Construction of a dynamic prediction model for diabetic retinopathy risk in adults with adult-onset type 1 diabetes: a real-world longitudinal cohort study.

Objective: To construct and validate a dynamic prediction model for diabetic retinopathy (DR) risk in adults with adult-onset type 1 diabetes mellitus (T1DM). Methods: A longitudinal cohort of 572 adult-onset T1DM patients from a tertiary hospital in Tianjin, China (2014-2023) was analyzed. LASSO regression and multivariable Cox regression with time-varying covariates (duration and HbA1c) were used to identify independent predictors of DR and to construct a nomogram. Model performance was assessed using the concordance index (C-index), area under the receiver operating characteristic curve (AUC), Brier score, calibration curves, Hosmer-Lemeshow test, and decision curve analysis (DCA). Internal validation was performed by bootstrap resampling (1000 replicates). Results: Multivariable Cox regression identified diabetic peripheral vascular disease, diabetic kidney disease, metabolic bone disease, dynamic disease duration, time-varying HbA1c, body mass index (BMI), and gender as independent predictors of DR. The dynamic model had a C-index of 0.704 (corrected: 0.681), with 1-, 2-, and 3-year AUCs of 0.704, 0.713, and 0.732, respectively. Brier scores were all <0.1. Calibration was acceptable at 2 and 3 years; the 1-year calibration slope was elevated (1.895). Hosmer-Lemeshow test P-values were >0.05 at all time points. The dynamic model outperformed a traditional static model in both discrimination and calibration. DCA showed favorable net clinical benefit for 2- and 3-year predictions. Conclusion: This study demonstrates the feasibility of dynamic DR risk assessment in adult-onset T1DM. The proposed nomogram may serve as an auxiliary tool for predicting 2- to 3-year DR risk, supporting nurses in risk stratification and individualized screening guidance.