Proof of Concept and dose estimation with binary responses

This article suggests a unified framework for testing Proof of Concept and estimating a target dose for the benefit of a more comprehensive, robust and powerful analysis in phase II or similar clinical trials. From a pre-specified set of candidate models we choose the ones that best describe the observed dose-response. To decide which models, if any, significantly pick up a dose effect we construct the permutation distribution of the maximum penalized deviance difference over the candidate set. This allows us to find critical values and multiplicity adjusted p-values that control the familywise error rate of declaring any spurious effect in the candidate set as significant. Model averaging is then used to estimate a target dose. Popular single or multiple contrast tests for Proof of Concept, such as the Cochran-Armitage, Dunnett or Williams tests are only optimal for specific dose-response shapes, suffer from a certain arbitrariness in selecting contrast coefficients and do not provide target dose estimates with confidence limits. A thorough evaluation and comparison of our approach to these tests reveals that its power is as good or better in detecting a dose-response under various shapes, with many more additional benefits: It incorporates model uncertainty in Proof of Concept decisions and target dose estimation, yields confidence intervals for target dose estimates, allows for adjustments due to covariates and extends to more complicated data structures. We illustrate our method with the analysis of a Phase II clinical trial. Copyright c © 2008 John Wiley & Sons, Ltd.