Comment of the Day: Phil Koop https://www.bradford-delong.com/2020/04/what-is-the-real-prevalence-of-coronavirus-across-states.html#comment-6a00e551f080038834025d9b4755b8200c: ' Adjusting for lag presumably requires both a model and longitudinal data to input to the model. So the starting point is with cross-sectional random samples. If we guess the true rate of infections is about 3%, then we need a test with very good specificity. An RNA test, even if had 100% specificity, could only tell us about current infections, not cumulative infections. A serologic test can tell us about cumulative infections as of 2-3 weeks ago, provided it has good enough specificity. There is a list of serologic tests here: https://www.centerforhealthsecurity.org/resources/COVID-19/serology/Serology-based-tests-for-COVID-19.html. Some of these are claimed to have 100% specificity, but of those, none are yet approved for use in the US. Of the tests approved for use in the US, the only one with a listed specificity is 95.6% (a suspiciously precise number, given it was tested in "a total of 128 COVID19 positive patients, and 250 COVID19 negative patients (as detected by RT-qPCR).") 95.6%, even if correct, would not be nearly good enough. I think that whatever is right is going to have to start with a good-enough test....
#commentoftheday #2020-04-17
