Statistical Methodology for Design and Analysis of Trials with Biomarker-Guided Strategies
Precision medicine has emerged from the awareness that many human diseases are intrinsically heterogeneous among patients. Instead of using "one-size-fits all" treatment selection, it is of interest to identify subpopulations who will benefit more from the targeted therapies based on their heterogeneous molecular information and provide estimation of localized treatment benefit for the targeted population instead of providing population average treatment effect. However, effectively identifying subpopulations in personalized medicine with targeted therapies and distinguishing predictive biomarkers from prognostic remain challenging when treatment effect is heterogeneous among the populations. In my dissertation, I present a general approach for estimating the mean localized predictive treatment benefit for biomarker-guided strategies when evaluated in the context of a predictive biomarker validation design. Then the approach is expanded to a retrospective study with adjustment for selection bias. The methodology is also demonstrated through a case study of lower grade glioma and applied to a Bayesian imaged-guided adaptive design. Furthermore, I present a Bayesian predictive model to identify subpopulations. The approach enables personalized treatment selection for future patients based on molecular measurements, clinical responses and treatment histories from historically treated patients. Using this approach, we provide point estimation for posterior predictive probability to respond for each treatment, type I error rate, power and classification accuracy for identifying the subpopulations. The method is also demonstrated through a case study of non-small-cell lung cancer. Lastly, this method is further demonstrated to minimize the local treatment randomization imbalance, and compared to Bayesian response-adaptive design.^
Huang, Meilin, "Statistical Methodology for Design and Analysis of Trials with Biomarker-Guided Strategies" (2017). Texas Medical Center Dissertations (via ProQuest). AAI10642428.