Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Environmental and entomological factors determining Ross River virus activity in the River Murray Valley of South Australia

Environmental and entomological factors determining Ross River virus activity in the River Murray... Ross River (RR) virus causes the most common arthropod‐borne human infection in Australia. Symptoms of infection in humans typically comprise fever, arthralgia and rash, with sequelae sometimes including on‐going fatigue and joint pain. RR virus is maintained in enzootic cycles in non‐human populations by vector mosquitoes. Transmission is supported in a variety of ecological contexts, meaning that the vectors, zoonotic reservoirs and environmental drivers of outbreaks vary throughout Australia. An analysis of RR virus outbreaks in Australia from 1886 to 1998 confirmed that a variety of epidemiologies exist for this pathogen, meaning that generic tools for epidemic prediction are probably of limited use. Despite this variability, it was clear that above average rainfall preceding epidemics was a common theme in all climatic types. A relationship between RR virus activity and the Southern Oscillation Index (a partial predictor of rainfall patterns in Australia) has also been demonstrated. Descriptive models for RR virus activity have been developed for several regions. Such models are most effective when developed for specific regions, and their effectiveness decreases when they are applied elsewhere. Models based on climatic data have been developed previously for south‐eastern Australia with satisfactory sensitivity and specificity for retrospectively predicting epidemics. Further models were developed for south‐western Australia, and were enhanced by the inclusion of mosquito surveillance data. Differences in mosquito community structure along the River Murray Valley of SA exist between local government areas (LGAs) (SRF unpubl. data). This fact, along with previous demonstration of variation in RR virus activity at small spatial scales, led us to pursue analysis at the local government scale. Vector‐borne disease control in Australia is primarily managed by local governments, with the responsibility falling to environmental health officers employed in each LGA. Mosquito surveillance is also typically carried out by these same officers. Therefore, this spatial scale is the most relevant to those charged with evidence‐based management of vector‐borne disease. RR virus is endemic in SA. Since 1993, annual statewide incidence has ranged from 1.4 to 52.7 per 100,000, with epidemics recorded in 1993, 1997, 2000 and 2006. Certain regions of SA have a higher historical incidence of RR virus notification, especially townships along the River Murray, which extends from the border with New South Wales and Victoria in the east, to the river mouth in the south ( Figure 1 ). In the 2005/6 reporting period, the notification rates in the Murray Valley of SA were 157.1 per 100,000, the third highest regional rate nationally. 1 Local government areas along the River Murray in South Australia. Dashed line represents course of the river and extent of lakes near the mouth. Local Government Areas: RP: Renmark‐Paringa, LW: Loxton‐Waikerie, BB: Berri‐Barmera, MM: Mid‐Murray, MB: Murray Bridge, CO: Coorong. Beyond a general ‘rule of thumb’ that RR virus epidemics in SA usually occur every three to four years following years of above average rainfall, there are no formal indicators currently used to predict RR activity levels in this state. Despite its validity, health authorities in SA do not use the existing predictive model for south‐eastern Australia, possibly because of a lack of awareness of it and the broad spatial scale for which it was devised. The initial aim of this work was to determine whether a combination of environmental and mosquito abundance variables could be used to explain fluctuations in RR virus activity in the River Murray Valley of SA, and then to develop models at the LGA spatial scale based on these variables. It was hoped that in the course of developing models at this scale, greater understanding of epidemiology would result, enabling the development of novel, locally specific strategies for epidemic mitigation. Models developed at the LGA‐scale should have better resolution for guiding local government health interventions and therefore have greater utility. Moreover, such models should be meaningful for local governments in terms of input data and timeliness to ensure that they are used routinely and are effective. Methods The River Murray Valley of SA For the purposes of this study, this region comprises the townships, agricultural, forested and recreational lands that occur along the length of the river in SA. The region is divided into six LGAs ( Figure 1 ), with a total residential population of approximately 65,000 people. Our analyses were conducted for three LGAs along the River Murray Valley: Renmark‐Paringa, Mid‐Murray, Coorong ( Figure 1 ). These LGAs were chosen as they had sufficient notification numbers along with environmental and entomological data available to permit analysis. These LGAs are dispersed along the Valley, thus are likely to encompass the diversity of virus ecology present in the region. Ross River virus human notification data De‐identified notification data of RR virus positive serology were provided by the SA Department of Health for the period December 1999 to February 2006. This period was chosen because mosquito community data were available for the same period. The following information was presented for each record: Date of illness onset, date of notification, age in years, and town of residence. Crude incidence per 100,000 population was calculated for each LGA using tallied notifications for each season (summer, autumn, winter, spring) for the period December 1999 to February 2006 and the estimated residential population of each LGA for each year ( Table 1 ). 1 Crude Seasonal notification rates of RR virus notification in three LGAs in the River Murray Valley of SA, 1999‐2006. Total cases in parentheses. Season Renmark‐Paringa Mid‐Murray Coorong Summer 1999‐2000 192.6 (19) 60.1 (5) 0.0 Autumn 2000 212.9 (21) 0.0 0.0 Winter 2000 10.1 (1) 0.0 0.0 Spring 2000 20.3 (2) 0.0 184.4 (11) Summer 2000/01 10.2 (1) 118.9 (10) 84.5 (5) Autumn 2001 10.2 (1) 11.9 (1) 0.0 Winter 2001 0.0 0.0 0.0 Spring 2001 20.3 (2) 0.0 0.0 Summer 2001/02 0.0 11.9 (1) 16.9 (1) Autumn 2002 0.0 0.0 0.0 Winter 2002 10.1 (1) 0.0 0.0 Spring 2002 0.0 0.0 0.0 Summer 2002/03 10.3 (1) 0.0 0.0 Autumn 2003 0.0 0.0 0.0 Winter 2003 0.0 0.0 0.0 Spring 2003 0.0 11.9 (1) 0.0 Summer 2003/04 10.3 (1) 11.9 (1) 0.0 Autumn 2004 0.0 0.0 0.0 Winter 2004 0.0 0.0 0.0 Spring 2004 10.3 (1) 0.0 0.0 Summer 2004/05 10.3 (1) 0.0 0.0 Autumn 2005 10.3 (1) 11.9 (1) 17.4 (1) Winter 2005 10.3 (1) 11.9 (1) 17.4 (1) Spring 2005 51.3 (5) 0.0 87.0 (5) Summer 2005/06 153.8 (15) 71.3 (6) 696.1 (40) Median estimated residential population 9758 8415 5901 Direct standardised notification rate 51.7 18.6 68.3 For the purposes of these analyses, the source of suspected infection was assumed to be the place of residence. Data in the ‘source’ field in de‐identified records were frequently absent or non‐sensical (e.g. regions where RR virus is not known to occur), and were subject to ill‐defined self‐reporting bias by patients who offer recollections of where they can remember being bitten by mosquitoes. Thus, residential location data were considered more reliable, and most likely to be the source of infection. This approach will probably underestimate infection rates in the study areas. Mosquito community data These were obtained from the River Murray Mosquito Surveillance Project; a collaboration between the University of South Australia and LGAs in the River Murray Valley of SA. Adult female mosquito collections from carbon‐dioxide baited miniature light traps were made at five locations within each LGA from December 1999, every three weeks in each spring, summer and autumn). The locations were dispersed evenly throughout the LGAs and were positioned with close proximity to both human residences and known mosquito breeding grounds. A full description of mosquito community analysis in the River Murray Valley will be reported elsewhere. Meteorological data Daily rainfall and temperature records were obtained from the Australian Bureau of Meteorology (www.bom.gov.au), for recording stations within (or nearest to) each LGA: Renmark Aerodrome (Renmark‐Paringa), Murray Bridge (Mid‐Murray) and Meningie (Coorong). Rainfall variables were converted into seasonal rainfall relative to the historic mean (i.e. seasonal rainfall/historic mean). Mean daily minimum and maximum temperatures were calculated for each season. Determining significant factors in RR virus activity Separate analyses were conducted for each LGA. Crude seasonal RR virus notification rates were compared with mosquito abundance, rainfall and temperature variables ( Table 2 ). River height was included as a variable for the Mid‐Murray LGA as high river levels in this region are known to cause inundation of mosquito breeding sites (SRF pers. obs.). This phenomenon is not as apparent in Renmark‐Paringa where the river is heavily regulated by locks, and not in the Coorong LGA where the river terminates in lakes. Stepwise Multiple Regression (STATA 9.0) was used. Significant factors ( P < 0.05) were included in the models, which were in the form of equations: 2 Explanatory variables included in multiple regression modelling for three LGAs. Variable Description Unit Calculation method Tmax Mean max daily temp. °C Mean for season Tmin Mean min daily temp. °C Mean for season Ravg Seasonal rainfall relative to historical average ‐ Total rainfall divided by historical Mean for season RavgLAG3 Rainfall previous three months ‐ Rainfall previous three months RIHEmorg a River height at Morgan pump m Mean for season Note: a) Variable only included in analysis for Mid‐Murray LGA where N is the predicted number of notifications for the season, a, b 1…n are constants derived from the data, and x 1…n are variables specific for each prediction. Additional analysis (Stepwise Multiple Regression) was performed on combined notification data from all three LGAs to determine if a single descriptive model for the entire Murray Valley in SA could be created. Factors included in this model were climatic variables ( Table 2 ), three‐month lagged mosquito abundance variables, and LGA population size to account for any scale‐related differences in disease risk between locations. Only lagged mosquito abundance variables were included to determine whether such a model may have some use as a predictive tool. Results Stepwise multiple regression with entomological and environmental variables for each of the three LGAs resulted in expressions that explained a significant amount of variation (R 2 = 0.77–0.98) in RR virus incidence in the River Murray Valley ( Table 3 ). 3 Co‐efficients, CI 95 , P values and R 2 of explanatory variables for RR virus descriptive models in the Murray Valley of SA. LGA variable Coeff. CI 95 P R 2 Renmark‐Paringa CxglLAG3 52.99 (45.86, 60.14) <0.001 0.98 Cqli 2.99 (1.78, 4.22) 0.001 Ravg 28.45 (7.53, 49.37) 0.015 Tmin ‐11.17 (‐16.37, ‐5.98) 0.001 constant 70.84 (16.49, 125.19) 0.018 Mid‐Murray AecaLAG3 1.03 (0.77, 1.29) <0.001 0.98 RIHEmorg 177.14 (145.01, 209.27) <0.001 constant ‐578.57 (‐683.93, ‐473.21) <0.001 Coorong AecaLAG3 1.53 (0.95, 2.10) <0.001 0.77 Ravg 120.29 (4.15, 236.44) 0.043 constant ‐145.92 (‐261.72, ‐30.11) 0.017 Murray Valley (all LGAs) CxglLAG3 9.69 (4.54, 14.85) <0.001 AecaLAG3 0.591 (0.28, 0.90) <0.001 0.52 constant ‐8.12 (‐29.32, 13.08) 0.446 In Renmark‐Paringa RR activity could be described by a combination of rainfall, minimum temperature, Coquillettidia linealis (Cqli) abundance and time‐lagged Culex globocoxitus (Cxgl) abundance ( Table 3 ). In the mid‐Murray LGA, river height at Morgan and time‐lagged Aedes camptorhynchus (Aeca) abundance were significant factors. Time‐lagged Ae. camptorhynchus abundance and rainfall were significant factors in the Coorong LGA. Overall, RR virus activity throughout the Murray Valley in SA was described by a single model (R 2 = 0.52) in which time‐lagged Ae. camptorhynchus and Cx. globocoxitus abundance were the only significant factors. Residential population size was not a significant factor in the analysis ( p =0.51), indicating that no significant scale effects were contributing to RR virus activity along the Valley. Discussion Models that explain a significant amount of variation in RR notification rates in three LGAs in the River Murray Valley of SA since 1999 have been developed. These models are regionally specific for the climatic and mosquito community variation that exists along the Valley in SA and identify significant environmental and entomological factors associated with RR virus activity. Furthermore, the creation of these models demonstrates that a combination of environmental and entomological factors can be used to describe variation in RR virus activity. Other studies have demonstrated that the addition of mosquito abundance data can be used to create models with greater descriptive power than those using climatic data alone. There are similarities between our findings and previously published RR virus models. Firstly, the geographic variation in environmental factors described here is consistent with the findings from an analysis of outbreaks in 90 LGAs in Queensland. In that study, different predictive models were developed for four regions of grouped LGAs. Secondly, the significant factors identified here are similar to those identified in other regions. Most notably, increased rainfall was associated with increased probability of disease outbreaks in Queensland. Positive associations with seasonal rainfall and time‐lagged mosquito abundance in Brisbane Qld, and in south‐western WA have also been previously reported. Above average rainfall has been consistently associated with RR virus activity. However, rainfall was not a significant factor determining RR virus activity in the mid‐Murray LGA, where river height was seemingly more important. River Murray levels are only partially influenced by local and catchment‐wide rainfall. The river is heavily regulated through a series of locks, and through irrigation and so‐called anthropogenic ‘environmental flows’ into the floodplain. In these ways there is significant human influence on river height meaning that future management of this regulated river will likely have some impact on RR virus activity. Elevated river height has been previously associated with increased RR activity in Victoria. While not demonstrating causative relationships, several mosquito species were significantly associated with increased virus activity. In the Renmark‐Paringa LGA, Cq. linealis , known to be an efficient vector of RR virus, was found to be a