In dose-response studies, researchers use regression to describe a dose-response
relationship and estimate a benchmark dose. Depending on an experimental design
with living animals, we can gain more information about the benchmark dose with a
fixed number of animals or the same amount of information with fewer animals. It is
not only a mathematical and statistical problem but also an ethical problem. We
demonstrate the potential benefit of an adaptive design which starts with a portion
of animals to update information then uses remaining animals to maximize
information about the benchmark dose. Simulation studies show that even a
4-phase locally optimal design is not robust, but a 2-phase Bayesian optimal
design (BOD) is more robust than a 1-phase BOD, and a 4-phase BOD is
not necessarily better than a 2-phase BOD. Since more phases require a
longer experimental time, a practical recommendation is to use a 2-phase
BOD.
