Meta-Analysis and Assessment of Publication Bias

摘要

The objective of the research was to investigate factors influencing publication of research findings, in particular, the effect of 'significant results on publication.' First, a large international survey of obstetricians and pediatricians was conducted to identify unpublished perinatal clinical trials. The results indicated that retrospective identification of completed trials is extremely difficult, if not impossible. A second study followed abstracts of randomized trials in order to examine factors related to subsequent publication. Publication was not associated with trial quality or statistical significance of the findings. There was a possible association between sample size and final publication of results. A third project involved the followup of all studies initiated at Johns Hopkins Schools of Medicine and Public Health prior to 1981. Findings indicated that there is a publication bias, such that investigators tend to selectively publish results that are statistically significant. Other factors found to be related to publication were having external funding and being a multicenter study. Publication was not associated with sample size, more than one study group, or type of study design.