Associations of Cumulative and Point-in-Time Neighborhood Poverty and Walkability with Body Mass from Age 30 to 39

Abstract

Few studies examining the effects of neighborhood exposures have accounted for longitudinal residential history. This study examined associations of body mass index (BMI, kg/m²) with neighborhood-level walkability and poverty, both assessed concurrently and cumulatively in the years leading up to BMI assessment. Participants (N = 808) were from a cohort study of individuals originally recruited from public schools in Seattle, Washington, in fifth grade in 1985. Height and weight for BMI were obtained at four assessments at ages: 30 (in 2005), 33, 35, and 39. Participants also completed residential timelines listing each address where they lived from ages 28 to 39, creating a continuous record of addresses and moves. Neighborhood-level walkability and poverty were based on census block groups of each address. Generalized estimating equation models estimated associations of standardized neighborhood variables, both at point-in-time concurrently with assessment of BMI and cumulatively up to the time of BMI assessment. Mean BMI across observations was 28.8 (SD = 7.1). After adjusting for covariates, cumulative walkability was associated with lower BMI (b =  − 0.28; 95% CI: − 0.55, − 0.02), and cumulative neighborhood poverty was associated with higher BMI (b = 0.35; 95% CI: 0.09, 0.60). When examining point-in-time concurrent walkability and poverty with BMI, adjusted associations were close to the null and non-significant. This study provides evidence for a significant role of cumulative exposure to neighborhood built and socioeconomic environments predicting BMI. It underscores the relative strength and importance of cumulative assessments to capture neighborhood exposure not captured through point-in-time assessments.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11524-022-00688-6.

Introduction

The prevalence of obesity remains historically high in the USA, with roughly 73% of adults meeting the criteria for overweight or obese [1]. Elevated body mass index (BMI) is a risk factor for a variety of chronic conditions, contributing to substantial disability and years of life lost [2]. Research examining the relationship of BMI and other health conditions with place—the social and physical environments in which people live—has increasingly recognized the importance of these environments for health and well-being. Contextual features in disadvantaged neighborhoods, such as: limited access to recreational facilities, parks, and healthy food, and lack of safe places to walk may affect physical activity and dietary behaviors associated with BMI [3]. Substantial evidence also links neighborhood concentrations of poverty with greater exposure to chronic stressors (e.g., higher crime, violence, and residential segregation; poor social capital and social support; and low esthetic quality) [3–5]. These stressors can in turn increase the risk of obesity by interfering with sleep that is essential for maintaining healthy body weight and contributing to processes that lead to maladaptive behaviors [6].

Policy initiatives have emphasized improvements in environmental factors, such as neighborhood walkability, as a way to reduce obesity at the population level [7, 8]. Neighborhood walkability, typically using indices including proximity to destinations, a supportive pedestrian environment, and design features such as street connectivity [9–11], has been linked to physical activity and lower prevalence of obesity, although results have been inconsistent [11–14]. Similarly, most studies examining the relationship between neighborhood poverty and obesity have reported significant associations; however, others have reported null or weak associations [15]. Several gaps in the existing literature limit our understanding of the impact of neighborhood factors on obesity. First, most research linking neighborhood poverty and walkability to obesity is based on cross-sectional studies [11–14]. This prevents the understanding of temporal precedence. Second, of available longitudinal studies, few have captured the composite history of neighborhood context as people move through different residences and experience various neighborhoods over time [16]. Most studies do not report on residential mobility, despite 12% of US residents moving each year [13, 17]. Moreover, although studies examining the effects of relocation can provide information on how a change in neighborhood context can shape the risk of obesity, these studies typically do not address the cumulative effects of neighborhood context. Detrimental exposures can compound to affect health over time, such that the risk of obesity is expected to be higher among those who accumulate exposure over time to less walkable neighborhoods. Conversely, health outcomes are relatively better among people who consistently reside in advantaged or more walkable neighborhoods [18, 19]. Research that fails to account for the cumulative history of residential context may underestimate the role of place on health [16].

To address these needs, the present study used data from a longitudinal study to examine associations of neighborhood-level poverty and walkability with BMI during the 30s using two different temporal measurements of neighborhood factors: (1) assessed concurrently with assessment of BMI and (2) assessed cumulatively from age 28 up to the time of BMI assessment. We also assessed whether associations varied by sex given previous investigations which observed different impacts of neighborhood socioeconomic disadvantage [20–22] and walkability [11, 12, 23] on obesity for men versus women. As secondary analyses, we also explored how associations between neighborhood-level factors and BMI might be explained by early life factors, including neighborhood poverty and receiving free or reduced-price school lunch in childhood.

