URL :
https://doi.org/10.1016/ S1473-3099(20)30144-4

Type d’article :
Article peer-reviewed

Type de contenu :
Modèle, Estimation de paramètres, Données Epidémiologiques, Prédictions

Thème :
Epidémiologie

Que retenir ? :
Combining 4 publicly available datasets and using a stochastic SEIR model, the authors establish that R0 decreased in Wuhan following travel restrictions, and correctly fit data on imported cases in several countries. They also predict a probability >50% of a new outbreak as soon as 4 cases are reported.

Description de l’article :
The authors introduce a stochastic transmission model (geometric random walk process): it is an SEIR model accounting for delays in symptoms onset and reporting, uncertainty in observations modeled by a Poisson observed process of newly symptomatic cases, reported onsets of new cases, reported confirmation of cases, and a binomial observation process for infection prevalence on evacuation flights. The incubation period is assumed to be Erlang distributed with mean 5.2 days, and delay from onset to isolation is assumed to be Erlang distributed with mean 2.9 days. The delay from onset to reporting is assumed to be exponentially distributed with mean 6.1 days. The model has 3 unknown parameters that are estimated: magnitude of temporal variability in transmission, proportion of cases that would eventually be detectable, and relative probability of reporting a confirmed case within Wuhan compared with an internationally exported case that originated in Wuhan.

The authors use 4 publicly available datasets: daily number of new internationally exported cases (or lack thereof), by date of onset, as of Jan 26, 2020; daily number of new cases in Wuhan with no market exposure, by date of onset, between Dec 1, 2019, and Jan 1, 2020; daily number of new cases in China, by date of onset, between Dec 29, 2019, and Jan 23, 2020; and proportion of infected passengers on evacuation flights between Jan 29, 2020, and Feb 4, 2020. Once Rt is estimated, they use a branching process with a negative binomial offspring distribution to calculate the probability an introduced case would cause a large outbreak.

They estimate a median daily R0, going from 2.35 (1.15 ; 4.77) one week before travel restrictions to 1.05 (0.41 ; 2.39) one week after. Results suggest there were around ten times more symptomatic cases in Wuhan in late January than were reported as confirmed cases. The authors estimate that a single introduction of SARS-CoV-2 with SARS-like or Middle East respiratory syndrome (MERS)-like individual-level variation in transmission would have a 17% to 25% probability of causing a large outbreak. And assuming SARS-like variation and Wuhan-like transmission, once 4 or more infections have been introduced into a new location, there is an over 50% chance that an outbreak will occur.

Commentaire : The article is clearly written. The model and the datasets are well introduced, and results are very easy to find and understand. The authors estimated several parameters of importance. In particular they estimated the early dynamics of the virus in Wuhan (number of symptomatic cases). They also correctly fitted the number of cases exported from Wuhan to other countries, with notably bad results for France, USA and Australia, that may be due to restrictions measures taken in these countries as the epidemic was spreading worldwide. They also predicted that the probability for a new outbreak was high everywhere new cases were to be introduced. The methods are clearly presented and several parameter values are given (either estimated or obtained from the literature). Limitations: The main limitation is the fact that this work focuses on early dynamics of the virus, which have been later redefined. For instance, plausible biological parameters for SARS-CoV-2 have been used based on evidence in March, but these values should be refined as more comprehensive data are now available. Regarding the results, the model did not predict the recent slowdown in cases, suggesting that transmission might have declined more than the model estimated during early February, 2020.