Multivariate Analysis of Adverse Events

Abstract

Clinical trial reports typically provide univariate data on adverse events in the form of rates. Little or no consideration is given to providing data on syndromes or constellations of adverse events in clinical trials. We describe three methods for analyzing adverse events; these methods focus on constellations of events within the same patient. A computer algorithm enumerates the constellations of adverse events in the data and counts the number of patients in each constellation. The proposed algorithm also preserves the identifiers of patients in the constellation and other covariates for further analysis. A log-linear model is used to estimate the magnitude of association of two or more adverse events by analyzing the count of patients in each of the possible subsets of the constellation. These methods make a unique contribution to the determination of safety in the development of new therapeutic agents.