IJE Advance Access originally published online on January 4, 2006
International Journal of Epidemiology 2006 35(2):315-322; doi:10.1093/ije/dyi284
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Published by Oxford University Press on behalf of the International Epidemiological Association 2006
Article |
Towards landscape design guidelines for reducing Lyme disease risk
1 National Health and Environmental Effects Research Laboratory, Office of Research and Development, US Environmental Protection Agency, Research Triangle Park, NC, USA
2 Department of Epidemiology, School of Public Health, University of North Carolina at Chapel Hill, NC, USA
* Corresponding author. US EPA (B343-06), Research Triangle Park, NC 27711, USA. E-mail: jackson.laura{at}epa.gov
 |
Abstract |
|---|
 Top Abstract MethodsResultsDiscussionReferences |
|---|
Background Incidence of Lyme disease in the US continues to grow. Low-density development is also increasing in endemic regions, raising questions about the relationship between development pattern and disease. This study sought to model Lyme disease incidence rate using quantitative, practical metrics of regional landscape pattern. The objective was to progress towards the development of design guidelines that may help minimize known threats to human and environmental health.
Methods Ecological analysis was used to accommodate the integral landscape variables under study. Case data derived from passive surveillance reports across 12 counties in the US state of Maryland during 1996–2000; 2137 cases were spatially referenced to residential addresses. Major roads were used to delineate 514 landscape analysis units from 0.002 to 580 km2.
Results The parameter that explained the most variation in incidence rate was the percentage of land-cover edge represented by the adjacency of forest and herbaceous cover [R2 = 0.75; rate ratio = 1.34 (1.26–1.43); P < 0.0001]. Also highly significant was the percentage of the landscape in forest cover (cumulative R2 = 0.82), which exhibited a quadratic relationship with incidence rate. Modelled relationships applied throughout the range of landscape sizes.
Conclusions Results begin to provide quantitative landscape design parameters for reducing casual peridomestic contact with tick and host habitat. The final model suggests that clustered forest and herbaceous cover, as opposed to high forest-herbaceous interspersion, would minimize Lyme disease risk in low-density residential areas. Higher-density development that precludes a large percentage of forest-herbaceous edge would also limit exposure.
Keywords Borrelia burgdorferi, ecology, ecological design, GIS, Ixodes scapularis, landscape design, landscape ecology, Maryland
Accepted 11 November 2005
Lyme disease is a potentially serious, though non-lethal, illness caused by the bacterium Borrelia burgdorferi. The pathogen is transmitted during the blood meals of Ixodes (hard-shelled) ticks. The typical indicator of infection is erythema migrans, a bulls-eye rash around the locus of the bite, and is usually accompanied by flu-like symptoms including fatigue, fever, headache, and stiffness. If initial symptoms are not treated with antibiotics, infection can progress into severely debilitating musculoskeletal, cardiac, and neurological ailments.1 Lyme disease is the most common vector-borne illness in the US, with 23 763 cases reported in 2002.2 Its incidence is increasing2 and may be exacerbated by global climate change3 and low-density residential development.4–6 States in the North-east and upper Midwest report the most cases; in these areas, Ixodes scapularis is the primary vector. Case reports are also elevated in a few areas of the North-west. Here, Ixodes pacificus is responsible for human infection; however, Ixodes neotomae is the Western vector involved in maintaining B. burgdorferi in the environment.1
In north-eastern US, the massive dual forces of reforestation and suburbanization have increased human exposure to forested habitat. The occurrence of Lyme disease results in part from these land-cover changes that bring humans into proximity with the vector and its native hosts. Although I. scapularis has a broad host range,7,8 its principal host in the larval and nymphal stages is the white-footed mouse (Peromyscus leucopus);7,9,10 its preferred host as an adult is the white-tailed deer (Odocoileus virginianus).11,12 Movements of deer determine the vector's presence in the landscape,7,11,12 given a cool, moist microhabitat and the availability of hosts for immature stages.13
Both O. virginianus and P. leucopus thrive in heterogeneous landscapes created by land-cover modification. Deer require a mix of forest cover and open areas with tender vegetation14; the white-footed mouse is highly opportunistic and will inhabit forest edges as well as islands of woodland too small or inaccessible to sustain populations of other forest species.15,16 All of these habitats can be coincident with or adjacent to low-density residential areas. Research throughout north-eastern US4,17–21 suggests that Lyme disease cases are associated primarily with exposure around the home (peridomestic exposure). Widespread trends in low-density development22,23 raise the potential for increased human exposure to ticks and resulting infection in areas endemic for Lyme disease.
