A systematic review of the influence of the retail food environment around schools on obesity-related outcomes


  • J. Williams,

    Corresponding author
    1. British Heart Foundation Health Promotion Research Group, Nuffield Department of Population Health, University of Oxford, Oxford, UK
    • Address for correspondence: Ms J Williams, British Heart Foundation Health Promotion Research Group, Nuffield Department of Population Health, University of Oxford, Old Road Campus, OX3 7LF Oxford, UK.

      E-mail: julianne.williams@dph.ox.ac.uk

    Search for more papers by this author
  • P. Scarborough,

    1. British Heart Foundation Health Promotion Research Group, Nuffield Department of Population Health, University of Oxford, Oxford, UK
    Search for more papers by this author
  • A. Matthews,

    1. British Heart Foundation Health Promotion Research Group, Nuffield Department of Population Health, University of Oxford, Oxford, UK
    Search for more papers by this author
  • G. Cowburn,

    1. British Heart Foundation Health Promotion Research Group, Nuffield Department of Population Health, University of Oxford, Oxford, UK
    Search for more papers by this author
  • C. Foster,

    1. British Heart Foundation Health Promotion Research Group, Nuffield Department of Population Health, University of Oxford, Oxford, UK
    Search for more papers by this author
  • N. Roberts,

    1. Bodleian Health Care Libraries, University of Oxford, Oxford, UK
    Search for more papers by this author
  • M. Rayner

    1. British Heart Foundation Health Promotion Research Group, Nuffield Department of Population Health, University of Oxford, Oxford, UK
    Search for more papers by this author


The high prevalence of childhood obesity has led to questions about the influence of ‘obesogenic’ environments on children's health. Public health interventions targeting the retail food environment around schools have been proposed, but it is unclear if they are evidence based. This systematic review investigates associations between food outlets near schools and children's food purchases, consumption and body weight. We conducted a keyword search in 10 databases. Inclusion criteria required papers to be peer reviewed, to measure retailing around schools and to measure obesity-related outcomes among schoolchildren. Thirty papers were included. This review found very little evidence for an effect of the retail food environment surrounding schools on food purchases and consumption, but some evidence of an effect on body weight. Given the general lack of evidence for association with the mediating variables of food purchases and consumption, and the observational nature of the included studies, it is possible that the effect on body weight is a result of residual confounding. Most of the included studies did not consider individual children's journeys through the food environment, suggesting that predominant exposure measures may not account for what individual children actually experience. These findings suggest that future interventions targeting the food environment around schools need careful evaluation.


adjusted odds ratio


body mass index


convenience store


fast food


fast food restaurant


food outlet


food retail index


healthy eating index


healthy food availability retail index


high in fat, sugar or salt


healthy fitness zone


incidence rate ratio


odds ratio




standard error






The prevalence of childhood obesity in the world has increased dramatically over the past three decades and is considered by the World Health Organization to be one of the most serious public health problems of the 21st century [1, 2]. Overweight or obese children are likely to remain overweight as adults and have an increased risk of developing chronic conditions such as cardiovascular disease or type 2 diabetes. Swinburn and Egger coined the term the ‘obesogenic environment’ in 1997, and since then a growing body of research has looked at ways that external factors (such as access to food outlets) may influence dietary behaviours [3].

Despite significant methodological and conceptual limitations in research about the environment and health [4-8], there has been interest in potential environmental interventions to support healthy dietary behaviours [9, 10]. This has led to regulation of the food environment within schools [11] – but these policies aimed at improving the food environment for children do not generally extend beyond school boundaries. Planning or licensing controls to restrict unhealthy food retailing operations around schools have been proposed (and in a few cases implemented) in the UK, United States and Australia [12-20], but it is unclear whether such interventions are effective. Some of this lack of clarity is due to a conflicted and equivocal evidence base.

Existing systematic reviews

Despite a growing body of primary research examining the retail food environment surrounding schools and its potential influence on children, we were unable to find any systematic reviews that focus specifically on food retailing around schools and its associated outcomes among schoolchildren. Existing reviews have considered the broader subject of possible environmental determinants of health [4, 5, 7, 8, 21-23], but they have not focused specifically on the retailing around schools. For the first time, our review tackles this knowledge gap by examining associations between these environmental exposures and obesity-related outcomes, as well as how they were defined and measured.

Aim of this review: focusing on school food environment studies

The primary aim of this systematic review was to examine the associations between the retail food environment around schools and dietary intake, weight status or food purchasing behaviour among school-age children. Our hypothesis was that the food environment around schools influences food purchasing behaviour of schoolchildren at three points in the day: (i) on the journey to school; (ii) at lunchtime during ‘breaks’ from school and (iii) on the journey from school. We also hypothesize that the influence on food purchasing behaviour results in changes in dietary intake and changes in weight status. Our secondary aim was to catalogue and critique the various methods employed within this body of literature.


We developed a full protocol that is available from the authors on request.

Search strategies

We conducted a search using a combination of free-text terms and subject headings to describe schools and schoolchildren, the retail food environment and our outcomes of interest: food purchasing, food consumption and body weight (please see Supporting Information Appendix S1 for the Medline strategy). The following publication databases were searched from database inception to 24 October 2013: MEDLINE (OvidSP, 1946-), EMBASE (OvidSP, 1974-), Global Health (OvidSP, 1973-), CINAHL (EBSCOHost, 1982-), Education Resources Information Centre (ERIC, Proquest, 1966-), Web of Science (Thomson Reuters, 1945-), the Cochrane Public Health Group Specialized Register, PsychINFO(OvidSP, 1967-), Dissertations & Theses (Proquest, 1637-), LILACS(Virtual Health Library) and Science Direct. Additionally, we hand-searched the reference lists of articles for additional relevant papers with an end search date of October 2013. We did not conduct a Cochrane review because of the small number of intervention studies at present and the observational nature of most of the studies we were considering.

