Food Environments Within and Outside of Schools Play a Critical Role in Curtailing the Rise in Obesity among School-Aged Children over Time

Abstract

Background

Sound evidence for effective community-based strategies is needed to curtail upward trends in childhood obesity in the United States (US).

Objectives

The aim of the study was to assess the association between school and community food environments and the prevalence of obesity over time.

Methods

Data were collected from K–12 schools in 4 low-income New Jersey cities in the US. School-level obesity prevalence, calculated from nurse-measured heights and weights at 4 time points, was used as the outcome variable. Data on the school food environment (SFE) measured the healthfulness of school lunch and competitive food offerings annually. The community food environment (CFE), i.e., the number of different types of food outlets within 400 m of schools, was also captured annually. The count and presence of food outlets likely to be frequented by students were calculated. Exposure to composite environment profiles both within schools and in communities around schools was assessed using latent class analysis. Data from 106 schools were analyzed using multilevel linear regression.

Results

The prevalence of obesity increased from 25% to 29% over the course of the study. Obesity rates were higher in schools that had nearby access to a greater number of limited-service restaurants and lower in schools with access to small grocery stores and upgraded convenience stores participating in initiatives to improve healthful offerings. Interaction analysis showed that schools that offered unhealthier, competitive foods experienced a faster increase in obesity rates over time. Examining composite food environment exposures, schools with unhealthy SFEs and high-density CFEs experienced a steeper time trend (β = 0.018, P < 0.001) in obesity prevalence compared to schools exposed to healthy SFE and low-density CFEs.

Conclusions

Food environments within and outside of schools are associated with differential obesity trajectories over time and can play an important role in curtailing the rising trends in childhood obesity.

Introduction

The prevalence of obesity among school-age children in the United States (US) continues to grow [1] and exhibits significant sociodemographic disparities, as children from low-income households experience obesity at nearly twice the rates observed among children from higher-income households [2]. Racial/ethnic disparities are just as alarming, with substantially higher rates among Hispanic and non-Hispanic Black children than non-Hispanic White and non-Hispanic Asian children [1].

The etiology of childhood obesity is complex and includes a wide array of factors across the social-ecological spectrum, ranging from genetic disposition to the influence of family, school, community, and social policy [3]. Following recommendations from the Centers for Disease Control and Prevention (CDC) [4] and the National Academies of Medicine [3], public health approaches have increasingly focused on policies and interventions that influence contextual elements in children’s food environments within schools and communities [[5], [6], [7], [8], [9]]. Numerous such initiatives in the US, including the Healthy Hunger-Free Kids Act (HHFKA), a federal policy aimed at improving the nutritional quality of school meals [10], the CDC-supported state and local obesity prevention programs that promote healthy eating and active living [11], and the Robert Wood Johnson Foundation’s Healthy Kids, Healthy Communities initiative designed to support healthier communities [12], among numerous other community-level efforts [13,14], have successfully improved children’s food environments. For instance, healthier reimbursable meals and a la carte options have become more available in schools over time [15], and convenience stores participating in community initiatives have been shown to carry healthier foods [16].

A few large-scale research studies have examined the relationship between obesity trends and children’s exposure to various school and community-based obesity prevention programs and policies implemented by communities. The Healthy Communities Study [17] assessed the impacts of various programs and policies implemented over a 10-y period in 130 communities across the US on K–8th grade children’s weight. Study findings suggested that more intensive programs and policies—those with wider reach, of longer duration, and with behavior change-focused strategies—were associated with lower child BMI (in kg/m²) [18]. The Childhood Obesity Declines Project [19,20] examined community strategies that may have contributed to observed declines in childhood obesity prevalence in certain communities. Findings indicated that the layering of strategies (e.g., farm-to-school programs, improving school food nutrition standards, active design guidelines for new city construction) across different social-ecological spheres of influence may have a synergistic impact in reducing childhood obesity [20]. Both studies provide important insights about the potential of school and community-based obesity prevention efforts, but aspects of the research designs, including reliance on informant recollections to identify program and policy changes implemented over an extended period of time, undermined the studies’ ability to make causal inferences. In addition, these studies did not measure actual changes in the environment or an individual’s likely exposure to the identified program and policy changes, limiting the ability to identify specific interventions or combinations of interventions that were associated with changes in children’s weight outcomes.

