A simple spatio-temporal algorithm for disease surveillance using routinely collected data

In Australia diagnostic data from medical practitioners and laboratories for over 60 different notifiable diseases are reported at a national level and compiled into the National Notifiable Diseases Surveillance System (NNDSS). Considerable time and resources are invested in the collection of these disease notification data. Due to the large number of diseases under surveillance, the performance of comprehensive daily analyses for the early detection of disease outbreaks is a time-consuming process. We aim to develop tools to allow epidemiologists to make better use of existing disease surveillance data. We explore the application of automated spatio-temporal algorithms for disease surveillance to assist epidemiologists to monitor large volumes of routinely collected data. A simple surveillance algorithm based on a Bayesian space-time hidden Markov model was developed to identify spatio-temporal aberrations in surveillance data. The algorithm monitors the distribution of notified cases of disease based on the date of diagnosis and post code of residence, and uses probability concepts to expressthe uncertainty associated with the likelihood of an outbreak. To illustrate the method developed we apply the algorithm to hepatitis A diagnoses as a case study, and assess the ability of the algorithm to identify events of concern to epidemiologists. This paper describes the algorithm developed and evaluates the ability of the algorithm to detect disease outbreaks before they become widespread