Multicenter validation of a severity index model for predicting postoperative acute kidney injury

BACKGROUND: Existing models for predicting postoperative acute kidney injury (AKI) after non-cardiac surgery are often complex and insufficiently validated for broad clinical use. We developed and externally validated a simple yet accurate model for predicting both overall and critical AKI that can be readily applied in routine practice. METHODS: The severity index model for AKI (SIM-AKI) was developed using data from 191,938 patients undergoing non-cardiac surgery at a tertiary hospital and externally validated using three independent datasets from other tertiary hospitals (n = 118,047; 86,092; 3727). Variables were selected using least absolute shrinkage and selection operator regression with 10-fold cross-validation, and predictor stability was assessed using backward elimination across 100 bootstrap resamples before multinomial logistic regression modeling. Model performance was evaluated using the C-statistic for discrimination, calibration plots, Brier scores, and decision curve analysis (DCA) for clinical utility. RESULTS: The SIM-AKI model incorporated age, sex, diabetes mellitus, American Society of Anesthesiologists classification, cancer surgery, emergency status, major abdominal surgery, anemia, hypoalbuminemia, estimated glomerular filtration rate, intraoperative transfusion, and operation time. For overall AKI, C-statistics were 0.801 (95% CI 0.796-0.806) in development and 0.754, 0.742, and 0.759 in validation cohorts. For critical AKI, C-statistics were 0.838 (95% CI 0.826-0.850) in development and 0.796, 0.805, and 0.767 in validation cohorts, demonstrating good calibration and clinical benefit in DCA. The SIM-AKI compared favorably with existing AKI prediction models in discrimination. CONCLUSION: SIM-AKI may serve as a reliable perioperative tool for predicting the risk of both overall and critical postoperative AKI in patients undergoing non-cardiac surgery.