An approach to select the appropriate statistical method for testing bioequivalence

Since most bioavailability studies are usually done with only a limited number of volunteers (usually 10-30), the statistical properties of the calculated bioavailability parameters are not well defined. The established statistical methods to test bioequivalence are usually based on either the assumption of normality or a symmetrical distribution. However, the decision of which method to apply, depends primarily on the distributional assumption of the data. In this study, an approach is followed where the small data base of a limited number of volunteers is expanded by adding pseudo-volunteers by "bootstrap" simulations. From such a larger data base it is easier to determine the statistical distributional properties of bioavailability parameters, which in its turn leads to the identification of an appropriate statistical method. With more certainty on which statistical method to apply, the original data can be used more effectively in testing for bioequivalence. In this paper, comparisons are made between the distributions of bioavailability parameters of an actual 60-volunteer study and those of two simulated data sets. Each such data set contained a random sample of 10 volunteers each (from the 60 volunteers), together with 50 pseudo-volunteers. These 50 volunteers were simulated from the random sample of 10 real volunteers. Good correspondences were obtained when comparing these two data sets with the original data, which indicated the validity to use this approach in bioavailability studies where a small number of volunteers had been used. This method proved useful to define the distributional properties for a relative small number of parameter-values available.