targeting miR‐551b‐5p in diabetic cardiomyopathy**. *Theranostics* (2019) **9** 4558-4566. PMID: 31285779
49. Wu Y, Zhang F, Li X. **Systematic analysis of lncRNA expression profiles and atherosclerosis‐associated lncRNA‐mRNA network revealing functional lncRNAs in carotid atherosclerotic rabbit models**. *Funct Integr Genomics* (2020) **20** 103-115. PMID: 31392586
50. Bai HL, Lu ZF, Zhao JJ. **Microarray profiling analysis and validation of novel long noncoding RNAs and mRNAs as potential biomarkers and their functions in atherosclerosis**. *Physiol Genomics* (2019) **51** 644-656. PMID: 31682178
51. Meng F, Yan J, Ma Q. **Expression status and clinical significance of lncRNA APPAT in the progression of atherosclerosis**. *PeerJ* (2018) **6**. PMID: 29372117
52. Bian X, Sun YM, Wang LM, Shang YL. **ELK1‐induced upregulation lncRNA LINC02381 accelerates the osteosarcoma tumorigenesis through targeting CDCA4 via sponging miR‐503‐5p**. *Biochem Biophys Res Commun* (2021) **548** 112-119. PMID: 33640603
53. Sun Y, Wang XJ, Bu XY. **LINC02381 contributes to cell proliferation and hinders cell apoptosis in glioma by transcriptionally enhancing CBX5**. *Brain Res Bull* (2021) **176** 121-129. PMID: 34274429
54. Jafarzadeh M, Soltani BM. **Long noncoding RNA LOC400043 (LINC02381) inhibits gastric cancer progression through regulating Wnt signaling pathway**. *Front Oncol* (2020) **10**. PMID: 33194632
55. Jafarzadeh M, Soltani BM, Soleimani M, Hosseinkhani S. **Epigenetically silenced LINC02381 functions as a tumor suppressor by regulating PI3K‐Akt signaling pathway**. *Biochimie* (2020) **171‐172** 63-71
56. Wang J, Zhao Q. **Linc02381 exacerbates rheumatoid arthritis through adsorbing miR‐590‐5p and activating the mitogen‐activated protein kinase signaling pathway in rheumatoid arthritis‐fibroblast‐like synoviocytes**. *Cell Transplant* (2020) **29**
57. Luo G, He K, Xia Z, Liu S, Liu H, Xiang G. **Regulation of microRNA‐497 expression in human cancer**. *Oncol Lett* (2021) **21** 23. PMID: 33240429
58. Sidorkiewicz I, Niemira M, Maliszewska K. **Circulating miRNAs as a predictive biomarker of the progression from prediabetes to diabetes: outcomes of a 5‐year prospective observational study**. *J Clin Med* (2020) **9** 2184. PMID: 32664305
59. Zhang ZK, Li J, Guan D. **A newly identified lncRNA MAR1 acts as a miR‐487b sponge to promote skeletal muscle differentiation and regeneration**. *J Cachexia Sarcopenia Muscle* (2018) **9** 613-626. PMID: 29512357
60. Wang Y, Lifshitz L, Gellatly K. **HIV‐1‐induced cytokines deplete homeostatic innate lymphoid cells and expand TCF7‐dependent memory NK cells**. *Nature Immunol* (2020) **21** 274-286. PMID: 32066947
61. Ren J, Qi H, Song C. **LncRNA4930473A02Rik promotes cardiac hypertrophy by regulating TCF7 via sponging miR‐135a in mice**. *Cell Death Discov* (2021) **7** 378. PMID: 34876564
62. Zafar U, Khaliq S, Ali Z, Lone K. **TCF7‐L2 rs7903146 polymorphism in metabolic syndrome with and without acute coronary syndrome**. *J Pak Med Assoc* (2020) **70** 1774-1778. PMID: 33159751
63. Ma Z, Wang C, Bai X. **TCF7 is highly expressed in immune cells on the atherosclerotic plaques, and regulating inflammatory signaling via NFκB/AKT/STAT1 signaling**. *Biosci Rep* (2022) **42**. PMID: 35792753
|
---
title: Correlation of apolipoprotein A‐I with T cell subsets and interferon‐ү in coronary
artery disease
authors:
- Xinlin Xiong
- Zonggang Duan
- Haiyan Zhou
- Guangwei Huang
- Li Niu
- Zhenhua Luo
- Wei Li
journal: Immunity, Inflammation and Disease
year: 2023
pmcid: PMC10013138
doi: 10.1002/iid3.797
license: CC BY 4.0
---
# Correlation of apolipoprotein A‐I with T cell subsets and interferon‐ү in coronary artery disease
## Abstract
APOAI was negatively correlated with CD4+ T cells, and IFN‐γ, and positively correlated with CD8+T cells in CAD. Together with the findings of previous studies, the present results provided novel important information for the modulatory action and anti‐inflammatory response between APOAI and T cell subsets, and inflammatory markers in CAD.
### Background
The association of Apolipoprotein A‐I (APOAI) with T cell subsets and interferon‐ү (IFN‐γ) in patients with coronary artery disease (CAD) has been not reported. Thus, this study aimed to investigate the association of APOAI with T cell subsets and IFN‐γ in CAD.
### Methods
This study included a total of 107 patients with CAD including acute coronary syndrome and chronic coronary syndrome. T cell subsets, and CD3‐CD56+ natural killer cells were quantified by flow cytometric analysis. The serum concentrations of IFN‐ү were measured by enzyme‐linked immunosorbent assay. Lipid profiles, C‐reactive protein (CRP), and fibrinogen were measured in the clinical laboratory. Clinical data was obtained duration hospitalization.
### Results
The CD4+ T cells were higher in patients of the low‐APOAI group (<median: 1.2 mmol/L) than in patients of the high‐APOAI group(≥median: 1.2 mmol/L) ($p \leq .05$). The CD8+ T cells were lower in patients of the low APOAI group than in patients of the high‐APOAI group ($p \leq .05$). APOAI was inversely associated with CD4+ T cells, IFN‐γ, and was positively associated with CD8+ T cells ($p \leq .05$). No correlation was observed between CD3 + CD56+ cells, regulatory T cells (Tregs), and CD3‐CD56+ natural killer cells and APOAI ($p \leq .05$). The high‐density lipoprotein cholesterol (HDL‐C) was also inversely associated with CD4+ T cells ($p \leq .05$), and positively associated with CD8+ T cells ($p \leq .05$). Lastly, APOA1 and HDL‐C did not correlated with fibrinogen and CRP ($p \leq .05$).
### Conclusion
The present study demonstrated the correlation of APOAI with T cell subsets and IFN‐γ in CAD. These results provided novel information for the regulatory action between APOAI and T cell subsets and inflammatory immunity in CAD.
## INTRODUCTION
Coronary artery disease (CAD) results from the comprehensive action of multiple factors such as diabetes mellitus, smoking, dyslipidemia, and obesity, and is linked to high morbidity and mortality worldwide. 1, 2 Accumulating studies have demonstrated that atherosclerosis is an inflammatory immune cell‐ and lipid‐driven disease. 2, 3 Immune response and inflammation are associated with CAD development, which is characterized by the formation of foam cells, infiltration of immune cells and cytokines, and lipid deposition. 3 White blood cells could increase the risk for significant coronary stenosis, noncalcified plaques, and cardiovascular diseases. 4 Moreover, dyslipidemia plays an important role in the initiation and development of atherosclerosis. Atherogenic lipids such as low‐density lipoprotein cholesterol (LDL‐C) are modified by oxidation and deposited in the sub‐endothelial area. Oxidized LDL can promote the expression of adhesion molecules on endothelial and smooth muscle cells, which can attract monocytes and lymphocytes into the vascular wall. Oxidized LDL activates T cells and macrophages. Finally, atherosclerotic plaques form, and these atherosclerotic lesions contain T cells, cytokines, accumulative lipids, and foam cells. 2, 3, 5 Apolipoprotein A‐I (APOAI), the major structural and functional protein of high‐density lipoprotein‐cholesterol (HDL‐C), constitutes approximately $70\%$ of HDL protein. 6 APOAI could increase the cholesterol efflux from peripheral cells, and APOAI transfers peripheral cholesterol incorporated into HDL with the help of the ATP‐binding cassette transporter protein A1 from the peripheral tissues or cells to the liver, a process called reverse cholesterol transport. 6 Thus, APOAI could regulate the cholesterol homeostasis in cells.
Increasing evidence have shown that APOAI or HDL has an immunomodulatory effect on innate and adaptive immune responses such as monocytes, macrophages, neutrophils, and T cells. 7, 8, 9 These immunomodulatory roles have been elaborated in some diseases such as experimental arthritis, 10, 11 colitis, 12 systemic lupus erythematosus, 13 graft‐versus‐host disease, 14 autoimmune encephalomyelitis in mice models. 15 Besides, in human studies, the associations of AOPAI or HDL with sepsis, acute pancreatitis, autoimmune disease, cancers have been reported. 16, 17, 18, 19, 20 Manipulating APOAI has also been thought to be a novel treatment strategy for CAD. A study demonstrated that the use of APOAI mimetic peptides could improve atherosclerosis and plaque progression. 21 A study of patients with hypertension also showed a negative association of HDL‐C level with leukocytes and lymphocytes. 22 Another study revealed that HDL‐C was inversely associated with white blood cells in patients with CAD. 23 However, to the best of our knowledge, the association of APOA1 or HDL‐C with T cell subsets and interferon‐ү (INF‐ү) in patients with CAD has been not reported. The correlation of APOA1 with T cell subsets and inflammatory makers in patients with CAD should be elucidated. Accordingly, this study investigated the association of APOAI with T cell subsets and IFN‐γ in CAD.
## Patients
This study included a total of 107 patients who had undergone coronary angiography at the Affiliated Hospital of Guizhou Medical University (from May 2021 to August 2022). The diagnostic criteria for CAD was based on the combination of ischemic clinical manifestations, electrocardiograms and troponin alteration, and coronary angiogram findings. 24, 25, 26, 27, 28 The patients with CAD had at least one main coronary artery stenosis with a luminal diameter of ≥$50\%$. CAD was composed of acute coronary syndrome (ACS) and chronic coronary syndrome (CCS). The median age of patients with CAD was 63.00 (55.00, 72.00) years. Patients with CAD who had either of the following medical histories were excluded: human immunodeficiency virus infection, malignancy, valvular heart disease, organ dysfunction (such as severe liver, or renal dysfunction), septicemia, and steroid therapy. This study was approved by the Ethics Committee of the Affiliated Hospital of Guizhou Medical University. Informed consent was obtained from all participants. This study was also in compliance with the ethical standards of Declaration of Helsinki.
## Clinical data collection
Demographic data, such as age, sex, and medical history, and anthropometric parameters, such as body mass index (BMI), weight, and height, were obtained from the electronic medical system. Lipid profiles, C‐reactive protein (CRP), and fibrinogen were measured in the clinical laboratory.
## Flow cytometric analysis of lymphocytes in the peripheral blood
Heparin‐anticoagulated whole blood was incubated with fluorophore‐conjugated antibodies against CD3, CD4, CD8, CD25, CD127, and CD56 at room temperature for 20 min. After erythrocytes were lysed using red blood cell lysate, the stained cells were washed once. The resuspended cells were analyzed by flow cytometry using the BD FACSCelesta flow cytometer equipped with Diva software. The CD3, CD4, CD8, CD25, and CD56 antibodies were purchased from BD Biosciences, and the antibody against CD127 was purchased from BioLegend.
## Detection of IFN‐γ by enzyme‐linked immunosorbent assay (ELISA)
Blood samples were collected in a tube without anticoagulants, and the sera were separated, aliquoted, and stored at −80°C until analysis. The serum concentrations of IFN‐γ (Multiscience, China) were measured by ELISA based on the manufacturer's instructions.
## Statistical analyses
Continuous variables were presented as mean ± standard deviation or medians with interquartile range. Categorical variables were presented as proportions. The χ 2 or Fisher's exact test was adopted for the comparisons of categorical variables. Comparison of continuous variables with normal distribution between two groups was carried out by Student's t‐test, or comparison of the nonparametric variables was carried out by Mann−Whitney U tests. The association of APOAI with T cell subsets, CD3‐CD56+ cells, fibrinogen, and IFN‐ү were analyzed by spearman's correlation coefficient. The multivariable line regression analysis was used to assess the association of T cell subsets, and CD3‐CD56+ cells with APOAI and HDL‐C levels. APOAI and HDL‐C were treated as independent variables and forcibly entered into the model, respectively. Other covariates were entered into the model with the stepwise method. All data were analyzed using IBM SPSS Statistics for Windows, version 26.0 (IBM Corp.). Values with a two‐sided p Value of <.05 were considered significant.
## Baseline characteristics of patients with CAD according to APOAI median
These patients were stratified into low‐APOAI (<1.2 mmol/L) and high‐APOAI groups (≥1.2 mmol/L) according to APOAI median (1.2 mmol/L). The baseline characteristics of the patients are shown in Table 1. Patients in the low‐APOAI group had a higher prevalence of smoking, and higher frequency of male than those in the high‐APOAI group ($p \leq .05$). The HDL‐C and total cholesterol (TC) levels were decreased in low‐APOAI group, compared with high‐APOAI group. However, no differences were noted in the distribution of patients with ACS and CCS, diabetes mellitus, hypertension, statin use, triglyceride, age, and BMI between two groups.
**Table 1**
| Characteristics | Low‐apolipoprotein A‐I group (<1.2 mmol/L, n = 53) | High‐apolipoprotein A‐I group (≥1.2 mmol/L, n = 54) | p |
| --- | --- | --- | --- |
| Age (years) | 61.00 (54.50−70.50) | 66.50 (54.50−73.00) | NS |
| BMI (kg/m2) | 25.04 ± 3.04 | 24.22 ± 3.21 | NS |
| Male (%) | 43 (81.10) | 34 (63.00) | .036 |
| Hypertension, n (%) | 27 (50.90) | 36 (66.70) | NS |
| diabete, n (%) | 16 (30.20) | 17 (31.50) | NS |
| smoking, n (%) | 35 (66.00) | 19 (35.20) | .001 |
| ACS/CCS | 35/18 | 37/17 | NS |
| Statins, n (%) | 15 (28.30) | 13 (24.10) | NS |
| Total cholesterol | 3.93 ± 1.15 | 4.44 ± 1.06 | .020 |
| Triglyceride | 1.41 (0.84−2.40) | 1.61 (1.04−2.63) | NS |
| HDL‐C | 0.89 (0.76−1.00) | 1.11 (1.02−1.35) | <.001 |
## Distribution of T cell subsets, and CD3‐CD56+ cells according to APOAI median
The gating strategy of T cell subsets, and CD3‐CD56+ cells is presented in Figure 1. The number of CD4+ T cells was higher in the low‐APOAI group than in the high‐APOAI group ($p \leq .05$) (Figure 2A). The number of CD8+ T cells was lower in the low‐APOAI group than in the high‐APOAI group (Figure 2B). No difference was noted in CD3 + CD56+ T cells, CD3‐CD56+ cells, CD3 + CD56‐ T cells, and regulatory T cells (Tregs) between two groups (Figure 2C–F).
**Figure 1:** *The gating strategy of the T cell, its subsets, and CD3‐CD56+ cells. (A) CD3/CD56 dot plot was used to identify CD3 + CD56‐cells, CD3‐CD56+ cells, and CD3 + CD56+ cells. (B) CD4/CD8 dot plot was used to identify CD4+ T cells and CD8+ T cells. (C) CD4 + CD25 + CD127‐/low T cells(Tregs).* **Figure 2:** *Differences in T cell subsets, and CD3‐CD56+ cells according to apolipoprotein A‐I median. (A) CD4+ T cells, (B) CD8+ T cells, (C) CD3 + CD56‐T cells, (D) CD4 + CD25 + CD127‐/low(Tregs), (E) CD3 + CD56+ T cells, (F) CD3‐CD56+ cells. (*Indicates p < .05). CD3+ T cells included CD3 + CD56‐T cells and CD3 + CD56+ T cells; CD56+ cells included CD3‐CD56+ cells and CD3 + CD56+ T cells.*
## Spearman's correlation analysis of APOAI with T cell subsets, and CD3‐CD56+ cells
The APOAI was associated with CD4+ T cells (r = −0.286, $$p \leq .003$$, Table 2 and Figure 3A), and CD8+ T cells ($r = 0.250$, $$p \leq .009$$, Table 2 and Figure 3B) in patients with CAD. However, CD3‐CD56+ cells, CD3 + CD56− T cells, Tregs, and CD3 + CD56+ T cells did not correlate with APOAI (Table 2).
## The Independent correlations between APOAI and T cell subsets, and CD3‐CD56+ cells
We performed a multivariable stepwise regression analysis to assess the correlation of APOAI with T cell subsets, and CD3‐CD56+ cells. After taking into account these variables including age, sex, BMI, type of CAD (ACS and CCS), hypertension, diabetes mellitus, smoking, triglyceride (TG), TC, and statin use. APOAI was negatively related to CD4+ T cells (β = −.286, $$p \leq .005$$) and positively related to CD8+ T cells (β =.341, $$p \leq .001$$)(Table 3), whereas we found that APOAI was not associated with CD3‐CD56+ cells, CD3 + CD56+ T cells, CD3 + CD56− T cells, and Tregs (Table 3).
**Table 3**
| Dependent variables | Other variables included in the models | Standardization coefficient β with apolipoprotein A‐I as independent variables | p |
| --- | --- | --- | --- |
| CD3 + CD56− T cells | Covariates | 0.030 | 0.762 |
| CD4+ T cells | Covariates | −0.286 | 0.005 |
| CD8+ T cells | Covariates | 0.341 | 0.001 |
| CD4 + CD25 + CD127low/− cells (Tregs) | Covariates | 0.071 | 0.47 |
| CD3 + CD56+ T cells | Covariates | −0.030 | 0.762 |
| CD3‐CD56+ cells | Covariates | 0.026 | 0.786 |
## The Independent correlations between HDL‐C with T cell subsets, and CD3‐CD56+ cells
HDL‐C consists mainly of APOAI, 6 and HDL‐C is closely related to APOAI ($r = 0.777$, $p \leq .001$, Figure 4A). Thus, we also assessed the correlation of HDL‐C with T cell subsets, and CD3‐CD56+ cells, which will further prove the correlation of APOAI with T cell subsets, and CD3‐CD56+ cells from the perspectives of HDL‐C. We performed multiple line stepwise regression analysis after controlling for age, sex, BMI, hypertension, diabetes mellitus, smoking, CAD type including ACS and CCS, TG, TC, and statin use. The HDL‐C was found to be inversely associated with CD4+ T cells (β = −.224, $$p \leq .024$$), and positively related to CD8+ T cells (β =.259, $$p \leq .010$$). However, a significant association of CD3‐CD56+ cells, CD3 + CD56+ T cells, CD3 + CD56− T cells, and Tregs with HDL‐C was also not observed (Table 4).
**Figure 4:** *The association of Apolipoprotein A‐I with high density lipoprotein cholesterol (HDL‐C), and Interferon‐ү (IFN‐ү). Apolipoprotein A‐I positively correlated with HDL‐C (A). Apolipoprotein A‐I negatively correlated with IFN‐ү (B). HDL‐C did not correlate with IFN‐ү (C). IFN‐ү values were logarithmically transformed.* TABLE_PLACEHOLDER:Table 4
## Correlation of makers of inflammation with APOAI
We examined the association of APOAI with IFN‐γ, CRP, and fibrinogen. IFN‐γ data of 88 patients were available. Spearman's correlation analysis showed that APOAI correlated with IFN‐γ (r = −0.287, $p \leq .05$, Figure 4B). CRP data of 55 patients were available. APOAI did not correlate with CRP ($p \leq .05$). Additionally, HDL‐C was not associated with IFN‐γ (Figure 4C) and CRP ($p \leq .05$). Lastly, APOA1 and HDL‐C did not correlated with fibrinogen ($p \leq .05$).
## DISCUSSION
To the best of our knowledge, this is the first study to find the correlation of APOAI with T cell subsets and IFN‐ү in patients with CAD. The major finding of this study was that the APOAI and HDL‐C levels were inversely and independently correlated with CD4+ T cells, and positively and independently correlated with CD8+ T cells in patients with CAD. APOAI levels were aslo negatively associated with IFN‐γ. Furthermore, we did not find the association of APOAI and HDL‐C with CD3‐CD56+ cells, CD3 + CD56+ T cells, CD3 + CD56− T cells, Tregs, and fibrinogen. Nevertheless, this study may provide new insights concerning the association of APOAI with immune inflammation in CAD.
T cells, marked by CD3, include CD4+ T cells, CD8+ T cells, and CD3 + CD56+ T cells. Moreover, CD4+ T cells were classified into CD4+ T helper cells (Th) such as Th1, Th2, Th17, and Tregs. 5 Numerous studies have shown that T cells and its subsets were associated with CAD or atherosclerosis. T cells including CD4+ T cells and CD8+ T cells were found to be increased in atherosclerotic lesions. 5, 29 An earlier study reported that T cells were found in plaques of human carotid artery specimens. 30 Another study reported that CD4+ T cells were major proatherogenic cells in a mice model of atherosclerosis. 31 The transfer of CD4+ T cells from apoE−/− mice into immunodeficient apoE−/− mice could promote the progression of atherosclerotic lesions. 32 The absence of CD4+ T cells in apoE‐knockout mice could ameliorate atherosclerosis. 31 A study revealed that CD8+ T cells played an obvious role in advanced atherosclerotic plaques. 33 CD8+ T cells accounted for approximately $50\%$ of the lymphocytes in advanced lesions in human arterial tissue. 33 The depletion of CD8+ T cells in apo E‐deficient mouse model improved atherosclerosis. 34 The above findings indicated that T cells play important roles in the pathogenesis of CAD and atherosclerosis.
An animal study indicated that APOAI could inhibit the proliferation of CD4+ memory T cell subsets (CD4 + CD44high T cells) in the lymph nodes in the LDLr −/–, APOAI −/− (DKO) mouse model. 35 Another study reported that APOAI deficiency could expand T cells in the mouse model. 36 A study showed that synthetic peptides of APOAI could inhibit allogeneic T‐lymphocyte proliferation in vitro. 37 The above studies have suggested that APOAI inversely regulated T cells.
Several studies have reported the correlation of APOAI or HDL‐C with T cell subsets. Zhao et al. reported that HDL‐C was inversely associated with CD3+ T cells and CD4+ T cells in elderly patients with type 2 diabetes mellitus (T2DM). 38 Another study reported that HDL‐C was inversely associated with CD4+ T subsets including interleukin (IL)‐8‐expressing CD4+ T cells and IL‐17‐expressing CD4+ T cells in diabetes mellitus patients with CAD. However, the association was not observed among these participants with T2DM, T2DM without CAD, and healthy controls. 39 Moreover, Guasti et al. found that HDL and APOAI were not associated with CD4+ T cells in patients with dyslipidemia and healthy controls. 40 In present study, we found a negative and independent correlation of APOAI or HDL‐C with CD4+ T cells in CAD. Based on the results of our study and those of the aforementioned studies, we speculated that APOAI might negatively regulate CD4+ T cells in CAD. The interaction of APOAI with T cells may be involved in the pathogenesis of CAD or atherosclerosis.
Tregs play important roles in atherosclerosis. It could be found in atherosclerotic plaques. Multiple studies have shown that Tregs helped to control the progression of atherosclerotic lesions. 41, 42, 43 Although several studies have assessed the correlation of HDL‐C or APOAI with Tregs in different patients including those with diabetes, dyslipidemia, and healthy individuals, the results remain controversial. 40, 44, 45, 46 Studies have shown that FOXP3+ regulatory T cells were positively associated with HDL‐C in healthy participants and in participants with T2DM. 44, 45 Others reported contrary results. Wigren et al. reported that they did not find a significant association of Tregs with lipid profiles including HDL‐C, LDL‐C, and TG in a prospective cohort study. 46 In present study, we found no association of APOAI or HDL‐C with Tregs in CAD. Our findings were in agreement with the study by Wigren et al. 46 The difference in the results of various studies is probably related to the differences in the study participants, and diseases state.
A previous literature reported that APOAI had anti‐atherosclerotic action by some functions such as reverse cholesterol transport, inhibiting the expression of cell adhesion molecules, depleting cholesterol of lipid rafts, and exerting antioxidative effects. 9, 47 Using ATP binding cassette transporters A1 and G1 (Abca1/g1)‐deficient mice, Westerterp M et al. found that reduced cholesterol efflux in dendritic cells could lead to their activation and production of inflammatory cytokine. 48 Cholesterol accumulation in antigen‐presenting cells could promote T cell priming, APOAI treatment improved reverse cholesterol transport and the immune disorder. 49 Our findings, in combination with those of previous studies, 23, 35, 36, 37 indicated that the regulation of APOAI to CD4+ T cells may be another key step and mechanism for anti‐atherosclerosis in CAD. We speculated that the negative correlation of APOAI with CD4+ T cells is caused by APOAI or HDL‐C's inhibition of the ability of antigen‐presenting cells to present antigen to T cells, influencing cholesterol efflux, lipid raft disruption or production of some inflammatory cytokines. 10, 37, 47, 48, 49, 50, 51 Interestingly, we found that APOAI and HDL‐C were positively correlated with CD8+ T cells, because human specimens and animal experiments revealed that CD8+ T cells contributed to the development of atherosclerosis. 33, 34 This appeared to be contrary to the anti‐atherosclerotic role of APOAI and HDL‐C. An earlier study showed that HDL‐C was negatively correlated with the risk of CAD, 52 whereas other studies have reported that the upregulation of HDL‐C level did not lower the risk of CAD. 53, 54 One of the possible reasons for this result is the positive regulatory role of APOAI and HDL‐C with CD8+ T cells. However, the precise mechanisms warrant further investigation in different diseases, especially CAD.
IFN‐γ has important proatherogenic effects. IFN‐γ is produced by various cells including CD4+ T cells and CD8+ cells. In mouse models, exogenous IFN‐γ administration increases the atherosclerotic lesions. 55 By contrast, the loss of IFN‐γ signaling can reduce atherosclerotic lesions in ApoE−/− mice and LDLR−/− mice. 56, 57 In patients with CAD, the evidence from the study showed that IFN‐γ was obviously increased in the CAD group compared with the control group and was associated with the occurrence of CAD. 58, 59 IFN‐γ has also been characterized as a functional and hallmark cytokine of Th1 cells and can induce Th1 differentiation. In human atherosclerotic lesions, Th1 cells are predominant effector cells, which are associated with the progression of the lesion. 60 Moreover, IFN‐γ modulates other inflammatory cells including monocytes, macrophages, and foam cells to take part in and promote atherosclerosis. 61 IFN‐γ exposure can lead to endothelial dysfunction of human coronary artery endothelial cells. 62 HDL and APOAI could enable the inhibition of inflammation. They could prohibit the secretion of chemokines such as CCL2 and CX3CL1, reduce their receptor expression, and adhesion molecule expression on endothelial cells. 6 Besides, APOAI suppressed proinflammatory cytokines secreted by monocytes by blocking the interaction between monocytes and stimulated T cells. 63 A study reported that HDL treatment inhibit IL‐6, and tumor necrosis factor‐α, and IL‐4 production by Mycobacterium tuberculosis‐infected macrophages. 64 IFN‐γsecretion was suppressed by HDL or APOAI treatment. 10, 37, 64 Moreover, APOA1 or HDL reduced the production of IFN‐ү by CD8+ T cells. 14 However, previous studies have not reported the association of APOAI or HDL‐C with IFN‐γ in patients with CAD. In present study, we found that APOAI was also inversely related to the IFN‐γ level, together with the above studies, suggesting that APOAI may exert anti‐inflammatory response, at least in part, by negatively regulating the production of IFN‐γ in CAD. However, the association between HDL‐C and IFN‐γ was not observed, which was inconsistent with the correlation between APOAI and IFN‐γ. The inconsistent result may be attributed to that HDL included various ingredients such as apolipoprotein A‐II, apolipoprotein E, 6 which could interfere with the association of HDL‐C with IFN‐γ. Moreover, CRP was thought to be an inflammatory marker, as it has been reported in some inflammatory diseases including CAD. We also assessed the association of APOAI and HDL‐C with CRP, whereas we did not find a significant correlation of APOAI or HDL‐C with CRP. These findings, in combination with those of previous studies, indicated that the regulatory correlation of APOAI and HDL‐C with T cell subsets, and IFN‐γ may play an important role in the pathogenesis of CAD. Fibrinogen was not only inflammatory representation but also coagulation markers. 65 In present study, however, a significant correlation of AOPAI and HDL‐C with Fibronogen was not observed.
This study has some limitations. First, T cells could be further divided into subtypes such as IFN‐ү‐secreting CD4+ T cells (Th1), Th2, Th17, and IFN‐ү‐secreting CD8+ T cells. The correlation of APOAI with these subtypes of T cells deserves much further exploration. Second, we analyzed regulatory CD4+ T (Treg) cells based on the low expression of CD127 and the expression of CD25, but not directly using the intracellular expression of FoxP3. Even if this is a widely approach used to identify Treg, the prototypical marker of *Treg is* the transcription factor, FoxP3. Thirdly, we have only analyzed a subset of Treg, the association of APOAI with other Treg cells such as IL‐10‐secreting Treg cells (Tr1) cells remains to be further studied. Moreover, in patients with CAD, besides IFN‐ү, many other cytokines are abnormally secreted by immune cells, whether APOAI also could inhibit these cytokines remains to be further studied. Finally, we did not obtain the medication duration and dosage of statins use, which may limit further correlation analysis.
## CONCLUSIONS
In conclusion, we firstly revealed that APOAI was negatively correlated with CD4+ T cells, and IFN‐γ, and positively correlated with CD8+ T cells in CAD. Together with the findings of previous studies, the present results provided novel important information for the modulatory action and anti‐inflammatory response between APOAI and T cell subsets, and inflammatory markers in CAD. However, the exact signal pathways and molecular mechanisms between APOAI and T cells and inflammatory markers require further investigation in patients with CAD or atherosclerosis. Our ongoing studies will elucidate these issues.
## AUTHOR CONTRIBUTIONS
Xinlin Xiong: Conceptualization; data curation; formal analysis; investigation; methodology; project administration; software; validation; writing—original draft; writing—review and editing. Zonggang Duan: Conceptualization; formal analysis; methodology; software. Haiyan Zhou: Conceptualization; formal analysis; methodology; software; validation; writing—original draft; writing—review and editing. Guangwei Huang: Formal analysis; methodology; software. Li Niu: formal analysis; investigation; methodology; software. Zhenhua Luo: Conceptualization; formal analysis; funding acquisition; methodology; project administration; resources; supervision; validation; writing—original draft; writing—review and editing. Wei Li: Conceptualization; data curation; formal analysis; funding acquisition; investigation; methodology; project administration; resources; supervision; validation; visualization; writing—original draft; writing—review and editing.
## CONFLICT OF INTEREST STATEMENT
The authors declare no conflict of interest.
## ETHICS STATEMENT
This study was approved by the Ethics Committee of the Affiliated Hospital of Guizhou Medical University (approval number: 2021010K).
## DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author, upon reasonable request.
## References
1. Ji Y, Bai C. **Research progress of hypertriglyceridemia and coronary heart disease**. *Heart Mind* (2018) **2** 40-44. DOI: 10.4103/hm.hm_2_19
2. Ali AH, Younis N, Abdallah R. **Lipid‐Lowering therapies for atherosclerosis: statins, fibrates, ezetimibe and PCSK9 monoclonal antibodies**. *Curr Med Chem* (2021) **28** 7427-7445. DOI: 10.2174/0929867328666210222092628
3. Malekmohammad K, Bezsonov EE, Rafieian‐Kopaei M. **Role of lipid accumulation and inflammation in atherosclerosis: focus on molecular and cellular mechanisms**. *Front Cardiovasc Med* (2021) **8**. DOI: 10.3389/fcvm.2021.707529
4. Kim JH, Lim S, Park KS, Jang HC, Choi SH. **Total and differential WBC counts are related with coronary artery atherosclerosis and increase the risk for cardiovascular disease in Koreans**. *PLoS One* (2017) **12**. DOI: 10.1371/journal.pone.0180332
5. Neupane R, Jin X, Sasaki T, Li X, Murohara T, Cheng XW. **Immune disorder in atherosclerotic cardiovascular disease ‐ clinical implications of using circulating T‐cell subsets as biomarkers**. *Circ J* (2019) **83** 1431-1438. DOI: 10.1253/circj.CJ-19-0114
6. van der Vorst EPC. **High‐density lipoproteins and apolipoprotein A1**. *SubCell Biochem* (2020) **94** 399-420. DOI: 10.1007/978-3-030-41769-7_16
7. Murphy AJ, Woollard KJ, Suhartoyo A. **Neutrophil activation is attenuated by high‐density lipoprotein and apolipoprotein A‐I in in vitro and in vivo models of inflammation**. *Arterioscler Thromb Vasc Biol* (2011) **31** 1333-1341. DOI: 10.1161/ATVBAHA.111.226258
8. Trakaki A, Marsche G. **Current understanding of the immunomodulatory activities of high‐density lipoproteins**. *Biomedicines* (2021) **9**. DOI: 10.3390/biomedicines9060587
9. Fernandes das Neves M, Batuca JR, Delgado Alves J. **The role of high‐density lipoprotein in the regulation of the immune response: implications for atherosclerosis and autoimmunity**. *Immunology* (2021) **164** 231-241. DOI: 10.1111/imm.13348
10. Tiniakou I, Drakos E, Sinatkas V. **High‐density lipoprotein attenuates Th1 and th17 autoimmune responses by modulating dendritic cell maturation and function**. *J Immunol* (2015) **194** 4676-4687. DOI: 10.4049/jimmunol.1402870
11. Wang Y, Lu S, Zhang G. **Anti‐inflammatory effects of HDL in mice with rheumatoid arthritis induced by collagen**. *Front Immunol* (2018) **9**. DOI: 10.3389/fimmu.2018.01013
12. Gerster R, Eloranta JJ, Hausmann M. **Anti‐inflammatory function of high‐density lipoproteins via autophagy of IκB kinase**. *Cell Mol Gastroenterol Hepatol* (2015) **1** 171-187.e1. DOI: 10.1016/j.jcmgh.2014.12.006
13. Black LL, Srivastava R, Schoeb TR, Moore RD, Barnes S, Kabarowski JH. **Cholesterol‐independent suppression of lymphocyte activation, autoimmunity, and glomerulonephritis by apolipoprotein A‐I in normocholesterolemic Lupus‐Prone mice**. *J Immunol* (2015) **195** 4685-4698. DOI: 10.4049/jimmunol.1500806
14. Chagué C, Gautier T, Dal Zuffo L. **High‐density lipoprotein infusion protects from acute graft‐versus‐host disease in experimental allogeneic hematopoietic cell transplantation**. *Am J Transplant (AJT)* (2022) **22** 1350-1361. DOI: 10.1111/ajt.16960
15. Hong S, Niu M, Meng D. **High‐density lipoprotein reduces microglia activation and protects against experimental autoimmune encephalomyelitis in mice**. *Int Immunopharmacol* (2022) **105**. DOI: 10.1016/j.intimp.2022.108566
16. Barlage S, Gnewuch C, Liebisch G. **Changes in HDL‐associated apolipoproteins relate to mortality in human sepsis and correlate to monocyte and platelet activation**. *Intensive Care Med* (2009) **35** 1877-1885. DOI: 10.1007/s00134-009-1609-y
17. Peng YS, Chen YC, Tian YC. **Serum levels of apolipoprotein A‐I and high‐density lipoprotein can predict organ failure in acute pancreatitis**. *Crit Care* (2015) **19** 88. DOI: 10.1186/s13054-015-0832-x
18. Madsen CM, Varbo A, Nordestgaard BG. **Low HDL cholesterol and high risk of autoimmune disease: two Population‐Based cohort studies including 117341 individuals**. *Clin Chem* (2019) **65** 644-652. DOI: 10.1373/clinchem.2018.299636
19. Pedersen KM, Çolak Y, Bojesen SE, Nordestgaard BG. **Low high‐density lipoprotein and increased risk of several cancers: 2 population‐based cohort studies including 116,728 individuals**. *J Hematol Oncol* (2020) **13** 129. DOI: 10.1186/s13045-020-00963-6
20. Nishiyama H, Funamizu T, Iwata H. **Low apolipoprotein A1 was associated with increased risk of cancer mortality in patients following percutaneous coronary intervention: a 10‐year follow‐up study**. *Int J Cancer* (2022) **151** 1482-1490. DOI: 10.1002/ijc.34164
21. Gou S, Wang L, Zhong C. **A novel apoA‐I mimetic peptide suppresses atherosclerosis by promoting physiological HDL function in apoE‐/‐ mice**. *Br J Pharmacol* (2020) **177** 4627-4644. DOI: 10.1111/bph.15213
22. Liu Y, Kong X, Wang W. **Association of peripheral differential leukocyte counts with dyslipidemia risk in Chinese patients with hypertension: insight from the China stroke primary prevention trial**. *J Lipid Res* (2017) **58** 256-266. DOI: 10.1194/jlr.P067686
23. Gao M, Zhao D, Schouteden S. **Regulation of high‐density lipoprotein on hematopoietic stem/progenitor cells in atherosclerosis requires scavenger receptor type BI expression**. *Arterioscler Thromb Vasc Biol* (2014) **34** 1900-1909. DOI: 10.1161/ATVBAHA.114.304006
24. Xiong X, Luo Z, Zhou H. **Downregulation of TIGIT expression in FOXP3 + regulatory T cells in acute coronary syndrome**. *J Inflamm Res* (2022) **15** 1195-1207. DOI: 10.2147/JIR.S351364
25. Knuuti J, Wijns W, Saraste A. **2019 ESC guidelines for the diagnosis and management of chronic coronary syndromes**. *Eur Heart J* (2020) **41** 407-477. DOI: 10.1093/eurheartj/ehz425
26. Thygesen K, Alpert JS, Jaffe AS. **Fourth universal definition of myocardial infarction (2018)**. *J Am Coll Cardiol* (2018) **72** 2231-2264. DOI: 10.1016/j.jacc.2018.08.1038
27. Collet JP, Thiele H, Barbato E. **2020 ESC guidelines for the management of acute coronary syndromes in patients presenting without persistent ST‐segment elevation**. *Eur Heart J* (2021) **42** 1289-1367. DOI: 10.1093/eurheartj/ehaa575
28. Ibanez B, James S, Agewall S. **2017 ESC guidelines for the management of acute myocardial infarction in patients presenting with ST‐segment elevation: the task force for the management of acute myocardial infarction in patients presenting with ST‐segment elevation of the European society of cardiology (ESC)**. *Eur Heart J* (2018) **39** 119-177. DOI: 10.1093/eurheartj/ehx393
29. Moreira DM, da Silva RL, Vieira JL, Fattah T, Lueneberg ME, Gottschall CAM. **Role of vascular inflammation in coronary artery disease: potential of anti‐inflammatory drugs in the prevention of atherothrombosis**. *American Journal of Cardiovascular Drugs* (2015) **15** 1-11. DOI: 10.1007/s40256-014-0094-z
30. Lehtonen‐Smeds EMP, Mäyränpää M, Lindsberg PJ. **Carotid plaque mast cells associate with atherogenic serum lipids, high grade carotid stenosis and symptomatic carotid artery disease**. *Cerebrovasc Dis* (2005) **19** 291-301. DOI: 10.1159/000084497
31. Zhou X, Robertson AKL, Rudling M, Parini P, Hansson K. **Lesion development and response to immunization reveal a complex role for CD4 in atherosclerosis**. *Circ Res* (2005) **96** 427-434. DOI: 10.1161/01.RES.0000156889.22364.f1
32. Zhou X, Nicoletti A, Elhage R, Hansson K. **Transfer of CD4(+) T cells aggravates atherosclerosis in immunodeficient apolipoprotein E knockout mice**. *Circulation* (2000) **102** 2919-2922. DOI: 10.1161/01.cir.102.24.2919
33. Gewaltig J, Kummer M, Koella C, Cathomas G, Biedermann BC. **Requirements for CD8 T‐cell migration into the human arterial wall**. *Hum Pathol* (2008) **39** 1756-1762. DOI: 10.1016/j.humpath.2008.04.018
34. Kyaw T, Winship A, Tay C. **Cytotoxic and proinflammatory CD8 + T lymphocytes promote development of vulnerable atherosclerotic plaques in apoE‐deficient mice**. *Circulation* (2013) **127** 1028-1039. DOI: 10.1161/CIRCULATIONAHA.112.001347
35. Wilhelm AJ, Zabalawi M, Grayson JM. **Apolipoprotein A‐I and its role in lymphocyte cholesterol homeostasis and autoimmunity**. *Arterioscler Thromb Vasc Biol* (2009) **29** 843-849. DOI: 10.1161/ATVBAHA.108.183442
36. Ouweneel AB, Reiche ME, Snip OSC. **Apolipoprotein A1 deficiency in mice primes bone marrow stem cells for T cell lymphopoiesis**. *J Cell Sci* (2022) **135**. DOI: 10.1242/jcs.258901
37. Hsieh GR, Schnickel GT, Garcia C, Shefizadeh A, Fishbein MC, Ardehali A. **Inflammation/oxidation in chronic rejection: apolipoprotein a‐I mimetic peptide reduces chronic rejection of transplanted hearts**. *Transplantation* (2007) **84** 238-243. DOI: 10.1097/01.tp.0000268509.60200.ea
38. Zhao C, Lou F, Li X. **Correlation of CD3 + /CD4 + , and serum CK‐18 fragment levels with glucose and lipid metabolism in elderly type 2 diabetes patients with nonalcoholic fatty liver disease**. *Am J Transl Res* (2021) **13** 2546-2554. PMID: 34017413
39. Mahmoud F, Al‐Ozairi E. **Inflammatory cytokines and the risk of cardiovascular complications in type 2 diabetes**. *Dis Markers* (2013) **35** 235-241. DOI: 10.1155/2013/931915
40. Guasti L, Maresca AM, Schembri L. **Relationship between regulatory T cells subsets and lipid profile in dyslipidemic patients: a longitudinal study during atorvastatin treatment**. *BMC Cardiovasc Disord* (2016) **16** 26. DOI: 10.1186/s12872-016-0201-y
41. Ait‐Oufella H, Salomon BL, Potteaux S. **Natural regulatory T cells control the development of atherosclerosis in mice**. *Nature Med* (2006) **12** 178-180. DOI: 10.1038/nm1343
42. Dinh TN, Kyaw TS, Kanellakis P. **Cytokine therapy with interleukin‐2/anti‐interleukin‐2 monoclonal antibody complexes expands CD4 + CD25 + Foxp3+ regulatory T cells and attenuates development and progression of atherosclerosis**. *Circulation* (2012) **126** 1256-1266. DOI: 10.1161/CIRCULATIONAHA.112.099044
43. Pedicino D, Giglio A, Ruggio A. **Inflammasome, T lymphocytes and Innate‐Adaptive immunity crosstalk: role in cardiovascular disease and therapeutic perspectives**. *Thromb Haemostasis* (2018) **118** 1352-1369. DOI: 10.1055/s-0038-1666860
44. Rodríguez‐Perea AL, Montoya CJ, Olek S, Chougnet CA, Velilla PA. **Statins increase the frequency of circulating CD4 + FOXP3 + regulatory T cells in healthy individuals**. *J Immunol Res* (2015) **2015**. DOI: 10.1155/2015/762506
45. Cheng L, Yang F, Cao X. **The effect of short‐term intensive insulin therapy on circulating T cell subpopulations in patients with newly diagnosed type 2 diabetes mellitus**. *Diabetes Res Clin Pract* (2019) **149** 107-114. DOI: 10.1016/j.diabres.2019.02.007
46. Wigren M, Björkbacka H, Andersson L. **Low levels of circulating CD4 + FoxP3 + T cells are associated with an increased risk for development of myocardial infarction but not for stroke**. *Arterioscler Thromb Vasc Biol* (2012) **32** 2000-2004. DOI: 10.1161/ATVBAHA.112.251579
47. Hu J, Xi D, Zhao J. **High‐density lipoprotein and inflammation and its significance to atherosclerosis**. *Am J Med Sci* (2016) **352** 408-415. DOI: 10.1016/j.diabres.2019.02.007
48. Westerterp M, Gautier EL, Ganda A. **Cholesterol accumulation in dendritic cells links the inflammasome to acquired immunity**. *Cell Metab* (2017) **25** 1294-1304.e6. DOI: 10.1016/j.cmet.2017.04.005
49. Ito A, Hong C, Oka K. **Cholesterol accumulation in CD11c + Immune cells is a causal and targetable factor in autoimmune disease**. *Immunity* (2016) **45** 1311-1326. DOI: 10.1016/j.immuni.2016.11.008
50. Kim KD, Lim HY, Lee HG. **Apolipoprotein A‐I induces IL‐10 and PGE2 production in human monocytes and inhibits dendritic cell differentiation and maturation**. *Biochem Biophys Res Commun* (2005) **338** 1126-1136. DOI: 10.1016/j.bbrc.2005.10.065
51. Yu B, Wang S, Peng D, Zhao S. **HDL and immunomodulation: an emerging role of HDL against atherosclerosis**. *Immunol Cell Biol* (2010) **88** 285-290. DOI: 10.1038/icb.2009.112
52. Ahmed HM, Miller M, Nasir K. **Primary low level of High‐Density lipoprotein cholesterol and risks of coronary heart disease, cardiovascular disease, and death: results from the multi‐ethnic study of atherosclerosis**. *Am J Epidemiol* (2016) **183** 875-883. DOI: 10.1093/aje/kwv305
53. Nicholls SJ. **CETP‐Inhibition and HDL‐Cholesterol: A story of CV risk or CV benefit, or both**. *Clin Pharm Ther* (2018) **104** 297-300. DOI: 10.1002/cpt.1118
54. Voight BF, Peloso GM, Orho‐Melander M. **Plasma HDL cholesterol and risk of myocardial infarction: a mendelian randomisation study**. *The Lancet* (2012) **380** 572-580. DOI: 10.1016/S0140-6736(12)60312-2
55. Whitman SC, Ravisankar P, Elam H, Daugherty A. **Exogenous Interferon‐γ enhances atherosclerosis in apolipoprotein E−/− mice**. *Am J Pathol* (2000) **157** 1819-1824. DOI: 10.1016/s0002-9440(10)64820-1
56. Gupta S, Pablo AM, Jiang X, Wang N, Tall AR, Schindler C. **IFN‐gamma potentiates atherosclerosis in ApoE knock‐out mice**. *J Clin Invest* (1997) **99** 2752-2761. DOI: 10.1172/JCI119465
57. Buono C, Come CE, Stavrakis G, Maguire GF, Connelly PW, Lichtman AH. **Influence of Interferon‐γ on the extent and phenotype of Diet‐Induced atherosclerosis in the LDLR‐deficient mouse**. *Arterioscler Thromb Vasc Biol* (2003) **23** 454-460. DOI: 10.1161/01.ATV.0000059419.11002.6E
58. Yang Z, Shi L, Xue Y. **Interleukin‐32 increases in coronary arteries and plasma from patients with coronary artery disease**. *Clin Chim Acta* (2019) **497** 104-109. DOI: 10.1016/j.cca.2019.07.019
59. Wang H, Zhang N, Li G, Xu B. **Proinflammatory cytokine IFN‐γ, lncRNA BANCR and the occurrence of coronary artery disease**. *Life Sci* (2019) **231**. DOI: 10.1016/j.lfs.2019.05.066
60. Frostegård J, Ulfgren AK, Nyberg P. **Cytokine expression in advanced human atherosclerotic plaques: dominance of pro‐inflammatory (Th1) and macrophage‐stimulating cytokines**. *Atherosclerosis* (1999) **145** 33-43. DOI: 10.1016/s0021-9150(99)00011-8
61. Gisterå A, Hansson GK. **The immunology of atherosclerosis**. *Nat Revi Nephrol* (2017) **13** 368-380. DOI: 10.1038/nrneph.2017.51
62. Lee LYH, Oldham WM, He H. **Interferon‐γ impairs human coronary artery endothelial glucose metabolism by tryptophan catabolism and activates fatty acid oxidation**. *Circulation* (2021) **144** 1612-1628. DOI: 10.1161/CIRCULATIONAHA.121.053960
63. Hyka N, Dayer JM, Modoux C. **Apolipoprotein A‐I inhibits the production of interleukin‐1β and tumor necrosis factor‐α by blocking contact‐mediated activation of monocytes by T lymphocytes**. *Blood* (2001) **97** 2381-2389. DOI: 10.1182/blood.v97.8.2381
64. Inoue M, Niki M, Ozeki Y. **High‐density lipoprotein suppresses tumor necrosis factor alpha production by mycobacteria‐infected human macrophages**. *Sci Rep* (2018) **8** 6736. DOI: 10.1038/s41598-018-24233-1
65. Wolberg AS, Sang Y. **Fibrinogen and factor XIII in venous thrombosis and thrombus stability**. *Arterioscler Thromb Vasc Biol* (2022) **42** 931-941. DOI: 10.1161/ATVBAHA.122.317164
|
---
title: 'Association between thyroid dysfunction, metabolic disturbances, and clinical
symptoms in first-episode, untreated Chinese patients with major depressive disorder:
Undirected and Bayesian network analyses'
authors:
- Pu Peng
- Qianjin Wang
- Xiao E Lang
- Tieqiao Liu
- Xiang-Yang Zhang
journal: Frontiers in Endocrinology
year: 2023
pmcid: PMC10013149
doi: 10.3389/fendo.2023.1138233
license: CC BY 4.0
---
# Association between thyroid dysfunction, metabolic disturbances, and clinical symptoms in first-episode, untreated Chinese patients with major depressive disorder: Undirected and Bayesian network analyses
## Abstract
### Aims
Thyroid dysfunction and metabolic disturbances are common in major depressive disorder (MDD) patients. We aimed to assess the relationship between thyroid dysfunction, metabolic disturbances, and clinical symptoms in Chinese first-episode, drug-naïve (FEDN) MDD patients using undirected and Bayesian network methods.
### Methods
1718 FEDN MDD patients were recruited. Serum levels of free triiodothyronine (FT3), free thyroxine (FT4), thyroid stimulating hormone (TSH), anti-thyroglobulin (TgAb), thyroid peroxidases antibody (TPOAb), total cholesterol (TC), total triglycerides (TG), high-density lipoprotein (HDL-C), low-density lipoprotein (LDL-C), and glucose were assessed. Blood pressure and body mass index were measured. Hamilton Rating Scale for Depression (HAMD), Hamilton Rating Scale for Anxiety, and positive subscale of Positive And Negative Syndrome Scales were used to detect clinical symptoms. An undirected network with EBICglasso default and a directed acyclic graph (DAG) using the Bayesian network approach was conducted.
### Results
The prevalence rates of clinical symptoms, thyroid dysfunction, and metabolic dysfunction were as follows: anxiety ($$n = 894$$, $52\%$), psychotic symptoms (171, $10\%$), subclinical hypothyroidism (SCH, $$n = 1041$$, $61\%$), abnormal TgAb ($$n = 297$$, $17\%$), abnormal TPOAb ($$n = 438$$, $25\%$), hyperthyroidism ($$n = 5$$, $0.3\%$), hypothyroidism ($$n = 3$$, $0.2\%$), hyperglycemia ($$n = 241$$, $14\%$), hypertriglyceridemia ($$n = 668$$, $39\%$), low HDL-C ($$n = 429$$, $25\%$), hypercholesterolemia (421, $25\%$), abnormal TC (357, $21\%$), abnormal LDL-C (185, $11\%$). overweight or obesity ($$n = 1026$$, $60\%$), and hypertension ($$n = 92$$, $5.4\%$). Both networks demonstrated serum TSH and TC levels and the severity of depression played an important role in the pathophysiology of MDD.
### Conclusions
MDD patients may have thyroid and metabolic dysfunction in the early stage. Targeting hypercholesterolemia, depressive symptoms, and SCH in MDD patients may hold promise in reducing clinical symptoms, metabolic disturbances, and thyroid dysfunction.
## Introduction
Major depressive disorder (MDD) is the most common and severe psychiatric disorder, which leads to a significant impact on quality of life and functioning [1, 2]. It is the leading cause of disability-adjusted life years in adults and affects approximately 300 million people worldwide [3]. Emerging studies have demonstrated biological changes in MDD patients, such as impaired glucose and lipid metabolism, thyroid dysfunction, and obesity (4–6). Previous studies have found that metabolic disorders and/or thyroid dysfunction increase the risk of MDD [4, 7, 8]. More importantly, a growing number of studies have shown that metabolic disorders and thyroid dysfunction are positively associated with poor response to antidepressants [9], a higher risk of readmission [10], and more severe comorbidities such as suicide [11] and anxiety [12]. Taken together, these findings highlight the important role of metabolic disorders and thyroid dysfunction in the pathophysiology of MDD.
Despite the great interest in the abnormalities of metabolism and thyroid function in MDD patients, there are some important limitations of the current evidence. First, most prior studies have had relatively limited sample sizes and have ignored potential confounding factors such as antidepressants and disease courses [13, 14]. Second, although studies in the general population have demonstrated substantial associations between metabolic disorders and thyroid dysfunction (15–19), they are most often studied alone in patients with MDD. Clarifying whether and how metabolic disorders, thyroid disorders, and clinical symptoms reinforce and interact with each other in patients with MDD will deepen our understanding of the pathophysiology of MDD.
Network analysis is a promising tool for understanding the associations between variables [20, 21]. In the network model, each variable is represented as a “node”. The edges represent the unique association between two variables after adjusting for the rest of the variables within the network. Through network analysis, we can estimate the interaction between thyroid dysfunction, metabolic disturbance, and clinical symptoms in MDD patients in more detail. It allows us to identify the most influential variables in the network, which may serve as the target for clinical interventions [20].
To date, network approaches have been used to identify associations between pro-inflammatory proteins, lipid markers, genetic factors, and depressive symptoms in patients with MDD [22, 23]. However, no previous studies have assessed networks of thyroid dysfunction, metabolic disorders, and clinical symptoms in patients with MDD. To fill this gap, we conducted the present study. We recruited a large sample of first-episode drug-naïve (FEDN) patients with MDD, which minimized the impact of medical therapy, disease duration, and comorbidities. We conducted two networks. The first network was undirected and was used to explore the most relevant connections and core variables in the thyroid-metabolism-clinical network. For the second network, we performed a Bayesian network analysis to obtain directions of potential causal relationships between thyroid dysfunction, metabolic disorders, and clinical symptoms.
## Study setting and procedure
FEDN MDD outpatients were recruited during 2015-2017 at the Department of Psychiatry, First Hospital, Shanxi Medical University. Participants should meet the following criteria: [1] diagnosis of MDD by two independently trained clinical psychiatrists according to the Structured Clinical Interview for DSM-IV (SCID); [2] 17-item Hamilton Rating Scale for Depression (HAMD) total score ≥ 24; [3] age between 18-60 years old and Han Chinese; [4] depressive symptoms were first-episode without any prior medications, including antidepressant, antipsychotic, or thyroid treatment medication; and [4] MDD of disease duration was no more than 24 months.
Exclusion criteria included: [1] the presence of any other major DSM-IV Axis I disorder based on SCID; [2] the presence of serious physical illness, such as organic brain disease or severe infection; [3] any substance abuse or dependence other than tobacco; [3] pregnant or lactating women; and [4] refusal to give informed consent.
The work described has been carried out in accordance with The Code of Ethics of the World Medical Association (Declaration of Helsinki). The study was approved by the Institutional Review Board (IRB) of the First Hospital of Shanxi Medical University (No. 2016-Y27). All participants gave their written informed consent.
## Basic information
The demographic information including gender, age, married status, education level, age at first episode onset, and duration of MDD courses were collected through self-administrated questionnaires.
## Clinical assessments
Two trained psychiatrists independently administrated the SCID to each participant. They assessed depression, anxiety, and psychotic symptoms using the Chinese version of the HAMD, the Hamilton Anxiety Rating Scale (HAMA), and the Positive and Negative Symptom Scale (PANSS) positive subscale, respectively [24, 25]. Following the previous study [26], patients were divided into the without or possible anxiety group (HAMA scores 0-13), mild to moderate anxiety group (HAMA scores 14-20), significant anxiety group (HAMA scores 21-28), and severe anxiety group (HAMA scores >28) based on the HAMA scores. A cutoff point of 15 was used to determine the presence of psychotic symptoms [27]. The inter-observer correlation coefficients of the scores on the three scales were >0.8.
## Thyroid function and metabolic parameters
Before participants received any medication, their fasting blood samples were collected between 6 am and 8 am. Serum levels of the following biochemical parameters were assessed: free triiodothyronine (FT3), free thyroxine (FT4), thyroid stimulating hormone (TSH), anti-thyroglobulin (TgAb), thyroid peroxidases antibody (TPOAb), total cholesterol (TC), total triglycerides (TG), high-density lipoprotein (HDL-C), low-density lipoprotein (LDL-C), and fasting glucose. Thyroid hormones (TSH, TPOAb, TgAb, FT3, and FT4) were measured on a Roche C6000 Electrochemiluminescence Immunoassay Analyzer (Roche Diagnostics, Indianapolis, IN, USA), while metabolic parameters (HDL-C, LDL-C, TC, TG, and glucose) were assessed on a Cobas E610 (Roche, Basel, Switzerland) in the laboratory of Shanxi Medical University. The blood pressure, height, and weight were taken by trained nurses. Body mass index (BMI) was calculated with the equation: BMI= Weight (kg)/Height (m)2. The definition of metabolic disturbances and thyroid dysfunction were as follows: [1] overweight or obesity: BMI≥24; [2] hyperglycemia: glucose≥6.1mmol/L; [3] hypertension: SBP≥140 mmHg and/or DBP≥90mmHg; [4] hypertriglyceridemia: TG≥2.3 mmol/L; [5] low HDL: HDL-C ≤ 1.0 mmol/L; [6] hypercholesterolemia: TC≥6.2 mmol/L or LDL-C≥4.1 mmol/L; [7]abnormal TgAb: TgAb≥115 IU/L; [8] abnormal TPOAb: TPOAb ≥34 IU/L; [9] subclinical hypothyroidism (SCH): TSH >4.2 mIU/L with a normal fT4 concentration (10–23 pmol/L); [10] hyperthyroidism: TSH<0.27 mIU/L and FT4 over 23 pmol/L, and [11] hypothyroidism: TSH >4.2 mIU/L with a low FT4 concentration (<10 pmol/L).
## Statistical analysis
According to the Shapiro-Wilk test, the continuous variables in this study were not normally distributed. Therefore, we denoted continuous variables as the median and interquartile range (IRQ; 25-$75\%$) and categorical variables as frequencies and percentages. All statistical analyses were performed on R(ver.4.20). We conducted the partial correlation test to evaluate the association between clinical symptoms (HAMA, HAMD, PANSS), thyroid hormones (TPOAb, TgAb, TSH, FT3, FT4), and metabolic parameters (HDL-C, LDL-C, TC, TG, glucose, BMI, SBP, and DBP). Basic information including age, age at first episode onset, illness duration, gender, education, and married status were controlled as covariates.
## Undirected network analysis
Thyroid hormones, metabolic parameters, and clinical symptoms were included in the network analyses. Following previous studies, we converted the data set to normal using a non-normal transformation through the R package “huge” [28]. The network was estimated using the EBICglasso model, which is the most widely used model in psychopathology networks [29, 30]. The R packages “qgraph” and “bootnet” were applied for the visualization of the network. In the network, each “node” represented one item of thyroid hormones, metabolic parameters, and clinical information. After adjusting the other nodes in the network, the unique connection between two nodes is visualized as an “edge” between the nodes. The thickness and color of the edges indicate the strength and direction of the relationship. The thicker the edge, the stronger the association. Red and blue edges represent negative and positive associations, respectively.
We calculate the centrality index “strength” to identify the central nodes in the network. The strength is the sum of the absolute edge weights that directly connect a node to the other nodes in the network. Nodes with the highest strength are considered central nodes, and they have the strongest influence on other variables in the network. Central nodes are important in the creation of networks (31–33) and can be used as potential targets for clinical interventions. We also assessed the “predictability” of each node by Rpackage “MGM”. Similar to the adjusted R2 in regression models, the predictability of a node indicates the extent to which the variance of a node can be predicted by other nodes in the network [34]. Nodes with high predictability might be easily changed by changing their related nodes.
The case-dropping procedure was applied to evaluate the stability of our network. The correlation stability coefficient (CS-C) of node strength was calculated, with CS-C above 0.5 implying high stability. To test the accuracy of edges estimated within the network, we conducted the nonparametric bootstrapping with 1000 bootstrap samples.
We used a case-dropping procedure to assess the stability of our network. The correlation stability coefficient (CS-C) of the node strength was calculated, and a CS-C above 0.5 indicated high stability. To test the accuracy of the estimated edges within the network, we performed nonparametric bootstrapping with 1000 bootstrap samples.
We further compared our networks with the Rpackage “Network Comparison Tool”. Three subgroups were analyzed including gender (female versus male), illness duration (≤5 months versus >5 months), and age (18-45 years old versus 46-60 years old). The overall strength (absolute sum value of all edge weights) and network structure (distribution of edge weights) were evaluated between subgroups of networks.
## Bayesian network analysis
Finally, we performed directed acyclic graphs (DAG) to assess the putative direction of causal relationships between thyroid dysfunction, metabolic disorders, and clinical symptoms. DAG is an emerging method for network analysis. It allows us to detect and represent the most likely direction of causality between variables based on the conditional dependence between each pair of variables in the presence of other variables in the network [35, 36]. More importantly, studies have shown that variables at the top of the DAG may have a higher predictive priority and more salience and should receive higher priority in treatment [37, 38]. Following previous studies [36, 39], we performed DAG by the R package “bnlearn”. A hill-climbing algorithm was chosen to estimate the network. According to the protocol of Scutari M & Nagarajan R [40], the optimal cut point method was used to retain edges with high sensitivity and specificity.
