Vessel Density and Caliber Discriminate between Normal Retina and Clinical Stages of Diabetic Retinopathy

Abstract

Purpose : To assess the utility of SD-OCTA derived metrics to determine diabetic retinopathy (DR) severity.

Methods : We carried out a multicenter, retrospective, cross-sectional study using OCTA (AngioPlex^TM Carl Zeiss Meditec, Inc). A total of 538 eyes in 349 subjects [102 non-diabetic, 103 mild non-proliferative DR (NPDR), 41 moderate NPDR, 32 severe NPDR, and 71 proliferative DR] were classified according to DR status. Vessel skeletal density (VSD) and vessel diameter index (VDI) were calculated on the superficial retinal layer (SRL). The subjects were divided into initial (227 patients) and validation cohorts (122 patients) for modeling. Multinomial logistic regression models were generated, and the performance of each model was quantified by calculating its area under the curve (AUC).

Results : There is a significant increase in the odds of DR for every 0.001 unit decrease in VSD. Specifically, the odds of having mild, moderate, severe NPDR, and PDR compared to normal subjects increased by 12% (95% CI: 8-19%), 25% (95% CI: 18-33%), 27% (95% CI: 19-35%), and 32% (95% CI: 25-41%) [all p<0.001] respectively, after adjusting for comorbidities from past medical history. In addition, there is a significant increase in the odds of DR for every 0.01 unit increase in VDI. The odds of having mild, moderate, severe NPDR, and PDR compared to normal subjects increased by 5% (95% CI: 1-10%), 17% (95% CI: 11-23%), 16% (95% CI: 10-23%), and 21% (95% CI: 15-28%), respectively [p=0.02 for mild, p<0.001 for moderate, severe, and PDR]. The AUC graphs in both cohorts showed excellent discrimination of normal subjects vs. any DR for the covariate model (AUC =.89 and AUC =.80 for the initial and validation cohort respectively) and furthermore the models containing VSD and VSD with VDI significantly contributed to the discrimination as seen in Figure 1. For mild vs. more severe DR and for mild/moderate/severe vs PDR, we also confirmed excellent discrimination, but the VSD or VDI did not significantly increase the AUC compared to the covariate only model in the validation cohort.

Conclusions : VSD and VDI are significantly correlated with the presence or absence of DR and DR severity. These values may be used with other clinically relevant variables to determine DR severity.

This is a 2020 ARVO Annual Meeting abstract.

 

Figure 1: AUC of all diabetic retinopathy vs. normal in the initial and validation cohort models. Model 1 refers to a covariate only model.