Epidemiological forecasting and the ecology of Ross River virus across Australia

Understanding the epidemiological mechanisms that drive disease incidence and forecasting incidence are major goals in public health. This is particularly relevant for mosquito-borne pathogens that cause human disease burden. In terms of incidence, Australia's most epidemiologically important mosquito-borne diseases is the Ross River virus (RRV, Togoviridae: Alphavirus). Ross River virus disease occurs widely across Australia with a significant burden to human health through epizootic spillover. Owing to its notifiable status, there is high-quality national health surveillance. In this thesis, my research focused on improving forecasting of RRV and understandings of the mechanisms driving patterns of RRV incidence. In Chapter 2, I describe research that I did on predictive modelling of RRV transmission in epidemic centres across Victoria determining environmental and meteorological factors associated with RRV disease incidence. I identified distinct similarities and differences in factors useful for predicting RRV incidence and outbreaks across Victoria, using readily available and inexpensive climate and meteorological data. The results highlighted new meteorological factors (i.e., rate of evapotranspiration) important in forecasting RRV incidence. The methodologies I used have formed the Ross River virus Outbreak Surveillance System for the Victorian Department of Health to monitor and provide intelligence for public health management action. The applied epidemiological implications from this study are processes and procedures to establish accessible predictive disease surveillance tools to aid public health management. This research documented how epidemiological studies can be integrated into public health systems. Having established an approach to construct a predictive surveillance system across multiple epidemic regions, in Chapter 3 I evaluated how the choice of the statistical model used in predictive modelling affects the accuracy of predictions. This was done to optimise methods for predicting RRV incidence and epidemics across Victoria and Western Australia. I determined that regional differences affect the choice of the statistical model used in establishing a predictive surveillance system. Current epidemiological studies forecasting RRV incidence often assume the best fit model for predicting RRV incidence is also the best for predicting RRV outbreaks, or vice versa. However, this assumption is rarely evaluated, and I found that the models used for predicting disease incidence may not be suitable for predicting disease outbreaks. Moreover, my research demonstrated that relative disease activity (i.e., greater number of notifications/incidence) does not lead to greater model predictive accuracy, and that, predictive accuracy is strongly influenced by model choice. Information gained from this research can be directly applied to improving and updating current RRV predictive surveillance systems. By using a common systematic approach to developing predictive disease surveillance systems, factors associated with disease can more reliably be assessed and compared across regions in relation to RRV epidemiology. In Chapter 4, I used distributed lag non-linear models to assess meteorological factors in non-linear exposure-response relationships for RRV incidence. My findings illustrated that the effect of meteorological factors on RRV incidence risk is non-linear with multiple delayed effects. I showed that there are optimal levels of exposure to meteorological factors that affect the risk of RRV incidence. My findings have the potential to allow public health personnel/departments to assess how future climate predictions may influence RRV disease activity across Australia with shifting climates. Currently predictive disease modelling of RRV often over or under predict disease activity, likely owing for complex dynamics not captured with linear predictor relationships. This research is foundational for exploring these methods as an approach to improve upon current predictive methods. In Chapter 5 I shifted my focus to improving upon mechanistic understanding of vector and host life-history traits and their importance in RRV pathogen transmission. Using empirically founded seasonally forced ordinary differential equation models, fitted to human RRV notification data, I investigated deterministic mechanisms of RRV transmission. I investigated mosquito vector, reservoir host, and transmission factors that shape RRV notifications among epidemic centres in Australia. I found that the combination of transmission mechanisms for RRV to be similar across sites supporting RRV as a multi-vector multi-host pathogen. My findings also support incidence of RRV as being largely underreported. By improving understanding of the fundamental transmission mechanisms of RRV my models have potential to aid current and future control strategies to limit the burden of human disease. Mosquito-borne diseases, such as Ross River virus disease, present an important issue for public health across Australia. The research presented in this thesis has enhanced forecasting and understanding of the mechanisms driving RRV disease incidence. The research contributions I have made help with applied epidemiological approaches in public health settings to improve population health management, such as through public health warnings associated with disease incidence, and with the control of mosquito vectors.