Bayesian modelling of herd-level infection dynamics in cattle: Local spread as the primary driver of Salmonella Dublin persistence on Öland

Abstract

Salmonella Dublin (S. Dublin), a zoonotic serotype adapted to cattle, causes animal welfare issues and economic losses. The disease has proven particularly challenging to control in Öland, Sweden. This study uses Bayesian simulation-based inference of bulk tank milk sample results to analyse the S. Dublin infection dynamics in Öland cattle. The infection process was formulated as a dynamic state-space model and particle Markov-chain Monte Carlo methods were applied to infer the underlying infection dynamics and estimate the basic reproduction number ($R_0$) as well as the effective reproduction number ($R_t$). These metrics provide insight into transmission dynamics, enabling assessment of the effectiveness of the current S. Dublin control in Swedish cattle and identification of interventions that may reduce the prevalence. The results show that most holdings on Öland have $R_0 < 1$, indicating that infection is expected to die out after introduction. However, in a subset of holdings $R_0 > 1$, and there the risk for spread of S. Dublin is higher. Furthermore, the analysis reveals that on average, $R_t \approx 1$, suggesting a stable endemic presence unless effective interventions are implemented. In addition, the results show that it is insufficient to restrict the movements of infected cattle on Öland to bring $R_t < 1$, as local spread and within-herd transmission contribute equally to the force of infection (approximately 50% each). These findings demonstrate how Bayesian data-driven analysis can support evidence-based decision making for the control and eradication of S. Dublin in cattle.