Changes in the Recreational Built Environment and Youth BMI

Abstract

Objective:

Many cities have implemented programs to improve the recreational built environment. We evaluated whether neighborhood recreational built environmental changes are associated with change in body mass index (BMI).

Methods:

We performed a longitudinal assessment of association between the recreational built environment and BMI percent of 95^th percentile (BMIp95). Patient data from 2012–2017 were collected from electronic medical records including height, weight, sex, race/ethnicity, insurance type and address. BMIp95 was calculated. Environmental data including sidewalks, trails, Healthy Mile Trails and parks were collected. Patients’ neighborhood environments were characterized using proximity of features from home address. Multilevel linear regressions with multiple encounters per patient estimated effects of recreational features on BMIp95 and stratified models estimated effect differences.

Results:

Of 8,282 total patients, 69.0% were racial/ethnic minorities, half were insured by Medicaid, and 29.5% changed residence. Median BMIp95 was 86.3%. A decrease in BMIp95 was associated with park proximity in the full cohort (−2.85; 95% CI: −5.47, −0.24; p=0.032), children with obesity at baseline (−6.50; 95% CI: −12.36, −0.64; p=0.030) and privately insured children (−4.77; 95% CI: −9.14, −0.40; p=0.032). Healthy Mile Trails were associated with an increase in BMIp95 among children without obesity (1.00; 95% CI 0.11, 1.89; p=0.027) and children living in higher income areas (6.43; 95% CI: 0.23, 12.64; p=0.042).

Conclusions:

Differences in effect indicate that built environment changes may improve or exacerbate disparities. Improving obesity disparities may require addressing family-level barriers to the use of recreational features in addition to proximity.

What’s New:

The effects of increased access to recreational features on child BMI differ by insurance payor, baseline weight status and neighborhood characteristics. These differential effects, which may improve or worsen existing disparities, should inform the development and evaluation of recreational environment initiatives.

Introduction

Child obesity and overweight remain a public health priority affecting one third of children in the U.S.¹ Black, Latinx and low-income children are disproportionately affected by obesity, increasing their risk of multiple co-morbidities including cardiovascular disease and depression.² In addition to family and individual level factors, neighborhood characteristics including aspects of the recreational and food environment reinforce racial and socioeconomic disparites in child obesity.³

Multiple studies have noted associations between body mass index (BMI) and recreational environmental characteristics.⁴ The majority of literature in children focuses on parks or green spaces and features which enhance neighborhood walkability such as sidewalks and trails.⁴ Parks and green spaces have been shown to have a beneficial effect on BMI by increasing physical activity.⁵ Sidewalks and trails have also been associatied with physical activity and weight status in a handful of studies, however there is overall less consistent eveidence supporting a beneficial effect on BMI.⁶ Despite this, sidewalks and trails remain a key aspect of policy proposals aiming to promote physical activity and active transit potentially because these features are often feasible additions existing neighborhoods.⁷ However, these potentially beneficial recreational features are not equitably distributed or maintained.³ Safe, attractive recreational features which appear to be beneficial for child BMI are more likely to be located in higher socioeconomic status (SES), majority white neighborhoods.³

Cities across the United States have implemented programs to improve walkability and access to parks and other green spaces.⁸ While several programs to improve access to trails, sidewalks and other recreational opportunities have yielded evidence of increased physical activity, the majoirty of studies evaluating these natural experiments focus on adults and lack longitudinal BMI data to help determine program efficacy.⁸ Of the pediatric studies, few examine differences in the effect of recreational built environment changes by race/ethnicity, neighborhood social characteristics and whether the child changed residence during the study period (residential stability).⁹ Changing residence is common among children and may affect the relationship between the built environment and child weight status.¹⁰

Evaluating the effects of citywide recreational environment changes is important because these efforts may not have the same effect on child BMI across socioeconomic strata, race/ethnicity and sex due to differences in neighborhood perceptions, time available for recreation and cultural norms.^11,12 This potential differential highlights the importance of determining the effects of recreational built environment changes across a diverse cohort, in order to determine their efficacy and potential impact on child obesity disparities. Furthermore, there is dearth of literature focused on the effect of built environment characteristics on obesity risk among low-income, Black and Latinx children.⁹

