AHA Scientific Sessions 2026 Abstract · Nov 2, 2026
AI-Derived Retinal Cardiovascular Risk Scores Predict Stroke and TIA in a Multiethnic Cohort: A Longitudinal Oculomics Study
By Injety R, Yoon HJ, Eveson D et al.
Introduction/Background: Stroke prevention depends on the timely identification of individuals at elevated vascular risk. Retinal vasculature provides a non-invasive window into systemic microvascular and macrovascular health. Artificial intelligence-based estimation of cardiovascular (CVD) risk from retinal images could serve as a proxy for predicting stroke risk. Research Questions/Hypothesis: AI-derived retinal CVD risk scores will significantly predict incident stroke and TIA and improve risk stratification beyond traditional clinical factors, enabling earlier and more precise identification of high-risk individuals. Methods/Approach: We analysed longitudinal colour fundus images from participants with stroke, TIA, or controls using Mediwhale’s Dr.Noon CVD risk score. For participants with repeated retinal images, CVD scores were modelled as time-updated covariates in time-varying Cox proportional hazards models. Four models were fitted: Stroke+Control, Stroke+TIA+Control, TIA+Control, and Stroke+TIA. Kaplan-Meier curves were generated using participants' first images and stratified by CVD quintiles; differences between survival distributions were assessed using multivariate log-rank testing. Results/Data: The analysis included 6,535 longitudinal images from 1,802 participants from the University Hospitals of Leicester, NHS Trust, Stroke and TIA clinic, with 645 stroke and 508 TIA events. Median age was 70 years, and 60.3% were male; with multiethnic representation included 66.1% White and 29.5% Asian participants; over 75% had high CVD scores (>40.74). CVD risk scores increased progressively prior to stroke diagnosis. Stroke risk was independently associated with CVD score (HR = 1.36, 95% CI: 1.25-1.48, p < 0.005), male sex (HR = 1.45, 95% CI: 1.21-1.73, p < 0.005), and age (HR = 1.02 per year, 95% CI: 1.02-1.03, p < 0.005). Similar associations were observed across combined outcome models. TIA risk was independently associated with CVD score (HR = 1.24, 95% CI: 1.13-1.36, p < 0.005) and age (HR = 1.04, 95% CI: 1.03-1.05, p < 0.005). Kaplan-Meier analysis demonstrated significant separations across CVD score quintiles (overall log-rank statistic 23.97, p < 0.005), with participants in the lowest quintile showing approximately twice the stroke-free survival probability compared with those in the highest quintile. Conclusion(s): AI-derived retinal CVD risk scores independently predict stroke and TIA risk and stratify stroke-free survival across quintiles. These findings support retinal imaging as a non-invasive, scalable tool for early risk identification and targeted prediction, with potential for integration into future clinical and population health pathways.