Analysis of Longitudinal Studies With Repeated Outcome Measures: Adjusting for Time-Dependent Confounding Using...

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Estimating causal effects of time-varying exposures with longitudinal data is challenging due to time-varying confounders.
• Standard regression methods can produce biased results in the presence of such confounders.

Purpose of the Study:

• To present sequential conditional mean models (SCMMs) as an alternative to marginal structural models for estimating causal effects.
• To demonstrate how standard regression can be adapted to handle time-dependent confounding.

Main Methods:

• Sequential conditional mean models (SCMMs) fitted using generalized estimating equations.
• Propensity score adjustment integrated into SCMMs for improved robustness.
• Comparison of SCMMs with marginal structural models via simulation studies.

Main Results:

• SCMMs provide precise causal effect estimations, even with time-dependent confounding.
• Propensity score adjustment enhances SCMM robustness against model misspecification.
• SCMMs accommodate continuous exposures and interactions effectively.

Conclusions:

• SCMMs offer a viable and robust approach for causal inference in longitudinal epidemiological studies.
• The method allows for precise estimation of total effects and introduces a new test for direct effects.