Obesogenic environments and cardiovascular disease: a path analysis using US nationally representative data

CVD has been the leading cause of death in the United States in recent decades, with tremendous health and economic burdens in the US and globally [6]. The most common types of CVD are coronary heart disease, heart failure, arrhythmia, and stroke [6]. CVD is caused by a combination of genetic, lifestyle, and environmental risk factors [6]. Most of these risk factors are modifiable. For example, CVD could be largely prevented by maintaining a healthy lifestyle, including not smoking, abstaining from or consuming only moderate amounts of alcohol, eating a healthy diet, and engaging regularly in moderate physical activity [10]. In addition, many of these health behaviors co-occur [7]. A US survey reported that 52% of participants had two or more health-compromising behaviors, including physical inactivity, smoking, heavy drinking, and having overweight [11]. Díaz-Gutiérrez and colleagues quantified the co-occurrence of these health-compromising behaviors (including smoking, physical activity, diet, alcohol intake, television exposure, and nap habit), and found that individuals who engaged in fewer health-promoting behaviors have higher CVD risk [5].

In addition to behaviors, some physiological indicators are associated with elevated risk of CVD. Multi-system physiological dysregulation is a sub-clinical state that occurs when long-term stressors accumulate over the life course and disrupt the regulation of multiple physiological systems [12]. The indicators for physiological dysregulation—including BMI, blood pressure, cholesterol, Hemoglobin A1c (HbA1c), and C-reactive Protein (CRP)—may predict future risk of CVD and poor health [13, 14]. For example, among older adults, higher total cholesterol, blood pressure, or BMI is associated with higher CVD risk [15]. As with health behaviors, biological CVD risk factors tend to co-occur [9]. A retrospective study of older adults found that 34% of participants presented with three or more biological CVD risk factors [8]. As the number of biological risk factors with values in risky ranges increases, so does CVD risk [14].

The risk of CVD is influenced not only by individual-level risk factors, but also by the broader environment in which individuals live. Obesogenic environments, defined as “the sum of influences that surroundings, opportunities, or conditions of life have on promoting obesity in individuals or a population [16],” are associated with increased risk of CVD among the residents [17]. For example, people living in neighborhoods with reduced access to grocery stores and supermarkets [1], and a higher density of fast food restaurants [1, 2, 18, 19] and convenience stores [20], have a higher risk of myocardial infarction, coronary heart disease, congestive heart failure, and stroke. Additionally, reduced access to parks and recreation facilities increases the odds of these same conditions [1, 18, 20].

It appears that obesogenic environments may increase CVD risk through their associations with health-compromising behaviors and physiological dysregulation. First, obesogenic environments are linked to health-related behaviors such as diet and physical activity [2]. These health behaviors are, in turn, associated with CVD risk [5]. For example, access to fast food restaurants and convenience stores is associated with excessive consumption of sugar-sweetened beverages and insufficient consumption of whole grains and vegetables [4, 21]. Conversely, closer proximity to supermarkets is linked to lower fat consumption [3] and greater consumption of fruits and vegetables [22]. Moreover, limited access to recreation facilities such as gyms and parks is associated with physical inactivity [23].

Furthermore, living in economically disadvantaged neighborhoods is associated with limited healthful food and physical activity resources which is, in turn, associated with increased alcohol and cigarette consumption [24]. Well-established conceptual frameworks assist with understanding this association. First, people living in economically disadvantaged neighborhoods have a higher risk of psychological distress [25]. Second, suggested by the tension reduction hypothesis, alcohol and cigarette consumption help people reduce stress [26]. It is therefore plausible that individuals living in obesogenic environments consume alcohol and tobacco products both because of their relative availability and as a means to relieve stress, and these compromising behaviors are associated with increased risk for CVD [6].

Additionally, residents living in obesogenic environments exhibit elevated physiological dysregulation [2]. For example, people living in neighborhoods with greater access to convenience stores or fast food restaurants, and those with limited access to grocery stores or farmers markets have higher BMI [2, 27], greater hypertension risk [2], higher HbA1c, and increased blood glucose levels, which indicate increased risk of Type 2 diabetes [2, 28]. Proximity to supermarkets is associated with lower prevalence of obesity and overweight; conversely, the presence of convenience stores is associated with a higher prevalence of overweight and obesity [29]. Moreover, environments with fewer facilities for physical fitness are associated with higher BMI and higher prevalence of hypertension [30]. Thus, the association between environmental obesogenicity and physiological dysregulation may be due to engaging in health-compromising behaviors, such as maintaining an unhealthy diet [4].

The complex interrelationships among individual-level and environmental-level risk factors and CVD suggest at least two potential pathways linking environmental obesogenicity and CVD [31]. The behavioral pathway suggests that limited access to healthy foods and facilities for physical fitness predict poor diet [32] and inadequate physical activity [23]. The physiological pathway suggests that living in obesogenic environments may contribute to the gradual deterioration of human physiological regulatory systems, perhaps through stress-related or psychosocial processes [12]. An alternative plausible conceptualization is serial mediation, in which the accumulation of health-compromising behaviors predict the early precursors of physiological dysregulation, which in turn, relates to cardiovascular health [4]. Combined, these behavioral and physiological processes may increase residents’ risk for CVD.