From 2006 to 2017, the city of Durham in North Carolina implemented several citywide initiatives to improve the recreational built environment including the addition of new trails, parks, bike lanes and sidewalks.^13,14 Durham city public records indicate that from 2012 to 2017 the city added at least 19 miles of trails, over 17,000 linear feet of new sidewalks, and added or augmented at least 3 parks.¹⁴ Additionally, one-mile marked walking loops of continuous sidewalk called “Healthy Mile Trails” were built by a community coalition to increase recreational opportunities starting in 2012.¹⁵ In this study, we aimed to take advantage of a natural experiment and existing longitudinal data to determine whether neighborhood recreational built environment changes (parks, trails, sidewalks and Healthy Mile Trails) during this time were associated with a change in child BMI percent of the 95^th percentile. We also aimed to describe differences in the effect of recreational built environment changes by individual and neighborhood characteristics.

Methods

Study Population

Duke Children’s Primary Care is the largest single provider of pediatric primary care in the city of Durham, North Carolina with three sites within the city that see over 15,000 children who reside in Durham per year (approximately 25% of Durham children).¹⁶ All well visits for children under the age of 13 are labeled well child visits within our electronic medical record. Address of residence, sex, height, weight, race/ethnicity and type of insurance were collected from the electronic medical records of children presenting for annual well visits aged 5–12 years at the time of initial visit from 2012 to 2017 with residences within the city of Durham. Children’s clinical record data were excluded if they had: (1) a residential address outside of the city of Durham at any time from 2012 to 2017; (2) a physiologically implausible BMI measurement based on 2000 CDC Growth Chart reference data¹⁷ or (3) fewer than three well child visit height and weight measurements between 2012 and 2017 in 3 separate years. A small portion (3.1%) of our cohort was underweight (BMI <5^th percentile). We include underweight children in the analysis presented herein as our results were unchanged when we excluded underweight children. Similarly, as our cohort is based on well-child visits across high volume clinics we did not exclude children with chronic health problems.

Measures

BMI percent of the 95^th percentile (BMIp95) is used to overcome the mathematical limitations of BMI percentile when examining child obesity.^18,19 It is calculated by dividing each child’s BMI by the BMI which represents the 95^th percentile for that child.^18,19 We calculated BMIp95 using measured height, measured weight, age and sex at each well child visit using 2000 CDC Growth Chart Reference Data.¹⁷ Measures from each annual well child visit during the study period were included in the analysis. We collected data on each child’s age, sex and insurance type ( a proxy of child socioeconomic status²⁰) based on evidence that these factors may confound the relationship between BMI and the environment.³

Environmental data, described in Table 1, were extracted from Durham Parks and Recreation, the U.S. Census’ American Community Survey and Durham city maps created by Dataworks NC.^21,22 Similar to prior studies, the child’s surrounding recreational environment was characterized based on the presence of features within a 400m street network-based buffer around the child’s residence at the time of each clinic visit.^23,24 As the majority of parks, trails and healthy mile trails have multiple entry opportunities along their perimeter we defined access by the presence of any aspect of the feature within the child’s road network buffer.²⁴ Sidewalk access was defined as the number of unique sidewalk segments within the child’s road network buffer.

Table 1:

Environmental Exposures and Assignment to Patient Data

Environmental Exposure	Definition	Source	Data Availability	Assignment of Exposure to Patient Data
Defined via Road Network Buffera
Healthy Mile Trails	Presence within buffer	City of Durham	2012, 2013, 2014, 2015, 2016, 2017	Annual
Trails	Presence within buffer	Durham Parks and Recreation	2014 and 2017	2014: Patient data from 2012 to 2015 2017: Patient data from 2016 to 2017
Parks	Presence within buffer	Durham Parks and Recreation	2013 and 2017	2013: Patient data from 2012 to 2015 2017: Patient data from 2016 to 2017
Sidewalks	Number of sidewalk segments within buffer	City of Durham	2014 and 2017	2014: Patient data from 2012 to 2015 2017: Patient data from 2016 to 2017
Defined via Census Tract of Residenceb
Neighborhood socioeconomic status and racial/ethnic composition		American Community Survey	2010	All patient data
Community Reported Crimec		City of Durham	2014–2015	All patient data

