Rates of obesity and associated co-morbidities have been escalating steadily over the past several decades(Reference Hedley, Ogden, Johnson, Carroll, Curtin and Flegal1, 2). Many researchers contend that environmental changes drive the obesity epidemic(Reference Flegal, Carroll, Kuczmarski and Johnson3, Reference Hill, Wyatt, Reed and Peters4). The environmental experience of Americans includes few opportunities for physical activity and an overabundance of high-energy foods. Widespread and profound societal changes during the last few decades have increased the availability of cheap, convenient, energy-dense food in larger portions(Reference Schmitz and Jeffery5, Reference Johnson-Taylor and Everhart6). These environmental changes have coincided with the grim trajectory of obesity in the USA(Reference Binkley, Eales and Jekanowski7).
Interventions targeting individual activity and dietary behaviours have had limited effectiveness(Reference Summerbell, Waters, Edmunds, Kelly, Brown and Campbell8). While many health organizations call for environmental interventions(Reference Kumanyika9, 10), the evidence base linking obesity and the food environment is in an early stage. A review of studies examining this association needs to be conducted to better understand the state of this research.
The food environment involves the sources of energy and nutrients and the circumstances surrounding their procurement and consumption. Researchers have used the ecological framework to describe the multiple levels of the food environment(Reference Glanz, Sallis, Saelens and Frank11, Reference Booth, Sallis and Ritenbaugh12). The model of community nutrition environments developed by Glanz et al.(Reference Glanz, Sallis, Saelens and Frank11) identifies four sub-environments: community, organization, consumer, and information nutrition environments. While each level impacts individual behaviour, the community and consumer environments have been recognized as high research priorities owing to their potentially far-reaching effects(Reference Glanz, Sallis, Saelens and Frank11).
The present review analyses studies that have measured the relationship between obesity and the community and/or consumer food environments, as defined by Glanz et al.(Reference Glanz, Sallis, Saelens and Frank11). The community environment involves the ‘number, type, location, and accessibility of food outlets’ in a location(Reference Glanz, Sallis, Saelens and Frank11). The community food environment, also termed the neighbourhood food environment, is frequently measured using proximity or density measures of food outlets. Proximity is the distance between a food outlet and another location such as an individual’s residence. Density is the number of food outlets in a defined area surrounding a location. Food outlets predominantly include stores and restaurants. Food stores can be classified by size and food selection in descending order as follows: supermarkets, grocery stores and convenience stores. Restaurants include full service and limited service establishments(13). The consumer environment involves the conditions that consumers encounter inside the individual food outlets, including ‘price, promotion, and placement’ of different foods(Reference Glanz, Sallis, Saelens and Frank11).
Materials and methods
The primary objective of the current review was to identify and evaluate studies that examined the relationship between BMI and environmental variables related to food outlets, including their geographic arrangement and consumer conditions. The inclusion criteria included measurement of the following main outcome variables:
1. BMI as a continuous or categorical variable computed using the formula and weight status ranges outlined by the Centers for Disease Control and Prevention (CDC)(14).
2. Physical measurement of environmental variables related to food outlets.
BMI was selected as the outcome for obesity because it is a widely accepted and validated unit of measurement correlated with body fat(14–Reference Garrow and Webster16). The standard weight status categories for adults included normal (BMI = 18·5–24·9 kg/m2), overweight (BMI = 25·0–29·9 kg/m2) and obese (BMI ≥ 30·0 kg/m2)(17). For children and adolescents aged 2 to 20 years, the weight status categories were based on the CDC BMI-for-age and sex percentile ranges, including ‘at risk for overweight’ (85th–95th percentile) and ‘overweight’ (>95th percentile)(18). The review only included studies that utilized measures of the physical environment rather than individual perceptions of the food environment. While perceptions are important in influencing behaviour, environmental perceptions may not accurately reflect the actual conditions(Reference Kirtland, Porter, Addy, Neet, Williams, Sharpe, Neff, Kimsey and Ainsworth19). Due to the limited research on the community and consumer food environment, the review included studies that involved subjects of all age ranges and geographic locations. Studies were excluded from the review if they were not in the English language or did not involve human subjects.
