Abstract
Salmonella Dublin, a cattle-adapted serovar of Salmonella enterica ssp. enterica with zoonotic potential, has become a pathogen of increasing concern for the dairy industry in Canada and beyond. A comprehensive understanding of its epidemiology is crucial to guide effective surveillance and control. The objectives of this serial cross-sectional study were to estimate the true herd-level seroprevalence of Salmonella Dublin and to characterize the spatial distribution of Salmonella Dublin-seropositive dairy herds in Alberta, Canada. Bulk tank milk (BTM) samples were collected from all active dairy herds in Alberta, Canada, in June (n = 462) and December (n = 451) 2024, and PrioCHECK ELISA was used to detect antibodies against Salmonella Dublin. Herds were classified as positive if ≥ 1 sample had a percent positivity ≥ 35. True seroprevalence was estimated using a Bayesian latent class model accounting for imperfect sensitivity and specificity. Spatial patterns were evaluated using a Bayesian geostatistical model, and a spatial scan statistic was used to detect localized clusters. The estimated true prevalence of Salmonella Dublin-seropositive herds in Alberta was 14.9% (95% Bayesian credible interval: 7.3-28.2%), with regional estimates of 8.0% (95% BCI = 1.6-22.6%), 13.9% (95% BCI = 4.2-31.8%), and 20.9% (95% BCI = 8.5-42.5%) for the north, central, and south regions of Alberta, respectively. The continuous spatial risk surface revealed that the true herd seropositivity varied markedly across Alberta, with some areas having much higher predicted seropositivity probabilities than others. Furthermore, 2 localized clusters of higher seropositivity were detected: a prominent cluster (95-km radius) in southwestern Alberta and a relatively smaller cluster (19-km radius) in a high-density dairy corridor in the central region. These findings indicate that Salmonella Dublin exposure is moderate in Alberta, with marked regional heterogeneity and pronounced spatial gradients in herd-level seropositivity. Incorporating spatial context into surveillance and control may improve targeting of interventions aimed at reducing within-region transmission. These findings also illustrate the value of Bayesian latent class modeling for generating bias-adjusted true seroprevalence estimates in the absence of a gold-standard diagnostic test.