FRIDAY, January 30, 2004
Time: 1:30 - 2:30 PM
Constant Hall Room 1043

Dr. James L. Kepner
Director, CCSG Biostatistics/Biomathematics Core Resource
Chair, Department of Biostatistics
Roswell Park Cancer Institute
Professor of Biostatistics
SUNY at Buffalo

A common statistical application is to show that a treatment causes a change in a binomial proportion. That is, it is of interest to test H₀: p = p₀ versus one of the alternatives, H₁: p > p₀, H₁: p < p₀, or H₁: p != p₀, where p₀ is known. The required sample size is a function of the magnitude of the Type I and Type II errors. In this presentation, we investigate the problem of determining the sample size when using exact k-stage group sequential designs for detecting these and other similar hypotheses. Several results described are extraordinary and surprising. The following topics will be discussed.

• Limitations of using asymptotic methods.

• A web-based application that allows users to rapidly find exact Type 1, 2, and 3 group sequential k-stage designs that satisfy user defined parameters when k = 1, 2, and 3 will be demonstrated via the Internet.

• Constraints on the magnitude of k will be described in the context of the theorems of Kepner and Chang (2003).

• Designs for Phase II clinical trials for which improved efficacy and comparable or reduced toxicity are desired endpoints.

• Uniformly most powerful k-stage exact group sequential designs.