Over 65% of all women in the United States are overweight or obese,1 and 71.7% of women 40–59 years old report being overweight and 39.5% report being obese, placing this population at high risk of the well-characterized health effects of overweight and obesity,2 including cardiovascular disease,3 cancer,4 and all-cause mortality.5 Researchers are increasingly applying an ecosocial lens to understand risk factors for obesity, investigating how individual, social, and environmental factors interact to drive health.6 In recent years, research has suggested that the built environment (e.g., neighborhood walkability),7–9 natural vegetation (i.e., greenness),10 and air pollution11–13 are associated with body mass index (BMI). These environmental factors are promising targets because they are ubiquitous exposures that can be modified broadly through changes to policy on development, land use, transportation, and environmental regulation.6,7,14,15 However, these exposures may have complex interrelationships, and few studies have examined these exposures in concert to isolate independent associations of each exposure with BMI.
The literature on environmental determinants of BMI is rapidly growing. Over the past decade, there has been a dramatic increase in studies examining how factors of the built environment, such as nearby destinations to walk to, may create opportunities for higher levels of walking and subsequent lower levels of obesity.8 Alternatively, there has been a growing interest in how natural environments might affect health behaviors and outcomes, including obesity. Greenness is an ecologic exposure that is thought to decrease BMI by providing a location for physical activity.16 Studies show that, in general, greater neighborhood greenness is associated with lower likelihood of overweight or obesity.10,17 Recent studies in children have also implicated air pollution as a risk factor for obesity.11,12 In research focusing on adults, one study observed elevated findings.13 Another study in adults showed a positive association between roadway proximity and adiposity but no association between particulate air pollution exposure and BMI.18
While evidence builds regarding the associations between BMI and exposure to neighborhood walkability, greenness, and air pollution, few studies have examined all of these factors simultaneously. Spatial features associated with walkability may be correlated with spatial patterns in air pollution levels. In addition, greenness may buffer exposure to air pollution, but greenness may be lower in more walkable places. Research has highlighted these correlations,19–22 but no study has examined the joint and independent associations between these exposures and BMI. Moreover, previous studies have not examined whether the assumption of linearity holds when evaluating these associations. If this is not the case, and the association between BMI and these exposures changes across the distribution of the exposures, then previously observed and reported effect estimates might be biased.
In this study, we examine environmental correlates of BMI using data from a subset of the Nurses’ Health Study, a cohort study of adult women, a population that is likely particularly susceptible to obesity-related health effects.2 Our primary objective was to examine the joint, potentially nonlinear, relationships between greenness, neighborhood walkability, and air pollution and BMI across the northeastern United States.
We used data from the Nurses’ Health Study, a nationwide prospective cohort assessing a wide variety of risk factors for chronic disease among women. In 1976, 121,701 female registered nurses aged 30–55 years from 11 states returned an initial questionnaire ascertaining a variety of health-related exposures. The cohort has been continuously followed with biennial questionnaires. Response rates over follow-up have been ≈90%. Data for this analysis were taken from 2006, the most recent year for which air pollution and walkability data were available at the time of analysis. All available residential addresses in 2006 were geocoded to obtain latitude and longitude. Approximately 90% were successfully matched to the street address. For the current analyses, we included all women who completed the 2006 questionnaire, lived in the northeastern United States (Figure 1) where satellite air pollution predictions were available, and had an address geocoded to the street level. The study was approved by the Institutional Review Board of Brigham and Women’s Hospital, Boston, MA, and informed consent was implied through return of the questionnaires.
Outcome Assessment: BMI
The study participants were asked about their weight at the 2006 questionnaire. We used information on self-reported height, collected at enrollment, and weight in 2006 to calculate BMI as weight/height2 (kg/m2). A validation study of 184 Nurses’ Health Study participants showed that self-reported weights were highly correlated with measured weights (r = 0.96; mean difference, 1.5 kg).23
Neighborhood walkability is often defined based on land use patterns, transportation systems, and urban design.24 Although measures of neighborhood walkability have not been standardized, the measures of population density, land-use mix, and street connectivity have commonly been linked to health outcomes.7,8 We created a consolidated walkability index using Z scores for these neighborhood components. This index reflects the built environment as a latent construct comprising information from various correlated individual factors. Using US Census25 and commercially available geographic information systems (GIS) data, we created residence-level measures of population density, business counts (as a proxy for accessibility), and intersection counts (as a proxy for street connectivity) linked to the participants’ geocoded 2006 mailing addresses. Population density was calculated based on the 2000 Census population density in the tract of residence in 2006. To develop business count and intersection count measures, we created 1200 m line-based network buffers around each home address; we previously identified this buffer as the most relevant to BMI-related associations within nurses living in the United States.26 Briefly, we used ArcGIS software to identify the street network within 1200 m from each participant’s mailing address and included a 50 m buffer on either side of the road. Interstates and ramps were excluded from the networks, as we were interested in walkable streets only. We used this line-based network buffer to focus on areas people can access by walking27 (eFigure 1; http://links.lww.com/EDE/B244). Destination counts were measured as the counts of all stores, facilities, and services in a participant’s buffer using the commercially available 2009 infoUSA spatial database on “points of interest,” which includes grocery stores, restaurants, banks, hotels, hospitals, libraries, and so on (ESRI 2008). Intersection counts were quantified as the number of ≥3-way intersections within each buffer. Higher intersection count increases the efficiency of walking to destinations.28 For the intersection count and road network information, we used ArcGIS 8.3 StreetMap USA from ESRI, which is based on 2007 US Census TIGER (www.census.gov/geo/www/tiger) and TeleAtlas data (www.teleatlas.com).
Finally, Z scores were created to standardize each individual component of the built environment, namely population density, business count, and street connectivity. These scores had a mean of 0 and a standard deviation of 1, which we summed to create a walkability index combining these three built environment components. The final walkability index had a mean of 0 (range: −1.7, 34.1), with higher values indicating higher levels of neighborhood walkability. Similar indices have been used previously.29–31
Exposure to green, natural areas around each participant’s home address was estimated using a satellite image–based vegetation index. Chlorophyll in plants strongly absorbs visible light (0.4–0.7 μm) for use in photosynthesis, while leaves reflect near-infrared light (0.7–1.1 μm). The normalized difference vegetation index (NDVI) calculates the ratio of the difference between the near-infrared region and red reflectance to the sum of these two measures, ranging from −1.0 to 1.0, with larger values indicating higher levels of vegetative density.32 For this study, we used data from the Moderate-resolution Imaging Spectroradiometer (MODIS) from NASA’s Terra satellite. MODIS provides images every 16 days at a 250 m resolution.33 We used ArcGIS software to link summertime (July 2006) NDVI data to each participant’s geocoded address as a measure of greenness directly accessible outside the home. Summertime NDVI was chosen as it provides the peak variation for greenness in the northeastern United States. We used the 250 m resolution as a measure of greenness directly accessible outside each home. In sensitivity analyses, we also modeled greenness using mean NDVI data in the 1250 m area surrounding each address to capture greenness within a 10- to 15-minute walk of the home.34
We assessed particulate air pollution exposure using a satellite-based spatiotemporal model that predicts daily ambient concentrations of particulate matter with an aerodynamic diameter less than 2.5 μm (PM2.5) specific to each participant’s residence.
The exposure assessment was based on a hybrid model incorporating daily satellite remote sensing data at 1 km spatial resolution.35 The model makes use of NASA’s MAIAC (Multi-angle Implementation to Atmospheric Correction) algorithm36 providing 1-km resolution aerosol optical depth data. We used a mixed model approach to regress daily PM2.5 mass concentrations from US Environmental Protection Agency (EPA) Air Quality System (AQS) and Interagency Monitoring of Protected Visual Environments (IMPROVE) network locations on aerosol optical depth (AOD), spatial, and temporal predictors. For days when aerosol optical depth data were not available for some grid cells, we fit a generalized additive model with a thin plate spline term of latitude and longitude to interpolate PM2.5. The mean out-of-sample R2 from 10-fold cross validation was 0.88, and the slope of the predicted versus withheld observations was 0.99, showing excellent prediction ability and no bias. We assigned each participant a PM2.5 exposure calculated as the calendar-year average of daily model predictions for 2006 based on the 1 km cell in which their address was located.
