This is a phenomenal article on the potential of different sorts of testing, and the ways that we are limited by demands for sensitivity and specificity. There's one important thing it misses out, which I will come to later in this thread 1/n theatlantic.com/health/archive…
Different sorts of questions need different sorts of tests. The US has got hung up on tests and testing, to the extent that it can seem some people think that testing is on its own a sufficient pandemic response. It’s not. It’s just keeping score 2/n
The crucial thing is not the test itself it is what you do in response. For instance to ensure appropriate treatment of cases in a healthcare setting, you want a very sensitive and specific test – meaning you can trust the result because getting it wrong matters 3/n
But if you are screening out in the community you might want to pool tests. That would be useful in schools, because if the test came back +ve you could send kids home without needing to know exactly which student is infected (that can be cleared up later with PCR testing) 4/n
My colleague @michaelmina_lab has been arguing eloquently in the article and elsewhere for the benefit of a test that is highly specific but not so sensitive, and cheap and quick enough it could be taken daily nytimes.com/2020/07/03/opi… 5/n
(These tests exist – an eg mentioned in the article are those made by @E25Bio. The co-founder and CEO of which is Bobby Brooke Herrera who took my class what feels like yesterday. Hi Bobby!) 6/n
Such tests run into trouble with regulators because of the lack of sensitivity. A fraction of people who are genuinely infected will get a false negative. This is obviously not ideal, but let’s unpack it for the purposes of screening. Imagine a test with 50% sensitivity 7/n
Now imagine if the tests were really being done daily. While the test might come back (wrongly) negative on the first day, it would be unlikely to come back negative throughout infection. 8/n
Also, even if you only detected 50% of cases and got them isolated in time to limit onward transmission, that would be a *lot* better than most current contact tracing efforts nytimes.com/2020/07/31/hea… 9/n
And as Mike argues there is a good reason to think this sort of test would be more sensitive the more contagious a person is. Full disclosure, Mike is a colleague. That’s not influencing my opinion – I can happily disagree with him🙂 but I think this is reasonable 10/n
The thing the article misses is another advantage of such tests, as we (me + @baym + @LeeKShaffer) show in this preprint (🚨 preprint!) they actually take advantage of the mathematics of viral transmission and over dispersion in transmission.
dash.harvard.edu/handle/1/37363… 11/n
dash.harvard.edu/handle/1/37363… 11/n
SARS-CoV-2 transmits in clusters (technical term overdispersion): a minority of infections are responsible for the majority of transmission and most infections don’t transmit. And it’s transmission we care about. So why does that work well with these tests? 12/n
Because where there’s 1 transmission event, there’s likely more. Imagine using these tests w contact tracing. While 1 test might be a false -ve it’s unlikely all of them will be, if a contact comes back +ve you know to quarantine and investigate all who share the exposure 13/n
It’s may also help ‘cluster busting’ backwards contact tracing strategies. Any case had to transmit from someone, who most likely transmitted to at least one other person. Tracking them quickly is key. Again the speed of the test matters medrxiv.org/content/10.110… 14/n
But if regulatory authorities demand very high sensitivity and specificity, regardless of whether the test is for screening or diagnostics, that will mean we can’t do this. And it won’t be because it’s not possible 15/n
As the article points out, we should’ve taken the virus seriously in the first place. Sure testing costs money. How much do you think all the testing we didn't do in the spring is costing us now? washingtonpost.com/outlook/corona… 16/n
And everyday of the pandemic in the US the economy hemorrhages more money. Money that could be spent on lots of cheap imperfect tests, because in this case the perfect may be the enemy of the good 17/end
