Marginal structural models in clinical research: when and how to use them?

Area of Science:

• Epidemiology
• Biostatistics
• Health Research Methods

Background:

• Standard confounding control methods are inadequate when time-varying confounders are affected by prior treatment.
• Such covariates act as both confounders and mediators, complicating causal inference in longitudinal data.

Purpose of the Study:

• To introduce and explain the methodology of marginal structural models.
• To provide a framework for appropriate confounding control in longitudinal studies with time-varying treatments and confounders.

Main Methods:

• Marginal structural models employ a two-step estimation procedure.
• Weights are calculated for each observation to balance covariate distributions across treatment groups in a target population.
• The outcome is then estimated using these weights to adjust for time-varying confounding.

Main Results:

• This weighting strategy effectively addresses confounding by time-varying covariates affected by treatment.
• It allows for unbiased estimation of treatment effects in complex longitudinal settings.

Conclusions:

• Marginal structural models are a powerful tool for causal inference in observational longitudinal studies.
• They are particularly useful when dealing with time-varying confounding influenced by prior treatment exposures.