Related Experiment Videos
Machine Learning-Based Risk Stratification Model to Guide Cardiopulmonary Exercise Training in Cardiovascular
Ha Eun Na1, Gi Pyo Lee2, Young Jae Kim3
1Department of Physical Medicine and Rehabilitation, Gachon University Gil Medical Center, Incheon, Korea.
Yonsei Medical Journal
|July 24, 2026
Summary
Machine learning models standardize cardiovascular risk stratification for cardiopulmonary exercise training. An AI model achieved 79.31% accuracy, improving patient safety and treatment efficacy in cardiac rehabilitation.
Area of Science:
- Cardiology
- Artificial Intelligence
- Exercise Physiology
Background:
- Precise risk stratification is vital for cardiovascular patients undergoing cardiac rehabilitation to minimize exercise risks and enhance treatment effectiveness.
- Current risk classification methods vary across organizations, highlighting the need for standardized assessment of patient conditions and medical test results.
- Developing a consistent approach to risk stratification is essential for optimizing cardiopulmonary exercise training protocols.
Purpose of the Study:
- To develop and validate machine-learning (ML) models for standardizing risk classification in cardiopulmonary exercise training for cardiovascular patients.
- To create an AI-driven decision-support tool that assists clinicians in accurately categorizing exercise-related risk types.
- To improve the safety and efficacy of cardiac rehabilitation programs through enhanced risk stratification.
Main Methods:
- Retrospective data from 1163 patients across three institutions were utilized to train ML models.
- Data pre-processing included imputation for missing values, resulting in 46 variables for analysis.
- Models evaluated included logistic regression, support vector machines, extreme gradient boosting (XGBoost), and random forests, with hyperparameter optimization via grid search and feature selection using LASSO.
Main Results:
- XGBoost demonstrated the highest performance among initial models, achieving 76.72% accuracy and an AUC of 88.76%.
- Feature selection using LASSO improved model performance, with the optimal model (α=0.001) reaching 79.31% accuracy and an AUC of 89.40%.
- Key features identified for risk stratification included ventilatory efficiency, maximal systolic blood pressure, maximal oxygen consumption, and weight.
Conclusions:
- A robust ML model was successfully developed for categorizing exercise-related risk types in cardiopulmonary rehabilitation.
- This AI tool serves as a valuable decision-support system for clinicians.
- The model facilitates safer and more effective exercise training for cardiovascular patients in rehabilitation settings.