A spatio-temporal hierarchical Markov switching model for the early detection of influenza outbreaks

Rapidly detecting the beginning of influenza outbreaks helps health authorities to reduce their impact. Accounting for the spatial distribution of the data can greatly improve the performance of an outbreak detection method by promptly detecting the first foci of infection. The use of Hidden Markov chains in temporal models has shown to be great tools for classifying the epidemic or endemic state of influenza data, though their use in spatio-temporal models for outbreak detection is scarce. In this work, we present a spatio-temporal Bayesian Markov switching model over the differentiated incidence rates for the rapid detection of influenza outbreaks. This model focuses its attention on the incidence variations to better detect the higher increases of early epidemic rates even when the rates themselves are relatively low. The differentiated rates are modelled by a Gaussian distribution with different mean and variance according to the epidemic or endemic state. A temporal autoregressive term and a spatial conditional autoregressive model are added to capture the spatio-temporal structure of the epidemic mean. The proposed model has been tested over the USA Google Flu Trends database to assess the relevance of the whole structure.