ABSTRACT

Here we will describe the four methods proposed by Potvin et al. (2008). For each method it is assumed that an interim analysis will be done after a prespecified number of subjects (n1) have completed both periods. At the interim, the unblinded log-transformed data are analyzed and an estimate of the treatment difference and within-subject variance is obtained. Depending on the method, one or more of the following are done: the TOST procedure is applied; the power of the study is recalculated; the sample size is recalculated, and the study stops or continues. If the study continues, with or without a sample size re-estimation, the TOST procedure is applied to the complete set of data from the study. Potvin et al. (2008) compared the methods using simulation under the assumption that the true ratio of the Test and Reference means on the natural scale was either 0.95 (a value consistent with approximate ABE) or 1.25 (i.e., a value consistent with the null hypothesis).