In recent days, the calculators of the health authorities have had to do it several times to display good results. In its daily report Tuesday, the Directorate General of Health (DGS) reported new death figures due to the coronavirus in France, with an important correction: the day before, 342 deaths had been counted too many in Ehpad and other establishments medico-social.
It is not the first time that such an anomaly has occurred. Last week, the numbers were also overestimated twice: on May 12, 15 more deaths were counted. On May 14, 14 deaths were wrongly counted.
Where do these differences come from? For the former, the Directorate General of Health explained that it had corrected errors in the "data from several establishments in Ile-de-France", with corrections "made by the Agence Régionale de Santé d'Ile-de- France ". For those of Tuesday, she says that there was "a counting error (anomaly of 342) in the death toll of medico-social establishments".
Problems in the lifts
In principle, Santé Publique France, the institute which collects this data and transmits it to the DGS, collects the figures via several data feeds. On the one hand, it records “deaths occurring in hospital, deaths linked to Covid-19 certified by CépiDC (Center for epidemiology of medical causes of death) of Inserm, and the number of deaths occurring in communities of elderly people through a dedicated application, ”she explains in a dedicated press release.
On the other, it determines "excess mortality" from "surveillance of all-cause mortality, based on civil status data transmitted to the National Institute of Statistics and Economic Studies (Insee) ".
What, sometimes, generate errors. Bad civil status registrations can in fact distort the final figures. As Mediapart notes, some town halls count as “deaths at home” deaths that actually occurred in retirement homes. With, as a consequence, oversights or wrong calculations of excess mortality rate, influencing the calculations carried out by Inserm and INSEE, which also work from these data ... And which therefore complement the work of Public Health France. Contacted, the institute was unable to answer our questions on Wednesday.
Assess the trend on D-7
As data science engineer Guillaume Rozier, who notably created CovidTracker, a platform used to analyze the evolution of the epidemic in figures, pointed out on Twitter, the data also depends on feedback from the local to the national level. These transmissions tend to decrease at weekends and to increase sharply at the start of the week, as a catch-up.
Newsletter - Essential news
Every morning, the news seen by Le ParisienI'm registering
Your email address is collected by Le Parisien to allow you to receive our news and commercial offers. Find out more
Can we therefore trust these figures entirely, if they are also random? Yes, by changing focal length and comparing, for example, the figures on D-7. With this method, we can see that, despite some day-to-day increases, the general downward trend in deaths, hospitals and other cumulative establishments, remains constant. "Deaths have only been going down for weeks," Guillaume Rozier already noted on May 12. The Parisian dashboard visualizations confirm this trend.
And that's the general trend. A stick is obtained by calculating the average (centered) over 7 days.
Deaths have been going down for weeks. And today's small increase has very little influence over the general trend… pic.twitter.com/inlEHShr68
This is also what the General Directorate of Health does by assessing hospitalization trends: “18,468 people are hospitalized for a COVID-19 infection (vs 21,595 a week ago on May 12) and 506 new admissions were recorded in 24 hours (vs. 670 a week ago), "she said in her statement Tuesday. Again, the decline in daily readings continues, suggesting a weaker circulation of the virus.
To ensure this continuous decrease after deconfinement, it is better to be patient. "We must wait until the weekend" of May 25 to confirm this, said Geneviève Chêne, director general of Public Health France, in our columns, Tuesday.