Semi-automatic translation of medicine usage data (in Dutch, free-text) from Lifelines COVID-19 questionnaires to ATC codes

Abstract

The mapping of human-entered data to codified data formats that can be analysed is a common problem across medical research and health care. To identify risk and protective factors for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) susceptibility and coronavirus disease 2019 (COVID-19) severity, frequent questionnaires were sent out to participants of the Lifelines Cohort Study starting 30 March 2020. Because specific drugs were suspected COVID-19 risk factors, the questionnaires contained multiple-choice questions about commonly used drugs and open-ended questions to capture all other drugs used. To classify and evaluate the effects of those drugs and group participants taking similar drugs, the free-text answers needed to be translated into standard Anatomical Therapeutic Chemical (ATC) codes. This translation includes handling misspelt drug names, brand names, comments or multiple drugs listed in one line that would prevent a computer from finding these terms in a simple lookup table. In the past, the translation of free-text responses to ATC codes was time-intensive manual labour for experts. To reduce the amount of manual curation required, we developed a method for the semi-automated recoding of the free-text questionnaire responses into ATC codes suitable for further analysis. For this purpose, we built an ontology containing the Dutch drug names linked to their respective ATC code(s). In addition, we designed a semi-automated process that builds upon the Molgenis method SORTA to map the responses to ATC codes. This method can be applied to support the encoding of free-text responses to facilitate the evaluation, categorization and filtering of free-text responses. Our semi-automatic approach to coding of drugs using SORTA turned out to be more than two times faster than current manual approaches to performing this activity. Database URL https://doi.org/10.1093/database/baad019

Disentangling the heterogeneity of multiple sclerosis through identification of independent neuropathological dimensions

“…interferon beta-1a) were manually transformed into one word (connected with dashes). The resulting drug texts were matched to standard Anatomical Therapeutic Chemical (ATC) codes using the MOLGENIS SORTA parser (version 10.1.0) [ 45 , 60 ], modified for drug to ATC matching by Kellmann et al [ 26 ], with the confidence threshold for the similarity score set to 78.5%. For each timepoint, higher ontology level ATC codes were removed if a more specific ATC code (within that same class) was matched as well, to prevent double matching/counting of drug texts.…”