Autonomous artificial intelligence (AI) systems for diagnosing diabetic eye disease (DED) have the potential to increase DED screening rates in underserved youth, while also saving costs for patients and caregivers, according to a study published in Nature Communications.1
The AI for Children’s diabetiC Eye ExamS (ACCESS) study also found sharing a diagnosis of referable disease at the point-of-care was associated with a higher rate of follow-through with eye care providers for management and treatment.
Early detection and treatment are key to preventing DED progression. The challenge is to ensure people with diabetes undergo regular screening.
ACCESS was a parallel randomised controlled trial run at two Johns Hopkins Paediatric Diabetes Centre sites in the United States. The trial randomised youth (ages eight to 21 years) with type 1 and type 2 diabetes to intervention (autonomous artificial intelligence diabetic eye exam at the point of care), or control (scripted eye care provider referral and education).
The AI systems used a camera operator to obtain point-of-care fundus images, which were interpreted by an AI algorithm to provide a diagnosis without human oversight. Review by a retina specialist found the AI images demonstrated an estimated sensitivity of 100% and specificity of 78.9.
FAVOURABLE RESULTS
The primary outcome for this trial was diabetic eye exam completion rate within six months.
In the intervention arm, 81/81 (100%) participants completed their diabetic eye exams, equating to a primary care gap closure rate of 100% (95%CI: 96%, 100%). All images were diagnostic for the AI system (output was either ‘DED present’ or ‘DED absent’). In the control arm, 18/82 completed the diabetic eye exam within six months, so the primary care gap closure rate was 22% (95%CI: 14%, 32%).
The difference of 78% (95% CI: 69%, 87%) in gap closure between control and intervention groups was described as ‘statistically significant’ (p < 0.001).
The secondary outcome was the proportion of participants who had a documented diabetic eye exam by an eye care professional (ECP) if deemed appropriate.
In the intervention arm, 25 participants received a ‘DED present’ output, and received the referral intervention. Of these, 16/25 attended an ECP visit within six months, for a follow-through completion rate of 64% (95%CI: 43%, 81%). In the control arm, 18 participants visited the ECP, for a follow-through completion rate of 22% (95%CI: 14%, 32%) and none had DED. The difference of 42% (95%CI: 21%, 63%) in followthrough completion rates between control and intervention groups was significant (p < 0.001).
Autonomous AI screening generated favourable feedback, with 92.5% of participants satisfied with the length of time it took to complete the exam, and 96% satisfied with the experience.
Of those in the intervention arm, 85% reported they would choose the AI-based eye exam in the future, while just 57% said they would choose an ECP-based diabetic eye exam.
Reference
Wolf, R.M., Channa, R., Liu, T.Y.A. et al. Autonomous artificial intelligence increases screening and follow-up for diabetic retinopathy in youth: the ACCESS randomized control trial. Nat Commun 15, 421 (2024). https://doi.org/10.1038/s41467-023-44676-z.