This study sought to quantify the role of land-cover pattern in endemic areas. We used an ecological design to capture large-scale zoonotic processes; this approach is recommended when an exposure is hypothesized to operate at a societal or other supra-individual level.24–26 Numerous studies related to Lyme disease have addressed land-cover variables on the scale of the individual residential parcel.4,5,19–21,27,28 Interactions among humans, vectors, and their hosts may indeed occur within the parcel, given suitable habitat and connectivity with adjacent land. However, the parcel unit fails to capture relevant land-cover pattern at larger scales that reflect the context of the home within broader host habitats, and also within municipal infrastructure and ordinances, and the culture of land-use allocation at the county or regional level. These factors illustrate that peridomestic exposure involves processes operating beyond the individual home parcel.
Our research focused particularly on the degree of interspersion between forested land and land that has been cleared for lawn, pasture, or other herbaceous cover. Previous studies have explored associations between forest edge and risk of B. burgdorferi infection.27–29 However, with one exception,30 they used small analysis units or small sample sizes, or they provided little quantification of the amount or configuration of forest edge associated with the dependent variable, whether vector abundance or infection; canine seroprevalence; or human case rates. Das et al.30 related the abundance of female I. scapularis on slain deer to the amount of forest edge in more than a hundred 10 km2 landscapes. Our research differs from their work by encompassing a much larger study area, addressing a broader spectrum of land-cover modification, and seeking associations with disease rate rather than vector abundance, among other characteristics.
Certainly, factors in addition to land-cover pattern are relevant in the spatial and temporal variability of B. burgdorferi infection. Abundance of I. scapularis and its hosts has been correlated with production cycles of acorns, a critical host food.31–33 Case rates have also been linked to climatic variability.34,35 Nevertheless, existing research indicates the significant role of land-cover pattern, a variable that is amenable to management.
|
Methods |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
We selected the US State of Maryland for study owing to endemic Lyme disease and land development there, as well as the availability of relevant health, environmental, and planning data. Research focused on 12 counties: Anne Arundel, Baltimore, Calvert, Carroll, Cecil, Charles, Frederick, Harford, Howard, Montgomery, Prince Georges, and St Mary's, plus the City of Baltimore (Figure 1). This contiguous area encompasses
15 000 km2 and exhibits a wide range of Lyme disease rates and development pressures from suburbanization around Washington DC, Annapolis, and Baltimore.
|
The Maryland Department of Health and Mental Hygiene provided data on 2543 Lyme disease cases reported through passive surveillance in the study area during 1996–2000. For reporting purposes, the US Centers for Disease Control and Prevention define a case as a person with erythema migrans or with at least one late-stage manifestation that is laboratory confirmed.36 Reports of such cases were available throughout the study area. However, Lyme disease is frequently under-reported.37,38 Reasons for uniform under-reporting include asymptomatic infection and empirical treatment. Ecological analysis can detect a stronger signal from uniformly weak data than a design centred on individual cases.24,25 In aggregating our case data, we compensated in part for under-reporting that may have occurred uniformly throughout the study area. Where under-reporting relates to socioeconomic status, as in failure to present,39 it is presumably less uniform. To account for differential socioeconomic status, we included median annual household income from the 2000 US Census40 as an explanatory variable.
Reported cases in the study area reflected the typical bimodal age distribution of cases reported nationally.2,41 Out of 2531 cases (99.5%) where age was reported, 699 (27.6%) were children aged 3–12. The second, smaller peak contained 573 cases (22.6%) aged 44–59. We considered restricting analysis to those cases younger than 5 or older than 65 (n = 405), since exposure away from home may be least likely for these age groups, but we decided to include cases of every age to maximize sample size. We mapped case residences in ArcGIS,42 first using a spatial database of parcel addresses from the Maryland Department of Planning and then using an electronic street atlas from TeleAtlas, Inc.43 for case addresses not found in the parcel database. We successfully mapped the residences of 2149 cases (85%). The deletion of second reports for 12 individuals resulted in 2137 cases available for spatial analysis.
