Abstract / Summary
Differentiating diabetic kidney disease (DKD) from non-diabetic kidney disease (NDKD) is challenging. Kidney biopsy remains the gold standard. The primary objective was to develop a clinically useful risk prediction model to distinguish DKD from NDKD. We retrospectively analyzed 80 adults with diabetes who underwent kidney biopsy. Multivariable logistic regression was performed in 72 patients with complete data to identify predictors of diabetic kidney disease (DKD). A point-based risk score was derived from regression coefficients and evaluated using receiver operating characteristic (ROC) analysis. Of the 72 patients, 34 (47.2%) had DKD and 38 (52.8%) had non-diabetic kidney disease (NDKD). DKD was associated with longer diabetes duration (168 vs. 60 months, p < 0.001), higher insulin use (74.4% vs. 26.8%, p < 0.001), and more frequent retinopathy (59.0% vs. 24.4%, p = 0.0036). Diabetes duration ≥ 10 years and serum creatinine > 120 µmol/L were the strongest independent predictors of DKD. A simplified six-variable risk score (0–15 points) demonstrated excellent discrimination (AUC 0.913). A cutoff > 4 points provided 90.5% sensitivity and 50.0% specificity, whereas ≥ 10 points yielded 76.2% sensitivity and 100.0% specificity. DKD probability increased progressively across low-, intermediate-, and high-risk groups, with all high-risk patients having biopsy-proven DKD. Not applicable.