Algorithms for finding locally and Bayesian optimal designs for binary dose-response models with control mortality

Algorithms for finding optimal designs for three-parameter binary dose-response models that incorporate control mortality are described. Locally and Bayesian optimal designs for models with a range of link functions are considered. Design criteria looked at include D-optimal, D-A-optimal and V-optimal designs, together with D-s-optimal designs where the control mortality parameter is regarded as a nuisance parameter. The range of prior distributions for the Bayesian optimal designs includes uniform, trivariate normal and a combination of a bivariate normal prior for the parameters of the underlying dose-response with an independent uniform prior for the control mortality parameter. (c) 2004 Elsevier B.V. All rights reserved.