A scalable approach for short-term disease forecasting in high spatial resolution areal data

Area of Science:

• Epidemiology
• Biostatistics
• Geographic Information Systems (GIS)

Background:

• Short-term disease forecasting is crucial for health planning.
• Standard spatiotemporal models face computational challenges with high-dimensional areal data.
• Accurate predictions are needed for large geographical areas.

Purpose of the Study:

• To introduce a novel
• divide-and-conquer
• method for short-term predictions in high-dimensional areal data.
• To evaluate the predictive performance of this new method against classical spatiotemporal models.
• To apply the method to cancer mortality data for all municipalities in continental Spain.

Main Methods:

• A new "divide-and-conquer" spatiotemporal modeling approach was developed.
• The method was validated using cancer mortality data across 7907 Spanish municipalities.
• Models were implemented within a Bayesian framework using integrated nested Laplace estimation (INLA).

Main Results:

• The proposed "divide-and-conquer" method demonstrated superior predictive performance.
• Outperformed traditional models in terms of mean absolute error (MAE), root mean square error (RMSE), and interval score.
• Achieved better accuracy in forecasting cancer mortality 1, 2, and 3 years ahead.

Conclusions:

• The "divide-and-conquer" approach offers a computationally efficient and accurate solution for short-term disease forecasting in high-dimensional areal data.
• This method enhances decision-making in public health planning by providing reliable predictions.
• The findings support the use of this novel approach for epidemiological surveillance and health resource allocation.