Abstract
For many medical and psychological problems, multiple treatment alternatives are available. Given data from a randomized controlled trial, an important challenge is to estimate an optimal treatment regime, that is, an optimal decision rule that specifies for each patient the most effective treatment alternative given that patient’s pattern of pretreatment characteristics. In a first part of this talk, I will discuss different possible specifications of the concept of treatment regime optimality in terms of two key elements that are involved in any optimization problem: the criterion function that is to be optimized and the search space. Subsequently, I will briefly outline two possible types of strategies that may be adopted in optimal treatment regime estimation, along with a few challenges that show up in this estimation. In a second part of this talk, I will address optimal treatment regime estimation in the special case of a mean-based criterion function, a search space comprising tree-based treatment regimes, and k>2 treatment alternatives. The estimation problem at hand can be shown to come down to a classification problem with a unit- and class-dependent misclassification cost, that is, a misclassification cost that may depend both on the object that is misclassified and the class to which it is erroneously assigned. This estimation can be achieved by means of a shrewd and novel type of application of a mainstream R package for tree building, rpart, via a user-defined splitting function and a rectangular misclassification cost matrix. I will illustrate with an application on the search for an optimal tree-based treatment regime in a randomized controlled trial on k=3 different types of after-care for younger women with early-stage breast cancer.
Organizzatore
Christian Hennig