| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
 |
ABSTRACT |
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
|
INTRODUCTION |
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
The relationships between environmental factors and vector-borne disease have received much attention, and RRV disease is no exception. Various investigators have reported positive correlations between RRV disease incidence and increased rainfall, temperature, humidity, and sea level, while others have reported associations between RRV disease and the Southern Oscillation Index and Quasi-Biennial Oscillation.4–10 One of the problems with assessing the relationship between environmental factors and disease incidence across large geographic regions, or comparing between regions, is the assumption that underlying factors such as mosquito species present, human reporting behavior, and lifestyle activities are the same across the region, or that their effect is negligible. Such analysis also assumes that the relationship between environmental factors and RRV incidence is constant across the region being considered. An analysis of RRV disease in Queensland, Australia indicated that rainfall and high tidal level were significant factors in the transmission of RRV.5 However, this analysis is biased towards the southeast corner of the state where most of the population reside, but has a considerably lower RRV disease incidence than other regions of Queensland.11 Whelan and others highlight the differences in the disease-environment relationship that can occur over relatively short distances by reporting that summer rainfall exceeding 100 mm was a good predictor of increased RRV risk in Alice Springs, but in Tennant Creek (500 km north) increased RRV risk was associated with summer rainfall > 400 mm.6 More importantly, summer rainfall in Alice Springs was a very good indicator of another mosquito-borne disease, Murray Valley encephalitis (MVE), but in Tennant Creek MVE seroconversion was independent of summer rainfall.6
One way to overcome the assumption of uniformity within and between regions is to assess the relationship between local anomalies in disease incidence and environmental factors. This method assumes that the underlying vector population and human behavior patterns within a region are relatively constant over time, allowing the environmental factors causing disease outbreaks to be more easily identified. Although environmental factors that generally affect the vector population appear to be associated with increases in disease incidence, the specific factors important in outbreaks of mosquito-borne disease appear to differ between diseases, and also between regions. Increased rainfall, runoff, and temperature have been linked to epidemics of western equine encephalomyelitis, while increased epidemic risk of St. Louis encephalitis is associated with low rainfall and runoff and high temperature.12 Both of these diseases are maintained in an enzootic cycle involving Culex mosquitoes. For malaria, increased minimum temperatures in the months leading up to the transmission season accounted for much of the variance when abnormally high numbers of cases were detected in Madagascar.13
Outbreaks of RRV disease in various regions of Australia have been linked to increased rainfall, the Southern Oscillation index in January and September of the previous year, and early spring rainfall and spring/summer temperature.7,9,14 Several investigators have highlighted the importance of different factors in different geoclimatic regions.7,14 Although most of the RRV disease cases in Australia occur in more tropical areas, little has been published on the relationship between RRV disease outbreaks and environmental factors in northern Australia.
We have previously developed a robust method to detect RRV disease outbreaks, and applied this to 10 years of notification data for Queensland.11 This methodology provides the platform required to investigate factors that may be associated with disease outbreaks. In the current study, we investigated the relationship between localized RRV disease outbreaks in Queensland and local weather patterns. Specifically, we have concentrated on whether anomalies in local weather variables are associated with outbreak occurrence.
|
METHODS |
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
Ross River virus disease notification data was the same as that used previously.11 Briefly, the data consist of 25,135 RRV disease notifications that were reported to Queensland Health between June 1991 and May 2001. This notification data was geocoded using the 2001 Australian census boundaries. Patient data were then grouped by age, sex, and LGA of residence. Population estimates for each LGA were obtained from the Australian Bureau of Statistics.
Previously developed methods were used to identify outbreaks of RRV disease in each season within each of the 125 LGAs in Queensland across the 10-year study period (June 1991 to May 2001).11 In summary, the number of notifications received within a LGA in each season was compared with that expected based on a measure of the long-term incidence rate for the LGA and the corresponding population. Outbreaks were declared if there was a < 1% chance of obtaining the observed number based on a Poisson model. For each season of each year, every LGA was classified as 0 (no RRV disease outbreak) or 1 (RRV disease outbreak). Only those outbreaks that commenced in the season of interest were considered in this study; those that represented a continuation of an outbreak commencing in a previous season were excluded.
