Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2025 Jun 18.
Published in final edited form as: J Behav Med. 2024 Aug 7;47(6):965–979. doi: 10.1007/s10865-024-00505-2

The moderating effect of access to food facilities and recreational activity space on mHealth multiple health behavior change intervention

Samuel L Battalio 1, Benjamin W Barrett 1, Ivelina I Arnaoudova 1, David J Press 1,2, Donald Hedeker 3, Angela Fidler Pfammatter 4, Kiarri N Kershaw 1, Bonnie Spring 1
PMCID: PMC12175576  NIHMSID: NIHMS2087518  PMID: 39110353

Abstract

Objective

To evaluate whether the neighborhood social and built environment moderates response to a mobile health multiple health behavior change intervention targeting fruit/vegetable intake, sedentary behavior, and physical activity.

Methods

Participants were 156 Chicago-residing adults with unhealthy lifestyle behaviors. Using linear mixed models, we evaluated whether access to food facilities (fast food restaurants and grocery stores) and recreational activity spaces (gyms and parks) moderated the difference in behavior change between the active intervention condition relative to control. Using spatial data analysis (cross K functions), we also assessed whether participants who achieved goal levels of behaviors (“responders”) were more or less likely than those who did not achieve intervention goals (“non-responders”) to reside near fast food restaurants, grocery stores, gyms, or parks.

Results

According to linear mixed models, none of the neighborhood social and built environment factors moderated the difference in behavior change between the active intervention condition and the control condition (Likelihood Ratio (χ2[1] = 0.02–2.33, P-values > 0.05). Cross K functions showed that diet behavior change responders were more likely than non-responders to reside near fast food restaurants, but not grocery stores. The results for activity behavior change were more variable. Sedentary screen time responders were more likely to reside around recreational activity spaces than non-responders. Moderate-vigorous physical activity responders had greater and lesser clustering than non-responders around parks, dependent upon distance from the park to participant residence.

Conclusions

A complex relationship was observed between residential proximity to Chicago facilities and response to multiple health behavior change intervention. Replication across diverse geographic settings and samples is necessary.

Keywords: Health disparities, mHealth, Health Behavior Change, Physical activity, Diet

Introduction

Poor-quality diet, physical inactivity, and high sedentary screen time are among the most prevalent and impactful lifestyle factors implicated in morbidity and mortality in the United States (US) (Mokdad et al., 2004; Sotos-Prieto et al., 2017; Stamatakis et al., 2019). These health risk behaviors are more prevalent in low socioeconomic and racial and ethnic minoritized populations, exacerbating health disparities (Braveman et al., 2011; Hawes et al., 2019; Satia, 2009). Various multilevel social determinants of health (SDOH) cause health disparities by, in part, facilitating health risk behaviors and impeding health promoting behaviors (Darmon & Drewnowski, 2008; Hawes et al., 2019; Satia, 2009). Researchers have made significant progress in understanding the impact of multilevel SDOH on health behaviors in non-interventional contexts. However, there is a critical need to apply modern SDOH frameworks to develop and enhance health behavior change interventions.

Empirically- and theoretically-supported SDOH frameworks, such as those proposed by the National Institute on Minority Health and Health Disparities (Alvidrez et al., 2019) and the World Health Organization (WHO, 2010), emphasize various domains of SDOH that act across multiple levels of influence. SDOH frameworks encourage consideration of systemic, environmental factors that act “upstream” to influence behavior. In particular, they highlight features of the neighborhood social and built environment that can act as barriers or facilitators to health behavior enactment (Alvidrez et al., 2019; Calloway et al., 2019; Hawes et al., 2019).

A feature of the neighborhood social and built environment that has been shown to influence health behavior is degree of access to relevant resources and facilities. Two domains of access have been particularly well-documented in observational studies to have associations with health behavior: (1) food access is associated with diet-related lifestyle behaviors, and (2) access to recreational activity space is associated with activity-related lifestyle behaviors (Ghosh-Dastidar et al., 2017; Zenk et al., 2019). To date, food access has been conceptualized in terms of ability to readily obtain unhealthy fast foods and to access grocery stores containing affordable, healthy foods. Although complex and mixed, evidence suggests that living in neighborhoods with less access to grocery stores and greater access to fast foods is associated with greater odds of eating an unhealthy diet and increased morbidity and mortality (Ghosh-Dastidar et al., 2017; Zenk et al., 2019). Regarding recreational activity space, research has shown that access to facilities that are conducive to physical activity, primarily gyms and parks, is associated cross-sectionally with greater levels of physical activity and reduced sedentary time, as well as a reduced rate of decline in physical activity over time (Christian et al., 2017; Ranchod et al., 2014; Smith et al., 2017).

Although it is well-established that limited access to environmental resources is associated with problematic health behaviors, there is a notable gap in research regarding whether and how access to these resources affects the effectiveness of health promotion interventions in promoting positive lifestyle changes. To fill this gap, researchers have recently proposed and begun testing conceptual models of the pathways by which social and built environment factors might interact to moderate intervention-facilitated health behavior change (McCormack et al., 2022). These conceptual models draw from and expand upon existing frameworks, such as Grossman’s Model of Health Promotion (Grossman, 2017), epigenetic models of gene-environment interaction (Ottman, 1996), and ecological models (Catford et al., 2022). Although the models propose slightly different names for each pathway, they are functionally consistent with the following: (1) Blunting Pathway, through which less supportive environments blunt an individual’s ability to benefit from the health behavior intervention, (2) Compensation Pathway, through which the intervention helps to compensate for a less supportive environment to produce disproportionally greater intervention-facilitated behavioral improvement relative to those in supportive environments, or (3) Invariant Pathway, through which the environment and intervention-facilitated behavior change do not meaningfully interact. Precise theoretical mechanisms underlying each of these pathways have not yet been articulated. Rather, they provide a conceptual framework for describing how interventions may or may not interact with health promotion interventions. To date, empirical examinations of each of these pathways have produced mixed results, and no pathway has sufficient evidence to stand out as a frontrunner (McCormack et al., 2022).

Most of the research in this content area has used observational data whereby relationships between access to environmental resources and individual health behaviors have been examined outside of the context of an experimental manipulation (Caspi et al., 2012; Raskind et al., 2020; Smith et al., 2019). Clearly, except for natural experiments (MacMillan et al., 2018), it is challenging to manipulate environmental factors in clinical trials. An important intermediate step, however, is to conduct secondary analyses of clinical trials to examine theoretical pathways by which the neighborhood environment may influence intervention facilitated lifestyle changes at the individual level. Though not a clinical trial, one secondary analysis has examined the moderating effect of environmental resource access on individual level, intervention-facilitated health behavior change. This study leveraged electronic health record data to examine whether Department of Veterans Affairs MOVE! Weight management program to examine whether access to recreational activity spaces and food resources significantly moderated changes in BMI for Veterans with obesity (Zenk et al., 2019). Consistent with the invariant pathway, they found that neither class of environmental resource significantly moderated changes in BMI (Zenk et al., 2019). However, to our knowledge, no studies have examined the moderating effect of resources access for individual-level health behavior change interventions targeting public health guideline adherence of diet and activity behaviors delivered remotely using mHealth tools and telecoaching.

