Toward an explainable AI-Based clinical decision support system for predicting adverse outcomes in Rhabdomyolysis

Rhabdomyolysis is a severe condition with high morbidity and mortality, driven by complications like acute kidney injury. Early risk stratification remains challenging as traditional scores fail to capture complex data patterns. This study lays the foundation for an explainable AI (XAI)-based clinical decision support system (CDSS) by developing a machine learning model to predict the composite outcome of renal replacement therapy or 90-day mortality. Using routinely available admission data from 1031 adults, we applied multivariate imputation, Boruta feature selection, and ADASYN for class imbalance. The CatBoost model achieved the highest discrimination (AUC = 0.942, 95% CI: 0.904-0.980), accuracy (0.913), and maintained good calibration (Brier score = 0.080). Shapley additive explanations (SHAP) identified creatinine, troponin T, and albumin as key predictors, validating clinical plausibility and enabling instance-level explanations for CDSS deployment. Decision-curve analysis confirmed superior net benefit against treat-all or treat-none strategies across clinically relevant thresholds. We propose a framework for integrating this interpretable model into electronic health records to provide real-time risk scores at the point of care. Calibration drift in older adults highlights the need for age-specific refinement, underscoring the value of transparent, evaluable AI in clinical informatics.