SEARCH

SEARCH BY CITATION

Keywords:

  • Aedes albopictus;
  • dengue;
  • basic reproductive rate;
  • spatial analysis;
  • land cover;
  • land use
  • Aedes albopictus;
  • dengue;
  • taux de reproduction de base;
  • analyse spatiale;
  • revêtement du sol;
  • utilisation des terres
  • Aedes albopictus;
  • dengue;
  • tasa reproductiva básica;
  • análisis espacial;
  • cobertura del suelo;
  • uso de la tierra

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Case study: dengue virus transmission in Hawai’i
  5. Data and methods
  6. Results
  7. Discussion and conclusion
  8. Acknowledgements
  9. References

Vector-borne diseases persist in transmission systems that usually comprise heterogeneously distributed vectors and hosts leading to a highly heterogeneous case distribution. In this study, we build on principles of classical mathematical epidemiology to investigate spatial heterogeneity of disease risk for vector-borne diseases. Land cover delineates habitat suitability for vectors, and land use determines the spatial distribution of humans. We focus on the risk of exposure for dengue transmission on the Hawaiian island of Oahu, where the vector Aedes albopictus is well established and areas of dense human population exist. In Hawai’i, dengue virus is generally absent, but occasionally flares up when introduced. It is therefore relevant to investigate risk, but difficult to do based on disease incidence data. Based on publicly available data (land cover, land use, census data, surveillance mosquito trapping), we map the spatial distribution of vectors and human hosts and finally overlay them to produce a vector-to-host ratio map. The resulting high-resolution maps indicate a high spatial variability in vector-to-host ratio suggesting that risk of exposure is spatially heterogeneous and varies according to land cover and land use.

Les maladies transmisses par des vecteurs persistent dans les systèmes de transmission qui impliquent habituellement des vecteurs et des hôtes distribués de façon hétérogène, menant à une distribution très hétérogène des cas. Dans cette étude, nous nous appuyons sur les principes de l’épidémiologie mathématique classique pour examiner l’hétérogénéité spatiale du risque de maladie pour les maladies à transmission vectorielle. Le revêtement du sol délimite l’habitat convenable pour les vecteurs et l’utilisation des terres détermine la répartition spatiale de l’homme. Nous nous concentrons sur le risque d’exposition pour la transmission de la dengue sur l’île hawaïenne d’Oahu, où le vecteur Aedes albopictus est bien établi et des zones de population humaine dense existent. A Hawaï, le virus de la dengue est généralement absent, mais se propage occasionnellement lors de son introduction. Il est donc pertinent d’étudier les risques, mais il est difficile de le faire sur la base de données d’incidence de la maladie. Sur base de données accessibles publiquement (le revêtement du sol, l’utilisation des terres, les données de recensement, la surveillance des moustiques piégés), nous avons cartographié la distribution spatiale des vecteurs et des hôtes humains, et enfin les avons superposé afin de produire une carte du rapport vecteur/hôte. Les cartes obtenues avec une haute résolution indiquent une forte variabilité spatiale dans le rapport vecteur/hôte ce qui suggère que le risque d’exposition est spatialement hétérogène et varie selon le revêtement du sol et l’utilisation des terres.

Las enfermedades de transmisión vectorial persisten en sistemas de transmisión que usualmente incluyen vectores con distribución heterogénea y hospederos que presentan una distribución de casos altamente heterogénea. En este estudio nos basamos en los principios matemáticos clásicos de la epidemiología para investigar la heterogeneidad espacial del riesgo de enfermedad de las enfermedades vectoriales. La cobertura de suelos delinea la idoneidad del hábitat para los vectores, y el uso de la tierra determina la distribución espacial de los hombres. Nos enfocamos en el riesgo de exposición para la transmisión del dengue en la isla Hawaiana de Oahu, en donde el vector Aedes albopictus está bien establecido y donde existen áreas densas de población humana. En Hawai, el virus del dengue está por lo general ausente, aunque ocasionalmente surge cuando es introducido. Por lo tanto es relevante investigar el riesgo, pero es difícil hacerlo basándose en datos de incidencia de la enfermedad. Basándonos en los datos públicos disponibles (cobertura de suelos, uso de la tierra, datos censales, vigilancia de trampas de mosquitos), hemos mapeado la distribución espacial de los vectores y de los hospederos humanos, y finalmente los hemos sobrepuesto para producir un mapa del ratio vector-hospedero. Los mapas resultantes, de alta resolución, indican que hay una variabilidad espacial alta en el ratio vector – hospedero, sugiriendo que el riesgo de exposición es espacialmente heterogéneo y varía de acuerdo con la cobertura del suelo y el uso de la tierra.


Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Case study: dengue virus transmission in Hawai’i
  5. Data and methods
  6. Results
  7. Discussion and conclusion
  8. Acknowledgements
  9. References

The distribution of vector-borne diseases tends to be highly spatially heterogeneous, varying according to the often-heterogeneous spatial distribution of transmission systems components: vectors, pathogens and hosts. Vectors and hosts depend on spatially diverse environmental conditions, including land cover. Different land uses modify contact of susceptible humans with infectious vectors (Vanwambeke et al. 2007a) and modify human cases distribution (Norris 2004; Vanwambeke et al. 2006; Linard et al. 2007). Heterogeneity in the risk of disease transmission results from spatial heterogeneity in both land cover and land use. Understanding sources of spatial heterogeneity can contribute to disease prevention and control. In this context, a logical first step is to produce high-resolution maps detailing the spatial heterogeneity of the system. In this study, we combine publicly available data on land cover, land use and human populations with mosquito surveillance data to produce a spatially disaggregated map representing the vector-to-human host ratio, a quantity originating from process-based epidemiological models of mosquito-borne diseases. The vector-to-host ratio used here allows an evaluation of the level of risk of contact in the absence of the pathogen as it does not rely on sparse epidemiological data. Such evaluations are particularly important in areas that experience occasional epidemic transmission. In such cases, the pathogen may not reach all susceptible population or areas, making epidemiological data an incomplete reflection of areas where conditions are suitable for vector and host presence.

Two major approaches have been used to investigate environmental factors related to vector-borne transmission of diseases: empirical statistical models and mechanistic models. Empirical statistical models are based on inductive approaches and quantitative statistical analyses, including in a spatially explicit context (Vanwambeke et al. 2007a, 2010. See also Eisen & Lozano-Fuentes 2009 for a review applied to Aedes aegypti and dengue virus). The drawback of these models is that their inference is limited to the domain of the data used to construct them. Diverse spatial data exist, varying in extent and resolution, but remotely sensed data provide extensive coverage and rich information content. They have been successfully used in epidemiology and vector-borne disease ecology (Herbreteau et al. 2005, 2007), particularly for mapping vector or disease case distribution. Such data provide a higher level of detail than census-like data, which aggregate spatially heterogeneous variables at the level of administrative units and are difficult to associate with human activities (Rindfuss et al. 2003). Census-like data are also focused on residence. However, humans spend time in other places with potentially profound implications for infectious disease epidemiology (Adams & Kapan 2009; Stoddard et al. 2009).

Mechanistic models for mosquito-borne diseases, such as the susceptible-infected-recovered model (SIR) modified for vector-borne disease (Aron & May 1982; Rogers 1988; Anderson & May 1991; Rogers et al. 2002; Snow & Gilles 2002; Adams & Kapan 2009), include explicit details of the transmission processes and express the number of infections per infected host (e.g. the basic reproductive rate R0) in terms of biologically meaningful parameters. R0 increases with vector biting rate, vector survival and proportion of bites infecting the host or vector. Parameters relating to infection or recovery probabilities are less influenced by extrinsic environmental parameters. Mosquito activity and survival are influenced by spatially variable climatic parameters. Mosquitoes tend to bite more often, and the extrinsic incubation period becomes shorter at higher temperatures, but mortality may rise at lower relative humidity and past a temperature threshold (Watts et al. 1987; Focks et al. 1993; Jetten & Focks 1997), leading to difficulties in generalizing the effects of changes in weather and/or climate on vector-borne disease (Rogers et al. 2002).

Thus, SIR models can be difficult to relate to environmental factors and spatially explicit knowledge of vectors and host populations at risk. We address this difficulty by relating vector abundance to host density (considering here exclusively human hosts), summarized as the vector-to-host ratio (Rogers et al. 2002), to determine the level of risk at the local, landscape scale. In the basic model, it is expected that a high vector-to-host ratio (numerous mosquitoes per host) increases the risk of exposure, while a low ratio (few mosquitoes per host) reduces it. The vector-to-host ratio can be highly spatially heterogeneous, in relation to the spatial distribution of mosquitoes (Vanwambeke et al. 2007a), and human habitat and activities (Linard et al. 2009). Attempts at developing spatially explicit mechanistic models have been limited, particularly at the landscape scale where human activities must be considered. Existing examples focus on veterinary diseases and coarser scales such as for bluetongue virus at the national scale (Hartemink et al. 2009a) and canine leishmaniasis at the regional scale (Hartemink et al. 2009b).

Both empirical and process-based approaches therefore have distinct but complementary advantages, the former being grounded on real-world observations, the latter explicitly detailing transmission processes. Another approach considers the disease ecological niche (Peterson 2008), and the area of high potential risk, as a function of the actual distribution of host and vector. But niche approaches necessitate disease presence records, or focus on the vector distribution, and thereby exclude the human role in transmission (e.g. Peterson et al. 2005; Moffett et al. 2007). In this study, we build on empirical, mechanistic and niche approaches by combining disaggregated human population density data with coarse estimates of mosquito population abundance to produce a spatially explicit map of risk of exposure, focusing on the risk of dengue transmission in Hawai’i.

