The relationship between retinal vascular parameters measured by artificial intelligence and the risk of new-onset cardiovascular disease in a Chinese community-based population

Background Studies on the association between retinal vascular diameter and tortuosity and the risk of new-onset cardiovascular disease (CVD) are controversial. The predictive value of retinal vascular parameters, measured through artificial intelligence (AI), for incident CVD risk in the community population, remains to be explored. Objectives Our study aims to investigate the association between retinal vascular parameters (diameter and tortuosity) measured through AI, and the risk of new-onset CVD in the Chinese community-based population. Methods We conducted a prospective cohort study on subjects from a Beijing community in China who were free of prevalent stroke or myocardial infarction and underwent retinal photography at baseline. By deep-learning algorithm, quantitative retinal vascular parameters were measured 0.5 to 1 disc diameter from the optic disc margin, including central retinal arteriolar equivalent (CRAE), central retinal venular equivalent (CRVE), arteriolar-to-venular ratio (AVR), curvature tortuosity arteriole (cTORTa), and curvature tortuosity venule (cTORTv). The primary endpoint was a new-onset major adverse cardiovascular event (MACE), defined as a composite of cardiovascular death, first non-fatal stroke, and first non-fatal acute myocardial infarction (AMI). The secondary endpoints included cardiovascular death, first stroke, and first AMI. Cox proportional-hazards regression analyses were performed. Results A total of 3,585 participants were included. The mean age was 56.49 ± 8.94 years, and 2,335 (65.1%) were female. Prevalence of prior hypertension and diabetes was 1594 (44.5%) and 676 (18.9%), respectively. During the follow-up (median, 7.53 years), there were 293 cases of new-onset MACE (8.2%), 44 cardiovascular deaths (1.2%), 226 strokes (6.3%), and 57 AMIs (1.6%). The Kaplan–Meier curves showed significant differences in cumulative hazards of new-onset MACE among tertiles of CRAE, CRVE, AVR, and cTORTa, but not for cTORTv (Figure 1). In Cox proportional-hazards models (Figure 2) adjusting for covariates, CRAE was negatively associated with the risk of new-onset MACE (per 10 μm increase, hazard ratio [HR]=0.95, 95% confidence interval [CI]: 0.92-0.99, P=0.009) and the risk of new-onset stroke (per 10 μm increase, HR=0.96, 95%CI: 0.92-1.00, P=0.04). Similarly, AVR was negatively associated with new-onset MACE risk (per 0.1 increase, HR=0.82, 95%CI: 0.72-0.94, P<0.001) and new-onset stroke risk (per 0.1 increase, HR=0.84, 95%CI: 0.73-0.98, P=0.022). No significant associations were observed between CAVE, cTORTa, and cTORTv with any of the primary or secondary endpoints. Conclusions CRAE and AVR, measured through AI, are associated with the risk of new-onset MACE in the Chinese population. These findings suggest the potential for AI-measured parameters in retinal vasculature to aid in stratifying CVD risk in the general population.