Based on these considerations, we evaluated the degree to which access to specific neighborhood facilities (fast food/grocery stores and gyms/parks) moderated the response to an mHealth multiple health behavior change intervention targeting diet, moderate-vigorous physical activity (MVPA), and sedentary behavior. Under the umbrella of this parent aim, we examined two related yet distinct questions: (1) ‘To what extent does distance of residence from neighborhood facilities moderate the difference in target behavior change between a diet and activity health behavior change intervention condition vs an active control condition?’, and (2) ‘Among recipients of a diet and activity health behavior change intervention, are responders more or less likely than non-responders to reside near relevant neighborhood facilities?’. We hypothesized, based on prior research, that access to food facilities (fast food restaurants and grocery stores) would moderate changes in diet behaviors (consumption of [1] fruits and vegetables and [2] saturated fat), and access to recreational activity spaces (gyms and parks) would moderate changes in activity behaviors (quantity of [1] sedentary leisure screen time and [2] MVPA). Informed by the behavior-environment pathways, we tested two alternative hypotheses about moderating effects: (1) blunting hypothesis (corresponding to the bunting pathway) or (2) compensation hypothesis (corresponding to the compensation pathway). The blunting hypothesis suggests that living in a less supportive environment blunts an individual’s ability to benefit from the health behavior intervention. Accordingly, we would expect participants with less supportive environments to respond less favorably to the intervention than those living in a more supportive environment. The compensation hypothesis suggests that the intervention helps to compensate for a less supportive environment to produce disproportionally greater intervention-facilitated behavioral improvement relative to those in supportive environments. Accordingly, we would expect participants living in a less supportive environment to benefit more (respond more favorably) from the intervention than those living in a more supportive environment.

Methods

The current study used data collected from the Make Better Choices 2 trial (Pellegrini et al., 2015; Spring et al., 2018), a three-arm randomized-controlled trial conducted from 2012 to 2014 that evaluated a remotely-delivered, technology-supported multiple health behavior change intervention targeting improvements in diet, physical activity, and sedentary screen time. The trial’s methods have been described in detail elsewhere (Pellegrini et al., 2015; Spring et al., 2018).

Briefly, three interventions used remote-coaching and a mobile smartphone application to target multiple health risk behaviors. Two aimed to foster healthy increases in fruit and vegetable intake and MVPA, and to reduce sedentary leisure screen time. One intervention simultaneously targeted all behaviors from intervention onset (Simultaneous), while the other targeted diet and sedentary screen time first, and subsequently incorporated physical activity (Sequential). The third intervention, an active control condition, targeted decreasing stress and increasing sleep (Control). All participants were provided with a smartphone application that enabled them to self-monitor their targeted behaviors relative to daily goals; their behavioral data were transmitted to a coach who used the data to personalize telephone-delivered behavioral advice. Health behavior goals were systematically tapered up during an initial 3-month period, after which point participants were instructed to maintain health behavior changes throughout a follow-up period which lasted until 9 months.

(Note that, in the interventions targeting improvements in diet and physical activity, participants were given explicit goals for fruits and vegetables, sedentary screen time, and physical activity. No explicit goals were assigned for saturated fat because of prior evidence showing that the desired reduction in fat intake is achieved incidentally as decreasing hand-to-mouth snacking complements the reduction in recreational screen time (with which it would previously have been paired) and as increasing high-fiber fruits and vegetables substitutes for and crowds out higher fat foods (Spring et al., 2012).

Since the primary results from the trial indicated that the Simultaneous and Sequential interventions produced statistically equivalent improvements in each health behavior domain (Spring et al., 2018), we treated these as a single condition in the current study. Participants were intervened upon for a 12-week period, during which time behavioral goals were systematically tapered up.

Participants

Participants were Chicago-area adults with unhealthy lifestyle behaviors who sought enrollment in a trial of remotely delivered, technology-supported health promotion intervention. Specific inclusion criteria were: (1) between 18 and 65 years old, (2) eating < 5 servings of fruits and vegetables per day, (3) eating ≥ 8% daily calories from saturated fat, (4) engaging in < 150 min of MVPA per week, and (5) engaging in > 120 min per week of leisure screen time (e.g., television, movies, videogames, recreational internet). To measure neighborhood characteristics within Chicago, we excluded participants from the original sample who lived outside of Chicago city limits. As such, the current analytic sample comprised 156 out of the original 212 participants that were included in the primary outcome paper’s analyses (74.5%).

Measures

Individual sociodemographic characteristics.

Participants completed a sociodemographic questionnaire at baseline. Specifically, participants self-reported: (1) sex (male or female), (2) age, (3) race (White/Caucasian, Black/African American, Native American, Asian, Other, More than one) and ethnicity (Hispanic/Latino, Not Hispanic/Latino), (4) estimated annual household income (reported in brackets of $10k, starting at <$15k and ending at >$75k), and (5) highest level of education (some high school, high school degree or G.E.D., trade school or specialty training, some college, Associate’s degree, Bachelor’s degree, some graduate school, Master’s degree, or professional degree).

Health behavior change.

We computed each behavior change target using the same procedures as were used in the primary outcome publication (Pellegrini et al., 2015; Spring et al., 2018). Participants across all conditions recorded and self-monitored each criterion behavior change target, without receiving feedback, for approximately one week at four assessment time periods: (1) immediately prior to intervention, (2) immediately following the 12-week coaching intervention period, (3) 6-months after the intervention period, and (4) 9-months after the intervention period. Specifically, dietary intake and sedentary leisure screen time were self-reported via the custom smartphone application. MVPA was computed using the output from a wrist-worn accelerometer and supplemented with app self-report in instances where the participant perceived the accelerometer had not accurately captured physical activity (e.g., nonwear during water activity like swimming, or battery failure).

Intervention Responder Status.

Intervention responder status was computed using the recommended goal levels of behavior that were pre-specified by the intervention: (1) ≤ 90 min per day of sedentary screen time, (2) ≥ 5 servings of fruits and vegetables per day, and (3) ≥ 150 min per week of MVPA, and (4) < 8% of calories from saturated fat. In the current study analysis, participants in the active condition who reached these goal levels of behavior by the post-intervention timepoint (3-month) assessment period were considered responders and those who did not were non-responders.

Chicago facilities.

We acquired the locations of Chicago fast food restaurants and gyms in existence between 2012 and 2014 from a search on the Reference USA database of US historical businesses (http://www.referenceusa.com/Home/Home), and manually validated returned results. Data on the location of Chicago grocery stores in 2011 was obtained from open access data made available from a prior study (Kolak et al., 2018). Boundaries of Chicago Park District properties were identified via data from the Chicago Data Portal (https://data.cityofchicago.org/) and are current as of 2016. We placed points around the boundary of each of the 614 Chicago parks at 100 m between points.

Neighborhood Socioeconomic Status (SES).

