Log Gaussian Cox processes and spatially aggregated disease incidence data

Area of Science:

• Epidemiology
• Spatial Statistics
• Biostatistics

Background:

• Traditional disease mapping models assume uniform risk within reporting regions.
• Common models like Besag-York-Mollié rely on the Markov assumption, simplifying computation but limiting spatial precision.
• Existing methods often fail to capture fine-scale spatial variations in disease risk.

Purpose of the Study:

• To present a spatially continuous log-Gaussian Cox process methodology for modeling aggregated disease incidence data.
• To improve upon the limitations of traditional region-based disease mapping models.
• To accurately estimate disease risk and identify high-risk areas using precise spatial locations.

Main Methods:

• Developed a log-Gaussian Cox process model for spatially continuous data.
• Employed a data augmentation step within a Markov chain Monte Carlo algorithm to sample exact locations.
• Modeled exact disease incidence locations using the proposed continuous spatial process.

Main Results:

• The log-Gaussian Cox process model demonstrated superior performance compared to the Besag-York-Mollié model in simulation studies.
• The methodology was successfully applied to syphilis risk data in North Carolina.
• Identified areas with significantly elevated syphilis risk beyond that explained by known social factors.

Conclusions:

• The log-Gaussian Cox process offers a more accurate and flexible approach to disease mapping than traditional aggregated models.
• This continuous spatial modeling framework enhances the identification of localized disease hotspots.
• The method provides valuable insights for targeted public health interventions and disease control strategies.