Interim Analysis in Adaptive Randomized Clinical Trials Open Access
Downloadable ContentDownload PDF
Clinical trials usually take a period of time to recruit volunteers, and it becomes a steady accumulation of data throughout the durations. This accrual of data naturally enables us to use what is learned during the course of a study or clinical development program to make adjustments that are beneficial to the trial. Interim analysis keeps the decision process free of conflict of interest while considering cost, resources, and meaningfulness of the project. Whenever necessary, interim analyses can also comprise a series of sequential monitoring and sample size re-estimation, and become a standard technique in conducting clinical trials. The principal goals are to stop recruiting volunteer and the trials as soon as there is sufficient evidence to reach a firm conclusion; to adjust the originally planned sample size to provide adequate power using interim findings.Traditionally, the sample size of a trial is determined in advanced and data is collected from all subjects before analysis proceeds. Over the past decades, many strategies have been proposed and rigorous theoretical groundings have been provided to conduct sample size re-estimation. However, the application of these methodologies has not been extended to take care of trials with adaptive designs. Our goal is to combine interim analysis and adaptive randomization design and utilized the advantages from both. Specifically, we aim to fill the gap by proposing several strategies to conduct interim analysis along with either sequential monitoring or sample size re-estimation on three types of adaptive randomization procedures: response-adaptive, covariate-adaptive and covariate-adjusted response-adaptive design. In addition, the statistical inference of interim analysis on adaptive designs will be studied thoroughly in each case. In Chapters 2 and 4, interim analysis with two-stage sample size re-estimation on response-adaptive and covariate-adaptive design are proposed, respectively. For ethical and economical concerns, we also use multiple stopping criteria with the allowance of early termination. In Chapter 3, we extend the work to incorporate sample size re-estimation under response-adaptive design in multiple stages and introduce the concept of "self-designing'' trial. The inference is studied for the hypothesis testings and we show that the test statistics for each stage are independently normally distributed though the test statistic on the following stage uses data from previous stages. In Chapter 5, we apply sequential monitoring on covariate-adjusted response-adaptive design and prove that the sequential test statistics follow Markov process asymptotically. In such case, critical boundaries could be derived to preserve the correct type I error probability. We found that the procedures we proposed in this work have advantages in many aspects, compared to fixed sample design and complete randomization procedure. First, type I error can be well control at desired level. Second, power has increased with adjusted sample size under multiple scenarios that are commonly seen in practice. Third, sample size has decreased when we underestimate the treatment effect prior to the start of trial. Next, duration of trials has shortened. Moreover, failure rate of the entire trial has also reduced. These advantages are evidenced by both theoretical and numerical justifications.