Inclusion/exclusion criteria

Studies were required to include at least one measurement of the school food environment. We defined this as the retailing in the area surrounding schools that schoolchildren encounter either on the journey to or from school, or at a lunchtime break from school. We used this definition because we wanted to consider environmental exposures that children were likely to encounter during the school day. This definition included food stores (e.g. supermarkets, convenience stores, farmers' markets) and catering outlets (e.g. fast food, full-service restaurants) but excluded food provision within the school building (e.g. cafeterias, vending machines, school tuckshops). Additionally, we required studies to include outcome data for schoolchildren 5–18 years old. The outcome data needed to include at least one of the following: (i) food purchases; (ii) dietary intake and (iii) body weight.

Study selection

One researcher examined the titles, abstracts and full-text articles. After the first researcher scanned titles and identified exclusions, a second researcher checked a 10% sample of exclusions and identified three papers where there was some disagreement. The title scan was then conducted for a second time, and the second researcher checked a different sample of exclusions and there was complete agreement. The same two researchers reviewed and cross-checked abstracts and full papers.

Classifying and coding the studies

We initially planned to group the studies by exposure and outcome and then, if possible, to perform a meta-analysis of the results. However, because of differences among study research questions, exposure measurements, outcome measurements and methods, formal meta-analysis was not possible, so we followed a semi-quantitative procedure used by Sallis et al. [24] and Dunton et al. [25]. For each study, we identified how the food environment was defined and measured (e.g. type of food outlet, the size of the school neighbourhood) and whether or not it was associated with increased frequency of food purchases, increased consumption of specific foods or increased body weight. We identified whether or not the finding was statistically significant, which we defined as a result that confirmed the hypothesis and had an associated P value of less than or equal to 0.05.

The aim of this semi-quantitative method was to allow a rapid assessment of the strength of the evidence of an association between the exposure and the outcomes of interest by reducing a range of results from heterogeneous analytical designs to two binary questions: Did the study show a positive association between the school food environment and the outcome of interest? If so, was this finding statistically significant (P < 0.05)?

Quality assessment

We assessed study quality using standard criteria for reviewing primary research papers that are not randomized controlled trials and following the guidelines presented by Zaza et al. [26, 27]. Because of the heterogeneity of study designs and the lack of a robust framework for ranking studies, we adopted a descriptive approach. Quality was assessed according to study methods (e.g. use of random sampling, use of objective or validated outcome measures, controlling for potential confounders) and reporting (e.g. defining exposure and outcome measures, describing the sample) (see Supporting Information Appendix S2).

After the team established the quality assessment criteria, one researcher completed an initial evaluation of the studies. A second researcher independently completed quality assessments for a 10% sample of the papers and the scores were checked for inter-rater reliability. The quality checks were sent to the corresponding authors of the included studies for verification.


The search retrieved 5,789 articles (see Fig. 1). Results come from 30 papers and 29 studies, featuring results from more than 10,000 schools and 1.5 million students (see Table 1).

Figure 1.

Processing the articles for inclusion in this review.

