text
stringlengths 1
6.27k
| id
int64 0
3.07M
| raw_id
stringlengths 2
9
| shard_id
int64 0
0
| num_shards
int64 16
16
|
|---|---|---|---|---|
Levinas [77], where responsibility should be a prerogative in the face of the other affected by vulnerability and suffering. Study Limitations There are certain limitations to this study that must be noted. First, the findings, such as those of previous qualitative studies, cannot be generalized to the target population. However, there are broad, phenomenological insights into how individuals with COPD perceive the important, universal and recurring features of their experience of dignity. Second, the participants in this study were chosen using a purposive selection method. The number of participants was deemed sufficient to achieve variance and maintain depth in the study. After 17 interviews, data saturation was reached; nonetheless, three extra interviews were conducted to ensure that further coding was not possible and that no new themes were identified. Although results were based on a single interview per patient, the material was sufficiently substantial for the researchers to reach a shared interpretation of the data. In addition, several strategies were used by the researchers to reduce the possibility of biased judgements and idiosyncratic interpretations. Further research should investigate the influence of culture on dignity experiences in other groups of advanced ill patients to increase the transferability of findings. The recruitment process produced a sample of participants who were not receiving palliative care. This could imply a lesser understanding of the role palliative care may play in advanced COPD as compared to that from a more diverse sample of respondents. Furthermore, specific participant characteristics were not evaluated, such as economic status or religious beliefs. Restrictions associated with | 3,069,900 | 254900742 | 0 | 16 |
the COVID-19 pandemic hampered both the recruitment and interviewing processes. Nonetheless, with the support of partners and additional precautions, the project persisted, and the findings were not affected by the pandemic conditions. Implications for Practice The findings of this study have important clinical and research implications. Assisting patients with advanced chronic illness to live and believe in their ability to live while ill is a critical component of comprehensive and integrative care. Lifeworld existentials provide researchers with a method of inquiry that is congruent with humanistic practices and that acknowledges the uniqueness of the individual within the world [78]. Confronting existential concerns about life and death, which can have positive effects, should be included in professional and informal support systems. In addition, access to social and healthcare services, which presently tend to be fragmented, needs to be addressed. We encourage healthcare providers to conduct these discussions with patients and families in order to better foresee their individual needs. Further research should concentrate on developing, implementing and evaluating supportive interventions aimed at improving people's psychosocial and spiritual well-being as they approach death. Given that many factors impact the sense of dignity, making care plans informed by a phenomenological stance would help defend the dignity of patients (and their families), thereby ensuring a higher quality of life. Additional qualitative research is needed to better understand the significance and value of dignity as a health resource in people's lives, and how boosting dignity might help individuals feel better. Finally, we must improve our understanding of the provision of palliative | 3,069,901 | 254900742 | 0 | 16 |
care in advanced COPD and how these services can be implemented successfully. Conclusions To summarize, this study provided an innovative contribution to our understanding of the relevance of dignity perception in the lives of advanced COPD patients. Patients needed to adopt this principle in their lives which will have a positive impact on their well-being. Our findings illustrate a set of holistic issues that had an impact on participants' dignity needs, including four closely intertwined constituents: "Lived body-balancing between sick body and willingness to continue"; "Lived relations-balancing between self-control and belongingness"; "Lived Time-balancing between past, present and a limited future"; and "Lived space-balancing between safe places and non-compassionate places". This research showed that exploring the notion of dignity via patient discourses may help acquire knowledge about dignity from an inner viewpoint, a knowledge that healthcare professionals and educators should integrate into their clinical and educational practices. Because dignity is frequently endangered when patients are very sick, professionals must support and strengthen patients to live and relieve their suffering so they can be themselves and feel dignity. Similarly, as revealed in this study, it is critical to help patients discover their inner resources that develop and promote dignity, including elements such as a "purpose in life" and "love as a restorative energy". More research is needed to investigate innovative approaches to manage complex care in advanced COPD and to clarify how palliative care can fit into this complex care network. The flux of demands in COPD, as illustrated in this phenomenological study, requires service support and flexibility | 3,069,902 | 254900742 | 0 | 16 |
so that both professionals and patients can adjust to the disease's unanticipated yet rising demands over time. This requires an interdisciplinary approach, incorporating health and social fields. Informed Consent Statement: Informed consent was obtained from all subjects involved in the study. Participation in the study was completely voluntary and anonymous. Implicit consent for the project was assumed when study participants completed the survey. Participants received no compensation. Data Availability Statement: All data generated or analysed during this study are included in this article. | 3,069,903 | 254900742 | 0 | 16 |
Strategies of elite Chinese gymnasts in coping with landing impact from backward somersault This study aimed to investigate how elite Chinese gymnasts manage the landing impact from a backward somersault. Six international-level male gymnasts performed backward somersault tests with a synchronous collection of kinematics (250 Hz), ground reaction forces (1,000 Hz), and surface electromyography (EMG) (2,000 Hz). A 19-segment human model was developed and lower extremity joints torques were calculated by means of a computer simulation. The angles of the lower extremity joints initially extended and then flexed. These angular velocities of extension continued to decrease and the joint torques changed from extensor to flexor within 100 ms before touchdown. The angles of the hips, knees, and ankles flexed rapidly by 12°, 36°, and 29°, respectively, and the angular velocities of flexion, flexor torque, and EMG peaked sharply during the initial impact phase of the landing. The angles of the hips, knees, and ankles flexed at approximately 90°, 100°, and 80°, respectively. The torques were reversed with the extensor torques, showing a relatively high level of muscle activation during the terminal impact phase of the landing. The results showed that the international-level gymnasts first extended their lower extremity joints, then flexed just before touchdown. They continued flexing actively and rapidly in the initial impact phase and then extended to resist the landing impact and maintain body posture during the terminal impact phase of the landing. The information gained from this study could improve our understanding of the landings of elite gymnasts and assist in injury prevention. | 3,069,904 | 204940816 | 0 | 16 |
INTRODUCTION Gymnastics is a popular sport with 50 million participants worldwide (Slater et al., 2015). Each gymnastics routine ends with a landing, and successful landings (without taking a step or falling) are a key factor for motion evaluation. However, there is a high incidence of injury reported in gymnasts while landing, especially to the lower extremity joints (Westermann et al., 2015). A better understanding of the potential injury mechanisms could help prevent certain injuries and change the design of the training schemes, thereby improving performance. Gymnasts were reported to bear high frequency (over 200 times a week) landing impact loads (Gittoes & Irwin, 2012), with the peak vertical ground reaction force (vGRF) reaching 7.1-15.8 times the athlete's body weight (BW) (Slater et al., 2015). This repetition and large GRF has caused a high rate of lower extremity injuries in gymnasts (Mills, Pain & Yeadon, 2009). It has also been suggested that there is a correlation between the high injury incidence and the excessive load to the lower extremities of gymnasts (Daly, Bass & Finch, 2001;Wade et al., 2012). Furthermore, the Code of the International Gymnastics Federation (FIG) requires that there be no excessive knee flexion during landing from gymnasts for aesthetic purposes (FIG, 2017). Several studies demonstrated that gymnasts produced greater peak vGRF than recreational athletes in drop landings, which are considered the stiff lower extremity landing techniques, with knee flexion less than 90 (Christoforidou et al., 2017;Devita & Skelly, 1992;Seegmiller & McCaw, 2003). More specifically, this landing pattern increased leg stiffness and is a potential | 3,069,905 | 204940816 | 0 | 16 |
contributing factor for injury (Butler, Crowell & Davis, 2003). Bradshaw & Hume (2012) investigated the changes in the landings of two gymnasts over a period of 8 years. Both gymnasts showed an increased ankle plantar-flexion stiffness by 10.8 and 13.9 kN/m, respectively, with both gymnasts reporting severe pain in one or both heels over this period of time. Furthermore, other contributing factors to injuries in these gymnasts, including internal factors such as anatomical differences, neuromuscular function, strength, and leg stiffness, were observed. External factors, such as the landing of complex tasks, exposure time, the training environment, and varied competition were considered (Bradshaw & Hume, 2012). Few studies have attempted to explore strategies for reducing the impact force and the incidence of injury. One study found that increasing the flexion of the lower extremity joints during landing could effectively reduce the impact load (Slater et al., 2015). However, this may lead to compensatory muscle and ligament injuries, thus affecting the landing stability, especially during a high-speed landing impact (Bradshaw & Hume, 2012;Tant, Wilkerson & Browder, 1989). Modifying the material composition of the gymnastics landing mat may reduce the vGRF, but increase the internal load (muscle forces and joint reaction forces) to the lower extremity joints (Mills, Pain & Yeadon, 2009) and the potential for subtalar and ankle instability during the landing (McNitt-Gray, 2000). It is worth noting that previous studies generally analyzed the whole impact phase of the landing, defined from the initial ground contact to the maximal knee flexion (Christoforidou et al., 2017), the maximal descending | 3,069,906 | 204940816 | 0 | 16 |
height of the center of mass (Caine et al., 2003), or the local minima in the vGRF (McNitt-Gray et al., 2001). Participants in these studies were collegiate students or young athletes who did not participate in international-or national-level gymnastics competitions. Therefore, it is unclear what strategies elite international-level gymnasts used and whether they adapted initial (from the touchdown to the peak vGRF) and terminal impact-phase (from the peak vGRF return to their BW) strategies for the flexion/extension of the lower extremity joints before the touchdown of the landing. Moreover, the lower extremity joints approach full extension and then flexion just before touchdown during the flight phase of drop landings (McNitt-Gray, Yokoi & Millward, 1993). However, it is not clear whether the same strategies would be seen before the touchdown of landings from other completed tasks in gymnastics. Using an in-vivo implanted sensor to test the internal load of the human lower extremity would be useful but there are ethical limitations to this methodology. A computer simulation of the human body, however, provides a practical approach to explore the characteristics of body motion and has been widely used to analyze body movement in humans (Pandy, 2001). The backward somersault (BS) is one of the most basic and common movements for gymnasts for developing difficulty movement and combined motion, and is used frequently in gymnastics training and competitions. The aim of this study was to investigate how elite gymnasts manage the landing impact from a BS and achieve a safe, aesthetic, and stable performance. It is our hypothesis | 3,069,907 | 204940816 | 0 | 16 |
that elite gymnasts utilize varied flexion/extension strategies to control their lower extremity joints during the different landing phases (before touchdown, initial, and terminal impact-phases of the landing) of the BS. Therefore, a study of the basic movements would help us understand the characteristics and injury mechanisms of the more complex gymnastics landing of the same type. Participants Six international-level male gymnasts from the Chinese national team competing in World Cups and/or Championships, with no musculoskeletal injuries for at least 6 months prior, participated in the study (mean ± standard deviation (SD) age: 17.3 ± 1.3 years, height: 165.7 ± 5.0 cm, body mass: 57.3 ± 3.9 kg). All participants were familiarized with the procedures in advance and informed consents were signed. The study was approved by the Ethical Advisory Committee of the China Institute of Sport Science in accordance with the regulations set forth by the Declaration of Helsinki. Procedure The experiment was conducted in the biomechanics laboratory of the China Institute of Sport Science. A 9-camera Qualisys Oqus motion system (250 Hz, Gothenburg, Sweden) was used to capture the 3D motion data. The standard reflective markers (diameter: 16 mm) were placed at the head, cervical vertebrae (CV7), scapula-inferior angle, thoracic vertebrae (TV10), shoulder, elbow, wrist, anterior superior iliac spine, posterior superior iliac spine, knee, ankle, metatarsal-phalangeal joints, heel, and toes on both sides of the body (Figs. 1A-1C). The marker placements were referenced from the CAST full body marker set (Sint, 2007). A Kistler force plate (400 × 600 mm), located beneath a landing mat | 3,069,908 | 204940816 | 0 | 16 |
(five cm thick), was used to collect vGRF data (1,000 Hz) and was surrounded by an ethylene-vinyl acetate insole mat. The vGRF was reported to decrease by 5% on the force plate when the landing mats were up to 12 cm thick (McNitt-Gray et al., 2001). Surface muscle activity signals were recorded (2,000 Hz, Bagnoli 8 Desktop electromyography (EMG) System; Delsys Inc, Boston, MA, USA) from the rectus femoris (RF), biceps femoris (BF), tibialis anterior (TA), and lateral gastrocnemius (LG) on the two lower extremities of the gymnasts as per the SENIAM guidelines (Hermens et al., 1999). The Qualisys Oqus motion capture system, the Kistler force plate, and the surface EMG system were all synchronized during the data collection. The gymnasts completed an initial warm-up exercise (15 min of jogging, jumping, and stretching) and then each participant performed three successful trials of BS without taking a step or hop. The BS was initiated by jumping from the ground next to the force plate with bare feet (Fig. 2). The best trial for each participant was chosen by two national-level judges using the Code of Points of FIG for further analysis. Experiment data reduction and analysis The 3D motion data was processed using Qualisys Track Manager Software, following a 10 Hz low-pass cut-off filter (Slater et al., 2015). The joint angles with its angular velocity were calculated between two lines in space based on three-dimension trajectories. The vGRFs were filtered using a low-pass cut-off at 50 Hz (Slater et al., 2015). The peak vGRFs were normalized with | 3,069,909 | 204940816 | 0 | 16 |
the BW of each gymnast. The touchdown was identified as the first frame when vGRF exceeded 10 N (Christoforidou et al., 2017). Raw EMG signals were fully wave rectified and band-pass filtered by 10-400Hz (Van Dieën et al., 2009). The pre-activation phase (T0) was defined as the 100 ms duration before touchdown (Komi & Bosco, 1987). The EMG processing was conducted in the pre-activated and two impact phases of the landing (T1: initial impact-phase, from contact to the peak vGRF; T2: terminal impact-phase, from the peak vGRF return to their BW) (Fig. 2). The EMG signals were normalized by the peak EMG of each trial. Antagonist-agonist co-activation was calculated as TA to LG normalized EMG for ankle, and BF to RF normalized EMG for knee (Ruan & Li, 2010). All of the experimental data for the gymnasts were averaged within 20 ms intervals and then the data was reported as mean ± SD using descriptive statistics. Computer simulation and validation The experimental findings from the elite gymnasts showed a low discrete degree, with the mean standard error within six participants of 0.6 BW (peak vGRF), 2.7 ms (time to peak vGRF) and ranging from 2 to 10 in the lower extremity angles. One gymnast with intermediate experiment results was chosen for modeling and simulation. LifeMod (LifeModeler, Inc., San Clemente, CA, USA) is an advanced multi-body computer simulation software system commonly used in human movement simulation with Automatic Dynamic Analysis of Mechanical Systems (ADAMS) as the dynamics modeling engine. The GeBod database (BRG.LifeMOD TM ) was used | 3,069,910 | 204940816 | 0 | 16 |
to develop a 19-segment and 50 degree of freedom rigid-body model based on age (17 years), weight (63 kg), and height (1.68 m) data from the selected gymnast. The model consisted of the head, neck, upper torso, central torso, lower torso, scapulas, upper arms, lower arms, hands, upper legs, lower legs, and feet ( Fig. 1D) (Serveto et al., 2010). A model of the gymnastics matting with a dimension of 2 × 2 × 0.05 m (length × width × height) was developed using MSC.ADAMS (MSC Software Corp. acronym of Automated Dynamic Analysis of Mechanical Systems) software. The basis for the mechanical properties of the landing mat was obtained by an optimization algorithm. The model was then validated by Figure 2 The demonstration of backward somersault landing. T0: The pre-activation phase was defined as 100 ms preceding ground contact; T1: initial impact-phase, from the first touchdown to the peak vertical ground reaction force (vGRF); T2: terminal impact-phase, from the peak vGRF to the vGRF equaling to body weight. Full-size DOI: 10.7717/peerj.7914/ fig-2 coefficients of multiple correlations (CMC) between the simulation and actual results, with the specific algorithm described in detail in our previous study (Xiao et al., 2017). After verifying the reliability of the model, the joint torques of the hip, knee, and ankle joints were conducted using computer simulation. RESULTS The measured angles of the hip, knee, and ankle joints were first extended (mean 11 , 10 , and 6 , respectively) and then flexed during T0 in the six gymnasts (Figs. 3A-3C). The | 3,069,911 | 204940816 | 0 | 16 |
