1116-OR: An Interpretable Machine-Learning Model to Differentiate Diabetic and Nondiabetic Kidney Disease in Type 2 Diabetes

Introduction and Objective: Diabetic kidney disease (DKD) is a major complication of type 2 diabetes (T2D), yet non-diabetic kidney disease (NDKD) is common. Although renal biopsy remains the diagnostic gold standard, its invasive nature limits routine use. We aimed to develop an interpretable machine learning (ML) model to noninvasively differentiate DKD from NDKD. Methods: From 19,404 renal biopsy cases across 14 centers in China, 2,047 patients with T2D and definitive pathological diagnoses were included. Six ML algorithms were trained using routinely available clinical variables optimized by recursive feature elimination. Model performance was assessed by area under the receiver operating characteristic curve (AUC-ROC), and interpretability was evaluated using SHapley Additive exPlanations (SHAP). The best-performing model was externally validated and deployed as a web-based tool. Results: Of the 2,047 biopsy-confirmed patients with T2D, 722 (35.3%) had DKD, 999 (48.8%) had NDKD, and 326 (15.9%) had mixed pathology. The Random Forest model achieved the highest discriminative performance (AUC 0.90 [95% CI 0.88-0.92], sensitivity 0.77, specificity 0.90). Performance remained robust in external validation (AUC 0.86). Diabetic retinopathy, diabetes duration, HbA1c, and serum creatinine were the strongest predictors of DKD. SHAP analysis identified diabetic retinopathy as the most influential positive contributor, whereas higher total cholesterol and BMI were inversely associated. The final model was deployed as an interactive web application (https://diagnosis-of-dkd-and-ndkd.streamlit.app/). Conclusion: This interpretable ML model accurately distinguishes DKD from NDKD using routine clinical data, potentially reducing reliance on renal biopsy and supporting personalized clinical decision-making in T2D. Disclosure J. Yang: None. F. Yang: None. J. Zhang: None. M. Li: None. X. Li: None. Funding National Natural Science Foundation of China (82570960)