Comparative evaluation of international and population-specific risk prediction models for type 2 diabetes: evidence from the kharameh cohort of the PERSIAN study

Various models have been developed for predicting the risk of type 2 diabetes. This study was designed to compare the performance of international and population-specific models in an Iranian population. This cross-sectional analysis examined data from 10,663 participants aged 40-70 years from the Kharameh cohort. The performance of five type 2 diabetes risk prediction models-FINDRISC, ADA, AUSDRISK, the American risk score, and a new logistic regression model was compared using statistical metrics (sensitivity, specificity, AUC) and Decision Curve Analysis to assess clinical utility. As glucose metabolism status worsened from normal to diabetes, mean age, body mass index, waist circumference, and blood pressure increased significantly. The American model showed the highest discriminative ability (AUC = 0.79) and provided the greatest standardized net benefit. The FINDRISC model showed acceptable performance (AUC = 0.72). The ADA and AUSDRISK models performed weaker (AUC = 0.64 and 0.68, respectively). The new logistic model, with an AUC of 0.71, demonstrated high sensitivity (0.82) but low specificity (0.52), making it primarily suitable for high-sensitivity initial screening. The American risk score was the most effective tool, while the lab-free FINDRISC model served as a practical alternative for primary care screening at lower risk thresholds.