A Doubly Robust Estimator of the Attributable Benefit of a Dynamic Treatment Regime

• Activity Number: 296
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Jason Brinkley*+ and Anastasios Tsiatis
• Companies: East Carolina University and North Carolina State University
• Keywords: Attributable Benefit ; Treatment Regime ; Causal Inference ; Doubly Robust

Abstract

It can sometimes be the case that there is no general consensus for treating patients with a particular disease or disorder. In such cases the best strategy may be to devise a regime or policy that individualizes treatment based on patient covariates and risk factors. Attributable benefit measures the proportion of poor outcomes that could have been prevented had such a policy been implemented. When estimating these quantities from observed data, model misspecification creates bias in the estimation of both the treatment policy and that policy's attributable benefit. Using the idea of potential outcomes and notions from causal inference we propose an estimator for this measure that augments an outcome regression model with a propensity score model for treatment. The result is an estimate of the regime's attributable benefit that is doubly robust in terms of model misspecification.