An intuitive Bayesian spatial model for disease mapping that accounts for scaling.

Area of Science:

• Geographical Epidemiology
• Bayesian Statistics
• Disease Mapping

Background:

• Bayesian hierarchical models are standard for disease mapping.
• Existing models like the BYM (Besag, York and Mollié) model have issues with confounding spatial and unstructured components, complicating hyperprior definition.

Purpose of the Study:

• To present a novel parameterisation of the BYM model for improved parameter control and interpretable hyperpriors.
• To address the need for scaled spatial components to facilitate hyperprior assignment across different spatial structures.

Main Methods:

• Introduced a reparameterised BYM model with independent hyperparameters.
• Incorporated a scaled spatial component for transferable hyperpriors.
• Derived penalised complexity priors based on information-theoretic distance for clear interpretation.

Main Results:

• The new model formulation allows hyperparameters to be viewed independently.
• It facilitates the assignment of interpretable and transferable hyperpriors.
• Simulation studies demonstrate good learning abilities, shrinkage behavior, and competitive model choice performance.

Conclusions:

• The proposed BYM model parameterisation enhances interpretability of parameters and hyperpriors.
• This offers a more flexible and understandable approach to disease mapping within geographical epidemiology.
• The model performs comparably to existing methods while providing superior interpretability.