DREAM: Building Equitable Predictive Models for Personalized Diabetes Pharmacotherapy

Although the American Diabetes Association (ADA) guidelines help clinicians choose medications for type 2 diabetes based on comorbidities, they primarily offer broad class-based recommendations. However, more granular decisions are needed to account for intra-class drug differences, considering factors like HbA1c reduction, weight impact, hypoglycemia risk, cost, and side effects. Thus, it is crucial to explore whether real-world data could provide insights beyond what is currently available in the ADA guidelines to support personalized clinical decisions. This K01 project aims to address these gap by developing Diabetes Recommendations Enhanced with AI Models (DREAM), a set of predictive models leveraging large, cross-institutional electronic health record (EHR) datasets, including Epic Cosmos, TriNetX, and AHEAD datasets. These models will predict HbA1c, weight, and hypoglycemia events 6 months into the future for each ADA-guideline compliant medication option. The project has three specific aims: (1) Develop DREAM 1.0, a set of predictive models to estimate individualized outcomes for pharmacotherapy options, including HbA1c, weight, and hypoglycemia; (2) Characterize variation in diabetes pharmacotherapy across patient subgroups using clustering methods to identify prescription patterns by demographic and clinical characteristics; and (3) Improve predictive model performance across patient subgroups. In addition to the research aims, this K01 award will provide Dr. Polina Kukhareva with training in diabetes pharmacotherapy, artificial intelligence, and subgroup-specific performance evaluation. Her four key career goals are: (1) gaining expertise in diabetes pharmacotherapy through specialized courses and clinical experiences; (2) acquiring advanced AI skills for EHR data analysis, including deep learning and clustering, and learning bias mitigation techniques; (3) strengthening understanding of clinical variability and its implications for AI development; and (4) building research leadership and grant-writing skills to lead multidisciplinary research teams and prepare for NIH funding. Dr. Daniel Malone, a recognized leader in pharmacotherapy and pharmacoepidemiology, will serve as Dr. Kukhareva’s primary mentor, supported by an interdisciplinary team of experts in AI and diabetes management. This mentorship, combined with the University of Utah's strong research environment, will equip Dr. Kukhareva with the tools and expertise to become a leader in developing AI solutions for personalized type 2 diabetes care.