iDR-DM-SCORE: Machine Learning Prognostic Model for Diabetic Retinopathy Risk Prediction among Patients with Type 2 Diabetes Mellitus

About the Project

This retrospective cohort study developed and validated machine-learning survival models to predict five-year diabetic retinopathy risk among patients with type 2 diabetes in Johor, Malaysia. Registry data included 47,953 development, 18,832 internal-validation, and 20,220 temporal-validation patients. Using 16 routine predictors, gradient-boosted survival performed best, achieving a C-index of 0.640 and AUC of 0.659. Discrimination declined during internal and temporal validation. Nephropathy, diabetes duration, and age were the strongest predictors. The model showed modest, reproducible risk ranking and could be applied in clinical settings in the future following further validation and recalibration.