1. Applying and developing simulation models to evaluate the long-term value of diabetes interventions and policies. I have contributed to multiple projects using patient-level simulation models to evaluate the long-term clinical and economic value of diabetes interventions. My work has included cost-effectiveness analyses comparing newer glucose-lowering therapies, including GLP-1RA and SGLT2i, with traditional drug classes, as well as analyses estimating value-based price benchmarks for glucose-lowering medications selected under the Inflation Reduction Act. Through these studies, I applied the first U.S.- ased patient-level BRAVO diabetes microsimulation model and developed expertise in using simulation models to translate trial, observational, and economic data into policy-relevant evidence. Supported by a PhRMA Foundation Predoctoral Fellowship, I developed the first U.S.-based microsimulation model designed to project multimorbidity outcomes among Veterans. This work advances decision modeling science by providing tools to quantify the long-term benefits and costs of prevention and treatment strategies in populations with complex chronic disease burdens, thereby strengthening the link between clinical evidence, health economics, clinical decision-making, and policy-making in multimorbidity care.
2. Generating real-world evidence on medication use, access, safety, and disparities in diabetes care. I have also made important contributions to understanding real-world medication use, access, off-label prescribing, and disparities in diabetes care. Using Epic Cosmos data, I led a national study showing increasing off-label use of GLP-1 receptor agonists and SGLT2 inhibitors among people with type 1 diabetes. This work, published in JAMA, highlighted an emerging clinical practice pattern and raised important safety considerations, including the need for careful monitoring of adverse events such as diabetic ketoacidosis and other treatment-related risks. Beyond off-label use, I have conducted multiple studies quantifying disparities in the adoption of newer glucose-lowering medications and diabetes management strategies, with attention to racial, ethnic, socioeconomic, and insurance-related gaps in access and outomes. These studies contribute to the growing evidence base on inequities in diabetes pharmacotherapy and identify opportunities for more equitable implementation of effective treatments. I am currently a coinvestigator on an NIDDK-funded R01 project that uses innovative machine-learning approaches to inform precision public health strategies designed to reduce barriers to newer diabetes medications. Through this work, I aim to generate evidence that not only describes disparities, but also informs actionable strategies to improve access, safety, and outcomes for populations most affected by diabetes.
3. Using causal machine learning to identify heterogeneous treatment effects and support precision intervention strategies. A major contribution of my work has been the application and advancement of causal machine-learning methods to identify heterogeneity in treatment effects across clinically meaningful subgroups. I have led and contributed to studies using causal forests, causal trees, and related methods to identify patients who are more likely to benefit from, or experience harm with, specific treatments and interventions. This body of work aims to move beyond one-size-fits-all treatment strategies by generating evidence to support precision medicine and precision public health. For example, in a reanalysis of the ACCORD MIND trial, my work showed that renal function modified the effect of intensive glucose control on cognitive decline. In a reanalysis of the HARMONY Outcomes trial, we identified diabetes duration and HbA1c as key modifiers of the cardiovascular benefit of albiglutide. Other analyses have identified subgroups with greater benefit from intensive glycemic or blood pressure control, highlighting the clinical importance of individualized treatment decisions. I have also extended this approach to population-level prevention by evaluating heterogeneity in the long-term effectiveness of the VA MOVE! lifestyle intervention. This work showed that individuals with higher baseline cardiometabolic risk may derive greater long-term benefit, suggesting that targeting high-risk and high-benefit populations could improve the efficiency and equity of prevention programs. Together, these studies demonstrate how causal AI methods can generate actionable evidence for tailoring interventions to the right populations.
4. Advancing EHR-based methods for diabetes surveillance and population health research. A central theme of my research has been developing methods that improve the use of EHRs for epidemiologic surveillance and real-world evidence generation. I led the development of an automated computable phenotype to identify diabetes cases among children and adolescents using EHR data, helping address a major methodological challenge in pediatric diabetes research: reliably distinguishing diabetes status and subtype in routinely collected clinical data. Building on this work, I developed a novel health system penetration-based approach to estimate type 1 diabetes prevalence using EHR data. This method addresses partial observation bias arising from the fragmented U.S. healthcare system, where no single health system captures all individuals in a geographic population. By improving the validity and generalizability of EHR-based estimation, this work expands the capacity of real-world data to support disease surveillance, monitor population health trends, and generate timely evidence across diverse geographic regions. Collectively, these contributions provide practical tools for transforming EHR data into scalable infrastructure for diabetes surveillance and public health research.