significant factor. Conversely, time‐lagged Cx. globocoxitus abundance was also found to be a significant predictor of activity, despite not being considered a competent vector of RR virus. It is possible that this species acts as an indicator for activity of other mosquitoes of public health significance. In the mid‐Murray and Coorong models, Ae. camptorhynchus abundance was found to be a significant factor. RR virus has been detected in numerous field collections of this species, which is a competent vector, and is suggestive of a key role in transmission in the Middle and Lower parts of the River Murray in SA. The creation of a model for overall RR virus activity throughout the Valley was successful. The overall model included just two factors, time‐lagged Ae. camptorhynchus and Cx. globocoxitus abundance. Therefore, such a model may have some use as an RR activity predictive tool for the Valley generally, provided it is properly validated. However, its reduced descriptive ability (R 2 = 0.52) compared with models for individual LGAs (R 2 = 0.77–0.98) highlights the regional differences in RR virus ecology that exist within the Valley in SA. There are limitations to these models in that potential changes in the immune status of the human population, and in the ecology of vertebrate virus reservoirs such as macropod marsupials have not been taken into account. The specificity of the models for particular LGAs means they are limited in geographic applicability. In this study, we demonstrated that local rainfall is an important factor in determining RR activity in two LGAs. However, rainfall was not a significant factor in the mid‐Murray LGA, whereas river level was. Variability in the important mosquito species in different LGAs was also demonstrated. While we have not provided evidence of a causal relationship between particular mosquito species and RR virus transmission, we have described plausible differences in local RR virus ecology. The development of these descriptive models for RR activity in specific regions of the River Murray valley has potential public health management implications. By demonstrating different descriptive models for different areas of the Valley we have determined that regional variability in the determinants of RR virus activity is likely. Through awareness of these regionally‐specific factors, local health officers may now be better informed in determining seasonal risk of RR activity, and in devising appropriate management strategies. These may include the targeted control of particular mosquito species, and timing of control efforts in response to higher rainfall or river levels. The descriptive models presented here may provide the basis for RR virus activity forecasting through the collection of mosquito surveillance and meteorological data. Conversion of these descriptive models into epidemic forecasting tools will require robust definition of epidemic activity using established methods and continued investment in rigorous mosquito surveillance. Furthermore, rainfall forecasts would be required for the Renmark‐Paringa and Coorong model, meaning that the accuracy of such long range forecasting (provided by the Australian Bureau of Meteorology) would need to be evaluated. The applicability of the three models described here to neighbouring River Murray LGAs also requires evaluation. For any predictive models to become part of effective warning systems, they must provide reliable forecasts with clear parameters for triggering response activities. The forecasts should also be timely, and personnel should be sufficiently resourced, with the capacity and political will to undertake responses. Heat health warning systems with such qualities are used in Europe, with at least some effectiveness demonstrated. In the case of RR virus activity, local governments in partnership with State governments bear responsibility for any interventions. The LGA‐scale models provided here can provide the foundation for an effective RR virus early warning system for deployment at the local level, but only once criteria for action, lines of responsibility and required response resources have been determined. Acknowledgements Funding was provided by the SA Department of Health (Environmental Health Branch), and A. Koehler (Communicable Diseases Branch) provided historical arbovirus notification data. Renmark‐Paringa, Mid‐Murray, and Coorong local governments provided cooperation and support. J. Wakefield, M. Stephenson and A. Snell are thanked for their timely help. The Australian Bureau of Meteorology, the Australian Bureau of Statistics and the SA Department of Water, Land and Biodiversity Conservation all provided data without charge. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Australian and New Zealand Journal of Public Health Wiley

Environmental and entomological factors determining Ross River virus activity in the River Murray Valley of South Australia

Download PDF
5 pages

Loading next page...
