Relationship between a Categorical Diabetic Retinopathy Severity Score and Quantitative Assessments of Diabetic Lesions

Abstract

Purpose : The qualitative, categorical Diabetic Retinopathy (DR) Severity Scale (DRSS) derived from the Early Treatment Diabetic Retinopathy Study (ETDRS) is commonly used to grade the severity of DR in clinical trials. In this study, we correlate this subjective DRSS score with precise quantifications of numbers and surface area of DR lesions on fundus images.

Methods : In this retrospective study, we collected ultra-widefield (UWF) pseudocolor images from adult diabetic patients. Eyes with evidence of retinal diseases aside from DR, history of ocular treatment for DR, or poor-quality images which precluded assessment of DR severity were excluded. A mask corresponding to the area circumscribed by the 7 standard ETDRS fields was applied to the UWF images, and the DRSS grade (no DR, mild, moderate (mod), or severe non-proliferative DR [NPDR], proliferative DR [PDR]) was assessed according to the ICDR protocol by two masked graders. Within the same ETDRS seven-fields, intra-retinal hemorrhages (H), microaneurysms (Ma), intra-retinal microvascular abnormalities (IRMA), and venous beadings (VB) were manually segmented. The numbers and surface area of these lesions were computed and correlated against the DRSS score using ANOVA with posthoc Bonferroni.

Results : Among 258 diabetic eyes (163 patients), 176 eyes (83 right eyes) of 113 patients (68 females) with a mean age of 59.18 ± 15.57 years were included. 4% were graded as no DR, 12.5% as mild NPDR, 69.3% as mod NPDR, 6.3% as severe NPDR, and 8% as PDR. There was strong intergrader agreement for the DRSS (Cohen’s Kappa = 0.95, p < 0.05). H/Ma area, H count, Ma count, and IRMA count were generally significantly different (p < 0.05) between DRSS levels (Figure 1). Specifically, these DR lesions' counts/area increased with DRSS level up to severe NPDR, but the lesions counts/area then decreased in PDR eyes. There was no significant difference in VB count between the DRSS levels (p = 0.42).

Conclusions : While DR lesions counts/area generally show a positive correlation with the DRSS level, lesion counts seem to decrease from severe NPDR to PDR. This seemingly paradoxical reduction in lesions may reflect the impact of progressive capillary nonperfusion.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.

 

Figure 1. Box and whisker plots of quantitative DR metrics for different DRSS levels. Red dot: mean; grey rectangle: confidence interval; IRMA: intraretinal microvascular abnormality.