Estimation of Accuracy of Diagnostic Tests Without a Gold Standard Using Latent Class Models

• Activity Number: 265
• Type: Invited
• Date/Time: Tuesday, August 5, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Mental Health Statistics Section
• Author(s): Xiao-Hua Andrew Zhou*+ and Zheyu Wang
• Companies: University of Washington and Johns Hopkins University
• Keywords: ROC Curves ; Diagnositc Tests ; Latent Class Models

Abstract

Estimation of the accuracy of diagnostic tests, such as sensitivity, specificity, and ROC curves, requires the truth of the disease status via a gold standard. However, in many studies, the gold standard may be hard to obtain, due to cost constraints, concerns about the invasive nature of the gold standard or a lack of technology needed to obtain the gold standard. For example, the current gold standard on diagnosis of Alzheimer's disease (AD) cannot be established until a patient has died and a neuropathological examination has been conducted. In this talk, we describe some new method for estimating the accuracy of diagnostic tests in the absence of the gold standard and give the necessary identifiability conditions for parameters of interest. Finally, we illustrate the application of the proposed method in a clinical study on dementia.