Methods

Sample and Setting

This study used data from the Seattle Social Development Project (SSDP), a longitudinal cohort study of the development of prosocial and antisocial health-related behaviors. Individuals were followed prospectively since 1985, when they were in fifth grade (age 10), to 2014, when they were approximately 39 years old. Participants were recruited from 18 Seattle public elementary schools that served higher crime neighborhoods. However, due to mandated busing within the Seattle school district in the 1980s aimed at reducing racial segregation in schools, the sample included students from a variety of neighborhoods. Of the population of fifth-grade students in these schools (N = 1035), 808 (77%) students and their parents consented to participate in the longitudinal study. For these analyses, data on adult neighborhood exposures corresponded to a continuous record of residential addresses from ages 28 to 39, and other covariates and/or outcomes were from self-report surveys and interviewer-assessed health measures when participants were approximately ages 27 (year 2002), 30, 33, 35, and 39. Follow-up rates were high at each of these assessments (94%, 91%, 92%, 82%, and 88%, respectively). Most adult participants completed secured, password-protected web-based surveys, though options for paper, phone, or in-person surveys were available upon request. For childhood covariates included in secondary analyses, we also used data from annual assessments (of parents and children) from fifth through ninth grade. The childhood surveys were administered via a paper-and-pencil questionnaire. This study was approved by the University of Washington Institutional Review Board.

Measures

To obtain a longitudinal record of neighborhood environments, at ages 30, 33, and 39, participants completed a life-history calendar that indicated major life events. As part of this calendar, participants were asked to report the month and year of each residential move and the address of each residence where they lived since the prior assessment (in the case of the age 30 survey, since age 28). From this, we derived a continuous record of residential addresses and moves from ages 28 to 39. Each address was geocoded and linked to its corresponding census block group. This created the ability to link survey data, neighborhood demographics from the census, and parcel-level observation-specific measures of walkability. A total of 2291 addresses were geocoded across the adult survey waves. Of these addresses, 1361 were located within King County in Washington State, where Seattle is located. Neighborhood walkability and poverty measures were calculated for each of the 2291 geocoded addresses. In addition, during childhood, parents of participants provided their residential address at the time of each survey. Residential addresses at the time of the age 10 (1985) childhood survey wave were also geocoded to obtain objective measures of childhood neighborhood environment.

Neighborhood Poverty

The percentage of households in the block group living at or below the poverty threshold was obtained from the American Community Survey (ACS) of the US Census. For adulthood neighborhood poverty, we used 5-year ACS averaged estimates from 2010 to 2014 for addresses from 2011 to 2014. For secondary analyses, we included childhood neighborhood poverty data, which were drawn from the 1990 US Census, by taking the mean of block-group-level poverty at grades 5, 6, 7, and 9.

Neighborhood Walkability

Indicators of neighborhood walkability were available from the Smart Location Database (SLD) version 2.0, which was released by the Environmental Protection Agency in 2013. Eight different SLD variables at the block-group level were used to create a walkability index. Variables selected represent the five “D” variables commonly used to model walkability (i.e., density, diversity of land use, design, distance to transit, and destination accessibility)[24, 25] and were informed by correlations found in prior studies with: physical activity, sedentary behavior, and body weight [9, 26–28]. Those variables were: gross residential density (representing density), number of household workers per job (diversity of land use), 8-tier employment entropy (0–1 normalized proxy measure of mixing of activities and land uses, representing diversity of land use), road network density (design), intersection density (design), proportion of employment within the block within a half-mile of a transit stop (distance to transit), jobs within 45 min by car (destination accessibility), and employment accessibility by automobile expressed as a ratio of total employment accessibility (destination accessibility) [25]. Each of these variables was standardized (mean = 0, standard deviation [SD] = 1), and the mean of the 8 standardized variables was used to create the walkability index score. Similar indices have been used in other studies [29–31].