Although many policy and intervention initiatives have focused on improving both school food offerings and the availability of healthy food outlets in communities [15,16,21,22], sound evidence of their effectiveness in curtailing rising trends in obesity is limited. The current study aims to examine the association between school and community food environments and trajectories of obesity prevalence and to identify specific elements of children’s school and community environmental exposures that merit inclusion in interventions. Results from the current study can fill a critical gap in knowledge and inform future policies and interventions in schools and communities to reduce the prevalence of childhood obesity.

Methods

Using a longitudinal design tracking schools over time, data were collected at multiple time points from a sample of K–12 public schools and surrounding communities in 4 New Jersey cities in the US with predominantly low-income populations over 7 y from school year (SY) 2013–14 to SY 2019–2020. Data collected in schools included professionally measured de-identified student heights and weights at 4-time points over the study period and information on types of food served in schools to capture the school food environment (SFE) each year. Data collected in communities included counts of various types of food outlets around schools for each year of the study, to capture the community food envrironment (CFE). Given the de-identified nature of student-level data, the study tracked schools over time rather than students; as a result, all outcome and exposure variables referred to the school level, and schools were the units of analysis.

Datasets and variables

Student variables and school demographics

All public schools in New Jersey are required to measure the heights and weights of students enrolled in grades K through 12 every SY [23]. The study team obtained nurse-measured height and weight data for SYs 2013–14, 2015–16, 2017–18, and 2019–20 from 141 public schools located in Camden, New Brunswick, Newark, and Trenton. In addition to student height and weight, the measurement records from the schools also included the sex, race, and age of the students. Following standard CDC procedures [24], age- and sex-specific BMI percentiles and modified z-scores were calculated for each child in the sample. Using the BMI percentile variable, we created 3 binary (yes/no) variables indicating whether the child had overweight/obesity (BMI ≥85th percentile), obesity (BMI ≥95th percentile), or severe obesity (BMI ≥120% of the 95th percentile) [25].

To aggregate individual-level weight outcomes to the school level, we ran a series of regression models (ordinary least squares linear regression for continuous outcomes and logit regression for binary outcomes), specifically 1 per school, with the individual-level weight measures (obesity status, modified BMI z-score) as the outcome, a dummy variable for the SY, and individual-level child sex, age, and race as control variables. We then used postestimation commands (i.e., margins in Stata 17 [26]) to obtain the adjusted mean (or prevalence) of the weight outcomes at the school level (e.g., percent students with obesity or mean modified BMI z-scores) for each SY for which the school provided the data. This step was necessary to adjust for potential demographic changes in the student body composition across SYs within each school. In most schools, such differences over the years were minimal among the students for whom BMI data were available; in these instances, the adjusted means were virtually identical to the observed (i.e., unadjusted) means.

Data from the National Center for Education Statistics common core data repository captured school-level factors, such as the proportion of students eligible for free or reduced-price meals as a proxy for school-level income, total student enrollment, and racial/ethnic composition of enrolled students, for each of the SYs under investigation [27]. Lastly, we obtained information from the New Jersey Department of Education and Food Research Action Center websites about whether the schools in our sample participated (yes/no) in the Community Eligibility Provision (CEP) for each SY [28,29]. CEP, a policy implemented by the USDA, allows schools in lower-income communities to offer no-cost meals to all enrolled students [30].

SFE

To characterize the food environment in schools, we sent all public schools in the study cities a survey that included questions about foods and beverages available to students in various school venues (e.g., cafeterias and vending machines). Survey questions were derived from prior studies [[31], [32], [33]] and have been used in other published research [15,34,35]. Surveys were sent to the school nurse, and SFE questions were completed in consultation with the school food service professionals. The survey could be completed on paper or using an online survey platform (Qualtrics). Respondents were prompted to answer questions pertaining to both the current and the previous SY.

Survey questions captured the availability of food and beverages in 3 separate domains: 1) the National School Lunch Program (NSLP), 2) a la carte items sold in the school cafeteria that are not part of the NSLP, and 3) vending machines. For each domain, we created 2 indices, 1 capturing the number of healthy foods and beverages and 1 capturing the number of unhealthy foods and beverages available to students. Appendix A includes the full list of items included within each index. Additional information on the food environment index creation is available elsewhere [15]. To capture the healthfulness of the SFE, we created 2 summary measures (range 0–1), with higher scores corresponding to healthier environments. The first, the NSLP healthy scale, was the ratio of the number of healthy NSLP items to the total (healthy + unhealthy) number of NSLP items. For instance, a score of 0.5 indicates that half of the NSLP items offered are healthy. The second, the Competitive food healthy scale, was created in a similar fashion and included the remaining 2 domains—a la carte offerings and vending machines. Again, the number of healthy items was divided by the total number of items (healthy + unhealthy), creating a scale representing the proportion of healthy items available as competitive foods.

