The impact of poor childhood nutrition has consequences for both short-term and long-term health(Reference Williams and Strobino1, Reference Mikkilä, Räsänen and Raitakari2) and the escalating rates of obesity in childhood are becoming a significant public health issue(Reference Cali and Caprio3). A number of factors contribute to poor food intake, including socio-economic status, parental and child food choice and preferences, and the accessibility of foods. There is growing evidence to show that those of lower socio-economic status have to spend a higher proportion of their income on food(Reference Friel, Walsh and McCarthy4) and tend to consume less healthy diets(Reference Estaquio, Druesne-Pecollo and Latino-Martel5, Reference Martikainen, Brunner and Marmot6).
Research in the USA has shown that people living in areas of low socio-economic status have poorer access to supermarkets, resulting in less access to healthier foods, and the foods that could be purchased were more expensive(Reference Wang, Kim and Gonzalez7). Other research shows that those living in deprived areas had much greater access to fast-food retailers and other retailers selling less healthy products(Reference Macdonald, Cummins and Macintyre8–Reference Kipke, Iverson and Moore11). To date there has been limited and inconsistent evidence that these ‘food deserts’, where residents have limited access to healthy food, exist in the UK(Reference Macintyre12). However, the available previous research has focused on more densely populated urban areas(Reference Pearson, Russell and Campbell13) and there may be particular problems with access to or availability of healthier foods for those living in smaller towns and villages. Furthermore, there have been few studies on the relationship between neighbourhood food outlets and food intake in children, although the findings from the limited available evidence suggest that children living in areas with higher densities of fast-food retailers tend to make poorer dietary choices(Reference Timperio, Ball and Roberts14–Reference Jago, Baranowski and Baranowski16).
While there has been some previous research on the role of the neighbourhood environment in determining food choice, there have been limited studies employing Geographic Information Systems (GIS)(Reference Timperio, Ball and Roberts14, Reference Jago, Baranowski and Baranowski16–Reference Larsen and Gilliland19) in order to produce objective indicators of different domains of the local food environment, such as distance to and density of food outlets. Furthermore, all previous studies of children that have used GIS data have focused only on fruit and vegetable consumption(Reference Timperio, Ball and Roberts14, Reference Jago, Baranowski and Baranowski16, Reference Hackett, Boddy and Boothby18), whereas the food environment is likely to influence other food choices as well.
Since childhood nutrition is important for future health, and rates of childhood obesity are rising, it is important to understand the relationship between food intake and food access among this group. Therefore the present study was undertaken to investigate the relationship between distance to and density of foods in the local neighbourhood and both healthy and less healthy food choices in a population of 9- to 10-year-old children in the UK.
Methods
SPEEDY
The SPEEDY (Sport, Physical Activity and Eating behaviour: Environmental Determinants in Young people) study was set up to quantify the potential correlates of levels of physical activity and dietary habits in 9- to 10-year-old schoolchildren (Year 5) in the county of Norfolk, England. A detailed description of the methods adopted has been published previously(Reference van Sluijs, Skidmore and Mwanza20). Ethical approval for the SPEEDY study was obtained from the University of East Anglia local research ethics committee.
Sampling
A cluster sampling strategy based on schools was adopted. Schools in Norfolk with eligible children attending were sampled to attain heterogeneity in location. The urban/rural status of all potentially eligible schools was determined using Bibby and Shepherd’s classification of rurality(Reference Bibby and Shepherd21), and one of four area profiles, ‘Urban’ (>10 000 residents), ‘Town and fringe’, ‘Village’ and ‘Hamlet and isolated dwelling’, was assigned to each. In total, 157 schools were approached and measurements were conducted at ninety-two schools. All Year 5 children within participating schools (n 3619) were invited to participate and parental informed consent was obtained from 2064 children (57·0 % response rate).
