Kilpatrick S J
Department of Biostatistics, Medical College of Virginia, Richmond 23298-0032.
Int Surg. 1992 Apr-Jun;77(2):131-3.
This paper demonstrates that Wald's meta-analysis is an unadjusted combination of unadjusted estimates. There are many differences among the design and implementation of the original studies combined here. One consequence of these differences is that studies of spousal smoking and lung cancer mortality conducted in the United States have a lower combined RR than those done elsewhere. Here it is shown that adjustment for this stratification results is a non significant finding as to the association of ETS with lung cancer. Likewise, the addition, for illustration, of Varela's Ph.D. thesis that is the largest case-control study to date, changes the overall results from a significant to a non-significant result. This paper does not attempt to update Wald's meta-analysis with more recent studies. This is being done, I believe by the Environmental Protection Agency. I have given here my reasons for suspecting the conclusions of meta-analysis of non-randomized studies and the illustrations provided here of the effect of publication bias and covariate adjustment support this view.
本文表明，瓦尔德的荟萃分析是未经调整的估计值的未经调整的组合。这里合并的原始研究在设计和实施方面存在许多差异。这些差异的一个后果是，在美国进行的配偶吸烟与肺癌死亡率的研究的合并相对危险度低于在其他地方进行的研究。这里表明，针对这种分层结果进行调整对于环境烟草烟雾与肺癌的关联而言是一个无显著意义的发现。同样，举例来说，纳入瓦雷拉的博士论文（这是迄今为止最大的病例对照研究）后，总体结果从显著变为无显著意义。本文并未尝试用更新的研究来更新瓦尔德的荟萃分析。我认为环境保护局正在做这件事。我在此给出了怀疑非随机研究荟萃分析结论的理由，并且这里提供的发表偏倚和协变量调整影响的例证支持了这一观点。
