| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
 |
ABSTRACT |
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
|
INTRODUCTION |
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Dengue is transmitted to humans, the only known reservoir of the disease, by Aedes mosquitoes. Adult female Aedes mosquitoes acquire the dengue virus by biting an infected human during the viremic phase, which usually lasts for 4–5 days, but may last up to 12 days.2,3 The virus is transmitted to other persons via bites from infected mosquitoes.
Aedes aegypti and Ae. albopictus have different breeding and biting patterns and these differences could influence potential for dengue transmission. These patterns could be relevant to efforts in progress to better define and control transmission of dengue viruses in Bangladesh.4–6 Efforts are also in progress to better define and control transmission of dengue viruses in Bangladesh.
The principle breeding sites of Aedes mosquitoes include containers such as cans, pots, earthen jars, plant pot bases, and drums, as well as tires and tree holes.7–9 Breeding sites of Aedes mosquitoes vary spatially.10 In a dengue control program, it is useful to identify precise areas of concentrated Aedes breeding. This information is particularly valuable if it can be linked with dengue incidence, since both the mosquito species and infected people are necessary for continued transmission. Areas with dense populations near breeding sites require interventions to minimize transmission.
Geographic information systems (GISs) are making practical contributions to public health.11 Identifying high-risk areas can be particularly useful for prioritizing prevention in public health decision-making.12–17 This study used a GIS to identify spatial risk for dengue. We measured spatial associations between clusters of dengue cases and vector populations to identify areas of increased transmission. We produced a risk map so local health officials can target specific areas for focused interventions.
|
MATERIALS AND METHODS |
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
During the survey, the household head or other adult resident was asked whether anyone in the household had dengue/ dengue hemorrhagic fever diagnosed at health facilities during the past two months. We used these data to identify self-reported cases of dengue.
Field workers also looked for containers with standing water and for Aedes larvae within containers inside, outside, and on the roof of each household. A total of 44,532 larvae were collected, and were reared to their adult stage whenever possible for species identification. We used total vector populations (Ae. aegypti and Ae. albopictus) per household for the calculations to define clusters of vector populations.
We mapped each household using Magellan (San Dimas, CA) 315/320 global positioning system (GPS) receivers. An accuracy assessment of the receiver provided no differences in the reading obtained from Geoexplorer (Trimble Navigation, Ltd., Sunnyvale, CA) (a relatively expensive GPS receiver), which suggested that it was reasonable to use the less expensive receivers. Of the 9,462 houses, the GPS readings of 170 houses were found to be inaccurate (did not coincide within the Dhaka metropolitan area); thus, they were excluded while constructing the GIS. Two houses provided the same geographic coordinates in eight cases since they were observed to be very close to each other. A single point was referenced for each of those pairs; the household level data of the pairs were aggregated accordingly. Ultimately, we could construct the GIS with 9,284 reference points for houses. The spatial distribution of the houses surveyed under the entomologic assessment study is shown in Figure 1
.
|
Clusters of dengue cases. We used Kernel estimation to demonstrate areas of higher intensity of dengue cases. Each kernel takes account of observed data around a point of measurement, thus providing smooth estimates of univariate probability densities from a sample of observations.21–24 The mathematical expression for estimating kernel of the intensity of dengue cases at point s is defined by
 |
where ei,.....,en are the observed cases at locations si,....,sn respectively; ei has a value of 1 if a case is observed at si and 0 otherwise. The further a point is from the point of measurement (kernel), the lesser the weight value (w) for computing the kernel. Thus, the kernel weight (wi) for the location si is computed by
 |
where 
is the maximum pixel distance (20 in this study [21 x 21-pixel window]) and di is the pixel lag between s and si. We chose 21 x 21-pixel window representing a 44,100 m2 (210 meters x 210 meters) area. This provided an average of six data points for computing the intensity at each s, which yielded the best estimation for addressing local level (neighborhood-based) variation in the data. Note that in a 10,000 m2 (100 meters x 100 meters) area, there would be only 2 data points, and 12 data points would have been provided if we covered a size of 90,000 m2 (300 meters x 300 meters) area. If we select a 10,000 m2 area, the data would display more individual variability, whereas a 90,000 m2 area would tend to blur local level variation in the data.
For 464 households, the respondents did not know whether dengue had occurred within the household; thus, 8,820 households were included in the analysis. The intensity of dengue cases was then computed at a regular interval of 210 meters. Finally, using a contouring algorithm25 with the regularly spaced data, we created the surface of intensity of cases from where the clusters of higher intensity of cases were isolated. To isolate the clusters, we used an iterative process beginning with intensity, 
(s), of 0.10, increasing stepwise. At intensity of 0.58, the iterative process yielded a visibly distinguishable threshold for defining higher intensity areas of dengue cases.