^aBased on 400 meter walking buffer based on the road network using address of residence at each well child visit

^bBased on census tract of residence at baseline well child visit

^cCommunity reported crime incidents per square mile per census tract

Changes in the environmental features were calculated over time based on data availability (Table 1). For Healthy Mile Trials, a new trail was added annually during the study period and location data was available for each of the years from 2012 to 2017. Thus clinical data was matched to Healthy Mile Trail locations based on the presence of a Healthy Mile Trail within the child’s buffer during the calendar year of the clinic visit. Trail maps for Durham city trails were available in 2014 and 2017. The presence or absence of a trail within each child’s network buffer was determined using trail locations from the 2014 map from the beginning of the clinical data (2012) until 2015. Trails locations from the 2017 map were used from 2016 through the end of the clinical data (2017). Parks and sidewalks were matched to clinical data in a manner similar to trails using data from two years during the study period, summarized in Table 1.

Neighborhood demographic characteristics included racial/ethnic composition, household median income and community reported crime. These variables were selected because they may modify the relationship between the built environment and BMI.^24,25 They were matched to clinical data based on children’s census tract of residence at the time of their baseline well child visit. Household median income and racial/ethnic composition were derived from the American Community Survey (2010 single year estimate).²¹ Neighborhood community reported crime data were procured from publicly available data from the City of Durham (2014–2015). Community reported crime data represents the number of incidents of community reported crime per square mile for the child’s census tract of residence.²⁶

Statistical Analysis

We used QGIS Geographic Information System (Open Source Geospatial Foundation Project (2018)) to perform all geospatial calculations. We described the characteristics of children and their residential environments in the study sample using summary statistics. We estimated the average effect of recreational built environmental changes (i.e. Healthy Mile Trail, trail, park, and sidewalk access) on change in BMIp95. Next, we assessed heterogeneity of this effect by estimating stratum-specific effects by baseline weight status, child residential stability during the study period, sex, race/ethnicity, and insurance type. Finally, we assessed differences in this effect by estimating effects within strata of neighborhood-level demographic characteristics median household income, percent non-white, and community reported crime. BMIp95 was normally distributed permitting use of linear regression modeling. Correlation in outcomes by neighborhood were assessed using intraclass correlations. Neighborhood-level interclass correlations for each outcome variable were less than 0.1, so we did not cluster within neighborhoods.²⁷

The effects of neighborhood recreational built environment changes on BMIp95 were estimated using multilevel linear regression, with repeated measures nested within individuals, who are nested within neighborhoods. ²⁸ Models included a random intercept at the individual level and time as years since study entry. For each environmental factor, the estimated regression coefficient represents the change in BMIp95 associated with a single additional Healthy Mile Trail, park, trail or unique sidewalk segment to the child’s road network buffer.

All analyses were completed by using R version 3.5.2.²⁹ P values of ≤.05 were considered statistically significant. This study was approved by the Institutional Review Boards of Duke University and the University of North Carolina at Chapel Hill.

Results

Table 2 shows descriptive characteristics of the 8,282 children who met inclusion criteria. The median baseline age of children in the sample was 6.0 years. The cohort was racially and ethnically diverse, mirroring the racial composition of Durham, NC, including 37.8% identifying as non-Hispanic black and 15.9% Hispanic. Half of children were insured by Medicaid. Their median baseline BMIp95 was 86.3 and 18% of children had obesity (BMI percentile ≥95 percentile). Approximately thirty percent of children had two or more addresses of residence from 2012–2017. At baseline, children had limited access to recreational environmental features, with 50% of children having fewer than one Healthy Mile Trail, trail or park within their walking buffer at baseline. There were relatively modest changes in access to Healthy Mile Trails and parks during the study period with an increase of less than 1 Healthy Mile Trail or park per 100 children from 2012–2017. Increases in access to trails and sidewalks were more robust with increases of 2.8 trails per 100 children and 5.4 unique sidewalk segments per 100 children from 2012–2017.

