Ross River Virus Disease Clusters and Spatial Relationship with Mosquito Biting Exposure in Redland Shire, Southern Queensland, Australia
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
MODELING/GIS, RISK ASSESSMENT, ECONOMIC IMPACT Ross River Virus Disease Clusters and Spatial Relationship with Mosquito Biting Exposure in Redland Shire, Southern Queensland, Australia P. A. RYAN,1 D. ALSEMGEEST,2 M. L. GATTON,1, 3 AND B. H. KAY1 J. Med. Entomol. 43(5): 1042Ð1059 (2006) Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 ABSTRACT The spatial heterogeneity in the risk of Ross River virus (family Togaviridae, genus Alphavirus, RRV) disease, the most common mosquito-borne disease in Australia, was examined in Redland Shire in southern Queensland, Australia. Disease cases, complaints from residents of intense mosquito biting exposure, and human population data were mapped using a geographic information system. Surface maps of RRV disease age-sex standardized morbidity ratios and mosquito biting complaint morbidity ratios were created. To determine whether there was signiÞcant spatial variation in disease and complaint patterns, a spatial scan analysis method was used to test whether the number of cases and complaints was distributed according to underlying population at risk. Several noncon- tiguous areas in proximity to productive saline water habitats of Aedes vigilax (Skuse), a recognized vector of RRV, had higher than expected numbers of RRV disease cases and complaints. Disease rates in human populations in areas which had high numbers of adult Ae. vigilax in carbon dioxide- and octenol-baited light traps were up to 2.9 times those in areas that rarely had high numbers of mosquitoes. It was estimated that targeted control of adult Ae. vigilax in these high-risk areas could potentially reduce the RRV disease incidence by an average of 13.6%. Spatial correlation was found between RRV disease risk and complaints from residents of mosquito biting. Based on historical patterns of RRV transmission throughout Redland Shire and estimated future human population growth in areas with higher than average RRV disease incidence, it was estimated that RRV incidence rates will increase by 8% between 2001 and 2021. The use of arbitrary administrative areas that ranged in size from 4.6 to 318.3 km2, has the potential to mask any small scale heterogeneity in disease patterns. With the availability of georeferenced data sets and high-resolution imagery, it is becoming more feasible to undertake spatial analyses at relatively small scales. KEY WORDS mosquito, Ross River virus, spatial statistics, SaTScan, vector-borne disease Vector populations are spatially heterogeneous in Thompson et al. 1997, Carter et al. 2000, Brooker et al. their densities, and this heterogeneity, together with 2004). In Maputo, Mozambique, malaria prevalence temporal changes in abundance, constitute important was 6.2 times higher for individuals living within 200 m elements in insect population dynamics (Liebhold et of anopheline mosquito breeding sites than for indi- al. 1993, Nestel and Klein 1995, Papadopoulos et al. viduals living ⬎500 m away (Thompson et al. 1997). 2003, Ryan et al. 2004). Tools such as global positioning Malaria transmission, at least in Africa, is focused systems, geographic information systems (GIS), geosta- around speciÞc mosquito breeding sites and can nor- tistics, and remote sensing have been used to investigate mally occur only within a few hundred meters to 1 km the spatial determinants of insect distributions, and the from the breeding sites, and it rarely in excess of 2Ð3 relationships between vector abundance and disease in- km (Carter et al. 2000). It has been proposed that cidence. If used soundly, these tools have the potential accurately targeted interventions to reduce transmis- to identify high-risk areas and allow control measures to sion rates can be expected to give greatly improved be targeted for maximum effect. levels of malaria control compared with untargeted Small-scale variations in the density of anopheline strategies (Carter et al. 2000). vectors have been shown to be important determi- Small-scale heterogeneity in dengue vector abun- nants of malaria transmission risk (Trape et al. 1992, dance and disease transmission has been examined (Waterman et al. 1985, Morrison et al. 1998, Russell et 1 Queensland Institute of Medical Research and Australian Centre al. 2002, Ali et al. 2003, Getis et al. 2003, Morrison et al. for International and Tropical Health and Nutrition, P.O. Royal Bris- 2004, Van Benthem et al. 2005), as has the landscape bane Hospital, Brisbane QLD 4029, Australia. 2 Redland Shire Council, P.O. Box 21, Cleveland 4163, Australia. ecology of western equine encephalomyelitis and St. 3 School of Population Health, University of Queensland, Herston Louis encephalitis viruses in California (Reisen et al. Rd., Herston QLD 4006, Australia. 1995a,b; Lothrop and Reisen 1999) and spatial deter- 0022-2585/06/1042Ð1059$04.00/0 䉷 2006 Entomological Society of America
September 2006 RYAN ET AL.: SPATIAL ANALYSIS OF ROSS RIVER VIRUS DISEASE CASES 1043 minants of West Nile virus transmission (Brownstein a total area of 539 km2, and the majority (60%) of this et al. 2002, Mostashari et al. 2003, Ruiz et al. 2004). area is made up of the Southern Moreton Bay Islands. In Australia, Ross River virus (family Togaviridae, Ninety-Þve percent of the 114,486 residents live on the genus Alphavirus, RRV) disease is the most common mainland, with the remaining population distributed mosquito-borne disease with ⬇5,000 human cases re- between the Southern Moreton Bay Islands of North ported annually (Russell 2002). Infection is not life Stradbroke, Coochiemudlo, Macleay, Lamb, Karra- threatening; however, the debilitation associated with garra, and Russell (Australian Bureau of Statistics clinical disease (polyarthritis, fever, and rash) is of 2003). considerable social and economic concern for local Redland Shire has a warm and humid subtropical communities. Although RRV has been isolated from 42 climate, characterized by mean daily maximum sum- mosquito species, with 10 able to transmit under lab- mer (DecemberÐJanuary) and winter (JuneÐAugust) oratory conditions, only a relatively small number of temperatures of 28 Ð29 and 20 Ð21⬚C, respectively species are regarded as being primary vectors in dif- (Commonwealth Bureau of Meteorology 2004). The Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 ferent regional environments in Australia (Russell area has a mean annual rainfall total of 1284 mm, with 2002). In coastal areas of Queensland, Aedes vigilax 74% of this rainfall occurring between November and (Skuse) is considered to be the major vector (Kay and May (Commonwealth Bureau of Meteorology 2004). Aaskov 1989). The freshwater species Culex annuliros- RRV Notifications. NotiÞcations of RRV disease tris Skuse is the principal inland vector, but a number from Redland Shire residents for the period June 1991 of other species, including Aedes notoscriptus (Skuse), to May 2001 were obtained from Queensland Health. Verrallina funerea (Theobald), and Aedes procax For each case, data were obtained for date of disease (Skuse), have been shown to be involved in virus onset along with the patientÕs sex, age, and approxi- transmission during epidemics in urban areas (Ritchie mate residential address (suburb, street, and house et al. 1997). The natural vertebrate hosts are assumed number ⫾ 6). The masked address was then entered to be macropodids (e.g., kangaroos and wallabies) into a GIS (ArcView GIS 3.2a; ESRI 1996a) maintained (Doherty 1972, Kay et al. 1986), although pteropodid by Redland Shire Council. Five hundred and seventy- fruit bats, horses, brushtail possums, and humans have Þve notiÞcations of RRV disease were provided of been implicated (Rosen et al. 1981, Kay and Aaskov which 522 (91%) cases were able to be mapped. The 1989, Ryan et al. 1997, Harley et al. 2000, Boyd et al. 53 cases that were unable to be mapped because of 2001). There is evidence that RRV survives in arid and missing data were generally distributed throughout semiarid areas by transovarial transmission (Russell the study area, with coverage rates in the 12 statistical 2002); however, transmission can occur throughout local areas (SLA) (Fig. 1) of 77Ð100%. the year in tropical and subtropical coastal areas in Cases were Þrst mapped to the level of SLA. Sta- Queensland. tistical local areas were deÞned by the Australian Bu- Our aim was to deÞne the spatial distribution of reau of Statistics (ABS 1991Ð2000) and represented RRV disease in Redland Shire in southern Queens- the smallest geographic areas for which estimated res- land. To determine whether there was signiÞcant vari- ident populations were available. The SLAs ranged in ation in the risk of RRV disease and exposure of hu- area from 4.6 km2 (Ormiston) to 318.3 km2 (southern mans to mosquito populations, we mapped RRV Moreton Bay Islands, Balance SLA), and had human disease notiÞcations and mosquito biting complaints, populations of between 3,496 (Thorneside) and 18,359 respectively, from residents in Redland Shire. Surface (Alexandra Hills) individuals (Australian Bureau of maps of RRV disease age-sex standardized morbidity Statistics 2003). To facilitate spatial analyses at a Þner ratios and mosquito biting complaint morbidity ratios resolution, we overlaid a 500- by 500-m grid (Fig. 1) were created, and the spatial scan analysis method was on the Redland Shire area and