We estimated exposure to nitrogen dioxide (NO2) from the US EPA Air Quality System database (http://www.epa.gov/ttn/airs/airsaqs/detaildata/downloadaqsdata.htm) and estimated monthly averages for 2006 at each monitoring location across the northeastern United States for months during which at least 75% of the observations were available at each monitor. We assigned the weighted monthly average concentration of NO2 from the five nearest monitors to each participant’s residence, excluding participants who did not live within at least 50 km of a monitor. We used the inverse of the square of the distance between residences and monitors as the weights.37 We allowed different monitors to be selected for each month, depending on data availability. We then calculated an annual average for 2006 at each participant’s residence.
Information on factors related to both environmental measures and BMI that could be potential confounders was available. We adjusted for age (continuous), race (White/Black/Asian/Other), and smoking status (current/former/never). To adjust for individual-level socioeconomic status (SES), we included information on husband’s education. To adjust for area-level SES, we used information from the 2000 Census on tract-level median household income (P053001) and home value (H085001), as well as the proportion of census tract residents without a high school diploma (P037001).25 All reported results refer to models adjusting for the above covariates, unless otherwise specified.
We first examined the means and distributions of each environmental measure and estimated Spearman correlation coefficients between each measure. Because these correlation coefficients do not capture potentially nonlinear relationships, we also fit a generalized additive model using walkability as the dependent variable and greenness, NO2, and PM2.5 as the independent variables and fit penalized splines to detect any potential deviations from linearity.
We then employed generalized additive models to assess the association between each exposure of interest (NO2, PM2.5, greenness, and walkability) and BMI separately, allowing for deviations from linearity using penalized splines. If the generalized cross-validation criterion indicated that a linear association was a better fit for any of these exposures, these were included in the model as linear terms, otherwise we used the generalized cross-validation-selected number of degrees of freedom (df). When we included untransformed BMI as the dependent variable in the models, the regression diagnostics showed a poor fit and violated assumptions of normally distributed residuals and homoscedasticity. We, therefore, used the natural logarithm of BMI as the dependent variable, resulting in no violations. To assess independent associations between these correlated exposures and BMI, we also included all four exposures in the same generalized additive model simultaneously.
We assessed potential interactions between pairs of exposures of interest. If both exposures (main associations) were linearly associated with BMI, we included an interaction term between the two in the model. If at least one exposure was nonlinearly associated with BMI, then we included a penalized spline surface in the model. In this case, we compared Akaike’s Information Criterion between the model with separate splines for each exposure versus the model with the spline surface to assess better fit. If the model with the surface spline was a better fit, that would suggest an interaction between the two variables in the surface spline.
Based on prior studies showing differential relationships between walkability and health by neighborhood SES,38,39 we also assessed whether age, census tract median household income, median home value, proportion of residents without a high school diploma, or urbanicity modified the walkability–BMI association. Urbanicity was defined by the participant’s residence in an urban (urban area ≥50,000 people) versus non-urban (suburban, i.e., urban cluster of 10,000–49,999; or small town/rural, i.e., urban cluster of <10,000) census tract.40
If linear, results are presented as percent lower or higher BMI per interquartile range width increase in exposure; otherwise, we present plots depicting the nonlinear dose–response curves. All statistical analyses were conducted using SAS, version 9.3 (SAS Institute Inc., Cary, NC) and R, version 3.2.0 (Foundation for Statistical Computing, Vienna, Austria).
There were a total of 23,435 women eligible for this study (Table). The mean age among participants was 72 years, and the average participant was normal weight (mean BMI 27). The majority of participants were white, lived in metropolitan areas, were past smokers, and had husbands with more than a high school education. We also show participant characteristics by BMI cut points for overweight and obese.41 Obese participants were more likely to live in areas with a higher walkability index, with higher PM2.5, and with lower area-level SES, were less likely to be white, and had husbands with lower educational attainment.
The correlation structure across the exposures of interest, as well as the census tract variables, is presented in eFigure 2; http://links.lww.com/EDE/B244. PM2.5 and NO2 were moderately positively correlated with each other and were also moderately correlated with walkability. Greenness was negatively correlated with all other exposures. These relationships, however, were not linear (eFigure 3; http://links.lww.com/EDE/B244).
