Should unpublished data be included in meta-analyses? Current convictions and controversies

To identify the extent to which meta-analyses currently include unpublished data and whether editors, meta-analysts, and methodologists believe unpublished material should be included.

This article describes two related studies: a literature review and a cross-sectional survey. SAMPLE SELECTION: For the literature review, we identified all articles indexed by the key word meta-analysis from January 1989 to February 1991 and determined whether unpublished material had been searched for, obtained, and included in the meta-analyses. For the cross-sectional survey, we surveyed authors of these meta-analyses, authors of articles addressing methodological issues in meta-analysis published during the same period, and editors of journals in which both types of articles were published.

We asked the respondents about their attitudes concerning inclusion of unpublished data and publication of articles from which data had previously been included in a scientific overview.

Inclusion of unpublished data and opinions about whether unpublished material should be included in overviews and whether prior inclusion of data in an overview should bear on publication.

Of 150 meta-analyses, 46 (30.7%) included unpublished data in their primary analysis. Of authors surveyed, 85% responded. Of the meta-analysts and methodologists, 77.7% felt that unpublished material should definitely or probably be included in scientific overviews; this was true of 46.9% of the editors. A total of 86.4% of the meta-analysts and methodologists and 53.2% of the editors felt that inclusion of data in a prior overview should have no bearing on full publication of original research.

While inclusion of unpublished data in scientific overviews remains controversial, most investigators directly involved in meta-analysis believe that unpublished data should not be systematically excluded. The most valid synthesis of available information will result when meta-analysts subject published and unpublished material to the same rigorous methodological evaluation and present results with and without unpublished sources of data.