Statistical Hypothesis Testing for Personalizing Treatment
Title: Statistical hypothesis testing for personalizing treatment
Advisor: Professor Susan Murphy
Committee Members: Professor Xuming He, Professor Naisyin Wang, Professor Jeremy Taylor, Assistant Professor Ambuj Tewari
Abstract: Personalized treatment uses a decision rule that inputs patients’ characteristics and outputs recommended treatments. Clinical Investigators are often interested in whether a particular biomarker is useful for personalizing treatment. We define a discretevalued biomarker as useful in personalized decision making if for a particular value of the biomarker, there is sufficient evidence to recommend one treatment, while for other values of the biomarker, 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 procedure to test whether a biomarker is useful for personalized decision making. A set of clinical decisions can be reached based on the procedure. We choose critical values of the two-stage procedure to properly control selected error rates. The proposed procedure is compared with the qualitative interaction test of Gail and Simon, as well as some of the current practices of subgroup analysis in clinical trials.