digitalepidemiologylab / swisscovid_efficacy

4 stars 0 forks source link

Easy to misinterpret the situation of the 26 reporting SwissCovid as cause for the test #1

Open pdehaye opened 4 years ago

pdehaye commented 4 years ago

The preprint says:

Screenshot 2020-09-13 at 12 21 12

I understand the form to be this one:

Screenshot 2020-09-13 at 12 21 49

To best understand that number of 26, it might be necessary to explicit the rules for getting a test.

According to the official strategy (at least in the cantons I am aware of, which are the big ones), someone who has been in close contact with a positive case and found by the health authorities/contact tracers will be put in quarantine. For as long as they don't exhibit symptoms, they will not be tested, with two exceptions:

I understand the first case to be relatively infrequent. The second case could occur for instance if a husband and wife both have the app installed. The husband gets infected, put in isolation, the wife is easily found by contact tracers, and gets put in quarantine right away. The wife is told she won't be tested until she has symptoms, unless she gets a SwissCovid notification. A few hours later she does (husband's TEKs). She goes to her doctor. There is no box "close contact with infected", and she has received a notification, so the doctor ticks that box.

This situation is probably much more frequent than one might think: clustering must occur due to natural population distribution, but also due to the proximity of the virus ("my friends are getting the virus, I better install the app!").

It looks thus plausible that the gist of the preprint is wrong (or the discourse around it). These 26 were not necessarily unknown to the contact tracers, on the contrary. It is quite plausible they were known and already in quarantine, and that a notification actually complicates the management of their situation as it increases confusion. This is particularly relevant as the 26 is imputed to 37, then compared to manual contact tracing. It would be odd if a big part of the credit claimed by SwissCovid was actually coming from cases already identified by manual contact tracing, especially given that this is meant as an assessment of the complementarity of the app to classic manual contact tracing.

Note that this is far from nitpicking. One of the authors of this paper (EB) participated two days ago in a public seminar. In his presentation and follow up answers, he gave what I would qualify as a misleading impression.

At one point he states:

26 RT-PCR+ cases found that would not have been tested without the app!

Screenshot 2020-09-13 at 14 50 55

At another point, another participant (VC) has an exchange with EB:

VC: Is 26 notifications enough to justify this deployment? EB: It's not 26 notifications, it's 26 people who know they are sick even though they are not symptomatic because they have been tested, and the reason they have been tested is because they have a notification from the app. That is the definition of the efficacy of proximity tracing. It's to break the infection chains [..]. VC: Were those people notified otherwise? Did they know that they were already in contact with other cases? EB: That is not what the form suggests. That is not what the data suggests. This is Swiss clinicians' reports, aggregated at the Federal Office of Public Health. [unintelligible]

The form does not suggest that they were not notified otherwise. The form does not ask this, and doesn't leave the opportunity to express this. The form is asking for the reason for getting the test, and being put in quarantine is not enough of a reason according to current policy.

Given the confusion coming from at least one of the authors, the text should be clarified.

vaudenay commented 4 years ago

This number 26 was announced (probably in the same misleading way) to be 13 by one coauthor one week before this peprint. https://www.srf.ch/play/tv/tagesschau-spezial/video/medienkonferenz-des-bundesamts-fuer-gesundheit-bag?urn=urn:srf:video:6af1ebc6-485c-4b4c-96eb-26626a09e4be

How come that this 13 was doubled between August 28 and September 4?

pdehaye commented 4 years ago

In RTS TV show Mise au Point, the same author mentioned by @vaudenay made again the same unjustified logical leap (episode of 2020.09.13, towards end of first part).

pdehaye commented 4 years ago

At least one author of the preprint has acknowledged publicly that there is no evidence those notified by the app were not already aware they were at risk.