Urban Tree Canopy and Asthma, Wheeze, Rhinitis, and Allergic Sensitization to Tree Pollen in a New York City Birth Cohort

1In the United States, a disproportionate burden of severe childhood asthma affects families in poverty, racial or ethnic minorities, and inner-city communities. Recent evidence points to intraurban variability in exposure to air pollution, allergens, environmental tobacco exposure, lifestyles (e.g., sedentary behavior, diet, obesity), and aspects of the social environment that may affect the incidence of asthma or allergic sensitization, a key asthma risk factor.

2Urban landscape elements, particularly trees, have the potential to affect air flow, air quality, and the production of aeroallergens. Air quality and respiratory health improvements are among the anticipated economic and social benefits of the urban forest. Trees may cause small reductions in particulate matter and ozone concentrations, though evidence directly linking trees to neighborhood air quality is limited. However, one study reported that more vegetation within a 100-m or 250-m buffer around the home was associated with lower personal exposure to particulate matter ≤ 2.5 µm (PM2.5). Some tree species also produce allergens, which may trigger seasonal allergic rhinoconjunctivitis or asthma exacerbations in sensitized individuals. Short-term variation in pollen concentration has been associated with allergy medication purchases, asthma symptoms, and asthma-related emergency department visits.

3The role of the urban forest in asthma and allergy prevalence warrants further investigation, especially as governments and organizations in New York City and around the world invest substantial resources toward large-scale tree planting efforts. We use data from a birth cohort based in New York City to examine the relationship of total urban tree canopy coverage with childhood asthma, wheeze, rhinitis, and allergic sensitization at the individual level.

4Study population: the CCCEH birth cohort in New York City. The Columbia Center for Children’s Environmental Health (CCCEH) birth cohort recruited a convenience sample of pregnant women through prenatal clinics. All participants were African American or Dominican and lived in economically disadvantaged areas of New York City (Northern Manhattan and the Bronx). Pregnant women with HIV were excluded. The 727 cohort births occurred during 1998–2006. Children have been followed up through a series of clinical visits for neurodevelopmental, growth, respiratory, and other outcomes. The number of individuals assessed for each outcome depended on loss to follow-up and on the correspondence between data collection efforts for fixed calendar years and ages of the cohort children. Women provided written informed consent at each visit, children also provided assent if they were ≥ 7 years of age, and all study procedures were approved by the Columbia University Medical Center Institutional Review Board.

5Outcome assessment. Asthma, wheeze, and rhinitis were assessed by parental report collected using the previously validated Brief Respiratory Questionnaire (BRQ) and International Study of Asthma and Allergies in Childhood (ISAAC) questionnaire. The BRQ asks whether the child has ever been diagnosed by a physician as having asthma. The ISAAC questionnaire asks about wheezing or whistling in the chest during the preceding 12 months, and whether the child has ever had “hay fever,” the lay term used for rhinitis in the ISAAC questionnaire. Questionnaires were completed at approximately 5 years of age (mean ± SD = 60.0 ± 3.7 months) and 7 years (mean± SD = 84.1 2.7 months).

6The study visit at 7 years of age included serum IgE antibody testing. A positive IgE test may indicate allergic sensitization of clinical importance, but may also be present in asymptomatic individuals. Assessment of IgE was conducted with ImmunoCAP® (Phadia, Uppsala, Sweden) for nine specific allergens: German cockroach, mouse urine proteins, dust mites (Dermatophagoides farinae), cat dander, dog dander, mold, common ragweed, mixed grass pollen (Gx2), and mixed tree pollen (Tx1). The mixed tree pollen used for IgE testing included the following species: Acer negundo (boxelder), Betula verrucosa (European white birch), Juglans californica (California black walnut), Quercus alba (white oak), and Ulmus americana (American elm). Children were classified as sensitized to an allergen based on a specific IgE antibody response ≥ 0.35 IU/mL against the allergen.

