BayesGmed: An R-package for Bayesian causal mediation analysis

Area of Science:

• Causal Inference
• Bayesian Statistics
• Health Services Research

Background:

• Causal mediation analysis has seen significant growth, but existing frequentist tools may lack robustness with small sample sizes.
• A Bayesian approach offers a more stable alternative for analyzing mediation effects.

Purpose of the Study:

• To introduce BayesGmed, an R-package for Bayesian causal mediation analysis using the Bayesian g-formula.
• To demonstrate the application of this new methodology in analyzing a randomized controlled trial for chronic pain.

Main Methods:

• Developed BayesGmed, an R-package implementing Bayesian causal mediation models.
• Applied the package to secondary data from the MUSICIAN study, a trial of remotely delivered cognitive behavioural therapy (tCBT).
• Utilized informative priors for probabilistic sensitivity analysis regarding causal identification assumptions.

Main Results:

• tCBT improved self-perceived health status compared to treatment as usual (TAU).
• Fear of movement, passive coping, and sleep problems were associated with lower odds of positive health change.
• BayesGmed indicated no statistically significant mediated effects, though results were comparable to existing R packages.
• Sensitivity analysis confirmed the persistence of direct and total effects of tCBT even with potential unmeasured confounding.

Conclusions:

• BayesGmed provides a valuable open-source tool for Bayesian causal mediation analysis.
• The study highlights the importance of considering Bayesian methods for robust mediation analysis, particularly in clinical research.
• The findings underscore the complex interplay of factors influencing treatment outcomes in chronic pain management.