Abstract:
|
Application of disproportionality analysis (DPA) methods to spontaneous spontaneous reporting system databases has been recognized for its great value in detecting significant adverse drug event (ADE) associations. However, currently, none of the DPA methods can control or estimate the false positive. Here, we propose a three-component mixture distribution for the relative risk of drug-ADE association. Because one of the mixture components is assumed as the null, local false discovery rate (LFDR) can be estimated for the drug-ADE associations. An empirical Bayes estimation and inference procedure was developed for this mixture model. In particular, we further simplified the statistical inference of this mixture model through a conditional likelihood function. Our analysis of a subset of FDA Adverse Event Reporting System (FAERS) data demonstrated that LFDR-based ranking was more sensitive in selecting true drug-ADE signals then the existing DPA methods, while it was less sensitive to small sample size comparing to the other methods. Our simulation studies further evaluated the bias of LFDR estimation, and demonstrated the superior power of our method than the other DPA methods.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.