Bayesian tele-connected spatial clustering of multivariate spatial data with applications to disease-mapping

Area of Science:

• Biostatistics
• Spatial Epidemiology
• Geographic Information Systems (GIS)

Background:

• Spatial clustering is vital for disease mapping, identifying geographic variations in disease incidence or mortality.
• Understanding multivariate disease patterns requires methods accounting for spatial dependence and inter-disease relationships.

Purpose of the Study:

• To propose a novel Bayesian spatial clustering method for multivariate spatial disease data.
• To develop a flexible model capable of identifying contiguous and non-contiguous spatial clusters.
• To estimate cluster-specific disease patterns and dependencies among multiple disease variables.

Main Methods:

• Development of a Bayesian hierarchical model incorporating a novel random tele-connected graph partition prior.
• The prior allows for an unknown number of clusters, encouraging local contiguity while enabling remote regions to cluster.
• Implementation of a tailored Markov chain Monte Carlo (MCMC) algorithm with efficient doubly split-merge samplers and graph algorithms for posterior inference.

Main Results:

• The proposed method effectively detects spatial clusters in multivariate disease data.
• It successfully estimates cluster-specific disease patterns and spatial dependence across multiple disease variables.
• Simulation studies validated the method's performance.

Conclusions:

• The novel Bayesian spatial clustering method provides a robust framework for analyzing complex multivariate spatial disease data.
• It enhances disease mapping by identifying geographically nuanced patterns and relationships.
• The method was successfully applied to prostate cancer mortality rate decline in the southern U.S.