Interpretable gradient boosting machine model for predicting in-hospital mortality in sepsis-induced myocardial injury: a multicenter development, validation, and web-based clinical implementation.

Background: Sepsis-induced myocardial injury (SIMI) is a life-threatening complication of sepsis, associated with high in-hospital mortality. Current risk prediction models for SIMI lack interpretability and multi-center validation, necessitating advanced analytical approaches to improve risk stratification and clinical decision-making. Method: In this study, LASSO regression for feature selection, eight machine learning algorithms were evaluated, including Gradient Boosting Machine (GBM), XGBoost, and Logistic Regression. Model performance was assessed via AUC, sensitivity, specificity, and calibration curves. SHapley Additive exPlanations (SHAP) were used to interpret feature contributions, and recursive feature elimination optimized the model for clinical usability. Result: The GBM model demonstrated optimal performance with an internal validation AUC of 0.751 (95% CI: 0.614-0.867), outperforming other algorithms (e.g., SVM: 0.733, LR: 0.730). External validation in eICU and the Chinese cohort yielded AUCs of 0.924 and 0.703, respectively, confirming generalizability. Key predictors identified by SHAP included APS III Score, Hypertension, Albumin, Diabetes, SOFA Score, ALT, RBC, and Lactate. A simplified model with five variables (age, SOFA score, APS III score, albumin, ALT) achieved an AUC of 0.789 and was deployed on a user-friendly platform (https://anyuanning.shinyapps.io/SIMI/), enabling real-time risk assessment for clinicians. Conclusion: This multicenter study developed an interpretable predictive model to predict in-hospital mortality in patients with sepsis-induced myocardial injury (SIMI) and deployed it on a user-friendly platform, potentially improving patient outcomes through early risk assessment.