Comparison of Outlier Detection Methods in Crossover Design Bioequivalence Studies

Abstract

The significance of bioequivalence (BE) studies is rising due to large scale production and utilization of generic products all over the world. The correct identification of outlying data in BE studies is substantial for deciding two products either bioequivalence or bioinequivalent. For the detection of outliers in BE studies with the crossover designs different methods have been suggested in the literature. In the present work, we compared three outlier detection tests; (i) the Likelihood distance (LD) test (ii) the estimated distance (ED) test and the principal component analysis (PCA) test. In this work, the PCA test has been first time compared with the LD and ED test. For the purpose of comparison, we used two-way and three-way BE crossover data sets on linear and logarithmic scales. During the course of work it was found interesting and note-worthy that the performances of the ED and PCA tests in the sense of outlier detection are better than the LD test and this performance persists even for the log-transformed data. The results of our simulation study also indicated that the performance of the ED test for outliers’ identification is better than the other two tests.