then mapped the RRV used to test whether the number of cases and com- disease cases to each of these grids. To estimate the plaints was distributed according to underlying pop- resident population in each of the 500- by 500-m ulation at risk. The spatial relationship between the square grids, we obtained the centroid coordinates of distributions of RRV disease cases, mosquito biting each occupied residential premises in Redland Shire, complaints, and the numbers of adult Ae. vigilax also each year between 1991Ð1992 and 2000 Ð2001 from a were examined. Based on these historical patterns of GIS (ArcView GIS 3.2a) maintained by Redland Shire RRV transmission, we then estimated the potential Council. The resident population in each SLA was impact of future population growth in Redland Shire then distributed across the respective 500- by 500-m on the numbers of RRV disease notiÞcations. grids in direct proportion to the number of occupied residential premises contained in each grid. Analyses were conducted using ArcView GIS 3.2a and ArcView Spatial Analyst extension (ESRI 1996b). Materials and Methods Spatial Analysis of RRV Disease Patterns. To deter- Study Area. Redland Shire (153⬚ 25⬘ E, 27⬚ 34⬘ S) is mine whether there were differences in the notiÞca- a local government administrative area located in the tion rates among the 12 SLAs, age-sex standardized southeast of Queensland. The area is bounded by morbidity ratios and conÞdence intervals were calcu- Brisbane to the west, Logan and Gold Coast cities to lated for each SLA by using the whole Redland Shire the south, and Moreton Bay and the PaciÞc Ocean to population as the reference group. The expected num- the north and east, respectively (Fig. 1). The Shire has ber of notiÞcations for each area was calculated from
1044 JOURNAL OF MEDICAL ENTOMOLOGY Vol. 43, no. 5 Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 Fig. 1. Location of Redland Shire (shaded and diagonal line Þlled areas) in southeastern Queensland, Australia. SLA boundaries are shown for the 11 SLAs located on the mainland, and the balance SLA comprising the southern Moreton Bay Islands. Populated areas were overlaid with a 500- by 500-m grid to facilitate SaTScan cluster analyses of RRV disease cases and mosquito biting complaints. age-sex-speciÞc incidence rates for the general pop- RRV disease patterns also were examined using the ulation applied to the particular age distribution in spatial scan statistic (Kulldorff and Nagarwalla 1995, each area. Ninety-Þve percent conÞdence limits (CL) Hjalmars et al. 1996). This method was used to scan the for the expected numbers of cases were calculated whole area for possible disease clusters and identify based on a Poisson distribution model. Standardized the approximate location of statistically signiÞcant morbidity ratios and 95% conÞdence intervals were clusters. Brießy, using the SaTScan software, version calculated by dividing the actual number of cases by 5.0 (http://satscan.org), a circular window was im- the expected number of cases, and the 95% CL for the posed on the map and its center was moved over the expected number of cases, in each area, respectively. area so that at any given position, the window included
September 2006 RYAN ET AL.: SPATIAL ANALYSIS OF ROSS RIVER VIRUS DISEASE CASES 1045 different sets of neighboring 500- by 500-m grids. Sep- ber of sampling occasions in which the estimated arate analyses were conducted using maximum-sized ranked value of each cell (100 by 100 m) exceeded the windows with radii of 0, 1,000, 2,000, and 3,000 m to 75th percentile. Of the three species examined, Co- determine whether the clustering of cases was related quillettidia linealis (Skuse), Cx. annulirostris, and Ae. to the spatial scale of the analyses. For each window vigilax, weekly distribution patterns were most stable size, the method created a circular window at each for Ae. vigilax, with a consistent pattern of relatively grid node, and then tested the null hypothesis against high (⬎75th percentile) numbers in the northern, the alternative hypothesis that there is an elevated risk central, and southern areas on the mainland, and on of RRV disease within, compared with outside the Macleay Island. Given the stability of Ae. vigilax spatial window. Using the SaTScan program, a likelihood ra- patterns over the two consecutive seasons, we com- tio test statistic was calculated for each window and its bined the data from 1999 Ð2000 and 2000 Ð2001 and distribution under the null hypothesis was obtained by created a probability map based on 15 wk of light trap repeating the calculation on 999 random replications sampling (Fig. 2). These analyses were limited to Ae. Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 of the data set generated under the null hypothesis. vigilax as the probability maps for other mosquito The P value for each cluster was obtained through species were not reliable (Ryan et al. 2004). The map Monte Carlo simulation, by comparing the rank of the was used to deÞne areas with consistently high (⬎75th likelihood test statistic from the real data set with the percentile) numbers of Ae. vigilax, and each cell was likelihood test statistics from the random data sets. The assigned a probability that mosquito numbers will be null hypothesis of no clusters was rejected at an ␣ level ranked in the top 25% of values. The probability values of 0.05 exactly, if the simulated P ⱕ 0.05 (Dwass 1957). were grouped into Þve categories: 0 Ð⬍20, 20 Ð⬍40, Mosquito Biting Complaints. Complaints from Red- 40 Ð⬍60, 60 Ð⬍80, and 80 Ð100%. Seven noncontiguous land Shire residents of mosquito biting were examined geographic areas were found to have ⱖ20% chance of as potential indicators of risk of mosquito borne dis- having high Ae. vigilax numbers. We then compared ease. Council records the period June 1991 to May the age-sex standardized RRV disease incidence rates 2001 were examined for complaints from residents of in the areas with 0 Ð⬍20% chance of having high Ae. mosquito biting activity. Each complaint record was vigilax numbers (i.e., 80 Ð100% chance of never having screened to ensure the complainant stated speciÞcally high vector numbers), to the RRV incidence in areas that there were high levels of adult mosquito biting. with 20 Ð⬍40, 40 Ð⬍60, 60 Ð⬍80, and 80 Ð100% chance Each validated complaint was geocoded according to of having high Ae. vigilax numbers. Analyses were the street address at which the reported mosquito limited to the mainland and Macleay Island areas of biting occurred. In total, 1,263 validated complaints Redland Shire as mosquito abundance data were not were obtained, and all of these complaints were able available for the remainder. To determine whether to be mapped to the approximate address (suburb, there was a consistent relationship in each of the seven street, and house number ⫾ 6) at which the reported geographic areas deÞned in Fig. 2, separate analyses mosquito biting occurred. To determine whether were undertaken for each area. Analyses also were there were differences in complaint rates from pop- undertaken to determine whether there was any as- ulations in different areas of Redland Shire, we ana- sociation between Ae. vigilax abundance and the num- lyzed the data as for the RRV disease notiÞcation data. bers of mosquito biting complaints. Spatial Relationship between RRV Disease and Impact of Future Population Growth on RRV Dis- Mosquito Biting Complaint Patterns. To determine ease Morbidity Patterns. Population projections for whether there was a spatial association between RRV 2001Ð2021 were obtained for each of the SLAs in disease incidence and mosquito biting complaints re- Redland Shire (Queensland Government 2005). To ceived from residents, we plotted the SaTScan results estimate the potential future RRV disease morbidity in for each of the four scanning windows. The RRV Redland Shire, we applied the long-term average in- disease standardized morbidity ratios SMR) and the cidence rates for each SLA between 1991 and 2001 to mosquito biting complaint morbidity ratios (MR) cal- estimated populations in each SLA in 2021. This was culated for each grid coordinate were plotted, and the based on the assumption that the age-sex proÞles of statistical results for each ratio, obtained from Monte each SLA remained constant and the long-term (1991Ð Carlo simulations, are presented. 2001) RRV disease incidence rates for each SLA re- To determine whether there was a spatial associa- ßected the pattern of disease in 2021. tion between areas with high numbers of RRV vectors and high RRV disease incidence rates, we used the adult mosquito abundance maps from Ryan et al. Results (2004) to deÞne areas of high vector abundance. Brießy, Ryan et al. (2004) used an array of 81 carbon The 10-yr RRV disease incidence rate in Redland dioxide and octenol-baited light traps to obtain weekly Shire was 52.1 cases per 100,000 person years (py), samples of adult mosquitoes from the mainland and with yearly rates of between 9.3 (1992Ð1993) and 121.4 Macleay Island areas of Redland Shire during 1999 Ð (1993Ð1994) cases per 100,000 py. The 10-yr mosquito 2000 (11 weekly samples) and 2000 Ð2001 (four weekly biting complaint rate was 126.1 complaints per 100,000 samples). Kriged mosquito counts from weekly col- py, with yearly rates ranging between 25.1 (1999 Ð lections were converted to percentile rank, and prob- 2000) and 387.5 (1993Ð1994) complaints per 100,000 ability maps were then calculated based on the num- py.