Case study: dengue virus transmission in Hawai’i

  1. Top of page
  2. Summary
  3. Introduction
  4. Case study: dengue virus transmission in Hawai’i
  5. Data and methods
  6. Results
  7. Discussion and conclusion
  8. Acknowledgements
  9. References

About half of the world’s population is potentially exposed to dengue virus, resulting in an estimated 50–100 million infections yearly. Of these, 250 000–500 000 cases annually manifest as severe dengue haemorrhagic fever, one of the most serious arboviral diseases, mostly fatal in children (WHO, 1997). Aedes aegypti is generally considered the main vector and is responsible for many epidemics worldwide. Aedes aegypti is a domestic, anthropophilic mosquito that thrives in and around human dwellings where a variety of sources of standing water (e.g. water storage containers) provide ideal habitat for its larval stage. Aedes albopictus is also a vector of dengue – its larvae prefer discarded debris such as tyres, as well as tree holes and other vegetation-associated ‘containers’ that readily capture rainwater. As Ae. albopictus is generally a much less aggressive biter of humans and not restricted to taking human blood meals, it is considered a less important dengue vector (Chung & Pang 2002). However, in various areas of Asia, Ae. albopictus does primarily or exclusively cause dengue transmission (Gratz 2004).

In Hawai’i (Figure 1), Ae. aegypti was widespread by 1892 (Perkins 1913) and caused early epidemics. It was controlled in the 1940s (Nakagawa & Hirst 1959) and is now only readily found on the leeward side of the island of Hawai’i (Hawai’i Department of Health, Vector Control Branch unpublished data communicated by P. Yang, 2008, Larish & Yang 2009; Winchester and Kapan, Personal Observation). Aedes albopictus was recorded in 1892 (Perkins 1913; Usinger 1944) and was widespread by 1896 (Perkins 1913; Joyce 1961). Aedes albopictus is currently the only widespread dengue vector in Hawai’i and caused the most recent dengue outbreak in 2001 (Effler et al. 2005).

image

Figure 1.  Map of the state of Hawai’i.

Download figure to PowerPoint

As evidenced by the recent (2001) and previous epidemics from the 1840s, 1903, 1912, and 1943–1944 (Wilson 1904; Gubler 1997), local transmission of dengue virus has occurred throughout the state. In the most recent Oahu outbreak, which likely originated in Tahiti (Effler et al. 2005; Imrie et al. 2006), most cases were found in areas with sparse settlements rather than the state’s main urban centre Honolulu. Based on climatic thresholds for Aedes survival, almost the whole of Oahu is suitable for dengue transmission (Kolivras 2006). However, this does not consider the location of at-risk humans, nor suitable habitats. Another analysis of environmental risk factors in Maui (Napier 2003) indicated that cases concentrated in areas with 1500–5000 mm of annual precipitation, tropical rainforest bordered by grassland, ‘rural demographic types’ and high mosquito populations, but human population density was not considered. Here, spatially explicit information (mosquito abundance, land cover, human population density and land use) was used to evaluate the mosquito-to-human ratio, a measure that should positively relate to the risk for dengue transmission. The analysis focused on the heterogeneously distributed residents, and not on tourists and holiday-goers, as they represent a much more volatile population for whom the risk is considered to be extremely low in Hawai’i (Smith et al. 2005; but see Jelinek et al. 1998 for a potential counterexample). Hotels, which are nearly all situated in Honolulu, apply strict mosquito control.

Data and methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Case study: dengue virus transmission in Hawai’i
  5. Data and methods
  6. Results
  7. Discussion and conclusion
  8. Acknowledgements
  9. References

Land cover data were retrieved from the National Oceanic and Atmospheric Administration Coastal Services Centre (http://csc.noaa.gov/crs/lca/hawaii.html). The data have a 30-m resolution, date from 2000 and include 13 land cover classes. Population census data were obtained from the U.S. census bureau (census.gov) at the census-block level. The TIGER database (Topologically Integrated Geographic Encoding and Referencing system) allowed mapping of census blocks and provided data on places of interest such as beach parks, natural reserves, etc. The beach map issued by the Life Guard Association (LGA) on beaches of Oahu was used in complement (oceansafety.soest.hawaii.edu). A digital elevation model at 30-m resolution was downloaded from the US Geological Survey website (usgs.gov). Mosquito trapping data were obtained from the Hawai’i Department of Health (DoH) in Oahu (P. Yang, personal communication, 2008). Trapping was only carried out below 300 m above sea level.

Human density, land use

Residential density

Population density by census-block level provides a fair idea of the most populated places, but many census blocks cover large areas where few people live. We combined the census information at census-block level with land cover information on built-up areas to calculate a finer spatial population density (30-m resolution), using dasymetric mapping (Mennis 2003). Although two land cover classes of built-up areas were found (high- and low-intensity development), the population density in census blocks entirely covered by one of the two classes did not differ between classes (Wilcoxon rank sum test P = 0.20). Population was therefore reallocated to developed areas regardless of their development intensity. Because of the pronounced topography, isolated dwellings are rare, and our assumption is realistic. Existing population data sets (e.g. GPW, 12-km resolution in the United States, or LandScan, 1-km) do not have as high a resolution.

Recreational density

Places of interest in the TIGER data where people are likely to gather in outdoor settings for recreational purposes were extracted (beach parks, sports grounds, parks and gardens, playgrounds). Landmarks recorded as points in the database were converted to circles with a 100 metre radius. Beach parks missing from the TIGER map were included based on the land cover map combined with the LGA beach map (Figure 2). In the absence of further data, we assumed 10% of the total population was evenly spread between the recreational landmarks, with one-third of people in recreational areas other than beaches and two-thirds of people on beaches. The recreational density map was produced by adding 10% of people (attending recreational activities) to the basic residential density map decreased by 10%. In both cases, the calculated human density was divided into quartiles for combination with the categorized mosquito abundance.

image

Figure 2.  Recreational areas, Oahu, HI.

