Retinomics as a Tool for Glaucoma Prediction

To investigate retinomic changes preceding glaucoma onset and explore their predictive value.

A population-based, prospective cohort study.

A total of 40 949 adults from the UK Biobank, all with eligible color fundus photography (CFP) data and OCT data and without baseline glaucoma, were included in this study.

We used baseline values of retinomics, a composite set of quantitative retinal imaging biomarkers including 135 retinal vascular measurements extracted with the Retina-based Microvascular Health Assessment System from CFP and 21 OCT-derived retinal layer measurements. After least absolute shrinkage and selection operator feature selection, Cox regression was used to assess associations with incident glaucoma, and a gradient boosting machine model was applied to evaluate predictive performance.

Glaucoma status.

During a median follow-up of 12.49 years, 653 of 40 949 participants developed glaucoma. After adjusting for age, sex, ethnicity, education, smoking behavior, alcohol consumption, physical activity, hypertension, obesity, glycated hemoglobin, and intraocular pressure, 18 of the 48 least absolute shrinkage and selection operator-identified retinal parameters showed statistically significant associations with incident glaucoma, with each standard deviation change associated with 8.2% to 26.4% increased risk. These findings highlighted novel predictors beyond conventional parameters, including vascular network simplification and inner nuclear layer-related thickening. For a 12.49-year incident glaucoma prediction, simply using age, sex, and retinomic features, we achieved a concordance index of 0.767. After being stratified into 3 risk groups, the highest risk group showed a hazard ratio of 8.72 (95% confidence interval: 6.59-11.54) against the lowest risk group.

Our study revealed retinal vascular and neural alterations associated with increased risk of incident glaucoma. In addition, our study showed that retinomics can serve as an effective biomarker for identifying individuals at high risk of developing glaucoma. The simplicity (age, sex, and basic imaging) of our model along with its satisfactory risk stratification performance for long-term incident glaucoma suggest that it can be used to distinguish those patients who are most suitable for early therapeutic intervention to prevent blindness or severe visual impairment at a population level.

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.