Key result
XGBoost model replicates clinician-assigned CV risk classifications for exercise training with ~79% accuracy.
Why the study?
Existing risk classification criteria for cardiac rehabilitation vary among organizations, necessitating standardized risk classification for cardiopulmonary exercise training.
Can machine learning models accurately replicate clinician-assigned cardiovascular risk stratification for cardiopulmonary exercise training in cardiovascular patients?
Observational (n=1,163)
Yes
Can machine learning models accurately replicate clinician-assigned cardiovascular risk stratification for cardiopulmonary exercise training in cardiovascular patients?
Effect estimate: Accuracy 79.31% (95% CI 71.55-86.21)
A machine-learning model using XGBoost and LASSO feature selection can accurately categorize exercise-related cardiovascular risk types to guide cardiopulmonary rehabilitation.
ML models may aid risk stratification in cardiac rehab; hypothesis-generating pending prospective validation.
Purpose:In cardiovascular patients, precise risk stratification is crucial for minimizing exercise-related risks and optimizing treatment efficacy during cardiac rehabilitation.Existing classification criteria vary among organizations, necessitating a comprehensive assessment of the underlying conditions and medical tests.We developed machine-learning models to assist in standardizing risk classification for cardiopulmonary exercise training.Materials and Methods: Using retrospective data from 1163 patients across three institutions, we developed an AI model to reproduce clinician-assigned risk stratification for exercise-related cardiovascular events.Data pre-processing involved excluding variables with numerous missing values and conducting multiple imputation using chained equations, resulting in 46 variables for analysis.Logistic regression, support vector machines, extreme gradient boosting (XGBoost), and random forests were employed, optimizing hyperparameters via a grid search.In feature selection, the least absolute shrinkage and selection operator (LASSO), with a range of regularization strengths, was applied to the L1 term to determine the optimal penalty value.Results: Several machine-learning models were evaluated for their ability to replicate clinician-assigned cardiovascular risk-type classifications.XGBoost achieved the best performance with 76.72% accuracy and an area under the curve (AUC) of 88.76%.After applying LASSO-based feature selection, performance improved, with the best result (α=0.001)showing 79.31% accuracy and an AUC of 89.40%.Feature importance analysis identified ventilatory efficiency, maximal systolic blood pressure, and maximal oxygen consumption as key features for risk stratification based on aerobic exercise load test data, along with weight. Conclusion:A machine-learning model was established to categorize exercise-related risk types during cardiopulmonary rehabilitation, serving as a decision-support tool for the safe and effective training of cardiovascular patients.
No takes yet. Share an insight, caveat, or question.
Na et al. (2026) conducted an observational in Cardiovascular disease (n=1,163). XGBoost machine learning model vs. Clinician-assigned risk stratification was evaluated on Accuracy of risk-type classification (Accuracy 79.31%, 95% CI 71.55-86.21). An XGBoost machine learning model using LASSO-selected features achieved 79.31% accuracy and an AUC of 89.40% in replicating clinician-assigned cardiovascular risk classifications for exercise training.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: