Objectives Marginal structural models (MSMs) in longitudinal studies are needed when time-varying confounders are themselves predicted by previous exposure, and are intermediate variables on the pathway between exposure and disease. The epidemiologist is left with the unenviable choice of adjusting or not for the confounder/intermediate variable. An example would be whether aspirin decreases cardiovascular mortality, in which the confounder/intermediate variable is cardiovascular morbidity.
Method MSMs use inverse-probability weights based on an ‘exposure’ model which assesses the probability that each subject has received their own exposure and confounder history up to time t, with the follow-up period divided into T (t=1 to T) categories. These weights are then used in standard regression models (eg., pooled logistic regression models across T categories) relating exposure to disease. Their use creates a pseudo-population where time-varying confounding is eliminated.
Results Empirical results show that standard methods to control for time-varying confounders can result in bias towards the null, compared to MSMs. A recent simulation study showed MSMs lead to unbiased results under a variety of assumptions.
Conclusions Some studies have used somewhat different but related methods (“g-estimation”) to account for the healthy worker survivor effect, where employment status is a time-varying confounder which predicts future exposure and may predict disease, but may also act as an intermediate variable because prior exposure may cause illness which results in leaving employment. Here we will present an overview of MSMs and the related g-estimation models.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.