Land-cover data derived from Landsat 7 Enhanced Thematic Mapper 30-m resolution satellite imagery, primarily from the year 2000.44 Initially available in 13 land-use/land-cover categories for the study area, the data were reduced to four classes of land cover: water, herbaceous, forest, and developed (
20% impervious surface). Herbaceous cover comprises half of the land area in the study, and includes pasture, row crops, and lawn. Forest covers 40% of the land area; the developed category accounts for the remaining 10%. Both forest and herbaceous cover may include low-density development (<20% impervious surface). We used the parcel database to help locate and quantify the low-density development. Residential land-use represents <15% of the forest class, while residences, together with constructed open space (e.g. parks, golf courses, airports), comprise almost one-third of the herbaceous class.
We partitioned the study area into 514 exposure landscapes using federally funded roads from the National Transportation Atlas.45 Major roads were thought to constrain most peridomestic activity and serve at least as partial barriers to deer. We made the assumption that major roads were more appropriate for delineating independent environmental analysis units than administrative or other non-physical boundaries with even less relevance to animal movement. The landscapes range in extent from 0.002 to 580 km2 (Figure 2).
|
We computed the population count and median annual household income for each landscape using area-weighted averaging of census data at the block-group level.6 We calculated the incidence rate for each landscape during 1996–2000 by using the population data as adjusted, and assuming population stability for the five study years.46 We defined incidence rate as the reported new cases per 1000 people over 5 years, expressed as cases per 5000 person-years. Cases (n = 146) located beyond the perimeter of our road-defined frame were dropped from the study, resulting in a final count of 1991 cases.
Metrics of land-cover pattern were calculated in Fragstats 3.3.47 They included landscape area; area and percentage of landscape in forest, herbaceous, and developed cover types; number of forest patches <2 ha; length of edge around patches of each cover type and across the total landscape; and an edge-contrast index that captured the percentage of all edge length between land-cover types that represents forest-herbaceous adjacency. We summarize the calculation of this index as follows:
 |
We evaluated statistical relationships with regression analysis using S-plus 6.148 with the MASS library module.49 We used a negative binomial generalized linear model to accommodate overdispersion in the data.50 As the traditional R2 statistic is inappropriate for negative binomial models,51,52 we used an alternate calculation. The method we chose is based on the model overdispersion parameter (
) instead of differences in sums of squares. While not directly comparable, it exhibits similar key traits such as (i) returning values between zero and one and (ii) consistently quantifying improvements in fit when variables are added in any order to a model.51,52 We also used the Akaike information criterion (AIC) to evaluate goodness-of-fit.50,53,54 Criteria for selecting the final model included (i) highest R2 value, (ii) lowest AIC value, (iii) coefficient correlations <0.4, and (iv) 95% of residuals between –2.0 and 2.0.
|
Results |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Reported Lyme disease cases across the 514 exposure landscapes for the combined 5 year study period resulted in a range of incidence rates from 0 to 13 cases per 5000 person-years. All but four landscapes exhibited ranges from 0 to 3.5 cases per 5000 person-years. No cases were reported in 317 landscapes (62%). No cases were reported in landscapes <0.3 km2 (n = 150). Cases were observed in all landscapes >35 km2.
The land-cover characteristic that explained the most variability in Lyme disease incidence rate was the edge-contrast index—the relative measure of forest-herbaceous edge. The relationship was log-linear; in the final model, every 10% increase in the value of the index corresponded to a 34% increase in incidence rate (Table 1).
|
Adding percentage forest cover improved the fit of the model, reducing AIC from 1324 to 1273 and increasing R2 from 0.75 to 0.82 (Table 1). Percentage forest exhibited a quadratic relationship with incidence rate, with the latter peaking at 53.5% forest in the final model. Percentage herbaceous cover also exhibited a quadratic relationship with incidence rate. However, this variable was highly negatively correlated with percentage forest and had lower explanatory power, so it was not included in the final model.
The inclusion of median annual household income resulted in our strongest model, further reducing AIC to 1257 and increasing R2 to 0.85 (Table 1). The relationship of this variable to incidence rate was quadratic as well, with rate peaking in the final model at 80 000 USD per year.
We found a weak positive relationship between incidence rate and landscape area [rate ratio (RR) = 1.080 (1.049–1.117), P < 0.05; units in 10 km2]. The relationship became weakly negative [RR = 0.988 (0.978–0.999), P < 0.05] when we evaluated this variable with those presented above. Similarly, we found a weak positive relationship between incidence rate and number of forest patches <2 ha [RR = 1.017 (1.010–1.024), P < 0.05; units in tens of patches]. This relationship also turned weakly negative [RR = 0.997 (0.994–0.999), P < 0.05] when we evaluated this variable with those in our best model. Consequently, we dropped both variables from further consideration.