 
/lp/wiley/environmental-and-entomological-factors-determining-ross-river-virus-LqnmIyvWah

References (30)

Publisher
Wiley
Copyright
© 2009 The Authors. Journal Compilation © 2009 Public Health Association of Australia
ISSN
1326-0200
eISSN
1753-6405
DOI
10.1111/j.1753-6405.2009.00390.x
pmid
19630851
Publisher site
See Article on Publisher Site

Abstract

Ross River (RR) virus causes the most common arthropod‐borne human infection in Australia. Symptoms of infection in humans typically comprise fever, arthralgia and rash, with sequelae sometimes including on‐going fatigue and joint pain. RR virus is maintained in enzootic cycles in non‐human populations by vector mosquitoes. Transmission is supported in a variety of ecological contexts, meaning that the vectors, zoonotic reservoirs and environmental drivers of outbreaks vary throughout Australia. An analysis of RR virus outbreaks in Australia from 1886 to 1998 confirmed that a variety of epidemiologies exist for this pathogen, meaning that generic tools for epidemic prediction are probably of limited use. Despite this variability, it was clear that above average rainfall preceding epidemics was a common theme in all climatic types. A relationship between RR virus activity and the Southern Oscillation Index (a partial predictor of rainfall patterns in Australia) has also been demonstrated. Descriptive models for RR virus activity have been developed for several regions. Such models are most effective when developed for specific regions, and their effectiveness decreases when they are applied elsewhere. Models based on climatic data have been developed previously for south‐eastern Australia with satisfactory sensitivity and specificity for retrospectively predicting epidemics. Further models were developed for south‐western Australia, and were enhanced by the inclusion of mosquito surveillance data. Differences in mosquito community structure along the River Murray Valley of SA exist between local government areas (LGAs) (SRF unpubl. data). This fact, along with previous demonstration of variation in RR virus activity at small spatial scales, led us to pursue analysis at the local government scale. Vector‐borne disease control in Australia is primarily managed by local governments, with the responsibility falling to environmental health officers employed in each LGA. Mosquito surveillance is also typically carried out by these same officers. Therefore, this spatial scale is the most relevant to those charged with evidence‐based management of vector‐borne disease. RR virus is endemic in SA. Since 1993, annual statewide incidence has ranged from 1.4 to 52.7 per 100,000, with epidemics recorded in 1993, 1997, 2000 and 2006. Certain regions of SA have a higher historical incidence of RR virus notification, especially townships along the River Murray, which extends from the border with New South Wales and Victoria in the east, to the river mouth in the south ( Figure 1 ). In the 2005/6 reporting period, the notification rates in the Murray Valley of SA were 157.1 per 100,000, the third highest regional rate nationally. 1 Local government areas along the River Murray in South Australia. Dashed line represents course of the river and extent of lakes near the mouth. Local Government Areas: RP: Renmark‐Paringa, LW: Loxton‐Waikerie, BB: Berri‐Barmera, MM: Mid‐Murray, MB: Murray Bridge, CO: Coorong. Beyond a general ‘rule of thumb’ that RR virus epidemics in SA usually occur every three to four years following years of above average rainfall, there are no formal indicators currently used to predict RR activity levels in this state. Despite its validity, health authorities in SA do not use the existing predictive model for south‐eastern Australia, possibly because of a lack of awareness of it and the broad spatial scale for which it was devised. The initial aim of this work was to determine whether a combination of environmental and mosquito abundance variables could be used to explain fluctuations in RR virus activity in the River Murray Valley of SA, and then to develop models at the LGA spatial scale based on these variables. It was hoped that in the course of developing models at this scale, greater understanding of epidemiology would result, enabling the development of novel, locally specific strategies for epidemic mitigation. Models developed at the LGA‐scale should have better resolution for guiding local government health interventions and therefore have greater utility. Moreover, such models should be meaningful for local governments in terms of input data and timeliness to ensure that they are used routinely and are effective. Methods The River Murray Valley of SA For the purposes of this study, this region comprises the townships, agricultural, forested and recreational lands that occur along the length of the river in SA. The region is divided into six LGAs ( Figure 1 ), with a total residential population of approximately 65,000 people. Our analyses