(P-IT-1) Assessing Methods to Adjust for Assay Differences across COVID-19 Serological Surveillance Studies in Canada

Previous research found substantial variability in SARS-CoV-2 seroprevalence estimates using various serological assays. Notably, the Roche Elecsys^® Anti-SARS-CoV-2 anti-nucleocapsid (anti-N) assay showed less waning in sensitivity with time since infection compared to the Abbott SARS-CoV-2 IgG Assay anti-N assay, limiting comparability of serosurveillance studies using different assays.

Study

Design/Methods: We analyzed two residual blood serosurveillance datasets in Alberta, Canada: 124,008 blood donor plasma samples tested with Roche Elecsys^® Anti-SARS-CoV-2 (anti-N) by Canadian Blood Services (CBS) and 214,780 outpatient lab plasma or sera samples tested with the Abbott SARS-CoV-2 IgG Assay (anti-N) by Alberta Precision Labs (APL). In each dataset, we estimated crude seroprevalence, and seroprevalence with the Rogen-Gladen (RG) adjustment for previously reported sensitivity and specificity using Bayesian credible intervals. Using data from samples tested with both assays early in the pandemic (in 2020), we developed linear regression models to transform CBS Roche qualitative results to Abbott and applied the Abbott cutoff index and made the opposite transformation to convert APL to Roche. Finally, we developed generalized linear models using data collected in the early pandemic to estimate the probability a tested sample was tested 0-3, 4-6, 7-9, or >9 months from infection and applied this model to the CBS and APL datasets to estimate rolling incidence from 2021 to 2023.
 

Results/Findings:

The datasets had similar seropositivity earlier in the pandemic but diverged after May 2022 due to greater waning sensitivity in the Abbott assay used by APL (Figure A).  RG and Bayesian adjustments had minimal impact, increasing sero-prevalence estimates by 2.04% for the CBS Alberta subset and 7.76% for APL (Panel A). Our regression-based approaches increased concordance between estimated seropositivity (Panel B) and led to more similar estimates of rolling incidence (Panel C). The gap between CBS and APL seropositivity estimates for October 2022 was 48% using raw seropositivity, 15% after transforming CBS to Abbott, and 14% after transforming APL to Roche.

Conclusions: Serosurveillance estimates are sensitive to choice of assay, and the most common adjustment methods (RG) do little to make estimates more concordant, particularly later in the pandemic when differences in waning sensitivity become important. Our regression-based approaches improve comparability across studies using different assays. Rolling incidence may be a more relevant population health measure in the post-pandemic era where history of SARS-CoV-2 is ubiquitous. These methods may help to integrate and compare serosurveillance data in blood donors with other populations that use different assays.