Empirical Comparison of Different Drug Safety Signal Detection Assays Based on Incident User Designs

• Activity Number: 50
• Type: Invited
• Date/Time: Sunday, August 1, 2010 : 4:00 PM to 5:50 PM
• Sponsor: ENAR
• Author(s): Maurice Alan Brookhart*+ and Eric Brinsfield
• Companies: Brigham and Women's Hospital/Harvard Medical School and SAS Institute
• Keywords: epidemiology ; pharmacoepidemiology ; drug safety

Abstract

We describe a general approach for detecting drug safety signals in large healthcare databases based on an incident user design combined with propensity score (PS) methods to adjust for confounding by baseline covariates. We empirically compare various drug safety signal detection assays (tests) based on different implementations of this general approach. The implementations vary in how the cohort is defined (cohort entry occurring at first use vs. a run-in period); how the PS is specified; whether the PS is truncated to remove non-overlap; how the PS is used (inverse-probability weighting vs. SMR weighting); and how follow-up and censoring are handled. The methods are compared in the setting of known or strongly suspected drug-outcome relations. In several large healthcare databases across a range of drug-outcome pairs, we report the sensitivity and specificity of the different assays