Approximate Bayesian inference for multivariate point pattern analysis in disease mapping

Area of Science:

• Environmental Epidemiology
• Spatial Statistics
• Bayesian Modeling

Background:

• Georeferenced disease mapping is crucial for public health.
• Analyzing multivariate case-control data with spatial risk factors presents challenges.
• Existing methods may not fully capture complex spatial variations and pollution source effects.

Purpose of the Study:

• To develop a novel Bayesian disease mapping framework for georeferenced case-control data.
• To estimate spatial variation and pollution source-related risks using SPDEs and INLA.
• To identify high-risk areas unexplained by covariates.

Main Methods:

• Utilized stochastic partial differential equations (SPDEs) for spatial modeling.
• Employed integrated nested Laplace approximation (INLA) for efficient model fitting.
• Modeled case and control intensities using Log-Gaussian Cox processes with baseline, disease-specific, and covariate effects.
• Incorporated various exposure models for pollution sources (fixed, random walk, Gaussian process).

Main Results:

• The novel framework successfully estimated spatial disease variation and pollution source risks.
• Residual spatial terms identified areas with unexplained high risk.
• Analysis of cancer data in Alcalá de Henares indicated a potential increased risk near specific polluting industries.

Conclusions:

• The proposed SPDE-INLA approach offers a robust method for Bayesian disease mapping.
• This framework effectively integrates spatial effects, pollution exposure, and covariates.
• Findings highlight the need to consider industrial proximity in public health assessments for cancer risk.