Estimating landscape spread of a low-prevalence disease using multiple surveillance methods.
Abstract
Diseases occurring at low prevalence in free-ranging wildlife can be difficult to study and predict at a landscape scale. This problem is compounded by imperfect pathogen detection. We used data from multiple surveillance methods to determine factors influencing spread of a low-prevalence disease, bovine tuberculosis (bTB), among white-tailed deer (Odocoileus virginianus) in a bTB endemic area over a 10-year period. We fit dynamic occupancy models to predict spatial spread of bTB through a county in Michigan, USA, within the bTB endemic area, while accounting for imperfect detection and a suite of ecological processes potentially driving bTB spread. Detection by all surveillance methods was low overall (posterior mean detection probabilities 0.01-0.03). Probability of bTB occurrence was 1.46 times as likely in the historical centre of the outbreak compared to outside. Deer density had a weakly positive effect on bTB spread. The probability of a site being newly infected with bTB annually was low (posterior mean colonization probabilities 0.08-0.27), while the probability of persistence after initial infection was high (posterior mean persistence probability 0.67). Synthesis and applications: We used three surveillance methods to estimate probability of presence and spread of a low-prevalence pathogen at a landscape scale under incomplete sampling coverage. Our methods are relevant for determining spatial risk of any wildlife disease that is monitored using multiple methods of surveillance, which is important for prioritization of management actions.