2198-P: Longitudinal Validation of a Retinal AI Score for Cardiovascular Risk Monitoring: Validation in Korean and UK Populations

Introduction and Objective: While retinal AI enables non-invasive cardiovascular disease (CVD) screening, its utility for monitoring longitudinal changes remains unverified. We validated a retinal AI score (Dr. Noon CVD) in a Korean population and verified progression patterns using the UK Biobank (UKBB). Methods: We analyzed data from the GreenCross (GC) checkup center (Korea, 2022-2024) and the UKBB. Participants free of baseline CVD with ≥2 fundus exams (≥30 days interval) were included. In GC, missing smoking data were imputed as non-smoker if absent across all visits, or the recorded status otherwise. We assessed AI score’s accuracy for high risk (PREVENT score ≥7.5%) and the correlation between changes in AI and PREVENT scores in GC. Annual progression rates were verified in both datasets. Results: 18,084 (GC) and 1,745 (UKBB) participants were included. Dr. Noon CVD achieved an AUROC of 0.950, and longitudinal score changes correlated strongly with PREVENT changes (r=0.67, p<0.001) in GC. Median annual progression in low-risk (<31) groups was 1.21 [IQR -0.49-3.81] (GC) and 2.03 [1.02-3.62] points/year (UKBB). Rapid progression cutoffs (top 5%) were 16.70 and 5.37 points/year, respectively. Conclusion: Dr. Noon CVD accurately identifies high-risk individuals and effectively tracks dynamic risk changes. The longitudinal progression patterns, verified across Korean and UK populations, support its utility for non-invasive risk monitoring. Disclosure D. Nam: Employee; Current; Mediwhale Inc. J. Seo: Employee; Current; Mediwhale Inc. S. Thakur: Employee; Current; Mediwhale Inc. S. Nusinovici: Consultant; Current; Mediwhale Inc. M. Jang: Employee; Current; Mediwhale Inc. T.H. Rim: Stock/Shareholder; Current; Mediwhale Inc.