Making diagnoses with multiple tests under no gold standard

摘要: In many applications, it is common to have multiple diagnostic tests on each subject. When there are available, combining incorporate information from various aspects in subjects may be necessary order obtain a better diagnostic. For continuous tests, the presence of gold standard, we could combine linearly [59] or sequentially [64], using some risk score as studied [36]. The however, not always available practice. This dissertation concentrates deriving classification methods based absence standard. Motivated by lab data set consisting two testing for an antibody 100 blood samples, first develop mixture model four bivariate normal distributions with probabilities depending two-stage latent structure. proposed structure biological mechanism tests. A Bayesian method incorporating prior derived utilizing decision theory. illustrated motivating example, and properties estimation described via simulation studies. Sensitivity choice distribution also studied. We investigate general problem without any standard reference test. thoroughly study existing optimal rules cor-