Outlier Detection in Average Bioequivalence Study Using Method of Bootstrap

Abstract

A generic drug is defined as "a drug product that is comparable to brand/reference drug product in dosage form, strength, route of administration, quality and performance characteristics, and intended use." The average bioequivalence criterion stipulates that two drugs are to be considered bioequivalent when the 90% confidence interval, for the ratio of Geometric means (Test/Reference) is between 80% and 125%. The result is expressed as average bioequivalence (ABE), as it compares average values for both test and reference drug bio availabilities. The bioavailability of a drug is defined as the rate and extent to which the active drug ingredient from a drug product is absorbed and becomes available at the site of drug action.  A bioequivalence study data which contains a statistical outlier, may affect the inference of the study. In this paper, the Bootstrap method is applied for detection of an outlier subject or unusual subject in a two way crossover design in bioequivalence study as it is the most acceptable design in BE studies which separates intra subject variability from inter subject variability.  Five data sets are simulated, for a two way crossover design, assuming Bivariate Log Normal distribution for pharmacokinetic parameter Cmax. For each data set, statistical outlier is identified using method based on statistical theory of Bootstrapping. In addition, statistical inference using above method is compared with inferences drawn from more commonly used methods of outlier detection such as i) Maximum Normalized Residual Method (MNR) ii) Estimates Distance Method (ED) iii) Lund’s Method. It is observed that inferences drawn from the existing methods of Outlier detection viz. Estimates Distance, Lund’s Method and MNR (Maximum Normalized residual) method are not consistent. Lund’s method is less likely to detect “true” outlier whereas Estimates distance method is more likely to indicate a subject as an outlier in the absence of “true” outlier. The method based on bootstrapping is more robust in identifying the “true” outlier.