1046 JOURNAL OF MEDICAL ENTOMOLOGY Vol. 43, no. 5 Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 Fig. 2. Probability that kriged adult Ae. vigilax numbers throughout mainland and Macleay Island areas of Redland Shire exceeded the 75th percentile each week, during 2000 (11 wk of sampling) and 2001 (4 wk of sampling) (data from Ryan et al. 2004). Areas 1Ð7 refer to noncontiguous areas in which the probability of Ae. vigilax numbers exceeding the 75th percentile was ⱖ20%. Analysis of RRV Disease and Mosquito Complaint on the mainland (Victoria Point and Redland Bay), Patterns at the SLA Level. There were large variations and the Moreton Bay Islands (Balance), were found to in the RRV disease incidence rates among individual have signiÞcantly (P ⬍ 0.05) higher than expected SLAs (Table 1; Fig. 3), and these values ranged from numbers of cases (SMRs 1.50 Ð2.77), four had signiÞ- 29.6 (Thorneside) to 144.2 (Balance) cases per 100,000 cantly (P ⬍ 0.05) lower than expected numbers of py. Three LGAs comprising the southern coastal areas cases (Alexandra Hills, Birkdale, Capalaba and Cleve-
September 2006 Table 1. Age-sex standardized RRVdisease incidence rates and SMR, and mosquito biting complaint incidence rates and MR in Redland Shire, 1991–2001 RRV disease notiÞcations (June 1991ÐMay 2001) Mosquito biting complaints (June 1991ÐMay 2001) Pop Pop SLA Observed Expected Rate c 95% CId Observed Expected Rate (py)a SMR (py)a MRf (95% CI)d cases casesb (/100,000 py) complaints complaintse (/100,000 py) Alexandra Hills 166,632 54 81.7 34.5 0.66 (0.54Ð0.86) 166,632 17 210.2 10.2 0.08 (0.07Ð0.09) Balance 50,925 84 30.4 144.2 2.77 (2.00Ð4.42) 50,925 310 64.3 608.5 4.82 (3.88Ð6.46) Birkdale 115,062 37 59.3 32.5 0.62 (0.49Ð0.84) 115,062 68 145.1 59.1 0.47 (0.40Ð0.56) Capalaba 164,027 50 81.2 32.1 0.62 (0.51Ð0.79) 164,027 19 206.9 11.6 0.09 (0.08Ð0.11) Cleveland 111,101 41 61.1 35.0 0.67 (0.53Ð0.91) 111,101 115 140.2 103.5 0.82 (0.70Ð0.99) Ormiston 34,802 16 18.7 44.5 0.85 (0.57Ð1.60) 34,802 52 43.9 149.4 1.19 (0.91Ð1.68) Redland Bay 58,470 66 31.6 108.8 2.09 (1.54Ð3.30) 58,470 410 73.6 702.3 5.57 (4.51Ð7.32) Sheldon-Mt Cotton 34,051 27 17.8 79.3 1.52 (1.00Ð3.00) 34,051 21 42.9 61.7 0.49 (0.38Ð0.70) Thorneside 34,777 10 17.6 29.6 0.57 (0.37Ð1.11) 34,777 35 43.9 100.6 0.80 (0.61Ð1.13) Thornlands 73,306 30 38.8 40.4 0.77 (0.58Ð1.15) 73,306 19 92.4 25.9 0.21 (0.17Ð0.26) Victoria Point 93,223 75 50.1 78.0 1.50 (1.17Ð2.08) 93,223 94 117.6 100.8 0.80 (0.68Ð0.98) Wellington Point 64,947 32 33.7 49.5 0.95 (0.71Ð1.46) 64,947 103 81.9 158.6 1.26 (1.03Ð1.61) Total 1,001,323 522 52.1 1,001,323 1263 126.1 a Populations expressed as the number of py. b Expected number of cases calculated by applying age-sex speciÞc RRV disease incidence rates from the whole Redland Shire population, to the population in the statistical local area. c Standardized morbidity ratio calculated by dividing the observed number of cases by the expected number of cases. d Ninety-Þve percent conÞdent interval (CI) for the standardized morbidity ratio or morbidity ratio. e Expected number of complaints calculated by applying mosquito biting complaint incidence rates from the whole Redland Shire population, to the populations in the 12 SLA. f Morbidity ratio calculated by dividing the observed number of complaints by the expected number of complaints. RYAN ET AL.: SPATIAL ANALYSIS OF ROSS RIVER VIRUS DISEASE CASES 1047 Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021
1048 JOURNAL OF MEDICAL ENTOMOLOGY Vol. 43, no. 5 Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 Fig. 3. Relationship between annual mosquito biting complaint morbidity ratios (x-axis) and RRV disease age-sex standardized morbidity ratios (y-axis) in statistical local areas in Redland Shire between 1991 and 2001. land; SMRs 0.62Ð 0.67), and the remainder were not expected cases) for RRV disease for each of the grid signiÞcantly different (SMRs 0.57Ð1.52) from the squares ranged between 0 and 250; however, when overall Redland Shire rate. A similar pattern was found these were tested against the null hypothesis, i.e., the with mosquito biting complaints, with higher than observed counts in each grid were proportional to the expected numbers of complaints from residents in age-sex adjusted population at risk in each grid, only Balance and Redland Bay SLAs (MRs 4.80 Ð5.86) and eight clusters of signiÞcantly higher than expected signiÞcantly (P ⬍ 0.05) lower than expected numbers numbers of cases were found (Fig. 4). The number of in the remaining areas (MRs 0.08 Ð 0.65), except for observed cases in each of these eight grids ranged Wellington Point, Ormiston, Cleveland, and Victoria between 3 and 12, compared with the expected num- Point (MRs 0.70 Ð1.54) (Table 1; Fig. 3). bers of 0.02Ð2.57. The difference between the total The temporal relationship between RRV disease observed number of cases (49) and the total expected cases and mosquito biting complaints was examined at number (5.77) from these eight grids was 43.23 cases, the LGA level (Fig. 3). There was a clear positive which represented only 2.3% of the total number of association between annual RRV disease standardized RRV disease cases in Redland Shire. There did not morbidity ratios and mosquito biting complaint mor- seem to be any spatial relationship among these eight bidity ratios at the SLA level. disease clusters (Fig. 4). In comparison, when we Analysis of RRV Disease and Mosquito Complaint increased the scan window to 1,000, 2,000, and 3,000 m, Patterns by Using Spatial Scan Method. The number the number of overlapping clusters increased to 28, 60, and location of clusters of higher than expected num- and 95, respectively, and the total observed numbers bers of RRV disease cases was dependent on the size (95, 128, and 204, respectively) and expected numbers on the scan window (i.e., 0-, 1,000-, 2,000-, or 3,000-m (30.30, 49.89, and 102.90, respectively) of RRV cases radii) used in the SaTScan analyses. With a 0-m radius also increased. The difference between the total ex- scan window, the observed number of RRV cases in pected and total observed numbers of RRV cases in each of the 500- by 500-m grids ranged between 0 and these areas ranged from 64.7 to 101.1 and represented 12. Standardized morbidity ratios (observed cases/ 12.4 Ð19.4% of the total number of RRV disease cases
September 2006 RYAN ET AL.: SPATIAL ANALYSIS OF ROSS RIVER VIRUS DISEASE CASES 1049 Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 Fig. 4. Age-sex SMR for each of 809 grids (500 by 500 m) throughout Redland Shire, by using overlapping windows with radii of 0, 1,000, 2,000, and 3000 m. SMRs were calculated by dividing the observed number of RRV disease cases by the expected number RRV disease cases. Statistical analyses were completed using the SaTScan analysis package using scan windows of maximum sizes of 0, 1,000, 2,000, and 3,000 m. Statistical signiÞcance was evaluated with Monte Carlo simulation. Grids located within statistically signiÞcant clusters are shaded. in Redland Shire. At these larger scanning window using a scan window of radius ⬍1,000 m, and one of sizes (1,000 Ð3,000 m), there were consistent patterns these was signiÞcant at the remaining scan window of signiÞcantly higher than expected numbers of RRV sizes (0, 2,000, and 3,000 m). The latter cluster was disease cases in four areas: Victoria Point and Redland limited to a single 500- by 500-m grid in Sheldon-Mt Bay on the mainland; Southern Moreton Bay Islands of Cotton, which had three RRV disease cases (single Macleay, Lamb, Karragarra, Russell; and Dunwich on cases reported in 1991, 1994, and 1995) from a total North Stradbroke Island. Two statistically signiÞcant population of six people. Although SMRs for RRV clusters also were found in inland areas (Fig. 4) by disease were consistently ⬎1.0 in surrounding areas in
