The Contribution of Individual Exercise Training Components to Clinical Outcomes in Randomised Controlled Trials of Cardiac Rehabilitation: A Systematic Review and Meta-regression

Inclusion Criteria

Studies were eligible if they were randomised controlled trials with at least one arm that compared exercise-based cardiac rehabilitation to usual care, and which reported at least one of the following outcomes: total mortality, cardiovascular mortality, myocardial infarction, coronary artery bypass graft (CABG) or percutaneous coronary intervention (PCI). Trial participants could comprise men or women of any age who had been diagnosed with coronary heart disease, suffered a myocardial infarction, or undergone either CABG or PCI procedures.

Cardiac rehabilitation could have been provided in any setting (e.g. home, community or outpatient centre) but must have involved the prescription of a structured exercise program (either supervised or unsupervised), with or without the addition of lifestyle modification and counselling. Unsupervised home-based interventions were required to comprise a structured and detailed exercise prescription for participants (e.g. specific intensity, frequency and duration of individual sessions) with regular staff review, in a similar manner to centre-based programs. Including only structured exercise interventions (and excluding those which offered only general exercise advice, e.g. perform 150 min of exercise per week, walk daily) allowed for a more robust examination of the effect of specific exercise variables on outcomes.

Trials with a follow-up period of <3 months, with inadequate randomisation techniques, or those reporting on heart failure programs, were excluded. Trials published in abstract form only (from conference proceedings) were eligible for inclusion only if authors responded to email requests for further study information to determine if they met all eligibility criteria.

Search Strategy and Selection of Studies

We conducted a structured search (last run on 28 January 2016) of the following databases: PubMed, Allied and Complementary Medicine, EMBASE, PEDro, Science Citation Index Expanded (via Web Of Science), CINAHL, The Cochrane Library and SPORTDiscus. The search strategy was developed in conjunction with an experienced medical librarian, consisted of a variety of exercise-based rehabilitation terms combined with coronary heart disease descriptors and used methodological filters to limit the results to randomised controlled trials, meta-analyses and systematic reviews. Variations of the following were searched, using a combination of text words and index terms: coronary artery disease, ischemic heart disease, cardiovascular disease, myocardial infarction, angina, CABG, PTCA, rehabilitation, cardiac rehabilitation, exercise, physiotherapy, physical fitness, exercise training, training program, aerobic exercise, randomised controlled trial, random, controlled trial, meta analysis, systematic review and clinical trial (Additional file 1: Appendix S1 contains full search strategy).

In order to identify further trials, the reference lists of all eligible studies were hand-searched, as well as those of previously published cardiac rehabilitation meta-analyses and systematic reviews. The process was supplemented further with searches of conference proceedings in the field from the previous 2 years (World Congress of Cardiology, American Heart Association Scientific Sessions, European Society of Cardiology Congress and EuroPrevent), and by forward and backward citation tracking of all eligible studies in Web of Science. No restrictions were placed on language or date of publication.

The titles and abstracts of all articles retrieved from electronic searching and other sources were screened (by BA) for eligibility against the pre-specified inclusion criteria, and full-text publications were obtained for any potentially relevant studies. Any uncertainties regarding study eligibility were resolved through discussion with two other reviewers (PG, TH) until a consensus was reached.

Data Extraction

A standardised data extraction form (available on request) was used to extract the following: participant characteristics (gender, age, diagnosis), intervention(s) characteristics, follow-up duration, all relevant clinical outcomes reported, the type of care provided to the usual care comparison group, cardiovascular medication usage and details about the methodological quality of the trial. If multiple time points were reported, data from all which were greater than 3 months were extracted. Where multiple intervention arms were compared with usual care within a single study, each exercise arm was considered a separate intervention and data extracted accordingly.

In order to extract the individual exercise characteristics (used as co-variates in meta-regression analysis) from each included intervention in a standardised manner, a modified version of the Template for Intervention Description and Replication checklist was used [19] (Additional file 1: Table S2). Where the individual components of exercise interventions were missing or described in insufficient detail, attempts were made to locate this information in additional publications or by contacting each corresponding author (n = 62) via email. This process is described in further detail elsewhere [11]. While 28 authors responded to these requests and provided missing intervention details, 14 did not respond to numerous reminders. We were unable to locate contact details for a further 20 authors. A number of authors were also contacted to clarify outcome data.

