PLoS ONE (Jan 2021)

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) seroprevalence: Navigating the absence of a gold standard.

  • Sahar Saeed,
  • Sheila F O'Brien,
  • Kento Abe,
  • Qi-Long Yi,
  • Bhavisha Rathod,
  • Jenny Wang,
  • Mahya Fazel-Zarandi,
  • Ashleigh Tuite,
  • David Fisman,
  • Heidi Wood,
  • Karen Colwill,
  • Anne-Claude Gingras,
  • Steven J Drews

DOI
https://doi.org/10.1371/journal.pone.0257743
Journal volume & issue
Vol. 16, no. 9
p. e0257743

Abstract

Read online

BackgroundSevere acute respiratory syndrome coronavirus 2 (SARS-CoV-2) seroprevalence studies bridge the gap left from case detection, to estimate the true burden of the COVID-19 pandemic. While multiple anti-SARS-CoV-2 immunoassays are available, no gold standard exists.MethodsThis serial cross-sectional study was conducted using plasma samples from 8999 healthy blood donors between April-September 2020. Each sample was tested by four assays: Abbott SARS-Cov-2 IgG assay, targeting nucleocapsid (Abbott-NP) and three in-house IgG ELISA assays (targeting spike glycoprotein, receptor binding domain, and nucleocapsid). Seroprevalence rates were compared using multiple composite reference standards and by a series of Bayesian Latent Class Models.ResultWe found 13 unique diagnostic phenotypes; only 32 samples (0.4%) were positive by all assays. None of the individual assays resulted in seroprevalence increasing monotonically over time. In contrast, by using the results from all assays, the Bayesian Latent Class Model with informative priors predicted seroprevalence increased from 0.7% (95% credible interval (95% CrI); 0.4, 1.0%) in April/May to 0.7% (95% CrI 0.5, 1.1%) in June/July to 0.9% (95% CrI 0.5, 1.3) in August/September. Assay characteristics varied over time. Overall Spike had the highest sensitivity (93.5% (95% CrI 88.7, 97.3%), while the sensitivity of the Abbott-NP assay waned from 77.3% (95% CrI 58.7, 92.5%) in April/May to 64.4% (95% CrI 45.6, 83.0) by August/September.DiscussionOur results confirmed very low seroprevalence after the first wave in Canada. Given the dynamic nature of this pandemic, Bayesian Latent Class Models can be used to correct for imperfect test characteristics and waning IgG antibody signals.