Leveraging Artificial Intelligence to Prevent Vision Loss from Diabetes Among Socioeconomically Disadvantaged Communities

PROJECT SUMMARY Diabetes mellitus affects 37 million US adults and is the leading cause of vision loss among adults aged 18-64-years. Countries such as the UK that have robust eye screening and treatment programs successfully prevent blindness from diabetes. In the US, where screening programs have been less successful, usual-care screening

involves a primary care provider referring patients with diabetes to an eye care provider for a dilated eye exam. Barriers to usual-care screening include transportation, cost, and the time required for patients to make and attend this separate eye care appointment. Racial/ethnic minorities and socioeconomically disadvantaged

communities, such as individuals on Medicaid, are more affected by these challenges, resulting in lower screening rates and higher rates of preventable vision loss. Thus, an urgent need exists for a program that equitably improves eye screening and follow-up eye care rates for patients with diabetes. The overall objective

of this proposal is to investigate an FDA-approved artificial intelligence (AI)-based system that allows primary care providers to identify diabetic eye disease at the primary care clinic without the need for oversight by an eye care provider. The novel intervention we are testing, AI-BRIDGE (Artificial Intelligence-Based point of caRe,

Incorporating Diagnosis, SchedulinG, and Education), is an autonomous AI-based protocol that provides screening for diabetic eye disease at primary care visits, as well as culturally adapted patient education on diabetic eye disease and, if a patient screens positive, assistance with scheduling an in-person, follow-up eye

care visit. To achieve our objective, we will carry out 2 specific aims: (1) Determine whether, relative to usual- care screening, AI-BRIDGE improves eye screening and follow-up care rates across races/ethnicities and reduces racial/ethnic disparities in screening rates. To do so, we will work with stakeholders to adapt AI-BRIDGE

to an underserved primary care setting and then conduct a stepped-wedge, cluster randomized controlled trial of the adapted intervention in partnership with 10 clinics that are Federally Qualified Healthcare Centers, providing primary care to medically underserved communities. (2) Identify determinants of, and strategies to

promote, effective and equitable implementation of AI-BRIDGE using a mixed methods approach. We hypothesize that factors such as organizational leadership’s commitment to the intervention, competing demands on the clinic, and patient and provider perceptions of AI will contribute to adoption of AI-BRIDGE. To test this

hypothesis, we will conduct semi-structured interviews of patients, clinic leadership, and providers to identify barriers and facilitators, and then work with stakeholders to identify strategies to address the barriers identified. This work is innovative because it is the first-ever randomized controlled trial that (1) evaluates whether AI can

improve equity in eye screening and follow-up care and (2) identifies barriers, facilitators, and strategies to successfully implement this screening strategy in an underserved setting. It is significant because a scalable, AI- based eye screening and follow-up program could reduce disparities in vision loss and ensure that, in the future,

diabetic eye disease is no longer the leading cause of blindness in the US.