Objective To study the application of latent class method in the evaluation of accuracy for diagnostic test without a gold standard.
Methods The principle, design and evaluation based on latent class model were considered in diagnostic tests without a gold standard. An example with two population and two tests was demonstrated for the application of latent class.
Results There were at least two populations and two tests, or one population and three tests in classic statistics to satisfy the model identifiability for dichotomization response variable, under the assumption of independence and invariability of accuracy. Bayesian statistics could be used in non-identifiability model with the introduction of prior distribution, but prior dependence became more obvious in the lower prevalence.
Conclusion It is helpful to evaluate the accuracy of diagnostic test without a gold standard using latent class method, however, the appropriate design and method should be selected carefully.
DownLoad: