Cardiovascular Risk Stratification Using Artificial Intelligence-Derived Retinal Imaging and SCORE2 in Untreated Dyslipidemia: A UK Biobank Prospective Cohort Study

We evaluated the prognostic performance of Dr. Noon CVD, an artificial intelligence (AI)-derived retinal imaging model, in 40,727 UK Biobank participants with untreated dyslipidemia. We assessed 5- and 10-year incident cardiovascular events and the model's incremental value beyond SCORE2. After adjustment for demographic, clinical, and metabolic risk factors, higher Dr. Noon CVD scores were independently associated with increased CVD risk; the hazard ratio was 1.51 (95% confidence interval 1.16–1.95) for the high-risk group and 1.75 (1.28–2.40) for an exploratory very-high-risk group. Adding the AI model to SCORE2 significantly improved discrimination (C-index improvement 0.025) and reclassification (net reclassification improvement 0.262; both p  < 0.001). Risk stratification remained effective even within the SCORE2 high-risk subgroup. These findings demonstrate that AI-derived retinal imaging independently predicts CVD outcomes and enhances standard risk assessment, offering a non-invasive strategy to guide treatment in individuals with dyslipidemia.