Data from Queensland weather stations with at least 30 years of records were obtained from the Australian Bureau of Meteorology.15 Data for the average minimum and maximum daily temperatures, the number of days with temperatures > 35°C, total rainfall, and number of days with rain were extracted for each month of every recording year. Seasonal values were calculated by averaging (for minimum and maximum temperature) or adding (for rainfall and counts of temperature and rainfall occurrences) the values for the months in each season.
For each weather station, the raw data for each month/season in the study period (1991–2001) were converted to percentiles based on all of the data recorded at the weather station for the variable of interest. Several statistics were calculated from this standardized data (Table 1
) and formed the basis of the analysis conducted.
|
Logistic regression was used to investigate the ability of climatic variables to predict RRV disease outbreaks in either summer (December to February) or autumn (March to May). Only LGAs that contained both outbreak and climatic data were included in the analysis. Due to the high correlation between some variables, the variance inflation factor (VIF) was used to assess the degree of multicollinearity.16 Only those variables that produced a VIF < 2.5 were included in the analysis (Table 1
).
As a first step, data from all LGAs with suitable climatic data were included in a logistic regression model. Subsequent analysis used smaller subsets of LGAs. Analysis of outbreaks in individual LGAs was not conducted since each LGA had a maximum of 10 data points for each season, and only a few (if any) of these were classified as RRV disease outbreaks. As a starting point, LGAs were grouped into subsets based on the results of a hierarchical cluster analysis of seasonal data for the total rainfall and average maximum and minimum daily temperatures for each LGA. A similar analysis was conducted using rainfall data only. This analysis grouped LGAs with similar weather patterns with no consideration of the RRV activity.
An iterative process was then used to determine the members of each LGA subset. First, a significant model was developed using forward stepwise logistic regression for the neighboring LGAs in the previously identified clusters. A cluster was only considered for regression analysis if there were at least three data points classified as RRV disease outbreaks. The significance/accuracy of this model was then monitored as new LGAs on the boundary of the region were added to the subset or existing boundary members removed. New LGAs were incorporated into the group if the accuracy of the model increased or remained fairly constant upon their inclusion. After the members of the group were finalized, forward stepwise regression was again conducted to ensure that the first model obtained was still the best. If this was not the case, the LGA selection process was repeated using the previously finalized subset as a starting point for the new model. In all cases, regression models were only reported if the overall accuracy of the model was greater than 90%, or if the sensitivity was greater than 60%. In all cases the HosmerLemeshow Goodness-of-fit statistic was > 0.05.17
All statistical analysis was conducted using SPSS for Windows Version 9.0.1 (SPSS Inc., Chicago, IL). The location and area of the LGAs, and location of weather stations was visualized using ArcView GIS Version 3.2a (Environmental Systems Research Institute Inc., Redlands, CA).
|
RESULTS |
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
).
|
).
When statistical analysis was conducted using all LGAs within Queensland that had appropriate weather data, no significant logistic regression model could be obtained using the independent variables listed in Table 1 for either summer or autumn. Thus, no single weather variable, or combination of variables, could accurately predict RRV outbreaks across the entire state.
Hierarchical cluster analysis using rainfall data for 81 LGAs produced similar groupings of LGAs compared with when both rainfall and minimum and maximum temperature data were used as the independent variables. A total of eight clusters were identified; however, some of these had insufficient numbers of RRV disease outbreaks to be analyzed further. The largest of these clusters contained 18 LGAs and was located in western Queensland, a region that had few RRV disease outbreaks during the study period (Figure 1). Similarly, another large cluster of 10 LGAs in far northern Queensland had insufficient numbers of outbreaks to be analyzed. As a result, the logistic regression analysis conducted on subsets of LGAs was restricted mainly to the eastern part of the state.
