Marginal structural models for estimating effect modification.

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Marginal structural models (MSMs) are increasingly used in observational studies to adjust for confounding.
• Estimating effect modification is crucial for understanding complex relationships in health research.

Purpose of the Study:

• To propose and evaluate Marginal Structural Models (MSMs) for estimating effect modification in observational cohort and case-control studies.
• To extend the application of MSMs beyond confounding adjustment to effect modification analysis.

Main Methods:

• Derived MSMs for effect modification using the potential outcome model framework.
• Applied the proposed MSMs to both a cohort study and a case-control study dataset.
• Utilized logistic MSMs for analysis in both study designs.

Main Results:

• In cohort studies, logistic MSMs accurately estimated effect modification, aligning with standard regression analysis.
• In case-control studies, effect modification estimation involved a ratio of model estimates from case and control data.
• The proposed MSM approach in case-control studies showed potential for bias.

Conclusions:

• Marginal structural models (MSMs) offer a viable method for epidemiological researchers to estimate effect modification.
• Researchers using MSMs in case-control studies must assess potential bias by applying logistic MSMs to control data.
• The study highlights the utility and limitations of MSMs for effect modification in different observational designs.