[Paper Review] Semiparametric Bayesian Inference for the Transmission Dynamics of COVID-19 with a State-Space Model

Tianjian Zhou, Yuan Ji|arXiv (Cornell University)|Jun 10, 2020

COVID-19 epidemiological studies45 references4 citations

TL;DR

This paper proposes a semiparametric Bayesian state-space model for COVID-19 transmission dynamics, using Gaussian process regression to flexibly estimate time-varying transmission rates and parallel-tempering MCMC for efficient posterior sampling. The method captures nonlinear trends in the effective reproduction number and predicts future case counts, revealing a declining $ρ_e$ across six U.S. states linked to public health interventions.

ABSTRACT

The outbreak of Coronavirus Disease 2019 (COVID-19) is an ongoing pandemic affecting over 200 countries and regions. Inference about the transmission dynamics of COVID-19 can provide important insights into the speed of disease spread and the effects of mitigation policies. We develop a novel Bayesian approach to such inference based on a probabilistic compartmental model using data of daily confirmed COVID-19 cases. In particular, we consider a probabilistic extension of the classical susceptible-infectious-recovered model, which takes into account undocumented infections and allows the epidemiological parameters to vary over time. We estimate the disease transmission rate via a Gaussian process prior, which captures nonlinear changes over time without the need of specific parametric assumptions. We utilize a parallel-tempering Markov chain Monte Carlo algorithm to efficiently sample from the highly correlated posterior space. Predictions for future observations are done by sampling from their posterior predictive distributions. Performance of the proposed approach is assessed using simulated datasets. Finally, our approach is applied to COVID-19 data from four states of the United States: Washington, New York, California, and Illinois. An R package BaySIR is made available at https://github.com/tianjianzhou/BaySIR for the public to conduct independent analysis or reproduce the results in this paper.

Motivation & Objective

• To develop a flexible, Bayesian approach for modeling time-varying transmission dynamics of COVID-19 using limited daily confirmed case data.
• To account for undocumented infections, which are often omitted in standard models but critical due to testing limitations and asymptomatic spread.
• To estimate the effective reproduction number ($\mathcal{R}_e$) with coherent uncertainty quantification using informative priors and posterior predictive distributions.
• To improve computational efficiency and mixing in MCMC sampling through a parallel-tempering algorithm for high-dimensional, correlated posterior spaces.
• To provide a generalizable, publicly available modeling framework applicable to regions with limited epidemiological data.

Proposed method

• A probabilistic extension of the classic SIR model is used, incorporating susceptible (S), undocumented infectious (I^U), documented infectious (I^D), and recovered (R) compartments.
• The transmission rate $\beta_t$ is modeled nonparametrically via a Gaussian process prior to capture nonlinear, non-monotonic temporal changes without parametric assumptions.
• The model uses a state-space formulation where transitions between compartments are governed by binomial likelihoods based on infection and diagnosis probabilities.
• A parallel-tempering Markov chain Monte Carlo (PTMCMC) algorithm is employed to enhance mixing and convergence in high-dimensional, correlated posterior distributions.
• Future predictions are generated by sampling from the posterior predictive distribution of daily confirmed cases.
• The model allows for covariate incorporation in transmission and diagnosis rate equations, though only intercept and time trends are used due to identifiability and model simplicity.

Experimental results

Research questions

• RQ1 How can we flexibly estimate time-varying transmission rates of COVID-19 without assuming a specific parametric form?
• RQ2 To what extent do undocumented infections influence the accuracy of transmission dynamics inference?
• RQ3 How can we improve MCMC sampling efficiency in high-dimensional, correlated posterior spaces common in epidemiological models?
• RQ4 What are the trends in the effective reproduction number ($\mathcal{R}_e$) across U.S. states, and how do they relate to mitigation policies?
• RQ5 How can we generate reliable, uncertainty-quantified forecasts of future case counts from limited daily case data?

Key findings

• The effective reproduction number ($\mathcal{R}_e$) showed a consistent decreasing trend across all six U.S. states—Washington, New York, California, Florida, Texas, and Illinois—suggesting the impact of public health interventions.
• The model successfully captured nonlinear changes in transmission rates through Gaussian process priors, avoiding restrictive parametric assumptions.
• The parallel-tempering MCMC algorithm significantly improved convergence and mixing compared to standard MCMC, enabling reliable posterior inference.
• Posterior predictive distributions provided forecasted case counts that reflected current transmission trends, though with caution due to reliance on extrapolation of human behavior.
• The model's performance was validated on simulated datasets, demonstrating robustness and accuracy in parameter estimation.
• An R package, BaySIR, was developed and made publicly available at https://github.com/tianjianzhou/BaySIR for reproducibility and independent analysis.