Despite evidence that various aspects of the environment are related to health behaviors and physiological functioning, few studies have investigated multiple behavioral and physiological pathways through which residential environments are linked to residents’ cardiovascular health. This study uses a representative sample of US adults aged 51 and over, a group of individuals who may be less mobile than younger adults and rely more heavily on neighborhood resources [33], to test whether the association between environmental obesogenicity and CVD can be explained by health-compromising behaviors or multi-system physiological dysregulation as parallel pathways, or if these factors operate as serial mediators (see Fig. 1). We hypothesized that the association between environmental obesogenicity and CVD was partially explained by 1) health-compromising behaviors, 2) physiological dysregulation, both individually and 3) in serial order (environment health behaviors physiological dysregulation CVD).

To test the hypotheses, we conducted a cross-sectional observational study using data from the Health and Retirement Study (HRS), an ongoing longitudinal survey of more than 20,000 US older adults [34]. The target population for HRS includes all older adults (≥ 51) living in the contiguous United States (including 48 States and the District of Columbia). The HRS used multi-stage area probability sample design to recruit participants to ensure the representativeness of the sample [35]. In 1992, HRS began collecting data on health conditions, health behaviors, and socioeconomic circumstances associated with aging. Since HRS inception, refresher cohorts have been recruited to include representative samples of the Asset and Health Dynamics (1993), Children of the Depression (1998), War Babies (1998), Early Baby Boomers (2004), Mid Baby Boomers (2010), and Late Baby Boomers (2016) [34]. After the initial 1992 interview, the participants were interviewed every two years. The response rate of each wave was between 80%—90% [36]. Since 2006, HRS has expanded data collection to include blood, saliva, and anthropometric measures as a part of enhanced face-to-face (EFTF) interviews [36]. Half of the sample was randomly selected to complete the EFTF in 2006, and the other half completed it in 2008. Both half-samples repeat the EFTF every four years. In 2013, HRS conducted a Health Care and Nutrition Mail Study (HCNS) in which HRS researchers mailed the Harvard food frequency questionnaire (FFQ) to a subsample of HRS respondents. HRS health records were linked to a Contextual Data Resource (CDR) including the USDA Food Environment Atlas [37], the US Uniform Crime Reporting Program [38], and the American Community Survey data sets via geographic identifiers [39].

The analytical sample of the current study included participants from all 48 contiguous states and the District of Columbia residing in 675 counties. The characteristics of the full study sample and subsamples stratified by tertiles of environmental obesogenicity are displayed in Table 4.

All analyses were conducted in SAS 9.4 using the Hayes Process Macro V2.13, a path analysis modeling tool that uses ordinary least squares (OLS) and logistic regressions for estimating direct and indirect effects in parallel and serial mediation models [48]. We used Process Macro Model 6 to conduct parallel and serial mediation analyses to test the extent to which health-compromising behaviors, physiological dysregulation, or both in serial order (environment → health behaviors → physiological dysregulation → CVD) explain the hypothesized relationship between environmental obesogenicity and CVD. The current study examined serial mediation, in which health-compromising behaviors preceded physiological dysfunction. We also examined a reverse mediation pathway (obesogenicity → dysfunction → behaviors → CVD) since people with physiological dysregulation may tend to be more physically inactive (see Supplemental Table 3). Results indicated that compared to parallel mediation pathways, both serial mediation pathways explained a negligible portion of the total effects (< 0.3%). As such, and because individuals' behaviors generally don't change substantially after physiological complications [49], we only report results of our originally hypothesized model in which health-compromising behaviors preceded physiological dysregulation.

Individual-level age, sex, education, race/ethnicity, and county-level education and population density were included as controls. Continuous variables were standardized prior to mediation analysis. Coefficients presented in results are therefore partially standardized regression coefficients (B). Observations with any missing data in analytical variables were excluded from the analyses, leaving a final sample size of 12,482. To avoid temporal confounding, we employed a fully lagged mediation design, i.e., the independent variable neighborhood obesogenicity contained indices available between 2009 and 2012; the outcome variable CVD was assessed in 2016; the two mediators – health-compromising behaviors and physiological dysregulation – were measured in 2012–2013 and 2014–2016, respectively. As such, we prioritized a temporal schema whereby the predictor preceded the mediators, which in turn, preceded the outcome. In other words, contemporaneous effects between mediators and outcomes were not permitted. Data sources and available years of each indicator are presented in Supplemental Table 2.

Consistent with previous studies [1, 2, 18, 20], we found that people living in more obesogenic environments had greater odds of reporting CVD. However, we went beyond existing research by assessing environmental obesogenicity using nine indicators to capture the multidimensional nature and complexity of peoples’ environments. Combined, this collection of environmental indicators helped to partially explain the commonly co-occurring behavioral and physiological risk factors observed among residents of these environments.

