Methodology Response

• How to make a clinical trial most informative with respect to certain trial objectives (for example estimation of the MED)? • In this talk: Combine ideas from optimal design theory and Bayesian statistics, to obtain adaptive designs and acknowledge underlying model uncertainty. 3 Response Notation and framework • Assume that the primary endpoint Y is observed for k parallel groups corresponding to doses d1,..., dk (d1 typically placebo) and Yij = f(di,θ)+ij, ij ind ∼ N (0, σ2) j = 1,..., ni, i = 1,..., k • Here f(di,θ) denotes a (typically) non-linear dose-response model function (e.g. logistic or the sigmoid Emax model) • Model uncertainty regarding the correct model f (sometimes even monotonicity questionable). ↪ → Use candidate set of modelsM = {M1,...,MM} with model functions f1(d,θ1),..., fM(d,θM). 4 Response Notation and framework II • Let ∆ denote clinically relevant effect; we aim at estimating the MED, i.e., the smallest dose achieving ∆ on top of placebo: MED = min d∈(d1,dk] {f(d,θ)> f(d1,θ) + ∆} • How to choose the allocation weights w = (w1,..., wk) ′ of the total number of patients N to the dose-groups of the study to obtain most information of the MED