Practical pharmacovigilance analysis strategies

Abstract

Purpose: To compare two recently proposed Bayesian methods for quantitative pharmacovigilance with respect to assumptions and results, and to describe some practical strategies for their use.

Methods: The two methods were expressed in common terms to simplify identifying similarities and differences, some extensions to both methods were provided, and the empirical Bayes method was applied to accumulated experience on a new antihypertensive drug to elucidate the pattern of adverse-event reporting. Both methods use the logarithm of the proportional risk ratio as the basic metric for association.

Results: The two methods provide similar numerical results for frequently reported events, but not necessarily when few events are reported. Using a lower 5% quantile of the posterior distribution gives some assurance that potential signals are unlikely to be noise. The calculations indicated that most potential adverse event-drug associations that were well-recognized after 6 years of use could be identified within the first year, that most of the associations identified in the first year persisted over time. Other insights into the pattern of event reporting were also noted.

Conclusion: Both methods can provide useful early signals of potential drug-event associations that subsequently can be the focus of detailed evaluation by skilled clinicians and epidemiologists.