In contrast to the Queensland-wide analysis, significant regression models could be obtained when smaller subsets of LGAs were considered. Analysis of RRV disease outbreaks commencing in summer indicated that significant models could be developed for four subsets of LGAs (Regions 1–4; Figures 2 and 3 and Table 2). Although these four regions represent only 6.4% of the area of Queensland, they are located in the more populated regions, incorporating 80% of the Queensland population (Table 3). No regression models meeting our criteria for summer RRV disease outbreaks could be developed outside these four regions. Overall, the sensitivity of the models (the probability of predicting an outbreak when one actually occurred) varied from 0.60 to 0.83, while the specificity (the probability of correctly predicting a "no outbreak" year) was higher, ranging from 0.94 to 1.00.
|
|
|
|
and Table 2). These regions were larger than those developed for summer (Table 3); however, the sensitivity of the models was less (Table 2). In Region 5, both rainfall and temperature variables appeared important in predicting disease outbreaks, while in the more northern region (Region 6), only elevated temperature was predictive of disease outbreaks (Table 2 and Figure 4). In both regions, an additional variable indicating the relative level of RRV disease in the preceding summer was included in the initial stepwise regression analysis; however, this variable was not significant in either model.
|
), with Region 2 having substantially more rain over summer than the other regions (Figure 5). There was also an increasing trend for higher age-/sex-adjusted RRV disease incidence rates with decreasing latitude (Table 3).
|
). The models were generally better at predicting outbreaks during larger epidemics (when most of the LGAs within the region experienced an outbreak) compared with isolated outbreaks. However, each was able to correctly predict at least one outbreak during predominantly non-outbreak years.
|
|
DISCUSSION |
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
Several regions were identified within the state where strong relationships between climatic variables and RRV disease outbreaks were detected. For outbreaks commencing in summer, increased outbreak risk in the southern part of the state (Region 1) was associated with increased spring temperatures and early summer rainfall. In the central coastal region (Regions 2 and 3), rainfall was the most important predictor of RRV disease outbreaks. For the most northern region (Region 4), rainfall was still the most important factor; however, increased spring temperatures acted to decrease the outbreak risk. When outbreaks commencing in autumn were considered, southern areas of the state (Region 5) experienced an increased risk of RRV disease outbreaks with increased rainfall; however, this was mediated by occurrences of temperatures > 35°C that reduced outbreak risk. For the central coastal region (Region 6), increased temperature was the major risk factor for disease outbreaks. The different risk factors identified in the different regions possibly reflect the underlying climatic environments, and/or different vector populations and habitats.
High rainfall increases the availability of aquatic habitats potentially resulting in increased numbers of Culex annulirostris Skuse, Ochlerotatus vigilax (Skuse), Oc. procax (Skuse), and Oc. normanensis (Taylor), all of which are considered major RRV vectors in Queensland.2 Higher water temperatures generally result in shorter immature development times, although extreme temperatures (> 30°C) can increase mosquito mortality.18–20 Higher temperatures also increase the rate of evaporation of temporary habitats. In the southern regions of the state where winter and spring temperatures are considerably less than the northern regions, increased spring temperatures may act to stimulate early mosquito breeding. Since the vectorial capacity increases exponentially rather than linearly through the transmission season, a small change in the start of the transmission season could have noticeable effects on the resulting number of mosquitoes and subsequent level of disease.21
The climatic variables considered in this analysis were selected to reflect different temporal patterns of climatic activity. The seasonal summaries indicate a prolonged trend that may occur over several months, whereas minimum and maximum statistics can potentially detect smaller, more isolated events that might not have a large impact on the seasonal statistic. The inclusion of the number of days with temperature > 35°C provided information about extreme temperatures; however, there was no corresponding indicator for the frequency of cooler periods. As a result of the variables considered, we have identified climatic conditions related to unusually high RRV disease prevalence; however, it was not possible to attribute these outbreaks to specific isolated events.
Although the conceptual relationships between environmental conditions and mosquito numbers appear simple, an attempt to predict adult mosquito numbers using variables such as rainfall, tides, and temperature had limited success.22 The relationships between adult mosquito abundance and weather variables were complex, with no clear association between the variables and size of mosquito population. Superimposed on these environmental-vector interactions, vector competence is also effected by temperature, with increased competence at moderate temperatures (18–25°C) compared with a higher temperature (32°C).23 Due to the large differences in the underlaying climatic conditions across the state, and the complex relationships between environmental conditions, mosquito bionomics, and virus transmission dynamics, we were not surprised to find different relationships between environmental conditions and RRV disease outbreaks in different regions of the state.
