Bayesian Causal Mediation Analysis for Group Randomized Designs with Homogeneous and Heterogeneous Effects:...

Area of Science:

• Biostatistics
• Causal Inference
• Bayesian Statistics

Background:

• Causal mediation analysis is crucial for understanding intervention mechanisms.
• Group-randomized designs present unique challenges for mediation analysis.
• Heterogeneous treatment or mediator effects complicate standard causal inference.

Purpose of the Study:

• To present a fully Bayesian approach for causal mediation analysis in group-randomized designs.
• To integrate Bayesian inferential methods with G-computation to address effect heterogeneity.
• To provide a robust framework for analyzing complex intervention effects.

Main Methods:

• A novel Bayesian causal mediation framework was developed.
• The approach combines Bayesian inference with G-computation.
• A simulation study and a case study were used for validation.

Main Results:

• The proposed Bayesian method demonstrates excellent frequentist properties.
• The approach performs well with small sample sizes when using informative priors.
• It accurately accounts for heterogeneous treatment or mediator effects without bias.

Conclusions:

• The developed Bayesian causal mediation analysis is a powerful tool for group-randomized designs.
• This method effectively handles heterogeneous effects, improving causal inference.
• It offers a robust and unbiased approach for complex intervention studies.