Our lab focuses on early changes in the retinas of patients with diabetes and leveraging technology to reduce vision loss from diabetes. Diabetes is the leading cause of vision impairment among working-age adults in the United States, contributing significantly to both individual and public health burdens.
Diabetes Registry
This study aims to establish a research data repository and recruitment registry for patients with diabetes. Eligible participants include individuals diagnosed with diabetes, including minors, who consent to participate in the study. Participants are approached for enrollment in the study upon visiting the UW Health Eye Clinic. After consent is obtained, they undergo research imaging, including RetEval Flash Electroretinography (RetEval), optical coherence tomograohy and fundus photography. The study team follow up with participants once a year in line with standard care protocols, including RetEval testing and fundus photography to track changes in the participants’ eye health over time.
Projects on retinal changes in patients with diabetes using large-scale datasets
1) Assessing retinal layer changes in patients with diabetes using the UK Biobank dataset
Diabetes mellitus is a major global health issue and a leading cause of vision loss due to its damaging effects on the retina, known as diabetic retinopathy. These retinal changes often begin before visible symptoms appear, emphasizing the need for early detection methods.
The UK Biobank, a large-scale biomedical database from over 500,000 participants, provides a powerful resource for investigating the ocular effects of diabetes in a real-world population. By leveraging optical coherence tomography (OCT) imaging data within the UK Biobank, we assess the thickness and integrity of total and individual retinal layers in diabetic versus non-diabetic cohorts. This approach enables the identification of subclinical retinal changes associated with diabetes, even in the absence of apparent retinopathy. This could lead to earlier interventions and personalized treatments, ultimately helping to prevent vision loss and improve outcomes for diabetic patients.
2) All of US dataset
The All of Us dataset is a rich, longitudinal health resource capturing data from a diverse and nationally representative population across the United States. Using this dataset, we examined national patterns of diabetic eye screening, including both in-person dilated eye exams and teleophthalmology exams, and identified factors associated with screening receipt among individuals with diabetes. Additionally, we evaluated the national trend of teleophthalmology exams over time.
Leveraging artificial intelligence to prevent vision loss from diabetes
This study is a stepped-wedge cluster randomized pragmatic clinical trial conducted among patients with diabetes across nine Federally Qualified Health Centers (FQHCs). The aim is to evaluate whether AI-BRIDGE, an autonomous artificial intelligence (AI)-based protocol for screening diabetic eye disease, improves screening and follow-up eye care compared to usual care. Additionally, the study seeks to identify key characteristics and strategies to promote the effective implementation of AI-BRIDGE in clinical practice.
The AI-BRIDGE intervention consists of three main components: (1) the acquisition of eye photos and the autonomous, AI-based identification of referrable or non-referrable eye disease at the primary care clinic, without human oversight, (2) the provision of culturally adapted educational materials for patients, identical to those given in the usual-care group, and (3) clinic champion assisted scheduling of follow-up eye care visits if they have referrable eye disease. In the usual-care screening group, patients with diabetes will follow the clinic’s standard practice, where the primary care provider recommends an annual eye exam. This requires the patient to make a separate visit to an eye care provider, with clinic staff offering scheduling assistance. During their primary care visit, patients will receive culturally adapted educational materials about diabetic eye disease.
Culturally adapted educational materials were revised after meeting with the OCHIN Patient Engagement Panel to incorporate their insights, ensuring that language and imagery reflect the diverse backgrounds and values of the target community.
Image-based genome-wide association study (iGWAS)
This is a collaborative study between the University of Wisconsin (UW) and UTHealth. The objective is to develop deep learning-based endophenotypes from retinal images, utilizing data from de-identified, publicly available datasets, including the UK Biobank and the DRCR Retina Network. Fundus images from the DRCR Retina Network, specifically those with a diagnosis of diabetic retinopathy, were annotated by the UW grading center. Using these annotated images, we developed AI algorithms to classify microaneurysms and hemorrhages.
Evaluating the Impact and Cost-Effectiveness of AI-Based Screening for Diabetic Eye Disease
We evaluated the effectiveness of AI screening using a Markov model known as CAREVL. Results showed that AI-based screening led to fewer cases of vision loss over a five-year period, 1535 cases per 100,000 patients compared to 1625 cases in the traditional eye care provider group. This reduction suggests that AI screening could potentially prevent vision loss in approximately 27,000 Americans over five years. These findings are detailed in a study (DOI: 10.1038/s41746-023-00785-z).
Cost-effectiveness is a key aspect of AI-based diabetic eye screening. We evaluated this through two separate studies. One study found that point-of-care diabetic retinopathy screening using autonomous AI systems is both effective and cost-saving for children with diabetes and their caregivers, particularly when adherence to recommended screening guidelines is met. In a second analysis, we compared the cost-effectiveness of AI-based screening with traditional screening by eye care providers (ECPs), focusing on the first year of implementation from a health system perspective. The results showed that depending on the size of the health system and patient volume, AI-based screening consistently reached more patients and offered greater cost savings.
