Data are often clustered in correlated groups in medical studies. Endpoints of interest may include the cluster size and/or one or more binary outcomes on each unit within a cluster. In vaccine and toxicology applications the effect of an exposure on the cluster size can be assumed to be isotonic. A variety of resampling methods are proposed for imputing the potential cluster sizes in the absence of exposure under this assumption. Standard models can then be fit to estimate a dose-response relationship or to test for an overall toxic effect. These methods are applied to data sets and are evaluated through simulation studies. Fully nonparametric and Bayesian alternatives are proposed.

Back to Workshop Schedule