In the final model, 95% of the deviance and Pearson residuals fell between –2.0 and 2.0. Pearson residuals beyond this range represented under-estimations exclusively; their spatial distribution did not suggest clustering or other patterns. The largest residuals occurred in landscapes between 0.3 and 10 km2. Our model accurately described incidence rate in all landscapes <0.3 km2. All coefficient correlations (except between terms and their square) were <0.32.
|
Discussion |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Half-forested landscapes with a large percentage of forest-herbaceous edge are statistically associated with the highest Lyme disease rates in our study area. These findings begin to quantify parameters for landscape design that may reduce B. burgdorferi exposure. Figure 3 presents two hypothetical landscapes with 50% forest cover, but different amounts of forest-herbaceous interspersion. Of the two, the landscape on the right would offer more sheltered foraging opportunities for O. virginianus and P. leucopus, which transport I. scapularis. This same landscape would also provide greater peridomestic access to forest habitat for low-density suburban and rural residents whose parcels are dominated by lawn or other herbaceous cover. By increasing the proximity of these residents to forest hosts, this landscape configuration would facilitate casual peridomestic exposure. Alternatives to this facultative design are landscapes with highly clustered forest and herbaceous cover (Figure 3, at left), as well as landscapes with sufficient higher-density development to preclude a large percentage of forest-herbaceous edge.
|
Notably, our model applies to landscapes that range over several orders of spatial magnitude. Such stability is somewhat unusual in geographic research55–57 and may reflect our use of environmentally meaningful analysis units to capture host habitat and human contact with it.58,59 A multi-scale design is recommended in landscape analyses to determine the sensitivity of land-cover metrics to environmental processes under study.60 With such a design, scale itself must be included as a possible covariate in any models.60 Our results indicate that landscape area, independent of land-cover attributes, is not significant in Lyme disease rate.
Comparisons with previous studies
Das et al.30 also found their response variable to be significantly positively correlated with percentage forest edge. However, they reported negative correlations with percentage forest, agricultural, and low-density residential land instead of quadratic relationships. One explanation may be that they studied hunting range, rather than a broad spectrum of land-cover modification. Additionally, their analysis units were uniformly set to 10 km2, precluding an assessment of the importance of spatial scale to the results.
We were not able to corroborate preliminary findings by Allan et al.61 that landscapes with patches <2 ha pose a greater risk for Lyme disease. Their reasoning was that small forest patches support a lower diversity of host species and thus are dominated by P. leucopus, the primary reservoir for B. burgdorferi. However, the significance of forest edge in our study may be consistent with this diversity hypothesis. Many disturbance-tolerant forest species exhibit some competence for B. burgdorferi.62 Non-competent hosts include disturbance-intolerant songbirds,62 which are absent in edge habitat and therefore unavailable to serve in a zooprophylactic capacity.
Evaluation of model fit
The low values for 95% of our residuals indicate a balanced influence of individual observations.63 Our model closely fit the observed rates in landscapes of all sizes. Over-estimations were small in magnitude, suggesting that the income variable adjusted for demographic non-uniformity in case reporting.
We propose that the largest residual values representing under-estimations are a function of our variable spatial delineation of the peridomestic environment. For any exposure that did occur away from home, it is more likely also to have taken place outside the home landscape if the home landscape was small. In contrast, cases reported in large landscapes could have resulted from peridomestic exposure or potentially from activities away from home but still within the home landscape. This broad definition of ‘home’ in large landscapes could explain why the model exhibits very few large under-estimations in analysis units >10 km2; these may be sufficiently large to encompass peridomestic as well as other types of exposures.
There are two plausible components to the high model accuracy observed in landscapes <0.3 km2, for which no cases were reported: (i) they are primarily urbanized and contain little or no exposure habitat and (ii) urban populations are more likely to be of low socioeconomic status with less opportunity for exposure away from home.
Limitations and implications
The strength of our model may be affected by our reliance on remote land-cover imagery, use of area-weighting in the spatial re-distribution of census data, and assumptions about major roads as barriers to deer and human peridomestic activity. Analysis did not include an evaluation of microhabitat, despite its importance to vector survival.5,30,64,65 Instead, our ecological approach relied on hierarchy theory66–68 to subsume localized factors within macro-scale variables. We also used 2000 land-cover data to describe habitat potentially associated with exposure that occurred during 1996–2000. As a result, our characterization of the residential landscape may not apply equally to all cases reported during the study period. These potential sources of error, plus the roles of plant productivity31–33 and climate34,35 in the variability of Lyme disease rate and vector abundance, would increase the overall random error in our model. Therefore, the actual relationship between Lyme disease and land-cover pattern may be even stronger than what we found.