7Exposure assessment: urban tree canopy coverage. A detailed and highly accurate characterization of the urban tree canopy was conducted for the years 2001 and 2010. The 2010 tree canopy was mapped and used as the basis for estimating the distribution of tree canopy in 2001, which has a better temporal correspondence with the prenatal addresses. The 2010 tree canopy data layer was developed using an automated object-based image analysis approach in eCognition Developer 8.64 software (Trimble, Westminster, CO) that combined high-resolution Light Detection and Ranging (LiDAR) data from an April 2010 flight over New York City, color infrared aerial imagery from spring 2008 (Sanborn Map Company, Colorado Springs, CO), and ancillary vector data from the New York City Department of Parks and Recreation, Department of City Planning, and Department of Information Technology and Telecommunication; New York State Department of Conservation; and U.S. Fish and Wildlife Service; the data derived from these sources are described in detail by. The automated approach was followed by a manual review of the data as described previously. Color infrared summer (leaf-on) and true color spring (primarily leaf-off) high-resolution aerial imagery from the year 2001 (New York City Department of Information Technology and Telecommunication, unpublished data) was used as the basis for manually modifying the 2010 tree canopy layer to 2001 conditions, once again at a scale of 1:1,000. There was a high level of agreement using the 2001 or 2010 versions of the tree canopy layer (correlation = 0.98 for 0.25-km buffers around prenatal addresses). Tree canopy coverage was defined as the estimated percent of land area covered by tree canopy.

8Neighborhood definition and characteristics using GIS (geographic information system). We focused primarily on prenatal address (reported at the time of recruitment during the third trimester of pregnancy and available for 99.6% of the cohort) to estimate exposures before the onset of the health outcomes. Geocoding was done using Geosupport software, a highly accurate parcel-based address matching program developed by the New York City Department of City Planning. Circular buffers with a radius of 0.25 km were created as the primary neighborhood definition to capture small-scale spatial variation in tree canopy coverage that may be relevant for personal air pollution and aeroallergen exposures. A spatial overlay function within ArcGIS version 10.0 (ESRI, Redlands, CA) was used to intersect these buffers with the tree canopy layer and other geographies used to measure potential confounders: park coverage, traffic volume, and residential composition.

9Park coverage was defined as the percent of land area covered by park land based on 2008 data from the New York City Department of Parks and Recreation. Average daily traffic volume was estimated by combining U.S. Census Bureau Feature Class Codes with street network data assembled in 2002 by TeleAtlas from federal, state, county, and city departments (TeleAtlas North America Inc., Lebanon, NH). Census block group data from the 2000 U.S. Census were used to measure neighborhood demographic and socioeconomic characteristics (including percent poverty, defined as the percentage of residents with incomes below the federal poverty line), with block group–level data aggregated to neighborhood buffers using aerial weighting interpolation.

10For most in this cohort, the address provided during the third trimester of pregnancy was different from the geocoded address at 5 or 7 years of age; 62% had moved to a new address before their age 5 visit, and 67% had moved to a new address before their age 7 visit. Therefore, alternate neighborhood specifications for sensitivity analyses included estimated tree canopy exposures based on addresses reported at age 5 or 7 years instead of the prenatal address as well as an alternate geographic scale (using larger 1.0-km buffers). Tree canopy coverage and other neighborhood characteristics for larger buffers or in buffers constructed around the age 5 and age 7 addresses were defined using methods parallel to those for the main analysis, except that tree canopy estimates for 2010 (rather than 2001) were used for buffers surrounding age 5 and age 7 addresses.

11Covariates collected via survey and personal monitoring. During the third trimester of pregnancy, expectant mothers were interviewed to assess sociodemographic characteristics including Medicaid enrollment and ethnicity. Maternal asthma history was assessed by self-report of a physician diagnosis. Environmental tobacco smoke exposure was classified based on the mother’s report of a smoker in the household during pregnancy. Mothers were excluded from the cohort study if they reported active smoking during an initial screening interview. However, mothers who were enrolled and subsequently determined to be active smokers during pregnancy (based on subsequent self-report, medical record information, or cord blood cotinine > 25 ng/mL) were included in the current analyses and classified as active smokers during pregnancy. Maternal ambient airborne polycyclic aromatic hydrocarbon (PAH) exposure during the third trimester was classified as high if the personal air monitor levels were > 2.26 g/m3.

12Statistical analyses. Analyses were conducted in Stata 12.0 (StataCorp, College Station, TX), with p < 0.05 interpreted as statistical significance. Regression models used robust standard errors that account for clustering of observations within community districts (59 named areas of New York City, governed by community boards). Relative prevalences [referred to as relative risks (RR) for simplicity] were estimated using a log link and Poisson working variance distribution. Percent tree canopy coverage in the residential neighborhood was rescaled to have an SD of 1. RRs presented in the tables can be interpreted as the prevalence ratio comparing children whose prenatal neighborhoods differed by 1 SD in tree canopy coverage. Covariates included sex, age at the time of outcome measurement, ethnicity, maternal asthma, previous birth, other previous pregnancy, Medicaid enrollment, tobacco smoke in the home, active maternal smoking, and the following characteristics of 0.25-km buffers: population density, percent poverty, percent park land, and estimated traffic volume.

