Novel Methods for Estimating Causal Effects Under an Additive Hazards Model

• Activity Number: 211
• Type: Invited
• Date/Time: Monday, July 30, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Caleb Miles*+ and Eric Tchetgen and Stijn Vansteelandt
• Companies: Harvard University and Harvard School of Public Health and Ghent University

Abstract

We introduce a number of new analytic techniques for making inferences about a treatment effect under a partially additive hazards model while appropriately accounting for confounding by a large number of risk factors. The new estimators demonstrate a variety of dimension reduction approaches to incorporate a high dimensional vector of confounders in a survival context. A propensity score approach is introduced, in which the treatment mechanism is modeled parametrically and is subsequently used to recover a nearly unbiased estimate of a treatment effect on the hazards difference scale provided modeling error is absent. The approach holds appeal in that it obviates the need for explicit modeling of the association between confounders and failure time. A variety of extensions will be discussed including the development of doubly robust estimators and extensions to the time-varying setting.