Neighborhood data on socioeconomic status (SES), was defined based on the census tract of the participant’s home according to boundaries from the 2010 US Decennial Census 2010 (US Census Bureau, 2010). Continuous measures obtained from the below sources were then scaled into quintiles based on the distribution of the Chicago metropolitan area level. For neighborhood SES, we used a publicly available measure based on the 1990 and 2000 U.S. Censuses and the 2008–2012 American Community Survey, which ranges from 0 to 100 with 50 being the national average. The authors, Miles et al., used an unconstrained single factor model according to the Eq. (1[ln{median household income}])+(−1.129[ln{% female-headed households}])+(−1.104[ln{%workers ≥ 16 years who are unemployed}])+(−1.974[ln{% of households in poverty}]) + 0.451([% high school grads but not bachelors holders] + 2[% bachelors holders]) (Miles et al., 2016).

Statistics

First, we evaluated the analytic sample by computing descriptive statistics for individual- and neighborhood-level characteristics.

Linear Mixed Models.

Next, we conducted a series of linear mixed models to evaluate the moderating effect of access to food facilities (fast food restaurants, grocery stores) and recreational activity spaces (gyms, parks) on health behavior improvement during the intervention. Health behavior improvement was evaluated using the same operationalization of the outcome behaviors as was used in the primary outcome publication (Spring et al., 2018). Specifically, outcome behaviors were transformed onto a common Z-score scale (see Spring et al. (2018) for more details), and then change from baseline was computed and used as the criterion in the linear mixed models. Models included a random person-level intercept and slope for the time effect (collection period, coded such that the first follow-up period was 0, the second follow-up period was 1, and the third follow-up period was 2). Fixed effects included the time effect, the condition variable (dummy coded such that the control [stress and sleep] condition was the reference, and the combined active condition was the comparator), the neighborhood SES control, and a main effect for each relevant facility access moderator (fast food and grocery store access for diet outcomes, gym and park access for activity outcomes). For the linear mixed models, facility access was operationalized as the Euclidian distance (in kilometers) to the nearest relevant facility.

To evaluate whether different types of facility access moderated the effect of the behavior change interventions on diet and activity changes, we ran eight separate additional models, one for each permutation of outcome and the relevant facility pair. In each additional model, we added to the base model (described above) an interaction between the facility access moderator and condition. We computed a likelihood ratio test to determine whether the inclusion of the interaction produced a significant improvement in model fit, relative to the base model. The likelihood ratio test corresponding to the inclusion of each facility access moderator, and the estimates corresponding to the interaction term, served as the primary indicators of a moderating effect of the environmental factor on the association between condition and improvement in each criterion behavior domain.

Cross K Functions.

Finally, we assessed residential clustering of participants who underwent the active condition (N = 121) around each Chicago facility (fast food, grocery stores, gyms, and parks) via cross K functions. Cross K functions are a spatial data analysis method used to assess the spatial dependence of data points (Dixon, 2002). Up to a maximum distance of 3 km (Kelman et al., 2019), we assessed the residential clustering between participants and Chicago facilities at various distances from a facility to a participant residential address.

We estimated cross K functions for responders and non-responders (at 12 weeks of follow-up) within each health behavior change goal. These analyses were conducted between relevant facility and behavior change outcome pairs (fast food restaurants and grocery stores for diet outcomes, gyms and parks for activity outcomes). We also evaluated the difference in cross K functions between responders and non-responders for each goal, to determine if, for example, fruit and vegetable consumption responders tended to reside near Chicago grocery stores more than fruit and vegetable consumption non-responders.

To determine if residential clustering between participants and Chicago facilities was meaningful (indicative of spatial dependence), we compared the cross K function results to clustering results that would be expected if participants and Chicago facilities were spatially independent. We used Monte Carlo random labeling (Dixon, 2002), which simulates spatial independence by randomly labeling each point as a responder, non-responder, or Chicago facility (e.g., grocery store), and calculating the (difference in) cross K functions over 100 trials. Ranking each trial and taking the 97.5th and 2.5th percentile (difference in) cross K functions provides upper and lower limits of 95% significance. Thus, if observed (difference in) cross K functions fall outside of these 95% significance boundaries, this indicates a cross K function or difference in cross K functions that significantly deviates from one that might be expected under spatial independence (and therefore the cross K function or difference in cross K functions can be considered statistically significant). For more technical information on the cross K function method, please see the Supplementary Materials.

The results of the difference in cross K functions analyses are presented in Figs. 1 and 2. Within these figures, the x-axis represents the distance in kilometers from Chicago facilities (h within the Supplementary Materials equation). The y-axis represents the value (at various distances from Chicago facilities) of the difference between the cross K function of Chicago facilities and responders, and the cross K function of Chicago facilities and non-responders. Here, the absolute value of the y-axis is less important than the overall pattern of the difference in cross K functions line – whether this line is below or above zero on the y-axis, and whether this line is within or outside of the 95% significance boundaries. The difference in cross K functions line is the solid black line in Figs. 1 and 2 and shows, as a function of distance from Chicago facilities, whether responders tend to cluster around Chicago facilities more so than non-responders. The 95% significance boundaries, estimated via Monte Carlo random labeling and simulation, are represented by the dashed red lines in Figs. 1 and 2. The upper dashed red line is the 97.5th percentile difference in cross K functions from the simulation, and the lower dashed red line is the 2.5th percentile difference in cross K functions from the simulation. Together, these 95% significance boundaries envelope the area on the figures, as a function of distance from Chicago facilities, where a difference in cross K functions might be observed under spatial independence between Chicago facilities and participants. Thus, if the difference in cross K functions line (solid black line) falls outside of these 95% significance boundaries, this indicates significant spatial dependence between Chicago facilities and participants. Finally, the dotted blue line marks a y-axis value of zero and is simply intended as a visual aid.

Fig. 1.

Fig. 1

Open in a new tab

Difference in cross K functions for responders vs. non-responders as defined by accomplishment of diet behavior change goals. Cross K functions are between participants and food facilities

Fig. 2.

Fig. 2

Open in a new tab

Difference in cross K functions for responders vs. non-responders as defined by accomplishment of activity behavior change goals. Cross K functions are between participants and recreational activity space facilities

Linear mixed model analyses were computed in R version 4.1.1, ‘lme4’ package. The (difference in) cross K function analyses were conducted using ArcGIS Pro version 2.7.0 (Redlands, CA: Environmental Systems Research Institute), and R version 4.1.1, ‘spatstat’ package (Baddeley et al., 2015).

Hypothesis Testing.

Support for either alternative hypothesis (e.g., the blunting hypothesis vs. the compensation hypothesis) was examined within the same parent statistical models. For the cross K functions, support for either hypothesis was first determined by whether there appeared to be meaningful differences between degree of spatial clustering for responders vs. non-responders. Then, for instances in which there was determined to be meaningful clustering, the direction of clustering (lesser vs. greater clustering) indicated which of the hypothetical pathways was supported. For example, lesser clustering around grocery stores for intervention non-responders for fruit and vegetable consumption would indicate support for the blunting hypothesis (in other words, less access to grocery stores is associated with not achieving fruit and vegetable consumption goals, thereby reflecting a blunting effect of the benefit of the intervention). Likewise, for linear mixed models, the direction of the slope for a significant interaction effect between intervention condition and facility access measure indicates which hypothetical pathway would be supported. For example, an interaction indicating a significantly lower rate of improvement in fruit and vegetable consumption among individuals in the active intervention condition who lived relatively farther from the nearest grocery store would provide support for the blunting hypothesis (in other words, the benefit of the active intervention condition relative to control is blunted for those who have less access to a beneficial neighborhood resource).

