Screening for stratification in two-phase ('two-stage') epidemiological surveys.

Screening is undertaken not only for the purpose of giving therapeutic treatment to those among screen positive subjects found to be diagnosis positive, but it is also used in epidemiology as the first phase of multi-phase sampling designs. We review the use of such designs in epidemiology, and in psychiatric epidemiology in particular, focusing on two-phase or double sampling. We then compare a variety of approaches to the statistical analysis of data from such designs including the use of sampling weights, Gibbs sampling, full maximum likelihood for random effects logistic regression and a simple E-M algorithm for incomplete data. The methods are illustrated using data from a recent multi-phase study of psychiatric morbidity in northern Spain.