Statistical tests for detecting reference product change in biosimilar studies

Abstract

For the biosimilarity assessment between a test product and a reference product, the US Food and Drug Administration (FDA) recommends a stepwise approach for obtaining totality-of-the-evidence in support of regulatory approval of the submission. The stepwise approach starts with conducting an analytical similarity assessment of certain critical quality attributes that are relevant to clinical outcome; however, potential drift in mean response and/or variability associated with the reference product over time may be observed. When there is a drift, it is of interest to determine which lots should be used for analytical similarity assessments. In this article, statistical tests for detecting possible drifts in mean and/or variability are derived. In summary, a statistical method is proposed that performs well to detect drift in the mean/variability of biological products while controlling false positive rates and maintain desired true positive rates even when the number of lots is large.