Modeling Without a Gold Standard: Stratification with Stratum-Dependent Parameters

Abstract

Bayesian latent-class models have been widely applied for assessing the performance of diagnostic tests in the absence of a gold standard. We provide a short discussion on identifiability issues appearing under the absence of a gold standard, and construct an extension of the well-known Hui–Walter stratification model which allows for stratumdependent parameters. We illustrate our approach using a Chagas disease case study on blood donors from Brazil.

Keywords

• absence of a gold standard
• diagnostic test
• identifiability
• sample size
• stratification