Impact of the built, social, and food environment on long‐term weight loss within a behavioral weight loss intervention

Abstract Background Behavioral weight loss interventions can lead to an average weight loss of 5%–10% of initial body weight, however there is wide individual variability in treatment response. Although built, social, and community food environments can have potential direct and indirect influences on body weight (through their influence on physical activity and energy intake), these environmental factors are rarely considered as predictors of variation in weight loss. Objective Evaluate the association between built, social, and community food environments and changes in weight, moderate‐to‐vigorous physical activity (MVPA), and dietary intake among adults who completed an 18‐month behavioral weight loss intervention. Methods Participants included 93 adults (mean ± SD; 41.5 ± 8.3 years, 34.4 ± 4.2 kg/m2, 82% female, 75% white). Environmental variables included urbanicity, walkability, crime, Neighborhood Deprivation Index (includes 13 social economic status factors), and density of convenience stores, grocery stores, and limited‐service restaurants at the tract level. Linear regressions examined associations between environment and changes in body weight, waist circumference (WC), MVPA (SenseWear device), and dietary intake (3‐day diet records) from baseline to 18 months. Results Grocery store density was inversely associated with change in weight (β = −0.95; p = 0.02; R 2 = 0.062) and WC (β = −1.23; p < 0.01; R 2 = 0.109). Participants living in tracts with lower walkability demonstrated lower baseline MVPA and greater increases in MVPA versus participants with higher walkability (interaction p = 0.03). Participants living in tracts with the most deprivation demonstrated greater increases in average daily steps (β = 2048.27; p = 0.02; R 2 = 0.039) versus participants with the least deprivation. Limited‐service restaurant density was associated with change in % protein intake (β = 0.39; p = 0.046; R 2 = 0.051). Conclusion Environmental factors accounted for some of the variability (<11%) in response to a behavioral weight loss intervention. Grocery store density was positively associated with weight loss at 18 months. Additional studies and/or pooled analyses, encompassing greater environmental variation, are required to further evaluate whether environment contributes to weight loss variability.


| INTRODUCTION
Despite the short-term effectiveness of lifestyle interventions for weight loss, many individuals regain significant weight within a 1-year period and there is significant inter-individual variability in weight loss response. 1 The National Institutes of Health (NIH) Accumulating Data to Optimally Predict obesity Treatment (ADOPT) Core Measures Project was created to identify factors that predict this variability in response to obesity treatment. ADOPT identifies four domains of focus: biological, behavioral, psychosocial, and environmental. 2 While several studies have evaluated the contribution of biological, behavioral, and psychosocial factors, fewer have evaluated the role of environment. 3 The built environment describes human-made aspects of community design and includes urbanicity (rural vs. urban) and walkability (how friendly an area is for walking). 4 The social environment describes the makeup of a neighborhood's culture, groups, relationships, and social processes and includes factors like crime (crimes against persons and property) and Neighborhood Deprivation Index (NDI; composed of 13 social economic status (SES) factors). 5 Lastly, the community food environment is a broad term that describes the distribution, number, type, and location of food sources and includes density of convenience stores, grocery stores, and limited-service restaurants (LSRs). 6 It has been hypothesized that environment can influence obesity outcomes. 7 Observational studies have found that obesity was inversely associated with walkability, 8 urbanicity (with more urban populations having a lower prevalence of obesity), 9 SES, 10 and the presence of grocery stores, 11 but positively associated with increasing crime, 12 and presence of convenience stores. 11 However, the role of environment on weight loss outcomes within the context of a behavioral weight loss intervention has not been well established.
There have been three prior interventional studies that have evaluated the role of environment. Mench et al. found that rural versus urban status did not moderate weight loss or changes in selfreported physical activity (PA) among a sample of 492 adults who participated in a 6-month weight loss intervention. 13 In another study of 114,256 participants from the U.S. Department of Veterans Affairs (VA) MOVE! weight management program, there was no association between walkability, park access, or fitness facility access and weight change at 6, 12, 18, or 24 months. 14 In terms of the social environment, a third study by Mendez et al. found no association between SES (poverty rate, neighborhood income) and changes in weight over 6 months during a behavioral weight loss intervention among 127 adults. 15 Lastly, while the community food environment moderated weight loss response among men, but not women, in the VA MOVE! program at 6 months, 16 community food environment was not associated with changes in weight at 24 months. 17 One explanation for this lack of association between environment and weight loss outcomes is that factors other than environment, such as psychosocial (motivation, self-regulation), biological (changes in resting energy expenditure, appetite) and/or behavioral (sleep, timing of eating) may have a greater contribution to weight loss outcomes. 2 Alternatively, the lack of association could be due to the heterogeneity in methods 18 and/or a lack of focus on local environmental supports within the behavioral intervention content. Thus, additional research is needed to explore the association between environment and responses to behavioral weight loss interventions.
The purpose of this study was to conduct a secondary analysis of data from an 18-month behavioral weight loss intervention to explore whether environmental factors, recommended by ADOPT, 3 are associated with changes in weight at 18 months. The hypotheses were that 1) greater levels of urbanicity and/or walkability would be associated with greater weight loss, 2) higher levels of crime and/or NDI would be associated with less weight loss, and 3) density of convenience stores, and/or LSRs would be associated with less weight loss, while density of grocery stores would be associated with greater weight loss at 18 months. The association between environment and changes in waist circumference (WC), PA, and dietary intake were also explored.

