Computational methods for comparing two binary diagnostic tests in the presence of partial verification of disease (Q638052)

The authors consider data of application of two binary diagnostic tests of some disease to a set (sample) of patients. The true status of the patients is known for only a subsample. The problem is to estimate the sensitivity and specificity of the tests. The Vacek model of conditional dependence is assumed for the tests results where the dependence is modeled as a multiplicative factor. A EM-type algorithm is proposed for the calculation of the maximum likelihood estimates and a SEM algorithm is described for the estimation of the covariances. Results of simulations and application to the diagnostic of coronary stenosis are presented.