corresponding joint angles rapidly flexed by 2 , 36 , and 29 during T1 and maintained at around 90 , 100 , and 80 during T2, respectively. The angular velocities of extension continued to decrease during T0, and the angular velocities of flexion reached their peaks during T1 and gradually approached zero during T2 (Figs. 4A-4C). The eight muscles in the bilateral lower extremity were pre-activated during T0 (Figs. 5A-5D). The EMG amplitude of most muscles increased from T0 to T1, and reached their maximum near the peak vGRF. They still maintained a high-level of activation during T2. The CMC between the measured lower extremity joint angles and the simulation result were calculated for model validation, with the left knee (CMC = 0.95), right knee (CMC = 0.93), and left and right ankles (CMC = 0.85) (Figs. 6A and 6B). The CMCs Figure 3 Average integrated (20 ms bins) angles of lower extremity joints (A-C) and vertical ground reaction force (vGRF) (D) during backward somersault landing (n = 6). The whole landing process was divided into three phases by solid vertical lines. T0: The pre-activation phase was defined as 100 ms preceding ground contact; T1: initial impact-phase, from the first touchdown to the peak vGRF; T2: terminal impact-phase, from the peak vGRF to the vGRF equaling to body weight (Mean: solid vertical line, standard deviation: dotted vertical line). Full-size DOI: 10.7717/peerj.7914/ fig-3 greater than 0.75 indicated good correlations (Collins et al., 2009). The difference between the simulated (11.9 BW) and measured peak vGRF (12.5 BW) | 3,069,912 | 204940816 | 0 | 16 |
was 4.6%, which was less than 10% and was considered to be an accurate representation (King, Wilson & Yeadon, 2006). Therefore, these results could be used to validate the model. The angles and angular velocities of the lower extremity joints were consistent with the results of Figs. 3A-3C and 4A-4C, respectively (Figs. 6A-6D). The torques of the lower extremity joints were initially dominated by extensor during T0 (plantar flexor for the ankle) (Figs. 6E and 6F), and reversed to flexor, reaching their maximum (dorsiflexion for the ankle) during T1. The torques quickly reversed again to the extensor and reached their peak during T2 (plantar flexor for the ankle). The torque of the knee joints reached their maximum and in the ankle joints the torques remained slight. DISCUSSION From the best of our knowledge, this is the first investigation to reveal the flexion/extension strategies of the lower extremity joints from flight to the initial and terminal impact-phases of BS landings in elite gymnasts. The study quantified the kinematics, kinetics, and muscle activation characteristics of each phase of the BS in elite gymnasts. The findings could enhance our understanding of the gymnastics landing. The lower extremity joints of the gymnasts first extended and then flexed actively during the preparation phase for the touchdown (T0). Increased angles of the lower extremity joints were conducive to the body's extension, which may contribute to increasing the moment of inertia of the body. As the moment of inertia increases, the gymnasts can reduce their angular velocity in preparation for touchdown (McNitt-Gray, 2000). | 3,069,913 | 204940816 | 0 | 16 |
It was noteworthy that the angular velocities of the knee and ankle joints changed from extension to flexion before touchdown. The lower extremity joints actively flexed, which was first seen in the ankle joint, followed by the knee and hip joints. Gittoes et al. (2011) suggested that the female gymnasts' knees and ankles were also flexing at the touchdown of the backward rotating pike and tuck dismounts, but there was no information about whether the joints flexed before touchdown. Furthermore, the antagonist and agonistic muscles of the lower extremity joints were pre-activated, which may play an important role in regulating the lower extremity stiffness (Christoforidou et al., 2017). Muscle contraction is an essential factor for producing lower extremity joint torque in the flight before landing. The lower extremity joints experienced the process from extension to flexion because these torques changed from extensor to flexor. Previous studies have focused on the torque of the lower extremity joints after touchdown (McNitt-Gray et al., 2001;Figure 5 Average integrated (20 ms bins) coactivation of root mean square of normalized EMG (EMG RMS ) (n = 6). Positive EMG RMS means antagonist of the joint, and negative EMG RMS means the agonist of the joint. The EMGs were normalized to the peak EMG during the landing phase. RF, rectus femoris; BF, biceps femoris; TA, tibialis anterior; LG, lateral gastrocnemius. The coactivation was defined as the TA to LG EMG for the ankle (A and C), and the BF to RF EMG for the knee (B and D). The whole landing process | 3,069,914 | 204940816 | 0 | 16 |
was divided into three phases by solid vertical lines. T0: The pre-activation phase was defined as 100 ms preceding ground contact; T1: initial impact-phase, from the first touchdown to the peak vertical ground reaction force (vGRF); T2: terminal impact-phase, from the peak vGRF to the vGRF equaling to body weight (Mean: solid vertical line, standard deviation: dotted vertical line). Full-size DOI: 10.7717/peerj.7914/fig-5 Figure 6 The angles (A and B), angular velocities (C and D) and torques (E and F) of lower extremity joints for one of the gymnasts during backward somersault landing (n = 1). The skeleton models showed four body postures of the landing (100 ms prior touchdown, touchdown, peak vertical ground reaction force (vGRF), and vGRF equal to body weight, respectively). The whole landing process was divided into three phases by solid vertical lines. T0: The pre-activation phase was defined as 100 ms preceding ground contact; T1: initial impact-phase, from the first touchdown to the peak vGRF; T2: terminal impact-phase, from the peak vGRF to the vGRF equaling to body weight. Full-size DOI: 10.7717/peerj.7914/ fig-6 Mills, Pain & Yeadon, 2009;Verniba et al., 2017), but there is little knowledge about the torque of the joints in the pre-landing phase. Our results indicated that in preparation for the landing, the lower extremity joints of the gymnast would first be extended and then actively flexed just before touchdown. During the initial impact-phase of the landing (T1), the lower extremity joints continue flexing actively and rapidly. T1 has a very short duration (22.8 ± 6.7 ms), | 3,069,915 | 204940816 | 0 | 16 |
which was consistent with the results of a previous study (Slater et al., 2015) where the angles of the lower extremity joints flexed rapidly. Furthermore, the angular velocities of the hip joints changed from extension to flexion and the angular velocities of the knee and ankle joints further flexed and reached their peak values. It is significant to note that the peak angular velocities of the ankle joint occurred before touchdown, but the peak angular velocities of the knee joint occurred before the peak of the vGRF and the peak angular velocities of the hip joint occurred near the peak vGRF. They reached peak values successively, which may be because the ankle is the initial interface with the ground, and the hip is the proximal joint, which properly reflects the synergy of multiple joints (Gittoes et al., 2011). McNitt-Gray et al. (2001) suggested that only the flexor torques of the hip joints were generated in a short time (about 20 ms) after touchdown and the other joints of the lower extremities maintain extensor torques during the landing. However, our results confirmed that all torques of the lower extremity joints were flexor torques in T1. This difference in lower extremity joint torques might be due to the landing being performed by different levels of participants (collegiate male gymnasts in their study against the international-level gymnast in this study). Therefore, the authors speculated that the participants (international-level gymnasts) in this study would represent an altered landing control strategy to help improve performance. Additionally, the results of muscle activation | 3,069,916 | 204940816 | 0 | 16 |
showed that the EMG amplitudes of the antagonistic and agonistic muscles of the lower extremity joints were close to reaching their peak values synchronously, and therefore this regulated the lower extremity stiffness to accommodate the rapidly increased vGRF (Kramer et al., 2012). A previous study indicated that the activity levels of the lower extremity muscles positively correlated with the peak vGRF during the landing absorption phase, which contributed to absorbing the impact of the landing and preventing injury to the lower extremities (Iida et al., 2011). Therefore, in order to quickly absorb the landing impact and prevent injuries during T1, the authors suggest that the lower extremity joints of gymnasts may flex continuously and actively, while increasing the activity levels of the lower extremity muscles. During the terminal impact-phase of the landing (T2), the gymnasts in this study began to actively extend their lower extremity joints to resist the impact force and maintain body posture. The angles and angular velocities of the lower extremity joints flexed minutely. Finally, the lower extremity joints stabilized and maintained a small flexion, thus these gymnasts met the Code of Points of FIG, which requires that the joints should not be excessively flexed (FIG, 2017). Furthermore, the EMG of the lower extremity muscles remained at a high level of activation, and the extensor torques of the lower extremity joints (plantar flexion torque of ankle) increased simultaneously, reaching their peak values during T2. Therefore, accomplishing the landing task with a small flexion of the lower extremity joints may allow the gymnasts to | 3,069,917 | 204940816 | 0 | 16 |
generate the joint torques needed to compensate for the large vGRF. Appropriate muscle length may be the key to producing greater joint torques (McNitt-Gray, 2000). Marchetti et al. (2016) found that the highest overall muscle activation of the lower extremities was generated at a 90 knee joint angle during a back squat, which was close to the knee angles of this study during T2. Devita & Skelly (1992) suggested that a stiff landing in which there is less than 90 of knee flexion had larger GRFs than a soft landing; the ankle plantar flexors produced a larger torque, and the ankle muscles absorbed a greater amount of the impact forces upon landing. The ankle is more likely to be injured in this condition, and this was consistent with the findings of an epidemiological investigation (Kerr et al., 2015). The study indicated that the gymnasts first extend their lower extremity joints before touchdown, then flex actively and rapidly in anticipation of the upcoming touchdown and the initial impact phase, and again extend during the terminal impact phase of the landing. These landing strategies could effectively mitigate some of the landing impact and are conducive to injury prevention. There are a few limitations in this study. First, this study is limited to a population of international-level male gymnasts, but we acknowledge that there might be differences in lower extremity kinematics and kinetics between different genders during their landings (Haines et al., 2011). Secondly, the sample size (six international-level gymnasts) of this study is limited and the authors did | 3,069,918 | 204940816 | 0 | 16 |
not consider more gymnasts of different levels, which was the original intent in order to control variables and reveal the landing strategies primarily in international-level gymnasts. However future studies should include male and female gymnasts of different levels. CONCLUSIONS The study quantified the lower extremity kinematics, kinetics, and muscle activation of international-level gymnasts during BS landings. Gymnasts first extend their lower extremity joints to increase the moment of inertia, thus better reducing the body's angular velocity before touchdown. The lower extremity joints flex actively just before touchdown, continue flexing actively and rapidly in the initial impact phase, and then extend to resist the impact force and maintain body posture during the terminal impact phase of the landing. This is the first study to investigate altered flexion/extension strategies of the lower extremity joints during different phases of the landing in gymnastics, thereby having the potential to expand the current understanding of the landing process of gymnastics and to aid in the prevention of injuries. | 3,069,919 | 204940816 | 0 | 16 |
Evaluating the Underlying Gender Bias in Contextualized Word Embeddings Gender bias is highly impacting natural language processing applications. Word embeddings have clearly been proven both to keep and amplify gender biases that are present in current data sources. Recently, contextualized word embeddings have enhanced previous word embedding techniques by computing word vector representations dependent on the sentence they appear in. In this paper, we study the impact of this conceptual change in the word embedding computation in relation with gender bias. Our analysis includes different measures previously applied in the literature to standard word embeddings. Our findings suggest that contextualized word embeddings are less biased than standard ones even when the latter are debiased. Introduction Social biases in machine learning, in general and in natural language processing (NLP) applications in particular, are raising the alarm of the scientific community. Examples of these biases are evidences such that face recognition systems or speech recognition systems work better for white men than for ethnic minorities (Buolamwini and Gebru, 2018). Examples in the area of NLP are the case of machine translation that systems tend to ignore the coreference information in benefit of a stereotype (Font and Costa-jussà, 2019) or sentiment analysis where higher sentiment intensity prediction is biased for a particular gender (Kiritchenko and Mohammad, 2018). In this work we focus on the particular NLP area of word embeddings (Mikolov et al., 2010), which represent words in a numerical vector space. Word embeddings representation spaces are known to present geometrical phenomena mimicking relations and analogies between words (e.g. | 3,069,920 | 121125604 | 0 | 16 |
man is to woman as king is to queen). Following this property of finding relations or analogies, one popular example of gender bias is the word association between man to computer programmer as woman to homemaker (Bolukbasi et al., 2016). Pre-trained word embeddings are used in many NLP downstream tasks, such as natural language inference (NLI), machine translation (MT) or question answering (QA). Recent progress in word embedding techniques has been achieved with contextualized word embeddings (Peters et al., 2018) which provide different vector representations for the same word in different contexts. While gender bias has been studied, detected and partially addressed for standard word embeddings techniques (Bolukbasi et al., 2016;Zhao et al., 2018a;Gonen and Goldberg, 2019), it is not the case for the latest techniques of contextualized word embeddings. Only just recently, Zhao et al. (2019) present a first analysis on the topic based on the proposed methods in Bolukbasi et al. (2016). In this paper, we further analyse the presence of gender biases in contextualized word embeddings by means of the proposed methods in Gonen and Goldberg (2019). For this, in section 2 we provide an overview of the relevant work on which we build our analysis; in section 3 we state the specific request questions addressed in this work, while in section 4 we describe the experimental framework proposed to address them and in section 5 we present the obtained and discuss the results; finally, in section 6 we draw the conclusions of our work and propose some further research. Background In this | 3,069,921 | 121125604 | 0 | 16 |
section we describe the relevant NLP techniques used along the paper, including word embeddings, their debiased version and contextualized word representations. Words Embeddings Word embeddings are distributed representations in a vector space. These vectors are normally learned from large corpora and are then used in downstream tasks like NLI, MT, etc. Several approaches have been proposed to compute those vector representations, with word2vec (Mikolov et al., 2013) being one of the dominant options. Word2vec proposes two variants: continuous bag of words (CBoW) and skipgram, both consisting of a single hidden layer neural network trained on predicting a target word from its context words for CBoW, and the opposite for the skipgram variant. The outcome of word2vec is an embedding table, where a numeric vector is associated to each of the words included in the vocabulary. These vector representations, which in the end are computed on co-occurrence statistics, exhibit geometric properties resembling the semantics of the relations between words. This way, subtracting the vector representations of two related words and adding the result to a third word, results in a representation that is close to the application of the semantic relationship between the two first words to the third one. This application of analogical relationships have been used to showcase the bias present in word embeddings, with the prototypical example that when subtracting the vector representation of man from that of computer and adding it to woman, we obtain homemaker. Debiased Word Embeddings Human-generated corpora suffer from social biases. Those biases are reflected in the cooccurrence statistics, | 3,069,922 | 121125604 | 0 | 16 |
and therefore learned into word embeddings trained in those corpora, amplifying them (Bolukbasi et al., 2016;Caliskan et al., 2017). Bolukbasi et al. (2016) studied from a geometrical point of view the presence of gender bias in word embeddings. For this, they compute the subspace where the gender information concentrates by computing the principal components of the difference of vector representations of male and female gender-defining word pairs. With the gender subspace, the authors identify direct and indirect biases in profession words. Finally, they mitigate the bias by nullifying the information in the gender subspace for words that should not be associated to gender, and also equalize their distance to both elements of gender-defining word pairs. Zhao et al. (2018b) proposed an extension to GloVe embeddings (Pennington et al., 2014) where the loss function used to train the embeddings is enriched with terms that confine the gender information to a specific portion of the embedded vector. The authors refer to these pieces of information as protected attributes. Once the embeddings are trained, the gender protected attribute can be simply removed from the vector representation, therefore eliminating any gender bias present in it. The transformations proposed by both Bolukbasi et al. (2016) and Zhao et al. (2018b) are downstream task-agnostic. This fact is used in the work of Gonen and Goldberg (2019) to showcase that, while apparently the embedding information is removed, there is still gender information remaining in the vector representations. Contextualized Word Embeddings Pretrained Language Models (LM) like ULMfit (Howard and Ruder, 2018), ELMo (Peters et | 3,069,923 | 121125604 | 0 | 16 |
al., 2018), OpenAI GPT (Radford, 2018;Radford et al., 2019) and BERT (Devlin et al., 2018), proposed different neural language model architectures and made their pre-trained weights available to ease the application of transfer learning to downstream tasks, where they have pushed the state-of-the-art for several benchmarks including question answering on SQuAD, NLI, cross-lingual NLI and named identity recognition (NER). While some of these pre-trained LMs, like BERT, use subword level tokens, ELMo provides word-level representations. and Liu et al. (2019) confirmed the viability of using ELMo representations directly as features for downstream tasks without re-training the full model on the target task. Unlike word2vec vector representations, which are constant regardless of their context, ELMo representations depend on the sentence where the word appears, and therefore the full model has to be fed with each whole sentence to get the word representations. The neural architecture proposed in ELMo (Peters et al., 2018) consists of a character-level convolutional layer processing the characters of each word and creating a word representation that is then fed to a 2-layer bidirectional LSTM (Hochreiter and Schmidhuber, 1997), trained on language modeling task on a large corpus. Research questions Given the high impact of contextualized word embeddings in the area of NLP and the social consequences of having biases in such embeddings, in this work we analyse the presence of bias in these contextualized word embeddings. In particular, we focus on gender biases, and specifically on the following questions: • Do contextualized word embeddings exhibit gender bias and how does this bias compare | 3,069,924 | 121125604 | 0 | 16 |
