Contribution of Stroke to the Cochrane Stroke Group Trials Register.

Contribution of Stroke to the Cochrane Stroke Group Trials Register To the Editor: The formulation of evidence-based stroke care practice guidelines is guided by the results of randomized trials. We sought to determine what proportion of all trial reports are published in Stroke, and what proportion might be missed by electronic search methods. One of us (B.T.) performed a page-by-page search of every issue of Stroke (1970–2001), applying standard criteria1 to identify all reports of relevant randomized trials and controlled clinical trials appearing in articles, editorials, letters, conference abstracts, or any other text. Reports of trials of interventions for the treatment, rehabilitation, and secondary prevention of stroke are included in the Cochrane Stroke Group Trials Register. Reports of trials outside the scope of the Group are forwarded for inclusion in the Cochrane Central Register of Controlled Trials in the Cochrane Library. We developed a detailed search strategy for MEDLINE, which includes controlled vocabulary and free text terms, and extends to 56 lines of search statements. The strategy is published in the Stroke Group’s module in the Cochrane Library2 and is designed to be highly sensitive and optimally specific to detect reports of randomized trials and controlled clinical trials. We compared the lists of reports found by handsearching (the gold standard search) with the list identified by the MEDLINE search of Stroke for publication years 1970 to 2001. Stroke is the most important single source of relevant trial reports in the Stroke Group Trials Register, accounting for 803/5771 (14%) of all trial reports in our Register to December 2001. The complete retrospective handsearch identified 803 relevant stroke trial reports including 365 original articles, 382 conference abstracts (and other abstracts), and mentions of trials in letters in 56. We identified a further 67 reports of trials in patients with nonstroke conditions. The Abstracts of Literature section of the journal (now discontinued) was a useful source of an additional 331 reports of stroke trials published in other journals. The Figure shows the increasing numbers of the different categories of stroke trial reports published in the journal since 1970. Three major Stroke conferences published their proceedings in the journal in 2000, which accounts for the large number of abstracts found in that year. Since 1970, Stroke has included proceedings from 44 conferences. The search of MEDLINE identified 1605 references published in Stroke from 1970 to 2001. A comparison with our gold standard handsearch showed that the MEDLINE search identified 353 of the 803 found by handsearching (sensitivity 44%). The remaining 1252 were not reports of relevant trials (precision 22%). Of the 450 trial reports in Stroke not identified by the MEDLINE search, 376 were abstracts, 42 were original articles, and 32 were letters. Stroke is an increasingly rich source of reports of stroke trials, and the journal accounts for a significant proportion of all trial reports in our Register. The search also allowed us to assess the sensitivity of our electronic search strategy. Although a significant number of the reports of trials in Stroke were not identified by the MEDLINE search, 91% of the missed reports were either letters or abstracts from sections of the journal that are not indexed in MEDLINE. Our aim is to identify all reports of relevant trials from the planning stage to full publication and, whilst a search of MEDLINE will identify major trial results, such indirect reports may be the only clue to the existence of an unpublished or ongoing major trial. Many of these trials may, of course, be subsequently published in full elsewhere, and we are planning to investigate and characterize those trial reports, published as abstracts or letters, that never come to full publication. Equally, it is well known that many trials (especially “negative” ones) may never be published as full papers. We have been able to demonstrate, using Stroke as an example, that handsearching important specialist journals is an essential component in our search for trials for systematic reviews, and limiting the search to MEDLINE may result in exclusion of potentially relevant trials. The Stroke Group Trials