AI screening tool increases eye exam referrals for African American patients

Investigators at the Wilmer Eye Institute, Johns Hopkins Medicine, led by principal investigator T.Y. Alvin Liu, M.D., published a retrospective analysis of 3,745 adult diabetes patients seen at Johns Hopkins primary care sites between August 2020 and September 2022 in npj Digital Medicine on April 13, 2026. Of these, 3,352 patients (mean age 60.6) received referrals from primary care providers and 393 patients (mean age 61.6) received a recommendation from an FDA-approved AI-assisted screening program. The study specifically examined two historically underserved groups: African American patients and patients covered by Medicaid.

Patients screened with the AI tool had retinal images captured with a specialized camera and analyzed in real time during the same primary care visit; if diabetic retinopathy was detected, patients were informed and given a same-day referral to the Wilmer Eye Institute or another eye care specialist of their choice. That immediate, on-the-spot result is the key mechanical difference from a standard PCP referral, which asks a patient to schedule a separate exam based on risk rather than a confirmed finding.

Comparing referral methods, researchers found a higher share of African American patients got an eye exam referral via the AI tool than via standard PCP referral (64.9% vs. 44.4%); patients with hypertension (89.6% vs. 82.6%) and chronic kidney disease (26.2% vs. 20.9%) showed a similar gap. Medicaid-coverage referral rates, by contrast, were comparable regardless of method (0.8% vs. 0.6%), meaning the AI tool narrowed the racial referral gap without similarly narrowing the insurance-based one. Patients who both used the AI tool and attended their eye exam were 15% more likely to be African American. The study builds on earlier work by Liu's team showing population-level referral increases from the same AI tool.

The result suggests point-of-care, same-day AI screening can close specific access gaps that delayed, judgment-based referral pathways do not, but the effect is not uniform across every disparity dimension. Teams deploying similar real-time screening tools in underserved settings should measure disparity-specific outcomes directly (as this study did for race, insurance status, hypertension, and kidney disease) rather than assuming a single equity benefit generalizes across subgroups.

Liu says further work is needed to determine whether increased referral and exam attendance translates into better long-term vision outcomes, and whether the effect replicates at other health systems beyond Johns Hopkins' community-based primary care sites.