Mapping Medication Orders to RxNorm Concepts

Authors

Jenny Staab, Kaiser Permanente NorthwestFollow
Andy Karnopp, Kaiser Permanente Northwest

Publication Date

8-15-2016

Keywords

RxNorm, pharmacy

Abstract

Background/Aims: There is a multitude of coding or classification systems for medications. While dispensed medications have traditionally been identified using National Drug Codes (NDCs), there has been no equally universal vocabulary for ordered medications. However, RxNorm, a nonproprietary drug terminology system developed by the National Library of Medicine, has become increasingly popular. Both PCORnet and the Observational Medical Outcomes Partnership’s Common Data Models have adopted RxNorm as a standard vocabulary for medications. RxNorm concepts range in specificity from branded packaged drugs to simple ingredients and can be linked to many proprietary and nonproprietary vocabularies such as NDCs or Medi-Span Generic Product Identifier (GPIs). We developed a method to map drugs from drug names and other drug terminologies to RxNorm concept unique identifiers (CUIs).

Methods: Data on ordered medications were extracted from Kaiser Permanente Northwest’s Clarity database. These data contained text strings identifying drugs and generic ingredient names as well as codes from different drug vocabularies including RxNorm, GCN Sequence Numbers, GPIs and NDCs. These vocabularies were available for overlapping subsets of the medications list. We used directly available CUIs and crosswalks linking other vocabularies to RxNorm. For drugs that could not be mapped satisfactorily, we also employed approximate string matching using RxMix. Potential concept matches were ranked using a scoring algorithm and the best match chosen. CUIs deemed highly reliable, such as those provided directly from Clarity or deemed a perfect match by RxMix, scored higher. CUIs scored lower if they were derived from obsolete NDCs or scored poorly on RxMix.

Results: We found CUIs for over 95% of orders; 75% of matches were fully specified drugs, the remaining were mostly ingredients. About 10% of CUIs were found only via string match, with an average confidence score of 75%.

Conclusion: There are multiple ways to map medications to RxNorm concepts. Using direct crosswalks based on standardized terminology is preferable, as it affords higher certainty. However, approximate string matching can provide mappings for medications without available standardized terminology. It also can improve data quality by mapping to more specific RxNorm concepts and supporting validation of the direct mapping results. The adopted level of certainty for inclusion in the dataset should be well documented.

Recommended Citation

Staab J, Karnopp A. Mapping medication orders to RxNorm concepts. J Patient Cent Res Rev. 2016;3:223.