Sequential Experimentation In Clinical Trials PDF Download

Randomized trials are the gold standard for the clinical assessment of a new treatment compared to a placebo or standard of care. Often in clinical trials, patients are accrued sequentially rather than all at once. Thus, the data from such a trial becomes available sequentially to the researcher. Monitoring and testing the accrued data throughout a trial and making decisions based on on such tests that could terminate the trial early is called sequential testing. Designing and analyzing such sequential trials has garnered much attention in the statistical literature over the last 50+ years. The added flexibility and benefi ts from such a trial do not come free-of-cost. Careful considerations in the design, careful monitoring of the data throughout, and careful analysis of the data at the conclusion are necessary to preserve the integrity of such a sequential clinical trial. This thesis will be mostly concerned with a special form of sequential testing called a group sequential procedure. Such procedures have the benefi t of a reduction in expected sample size while not being burdened by continual monitoring of the data after every observation. Special topics of group sequential procedures include the concepts of overrun, secondary endpoints and adaptive group sequential procedures. Overrun is the accrual of data after the decision to terminate the trial has been reached. We investigate and compare popular approaches to the incorporation of such data into the final analysis. Through a simulation study, it is found that a random weighting of the p-values from the data up to the termination of the trial and the overrun data based the sample sizes for such data under the Sample Mean Ordering of the outcome space leads to the shortest average con fidence intervals while maintaining the nominal coverage probability. Most clinical trials are designed and evaluated using a primary endpoint for the treatment eff ect. Some trials have secondary endpoints to assess either safety or additional clinical benefi ts beyond the primary outcome. We consider the design and analysis of group sequential trials when both a primary and secondary endpoint are of interest. Our investigations are done in the setting of a gatekeeping procedure. We are able to unify and generalize global proofs to certain propositions made by other researchers when we consider testing both a primary and secondary endpoint. We further investigate secondary inference in the form of con fidence interval construction through an extensive simulation study. We find that the approach of Whitehead et al. (2000) outperforms existing methods for the settings considered. Adaptive clinical trials seek to modify some aspect of the trial after an interim look at the data in order to improve the odds of a successful trial by the end. We compare some popular choices of adaptive Phase II two-stage designs and introduce a new design while evaluating operating characteristics (Type I error, Type II error and expected sample sizes). Majority of the literature focuses on minimizing the expected sample size under the null hypothesis only. Our new Quasi-Symmetric n2-design seeks to substantially reduce the expected sample size under the parameter values close to the design alternative while minimally increasing expected sample size under the design null. We evaluate and compare such a design to existing methods.