Background: It is generally acknowledged that searching for studies to populate a systematic review should be as comprehensive as resources allow. Where choices have to be made between available databases, little evidence-based guidance exists about how to prioritize resources to search or when to stop searching.

Objectives: This study aimed: (i) to compare the individual contribution of databases searched to identify studies meeting the inclusion criteria for a recently completed systematic review; (ii) to identify the best combination of databases to retrieve all relevant studies.

Methods: A systematic review of the non-pharmacological management of frozen shoulder was used as a case study. The total number of studies included in the review provided the reference standard. A spreadsheet was used to record for each included study a) whether it was retrieved by the appropriate search strategy and b) whether it was indexed on each database. Yields and number needed to read (NNR) from different combinations of databases were investigated.

Results: The review included 27 randomized controlled trials (RCTs), one controlled clinical trial (CCT) and two case series. Of the 28 controlled trials, 27 were indexed in at least one of sixteen analysed databases. Recall for individual databases ranged from 0% to 90% (median 40%), and precision from 0.0% to 4.3% (median 1.2%). For the strategies used, the best combination of resources to retrieve all studies was either CENTRAL, EMBASE, Science Citation Index, and reference checking; or CENTRAL, EMBASE, MEDLINE (including PREMEDLINE), and reference checking.

Conclusions: Of the databases searched, CENTRAL displayed the best combination of recall and precision. In terms of combinations of databases a minimum of three databases, including CENTRAL, plus reference checking were required to retrieve all included studies.