Waist Circumference Modifies the Association Between a Deep Learning-Derived Retinal Biomarker and Coronary Artery...

Background: The deep learning-derived retinal cardiovascular risk index (Reti-CVD) is a deep learning-derived retinal biomarker calculated from non-mydriatic fundus photographs for cardiovascular risk assessment. This study examined whether obesity phenotype, particularly central adiposity, modifies the association between Reti-CVD and coronary artery calcium score (CACS) in asymptomatic adults undergoing routine health screening. Methods: We retrospectively analyzed 237 Korean adults who underwent fundus photography for Reti-CVD assessment and cardiac computed tomography for CACS measurement. Abdominal obesity was defined as waist circumference (WC) ≥ 90 cm in men and ≥85 cm in women, and general obesity as body mass index (BMI) ≥ 25 kg/m². Multivariable linear regression models with sequential adjustment were used to evaluate the association between Reti-CVD and CACS. Effect modification was assessed using interaction terms for Reti-CVD×WC and Reti-CVD×BMI. Discriminatory performance for coronary calcification, defined as CACS > 0, was evaluated using the area under the receiver operating characteristic curve (AUC). Results: Abdominal obesity was present in 78 participants (32.9%), and general obesity in 102 (43.0%). Participants with CACS > 0 had significantly higher Reti-CVD scores than those with CACS = 0 (0.15 ± 0.09 vs. 0.09 ± 0.05; p < 0.001). Reti-CVD remained positively associated with CACS after adjustment for metabolic and lifestyle factors. In fully adjusted models, WC significantly moderated this association (interaction p = 0.0288), whereas BMI did not (interaction p = 0.5381). Overall discrimination for CACS > 0 was moderate (AUC = 0.735) and numerically higher in participants with abdominal obesity than in those with normal WC (0.787 vs. 0.695). Conclusions: Reti-CVD is independently associated with coronary calcification, and WC-based central adiposity modifies this relationship. Incorporating obesity phenotype may improve personalized interpretation of retinal biomarker-based cardiovascular risk assessment.