In order to find the most comprehensive research, academic database searches, Internet searches and reference list reviews were performed. All searches were conducted in 2006. The following keywords and phrases were searched in multiple combinations: food environment, nutrition environment, environment, food availability, food outlets, restaurants, obesity, and overweight. The health science databases searched included MEDLINE, CINAHL, PubMed, Cochrane Central Register of Controlled Trials and CRISP. Database searches were limited to human subjects, English language and research when permitted by the search engine. The database searches resulted in over 4975 articles for initial screening. After reviewing titles and abstracts, only four of these articles met the sampling criteria and were eligible for review. General Internet search engines and health-related websites, such as the National Guideline Clearing House, WHO and the CDC websites, were also searched. The websites supplied additional commentary and recommendations for the food environment, but no new research articles were found.
Other methods of locating pertinent research included professional sources and searches of key article reference lists. Over the past 3 years, the author of the review has researched the food environment and discovered articles through professional conferences, presentations and recommendations. Recently published articles were brought to the author’s attention in this manner. Reference lists of articles previously identified for the review and other key commentary and review articles were also searched. Three articles that had not surfaced in the database and Internet searches were found using these methods.
Many potential studies were identified that did not satisfy all of the review criteria. Some studies measured the community or consumer environment, but did not include BMI as an outcome variable(Reference Reidpath, Burns, Garrard, Mahoney and Townsend20–Reference Laraia, Siega-Riz, Kaufman and Jones32). Other studies were excluded because they only measured individual perceptions of the food environment and not the physical environment(Reference Boehmer, Lovegreen, Haire-Joshu and Brownson33, 34). Studies that measured government influences(Reference Rabin, Boehmer and Brownson35) or other organizational settings, such as school, worksite or home food environments(Reference French and Stables36, Reference Campbell, Crawford and Ball37), surfaced in the search, but were excluded because they did not measure the community or consumer food environments as defined by Glanz et al.(Reference Glanz, Sallis, Saelens and Frank11).
The literature search identified seven studies that measured both BMI and a variable in the community or consumer food environment(Reference Burdette and Whitaker38–Reference Sturm and Datar44). The food environments were measured in several different ways, such as number of outlets per capita, proximity to or density of fast-food restaurants(Reference Burdette and Whitaker38, Reference Jeffery, Baxter, McGuire and Linde40, Reference Maddock41, Reference Simmons, McKenzie, Eaton, Cox, Khan, Shaw and Zimmet43), food stores(Reference Inagami, Cohen, Finch and Asch39, Reference Morland, Diez Roux and Wing42) or both restaurants and stores(Reference Sturm and Datar44). One study incorporated elements of the consumer food environment by analysing food price data(Reference Sturm and Datar44). All of these studies measured demographic variables, such as age, sex, race, ethnicity and varying socio-economic variables. The studies found mixed results and were subject to various limitations. A summary table provides a brief overview of all seven studies as they relate to the food environment.
Results
In 2004, Burdette and Whitaker(Reference Burdette and Whitaker38) studied the association of overweight in pre-school children and the proximity of fast-food restaurants to their residences. The cross-sectional study collected information from 7020 children, aged 3 to 5 years, from low-income families in Ohio. Weight and height was measured at a WIC (Special Supplemental Nutrition Program for Women, Infants, and Children) visit to determine BMI percentiles. The food environment was determined by examining the distance of the nearest fast-food restaurant to each child’s residence and presence of fast-food restaurants in their neighbourhood. The fast-food restaurant locations were obtained from US Yellow Page listings. Neighbourhoods were defined by political jurisdiction boundaries. Spatial relationships were calculated using Geographical Information Systems (GIS). There was no difference in mean distance to the nearest fast-food restaurant when comparing overweight with non-overweight children (0·70 (sd 0·40) v. 0·69 (sd 0·38) miles, P = 0·91, t test) and when comparing children at risk for overweight and not at risk (0·69 (sd 0·39) v. 0·70 (sd 0·38) miles, P = 0·43). Comparing children who were overweight and non-overweight, there was no significant difference in the percentage living in neighbourhoods without fast-food restaurants (44·0 % v. 44·5 %, P = 0·84)(Reference Burdette and Whitaker38).