Because of the skewed distribution of the walkability index (eFigure 4; http://links.lww.com/EDE/B244), 99% of participants lived in an area with a walkability index below 10. Thus, Figure 2D (top and bottom) presents the association between walkability and BMI at walkability index levels below 10. eFigure 5; http://links.lww.com/EDE/B244 shows the association across the entire distribution of walkability index.
When assessing the association of each exposure of interest with BMI in separate models, we observed nonlinear associations with PM2.5 and walkability, a marginal nonlinear association with greenness (Figure 2B–D, top), and no association with NO2 (Figure 2A, top). Specifically, at lower levels of PM2.5 (<10 μg/m3), increases in particle concentrations were associated with increased BMI, while this association reached a plateau at higher particle concentrations (Figure 2A, top). We observed an inverse U-shaped association with greenness; at lower greenness levels, increases in greenness were associated with increased BMI, while at higher greenness levels, increases in greenness were associated with lower BMI (Figure 2C, top). Finally, we observed a nonlinear association between the walkability index and BMI (Figure 2D, top).
When we included all four exposures of interest together in the same model, however, the observed associations between BMI and NO2, PM2.5, and greenness changed. We observed no change in BMI (0.01%; 95% confidence interval [CI] = −0.36%, 0.37%) per interquartile range width (IQRw) increase in NO2 (5.3 ppb), 0.30% higher BMI (95% CI = −0.13%, 0.73%) per IQRw increase in PM2.5 (2.0 μg/m3), and 0.15% lower BMI (95% CI= −0.53%, 0.23%) per IQRw increase in greenness (0.14; Figures 2A–C, bottom). However, the nonlinear association between walkability and BMI remained consistent after adjusting for NO2, PM2.5, and greenness, with an estimated df = 3.78 (Figure 2D, bottom). At low levels of walkability (walkability index below 1.8, representing the lower 89th percentile of our population), we observed a positive association between walkability and BMI where a 10 percentile increase in walkability right before this level (i.e., an increase from the 79th to the 89th percentile of the walkability index) was associated with a 0.03% increase in log BMI. However, walkability was negatively associated with BMI at high walkability levels (walkability index above 1.8, representing the top 11th percentile of our population), where a 10 percentile increase in walkability right after this level was associated with a 0.84% decrease in log BMI. Results were consistent when considering greenness at a 1250 m buffer size (eFigure 6; http://links.lww.com/EDE/B244).
We observed no interactions between the exposures of interest, with the exception of a potential interaction between PM2.5 and walkability. The model containing a surface spline for PM2.5 and walkability yielded a slightly better fit based on Akaike Information Criterion values. We observed that at low PM2.5 levels, higher walkability held a moderate nonlinear association with lower BMI; however, at higher levels of PM2.5, this nonlinear relationship was stronger, suggesting a likely moderate interaction (eFigure 7; http://links.lww.com/EDE/B244).
Finally, we did not observe any interactions by area-level SES variables. In analyses stratified by urbanicity, the shape of the curve between walkability and BMI for urban participants was unchanged (eFigure 8; http://links.lww.com/EDE/B244). In the nonurban participants, we observed no relationship between walkability and BMI. We did detect an interaction by participant age. Specifically, we observed that among participants at the lower end of the age range, there was a nonlinear association between walkability and BMI, but this relationship was not observed in the oldest participants (greater than approximately 71 years; eFigure 9; http://links.lww.com/EDE/B244). To clarify the differences across age, we show results in eFigure 10; http://links.lww.com/EDE/B244 stratified by age quartiles.
We conducted a study in the northeastern United States to assess independent associations between air pollution, walkability, and greenness and BMI in a cohort of female nurses. Although in individual models we observed nonlinear associations with PM2.5, greenness, and walkability, when we simultaneously assessed all exposures, only the association between walkability and BMI remained. We observed a threshold for the relationship between walkability and BMI; increasing walkability was associated with increasing BMI at lower levels of walkability, while increasing walkability was linked to lower BMI in areas of higher walkability. We also observed a suggestive interaction between PM2.5 and walkability, where the relationship between walkability and BMI was stronger at higher levels of PM2.5. Finally, the nonlinear relationship between walkability and BMI existed only among younger participants.
