METHODS
We have identified recent or ongoing systematic reviews that focused on deprescribing research-related questions using PROSPERO and active monitoring of MEDLINE via PubMed. A call for additional candidate reviews was sent out via Twitter. Eligible systematic review had to use the deprescribing definition developed by Reeve et al. 6, and must have developed a search strategy for MEDLINE via PubMed and Embase via Embase.com independently from the tested deprescribing search filters, i.e. authors were not aware of the maximized sensitivity, or USDeN search filters when they developed their search strategies.
First, authors were contacted and gave their consent to the use of their original search strategies. Information provided by systematic review authors included original search strategies for MEDLINE and Embase, a flowchart of the selection process, and data on excluded and included studies.
Then, we studied the deprescribing maximized sensitive filters for MEDLINE and Embase4 and the USDeN deprescribing filter for MEDLINE5 (table S1). One author (TM) implemented the deprescribing filters in each search strategy. For that purpose, all terms originally used by systematic review authors that referred to deprescribing were removed and replaced by the tested deprescribing filter. We thus obtained two implemented search strategies for MEDLINE, and one implemented search strategy for Embase, for each systematic review. MEDLINE and Embase were then searched using both the original and implemented search strategies on the same date. Articles retrieved from each strategy were included for performance calculation. Systematic review authors were then asked to complete the same selection process described in their original methods and to provide the list of excluded articles after title/abstract screening, after full-text screening and the list of included articles.
Performances of implemented search strategies were calculated and compared to performances of original search strategies. Performances calculated were sensitivity, i.e. proportion of articles included among relevant articles indexed in the database, precision, i.e.proportion of articles included among all articles retrieved by the search strategy, and the number needed to read (NNR) in title, abstract or full text to include one relevant article, i.e. the number of articles that need to be read to include one articles7.