Bayesian spatial modeling of disease risk in relation to multivariate environmental risk fields

Area of Science:

• Environmental health
• Spatial statistics
• Developmental toxicology

Background:

• Prenatal exposure to environmental chemicals poses risks to child development.
• Assessing these risks is complex due to the spatial nature of chemical exposure and outcome data.
• Existing methods struggle with interpolating environmental data and estimating health risks simultaneously.

Purpose of the Study:

• To develop a Bayesian joint model for interpolating soil chemical concentrations.
• To simultaneously estimate the risk of mental retardation and developmental delay (MRDD) associated with these exposures.
• To evaluate a low-rank Kriging method for computational efficiency in spatial interpolation.

Main Methods:

• Developed a Bayesian joint model integrating spatial interpolation and risk assessment.
• Employed a low-rank Kriging method for interpolating multiple soil chemistry fields.
• Conducted sensitivity analyses on bivariate smoothing parameters (knots and smoothing parameter).

Main Results:

• The Bayesian joint model successfully interpolates soil chemical fields and estimates MRDD risk.
• Low-rank Kriging provides a computationally efficient alternative to full-rank Kriging for spatial interpolation.
• Sensitivity analyses indicate that the choice of knots in low-rank Kriging requires careful consideration.

Conclusions:

• A Bayesian joint model is effective for analyzing the spatial relationship between environmental chemicals and developmental outcomes.
• Low-rank Kriging offers a practical approach to reduce computational demands in spatial environmental health studies.
• Further research should focus on optimizing knot selection for robust spatial surface estimation.