We observed a threshold relationship for walkability, where increasing walkability below a walkability index of 1.8 was associated with increasing BMI, while above a walkability index of 1.8, increasing walkability was associated with lower BMI. About 11% of our sample lived in areas with a walkability index of 1.8 or higher, which ranged from smaller towns such as Allentown, PA (579.2 people per square mile) to larger cities such as Boston, MA (1,685.1 people per square mile), Philadelphia, PA (1,323.1 people per square mile), and New York City, NY (8,158.7 people per square mile).25 The concept of a threshold is consistent with the theory that when the built environment reaches a certain density and connectivity with sufficient land-use mix, it becomes more convenient to walk or take transit than to drive.42 Subsequently, residents may be more physically active and may maintain lower BMI. Indeed, these thresholds have been demonstrated in previous empirical analyses of the built environment and physical activity.43
There was no association between walkability and BMI in analyses restricted to participants living in nonurban areas. This is likely owing to the narrow range of walkability in nonurban areas, as well as the small number of participants living in nonurban areas. We found that the nonlinear relationship between walkability and BMI was limited to younger participants. This finding is consistent with the idea that BMI represents adiposity in younger adults, but BMI may incorporate sarcopenia, or decreased muscle mass, and have lower validity in older adult populations.44
Previous studies have investigated environmental influences on BMI, with diverse findings. There is an extensive body of literature examining the relationship between neighborhood walkability and BMI.7 Consistent with our findings, a recent review of natural experiments, longitudinal studies, and studies applying advanced analytical techniques to examine the built environment and BMI concluded that the evidence supports altering the design of neighborhoods to combat obesity.9
Less consensus exists regarding natural environments and BMI. Results suggest that there may be an inverse association between greenness and overweight/obesity45–47; however, gender-specific estimates have been shown to vary across studies.17,48 Other studies have observed either no association between greenness and BMI,49–51 or, contrary to a priori hypotheses, that greenness was associated with increased odds of overweight and obesity.52 Particle exposure has also been associated with elevated central obesity in adults.13 Most of these analyses controlled for a range of potential confounders but rarely adjusted for neighborhood walkability, greenness, or air pollution exposure.
Although there is evidence for associations between air pollution, walkability, and greenness and BMI,7,10–13 previous studies have assessed these associations either linearly or using categorical variables. These approaches either ignore any potential deviations from linearity in the dose–response curve or lack efficiency. Moreover, previous studies have ignored the correlations across environmental factors potentially associated with BMI. In this study, we accounted for these interrelationships and were able to assess independent nonlinear associations between these factors and BMI. Furthermore, we showed that because the relationship across these exposures is not linear (eFigure 2; http://links.lww.com/EDE/B244), failure to adjust for some of these environmental factors would lead to nonlinear confounding and result in biased observed associations. For instance, when we assessed the PM2.5–BMI association without adjusting for walkability, NO2, and greenness, we observed a nonlinearly increasing dose–response curve (Figure 2B, top). When we included these factors in the same model simultaneously, however, the association between PM2.5 and BMI was no longer evident (Figure 2B, bottom).
Our study has some limitations. In this cross-sectional study, we were unable to assess the association with incidence of obesity but only with current BMI. Even though reverse causation is a concern in cross-sectional studies, such a scenario is not likely in this analysis, as we have shown in previous analyses in this cohort that BMI does not predict neighborhood choice.53 Moreover, a recent review of prospective studies found that the use of prospective study designs and advanced methods of analysis show similar associations between walkability and obesity when compared with cross-sectional analyses.9 We used self-reported weight and height information to calculate BMI, which might have resulted in outcome misclassification. Self-reported weights, nevertheless, have been found to be highly reliable,23 and any misclassification is likely nondifferential with respect to exposure. Exposure measurement error is also likely for all examined exposures. For example, while many studies have examined measures of the neighborhood built environment, there is no consensus on what metrics best represent neighborhood walkability. In addition, NO2 exposures were assessed using a weighted average of concentrations measured at monitors, which is expected to bias the estimated effects towards the null54 and could be one reason why we observed no association between NO2 and BMI. Nevertheless, in our study, exposure measurement error is expected to be nondifferential because all exposures were assessed independent of BMI. For exposures with linear relationships, any resulting bias would likely be toward the null.54 However, the behavior of such error in nonlinear settings, and especially in the presence of nonlinear confounding measured with error, has not been widely explored. Because participants were only asked about recreational physical activity and not routine travel-related activity that might be linked to neighborhood walkability, we were unable to assess potential mediation by physical activity. Participants in this study were nurses at the time of recruitment, were over 90% white, and were all female. While this might restrict the generalizability of these results to the general population, it also limits the potential for confounding by SES, race, and sex. Finally, the correlation structure across environmental factors is likely to vary across locations, and therefore, our results should be replicated in other parts of the country, as well as other parts of the world with a different urban design and among populations of different ages (for instance, children, who are under development).
