A Bayesian two-stage spatially dependent variable selection model for space-time health data.

Area of Science:

• Epidemiology
• Biostatistics
• Geographic Information Science

Background:

• Traditional epidemiological models often overlook spatial and temporal variations in risk factor effects.
• Health outcome associations with risk factors can exhibit complex, location- and time-dependent patterns.
• Existing methods may struggle to account for spatially varying temporal dependencies in covariate effects.

Purpose of the Study:

• To develop a Bayesian two-stage approach for analyzing space-time health data.
• To identify spatially varying subsets of regression coefficients that capture temporal dependence.
• To reduce spatial confounding bias in epidemiological modeling.

Main Methods:

• Proposed a Bayesian two-stage spatially dependent variable selection method.
• Implemented a common temporal dependence structure for regression coefficients.
• Utilized a simulation study to validate the model's performance.
• Applied the model to lung cancer inpatient data in Georgia, USA.

Main Results:

• The proposed two-stage model effectively identifies spatially varying risk factor effects.
• Demonstrated the reduction of spatial confounding bias.
• Simulation results confirmed the model's performance in space-time analysis.
• The application to lung cancer data provided insights into regional variations.

Conclusions:

• The Bayesian two-stage approach offers a robust framework for space-time epidemiological modeling.
• This method enhances the understanding of how risk factors impact health outcomes across diverse geographic areas and time periods.
• The findings contribute to more accurate risk assessment and targeted public health interventions.