Power.

Given that the moderation analyses were not primary outcomes of the parent trial, and therefore not accounted for in the sample size estimation for this trial, we computed power estimates to identify the size of the moderating effect that would be detectable with the current sample size. We computed Power using G*Power 3.1.9.4 to estimate the power to detect a difference between two slopes for the active and control condition, using α = 0.05 and Power (1 - β) = 0.80 (both of which are standard and conservative estimates). Based on these parameters, we would expect to be able to detect a standardized effect size of approximately Cohen’s D ~ 0.53. In other words, if significant moderating effects did exist, and we did not detect them, those moderation effect sizes would likely fall below the effect size of Cohen’s D ~ 0.53. It is important to note that moderation effects are typically small, and so a lack of finding could well reflect a lack of power.

Results

Participant characteristics

Participant characteristics are reported in Table 1. The sample consisted of 156 middle-aged adults (mean age 39.38 ± 12.32yrs) of primarily female sex (75%). Most participants identified as Black (n = 83; 53%), and 35% (n = 56) identified as White. In terms of ethnicity, 17 (11%) identified as Hispanic/Latino. The sample was also moderately highly educated, as 59% of participants reported having obtained a college degree or higher. 41% reported an annual household income of $45k or greater, suggesting that approximately half the sample exceeded the Chicago population median household income. In terms of neighborhood socioeconomic status (nSES), the mean of the current sample was 56.6 ± 19.0, 6.6 points above the national average according to the normative reference data.

Table 1.

Baseline demographics

Variable Total (n = 156) Active (n = 121) Control (n = 35)
Age (years), mean (SD) 39.3 (12.3) 38.8 (12.6) 41.2 (11.1)
Body mass index (kg/m2), mean (SD) 34.4 (9.0) 33.7 (8.6) 36.5 (10.0)
Sex, n (%)
Male 39 (25.0) 29 (18.6) 10 (6.4)
Female 117 (75.0) 92 (59.0) 25 (16.0)
Race, n (%)
Black 83 (53.2) 64 (41.0) 19 (12.2)
White 54 (34.6) 41 (26.3) 13 (8.3)
Asian 4 (2.6) 3 (1.9) 1 (0.6)
Other or multiple 15 (9.6) 13 (8.3) 2 (1.3)
Ethnicity, n (%)
Hispanic/Latino 16 (10.3) 12 (7.7) 4 (2.6)
Not Hispanic/Latino 134 (85.9) 104 (66.7) 30 (19.2)
Education, n (%)
College degree 92 (59.0) 69 (44.2) 23 (14.7)
No college degree 64 (41.0) 52 (33.3) 12 (7.7)
Annual Household Income, n (%)
$45,000 + 63 (40.4) 46 (29.5) 17 (10.9)
< $45,000 93 (59.6) 75 (48.1) 18 (11.5)
Neighborhood Socioeconomic Status (nSES), mean (SD) 56.6 (19.0) 55.5 (20.8) 57.0 (18.6)
Distance to Nearest Facility (km)
Gym 0.54 (0.43) 0.54 (0.40) 0.56 (0.51)
Park 0.17 (0.11) 0.16 (0.12) 0.20 (0.11)
Fast Food 0.27 (0.18) 0.26 (0.16) 0.32 (0.24)
Grocery Store 0.64 (0.48) 0.62 (0.45) 0.72 (0.55)
Open in a new tab

SD = Standard deviation

Linear mixed models

Results of the linear mixed models that we used to evaluate the moderating effect of each SDOH on the intervention’s improvement of each of the four individual criterion health behaviors are reported in Table 2. We observed no evidence that any of the individual- or neighborhood-level SDOH measures provided a statistically significant moderating effect on any criterion health behavior examined (χ2[1] = 0.02–2.33, P-values > 0.05).

Table 2.

Summary of interaction effects from linear mixed models used to evaluate the moderating effect of access to facilities on improvement in each behavioral criterion across intervention and follow-up

Neighborhood Facility Access Measure Criterion Domain
Activity Domains
 Diet Domains
Moderate-Vigorous Physical Activity Sedentary Screen Time Fruit and Vegetable Servings Saturated Fat Percent
Fast Food −1.06 [−2.44, 0.32]
χ2[1] = 2.33, p = .13		 0.85 [−0.64, 2.33]
χ2[1] = 1.30, p = .25
Grocery Stores −0.09 [−0.70, 0.51]
χ2[1] = 0.09, p = .76		 −0.31 [−0.95, 0.33]
χ2[1] = 0.95, p = .33
Parks −1.46 [−4.73, 1.81]
χ2[1] = 0.78, p = .38		 −0.41 [−4.47, 3.66]
χ2[1] = 0.04, p = .84
Gyms −0.12 [−0.74, 0.51]
χ2[1] = 0.13, p = .72		 −0.05 [−0.77, 0.66]
χ2[1] = 0.02, p = .88
Open in a new tab

Note: Values are interaction fixed effects (with 95% confidence intervals in brackets) and the corresponding Likelihood Ratio Test evaluating the inclusion of the interaction fixed effect in the model

Cross K functions

We assessed clustering among 121 participant’s residential addresses, 866 fast food restaurants, 143 grocery stores, 284 gyms, and 8313 park boundary points. The Supplementary Material displays cross K functions between each relevant facility and behavior change outcome pair (fast food restaurants and grocery stores for diet outcomes, gyms and parks for activity outcomes). These figures allow for visualization of the cross K function “main effects” that were used in the difference comparisons of Figs. 1 and 2. Across all “main effects”, responders and non-responders for all goals and between all facilities clustered statistically significantly more across examined distances than would be expected under spatial independence, with the notable exceptions of FV non-responders and grocery stores and sedentary screen time non-responders and gyms, where the participant clustering did not consistently differ significantly from what would be expected under spatial independence. Figure 1 displays the results for the differences in cross K functions between responders and non-responders for the food facilities (fast food restaurants and grocery stores) and diet behavior change goals (consumption of fruits and vegetables and saturated fat). Figure 2 shows the differences in cross K functions between responders and non-responders for the recreational activity space facilities (gyms and parks) and activity behavior change goals (quantity of sedentary screen time and MVPA).

Fruit and Vegetable Responder Status and Diet-related Facility Access.

Differences in cross K functions showed that fruit and vegetable responders (consumption of ≥ 5 servings of fruits and vegetables per day) tended to statistically significantly cluster more around fast food restaurants relative to non-responders at distances up to approximately 1.40 km from fast food restaurants. For grocery stores, there was no statistically significant difference in clustering between responders and non-responders.

Saturated Fat Responder Status and Diet-Related Facility Access.

Saturated fat responders (consumption of < 8% daily calories from saturated fat) tended to statistically significantly cluster around fast food restaurants to a greater extent than non-responders at distances of about 0.60–3.00 km. Saturated fat responders tended not to significantly cluster more around grocery stores as compared to non-responders for most distances.

Sedentary Screen Time Responder Status and Recreational Activity Space Access.

