Presentation (CTI)

OBJECTIVES: Outcomes research using real-world data requires reproducible methods to reconcile heterogeneous drug terminologies and construct consistent, analyzable datasets. Medication records in regional electronic health record (EHR) systems are frequently non-curated and inconsistently coded. Existing crosswalks are often applied in an ad hoc manner, limiting reproducibility and pharmacologic interpretation. We propose an ontology-informed, layered mapping strategy that supports analytic feasibility while establishing a reproducible conceptual foundation for addressing interoperability challenges.
METHODS: We analyzed 2020-2024 discharge medication records in a regional EHR database, excluding vaccines, supplements, and inpatient-only drugs. An ontology-informed mapping assigned complementary conceptual roles to drug terminologies instead of treating them as interchangeable. RxNorm served as an anchor for identifying drug use at the active-ingredient level. The Anatomical Therapeutic Chemical (ATC) classification was incorporated as an abstraction layer to encode pharmacologic properties. National Drug Codes (NDCs) were mapped to RxNorm through the National Library of Medicine’s crosswalk. RxNorm further supported reconciliation with Multum-coded entries in EHR, which weren’t compatible with standard MMSL identifiers as distributed via RxNorm. This layered strategy enables consistent analytic dataset construction while preserving both ingredient-level specificity and mechanism-based classification.
RESULTS: Medication records from 29,123 patients were coded as Multum (62.2%), NDC (27.3%), RxNorm (9.6%), or unknown (0.9%). Multum-coded records predominated and exhibited substantial heterogeneity. Applying the proposed strategy, 5,806 unique Multum-coded medication records were assigned initial RxNorm and corresponding ATC representations. Line-by-line reviews resulted in modifications to 3,712 records (63.9%), including 1,146 exclusions, 1,556 RxNorm corrections, and 2,757 ATC code corrections, reflecting inconsistencies in ingredient identification and pharmacologic classification.
CONCLUSIONS: Applying an ontology-informed approach to medication interoperability, this method provides a reproducible basis for reconciling non-curated EHR medication data. It supports analytic readiness while establishing a transferable conceptual approach that can be reapplied when encountering similar terminology heterogeneity and extended in the future to ontology-based pharmacoepidemiology and outcomes research.