Fieller's confidence intervals for the ratio of two means in the assessment of average bioequivalence from crossover data.

The two-period crossover design is the most commonly used study design for bioequivalence of one test formulation to be assessed in comparison to one reference formulation. Consequently, in this paper, all derivation is based on this particular design. It is assumed that for the underlying statistical model the usual assumptions of normality and additivity are satisfied on the original scale of measurement and that it is wanted to base the assessment of average bioavailability on the ratio of the unknown population means for the test and reference formulation. The purpose of this paper is to illustrate that it is reasonable to assume a uniform covariance structure for the two-period crossover design, because the demand of equal variability in bioavailabilities, in addition to equal average bioavailabilities, for the reference and test formulation makes the assumption of uniform covariance structure very realistic, and also because the properties of a decision rule based upon a Fieller's confidence interval under a uniform covariance structure are competitive with those of the corresponding rule based on a general covariance structure.