Discriminating Drug Safety Profiles in Polytherapy

DESCRIPTION (provided by applicant): Statistical data mining applied to large databases of spontaneous post-marketing adverse event reports has demonstrated substantial utility in supporting pharmacovigilance efforts at government health authorities and pharmaceutical manufacturers. The goal of this project is to develop and validate advanced statistical methods for assessing the true association strengths between specific drugs and specific events, accounting for significant polytherapy within and across individual reports. These methods will support generating more accurate drug safety profiles that are corrected for confounding induced by the presence of other drugs throughout the database. Phase I demonstrated that logistic regression is a capable and generalizable method for providing intuitive and clinically reasonable safety signal scores in situations involving complex polytherapy where simpler non-model-based techniques are not reliable. Phase II focuses on design, prototyping, and full-scale development of a suitable high-performance, multivariate Bayesian logistic regression method; creation of an enhanced version of applicant's WebVDME safety data mining system incorporating a new regression-based computational core; and beta testing the new software with participation by academia, pharmaceutical companies, and FDA. The key Public health benefit is an improved ability to generate and interpret signals of potential drug safety problems, and to prioritize efforts to investigate and evaluate these signals.