Adaptive Trial Design

The planning of a randomised clinical trial typically involves the knowledge of several basic characteristics. For example, for comparing means in a two-group parallel design the knowledge of the effect size and its variability enables calculation of the sample size needed for achieving a specified power when using a test at a given significance level. Misspecification of the designing parameters yields an underpowered or overpowered study.

Interim analyses in group sequential trials provide the possibility of determination of efficacy and safety prior to the planned end of the study. The practical use of group sequential designs, however is limited, since they do not allow for data-driven sample size reassessments or other design changes. E.g., if the treatment effect was overestimated and/or the variance turned out being larger than anticipated the study will not have enough power to yield a significant result.

New adaptive (flexible) study designs allow for correct data-driven re-estimation of the sample size controlling the type I error rate. In particular, redesigning the sample size in an interim analysis based on the results observed so far considerably improves the power of the trial since the best available information at time is used for the sample size adjustment, thus reducing the risk of a false-negative study outcome.

In recent years, several methods were proposed that enable a flexible design through the use of adaptive interim analyses while maintaining the type I error rate. A strategy that copes well with the demands of practice is based on combining the p-values obtained from the separate stages by use of the inverse normal method. This strategy was proposed for two-stage designs by Bauer and Köhne (1994) and more generally for multistage designs by Lehmacher and Wassmer (1999). You may download a recently written survey paper and see the details.

ADDPLAN Adaptive Designs - Plans and Analyses

A comprehensive software package for planning, simulating and analyzing adaptive studies is distributed by ADDPLAN GmbH (www.addplan.org).

Literature

An up to date selection of the literature published in the context of sample size reestimation and adaptive designs is provided upon request by Reinhard Eisebitt.