Figure 3 illustrates the DAG of thyroid disorders, metabolic disorders, and clinical symptoms in patients with FEDN MDD. Consistent with the undirected network analysis, HAMD, TSH, and TC were located at the top of the DAG, indicating that they may trigger and maintain metabolic disorders, thyroid dysfunction, and clinical symptoms in patients with FEDN MDD. Similarly, we found that TSH was closely associated with metabolic disorders. It was upstream of glucose disturbance, hypertension, obesity, and HDL-C. The DAG also suggests that TSH may contribute to more severe psychotic symptoms.
**Figure 3:** *The Bayesian network of thyroid hormones, metabolic parameters, and clinical symptoms. HAMD, Hamilton Depression Rating Scale; HAMA, Hamilton Anxiety Rating Scale; PANSS, the Positive and Negative Syndrome Scale; TSH, thyroid-stimulating hormone; FT3, free triiodothyronine; FT4, free thyroxine; TgAb, antithyroglobulin; TPOAb, thyroid peroxidases antibody; TC, total cholesterol; HDL-C, high-density lipoprotein; LDL-C, low-density lipoprotein; TG, total triglycerides; BMI, body mass index.*
## Sample characteristics
A total of 1718 patients (female: 1130, male: 588) with FEDN MDD were recruited (Table 1). The median age at the first episode onset was 34 [23, 41] years. $43\%$ ($$n = 740$$), $32\%$ ($$n = 549$$), and $25\%$ ($$n = 429$$) of the participants were between 18-30, 31-45, and 46-60 years old, respectively. 171 ($10\%$) suffered from psychotic symptoms. According to HAMA scores, the participants could be divided into without or possible anxiety ($$n = 1$$, $0.1\%$), mild to moderate anxiety ($$n = 853$$, $47.9\%$), significant anxiety ($$n = 849$$, $49.4\%$), and severe anxiety ($$n = 45$$, $2.6\%$). The prevalence of thyroid dysfunction and metabolic disorders was as follows: SCH ($$n = 1041$$, $61\%$), TgAb abnormalities ($$n = 297$$, $17\%$), TPOAb abnormalities ($$n = 438$$, $25\%$), hyperthyroidism ($$n = 5$$, $0.3\%$), hypothyroidism ($$n = 3$$, 0. $2\%$), hyperglycemia ($$n = 241$$, $14\%$), hypertriglyceridemia ($$n = 668$$, $39\%$)), low HDL-C ($$n = 429$$, $25\%$), hypercholesterolemia (421, $25\%$), abnormal TC (357, $21\%$), abnormal LDL-C (185, $11\%$), overweight or obese ($$n = 1026$$, $60\%$) and hypertensive ($$n = 92$$, $5.4\%$).
**Table 1**
| Variable | Overall, N = 1,7181 |
| --- | --- |
| Age, year | 34 (23, 45) |
| Age groups | Age groups |
| 18-30 | 740 (43%) |
| 31-45 | 549 (32%) |
| 46-60 | 423 (25%) |
| Duration, month | 5 (3, 8) |
| Age at first episode onset, year | 34 (23, 45) |
| Gender | |
| Male | 588 (34%) |
| Female | 1,130 (66%) |
| Education | Education |
| Below college | 1,173 (68%) |
| College or above | 545 (32%) |
| Married | 1,216 (71%) |
| PANSS | 7 (7, 7.8) |
| Psychotic symptom | 171 (10.0%) |
| HAMD | 30 (28, 32) |
| HAMA | 21 (18, 23) |
| Anxiety | Anxiety |
| Without or possible anxiety | 1 (0.1%) |
| Mild to moderate anxiety | 853 (47.9%) |
| Significant anxiety | 849 (49.4%) |
| Severe anxiety | 45 (2.6%) |
| TSH, uIU/L | 4.91 (3.11, 6.66) |
| TgAb, IU/L | 21 (14, 44) |
| TPOAb, IU/L | 17 (12, 35) |
| FT3, pmol/L | 4.92 (4.38, 5.41) |
| FT4, pmol/L | 16.5 (14.4, 18.7) |
| Glucose, mmol/L | 5.34 (4.94, 5.80) |
| TC, mmol/L | 5.22 (4.46, 6.00) |
| HDLC, mmol/L | 1.23 (1.01, 1.42) |
| TG, mmol/L | 1.97 (1.40, 2.77) |
| LDLC, mmol/L | 2.96 (2.38, 3.52) |
| BMI, kg/m2 | 24.23 (23.22, 25.60) |
| SBP, mmHg | 120 (112, 127) |
| DBP, mmHg | 76 (70, 80) |
| Abnormal TgAb | 297 (17%) |
| Abnormal TPOAb | 438 (25%) |
| SCH | 1,041 (61%) |
| Hyperthyroidism | 5 (0.3%) |
| Hypothyroidism | 3 (0.2%) |
| Hyperglycemia | 241 (14%) |
| Low HDL | 429 (25%) |
| Overweight or obesity | 1,026 (60%) |
| Hypertriglyceridemia | 668 (39%) |
| Abnormal TC | 357 (21%) |
| Abnormal LDL-C | 185 (11%) |
| Hypertension | 92 (5.4%) |
| Hypercholesterolemia | 421 (25%) |
## Correlation between thyroid dysfunction, metabolic disturbances, and clinical symptoms
We found that clinical symptoms, metabolic disturbances, and thyroid dysfunction were strongly correlated after adjusting for basic information (Table S1). Specifically, the partial correlation test demonstrated substantial inter-relationships between three clinical symptoms: HAMA-HAMD ($r = 0.615$), HAMD-PANSS ($r = 0.539$), and HAMA-PANSS ($r = 0.610$) (all $p \leq 0.001$). Notably, TSH levels were strongly and positively associated with HAMD ($r = 0.466$), HAMA ($r = 0.344$), PANSS ($r = 0.367$), glucose ($r = 0.444$), TC ($r = 0.545$), SBP ($r = 0.551$), and DBP ($r = 0.356$) (all $p \leq 0.001$). We also found a solid positive relationship between TC and HAMD ($r = 0.553$, $p \leq 0.001$).
## Undirected network of thyroid dysfunction, metabolic disturbance, and clinical symptoms in FEDN MDD patients
Figure 1 illustrates the network of thyroid dysfunction, metabolic disorders, and clinical symptoms in patients with FEND MDD. The network consists of 16 nodes with a density of 0.44 ($\frac{53}{120}$). Visually, the node TSH is located in the center of the network. It shows a strong relationship with metabolic disorders (i.e. SBP, Glucose, and TC). It was also positively correlated with HAMA, HAMD, and PANSS. We also observed a strong relationship between HAMA, HAMD, and PANSS. The ten strongest edges in the network were SBP-DBP, followed by TPOAb-TgAb, TC-LDL-C, HAMD-PANSS, HAMA-PANSS, TSH-Glucose, HAMD-HAMA, HAMD-TC, TSH-TC, and TSH-TC. According to the results of the nonparametric bootstrap procedure, these edges were statistically stronger than the other edges within the network (Figure S1). The value of the edges is listed in Table S2.
**Figure 1:** *The undirected network of thyroid hormones, metabolic parameters, and clinical symptoms The blue, green, and orange nodes represent thyroid hormones, metabolic parameters, and clinical profiles, respectively. Blue edges indicate positive association while red edges indicate negative association. Thicker edge implies a stronger association. HAMD, Hamilton Depression Rating Scale; HAMA, Hamilton Anxiety Rating Scale; PANSS, the Positive and Negative Syndrome Scale; TSH, thyroid-stimulating hormone; FT3, free triiodothyronine; FT4, free thyroxine; TgAb, antithyroglobulin; TPOAb, thyroid peroxidases antibody; TC, total cholesterol; HDL-C, high-density lipoprotein; LDL-C, low-density lipoprotein; TG, total triglycerides; BMI, body mass index.*
The centrality plot (Figure 2) shows that TSH, TC, and HAMD scores have the highest strength, implying that they are the central nodes in the network. They have a statistically higher strength than the other nodes (Figure S2). The predictability of the nodes within the network is shown in Table S3. Notably, the predictability of HAMD was 0.604, HAMA was 0.456, and PANSS was 0.516, indicating that half of the severity of depression, anxiety, and psychotic symptoms could be explained by the other nodes in the network.
**Figure 2:** *The central nodes of the network. The X-rays represented the strength of each node. Nodes with higher strength have stronger impact in other nodes within the network. HAMD, Hamilton Depression Rating Scale; HAMA, Hamilton Anxiety Rating Scale; PANSS, the Positive and Negative Syndrome Scale; TSH, thyroid-stimulating hormone; FT3, free triiodothyronine; FT4, free thyroxine; TgAb, antithyroglobulin; TPOAb, thyroid peroxidases antibody; TC, total cholesterol; HDL-C, high-density lipoprotein; LDL-C, low-density lipoprotein; TG, total triglycerides; BMI, body mass index.*
The network had good stability and accuracy (Figures S3, S4). The CS-C was 0.75, indicating that the network was highly correlated with the original network even after discarding $75\%$ of the original data ($r = 0.7$).
## Network comparison test
We compared the networks of thyroid dysfunction, metabolic disorders, and clinical symptoms according to gender, age, and illness duration (Figure S5). We did not observe any differences in the overall strength and network structure of the networks between the three subgroups.
## Discussion
To our knowledge, this is the first study to describe the relationship between thyroid dysfunction, metabolic disorders, and clinical symptoms in a large sample of FEDN MDD patients by a network approach. Our study highlights the common metabolic burden and thyroid dysfunction in patients with MDD. Both the undirected network and DAG showed a dominant role of TC, TSH, and depression severity in patients with MDD. Therefore, targeting abnormal TSH, TC, and depressive symptoms may hold great promise for reducing MDD-related thyroid dysfunction, metabolic disturbances, and clinical symptoms.
Our study suggested high TSH levels (i.e., SCH) played a vital role in triggering metabolic disturbances and clinical symptoms in FEDN MDD patients. In our study, the prevalence of SCH in our sample was $61\%$, much higher than that in the *Chinese* general population (range: $3.4\%$-$12.93\%$) (42–44). More importantly, we found that TSH levels were highly correlated with the severity of depressive symptoms in FEDN MDD patients. The shared biological mechanisms between MDD and SCH, such as disturbances in the hypothalamic–pituitary–adrenal axis and changes in the hormone levels including somatostatin and serotonin might contribute to their tight associations [45]. In recent years, studies also indicated that the Wnt/β-catenin pathway might play a role in the association between SCH and depression. Interestingly, the DAG indicated that the TSH levels might be the outcome rather than a cause of depressive symptoms in MDD. However, there were very few, if any, longitudinal studies that evaluated whether depressive symptoms could predict increased TSH levels. Further studies are needed to validate our findings.
In addition to depressive symptoms, we demonstrated a strong positive correlation between TSH levels and psychotic symptoms in patients with FEDN MDD. To date, only a few studies have evaluated their relationship and have yielded inconsistent results. For example, Liu et al. found that serum TSH levels were independently associated with psychotic symptoms in 1279 patients with MDD [11]. Another large study of 1410 patients with MDD showed that hypothyroidism was positively associated with psychotic symptoms [41]. However, Contreras et al. found no difference in TSH blunting in MDD patients with and without psychotic symptoms [46]. A possible explanation might be the complicated medication in the treatment of MDD and the varied sample size, which might impact the association between psychotic symptoms and TSH levels. Furthermore, most of the current evidence on the association between TSH and psychosis came from the observational cross-sectional study. The underlying mechanism is still largely unexplored. Hence, further studies were still needed to confirm our findings.
Consistent with previous studies in the general population [18, 47], high TSH levels exhibited substantial associations with impaired metabolism, including hypertension, hyperglycemia, obesity, and hypercholesterolemia. To date, only a few studies have investigated their relationship in the context of MDD [48, 49], and these studies reported similar results to ours. For example, Kim et al. found that SCH increased the risk of metabolic syndrome (MetS) approximately 7-fold in Korean adults with depression [48]. Zhao et al. found that SCH was independently associated with BMI, TC, and LDL-C in MDD patients [49]. A few possible explanations might account for the high co-occurrence of SCH and metabolic disturbances. First, studies have suggested that high TSH levels could directly impact lipid metabolism by binding the TSH receptor [50]. High TSH levels could promote cholesterol synthesis and inhibit cholesterol clearance, which is independent of FT3 and FT4 [50]. Second, recent studies suggested that hepatic endoplasmic reticulum stress induced by SCH played an important role in dyslipidemia and impaired glucose metabolism [51, 52]. Interestingly, a strong association between SCH and metabolic disorders has been reported in patients with other psychiatric disorders. For example, a recent study showed higher serum TSH levels in schizophrenia patients with comorbid MetS [53]. Taken together, the co-occurrence of SCH and metabolic disorders may be common in psychiatric patients, which calls for regular screening of this population.
Further, we found that high TC levels were another important factor in the network of thyroid dysfunction, clinical symptoms, and metabolic disorders in patients with FEDN MDD. Notably, it was highly correlated with the severity of depression in this study, which is consistent with some previous studies (54–56). There are several explanations for their strong association. First, MDD patients with more severe depressive symptoms may have a more irregular lifestyle, such as late nights and an unhealthy diet, which leads to increased TC levels. Second, inflammation may act as a bridge between hypercholesterolemia and MDD. Hypercholesterolemia has been found to trigger the activation of the NLRP3 inflammasome, which leads to a chronic state of inflammation [57]. A growing number of studies suggest a bidirectional relationship between MDD and inflammation [58, 59]. However, we did not collect the variables associated with inflammation. Therefore, further studies are needed to test our hypothesis.
The present study has several important clinical implications. First, our study highlights the heavy metabolic burden and thyroid dysfunction in patients with FEDN MDD, which calls for regular screening of this population. Second, we found that greater depressive symptoms, higher TSH levels, and higher TC levels may play an important role in metabolic disturbances, thyroid dysfunction, and clinical symptoms, which should be given higher priority in treatment. Several clinical trials have demonstrated the potential effectiveness of anti-SCH and anti-hypercholesterolemia in the treatment of MDD (60–62), which supported our findings. For example, several studies have demonstrated the efficacy and safety of statins (one of the most widely used drugs for the treatment of hypercholesterolemia) in MDD [60, 61]. A recent review suggests that thyroid hormone therapy may be a promising strategy for refractory MDD [62]. Thyroid hormone therapy was also found to improve lipid metabolism in SCH patients [63]. However, it should be noted that the results of thyroid hormone therapy in the treatment of MDD are inconsistent [64, 65], which warranted further studies.
The study has the following limitations. First, it should be acknowledged that the study was observational and cross-sectional, which inhibited us from drawing a causal relationship between clinical symptoms, metabolic disturbances, and thyroid dysfunction among patients with MDD. Second, despite the large sample size, this study was conducted in a single hospital. Whether our findings can be generalized to other populations remains unclear. Third, we did not control for several important potential confounders, such as diet, smoking, drinking, and pro-inflammatory factors. Therefore, further longitudinal studies which provide a more comprehensive assessment of lifestyle factors are needed to validate our findings.
In summary, this study described the association between thyroid dysfunction, metabolic disorders, and clinical symptoms in a large sample of FEDN MDD patients by both undirected and Bayesian network approaches. Both networks suggest that TSH, TC, and depression severity are of high importance in patients with MDD and should be treated with high priority.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by Institutional Review Board (IRB) of the First Hospital of Shanxi Medical University. The patients/participants provided their written informed consent to participate in this study.
## Author contributions
PP: Formal analysis, Writing - original draft. QW: Writing – review & editing. XL: Writing – review & editing. TL: Conceptualization, Writing–review & editing. X-YZ: Conceptualization, Writing–review & editing. All authors contributed to the article and approved the submitted version.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fendo.2023.1138233/full#supplementary-material
## References
1. Malhi GS, Mann JJ. **Depression**. *Lancet* (2018) **392**. DOI: 10.1016/S0140-6736(18)31948-2
2. Otte C, Gold SM, Penninx BW, Pariante CM, Etkin A, Fava M. **Major depressive disorder**. *Nat Rev Dis Primers* (2016) **2** 16066. DOI: 10.1038/nrdp.2016.66
3. **GBD 2019 diseases and injuries collaborators. global burden of 369 diseases and injuries in 204 countries and territories, 1990-2019: a systematic analysis for the global burden of disease study 2019**. *Lancet* (2020) **396**. DOI: 10.1016/S0140-6736(20)30925-9
4. Tang R, Wang J, Yang L, Ding X, Zhong Y, Pan J. **Subclinical hypothyroidism and depression: A systematic review and meta-analysis**. *Front Endocrinol (Lausanne)* (2019) **10**. DOI: 10.3389/fendo.2019.00340
5. Vancampfort D, Stubbs B, Mitchell AJ, De Hert M, Wampers M, Ward PB. **Risk of metabolic syndrome and its components in people with schizophrenia and related psychotic disorders, bipolar disorder and major depressive disorder: a systematic review and meta-analysis**. *World Psychiatry* (2015) **14**. DOI: 10.1002/wps.20252
6. Vancampfort D, Correll CU, Wampers M, Sienaert P, Mitchell AJ, De Herdt A. **Metabolic syndrome and metabolic abnormalities in patients with major depressive disorder: a meta-analysis of prevalences and moderating variables**. *Psychol Med* (2014) **44**. DOI: 10.1017/S0033291713002778
7. Han K-M, Kim MS, Kim A, Paik J-W, Lee J, Ham B-J. **Chronic medical conditions and metabolic syndrome as risk factors for incidence of major depressive disorder: A longitudinal study based on 4.7 million adults in south Korea**. *J Affect Disord* (2019) **257**. DOI: 10.1016/j.jad.2019.07.003
8. Fugger G, Dold M, Bartova L, Kautzky A, Souery D, Mendlewicz J. **Comorbid thyroid disease in patients with major depressive disorder - results from the European group for the study of resistant depression (GSRD)**. *Eur Neuropsychopharmacol* (2018) **28**. DOI: 10.1016/j.euroneuro.2018.03.011
9. Hough CM, Bersani FS, Mellon SH, Morford AE, Lindqvist D, Reus VI. **Pre-treatment allostatic load and metabolic dysregulation predict SSRI response in major depressive disorder: a preliminary report**. *Psychol Med* (2021) **51** 2117-25. DOI: 10.1017/S0033291720000896
10. Yang L, Yang X, Yang T, Wu X, Sun P, Zhu Y. **The effect of thyroid function on the risk of psychiatric readmission after hospitalization for major depressive disorder**. *Psychiatry Res* (2021) **305**. DOI: 10.1016/j.psychres.2021.114205
11. Liu W, Wu Z, Sun M, Zhang S, Yuan J, Zhu D. **Association between fasting blood glucose and thyroid stimulating hormones and suicidal tendency and disease severity in patients with major depressive disorder**. *Bosn J Basic Med Sci* (2022) **22**. DOI: 10.17305/bjbms.2021.6754
12. Bagherniya M, Khayyatzadeh SS, Avan A, Safarian M, Nematy M, Ferns GA. **Metabolic syndrome and its components are related to psychological disorders: A population based study**. *Diabetes Metab Syndr* (2017) **11 Suppl 2**. DOI: 10.1016/j.dsx.2017.04.005
13. Caye A, Pilz LK, Maia AL, Hidalgo MP, Furukawa TA, Kieling C. **The impact of selective serotonin reuptake inhibitors on the thyroid function among patients with major depressive disorder: A systematic review and meta-analysis**. *Eur Neuropsychopharmacol* (2020) **33**. DOI: 10.1016/j.euroneuro.2020.01.011
14. Lee SH, Paz-Filho G, Mastronardi C, Licinio J, Wong M-L. **Is increased antidepressant exposure a contributory factor to the obesity pandemic**. *Transl Psychiatry* (2016) **6** e759. DOI: 10.1038/tp.2016.25
15. Liu F-H, Hwang J-S, Kuo C-F, Ko Y-S, Chen S-T, Lin J-D. **Subclinical hypothyroidism and metabolic risk factors association: A health examination-based study in northern Taiwan**. *BioMed J* (2018) **41**. DOI: 10.1016/j.bj.2018.02.002
16. Li R-C, Zhang L, Luo H, Lei Y, Zeng L, Zhu J. **Subclinical hypothyroidism and anxiety may contribute to metabolic syndrome in sichuan of China: a hospital-based population study**. *Sci Rep* (2020) **10** 2261. DOI: 10.1038/s41598-020-58973-w
17. Ding X, Zhao Y, Zhu C-Y, Wu L-P, Wang Y, Peng Z-Y. **The association between subclinical hypothyroidism and metabolic syndrome: an update meta-analysis of observational studies**. *Endocr J* (2021) **68**. DOI: 10.1507/endocrj.EJ20-0796
18. He J, Lai Y, Yang J, Yao Y, Li Y, Teng W. **The relationship between thyroid function and metabolic syndrome and its components: A cross-sectional study in a Chinese population**. *Front Endocrinol (Lausanne)* (2021) **12**. DOI: 10.3389/fendo.2021.661160
19. Nakajima Y, Yamada M, Akuzawa M, Ishii S, Masamura Y, Satoh T. **Subclinical hypothyroidism and indices for metabolic syndrome in Japanese women: one-year follow-up study**. *J Clin Endocrinol Metab* (2013) **98**. DOI: 10.1210/jc.2013-1353
20. McInerney AM, Lindekilde N, Nouwen A, Schmitz N, Deschênes SS. **Diabetes distress, depressive symptoms, and anxiety symptoms in people with type 2 diabetes: A network analysis approach to understanding comorbidity**. *Diabetes Care* (2022) **45**. DOI: 10.2337/dc21-2297
21. Borsboom D, Deserno MK, Rhemtulla M, Epskamp S, Fried EI, McNally RJ. **Network analysis of multivariate data in psychological science**. *Nat Rev Methods Primers* (2021) **1** 1-18. DOI: 10.1038/s43586-021-00055-w
22. Zainal NH, Newman MG. **Prospective network analysis of proinflammatory proteins, lipid markers, and depression components in midlife community women**. *Psychol Med* (2022) **4** 1-12. DOI: 10.1017/S003329172200232X
23. Garcia-Mondragon L, Konac D, Newbury JB, Young KS, Ing A, Fürtjes AE. **Role of polygenic and environmental factors in the co-occurrence of depression and psychosis symptoms: a network analysis**. *Transl Psychiatry* (2022) **12** 259. DOI: 10.1038/s41398-022-02022-9
24. Duan Y, Wei J, Geng W, Jiang J, Zhao X, Li T. **Research on cognitive function in anxious depression patients in China**. *J Affect Disord* (2021) **280**. DOI: 10.1016/j.jad.2020.11.078
25. Li Z, Xue M, Zhao L, Zhou Y, Wu X, Xie X. **Comorbid major depression in first-episode drug-naïve patients with schizophrenia: Analysis of the depression in schizophrenia in China (DISC) study**. *J Affect Disord* (2021) **294**. DOI: 10.1016/j.jad.2021.06.075
26. Zhou Y, Wang Q, Ren H, Yang WFZ, Ma Y, Wu Q. **Prevalence and related factors of anxiety in first episode and drug naïve Chinese han outpatients with psychotic major depression**. *J Affect Disord* (2022) **316**. DOI: 10.1016/j.jad.2022.08.018
27. Peng P, Wang Q, Ren H, Zhou Y, Hao Y, Chen S. **Association between thyroid hormones and comorbid psychotic symptoms in patients with first-episode and drug-naïve major depressive disorder**. *Psychiatry Res* (2023) **320** 115052. DOI: 10.1016/j.psychres.2023.115052
28. Skjerdingstad N, Johnson MS, Johnson SU, Hoffart A, Ebrahimi OV. **Feelings of worthlessness links depressive symptoms and parental stress: A network analysis during the COVID-19 pandemic**. *Eur Psychiatry* (2021) **64**. DOI: 10.1192/j.eurpsy.2021.2223
29. Epskamp S, Borsboom D, Fried EI. **Estimating psychological networks and their accuracy: A tutorial paper**. *Behav Res* (2018) **50** 195-212. DOI: 10.3758/s13428-017-0862-1
30. Hevey D. **Network analysis: a brief overview and tutorial**. *Health Psychol Behav Med* (2018) **6**. DOI: 10.1080/21642850.2018.1521283
31. Wang Y, Hu Z, Feng Y, Wilson A, Chen R. **Changes in network centrality of psychopathology symptoms between the COVID-19 outbreak and after peak**. *Mol Psychiatry* (2020) **25**. DOI: 10.1038/s41380-020-00881-6
32. Fonseca-Pedrero E. **Network analysis: A new way of understanding psychopathology**. *Rev Psiquiatría Y Salud Ment (English Ed)* (2017) **10**. DOI: 10.1016/j.rpsmen.2017.10.005
33. Malgaroli M, Calderon A, Bonanno GA. **Networks of major depressive disorder: A systematic review**. *Clin Psychol Rev* (2021) **85**. DOI: 10.1016/j.cpr.2021.102000
34. Ren L, Wang Y, Wu L, Wei Z, Cui L-B, Wei X. **Network structure of depression and anxiety symptoms in Chinese female nursing students**. *BMC Psychiatry* (2021) **21** 279. DOI: 10.1186/s12888-021-03276-1
35. McNally RJ. **Can network analysis transform psychopathology**. *Behav Res Ther* (2016) **86** 95-104. DOI: 10.1016/j.brat.2016.06.006
36. Jones PJ, Mair P, Riemann BC, Mugno BL, McNally RJ. **A network perspective on comorbid depression in adolescents with obsessive-compulsive disorder**. *J Anxiety Disord* (2018) **53** 1-8. DOI: 10.1016/j.janxdis.2017.09.008
37. França AB, Gordon AL, Samra R, Rodolpho Duarte ES, Jacinto AF. **Symptoms of mood disorders in family carers of older people with dementia who experience caregiver burden: a network approach**. *Age Ageing* (2020) **49**. DOI: 10.1093/ageing/afaa008
38. Zhou J, Fan A, Zhou X, Pao C, Xiao L, Feng Y. **Interrelationships between childhood maltreatment, depressive symptoms, functional impairment, and quality of life in patients with major depressive disorder: A network analysis approach**. *Child Abuse Negl* (2022) **132**. DOI: 10.1016/j.chiabu.2022.105787
39. McNally RJ, Mair P, Mugno BL, Riemann BC. **Co-Morbid obsessive-compulsive disorder and depression: a Bayesian network approach**. *Psychol Med* (2017) **47**. DOI: 10.1017/S0033291716003287
40. Scutari M, Nagarajan R. **Identifying significant edges in graphical models of molecular networks**. *Artif Intell Med* (2013) **57**. DOI: 10.1016/j.artmed.2012.12.006
41. Sharif K, Tiosano S, Watad A, Comaneshter D, Cohen AD, Shoenfeld Y. **The link between schizophrenia and hypothyroidism: a population-based study**. *Immunol Res* (2018) **66**. DOI: 10.1007/s12026-018-9030-7
42. Gu F, Ding G, Lou X, Wang X, Mo Z, Zhu W. **Incidence of thyroid diseases in zhejiang province, China, after 15 years of salt iodization**. *J Trace Elem Med Biol* (2016) **36** 57-64. DOI: 10.1016/j.jtemb.2016.04.003
43. Zhang J, Gu Y, Meng G, Zhang Q, Liu L, Wu H. **Association between dietary onion intake and subclinical hypothyroidism in adults: a population-based study from an iodine-replete area**. *Endocrine* (2021) **74**. DOI: 10.1007/s12020-021-02790-2
44. Li Y, Teng D, Ba J, Chen B, Du J, He L. **Efficacy and safety of long-term universal salt iodization on thyroid disorders: Epidemiological evidence from 31 provinces of mainland China**. *Thyroid* (2020) **30**. DOI: 10.1089/thy.2019.0067
45. Loh HH, Lim LL, Yee A, Loh HS. **Association between subclinical hypothyroidism and depression: an updated systematic review and meta-analysis**. *BMC Psychiatry* (2019) **19**. DOI: 10.1186/s12888-018-2006-2
46. Contreras F, Menchon JM, Urretavizcaya M, Navarro MA, Vallejo J, Parker G. **Hormonal differences between psychotic and non-psychotic melancholic depression**. *J Affect Disord* (2007) **100** 65-73. DOI: 10.1016/j.jad.2006.09.021
47. Chang C-H, Yeh Y-C, Caffrey JL, Shih S-R, Chuang L-M, Tu Y-K. **Metabolic syndrome is associated with an increased incidence of subclinical hypothyroidism - a cohort study**. *Sci Rep* (2017) **7** 6754. DOI: 10.1038/s41598-017-07004-2
48. Kim M-D, Yang H-J, Kang NR, Park JH, Jung Y-E. **Association between subclinical hypothyroidism and metabolic syndrome among individuals with depression**. *J Affect Disord* (2020) **264**. DOI: 10.1016/j.jad.2019.11.080
49. Zhao S, Chen Z, Wang X, Yao Z, Lu Q. **Increased prevalence of subclinical hypothyroidism in female hospitalized patients with depression**. *Endocrine* (2021) **72**. DOI: 10.1007/s12020-020-02490-3
50. Liu H, Peng D. **Update on dyslipidemia in hypothyroidism: the mechanism of dyslipidemia in hypothyroidism**. *Endocr Connect* (2022) **11**. DOI: 10.1530/EC-21-0002
51. Xu C, Zhou L, Wu K, Li Y, Xu J, Jiang D. **Abnormal glucose metabolism and insulin resistance are induced**. *Front Endocrinol (Lausanne)* (2019) **10**. DOI: 10.3389/fendo.2019.00303
52. Zhou L, Ding S, Li Y, Wang L, Chen W, Bo T. **Endoplasmic reticulum stress may play a pivotal role in lipid metabolic disorders in a novel mouse model of subclinical hypothyroidism**. *Sci Rep* (2016) **6**. DOI: 10.1038/srep31381
53. Kalinowska S, Trześniowska-Drukała B, Safranow K, Pełka-Wysiecka J, Kłoda K, Misiak B. **Association between thyroid function and metabolic syndrome in male and female schizophrenia patients**. *Psychiatry Res* (2019) **274**. DOI: 10.1016/j.psychres.2019.02.029
54. Wagner CJ, Musenbichler C, Böhm L, Färber K, Fischer A-I, von Nippold F. **LDL cholesterol relates to depression, its severity, and the prospective course**. *Prog Neuropsychopharmacol Biol Psychiatry* (2019) **92**. DOI: 10.1016/j.pnpbp.2019.01.010
55. Park JH, Jung SJ, Jung Y, Ahn SV, Lee E, Kim HC. **Association between the change of total cholesterol during adolescence and depressive symptoms in early adulthood**. *Eur Child Adolesc Psychiatry* (2021) **30**. DOI: 10.1007/s00787-020-01511-w
56. Beydoun MA, Beydoun HA, Dore GA, Fanelli-Kuczmarski MT, Evans MK, Zonderman AB. **Total serum cholesterol, atherogenic indices and their longitudinal association with depressive symptoms among US adults**. *Transl Psychiatry* (2015) **5**. DOI: 10.1038/tp.2015.4
57. Haneklaus M, O’Neill LAJ. **NLRP3 at the interface of metabolism and inflammation**. *Immunol Rev* (2015) **265** 53-62. DOI: 10.1111/imr.12285
58. Colasanto M, Madigan S, Korczak DJ. **Depression and inflammation among children and adolescents: A meta-analysis**. *J Affect Disord* (2020) **277**. DOI: 10.1016/j.jad.2020.09.025
59. Beurel E, Toups M, Nemeroff CB. **The bidirectional relationship of depression and inflammation: Double trouble**. *Neuron* (2020) **107**. DOI: 10.1016/j.neuron.2020.06.002
60. De Giorgi R, Rizzo Pesci N, Quinton A, De Crescenzo F, Cowen PJ, Harmer CJ. **Statins in depression: An evidence-based overview of mechanisms and clinical studies**. *Front Psychiatry* (2021) **12**. DOI: 10.3389/fpsyt.2021.702617
61. Kim S-W, Kang H-J, Jhon M, Kim J-W, Lee J-Y, Walker AJ. **Statins and inflammation: New therapeutic opportunities in psychiatry**. *Front Psychiatry* (2019) **10**. DOI: 10.3389/fpsyt.2019.00103
62. Bauer M, Whybrow PC. **Role of thyroid hormone therapy in depressive disorders**. *J Endocrinol Invest* (2021) **44**. DOI: 10.1007/s40618-021-01600-w
63. Zhao C, Wang Y, Xiao L, Li L. **Effect of levothyroxine on older patients with subclinical hypothyroidism: A systematic review and meta-analysis**. *Front Endocrinol (Lausanne)* (2022) **13**. DOI: 10.3389/fendo.2022.913749
64. Zhao T, Chen BM, Zhao XM, Shan ZY. **Subclinical hypothyroidism and depression: a meta-analysis**. *Transl Psychiatry* (2018) **8** 239. DOI: 10.1038/s41398-018-0283-7
65. Wildisen L, Feller M, Del Giovane C, Moutzouri E, Du Puy RS, Mooijaart SP. **Effect of levothyroxine therapy on the development of depressive symptoms in older adults with subclinical hypothyroidism: An ancillary study of a randomized clinical trial**. *JAMA Netw Open* (2021) **4**. DOI: 10.1001/jamanetworkopen.2020.36645
|
---
title: Does Long-Term Post-Bariatric Weight Change Differ Across Antidepressants?
authors:
- David E. Arterburn
- Matthew L. Maciejewski
- Theodore S. Z. Berkowitz
- Valerie A. Smith
- James E. Mitchell
- Chuan-Fen Liu
- Adenike Adeyemo
- Katharine A. Bradley
- Maren K. Olsen
journal: Annals of Surgery Open
year: 2022
pmcid: PMC10013150
doi: 10.1097/AS9.0000000000000114
license: CC BY 4.0
---
# Does Long-Term Post-Bariatric Weight Change Differ Across Antidepressants?
## Abstract
Mini-abstract: Bariatric surgery induces significant weight loss, but outcomes are highly variable. We found that sleeve gastrectomy and gastric bypass patients taking bupropion had greater weight loss than those taking selective serotonin reuptake inhibitors, although these differences may wane over time. Bupropion may be the first-line antidepressant of choice among patients considering bariatric surgery.
