All Times EDT
Keywords: Dissolution Profile, Variance Heterogeneity, Multivariate Statistical Distance
In vitro dissolution testing is critical for drug quality control and bioequivalence assessment. The similarity factor f2 is the standard approach for comparing low-variable dissolution profiles. For highly variable dissolution profiles, multivariate model-independent procedures are recommended by FDA and EMA guidelines, with Multivariate Statistical Distance (MSD) being the most popular method. However, these approaches have been developed with the underlying assumption of homogeneous test-reference variances. For example, a common covariance matrix is assumed for test and reference batches in MSD, which, however, may be suboptimal if the true variabilities are different between test and reference batches. To our knowledge, no research has been conducted to systematically compare and discuss the appropriateness of the available methods when the homoscedasticity assumption is violated in the dissolution data. And there are no multivariate model-independent methodologies being developed yet to account for the unequal variances in this setting.
Motivated by an inhouse dissolution dataset where the test and reference batches have unequal variabilities, we studied the performances of several existing approaches in such scenario. We also employed a distance metric (other than the commonly used Mahalanobis distance) that can account for unequal covariances of the test and reference batches. We then compared its performance with existing methods under various simulation settings. In addition, we revisited the definition of “equivalence” and refined the process of dissolution testing to not only consider the difference in the profile means, but also account for the variance heterogeneity in a systematic way. Based on our findings, we conclude that ignoring the unequal variabilities between test and reference batches may lead to misleading testing outcomes, and we propose a new dissolution testing process to handle this issue.