PhRMA Foundation’s Challenge Awards are competitions that invite researchers to submit papers or research proposals addressing pressing issues in research and health care.

Hsu is one of three trainees whose paper received a $5,000 Trainee Challenge Award as part of a Journal of Managed Care & Specialty Pharmacy (JMCP) themed issue on artificial intelligence (AI) in managed care pharmacy.

“This recognition increases the visibility of my work, supports new interdisciplinary collaborations, and strengthens my professional development at the intersection of medication safety, health informatics and AI,” said Hsu. “It also advances my long-term goal of becoming an independent investigator who develops evidence-based tools to improve medication use, care delivery and health outcomes among high-risk populations.”

Hsu is first author of the paper, “An informatics framework to harmonize electronic health record medication data for managed care analytics and artificial intelligence applications,” published in JMCP’s themed issue on AI in managed care pharmacy, a collection of nine articles exploring current applications of AI in the field. Co-authors include Gabe Gazetta de Araujo, PhD, a postdoctoral associate in pharmacy practice; Chi-Hua (Richard) Lu, PharmD/MS, research scientist in pharmacy practice; and David Jacobs, PharmD, PhD, associate professor of pharmacy practice.

In their paper, Hsu and colleagues examine a key challenge: medication information is often recorded differently across electronic health record systems, limiting its reliability for research, managed care analytics and AI development and application. Their study developed and evaluated a two-layer informatics framework that converts heterogeneous medication identifiers into standardized representations: first at the medication-ingredient level using RxNorm ingredient concept, then at the pharmacologic-class level using the Anatomical Therapeutic Chemical classification system. The framework was applied to 214,080 discharge medication records from adults aged 65 years or older hospitalized at Buffalo General Medical Center between 2020 and 2024. It combined standardized terminology crosswalks, medication-name matching and pharmacist validation to identify and correct incomplete or inaccurate mappings. 

Their findings are significant because reliable real-world analytics require medication data that are clinically and semantically consistent. Incomplete or inaccurate identifiers can distort drug-exposure estimates, safety assessments and population-level findings. Their work demonstrates a transparent approach for standardizing heterogeneous electronic health record medication data into meaningful ingredient and pharmacologic-class representations. In managed care pharmacy, such harmonization can strengthen medication-use evaluations, safety surveillance, risk stratification, comparative effectiveness research and quality improvement. AI applications depend on standardized inputs because medication-mapping errors can propagate into risk predictions and decision-support tools. Their findings underscore the importance of clinical validation and pharmacist oversight in developing trustworthy and transportable applications.

Their findings are also significant in addressing data quality issues for real-world analytic and AI development. The team combined terminology crosswalks, structured medication-name reconciliation and pharmacist validation rather than relying on automated mapping alone. Their study demonstrates that high initial mapping rates can conceal clinically meaningful classification errors involving branded products, medication names, routes and indications. By making medication data more accurate, transparent, and interoperable, the framework provides a practical foundation for reproducible real-world evidence, managed care analytics and trustworthy AI development and applications across health care systems.

The team acknowledges the UB Institute for Health Informatics for its assistance with data provision and for maintaining a secure environment for data processing and analysis.