Background

It is estimated that 97 million adults in the United States are either overweight or obese [1]. Obesity, defined as a body mass index (BMI) of 30 kg/m2 or higher in adults, is a complex disease that arises from interactions between multiple genes, as well as behavioral and environmental factors [1]. Furthermore, obesity is a serious risk factor for many chronic conditions (diabetes, hypertension, hypercholesterolemia, stroke, heart disease, certain cancers, and arthritis) and has been reported to markedly decrease life expectancy [2–4]. The prevalence of obesity was relatively stable between 1960 and 1980, but has dramatically increased over the past 20 years [5]. Although the health risks of obesity are well established, there is less certainty about the management of the disease. Lifestyle modification programs to address obesity prevention and weight loss have achieved only moderate success, particularly interventions for long-term weight loss [6].

New public health approaches to the obesity problem are urgently required. One factor that may play a role in the risk of obesity is the neighborhood environment [7, 8]. Neighborhood socioeconomic conditions are known to affect health even after controlling for individual-level socio-demographic factors [9–11]. Recent data suggests that the neighborhood environment may influence risk of chronic diseases such as cardiovascular disease, type 2 diabetes, and related health behaviors such as decreased levels of physical activity [12–22]. A few studies have demonstrated that living in neighborhoods with low socio-economic status (SES) is associated with an increased risk of obesity [23, 24]. However, there are few data on neighborhood and weight-control behaviors. Investigating novel correlates of weight control is necessary given there is overwhelming evidence showing that weight loss is associated with marked improvement in health status, particularly, blood pressure and glucose control [25–27].

Therefore, we conducted a multi-level analysis to examine the relationship between neighborhood SES and weight-related variables at baseline among overweight participants with type 2 diabetes enrolled in the Look AHEAD study. We hypothesized that: 1) poorer neighborhood SES would be associated with poorer eating patterns and weight control behaviors independent of individual-level socio-economic status and 2) more availability of stores with healthy options (i.e. food stores) in the neighborhood would be associated with better eating patterns and weight control behaviors independent of individual-level socio-economic status.

Methods

Study Population of the Parent Study

The primary objective of the Look AHEAD study [28] is to examine, in overweight volunteers with type 2 diabetes, the long-term effects of an intensive lifestyle intervention program designed to achieve and maintain weight loss by decreased caloric intake and increased physical activity. The intervention group is compared to a control condition involving a program of diabetes education and support. The primary basis for the comparison is the incidence of serious cardiovascular events. Other outcomes, including cardiovascular disease risk factors, diabetes-related metabolic factors and complications, and the cost-effectiveness of the intensive intervention are also studied. Participants are 5,145 volunteers with type 2 diabetes who are 45-75 years of age and overweight or obese (body mass index [BMI] ≥ 25 kg/m2).

Study Population of the Ancillary Study

This ancillary study was conducted at baseline using Look AHEAD participants at 4 clinical sites; Baltimore(n = 302), Philadelphia(n = 293), Pittsburgh(n = 321), and New York(n = 303). Sites were chosen because of their close geographic proximity and similar demographic profile. The total study sample for this ancillary consists of 1219 participants with complete data on neighborhood environment. Addresses were used to identify the corresponding census tracts for each participant (neighborhood) as defined by the 2000 Census using a process called geocoding and software program ArcGIS™. The program matches imported addresses to geographic maps and other geographic data. Matches are rated with scores from 0 (no match) to 100 (perfect match); we accepted matches with 80% certainty or more. Once we identified the census tracts and corresponding data for each participant, these data were linked to the individual-level data collected during the Look AHEAD trial. A description of all of the main variables used in this analysis is summarized in Table 1.

Table 1 Selected Characteristics of 1219 Look AHEAD participants
Full size table

Main Data Sources

Data are derived from the 2000 US Census long form and include demographic characteristics (age, race, sex), housing characteristics (housing structure, number of rooms, telephone surface), economic characteristics (occupation, place of work and journey to work) and financial characteristics (value of home, rent, utilities cost) for each census tract.

