Development and validation of a machine learning-based prediction model for 3-year major adverse cardiovascular events in patients undergoing coronary artery bypass grafting: a multicenter study

Major adverse cardiovascular events (MACE) represent a residual risk following coronary artery bypass grafting (CABG) and adversely affect long-term prognosis. Accurate risk prediction is critical for postoperative risk stratification and secondary prevention. We aimed to develop and validate a machine learning model for early postoperative estimation of 3-year MACE risk after CABG. This multicenter retrospective study enrolled 3,930 adults undergoing CABG between January 2019 and December 2022 at eight cardiac centers in Jiangsu Province, China. We collected 232 preoperative, intraoperative, and early postoperative variables. Data from 3,018 patients at Nanjing First Hospital were divided into training and internal validation sets; data from 912 patients at seven other centers formed the external validation set. Features were selected using minimum redundancy maximum relevance (mRMR) followed by least absolute shrinkage and selection operator (LASSO) regression. Ten machine learning models were developed and compared. Performance was evaluated using the area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUC-PR), accuracy, sensitivity, specificity, F1-score, Brier score, calibration curves, and decision curve analysis. SHapley Additive exPlanations (SHAP) was used to interpret the final model. The 3-year MACE incidence was 16.41% (645/3,930). Seventeen predictors were selected. The Extremely Randomized Trees (ET) model, selected based on internal validation performance, achieved AUROCs of 0.90 (95% CI: 0.87–0.93) in the internal validation set and 0.82 (95% CI: 0.78–0.86) in the external validation set, with an external AUC-PR of 0.59. At the prespecified probability threshold of 0.50, external accuracy, sensitivity, specificity, and F1-score were 0.87, 0.54, 0.94, and 0.59, respectively. External Brier score, calibration intercept, and slope were 0.100, −0.20, and 0.80, respectively. Postoperative left ventricular ejection fraction was the most important predictor, followed by preoperative neutrophil percentage and preoperative New York Heart Association (NYHA) class. The ET model provided externally validated estimates of 3-year MACE risk after CABG. The model is intended for early postoperative risk stratification once required postoperative inputs are available, although threshold-specific sensitivity was limited. SHAP analysis described predictor contributions, and a web-based calculator was developed for further evaluation. Prospective evaluation and threshold validation are required before clinical implementation.