Robust designs in non-inferiority three arm clinical trials with presence of heteroscedasticity

In this paper, we describe an adjusted method to facilitate a non-inferiority trial by a three-arm robust design. Because local optimal designs derived in Hasler et al. [2007] require knowledge about the ratios of the population variances and are not necessarily robust with respect to possible misspecifications, a maximin approach is adopted. This method requires only the specification of an interval for the of variance ratios and yields robust and efficient designs. We demonstrate that a maximin optimal design only depends on the boundary points specified for the intervals of the variance ratios and receive numerical and analytical solutions. The derived designs are robust and very efficient for statistical analysis in non inferiority three arm trials.