Publication bias: the problem that won't go away.

Conclusions about the efficacy and safety of medical interventions are based on data presented in the scientific literature. The validity of these conclusions is threatened if publication bias results from investigators or editors making decisions about publishing study results on the basis of the direction or strength of the study findings. This paper reports meta-analyses performed using data from four prospective investigations in which a total of 997 initiated studies were followed to learn of study results, publication status, and reasons for nonpublication. The analysis indicates that there is a positive association between "significant" study results and publication (OR = 2.88; 95% confidence interval [CI] 2.13 to 3.90). When the analysis was restricted to controlled trials (n = 280), an even stronger relationship between "significant" results and publication was observed (OR = 6.15; 95% CI 2.24 to 16.92), with randomized trials (n = 200) apparently no less susceptible to publication bias than controlled trials in general (OR = 8.72; 95% CI 1.91 to 39.81). In every case, failure to publish was investigator-based, and not due to editorial decisions. The results of clinical trials should not be suppressed in this way. Development of registration systems for randomized trials is essential if this problem is to be minimized in future.