Statistical software for analyzing the health effects of multiple concurrent exposures via Bayesian kernel machine...

Area of Science:

• Environmental Epidemiology
• Biostatistics
• Computational Biology

Background:

• Estimating health effects of multi-pollutant mixtures is crucial in environmental epidemiology.
• Bayesian kernel machine regression (BKMR) is a novel method for analyzing complex exposure-response functions.
• Previous BKMR applications were limited by software availability, computational efficiency, and outcome variable types.

Purpose of the Study:

• To address limitations of the BKMR method by developing open-source software and extending its applicability.
• To enhance the usability of BKMR for analyzing health effects of environmental exposures.
• To provide tools for visualizing complex exposure-response relationships and handling various outcome types.

Main Methods:

• Introduced an open-source R package, 'bkmr', for BKMR implementation.
• Developed methods for visualizing high-dimensional exposure-response functions and estimating summaries.
• Implemented probit regression for binary outcomes and a Gaussian predictive process approach for computational efficiency.

Main Results:

• The BKMR implementation accurately estimated health effects of multi-pollutant mixtures with nonlinear dose-response functions.
• The Gaussian predictive process approach significantly reduced runtime with minimal accuracy loss.
• Probit BKMR successfully identified key exposure variables and provided interpretable results for binary outcomes.

Conclusions:

• The developed BKMR software and extended methodology make this powerful tool accessible for diverse epidemiological studies.
• This integrated suite of tools facilitates the analysis of complex health effects from multiple environmental risk factors.
• The advancements enable broader application of BKMR in understanding the health impacts of environmental exposures.