A comparison of conditional autoregressive models used in Bayesian disease mapping

Area of Science:

• Epidemiology
• Spatial statistics
• Biostatistics

Background:

• Disease mapping identifies geographical areas with high disease risk.
• Bayesian hierarchical models with spatial random effects are standard tools.
• Conditional autoregressive (CAR) priors are commonly used for spatial effects.

Purpose of the Study:

• To critique four prevalent CAR models for Bayesian disease mapping.
• To assess the appropriateness of these CAR models through simulation.
• To apply selected models to real-world cancer incidence data.

Main Methods:

• Critique of four common Conditional Autoregressive (CAR) models.
• Simulation study to evaluate model performance and appropriateness.
• Application of models to cancer incidence data in Greater Glasgow (2001-2005).

Main Results:

• The simulation study provides insights into the performance of different CAR models.
• Comparative analysis of model fit and parameter estimation.
• Identification of suitable models for specific disease mapping scenarios.

Conclusions:

• The choice of CAR model impacts disease risk estimation in spatial epidemiology.
• Simulation results guide the selection of appropriate models for disease mapping.
• The study offers practical insights for analyzing cancer incidence data in Greater Glasgow.