Body Mass

At the time of each adult survey wave, trained data collection staff arranged home visits for each participant to obtain additional health measures. During the age 27, 30, and 33 visits, participants self-reported their height and weight verbally to the interviewer in a face-to-face setting. During the age 35 and 39 visits, interviewers obtained objective measures of weight and height of participants (using scales and height rulers). Weight and height measurements were used to calculate BMI (weight in kilograms/[height in meters]²) at each of the study waves. Correlations between BMI based on in-person self-report at age 33 and BMI based on objective measures at 35 were high (0.90), suggesting accurate reporting in the earlier survey waves (see supplemental table for more details).

Covariates

Sociodemographic characteristics including: sex, race/ethnicity, annual household income at age 27, and highest level of education completed as of age 27 were included in analyses. As an indicator of childhood socioeconomic status, a variable was created for participation in the National School Lunch/Breakfast Program at grade 5, 6, or 7, with participation data obtained from schools. Eligibility for the school lunch program is often based on Federal income standards or participation in other Federal assistance programs.

Statistical Analysis

To account for clustering of observations within individuals over time, Generalized Estimating Equations (GEE) models were used to examine associations between neighborhood characteristics and BMI. Exchangeable correlation structure and robust standard errors were specified. Two primary sets of analyses were performed. To assess point-in-time associations across adulthood, the covariate of interest was the value of the neighborhood factor at the time of the study assessment, t, of BMI (e.g., neighborhood poverty at age 30 and BMI at age 30, neighborhood poverty at age 33 and BMI at age 33). The second set of analyses examined the cumulative exposure to the neighborhood characteristic from age 28 up to and including the study assessment, t. This cumulative exposure was calculated as the time-weighted mean of the neighborhood characteristic (e.g., poverty) across all months for the different residences from age 28 through the time of assessment, t. For example, if a participant lived in two different neighborhoods from age 28 to 30 and 75% of those months were in neighborhood A and 25% of those months were in neighborhood B, then the cumulative exposure value for neighborhood poverty at age 30 would be 0.75 times the poverty value in neighborhood A + 0.25 times the poverty value in neighborhood B. To aid interpretation, the point-in-time and cumulative neighborhood covariates were included in statistical models as standardized variables with a mean of 0 and standard deviation (SD) of 1.

In all models, we adjusted for biological sex, race (White [reference], African American, Native American, Asian American), and study assessment. We further adjusted for variables at the age 27 assessment: BMI as ordinal categories (< 25, 25–29.99, 30 + kg/m²), annual household income, and highest level of education completed as ordinal categories (< high school, high school degree, some college, Bachelor’s degree or higher). Finally, in additional models, we further adjusted for neighborhood poverty and receiving free or reduced-fee school lunch to understand the extent to which associations were explained by earlier childhood environments. To test for differential associations between the neighborhood factors and BMI according to sex, we included sex-by-neighborhood factor interaction terms.

Missing Data

Because of missing data in covariates and outcomes for the 808 study participants over the course of the study, we used Multiple Imputation with Chained Equations (MICE) to reduce potential bias and boost statistical power. MICE allowed us to account for different distributional forms (e.g., categorical, count, continuous) of the variables in the data. To improve the calculation of imputed values, we included all study variables in regression models as well as auxiliary variables that were correlated with the primary study variables including blood pressure status. We created 20 imputed datasets, and GEE models were run across each of the 20 datasets. Summary estimates and standard errors were calculated based on Rubin’s rules that account for between- and within-imputation variability.

Results

Demographic characteristics are presented in Table 1. The sample was evenly distributed with regards to sex. Almost half (47.2%) of the sample were White, 25.6% were African American, 5.3% were Native American, and 21.9% were Asian American. The highest level of education for approximately 40% of the sample was completion of high school or less. At age 27, just over half had a BMI ≥ 25 kg/m². In childhood, 52.4% received free or reduced-priced school lunch, and the mean census block group percentage of households living at or below poverty was 14.6 (SD = 13.6).

Table 1.

