Machine Learning Model Predicts Non-Responders to Cardiac Rehabilitation

Machine Learning Predicts Cardiac Rehab Outcomes

A new machine learning model can predict which patients with coronary artery disease are likely to experience little or no cardiorespiratory fitness improvement from standard exercise-based cardiac rehabilitation, according to a study published on May 5, 2026, in the Journal of Sport and Health Science. Researchers from the University of Witten/Herdecke and DRV Clinic Königsfeld in Germany, alongside collaborators from FORTH in Greece, developed and evaluated ten algorithms to address a persistent clinical gap where roughly one in five patients fails to show meaningful aerobic capacity gains after completing supervised training programs.

Key Clinical Takeaways

• A machine learning Random Forest model accurately classifies cardiac rehabilitation responders and non-responders with 77 percent accuracy prior to training.
• The study analyzed 353 patients with coronary artery disease who completed three to four weeks of inpatient rehabilitation following heart attacks or surgical procedures.
• Baseline indicators such as breathing efficiency during exercise testing and arterial stiffness via pulse wave analysis proved more predictive than traditional factors like age or disease severity.

The Challenge of Inter-Individual Variability

Exercise training serves as a cornerstone of cardiac rehabilitation for individuals recovering from heart attacks, angioplasty, stent placement, or bypass surgery. Regular supervised physical activity improves peak oxygen uptake—measured as V̇O₂peak—which stands as a primary indicator of long-term cardiovascular survival and functional capacity. Yet, significant inter-individual variability remains a challenge in clinical cardiology. A substantial proportion of participants achieve minimal or zero change in V̇O₂peak after standard interventions, leaving them at a higher risk for future cardiovascular events.

Random Forest Model Outperforms Standard Metrics

To tackle this limitation, the research team analyzed baseline data from 353 patients undergoing three to four weeks of inpatient rehabilitation. The evaluation utilized cardiopulmonary exercise testing and non-invasive pulse wave analysis to measure arterial stiffness and vascular function. Out of ten tested machine learning algorithms, the Random Forest model emerged as the strongest performer. By aggregating outputs from multiple decision trees, the model identified complex non-linear relationships among biological variables to classify responders and non-responders with 77 percent accuracy before training commenced.

Physiological Dynamics Drive Predictions

Standard clinical characteristics at baseline—including age, sex, body mass index, and the specific severity of underlying heart disease—did not reliably separate responders from non-responders. Instead, explainable artificial intelligence techniques, specifically SHAP analysis, revealed that physiological dynamics drove the predictions. Patients who utilized more ventilation per unit of oxygen consumed, possessed less breathing reserve, or demonstrated elevated pulse wave velocity exhibited a lower likelihood of fitness improvement. Additionally, specific pharmacological treatments, including angiotensin II receptor blockers and calcium channel blockers, influenced the algorithmic predictions.

Corresponding author Prof. Boris Schmitz of the University of Witten/Herdecke noted that utilizing data routinely collected at the start of rehabilitation can transform clinical care pathways. Identifying non-responders early allows multidisciplinary care teams to pivot away from a one-size-fits-all approach. Investigators caution that the model requires further validation across older demographics, diverse clinical comorbidities, and alternative rehabilitation program structures.

*Disclaimer: The information provided in this article is for educational and scientific communication purposes only and does not constitute medical advice. Always consult with a qualified healthcare provider regarding any medical condition, diagnosis, or treatment plan.*