Twitter | Search | |
This is the legacy version of twitter.com. We will be shutting it down on 15 December 2020. Please switch to a supported browser or device. You can see a list of supported browsers in our Help Center.
Dan Larremore
Asst. Prof, University of Colorado Boulder CS and . Math, infectious diseases, networks, and computational social science.
3,409
Tweets
498
Following
4,371
Followers
Tweets
Dan Larremore 2h
This is certainly one potential problem, yes. However, if the alternative is blanket lockdowns or curfews (quarantine everyone, regardless of whether they carry the virus), then having *some* false positives who quarantine & wait for a confirmatory PCR is a better option, IMO.
Reply Retweet Like
Dan Larremore 3h
Replying to @svscarpino
If Maine has 100 freezers, and they buy -80 freezers, how many freezers do they have?
Reply Retweet Like
Dan Larremore 3h
Replying to @dsracoon
Yes, exactly. Screening = olfaction test, with a followup PCR if the olfaction test is failed. And the dashed yellow line is just weekly PCR with a 1 day turnaround.
Reply Retweet Like
Dan Larremore 3h
Oh oh, to clarify: In my thought experiment, I meant a test that tells negatives that they are negative (high specificity), but when it comes to finding the positives, it only gets 50%. I wasn't totally clear! Sorry.
Reply Retweet Like
Dan Larremore 14h
I. Love. This. "100 enthusiastic amateur fishermen will catch more bass in a day than the world's top bass pro."
Reply Retweet Like
Dan Larremore 18h
Replying to @Deadly_Statins
That contrasts with the literature, but actually for an interesting reason: The symptom is only moderate prevalence in self-reporting surveys, but increases in prevalence when you used a standardized test. (Various references in the paper if interested.)
Reply Retweet Like
Dan Larremore 19h
Replying to @ml_barnett
These are really good points. The engineering is critical. Randomized order, various smells, possible non-smells among the correct answers. And, as you said, cultural matching: "orange" may not be a touchstone smell for a lot of people.
Reply Retweet Like
Dan Larremore 19h
Replying to @janetrosenbaum
The test that I tried involved a card with scents on it (scratch/sniff) and a matched multiple choice quiz. I had to identify the smells in order to pass. This would probably not work with kids who are too young to ID lots of smells, but could work with older kids, I think.
Reply Retweet Like
Dan Larremore 19h
Great question. You're right: not everyone would be able to be able to participate in this kind of screening. Others include those with anosmia due to concussion / head trauma, those with rhinorrhea (stuffy/runny nose), and older adults (links to alzheimer's / Parkinsons, IIRC).
Reply Retweet Like
Dan Larremore 19h
Replying to @essexie @DrCSyn
I think that's a good way of putting it. So the idea with repeated screening tests is: what if we think about the efficacy of the testing regimen, not the accuracy of a single test? The key with the repeated testing is about the impact that the testing has on community spread.
Reply Retweet Like
Dan Larremore 20h
Replying to @TJargstorff
Agreed, 100%. If the specificity of the imagined test is really awful, this thought experiment would be a disaster! 🤯😭
Reply Retweet Like
Dan Larremore 20h
Replying to @DrCSyn
Not a perfect analogy, but the idea is the same: We want to remove a pathogen where (1) each dose/wave is imperfect, and (2) the pathogen spreads between waves. In other words, we already sort of "get" how a regimen works, through imperfect action, repeated to great effect.
Reply Retweet Like
Dan Larremore 20h
Replying to @DrCSyn
Ah—different scope: At the level of the infection in your body, each dose of antibiotic is imperfect. It kills only a fraction of the bacteria. But repeated doses of antibiotic do the trick, and those that persisted through dose 1 multiply, but not enough to overcome dose 2, etc.
Reply Retweet Like
Dan Larremore 20h
Clear & crisp analysis of Slovakia's mass screening work here, with Martin Pavelka, , & team. Combined empirical data with models to estimate effects. "In the two weeks between the pilot and round 2 infection prevalence decreased by 82%."
Reply Retweet Like
Dan Larremore 20h
Replying to @KatyMcconkey
Yes! And, controlled studies show that if you further give people a test that asks them to identify unlabeled smells, in a standardized way, the prevalence is higher still. It's a very interesting symptom, for screening potential.
Reply Retweet Like
Dan Larremore 22h
With this reasoning, we would halt the use of any medicine that didn't work on the first dose. Antibiotics, chemotherapies, radiation—halted, until a single dose works perfectly. No. These all work as regimens, as designed. Each layer is imperfect, but repeated, they work.
Reply Retweet Like
Dan Larremore 22h
Replying to @DanLarremore
The core concept behind repeated screening tests is that EVEN IF a test's sensitivity isn't perfect, its repeated use will drive the epidemic down. Think about it: why do we complete a course of antibiotics? Because we clear that infection with repeated doses—not just one. 4/4
Reply Retweet Like
Dan Larremore 22h
Replying to @DanLarremore
The idea of using repeated rapid antigen tests is like this, except that that 50 catch / 50 miss number is even better. Depending on the test, and where you draw the line, it could be 70/30, 90/10, or better. So if we 90/10 every week for a few weeks, this has big impacts. 3/
Reply Retweet Like
Dan Larremore 22h
Replying to @DanLarremore
If we repeat this process, over the course of a few months, we will have missed a lot of infections! BUT, we will ALSO have identified enough infections, and broken enough transmission chains, to reduce viral spread. The key? The 50% filter gets used over and over. 2/
Reply Retweet Like
Dan Larremore 22h
A thought experiment. Suppose we had a COVID test that had a 50/50 chance of identifying someone who was infected. And now imagine that everyone in a city uses this test. We identify 50% of the cases but miss the other 50%. We repeat this a week later. Find 50%, miss 50%. 1/
Reply Retweet Like