High-dimensional Bayesian mediation analysis with adaptive Laplace priors

Area of Science:

• Environmental Epidemiology
• Biostatistics
• Genomic Medicine

Background:

• Mediation analysis is crucial for understanding indirect effects in exposure-outcome relationships.
• Bayesian methods are well-suited for mediation analysis due to their hierarchical modeling capabilities.
• High-dimensional mediators pose challenges for traditional mediation analysis.

Purpose of the Study:

• To introduce an adaptive Bayesian mediation analysis method for high-dimensional mediators.
• To apply this method to investigate racial disparities in triple-negative breast cancer (TNBC) diagnosis stage.
• To identify environmental and clinical mediators contributing to these disparities.

Main Methods:

• Developed an adaptive Bayesian mediation analysis incorporating adaptive Laplace priors.
• Applied a penalization function on direct and indirect effects for robust estimation.
• Utilized a linked dataset of TNBC patients (2010-2017) and hazardous air pollutant emissions.

Main Results:

• The adaptive method effectively handles high-dimensional mediators and enhances statistical robustness.
• A portion of the racial disparity in TNBC diagnosed stage was explained by identified variables.
• Key mediators and confounders included age of diagnosis, insurance status, tumor grade, and Naphtha air concentration.

Conclusions:

• The novel adaptive Bayesian mediation analysis provides a powerful tool for complex epidemiological studies.
• Environmental factors, specifically Naphtha exposure, alongside clinical variables, contribute to racial disparities in TNBC.
• This research highlights the importance of integrated environmental and clinical data for understanding health inequities.