Bayesian SIR model with change points with application to the Omicron wave in Singapore

Jiaqi Gu, Guosheng YinScientific 2022 · DOI

Abstract: The Omicron variant has led to a new wave of the COVID-19 pandemic worldwide, with unprecedented numbers of daily confirmed new cases in many countries and areas. To analyze the impact of society or policy changes on the development of the Omicron wave, the stochastic susceptible-infected-removed (SIR) model with change points is proposed to accommodate the situations where the transmission rate and the removal rate may vary significantly at change points. Bayesian inference based on a Markov chain Monte Carlo algorithm is developed to estimate both the locations of change points as well as the transmission rate and removal rate within each stage. Experiments on simulated data reveal the effectiveness of the proposed method, and several stages are detected in analyzing the Omicron wave data in Singapore.

Introduction

The classic Susceptible-Infected-Removed (SIR) model is a cornerstone of mathematical epidemiology. However, its standard formulation typically assumes that the disease transmission rate ($\beta$) and the removal rate ($\gamma$) remain constant throughout an outbreak. This assumption is often violated in long-term pandemics like COVID-19, where government interventions (lockdowns, mask mandates), individual behavioral shifts (social distancing), and the scaling of medical resources cause these rates to fluctuate significantly over time.

This paper addresses the challenge of modeling these "non-stationary" dynamics by introducing a stochastic SIR model that incorporates "change points." The goal is to detect precisely when these shifts occur and how the transmission and recovery rates evolve within each distinct stage of a pandemic wave.

Methodology: Stochastic SIR with Change Points

The researchers propose a discrete-time stochastic SIR model. Unlike deterministic differential equations, this model treats daily new infections and removals as binomial random variables, capturing the inherent randomness of daily reporting and transmission events.

Model Framework

The study period is partitioned into $K$ stages using a latent binary indicator vector ($\delta$), where a value of 1 indicates the start of a new stage (a change point). Within each stage, the parameters for transmission and removal are assumed to be distributionally homogeneous, allowing the model to adapt to the evolving nature of the virus and the public health response.

Stochastic SIR model with change points: (a) State transitions over time; (b) Transmission and removal rate shifts at detected change points.

Bayesian Inference and the Add-Delete-Swap Algorithm

To estimate the posterior distributions of the change points and the associated rates, the authors employ a hierarchical Bayesian framework. Navigating the complex space of potential partitions requires an efficient sampling strategy. The authors developed a specialized Markov Chain Monte Carlo (MCMC) move set known as the Add-Delete-Swap algorithm.

The Add-Delete-Swap MCMC algorithm for identifying optimal change point locations by proposing moves to add, remove, or shift indicators.

This algorithm allows the model to dynamically explore different numbers and locations of change points. Following the MCMC sampling, a clustering aggregation method is used to identify the single most representative set of change points, serving as the Bayes estimator for the pandemic's timeline.

Key Results

Simulation Performance

The model was first validated using simulated datasets across three scenarios with varying signal strengths. The model demonstrated high accuracy in recovering the "ground truth" partitions. In the most distinct scenario, the model achieved an Adjusted Rand Index (ARI) of 0.960, indicating near-perfect stage identification.

Singapore Omicron Wave Case Study

The model was applied to daily confirmed cases and removal data (recoveries and deaths) in Singapore from January 3 to April 21, 2022. The analysis identified four distinct change points, dividing the wave into five stages:

  1. Stage 1 (Initial Outbreak): Characterized by high transmission and low removal rates as the variant first spread.
  2. Stage 2 (Response): Transmission began to dip as public self-protection increased and medical resources were deployed.
  3. Stage 3 (Peak): Transmission dropped sharply following the wave's peak.
  4. Stage 4 (Reporting Anomaly): Interestingly, the model detected a sharp dip in removal rates between March 23 and April 2. This corresponded to a real-world period where recovery cases were not reported, demonstrating the model's utility in flagging data quality issues.
  5. Stage 5 (Stabilization): The wave stabilized with low transmission and consistent removal rates.

Discussion and Conclusion

By allowing $\beta$ and $\gamma$ to be time-varying, this model provides a much more flexible and realistic fit for real-world data than standard SIR models. It successfully quantifies social feedback loops, showing how transmission rates drop as populations react to an outbreak.

While highly effective for retrospective analysis, future work is needed to adapt the model for real-time detection, which is complicated by reporting delays. Additionally, the authors suggest that incorporating non-homogeneous mixing—such as interpersonal contact networks—could further refine the model's accuracy in predicting the spread of future variants.

Summary updated

Abstract

The Omicron variant has led to a new wave of the COVID-19 pandemic worldwide, with unprecedented numbers of daily confirmed new cases in many countries and areas. To analyze the impact of society or policy changes on the development of the Omicron wave, the stochastic susceptible-infected-removed (SIR) model with change points is proposed to accommodate the situations where the transmission rate and the removal rate may vary significantly at change points. Bayesian inference based on a Markov chain Monte Carlo algorithm is developed to estimate both the locations of change points as well as the transmission rate and removal rate within each stage. Experiments on simulated data reveal the effectiveness of the proposed method, and several stages are detected in analyzing the Omicron wave data in Singapore.
BibTeX
@inproceedings{gu2022bayesian,
  author = {Jiaqi Gu and Guosheng Yin},
  title = {{Bayesian SIR model with change points with application to the Omicron wave in Singapore}},
  booktitle = {Scientific},
  year = {2022},
  doi = {10.1038/s41598-022-25473-y},
}
Figure