Cardiac Rehabilitation Programs and their Core Components for Coronary Heart Disease: A Systematic Review and Network Meta-Analysis

Transcription

1 Cardiac Rehabilitation Programs and their Core Components for Coronary Heart Disease: A Systematic Review and Network Meta-Analysis by Nader N. Kabboul A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy, Graduate Department of Pharmaceutical Sciences, University of Toronto Copyright by Nader N. Kabboul «2016»

2 Cardiac Rehabilitation Programs and their Core Components for Coronary Heart Disease: A Systematic Review and Network Meta-Analysis Abstract Nader N. Kabboul Doctor of Philosophy Graduate Department of Pharmaceutical Sciences University of Toronto «2016» Background Although earlier reviews establish that patients with coronary heart disease (CHD) respond well to cardiac rehabilitation (CR), the comparative effectiveness of different CR programs or their individual core components of CR have not been evaluated. Methods A systematic review and network meta-analysis (NMA) of randomized controlled trials (RCTs) evaluating the core components of CR (nutritional counseling (NC), risk factor modification (RFM), psychosocial management (PM), patient education (PE), and exercise training (ET)) was undertaken. The core components were evaluated individually and as part of broader CR treatment categories ( exercise-only CR programs, comprehensive CR programs, and secondary prevention programs without exercise ). Published RCTs were identified from database inception dates (Medline, Medline In-Process, the Cochrane database, Embase, CINAHL, Sci-Expanded, PsychINFO, the Web of Science) to July Endpoints included measures of mortality (all-cause and CV) and morbidity (myocardial infarction (MI), revascularization, and hospitalization). Hazard ratios (HR) and 95% credible intervals (CrIs) were used as summary measures. ii

3 Findings 136 trials (50,054 participants) and 169 trials (62,149 participants) were included for the NMAs of the overall CR treatment strategies and for the core components of CR respectively. Exercise-only and comprehensive CR programs significantly reduced the hazards of all-cause mortality, CV mortality, total MI, fatal MI, non-fatal MI, revascularization, total and CV hospitalization by 25-54% when compared to usual care, and significantly reduced the hazard of all-cause mortality by 23%-26% when compared to secondary prevention programs without exercise. Secondary prevention programs without exercise significantly reduced the hazards of all-cause and CV mortality, total MI, non-fatal MI, all-cause and CV hospitalization by 20%-43% when compared to usual care. The core component of PM significantly reduced the hazards of all-cause mortality, all-cause and CV hospitalization by 31%, 33% and 50% respectively. RFM and PE significantly reduced the hazards of total MI and all-cause hospitalization by 37% and 23% respectively. Interpretation These findings confirm the central role of CR in patients with CHD, emphasize the central role of exercise training in CR, elevate the role of other key CR core components, especially psychosocial management, and establish the superiority of exercise-based CR programs to secondary prevention programs without exercise. iii

4 Acknowledgments I acknowledge support from the University of Toronto, Department of Pharmaceutical Sciences, the Toronto Rehabilitation Institute (TRI), the Institute for Health Policy, Management and Evaluation (HPME) and the Toronto Health Economics and Technology Assessment Collaborative (THETA). The opinions, results and conclusions reported in this paper are those of the authors and are independent from these organizations. As such, no endorsement by these organizations is intended or should be inferred. I sincerely thank Dr. Murray Krahn and Dr. George Tomlinson for their support, encouragement, trust and inspiration, without which this work would not have been possible. I also acknowledge the support that I received from Dr. Sherry Grace, Dr. David Alter, Dr. Jeffrey Hoch, Dr. David Naimark, and Dr. Harindra Wijeysundera. I especially thank Troy Francis for his support, dedication, perseverance, and contributions to this paper, without which this work would not have been possible. Last, but definitely not least, I thank my wife, Tamara Daou-Kabboul, my daughter, Maya Noelle Kabboul, my mother Nawal Kabboul, and my mother-in-law, Nawal Daou, for their patience, support, understanding and encouragement, without which this work would not be possible. I also acknowledge the critical academic contributions from my wife to this paper. iv

5 Table of Contents Acknowledgments... iv List of Tables... ix List of Figures... x List of Appendices... xi 1 Introduction Coronary Heart Disease (CHD) Risk Factors for CHD Cardiac Rehabilitation (CR) for CHD The Core Components of CR for CHD More Recent Findings for CR in CHD Heterogeneity and Complexity: Reasons for the Demise of CR? Fortunately, CR is not Alone As a Complex Intervention These Problems Are Inherent to Using Traditional Statistical Approaches to Evaluate Complex Interventions New Statistical Approaches for Complex Interventions Applying New Statistical Approaches in Reviews of CR Cardiac Rehabilitation Programs and their Core Components for Coronary Heart Disease: A Study Protocol for a Systematic Review and Network Meta-Analysis Objectives Methods Types of studies Types of participants Definition of Cardiac Rehabilitation and its Core Components Types of interventions: Core Components of Cardiac Rehabilitation Types of outcome measures Electronic searches Searching other resources Study Selection, Data Collection and Analysis Selection of Studies Data extraction and management Assessment of risk of bias in included studies v

6 2.3.4 Missing data Assessment of reporting biases Data synthesis Statistical Analysis Assessment of Heterogeneity, Inconsistency and Model Fit Model 1 Direct Head-to-Head Comparison Model 2 Network Meta-Analysis Model 3 Network Meta-Analysis ( Welton Method ) Comparative Effectiveness of Exercise- Only and Comprehensive Cardiac Rehabilitation Programs on Mortality and Morbidity: A Systematic Review and Network Meta- Analysis Introduction Methods Search strategy and selection criteria Inclusion and exclusion criteria Study selection Data extraction process and quality assessment Data synthesis and analysis Results Network Meta- Analysis Effect of CR Strategies Discussion The Comparative Effectiveness of the Core Components of Cardiac Rehabilitation on Mortality and Morbidity: A Systematic Review and Network Meta- Analysis Introduction Methods Search strategy and data sources Inclusion and exclusion criteria Study selection Data extraction process and quality assessment Data synthesis and analysis Results Network Meta- Analysis Effects of Core CR Components Discussion vi

7 4.4.1 Implications Limitations The Cost- Effectiveness of the Core Components of Cardiac Rehabilitation in Coronary Heart Disease: A Study Protocol Background Statement of Problem and Objectives Study Design and Outcomes Economic Assumptions Patients and Setting (Base Case) Data Sources Treatment Strategies Model Structure Probabilities and Hazard Ratios Utilities Costs Verification, External Validation and Calibration Sensitivity Analysis Subgroup analyses Deterministic Scenario Probabilistic Synthesis Novel Findings Implications for Clinical Practice Implications for Health Policy Makers Foci for Future Research Application of NMA Methods in Reviews of CR Natural Extensions of NMA Methods to CR Real World Outcomes Future Randomized Controlled Trials of Exercise- Based CR Cost Effectiveness Evaluation of CR Core Components Conclusion vii

8 References.100 Appendices viii

9 List of Tables Page Chapter 3 Table 1: Trial characteristics (overall and by endpoint) 51 Table 2: Trial characteristics by treatment comparison Chapter 4 Table 1: Trial characteristics (overall and by endpoint) 70 Table 2: Core CR components evaluated by study arms of randomized controlled trials (overall and by endpoint). 71 Table 3: Deviance information criterion and between-trial heterogeneity for each endpoint 72 Table 4: Proportions of simulations from model 2 in which each core component was most effective ix

10 List of Figures Page Chapter 3 Figure 1: Study selection (PRISMA flow diagram).. 48 Figure 2: Risk of bias summary. 49 Figure 3: Network diagram of direct comparisons across treatment strategies (all-cause mortality and CV mortality) Figure 4: Comparative effectiveness of each CR treatment strategy by endpoint. 50 Chapter 4 Figure 1: Study selection (PRISMA flow diagram).. 68 Figure 2: Risk of bias summary. 69 x

11 List of Appendices Page Chapter 2 Appendix A: Search strategy Chapter 3 Appendix A: rjags code for Network Meta-Analysis (NMA) 151 Appendix B: List of excluded studies, with reasons/evidence for exclusion (full-text review subset only) 153 Appendix C: Characteristics of included studies. 193 Appendix D: Risk of bias table for each study 206 Appendix E: Networks of CR treatment strategies by endpoint. 211 Appendix F: All model results. 212 Chapter 4 Appendix A: rjags code for Welton model 217 Appendix B: Incorporating follow-up time using an underlying Poisson distribution Appendix C: Detailed explanation of 4 statistical models 223 Appendix D: Examples of how treatments were coded across study arms 224 Appendix E: Model 1 and model 2 results xi

12 1 1 Introduction 1.1 Coronary Heart Disease (CHD) Coronary heart disease (CHD) is the leading cause of mortality worldwide, estimated to cause one-third (7 million) of deaths globally every year (1). Despite decreases in age-adjusted mortality rates in developed countries in recent decades, it still accounts for 1 in 5 deaths in men and 1 in 6 deaths in women in Europe (2, 3). In the US, 1 in 4 men and 1 in 3 women still die within a year of having their first heart attack (1, 4). Not surprisingly, the decreases in death rates from CHD in developed countries have resulted in increasing morbidity. With better diagnosis, treatment and prevention, the majority of CHD patients are surviving their first heart attack (5). Within 5 years of surviving a first acute MI, it is estimated that 15% to 22% of men and women over the age of 45 will suffer a second MI or will die of CHD (6). As a result, CHD poses a growing and significant economic burden in developed countries, estimated to have cost Germany US$ 74 billion in 1996 (1), US$ 393 billion to the US in 2005 (4), and is estimated to have cost China US$ 558 billion in the last 10 years (1). The average lifetime costs of CHD per patient is US$ 82,000, two thirds of which are due to lost productivity caused by short-term and long-term disability (1). CHD respects no borders (1). It is estimated that more than 60% of the global burden of CHD occurs in developing countries and it is expected that > 80% of the future increase in CHD mortality will happen in low and middle income countries (1, 7, 8). Given that the burden of CHD is projected to nearly double from around 47 million disability-adjusted life years (DALYs, defined as healthy years of life lost ) in 1990 to 82 million DALYs in 2020 worldwide (1), the

13 2 situation will become an increasing priority for any developing country. Even in developing countries with a low risk of CHD, the risk of having a heart attack increases when people immigrate to developed countries, implicating lifestyle as a major culprit for CHD risk. For example, despite Japan having a low rate of CHD, Japanese immigrants to the US have been found to have a gradually increasing risk of CHD over time, eventually approaching the rates of people born in the US (1). Lessons learned in developed countries will be critical to the success of countries about to embark on the journey of intervening against rising CHD mortality and morbidity. For those living with this chronic disease globally, it is not surprising that CHD is also the leading cause of potentially preventable decreases in quality of life (9). Patients living with CHD can have difficulties performing everyday activities such as housework and cooking meals. CHD can also impair sexual function (10). 1.2 Risk Factors for CHD International guidelines on cardiovascular disease have targeted patients with CHD, like those that have had a myocardial infarction (MI) or have angina pectoris (AP), as top priority groups for preventative strategies (11). Controlling this epidemic requires a multifaceted strategy that targets currently recognized and modifiable risk factors for CHD (12). Over 300 different risk factors for CHD have been identified (1). While genetic factors play a part, 80-90% people dying of CHD have one or more major risk factors that are influenced by their lifestyle (1). Yusuf (12) has identified abnormal lipids, smoking, hypertension, diabetes, abdominal obesity, psychosocial factors, consumption of fruits and vegetables, moderate alcohol consumption and regular physical activity to account for > 90% of the population attributable risk (PAR) of future

