Detecting disease progression from animal movement using hidden Markov models

Detecting infectious disease in wildlife is critical for conservation, management of reintroduction programmes, and to reduce the risk of spillover into livestock and humans. However, collecting diagnostic samples from free-ranging animals is logistically difficult and costly. Many pathogens alter host behaviour, including reductions in movement, suggesting that animal tracking data could offer a way to infer infection status. We develop a modelling framework that links animal movement to disease processes using hidden Markov models (HMMs). Infection status is treated as a hidden (unobserved) state, while movement patterns, such as step lengths and turning angles, serve as state-dependent observations. This structure is similar to epidemiological compartmental models (‘Susceptible-Infected-Recovered; SIR’), allowing movement data to be formally connected to disease progression. We test multiple model formulations, including (1) constrained state transition probabilities to preclude or include recovery, (2) covariate effects to test whether ecological factors influence infection risk and (3) hierarchically structured HMMs (HHMMs) to distinguish movement responses at short and long temporal scales. We apply this framework to GPS movement data from 84 reintroduced scimitar-horned oryx (Oryx dammah) in Chad. During the study period, 38 individuals were confirmed dead and 6 were sampled for pathogens (e.g. Rift Valley Fever, Peste des Petits Ruminants, Babesiosis). Models that included biologically realistic constraints on transitions among infection states successfully detected disease-associated reductions in movement and aligned with veterinary necropsy and diagnostic reports. Unconstrained models performed poorly, misclassifying individuals and failing to detect disease progression. Simulation results provided further evidence that constrained HMMs can infer susceptible, infected and recovered states under appropriate model specifications. Synthesis and applications. We demonstrate how (H)HMMs can be tailored to different epidemiological scenarios and provide a template workflow for developing and selecting Hidden Markov models to infer disease status from animal movement data. Identifying infection before mortality occurs offers a valuable early-warning tool for population managers, reduces reliance on difficult and costly field testing and improves surveillance strategies for vulnerable and reintroduced populations.