The CDC tests a few tens of thousand blood samples each month, looking for the antibodies that show that a person had some prior infection with COVID-19. This set of blood samples is a sample of convenience — the blood in question was typically drawn for routine lab testing, and may or may not come close to presenting the actual cross-section of the U.S. population. But it provides our best estimate of cumulative total of true number of total COVID-19 infections that have occurred in the U.S.
The method isn’t perfect. Persons getting blood drawn for routine testing are not necessarily a random cross-section of the U.S. population. By design, it will not separately count re-infections, even though those are counted as additional infections in the U.S. test count data. (You either have antibodies or not, per these tests.) And post-infection antibody levels may decline rapidly enough over time we may begin to lose count of some of those who were infected early in the pandemic.
That said, this is the best we’ve got for estimating the actual count of infections. As opposed to the count of officially reported infections.
Prior to this most recent month, the seroprevalence data have been pretty boring. Aside from a chance in methods that means you can’t look back prior to September 2021, the results consistently showed more-or-less one un-reported infection for every reported infection. That is, the gray “ratio” line below hovered around a value of 1.0, plus or minus.
The most recent round of survey data — for the month ending at the end of February 2022, is noticeably different.
Source: Calculated from CDC seroprevalence survey data, COVID data tracker accessed 5/2/2022.
With this most recent round, that ratio popped up to about 1.4 to 1. Taken at face value, there was a large jump in the number of actual infections, relative to the cases reported in the official U.S. COVID-19 statistics. Or, putting that another way, a large jump in unreported infections.
On the one hand, it’s tough to make too much out of one month of data. The CDC publishes standard errors with these, but they seem to purely “statistical” standard errors (i.e., they only account for the size of the sample). They don’t seem to try to account for possible “structural” errors (e.g., that the representativeness of the sample may vary from month to month).
On the other hand, this is cumulative. When you see a change, it means that the change in the most current month was so large that you can see it, even through you are totaling up cases all the way back to the start of the pandemic.
And, ultimately, that’s what makes the most recent data (from blood drawn in February 2022) interesting. There seemed to be a lot more un-reported cases. The ratio of un-reported to reported total COVID-19 cases rose to 1.4, from previous highs around 1.2.
This is something that I’ve been on the lookout for, in the context of the growth of over-the-counter (home) testing. E.g., Post #1431. The one real-world example, where one New York county asked residents to self-report positives from home tests, found that positives on home tests accounted for about 20% of total positives (including official tests and self-reported over-the-counter tests).
The growth of re-infections under Omicron adds a small note of caution, however. Last time I looked, re-infections accounted for 8 percent of all infections under Omicron, up from just one or two percent from prior variants. That said, back-of-the-envelope, those reinfections (not captured separate in the seroprevalence data) wouldn’t be nearly large enough to account for the jump in the ratio of unreported to reported tests.
In any case, this February 2022 seroprevalence survey is the first hard data suggesting that there has been recent strong growth in un-reported COVID-19 cases in the U.S. That really wouldn’t be a surprise, as (e.g.) the U.K. seems to have experienced a pretty significant divergence between the official count of new positive tests, and the actual new infection rate (Post #1478).
If this truly reflects an upsurge in unreported cases, this divergence should widen over the next few months. I guess I’ll check in a month from now and see where this stands.
To me, the interesting question really relates to average severity of illness. We’re still below 2000 new hospitalizations a day in the U.S. Is that because there really are that few new cases? Or are there lots of new cases, but some factors (e.g., vaccination rate, prior infection rate, or changes in the virus itself) are keeping the hospitalization rate down.