Download figure to PowerPoint

Mosquito abundance, land cover

The DoH of Hawai’i routinely traps mosquitoes, focusing on ports of entry and built-up areas, i.e. areas used by humans, and is therefore appropriate for this study. New Jersey light traps were used in 64 locations (December 2005 – October 2008) and gravid traps in 35 locations (December 2004 – October 2008) (Figure 3). Traps were checked once a week (New Jersey traps) or every 2–3 days (gravid traps). The total number of female Ae. albopictus was divided by the total number of trapping days to allow comparison across sites and trap types. Trapping performance was compared in eight locations where both types of traps were placed. Although the levels obtained are very different (mean females/per trapping day = 0.01 for New Jersey traps and 0.59 for gravid traps), they are highly correlated (r = 0.8). Therefore, the number of females/trapping day data was divided in three abundance categories using quantiles (calculated per trap type) and then merged. Environmental predictors were consistent when recoding data by trap type, season or presence/absence. The number of females per trapping day was not spatially correlated, as verified using Moran’s I index.

image

Figure 3.  Mean number of female Aedes albopictus trapped per day in New Jersey and gravid traps and trap location on Oahu, HI.

Download figure to PowerPoint

Environmental predictors of mosquito abundance

To capture climatic variability, which is non-limiting for Ae. albopictus (Kolivras 2006), altitude, slope and aspect were calculated for each trap. Environmental factors potentially important to mosquitoes were also described using hexagonal buffers of 50, 100 and 500 metres around traps, covering the estimated flight range of Ae. albopictus (Bonnet & Worcester 1946; Niebylski & Craig 1994; Honorio et al. 2003). Mean slope, proportions of each land cover class, urban classes (‘high-intensity developed’, ‘low-intensity developed’) and vegetation classes (‘scrub/shrub’, ‘evergreen forest’) were calculated within these buffers. Because Ae. albopictus larvae appear to respond to fragmentation (Vanwambeke et al. 2007b), fragmentation indices were calculated for all patches of evergreen forest, scrub/shrub, wetlands and urban land cover classes, and their mean and standard deviations were calculated in the buffers. Indices included patch area, the distance to the nearest neighbour of the same land cover class and the shape index (McGarigal & Marks 1994). A value of 1 indicates the most compact shape (a square), and increasing values over 1 indicate less compact and more complex shapes. To represent the diversity and balance of land cover types, the Shannon index of diversity of land cover was also calculated within these buffers, with higher values indicating greater imbalance between the land cover classes.

Environmental predictors for mosquito abundance categories were investigated using ordinal logistic regression (Kirkwood & Sterne 2003). As 124 variables were available but contained redundancy, variables were first examined in models with one independent variable. Variables significant at the 0.05-level were then introduced in a multiple model and retained if they were significant (0.05-level). All variables were then included one-by-one to test for interactions. The multiple model was checked for collinearity and confounding variables.

Combining information

The island was subdivided into smaller landscapes (1 ha, or 100-m resolution) based on a set of regularly spaced points and for which the values of the variables included in the final regression model were calculated. A continuous map of predicted abundance classes for each 100 m pixel was then produced using the predicted probabilities of the model. Human density was divided in quartiles as explained previously. Both maps were then combined in a map with 12 categories (the product of three vector abundance and four human host density categories) of vector-to-host ratio. Frequencies of the vector-to-host ratio categories were analyzed and compared.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Case study: dengue virus transmission in Hawai’i
  5. Data and methods
  6. Results
  7. Discussion and conclusion
  8. Acknowledgements
  9. References

Human density

Dasymetric mapping yielded higher spatial detail than census blocks. Census blocks vary hugely in size (range 0.04 ha to over 7000 ha) but most are between 1 and 10 ha. Settlements occupied a small fraction of some of the blocks. Residential areas represented 14% (21903 ha) of Oahu vs. 63% if considering the area of census blocks for the same population (Figure 4). Recreational areas covered 2% (2947 ha) of the island area. Of recreational areas, 88% (2584 ha) did not overlap with residential areas.

image

Figure 4.  Human population density (recreational and residential), Oahu, HI.

Download figure to PowerPoint

In a 2004 survey, 30% of 700 Oahu residents interviewed (QMARK 2005) went to the beach more than 150 times, i.e. around three times a week or more, 53% <150 times and 17% never over the previous 12 months. An average estimated 10% of the population visiting the beach on a weekend day is therefore reasonable.

Environmental factors and Ae. albopictus abundance

The multiple model (Table 1) included as positive and significant (P < 0.05) predictors the Shannon diversity index in a 500-m buffer, the standard deviation of the area of urban patches within a 500-m buffer and the proportion of tree vegetation in a 50-m buffer. The pseudo-R2 of the model was 0.29, indicating a good fit (Wrigley 1985). This model was used to produce a continuous mosquito abundance map. The high abundance class covered most of the study area (88%, Figure 5). Comparing the predicted abundance class to the abundance class recorded at trapping sites yielded a global accuracy of 54% and a kappa index of accuracy of 0.53, therefore in fair to good agreement with the mosquito trapping data (Kirkwood & Sterne 2003). The weighted kappa, which considers misclassifications between neighbouring classes as partial agreement, was 0.75.