Over the course of the study, 129 schools provided student heights and weights and SFE data. Of these, 23 schools were excluded because they only provided data for 1 SY. Of the remaining 106 schools, 26 provided data for 2 SYs, 39 for 3 SYs, and 41 for 4 SYs, for a total of 333 data points (72 in SY 2013–14, 91 in 2015–16, 93 in 2017–18, and 77 in 2019–20). Differences in a number of schools across SYs with complete data resulted from schools opening and closing because of the renovations and/or district student enrollment needs and from nonresponse. Mean response rates were 71% for height and weight data and 91% for SFE surveys.

CFE

Data on the characteristics and location of food outlets located in study cities and in a 1.6 km radius around the city boundary were obtained from 2 commercial data companies, InfoUSA [36] and Trade Dimensions/Nielsen [37], for each year during the study period. Using a systematic classification protocol, food outlets in these databases were de-duplicated, geocoded, and categorized into 4 groups likely to be frequented by school-age children: small grocery stores, convenience stores, upgraded convenience stores, and limited-service restaurants [8,38]. Supermarkets were omitted because of their consistently low proximity to schools in the study cities. Small grocery stores were characterized as stores with sales between $1 million and $2 million that carried ≥3 of the following healthy options, which were confirmed through phone calls to the store: 5 different types of fruits, 5 different types of vegetables, low-fat or skim milk, and fresh or frozen meat. Convenience stores were defined as: 1) stores with annual sales below $1 million; 2) stores from larger chains such as 7-Eleven or Wawa; or 3) stores with sales between $1 million and $2 million but not meeting the criteria to be classified as small grocery stores. Upgraded convenience stores were a special group of small stores that participated in local healthy corner store initiatives aimed at increasing the availability of healthy items in small stores. Information on upgraded convenience stores was obtained annually from local agencies spearheading the initiatives. Given the prominence of this initiative in all the study cities, we removed these outlets from the convenience store category and classified them separately as upgraded convenience stores [21]. Limited-service (i.e., fast-food) restaurants were characterized as establishments where patrons paid before rather than after, eating. To characterize the CFE around each school, consistent with previous literature [8,39,40], we calculated the number of convenience stores, limited-service restaurants, small grocery stores, and upgraded stores within easy walking distance (400 m roadway network) from each school using ArcGIS (ArcGIS Desktop 10.2, Esri, (2013) Redlands, CA), applying methodology described elsewhere [8]. Because small grocery stores and upgraded convenience stores were far less prevalent, in the analysis, we used 2 binary variables indicating the presence (yes/no) of these outlet types. The count variables for limited-service restaurants and convenience stores were topcoded (i.e., values were recoded into the mean plus 3 SDs if they were above this threshold) to reduce the influence of extreme values on the estimated coefficients.

Statistical analysis

All analyses were run using Stata 17 [26], with 2-sided alpha = 0.05. The proportion of missing data was minimal, and the analyses were based on complete data from 106 schools over the study years, incorporating data from over 19,000 students/y. Univariate and bivariate statistics were used to analyze the distributions of the study variables.

Regression analysis

To model the trend in the prevalence of obesity (the main outcome) at the school level, we utilized a multilevel linear regression model (command mixed) with a random intercept to account for repeated measurements. The option cluster was added to adjust for the nested structure of the data at the school-district level (with 106 schools from 4 school districts). The time variable indicated the SY, coded as 0 = 2013–14, 2 = 2015–16, 4 = 2017–18, and 6 = 2019–20, and was used as a continuous variable so that the estimated coefficient represented the predicted yearly change in the outcome. The main exposure variables were the NSLP healthy scale and the competitive food healthy scale, which together characterized the SFE, as well as the number of convenience stores, the number of limited-service restaurants, and the 2 binary variables indicating the presence of small grocery stores and upgraded convenience stores, respectively. The last 4 variables together characterized the CFE within a 400-m roadway network around each school. To account for variation in obesity prevalence by school-level factors, control variables in the analyses included an indicator of school level (elementary compared with middle and high schools), the proportion of enrolled students eligible for free and reduced-price meals, and the proportion of enrolled students from different racial/ethnic groups. Because there was considerable between-school variation in the number of students for whom height and weight data were available (first quartile: 77 students, median: 152 students, third quartile: 397 students), frequency weights to adjust for these differences in sample size across schools were created and used in all regression models. Results from this initial model were used to obtain the overall linear trend in obesity over the 7-y study period, net of the SFE, the CFE and school characteristics.