to standard and debiased word embeddings? • Do different evaluation techniques identify bias similarly and what would be the best measure to use for gender bias detection in contextualized embeddings? To address these questions, we adapt and contrast with the evaluation measures proposed by Bolukbasi et al. (2016) and Gonen and Goldberg (2019). Experimental Framework As follows, we define the data and resources that we use for performing our experiments. The approach motivation is applying the experiments on contextualized word embeddings. We worked with the English-German news corpus from the WMT18 1 . We used the English side with 464,947 lines and 1,004,6125 tokens. To perform our analysis, we used a set of lists from previous work (Bolukbasi et al., 2016;Gonen and Goldberg, 2019). We refer to the list of definitional pairs 2 as 'Definitonal List' (e.g. shehe, girl-boy). We refer to the list of female and male professions 3 as 'Professional List' (e.g. accountant, surgeon). The 'Biased List' is the list used in the clustering experiment and it consists of biased male and female words (500 female biased tokens and 500 male biased token). This list is generated by taking the most biased words, where the bias of a word is computed by taking its projection on the gender direction ( − → he-−→ she) (e.g. breastfeeding, bridal and diet for female and hero, cigar and teammates for male). The 'Extended Biased List' is the list used in classification experiment, which contains 5000 male and female biased tokens, 2500 for each gender, generated in the | 3,069,925 | 121125604 | 0 | 16 |
same way of the Biased List 4 . A note to be considered, is that the lists we used in our experiments (and obtained from Bolukbasi et al. (2016) and Gonen and Goldberg (2019)) may contain words that are missing in our corpus and so we cannot obtain contextualized embeddings for them. Among different approaches to contextualized word embeddings (mentioned in section 2), we choose ELMo (Peters et al., 2018) as contextualized word embedding approach. The motivation for using ELMo instead of other approaches like BERT (Devlin et al., 2018) is that ELMo provides word-level representations, as opposed to BERT's subwords. This makes it possible to study the word-level semantic traits directly, without resorting to extra steps to compose word-level information from the subwords that could interfere with our analyses. Evaluation measures and results There is no standard measure for gender bias, and even less for such the recently proposed contextualized word embeddings. In this section, we adapt gender bias measures for word embedding methods from previous work (Bolukbasi et al., 2016) and (Gonen and Goldberg, 2019) to be applicable to contextualized word embeddings. We start by computing the gender subspace from the ELMo vector representations of genderdefining words, then identify the presence of direct bias in the contextualized representations. We then proceed to identify gender information by means of clustering and classifications techniques. We compare our results to previous results from debiased and non-debiased word embeddings (Bolukbasi et al., 2016) . Bolukbasi et al. (2016) propose to identify gender bias in word representations by computing | 3,069,926 | 121125604 | 0 | 16 |
the direction between representations of male and female word pairs from the Definitional List ( − → he-−→ she, −−→ man-− −−−− → woman) and computing their principal components. Detecting the Gender Space In the case of contextualized embeddings, there is not just a single representation for each word, but its representation depends on the sentence it appears in. Hence, in order to compute the gender subspace we take the representation of words by randomly sampling sentences that contain words from the Definitional List and, for each of them, we swap the definitional word with its pair-wise equivalent from the opposite gender. We then obtain the ELMo representation of the definintional word in each sentence pair, computing their difference. On the set of difference vectors, we compute their principal components to verify the presence of bias. In order to have a reference, we computed the principal components of representation of random words. Similarly to Bolukbasi et al. (2016), figure 1 shows that the first eigenvalue is significantly larger than the rest and that there is also a single direction describing the majority of variance in these vectors, still the difference between the percentage of variances is less in case of contextualized embeddings, which may refer that there is less bias in such embeddings. In the right graph of the figure, we can easily note the difference in the case of random, where the data is not concentrated in a specific direction, as the weight is spread among all components. A similar conclusion was stated in the | 3,069,927 | 121125604 | 0 | 16 |
recent work (Zhao et al., 2019) where the authors applied the same approach, but for gender swapped variants of sentences with professions. They computed the difference between the vectors of occupation words in corresponding sentences and got a skewed graph where the first component represent the gender information while the second component groups the male and female related words. Direct Bias Direct Bias is a measure of how close a certain set of words are to the gender vector. To compute it, we extracted from the training data the sentences that contain words in the Professional List. We excluded the sentences that have both a professional token and definitional gender word to avoid the influence of the latter over the presence of bias in the former. We applied the definition of direct bias from Bolukbasi et al. (2016) on the ELMo representations of the professional words in these sentences. where N is the amount of gender neutral words, g the gender direction, and w the word vector of each profession. We got direct bias of 0.03, compared to 0.08 from standard word2vec embeddings described in Bolukbasi et al. (2016). This reduction on the direct bias confirms that the substantial component along the gender direction that is present in standard word embeddings is less for the contextualized word embeddings. Probably, this reduction comes from the fact that we are using different word embeddings for the same profession depending on the sentence which is a direct consequence and advantage of using contextualized embeddings. Male and female-biased words clustering. | 3,069,928 | 121125604 | 0 | 16 |
In order to study if biased male and female words cluster together when applying contextualized embeddings, we used k-means to generate 2 clusters of the embeddings of tokens from the Biased list. Note that we cannot use several representations for each word, since it would not make any sense to cluster one word as male and female at the same time. Therefore, in order to make use of the advantages of the contextualized embeddings, we repeated 10 independent experiments, each with a different random sentence of each word from the list of biased male and female words. Among these 10 experiments, we got a minimum accuracy of 69.1% and a maximum of 71.3%, with average accuracy of 70.1%, much lower than in the case of biased and debiased word embeddings which were 99.9 and 92.5, respectively, as stated in Gonen and Goldberg (2019). Based on this criterion, even if there is still bias information to be removed from contextualized embeddings, it is much less than in case of standard word embeddings, even if debiased. The clusters (for one particular experiment out of the 10 of them) are shown in Figure 2 after applying UMAP to the contextualized embeddings. Classification Approach In order to study if contextualized embeddings learn to generalize bias, we trained a Radial Basis Function-kernel Support Vector Machine classifier on the embeddings of random 1000 biased words from the Extended Biased List. After that, we evaluated the generalization on the other random 4000 biased tokens. Again, we performed 10 independent experiments, to guarantee randomization | 3,069,929 | 121125604 | 0 | 16 |
of word representations. Among these 10 experiments, we got a minimum accuracy of 83.33% and a maximum of 88.43%, with average accuracy of 85.56%. This number shows that the bias is learned in these embeddings with high rate. However, it learns in a lower rate than the normal embeddings, whose classification reached 88.88% and 98.25% for debiased and biased versions, respectively. K-Nearest Neighbor Approach To understand more about the bias in contextualized embeddings, it is important to analyze the bias in the professions. The question is whether these embeddings stereotype the professions as the normal embeddings. This can be shown by the nearest neighbors of the female and male stereotyped professions, for example 'receptionist' and 'librarian' for female and 'architect' and 'philosopher' for male. We applied the k nearest neighbors on the Professional List, to get the nearest k neighbor to each profession. We used a random representation for each token of the profession list, after applying the k nearest neighbor algorithm on each profession, we computed the percentage of female and male stereotyped professions among the k nearest neighbor of each profession token. Afterwards, we computed the Pearson correlation of this percentage with the original bias of each profession. Once again, to assure randomization of tokens, we performed 10 experiments, each with different random sentences for each profession, therefore with different word representations. The minimum Pearson correlation is 0.801 and the maximum is 0.961, with average of 0.89. All these correlations are significant with p-values smaller than 1 × 10 −40 . This experiment showed | 3,069,930 | 121125604 | 0 | 16 |
the highest influence of bias compared to 0.606 for debiased embeddings and 0.774 for biased. Figure 3 demonstrates this influence of bias by showing that female biased words (e.g. nanny) has higher percent of female words than male ones and viceversa for male biased words (e.g. philosopher). Conclusions and further work While our study cannot draw clear conclusions on whether contextualized word embeddings augment or reduce the gender bias, our results show more insights into which aspects of the final contextualized word vectors get affected by such phe-nomena, with a tendency more towards reducing the gender bias rather than the contrary. Contextualized word embeddings mitigate gender bias when measuring in the following aspects: 1. Gender space, which is capturing the gender direction from word vectors, is reduced for gender specific contextualized word vectors compared to standard word vectors. 2. Direct bias, which is measuring how close set of words are to the gender vector, is lower for contextualized word embeddings than for standard ones. 3. Male/female clustering, which is produced between words with strong gender bias, is less strong than in debiased and non-debiased standard word embeddings. However, contextualized word embeddings preserve and even amplify gender bias when taking into account other aspects: 1. The implicit gender of words can be predicted with accuracies higher than 80% based on contextualized word vectors which is only a slightly lower accuracy than when using vectors from debiased and non-debiased standard word embeddings. 2. The stereotyped words group with implicitgender words of the same gender more than in the | 3,069,931 | 121125604 | 0 | 16 |
case of debiased and non-debiased standard word embeddings. While all measures that we present exhibit certain gender bias, when evaluating future debiasing methods for contextualized word embeddings it would be worth putting emphasis on the latter two evaluation measures that show higher bias than the first three. Hopefully, our analysis will provide a grain of sand towards defining standard evaluation methods for gender bias, proposing effective debiasing methods or even directly designing equitable algorithms which automatically learn to ignore biased data. As further work, we plan to extend our study to multiple domains and multiple languages to analyze and measure the impact of gender bias present in contextualized embeddings in these different scenarios. | 3,069,932 | 121125604 | 0 | 16 |
Deciphering the Molecular Mechanism Underlying African Animal Trypanosomiasis by Means of the 1000 Bull Genomes Project Genomic Dataset Simple Summary Climate change is increasing the risk of spreading vector-borne diseases such as African Animal Trypanosomiasis (AAT), which is causing major economic losses, especially in sub-Saharan African countries. Mainly considering this disease, we have investigated transcriptomic and genomic data from two cattle breeds, namely Boran and N‘Dama, where the former is known for its susceptibility and the latter one for its tolerance to the AAT. Despite the rich literature on this disease, there is still a need to investigate underlying genetic mechanisms to decipher the complex interplay of regulatory SNPs (rSNPs), their corresponding gene expression profiles and the downstream effectors associated with the AAT disease. The findings of this study complement our previous results, which mainly involve the upstream events, including transcription factors (TFs) and their co-operations as well as master regulators. Moreover, our investigation of significant rSNPs and effectors found in the liver, spleen and lymph node tissues of both cattle breeds could enhance the understanding of distinct mechanisms leading to either resistance or susceptibility of cattle breeds. Abstract African Animal Trypanosomiasis (AAT) is a neglected tropical disease and spreads by the vector tsetse fly, which carries the infectious Trypanosoma sp. in their saliva. Particularly, this parasitic disease affects the health of livestock, thereby imposing economic constraints on farmers, costing billions of dollars every year, especially in sub-Saharan African countries. Mainly considering the AAT disease as a multistage progression process, we previously performed upstream analysis to | 3,069,933 | 248791667 | 0 | 16 |
identify transcription factors (TFs), their co-operations, over-represented pathways and master regulators. However, downstream analysis, including effectors, corresponding gene expression profiles and their association with the regulatory SNPs (rSNPs), has not yet been established. Therefore, in this study, we aim to investigate the complex interplay of rSNPs, corresponding gene expression and downstream effectors with regard to the AAT disease progression based on two cattle breeds: trypanosusceptible Boran and trypanotolerant N’Dama. Our findings provide mechanistic insights into the effectors involved in the regulation of several signal transduction pathways, thereby differentiating the molecular mechanism with regard to the immune responses of the cattle breeds. The effectors and their associated genes (especially MAPKAPK5, CSK, DOK2, RAC1 and DNMT1) could be promising drug candidates as they orchestrate various downstream regulatory cascades in both cattle breeds. Introduction Trypanosomiasis is a deadly neglected tropical disease that affects the health of several mammalian species, including cattle, horses and humans. When it affects the health of humans, this disease is commonly known as 'sleeping sickness' [1]. On the other hand, African Animal Trypanosomiasis (AAT), also known as nagana (which means 'useless' in the Zulu language), affects the health of livestock and it is spread by the tsetse fly carrying salivarian trypanosomes [2][3][4]. It prevails extensively in 40 sub-Saharan African countries and accounts for huge economic losses to farmers, particularly affecting meat and milk production [5,6]. Therefore, it has gained socio-economic importance as it retards the agricultural development of several regions in those areas [7]. Particularly, AAT is caused by different Trypanosoma species, including Trypanosoma congolense, | 3,069,934 | 248791667 | 0 | 16 |
Trypanosoma vivax and Trypanosoma brucei spp. [7]. Out of them, Trypanosoma congolense is regarded as the most serious pathogen for livestock. In humans, these unicellular protozoans cause various diseases; for example, T. brucei causes sleeping sickness, which alters the sleep-wake cycle by interfering the circadian rhythm [8,9], whereas T. cruzi causes Chagas disease or American trypanosomiasis [10,11]. Trypanosomes infect a wide range of hosts and are transmitted into the bloodstream of the mammalian host [12][13][14][15]. When the tsetse fly transmits the trypanosomes into the body of the cattle, the parasite first infects the skin resulting in the lesions due to local host immune responses. Afterwards, it enters the blood circulation via lymphatic vessels [16][17][18][19]. Important symptoms primarily observed in animals after being infected with the most pathogenic T. congolense include anaemia, loss of body conditions, thrombocytopenia [20], lymphopenia, immunosuppression [21][22][23] and other secondary infections [24]. Few West African cattle breeds like N'Dama can control the development of the disease AAT, in contrast to the other breeds such as Boran [25]. As a trait, trypanotolerance is the ability to control parasitemia (development of parasites) and the associated anaemia [12][13][14][15]. Harnessing the genetic potential of trypanotolerant breeds like N'Dama, recent studies [26][27][28][29] have focussed on investigating the trait of trypanotolerance. Mainly considering the trait of trypanotolerance, several researchers [29][30][31][32][33][34][35] have performed different types of analysis based on either gene expression data sets or genotype × phenotype data sets from the cattle breeds, namely trypanosusceptible Boran and trypanotolerant N'Dama (for a short overview, see [26,36]). Among these previous studies | 3,069,935 | 248791667 | 0 | 16 |
[29][30][31][32][33][34][35], especially, Hanotte et al. [30] performed genome-wide analyses and identified genomic regions to reveal the genetic differences between the cattle breeds related to the trait of trypanotolerance. In this regard, Noyes et al. [34] analysed the gene expression dataset to identify differentially expressed genes that responded to trypanosome infection to differentiate between the susceptible and tolerant cattle breeds. To this end, Mekonnen et al. [29] investigated the genetic background of N'Dama along with other cattle breeds. Moreover, O'Gorman et al. [33] and Gautier et al. [35] conducted the genetic and expression analyses to identify the significant chromosomal regions which could affect the susceptibility/tolerance of the cattle breeds. To decipher the underlying regulatory mechanisms determining trypanosusceptibility/trypanotolerance of these cattle breeds, we have recently analysed a time-series gene expression data set of the two cattle breeds [37,38]. Particularly, by considering the AAT disease development as a multi-stage progression process, we investigated Monotonically Expressed Genes (MEGs) to capture the complete progression process of the disease. As a result of our previous studies [37,38], we reported several transcription factors (TFs), their co-operations and master regulators governing the upstream molecular mechanism during the infection. Despite the rich literature on this disease, there is still a need for further investigation of genetic mechanisms of the regulatory processes addressing the complex interplay between regulatory SNPs, their corresponding gene expression and the downstream effectors in association with the AAT disease. Recent progress in molecular biology created the opportunity to use heterologous animal models to investigate complex traits and genetics underlying the disease mecha-nisms | 3,069,936 | 248791667 | 0 | 16 |