1050 JOURNAL OF MEDICAL ENTOMOLOGY Vol. 43, no. 5 Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 Fig. 4. Continued. Sheldon-Mt Cotton, these were not statistically sig- observed for RRV disease. For example, at a 0-m radius niÞcant. A similar trend was found on Coochiemudlo scan window, the observed number of mosquito biting Island and in northern areas in Wellington Point and complaints in each of the 500- by 500-m grids ranged Ormiston. between 0 and 58, and 35 clusters of signiÞcantly Geographic clustering of higher than expected higher than expected numbers of complaints were numbers of mosquito biting complaints also was found (Fig. 5). The number of observed complaints in found; however, the variation in the rate of mosquito each of these 35 grids ranged between 3 and 58, com- biting complaint per population was greater than that pared with the expected numbers of 0.05Ð 6.25. The
September 2006 RYAN ET AL.: SPATIAL ANALYSIS OF ROSS RIVER VIRUS DISEASE CASES 1051 Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 Fig. 5. Mosquito biting complaint MR for each of 809 grids (500 by 500 m) throughout Redland Shire, by using overlapping windows with radii of 0, 1,000, 2,000, and 3,000 m. Morbidity ratios were calculated by dividing the observed number of mosquito biting complaints by the expected number of complaints. Statistical analyses were completed using the SaTScan analysis package using scan windows of maximum sizes of 0, 1,000, 2,000, and 3,000 m. Statistical signiÞcance was evaluated with Monte Carlo simulation. Grids located within statistically signiÞcant clusters are shaded. difference between the total observed number of Shire. In comparison, when we increased the scan complaints (594) and the total expected number window to 1,000, 2,000, and 3,000 m, the number of (44.0) from these 35 grids was 550, which represented overlapping clusters increased to 141, 215, and 233, 43.5% of the total number of complaints in Redland respectively, and the observed numbers (845, 907, and
1052 JOURNAL OF MEDICAL ENTOMOLOGY Vol. 43, no. 5 Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 Fig. 5. Continued. 984, respectively) and expected numbers (164.72, quito biting complaints in Redland Shire. The majority 251.79, and 326.93, respectively) of complaints also of mosquito biting complaints (93.9%) came from increased. The difference between the total expected residents living within two km of the coast, and of and total observed numbers of mosquito biting com- these, most (60.3%) were from residents in coastal plaints in these areas ranged from 655.21 to 680.28 and areas in Redland Bay and the southern Moreton Bay represented 51.9 Ð53.9% of the total number of mos- Islands.
September 2006 RYAN ET AL.: SPATIAL ANALYSIS OF ROSS RIVER VIRUS DISEASE CASES 1053 Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 Fig. 6. Relationship between RRV disease age-sex SMR and mosquito biting complaint MR calculated for each of 809 grid nodes (500 by 500 m) throughout Redland Shire, by using overlapping buffer windows with radii of 0, ⬍1,000, ⬍2,000, and ⬍3,000 m. Each SMR and MR was examined to determine whether the observed number of RRV disease cases and number of complaints, respectively, was signiÞcantly greater than the expected numbers. Analyses were completed using SaTScan analysis package and statistical signiÞcance was evaluated with Monte Carlo simulation. Statistical signiÞcance was reported as white circles, SMR and MR not signiÞcantly ⬎1.0 (P ⬎ 0.05); gray circles, SMR signiÞcantly ⬎1.0 (P ⱕ 0.05) and MR not signiÞcantly ⬎1.0 (P ⬎ 0.05); gray triangles, SMR not signiÞcantly ⬎1.0 (P ⬎ 0.05) and MR signiÞcantly ⬎1.0 (P ⱕ 0.05); and black circles, SMR signiÞcantly ⬎1.0 (P ⱕ 0.05) and MR signiÞcantly ⬎1.0 (P ⱕ 0.05). Relationship between RRV Disease Incidence and ters of higher than expected numbers of mosquito Mosquito Biting Exposure at Different Spatial Scales. biting complaints (Figs. 4 and 5). To determine whether there was a spatial association The strength of the relationship between RRV in- between RRV disease incidence and mosquito biting cidence and mosquito biting exposure, and the sensi- complaints received from residents, we plotted the tivity of mosquito biting complaints as a marker of SaTScan results for each of the four scanning windows increased risk of RRV disease, was dependent on the (Fig. 6). The RRV disease SMR and the mosquito size of the SaTScan scanning window. For example, biting complaint MR calculated for each grid coordi- with a 0-m scanning window (i.e., cases and com- nate was plotted, and the statistical results for each plaints mapped to 500- by 500-m grids and SMRs and ratio, obtained from SaTScan Monte Carlo simula- MRs calculated for each grid separately), there were tions, are presented. There was a clear, positive rela- eight grids that had signiÞcantly higher than expected tionship between RRV disease incidence and mos- numbers of RRV disease cases (Fig. 6). Of these grids, quito biting exposure in humans, and a spatial only three had signiÞcantly higher than expected association between the location of clusters of higher numbers of mosquito biting complaints, and three than expected numbers of RRV disease cases and clus- grids had no reported mosquito biting complaints at
1054 JOURNAL OF MEDICAL ENTOMOLOGY Vol. 43, no. 5 Table 2. Age-sex standardized RRV disease incidence rates and SMR in mainland and Macleay Island areas of Redland Shire versus probability of the area having consistently high (>75th percentile) numbers of Ae. vigilax mosquitoes Probability (%) RRV rate Areaa Ae. vigilax no. Area (ha) Pop (py)c Cases Expd SMRe P valuef (/100,000 py) ⱖ75th percentileb Reference pop 0Ð⬍20 15,748 676,776 263 263.0 39.1 All areas 20Ð⬍40 2,374 134,244 61 52.1 45.7 1.2 ⬎0.05 40Ð⬍60 1,106 48,215 25 19.2 50.9 1.3 ⬎0.05 60Ð⬍80 1,687 56,249 49 22.5 85.3 2.2 ⬍0.001 80Ð100 3,997 46,349 56 19.0 115.3 2.9 ⬍0.001 Area 1 20Ð⬍40 895 60,059 27 22.9 46.2 1.2 ⬎0.05 40Ð⬍60 266 4,460 1 1.7 22.8 0.6 ⬎0.05 Area 2 20Ð⬍40 400 47,479 13 18.4 27.6 0.7 ⬎0.05 40Ð⬍60 274 21,699 4 8.6 18.3 0.5 ⬎0.05 60Ð⬍80 659 42,915 24 17.0 55.1 1.4 ⬎0.05 Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 80Ð100 362 23,493 18 9.3 76.0 1.9 ⬍0.01 Area 3 20Ð⬍40 33 756 0 ⬍0.1 0 0 ⬎0.05 Area 4 20Ð⬍40 343 9,226 7 3.7 73.3 1.9 ⬎0.05 40Ð⬍60 230 20,273 16 8.2 76.2 1.9 ⬍0.01 60Ð⬍80 255 1,627 3 0.7 179.4 4.6 ⬍0.05 