URL

https://www.medrxiv.org/content/10.1101/2020.11.05.20226597v2

Type d’article

Preprint

Thème

Modelisation of the dynamics of COVID-19

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

This article proposes a new criterion called the acceleration index to track the dynamics of the pandemic in real time. More specifically, this index along with illustrative plots indicates if the pandemic is accelerating (with more new cases per new test) or if it is decelerating (with less new cases per new test).

Objectifs de l’étude / Questions abordées

The objective is first to propose a new index, based on real-time data (number of cases, number of tests) to assess the level of acceleration/deceleration of the disease. Second, to derive a test strategy based on the new index.

Méthode

The authors propose a simple way to look at the data in real time by plotting the number of cases against the number of tests. Then the derivative of the curve gives information about the acceleration or deceleration of the pandemic. The acceleration index is developed based on the elasticity concept used in economy. The plot and the index are illustrated on the French data (Agence Santé Publique France) from May 13th 2020 to October 25th 2020. More refined analyses are also presented on the French data, over age groups and space.

Résultats principaux

The first plots on the pandemic in France, from May 13th 2020 to October 25th 2020, clearly reveal that the pandemic was decreasing between May 13th and June 13th, then it slightly started to increase between June 13th and July 13th, while from August 13th onwards acceleration was clearly taking place. It is also shown that the acceleration index started to precisely increase between July 7th and July 23rd which was an early warning of the future rise in the average positivity rate (which started two weeks later). Using the acceleration index by age groups shows that the pandemic acceleration has been stronger than national average for the [59−68] age group, and especially the 69 and older age groups, since early September, the latter being associated with the strongest acceleration index, as of October 25th. In contrast, acceleration among the [19−28] age group has been the lowest and about half that of the [69−78] age group (which was a surprising and interesting result).

The authors propose an acceleration-based test strategy, instead of a population-based test strategy, as currently performed in France. Their strategy aims at answering the question of how to geographically allocate a limited number of tests. They determine the relevance of allocating a certain number of tests to a French department based on the acceleration index and they evaluate the feedback effect that testing would have on the virus spread. The authors compare both strategies (population- and acceleration-based) using the Jensen-Shannon divergence.

Commentaire / brève évaluation, limites, ouvertures possibles

This article provides an interesting way to look at the data. It shows that following the number of cases, of tests and even the positivity rate of tests over time is far from the best way to measure the acceleration/deceleration of the virus spread from the results of performed tests. Instead, the authors’ plots and acceleration index take into account the variability in the number of performed tests. The authors argue that the acceleration index would have been able to detect the rise in the average positivity rate that occurred in August 2020 in France sooner and propose to use it to detect future changes (deceleration/acceleration) in the pandemic and to adopt public health policies. The view that young people such as college students have contributed a lot to the resurgence of the virus since last summer is refuted from this article as it is shown that the [19−28] age group had the lowest acceleration index as compared to the [59−68] and [69−78] age groups.

However, the justification of this indicator and the proposed test strategy seems to rely on the hypothesis that tests are performed at random in the population and hence that this sampling is stable over time. This is not obvious and should be discussed, since reasons why people are tested may vary over time (for example before holidays). Therefore, the use of the acceleration index in designing optimal testing strategies would necessitate additional investigations concerning the robustness of this index to possible testing biases.

One could also notice, that, due to the definition of the acceleration index, the proposed testing strategy is based rather on the ratio between the daily number of positive tests and the daily total number of tests, and not on the acceleration index itself. Another point could be noticed: the acceleration index is an estimator of the final slope of the curve given by the ratio between positive tests and total number of tests. The estimator used by the authors is based only on the last two points of this curve, and is therefore quite noisy, as can be seen in several figures of the article. It might be a good idea to use more than two time points to construct the acceleration index.