Table 1. Description of included studies on associations between food outlets around schools and student food purchases, consumption and body weight
Author, yearCountryAge in years (grade)aNumber of students (schools)ExposureType of food outletOutcomeCovariates/stratification
  1. Buffer size in bold indicates the buffer distance that we used in our analysis.
  2. aWhen papers reported student grade level only, we inferred age in years from the grade described in parentheses.
  3. bThe walkshed is the territory within a school's catchment that encompasses only those students living within walking distance.
  4. cOutcome was percentage of students falling within a ‘healthy fitness zone’, which includes both fitness measures and BMI.
  5. BMI, body mass index; CS, convenience store; F, fruit; FF, fast food; FFR, fast food restaurant; FO, food outlet; FSM, free school meal; HEI, healthy eating index (a composite variable that reflects overall diet quality); M, measured; SES, socioeconomic status; SM, supermarket; SR, self-report; SSB, sugar-sweetened beverage; V, vegetable.
An 2012 [46]United States5–1713,462GIS: density within 0.1, 0.5, 1.0 and 1.5 mile circular buffer of schoolCS, FFR, grocery stores and SMs , small food storesDietSR: F, V, FF, juice, milk, soda, high-sugar foodsAge, gender, household size, education, parent weight, race/ethnicity, survey wave
Buck 2013 [67]Germany6–9610GIS: clustering around schools, food retail index (kernel density estimates of FOs per km2)Bakeries, FFR, kiosks, SMsBMIM, DietSR: Junk food (SSB, chocolate, crisps, etc.)Age, sex, household income, parent education, under and over-reporting
Chiang 2011 [38]Taiwan6–132,283GIS: density within 500-m circular buffer of schoolCS, FFR, fresh produce markets, street vendorsBMIMAge, ethnicity, father's education, household income, pocket money, birth weight, time spent watching TV on weekdays, diet quality, region
Currie 2010 [43]United States14–15 (9)8,373GIS: presence within 0.1, 0.25 and 0.5 mile straight line bufferFFRBody fatMCensus demographics of nearest block, ethnicity, free school meals, school characteristics, school test scores, student : teacher ratio
Davis 2009 [39]United States12–17 (7–12)529,367GIS: presence within 0.25, 0.25–0.5 and 0.5–0.75 mile straight line buffer. Density within 3 milesFFR, ‘other restaurants’BMISR and diet: F, V, juice, soda, fried potatoesAge, gender, grade, physical activity, FSM eligibility, race/ethnicity, school location type, school type
Forsyth 2012 [68]United States11–14 (6−9)2,724 (20)GIS: Density within 800-m street network bufferFFR: traditional, pizza, subs/sandwiches, other FFDiet: FFEthnicity/race, grade level, gender, SES
Gebremariam 2012 [30]Norway11–12 (6)1,425 (35)Survey of school staff: presence ‘within walking distance from school’‘Food outlets where food or drinks could be purchased’DietSR: F, V, snacks, SSB, fruit drinksCanteen/food booth at school, food outlets present, gender, parent education, school nutrition committee, school's perceived responsibility for student diet, two parents
Gilliland 2012 [34]Canada10–141,048 (28)GIS: presence within 500 and 800-m straight line buffer, street network buffer and school walkshedbCS, FFRBMISRAge, sex
Grier 2013 [40]United States12–171,000GIS: straight line distance to closest outletFFRBMISR and dietSR: sodaAge, grade, sex, physical activity, race/ethnicity, school time, per cent eligible for FSM, school urbanicity
Harris 2011 [69]United States14–17 (9–12)552 (11)GIS: density within 2 km (1.24 mile) straight line buffer of school, distance to closest storeBagel shops, bakeries, coffee shops, FFR (burger/fries or Mexican), fried chicken restaurant, ice cream shops, pizza parlours, sandwich/sub shops, sit-down restaurants, snack barsBMISRAge, birth weight, diet quality, ethnicity, father's education, household income, pocket money, region, time spent watching TV on weekdays
Harrison 2011 [33]England9–101,995GIS: density within 800-m pedestrian network buffer weighted sum of the distance to every facility within 6 km of home and school‘Healthy outlets’ (SMs and green grocers), ‘unhealthy outlets’ (CS and takeaway)Fat mass indexMAge, sex, parent education, mode of travel to school
He 2012 [45]Canada11–13 (7–8)810 (21)GIS: density within 1-km straight line buffer; shortest network distance to nearest outletCS, FFRFood purchaseSRMode of transportation, father's education, land use mix
He 2012 [35]Canada11–13 (7–8)810 (21)GIS: density within 1-km straight line buffer; shortest network distance to nearest outletCS, FFRDietSR: HEIGender, grade level, neighbourhood distress score, annual family income, ethnicity, family structure, parent education
Heroux 2012 [65]Canada, Scotland, United States13–1526,778 (687)GIS: density within 1-km straight line bufferCS, chain FFR restaurants and cafésBMISRFamily affluence, grade, sex
Howard 2011 [44]United States14–15 (9)(879)GIS: Presence within 800-m network bufferCS, FFRBMIMEthnicity, percentage of students receiving free meals, urbanicity
Langellier 2012 [70]United States10–15 (5–9)(1,694)GIS: presence within 800-m network bufferCorner stores, FFRBMIMEligibility for title 1 funding, race/ethnicity, school type, urbanicity
Laska 2010 [71]United States11–18334GIS: density within 800, 1,600 and 3,200 m network bufferBakeries/doughnut shops, FFR, gas stations, grocery stores, variety storesBMISRAge, parent education, school and area-level SES, sex
Leatherdale 2011 [72]Canada9–13 (5–8)2,429 (30)GIS: density within 1-km straight line bufferAny retail facilities, CS, FFR, grocery storesBMISREthnicity, gender, grade, physical activity
Li 2011 [36]China11–171,792 (30)Survey of school staff: ‘presence within 10-min walk of school’Western FFRBMIMAge, household wealth, parent BMI, parent education
Nixon 2011 [41]United States14–15 (9)(41)GIS: density within 400- and 800-m straight line buffer, closest facility, degree of clustering around schoolsFFRBMISRSchool lunch policy, percentage of students receiving free meals, race/ethnicity, percentage of students in talented education program, parent education level
Park 2013 [37]South Korea9–15 (4–9)1,342Survey: density within 500-m radius of schoolSM, traditional markets, F and V markets, street vendors, snack bars, CS, FFO, doughnuts, ice cream, bakery shops, full-service restaurantsBMIM, HEIAge, sex, screen time, family affluence, mother's employment, school nutrition environment (composite index), social safety net program participants
Richmond 2013 [73]United States11–14 (6–8)18,281 (47)GIS: density within a 1,500-m straight line bufferFFR, CSDietSR: SSBAge, sex, race/ethnicity, percentage of students receiving free school meals
Rossen 2013 [28]United States8–13319GIS: mean healthy food availability index (HFAI), density of outlets within 100 m of shortest street network path between home and schoolCS, SM/GS, CS, restaurants (full service or carry-out), gas stationsBMIM , waist circumference (baseline and 1 year)Age, gender, race/ethnicity, number of siblings, receipt of free or reduced price lunch, walking to school status, distance to school (log km), school violence strata, census-tract deprivation index
Sánchez 2012 [42]United States10–15 (5–9)926,018 (6,362)GIS: density within 800-m straight line buffer around schoolCS, FFRBMIMAge, sex, school-level characteristics and interactions with race/ethnicity
Schafft 2009 [74]United States10–13 (5, 7)243 school districtsGIS: absence of ‘large grocery store’ within 10 mile straight line buffer around ‘population based centroid’ of the school districtLarge grocery store: grocery or retail food store with more than 50 employeesBMIMedian family income, per cent mobile home residence, per cent incomplete kitchen
Seliske 2009 [75]Canada11–167,281 (178)GIS: density within 1 and 5 km straight line bufferCS, doughnut/coffee shops, FFR, full-service restaurants, sub/sandwich shopsBMISR cBMI, family affluence scale, physical activity, sex, urbanicity
Smith 2013 [29]England11–161,382 (29)GIS: density within 400 and 800-m road network buffer, median and minimum distance to grocer or TATA, grocer/SM/CSDiet quality: ‘Healthy’ or ‘Unhealthy’ (aggregate score)Age, gender, FSM eligibility, ethnicity, school-level deprivation
Svastisalee 2012 [47]Denmark11–15 (5–9)6,034 (80)GIS: ‘Exposure:’ number of FOs divided by total road segments within 300 m of schoolsFFR, SMsDietSR: F, VAge, family social class, sex
Timperio 2009 [32]Australia5–12816GIS: density within 50 m buffer along route to school, Presence of FO along routeCafes, FFR, restaurants, takeaway storesDietSR: FF or takeawayAge, SES
van der Horst 2008 [31]Netherlands12–151,293 (15)GIS: density within 500-m straight line bufferBakery, FFR, fruit/vegetable store, large SM, small food storeDietSR: SSB and snacksDate of measurement, ethnicity, education

General characteristics of included studies

The earliest publication was in 2008 and about three quarters of the papers (n = 23) were published between 2011 and 2013. The papers were largely cross-sectional, but there were two longitudinal exceptions from Rossen et al. [28] and Smith et al. [29]. Most of the studies took place in North America (United States: n = 14; Canada: n = 5) but there were also studies from Europe (n = 6), Asia (n = 3) and Australia (n = 1). One multi-country study from the United States, Scotland and Canada was also included. Participant age ranged from 5 to 17 years. Sample sizes ranged widely from 334 to 926,018 students and more than three quarters of the studies had more than 1,000 students. Most of the papers did not explicitly identify the theoretical model informing their work, but those that did [30, 31] cited social ecological models.

