Targeted Data Adaptive Estimation of the Causal Dose Response Curve

• 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): Ivan Diaz*+ and Mark van der Laan
• Companies: University of California at Berkeley and University of California at Berkeley School of Public Health

Abstract

The definition of a causal inference question under a non-parametric structural equation model requires the specification of a set of intervention variables usually called treatments or exposures. In spite of the fact that a large portion of the treatment variables are of a continuous nature, causal inference methods have centered the attention on the definition and estimation of parameters for either binary or categorical exposures. In this work we discuss the estimation of the causal dose-response curve (CDRC) for continuous treatments. The most widely used method for estimating the CDRC is given by marginal structural models (MSMs), and requires an a priori specification of a functional form (i.e., a model). MSMs estimates are valid if 1) the model is correct, or 2) the parameter is defined as a projection of the true CDRC on the model. However, MSMs are not satisfactory when the research interest relies on estimating the true CDRC. We present an algorithm that uses cross-validated targeted maximum likelihood estimators (CV-TMLE) of the risk of an estimator among a library of candidate estimators of the CDRC.