Our model does not distinguish sub-classes of herbaceous cover. Edge between forest and lawn is the exposure equivalent to edge between forest and pasture, for example. Any preliminary inference for landscape management about the forest-herbaceous edge would equate agricultural and low-density developed land uses. From an environmental perspective, forests bordering either of these land uses are attractive to opportunistic host species. More people are likely to encounter forest habitat that borders low-density development, which is increasing across the study area, than to encounter the forested edge of agricultural fields. Therefore, our findings suggest that landscape design to minimize exposure in the study area would involve managing the extension of low-density development into forested areas. Model validation with more recent data is necessary to strengthen this inference; data from other endemic areas would be necessary to extrapolate findings geographically.
Concern about the role of low-density development in human illness is not restricted to Lyme disease.69 Babesiosis and human anaplasmosis, both Ixodes-borne, are also on the rise.70,71 The spread of residential areas into previously isolated wildlife habitat may have contributed to the emergence of Hanta virus in the US South-west.72 The Biodiversity Project73 has suggested that development-associated destruction of habitat for mosquito predators such as fish, reptiles, and bats has exacerbated mosquito-associated encephalitis. A growing body of literature indicates that low-density development also adversely affects physical fitness, pedestrian safety, accessibility for the elderly and disabled, community vibrancy, and even mental health.74–77
High-density development can certainly carry health risks, particularly those from air and water pollution. However, low-density development has become the US standard, so its adverse effects on human health are increasing. Furthermore, low-density development requires more land, pavement, and other resources per capita than traditional development patterns.78 Its environmental effects are numerous and well-known; they include increases in polluted runoff, erosion, both flooding and drought, and the extirpation of disturbance-sensitive wildlife.79–81
Previous researchers4,19,82 have suggested domestic precautions for diminishing Lyme disease risk in low-density forested residential areas. Others6,61 have recognized that risk reduction may be most effective at the community level. Clustered development and moderate residential densities would minimize forest fragmentation and the resulting forest-herbaceous edge habitat that is strongly implicated in exposure. Our findings begin to quantify thresholds of deforestation where we observe increased incidence rates and propose a landscape unit for planning development that preserves forest interior relative to forest edge. Designing the built environment for improved public health is a strategy that dates to the origins of the public health profession. Despite having faded from consideration, the built environment may again represent a powerful resource not only for reducing Lyme disease, but zoonotic diseases in general, as well as additional threats to both human and environmental health.
|
Acknowledgments |
|---|
We thank Karon Damewood and Dale Rohn of the Maryland Department of Health and Mental Hygiene; Rich Hall and Lynda Eisenberg of the Maryland Department of Planning; and Jim Wickham, Vasu Kilaru, and Tim Wade, of the US EPA, for providing essential datasets. Address-matching was performed by Lockheed Martin Services, Inc. under US EPA contract #68-W7-0055.
This study was funded by the US Environmental Protection Agency. A previous version of this article was reviewed by the National Health and Environmental Effects Research Laboratory and approved for publication. Approval does not signify that the contents reflect the views of the Agency, nor does mention of trade names or commercial products constitute endorsement or recommendation for use.
|
References |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
1 Steere AC. Lyme disease. N Engl J Med 2001;345:115–25.
2 Centers for Disease Control and Prevention. Lyme Disease—United States, 2001–2002. MMWR Morb Mortal Wkly Rep 2004;53:365–69.[Medline]
3 Estrada-Pena A. Increasing habitat suitability in the United States for the tick that transmits Lyme disease: a remote sensing approach. Environ Health Perspect 2002;110:635–40.[ISI][Medline]
4 Maupin GO, Fish D, Zultowsky J, Campos EG, Piesman J. Landscape ecology of Lyme disease in a residential area of Westchester County, New York. Am J Epidemiol 1991;133:1105–13.
5 Glass GE, Schwartz BS, Morgan JM III, Johnson DT, Noy PM, Israel E. Environmental risk factors for Lyme disease identified with geographic information systems. Am J Public Health 1995;85:944–48.
6 Cromley EK, Cartter ML, Mrozinski RD, Ertel S-H. Residential setting as a risk factor for Lyme disease in a hyperendemic region. Am J Epidemiol 1998;147:472–77.
7 Piesman J, Spielman A. Host-associations and seasonal abundance of immature Ixodes dammini in southeastern Massachusetts. Ann Entomol Soc Am 1979;72:829–32.
