SMOOTHED ANOVA WITH SPATIAL EFFECTS AS A COMPETITOR TO MCAR IN MULTIVARIATE SPATIAL SMOOTHING

Area of Science:

• Spatial statistics
• Disease mapping
• Geographical Information Systems (GIS)

Background:

• Geographical Information Systems (GIS) are crucial for analyzing complex spatial datasets, particularly for disease mapping.
• Multivariate conditionally autoregressive (MCAR) models are used for smoothing disease maps and accounting for inter-disease associations but present interpretation and estimation challenges.
• Existing methods lack sufficient structure for complex datasets, necessitating simpler, interpretable alternatives.

Purpose of the Study:

• To develop a simpler and more interpretable alternative to the MCAR model for smoothing spatial disease data.
• To extend the Smoothed ANOVA (SANOVA) technique to incorporate spatial structures using a conditional autoregressive (CAR) model for lattice data.
• To compare the performance of the proposed SANOVA approach against the MCAR model using simulations and a real-world cancer surveillance dataset.

Main Methods:

• Development of a novel Smoothed ANOVA (SANOVA) approach for spatial random effects, leveraging spatial structure.
• Extension of SANOVA to handle situations with a spatial lattice factor (smoothed via CAR model) and a categorical factor (e.g., cancer type).
• Comparative analysis using simulation studies and a cancer surveillance dataset of 3-cancers across 87 Minnesota counties.

Main Results:

• The SANOVA approach provides a simpler and more intelligible structure compared to the MCAR model.
• SANOVA demonstrates competitive performance against MCAR in both simulation studies and the analysis of the cancer surveillance dataset.
• The proposed method effectively smooths spatial variations and accounts for disease associations, offering a viable alternative for disease mapping.

Conclusions:

• The SANOVA approach offers a practical and effective alternative to MCAR for disease mapping and spatial analysis.
• This method simplifies the analysis of complex spatial disease data while maintaining analytical rigor and interpretability.
• SANOVA enhances hypothesis generation for geographical variations in disease risk by providing a more accessible modeling framework.