Big problems in spatio-temporal disease mapping: Methods and software.

Area of Science:

• Epidemiology
• Statistical Modeling
• Geospatial Analysis

Background:

• Spatio-temporal models for areal data are vital in epidemiology.
• Large datasets present computational challenges for traditional modeling approaches.
• Accurate relative risk estimation is critical for public health interventions.

Purpose of the Study:

• To propose a general procedure for analyzing high-dimensional spatio-temporal areal data.
• To address computational limitations in fitting complex hierarchical models.
• To enhance mortality/incidence relative risk estimation.

Main Methods:

• Developed a pragmatic procedure for fitting hierarchical spatio-temporal models to large areal datasets.
• Utilized integrated nested Laplace approximations over spatial partitions.
• Employed parallel and distributed computing strategies to accelerate Bayesian model fitting.

Main Results:

• Demonstrated superior performance of the proposed method over classical global models using simulated and real data.
• Implemented the methodology in the open-source R package bigDM.
• Included user-friendly vignettes to facilitate adoption by non-expert users.

Conclusions:

• The proposed scalable methodology enables reliable risk estimates.
• Facilitates the fitting of Bayesian hierarchical spatio-temporal models for high-dimensional data.
• Offers a computationally efficient solution for complex epidemiological analyses.