Prediction of retinopathy risk: A prospective cohort study in China

Aim: To identify risk factors for retinopathy and to develop a nomogram for individualised risk prediction in a multi-ethnic Chinese cohort. Methods: Data were derived from the SENSIBLE-Cohort, excluding participants with retinopathy at baseline. Two nomograms were constructed: one using baseline data only (Baseline), and one incorporating baseline and follow-up data (Combination). Predictor selection involved Cox regression, Boruta, least absolute shrinkage and selection operator (LASSO), and recursive feature elimination (RFE). Model performance was evaluated using Harrell's C-index, confusion matrix, and Brier Score. The receiver operating characteristic (ROC) curves, the area under the ROC curve (AUC), the DeLong test, and the decision curve analysis (DCA) were used for comparative assessment. Results: A total of 2,447 participants were included (mean age: 53.0 ± 8.6 years; 66.1 % female; BMI: 25.4 ± 3.5 kg/m²), including 1,380 with normal glucose tolerance, 762 with prediabetes, and 305 with diabetes. During follow-up, 144 (5.9 %) people developed retinopathy. Key predictors included BMI, waist-to-hip ratio, triglycerides, systolic and diastolic blood pressure, hypertension history, and ethnicity. The Combination nomogram showed superior discrimination compared to the Baseline nomogram (AUC: 0.75 vs. 0.64, P < 0.001) and demonstrated balanced sensitivity and specificity. DCA demonstrated greater clinical utility of the Combination nomogram across a range of risk thresholds. Conclusion: The Combination nomogram enables early retinopathy risk stratification using accessible clinical data. It may support personalised screening and introduces the broader concept of metabolic retinopathy.