Methods for defining and measuring the school food environment

Studies varied in their methods of constructing exposure measures in terms of the level of the exposure (whether or not it accounted for individual variation) and the source of information (primary vs. secondary sources).

Level of exposure: area-level vs. individual-level exposures

Area-level exposures were based on a static area such as a buffer around the school or the school's census tract. Most of the included studies used area-level measures, defined at the level of the school, which meant that all students attending the same school had a shared exposure value (n = 21). The alternative approach of using an individual-level exposure, where quantification of food outlets accounted for individual factors such as a student's home address, was used by nine studies. Three papers [28, 32, 33] accounted for the student's journey through the food environment by taking the student's school and residential address and calculating the number of outlets falling along the route between the two locations. Gilliland et al. [34] used multilevel structural equation modelling techniques to simultaneously test the effects of the school-environment and home-environment predictors on body mass index (BMI) scores and He et al. [35] calculated individual participants' ‘junk food density’ based on the density of stores around both students' home and school address.

Geographical information systems vs. survey-based measures of exposure

The predominant method of characterizing the food environment exposure was by using geographical information systems (GIS) (n = 27). Most commonly, this was done through use of a software program to construct a buffer zone (straight line, street or pedestrian network) around the child's school or the route between home and school and then counting the number of food outlets within that area (density). For studies using this density method, the buffer distances ranged from 0.1 to 3.0 miles (about 160–4,800 m) for the area around schools and from 50 to 100 m for the area around routes. For the former category, the most frequently used buffer distance was half a mile (about 800 m). Most of the papers using buffer zones provided rationale for using the buffer distance that they did (n = 25) and for about one-third of them (n = 12), at least one of the cited reasons was to be consistent with earlier studies. Another GIS method calculated the distance from the school to the nearest outlet (proximity). For these GIS-based studies, information about the locations, names and types of food outlets came predominantly from large secondary data sources including private companies and local business directories (n = 18) or public records such as census data, tax registry documents or government food premise databases (n = 8). Harris et al. collected store data using a global positioning system (GPS) unit and adding these geo-referenced points to a digital map.

Subjective measures of the retail food environment included the use of questionnaires. Two studies identified food outlets via a questionnaire in which school administrators identified the presence of food outlets ‘within walking distance’ [30] or within ‘ten minutes' walk’ [36] of the school. Park et al. [37] used an audit tool to record observations of the various types of food outlets found within a 500-m radius of the school.

Defining types of food outlets

Food outlet definitions and categories varied between papers and, in the instances when they were explicitly defined, often depended on the definitions provided by the original data source. A range of food outlets were included, but most of the studies narrowed their measures to a few specific types. The most common types of outlets to be included were fast food restaurants (n = 23), convenience stores (n = 10), supermarkets (n = 6) and grocery stores (n = 7).

Types of outcomes: food purchasing behaviour, consumption and body weight

Of the three outcomes we considered in this review, the most common was body weight, with 20 papers evaluating environmental associations with BMI (n = 18) and fat mass (n = 2). The second most common outcome was food consumption, with 14 papers evaluating associations between the environment and diet. Food consumption was predominantly assessed as daily or habitual consumption (rather than food consumption at school). A range of specific foods were measured, but the most frequently evaluated were fruit and vegetables (n = 4), soda or sugar-sweetened beverages (n = 7), or fast food (n = 4). Three papers used a composite variable such as a Healthy Eating Index (HEI; n = 2) or a healthy diet score (n = 1). Of the three types of outcomes we considered, food purchases were measured least frequently, with only one paper including it as an outcome. This measure was based on participant's self-report of purchasing fast food at least once in the previous week.

Quality assessment

We assessed the quality of studies using 13 criteria that included whether or not studies randomly selected participants, provided clear definitions of the study area, validated their exposure and outcome measurements, or attempted to control for potential confounders. When it was applicable, most papers randomly selected schools (n = 18) and students (n = 19), and defined the area of measurement (i.e. the ‘school neighbourhood’) in terms of a defined spatial size (n = 27). Nine studies validated their exposure measures via ground-truthing and three via Google Maps. Nine of the 14 studies measuring diet used a validated instrument. Twelve of the 20 studies with BMI or weight as an outcome used objective measures and eight relied on self-report. Almost all of the studies adjusted for potential confounders in their final analysis with the most common adjustments for socioeconomic status (n = 26), race/ethnicity (n = 20) and urbanicity/population density (n = 8).

Results from the included studies

The results below are organized according to their outcome measures. Because of the heterogeneity in study design, we report the following results in terms of increased food purchases, increased consumption or increased body weight. We chose an arbitrary level of significance (P = 0.05).

