Back to Annual Meeting Page		American Public Health Association 133rd Annual Meeting & Exposition December 10-14, 2005 Philadelphia, PA

Abstract #120589

Alan Hubbard, PhD, Biostatistics and Environmental Health, University of California, Berkeley, 735 University Hall, 2199 Addison Street, Berkeley, CA 94720, 510-642-8365, hubbard@stat.berkeley.edu

Marginal structural models (MSM) provide a powerful tool for estimating the causal effect of a treatment or risk variable on the distribution of a disease outcome in a population. These models, as originally introduced by Robins (Robins, 1998; Robins, 2000), model the marginal distributions of treatment-specific counterfactual outcomes, possibly conditional on a subset of the baseline covariates. Marginal structural models are particularly useful in the context of longitudinal data structures, in which each subject's treatment and covariate history are measured over time, and an outcome is recorded at a final time point. In addition to the simpler, inverse probability of treatment weighted estimators, more general (and robust) estimators have been developed.(Robins, 2000; Neugebauer and van der Laan, 2004; Yu and van der Laan, 2003). In many applications one is interested in modelling the difference between a treatment-specific counterfactual population distribution and the actual population distribution of the target population of interest. We focus on intervention models estimating the effect on an intervention in terms of a difference of means, ratio in means (e.g., relative risk if the outcome is binary), a so called switch relative risk for binary outcomes, and difference in entire distributions as measured by the quantile-quantile function. In addition, we provide a class of inverse probability of treatment weighted estimators, and double robust estimators of the causal parameters in these models. We illustrate the methodology with both simulations and a data analysis.

Learning Objectives: