Machine learning for cardiac death prediction after myocardial infarction: the power of echocardiographic data.

Abstract

Background Identifying patients at high risk of cardiac death after myocardial infarction (MI) is crucial for guiding post-MI management. Echocardiography during the index admission enables assessment of cardiac function and haemodynamics providing important prognostic information. We sought to use machine learning (ML) to identify key echocardiographic predictors of cardiac mortality following acute MI. Methods Retrospective, single-centre study of 3,202 consecutive patients admitted with acute MI and followed-up for a median of 4.5 years. All patients underwent comprehensive transthoracic echocardiography within 24 hours of the index admission. A boosted ensemble algorithm (XGBoost) was trained to predict cardiac death using clinical and echocardiographic data. The SHAP (SHapley Additive exPlanations) method was used to identify the most important predictors in the ML models. Results XGBoost exhibited the highest performance for prediction of cardiac death, with an AUC of 0.790 and accuracy of 95.4%, outperforming conventional logistic regression (AUC 0.733, accuracy 80.5%). [Formula presented] Conclusion Machine learning models, particularly XGBoost, can effectively predict cardiac mortality after MI using echocardiographic and clinical data. The medial E/e' ratio emerged as the strongest echocardiographic predictor, highlighting the importance of diastolic function assessment in risk stratification. These findings may aid the development of targeted interventions in high risk patients.Copyright © 2025 American College of Cardiology Foundation