Bayesian Forecasting of Mortality Rates for Small Areas Using Spatiotemporal Models

Area of Science:

• Biostatistics
• Epidemiology
• Demography

Background:

• Accurate estimation and prediction of subnational mortality rates are crucial for analyzing health inequalities.
• Standard statistical methods often struggle with noisy subnational data, leading to unreliable estimates.
• Existing models may not adequately capture regional variations in mortality patterns.

Purpose of the Study:

• To develop a robust Bayesian hierarchical model for predicting subnational mortality rates at a small area level.
• To incorporate spatial components and smooth data across time, age, and neighboring regions for improved accuracy.
• To enhance prediction reliability through Bayesian stacking and provide uncertainty quantification.

Main Methods:

• A Bayesian hierarchical model framework integrating demographic and epidemiological concepts.
• Inclusion of a spatial component to account for regional heterogeneity in mortality.
• Application of Bayesian stacking with leave-future-out validation for model selection and robustness.
• Forecasting mortality rates for 96 regions in Bavaria, Germany, disaggregated by age and sex.

Main Results:

• The proposed model demonstrates superior predictive performance compared to standard models lacking a regional component on held-out data.
• Posterior predictive checks confirm the model's ability to capture essential data features for forecasting.
• The method provides prediction intervals, offering crucial uncertainty estimates for the forecasts.

Conclusions:

• The Bayesian hierarchical model offers a reliable framework for predicting subnational mortality rates, particularly in data-limited or noisy settings.
• This approach enhances the study of health inequalities by providing more accurate small-area mortality estimates.
• The model's spatial component and smoothing techniques improve forecast robustness and reliability.