Approximate methods in Bayesian point process spatial models.

Area of Science:

• Spatial epidemiology
• Statistical modeling
• Disease surveillance

Background:

• Point process models are crucial for analyzing disease incidence in spatial epidemiology.
• Comparing various models, from approximate to exact methods, is essential for accurate disease mapping and risk assessment.
• Understanding spatial patterns of disease requires robust statistical frameworks.

Purpose of the Study:

• To compare the performance of different point process models in spatial epidemiology.
• To evaluate models for their ability to estimate distance effects and explain spatial heterogeneity.
• To assess model accuracy using real-world data (Lancashire larynx cancer) and simulations.

Main Methods:

• Comparison of approximate methods (Poisson process, discretized window models) and an exact method (conditional logistic model).
• Application to the Lancashire larynx cancer dataset.
• Conducting a simulation study to assess parameter recovery.

Main Results:

• Conditional logistic and binomial models performed well in estimating the distance effect of larynx cancer incidence from an incinerator.
• Poisson and log Gaussian Cox process models, applied to a discretized window, provided the best estimates for explaining spatial heterogeneity.
• Model performance varied depending on whether the focus was on distance effects or spatial variation.

Conclusions:

• The choice of point process model impacts the accuracy of spatial epidemiological analyses.
• Conditional logistic and binomial models are suitable for analyzing proximity-related disease risks.
• Discretized Poisson or log Gaussian Cox models are effective for capturing broader spatial variations in disease incidence.