NOTE: This study has been widely criticized for drawing conclusions that the data doesn’t support. In particular, the false positive rate of the antibody test is too high to guarantee a nonzero result, let alone the very high results the authors report. You can find critiques all over the internet, but Andrew Gelman has a very detailed write-up here.
This doesn’t mean the conclusion of this study is wrong. Their point estimate is still what it is. However, we clearly need more antibody tests with larger sample sizes and more geographic diversity. I imagine we’ll get both before long.
One of the key metrics for the coronavirus pandemic is the case fatality rate. That is, out of all cases of people infected with COVID-19, how many die? This is difficult to measure because we know that our measure of the number of infections is wildly unreliable. However, antibody tests are starting to become available that allow us to test an entire population and get a solid estimate of how many people have ever been infected. A team of researchers did this for Santa Clara County in California recently and came up with this:
Our study included 3,439 individuals that registered for the study and arrived at testing sites. The total number of positive cases by either IgG or IgM¹ in our unadjusted sample was 50, a crude prevalence rate of 1.50%. After weighting our sample to match Santa Clara County by zip, race, and sex… the estimated prevalence was 2.49% under the S1 scenario, 4.16% under the S2 scenario, and 2.75% under the S3 scenario.
….After adjusting for population and test performance characteristics, we estimate that the seroprevalence of antibodies to SARS-CoV-2 in Santa Clara County is between 2.49% and 4.16%.
The “official” number of confirmed coronavirus cases in the US is about 600,000. If we take a middle estimate of 3.3 percent from this study and extrapolate it to the entire country, the true number of cases is about 11 million.
But what’s the death toll? The official number right now is about 30,000, but let’s be conservative and assume the real number is twice as high. This gives us 60,000 deaths out of 11 million cases, for a true CFR of 0.5 percent. This is much lower than previous estimates from around the world.
A few days ago I suggested that the initial models of the death toll from COVID-19 may have been too high “for reasons we don’t yet understand,” but this might be the reason: the assumed CFR in these models was not just 2x or 3x too high, it was 10x or 20x too high. Using the correct CFR, the death toll projections should always have been much lower, and that in turn means the effects of social distancing measures have been overestimated. Maybe.
As usual, this is armchair epidemiology, not to be taken too seriously. It’s also based on a single study. Further studies may show different results. Still, it’s pretty suggestive.
¹Hey! I recognize that. My particular version of multiple myeloma is IgG light kappa, which means it’s the IgG immunoglobulins that are cancerous.