Abstract
Since there is often different information contained in observations from different data sources, outbreak detection systems should be multivariate by nature. Experience from public health shows that, in reality they often fail to achieve acceptable sensitivity while retaining manageable false alert rates. A valuable alternative to classical \"outbreak detection\" is \"outbreak prediction\" based on suitably selected model. We think that such an approach is particularly promising for multivariate surveillance. We propose to use Swiss multivariate surveillance data to develop model-based predictive methods which can be used to inform decisions about animal health.