Because of a diverse range of exposures, outcomes, levels of adjustments and the number of analyses reported by individual studies, and to avoid over-representing results from papers that reported many results, we used the following criteria to determine which results to feature in Tables 2-4. When papers presented results using multiple levels of adjustment, we took the most adjusted. When results were stratified using categorical variables (e.g. ethnicity), we included all results, but when they were stratified using ordered variables (e.g. grade or social class) we took the result from the highest and lowest levels only. When papers presented results using multiple exposure measures (varying buffer sizes and types, GIS methods, and means of quantifying food outlets), we included the network buffer size closest to 800 m and the ‘density’ variable that accounted for the most individual-level variation. When papers presented results of multiple outcomes related to weight (BMI, waist circumference, triceps skinfold thickness), we used the outcome closest to our primary outcome of interest (BMI). All of the results (both included and excluded) have been provided (Supporting Information Appendixes S3–S5).

Table 2. Summary of findings: food outlets around schools and student body weight
AuthorType of food outletOutcomeIncreases weight?P < 0.05
  1. aApproximate: rounded from ½ mile (804.7 m).
  2. bMeasure is the distance from food outlet and weight outcome or the absence of food outlet and weight outcome.
  3. AOR, adjusted odds ratio; APR, adjusted prevalence ratio; BMI, body mass index; CS, convenience store; FFR, fast food restaurant; FO, food outlet; FRI, food retail index (# of FOs per 1,000 residents); HFAI, healthy food availability index (based on the availability of foods from eight food groups within each outlet); HFZ, healthy fitness zone (accounts for school fitness levels and student BMI); IRR, incidence rate ratio; OR, odds ratio; OW, overweight; SE, standard error.
Buck 2013 [67]  βP value  
FRIBMI z score0.1100.17YesNo
Chiang 2011 [38]# within 500 mBMI z scoreβ   
CSBoys0.010 YesNo
FF 0.080 YesYes
CSGirls0.020 YesNo
FF 0.030 YesNo
Currie 2009 [76]FO within 800 ma% obeseβSE  
FFRNinth graders−0.03910.4475NoNo
Other 0.46380.4881YesNo
FFRFifth graders0.43410.1844YesYes
Other FO 0.28790.2312YesNo
Davis 2009 [39]FO within 800 maBMIb95% CI  
FF 0.100.03, 0.16YesYes
Other FO 0.080.01, 0.14YesYes
Gilliland 2012 [34]FO within school walkshedBMI z scoreEstimateSE  
FFR 0.0730.034YesYes
Presence of CS (school walkshed) 0.0200.021YesNo
Grier 2013 [77]DistanceΒ95% CI   
FFRBMI−0.050−.10, .00YesbYes
Harris 2011 [69]# within 2 kmBMIβP  
Restaurants 0.0100.31YesNo
Pre-packed food stores 3 × 10−40.96YesNo
Grocery stores 0.0460.53YesNo
Other stores 0.0200.78YesNo
Stores overall 0.0000.66YesNo
Harrison 2011 [33]School access (high vs. low)FMI for girlsB95% CI  
Healthy FOsCar, bus or train0.020−0.068, 0.110YesNo
Unhealthy FOs0.010−0.107, 0.130YesNo
Healthy FOsWalk or cycle−0.090−0.183, −0.006NoNo
Unhealthy FOs0.1400.009, 0.280YesYes
Route to school access (present vs. not)     
Healthy FOs presentCar, bus or train−0.021−0.104, 0.062NoNo
Unhealthy FOs present0.041−0.029, 0.110YesNo
Healthy FOs presentWalk or cycle−0.0320.143, 0.078NoNo
Unhealthy FOs present 0.007−0.068, −0.082YesNo
Heroux 2012 [65]# within 1 km (ref: 0)OW/obesityOR95% CI  
All FOs (5+)Canada0.970.80, 1.18NoNo
CS (5+) 1.000.79, 1.26NoNo
FFR (5+) 0.810.63, 1.06NoNo
Cafes (3+) 0.790.53, 1.21NoNo
All FOs (5+)Scotland0.890.61, 1.29NoNo
CS (5+) 1.050.61, 1.80YesNo
FFR (5+) 0.600.32, 1.15NoNo
Cafes (3+) 0.660.42, 1.03NoNo
All FOs (5+)United States1.010.84, 1.23YesNo
CS (5+) 1.110.87, 1.40YesNo
FFR (5+) 0.990.81, 1.22NoNo
Cafes (3+) 0.980.66, 1.41NoNo
Howard 2011 [44]FO within 800 m% OWβSE  
FFR −0.0100.58NoNo
CS 0.0500.59YesYes
SM −0.0100.65NoNo
Langellier 2012 [70]FO within 800 ma% OWβSE  
Corner store or liquor store 1.630.61YesYes
FFR 0.350.52YesNo
Laska 2010 [71]Presence within 800 m β95% CI  
Any restaurantBMI z score−0.28−0.50, −0.07NoYes
Leatherdale 2011 [78]# within 1 kmOW (vs. normal weight)AOR95% CI  
Gas stations 1.460.79, 2.68YesNo
FFO 0.960.82, 1.13NoNo
Bakeries/doughnut shops 0.890.68, 1.15NoNo
Variety stores 0.820.59, 1.13NoNo
Grocery stores 1.100.86, 1.42YesNo
Li 2011 [36]# within 10 min walk (ref: 0)BMIβ95% CI  
FFR (1) 0.60−0.02, 1.1YesNo
FFR (≥2) 0.800.1, 1.4YesYes
Nixon 2011 [41]FFR clustering% not within HFZaMoran's I indexaP value  
400 m 1.24P < 0.01YesYes
800 m 0.37P < 0.05YesYes
Park 2013 [37]FO density (low: ref)ObeseOR95% CI  
Markets (SM, traditional, FV) 1.04.99, 1.11YesNo
Street vendors, snack bars, CS 0.98.95,1.01NoNo
FFR, doughnuts, ice cream, bakery shops 1.021.00,1.04YesYes
Full-service restaurants 0.99.98, 1.01NoNo
Rossen 2013 [79]FO within 100-m path to school1 year changeb95% CI  
HFAIaBMI−0.15−0.26, −0.13NoYes
Sánchez 2012 [42]Presence within 800 m% OWAPR95% CI  
FFR (≥1 vs. 0) 1.021.01, 1.03YesYes
 White1.021.00, 1.04YesYes
 Hispanic1.021.01, 1.03YesYes
 Black1.031.00, 1.06YesYes
 Asian0.940.91, 0.97NoYes
CS (per additional FO) 1.011.00, 1.01YesYes
 Fifth grade1.011.00, 1.02YesYes
 Ninth grade1.000.99, 1.01NoNo
Schafft 2009 [74]Absence within 10 miles% OW/at riskbSE  
Large grocery or SM 0.0440.020NobYes
