Improving Bayesian Methods in Pharmacovigilance with Drug Hierarchies

• Activity Number: 80
• Type: Contributed
• Date/Time: Sunday, July 29, 2012 : 4:00 PM to 5:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Trevor Shaddox*+ and Marc A Suchard
• Companies: University of California at Los Angeles and University of California at Los Angeles
• Keywords: pharmacovigilance ; hierarchy ; Bayesian

Abstract

Adverse drug events (ADEs) are a serious public health risk. The advent of ADE data has inspired post-approval early detection methods like the Bayesian self-controlled case series (BSCCS) regression model. However, this model previously did not account for the relatedness of drugs. We incorporated the Anatomical Therapeutic Chemical Classification drug hierarchy into the cyclic coordinate descent algorithm used by the BSCCS model. Specifically, we either modified the prior probabilities or used the lasso method when either normal or Laplacian priors, respectively, were placed over the regression coefficients. A greedy multidimensional cross-validation algorithm was developed for parameter selection. Using simulated data, the adjusted normal prior probabilities model showed enriched signal from drugs in dangerous classes relative to innocuous drugs. The lasso model shrank the estimated risk of innocuous drugs to zero with little effect on dangerous drugs. The greedy parameter estimation algorithm yielded a 10-fold speed increase versus an exhaustive parameter search. This research demonstrates that leveraging drug hierarchies is feasible and beneficial.