Supplemental Digital *Content is* available in the text.
### Objectives:
We sought to evaluate whether weight change up to 5 years after bariatric surgery differed by antidepressant class taken before surgery.
### Background:
Bariatric surgery induces significant weight loss, but outcomes are highly variable. The specific type of antidepressant used prior to surgery may be an important factor in long-term weight loss.
### Methods:
This retrospective cohort study from 2000 to 2016 compared the 5-year weight loss of 556 Veterans who were taking antidepressant monotherapy (bupropion, selective serotonin reuptake inhibitors [SSRIs], or serotonin-norepinephrine reuptake inhibitors [SNRIs]) before bariatric surgery (229 sleeve gastrectomy and 327 Roux-en-Y gastric bypass) versus 556 matched nonsurgical controls.
### Results:
Patients taking bupropion before sleeve gastrectomy had greater differential weight loss between surgical patients and matched controls than those taking SSRIs at 1 (8.9 pounds; $95\%$ confidence interval [CI], 1.6–16.3; $$P \leq 0.02$$) and 2 years (17.6 pounds; $95\%$ CI, 5.9–29.3; $$P \leq 0.003$$), but there was no difference at 5 years (11.9 pounds; $95\%$ CI, –8.9 to 32.8; $$P \leq 0.26$$). Findings were similar for gastric bypass patients taking bupropion compared to SSRIs at 1 (9.7 pounds; $95\%$ CI, 2.0–17.4; $$P \leq 0.014$$), 2 (12.0 pounds; $95\%$ CI, –0.5 to 24.5; $$P \leq 0.06$$), and 5 years (4.8 pounds; $95\%$ CI, –16.7 to 26.3; $$P \leq 0.66$$). No significant differences were observed comparing patients taking SNRI versus SSRI medications.
### Conclusions:
Sleeve gastrectomy and gastric bypass patients taking bupropion had greater weight loss than those taking SSRIs, although these differences may wane over time. Bupropion may be the first-line antidepressant of choice among patients with severe obesity considering bariatric surgery.
## INTRODUCTION
Bariatric surgery is currently the most effective intervention for inducing long-term weight loss among patients with severe obesity.1 However, the long-term weight trajectories of postsurgical patients vary widely,2 and there is a clear need for research that identifies modifiable preoperative clinical factors that can maximize long-term weight loss.
Depression is highly prevalent among patients who undergo bariatric surgery, with preoperative prevalence estimates ranging from $35\%$ to $44\%$.3,4 *Evidence is* mixed regarding the impact of depression on weight loss after bariatric surgery, with some research suggesting no effect and other studies finding reductions in weight loss.3–6 This literature is difficult to synthesize, owing to many studies with short durations of follow-up and wide variations in the types of antidepressant medications evaluated. Another consideration is the finding that bariatric surgery appears to decrease the absorption of some antidepressants,7 which might change their clinical effectiveness as well as alter their effects on weight loss.5–7 Different classes of antidepressant medications are known to have a clinically meaningful impact on long-term weight trajectories in general populations of adults with depression.8 Among the most commonly prescribed antidepressant medications, the norepinephrine and dopamine reuptake inhibitor (NDRI) medication, bupropion, is associated with weight loss, while selective serotonin reuptake inhibitors (SSRIs) and serotonin-norepinephrine reuptake inhibitors (SNRIs) have been largely found to be weight neutral or promote small amounts of weight gain. Given the high prevalence of depression among patients undergoing bariatric surgery, it could be clinically helpful to know if the preoperative choice of antidepressant medication has a significant impact on long-term surgical weight loss outcomes.
A few prior studies have attempted to identify whether antidepressant use is associated with weight change in surgical patients.9–12 Two studies found that patients taking antidepressants before surgery had similar weight loss at 1 year compared to patients not taking antidepressants,9,12 but 1 study found that patients continuing antidepressants after surgery lost less weight at 24 months than patients not continuing antidepressants.11 Patients taking SNRIs lost more weight than patients taking SSRIs in 1 study,12 but lost less weight than patients not taking antidepressants after surgery in another study.11 These studies tended to only examine follow-up up to 1 year, focused on a single surgical procedure (most often gastric bypass) from a single site, and did not contrast weight change of surgical and matched nonsurgical control patients.
To address these gaps, we conducted a retrospective observational study involving a multisite cohort of Veterans with depression undergoing bariatric surgery matched to nonsurgical controls. The purpose was to examine whether postsurgical weight change differed between surgical patients and nonsurgical controls on 3 classes of antidepressants. We compared patients who were prescribed bupropion and SNRIs before surgery to those taking the more commonly prescribed SSRIs at baseline. We hypothesized that bupropion use at baseline would be associated with more weight loss than SSRIs and that SSRI and SNRI medications would have similar weight loss.
## Study Design and Study Population
In this retrospective cohort study based on data from Veterans Affair’s (VA’s) Corporate Data Warehouse that aggregates electronic health record data from all VA medical centers nightly, we initially identified 10,653 Veterans who underwent a bariatric operation at any VA bariatric center between October 1, 2000, and September 30, 2016. The Institutional Review Boards of the Durham VA and Kaiser Permanente Washington Health Research Institute approved this study. We excluded those patients who had reoperations, did not have documented obesity, had preoperative cancer, had other medical exclusions, received uncommon bariatric procedures, and patients who either [1] did not have depression, [2] were not receiving antidepressant treatment, or [3] or were receiving multiple antidepressant medications at the time of surgery (eFigure 1, http://links.lww.com/AOSO/A89). The final surgical cohort included 229 patients who underwent a laparoscopic sleeve gastrectomy and 327 patients who underwent a Roux-en-Y gastric bypass operation who were receiving antidepressant monotherapy at the time of surgery. The cohort was stratified based on antidepressant class into those taking only 1 of 3 antidepressant classes: SSRI, SNRI, or NDRI only. Antidepressant use was defined as having 2 or more fills of 30+ days duration in the 180 days prior to their surgery date (see eTable 1, http://links.lww.com/AOSO/A89 for specific medications included). Medication data was extracted from VA’s Pharmacy Benefits Management databases.
These surgical patients were matched 1-to-1 to comparable nonsurgical controls using sequential stratification matching to accommodate the time-varying nature of surgical eligibility characteristics (eg, body mass index [BMI]) after being identified from the VA electronic health record.13–15 Data were organized into a series of n-of-1 randomized clinical trials, in which the trial start date was the surgical date of each patient. Matching was conducted within each of the antidepressant classes. For example, for each surgical patient with SSRI monotherapy, we identified a pool of eligible potential control patients with SSRI monotherapy who had not yet had a bariatric surgical procedure but had similar binary characteristics associated with the outcomes and the likelihood of receiving a surgical procedure (eg, sex, race/ethnicity [White or non-White], diabetes diagnosis [all-time lookback], chronic prescription opioid use, unhealthy alcohol use [2-year lookback], and alcohol use disorder [2-year lookback] documented in the VA databases). Surgical patients and potential control patients were required to be within 5 years of age and have similar baseline BMI measurements (calculated as weight in kilograms divided by height in meters squared) within 6 months before the surgical date. Surgical patients without a representative match were excluded (eFigure 1, http://links.lww.com/AOSO/A89). The majority ($$n = 401$$ of 556) of surgical patients had more than 1 eligible match. Among those with more than 1 potential match, the match with most weight measurements during the study period was chosen as the best match. The final control cohort included 229 patients who matched to 229 laparoscopic sleeve gastrectomy patients and 327 patients who matched to 327 Roux-en-Y gastric bypass patients.
## Weight Change Outcome
Follow-up weight data up to 6 years after the index date were obtained from measurements recorded in the electronic health record during outpatient visits between October 1, 1999, and March 31, 2020. For each individual surgical patient and match, the analytic model included all valid weights measured from 1 year prior to the surgical patient’s date of surgery through 5 years after surgery. The primary comparisons of interest were the difference in change in weight between surgical and nonsurgical patients from baseline (index or surgery) to 1, 2, and 5 years after surgery, comparing patients taking bupropion and SSRIs at baseline and those taking SNRIs and SSRIs at baseline, separately.
## Statistical Analysis
Descriptive summary statistics and individual trajectory plots and penalized B-spline plots by antidepressant class and surgery type were used to guide initial model specification. As expected, surgical patients’ weights decreased after surgery, with a leveling off between 6 and 18 months postsurgery, depending upon the type of surgery. Weight trajectories among the matched nonsurgical controls showed little change over time.
Potential mean model specification included quadratic, cubic, quartic, and parametric piece-wise linear splines, with separately specified random effects for cases and matches. Final model selection was guided by lowest Akaike information criterion indices. Our final model included separate intercepts for each antidepressant class within surgical cases and matches, linear time, linear time interactions for each antidepressant class, linear piece-wise splines at surgery, 6 months, 1, 2, 3, and 4 years for the surgical cases, and interactions of the piece-wise splines with antidepressant class. Random effect specification included intercepts, linear, and quadratic time, grouped by surgical case indicator. This model specification resulted in 6 estimated mean trajectories over the 5-year study period—one for each antidepressant class, by surgical case and match and was fit with PROC MIXED in SAS v9.4 (Cary, NC). Estimate statements were used to generate estimated change in weights in pounds at 1, 2, and 5 years of follow-up within each antidepressant class and case or match subgroup and to estimate comparisons of cases and matches over time between the antidepressant classes.
## Descriptive Statistics
Patients undergoing sleeve gastrectomy and controls taking antidepressants before surgery were well balanced on the covariates used for matching (top of Table 1), with an average age of 52–53 years, average BMI of 42.4–43.6 kg/m2 and $25\%$–$39\%$ female. The SSRIs were more commonly taken at baseline ($$n = 162$$ sleeve patients) than either SNRIs ($$n = 39$$ sleeve patients) or NDRIs ($$n = 28$$ sleeve patients). There were several covariates (eg, marital status, diagnosed anxiety, diagnosed post-traumatic stress disorder) not used in matching that were imbalanced (bottom of Table 1). A smaller proportion of patients taking bupropion were of White race or had diagnosed diabetes than patients taking SNRIs or SSRIs.
**TABLE 1.**
| Variables | NDRIs Only | NDRIs Only.1 | SNRIs Only | SNRIs Only.1 | SSRIs Only | SSRIs Only.1 |
| --- | --- | --- | --- | --- | --- | --- |
| Variables | Sleeve (n = 28) | Controls (n = 28) | Sleeve (n = 39) | Controls (n = 39) | Sleeve (n = 162) | Controls (n = 162) |
| Variables used in match | Variables used in match | Variables used in match | Variables used in match | Variables used in match | Variables used in match | Variables used in match |
| Female, N (%) | 11 (39.3) | 11 (39.3) | 10 (25.6) | 10 (25.6) | 42 (25.9) | 42 (25.9) |
| Age, mean (SD) | 52.9 (8.2) | 53.4 (8.4) | 51.9 (10.5) | 53.0 (10.5) | 53.9 (9.8) | 53.9 (10.0) |
| BMI, mean (SD) | 43.6 (5.6) | 42.7 (5.2) | 42.7 (5.0) | 42.4 (5.0) | 43.3 (5.9) | 42.9 (5.4) |
| Race, White, N (%) | 21 (75.0) | 21 (75.0) | 35 (89.7) | 35 (89.7) | 142 (87.7) | 142 (87.7) |
| Diagnosed diabetes, N (%)* | 10 (35.7) | 10 (35.7) | 23 (59.0) | 23 (59.0) | 92 (56.8) | 92 (56.8) |
| Alcohol misuse, N (%) | 0 (0.0) | 0 (0.0) | 1 (2.6) | 1 (2.6) | 18 (11.1) | 18 (11.1) |
| Chronic prescription opioid use, N (%) | 6 (21.4) | 6 (21.4) | 18 (46.2) | 18 (46.2) | 57 (35.2) | 57 (35.2) |
| Diagnosed alcohol use disorder, N (%)* | 1 (3.6) | 1 (3.6) | 1 (2.6) | 1 (2.6) | 17 (10.5) | 17 (10.5) |
| Diagnosed opioid use disorder, N (%)* | 0 (0.0) | 0 (0.0) | 0 (0.0) | 0 (0.0) | 0 (0.0) | 0 (0.0) |
| Variables not used in match | Variables not used in match | Variables not used in match | Variables not used in match | Variables not used in match | Variables not used in match | Variables not used in match |
| DCG risk score, mean (SD) | NA (NA) | NA (NA) | NA (NA) | NA (NA) | 0.3 (NA) | 0.9 (NA) |
| Nosos risk score, mean (SD) | 1.8 (1.1) | 1.9 (1.0) | 1.9 (1.0) | 1.8 (1.0) | 2.0 (1.1) | 1.8 (1.1) |
| Married, N (%) | 16 (57.1) | 12 (42.9) | 24 (61.5) | 15 (38.5) | 85 (52.5) | 65 (40.1) |
| Previously married, N (%) | 7 (25.0) | 9 (32.1) | 11 (28.2) | 12 (30.8) | 58 (35.8) | 53 (32.7) |
| Unmarried or unknown, N (%) | 5 (17.9) | 7 (25.0) | 4 (10.3) | 12 (30.8) | 19 (11.7) | 44 (27.2) |
| VA outpatient mental health visits, mean (SD) | 11.6 (9.3) | 18.0 (29.6) | 18.6 (13.9) | 14.5 (13.6) | 14.6 (17.6) | 14.0 (21.5) |
| Diagnosed PTSD, N (%)* | 6 (21.4) | 12 (42.9) | 19 (48.7) | 14 (35.9) | 58 (35.8) | 57 (35.2) |
| Diagnosed cannabis disorder, N (%)* | 0 (0.0) | 0 (0.0) | 0 (0.0) | 3 (7.7) | 1 (0.6) | 3 (1.9) |
| Diagnosed other drug disorder, N (%)* | 2 (7.1) | 2 (7.1) | 0 (0.0) | 1 (2.6) | 7 (4.3) | 4 (2.5) |
| Diagnosed anxiety, N (%)* | 8 (28.6) | 9 (32.1) | 11 (28.2) | 15 (38.5) | 44 (27.2) | 37 (22.8) |
| Diagnosed bipolar, N (%)* | 2 (7.1) | 3 (10.7) | 2 (5.1) | 2 (5.1) | 9 (5.6) | 13 (8.0) |
| Diagnosed psychosis, N (%)* | 0 (0.0) | 0 (0.0) | 0 (0.0) | 1 (2.6) | 2 (1.2) | 2 (1.2) |
| Diagnosed schizophrenia, N (%)* | 1 (3.6) | 1 (3.6) | 0 (0.0) | 0 (0.0) | 2 (1.2) | 3 (1.9) |
| Diagnosed eating disorder, N (%)* | 3 (10.7) | 0 (0.0) | 0 (0.0) | 0 (0.0) | 2 (1.2) | 1 (0.6) |
| Diagnosed tobacco use disorder, N (%)* | 3 (10.7) | 4 (14.3) | 6 (15.4) | 7 (17.9) | 19 (11.7) | 23 (14.2) |
| Diagnosed GERD, N (%)* | 13 (46.4) | 10 (35.7) | 14 (35.9) | 9 (23.1) | 48 (29.6) | 39 (24.1) |
Patients undergoing gastric bypass and matched controls taking antidepressants before surgery were also well matched on the covariates used for matching (top of Table 2), with an average age of 51–53 years, average BMI of 41.8–46.4 kg/m2 and $28\%$–$34\%$ female. The SSRIs were more commonly taken at baseline ($$n = 266$$ gastric bypass patients) than either SNRIs ($$n = 32$$ gastric bypass patients) or NDRIs ($$n = 29$$ gastric bypass patients). Again, there were several covariates (eg, marital status, diagnosed post-traumatic stress disorder, smoking status) not used in matching that were imbalanced (bottom of Table 2). Both the sleeve and gastric bypass patients and matched controls had similar numbers of weight observations and rates of follow-up among the antidepressant therapy classes (eTable 2, http://links.lww.com/AOSO/A89).
**TABLE 2.**
| Variables | NDRIs Only | NDRIs Only.1 | SNRIs Only | SNRIs Only.1 | SSRIs Only | SSRIs Only.1 |
| --- | --- | --- | --- | --- | --- | --- |
| Variables | RYGB (n = 29) | Controls (n = 29) | RYGB (n = 32) | Controls (n = 32) | RYGB (n = 266) | Controls (n = 266) |
| Variables used in match | Variables used in match | Variables used in match | Variables used in match | Variables used in match | Variables used in match | Variables used in match |
| Female, N (%) | 8 (27.6) | 8 (27.6) | 11 (34.4) | 11 (34.4) | 83 (31.2) | 83 (31.2) |
| Age, mean (SD) | 50.9 (6.6) | 51.3 (8.0) | 51.2 (8.3) | 51.4 (8.1) | 53.0 (8.0) | 52.9 (8.3) |
| BMI, mean (SD) | 44.1 (6.0) | 41.8 (4.3) | 45.0 (6.2) | 43.8 (5.5) | 46.4 (7.2) | 45.3 (6.5) |
| Race, White, N (%) | 22 (75.9) | 22 (75.9) | 28 (87.5) | 28 (87.5) | 227 (85.3) | 227 (85.3) |
| Diagnosed diabetes, N (%)* | 17 (58.6) | 17 (58.6) | 19 (59.4) | 19 (59.4) | 155 (58.3) | 155 (58.3) |
| Alcohol misuse, N (%) | 0 (0.0) | 0 (0.0) | 1 (3.1) | 1 (3.1) | 7 (2.6) | 7 (2.6) |
| Chronic prescription opioid use, N (%) | 13 (44.8) | 13 (44.8) | 14 (43.8) | 14 (43.8) | 91 (34.2) | 91 (34.2) |
| Diagnosed alcohol use disorder, N (%)* | 2 (6.9) | 2 (6.9) | 2 (6.3) | 2 (6.3) | 20 (7.5) | 20 (7.5) |
| Diagnosed opioid use disorder, N (%)* | 0 (0.0) | 0 (0.0) | 1 (3.1) | 1 (3.1) | 1 (0.4) | 1 (0.4) |
| Variables not used in match | Variables not used in match | Variables not used in match | Variables not used in match | Variables not used in match | Variables not used in match | Variables not used in match |
| DCG risk score, mean (SD) | 0.7 (0.5) | 0.6 (0.4) | 0.5 (0.3) | 0.4 (0.1) | 0.8 (0.6) | 0.9 (0.6) |
| Nosos risk score, mean (SD) | 2.0 (0.7) | 2.2 (1.6) | 1.9 (0.8) | 1.8 (0.9) | 1.9 (1.0) | 1.9 (1.3) |
| Married, N (%) | 10 (34.5) | 12 (41.4) | 14 (43.8) | 13 (40.6) | 150 (56.4) | 116 (43.6) |
| Previously married, N (%) | 15 (51.7) | 10 (34.5) | 13 (40.6) | 12 (37.5) | 80 (30.1) | 67 (25.2) |
| Unmarried or unknown, N (%) | 4 (13.8) | 7 (24.1) | 5 (15.6) | 7 (21.9) | 36 (13.5) | 83 (31.2) |
| VA outpatient mental health visits, mean (SD) | 15.8 (19.9) | 21.3 (26.8) | 13.9 (14.1) | 12.6 (11.7) | 12.3 (16.8) | 11.6 (24.6) |
| Diagnosed PTSD, N (%)* | 7 (24.1) | 7 (24.1) | 12 (37.5) | 16 (50.0) | 86 (32.3) | 70 (26.3) |
| Diagnosed cannabis disorder, N (%)* | 0 (0.0) | 0 (0.0) | 0 (0.0) | 0 (0.0) | 4 (1.5) | 5 (1.9) |
| Diagnosed other drug disorder, N (%)* | 1 (3.4) | 2 (6.9) | 2 (6.3) | 1 (3.1) | 9 (3.4) | 11 (4.1) |
| Diagnosed anxiety, N (%)* | 8 (27.6) | 8 (27.6) | 7 (21.9) | 12 (37.5) | 51 (19.2) | 55 (20.7) |
| Diagnosed bipolar, N (%)* | 5 (17.2) | 2 (6.9) | 2 (6.3) | 4 (12.5) | 19 (7.1) | 12 (4.5) |
| Diagnosed psychosis, N (%)* | 0 (0.0) | 1 (3.4) | 1 (3.1) | 0 (0.0) | 2 (0.8) | 4 (1.5) |
| Diagnosed schizophrenia, N (%)* | 0 (0.0) | 2 (6.9) | 0 (0.0) | 0 (0.0) | 1 (0.4) | 18 (6.8) |
| Diagnosed eating disorder, N (%)* | 0 (0.0) | 1 (3.4) | 1 (3.1) | 1 (3.1) | 7 (2.6) | 0 (0.0) |
| Diagnosed tobacco use disorder, N (%)* | 3 (10.3) | 6 (20.7) | 3 (9.4) | 5 (15.6) | 20 (7.5) | 39 (14.7) |
| Diagnosed GERD, N (%)* | 7 (24.1) | 10 (34.5) | 13 (40.6) | 8 (25.0) | 72 (27.1) | 61 (22.9) |
## Differences in Weight Change by Antidepressant Monotherapy in Sleeve Gastrectomy Cohort
In the sleeve gastrectomy cohort, mean weights of surgical and nonsurgical patients were similar at baseline (Fig. 1), but surgical patients’ mean weight change 1 year after surgery was 59–69 pounds lower than matched controls (depending on their antidepressant used at baseline; see top of Table 3). Compared with patients taking SSRIs at baseline, patients taking NDRIs at baseline had greater differential weight loss between surgical patients and matched controls at 1 year (8.9 pounds; $95\%$ confidence interval [CI], 1.6–16.3; $$P \leq 0.02$$; Table 3).
Sleeve gastrectomy patients’ weight 2 years after surgery was 49–68 pounds lower than matched controls (depending on antidepressant). The relative improvement in weight loss among sleeve gastrectomy patients compared with matched controls was again greater for those taking NDRIs at baseline compared with those taking SSRIs (17.6 pounds; $95\%$ CI, 5.9–29.3; $$P \leq 0.003$$).
Five years after surgery, sleeve gastrectomy patients’ weight change was 30–47 pounds lower than matched controls (depending on antidepressant), but there were no statistically significant differences in weight loss between surgical and matched nonsurgical patients comparing those taking NDRI and those taking SSRIs at baseline (–4.9 pounds for SSRI vs SNRI [$95\%$ CI, –23.3 to 13.4; $$P \leq 0.60$$] and 11.9 pounds for NDRI vs SSRI [$95\%$ CI, –8.9 to 32.8; $$P \leq 0.26$$]).
Finally, comparing patients taking SSRI and SNRIs, there were no significant differences between sleeve gastrectomy patients and matched controls in 1-, 2-, or 5-year weight loss outcomes.
## Differences in Weight Change by Antidepressant Monotherapy in Gastric Bypass Cohort
In the gastric bypass cohort, mean weights of surgical and nonsurgical patients were similar at baseline for patients taking SSRIs or NDRIs (Fig. 2). Surgical patients’ change in weight from baseline to 1 year after surgery was 87 to 97 pounds lower than matched controls (depending on antidepressant; see bottom of Table 3). Compared with patients taking SSRIs at baseline, those taking NDRIs had greater differential weight loss between surgical patients and matched controls at 1 year (9.7 pounds; $95\%$ CI, 2.0–17.4; $$P \leq 0.014$$; Table 3).
**FIGURE 2.:** *Estimated changes in weight over time among gastric bypass patients and nonsurgical matches taking 1 of 3 antidepressants before surgery. lbs indicates pounds.*
Gastric bypass patients’ change in weight from baseline to 2 years after surgery was 79–96 pounds lower than matched controls (depending on antidepressant; Table 3), and compared with patients taking SSRIs at baseline, differential weight loss at 2 years between surgical patients and matched controls was 12.0 pounds greater for the NDRI cohort ($95\%$ CI, –0.5 to 24.5; $$P \leq 0.06$$; Table 3), although this difference was not statistically significant. Five years after surgery, gastric bypass patients’ change in weight from baseline was 68–75 pounds lower than matched controls (depending on antidepressant) and, by 5 years, surgical-matched weight improvements were not significantly different across patients taking NDRIs versus SSRIs (4.8 pounds; $95\%$ CI, –16.7 to 26.3; $$P \leq 0.66$$; Table 3).
There were no significant differences in 1-, 2-, or 5-year weight loss outcomes comparing gastric bypass patients and matched control taking SSRIs versus those taking SNRIs at baseline.
## COMMENT
This study tested the association between taking 3 classes of antidepressant medications prior to bariatric surgery and weight loss in the 5 years after surgery. Specifically, the study compared weight loss in those undergoing bariatric surgery and matched nonsurgical controls in patients taking bupropion, SNRIs, or SSRIs at baseline. We found that patients taking bupropion (a NDRI medication) had greater weight loss compared with nonsurgical controls than patients taking SSRIs at 1 and 2 years after surgery. However, differences resolved by 5 years. Those taking SNRIs had similar weight loss as those taking SSRIs. Our findings were similar for patients who had sleeve gastrectomy and gastric bypass surgery. Taken as a whole, this study suggests that the choice of antidepressant medication prior to bariatric surgery may have an important impact on weight loss outcomes. This finding is clinically relevant because as many as one-third to one-half of bariatric patients have a history of major depression at baseline, and preoperative antidepressant treatment is common.3,4 This study improves upon prior studies that had limited follow-up to just 129,12 or 24 months11 after surgery and examined only sleeve gastrectomy or gastric bypass. In addition, we compared the weight change of surgical patients taking each class of antidepressants with matched nonsurgical controls taking the same antidepressant class to have a counterfactual weight trend in contrast to the trends for each antidepressant class; none of the prior studies included nonsurgical controls. This design difference is important because our findings also suggest that bariatric surgery is more effective in promoting long-term weight loss compared than usual care regardless of the use of antidepressant medications.
Antidepressant medications are widely believed to have similar efficacy for the treatment of depression.16 Thus, the decision about which antidepressant medication to use often revolves around the expected side effects of the medications, their costs, and patient preferences.17 Prior research has demonstrated that, for adults, initiation of bupropion was associated with 7.1 pounds greater weight loss at 2 years ($95\%$ CI, –11.3 to –2.8; $P \leq 0.01$) than people who initiated fluoxetine (a SSRI).18 The current study extends these findings to patients taking bupropion at the time of bariatric surgery. Given that bupropion is the only antidepressant associated with long-term weight loss, these studies suggest that bupropion should be the first-line drug of choice for all overweight and obese patients—including those seeking bariatric surgery—unless there are other existing contraindications (eg, history of seizure disorder, anorexia nervosa or bulimia, or patients undergoing abrupt discontinuation of ethanol or sedatives including anticonvulsants, barbiturates, or benzodiazepines).
Several factors likely account for a majority of patients in the study being prescribed SSRIs. These may include concerns about the seizure risk with bupropion, particularly at higher dosages,19 and greater risk of adverse effects with bupropion in the elderly.20 Other possible reasons may include the greater toxicity of bupropion in overdose situations, such as in adolescents,21 and some evidence of greater acute improvement using an SSRI for suicidal ideation when present in those with depression.22 Given these trade-offs, clinicians should engage patients in a shared decision-making conversation about their choice of antidepressant treatment, taking into consideration the above potential for adverse effects from bupropion along with the potential benefits in terms of weight loss.
This study has important limitations. First, our study had a relatively small sample size and needs to be replicated in larger samples with 5-year follow-up. We found that the relative benefit of bupropion versus SSRIs was not statistically significant at 5 years, although the point estimates suggested a significant weight loss benefit of bupropion could be observed in a larger cohort. A related limitation is that we only considered antidepressant medications that were prescribed at baseline and did not have the resources to consider whether the duration or dose of antidepressant treatment after surgery may significantly influence the weight trajectory. It is also common for patients to switch antidepressant medications, and our study did not measure changes in antidepressant medications after surgery. Future research should investigate the time-varying influence of exposure to different antidepressant medications as a potential contributor to weight regain and long-term weight loss maintenance after bariatric surgery. Our sample sizes also were not large enough to permit comparisons of NDRI and SNRI medications.