Table 2:

Patient Characteristics, Durham Youth (Ages 5 to 12) from 2012 to 2017 (N = 8,282)

Characteristic	N (%)
Sex
Female	3,965 (47.9)
Male	4,317 (52.1)
Race/Ethnicity
American Indian or Alaska Native, Non-Hispanic	21 (0.3)
Asian, Non-Hispanic	396 (4.8)
Black or African American, Non-Hispanic	3131 (37.8)
Multiracial or 2 or More Races, Non-Hispanic	257 (3.1)
Native Hawaiian or Other Pacific Islander, Non-	14 (0.2)
Hispanic	1318 (15.9)
White or Caucasian, Non-Hispanic	2298 (27.7)
Other, Non-Hispanic	111 (1.3)
Missing or Unknown	736 (8.9)
Insurance Status
Medicaid	4,080 (49.3)
Private	4,202 (50.7)
Changed Address of Residence from 2012–2017	2,444 (29.5)
Obesity at Baseline Visit (≥95th percentile)	1488 (18.0)
	Median (25th, 75th percentile)
Age at Baseline Visit	6.0 (5.0, 8.0)
Baseline Percent 95th Percentile	86.3 (79.7, 95.4)
Baseline Built Environment Features a
Healthy Mile Trail Access	0.0 (0.0, 0.3)
Trail Access	0.0 (0.5, 2.4)
Park Access	0.0 (0.1, 1.4)
Sidewalk Access	1.0 (0.1, 1.8)
Modifying Built Environment Characteristics
Community Reported Crime (2014–2015) c	97.0 (12.2, 695.2)
Census Tract Median Household Income (2010)	$47,450 (28,186, 77,734)
Census Tract Percent Non-white (2010)	57.1 (27.8, 86.8)
	Difference (Range)
Change in Access to Built Environment Measures 2012–2017 b
Healthy Mile Trail Access	0.3 (0.0, 0.4)
Trail Access	2.8 (1.0, 4.4)
Park Access	0.2 (0.0, 0.2)
Sidewalk Access	5.4 (2.0, 8.5)

^aNumber in walking buffer

^bNumber in walking buffer per 100 children

^cCommunity reported crime incidents per square mile per census tract

As shown in Table 3, in the full cohort, parks were associated with a change in BMIp95 of −2.85 (95% CI: −5.47, −0.24; p=0.032). Healthy mile trails, sidewalks and trails were not associated with change in BMIp95. Analyses were also performed with adjustment for baseline BMIp95 with identical results.

Table 3:

Estimated Associations Between Change in Built Environment Characteristics and Percent of 95^th BMI Percentile, Stratified by Patient Characteristics, Durham Youth (Ages 5 to 12) from 2012 to 2017 (N = 8,282)^a,b

