Extreme value modelling of laboratory safety data from clinical studies

Abstract

Most clinical studies collect several safety-related laboratory variables. Generally, it is the extreme values of these variables that indicate potential safety issues. We illustrate the novel application of extreme value modelling to such data, with the aim of predicting the incidence of severe adverse drug reactions. By applying the methods to a clinical trial data set, we identify a dose-response relationship and use Bayesian techniques to identify a potential safety concern by making predictions from the fitted model, despite the small sample size.