Machine learning-based hybrid risk estimation system (ERES) in cardiac surgery: Supplementary insights from the ASA score analysis

Abstract

Accurate prediction of postoperative mortality risk after cardiac surgery is essential to improve patient outcomes. Traditional models, such as EuroSCORE I, often struggle to capture the complex interactions among clinical variables, leading to suboptimal performance in specific populations. In this study, we developed and validated the Ensemble-Based Risk Estimation System (ERES), a machine learning model designed to enhance mortality prediction in patients undergoing coronary artery bypass grafting and/or valve surgery. A retrospective analysis of 543 patients was performed using six machine learning algorithms applied to preoperative clinical data to assess predictive accuracy and clinical outcomes. Feature selection techniques, including Gini importance, Recursive Feature Elimination, and Adaptive Synthetic Sampling, were employed to improve accuracy and address class imbalance. ERES, which utilizes 15 key features, demonstrated superior predictive performance compared to EuroSCORE I. Calibration plots indicated more accurate probability estimates, whereas SHAP analysis identified creatinine, age, and left ventricular ejection fraction as the most significant predictors. The decision curve analysis further confirmed the superior clinical utility of ERES across a range of decision thresholds. Additionally, although the American Society of Anesthesiologists (ASA PS) score had limited predictive power independently, its combination with EuroSCORE I enhanced the predictive performance. Integrating machine learning models like ERES into clinical practice can improve decision making and patient outcomes although external validation is warranted for broader implementation.

Fig 1. ERES hybrid model development process integrating machine learning models and feature selection.

Fig 2. Comparison of performance metrics between previous and current studies.

Fig 3. AUC curves (A) and model performance comparisons (B).

Fig 4. Key predictors of Gini importance.

Fig 5. Calibration curves comparing EuroSCORE I and ERES models with Brier scores.

Fig 6. Performance of the ERES model with 95% confidence intervals (A) and precision-recall curve for ERES model (B).

Fig 7. Contributions of input features to mortality prediction in the ERES model.

Fig 8. Decision curves illustrating the clinical utility of EuroSCORE I, ERES, and ML models in forecasting postoperative mortality are presented.

Fig 9. ROC Curve for mortality prediction using ASA classification (A) Mortality prediction performance: ASA, EuroSCORE, and combined models. (B).