Abstract
Dengue virus (DENV) is an increasingly important human pathogen, with already half of the globe's population living in environments with transmission potential. Since many cases are missed by direct detection methods (RT-PCR or antigen tests), serological assays play an important role in the diagnostic process. However, individual assays can suffer from low sensitivity and specificity and interpreting results from multiple assays remains challenging, particularly because interpretations from multiple assays may differ, creating uncertainty over how to generate finalized interpretations. We develop a Bayesian mixture model that can jointly model data from multiple paired serological assays, to infer infection events. We first test the performance of our model using simulated data. We then apply our model to 677 pairs of acute and convalescent serum collected as a part of illness and household investigations across two longitudinal cohort studies in Kamphaeng Phet, Thailand, including data from 232 RT-PCR confirmed infections (gold standard). We compare the classification of the new model to prior standard interpretations that independently utilize information from either the hemagglutination inhibition assay (HAI) or the enzyme-linked immunosorbent assay (EIA). We find that additional serological assays improve accuracy of infection detection for both simulated and real world data. Models incorporating paired IgG and IgM data as well as those incorporating IgG, IgM, and HAI data consistently have higher accuracy when using PCR confirmed infections as a gold standard (87-90% F1 scores, a combined metric of sensitivity and specificity) than currently implemented cut-point approaches (82-84% F1 scores). Our results provide a probabilistic framework through which multiple serological assays across different platforms can be leveraged across sequential serum samples to provide insight into whether individuals have recently experienced a DENV infection. These methods are applicable to other pathogen systems where multiple serological assays can be leveraged to quantify infection history.