The Pulse
AI Eye Exams Narrow Youth Diabetes Screening Gap in Trial
A prospective study of 397 young people with diabetes found that point-of-care AI eye exams brought screening completion to 99.5% among White participants and 98% among non-White participants. The Nature Communications study also reports th
AI.info Team · Written with AI · fact-checked ·
Autonomous AI eye exams completed during diabetes care visits brought retinal screening rates for White and non-White youth to nearly the same level, according to a prospective study published October 23 in Nature Communications. Among 397 participants, 99.5% of the White cohort and 98% of the non-White cohort completed the AI exam after the intervention. Before it, the White group was more likely to have had a diabetic eye exam previously: 79.8%, compared with 61.8% of non-White participants.
“Minority and underserved youth are less likely to undergo diabetic retinopathy screening, yet more likely to have diabetic retinopathy,” the authors write. The ACCESS2 study tested whether offering autonomous screening during diabetes visits could narrow that difference.
Screening completion rose in both groups
The study enrolled White and non-White youth with type 1 diabetes, who made up 72% of participants, and type 2 diabetes, who made up 28%. Participants received autonomous AI diabetic eye exams and optical coherence tomography imaging during diabetes care visits. The study’s primary outcome was whether screening completion became equivalent between the two racial groups.
The reported completion rates—99.5% for White youth and 98% for non-White youth—were close after the AI intervention, in contrast with the gap in prior exam history. The authors describe the result as equivalent screening completion after implementation. It shows that the exam was completed by nearly all participants in both groups in this study; it does not establish that screening rates will be the same in other clinics or populations.
Accuracy estimates carry a wide confidence interval
The researchers also compared the AI results with standard grading for more than mild diabetic retinopathy. They report 100% sensitivity, with a 95% confidence interval from 67.6% to 100%, and 76% specificity, with a confidence interval from 71.4% to 79.9%. The sensitivity result means the system detected all cases identified by the comparison standard in the study sample, but the broad interval leaves considerable uncertainty about how well that figure would hold in a larger group.
Specificity of 76% indicates that the system did not correctly identify every participant without more than mild disease. The paper’s diagnostic results therefore warrant a more measured reading than the 100% sensitivity figure alone might suggest. Screening completion was the main result; the study did not report that the intervention improved vision or reduced later eye disease.
Digital Diagnostics supplied the AI readings
The paper says Digital Diagnostics provided the autonomous AI interpretations in kind. Coauthor Michael D. Abramoff disclosed that he is an investor, director and consultant to the company and holds relevant patents assigned to Digital Diagnostics and the University of Iowa. The study was supported by the National Eye Institute and Breakthrough T1D.
That disclosure provides important context for interpreting the diagnostic findings, alongside the study’s limited sensitivity estimate. The result most clearly supported by the reported data is narrower: among the 397 youth enrolled, completion of an AI eye exam during diabetes care visits reached near-parity between White and non-White participants. The trial is registered as NCT05463289.
“Minority and underserved youth are less likely to undergo diabetic retinopathy screening, yet more likely to have diabetic retinopathy.”
Risa M. Wolf et al., Nature Communications