Non-publication and publication bias in reproductive medicine: a cohort analysis.

STUDY QUESTION

Does publication bias or non-publication exist in fertility trials presented as conference abstracts?

SUMMARY ANSWER

This study did not detect any publication bias; however, it did identify a high level of non-publication, with only 49% of abstracts reaching full-text publication four or more years after abstract presentation.

WHAT IS KNOWN ALREADY

Systematic reviews of randomized controlled trials (RCTs) are the foundation of evidence based medicine. Non-publication or publication deficit refer to the failure to publish trial results. A publication bias exists when there is any tendency on the parts of the investigators or editors to fail to publish study results on the basis or strength of the study findings. Both present a serious problem for researchers, clinicians and policymakers alike, and ultimately impact on patient care.

STUDY DESIGN, SIZE, DURATION: A retrospective cohort study identified 337 fertility RCTs presented as conference abstracts between 2007 and 2010, as captured by an electronic search of the Cochrane Gynaecology and Fertility Database. After excluding ineligible trials and duplicates, 224 abstracts remained.

PARTICIPANTS/MATERIALS, SETTING, METHODS: A search for the full-text papers of each abstract was undertaken in Pubmed, MEDLINE, Embase, CINAHL and Google in May 2015 using a probabilistic approach. Trial authors were contacted to query the publication status of abstracts when no full-text was identified. The association between individual variables and the probability of publication, and time to publication, was assessed using logistic regression and Cox regression, respectively.

MAIN RESULTS AND THE ROLE OF CHANCE

Of the 224 included abstracts, only 110 (49%; 95% CI: 42.6, 55.6) were found to be published as full-text articles. Publication bias was not identified in this cohort; studies with positive results had a similar probability of reaching full-text publication 52/113 (46%; 95% CI: 37.0, 55.3) as studies with non-positive (negative or null) results 58/111 (52%; 95% CI: 17.8, 33.9) (adjusted odds ratio (AOR): 1.02; 95% CI: 0.53, 1.97). Similarly, the time from abstract presentation to full-text publication was similar in studies with positive and non-positive results. Oral presentations were more likely to be published, and to be published sooner, than poster presentations (poster presentation AOR: 0.31; 95% CI: 0.15, 0.61 and adjusted hazard ratio (AHR): 0.57; 95% CI: 0.38, 0.86). Studies that were not registered were less likely to be published and to have delayed publication, than studies which were registered either prospectively or retrospectively (AOR: 0.14; 95% CI: 0.04, 0.44 and AHR: 0.43; 95% CI: 0.25, 0.72). Abstracts which were presented a longer time ago also had a higher probability of reaching full-text publication (P  = 0.01).

LIMITATIONS, REASONS FOR CAUTION: Commencing with a cohort of RCTs from ethics committee registers may provide a better picture of the extent of non-publication and publication bias, as not all trials reach the stage of abstract presentation. It is also possible that the search did not identify all published trials, as some may have been published after the follow-up period.

WIDER IMPLICATIONS OF THE FINDINGS

This study did not identify any publication bias. However, only half of the abstracts in this cohort have been published as full-text articles, four or more years after their presentation at a conference. This is similar to publication rates reported previously for fertility trials, and is despite increasing awareness of the importance of publishing trial results, and subsequent requirements for all RCTs to be registered prior to trial initiation. A better understanding of the reasons for non-publication should assist in facilitating the prompt full-text publication of RCTs in the future.

STUDY FUNDING/COMPETING INTEREST(S): Funding provided from the University of Auckland. All authors declare they have no conflicts of interest in relation to this article.

TRIAL REGISTRATION NUMBER

Not applicable.