Geographically weighted Poisson regression for disease association mapping.

Area of Science:

• Spatial statistics
• Geographical epidemiology
• Statistical modeling

Background:

• Traditional global modeling of spatial data often assumes stationary processes, which may not accurately reflect reality.
• Disease mapping and analysis frequently encounter spatially non-stationary processes.
• Identifying localized relationships between disease rates and socio-economic factors is crucial for public health.

Purpose of the Study:

• To introduce Geographically Weighted Poisson Regression (GWPR) as a novel statistical tool for analyzing disease maps.
• To assess spatial variations in Poisson regression parameters for non-stationary processes.
• To test the assumption of spatial stationarity in disease modeling and identify areas with significant localized relationships.

Main Methods:

• The study employs Geographically Weighted Poisson Regression (GWPR), a conditional kernel regression technique.
• Spatial weighting functions are utilized to estimate local variations in Poisson regression parameters.
• A semi-parametric variant of GWPR is also described.

Main Results:

• GWPR successfully depicts spatial variations in the relationships between disease rates and socio-economic characteristics.
• Significant spatial variations were found in the relationships between working-age mortality and occupational segregation/unemployment in Tokyo.
• Application of traditional global models to the Tokyo data would yield misleading results due to spatial non-stationarity.

Conclusions:

• GWPR offers a valuable statistical approach for analyzing disease maps with spatially non-stationary processes.
• The method effectively identifies localized relationships and exceptions that traditional global models might miss.
• GWPR provides disease analysts with enhanced tools for investigating spatial variations in health outcomes.