Objectives: To compare reporting and methodological quality, pooled treatment effects and evidence of publication bias of meta-analyses from the Cochrane Database of Systematic Reviews with meta-analyses published in leading general and specialist English-language journals.
Methods: Handsearch of years 1993-1997 of four general medicine journals (Ann Intern Med, BMJ, Lancet, JAMA,) and four specialist journals (Am J Cardiol, Cancer, Circulation, Obstet Gynecot) for reports of meta-analyses that reported individual trial results on binary endpoints for at least 5 controlled trials. For each journal meta-analysis we randomly selected a Cnchrane meta-analysis, matching for the number of trials combined in the main analysis. We used the same standardised instrument which covered the search strategy, quality assessment, data extraction and analysis, and the presentation and interpretation of findings for Cochrane and journal meta-analyses. We used a regression test for publication bias (Egger et al, 1997) to assess the likely presence or absence of bias associated with the conduct and dissemination of small trials (publication bias and related biases).
Results: We identified 135 journal meta-analyses, 78 of which reported results on binary endpoints from individual trials. The distribution of the number of trials included in the main analysis is shown in the figure.
Discussion: Preliminary results indicate that both reporting and methodological quality are higher among Cochrane meta-analyses, that evidence of bias is more prevalent among journal meta-analyses and that the latter are more likely to report statistically significant, beneficial effects of interventions. The work is in progress; final results and conclusions will be presented at the Colloquium.