Biomedical research and clinical trials using restricted mean survival time

Abstract

Modern oncology study is rapidly advancing in the present era. Given the distinctive attributes of cancer therapy, including its extended treatment cycles, the development of novel biomedical research and the design of clinical trials becomes essential. This thesis proposes new statistical methods handling survival outcomes based on the theory of survival analysis and restricted mean survival time. These methods can be applied in the drug development stage for the screening of biomarkers, and the subsequent designing of new clinical trials. In the stage of feature screening, a robust model-free screening approach for right-censored survival data is proposed, addressing the challenge of detecting continuous variables with nonlinear or non-monotonic relationships with the response variable based on a new measure utilizing restricted mean survival time (RMST). Sure screening property is established, along with the introduction of an iterative screening procedure to handle multicollinearity among high dimensional covariates. The method’s potential in handling interval-censored failure time data is also explored, yielding promising results. However, the feature screening methods often involve manually setting the number of variables, resulting in subjective outcomes and the inclusion of many unimportant variables. Therefore, a new variable selection method called SurvClipper that controls the false discovery rate (FDR) is developed based on the proposed measure to detect prognostic biomarkers. To enhance the efficiency of SurvClipper, we introduce patient information and historical studies into the selection and develop a weighted method called wSurvClipper, which leverages previous research findings and can identify more accurate sets of variables, also with a controlled FDR. In terms of conducting clinical trials for survival endpoint, a novel method of designing multi-arm clinical trials based on RMST is proposed that can be applied in both phase II/III settings using a global χ2 test as well as a modelling based multiple comparison procedure. The framework provides a closed form sample size formula built upon multi-arm global test and a sample size determination procedure based on multiple-comparison in the phase II dosefinding study. The proposed method enjoys strong robustness and flexibility as it requires less a priori set-up than conventional work, and obtains smaller sample size while achieving the target power. Driven by evolving FDA recommendations, modern clinical trials demand innovative designs that strike a balance between statistical rigor and ethical considerations. Covariate-adjusted response-adaptive (CARA) designs bridge this gap by utilizing patient attributes and responses to skew treatment allocation in favor of the treatment to be best for an individual patient’s profiles. To address the limitation that existing CARA designs for survival outcomes often hinge on specific parametric models, a novel CARA design for survival outcomes (CARAS), based on the Cox model and an innovative variance estimator is proposed. This method addresses issues of model misspecification and enhances the flexibility of the design. We also propose a group sequential overlap-weighted log-rank test to preserve type I error rate in the context of group sequential trials using CARAS. Extensive simulation studies demonstrate the clinical benefit, statistical efficiency, and robustness to model misspecification of the proposed method compared to traditional randomized controlled trial designs and response-adaptive randomization designs.