Abstract
BACKGROUND: To develop a retinal age prediction model based on a foundation model using fundus images and to determine the association between gamma-glutamyl transferase (GGT) levels and the retinal age gap. METHODS: A total of 36,044 fundus images with reasonable quality from 9,752 participants in the Jidong Eye Cohort Study were included in this study. Of these images, 8,869 fundus images from 3,010 healthy individuals were used to train and validate the model based on the foundation model RETFound for age prediction using 10-fold cross-validation. A total of 4,081 fundus images from 4,081 participants who were enrolled from May to October 2023 had available GGT data, and these images were used to investigate the association between GGT levels and the retinal age gap. RESULTS: The trained model in this study achieved excellent performance, with a mean absolute error (MAE) of 2.42 ± 0.08 years. The mean age of the participants in the analysis dataset was 43.7 ± 10.4 years, and 1987 (48.7%) participants were women. The multivariable βs and 95% confidence intervals (CIs) of the retinal age gap in the second, third, and fourth GGT quartiles compared with the lowest GGT quartiles were 0.42 (0.08-0.77), 0.54 (0.15-0.92), and 0.72 (0.29-1.14) (P for trend = 0.001), respectively, in the fully adjusted model (adjusted for age, sex, current smoking status, current drinking status, body mass index, hypertension, diabetes, dyslipidemia, and serum uric acid). CONCLUSION: Increased GGT levels were significantly associated with accelerated retinal aging as quantified by the retinal age gap. Our findings indicate that elevated GGT levels may have an adverse effect on the aging process.