PubMed ‘Early Alerts’: Towards Better Precision of Literature Searching for Pharmacovigilance Information based on an Assessment of Relevance Feedback
Thursday, September 15, 2016 — Poster Session II
- AM Ripple
- AF Sorbello
- S Haider
- O Bodenreider
PubMed 'Early Alerts' provide FDA reviewers with weekly topical searches of the newest submitted citations to PubMed to support detection of emerging adverse drug events. We seek to increase the precision based on an assessment of relevance feedback for a subset of retrieved citations for the antidiabetic medications. Using a search optimized for recall and focusing on the text word fields, four regulatory evaluators assessed 30 citations each for relevance to drug safety and efficacy drawn from a random sample of 120 new reports. Candidate precision terms were identified by word frequency analysis using Word Counter Tool, an online word counter. We performed a differential frequency analysis of text word occurrences for the relevant compared to non-relevant citations. Based on reviewers' feedback, half of the reports in the 120 citations were relevant to safety, efficacy, or both, whereas the remainder were non-relevant for both safety and efficacy. The candidate precision words 'efficacy', 'safety', 'adverse', and 'risk' were identified by expert opinion from the titles or abstracts of the relevant citations among frequently occurring words. 'Efficacy, 'safety', and 'risk' had no title occurrences in the non-relevant citations. However, 'risk' and 'safety' are not discriminant, because they also occurred in the abstracts of non-relevant citations, leaving 'adverse' as the only candidate precision word derived from the abstract field. Keywords field did not yield any candidate precision text words. We identified several candidate text words to potentially increase the precision of our search strategy for pharmacovigilance information, but further validation is required.
Category: Research Support Services