Interaction analysis

To understand whether and how the healthfulness of the SFE and the CFE was associated with the trend in obesity, we expanded the model described above by adding a set of interaction terms between the time variable and each of the 6 variables characterizing the SFE and the CFE. These interaction terms allowed the time trend to differ across values of the 6 food environment variables. To ensure that the findings from the interaction model were robust, we ran additional models using 3 different outcome variables (proportion with overweight/obesity; proportion with severe obesity; modified BMI z-score – all continuous variables) and another model adding a binary time-varying control variable indicating whether the school participated in CEP.

Latent class analysis

Next, latent class analysis (LCA) was used to group the sample schools based on the characteristics of their food environment. The goal was to analyze the trajectory of obesity prevalence for schools with different food environment profiles. LCA identifies underlying categorical latent variables that are constructed based on the distribution of directly measured variables [41]. Through LCA (see Appendix B for additional details), we identified the following 4 classes to categorize schools in the sample: 1) Healthier SFE and low-density CFE; 2) Unhealthier SFE and low-density CFE; 3) Healthier SFE and high-density CFE; 4) Unhealthier SFE and high-density CFE with fewer upgraded stores (see Table B1 with the descriptive statistics). Based on these 4 classes, we created a 4-category variable named “environment type.” This variable was used as a predictor in the regression model with interactions described above in lieu of the single food environment predictors. After running the regression model, postestimation commands were used to summarize the regression results by showing the estimated trajectory of obesity prevalence for schools that were exposed to the 4 different food environment types while setting all other variables at their mean. The 4 slopes were predicted using the margins command set to estimate the marginal effects of the “environment type” variable. With this command, we also assessed whether each slope was different from zero based on z-tests.

TABLE 1.

Descriptive characteristics of the sample. Mean (SD) for continuous variables and percentage for proportions and categorical variables

	2013–14	2015–16	2017–18	2019–20
School demographics
School level
Primary (%)	76	75	73	70
Secondary (%)	24	25	27	30
Students eligible for FRPM (%)	90	76	78	77
Hispanic students (%)	47	47	51	54
Non-Hispanic Black students (%)	49	49	44	41
Non-Hispanic White/other students (%)	4	4	5	5
Students with overweight/obesity (%)	43.5	45.2	43.9	48.2
Students with obesity (%)	25.3	27.3	25.5	28.8
Students with severe obesity (%)	9.1	10.0	10.3	11.3
Modified BMI z-scores	0.72 (0.2)	0.78 (0.3)	0.70 (0.3)	0.85 (0.3)
Food environment in school
NSLP – healthy scale (range 0–1)	0.71 (0.1)	0.70 (0.1)	0.74 (0.1)	0.76 (0.1)
Competitive foods – healthy scale (range 0–1)	0.63 (0.3)	0.65 (0.2)	0.68 (0.3)	0.67 (0.2)
Food environment outside schools (400 m)
Limited-service restaurants (range 0–16)	2.5 (3.4)	2.4 (2.9)	2.4 (3.1)	2.7 (3.3)
Convenience stores (range 0–6)	2.1 (1.7)	2.2 (1.7)	1.6 (1.5)	1.4 (1.4)
Presence of small grocery store (% yes)	18	27	26	19
Presence of upgraded store (% yes)	17	18	34	34
Total number of schools (N)	72	91	93	77

FRPM, free and reduced-price meals; NSLP, national school lunch program.

Analyzing overall healthy compared with overall unhealthy environmental profiles

Finally, a similar approach based on postestimation commands was used to compare obesity trajectories over time in a scenario where all aspects of the food environment were set to either the healthy or the unhealthy ends of the distribution for each of the 6 food environment variables. This approach complements the LCA approach, which uses empirically derived classes, by creating policy-focused a priori categories. Specifically, this complementary analysis highlights the potential impacts of policies that move schools from very poor to excellent performance across all 6 environment variables simultaneously. Using regression results from the model that included interaction terms for all 6 elements of the food environment with time, we compared how 2 contrasting scenarios were associated with different trends in obesity over time. To define a healthy and an unhealthy profile for each of the 6 exposure variables measuring the healthfulness of the SFE and the CFE, we used specific values from the overall distribution for each of these variables, as shown in Box 1 and fully explained in Appendix C. This approach, based on previous literature [42], was used to create 2 contrasting composite environment profiles: a composite healthier food environment profile, where all elements of the food environments within schools and in communities around schools are healthier, and a composite unhealthier food environment profile, where all these elements are unhealthier.

