Population-based seroprevalence surveys can provide useful estimates of the number of individuals previously infected with SARS-CoV-2 and still susceptible as well as contribute to better estimates of the case fatality rate and other measures of COVID-19 severity. No serological test is 100% accurate, however, and the standard correction that epidemiologists use to adjust estimates relies on estimates of the test sensitivity and specificity often based on small validation studies. This paper develops a fully Bayesian approach to adjust observed prevalence estimates for sensitivity and specificity. Application to a seroprevalence survey conducted in New York State in 2020 demonstrates that this approach results in more realistic – and narrower – credible interval than the standard sensitivity analysis using confidence interval endpoints. In addition, the model permits incorporating data on the geographical distribution of reported case counts to create informative priors on the cumulative incidence to produce estimates and credible intervals for smaller geographic areas than often can be precisely estimated with seroprevalence surveys.
Adjusting COVID-19 seroprevalence survey results to account for test sensitivity and specificity