Autonomous AI to mitigate disparities for diabetic retinopathy screening in youth during and after COVID-19

点击翻译按钮获取中文摘要

Project Summary COVID-19 has led to disruptions and delays in routine pediatric care. For children with diabetes who see their diabetes team every 3 months, this has been reduced or transitioned to telemedicine due to COVID- 19. However, those without technology and internet capabilities, namely low income and minority youth, are less likely to participate in telemedicine and may see their diabetes team less frequently during the pandemic. Screening for diabetes complications, such as diabetic retinopathy (DR), is generally fulfilled by a separate visit to an eye-care professional (ECP), and is also less likely to occur during COVID-19. Diabetic retinopathy affects 4-15% of youth with type 1 and type 2 diabetes and is a leading cause of blindness in adults as early as age 20. Yearly screening for DR is recommended, but only 35-72% of youth undergo screening, with minority youth and children from lower socioeconomic backgrounds less likely to undergo screening. Early detection of DR through screening prevents progression to vision loss. The current standard of care for pediatric DR screening is referral to an ECP for a dilated eye exam. Recently, the FDA approved the first autonomous artificial intelligence (AI) software that interprets retinal images taken with a non-mydriatic fundus camera, providing an immediate result for DR screening at the point of care (POC) for adults with diabetes. In a pilot study at our institution, we were the first to implement this technology in pediatrics, demonstrating safety, effectiveness and equity, and cost-savings to the patient. We also found that minority youth, those with lower household income and Medicaid insurance were less likely to undergo recommended screening, yet were more likely to have DR. This is likely to worsen due to the disparate effects of COVID-19. We hypothesize that implementing POC autonomous AI in the diabetes care setting will increase DR screening rates in youth with diabetes, mitigate disparities in access to screening, and be cost-effective to the health care system now and beyond the COVID-19 pandemic. In this proposal, we will first determine (Aim1) in a randomized control trial at two clinic sites if autonomous AI increases screening compared to ECP, and if those who screen positive by AI are more likely to go for follow-up at the ECP. In the second phase of this proposal (Aim2) we will perform a prospective observational trial of AI screening to determine if AI mitigates disparities in screening, and improves the proportion of at-risk, minority and low income, youth who go for follow-up if their AI screen is positive. In Aim 3, we will use a decision model to determine if AI is cost-effective and cost-savings to the health care system. If AI is shown to increase screening rates while mitigating disparities in access to care, it has the potential to reshape screening methods now and in the future.