A mixture model to estimate SARS-CoV-2 seroprevalence in Chennai, India

Background Serological assays used to estimate SARS-CoV-2 seroprevalence rely on manufacturer cut-offs established based on more severe early cases who tended to be older. Methods We conducted a household-based serosurvey of 4,677 individuals from 2,619 households in Chennai, India from January to May, 2021. Samples were tested for SARS-CoV-2 IgG antibodies to the spike (S) and nucelocapsid (N) proteins. We calculated seroprevalence using manufacturer cut-offs and using a mixture model in which individuals were assigned a probability of being seropositive based on their measured IgG, accounting for heterogeneous antibody response across individuals. Results The SARS-CoV-2 seroprevalence to anti-S and anti-N IgG was 62.0% (95% confidence interval [CI], 60.6 to 63.4) and 13.5% (95% CI, 12.6 to 14.5), respectively applying the manufacturer’s cut-offs, with low inter-assay agreement (Cohen’s kappa 0.15). With the mixture model, estimated anti-S IgG and anti-N IgG seroprevalence was 64.9% (95% Credible Interval [CrI], 63.8 to 66.0) and 51.5% (95% CrI, 50.2 to 52.9) respectively, with high inter-assay agreement (Cohen’s kappa 0.66). Age and socioeconomic factors showed inconsistent relationships with anti-S IgG and anti-N IgG seropositivity using manufacturer’s cut-offs, but the mixture model reconciled these differences. In the mixture model, age was not associated with seropositivity, and improved household ventilation was associated with lower seropositivity odds. Conclusions With global vaccine scale-up, the utility of the more stable anti-S IgG assay may be limited due to the inclusion of the S protein in several vaccines. SARS-CoV-2 seroprevalence estimates using alternative targets must consider heterogeneity in seroresponse to ensure seroprevalence is not underestimated and correlates not misinterpreted.