Inferring time of infection from field data using dynamic models of antibody decay

Studies of infectious disease ecology often rely heavily on knowing when individuals were infected, but estimating this time of infection can be challenging, especially in wildlife. Time of infection can be estimated from various types of data, with antibody level data being one of the most promising sources of information. The use of antibody levels to back-calculate infection time requires the development of a host-pathogen system-specific model of antibody dynamics, and a leading challenge in such quantitative serology approaches is how to model antibody dynamics in the absence of experimental infection data. Here, we present a way to do this in a Bayesian framework that facilitates the incorporation of all available information about potential infection times. We apply the model to estimate infection times of Channel Island foxes infected with Leptospira interrogans, leading to reductions of 51-92% in the window of possible infection times. Using simulated data, we show that the approach works well across a broad range of parameter settings and can lead to major improvements of infection time estimates that depend on system characteristics such as antibody decay rate and variation in peak antibody levels after exposure. The method substantially simplifies the challenge of modeling antibody dynamics in the absence of individuals with known infection times, opens up new opportunities in wildlife disease ecology, and can even be applied to cross-sectional data once the model is trained.