Development and validation of search filters to retrieve medication discontinuation articles in Medline and Embase

BackgroundMedication discontinuation studies explore the outcomes of stopping a medication compared to continuing it. Comprehensively identifying medication discontinuation articles in bibliographic databases remains challenging due to variability in terminology.ObjectivesTo develop and validate search filters to retrieve medication discontinuation articles in Medline and Embase.MethodsWe identified medication discontinuation articles in a convenience sample of systematic reviews. We used primary articles to create two reference sets for Medline and Embase, respectively. The reference sets were equally divided by randomization in development sets and validation sets. Terms relevant for discontinuation were identified by term frequency analysis in development sets and combined to develop two search filters that maximized relative recalls. The filters were validated against validation sets. Relative recalls were calculated with their 95% confidences intervals (95% CI).ResultsWe included 316 articles for Medline and 407 articles for Embase, from 15 systematic reviews. The Medline optimized search filter combined 7 terms. The Embase optimized search filter combined 8 terms. The relative recalls were respectively 92% (95% CI: 87-96) and 91% (95% CI: 86-94).ConclusionsWe developed two search filters for retrieving medication discontinuation articles in Medline and Embase. Further research is needed to estimate precision and specificity of the filters.