Examination of children at school
Data collection was performed during the summer term of 2007 (April to July), and participating children were visited at school by teams of two or more trained research assistants. They collected a range of data according to standard operating procedures including anthropometry, demographic information, school-level information, and details of children’s home and neighbourhood environment. A questionnaire was also completed by a parent or main carer of each child.
Children self-reported their date of birth, and age was calculated at time of examination. The parent or main carer self-reported their age at leaving full-time education (in categories). Simple non-invasive anthropometry measures were conducted using standardised procedures, with children dressed in light clothing. Portable Leicester height measures (Seca, Birmingham, UK) were used to measure height to the nearest millimetre. A non-segmental bio-impedance scale (type TBF-300A; Tanita, Tokyo, Japan) was used to measure weight (to the nearest 0·1 kg) and impedance. BMI was calculated as weight (in kilograms) divided by the square of height (in metres). Obesity status was determined using gender- and age-dependent cut-off points(Reference Cole, Bellizzi and Flegal22). Index of Multiple Deprivation (IMD) scores(Reference Noble, Wright and Dibben23), a measure of area material deprivation, were assigned to each child based on the location of their residential postcode and results were allocated into quintiles of deprivation (NQD), using the national categories(Reference Noble, Wright and Dibben23). A higher IMD score reflects a higher level of deprivation.
Dietary assessment
Food choice was assessed using an adapted version of the Health Behaviour in School Children (HBSC) questionnaire, which was designed to collect self-reported information on food choices in children in Europe and North America(Reference Vereecken and Maes24). It consists of a fifteen-item questionnaire of the most commonly consumed foods in participating countries that has been previously validated and can be used for ranking participants for most food items(Reference Vereecken and Maes24). Adaptations to the items to increase applicability to our age group included removing alcohol consumption, adding fruit juice consumption, and separating sweets and chocolate into two separate items. The children reported the frequency of consumption on a 7-point scale (‘never’, ‘<1/week’, ‘1/week’, ‘2–4 d/week’, ‘5–6 d/week’, ‘once a day, every day’, ‘every day, more than once’)(Reference Vereecken and Maes24).
Food accessibility
The computation of the accessibility measures was undertaken in the ArcGIS package (ESRI Inc., Redlands, CA, USA). All facilities in the study area serving food for consumption off the premises were mapped based on their location identified in the UK Ordnance Survey ‘Points of Interest’ database. ‘Points of Interest’ provides data on the location of commercial facilities, which are coded according to type. For this work we classified each food retailer according to whether it was a supermarket (a large food store selling a wide variety of foods), a convenience store (a small neighbourhood retailer with more restricted choice), a takeaway (a retailer selling hot food for consumption off the premises) or another type of food shop. Other food retailers included greengrocers, bakers, butchers, delicatessens and other independent food retailers.
Two measures of accessibility were computed for each outlet, both based on a pedestrian network created in the GIS using the location of roads and public footpaths. The first measure was the network walking distance from the child’s home to the nearest food retailer of each type, which was calculated in kilometres. The second measure was of the density (per square kilometre) of each type of facility within the neighbourhood of each child. The neighbourhood was delineated as the area within 800 m (roughly equating to a 10 min walk) along the network from a child’s home.
Statistical analyses
Descriptive data were summarised as means with standard deviations or percentages. Differences in overweight and obesity between boys and girls were investigated using the χ 2 test, while all other gender differences were determined using Student’s t test.
Regression analyses were performed using generalized estimating equations and included a food environment variable (either distance to or density or food outlet) and a single food variable. A two-level hierarchical structure was used to allow for the clustering in food consumption behaviours among children attending the same schools(Reference Goldstein25). The influence of confounding variables was assessed by fitting two models, with Model 1 including gender and clustered by school. As both individual- and neighbourhood-level deprivation are associated with food choice, Model 2 was adjusted for gender, BMI, parental education, IMD and urban/rural location and also clustered by school. All regression coefficients are presented as portions per week. The β coefficients are presented per kilometre increase for distance variables, and per outlet per square kilometre increase for density variables. All analyses were performed using the STATA statistical software package version 10 (StataCorp LP, College Station, TX, USA).
