Article type
Year
Abstract
Background: When reviewers search MEDLINE they must ensure that all high quality original articles on a particular topic are retrieved as failure to identify all possibly relevant reports could result in bias. When searching MEDLINE unaided key articles are often missed and many articles that are retrieved are irrelevant to the search. Search strategies that accurately retrieve citations of high quality therapy and diagnosis studies from MEDLINE will enhance the searching efforts of Cochrane reviewers.
Objective: To develop optimal search strategies in MEDLINE for retrieving sound clinical studies of the treatment and diagnosis of health disorders.
Methods: The study design was modeled on that of diagnostic test studies, with search terms as "tests" and hand searches of journals using explicit criteria for studies of treatments and diagnostic procedures as the "gold standard." 6 research assistants who had been intensively calibrated reviewed all issues of 161 journals for the publishing year 2000. Original and review articles of treatment and diagnosis were categorized as "pass" or "fail" for methodologic rigor. "Pass" rigor criteria for treatment articles were random allocation of participants to comparison groups and at least 80% follow up. "Pass" rigor criteria for diagnosis articles were inclusion of a spectrum of participants; objective diagnostic standard or current clinical standard; participants received both the new test and the diagnostic standard; and interpretation of diagnostic standard without knowledge of test result and vice versa. The sensitivity, specificity, precision, and accuracy of 4,862 unique search terms in 18,404 combinations ("tests") for treatment and in 17,287 combinations for diagnosis were determined by comparison with a manual review of all articles (the "gold standard") in 161 journals published during the year 2000 (49,028 articles).
Results: 1587 (24.2%) of 6568 treatment articles and 147 (18.9%) of 778 diagnosis articles met basic criteria for scientific merit. For the detection of high quality original studies combinations of search terms reached peak sensitivities of 99.3% at a specificity of 70.4% for treatment studies, and peak sensitivities of 98.6% at a specificity of 74.3% for diagnosis studies. The search strategies are presented in the table and generally outperform other validated search strategies. Three MEDLINE services (PubMed, Ovid, and SKOLAR) have incorporated these high sensitivity strategies.
Conclusions: New empirical search strategies have been developed to optimize retrieval of articles from MEDLINE reporting high-quality clinical studies of treatment and diagnosis.
Acknowledgements: Funded by the U.S. National Library of Medicine. Researchers involved in the study were Angela Eady, Brian Haynes, Susan Marks, Ann McKibbon, Doug Morgan, Cindy Walker-Dilks, Stephen Walter, Stephen Werre, Nancy Wilczynski, and Sharon Wong.
Objective: To develop optimal search strategies in MEDLINE for retrieving sound clinical studies of the treatment and diagnosis of health disorders.
Methods: The study design was modeled on that of diagnostic test studies, with search terms as "tests" and hand searches of journals using explicit criteria for studies of treatments and diagnostic procedures as the "gold standard." 6 research assistants who had been intensively calibrated reviewed all issues of 161 journals for the publishing year 2000. Original and review articles of treatment and diagnosis were categorized as "pass" or "fail" for methodologic rigor. "Pass" rigor criteria for treatment articles were random allocation of participants to comparison groups and at least 80% follow up. "Pass" rigor criteria for diagnosis articles were inclusion of a spectrum of participants; objective diagnostic standard or current clinical standard; participants received both the new test and the diagnostic standard; and interpretation of diagnostic standard without knowledge of test result and vice versa. The sensitivity, specificity, precision, and accuracy of 4,862 unique search terms in 18,404 combinations ("tests") for treatment and in 17,287 combinations for diagnosis were determined by comparison with a manual review of all articles (the "gold standard") in 161 journals published during the year 2000 (49,028 articles).
Results: 1587 (24.2%) of 6568 treatment articles and 147 (18.9%) of 778 diagnosis articles met basic criteria for scientific merit. For the detection of high quality original studies combinations of search terms reached peak sensitivities of 99.3% at a specificity of 70.4% for treatment studies, and peak sensitivities of 98.6% at a specificity of 74.3% for diagnosis studies. The search strategies are presented in the table and generally outperform other validated search strategies. Three MEDLINE services (PubMed, Ovid, and SKOLAR) have incorporated these high sensitivity strategies.
Conclusions: New empirical search strategies have been developed to optimize retrieval of articles from MEDLINE reporting high-quality clinical studies of treatment and diagnosis.
Acknowledgements: Funded by the U.S. National Library of Medicine. Researchers involved in the study were Angela Eady, Brian Haynes, Susan Marks, Ann McKibbon, Doug Morgan, Cindy Walker-Dilks, Stephen Walter, Stephen Werre, Nancy Wilczynski, and Sharon Wong.