How does Poisson kriging compare to the popular BYM model for mapping disease risks?

Area of Science:

• Spatial statistics
• Geostatistics
• Epidemiology

Background:

• Geostatistical techniques can account for spatially varying population sizes and patterns in disease rate mapping.
• Poisson kriging is an alternative to Bayesian spatial models, offering easier implementation and better handling of geographical unit variations.
• A comparison between Poisson kriging and Bayesian spatial models (like BYM) in simulation studies is needed to understand their accuracy and precision in disease risk estimation.

Purpose of the Study:

• To compare the performance of Poisson kriging and Bayesian spatial models (Besag, York and Mollie - BYM) in mapping disease rates.
• To evaluate the accuracy and precision of disease risk estimates generated by these two methodologies.
• To assess the impact of geographical unit heterogeneity on the performance of the models.

Main Methods:

• Application of the Besag, York and Mollie (BYM) model and Poisson kriging (point and area-to-area) to age-adjusted lung and cervix cancer mortality rates.
• Analysis of data from two contrasted county geographies: Indiana (uniform counties) and Western US (heterogeneous counties).
• Simulation studies to compare the accuracy, precision, and discrimination capabilities of the methods.

Main Results:

• The statistical methodology (geostatistical vs. Bayesian) had a greater impact on results than spatial support (point vs. area).
• Poisson kriging yielded smaller prediction errors, more precise probability intervals, and better discrimination between high and low mortality risk counties.
• Differences were more pronounced in the heterogeneous Western US dataset, where BYM produced smoother risk surfaces and Poisson kriging variances increased in large, sparsely populated counties.

Conclusions:

• Poisson kriging provides greater flexibility in modeling spatial risk structures and generates less smoothing than the BYM model.
• The geostatistical approach, particularly area-to-area Poisson kriging, is beneficial in heterogeneous geographies.
• Public health officials should consider the spatial and distributional assumptions of Bayesian models and explore alternatives like Poisson kriging for improved identification of high-risk areas.