Results
Of the 2064 children recruited to the study, GIS and IMD data were available for 2023 (98 %) children and 2056 (99·6 %) completed the FFQ. BMI Z-scores were calculated for 93 % of children (n 1923) and data on parental education was available for 91·4 % of participants (n 1887). The number of participants with data for all variables was 1721 (83 % of the original sample).
In our study population, 77 % of children were of normal weight, with 23 % defined as overweight or obese (Table 1). There were gender differences, with a lower percentage of girls being of normal weight compared with boys (P = 0·005) and significant differences in intakes of all foods apart from fruit juices, crisps and white bread (Table 1). The distribution of IMD scores (Table 2) shows that 21 % of the children lived in households that were located in areas falling in NQD categories 1 and 2 (most deprived), while 47 % were in categories 4 and 5 (least deprived). Only 20 % of the children lived in households where parental education continued after the age of 18 years. About 39 % of participants lived in an urban area and the majority of subjects lived over a kilometre from their nearest food retailer or takeaway. Mean density of food outlets ranged from 0·3 to 1·9/km2, with the accessibility and provision of supermarkets being the poorest (Table 2).
Table 1 Rates of overweight and obesity and frequency of food consumption in the study population: boys and girls aged 9–10 years, Norfolk, England, April–July 2007
IOTF, International Obesity Taskforce.
†Obesity status was determined using gender- and age-dependent cut points according to the method of Cole et al.Reference Cole, Bellizzi and Flegal(22).
‡P for the difference between boys and girls.
Table 2 Characteristics of the study population and access to food outlets: boys and girls aged 9–10 years, Norfolk, England, April–July 2007
Tables 3 and 4 show the relationship between weekly food intake frequency and distance to and density of food outlets, respectively. Distance to and density of food outlets did not appear to impact on the consumption of the following foods included in the HBSC questionnaire: whole milk, semi-skimmed or skimmed milk, brown or wholemeal bread, cheese or sugar-free soft drinks (data not shown).
Table 3 The relationship between distance to food outlets and food intake frequency per week: boys and girls aged 9–10 years, Norfolk, England, April–July 2007
Model 1: adjusted for gender, clustered by school.
Model 2: adjusted for BMI, parental education, Index of Multiple Deprivation score, location and gender, clustered by school.
Coefficient was significant: *P < 0·05, **P < 0·01.
Table 4 The relationship between neighbourhood density of food outlets and food intake frequency per week: boys and girls aged 9–10 years, Norfolk, England, April–July 2007
Model 1: adjusted for gender, clustered by school.
Model 2: adjusted for BMI, parental education, Index of Multiple Deprivation score, location and gender, clustered by school.
Coefficient was significant: *P < 0·05, **P < 0·01.
Supermarkets
Living further away from a supermarket was associated with a generally more favourable food intake; fruit intake was significantly higher (0·11 portions/week increase per 1 km increase in distance from a supermarket after multivariate adjustment, P < 0·05) as was vegetable intake (0·11 portions/week, P < 0·05), while white bread intake was significantly lower (−0·11 portions/week, P < 0·05; Table 3). Higher density of supermarkets in a neighbourhood was related to an increase in vegetable consumption (0·31 portions/week per unit increase in food outlets in the surrounding 1 km2 after multivariable adjustment, P < 0·05). A similar increase in fruit intake was observed although this did not reach statistical significance. There were also significant increases in intake of sweets, sugary soft drinks, breakfast cereals and white bread for each additional supermarket per square kilometre in Model 1, but these became non-significant in Model 2.
Convenience stores
Living further away from a convenience store was associated with a more favourable food intake (Table 3), although the finding of increased vegetable intake with increasing distance became non-significant after full adjustment. Nevertheless, reported consumption of crisps, chips, sweets, chocolate and white bread all declined with increasing distance from convenience stores in both models. Associations with the density of convenience store provision were less strong, although an increased density of convenience stores in the local area was associated with a significant increase in consumption of fruit juices (Table 4).
