An adaptive method for establishing a dose-response relationship.

A multistage sampling method is proposed in dose-response trials, where dose adaptions can be performed in the preplanned adaptive interim analyses. The overall test for proving a dose-response relationship is performed by Fisher's product criterion for the p-values from the separate tests of a dose effect in the disjoint samples at the different stages. Based on these p-values decision boundaries for early stopping with the rejection of the global null hypothesis of no existing dose-response relationship are introduced. The power of the adaptive two-stage procedure using a particular adaptation rule is compared with the power of the test for a linear trend under analysis of variance assumptions in extensive simulations. The bias in estimation is also quantified. This procedure could be used for establishing a dose-response relationship without including a placebo treatment.