Background: A comprehensive literature search underpins any systematic review. Most research on searching for diagnostic test accuracy studies has focused on search filters; these tend to produce unacceptably low sensitivity.
Objectives: To examine the value of searching a wide range of databases and additional sources, compared to more limited searches, in identifying diagnostic test accuracy studies for inclusion in systematic reviews.
Methods: We included eight diagnostic systematic reviews that carried out extensive, well-documented literature searches that were not limited using terms relating to diagnosis, and for which full results were available. For each of the databases listed in the table and for each review, we classified primary studies as: (1) on the database and identified by original searches, (2) on the database but not identified, or (3) not on the database. We extracted the number of hits for each review on each database and the number of studies not identified.
Results: Results are currently available for four reviews. The table summarises the results. Searching PASCAL, LILACS and the Cochrane Central Register of Controlled Trials (CENTRAL) only contributed one additional study to one of the reviews. The other four databases each contributed relevant studies that would have been missed had all databases not been searched. However, 4-35% of relevant studies were missed by the original searches on these databases. The proportions of missed studies were much higher for Science Citation Index and BIOSIS than for MEDLINE and EMBASE. Three of the four reviews included studies that were not identified from any of the databases searched.
Conclusions: Searches for diagnostic test accuracy studies should use several general medical databases. Even sensitive searches do not identify all the known relevant studies. Additional methods to identify studies, such as screening references and searching grey literature, are required.
Table: Results of searches for four reviews combined (179 studies)