GPADRlex: Grouped Phrasal Adverse Drug Reaction lexicon

Identification and monitoring of adverse drug reactions (ADRs) of marketed drugs from digital media can complement traditional pharmacovigilance activities. A phrasal lexicon is the need of systems that would monitor the ADRs from digital media. In this study, we compiled a large phrasal lexicon from the FDA Adverse Event Reporting System (FAERS) that is specific for ADRs, descriptive in nature and wherein the phrases representing the same ADR are grouped together using semantic similarity based hierarchical clustering. We explored six different methods to compute the phrasal semantic similarity and seven different linkage methods. We compared our automatic clustering with manually curated clusters on a subset of a previous ADR lexicon. Our results show that nine clustering schemes produced significantly better clusters than random clustering. Moreover, we provide a mechanism to generate user-defined k clusters from the 19,585 phrasal ADRs in our lexicon.