Stagewise, group sequential experimental designs for quantal responses. One-sample and two-sample comparisons

Abstract

The use of stagewise, group sequential experimental designs with dichotomous responses in toxicity or drug screening programs is discussed. Such designs represent a compromise between the standard, fixed sample size designs and fully sequential designs. Stagewise group sequential designs place specified numbers of animals on test at each stage, up to a maximum number of stages. The greatest increases in sample size efficiency occur with small numbers of stages, particularly when going from one stage to two. Two-stage designs can result in a 15 to 20 percent reduction in average sample size. Five-stage designs can result in a 30 to 40 percent reduction in average sample size, with no appreciable decrease in Type 1 error or power. Examples of the efficiencies that arose in actual screening programs are given. This paper demonstrates that the routine use of stagewise, group sequential designs in standardized screening protocols can result in substantial savings in animal use with virtually no sacrifice of statistical sensitivity.