Purpose The clinical endpoints used in TE are often highly variable and require a significant number of patients in order to have sufficient power to demonstrate equivalence. A major challenge in those trials is to obtain an accurate estimate of the coefficient of variation (CV) which is essential to calculate the sample size required. Some methods will be presented and compared in terms of overall power, protection of Type 1 error and average sample size required.
Methods Two methods were compared using simulations of normally distributed data. The first method (Potvin et al. AAPS Annual Meeting 2017) used a blinded re-estimation of sample size at 25% and 50% interim analyses. The second method follows the same principle as the Potvin’s method C (Potvin et al., 2008) and thus allow for sample size re-estimation at interim analysis and assessment of bioequivalence at both interim and final analysis. For both methods, the assumptions tested were TE of the test and reference products and superiority of T and R vs. Placebo. TE was based on the T/R ratio with the 90% confidence intervals (CI) for the ratio falling within the 80.00-125.00% limits. An analysis of covariance (ANCOVA) model was used. The corresponding 90% CIs was calculated using Fieller’s theorem. To ensure adequate study sensitivity, T and R should be both statistically superior in terms of differences to placebo (p<0.05, 2-sided) with regard to the primary endpoint. Superiority was assessed using a similar ANCOVA model as described above, for both pairs of compared treatments.
Conclusion Both methods control for inflation of Type 1 error and provide acceptable power. The second method is more flexible, since it allows for sample size re-estimation and possibility to stop for equivalence at interim assessment. These methods represent interesting and novel solutions for TE trials for which there is uncertainty in the initial sample size estimate. Interim analyses can be performed in order to re-adjust the sample size in a blinded fashion, while protecting the Type 1 error to a maximum of 5% and insuring an adequate power. The first method is done without adjusting the Type 1 error for interim looks, since blinded estimates of CVs are being performed whereas the second method requires an adjustment of the nominal Type 1 error.