Friede, T. and Kieser, M. A. (2001) A loss function based approach for dose selection in two-stage dose-response trials. Journal of Epidemiology and Biostatistics, 6 (4). pp. 317-324.Full text not available from this repository.
Background Adaptive two-stage designs are a flexible tool in drug development and have the potential for an improvement of power over one-stage designs. In an adaptive two-stage dose-response trial, the dose-response relationship is examined in a preplanned interim analysis. If efficacy has not yet been proved, a subset of the most favourable doses may be selected for the second stage. This design offers the opportunity to combine dose selection and proof of efficacy within a single trial. Methods We consider a change-point regression model describing the dose-response relationship. In this framework, selection of the most favourable dose can be achieved by estimating the change point in the regression model. We introduce a change-point estimator that can be optimised by a loss-function based approach, taking into account both efficacy and safety aspects. We investigate the power characteristics of our approach. Results The proposed procedure performs well with regard to statistical power. The proposal demonstrates the feasibility of simultaneous modelling of efficacy and safety aspects by a loss-function based approach. Conclusion Adaptive two-stage designs, in conjunction with an elaborated dose-selection rule, can support the decision about the suitable dose to use, leading to a considerable gain in power (or saving in sample size) and possibly speeding up the time-to-market in drug development.
|Journal or Publication Title:||Journal of Epidemiology and Biostatistics|
|Uncontrolled Keywords:||Change ; Point ; Estimator ; Dose-RESPONSE ; Relationship ; Adaptive ; Design ; Interim ; Analysis|
|Subjects:||Q Science > QA Mathematics|
|Departments:||Faculty of Science and Technology > Mathematics and Statistics|
|Deposited On:||21 Nov 2008 12:42|
|Last Modified:||19 Jan 2017 01:15|
Actions (login required)