ESTIMATION of SARS-CoV-2 CUMULATIVE INCIDENCE: AN APPLICATION of MIXTURE MODELING

ABSTRACT

Background: With global vaccine scale-up, the utility of the more stable anti-S IgG assay in seroprevalence studies is limited. P population prevalence estimates of anti-N IgG SARS-CoV-2 using alternate targets (eg, anti-N IgG) will be critical for monitoring cumulative SARS-CoV-2 incidence., We demonstrate the utility of a Bayesian approach that accounts for heterogeneities in SARS-CoV-2 seroresponse (eg, must consider mild infections and/or antibody waning) to ensure anti-N IgG prevalence is not underestimated and correlates not misinterpreted. Methods: We sampled 4,828 participants from 2,723 households across 100 unique geospatial locations in Chennai, India, from Jan-May, 2021 when <1% of the general population was vaccinated. All samples were tested for SARS-CoV-2 IgG antibodies to S and N using the Abbott ARCHITECT. We calculated prevalence using manufacturer cut-offs and applied a Bayesian mixture model. In the mixture model, individuals were assigned a probability of being seropositive or seronegative based on their normalized index value, accounting for differential immune response by age and antibody waning. Regression analyses to identify correlates of infection defined seropositivity by manufacturer cut-offs and the mixture model. Results: The raw SARS-CoV-2 seroprevalence using IgG to S (cutoff=50) and N (cutoff=1.4) were 61.9% (95% confidence interval [CI] 60.5-63.3%) and 13.7% (CI 12.8-14.7%), respectively with a correlation of 0.33. With the mixture model, anti-N IgG prevalence was 65.4% (95% credible interval [CrI] 61.8-68.9). Correlates of anti-N IgG positivity differed qualitatively by the two approaches (Table). Using the manufacturer cut-off, income loss during the pandemic, household crowding and lack of air conditioning were associated with significantly lower anti-N prevalence. By contrast, in the mixture model, many measures of lower socioeconomic status were associated with higher prevalence, associations that were comparable when anti-S was the outcome. The age pattern differed between approaches the mixture model identified that individuals aged >50 had the lowest seroprevalence, but the highest immune response to infection. Conclusion: With global vaccine scale-up, population prevalence estimates of anti-N IgG will be critical for monitoring cumulative SARS-CoV-2 incidence. We demonstrate the utility of a Bayesian approach that accounts for heterogeneities in SARS-CoV-2 seroresponse to improve accuracy of anti-N IgG prevalence estimates and associated correlates.