Machine Learning-Based Prediction of Early Left Ventricular Function After STEMI.

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Left ventricular (LV) function and lactate dynamics are crucial prognostic indicators post-ST-segment elevation myocardial infarction (STEMI).
• Early identification of patients at risk for impaired LV function or hypoperfusion is vital for improving outcomes.
• Traditional statistical methods have limitations in predicting these outcomes, often due to small sample sizes and categorical data.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting continuous left ventricular ejection fraction (LVEF) and lactate levels in STEMI patients.
• To enhance the accuracy of risk stratification beyond traditional methods using routinely available data.
• To assess the performance and interpretability of different machine learning algorithms for this application.

Main Methods:

• Retrospective analysis of 2132 STEMI patients, with 1608 included after data preprocessing.
• Training of Decision Tree, Random Forest, and XGBoost regression models using 38 demographic, clinical, procedural, and laboratory variables.
• Evaluation of model performance using metrics like MSE, RMSE, MAE, R², and MAPE, with SHAP for interpretability.

Main Results:

• Ensemble models, particularly XGBoost, demonstrated superior performance in predicting LVEF (R² = 0.35).
• Lactate prediction showed moderate accuracy (R² = 0.42 for admission, 0.47 for peak levels).
• Key predictors identified include cardiogenic shock, left anterior descending (LAD) culprit lesions, and peak lactate levels.

Conclusions:

• Machine learning enables individualized prediction of LV function and lactate dynamics post-STEMI using routinely collected data.
• XGBoost models show consistent, clinically meaningful predictive performance and generalizability.
• These ML models support early, data-driven risk stratification in acute cardiac care settings.