Outcomes

Our primary outcome was cardiovascular mortality. Secondary outcomes were total mortality, myocardial infarction, CABG and PCI procedures. All outcomes were assessed for the time period beginning with randomisation until the end of last reported follow-up, thereby including events which occurred before, during and after the exercise intervention period for all trials. We chose not to use a combined endpoint for analysis, as although it may have increased the statistical power required to detect treatment effects, it can also be problematic in interpretation and generate misleading conclusions, particularly about the strength of reductions in mortality [20, 21]. For this reason, the results of each outcome are presented separately.

Risk of Bias of Included Trials

The quality of each included trial was assessed according to the Cochrane Risk of Bias Tool [22]. Trials were rated as having a low, high or unclear risk of bias for the following criteria: sequence generation (adequate randomisation methods described), concealment of allocation, study blinding (of participants, personnel and outcomes), incomplete outcome data (participant attrition) and selective outcome reporting. Blinding of participants to exercise interventions is virtually impossible; however, the clinical outcomes in our analyses are unlikely to be influenced by knowledge of group allocation. We therefore considered the criterion of blinding in terms of outcome assessment. The degree of participant attrition was assessed at the outcome, rather than the study level. For example, a substantial number of participants may not have completed the trial (high-risk attrition at study level) but the authors were able to determine from medical records whether or not these missing participants were deceased at the trial end (low-risk attrition for mortality outcome).

Statistical Analysis

Both meta-analysis and meta-regression techniques were carried out in order to synthesise the individual trial data for each intervention and then examine the effect of individual intervention characteristics on each outcome. All comparisons of exercise-based cardiac rehabilitation versus usual care are expressed as relative risks (RRs) with 95% confidence intervals (CI) which were calculated from reported event and population data in each trial. Effects were considered statistically significant at a p value of <0.05. In order to minimise the bias due to missing participant data across a number of trials, the primary meta-analysis was conducted using a complete case analysis, as proposed in the Cochrane Handbook [23] and by Akl et al. [24]. This method uses, as the denominator, only those participants with complete outcome data recorded. Meta-analyses were performed in RevMan (Version 5.3, Cochrane Collaboration), with the relative risks of the outcome in each study pooled using a Mantel-Haenszel random effects model. Both the meta-analysis and meta-regression were performed using random effects models as clinical heterogeneity was expected given the wide variety of interventions and patients represented in the included trials. We tested for this heterogeneity in each meta-analysis using the Cochrane Q statistic and also calculated the I ² to quantify the percentage of variation in effectiveness across studies which would be considered beyond a chance finding. Potential for publication bias was examined using funnel plots.

In trials with multiple intervention arms, combining these into a single group is the recommended approach for meta-analysis to avoid unit-of-analysis errors arising from correlated intervention effects. Using this approach however does not allow an investigation into the differences between treatment arms required by this analysis. Hence, we chose to use another recognised approach, dividing proportionally the number of events and total population of the shared usual care group between intervention arm comparisons, avoiding ‘double-counting’ while still allowing exploration of heterogeneity across differing interventions [23].

We also used approaches proposed by Akl et al. [25] to test the robustness of effect estimates to missing participant data (due to withdrawal, drop-out, etc.). This was done by conducting two different sensitivity analyses using plausible assumptions to impute missing participant event data for each trial. The first assumed that the incidence of events in all missing participants was the same as that observed among those followed up in the control arm of the same trial. The second approach imputed missing participant data relative to those with complete data in the same trial arm, assuming that those missing in the intervention arm had an increased event incidence of 1.5 compared to those followed up, and the control arm the same event incidence (1.0) in those with complete and missing data. We additionally performed further sensitivity analyses excluding those trials assessed at high risk of bias for missing outcome data and those published in only abstract form or as a doctoral thesis.

Trial Selection

Sixty-eight publications [34–101] met our inclusion criteria (see Fig. 1 for PRISMA flow chart), reporting on clinical outcomes and follow-up of 69 different trials (one publication [98] reported outcomes of a collaborative study which involved several individual trial centres). Several trials included multiple intervention arms [39, 71, 72, 87] resulting in a total of 72 individual exercise interventions.

Trial Characteristics

Participant Characteristics

Of the 13,423 participants with coronary heart disease included in all trials, the mean age in individual trials ranged from 49 to 80 years, with an average age of 54 years (Table 1). The majority of participants were male (83%), with only 10 trials containing more than 25% of female participants and 26 trials (38%) comprising male participants only. A large proportion of trials (43/69; 62%) included only patients after myocardial infarction; however, those published from 1990 onward often included patients diagnosed with coronary artery disease (n = 8; 12%), post-PCI (n = 5; 7%), post-CABG (n = 2; 3%) or a combination of these cardiac aetiologies (n = 11; 16%).

Intervention Characteristics

Despite our repeated attempts to obtain missing information [11], many details about the individual components of exercise interventions remained unknown, with 22% of interventions with missing details for at least one characteristic of session time, session frequency or exercise intensity. Hence, while these trials and interventions are included in the overall meta-analysis, they could not be entered into the meta-regression where these co-variates were missing.

Usual Care Comparisons

The care provided in the usual care arm of trials varied greatly (Table 1). For many comparisons (n = 31; 41%), the exact nature of care received in this arm was unspecified, described only as usual care in the hands of a physician or local health service. Participants in the control arm of 10 other comparisons were reported to receive ‘standard cardiological care’, which usually consisted of guideline-based treatment, including regular cardiology or nursing review, and medication titration. The remaining 31 control comparison groups also received either usual or cardiological care, with the addition of risk factor and/or general physical activity advice (e.g. eat less fat, walk daily). Unlike home-based intervention arms, these control arm interventions did not take the form of a prescribed and structured program but consisted of printed or on-line material, a single education session, face-to-face discussion or personalised advice. The first two types of control arms were classified as ‘usual care’ in this review, whereas we considered participants in the latter type of trials to have been delivered ‘usual care plus lifestyle advice’ in the control arm.

Guideline-based medication regimens which included statin therapy were reportedly used at baseline, during the trial, or throughout follow-up, in 24 (34%) of the control arm comparisons. In a further seven comparisons (10%), statin therapy was not explicitly reported but probable given the dates of participant recruitment. Lipid-lowering therapy was either not begun in the control arm, or reported as not used, in 17 comparisons (24%) and was unlikely used in a further 24 comparisons with recruitment and follow-up periods preceding 1994. Overall, this resulted in statin use in the control arm (matched with comparable use in the intervention arm) for 40% of comparisons, 56% of comparisons without statin use in either arm and three trials [46, 48, 60] in which the intervention arm included a specific pharmacological lipid-lowering strategy compared to the control arm (Table 1).

Publication Bias and Quality Assessment

Primary Outcome: Cardiovascular Mortality

Cardiovascular mortality, reported in 31 trials (comprising 44 different interventions and 6926 participants), showed a statistically significant reduction with exercise-based cardiac rehabilitation compared to usual care over a follow-up period of 10 years in length (RR 0.74, 95% CI 0.65 to 0.86, p < 0.0001) (Fig. 2). When considering follow-up out to 19 years (additional information provided in two trials), a smaller reduction in risk was observed (RR 0.82, 95% CI 0.73 to 0.91, p = 0.0004). No evidence of heterogeneity was observed in either analysis (I ² = 0%). Sensitivity analysis also demonstrated the effect estimates to be robust to missing participant data (Additional file 1: Table S6a) and publication type (Additional file 1: Table S7a).

Analysis of Intervention Components

Subgroup analyses with interventions stratified based on population diagnosis (I ² = 0%, p = 0.60 for interaction effect), minutes of prescribed exercise per week (I ² = 0%, p = 0.81), type of usual care (I ² = 0%, p = 0.72) or the use of lipid-lowering therapy (I ² = 0%, p = 0.60) did not demonstrate any differences in effect on cardiovascular mortality (Additional file 1: Table S8a).

Univariate meta-regression analysis found no significant effect of the year of recruitment to the trial, exercise mode and exercise provider; how the exercise was delivered; or the prescribed intervention duration, session frequency, time or intensity on cardiovascular mortality outcomes (Additional file 1: Table S9a). A significant relationship was however observed between the level of adherence to the exercise intervention and cardiovascular mortality (Fig. 3), with a 28% reduction in relative risk (RR 0.72, 95% CI 0.52–0.99, p = 0.045) observed for those comparisons reporting high levels of adherence to the prescribed exercise intervention compared to those reporting only moderate levels of adherence. No other co-variates met the pre-specified p value cut-off for entry into the multivariate model.

Secondary Outcome: Total Mortality

The pooled meta-analysis found a 10% reduction in overall mortality risk with exercise-based cardiac rehabilitation compared to usual care over a follow-up period of 10 years (Fig. 4). When extending the follow-up period out to 19 years, no demonstrable reduction in overall mortality was observed (RR 0.97, 95% CI 0.90 to 1.04, p = 0.37). These intervention effects were also less robust to missing data with the sensitivity analysis suggesting a trend towards non-significance (Additional file 1: Table S6b). No evidence of heterogeneity was observed in either analysis (I ² = 0%).

Secondary Outcome: Myocardial Infarction

Random effects meta-analysis found that exercise-based cardiac rehabilitation produced a 20% reduction in myocardial infarction events compared to usual care (Fig. 4). This effect was also robust to missing data and publication type in sensitivity analysis (Additional file 1: Table S6c; Additional file 1: Table S7c), and no evidence of heterogeneity was observed (I ² = 0%).

Analysis of Intervention Components

Subgroup analyses with interventions stratified based on population diagnosis, minutes of prescribed exercise per week and type of usual care did not demonstrate any differences in effect on myocardial infarction events (Additional file 1: Table S8c). Low to moderate heterogeneity (I ² = 33%, p = 0.23 for interaction effect) was observed among subgroups divided by use of concomitant lipid-lowering therapy, which was largely due to the difference in effect observed in the subgroup of two trials providing statin therapy as part of the intervention arm.

Univariate meta-regression analysis found only one co-variate that significantly predicted myocardial infarction outcomes (Additional file 1: Table S9c). A significant positive relationship was observed between the total time prescribed for exercise each session and the risk of myocardial infarction (Fig. 5). That is, for every 1 min increase in time (between 25 and 90 min), the relative risk of myocardial infarction with the exercise intervention vs usual care increased by 1%. While several co-variates met the pre-specified p value cut-off for entry into the multivariable model, after backwards elimination of non-significant predictors, only exercise time remained. Due to a moderate pairwise correlation between session time and session frequency (r = −0.512), some potential multicollinearity may have been observed between these two co-variates in the multivariable analysis. However, repeating the analysis accounting for this fact did not produce different results about potential predictors of myocardial infarction outcomes.

Secondary Outcome: Coronary Artery Bypass Grafting

Random effects meta-analysis found no demonstrable reduction in the number of participants who required coronary artery bypass grafting (CABG) in the exercise-based cardiac rehabilitation interventions compared to usual care (Fig. 4). Sensitivity analysis for missing outcome data did not change this effect estimate (Additional file 1: Table S6d), and no evidence of heterogeneity was observed (I ² = 0%).

Secondary Outcome: Percutaneous Coronary Intervention

The pooled meta-analysis found no difference between exercise-based cardiac rehabilitation and usual care in reducing PCI events (Fig. 4), with some possibly important heterogeneity observed (I ² = 37%, p = 0.06). Sensitivity analysis for missing outcome data did not change this effect estimate or substantially reduce the degree of heterogeneity (Additional file 1: Table S6e). Excluding trial data obtained only in abstract form reduced heterogeneity without substantially changing the effect estimate (Additional file 1: Table S7d).

Analysis of Intervention Components

Subgroup analyses with interventions stratified based on population diagnosis, type of usual care or minutes of prescribed exercise per week did not demonstrate any differences in effect on the number of PCI procedures performed (Additional file 1: Table S8e). Some non-significant heterogeneity (I ² = 21%, p = 0.28) was observed for subgroups stratified by use of lipid-lowering therapy. However, stratifying the overall meta-analysis based on this factor could not account for the overall heterogeneity observed.

Univariate meta-regression analysis found no significant effect of the year of recruitment to the trial, exercise mode, format of exercise training or prescribed intervention duration, session frequency or time on PCI outcomes (Additional file 1: Table S9e). A significant positive relationship was however observed between the highest level of intensity prescribed and the risk of PCI (Fig. 6). That is, for every 1% increase in maximal heart rate prescribed (between 60 and 91% HR_max), the relative risk of PCI with the intervention compared to usual care increased by 5%. Additionally, in interventions where the team providing exercise training included medical practitioners, as opposed to nursing or allied health staff alone, the risk of PCI was significantly decreased (RR 0.30, 95% CI 0.14–0.62, p = 0.004). None of the co-variates which met the pre-specified p value cut-off for entry into the multivariate model remained after backwards elimination of non-significant predictors.

Exercise-Based Cardiac Rehabilitation in the Modern Treatment Era

It has been previously suggested that the benefits of exercise-based cardiac rehabilitation over usual care may be incremental in the modern era, with the advent of lipid-lowering therapy and optimal medical treatment [8, 97]. Previous meta-analyses [8, 10, 15, 102] have examined this hypothesis with subgroup analysis comparing the outcomes of trials published before and after 1994/95, with this date acting as a surrogate for changing practices in treatment reported after this time [30]. While these studies found no significant subgroup differences and consequently claimed the benefits of cardiac rehabilitation to be maintained in the modern era, the use of publication date in this manner has several limitations. Firstly, it is a crude measure of when a trial was conducted and may not accurately reflect treatment practices at the time of publication. Trials with long recruitment and follow-up periods, along with delays in publication, mean that not all post-1994 trials were necessarily conducted in the modern treatment era. For example, one trial published in 2002 [68] reported a 10-year follow-up of patients recruited in 1984 and 1985 and was conducted wholly in the pre-statin era. Additionally, at least one study published post-1995 [37] explicitly excluded the use of lipid-lowering therapy. Including these trials in their respective year-based subgroups may not provide a true reflection of changes in care. Furthermore, even in the modern treatment era, the level of usual care provided to the control arm can vary widely, from follow-up with a physician to educational packages or sessions and targeted guideline advice about lifestyle modification. This is an important consideration as Janssen et al. [28] have previously shown that there may be a dilution of effectiveness for secondary prevention interventions when compared to control groups where usual care is optimal.

Therefore, to assess the impact of cardiac rehabilitation treatment in the modern era, we examined the effects of both usual care and lipid-lowering therapy separately, as well as performing meta-regression analysis based on the median year of reported trial recruitment. Subgroup analysis was not able to replicate the findings of Janssen, with no significant differentiation in the effects of cardiac rehabilitation with the varying levels of usual care provided. This seems to support the importance of providing structured exercise training even in the presence of other lifestyle change measures, such as physical activity advice, education or counselling. This is an important consideration given that the level of usual care provided in clinical practice may often be less ideal than that delivered in clinical trials. While subgroup analysis was not able to display a significantly different effect of cardiac rehabilitation in programs with lipid-lowering therapy compared to those without, we did observe some evidence that total mortality was less likely to be reduced by cardiac rehabilitation in the presence of lipid-lowering therapy. This was more pronounced when excluding those studies with a specific lipid-lowering strategy in the intervention arm (p for subgroup differences = 0.07, I ² = 70%).

While this attenuation of effect may be due to statin therapy alone, it is more likely confounded by other improvements in medical management and patient demographics in these later trials which included lipid-lowering therapy. Essentially, as hypothesised, statins may be acting as a surrogate marker for improved medical management over time. This argument is in agreement with the most recent Cochrane review on the topic where a trend was observed via meta-regression for smaller reductions in total mortality with cardiac rehabilitation trials published in more recent years [8]. However, our analysis using a more clinically meaningful co-variate for trial year found no trend or significant association between the median year of participant recruitment to the trial and the subsequent risk of mortality. Together, these findings suggest that while pharmaceuticals may affect the ability of cardiac rehabilitation to reduce overall mortality to some extent, the importance of this intervention in reducing clinical events in the modern treatment era does not appear to be greatly diminished.

The Impact of Exercise Intervention Characteristics and Prescribed ‘Dose’

Our findings extend those of previous meta-analyses of exercise training in cardiac rehabilitation, which have also found no effect of prescribed exercise dose (reported as program duration × session time × session frequency) on clinical outcomes [8, 10, 15]. However, analysing the effect of exercise interventions based on such a crude overall measure of ‘dose’ may have acted to mask the effects of the individual components in these earlier analyses, and none took into account varying exercise intensities. Additionally, the level of poor intervention reporting we observed in the majority of trials would have hampered the ability of researchers to obtain all three elements required for the calculation of dose (e.g. the most recent Cochrane review [8] was only able to examine the effect of 18 doses of exercise on cardiovascular mortality out of 24 trials). Our use of individual exercise components, along with efforts to obtain missing descriptions, meant that we were able to explore a greater range of dose comparisons in analysis, with increased power to detect a true effect. The finding that few differences in effect were present, despite these improvements in methodology, supports the conclusions of the original meta-analyses. Additionally, they are consistent with a systematic review which found no effect of exercise session frequency or duration on improvements in fitness in cardiac rehabilitation patients [18].

It could be argued that energy expenditure expressed as either MET-hours or kilocalories would have provided the most useful measure of exercise dose accounting for various combinations of intensity, frequency and duration of exercise sessions prescribed within an intervention. However, the combination of missing intervention characteristics and a lack of knowledge about baseline VO₂ values in the majority of included trials meant that we were unable to obtain the necessary information to calculate energy expenditure within most individual interventions. Consequently, we could not examine the relationship between this measure of exercise dose and clinical outcomes. Improving the reporting of intervention characteristics and baseline exercise data in future trials would help to address this issue.

While meta-regression analyses provide a large amount of evidence based on indirect comparisons of exercise training interventions within the cardiac population, relatively few trials exist which provide direct, head-to-head comparisons of variations in individual exercise components. Our findings however are largely in agreement with this small body of literature which has found no effect of interventions with differing durations [104], frequencies [105, 106], session times [107] or proportions of hospital-based sessions [108–110], on fitness, risk factors or quality of life. While avoiding the potential aggregation bias of subgroup analysis and allowing a more robust comparison of intervention characteristics, these trials are unfortunately limited by short follow-up periods and generally report only on risk factors or biological parameters, making it difficult to assess their impact on the long-term clinical outcomes observed in our analysis.

In our analysis, we did not observe higher intensity interventions to be any more effective than lower intensity ones in reducing mortality or preventing myocardial infarction in the long term. This differs to other research which suggests that increased exercise intensity is an important factor in achieving superior outcomes in patients with cardiovascular disease. Systematic reviews have shown high-intensity interval training (HIIT) to be better than moderate-intensity continuous exercise training at improving fitness (VO₂ peak) in patients with coronary heart disease [111, 112] or lifestyle-induced chronic disease [113]. Increased exercise intensity was also associated with increased gains in fitness in a recent systematic review [18] and in a large cohort of cardiac rehabilitation patients [114]. However, our review included only eight trials which specifically used interval training protocols, and the intensities prescribed during work intervals were generally lower than those observed in the HIIT literature. Additionally, all trials included in our analysis compared changes in clinical endpoints rather than fitness. While cardiorespiratory fitness is recognised as an important prognostic factor for mortality [115, 116] and is associated with decreased incidence of myocardial infarction and revascularisation in patients with CAD [117], a direct link between exercise intensity and clinical endpoints is yet to be made, as most trials are limited by short-term follow-up (4 months or less) and are underpowered to detect changes in clinical endpoints. One trial which did report an extended follow-up period of 4 years [118] found no difference in the rate of recurrent myocardial infarction (fatal and non-fatal) between patients randomised to low or high intensity exercise.

We did however observe evidence of a statistically significant relationship between the highest intensity of exercise prescribed to participants and an increased risk of PCI events. While we were unable to find similar reports of this phenomenon in previous research, one possible explanation may be that higher intensity exercise causes higher myocardial oxygen demands (as a result of increasing heart rate, myocardial contractility, ventricular work and blood pressure) which if not met, may elicit myocardial ischaemia not otherwise observable at lower intensities [119, 120]. If observed, these symptoms may trigger further examination and invasive intervention, given that trial participants are reviewed regularly, and that PCI (like CABG) is a physician-driven endpoint which may be prone to clinically subjective decisions [121] (an elective rather than a spontaneous event like death or infarction). Additionally, PCI (as opposed to CABG) would most likely be considered the primary method of revascularisation in this population, given that most of those enrolled in the trials would be at low subsequent cardiovascular risk [122]. It is also important to note that only three of the trials reporting PCI outcomes prescribed higher training intensities in an interval format [75, 89, 95], and it may be that periods of intermittent recovery (as used in HIIT) are needed to mediate the potential negative effect of higher intensity exercise.

Why an increase in the risk of MI should occur with longer prescribed session times is also not clear. One previous trial in coronary patients has found exercise durations extended to 60 min blunted the beneficial effects on antioxidants and vasculature observed with 30 min of training [123]. Conversely, 40 and 60 min of training have been shown to be equally effective in improving exercise capacity and lipid profile [107]. We did observe some potential confounding between session frequency and session time, with longer session times associated with less frequent exercise sessions. Despite this, however, session frequency alone was not predictive of MI outcomes, and the relationship between session time and myocardial infarction remained after adjusting for the frequency of sessions. Further investigation of the impact of session time on MI outcomes may therefore be warranted.

Exercise Adherence and Mortality

Our finding of improved mortality outcomes with increased adherence to exercise training is supported by a similar observation in several cohort studies involving cardiac rehabilitation program attendees. Even when matched for prognostic factors, users of more cardiac rehabilitation sessions (≥70%) were less likely to die over the subsequent 5 years than lower users [124], and program completers had a reduced risk of death and hospitalisation than non-completers [125]. Additionally, Hammill [126] observed what appeared to be a dose-response gradient between sessions attended and the risk of myocardial infarction and mortality. Unfortunately, all these studies may have been subject to confounding if sicker participants were those who stopped attending earlier. Additionally, they were all based on adherence to the standardised US model of 36 supervised on-site sessions, meaning that outcomes observed may not necessarily be due to adherence alone, but the extra volume of training received. Given however that the results we observed in our study arose from adherence to programs of varying lengths and types and we found no effect of other training characteristics on mortality, adherence alone does appear to be an important factor.

While exercise adherence in itself may be an important driver behind reduced mortality, our findings could also be attributed to a phenomenon known as the ‘healthy adherer effect’ [127]. That is, those participants who complete more exercise sessions are also those who are more likely to adhere to medication regimens and perform other healthy lifestyle changes. Consequently, good exercise adherence could be considered a marker for adherence to other positive health behaviours which may also decrease the risk of mortality and clinical events. This has been previously observed by others, with increased attendance at on-site cardiac rehabilitation sessions related not only to reductions in all-cause mortality, re-admissions and major events, but also to greater improvements in clinical risk profile and secondary prevention behaviours such as medication and dietary adherence [128, 129]. Additionally, some ‘clustering’ of good lifestyle behaviours (diet, exercise, smoking cessation) was observed in survivors of MI, where increased adherence led to a decrease in mortality, myocardial infarction and stroke [130]. It could also be postulated however that cardiac rehabilitation programs in themselves promote adherence, by providing a structured learning environment which offers the frequent contact, skills training and self-efficacy enhancing care required for successful adherence [131]. Furthermore, knowledge about personal risk factors (usually provided at CR sessions) has been shown to correlate to a patient’s ability to achieve treatment goals [132]. Consequently, attending an increasing number of sessions provides increased chances to gain these skills and knowledge. Despite these difficulties in separating out the effect of exercise adherence, it does appear to play a crucial role in clinical outcomes, which has important implications for program design.

Implications for Practice

There is a growing concern that cardiac rehabilitation needs to be re-engineered for the future, with a decreased reliance on traditional models of care [29, 133]. In order to do this, it is essential to understand the interaction between program components and patient outcomes to ensure essential elements are retained. In terms of exercise prescription at least, our findings appear to align well with the current impetus to provide flexible, individualised and ‘menu-based’ models of care tailored to circumstances and individual patient needs [134, 135]. While a number of the exercise training programs contained within our meta-regression were those of the traditional hospital-supervised, moderate-to-vigorous intensity format, many alternate styles were also observed. All appeared equally effective in reducing mortality, and most also in reducing the risk of subsequent myocardial infarction. It may be more important therefore to design exercise programs in whichever format that may help to promote increase participant adherence, rather than being only concerned with the details of specific intervention components.

By considering alternative evidence-based models of exercise training, and offering flexible program design in order to increase ongoing adherence to exercise, programs may also attract an increased number of participants to their service. While a lack of referral is often cited as a significant barrier to the uptake of cardiac rehabilitation, other well-known barriers include issues with the program itself such as exercise training location, session availability or the frequency and duration of attendance required [133, 136]. Knowledge that program design can be flexible, without greatly impacting on the clinical outcomes expected, may help to reassure clinicians when prescribing exercise training. Additionally, flexibility in design may overcome the resourcing issues faced by many programs in the current economic climate [137].

Future Questions

It appears obvious that structured exercise training has clear benefits over usual care, even when this includes physical activity advice, yet the question remains as to the minimal threshold of exercise intensity and volume to produce these changes. This is a pertinent fact to consider given the rise of health coaching interventions in recent years which provide exercise interventions in a format somewhere between general advice and structured exercise intervention [138, 139]. Additionally, this review appears to support the hypothesis that the threshold for global health benefits may occur at low levels of intensity in people who are deconditioned. The minimum intensities in some included trials were lower than the aerobic intensity threshold for fitness improvements of 45% VO₂R/60%HR_max proposed by Swain and Franklin [140], and consequently, it remains unclear at what point significant improvements in clinical outcome fail to occur.

Strengths and Limitations

Our study is strengthened by the use of a broad search strategy with no limitations imposed on language or date of publication. We also performed sensitivity analysis to examine the effects of missing data on all outcomes. Additionally, by using an exhaustive process to contact trial authors for missing intervention details, we were able to provide a more complete picture of exercise training than has ever been previously examined. Our analysis contains a wide range of interventions, allowing a comparison of varying exercise characteristics across a number of clinical outcomes.

The strength of our findings however is limited by the quality of the included trials. Due to poor reporting, many were rated at unclear risk of bias, and missing data were observed in a number of trials. Importantly, however, imputation of these missing data in sensitivity analysis did not significantly change the effect of cardiac rehabilitation on clinical outcomes. Many intervention details also remained missing, limiting the amount of data available to perform meta-regression. In particular, attempts at multivariable analysis (to account for any potential confounding due to correlated co-variates) were limited by the low number of interventions reporting all exercise co-variates of interest. It is not clear if these missing variables would have changed the outcome of the analysis.

It must also be recognised that the results of subgroup analyses and meta-regression performed in our study are subject to the challenges associated with interpreting complex interventions in such a manner. Separating the effects of all potential effect modifiers is difficult, particularly where this information may be unreported or unaccounted for. It must also be considered that these component parts alone may not be enough to explain the power of an intervention. A complex intervention such as cardiac rehabilitation may be more than the sum of its parts, and in reducing it down into these parts, we may lose the essence of what makes it a successful system [141]. The outcomes of such interventions may also be context-dependant, and it is not yet clear how this can be accounted for in analyses [142].

Finally, while this review contains a large number of trials from many countries, two thirds of them are more than 15 years old, and the majority of participants are still middle-aged men. Therefore, the generalisability of these findings to modern cardiac rehabilitation practice including participants of more varied aetiologies, elderly patients and women may be somewhat compromised.