Purchase this article with an account.
Karen Chen, Eliot R Dow, Nergis C Khan, Marcie Levine, Chandrashan Perera, Anuradha Phadke, Jimmy Dang, Kirsti Weng, Diana V Do, Vinit B Mahajan, Prithvi Mruthyunjaya, Kapil Mishra, Theodore Leng, David Myung; Integration of Artificial Intelligence into a Telemedicine-Based Diabetic Retinopathy Screening Program. Invest. Ophthalmol. Vis. Sci. 2022;63(7):1422 – A0118.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Annual diabetic retinopathy (DR) screenings are critical for preventing blindness in patients with diabetes. Our goal was to successfully integrate artificial intelligence-based image interpretation software into a telemedicine-based diabetic retinopathy screening program at primary care clinics.
The Byers Eye Institute at Stanford (BEIS) partnered with University HealthCare Alliance (UHA) to integrate IDx-DR, an FDA-cleared AI diagnostic system that autonomously detects diabetic retinopathy (DR) in fundus images, into a pre-existing teleophthalmology workflow. Patients without a prior DR diagnosis or a DR exam in the past year were offered the opportunity to have retinal photographs taken at the end of their primary care visit. The AI-human hybrid workflow involves interpretation by the IDx system. Images deemed ungradable by the AI software were then sent for interpretation by a retina specialist at the BEIS reading center. Patients were referred for in-person exam if either the AI or the human reader detected more than mild DR (mtmDR) in the images.
From April 2021 to December 2021, a total of 550 patients with diabetes at four primary care sites opted for DR screening using the AI-human hybrid workflow. Of these, 72 patients screened positive for mild or worse DR (13%) and 425 patients screened negative (77%). Average gradeability ranged by site from between 80% and 90%.At each site, the percentage of the diabetic patient population up-to-date with recommended eye exams was measured in accordance with the HEDIS measure specification for Comprehensive Diabetes Care (CDC). A target of 67.89% was chosen, reflecting the 90th percentile HEDIS national benchmark. The percentage of patients adherent with annual diabetic eye exams reached a peak of 71.1% across the four sites after integration of AI into the telemedicine workflow. These outcomes demonstrate that integration of the AI-human hybrid workflow resulted in increased patient adherence with annual diabetic eye exams.
An AI-human hybrid workflow for detecting referral-warranted DR was successfully implemented in the primary care setting and resulted in improved patient adherence and quality measures associated with annual diabetic eye exams.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.
This PDF is available to Subscribers Only