Tracking and Predicting COVID-19 Epidemic in China Mainland

Haoxuan Sun, Yumou Qiu, Han Yan, Yaxuan Huang, Yuru Zhu, Song Xi ChenmedRxiv 2020 · DOI

Abstract

By proposing a varying coefficient Susceptible-Infected-Removal model (vSIR), we track the epidemic of COVID-19 in 30 provinces in China and 15 cities in Hubei province, the epicenter of the outbreak. It is found that the spread of COVID-19 has been significantly slowing down within the two weeks from January 27 to February 10th with 87.0% and 84.3% reductions in the reproduction number $R_0$ among the 30 provinces and 15 Hubei cities, respectively. This suggests the extreme control measures implemented since January 23, which include cutting off Wuhan and many other cities and towns, a great public awareness and high level of self isolation at home, have contributed to a substantial decline in the reproductivity of the COVID-19 in China. We predict that Hubei province will reach its peak between February 20 and 22, 2020, and if the removal rate can be increased to 0.1, the epidemic outside Hubei province will end in May 2020, and inside Hubei in early June.

Introduction

Published in the early weeks of the COVID-19 pandemic (February 2020), this research provides a rigorous statistical post-mortem and forecast of the initial outbreak in China. By moving away from static epidemiological models, the authors quantify exactly how effective the "extreme control measures" were in bending the curve. The study serves as an early quantitative validation of aggressive public health interventions.

The Problem with Static Modeling

Standard epidemiological models, such as the classic SIR (Susceptible-Infected-Removed) or SEIR (Susceptible-Exposed-Infected-Removed) frameworks, typically assume that the parameters governing disease spread—specifically the transmission rate—remain constant over time. While these assumptions hold for natural, unmitigated outbreaks, they are insufficient during a global health crisis where government interventions, widespread mask-wearing, and self-isolation drastically alter the environment.

The authors argue that a constant $R_0$ (the basic reproduction number) is an overly idealistic assumption for a dynamic crisis. To address this, they developed a model capable of tracking the changing speed of the epidemic in real-time, allowing for an assessment of whether containment measures were working and providing a basis for predicting the peak of infections.

Methodology: The vSIR Model

The core innovation of this research is the varying coefficient SIR (vSIR) model. This approach introduces several key statistical refinements:

  • Time-Varying Parameters: Unlike standard models, the infectious rate $\beta(t)$ and the reproduction number $R_0(t)$ are treated as dynamic functions of time.
  • Locally Weighted Regression: The authors utilized a nonparametric approach to estimate these parameters from daily reported data. This method "smooths" daily reporting noise while capturing the underlying trend of the intervention’s impact.
  • Reciprocal Regression for Prediction: To forecast future trends, the declining infection rate was modeled using a reciprocal regression ($1/t$ relationship). This captures the phenomenon where, as control measures tighten, the rate of new infections drops toward zero.
  • Bootstrap Inference: The study employed bootstrap resampling to generate 95% confidence intervals, providing a necessary measure of uncertainty for predicted peak and end dates.

Data Sources

The study integrated data from the National Health Commission of China (NHCC), covering daily records of infected, deceased, and recovered patients across 30 provinces and 15 cities in Hubei. Additionally, detailed case data from Shenzhen (391 cases) and Shaoyang (100 cases) were used to estimate the "infectious duration"—the critical window from symptom onset to hospital admission and quarantine.

Key Insights and Results

The "Two-Week Turnaround"

The vSIR model revealed a dramatic shift in transmission dynamics between January 27 and February 10, 2020. During this period, $R_0$ plummeted from levels as high as 7.0 to near or below 1.0. Specifically, $R_0$ dropped by 87.0% across the 30 provinces and 84.3% within Hubei cities. This provided early, quantitative evidence that the lockdowns and social distancing measures were effectively curtailing the virus.

Crossing the Critical Threshold

By February 10, the average $R_0$ for the 30 provinces had dropped to 0.77. Falling below the critical threshold of 1.0 is the mathematical turning point where an epidemic begins to die out rather than expand exponentially.

Peak and End-Date Predictions

The model accurately forecasted that Hubei province would reach its peak of active infections between February 20 and 22, 2020. Furthermore, the study highlighted the sensitivity of the timeline to the "removal rate" ($\gamma$). If the rate of diagnosis and quarantine could be maintained or increased to 0.1 (averaging 10 days from diagnosis to removal), the epidemic was predicted to end in May 2020 for most of China and early June for Hubei.

Discussion and Limitations

While the vSIR model provided a robust framework for tracking the outbreak, the authors noted several caveats. Early data from Hubei and Wuhan were likely subject to reporting delays or incompleteness, which could impact early-stage estimates.

Furthermore, the paper warned of the "impatient population" risk: the danger that individuals might abandon self-isolation prematurely due to economic pressure or fatigue, potentially triggering a resurgence. Finally, while the infection rate was modeled as time-varying, the removal rate was treated as relatively constant in the short term, suggesting an area for future model refinement as medical treatments and hospital capacities evolved.

Conclusion

The vSIR model demonstrates that aggressive public health interventions can drastically alter the trajectory of an epidemic. By quantifying the reduction in $R_0$, this research provided a blueprint for real-time monitoring of COVID-19 and underscored the importance of sustained control measures to reach the end of the outbreak.

Summary updated

Abstract

Abstract By proposing a varying coefficient Susceptible-Infected-Removal model (vSIR), we track the epidemic of COVID-19 in 30 provinces in China and 15 cities in Hubei province, the epicenter of the outbreak. It is found that the spread of COVID-19 has been significantly slowing down within the two weeks from January 27 to February 10th with 87.0% and 84.3% reductions in the reproduction number R 0 among the 30 provinces and 15 Hubei cities, respectively. This suggests the extreme control measures implemented since January 23, which include cutting off Wuhan and many other cities and towns, a great public awareness and high level of self isolation at home, have contributed to a substantial decline in the reproductivity of the COVID-19 in China. We predict that Hubei province will reach its peak between February 20 and 22, 2020, and if the removal rate can be increased to 0.1, the epidemic outside Hubei province will end in May 2020, and inside Hubei in early June.
BibTeX
@misc{sun2020tracking,
  author = {Haoxuan Sun and Yumou Qiu and Han Yan and Yaxuan Huang and Yuru Zhu and Song Xi Chen},
  title = {{Tracking and Predicting COVID-19 Epidemic in China Mainland}},
  year = {2020},
  doi = {10.1101/2020.02.17.20024257},
  howpublished = {medRxiv},
}
Figure