A0660
Title: Optimal design for parameters estimation in generalized linear models with treatment-by-covariate interactions
Authors: Rosamarie Frieri - University of Bologna (Italy) [presenting]
Alessandro Baldi Antognini - University of Bologna (Italy)
Maroussa Zagoraiou - University of Bologna (Italy)
Abstract: In an experiment aimed at comparing multiple treatments, especially in the clinical context, subject covariate information is often available and should be incorporated not only in the data analysis at the end of the study but also into the treatment allocation scheme. With the advances in medical research and the advent of precision medicine, scientists have identified many new covariates/biomarkers that may be linked with certain diseases and could strongly influence patients' responses to treatments. The D- and A-optimal designs are derived for parameter estimation for generalized linear models with treatment-by-covariate interactions. The optimal designs require a set of equality constraints involving i) the unknown model parameters and ii) the subject covariate values. Since such conditions are not directly attainable, the optimum can be achieved sequentially by adopting a novel class of covariate-adjusted response-adaptive randomization, which aims at minimizing, at each step of the sequential procedure, the Euclidean distance between the current allocation and the optimum. The performance of the proposed approach in terms of estimation efficiency is assessed both theoretically and through an extensive simulation study.