80Ð100 310 5,730 13 2.3 219.8 5.6 ⬍0.001 Area 5 20Ð⬍40 118 3,236 2 1.3 59.7 1.5 ⬎0.05 Area 6 20Ð⬍40 585 13,489 12 5.5 85.0 2.2 ⬍0.01 40Ð⬍60 335 1,784 4 0.7 214.3 5.5 ⬍0.01 60Ð⬍80 773 11,707 22 4.8 179.7 4.6 ⬍0.001 80Ð100 2326 5,696 9 2.3 151.6 3.9 ⬍0.001 Area 7 80Ð100 999 11,431 16 5.1 122.7 3.1 ⬍0.001 a Areas deÞned in Fig. 2. b Based on data from Ryan et al. (2004), probability maps were used to deÞne areas with consistently high (⬎75th percentile) numbers of Ae. vigilax (Fig. 2). The probability values were grouped into Þve categories: 0 Ð⬍20, 20-⬍40, 40 Ð⬍60, 60-⬍80, and 80 Ð100%. Seven noncontiguous, geographic areas were found to have ⱖ20% chance of having high Ae. vigilax numbers. c Populations expressed as the number of py. d Expected numbers of cases calculated by applying age-sex speciÞc RRV disease incidence rates from the pop residing in the 0 Ð⬍20% area, to populations in the 20 Ð⬍40, 40 Ð⬍60, 60 Ð⬍80, and 80 Ð100% areas. e Standardized morbidity ratio calculated by dividing the actual number of cases by the expected number of cases. f Probability that the SMR is signiÞcantly greater than or ⬍1.0, based on the assumption that the observed cases followed a Poisson distribution with a mean equal to the expected number of cases. all. In comparison, when larger sized scanning win- chance of having high Ae. vigilax numbers were not dows were used, the sensitivity of mosquito biting associated with increased risk of RRV disease. Of the complaints as a marker for RRV disease risk increased. seven noncontiguous areas in Fig. 2, four had a positive For example, for a 2,000-m scanning window analysis, relationship between high Ae. vigilax abundance and 60 windows had signiÞcantly higher than expected RRV disease risk (areas 2, 4, 6, and 7). The three numbers of RRV disease cases. Of these windows, 43 remaining areas had ⬍60% chance of having high Ae. (72%) also had signiÞcantly higher than expected vigilax number (areas 1, 3, and 5) and relatively small numbers of mosquito biting complaints. This propor- human populations (areas three and 5). In areas in tion increased to 80% when a 3,000-m scanning win- Victoria Point (area 4) and Redland Bay (area 6) with dow was used. Based on 3,000-m scanning window ⬎60% chance of having high Ae. vigilax numbers, the results, we identiÞed 290 overlapping, circular areas RRV disease incidence rates were 3.9 Ð5.6 times the with mosquito biting complaint MRs of ⬎1.0. This disease rate (39.1 cases per 100,000 py) in the area that combined area contained 83% of the windows (95) never or rarely (0 Ð⬍20% chance) had high Ae. vigilax with signiÞcantly higher than expected numbers of numbers. RRV disease cases. High rates of mosquito biting complaints also were Spatial Relationships between RRV Disease–Mos- quito Biting Complaints and Ae. vigilax Adult Abun- associated with high Ae. vigilax adult abundance (Ta- dance. Increased risk of RRV disease in humans was ble 3). Overall, areas with 40 Ð⬍60% chance of high Ae. associated with high Ae. vigilax adult abundance (Ta- vigilax numbers had almost double the numbers of ble 2). Overall, based on the Þve probability categories complaints compared with the reference (0 Ð⬍20%) (0 Ð⬍20, 20 Ð⬍40, 40 Ð⬍60, 60 Ð⬍80, and 80 Ð100%) population. In areas with 60 Ð⬍80 and 80 Ð⬍100% that were used to deÞne areas with consistently high probabilities, the mosquito biting complaint MRs in- (ⱖ75th percentile) numbers of Ae. vigilax (Fig. 2), we creased to 11.8 and 12.8, respectively. The relationship found areas with a 60 Ð⬍80% and 80 Ð100% chance of between high Ae. vigilax numbers and complaints was having high Ae. vigilax numbers had 2.2 and 2.9 times consistent in all areas except for area 1 (Fig. 2; Table the risk of RRV disease, respectively, compared with 3). In areas in Victoria Point (area 4) and Redland Bay areas that had 0 Ð⬍20% chance of having high Ae. (area 6) with ⱖ60% chance of having high Ae. vigilax vigilax numbers. Areas with 20 Ð⬍40 and 40 Ð⬍60% numbers, the mosquito biting complaint incidence
September 2006 RYAN ET AL.: SPATIAL ANALYSIS OF ROSS RIVER VIRUS DISEASE CASES 1055 Table 3. Mosquito biting complaint incidence rates and MR in mainland and Macleay Island areas of Redland Shire, versus probability of the area having consistently high (>75th percentile) numbers of Ae. vigilax mosquitoes Probability (%) Area Pop Complaint rate Areaa Ae. vigilax numbers Complaints Expd MRe P valuef (ha) (py)c (/100,000 py) ⱖ75th percentileb Reference pop 0Ð⬍20 15,748 676,776 315 315 46.5 All areas 20Ð⬍40 2,374 134,244 80 62.5 59.6 1.3 ⬍0.05 40Ð⬍60 1,106 48,215 41 22.42 85.04 1.8 ⬍0.001 60Ð⬍80 1,687 56,249 310 26.8 551.1 11.8 ⬍0.001 80Ð100 3,997 46,349 277 12.8 597.6 12.8 ⬍0.001 Area 1 20Ð⬍40 895 60,059 35 28.0 58.3 1.3 ⬎0.05 40Ð⬍60 266 4460.0 1 2.1 22.4 0.5 ⬎0.05 Area 2 20Ð⬍40 400 47,479 22 22.1 46.3 1.0 ⬎0.05 40Ð⬍60 274 21,699 17 10.1 78.3 1.7 ⬍0.05 60Ð⬍80 659 42,915 80 20.0 186.4 4.0 ⬍0.001 Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 80Ð100 362 23,493 54 10.9 229.9 4.9 ⬍0.001 Area 3 20Ð⬍40 33 756 4 0.4 529.2 11.4 ⬍0.001 Area 4 20Ð⬍40 343 9,226 1 4.3 10.8 0.2 ⬎0.05 40Ð⬍60 230 20,273 16 9.4 78.9 1.7 ⬍0.05 60Ð⬍80 255 1627 30 0.8 1,844.2 39.6 ⬍0.001 80Ð100 310 5730 4 2.7 87.3 1.9 ⬎0.05 Area 5 20Ð⬍40 118 3,236 2 1.5 61.8 1.3 ⬎0.05 Area 6 20Ð⬍40 585 13,489 16 6.3 118.6 2.6 ⬍0.001 40Ð⬍60 335 1,784 7 0.8 392.5 8.4 ⬍0.001 60Ð⬍80 773 11,707 200 5.5 1,708.4 36.7 ⬍0.001 80Ð100 2,326 5,696 148 2.7 2,598.2 55.8 ⬍0.001 Area 7 80Ð100 999 11,431 70 5.3 612.4 13.2 ⬍0.001 a Areas deÞned in Fig. 2. b Based on data from Ryan et al. (2004), probability maps were used to deÞne areas with consistently high (⬎75th percentile) numbers of Ae. vigilax (Fig. 2). The probability values were grouped into Þve categories: 0 Ð⬍20, 20 Ð⬍40, 40 Ð⬍60, 60 Ð⬍80, and 80 Ð100%. Seven noncontiguous, geographic areas were found to have ⱖ20% chance of having high Ae. vigilax numbers. c Populations expressed as the number of py. d Expected numbers of complaints calculated by applying complaint incidence rates from the pop residing in the 0 Ð⬍20% area, to populations in the 20 Ð⬍40, 40 Ð⬍60, 60 Ð⬍80, and 80 Ð100% areas. e Morbidity ratio calculated by dividing the actual number of complaints by the expected number of complaints. f Probability that the MR is signiÞcantly greater than or ⬍1.0, based on the assumption that the observed numbers of complaints followed a Poisson distribution with a mean equal to the expected number of cases. rates were 39.6 Ð55.8 times the rate in the area that Impact of Future Population Growth on RRV Dis- never or rarely (0 Ð⬍20% chance) had high Ae. vigilax ease Morbidity Patterns. The projected population for numbers (46.5 complaints per 100,000 py). Redland Shire for the year 2021(Queensland Govern- ment 2005) is 168,434 people, which represents an Table 4. Estimated resident population and numbers of RRV increase of 42.2% from the 2001 population of 118,408 disease cases in Redland Shire in 2021, based on average disease (Table 4). Projections for each of the 12 SLAs indicate incidence rates for each statistical local between 1991 and 2001 the largest population