Early prediction of acute kidney injury in traumatic and non‐traumatic rhabdomyolysis using an interpretable machine learning model: A multicenter study with external validation

Acute kidney injury (AKI) is a common and severe complication of rhabdomyolysis (RM), and early risk stratification remains challenging because of its multifactorial and heterogeneous nature. We developed and externally validated an interpretable machine learning (ML) model for early prediction of AKI in RM across traumatic and non‐traumatic etiologies. Data were obtained from four public critical care databases and a multicenter cohort from tertiary hospitals in China. A total of 1569 patients were included in the derivation cohort and 401 in the external validation cohort. Eighteen variables within 24 h of admission were used to train 12 ML models. Performance was assessed by area under the receiver operating characteristic curve (AUC), and interpretability was evaluated using SHapley Additive exPlanations. The random forest model achieved the best performance (AUC = 0.940) and was simplified into a five‐variable model including lactate dehydrogenase, serum creatinine, Alb, prothrombin time, and activated partial thromboplastin time. The final model achieved AUCs of 0.919 and 0.900 in internal and external validation, respectively, with consistent performance in non‐traumatic (0.911) and traumatic (0.882) subgroups. This interpretable model may support early risk stratification in patients with RM across different etiologies.