Interval Training Tops the Field for Boosting Fitness in Coronary Heart Disease, Huge Analysis Finds

For millions of people living with coronary heart disease, exercise is not optional advice from a well-meaning physician — it is a core component of treatment, as central to recovery as medication or stents. Yet a stubborn question has shadowed cardiac rehabilitation for decades: which kind of exercise actually works best? A sweeping new analysis published in BMC Medicine by researchers from Central China Normal University and Macao Polytechnic University takes one of the most ambitious runs at that question to date, combining a network meta-analysis of 129 randomised controlled trials with machine learning tools borrowed from artificial intelligence research. The answer, in short, is that interval aerobic training and combined aerobic-plus-resistance programmes lead the pack for improving peak oxygen uptake, the single most powerful predictor of survival in cardiac patients.

The study, led by Jingyi Xie and Zhide Liang with corresponding author Bin Wang, pooled data from 7,760 participants with coronary heart disease across 129 randomised controlled trials. That scale matters. Individual trials of exercise in cardiac patients are often small, brief, and heterogeneous, making it difficult to compare one training modality against another with any statistical confidence. By using a network meta-analysis — a technique that simultaneously compares multiple interventions, both through direct head-to-head trials and indirect comparisons across the network of studies — the team could rank exercise modalities that had rarely, if ever, been tested against each other in the same experiment.

The outcome the researchers focused on was peak oxygen uptake, or VO₂peak, the maximum rate at which the body can consume oxygen during incremental exercise. Cardiologists prize this measurement because it integrates the performance of the entire oxygen transport chain: the lungs, the heart’s pumping capacity, the vasculature, and the oxidative machinery of skeletal muscle. In patients with coronary heart disease, higher VO₂peak consistently tracks with lower mortality, fewer hospitalisations, and better quality of life, which is why guidelines from bodies such as the American College of Sports Medicine place cardiorespiratory fitness at the centre of exercise prescription.

The headline finding is that most exercise modalities outperformed control conditions, but two stood out. Interval aerobic exercise training — sessions built around alternating periods of higher and lower intensity — produced a mean improvement of 3.43 millilitres per kilogram per minute in VO₂peak compared with control, with a 95 percent credible interval of 2.87 to 3.99. Combined aerobic and resistance exercise training came next, with a mean difference of 2.49 (95 percent credible interval: 1.81 to 3.17). For context, even modest gains in peak oxygen uptake on the order of one metabolic equivalent are associated with clinically meaningful reductions in cardiovascular risk, so differences of this magnitude across an entire rehabilitation population are far from trivial.

Why might interval training hold an edge? The physiological rationale is that repeated bouts near a patient’s tolerance impose a potent stimulus on both central cardiovascular function and peripheral muscle oxidative capacity, while the recovery intervals allow total work to accumulate without excessive fatigue. The combined-training result is similarly intuitive: resistance work complements aerobic conditioning by improving skeletal muscle strength and mass, which can raise the ceiling on whole-body oxygen consumption. But the authors are careful to temper these rankings. They report substantial heterogeneity across trials and describe the certainty of the modality-specific comparisons as low, meaning the ordering of interventions should be read as exploratory rather than definitive.

What sets this analysis apart from earlier meta-analyses is its second analytical engine. The team deployed a meta-analytic random forest, or MARF, a machine learning model trained on study-level characteristics to predict how large an exercise effect would be, and then interpreted the model using Shapley Additive Explanations, or SHAP. SHAP values, a technique rooted in cooperative game theory, quantify how much each feature — such as participant age, intervention length, or training intensity — contributes to each prediction. This approach lets researchers probe heterogeneity in a way traditional subgroup analyses cannot, revealing which variables genuinely drive differences between trials rather than merely appearing to.

The machine learning exploration converged on a clear signal: intervention length emerged as an influential study-level moderator of effect. In plain terms, longer programmes were associated with bigger improvements in fitness. This aligns with the biology of training adaptation, where structural and enzymatic remodelling of heart and muscle tissue accrues over weeks and months, not days. It also carries a practical warning for health systems that fund short courses of cardiac rehabilitation and then discharge patients, potentially cutting the adaptation curve short at its steepest point.

The dose-response analyses added further nuance. Using meta-regression models, the researchers found that greater weekly exercise dose was generally associated with larger improvements in VO₂peak — more volume, more benefit, at least across the range studied. Per-session dose, however, told a different story: the relationship was inverted-U-shaped, rising to a peak and then declining. That shape suggests there is an optimal amount of work per session beyond which additional load yields diminishing, or even counterproductive, returns — a pattern consistent with the idea that overly demanding single sessions may compromise quality, adherence, or safety in a cardiac population.

The authors frame their conclusions with deliberate caution. Exercise-based cardiac rehabilitation does improve VO₂peak in adults with coronary heart disease, they conclude, but the substantial heterogeneity and low-certainty evidence limit confidence in the precise ranking of modalities. They recommend that exercise dose and programme duration be considered alongside patient characteristics, comorbidity burden, and safety when designing rehabilitation programmes, and they stress that these exploratory, aggregate-level findings require prospective validation before they can guide individualised exercise prescriptions. Notably, the research received no specific external funding, and the authors declare no competing interests.

Even with those caveats, the study marks a methodological milestone. Pairing network meta-analysis with interpretable machine learning offers a template for how exercise science — a field drowning in small, heterogeneous trials — might extract more reliable guidance from the evidence it already has. For clinicians, the actionable takeaways are refreshingly concrete: favour interval-based aerobic work or combined aerobic-resistance programmes where feasible, extend programme duration where possible, and think in terms of weekly dose rather than pushing any single session to its limit. For patients, the message is simpler and older than any algorithm: the exercise you do consistently, at an appropriate dose, over a long enough period, is what reshapes the heart’s capacity — and with it, the odds of a longer life.

Subject of Research: Comparative effects of exercise modalities on peak oxygen uptake in coronary heart disease rehabilitation

Article Title: Exercise-based management of patients with coronary heart disease: an integrated network meta-analysis and machine learning exploration

Article References: Xie, J., Liang, Z., Wang, B., Cai, X., & Gan, Z. (2026). Exercise-based management of patients with coronary heart disease: an integrated network meta-analysis and machine learning exploration. BMC Medicine. https://doi.org/10.1186/s12916-026-05229-5

Image Credits: AI Generated

DOI: 10.1186/s12916-026-05229-5

Keywords: coronary heart disease, cardiac rehabilitation, peak oxygen uptake, interval training, network meta-analysis, machine learning, random forest, SHAP, exercise dose-response, aerobic exercise, resistance training, VO2peak