Modelling the presence of disease under spatial misalignment using Bayesian latent Gaussian models

Area of Science:

• Environmental epidemiology
• Geostatistics
• Bayesian statistics

Background:

• Spatial disease modelling is crucial for understanding disease incidence and risk factors.
• Geostatistical models are increasingly used for disease risk estimation and prediction.
• Misalignment between covariate data and observed locations poses a challenge in spatial modelling.

Purpose of the Study:

• To develop and present Bayesian spatial models to address data misalignment in disease incidence modelling.
• To incorporate geographical and environmental characteristics into hierarchical Bayesian spatial models.
• To provide a framework for estimating disease risk and identifying risk factors even with partially different covariate information.

Main Methods:

• Hierarchical Bayesian spatial models were employed to model disease presence/absence.
• Two models were developed: one accounting for covariate uncertainty and one without.
• Latent Gaussian models and integrated nested Laplace approximation were used for Bayesian inference.
• Stochastic partial differential equations were utilized for implementing the spatial effect.

Main Results:

• The developed models effectively handle misalignment issues in spatial disease modelling.
• Bayesian inference and prediction were successfully performed using the latent Gaussian model framework.
• The methodology was validated using data on Fasciola hepatica in Galicia, Spain.

Conclusions:

• The proposed Bayesian spatial models offer a robust solution for disease incidence modelling with potentially misaligned covariate data.
• The integrated nested Laplace approximation and stochastic partial differential equation approach provide efficient computational tools.
• This approach enhances the ability to predict disease risk and identify environmental risk factors in epidemiological studies.