Background: Meta-analyses based on published studies are probably subject to a publication bias of unknown size. The aim of the present study was to estimate the size of this bias for selektiv serotonin reuptake inhibitors (SSRI) by comparing pooled results obtained from published studies with pooled results obtained from studies submitted to the Swedish drug regulatory agency.

Methods: All placebo-controlled short-term studies included in the applications for marketing authorisation of SSRI in Sweden were identified (the applicant has an obligation to submit all studies). Published versions of the submitted studies were searched for through a computerised search in Medline, Embase and Psychological abstracts, by scrutinising reference lists and finally through enquires with the concerned pharmaceutical companies. The applications were submitted 8-10 years ago and publications were searched for up to the end of 1999. Pooled results based on submitted studies that were published were compared with pooled results based on all submitted studies.

Results: A considerable part of the submitted studies was never published. A clear tendency towards publishing more favourable results was seen. Pooled analyses based on only published studies overestimated the magnitude of the effect. For one of the compounds the difference in responder rate compared to placebo was approximately 10% units higher in the analysis based on published studies compared to the analysis based on all submitted studies.

Conclusion: Evidence of publication bias was seen in this study. This makes it difficult to draw conclusions about efficacy without knowledge of all performed studies. Measures to make all clinical trials known and available for the medical profession should be encouraged.