Responders to sedentary screen time behavior change (≤ 90 min per day of sedentary screen time) were statistically significantly more likely to cluster around gyms as compared to non-responders at all distances greater than about 0.60 km. For parks, we estimated statistically significantly greater clustering among responders relative to non-responders at most distances above 0.30 km from parks.

Moderate-Vigorous Physical Activity Responder Status and Recreational Activity Space Access.

MVPA responders (≥ 150 min per week of MVPA) did not exhibit statistically significantly greater clustering than non-responders around gyms at most distances, with some intermittent evidence of marginal clustering, particularly at 2.50–3.00 km from gyms. Responders also had statistically significantly greater clustering than non-responders around parks at distances of about 1.25–3.00 km. By contrast, non-responders clustered around parks significantly more than responders at about 0.25–1.05 km from parks.

Discussion

We evaluated the extent to which access to food facilities and recreational activity spaces moderated response to an mHealth multiple health behavior change intervention targeting physical activity, diet, and sedentary leisure screen time. Under the umbrella of this parent aim, we examined two related yet distinct questions: (1) ‘To what extent does distance of residence from neighborhood facilities moderate the difference between a diet and activity health behavior change intervention condition vs. an active control condition?’ and (2) ‘Among recipients of a diet and activity health behavior change intervention, are responders more or less likely than non-responders to reside near relevant neighborhood facilities?’. We explored two competing alternative hypotheses that each corresponded with a different theoretical pathway. The blunting hypothesis suggests that living in a less supportive environment blunts an individual’s ability to benefit from the health behavior intervention. Accordingly, we would expect participants with less supportive environments to respond less favorably to the intervention than those living in a more supportive environment. The compensation hypothesis suggests that the intervention helps to compensate for a less supportive environment to produce disproportionally greater intervention-facilitated behavioral improvement relative to those in supportive environments. Accordingly, we expected participants living in a less supportive environment to benefit more (respond more favorably) from the intervention than those living in a more supportive environment.

Findings related to the question of whether access to environmental resources moderated the difference between the active intervention condition vs. the control condition were null, providing support for neither the blunting nor compensation pathway. The finding of a lack of significant moderating effect on intervention response is consistent with some existing studies that have evaluated and found no moderating effects of some individual-level SDOH on response to health behavior change intervention (Alcántara et al., 2020). Although this finding could indeed reflect a lack of treatment moderation by environmental context, it is also possible that the null findings here reflect methodologic limitations. For example, moderation models using the sample sizes typical for clinical trials are underpowered to detect these effects. Indeed, most studies, including the present one, were only adequately powered to detect a medium effect on the grounds that an effect of that magnitude would be considered clinically meaningful. This degree of power may have been insufficient to detect small effects that are clinically subthreshold, but that could have significant population-level implications if existing interventions were to be scaled.

For the question, ‘Among recipients of a diet and activity health behavior change intervention, are responders more or less likely than non-responders to reside near relevant neighborhood facilities?’, findings produced evidence in support for both alternative hypotheses. Within individuals randomized to receive the active intervention condition, we saw evidence of statistically significant interactions between facility access and behavior change. These relationships varied across behavior and facility type and show mixed support for each proposed hypothetical pathway.

Findings for access to food facilities and diet behaviors showed that fruit and vegetable and saturated fat responders were more likely to reside near fast food restaurants. This finding is consistent with the compensation pathway, suggesting that the intervention helped to compensate for or buffer negative effects resulting from greater access to fast foods. Future experimental research is necessary to replicate our findings, and to further explore mechanisms that may explain moderation of the effect of a diet change intervention by its association with fast food restaurant access.

In contrast, there were no differences in fruit and vegetable or saturated fat responder status based on access to grocery stores, a finding that may be consistent with the invariant pathway. Our current findings progress existing literature by suggesting that proximity to grocery stores may not necessarily be the optimal predictor of intervention-facilitated diet behavior change. Although several observational studies suggest proximity to grocery stores is associated with higher quality diet consisting of greater fruit and vegetable consumption and lower saturated fat consumption (e.g., (Rose & Richards, 2004; Zenk et al., 2009), there is also a growing literature suggesting that proximity to grocery stores alone may not directly influence diet quality (e.g. (Dubowitz et al., 2015). Attempts to explain mixed findings about the role of grocery store access in diet have highlighted several nuances. Most studies caution against evaluating the presence of a grocery store alone without accounting for important characteristics that may influence dietary habits, such as typical produce prices, food quality, or access to ethnic specialty foods (Beydoun et al., 2008; Shier et al., 2022; Walker et al., 2010). Other findings suggest that individuals will travel relatively long distances to access groceries (Liu et al., 2015) or, conversely, may prefer smaller convenience stores or ethnic specialty stores (Shier et al., 2022). As such, it may also be that our findings do not truly support the invariant pathway, but rather that our operationalization of grocery store access was insufficient to capture the myriad factors that could influence intervention-facilitated changes in diet quality.

For sedentary leisure screen time responder status and both recreational activity space (gyms and parks), findings were consistent with the blunting pathway. Sedentary screen time non-responders were less likely than responders to reside near either gyms or parks. It follows that greater access to gyms and parks may be associated with improved intervention-facilitated improvements in sedentary screen time, whereas lesser access is associated with a relatively reduced interventional benefit. Observational research has regularly shown that greater access to recreational activity spaces is associated with less sedentary screen time. For example, parks are a common outlet for leisure physical activities that have been shown to be substitutes for unhealthy leisure sedentary leisure screen time (Huston et al., 2003). Further research on the mechanisms of these interactions is warranted, particularly in the context of intervention. This research should inform if and how mHealth interventions targeting sedentary leisure screen time change might be tailored to reduce the possible blunting effect of low access to recreational activity spaces.

Findings regarding relationships between MVPA responder status and recreational activity space access were mixed and particularly complex. MVPA responders and non-responders were not different in terms of their rates of residing near gyms. Again, although this finding could provide evidence for the invariant pathway, it is also possible that our measures were insufficient to capture the variety of factors that could influence the relationship between gyms and intervention-facilitated MVPA (Smith et al., 2019). For example, some research has suggested that gyms of sufficient quality to elicit engagement in MVPA may be too cost-prohibitive for individuals of lower SES (Schroeder et al., 2019). More research is necessary to determine how to assess optimal gym access and whether it meaningfully interacts with intervention-facilitated MVPA.

The relationship between MVPA response status and park access was particularly complex, such that non-responders were more likely to reside at relatively closer distances (up to about 1.5 km within the 3 km buffer) and responders were more likely to reside at relatively farther distances (between 1.5k and 3 km). Evidence is mixed regarding whether access to parks is associated with increased MVPA (Marquet et al., 2022; Nigg et al., 2022). A recent review found that numerous characteristics may influence the relationships between recreational activity spaces and activity behaviors (Motomura et al., 2022). For example, a given park may not have the necessary physical infrastructure for MVPA (e.g., open space, sports fields/courts, outdoor fitness equipment). Moreover, features of the surrounding built environment may influence whether the park is considered accessible and safe (e.g., crime, walkability) (Motomura et al., 2022). Our findings, when considered together with existing literature, indicate a need for more research that granularly assesses features of recreational activity spaces and their surrounding contexts to understand interactions with intervention-facilitated MVPA.

