Assessment of publication bias based on clinical trial registrations

Tags: Poster
Strandell A1, Bergh C1, Bernhardsson S1, Jivegård L1, Samuelsson O1, Sjögren P1, Sjövall H1, Svanberg T1
1HTA-centrum, Sahlgrenska University Hospital

Background: In the assessment of publication bias, a funnel plot is often used if at least 10 trials are available. If fewer studies are available, assessment of publication bias is more difficult. In the systematic reviews (SR) conducted at the HTA (Health Technology Assessment)-centrum in Region Västra Götaland, Sweden, we have searched for ongoing trials in the Clinical Trials database and reported relevant trials. This information was not used for assessment of publication bias during the study period.

Objectives: To study if www.clinicaltrials.gov is a useful tool for assessing publication bias in SRs.

Methods: We retrospectively reviewed all SRs conducted at the HTA-centrum during 2009 to 2016. Publication bias was reassessed, including the information found in Clinical Trials at the time when the SR was conducted. If the recruitment of patients in a registered study had been completed at the time of the search in www.clinicaltrials.gov, the publication status was investigated. The NCT number and/or the name of the principal investigator and the key words were used in PubMed, Embase, the Cochrane Library and Google Scholar to find publications within 2 years after the completion of the study. In each case it was evaluated whether publication bias was present and would have been a reason for downgrading the certainty of evidence.

Results: During the study period, 68 SRs were published and 2 of those had noted publication bias as a contributing reason for downgrading the certainty of evidence in any of the outcomes. Sixty-five SRs included a search in www.clinicaltrials.gov and in 48 of those there were relevant registered trials for the question at issue (median number of studies 4, range 1-98). In 24 SRs it was noted that at least 1 trial had completed recruitment (median 1, range 1-16), and 8 studies in 6 SRs had not been published 2 years later. In 3 cases publication bias was considered as being present and would have caused downgrading one level of the certainty of evidence.

Conclusions: The use of Clinical Trials may contribute to the overall assessment of publication bias and may influence the certainty of evidence.