By Chen-Wei Oh

Published on June 24, 2024

Introduction

The handling of outliers in bioequivalence (BE) studies is a critical aspect of ensuring the accuracy and reliability of study results. Various regulatory agencies have developed guidelines to address the identification and management of outliers, with differences in approaches and acceptance criteria. This article reviews the guidelines from ASEAN States, Australia, Canada, the East African Community, Egypt, the Eurasian Economic Union, the European Economic Area, New Zealand, the Russian Federation, WHO, Health Canada, and ANVISA, highlighting historical and contemporary methods used in outlier detection and management.

Historical Context

Historically, the Lund test [1] was a popular method for outlier identification. However, it has several limitations, such as:

• Insignificance if outliers have opposite signs (low/high values).
• Complex modifications.
• Declining regulatory acceptance.

Robust, nonparametric methods have gained traction as they address some limitations of parametric methods. Evaluations of approximately 300 studies using nonparametric methods revealed that the outcomes are often similar to parametric methods after outlier exclusion [2].

ASEAN and Global Guidelines

Applicants in ASEAN States, Australia, Canada, the East African Community, Egypt, the Eurasian Economic Union, the European Economic Area, New Zealand, the Russian Federation, and WHO are required to justify that the calculated intra-subject variability is a reliable estimate and not the result of outliers. This requirement reflects ongoing concerns about the handling of outliers, which remains an open issue.

Health Canada (June 2018)

Health Canada specifies that outlier identification must be performed before assessing the study for bioequivalence (i.e., before calculating the point estimate (PE) and confidence interval (CI)), as detailed in the Statistical Analysis Plan. The criteria for identifying outliers include:

• Outliers are identified if the studentized residual is outside ±3.
• Identified outliers may be removed from the analysis data set.
• The removal process is independent of treatment, and a subject identified as an outlier should be consistently considered an outlier for all pharmacokinetic (PK) metrics.

Health Canada's rules for handling outliers are as follows:

• ≤5% of subjects may be removed unless the study involves 20 or fewer subjects, in which case only one subject may be removed.
• If >5% of subjects are identified as outliers, the absolute studentized residuals are ranked, and the top 5% of subjects with the highest residuals are removed.