Seliske 2009 [75]Presence within 1 km (ref: 0 vs. high)OW vs. normalOR95% CI  
FFR 0.830.70, 0.98NoYes
Sub/sandwich shops 0.780.64, 0.93NoYes
Doughnut/coffee shops 0.810.68, 0.96NoYes
Total FRI 0.700.61, 0.81NoYes
Table 3. Summary of findings: food outlets around schools and student consumption or purchase of foods high in fat, sugar or salt (HFSS)
AuthorType of food outletOutcome  Increases consumptionP < 0.05
  1. aSSB, chocolate, nut-based spreads, crisps, chocolate bars, candies.
  2. bFruit juice, SSB, sugar-added cereals, chocolate, candy, etc.
  3. cDif P value: difference between those with 0 and those with 1 at P < 0.05; t-test.
  4. dTraditional FF: burgers and fries.
  5. eMediational effect of FO density on association of race/ethnicity and SSB consumption.
  6. fExposure is expressed as distance to food outlet.
  7. AOR, adjusted odds ratio; CS, convenience store; FF, fast food; FFR, fast food restaurant; FO, food outlet; FRI, food retail index (# of FOs per 1,000 residents); HFAI, healthy food availability index (based on the availability of foods from eight food groups within each outlet); OR, odds ratio; OW, overweight; SM, supermarket.
An 2012 [46]# within 500 m Child   
 High-sugar food0.9980.008NoNo
 Fast food0.9910.01NoNo
 High-sugar food0.9860.027NoNo
 Fast food0.9870.033NoNo
Small FOSoda1.0020.011YesNo
 High-sugar food0.9990.007NoNo
 Fast food1.0060.009YesNo
 High-sugar food1.0220.025YesNo
 Fast food1.0290.035YesNo
Large SMSoda0.9950.035NoNo
 High-sugar food0.9550.024NoNo
 Fast food1.0080.031YesNo
 High-sugar food1.0290.016YesNo
 Fast food0.9930.012NoNo
 High-sugar food1.0510.055YesNo
 Fast food1.0050.032YesNo
Small FOSoda1.0020.009YesNo
 High-sugar food1.0130.015YesNo
 Fast food1.010.009YesNo
 High-sugar food0.960.047NoNo
Fast food1.0420.043YesNo
Large SMSoda1.0380.039YesNo
High-sugar food1.0330.04YesNo
Fast food1.060.036YesNo
Buck 2013 [67]# per 1,000 people Exp βP value  
FRIJunk fooda1.040.57YesNo
 Simple sugar foodb0.990.87YesNo
Davis 2009 [39]Proximity# of servingsb95% CI  
FFRSoda0.02−0.01, 0.04YesNo
 Fried potatoes00.02, 0.02NoNo
Forsyth 2013 [80]# within 800 mAdjusted weekly frequency Dif P valuec  
FFR type Boys   
All types03.6   
 3+4.4 YesNo
 Trend P valuef0.644   
All types     
 3+3.2 NoNo
 Trend P valuef0.299   
Gebremariam 2012 [30]# within walking distance βSE  
Grier 2013 [81]Distance from school β95% CI  
Richmond 2013 [73]# within 1,500 mMediational effecteβSE  
FFR and CSSSB (servings per day)0.00010.001YesNo
Smith 2013 [29]Distance to school (min)Unhealthy dietβ95% CI  
Grocer (800 m) −0.001−0.003, 0.000YesfYes
Takeaway (800 m) −0.002−0.004, 0.000YesfYes
Timperio 2009 [32]Access along route to school AOR95% CI  
# of FF or TAConsumed ≥1/wk11.0, 1.0NoNo
van der Horst 2008 [31]# within 500 mLitres per dayβ   
SMSoft drinks0.077 YesNo
FFR −0.055 NoNo
Small food stores −0.259 NoYes
Food outlets and purchases of HFSS foods
He, 2012 [35]# within 1 kmPrevious weekOR95% CI  
FFRFF purchase1.41.1, 1.7YesYes
Table 4. Summary of findings: food outlets around schools and student consumption of fruit and vegetables or healthy eating indexes
AuthorType of food outletOutcome  Increases consumption?P < 0.05
  1. aApproximate: rounded from ½ mile (804.7 m).
  2. bDifference in HEI score compares the difference in scores between schools where nearest outlet was <1 km away and schools where nearest outlet was ≥1 km away.
  3. cExposure is expressed as distance to food outlet.
  4. dOutcome is infrequent consumption.
  5. AOR, adjusted odds ratio; CS, convenience store; FF, fast food; FFR, fast food restaurant; FO, food outlet; FRI, food retail index (# of FO's per 1000 resident); HE, healthy eating index, a composite variable based on habitual meal habits (e.g. skipping breakfast) or consumption (fruit, vegetables, milk, soda, FF, Ramen noodles, chips, fried food, etc.); HFAI, healthy food availability index (based on the availability of foods from eight food groups within each outlet); IRR, incidence rate ratio; OR, odds ratio; OW, overweight; SE, standard error; SM, supermarket; TA, takeaway.
An 2012 [46]FO within 800 ma Child   
Small FOFruits1.0020.005YesNo
Large SMFruits1.0090.016YesNo
Small FOFruits0.9960.007NoNo
Large SMFruits1.0200.021YesNo
Davis 2009 [39]P# of servingsb95% CI  
FFRFruit−0.02−0.04, 0.00NoYes
 Vegetables−0.02−0.03, 0.00NoYes
Gebremariam 2012 [30]FO within walking distance βSE  
Svastisalee 2012 [47]  Low family social class   
SMs (low vs. high)Infrequent consumptionAOR95% CI  
 Fruit1.170.89, 1.54YesdNo
 Vegetables1.330.92, 1.90YesdNo
FFR (high vs. low)Fruit1.320.98, 1.76NodNo
 Vegetables1.170.80, 1.71NodNo
  High family social class   
SMs (low vs. high)Infrequent consumptionAOR95% CI  
 Fruit1.080.80, 1.45YesdNo
 Vegetables1.040.80, 1.35YesdNo
FFR (high vs. low)Fruit1.230.89, 1.69NodNo
 Vegetables1.260.95, 1.66NodNo
Food outlets and composite variables
He 2012 [45]# within 1 kmHEIc scoreDaffSE  
FFR (0) (ref: ≥3) 2.751.06YesYes
FFR (1–2) (ref: ≥3) 0.661.14YesNo
Park 2013 [37]# within 500 mHEIcβSE  
Markets (SM, traditional, FV) −0.020.06NoNo
Street vendors, snack bars, CS 0.040.08YesNo
FFR, donuts, ice cream, bakery −0.130.07NoNo
Full-service restaurants 0.030.07YesNo
Smith 2013 [29]Minimum distanceHealthy dietβ95% CI  
Grocer (800 m) 0.0020.000, 0.003NocYes

