Estimating magnitude of publication bias in medical study areas and measures to counter it

点击翻译按钮获取中文摘要

Publication bias is a recognized phenomenon, i.e.studies with statistically significant or clear-cut results are more likely to be published than those finding no difference between the groups studied. Summarized odds ratio calculated from odds ratios of pubulished studies in a meta-analysis may be overestimated because of publication bias. Sugita et al.have developed a method of estimating indirectly the summarized odds ratio of all studies in a given research area, including not only those published but also those unpublished. In the present study, nonpublication in several medical study areas were investigated from the probability density function of all the studies using Sugita's method and a histogram of those published. Notable nonpublication of studies whose odds ratios were close to unity was shown from examples of studies on relationship between passive smoking and lung cancer and clinical trial studies on adjuvant chemotherapy for colorectal cancer, whereas nonpublication of studies whose summarized odds ratio was located far from unity was not detected from examples of studies on relationship between cryptorchidism and testicular cancer. In all of the study areas, however, the differences between the summarized odds ratios, either with or without the hypothetical unpublished studies, were not large. The small difference, however, should not be ignored, when the study area is recognized as a social problem. It is pointed out that some of epidemiological studies with odds ratios slightly less than unity and clinical studies with odds ratios slightly more than unity were not published in medical study areas for weak association.