Background: Systematic reviews (SRs) are increasingly viewed as useful decision-making tools yet the extent of SR publication bias is under-explored. Through my thesis, I aimed to investigate the extent of SR publication bias.
Methods: A conceptual model was derived from literature searches and one-on-one interviews and three studies were conducted: a cross-sectional study of 296 SRs indexed in MEDLINE and published in November 2004, an international survey of 625 corresponding or first authors of a published SR in 2005, and a retrospective cohort study of 411 Cochrane protocols from Issues 2-4, 2000 and Issue 1,2001 that were followed until Issue 1,2008 in The Cochrane Library.
Main findings: The interviewees reported 40 unpublished SRs and the conceptual model showed that publication bias can permeate all steps of the publication process, from conceptualization to ultimate effect on health outcomes. The cross-sectional study identified favourable results in 57.7% of Cochrane reviews and 64.3% of non-Cochrane reviews with a meta-analysis of the primary outcome and non-Cochrane reviews were twice as likely to have positive conclusions as Cochrane reviews (p-value≤0.05). In the international survey, participants reported 1405 published (median: 2.0, range: 1-150) and 199 unpublished (median: 2.0, range: 1-53) SRs. In the retrospective cohort study, 19.1% (71/372) of eligible Cochrane protocols remained unpublished and the median time to publication was 2.4 years (range: 0.15-8.96). A shorter time to publication was associated with the Cochrane review being subsequently updated versus not updated (n=100/372 Cochrane reviews that were updated, hazard ratio: 1.80 [95% confidence interval: 1.39-2.33 years]) and a longer time to publication was associated with the Cochrane review having two published versus one protocol (n=10/372 Cochrane reviews with two published protocols, 0.33 [0.12-0.90 years]).
General conclusions: Over 300 unpublished SRs were identified through the interviews conducted for the conceptual model and the three studies that comprised my thesis. Possible solutions for minimizing or avoiding SR publication bias include registration of SRs at inception, educating the research community about the importance of publishing SRs, and having a general online open-access journal with rapid peer review that is dedicated to only publishing the results of SRs (including their updates).
Identifer | oai:union.ndltd.org:uottawa.ca/oai:ruor.uottawa.ca:10393/29957 |
Date | January 2009 |
Creators | Tricco, Andrea C |
Publisher | University of Ottawa (Canada) |
Source Sets | Université d’Ottawa |
Language | English |
Detected Language | English |
Type | Thesis |
Format | 60 p. |
Page generated in 0.0229 seconds