In a similar study with adults, Jeffery et al.(Reference Jeffery, Baxter, McGuire and Linde40) investigated the relationship between BMI and living or working near various types of restaurants, which served as a proxy measure for ease of access. The outcome variable of BMI was determined by self-reported height and weight during a random digit-dial telephone survey of 1033 Minnesota residents. The restaurant locations were obtained through a public domain database. Food outlet density was mapped and calculated as the number of fast-food restaurants within a 0·5, 1·0 or 2·0 mile radius of home and work addresses by GIS. There was a significant inverse relationship between BMI and number of fast-food outlets (P = 0·008, linear regression models), other restaurants (P = 0·01) and total restaurants (P = 0·01) within 2 miles of work for men only. The study did not find a significant relationship between BMI and restaurant density around the home address for either women or men.
Simmons et al.(Reference Simmons, McKenzie, Eaton, Cox, Khan, Shaw and Zimmet43) examined the relationship between obesity and availability of takeaway outlets (take-out and fast-food outlets) and other restaurants. This cross-sectional study involved 1454 participants in Australia. The main outcome variables included BMI, calculated from measured height and weight, and number of local takeaway and sit-in restaurant outlets per 1000 residents. Unlike the other studies, some food outlets were identified via direct observation of the community, rather than relying solely on indirect databases. The food environment was assessed for three areal units: regional centre, large rural towns and small rural towns. The authors did not find an association between obesity and the density of takeaway outlets and restaurants. However, limited information was provided about the process of data collection and the results from the statistical analysis for the availability of food outlets.
In another cross-sectional study, Morland et al.(Reference Morland, Diez Roux and Wing42) studied the relationship between obesity, overweight and the presence of supermarkets, grocery stores and convenience stores in an individual’s residential census tract. The study included 10 763 individuals from four sites in the USA that participated in the Atherosclerosis Risk in Communities study. Obesity status was determined by BMI and calculated from weight and height measurements collected in 1993. The food environment outcome was the number and type of food stores classified using the 1997 North American Industry Classification System (NAICS) codes and geo-coded to each census track. The food store addresses were collected in 1999 from local and state government agencies. The authors used binomial regression analysis to estimate prevalence ratios (PR) of overweight and obesity associated with the presence of food outlets and to control for types of food stores, demographic variables and physical activity. The presence of supermarkets was inversely related to the prevalence of overweight (PR = 0·94; 95 % CI 0·90, 0·98) and obesity (PR = 0·83; 95 % CI 0·75, 0·92). The presence of convenience stores was associated with a higher prevalence of overweight (PR = 1·06; 95 % CI 1·02, 1·10) and obesity (PR = 1·16; 95 % CI 1·05, 1·27) compared with areas without convenience stores. However, the presence of grocery stores was not significantly associated with overweight status (PR = 1·03; 95 % CI 1·00, 1·07) or obesity (PR = 1·06; 95 % CI 0·99, 1·16).