Clusters of vector populations. To identify clusters of higher density of vector populations, we began by interpolating the data on vector population using a geostatistical method called kriging, the estimates of which are unbiased and have minimum variance.26,27 The method is based on the assumption of spatial autocorrelation of data, and the autocorrelation structure (spatial distribution of the data) is addressed through variogram modeling. The underlying assumption of the variogram is that two observations close together are more similar than those further apart. Using the variogram in combination with observed data, we used ordinary kriging to interpolate the data at regular intervals (which yields intermediate locations, much like the way kernal values are calculated earlier in this report). An ordinary kriged estimate of a variable Z at a point s is the weighted average of
 |
where Z(si) is the observed data value at points i, and 
i is the weight associated with the data at point i, which is obtained from a variogram modeling.
Since the observed data on a vector population are highly skewed, the logged (natural) values were used instead in the variogram function and in kriging the data. A grid cell of size 210 meters was chosen to allow the data to be interpolated at a regular interval of 210 meters. Again, using a contour algorithm, we created the surface of density of the vector population, and the areas of higher density of the vector population were isolated from the surface map. Using a visual process to discriminate high and low vector density areas, we arrived at a cut-off value of 3.26 (logarithmic scale) to define high-density vector areas.
Spatial analysis. Spatial regression. Distance surfaces were created for each of the cluster maps to be used in a regression model for analyzing spatial association between the clusters of dengue cases and vector populations. In the distance surface, the further a pixel was from the cluster, the higher the attribute value was carried by the pixel. Distance surface was also created for the four major hospitals (Holy Family Red Crescent Hospital, Dhaka Medical College Hospital, Bangladesh Institute of Research for Diabetics and Endocrinal Medicine Hospital, and Bangabandhu Sheikh Mujib Hospital) where many dengue patients were admitted in 2000. The hospital distance surface was considered as the covariate in the multiple regression model.
The coefficient of determination (r2) was computed to determine whether variances in clustering of dengue cases could be explained by independent factors (clusters of high density vector populations and proximity to the four major hospitals). In the regression models, the distance surface of dengue cases was entered as a dependent variable and the high density vector population as the independent variable.
|
RESULTS |
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
The spatial pattern map shows that self-reported dengue cases are clustered in the southern part of Dhaka in areas where the four major hospitals are located (Figure 2). The results of the simple regression analysis are shown in Table 1. Approximately 54% of the variance (r2 = 0.54) in clustering of cases is explained by the distance to the hospitals; i.e., the greater the distance from the hospitals, the fewer the clusters of cases (Figure 3).
|
|
|
). In contrast, the presence of Ae. aegypti in the household premises was not associated with self-reported disease (Table 3). Using a GIS tool, we created a buffer of 50 meters around each of the households. Households within each buffer that reported a case of dengue and had Aedes larvae detected within the 50-meter buffer were counted. The neighborhood assessment was consistent with the household data in that the risk of acquiring dengue was increased when Ae. albopictus was present in and around neighbors’ premises (within 50 meters of the home) (RR = 1.24, 95% CI = 1.02–1.51) (Tables 4 and 5). The spatial pattern of households with Ae. albopictus and dengue cases is shown in Figure 4.
|
|
|
|
|
). The mean population density was determined for the areas around each of the 190 case clusters. The median population density was 50,829/km2 for the case clusters compared with a median population density of 62,061/km2 for the 90 wards within Dhaka.
|
). When distance to the major hospitals was controlled using a multivariate model, approximately 56% of the variance in the clustering of dengue is explained by the clusters of vector population. Thus, the clustering of cases is linked to clustering of Aedes.
The influence of hospital distance on spatial association between clusters of dengue and vector population was further investigated by analyzing spatial subsets. Fourteen buffer zones starting from 1 to 14 km of radius from the center of the four major hospitals were created. Separate regressions were modeled with the data of the fourteen spatial subsets. The first data set included a 1-km2 area around the four hospitals and the last included a 14-km2 area around the hospitals. Eventually, there were more data values for the larger buffer zones. No significant spatial association was observed between the clusters when the analysis was restricted to a small (< 5 km radius) or a large buffer zone (> 11 km radius) (Figure 3). The buffer zone of 8-km radius shows the highest degree of spatial association (coefficient of determination = 0.70) between the clusters.
|
DISCUSSION |
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
This study identified clusters of dengue cases and high vector populations. Areas where the clusters of dengue cases and high vector populations either intersect or are close to each other may be subject to further high transmission of dengue, since both infected persons and vector populations are needed for transmission of the disease. Thus, our observation that population density did not independently correlate with incidence of disease is not surprising, since presence of an infected vector population is the critical factor.