Table 1.   Multiple ordinal logistic model for Aedes albopictus abundance
VariableCoefficient (P-value)
  1. SD, standard deviation.

Shannon diversity index, 500-m buffer3.4156 (0.04)
SD of urban patches areas, 500-m buffer0.0002 (<0.01)
Proportion of tree vegetation, 50-m buffer3.8055 (<0.01)
image

Figure 5.  Predicted Aedes albopictus abundance, Oahu, HI.

Download figure to PowerPoint

Vector/host map

About 17% of the areas considered (with humans and below 300 m) had high predicted Ae. albopictus abundance and low residential human population density, therefore a high vector-to-host ratio (Figure 6a; Table 2). This percentage decreased to 15% when considering the human population distribution modified for recreational use (Figure 6a, Table 2). Ninety-two percentage of recreational area cells were in the high mosquito abundance class and second lowest human density class (Table 2). Overall, areas with people mostly had high mosquito abundance (residential: 86%; recreational overall map 87%, recreational exclusively: 98%). Areas of lower mosquito abundance were mostly found within areas with higher human population densities (residential: 5%; recreational overall map 4%, recreational exclusive: <1%). In Honolulu (Figure 6b), areas such as Manoa had lower human and high mosquito density/abundance, while Kalihi had high mosquito abundance and high human density. Areas such as Kaimuki had low mosquito abundance and high human density. Beach parks were all in the medium and high mosquito abundance categories (e.g. Figure 6c: Haleiwa area beach parks), while recreational areas of low and medium mosquito abundance were mostly urban parks. Areas where infections were recorded as mapped in Effler et al. (2005) had higher vector-to-host ratios (high mosquito abundance category (residential: 89%; recreational overall map 98%) and lower human density). The coarse map does not permit a more detailed comparison.

image

Figure 6.  (a) Predicted vector-to-host ratio, Oahu, HI. Mosquito abundance classes are represented with one set of colours each. Lighter colours then indicate lower categories of human population density, thus higher vector to host density. (b) Predicted vector-to-host ratio, Honolulu, Oahu, HI. Mosquito abundance classes are represented with one set of colours each. Lighter colours then indicate lower categories of human population density, thus higher vector to host density. (c) Predicted vector-to-host ratio, Haleiwa, Oahu, HI. Mosquito abundance classes are represented with one set of colours each. Lighter colours then indicate lower categories of human population density, thus higher vector to host density.

Download figure to PowerPoint

Table 2.   Percentage area in each vector-to-host category, (a) residential population, (b) recreational population (residential map modified for the recreational use of the land), (c) recreational population excluding all other areas
Human population density quartile1234Total
(a)
 Low Aedes albopictus abundance1.260.531.101.894.78
 Medium Ae. albopictus abundance1.821.792.552.919.07
 High Ae. albopictus abundance17.2926.2923.4819.0886.14
 Total20.3728.6127.1323.88100
(b)
 Low Ae. albopictus abundance1.120.510.991.694.31
 Medium Ae. albopictus abundance1.631.712.292.608.24
 High Ae. albopictus abundance15.1534.0521.0517.1987.45
 Total17.9136.2724.3321.49100
(c)
 Low Ae. albopictus abundance00.020.040.120.18
 Medium Ae. albopictus abundance00.740.190.391.32
 High Ae. albopictus abundance0.5191.573.502.9298.50
 Total0.5192.343.733.42100

Discussion and conclusion

  1. Top of page
  2. Summary
  3. Introduction
  4. Case study: dengue virus transmission in Hawai’i
  5. Data and methods
  6. Results
  7. Discussion and conclusion
  8. Acknowledgements
  9. References

The value of disaggregating spatial data on land cover (reflecting mosquito habitat) and land use (reflecting human activities) for understanding the spatial heterogeneity in disease transmission risk was investigated. Based on empirical methods and an approach combining process-based and niche models, we produced a high-resolution map of vector-to-host ratio as an indicator of the conditional risk of exposure.

Highest human densities on Oahu are found in Honolulu, but smaller settlements occur along most of the shore, where diverse environments surround them. Populations in these areas can increase during recreational activities such as beachgoing. Protection against mosquito bites in these areas tends to be minimal, and, while exposure on open beaches would be limited, it might increase substantially under tree vegetation where people typically seek shade.

The model for Ae. albopictus was consistent with existing knowledge of its habitat and ecology (e.g. Barker et al. 2003; Vanwambeke et al. 2007b), although the role of landscape configuration has been sparsely investigated in the past. Our analysis predicted highest abundance of Ae. albopictus in places with mixed built-up areas and vegetation. On Oahu, these conditions are widespread, except in densely built-up areas of Honolulu, which currently have low Ae. albopictus abundance and where conditions might be more favourable for Ae. aegypti, a highly competent dengue vector that is currently not present on Oahu.