In terms of our methodology, it is important to highlight that we assessed each study question using a distinct statistical approach, each with its own unique capabilities and nuances. These methods are not intended to be directly comparable. Rather, they each attempt to answer unique yet synergistic questions about the role of neighborhood resource access in multiple health behavior change intervention. As with any empirical study, although the answers to each question that we arrived at here could reflect the reality of the studied phenomena, they could also reflect the suitability of the methodologies used to answer the questions.

We used linear mixed models to assess the extent to which distance of residence from neighborhood facilities moderates the difference between a diet and activity health behavior change intervention condition vs. an active control condition. Linear mixed moderation models contrast participants who were randomized into the active conditions versus those who were randomized into the contact control condition. They assess whether the effect of access to resources moderates the intervention effect (explains part of the difference between the active condition vs the control condition). These models use the whole sample (both the active and control conditions).

We used cross K functions to assess, among recipients of a health behavior change intervention condition, whether responders vs. non-responders are more or less likely to reside near relevant neighborhood facilities. Cross K functions use only participants from the active intervention condition. They examine whether those who respond favorably to the active intervention condition are more or less likely to reside near neighborhood resources than those who did not respond favorably to the active intervention. We operationalized favorable intervention response as achieving health behavior goals specified during the intervention (e.g., ≤ 90 min per day of sedentary screen time, ≥ 5 servings of fruits and vegetables per day, and ≥ 150 min per week of MVPA, and < 8% of calories from saturated fat). As such, findings from the cross K functions evaluate individual differences in intervention response among those who are enrolled in the active intervention, as a function of their degree of access to resources. The sample for these analyses includes only those enrolled in the active intervention condition, and not the control condition.

In important ways, the current study builds upon past research that has evaluated the moderating effect of SDOH on the response to health behavior change interventions. This study is the first, to our knowledge, to evaluate the moderating effect of the neighborhood social and built environment on responsiveness to an mHealth intervention that leveraged both mobile technologies and remote coaching to target physical activity, diet, and sedentary screen time. The already complex interplay of resource access and intervention-facilitated health behavior change is further complicated by the recent trend among health behavior change interventions to incorporate mobile health (mHealth) tools and consumer information technologies, such as smartphone apps and worn sensors (Kumar et al., 2013). This rapid and widespread introduction of mHealth tools may create more ways in which the social environment interacts with intervention-facilitated health behavior change. Indeed, consistent with the Compensation Pathway, some have hypothesized that mHealth technologies can help to buffer or circumvent SDOH-related barriers among socially disadvantaged populations (Bakken et al., 2019). In line with this hypothesis, some pilot work shows preliminary efficacy of mHealth interventions that were designed to promote health in specific socially disadvantaged populations (Nollen et al., 2014), although this work has used small and homogenous samples. In contrast, consistent with the Blunting Pathway, some research suggests that mHealth tools might amplify intervention effects only for those with existing social privilege while blunting intervention effects for socially disadvantaged populations with technology-specific barriers, such as low technology literacy, diminished access to technology or internet connection, or general mistrust of technology (Bailey et al., 2015). To achieve health equity, it is critical that researchers understand the role of the social and built environment in mHealth intervention-facilitated health behavior change. Identifying how the features of the social and built environment moderate the response to mHealth multiple health behavior change interventions can begin to inform the selection of intervention components that are optimally tailored to the contexts of socially disadvantaged populations. Such information could also safeguard against the possibility of proliferating mHealth intervention components that may inadvertently worsen inequities.

This study also was the first to our knowledge to use a novel class of geospatial analysis, difference in cross K functions, to evaluate differences in spatial clustering by intervention response in a multiple health behavior change intervention. In doing so, our study also builds upon past research on the interplay between SDOH and multiple health behavior change intervention by studying neighborhood-level environmental factors, particularly by using analytics designed to understand relationships among spatial variables. Although past research has examined the moderating effects of SDOH on the response to a health behavior change intervention, those studies only evaluated a limited subset of SDOH, and exclusively at the individual-level (Alcántara et al., 2020). The current study evaluated the moderating effects of access to facilities shown to be associated with diet and activity health behaviors. These are important progressions, given emerging evidence and theoretical frameworks that highlight the importance of extra-individual SDOH, such as features of the neighborhood social and built environment.

Several important limitations should be considered when interpreting the results of this study. A common limitation of health behavior change interventions is the lack of data from members of the population who did not agree to participate (e.g., selection bias). It is likely that selection bias influenced the present findings, threatening their generalizability. Sociodemographic features impact health behaviors at various levels of spatial aggregation from household, street, block group, tract, and above, and may vary for different communities. Additionally, there are numerous ways to operationalize facility types and their qualities that may not be accounted for in our spatial measures. For example, distance-based measures likely do not entirely capture the breadth of components, such as affordability accommodation, or acceptability, that could influence one’s ability to meaningfully use a resource (Black et al., 2014; Caspi et al., 2012; Smith et al., 2019). Relatedly, there are likely to be numerous unobserved confounders at both the environmental and individual levels that could influence the outcomes measured in this study (Nigg et al., 2022). It is therefore possible that the geographic measures in our study may not have effectively captured environmental characteristics relevant to health behavior change moderation. Further work may elucidate whether sociocultural and built environment contextual features moderate health behavior change interventions at differing levels of geographic detail and using a variety of measurement approaches. As noted previously, the current study was not adequately powered to detect small moderating effects. Additionally, although the sample was racially diverse and geographically representative of the Chicago-area, results may not generalize to rural or suburban settings. As such, replication using large samples from diverse sociocultural settings and evaluating a broad array of SDOH is necessary to clarify the generalizability of the current study’s findings. Finally, perhaps the most critical limitation to these analyses is that they are secondary to the primary aims of the trial, and as such the sample, design, and other elements are not optimal for determining the true impact of the neighborhood social and built environment on intervention response. Moving forward, it would be advantageous for researchers to design studies for the explicit purpose of addressing questions regarding differential responding according to the social and built environment, rather than relying solely on secondary analyses of existing datasets. These findings provide preliminary evidence that such studies should be powered to detect small effect sizes.

Conclusion

The current study evaluated the extent to which access to food facilities and recreational activity spaces moderated response to an mHealth intervention that leveraged both mobile technologies and remote coaching to target physical activity, diet, and sedentary leisure screen time. For the study question, “To what extent does distance of residence from neighborhood facilities moderate the difference between a diet and activity health behavior change intervention condition vs an active control condition?”, findings did not support a moderating effect of access to food facilities or recreational activity spaces on the response to interventions targeting health behavior improvement for any of the four behavioral domains examined (increase in fruit and vegetable servings, reduction in saturated fat consumption, increase in MVPA, and decrease in sedentary leisure screen time). For the study question, “Among recipients of a diet and activity health behavior change intervention, are responders more or less likely than non-responders to reside near relevant neighborhood facilities?”, findings indicated a complex relationship between spatial proximity to facilities and intervention response. This relationship varied by diet or activity behavior and facility type. These results suggest that mHealth intervention may help to counteract detrimental effects of heightened access to fast food and facilitate greater improvements in saturated fat and fruit and vegetable consumption for individuals with greater fast food access. Conversely, limited access to recreational activity spaces may blunt or diminish intervention-facilitated improvement in sedentary leisure screen time. MVPA responders had greater and lesser clustering than non-responders around recreational activity spaces, dependent upon distance from activity space to participant residence. Finally, access to grocery stores was not associated with responder status, suggesting a lack of a meaningful interaction with intervention-facilitated diet behaviors.

