Marginal structural models for the estimation of direct and indirect effects

Area of Science:

• Causal inference
• Epidemiology
• Statistical modeling

Background:

• Estimating controlled direct effects is crucial in causal inference.
• Marginal structural models (MSMs) with inverse probability of treatment weighting (IPTW) are standard for this.
• Extending MSMs to natural direct and indirect effects requires careful model specification.

Purpose of the Study:

• To present a method for estimating natural direct and indirect effects using MSMs.
• To adapt MSMs for estimating these specific causal effects in observational studies.
• To highlight the importance of conditioning MSMs on covariates for accurate estimation.

Main Methods:

• Fitting two marginal structural models (MSMs).
• The first MSM models the effect of treatment and mediator on the outcome.
• The second MSM models the effect of treatment on the mediator, with both models conditioned on covariates.

Main Results:

• The proposed method allows for the estimation of natural direct and indirect effects.
• Conditioning MSMs on covariates improves the accuracy of effect estimation compared to standard epidemiologic approaches.
• This provides a robust framework for dissecting causal pathways.

Conclusions:

• Marginal structural models can be extended to estimate natural direct and indirect effects.
• Conditioning these models on covariates is essential for valid estimation.
• This methodology enhances causal inference in complex scenarios involving mediators.