Identifying potentially asymptomatic people

[chegli]

Hi,

I was curious as to your opinion on the idea presented below. In case this model for contact tracing was previously considered but not adopted I would be very interested to know the reasons.

---

Current model 1: Symptomatic contact tracing

This is the current model implemented by most contact tracing apps IIUC: A person has tested positive for Covid-19, informs the app about this fact, all people verify if they were in proximity/contact to that person in the last eg 48-72 hours and alert the user if that's the case.

• In this model, based on a known infected person, potentially newly infected people are warned of having gotten the infection (but probably not being infectious yet as it's only eg 48 hours since).
• Even if being near a person isn't a 100% guarantee for getting infected, this has a somewhat high likelihood given the infectiousness of Covid-19. Thus this is a highly relevant information for those participating via the app.

New (?) model 2: Identifying potentially asymptomatic people

Now coming to the new model I'd like to suggest: this model assumes an asymptomatic person that is spreading the virus unknowingly. This represents the opposite contact tracing to Model 1.

• Given the person is asymptomatic the assumption is the person is not being tested (otherwise this becomes the model 1 above). Reasons for not being tested would be (a) because the person doesn't notice anything being asymptomatic and/or (b) limited testing capabilities/not testing random people.

• Now let's assume this asymptomatic person has infected other people in the last 1 or 2 weeks. Those other people might have meanwhile become symptomatic and got tested. Thus those other people, if they also have this app, would inform the app about their positive test. But this would happen with the incubation delay obviously, so we're talking few days maybe 2 weeks.

• And let's assume it is not only 1 person that got infected (by this same asymptomatic spreader) but say 2 or 3, they all become symptomatic, all get tested positive, all inform the app.

Under the above assumptions the algorithm below would be able to detect a potentially asymptomatic spreader with a certain, say medium likelihood: obviously only a test can ultimately verify if a person is positive or not, but with this the app could inform the person that they should consider getting tested as they might be spreading asymptomatic/unknowingly.

Waiting for more than 1 such potential spreading events, perhaps waiting for 2 or 3 instead, should increase the likelihood and remove false positive alerts.

Algorithm

• A person testing positive to Covid-19 enters this information in the app with an additional flag : whether or not they know from where they are likely to have been infected. This algorithm is only interested in cases where the origin is unknown (otherwise this person/infection is skipped).
• The infection case (with flag "unknown origin") is uploaded (to the same location probably as in Model 1), everybody else downloads this (also same as in Model 1).
• Now unlike in Model 1 where the app goes back eg 48 hours, in this case the app goes back eg 4-14 days to see if the local mobile phone had proximity/a contact with the meanwhile "positive tested person with unknown origin".
• If the above is the case, ie the person is in the local contact list 4-14 days back, the app takes note of this as one case of potential asymptomatic spreading from the local person a (4-14 days) later positively tested person.
• If the above repeats for 2 or 3 more cases (within a certain time period where asymptomatic spreading is biologically possible that is), the app becomes confident enough that the holder of the local mobile phone might actually indeed be this "unknown origin" for others and shows an alert saying, as mentioned, something like: "you might be an asymptomatic carrier of Covid-19, please consider getting tested asap".
• Already warning after 1 such case might create too many false positive alerts, waiting for too many such cases (say 4 or more) however might suppress (ie result in false negatives). Perhaps 2 or 3 is a good number therefore.

---

Looking forward to hearing your opinion on this approach.

[kollokollo]

I like the idea. This would add a completely new feature to the tracking app.

[ftyl]

There is one flaw to this idea: the current plementation of DP-3T takes a much longer period than last X hours into account. All contacts of previous days are contacted.

Therefor:

1. person A is asymptomatic but infections
2. person B becomes infected by person A
3. person B 3 days later becomes symptomatic, goes test and marks positiv via the app
4. person B's app would nowsend a (SKt, t) packet to the backend
5. the backend distributes this to anyone that asks, and person A will receive this (SKt, t) and with all likeliness is going to be warned about the contact 3 days before as if B was infected and A might have been infected by B.

Therefor, A is now warned, will get tested (optimally), positive result, marks it and person A's app uploads (SKt, t) packet and it gets spread around, causing all contacts to be informed.

Therefor the asymptomatic spreader is found sooner or later with relatively high likeliness if that spreader has infected others which developed symptoms.

It might be interesting to warn a user in case their app gets a high amount of positive contact warnings in a short period of time, however it is not clear that this means the user is an asymptomatic spreader or simply is unlucky. Either way though, the user will be informed of a potentially positive contact so not sure if anything really needs to (or can) be added to "achieve" what you envision.

[chegli]

Thx for the feedback! Indeed my assumption for some reason was that the app would only warn people that were in contact 48 hours before the onset of the symptoms (which I thought was plausible as it seems you can start spreading the virus 48-72h before the symptoms). If the app already goes back considerably more then my approach would not work indeed.

How much does it go back atm?

[chegli]

@ftyl according to the whitepaper

The health authority is responsible for informing patients of (positive) test results, authorizing uploads from phones to the backend, and determining the contagious window, i.e., during what time the patient was contagious and might have infected others. Epidemiologists estimate that the contagious window starts 1 to 3 days before the onset of symptoms. The start of the contagious window determines for which time frame phones upload information.

From which I'd derive that contacts are only analysed within this contagious window.

What I'm suggesting is looking 6-14 days before the onset of infection, which is before this contagious window.

[peterboncz]

Hi, an earlier issue promoting this idea is https://github.com/DP-3T/documents/issues/242