Variance partitioning in spatio-temporal disease mapping models.

Area of Science:

• Statistics
• Epidemiology
• Biostatistics

Background:

• Bayesian disease mapping is crucial for understanding spatio-temporal variations in disease risk.
• Traditional models face challenges with prior elicitation for random effect precision parameters, hindering interpretability.

Purpose of the Study:

• To introduce a reparametrized spatio-temporal interaction model, the variance partitioning model, to enhance the interpretability of Bayesian disease mapping.
• To simplify prior elicitation using a mixing parameter and penalized complexity priors.

Main Methods:

• The variance partitioning model utilizes Kronecker product intrinsic Gaussian Markov random fields.
• A key feature is a mixing parameter that balances main and interaction effects, controlling total variance.
• Penalized complexity priors are employed for intuitive prior information coding.

Main Results:

• The variance partitioning model offers enhanced interpretability compared to standard spatio-temporal models.
• The mixing parameter provides a clear way to assess the contribution of different effects to disease risk variation.
• Case studies demonstrate the practical advantages and improved understanding offered by the model.

Conclusions:

• The variance partitioning model provides a more intuitive and interpretable approach to Bayesian disease mapping.
• It effectively addresses the challenges of prior elicitation in complex spatio-temporal models.
• This facilitates a better understanding of disease risk dynamics over time and space.