[39][40][41]. Remarkably, integratomics is fast becoming the latest trend in omics research while integrating a variety of omics data (such as genomic, transcriptomic and proteomic data), irrespective of the species [42]. Access to genome sequences of species like cattle unlocked the potential for integrating transcriptomic and genomic data. The information about effectors, which are end products located several steps downstream and regulate the functioning of multiple signal transduction pathways, is pivotal for understanding the complex molecular mechanisms such as the response of the cell to an extracellular pathogen. In silico study of the candidate, MEGs were undertaken to identify the novel trypanotolerance-associated rSNPs and downstream effectors. The candidate MEGs from our analysis of effectors were analysed for their gene expression profiles. To address this missing point of previous studies, we applied an integratomics approach to study the complex interplay of biological processes orchestrated by rSNPs, genes and downstream effectors during the AAT disease progression. For this purpose, we performed integrated systems biology and bioinformatics approaches while incorporating the transcriptomic data [34] and genomic data from the 1000 Bull Genome Project [43] for both cattle breeds. To examine the combinatorial interplay, we firstly identified the regulatory SNPs (rSNPs), which are located in the promoter regions of the MEGs and which, as per definition, exert a strong influence on the binding affinity of the TFs either by the deletion or the creation (gain/loss) of a transcription factor binding site (TFBS) [44][45][46]. In accordance with previous studies on the rSNPs [47,48], it is today well-known that the rSNPs based | 3,069,937 | 248791667 | 0 | 16 |
on their consequences can influence and change individual steps of gene expression. Subsequently, we extracted for each tissue (liver, spleen and lymph node) the MEGs harbouring the regulatory SNPs in their promoters by manually studying their gene expression profiles during the AAT disease progression. Finally, we explored the corresponding downstream effectors that have a pronounced effect on the activation and regulation of a multitude of downstream signalling pathways. Taken together, our findings provide a multifaceted glimpse of (i) the regulatory SNPs governing the susceptibility/tolerance mechanism of the cattle breeds; (ii) downstream effectors associated with the MEGs harbouring rSNPs, and their biological and immune-related functions, which could potentially distinguish the susceptibility/tolerance mechanism of cattle breeds to AAT disease; (iii) deciphering novel hypotheses and potential targets for breeding goals and therapeutic implications. Materials and Methods In this section, we illustrate an overview of the analyses to highlight the difference between our previous studies [37,38] and the current study. Simultaneously this overview shows how this present study complements our previous studies. Figure 1 outlines the overview of our analyses. Monotonically Expressed Genes In this study, we investigate the complex interplay of regulatory SNPs (rSNPs), the related gene expression and their corresponding downstream effectors. A time-series microarray data set, originally published by Noyes et al. (http://www. ebi.ac.uk/arrayexpress/, accession no. E-MEXP-1778, accessed on 12 March 2019) [34], has been analysed [37] to identify the Monotonically Expressed Genes (MEGs), expressed either with increasing or decreasing patterns during a biological process or a disease. The data set consisted of the gene expression recordings | 3,069,938 | 248791667 | 0 | 16 |
from three tissues (liver, spleen and lymph node) of two cattle breeds: trypanotolerant N'Dama and trypanosuceptible Boran. In this experiment, tissue harvest was performed on days 0, 21 and 35. Only the liver tissue samples were collected at additional time points such as days 12, 15, 18, 26, 29 and 32. Readers who are interested in this analysis and the identification of MEGs are kindly referred to [37]. We use these identified MEGs in our further analysis. The numbers of MEGs are provided in Table 1 and the lists of MEGs are provided in Supplementary File S1. highlighted the transcription factor co-operations associated with the AAT disease progression [37]. In our second study (top box in black dashed lines), we performed an upstream analysis to detect master regulators and over-represented upstream pathways related to AAT [38]. In the current study (bottom box in red dashed lines), we focus on the downstream analysis to decipher the complex interplay of regulatory SNPs (rSNPs), their related gene expression and their corresponding downstream effectors, which regulate a multitude of signal transduction pathways during the AAT disease progression. Genotype Data The genotype-phenotype data set of the cattle breeds Boran and N'Dama used in this study are a part of the 1000 Bull Genomes Project [43]. The genotype data contains for 23 animals (11 Boran and 12 N'Dama) 783,637 variants that are located in the promoter regions covering from −1000 bp to 0 bp relative to the transcription start sites of the MEGs. Furthermore, we considered the resistance of the cattle breed | 3,069,939 | 248791667 | 0 | 16 |
as a qualitative phenotype and assigned '0' and '1' to represent the disease phenotypes for resistance and susceptibility, respectively. Similar to our previous studies [46,49], for the purpose of quality control, filtering of genotype data was then carried out to remove the SNPs with a minor allele frequency (MAF) less than 0.1, call rate less than 0.95 and which significantly deviated from Hardy-Weinberg Equilibrium (p < 1× 10 −8 ). After this filtering, the data set contained about 19,330 SNPs and 23 animals for further analyses. We performed a Genome-Wide Association analysis using PLINK 1.9 software [50]. The genotype-phenotype association was evaluated with PLINK by chi-squared allelic test. As suggested by Heinrich et al. [46], we used the false discovery rate (FDR) of 0.1 to control the type I error rate. Identification of Regulatory SNPs In previous studies [44,46], an SNP is defined to be a regulatory SNP (rSNP) if it is located in the promoter region of a gene and if it affects the binding affinity of one or more transcription factors (TFs) to their respective binding sites which leads to the gain/loss of TFBSs. According to the rSNP detection pipeline, we extracted the flanking sequence of ±25 bp for each selected SNP. Finally, we scanned the flanking sequences of the SNPs for both alternate and reference alleles using the MATCH TM program [51] and thus classified an SNP as rSNP if it leads to gain and loss of a TFBS. Finding the Effectors Taking the rSNPs into account, we filtered the list of MEGs | 3,069,940 | 248791667 | 0 | 16 |
under study that harbour at least one rSNP within their promoter. Using the filtered list of MEGs for each tissue individually, we employed the systems biology platform geneXplain [52] to identify the effector molecules. Effectors are important signalling molecules that are end products located several steps downstream and regulate the functioning of a multitude of signalling cascades. With regard to AAT disease, the knowledge about the effectors could provide promising information to decipher their complex interplay with rSNPs and the corresponding MEGs. The identification of effectors was performed using the 'Effector search' function on the geneXplain platform, which utilises the TRANSPATH ® database [53] for searching the downstream effectors regulated by the input set of MEGs. Results and Discussion By analysing regulatory SNPs (rSNPs), the related gene expression profiles of MEGs and their associated downstream effectors, we established their complex interplay involved in the AAT disease progression for both cattle breeds. For this purpose, we firstly performed a genome-wide association analysis and obtained 19,330 significant SNPs, out of which 1849 SNPs have been further classified as rSNPs. Uncovering disease-related SNPs is recently gaining utmost importance as they can have an impact on the disease progression and also on how the infected individual responds to the infection [54][55][56][57][58]. In particular, rSNPs are of great interest as they could be causal and thus alter the protein-DNA interaction. Afterwards, considering the MEGs of interest, which harbour at least one rSNP in their promoter regions, we created for each tissue a filtered list of monotonically expressed genes. Finally, using these | 3,069,941 | 248791667 | 0 | 16 |
lists of MEGs obtained for each tissue (liver, spleen and lymph node) for both cattle breeds, we identified the downstream effectors to investigate further the underlying molecular mechanisms that orchestrate differences in the level of tolerance of the cattle breeds to AAT. The numbers of rSNPs and MEGs of interest are given in Tables 2 and 3, respectively. The list of respective rSNPs and MEGs are provided as Supplementary Files S2 and S3. Table 3. Numbers of MEGs under study harboring at least one rSNP in their promoter region, for liver-, spleen-and lymph node-tissues for the cattle breeds Boran and N'Dama. Identification of Downstream Effectors We employed the "Effector Search" algorithm from the geneXplain platform [52] using the tissue-based MEG sets of interest for the computational identification of downstream effectors. From this analysis, we obtained a total of 18 effectors that are unique for the breeds and the three tissues (given in Table 4). Remarkably, the effectors obtained are completely different between the susceptible and tolerant cattle breeds. Downstream Effectors for Liver Tissue The analysis of the MEGs for the liver tissue resulted in the detection of three effectors for Boran (namely SRF, PKCδ and a complex of proteins ITK, LCK, PLCγ and SLP76) and N'Dama (p85α, chTOG:H3F3A and TF2-1). Serum response factor (c-fos serum response element-binding transcription factor) is a transcription factor belonging to the MADS (MCM1, Agamous, Deficiens and SRF) box superfamily. It is mainly involved in the regulation of immediate-early genes and takes part in important cellular processes like cell differentiation, cell growth | 3,069,942 | 248791667 | 0 | 16 |
and apoptosis. The gene encoding this protein serves as the major target for several signalling pathways, in particular, the mitogen-activated protein kinase pathway (MAPK) that plays a significant role in the immune surveillance mechanism supporting the trypanosome infection [59]. Therefore, the SRF protein could be directing the immune evasion, thereby assisting susceptibility of the cattle breed in AAT disease progression. The second effector, PKCδ, found in Boran's liver tissue, has been reported as the marker of inflammation and plays an essential role in tuberculosis disease progression in humans [60]. This could be an important hint for the AAT disease progression in the susceptible cattle breed Boran. Moreover, the third effector consists of four proteins, namely ITK, SLP 76, LCK and PLCγ1. Inducible T-cell kinase (ITK) belongs to the Tec family of non-receptor tyrosine kinases, which are expressed in immune cells like mast cells and T cells. It plays a critical role in T-lymphocyte development and functioning and is involved in regulating T-cell receptor signalling. Furthermore, it is activated with respect to antigen receptors, for example, T-cell receptor stimulation [61][62][63]. It is reported to be important for the replication of the virus inside the infected host cells [64], elucidating its role in supporting the pathogen infection in AAT. SH2-domain-containing leukocyte protein of 76 kDa (SLP 76) is one of the key adaptor proteins expressed only in the haematopoietic part of the immune cells such as monocyte, granulocyte and T lymphocyte lineage [65]. The protein SLP 76 plays a crucial role in the regulation of several signalling cascades | 3,069,943 | 248791667 | 0 | 16 |
[66]. Additionally, its expression is regulated during T cell maturation and activation [65]. This demonstrates the close association of the protein SLP 76 with the haematopoiesis and generation of immune responses relating to anaemia in AAT disease, an important hallmark of AAT. The association of LCK (lymphocyte-specific cytoplasmic protein-tyrosine kinase) to CD4 and CD8 is necessary for antigen-specific T cell development and activation [67]. Of particular interest, phospholipase C gamma 1 (PLCγ1) signalling is important for several physiological processes like cell differentiation [68,69]. In our analysis, we found an effector as a complex of chTOG and H3F3A for the liver tissue of N'Dama. The chTOG is a human TOG protein, reported as a mitotic error correction factor playing an important role in accurate chromosome segregation during cell division [70]. Further, H3F3A belongs to the group of basic nuclear histone proteins supporting the structure of the chromosome, thereby maintaining the genome integrity [71]. Another effector, TF2-1, found in the liver tissue of N'Dama, is a non-infectious and intracellular retrotransposon [72]. However, both of these effectors were not illustrated in relation to host-pathogen interaction, and thus, their potential roles in AAT disease progression are not studied. On the other hand, the third effector p85 α, is an adapter subunit of heterodimer phosphatidylinositol 3-kinase, which is involved in the production of phospholipids. By interacting with other proteins such as p110 α and PTEN, p85 α could regulate the PI3K pathway either in a positive or negative manner [73]. Due to the importance of the phosphatidylinositol 3-kinase (PI3K) signalling pathway | 3,069,944 | 248791667 | 0 | 16 |
in many diseases [74], the regulatory activity of p85 α is gaining importance in response to infections as well. This demonstrates the role of p85 α during AAT infection, which might play a crucial part in trypanotolerance of N'Dama by maintaining the lipid synthesis in the host's liver intact without interruption from the pathogenic attack. The first two effectors are a complex of two proteins: HEXIM1 and p53. Hexamethylene bisacetamide-inducible protein 1 (HEXIM1) protein encoded by HEXIM1 is known for its role in the regulation of gene expression, especially with regard to innate immunity [75]. Particularly, it has been reported in the Trypanosoma cruzi infection, in association with splenomegaly in the Hexim1 +/− mice. It was shown how the downregulation of HEXIM1 protects the host against T. cruzi infection [76]. This hints at the functioning of HEXIM1 during the infection process by increasing inflammation. Another part of the protein complex, p53, identified for the spleen tissue, acts as a tumour suppressor protein in humans, therefore called as "guardian of the genome" [77,78]. In recent studies, it has been demonstrated that p53 regulates inflammation [79] which is highly associated with AAT. Especially in a study involving bacterial infection [80], deletion or inhibition of p53 resulted in the clearance of extracellular bacteria, which reveals the regulatory role of p53 in the defence against extracellular pathogens establishing the modulation of microbicidal function. Another effector found in the spleen tissue, DNA Protein Kinase, has been reported for its roles in regulating metabolic pathways, particularly in fatty acid synthesis [81]. It | 3,069,945 | 248791667 | 0 | 16 |
is one of the key players responding to DNA damage and in IRF-3-dependent innate immunity [82]. Especially, DNA Damage Response PK has been studied as a driver in evading host immunity [83] and in developmental transitions occurring between the vector and the host [84]. This effector could play a role in immune evasion, thereby supporting the trypanosome infection and increasing the susceptibility of Boran. For the spleen tissue of N'Dama, the identified effectors, including LYZL2 isoform 2, PON2 isoform 1 and WSX1 are complexes of LRP11 protein. LRP11 plays a key role in the development of stress responses in mice, as suggested by Xu et al. in [85]. It is wellknown that through the activation of the stress response, the host's body provides energy immediately available for immune responses against the parasitic infection, therefore benefitting the host to recover earlier [86]. LYZL2 identified as one of the effectors, exhibits lysozyme activity, which functions as bacteriolytic factors [87] and they are mainly involved in the host defence. Their biological function in relation to parasitic infection has not been largely studied yet. Interestingly, we found Paraoxonase 2 (PON2) in the spleen tissue of N'Dama, which is an intracellular membrane protein exerting anti-oxidant functions [88]. Macrophages are key players against extracellular and intracellular pathogens. In this regard, PON2 has been studied for their expression in the macrophages [89]. In a study involving bacterial infection with Pseudomonas aeroginosa, the role of PON2 in the innate immune defence has been demonstrated [90]. The next effector, WSX1, is a class I cytokine | 3,069,946 | 248791667 | 0 | 16 |
receptor for IL27 and is predominantly expressed in lymphoid tissues like the spleen and lymph nodes [91]. Being the IL27 receptor, WSX1 has been studied to be associated with the IL27 signalling pathway. It is further involved in the regulation of Th1-type adaptive immune responses and also of the cells of the innate immune system [92]. Villarino et al. reported in their study [93] that WSX1 is necessary for resistance to parasitic infection from Toxoplasma gondii. Particularly, this could provide an important hint on the functioning of WSX1 in resistance of N'Dama to AAT disease. Downstream Effectors for Lymph Node The analysis of the MEGs of lymph node tissue reveals the effectors, namely LIMP-2:Prpf8, VICKZ3:Prpf8 and SNRPGP15:Prpf8, for Boran and the effectors Ssu72, MTMR4, Clathrin LCb for N'Dama. Considering the biological roles of effector LIMP-2, it is a type III glycoprotein belonging to the CD36 superfamily of scavenger proteins, facilitating the transport of the acid hydrolase β-glucocerebrosidase (GC) [94]. This protein provides a strong connection between cholesterol export and innate immunity [95,96] as lipids play crucial roles in the multiplication of the trypanosome infection cycle. Therefore, the LIMP2 protein might be a strong candidate protein crucial for establishing the infection, thereby making the cattle breed Boran susceptible to AAT. Another effector, VICKZ3, for the lymph node issue of Boran, belongs to the family of RNA binding proteins and is expressed in the developing central nervous system [97] during embryogenesis. This group of proteins are associated with the regulation of RNA and are involved in controlling the | 3,069,947 | 248791667 | 0 | 16 |
