Feder P I, Hobson D W, Olson C T, Joiner R L, Matthews M C
Battelle, Columbus, OH 43201.
Neurosci Biobehav Rev. 1991 Spring;15(1):109-14. doi: 10.1016/s0149-7634(05)80101-0.
A principal design objective of many dose-response studies is to estimate extreme percentiles of a dose-response distribution, e.g., the ED95 dose for a particular drug therapy, as precisely as feasible using the smallest number of experimental subjects possible. Such a design requirement necessitates that allocation of subjects to drug doses be carried out in a stagewise fashion to maximize the information obtained from each subsequent experimental observation in light of what has previously been determined. This paper describes and illustrates specialized methods and associated computer programs to evaluate, on a stagewise basis, the anticipated relative sensitivities of alternative experimental plans in the case of dichotomous responses. Following each stage of experimentation, the current estimates of the dose-response distribution parameters, as well as the uncertainties in these estimates, are updated and are used to assign subjects to experimental dose levels for the next stage of testing. Competing dose allocations are compared with respect to anticipated improvement in estimation precision. The adoption of such a stagewise dose allocation strategy is illustrated by example.