Moreover, we extended the literature by demonstrating that both health behaviors and physiological dysregulation are implicated in this association. Our study examined a wider array of variables capturing a multitude of health-compromising behaviors and aspects of physiological dysregulation to better represent these commonly comorbid factors [7‐9], whereas most prior studies typically included a single health-related behavior [31] or one biomarker [50]. We observed that residents of obesogenic environments engaged in more health-compromising behaviors [2, 4, 22, 23, 29, 51] and had greater physiological dysregulation [2, 3, 27, 28, 30] than those living in less obesogenic environments. Both behaviors and physiological dysregulation, in turn, were associated with CVD. While most previous studies did not formally test indirect effects and did not assess whether a significant proportion of variance was explained [2, 31], this study used path analysis to estimate the total, direct, and indirect effects through each pathway and the significance of each. We found that health-compromising behaviors and physiological dysregulation each explained similar proportions of the total effect (3.3% vs 3.3%), indicating that both health-compromising behaviors and physiological dysregulation partially explain the obesogenicity-CVD link. This was comparable to the study by Yang et al., which found that physiological risk factors explained between 3 and 11% (depending on the physiological measure) of the link between residential greenness and CVD risk [50]. Although our study examined obesogenicity rather than greenness, our findings are generally consistent with those of Yang et al., that physiological risk factors explained a small portion of the environment-CVD link. It is plausible that other potential pathways, though not the focus of the current study, may also explain the obesogenicity-CVD relation, such as stress [25]. The purpose of the current study was to investigate specific mediators that would inform the development of targeted interventions to reduce CVD, but not to identify a model that completely explains the obesogenic environment – CVD relationship. Future research should investigate other potential pathways that contribute to the relationship.

Finally, results suggested that behavioral and physiological factors each partially, both in tandem and – to a much smaller degree – in serial order, explained the significant association between environmental obesogenicity and CVD. This finding made a major contribution to the literature by enhancing scholars’ understanding of the conceptual links between these factors. Although conventional logic suggested that behaviors led to physiological dysregulation which affected health, our findings showed that this serial mediation conceptualization was not a substantially better model for these associations. That is, physiological dysregulation may also operate as a separate mechanism from behaviors (potentially through other pathways such as stress). It is plausible that the time lag between our measures is not sufficiently long enough to capture physiological changes resulting from behaviors, and future research should continue to compare alternative conceptualizations of these associations, as well as additional pathways linking environment and CVD.

Results should be interpreted with several limitations in mind. First, although we modeled the environmental obesogenicity measure after the COEI [40], data on walkability and access to parks which were included by COEI, are not available in the HRS. However, based on the USDA Food Access Research Atlas [37], our measure of environmental obesogenicity included two additional variables: county-level median household income and census tract-level proportion of residents with low-access to healthful food stores, which combined urban/rural-adjusted distance to the closest healthful food venue and average vehicle access. The former was an indicator of area-level food and physical activity resources, and the latter was used to define low access to healthy food outlets by the USDA Food Access Research Atlas [37]. The present study extended COEI to a representative older population with a wider array of mediators and outcomes. Second, the dietary information available in HRS came from a non-representative subsample of participants in the HRS HCNS. In our final analytical sample, 5,557 people had dietary information, so the health-compromising behavioral index was assessed without diet information from many participants in this study (summed across smoking, drinking, and physical activity). Third, to test the serial mediation pathway, we employed a fully lagged mediation design, i.e., the measure of environmental obesogenicity preceded two mediators, which preceded the measure of CVD (see Supplemental Table 2 for a description of data availability across HRS waves of data collection). This lagged mediation design, however, introduced a potential bias related to residential mobility between the long study period. However, based on the results of the sensitivity analysis, which only included people who did not move during the study period (n = 7,746), there was still a significant association between environmental obesogenicity and CVD, and significant indirect effects through health-compromising behaviors and physiological dysregulation, combined. Furthermore, CVD and health-compromising behaviors were self-reported in the present study, inviting bias related to several factors including memory, social desirability, and perhaps most importantly, variability in access to regular care. However, we believe that concern regarding self-report bias is somewhat minimized given that the current study found a significant association between the objectively measured physiological dysregulation composite variable and self-reported CVD. Indeed, this finding supports the future use of self-reported CVD data in instances when more objective physician reports are unavailable. Moreover, we found a modest effect of neighborhood environment on health, consistent with research showing that neighborhood effects on health are typically small [52]. Last, study variables used in the present study were at multiple levels, i.e., predictors were at county/tract level, mediators and outcomes were at individual level, and covariates were at both levels. Including variables at different levels of aggregation may lead to the modifiable areal unit problem [53], which may introduce challenges to interpretation of the findings. Despite these limitations, this study went beyond existing research in the robustness of measures and statistical methods used, providing valuable insight into the roles of health behaviors and physiological dysregulation in the obesogenicity-CVD association.