How many cases of COVID-19 are there?

There is currently a wave of panic around a potential 10-times the number of COVID-19 cases in the general population. I suggest this is a consequence of a poor study methodology.

The study tests for the presence of COVID-19 antibodies in blood taken from blood donors and others who mostly have either not been tested or have tested negative. It found that there are a very large number of people who show some levels of antibodies – roughly 10 times the number of people recorded as having contracted COVID-19.

The problem is that the antibody test isn’t perfect. At best it is perhaps 95% accurate. That means that for every thousand people tested it could falsely report that 50 of them have antibodies (a false positive) or If could miss 50 of them who do (a false negative).

Estimates from other sources suggest that as many as 35% of COVID-19 cases are unreported, so currently perhaps 1% of the American population have had COVID-19. However most of them have not given blood, so the number of potential test subjects who where never tested but who had or have COVID-19 is about 3 in a thousand.

The study found antibodies in about 10 times that number. However 10 times that number is squarely in the expected range of false positives. In other words, the study likely only measured the false positive rate of the test, not the prevalence of COVID-19 in the general population.

This is distinctly different from the test for COVID-19. While the COVID-19 test has similar error rates, a positive test for COVID-19 is followed up by a second, different and more accurate, test. Only if you test positive on both tests are you diagnosed as having COVID-19. Also, the COVID-19 is mainly given to people who have either symptoms or reason to believe they may have been exposed. That alone makes it inherently more accurate.

Conversely, testing negative doesn’t mean you don’t have COVID-19. That’s why it’s important to treat everyone as if they were a carrier. Testing isn’t perfect. You always need to be aware of the error rates.