Key result
Hill climbing machine learning ensemble outperforms single classifiers, predicting myocardial injury with ~90% accuracy.
Why the study?
Myocardial injury is an early manifestation of myocarditis, and routine physical examination data offer a valuable resource to develop a machine learning model for early risk assessment.
Does a hill climbing-based machine learning ensemble model using routine physical examination data improve the risk stratification of biochemical myocardial injury (hs-cTnT > 14.1 pg/mL) compared to individual machine learning classifiers?
Observational (n=1,827)
No
Does a hill climbing-based machine learning ensemble model using routine physical examination data improve the risk stratification of biochemical myocardial injury (hs-cTnT > 14.1 pg/mL) compared to individual machine learning classifiers?
Absolute Event Rate: 81.8% vs 81.6%
p-value: p=0.041
A hill climbing-based machine learning ensemble model utilizing eight routine physical examination biomarkers accurately stratifies biochemical myocardial injury risk (hs-cTnT > 14.1 pg/mL), offering a practical screening tool to guide targeted cardiovascular evaluations.
May aid myocardial injury risk stratification with routine data; extends ensemble ML methods but leaves open prospective outcome trials.
Inflammatory diseases such as myocarditis, which can be caused by a variety of factors, severely impact cardiac function and may lead to heart failure and death. Myocardial injury is widely recognized as an early and common pathological manifestation in the progression of myocarditis. With the widespread adoption of health check-ups, physical examination data has become a valuable resource for the early detection of diseases. This study aims to develop a machine learning-based model for assessing the risk of myocardial injury using routine physical examination data. Participants recruited in 2024 and 2025 were used for model development and temporal validation, respectively. Variable and sample selection were performed based on missing values, followed by multiple imputation via chained equations. A hill climbing algorithm was utilized to optimize model selection and parameter tuning, resulting in the integration of five ML models: Random Forest (RF), Support Vector Machine (SVM), Logistic Regression (LR), Classification and Regression Trees (CART) Classifier, and Light Gradient Boosting Machine (LGBM) Classifier. Model performance was evaluated using accuracy, precision, recall, F1-score, AUC, calibration plot and Brier score. Feature importance was assessed using the SHapley Additive exPlanation (SHAP) method. Performance evaluation results demonstrated that the ensemble model showed modest but statistically supported improvement over all individual classifiers in risk prediction of myocardial injury, as verified by paired DeLong test for AUC and McNemar’s test for classification discordance. The ensemble model achieved 81.8% AUC (0.2-2.0 increase), 89.7% accuracy (0.2%-2.7% improvement), 70.1% recall (2.9%-15.2% increase), and 75.8% F1-score (1.7%-8.2% higher) on the test set, while it retained stable performance on the temporal validation cohort with an AUC of 0.812 (95% CI: 0.768–0.856) and accuracy of 88.8% (95% CI: 0.847–0.929). Ablation experiments confirmed that each base classifier contributed predictive gain to the integrated framework, and our hill climbing-based weight adjustment strategy yielded highest overall performance compared with fixed equal-weight fusion. The ensemble model proposed in this study provides support for the risk stratification of biochemical myocardial injury. High-risk individuals identified by this model allows targeted further myocarditis-related examinations, facilitating early screening and timely early intervention. The final prediction model was implemented into a web application to enhance its practicality in clinical settings.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) conducted an observational in Biochemical myocardial injury (n=1,827). Hill climbing-optimized machine learning ensemble model (integrating RF, SVM, LR, CART, and LGBM) vs. Individual baseline machine learning classifiers (CART, LGBM, RF, LR, SVM) was evaluated on Area under the receiver operating characteristic curve (AUC) for predicting biochemical myocardial injury (hs-cTnT > 14.1 pg/mL) on the internal test set (95% CI 0.783-0.853, p=0.041). The hill climbing machine learning ensemble model achieved an AUC of 0.818 (95% CI 0.783-0.853) and 89.7% accuracy for predicting myocardial injury, outperforming single classifiers (p<=0.041).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: