Deep learning–derived retinal age gap and its associations with lifestyle, systemic, and ocular health in a health screening cohort

Individuals of the same chronological age differ in biological aging, and scalable, noninvasive markers are needed. The deep learning–derived retinal age gap (RAG) is a promising measure of retinal aging, but its associations with real-world health determinants remain unclear. We developed a multi-task model to predict retinal age using 29,530 fundus images from 7535 participants in a health screening cohort and evaluated RAG in cohort A for lifestyle, socioeconomic, and systemic factors ( n = 5606) and cohort B for ocular diseases ( n = 1810). The bias-corrected multi-task model trained on mixed data achieved the best performance, with a mean absolute error of 2.656 years and a Pearson correlation of 0.921 in cohort A and 2.529 years and 0.938 in cohort B. Higher RAG was significantly associated with smoking (ex-smokers, β = +0.46 years; current smokers, β = +0.50 years) and with clinical diabetes (+2.52 years); both survived false discovery rate (FDR) and Bonferroni correction, and the diabetes association persisted across all sequential covariate-adjustment sets. Married participants had lower RAG ( β = −0.46 years), significant after FDR correction only. Hypertension and hyperlipidemia were not associated with RAG. In cohort B, RAG was significantly higher in eyes with age-related macular degeneration ( β = +0.60 years) and cataract ( β = +1.86 years) than in normal controls, both surviving corrections. RAG, an imaging-derived age-prediction residual, is therefore associated with lifestyle, systemic, and ocular health. Whether it reflects biological aging requires longitudinal validation against established aging biomarkers; at present, RAG suits population-level characterization better than individual-level risk stratification.