Box 1. Description of the composite healthier food environment and the composite unhealthier food environment profiles.

The composite healthier food environment profile had the following characteristics:	The composite unhealthier food environment profile had the following characteristics:
NSLP Healthy Scale = 0.89 (90th percentile)	NSLP Healthy Scale = 0.61 (10th percentile)
Competitive Food Healthy Scale = 1 (90th percentile)	Competitive Food Healthy Scale = 0.37 (10th percentile)
Number of limited-service restaurants = 0 (10th percentile)	Number of limited-service restaurants = 8 (90th percentile)
Number of convenience stores = 0 (10th percentile)	Number of convenience stores = 3 (90th percentile)
Presence of small grocery stores = 1 (Yes)	Presence of small grocery stores = 0 (No)
Presence of upgraded stores = 1 (Yes)	Presence of upgraded stores = 0 (No)

NSLP, national school lunch program.

Alt-text: Box 1

We then estimated the obesity prevalence for each study year for these 2 composite profiles to compare the obesity trends associated with the composite healthier and unhealthier food environment profiles.

Results

Table 1 shows the sample characteristics at each measurement period during the study. Most schools were elementary schools (≥70%), and the majority of students were Hispanic or non-Hispanic Black and eligible for free and reduced-price meals. Over time, all 4 weight outcome measures tended to increase, indicating an overall upward trend in overweight, obesity, severe obesity, and modified BMI z-scores.

The adjusted base model (Table 2) shows that obesity became more prevalent in our sample, with an estimated increase of 0.63% points every year, or ∼3.78 (0.63 ∗ 6)% points over the study period. Overall, obesity rates tended to be higher in schools that had nearby access to limited-service restaurants and lower in schools with access to small grocery stores and upgraded convenience stores. Obesity rates were higher in schools with a higher proportion of enrolled students eligible for free and reduced-price meals and in schools with a higher proportion of Black and Hispanic students.

TABLE 2.

Results from the base (i.e., without interactions) linear regression model analyzing obesity rate over time

	Coefficient	95% CI - low	95% CI - high	P value
Time (yearly change)	0.006	0.000	0.012	0.044
NSLP - healthy scale	–0.035	–0.107	0.037	0.344
Competitive foods - healthy scale	0.008	–0.050	0.067	0.780
Limited-service restaurants	0.002	0.000	0.004	0.024
Convenience stores	–0.002	–0.004	0.001	0.143
Presence of small grocery store	–0.010	–0.016	–0.004	0.001
Presence of an upgraded store	–0.023	–0.036	–0.011	<0.001
School level (reference: primary)
Secondary	0.022	–0.010	0.055	0.182
% of students eligible for FRPM	0.088	0.011	0.165	0.025
% of Hispanic students	0.156	0.131	0.182	<0.001
% of non-Hispanic Black students	0.088	0.066	0.109	<0.001

CI, confidence interval; FRPM, free and reduced-price meals; NSLP, national school lunch program.

CI, confidence interval; FRPM, free and reduced-price meals; NSLP, national school lunch program.

Table 3 presents the results from the interaction models, which allowed the time trend to vary across each food environment variable. The interaction term between year and the competitive foods healthy scale was negative and significant (β = –0.013; P = 0.038), indicating that the time trend in obesity is expected to be smaller for higher (i.e., healthier) values of the competitive food healthy scale, or that schools with unhealthier competitive food offerings are expected to experience a faster increase in obesity rates over time. The interaction terms for the other individual food environment variables were not significant. The main coefficients of the food environment variables in this model indicate whether there was an association between said variables and obesity rate at baseline (i.e., when the time variable = 0). For instance, schools with ≥1 upgraded store within 400 m tended to have a lower obesity rate at baseline (β = –0.035; P < 0.001), whereas a higher number of limited-service restaurants was associated with higher obesity prevalence (β = 0.002; P = 0.001). In sensitivity analysis, models estimated using the 3 different outcomes and with CEP as a covariate yielded similar results.

TABLE 3.

