Background: Searching for existing systematic reviews is a key step in the development of most guidelines, but retrieving all relevant reviews involves substantial work. In the context of the venous thromboembolism (VTE) guidelines of the American Society of Hematology we compared two different approaches.

Objectives: To compare a traditional search approach to an alternative approach using Epistemonikos database (www.epistemonikos.org) to identify systematic reviews relevant for a guideline.

Methods: We selected all of the questions (n=27) from one guideline (VTE treatment). The traditional approach included searches in MEDLINE, EMBASE and Cochrane, using a filter adapted from several sources (SIGN filters, KSA guidelines, NICE 144) plus filters to identify records relevant for VTE. The same terms for VTE were adapted to Epistemonikos syntaxis for the alternative approach. For both approaches, at least two researchers screened records. In the traditional approach potentially eligible articles were evaluated in full text for inclusion. In the alternative approach one review per question (index review) was selected, and reviews that shared at least one included study were evaluated for inclusion. The final reference selection was done by the chapter methodologist for both approaches. We calculated recall (included reviews/reviews identified by any approach) and search efficiency (included reviews/ initial number of records).

Results: Traditional approach returned 7678 citations, 159 full texts were retrieved, and 38 reviews were finally included. Epistemonikos approach returned 4434 citations, from which 406 were pre-selected (20 index reviews selected), 153 full texts were retrieved, and 94 were finally included.
Recall was 39.6% (38/96) for traditional approach vs. 97.9% for alternative. Search efficiency was 0.0049 (38/7678) for traditional approach and 0.0211 for alternative (94/4434).

Conclusions: An alternative approach using Epistemonikos database was more sensitive and efficient than the traditional approach. These conclusions have important implications for improving efficiency and feasibility of guidelines, but need further validation.