Publication bias in the medical literature: A review by a Canadian research ethics board

BACKGROUND: We reviewed the publication record of all protocols submitted to the Capital District Health Authority Research Ethics Board (REB) in Halifax, Nova Scotia, for the period 1995-1996. Because of a heightened awareness of the issue, we hypothesized that there would be less publication bias (a failure to report negative results) and a higher publication rate from completed studies, than previously reported. METHODS: Closed studies were identified from the REB database. Publications were identified by the investigators, requests from sponsors, and a literature review. For each publication, we identified authors, title, journal, number of subjects enrolled, and whether or not the publication was a report of a randomized clinical trial. Comparisons were done using a Student's t test, the Chi-square statistic, or Fisher's exact test as appropriate. RESULTS: From the database of closed studies, 106 remained unpublished, while completed investigations resulted in 84 publications (44% publication rate). The median time to publication was 32.5 months. Publication of statistically significant results occurred in 71/84 trials. Publication of protocols submitted by departments ranged from 91% (anesthesia; 10/11) to 25% [nursing; 2/8 (P<0.05)]. Trials investigating new drugs in Phase 3 or 4 studies were more likely to be published than trials investigating agents in Phase 1 or 2 (P<0.05), and were less likely to be published if sponsored by a pharmaceutical company (P<0.05). CONCLUSIONS: Publication bias continues to be a problem, particularly for early phase investigative studies. Our results suggest that a different approach is required to reduce publication bias. The role that REBs and peer-reviewed journals might play requires further exploration.