On Mapping Textual Queries to a Common Data Model

The widespread adoption of Electronic Health Records (EHRs) has enabled data-driven approaches to clinical care and research. However, the performance and generalizability of those approaches are severely hampered by the lack of syntactic and semantic interoperability of EHR data across institutions. Towards resolving this problem, Common Data Models (CDMs) can be used to standardize the clinical data in clinical data repositories. In this paper, we described our mapping of entity mention types from patient-level information retrieval queries to an empirical subset of Observational Medical Outcomes Partnership (OMOP) CDM data fields. We investigated the empirical data model by annotating multi-institutional clinical data requests in free text and comparing the distributions of data model fields. The similar distribution of the entity mention types from two different sites indicates that the data model is generalizable for multi-institutional cohort identification queries.