Spatial mixture multiscale modeling for aggregated health data.

Area of Science:

• Spatial epidemiology
• Geostatistics
• Biostatistics

Background:

• Spatial epidemiology aims to understand disease risk patterns geographically.
• Convolution models with correlated (CH) and uncorrelated heterogeneities (UH) are common but may vary regionally.
• Investigating the predominance of CH or UH components across different data scales is crucial.

Purpose of the Study:

• To develop and evaluate novel spatial mixture multiscale models for disease risk analysis.
• To explore how correlated and uncorrelated heterogeneities influence disease patterns at multiple scales.
• To identify the most effective model for capturing scale-dependent disease risk variations.

Main Methods:

• Proposed four spatial mixture multiscale models by combining spatially varying probability weights of CH and UH.
• Model 1: Independent mixture convolution models at each scale.
• Models 2-4: Introduced linkages between scales via shared CH, UH, or both components.

Main Results:

• The fourth model, sharing both CH and UH components across scales, demonstrated superior performance.
• The second model, linking scales via a shared UH component, was the second-best performing model.
• Model evaluation was conducted using both real and simulated spatial epidemiological data.

Conclusions:

• The proposed fourth model effectively accounts for scale effects by simultaneously sharing correlated and uncorrelated heterogeneities.
• Integrating information across multiple scales, particularly through shared components, enhances the accuracy of spatial disease risk modeling.
• The findings highlight the importance of considering scale-dependent heterogeneity in spatial epidemiological studies.