Regency EF
An Algorithm for Post-Processing Medication Information Extracted by Natural Language Processing Systems from Electronic Health Records (304045)
Cole Beck, Vanderbilt University Medical Center
*Leena Choi, Vanderbilt University Medical Center
Elizabeth McNeer, Vanderbilt University Medical Center
Hannah Weeks, Vanderbilt University Medical Center
Michael Williams, Vanderbilt University Medical Center
Keywords: electronic health records, post-processing algorithm, natural language processing, medication data, pharmacokinetic, pharmacodynamic, EHR
As longitudinal medication data are often available in electronic health records (EHRs), large-scale population studies examining drug exposure effects on health outcomes can be performed using EHRs, if all relevant information can be accurately and efficiently retrieved. However, obtaining precise medication information from EHRs is challenging, as a large portion of dose information must be extracted from unstructured clinical notes. Although there are several natural language processing systems (NLPs) that may be used to extract medication information, the raw output directly obtained from NLPs is not usable for the analyses. First, the raw output should be parsed to get individual entities (e.g., drug name, strength, dose amount, frequency). Then, the parsed individual entities should be paired up to define dose given intake at each day, which requires a sophisticated post-processing algorithm. We have developed an algorithm for this purpose using clinical notes from Vanderbilt University Medical Center EHRs. This algorithm would be useful to provide medication information critical for diverse clinical research, including population pharmacokinetic and pharmacodynamic studies.