Analysis of Ungradable Images in a Diabetic Retinopathy Telemedicine Screening Study

Abstract

Purpose :
Telemedicine diabetic retinopathy (DR) screening has become increasingly valuable in underserved communities. Poor photo quality significantly hampers screening efforts and can occur due to technical error or pathologies such as cataract, small pupils, or even vitreous hemorrhage. We performed a retrospective study to determine the final diagnoses at an in-person dilated fundus exam of patients who had ungradable telemedicine screening exams. We then compared them to patients with gradable screening exams.

Methods : From March 2018 to March 2020, we evaluated 1902 adult patients with diabetes who presented to one of several primary clinics in North Philadelphia. Non-dilated fundus photos were taken by a trained technician and read remotely by an optometrist. Patients who screened positive for DR, or had ungradable images due to poor view of the fundus were referred to an ophthalmologist for a dilated fundus exam. Initial screening results were compared to final diagnoses, and statistical analyses were performed using Fisher's exact test.

Results :
133 patients attended a follow-up appointment, of which 48 (36%) were referred for DR and 85 (64%) for ungradable images (Table 1). 33/48 (69%) patients who had DR on screening had some degree of DR on the final exam, although only 17/48 (35%) received the same grading on both exams. In comparison, only 16/85 (19%) patients who had ungradable photos had DR on their final exam (p<0.001).
Of those who had ungradable images who did have DR on final diagnosis, there was a significant difference in distribution of disease severity compared to those who had gradable images (Table 2; p=0.039). In those with ungradable images, there was a higher percentage of patients that had proliferative DR (44% vs. 15%) and a lower percentage that had mild NPDR (19% vs. 52%). However, the result was only marginally significant due to small sample size when the 1 unspecified DR patient was removed (p=0.057).

Conclusions : Telemedicine is less effective in identifying patients who have DR when the image is ungradable. However, these patients may present with more severe pathology that worsen image quality, highlighting the importance of follow-up in these patients.

This is a 2021 ARVO Annual Meeting abstract.