Identifying Indications for Drugs in Drug Package Inserts

Linking drugs and their indications can helpt improve the quality of electronic medical records (EMRs) and facilitate clinical decision support. We investigate an automatic approach to extraction of indications for marketed drugs from drug package inserts. Our aim is to build a reference table that associates medications and their indications. This resourcecould facilitate maintenance of medication and problem lists in the EMR. We are developing modules for downloading drug package inserts from the DailyMed site, parsing the XML file for each drug, and submitting the resulting texts to MetaMap. All relevant information will be represented by codes from standard terminology systems. Introduction Accurate and well-managed medication and problem lists are critical to success of Clinical Decision Support systems, pharmacovigilance, and other patient safety efforts. [1] Medication reconciliation depends on an accurate and up-to-date medication list. Previous research has shown that problem list and medication list data can be used to mutually validate each other. [2] One prerequisite to using this approach is to know what drugs are prescribed for what problems. Previous indications lists have been mostly hand-crafted, which are tedious to create and maintain. We are exploring a novel way to automatically extract drug indications from authoritative information in the FDA approved package inserts. Our aim is to establish a reference table linking drugs and their indications using identifiers from standard terminologies. Methods The DailyMed website provides access to thousands of package inserts. Each drug is linked to its RxNorm identifier (RXCUI). [3] We are developing modules for: (1) downloading drug package insert information in the Extensible Markup Language (XML) format available from the DailyMed website; (2) parsing these XML files and extracting free text in the indications section which is identified by the LOINC code 34067-9 FDA package insert Indications & usage section (3) dividing the indication text into smaller chunks making use of XML tags (e.g. titles, paragraphs or list items) to facilitate subsequent processing; (4) t processing and identifying the indications concepts (represented by their Unified Medical Language System (UMLS®) concept unique identifiers (CUIs) ) using MetaMap [4]. Results We tested a pool of 329 of the most frequently prescribed mono-ingredients drugs. The average size of indication sections in the sample files is 1176 characters. Further subdivision of these yielded 2242 text chunks. The chunking step allows to better situate the extracted concepts in the context of the insert. This information will be used in selecting final intervention candidates for a manual review of the reference table. The average size of these text chunks is 210 characters. The observed MetaMap processing time is significantly related to the number of indication terms within the chunk, rather than the size of the chunk file. In less than 1% of testing samples there were no UMLS concepts identified. DiscussionWe started exploration of automatic methods forbuilding a reference table containing indications formarketed drugs. The relations extracted from thedrug package inserts are stored as pairs of drugRXCUIs [5] and CUIs of the associated indications.Our next step is to validate the results by comparingthem with other publicly available sources of similarinformation e.g. NDF-RT (National Drug FileReference Terms) [6], and refine the algorithmaccordingly. References1. Joint Commission. 2010 National Patient SafetyGoals for Hospitals. Available:http://www.jointcommission.org/PatientSafety/NationalPatientSafetyGoals/http://www.nlm.nih.gov/research/umls/rxnorm/.2. Jao CS, Hier DB, Galanter WL. Using clinicaldecision support to maintain medication andproblem lists: A pilot study to yield higher patient safety. 2008 IEEE InternationalConference on Systems, Man and Cybernetics,10-16 October 2008, Singapore, 739-743.3. http://dailymed.nlm.nih.gov/.4. http://mmtx.nlm.nih.gov/.5. http://www.nlm.nih.gov/research/umls/rxnorm/overview.html.6. http://www.nlm.nih.gov/research/umls/sourcereleasedocs/2008AB/NDFRT/.