My research applies data science methods to improve emergency department (ED) operations and diagnostic quality, with a focus on closing the feedback gap between clinical decisions and patient outcomes. I lead analytic work within the Michigan Emergency Department Improvement Collaborative (MEDIC), a statewide quality collaborative whose registry captures structured clinical and process data from 50+ Michigan EDs and thousands of clinicians. This multi-site infrastructure enables population-level studies of practice variation, diagnostic decision-making, and care delivery efficiency.
A major line of my work uses the MEDIC registry to evaluate diagnostic quality in the ED workup of pulmonary embolism (PE). I helped develop a multi-stage decision-framework approach that decomposes the diagnostic process into discrete, measurable decision moments, allowing quantitative assessment of guideline-concordant care across large clinician populations. This framework supports both observational research and the design of precision feedback interventions — individualized, data-driven performance reports delivered to clinicians to promote evidence-based practice change.
I also develop AI pipelines for clinical text, including a hybrid de-identification system for radiology reports that combines spaCy named-entity recognition, rule-based pattern matching (philter), and regular expressions to enable safe use of free-text clinical data in research. These pipelines interface with large language model APIs for downstream analytic tasks such as classification and extraction.
Across these projects, I use Python for data engineering, NLP, and pipeline development; R and Stata for statistical modeling; and SQL-based querying of enterprise clinical data warehouses (HSDW, Epic Caboodle) for cohort construction, location-event modeling, and capacity analytics. My work sits at the intersection of health informatics, clinical epidemiology, and implementation science, with the shared goal of building learning health system infrastructure that translates routinely collected data into actionable knowledge for frontline clinicians.