The difference between the vector-to-host maps indicated the relevance of investigating the distribution of human activities, which expose them to diverse environments and vector abundances. The value of spatially disaggregated information is obvious from the range of situations encountered even when only looking at exposure for residence and recreation. The results could be refined by studying mosquito bite exposure, or by localizing people for other land uses (e.g. school and work) as well as activities that vary throughout the day along with human behaviour (e.g. the activity space model of Stoddard et al. 2009). Inclusion of tourists would further improve the picture of human distribution. Also, the categorized vector-to-host ratio turns out to be delicate to interpret in between the extremes, and classes cannot be considered as a straight gradation. Using continuous variables rather than categories would improve the result substantially but would necessitate a new mosquito sampling study. The cases recorded on Oahu in 2000–2001 (Effler et al. 2005) were found in areas of lower population density and high Ae. albopictus abundance. Other relevant elements of human lifestyles (e.g. house screening, air conditioning) were not considered. However, although this was the major factor behind low dengue transmission in Texas relative to Mexico, which have similar environments (Reiter et al. 2003), a study during the last epidemic in Maui did not identify any such risk factor (Hayes et al. 2006). The cases recorded in 2000–2001 were on the more humid windward side of Oahu, which might be more favourable to mosquito survival. Although Kolivras (2006) indicated that climatic conditions for Ae. albopictus are suitable all around the island, life span is critical to meet the requirements of dengue’s extrinsic incubation period (8–14 days, Gubler et al. 2007). Mosquitoes may survive longer in the more humid windward areas and be more likely to fully incubate and disseminate dengue for longer, leading to a greater vectorial capacity.

Our study shows great potential in seeking to further parameterize mechanistic models (here, we start with the vector-to-host ratio) in a spatially explicit empirical context. This approach considered both the distributions of vector and host to predict areas with increased risk of transmission if the virus were to be reintroduced. This has the distinct benefit of a forward prediction of where transmission is more likely, without relying on retrospective epidemiological data, which do not cover all areas where a virus could spread, but only areas where transmission eventually took place. Despite being based on standard, low-resolution data sets, we produced a spatially detailed, comprehensive hypothesis representing the spatial heterogeneity of vector-to-host ratio. The approach also allowed us to look separately at land cover and land use, two facets of landscape that reflect distinct aspects of any vector-borne disease transmission cycle affecting vector and host, respectively. A major challenge with the results is the interpretation of categorized vector-to-host ratio. While a higher vector-to-host ratio is expected to lead to a higher number of bites per host and therefore a greater risk, a low vector-to-host ratio can be more difficult to interpret, as in situations where the host is sparse, vectors may have increased difficulty finding new susceptible hosts. The Hawai’i DoH Vector Control Branch data are collected for invasive species surveillance (vectors and pathogens alike), not exclusively targeting dengue or dengue vectors. Also, the data were used in a raster format, ignoring other topologies relevant to human activities, such as networks, better suited to represent movements. Recent studies highlight how movements of infected humans may overwhelmingly impact the spread in space and time of infection, including models of dengue transmission in urban networks and how variation of mosquito bite exposure risk across multiple sites will substantially increase R0 (Stoddard et al. 2009). Such studies underline the value of a spatially disaggregated approach such as ours and highlight the value of directly accounting for human movements to elucidate disease transmission risk (Adams & Kapan 2009).

Overall, our study points to the value of even relatively low-resolution spatially explicit proxies of host and vector abundance to generate preliminary predictions of disease risk from either standard empirical or more mechanistic process-based models. We combined both approaches and obtained a rich, disaggregated map that highlights areas where virus transmission between abundant mosquitoes and humans is most likely, depending on humans’ use of the land.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Case study: dengue virus transmission in Hawai’i
  5. Data and methods
  6. Results
  7. Discussion and conclusion
  8. Acknowledgements
  9. References

The authors thank Pingjun Yang from the Hawai’i Department of Health for providing the mosquito trapping data and Jon Winchester for helping with retrieving the data, providing historical references as well as comments on this manuscript. Thanks also go to Bruce Wilcox and Hilary Ranson for comments on a previous version of the manuscript. SOV was a Fulbright scholar at the University of Hawai’i at Manoa. This research was partly funded by University of Hawai’i, National Science Foundation Graduate Education and Research Traineeship Program in Ecology, Conservation and Pathogen Biology (see http://www2.jabsom.hawaii.edu/igert/) (NSF IGERT 0549514). Support also came from NSF-EPSCoR IMUA II (05546557), U.S. National Institutes of Health (P20RR018727, U54AI065359, G12RR003061) and U.S. Department of Defense (06187000).

