AHA Scientific Session Abstract, November 2023
Our previous work led to developing a deep learning algorithm for retinal images, Reti-CVD, which effectively predicted cardiovascular disease (CVD) events in individuals without CVD history, leveraging coronary artery calcium (CAC) scores for algorithm training.
Hypothesis
This study aims to assess the capability of deep learning-assisted retinal imaging to predict CVD events among prediabetic and diabetic patients using the data from the UK Biobank.
Methods
Our study included prediabetic and diabetic patients from the UK Biobank. Reti-CVD scores were calculated and categorized into three risk groups – low (n=550), moderate (n=276), and high (n=275), based on the 50th and 75th percentiles, following a 2:1:1 ratio. To assess the Reti-CVD’s ability in predicting fatal and non-fatal CVD events, we performed a survival analysis on the longitudinal data from the UK Biobank using Cox proportional-hazards models and hazard ratios (HRs).
Results
Among the 1101 prediabetic or diabetic patients at the onset, 138 (12.5%) experienced CVD events. According to Reti-CVD scores, these events were found as 8.2% (45/550), 15.2% (42/276), and 18.5% (51/275) in the low, moderate, and high-risk groups over a median follow-up period of 11 years, respectively. After adjusting for factors such as age, gender, hypertensive medication use, statin use, and smoking history, a significant association was observed between the Reti-CVD and the incidence of CVD events (HR=1.57, 95% CI, 1.00-2.47 for the moderate-risk group; HR=1.88, 95% CI, 1.19-2.98 for the high-risk group compared to the low-risk group). An increasing HR trend of 1.36 (95% CI, 1.09-1.70) was observed across risk groups in the prediction of CVD events.
Conclusions
The Reti-CVD offers a valuable tool for risk stratification among prediabetic and diabetic patients, indicating its potential in managing these high-risk groups.
