Bayesian Causal Inference Analyses with Unmeasured Confounders

• Activity Number: 441
• Type: Contributed
• Date/Time: Tuesday, August 2, 2016 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Negar Jaberansari* and Bin Huang
• Companies: University of Cincinnati and Cincinnati Children's Hospital Medical Center
• Keywords: Bayesian inference ; Causal inference ; Comparative clinical effectiveness ; Observational studies ; Propensity score

Abstract

Bayesian analyses have the advantage of combining data from different sources. We consider a setting where some important confounders are not observed in a large primary data. However, we also have access to a smaller but richer dataset, which offers information on the unmeasured confounders. This is the same setting considered previously (McCandless et al., 2012). Here we investigate McCandless et al. method under different modeling strategies using simulation studies. Our simulation results suggest improved performance in terms of estimating averaged causal treatment effect and predicting potential outcomes, when better modeling strategies are used. We apply the method to analyze a large patient network registry data. The study is aimed to evaluate the comparative effectiveness of the use of combination of biological and non-biological disease modification anti-rheumatic drug (DMARD) vs. non-biological DMARD only, in children with newly onset of juvenile idiopathic arthritis.