When meta-analyses fall short: the need to reframe evidence generation through real-world data in cardiac rehabilitation

This editorial refers to ‘Exercise-based cardiac rehabilitation for coronary heart disease: the CaReMATCH individual participant data meta-analysis’, by N.A. Stens et al., https://doi.org/10.1093/eurjpc/zwaf629. Cardiac rehabilitation (CR) emerged in the mid-20th century as a supervised exercise programme for patients recovering from acute myocardial infarction, giving rise to the term exercise-based CR. Since then, it has evolved into comprehensive CR, encompassing several core components, and is now recognized as a patient-centred, evidence-based strategy for secondary prevention of cardiovascular diseases.1 In patients with coronary artery disease (CAD), CR improves physical and psychological function, adherence to pharmacotherapies, risk-factor control, and health-related quality of life (HRQoL), while reducing hospital readmissions and cardiovascular mortality.1,2 Consequently, CR holds a Class I, Level of Evidence A recommendation in both European and American guidelines.1,3 Nevertheless, in the modern era of cardiovascular care, defined by potent lipid-lowering therapies, advanced antithrombotic regimens, and early revascularization, the cardiovascular community continues to debate whether traditional, exercise-based CR provides additional benefit beyond these contemporary interventions. Importantly, comprehensive CR has been associated with a reduction in all-cause mortality even under modern management conditions. However, this evidence is derived primarily from systematic reviews that include observational studies, which remain susceptible to residual confounding.4 Against this background, the individual participant data (IPD) meta-analysis by Stens et al.5 provides a timely and methodologically robust reassessment of the effectiveness of exercise-based CR in patients with CAD. The analysis pooled data from contemporary randomized controlled trials (RCTs) published since 2010, reflecting the evolution of CR towards more flexible and digitally supported delivery models. Most interventions were home-based or hybrid programmes, integrating telemonitoring or remote supervision rather than traditional centre-based exercise training. Across studies, exercise modalities and intensities varied considerably, from conventional aerobic and resistance training to low-intensity mind–body approaches. In several studies, the intervention primarily consisted of walking at low-to-moderate intensity, which aligns more closely with general physical activity recommendations than with individually prescribed exercise training. Notably, two studies consisted of exercise-only interventions, and none could be considered comprehensive CR. The largest contributing trial, the Yoga-CaRe study, evaluated a structured, yoga-based programme in an Asian post-myocardial infarction population. Across all included trials, CR was associated with reduced all-cause and cardiovascular hospitalizations and improved HRQoL, whereas effects on major adverse cardiovascular events and mortality were neutral. A ceiling effect may have been present, which could explain why benefits were mainly observed among those with low baseline exercise capacity, reduced LVEF, and a higher comorbidity burden. The authors, an experienced group with longstanding contributions to CR research, present their work with rigour and transparency. Nonetheless, several aspects merit consideration. Approximately 80% of participants in the meta-analysis were drawn from a single large trial, the Yoga-CaRe study, conducted in a predominantly Asian, male, post-myocardial infarction population. As a result, >90% of the pooled cohort represented acute coronary syndromes, while women and patients with chronic coronary syndromes were markedly under-represented. The yoga-based intervention itself, though innovative, differed from conventional CR in exercise type, intensity, and delivery mode, combining low-to-moderate intensity endurance and resistance components, largely delivered in a hybrid home-based setting with lower session frequency than standard European programmes. Moreover, major cardiovascular events and HRQoL were assessed at different follow-up intervals, and hospitalizations were not pre-specified endpoints, introducing potential inconsistencies across studies. It is also noteworthy that the largest classical RCT on comprehensive CR, the RAMIT trial,6 which reported neutral effects on mortality, was not included in the present IPD analysis. This omission underscores the difficulties of integrating all relevant contemporary evidence, as source data are often incomplete or inaccessible. Together, these factors contribute to heterogeneity and limit the generalizability of pooled outcomes to non-Asian populations, women, and patients with chronic coronary syndromes. These limitations reflect broader challenges in contemporary CR research. While meta-analyses of both trial-and patient-level data have consistently confirmed some benefits of exercise-based CR, conducting additional large-scale RCTs has become increasingly difficult, both ethically and practically. After decades of accumulated evidence supporting its efficacy, true equipoise is limited, and many patients are understandably reluctant to be randomized to non-rehabilitation control groups. This lack of equipoise contributed to the recruitment difficulties and eventual early termination of the RAMIT trial, despite its robust design.6 Consequently, studies published since 2010 have shifted focus away from re-testing classical, centre-based CR. Instead, they explore alternative delivery models, home-based, hybrid, digitally supported, or non-traditional interventions such as yoga. While this diversification enhances accessibility, it also increases intervention heterogeneity, complicating pooled analyses. Furthermore, the European Association of Preventive Cardiology has established standards defining both conventional CR and cardiac telerehabilitation.7,8 This raises the question of how innovative an intervention can be while still being accurately classified as CR. The limited availability of source data and incomplete reporting of outcomes further impede comprehensive meta-analyses, underscoring the need for new research strategies. Rehabilitation is inherently a complex intervention, combining behavioural, educational, psychosocial, and organizational components. Future studies should therefore align with the updated Medical Research Council framework for complex interventions, which emphasizes understanding how, for whom, and under what circumstances interventions achieve benefit.9 Key recommendations include defining a clear programme theory linking intervention components to outcomes and context; incorporating stakeholder and patient perspectives at every stage; considering contextual and system-level factors that influence implementation; allowing iterative refinement and flexible adaptation; and embedding economic evaluation to support scalability. Applying this framework will move CR research beyond isolated efficacy testing towards understanding real-world effectiveness within diverse healthcare ecosystems. To complement this shift, the field must embrace systematic collection and integration of real-world data. Existing national and international CR registries provide an excellent foundation but remain fragmented.10 Harmonizing data standards, outcome definitions, and reporting structures across registries will enable large-scale analyses encompassing diverse populations and healthcare systems. Linking registry data with electronic health records and digital health tools will open the door to big-data analytics, facilitating risk prediction, benchmarking, and personalized rehabilitation pathways. Emerging analytical frameworks such as TRIALSCOPE demonstrate how real-world data can approximate or even simulate RCTs.11 Leveraging biomedical language models, employing advanced probabilistic modelling and using causal-inference techniques, TRIALSCOPE structured one million oncology patient records and reproduced hazard ratios from 11 lung cancer trials, achieving statistical equivalence with the original RCT results. Moreover, it simulated unperformed or prematurely terminated trials, such as the EMPHASIS study, validating its predictive accuracy. These advances illustrate a paradigm shift: real-world data, when analysed through causal frameworks, can emulate trial conditions and accelerate evidence generation in complex, ethically constrained fields. Applying such approaches to CR could enable large-scale virtual evaluations of effectiveness across diverse delivery models, thereby complementing conventional research and bridging evidence gaps (Figure 1). Evolving evidence pathways in cardiac rehabilitation. In summary, Stens et al. are to be congratulated for addressing an important question in CR despite scarce and heterogeneous source data. Their rigorous and transparent work advances the field and highlights both the potential and limitations of traditional evidence synthesis. Moving forward, modern data analytic approaches, combining registry harmonization, real-world evidence, and causal modelling, will likely prove more suitable for evaluating complex and multifaceted interventions like CR. As these methods mature, they should be integrated into future ESC frameworks for grading evidence,12 ensuring that evolving forms of clinical proof for CR are appropriately recognized in the cardiovascular community.