Estimation of stratified seroprevalence directly from raw serological assay measurements with multi-level Bayesian mixture modelling

The article describes a new statistical method for estimating seroprevalence, the proportion of people with antibodies against infections. Traditional approaches analyze serological tests sequentially in multiple steps, leading to information loss and uncertainty. The research team developed a Bayesian model called dvsb that combines all analysis steps into a single hierarchical model. This method better propagates uncertainty through all parts of the analysis and allows classification thresholds to be adapted to individual subpopulations. When tested on simulated data, the traditional approach produced estimates far from true values, while the new dvsb method was markedly more accurate. The method was validated on a real dataset with 72,863 measurements for 21,391 samples for Lassa fever antibodies. The code is available on GitHub.