|
Activity Number:
|
176
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 4, 2003 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #302128 |
|
Title:
|
Optimal Conditional Error Functions for the Control of Conditional Power
|
|
Author(s):
|
Werner Brannath*+ and Peter Bauer
|
|
Companies:
|
University of Vienna and Instit Fur Medizinische Statistik Der Universitat Wien
|
|
Address:
|
Department of Medical Statistics, Vienna, , A-1090, Austria
|
|
Keywords:
|
adaptive design ; clinical trial ; group sequential test ; maximum likelihood ; Neyman-Pearson lemma ; sample size recalculation
|
|
Abstract:
|
Flexible designs have been suggested which allow for data-dependent adaptations of sample sizes at an interim analysis without compromising the Type I error rate. In particular, the control of conditional power at an interim analysis can be achieved. The conditional power is the probability to reject the null hypothesis given the results of the interim analysis and the alternative of interest. This conditional rejection probability is important for many clinical trials, as investigators and sponsors often aim to achieve sufficient conditional power. Various choices of flexible designs have been suggested. However, an "optimal design'' with control of conditional power has not be given yet. We identify (in an exact manner) those two-stage designs which minimize the expected sample size when sample sizes are reassessed for conditional power. The minimization is done under either a single alternative or a weighted average of several alternatives. The solution is based on a Neyman-Pearson lemma argument. We consider also a maximum likelihood method. The methods are compared with each other, and to more "classical" two-stage designs. We further discuss some application issues.
|
