Optimized search strategy for detecting scientifically strong studies on treatment through PubMed

Abstract

Our study was designed to optimize the search strategies based on the work of Haynes et al. for detecting randomized controlled trials (RCTs) through PubMed. In particular, we aimed to improve precision for broad and narrow searches on interventional studies. We used in addition to the string suggested by the Hedge Team the following: {NOT ((animals [mh] NOT humans [mh]) OR (review [pt] OR meta-analysis [pt]))} and tested its effectiveness. The search was carried out on a year's worth of articles from the PubMed database. We analyzed 35,590 bibliographic citations about four relevant major topics in internal medicine (hypertension, diabetes, heart failure, and hepatitis). Precision, percentage gain between the Hedge Team search strategies and the new one were computed and reported in the text. Moreover, a pooled analysis was carried out in terms of absolute precision difference. We observed better precision for both broad and narrow searches. However, effective gain resulted only for broad searches. In this case, bibliographic citation recall effectively reduced (-24 to -35 % retrieved citation with a gain of 32-54 %) without loss of information. The search strategy improved broad searches regarding each of the four considered topics. We think this new search strategy, based on a previous work of the Hedge team, could be a step forward and can save some time by researchers.