Expanding cholera serosurveillance to vaccinated populations

Mass oral cholera vaccination campaigns targeted at subnational areas with high incidence are central to global cholera elimination efforts. Serological surveillance offers a complementary approach to address gaps in clinical surveillance in these regions. However, similar immune responses from vaccination and infection can lead to overestimates of the incidence of infection. To address this, we analyzed antibody dynamics in infected and vaccinated individuals to refine seroincidence estimation strategies for partially vaccinated populations. We tested 757 longitudinal serum samples from confirmed Vibrio cholerae O1 cases and uninfected contacts in Bangladesh as well as vaccinees from Bangladesh and Haiti, using a multiplex bead assay to measure IgG, IgM, and IgA binding to five cholera-specific antigens. Infection elicited stronger and broader antibody responses than vaccination, with rises in cholera toxin B-subunit (CTB) and toxin-coregulated pilus A (TcpA) antibodies uniquely associated with infection. Previously proposed random forest models frequently misclassified vaccinated individuals as recently infected (over 20% at some time points) during the first 4 months post-vaccination. To address this, we developed new random forest models incorporating vaccinee data, which kept false-positive rates among vaccinated (1%) and unvaccinated (6%) individuals low without a significant loss in sensitivity. Simulated serosurveys demonstrated that unbiased seroincidence estimates could be achieved within 21 days of vaccination campaigns by ascertaining the vaccination status of participants or applying updated models. These approaches to overcome biases in serological surveillance enable reliable seroincidence estimation even in areas with recent vaccination campaigns enhancing the utility of serological surveillance as an epidemiologic tool in moderate-to-high cholera incidence settings.

Serological surveillance can improve how we monitor cholera in high-burden areas where clinical surveillance is limited. However, vaccination can produce immune responses similar to infection, leading to overestimates in seroincidence. This study extends seroincidence estimation techniques using machine learning models to partially vaccinated populations. We analyzed antibody dynamics from vaccinated and infected individuals to develop methods that reduce the misclassification of vaccinated individuals as recently infected. These methods enable reliable seroincidence estimates in areas with recent vaccination campaigns, providing a step toward better epidemiologic monitoring in the context of global cholera control initiatives. Studies in other populations are needed to further validate our results and understand their generalizability.