Walter S D
Department of Clinical Epidemiology and Biostatistics, McMaster University, Hamilton, Ontario, Canada.
Epidemiology. 1999 Jan;10(1):67-72.
Estimation of sensitivity and specificity for diagnostic or screening tests usually requires independent confirmation of subjects as diseased or nondiseased using a gold standard. In practice, however, application of the confirmatory procedure is usually limited to individuals with one or more positive test results. For situations in which two initial tests are applied, recent literature has shown that one can use the data from confirmed disease cases to estimate the ratio of test sensitivities and the information from confirmed noncases to estimate the ratio of false-positive rates. In this paper, I show that estimates of sensitivity and specificity can be obtained for each test separately, together with an estimate of the disease prevalence. The only additional information required compared with previous methodology is the total number of individuals tested, a quantity that is usually readily available. The assumption that the test errors are independent is required. Although specific patterns of test errors cannot be identified, the overall assumption can be tested using goodness of fit. I illustrate the methods using data on breast cancer screening. Provision of sensitivity and specificity estimates for each test separately provide considerably greater insight into the data than previous methods.
诊断或筛查试验的敏感性和特异性估计通常需要使用金标准对受试者是否患病进行独立确认。然而，在实际操作中，确认程序通常仅适用于一项或多项检测结果呈阳性的个体。对于应用两项初始检测的情况，近期文献表明，人们可以利用确诊病例的数据来估计检测敏感性的比率，并利用确诊非病例的信息来估计假阳性率的比率。在本文中，我表明可以分别获得每项检测的敏感性和特异性估计值，以及疾病患病率的估计值。与先前的方法相比，唯一需要的额外信息是检测个体的总数，这一数量通常很容易获得。需要假设检测误差是独立的。尽管无法识别特定的检测误差模式，但总体假设可以通过拟合优度检验进行验证。我使用乳腺癌筛查数据来说明这些方法。分别为每项检测提供敏感性和特异性估计值，比以前的方法能更深入地洞察数据。
