Key results
Using longitudinal data, our findings demonstrate few associations between the food environment and measures of adolescent weight status. Where we detected associations, they mostly fell in the hypothesised direction. The presence of a fast food outlet along the estimated travel route between school and home, and the presence of ‘other retail outlets’ near schools were both positively associated with weight status. While upon initial inspection their effect sizes (i.e. coefficients) appear small and not clinically significant, they refer to the change in weight status for an additional outlet. When considered alongside an additional 10 outlets, the effect size is clinically significant. Such a finding would be appropriate for the number of fast food outlets along the travel route since the interquartile range (IQR) was 7–23 (although the coefficient was not significant for BMI SDS). However, for the number of ‘other retail outlets’ it is a less suitable observation since there does not appear to be this magnitude of variation (IQR 6–9).
Limitations
Longitudinal data were not used for the exposure variables since they were not available. Food retail environments do not change quickly (particularly in comparison to the time period of our study), suggesting that our data are appropriate and the impact of bias on our results will be low [
30]. Information on key covariates such as diet, physical activity or parental characteristics, which have all been shown to be associated with adolescent weight status, were not available. These unknown characteristics may confound the associations observed. Data collection was also unbalanced, with varying sample sizes between years which may introduce bias into our estimates.
We used buffers to measure exposure to aspects of the food environment. It is unlikely that these buffers correspond to the actual food environments individuals engage and interact with [
31]. While we find little evidence of the role of geographical context in our study, the role geography plays is complex and likely to operate at different interacting scales (i.e. local vs regional food environments). It is also difficult to measure the exact environments individuals engage in without using GPS data. Our use of buffers was to provide a proxy measure to capture different routes or uses of environments, but they are only estimates. As such, we should acknowledge that the use of single point estimates to measure the ‘home’, ‘school’ and ‘commuting’ environments may introduce the locational fallacy, since we do not know the exact exposure of individuals to their food environments. We did calculate smaller buffers (500 m) for the home and school environment to test the impact of the more immediate environment, however this did not alter our findings (results not shown). Travel routes were also estimated since the data were not available limiting the observations that can be inferred from them. Since one of the positive associations found was along the travel route, we can only suggest that this is an interesting avenue for future research to explore in greater detail.
Our study focuses only on one aspect of accessibility (i.e. geographical proximity), however it is important to account for broader aspects of the food environment. Home and school characteristics themselves have direct impacts of dietary choices [
32], and they may influence how individuals engage with surrounding neighbourhood features. In this vein, characterising the types of food sold by outlets, the price of food options (also linked to an individual’s food purchasing power) or their opening hours will be important in providing greater detail on the role of environmental features.
Finally, our study was purely descriptive in exploring the association between the density of features of the food environment and measures of weight status. While this is useful for initially understanding if any association(s) exist, and follows previous studies (e.g. 15,28), there is greater need in future research to explore the causal pathways and mechanisms through which the food environment may influence weight status.
Interpretation
Our study makes a novel contribution to the literature building on previous studies by using longitudinal data across the period of secondary school, an important period in the anthropometric development in children and one which is key for their risk of obesity during adolescence and into adulthood [
10]. We examined two measures of weight status and multiple spatial contexts (home, school and travel route environments) to capture the wider experience of food environments.
The majority of studies examining the role of food environments on weight status are cross-sectional and therefore less able to draw out cause and effect (even if analyses are carried out within a causal framework). Findings from cross-sectional studies are ambiguous with some reporting positive associations between fast food outlets and weight status [
17‐
20], but a comparable number finding no association (19–21). The lack of consistency across the literature may be a symptom of residual confounding across studies, suggesting the importance of correct study design when using observational data [
33]. A null association may simply be because there is too much ‘noise’ in the data to be able to observe the true effect. It is important for future research to develop stronger causal models that can be evaluated will help to lessen the impact of residual confounding.
To our own knowledge, our study is the first longitudinal analysis using UK data. Sturm and Datar [
21] investigated the association between food outlet density and changes in BMI over 1 and 3 years among elementary school children in the USA, finding null associations [
21]. Although interestingly they did report differential gains in BMI according to geographical variation in fruit and vegetable prices, possibly suggesting that additional measures beyond simple density metrics are required to expose geographical associations. Adolescents develop at different stages, so taking a longitudinal approach is more appropriate than a snap shot of cross sectional data. In addition to changes in physical development, adolescent behaviours are likely to change during this period as they transition toward independence. The longitudinal data will go some way to encapsulate this. That being said, our results using longitudinal data do match results from the baseline cross-sectional analysis reported previously [
22].
Our findings have important policy implications. The inconsistency of evidence from our study and throughout the literature between the density of fast food outlets and weight status suggests that efforts to restrict the locations of new fast food outlets may be of limited value as a standalone intervention to tackle childhood obesity [
13‐
16]. We do not contest that strategies to improve the food environment surrounding homes and school are logical. There have been long term shifts of increasing access to unhealthy foods corresponding to increased prevalence of obesity [
30]. Rather policies need to be move beyond simply restricting their numbers alone. Table
2 demonstrated that while in our study the median number of fast food outlets surrounding homes, schools and travel route was similar (12, 12 and 11 respectively), the interquartile ranges suggests that there was considerable variability. It could be hypothesised that restricting the location of new fast food outlets in areas with few fast food outlets may have a larger effect (assuming that a community has enough food sources). However, the majority of areas could already be at a level of saturation whereby restricting further outlets will likely have little impact.
If Local Authorities are unable to make significant changes to the spatial availability of unhealthy foods, then they need to consider alternative strategies dealing with the existing environments. Licensing may offer one approach that could be effective at addressing the whole food environment, such as reducing opening hours, encouraging firms to use less cooking oil or introducing price subsidies for healthy food options [
34‐
36]. Similar restrictive licencing policies for alcohol sale, implemented at the local authority level [
37], have been successful in reducing purchases and consumption. In terms of national policy, minimum pricing of alcohol has been estimated to reduce alcohol consumption and subsequent health-related harms [
38]; similar policy interventions directed towards fast food outlets may help reduce the purchase and consumption of unhealthy food. Developing local food systems which promote healthier choices, as well as dealing with wider issues such as food insecurity or the social determinants of poor dietary choices, may be more appropriate policy strategies than simply restricting the location or density of types of outlets. Examining the contribution of these policy scenarios for tackling obesity represent useful avenues for future research.