Results from the linear regression model with interaction terms examining obesity trends over time; the main predictors are the elements of the food environment

	Coefficient	95% CI - low	95% CI - high	P value
Time	0.008	–0.005	0.021	0.250
NSLP - healthy scale	–0.049	–0.195	0.098	0.515
Interaction: year ∗ NSLP - healthy scale	0.005	–0.022	0.033	0.706
Competitive foods - healthy scale	0.045	–0.041	0.132	0.304
Interaction: year ∗ competitive foods - healthy scale	–0.013	–0.024	-0.001	0.038
Limited-service restaurants	0.002	0.001	0.004	0.001
Interaction: year ∗ limited-service restaurants	0.000	–0.001	0.000	0.461
Convenience stores	–0.005	–0.011	0.002	0.144
Interaction: year ∗ convenience stores	0.001	–0.001	0.003	0.171
Presence of small grocery store	–0.010	–0.026	0.007	0.252
Interaction: year ∗ presence of small grocery store	0.000	–0.005	0.005	0.928
Presence of an upgraded store	–0.035	–0.053	–0.017	<0.001
Interaction: year ∗ presence of upgraded store	0.004	–0.002	0.009	0.220
School level (reference: primary)
Secondary	0.021	–0.010	0.053	0.189
% of students eligible for FRPM	0.095	0.021	0.169	0.011
% of Hispanic students	0.154	0.135	0.174	<0.001
% of non-Hispanic Black students	0.089	0.073	0.106	<0.001

CI, confidence interval; FRPM, free and reduced-price meals; NSLP, national school lunch program.

CI, confidence interval; FRPM, free and reduced-price meals; NSLP, national school lunch program.

Table 4 reports the coefficients estimated from the regression model in which the food environment was characterized by the 4-category variable derived from LCA. Compared to the reference group (i.e., schools with healthy SFE and low-density CFE), schools exposed to the unhealthy SFE and high-density CFE with fewer upgraded stores had a lower obesity prevalence at baseline but a much steeper time trend (β = 0.018, P < 0.001). The other 2 categories did not differ from the reference group in the baseline value or in the slope. The coefficients of all other predictors were similar in magnitude and statistical significance to the coefficients from the model reported in Table 3. Figure 1 shows the predicted obesity prevalence for the 4 groups over the 4-time points, highlighting the upward trend experienced by schools with an unhealthy SFE and a high-density CFE with fewer upgraded stores. The trajectories for schools in the other 4 groups were approximately flat, with their slopes not significantly different from 0.

TABLE 4.

Results from the linear regression model with interaction terms examining obesity trends over time; the main predictor is the environment type, a 4-category variable from latent class analysis

	Coefficient	95% CI - low	95% CI - high	P value
Time	0.004	–0.003	0.010	0.297
Environment type (reference: healthy SFE; low-density CFE)
Unhealthy SFE; low-density CFE	–0.014	–0.067	0.039	0.611
Healthy SFE; high-density CFE	0.026	–0.006	0.059	0.115
Unhealthy SFE; high-density CFE1	–0.050	–0.096	–0.004	0.033
Interaction terms: time ∗ latent class
Time ∗ unhealthy SFE; low-density CFE	0.001	–0.005	0.007	0.674
Time ∗ healthy SFE; high-density CFE	–0.010	–0.022	0.001	0.080
Time ∗ unhealthy SFE; high-density CFE1	0.014	0.007	0.022	<0.001
School level (reference: primary)
Secondary	0.031	0.000	0.063	0.053
% of students eligible for FRPM	0.106	0.018	0.195	0.019
% of Hispanic students	0.137	0.107	0.168	<0.001
% of non-Hispanic Black students	0.070	0.044	0.096	<0.001

CFE, community food environment; CI, confidence interval; FRPM, Free and reduced-price meals; SFE, school food environment.

¹The full label is as follows: Unhealthy SFE; high-density CFE with fewer upgraded stores.

FIGURE 1.

Predicted probability of obesity for schools based on their food environment type.¹ Notes: The estimated slopes and P values for the 4 groups are the following: Group 1: β = 0.004, P = 0.297; group 2: β = 0.005, P = 0.254; group 3: β = –0.007, P = 0.436; group 4: β = 0.018, P < 0.001. The full label for category 4 is as follows: Unhealthy SFE; high-density CFE with fewer upgraded stores.

CFE, community food environment; SFE, school food environment.

¹The environment type categories were determined through latent class analysis. The predicted obesity prevalence was calculated from the regression model reported in TABLE 4 for the 4 environment types while fixing all other predictors at their mean.

