Bayesian shared spatial-component models to combine and borrow strength across sparse disease surveillance sources

Area of Science:

• Epidemiology
• Biostatistics
• Veterinary Science

Background:

• Geographical disease risk analysis often suffers from data sparseness.
• Spatially structured bias can compromise the accuracy of disease risk assessments.
• Integrating multiple data sources is crucial for robust epidemiological studies.

Purpose of the Study:

• To investigate the utility of Bayesian shared spatial component models for analyzing geographical disease risk with sparse data.
• To correct for spatially structured bias using distinct data sources on related diseases.
• To apply and assess these models for scrapie infection risk in sheep in Wales.

Main Methods:

• Development and application of Bayesian shared spatial component models.
• Modeling individual diseases separately using combined surveillance data.
• Jointly analyzing multiple diseases and their respective data sources.
• Utilizing pseudo cross-validatory predictive model checks for performance assessment.

Main Results:

• Demonstrated the effectiveness of Bayesian shared spatial component models in addressing data sparseness.
• Successfully corrected for spatially structured bias in disease risk analysis.
• Showcased improved inference by jointly modeling diseases and surveillance data.
• Evaluated and compared the predictive performance of various nested joint models.

Conclusions:

• Bayesian shared spatial component models offer a powerful approach to overcome data limitations in geographical disease risk analysis.
• Joint modeling of diseases and data sources enhances the reliability and accuracy of epidemiological insights.
• The developed methodology is particularly valuable for diseases with limited surveillance data, such as scrapie in sheep.