Publication bias in randomised trials of duloxetine for the treatment of major depressive disorder

Article type
Authors
Maund E1, Tendal B1, Hrobjartsson A1, Jørgensen KJ1, Lundh A1, Schroll B1, Gøtzsche PC1
1The Nordic Cochrane Centre
Abstract
Background: In order to reduce publication bias, researchers have called for many years for clinical study reports submitted in licensing applications to the regulatory authorities to be made publicly available. In 2010 an important precedent was set when the European Medicines Agency (EMA) granted researchers at The Nordic Cochrane Centre access to full clinical study reports and corresponding protocols.

Objectives: To assess and compare selective reporting in full clinical study reports and published articles in a cohort of RCTs submitted to the EMA for marketing approval.

Methods: Unredacted clinical study reports and the corresponding protocols, approximately 13 500 pages, for all nine placebo controlled trials used in the regulatory submission for duloxetine for the treatment of major depressive disorder were obtained from the EMA in May 2011. Published articles, including pooled analyses, were identified through searching relevant literature databases; contacting the manufacturer; and through reference checking. Data of interest were: primary outcome (including type of analysis, and analysis population), and harms (deaths, serious adverse events, drug related serious adverse events, suicides, attempted suicides, number of discontinuations due to adverse events, and total number of adverse events). Relevant datawere extracted from protocols and clinical study reports by one set of independent observers, and from published articles by a second set of observers. These data were then compared.

Results: No major discrepancies were identified between protocols and clinical study reports. Sixty published articles were identified. There was evidence of both publication bias and selective reporting. Harms, in particularly serious adverse events and discontinuation emergent adverse events, were poorly reported in published articles (see Table 1).

Conclusions: Whilst there was no evidence of selective reporting in clinical study reports, there was evidence of publication bias and selective reporting in publications. Clinical study reports contained important data on harms that were not available in publications.