Simple Instrumental Variable Regression for Censored Data with Exposure-Dependent Censoring

• Activity Number: 10
• Type: Invited
• Date/Time: Sunday, August 9, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Kwun Chuen Gary Chan* and Dylan Small and Yijian Huang
• Companies: University of Washington and University of Pennsylvania and Emory University
• Keywords: accelerated failure time model ; additive hazards model ; instrumental variable ; exposure-dependent censoring

Abstract

Two stage least squares for uncensored linear models can be extended to censored survival outcome, at the expense of a stronger censoring assumption that precludes exposure-dependent censoring. In order to incorporate this practically important situation, complicated methodologies has been proposed that typically requires additional modeling assumptions. We propose a simple two-stage instrumental variable method that can yield consistent estimators under exposure-dependent censoring. A simple asymptotic variance estimator is also available due to orthogonality between estimating equations.