AI-driven eye screening aims to close diabetes vision gap in community clinics

By integrating autonomous AI screening directly into safety-net primary care, researchers hope to speed the diagnosis of diabetic retinopathy, enhance referral follow-up, and help millions at risk of vision loss receive early treatment.

Trial: Diabetic retinopathy screening among patients with federally certified health centers using point-of-care artificial intelligence. Image credit: Anukool Manoton / Shutterstock

In a recent trial protocol published in JAMA Network Opena team of researchers will test whether the point of care is autonomous All included screening integrated into a federally certified health center (FQHC) workflows improve diabetic retinopathy (Dr) completion of screening, speeds diagnosis and enhances referral follow-up and patient experience.

Background

One in four people with diabetes in US shows early signs Drhowever symptoms are often invisible until vision is threatened. Annual eye exams prevent preventable blindness, but many patients miss referrals due to time, travel and cost barriers. FQHCthey serve 32.5 million people who face these barriers. Autonomic All included that reads bottom photographs during a routine visit can return a result on the spot, turning a missed referral into same-day action. However, fairness, workflow adaptation, and patient trust remain open questions, which urgently require further research.

About the study

This randomized, open-label, patient-level, parallel-group study enrolled adults with diabetes who had not completed Dr screening within 11 months in both FQHC clinics at San Ysidro Health in San Diego County, California. Recruitment began in June 2024 and is expected to end in August 2025, with follow-up through February 2026. Participants provide consent, complete baseline surveys, and are randomized to point-of-care screening with an autonomous AI diabetic retinopathy system (AI-DRS) or to usual care referral. In the intervention arm, non-mydriatic fundus photographs are taken on a Topcon TRC-NW400 camera by trained personnel after a 2-day skills training program. Images are analyzed by an algorithm cleaned by FDA (EyeArt; Eyenuk Inc). The algorithm classifies more than mild diabetic retinopathy (mtmDR) and sight-threatening diabetic retinopathy (VTDR) three failed quality attempts yield an ungradable result.

The results are entered into the electronic health record (EHR) through HL7 integration, enabling referrals based on risk stratification for primary care professionals (PCP) review, immediate retinal referral for positives and urgent ophthalmology referral for ungraded images. Both study arms receive standard referrals and participant navigation support for appointment booking. Outcomes at 90 and 180 days include completion of screening (primary), stage at first detection, completion of referral, and patient-reported knowledge, attitudes, self-efficacy, and satisfaction. The protocol follows Standard Protocol Details: Recommendations for Interventional Trials (SPIRIT) and guided by the Pragmatic Robust Implementation and Sustainability Model (PRISM).

Study results

This protocol predetermines the results and analytical design. The primary endpoint is its completion Dr screening within 90 days, a patient-centered measure linked to early diagnosis and treatment. Secondary endpoints include; Dr stage at first detection, completion of risk-stratified referrals, and changes in point-of-care knowledge, attitudes, and self-efficacy All included. EHR The integration allows tracking of orders, incidence rates, and International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) codes after eye care visits. Implementation metrics such as successful image acquisition, nongradable rates after three attempts, and camera time per patient quantify feasibility in busy clinics. Patient-reported measures adapted from validated instruments assess confidence and comfort with autonomy All included as regular care.

The evidence that motivates the intervention comes from previous evaluations of it FDA-cleaned up AI-DRS. Diabetic Retinopathy Versus Early Treatment Study (ETDRS) reference standard, the sensitivity for mtmDR was 96% (specificity, 88%), and the sensitivity for VTDR was 97% (specificity, 90%), with definitive results for more than 97% of eyes, usually without dilatation. In more than 100,000 actual encounters, sensitivity and specificity were approximately 91% for mtmDR detection. These performance characteristics suggest that field screening can accurately assess patients and reduce missed referrals.

Workflow integration is central to impact. Orders come from EHR and route to AI-DRS servant; the client guides acquisition with real-time quality feedback. After analysis, the results are returned to EHR to urge PCP risk-based review and referrals. Patients positive for VTDR or mtmDR get immediate retinal appointment (target ≤ 24 hours). Patients with ungraded images, often indicative of underlying pathology, receive urgent ophthalmology appointments (target ≤ 72 hours). Patients negative for both mtmDR and VTDR are scheduled for a follow-up in 12 months. This closed-loop design is intended to shrink diagnostic delays and increase adherence.

To support the report, Dr findings are mapped ICD-10 codes, covering both nonproliferative and proliferative stages, with and without macular edema. This allows for consistent documentation, auditing and population-health dashboards within FQHC systems. Analyzes will use regression models with covariate adjustment for clinic location, demographics and clinical risk, with differences in differences for survey changes. The application will be evaluated through PRISM areas (approach, adoption, implementation and maintenance) to understand what it takes to sustain All included-Enabled check in safety net settings. If the AI-DRS meets or exceeds screening benchmarks, the model could help clinics meet performance targets such as the 63% “high performance level” as defined in the protocol PRISM results.

Finally, the protocol includes security supervision and monitoring. Although no formal safety endpoints are planned, adverse events will be recorded and reviewed by a Data Safety Monitoring Committee. Algorithms remain static during testing to avoid performance drift, and staff training emphasizes proficiency checks and troubleshooting for reliable image acquisition. Generalizability may be limited by reliance on Epic EHR integration, equipment and licensing costs and trained personnel requirements.

conclusions

This pragmatic protocol evaluates whether the embedding is autonomous AI-DRS within FQHC primary care, with support FDA customs clearance, ETDRS-referential accuracy and HL7– enabled EHR interoperability, can lift Dr completing screening, speeding up diagnosis and enhancing referral compliance. Returning results in minutes and matching findings to ICD-10 for consistent reporting, the model aims to reduce avoidable vision loss while simplifying workflows for PCPsmall. Guided by SPIRIT for transparent reporting and PRISM for real-world implementation, the approach offers a practical roadmap, including education, closed-loop referrals, and quality monitoring, that safety net clinics can adapt to improve population eye health at scale.

Trial registration: NCT06721351. Two co-authors are employees of Eyenuk Inc. and cite related patents. other authors report no relevant disclosures.