Search

In the European Union, bioequivalence (BE) for narrow therapeutic index (NTI) drugs is currently demonstrated when the 90% confidence interval for the ratio of the population geometric means of the test and reference products for AUC, and in some cases for Cmax, falls within the acceptance range of 90.00% to 111.11%. However, meeting this requirement results in an increased difficulty of demonstrating BE and a need for clinical trials with larger subject sample sizes, especially for medium-to-high variability drugs. To address this challenge, a scaled average BE based on the reference product within-subject variability for narrowing the acceptance range of NTI drugs was recently proposed. However, this approach showed increased type I error (T1E), especially close to the cut-off point between the unscaled and scaled portions of the method. Based on simulations, this limitation can be overcome by predefining the protocol the path to be followed: either the fixed 90.00–111.11% acceptance range approach or the previously proposed scaled average BE approach with a slight adjustment of the one-sided significance level α to 0.042 for a 2 × 3 × 3 partial replicate design and without a lower cut-off point. This results in a mixed approach allowing to reduce the sample size whilst not inflating the T1E. [ABSTRACT FROM AUTHOR]

In the European Union, bioequivalence (BE) for narrow therapeutic index (NTI) drugs is currently demonstrated when the 90% confidence interval for the ratio of the population geometric means of the test and reference products for AUC, and in some cases for Cmax, falls within the acceptance range of 90.00% to 111.11%. However, meeting this requirement results in an increased difficulty of demonstrating BE and a need for clinical trials with larger subject sample sizes, especially for medium-to-high variability drugs. To address this challenge, a scaled average BE based on the reference product within-subject variability for narrowing the acceptance range of NTI drugs was recently proposed. However, this approach showed increased type I error (T1E), especially close to the cut-off point between the unscaled and scaled portions of the method. Based on simulations, this limitation can be overcome by predefining the protocol the path to be followed: either the fixed 90.00–111.11% acceptance range approach or the previously proposed scaled average BE approach with a slight adjustment of the one-sided significance level α to 0.042 for a 2 × 3 × 3 partial replicate design and without a lower cut-off point. This results in a mixed approach allowing to reduce the sample size whilst not inflating the T1E.