Diabetic Retinopathy Assessment Variability Among Eye Care Providers in an Urban Teleophthalmology Program

Background: Teleophthalmology is an evidence-based method for diabetic eye screening. It is unclear whether the type of eye care provider performing teleophthalmology interpretation produces significant variability. Introduction: We assessed grading variability between an optometrist, general ophthalmologist, and retinal specialist using images from an urban, diabetic retinopathy teleophthalmology program. Methods: Three readers evaluated digital retinal images in 100 cases (178 eyes from 90 patients with type 2 diabetes). Fisher's exact test, percent agreement, and the observed proportion of positive (P pos ) or negative agreement (P neg ) were used to assess variability. Results: Among cases deemed gradable by all three readers (n = 65), there was substantial agreement on absence of any retinopathy (88% ± 4.6%, P neg = 0.91–0.95), presence of moderate nonproliferative or worse retinopathy (87% ± 3.9%, P pos = 0.67–1.00), and presence of macular edema (99% ± 0.9%, P pos = 0.67–1.00). There was limited agreement regarding presence of referable nondiabetic eye pathology (61% ± 11%, P pos = 0.21–0.59) and early, nonroutine referral for a follow-up clinical eye exam (66% ± 8.1%, P pos = 0.19–0.54). Among all cases (n = 100), there was acceptable agreement regarding which had gradable images (77% ± 5.0%, P pos = 0.50–0.90). Discussion: Inclusion of multiple types of eye care providers as teleophthalmology readers is unlikely to produce significant variability in the assessment of diabetic retinopathy among high-quality images. Greater variability was found regarding image gradability, nondiabetic eye pathology, and recommended clinical referral times. Conclusions: Our results suggest that more extensive training and uniform referral standards are needed to improve consensus on image gradability, referable nondiabetic eye pathology, and recommended clinical referral times.