An additional limitation was that patients were not randomized to bariatric surgery or to the specific antidepressant classes examined in this study, so unobserved confounding may persist, and causality cannot be inferred from this study. Surgical and nonsurgical patients were matched on several covariates that improved balance in those covariates, but several other covariates not included in the match remained imbalanced. Finally, this is a cohort of Veterans who are older and predominantly male, so results may not generalize to non-Veteran cohorts that often comprise 30–40-year-old women.
In conclusion, we found a significant association between the use of the antidepressant bupropion and superior 1-to-2 year weight loss among patients with severe obesity undergoing either gastric bypass or sleeve gastrectomy, which are the 2 most common bariatric procedures. Specifically, we found that bupropion was associated with greater weight loss than SSRIs and that SNRIs and SSRIs had similar weight loss after bariatric surgery. Therefore, we have assumed that bupropion is likely to achieve superior weight loss to SNRIs, although this assumption should be tested in larger studies. Our findings are supported by prior research on antidepressants and longitudinal weight change and suggest that bupropion should be considered the first-line antidepressant of choice for patients undergoing bariatric surgery.
## References
1. Arterburn DE, Telem DA, Kushner RF. **Benefits and risks of bariatric surgery in adults: a review.**. *JAMA* (2020) **324** 879-887. PMID: 32870301
2. Courcoulas AP, King WC, Belle SH. **Seven-year weight trajectories and health outcomes in the Longitudinal Assessment of Bariatric Surgery (LABS) study.**. *JAMA Surg* (2018) **153** 427-434. PMID: 29214306
3. Fisher D, Coleman KJ, Arterburn DE. **Mental illness in bariatric surgery: a cohort study from the PORTAL network.**. *Obesity (Silver Spring)* (2017) **25** 850-856. PMID: 28440047
4. Susmallian S, Nikiforova I, Azoulai S. **Outcomes of bariatric surgery in patients with depression disorders.**. *PLoS One* (2019) **14** e0221576. PMID: 31454382
5. Müller M, Nett PC, Borbély YM. **Mental illness has a negative impact on weight loss in bariatric patients: a 4-year follow-up.**. *J Gastrointest Surg* (2019) **23** 232-238. PMID: 30091038
6. Pedro J, Neves JS, Ferreira MJ. **Impact of depression on weight variation after bariatric surgery: a three-year observational study.**. *Obes Facts* (2020) **13** 213-220. PMID: 32229734
7. Roerig JL, Steffen KJ, Zimmerman C. **A comparison of duloxetine plasma levels in postbariatric surgery patients versus matched nonsurgical control subjects.**. *J Clin Psychopharmacol* (2013) **33** 479-484. PMID: 23771193
8. Arterburn D, Sofer T, Boudreau DM. **Long-term weight change after initiating second-generation antidepressants.**. *J Clin Med* (2016) **5** E48
9. Love RJ, Love AS, Bower S. **Impact of antidepressant use on gastric bypass surgery patients’ weight loss and health-related quality-of-life outcomes.**. *Psychosomatics* (2008) **49** 478-486. PMID: 19122124
10. Malone M, Alger-Mayer SA, Polimeni JM. **Antidepressant drug therapy does not affect weight loss one year after gastric bypass surgery.**. *Obes Surg* (2011) **21** 1721-1723. PMID: 21234700
11. Plaeke P, Van Den Eede F, Gys B. **Postoperative continuation of antidepressant therapy is associated with reduced short-term weight loss following Roux-en-Y gastric bypass surgery.**. *Langenbecks Arch Surg* (2019) **404** 621-631. PMID: 30969361
12. Hawkins M, Leung SE, Lee A. **Psychiatric medication use and weight outcomes one year after bariatric surgery.**. *Psychosomatics* (2020) **61** 56-63. PMID: 31806241
13. Kennedy EH, Taylor JM, Schaubel DE. **The effect of salvage therapy on survival in a longitudinal study with treatment by indication.**. *Stat Med* (2010) **29** 2569-2580. PMID: 20809480
14. Li YP, Propert KJ, Rosenbaum PR. **Balanced risk set matching.**. *J Am Stat Assoc* (2001) **96** 870-882
15. Lu B. **Propensity score matching with time-dependent covariates.**. *Biometrics* (2005) **61** 721-728. PMID: 16135023
16. Qaseem A, Snow V, Denberg TD. **Using second-generation antidepressants to treat depressive disorders: a clinical practice guideline from the American College of Physicians.**. *Ann Intern Med* (2008) **149** 725-733. PMID: 19017591
17. Gartlehner G, Gaynes BN, Hansen RA. **Comparative benefits and harms of second-generation antidepressants: background paper for the American College of Physicians.**. *Ann Intern Med* (2008) **149** 734-750. PMID: 19017592
18. Blumenthal SR, Castro VM, Clements CC. **An electronic health records study of long-term weight gain following antidepressant use.**. *JAMA Psychiatry* (2014) **71** 889-896. PMID: 24898363
19. Carvalho AF, Sharma MS, Brunoni AR. **The safety, tolerability and risks associated with the use of newer generation antidepressant drugs: a critical review of the literature.**. *Psychother Psychosom* (2016) **85** 270-288. PMID: 27508501
20. Sobieraj DM, Martinez BK, Hernandez AV. **Adverse effects of pharmacologic treatments of major depression in older adults.**. *J Am Geriatr Soc* (2019) **67** 1571-1581. PMID: 31140587
21. Overberg A, Morton S, Wagner E. **Toxicity of bupropion overdose compared with selective serotonin reuptake inhibitors.**. *Pediatrics* (2019) **144** e20183295. PMID: 31278211
22. Grunebaum MF, Ellis SP, Duan N. **Pilot randomized clinical trial of an SSRI vs bupropion: effects on suicidal behavior, ideation, and mood in major depression.**. *Neuropsychopharmacology* (2012) **37** 697-706. PMID: 21993207
|
---
title: 'Remission and Relapse of Hypertension After Bariatric Surgery: A Retrospective
Study on Long-Term Outcomes'
authors:
- David P. Fisher
- Liyan Liu
- David Arterburn
- Karen J. Coleman
- Anita Courcoulas
- Sebastien Haneuse
- Eric Johnson
- Robert A. Li
- Mary Kay Theis
- Brianna Taylor
- Heidi Fischer
- Julie Cooper
- Lisa J. Herrinton
journal: Annals of Surgery Open
year: 2022
pmcid: PMC10013161
doi: 10.1097/AS9.0000000000000158
license: CC BY 4.0
---
# Remission and Relapse of Hypertension After Bariatric Surgery: A Retrospective Study on Long-Term Outcomes
## Abstract
### Objectives:
To compare hypertension remission and relapse after bariatric surgery compared with usual care.
### Background:
The effect of Roux-en-Y gastric bypass and sleeve gastrectomy on hypertension remission and relapse has not been studied in large, multicenter studies over long periods and using clinical blood pressure (BP) measurements.
### Methods:
This retrospective cohort study was set in Kaiser Permanente Washington, Northern California, and Southern California. Participants included 9432 patients with hypertension 21–65 years old who underwent bariatric surgery during 2005–2015 and 66,651 nonsurgical controls matched on an index date on study site, age, sex, race/ethnicity, body mass index, comorbidity burden, diabetes status, diastolic and systolic BP, and number of antihypertensive medications.
### Results:
At 5 years, the unadjusted cumulative incidence of hypertension remission was $60\%$ ($95\%$ confidence interval [CI], 58–$61\%$) among surgery patients and $14\%$ ($95\%$ CI, 13–$14\%$) among controls. At 1 year, the adjusted hazard ratio for the association of bariatric surgery with hypertension remission was 10.24 ($95\%$ CI, 9.61–10.90). At 5 years, the adjusted hazard ratio was 2.10 ($95\%$ CI, 1.57–2.80). Among those who remitted, the unadjusted cumulative incidence of relapse at 5 years after remission was $54\%$ ($95\%$ CI, 51–$56\%$) among surgery patients and $78\%$ ($95\%$ CI 76–$79\%$) among controls, although the adjusted hazard ratio was not significant (hazard ratio, 0.71; $95\%$ CI, 0.46–1.08).
### Conclusions:
Bariatric surgery was associated with greater hypertension remission than usual care suggesting that bariatric surgery should be discussed with patients with severe obesity and hypertension. Surgical patients who experience remission should be monitored carefully for hypertension relapse.
## INTRODUCTION
Obesity is an antecedent of primary hypertension in 65–$75\%$ of cases.1 Since 1980, the global incidence of obesity and related comorbidities have more than doubled.2 Medical management of hypertension includes weight loss, lifestyle modification, and medications. Improving the success of weight loss efforts could greatly reduce the risk of hypertension with a 2003 meta-analysis of 25 randomized controlled trials showed a reduction of 1 mm Hg blood pressure (BP) for each kilogram of weight loss.3 However, lifestyle modifications frequently fail to generate significant sustained weight loss for most people.4 Roux-en-Y gastric bypass (RYGB) and sleeve gastrectomy (SG) both result in significant and sustained weight loss.5,6 However, the effect of these surgeries on hypertension is less clear.7,8 Although remission of hypertension has been observed in the short-term after bariatric surgery, few long-term studies have examined hypertension after bariatric surgery.9 Furthermore, some studies have relied on patient self-report of their BP rather than clinical measurements and used unclear definitions of hypertension remission and relapse.8 In addition, many studies have been set in small, single-institution surgery cohorts.9 To address these gaps in the literature, a multisite, retrospective, community-based cohort study was conducted to investigate the association between bariatric surgery and long-term remission and relapse of hypertension using BP measurements recorded in the electronic medical record.
## Setting
The study was set in 3 Kaiser Permanente regions: Washington (formerly Group Health), Northern California, and Southern California.5,10–15 Study procedures were approved by each site’s Institutional Review Board. In these regions, Kaiser Permanente provides prepaid, comprehensive, and integrated care to >8 million members. During the study period, >$98\%$ of bariatric operations were performed laparoscopically. The health systems did not provide coverage for medically supervised weight loss such as meal replacement or pharmacotherapy, resources such as group classes were provided. Patients with hypertension receive medications and monitoring.
## Study Population
The cohort study included patients 21–65 years old during January 1, 2005, through September 30, 2015, with body mass index (BMI) ≥35 kg/m2 who had hypertension. Each study subject was assigned an index date. For bariatric surgery patients, the index date was defined as the date of surgery. The same date was defined as the index date for individually matched nonsurgical controls. Hypertension at the index date was defined as (a) ≥2 consecutive measurements of BP ≥$\frac{140}{90}$ at least 1 week apart or (b) use of antihypertensive medication on the index date plus ≥1 outpatient hypertension diagnosis in the year before the index date. Values for BP were determined using the 2 most recent measurements before the index date. However, BP measurements during days 0–7 before the index date were not used because anticipation of surgery could have affected the patient’s BP.
Enrollment for a minimum of 12 months before the index date was required to determine exclusion criteria. Exclusion criteria included pregnancy, eclampsia or preeclampsia (International Classification of Diseases, 9th edition [ICD-9] diagnostic code 642.XX), cancer (140–209.XX), HIV (042, V08), major organ transplantation (V42.0,.1,.6,.7,.8; procedure codes 33.5, 33.6, 37.5, 41.0, 50.5, 52.8, 55.6), end-stage renal disease (V45.1; procedure codes 38.95, 39.95, 54.98), heart bypass surgery (V56.XX), respiratory failure (518.XX), and liver abscess or sequelae of chronic liver disease (572.XX). Patients with resistant hypertension, defined as 1 measurement of BP ≥$\frac{140}{90}$ mm Hg closest to the index date with ≥3 classes of antihypertensive medications or any BP level with use of ≥4 classes of medications were excluded as well, because such patients are unlikely to respond to weight loss.
After applying these exclusion criteria, patients having an RYGB or SG operation were selected for study. Other bariatric operations were not included because the health systems performed few operations such as laparoscopic adjustable banding. Bariatric operations were identified using a combination of patient registries, chart review, and procedure codes (RYGB, ICD-9 44.31, 44.38, 44.39 and Common Procedural Terminology, 4th edition 43633, 43644, 43645, 43844, 43846, 43847, S2085; SG, ICD-9 43.82 and 43.89 and Common Procedural Terminology, 4th edition 43775).
Up to 10 matched nonsurgical controls were matched to each surgical case using the following procedure. First, procedure codes were used to verify that the control patient had not had a bariatric procedure at any time during their health plan membership. Second, for each surgical patient, a pool of potential controls who were enrolled on the patient’s date of bariatric surgery were identified and assigned their index dates. Controls who met the exclusion criteria were removed from the pool. Third, matching of the potential control to the surgery patient was evaluated on the basis of study site, age, BMI (categories: 35.0–39.9, 40.0–49.9, ≥50.0 kg/m2), sex, race/ethnicity, Elixhauser score (≤1 unit difference between surgical and nonsurgical patients), uncontrolled BP status (BP ≥$\frac{140}{90}$; yes or no), number of distinct hypertension medication classes (0–6), and diabetes status (yes or no). Finally, the Mahalanobis distance between each surgical patient and their potential control was calculated using age, BMI, Elixhauser score, diastolic and systolic BPs, and the number of hypertension medication classes. A maximum of 10 matches were selected for each surgical patient using those with the shortest Mahalanobis distance, requiring that no control be used for >1 bariatric surgery patient.
## Data Sources and Variable Construction
As with these investigators’ previous studies, 5,10–15 electronic medical records, insurance claims, and other data systems were used to extract enrollment, insurance coverage, demographics, BMI, BP, medications, deaths, outpatient, inpatient, and emergency department encounters, and diagnosis and procedure codes during the study period and the year before the index date. These data were used to construct study variables (Supplemental Table 1, http://links.lww.com/AOSO/A113).
BP measurements were obtained from outpatient encounters and did not include visits to the emergency room, urgent care, or surgical outpatient settings. When multiple BP measurements were recorded on the same date, the lowest reading was used. Single-class oral BP medications were grouped in 5 categories: diuretics, calcium channel blockers, angiotensin-converting enzyme inhibitors and angiotensin II receptor blockers, alpha blockers, and beta blockers, with other combination medications counted as a sixth category.
Matching variables (age, BMI, sex, race/ethnicity, Elixhauser comorbidities, BP measurements, number of medication classes, and diabetes status) were obtained from the electronic medical record. In addition, self-reported smoking history (ever, never, unknown) at baseline was obtained from vital signs.
The outcome variable “remission of hypertension” was defined as occurring ≥125 days after the end date of the last antihypertensive medication fill followed by 2 normal BP measurements at least 7 days apart, without an elevated BP between them. The date of the second normal BP measurement was used as the date of remission. One hundred twenty-five days was used to confirm that the event was remission and not a gap in medication use, since most patients receive a 100-day supply of medication. The outcome variable “relapse of hypertension” was defined for those who went into remission as of the date that the patient either restarted antihypertensive medication or the date of the second consecutive measurement of BP ≥$\frac{140}{90}$ at least 7 days apart, without a normal BP measurement between them.
Censoring variables included disenrollment and date of death as recorded in state vital statistics files.
## Statistical Analysis
Missing values on smoking and race/ethnicity were imputed, after matching, using the predictive mean matching method.16 Imputation used surgery-control status (RYGB, SB, control), study site, year, age, sex, baseline BMI, systolic and diastolic BP, number of antihypertensive medication classes, Elixhauser score, and number of baseline patient, outpatient, and emergency department visits.
For modeling of remission of hypertension at 5 years, Cox proportional hazards analysis was used to estimate the hazard ratio (HR) and $95\%$ confidence interval (CI) after adjusting for all matching variables as well as smoking history, Elixhauser score, and comorbid cardiovascular, pulmonary, renal, and mental health disease. Days since the index date was used the time axis, and follow-up started on the index date and ended on the date of outcome, or upon censoring due to the diagnosis of cancer, death, disenrollment from the health plan, the end of the study on September 30, 2015, or the end of the follow-up period, which was 5 years for the primary analysis. The proportional hazards assumption was assessed by visually inspecting a plot of the log(-log(survival)) versus the log of survival and was found to be nonproportional, therefore, step functions were used.17 Among those who experienced remission, relapse of hypertension at 5 years was modeled using Cox proportional hazards analysis, adjusting as described in the preceding paragraph. Days since the remission was used the time axis, and follow-up started on the date of remission and ended on the date of outcome, or upon censoring as described in the preceding paragraph.
The cumulative incidence (%) of remission and relapse at 1, 5, and 7 years was calculated using Kaplan-Meier methods. In addition, the average number of antihypertensive medication classes dispensed per person to the surgical and nonsurgical cohort during each month of the follow-up period was calculated by adding the total number of medication classes (up to 6) dispensed to each patient and dividing by the total number of patients, restricting to patients who were under observation in that month. P values were computed using Student t test. All analyses were conducted using SAS version 9.4 (Cary, NC) and R 3.4.4.
## RESULTS
The number of patients 21–65 years old during the accrual period with BMI ≥35 kg/m2 who underwent RYGB or SG and had ≥1 year of enrollment before their surgery date was 34,824. Of these, 12,075 ($35\%$) had hypertension at baseline. Of those with hypertension, 493 ($4\%$) were pregnant or had another disqualifying condition and 1932 ($17\%$) had resistant hypertension, leaving 9650 eligible surgery patients. Of these, 9432 ($98\%$) were successfully matched to at least 1 nonsurgical control, with the total number of controls being 66,651 (ratio, 1:7.1).
Although bariatric surgery patients and nonsurgical controls were closely matched at the time of study enrollment, surgery patients more often had a BMI ≥50 ($17\%$ vs $12\%$) (Supplemental Table 2, http://links.lww.com/AOSO/A113). About one-quarter of surgery patients and nonsurgical controls had systolic BP ≤139 and diastolic BP ≤89 with the use of 1 class of antihypertensive medication, and about $50\%$ had systolic BP ≤139 and diastolic BP ≤89 with the use of 2 classes of antihypertensives. The most frequently used medications were diuretics and angiotensin-converting enzyme inhibitors/angiotensin II receptor blockers.
For the analysis of remission, among patients whose follow-up began 1 year or more before the end of study, the proportion with at least 1 year of follow-up was $92.8\%$ among the 9432 surgery patients and $89.0\%$ among the 66,651 controls. At 5 years, these percentages were $71.7\%$ and $66.5\%$ (Supplemental Table 3, http://links.lww.com/AOSO/A113). For the analysis of relapse, retention differed slightly because entry into follow-up began on the remission date and not the index date.
Remission of hypertension was noted in 4377 surgery patients and 5673 controls over the course of follow-up. At 1 year, the unadjusted cumulative incidence of remission was $28\%$ ($95\%$ CI, 27–$29\%$) among surgery patients and $3\%$ ($95\%$ CI, 3–$3\%$) among controls (Table 1). At 3 years, it was $55\%$ ($95\%$ CI, 53–$56\%$) among patients and $9\%$ ($95\%$ CI, 9–$10\%$) among controls. At 5 years, it was $60\%$ ($95\%$ CI, 58–$61\%$) among surgery patients and $14\%$ ($95\%$ CI, 13–$14\%$) among controls. The adjusted HR for the association of bariatric surgery with hypertension remission in the first year following the index date was 10.24 ($95\%$ CI, 9.61–10.90) (Fig. 1 and Supplemental Table 4, http://links.lww.com/AOSO/A113). The adjusted HR declined each year, but even in year 5, surgery was associated with a 2.10-fold greater incidence of remission ($95\%$ CI, 1.57–2.80).
Relapse was noted in 1639 of the 4377 surgery patients who remitted and 3296 of the 5673 controls who remitted. At 1 year, the unadjusted cumulative incidence of relapse was $21\%$ ($95\%$ CI, 27–$29\%$) among surgery patients and $44\%$ ($95\%$ CI, 43–$46\%$) among controls (Table 1). At 3 years, it was $41\%$ ($95\%$ CI, 40–$43\%$) among surgery patients and $68\%$ (66–$69\%$) among controls by 3 years. Among those who remitted at any time during follow-up, bariatric surgery was associated with $66\%$ lower risk of relapse during the first year following remission (adjusted HR, 0.34; $95\%$ CI, 0.32–0.37) (Fig. 2 and Supplemental Table 4, http://links.lww.com/AOSO/A113). This association was statistically significant through 4 years after remission but by year 5, it was no longer statistically significant (HR, 0.71; $95\%$ CI, 0.46–1.08).
**FIGURE 2.:** *Hypertension relapse: Adjusted hazard ratios and 95% CIs for the association of bariatric surgery with relapse among those who remitted, 4377 bariatric surgery patients and 5673 matched nonsurgical controls. (1) Hypertension was defined as ≥2 measurements of BP ≥140/90 at least 1 week apart with no nonelevated BP measures between them or use of antihypertensive medication on the date of surgery plus ≥1 outpatient hypertension diagnosis in the year before surgery. (2) For each patient who underwent bariatric surgery, up to 10 nonsurgical controls matched on the surgery/index date on site, age, sex/gender, race/ethnicity, BMI, Elixhauser score, diabetes status, uncontrolled BP status, diastolic and systolic BPs, and the number of hypertension medication classes were identified. (3) The model was adjusted for the matching variables (study site, age, BMI, sex/gender, race/ethnicity, Elixhauser score [≤1 score], uncontrolled BP status, diastolic and systolic BPs, number of distinct hypertension medication classes, and diabetes status) plus index year (continuous), smoking history, and specific physical and mental health comorbidities.*
The average number of medication classes dispensed in the controls, per person per month, appeared to decline from 1.5 to 1.4 during the 6 months after the index date but then returned slowly to 1.5 by 5 years (Fig. 3). This contrasts ($P \leq 0.0001$) with the sharper decrease in the surgical group from 1.5 to 0.5 in the first year, followed by an increase to 0.7 medications by 5 years.
**FIGURE 3.:** *Average number of antihypertensive medication classes dispensed per patient per month. (1) Hypertension was defined as ≥2 measurements of BP ≥140/90 at least 1 week apart with no nonelevated BP measures between them or use of antihypertensive medication on the date of surgery plus ≥1 outpatient hypertension diagnosis in the year before surgery. (2) For each patient who underwent bariatric surgery, up to 10 nonsurgical controls matched on the surgery/index date on site, age, sex/gender, race/ethnicity, BMI, Elixhauser score, diabetes status, uncontrolled BP status, diastolic and systolic BPs, and the number of hypertension medication classes were identified. (3) Hypertensive medication classes included beta blocker, angiotensin II receptor antagonists, ACE inhibitors, calcium channel blocker, diuretics, and other antihypertensive medications. ACE indicates angiotensin-converting enzyme.*
## DISCUSSION
In this large, retrospective cohort study of patients with severe obesity and hypertension, the 5-year unadjusted cumulative incidence of hypertension remission was $60\%$ ($95\%$ CI, 58–$61\%$) in 9432 bariatric surgery patients and $14\%$ ($95\%$ CI, 13–$14\%$) in 66,651 matched nonsurgical controls. The adjusted HR for the association of bariatric surgery with hypertension remission was 10.24 ($95\%$ CI, 9.61–10.90) in the first year and 8.85 ($95\%$ CI, 8.24–9.50) in the second year but diminished by 7 years, when the difference was no longer statistically significant. At 1 year after remission, the risk of relapse differed between surgical and nonsurgical groups, such that bariatric surgery was associated with $66\%$ lower risk of relapse (adjusted HR, 0.34; $95\%$ CI, 0.32–0.37). This difference narrowed after the first year but remained significant until 5 years after remission (surgical patients, $54\%$; controls, $78\%$; adjusted HR, 0.71; $95\%$ CI, 0.41–1.08). Bariatric surgery was also associated with a striking reduction in the use of antihypertensive medications from an average of 1.5 medications per month before surgery to 0.7 per month at 5 years, while the number of medications in controls did not change appreciably. The study was novel because of its large size and 7-year duration of follow-up. The analysis of long-term changes in antihypertensive medications was novel as well, indicating a long-term benefit of surgery for reducing the need for antihypertensive medications.
These findings are also consistent with prior research. A recent systematic review found that bariatric surgery was associated with hypertension remission at 1 year ranging from $43\%$ to $83\%$.7 Remission in this study was lower, at $28\%$, possibly because of the strict definition of remission, which required 125 days since the end of the last antihypertensive prescription followed by 2 consecutive normal BP readings separated by at least 1 week. This definition was used to reflect current prescribing practices in the participating health systems, which frequently issue a 100-day supply of hypertension medications to improve adherence and control. In the Swedish Obesity Study, 2-year incidence of hypertension was $62\%$ lower in bariatric patients than controls (odds ratio, 0.38; $95\%$ CI, 0.22–0.65)18 and differences in BP between surgical and control patients were larger at 2 years than at 10 years of follow-up.19 Hypertension control rates in the participating health systems are some of the highest reported among all health plans in the United States.20 This could have attenuated the differences in hypertension control between the bariatric surgery patients compared to controls given that there are aggressive efforts across all patient populations to achieve hypertension control and remission. These findings are similar to a recent clinical trial focused specifically on the 3-year hypertension outcomes of 100 patients randomized to either RYGB plus medical treatment for hypertension compared to medical treatment alone.9 At 1 year, half of the RYGB patients experienced remission compared with none of the medically treated patients. At 3 years, $35\%$ of RYGB patients had continuous (ie, durable) remission compared with none of the medically treated patients. These findings are largely consistent with the results of this study at 3 years, with $55\%$ of surgical patients experiencing a remission by 3 years and $41\%$ of those subsequently relapsing within 3 years after remission, resulting in a $32\%$ durable remission at 3 years.
The primary purpose of this report was to compare bariatric surgery patients to patients who did not get surgery because this is relevant to patients early in their decision-making process and addresses the question, is bariatric surgery more effective than nonsurgical treatment for improving my hypertension? Once patients have decided to consider bariatric surgery, the question then shifts towards selection the optimal procedure to improve health. To address this second questions, members of our team recently published a report comparing the effectiveness of RYGB and SG on hypertension using a subset of the data included in this analysis, of whom 1778 received RYGB and 3186 received SG, with no nonsurgical control included in the study (Reynolds). The earlier study observed no difference in the incidence of remission between the 2 bariatric procedures, although RYGB was associated with lower systolic and diastolic BP measurements over time. In both reports, nearly all remissions occurred in the first year, with some patients subsequently relapsing. In the earlier report, at 5 years, $43\%$ of patients had remitted, although $25\%$ relapsed, leaving $18\%$ in remission. In the present analysis, at 5 years, $60\%$ of surgery patients and $14\%$ of controls remitted (adjusted HR, 2.10; $95\%$ CI, 1.57–2.80), and among those who remitted at any time during follow-up, bariatric surgery was associated with $66\%$ lower risk of relapse during the first year following remission (adjusted HR, 0.34; $95\%$ CI, 0.32–0.37). In addition, the present report presents longitudinal information on use of antihypertensive drugs that was not reported in our earlier study.
In this same cohort, we previously reported on weight loss finding that RYGB patients had $28.4\%$ total weight loss (TWL), SG $23.0\%$TWL, and nonsurgical patients $0.2\%$TWL (0.1, 0.4) at 1 year and at 5 years, RYGB had $21.7\%$TWL (21.5, 22.0), SG $16.0\%$TWL (15.4, 16.6), and nonsurgical patients $2.2\%$TWL (2.0, 2.5). It is likely that weight changes were the main driver in differences in the rates of hypertension (HTN) remission and relapse across surgical and nonsurgical patients. However, attribution of HTN control to weight loss is an important question, that is, beyond the scope of the current study. Understanding the mediating effect of weight loss on HTN should be the subject of future research.
Although gains in remission and relapse in bariatric surgery patients compared to controls diminished by 5 years, reduction in use of antihypertensive medications continued. This is important for reducing medication side effects. In addition, for patients who experience diabetes relapse, the benefit of bariatric surgery on microvascular complications is retained long-term, provided they remain in remission for at least 1 year.21,22 This could also be true for complications resulting from prolonged hypertension.
An initial drop in the number of medication classes used per month during the first 6 months of follow-up was observed in the control group. This likely was an artifact of the definition used for hypertension, which included patients who used an antihypertensive on their index date but who may have skipped days or otherwise been nonadherent to their medication during follow-up.
The control group received usual medical care that generally did not include intensive medical or lifestyle interventions for weight loss. Past studies have compared bariatric surgery to intensive medical and lifestyle interventions and reported similar results, that is, rates of hypertension remission are higher and antihypertensive medication use is lower in surgical compared with nonsurgical patients.9,23–25 Strengths of the study include the sample size, long follow-up, multiple study sites, and comprehensive information. The study population was diverse with $17\%$ black and $27\%$ Hispanic patients. Baseline hypertension, remission, and relapse were rigorously defined. The main limitation was lack of randomization. Only 1–$2\%$ of patients with severe obesity and hypertension underwent bariatric surgery, and surgical patients likely differed from controls in ways that cannot be assessed because the information was not available. Thus, the study could be biased if, at baseline, the bariatric surgery patients differed from nonsurgical controls with respect to risk factors that are related both to the decision to undergo surgery and to later remission and relapse of hypertension. It is difficult to identify a such a risk factor, particularly one that would lead to the large effects that were observed. Nonetheless, the study results should be confirmed in a randomized controlled trial.
The study also had limitations. Hypertension remission and relapse were not evaluated in relation to weight loss. Also, the patients in this study were from integrated healthcare systems that prepare patients for surgery through careful chronic disease management, possibly attenuating the effects of surgery on hypertension. Our system uses population management to assure frequent BP measurement and medication compliance, with rates of hypertension control in the general membership of $84.1\%$ among commercial patients, $87.2\%$ among Medicare patients, and $84.6\%$ among Medicaid patients. Additionally, in prior research, we reported that this population is $90\%$ commercially insured, with $6\%$ Medicare and $4\%$ Medicaid; thus, we had insufficient data to investigate differences in outcomes across different insurance types, and our findings may not generalize to other health care settings. Furthermore, the study excluded patients with high BP on 3 medications, and all patients on 4 medications; however, it is possible that these patients could benefit from bariatric surgery and future studies should seek to investigate this question.26 Finally, our study lacked resources to assess the mediating effect of weight loss and regain on hypertension remission and relapse, which should be the subject of future research.27 Two thirds of adults with severe obesity who seek bariatric surgery have hypertension, which is associated with an increased risk of cardiovascular disease and early mortality.28 *In this* observational comparative-effectiveness study with long-term follow-up and careful matching of surgical and nonsurgical patients, bariatric surgery was strongly associated with hypertension remission over 5 years. These findings suggest that patients with severe obesity and hypertension may benefit from bariatric surgery. Healthcare systems should discuss bariatric surgery more often as a treatment for hypertension. Notwithstanding, surgical and nonsurgical patients who experience remission should be closely monitored because of the high rate of relapse.
