A simple and accessible diabetic retinopathy risk prediction model: Establishment and validation in a hospital-based cohort of type 2 diabetes patients

AIMS: Diabetic retinopathy (DR) is a leading cause of vision loss, with early detection challenging due to asymptomatic progression and limited predictive tools. To address this, we aimed to develop and validate a risk nomogram for DR prediction in type 2 diabetes patients. METHODS: In this retrospective cohort study of 70,073 patients with type 2 diabetes admitted from 2013 to 2019, 2,585 patients were included after exclusions. Patients were randomly assigned to derivation (2/3) and validation (1/3) sets. The prediction model was derived using Cox proportional hazards regression. A nomogram was developed and evaluated for discriminatory capacity and calibration accuracy. RESULTS: Among 2585 participants (mean age 59 years), 220 (8.5 %) developed retinopathy over a median follow-up of 34 months. We identified key predictors: glycated haemoglobin A1c, serum urea, and diabetes duration. Predictive models for 1-, 3-, and 5-year retinopathy-free survival were constructed and presented as a nomogram, demonstrating good discriminatory power (AUC: 0.941, 0.886, 0.594 in derivation; 0.747, 0.736, 0.670 in validation). Calibration plots further corroborated the improved fit for 3- and 5-year models. CONCLUSIONS: The proposed model shows promise for guiding early interventions and improving outcomes. Further external validation is needed to confirm its applicability across diverse populations.