Online Program

Hypothesis Testing for Personalizing Treatment

*Huitian Lei, University of Michigan 
Susan A Murphy, University of Michigan 

Keywords: Personalized treatment, Subgroup analysis, Hypothesis testing

In personalized treatment the recommended treatment is based on patient characteristics. Given pre-specified subgroups, we define the subgroup indicator as useful in personalized decision making if for particular subgroups there is sufficient evidence to recommend one treatment, while for other subgroups, either there is sufficient evidence to recommend a different treatment, or there is insufficient evidence to recommend a particular treatment. We propose a two-stage hypothesis testing procedure to evaluate if a subgroup indicator is useful in personalized decision making. In the first stage of the procedure, we utilize the test statistic for testing treatment-subgroup interaction. If the first stage test statistic exceeds the critical value, we proceed to the second stage of the procedure and utilize test statistics for testing subgroup treatment effects. We control a generalized Type I error rate. We illustrate the proposed procedure using data from a depression study involving the medication, Nefazodone and a combination of a behavorial therapy with Nefazodone.