Inagami et al.(Reference Inagami, Cohen, Finch and Asch39) explored the relationship between individuals’ BMI and proximity to and deprivation of their selected food store census tract. The study used a cross-sectional design and involved 2144 individuals in the Los Angeles Family and Neighborhood Study with a predominantly Latino population. Height and weight were self-reported and BMI was calculated as a continuous outcome. The community food environment was measured as the centroid-to-centroid distance from residential and selected grocery store census tracts (distance between the geometric centres). The difference between ‘disadvantage scores’ of the residential and the selected grocery store census tracts obtained from 2000 US Census information served as a proxy for grocery store quality. Through multilevel linear regression models, the authors found that BMI was greater for individuals who selected grocery stores in more disadvantaged neighbourhoods than their residential neighbourhood (data were not provided in the article). Exposure to neighbourhoods of worship, entertainment, medical care and work were not associated with BMI. In addition, the grocery store neighbourhood scores averaged for each census tract improved the model more than the individual scores. These results indicate that individual and group-level grocery store selection may be related to BMI. A distance of greater than or equal to 1·76 miles from home to grocery store was an independent predictor of a BMI increase of 0·775 units or ‘4·65 pounds for a 5′5″ person’ (P < 0·05)(Reference Inagami, Cohen, Finch and Asch39).
Diverging from the other studies, Maddock(Reference Maddock41) examined the relationship between obesity and the prevalence of fast-food restaurants with a cross-sectional analysis at the state level. Mean BMI for adults in fifty states excluding Alaska and including the District of Columbia were calculated from self-reported weight and height in the 2000 Behavioral Risk Factor and Surveillance System data. The food environment variables were square miles per fast-food restaurant and number of residents per fast-food restaurant by state. They were identified using data from the US Yellow Pages and the 2000 US Census. The author found a significant negative correlation between number of residents per fast-food restaurant and obesity prevalence (r = −0·53, P < 0·001): as the number of residents per fast-food restaurant decreased the percentage of obesity increased. The bivariate correlation between square miles per restaurant and obesity was not significant (r = −0·20, P < 0·16)(Reference Maddock41). Using multiple hierarchal regression analysis, the model that included population density, demographic variables, physical inactivity and fruit and vegetable intake explained 64 % of the state obesity rates (P < 0.001, se 0·001). Adding the food environment variables to the model explained an additional 6 % of the variance, which was a significant improvement from the model without the environmental variables (F change(2,38) = 3·96, P < 0·01). As the residents per fast-food restaurant and the square miles per fast-food restaurants decreased, state-level obesity prevalence increased.
Sturm and Datar(Reference Sturm and Datar44) performed another national analysis of the community food environment, examining the relationship between food outlet density, food prices and change in BMI over four years among 13 282 elementary-school children in the Early Childhood Longitudinal Study. BMI change over one and three years was measured at the individual level. The US Census Bureau’s 1999 Zip Code Business Patterns provided the per capita number and types of food outlets in each child’s residential and school zip code areas. Similar to Morland et al.(Reference Morland, Diez Roux and Wing42), Sturm and Datar used the 1997 NAICS to code the types of food outlets. In addition, the American Chamber of Commerce Research Association provided data on food prices for each Metropolitan Statistical Area. The results revealed that a higher per capita number of fast-food restaurants was associated with faster BMI gain, but was not statistically significant (P < 0·10). Lower fruit and vegetable prices predicted a significantly lower gain in BMI over one (P = 0·016, se 0·022, multivariate analysis) and three years (P < 0·001, se 0·033). However, there were no significant differences for dairy prices (P > 0·10) or fast food prices (P > 0·10). Lower meat prices were related to an increase in BMI, but were not statistically significant over one (P = 0·095, se 0·022) or three years (P = 0·414, se 0·031). Children below the poverty line (P = 0·05, se 0·063), children at risk for overweight (P = 0·01, se 0·070) and Hispanics (P = 0·01, se 0·063) had a stronger effect of fruit and vegetable prices than the general population; however, subgroup analyses were not sufficiently powered to show statistical significance.
Overall limitations
The literature examining obesity and the community and/or consumer food environment shares many limitations. Six of the seven studies were cross-sectional, examining all variables at one point in time(Reference Burdette and Whitaker38–Reference Simmons, McKenzie, Eaton, Cox, Khan, Shaw and Zimmet43). Cross-sectional designs have a limited ability to assess a temporal relationship(Reference Gordis45). The studies did not follow the individuals over time or account for the length of time the individuals lived in the communities. Exposures were measured in the most recent environment, which may not represent the majority of the lifetime exposure as noted by Burdette and Whitaker(Reference Burdette and Whitaker38). The results can suggest avenues of research that show promise for detecting an aetiological relationship, but cannot determine causation.
Six of the seven studies used secondary aggregate data or partnered with other ongoing research studies, which limited the collection of data about the variables under study(Reference Burdette and Whitaker38, Reference Inagami, Cohen, Finch and Asch39, Reference Maddock41–Reference Sturm and Datar44). The secondary data included large data sets, such as the US Census and national longitudinal studies, making small-scale analysis difficult. In addition, combining multiple databases that have different data collection periods erroneously assumes that dynamic variables like the food environment and BMI remain constant over a period of time greater than 1 year(Reference Morland, Diez Roux and Wing42). In addition, some of the data sets used in the studies had limited sample sizes in subpopulations of racial or ethnic minorities(Reference Maddock41, Reference Morland, Diez Roux and Wing42). Only two studies examined children(Reference Burdette and Whitaker38, Reference Sturm and Datar44), indicating a need for more evidence on the effect of the food environment on minority populations and children. Some studies did not measure or control for key confounders or mediating variables such as dietary patterns(Reference Morland, Diez Roux and Wing42, Reference Sturm and Datar44), physical activity levels(Reference Burdette and Whitaker38, Reference Inagami, Cohen, Finch and Asch39) and/or sociodemographic characteristics(Reference Simmons, McKenzie, Eaton, Cox, Khan, Shaw and Zimmet43).
Several limitations were found in the measurement of the food environment. All seven of the studies focused primarily on the community food environment in measuring the geographic proximity, density or number of outlets per capita. The exceptions include Sturm and Datar’s measurement of food prices(Reference Sturm and Datar44) and Inagami et al.’s measurement of the deprivation score of the grocery store census tracts as a proxy for grocery store quality(Reference Inagami, Cohen, Finch and Asch39). No studies directly measured the consumer environment at an individual food outlet level and its relationship with obesity, highlighting a large gap in the literature.
All but one of the studies defined communities with administrative units, including census tracts(Reference Inagami, Cohen, Finch and Asch39, Reference Morland, Diez Roux and Wing42), political jurisdictions(Reference Burdette and Whitaker38), shires(Reference Simmons, McKenzie, Eaton, Cox, Khan, Shaw and Zimmet43), metropolitan areas(Reference Sturm and Datar44), zip codes(Reference Sturm and Datar44) or states(Reference Maddock41). Many of these areal units do not have a conceptually founded geographic context to the health outcome of overweight or obesity. In defining communities with these administrative boundaries, the actual shopping or dining locations of individuals living or working in those areas could be misidentified, as noted by Morland et al.(Reference Morland, Diez Roux and Wing42). In addition, since objects are spatially related and not independent, many analyses should have controlled for spatial autocorrelation(Reference Wong46).
With the exception of Sturm and Datar(Reference Sturm and Datar44), the studies examined food stores or restaurants and placed emphasis on fast-food restaurants. While many organizations play a role in the food environment, it may be useful to examine all types of food outlets and their geographic arrangement in one analysis to synthesize the community food environment. In addition, the categorization of food outlets was inconsistent between studies. The 1997 NAICS codes were used to classify types of restaurants and stores in two studies(Reference Morland, Diez Roux and Wing42, Reference Sturm and Datar44). Other studies did not specify their classification system, did not distinguish between different types of outlets or stipulated their own definitions of various outlets.
