Association between trial registration and positive study findings: cross sectional study (Epidemiological Study of Randomized Trials—ESORT)BMJ 2017; 356 doi: https://doi.org/10.1136/bmj.j917 (Published 14 March 2017) Cite this as: BMJ 2017;356:j917
All rapid responses
We read with great interest the cross sectional study of Odutayo et al. analysing whether trial registration is associated with positive study findings (1). They included all PubMed-indexed randomised controlled trials published in December 2012, providing indexation by November 2013. Among all eligible studies, they found 593 registered and 529 non-registered trials. Although registered trials had a slight trend towards positive study findings, no clear effect was identified, nor was there conclusive evidence with regard to the effect of funding sources. Due to the inclusion of a large sample and careful analyses, this study marks an important contribution.
However, a relevant risk of bias remains unaccounted for. As the authors themselves point out, unpublished studies were not included in the data analysis, because PubMed obviously does not index unpublished findings. Since the most important driver for non-publication of negative research results – i.e. publication bias – was not addressed, the finding of no difference between registered and non-registered trials is hardly surprising. Due to publication bias, positive results will more likely make it into indexed journals no matter whether they are registered or not. The finding of a (small) association between registration and positive study findings may be related to more thorough study planning and preparation of registered trials, especially multicentre trials, typically involving a team of statisticians (and substantial funding). Moreover, multicentre trials often analyse several outcomes, and Odutayo et al. defined a positive study finding if at least one primary outcome was significant (1).
In order to address the very relevant question if a real difference in publication of positive study findings between registered and non-registered trials exists, one would certainly need access to all unpublished data, which is impossible, thus the study aim cannot be addressed. To clarify, we absolutely agree with the authors that trials should be registered and we support the attitude that journals should make trial registration mandatory for article submission. In addition, we underline that all results should be made accessible after trial registration, e.g. if the results do not make it into a journal within a certain period after registration (e.g. 5 years for a single-centre, randomised controlled trial), they should be published in an online database of the trial registry. If trials are registered prior to data collection and their results are published after a certain time interval, this would ensure that the data are accessible to the scientific community. More importantly, editors and reviewers should encourage publication of negative study findings, providing satisfactory quality of trials.
Odutayo A, Emdin CA, Hsiao AJ, et al. Association between trial registration and positive study findings: cross sectional study (Epidemiological Study of Randomized Trials—ESORT). BMJ 2017;356:j917.
Competing interests: No competing interests