Characteristics of the study sample

Characteristic	Mean (SD) or % N = 808
Female sex	49.0
Race
White	47.2
African American	25.6
Native American	5.3
Asian American	21.9
Age 27 annual household income, $ thousands, in 2002	49.0 (40.8)
Age 27 body mass index, kg/m2
< 25.0	48.5
25.0–29.9	29.7
30.0 +	21.8
Age 27 highest level of education completed
Less than high school	25.6
High school	15.9
Some college	38.8
Bachelor’s degree or higher	19.7
Mean census block group percentage of households living at or below poverty during childhood	14.6 (13.6)
Childhood receipt of free or reduced-priced school lunch	52.4

Across all observations from ages 30 to 39, the mean BMI was 28.8 kg/m² (SD = 7.1). The average within-person SD of BMI across observations was 2.4 (range: 0.3 to 10.2), which suggests that there was both within- and between-person variability in BMI in this study sample. For the neighborhood-level factors, across all observations, the average census block group percentage of residents living in poverty was 11.6 (SD = 10.2), and the average of the standardized walkability scores was − 0.02 (SD = 0.41). There was also substantial within-person variability for neighborhood poverty (mean within-person SD = 6.2; range: 0, 22.0) and walkability (mean within-person SD = 0.22; range: 0, 1.13). Neighborhood poverty and higher walkability were modestly correlated (r = 0.18).

Table 2 shows model results for the association between standardized point-in-time neighborhood covariates and concurrent BMI from age 30 to 39 with columns including different sets of covariates. In models without adjustment for covariates (Models 1 and 2), neighborhood walkability was associated with lower BMI (b =  − 1.13; 95% CI: − 1.63, − 0.64). After adjustment of covariates including BMI at age 27 (Model 4) and further adjustment for neighborhood poverty (Model 5), this association was attenuated and no longer statistically significant. Neighborhood poverty showed no significant association with BMI in any unadjusted or adjusted models. There was no substantial evidence for moderation of neighborhood poverty-BMI and walkability-BMI associations by sex (interaction-p-values > 0.4).

Table 2.

Model results for association between point-in-time neighborhood factors and concurrent BMI

	Model 1		Model 2		Model 3		Model 4		Model 5
	b	95% CI	b	95% CI	b	95% CI	b	95% CI	b	95% CI
Intercept	28.83	28.35, 29.31	28.81	28.33, 29.29	21.77	20.45, 23.09	21.82	20.50, 23.14	21.82	20.50, 23.14
Neighborhood covariates
Neighborhood poverty	− 0.05	− 0.22, 0.12			− 0.06	− 0.22, 0.10			− 0.03	− 0.19, 0.13
Walkability			− 1.13*	− 1.63, − 0.64			− 0.17	− 0.37, 0.03	− 0.16	− 0.36, 0.04
Individual covariates
Male sex					0.72*	0.06, 1.38	0.72*	0.06, 1.38	0.72*	0.06, 1.38
Race
White (ref)
Black					0.25	− 0.61, 1.11	0.24	− 0.61, 1.1	0.25	− 0.60, 1.11
Native American					− 0.78	− 2.13, 0.57	− 0.79	− 2.13, 0.56	− 0.78	− 2.12, 0.57
Asian					− 0.92*	− 1.69, − 0.16	− 0.92*	− 1.68, − 0.16	− 0.92*	− 1.68, − 0.16
Age 27 income					0.00	− 0.01, 0.01	0.00	− 0.01, 0.01	0.00	− 0.01, 0.01
Age 27 education					− 0.30	− 0.61, 0.02	− 0.27	− 0.58, 0.04	− 0.28	− 0.59, 0.03
Age 27 BMI					6.30*	5.82, 6.78	6.28*	5.80, 6.76	6.28*	5.80, 6.76
Study wave					0.59*	0.48, 0.70	0.57*	0.46, 0.69	0.57*	0.46, 0.69

^*p < 0.05

The next set of analyses examined the association between standardized cumulative exposure to neighborhood factors from age 28 up to and including the time of BMI assessment (Table 3). Prior to adjusting for covariates (Models 1 and 2), cumulative neighborhood poverty was associated with increased BMI (b = 0.45; 95% CI: 0.15, 0.75), and cumulative neighborhood walkability was associated with decreased BMI (b =  − 0.86; 95% CI: − 1.21, − 0.52). These associations became attenuated after adjustment for covariates through Models 3 and 4 but remained statistically significant for cumulative poverty (b = 0.29; 95% CI: 0.04, 0.53). When cumulative neighborhood poverty and walkability were included together in the model (Model 5), both showed a statistically significant association with BMI. Sex did not moderate these associations (interaction-p values > 0.3).

Table 3.

