Costanza M C
Medical Biostatistics/Biometry Facility, University of Vermont, Burlington 05405, USA.
Prev Med. 1995 Sep;24(5):425-33. doi: 10.1006/pmed.1995.1069.
Matching is an intuitively appealing design strategy for ensuring balance on one or more potential confounding variables, usually either among subjects who were exposed or unexposed to a suspected risk factor for disease in a cohort study or between diseased and nondiseased subjects in a case-control study. But does matching always automatically "control" confounding and is it always as good a strategy as it seems? It is the intention of this review to shed light on these questions primarily through illustrative examples of the effects of matching on the validity of point estimates of the odds ratio between exposure and disease status in both types of study designs. It is seen that the results of matching are more or less in line with expectations in cohort studies, but that matching can lead to unexpected results in case-control studies. In a case-control study, confounding is not automatically controlled by matching per se; rather, matching and a statistical analysis that properly accounts for the matching are needed to obtain a valid estimate of effect in a case-control study design.
匹配是一种直观上有吸引力的设计策略，用于确保在一个或多个潜在混杂变量上达到平衡，通常是在队列研究中暴露或未暴露于疾病可疑危险因素的受试者之间，或者在病例对照研究中患病与未患病的受试者之间。但是匹配是否总能自动“控制”混杂因素，它是否总是像看起来那样是一个好策略呢？本综述的目的主要是通过匹配对两种研究设计中暴露与疾病状态比值比的点估计有效性影响的示例，来阐明这些问题。可以看出，在队列研究中匹配结果或多或少符合预期，但在病例对照研究中匹配可能会导致意外结果。在病例对照研究中，匹配本身并不能自动控制混杂因素；相反，在病例对照研究设计中，需要匹配以及能恰当考虑匹配因素的统计分析，才能获得有效的效应估计值。
