Estimation of incubation period distribution of COVID-19 using disease onset forward time: a novel cross-sectional and forward follow-up study

Jing Qin, Chong You, Qiushi Lin, Taojun Hu, Shicheng Yu, Xiao‐Hua ZhoumedRxiv 2020 · DOI

Abstract

Based on the estimated incubation distribution in this study, about 10% of patients with COVID-19 would not develop symptoms until 14 days after infection. Further study of the incubation distribution is warranted to directly estimate the proportion with long incubation periods.

Introduction

Accurately estimating the incubation period—the time from infection to symptom onset—is the cornerstone of effective public health policy. It dictates the length of quarantines, helps calculate the reproductive number ($R_0$), and informs contact tracing. Early in the COVID-19 pandemic, estimates were often hindered by recall bias (patients misremembering exposure dates) and small sample sizes.

This research addresses these challenges by applying renewal theory to travel data from the early days of the pandemic. By analyzing travelers leaving Wuhan before the January 2020 lockdown, researchers estimated a median incubation period of approximately 8.13 days—significantly longer than early estimates of 4–5 days.

Methodology: The Forward Time Approach

The researchers treated the development of the disease as a stochastic renewal process. Instead of trying to find the exact moment of infection (the "backward time"), they focused on the forward recurrence time: the duration between a person’s departure from Wuhan and their subsequent symptom onset.

Conceptual diagram of backward and forward time relative to departure from Wuhan.

Key technical aspects include:

  • Renewal Theory: Assuming the infection rate in Wuhan reached a steady state by late January, the statistical properties of "time since infection" and "time until symptoms" become mathematically related.
  • Length-Biased Sampling Correction: The authors accounted for the fact that people with longer incubation periods are more likely to be captured in a cross-sectional "snapshot" of travelers.
  • Distribution Fitting: They assumed the incubation period followed a Weibull distribution and used Maximum Likelihood Estimation (MLE) to fit the observed forward times of 1,211 cases.

Datasets

The study utilized publicly available data from provincial and municipal health commissions in China and international health ministries. From an initial pool of 12,963 confirmed cases outside Hubei province, a final cohort of 1,211 cases was selected. These individuals left Wuhan between January 19 and January 23, 2020, were asymptomatic at the time of departure, and later developed symptoms. This specific window was chosen to ensure infections likely occurred within Wuhan before the public was fully aware of the epidemic's severity.

Key Results and Comparison

The study found that the incubation period for COVID-19 was longer than previously reported in high-profile early studies (like those in the NEJM or The Lancet).

  • Median Incubation: 8.13 days (95% CI: 7.37–8.91).
  • Mean Incubation: 8.62 days (95% CI: 8.02–9.28).
  • 90th Percentile: 14.65 days (meaning 10% of cases take longer than this).
  • 99th Percentile: 20.59 days.
Comparison of incubation period estimates across SARS, MERS, and COVID-19 studies.

As shown in the table above, other early COVID-19 studies estimated the mean at approximately 5.2 to 5.8 days. The authors argue their larger sample size and bias-correction methods provide a more robust picture of the virus's behavior.

Discussion and Limitations

The most striking insight is the "long tail" of the distribution. While a 14-day quarantine was the global standard, this model suggests that 1 in 10 infected individuals would still be in the incubation phase when their 14-day isolation ended. This highlights a potential gap in early containment strategies.

However, several limitations apply:

  • Equilibrium Assumption: If the outbreak was growing exponentially rather than being in a steady state, the results might be biased.
  • Imported Cases: There is a small risk that some individuals were infected at their destination, which would lead to an overestimation.
  • Variant Specificity: These findings represent the ancestral strain of SARS-CoV-2; subsequent variants like Delta and Omicron have shown different incubation characteristics.
  • Asymptomatic Carriers: The study only includes those who eventually developed symptoms and does not account for "true" asymptomatic cases.

Summary updated

Abstract

Based on the estimated incubation distribution in this study, about 10% of patients with COVID-19 would not develop symptoms until 14 days after infection. Further study of the incubation distribution is warranted to directly estimate the proportion with long incubation periods.
BibTeX
@misc{qin2020estimation,
  author = {Jing Qin and Chong You and Qiushi Lin and Taojun Hu and Shicheng Yu and Xiao‐Hua Zhou},
  title = {{Estimation of incubation period distribution of COVID-19 using disease onset forward time: a novel cross-sectional and forward follow-up study}},
  year = {2020},
  doi = {10.1101/2020.03.06.20032417},
  howpublished = {medRxiv},
}
Figure