Inverse sampling and triangular sequential designs to compare a small proportion with a reference value

Abstract

Inverse sampling and formal sequential designs may prove useful in reducing the sample size in studies where a small population proportion p is compared with a hypothesized reference proportion p0. These methods are applied to the design of a cytogenetic study about chromosomal abnormalities in men with a daughter affected by Turner's syndrome. First it is shown how the calculated sample size for a classical design depends on the parameterization used. Later this sample size is compared with the required sample size in an inverse sampling design and a triangular sequential design using four different parameterizations (absolute differences, log-odds ratio, angular transform and Sprott's transform). The expected savings in sample size, when the alternative hypothesis is true, are 20% of the fixed sample size for the inverse sampling design and 40% for the triangular sequential design.