For the uninitiated, the FDA faced criticism initially by being too slow to authorize PCR tests for SARS-CoV-2 and issued a much more broad EUA for serological tests. However, serological tests need to be run on known positive and negative patients (already tested) to validate their accuracy. Because infection with SARS-CoV-2 is relatively uncommon, and because the EUA is broad, tests that offer the statistical value of a coin flip have been authorized and used. This is a problem if you’re trying to make health decisions.
So, the FDA is trying to advise people on this risk by providing tools to judge the performance of serological tests to help clear up the confusion. This is because many tests have rated performance based on a very small number of people and the error bars are relatively large.
Even the worst tests here provide a lot more information than a coin flip, as I understand it. A 50% PPV at a prevalance of 5% means that if the odds of someone having had coronavirus were 5% going into the test and they get a positive result, you can immediately update your estimate of the odds of them having had it to 50%, which isn't nothing.
It looks like this technology is getting debugged. Some of the tests are well above 99% for both specificity and sensitivity. Some aren't, and they will probably be withdrawn from the market.
So from what I heard the other day, those >99% Numbers were from a subset of the actual sample, which they argued was because of the requisite time lag for seroconversion but may just be p-hacking.
So, the FDA is trying to advise people on this risk by providing tools to judge the performance of serological tests to help clear up the confusion. This is because many tests have rated performance based on a very small number of people and the error bars are relatively large.