## References
1. Kung HC, Xu J. **Hypertension-related mortality in the United States, 2000–2013.**. *NCHS Data Brief* (2015) **193** 1-8
2. Gregg EW, Shaw JE. **Global health effects of overweight and obesity.**. *N Engl J Med* (2017) **377** 80-81. PMID: 28604226
3. Neter JE, Stam BE, Kok FJ. **Influence of weight reduction on blood pressure: a meta-analysis of randomized controlled trials.**. *Hypertension* (2003) **42** 878-884. PMID: 12975389
4. Dansinger ML, Tatsioni A, Wong JB. **Meta-analysis: the effect of dietary counseling for weight loss.**. *Ann Intern Med* (2007) **147** 41-50. PMID: 17606960
5. Arterburn DE, Johnson E, Coleman KJ. **Weight outcomes of sleeve gastrectomy and gastric bypass compared to nonsurgical treatment.**. *Ann Surg* (2021) **274** e1269-e127. PMID: 32187033
6. Maciejewski ML, Arterburn DE, Van Scoyoc L. **Bariatric surgery and long-term durability of weight loss.**. *JAMA Surg* (2016) **151** 1046-1055. PMID: 27579793
7. Climent E, Goday A, Pedro-Botet J. **Laparoscopic Roux-en-Y gastric bypass versus laparoscopic sleeve gastrectomy for 5-year hypertension remission in obese patients: a systematic review and meta-analysis.**. *J Hypertens* (2020) **38** 185-195. PMID: 31633582
8. Buchwald H, Avidor Y, Braunwald E. **Bariatric surgery: a systematic review and meta-analysis.**. *JAMA* (2004) **292** 1724-1737. PMID: 15479938
9. Schiavon CA, Bhatt DL, Ikeoka D. **Three-year outcomes of bariatric surgery in patients with obesity and hypertension: a randomized clinical trial.**. *Ann Intern Med* (2020) **173** 685-693. PMID: 32805133
10. Coleman KJ, Huang YC, Hendee F. **Three-year weight outcomes from a bariatric surgery registry in a large integrated healthcare system.**. *Surg Obes Relat Dis* (2014) **10** 396-403. PMID: 24951065
11. Arterburn D, Powers JD, Toh S. **Comparative effectiveness of laparoscopic adjustable gastric banding vs laparoscopic gastric bypass.**. *JAMA Surg* (2014) **149** 1279-1287. PMID: 25353723
12. Fisher DP, Johnson E, Haneuse S. **Association between bariatric surgery and macrovascular disease outcomes in patients with type 2 diabetes and severe obesity.**. *JAMA* (2018) **320** 1570-1582. PMID: 30326126
13. Li R, Fisher D, Dutta S. **Bariatric surgery results: reporting clinical characteristics and adverse outcomes from an integrated health care system.**. *Surg Obes Rel Dis* (2015) **11** 1119-1125
14. Li RA, Liu L, Arterburn D. **Five-year longitudinal cohort study of reinterventions after sleeve gastrectomy and Roux-en-Y gastric bypass.**. *Ann Surg* (2021) **273** 758-765. PMID: 31188199
15. O’Brien R, Johnson E, Haneuse S. **Microvascular outcomes in patients with diabetes after bariatric surgery versus usual care: a matched cohort study.**. *Ann Intern Med* (2018) **169** 300-310. PMID: 30083761
16. **Inputing Missing Data With R; MICE Package.**. (2015)
17. Hosmer DW, Lemeshow S, May S. *Applied Survival Analysis: Regression Modeling of Time to Event Data* (2000)
18. Sjöström CD, Lissner L, Wedel H. **Reduction in incidence of diabetes, hypertension and lipid disturbances after intentional weight loss induced by bariatric surgery: the SOS Intervention Study.**. *Obes Res* (1999) **7** 477-484. PMID: 10509605
19. Sjöström L, Lindroos AK, Peltonen M. **Lifestyle, diabetes, and cardiovascular risk factors 10 years after bariatric surgery.**. *N Engl J Med* (2004) **351** 2683-2693. PMID: 15616203
20. Jaffe MG, Young JD. **The Kaiser Permanente Northern California Story: improving hypertension control from 44% to 90% in 13 years (2000 to 2013).**. *J Clin Hypertens (Greenwich)* (2016) **18** 260-261. PMID: 26939059
21. Arterburn D, Bogart A, Coleman KJ. **Comparative effectiveness of bariatric surgery vs. nonsurgical treatment of type 2 diabetes among severely obese adults.**. *Obes Res Clin Pract* (2013) **7** e258-e268. PMID: 24306153
22. Coleman KJ, Haneuse S, Johnson E. **Long-term microvascular disease outcomes in patients with type 2 diabetes after bariatric surgery: evidence for the legacy effect of surgery.**. *Diabetes Care* (2016) **39** 1400-1407. PMID: 27271192
23. Schauer PR, Bhatt DL, Kirwan JP. **Bariatric surgery versus intensive medical therapy for diabetes - 5-year outcomes.**. *N Engl J Med* (2017) **376** 641-651. PMID: 28199805
24. Mingrone G, Panunzi S, De Gaetano A. **Bariatric-metabolic surgery versus conventional medical treatment in obese patients with type 2 diabetes: 5 year follow-up of an open-label, single-centre, randomised controlled trial.**. *Lancet* (2015) **386** 964-973. PMID: 26369473
25. Pareek M, Bhatt DL, Schiavon CA. **Metabolic surgery for hypertension in patients with obesity.**. *Circ Res* (2019) **124** 1009-1024. PMID: 30920920
26. Schiavon CA, Pio-Abreu A, Drager LF. **Bariatric surgery for resistant hypertension: working in progress!**. *Curr Hypertens Rep* (2020) **22** 55. PMID: 32671558
27. Shaw KM, Handler J, Wall HK. **Improving blood pressure control in a large multiethnic California population through changes in health care delivery, 2004-2012.**. *Prev Chronic Dis* (2014) **11** E191. PMID: 25357259
28. Courcoulas AP, King WC, Belle SH. **Seven-year weight trajectories and health outcomes in the Longitudinal Assessment of Bariatric Surgery (LABS) study.**. *JAMA Surg* (2018) **153** 427-434. PMID: 29214306
|
---
title: Vitamin D status, including serum levels and sun exposure are associated or
correlated with bone mass measurements diagnosis, and bone density of the spine
authors:
- Adeleh khodabakhshi
- Sayed Hossein Davoodi
- Farhad Vahid
journal: BMC Nutrition
year: 2023
pmcid: PMC10013235
doi: 10.1186/s40795-023-00707-y
license: CC BY 4.0
---
# Vitamin D status, including serum levels and sun exposure are associated or correlated with bone mass measurements diagnosis, and bone density of the spine
## Abstract
### Background
Osteoporosis is a health complication worldwide, especially in developing countries. The prevalence was reported to be $18.3\%$ globally. While the effect of biochemical factors on fracture risk/odds has been documented, the association/correlation between serum 25(OH)D levels, vitamin D dietary intake, and sun exposure with bone mineral density (BMD) remains controversial. This study aimed to evaluate the association and correlation between vitamin D status, including serum levels, dietary intakes, and sun exposure with BMD. We hypothesized that vitamin D-related factors would have different correlations/associations with BMD, which would help better evaluate future studies’ results.
### Methods
A total of 186 individuals were included in this study (winter 2020). BMD was measured by Dual-energy X-ray absorptiometry. Blood serum levels of 25(OH)D, phosphorus, calcium, parathyroid hormone (PTH), and calcitonin were tested using standard lab tests. Valid and reliable questionnaires were used for sun exposure assessment and dietary intakes.
### Results
There was a significant protective association between spine BMD (classifications, two groups) (OR = 0.69, $95\%$CI: 0.50–0.94; p-value = 0.023), BMD diagnosis (classifications, two groups) (OR = 0. 69, $95\%$CI: 0.49–0.87; p-value = 0.036) and sun exposure. There was a significant and moderate correlation between Spine measurements (Spine BMD: Pearson correlation coefficient = 0.302, p-value = 0.046; Spine T-score: Pearson correlation coefficient = 0.322, p-value = 0.033, Spine Z-score: Pearson correlation coefficient = 0.328, p-value = 0.030) and serum 25(OH)D. In addition, participants with osteopenia and osteoporosis significantly consume a higher amount of soluble fiber than the normal BMD group. There was no significant correlation between vitamin D intake and BMD.
### Conclusion
In conclusion, serum 25(OH)D levels and sun exposure are correlated and associated with BMD. However, prospective studies are needed to investigate the association between dietary vitamin D intake and BMD.
## Background
As a systemic disease, osteoporosis is characterized by microarchitectural deterioration of bone tissue and low bone mass [1]. It is a crucial public health problem worldwide, especially in developing countries, so the prevalence of osteoporosis globally was reported to be $18.3\%$ [2]. According to estimates in Iran, about $17\%$ of the general population over 30 years have osteoporosis, and about $35\%$ suffer from osteopenia [3]. If identified early in its course, as it is a major leading cause of bone fragility fractures, many of the fractures can be prevented [4]. Dietary and lifestyle-related factors such as calcium and/or vitamin D deficiency, little or no exercise (sedentary lifestyle), especially weight-bearing exercise, alcohol abuse, smoking, genetic factors, and environmental and hormonal factors, among others, affect bone mineral density (BMD) [5, 6].
While the effect of biomarkers on fracture risk/odds has been documented in some previous studies, the association/correlation between serum 25(OH)D levels, dietary intake, and sun exposure with BMD remains controversial [7, 8]. Although a positive association between low serum vitamin D and low BMD was found in several studies [9–11], other studies did not show any significant association between these two parameters [7, 12, 13].
Until recently, in some countries, such as the UK, vitamin D and/or calcium supplementations were the first treatment choice for preventing/controlling fractures in the elderly [14]. However, the Randomised Evaluation of Calcium Or vitamin D (RECORD) trial questioned/criticized the importance of vitamin D, and apparently, this strategy may not be sufficient to avert further fractures in the ‘healthy’ elderly [15]. Some other randomized controlled trials also were not able to show an advantage in fracture reduction with vitamin D supplementation [16, 17]. However, a meta-analysis of randomized controlled trials proposed that 20 µg/day (800 IU/day) of vitamin D is necessary to demonstrate any advantage [18].
Nevertheless, low vitamin D levels is associated/correlated with higher odds/risk of bone loss, bone turnover, and other bone-related disorders [19]. On the other hand, it seems diet attenuates the seasonal variation of vitamin D levels at the northern latitude, where the quality of sunlight for vitamin D production decreases [19]. Therefore, it might be a comprehensive and advantageous solution to consider all the factors involved in vitamin D status, including exposure to sunlight, dietary intake (with or without supplementation), and serum vitamin D levels to assess its effect on bone health or even other vitamin-related diseases.
Considering that, this study aimed to evaluate the association and correlation between vitamin D status, including serum levels, dietary intakes, and sun exposure with BMD.
## Study population
Protocol and design of study previously published elsewhere [8]. Briefly, this study was conducted on 186 Sirjan Gol Gohar Company staff in the winter of 2020. An invitation letter was circulated to all staff, inviting them to participate in the study. Then, individuals who accepted the invitation (responded to the initial letter) and had the inclusion criteria (see below) were included in the survey [8]. Written informed consent was obtained from all participants. The study protocol and design were approved by the Kerman University of Medical Sciences ethics committee board (IR.KMU.REC.1399.156). All methods were performed in accordance with the Declaration of Helsinki. A trained professional filled out a general questionnaire for all participants, including general characteristics and medical history.
## Inclusion and exclusion criteria
Individuals with pregnancy and lactation, diseases interfering with vitamin D absorption/metabolisms such as chronic pancreatitis, inflammatory bowel disease (IBD), resection of part of the intestine or stomach, as well as individuals with hyperparathyroidism, renal failure, advanced liver failure, rheumatoid arthritis, and those who took calcium supplements at least once a day and vitamin D supplements over the past two weeks, and vitamin D ampules over the past six months, individuals smoking more than 10 cigarettes/day and consuming alcohol for more than 5 years and more than a glass/day or individuals with addiction to any drugs were excluded from the study [8].
## Blood samples
In a fasting state, seven milliliters (ml) of blood were taken from the individuals. Blood samples were immediately centrifuged and stored at -80 °C. The ELISA method used a Monobind kit made in the USA to measure serum 25(OH)D. In addition, serum calcium and phosphorus were measured using an Auto Analyser (Hitachi, Germany) photometry method. Serum PTH and calcitonin were measured by the Chemiluminescence method (Siemens kit, Germany).
## Dietary intake
Participants’ dietary intakes were estimated by semi-quantitative and valid Food Frequency Questionnaires (FFQ) [20]. A nutritionist completed the questionnaire. Portion size in FFQ was converted to grams per day using household measures. Subsequently, the Nutritionist IV software was applied to extract macro and micronutrients daily intake, including vitamin D [8].
## Sun exposure
Using a valid and reliable questionnaire, sun exposure was estimated. The questionnaire included questions about the amount of exposure to outdoor sunlight (on weekdays and weekends), applying sunscreen creams, and the parts of the body exposed to sunlight during outdoor sunlight [21, 22].
## BMD
An experienced and trained technician assessed hip, femoral neck, and lumbar spine (L1–4) areal BMD g/cm2 by Dual-energy x-ray absorptiometry (Hologic Horizon WI, USA). According to the World Health Organization (WHO) classification system, osteoporosis was classified as T-score ≤ − 2.5, osteopenia as − 2.5 < T-score < − 1, and normal as T-score ≥ − 1 [23].
## Statistical analyses
Before choosing statistical tests, the normality of continuous variables was checked by the Q-Q plot and Kolmogorov-Smirnov test. If the variables were not normal, they were log-transformed. An Independent sample t-test was used for continuous variables, and chi-square analyses were used for categorical variables. Bivariate correlation (variables categorized), Spearman’s rho, was used to investigate the correlation between classified/categorized variables. Partial correlation controlled for BMI, age, PTH, and calcitonin was applied to investigate the correlation between two continuous variables while taking away the effects of another variable, or several other variables, on these correlations. Logistic regression models adjusted for age, BMI, PTH, and Calcitonin were used to investigate the association between vitamin status, dietary intake, serum levels, and sun exposure with BMD measurements including spine, total hip, and femoral neck and BMD diagnosis. Data were analyzed with SPSS (IBM, Chicago, IL, USA) version 25.0. A p-value of < 0.05 (2-sided) was considered statistically significant. Benjamini–Hochberg correction was applied to all p-values, and all p-values are displayed after this correction.
## Distribution of basic characteristics and their comparison
The distribution of anthropometric, socioeconomic, and serum indicators of participants is shown in Table 1. Based on Table 1, there was no significant difference between the normal BMD group and participants with osteopenia and the osteoporosis group in terms of baseline measurements. A comparison of participants’ macro-and micronutrient daily intake is represented in Table 2. According to Table 2, except for soluble fiber (normal BMD group 0.16 ± 0.09 vs. osteopenia and osteoporosis group 0.26 ± 0.18), there was no significant in terms of dietary intakes in the two groups. In addition, Table 2 shows that participants with osteopenia and osteoporosis consume significantly higher amounts of soluble fiber than the normal BMD group.
Table 1Distribution of anthropometric, socioeconomic, and serum indicators of participantsMean ± SD or N (%)P-value*NormalOsteopenia and OsteoporosisTotalAge (year)34.6 ± 9.236.6 ± 6.335.9 ± 7.70.320BMI (Kg/m2)26.2 ± 3.224.9 ± 3.226.1 ± 3.70.131PTH (pg/mL)43.1 ± 19.141.5 ± 22.344.3 ± 21.40.770Serum calcium (mg/dL)9.6 ± 0.49.74 ± 0.59.76 ± 0.50.418Serum phosphorus (mg/dL)3.1 ± 0.53.1 ± 0.33.2 ± 0.40.841Calcitonin (pg/mL)5.6 ± 2.66.0 ± 3.85.2 ± 2.80.663Serum D3 (ng/mL)27.4 ± 13.427.09 ± 15.426.4 ± 13.50.922Gender-Men25 ($96.2\%$)36 ($97.3\%$)61 ($96.8\%$)0.659Smoking-No18 ($69.2\%$)28 ($75.7\%$)49 ($77.7\%$)0.361Marital status-Married17 ($65.4\%$)30 ($81.1\%$)47 ($74.6\%$)0.377Education-University degree25 ($96.2\%$)36 ($97.3\%$)61 ($96.8\%$)0.419Exposure to sunlight0.057 ■ ˃30 min07 ($18.9\%$)7 ($18.9\%$) ■ 31–60 min03 ($8.1\%$)3 ($8.1\%$) ■ 2 h2 ($7.7\%$)4 ($10.8\%$)6 ($18.5\%$) ■ 3 h3 ($11.5\%$)3 ($8.1\%$)6 ($19.6\%$) ■ 4 h4 ($15.4\%$)5 ($13.5\%$)9 ($28.9\%$) ■ 5 h3 ($11.5\%$)03 ($11.5\%$) ■ 6 h4 ($15.4\%$)7 ($18.9\%$)11 ($34.3\%$)* Independent sample t-test was used for comparing continuous variables. Chi-square analyses were used for comparing categorical variables. BMI = body mass index, PTH = parathyroid hormone. Benjamini–Hochberg correction was applied to all p-values: all p-values are displayed after this correction.
Table 2Comparison of participants’ macro-and micronutrient daily intakeMean ± SDP-value*NormalOsteopenia and OsteoporosisTotalTotal energy (kcal)1540 ± 629.81680 ± 691.41617 ± 569.80.511Total protein (g)64.2 ± 28.167.8 ± 25.266.2 ± 26.20.674Total carbohydrate (g)198.1 ± 95.0236.5 ± 136.4219.2 ± 119.70.319Total fat (g)56.1 ± 26.953.2 ± 26.254.5 ± 26.20.737Cholesterol (mg)379.3 ± 448.7314.3 ± 260.5343.6 ± 345.10.570SFA (g)17.1 ± 6.418.1 ± 9.217.6 ± 8.00.685MUFA (g)19.1 ± 10.218.5 ± 11.118.8 ± 10.60.860PUFA (g)13.4 ± 8.510.9 ± 4.212.1 ± 6.50.234MFA (g)17.6 ± 9.917.0 ± 10.117.3 ± 9.90.840PFA2 (g)11.3 ± 8.29.3 ± 4.110.2 ± 6.30.325PFA3 (g)0.87 ± 0.300.76 ± 0.520.81 ± 0.440.457PFA5 (g)0.16 ± 0.160.09 ± 0.80.12 ± 0.130.089PFA6(g)3.9 ± 5.23.0 ± 2.13.4 ± 3.80.424Sodium (mg)1355 ± 671.91415 ± 898.11388 ± 795.00.816Potassium (mg)2047 ± 840.12416 ± 887.72250 ± 875.60.188Vitamin A (RAE)306.4 ± 184.0348.5 ± 219.7329.6 ± 203.00.522Beta-carotene (µg)690.2 ± 401.7845.5 ± 502.9775.6 ± 461.20.295Alpha-carotene (µg)38.1 ± 26.463.5 ± 53.352.1 ± 44.70.072Lutein (µg)829.2 ± 475.1906.7 ± 520.6871.8 ± 495.90.629Betacryptox (µg)201.1 ± 139.9321.9 ± 265.7267.5 ± 224.20.090Vitamin C (mg)57.6 ± 30.780.8 ± 51.070.4 ± 44.10.099Calcium (mg)674.6 ± 233.3776.2 ± 381.1730.5 ± 323.30.329Iron (mg)11.1 ± 4.812.7 ± 5.711.9 ± 5.30.326Vitamin D (µg)1.8 ± 1.62.1 ± 2.71.9 ± 2.20.707Vitamin E (mg)8.7 ± 8.36.9 ± 3.67.7 ± 6.20.372Alpha-tocopherol (mg)5.5 ± 5.54.4 ± 2.54.9 ± 4.10.403Thiamine (mg)1.49 ± 0.671.71 ± 0.991.61 ± 0.860.441Riboflavin (mg)1.47 ± 0.751.53 ± 0.681.50 ± 0.710.790Niacin (mg)14.0 ± 6.316.0 ± 7.815.1 ± 7.10.389Vitamin B6 (mg)1.11 ± 0.341.28 ± 0.581.21 ± 0.490.277Total folate (µg)413.1 ± 184.5490.7 ± 288.0455.7 ± 247.10.329Folate DFE (µg)510.5 ± 269.9615.5 ± 455.3568.3 ± 382.30.395Vitamin B12 (µg)3.89 ± 1.714.86 ± 2.774.42 ± 2.380.203Biotin (µg)25.1 ± 23.322.9 ± 14.023.9 ± 18.50.719Pantothenic acid (mg)4.6 ± 2.15.1 ± 2.64.8 ± 2.30.540Vitamin K (µg)86.1 ± 57.097.0 ± 69.892.1 ± 63.80.596Phosphorous (mg)1124 ± 412.61209 ± 440.11170 ± 424.60.537Magnesium (mg)268.4 ± 131.3294.8 ± 108.9282.9 ± 118.60.490Zinc (mg)10.8 ± 5.412.6 ± 5.011.8 ± 5.20.301Copper (mg)1.03 ± 0.441.18 ± 0.461.11 ± 0.450.321Manganese (mg)3.5 ± 1.24.2 ± 3.33.9 ± 2.60.372Selenium (µg)102.1 ± 56.1106.7 ± 50.4104.6 ± 52.40.783Total fiber (gr)14.1 ± 5.316.5 ± 6.515.4 ± 6.00.219Soluble fiber (gr)0.16 ± 0.090.26 ± 0.180.22 ± 0.15 0.043 Insoluble fiber (gr)1.12 ± 0.511.44 ± 0.731.30 ± 0.660.120Crude fiber (gr)12.2 ± 5.413.9 ± 13.313.2 ± 10.40.615Total sugar (gr)49.7 ± 22.058.8 ± 23.154.5 ± 22.80.228Caffeine (mg)60.8 ± 51.198.2 ± 105.581.4 ± 86.50.177*Independent sample t-test was used for comparing continuous variables.*Benjamini–Hochberg correction was applied to all p-values: all p-values are displayed after this correction; significant values are given in bold. SFA = saturated fatty acid, MUFA = monounsaturated fatty acid, PUFA = polyunsaturated fatty acid, PFA = Polyunsaturated fatty acid,
## Correlations
Partial and bivariate correlations between serum 25(OH)D and BMD are shown in Table 3. According to Table 3, in the partial correlation model controlled for BMI, age, PTH, and calcitonin, there is a significant and moderate correlation between Spine measurements (Spine BMD: Pearson correlation coefficient = 0.302, p-value = 0.046; Spine T-score: Pearson correlation coefficient = 0.322, p-value = 0.033, Spine Z-score: Pearson correlation coefficient = 0.328, p-value = 0.030) and serum 25(OH)D. The partial and bivariate correlation between vitamin D intake and BMD are shown in Table 4. According to Table 4, there was no significant correlation between vitamin D intake and BMD. Partial and bivariate correlations between sun exposure and BMD are shown in Table 5. Table 5 shows that only in bivariate models (BMD are classifications, two groups) without controlling for any confounder factor, there is a significant, moderate, and negative correlation between Spine BMD (correlation coefficient=-0.355, p-value = 0.017), BMD diagnosis (correlation coefficient=-0.326, p-value = 0.029) and sun exposure (Table 5).
Table 3Partial and bivariate correlation between serum 25-hydroxyvitamin D3 and bone mass measurements (BMD).VariablesModel AModel BCorrelationP-Value*CorrelationP-Value*Spine BMD 0.302 0.046 -0.1190.353Femoral neck BMD0.0290.8500.0960.456Total hip BMD0.0330.8300.1480.246BMD diagnosis-0.0130.918Spine T-Score 0.322 0.033 Femoral neck T-score0.0640.681Total hip T-score0.0990.521Spine Z-score 0.328 0.030 Femoral neck Z-score0.0670.664Total hip Z-score0.0710.645Mode A: Partial correlation controlled for BMI, Age, PTH, and Calcitonin. Model B: Bivariate correlation (variables categorized).BMI = body mass index, PTH = parathyroid hormone.*Benjamini–Hochberg correction was applied to all p-values: all p-values are displayed after this correction; significant values are given in bold.
Table 4Partial and bivariate correlation between vitamin D intake and bone mass measurements (BMD).VariablesModel AModel BCorrelationP-Value*CorrelationP-Value*Spine BMD-0.0560.7850.2200.172Femoral neck BMD-0.0170.9350.0540.739Total hip BMD-0.1030.6170.2300.153BMD diagnosis0.1920.235Spine T-score-0.0630.670Femoral neck T-score-0.0090.967Total hip T-score-0.1110.591Spine Z-score-0.0490.811Femoral neck Z-score-0.0120.953Total hip Z-score-0.1090.506Mode A: Partial correlation controlled for BMI, Age, PTH, and CalcitoninModel B: Bivariate correlation (variables categorized).BMI = body mass index, PTH = parathyroid hormone.*Benjamini–Hochberg correction was applied to all p-values: all p-values are displayed after this correction.
Table 5Partial and bivariate correlation between sun exposure and bone mass measurements (BMD).VariablesModel AModel BCorrelationP-Value*CorrelationP-Value*Spine BMD0.1710.366 -0.355 0.017 Femoral neck BMD0.0340.8590.0530.730Total hip BMD-0.0020.9920.0240.875BMD diagnosis -0.326 0.029 Spine T-score0.1740.377Femoral neck T-score0.0310.870Total hip T-score-0.0250.895Spine Z-score0.1760.351Femoral neck Z-score0.0320.876Total hip Z-score0.0020.990Mode A: Partial correlation controlled for BMI, Age, PTH, and CalcitoninModel B: Bivariate correlation (variables categorized).BMI = body mass index, PTH = parathyroid hormone.*Benjamini–Hochberg correction was applied to all p-values: all p-values are displayed after this correction; significant values are given in bold.
In addition, Fig. 1 represents the correlation matrix between vitamin D status, including serum vitamin D, dietary intake, and sunlight exposure.
Fig. 1Correlation matrix between vitamin D status, including serum vitamin D, dietary intake, and sunlight exposure
## Regression models
Association (OR and $95\%$ CI) between serums 25(OH)D, vitamin D intake, sun exposure, and BMD are shown in Table 6. According to Table 6, in regression logistic multivariable models adjusted for BMI, age, PTH, and calcitonin, there was a significant protective association between spine BMD (classifications, two groups) and serums 25(OH)D (OR = 0.92, $95\%$CI: 0.86–0.99; p-value = 0.025) and between BMD diagnosis (classifications, two groups) and sun exposure (OR = 0.51, $95\%$CI: 0.24–0.98; p-value = 0.049). In addition, Table 6 showed that in regression logistic crude models, there was a significant protective association between spine BMD (classifications, two groups) (OR = 0.69, $95\%$CI: 0.50–0.94; p-value = 0.023) BMD diagnosis (classifications, two groups) (OR = 0. 69, $95\%$CI: 0.49–0.87; p-value = 0.036) and sun exposure (Table 6). According to Table 6, there was no significant association between vitamin D intake and BMD in regression logistic multivariable and crude models.
Table 6Association (OR$95\%$CI) between serums vitamin D3, vitamin D intake, sun exposure, and bone mass measurements (BMD).CategoriesSerums vitamin D3Vitamin D intakeSun exposureModel AModel BModel AModel BModel AModel BOR ($95\%$CI)P-Value*OR ($95\%$CI)P-Value*OR ($95\%$CI)P-Value*OR ($95\%$CI)P-Value*OR ($95\%$CI)P-Value*OR ($95\%$CI)P-Value*Spine0.97 (0.94–1.01)0.200 0.92 (0.86–0.99) 0.025 1.10 (0.82–1.49)0.4990.68 (0.22–2.10)0.506 0.69 (0.50–0.94) 0.023 0.72 (0.46–1.07)0.111Total hip1.02 (0.98–1.06)0.2541.00 (0.94–1.05)0.9381.22 (0.89–1.67)0.2051.01 (0.99–1.04)0.9871.02 (0.74–1.41)0.8721.24 (0.59-2,63)0.562Femoral neck1.01 (0.97–1.05)0.4521.00 (0.95–1.05)0.8331.05 (0.79–1.39)0.7330.97 (0.29–3.21)0.9671.06 (0.80–1.42)0.6620.97 (0.64–1.47)0.905BMD diagnosis0.99 (0.96–1.03)0.9210.98 (0.93–1.02)0.4171.05 (0.78–1.42)0.7010. 99 (0.31–3.17)0.994 0. 69 (0.49–0.87) 0.036 0.51 (0.24–0.98) 0.049 Mode A: Crude modelsModel B: Models adjusted for age, BMI, PTH, and Calcitonin.* Logistic regression models.*Benjamini–Hochberg correction was applied to all p-values: all p-values are displayed after this correction; significant values are given in bold.
## Discussion
According to the result of our study, there is a significant and moderate correlation between Spine BMD and serum 25(OH)D. In addition, there is a significant, moderate, and negative correlation between Spine BMD and BMD diagnosis (osteopenia and osteoporosis) with sun exposure. The results of the correlation between serum 25(OH)D levels and BMD values are found to be controversial [8]. While certain studies have failed to find any association between these two variables, others have suggested positive correlations between serum 25(OH)D levels and BMD values.
In line with our finding, Khashayar et al. reported 25(OH)D levels were inversely correlated with BMD values at the total hip and spine in both sexes [24]. In addition, Kamineni concluded Vitamin D deficiency coexists with low BMD [25]. They concluded that vitamin D insufficiency is among the common risk factor for osteoporosis-related to low bone mass and increased bone remodeling [25]. Contrary to these findings, a study on patients with low BMD in the Southeast Asian population concluded that there is no direct association between serum 25(OH)D levels and BMD [26]. Another study revealed no association between BMD and serum vitamin D levels [27].
In addition, Chhantyal et al. reported that free vitamin D was significantly related to lumbar BMD; however, there was no significant association between BMD at different sites as well as fragile vertebral fracture total serum with vitamin D levels [28].
Moreover, our results suggest that sunlight exposure reduced the risk of osteoporosis and osteopenia and increased BMD. This finding aligns with previous studies exploring the links between sunlight exposure and BMD and osteoporosis [29, 30].
Although, in a previous study, we showed a correlation between some factors with vitamin D [31, 32]. Nevertheless, the association between fracture and total vitamin D remains controversial and unclear.
Undoubtedly, osteoporosis is a widely known predisposing factor for fracture, and vitamin D deficiency has been assumed to be a predictor for osteoporotic fractures [33]. Furthermore, vitamin D insufficiency was regarded as an important risk factor for fragile vertebral fractures in women and men [34]. A study of community-dwelling postmenopausal women found that sufficient vitamin D status might decrease—the risk of future fracture risk [35].
Discrepancies and inconsistency between studies may be attributed to (a) many of these population-based studies have recruited subjects with relatively good health status and, therefore, the lower prevalence of severe vitamin D deficiency and osteoporosis; (b) this study’s sites used for densitometry measurement affect the possible association between 25(OH)D and BMD; (c) also, sex, age, and physical activity vary in these studies.
Surprisingly, there was no significant difference between dietary intakes in the two groups in our study. Still, participants with osteopenia and osteoporosis significantly consumed a higher amount of soluble fiber than the normal BMD group. On the contrary, in the Framingham Offspring Study, associations with hip bone loss were not observed for women, although higher dietary fiber intake may modestly lower bone loss in men at the hip [36]. Data about the relation between fiber and bone turnover biomarkers showed either an increase, decrease, or no changes in bone formation and resorption markers [36].
