Estimating Diagnostic Accuracy from Designs with no Gold Standard, Partial Gold Standard, or Imperfect Gold Standard Evaluation
*Paul S Albert, NCI 

Keywords: latent class models, diagnostic accuracy, semi-latent class models,

Interest often focuses on estimating sensitivity and specificity of a group of raters or a set of new diagnostic tests in situations where gold standard evaluation is invasive or expensive. As shown by Albert and Dodd (2004), these approaches may be problematic due to a lack of robustness to modeling assumptions and the inability to distinguish between competing models for the dependence between tests. In this talk, I will propose two alternatives to latent class models. First, semi-latent class models are proposed for estimating diagnostic accuracy when gold standard tests are available on a fraction of individuals. Second, an approach is proposed when an imperfect reference standard is observed along with diagnostic accuracy estimates of the imperfect test. Using simulations and data analyses, we show the advantages of these approaches over latent class models without a gold standard.