Large-Scale Regression-Based Pattern Discovery in International Adverse Drug Reaction Surveillance

This paper demonstrates the first use of shrinkage logistic regression as a pattern discovery method in international adverse drug reaction surveillance. This novel method is compared to bivariate pattern discovery, the standard approach in the application domain. Our results show that regression can eliminate false positives and false negatives due to the impact of other covariates, and that it can retrospectively detect established drug safety issues earlier than bivariate pattern discovery. However, regression cannot completely replace bivariate methods, for two reasons: its failure to identify some established drug safety concerns as early as the routinely used bivariate measures, and the relative intransparency of the procedure to estimate the regression coefficients. This suggests both approaches be used in parallel.