2021 FDA Science Forum
Use of AUC as a Discriminating Measure for Identifying Data Separation of PK/PD Measurements in Bioequivalence Studies
- Authors:
- Center:
-
Contributing OfficeCenter for Drug Evaluation and Research
Abstract
In bioequivalence (BE) studies, data separation observed in the bioanalytical order of pharmacokinetics (PK) or pharmacodynamic (PD) measurements could be one of the issues in data analysis. Recently, there were cases of BE study data showing potential PK measurement separation along with observed values’ bioanalytical order and this type of issue becomes a challenge for researchers to provide reasonable explanations of the phenomenon. In this research project, the receiver operating characteristic (ROC) curve was applied to see how likely a separation between two groups could occur as a tool for analyzing a classification performance in binary classification (two groups) problems. Specifically, the area under the ROC curve (AUC) was used to evaluate a classification performance or a discrimination degree of two groups, where it took probabilities of an observation belonging to each group into consideration.
In this problem, a classification rule (i.e., ratio of test to reference means ? 1) was defined to allocate an individual measurement to one of the two classes and the corresponding AUC was computed accordingly, where a larger AUC value (0.5?AUC?1) indicated a stronger separation. A simulation study was conducted based on a case example to evaluate how likely its AUC could happen compared to all possible AUCs of the considered case example. The AUC based analysis can help evaluate BE study data with any potential data separation of PK/PD measurements along with observed values’ bioanalytical order.