14 3 myocardial infarctions worldwide in both sexes, at all ages and in all regions. Other risk factors for CHD include poverty, low educational status, poor mental health (depression), inflammation and blood clotting disorders (1). The effective prevention of future CHD needs a global strategy that is based on intervening against the most culprit of cardiovascular risk factors in different geographic regions (12). Primary prevention interventions aimed at the general public, especially patients at high risk for future CV events, and secondary prevention for patients with established CHD are critical components of this global strategy (13). 1.3 Cardiac Rehabilitation (CR) for CHD CR is defined as a medically supervised program designed to optimize a cardiac patient s physical, psychological, and social functioning, in addition to stabilizing, slowing, or even reversing the progression of the underlying atherosclerotic processes, thereby reducing death and disability (14, 15). CR is designed to optimize the secondary prevention of CHD, and is indicated for patients that have recently had a myocardial infarction, have undergone coronary artery bypass graft (CABG) or percutaneous coronary intervention (PCI), have received heart transplants, diagnosed with heart failure (HF) or peripheral artery disease (PAD) and other forms of cardiovascular disease (CVD) like valvular heart disease.(14) Along with evidence-based revascularization procedures and pharmacological treatments, CR is considered to be an essential part of the contemporary care of heart disease and is especially a priority in developed countries with a high prevalence of CHD (16-20). The benefits of CR programs on CHD risk factors, morbidity and mortality in a secondary prevention setting have long been recognized, and these programs are widely available today (21, 22). International clinical guidelines consistently identify exercise training as the central element of CR (16-18, 23, 24). The earliest CR programs were exclusively exercise based (25). More recently, the AHA, the American Association of

15 4 Cardiovascular and Pulmonary Rehabilitation (AACVPR) (16, 23), and the Agency for Health Care Policy and Research (26) have published consensus statements that conclude that CR programs should offer a multifaceted and multidisciplinary approach to overall CHD risk reduction (16). CR is a Class I recommendation in many developed countries (20, 27-31). Most recently, there has been another global effort between the AACVPR, the ACC Cardiology Foundation, the AHA Task Force on Performance Measures, the American College of Chest Physicians, American College of Chest Physicians, the American College of Sports Medicine, the American Physical Therapy Association, the Canadian Association of Cardiac Rehabilitation, the Clinical Exercise Physiology Association, the European Association for Cardiovascular Prevention and Rehabilitation (EACVPR), the Inter-American Heart Foundation, the National Association of Clinical Nurse Specialists, the Preventive Cardiovascular Nurses Association, and the Society of Thoracic Surgeons. This international collaboration has resulted in another consensus guideline urging international healthcare providers and policy makers to ensure the delivery of CR and its core components to patients living with CHD (32, 33) Such international guidelines for CR are supported by a large body of randomized evidence highlighting the efficacy of CR programs on CHD risk factors, health behavior, mortality, morbidity and health-related quality of life (HRQOL), showing that the efficacy of CR is at least comparable to the efficacy of other evidence-based revascularization procedures and pharmacological interventions (22, 34-40). Furthermore, randomized evidence has shown that the significant benefits of CR are provided incrementally above and beyond background usual care, revascularization and pharmacological care (13, 22). Reviews have shown that exercisebased CR significantly reduces mortality (total and/or cardiovascular) and re-hospitalization rates

16 5 by 20% to 30%, and re-infarction rates by as much as 50% (13, 22, 41, 42). A recent systematic review of RCTs published only in the last 10 years showed that lifestyle modification programs significantly reduced all-cause mortality by 34%, cardiac mortality by 48%, and cardiac readmissions and non-fatal MIs by 35% (25). 1.4 The Core Components of CR for CHD CR is a complex intervention that targets CHD risk factors, encompassing a variety of core components including nutritional counseling, patient education, exercise training, risk factor modification, and psychosocial management (16). Both the American Heart Association (AHA) and the American Association for Cardiovascular and Pulmonary Rehabilitation (AACFPR) have recognized the need for CR programs to contain these core components to reduce cardiovascular risk, foster healthy behaviors, reduce disability and to promote an active lifestyle for patients with CHD. In addition to reviews of exercise-based CR, the core components of CR have also been the subject of recent systematic reviews and meta-analyses. Reviews have shown that combined lifestyle, dietary and exercise interventions significantly reduce mortality by 25% to 40%,(34, 43) and re-infarction / hospitalization rates by 48%; (25, 44) patient education interventions significantly improve HRQOL and reduce subsequent healthcare utilization;(38) physical activity alone reduced mortality by 20% to 30%;(34, 42) psychological intervention significantly improved depression and anxiety;(45, 46) and smoking cessation interventions significantly increase smoking abstinence rates subsequently leading to significant reductions in all-cause mortality of 36%.(36)

17 6 The core components of CR have been studied across a range of settings and delivery modalities. They include clinic based settings (22), community-based settings (39, 47), across a variety of delivery modalities including in-person (40), patient self-management (48), telehealth, including phone, internet and videoconferencing patients (49-52). Within intervention variability has also been explored in the core components of exercise training and psychological counseling. Within exercise training, a few reviews have explored the potential differential efficacy of aerobic training vs. non-aerobic training vs. interval training vs. resistance training on CHD outcomes ((53-55), and furthermore, have also explored the efficacy of alternative nontraditional exercise based interventions of Tai Chi and Qigong in patients with CHD (56, 57). Within psychological counseling, the comparative effectiveness of educational, behavioral, cognitive, relaxation and support interventions on mortality and morbidity have also been explored (45). Despite the positive impact of the core components of CR in CHD, recent evidence suggests there may be significant variation in how successfully they are implemented in CR programs. (58) A recent survey in the US has shown that this variation is especially true for the non-exercise based core components of CR, such as nutritional counseling and psychosocial management. (59) The survey showed that all CR programs seemed to be offering group-based nutritional counseling lectures and 87% of the programs having an on-call registered dietitian (RD). However, only 27% of these programs provided individual one-on-one nutritional counseling to their participants. In the remaining 73% of programs, only 43% of overweight patients and 63% of obese patients received individual nutritional counseling sessions. Furthermore, less than 15% of all programs provided overweight and obese patients with individual weight loss counseling, while only 60% of programs implemented weight loss goals as a strategy. Similar trends were found for the core component of psychosocial management,

18 7 showing that only 37% of CR programs screened for depression. It also suggested that 88% of CR programs only offered one group lecture on stress management and/or a related psychosocial topic. This survey clearly shows the urgency required for quality improvement in the provision of the core components of CR to patients with CHD. 1.5 More Recent Findings for CR in CHD Despite many earlier reviews showing statistically significant benefits of exercise-based CR and its general core components, on CHD risk factors, health behavior, mortality, morbidity, and HRQOL, more recent reviews have found conflicting results on mortality. Along with the null results of recent large RCTs of CR in patients with CHD, this has led to a recent and major challenge to the wider adoption of CR in patients with CHD. Today, there is an emerging belief that major advances in the post-mi pharmacological management in recent decades may have nullified any effect of CR on mortality in patients with established CHD (60, 61). Researchers have argued that any review that includes early trials remains heavily weighted by early trials, when mortality was high and with radical changes in the clinical management of MI like the introduction of coronary care units in 1970, early mobilization in 1975, thrombolysis in 1990, primary angioplasty in 2000, aspirin in 1980, beta blockers in 1985, ACEinhibitors in 1995 and statins in 1994, the context of modern CR has changed and the validity of previous conclusions should be called into question. The notion of zero remaining residual risk after modern clinical management is supported by 4 systematic reviews showing that CR did not significantly improve mortality in patients with CHD, one of which suggested a mild 6% increase in the relative risk of mortality that was associated with CR (15, 62-64).

19 8 Some larger RCTs of CR in patients with CHD have also supported these conclusions. A recent RCT of comprehensive CR in 1813 patients with CHD failed to significantly improve mortality, morbidity, psychological well-being, lifestyle or HRQOL at 1, 2, and 7-9 years (61). A few older and larger RCTs have come to similar conclusions (65-68). The largest RCT in CR, which was conducted 30 years ago in 2605 patients with CHD, also failed to establish a significant effect of comprehensive CR programs on mortality and morbidity after an acute MI (69). Most reviews, positive or neutral, conclude that handling the heterogeneity of CR offered in many randomized controlled trials has been a pressing challenge to the interpretation of the efficacy results of different combinations of exercise-based CR interventions and secondary prevention programs without exercise (13, 25, 34, 38, 39, 41, 43, 44, 63, 70, 71). Recent RCTs have also acknowledged smaller than expected sample size, insufficient program intensity, considerable heterogeneity amongst included CR programs and the lack of consensus on critical program-level factors (duration, frequency and intensity) as other potential contributors to their null findings (15, 61, 72). 1.6 Heterogeneity and Complexity: Reasons for the Demise of CR? In addition to the heterogeneity of CR interventions, and the heterogeneity of usual care interventions offered in control arms of international RCTs, many other factors have contributed to heterogeneity in different systematic reviews and meta-analyses. These include factors like the program s intensity, provision of booster sessions and relapse prevention, modes of intervention, delivery type (e.g. face-to-face, internet or telephone), types of participants included, study quality, the lack of specificity of behavioral or psychosocial or lifestyle

20 9 interventions in RCTs, study design, patient populations included, length of follow-up, types and quality of data reported, outcome reporting, outcome definitions, methods reporting, diagnostic criteria used, baseline CHD risk of included patients, sample size, format (e.g. group or individual), number and duration of sessions provided, adherence, loss to follow up, and heterogeneity of personnel delivering the interventions (22, 25, 34, 41, 43, 44, 63, 70, 71, 73). This body of research has also identified treatment setting, treatment timing, treatment duration, background patient habits, intervention characteristics, outcomes evaluated and length of follow up as factors that significantly moderate the treatment effectiveness of CR on mortality and morbidity (25). It has been argued that such heterogeneity has the benefit of better external validity to a wider population of CHD patients in real-world clinical practice (38). It has also been argued that all of this heterogeneity may well reflect only the variety of CR rehabilitation interventions that have been used in previous RCTs (39). It has also long been recognized that such nondifferential misclassification of dichotomous and/or continuous variables in the intervention and control arms of individual RCTs results in an underestimation of the hypothesized relationship between the exposure and the outcome, biasing any significant findings to the null hypothesis (74). In the context of reviews of RCTs of CR, the heterogeneity of interventions being evaluated in both the intervention and control arms of included RCTs, are likely biasing previous risk reduction estimates to the null hypothesis for two reasons. First, the potential combinations of the core components of CR are numerous and excluding some of the core components in any included RCTs may be attenuating the risk reduction estimates of CR in those RCTs. Second, RCTs that formally randomize patients to active interventions (i.e. patient education or exercise)

21 10 in their control arms, also work to attenuate risk reduction estimates in respective RCTs. Yet, in most reviews, the interventions given to patients in the control arms of RCTs are treated/defined as if they were usual care. These arguments alone may well preclude any premature recent conclusions tempering the continued utility of CR in CHD, and are challenges that can be addressed with mixed treatment comparisons meta-analysis methods (also known as network meta-analysis or indirect mixed treatment comparisons).(75) Recently, Anderson (76) recommended that future systematic reviews and meta-analyses of CR in patients with CHD need to explore the complexity and heterogeneity of CR programs in the literature by stratifying patient populations and intervention types, and to explore the association between different intervention characteristics and outcomes across trials. In an effort to evaluate the comparative effectiveness of different CR interventions, they advocate for the use of appropriate indirect comparison methods to start accounting for the heterogeneity of CR interventions in both the intervention and comparator arms of RCTs (76). This is in line with other recent recommendations to open the black box of CR in order to start to determine the incremental benefits of various components of CR for patients with established CHD (13, 77). 1.7 Fortunately, CR is not Alone As a Complex Intervention The challenges of heterogeneity and intervention complexity seen in the evidence evaluating CR interventions are common to evaluations of all complex interventions. Complexity has been defined as a scientific theory, which asserts that some systems display behavioral phenomena that are completely inexplicable, by any conventional analysis of the systems constituent parts.(78-80) Experts have argued that the greater the difficulty in defining precisely what exactly are the active ingredients of an intervention and how they related to each other, the

22 11 greater the likelihood that you are dealing with a complex intervention. Complex health interventions have been defined as interventions with interconnecting parts within experimental and control interventions. They have been further defined by the number and difficulty of behaviors of those delivering or receiving the intervention, number of groups/organizational levels targeted, number and variability of outcomes evaluated, and the degree of flexibility permitted for the intervention. As such, evaluating these interventions in the context of their interconnecting parts has been difficult because of the problems involved in developing, identifying, documenting and reproducing the intervention (77, 79). These challenges have become increasingly problematic as healthcare interventions have moved along the spectrum from simpler interventions, like a single drug, toward more complex interventions like CR (81). The challenges for conducting systematic reviews of complex interventions involve establishing exactly how to credibly incorporate the complexity perspective into a review s research question and methods, and the ability of the review to identify, analyze and integrate the heterogeneous body of evidence in order to understand the processes and outcomes. A further challenge involves presenting the results of such systematic reviews in a way that is meaningful to practitioners, researchers and decisionmakers (81). The evaluation of complex interventions has especially posed unique challenges to systematic reviews. Until recently, as seen in the body of systematic reviews and meta-analyses of CR in patients with established CHD, debates have tended to focus around describing this complexity rather than proposing guidance on what to do about it.(81) In response to these challenges, recent pragmatic approaches to conducting systematic reviews of complex interventions have been proposed (81). As seen in RCTs of CR in patients with established

23 12 CHD, there are many sources of heterogeneity across RCTs of complex interventions, which fall into two broad categories; characteristics of the intervention itself and the characteristics of the causal pathway between the intervention and the outcomes evaluated (81). Indeed, the causual pathways between CR (and its core components) and outcomes have been previously described, and include its effects on heart and coronary vasculature and direct improvements in CHD risk factors. (27)_Additionally, the research question itself may be a question of a package of interventions rather than a single intervention, adding to the challenges of evaluating complex interventions in systematic reviews. Examples of heterogeneity in evaluating complex interventions are similar to those seen in reviews of CR in patients with CHD. They include multiple components (79), number of groups targeted, organizational levels targeted (78), the degree of flexibility of the intervention permitted with patients (78, 82), self-organization, adaptivity, and evolution over time (82, 83). Examples of heterogeneity in the characteristics of the intervention s causal pathway in a particular disease include linearity assumptions ((82, 83), different mediators and moderators of effect like background characteristics of the patient and their environment (84), positive feedback loops (e.g. exercise becoming part of the norm today, despite being less mainstream in the past) (85), synergy between component parts of the intervention (82), number and variability of outcomes evaluated (78), the connectivity of interventions in a system (e.g. physical activity counseling and exercise training) (83), and the interaction with context and the capability created from this interaction, which is very susceptible to different contexts (e.g. policy timing, staffing levels) (82, 86). Given the importance of these sources of heterogeneity and intervention complexity to understanding a complex intervention like CR, and how it works, it will be critically important to identify them in CR and credibly evaluate them in future systematic reviews (81).

24 These Problems Are Inherent to Using Traditional Statistical Approaches to Evaluate Complex Interventions Traditional approaches in systematic reviews and meta-analyses are well-established, and methods for evaluating and summarizing large bodies of evidence for a given patient population have long been used (87). Standard meta-analytical methods are typically restricted to the comparison of 2 interventions using direct head-to-head evidence alone (45). This approach has been used for evaluating the efficacy of CR and seeks to understand the average treatment effect of the intervention as a whole (81). However, in the case of systematic reviews and metaanalyses of complex interventions like CR, no two interventions are exactly alike across RCTs evaluated. Thus, traditional methods have successfully answered the question of What is the average effect (or range of effects) of the range of the components of a complex intervention on the outcomes of interest? Random effects models have been used to account for the heterogeneity of individual RCT designs and interventions (87). A key limitation of standard (or pairwise) approaches for meta-analyses is that they can only compare two interventions at a time (88). As with the body of evidence for the general core components of CR in patients with CHD described above, review questions on the individual core components begin to proliferate, nullifying the existence of a single research question, fragmenting the results of the available evidence, and providing only partial information with each additional review.(40, 89) Finally, reviews have only attempted to account for the heterogeneity in the intervention arms of included RCTs and failed to also account for the heterogeneity of interventions in the comparator arms. Simple questions about complex interventions can be legitimately answered in this way using traditional statistical approaches. By lumping together all intervention arms as the same

25 14 treatment, they help to answer questions like Are these interventions generally effective? This can provide valuable information to clinicians, researchers and decision makers.(81) Although this is of interest, these approaches fall short of answering other important questions like Which of these interventions should be implemented?, Which type of intervention has the greatest probability of being most effective?, and With which types or combinations of components do interventions have the greatest probability of being most effective?. As such, definitive conclusions made in the face of the challenges posed by using a simple approach to evaluating complex interventions, as with CR in patients with CHD, are almost always flawed and premature. 1.9 New Statistical Approaches for Complex Interventions In the case of CR in patients with established CHD, traditional approaches to evaluating complex interventions have posed significant challenges to the interpretation of this rich body of evidence. As such, more complex approaches to handling the complexity of CR are needed (76, 90). Network meta-analysis has been recommended to address such challenges (76). This method is most commonly based on a framework of Bayesian statistics and has been demonstrated to be a valuable approach for complex interventions, with evidence of increasingly successful use in the literature (45, 91-95). Network meta-analysis, otherwise referred to as multiple treatment meta-analysis or mixed-treatment comparison, has been developed to assess the relative effectiveness of several interventions to synthesize evidence across a network of randomized trials, while fully respecting the randomized structure of the evidence (45, ). The method is based on the simultaneous analysis of direct evidence (studies directly comparing the interventions of interest) and indirect

26 15 evidence (studies that are comparing interventions of interest via a common comparator) ( ). Evaluated in this way, network meta-analysis methods are perfectly suited to answer questions like which type of intervention has the greatest possibility of being most effective? or which combinations of components have the greatest probability of being most effective?, the answers of which are needed to address the known challenges in the body of evidence evaluating CR in patients with CHD (45). Although most of the assumptions inherent to network meta-analysis are shared with traditional pairwise meta-analysis approaches, the method is not perfect and has been criticized (106). Numerous research articles have aimed to clarify the method to the scientific and clinical community and have also aimed to address common criticisms (88). These have included recommendations to elucidate the strength and geometry of direct and indirect comparisons, otherwise known as network geometry of the evidence;(103) on the justification of the equal power of traditional and network meta-analysis to provide equally valid interpretations of treatments that are randomized within but not across trials;(107) on ensuring appropriate intervention identification, to address transitivity (if a is related to b, and b is related to c, then a is related to c ), similarity and consistency of the direct and indirect comparisons from included trials; (88, 96, ) on when to rely on direct, indirect and mixed evidence reviews; (88, 115) on evaluating bias and the interpretation of results; (113, ) on the presentation of results in a clinically relevant way; (113, ) and on ensuring the internal and external validity of ranking interventions in order of the probability of being most effective (125). Furthermore, clearly detailed research protocols for systematic reviews and network meta-analyses are now strongly encouraged to promote transparency and accountability. Such protocols are now required to provide ad hoc methodology to addressing transitivity and consistency, listing the factors that are known to introduce heterogeneity and inconsistency,

27 16 detailed conditions under which analyses will be used to synthesize results, and specifying how inconsistency will be addressed if found (88). Online registries and databases like PROSPERO ( and The Cochrane Library ( provide natural destinations for publication of these details. The PRISMA preferred reporting items for systematic reviews and meta-analyses now encourage all researchers to provide registration information for their reviews, if available. Analyses for network meta-analyses are typically done in software based on BUGS language, like WinBUGS (Medical Research Council s Biostatistics Unit, Cambridge, United Kingdom) for example, and challenges in accessing this technology have been fully alleviated by freely providing the such programs and numerous codes for analyzing different types of data online (126, 127). Although the methods for network meta-analysis will continue to develop as its use grows in the literature, the basics of ensuring a comprehensive search of the literature, careful assessment of the body of evidence with respect to transitivity and consistency assumptions, and thoughtful discussion of the study-level biases on the effect estimates can serve to maximize the transparency of a network meta-analysis and serve to avoid errors in interpretation (128) 1.10 Applying New Statistical Approaches in Reviews of CR Although the application of network meta-analysis and Bayesian methods have been increasingly used to evaluate the comparative effectiveness of pharmacological interventions (92, 94, 95, ) and non-pharmacological interventions (134, 135) in medical journals, our group is not aware of any applications in reviews of CR. One review uses network meta-analysis methods to

28 17 evaluate the complex intervention of psychological counseling in patients with CHD, but psychological management represents only one of the core components of CR (45). Thus, our group aimed to start to address the identified challenges of heterogeneity and intervention complexity of CR in patients with CHD, by carrying out a systematic review and network meta-analysis of all RCTs in this area, in order to evaluate the comparative effectiveness of different CR programs, and their core components, on mortality and morbidity in patients with CHD.

29 18 2 Cardiac Rehabilitation Programs and their Core Components for Coronary Heart Disease: A Study Protocol for a Systematic Review and Network Meta-Analysis Authors: Kabboul, Nader N., PhD (Candidate) a, d ; Tomlinson, George, PhD a, b, c, e ; Francis, Troy MSc (Candidate) a,d ; Grace, Sherry, L., PhD b, c, e, f, g ; Hoch, Jeffrey, PhD a, c, e ; Daou-Kabboul, Tamara, MSc h ; Bielecki, Joanna M. BSc, MISt a, d ; Pechlivanoglou, Petros PhD a, d ; Rac, Valeria MD, PhD a, d ; Naimark, David M., MD, MSc a-c, e ; Wijeysundera, Harindra C., MD, PhD a-c, e ; Alter, David A., MD, PhD b, c, e, f ; Krahn, Murray, MD, MSc a-e Affiliations: a Toronto Health Economics and Technology Assessment (THETA) Collaborative; b Faculty of Medicine, University of Toronto; c University Health Network; d Faculty of Pharmacy, University of Toronto; e Institute of Health Policy, Management and Evaluation (IHPME); f The Cardiac Rehabilitation and Secondary Prevention Program, Toronto Rehabilitation Institute (TRI); g York University; h Bridgeport University.. Toronto, Ontario, Canada.

30 19 ABSTRACT Background: Previous reviews of cardiac rehabilitation (CR) in patients with coronary heart disease (CHD) have only evaluated CR programs as a whole and have not compared the individual contributions of secondary prevention programs without exercise, exercise-only CR programs and comprehensive CR programs to each other, nor have they explored the comparative effectiveness of their individual CR core components. Objective: To assess the comparative effectiveness of different CR programs to each other and to assess the comparative effectiveness of their core components to usual care. The results of these assessments were used to describe an approach to conduct a cost-effectiveness of different CR treatment interventions when added to UC (UC) in patients with CHD. Types of Studies: Randomized Controlled Trials (RCTs) with at least 6 months of follow-up. Types of Patients: Adult men and women, in both hospital-based and community-based settings, who have had a myocardial infarction (MI), or who had undergone revascularization (coronary artery bypass grafting (CABG), percutaneous coronary intervention (PCI)), or who have angina pectoris or coronary artery disease defined by angiography were included. Types of Outcome Measures: Mortality (all-cause and cardiovascular (CV)) and morbidity (myocardial infarctions (total, fatal, non-fatal), revascularizations (total, CABG, PCI) and hospitalizations (all-cause and CV). Data Sources: Medline, Medline In-Process, the Cochrane database, Embase, CINAHL, Sci- Expanded, PsychINFO, the Web of Science, from their inception to July 2014.