Food outlets and body weight

Twenty papers looked at the relationship between food outlets and body weight. Of the 72 associations (reported in Table 2), 43 showed a positive relationship between body weight and exposure to food outlets. Nineteen of these positive relationships were significant, with most in the expected direction after adjustments. These included positive associations between exposure to fast food outlets and BMI [34, 36, 38-40], obesity [37] and the proportion of overweight [41, 42] or obese [43] students. Positive associations were also observed between the presence of ‘unhealthy outlets’ (convenience stores and takeaways) and adiposity among girls who walk or cycle to school [33] or convenience stores and proportion of overweight students [42, 44].

Food outlets and food purchases

Although three studies reported measuring food purchases, only one paper provided results. He et al. found that high fast food outlet density was positively correlated with student report of fast food purchases in the past week and this was significant (P < 0.05) [45] (see Table 3).

Food outlets and consumption of foods high in fat, sugar or salt

Ten papers measured associations between food outlets and consumption of foods high in fat, sugar or salt, the most common of which were sugar-sweetened beverages (n = 6) and ‘fast-food’ (including fried potatoes) (n = 4) or an aggregate variable that took these foods into account) (see Table 3). In total, 54 associations between these foods and retail outlets were reported and in about half (n = 28), food outlets were associated with increased consumption. However, only two of these results were significant (P < 0.05); Smith et al. found that unhealthy diet scores (reflecting frequency of consuming crisps, sweets, biscuits, fried food, fizzy drinks) were negatively correlated with the minimum distance to grocery stores and takeaways within 800 m [29].

Food outlets and consumption of fruits, vegetables or overall diet quality

Four papers considered associations between food outlets and fruit and vegetable consumption (see Table 4) [30, 39, 46, 47]. A total of 32 associations were reported and in about half (n = 18), exposure to food outlets was associated with increased consumption of fruit and vegetables. Three of these associations were significant (P < 0.05) and they all related to fast food outlets. An [46] observed positive association between the presence of fast food outlets and vegetable consumption among adolescents and Davis [39] observed a negative association between proximity to fast food and fruit or vegetable consumption.

Food outlets and healthy eating indexes

Three papers included composite variables that reflected overall diet quality [29, 37, 45] (see Table 4). Of seven associations, four were positively correlated with increased healthy eating scores. Among these, there were two significant (P < 0.05) findings. He et al. [35] looked at associations between food outlets around schools and the HEI score, which reflects overall diet quality, and found that students attending schools with a convenience store or fast food outlet farther than 1 km away had a significantly higher HEI score than students with an outlet within 1 km [35]. Smith et al. found a positive correlation between distance to grocers and healthy diet scores.


Principal findings

This review examined associations between the food environment around schools and children's food purchases, consumption or body weight. The methods for defining and measuring the food environment varied widely between studies and few consistent findings emerged. We found little reported evidence for an effect of the school food environment on food consumption patterns and limited evidence of an effect on food purchases, but some evidence of an effect on body weight. However, these results should be interpreted cautiously. These studies were observational and therefore susceptible to confounding. With only two exceptions (from the longitudinal studies of Smith et al. and Rossen etal.), the evidence base is composed almost entirely of cross-sectional data. Measurement bias is likely, particularly with the diet-related outcomes, where misreports have been shown to vary children's characteristics (age, sex, weight) and social factors [48]. Reporting bias is possible, which is suggested by the fact that several papers reported significant results only.

Strengths and weaknesses

We were unable to assess pooled effects as there were many definitions and measures of the food environment surrounding schools [6]. One strength of this review was that it provided some focus by honing in on one specific element of the food environment – the presence of retail food outlets in the area surrounding schools. However, this strength was also a weakness; this definition does not account for all of the other relevant obesogenic environments that a child will encounter over the course of a day [49, 50] and it prevented us from considering research about the other elements of food access, such as availability, accessibility, affordability and accommodation [5, 7, 51]. The recent review by Caspi et al. provides a helpful overview of these other influences [5]. Additionally, the focused nature of this review kept us from considering the environment within retail outlets (e.g. product availability or placement within stores), but another recent review by Ni Mhurchu et al. suggests that this aspect of the food environment is not consistently associated with dietary outcomes [52]. As here, methodological heterogeneity makes it difficult to draw firm conclusions.

