Deep learning-based cardiovascular risk stratification for stage 1 hypertension using retinal photographs

Abstract

Purpose

The advent of deep learning (DL) algorithms has now made it possible to predict the risk of cardiovascular disease (CVD) using retinal images. We had previously developed a retina-based DL model, Reti-CVD, which successfully predicted future CVD incidents in a longitudinal study. According to US guidelines, adults with stage 1 hypertension (HTN) who have an estimated 10-year atherosclerotic cardiovascular disease (ASCVD) risk of 10% or higher should be managed initially with blood pressure (BP) lowering medication. This study aims to investigate to use Reti-CVD score for decisions on BP-lowering medication among adults with stage 1 HTN.

Methods 

Our study included 7664 adults with stage 1 HTN from the UK Biobank. Reti-CVD scores were calculated and categorized into two risk groups – low (n=6433), and high (n=1231), based on the same number of participants with 0 to 10% and ≥10% 10-year ASCVD risk according to Pooled Cohort Equation (PCE). To assess the Reti-CVD’s ability to predict fatal and non-fatal CVD events, we performed a survival analysis on the longitudinal data using Cox proportional-hazards models and hazard ratios (HRs).

Results 

Among the 7664 participants with stage 1 HTN, 125 (1.6%) had nonfatal and fatal ASCVD events during the 10-year follow-up. Based on the standard of care (PCE), those with ≥10% ASCVD risk group (4.5 CVD incidence rate per 1000 person-years[IR]) showed significantly increased risk of ASCVD events (HR=3.87, 95%CI:2.71-5.53) compared with <10% ASCVD risk group (1.2-CVD IR). In comparison to the low-risk group (1.2-CVD IR), the high-risk group (4.2-CVD IR) of Reti-CVD was significantly associated with increased risk of ASCVD events (HR=3.48, 95%CI: 2.43-4.99) in the multivariable Cox model. The two-tier Reti-CVD (low- vs high-risk) showed prognostic performance with a concordance index of 0.620 (95%CI:0.577-0.663).

Conclusions 

With DL, retinal photographs could be used as a surrogate marker for cardiovascular risk stratification for stage 1 HTN given its compatibility with the estimated 10-year ASCVD risk predictions from blood tests. Our findings suggest that contactless and non-invasive retinal photography may be adopted as a supportive and adjunctive modality to aid decision-making in starting BP-lowering medication for adults with stage 1 HTN, especially, patients who fail to meet the <130/80 mmHg treatment goal even after nonpharmacologic therapy.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.