increase will be in the Thorn- lands SLA with an estimated population of 24,158 2021 estimated people, representing an increase of 216.5% from the Estimated % pop no. RRV SLA resident increase disease cases 2001 population (7,632). Based on the long-term av- pop 2021a (2001Ð2021) erage incidence rates for each SLA between 1991 and no.b %c 2001 (Table 1) and the estimated populations in 2021, Alexandra Hills 17,396 -6.0 6.0 (6.3) we estimate that there will be ⬇95 RRV disease cases Balance 8,715 41.2 12.6 (13.2) Birkdale 16,028 16.7 5.2 (5.5) per year and an average Shire incidence rate of 56.3 Capalaba 17,824 -0.4 5.7 (6.0) cases per 100,000 py. The population growth in SLAs Cleveland 16,405 20.1 5.7 (6.1) (Balance, 41.2%; Redland Bay, 128.8%; Sheldon-Mt Ormiston 7,417 44.3 3.3 (3.5) Cotton, 134.4%; and Victoria Point, 39.1%) with higher Redland Bay 16,138 128.8 17.6 (18.5) Sheldon-Mt Cotton 10,313 134.4 8.2 (8.6) than average RRV disease rates, may result in an 8% Thorneside 4,129 17.0 1.2 (1.3) increase in the average risk of RRV disease in Redland Thornlands 24,158 216.5 9.8 (10.3) Shire in 2021, compared with long-term rate between Victoria Point 16,851 39.1 13.1 (13.9) 1991 and 2001. Wellington Point 13,060 52.0 6.5 (6.8) Total 168,434 42.2 94.9 a Discussion Projected populations for Redland Shire in 2021 (Queensland Government 2005). Our analyses of RRV disease and mosquito biting b Estimated numbers of RRV disease cases calculated by multiply- ing long-term RRV disease incidence rates (Table 1) by the projected complaint patterns, together with published data on populations in the 12 SLA. vector abundance (Ryan et al. 2004), have highlighted c Estimated number of RRV disease cases in each SLA divided by the interaction between human and mosquito vector total estimated number from Redland Shire in 2021. populations in urban areas. There was considerable
1056 JOURNAL OF MEDICAL ENTOMOLOGY Vol. 43, no. 5 heterogeneity in both human and vector population centage of the incidence of RRV disease in the densities in Redland Shire, and together, these two mainland and Macleay Island areas (both in areas with factors explained a signiÞcant amount of the variation and without high Ae. vigilax numbers) that would be in RRV morbidity. It is clear from Ryan et al. (2004) eliminated if targeted adult mosquito control activities that the distribution of adult Ae. vigilax was not uni- were undertaken in areas with consistently high form throughout Redland Shire. Several noncontigu- (ⱖ60% chance) numbers of Ae. vigilax. If we were able ous areas (Wellington Point, Victoria Point, Redland to reduce the risk factors for RRV disease in the areas Bay, and the Southern Moreton Bay Islands) in prox- with consistently high numbers of Ae. vigilax to the imity to productive saline water habitats of this species same levels as those in other areas in Redland Shire, had relatively high numbers of adult mosquitoes. Al- then we could expect to reduce the RRV disease though most vector-borne disease transmission cycles incidence by an average of 13.6%. This relatively small are complex and density of vector mosquitoes is not reduction in RRV morbidity is due to the fact that the always correlated with pathogen transmission inten- majority of people in Redland Shire reside in areas Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 sity (Beier et al. 1999), the current study has shown where Ae. vigilax numbers are not extremely high and that areas with consistently high numbers of adult Ae. the risk of RRV disease is relatively low. Interestingly, vigilax (Table 2) have higher than expected numbers when we calculated the population attributable risk of RRV disease cases. RRV disease rates in these areas percentage for mosquito biting complaint incidence are up to 2.9 times those in areas which rarely had high that was associated with areas with consistently high Ae. vigilax numbers. It should be noted, however, that Ae. vigilax numbers, we estimated that 58.4% of the a range of mosquito species, in addition to Ae. vigilax, mosquito biting complaint incidence could be elimi- were probably involved in RRV transmission in Red- nated if areas with consistently high numbers of Ae. land Shire (Ryan et al. 2004). These species include vigilax were targeted. Basically, the majority of RRV the brackish water species Ve. funerea and the fresh- disease cases occurred in areas with relatively low Ae. water species Cx. annulirostris, Cq. linealis, Ae. noto- vigilax numbers, whereas the majority of mosquito scriptus, and Ae. procax (Ryan et al. 2000, 2004; Jeffery biting complaints occurred in areas with consistently et al. 2002a,b). Although these species were abundant high numbers of Ae. vigilax adults. Therefore, a tar- in some areas in Redland Shire, the lack of signiÞcant geted approach to adult mosquito control in areas with autocorrelation in trap catches or the high seasonal high Ae. vigilax numbers will probably have a signif- variability in the spatial patterns meant that probabil- icant effect on the number of mosquito biting com- ity maps to deÞne areas with consistently high num- plaints, but a limited effect on the number of RRV bers were not reliable (Ryan et al. 2004). Surveillance disease cases. activities to deÞne the seasonal and spatial distributions Although targeted control of adult Ae. vigilax in of these mosquito species are required. In addition, het- high-risk areas may result in modest reductions in RRV erogeneity in the density of vertebrate hosts of RRV such disease morbidity, effective control of immature as macropods and possums (Kay et al. 1986, Boyd et al. stages in productive saline water habitats may result in 2001) probably occurred throughout the study area, and widespread reductions in adult Ae. vigilax densities this also may have an affect on virus ampliÞcation and the throughout wider areas in Redland Shire. Ae. vigilax spillover of virus in human populations. adults are known to disperse inland on purposive It has been proposed that accurately targeted in- ßights (Marks 1969, Lee et al. 1984), and in Redland terventions in areas around speciÞc mosquito breed- Shire upwards of 300 adults were collected in carbon ing sites can be expected to give greatly improved dioxide- and 1-octen-3-ol-baited light traps per night, levels of malaria control compared with untargeted in areas located ⬎10 km from the nearest saline water strategies (Carter et al. 2000). In terms of resources habitat (P.A.R. and D.A., unpublished data). There- and environmental impacts, it is desirable to target fore, it would be more prudent to focus control mea- control activities to speciÞc areas. Although some ar- sures against immature stages in aquatic environ- eas in Redland Shire had consistently high numbers of ments, before emergence and dispersal of adult Ae. vigilax and high risk of RRV disease in humans, it mosquitoes. This is consistent with most local govern- is clear that targeted control of adult Ae. vigilax in ment mosquito control programs in coastal areas of these areas alone would not be effective in preventing southeastern Queensland (Bell 