Model results for association between cumulative neighborhood factors and BMI

	Model 1		Model 2		Model 3		Model 4		Model 5
	b	95% CI	b	95% CI	b	95% CI	b	95% CI	b	95% CI
Intercept	28.83	28.35, 29.31	28.83	28.36, 29.31	21.72	20.39, 23.04	21.80	20.48, 23.12	21.76	20.44, 23.09
Neighborhood covariates
Neighborhood poverty	0.45*	0.15, 0.75			0.29*	0.04, 0.53			0.35*	0.09, 0.60
Walkability			− 0.86*	− 1.21, − 0.52			− 0.19	− 0.45, 0.07	− 0.28*	− 0.55, − 0.02
Individual covariates
Male sex					0.71*	0.05, 1.37	0.71*	0.05, 1.37	0.70*	0.04, 1.36
Race
White (ref)
Black					0.12	− 0.74, 0.97	0.24	− 0.62, 1.09	0.10	− 0.75, 0.95
Native American					− 0.97	− 2.33, 0.4	− 0.78	− 2.13, 0.57	− 0.96	− 2.33, 0.42
Asian					− 0.96*	− 1.73, − 0.2	− 0.94*	− 1.7, − 0.17	− 0.98*	− 1.74, − 0.21
Age 27 income					0.00	− 0.01, 0.01	0.00	− 0.01, 0.01	0.00	− 0.01, 0.01
Age 27 education					− 0.25	− 0.56, 0.07	− 0.26	− 0.58, 0.05	− 0.20	− 0.52, 0.11
Age 27 BMI					6.27*	5.79, 6.75	6.27*	5.8, 6.75	6.24*	5.76, 6.72
Study wave					0.60*	0.49, 0.71	0.58*	0.47, 0.69	0.58*	0.47, 0.69

^*p < 0.05

Finally, we explored whether associations of cumulative neighborhood factors with BMI remained after adjustment for childhood factors including childhood neighborhood poverty and receiving free or reduced-price school lunch (Table 4). In an initial model with the childhood factors and demographic characteristics only (Model 1), participants living in higher poverty neighborhoods during childhood had a moderately higher BMI, but the difference was not statistically significant. Those who received free or reduced-priced school lunch also had moderately higher adulthood BMI, but this also did not reach statistical significance at p < 0.05. When all adulthood covariates were included (Model 2), the associations of childhood factors were attenuated, but cumulative neighborhood poverty and walkability in adulthood remained significantly associated with BMI.

Table 4.

Model results for association between cumulative neighborhood factors and BMI after including the childhood factors

	Model 1		Model 2
	b	95% CI	b	95% CI
Intercept	25.98	24.49, 27.47	21.40	20.02, 22.77
Childhood covariates
Childhood neighborhood poverty	0.50	− 0.04, 1.04	− 0.17	− 0.54, 0.21
Free or reduced-priced lunch	0.99	− 0.06, 2.04	0.70	− 0.06, 1.46
Adulthood neighborhood covariates
Neighborhood poverty			0.34*	0.09, 0.60
Walkability			− 0.29*	− 0.55, − 0.02
Individual covariates
Male sex	0.01	− 0.89, 0.92	0.66*	0.00, 1.33
Race
White (ref)
Black	2.14*	0.78, 3.51	− 0.04	− 0.98, 0.90
Native American	− 0.04	− 2.12, 2.03	− 1.03	− 2.41, 0.35
Asian	− 2.14*	− 3.32, − 0.96	− 1.13*	− 2.01, − 0.24
Age 27 income			0.00	− 0.01, 0.01
Age 27 education			− 0.14	− 0.47, 0.19
Age 27 BMI			6.26*	5.78, 6.74
Study wave	0.63*	0.55, 0.72	0.58*	0.47, 0.69

^*p < 0.05

Discussion

This longitudinal study that captured detailed residential history on participants over 11 years from age 28 to 39 showed that cumulative exposure over time to physical and social environments, including neighborhood walkability and poverty, was associated with adulthood BMI. Specifically, living in more walkable residential environments over time was associated with lower BMI, while living in neighborhoods with higher levels of poverty over time was associated with higher BMI. These associations were observed even when adjusting for age 27 BMI as well as demographic covariates including: sex, race, adult income, and education. Additional analyses showed that childhood SES factors did not explain these associations. In contrast, when examining these neighborhood exposures concurrently with BMI, there were no statistically significant associations.

