Authors
Zilan Chai, Arce Domingo-Relloso, Katrina L Devick, Maitreyi Mazumndar, Ana Navas-Acien, Brent A Coull, Linda Valeri
Published in
International journal of epidemiology. Volume 55. Issue 6. Oct 14, 2026.
Abstract
Machine learning is becoming increasingly used in high-dimensional mediation settings. In this work, we present software that implements the Bayesian Kernel Machine Regression-Causal Mediation Analysis (BKMR-CMA) algorithm, which extends the popular BKMR methodology to causal mediation analysis in the presence of multiple correlated exposures. Its functions are implemented in the R package causalmixtures.
The mediation.bkmr command conducts a mediation analysis evaluating the potential intermediate role of a variable on the association between a mixture of exposures and an outcome. The CDE.bkmr command computes controlled direct effects when fixing the mediator to different levels.
The BKMR-CMA method, implemented in the mediation.bkmr and the CDE.bkmr commands from the causalmixtures R package, is able to decompose the total effect into the natural direct and the natural indirect effect, while evaluating interventions on the mediator, incorporating nonlinear associations between exposure, mediator, and outcome, accounting for correlations between exposures, testing for interactions, and providing tools for detailed visualization of results.
The causalmixtures R package is freely available for download on Github: https://arcedr.github.io/causalmixtures/. Details of implementation can be found in the quick start guide: https://arcedr.github.io/causalmixtures/articles/BKMRCMA_QuickStart.html.
PMID:
42966014
Bibliographic data and abstract were imported from PubMed on 01 Nov 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 41
- Comments 0