Keywords: Type I error protection, simultaneous inference, benefiting subgroup, subgroup analysis, multiple endpoints, Bayesian inference
Many new experimental treatments outperform the current standard only for a subset of the population. The credible subgroups method provides a pair of bounding subgroups for the benefiting subgroup constructed so that one contains *only* patients with an expected benefit, and the other contains *all* patients with an expected benefit. However, when more than two treatments and multiple endpoints are under consideration, there are many possible requirements for a particular treatment to be beneficial, and multiple approaches to adjusting for multiple comparisons. In this talk, we extend the credible subgroups method to handle such cases, investigate the extended method's performance via simulation, and apply it to an Alzheimer's disease treatment trial. Our results account for multiplicity while showing patient covariate profiles that are (or are not) likely to be associated with treatment benefit. Time permitting, we also discuss recent nonparametric extensions that permit spline-type flexibility in the modeled mean response, as well as an R package we have developed, credsubs, which leverages NIMBLE and shiny to deliver credible subgroup results in a user-friendly format.