Abstract
PURPOSE: To evaluate factors associated with retinal nerve fiber layer (RNFL) thickness in the Canadian Longitudinal Study on Aging (CLSA) using the Machine-to-Machine (M2M) deep learning model applied to fundus photographs. DESIGN: Cross-sectional study. SUBJECTS: Participants from the baseline Comprehensive Cohort of the CLSA. METHODS: This study included 28,114 CLSA participants aged 45 to 85 years with gradable baseline fundus photographs. The M2M model, trained on optical coherence tomography (OCT) data, was applied to estimate RNFL thickness from disc-centered fundus images. For participants with images from both eyes, the mean RNFL thickness of the 2 eyes was used. Associations between M2M-predicted RNFL thickness and age, sex, ethnicity/race, and self-reported glaucoma were analyzed using linear regression models adjusted for covariates. MAIN OUTCOME MEASURES: M2M-predicted RNFL thickness, age, age groups, sex, ethnicity/race, and self-reported glaucoma. RESULTS: The mean age of participants was 62.6 ± 10.1 years, and 51% were women. Self-reported glaucoma was present in 4.8% of the participants. The mean M2M-predicted RNFL thickness was 90.9 ± 9.2 µm. Age was inversely associated with RNFL thickness (Pearson's r = -0.16; p < .001), with each additional year associated with a 0.15 µm decrease (p < .001); after adjustment for covariates, the association remained significant (β = -0.11; p < .001). Participants with self-reported glaucoma exhibited significantly thinner RNFL (82.6 ± 12.9 µm) compared to those without (91.4 ± 8.7 µm; p < .001). RNFL thickness was slightly greater in women than in men (p < .001), and differences were observed across ethnicity/race groups (p < .001). CONCLUSIONS: The M2M model provided robust estimates of RNFL thickness from fundus photographs in a large population-based cohort. The observed associations between RNFL thickness, age, and glaucoma status were consistent with previous OCT-based findings, supporting the utility of the model for scalable structural assessments in epidemiological studies.