Biostatistics Seminar
A word of caution regarding stabilized weights when using marginal structural models to estimate the effect of exposure history
Denis Talbot, PhD
Biostatistics Professor, Department of Social and Preventive Medicine, Laval University
Abstract:
Marginal structural models (MSMs) are commonly used to estimate the causal effect of a time-varying exposure in presence of time-dependent confounding. These models have gained vast popularity over the recent years, in part due to their ease of implementation. Indeed, the most common implementation of MSMs consists of fitting a weighted outcome model (e.g. a linear regression), where observations are weighted according to their inverse-probability-of-treatment weights. Users must however be wary of a number of subtle issues regarding MSMs that might impact the validity of the inferences. In this talk, I first introduce MSMs and then present one such issue: the stabilization of the weights. This stabilization is usually recommended to reduce variability, but I provide a simple example and some simulation results where it is also observed to yield biased inferences.
Bio:
