Bayesian mapping of mortality clusters

Area of Science:

• Spatial statistics
• Bayesian modeling
• Public health surveillance

Background:

• Disease mapping identifies geographic patterns in health outcomes.
• Existing methods struggle to simultaneously identify spatial clusters and contributing diseases.
• Accurate disease mapping requires understanding both location and cause.

Purpose of the Study:

• To develop a multivariate Bayesian model for spatial mortality cluster detection.
• To simultaneously identify cluster boundaries and the diseases driving them.
• To incorporate external covariates for enhanced disease mapping.

Main Methods:

• Introduced 'perla', a multivariate Bayesian model for clustering areas by mortality rates.
• Utilized a stick-breaking formulation of the multinomial distribution for spatial structure.
• Employed global-local shrinkage priors and a Markov chain Monte Carlo algorithm for inference.

Main Results:

• The 'perla' model effectively clusters areas based on multiple causes of death.
• It successfully identifies diseases contributing to mortality clusters.
• The model demonstrated flexibility and effectiveness in Italian and US county case studies.

Conclusions:

• 'Perla' offers a novel solution for simultaneous spatial mortality cluster detection and disease attribution.
• The methodology enhances disease mapping by integrating spatial data and covariates.
• This approach provides valuable insights for targeted public health interventions.