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Case study: dengue virus transmission in Hawai’i
  5. Data and methods
  6. Results
  7. Discussion and conclusion
  8. Acknowledgements
  9. References
  • Aron JL & May RM (1982) The population dynamics of malaria. In: Population Dynamics of Infectious Diseases (ed. RMAnderson) Chapma & Hall, London, pp. 139179.
  • Adams B & Kapan DD (2009) Man bites mosquito: understanding the contribution of human movement to vector-borne disease dynamics. PLoS ONE, 4, e6763.
  • Anderson RM & May RM (1991) Infectious Diseases of Human Dynamics and Control. Oxford University Press, Oxford.
  • Barker CM, Paulson SL, Cantrell S & Davis BS (2003) Habitat preferences and phenology of Ochlerotatus triseriatus and Aedes albopictus (Diptera:Culicidae) in Southwestern Virginia. Journal of Medical Entomology, 40, 403410.
  • Bonnet DD & Worcester DJ (1946) The dispersal of Aedes albopictus in the territory of Hawaii. American Journal of Tropical Medicine and Hygiene S1–26, 465476.
  • Chung YK & Pang FY (2002) Dengue virus infection rate in field populations of female Aedes aegypti and Aedes albopictus in Singapore. Tropical Medicine & International Health, 7, 322330.
  • Effler PV, Pang L, Kitsutani P et al. (2005) Dengue fever, Hawaii, 2001–2002. Emerging Infectious Diseases, 11, 742749.
  • Eisen L & Lozano-Fuentes S (2009) Use of mapping and spatial and space-time modeling approaches in operational control of Aedes aegypti and dengue. PLoS Neglected Tropical Diseases, 3, e411.
  • Focks DA, Haile DG, Daniels E & Mount GA (1993) Dynamic life table model for Aedes aegypti (Diptera: Culicidae):Analysis of the literature and model development. Journal of Medical Entomology, 30, 10031028.
  • Gratz NG (2004) Critical review of the vector status of Ae. albopictus. Medical and veterinary entomology, 18, 215227.
  • Gubler DJ (1997) Dengue and dengue hemorrhagic fever: its history and resurgence as a global public health problem. In: Dengue and Dengue Hemorrhagic fever (eds DJGubler & GKunos) CAB International, London, UK, p. 122.
  • Gubler DJ, Kuno G & Markoff L (2007) Flaviviruses. In: Fields Virology, 5th edn. (eds DMKnipe & PMHowley) Lippincott, Williams & Wilkins, Philadelphia, p. 11531252.
  • Hartemink NA, Purse BV, Meiswinkel R et al. (2009a) Mapping the basic reproduction number (R0) for vector-borne diseases: a case study on bluetongue virus. Epidemics, 1, 153161.
  • Hartemink NA, Vanwambeke SO, Heesterbeek JAP et al. (2009b) Modelling and mapping the basic reproduction number R0 for canina leishmaniasis: a case study for a region in South West France. In: Vector-Borne Diseases: The Basic Reproduction Number R0 and Risk Maps (ed. NAHartemink) PhD Dissertation, University of Utrecht, Utrecht, p. 111138.
  • Hayes JM, Rigau-Perez JG, Reiter P et al. (2006) Risk factors for infection during a dengue-1 outbreak in Maui, Hawaii, 2001. Transactions of the Royal Society of Tropical Medicine and Hygiene, 100, 559566.
  • Herbreteau V, Salem G, Souris M & Hugot J-P (2005) Sizing up human health through remote sensing: uses and misuses. Parassitologia, 47, 6379.
  • Herbreteau V, Salem GR, Souris M, Hhugot J-P & Gonzalez J-P (2007) Thirty years of use and improvement of remote sensing, applied to epidemiology: from early promises to lasting frustration. Health & Place, 13, 400403.
  • Honorio NA, Da Silva Costa W, Leite PJ, Monteiro Gonçalves J, Lounibos LP & Lourenço-De-Oliveira R (2003) Dispersal of Aedes aegypti and Aedes albopictus (Diptera: Culicidae) in an urban endemic dengue area in the State of Rio de Janeiro, Brazil. Memorias do Instituto Oswaldo Cruz, 98, 191198.
  • Imrie A, Zhao Z, Bennett SN, Kitsutani P & Laille M (2006) Molecular epidemiology of dengue in the Pacific: introduction of two distinct strains of the dengue 2 type-1 virus into Hawaii. Annals of Tropical Medicine and Parasitology, 100, 327336.
  • Jelinek T, Dobler G & Nothdurft HD (1998) Evidence of dengue virus infection in a German couple returning from Hawaii. Journal of Travel Medicine, 5, 4445.
  • Jetten TH & Focks DA (1997) Potential changes in the distribution of dengue transmission under climate warming. American Journal of Tropical Medicine and Hygiene, 57, 285297.
  • Joyce CR (1961) Potentialities for accidental establishment of exotic mosquitoes in Hawaii. Proceedings of the Hawaiian Entomological Society, 17, 403413.
  • Kirkwood BR & Sterne JAC (2003) Essential Medical Statistics. Blackwell Science, Malden.
  • Kolivras KN (2006) Mosquito habitat and dengue risk potential in Hawaii: a conceptual framework and GIS application. The Professional Geographer, 58, 139154.
