Evaluation of PubMed filters used for evidence-based searching: validation using relative recall.

TitleEvaluation of PubMed filters used for evidence-based searching: validation using relative recall.
Publication TypeJournal Article
Year of Publication2009
AuthorsHoogendam A, de Vries Robbé PF, Stalenhoef AFH, Overbeke JAPM
JournalJournal of the Medical Library Association : JMLA
Volume97
Issue3
Pagination186-93
Date Published2009 Jul
ISSN1558-9439
KeywordsAbstracting and Indexing as Topic; Evidence-Based Medicine; Information Storage and Retrieval; PubMed
Abstract

OBJECTIVES: The research sought to determine the value of PubMed filters and combinations of filters in literature selected for systematic reviews on therapy-related clinical questions.

METHODS: References to 35,281 included and 48,514 excluded articles were extracted from 2,629 reviews published prior to January 2008 in the Cochrane Database of Systematic Reviews and sent to PubMed with and without filters. Sensitivity, specificity, and precision were calculated from the percentages of unfiltered and filtered references retrieved for each review and averaged over all reviews.

RESULTS: Sensitivity of the Sensitive Clinical Queries filter was reasonable (92.7%, 92.1-93.3); specificity (16.1%, 15.1-17.1) and precision were low (49.5%, 48.5-50.5). The Specific Clinical Queries and the Single Term Medline Specific filters performed comparably (sensitivity, 78.2%, 77.2-79.2 vs. 78.0%; 77.0-79.0; specificity, 52.0%, 50.8-53.2 vs. 52.3%, 51.1-53.5; precision, 60.4%, 59.4-61.4 vs. 60.6%, 59.6-61.6). Combining the Abridged Index Medicus (AIM) and Single Term Medline Specific (65.2%, 63.8-66.6), Two Terms Medline Optimized (64.2%, 62.8-65.6), or Specific Clinical Queries filters (65.0%, 63.6-66.4) yielded the highest precision.

CONCLUSIONS: Sensitive and Specific Clinical Queries filters used to answer questions about therapy will result in a list of clinical trials but cannot be expected to identify only methodologically sound trials. The Specific Clinical Queries filters are not suitable for questions regarding therapy that cannot be answered with randomized controlled trials. Combining AIM with specific PubMed filters yields the highest precision in the Cochrane dataset.

DOI10.3163/1536-5050.97.3.007
Alternate JournalJ Med Libr Assoc