Use of machine learning to predict peak cardiorespiratory fitness in patients with cardiovascular disease

Abstract Background Cardiorespiratory fitness (CRF) is a strong predictor of mortality and risk of cardiovascular disease (CVD). Little is known, however, about the factors influencing improved CRF in patients participating in cardiac rehabilitation (CR) programs. This study aimed to develop machine learning (ML) models to predict peak CRF before and after CR. Methods This study was retrospective cross-sectional study. Data from 162 patients with CVD were analyzed. Two predictive tasks were employed: Task 1 estimated peak oxygen consumption (VO₂ peak) using baseline clinical and functional data, and Task 2 predicted changes in VO₂ peak after CR by additionally considering inter-visit exercise quantities and pre-CR cardiopulmonary exercise test results. Four linear regression models and six ML models were trained and validated through 5-fold cross-validation. Results The CatBoost and XGBoost models exhibited the highest predictive performance on both tasks, effectively forecasting VO₂ peak values before and after CR. Task 1 highlighted the importance of the six-minute walk distance, age, Korean Activity Scale Index, and hand grip strength in predicting the initial VO₂ peak. Task 2 highlighted the importance of the pre-CR VO₂ peak in predicting the post-CR VO₂ peak and ΔVO₂ peak, although the direction of the correlation was opposite. Conclusions The application of ML models provides a powerful tool for predicting peak CRF in patients with CVD undergoing CR, both at the initial assessment and after completing rehabilitation programs.