The cochrane oral health group trials register: electronic searching and beyond.

and indexing terms assigned by professional indexers (in MEDLINE these are Medical Subject Headings or MeSH). Essentially, for the researcher, electronic searches chiefly rely on two things: • the controlled vocabulary (in MEDLINE MeSH) terms assigned to the article by professional indexers; and • descriptors (text words) used by author/s in the title and abstract. Where the detail of the study design is “lost” in the main body of the paper, this information will not be available in the electronic record and, further, is more likely to be missed by the indexers. Lack of detail in the title and abstract of the paper will influence the results of a search, and as by no means all articles have abstracts, this limits the search potential even further. Increasingly, journals are insisting on structured abstracts, where the author systematically discloses the methodology of the study in the abstract. This is one of the recommendations of the CONSORT guidelines and should improve the yield and quality of electronic searches in the future. But for those records with no abstracts and already indexed in MEDLINE, for example, handsearching the relevant journal may be the only sure way of identifying them. Another source of trials that slip through the electronic search net are conference abstracts that are published in journals but are not indexed in the main bibliographic databases such as MEDLINE. Some of the abstracts will, in due course, appear as fully published papers. It has been estimated that between approximately 30 and 60 percent of abstracts presented at scientific meetings eventually get published as full journal articles, depending upon the clinical specialty. The percentage of oral health conference abstracts reaching full publication is as yet unclear. Unpublished trials are an essential component of systematic reviewing if the effect of publication bias (defined as “the tendency on the parts of investigators, reviewers, and editors to submit or accept manuscripts for publication based on the direction or strength of the study findings”) is to be minimized. Conference abstracts provide an important tracking system to identify both published and unpublished trials that are paramount to unbiased reviews. Several electronic resources are now available to search for conference abstracts. Two of these are the British Library’s Inside Web service (www.bl.uk/services/current/inside.html) and the CSA Internet Database Service, which is located in Bethesda, Maryland, with offices in Hong Kong, France, and the UK (www.csa1.co.uk). Another limitation of electronic searching is that not all journals are indexed in MEDLINE and, in particular, non-English language references are underrepresented. There is good evidence that research findings showing statistically significant results are more likely to be published in English language journals. By excluding non-English language articles, the results of a systematic review are again susceptible to publication biases. The study by Suarez-Almazor et al. to compare the performance of MEDLINE and EMBASE in identifying English language controlled clinical trials found that even using these two databases, with dedicated compreTable 1. Identification of clinical trials through handsearching and MEDLINE Method of identification Total Percent Handsearching and Medline 105 34.5 Handsearching only 161 53.0 Medline only 38 12.5 Total 304 100.0 Table 2. Studies identified as clinical trials by MEDLINE alone Identification by MEDLINE only Total Percent Confirmed as clinical trials 6 15.8 False positives (not clinical trials) 32 84.2 Total 38 100.0