AI and the World's Hard Problems · Medicine & drug discovery · Nature (computational biology) · 2026-10-21
Machine learning-based ensemble model for the risk stratification of myocardial injury in physical examination data
A machine learning ensemble stratifies the risk of myocardial injury, an early and common manifestation of myocarditis, from routine health check-up data. A hill climbing algorithm guided the selection and tuning of five integrated models (random forest, SVM, logistic regression, CART, and LightGBM), achieving 81.8% AUC and 89.7% accuracy on the test set, a modest but statistically supported gain over the individual classifiers. Performance remained stable in temporal validation on participants recruited in 2025, with an AUC of 0.812.