1989), which involve RRV disease. For example, based on the 60 Ð⬍80 and the application of various formulations of Bacillus thu- 80 Ð100% probability categories that were used to de- ringiensis variety israelensis and (S)-methoprene to Þne areas with consistently high numbers of Ae. vigilax saline water habitats. (Fig. 2), we calculated RRV disease relative risks of 2.2 Our mapping of RRV disease cases was to the pa- and 2.9, respectively, compared with the risk of disease tientsÕ usual place of residence, and we recognize that in areas that had 0 Ð⬍20% chance of having high num- this may not necessarily be the same location at which bers of Ae. vigilax. The areas with consistently high Ae. the person was infected. The retrospective nature of vigilax numbers comprised 5,684 ha or 22.8% of the the study prevented us from ascertaining whether total mainland and Macleay Island area, yet only 10.7% patients had traveled before RRV disease onset and of the human population resided in these areas. To whether these areas represented potential foci of RRV estimate the potential beneÞt of targeted adult mos- transmission. Therefore, if we accept that a proportion quito control in these areas, we calculated the popu- of the RRV disease cases were infected in areas other lation attributable risk percentage, which is the per- than where they lived, and we assume that peoplesÕ
September 2006 RYAN ET AL.: SPATIAL ANALYSIS OF ROSS RIVER VIRUS DISEASE CASES 1057 travel habits are fairly uniform throughout the Shire, window but instead use a ßexibly shaped spatial scan then our mapping of disease cases to usual place of statistic that can detect irregular shaped clusters may residence probably underestimated the heterogeneity prove useful for detecting irregular shaped clusters of in disease risk. Heterogeneity in RRV disease risk has vector-borne disease (Ozdenerol et al. 2005, Tango been noted in other areas, such as Brisbane and Dar- and Takahashi 2005). win, where higher than expected numbers of RRV There was a clear spatial association between re- disease cases were reported from areas near wetlands ports of mosquito biting exposure from residents and and other major mosquito breeding sites (Whelan et RRV disease incidence. Although tolerance to mos- al. 1997, Muhar et al. 2000). quito attack is likely to vary from person to person and Information on the heterogeneity in vector borne complaints of mosquito biting may not always be due disease risk also can lead to a more informed deci- to attack by competent RRV vectors, in Redland Shire sion-making framework to help balance the needs at least, reports of mosquito biting exposure was found for human development, public health, and respon- to be a sensitive indicator of elevated risk of RRV Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 sible land use. Based on the long-term average inci- disease. Areas with high numbers of mosquito biting dence rates between 1991 and 2001 and the 2021 complaints should be surveyed to determine whether population projections for each of the 12 SLAs in there is exposure to medically important species, and Redland Shire, we estimated that there will be ⬇95 if necessary, additional mosquito control measures can RRV disease cases per year and an average Shire in- be implemented. cidence rate of 56.3 cases per 100,000 py. The popu- At the larger scanning window sizes (1,000 Ð3,000 lation growth in SLAs with higher than average RRV m) there were consistent patterns of signiÞcantly disease rates may result in an 8% increase in the av- higher than expected numbers of RRV disease cases in erage risk of RRV disease in Redland Shire in 2021, four areas: Victoria Point and Redland Bay on the compared with the long-term rate between 1991 and mainland; Southern Moreton Bay Islands of Macleay, 2001. Although land use change in these areas can Lamb, Karragarra, Russell; and Dunwich on North potentially have a large impact on local ecology and Stradbroke Island. The disadvantage of mapping cases habitats that effect mosquito abundance, species com- to the SLA level boundaries was the potential to mask position, and ultimately pathogen transmission (Nor- any patterns within the SLA administrative area. This ris 2004), urban planning usually occurs without con- was evident for mosquito biting complaints in the sideration for environmental affects that can lead to Wellington Point area, which were not signiÞcantly increased vector-borne disease burden. Incorporation different from the expected numbers when analyzed of basic surveillance data on vectors and vector-borne by SLA, yet SaTScan analyses indicated higher than diseases into a land use planning framework has the expected numbers of complaints in coastal areas (Fig. potential to guide development into areas of low dis- 5). Similarly with RRV and mosquito biting complaints ease risk or alternatively identify the need for miti- from the Southern Moreton Bay Islands (Balance gation strategies to reduce future disease burden in SLA), where the SaTScan analysis method was able to these populations. identify speciÞc areas with higher than expected num- With the aim of providing practical information to bers of cases (Macleay, Lamb, Karragarra, and Russell local governments on methods for mapping vector- Islands, and Dunwich) and complaints (Macleay, borne disease patterns, we examined RRV cases and Lamb, Karragarra, and Russell Islands). Based on SLA mosquito complaints patterns at two spatial scales: 1) level analysis, one could only assume homogeneity in broad patterns, using predeÞned SLA level boundaries cases and complaints throughout the region and this as deÞned by the Australian Bureau of Statistics; and was clearly not the case (i.e., no clusters of signiÞ- 2) Þner patterns ranging from individual 500- by 500-m cantly higher than expected numbers of cases or com- grids, through to circular windows with radii ranging plaints in the northern population centers of Amity or from 1,000 to 3,000 m (SaTScan analyses). Generally, Point Lookout). results from the two approaches were in agreement, As demonstrated above, the issue of spatial scale is with higher than expected RRV disease cases and important for mapping of vector-borne disease pat- mosquito biting complaints from Victoria Point, Red- terns. Clustering of vector-borne disease cases can be land Bay, and the Southern Moreton Bay Island areas. a result of broad environmental risk factors such as Given the fact that RRV disease notiÞcation data are weather events, which can be correlated with arbo- now provided to local governments according to the virus disease incidence over several hundred kilome- suburb in which the patient resided, mapping cases to ters (Gatton et al. 2004, 2005), through to clustering of the level of SLA is probably sufÞcient to highlight cases over several kilometers such as those found in broad differences in RRV disease incidence. Depend- southern coastal areas in Redland Shire, which had ing on the size and shape of the SLA, additional anal- consistently high Ae. vigilax numbers. In Dakar, Sene- yses using SaTScan or similar cluster detection soft- gal, Plasmodium falciparum prevalence in humans and ware may be undertaken. One potential limitation of their proximity to Anopheles arabiensis Patton imma- SaTScan is the fact that it uses a circular window to ture habitats was correlated over several hundred scan for clusters, which makes it difÞcult to correctly meters (Trape et al. 1992), as opposed to dengue cases detect noncircular based clusters such as those that in Florida, Puerto Rico, which showed signiÞcant clus- may occur along rivers or transport routes. Several tering within households and at very short distances new methods that do not rely on a circular scanning (⬍10 m) (Morrison et al. 1998). The use of arbitrary