Figure 2 shows the linear trends in obesity for the composite healthier food environment and the composite unhealthier food environment profiles, estimated from the interaction model reported in Table 3. If schools were to experience an overall composite unhealthier food environment profile, where all 6 elements of the food environment had unhealthy values (see Box 1 above and Appendix C), the obesity trend among enrolled students was expected to be positive and significant (β = 0.009, P = 0.020). In contrast, if schools were to experience an overall composite healthier food environment profile, where all 6 elements of the food environment had healthy values, the obesity trend among students was expected to not differ from 0 (β = 0.003, P = 0.386). Further, as can be seen in the diverging trends in Figure 2, the estimated obesity prevalence for the 2 environment profiles was not different for the first 2 time points, SY 2013–14 and 2015–16, but it was significantly different in the later years, SY 2017–18 and SY 2019–20.

FIGURE 2.

Predicted probability of obesity for schools with composite healthier food environment and composite unhealthier food environment profiles.¹

¹The characteristics of the composite healthier and unhealthier environment profiles are described in Appendix C.

∗Indicates that there is a significant difference (P < 0.05) between the estimated obesity rate associated with the healthier food environment and the estimated obesity rate associated with the unhealthier food environment. These comparisons were based on z-tests for differences in proportion, obtained from the lincom command analyzed after the margins command in Stata 17.

Discussion

We used longitudinal data collected from K–12 schools over a 7-y period to examine the influences of the food environment within schools and in the community around schools on student weight status. A significant upward trend in student obesity prevalence was observed across all schools in the sample over the study period. However, our analyses consistently showed that the estimated obesity prevalence trajectories differed significantly based on the SFE and the CFE, with unhealthier environments corresponding to an upward trend, whereas the trend was mostly flat for schools experiencing healthier environments.

The finding that obesity rates among schools in our sample tended to increase over time is consistent with trends observed in the US [[43], [44], [45]]. Previous research showed particularly marked increases in obesity among non-Hispanic Black and Hispanic youth[43,45,46] and among youth from lower-income households [2]. Because our sample consisted predominantly of Black and Hispanic students living in lower-income communities, the detected increase in obesity prevalence is not surprising.

Individual elements of the CFE around schools were associated with student weight outcomes. Specifically, schools’ proximity to a larger number of limited-service restaurants was associated with a higher prevalence of obesity overall, whereas the presence of upgraded convenience stores and small grocery stores was associated with a lower prevalence of obesity. These findings are consistent with previous literature showing that nearby access to small grocery stores that sell healthy items is associated with more favorable child weight outcomes [6,8], and limited-service restaurant density around schools and children’s homes is associated with a higher prevalence of obesity [6,47,48]. Findings from the interaction analyses reported in Table 3 demonstrated that compared to schools that offered a healthier competitive food environment, the availability of a greater proportion of unhealthy competitive foods (i.e., foods sold a la carte during lunch and in vending machines) was associated with a steeper increase in obesity prevalence among enrolled students over time. Schools are a significant source of students’ dietary intake, and many students purchase foods outside of reimbursable school meals [49]. In SY 2014–15, as part of the HHFKA, the Smart Snacks in School policy established minimum nutritional standards for all foods and beverages sold on school campuses outside of reimbursable school meals [50]. Although the Smart Snack policy improved the nutrition quality of competitive foods sold [15,34], these options continue to be less healthy than the foods served as part of reimbursable school meals [51]. Contrary to our findings, a recent systematic review examining the impacts of competitive food on the weight of adolescents did not detect any conclusive associations; however, the majority of the 26 studies eligible for inclusion relied upon cross-sectional (73%) or self-reported data (58%) [52].

The lack of association between healthy NSLP items and obesity observed in this study is likely attributable to the fact that the study took place after implemention of the the HHFKA school meal guidelines. Similar to the vast majority of schools nationwide that successfully implemented the HHFKA guidelines [53], schools in our sample provided healthy lunch meals [15,34]. The limited variation in the degree of healthfulness of NSLP foods may explain the lack of significant association between NSLP and obesity prevalence, even though the direction of association was in the expected direction.