31 20 Study Selection, Data Collection, Risk of Bias, and Statistical Analysis: Systematic review and network meta-analysis, using standard and established approaches. Binomial likelihood models for binary outcomes at different follow-up times were used to analyze included RCTs. Statistical Software: All models were fitted by using Bayesian inference computed with Monte Carlo Markov chain simulation in jags software in R (R Package (version 3.1.2)) and rjags (R package version 3-13) and OpenBUGS version

32 Objectives I. To assess the comparative effectiveness of different CR programs ( secondary prevention programs without exercise, exercise-only CR programs and comprehensive CR programs) on mortality and morbidity in patients with established coronary heart disease (CHD). II. To assess the comparative effectiveness of the core components of CR on mortality and morbidity in patients with established CHD. III. To describe an approach to assess the cost-effectiveness of the combinations of CR core components when added to usual care (UC) in patients with established CHD. 2.2 Methods Types of studies Randomized controlled trials (RCTs) evaluating any combination of the core components of CR, reporting one or more of the primary outcomes sought in this systematic review, with a followup period of at least six months Types of participants Adult men and women, in both hospital-based and community-based settings, who have had a myocardial infarction (MI), or who had undergone revascularization (coronary artery bypass

33 22 grafting (CABG), percutaneous coronary intervention (PCI)), or who have angina pectoris or coronary artery disease defined by angiography were included. Studies with participants following heart valve surgery, with heart failure, with heart transplants or implanted with either cardiac resynchronization therapy (CRT) or implantable defibrillators (ICD) were excluded. Studies of participants who completed cardiac rehabilitation programs prior to randomization and studies randomizing patients prior to cardiovascular surgery were also excluded, as were studies evaluating the same intervention in both arms Definition of Cardiac Rehabilitation and its Core Components Cardiac rehabilitation is a medically supervised program designed to optimize a cardiac patient s physical, psychological, and social functioning, in addition to stabilizing, slowing, or even reversing the progression of the underlying atherosclerotic processes, thereby reducing death and disability (14, 15) Types of interventions: Core Components of Cardiac Rehabilitation The core components of cardiac rehabilitation/secondary prevention programs include (16, 23): I. Nutritional Counseling (NC) II. Risk Factor Management (RFM) a. Lipids, Hypertension, Weight, Diabetes, and Smoking III. IV. Psychosocial Management (PM) Physical Activity Counseling and/or Patient Education (PE) V. and Exercise Training (ET)

34 23 All of the possible combinations of the core components of CR above were classified as follows, and identified in each of the active and control arms of included RCTs. 1 UC (UC) 17 NC + RFM + PM 2 Nutritional Counseling (NC) 18 NC + RFM + PE 3 Risk Factor Modification (RFM) 19 NC + RFM + Exer 4 Psychosocial Management (PM) 20 RFM + PM + PE 5 Patient Education (PE) 21 NC + PM + Exer 6 Exercise Training (Exer) 22 NC + PE + Exer 7 NC + RFM 23 RFM + PM + PE 8 NC + PM 24 RFM + PM + Exer 9 NC + PE 25 RFM + PE + Exer 10 NC + Exer 26 PM + PE + Exer 11 RFM + PM 27 NC + RFM + PM + PE 12 RFM + PE 28 NC + RFM + PM + Exer 13 RFM + Exer 29 NC + RFM + PE + Exer 14 PM + PE 30 NC + PM + PE + Exer 15 PM + Exer 31 RFM + PM + PE + Exer 16 PE + Exer 32 NC + RFM + PM + PE + Exer The combinations of the core components of CR above were initially classified into secondary prevention programs without exercise, exercise-only CR programs and comprehensive CR programs. Comprehensive CR programs were programs that included exercise plus any other of the core components of CR. The core component of physical activity

35 24 counseling was re-classified as patient education. The individual effects of the core components were then evaluated individually. Usual care / standard of care could include standard medical care, such as background drug therapy at the time of randomization for each RCT, but were not randomized to receive any of the core components of CR, drug therapy and/or surgery. The AACVPR definitions for each intervention were used (16) Types of outcome measures All clinical events, continuous outcomes and sample sizes reported post-randomization in each of the active and control arms of all included RCTs, were extracted for this review. No maximum was imposed on the length of follow-up time, which was also extracted for use in the statistical analyses (see Statistical Analysis below). Primary outcomes I. All-cause mortality II. Cardiovascular (CV) mortality Secondary Outcomes I. Total myocardial infarction (MI) Fatal MI Non-fatal MI II. Total revascularizations Coronary artery bypass graft (CABG) Percutaneous coronary intervention (PCI)

36 25 III. All-cause hospitalizations IV. CV hospitalizations Electronic searches Randomized controlled trials were sought from a full systematic search of the Cochrane Central Register of Controlled Trials (CENTRAL), Health Technology Assessment (HTA), Cochrane Database of Systematic Reviews (CDSR), the Cochrane Methodology Register, NHS Economic Evaluation Database (NHS EED), and Database of Abstracts of Reviews of Effects (DARE) databases in The Cochrane Library, MEDLINE, MEDLINE IN-PROCESS, EMBASE, CINAHL, PsychINFO, and Web of Science (WOS), all from their inception to July 16 th, Search strategies were designed with reference to those of the previous systematic reviews evaluating the core components of CR (22, 36-39, 71). MEDLINE, EMBASE and CINAHL were searched using a strategy combining selected MeSH terms and free text terms relating to the core components of CR and coronary heart disease with RCT filters. The MEDLINE search strategy was translated into the other databases using the appropriate controlled vocabulary as applicable. The search strategy for RCTs evaluating smoking cessation counseling was designed with reference to the search strategy developed by Tobacco Addiction Group specialized register (136). This has been developed from electronic searching of MEDLINE, EMBASE and PsycINFO and the Cochrane Central Register of Controlled Trials (CENTRAL) together with hand searching of specialist journals, conference proceedings and reference lists of previous trials and overviews in smoking cessation. The following MeSH terms to identify potentially relevant trials in the register were used: physician-patient-relations or physicians or family-

37 26 practice or physician s-role. Trials with the words GP or general practice or physician* in the title or abstract were also checked. Searches were limited to randomized controlled trials and a filter will be applied to limit by humans. Consideration was given to variations in terms used and spellings of terms in different countries so that studies were not missed by the search strategy because of such variations. See Chapter 1 Appendix A for a list of the search strategies used Searching other resources Reference lists of all included systematic reviews, meta-analyses and included trials published since inception of any of the above databases to July 16 th, 2014 were fully screened, and relevant titles were imported for evaluation of their eligibility for this systematic review. 2.3 Study Selection, Data Collection and Analysis Selection of Studies The titles and abstracts of all citations identified by the electronic searches were examined for possible inclusion by two reviewers (NNK and TF) working independently. Full publications of potentially relevant studies were retrieved and reviewed by two reviewers (NNK and TF) who then independently determine study eligibility using a standardized inclusion form. Any disagreements about study eligibility were resolved by discussion and, if necessary, a third reviewer (MDK) was asked to arbitrate.

38 Data extraction and management Data from included studies were extracted independently by two reviewers (NNK or TF) using a standardized data extraction tool (the Cochrane Collaboration s recommended tool). Any disagreements about data were resolved by discussion and, if necessary, a third reviewer (MDK) was asked to arbitrate. If data is presented numerically (in tables or text) and graphically (in figures), the numeric data was used because of possible measurement error when estimating from graphs. If data was presented only in graphs, the graphical data was estimated accordingly and independently by each reviewer (NNK or TF). Any remaining discrepancies were resolved by the third reviewer (MDK) Assessment of risk of bias in included studies Two reviewers (NNK and TF) independently assessed the risk of bias in included studies using the Cochrane Collaboration s recommended tool, which is a domain-based critical evaluation of the following domains: sequence generation; allocation concealment; blinding of outcome assessment; incomplete outcome data; and selective outcome reporting (137). Any disagreements were resolved by a third reviewer (MDK). Assessments of risk of bias were provided in the risk of bias table for each study Missing data If there were multiple reports of the same study, the duplicate publications were scanned for additional data. Outcome results were extracted at all follow-up points post-randomization, however, the results closest to two years were used for the network meta-analyses. Study authors were contacted where necessary to provide additional information.

39 Assessment of reporting biases Non-English studies were excluded Data synthesis Data were processed in accordance with the Cochrane Handbook for Systematic Reviews of Interventions (137). Data synthesis and analyses were done using Microsoft Excel software, R, JAGS, and OpenBUGS version ( (138)) Statistical Analysis Binary Outcome Data Collected at Different Follow-Up Time Points: Ten binary outcome measures were included in our review: mortality (all-cause and CV), myocardial infarctions (total, fatal and non-fatal), revascularizations (total, CABG and PCI), and hospitalizations (allcause and CV). Effects on dichotomous outcomes for each comparison were expressed as Hazard Ratios (HRs) with 95% credible intervals (CrIs) using a Bayesian statistical framework via a complementary log-log binomial model (see Statistical Analysis below). ( ) Credible intervals are to Bayesian statistics as confidence intervals are to traditional frequentist statistics. For all of the statistical models in this analysis, the effectiveness of different CR programs and their core components were compared simultaneously for each of the outcomes in this review using previously described approaches (45, ). In summary, the interventions were first classified as (1) secondary prevention programs without exercise ; (2) exerciseonly CR programs; and (3) comprehensive CR programs. These groups comprised the treatments compared in the head-to-head meta-analyses in model 1 (inconsistency model) below as well as the network meta-analyses in model 2 (consistency model). Models 1 and 2 use a

40 29 generalized linear model framework to synthesize data from randomized controlled trials. The model takes the form of a linear regression for both fixed and random effects synthesis for binomial data while accounting for different follow up times (complimentary log-log model, using an underlying Poisson distribution). The model was also applied to pairwise meta-analysis (inconsistency model, Model 1) and applied to indirect comparisons involving multi-arm RCTs (consistency model, Model 2). In model 3, we did form groups of interventions, but characterized them along their 5 key CR core components and estimated the contributions of these components on the same endpoints. Model 3 leverages previously described mixed treatment comparison meta-analysis methods that are geared towards exploring the efficacy of interventions with different components and combination components Assessment of Heterogeneity, Inconsistency and Model Fit Standard approaches for testing for heterogeneity, consistency and model fit were applied to this NMA.( ) Heterogeneity amongst included head-to-head studies were also be explored qualitatively (by comparing the characteristics of included studies) and if needed, quantitatively (using the estimated between-study standard deviation, Cochran s Q, and the I 2 statistic). The l 2 statistic was used to estimate the degree of heterogeneity. This measure describes the percentage of total variation across studies that results from heterogeneity rather than chance. A value of 25% is considered to indicate low heterogeneity, 50% moderate heterogeneity and 75% high heterogeneity (111). Data were pooled for analysis if patient populations are similar enough within each outcome measure to justify pooling (determined by consensus). Data were not pooled for meta-analysis if a high degree of heterogeneity is detected. Random-effects models were generally preferred given the complex nature of CR as an intervention in patients with CHD, and were used if significant heterogeneity is detected (103).

41 30 All NMA models were compared to fixed-effects models to evaluate their comparative heterogeneity, consistency and model fit. Funnel plots and the Egger test were used to examine publication and small study bias where possible (107) Model 1 Direct Head-to-Head Comparison This model was based on the direct head-to-head pair-wise Bayesian statistical analysis (142). Primarily, studies comparing different CR programs to usual care were analyzed. If possible, studies evaluating CR interventions to active control arms (if they were sufficiently represented in the control arms of included RCTs) were also analyzed. Active control interventions could theoretically include one or more of the core components of CR Model 2 Network Meta-Analysis This model was based on previously described methods for network meta-analysis (97). We plotted the network geometry of trials to ensure that the trials are connected by at least one common treatment (81), and exclude any trials that are not connected to the network. We compared the effectiveness of CR treatment strategies on outcomes using a random effects model using jags software in R (R Package (version 3.1.2)) and rjags (R package version 3-13) and OpenBUGS ( (138)). Each analysis was based on non-informative priors for effect sizes and precision. Convergence and inferences were confirmed over a sufficiently long burn-in phase (20,000 iterations) and were based on an additional 30,000-simulation phase. We used recently proposed methods of analysis to assess for the presence of small-study effects and publication bias (118). Direct estimates were obtained from trials directly comparing any of the CR interventions (inconsistency model), whereas

42 31 indirect estimates were obtained by re-running the models excluding trials directly comparing any of the CR interventions. Transitivity of included studies was evaluated qualitatively. The consistency of the networks was checked several ways (113, 143). The distinction between transitivity and consistency can be viewed as analogous to the distinction between clinical/methodological heterogeneity and statistical heterogeneity seen in standard frequentist meta-analyses (88). OpenBUGS programs can be downloaded from The Monte Carlo Markov chain simulation framework of jags software in R (R Package (version 3.1.2)) and rjags (R package version 3-13) and OpenBUGS allowed us to present summaries that are of key interest, such as the probability that a particular intervention is the most effective. This was calculated by recording the proportion of MC iterations in which a given intervention yields the greatest relative effect. (119, 144) We reported the full posterior distribution for each comparison, but as a guide to interpretation, a result for which the 95% CrI does not include the null effect was regarded as statistically significant Model 3 Network Meta-Analysis ( Welton Method ) This model was based on the methods used by Welton, Caldwell (45), in a network metaanalysis of a complex intervention (psychological counseling) in patients with CHD. Welton, Caldwell (45) explored the comparative effectiveness of the core components of psychological counseling using four statistical models; a single-effect model, an additive main effects model, a 2-way interaction model and a full-interaction model. These models were compared for

43 32 goodness of fit using the deviance information criterion (DIC), whereby models with lower DIC were preferred (145) All models were fitted by using Bayesian inference computed with Monte Carlo Markov chain simulation in jags software in R (R Package (version 3.1.2)) and rjags (R package version 3-13) and OpenBUGS version ( (138)). All baseline and intervention effect parameters were given flat normal (0,1000) priors and the between-study standard deviation flat uniform distributions with an appropriately large range given the scale of measurement. Convergence was assessed by using the Brooks-Gelman-Rubin diagnostic (146) in OpenBUGS, and in all cases a sufficient burn-in period were followed by a further sample that is large enough that MCMC error is less than 0.5% of the posterior means of HRs or standardized mean differences. We estimated the predicted effect of a program with the optimal combination of the core components based on the Welton model. 2.4 Summary In summary, our group searched the literature for relevant RCTs of CR using a comprehensive literature search strategy from numerous databases from their inception, utilizing the latest Cochrane methods for the literature search, study selection, data collection, data analysis, risk of bias assessments, and data synthesis; utilized the most respected international CR guidelines to identify eligible CR patient populations, eligible CR interventions, and the most important outcomes in this patient population; and utilized the most up to date network meta-analysis methods developed specifically for complex interventions like CR, RCTs that have different lengths of follow up, and methods that allowed for many different views of the data evaluation (direct/indirect comparisons, classes of CR interventions, CR core components). These

44 33 approaches were specifically targeted at solving some of the problems that exist in previous reviews of CR.

