Comparing Machine Learning Models for Predicting Mortality after Myocardial Infarction: A Systematic Review and Meta-analysis.

Introduction: Accurate prediction of mortality following myocardial infarction (MI) is critical for timely identification of high-risk patients and optimization of interventions. Conventional statistical models are commonly used; however, advanced machine learning (ML) methods are being increasingly recognized. This meta-analysis aimed to systematically evaluate and compare the predictive performances of various ML models. Methods: A systematic search of the Medline (via PubMed), Embase, Scopus, and Web of Science databases was conducted up to January 9, 2025. A total of 14933 articles were identified, of which 330 underwent a full-text review and 69 met the inclusion criteria. The meta-analysis was conducted using a bivariate random-effects model in the 'midas' package of STATA 14. Subgroup analyses were conducted based on the follow-up duration and selected clinical features. The risk of bias was assessed using the QUAPAS. Publication bias and evidence certainty were assessed using Deeks' funnel plots and GRADE framework, respectively. Results: Gradient Boosting Machines (GBM), Single Decision Tree Models, and Random Forest models yielded similarly high predictive accuracies. Advanced GBMs, particularly XGBoost (AUC = 0.90, 95% CI: 0.87-0.92; sensitivity = 0.78, 95% CI: 0.74-0.82; specificity = 0.87, 95% CI: 0.83-0.89), showed the highest evidence certainty due to precision and minimal publication bias. Across advanced GBMs, adding echocardiographic parameters increased the sensitivity from 0.77 to 0.83 and specificity from 0.85 to 0.90, indicating a clinically meaningful yet resource-dependent gain in discrimination. Conclusions: Advanced Gradient Boosting Machines, particularly XGBoost, currently provide the most reliable mortality predictions in patients with MI. Future research should emphasize external validation, transparent reporting of feature selection, detailed data preprocessing, and dedicated studies on populations with NSTEMI.