Retinal biomarkers for systemic diseases: an oculome-wide association study in 164,784 individuals

Abstract

Purpose : The advancement of high-resolution imaging techniques in ophthalmology, combined with the growing ability of deep-learning algorithms, has enabled discoveries of many retinal vascular and neurosensory biomarkers. This study aims to systematically examine the relationship between these biomarkers and systemic diseases to guide further research in the field of oculomics, the study of eye-body relationships.

Methods : This cross-sectional analysis involved 61,718 individuals from the UK Biobank cohort and 103,066 individuals from the AlzEye cohort after excluding data on image quality. 43 systemic diseases were curated based on the positive predictive values of their ICD10 codes. Quantitative biomarkers were extracted from color fundus photography (CFP) and optical coherence tomography (OCT) by externally validated machine learning algorithms, giving the thicknesses of nine retinal layers in three rings, 24 features of the retinal vasculature, and 6 features of the optic nerve. Logistic regression was used to test the associations between the systemic diseases and retinal biomarkers while controlling for age, age-squared, sex, and ethnicity. Bonferroni correction set a p-value cut-off of 1.66E-05 for statistical significance.

Results : 588 significant associations were found between systemic diseases and retinal biomarkers. On the feature side (number of associations), retinal fractal dimension (29), vessel density (29) and venular width (24) were associated with the highest number of systemic diseases. In OCT, the associations were most prominent in the thickness of the entire inner retina (25), outer photoreceptor layer (24), and inner ganglion cell-inner plexiform layer (24). On the disease side, renal (34), cardiovascular (20.8), and endocrine (17) pathologies had the highest average number of associations. The top diseases were diabetic mellitus (46), hypertension (44), heart failure (36), acute kidney injury (34), and obesity (30). Hierarchical clustering differentiated closely related pathologies to facilitate disease subtyping.

Conclusions : Our findings support the value of oculomics research by demonstrating the retinal biomarkers’ multisystemic utility. High-potential eye-body associations can be investigated by future studies to fully determine their diagnostic and prognostic value.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

 

Summary tile-plots for associations between retinal biomarkers and diseases