We also used data from the 2004 Consumer Expenditure database which outlines the locality of food stores using a multi-level hierarchical classification system. The data are derived from an extensive modeling effort using the 1994, 1995, 1996, 1997, and 2000 Consumer Expenditure Survey data from the Bureau of Labor Statistics (BLS), in addition to the latest 1998 overview data. The BLS survey averages over 5,000 households four times a year using a rotating sampling frame. We used aggregate data at the census tract level (estimated at 3000-5000 persons).

Participants in the Look AHEAD study underwent extensive data collection at baseline, including interview, physical examination, and blood and urine assays[28]. Although the trial will last over 10 years, this manuscript is restricted to data collected at baseline only. The parent study was approved by the Johns Hopkins Western Institutional Review Board and all participants signed written informed consent to participate in the study.

Key Independent Variables

Using the Census data, indices of neighborhood socio-economic status developed by Diez-Roux and Winkleby/Cubbin, were created using variables such as the % of persons living below poverty, % of adults with a college degree, median household income, % of persons earning interest income, % of adults in executive/managerial occupations, and % of adults who are unemployed. After considering these measures used in previous studies [29–31], we ultimately decided on the single item "% of individuals in the census tract living below the federal poverty line" because this measure is highly correlated with other census-based indices and has been shown to be similarly predictive of health outcomes [31].

Data on food availability in the census tract was categorized using the North American Industry Classification System (NAICS) definitions into: 1) food storesthese establishments retail food and beverages merchandise from fixed point-of-sale locations. Establishments in this subsector have special equipment (e.g., freezers, refrigerated display cases, refrigerators) for displaying food and beverage goods. They have staff trained in the processing of food products to guarantee the proper storage and sanitary conditions required by regulatory authority; includes grocery stores and supermarkets; 2) convenience storesthese establishments primarily engaged in retailing a limited line of goods that generally includes milk, bread, soda, and snacks; and 3) restaurantsthese establishments primarily engaged in providing food services to patrons who order and are served while seated (i.e. waiter/waitress service and pay after eating. they may provide this type of food service to patrons in combination with selling alcoholic beverages, providing carry out services, or presenting live non-theatrical entertainment; includes full-service, fast food, and carryout.

Key Dependent Variables

Dependent variables included eating patterns, weight loss control practices, and BMI. Variables capturing participant eating patterns consisted of reports of eating breakfast, lunch, and dinner and whether they ate at fast-food or non-fast food restaurants. Dietary intake, daily total fat, saturated fat, and fruit and vegetable intake was measured on a sub-set of individuals at baseline (629 participants in the ancillary study). Questions related to attempting weight loss and participating in weight loss programs and weight loss control practices related to food, physical activity, and weight control resource use were also examined. Examples of resources use include purchases of exercise equipment or weight loss program membership. For multivariate analysis purposes, individual questions were summarized to create separate scores for resource use related to physical activity, weight control, and food preparation. BMI was calculated using measured height and weight.

Statistical Analysis

In this analysis, the main independent variables were the neighborhood factors and the main dependent variables were individual-level weight-related variables from the Look AHEAD study. Descriptive statistics were used to describe the study population.

Multi-level analyses were used to analyze the aggregate and individual level data [32–34]. Recognizing that when studying group-level variables, individuals are nested within those groups, multi-level analyses are designed to account for this clustering. Specifically, the model building first identifies the most predictive set of individual-level variables. Then aggregate-level variables are added. At each level, all variables and their interaction effects are tested. Random effects terms are then added as additional parameters to account for extra area-level variability not explained by the model and included variables (overdispersion) [32].

In the current study, the association between neighborhood factors (% poverty, food availability for the census tract) and individual-level weight-related outcomes were determined while accounting for individual level SES (personal income and education). This enabled us to determine the independent effects of the neighborhood SES. Other potential confounders included in the models were: age, sex, and race. All analyses were conducted using STATA statistical software, version 9.

