Kleijnen J, Knipschild P
Department of Epidemiology and Biostatistics, University of Limburg, Maastricht, The Netherlands.
Arzneimittelforschung. 1992 May;42(5):587-91.
Publication bias occurs if the results from studies which have not been published are different from the published ones. From a Bayesian viewpoint, it also concerns non-publication of studies with similar results as the published ones because the strength of the evidence will be influenced. Publication bias complicates the interpretation of reviews and meta-analyses. If favourable results are published more often there will be an overestimation of the effects of a treatment. There have been several attempts to assess the magnitude of publication bias. Unpublished trials could be identified by means of a survey among researchers, and the results could subsequently be compared with the outcomes of published trials. Also, the results from published trials could be compared with trials from a registry. Furthermore, the results from registered but unpublished trials could be compared with those of registered and subsequently published trials. Studies addressing publication bias have shown that it is a serious problem which complicates the interpretation of reviews. In assessments of publication bias other factors must be taken into account. These include the mode of publication: refereed journals, other journals, books, etc. Differences could also be related to the quality of trials. Finally, the source of funding may influence both the results and subsequent publication. Publication bias can only be avoided by registration of all trials before data collection is started; several of such registries have already been installed. Perhaps, if more of such registries exist, reviewers could only use registered trials for their main conclusions. All other information could then be considered sensitive to publication bias.
如果未发表研究的结果与已发表研究的结果不同，就会出现发表偏倚。从贝叶斯观点来看，它还涉及与已发表研究结果相似的研究未被发表的情况，因为证据的力度会受到影响。发表偏倚使综述和荟萃分析的解读变得复杂。如果阳性结果发表得更频繁，那么对一种治疗效果的估计就会过高。已经有几次评估发表偏倚程度的尝试。未发表的试验可以通过对研究人员进行调查来识别，随后将结果与已发表试验的结果进行比较。此外，已发表试验的结果可以与来自登记处的试验进行比较。而且，已登记但未发表试验的结果可以与已登记并随后发表试验的结果进行比较。针对发表偏倚的研究表明，这是一个严重的问题，会使综述的解读变得复杂。在评估发表偏倚时，必须考虑其他因素。这些因素包括发表方式：同行评审期刊、其他期刊、书籍等。差异也可能与试验质量有关。最后，资金来源可能会影响结果及后续发表。只有在开始数据收集之前对所有试验进行登记，才能避免发表偏倚；已经设立了几个这样的登记处。也许，如果有更多这样的登记处，综述作者在得出主要结论时只能使用已登记的试验。那么所有其他信息都可能被认为容易受到发表偏倚的影响。
