Analysis and prediction of improved SEIR transmission dynamics model: taking the second outbreak of COVID-19 in Italy as an example

Ming Lü, Xu-yang Zheng, Weinan Jia, Chun-zhi TianFrontiers in Public Health 2023 · DOI

Abstract

This study aimed to predict the transmission trajectory of the 2019 Corona Virus Disease (COVID-19) and analyze the impact of preventive measures on the spread of the epidemic. Considering that tracking a long-term epidemic trajectory requires explanatory modeling with more complexities than short-term predictions, an improved Susceptible-Exposed-Infected-Removed (SEIR) transmission dynamic model is established. The model depends on defining various parameters that describe both the virus and the population under study. However, it is likely that several of these parameters will exhibit significant variations among different states. Therefore, regression algorithms and heuristic algorithms were developed to effectively adapt the population-dependent parameters and ensure accurate fitting of the SEIR model to data for any specific state. In this study, we consider the second outbreak of COVID-19 in Italy as a case study, which occurred in August 2020. We divide the epidemic data from February to September of the same year into two distinct stages for analysis. The numerical results demonstrate that the improved SEIR model effectively simulates and predicts the transmission trajectories of the Italian epidemic during both periods before and after the second outbreak. By analyzing the impact of anti-epidemic measures on the spread of the disease, our findings emphasize the significance of implementing anti-epidemic preventive measures in COVID-19 modeling.

Introduction

Classical epidemiological models often struggle with long-term predictions because they treat parameters—such as the contact rate or recovery rate—as constants. In reality, these factors are highly fluid, shifting in response to government mandates, improvements in medical care, and changes in public behavior. The authors of this study address the "long-term tracking" gap, specifically focusing on the challenge of modeling a resurgence or "second wave." Using the second outbreak of COVID-19 in Italy (August 2020) as a case study, they developed a model that doesn't just fit historical data but remains robust as the social and medical landscape evolves.

The Improved SEIR Framework

The researchers proposed an enhanced SEIR model that expands the traditional four categories into a more granular system to better reflect real-world dynamics. This model distinguishes between asymptomatic ($A$) and symptomatic ($I$) infections and adds specific categories for hospitalized ($H$), deceased ($D$), quarantined susceptible ($S_q$), and quarantined exposed ($E_q$).

Figure 1: Population transformation diagram of the improved SEIR model, illustrating the flow between various compartments including quarantined and asymptomatic states.

Instead of fixed values, the model utilizes time-varying functions for key parameters:

  • Contact Rate ($c$): Modeled as a decaying function that reaches a minimum steady state ($b$) to reflect social distancing and lockdown measures.
  • Detection Capability ($T$): Modeled to reflect improvements in testing and diagnostic efficiency over time.
  • Recovery Rate ($\gamma$): Modeled using a nonlinear regression equation to account for improvements in clinical treatment.

Methodology and Optimization

To calibrate these dynamic parameters, the study integrated machine learning and heuristic algorithms. Gradient Descent was employed to solve regression coefficients for the daily recovery rate, ensuring the model could track the increasing efficiency of medical interventions.

Figure 6: Comparison of forecast and true values for the recovery rate over time, demonstrating the model's ability to capture temporal shifts in medical outcomes through regression.

Additionally, a Genetic Algorithm (GA) was used to calibrate state-dependent parameters, such as the infection rate and quarantine ratio. This heuristic approach treats sets of parameters as "agents" and evolves them over successive generations to find the optimal fit for the specific regional data.

Case Study: Italy’s Second Wave

The study utilized official daily COVID-19 data from Italy, covering February to September 2020. The analysis was divided into two stages:

  1. Initial Wave (Feb–July): Data from February to May was used for training, while June to July served as the validation set. During this stage, actual daily cumulative cases stayed almost entirely within the model’s 95% confidence interval.
  2. Second Outbreak (Aug–Sept): Data from August was used to refit the model to account for the resurgence. The model correctly predicted that the situation would stabilize by late October 2020, with cumulative cases peaking around November 1st.

Policy Insights and Sensitivity Analysis

A key finding of the research is the quantifiable impact of specific interventions. Sensitivity analysis revealed that the track quarantine ratio ($q$) is a critical lever; even small increases in the efficiency of identifying and isolating close contacts significantly lowered the peak of cumulative infections. Conversely, reducing the track quarantine ratio to 0.6$q$ led to a sharp spike in predicted infections.

Furthermore, the results suggest that lowering the proportion of population movement (represented by the minimum contact rate $b$) can achieve outcomes similar to complete lockdowns. This provides a mathematical basis for choosing "targeted" restrictions over "blanket" lockdowns to balance public health and economic activity.

Conclusion

The improved SEIR model demonstrates that long-term predictive accuracy requires models to "learn" how parameters change over time. While the current model assumes a closed system without migration, it provides a robust framework for simulating "what-if" policy scenarios. Future iterations may incorporate spatial distribution and transportation networks to account for viral movement between regions, allowing for even more precise local interventions.

Summary updated

Abstract

This study aimed to predict the transmission trajectory of the 2019 Corona Virus Disease (COVID-19) and analyze the impact of preventive measures on the spread of the epidemic. Considering that tracking a long-term epidemic trajectory requires explanatory modeling with more complexities than short-term predictions, an improved Susceptible-Exposed-Infected-Removed (SEIR) transmission dynamic model is established. The model depends on defining various parameters that describe both the virus and the population under study. However, it is likely that several of these parameters will exhibit significant variations among different states. Therefore, regression algorithms and heuristic algorithms were developed to effectively adapt the population-dependent parameters and ensure accurate fitting of the SEIR model to data for any specific state. In this study, we consider the second outbreak of COVID-19 in Italy as a case study, which occurred in August 2020. We divide the epidemic data from February to September of the same year into two distinct stages for analysis. The numerical results demonstrate that the improved SEIR model effectively simulates and predicts the transmission trajectories of the Italian epidemic during both periods before and after the second outbreak. By analyzing the impact of anti-epidemic measures on the spread of the disease, our findings emphasize the significance of implementing anti-epidemic preventive measures in COVID-19 modeling.
BibTeX
@inproceedings{lu2023analysis,
  author = {Ming Lü and Xu-yang Zheng and Weinan Jia and Chun-zhi Tian},
  title = {{Analysis and prediction of improved SEIR transmission dynamics model: taking the second outbreak of COVID-19 in Italy as an example}},
  booktitle = {Frontiers in Public Health},
  year = {2023},
  doi = {10.3389/fpubh.2023.1223039},
}
Figure