Abstract
The basic reproduction number $\mathcal{R}_0$ of the coronavirus disease 2019 has been estimated to range between 2 and 4. Here we used a SEIR model that properly accounts for the distribution of the latent period and, based on empirical estimates of the doubling time in the near-exponential phases of epidemic progression in several locations, we estimated that $\mathcal{R}_0$ lies in the range 4.7–11.4. We explained this discrepancy by performing stochastic simulations of model dynamics in a population with a small proportion of super-spreaders. The simulations revealed two-phase dynamics, in which an initial phase of relatively slow epidemic progression diverts to a faster phase upon appearance of infectious super-spreaders. Early estimates obtained for this initial phase may suggest lower $\mathcal{R}_0$.
Introduction
At the onset of the COVID-19 pandemic, the World Health Organization and early epidemiological studies estimated the basic reproduction number ($\mathcal{R}_0$) to be between 2 and 4. These figures suggested a virus that could be contained through moderate social distancing and contact tracing. However, the rapid surge of cases in regions like Italy and New York State—where healthcare systems were overwhelmed despite early interventions—indicated a much more aggressive transmission profile.
This research identifies two critical gaps in early modeling that led to these underestimations. First, standard models often oversimplify the latent period (the time an individual is exposed but not yet infectious). Second, early data often missed the "two-phase" nature of the outbreak, where initial slow growth suddenly accelerates following super-spreading events.
Methods
The researchers employed a modified SEIR (Susceptible-Exposed-Infectious-Removed) model to re-evaluate the virus's reproductive potential.
- Erlang Distributions: Rather than assuming a simple exponential distribution for transitions between states, the authors used Erlang distributions ($m=6$ for the exposed state). This more accurately reflects the biological reality that the probability of becoming infectious peaks after a specific duration rather than being constant from the moment of exposure.
- Doubling Time Analysis: The study analyzed empirical data from eight regions (China, Italy, Spain, France, UK, Germany, Switzerland, and New York State). Doubling times ($\mathcal{T}_d$) were calculated using linear regression on the logarithm of cumulative cases and deaths during the peak exponential growth phase.
- Stochastic Simulations: To model the impact of super-spreaders, the team ran 5,000 simulations using the Gillespie algorithm. They tested scenarios where a tiny fraction of the population (1%) was responsible for a disproportionate majority (66.7%) of infections.
Key Results
Analysis of the empirical data revealed remarkably short doubling times across all studied regions, typically ranging from 1.86 to 2.96 days. When these doubling times are applied to a model with a realistic latent period distribution, the estimated $\mathcal{R}_0$ increases dramatically.
The study estimates that the true $\mathcal{R}_0$ for COVID-19 lies between 4.7 and 11.4. This revised range explains why moderate restrictions were insufficient to halt the initial spread. For instance, in countries like Italy and Spain, lockdowns needed to achieve at least a five-fold reduction in the infection rate to stop growth.
Key Insights: The Role of Model Structure and Super-Spreaders
A central finding of the paper is that for any observed doubling time, the calculated $\mathcal{R}_0$ is highly sensitive to the assumed distribution of the latent period. As shown in the theoretical curves, models that account for a more structured latent period (higher $m$ values in the Erlang distribution) require a higher $\mathcal{R}_0$ to produce the same observed growth rate.
Furthermore, the stochastic simulations demonstrated that super-spreaders act as "phase shifters." An epidemic can "simmer" at a low growth rate for weeks during the initial transmission chains. The moment a super-spreader event occurs, the doubling time drops sharply. If researchers estimate $\mathcal{R}_0$ during the initial "simmering" phase, they will significantly underestimate the virus's explosive potential once it reaches the general population.
Discussion and Implications
These findings have profound implications for public health policy, particularly regarding herd immunity. If $\mathcal{R}_0$ is approximately 3, the herd immunity threshold is roughly 67%. However, if $\mathcal{R}_0$ is closer to 10, the threshold for population-level protection jumps to 90%, necessitating much higher vaccination coverage.
While the study provides a more robust framework for understanding early pandemic dynamics, it acknowledges limitations, such as the assumption that the latent period is identical to the incubation period and the use of a "perfectly mixed" population model. Future work focusing on spatial heterogeneity and social network structures could further refine these estimates.
Summary updated