Improving SARS-CoV-2 cumulative incidence estimation through mixture modelling of antibody levels

Abstract As countries decide on vaccination strategies and how to ease movement restrictions, estimates of cumulative incidence of SARS-CoV-2 infection are essential in quantifying the extent to which populations remain susceptible to COVID-19. Cumulative incidence is usually estimated from seroprevalence data, where seropositives are defined by an arbitrary threshold antibody level, and adjusted for sensitivity and specificity at that threshold. This does not account for antibody waning nor for lower antibody levels in asymptomatic or mildly symptomatic cases. Mixture modelling can estimate cumulative incidence from antibody-level distributions without requiring adjustment for sensitivity and specificity. To illustrate the bias in standard threshold-based seroprevalence estimates, we compared both approaches using data from several Kenyan serosurveys. Compared to the mixture model estimate, threshold analysis underestimated cumulative incidence by 31% (IQR: 11 to 41) on average. Until more discriminating assays are available, mixture modelling offers an approach to reduce bias in estimates of cumulative incidence. One-Sentence Summary Mixture models reduce biases inherent in the standard threshold-based analysis of SARS-CoV-2 serological data.