Identifying clusters in Bayesian disease mapping.

Area of Science:

• Spatial epidemiology
• Statistical modeling
• Public health

Background:

• Disease mapping estimates spatial disease risk patterns across areal units.
• Identifying elevated disease risk clusters is crucial for public health interventions.
• Existing Bayesian models struggle to define the spatial extent of high-risk clusters.

Purpose of the Study:

• To develop a novel two-stage methodology for identifying and delineating high-risk disease clusters.
• To overcome the limitations of current Bayesian models in defining cluster boundaries.
• To improve the precision of spatial epidemiology for public health action.

Main Methods:

• A two-stage approach combining a spatially adjusted hierarchical agglomerative clustering algorithm with Poisson log-linear models.
• Stage one: Clustering pre-study data to generate potential cluster structures.
• Stage two: Fitting separate models for each structure to allow risk changes at cluster boundaries, selecting the best structure using the Deviance Information Criterion.

Main Results:

• The proposed methodology effectively identifies and delineates spatial disease clusters.
• Simulation studies confirmed the efficacy of the two-stage approach.
• Application to respiratory disease data in Glasgow illustrated its practical utility.

Conclusions:

• The novel two-stage method enhances disease mapping by accurately defining high-risk cluster extents.
• This improved spatial delineation facilitates more targeted and effective public health interventions.
• The methodology offers a valuable advancement for spatial epidemiology and risk assessment.