Dengue clusters were also less frequently observed in areas far from the four major hospitals, often despite the presence of dense vector populations. Dengue cases that had occurred in those areas may not have been diagnosed or recognized because many people who live further away may not reach a hospital, which is the setting where dengue diagnosis is most often made in Dhaka. This may have weakened the association we observed between the clusters of dengue cases and vector populations. The poor spatial association when considering larger distances (>11 km) from hospitals suggests that hospital-based surveillance for dengue would substantially underestimate the true disease burden.
Weak spatial associations between dengue clusters and vector populations at very short distances from hospitals may be attributed to the characteristics of these areas. The hospitals are located in commercial areas where relatively fewer people live. Since dengue infection appears to be a recent phenomenon in Bangladesh, it is possible that the proportion of Aedes mosquitoes that are infected with dengue virus is not homogenous around the city; thus, the relationship between vector density and disease incidence might not yet be consistent.
This study was conducted early in the dengue outbreak before geographic information on laboratory-confirmed cases was routinely collected during surveillance. We relied on self-reporting of cases, which limits the precision of our findings since quality of diagnosis may vary by health facilities and patient recall and/or comprehension may not always be optimal. Self-reporting may result in pseudo-clustering because residents living near a true case of dengue may have been more aware of the disease and might have been more likely to assume that they also had dengue fever.
Mayer31 suggested that "the human-environment relationship, if disturbed enough by major changes in land use, migration, population pressure, or other stressors can show significant maladaptation, as manifested by the appearance or diffusion of new diseases." Thus, the clustering of vector populations in distinct sectors of the city suggests that there might have been some behavioral or environmental processes operating at neighborhood levels that supported breeding of Aedes mosquitoes and/or transmission of dengue viruses. Studies are underway to define factors responsible for clustering of vectors. Multi-temporal satellite imagery can be used to model the local level environmental characteristics (such as vegetation type, bodies of water, densely populated communities) that might promote breeding of disease-transmitting vectors.32
Until a safe and effective vaccine is available, controlling Aedes mosquitoes and improved case management are the principal options available for reducing the burden of dengue in Dhaka. An integrated control measure33 targeting high-risk areas is a logical approach to minimize the impact of disease, given the limited resources of this country. This paper identifies a strategy of using spatial risk analysis with human disease and vector surveillance data for targeting areas for integrated interventions to reduce burden of disease during outbreaks of dengue and dengue hemorrhagic fever.
Received December 2, 2002. Accepted for publication July 23, 2003.
Acknowledgments: We are grateful for the cooperation and collaboration of the scientists at International Centre for Diarrhoeal Disease Research, Bangladesh (ICDDR,B). We are also grateful to Touhid Uddin Ahmed, Professor M. Ismail Hossain, and Dr. Manjur A. Chowdhury for their technical assistance on entomologic data collection. We thank Zahirul Huq and Mamunur Rashid for their assistance in constructing the GIS database.
Financial support: This research was funded by the United States Agency for International Development, Duncan Brothers, Ltd., the American Express Foundation, the Department for International Development of the United Kingdom, and the ICDDR,B: Centre for Health and Population Research, which is supported by the countries and agencies that share its concern for the health problems of developing countries. Current donors providing unrestricted support include the aid agencies of the Governments of Australia, Bangladesh, Belgium, Canada, Japan, the Netherlands, Sweden, Sri Lanka, Switzerland, the United Kingdom, and the United States of America; international organizations include United Nations Children’s Fund.
Authors’ addresses: Mohammad Ali, International Vaccine Institute, Seoul National University Campus, Shillim-dong Kwanak-ku, Seoul, Republic of Korea 151-742, Fax: 82-2-872-2803, E-mail: mali{at}ivi.int. Yukiko Wagatsuma and Robert F. Breiman, Centre for Health and Population Research, International Centre for Diarrhoeal Disease Research, Bangladesh (ICDDR,B), Mohakhali, Dhaka, Bangladesh, E-mail: breiman{at}icddrb.org. Michael Emch, Department of Geography, Portland State University, Portland, OR 97207-0751.
|
REFERENCES |
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  C.-C. Kan, P.-F. Lee, T.-H. Wen, D.-Y. Chao, M.-H. Wu, N. H. Lin, S. Y.-J. Huang, C.-S. Shang, I-C. Fan, P.-Y. Shu, et al. Two Clustering Diffusion Patterns Identified from the 2001-2003 Dengue Epidemic, Kaohsiung, Taiwan Am J Trop Med Hyg, September 1, 2008; 79(3): 344 - 352. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. Chansang and P. Kittayapong Application of Mosquito Sampling Count and Geospatial Methods to Improve Dengue Vector Surveillance Am J Trop Med Hyg, November 1, 2007; 77(5): 897 - 902. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
