Concordance of reporting of primary outcomes in abstracts and full text of randomized clinical trials: Is there an outcome reporting bias?

Article type
Authors
Ribic C1, Lamontagne F2, Johnston B3, Briel M4, You J1, Sun X5, Busse J6, Bassler D7, Vera C8, Al-Shurafa M1, Mills E9, Guyatt G1, Cook D1, Akl E10
1McMaster University, Canada
2University of Sherbrooke, Canada
3University of Alberta, Canada
4University Hospital Basel, Basel
5Sichuan University, China
6The Institute for Work and Health, Canada
7University Children’s Hospital, Germany
8Pontificia Universidad Catolica de Chile, Chile
9Centre for International Health and Human Rights, Canada
10State University of New York at Buffalo, USA
Abstract
Background: Reporting the results of pre-specified primary outcomes in the full text of randomized clinical trials (RCTs) but not in the associated abstract, or reporting primary outcomes in the abstract that are not reported in the full text, represents a potential source of publication bias.

Objectives: To assess the concordance in reporting primary outcomes between abstracts and full text of RCTs published in top medical journals.

Methods: We identified all intervention RCTs published from 2005 to 2007 in the NEJM, JAMA, Lancet, BMJ, or Annals of Internal Medicine. Ineligible trials were cross over and cluster designs. Teams of two reviewers appraised the abstract then, separately, the full text of each selected article using pretested, standardized forms. Abstractors were blinded to outcomes reported in full text when screening abstracts. We calculated the concordance of primary outcome reporting between abstracts and full texts.

Results: Of 709 eligible RCTs, abstracts reported one primary outcome (55.1%, n = 391), multiple primary outcomes (25.6%, n = 182) or no primary outcome (19.2%, n = 136). Of abstracts reporting one primary outcome, 97.4% (n = 381) had concordant reporting in the full text. We found 2.6% (n = 10) of RCTs had discordant reporting of single primary outcomes in abstracts compared to full text. Of these, 6trials reported a single primary outcome in the abstract and pre-specified multiple primary outcomes in the full text, while 4 trials reported a single primary outcome in the abstract and did not specify a primary outcome in the full text.

Conclusions: In this representative heterogeneous sample of RCTs published in high-impact journals, there is a high concordance between single primary outcomes reported in abstracts and those specified in the full text. However, approximately one-fifth of the abstracts did not specify a primary outcome, potentially introducing bias in primary outcome reporting.