Latent likelihood ratio tests for assessing spatial kernels in epidemic models.

Area of Science:

• Epidemiology
• Mathematical Modeling
• Spatial Statistics

Background:

• Accurate spatio-temporal epidemic models depend on appropriate transmission kernel selection.
• Kernel choice significantly impacts control strategy design by influencing estimated transmission distances.
• Existing model criticism methods require further development for spatial epidemic models.

Purpose of the Study:

• Introduce and evaluate latent likelihood ratio tests for assessing spatial kernel validity in epidemic models.
• Compare the power of latent likelihood ratio tests against infection-link residuals for detecting kernel mis-specification.
• Provide a computationally efficient alternative to Bayesian methods for spatial kernel assessment.

Main Methods:

• Developed latent likelihood ratio tests using likelihood-based discrepancy variables.
• Compared the performance of latent likelihood ratio tests with infection-link residuals.
• Utilized simulated epidemic data to assess model mis-specification detection capabilities.

Main Results:

• Latent likelihood ratio tests demonstrate higher power in detecting kernel mis-specification compared to infection-link residuals.
• The new tests are particularly effective when mis-specification is modest.
• The approach avoids the computational complexity and prior sensitivity issues associated with fully Bayesian methods.

Conclusions:

• Latent likelihood ratio tests provide a powerful and practical tool for scrutinizing spatial kernels in epidemic models.
• This method enhances the reliability of epidemic models used for public health interventions.
• The findings support the use of latent likelihood ratio tests for improving the accuracy of epidemiological predictions.