검색 팝업 닫기
Conference Abstract Biological Age

Comparison of deep-learning based retinal imaging biomarkers for their association with incident cardiovascular diseases

ARVO Annual Meeting Abstract
저자

Faye Ng Yu Ci; Cancan Xue; Yih Chung Tham; Crystal Chong; Marco Chak Yan Yu; Simon Nusinovici; Tyler Rim; Carol Yim-lui Cheung; Tien Yin Wong; Ching-Yu Cheng

Abstract

Purpose 

Retinal images provide critical information about vasculature changes and cardiovascular health, offering a non-invasive imaging approach for predicting major cardiovascular disease (CVD). In this study, we compare three deep-learning (DL) based biomarkers: RetiAGE, RetiPhenoAge, and retinal vessel calibers measured using a DL approach to assess their association with the risk of incident CVD events.

Methods 

We performed prospective cohort analyses using retinal images of 33,057 subjects from the UK Biobank dataset to derive RetiAGE, RetiPhenoAge, and retinal vessel calibers such as central retinal arteriolar and venular equivalents (CRAE and CRVE) and arterio-venous ratio (AVR). Incident CVD events were identified using ICD codes as per the American Heart Association’s definition. We examined the association of each biomarker with incident CVD events using the Cox proportional-hazards regression model, adjusting for age, sex, body mass index, current smoking status, alcohol intake, hypertension, diabetes, dyslipidemia and anti-cholesterol medications. The continuous net reclassification index (NRI) of each biomarker against the pooled cohort equation (PCE) score was calculated.

Results :

 Among the 33,057 subjects (age:56.5±8.2 years, female:55.2%), 1,274 (3.85%) had incident CVD events during the 12 years of follow-up. After adjusting for age, sex and other covariates, RetiAGE (per SD increase: HR=1.09; 95%CI=1.02-1.15) and RetiPhenoAge (HR=1.32; 95%CI=1.20-1.46) were significantly associated with increased risk of incident CVD events, with RetiPhenoAge having the larger effect size. Significant classification improvement was observed after adding RetiAGE (NRI=33.9%; 95%CI=27.1-40.0; P<0.001) and RetiPhenoAge (NRI=30.7%; 95%CI=24.8-37.0; P<0.001) to PCE. Narrower arteries (HR=0.81; 95%CI=0.76-0.87), wider venules (HR=1.15; 95%CI=1.07-1.23), and smaller AVR (HR=0.86; 95%CI=0.81-0.90) were associated with a higher risk of CVD in the unadjusted model.

Conclusions :

 RetiPhenoAge demonstrated a stronger association with future CVD risk compared to RetiAGE and retinal vessel measurements, highlighting its potential to effectively risk stratify patients for future CVD events against existing risk scores. This provides a convenient and non-invasive approach to risk profiling patients for CVD treatments and interventions.

This abstract was presented at the 2025 ARVO Annual Meeting, held in Salt Lake City, Utah, May 4-8, 2025.