Our findings also highlight important methodological issues (power, study design, spatial measurement practices) that pervade research. These issues limit current understanding of how mHealth multiple health behavior change interventions interface with neighborhood social and built environment features that might influence intervention impact on health behavior change. Although additional secondary analyses of interventional studies may be beneficial, researchers should consider broadening the tools used to deconstruct this issue. In addition to examining the effects of context in trials, researchers should also leverage complementary tools, such as mixed methods, natural experiments, and propensity score matching. It is essential to conduct studies explicitly designed and adequately powered to identify differential responses influenced by theoretically-selected SDOH at various levels. These studies should carefully consider analytic methods and spatial measurement. Such research is necessary to understand how environmental factors, mHealth tools, and telecoaching interact to equitably promote health behavior change.

Supplementary Material

technical appendix
cross K figure supplemental

Acknowledgements

This work was supported by the National Cancer Institute [grant number T32CA193193] and NHLBI (R01 075451-09 and F31 162555).

Disclosures

D.J. Press reports other support from Genentech (A Member of the Roche Group) outside the submitted work. Otherwise, authors do not have disclosures to report.

References

  1. Alcántara C, Diaz SV, Cosenzo LG, Loucks EB, Penedo FJ, & Williams NJ (2020). Social determinants as moderators of the effectiveness of health behavior change interventions: Scientific gaps and opportunities. Health Psychology Review, 14(1), 132–144. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Alvidrez J, Castille D, Laude-Sharp M, Rosario A, & Tabor D (2019). The national institute on minority health and health disparities research framework. American Journal of Public Health, 109(S1), S16–S20. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Baddeley A, Rubak E, & Turner R (2015). Spatial point patterns: Methodology and applications with R. CRC. [Google Scholar]
  4. Bailey SC, O’conor R, Bojarski EA, Mullen R, Patzer RE, Vicencio D, Jacobson KL, Parker RM, & Wolf MS (2015). Literacy disparities in patient access and health-related use of I nternet and mobile technologies. Health Expectations, 18(6), 3079–3087. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Bakken S, Marden S, Arteaga SS, Grossman L, Keselman A, Le PT, Creber RM, Powell-Wiley TM, Schnall R, & Tabor D (2019). Behavioral interventions using consumer information technology as tools to advance health equity. American Journal of Public Health, 109(S1), S79–S85. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Beydoun MA, Powell LM, & Wang Y (2008). The association of fast food, fruit and vegetable prices with dietary intakes among US adults: Is there modification by family income? Social Science & Medicine, 66(11), 2218–2229. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Black C, Moon G, & Baird J (2014). Dietary inequalities: What is the evidence for the effect of the neighbourhood food environment? Health & Place, 27, 229–242. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Braveman P, Egerter S, & Williams DR (2011). The Social Determinants of Health: Coming of Age. Annual Review of Public Health, 32(1), 381–398. 10.1146/annurev-publhealth-031210-101218 [DOI] [PubMed] [Google Scholar]
  9. Bureau, U. S. C. Census 2010. Population counts (Social Explorer). [Google Scholar]
  10. Calloway EE, Parks CA, Bowen DJ, & Yaroch AL (2019). Environmental, social, and economic factors related to the intersection of food security, dietary quality, and obesity: An introduction to a special issue of the translational behavioral medicine journal. Translational Behavioral Medicine, 9(5), 823–826. [DOI] [PubMed] [Google Scholar]
  11. Caspi CE, Sorensen G, Subramanian S, & Kawachi I (2012). The local food environment and diet: A systematic review. Health & Place, 18(5), 1172–1187. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Catford JA, Wilson JR, Pyšek P, Hulme PE, & Duncan RP (2022). Addressing context dependence in ecology. Trends in Ecology & Evolution, 37(2), 158–170. [DOI] [PubMed] [Google Scholar]
  13. Christian H, Knuiman M, Divitini M, Foster S, Hooper P, Boruff B, Bull F, & Giles-Corti B (2017). A longitudinal analysis of the influence of the neighborhood environment on recreational walking within the neighborhood: Results from RESIDE. Environmental Health Perspectives, 125(7), 077009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Darmon N, & Drewnowski A (2008). Does social class predict diet quality? The American Journal of Clinical Nutrition, 87(5), 1107–1117. [DOI] [PubMed] [Google Scholar]
  15. Dixon PM (2002). Ripley’s K function. Encyclopedia of Environmetrics, 3, 1796–1803. [Google Scholar]
  16. Dubowitz T, Ghosh-Dastidar M, Cohen DA, Beckman R, Steiner ED, Hunter GP, Flórez KR, Huang C, Vaughan CA, & Sloan JC (2015). Diet and perceptions change with supermarket introduction in a food desert, but not because of supermarket use. Health Affairs, 34(11), 1858–1868. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Ghosh-Dastidar M, Hunter G, Collins RL, Zenk SN, Cummins S, Beckman R, Nugroho AK, Sloan JC, & Dubowitz T (2017). Does opening a supermarket in a food desert change the food environment? Health & Place, 46, 249–256. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Grossman M (2017). The demand for health: A theoretical and empirical investigation. Columbia university. [Google Scholar]
  19. Hawes AM, Smith GS, McGinty E, Bell C, Bower K, LaVeist TA, Gaskin DJ, & Thorpe RJ (2019). Disentangling race, poverty, and place in disparities in physical activity. International Journal of Environmental Research and Public Health, 16(7), 1193. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Huston SL, Evenson KR, Bors P, & Gizlice Z (2003). Neighborhood environment, access to places for activity, and leisuretime physical activity in a diverse North Carolina population. American Journal of Health Promotion, 18(1), 58–69. [DOI] [PubMed] [Google Scholar]
  21. Kelman J, Pool LR, Gordon-Larsen P, Carr JJ, Terry JG, Rana JS, & Kershaw KN (2019). Associations of unhealthy food environment with the development of coronary artery calcification: The Cardia study. Journal of the American Heart Association, 8(4), e010586. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Kolak M, Bradley M, Block D, Pool L, Garg G, Toman CK, Boatright K, Lipiszko D, Koschinsky J, Kershaw K, Carnethon M, Isakova T, & Wolf M (2018). Chicago supermarket data and food access analytics in census tract shapefiles for 2007–2014. Data Brief, 21, 2482–2488. 10.1016/j.dib.2018.11.014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Kumar S, Nilsen WJ, Abernethy A, Atienza A, Patrick K, Pavel M, Riley WT, Shar A, Spring B, & Spruijt-Metz D (2013). Mobile health technology evaluation: The mHealth evidence workshop. American Journal of Preventive Medicine, 45(2), 228–236. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Liu JL, Han B, & Cohen DA (2015). Beyond Neighborhood Food environments: Distance Traveled to Food Establishments in 5 US Cities, 2009–2011. Preventing Chronic Disease, 12, E126. 