Retrieval of Overviews of Systematic Reviews in MEDLINE Was Improved by the Development of an Objectively Derived and Validated Search Strategy

J Clin Epidemiol. 2016 Jun;74:107-18. doi: 10.1016/j.jclinepi.2015.12.002. Epub 2015 Dec 23.

Abstract

Objectives: Locating overviews of systematic reviews is difficult because of an absence of appropriate indexing terms and inconsistent terminology used to describe overviews. Our objective was to develop a validated search strategy to retrieve overviews in MEDLINE.

Study design and setting: We derived a test set of overviews from the references of two method articles on overviews. Two population sets were used to identify discriminating terms, that is, terms that appear frequently in the test set but infrequently in two population sets of references found in MEDLINE. We used text mining to conduct a frequency analysis of terms appearing in the titles and abstracts. Candidate terms were combined and tested in MEDLINE in various permutations, and the performance of strategies measured using sensitivity and precision.

Results: Two search strategies were developed: a sensitivity-maximizing strategy, achieving 93% sensitivity (95% confidence interval [CI]: 87, 96) and 7% precision (95% CI: 6, 8), and a sensitivity-and-precision-maximizing strategy, achieving 66% sensitivity (95% CI: 58, 74) and 21% precision (95% CI: 17, 25).

Conclusion: The developed search strategies enable users to more efficiently identify overviews of reviews compared to current strategies. Consistent language in describing overviews would aid in their identification, as would a specific MEDLINE Publication Type.

Keywords: MEDLINE; Overviews of systematic reviews; Search filter; Search strategy design; Sensitivity; Text mining.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Data Mining
  • Evidence-Based Medicine
  • Humans
  • Information Storage and Retrieval / methods*
  • Information Storage and Retrieval / statistics & numerical data*
  • MEDLINE / statistics & numerical data*
  • Reproducibility of Results
  • Review Literature as Topic*
  • Sensitivity and Specificity