45 34 3 Comparative Effectiveness of Exercise-Only and Comprehensive Cardiac Rehabilitation Programs on Mortality and Morbidity: A Systematic Review and Network Meta-Analysis Authors: Kabboul, Nader N., PhD (Candidate) a, d ; Tomlinson, George, PhD a, b, c, e ; Francis, Troy MSc (Candidate) a,d ; Grace, Sherry, L., PhD b, c, e, f, g ; Hoch, Jeffrey, PhD a, c, e ; Pechlivanoglou, Petros, PhD a, d ; Rac, Valeria, MD, PhD a, d ; Daou-Kabboul, Tamara, MSc h ; Bielecki, Joanna M. BSc, MISt a, d ; Naimark, David M., MD, MSc a-c, e ; Wijeysundera, Harindra C., MD, PhD a-c, e ; Alter, David A., MD, PhD b, c, e, f ; Krahn, Murray, MD, MSc a-e Affiliations: a Toronto Health Economics and Technology Assessment (THETA) Collaborative, b Faculty of Medicine, c University Health Network, d Faculty of Pharmacy, e Institute of Health Policy, Management and Evaluation (IHPME), f The Cardiac Rehabilitation and Secondary Prevention Program, Toronto Rehabilitation Institute (TRI), University of Toronto, g York University, h Bridgeport University Toronto, Ontario, Canada This chapter will be submitted for publication upon its final approval from my thesis advisory committee.

46 35 ABSTRACT Background Participation in exercise-based cardiac rehabilitation (CR) programs has been shown to reduce mortality and morbidity when compared to usual care (UC). Notwithstanding any contraindications to exercise prescriptions, most CR practice guidelines promote comprehensive programs, delivering exercise plus other secondary prevention services. However, the comparative advantages of different CR programs to each other have not been empirically evaluated. Methods A network meta-analysis of randomized controlled trials (RCTs) evaluating CR programs, identified from published sources between database inception dates and July 2014, was undertaken. CR interventions evaluated in the intervention and control arms were grouped into exercise-only CR programs, comprehensive CR programs and secondary prevention programs without exercise. Outcomes included mortality and morbidity measures. Hazard ratios and 95% credible intervals were computed. Effects of different CR programs were compared to each other and to UC. Findings 136 trials (50,054 participants) were included. Compared to UC, exercise-only and comprehensive CR programs significantly reduced the hazards of mortality (all-cause and cardiovascular), myocardial infarction (MI) (total and fatal), total revascularization, and hospitalization (total and cardiovascular) by 32-54%. Secondary prevention programs without exercise significantly reduced the hazards of mortality (all-cause and cardiovascular), MI (total and non-fatal), and hospitalization (total and cardiovascular) by 20%-43%. Compared to secondary prevention programs without exercise, the exercise-only and comprehensive CR programs significantly reduced the hazards of mortality (all-cause and cardiovascular) by 23-

47 36 26%. The analyses were underpowered to detect any potential advantages of comprehensive above and beyond exercise-only CR programs. Interpretation These findings reiterate the central role of CR programs for reducing mortality and morbidity.

48 Introduction Coronary heart disease (CHD) is one of the most prevalent health conditions globally, and is a major cost to healthcare systems.(147) According to the World Health Organization, patients living with CHD accounted for 5.8% of total disability-adjusted life years globally in 2011.(147) Cardiac rehabilitation (CR) is designed to optimize secondary prevention of CHD.(14) Previous reviews have shown that participation in exercise-based CR is associated with 20-30% reductions in mortality and morbidity.(13, ) CR has evolved from an exercise-focused program, to a multi-component model of care delivering comprehensive risk reduction. Indeed, CR clinical practice guidelines promote standard delivery of comprehensive CR.(14, 16, 31, 153, 154) Overall, there is general consensus that comprehensive CR consists of not only exercise training, but also patient education, nutritional counseling, risk factor modification, and psychosocial management. (16, 154) While these components are shown to be effective outside of the CR setting, arguably there is little empirical testing of the benefit of comprehensive CR over and above traditional exerciseonly programs. Given the incremental costs to deliver these ancillary components and the importance of providing evidence-based care, it is important to test their comparative effectiveness, both to each other, as well as to usual care. While previous reviews have evaluated the effectiveness of different CR programs in comparison to usual care in patients with CHD, (13, ) they have not compared their comparative effectiveness to each other. Furthermore, the nature of secondary prevention services applied in the control arms of included RCTs has yet to be taken into consideration. These limitations are inherent to using traditional pairwise statistical methods when applied to

49 38 complex interventions like CR, and have likely biased previous estimates toward the null hypothesis.(75) As such, the use of network meta-analysis (NMA) methods for reviews of the complex intervention of CR has recently been recommended.(76) Accordingly, to address these limitations, and act on recent recommendations, a systematic review and NMA of different CR programs on mortality and morbidity was undertaken. 3.2 Methods This systematic review and NMA of different CR programs for patients with CHD was undertaken in accordance with the Preferred Reporting Items for Systematic Reviews and Meta- Analyses (PRISMA) guidelines and the extension PRISMA statement for NMA.(155, 156) Search strategy and selection criteria The literature search strategy was developed and executed by an information scientist (JB). The following databases were searched from their inception to July 2014: Medline, Medline In- Process, the Cochrane database, Embase, CINAHL, Sci-Expanded, PsychINFO, the Web of Science. The search strategies are shown in chapter 1 appendix A Inclusion and exclusion criteria Randomized controlled trials (RCTs) had to include at least one of the core components of CR (as defined by American Association of Cardiovascular and Pulmonary Rehabilitation),(16) namely patient education, nutritional counseling, risk factor modification, psychosocial management, physical activity counseling and exercise training, or any combination thereof,(16, 23) and were grouped into exercise-only CR programs, comprehensive CR programs (i.e., exercise training plus secondary prevention services) or secondary prevention programs without

50 39 exercise treatment groups. These treatment groups were applied to the intervention and control arms of included RCTs. Usual care could include standard medical care, such as evidence-based medications at the time of randomization, but participants could not be randomized to receive surgery or drug therapy. The trials had to include adults who have had a myocardial infarction (MI), who had undergone revascularization (coronary artery bypass grafting (CABG), percutaneous coronary intervention (PCI)), or whom had angina pectoris or coronary artery disease defined by angiography. Studies also had to report mortality or morbidity outcomes and six or more months of follow-up to be included in this review. The co-primary outcomes were all-cause and cardiovascular (CV) mortality. Secondary pre-specified outcomes were: total MI, fatal MI, nonfatal MI, total revascularization, CABG, PCI, as well as all-cause and CV hospitalization. Studies of participants participating in CR following heart valve surgery, heart failure, and heart transplant or implanted with either cardiac resynchronization therapy or implantable defibrillators were excluded. Studies of participants who completed a CR program prior to randomization, or which randomized participants prior to cardiovascular surgery, or evaluating the same CR strategy in the intervention and control arms were excluded. Non-English studies were excluded Study selection Two investigators (NNK and TAF) first independently reviewed the titles and abstracts of all identified citations. Full-texts of potentially eligible citations were then considered to establish whether they met the inclusion criteria. The same two investigators also searched the reference

51 40 lists of relevant reviews and included studies. Any disagreements were resolved by consensus at each stage of the review Data extraction process and quality assessment Using a standardized data abstraction sheet, two investigators (NNK and TF) also independently extracted the data and independently assessed the risk of bias for each included study using the Cochrane risk of bias assessment tool. (157) Masking was deemed complete when outcome assessors were masked. Patient or performing physician masking were not deemed to be relevant because of the procedural nature of the interventions Data synthesis and analysis Bayesian statistical analyses of both fixed-effects and random-effects NMA were performed. (139) The latter accounts for the correlation between trial-specific effects in trials with more than two arms.(158) Given variability in the length of follow-up of RCTs, hazard ratios (HR) and 95% credible intervals (CrIs) were used as summary statistics. Non-informative priors were used in a complementary log-log binomial likelihood model that accounted for the different follow-up times using an underlying Poisson distribution.(139) The effects of the different CR programs were compared to each other, and to usual care, for each outcome using Markov chain Monte Carlo implemented in jags software in R (R Package (version 3.1.2)) and rjags (R package version 3-13) (chapter 3 appendix A). The first 10,000 iterations were discarded, and 20,000 further iterations were run. Three chains with different initial values were run simultaneously to assess convergence using the Gelman-Rubin diagnostic trace plots. Heterogeneity, model fit, and consistency (between the NMA and the

52 41 pairwise comparison results) were assessed using standard approaches.(140, 141) The principal analyses were computed based on the intention-to-treat populations. The clinical follow-up period closest to two years of follow-up was considered for our analysis. 3.3 Results Figure 1 displays the process of study identification and selection, with reasons for exclusion. A list of excluded studies with reasons is shown in chapter 3 appendix B. There were 136 RCTs with 50,054 participants included in the NMA. Characteristics of included studies are shown in chapter 3 appendix C. Risks of bias assessments for each study are shown in chapter 3 appendix D, with an overall summary shown in figure 2. Included RCTs were conducted between 1975 and 2014, most commonly conducted in the United States (n=36, 26%) and the United Kingdom (n=15, 11%). Tables 1 and 2 provide RCT characteristics, overall, by outcome and by possible treatment group comparison. By outcome, 111 RCTs (80.4%) with 43,885 participants reported the primary outcome of all-cause mortality, while 32 RCT s (23.5%) with 11,496 participants reported the other primary outcome of CV mortality. The secondary outcomes were reported in RCTs (10.3% %) with 3,671-18,076 participants. The mean duration of follow-up was 26.4 months (IQR; 12 to 26 months). Table 2 provides RCT characteristics by possible treatment group comparison. The largest number of RCTs (n=64; 47.1%) and number of participants randomized (n=35,599) tested secondary prevention programs without exercise to usual care. This was followed by RCTs comparing exercise-only CR programs to usual care (25 RCTs (18.4%) in 5,099 participants), and RCTs comparing exercise-only and comprehensive CR programs to secondary prevention programs without exercise. Only 3

53 42 RCTs (2.2%) in 471 participants compared comprehensive to exercise-only CR programs, the only remaining possible comparison. Finally, with regard to overall participant characteristics in included RCTs, the median age was 57.0 years (IQR; 54 to 63) and the median percentage of males in the trials was 81% (IQR; 72% to 98%). Forty-six percent of trials included only post-mi patients. There were no apparent differences in participant characteristics by outcome (table 1). However, RCTs of secondary prevention programs without exercise were more likely to include more females, and possibly older patients (table 2) Network Meta-Analysis Figure 3 shows the networks of comparisons among CR strategies for the co-primary outcomes of all-cause and CV mortality. Networks of comparisons for all secondary outcomes can be found in chapter 3 appendix E. Included RCTs provided direct comparisons between all of the treatment groups for each endpoint, except for the secondary outcomes of fatal MI and PCI, which were missing RCTs with direct comparisons between comprehensive and exerciseonly CR programs. As can be seen in chapter 3 appendix F, the model residual deviances for each outcome compared well to the number of treatment groups evaluated under each outcome, suggesting good model fit for all models.(139) There were small differences in model fit paramaters between the consistency (NMA) model and the inconsistency (head to head) models, all of which generally favored the consistency model. The general model fit of the inconsistency models, when compared to the consistency models, suggests no inconsistency in the overall results. Outcome specific considerations for each treatment comparison are provided respectively below.

54 Effect of CR Strategies When compared to usual care, exercise-only CR programs impacted the most number of outcomes evaluated in this review, significantly reducing the hazards of all-cause mortality, CV mortality, total MI, fatal MI, total revascularization, all-cause hospitalization and CV hospitalization by 41%, 41%, 43%, 65%, 35%, 39% and 54% respectively (figure 4). As can be seen from chapter 3 appendix F, caution is warranted when interpreting the significant results of exercise-only CR programs on total revascularization (considerable heterogeneity detected) and on CV hospitalization (inconsistency detected).(113, 141, 159) When compared to usual care, secondary prevention programs without exercise significantly reduced the hazards of all-cause mortality, CV mortality, total MI, non-fatal MI, allcause hospitalization and CV hospitalization by 20%, 30%, 28%, 33%, 36%, and 43% respectively (figure 4). As can be seen from chapter 3 appendix F, only the significant results on total MI were consistent and had an acceptable level of between-trial heterogeneity.(140) Caution is warranted when interpreting the significant results of secondary prevention programs without exercise on all remaining outcomes (inconsistency and/or considerable between-trial heterogeneity detected). When compared to usual care, comprehensive CR programs significantly reduced the hazards of all-cause mortality, CV mortality, total MI, all-cause hospitalization and CV hospitalization by 38%, 42%, 45%, 52% and 67% respectively (figure 3). As can be seen in chapter 3 appendix F, only the significant results all-cause hospitalization were consistent and had acceptable levels of between-trial heterogeneity.(141) Caution is warranted when interpreting the significant results for comprehensive CR programs on all remaining outcomes (inconsistency and/or considerable between-trial heterogeneity detected).

55 44 When compared to secondary prevention programs without exercise (instead of usual care), the exercise-only and comprehensive CR programs significantly reduced the hazard of all-cause mortality by 23%-26% (chapter 3 appendix F). The results were consistent, (141) however, had considerable levels of between-trial heterogeneity.(140) Caution is warranted when interpreting these results. The results for the comparison between comprehensive and exercise-only CR programs were not significant (chapter 3 appendix F). However, inconsistency, considerable heterogeneity and large credible intervals were observed, suggesting that the results were not precise enough to conclude any potential advantage of comprehensive CR programs to exercise-only Cr programs. It should be noted that only 3 RCTs in 471 participants were available for these analyses. 3.4 Discussion The results of this initial NMA of different CR programs confirm that the provision of any CR in addition to usual care in patients with CHD is associated with significantly lower mortality and morbidity. Results also show that exercise-only and comprehensive CR programs are superior to secondary prevention programs without exercise. This is a novel finding, because previous reviews have only compared CR programs to usual care.(13, ) Unfortunately, the results were not precise enough to conclude or preclude any advantage of comprehensive CR programs when compared to exercise-only CR programs and more research is warranted. The significant reductions in all-cause mortality related to exercise-based CR programs ( exercise-only CR programs and comprehensive CR) found in this review contrast the nonsignificant associations reported in 2 recent reviews of CR,(151, 152) and are larger in

56 45 magnitude than those reported in other reviews.(13, 15, 148, 149, 160) The significant effects on CV mortality, total MI, fatal MI and all-cause hospitalization seen in this review are also larger in magnitude than those previously reported;(13, 15, , 160) Anderson, Oldridge (152) do however report similar results for CV mortality for trials with follow-up longer than 3 years. The significant effects of exercise-based CR on revascularization seen in this review have not been previously reported,(15, 151, 152, 160) and any effects on CV hospitalizations have not been previously evaluated. By acting on recent recommendations for reviews to account for the heterogeneity of CR interventions in both the intervention and control arms of included RCTs, and by accounting for the different length of follow-up of included RCTs,(76) via standard NMA methods,( ) it is likely that this review better elucidated the effects of exercise-based CR when compared to the traditional meta-analytical methods applied in previous reviews. These results are some of the first to begin to open the black box of CR using this novel, arguably more rigorous, statistical approach that is better catered to complex interventions, like CR.(77, 152) It is interesting to note the similarity in the magnitude of the results for all-cause mortality, CV mortality and total MI herein when compared to one previous review.(41) By excluding RCT s with exercise in their control groups, Lawler et al. (2011) effectively removed any effects of control exercise interventions have towards a null hypothesis; a phenomenon that NMA methods also accounts for. 17 RCTs with exercise in the control group were also excluded from this review (figure 1), and although 3 RCTs comparing comprehensive to exercise-only CR programs were included, standard NMA methods are well equipped to account for active interventions employed in the control arms of these RCTs.(75, ) The inclusion of these RCTs in other reviews has likely resulted in bias towards the null hypothesis given the limitations of traditional meta-analytic methods.(75)

57 46 The finding that exercise-only and comprehensive CR programs significantly reduced the hazard of all-cause mortality by 23-26% when compared to secondary prevention programs without exercise is also interesting. With major advances in the medical management of CHD in the new century, many recent RCTs evaluating exercise-based CR have randomized participants to non-exercise based secondary prevention interventions in their control arms.(61, ) Based on the results of the NMA, any two-year trial comparing exercise-based CR programs to non-exercise based secondary prevention interventions, with 80% power to detect a 23% reduction in the hazard of all-cause mortality seen in this review, would require > 5,000 CHD participants. Such a trial would be more than twice the size of the largest of recent RCTs in this area.(61, ) As with previous reviews of CR, this review has several limitations. First, included RCTs rarely gave enough detail to assess the adequacy of their potential risk of bias and only a minority of trials was judged to be adequate in terms of their quality. The difficulty faced in truly blinding patients and providers to CR interventions in RCTs will persist, however, future trials must continue to adhere to current standards for conducting and reporting RCTs.( ) Second, despite accounting for some of the heterogeneity in CR interventions by evaluating distinct CR programs to each other and to usual care, a key challenge to interpreting the results of this review continued to be the unexplained between-trial heterogeneity of RCTs in this area. This made it difficult to interpret most of the significant results for comprehensive CR programs and secondary prevention programs without exercise. As recommended by Anderson and Taylor (2014), future network meta-analyses methods should continue to shed light on additional sources of heterogeneity inherent to RCTs of CR. This should include exploring CR in distinct CHD patient populations that have received it and should further extend

58 47 investigations into the comparative effectiveness of the individual core components of CR evaluated in RCTs. A recent network meta-analysis of psychological interventions in patients with CHD is one such example.(75) Finally, given that 81% of included patients were males, exploring CR specifically in females should help increase the external validity of these results to the entire scope of eligible patient populations in future reviews. Third, only 3 trials in 471 participants evaluated comprehensive CR programs to exercise-only CR programs in this review. As a result, there may have been insufficient data to reach any definitive concludsions for this treatment comparison. However, there were many indirect comparisons that also evaluated this combination of CR programs to each other. Only a high quality, appropriately powered RCT is needed to potentially address this outstanding research question. The results of such a RCT should also serve to advise policy makers on any added benefits of providing exercise and non-exercise core components in combination to patients with CHD. The findings herein suggest that there seems to be little to choose between exerciseonly CR programs, secondary prevention programs without exercise and comprehensive CR programs in patients with CHD, unless there are patient/provider barriers to an exercise prescription. They also show that contemporary RCTs that employ non-exercise secondary prevention interventions in their control arms were likely under powered to detect significant effects on mortality and morbidity. Clearly, CR programs that provide exercise or non-exercise secondary prevention services to referred patients with CHD warrant continued clinical and policy focus, as they are improving important outcomes in this patient population.

59 48 64,444$records$identified$$ $ 63,197$through$database$search$ $$$$$$$$$$$$$$$1,247$through$other$sources$$$ $ 49,684$records$after$duplicates$removed$ 44,253$records$excluded$ 5,431$full@text$articles$assessed$for$eligibility$ $ 5,249$full@text$articles$excluded$ $$$$$$$$$3,057$not$randomized$$ $$$$$$$$$$$$707$ineligible$patient$population$ $$$$$$$$$$$$529$study$duration$<$6$months$$ $$$$$$$$$$$$511$do$not$report$outcome$$ $$$$$$$$$$$$321$ineligible$intervention$ $$$$ 56$non@English$ $ 53$same$intervention$both$arms$ $$ $$$$$$$$$$36$$ non@ex $vs$ non@ex $ $ $$$$$$$$$$11$$ ex@only $vs$ ex@only $ $ $$$$$$$$$$$$6$$ ex@plus $vs$ ex@plus $ $$$$ 10$randomized$after$CR$$ $$$$ $$5$randomized$before$surgery$ 182$reports$of$136$studies$included$in$ quantitative$synthesis$(network$meta@analysis)$ $ Figure 1: Study selection (PRISMA Flow Diagram) PRISMA = Preferred Reporting Items for Systematic reviews and Meta-Analyses. Non-ex = secondary prevention programs without exercise ; Ex-only = exercise only CR programs; Ex-plus = comprehensive CR programs; CR = cardiac rehabilitation

60 49 Figure 2: Risk of bias summary All#Cause)Mortality) CV"Mortality" NON# EX( 15# 54# 11# 6" 9" 3" 10# 3# 4" EX# PLUS( 2" 18# UC( 8" EX# ONLY( Figure 3: Network diagram of direct comparisons across treatment strategies (all- cause & CV Mortality) The width of the lines represents the number of trials in which a direct comparison is made (number of trials also provided alongside lines). The size of each square represents the number of people who received each treatment. UC = UC. Non- Ex = secondary prevention programs without exercise. Ex- Only = exercise- only CR programs. Ex- Plus = comprehensive CR programs. CR = cardiac rehabilitation.

61 50 All Cause Mortality Non Ex Ex Only Ex Plus CV Mortality Non Ex Ex Only Ex Plus Total MI Non Ex Ex Only Ex Plus Fatal MI Non Ex Ex Only Ex Plus Non Fatal MI Non Ex Ex Only Ex Plus Revascularization Non Ex Ex Only Ex Plus CABG Non Ex Ex Only Ex Plus PCI Non Ex Ex Only Ex Plus All Cause Hospitalization Non Ex Ex Only Ex Plus CV Hospitalization Non Ex Ex Only Ex Plus Hazard Ratio (95% CrI) Figure 4: Comparative effectiveness of each CR treatment strategy by endpoint Reference group = usual care; Non- ex = secondary prevention programs without exercise ; Ex- only = exercise- only CR programs; Ex- plus = comprehensive CR programs; 95% CrI = 95% credible interval; CV = cardiovascular; MI = myocardial infarction; CABG = coronary artery bypass graft; PCI = percutaneous coronary intervention

62 51 # N Post- MI trials(%) All Endpoints , (46%) Primary Endpoint All- Cause Mortality , (50%) Secondary Endpoints CV Mortality 32 11, (56%) Any MI 53 18, (57%) Fatal MI 14 3,671 9 (64%) Non- Fatal MI 27 6, (59%) Revascularization 43 12, (58%) CABG 36 8, (58%) PCI 26 9, (42%) Any Hospitalization 44 16, (41%) CV Hospitalization 23 6, (43%) Age (Range) (IQR) 57.0 (48-84) (54-63) 57.5 (48-84) (54-64) 55.5 (48-80) (53-57) 54.7 (48-80) (53-57) 53.0 (48-64) (51-57) 53.5 (48-62) (52-56) 55.4 (50-71) (52-58) 54.4 (50-71) (52-58) 56.9 (50-71) (53-59) 57.4 (50-76) (55-63) 57.7 (50-76) (56-60) Male (Range) (IQR) 81% (14-100) (72-98) 81% (24-100) (71-98) 89% (24-100) (77-100) 87% (35-100) (77-100) 100% (77-100) (89-100) 91% (56-100) (78-100) 87% (56-100) (78-100) 86% (56-100) (78-97) 83% (56-100) (77-100) 79% (14-100) (71-90) 78% (14-100) (72-93) FU (Months) (Range) (IQR) 26.4 (6-300) (12-26) 29 (6-300) (12-36) 32.6 (6-120) (12-50) 36.6 (6-300) (12-50) 50.7 (6-120) (24-69) 27.0 (6-120) (8-36) 30.7 (6-300) (6-24) 32.3 (6-300) (10-27) 30.1 (6-300) (6-12) 18.3 (6-108) (8-22) 20.0 (6-108) (6-23) # = Number of trials reporting endpoint. N = number of patients randomized. MI = myocardial infarction. Age = median of mean age. Male = median % of males. Duration = mean intervention duration (weeks). FU = mean length of follow up (months). IQR = interquartile range. CV = cardiovascular. CABG = coronary artery bypass surgery. PCI = percutaneous coronary intervention. Table 1: Trial characteristics (overall and by endpoint)

63 52 Treatment 1 Treatment 2 # N Post- MI trials (%) UC Non- Ex 64 35, (41%) UC Ex- Only 25 5, (68%) UC Ex- Plus 11 1,697 3 (27%) Non- Ex Ex- Only 16 4,102 8 (50%) Non- Ex Ex- Plus 17 2,456 7 (41%) Ex- Only Ex- Plus (66%) Age (Range) (IQR) 60.2 (50-80) (57-65) 56.0 (48-84) (52-58) 54.5 (53-74) (54-58) 54.3 (48-74) (51-61) 54.1 (51-77) (53-57) 59.4 (55-64) (57-61) Male (Range) (IQR) 72% (0-100) (61-83) 87% (43-100) (80-100) 85% (0-100) (76-97) 88% (66-100) (80-100) 85% (14-100) (78-100) 93% (93-93) (93-93) FU (Months) (Range) (IQR) 20.7 (6-108) (6-18) 28 (6-108) (12-36) 51.3 (6-300) (13-48) 24.8 (6-108) (8-30) 31.1 (6-120) (12-24) 30 (6-60) (15-42) # = Number of trials reporting endpoint. N = number of patients randomized. MI = myocardial infarction. Age = median of mean age. Male = median % of males. Duration = mean intervention duration (weeks). FU = mean length of follow up (months). UC = usual care. Non- Ex = secondary prevention programs without exercise. Ex- Only = exercise- only CR programs. Ex- Plus = comprehensive CR programs. Table 2: Trial characteristics by treatment comparison

64 53 4 The Comparative Effectiveness of the Core Components of Cardiac Rehabilitation on Mortality and Morbidity: A Systematic Review and Network Meta-Analysis Authors: Kabboul, Nader N., PhD (Candidate) a, d ; Tomlinson, George, PhD a, b, c, e ; Francis, Troy A., MSc (Candidate) a,d ; Grace, Sherry, L., PhD b, c, f, g ; Hoch, Jeffrey, PhD a, c, e ; Pechlivanoglou, Petros, PhD a, d ; Rac, Valeria, MD, PhD a, d ; Daou- Kabboul, Tamara, MSc h ; Bielecki, Joanna M. BSc, MISt a, d ; Naimark, David M., MD, MSc a- c, e ; Wijeysundera, Harindra C., MD, PhD a- c, e ; Alter, David A., MD, PhD b, c, e, f ; Krahn, Murray, MD, MSc a- e Affiliations: a Toronto Health Economics and Technology Assessment (THETA) Collaborative, b Faculty of Medicine, c University Health Network, d Faculty of Pharmacy, e Institute of Health Policy, Management and Evaluation (IHPME), f The Cardiac Rehabilitation and Secondary Prevention Program, Toronto Rehabilitation Institute (TRI), University of Toronto, g York University, h Bridgeport University Toronto, Ontario, Canada This chapter will be submitted for publication upon its final approval from advisory committee.

65 54 ABSTRACT Background Although cardiac rehabilitation (CR) participation has been shown to reduce mortality and morbidity in patients with coronary heart disease (CHD), the comparative effectiveness of each of its core components has not been evaluated. Methods A systematic review and network meta-analysis (NMA) of randomized controlled trials (RCTs) evaluating the core components of CR (nutritional counseling (NC), risk factor modification (RFM), psychosocial management (PM), patient education (PE), and exercise training (ET)) was undertaken. Published RCTs were identified from database inception dates to July Endpoints included measures of mortality (all-cause and cardiovascular (CV)) and morbidity (total, fatal and non-fatal myocardial infarction (MI), total revascularization, coronary artery bypass surgery, percutaneous coronary intervention, and hospitalization (all-cause and CV)). Hazard ratios (HR) and 95% credible intervals (CrIs) were used as summary measures. Effects were compared to usual care. Findings 169 RCTs (62,149 participants) were included. ET significantly reduced the hazards of all-cause mortality (HR=0.75; 95% CrI ( )), total MI (HR = 0.64; 95% CrI ( )), total hospitalization (HR = 0.58; 95% CrI ( )) and CV hospitalization (HR = 0.48; 95% CrI ( )) respectively. PM significantly reduced the hazards of all-cause mortality (HR = 0.69; 95% CrI ( )), all-cause hospitalization (HR = 0.67; 95% CrI ( )) and and CV hospitalization (HR = 0.50; 95% CrI ( )). RFM and PE significantly reduced the hazards of total MI (HR = 0.63; 95% CrI ( )) and all-cause hospitalization (HR = 0.77; 95% CrI ( )) respectively.

66 55 Interpretation These findings reiterate the central role of ET in CR, and provide evidence for the CR core components of PM, RFM, and PE. 4.1 Introduction Coronary heart disease (CHD) is one of the most prevalent health conditions globally, and is the leading cause of mortality. (147) CHD accounted for 5.8% of total disability-adjusted life years in 2011, and for one-third of all deaths every year.(147) Cardiac rehabilitation is designed to optimize secondary prevention of CHD. (14, 16) Reviews have established that CR participation is associated with 20-30% lower mortality and morbidity (13, 149). CR has evolved from an exercise-focused program, to a multi-component model of care delivering comprehensive risk reduction. Indeed, learned CR societies have published statements on the so-called core components of CR,(14, 16, 31, 153, 154, 170) ensuring delivery of all evidence-based secondary prevention recommendations. These core components have also been internationally agreed through the International Council of Cardiovascular Prevention and Rehabilitation charter.(171) Overall, there is general consensus that CR consist of approximately 5 core components, namely patient education, nutritional counseling, risk factor modification, psychosocial management, and exercise training. (16, 154) While each of the components are supported by evidence of benefit,(172) reviews of the effectiveness of CR to date have not considered the individual components (except exercise). Expert recommendations for each core component should be tested, in such a way that the complexity of CR can be considered.(77) Accordingly, the objective of this study was to evaluate the comparative effectiveness of the individual core components of CR.

67 Methods Network meta-analysis (NMA) was used to test the comparative effectiveness of the 5 CR components. The systematic review was undertaken in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and the extension PRISMA statement for NMA.(155, 156) Search strategy and data sources The literature search strategy was developed and executed by an information scientist (JB). The following databases were searched from their inception to July 2014: Medline, Medline In- Process, the Cochrane database, Embase, CINAHL, Sci-Expanded, PsychINFO, the Web of Science. The search strategies can be found in chapter 1 appendix A Inclusion and exclusion criteria Randomized controlled trials (RCTs) evaluating any combination of the core components of CR were eligible for inclusion. Participants were adults who had had a MI, or who had undergone revascularization (coronary artery bypass grafting (CABG), percutaneous coronary intervention (PCI)), or whom had angina pectoris or coronary artery disease defined by angiography. Studies had to include at least one of the core components of CR (as defined by American Association of Cardiovascular and Pulmonary Rehabilitation),(16) namely patient education, nutritional counseling, risk factor modification, psychosocial management, physical activity counseling and exercise training, or any combination thereof. (16, 23) Patient education and physical activity counseling were combined into a single intervention: patient education. Usual care could include standard medical care, such as evidence-based medications at the time of

68 57 randomization, but participants could not be randomized to any of the core components of CR or to drug therapy or to surgery. Studies also had to report mortality or morbidity outcomes, assessed after six or more months of follow-up. The co-primary outcomes were all-cause and cardiovascular (CV) mortality. Secondary pre-specified outcomes included total MI, fatal MI, non-fatal MI, total revascularization, CABG, PCI, as well as all-cause and CV hospitalization. Studies of participants participating in CR following heart valve surgery, heart failure, heart transplants or implanted with either cardiac resynchronization therapy or implantable defibrillators were excluded. Studies of participants who completed a CR program prior to randomization, randomized participants prior to cardiovascular surgery, or evaluated the same CR treatment strategy in both arms were excluded, as were non-english studies Study selection Two investigators (NNK and TAF) first independently reviewed the titles and abstracts of all identified citations. Full-texts of potentially eligible citations were then considered to establish whether they met the inclusion criteria. These 2 investigators also searched the reference lists of relevant reviews and included studies. Any disagreements were resolved by consensus at each stage of the review Data extraction process and quality assessment Using a standardized data abstraction sheet, two investigators (NNK and TAF) also independently extracted the data for each included study, and independently assessed the risk of bias using the Cochrane risk of bias assessment tool. (157) Masking was deemed complete when

69 58 outcome assessors were masked. Patient or performing physician masking were not deemed to be relevant because of the procedural nature of the interventions Data synthesis and analysis A Bayesian random-effects NMA model was computed, (139) which accounted for the correlation within a trial and random effects of trials with more than two arms. (158) As there was variability in the length of follow-up of included RCTs (6-300 months), hazard ratios (HR) and 95% credible intervals (CrIs) were used as summary statistics. Non-informative priors were used in a complementary log-log binomial likelihood NMA model that accounted for the different follow-up times using an underlying Poisson distribution (chapter 4 appendix B).(139) Four different models of the effects of the core components of CR were computed using previously established NMA methods: a single-effect model, an additive main effects model, a two-way interaction model, and a full-interaction model. A more detailed description of these four models can be found in chapter 4 appendix C. (75) In summary, the single-effect model (model 1), all CR interventions were grouped together as a single treatment (overall CR). This is similar to traditional head-to-head CR versus usual care meta-analytical methods used in previous reviews. The additive main effects model (model 2) assumed a separate treatment effect for each of the core components and for each of their additive combinations. The two-way interaction model (model 3) was an extension of the main effects model (model 2) and allowed for bigger (synergistic) or smaller (antagonistic) effects for each potential pair of CR core components. Finally, the full interaction model (model 4) assumed that each possible combination of the core components of CR was distinct. Two examples of how treatments were coded across their 5 CR core component dimensions across study arms are provided in chapter 4 appendix D.

70 59 The effect of each core CR component compared to usual care was estimated for each outcome using Markov chain Monte Carlo implemented in jags software in R (R Package (version 3.1.2) and rjags (R package version 3-13) (chapter 4 appendix A). The first 100,000 iterations were discarded, and all results were based on a further sample of at least 50,000 iterations. Three chains with different initial values were run simultaneously to assess convergence using the Gelman-Rubin diagnostic trace plots. Heterogeneity and model fit were assessed using standard approaches.(75, 140, 141) The Monte Carlo Markov chain simulation framework also allowed for the presentation of other summaries of key clinical and policy interest, such as the probability that a particular core component is most effective by outcome evaluated. Analyses were done in the intention-to-treat populations. The clinical follow-up period closest to two years of follow-up was considered for analysis. 4.3 Results Figure 1 displays the process of study identification and selection. A list of excluded studies with reasons for exclusion is shown in chapter 3 appendix A. There were 169 RCTs assessing 62,149 participants included in the NMA. Characteristics of included studies are shown in chapter 3 appendix B. Risks of bias assessments for each study are shown in chapter 3 appendix C, with an overall summary shown in figure 2. Included RCTs were undertaken between 1974 and 2014, most often in the United States (n=48, 28%) and the United Kingdom (n=17, 10%). Characteristics of included RCTs overall and by outcome can be found in table 1. The mean duration of follow-up was 25.9 months (IQR; 12 to 36 months). Overall, 138 (81.7%) RCTs with 54,084 participants reported the primary outcome of all-cause mortality, while 43

71 60 (25.4%) RCTs with 15,884 participants reported the other primary outcome of CV mortality. The secondary endpoints were reported in RCTs (10.1% %) with 5,212-23,254 participants. The number of RCT arms evaluating different combinations of the core components (overall and outcome-specific) can be found in table 2. The majority of included RCTs were designed with two arms (n=167; 98.8%), and two RCTs had three arms. Overall, patient education was the most-frequently evaluated individual core component of CR (71 RCT arms), followed by psychosocial management (23 RCT arms). The combination of exercise training and patient education (44 RCT arms) was the most frequently evaluated combination of core components followed by psychosocial management and patient education (16 RCT arms), the comprehensive combination of all core components (10 RCT arms), and the combination of risk factor modification and patient education (9 RCT arms). Usual care (no CR) was evaluated in the control arms of 100 (59.2%) included RCTs. Finally, with regard to participant characteristics in included RCTs (table 1), the median of mean ages was 58.0 years (Interquartile Range (IQR); 54 to 63) and the median mean percentage of males in the trials was 79% (IQR; 68% to 94%). Forty-one percent of trials included only post-mi patients Network Meta-Analysis Table 3 shows results of the NMA. The single-effect model (i.e., model 1; for all endpoints) and the additive main effects model (i.e., model 2; for all endpoints except for CV mortality, fatal MI, and PCI) were the best-fitting models for the data (DIC increase of > 4 points indicates poorer model fit). Furthermore, model 2 explained more of the between-trial

72 61 heterogeneity than model 1. The other models were more complex without explaining more of the between-trial heterogeneity or improving model fit. For all-cause mortality, the two-way interaction model (i.e., model 3) was also a good fit to the data; however, as with other outcomes, there was no evidence for synergy/attenuation between paired core components of CR. It should be noted that there is likely not enough data available to detect such synergy/attenuation in the two-way interaction (model 3) or the full-interaction (model 4) models. Therefore, only the results of the additive main effects model (i.e., model 2) are reported for all endpoints except for CV mortality, fatal MI and PCI, for which we could only report the results from the main effects model (model 1). When compared to UC, the single effects model (model 1) showed that CR overall significantly reduced the hazards of CV mortality, fatal MI and PCI by 38% (HR=0.62; 95%CrI ( )), 61% (HR=0.39; 95%CrI ( )), and 29% (HR=0.71; 95%CrI ( )) respectively (chapter 4 appendix E) Effects of Core CR Components When compared to usual care, the core component of exercise training impacted the largest number of outcomes in this review, significantly reducing the hazards of all-cause mortality, total MI, all-cause and CV hospitalization by 25%, 36%, 42% and 52% respectively (chapter 4 appendix E). However, the between-trial heterogeneity approximated the mean estimate of the log HR for all-cause mortality, suggesting caution is warranted when interpreting this result (chapter 4 appendix E). Next, the core component of psychosocial management significantly reduced the hazards of all-cause mortality, all-cause and CV hospitalization by 31%, 33% and 50% respectively

73 62 (chapter 4 appendix E). However, the between-trial heterogeneity for all-cause hospitalization was larger than the mean estimate of the log HR, suggesting that caution is warranted in interpreting this result (chapter 4 appendix E). Finally, the core components of risk factor modification and patient education each significantly reduced the hazard of one secondary outcome. The core component of risk factor modification significantly reduced the hazard of total MI by 37% (chapter 4 appendix E). The core component of patient education significantly reduced the hazard of all-cause hospitalization by 23% (chapter 4 appendix E). However, caution is warranted when interpreting these two results given the considerable between-trial heterogeneity in RCTs reporting these two outcomes (chapter 4 appendix E). The core component of nutrition counselling did not significantly reduce the hazard of any outcome. Table 4 shows the probability that each of the CR components was the most effective component for each of the outcomes evaluated in this review. As displayed, usual care was the least effective strategy for patients with CHD across all endpoints. With regard to the primary outcomes, the CR core component of psychosocial management was the most effective for reducing the hazard of all-cause mortality. With regard to secondary endpoints, the CR core component of exercise training was the most effective for reducing the hazards of all-cause and CV hospitalization. The CR core component of risk factor modification was most effective for reducing the hazard of total MI. The CR core component of nutritional counseling was most effective for reducing the hazard of non-fatal MI, revascularization and CABG.

74 Discussion By acting on recent recommendations that CR reviews consider the heterogeneity of core components in the active and control arms of RCTs, and by accounting for differing lengths of follow-up in CR RCTs,(173) via standard NMA methods,( ) this review has better elucidated the true effectiveness of CR and its core components in patients with CHD. The results of this first ever NMA or CR core components establishes that the delivery of each of the recommended core components is associated with reductions in mortality and morbidity. In order of decreasing effectiveness, the core components of exercise training, psychosocial management, risk factor modification, and patient education each significantly reduced the hazard of at least one of the mortality or morbidity endpoints by 23-57%. Usual care was the least effective strategy in reducing mortality and morbidity when compared to all other strategies. The effects of CR overall on mortality and morbidity reduction reported in this review are greater than those reported in the most recent meta-analysis evaluating exercise-based CR.(152) For instance, in the current review, CR was associated with 38% lower CV mortality, compared to 26% in the last Cochrane review. Moreover, while the previous review aggregated fatal and non-fatal MI and found no significant effect of CR (except with follow-up of more than 3 years), herein the hazard of fatal MI was significantly reduced by 61%. Similarly, the Cochrane review identified no significant effect of CR in reducing PCI, however herein a 29% reduction in the hazard of PCI was reported. Finally, although the current review and the Cochrane review report significant reductions in re-hospitalization, the magnitude of reported reductions were only similar to those reported for the core component of patient education in this review (~20%). However, the core components of exercise training and psychological management were shown to significantly reduced re-hospitalizaiton by 33%-52% in this review. It is likely that the NMA

75 64 approach in this review better elucidated the effects of CR and its individual core components when compared to the traditional meta-analytic methods applied in previous reviews. As has been demonstrated in previous reviews,(13, 15, , 174) the results herein confirm the centrality of the exercise component of CR in reducing mortality and morbidity in patients with CHD. The results also provide evidence to support other core components of CR, particularly psychosocial management. Previous reviews of psychosocial management in patients with CHD have only reported non-significant 7-20% reductions in all-cause and CV mortality, and had not evaluated its effects on hospitalization (all-cause or CV).(175) The beneficial effects of psychosocial management observed using NMA methods were incredibly compelling, and warrant further investigation. Given secondary prevention guidelines for CHD,(172) it is likely that many of the core components of CR are being increasingly provided via usual care in clinical practice. Thus, the significant results seen in this review suggest incremental, and synergistic, benefit of CR above and beyond evidence-based usual care. Conversely, the lack of statistical significance for any core component, as with the core component of nutritional counseling seen in this review, does not imply a lack of benefit altogether. Rather, it may suggest that nutritional counseling is being provided sufficiently in cardiology practice more broadly. The positive but non-significant trends detected in this review for nutritional counseling on non-fatal MI, revascularization and CABG should be noted and warrant further replication.

76 Implications Results of this review support the continued provision of the core components of CR to all indicated patients. Given that CR is chronically under-resourced, it is known that many programs do not have the capacity to deliver all guideline-recommended core components. In particular, many programs have limited human health resources in the area of psychosocial management.(176, 177) This is disconcerting given the findings herein. Policy-makers have a key role to play in ensuring CR is a standard model of care delivered with exercise, psychosocial management, risk factor modification and patient education. These results should also be used to inform CR guidelines regarding core components (preferably using TIDieR guidance),(178) with this better evidence applied to support component recommendations. Which combination of core components will provide the best clinical and cost effectiveness, and whether they will have an additive effect when combined together (especially when added to exercise alone), should be investigated Limitations This review has several limitations. First, information provided in included RCTs was often insufficient to assess their risk of bias, and only a minority of trials was judged to be adequate in terms of quality. That patients and providers cannot be blind to arm allocation in CR RCTs cannot be overcome, however, future trials must aspire to the highest standards for conducting and reporting RCTs.( ) Second, despite reducing between-trial heterogeneity by accounting for the distinct core components of CR and the different lengths of follow-up in included RCTs, a key challenge to interpreting the results of this review continued to be some unexplained between-trial

77 66 heterogeneity. Future research is needed utilizing NMA methods to further investigate other sources of heterogeneity in CR RCTs. As previously recommended,(76, 77) this could include patient populations that receive CR, the intensity/dose of CR interventions provided, CR setting (center-based, home-based, community), and other possible study-level covariates (e.g., year of recruitment, country). There is however a limitation on the number of parameters that can be included in the statistical models, given the number of data points and the data structure.(75, 109) Finally, it is expected that the standards of usual care for patients with CHD also differ by country, and as such, this may have also contributed to the remaining unexplained between-trial heterogeneity in this review. Such non-differential misclassification across included RCTs from different countries is likely to bias the results towards the null hypothesis, without introducing any systematic bias.(75) Third, although the additive main effects model had the best fit for most of the outcomes evaluated in this review, the differences in model fit statistics were not large, and should not solely serve to inform any assumption of additive effects of the core components of CR in the real world. Only a high quality, appropriately powered RCT, could potentially address this outstanding research question of comprehensive CR programs in comparison to exercisetraining alone CR programs. The results herein could inform its design. In conclusion, using a novel approach, which takes into consideration the core components of CR, this review has reiterated the significant benefits of CR participation in reducing mortality and morbidity. The findings herein confirm the centrality of exercise training as the key component of CR, and also provide strong evidence for the benefit of the other CR components, particularly psychosocial management. Policies are needed to standardize delivery

78 67 of comprehensive CR, ensuring delivery of these beneficial core components to all CHD patients.

79 68!! 64,444$records$identified$$ $ 63,197$through$database$search$ $$$$$$$$$$$$$$$1,247$through$other$sources$$$ $ 49,684$records$after$duplicates$removed$ $ 44,253$records$excluded$ 5,431$full@text$articles$assessed$for$eligibility$ $ 5,219$full@text$articles$excluded$ $$$$$$$$$3,057$not$randomized$$ $$$$$$$$$$$$707$ineligible$patient$population$ $$$$$$$$$$$$529$study$duration$<$6$months$$ $$$$$$$$$$$$511$do$not$report$outcome$$ $$$$$$$$$$$$321$ineligible$intervention$ $$$$ 56$non@English$ $ 23$same$intervention$both$arms$ $$ $$$$$$$$$$14$ET$ $$$$$$$$$$$$4$PE$ $ $$$$$$$$$$$$1$NC$+$PE$ $ $$$$$$$$$$$$1$RFM$+$PE$ $ $$$$$$$$$$$$1$NC$+$RFM$+$PE$ $ $ $$$$$$$$$$$$1$NC$+$RFM$+$PE$+$PM$+$ET$ $$$$ 10$randomized$after$CR$$ $$$$ $$5$randomized$before$surgery$ $ 212$reports$of$169$studies$included$in$ quantitative$synthesis$(network$meta@analysis)$ $ Figure 1: Study selection (PRISMA Flow Diagram) PRISMA = Preferred Reporting Items for Systematic reviews and Meta-Analyses. ET = exercise training; PE = patient education; NC = nutritional counselling; RFM = risk factor modification; PM = psychosocial Management; CR = cardiac rehabilitation.

80 69 Selective!Reporting! Incomplete!Outcome!Data! Blinding!Outcome!Assessement! Blinding!Participants!&!Personnel! Allocation!Sequence!Concealment! Sequence!Generation! 0%# 25%# 50%# 75%# 100%# High!Risk! Low!Risk! Unclear!! Figure 2: Risk of bias summary!!