Our study had its strengths included; this is the first study in Iran that considers all factors related to vitamin D status. Given the geographical location and the high prevalence of vitamin D deficiency in Iran, this study can help interpret the situation of vitamin D deficiency in Iran and countries with similar geographical conditions. Studies have shown that measurement methods can partially explain the lack of correlation between factors [36]. Another strength of our research is using standard methods to measure serum vitamin D and diagnose bone problems. In addition, the use of a valid FFQ and its completion by a nutritionist also assured us that the recall bias, one of the most common biases in retrospective studies, has been minimized.
Like any other study, our study had its limitations. One of our study s limitations was sample loss. So that some patients did not go to the BMD measurement center due to the COVID-19 situation (quarantine); since this problem was not anticipated at the time of study design, the COVID-19 pandemic outbreak also affected our sampling, and we lost some participants for the final analysis. To this end, modification in sampling protocols may be necessary for future studies. Therefore, risk management and quality assurance should be done more carefully and revised for future studies. Another limitation of our study was that it was not representative, so regarding variables such as age and gender, our study participants were not representative of the general population. Since this study was only a pilot study and the study population was deliberately selected from Sirjan Gol Gohar Company staff to highlight the job status more. Therefore, future studies with a large sample size and considering age and sex, and other confounding factors are necessary to confirm the results of our study. Another limitation of our study was the high risk of recall bias due to its retrospective nature. However, by taking the help of trained experts to collect data and complete the questionnaires, we were able to minimize this bias to a certain extent. On the other hand, using blood samples and serum levels of indicators allowed us to examine the data more precisely.
## Conclusion
In conclusion, although the results of our study showed a significant association/correlation between some components of vitamin D status, such as exposure to sunlight or serum levels, we failed to demonstrate the association between dietary vitamin D intake and BMD. Nevertheless, our results support previous studies, which concluded that serum 25(OH)D levels and sun exposure are correlated with bone mass. Future prospective studies considering confounding factors are recommended to confirm the results and elucidate possible mechanisms.
## References
1. 1.Saag KG, Morgan SL, Julian B. Osteopenic bone diseases. Clinical Primer of Rheumatology. 2003:278–285.
2. Salari N, Ghasemi H, Mohammadi L, Rabieenia E, Shohaimi S, Mohammadi M. **The global prevalence of osteoporosis in the world: a comprehensive systematic review and meta-analysis**. *J Orthop Surg Res* (2021.0) **16** 1-20. PMID: 33397415
3. Irani AD, Poorolajal J, Khalilian A, Esmailnasab N, Cheraghi Z. **Prevalence of osteoporosis in Iran: a meta-analysis**. *J Res Med sciences: official J Isfahan Univ Med Sci* (2013.0) **18** 759
4. 4.Peck W. Consensus development conference: diagnosis, prophylaxis, and treatment of osteoporosis. Am J Med. 1993;94(6):646–650.
5. Ardawi MSM, Maimany AA, Bahksh TM, Nasrat HA, Milaat WA, Al-Raddadi RM. **Bone mineral density of the spine and femur in healthy Saudis**. *Osteoporos Int* (2005.0) **16** 43-55. DOI: 10.1007/s00198-004-1639-9
6. Keramat A, Patwardhan B, Larijani B, Chopra A, Mithal A, Chakravarty D. **The assessment of osteoporosis risk factors in iranian women compared with indian women**. *BMC Musculoskelet Disord* (2008.0) **9** 1-10. DOI: 10.1186/1471-2474-9-28
7. Chhantyal K, He L, Mo J, Yin M, He T, Chen Y. **Free vitamin D correlate better with bone mineral density and thoracolumbar junction osteoporotic vertebral fractures than serum vitamin D**. *BMC Musculoskelet Disord* (2020.0) **21** 164. DOI: 10.1186/s12891-020-3179-7
8. Khodabakhshi A, Mahmoudabadi M, Vahid F. **The role of serum 25 (OH) vitamin D level in the correlation between lipid profile, body mass index (BMI), and blood pressure**. *Clin Nutr ESPEN* (2022.0) **48** 421-6. DOI: 10.1016/j.clnesp.2022.01.007
9. Choi S-W, Kweon S-S, Choi J-S, Rhee J, Lee Y-H, Nam H-S. **The association between vitamin D and parathyroid hormone and bone mineral density: the Dong-gu Study**. *J Bone Miner Metab* (2016.0) **34** 555-63. DOI: 10.1007/s00774-015-0696-9
10. Nguyen HT, von Schoultz B, Nguyen TV, Dzung DN, Duc PT, Thuy VT. **Vitamin D deficiency in northern Vietnam: prevalence, risk factors and associations with bone mineral density**. *Bone* (2012.0) **51** 1029-34. DOI: 10.1016/j.bone.2012.07.023
11. Pourhashem Z, Bayani M, Noreddini H, Bijani A, Hosseini SR. **Prevalence of osteoporosis and its association with serum vitamin D level in older people in Amirkola, North of Iran**. *Caspian J Intern Med* (2012.0) **3** 347. PMID: 26557284
12. Zhen D, Liu L, Guan C, Zhao N, Tang X. **High prevalence of vitamin D deficiency among middle-aged and elderly individuals in northwestern China: its relationship to osteoporosis and lifestyle factors**. *Bone* (2015.0) **71** 1-6. DOI: 10.1016/j.bone.2014.09.024
13. Bruno AG, Burkhart K, Allaire B, Anderson DE, Bouxsein ML. **Spinal loading patterns from biomechanical modeling explain the high incidence of vertebral fractures in the thoracolumbar region**. *J Bone Miner Res* (2017.0) **32** 1282-90. DOI: 10.1002/jbmr.3113
14. Chapuy MC, Arlot ME, Delmas PD, Meunier PJ. **Effect of calcium and cholecalciferol treatment for three years on hip fractures in elderly women**. *BMJ: Br Med J* (1994.0) **308** 1081. DOI: 10.1136/bmj.308.6936.1081
15. Group RT. **Oral vitamin D3 and calcium for secondary prevention of low-trauma fractures in elderly people (randomised evaluation of calcium or vitamin D, RECORD): a randomised placebo-controlled trial**. *The Lancet* (2005.0) **365** 1621-8. DOI: 10.1016/S0140-6736(05)63013-9
16. Porthouse J, Cockayne S, King C, Saxon L, Steele E, Aspray T. **Randomised controlled trial of calcium and supplementation with cholecalciferol (vitamin D3) for prevention of fractures in primary care**. *BMJ* (2005.0) **330** 1003. DOI: 10.1136/bmj.330.7498.1003
17. Jackson RD, LaCroix AZ, Gass M, Wallace RB, Robbins J, Lewis CE. **Calcium plus vitamin D supplementation and the risk of fractures**. *N Engl J Med* (2006.0) **354** 669-83. DOI: 10.1056/NEJMoa055218
18. Bischoff-Ferrari HA, Willett WC, Wong JB, Giovannucci E, Dietrich T, Dawson-Hughes B. **Fracture prevention with vitamin D supplementation: a meta-analysis of randomized controlled trials**. *JAMA* (2005.0) **293** 2257-64. DOI: 10.1001/jama.293.18.2257
19. Macdonald HM, Mavroeidi A, Barr RJ, Black AJ, Fraser WD, Reid DM. **Vitamin D status in postmenopausal women living at higher latitudes in the UK in relation to bone health, overweight, sunlight exposure and dietary vitamin D**. *Bone* (2008.0) **42** 996-1003. DOI: 10.1016/j.bone.2008.01.011
20. Asghari G, Rezazadeh A, Hosseini-Esfahani F, Mehrabi Y, Mirmiran P, Azizi F. **Reliability, comparative validity and stability of dietary patterns derived from an FFQ in the Tehran lipid and glucose study**. *Br J Nutr* (2012.0) **108** 1109-17. DOI: 10.1017/S0007114511006313
21. Bodekaer M, Harrison GI, Philipsen P, Petersen B, Triguero-Mas M, Schmalwieser A. **Personal UVR exposure of farming families in four european countries**. *J Photochem Photobiol B* (2015.0) **153** 267-75. DOI: 10.1016/j.jphotobiol.2015.10.002
22. Baradaran Mahdavi S, Mansourian M, Shams E, Qorbani M, Heshmat R, Motlagh ME. **Association of sunlight exposure with sleep hours in iranian children and adolescents: the CASPIAN-V Study**. *J Trop Pediatr* (2020.0) **66** 4-14. PMID: 31098631
23. Lu Y, Genant HK, Shepherd J, Zhao S, Mathur A, Fuerst TP. **Classification of osteoporosis based on bone mineral densities**. *J Bone Miner Res* (2001.0) **16** 901-10. DOI: 10.1359/jbmr.2001.16.5.901
24. Khashayar P, Meybodi HRA, Hemami MR, Keshtkar A, Dimai HP, Larijani B. **Vitamin D status and its relationship with bone mineral density in a healthy iranian population**. *Revista Brasileira de Ortopedia (English Edition)* (2016.0) **51** 454-8. DOI: 10.1016/j.rboe.2015.09.011
25. Kamineni V, Latha AP, Ramathulasi K. **Association between serum 25-hydroxyvitamin D levels and bone mineral density in normal postmenopausal women**. *J mid-life health* (2016.0) **7** 163. DOI: 10.4103/0976-7800.195694
26. Chandran M, Hoeck H, Wong H, Zhang R, Dimai H. **Vitamin D status and its relationship with bone mineral density and parathyroid hormone in southeast asian adults with low bone density**. *Endocr Pract* (2011.0) **17** 226-34. DOI: 10.4158/EP10202.OR
27. Alkhenizan A, Mahmoud A, Hussain A, Gabr A, Alsoghayer S, Eldali A. **The relationship between 25 (OH) D levels (vitamin D) and bone mineral density (BMD) in a saudi population in a community-based setting**. *PLoS ONE* (2017.0) **12** e0169122. DOI: 10.1371/journal.pone.0169122
28. He L, Mo J, Yin M, He T, Chen Y, Yang Y. **Free vitamin D correlate better with bone mineral density and thoracolumbar junction osteoporotic vertebral fractures than serum vitamin D**. *BMC Musculoskelet Disord* (2020.0) **21** 1-10
29. Kanemura H, Hatakeyama K, Sano F, Yagasaki H, Sugita K, Aihara M. **Effect of sunlight exposure on bone mineral density in children with severe disability**. *Neuropediatrics* (2016.0) **47** 233-7. DOI: 10.1055/s-0036-1584083
30. Min C-Y, Yoo D-M, Choi H-G. **Associations between Physical Activity, Sunshine duration and osteoporosis according to obesity and other Lifestyle factors: a nested case–control study**. *Int J Environ Res Public Health* (2021.0) **18** 4437. DOI: 10.3390/ijerph18094437
31. 31.Khodabakhshi A, Mahmoudabadi M, Vahid F. The role of serum 25 (OH) vitamin D level in the correlation between lipid profile, body mass index (BMI), and blood pressure. Clinical Nutrition ESPEN. 2022.
32. Doaei S, Jarrahi S, Torki S, Haghshenas R, Jamshidi Z, Rezaei S. **Serum vitamin D level may be associated with body weight and body composition in male adolescents; a longitudinal study**. *Pediatr Endocrinol Diabetes Metabolism* (2020.0) **26** 125-31. DOI: 10.5114/pedm.2020.97466
33. Endo N, Oinuma T. **Serum 25-OHD insufficiency as a risk factor for hip fracture**. *J Bone Miner Metab* (2007.0) **25** 147-50. DOI: 10.1007/s00774-007-0749-9
34. Maier G, Seeger J, Horas K, Roth K, Kurth A, Maus U. **The prevalence of vitamin D deficiency in patients with vertebral fragility fractures**. *The Bone & Joint Journal* (2015.0) **97** 89-93. DOI: 10.1302/0301-620X.97B1.34558
35. LeBoff MS, Kohlmeier L, Hurwitz S, Franklin J, Wright J, Glowacki J. **Occult vitamin D deficiency in postmenopausal US women with acute hip fracture**. *JAMA* (1999.0) **281** 1505-11. DOI: 10.1001/jama.281.16.1505
36. Dai Z, Zhang Y, Lu N, Felson DT, Kiel DP, Sahni S. **Association between Dietary Fiber Intake and Bone loss in the Framingham offspring study**. *J Bone Miner Res* (2018.0) **33** 241-9. DOI: 10.1002/jbmr.3308
|
---
title: Optimizing the biodegradability and osteogenesis of biogenic collagen membrane
via fluoride-modified polymer-induced liquid precursor process
authors:
- Xiyan Li
- Chuangji Li
- Mengxi Su
- Xinyi Zhong
- Yihan Xing
- Zhengjie Shan
- Shoucheng Chen
- Xingchen Liu
- Xiayi Wu
- Quan Liu
- Ye Li
- Shiyu Wu
- Zhuofan Chen
journal: Science and Technology of Advanced Materials
year: 2023
pmcid: PMC10013244
doi: 10.1080/14686996.2023.2186690
license: CC BY 4.0
---
# Optimizing the biodegradability and osteogenesis of biogenic collagen membrane via fluoride-modified polymer-induced liquid precursor process
## ABSTRACT
Biogenic collagen membranes (BCM) have been widely used in guided bone regeneration (GBR) owing to their biodegradability during tissue integration. However, their relatively high degradation rate and lack of pro-osteogenic properties limit their clinical outcomes. It is of great importance to endow BCM with tailored degradation as well as pro-osteogenic properties. In this study, a fluoride-modified polymer-induced liquid precursor (PILP) based biomineralization strategy was used to convert the collagen membrane from an organic phase to an apatite-based inorganic phase, thus achieving enhanced anti-degradation performance as well as osteogenesis. As a result, three phases of collagen membranes were prepared. The original BCM in the organic phase induced the mildest inflammatory response and was mostly degraded after 4 weeks. The organic-inorganic mixture phase of the collagen membrane evoked a prominent inflammatory response owing to the fluoride-containing amorphous calcium phosphate (F-ACP) nanoparticles, resulting in active angiogenesis and fibrous encapsulation, whereas the inorganic phase induced a mild inflammatory response and degraded the least owing to the transition of F-ACP particles into calcium phosphate with high crystallinity. Effective control of ACP is key to building novel apatite-based barrier membranes. The current results may pave the way for the development of advanced apatite-based membranes with enhanced barrier performances.
## GRAPHICAL ABSTRACT
## Introduction
Barrier membranes have been widely used in guided bone regeneration (GBR) and guided tissue regeneration (GTR). They create a stable healing space for bone defects by isolating soft tissue ingrowth [1]. Compared with other types of barrier membranes, biogenic collagen membranes (BCM) have a substantial advantage owing to their biodegradability during tissue integration. With a unique biologically derived smooth-rough membrane structure, the rough side of the BCM favors protein adsorption and cell attachment, whereas the smooth side prevents epithelial and fibroblast cells from entering the defected area [2]. However, this dual-functional effect gradually weakens with membrane degradation, a double-edged sword, which is a crucial property of BCM.
For clinical cases with large defects and requirements for long-term space maintenance, the degradation rate of the collagen membrane is relatively fast and may not provide sufficient barrier function. Following degradation, the space is filled with soft tissue owing to the lack of pro-osteogenic properties of the collagen membrane. These factors reduce the clinical effects of the GBR technique [3]. Therefore, it is very important to endow the collagen membrane with tailored degradation and pro-osteogenic properties.
At present, collagen membrane modification is mainly based on a cross-linking strategy combined with the addition of growth factors. The cross-linking technique using glutaraldehyde, 1-Ethyl-3-(3-dimethylaminopropyl) carbodiimide hydrochloride, and ribose has been shown to enhance the anti-degradation properties of collagen-based materials [4–6]. Growth factors have been used to endow the collagen membrane with additional biological functions, such as stromal cell-derived factor-1 alpha [7], basic fibroblast growth factor [8], interleukin-4, and interferon-gamma [9,10]. However, few cross-linking agents have low cytotoxicity [6], and the exogenous protein products result in issues, such as evoking additional oxidative stress on cells as well as the high cost and complex fabrication of protein additives [11]. Therefore, further improvements are required to optimize the properties of the collagen membrane.
The essential components of GBR are apatite-based bone substitutes and barrier membranes. As insoluble compounds with a stable inorganic crystal phase, apatite-based bone substitutes have a strong resistance to degradation. Our laboratory data and previous histological observations confirmed that the bone substitute in maxillary sinus elevation surgery can be stable for more than 14 years [12]. Accordingly, we propose a new strategy to convert the collagen membrane from an organic phase to an apatite-based inorganic phase, aiming to endow the membrane with enhanced anti-degradation performance via this phase transition. Moreover, nutrient elements can be introduced to the modified inorganic crystal phase of the membrane to promote bone regeneration while avoiding the application of cross-linking agents and exogenous proteins.
The polymer-induced liquid precursor (PILP) mineralization technique can translate a collagen template into an apatite-based inorganic phase in a bottom-up manner while retaining the microscopic morphology of the organic template [13,14]. Saxena et al. proposed a fluoride-modified PILP strategy in 2018, which combined the concepts of crystal phase transition and fluorine nutrient element modification [15]. In our previous work, biogenic hydroxyapatite was successfully modified by fluorine and improved osteogenesis and immune regulation performance, which was confirmed by both in vivo and in vitro results [16–19]. Thus, we believe that fluoride ions are a suitable nutrient element for imparting osteogenic properties to apatite-based materials and that this modified PILP strategy is a feasible solution to meet our demands.
As the feasibility and biological effect of this organic-inorganic phase transition of the collagen membrane remains unknown, we first developed a biogenic-derived collagen membrane (BCM) as the organic phase and a fluoride-containing biomineralized BCM (F-mBCM) as an organic-inorganic mixture phase. Their degradation and osteogenic performance and the underlying mechanism related to the immune response were then explored. On this basis, the membrane was further transformed to a fully inorganic phase (SF-mBCM) following a re-evaluation of its biological effects. This phase transition strategy of collagen membranes aims to combine the advantages of absorbable and non-absorbable barrier membranes. It proposed a new solution for optimizing the degradation property of the collagen membrane and may pave a new way for the development of advanced barrier membranes.
## Preparation of BCM
Porcine peritonea were collected from the abdominal tissues of one-year-old black pigs following the removal of the fat tissue. Samples were divided into 4 cm × 4 cm pieces and decellularized using a chemical immersion procedure, as previously described [20,21]. In brief, the samples were treated with acid and alkali, hypertonic, and neutralization treatments, followed by dehydration and degreasing and were labeled as BCM. BCMs were fabricated after lyophilization in a vacuum freeze dryer (Christ, ALPHA 2-4, Germany) for 13 h.
## Preparation of F-mBCM
The biomimetic mineralization protocol is based on the principles of PILP [22]. Mineral solution A was prepared by mixing carboxymethyl chitosan (CMC) (920 mg/L) and CaCl2 (1332 mg/L) in 1000 mL of deionized water with stirring (600 rpm) until the CMC powder was completely dissolved. K2HPO4 (1460 mg/L) was dissolved in 1 L of deionized water to obtain mineral solution B. Solution B was slowly added to an equal amount of solution A under stirring. A series of fluoride concentrations were prepared (1250, 2500, and 5000 ppm F−) with deionized water and NaF. One milliliter of each NaF solution was added to 25 mL mineral solution B before mixing with solution A. The final F− concentrations of the mixed mineral solutions were 25, 50, and 100 ppm. Samples were then soaked in 50 mL of the final mineral solution for 5 days, followed by daily replacement of the mineral solution and lyophilization to obtain F25 (25 ppm), F50 (50 ppm), and F100 (100 ppm). F25, F50, and F100 were also termed as F-mBCM.
## Preparation of SF-mBCM
Samples from F25 were soaked in 0.25 mol/L NaF solution for 24 h and lyophilized. The samples were then sintered at 800°C for 2 h in an air atmosphere at a heating rate of 10°C min−1 in a muffle furnace (SGM6812BK, Sigma Furnace Industry, China) to get SF-25/SF-mBCM.
## Physicochemical characterization and biocompatibility evaluation
General views of all the samples, including the rough and smooth sides, were captured using an SLR camera (D160, Nikon, Japan). All the samples from each group were sputter-coated with gold and palladium. Both sides of the membrane were examined by scanning electron microscopy (SEM, SU8220, HITACHI, Japan). Energy dispersive spectrometry (EDS) was used to evaluate the elemental distribution of the membrane surface. Crystalline phases were examined using an X-ray powder diffractometer (XRD; Empyrean, Netherlands) and matched with the Joint Committee on Powder Diffraction Standards (JCPDS) database. Chemical bonding was characterized by a Fourier transform infrared spectrometer (FTIR, Nicolet6700-Continuum, Thermo Fisher, U.S.A.) with a resolution of 650–4000 cm−1. The thermal stability and proportion of the inorganic phase were tested by thermogravimetric analysis (TG), derivative thermogravimetry (DTG), and differential scanning calorimetry (DSC) with a synchronous thermal analyzer (Nicolet 6700, Thermo Fisher, U.S.A.) in an air atmosphere at a heating rate of 10°C/min, with the test temperature ranging from 22 to 1100°C.
The biocompatibility of the membrane was evaluated using Cell Counting Kit-8 (CCK-8) (Dojindo, Kumamoto, Japan) assay in Raw 264.7 cells (Chinese Academy of Sciences, Shanghai, China). Each sample was soaked in 2 mL complete medium for 24 h. The supernatant was collected and used as the conditioned medium. Cells were plated at a density of 2000 per well in 96-well plates and incubated with conditioned medium. The medium was replaced with a $10\%$ CCK-8 solution for 45 min on days 1, 3, and 5. A microplate reader (TECAN, Thermo Scientific) was used to measure the absorbance at 450 nm.
To demonstrate the biomimetic mineralization process, collagen membranes under PILP were collected after 12 h and 5 days. Samples were embedded in resin and cut into cross sections (EM UC7, Leica, Germany). The Cu mesh was examined by transmission electron microscopy (TEM, HT7800, HITACHI, Japan; Talos-F200S, FEI, U.S.A.) at an accelerating voltage of 80 kV. Selected area electron diffraction (SAED) was used to detect the early characteristics of the crystalline phase after 12 h. To identify the fluoride-amorphous calcium phosphate (F-ACP) nanoparticles in the conditioned medium, the membrane was soaked in 1 mL water for 24 h. The supernatant was dispersed on a Cu mesh. Transmission electron microscopy (TEM, Tecnai G2 F30, FEI, America) and SAED were used to examine F-ACP and its crystalline phase.
## Animal surgery
All experiments were approved by the Institutional Animal Care and Use Committee, Sun Yat-sen University, and performed according to the animal use protocol (No. SYSU-IACUC-2022-000673).
For the rat calvarial defect model: Eight-week-old male Sprague-Dawley (SD) rats (220–250 g) were used in this study. General anesthesia was administered before surgery using Zoletil®50 containing zolazepam and tiletamine (VIRBAC, France) at a dose of 60 mg/kg. The sample was soaked in 1 mL of tail blood for 15 min. The rat parietal bone was carefully exposed and two 5-mm critical-sized calvarial defects were prepared using a sterile trephine bur. The calvarial defects were covered with a membrane. In the blank group, the defect was solely filled with a blood clot. The incision was carefully sutured without moving the membrane. Each group contained at least 3 biological replicates. The animals were sacrificed after four weeks. For the rat subcutaneous model, pieces of 1 × 1 cm membranes (F25 and SF25) were inserted into the subcutaneous pouch on the left and right of the dorsum under general anesthesia. The rats were euthanized 3 days after the implantation surgery for subsequent analysis. The parietal bone with the defect area and the subcutaneous pouch with membranes were removed from the animal and fixed with paraformaldehyde for 24 h. Micro-CT (Bruker SkyScan 1276, 85 kV, 200 μA, 15 μm resolution) was used to evaluate osteogenesis and the residual amount of the membrane for the calvarial defects.
## Histological section staining
After micro-CT evaluation, the samples were decalcified using $4\%$ ethylenediaminetetraacetic acid for 4 weeks. The samples were then embedded in paraffin and cut into 4-µm slices using a rotary microtome (Leica, RM2255, Germany), followed by dewaxing and hydration. Hematoxylin and eosin (H&E) and Masson’s trichrome staining were used to evaluate the biocompatibility, integration, and degradation properties of the membrane. For H&E staining, the cell nuclei were stained with Mayer’s hematoxylin (Sigma-Aldrich, St. Louis, MO, U.S.A.), whereas the cell plasma and extracellular matrix were stained with eosin (Sigma-Aldrich). Masson’s trichrome staining (G1340, Solarbio, China) was performed according to the manufacturer’s instructions. Muscle fibers were stained red, and collagen fibers were stained green/blue. Immunohistochemistry (IHC) staining of α-smooth muscle actin (α-SMA) was applied. Endogenous peroxidase activity was eliminated by incubating in $3\%$ H2O2 for 15 min. The slides were blocked for 1 h and incubated with α-SMA antibodies (1:1000; Abcam, Cambridge, MA, U.S.A.) overnight at 4°C. The sections were then incubated with a secondary antibody (Gene Tech, Shanghai, China) for 30 min at room temperature. The antibody complexes were visualized by diaminobenzidine (DAB) solution (Gene Tech, Shanghai, China) and counterstained with Mayer’s hematoxylin for 90 s. Tartrate-resistant acid phosphatase (TRAP) (Servicebio, Wuhan, China) staining was performed to identify TRAP+ MNGCs. The sections were rehydrated and incubated with TRAP staining solution for 1 h at 37°C followed by staining with hematoxylin. Images were captured using a microscope slide scanner (Aperio AT2; Leica, Germany). For semi-quantitative analysis, the number of multinucleated giant cells (MNGC) around the residual material and the number of vessels were counted from three non-overlapping views (1 mm2) using Aperio ImageScope 12.3 (Leica Biosystems, Germany). The percentage of residual membrane area was calculated in three non-overlapping views (1 mm2) (Supplementary Figure S1).
## Cell culture
Raw 264.7 cells were cultured under standard conditions (37°C, with $5\%$ CO2 atmosphere and $100\%$ relative humidity) in Dulbecco’s modified Eagle’s medium (Thermo Scientific, Australia) containing $5\%$ fetal bovine serum (Thermo Scientific) and $1\%$ (v/v) penicillin/streptomycin (Thermo Scientific). The culture medium was changed every two days. The cells were gently scraped off and passaged once they reached $80\%$ confluency.
## RNA extraction and quantitative real-time PCR (RT-PCR)
Cells were stimulated with the conditioned medium of collagen membranes for 24 h. Total RNA was extracted using TRIzol reagent (Life Technologies, U.S.A.) following the manufacturer’s instructions. RNA was reverse-transcribed into single-stranded cDNA using the TaKaRa PrimeScriptTM Master Mix (Perfect Real Time Kit; Takara, Japan). RT-PCR was performed on an equivalent quantity of cDNA using Hieff® qPCR SYBR Green Master Mix (No Rox; Yeasen, Shanghai, China) according to the manufacturer’s instructions. The relative mRNA expression of inflammation-related genes [interleukin 1 beta (IL-1β), tumor necrosis factor-alpha (TNF-α), interleukin- 6 (IL-6), nuclear factor kappa-B1 (NFκB1), and interleukin 1 receptor antagonist (IL-1rn)], osteoclastogenesis-related genes [macrophage colony-stimulating factor (MCSF)], angiogenic factors [vascular endothelial growth factor (VEGF)], osteogenesis- and fibrosis-related genes [transforming growth factor-β1 (TGF-β1) and TGF-β3)], and collagenase [matrix metalloprotein 9, (MMP9)] were detected in the F-mBCM and blank groups. The relative mRNA expression of inflammation-related genes (IL-1β, TNF-α, IL-6, NFκB1, and IL-1rn), osteoclastogenesis-related genes (MCSF), osteogenesis- and fibrosis-related genes [TGF-β1, TGF-β3, and oncostatin M, (OSM)], and collagenase (MMP9) were detected in the SF25 and blank groups. The 2−ΔΔCt method was used to calculate relative gene expression after normalization against the housekeeping gene glyceraldehyde-3-phosphate dehydrogenase (GAPDH). Primers used in this study are listed in Supplementary Table S1.
## Statistical analysis
GraphPad Prism (version 8.0.2) software was used for statistical analysis. All data were shown as the mean ± standard deviation (SD). The one-factor analysis of variance (ANOVA) comparison test was used to evaluate the differences among groups followed by Tukey (multiple comparisons) post hoc test. For all tests, significance was considered at $P \leq 0.05.$ The significant difference between groups was marked as follows: *, $P \leq 0.05$, **, $P \leq 0.01$, ***, $P \leq 0.001$, ****, $P \leq 0.0001.$ Data without * were not significantly different.
## Physicochemical properties and biocompatibility of F-mBCM
The membrane morphologies were evaluated using an SLR camera and SEM. *The* general view of biomineralized F-mBCM is consistent with that of BCM. SEM showed extrafibrillar deposits on the fiber surface in the F-mBCM groups, whereas the number of deposits increased with an increase in the fluoride ion content (Figure 1(a,b)). EDS detected calcium and phosphate elements in the F0–F100 samples, indicating mineralization in collagen membranes. Fluoride was detected in the F25–F100 samples, indicating the successful doping of fluoride onto the biomineralized collagen membranes (Figure 1(c)). The crystal phase of F-mBCM was composed of Ca3(PO4)2. xH2O, and Ca5(PO4)3(OH), according to the XRD results. In groups F25, F50, and F100, as the degree of fluorination increased, the shift in the peak value was more pronounced in the enlargement of 50° (2θ), which indicated the successful doping of fluorine (Figure 1(d)). The FTIR spectrum of the BCM group showed obvious absorption peaks of the collagen amide bands at 1037, 1240, and 1552 cm−2, respectively. After mineralization, the absorption peaks of the amide disappeared, and the absorption peaks of phosphate appeared at 1030 cm−2, indicating the existence of calcium phosphate. Amide absorption peaks of amide, methylene and hydroxyl were observed at 1750, 2930, and 3730 cm−2, respectively, among all mineralized groups (Figure 1(e)). A decrease in cell vitality was observed in F100 after 5 days of stimulation in the cytotoxicity test, whereas the other samples showed no toxic effects after 5 days of treatment (Figure 1(f)). Figure 1.Physicochemical properties and biocompatibility of BCM, F0, F25, F50, and F100. ( a) General view and SEM of BCM, F0, F25, F50, and F100 of the rough side. ( b) General view and SEM of BCM, F0, F25, F50, and F100 of the smooth side. ( c) EDS shows the elemental distribution of collagen fiber in (a). ( d) XRD spectra of BCM, F0, F25, F50, and F100. ( e) FTIR spectra of BCM, F0, F25, F50, and F100. ( f) Cytotoxicity test of F-mBCM in Raw 264.7 by CCK-8 assay. * Significant differences compared with the blank group or between two groups. *, $P \leq 0.05$, **, $P \leq 0.01$, ***, $P \leq 0.001$, ****, $P \leq 0.0001$, data without * were not significantly different (NS). Error bars represented mean ± SD.