Other food retailers
For both distance and density, there were no consistent statistically significant associations with food intake (Tables 3 and 4).
Takeaways
Living further away from takeaways was associated with more favourable food intakes, as consumption of crisps, chocolate, sugary soft drinks and white bread was negatively associated in both models (Table 3). Associations were weaker for density (Table 4), although vegetable intakes were higher with increasing density (0·12 portions/week increase in vegetable consumption per outlet per km2 increase, P < 0·05).
Discussion
Our analyses suggest that both distance to and density of local food outlets were associated with food intake in children. Close proximity to supermarkets and convenience stores was generally associated with a higher prevalence of reporting of unhealthy food choices, and reduced consumption of fruits and vegetables. Closer proximity to convenience stores was also associated with higher reported intakes of snack foods, including sweets, chocolates and crisps. These associations were observed even after adjusting for major influences on food intake, including location, socio-economic status and BMI. A high density of supermarkets and convenience stores in the neighbourhood was associated with higher fruit and vegetable intakes, and this observation was statistically significant for supermarket access. However, the density of supermarkets was also associated with increased intakes of some unhealthy snacks, including sugary drinks and sweets. Although we found evidence of effects, overall the observed effect sizes were small, suggesting that distance to and density of food outlets in the neighbourhood play only a minor role in influencing food consumption patterns in these children.
This is one of the first studies to have taken an integrated approach to address the combined effects of the local food environment, such as distance to and density of food outlets in the local environment(Reference Friel, Walsh and McCarthy4, Reference Wang, Kim and Gonzalez7, Reference Kipke, Iverson and Moore11, Reference Pearson, Russell and Campbell13, Reference Larsen and Gilliland19). Our findings on the effects of distance and density are in agreement with previous research in other countries. With regard to distance to food outlets, in a US study, living closer to small food stores and fast-food retailers resulted in lower consumption of fruits and vegetables(Reference Jago, Baranowski and Baranowski16). To date there are fewer data available on neighbourhood density of food outlets and food choice in children, but recently Timperio et al.(Reference Timperio, Ball and Roberts14) found that a higher density of fast-food retailers and convenience stores was associated with lower levels of fruit and vegetable intake in Australian children. In another study, access to multinational fast-food outlets was not related to fruit consumption in adults, but living further away was associated with higher vegetable intakes, while access to locally operated fast-food outlets was not related to fruit and vegetable intake(Reference Pearce, Hiscock and Blakely17).
Our findings support the limited evidence from previous research in the UK, and suggest that proximity to a wide variety of food choices does not necessarily lead to optimal food intake(Reference Cummins and Macintyre26, Reference Dibsdall, Lambert and Bobbin27). Our results also concur with another UK study(Reference Hackett, Boddy and Boothby18) which showed that children from similar socio-economic backgrounds, but who lived in divergent environmental conditions, had differing eating habits. It was found that those children who lived in the area with a higher density, safety of access to and longer opening hours of shops made poorer food choices. In our work, the children who lived furthest from shops generally had the healthiest food choices, but the impact of both distance to and density of food outlets on food choice was small, suggesting that other factors may play a more important role in food choice in children. For example, the foods available at or around school may influence children’s food choice, and associations may possibly be influenced by fruit and vegetable preferences, but to date interventions to improve healthy food availability have had limited success(Reference Jago, Baranowski and Baranowski16, Reference Moore and Tapper28).
The finding that close proximity to supermarkets was related to reduced consumption of fruits and vegetables, while the density of supermarkets in the area was positively related to increased vegetable consumption, is difficult to explain. The implications of these findings are unclear, but the increased fruit and vegetable intake by children who lived further away from supermarkets may in part be explained by the high proportion of children who lived in rural communities in our study, who may have more access to locally produced produce/home-grown fruit and vegetables. The finding that high density was associated with increased intakes of both healthy and unhealthy food choices is also difficult to explain but may relate to increased availability, choice and bulk purchasing of all food types in supermarkets.