  • Larish LH & Yang P (2009) Occurrence of Aedes aegypti (Linnaeus) and Aedes albopictus (Skuse) (Diptera: Culicidae) in the Ka’u district of the Island of Hawaii. The Pan-Pacific Entomologist, 85, 169173.
  • Linard C, Lamarque P, Heyman P et al. (2007) Determinants of the geographic distribution og Puumala hantavirus and Lyme borreliosis infections in Belgium. International Journal of Health Geographics, 6, 15.
  • Linard C, Ponçon N, Fontenille D & Lambin EF (2009) Risk of malaria reemergence in southern France: testing scenarios with a multiagent simulation model. EcoHealth, 6, 135147.
  • McGarigal K & Marks BJ (1994) Fragstats: spatial pattern analysis program for quantifying landscape structure. Version 2.0.
  • Mennis J (2003) Generating surface models of population using dasymetric mapping. The Professional Geographer, 55, 3142.
  • Moffett A, Shackelford N & Sarkar S (2007) Malaria in Africa: vector specie’s niche models and relative risk maps. PLoS ONE, 9, e824.
  • Nakagawa PY & Hirst JM (1959) Current efforts in mosquito control in Hawaii. Mosquito News, 19, 6467.
  • Napier M. (2003) Application of GIS and modeling of dengue risk areas in the Hawaiian Islands. In: International Symposium on Remote Sensing of Environment. Honolulu, HI.
  • Niebylski ML & Craig GBJ (1994) Dispersal and survival of Aedes albopictus at a scrap tire yard in Missouri. Journal of the American Mosquito Control Association, 10, 339343.
  • Norris DE (2004) Mosquito-borne diseases as a consequence of land use change. EcoHealth, 1, 1924.
  • Perkins RCL (1913) Fauna Hawaiiensis or the Zoology of the Sandwich (Hawaiian Isles). Cambridge University Press, Cambridge.
  • Peterson AT (2008) Biogeography of diseases: a framework for analysis. Naturwissenschaften, 95, 483491.
  • Peterson AT, Martinez-Campos C, Nakazawa Y & Martinez-Meyer E (2005) Time-specific ecological niche modeling predicts spatial dynamics of vector. Transactions of the Royal Society of Tropical Medicine and Hygiene, 99, 647655.
  • QMARK (2005) Noneconomic value of Hawaii’s Coral Reef: results from telephone survey and focus groups. Qmark Polling & Research.
  • Reiter P, Lathrop S, Bunning M et al. (2003) Texas lifestyle limits transmission of dengue virus. Emerging Infectious Diseases, 9, 8689.
  • Rindfuss RR, Walsh SJ, Mishra V, Fox J & Dolcemascolo GP (2003) Linking household and remotely sensed data. Methodological and practical problems. In: People and the Environment. Approaches for Linking Household and Community Surveys to Remote Sensing and GIS (eds JFox, RRRindfuss, SJWalsh & VMishra) Kluwer Academic Publishers, Boston, p. 130.
  • Rogers DJ (1988) The dynamics of vector-transmitted diseases in human communities. Philosophical Transactions of the Royal Society of London B, 321, 513539.
  • Rogers DJ, Randolph SE, Snow RW & Hay SI (2002) Satellite imagery in the study and forecast of malaria. Nature, 415, 710715.
  • Smith CE, Tom T, Sasaki J, Ayers T & Effler PV (2005) Dengue risk among visitors to Hawaii during an outbreak. Emerging Infectious Diseases, 11, 750756.
  • Snow RW & Gilles HM (2002) The epidemiology of malaria. In: Essential Malariology (eds DAWarrel & HMGilles) Arnold, London, p. 85–106.
  • Stoddard ST, Morrison AC, Vazquez-Prokopec GM et al. (2009) The role of human movement in the transmission of vector-borne pathogens. PLoS Neglected Tropical Diseases, 3, e481.
  • Usinger RL (1944) Entomological phases of the recent dengue epidemic in Honolulu. Public Health Reports, 59, 423430.
  • Vanwambeke SO, Van Benthem BHB, Khantikul N et al. (2006) Multi-level analyses of spatial and temporal determinants for dengue infection. International Journal of Health Geographics, 5, 5.
  • Vanwambeke SO, Lambin EF, Eichhorn MP et al. (2007a) Impact of land-use change on dengue and malaria in Northern Thailand. EcoHealth, 4, 3751.
  • Vanwambeke SO, Somboon P, Harbach RE et al. (2007b) Landscape and land cover factors influence the presence of Aedes and Anopheles larvae. Journal of Medical Entomology, 144, 133144.
  • Vanwambeke SO, Šumilo D, Bormane A, Lambin EF & Randolph SE (2010) Landscape predictors of tick-borne encephalitis in Latvia: land cover, land use, and land ownership. Vector borne and Zoonotic Diseases, 10(5), 497506.
  • Watts DM, Burke DS, Harrison BA, Whitmire RE & Nisalak A (1987) Effect of temperature on the vector efficiency of Aedes aegypti for Dengue 2 virus. American Journal of Tropical Medicine and Hygiene, 36, 143152.
  • WHO (1997) Dengue Haemorrhagic Fever: Diagnosis, Treatment, Prevention and Control. World Health Organisation, Geneva.
  • Wilson GW (1904) Epidemic of dengue in the territory of Hawaii during 1903. Public Health Reports, 19, 6770.
  • Wrigley N. (1985) Categorical Data Analysis for Geographers and Environmental Scientists. Longman, London.