Identifying controlled studies of alcohol-impaired driving prevention: designing an effective search strategy

Abstract

Database search strategies for systematic reviews are often developed subjectively and may miss relevant studies. We examined whether objective methods, originally developed to identify systematic reviews, could be applied to a multidisciplinary topic, alcohol-impaired driving (AID). We performed word-frequency analyses of titles/abstracts/keywords on citations randomly selected from a ‘gold standard’ database of 131 controlled studies evaluating AID interventions. Identified words were tested in the Web of Science databases for sensitivity, precision, and feasibility. Words meeting varying sensitivity and precision thresholds were combined into search strategies. All threshold combinations with cumulative sensitivity of at least 75% retrieved ≥ 100,000 citations. The word frequency-based procedure was therefore modified, first by eliminating non-specific words, then by categorizing and combining the remaining words. With these modifications, sensitivity was 80.0% and precision 4.4%, retrieving 17,565 citations. Against a validation set, sensitivity was 85.7% and precision 3.5%. To identify studies on multidisciplinary topics, objective methods may be useful initially but may need to be modified to meet sensitivity and feasibility targets.