I know of a couple who bought a clock radio connected to the emergency weather system. When a notice went out, the radio would turn on so you could hear it. They thought it would be useful, as they lived in in a tornado area, and would like a warning if they were asleep.
Problem is, it wasn't that localized. It would wake them for alerts that were 100 miles away, and which didn't affect them.
They stopped using the alarm.
In general, people won't use a service like what you describe unless it's really actionable. And I don't think this system has a high false positive rate (and high false negative rate) so there isn't good actionable information beyond what general precautions (social distancing, good hand hygiene, etc.) offer.
- What percentage of people get fevers from other reasons? That is, even without bad actors, people are going to report fevers, and those might overwhelm the actual positives.
- How close does your infection model (X minutes within distance Y) match the actual infection mechanism and rate? I suspect it isn't that good, and will include many people who aren't infected.
> What percentage of people get fevers from other reasons?
Interesting. Other infectious diseases (flu/colds) seem to be going down in countries with lockdowns so this may not be such a big issue.
> - How close does your infection model (X minutes within distance Y) match the actual infection mechanism and rate? I suspect it isn't that good, and will include many people who aren't infected.
I pitched this idea to a leading epidemiologist and he thought the mechanism would work.
https://www.cdc.gov/flu/weekly/index.htm says (of the current US flu season): "CDC estimates that so far this season there have been at least 36 million flu illnesses, 370,000 hospitalizations and 22,000 deaths from flu."
If 50% of the 36 million flu illnesses had a fever, and the flu season is 23 weeks long (the records start in 201940), and the flu lasts for one week, then that's over 750,000 people each week with a fever, on average.
That number far outweighs the current numbers for people with COVID-19 symptoms, so gives a large false-negative signal in the current situation.
As the number of cases increase, the signal-to-noise ratio changes. But at some point, specialized per-person notification is no better than general advice to the entire population.
You mention that the numbers for flu/colds "seem to be going down in countries with lockdowns."
FWIW, it's also going down in the US without lockdowns. That's because we are approaching the end of the flu season. See the CDC chart just under the text "In the most recent three weeks, influenza A viruses are the most commonly reported influenza viruses in all age groups." in my earlier link.
Thanks, interesting. It might be the case that the app would only be useful once the levels of infection increase, or perhaps the app would only instruct the user to send an alert when they had developed a fever + dry cough, which seems more typical of this virus.
> But at some point, specialized per-person notification is no better than general advice to the entire population.
I take your point, but I think targetted "you have had a good chance of being exposed" is far more likely to result in action than "minimise contact with other people if possible".
I looked at that paper but it gave no indication of how the app was supposed to implement algorithmic contact tracing. I think it assume that that method would always be better than the current system? If so, there's no presented evidence that that's true.
I also can't figure out why you and are are downvoted for our low-key discussion in a days-old posting, so I've up-voted you.
Problem is, it wasn't that localized. It would wake them for alerts that were 100 miles away, and which didn't affect them.
They stopped using the alarm.
In general, people won't use a service like what you describe unless it's really actionable. And I don't think this system has a high false positive rate (and high false negative rate) so there isn't good actionable information beyond what general precautions (social distancing, good hand hygiene, etc.) offer.
- What percentage of people get fevers from other reasons? That is, even without bad actors, people are going to report fevers, and those might overwhelm the actual positives.
- How close does your infection model (X minutes within distance Y) match the actual infection mechanism and rate? I suspect it isn't that good, and will include many people who aren't infected.