Findings showing a role of cumulative, but not point-in-time, exposure to neighborhood physical and socioeconomic environments for BMI may provide insight into the inconsistent evidence in cross-sectional studies. It may take extended time and exposure—captured in cumulative assessments—for environmental influences on health behaviors, such as physical activity and dietary habits, to manifest in changes in BMI. Consistent with this, cross-sectional findings of associations between neighborhood walkability and physical activity—a more proximal outcome that is likely to be more immediately shaped by one’s current environment—tend to be more consistent relative to studies using weight-related measures as the outcome [12, 32]. Similarly, neighborhood disadvantage could have more temporally proximal impacts on behaviors that only later manifest in effects on BMI. For example, residents of disadvantaged neighborhoods tend to have less access to affordable healthy food [33, 34], lack of safety from crime [35], limited opportunities to be physically active, and higher levels of chronic stress [3, 36]. These factors can exacerbate the risk of obesity over time, emphasizing the importance of examining long-term neighborhood exposures [37].

If replicated, our findings could have important public health research implications. There is increased recognition that pedestrian-friendly infrastructure promotes physical activity and, in turn, reduces the prevalence of obesity and other chronic diseases. This study reported cumulative exposure to physical and social environments rather than point-in-time exposure to be associated with BMI. It may therefore be necessary to design the evaluation of place-based interventions to allow sufficient time of assessment to detect potential effectiveness in reducing obesity and other downstream outcomes, including diabetes and heart disease. Thus, long-term evaluations of such interventions may be required. Additionally, future studies could consider the combined effects of these neighborhood exposures on cardiometabolic outcomes.

There were notable strengths of this study. Few studies have tracked exposure to geospatial environments over a sufficient period of time to evaluate how cumulative, multi-year exposure to place may impact BMI. The geospatial measures of the built and social environment used in this study were objectively assessed. This study also had objective measures of BMI in later years of the study, with earlier in-person self-reported BMI highly correlated with the later objective measures. Additionally, this study adjusted for a number of longitudinally assessed covariates including indicators of socioeconomic status and earlier levels of BMI. There were, however, limitations that are important to consider. Although there was substantial sample dispersion to diverse neighborhoods throughout the USA (only 30% remained in Seattle by age 39), the sample originally recruited participants when they were elementary school students from public schools in an urban area, and, thus, results may not be generalizable to other populations (e.g., other age groups, other geographic regions). For example, compared to Washington State estimates from the 2002 ACS for adults 25 to 44 years old, participants in the sample had less educational attainment (26% vs. 9% less than high school education) [38] and slightly higher annual household income ($49,000 vs. $44,252) [39]. Additionally, most participants were racial minorities and from lower socioeconomic families. While this was not representative of the state, it was advantageous to include substantial subsamples from populations commonly underrepresented in other studies. While BMI is commonly used in most studies of physical health, it is a limited indicator of cardiometabolic health and mortality because it does not distinguish between fat and lean body mass [40, 41]. Future studies examining neighborhood factors may seek to obtain other measures of body composition and markers of cardiometabolic health. These measures may include waist circumference, body fat percentage, or cardiorespiratory fitness [40]. It is also possible that some participants were pregnant, influencing their BMI, although we did not have reason to expect disproportionate associations between pregnancy and neighborhood walkability or poverty. Indeed, at age 30, there was no statistically significant association of neighborhood walkability or neighborhood poverty with pregnancy during the past 3 years (p > 0.49). Lastly, an examination of potential mechanisms through which neighborhood walkability and poverty influenced BMI was beyond the scope of this study.

Conclusions

This study provides further evidence documenting the role of built and socioeconomic neighborhood environments on BMI. Sustained exposure over time to improvements in pedestrian infrastructure may be promising upstream approaches because of the large number of individuals that are typically impacted. Pedestrian infrastructure improvements have the additional advantage of potential health effects lasting well into the future. Study results suggest that such infrastructure improvements could have long-term impacts on residents’ behaviors which can manifest in significant changes in BMI.

Future observational studies may benefit from including longitudinal assessments of residential history in order to assess the role of cumulative exposure to neighborhood features. Similarly, natural experiments as well as randomized trials should seek to incorporate longer follow-ups in order to understand the cumulative impacts of place-based interventions on BMI and related health outcomes.

Supplementary Information

Below is the link to the electronic supplementary material.