13Covariate data were missing for maternal asthma in 30% of participants, and small numbers (< 1%) were missing data for tobacco smoke in the home and Medicaid enrollment. Although our main analytic strategy was a complete cases analysis (i.e., with observations that had missing data for any covariate excluded), we conducted sensitivity analyses with missing covariate data imputed using multiple imputation or with potential bias due to incomplete follow-up addressed through inverse probability weighting. Input variables for both methods included canopy coverage, all neighborhood covariates and individual characteristics included as model covariates, plus high prenatal PAH exposure, which may have predicted loss to follow-up in previous cohort analyses. Multiple imputation also used outcome data, and was implemented using chained equations through the ice command with parameter estimates combined across 20 imputed data sets using mi estimate in Stata. Inverse probability weights were estimated for each outcome using logistic regression models predicting successful follow-up and data collection for age 5 questionnaire data, age 7 questionnaire data, and age 7 IgE testing. Inverse probability weights ranged from 1.1 to 8.3. Models were evaluated using a Hosmer–Lemeshow test and visual inspection across deciles, both of which suggested good fit to the data (data not shown).

14Of 727 participants enrolled, 549 participants had outcome information available based on questionnaire data at 5 (n = 492) or 7 years of age (n = 427), or based on IgE testing at age 7 (n = 288) (Table 1). The prevalence of maternally reported physician-diagnosed asthma was 28% at age 5 and 36% at age 7. Of the children who completed IgE testing, 45% had allergic sensitization to one or more specific allergens. The most common allergic sensitizations were to German cockroach (31%) and mixed tree pollen (19%).

15The 0.25-km neighborhood buffers around the prenatal addresses (Figure 1) were characterized by 2–41% canopy coverage, with a mean ± SD of 15 ± 8%. Tree canopy coverage was negatively correlated with population density, and positively correlated with percent poverty, percent black race, daily traffic volume, and percent of area covered by park land (Table 2).

16Proportions of children with asthma, wheeze, or rhinitis did not appear to be related to quartiles of tree canopy coverage near the prenatal address (Table 3). Regression models adjusted for covariates indicated a significant positive association of tree canopy coverage with diagnosed asthma at 7 years of age (adjusted RR = 1.17; 95% CI: 1.02, 1.33) consistent with a 17% increase in the prevalence of asthma with each SD increase in tree canopy coverage (for tree canopy coverage, the SD was 8%). The association was similar but nonsignificant based on models employing either multiple imputation of missing data or inverse probability weighting to adjust for loss to follow-up (Figure 2A). Associations of tree canopy coverage with asthma at 5 years, and with wheeze at 5 or 7 years, were similar in magnitude, but not statistically significant [Table 4; see also Supplemental Material, Figure S1 (http://dx.doi.org/10.1289/ehp.1205513)]. RRs for associations with rhinitis were larger, but were imprecise and not significant due to the low prevalence of this outcome.

17The proportion of children with allergic sensitization to tree pollen increased with increasing quartiles of tree canopy coverage (Table 3, p = 0.03). In addition, each 1-SD increase in tree canopy coverage was associated with a 43% increase in the probability of IgE antibody response to the tree pollen mix (95% CI: 1.19, 1.72) after adjustment (unadjusted RR = 1.33; 95% CI: 1.16, 1.52) (Table 4). The association was similar following multiple imputation or inverse probability weighting, but was weaker and no longer significant when based on tree canopy distributions for addresses at ages 5 or 7 years, and when based on a larger 1-km buffer (Figure 2B). Tree canopy coverage also was positively associated with the IgE response to any of the nine allergens after (but not before) adjustment for covariates (RR = 1.20; 95% CI: 1.05, 1.37) (Table 4) and with the IgE response to several individual allergens when modeled separately [see Supplemental Material, Table S1 (http://dx.doi.org/10.1289/ehp.1205513)].

18Results from this birth cohort study did not support the hypothesized protective association of urban tree canopy coverage with asthma or wheeze. However, tree canopy cover within 0.25 km of the prenatal address was associated with higher prevalence of allergic sensitization to tree pollen at 7 years of age that was evident before and after covariate adjustment and when analyzed using alternate methods to address missing data and loss to follow-up. Allergic sensitization to tree pollen and other allergens as defined in this study does not necessarily indicate the presence of noticeable allergy symptoms such as rhinitis, which was relatively uncommon in this cohort and was not significantly associated with exposure.