Our study has several strengths. We used data from a large population of medical professionals with detailed information on potentially important confounders, including lifestyle, individual- and area-level factors. We used sophisticated methods to obtain exposure estimates at the residential addresses of the study participants, reducing potential exposure measurement error. We were also able to simultaneously examine the association with all four exposures of interest (NO2, PM2.5, greenness, and walkability) while allowing for nonlinear dose–response curves. We were also able to observe a nonlinear interaction between age and the walkability–BMI association. To our knowledge, no other study has assessed these correlated exposures simultaneously, nor allowed for nonlinear interactions, when examining the association between environmental factors and BMI.
In conclusion, we conducted a large cross-sectional study across the northeastern United States. We observed a robust nonlinear association between walkability and BMI, independent of air pollution and greenness. Our findings can help to clarify the relationship between environmental factors and BMI and highlights the importance of accounting for nonlinear confounding by interrelated environmental factors in studies of the environment and BMI.
1. Ogden CL, Carroll MD, Kit BK, Flegal KM. Prevalence of childhood and adult obesity in the United States, 2011-2012. JAMA. 2014;311:806–814.
2. Guh DP, Zhang W, Bansback N, Amarsi Z, Birmingham CL, Anis AH. The incidence of co-morbidities related to obesity and overweight: a systematic review and meta-analysis. BMC Public Health. 2009;9:88.
3. Hubert HB, Feinleib M, McNamara PM, Castelli WP. Obesity as an independent risk factor for cardiovascular disease: a 26-year follow-up of participants in the Framingham Heart Study. Circulation. 1983;67:968–977.
4. Calle EE, Rodriguez C, Walker-Thurmond K, Thun MJ. Overweight, obesity, and mortality from cancer in a prospectively studied cohort of U.S. adults. N Engl J Med. 2003;348:1625–1638.
5. Masters RK, Powers DA, Link BG. Obesity and US mortality risk over the adult life course. Am J Epidemiol. 2013;177:431–442.
6. Krieger N. Embodiment: a conceptual glossary for epidemiology. J Epidemiol Community Health. 2005;59:350–355.
7. Sallis JF, Floyd MF, Rodríguez DA, Saelens BE. Role of built environments in physical activity, obesity, and cardiovascular disease. Circulation. 2012;125:729–737.
8. Feng J, Glass TA, Curriero FC, Stewart WF, Schwartz BS. The built environment and obesity: a systematic review of the epidemiologic evidence. Health Place. 2010;16:175–190.
9. Martin A, Ogilvie D, Suhrcke M. Evaluating causal relationships between urban built environment characteristics and obesity: a methodological review of observational studies. Int J Behav Nutr Phys Act. 2014;11:142.
10. James P, Banay RF, Hart JE, Laden F. A Review of the Health Benefits of Greenness. Curr Epidemiol Rep. 2015;2:131–142.
11. McConnell R, Shen E, Gilliland FD, et al. A longitudinal cohort study of body mass index and childhood exposure to secondhand tobacco smoke and air pollution: the Southern California Children’s Health Study. Environ Health Perspect. 2015;123:360–366.
12. Jerrett M, McConnell R, Wolch J, et al. Traffic-related air pollution and obesity formation in children: a longitudinal, multilevel analysis. Environ Health. 2014;13:49.
13. Eze IC, Schaffner E, Foraster M, et al. Long-term exposure to ambient air pollution and metabolic syndrome in adults. PLoS One. 2015;10:e0130337.
