A note on statistical methods for assessing therapeutic equivalence.

The two one-sided tests procedure and the confidence interval approach are two commonly used statistical approaches for testing therapeutic equivalence or assessing bioequivalence. However, some confusion arises. For example, what is the difference between the two approaches, given the fact that in some cases the two approaches produce the same test? Should we use level 1-alpha or 1-2alpha when applying the confidence interval approach? When different confidence intervals are available, which confidence interval should be used? The purpose of this paper is to clarify this confusion. It is shown that the approach of using 1-alpha confidence intervals produces level alpha tests, but the sizes of these tests may be smaller than alpha, and that the use of 1-2alpha confidence intervals generally does not ensure that the corresponding test be of level alpha, although there are exceptional cases. The sizes of several tests obtained using different confidence intervals are also evaluated. Copyright 2002 Elsevier Science Inc.