Crossed Random Effects Modeling Approaches for Estimating Diagnostic Accuracy from Ordinal Tests When the True Ordinal Disease Status Is Unknown

• Activity Number: 178
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Yunlong Xie*+
• Companies: National Institutes of Health
• Keywords: gold standard ; latent class models ; crossed random effects modeling ; Monte-Carlo EM algorithm

Abstract

In diagnostic studies, a gold standard is often unavailable or difficult to obtain. Statistical methods based on latent class models (LCMs) are commonly used to estimate the diagnostic accuracy parameters when the true binary disease status is unknown. Recently Wang et al (Biostatistics 2011) proposed a modeling framework extending these LCMs to the case where the test results and unknown true disease status are both ordinal. They make the strong assumption of conditional independence where the test results are assumed independent given the true disease status. We relax this assumption and consider a crossed random effects modeling approach for this situation. A Monte-Carlo EM algorithm is proposed for parameter estimation. Simulation studies of the model indicate that the MCEM algorithm has good operating characteristics. The proposed approach is applied to data from the Physician Reliability Study (PRS) in staging of endometriosis.