Title: Data Mining for Drug Safety: Statistical Analyses of Spontaneous Reports and Clinical Safety Data
Speaker: William DuMouchel, Lincoln Technologies Division of Phase Forward, Inc.
Date: February 12, 2007 12:00 - 1:30 pm
Location: DIMACS Center, CoRE Bldg, Room 431, Rutgers University, Busch Campus, Piscataway, NJ
First I will review methods for analysis of databases of spontaneous adverse drug reaction reports and then focus on some new methods for analysis of clinical safety data. Analysis methods for adverse event frequencies in clinical trials are less developed than are analyses for efficacy. Accounting for multiple comparisons due to the many types of adverse events under observation is problematical, as is the question of how to group event types that seem medically related. Empirical Bayesian approaches to two problems will be developed and illustrated with examples. First, a clustering method for finding potential syndromes related to treatment is based on finding sets of events for which all pairs within the set have enhanced presence within the treatment group compared to the control group. Second, a hierarchical model for parallel logistic regressions allows analyses of medically related events to "borrow strength" from each other and also permits subgroup analyses that are designed to be resistant to the multiple comparisons fallacy.
see: DIMACS Computational and Mathematical Epidemiology Seminar Series 2006 - 2007