14. Durand CP, Andalib M, Dunton GF, Wolch J, Pentz MA. A systematic review of built environment factors related to physical activity and obesity risk: implications for smart growth urban planning. Obes Rev. 2011;12:e173–e182.
15. Pasanen TP, Tyrväinen L, Korpela KM. The relationship between perceived health and physical activity indoors, outdoors in built environments, and outdoors in nature. Appl Psychol Health Well Being. 2014;6:324–346.
16. Hartig T, Mitchell R, de Vries S, Frumkin H. Nature and health. Annu Rev Public Health. 2014;35:207–228.
17. Astell-Burt T, Feng X, Kolt GS. Greener neighborhoods, slimmer people? Evidence from 246,920 Australians. Int J Obes (Lond). 2014;38:156–159.
18. Li W, Dorans KS, Wilker EH, et al. Residential proximity to major roadways, fine particulate matter, and adiposity: The framingham heart study. Obesity (Silver Spring). 2016;24:2593–2599.
19. James P, Hart JE, Laden F. Neighborhood walkability and particulate air pollution in a nationwide cohort of women. Environ Res. 2015;142:703–711.
20. Hankey S, Marshall JD, Brauer M. Health impacts of the built environment: within-urban variability in physical inactivity, air pollution, and ischemic heart disease mortality. Environ Health Perspect. 2012;120:247–253.
21. Marshall JD, Brauer M, Frank LD. Healthy neighborhoods: walkability and air pollution. Environ Health Perspect. 2009;117:1752–1759.
22. Hystad P, Davies HW, Frank L, et al. Residential greenness and birth outcomes: evaluating the influence of spatially correlated built-environment factors. Environ Health Perspect. 2014;122:1095–1102.
23. Willett W, Stampfer MJ, Bain C, et al. Cigarette smoking, relative weight, and menopause. Am J Epidemiol. 1983;117:651–658.
24. Handy SL, Boarnet MG, Ewing R, Killingsworth RE. How the built environment affects physical activity: views from urban planning. Am J Prev Med. 2002;23(2 suppl):64–73.
25. US Census Bureau. US Census 2000. 2000.
26. James P, Berrigan D, Hart JE, et al. Effects of buffer size and shape on associations between the built environment and energy balance. Health Place. 2014;27:162–170.
27. Oliver LN, Schuurman N, Hall AW. Comparing circular and network buffers to examine the influence of land use on walking for leisure and errands. Int J Health Geogr. 2007;6:41.
28. Saelens BE, Handy SL. Built environment correlates of walking: a review. Med Sci Sports Exerc. 2008;40(7 suppl):S550–S566.
29. Kelly CM, Lian M, Struthers J, Kammrath A. Walking to Work: The Roles of Neighborhood Walkability and Socioeconomic Deprivation. J Phys Act Health. 2015;12(suppl 1):S70–S75.
30. Frank LD, Sallis JF, Saelens BE, et al. The development of a walkability index: application to the Neighborhood Quality of Life Study. Br J Sports Med. 2010;44:924–933.
31. Saarloos D, Alfonso H, Giles-Corti B, Middleton N, Almeida OP. The built environment and depression in later life: the health in men study. Am J Geriatr Psychiatry. 2011;19:461–470.
32. Kriegler F, Malila W, Nalepka R, Richardson W. Preprocessing transformations and their effects on multispectral recognition. Proceedings of the Sixth International Symposium on Remote Sensing of Environment. 1969:Ann Arbor, MI: The University of Michigan; 97–131.
33. Carroll ML, DiMiceli CM, Sohlber RA, Townshend JRG. 250m MODIS Normalized Difference Vegetation Index. 2004.College Park, MD: University of Maryland.
34. James P, Hart JE, Banay RF, Laden F. Exposure to greenness and mortality in a Nationwide Prospective Cohort Study of Women. Environ Health Perspect. 2016;124:1344–1352.
35. Kloog I, Chudnovsky AA, Just AC, Nordio F, Koutrakis P, Coull BA, Lyapustin A, Wang Y, Schwartz J. A new hybrid spatio-temporal model for estimating daily multi-year PM2.5 concentrations across northeastern USA using high resolution aerosol optical depth data. Atmos Environ. 2014;95:581–590.
