Lot consistency as an equivalence problem

One requirement for licensure of a vaccine in the United States is demonstration by the manufacturer of consistently produced lots of vaccine. Demonstration of consistency of manufacturing can be viewed as a multigroup equivalence problem. The standard statistical procedures for evaluating equivalence assume normally distributed data and define equivalence margins with respect to group means. As an alternative approach, we define a measure of the similarity among group distributions and their nonparametric estimators. Through computer simulations we explore the statistical properties of an equivalence test based on this estimator and compare them to the standard methods. Preliminary work suggests that a test of similarity can be useful in demonstrating equivalence when distributions are not normal or when there are differences in scale or shape. It appears to detect departures from equivalence that are not reflected by differences among group means.