Typical testing of multiple endpoints results in sequentially testing multiple hypotheses. While these can methods can create control of type I error they require a specification of order and ignore the possibility of borrowing of information across analyses.

In this talk we present some examples of analyzing a collection of endpoints using Bayesian hierarchical and cluster modeling and the ramifications to type I error. The advantageous are the ability to borrow information across endpoints in the effect of a device without an ordering of hypotheses. We explore the ramifications for type I error individually and across endpoints.