Retinal age gap in unilateral retinal diseases with unaffected fellow eye: phakia and vascular pathology impact retinal age prediction.

PURPOSE: To investigate retinal age gap (RAG), defined as the difference between the deep learning (DL)-predicted retinal age and chronological age, in five treatment-naïve unilateral retinal diseases and identify associated factors.

STUDY

DESIGN

Retrospective observational study.

METHODS

This study included patients with unilateral retinal diseases who had unaffected fellow eyes. Individuals with bilaterally healthy eyes served as controls. Retinal age was estimated using the publicly available DL model. RAG was calculated by subtracting chronological age from predicted retinal age. We compared RAG among affected, fellow, and normal eyes. Multivariable linear mixed models identified covariates associated with RAG across the whole cohort of eyes.

RESULTS

Overall, 873 patients and 222 controls were analyzed. Prediction accuracy for retinal age was highest in normal (r = 0.90), followed by fellow (r = 0.70) and affected eyes (r = 0.54). Phakia (standardized β = -0.34), greater myopia (β = -0.08), and right-eye laterality (β = -0.05) were independently associated with higher RAG. Phakic eyes showed higher RAG than pseudophakic (0.8 ± 7.8 vs. -9.1 ± 10.6 years, p < 0.001). After multivariable adjustment of three covariates, higher RAG in affected eyes versus fellow eyes was observed only in retinal vein occlusion (RVO) (mean difference = 1.42 years, p = 0.037), particularly in central RVO (p = 0.001).

CONCLUSIONS

Phakia had the strongest influence on RAG, warranting lens status considerations when developing or interpreting DL models. RAG may reflect retinal vascular pathology.