Fitting marginal structural models: estimating covariate-treatment associations in the reweighted data set can guide...

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Marginal structural models (MSMs) are widely used for causal inference from observational data.
• The validity of MSMs depends on unconfoundedness in inverse probability-of-treatment weighted datasets.
• Time-dependent treatments and covariates present unique challenges for maintaining unconfoundedness.

Purpose of the Study:

• To evaluate the unconfoundedness property of weights in MSMs with time-dependent treatment and covariates.
• To propose a framework for assessing sensitivity to weighting schemes.
• To identify and select optimal weighting strategies for improved causal inference.

Main Methods:

• Utilizing an observational study of intravenous immunoglobulin for juvenile dermatomyositis as a case example.
• Evaluating traditional methods for fitting the probability-of-treatment model.
• Augmenting the probability-of-treatment model to address residual confounding.

Main Results:

• Traditional probability-of-treatment models leave significant associations between treatment and covariates.
• Augmenting these models effectively reduces confounding.
• Altering the weighting scheme impacts treatment effect estimates.

Conclusions:

• Standard model-fitting strategies for probability-of-treatment models may fail to eliminate confounding.
• The proposed framework aids in detecting residual confounding.
• This approach facilitates the formulation of improved probability-of-treatment models for more reliable causal inference.