cellular processes like proliferation and translational repression [98]. Furthermore, the effector SNRPGP15 (Small Nuclear Ribonucleoprotein G-like protein 15) is a part of the spliceosome, which mainly takes part in RNA metabolism [99]. Finally, part of the protein complexes of all the three effectors is pre-mRNA processing factor 8 (Prpf8) is a highly conserved protein and known for its role in the pre-mRNA splicing process [100]. However, VICKZ3, SNRPGP15 and PRPF8 have not been largely studied in terms of host-pathogen interaction; therefore, their potential role in AAT disease progression is currently unknown. On the other hand, the effectors identified for the lymph node tissue of N'Dama suggest their crucial roles in immunity, bolstering the host's defence against the parasite. The effector Ssu72 is a dual protein phosphatase that plays a role in RNA processing. A recent study has associated the Ssu72 protein in macrophages with the process of immunometabolism [101], implicating a closer connection between immunity and trypanotolerance of N'Dama. The next effector, Myotubularin-related protein 4 (MTMR4), is an intracellular protein that exhibits lipid and protein phosphatase activities in several cellular functions. Especially MTMR4 is involved in the negative regulation of TGF-β signalling. During the infection of Trypanosoma cruzi, the role of TGF-β has been demonstrated to inhibit the functioning of immune effector cells and the production of interferon α, thereby resulting in the multiplication of the pathogen [102]. Therefore, MTMR4 indirectly assists the host in decreasing the pathogen numbers within the body, supporting the tolerance mechanism of the cattle breed N'Dama. Another effector, Clathrin, is a | 3,069,948 | 248791667 | 0 | 16 |
cytosolic protein made up of heavy and light chains. Clathrin light chains (LCb) are important components of clathrin-coated vesicles, especially necessary to uptake large foreign particles into the vesicles [103]. This effector found in lymph nodes could represent the engulfing of infectious parasites during the AAT disease in the body of N'Dama. In particular, the knowledge of these effectors provides essential information in distinguishing the downstream events underlying the susceptibility and tolerance mechanisms of the cattle breeds Boran and N'Dama, respectively. Further validation of the results from the molecular biology end is necessary to evaluate the biological importance of their functions in the AAT disease progression as well as to gain a comprehensive understanding of their roles in susceptibility/tolerance mechanisms of the cattle breeds. Gene Expression Profile Analysis of MEGs Harbouring rSNPs Using gene expression profiles, it is possible to gain insights into the differences in the expression levels under certain cellular conditions. Therefore, we were additionally interested in the gene expression profiles for the MEGs of interest to decipher their differentiation between the cattle breeds. For this purpose, we manually analysed and then annotated the gene expression profiles of MEGs for each tissue to investigate their expression patterns. A closer look at these gene expression profiles reveals the distinguishing expression patterns for five MEGs (namely MAPKAPK5, CSK, DOK2, RAC1 and DNTM1) expressed over several time points in the liver tissue of both breeds Boran and N'Dama (see Supplementary File S4). Interestingly, these genes are key players in the detection of effectors found in liver tissue | 3,069,949 | 248791667 | 0 | 16 |
(see Supplementary File S4). Gene expression profiles of other MEGs of interest are provided in Supplementary File S5. Figure 2 exemplarily shows the changes in the gene expression profile of MAPKAPK5 for liver tissue of both cattle breeds, harbouring rSNPs in its promoter region. Considering the biological roles, MAPKAPK5 (MAPK Activated Protein Kinase 5), encoded by the gene MAPKAPK5, is a serine/threonine-protein kinase that plays a major role in the posttranscriptional regulation of MYC, [104,105] which is intimately associated with immune evasion [106]. The protein encoded by the gene CSK plays a critical role in the activation of Tcells and is involved in several pathways, which include the regulation of Src family kinases [107]. Expression of DOK2 has been reported to regulate the cell cycle of haematopoietic stem cells. Furthermore, the inactivation of DOK2 also resulted in the aberrant activation of MAP kinase [108], implicating that their functional loss could exacerbate the AAT disease. The protein encoded by RAC1 (Rac Family Small GTPase 1) is important in regulating cellular processes like phagocytosis of apoptotic cells and binds to effector proteins in their active state [109]. DNMT1 plays a critical part in regulating the immune system and is regarded indispensable for the inhibition of Foxp3+Treg cells [110]. Conclusions Transcription factors are involved in regulating transcription processes by binding to short DNA sequences called transcription factor binding sites (TFBSs). In particular, single nucleotide polymorphisms (SNPs) are widely studied with regard to the disease mechanisms as they can have direct control over the disease susceptibility (causal polymorphisms). Importantly, regulatory | 3,069,950 | 248791667 | 0 | 16 |
SNPs (rSNPs) that are located in the regulatory regions like promoters can significantly affect the gene expression, especially by modifying the binding sites of the TFs. Knowledge about the rSNPs and their complex interplay with the corresponding gene expression and downstream effectors could reveal multiple disease-associated polymorphisms, which can be further used as targets in drug design and breeding programs. Taking the importance of rSNPs and their combinatorial interplay into account, we performed a systematic investigation of genomic and transcriptomic data of two cattle breeds, Boran and N'Dama, to unravel the underlying genetic mechanisms of AAT disease progression. Our findings provide mechanistic insights into significant rSNPs, which are harboured within the promoter regions of MEGs. Moreover, our further investigation of effectors found in the liver, spleen and lymph node tissues of both cattle breeds enhanced our understanding of distinct mechanisms leading to either resistance or susceptibility of cattle breeds. Our current study complements our previous studies, which mainly focused on the upstream events, including TFs and their co-operations as well as master regulators. Taken together, our findings provide a multifaceted glimpse of (i) novel insights into the rSNPs governing the susceptibility/tolerance mechanism of the cattle breeds; (ii) downstream effectors, particularly LYZL2, WSX1 and MTMR4 and their biological roles related to innate and adaptive immune responses during the AAT disease progression. | 3,069,951 | 248791667 | 0 | 16 |
Nanocatalysts Unravel the Selective State of Ag In the present work, we report on a comparative study of model catalysts during ethylene epoxidation reaction under industrially relevant conditions. The catalysts consist of Ag nanoparticles <6 nm and a reference sample ∼100 nm. Combining catalytic data with transmission electron microscopy, thermal desorption spectroscopy, and density functional theory allows us to show that catalytic performance is linked to the oxygen concentration in/on the Ag particles. Isotope experiments using 18O2 and C18O2 are conducted to gain insight into the nature and location of oxygen in/on the Ag nanoparticles. The oxygen species responsible for the CO2 formation and inhibition of the overall catalytic activity are identified, and the abundance of those species is shown to depend strongly on the pre‐treatment and reaction conditions, showing both are critical for effective oxygen management. By comparison with a conventional Ag/α‐Al2O3 catalyst, we demonstrate a low concentration of oxygen in/on Ag leads to the highest selectivity regardless of particle size. However, particle size dependent oxophilicity leads to significantly lower TOFs for the Ag nanoparticles. This study provides fundamental understanding of the performance of supported Ag particles in ethylene epoxidation and offers new strategies to improve performance under industrially relevant conditions. Introduction An intuitive approach to maximize the conversion of a catalyst is to increase its active surface area to provide more catalytically relevant reaction sites. This is accessible by, e. g., nanostructur-ing to increase the reactivity per unit mass relative to macroscopic crystals of the same substance. [1] Besides, the formation of nanoparticles of, | 3,069,952 | 216166674 | 0 | 16 |
typically, 1-10 nm gives rise to phenomena described as particle size (PS) effects. [2] This property has already been observed for a number of supported metal based catalysts like Co, [3] Ni, [4] Pd, [5] Pt, [6] Cu, [7] Ag [8] and Au. [9] The overall consensus from such observations is that the PS effect strongly depends on the kind of reaction as well as the active material, thereby leading to beneficial or detrimental effects. The possibility of a PS effect for Ag in the partial oxidation of ethylene to ethylene epoxide (EO) has long captivated researchers, as it could act on EO selectivity and or conversion by altering the nature of the oxygen species on and in the Ag particles. Yet, after more than 40 years of research [10] it is still debated if a true Ag PS effect even exists, much less whether it contributes to EO selectivity and its role in oxygen management of real catalysts. An overview of some selected results dealing with the influence of the Ag particle size for supported Ag catalysts in the epoxidation of ethylene is shown in Scheme 1. Wu and Harriott [10b] studied supported Ag on silica in the range of 3-50 nm Ag PS and observed the highest conversions of ethylene, X(C 2 H 4 ), at around 5-6 nm, but with very low selectivity to ethylene epoxide, S(EO), the latter steadily increased with increasing PS. Verykios et al. [11] investigated larger Ag particles, starting from 30 nm, and observed the same phenomenon for S(EO), | 3,069,953 | 216166674 | 0 | 16 |
with a minimum in X(C 2 H 4 ) in the range of 50-70 nm Ag PS. This is in reasonable agreement with the observations of Wu and Harriot. However, since Verykios et al. normalized their rates per square meter of "free-metallic-silver surface area" determined by selective oxygen chemisorption, any comparison in terms of X(C 2 H 4 ) has to be interpreted with caution. Investigations of Cheng and Clearfield [12] showed a maximum of S(EO) and X(C 2 H 4 ) at around 50 nm, with a decrease in S(EO) for larger PS in stark contrast to the aforementioned studies. Lee et al. [13] also observed a steady increase in S(EO) for increasing PS, but this time from 6-50 nm and in the range of 50-100 nm S(EO) remained unchanged. The X(C 2 H 4 ) showed a maximum at 40-50 nm in the case of corundum supported Ag and a plateau starting at around 60 nm for silica supported Ag. The Ag on corundum support performed similar to results reported by Cheng and Clearfield. Goncharova et al. [14] were able to show a steady increase in S (EO) from 10-100 nm, which is in line with that reported by Wu/Harriot and Verykios et al., and a maximum in X(C 2 H 4 ) at around 50 nm in agreement with Cheng/Clearfield. The listed results were obtained during a period of 20 years and it took another 17 years until the group of Petra de Jongh investigated the PS of Ag and its influence on | 3,069,954 | 216166674 | 0 | 16 |
the ethylene epoxidation reaction. In contrast to the previous investigations, the S(EO) at a constant low conversion of 2.8 % were compared, resulting in constant S(EO) values. The achieved X(C 2 H 4 ) were again in agreement with previous studies, reaching its maximum at 60-70 nm. As a consequence, no PS effect on the selectivity in the range of 20-200 nm Ag PS was observed. A very recent study of van Hoof et al. [15] identified an increase in selectivity with PS (from ca. 20-200 nm). The Ag weight-based activity decreased until 50 nm and was stable afterwards. Besides, the Ag surface area-based activity was stable until 50 nm and increased with the Ag PS. The proposed explanation involves the complex interplay of Ag bulk, Ag crystallite size and grain boundaries, which remained rather elusive. Since the synthesis of Ag particles < 10 nm is difficult to accomplish, only a limited number of studies are available. Demidov and co-workers [16] investigating the model catalyst Ag/HOPG which stayed inactive for an average Ag PS of 8 nm. In contrast, 40 nm Ag particles on HOPG showed the formation of ethylene epoxide. These results stand contradictory to the report of Fotopoulus et al., [17] where a catalyst with 9 nm Ag particles on MCM-41 showed a X(C 2 H 4 ) of up to 65 % with S(EO) of 30-35 %. Such reasonable performances were comparable to the reference Ag/α-Al 2 O 3 catalyst with Ag particle sizes > 60 nm. The essence of more than 40 | 3,069,955 | 216166674 | 0 | 16 |
years of research regarding the PS effect of Ag in the ethylene epoxidation reaction is unclear. The trends of the S(EO) and X(C 2 H 4 ) curves illustrated in Figure 1 are arbitrary with respect to the Ag PS regimes. A maximum for X(C 2 H 4 ) for Ag PS around 50-60 nm, as well as a poor S(EO) for small Ag particles seem to be consistent. But, the origin of the observed effects for different catalysts is still not clear and the observations often appear contradictory. An explanation for the inconsistent situation in terms of a Ag PS effect might be the broad PS distributions achieved so far. Furthermore, the achieved Ag PS were above the relevant range, up to 6 nm, for which a dependence on catalytic performance is expected. [8,18] Besides, related oxidation reactions might also be helpful and serve as orientation for any PS effect. Lei et al. [8] used Ag cluster-based catalysts supported on silicon wafer in the range of 0.5-3.5 nm. Those small clusters and particles were active and selective in epoxidation of propylene with high turnover rates. Comparable results were achieved in our recently published study [19] of supported Ag catalysts in the oxidation of CO. In-situ synthesized Ag clusters < 1 nm showed exceptionally high CO oxidation rates com-pared to Ag nanoparticles of 1-6 nm. Further, the Ag clusters and particles showed much higher CO oxidation rates than a Ag/α-Al 2 O 3 reference catalyst with 40 nm crystalline domain size. Since both studies showed that | 3,069,956 | 216166674 | 0 | 16 |
Ag nanoparticles are able to perform oxidation reactions, the motivation of this study was to transfer these performances to the epoxidation of ethylene and to answer the long standing question of a Ag PS effect. Unraveling this possible PS effect is expected to lead to a deeper understanding of the selective state of Ag in general, as the PS effect has been linked to the nature of the adsorbed oxygen species present on active catalysts. [16] Applying the recently introduced synthesis strategy, [19] SiO 2 supported Ag nanoparticles up to 6 nm with a narrow PS distribution are investigated. To ensure the relevance of the conducted study for any catalytic discussion, the samples are tested under industrially relevant conditions of high pressure. The performances of the Ag/SiO 2 catalysts are compared to a conventional Ag/α-Al 2 O 3 catalyst. All samples are thoroughly analyzed before and after catalytic testing by way of powder Xray diffraction (PXRD) and transmission electron microscopy (TEM). Further, thermal desorption spectroscopy (TDS) allows us to identify the relevant Ag-O interactions related to the catalytic results. Within this study the adding of promoters on the catalyst or using organochloride as co-feed is excluded (as used on the industrially applied catalyst), [20] since it influences the origin of the AgÀ O interaction and will be part of a separated manuscript. Experimental Section List of the used chemicals Synthesis of Ag/SiO 2 and Ag/α-Al 2 O 3 has already been described. [19] In short, a vacuum assisted impregnation technique was applied to achieve a controlled | 3,069,957 | 216166674 | 0 | 16 |
distribution on the support. For the silica (Aerosil® 300, Degussa, hydrophilic fumed silica powder, primary particles: spherical, 7-40 nm, no porosity) supported catalyst AgNO 3 was dissolved in water according to 5 wt.-% Ag loading (labeled as Ag5/SiO 2 ) and used as impregnation solution. The amount of H 2 O needed for impregnation was determined by determining the "solvent capacity volume". The impregnated support was dried, transferred into 100-200 μm sieve fraction and subsequently calcined at 600°C for 1 h in a rotating tube furnace with a constant flow of 21 % O 2 in Ar (300 ml · min À 1 ) with a heating rate of 2°C · min À 1 . For the synthesis of the industrial reference sample, 15 wt.-% Ag was loaded on α-Al 2 O 3 using an Ag oxalate based precursor according to patent literature, labeled as Ag15/α-Al 2 O 3 . [20] The α-Al 2 O 3 support was then impregnated with the silver oxalateethylenediamine solution followed by a calcination under air. Ethylene epoxidation was performed in a stainless steel plug flow reactor (inner diameter 4 mm) at 17.5 bar absolute pressure and a gas composition of 7/8/50/35 for O 2 /Ar/N 2 /C 2 H 4 at a GHSV of 4850 h À 1 . The temperature was raised stepwise with a 1°C · min À 1 heating rate and a dwell time of 6 hours for each temperature step. The composition of the exhaust gases was analysed online using a gas chromatograph (Agilent 7890 N) | 3,069,958 | 216166674 | 0 | 16 |
equipped with a FID and TCD detector. The samples were pre-treated for 24 hours at 210°C either in N 2 or synthetic air (named as O 2 ). For the kinetic analysis internal and external transport limitations, as well as heat transfer problems, are excluded by reference measurements ( Figure S12). Ethylene oxide decomposition tests were performed at 250°C under a constant flow of 1 % EO in He. The temperature is higher than the actual testing temperature to force the EO decomposition. All tested samples were pre-treated in synthetic air at 250°C for 10 min. The Products were analysed by GC-MS analysis. Powder X-ray diffraction (PXRD) patterns were recorded using a Bruker AXS D8 Advance II Theta/Theta diffractometer in Bragg-Brentano geometry using Ni filtered Cu K α1 + 2 radiation and a position sensitive LynxEye silicon strip detector. The sample powder was filled into the recess of a cup-shaped sample holder, the surface of the powder bed being flush with the sample holder edge (front loading). The resulting diffractograms were analyzed by full pattern fitting using the Topas software [21] to extract lattice parameters, crystallite sizes and Ag loading. Scanning transmission electron microscopy (STEM) imaging was performed using a double Cs corrected JEM-ARM200CF (Jeol) operated at 200 kV and equipped with HAADF (high angle annular dark-field) and BF (bright-field) detectors. Samples were prepared by direct deposition of dry powder onto a Quantifoil Au holey grid. For the resulting histograms, the diameter of 1000 particles for Ag5/SiO 2 and 200 particles for Ag15/α-Al 2 O 3 | 3,069,959 | 216166674 | 0 | 16 |
were measured. Thermal desorption spectroscopy (TDS) was applied for the temperature programmed desorption of oxygen. Therefore, a selfconstructed setup which enables the testing of powder samples was used. The setup is equipped with mass flow controllers, an IR- ChemCatChem Full Papers doi.org/10.1002/cctc.202000035 light furnace (Behr IRF 10) and a mass spectrometer (Pfeiffer Vacuum QME 200). The powder sample is placed on a small quartz-glass boat which is placed in a quartz tube (inner diameter of 14 mm, outer diameter of 20 mm, length of 450 mm) located inside the furnace and connected to the system using Ultra Torr vacuum fittings. Prior to the desorption experiment the samples were pre-treated at 1 bar in 25 % O 2 in Ar (synthetic air) at a flow of 100 ml · min À 1 for 12 h at 210°C, which cleans the surface of the sample and saturates the Ag with oxygen (surface and bulk). The gases are detected using the mass spectrometer leak valve. Afterwards the system is stepwise brought to 9 · 10 À 7 mbar and directly connected to the mass spectrometer. The desorption experiment is conducted at a heating rate of 25°C · min À 1 up to 700°C. All masses and the temperature are monitored online. Inductive coupled plasma -optical emission spectroscopy (ICP-OES) was used to determine the Ag loading of the catalysts. Therefore, the sample is solubilized using LiF, nitric acid and water at 230°C, diluted with water and analyzed with a Perkin Elmer ICP OES Optima 8300. Microcalorimetry was performed in a | 3,069,960 | 216166674 | 0 | 16 |
HT1000 (RT-1000°C) and MS70 (RT-100°C) Tian-Calvet calorimeter (Setaram) combined with a custom-designed high vacuum (HV) and gas dosing apparatus. The sample was placed in batch reactor. O 2 and C 2 H 4 adsorption experiments at 230°C were performed after cleaning the samples at 400°C for 15 h in H 2 (400 mbar). The reoxidation/regeneration at 350°C in O 2 was kept for 5 h (400 mbar). Density functional theory (DFT) calculations were performed with the Quantum ESPRESSO package [22] at the PBE level including dispersion corrections with the exchange-hole dipole moment (XDM) model. [23] Following earlier work, [24] PAW datasets were taken from PS library [25] and used with a kinetic energy cutoff of 30 Ry. A k-point mesh equivalent to (12 × 12) for the (1 × 1) Ag(111) surface unit cell was employed along with Marzari-Vanderbilt smearing with a 0.02 Ry smearing parameter. [26] Minimum energy paths were computed with the climbing image nudged elastic band method using eight images for each path. Results and Discussion The investigated catalysts consist of Ag particles on SiO 2 and α-Al 2 O 3 . The low surface area α-Al 2 O 3 support serves as industrial reference with large Ag particles. To stabilize Ag nanoparticles in the range of~2 nm a high surface area support material in the range of 300-400 m 2 · g À 1 has to be used. [19] To avoid EO isomerization and combustion, support materials in the ethylene epoxidation reaction have to be chemically inert and with no acid functional | 3,069,961 | 216166674 | 0 | 16 |
groups. The used SiO 2 material fulfills the given criteria (Table 1, 328 m 2 · g À 1 ) as thoroughly verified by EO decomposition experiments, conducted prior to the catalytic tests ( Figure S1, Table S1). In the following the non-reducible supports are interpreted as inert without any further impact on the results. Sample preparation and characterization According to an established synthesis protocol, [19] 5 wt.-% Ag supported on SiO 2 (Ag5/SiO 2 ) was prepared. In addition, a reference sample with 15.5 wt.-% Ag supported on α-Al 2 O 3 (Ag15/α-Al 2 O 3 ) was synthesized, following Rosendahl et al. [20] The corresponding PXRD are shown in Figure 1. The Ag15/α-Al 2 O 3 catalyst exhibits slightly broadened Ag reflections (in comparison to the Ag powder reference, Figure 1 green pattern) and reflections of α-Al 2 O 3 . The widening is explained by the reduced domain size, which was determined to be 39.4 � 4.2 nm by full pattern fitting. Ag5/SiO 2 shows a pronounced broadening of the Ag reflections, a direct indication of the smaller domain sizes of 6.5 � 0.7 nm. The SiO 2 support is responsible for the diffuse reflection visible in the range of 30-40°2Θ. The Ag lattice parameters are also determined for Ag5/ SiO 2 and Ag15/α-Al 2 O 3 as 4.089 � 0.012 Å and 4.08603 � 0.00009 Å, respectively. Within the uncertainty of the fitted results, the lattice parameters are in good agreement with the reported reference value of 4.086 Å [27] for Ag | 3,069,962 | 216166674 | 0 | 16 |
0 . Complementary to PXRD, STEM analysis was performed in order to extract the PS information illustrated in Figure Figure 2b shows in addition the element specific EDX maps (see also Figure S2). The main PS determined by STEM are larger than the domain sizes determined by PXRD, indicating that the fraction of larger particles have been formed by sintering. The Ag particles were sometimes irregularly shaped (non-spherical), which explains the broad distribution. For Ag5/ SiO 2 (Figure 2 c, d) the Ag particles are homogeneously distributed over the whole support with a very narrow size distribution (corresponding histogram inset Figure 2d, a PS of 2.3 nm and a standard deviation of 0.72 nm, see also Figure S3a). The absence of Ag particles larger than 5.8 nm is in good agreement with the results from PXRD (domain sizes of 6.5 � 0.7 nm). Since XRD is only sensitive to crystallites < 2-3 nm, the main fraction of the Ag PS stays invisible and the XRD domain size is interpreted as upper limit. An overview of the relevant parameters of the used catalysts is presented in Table 1. The Ag loading was experimentally determined by ICP-OES and resulted in 4.4 wt.-% Ag for the Ag5/SiO 2 sample and 12.4 wt.-% for Ag15/α-Al 2 O 3 . In addition, the quantitative PXRD analysis (Rietveld method) for Ag15/α-Al 2 O 3 determined the Ag loading to be 13.9 wt.-%. Generally, the quantitative assessments are in good agreement with the nominal values. Ethylene epoxidation tests All catalysts were tested in | 3,069,963 | 216166674 | 0 | 16 |
the epoxidation of ethylene to EO with applied gas feed, temperatures and pressures of the industrial process (without co-feeds, see also experimental part). [20] As a surface purification step, a pre-treatment temperature of 210°C was chosen, which corresponds to the decomposition temperature of both Ag 2 CO 3 (175°C-225°C [28] ) and Ag 2 O (~200°C [29] ). In the following, the pre-treatment atmospheres, which were varied for the Ag5/SiO 2 sample from N 2 to O 2 (synthetic air), is part of the label, i. e. Ag5/SiO 2 -N 2 or Ag5/SiO 2 À O 2 (this is not necessary for the reference catalyst since identical performances were obtained). Figure 3 shows S (EO) as a function of the oxygen conversion, X(O 2 ), at various temperatures (dwelled for 6 h each) for the Ag5/SiO 2 À N 2 , Ag5/ SiO 2 À O 2 catalyst and the Ag15/α-Al 2 O 3 reference. The performance of the Ag15/α-Al 2 O 3 catalyst in steady state follows a typical S to X behavior (lower selectivity at higher conversions) indicated by the grey line. Besides, the high selectivity towards EO is instantly reached and only at higher reaction temperatures (190 and 200°C) an increase in S(EO) as function of time is observed (see also Figure S4, due to decreasing CO 2 rates). The Ag5/SiO 2 À N 2 catalyst shows a pronounced activation period, for which the S(EO) steadily increases with time and temperature. The S(EO), e. g., increases from 48 % at 150°C (after 2.5 | 3,069,964 | 216166674 | 0 | 16 |
h dwell time) to a maximum of 64 % at 180°C, which is interpreted as successful activation. The corresponding X(O 2 ) shows only minor changes up to 180°C (~4-7 %). Further increasing the temperature leads to decreasing S(EO) and strong increase in X(O 2 ) following an expected S to X behavior. At~210°C Ag nanoparticles start to sinter and the Ag5/SiO 2 À N 2 catalyst loses activity (see Figure S5, decrease of EO and CO 2 rates). Since the deactivation of the Ag5/SiO 2 À N 2 catalyst starts already at 200°C, the same catalyst was tested again until 230°C (to trigger deactivation) and 180°C (to avoid deactivation). The corresponding PS analysis ( Figure S3c) after testing at elevated temperature shows a broad distribution of Ag particles until 40 nm (main PS~5 nm). This serves as a textbook example of sintered particles continuously growing in size. In contrast, the PS distribution from the catalyst tested until 180°C (highest S(EO)) remains narrow ( Figure S3b, main PS~2 nm, almost identical to the fresh sample S3a). The small increase of particles in the range of 1-2 nm is explained by the coalescence of Ag clusters still present after the calcination process. [19] Some of the Ag nanoparticles sintered, which is in line with the low Tammann temperature (< 100°C ) of unsupported 2 nm Ag nanoparticles, however, since the sintering of the Ag nanoparticles lead to a loss in activity without influencing the S(EO) ( Figure S5a 210°C) the insignificant change of the PS distribution is | 3,069,965 | 216166674 | 0 | 16 |
excluded as source of the activation process. [30] This is supported by the loss in S(EO) at 230°C ( Figure S3c) demonstrating that sintering and S(EO) are decoupled phenomena. Since we recently demonstrated that smaller Ag particles exhibit a stronger oxygen interaction, [19] it is reasonable to interpret the activation phase of the Ag5/SiO 2 À N 2 catalyst as being influenced by its AgÀ O chemistry. To gain further experimental evidence, the Ag5/SiO 2 catalyst was pre-treated in synthetic air (Ag5/SiO 2 À O 2 ) before testing. We exclude sintering effects by synthetic air treatments since the catalyst was already calcined at 600°C for 1 h. Figure 3 shows the direct comparison of Ag5/SiO 2 À N 2 (blue) and Ag5/SiO 2 À O 2 (green) catalysts. The Ag5/SiO 2 À O 2 catalyst shows poor S(EO) and X(O 2 ) at low temperatures. In comparison to the Ag5/SiO 2 À N 2 catalyst, its activation phase is more pronounced and prolonged. The completely activated state is not reached until 190°C (maximum in S(EO) of 61 %) and generally the Ag5/SiO 2 À O 2 catalyst Figure S2 for EDX). STEM ADF of Ag5/SiO 2 (C). The bright spots refer to Ag particles. For catalyst Ag5/SiO 2 also the particle size distribution is shown (D). is less active (i. e. at 200°C same S(EO) but only 50 % of the conversion) than its Ag5/SiO 2 À N 2 counterpart. However, upon increasing the temperature, the Ag5/SiO 2 À O 2 catalyst still reaches the | 3,069,966 | 216166674 | 0 | 16 |
typical S-X behavior (green line) seen for the other catalysts. In summary, the SiO 2 supported Ag nanoparticles show, after a distinct activation period, a promising catalytic performance in comparison to the Ag15/α-Al 2 O 3 reference. Since the increase in S(EO), and generally the different activation behavior of the catalysts, is not a sintering effect, the AgÀ O interaction, respectively stored oxygen in/on Ag, might be responsible for the observed effects. This is supported by studies which show that the pre-treatment also pre-determines the existence of highly stable oxygen species in/on the Ag nanoparticles difficult to allocate for oxidation reactions. [12][13]31] The existence of highly temperature stable oxygen species for Ag nanoparticles below 6 nm [19] might also influence the catalytic performance and be responsible for the diverse picture of catalytic activities in oxidation reactions as illustrated in Scheme 1. Impact of the Ag-oxygen interaction To have access to the oxygen concentration in/on Ag and the corresponding Ag-oxygen interaction, thermal desorption spectroscopy (O 2 -TDS) was applied. A dedicated setup for powder samples was used, which allows a sample pretreatment under 1 bar and the subsequent desorption experiment at 10 À 6 mbar, as bridge between conventional TPD [32] (desorption in to inert gas) and surface science related TDS [33] experiments. The calcined catalysts were in-situ pretreated under Ar (Ag5/SiO 2 À Ar 12 h at 210°C, analogue to Ag5/SiO 2 À N 2 ) or synthetic air (12 h at 210°C) and compared to the Ag15/α-Al 2 O 3 reference, pretreated under synthetic air | 3,069,967 | 216166674 | 0 | 16 |
as well (Figure 4a). The desorbed oxygen signal is normalized to the sample weight. For Ag5/ SiO 2 À O 2 , two regions in which oxygen desorbs can be clearly distinguished, with region I (200°C-500°C) having a maximum at 350°C and region II appearing for T > 540°C (where Ag nanoparticles already start to melt). The temperature range for region I is consistent with the oxygen present in surface reconstructions on low miller index surfaces (e. g. nucleophilic oxygen) known to participate in combustion in model studies. [34] From desorption temperature alone, we cannot determine if the oxygen desorbing in region I is also associated with oxygen on the unreconstructed surface, a species that may participate in EO formation. [35] The desorption temperature for region II is in line with dissolved oxygen and/or electrophilic oxygen participating in EO formation during temperature programmed reaction. [24,36] However, since the mentioned oxygen species were identified on model systems, their roles and relevance for supported Ag catalysts remains rather speculative and will not be further discussed. After pre-treatment in Ar the amount of O 2 desorption in region I is reduced by 54 %. Region II, however, stays almost unaffected, emphasized by the converging amount of desorbed oxygen at around 620°C. These different oxygen desorption signals indicate a critical correlation between the differences in activation behavior seen for the different pre-treatments, Ag5/ SiO 2 À N 2 and Ag5/SiO 2 À O 2 . The species desorbed in region I, with a high population on Ag5/SiO 2 À O | 3,069,968 | 216166674 | 0 | 16 |
2 , likely reacts with ethylene to contribute to the unselective total oxidation reaction to CO 2 (see CO 2 formation rates Figure S5). Further, this ChemCatChem Full Papers doi.org/10.1002/cctc.202000035 might be the origin of a possible blocking of the reaction sites and explain the lowered X(O 2 ). The oxygen species desorbing in region II are not affected by the Ar pre-treatments, consistent with its assignment to dissolved oxygen. [24,36] The O 2 desorption behavior of the Ag15/α-Al 2 O 3 À O 2 reference (pre-treated in synthetic air) is entirely different to the Ag5/SiO 2 À O 2 sample. The temperature of the maximum in oxygen desorption is shifted by~200°C to lower temperatures (T max = 165°C) and a second desorption event is located a high temperatures (likely related to dissolved oxygen species). Besides, the quantity of desorbed oxygen is significantly decreased for Ag15/α-Al 2 O 3 À O 2 (2.49 μmol(O)/g cat ) compared to Ag5/SiO 2 À O 2 (14.64 μmol(O)/ g cat ). Calculating the Ag : O ratio (based on the Ag loading of the samples) as indication for the oxygen concentration in/on Ag, the Ag nanoparticles are significantly enriched in oxygen by a factor of~20 (Ag : O = 30 vs. 600 for Ag15/α-Al 2 O 3 , only the first desorption events are integrated, for Ag5/SiO 2 range of~200-500°C and for Ag15/α-Al 2 O 3 the range of~100-300°C). Since the samples were pre-treated identically (saturated with oxygen at 210°C for 12 h) and the heating rates were also | 3,069,969 | 216166674 | 0 | 16 |
the same (25°C/min), a shift of the desorption temperature is interpreted as a change of the electronic structure of the Ag and the strength of the AgÀ O interaction, respectively. This is similar to TPD studies on supported Au samples, which also identified a size dependent desorption energy excluding support effects. [37] However, based on the different quantities of the desorption (Ag : O ratios, as a consequence of the stronger Ag : O interaction and the high dispersion on SiO 2 ) and the shift in the T max , the surface or subsurface location of the oxygen cannot be deduced. To aid in the assignment of the oxygen species observed in TDS, isotope exchange experiments with 18 O 2 were conducted inside the TDS setup. After pre-treating the Ag15/α-Al 2 O 3 and Ag5/SiO 2 catalysts for 12 h at 210°C in synthetic air, the exposure times of 18 O 2 was adjusted to 0 min, 10 min and 60 min (also at 210°C). Figure 4(b) and 4(c) show the desorption signal of m/z = 34 for Ag15/α-Al 2 O 3 and Ag5/SiO 2 , which represents the mixed labeled oxygen isotopomer ( 16 O 18 O). As a function of time, the mixed labeled oxygen increases (no change of m/z = 36, see also Figure S6), which implies that the oxygen (here: 18 O 2 ) is dissociatively activated, accumulating as atomic oxygen in/on Ag. The recombination of stored ( 16 O) and exchanged ( 18 O) oxygen corresponds to a second order | 3,069,970 | 216166674 | 0 | 16 |
desorption kinetic, [33c] independent of the Ag particles size and desorption temperature. The shift of the T max with an increased saturation of labeled oxygen to higher (Ag15/α-Al 2 O 3 , Figure 4b) and lower (Ag5/SiO 2 , Figure 4c) temperatures represents attractive (higher T) and repulsive (lower T) interactions of the adsorbed species. [33b] This seems to be conclusive since the oxygen concentration in/on Ag5/SiO 2 is significantly higher. Besides, the labeled 18 O oxygen species are also desorbing in the high temperature regions as mixed labeled 16,18 O 2 , which are likely related to subsurface or dissolved oxygen species. Since the exchange of oxygen was conducted at 210°C, diffusion processes are very likely involved. To elucidate the accessibility of the atomic oxygen species in/on Ag, a second experiment with C 18 O 2 was conducted. Such a test can efficiently identify the surface atomic oxygen species (all known forms of adsorbed atomic oxygen readily react with CO x to form surface carbonates), while dissolved and electrophilic oxygen has low propensity to form carbonates (Scheme 2). [38][39] To avoid the issue of carbonate formation and subsequent isotope mixing on the surface by contact to the environment, the precursor samples (e. g. AgNO 3 /SiO 2 ) were first in-situ calcined in synthetic air. The in-situ generated Ag5/ SiO 2 À O 2 and Ag15/α-Al 2 O 3 À O 2 catalysts were subsequently exposed to only C 18 O 2 at 40°C for 30 min to saturate the surface. While oxygen is able | 3,069,971 | 216166674 | 0 | 16 |
to migrate/diffuse into the TDS accessible sub-surface region of Ag during the in-situ generation step, the low temperature C 18 O 2 exposure ensures the surface limited carbonate formation cannot access these buried species. [ 16 O will be seen during TDS (e. g. 1 : 2 for tridentate or 1 : 1 for bidentate carbonate), however, since different types of carbonates might contribute a rather qualitative statement seems reliable. [33a] Conversely, if free surface is available or the oxygen is not accessible, C 18 O 2 will adsorb and desorb as m/z = 48 in TDS. [40] The dominant desorption species for Ag5/SiO 2 (Figure 5b) is identified as C 18 O 2 with a distinctly smaller desorption event for C 18 O 16 O both appearing at a desorption temperature in the range of surface carbonates or weakly bound CO 2 . [39b, [40][41] The fraction of oxygen species able to interact with adsorbed C 18 O 2 seems to be insignificant, also confirmed by reference measurements with pure C 18 O 2 . This means the stored oxygen in/on Ag, which forms under reaction conditions unselectively CO 2 , is due to the strong AgÀ O interaction of the nanoparticles unable to form carbonates. Saturating the surface of Ag15/α-Al 2 O 3 with C 18 O 2 offers a different picture (Figure 5a). At low temperatures again C 18 O 2 is mainly detected, but at elevated temperatures the amount of mixed labeled C 16,18 O 2 increases significantly. This is a clear | 3,069,972 | 216166674 | 0 | 16 |
indication that oxygen stored in the Ag15/α-Al 2 O 3 sample is accessible and, at least in parts, located on the surface. The quantities of the desorbed CO 2 traces (m/z = 46 and 48) was nearly equal for both catalysts, which is line with our interpretation that the amount of oxygen stored in/on Ag5/SiO 2 Scheme 2. Possible surface Ag 2 CO 3 formation on Ag nanoparticles via terminal or sub-surface oxygen. In the case of sub-surface oxygen C 18 O 2 might adsorb on the Ag surface and desorb also as C 18 O 2 . is not accessible for carbonate formation under the selected conditions. To test the necessity of stored oxygen in epoxidation, the influence of oxygen on the catalytic performance was further evaluated by TDS experiments simulating the activation period. The in-situ created Ag5/SiO 2 À O 2 catalyst was tested at 1 bar and reaction feed (C 2 H 4 : O 2 = 5 : 1) for 4 h at 230°C in the TDS setup ( Figure S7a). After reaching a steady state conversion, the resulting O 2 -TDS spectrum was recorded ( Figure 6). A significantly reduced amount of oxygen desorbed from the catalyst in region I (green and red curves Figure 6). In addition, a completely activated Ag5/SiO 2 catalyst tested under 17.5 bar was transferred to the TDS setup and an oxygen desorption experiment was conducted. The corresponding O 2 desorption signal is negligible, demonstrating a correlation between the selective catalytic performance and the oxygen poor | 3,069,973 | 216166674 | 0 | 16 |
