Obesity and the Built Environment: A Reappraisal

Home as the center for BE exposure

Interest in health geography began in the early 1990s as researchers sought to establish evidence for geographic disparities in health outcomes (12, 64). Concurrently, there was a renewed interest in social determinants of health as well as in the socioeconomic and demographic factors that influence where people live (12, 13). The earliest investigations into the BE and health often relied on neighborhood-level socioeconomic measures, such as indexes of socioeconomic position derived from the Census Bureau in the United States, the Townsend index, the Carstairs index, the Index of Multiple Deprivation in the United Kingdom, and the Socio-Economic Indexes for Areas in Australia, among others (65-70). These indexes were often composite indicators for area-level poverty or wealth, based on sociodemographic data on poverty, employment, education, and income, among other metrics (64-72). In these studies, an individual’s neighborhood was often defined by some administrative boundary (e.g., census tracts in the United States) in which the individual’s home was located (64, 71, 72). Often, actual addresses were geocoded to the centroid of the geographic administrative unit.

Early studies combined individual addresses with area-based, sociodemographic data. The density of fast food restaurants or supermarkets as well as structural environment features, such as land use mix or availability of parks and trails, was calculated per administrative unit (7, 13, 19, 58, 73). However, care was needed in choosing the most relevant level of aggregation because both the scale and shape of the defined geographic units could influence the observed association, introducing a form of statistical bias known as the modifiable areal unit problem (74). The mismatch in geographic scales was resolved through hierarchical modeling (12, 64, 75).

Later studies used prespecified buffer zones, such as 400- or 800-m buffers around an individual’s home using GIS software packages (7, 13, 19, 58, 73). Here, the issue was between the nearest supermarket and the actual destination supermarket. While the location of the nearest supermarket could be readily obtained from GIS data, the location of the destination supermarket could be ascertained only through a survey (19, 36, 76-78). Those studies undermined the main premise of density metrics by showing that people did not shop at the nearest supermarket and did not eat at the nearest fast food restaurant (19, 36, 76-78). Those findings were instrumental in shifting attention to AS captured in space and time.

The postulated links between home BE and individual or neighborhood-level health are summarized in Figure 1. Elements of BE exposure were linked cross-sectionally to the prevalence of obesity, hypertension, diabetes, dyslipidemia, and metabolic syndrome (2, 19, 34, 37, 61, 79, 80). With respect to the PA environment, studies have generally found that greater land use mix, higher residential density, higher walkability scores, and more green space were all associated with lower BMI. Other measures of the BE such as road traffic and noise have been associated with a higher likelihood of having hypertension or dyslipidemia (34). However, there was less evidence to link local PA resources, such as recreational facilities, with body weight or hypertension (34). Moreover, the few examples of longitudinal evaluations of the PA environment and changes in body weight have produced null or inconsistent findings (34, 81).

The evidence on the association between the food environment and obesity was even less consistent (2, 34). Studies have suggested that a higher density of supermarkets or full-service restaurants is associated with lower body weight whereas higher densities of convenience stores or fast food restaurants are associated with higher body weight (2, 20, 29-31, 34-41). However, other studies have not observed the same association between supermarket density and lower BMI (32, 33, 36, 39). Still more studies have been unable to replicate the findings between fast food restaurants or convenience stores and higher BMI (32, 39, 40).

Many early home-centric BE studies had several notable limitations with respect to their study design, measurement of the BE, and their consideration of BE–health pathways. First, the majority of early BE and health studies utilized a cross-sectional study design, which does not allow for a causal interpretation for observed associations between the BE and health (81). Second, home-centric analyses can capture the BE features and resources within an individual’s neighborhood; however, proximity does not allow one to infer usage (19, 36). For example, it was shown that people do not shop for food in their immediate neighborhood but will instead travel great distances to the supermarket or fast food restaurant of choice (36). Third, the choice of BE variables can influence the magnitude and direction of the observed association between BE and health; not all such metrics are predictive of health (82). Relatedly, investigators should be more transparent regarding the variables and methods used to quantify the BE (83). A recent systematic review examined 113 studies (encompassing 1,937 tests of association) that sought to evaluate the relationship between the retail food environment and health (83). The authors found that the reporting of methods was quite poor for more than half of the studies included in the review and that the wide diversity in methods led to an array of conclusions (83). Fourth, some of the theoretical pathways through which features of the BE become internalized to influence health are not always readily apparent (49). Additionally, as indicated in Figure 1, most early studies focused on the relationship between BE and health, with few evaluating its effect on any energy balance behaviors. Few early studies had any actual measures of diet or PA, and those that did usually evaluated these outcomes separately without consideration of the mediating role these energy balance behaviors might have on the BE–health relationship. Fifth, some studies do not adequately disentangle the physical attributes of the BE from the demographic and socioeconomic environments (84). For example, studies have shown that the socioeconomic indicator residential property value is associated with both perceived proximity to neighborhood PA environment features as well as health (31, 36, 40, 41, 76, 84-86).

Some of these limitations can be overcome through the usage of person-centric analyses that incorporate individual-level behaviors (e.g., diet, PA), demographics (e.g., race, culture), socioeconomic status (e.g., property value, educational attainment), and psychosocial factors (e.g., attitudes, perceptions). Researchers should also move to directly examine the intermediary pathways through which BE features might operate to influence health via person-centric attributes. These methodological additions will aid researchers in understanding how individual-level characteristics influence a person’s exposure to BE elements in space and time and how the BE hinders or helps facilitate health-promoting behaviors.

Person as the center of BE exposure

Acknowledging that an individual’s home neighborhood cannot capture the full extent of health-related BE exposure, researchers turned to more advanced GPS technologies to track movements in space and time (14, 15, 18, 44, 45, 47-51, 56). Cumulative mobility over a given time period is the basis for constructing AS areas (18). These AS-derived areas are then supplemented with travel diary data to provide context for the GPS tracks, such as the reason for the trip, identified destination, and any participation in PA or foods consumed (50).

Several different measures of AS have been used. The radius of gyration is defined as the average distance of a set of GPS points or tracks to the most frequented location (87). The standard deviational ellipse of a set of GPS-recorded points is defined as the area that covers approximately 68% of GPS points and that is centered on the average of the point pattern (88). Convex hulls are defined as polygons that contain all GPS points or tracks and have no angles greater than 180° (89). Studies have also utilized a nonparametric method known as 2D kernel density estimation in which a symmetrical kernel function is superimposed over a cluster of GPS points centered around its mean (90). The set of overlapping kernel density functions are then superimposed to create a continuous density surface (90). Other methods include mapping GPS points along street networks, buffered street networks, daily path areas, and more (18).

The literature examining whether BE features captured in an individual’s AS are more predictive of health than home-centric-defined features is relatively nascent (14, 15, 18, 44, 45, 47-51, 56). In some cases, AS-defined food and structural environment features have been shown to be more predictive of health behaviors than those captured in the home neighborhood. A study examining fast food outlet density and consumption of saturated fats, whole grains, and fruits and vegetables found that the density captured via the AS was positively associated with saturated fat intake and negatively associated with whole grains (18). No such associations were observed using fast food outlet density captured in the 0.5-mile street network buffer (18). However, two studies conducted in New York City produced counterintuitive results, with one finding that higher noise density in AS-defined neighborhoods was associated with lower systolic and diastolic pressure and the other finding that home-centric food environments were more predictive of BMI and blood pressure (48, 49).

These disparate findings may be demonstrative of the utility of AS BE measures in various settings or may highlight differences in the operationalization of the AS techniques. For example, Zenk et al. found that AS defined by daily path areas was predictive of diet but standard deviational ellipse was not (18). In addition, GPS monitoring and travel diaries place a heavy burden on participants and produce high volume data with thousands of data points. These data must then be cleaned and consolidated into useful metrics, for which there are many, with varying predictability based on setting and the health outcome(s) of interest. This varied predictability may also be due, in part, to the timing and duration of BE exposure covered by the AS measures. For example, a given GPS path may be predictive of PA or diet quality for that day or week but may be less predictive of long-term behavior or chronic diseases that are the result of the totality of behaviors over a protracted period of time (e.g., obesity, diabetes).

In order to overcome issues with the timing of BE exposures relative to the distal outcomes of interest, investigators have increasingly turned to EMA to capture real-time, proximal dietary and PA decision-making as well as mood as it occurs in a real-world setting (63, 91-95). EMA is able to capture dynamic behaviors as they occur throughout the day and can allow investigators to observe the array of risk factors that precipitate a change in behavior that is spatiotemporally linked to the GPS-derived area measures (63, 91-95). This method of data collection often uses short message services or applications via mobile devices and therefore it can often be coupled with GPS and accelerometer data. EMA represents an exciting and rapidly growing field of exposure and outcome assessment that is readily applicable to the study of the BE and health and that complements new and evolving GPS-derived metrics (63, 91-95).

Energy balance pathways

Any observable relationship between the BE and obesity should first, logically, have some observable effect on upstream energy balance behaviors, such as diet and PA (Figure 2) (11). Several studies have examined the role that the PA environment plays on PA and use of active modes of transport to commute to work or school. One feature of the PA environment that has been consistently linked to increased PA (specifically, walking) is neighborhood walkability (23-26). Walkability is typically defined by land use mix, residential density, retail floor area density, and street connectivity (27, 28). With respect to the food environment, several studies have examined whether the composition of healthy and unhealthy food outlets, often referred to as the Retail Food Environment Index, in the home neighborhood was associated with diet quality. Several studies have observed that proximity to supermarkets, grocery stores, and sit-down restaurants is associated with high diet quality (37, 38, 96-98), while proximity to convenience stores and fast food restaurants has been associated with poor-quality diets (2, 20, 99). However, as observed with obesity and diabetes, the relationship between the home food environment and diet quality has not always been consistent (2).

While there are several studies that have evaluated the effect of BE exposure on health and health behaviors separately, there are far fewer that have formally evaluated the potential mediating role of health behaviors in the BE–health relationship. International research examining the interplay among the BE, health behaviors, and health has provided evidence that PA mediates the effect of the PA environment on health (100-103). A study of Belgian adults examined the potential mediating effect of accelerometer-assessed PA and sedentary behaviors on the relationship between walkability and adiposity (100). They found that moderate-to-vigorous PA as well as walking and cycling for transport significantly mediated the relationship between walkability and both BMI and waist to height ratio; however, no such mediating relationship was found for sedentary behaviors (100). A study of New Zealand adults found that PA and sedentary behaviors played a significant mediating role on the effect of street connectivity and neighborhood destination accessibility on BMI and waist circumference (101). A third study found that light and moderate-to-vigorous PA mediated the relationship between elements of the urban walking environment and BMI among a group of Japanese adults aged 65 to 84 years (102). With respect to cardiometabolic outcomes, a study using a representative sample of adults in South Australia found that the effect of walkability and road network buffers on hemoglobin A_1c was significantly mediated by self-reported PA levels (103). However, it is worth noting that, since the walkability index has been derived from elements in the environment that relate to walking, it should be expected that, by design, walking and walkability will be correlated. Therefore, the observed associations between walkability and PA levels may represent a form of autocorrelation. To our knowledge there are few, if any, examples of formal mediational analyses of PA in the relationship between BE and obesity, diabetes, or metabolic syndrome in US-based adult populations. Moreover, there are no equivalent studies that have examined the mediating role of diet, food shopping, or cooking behaviors on the BE–health relationship either in the US or internationally. However, such studies evaluating mediational effects of dietary behaviors may not prove fruitful if BE exposure assessment relies on home-centric measures alone.

Early evaluations of the neighborhood BE assumed that, much like the PA environment, the health-relevant food environment comprised the food stores and restaurants around the home (2, 4, 8, 13, 19, 57). Yet the current body of evidence evaluating the food environment and health shows that this was an overly simplistic model (2, 4, 8, 13, 19, 36, 57). It is now understood that the built elements of the food environment constitute only a small slice of the wider context of the food environment. Other factors at the individual level (e.g., socioeconomic status, cultural and religious norms, perceptions, attitudes); at the store and restaurant level (e.g., affordability, availability, convenience, marketing); and even at the local, state, and federal levels (e.g., social, income, and food policies) are more predictive of where individuals shop (19, 36, 104-108). Many studies fail to account for the diet cost at various food outlets as well as differences in individual purchasing power. Moreover, home-centric measurement fails to capture meals that may be consumed in the vicinity of the workplace or school, leading to incomplete assessment of the food environment to which an individual is exposed. Therefore, more thorough investigations of the relationship between the food environment and health should include diet and diet cost in their analyses as well as extend capture of food environment exposures to where people live, work, and spend their leisure time.

Psychosocial pathways

Individual-level psychosocial factors such as perceptions (e.g., neighborhood attractiveness, availability of fresh produce), attitudes, social cohesion, and self-rated health and wellness can aid researchers in elucidating how individuals interact with and internalize their environment (Figure 3) (2, 109-115). Indeed, where objective, GIS-based assessments of the BE have failed to produce consistently significant associations with health outcomes, such as obesity and diabetes, perceived metrics such as the availability of supermarkets have been shown to be far more predictive (2). The collection of survey-based psychosocial measures may also assist in accurately capturing individual-level utilization of BE resources better than objective, GIS-based measures alone (2, 109).