Since students are exposed to a variety of elements in the environment simultaneously, we assessed the combined impacts of the SFE and the CFE by 1) empirically grouping the schools into 4 categories based on their exposure to both the SFE and the CFE and 2) by creating 2 conceptually-based overall composite healthier and unhealthier food environment profiles, where all elements of the environment were set to healthy or unhealthy ends of the food environment variable distributions. Both these analyses provided similar findings, suggesting that schools that are exposed to unhealthy food environments, both within and outside, were likely to experience the highest increase in obesity prevalence among enrolled students over time. These findings on the cumulative impact of specific constellations of food environments align with prior longitudinal research, including the Healthy Communities Study and the Childhood Obesity Declines Project [[18], [19], [20]], and provide more conclusive evidence. Unlike previous work, our analyses were based upon contemporaneously collected data on specific elements of the SFE and the CFE over a 7-y period. Because of the robust study design, we were able to examine the association of exposure to documented changes in the food environment with trends in obesity.

The finding that healthier composite food environments may curtail rising obesity trends informs the development of multi-faceted community interventions; indeed, they are increasingly recognized as critical to effective prevention [[54], [55], [56]]. Policies and interventions that improve SFE are warranted, as are policies and interventions that improve the CFE around schools. The USDA’s Food and Nutrition Service is proposing updates to school meal standards to reduce added sugars and sodium and increase whole grains [57]. Such changes are likely to improve student dietary intake and reduce risk of obesity. Potential strategies to improve the CFE around schools include zoning laws restricting unhealthy food outlets from being located near schools and interventions such as healthy corner store initiatives that increase the availability and promotion of healthier foods within food outlets frequented by students [39,[58], [59], [60], [61], [62], [63]]. Individual risk of obesity is subject to many factors at the individual (e.g., genetics), environmental, and policy levels [64]. Although the observed divergences in trends are modest, the favorable cumulative impact on obesity trends associated with improvements in aspects of the environment that are amenable to interventions promises to make a meaningful difference at the population level.

Limitations

Given the de-identified nature of our height and weight data, we were not able to track individual students over time. As a result, both the exposure variables and the outcome (i.e., school obesity prevalence) were assessed and tracked at the school level. Because most variation in weight status exists at the individual level, findings from our school-level analyses likely yield more conservative estimates than tracking students over time. In addition, schools included in the analytical sample serve a large proportion of low-income and racial/ethnic minority students and are located within densely populated urban settings. Although these findings provide persuasive evidence regarding environmental influences on students at disproportionate risk of experiencing obesity, they may not be generalizable to all students attending schools within the US. Lastly, although the analyses accounted for school level, it would be interesting to analyze whether the associations under examination were different for primary compared with secondary schools, as older students tend to be more independent and may interact differently with the CFE [39]. Our sample size, however, was too small to run the analysis separately by school level.

Conclusions

In conclusion, we found that obesity prevalence among children is significantly impacted by both the SFE and the CFE. Schools exposed to unhealthy food environments experience steeper obesity prevalence trajectories over time. Specifically, the availability of unhealthy competitive food options is detrimental to school-level obesity outcomes. Effective strategies are needed to ensure that nutrition guidance for competitive foods sold in schools is consistent with dietary guidelines and that such guidance is strictly followed. Similarly, restricting students’ access to food outlets around schools, particularly limited-service restaurants, would be beneficial. Outlets close to schools should be encouraged to stock and promote healthy options, as stores in our study that participated in the healthy convenience store initiatives (i.e., upgraded stores) did. Our findings provide strong evidence that food environments within and outside of school matter and can play an important role in curtailing rising childhood obesity trends.

Author contributions

The authors’ responsibilities were as follows– PO-V and MJY: conceptualized the research design, acquired funding, and supervised data collection and implementation of the study design; FA: conducted statistical analysis. All authors made significant contributions to operationalizing the study design and interpreting findings; PO-V, FA, and EMM: wrote the first draft of the manuscript; All authors reviewed and edited the first draft of the manuscript and gave final approval of the version to be published, agree to be accountable for all aspects of the work, and ensure that any potential questions related to the accuracy or integrity of any part of the work will be appropriately investigated and resolved; and all authors: read and approved the final manuscript.

Conflict of interest

The authors report no conflicts of interest.

Funding

This study was supported by grants from the Robert Wood Johnson Foundation, the Eunice Kennedy Shriver National Institute of Child Health and Human Development, NIH (1R01HD071583 and R01HD104708), and the National Heart, Lung, and Blood Institute, NIH (1R01HL137814). The funding sources had no role in the design or conduct of the study.

Data availability

The data described in the manuscript, the codebook, and the analytic code are not publicly available but will be provided upon request.

Appendix A. Supplementary data

The following is the Supplementary data to this article: