Evidence against cardiac rehabilitation
No trial of cardiac rehabilitation (CR) following myocardial infarction (MI) (not even the WHO European collaborative) demonstrates significant reduction of mortality, as do trials of secondary prevention. There is potential conflict of interest when therapists report self-evaluations. Reviews of published reports exaggerate publication bias. Meta-analyses show no significant effect of CR on mortality in recent years – since 1990, 23 trials, 6527 patients, relative risk 1.01 (0.88–1.15). It does no service to MI patients – or rehabilitation therapists – to repeat claims derived from poolings of historic trials, undertaken before many significant advances in diagnosis, acute treatment and effective secondary prevention. While CR has a role in good medical/nursing practice and continuity of care, rehabilitation therapists could be more effective elsewhere in the NHS.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Rehabilitation, to restore to health by training or therapy after illness, might not be expected to reduce mortality. Prevention or, as in this case following myocardial infarction (MI), secondary prevention to induce changes in lifestyles and treat risk factors does aim to reduce the risk—of further MI and death. The principal components of rehabilitation are health education, exercise training and psychological counselling and those of secondary prevention are risk factor control over the longer term (following education, advice and support provided in rehabilitation), revascularisation and pharmacotherapy (as indicated clinically). For purposes of this debate we will acknowledge that cardiac rehabilitation (CR) programmes include elements of both rehabilitation and secondary prevention—and thus might reduce mortality.
Many models have been promoted, ranging from weeks of inpatient therapy, training and advice by multidisciplinary teams to self-instruction at home with a booklet and cassette tape; a very heterogeneous collection of activities. Many of these have been evaluated by randomised controlled trials but most trials are too small to demonstrate whether or not rehabilitation programmes reduce mortality. Even the relatively large World Health Organization (WHO) European collaborative1 was small compared with trials of pharmaceuticals and revascularisations. While one cannot ‘prove’ a negative, accumulation and critical appraisal of relevant evidence can reduce uncertainty surrounding a ‘best estimate’ of effect.
Interpretation of findings of CR trials individually and collectively necessitates the appreciation of differences between trials of service provision and drug trials, and of limitations of the ‘art’ of reviewing (chapter 9).2 Briefly, service trials are invariably ‘doubly un-blind’ so it is somewhat easier to manipulate random allocation. The control group usually gets no intervention (seldom a ‘dummy’) so 'placebo’ or ‘Hawthorne’ effects cannot be excluded. ‘Contamination’ among controls can occur, as patients learn from ‘cases’ what they are missing and seek comparable alternatives elsewhere. Pioneer trials of innovative treatments, essential to introduce a new ‘medical technology’ and often challenging a prevailing conservative consensus, as we found with our earlier multicentre trial (chapter 10),2 tend to be ‘self-evaluations’ by the therapist with potential conflict of interest. Repetitions by ‘third party’ researchers may not confirm pioneer's findings.3 Selective reporting of outcomes—for example, omitting mortality—can influence conclusions.4 Publication bias by journals, researchers or funders (British Heart Foundation medical director, personal communication)5 ,6 tends to favour new and positive at the expense of repetitions and null findings.
Reviews have made important contributions to demonstrating efficacy and estimating effect size, when adding together trials of similar therapies, but it must be appreciated that they are observational studies. Reviews do not follow the scientific dictum, ‘first the hypothesis, then the test’ but rather ‘first the data, then selection of the acceptable’. Reviews that pool findings of highly disparate interventions as in CR require critical interpretation. It is questionable whether reviews should be universally ranked higher in the hierarchy of evidence than large primary trials. Reviews depend on what is in the literature and consequently accentuate publication bias. Funnel plots may help regarding suspicions of publication bias but they do not find missing trials. Selection is crucial; differences between different reviewers can be marked; in three reviews of psychological rehabilitation for cardiac patients only five of 61 trials were common to all. Good practice recommends setting inclusion criteria before searching the literature but it can never be truly prior. Even within a review selections can vary; in revision of that of exercise-based CR one third of pre-1999 trials were removed and replaced by a similar number of pre-1999 trials. 7 Reviewers, possibly without experience of conducting service trials, can exclude trials through misunderstanding reports, for example, confusing non-attendance at clinical review (‘missing value’ in a clinical measure) and loss to follow-up for mortality (by Office for National Statistics—or equivalent) and thereby excluding five (of 17) WHO centres, 8 the WHO report having already excluded seven of 24 collaborating centres.1 The simple addition of death rates (dead/entered) does not accommodate either right-censoring or different durations of follow-up. Several reviews of CR include frank errors, including double entry of a trial and wrong denominators in mortality tables. A radical but possibly rational step, when trial numbers and total patient numbers exceed satisfactory thresholds, might be to reclassify early small trials by pioneers as ‘hypothesis generating’ and exclude them from meta-analyses.
Exercise training has become the cornerstone of most CR programmes. 9 The first review in the Cochrane Collaboration Heart Group series reported a significant reduction of mortality in 12 ‘exercise only’ trials with OR of 0.74 (0.56 to 0.98), but less in 28 ‘exercise plus’ trials OR of 0.87 (0.71 to 1.05). 8 This apparent difference demands interpretation. While it is conceivable that the addition of other rehabilitation activities could reduce time spent on exercise and thus dilute a most active component, it should also be noted that the exercise-only trials generally predated the exercise-plus trials. The revised Cochrane review reported reduced total mortality beyond 12 months (17 of 47 trials and 5790 of 10 794 patients) at the borderline level of statistical significance—relative risk (RR) of 0.87 (95% CI 0.75 to 0.99). 7 For those with shorter follow-up (<12 months), despite counting 15 of 17 WHO centres and one other early trial a second time, the CI was wider and included the null. Trials included in the review date from 1965 to 2005 (median dates of patient entry), and mortality following MI has reduced progressively over these 40 years with revised definition (more sensitive cardio-specific enzyme markers) and radical advances in treatment. Pooling all trials reporting mortality in the later review and splitting around 1990, a date when guidelines widely recommended aspirin and β-blockers, into two groups (with similar patient numbers 4457 and 4714) yields RR before 1990 of 0.84 (0.72 to 0.97) and post-1990 of 0.99 (0.77 to 1.27).
This apparent effect on mortality in early trials, if not an artefact of publication bias, may indicate ‘real’ benefit attributable to the rehabilitation component, because there was then little effective secondary prevention. Alternatively, a possible beneficial effect may have been a consequence of the extra provision of secondary prevention medications as these became available (unfortunately not recorded), because patients included in trials tend to obtain better management than ‘normal’ and those in the rehabilitation arms of trials were being managed more closely than controls. Whether or not there is satisfactory explanation for a possible beneficial effect in early trials, there is no statistically significant effect of rehabilitation following MI on mortality in more recent trails. In the late 1990s the UK NHS, recognising important advances in the management of MI, commissioned a multicentre trial (8000 patients) of ‘comprehensive’ CR programmes as provided in the modern clinical context. 6 Regrettably for science, medical care, the NHS and patients, the trail was closed prematurely by the funders influenced by lobbyists with null findings, nonetheless consistent with those of contemporary trials and hence narrowing the confidence limits around a best estimate RR of 1.01 (0.88 to 1.15).
Reviews of the other two components of CR complement those for exercise. The meta-analysis of psychological rehabilitation reported in 17 trials (6852 patients; RR=0.89, 95% CI 0.75 to 1.05). 10 Although not statistically significant, this might imply a modest reduction. However, the two largest trials (82% of total weight) reported 0.99 and 1.01, respectively. That the ‘average’ is pulled away from the null so far by 15 small trials raises the question of potential bias. The review found some asymmetry in the funnel plot, compatible with a previous personal enquiry of authors of trials that omitted to report mortality, in which those who replied acknowledged that mortality was higher in the rehabilitation group.4 Furthermore, the two largest trials both entered patients after 1990, implying a time trend as observed in the exercise trials. With reservations over potential bias among small and predominantly early trials, we conclude that psychological interventions do not significantly reduce mortality.
The effect of the education component in typical rehabilitation programmes is not particularly easy to assess, as relatively few programmes are predominantly education based and relatively few trials have evaluated education alone (or as a principal component). Brown and colleagues11 identified 13 trials; six reported mortality (2330 patients but <40% MI), with an estimated RR of 0.79 (0.55 to 1.13).
While trials are the ‘gold standard’ for assessing efficacy and effectiveness, trials of CR generally have been small and even after 40 years the total trial database remains a relatively small proportion of the MI patient population. Observational studies could contribute but only with careful design, thorough analysis, critical appraisal and interpretation. Opportunistic analysis of data obtained for other purposes have limitations, for example, the analysis of a Medicaid patient dataset, which reported a significant reduction in mortality. 12 Although numbers were large, 70 000 matched pairs (37% MI), the analysis was unable to adjust for several highly relevant confounders including severity (eg, ejection fraction), laboratory results, lifestyle risk factors (eg, smoking) and medication. 13 Omissions of important prognostic measures render the findings almost uninterpretable. Mortality was higher than in modern trials as patients were older; therefore, if there were a mortality effect after adjustment for all potential confounders, comparison with pooled trial evidence could imply a relationship between risk and effect. In other words, CR could be more efficacious for high-risk patients.
Plausible mechanisms by which CR could reduce mortality have been postulated and significant changes in potential process variables (risk factors) have been observed, although relatively sparsely reported. 8 Risk factors may be more sensitive to CR interventions and improvements in risk factors could imply a small (currently not significantly demonstrable) beneficial effect on hard clinical endpoints. Whether improvements in risk factors can be attributed directly to CR programmes or to secondary prevention does not contradict the null findings on mortality among contemporary trials.
Experiment does not prove a negative but pooled evidence from trials undertaken in the past 20 years would imply that the effect, if any, on mortality following MI is small. Larger trails than any undertaken in the history of CR would be needed to reduce uncertainty significantly, to detect whether or not there could be a reduction of, say, 5%. Without more contemporary trial evidence there remains the possibility of a small effect, currently not demonstrable in the pooled data partly because of the relatively low mortality following MI at present. We have suggested elsewhere that any such small effect is less likely to lie with rehabilitation interventions directly than with rehabilitation being a vehicle for secondary prevention, through revascularisation and medication, lifestyle change and compliance with prescribed medication.
This is not to say that the emperor wears no clothes—but perhaps that he wears no glorious cloak. Claims that CR is as effective in reducing mortality as aspirin, β-blockers, statins and revascularisation are outdated—are not supported by contemporary trial evidence. None of the foregoing is to assert that CR is irrelevant—in some form; it provides continuity of care that is part of good medical and nursing practice, and feeds into secondary prevention. Patients, whether or not they attend, seek and deserve good communication with doctors and nurses regarding information, advice and reassurances and appreciate continuity of care following discharge after a life-threatening event, particularly early follow-up by a general practitioner. This can be done without dedicated programmes of outpatient education, exercise training and counselling. CR is not irrelevant but its provision could benefit from revision. The ‘life cycle of a health technology’ recognises difficulty to introduce when novel and potentially beneficial, delayed adoption not uncommonly after its peak usefulness and difficulty to replace (or reform) when established even when superseded by a newer technology. 14 With present clinical management the effect of CR on mortality in typical MI patients is not significant. The service implication might be to redeploy specialist, trained and experienced therapists into other patient groups with greater needs and greater potential benefits; and greater potential savings to the NHS. The introduction of CR following MI was innovative before the advent of effective secondary prevention; it is time for rehabilitation therapists to be innovative again—and move on.
Contributors RW wrote the first draft; RW and DJ agreed the final manuscript.
Competing interests None.
Provenance and peer review Commissioned; internally peer reviewed.