A Note on G-Estimation of Causal Risk Ratios

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• G-estimation is a semiparametric method for estimating exposure effects in epidemiologic studies.
• It possesses advantages over popular propensity score methods but is underutilized due to software limitations.
• Existing software for generalized estimating equations (GEE) is widely available.

Purpose of the Study:

• To highlight the underappreciated benefits of G-estimation in epidemiology.
• To provide a practical solution for implementing G-estimation using readily available software.
• To extend the methodology to handle complex scenarios involving time-varying confounders.

Main Methods:

• Demonstrate a technique to derive G-estimators for causal risk ratios.
• Utilize existing generalized estimating equations (GEE) software for implementation.
• Extend the G-estimation procedure to accommodate time-varying confounders.

Main Results:

• A straightforward method is presented for obtaining G-estimators of causal risk ratios.
• This approach leverages existing generalized estimating equations (GEE) software.
• The method is successfully extended to address time-varying confounding.

Conclusions:

• G-estimation can be practically implemented using standard generalized estimating equations (GEE) software.
• This facilitates the application of G-estimation in epidemiologic research.
• The approach is adaptable for complex causal inference problems with time-varying confounders.