gms | German Medical Science

37. Internationaler Kongress der Deutschen Ophthalmochirurgie (DOC)

15.05. - 17.05.2025, Nürnberg

Eyeing the heart: Early cardiovascular risk prediction with deep learning-based retinal insights

Meeting Abstract

Search Medline for

  • Yarn Kit Chan - Singapore General Hospital, Singapore, Singapur
  • Qingsheng Peng - Singapore National Eye Centre, Singapore, Singapur
  • Ching-Yu Cheng - Yong Loo Lin School of Medicine, Singapore, Singapur

37. Internationaler Kongress der Deutschen Ophthalmochirurgie (DOC). Nürnberg, 15.-17.05.2025. Düsseldorf: German Medical Science GMS Publishing House; 2025. DocFP 1.3

doi: 10.3205/25doc003, urn:nbn:de:0183-25doc0030

Published: May 13, 2025

© 2025 Chan et al.
This is an Open Access article distributed under the terms of the Creative Commons Attribution 4.0 License. See license information at http://creativecommons.org/licenses/by/4.0/.


Outline

Text

Purpose: Early detection and precise risk stratification of cardiovascular diseases (CVD) are critical in reducing morbidity, mortality, and healthcare costs. We utilized deep learning algorithms (DLAs) to identify biomarkers for early-stage CVD detection through retinal imaging.

Methods: We utilized RetiAGE, an advanced DLA that estimated biological age from retinal photographs to predict CVD risk. We utilized data from 57,297 patients from the UK Biobank and excluded images with poor quality and participants with CVD at baseline. RetiAGE scores were generated for the remaining samples. Statistical analysis included multivariable Cox proportional hazards (CoxPH) regression, adjusted for age and sex, to evaluate the utility of RetiAGE scores in predicting CVD outcomes. Subgroup analyses were performed to risk stratify CVD subtypes using RetiAGE quartiles.

Results: RetiAGE scores correlated with CVD events (z score = 3.11, p = 0.002). Our novel findings are that RetiAGE predicts 37 of 176 CVD subtypes. RetiAGE scores were highly correlated with risk for nonrheumatic aortic (valve) stenosis (HR 11.86, p = 0.01), cerebral infarction due to embolism of cerebral arteries (HR 6.86, p = 0.02), unstable angina (HR 5.73, p = 0.002), and essential (primary) hypertension (HR 3.01, p < 0.001). HR ranged from 2 to 12 in all subtypes with statistical significance.

We performed subgroup analyses within each CVD subtype after stratifying the entire cohort by quartiles based on RetiAge score, with higher quartiles associated with greater risk. Quartiles 2, 3, and 4 were compared to quartile 1 as baseline. In essential (primary) hypertension, the 2nd (HR 1.80, p < 0.0001), 3rd (HR 2.37, p < 0.0001), and 4th (HR 3.44, p < 0.0001) quartiles were significantly correlated with CVD. This trend of hazard ratios of CVD events increasing from quartiles 2 to 4 was also present in most other subtypes, showing that RetiAge can capture subclinical changes in the retina in early-stage CVD.

Conclusion: RetiAGE primarily predicts CVD subtypes with hemodynamic instability, and demonstrated good predictive accuracy across diverse CVD subtypes, paving the way for early risk stratification.