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

ABSTRACT

Background/Case Studies Multiple assays to detect SARS-COV-2 antibodies are available but no gold standard exists. Due to many factors including waning antibodies and differences in test designs, discordance between SARS-CoV-2 serology assays is common. Given these limitations we used multiple assays and methodological approaches to estimate SARS-COV-2 seroprevalence during the first COVID-19 wave in Canada. Study Design/Methods: This serial cross-sectional study was conducted using residual plasma from healthy blood donors between April-September 2020. Qualitative (Table Presented) assessment of SARS-CoV-2 IgG antibodies was based on four assays Abbott Architect SARS-Cov-2 IgG assay (target nucleocapsid) (Abbott-NP) and three in-house IgG ELISA assays (target spike glycoprotein (Spike), spike receptor binding domain (RBD), and nucleocapsid (NP)) based on thresholds set by the manufacture or 3-standarddeviations from the negative mean. We compared seroprevalence rates by multiple composite reference standards (CRS) and by a series of Bayesian Latent Class Models (BLCM) (using uninformative, weakly and informative priors). Using the BLCM we estimated assay characteristics, bimonthly to evaluate changes over time. Results/Findings: In total, 8999 blood samples were tested. The Abbott-NP assay consistently estimated seroprevalence to be lower than the ELISA-based assays. A priori, choosing a combination of 2 assays resulted in a range of seroprevalence estimates that ranged from 0.2% to 0.5% in April to 0.4% to 1.5% in September. From 16 possible diagnostic phenotypes, 13 were observed, only 33 samples (0.4%) were positive by all four assays. BLCM with non-informative priors provided the best model fit and predicted seroprevalence increased from 0.7% (95% CrI;0.6, 0.8%) in April/May to 1.0% (0.8, 1.1%) in June/ July to 1.5% (1.3, 1.8) in August/September. Assay characteristics varied considerably over time. Overall RBD had the highest sensitivity 82.2% (69.3, 92.9%) with a specificity of 99.6% (99.4, 99.7%). In contrast the sensitivity of the Abbott-NP assay was the lowest and waned from 63.2% (41.4, 83.1%) in April/May to 33.9% (19.7, 53.1%) by August/September. Conclusions: Regardless of the analytical method we found at the end of the first COVID-19 wave, SARSCoV- 2 seroprevalence among a healthy population of blood donors was low (<2%). While the sensitivity of all assays waned, the rates did vary. We found significant limitations to using a single assay to estimate SARSCoV- 2 seroprevalence in a low prevalence setting, such as healthy Canadian blood donors during the first wave of the COVID-19 pandemic.