8 Main AJ, Carey AB, Carey MG, Goodwin RH. Immature Ixodes dammini (Acari: Ixodidae) on small mammals in Connecticut, USA. J Med Entomol 1982;19:655–64.[ISI][Medline]
9 Spielman A, Levine JF, Wilson ML. Vectorial capacity of North American Ixodes ticks. Yale J Biol Med 1984;57:507–13.[ISI][Medline]
10 Adler GH, Telford SR III, Wilson ML, Spielman A. Vegetation structure influences the burden of immature Ixodes dammini on its main host, Peromyscus leucopus. Parasitology 1992;105:105–10.
11 Spielman A, Wilson ML, Levine JF, Piesman J. Ecology of Ixodes dammini-borne human babesiosis and Lyme disease. Annu Rev Entomol 1985;30:439–60.[ISI][Medline]
12 Wilson ML, Telford SR III, Piesman J, Spielman A. Reduced abundance of immature Ixodes dammini (Acari: Ixodidae) following elimination of deer. J Med Entomol 1988;25:224–28.[ISI][Medline]
13 Cortinas MR, Guerra MA, Jones CJ, Kitron U. Detection, characterization, and prediction of tick-borne disease foci. Int J Med Microbiol 2002;291(Suppl. 33):11–20.
14 Halls LK, ed. White-Tailed Deer: Ecology and Management. Harrisburg, PA: Stackpole Books, 1984.
15 Nupp TE, Swihart RK. Effects of forest fragmentation on population attributes of white-footed mice and eastern chipmunks. J Mammal 1998;79:1234–43.[CrossRef]
16 Krohne DT, Hoch GA. Demography of Peromyscus leucopus populations on habitat patches: the role of dispersal. Can J Zool 1999;77:1247–53.[CrossRef][ISI]
17 Falco RC, Fish D. Prevalence of Ixodes dammini near the homes of Lyme disease patients in Westchester County, New York. Am J Epidemiol 1988;127:826–30.
18 Nicholson MC, Mather TN. Methods for evaluating Lyme disease risks using geographic information systems and geospatial analysis. J Med Entomol 1996;33:711–20.[ISI][Medline]
19 Klein JD, Eppes SC, Hunt P. Environmental and life-style risk factors for Lyme disease in children. Clin Pediatr (Phila) 1996;35:359–63.
20 Orloski KA, Campbell GL, Genese CA et al. Emergence of Lyme disease in Hunterdon County, New Jersey, 1993: a case–control study of risk factors and evaluation of reporting patterns. Am J Epidemiol 1998;147:391–97.
21 Smith G, Wileyto EP, Hopkins RB, Cherry BR, Maher JP. Risk factors of Lyme disease in Chester County, Pennsylvania. Public Health Rep 2001;116 (Suppl. 1):146–56.[CrossRef][ISI][Medline]
22 Fulton W, Pendall R, Nguyen M, Harrison A. Who Sprawls Most? How Growth Patterns Differ Across the US. Washington, DC: The Brookings Institution, 2001. Available at: http://www.brookings.edu/dybdocroot/es/urban/publications/fulton.pdf (Accessed April 8, 2005).
23 US Environmental Protection Agency. Our Built and Natural Environments (EPA/231/R-01/002). Washington, DC: US Environmental Protection Agency, 2001.
24 Susser M. The logic in ecological: I. The logic of analysis. Am J Public Health 1994;84:825–29.
25 Susser M. The logic in ecological: II. The logic of design. Am J Public Health 1994;84:830–35.
26 Pekkanen J, Pearce N. Environmental epidemiology: challenges and opportunities. Environ Health Perspect 2001;109:1–5.[ISI][Medline]
27 Dister SW, Fish D, Bros SM, Frank DH, Wood BL. Landscape characterization of peridomestic risk for Lyme disease using satellite imagery. Am J Trop Med Hyg 1997;57:687–92.
28 Frank DH, Fish D, Moy FH. Landscape features associated with Lyme disease risk in a suburban residential environment. Landsc Ecol 1998;13:27–36.
