Let me add a few more nuggets of information on this emerging topic.
There are a couple of important caveats and cautions about the study I flagged yesterday from Santa Clara County. One is an important possible sample bias. The way the study was conducted was people were invited to participate via Facebook. But the people responded was on a first come first serve basis. In the nature of things, people who were maybe sick in February and really wanted to know if it was COVID19 would probably be more likely to participate. So that would be a sample bias pushing the study toward a higher estimate.
Others note a more general issue, which is that these tests are not that accurate. Here I’m not talking about the particular test used in the Santa Clara County study. I’m talking about all of them at this point in their development. There’s a whole complex discussion of this in various articles, which touch on both diagnostic and statistical issues, most of it beyond my comprehension. But I think I can summarize the gist like this. Let’s say you have a test which is 95% accurate for negative results. So if you test a hundred people who have never had COVID19, 95 will come back correctly as negative but five will be falsely positive. That’s good enough to generate some valuable information if your testing a population in which the disease is prevalent. If it comes back with 65% positive, it may be a little lower or a little higher. But you know a lot of people have had it. But if you’re sampling a population where only tiny percentages are positive then quite possibly the great majority of your positives are going to be false positives.
Don’t take the precise percentages I’m using here as the actual ones. I’m trying to illustrate the concept. But given that the tests are new and in some cases have relatively high error rates this is an issue if you’re finding that 2% or 3% have positives.
Finally there’s an entirely different, very small study from Massachusetts. Researchers from Mass General went out on to the street in Chelsea, Massachusetts (a particularly hard hit town just across the river from Boston proper) and got volunteers for pinprick serology tests. They tested 200 people and 30% were positive. That is an astonishingly high percentage.
These are all small studies, under conditions with various kinds of possible sample bias and the tests themselves are new and may have high error rates. So I would not put too much stock into any individual one. It seems certain we’ll get a lot more of these data points in the coming days and over time the picture will come into focus. But 30% is really a stunning number.