were conducted for three LGAs along the River Murray Valley: Renmark‐Paringa, Mid‐Murray, Coorong ( Figure 1 ). These LGAs were chosen as they had sufficient notification numbers along with environmental and entomological data available to permit analysis. These LGAs are dispersed along the Valley, thus are likely to encompass the diversity of virus ecology present in the region. Ross River virus human notification data De‐identified notification data of RR virus positive serology were provided by the SA Department of Health for the period December 1999 to February 2006. This period was chosen because mosquito community data were available for the same period. The following information was presented for each record: Date of illness onset, date of notification, age in years, and town of residence. Crude incidence per 100,000 population was calculated for each LGA using tallied notifications for each season (summer, autumn, winter, spring) for the period December 1999 to February 2006 and the estimated residential population of each LGA for each year ( Table 1 ). 1 Crude Seasonal notification rates of RR virus notification in three LGAs in the River Murray Valley of SA, 1999‐2006. Total cases in parentheses. Season Renmark‐Paringa Mid‐Murray Coorong Summer 1999‐2000 192.6 (19) 60.1 (5) 0.0 Autumn 2000 212.9 (21) 0.0 0.0 Winter 2000 10.1 (1) 0.0 0.0 Spring 2000 20.3 (2) 0.0 184.4 (11) Summer 2000/01 10.2 (1) 118.9 (10) 84.5 (5) Autumn 2001 10.2 (1) 11.9 (1) 0.0 Winter 2001 0.0 0.0 0.0 Spring 2001 20.3 (2) 0.0 0.0 Summer 2001/02 0.0 11.9 (1) 16.9 (1) Autumn 2002 0.0 0.0 0.0 Winter 2002 10.1 (1) 0.0 0.0 Spring 2002 0.0 0.0 0.0 Summer 2002/03 10.3 (1) 0.0 0.0 Autumn 2003 0.0 0.0 0.0 Winter 2003 0.0 0.0 0.0 Spring 2003 0.0 11.9 (1) 0.0 Summer 2003/04 10.3 (1) 11.9 (1) 0.0 Autumn 2004 0.0 0.0 0.0 Winter 2004 0.0 0.0 0.0 Spring 2004 10.3 (1) 0.0 0.0 Summer 2004/05 10.3 (1) 0.0 0.0 Autumn 2005 10.3 (1) 11.9 (1) 17.4 (1) Winter 2005 10.3 (1) 11.9 (1) 17.4 (1) Spring 2005 51.3 (5) 0.0 87.0 (5) Summer 2005/06 153.8 (15) 71.3 (6) 696.1 (40) Median estimated residential population 9758 8415 5901 Direct standardised notification rate 51.7 18.6 68.3 For the purposes of these analyses, the source of suspected infection was assumed to be the place of residence. Data in the ‘source’ field in de‐identified records were frequently absent or non‐sensical (e.g. regions where RR virus is not known to occur), and were subject to ill‐defined self‐reporting bias by patients who offer recollections of where they can remember being bitten by mosquitoes. Thus, residential location data were considered more reliable, and most likely to be the source of infection. This approach will probably underestimate infection rates in the study areas. Mosquito community data These were obtained from the River Murray Mosquito Surveillance Project; a collaboration between the University of South Australia and LGAs in the River Murray Valley of SA. Adult female mosquito collections from carbon‐dioxide baited miniature light traps were made at five locations within each LGA from December 1999, every three weeks in each spring, summer and autumn). The locations were dispersed evenly throughout the LGAs and were positioned with close proximity to both human residences and known mosquito breeding grounds. A full description of mosquito community analysis in the River Murray Valley will be reported elsewhere. Meteorological data Daily rainfall and temperature records were obtained from the Australian Bureau of Meteorology (www.bom.gov.au), for recording stations within (or nearest to) each LGA: Renmark Aerodrome (Renmark‐Paringa), Murray Bridge (Mid‐Murray) and Meningie (Coorong). Rainfall variables were converted into seasonal rainfall relative to the historic mean (i.e. seasonal rainfall/historic mean). Mean daily minimum and maximum temperatures were calculated for each season. Determining significant factors in RR virus activity Separate analyses were conducted for each LGA. Crude seasonal RR virus notification rates were compared with mosquito abundance, rainfall and temperature variables ( Table 2 ). River height was included as a variable for the Mid‐Murray LGA as high river levels in this region are known to cause inundation of mosquito breeding sites (SRF pers. obs.). This phenomenon is not as apparent in Renmark‐Paringa where the river is heavily regulated by locks, and not in the Coorong LGA where the river terminates in lakes. Stepwise Multiple Regression (STATA 9.0) was used. Significant factors ( P < 0.05) were included in the models, which were in the form of equations: 2 Explanatory variables included in multiple regression modelling for three LGAs. Variable Description Unit Calculation method Tmax Mean max daily temp. °C Mean for season Tmin Mean min daily temp. °C Mean for season Ravg Seasonal rainfall relative to historical average ‐ Total rainfall divided by historical Mean for season RavgLAG3 Rainfall previous three months ‐ Rainfall