19A previously observed inverse association between street trees and asthma was limited by an ecologic design, with the neighborhood unit of analysis being much larger than the 0.25-km buffer areas evaluated in this study, and did not consider the full geographic extent of urban trees, due to a focus exclusively on street trees. Most trees in New York City’s urban forest would not be included as street trees because they are in public parks or on private property. We hypothesized that urban tree canopy coverage would have a protective effect on asthma through a reduction in local air pollution exposure, but we did not observe any trends in the hypothesized direction, and some of our analyses indicated a positive association with the prevalence of asthma among children in our study population. The amount of pollution removed by the urban tree canopy may be too small to cause meaningful differences in air pollution exposure among New York City residents, particularly given spatial exposure gradients attributable to emissions from vehicles and other urban sources.

20Urban tree canopy coverage near the prenatal address was associated with IgE response to a tree pollen mix and other allergens (notably grass pollen, ragweed, and cat dander) at 7 years of age in this birth cohort study. Our assessment of tree canopy coverage and allergic sensitization did not allow us to determine the specificity of the association at the level of tree species or to evaluate the role of plant biodiversity, and we did not have data on personal or outdoor exposure to tree pollen, which has been previously evaluated alongside tree pollen sensitization as a predictor of rhinitis. The observed associations may be the result of multiple allergen exposures, or other geographically patterned exposures, that are higher in areas of New York City with more tree canopy coverage. However, the tree pollen mix used for the IgE assessment included three species (Acer negundo, Quercus alba, and Ulmus americana) that each make up at least 1% of the leaf area in New York City. In addition, there are cross-reactivities between the five species represented in the tree pollen mix and other tree species of the same genera.

21Randomized controlled studies of interventions to reduce multiple allergen exposures have suggested that early allergen exposures contribute to later development of allergic sensitization. Although geographic and temporal variation in early-life allergen exposure also occurs without clinical manipulation, there is limited evidence as to whether such variation predicts allergic sensitization. Tree pollen counts have been reported to be higher in the homes of children with allergic sensitization to tree pollen, but the interpretation of such findings remains unclear; in fact, one recent study suggests that quantitative IgE levels can be used as a biomarker for exposure to indoor allergens among sensitized individuals. It is unknown whether individuals without symptomatic allergic sensitization would likewise have higher IgE response and be more likely to exhibit false positive tests for allergic sensitization following allergen exposure, but if so an association between allergen exposure and seroatopy could be observed without having meaningful clinical consequences. Several studies of seasonal exposure patterns suggest an association between pollen exposure and sensitization that may be strongest in infancy rather than during pregnancy, but an association between novel allergen exposure and allergic sensitization may also persist into adulthood. Tree canopy exposure estimates based on the prenatal neighborhood in our study may be serving as a proxy marker of exposure during infancy or early childhood.

22Key strengths of this study include detailed assessment of the urban tree canopy using LiDAR and multispectral image processing to characterize buffer neighborhoods centered on prenatal home addresses. Furthermore, we used previously validated outcomes with relevance to asthma or allergic sensitization. In addition, the birth cohort design made it possible to control for detailed, prospectively collected family, home, and neighborhood characteristics that may confound the associations of interest.

23Our study was limited by the lack of individual measures of pollen exposure and measures of exposure to specific allergenic tree species. Estimates may be biased by unmeasured or residual confounding, or by cohort selection or attrition. In addition, the correlations among neighborhood characteristics, and particularly between tree canopy and park coverage, limits our ability to distinguish the importance of particular types of green space or vegetation. Measurement error may be present in the health outcomes or the tree canopy exposure metric. In particular, the accuracy of geographic exposure measures could be improved by incorporating other frequently visited locations. Finally, restriction to a population of African-American and Dominican families in low-income areas of New York City may limit the generalizability of our findings to other populations and geographic areas.

24Although asthma and allergy in children are important urban health concerns, our findings should be considered in the context of considerable environmental benefits of urban tree cover such as carbon sequestration, heat island reduction, energy conservation, and storm water management as well as potential health benefits suggested by other studies. Further study is needed to elucidate the influence, if any, of the urban forest on allergic and respiratory illness. Future research should examine spatial variation in tree species, pollen exposure, and air quality and their link to health across diverse populations and geographic settings.

25The original manuscript published online reported that testing used Tx8 instead of the correct Tx1, which included Juglans californica (California black walnut) and Ulmus americana (American elm) rather than Corylus avellana (common filbert) and Platanus x acerifolia (London planetree). The text has been corrected here.