Bayesian spatio-temporal modelling for infectious disease outbreak detection.

Area of Science:

• Epidemiology
• Biostatistics
• Computational Biology

Background:

• Bayesian analysis of infectious disease surveillance data commonly uses spatio-temporal models.
• Existing models face challenges with parameter identifiability, particularly concerning seasonality and outbreak components.

Purpose of the Study:

• To present a new, generally applicable methodology for spatio-temporal Bayesian analysis of infectious disease surveillance data.
• To develop computationally efficient inference techniques for improved outbreak detection and model comparison.

Main Methods:

• Developed a parsimonious representation of seasonality and a biologically informed outbreak component specification.
• Implemented computationally efficient Bayesian inference, including dynamic Hamiltonian Monte Carlo (HMC) and importance sampling for model evidence approximation.
• Integrated techniques for outbreak detection using marginal posterior probabilities.

Main Results:

• Successfully detected simulated outbreaks and demonstrated strong reliability in model comparison.
• Applied the methodology to invasive meningococcal disease data from 28 European countries, highlighting multi-country outbreaks.
• Model comparison analysis indicated superior performance of the new specification over previous approaches.

Conclusions:

• The new Bayesian methodology offers a robust and efficient framework for infectious disease surveillance and outbreak detection.
• The approach improves upon existing spatio-temporal modeling techniques, providing more reliable model comparison.
• Freely available R package facilitates the application of this advanced methodology in public health research.