Authors: Himali Saitwal, BHMS, MS
Primary Advisor: David Qing, MD (co-author)
Committee Members: Stephen Jones, MD (co-author); Elmer Bernstam, MD, MS (co-author); Todd R. Johnson, PhD (co-author)
Masters thesis, The University of Texas School of Biomedical Informatics at Houston.
Information about medications, such as drug class, physiologic effect, mechanism of action, different drug components, etc. has a wide range of uses, but is not yet available in a single terminological system or integrated set of systems. Thus using medication information requires considerable work mapping across terminological systems, on a case-by-case basis. This paper reviews the current state of medication terminological systems and presents a case study. We mapped medication codes in a clinical data warehouse to the UMLS and SNOMED-CT. We found that three methods were required to accurately map the majority of actively prescribed medications: (1) Automatic mapping, using existing connections between terminological systems; (2) Partially automated mapping that made use of medication names to propose possible alternative matches to a human expert who made the final mapping; and (3) Manual mapping in which an expert experienced with medications and terminological systems manually mapped each medication. The completeness and accuracy of automatic and manual mapping resulted in excellent and comparable numbers (each above 99%) as opposed to partially automated mapping (77% and 65%). These original mappings from all methods were verified through manual review followed by final corrections, which resulted in overall completeness of 99.73% with an accuracy of 100% for entire mapping analysis. Compound drugs were especially difficult to map, because only 7.5% could be mapped using the automatic method. At the start of the project we could not predict the number or kinds of methods that would be required. The mapping effort required a total of one FTE for approximately one year. We conclude that better automatic mapping methods and truly integrated medication terminological systems are needed.