|
Abstract:
|
While the narrative is built for the case of two binary endpoint the generalization for the higher dimension is straightforward. The primary goal is to establish (estimate) the dose-response relationship. The traditional optimal experimental design is concerned mainly with collective ethics: one tries to gain as much information as possible without explicit concern about patients in the trial, i.e. doing what is best for future patients. In contrast, the currently popular procedures for dose-finding studies are mainly concerned with individual ethics: doing what is best for current patients in the trial. The proposed approach attempts to compromise between these two extremes: we maximize information under constraints on a penalty function, which depends on efficacy and toxicity of each dose entering a design. Locally optimal, minimax, Bayesian designs are discussed.
|