previous three months RIHEmorg a River height at Morgan pump m Mean for season Note: a) Variable only included in analysis for Mid‐Murray LGA where N is the predicted number of notifications for the season, a, b 1…n are constants derived from the data, and x 1…n are variables specific for each prediction. Additional analysis (Stepwise Multiple Regression) was performed on combined notification data from all three LGAs to determine if a single descriptive model for the entire Murray Valley in SA could be created. Factors included in this model were climatic variables ( Table 2 ), three‐month lagged mosquito abundance variables, and LGA population size to account for any scale‐related differences in disease risk between locations. Only lagged mosquito abundance variables were included to determine whether such a model may have some use as a predictive tool. Results Stepwise multiple regression with entomological and environmental variables for each of the three LGAs resulted in expressions that explained a significant amount of variation (R 2 = 0.77–0.98) in RR virus incidence in the River Murray Valley ( Table 3 ). 3 Co‐efficients, CI 95 , P values and R 2 of explanatory variables for RR virus descriptive models in the Murray Valley of SA. LGA variable Coeff. CI 95 P R 2 Renmark‐Paringa CxglLAG3 52.99 (45.86, 60.14) <0.001 0.98 Cqli 2.99 (1.78, 4.22) 0.001 Ravg 28.45 (7.53, 49.37) 0.015 Tmin ‐11.17 (‐16.37, ‐5.98) 0.001 constant 70.84 (16.49, 125.19) 0.018 Mid‐Murray AecaLAG3 1.03 (0.77, 1.29) <0.001 0.98 RIHEmorg 177.14 (145.01, 209.27) <0.001 constant ‐578.57 (‐683.93, ‐473.21) <0.001 Coorong AecaLAG3 1.53 (0.95, 2.10) <0.001 0.77 Ravg 120.29 (4.15, 236.44) 0.043 constant ‐145.92 (‐261.72, ‐30.11) 0.017 Murray Valley (all LGAs) CxglLAG3 9.69 (4.54, 14.85) <0.001 AecaLAG3 0.591 (0.28, 0.90) <0.001 0.52 constant ‐8.12 (‐29.32, 13.08) 0.446 In Renmark‐Paringa RR activity could be described by a combination of rainfall, minimum temperature, Coquillettidia linealis (Cqli) abundance and time‐lagged Culex globocoxitus (Cxgl) abundance ( Table 3 ). In the mid‐Murray LGA, river height at Morgan and time‐lagged Aedes camptorhynchus (Aeca) abundance were significant factors. Time‐lagged Ae. camptorhynchus abundance and rainfall were significant factors in the Coorong LGA. Overall, RR virus activity throughout the Murray Valley in SA was described by a single model (R 2 = 0.52) in which time‐lagged Ae. camptorhynchus and Cx. globocoxitus abundance were the only significant factors. Residential population size was not a significant factor in the analysis ( p =0.51), indicating that no significant scale effects were contributing to RR virus activity along the Valley. Discussion Models that explain a significant amount of variation in RR notification rates in three LGAs in the River Murray Valley of SA since 1999 have been developed. These models are regionally specific for the climatic and mosquito community variation that exists along the Valley in SA and identify significant environmental and entomological factors associated with RR virus activity. Furthermore, the creation of these models demonstrates that a combination of environmental and entomological factors can be used to describe variation in RR virus activity. Other studies have demonstrated that the addition of mosquito abundance data can be used to create models with greater descriptive power than those using climatic data alone. There are similarities between our findings and previously published RR virus models. Firstly, the geographic variation in environmental factors described here is consistent with the findings from an analysis of outbreaks in 90 LGAs in Queensland. In that study, different predictive models were developed for four regions of grouped LGAs. Secondly, the significant factors identified here are similar to those identified in other regions. Most notably, increased rainfall was associated with increased probability of disease outbreaks in Queensland. Positive associations with seasonal rainfall and time‐lagged mosquito abundance in Brisbane Qld, and in south‐western WA have also been previously reported. Above average rainfall has been consistently associated with RR virus activity. However, rainfall was not a significant factor determining RR virus activity in the mid‐Murray LGA, where river height was seemingly more important. River Murray levels are only partially influenced by local and catchment‐wide rainfall. The river is heavily regulated through a series of locks, and through irrigation and so‐called anthropogenic ‘environmental flows’ into the floodplain. In these ways there is significant human influence on river height meaning that future management of this regulated river will likely have some impact on RR virus activity. Elevated river height has been previously associated with increased RR activity in Victoria. While not demonstrating causative relationships, several mosquito species were significantly associated with increased virus activity. In the Renmark‐Paringa LGA, Cq. linealis , known to be an efficient vector of RR virus, was found to be a significant factor. Conversely, time‐lagged Cx. globocoxitus abundance was also found to be a significant predictor of activity, despite not being considered a competent vector of RR virus. It is possible that this species acts as an indicator for activity of other mosquitoes of public health significance. In the mid‐Murray and Coorong models, Ae. camptorhynchus abundance was found to be a significant factor. RR virus has been detected in numerous field collections of this species, which is a competent vector, and is suggestive of a key role in transmission in the Middle and Lower parts of the River Murray in SA. The creation of a model for overall RR virus activity throughout the Valley was successful. The overall model included just two factors, time‐lagged Ae. camptorhynchus and Cx. globocoxitus abundance. Therefore, such a model may have some use as an RR activity predictive tool for the Valley generally, provided it is properly validated. However, its reduced descriptive ability (R 2 = 0.52) compared with models for individual LGAs (R 2 = 0.77–0.98) highlights the regional differences in RR virus ecology that exist within the Valley in SA. There are limitations to these models in that potential changes in the immune status of the human population, and in the ecology of vertebrate virus reservoirs such as macropod marsupials have not been taken into account. The specificity of the models for particular LGAs means they are limited in geographic applicability. In this study, we demonstrated that local rainfall is an important factor in determining RR activity in two LGAs. However, rainfall was not a significant factor in the mid‐Murray LGA, whereas river level was. Variability in the important mosquito species in different LGAs was also demonstrated. While we have not provided evidence of a causal relationship between particular mosquito species and RR virus transmission, we have described plausible differences in local RR virus ecology. The development of these descriptive models for RR activity in specific regions of the River Murray valley has potential public health management implications. By demonstrating different descriptive models for different areas of the Valley we have determined that regional variability in the determinants of RR virus activity is likely. Through awareness of these regionally‐specific factors, local health officers may now be better informed in determining seasonal risk of RR activity, and in devising appropriate management strategies. These may include the targeted control of particular mosquito species, and timing of control efforts in response to higher rainfall or river levels. The descriptive models presented here may provide the basis for RR virus activity forecasting through the collection of mosquito surveillance and meteorological data. Conversion of these descriptive models into epidemic forecasting tools will require robust definition of epidemic activity using established methods and continued investment in rigorous mosquito surveillance. Furthermore, rainfall forecasts would be required for the Renmark‐Paringa and Coorong model, meaning that the accuracy of such long range forecasting (provided by the Australian Bureau of Meteorology) would need to be evaluated. The applicability of the three models described here to neighbouring River Murray LGAs also requires evaluation. For any predictive models to become part of effective warning systems, they must provide reliable forecasts with clear parameters for triggering response activities. The forecasts should also be timely, and personnel should be sufficiently resourced, with the capacity and political will to undertake responses. Heat health warning systems with such qualities are used in Europe, with at least some effectiveness demonstrated. In the case of RR virus activity, local governments in partnership with State governments bear responsibility for any interventions. The LGA‐scale models provided here can provide the foundation for an effective RR virus early warning system for deployment at the local level, but only once criteria for action, lines of responsibility and required response resources have been determined. Acknowledgements Funding was provided by the SA Department of Health (Environmental Health Branch), and A. Koehler (Communicable Diseases Branch) provided historical arbovirus notification data. Renmark‐Paringa, Mid‐Murray, and Coorong local governments provided cooperation and support. J. Wakefield, M. Stephenson and A. Snell are thanked for their timely help. The Australian Bureau of Meteorology, the Australian Bureau of Statistics and the SA Department of Water, Land and Biodiversity Conservation all provided data without charge.

Journal

Australian and New Zealand Journal of Public Health – Wiley

Published: Jun 1, 2009

There are no references for this article.