Presentation (CTI)

OBJECTIVES: Diabetic retinopathy (DR) is a major microvascular complication of type 2 diabetes (T2D) and a leading cause of vision loss, and an indicator of increased cardiovascular risk. Early identification of patients at risk may support timely screening and intervention, but the performance of existing DR risk prediction equations in Taiwanese populations remains uncertain. We externally validated three established DR risk equations among Taiwanese adults with T2D
METHODS: Using electronic health records from National Cheng Kung University Hospital, we identified adults with T2D and no prior DR between 2019 and 2021. Baseline demographics, comorbidities, medication use, and laboratory measurements were ascertained within 1 year of the first T2D diagnosis (index date). Three equations, RECODe, CHIME, and the equation developed by Lian et al., were evaluated for predicting incident DR at 2 and 4 years. Incident DR was identified from outpatient, inpatient, and emergency department diagnoses. Performance was assessed using the area under the receiver operating characteristic curve (AUROC) for discrimination, and calibration slopes and intercepts for calibration.
RESULTS: Among 20,551 adults with T2D, the mean age was 65 years, the mean diabetes duration was 2.3 years, and the DR incidence rate was 0.93 per 100 person-years. For 4-year prediction, CHIME demonstrated better discrimination (AUROC, 0.69) than RECODe (0.60) and the Lian equation (0.57), with consistent findings at 2 years. CHIME also had calibration parameters closest to ideal among the three equations, with slopes/intercepts of 0.648/0.006 at 2 years and 0.576/0.006 at 4 years. However, CHIME modestly underestimated observed DR risk.
CONCLUSIONS: CHIME showed the best overall performance among the three equations in Taiwanese adults with T2D. However, its modest discrimination and residual miscalibration indicate that local recalibration or model updating is needed before routine clinical implementation.