Spatial modelling, analysis and prediction of Hendra disease outbreaks in South East Queensland, Australia

Spatial modelling, analysis and prediction of Hendra disease outbreaks in South East Queensland, Australia

Hendra Virus (HeV) is an emerging zoonotic disease that was first identified in 1994 and has only been found in Australia. It can be transmitted to other horses, humans and dogs with a high fatality rate of >79 per cent in horses and 57 per cent in humans giving it both veterinary and public health significance. Fruit bats (Pteropus spp.) commonly known as flying-foxes have been identified as the natural host of the virus. From 1994 to 2015 inclusive, there have been more than 70 sporadic confirmed cases of HeV infection in horses. All cases have occurred in Queensland and in north-east New South Wales. The research on the HeV has almost begun immediately after the first outbreak. Government organisations as well as scientists and academicians from a broad range of disciplines, including the animal health, environmental and social sciences, are working together to develop a 'One Health' approach that will help minimise the impact of HeV. This research uses a GIS-based spatial approach to research and determine the potential factors that can explain the dispersal of HeV outbreaks in the south east Queensland, Australia. The aim of this research is to identify the equine population ‘at risk’ and thus identifying the human population ‘at risk’ in the study area.

A preliminary spatial analysis examined the relationship between the Hendra disease outbreaks and the roosting sites of flying foxes in the study area. There are four main roosting site categories which are permanent (continuous or seasonal use), temporary (occupied or unoccupied), abandoned and destroyed. This analysis showed a strong relationship between the outbreak events and the existence of temporary and seasonal flying fox roosting sites within a 10 kms range. But very few disease outbreak incidents have a permanent roosting site in their range. This provided a strong case for further study into the seasonal behaviour of flying foxes, particularly in breeding season. This analysis revealed that variables such as species and their foraging range, breeding time, equine data, and environment aspects such as types of vegetation and seasonal changes could provide suitable factors for the determination of potential factors that can explain the dispersal of HeV outbreaks in the study area.

Based on the preliminary results, a further analysis was done on the roosting sites by considering factors such as the species of flying foxes, foraging range and pregnancy period. Global Moran’s I method (inverse distance conceptualisation) was used to identify the presence of significant spatial clustering of the three flying fox species at various foraging ranges (10, 20, 30, 40 and 50 kms) in the study area. Global Moran’s I revealed significant clustering of P. alecto and P.scapulatus species. The analysis of P. alecto species showed significant clustering at all foraging range intervals with high occurrence at 50 kms, which is their maximum foraging range. The results of P.scapulatus species showed maximum significant clustering occurring at 10 kms range. Kernel density estimation (KDE technique) analysis helped in establishing a strong relationship between P. alecto and P.scapulatus species density and the outbreak events in the study area and revealed the density hotspots of these species. Buffer analysis established an initial relationship between P. alecto and P. poliocephalus species birth periods and the outbreak incidents.

The ordinary least squares (OLS) regression analysis was carried out using the ‘incident rate’ as a dependent variable and black flying foxes, grey-headed flying foxes and pregnancy period as independent variables. This model has a statistically significant heteroscedasticity (p<0.05) which suggests the use of Robust P to determine the coefficient significance for consideration. Goodness-of-fit measure indicated a model performance of 0.7. Ordinary least squares (OLS) regression identified P. poliocephalus species as statistically significant at a global context across the study area. The variance inflation factor (VIF) values indicated no redundancy among the variables. Moran’s I test (Index = -0.02, P = 0.8) indicated no significant clustering among the residuals. An exploratory method approach was exercised to calibrate the model for local regression (GWR), which used the most significant exploratory variables that could explain the trends of dispersion of HeV in the study area. Geographically weighted regression (GWR) analysis performed to study the local spatial variations of the explanatory variables in the study area identified P. alecto and P. poliocephalus species as having a significant positive relationship in most of the regions. ‘Pregnancy/Birth period’ variable exhibited a significant negative relationship to the HeV incidents in the study area. The goodness-of-fit measure indicated an improvement from 0.7 (global model) to 0.8. Moran’s I test (Index = -0.02, P = 0.9) indicated no significant clustering among the residuals. The spatial variability of the local parameter estimates of each variable in the GWR model has been tested and a significant spatial variability was present in the variables.

An in-depth analysis was carried out to determine the correlation between food source vegetation and the flying foxes roosting sites in the study area. Using spatial analyst tools, the major vegetation subgroups (MVS) present within 20 kms range of P. alecto and P. poliocephalus roosting sites were identified. The identification of abundance of food sources for individual species within their minimum foraging range indicated a strong correlation between their site locations and the vegetation subgroups present. A 10 kms range vegetation study on the incident locations identified the presence of ‘food sources’ of both species. The clustering of the food resource vegetation present near the incidence was studied using Getis-Ord General G Statistic method, which indicated statistically high clustering with 99% confidence level at 3 kms distance threshold. A 10 kms range vegetation study on the equine properties in the study area identified the food source vegetation of both significant species. The clustering of the food source vegetation present near the equine properties was studied using high/low clustering/Getis-Ord General G Statistic method, which indicated statistically significant high clustering at 3, 5 and 10 kms distance thresholds. The vegetation analysis revealed a strong correlation between the roosting sites, food source vegetation and the equine properties.

Based on the above analysis, three prediction models were produced to identify the equine population ‘at risk’ in the study area. These models were based on the presence of the significant species identified in the GWR model and the clustering of their food source vegetation in statistically significant high clusters within 20 kms from the equine properties. Flowering season of the food source vegetation was considered as an additional risk factor. These models have successfully identified the equine population ‘at risk’. The risk percentage of a probable outbreak event varies for each equine property depending on their exact location and their contributing factors. The prediction model(s) is an effective tool to identify the potential population (both equine and human) ‘at risk’, which can assist with Health Service Planning, policy implications, decision making and ongoing disease surveillance. This research successfully established the correlation between the HeV outbreak events, flying fox species and their roosting sites, food source vegetation and seasons spatially. The factors influencing the dispersal of HeV outbreak events in the study area were understood. This study reveals the capability of GIS-based surveillance system to issue early warnings and precautionary measures to the identified population ‘at risk’. This research also makes evidence based practice of disease mitigation, planning and prevention and control strategies for HeV achievable.