The "epidemic" of obesity throughout the world has prompted a search for environmental causes, over and above individual-level explanations. Researchers have focused in particular on the potential influence of residential neighborhood environments on people's health behaviors, including dietary practices and physical activity patterns . Methodological developments using GIS and the availability of detailed spatial data have contributed to advances in this field. Using GIS-based methods, the present study examined associations between the local food environment and BMI for Japanese older adults. To our knowledge, this is the first study to test such associations with a sample of older Japanese adults.
By extending the scope of inquiry to a non-western country, our study adds to the evidence base for the potential influence of the food environment. Our findings suggest that the evidence accumulated so far in predominantly western settings cannot necessarily be generalized to other cultural contexts. For instance, we found an unexpected positive association between supermarket accessibility and BMI, which is contrary to what has been described in the US. In general, studies in the US have suggested that better access to supermarkets is related to healthier food intake and lower levels of obesity, since supermarkets tend to offer a variety of high-quality products at lower cost . Indeed, supermarkets are treated as a proxy for healthy food choices in the US to the extent that communities lacking in supermarkets have been labeled as "food deserts" .
We also found no main associations between fast food outlets/convenience stores and the average BMI of residents. Our null finding may be due to several factors, including a) differences in the range of foods sold in convenience stores (e.g., many Japanese convenience stores sell fresh produce), and/or b) the specificity of the association between fast food outlet exposure and the BMI of particular subgroups (e.g., the results might have been different if we had examined a school-aged population). We found a positive interaction between the BMI of individuals living alone and more proximate access to fast food outlets (p = 0.02), thus providing some support for the hypothesis that fast food stores contribute to the "obesogenic" environment. This result seems to be important because the number of older people living alone is increasing as the Japanese society rapidly ages. Taken together, our findings suggest that the extrapolation of research results for local food environments must be done cautiously when involving heterogeneous cultural settings.
According to the OECD Health Data, the population prevalence of obesity (BMI ≥ 30) is 3.4% in Japan, 24.0% in the UK, and 34.3% in the US (as of 2006). Considering that the Japanese population has such a low prevalence of obesity or being overweight, and that our respondents were older adults, our results need to be interpreted with care. Among our respondents, the proportions of those overweight and obese were only 20.8% and 2.2%, respectively (see Table 1). Meanwhile 7.4% of our sample was categorized as being underweight. Recent cohort studies on the association between BMI and all-cause mortality showed that being underweight was significantly associated with increased risk of mortality, while being overweight was not related to the risk of mortality among Japanese older adults (being obese were only partially and weakly associated with mortality) [18, 27]. For example, recent Japanese studies on "food deserts" have discussed the phenomenon for older adults as a hazard for under-nutrition, rather than a risk factor for being overweight or obese . Thus, if the better access to supermarkets reduced the risk for being underweight, increased BMI could contribute to better health consequences among older adults. In our study; however, the logistic regression analysis that used categorized BMI (i.e., underweight vs. overweight or obese), exhibited no significant association between access to supermarkets and being underweight. In contrast, number of supermarkets and distance to nearest supermarket were associated with being overweight or obese. The policy implications of our findings are therefore ambiguous, and we cannot determine (based on our findings among older adults alone) whether or not improving access to supermarkets would result in a net population health gain.
Our study has a number of notable strengths. First, the relatively large sample size contributed to statistical power in detecting associations, including the ability to check interactions between living status and food environment measures. Second, our characterization of the neighborhood environment was based on street networks (as opposed to circular buffers), producing a more accurate representation of accessibility to food outlets. The GIS approach has an additional advantage in assessing spatial variations that are independent of arbitrarily defined administrative boundaries [7, 29]. Lastly, our analyses simultaneously considered the local food environment and built environment for physical activity. Previous studies of neighborhood contexts have tended to focus on one or the other, but seldom considered the potential confounding of the food environment by built environment characteristics. To the extent that various "desirable" characteristics of neighborhoods tend to cluster, it is important to check that any putative influence of the food environment on BMI is not confounded by co-occurring built environment characteristics .
Our study is subject to a number of limitations. First, our study area may not be representative of Japan. Our findings need to be replicated in other areas, particularly in the larger metropolitan areas. Self-reported height and weight was also a limitation in this study, as they might have been misreported. Although our consideration of a potential confounding effect by the built environment for physical activity was a strength in this study, we only used two variables (population density, and parks or green spaces). We need to obtain a more comprehensive picture of neighborhood environment, including neighborhood safety and social cohesion/capital. R squared values in our models were small, indicating that the independent variables only partially explained the variance of BMI among respondents. Incorporating variables that are more directly related to BMI (e.g., dietary intake) should be considered in future studies.
Another limitation is the validity of data used to capture the local food environment. Some authors have questioned the validity of publicly available lists of food stores [2, 30]. In addition, a three to four year gap exists between the data for individuals in our sample and the Yellow Pages, from which food store information was collected. Some validation studies have supported using commercial databases [31, 32], while others have raised doubts about using them [33, 34]. Since the number of business establishments listed in the Yellow Pages exceeds that given in the Establishment and Enterprise Census of Japan  (p. 27), the risk for undercount error would be low. As for an overcount error, we confirmed that 92.5% of the stores in the study area were successfully found at the addresses given by the Yellow Pages. Generally, we consider using the Yellow Pages to capture the local food environment acceptable for our study.
The definitions of different categories of food stores and restaurants are also debatable [2, 36]. Although the industrial categories of the Yellow Pages are considered to be specific, the size and type of stores could be mixed, since explicit information about how store types are defined was not provided (e.g., floor areas or products). In the future, a culture-specific research approach might be useful for examining the consumer nutrition environment (i.e., what consumers encounter within and around a retail food outlet)  in a variety of localities.
Finally, with our cross-sectional data, causal inferences should not be made, and even with longitudinal data, the causal association between food outlets and behavior (dietary practices and BMI) can be problematic because of the residential selection and store location preferences . Thus, even if longitudinal data was available, a positive association between the opening of a local fast food outlet and subsequent increases in the BMI of nearby residents does not necessarily imply that one caused the other. Fast food retailers do not open new stores in random locations; rather, they are likely to carefully select locations based on a profit motive; i.e., based on the best available information on the residential demand for their products. To explore possible causal links between food environment and weight gain/lose, further investigation of shopping and eating behaviors (i.e., when, where, and what residents buy and eat) should be carried out in neighborhoods with a variety of local settings.
In summary, our study provides initial evidence for extending the investigation of retail environmental influences on obesity risk within a non-western setting. Our findings provide mixed support for the types of food environment measures commonly adopted in western studies. Importantly, our results suggest the need to develop culture-specific approaches for characterizing neighborhood contexts when hypotheses are extrapolated across national borders.