So what is the #seroprevalence of #SARS-CoV-2?
In short, we don’t know. As indicated above, a potential indication for antibody testing is to determine the percent of the population who have antibodies, which may be useful both in understanding the pandemic thus far and for predicting the future course of the outbreak. Although a number of studies have estimated seroprevalence, many of them have unfortunately received media attention inversely proportional to the quality of the methodology. For others, the quality of the methodology cannot be completely assessed because the authors have shared their results in media reports rather than scientific papers. Few large-scale studies exist and estimated prevalence ranges widely from 2% to over 30%.
The highest reported prevalence comes from a study of 200 people in Chelsea, MA. As covered in a prior FLARE (April 21), Chelsea has been hard hit by COVID-19. Researchers from MGH tested blood from 200 volunteers and found that 31% had antibodies. This research has not yet been published (it has been reported locally and was described by the Boston Globe) so it is not possible to thoroughly evaluate the methodology. The study has been criticized both for recruiting a non-random sample and for using a test with a specificity as low as 90%, potentially resulting in a high false positive rate.
Sampling biases may have confounded other widely-publicized seroprevalence studies as well. A study from Germany (to date only available as a preprint) reported a roughly 15% prevalence, sampling 919 individuals from 405 households (Streeck et al. 2020). These are clearly not independent samples - members of the same household are highly likely to have concordant antibody status. The lack of generalizability did not stop the authors from conducting extensive media interviews. Similar concerns have been expressed about a seroprevalence study (Bendavid et al. 2020) conducted in Santa Clara County, CA in which the authors reported a seroprevalence rate of 1.5%. They subsequently adjusted this for demographic and test performance criteria (sensitivity 83% and specificity 99%) to 2.8%. The study recruited subjects via Facebook ads and so it likely also suffers from selection bias. So far this study is also available only as a preprint. In another similarity to the German study, the methodologic shortcomings and lack of peer review have not stopped the authors from conducting extensive media tours.
Some of the same authors (Sood et al. 2020) attempted to address the flaws of their Santa Clara study by conducting one of the few studies on seroprevalence that has been subject to peer review. They used a test with an estimated sensitivity of 83% and a specificity of 99%. They used a randomly generated sample of residents in Los Angeles County, but only approximately 50% of invited participants agreed to be tested. The tested group failed to match county demographics and the proportion of positive tests did in fact vary by ethnic group. In addition, ~13% of their sample was symptomatic, likely indicating that symptomatic people were more likely to participate.
The largest seroprevalence study to date was conducted in Boise, Idaho (Bryan et al. 2020). These authors first carefully determined the specificity of their assay by testing a large number of serum samples predating SARS-CoV-2 and found only one false positive from 1020 samples, suggesting a specificity of ~99%. They then tested serum from patients with PCR confirmed SARS-CoV-2 infection and found a sensitivity of roughly 50% on day 7 from symptom onset, which rose to 100% by day 17 from symptom onset. They then tested 4,856 community members as part of an initiative to determine the community wide prevalence. While this is a large number of individuals, they were self-selected. The authors report a prevalence of 1.79% which should likely be regarded as something of an overestimate given sampling bias.
Other studies attempt to analyze the population prevalence of disease not with antibody testing, but by using various modeling techniques to extrapolate from hospitalization data. This was the approach used by Ceylan and colleagues to arrive at an estimated prevalence of 4.4% for the nation of #France (Ceylan 2020). Unfortunately, the relationship between hospitalized cases and prevalence of mild or asymptomatic infection was based on data from the Diamond Princess cruise ship outbreak, in which all passengers were tested. It is unclear whether data from this confined population is generalizable to the national level.
In sum, the available data on seroprevalence are of low quality and very few studies have been subjected to peer review . In our opinion, none are of high enough quality to drive evidence-based decision making. Given the intense public interest in this issue, it seems wisest for researchers to avoid speaking to the media prior to formally vetting their results . The uncertainty about the quality of different antibody tests (discussed above) adds yet another level of uncertainty about the prevalence of SARS-CoV-2.