state. The same behavior is observed for the Ag15/α-Al 2 O 3 sample, after in-situ activation under reaction feed (Figure S7b + c and Figure S8). Generally, activating a supported Ag catalyst, independent of the respective particle size, leads in the TDS accessible range to an oxygen poor and selective catalyst. The consecutive removal of the oxygen species (at first completely in region I where nucleophilic oxygen, and afterwards incompletely in region II where dissolved and/or electrophilic oxygen might be located) is tentatively interpreted as evidence for the important role of dissolved oxygen species in product formation. [15] To confirm the non-reversible character of the activation period, in particular for Ag nanoparticles, an increasing/decreasing/ increasing temperature testing protocol should lead to different catalytic performances in terms of activity and S(EO) for Ag5/ SiO 2 À N 2 . Figure S9a shows the results of the applied program. Within the first increasing branch of the temperature the development of the S(EO) is distinctly visible, starting at 150°C with around 20 % and finally reaching 58 %. The second increasing branch (ca. 25 to 35 h TOS) shows only minor changes in terms of S(EO) (58 to 62 %) as a function of the temperature, but significantly higher formation rates for EO than CO 2 . Since 170°C was selected three times as a reaction temperature, Figure S9b highlights the changes as a function of time. The continuously decreasing CO 2 formation rates and stable EO formation rates lead to an increased S(EO). This serves as (another) experimental | 3,069,974 | 216166674 | 0 | 16 |
evidence that activation means nonreversible consumption of unselective oxygen species towards a selective, oxygen poor state of Ag. Due to the insulating character of supported Ag particles and the oxygen of the supports (SiO 2 , α-Al 2 O 3 ), a discrimination in electrophilic or nucleophilic oxygen species was not possible. However, based on the discussed oxygen poor character of the Ag particles in the selective state, the concentration of atomic oxygen species on the surface and its stability seems to be critical. To elucidate if any oxygen, stored in/on Ag, is needed at all to activate the reactants, microcalorimetry is applied (Figure 7). Prior to a surface titration at 230°C with C 2 H 4 and O 2 , the oxygen in/on the supported Ag catalysts is removed under H 2 atmosphere at 400°C ("chemically cleaned"). Subsequent dosing of C 2 H 4 showed no measureable ΔH ads (differential heat of adsorption). Adding O 2 resulted in ΔH ads values between 55 kJ · mol À 1 (Ag5/ SiO 2 ) and 18 kJ · mol À 1 (Ag15/α-Al 2 O 3 ), still in the range where molecularly adsorbed O 2 is adsorbed. [19] In an attempt to dissolve/store again traces of oxygen in/on Ag, both samples were treated in O 2 at 350°C (above the Tamman temperature of Ag). Dosing again O 2 leads to significantly increased ΔH ads of 110 kJ · mol À 1 (Ag5/SiO 2 ) and 65 kJ · mol À 1 (Ag15/α-Al 2 O 3 ), in | 3,069,975 | 216166674 | 0 | 16 |
the range where O 2 is activated dissociatively. [19] This is interpreted as a successful incorporation of oxygen in/on Ag by high temperature activation (350°C) as responsible for the O 2 activation at reaction conditions. The ΔH ads for Ag nanoparticles is increased following its stronger AgÀ O interaction, interpreted as PS dependent oxophilicity. When titrating with C 2 H 4 again, an enormous heat evolution due to a chemical reaction with the stored atomic oxygen is observed (> 300 kJ · mol À 1 ). Obviously, a small quantity of dissolved oxygen appears to be enough to activate [24,36a] O 2 and to form Ag δ + O x to facilitate ethylene adsorption/reaction. [42] As a consequence, purely metallic Ag 0 (without Ag d-/O p-hybridized states, "closed dband") [43] is unable to activate the reactants. [34b] In turn, the oxygen traces needed for an active Ag δ + O x are within the detection limit/resolution of the TDS setup. [35d] Under the assumption that the surface sites quantified by ethylene adsorption (30.8 μmol/g cat ) [19] and the quantified atomic oxygen amount desorbing in region I (Figure 4a, 14.64 μmol/g cat ) is solely present on the surface, a coverage in the range of 0.5 monolayers (ML) is estimated. In the context of the almost absent oxygen desorption signal after reaching a selective state (Figure 6), the surface coverage is in the range of < 0.005 ML (< 1 % of the starting value, 0.07 μmol/g cat ). Kinetic Evaluation Based on the discussed catalytic | 3,069,976 | 216166674 | 0 | 16 |
investigation a kinetic analysis with respect to Arrhenius plots and corresponding apparent activation energies (E A ) of EO and CO 2 formation was conducted ( Figure 8). Figure 8(a) shows the E A of Ag5/SiO 2 and Ag15/α-Al 2 O 3 after reaching steady state ( Figure S4, 30-60 h TOS; Figure S9, 25-35 h TOS). The E A for EO are very similar (Ag15/α-Al 2 O 3 : 98 � 1 kJ · mol À 1 and Ag5/SiO 2 : 96 � 1 kJ · mol À 1 ) evidencing that the nature of the active Ag and the formation mechanism is identical independent of the Ag PS. The E A for CO 2 of the reference system is slightly higher than for the Ag nanoparticles (Ag15/α-Al 2 O 3 : 114 � 3 kJ · mol À 1 and Ag5/SiO 2 : 103 � 5 kJ · mol À 1 ), which might explain the lower S(EO) reasoned in the stronger AgÀ O interaction leading locally to an unwanted oxygen enrichment. This interpretation is supported by Figure 8(b) showing the E A of Ag5/SiO 2 À N 2 during and after the activation period. The E A for EO is identical to the values obtained from the activated sample (98 � 2 kJ · mol À 1 ). The E A for CO 2 is different for the activation phase (48 � 4 kJ · mol À 1 ) and the activated sample (100 � 2 kJ · mol À 1 ) extracted | 3,069,977 | 216166674 | 0 | 16 |
from two linear fits. This means, in the low temperature regime (160-180°C) when the sample is still rich in oxygen and unselective, the E A for CO 2 is significantly decreased and in the high temperature regime close to the steady state values (180-200°C) almost doubled. Since the ln(rate) of the EO formation follows during the entire temperature range a linear behavior, the unselective sites are removed during activation without their transformation into selective ones ( Figure S5). As a consequence, ethylene adsorbing on an oxygen enriched surface reacts unselectively to CO 2 (or to acetaldehyde and then CO 2 ) but on a surface poor in oxygen not necessarily to EO. This Figure 7. Microcalorimetry study of Ag5/SiO 2 and Ag15/α-Al 2 O 3 to probe the ability in activating O 2 and C 2 H 4 at 230°C after different pretreatment conditions (400°C in H 2 and 350°C in O 2 ). The initial differential heat of adsorption values are given in kJ · mol À 1 . Figure 8. Arrhenius plots and the resulting apparent activation energies (E A ) for EO and CO 2 formation on catalysts Ag5/SiO 2 and Ag15/α-Al 2 O 3 after reaching steady state (a). Arrhenius plots and the resulting E A for EO and CO 2 formation of Ag5/SiO 2 À N 2 during the activation phase and afterwards (b). evidences that different oxygen species are involved. Further, the selective transfer of an oxygen atom is part of the rate determining step. If the selective oxygen stems | 3,069,978 | 216166674 | 0 | 16 |
from an unreconstructed Ag surface, the sub-surface regime or is part of a SO x intermediate is beyond the scope of this study. However, calculating the turn over frequency (TOF, based on the ethylene adsorption capacities [19] ) for both catalysts in the selective regime at almost identical X(O 2 ) (Ag5/SiO 2 À N 2 at 190°C: X(O 2 ) = 11.80 and for Ag15/α-Al 2 O 3 at 170°C: X(O 2 ) = 11.54, see also Figure 3), the TOF for Ag15/α-Al 2 O 3 is almost three-times higher compared to Ag5/SiO 2 (for the EO and CO 2 formation, see Figure S10). In light of the oxygen poor Ag surfaces in the selective state, these results might be explained by different strength of the AgÀ O interaction leading to a decreased TOFs for the Ag nanoparticles independent of the reaction path (EO or CO 2 formation). This is generally named as particle size effect. A DFT perspective At first glance this need for low atomic oxygen concentration is at odds with older proposals suggesting surface oxides are needed for EO production. [16,36a,44] The reasons for the requirement of low atomic oxygen concentration can, however, be seen when considering how it reacts with adsorbed ethylene as a function of oxygen coverage. To do so it is helpful to begin by examining the high coverage limit of oxygen. The reaction between ethylene and atomic oxygen is often thought to proceed through an oxometallacycle (OMC) intermediate, [35a,44b] where the OC 2 H 4 fragment is bound | 3,069,979 | 216166674 | 0 | 16 |
to the surface through an AgÀ O and an AgÀ C bond ( Figure S11). In the OMC mechanism ethylene reacts with O to form an OMC, which then decomposes into EO or AcH (acetaldehyde). As AcH rapidly burns on Ag surfaces, the branching ratio to the two products places an upper limit on EO selectivity. On the bulk oxide surface the OMC mechanism is predicted to appear once surface O vacancies are present, and is predicted to favor AcH production by 31 kJ · mol -1 , [44b] making oxygen on the reduced oxide surface selective to AcH, and hence CO 2 , through the OMC mechanism. Similar behavior has been seen on the oxygen induced surface reconstructions (nucleophilic oxygen) that can form at the oxygen chemical potentials relevant for ethylene epoxidation. [34b,35b,c] In particular, both the oxygen reconstructed Ag(110) [34] and Ag(111) [34b] surfaces are found to selectively produce AcH/CO 2 . Thus, the (nucleophilic) oxygen concentration should be minimized to achieve high S(EO). On unreconstructed surfaces, adsorbed oxygen cab form EO at a comparable rate to AcH. [44b] Such a phase is also pertinent to catalysis; a low coverage of oxygen has been observed on unreconstructed Ag surfaces, again, at oxygen chemical potentials relevant for ethylene epoxidation. [35c,d] The unreconstructed atomic oxygen phase also likely plays a role in mediating the coverage of electrophilic oxygen during catalysis. [34b] Thus, it is critical to understand how atomic oxygen on the surface should be managed to achieve high S(EO). How the chemistry of this phase | 3,069,980 | 216166674 | 0 | 16 |
changes as a function of coverage is, however, unclear. To investigate the role of oxygen coverage on the OMC branching ratio we performed a series a DFT calculations including the exchange-hole dipole moment model for dispersion. [23] Here we examined OMC formation and its subsequent decomposition into AcH/EO using a (3 × 3) surface unit cell of Ag(111) with 1-4 oxygen adatoms on the surface, 1/9-4/9 mononlayer coverage (ML). (For completeness a (4 × 4) cell with 2/16 ML oxygen was also included.) The lower coverage range is near that observed on Ag surfaces, [35c,d] while the higher coverage is in the range of the maximum oxygen coverage seen in this work (~0.5 ML). In addition, kinetic Monte-Carlo modelling on the surface coverage of oxygen on Ag (111) identified a pressure dependent phase with a low coverage of oxygen < 0.05 ML. [45] Figure 9 shows the activation energies to EO and AcH computed on these surfaces plotted as a function of the corresponding heats of reaction. The stability of the OMC intermediate is interpreted as descriptor for selectivity towards EO. [46] Inspection of Figure 9 reveals the Bell-Evans-Polanyi principle holds for the OMC decomposition; the activation energy scales with the heat of reaction. Moreover, the activation energies also scale with the coverage of adsorbed oxygen, with the low coverage cases having nearly equivalent activation energies to EO and AcH. Conversely, at 4/9 ML oxygen coverage, approximately the maximum seen in this work, the activation energy to AcH is 15 kJ · mol À 1 below | 3,069,981 | 216166674 | 0 | 16 |
that to EO. This finding shows that, regardless of its nature, adsorbed atomic oxygen tends to favor AcH/CO 2 production at high oxygen coverage. This behavior can be rationalized by considering that the branching ratio in the OMC mechanism is mediated by the relative strength of its CÀ Ag and OÀ Ag bonds, with decomposition to EO favored by an increase in OÀ Ag bond strength relative to CÀ Ag. [47] Increasing the surface oxygen coverage will increase the amount of Ag δ + and O adatom repulsion, [43b] which will favor AcH production. This repulsive behavior is directly evidenced for the Ag5/SiO 2 nanoparticles by the TDS 18 O 2 exchange experiments. In short, the activation barriers calculated for the formation of EO vs. AcH as function of the oxygen coverage on the Ag surface gives a direct explanation for the experimental findings: increasing the concentration of atomic oxygen adsorbed on Ag reduces EO selectivity (Figure 3, 6 and 7) regardless of whether it is present in an unreconstructed adatom phase or as nucleophilic oxygen. Conclusion Based on the advanced synthesis of supported Ag nanoparticles < 6 nm with a narrow size distribution (Ag5/SiO 2 model catalysts), [19] we demonstrate its activity and selectivity in the ethylene epoxidation under industrially relevant conditions. The high strength of the AgÀ O interaction allowed a resolution of the activation period and the identification of the selective state of Ag. The concept of an oxygen poor and selective state of Ag is independent of the Ag particle size and | 3,069,982 | 216166674 | 0 | 16 |
transferable to a Ag15/α-Al 2 O 3 reference system. Complementary TDS experiments of Ag5/SiO 2 and Ag15/α-Al 2 O 3 extracted the differences in AgÀ O interaction also with respect to quantity and accessibility. The Ag nanoparticles exhibit a high strength in AgÀ O interaction, coupled to a comparably high oxygen concentration and low accessibility. These findings make the transformation of Ag nanoparticles into a selective catalyst challenging and serve as explanation on the diverse performances discussed in the literature. The present findings correlate well with a suite of spectroscopic studies aimed at finding the oxygen species responsible for selective oxidation. The standard picture on the roles of nucleophilic (combustion) and electrophilic [48] (selective) oxygen is oversimplified, since under reaction conditions nucleophilic oxygen does not appear to be necessary. Adsorbed atomic oxygen, as shown by isotope exchange experiments, can easily migrate into the sub-surface region [49] where it can modify the electronic structure of Ag. As a consequence, a key to a selective state of the catalyst is to minimize the atomic oxygen on the surface. In industrially applied systems this is realized by promoters. [50] Due to the strong AgÀ O interaction of nanoparticles (higher oxophilicity), not only the activation phase is intriguing, but also the oxygen poor Ag surface might accumulate locally an excess of adsorbed oxygen leading to a lowered S(EO) and E A of CO 2 . Further, the TOFs for the EO and CO 2 formation of the Ag nanoparticles are negatively affected (~1/3 compared to Ag15/α-Al 2 O 3 ) | 3,069,983 | 216166674 | 0 | 16 |
and a distinct PS effect is evidenced. In summary, the concept of highly dispersed nanoparticles creating more active sites is, particularly for Ag, however, not without pitfalls. The design challenge remains in the transformation of reactive sites into selective ones. The present case illustrates how the reaction of nanostructures with unexpected stability parameters may override the positive effect of higher dispersion. The intriguing AgÀ O chemistry in combination with a PS dependent activation led to the controversially discussed role of Ag nanoparticles and a PS effect, which has now been understood and clarified. | 3,069,984 | 216166674 | 0 | 16 |
Chronic disease risk factors associated with health service use in the elderly Background To examine the association between number and combination of chronic disease risk factors on health service use. Methods Data from the 1995 Nova Scotia Health Survey (n = 2,653) was linked to provincial health services administrative databases. Multivariate regression models were developed that included important interactions between risk factors and were stratified by sex and at age 50. Negative-binomial regression models were estimated using generalized estimating equations assuming an autoregressive covariance structure. Results As the number of chronic disease risk factors increased so did the number of annual general practitioner visits, specialist visits and days spent in hospital in people aged 50 and older. This was not seen among individuals under age 50. Comparison of smokers, people with high blood pressure and people with high cholesterol showed no significantly different impact on health service use. Conclusion As the number of chronic disease risk factors increased so did health service use among individuals over age 50 but risk factor combination had no impact. Background Chronic diseases are responsible for the majority of deaths worldwide [1]. The baby-boomer generation makes up a large portion of the North American population, and there will likely be an increase in absolute numbers of heart attacks and strokes as this group reaches the age traditionally associated with the onset of cardiovascular disease, even if the rates of these events decline [2][3][4]. Cardiovascular disease is not the only concern: other chronic diseases such as cancer, pulmonary diseases, diabetes and osteoporosis | 3,069,985 | 7344291 | 0 | 16 |
are likely to increase in number [5,6]. Risk factors such as smoking, high blood pressure, cholesterol, obesity, physical inactivity and diabetes are common to many of these chronic diseases. Primary prevention of even a few of these factors could result in an increase in disability-free years and ultimately, a reduction in health care costs [7][8][9]. Quantifying the impact of chronic disease risk factors on health care use can be used to estimate the return on investment of health promotion and other policies designed to prevent chronic disease. It also provides a powerful means of communicating the value of risk factor modification to the public and policy makers. However, estimating the effect of chronic disease risk factors on health service use is complex, and previous studies have been subject to some important limitations. In the absence of long-term longitudinal data to support a life course perspective, measuring the effects of risk factors within cohorts is difficult. There are tradeoffs between modelling the complexity of how risk factors interact, on the one hand, and conveying the results in a clear fashion on the other hand. As a result, risk factor measurement is often oversimplified, relying on the reporting of main effects only rather than incorporating higher order interactions to facilitate ease of interpretation. Such oversimplifications may bias results and underestimate the true impact of chronic disease risk factors on a variety of outcomes such as morbidity, mortality and health service use. The aim of this study was to measure the effect of chronic disease risk factors on health | 3,069,986 | 7344291 | 0 | 16 |
services utilization in Nova Scotia, Canada. The first objective of this study was to estimate the effects of number and combination of risk factors on health service utilization, as measured by family physician visits, specialist physician visits, and inpatient hospital bed days in different age groups of women and men. A second objective was to assess the relative importance of three major risk factors -smoking, high blood pressure and high cholesterol, on health care utilization. The approach we utilized overcomes the limitations noted by separating the task of modelling the relationship between risk factors and health services use from the task of summarizing, for communication purposes, variation in health services use by risk factor combination. This allows for a complicated modelling strategy to better describe the relationship between risk factor and service use with an accessible presentation of the results. Overview of study design A retrospective cohort study was conducted. Participants from the 1995 Nova Scotia Health Survey (NSHS) with information on risk factors were linked to provincial health services utilization data and followed prospectively for six years. Individuals were assessed based on their 1995 exposure to major risk factors for chronic disease. The outcome under study was health service utilization, as measured by the average annual number of GP visits, specialist visits and hospital bed days between the NSHS interview date in 1995 and March 31, 2001. The outcomes were compared for groups with varying numbers and combinations of risk factors. Sample The 1995 Nova Scotia Health Survey was a populationbased survey designed to provide | 3,069,987 | 7344291 | 0 | 16 |