	Change in BMI Percent 95 Associated with 1-Unit Increase
	Heathy Mile Trail Access	Trail Access	Park Access	Sidewalk Access
	Beta (95% CI) p-value	Beta (95% CI) p-value	Beta (95% CI) p-value	Beta (95% CI) p-value
All Children	0.34 (−0.61, 1.29) 0.480	−0.04 (−0.42, 0.34) 0.837	−2.85 (−5.47, −0.24) 0.032	0.09 (−0.20, 0.37) 0.566
Obesity
Obese (BMI Percentile ≥95)	−0.30 (−2.85, 2.25) 0.816	0.11 (−1.11, 1.32) 0.863	−6.50 (−12.36, −0.64) 0.030	0.00 (−0.91, 0.92) 0.999
Non-obese (BMI Percentile <95)	1.00 (0.11, 1.89) 0.027	−0.07 (−0.39, 0.25) 0.663	−1.67 (−4.20, 0.85) 0.194	0.10 (−0.14, 0.34) 0.401
Residence
Changed	0.14 (−1.06, 1.33) 0.823	−0.17 (−0.65, 0.31) 0.491	−2.17 (−5.91, 1.57) 0.255	−0.06 (−0.43, 0.31) 0.768
Did not change	0.82 (−0.99, 2.64) 0.375	0.42 (−0.26, 1.11) 0.228	−3.65 (−7.56, 0.26) 0.067	0.41 (−0.09, 0.91) 0.104
Sex
Female	0.32 (−0.96, 1.61) 0.620	0.05 (−0.49, 0.59) 0.864	−3.27 (−7.71, 1.17) 0.149	0.19 (−0.24, 0.63) 0.380
Male	0.34 (−1.06, 1.75) 0.630	−0.13 (−0.66, 0.39) 0.626	−2.71 (−5.95, 0.52) 0.100	0.00 (−0.38, 0.38) 0.989
Race/Ethnicity
Black, non-Hispanic	–0.64 (−1.94, 0.65) 0.328	−0.13 (0.78, 0.53) 0.705	−0.58 (−4.65, 3.48) 0.778	−0.03 (−0.51, 0.45) 0.901
Hispanic	1.71 (−0.50, 3.92) 0.129	0.11 (−0.66, 0.88) 0.781	−2.58 (−8.40, 3.25) 0.386	0.55 (−0.08, 1.19) 0.089
White, non-Hispanic	−0.80 (−5.32, 3.73) 0.730	−0.36 (−1.16, 0.43) 0.369	---	−0.09 (−0.57, 0.40) 0.723
Insurance
Medicaid	0.66 (−0.46, 1.79) 0.248	0.09 (−0.42, 0.61) 0.723	−2.52 (−5.89, 0.85) 0.143	0.10 (−0.29, 0.49) 0.627
Private	−2.15 (−4.30, 0.01) 0.051	−0.27 (−0.84, 0.29) 0.340	−4.77 (−9.14, −0.40) 0.032	−0.05 (−0.47, 0.37) 0.813
Age Group
5–9 years	0.29 (−0.68, 1.26) 0.556	−0.01 (−0.40, 0.38) 0.972	−2.70 (−5.30, −0.10) 0.042	0.13 (−0.17, 0.43) 0.391
10–12 years	1.35 (−3.06, 5.75) 0.549	−0.33 (−1.67, 1.01) 0.633	−23.64 (−61.66, 14.37) 0.223	−0.33 (−1.32, 0.67) 0.517

^aEstimate represents the effect of a 1 unit increase within a 400 meter walking buffer based on the road network

^bModels adjusted for age, sex, and insurance type as appropriate

Effect Modification by Patient Demographic Characteristics and Weight Status

The effect of recreational built environment features differed by child baseline weight status, insurance type and age controlling for child demographic characteristics, also shown in Table 3. Gaining access to one park was associated with a decrease in BMIp95 of 6.50 (95% CI: −12.36, −0.64; p=0.030) for children with obesity at baseline and a decrease of 4.77 among children with private insurance (95% CI: −9.14, −0.40; p=0.032). For children age 5–9 an additional park was associated with a decrease in BMIp95 of 2.70 (95% CI: −5.30, −0.10; p=0.042). Increased Healthy Mile Trail access was associated with an increase in BMIp95 of 1.00 (95% CI: 0.11, 1.89; p=0.027) for children without obesity at baseline. There were no differences in the effect of recreational built environment features on BMIp95 by race/ethnicity, sex or residential stability.

Effect Modification by Patient Neighborhood Characteristics

As shown in Table 4, the effect of park and Healthy Mile Trail access differed by child residential neighborhood characteristics controlling for age, sex and insurance type. Increased access to Healthy Mile Trails was associated with an increase in BMIp95 of 6.43 (95% CI: 0.23, 12.64; p=0.042) for children living in neighborhoods with median household income greater than or equal to the Durham city median household income of $54,062. Increased park access was associated with a decrease in BMIp95 of 5.45 in areas with a number of community reported crime events greater than or equal to the median crime rate for the study population (95% CI: −9.84, −1.06; p=0.015). There were no significant differences in the effect of built environment features by neighborhood percent non-white race.

