Tracking Dobrava-Belgrade Orthohantavirus in Alpine rodents: a detection-aware approach

Introduction Orthohantaviruses are rodent-borne zoonotic viruses maintained in wild small-mammal populations and transmitted to humans mainly through inhalation of aerosolised particles from contaminated rodent excreta. In Europe, Dobrava-Belgrade orthohantavirus (DOBV) is associated primarily with Apodemus mice and is relevant to wildlife health, occupational exposure, and One Health surveillance. However, pathogen surveillance in wildlife is often sparse, unevenly replicated, and affected by imperfect detection, meaning that non-detection cannot be interpreted directly as true absence. Materials and Methods We analysed 15 years of small-mammal surveillance data from 55 trapping sites in the Autonomous Province of Trento, northern Italy, to estimate seasonal DOBV-reactive antibody occurrence in sampled Apodemus hosts. The analytical unit was defined as site × year × season, and the response indicated whether at least one antibody-positive Apodemus individual was detected in a sampled site-period. The estimand therefore represents sampled-host serological occurrence, not active infection, viral shedding, or direct human risk. To account for imperfect detection, we applied a Bayesian spatiotemporal occupancy model separating latent serological occurrence from the observation process. Detection probability was modelled as a function of the number of tested Apodemus individuals. We then implemented a strictly out-of-fold residual-correction layer using elastic-net and random-forest learners to improve probability calibration for retrospective mapping. Spatial projections were restricted to predictors that could be applied consistently across the landscape. Results and Discussion Detection probability increased with sampling effort, but even moderate sampling left substantial scope for non-detection. The clearest ecological associations were positive gradients with lagged Apodemus density and the proportion of adult hosts, consistent with the role of host availability and demographic structure in cumulative exposure. Retrospective maps for 2003–2008 and 2010–2015 suggested a higher later-period serological signal, although these surfaces should be interpreted as calibrated sampled-host exposure maps rather than evidence of active viral circulation, geographic expansion, or spillover risk. This study provides a detection-aware framework for interpreting long-term wildlife pathogen surveillance data in heterogeneous Alpine landscapes. By distinguishing ecological occurrence from the observation process, the approach helps avoid conflating non-detection with absence and supports more robust prioritisation of ecopathological surveillance. More broadly, the work highlights the value of integrating reservoir ecology, statistical modelling, and wildlife health monitoring within a One Health perspective.