As noted earlier, given the heterogeneity of the studies and the wide range in the number of exposures, outcomes and analyses that individual papers reported, we did not include every single result that every paper provided in our overall assessment. We used a consistent and transparent approach to select results from studies so as to avoid conclusions being overweighted by studies that reported multiple findings from the same dataset. For example, for the BMI outcome, we reported 72 associations, with 43 showing a positive correlation with food outlets (28 of those being significant). Comparing these figures to all results reported (and featured in the Supporting Information Appendix S3), there were 142 associations, with 89 showing a positive correlation of weight with food outlets and 53 being significant. We have highlighted the instances when there were significant associations that varied from what we reported (either in terms of direction or significance) in the Supporting Information Appendix. Davis et al. [39] presented associations on the school food environment and body weight within three buffer sizes: 0–0.25 miles, 0.25–0.5 miles and 0.5–0.75 miles. We showed the results from 0.5 miles, which again were in the same direction of association as the other two buffer sizes, although the association between fast food outlets and BMI was not significant at the 0.5–0.75 mile area of exposure while it was significant at the two smaller sizes. Therefore, choosing to present the results as we did may have altered our assessment of the number of associations that are significant compared to if we had chosen to use the larger buffer. Finally, Currie [43] presented associations with exposure at 0.1, 0.25 and 0.5 mile buffers and we presented the latter. For exposure to ‘other restaurants’, the results are in the same direction and at the same significance level, but for fast food exposure, the associations were not significant at the smaller buffer sizes (as they were at the larger size for fifth graders). For dietary outcomes, please see the Supporting Information Appendix for a full list of results and how our inclusion decision may have altered the assessment. For example, Svastisalee et al. reported additional analyses assessing interactions between fast food and supermarkets and associations with fruit and vegetable consumption according to social class and found that children from low and middle social class backgrounds attending schools with high fast food and low supermarket exposure were most likely to report infrequent fruit intake. A final limitation is that despite a comprehensive search in 10 databases and hand-searching references, we failed to identify one paper that did not have MeSH headings attached. Fortunately, this paper was identified by a reviewer and it is represented here.

Implications for policy

Overall, this review did not find strong evidence at this time to support policies aimed at regulating food environments around schools. However, given that food retailing is already influenced by a number of other policy drivers (related to economics, antisocial behaviour, litter and pollution, food hygiene, etc.), it is important that broader public health evidence is also considered. However, it is not possible to draw conclusions until a higher quality evidence base is developed.

Implications for research

To improve the quality of the evidence base, future longitudinal data are required to account for changes that may occur in the food environment over time. As earlier reviews found [7], the research has relied on cross-sectional data with the most common approach to characterizing the retail food environment in this body of literature being to calculate the density or proximity of outlets within a buffer using indirect sources of food outlet data (such as directories or large databases). These methods bring up several questions about data accuracy and comprehensiveness, especially given that food outlet data are imperfect [53], which may have implications for exposure assessment accuracy. Questions also remain about which types of outlets to focus on. Earlier reviews noted a focus on fast food outlets and recommended that future studies include other types of outlets in their exposure measures [7], but we found that a much wider range of food outlet types were included, such as fast food, convenience stores, grocery stores and supermarkets. While this may provide a more comprehensive picture of the retail food environment, it brings up questions about the best way to classify a food outlet and how to compare results from studies using different classification systems. To enable between-study comparisons, future work should integrate validated classification systems into the design [54]. Future studies should also explore the capacity of alternative methods for validating exposure data, including Google Street View [55, 56].

Additionally, future work should also incorporate a child's usual mode of travel to and from school into decisions about appropriate buffer distances. We found only three of the studies in this review accounted for mode of travel in their final analyses. If buffers are to reflect the real ability of children to walk or cycle to school (and hence their real exposure to environments), it is important that studies account for transport exposure and adjust for active vs. motorized transport as Harrison's [33] study did. Capturing this individual-level data may become easier as advances in measurement technologies foster a new era of ‘people-based’ rather than ‘place-based’ exposure measures [57-59]. Promising examples include the use of GPS devices or interactive mapping tools to capture individual mobility patterns, characterize the individual's activity space and then quantify outlets within that space [60-62]. The specificity that individual-level measures of exposure to the food environment would allow is vital if we are to accurately measure what is likely to be a small-effect size.

In addition to improving these GIS-based measures of the food environment (e.g. density of food outlets), future work may benefit from collecting complementary measures of both qualitative (participant perception-based) and quantitative measures of food access [63].

Future research needs to collect outcome measures that are appropriate relative to the exposures. For example, all of the papers assessed daily or habitual diet patterns, but these outcomes cannot be linked to the school food environment without knowing the time or place of consumption, and where the food was originally sourced. Future studies concerned with specific environments should collect this additional contextual information.

The age range of included studies encompassed both primary and secondary school settings and there are potentially important theoretical differences regarding how age may influence a child's interaction with the food environment as he grows older and develops more autonomy. This may lead to differences in travel time, distance travelled, availability of pocket change and other factors.

Another issue related to between-country generalizability. As Feng noted in his review, most of the associations came from North America, but food environments vary between countries [6, 64]. It was promising to see that one included study by Heroux et al. [65] looked at between-country food environments and outcomes. Future work is needed to develop standardized tools to monitor local food environments across countries [66].


In conclusion, we did not find strong evidence at this time to justify policies related to regulating the food environments around schools. Our findings may provide some timely insight to debate about prevention of obesity among children. Future work with longitudinal cohorts and more refined exposure and outcome measures may lead to higher quality evidence that may inform more effective public health interventions. Additionally, these improvements will allow researchers to better understand how this particular component of the food environment in the school neighbourhood interacts with other components of a child's environment and investigate the effects this may have on obesity risk.

Conflict of interest statement

None declared.

Author contributions

PS was the PI, supervised the data collection and contributed to finalization of the manuscript. NR conducted the keyword search. JW and AM completed the data extraction. JW drafted the manuscript. AM, CF, GC, NR, PS and MR assisted with writing the manuscript.


Funding for the study was provided by the National Health Service (NHS) Berkshire (JW) and the British Heart Foundation (PS, AM, GC, CF, MR, JW).