A Marginalized Zero-Inflated Negative Binomial Model for Spatial Data: Modeling COVID-19 Deaths in Georgia

Area of Science:

• Biostatistics
• Epidemiology
• Spatial Statistics

Background:

• Spatial count data often exhibit excess zeros, common in disease mapping.
• Conventional zero-inflated models present interpretation challenges due to their mixture nature.
• Marginalized zero-inflated models offer a more interpretable alternative by directly modeling the mean.

Purpose of the Study:

• To develop a spatiotemporal marginalized zero-inflated negative binomial model for disease mapping.
• To extend marginalized zero-inflated models to incorporate spatial dependencies.
• To identify predictors of COVID-19 death rates using the developed model.

Main Methods:

• Developed a spatiotemporal marginalized zero-inflated negative binomial model.
• Incorporated region-level covariates, smooth temporal effects, and spatially correlated random effects.
• Employed a Bayesian approach using Gibbs sampling and Metropolis-Hastings steps for estimation.

Main Results:

• The model effectively captures spatiotemporal heterogeneity in COVID-19 death rates.
• Identified key predictors associated with COVID-19 mortality in Georgia.
• Demonstrated the utility of the marginalized approach in spatial count data analysis.

Conclusions:

• The proposed spatiotemporal marginalized model provides a robust and interpretable framework for analyzing excess zero count data.
• This methodology advances disease mapping by accounting for spatial and temporal complexities.
• The findings offer insights into factors driving COVID-19 mortality, aiding public health strategies.