Modelling spatial heterogeneity in exposure buffers and risk: a hierarchical Bayesian approach.

Area of Science:

• Spatial epidemiology
• Geographic Information Systems (GIS)
• Biostatistics

Background:

• Traditional place-based epidemiology uses fixed circular buffers to define exposure, which can lead to suboptimal inference.
• Buffer radii are often arbitrarily chosen, assuming a constant threshold for exposure effects across all locations.

Purpose of the Study:

• To develop a flexible, data-driven method for defining exposure in spatial epidemiology.
• To introduce spatially varying buffer radii (SVBR) allowing radii and exposure effects to vary spatially.
• To improve statistical inference in place-based epidemiological studies.

Main Methods:

• Developed a hierarchical Bayesian spatial change points approach.
• Treated buffer radii as unknown parameters that can vary spatially.
• Implemented the method in the R package EpiBuffer.

Main Results:

• Simulations showed SVBR improves estimation and inference compared to traditional methods.
• Application to healthcare access in Madagascar revealed spatial variation in the relationship between facility proximity and antenatal care usage.
• Demonstrated enhanced accuracy in defining exposure and quantifying its impact.

Conclusions:

• Spatially varying buffer radii (SVBR) offer a more accurate and flexible alternative to fixed buffers in epidemiological studies.
• The SVBR approach relaxes rigid assumptions, enabling data-driven exposure definition and impact quantification.
• This method enhances the reliability of spatial epidemiological findings and is available in the EpiBuffer R package.