Abstract
Purpose :
We aim to determine the effects of implementing an artificial intelligence (AI) diabetic retinopathy (DR) screening system on the rate of patient follow-up and time to follow-up in patients of the North Philadelphia community with more-than-mild (MTM) DR.
Methods :
A retrospective chart review of patients screened for DR with Eyenuk's (Los Angeles, CA) EyeArt software at three primary care offices between 10/1/2020–12/31/21. Inclusion criteria were diabetics 18 or older who had a DR screening. Photos were taken on a non-mydriatic fundus camera. EyeArt interpretation was based on the International Classification of Diabetic Retinopathy (ICDR) criteria. Results were reported as “referable” for uninterpretable images or MTM DR (ICDR 2-4 +/- evidence of macular edema) or “not referable” (ICDR 0-1 and no evidence of macular edema). Temple ophthalmology office staff reviewed the EyeArt database to schedule follow up appointments for referrable patients. Screening outcomes and time to follow up exams were recorded and compared to a control group that were interpreted by an optometrist without AI between 3/2018 and 3/2020.
Results :
41 patients were screened for DR at 3 clinic sites after implementing AI. 24/41 (58.5%) were referable by EyeArt. 17/41 (41.4%) patients were not referable. Of the referable patients, 3/24 (12.5%) were contacted by office staff, and all 3 completed a follow up exam. The average time to a follow-up exam was 126 +/- 25.7 days after the screening exam. Of 1902 screening exams prior to implementing EyeArt, 688 (35.1%) were referable. 157/688 (22.8%) had an appointment scheduled, and 133/157 (85.7%) showed up to their appointment. The average time from screening to interpretation was 30.5 +/- 43.2 days, and the average time from interpretation to appointment was 158.6 +/- 123.3 days. There was no significant difference between protocols for time to appointment (p=0.40).
Conclusions :
We found no difference in time to follow up exam after AI implementation, due to delays at the appointment scheduling step. AI interprets the images immediately, but the follow-up exam is not scheduled until the ophthalmology office staff are notified. To effectively utilize AI, involving PCP office staff in scheduling follow up exams immediately after EyeArt interpretation is crucial.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.