AI-guided integration of aging-related functional assessment to predict nonresponse to cardiac rehabilitation in heart failure.

BACKGROUND

Cardiac rehabilitation (CR) improves functional capacity and outcomes in patients with heart failure (HF). However, a clinically significant subgroup of patients exhibits limited improvements in aerobic capacity, indicating substantial heterogeneity in response and limiting the effectiveness of standardized rehabilitation programs.

OBJECTIVE

To develop an interpretable artificial intelligence (AI)-guided framework to identify predictors of nonresponse to CR and support personalized rehabilitation strategies in HF, with age-related functional decline as a key determinant of heterogeneity.

METHODS

Data on patient characteristics, exercise interventions, and response rates were extracted from Randomized controlled trials evaluating exercise-based CR in patients with HF that reported changes in peak oxygen uptake (VO₂peak) or functional capacity. Nonresponse was defined as <10% improvement in VO₂peak after CR. Coupling of eXtreme Gradient Boosting (XGBoost) and SHapley Additive exPlanations (SHAP) was introduced to characterize age-related determinants of nonresponse to CR in HF.

RESULTS

Across 38 randomized trials including 5,198 patients, 30-45% failed to achieve clinically meaningful improvements in VO₂peak after CR. Advanced age was consistently associated with attenuated benefit, with older patients showing reduced gains. Non-response was linked to higher age, lower baseline aerobic capacity, metabolic comorbidities, impaired skeletal muscle function, and adverse psychosocial status. An interpretable AI-guided approach integrating aging-associated functional and psychosocial parameters was introduced to CR in patients with HF. Principal predictors of nonresponse included impaired baseline aerobic capacity, metabolic comorbidities, diminished skeletal muscle function and adverse psychosocial factors, delineating an age-modulated vulnerability to impaired rehabilitation adaptation.

CONCLUSION

Age is associated with cardiac rehabilitation response in heart failure, with heterogeneous effects across outcomes. An AI-driven framework can be introduced to identify age-related non-response phenotypes, supporting precision rehabilitation strategies.