OBJECTIVES: The National Library of Medicine (NLM) RxMix interface and the National Drug Code (NDC) Directory by the Food and Drug Administration (FDA) are publicly available sources to identify prescription drug use in pharmacoepidemiology studies. This study compared the performance of NLM RxMix interface, FDA NDC Directory, and Micromedex Redbook^® to identify frequently prescribed medication records in Medical Expenditure Panel Survey Household Component (MEPS-HC) data using the NDC generic name mapping tables.

METHODS: This cross-sectional study included records from the 2017-2018 MEPS-HC prescribed medicine file. Records for the top 25 most prescribed medications were retained based on the generic name. Records with a missing NDC were removed. The generic name to NDC mapping tables was created from all three sources of NDC for the previously identified records. For each of the 25 most prescribed medications, the proportions of records identifiable in the MEPS using all three sources of NDC were compared.

RESULTS: The study identified an unweighted sample of 258,114 prescription records representing 2.63 billion annualized records nationally. Micromedex Redbook^® based NDC to generic name mapping tables identified 99.89% prescription records, whereas NLM and FDA-based mapping tables identified 93.16% and 93.23% records, respectively. The performance of NLM RxMix and FDA NDC directory-based mapping tables varied across different drugs. NLM RxMix had the lowest performance for citalopram (45.42%), alprazolam (68.65%), and metoprolol (82.05%), whereas the FDA NDC directory performance was lowest for levothyroxine (69.91%), acetaminophen (75.46%), and simvastatin (84.10%).

CONCLUSIONS: The coverage by the publicly available NLM RxMix and FDA NDC directory mapping table for the 25 most frequently prescribed medications was highly variable. Micromedex Redbook^®, in contrast, provides complete capture of drug use. Researchers should carefully select drug knowledge base sources for pharmacoepidemiology research as it can significantly impact the study populations and associated outcomes.