29 Daniel M, Kolar J, Zeman P, Pavelka K, Sadlo J. Tick-borne encephalitis and Lyme borreliosis: comparison of habitat risk assessments using satellite data. Cent Eur J Public Health 1999;7:35–39.[Medline]
30 Das A, Lele SR, Glass GE, Shields T, Patz J. Modelling a discrete spatial response using generalized linear mixed models: application to Lyme disease vectors. Int J Geog Inf Sci 2002;16:151–66.[CrossRef]
31 Ostfeld RS, Jones CG, Wolff JO. Of mice and mast: ecological connections in eastern deciduous forests. Bioscience 1996;46:323–30.[CrossRef][ISI]
32 Ostfeld RS. The ecology of Lyme disease risk. Am Sci 1997;85:338–46.
33 Jones CG, Ostfeld RS, Richard MP, Schauber EM, Wolff JO. Chain reactions linking acorns to gypsy moth outbreaks and Lyme disease risk. Science 1998;279:1023–26.
34 Subak S. Effects of climate on variability in Lyme disease incidence in the northeastern United States. Am J Epidemiol 2003;157:531–38.
35 McCabe GJ, Bunnell JE. Precipitation and the occurrence of Lyme disease in the northeastern United States. Vector Borne Zoonotic Dis 2004;4:143–48.[CrossRef][ISI][Medline]
36 Centers for Disease Control and Prevention. Case definitions for infectious conditions under public health surveillance. MMWR Morbid Mortal Wkly Rep 1997;46(RR-10):1–55.[Medline]
37 Coyle BS, Strickland GT, Liang YY, Pena S, McCarter R, Israel E. The public health impact of Lyme disease in Maryland. J Infect Dis 1996;173:1260–62.[ISI][Medline]
38 Meek JI, Roberts CL, Smith EV Jr, Carrter ML. Underreporting of Lyme disease by Connecticut physicians, 1992. J Public Health Manag Pract 1996;2:61–65.[Medline]
39 Fiscella K, Franks P, Gold MR, Clancy CM. Inequality in quality: addressing socioeconomic, racial, and ethnic disparities in health care. JAMA 2000;283:2579–84.
40 US Census 2000 Long Form (SF-3) Data. East Brunswick, NJ: Geolytics, Inc.
41 Centers for Disease Control and Prevention. Lyme Disease—United States, 2000. MMWR Morbid Mortal Wkly Rep 2002;51:29–31.[Medline]
42 ArcGIS Desktop for Windows, Version 8.3. Redlands, CA: Environmental Systems Research Institute Inc, 2002.
43 Maryland Street Coverage. Menlo Park, CA: TeleAtlas Inc, 2001.
44 Homer CG, Huang C, Yang L, Wylie B, Coan M. Development of a 2001 National Land-Cover Database for the United States. Photogramm Eng Rem S 2004;70:829–40.
45 Federal Highway Administration. National Highway Planning Network, Version 3.0. Washington, DC: Bureau of Transportation Statistics, 2001.
46 Kelinbaum DG, Kupper LL, Margenstern H. Epidemiologic Research: Principles and Quantitative Methods. Belmont, CA: Lifetime Learning Publications, 1982.
47 McGarigal K. Fragstats Spatial Pattern Analysis Program for Quantifying Landscape Structure, Version 3.3. Amherst, MA, 2002.
48 S-plus for Windows, Version 6.1. Seattle, WA: Insightful Corp, 2002.
49 Venables WN, Ripley BD. Modern Applied Statistics with S. New York: Springer, 2002.
50 Crawley MJ. Statistical Computing: An Introduction to Data Analysis Using S-Plus. West Sussex, England: John Wiley and Sons Ltd,2002.
51 Miaou S-P. Measuring the Goodness-of-Fit of Accident Prediction Models (FHWA-RD-96-040). Washington, DC: US Department of Transportation, Federal Highway Administration, 1996.
52 Vogt A, Bared JG. Accident Models for Two-Lane Rural Roads: Segments and Intersections (FHWA-RD-98-133). Washington, DC: US Department of Transportation, Federal Highway Administration, 1998.
53 Burnham KP, Anderson DR. Model Selection and Multimodel Inference. New York: Springer-Verlag, 2002.
54 Hoffman JP. Generalized Linear Models: An Applied Approach. Boston: Pearson Education Inc., 2004.
55 Openshaw S, Taylor PJ. The modifiable areal unit problem. In: Wrigley N, Bennett R (eds). Quantitative Geography: A British View. London: Routledge and Kegan Paul, 1981, pp. 60–69.
56 Meentemeyer V. Geographical perspectives of space, time, and scale. Landsc Ecol 1989;3:163–73.
57 Meade MS, Earickson RJ. Medical Geography. New York: Guilford Press, 2000.
58 Wiens JA. Spatial scaling in ecology. Funct Ecol 1989;3:385–97.
59 Jelinski DE, Wu J. The modifiable areal unit problem and implications for landscape ecology. Landsc Ecol 1996;11:129–40.
60 Li H, Wu J. Use and misuse of landscape indices. Landsc Ecol 2004;19:389–99.[CrossRef]
61 Allan BF, Keesing F, Ostfeld RS. Effect of forest fragmentation on Lyme disease risk. Conserv Biol 2003;17:267–72.[CrossRef]
62 Ostfeld RS, Keesing F. The function of biodiversity in the ecology of vector-borne zoonotic diseases. Can J Zool 2000;78:2061–78.[CrossRef][ISI]
63 Hosmer DW, Lemeshow S. Applied Logistic Regression. New York: Wiley, 1989.
64 Glass GE, Amerasinghe FP, Morgan JM III, Scott TW. Predicting Ixodes scapularis abundance on white-tailed deer using geographic information systems. Am J Trop Med Hyg 1994;51:538–44.
65 Bunnell JE, Price SD, Das A, Shields TM, Glass GE. Geographic information systems and spatial analysis of adult Ixodes scapularis (Acari: Ixodidae) in the middle Atlantic region of the USA. J Med Entomol 2003;40:570–76.[ISI][Medline]
66 Allen TFH, Starr TB. Hierarchy: Perspectives for Ecological Complexity. Chicago: University of Chicago Press, 1982.
67 O'Neill RV, DeAngelis DL, Waide JB, Allen TFH. A Hierarchical Concept of the Ecosystem. Princeton: Princeton University Press, 1986.
68 Urban DL, O'Neill RV, Shugart HH Jr. Landscape ecology: a hierarchical perspective can help scientists understand spatial patterns. Bioscience 1987;37:119–27.
69 Patz AP, Daszak P, Tabor GM et al. Unhealthy landscapes: policy recommendations on land use change and infectious disease emergence. Environ Health Perspect 2004;112:1092–98.[ISI][Medline]
70 Varde S, Beckley J, Schwartz I. Prevalence of tick-borne pathogens in Ixodes scapularis in a rural New Jersey county. Emerg Infect Dis 1998;4:97–99.[ISI][Medline]
71 Ijdo JW, Meek JI, Cartter ML et al. The emergence of another tickborne infection in the 12-town area around Lyme, Connecticut: human granulocytic ehrlichiosis. J Infect Dis 2000;181:1388–93.[CrossRef][ISI][Medline]
72 Hill K. Design and planning as healing arts: the broader context of health and environment. In: Johnson BR, Hill K (eds). Ecology and Design: Frameworks for Learning. Washington, DC: Island Press, 2002, pp. 203–14.
73 Biodiversity Project. Getting on message: making the biodiversity–sprawl connection. December, 2000. Available at: http://www.biodiversityproject.org/messagekit.htm (Accessed April 8, 2005).
74 Surface Transportation Policy Project. Aggressive driving: are you at risk? April, 1999. Available at: http://www.transact.org/report.asp?id=56 (Accessed April 8, 2005).
75 Ewing R, Schmid T, Killingsworth R, Zlot A, Raudenbush S. Relationship between urban sprawl and physical activity, obesity, and morbidity. Am J Health Promot 2003;18:47–57.[ISI][Medline]
76 Jackson LE. The relationship of urban design to human health and condition. Landsc Urban Plan 2003;64:191–200.[CrossRef]
77 Sturm R, Cohen DA. Suburban sprawl and physical and mental health. Public Health 2004;118:488–96.[CrossRef][ISI][Medline]
78 Wackernagel M, Rees WE. Our Ecological Footprint: Reducing Human Impact on the Earth. Philadelphia: New Society Publishers, 1996.
79 Jones CR, Clark CC. Impact of watershed urbanization on stream insect communities. Water Resour Bull 1987;23:1047–55.
80 Todd DA, Bedient PB, Haasbeek JF, Noell J. Impact of land use and NPS loads on lake quality. J Environ Eng 1989;115:633–49.
81 Schueler TR. The importance of imperviousness. Watershed Prot Tech 1994;1:100–11.
82 Ward SE, Brown RD. A framework for incorporating the prevention of Lyme disease transmission into the landscape planning and design process. Landsc Urban Plan 2004;66:91–106.[CrossRef]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  G. DAVEY SMITH Cultural climate, physical climate, life, and death Int. J. Epidemiol., April 1, 2006; 35(2): 211 - 212. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||