Detection of Outliers in Bioequivalence Studies Data Analysis with Williams Design

Abstract

Background: Drug Regulatory agencies all over the world generally discourage exclusion of outliers in a BE (BE) study; on the other hand in routine bio-statistical work we take these into the account. If the decision rules for identifying the outliers are clearly mentioned before the start of the study and laid down in protocol by the responsible biostatistician in collaboration with clinicians, the problem of outliers can be dealt smartly without jeopardizing the whole study for redoing. The purpose of this article is to introduce procedure for reliably detecting outlier subject(s) with Williams design. Experimental: Literature review reveals many different methods for the detection of outlier values in BE studies; most of them are for BE of two treatments. For BE studies with more than two treatments use of Williams design seems imperative; but inclusion and deletion of outlying subjects may lead to profound effect on conclusion of BE which in turn may be dangerous for the health. The suggested method is an adjustment to a previously introduced method using exploratory data analysis technique such as principle component analysis and Andrews curves.