Prediction Tools for Diabetic Retinopathy Among Veterans Using Machine Learning and Large-scale Data Analysis Techniques

Diabetic retinopathy is a form of microvascular end-organ damage that can result in vision loss and is an outcome that Veterans suffer at a higher rate than the general population living with diabetes. Current retinopathy screening programs lack precision which can mean that high-risk patients struggle to get the care they need when they need it. Access to screening is limited by a mismatch of demand and supply of available eye care providers or teleretina screening opportunities. There is currently very little individualization in diabetic retinopathy screening and management strategies. Therefore, precision medicine approaches which attempt to match optimal screening strategy to the risk factors of an individual patient would be ideal to address the unmet need for patient-centered diabetic retinopathy screening and care. The overall goal of this career development award (CDA) is to develop sophisticated bioinformatics and machine learning tools to identify Veterans at high risk for retinopathy progression while at the same time developing Dr Christopher’s expertise in advanced statistical and machine learning strategies. Doing so will support her career goal of enhancing ophthalmic outcomes for Veterans, specifically targeting diabetic retinopathy. This proposal will leverage a major asset of the VA healthcare system, the electronic medical record, with clinical and administrative information available as the largest integrated healthcare system in the country. In Aim 1, we will develop and validate a phenotype for identifying patients with diabetic retinopathy from electronic health record data. In Aim 2, we will apply machine learning techniques for retinal image analysis to generate prediction tools for future diabetic retinopathy at the time of a negative retinal photography screening. Finally, Aim 3 will describe the phenomenon of early worsening of retinopathy following rapid improvement in glycemic control. This has not previously been described in large cohorts of patients, and we will explore predictors including the relative importance of medication type and HgA1c. This proposal is innovative in its leveraging of the existing VA diabetic teleretina screening program to generate a very large repository of retinal photographs for study, its focus on risk factors and machine learning strategies to predict future retinopathy, and development of tools which will lead to personalization of retinopathy screening protocols at the individual Veteran level. The career development plan is strategically designed to achieve these aims via the proposed didactic, technical, and career development objectives. Specifically, advanced statistical training, Big Data infrastructure within healthcare, and machine learning and artificial intelligence for predictive analyses educational goals will be realized with a combination of coursework, training, and mentorship. The assembled team of mentors, each outstanding investigators in their own fields, will provide content, methodologic, and career development guidance to fill gaps in the applicant’s training and ensure progress towards independence as a clinician- investigator. Upon successful completion of the proposed scientific and training aims, the applicant will be well-equipped to utilize large-scale data analysis techniques and machine learning to improve ophthalmic care for Veterans for years to come. Project Number: 1IK2CX002805-01A1 | Fiscal Year: 2025 | NIH Institute/Center: Veterans Affairs (VA) | Principal Investigator: Karen Christopher | Institution: VA EASTERN COLORADO HEALTH CARE SYSTEM, Aurora, CO | Activity Code: IK2 | Study Section: Special Emphasis Panel[ZRD1 NURF-H (01)] View on NIH RePORTER: https://reporter.nih.gov/project-details/11052679