Constructing G-computation Estimators: Two Case Studies in Selection Bias.

Area of Science:

• Epidemiology
• Causal Inference
• Biostatistics

Background:

• G-computation is a valuable method for addressing epidemiological biases.
• Adapting g-computation for complex causal structures can be challenging.
• Translating causal diagrams into estimation strategies requires careful consideration.

Purpose of the Study:

• To demonstrate the adaptation of g-computation for specific selection bias scenarios in epidemiology.
• To provide practical guidance on implementing adapted g-computation estimators.
• To explore the theoretical and finite-sample properties of novel g-computation estimators.

Main Methods:

• The study adapted g-computation for two selection bias cases: treatment-induced selection and co-occurring biases without a joint adjustment set.
• Proposed estimators were expressed as stacked estimating equations for simplified theory and application.
• Simulations were used to illustrate the performance of the adapted estimators.

Main Results:

• G-computation can be effectively adapted to address complex selection biases in epidemiological studies.
• Stacked estimating equations provide a practical framework for implementing novel g-computation estimators.
• Simulations confirmed the utility and properties of the developed estimators.

Conclusions:

• Epidemiologists can translate causal identification strategies into practical g-computation estimators.
• The proposed methods facilitate the study of theoretical and finite-sample properties of novel causal estimators.
• This work enhances the application of g-computation in complex epidemiological research.