36. Lyapustin A, Martonchik J, Wang Y, Laszlo I, Korkin S. Multiangle implementation of atmospheric correction (MAIAC): 1. Radiative transfer basis and look-up tables. J Geophys Res Atmospheres. 2011;116(D3).
37. Volk HE, Lurmann F, Penfold B, Hertz-Picciotto I, McConnell R. Traffic-related air pollution, particulate matter, and autism. JAMA Psychiatry. 2013;70:71–77.
38. Steinmetz-Wood M, Kestens Y. Does the effect of walkable built environments vary by neighborhood socioeconomic status? Prev Med. 2015;81:262–267.
39. Sugiyama T, Howard NJ, Paquet C, Coffee NT, Taylor AW, Daniel M. Do relationships between environmental attributes and recreational walking vary according to area-level socioeconomic status? J Urban Health. 2015;92:253–264.
40. Morrill R, Cromartie J, Hart G. Metropolitan, urban, and rural commuting areas: toward a better depiction of the United States settlement system. Urban Geography. 1999;20:727–748.
41. Expert Panel on the Identification Evaluation and Treatment of Overweight in Adults. Clinical guidelines on the identification, evaluation, and treatment of overweight and obesity in adults: executive summary. Expert Panel on the Identification, Evaluation, and Treatment of Overweight in Adults. Am J Clin Nutr. 1998;68(4):899–917.
42. Forsyth AJ, Oakes JM, Schmitz KH, Hearst M. Does residential density increase walking and other physical activity? Urban Studies. 2007;44(4):679–697.
43. Moudon AV, Lee C, Cheadle AD, Garvin C, Johnson D, Schmid TL, et al. Operational definitions of walkable neighborhood: theoretical and empirical insights. J Phys Activity Health. 2006;3(suppl 1):S99–S117.
44. Hu FB. Obesity Epidemiology. 2008.Oxford: Oxford University Press.
45. Lovasi GS, Schwartz-Soicher O, Quinn JW, et al. Neighborhood safety and green space as predictors of obesity among preschool children from low-income families in New York City. Prev Med. 2013;57:189–193.
46. Bell JF, Wilson JS, Liu GC. Neighborhood greenness and 2-year changes in body mass index of children and youth. Am J Prev Med. 2008;35:547–553.
47. Liu GC, Wilson JS, Qi R, Ying J. Green neighborhoods, food retail and childhood overweight: differences by population density. Am J Health Promot. 2007;21(4 suppl):317–325.
48. Prince SA, Kristjansson EA, Russell K, et al. A multilevel analysis of neighbourhood built and social environments and adult self-reported physical activity and body mass index in Ottawa, Canada. Int J Environ Res Public Health. 2011;8:3953–3978.
49. Mowafi M, Khadr Z, Bennett G, Hill A, Kawachi I, Subramanian SV. Is access to neighborhood green space associated with BMI among Egyptians? A multilevel study of Cairo neighborhoods. Health Place. 2012;18:385–390.
50. Coombes E, Jones AP, Hillsdon M. The relationship of physical activity and overweight to objectively measured green space accessibility and use. Soc Sci Med. 2010;70:816–822.
51. Potestio ML, Patel AB, Powell CD, McNeil DA, Jacobson RD, McLaren L. Is there an association between spatial access to parks/green space and childhood overweight/obesity in Calgary, Canada? Int J Behav Nutr Phys Act. 2009;6:77.
52. Cummins S, Fagg J. Does greener mean thinner? Associations between neighbourhood greenspace and weight status among adults in England. Int J Obes (Lond). 2012;36:1108–1113.
53. James P, Hart JE, Arcaya MC, Feskanich D, Laden F, Subramanian SV. Neighborhood self-selection: the role of pre-move health factors on the built and socioeconomic environment. Int J Environ Res Public Health. 2015;12:12489–12504.
54. Kioumourtzoglou MA, Spiegelman D, Szpiro AA, et al. Exposure measurement error in PM2.5 health effects studies: a pooled analysis of eight personal exposure validation studies. Environ Health. 2014;13:2.