Keywords: Bayesian analysis, hurdle model, two-part model, Poisson regression, shared frailty model
Panel counts are often encountered in longitudinal studies where individuals are followed over time and the number of events occurring in time intervals, or panels, are recorded. This work discusses methods for situations where, in addition to the counts of events, we also record a mark, denoting a measure of prognostic factors or severity of the events. In many situations there is an association between the recurring processes and their marks. This occurs in the motivating example for this work, a study of the effects of two hormone therapy treatments in reducing vasomotor symptoms in healthy menstruating women prior to hysterectomy/ovariectomy for benign disease. We compare a joint analysis that models the event counts and their severities jointly through the use of shared or linking random effects, and compare to an approach that analyzes a composite measure built as the number of events multiplied by their severity.