D2EA: Depict the Epidemic Picture of COVID-19

Chenzhengyi Liu, Jingwei Zhao, Guohang Liu, Yuanning Gao, Xiaofeng GaoJournal of Shanghai Jiaotong University (Science) 2020 · DOI

Abstract

In the early months of 2020, researchers at Shanghai Jiao Tong University developed a specialized mathematical framework to model the COVID-19 outbreak in Hubei Province. This model, known as the Delay Differential Epidemic Analyzer (D2EA), extends the classic SEIR framework by incorporating quarantine dynamics and time-lagged transitions. By accounting for government interventions and administrative delays, the D2EA model provided accurate forecasts of the infection peak and offered a data-driven roadmap for the eventual reopening of society.

Introduction

Standard epidemiological models, such as the Susceptible-Exposed-Infectious-Recovered (SEIR) model, often struggle to capture the dynamics of an epidemic when aggressive government interventions are present. These traditional models typically assume "free mixing," where every individual in a population has an equal probability of contacting any other.

However, the early COVID-19 outbreak in Hubei Province broke these assumptions due to two primary factors:

  1. Massive Interventions: Lockdowns ("city blocking") and mandatory quarantines fundamentally altered social contact patterns.
  2. Time Delays: Significant lags existed between exposure, symptom onset, testing, and official confirmation or quarantine.

The D2EA model was designed to bridge this gap, incorporating the impact of the "Chunyun" (Lunar New Year) travel rush, lockdown measures, and the biological/administrative delays inherent in disease confirmation.

The D2EA Model Framework

The core innovation of the D2EA model lies in its transition from simple state categories to a more complex system that accounts for quarantine and testing status.

Quarantine States

Beyond the standard SEIR states, the model introduces five specific quarantine categories:

  • $Q_S$: Quarantined healthy individuals.
  • $Q_E$ & $Q_I$: Quarantined exposed and infectious individuals.
  • $Q_N$ & $Q_P$: Suspected cases that test negative or positive, respectively.
The D2EA model structure, incorporating quarantine states ($Q_S, Q_E, Q_I, Q_N, Q_P$) to account for government interventions.

Delay Differential Equations (DDEs)

Unlike Ordinary Differential Equations (ODEs), which rely solely on current states, D2EA utilizes Delay Differential Equations (DDEs) to incorporate past states. This is crucial for modeling the time $T$ required to confirm a case and the duration $\tau$ of quarantine. This approach reflects the reality of early pandemic testing bottlenecks, where laboratory capacity was a limiting factor in controlling the spread.

Dynamic Parameter Estimation

Rather than assuming static rates, the D2EA model employs dynamic parameters that evolve over time to reflect changing social conditions and medical practices.

Dynamic modeling of contact rate ($r$), recovery rate ($\gamma$), and quarantine rate ($\delta$) over time.
  • Contact Rate ($r$): Modeled as a piecewise function combining a Gaussian distribution (to capture the travel rush) and an exponential decay (to reflect lockdown effectiveness).
  • Recovery Rate ($\gamma$): Adjusted as medical understanding and treatment protocols improved.
  • Quarantine Rate ($\delta$): Observed to increase sharply around February 11, 2020, when clinical diagnosis (CT scans) began supplementing nucleic acid tests to accelerate confirmation.

Key Results and Insights

The D2EA model was fitted using real-world data from the National Health Commission of China and the CDC, focusing on Hubei's population of approximately 60 million.

  • Peak Prediction: The model accurately forecasted that accumulated confirmed cases in Hubei would peak at the end of February 2020, stabilizing around 65,000 cases. Historical data later confirmed this, with Hubei reaching approximately 67,000 cases in that timeframe.
  • Impact of Intervention: By comparing D2EA against a standard SEIR model without quarantine, the researchers found that interventions reduced the number of exposed and infectious individuals to just 1/10th of what they would have been otherwise.
  • Reopening Strategy: Sensitivity analysis suggested that Hubei should delay reopening until at least late March 2020 to prevent a second wave of infections.

Limitations and Future Work

While highly effective for short-term modeling of the first wave, the D2EA model has several limitations:

  • Homogeneity: It assumes a uniform population, neglecting differences in age-based susceptibility or geographic density (e.g., Wuhan vs. rural Hubei).
  • Reinfection: The model assumes short-term immunity, which may not hold for multi-year endemic scenarios.
  • Spatial Heterogeneity: Future iterations could improve by modeling the spread between different cities and incorporating the impact of medical breakthroughs like vaccines.

Summary updated

BibTeX
@article{liu2020d,
  author = {Chenzhengyi Liu and Jingwei Zhao and Guohang Liu and Yuanning Gao and Xiaofeng Gao},
  title = {{D2EA: Depict the Epidemic Picture of COVID-19}},
  journal = {Journal of Shanghai Jiaotong University (Science)},
  year = {2020},
  doi = {10.1007/s12204-020-2170-7},
}
Figure