Table 4:

Estimated Associations Between Change in Built Environment Characteristics and Percent of 95^th BMI Percentile, Stratified by Patient Neighborhood Characteristics, Durham Youth (Ages 5 to 12) from 2012 to 2017 (N = 8,282)^a,b

	Change in BMI Percent 95 Associated with 1-Unit Increase
	Heathy Mile Trail Access	Trail Access	Park Acess	Sidewalk Access
	Beta (95% CI) p-value	Beta (95% CI) p-value	Beta (95% CI) p-value	Beta (95% CI) p-value
Median Household Income c
<$54,062	0.05 (−0.98, 1.07) 0.926	−0.12 (−0.57, 0.33) 0.602	−2.99 (−6.10, 0.12) 0.060	0.01 (−0.31, 0.33) 0.962
≥$54,062	6.43 (0.23, 12.64) 0.042	−0.10 (−0.88, 0.67) 0.793	−3.23 (−8.83, 2.36) 0.257	−0.07 (−0.96, 0.83) 0.885
Percent non-white d
< 57.1%	−2.03 (−4.97, 0.91) 0.175	−0.02 (−0.53, 0.48) 0.925	−3.20 (−8.71, 2.31) 0.255	−3.20 (−8.71, 2.31) 0.255
≥ 57.1%	0.41 (−0.66, 1.48) 0457	−0.03 (−0.59, 0.54) 0.927	−2.99 (−6.12, 0.14) 0.061	0.12 (−0.25, 0.50) 0.517
Crime d
< 97	---	−0.18 (−1.37, 1.01) 0.769	−4.31 (−8.78, 0.16) 0.059	−0.51 (−2.25, 0.23) 0.565
≥ 97	0.52 (−0.62, 1.67) 0.371	−0.28 (−0.86, 0.30) 0.348	−5.45 (−9.84, −1.06) 0.015	−0.07 (−0.46, 0.33) 0.731

^aEstimate represents the effect of a 1 unit increase within a 400 meter walking buffer based on the road network

^bModels adjusted for age, sex, and insurance type

^cMedian for Durham County in 2010

^dMedian for study population

Discussion

Using longitudinal measures from a diverse cohort of children, we found that changes in the recreational built environment were associated with change in BMIp95. Parks were associated with a decrease in BMIp95 for the full study cohort with stratum specific effects among children with obesity at baseline, those with private insurance and children aged 5–9. We also observed a handful of relationships between BMIp95 and the recreational environment in an unexpected direction. Most notably, healthy mile trails were associated with increased BMIp95 among children living in higher income areas. Cumulatively, these results indicate that recreational built environment changes affect children differently by weight status as well as individual and neighborhood sociodemographic characteristics.

Our findings with regard to Healthy Mile Trails were surprising, as one would expect these one mile marked loops of sidewalk to promote physical activity and thus potentially lower BMI. However there are several potential explanations for these results. The first is the presence of unmeasured environmental factors such as the surrounding food environment which could have led to an increase in BMI for children living near Health Mile Trails. A second explanation, which could work in concert with the first, is that children or families with children may not find the Healthy Mile Trails attractive compared to the features offered by parks or playgrounds leading to low utilization of Healthy Mile Trails by our cohort.^30,31 Our study findings for both Healthy Mile Trails and sidewalks are similar to prior inquiries into the relationship between sidewalks and BMI showing either no association or a direct association between BMI and sidewalk density.^32,33 These studies along with our own reflect that Healthy Miles trails and sidewalks are not isolated environmental features. While they may promote physical activity, potential beneficial effects on BMI may be countered by surrounding environmental features such as restaurants and corner stores, many of which may be rendered more accessible via new sidewalks. ⁶ Subsequent examinations of the relationship between the addition of new sidewalks and change in child BMI warrant closer examination of surrounding environmental features as well as dietary and physical activity data in order to more clearly ascertain the effects of sidewalks and similar features like Healthy Mile Trails on child obesity.⁶

Our finding that parks were associated with decreased BMIp95 indicates the significance of parks as an environmental obesity determinant. Studies using direct observation or accelerometry have shown increases in child physical activity due to new or renovated parks.^34,35 Interventions which improve the proximity and quality of local parks have been effective across a broad array of racial and ethnic groups and children living in low SES neighborhoods.^34–36 In the present study, the overall increase in number of parks during our study period was modest, thus one must consider whether our findings are due to parks that were added or augmented during the study period or due to children who moved to neighborhoods in closer proximity to Durham parks. Importantly, our study adds to existing literature by demonstrating a particular benefit for parks among children with obesity supporting a role for park programming and park prescriptive programs as part of obesity treatment.³⁸ That parks were more beneficial for children with private insurance may signal differences in park utilization or quality by socioeconomic status. Safer, more attractive parks tend to be more accessible to children living in higher income households.³ Thus, the effect of parks on child BMI may be maximized by ensuring equitable quality and safety as well as geographic access.³⁹

