Abstract
Purpose :
Retinal vascular morphological features (RVMFs) derived from retinal fundus images serve as sensitive biomarkers for systemic vascular diseases. Existing literatures focused on only a few pre-selected RVMFs but overlooked the inter-correlation among all available RVMFs, which leads to inconsistent findings. Simultaneously analyzing all RVMFs can generate comprehensive digital profile for overall retinal vascular pattern (RVP) to provide reliable insights into the association with vascular outcomes.
Methods :
The study cohort consists all participants from GoDARTS (Genetics of Diabetes Audit and Research in Tayside, Scotland) who had their retinal images analyzed by VAMPIRE (Vascular Assessment and Measurement Platform for Images of the Retina). RVMF data were linked to electronic health records. Confirmatory factor analysis generated 22 latent factors from 149 RVMFs. K-means clustering was used to identify distinct RVP clusters based on the 22 latent factors. Cox proportional hazards model was used to compare ten-year incidences of vascular outcomes across RVP clusters, including a composite cardiovascular (CV) endpoint (fatal and non-fatal), hospitalization for heart failure (HHF), de novo chronic kidney disease (CKD), and all-cause death.
Results :
4103 participants with type 2 diabetes (age: 67 ± 11 years, 55% male) were included. K-means clustering identified three distinct RVP clusters (Figure 1). Individuals in cluster 2 characterized with greater venous tortuosity, while those in cluster 3 had wider vessels but greater arterial tortuosity. Comparing to cluster 1, those in cluster 2 had higher incidences of CV events (adjusted hazard ratio [aHR] = 1.27; 95% CI: 1.09 – 1.47; p = 0.002), de novo CKD (1.32 [1.07 – 1.63]; p = 0.011), and all-cause death (1.17 [1.03 – 1.34]; p = 0.02). Higher incidence of HHF was observed in cluster 3 (1.27 [1.03 – 1.57]; p = 0.028).
Conclusions :
The findings revealed heterogeneous RVPs in people with type 2 diabetes, and their associations to different vascular outcomes. This paves the way for further personalized risk assessment and targeted interventions. Further validation is needed to consolidate the findings.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.