Abstract
To reduce the biases of traditional survey-based methods, this paper proposes an epidemic model-based approach to inference the incubation period distribution of COVID-19 utilizing the publicly reported confirmed case number. We construct an epidemic model, namely SEAIR, and take advantage of the dynamic transmission process depicted by SEAIR to estimate the onset probability in each day of exposed individuals in eight impacted countries. Based on these estimations, the general incubation probability distribution of COVID-19 has been revealed. The proposed method can avoid several biases of traditional survey-based methods. However, due to the mathematical-model-based nature of this method, the inference results are somewhat sensitive to the setting of parameters. Therefore, this method should be practiced reasonably on the basis of a certain understanding of the studied epidemic.
Introduction
Determining the incubation period—the time between infection and the onset of symptoms—is critical for public health policy. It dictates the length of quarantine periods and informs epidemiological models of viral spread. Traditionally, these data are gathered through patient surveys. However, survey-based methods are prone to "noise," including recall bias (patients misremembering exposure dates), small sample sizes, and measurement errors from contact tracers.
This study proposes a "top-down" approach: inferring the biological distribution of the incubation period from aggregate daily confirmed case counts. By treating the population as a single system, this model-based approach bypasses individual-level errors and provides a more objective estimation of the virus's natural progression.
Methodology: The SEAIR Model
The researchers developed a modified SEAIR model, extending the classic Susceptible-Infectious-Removed (SIR) framework. The population is categorized into five distinct states:
- S (Susceptible): Healthy individuals at risk of infection.
- E (Exposed): Individuals who are infected but not yet symptomatic or infectious.
- A (Asymptomatic): Individuals who are infected and infectious but never develop symptoms.
- I (Infected): Individuals who are symptomatic and infectious.
- R (Removed): Individuals who have recovered or deceased.
A key innovation in this model is the introduction of a discrete random probability distribution for the time delay between the "Exposed" state and the onset of symptoms (the transition to "Infected" or "Asymptomatic"). To solve for these incubation probabilities, the transmission process was transformed into a system of multivariate linear equations. Given the high dimensionality and potential for local minima, the authors utilized the basin-hopping algorithm, a global optimization technique, to find the best fit for the observed data.
Data and Calibration
The study utilized publicly available data from January 21 to April 13, 2020, across eight countries: the US, Iran, Canada, Germany, Turkey, Austria, France, and Italy. The model was calibrated using a reproduction rate ($R_0$) of 3.77 and an asymptomatic proportion of 17.9%, based on early clinical studies and data from the Diamond Princess cruise ship.
Results and Sensitivity Analysis
The model consistently identified a peak in the incubation distribution on the fourth day across different countries, suggesting a robust biological signal. Using the primary model settings (containment parameter $\lambda = 0.1$), the study estimated a median incubation period of 4.76 days (95% CI: 4.68–4.84) and a mean (expectation) of 6.34 days (95% CI: 6.27–6.42).
The researchers also investigated the role of government containment efforts, represented by the parameter $\lambda$. As shown in the sensitivity analysis, the model's predictions shift as the damping effect of interventions changes. Interestingly, as government control weakens (higher $\lambda$), the model’s predictions align more closely with traditional survey results, suggesting that surveys may more effectively capture the "natural" incubation period when viral spread is less restricted.
Discussion and Limitations
While the SEAIR model-based approach successfully avoids the biases inherent in patient surveys, it is sensitive to input parameters. The accuracy of the resulting incubation distribution depends heavily on the initial assumptions for $R_0$ and the asymptomatic ratio ($\gamma$). Consequently, this method should be applied with a deep understanding of the specific epidemic's characteristics. Future work may involve refining these parameters as more biological data becomes available or adapting the model for different viral variants with distinct incubation profiles.
Summary updated