ABSTRACT
With the increasing interest in individualized medicine there is a greater need for robust statistical methods for prediction of optimal treatment based on the patient’s characteristics. When evaluating two treatments, one treatment may not be uniformly superior to the other treatment for all patients. A patient characteristic may interact with one of the treatments and change the effect of the treatment on the response. Clinical trials are also collecting more information on the patient. This additional information on the patients combined with the state of the art in model selection allows researchers to build better optimal treatment algorithms. In this chapter we introduce a methodology for predicting optimal treatment. The
methodology is demonstrated first on a simulation and then on a phase III clinical trial in neuro-oncology.
