Home > Research > Publications & Outputs > Outbreak response forecasting for vector borne ...

Associated organisational unit

Electronic data

  • chrisJ

    6.46 MB, PDF document

    Available under license: None

View graph of relations

Outbreak response forecasting for vector borne diseases: theileria orientalis (Ikeda) in NZ cattle

Research output: Contribution to conference - Without ISBN/ISSN Conference paper

Published
Publication date21/07/2016
<mark>Original language</mark>English

Abstract

Dynamical models of communicable diseases have become a prominent feature of national-level epidemic response. Developments in Bayesian inference have enabled these models to provide quantitative risk predic- tions in a real-time setting, learning from spatiotemporal data as it arrives from the field. However, these models rely heavily on accurate covariate data from which to make inference. Incursions of vector borne disease present a particular challenge in this respect, as exemplified by the recent introduction of Theileria orientalis (Ikeda), an obligate tick-borne disease of cattle, into New Zealand. Whereas the location of cattle and the animal movement network between farms is well recorded, little is known about the national scale ecology of the tick vector. This talk will present a Bayesian data assimilation approach to this problem, in which vector presence is modelled as a discrete-space latent process with a continuous-time seasonality. A joint likelihood function assimilates the epidemic data and results from a national disease surveillance pro- gramme designed for a different disease. A spatiotemporally inhomogeneous Poisson process is used to model the epidemic, with an a priori independent hierarchical binomial surveillance model. This joint model is fitted to observed case detection data using a non-centered trans-dimensional MCMC algorithm, integrating over the marginal posterior of the latent vector surface, censored herd infection times, and the presence of undetected infections. Importantly, the algorithm is implemented using GPGPU technology which acceler- ates within-chain likelihood calculations to an overnight timeframe. Finally, the predictive distribution is provided as a real time disease forecast for decision support purposes.