Attribution of multiple literature databases in systematic reviews for public health guideline development




Poster session 4


Tuesday 25 October 2016 - 15:30 to 16:00


All authors in correct order:

Kojimahara N1, Morizane T2, Shigekawa S1, Kawai F3, Sayama S3, Kato S1, Yamaguchi N4
1 Tokyo Women's Medical University, Japan
2 MINDS Guideline Center, Japan
3 St. Luke's International University, Japan
4 The Japan Medical Library Association, Japan
Presenting author and contact person

Presenting author:

Noriko Kojimahara

Contact person:

Abstract text
Background: Since 'Minds Handbook for Clinical Practice Guidelines Development 2014' was published, clinical practice guidelines (CPG) based on systematic reviews for clinical questions that necessitate comprehensive literature searches have become common in Japan. However, it is unclear which and/or how many literature databases should be used or to what extent refinement of search queries according to retrieval results is satisfactory for systematic reviews for public health guidelines.

Objectives: NICE guidance (PH19, 2009) examining interventions for long-term sickness and incapacity for work was used as a reference to investigate the performance of bibliographic databases in identifying the included studies,and the most effective combination of databases required to retrieve all included studies.

Methods: Authors searched the yield of included studies from 3 databases; MEDLINE, PubMed, and EMBASE and calculated the precision of each search strategy. We investigated differences between the presence of a record in a database and its retrieval and number needed to read (NNR). We applied a filter to pick up only randomized controlled trials.

Results: Thirty-two out of 45 included studies were present in MEDLINE, 32 in PubMed, and 29 in EMBASE. Combinations of PubMed and EMBASE identified 36 studies, most effectively. Only 12/45 studies had articles whose full texts were available free. NNR for MEDLINE was the lowest at 74.2 (2373/32), but 300.5 (9616/32) for PubMed. NNT for EMBASE after removing studies included in MEDLINE too, called 'only EMBASE', was 54.2 (275/5). Six studies were not found using our retrieval system, although all of them were observational studies.

Conclusions: Systematic reviews could produce biased conclusions if a search to identify eligible studies is not comprehensive. Compared to 80.3 (2331/29) for a simple EMBASE search, a combination of MEDLINE and 'only EMBASE' seemed to be more effective. In a systematic review of a range of interventions that were topics of one of NICE guideline regarding workplace health, at least two databases and reference checking were required to retrieve all included studies.