All of the studies utilized indirect methods to assess the environment, including telephone directories, census information or large databases. Simmons et al.(Reference Simmons, McKenzie, Eaton, Cox, Khan, Shaw and Zimmet43) used some direct observation of restaurant locations, but also depended on indirect methods. Indirect measurement methods are convenient and inexpensive, allowing researchers to study large samples with minimal resources. However, they can provide imprecise and outdated information as noted by Booth et al.(Reference Booth, Pinkston and Poston47). Along with the other limitations, this may have contributed to the lack of associations present in the limited research on the community food environment and obesity.
Conclusion
Research on obesity and the community food environment is in an early stage. Few studies have explored the link between obesity and the community food environment, and none has examined individual consumer environments. The seven pioneering studies utilized different methods and found mixed results. The samples, food environment variables, measurement methods and analysis varied substantially, making a comparison of the results difficult. Two studies did not find any significant association between obesity rates and the community food environment variable(Reference Burdette and Whitaker38, Reference Simmons, McKenzie, Eaton, Cox, Khan, Shaw and Zimmet43); however, five others found significant results(Reference Inagami, Cohen, Finch and Asch39–Reference Maddock41, Reference Morland, Diez Roux and Wing42, Reference Sturm and Datar44). Significant findings were related to presence of different types of food stores, fruit and vegetable prices, disadvantage of the food store neighbourhood, distance travelled to the food store and distribution of fast-food restaurants on a state-wide basis. Only one study found a significant inverse relationship between BMI and restaurant density around work, which was only present for men(Reference Jeffery, Baxter, McGuire and Linde40). All of the studies suffered from limitations, possibly limiting the generalizability and validity of the findings. While no overarching conclusions can be drawn at this juncture, these innovative studies form a firm foundation of research and establish a direction for future research.
Implications for future research
Future research should include more in-depth descriptive studies of small areal units, selected with conceptual significance, in order to identify precise factors in the community and consumer food environment that are correlated with obesity. The relationship between obesity and consumer food environments, including food price, placement, availability, quality and promotion in individual food outlets, should be explored. In addition, researching multiple environments in one design would allow for a comparison of the effects of different environments, such as the home, community and consumer environments. By designing a study that addresses multiple environments, the results can be placed within the ecological model to gain a broader understanding of the overlapping contexts of the food environment. After more precise risk factors are identified across multiple environments, longitudinal designs will be needed to understand long-term exposures and the causal pathway between the food environment and obesity.
There is a crucial need for studies that measure the environment through direct methods, such as individual inventory of neighbourhoods or food outlets. Community Service Unlimited provides an example of direct measurement of neighbourhoods with their extensive, community-based, food environment assessment(48). Glanz et al. developed the Nutrition Environment Measures Survey (NEMS) that reliably measures and scores store and restaurant environments(Reference Glanz, Clawson, Young and Carvalho49). Standardizing these tools and procedures will make it possible to compare similar food environments and research results across communities.
In addition to directly measuring the food environment, researchers should measure and adjust for all intervening individual-level factors, such as sociodemographics, dietary patterns, physical activity and physical inactivity. As discussed by Diez Roux(Reference Diez Roux50), contextual and multilevel analyses allow the assessment of individual-level characteristics as modifiers of the environmental effect. This level of analysis is necessary to identify the independent effect of the community environment on individual health outcomes like obesity(Reference Diez Roux50). Furthermore, future research should include racial and ethnic minorities and children, since they are under-represented in the current literature and suffer from disparities in health and environmental exposures(Reference Kumanyika and Grier51).
Future research exploring the relationship of obesity and the food environment should aim to address these recommendations and fill the gaps in the literature. It is important to build on the foundations provided by these studies and more extensively explore what makes up healthy food environments and how individuals interact with them. The food environment holds great promise for a lasting solution to the obesity epidemic, and more in-depth research is urgently needed.
Acknowledgements
The author is supported by the University of Pennsylvania, School of Nursing, as a doctoral student. The author declares no financial conflict of interest that would bias the information presented. Dr Charlene Compher was kind enough to read and comment on a draft of this manuscript.
