28 Jul 2016 10:00am

Marginal structural models: powerful tools to estimate the causal effect of time varying exposures and the optimal dynamic regime. SEMINAR CANCELLED


In the first part of this talk I will introduce one of the methods to estimate the causal effect of time-varying exposures: inverse probability of treatment weighting (IPTW) of marginal structural models (MSMs). These models proposed by Robins et al (2000) allow estimation of causal effects from observational data even in the presence of time-dependent confounders, intermediate variables that are also confounders. I will present the ideas underlying MSMs, describe how an observational study can be used to emulate a hypothetical randomised trial, discuss the kind of causal questions that can potentially be answered using MSMs and point out the challenges faced when applying these models to real data sets.

In the second part of this talk I will introduce dynamic treatment regimes (DTR), individually tailored treatments based on patient covariate history, the potential outcomes associated with DTR and the target parameter for inference when evaluating the causal effect of a DTR. I will describe an extension of MSMs suitable for estimating the optimal DTR from longitudinal data when the set of regimes of interest comprises simple rules that can be indexed by a Euclidean vector, and discuss some of the practical issues when applying dynamic MSMs.