Why is it difficult to accurately predict the COVID-19 epidemic?

Area of Science:

• Epidemiology
• Mathematical Modeling
• Infectious Disease Dynamics

Background:

• The COVID-19 pandemic, originating in Wuhan in December 2019, spurred numerous epidemic predictions with significant variations.
• Nonidentifiability in model calibration using confirmed-case data is a primary driver of these prediction discrepancies.

Purpose of the Study:

• To investigate the reasons behind the wide variations in COVID-19 epidemic model predictions.
• To evaluate the performance of different epidemiological models (SIR vs. SEIR) using confirmed-case data.
• To model the impact of control measures and predict future epidemic trajectories in Wuhan.

Main Methods:

• Utilized the Akaike Information Criterion (AIC) for robust model selection between Susceptible-Infectious-Recovered (SIR) and Susceptible-Exposed-Infectious-Recovered (SEIR) models.
• Calibrated epidemiological models using confirmed COVID-19 case data from Wuhan.
• Simulated epidemic scenarios to assess the effects of lockdown, quarantine, and return-to-work policies.

Main Results:

• Demonstrated that nonidentifiability in model calibration is the principal cause of prediction variability.
• Showcased that the simpler SIR model provided a superior fit to confirmed-case data compared to the more complex SEIR model.
• Presented model predictions for the Wuhan epidemic post-lockdown and analyzed the impact of stringent quarantine measures.

Conclusions:

• Simpler epidemiological models can be more reliable than complex ones when calibrated with confirmed-case data due to identifiability issues.
• Effective quarantine measures significantly altered the epidemic's time course in Wuhan.
• Modeling suggests potential for a second outbreak following the return-to-work period, necessitating continued vigilance.