ABSTRACT
In this chapter we discuss some nontraditional applications of optimal experimental design techniques. In Section 9.1 methods of selecting informative variables are considered as in Fedorov et al. (2006) [138]. In Section 9.2 we discuss and compare several types of adaptive designs, including those that allocate each new patient (or a cohort of patients) to the dose currently viewed as the best one, for example, the seemingly most efficacious. We call this type of allocation “best intention” designs. As it turns out, similar to other areas of life, the best intentions may lead to rather disastrous situations. Theoretical results originating in research on optimal/automatic control strategies suggest that the best intention approach needs modifications to insure convergence to the “best” dose. Monte Carlo simulations presented in Section 9.2 provide corroborating evidence that caution is needed: even for very simple models, best intention designs may converge to a wrong dose with non-zero probability.
