A simple hint to improve Robinson and Dickersin's highly sensitive PubMed search strategy for controlled clinical trials

Abstract

Systematic literature searches are of paramount importance for investigators involved in clinical research as well as for those overviewing clinical evidence.1 While several electronic databases are available to search for controlled clinical studies, PubMed, the online version of the US National Library of Medicine MEDLINE database, is probably one of the most complete and user-friendly.2 Nonetheless, significant expertise is needed to tackle the vast amount of information available in PubMed. Among other useful tools, explicit search strategies using Boolean operators, Medical Subject Heading (MeSH) terms and keywords have been proposed to improve the sensitivity and specificity of PubMed searches for clinical trials.3 One of the most sensitive as well as specific of such systematic search algorithms is the strategy originally proposed by Robinson and Dickersin in 2002, which has been adopted by several Cochrane Collaboration reviewers.4 However, even this well-tested strategy may lack optimal specificity, as many initially retrieved citations, often in the order of hundreds, eventually end up being non-pertinent or duplicate hits. Unfortunately, it is well known that the number of initial reports is one of the key factors determining length, complexity, and economic burden of a systematic review. Moreover, readers and reviewers of biomedical literature always hope to reduce the number-needed-to-read in order to pursue their scholarly projects or continuing medical education.5 Our aim was to devise a few simple ameliorations to the Robinson and Dickersin’s highly sensitive PubMed search strategy for controlled trials. We thus simply excluded noncontrolled and non-experimental studies, such as reviews, commentaries, practice guidelines, meta-analyses and editorials from the potential citations of interest, by using the ‘NOT’ Boolean operator. Letters or journal correspondence were instead considered as potential sources of randomized trial data and thus not explicitly excluded.4 Actually, we simply added to Robinson and Dickersin’s string the following: ‘NOT (comment[pt] OR editorial[pt] OR meta-analysis[pt] OR practice-guideline[pt] OR review[pt])’. The modified search strategy is thus the following (additions are in italics): (randomized controlled trial [pt] OR controlled clinical trial [pt] OR randomized controlled trials [mh] OR random allocation [mh] OR double-blind method [mh] OR single-blind method [mh] OR clinical trial [pt] OR clinical trials [mh] OR (‘clinical trial’ [tw] OR ((singl* [tw] OR doubl* [tw] OR trebl* [tw] OR tripl* [tw]) AND (mask* [tw] OR blind [tw])) OR (‘latin square’ [tw]) OR placebos [mh] OR placebo* [tw] OR random* [tw] OR research design [mh:noexp] OR comparative study [mh] OR evaluation studies [mh] OR follow-up studies [mh] OR prospective studies [mh] OR cross-over studies [mh] OR control* [tw] OR prospectiv* [tw] OR volunteer* [tw]) NOT (animal [mh] NOT human [mh]) NOT (comment[pt] OR editorial[pt] OR meta-analysis[pt] OR practice-guideline[pt] OR review[pt])). In order to test the performance of this simple modification of the Robinson and Dickersin’s algorithm, we compared the standard Robinson and Dickersin’s PubMed search strategy versus the modified algorithm hereby proposed in a pool of five systematic reviews already completed by our group.6–10 The median number (25–75% range) of initial citations was significantly lower using the modified search in comparison to Robinson and Dickersin’s standard PubMed algorithm, respectively 354 (55–1653) versus 393 (87–2041) (P = 0.043 at non-parametric Wilcoxon test). This translates into a median 19% (14–37) reduction in the number of initial citations. We also tested whether any pertinent trial, finally included in the meta-analyses published by our group, and originally retrieved by the standard Robinson and Dickersin’s search strategy, was missed by our more stringent modified algorithm. Actually, no pertinent study was excluded, implying unimpaired sensitivity. In conclusion, we believe that a minor modification to the well-established Robinson and Dickersin’s PubMed search strategy for clinical trials may significantly increase search specificity without any reduction in sensitivity.