URL

https://arxiv.org/pdf/2009.09422.pdf

Type d’article

Preprint

Thème

Stratégies de contrôle

Que retenir de cet article, en 1-2 phrases ?

In this work, the authors propose a Bayesian inference method to estimate the risk of any individual to be infected by COVID19 in order to optimize testing and quarantining strategies.

Objectifs de l’étude / Questions abordées

In this work, the authors:

  • propose two algorithms for estimating the individual risk of infection,
  • test the inference of risk on two types of epidemic models,
  • show that these methods can be used as an alternative of the current contact tracing to better control the epidemic.

Méthode

The authors estimate the risk of infection by Bayesian inference using as prior information the list of recent contacts, associated own risks and personal information (results of tests, presence of symptoms). They present two algorithms, based on message-passing principles, inspired by previous works and adapted to the contact tracing problem they focus on: the first one is based on Belief Propagation and is more accurate, the second one is a simpler approximation, based on the Mean-Field method. The authors simulated data with a varying number of “patients-0” and number of performed tests. Two epidemic models are compared: SIR spreading model and a more realistic one, the Oxford OpenABM model.

Résultats principaux

To measure the ability of the proposed method to control the epidemic, the authors compare both algorithms with Random Guessing (where the tested individuals are randomly chosen) and the actual Contact Tracing (where the individuals are ranked and tested according to their number of risked contacts) using the two epidemic models. They show that if the number of available tests is not too small, even if only tested positive individuals are confined, these strategies allow to control the epidemic spreading within 100 days. To a lesser extent, they show that this is also the case for more realistic epidemic scenarios, where the sensitivity of medical tests is not 100% and only a fraction of the population download the application.

Commentaire / brève évaluation, limites, ouvertures possibles

This paper is a nice work, well-written with strong simulation results. To improve the paper, one could consider adding some comparison with related studies , even if the authors indicate that this should be done in future work.