Notes on prevalence

One of the most remarkable aspects of this pandemic is that, exactly three months after the first case was detected in the USA, we still have almost no idea what the actual prevalence of the virus is in the general population.

This is likely to change very soon: several studies are underway. Abbott has apparently developed a serology test that it says has 100% sensitivity and 99.6% specificity (Sensitivity means the test detects the presence of antibodies triggered by the COVID-19 virus; specificity means it successfully avoids mistaking that virus for similar coronaviruses). This test was used in to test 200 people in Chelsea, Massachusetts, a working class high-density suburb of Boston that has had a severe outbreak.

None of these people had been tested for COVID-19. 32% appear to have contracted the virus. This suggests the ratio of infections to positive tests in the area is 17 to 1, (1.9% of the population had tested positive) while the current case fatality ratio would be around 0.3%, although that would probably go up a tenth of a percentage or two from currently unresolved cases that prove fatal.

Another striking result comes from Santa Clara County, where a test of volunteers who responded to a Facebook ad yielded an estimate that between 2.49% and 4.16% of the county’s nearly two million residents have been infected. What’s striking about this is that it represents a ratio of between 48 and 81 actual cases in the county’s population for every official case. This represents a CFR of around 0.1% to 0.2%, depending again on assumptions about unresolved cases. (There are some obvious methodological issues with this study, as it wasn’t a randomized group, and people who are more concerned about possibly having had the virus would no doubt be more likely to volunteer. Also the sample had to be weighted quite a bit to get population-representative numbers).

Some other prevalence numbers:

A German town found a 14% prevalence rate, and a 0.37% CFR.

Apparently 49% of the population of this Italian resort town is estimated to have been infected.

Obviously these numbers are all over the map, although all of them suggest that the true ratio of actual cases to positive tests is extremely high. It would be no surprise to discover in the next few days that twenty million or more Americans have already been infected. (The official count stands at 778,000, but testing volume remains pitifully inadequate.)

What are the implications of a true prevalence rate in the tens of millions, if this turns out to be the case? (Again, our knowledge regarding this critical point is likely to grow enormously in the next week or two).

. . . LA County just released a study indicating 4.1% of the county’s population has had the virus. This is 55 times higher than the number of people who have tested positive. It produces an initial imputed CFR of .09% to .15% out of 221K to 441K cases (again this would almost certainly rise with more complete data).