Our study has a number of strengths and limitations. Currently the relationship between food choice in the local environment is relatively under-investigated, with those studies that have been undertaken often not taking an integrated approach to address the combined effects of distance to and density of food outlets in the neighbourhood in relation to intake in children(Reference Friel, Walsh and McCarthy4, Reference Wang, Kim and Gonzalez7, Reference Kipke, Iverson and Moore11, Reference Pearson, Russell and Campbell13, Reference Larsen and Gilliland19). By using a validated survey instrument together with a well-characterised sample of children, and combining information on reported frequency of consumption of key food items with a detailed GIS database on the location, type and accessibility of food retailers, we have provided new evidence on associations between the neighbourhood food environment and dietary choice in UK children.
In terms of limitations, our study is cross-sectional in nature and therefore we are limited in our ability to ascribe causality to the associations detected. The data collection was carried out in the school summer term and, owing to greater availability and range of fruit and vegetables during this season(Reference Givens, Lu and Bartell29), recorded intakes may not reflect average intake over a year. Although we used a detailed and contemporary GIS database of food store locations, it is likely that some facilities we included were no longer open, while newly opened outlets were excluded. Our sampling strategy was designed to achieve maximum environmental heterogeneity in order to address questions relating to environmental influences on behaviour, and that led to deliberate oversampling from schools in rural areas, which may limit comparisons with other studies that have predominantly been undertaken in large urban areas. Proximity and density of food outlets may be indicative of a number of environmental factors, and although in our analyses we attempted to account for this by adjusting for IMD, there may be other environmental factors that have not been accounted for. Furthermore, as the SPEEDY study was carried out in an area of the UK with a low proportion of families from different ethnic backgrounds, our final sample is mainly white, with only 3·8 % of children from other ethnic backgrounds. Although this is representative of the Norfolk population, it means we were unable to examine how associations varied with ethnicity.
The short HBSC questionnaire was used to assess intake of the most commonly consumed foods rather than exact intake of nutrients, in order to provide simple information on trends in dietary intake. We used the HBSC FFQ because it is known that children’s recall of food intake is prone to reporting errors(Reference Domel30, Reference Crawford, Obarzanek and Morrison31) and it has been previously validated(Reference Vereecken and Maes24); patterns of food consumption in the present study are largely in agreement with more precise dietary methodology used in the UK National Diet and Nutrition Survey in children aged 7–10 years(Reference Gregory and Lowe32).
In conclusion, our findings show that the impact of both distance and density of food outlets on children’s food choice was small, with close proximity to supermarkets and convenience stores being associated with reduced consumption of fruits and vegetables, and proximity to convenience stores associated with higher intakes of snacks and confectionery. Conversely, a high neighbourhood density of supermarkets and convenience stores was associated with higher fruit and vegetable intakes and, for supermarkets only, higher intakes of snacks and confectionery. Despite the small magnitude of the associations we detected, the effects of individual foods combined could be important, particularly as even small differences in intake can contribute to the onset of childhood obesity.
Acknowledgements
This research was funded by the National Prevention Research Initiative (http://www.npri.org.uk) with support from the following organisations: British Heart Foundation; Cancer Research UK; Chief Scientist Office, Scottish Government Health Directorate; Department of Health; Diabetes UK; Economic and Social Research Council; Health & Social Care Research & Development Office for Northern Ireland; Medical Research Council; Welsh Assembly Government; and World Cancer Research Fund. None of the authors has a conflict of interest. S.G., E.v.S., P.S., A.C. and A.J. were responsible for conception of the study, study design, setup, and data collection and processing. F.H. participated in data collection and processing. P.S. and A.W. analysed the data and P.S. tabulated the data. P.S., A.W. and A.C. drafted the manuscript. I.H. assisted with drafting the manuscript. All authors were involved with data interpretation and critical revisions of the paper, and provided approval for publication.
