Statistical issues in individual and population bioequivalence

Abstract

The US Food and Drug Adm instration (FDA) has proposed new regulations that address the ‘prescribability’ and ‘switchability’ of new formulations of already- or nearly-approved drugs. These new criteria are known, respectively, as population and individual bioequivalence. The subject of bioequivalence has been fertile ground for statistical research over the past 3 decades, and the introduction of these new criteria has created new statistical problems to be explored. The focus of this investigation is the development of new tests for individual and population bioequivalence based on the generalized p-value (GPV ) methodology of Tsui and Weerahandi (JASA, 1989). We show that the GPV has good statistical properties and avoids a major shortcoming of other testing procedures for individual bioequivalence. In particular, the GPV is shown to be more powerful than any of the known testing procedures for population bioequivalence, including the test recommended by the FDA , for likely bioequivalence study designs. We also examine the asymptotic relative efficiencies of likely candidate designs for both individual and population bioequivalence, and compare these large-sample results with empirical calculations derived from power sim ulations. We also construct generalized confidence intervals for bioequivalence parameters and examine their properties.