On Monday, researchers in California released preliminary results from one of the first US surveys to test a random group of people for antibodies to SARS-CoV-2. Serologic surveys are a tool epidemiologists can use to deduce how many people have really contracted the virus, including people who never felt sick or who did but were never able to get a diagnostic test.
The survey included 863 adults in Los Angeles County, selected from a market research database to reflect the overall demographics of the county. The results of their blood tests, conducted on April 10 and 11, show that between 2.8% and 5.6% of the participants had antibodies to the virus. Accounting for a margin of error, that suggests 221,000 to 442,000 people in the county may have been infected—28 to 55 times higher than the county’s official count of confirmed cases.
The results, from researchers from Stanford University, the University of Southern California, and LA County’s Department of Public Health, are in line with another survey released on Friday by the same academic researchers, of 3,330 adults in Santa Clara county. That survey found an infection rate of 2.49% to 4.16%, equal to a caseload 50 to 85 times larger than the county’s official count.
These surveys, neither of which have been peer-reviewed, aren’t perfect. The Santa Clara survey was criticized by some epidemiologists because it recruited its participants with Facebook ads, meaning that those who responded were more likely to be people who suspected they were infected. The blood test used for both surveys claims to have a specificity rate (the rate at which it accurately identifies negative samples) of 99.5%. But when only a relatively small portion of the population has been infected, it’s possible for all serology tests, no matter how fine-tuned, to show a misleading number of false positives. As we explained in our primer on SARS-CoV-2 immunology:
Imagine a group of 100 people, 10 of whom are infected. You have a test that, similar to most of the serology tests being marketed today, is 90% effective in correctly identifying both positive and negative cases. That means you’ll get: 9 correct positives and 81 correct negatives, plus 9 false positives and 1 false negative. Of your 18 total positives, only 9 actually have the virus (and you’ve given the all-clear to one person who is actually infected). So a positive result is only accurate half the time. But try the math again when 40 people are infected, and the accuracy jumps to 89%.
Still, surveys like this are an important first look not only at the true scale of the pandemic, but at what the mortality rate may be. In a press conference, Barbara Ferrer, director of the LA County Department of Public Health, said that based on the survey’s prevalence estimate, the county’s mortality rate is between 0.1 and 0.2%, rather than above 4% as reflected by the official caseload. The Santa Clara survey also estimated a mortality rate between 0.1 and 0.2%.
The LA survey also suggested some intriguing differences between demographic groups: Men were three times more likely to be infected than women, and African Americans had the highest infection rate of any racial group. But Neeraj Sood, a USC professor of public policy who led the survey, said it was too soon to draw any conclusions from those findings.
Over the next few weeks, more data will be forthcoming from serology surveys currently being conducted by New York State, the CDC, and other researchers. Even if they also show an infection rate much higher than the official count, we’re still a very long way from the 60% to 80% exposure needed to reach herd immunity, Sood said.
“If only 4% of the population has been infected, that means we are very early in the epidemic and many more people in LA county could become infected,” Sood said. “And the number of deaths and hospitalizations and ICU visits will rise.”