| Description of the behavioral weight loss trial
A full description of the weight loss trial (NCT01985568) and primary results were published previously. 19 All participants provided written informed consent. Briefly, 170 adults with overweight/ 262 -TEWAHADE ET AL. obesity (age 18-55 years, Body Mass Index (BMI) 27-42 kg/m 2 , 84% female) were randomized 1:1, stratified by sex, to receive one of two 18-month group-based behavioral weight loss interventions: standard behavioral therapy (Standard) or sequential behavioral therapy (Sequential). Both randomized groups received an identical 18-month group-based behavioral weight loss program (weekly group meetings during months 0-6 followed by monthly group meetings during months 7-18 led by a registered dietitian), including a reduced calorie diet (1200-1800 kcal/day). 19 Targeted macronutrient content was 20%-30% fat, 50%-55% carbohydrates, and 20%-25% protein. Randomized groups differed only in the timing of exercise initiation. The Standard group received a supervised exercise program, progressing to 300 min/week of moderate intensity erobic activity during months 0-6, followed by unsupervized exercise during months 7-18. The Sequential group was asked to refrain from changing their exercise habits during months 0-6 and received an identical supervised exercise program during months 7-12, followed by unsupervized exercise during months 13-18. On completion of the 6-month supervised exercise phase, participants in both groups were instructed to continue 300 min/ week of moderate intensity activity and were provided continued access to the exercise facility for the remainder of the study. At 18 months, there were no differences between the Standard and Sequential in changes in weight, moderate-to-vigorous PA (MVPA), or energy intake (EI). 19

| Analytical sample
Participants were included if they completed the 18-month intervention (n = 120, 71% completion rate). Participants were excluded if they changed their home address during the intervention (n = 27).
Thus, of 170 participants randomized, 93 were included in the analysis.

| Anthropometrics
The primary outcome was change in weight (%) from baseline to 18 months. Body weight was measured using a calibrated digital scale (to the nearest 0.1 kg). Height was measured with stadiometer. WC (cm) was measured at the level of the superior iliac crest.

| Dietary energy and macronutrient intake
Dietary EI (kcal/d) and macronutrient intake (% kcal from fat, carbohydrates, and protein) were assessed using 3-day diet records.
Diet records were analyzed using Nutrition Data System for Research (version 2016; Nutrition Coordinating Center, University of Minnesota, Minneapolis, Minnesota) by blinded core laboratory staff.

| Environment variables
Each participant self-reported their home address at baseline and any changes in home address at the 18-month visit. Home address was used to determine participant geolocation (Census geocoding services). 20 22 All environmental variables were analyzed at the census tract level.

| Urbanicity
Urbanicity is a six category variable 21 using National Center for Education Statistics urban/rural locale definitions, applied to Census urban/rural population data. 23 Due to low participant counts in mixed (n = 5) and rural (n = 2) urbanicity, those categories were removed from analyses and treated as missing.

| Walkability
Walkability is composed of measures from block-level variables based on the 2010 census (2013 National Walkability Index dataset). 21 Block-level data came from the Environmental Protection Agency's Smart Location Mapping project 24 and was aggregated into tract-level variables. Walkability was examined both as a continuous variable and a categorical variable, categorized by tertiles based on the analytical sample (low: n = 26, 9.6 � 2.8 walkability score; medium: n = 26, 12.9 � 0.5 walkability score; and high: n = 28, 15.1 � 0.8 walkability score). total crime, personal crime (murder, rape, robbery, assault), and property crime (burglar, larceny, motor vehicle theft) by census tract. Per the licensing agreement with Applied Geographic Solutions, the study team was required to analyze the data using categories. Crime was sorted by low to high and secondarily by personal crime rate as several tracts had the same crime rate. 21 Cumulative population was then calculated as the tract's population plus the previous tract's cumulative population. The ration of cumulative population to one third of the total analytical sample population was calculated to create the crime tertiles ratio. A tract was defined as "low" crime for crime tertile ratio ≤1.0; "medium" for crime tertile ratios >1.0 and ≤ 2.0; and "high" crime for crime tertile ratios ≤3.0.

| Neighborhood Deprivation Index
NDI was created using factor analysis of tract-level variables at the national level. 21,22 NDI is composed of 13 SES factors broken down into four different categories: wealth and income, education, occupation, and housing conditions. 22 All variables used to create NDI were obtained from the Census Bureau's 5-year American Community Survey data for 2013-2017. 22 NDI ranges from −2.5 to +1.9, which is further categorized into quintiles, weighted by the census tract population (such that 20% of the population is in each quintile group). Categories include: least deprivation, below average deprivation, average deprivation, above average deprivation, most deprivation, and NDI not specified. NDI was analyzed based on quintiles to improve interpretability of results. Participants in the "NDI not specified" category (n = 3) were treated as missing in analyses.