10.5888/pcd12.150065 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. MacMillan F, George ES, Feng X, Merom D, Bennie A, Cook A, Sanders T, Dwyer G, Pang B, & Guagliano JM (2018). Do natural experiments of changes in neighborhood built environment impact physical activity and diet? A systematic review. International Journal of Environmental Research and Public Health, 15(2), 217. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Marquet O, Hirsch JA, Kerr J, Jankowska MM, Mitchell J, Hart JE, Laden F, Hipp JA, & James P (2022). GPS-based activity space exposure to greenness and walkability is associated with increased accelerometer-based physical activity. Environment International, 165, 107317. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. McCormack GR, Patterson M, Frehlich L, & Lorenzetti DL (2022). The association between the built environment and intervention-facilitated physical activity: A narrative systematic review. International Journal of Behavioral Nutrition and Physical Activity, 19(1), 86. 10.1186/s12966-022-01326-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Miles JN, Weden MM, Lavery D, Escarce JJ, Cagney KA, & Shih RA (2016). Constructing a time-invariant measure of the socio-economic status of US census tracts. Journal of Urban Health, 93, 213–232. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Mokdad AH, Marks JS, Stroup DF, & Gerberding JL (2004). Actual causes of death in the United States, 2000. Journal of the American Medical Association, 291(10), 1238–1245. [DOI] [PubMed] [Google Scholar]
  30. Motomura M, Koohsari MJ, Lin CY, Ishii K, Shibata A, Nakaya T, Kaczynski AT, Veitch J, & Oka K (2022). Associations of public open space attributes with active and sedentary behaviors in dense urban areas: A systematic review of observational studies. Health & Place, 75, 102816. [DOI] [PubMed] [Google Scholar]
  31. Nigg C, Niessner C, Burchartz A, Woll A, & Schipperijn J (2022). The geospatial and conceptual configuration of the natural environment impacts the association with health outcomes and behavior in children and adolescents. International Journal of Health Geographics, 21(1), 1–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Nollen NL, Mayo MS, Carlson SE, Rapoff MA, Goggin KJ, & Ellerbeck EF (2014). Mobile technology for obesity prevention: A randomized pilot study in racial-and ethnic-minority girls. American Journal of Preventive Medicine, 46(4), 404–408. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Organization, W. H. (2010). A conceptual framework for action on the social determinants of health. [Google Scholar]
  34. Ottman R (1996). Gene–environment interaction: Definitions and study design. Preventive Medicine, 25(6), 764–770. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Pellegrini CA, Steglitz J, Johnston W, Warnick J, Adams T, McFadden H, Siddique J, Hedeker D, & Spring B (2015). Design and protocol of a randomized multiple behavior change trial: Make better choices 2 (MBC2). Contemporary Clinical Trials, 41, 85–92. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Ranchod YK, Roux D, Evenson AV, Sánchez KR, B. N, & Moore K (2014). Longitudinal associations between neighborhood recreational facilities and change in recreational physical activity in the multi-ethnic study of atherosclerosis, 2000–2007. American Journal of Epidemiology, 179(3), 335–343. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3895096/pdf/kwt263.pdf [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Raskind IG, Kegler MC, Girard AW, Dunlop AL, & Kramer MR (2020). An activity space approach to understanding how food access is associated with dietary intake and BMI among urban, low-income African American women. Health & Place, 66, 102458. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Rose D, & Richards R (2004). Food store access and household fruit and vegetable use among participants in the US Food Stamp Program. Public Health Nutrition, 7(8), 1081–1088. [DOI] [PubMed] [Google Scholar]
  39. Satia JA (2009). Diet-related disparities: Understanding the problem and accelerating solutions. Journal of the American Dietetic Association, 109(4), 610. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Schroeder K, Klusaritz H, Dupuis R, Bolick A, Graves A, Lipman TH, & Cannuscio C (2019). Reconciling opposing perceptions of access to physical activity in a gentrifying urban neighborhood. Public Health Nursing, 36(4), 461–468. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Shier V, Miller S, & Datar A (2022). Heterogeneity in grocery shopping patterns among low-income minority women in public housing. BMC Public Health, 22(1), 1612. 10.1186/s12889-022-14003-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Smith M, Hosking J, Woodward A, Witten K, MacMillan A, Field A, Baas P, & Mackie H (2017). Systematic literature review of built environment effects on physical activity and active transport–an update and new findings on health equity. International Journal of Behavioral Nutrition and Physical Activity, 14(1), 1–27. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Smith L, Foley L, & Panter J (2019). Activity spaces in studies of the environment and physical activity: A review and synthesis of implications for causality. Health & Place, 58, 102113. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Sotos-Prieto M, Bhupathiraju SN, Mattei J, Fung TT, Li Y, Pan A, Willett WC, Rimm EB, & Hu FB (2017). Association of changes in diet quality with total and cause-specific mortality. New England Journal of Medicine, 377(2), 143–153. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Spring B, Schneider K, McFadden HG, Vaughn J, Kozak AT, Smith M, Moller AC, Epstein LH, DeMott A, & Hedeker D (2012). Multiple behavior changes in diet and activity: A randomized controlled trial using mobile technology. Archives of Internal Medicine, 172(10), 789–796. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Spring B, Pellegrini C, McFadden H, Pfammatter AF, Stump TK, Siddique J, King AC, & Hedeker D (2018). Multicomponent mHealth intervention for large, sustained change in multiple diet and activity risk behaviors: The make better choices 2 randomized controlled trial. Journal of Medical Internet Research, 20(6), e10528. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Stamatakis E, Gale J, Bauman A, Ekelund U, Hamer M, & Ding D (2019). Sitting time, physical activity, and risk of mortality in adults. Journal of the American College of Cardiology, 73(16), 2062–2072. [DOI] [PubMed] [Google Scholar]
  48. Walker RE, Keane CR, & Burke JG (2010). Disparities and access to healthy food in the United States: A review of food deserts literature. Health & Place, 16(5), 876–884. [DOI] [PubMed] [Google Scholar]
  49. Zenk SN, Lachance LL, Schulz AJ, Mentz G, Kannan S, & Ridella W (2009). Neighborhood retail food environment and fruit and vegetable intake in a multiethnic urban population. American Journal of Health Promotion, 23(4), 255–264. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Zenk SN, Tarlov E, & Kraft AN (2019). Environmental modification of adult weight loss, physical activity, and diet intervention effects. Geospatial approaches to energy balance and breast Cancer (pp. 255–281). Springer. [Google Scholar]
  51. Zenk SN, Tarlov E, Wing C, Slater S, Jones KK, Fitzgibbon M, & Powell LM (2019). Does the built environment influence the effectiveness of behavioral weight management interventions? Preventive Medicine, 126, 105776. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

technical appendix
NIHMS2087518-supplement-technical_appendix.docx (69.6KB, docx)
cross K figure supplemental
NIHMS2087518-supplement-cross_K_figure_supplemental.docx (547.9KB, docx)

RESOURCES