BETS: The dangers of selection bias in early analyses of the coronavirus disease (COVID-19) pandemic

Qingyuan Zhao, Nianqiao Ju, Sergio Bacallado, Rajen D. Shah

Abstract

The coronavirus disease 2019 (COVID-19) has quickly grown from a regional outbreak in Wuhan, China to a global pandemic. Early estimates of the epidemic growth and incubation period of COVID-19 may have been biased due to sample selection. Using detailed case reports from 14 locations in and outside mainland China, we obtained 378 Wuhan-exported cases who left Wuhan before an abrupt travel quarantine. We developed a generative model we call BETS for four key epidemiological events—Beginning of exposure, End of exposure, time of Transmission, and time of Symptom onset (BETS)—and derived explicit formulas to correct for the sample selection. We gave a detailed illustration of why some early and highly influential analyses of the COVID-19 pandemic were severely biased. All our analyses, regardless of which subsample and model were being used, point to an epidemic doubling time of 2 to 2.5 days during the early outbreak in Wuhan. A Bayesian nonparametric analysis further suggests that about 5% of the symptomatic cases may not develop symptoms within 14 days of infection and that men may be much more likely than women to develop symptoms within 2 days of infection.

Introduction

In the early stages of the COVID-19 pandemic, researchers faced the critical task of estimating the epidemic's growth rate and the virus's incubation period. These parameters are vital for public health planning, yet early estimates varied significantly. For instance, doubling times were reported anywhere between 2 and 7 days. This paper identifies systematic selection bias as the primary driver of these discrepancies.

Early datasets relied heavily on "exported cases"—individuals who traveled from Wuhan and were subsequently diagnosed elsewhere. These samples were subject to three major biases:

  1. The Travel Quarantine: The sudden lockdown of Wuhan on January 23, 2020, truncated the exposure window for travelers.
  2. Exponential Growth: In a rapidly expanding epidemic, individuals are statistically more likely to have been infected recently. Failing to account for this "exponential tilt" leads to an overestimation of the incubation period.
  3. Right-Truncation: Studies conducted early in the outbreak only observed cases confirmed by a specific cutoff date, missing those with longer incubation periods.

The BETS Generative Model

To address these biases, we proposed the BETS model, which tracks four key events: Beginning of exposure, End of exposure, time of Transmission (infection), and time of Symptom onset.

Unlike traditional models that assume infection times are uniformly distributed within an exposure window, BETS incorporates the epidemic's exponential growth directly into the likelihood function. By deriving closed-form formulas, we adjusted for the January 23rd travel ban and the fact that only symptomatic exported cases were observable.

Data and Geographic Scope

The study utilized a curated dataset of 1,460 confirmed cases across 14 locations, including 8 cities/provinces in mainland China and 6 international regions. A high-quality subset of 378 Wuhan-exported cases was selected for primary analysis because their travel history provided well-defined exposure windows.

Geographic distribution of locations used in the study, highlighting Wuhan as the epicenter.

Key Results and Correcting Bias

Our analysis revealed that some early influential studies arrived at roughly correct median incubation periods only through a "balanced out" coincidence: the underestimation from right-truncation was offset by the overestimation from ignoring exponential growth.

Symptom onset distributions across various locations. The dashed green line indicates the January 23 travel quarantine.

Epidemic Growth

Correcting for the travel quarantine, we consistently estimated a doubling time of 2.0 to 2.5 days (specifically ~2.1 days) across all models. This is significantly faster than the 6.4 days reported in some early high-profile publications, highlighting the extreme speed of the initial outbreak.

Incubation Period

  • Median: Approximately 4.5 days.
  • The 14-Day Tail: Bayesian nonparametric analysis suggests the incubation distribution has a heavier tail than standard Gamma distributions. Approximately 5% of symptomatic cases may not develop symptoms until after the standard 14-day quarantine period.
  • Demographics: Data suggests men may develop symptoms more rapidly (within 2 days) than women, though they also exhibit greater variability in incubation times.

Conclusion

The BETS model demonstrates that ignoring the dynamics of an expanding epidemic and the impact of sudden policy changes (like quarantines) leads to severe bias. The finding of a 2-day doubling time underscores the necessity of rapid, aggressive intervention in the face of such a fast-moving pathogen. The potential for a 14-day-plus incubation period in 5% of cases also suggests that quarantine protocols may need to be more flexible or extended for high-risk individuals.

The dataset and statistical tools are available in the R package bets.covid19.

Summary updated

Abstract

The coronavirus disease 2019 (COVID-19) has quickly grown from a regional outbreak in Wuhan, China to a global pandemic. Early estimates of the epidemic growth and incubation period of COVID-19 may have been biased due to sample selection. Using detailed case reports from 14 locations in and outside mainland China, we obtained 378 Wuhan-exported cases who left Wuhan before an abrupt travel quarantine. We developed a generative model we call BETS for four key epidemiological events---Beginning of exposure, End of exposure, time of Transmission, and time of Symptom onset (BETS)---and derived explicit formulas to correct for the sample selection. We gave a detailed illustration of why some early and highly influential analyses of the COVID-19 pandemic were severely biased. All our analyses, regardless of which subsample and model were being used, point to an epidemic doubling time of 2 to 2.5 days during the early outbreak in Wuhan. A Bayesian nonparametric analysis further suggests that about 5% of the symptomatic cases may not develop symptoms within 14 days of infection and that men may be much more likely than women to develop symptoms within 2 days of infection.
BibTeX
@article{zhao2021bets,
  author = {Qingyuan Zhao and Nianqiao Ju and Sergio Bacallado and Rajen D. Shah},
  title = {{BETS: The dangers of selection bias in early analyses of the coronavirus disease (COVID-19) pandemic}},
  journal = {The Annals of Applied Statistics},
  year = {2021},
  doi = {10.1214/20-aoas1401},
}
Figure