Sample size determination for equivalence assessment with multiple endpoints

Equivalence assessment between a reference and test treatment is often conducted by two one-sided tests (TOST). The corresponding power function and sample size determination can be derived from a joint distribution of the sample mean and sample variance. When an equivalence trial is designed with multiple endpoints, it often involves several sets of two one-sided tests. A naive approach for sample size determination in this case would select the largest sample size required for each endpoint. However, such a method ignores the correlation among endpoints. With the objective to reject all endpoints and when the endpoints are uncorrelated, the power function is the production of all power functions for individual endpoints. With correlated endpoints, the sample size and power should be adjusted for such a correlation. In this article, we propose the exact power function for the equivalence test with multiple endpoints adjusted for correlation under both crossover and parallel designs. We further discuss the differences in sample size for the naive method without and with correlation adjusted methods and illustrate with an in vivo bioequivalence crossover study with area under the curve (AUC) and maximum concentration (Cmax) as the two endpoints.