The forecasting of dynamical Ross River virus outbreaks: Victoria, Australia

Ross River virus (RRV) is Australia’s most epidemiologically important mosquito-borne disease.During RRV epidemics in the State of Victoria (such as 2010/11 and 2016/17) notifications canaccount for up to 30% of national RRV notifications. However, little is known about factors which canforecast RRV transmission in Victoria. We aimed to understand factors associated with RRVtransmission in epidemiologically important regions of Victoria and establish an early warningforecast system. We developed negative binomial regression models to forecast human RRVnotifications across 11 Local Government Areas (LGAs) using climatic, environmental, andoceanographic variables. Data were collected from July 2008 to June 2018. Data from July 2008 toJune 2012 were used as a training data set, while July 2012 to June 2018 were used as a testing dataset. Evapotranspiration and precipitation were found to be common factors for forecasting RRVnotifications across sites. Several site-specific factors were also important in forecasting RRVnotifications which varied between LGA. From the 11 LGAs examined, nine experienced an outbreakin 2011/12 of which the models for these sites were a good fit. All 11 LGAs experienced an outbreakin 2016/17, however only six LGAs could predict the outbreak using the same model. We documentsimilarities and differences in factors useful for forecasting RRV notifications across Victoria anddemonstrate that readily available and inexpensive climate and environmental data can be used to predict epidemic periods in some areas. Furthermore, we highlight in certain regions the complexityof RRV transmission where additional epidemiological information is needed to accurately predictRRV activity. Our findings have been applied to produce a Ross River virus Outbreak SurveillanceSystem (ROSS) to aid in public health decision making in Victoria