AI-assisted diabetic retinopathy screening in the primary care setting compared to the ophthalmology clinic setting

Abstract

Purpose : At Temple we are doing AI-assisted diabetic retinopathy (DR) photo screenings in our primary care setting (PCS) and ophthalmology clinic setting (OCS). The PCS has medical assistants with limited training as ophthalmic photographers who in general take relatively few diabetic screening photographs each month. The OCS has a trained ophthalmic photographer performing the diabetic screenings who has taken thousands of photographs. In this study we aim to compare DR screening outcomes in the PCS versus the OCS.

Methods : A retrospective chart review of DR screenings at nine primary care offices and two ophthalmology clinics with Eyenuk’s (Los Angeles, CA) EyeArt between 5/12/2021-12/28/2022 was conducted. Photo gradeability, the number of patients with a referable diagnosis of more than mild (MTM) DR (International Clinical Diabetic Retinopathy (ICDR) 2-4 +/- evidence of macular edema), the rate of referable patients scheduled for a follow-up appointment, the average time to contact referable patients after screening, and the rate of completed appointments from each clinic setting were compared by chi-square analysis.

Results : At the PCS, the ungradable photo rate was 42.5% (333/784) and 14.5% (82/564) at the OCS (p<0.0001). The number of patients with MTM DR was 105/784 (13%) at the PCS and 137/564 (24%) at the OCS (p<0.0001). Of the 105 patients with MTM DR at the PCS, 61 (58%) were scheduled for follow-up appointments, compared to 110/137 (80%) at the OCS (p=0.0002). Of those patients, 43/61 (70%) and 76/110 (69%) showed to their appointments at the PCS and OCS, respectively (p=0.8). The average time to contact the patient was 28.6 days at the PCS and 13.2 days at the OCS.

Conclusions : The significantly higher ungradable percentage at the PCS compared to the OCS suggests the PCS photographers need more training. The greater percentage of patients scheduled for follow-up appointments and the significantly shorter time to contact the patient at the OCS suggests the PCS needs improved coordination, potentially with a patient care navigator. Interestingly, the number of patients with MTM DR was almost twice as much at the OCS compared to the PCS. Further evaluation is needed to explain the difference between the groups.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.