Vector Analysis to Detect Hepatotoxicity Signals in Drug Development

Hepatotoxicity is a leading cause of discontinuing drugs in development and withdrawing drugs from clinical use. Most withdrawals are due to idiosyncratic hepatotoxicity which usually escapes detection in clinical trials because of its rarity. We investigated the potential of vector analysis of liver function tests to detect an early signal of idiosyncratic hepatotoxicity in clinical trials using data with an obvious drug-induced liver response. Serum samples collected serially during randomized, placebo-controlled trials of a compound that was subsequently discontinued because of hepatotoxicity were analyzed for alanine aminotransferase, aspartate aminotrans-ferase, γ-glutamyltransferase, and alkaline phosphatase. Vectors for combinations of tests were computed and displayed in two or three dimensions over the 6-week course of the trials. At baseline, short vectors were clustered within the normal range in the active- and placebo-treatment groups. By the third week, several vectors in the active-treatment group had elongated inside of the normal range and moved outside in the fourth week. Thereafter, the vectors began to return to the baseline range. These results suggest that vector analysis may be useful in detecting earlier or clinically obscure signals of hepatotoxicity in clinical trials. Learning Objectives Upon completion of this article, participants should be able to: Describe how vector (multivariate) analysis is used to detect hepatotoxicity in clinical trials Recognize why the usual univariate statistical methods of detecting abnormalities in clinical laboratory tests may miss important signals Conceptualize how motion in multidimensional data can define a signal that is not otherwise observable Explain why the usual normal limits may be missing hepatotoxicity signals both in space and time Target Audience This article is designed for clinicians, biostatisticians, drug safety analysts, epidemiologists, toxicologists, and pharmacologists.