Mortality risk prediction in NSTE-ACS following PCI: Insights from a real-world cohort

Abstract

Background Non–ST-segment elevation acute coronary syndrome (NSTE-ACS) is a major contributor to cardiovascular mortality, yet reliable tools for individualized mortality prediction remain limited. Machine learning offers the potential to enhance prognostic accuracy in this high-risk population. Methods A total of 1,495 patients with NSTE-ACS who underwent percutaneous coronary intervention (PCI) were retrospectively analyzed. Eight clinical and laboratory variables were selected through univariate and multivariate logistic regression. Five machine learning models-logistic regression, random forest, XGBoost, LightGBM, and naïve Bayes-were constructed. Model performance was evaluated using area under the curve (AUC) and calibration curves. Results Age, diabetes mellitus, and ejection fraction were identified as independent predictors of all-cause mortality. Among all models, LightGBM achieved the highest AUC (0.847), followed by XGBoost (0.822), both of which demonstrated superior discrimination and calibration compared to traditional logistic regression and other algorithms. Calibration analysis showed excellent agreement between predicted and observed mortality in both training and test cohorts. Conclusion Gradient boosting models, particularly LightGBM and XGBoost, significantly improve mortality prediction in NSTE-ACS patients after PCI. These models may facilitate more accurate risk stratification and guide personalized post-procedural management strategies in clinical practice.