Individual perceptions of the BE have been shown to be correlated with objective measures of the BE (109, 116) as well as measures of socioeconomic status (117), and they are predictive of health (2, 109-115, 118). One study found that a four-item food environment assessment tool, which measured an individual’s perceptions of healthy food access, matched very closely with actual access in a low-income population (116). However, the degree to which perceived and objective measures of the BE match up may vary by population in a systematic way (109). Caspi and colleagues found that mismatch between perceived and objective BE measures was high but also observed that perceived supermarket access was more predictive of fruit and vegetable consumption than objective measures such as supermarket proximity (109). A similar relationship has been observed with perceived and objective measures of the structural environment and levels of PA, with the perceived presence of facilities, sidewalks, shops, and area traffic each showing a positive association with level of PA (113). Likewise, other self-rated measures of BE, such as attractiveness, noise, and crime, were associated with individual property value, which in turn correlated highly to neighborhoods with a high prevalence of obesity (117).

Positive attitudes, self-efficacy, and social supports have been demonstrated to be positively associated with healthy eating and higher levels of PA (55, 77, 119-126). One study found that participants who strongly agreed that a healthy diet was important had a healthier diet overall compared with those participants who disagreed or strongly disagreed (127). This attitude–diet relationship was found to be present regardless of the average cost of a healthy diet at the supermarket (127). Studies examining psychosocial factors predictive of PA levels have found that self-efficacy and social supports are associated with greater levels of activity (55, 119, 125, 126). Interestingly, studies that have jointly examined both perceived and objective measures of the BE and psychosocial factors have found a synergistic relationship between the BE, self-efficacy, and social supports (55, 119, 125, 126). Those individuals possessing positive psychosocial characteristics and living in environments conducive to PA have shown the greatest levels of PA or participation in active transport to work or school (55, 119, 125, 126).

While psychological factors have been shown to be related to both BE features as well as health, more work is needed to understand how the perceived environment and positive attitudes mediate the effects of the BE on health. Furthermore, there is limited evidence evaluating the role that mental health and well-being play in mediating the effects of the BE on health. One study found that the presence of depressive symptoms changed both the strength and the direction of the association between the perceived structural environment and PA levels (128). To our knowledge, there are no studies that have examined how the effect of the food environment on diet quality and health outcomes is mediated by individual perceptions, attitudes, and mental health. Such studies should use dynamic, person-centric measures of the food environment to capture the full extent of exposure at home, work, school, and during commutes as well additional contextual factors such as prices, policies, marketing, and cultural and religious norms.

EMA, now viewed as a new and promising behavioral outcome of BE exposure (63), is concordant in space and time with spatiotemporally driven GPS measures of the BE exposome such as AS metrics. EMA has helped to shift attention from dietary quality or body weight to momentary, place-driven impulses, behavioral triggers, or environmental cues. These triggers may be predictive of longer-term health outcomes, such as obesity and diabetes. While the EMA may provide answers to the relationship between BE and risk factors for behavior change, such methods may have less utility in providing insights into how the neighborhood BE or larger BE exposome influences chronic diseases with protracted onset periods. Such long-term chronic diseases have etiologies that are highly complex and that result from the accumulation of exposures to the BE as well as socioeconomic status, behavior, and more (129).

Residential selection bias

Underpinning many residential BE exposures are the socioeconomic and sociodemographic factors that determine where individuals are able to live as well as whether and where they are able to relocate (19, 74, 130). Historical patterns of residential segregation, a prominent manifestation of structural racism, have dramatically shaped the human geography of much of the world and have been linked to health disparities (131, 132). Relatedly, inequities in educational and work opportunities as well as intergenerational accumulation of wealth also play a key role in where individuals locate as well as how they interact with the BE (133). Two individuals may occupy the same physical residential space or AS but experience very different social or economic environments (133). Therefore, careful confounding control is needed as there are several socioeconomic and sociodemographic factors that are associated with both where a person lives and the risk of obesity, which, in turn, may vary based on the shape and scale of the defined area (74, 130). Interestingly, the incongruence in area-level socioeconomic advantage or disadvantage between residential BE versus AS BE and its relationship to health is currently an active area of research that, to date, has produced mixed findings (91, 134).

It is also worth noting that much of the extant research into the relationship between the BE and health has been conducted in the US and other high-income countries (135). The investigations into whether such findings are applicable to low- and middle-income countries are still in their infancy (135). As with high-income countries, care must be taken in such evaluations to consider the larger social and political context behind where people live. Moreover, the development of new technologies to assess the BE should be flexible and adaptive enough for use in both settings.