1058 JOURNAL OF MEDICAL ENTOMOLOGY Vol. 43, no. 5 administrative boundaries for spatial analyses has the in Queensland, Australia. Am. J. Trop. Med. Hyg. 71: potential to mask any small-scale heterogeneity in 629 Ð 635. disease patterns. With the availability of georefer- Gatton, M., B. Kay, and P. Ryan. 2005. Environmental pre- enced data sets and high-resolution imagery, it is be- dictors of Ross River virus disease outbreaks in Queens- coming more feasible to undertake spatial analyses at land, Australia. Am. J. Trop. Med. Hyg. 72: 792Ð799. Getis, A., A. C. Morrison, K. Gray, and T. W. Scott. 2003. these relatively small scales. Characteristics of the spatial pattern of the dengue vec- tor, Aedes aegypti, in Iquitos, Peru. Am. J. Trop. Med. Hyg. 69: 494 Ð505. Acknowledgments Harley, D., S. Ritchie, D. Phillips, and A. van den Hurk. 2000. Mosquito isolates of Ross River virus from Cairns, Funding for this research was provided, in part, by an Queensland, Australia. Am. J. Trop. Med. Hyg. 62: 561Ð Arbovirus Prevention Research grant (ARBO 00-06) from 565. Queensland Health. Hjalmars, U., M. Kulldorff, G. Gustafsson, and N. Nagarwalla, N. 1996. Childhood leukaemia in Sweden: using GIS and Downloaded from https://academic.oup.com/jme/article/43/5/1042/881278 by guest on 03 January 2021 a spatial scan statistic for cluster detection. Stat. Med. 15: References Cited 707Ð715. Jeffery, J.A.L., P. A. Ryan, S. A. Lyons, P. T. Thomas, and B. H. Ali, M., Y. Wagatsuma, M. Emch, and R. F. Breiman. 2003. Kay. 2002a. Spatial distribution of Ross River virus and Use of a geographic information system for deÞning spa- Barmah Forest virus vectors on Russell Island, Moreton tial risk for dengue transmission in Bangladesh: role for Bay, Queensland. Aust. J. Entomol. 41: 329 Ð338. Aedes albopictus in an urban outbreak. Am. J. Trop. Med. Jeffery, J.A.L., P. A. Ryan, S. A. Lyons, and B. H. Kay. 2002b. Hyg. 69: 634Ð640. Vector competence of Coquillettidia linealis (Skuse) Australian Bureau of Statistics. 2003. Census of population (Diptera: Culicidae) for Ross River and Barmah Forest and housing, selected social and housing characteristics viruses. Aust. J. Entomol. 41: 339 Ð344. for statistical local areas, Queensland, 2001 (2015.1). Kay, B. H., and J. G. Aaskov. 1989. Ross River virus (epi- Commonwealth of Australia, Canberra, Australia. demic polyarthritis), pp. 93Ð112. In T. P. Monath [ed.], Beier, J. C., G. F. Killeen, and J. I. Githure. 1999. Entomo- The arboviruses: epidemiology and ecology, vol. 4. CRC, logic inoculation rates and Plasmodium falciparum ma- laria prevalence in Africa. Am. J. Trop. Med. Hyg. 61: Boca Raton, FL. 109Ð113. Kay, B. H., R. A. Hall, I. D. Fanning, P. Mottram, P. L. Young, Bell, K. M. 1989. Development and review of the contiguous and C. C. Pollitt. 1986. Experimental infection of verte- local authority group programme on saltmarsh mosquito brates with Murray Valley encephalitis and Ross River control. Arbovirus Res. Aust. 5: 168Ð171. viruses. Arbovirus Res. Aust. 4: 71Ð75. Boyd, A. M., R. A. Hall, R. T. Gemmell, and B. H. Kay. 2001. Kulldorff, M., and N. Nagarwalla N. 1995. Spatial disease Experimental infection of Australian brushtail possums, clusters: detection and inference. Stat. Med. 14: 799 Ð 810. Trichosurus vulpecula (Phalangeridae: Marsupialia), with Lee, D. J., M. M. Hicks, M. Griffiths, R. C. Russell, and E. N. Ross River and Barmah Forest viruses by use of a natural Marks. 1984. Monograph Series, Entomology Mono- mosquito vector system. Am. J. Trop. Med. Hyg. 65: 777Ð graph No. 2, The Culicidae of the Australasian Region, 782. vol. 3. pp 207Ð233, Australian Government Publishing Brooker, S., S. Clarke, J. K. Njagi, S. Polack, B. Mugo, B. Service, Canberra, Australia. Estambale, E. Muchiri, P. Magnussen, and J. Cox. 2004. Liebhold, A. M., R. E. Rossi, and W. P. Kemp. 1993. Geosta- Spatial clustering of malaria and associated risk factors tistics and geographic information systems in applied in- during an epidemic in a highland area of western Kenya. sect ecology. Annu. Rev. Entomol. 38: 303Ð327. Trop. Med. Int. Health 9: 757Ð766. Lothrop, H. D., and W. K. Reisen. 1999. A geographical Brownstein, J. S., H. Rosen, D. Purdy, J. R. Miller, M. Mer- information system to manage mosquito and arbovirus lino, F. Mostashari, and D. Fish. 2002. Spatial analysis of surveillance and control data in the Coachella Valley of West Nile virus: rapid risk assessment of an introduced California. J. Am. Mosq. Control Assoc. 15: 299 Ð307. vector-borne zoonosis. Vector Borne Zoonotic Dis. 2: Marks, E. N. 1969. Mosquitoes (Culicidae) on QueenslandÕs 157Ð164. coral cays. Qld Nat. 19: 94 Ð98. Carter, R., K. N. Mendis, and D. Roberts. 2000. Spatial tar- Morrison, A. C., A. Getis, M. Santiago, J. G. Rigau-Perez, and geting of interventions against malaria. Bull. World P. Reiter. 1998. Exploratory space-time analysis of re- Health Organ. 78: 1401Ð1411. ported dengue cases during an outbreak in Florida, Commonwealth Bureau of Meteorology. 2004. Climate Av- Puerto Rico, 1991Ð1992. Am. J. Trop. Med. Hyg. 58: 287Ð erages for Australian Sites-Averages for Redlands 298. (http://www.bom.gov.au/climate/averages/tables/cw_ Morrison, A. C., K. Gray, A. Getis, H. Astete, M. Sihuincha, 040265.shtml). D. Focks, D. Watts, J. D. Stancil, J. Olson, P. Blair, and Doherty, R. L. 1972. Arboviruses in Australia. Aust. Vet. J. T. W. Scott. 2004. Temporal and geographic patterns of 48: 172Ð180. Aedes aegypti (Diptera: Culicidae) production in Iquitos. Dwass, M. 1957. ModiÞed randomization tests for nonpara- Peru J. Med. Entomol. 41: 1123Ð1142. metric hypotheses. Ann. Math. Stat. 28: 181Ð187. Mostashari, F., M. Kulldorff, J. J. Hartman, J. R. Miller, and [ESRI] Environmental Systems Research Institute Inc. V. Kulasekera. 2003. Dead bird clusters as an early warn- 1996a. ArcView GIS 3.2a for Windows. ESRI, Redlands, ing system for West Nile virus activity. Emerg. Infect. Dis. CA. 9: 641Ð 646. [ESRI] Environmental Systems Research Institute Inc. Muhar, A., P.E.R. Dale, L. Thalib, and E. Arito. 2000. The 1996b. Using the ArcView Spatial Analyst. ESRI, Red- spatial distribution of Ross River virus infections in Bris- lands, CA. bane: signiÞcance of residential location and relation- Gatton, M., L. Kelly-Hope, B. Kay, and P. Ryan. 2004. Spa- ships with vegetation types. Environ. Health Prev. Med. tial-temporal analysis of Ross River virus disease patterns 4: 184 Ð189.
You can also read