a cross sectional picture of the health status, risk factors and preventive health practices of Nova Scotians [10]. In total, 5578 non-insti-tutionalized adults from all parts of the province were drawn from the provincial Medical Services Insurance database, a comprehensive list of all Nova Scotians eligible to use heath services. Canada has a comprehensive medical insurance system that provides all necessary medical services to all Canadians; as a result, the list of insured participants is considered to be a comprehensive registry of the population. From this sampling, 4360 people were located and 3227 participated. The current study was comprised of the 82% of survey respondents (n = 2658) who completed a clinic visit where blood pressure, BMI and blood cholesterol measurements were taken. Over 99% of all clinic respondents gave consent (n = 2653) to link their survey information with administrative health records. Weights applied to adjust for those we were unable to locate and propensity score weights adjusting for non-response showed no meaningful biases in cardiovascular risk factors among clinic participants compared to the general population [11]. A detailed methodology of the NSHS is described elsewhere [12]. Data The NSHS data was linked to provincial administrative health databases containing information on insured health services. Information was linked through the use of encrypted identification numbers to ensure confidentiality; no identifying information was available to the investigators. The study received ethics review and approval by the Dalhousie University Health Sciences Human Research Ethics Board. Two sources of health service information were utilized: the Hospital Discharge Abstract Database; | 3,069,988 | 7344291 | 0 | 16 |
and, the Medical Services Insurance (MSI) Physician Services File. The MSI Registry file was used to identify participants who died or migrated out of the province during the study period. Annual Hospital Bed Days were summed; individual hospital admissions with a length of stay longer than 60 days were excluded to eliminate hospitalizations that reflected patterns more similar to long term care. The MSI Physician Services File contained fee-for-service doctor visits, both to general practitioners and specialists. The number of visits by each participant each year was extracted. Chronic disease permeates several aspects of health service utilization and can be implicated in many diagnoses; therefore, all services for all ICD-9-CM diagnostic codes were included. This likely provided an overrepresentation of chronic disease health service use, however, the group with no risk factors provided baseline health service utilization unrelated to chronic disease against which all other risk factor profiles were compared. Measures Continuous measurement of variables such as blood pressure, cholesterol and obesity, annual GP visits, annual specialist visits and hospital bed days was used to enhance the ability of the statistical models to estimate the variance in utilization attributable to risk factors. Many of the variables in previous risk factor studies were aggregated into a few discrete categories rather than being left on an interval scale with many gradations. Unfortunately, categorization sacrifices the ability to measure the variation that is seen within each level of the variable. This variability may be crucial to predict service use or mortality outcomes precisely. Statistical Analysis Descriptive Statistics Descriptive statistics | 3,069,989 | 7344291 | 0 | 16 |
were calculated to characterize the distribution of risk factors in the sample population. Mean, median and percentage values were calculated. Complex Modelling The models employed person-years as the unit of analysis to measure the association between risk factors and health service utilization. Regressions models were estimated using generalized estimating equations (GEE); a first order autoregressive correlation matrix for errors was specified for the GEE procedure. The negative binomial distribution was selected for all outcomes, based on its fit with the observed distribution of health services use. The study participants were stratified into four categories by sex and age (dichotomized at 50 years). Final models were built for the 12 groups (3 outcomes × 4 age-sex strata) to best predict health service utilization in each stratum. We engaged in a process to build models that were parsimonious, yet sufficiently complex to capture the interactions between risk factors, and between risk factors and age. Initially, each of the 12 models was run with only main effects for risk factors, age and the socio-demographic variables. The relationship between the risk factors and outcomes is likely to vary with age, so interaction terms for each risk factor and age were included in the model. A number of two-way and three-way risk factor interaction terms, derived from previous reports of statistically significant interactions in chronic disease outcomes found in the literature, were also included [13][14][15][16][17][18]. Each interaction term was assessed for its contribution to the model based on the difference in Wald Chi-square statistic using a minimum significance level of 0.05. | 3,069,990 | 7344291 | 0 | 16 |
A set of regression models that described utilization as a function of risk factor combinations and age for each agesex stratum was developed in the modelling phase. The final models included two and three-way interaction terms between categorical and continuous variables. These models were different for each age-sex strata. Results Interpretation and Summary While the complex regression models are well suited for modelling variation in utilization associated with risk factors, they cannot be easily interpreted. The final step in the analysis focused on summarizing the results of the models in a meaningful way. An important methodological contribution of this study shows that modelling need not be compromised to facilitate the communication of results. All of the observations were divided into age and sex strata and the corresponding model developed for each stratum was estimated using the appropriate data. Predicted utilization estimates were obtained for each person-year observation and then summarized to obtain the mean estimated (predicted) health service use for patients with different numbers and combinations of risk factors. Possible confounding variables were held constant within each of the 12 strata to estimate only the effect of risk factors. These variables (age, geographic location, education, employment status and exposure) were held constant at the mean value for each stratum. For summary purposes, each participant was re-classified as high or low risk for each risk factor. The following cut points were used to designate study participants as high or low in risk status: • Hypertensive -mean systolic >= 140 mmHg or mean diastolic >= 90 mmHg. • | 3,069,991 | 7344291 | 0 | 16 |
High Cholesterol -total cholesterol >= 5.2 mmol/L • Smoker -currently smokes or quit <= 5 years prior to survey • Obesity -BMI >= 27 • Diabetes -Self-identified as having Type I or II diabetes and reported being seen by a medical professional about diabetes treatment • Physical Inactivity -Exercised less than three times per week. Each person year observation was evaluated based on the cut points outlined above and the predicted values were then averaged to obtain the mean number of annual doctor visits or annual hospital bed days for those individuals with 0, 1, 2, 3, and 4 + risk factors. Means were weighted to adjust for the complex sample design. Because age plays such a large role in the severity of chronic disease and level of service utilization, predicted values were generated holding age constant at 25, 33.3 (average age of under 50 cohort), 45, 55, 65.7 (average age of 50 & over cohort) & 75. Predicted data was not stratified by sex because similar utilization by risk factor combination was found between men and women. Individuals with 1, 2 and 3 risk factors were further stratified to assess which of the major risk factors traditionally examined in the cardiovascular and chronic disease literature (high blood pressure, high cholesterol, smoking) had the greatest impact on service utilization. This was done holding age constant only at 33.7 and 65.7 (the mean ages of the old and young strata). Software Data preparation was conducted using SAS Version 8. Descriptive statistics and the full analysis were conducted | 3,069,992 | 7344291 | 0 | 16 |
using STATA Version 8. Participant characteristics There were a total of 13676 person-years representing 2523 individuals included in all regressions. 130 individuals were removed from regressions due to missing data. Large differences were seen in prevalence of risk factors between age/sex strata, indicating that it was valuable to stratify by sex and at age 50 at the modelling stage (Table 1). Large variations in risk factor status and health service utilization were seen between men and women as well as between the young and old Regression Models Models were constructed for each outcome: number of annual GP visits, number of annual specialist visits, and number of annual hospital bed-days. The data indicated that risk factor synergies, whether they were interacting with other risk factors or age were not particularly important in predicting GP use. Age/risk interactions were more important than risk/risk interactions in predicting specialist visits but hospital visits were affected by both age/risk interactions and risk/risk interactions. The association between number of risk factors and utilization The number of chronic disease risk factors had an impact on health service use for individuals over the age 50, but not for those under age 50 (Table 2). This was true regardless of the type of health service use examined. Therefore, we focus our results with respect to number of risk factors on those over 50. There was a consistent upward trend in use as the number of risk factors increased at age 55, 65 and 75 for GP visits and after age 65 for hospital and specialist | 3,069,993 | 7344291 | 0 | 16 |
use ( Figure 1, Table 2). For example, individuals aged 55 with zero risk factors had an average of 5.67 (95%CI 5.37, 5.97) annual GP visits compared to 7.38 (95%CI 7.18, 7.59) annual visits in 55 year olds with four or more risk factors. The effect of risk factors on health service use in the '50 and over' population varied by age for hospital bed days and specialist visits but not for GP visits (Figure 1, Table 2). The slopes of the graphs of annual GP use at ages 55, 65 and 75 were similar indicating a proportionate rise in use of services regardless of age ( Figure 1 Table 2). Type of Risk Factors A comparison of the average number of doctor visits and hospital bed days for people with high blood pressure, high cholesterol or smoking in their risk profile showed that the specific combination of risk factors had little effect on service use (Table 3). Smoking, high blood pressure and high cholesterol status also had no effect on health service use in individuals less than 50 years. This result is not surprising since the number of risk factors did not predict health service use in individuals less than 50 years. Discussion Health service utilization was found to be dependent on the number of chronic disease risk factors only among individuals over the age of 50. People over the age of 50 with many risks factors (4+) went to the GP almost 30% more often than those with no risk factors; a similar result | 3,069,994 | 7344291 | 0 | 16 |
was found in annual specialist visits and hospital use. The effect of number of risk factors on specialist and hospital use was not seen until age 65. After age 75 the high-risk * Holding all study participants' age constant at the age specified, education, employment status and region constant at the sample proportions. Survey weights were applied to get weighted means and corrected standard errors group was found to have 40% higher specialist use and 350% higher hospital visits than the group with no risk factors. The finding that hospital use increased as the number of risk factors increased in older populations is consistent with previous work. Daviglus et al, found that costs related to hospital use increased significantly when comparing individuals with some risk to those with no risk [19]. Our sub-analysis by smoking, high blood pressure and cholesterol status indicated that the specific combination of risk factors was not important in predicting health service use. These results differed from Daviglus et al. who found that high blood pressure was less costly than smok-ing and high cholesterol. The differing results may be due to our inclusion of synergistic relationships between risk factors. Our method of separating service use modelling from results presentation may have allowed us to more accurately measure the impact of multiple risk factors present in an individual. Our approach suggests that it is the number of risks and not the type of risks that impact service use. Examination of the effect of risk factors on health service use under the age of | 3,069,995 | 7344291 | 0 | 16 |
50 revealed that the number of risk factors had no effect on the use of any type of health service. However, chronic disease manifests late in life after many years of exposure to risk factors. Due to this threshold effect, those with the highest risk profiles at younger ages are probably no more likely to have overt CVD, cancer or osteoporosis than those with no risks. It is, therefore, not surprising that young people with many risks had similar use to those with no risks. Surprisingly it was found that, for individuals under age 50, annual GP visits, specialist visits and, to some degree, hospital bed days decreased slightly as the number of risks increased. Regular visits to the GP often include messages of healthy lifestyle and primary prevention of disease. Our data could support two competing possibilities: more frequent visits to the GP encourage the development of fewer risk factors, or those with more risk factors are more likely to avoid their general practitioner. The models developed in this study cannot distinguish between these two explanations, and serves to highlight the possible endogenous relationship between risk factors and health service use that requires further exploration in future studies. Modelling three outcomes in two different sexes in two different age groups provided the opportunity to examine the risk factors that impact health service use in different age/sex groups of the population. The synergy of risk factors appeared to have the greatest effect on hospital use. That is, the largest number of risk factor interaction terms remained | 3,069,996 | 7344291 | 0 | 16 |
significant in the hospital use model compared to the other health services. Particularly in those aged 50 and older, many age by risk factor combinations as well as combinations of different risk factors had a significant association with hospital use. This was also seen, to a lesser degree, in hospital use in the younger age strata. Synergistic effects of risk factors by age and risk factor by risk factor also appeared to have an effect on specialist visits, much more so than general practitioner visits. In fact, there was no significant interaction term predicting general practitioner use in women aged 50 and older. It is not a surprising finding that hospital visits were more likely to be affected by risk factor interactions than doctor visits. GP visits are often routine check-ups and are just as likely to represent primary prevention of disease as opposed to Average Annual Predicted GP Visits by Age Figure 1 Average Annual Predicted GP Visits by Age. Limitations from this study may have arisen from the cross-sectional nature of the survey data and the selfreport of the variables. Measurement error may have been introduced with the self-reported variables such as diabetes [20]. However, the self-report limitation was not a factor in the measurement of BMI, cholesterol levels and blood pressure. These factors were measured in the clinic portion of the 95NSHS, which increased the accuracy of measurement of the level of risk in participants. Health services administrative data is more reliable source of utilization than self-reported service use [20]. It is important | 3,069,997 | 7344291 | 0 | 16 |
to note that standard errors and confidence intervals presented in the tables represent only the error from the variation between summarized predicted values. The independent variable coefficients or point estimates for each independent risk factor also include a standard error of the estimate, as do the mean values of systolic and diastolic blood pressure. The modelling/summarization approach allows for sophisticated modelling of interaction terms but sacrifices the ability to assess the additive effects of standard errors at multiple levels. Thus, confidence intervals and p-values were likely underestimated in the summarized results, and may not be appropriate for determining statistically significant differences between different numbers or combinations or risk factors. Conclusion The complex measurement and modelling presented in this study provides an example of how the complicated web of risk factors can be more accurately modelled in relation to health service use. Inclusion of six major risk factors for chronic diseases of the cardiovascular, respiratory, musculo-skeletal systems and cancer allows us to make broad inferences about the effect of chronic disease risk factors on health service utilization. The number of chronic disease risk factors had a big impact on health service utilization after the age of 50. As hypothesized, the *This was done holding all study participants' age constant at the stratum mean (65.7), education, employment status and region constant at the sample proportions. Survey weights were applied to get weighted means and corrected standard errors number of chronic disease risk factors was a better predictor of health service use than the combination of risk factors. Smoking, | 3,069,998 | 7344291 | 0 | 16 |
high blood pressure and high cholesterol had equal effects on all types of service use at all ages. This result differs from previous work perhaps because of the increased precision of modelling risk factors level and risk factor synergies. These results suggest that primary prevention of chronic disease risk factors should be conducted at a population level if we are ever to hope to reduce consumption of health service use. Although rates of risk factors, such as smoking, have decreased in recent years in young people there are still relatively high rates in the young adult population [21]. These will need to be addressed in order to stem the increase of chronic disease in the future. Given the age of the data, more than 10 years have passed since the time of the 95NSHS and chronic disease rates have been rising in that time frame. Thus, the impact on health service use today is a conservative estimate, and is likely to be greater than is reflected in these data. Central to improving the health of the population is assessment of our government policy initiatives for their impact on health. There is a need to develop long-term health plans that will be unaffected by budget cycles and high turnover rates. Policy changes are needed to reduce the prevalence of chronic disease rather than placing the responsibility solely on the individual. | 3,069,999 | 7344291 | 0 | 16 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.