## Formation of biomimetic mineralized collagen fibers
TEM was used to detect the intrafibrillar mineralization structure of F-mBCM. After 12 h of mineralization, TEM captured partially mineralized collagen in F25 and F50 with hierarchical intrafibrillar banding, whereas heavily mineralized fibrils were found in F100 (Figure 2(a)). The SAED patterns showed a polycrystalline morphology in the F25 and F50 groups, whereas F100 showed a monocrystalline morphology pattern (Figure 2(b)). After 5 days of mineralization, the cross-section of the collagen fibers was observed using TEM. Both intrafibrillar mineralization (yellow asterisks) and extrafibrillar mineralization (blue arrows) were observed around the fibers. The amount of extrafibrillar deposits (blue arrows) around the collagen fibers on the fiber surface increased as the fluoride ion content increased (Figure 2(c)). Figure 2.TEM showed the formation of biomimetic mineralized collagen fibers. ( a) TEM showed the initial intrafibrillar mineralization process of collagen fibers after biomineralization for 12 h. (b) SAED patterns of (a). ( c) TEM showed the cross-section of collagen fibers after biomineralization for 5 days.
## Osteogenesis, degradation, and integration evaluation of F-mBCM
Biological performance, including osteogenesis, degradation, and integration properties, was evaluated using a rat calvarial defect model. Micro-CT and Masson’s trichrome staining showed that the defects were partially repaired with new bone tissue in the BCM, F25, F50, and F100 groups, whereas few new bone islands were observed in the blank group (Figures 3(a,b) and 4(e)). Few blood vessels were observed in the blank and BCM groups, whereas more blood vessels with fibrous formation were found in all the F-mBCM groups around the membrane (Figures 3(d,e) and 4(f)). The contour of the residual mineralized membrane covered the bone defects, suggesting that the membrane position remained stable without deviation during the entire animal experiment (Figure 4(a)). In the BCM group, the membrane was mostly degraded and replaced with fibrous connective tissue. In the F-mBCM group, mineralized membrane fragments were retained with vascularized fibrous encapsulation formed around them (Figure 4(b)). TRAP+ MNGCs were observed around the residual fragments in the F-mBCM group, which were significantly higher than those in the BCM group (Figure 4(d,g)). No significant difference was found in the amount of residual membrane among F-mBCM groups (Figure 4(h), Supplementary Figure S1). Figure 3.Effects of F-mBCM on osteogenesis at the calvarial defect at 4 weeks following surgery. ( a) Micro-CT showed the new bone formation at the calvarial defect without barriers. ( b, c) Masson’s trichrome showed the (b) overview of calvarial defect areas and (c) higher magnification of new bone tissue (yellow asterisks). ( d) H&E staining and (e) α-SMA staining showed that few blood vessels were observed in blank and BCM groups, whereas more were found in F-mBCM groups (yellow arrows). Figure 4.Degradation and integration properties of F-mBCM at the calvarial defect at 4 weeks following surgery. ( a) Micro-CT revealed the residual membrane of F-mBCM over the calvarial defect (yellow area). ( b, c) H&E staining showed that the residual membrane was surrounded by fibrous tissue and MNGCs. ( d) TRAP staining showed the presence of TRAP+ MNGCs (blue arrows). Semi-quantitative analysis of (e) new bone volume, (f) the number of vessels, (g) the number of MNGCs, and (h) the percentage of residual membrane area. * Significant differences compared with the blank or BCM groups. *, $P \leq 0.05$, **, $P \leq 0.01$, ***, $P \leq 0.001$, ****, $P \leq 0.0001$, data without * were not significantly different (NS). Error bars represented mean ± SD.
## Effects of F-mBCM on modulating the immune response of macrophages
Because significant degradation of F-mBCM was observed in vivo, the effects of F-mBCM on the immune response of macrophages were further examined. Inflammation-related genes (IL-6, TNF-α, IL-1β, NFκB1, and IL-1rn) were significantly upregulated in the BCM and blank groups (Figure 5(a)). The expression of osteoclastogenesis-related genes (MCSF) was not significantly different from that of the control group. The expression of the angiogenic factor (VEGF) was upregulated in the F-mBCM groups without a significant difference. The fibrosis-enhancing factor, TGF-β1, was significantly upregulated in the F25 group, whereas TGF-β3 was significantly downregulated in both the BCM and F-mBCM groups. Collagenase (MMP9) was significantly upregulated in all F-mBCM groups (Figure 5(b)). Figure 5.The effect of F-mBCM on modulating immune response of Raw 264.7 in vitro. Relative mRNA expression of (a) inflammation-related genes (IL-1β, TNF-α, IL-6, NFκB1, and IL-1rn). ( b) *Osteoclast* genesis-related genes (MCSF), angiogenic factor (VEGF), osteogenesis- and fibrosis-related genes (TGF-β1 and TGF-β3), and collagenase (MMP9). ( c) TEM, (d EDS mapping, and (e) SAED showing the presence of F-ACP nanoparticles in the extracts of F-mBCM. * Significant differences compared with the blank group. *, $P \leq 0.05$, **, $P \leq 0.01$, ***, $P \leq 0.001$, ****, $P \leq 0.0001$, data without * were not significantly different (NS). Error bars represented mean ± SD.
To explore the causes of barrier membrane inflammation, we conducted a detailed literature review and found that amorphous calcium phosphate (ACP) nanoparticles are an important potential inflammatory factor. We detected the presence of fluoride-containing ACP (F-ACP) nanoparticles in the material extracts using TEM. The particle size increased from 30 to 200 nm with an increase in the fluoride ion concentration in the PILP system (Figure 5(c)). The EDS analysis showed that the elemental composition of the particles was calcium, phosphate, and fluoride (Figure 5(d)). SAED analysis showed that the crystalline phase was amorphous in F25, F50, and F100 (Figure 5(e)).
## Physicochemical properties and biocompatibility of SF-mBCM/SF25
High-temperature sintering was performed to stabilize the F-ACP nanoparticles on F-mBCM. After sintering at 800°C for 2 h, the general view of the generated SF25 was consistent with that of F25. SEM showed that the structure of the collagen membrane was well-preserved on both the rough and smooth sides, whereas the mineralized substances on the membrane surface partially fused to form rod-like shapes under high-temperature sintering. EDS analysis showed that the elemental composition of SF25 was calcium, phosphate, and fluoride (Figure 6(a)). XRD revealed a higher and narrower peak in the SF25 group than that in the F25 group, indicating the higher crystallinity of SF25 (Figure 6(b)). The FTIR spectrum of SF25 showed an absorption peak at 746 cm−2, which is possibly owing to the interaction between OH− and F− as further evidence of fluorine incorporation. The characteristic phosphate group bands at 1097 cm−2 and 1035 cm−2 were visible. This indicates a fluorohydroxyapatite composition in SF25 (Figure 6(c)). DSC results showed that a cold crystal peak was observed in the F25 group at 350°C, whereas an endothermic melting peak was observed in both the F25 and BCM groups. No obvious peaks were observed in the SF25 group (Figure 6(d)). The weight loss rates of SF25 differed from those of F25 and BCM. The latter two showed a peak at 75°C, representing the volatilization of water. Peaks of F25 and BCM at 340 and 300°C, respectively, represent the decomposition of organic components. No peak was found for SF25, indicating the absence of an organic component (Figure 6(e,f)). No toxic effects were found after 5 days of culture of Raw 264.7. SF25 significantly increased the cell viability on day 5 (Figure 6(g)). Figure 6.Physicochemical properties and biocompatibility of SF25. ( a) General view, SEM, and element distribution of SF25. ( b) XRD of BCM, F25, and SF25. ( c) FTIR of BCM, F25, and SF25. ( d) Cytotoxicity test of SF25 in Raw 264.7 by CCK-8 assay. ( e) DTG patterns. ( f) TG patterns. ( g) DSC patterns.
## Effects of SF25 on modulating the immune response
To determine the effects of SF25 on the immune response, macrophages were stimulated with SF25 extracts for 24 h. The expression of inflammation-related factors (IL-6, TNF-α, IL-1β, NF-κB1, and IL-1rn) was significantly reduced compared with that in the blank group. Osteoclastogenesis-related genes (MCSF) were significantly downregulated compared to those in the blank group. The fibrosis-enhancing factor, TGF-β1, was significantly downregulated, whereas TGF-β3 was significantly upregulated in the SF25 group. The collagenase (MMP9) was significantly downregulated in the SF25 group (Figure 7(a)). This decrease in immune response was also noticed using the rat subcutaneous model. H&E staining showed that more monocytes were observed in the F25 group compared with the SF25 group (Figure 7(b)). Figure 7.Effects of SF25 on modulating the immune response in vitro and in vivo. ( a) Relative mRNA expression of macrophage inflammation-related genes (IL-1β, TNF-α, IL-6, NF-κB1, and IL-1rn), osteoclastogenesis-related gene (MCSF), fibrosis-related genes (TGF-β1 and TGF-β3,) and collagenase (MMP9). ( b) H&E staining showed that more monocytes were observed around the residual membrane in a 3-day rat subcutaneous model. * Significant differences compared with the blank group. *, $P \leq 0.05$, **, $P \leq 0.01$, ***, $P \leq 0.001$, ****, $P \leq 0.0001$, data without * were not significantly different (NS). Error bars represented mean ± SD.
## Osteogenesis, degradation, and integration evaluation of SF25
In vivo experiments were performed to explore the osteogenesis, degradation, and integration properties of SF25. Newly formed bone lamellae were observed in H&E staining of the SF25 group, whereas the quantitative analyses of micro-CT showed a comparable amount of new bone formation in the SF25 and F25 groups (Figure 8(g)). Regarding the degradation and integration properties, the amount of residual membrane was significantly higher in the SF25 group than that in the F25 group (Figure 8(d,e,i), Supplementary Figure S1). The residual mineralized membrane in the SF25 group presented scattered strips, whereas that in the F25 group presented smaller fragments. More regular fibrous formation with fewer vessels and more TRAP+ MNGCs was found in SF25, whereas vascularized fibrous formation with fewer TRAP+ MNGCs was found in the F25 group (Figure 8(c,f,h,j)). Figure 8.Effects of SF25 on osteogenesis, degradation, and integration properties at the calvarial defect at 4 weeks following surgery. ( a) Micro-CT, (b) H&E staining, and (c) α-SMA staining showed newly formed bone formation (yellow asterisks) and blood vessels (yellow arrows) in SF25 compared with F25. ( d) Micro-CT, (e) H&E staining, and (f) TRAP staining revealed the residual membrane (blue asterisks), the presence of TRAP+ MNGCs (blue arrows), and the integration property of SF25 compared with those of the F25. Semi-quantitative analysis of (g) new bone volume, (h) the number of vessels, (i) the percentage of residual membrane area, and (j) the number of MNGCs. * Significant differences compared with the blank or F25 groups. *, $P \leq 0.05$, **, $P \leq 0.01$, ***, $P \leq 0.001$, ****, $P \leq 0.0001$, data without * were not significantly different (NS). Error bars represented mean ± SD.
## Discussion
In the present study, we developed three-phase barrier membranes using fluoride-modified PILP. Three kinds of membranes mediated diverse biological outcomes, which was closely related to ACP-related immunomodulation. The original BCM in the organic phase induced the mildest inflammatory response and degraded mostly after 4 weeks. The organic-inorganic mixture phase of the collagen membrane (F-mBCM) evoked a prominent inflammatory response owing to the F-ACP nanoparticles, resulting in fibrous encapsulation and active angiogenesis, whereas the inorganic phase (SF-mBCM) induced a mild inflammatory response and degraded the least owing to the transition of F-ACP particles into calcium phosphate with low solubility and high crystallinity during sintering.
## Fluoride-modified PILP-induced organic-inorganic phase transition of the collagen membrane
The transition from the organic to the inorganic phase via the fluoride-modified PILP method involved two processes. In the BCM stage, the membrane was composed of organic matter. After the PILP process, the membrane was composed of both organic and inorganic matter. After sintering, all organic matter disappeared. These results were confirmed by TG analysis (Figure 6(f)). In contrast, inorganic substances were generated after PILP, with a mixture of both amorphous and crystalline phases (Figures 2(b) and 5(e)). After sintering, increased crystallinity was observed using TEM and XRD (Figure 6(b)). Overall, the composition of the organic substances changed from $100\%$ to $0\%$, whereas the inorganic matter content increased from $0\%$ to $100\%$. BCM in an organic phase, F-mBCM with an organic-inorganic mixture phase, and SF-mBCM in an inorganic phase were successfully synthesized through PILP combined with a sintering strategy.
F-ACP nanoparticles were observed for the first time in the present fluoride-modified PILP system (Figure 5(c–e)), mediating subsequent intrafibrillar and extrafibrillar mineralization (Figure 2(a,c)). The formation of F-ACP proved that the process of fluorination simultaneously occurred with mineralization, and the degree of fluorination could affect the mineralization behavior and outcome. As fluoride concentration increased, the degree of extrafibrillar mineralization also increased (Figure 1(a,b)). More crystals were observed around the collagen fibers (Figure 2(c)). This was in accordance with the findings of Saxena et al. [ 15]. This mechanism may be attributed to the lower solubility of FHA compared with that of HA. Inadequate polymers in the mineral solution were not able to stabilize amorphous precursors from entering collagen fibrils, eventually leading to a higher degree of extrafibrillar mineralization.
## Phase transition induced diverse degradation behavior via ACP control
To determine the degradation behavior of F-mBCM and SF-mBCM, in vivo experiments were conducted. The results showed that the residuals of F-mBCM and SF25 were higher than those of BCM 4 weeks after surgery, which indicated that the PILP strategy could enhance barrier function (Figures 4(a) and 8(i)). Further analysis showed that, although both F-mBCM and SF25 improved the anti-degradation performance, the integrity of the SF25 membrane was significantly higher than that of F-mBCM (Figure 8(i)). The underlying mechanisms may be closely related to the F-ACP nanoparticles (Figure 5).
ACP can induce aseptic inflammation by mediating macrophage polarization [23]. It could change the morphology of macrophages as well as polarize macrophages into the M1 type, which weakens the osteogenic ability of BMSC. In response to different stimuli, primary macrophages can polarize to a pro-inflammatory M1 phenotype or anti-inflammatory wound-healing M2 phenotype [24]. In the early stages of the inflammatory phase, the inflammatory M1 phenotype is responsible for phagocytic fragmentation, whereas the late M1-to-M2-type transition promotes BMSC osteogenesis and biomineralization [25]. In the F-mBCM group, the expressions of TNF-α, IL-1β, IL-6, and NF-κB1 were significantly upregulated, revealing a prominent inflammatory response induced by released F-ACP (Figure 5(a)) [26,27]. This proves that macrophages were polarized to the M1 phenotype in the early stage, and further resulted in vascularized fibrous encapsulation and degradation (Figures 3 and 4). On the other hand, after the transition of F-ACP particles into calcium phosphate with low solubility and high crystallinity during sintering, the expression of M1-type related cytokines (TNF-α, IL-1β, IL-6, and NF-κB1) was significantly down-regulated in the SF25 group (Figure 7(a)) along with the number of monocytes around the membrane (Figure 7(b)). As a result, a low level of initial immune response avoided the induction of severe vascularized fibrous formation thus optimizing the biocompatibility of the membrane, leading to enhanced barrier performance (Figure 8(d,e,i)). These results collectively indicate that the control of F-ACP plays a central role in mediating diverse inflammatory responses as well as the degradation behavior of the PILP-based membrane.
It should be pointed out that the main observation time points of the in vivo model used in this study are 3 days and 4 weeks. However, the biogenic collagen membrane may change the cell composition at other time points and influence the cell state of bone regeneration, such as $\frac{6}{8}$ weeks, etc [21,28]. Therefore, further sample evaluation at more time points is still needed to reflect the degradation performance of materials.
## Osteogenesis mechanism of fluoride-modified biomineralized collagen membrane
Comparable new bone formation was observed in both F25 and SF25 groups with the BCM group. It should be noted that massive vascularized fibrous encapsulation was accompanied by new bone tissue in the F-mBCM group (Figure 4(b,c)), which may be related to the upregulation of TGF-β1, MMP9, and pro-inflammatory cytokines (Figure 5(a)). The expression of TGF-β1 can lead to the formation of fibrous and scar tissue through the activation of the classic TGF-β-Smad-MMP signaling pathway [29,30]. MMP9 can degrade the extracellular matrix by digesting solubilized collagen I and III monomers [31]. Upregulation of pro-inflammatory cytokines can also induce pre-osteoblasts to differentiate towards a fibroblast phenotype [32] (Figure 5(a)), leading to fibrous encapsulation. In the SF-mBCM group, the expression of TGF-β1, MMP9, and pro-inflammatory cytokines (TNF-α, IL-1β, and IL-6) were all significantly downregulated (Figure 7(a)), along with reduced angiogenesis and fibrous encapsulation in vivo (Figure 8(c,e)). Additionally, MCSF expression was downregulated. MCSF can induce osteoclastogenesis by binding to the colony-stimulating factor-1 receptor on macrophages and promoting osteoclast differentiation [33]. Therefore, the downregulation of MCSF also promoted bone formation in the SF-mBCM group (Figure 7(a)).
Although fluoride ions promote the osteogenesis performance compared with the blank group to some extent, there is no significant difference between the F-mBCM group and the conventional BCM group, which may be due to the lack of scaffolds with osteoconduction in the bone defect area, such as bone substitute, and the insufficient observation time. Some studies have achieved good osteogenic effects by adding bone substitutes along with barrier membranes [34], and some studies plugged the membrane into the defect as an osteoconduction scaffold itself [35]. Therefore, the use of membrane alone without placing any bone substitute materials to promote osteogenesis still needs further research in the current study model.
## Implications of the novel apatite-based barrier membranes
Generally, three strategies were used to prolong the degradation of a collagen membrane. One is to increase the crosslinking or mineralization degree [21], the other is to modulate the phagocytic function of macrophages and foreign body giant cells [36], and the third is to reduce the immunorecognition and related foreign body reaction of the membrane [37]. The current biomimetic mineralization strategy aims to increase the mineralization degree of the membrane. The increase in the collagen membrane’s overall mineralization significantly prolonged the membrane’s degradation process (Figure 4). However, obvious vascularized fibrous encapsulation was noticed, suggesting that excess inflammation existed. Such inflammation is closely related to the inflammatory effect of the release of ACP from biomineralized membrane [38], which is confirmed by TEM (Figure 5). Therefore, the simple biomineralization modification strategy is to some extent contrary to the concept of immunomodulation optimization. Effective control of ACP is key to building novel apatite-based barrier membranes. Future research will focus on the use of preparing organic-coated biomineralized membrane to control the release of ACP and optimize the immune microenvironment. Autologous blood or platelet-rich fibrin (PRF) is an ideal natural wrapping substance with superior biocompatibility. PRF exhibits anti-inflammatory activity and shifts macrophage polarization from the M1 to M2 phenotype [39]. In addition, PRF can decrease the inflammatory response in mesenchymal cells [40]. Cross-linking agents have also shown great potential for regulating target molecule release [41,42], indicating a possible solution for ACP control. These methods may create a favorable environment for tissue regeneration by controlling ACP release.
High-temperature sintering is also an effective way to keep ACP from being released via an amorphous-to-crystalline phase transition. Heat treatment is a common method for stabilizing ACP and improving the crystallinity of HA [43], which affects biological reactions, such as the adsorption of proteins and osteoblast precursor cells [44,45]. We further prepared a completely inorganic phase barrier membrane via the sintering method. By converting ACP into stable crystals, we downregulated the level of membrane inflammation, optimized the immune microenvironment, reduced the degree of vascularized fibrous encapsulation, and further improved the barrier performance. One of the main limitations of heat treatment is that it causes the collagen membrane to solely retain the rigid inorganic scaffold, thereby losing plasticity. Therefore, the sintering strategy has transformation value in non-degradable barrier membranes and personalized barrier stent substitutes.
## Conclusions
F-mBCM and SF-mBCM were prepared to examine the modified degradation behavior of the apatite-based barrier membrane. F-mBCM provoked a strong inflammatory response leading to vascularized fibrous encapsulation owing to the presence of F-ACP nanoparticles, whereas SF-mBCM induced a mild inflammatory response and significantly enhanced barrier performance via eliminating ACP particles. Effective control of ACP is key to building novel apatite-based barrier membranes. A new generation of collagen membranes should be further optimized via ACP control while promoting osteogenic properties.
## Disclosure statement
No potential conflict of interest was reported by the author(s).
## Data availability statement
The data that support the findings of this study are available from the corresponding authors, SW, ZC, upon reasonable request.
## Supplemental data
Supplemental data for this article can be accessed online at https://doi.org/$\frac{10.1080}{14686996.2023.2186690.}$
## References
1. Retzepi M, Donos N.. **Guided bone regeneration: biological principle and therapeutic applications**. *Clin Oral Implants Res* (2010.0) **21** 567-16. PMID: 20666785
2. Schlegel AK, Möhler H, Busch F. **Preclinical and clinical studies of a collagen membrane (Bio-Gide)**. *Biomaterials* (1997.0) **18** 535-538. PMID: 9105592
3. Lee SW, Kim SG. **Membranes for the guided bone regeneration**. *Maxillofac Plast Reconstr Surg* (2014.0) **36** 239-246. PMID: 27489841
4. Barnes CP, Pemble CW, Brand DD. **Cross-linking electrospun type II collagen tissue engineering scaffolds with carbodiimide in ethanol**. *Tissue Eng* (2007.0) **13** 1593-1605. PMID: 17523878
5. Chia-Lai PJ, Orlowska A, Al-Maawi S. **Sugar-based collagen membrane cross-linking increases barrier capacity of membranes**. *Clin Oral Investig* (2018.0) **22** 1851-1863
6. Veríssimo DM, Leitão RF, Ribeiro RA. **Polyanionic collagen membranes for guided tissue regeneration: effect of progressive glutaraldehyde cross-linking on biocompatibility and degradation**. *Acta Biomater* (2010.0) **6** 4011-4018. PMID: 20433958
7. Yu X, Sun H, Yang J. **Evaluation of bone-regeneration effects and ectopic osteogenesis of collagen membrane chemically conjugated with stromal cell-derived factor-1 in vivo**. *Biomed Mater* **15** 015009. DOI: 10.1088/1748-605X/ab52da
8. Chu C, Deng J, Liu L. **Nanoparticles combined with growth factors: recent progress and applications**. *RSC Adv* (2016.0) **6** 90856-90872. DOI: 10.1039/C6RA13636B
9. Spiller KL, Nassiri S, Witherel CE. **Sequential delivery of immunomodulatory cytokines to facilitate the M1-to-M2 transition of macrophages and enhance vascularization of bone scaffolds**. *Biomaterials* (2015.0) **37** 194-207. PMID: 25453950
10. Chu C, Deng J, Sun X. **Collagen membrane and immune response in guided bone regeneration: recent progress and perspectives**. *Tissue Eng Part B Rev* (2017.0) **23** 421-435. PMID: 28372518
11. Shi M, Chen Z, Farnaghi S. **Copper-doped mesoporous silica nanospheres, a promising immunomodulatory agent for inducing osteogenesis**. *Acta Biomater* (2016.0) **30** 334-344. PMID: 26596565
12. Ayna M, Açil Y, Gulses A. **Fate of a bovine-derived xenograft in maxillary sinus floor elevation after 14 years: histologic and radiologic analysis**. *Int J Periodontics Restorative Dent* (2015.0) **35** 541-547. PMID: 26133144
13. Olszta MJ, Douglas EP, Gower LB. **Scanning electron microscopic analysis of the mineralization of type I collagen via a polymer-induced liquid-precursor (PILP) process**. *Calcif Tissue Int* (2003.0) **72** 583-591. PMID: 12616327
14. Antebi B, Cheng X, Harris JN. **Biomimetic collagen-hydroxyapatite composite fabricated via a novel perfusion-flow mineralization technique**. *Tissue Eng Part C Methods* (2013.0) **19** 487-496. PMID: 23157544
15. Saxena N, Cremer MA, Dolling ES. **Influence of fluoride on the mineralization of collagen via the polymer-induced liquid-precursor (PILP) process**. *Dent Mater* (2018.0) **34** 1378-1390. PMID: 29935767
16. Liu Y, Liu C, Huang R. **Effects of fluorinated porcine hydroxyapatite on lateral ridge augmentation: an experimental study in the canine mandible**. *Am J Transl Res* (2020.0) **12** 2473-2487. PMID: 32655785
17. Wu S, Xia B, Mai S. **Sodium fluoride under dose range of 2.4–24 μM, a promising osteoimmunomodulatory agent for vascularized bone formation**. *ACS Biomater Sci Eng* **5** 817-830. DOI: 10.1021/acsbiomaterials.8b00570
18. Liu R, Qiao W, Huang B. **Fluorination enhances the osteogenic capacity of porcine hydroxyapatite**. *Tissue Eng Part A* (2018.0) **24** 1207-1217. PMID: 29376480
19. Qiao W, Liu Q, Li Z. **Changes in physicochemical and biological properties of porcine bone derived hydroxyapatite induced by the incorporation of fluoride**. *Sci Technol Adv Mater* (2017.0) **18** 110-121. DOI: 10.1080/14686996.2016.1263140
20. Shi YL, Zhi-Peng LI, Liu CW. **Study on physicochemical property and biocompatibility in vitro of biogenic collagen membrane**. (2018.0)
21. Wu Y, Chen S, Luo P. **Optimizing the bio-degradability and biocompatibility of a biogenic collagen membrane through cross-linking and zinc-doped hydroxyapatite**. *Acta Biomater* (2022.0) **143** 159-172. DOI: 10.1016/j.actbio.2022.02.004
22. Gower LB, Odom DJ. **Deposition of calcium carbonate films by a polymer-induced liquid-precursor (PILP) process**. *J Crystal Growth* (2000.0) **210** 719-734. DOI: 10.1016/S0022-0248(99)00749-6
23. Edwards FC, Taheri A, Dann SC. **Characterization of cytolytic neutrophil activation in vitro by amorphous hydrated calcium phosphate as a model of biomaterial inflammation**. *J Biomed Mater Res Part A* **96** 552-565. DOI: 10.1002/jbm.a.32937
24. Murray PJ, Allen JE, Biswas SK. **Macrophage activation and polarization: nomenclature and experimental guidelines**. *Immunity* **41** 14-20. DOI: 10.1016/j.immuni.2014.06.008
25. Zhang Y, Böse T, Unger RE. **Macrophage type modulates osteogenic differentiation of adipose tissue MSCs**. *Cell Tissue Res* (2017.0) **369** 273-286. PMID: 28361303
26. Cao S, Zhang X, Edwards JP. **NF-kappaB1 (p50) homodimers differentially regulate pro- and anti-inflammatory cytokines in macrophages**. *J Biol Chem* **281** 26041-26050. DOI: 10.1074/jbc.M602222200
27. Ghosh S, May MJ, Kopp EB. **NF-κB and REL PROTEINS: evolutionarily conserved mediators of immune responses**. *Annu Rev Immunol* (1998.0) **16** 225-260. PMID: 9597130
28. Jin S, Yang R, Chu C. **Topological structure of electrospun membrane regulates immune response, angiogenesis and bone regeneration**. *Acta Biomater* **129** 148-158. DOI: 10.1016/j.actbio.2021.05.042
29. Kim KK, Sheppard D, Chapman HA. **TGF-β1 signaling and tissue fibrosis**. *Cold Spring Harbor Perspect Biol* **10** a022293
30. Frangogiannis N. **Transforming growth factor-β in tissue fibrosis**. *J Exp Med* **217** e20190103. PMID: 32997468
31. Van Doren SR. **Matrix metalloproteinase interactions with collagen and elastin**. *Matrix Biol* (2015.0) **44–46** 224-231
32. Champagne CM, Takebe J, Offenbacher S. **Macrophage cell lines produce osteoinductive signals that include bone morphogenetic protein-2**. *Bone* (2002.0) **30** 26-31. PMID: 11792561
33. Yamashita M, Otsuka F, Mukai T. **Simvastatin inhibits osteoclast differentiation induced by bone morphogenetic protein-2 and RANKL through regulating MAPK, AKT and Src signaling**. *Regul Pept* **162** 99-108. DOI: 10.1016/j.regpep.2010.03.003
34. Cai Y, Guo J, Chen C. **Silk fibroin membrane used for guided bone tissue regeneration**. *Mater Sci Eng C Mater Biol Appl* **70** 148-154. DOI: 10.1016/j.msec.2016.08.070
35. Zhang J, Ma S, Liu Z. **Guided bone regeneration with asymmetric collagen-chitosan membranes containing aspirin-loaded chitosan nanoparticles**. *Int J Nanomedicine* (2017.0) **12** 8855-8866. PMID: 29276386
36. Fang J, Liu R, Chen S. **Tuning the immune reaction to manipulate the cell-mediated degradation of a collagen barrier membrane**. *Acta Biomater* (2020.0) **109** 95-108. PMID: 32268238
37. Eliezer M, Sculean A, Miron RJ. **Cross-linked hyaluronic acid slows down collagen membrane resorption in diabetic rats through reducing the number of macrophages**. *Clin Oral Investig* (2022.0) **26** 2401-2411
38. Chen L, Qiao P, Liu H. **Amorphous calcium phosphate NPs mediate the macrophage response and modulate BMSC osteogenesis**. *Inflammation* (2021.0) **44** 278-296. PMID: 32939669
39. Nasirzade J, Kargarpour Z, Hasannia S. **Platelet-rich fibrin elicits an anti-inflammatory response in macrophages in vitro**. *J Periodontol* (2020.0) **91** 244-252. PMID: 31376159
40. Kargarpour Z, Nasirzade J, Panahipour L. **Platelet-rich fibrin decreases the inflammatory response of mesenchymal cells**. *Int J Mol Sci* **22** 11333. DOI: 10.3390/ijms222111333
41. Solorio L, Zwolinski C, Lund AW. **Gelatin microspheres crosslinked with genipin for local delivery of growth factors**. *J Tissue Eng Regen Med* (2010.0) **4** 514-523. PMID: 20872738
42. Liu T, Dan W, Dan N. **A novel grapheme oxide-modified collagen-chitosan bio-film for controlled growth factor release in wound healing applications**. *Mater Sci Eng C Mater Biol Appl* **77** 202-211. DOI: 10.1016/j.msec.2017.03.256
43. Liu X, Zhao M, Lu J. **Cell responses to two kinds of nanohydroxyapatite with different sizes and crystallinities**. *Int J Nanomedicine* (2012.0) **7** 1239-1250. PMID: 22419871
44. Yang Y, Dennison D, Ong JL. **Protein adsorption and osteoblast precursor cell attachment to hydroxyapatite of different crystallinities**. *Int J Oral Maxillofac Implants* (2005.0) **20** 187-192. PMID: 15839111
45. Lee WH, Zavgorodniy AV, Loo CY. **Synthesis and characterization of hydroxyapatite with different crystallinity: effects on protein adsorption and release**. *J Biomed Mater Res Part A* (2012.0) **100** 1539-1549
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.