Artificial Intelligence Closes Care Gaps in Diabetic Eye Exams

Risa Wolf, MD

Credit: Johns Hopkins Medicine

Autonomous artificial intelligence (AI) diabetic eye exams increase screening rates for youth with diabetes and close further care gaps among a racially and ethnically diverse population versus the standard of care, according to new research.¹

Analyses of the ACCESS randomized controlled trial revealed these findings remained accurate despite an expansion in the standard of care referral among the control arm, consisting of deliberate education for the patient and caregiver on the importance of diabetic eye disease.

“With AI technology, more people can get screened, which could then help identify more people who need follow-up evaluation,” said lead study author Risa M. Wolf, MD, a pediatric endocrinologist at Johns Hopkins Children’s Center.² “If we can offer this more conveniently at the point of care with their diabetes doctor, then we can also potentially improve health equity, and prevent the progression of diabetic eye disease.”

Screening and early detection are factors that can prevent the progression of diabetic eye disease. Still, challenges, including a lack of access and education around their importance, have led to a care gap for many of the 34 million with diabetes in the US.³ These gaps are particularly notable for health disparities among racial and ethnic minorities and under resourced communities, who experience worse outcomes and a higher prevalence of diabetic eye disease.

Advances in telemedicine have improved early detection rates of diabetic eye disease – however, diagnostic autonomous AI systems for diabetic eye disease may represent the next generation of screening capabilities. Wolf and colleagues hypothesized that autonomous AI diabetic eye exams at the point-of-care would increase diabetic eye exam completion rates among a racially and ethnically diverse youth population.¹

The ACCESS trial was a parallel, randomized controlled trial that included youth with type 1 diabetes (T1D) (11 to 21 years) or type 2 diabetes (T2D) (8 to 21 years) if they met the criteria for diabetic eye disease screening per American Diabetes Association (ADA) 2021 guidelines, had no known diabetic eye disease, and had not had a diabetic eye exam within the last six months. Participants in ACCESS enrolled between November 2021 and June 2022, with follow-up completed by December 2022.

Enrolled participants were randomized 1:1 to the intervention arm, consisting of an autonomous AI diabetic eye exam at the point-of-care or the control arm, including scripted eye care provider referral and education. The pre-specified primary outcome was the completion rate of diabetic eye exams within six months of randomization, while the pre-specified secondary outcome was the amount who completed follow-through with an eye care provider if considered appropriate.

Overall, 177 individuals were eligible for study inclusion; 164 participants completed informed consent and were randomized to the intervention (n = 81) and control (n = 83) arms. Baseline characteristics were similar between groups, with a mean age of 15.2 years, 58% female participants, and a median duration of diabetes of 5.8 years. Data revealed that 79% of participants reported having a prior diabetic eye exam, while 21% experienced a care gap.

Regarding the primary outcome, the diabetic eye exam completion rate was significantly higher (100% [95% CI, 95.5 - 100]) in the intervention arm compared with the control arm (22% [95% CI, 14.2 - 32.4]). Analyses suggested the difference of 78% (95% CI, 69 - 87) in gap closure between the control and intervention groups was statistically significant (P <.001). Investigators noted no statistically significant differences by race, ethnicity, socioeconomic status, or education.

Within the intervention arm, 25 individuals received a “diabetic eye disease present” output, with 16 completing an eye care provider visit within six months for a follow-through completion rate of 64% (95% CI, 43 - 81). By comparison, 18 participants in the control arm visited an eye care provider for a follow-through completion rate of 22% (95% CI, 14 - 32), while none had a diabetic eye disease. Wolf and colleagues indicated the 42% difference (95% CI, 21 - 63) in follow-through completion between control and intervention arms was significant (P <.001).

Patients in ACCESS reported a high level of satisfaction with the autonomous AI-based screening, with 96% of participants satisfied with the experience and 85% of participants in the intervention arm indicating their decision to do the AI-based eye exam in the future.

“The high satisfaction and acceptance rates for autonomous AI in ACCESS, suggest that this racially and socioeconomically diverse patient population is comfortable with a 'computer' or autonomous AI diagnosing their disease,” Wolf and colleagues wrote.¹ “Importantly, the use of AI did not introduce health disparities into care-gap closure.”

References

1. _{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}
2. _{Medicine JH. Study finds AI-driven eye exams increase screening rates for youth with diabetes. Newswise. January 11, 2024. Accessed January 15, 2024. https://www.newswise.com/articles/study-finds-ai-driven-eye-exams-increase-screening-rates-for-youth-with-diabetes.}
3. _{Hill-Briggs, F. et al. Social determinants of health and diabetes: a scientific review. Diabetes Care 44, 258–279 (2020).}