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

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Accurate mortality prediction post-myocardial infarction (MI) is crucial for patient management.
• Conventional statistical models are often used, but machine learning (ML) methods show promise.

Purpose of the Study:

• To systematically evaluate and compare the predictive performance of various ML models for mortality after MI.
• To identify the most reliable ML models for predicting post-MI mortality.

Main Methods:

• A systematic literature search was conducted across major databases (Medline, Embase, Scopus, Web of Science).
• A meta-analysis using a bivariate random-effects model was performed on 69 eligible studies.
• Subgroup analyses and risk of bias assessments were conducted.

Main Results:

• Gradient Boosting Machines (GBM), Single Decision Tree, and Random Forest models demonstrated high predictive accuracies.
• Advanced GBMs, especially XGBoost, showed the highest certainty (AUC=0.90), precision, and minimal publication bias.
• Incorporating echocardiographic parameters into advanced GBMs improved sensitivity and specificity.

Conclusions:

• Advanced GBMs, particularly XGBoost, are the most reliable for predicting mortality in MI patients.
• Future research should focus on external validation, transparent reporting, and NSTEMI populations.