CMStatistics 2016: Start Registration
View Submission - CMStatistics
B0706
Title: Doubly-robust estimation of semiparametric additive hazards models Authors:  Oliver Dukes - Ghent University (Belgium) [presenting]
Stijn Vansteelandt - Ghent University and London School of Hygiene and Tropical Medicine (Belgium)
Torben Martinussen - University of Copenhagen (Denmark)
Abstract: Additive hazards models are becoming increasingly popular in survival analysis. Their parameters are easily interpretable in terms of relative survival risks, and are moreover collapsible, which makes the development of mediation and instrumental variable approaches more manageable. Estimation of the effect of an exposure in such models typically demands adjustment for a high-dimensional covariate. Standard estimation is then less desirable, as misspecification of the effect of these covariates may induce large bias in the exposure effect estimate. To overcome this, we consider a novel class of semiparametric additive hazards models which leave the effects of baseline covariates unspecified. We derive the efficient score for the exposure effect in these models, which we assume to be constant over time. We argue that using this score may not be advisable in practice, as it requires a model for the conditional distribution of the exposure, given covariates. We therefore derive the efficient score in a subclass of estimators, which require this conditional distribution merely to be correctly specified up to the mean. We show that the resulting estimator enjoys a double-robustness property. Finally, we give guidance on efficiency, and discuss how the proposed estimators lay the foundations for G-estimation of structural nested cumulative survival models for the effect of time-varying exposures.