Statistical Methods in Diagnostic Medicine

描述（由申请人提供）：诊断测试有两个目的：(1)提供有关患者病情的可靠信息；(2)影响医生管理患者的计划。只有当医生知道如何解释测试时，测试才能达到这些目的。通过评估检测的诊断准确性需要这些信息，即检测区分存在或不存在疾病的能力。当诊断试验的响应是连续的时，其诊断准确性最好由受试者工作特征（ROC）曲线表示。目前使用最广泛的估计带有协变量的ROC曲线的方法是参数或半参数方法。然而，估计的ROC曲线可能对链接函数的参数假设非常敏感。最后，目前大多数ROC研究采用固定样本设计，这是不有效的。虽然在ROC研究中使用组序贯设计有很多很好的理由，但由于缺乏在ROC研究中进行GSD的可用方法，组序贯设计尚未在ROC研究中广泛使用。在本提案中，我们将开发具有协变量的ROC曲线的非参数回归模型和在数据单调变换下不变的估计方法。我们还将开发具有协变量和估计方法的部分或全部ROC曲线区域的非参数回归模型。最后，我们将使用非参数估计的ROC曲线为比较ROC研究开发一般组序贯设计。

DESCRIPTION (provided by applicant): A diagnostic test has two purposes: (1) to provide reliable information about the patient's condition and (2) to influence the physician's plan for managing the patient. A test can serve these purposes only if the physicians know how to interpret it. This information is required through an assessment of the test's diagnostic accuracy, which is the ability of a test to discriminate between the presence or absence of the disease. When the response of a diagnostic test is continuous, its diagnostic accuracy is best represented by the receiver operating characteristic (ROC) curve. Currently most widely used methods for estimating ROC curves with covariates are parametric or semi-parametric. However, estimated ROC curves can be very sensitive to the parametric assumption on the link function. Finally, most current ROC studies employ fixed sample designs, which are not efficient. Although there are many good reasons for the use of group sequential designs in ROC studies, the group sequential design has not been used widely in ROC studies because of the lack of available methodology for conducting the GSD in ROC studies. In this proposal we will develop non-parametric regression models for ROC curves with covariates and estimation methods that are invariant under the monotone transformation of data. We will also develop nonparametric regression models for partial or full regions of ROC curves with covariates and estimation methods. Finally, we will develop general group sequential designs for comparative ROC studies using nonparametrically estimated ROC curves.