Interestingly, we were unable to find any significant relationship between climatic variables and RRV disease outbreaks in the northern (tropical) part of Queensland. This may reflect the small size of the dataset, or alternatively indicate that RRV disease outbreaks in this region do not have a simple relationship with temperature and rainfall events. Further investigation of the factors associated with RRV disease outbreaks in this region is warranted.
Our results quantify the effect of climatic factors on RRV disease outbreaks in Queensland and generally reflect the qualitative results previously reported using RRV epidemics in Australia.7 Since five of the seven Queensland outbreaks previously analyzed commenced in January or February, comparison with our results should concentrate on summer outbreaks. Our results also support the importance of rainfall in RRV disease incidence in the Townsville region,8 although our results also suggest that minimum temperature had a role in outbreaks in this region (Region 4).
Tide height (or sea level) has been associated with RRV disease incidence and outbreaks.5,7,8 We did not consider tide height in our analysis since the highest spring tides occur regularly during summer and winter each year.24 By comparing disease incidence within the same season over time, these regular high tides need not be explicitly considered. Likewise, we did not consider large-scale climatic phenomena such as El Niño, the Southern Oscillation Index, or the Quasi-Biennial Oscillation since these affect rainfall and/or temperature that were already included in the model. Additionally, the large geographic areas affected by these phenomena do not suit small-scale regional analysis in which some areas experience an outbreak while others do not.
The analysis conducted indicates that for certain areas of the state, consideration of recent environmental conditions could help local authorities identify periods of increased RRV risk. This information, combined with an efficient early case detection system, will allow health authorities to provide timely warnings to residents when it is most warranted, and also indicate periods when additional mosquito control efforts may be justified.
In this study, we investigated the relationship between RRV disease outbreaks and climatic variables in Queensland and showed that rainfall and temperature variables were good predictors of RRV disease outbreaks for certain areas within the state. Importantly, no one set of climatic variables was predictive of RRV disease outbreaks across the entire state, highlighting the need for regional assessment.
Received July 1, 2004. Accepted for publication September 1, 2004.
Acknowledgment: We thank Craig Davies (Communicable Diseases Unit, Queensland Health) for providing the RRV disease notification data.
Financial support: Michelle L. Gatton was supported by a University of Queensland Postdoctoral Research Fellowship.
Authors’ address: Michelle L. Gatton, Brian H. Kay, and Peter A. Ryan, Australian Centre for International and Tropical Health and Nutrition, Queensland Institute of Medical Research, PO Royal Brisbane Hospital, Herston, Brisbane, Queensland 4029, Australia, Telephone: 61-7-3362-0416, Fax: 61-7-3362-0104, E-mail: michellG{at}qimr.edu.au, Telephone: 61-7-3362-0350, Fax: 61-7-3362-0106, E-mail: brianK{at}qimr.edu.au, Telephone: 61-7-3362-0351, Fax: 61-7-3362-0106, E-mail: peterR{at}qimr.edu.au.
|
REFERENCES |
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  Ying Zhang, Peng Bi, and J. E. Hiller Climate Change and the Transmission of Vector-Borne Diseases: A Review Asia Pac J Public Health, January 1, 2008; 20(1): 64 - 76. [Abstract] [PDF]
|
|
|
|
 |
|
 W. HU, S. TONG, K. MENGERSEN, and B. OLDENBURG EXPLORATORY SPATIAL ANALYSIS OF SOCIAL AND ENVIRONMENTAL FACTORS ASSOCIATED WITH THE INCIDENCE OF ROSS RIVER VIRUS IN BRISBANE, AUSTRALIA Am J Trop Med Hyg, May 1, 2007; 76(5): 814 - 819. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
= region 1; • = region 2; 
= region 3; 
= region 4. Lower panels: 
= region 5; 
= region 6.