Abstract
The 2019-nCov outbreak in Italy presented significant challenges for early forecasting due to data noise and biological complexities. This study utilizes a modified SEIJR model with log-normal distributed time delays to better reflect clinical realities. By applying Particle Swarm Optimization (PSO), we performed an inversion analysis to estimate the outbreak's true start date and initial spreading velocity. Our findings indicate the virus began spreading around February 1, 2020, with an initial velocity of 0.330. While national lockdowns reduced this to 0.278, simulations suggest a threshold of 0.25 was necessary for containment by mid-April.
Introduction: Addressing the Inversion Problem
Early in the pandemic, official case counts often lagged behind the actual spread of the virus. Standard epidemiological models like SIR frequently fail to account for the non-linear nature of incubation and diagnosis periods. This paper addresses the "inversion problem": using observed confirmed cases to work backward and identify hidden parameters like the true start date ($T_0$) and the spreading velocity ($\alpha$).
Methodology: The SEIJR Model and PSO
The researchers expanded the traditional SEIR framework into a SEIJR model. This model includes a "J" category for confirmed/diagnosed individuals and incorporates log-normal distribution functions for transition periods, such as the time from infection to symptoms ($t_1$) and symptoms to diagnosis ($t_2$).
To solve the complex, non-differentiable equations of this model, the authors employed Particle Swarm Optimization (PSO). PSO is a metaheuristic algorithm that iteratively improves candidate solutions by simulating social behavior, such as bird flocking. This allowed for the precise estimation of $\alpha$ and $T_0$ without requiring gradient-based methods.
Key Results and Inversion Analysis
The inversion analysis pinpointed Italy's outbreak start to approximately February 1, 2020, roughly three weeks before the first official cluster was detected. The initial spreading velocity ($\alpha_0$) was calculated at 0.330, a value higher than most early-phase Chinese cities except Wuhan.
Following the national lockdown on March 9, the spreading velocity ($\alpha_1$) dropped to 0.278. This quantitative shift demonstrates the effectiveness of public health measures, though the study highlights that a further reduction to below 0.25 was critical for timely containment by mid-April.
Conclusion
By combining the SEIJR model with PSO-based inversion, this research provides a robust framework for understanding early outbreak dynamics. The ability to "look back" and identify the true start of an epidemic is vital for evaluating policy responses and preparing for future waves. The study underscores the high sensitivity of infection curves to even minor changes in spreading velocity, emphasizing the continued importance of strict public health precautions.
Code for this study is available at: https://github.com/Summerwork/2019-nCov-Prediction
Summary updated