Cox B, Elwood J M
Hugh Adam Cancer Epidemiology Unit, University of Otago, Dunedin, New Zealand.
Am J Epidemiol. 1991 Jan 15;133(2):202-7. doi: 10.1093/oxfordjournals.aje.a115858.
Precise measurement of confounding or effect modifying variables is seldom possible. Nondifferential misclassification of such a variable can introduce bias in the odds ratios within the strata of the confounding variable. For two strata, the effect on the odds ratios is dependent on separate measures of the degree of nondifferential misclassification of the confounding variable. The authors investigate the bias in the stratum-specific odds ratios from nondifferential misclassification of a confounder in a case-control study, and show how estimates of the true stratum-specific odds ratios can be derived. For one stratum, the degree of bias introduced in the odds ratio is dependent only on the sensitivity of the measure of the confounding variable while only the specificity of this measure determines the bias introduced in the other stratum-specific odds ratio. The results are applied to a case-control study of fenoterol prescription and asthma death.
很少能够精确测量混杂变量或效应修饰变量。对此类变量的非差异性错误分类可能会在混杂变量的各分层内的比值比中引入偏差。对于两个分层而言，对比值比的影响取决于对混杂变量非差异性错误分类程度的单独测量。作者们在一项病例对照研究中调查了由于混杂因素的非差异性错误分类而导致的分层特异性比值比中的偏差，并展示了如何推导真实分层特异性比值比的估计值。对于一个分层，比值比中引入的偏差程度仅取决于混杂变量测量的敏感度，而仅该测量的特异度决定了在另一个分层特异性比值比中引入的偏差。研究结果应用于一项关于非诺特罗处方与哮喘死亡的病例对照研究。