Prior studies have shown the effect of recreational built environment features on BMI may be less beneficial in higher crime areas compared to areas with lower crime.^25,40 Surprisingly, in the present study, increased park access was associated with a decrease in BMIp95 in areas with greater community reported crime than the study population median. This result may be attributable to the difference between crime and perceived safety.⁴¹ Our use of community reported crime as a measure may not reflect families’ perceptions of safety, a factor known to reduce park use.⁹ Perceptions of safety may not be reflected in crime data.^41,42 Differences in perception of recreational features warrant further study as they may contribute to disparities in the effects of initiatives to improve access to recreational opportunities.⁴³ Integrating longitudinal data describing family and child perceptions of parks and other recreational features with geographic and BMI data may yield a better understanding of the environmental characteristics which contribute to improved access among minority and low-income children in particular.

Nearly one third of children in our cohort changed residences during the 5-year study period. It has been noted that changing residence during childhood or adolescence may be protective against obesity due to exposure to improved community characteristics within the new neighborhood context.¹⁰ In our cohort, the effect of neighborhood environmental exposures on BMIp95 was relatively similar by residential stability. However, it is important to note that children who moved outside of the city of Durham were excluded from the study and children with one move were grouped with children with multiple moves over the study period. It will be important to include children with residential instability in future natural experiment studies with individual-level utilization data (i.e. accelerometers or pedometers) to better understand the variability of their neighborhood exposures over time and the effect of residential instability on the utilization of nearby recreational features.

Study Limitations and Strengths

Similar to other assessments of natural experiments, our study is limited by the inability to control for factors such as the placement and aesthetic condition of recreational environment features which may impact their use by children, adolescents and their families. Our results may also have been influenced by the distribution of our study population across the city of Durham, although our cohort is demographically representative of our city and includes at least one child from nearly every census tract in the city. A critical limitation of our work a lack of utilization data to describe the extent to which children utilized the built environment features studied. Our lack of utilization data increases the likelihood that our results are at least partially the result of unmeasured neighborhood or individual-level factors. We also note that we defined parks based on ownership by the Durham Department of Parks and Recreation, which excluded school greenspaces and informal private or public areas where children or families might be active. However, it is notable that Durham Parks and Recreation has the largest network of parks in the city and all are publicly accessible without cost. Lastly, our findings may also be the result of secular weight changes given the lack of a matched control group for comparison.

Natural experiments have the advantage of evaluating real-world changes at the population-level which can be translated into community or policy-level change.⁴⁴ Our study limitations are balanced by several strengths including our use of longitudinal measures allowing us to estimate the time-varying effect of both exposure and outcome. This study design renders our effect estimates less vulnerable to self-selection bias- whereby individuals prone to be more physically active live closer to recreational opportunities- than cross-sectional studies. Another strength of our study is our diverse cohort of children enabling the estimation of effects for subgroups of children with a higher prevalence of overweight and obesity. Finally, our study highlights the application of geographic data to health disparities and the potential of health system-community partnerships combining clinical and city data. These cross-sector data sharing partnerships can be leveraged to plan and evaluate environmental interventions, echoing a growing call for the application of geographic and social determinants to policy and clinical care.⁴⁵

Conclusion

Using longitudinal measures over a 5-year period, increased park access was associated with a decrease in child BMIp95. This effect was particularly evident among children with obesity at baseline. We also demonstrate that the effects of some recreational features differ by individual and neighborhood-level sociodemographic characteristics. The potential for a differential effect, which may improve or worsen existing disparities, should inform the way that recreational environment initiatives are planned and evaluated. Improving obesity disparities will likely require interventions which address multiple barriers to the use of recreational features and cross-sector partnerships to plan and evaluate interventions.

Abbreviations:

BMI

body mass index

BMIp95

body mass index percent 95^th percentile

SES

socioeconomic status