XGBoost based machine learning prediction model for major adverse cardiovascular events after PCI in STEMI patients

Abstract

ST-segment elevation myocardial infarction (STEMI) demands urgent reperfusion. Despite major strides in percutaneous coronary intervention (PCI), major adverse cardiovascular events (MACE) persist, calling for more comprehensive risk evaluation and management. Clinical data from 1,011 STEMI patients who underwent PCI were included in this study. A total of 37 clinical variables—including demographic characteristics, hemodynamic parameters, and laboratory indicators—were initially collected. Six machine learning algorithms, including random forest (RF), least absolute shrinkage and selection operator (Lasso), support vector machine (SVM), extreme gradient boosting (XGBoost), gradient boosting machine (GBM), and feedforward neural network (FNN), were employed to select key predictors and construct a MACE prediction model. Partial dependence plots (PDP) and Shapley additive explanations (SHAP) were subsequently used to interpret model variables and visualize the decision-making process. The XGBoost model, based on nine key predictors, demonstrated the best performance (AUC: 0.81 for the training set, 0.71 for the test set). SHAP analysis identified LCX occlusion, KILLIP classification, lymphocyte count, LVEF, AST, monocyte count, gender, alcohol consumption, and the neutrophil-to-lymphocyte ratio (NLR) were positively associated with the risk of MACE, whereas higher lymphocyte levels and male sex were negatively associated with the occurrence of MACE. PDP results revealed the synergistic effects of AST, LCX, monocyte count, LVEF, and lymphocyte count on MACE risk. This study provides a clinically promising predictive model for MACE risk assessment in STEMI patients post-PCI. The model, using interpretable methods, identifies key clinical indicators that critically influence MACE occurrence, offering valuable insights for clinical decision-making and patient management.