Date of Award

Fall 12-2018

Degree Name

Doctor of Philosophy (PhD)

Advisor(s)

RUOSHA LI, PHD

Second Advisor

DEJIAN LAI, PHD

Third Advisor

HAN CHEN, PHD

Abstract

Clinical trials are complicated and involve human beings. Therefore, lots of ethical and efficient objectives are expected to be achieved. These objectives include maximizing the power of detecting the treatment effects, assigning more patients to the better treatments, saving the cost and time, and controlling the type I error rate. A variety of adaptive designs have been proposed to achieve different aims, among which sequential monitoring and sample size re-estimation are very popular in real clinical trials. In addition, adaptive randomization designs sequentially update the allocation probability aiming to target different allocation proportions and achieve different aims. Hu and Rosenberger (2006), classified adaptive randomization design into four categories, i.e., permuted block randomization, covariate-adaptive randomization (CAR), response-adaptive randomization (RAR), and covariate-adjusted response-adaptive randomization. In this dissertation, I investigate the combination of sequential monitoring, sample size re-estimation, and two types of adaptive randomization designs, i.e., CAR and RAR. For RAR, I focus on urn models. For CAR, I study three scenarios depending on whether all, part, or none of the randomization covariates are included in the data analysis. I propose methods to control the type I error rate, offer the theoretical results, and perform comprehensive numerical studies to show that the methods can protect the type I error rate and have advantages over traditional designs.

Share

COinS