An ANCOVA Approach to Reducing the Residual Variance and Sample Sizes Required for Parallel Design Bioequivalence Studies
*Pina D'Angelo, Novum Pharmaceutical Research Services  Charles DiLiberti, Montclair Bioequivalence Services, LLC 


Parallel design bioequivalence (BE) studies, which are commonly used for long half-life drugs, often pose substantial challenges for sponsors in that they employ no replication, and thus cannot utilize the popular reference-scaled average bioequivalence (RSABE) method to control sample size. Furthermore, their sample sizes are dictated by between-subject variance, which is often substantially larger than the within-subject variance that dictates the sample sizes of crossover design studies. As a result, parallel design studies may easily require hundreds of subjects to achieve reasonable power. While this problem may arise for some solid oral dosage forms, it is a common problem for long-acting injectable formulations. Under some circumstances, the apparent terminal elimination rate constant (kel) is strongly correlated with the primary pharmacokinetic (PK) parameters AUC and Cmax, and yet is independent of the treatment effect. Therefore, incorporating ln(kel) as a covariate in an analysis of covariance (ANCOVA) of the ln-transformed PK parameters, i.e., ln(AUC) and ln(Cmax), provides an opportunity to dramatically reduce the residual variance and consequently, the sample size as well, for parallel design studies. Incorporating other, subject-related factors and covariates into the ANCOVA model can further reduce the residual variance and required sample size. The conditions under which this type of ANCOVA approach is valid, as well as practical considerations, such as how best to address missing covariate values, will be discussed and illustrated with simulations and actual case studies.