Efficient sequential Bayesian inference for state-space epidemic models using ensemble data assimilation.

Area of Science:

• Epidemiology
• Computational Biology
• Statistical Modeling

Background:

• Accurate estimation of epidemic states and parameters from noisy, incomplete data is crucial for infectious disease modeling.
• Traditional Bayesian methods, like Sequential Monte Carlo squared (SMC2), are computationally intensive, limiting their use in real-time outbreak response.
• Existing methods struggle with the partially observed and noisy nature of surveillance data.

Purpose of the Study:

• To develop a computationally efficient Bayesian inference framework for infectious disease modeling.
• To improve the speed of state-space model parameter and latent trajectory estimation.
• To provide a robust method for near-real-time outbreak analysis using imperfect surveillance data.

Main Methods:

• Proposed Ensemble SMC2 (eSMC2), a novel variant of the SMC2 algorithm.
• Replaced the inner particle filter of SMC2 with an Ensemble Kalman Filter (EnKF) for faster likelihood approximation.
• Incorporated an unbiased Gaussian density estimator and state-dependent observation variance to handle overdispersed epidemic data.

Main Results:

• eSMC2 demonstrated substantial computational gains compared to the standard SMC2 algorithm.
• Posterior estimates from eSMC2 were comparable to those obtained using SMC2.
• The method accurately recovered latent epidemic trajectories and key epidemiological parameters in simulations and real-world data.
• Applied eSMC2 to 2022 U.S. monkeypox incidence data, showing its practical utility.

Conclusions:

• eSMC2 offers an efficient and accurate framework for sequential Bayesian inference in infectious disease modeling.
• The method is particularly suitable for analyzing overdispersed incidence data common in public health surveillance.
• eSMC2 provides a viable alternative for near-real-time epidemic analysis and outbreak response.