| Community food environment
Counts of convenience stores and grocery stores were obtained from historical commercial business listings for the specified Standard Industrial Classification codes and chain names from the year 2019 for the state of Colorado from Data Axle USA ® as outlined by Jones et al. 25 Historical listings for the same state and time period for LSRs were purchased from Dun & Bradstreet. 25 Density of each food outlet type was then summarized by each tract for Colorado (number of stores/land area, mi 2 ). 21 In addition, the density of food outlet types for each tract, plus their neighboring tracts, was calculated to account for the idea that participants may shop for food outside of their immediate tract. This variable was subsequently used in sensitivity analyses.

| Assessment of covariates
Age, sex, race, ethnicity, and education were self-reported at baseline.

| Study participant characteristics
Baseline demographic characteristics of the analytical sample (n = 93) were similar to that of the randomized sample (n = 170, Table 1), except for age. Those excluded from the analysis (n = 77) were, on average, younger (mean � SD; 36.5 � 9.6 years) compared to the analytical sample (41.5 � 8.3 years; t value −3.63; p < 0.01).

| Environmental data
None of the tested covariates (age, sex, race, ethnicity, education, and randomized group) were associated with changes in outcome variables, thus these covariates were not considered further in these analyses. Environmental data are described in Table 2. Within the sample of 93 participants, 72 tracts were represented.

| Association between environment and change in anthropometric outcomes
There was a significant association between grocery store density and changes in weight (%) and WC (Table 3) (Table 3). After conducting a sensitivity analysis of the density of food outlets within each tract, plus the neighboring tracts, there was no association between community food environment variables and weight change at 18 months (Supplementary Table S1).

| Association between environment and change in PA outcomes
There was a significant inverse association between walkability and changes in total MVPA at 18 months: for every 1 unit increase in walkability, change in total MVPA at 18 months decreased by 3.76 min/day (R 2 = 0.079; Table 4 Figure S2A). There were no other significant associations between other environment variables and changes in PA outcomes (Table 4). In a post-hoc analysis, there was no correlation between walkability and average number of fitness center check-ins per month (r = −0.04, p = 0.67).

| Association between environment and change in dietary intake outcomes
Limited-service restaurant density was positively associated with change in % protein intake at 18 months (p = 0.046; R 2 = 0.051; Figure S2B). There were no other associations between environmental variables and changes in EI or dietary macronutrient content ( -267 that high levels of neighborhood crime were associated with a decrease in BMI. 12 Importantly, the way in which SES is measured is critical, and future studies should consider this when exploring the relationship between SES and weight. For example, a review of longitudinal cohort studies found that when SES is measured with occupation, lower SES was associated with greater weight change over time, but when SES was measured using income, findings were inconsistent. 10  ~52.5% carbohydrates, and~22.5% protein), though this did not translate to significant reductions in EI. In two cross-sectional studies, lower SES (occupation, housing, and education) was significantly associated with lower protein intake. 36,37 The authors are unaware of any other studies that have investigated the association between NDI and changes in dietary intake outcomes in the context of a behavioral weight loss intervention.
Adults living in a tract with a higher density of LSRs were more likely to increase their % protein intake at 18 months compared to adults living in tracts with a lower LSR density. In contrast to the present study's findings, Barnes et al. found no association between frequency of fast-food consumption and % protein intake, both crosssectionally and prospectively over 6 months in a free-living sample. 38 There was no association between convenience store density or grocery store density and changes in EI or dietary macronutrient intake.
Alternatively, Wedick et al. found that proximity to health food stores was associated with greater improvements in consumption of fiber and fruit and vegetables at 1 year among 204 adults with obesity and The following categories were removed from analyses due to the small sample sizes: NDI not at all for NDI.
Results indicate that environmental factors may account for some of the variability (<11%) in response to behavioral weight loss interventions. It will be important to continue to assess the impact of environmental factors on weight loss outcomes in additional, interventional studies with assessments that capture the dynamic, contextual environmental factors that influence EI and PA behavior.
Pooling data across multiple weight loss interventions will improve knowledge regarding whether environment moderates weight loss outcomes. This line of research could lead to policies that support environmental changes to enhance environmental supports (e.g., improving access to grocery stores) and/or the development of more

ACKNOWLEDGMENTS
We would like to thank our study participants and funding partners (NIH R01 DK 097266, NIH UL1 TR001082, NIH F32 DK122652, and NIH P30 DK048520). We would also like to thank Dr. Susan Czajkowski and Dr. William Klein for their review of the manuscript.
Lastly, we would like to thank Dr. Ian Buller for his work that informed our computation of the NDI variable used in the present manuscript (https://cran.r-project.org/web/packages/ndi/index.html).