Title: Detecting adverse drug reactions from pharmacovigilance databases
Authors: Kun Liang - University of Waterloo (Canada) [presenting]
Abstract: The World Health Organization and many countries have built pharmacovigilance databases to detect potential adverse reactions due to marketed drugs. Although a number of methods have been developed for early detection of adverse drug effects, the vast majority of them do not consider the multiplicity arising from testing thousands drug and adverse event combinations. We first derive the optimal statistic to maximize the power of detection while maintaining proper error rate. We then propose a nonparametric empirical Bayes method to estimate the optimal statistic and demonstrate its superior performance through simulation. Finally, the proposed method is applied to the pharmacovigilance database in the United Kingdom.