The results from the first blood surveys that test for evidence of antibodies to the new coronavirus have begun rolling in. They’ve been confirming earlier hints that in hard-hit places a significant share of people — 21.2% in the New York City survey — may have been infected with the virus, and that in most other places the percentages are still in the low single digits. They’re also making it ever clearer that the disease caused by the virus, Covid-19, is much, much deadlier than influenza.
“It’s just the flu, bro” has been a sort of know-nothing rallying cry for those skeptical of government lockdowns and social-distancing efforts this spring. But it has also been the base case of a few prominent experts, such as John Ioannidis of the Stanford University School of Medicine and former Swedish state epidemiologist Johan Giesecke.
One important thing to get straight here is exactly how dangerous seasonal influenza is. The fatality rate that has been cited most often (including by me) is 0.1%, which happens to be about what you get if you divide the Centers for Disease Control and Prevention’s estimates of U.S. influenza deaths over the past nine years by its estimates of symptomatic cases. But just as with the coronavirus, testing has shown that many people infected with influenza viruses develop no flu symptoms. In a Twitter thread from February that a reader pointed out to me this week, University of Oxford infectious disease epidemiologist Christophe Fraser estimated that the actual infection fatality rate (which I will refer to from now on as IFR) of seasonal influenza is 0.04%.
Fraser also speculated that influenza expert Lone Simonsen “may have more accurate numbers.” So I emailed Simonsen, a professor of population health sciences at Roskilde University in Denmark who has worked at the CDC and National Institutes of Health in the U.S., to ask. Her answer: Fraser’s estimate is spot on. Simonsen believes that the IFR for the coronavirus will eventually turn out to be on the low end of current estimates, possibly as low as 0.2% or 0.3%, but emphasized that this is “still far greater than … for seasonal influenza.”
What are the new coronavirus serology surveys saying about the infection fatality rate? Before I run through the numbers, some caveats: Because of the difficulties in getting a truly representative sample of the population and possible flaws with the tests, none of these estimates of the virus’s prevalence is necessarily right. None has been peer reviewed, either, and only one has even been presented in the form of a formal scientific paper. One bit of guidance is that surveys that find a large percentage of infected people are likely to be more reliable in calculating IFRs because tests that deliver false positives skew their results less — if 20% of the population is infected and 1% of the tests are false positives, that just changes the result from 20% to 21%; if only 1% is infected, the false positives double the apparent rate from 1% to 2%. Also, the IFRs calculated here are snapshots that miss out on deaths from the disease that haven’t happened yet or haven’t been reported; the influenza death totals used in calculating its fatality rate are more-complete after-the-fact estimates. Where I do the calculations myself, I use the death totals available as of Friday morning, even though the surveys may have happened earlier, so that counteracts this effect at least a little.
Here goes, in descending order of what percentage of the population is estimated to be infected.
- In New York, preliminary results of an ongoing state survey of people approached outside of grocery and big-box stores that were announced Thursday by Governor Andrew Cuomo found that 13.9% of those tested statewide had coronavirus antibodies, and that 21.2% of those tested in New York City did. The official estimate of Covid-19 deaths statewide is 15,740, which makes for an infection fatality rate of 0.58%. New York City adds in deaths from people who died at home and had Covid-like symptoms, for a city total of 15,411, which gives an infection fatality rate of 0.86%. Virus expert Trevor Bedford of the Fred Hutchinson Cancer Research Center in Seattle prefers using an estimate of the city’s excess deaths over normal (19,200 since March 11), which gives an IFR of 1.08%.
- In the German town of Gangelt, preliminary results of an ongoing University Hospital Bonn survey of a “representative set of households” found that 14% of the people tested had the antibodies for the coronavirus and an additional 1% were still infected with it, which according to the study’s authors came out to an IFR of 0.37%.
- In the canton of Geneva, Switzerland, preliminary results released Wednesday from a University Hospitals of Geneva survey of people randomly selected from recent participants in the canton’s annual health survey found an infection rate of 5.5%, which at a canton death total of 215 makes for an IFR of 0.8%.
- In Los Angeles County, preliminary results released Monday of a survey of residents “recruited via a proprietary database that is representative of the county population” that was conducted by the county health department and researchers from the University of Southern California, Stanford University and elsewhere estimated that from 2.8% to 5.6% of county residents were infected, which with the current county death total of 797 would imply an IFR range of 0.36% down to 0.18%.
- In the Netherlands, tests on donations to the blood-bank organization Sanquin in the first half of April (the testing continues) found that about 3% had coronavirus antibodies, which with the Dutch death total of 4,177 works out to an IFR of 0.81%.
- In Santa Clara County, California, a survey conducted by Stanford University researchers (one of whom is the aforementioned Ioannidis and two of whom also worked on the Los Angeles County survey) and others in conjunction with the county health department tested residents recruited via Facebook ads and found 1.5% to have coronavirus antibodies. The study’s authors then made adjustments based on county demographics and the characteristics of the tests they used that drove their estimate of the county’s infection rate up to a range of 2.49% to 4.16%. By their calculations, this implied an IFR of 0.12% to 0.2%. This is the only antibody study so far that’s been presented in the form of a scientific paper, enabling more scrutiny of methods than with the others, and the criticisms from epidemiologists and statisticians of the statistical adjustments have been so withering that I’ll also calculate the IFR based on the raw 1.5% positive result and the Santa Clara County death total of 95, which comes to 0.33%.
- In San Miguel County, Colorado (home of the Telluride Ski Resort and part-time home of the biotech entrepreneurs who funded the study), antibody tests were offered to all residents, and more than half have taken them. So far 0.5% of the tests have come back positive, and 1.5% “borderline.” No Covid-19 deaths have been reported in the county, so the IFR is zero. But if just one resident were to die of the disease, it would jump to about 1% if you count both positive and borderline test results as infections.
Those are all the survey results from antibody testing that I knew of as of Friday morning. There’s an antibody testing program in the Boise, Idaho, area, that so far shows a 1.75% positive rate, but it seems to be intended not as a survey but to let people who think they’ve been infected know if they have been. Stockholm’s Karolinska Institute medical university had released results Tuesday showing that 11% of Stockholm area blood donors had antibodies for the coronavirus, but retracted them Wednesday after discovering that an undetermined percentage of the blood samples were from people who knew they had Covid-19 and were donating to produce plasma for treatments for the disease.
So the range of IFRs derived from these surveys so far is 0.12% to 1.08%, and the latter result should probably be given much more credence than the former both because of the false-positives issue described above and the seeming flaws in the calculations used to arrive at 0.12%. The most exhaustive and up-to-date pre-serology-survey estimate of Covid-19’s IFR that I’m aware of, from a peer-reviewed article in Lancet Infectious Diseases by a group of researchers at Imperial College London, is 0.66%. If the IFR of the seasonal flu is 0.04%, these blood surveys show Covid-19 to be anywhere from three times deadlier to 27 times deadlier — and given the incompleteness of current death counts, the true range seems likely to be higher than that.
There are lots of factors that can affect these fatality rates, including demographics, underlying health issues in the population, the quality of the health-care system and how overwhelmed that health-care system is with coronavirus patients. Covid-19 severity is of course hugely dependent on age, with the estimated IFRs from the Lancet article mentioned above rising above 0.5% only after age 50 and topping 13% for those over 80. Severity that rises with age is customary with seasonal influenza as well, though, and risks from the coronavirus seem comparable to those from seasonal flu only for children. For every age group above 20 the fatality risk from Covid-19 — going by the estimates from the Lancet article and the CDC’s seasonal influenza estimates — is much, much higher. It’s not just the flu, bro.