MSPOCK: Alleviating Spatial Confounding in Multivariate Disease Mapping Models

Area of Science:

• Biostatistics
• Spatial Epidemiology
• Public Health

Background:

• Disease mapping uses spatial patterns to assess geographical disease tendencies.
• Multivariate shared component models analyze multiple disease types but can suffer from spatial confounding.
• Spatial confounding leads to misleading interpretations by correlating spatial random effects with fixed effects.

Purpose of the Study:

• Introduce Multiple SPOCK (MSPOCK) to address spatial confounding in multiple count data.
• Evaluate MSPOCK's effectiveness on synthetic data and real-world respiratory cancer incidence.
• Improve the reliability of spatial epidemiological models for public health applications.

Main Methods:

• Developed the Multiple SPOCK (MSPOCK) methodology, an extension of SPOCK for multiple count scenarios.
• Applied MSPOCK to model spatial patterns of respiratory system cancer incidence in California.
• Utilized synthetic data for initial validation and a real-world dataset for illustration.

Main Results:

• MSPOCK effectively tackles spatial confounding in multivariate disease mapping.
• The method demonstrated a reduction in posterior variance estimates for model parameters.
• Model interpretability was preserved following the MSPOCK correction.

Conclusions:

• MSPOCK offers a robust solution to spatial confounding in complex disease mapping scenarios.
• The methodology enhances the precision and reliability of spatial epidemiological analyses.
• MSPOCK is valuable for accurately assessing geographical disease tendencies and informing public health strategies.