Results

Selected Baseline Characteristics of Study Participants

Selected baseline characteristics of the study participants are presented in Table 1. Participants were on average 59.5 ± 6.7 years of age, 41% male, and 27% were Black/African American. About half of participants had at least some college education; the majority of participants had annual incomes > $40,000. All participants were at least overweight or obese (BMI > 25 kg/m2),eligibility criteria for Look AHEAD.

Participant neighborhoods were diverse. Of all the neighborhoods represented in the study, the mean % of those living below the federal poverty level was 11%. Neighborhoods on average had 1.3 ± 1.5 food stores, 0.6 ± 0.8 convenience stores, and 6.8 ± 10.8 restaurants (fast food and non-fast food). Overall, there were 920 unique census tracts represented in the study; Baltimore = 201, New York= 257, Philadelphia = 245, Pittsburgh = 217. The number of participants per census tract ranged from 1-6.

With respect to dietary intake and eating patterns, most participants did not meet the recommended intakes for total fat, saturated fat, and fruit and vegetables. Most participants reported eating breakfast, lunch and dinner every day, and about a quarter reported eating those meals at a fast food restaurant more than once a week.

The vast majority of participants reported that they were currently attempting weight loss (96%) and 62% reported that they had participated in a weight loss program before the study. Many reported various weight control strategies such as cutting out sweets and junk food from their diets (63%), increasing fruit and vegetables (65%), and increasing exercise levels (55.3). With respect to resources spent on food, physical activity, and weight control, the most common were purchase of food preparation equipment (51%), indoor exercise equipment (33.8%), and class membership services for weight loss (33%).

Association between Neighborhood and Eating Patterns

Table 2 outlines the association between neighborhood and eating patterns. Participants living in neighborhoods with more restaurantswere significantly more likely to eat breakfast and lunch at restaurants that were not fast food restaurants compared to those living in neighborhoods with fewer restaurants. Furthermore, they were significantly more likely to eat dinner 7 days per week.

Table 2 Prevalence Ratios and 95% confidence intervals for Neighborhood Indicators and Eating Patterns among 1219 participants in the Look AHEAD Study
Full size table

Association between Neighborhood and Weight Loss Control Practices

Neighborhood SES had little association with weight control practices (see Table 3). Those who lived in neighborhoods with more restaurants were less likely to be attempting weight loss, and more likely to participate in weight loss control practices related to food and physical activity than those who lived in neighborhoods with fewer restaurants. There was no significant association seen for neighborhood and BMI.

Table 3 Analyses for Neighborhood Indicators, and Weight Control and BMI among 1219 participants in the Look AHEAD Study
Full size table

Association between Neighborhood and Resource Use

Those living in neighborhoods with more poverty were significantly more likely to have purchased food preparation equipment in the past year compared to those living in neighborhoods with the least poverty (PR = 1.36, see Table 4).

Table 4 Analyses for Neighborhood Indicators, Weight Control Practices and Resource Use
Full size table

Association between Neighborhood and Dietary Intake

Contrary to our hypothesis, those who lived in poorer neighborhoods had lower intakes of total and saturated fat compared to those living in wealthier neighborhoods (Table 5). This did not appear to be influenced by total caloric intake.

Table 5 Adjusted Prevalence Ratios, β coefficients and 95% confidence intervals for Neighborhood Indicators and Dietary Intake among 629 participants in the Look AHEAD Study
Full size table

Discussion

Our results suggest that among this group of overweight adults with type 2 diabetes in the Look AHEAD study that: 1) the presence of more restaurants in the neighborhood was associated with eating at non-fast food restaurants and with participation in several food and physical activity weight control practices; 2) neighborhood SES was only associated with a few of the weight-related factors. These conclusions are supported by a study with a diverse range of neighborhoods, detailed individual-level data, and a large percentage of minority participants.

Studies of neighborhood and health have generally focused on the physical built environment and its relation to physical activity[35–41]. Some studies have evaluated neighborhood and dietary patterns[42–44], and more recent studies have evaluated obesity or weight status as an outcome [45–50]. Few studies, to date, have evaluated neighborhood SES or other characteristics with weight control practices. Our study, which was conducted within a large-scale weight loss trial, had the strength of including a wealth of individual-level data on eating patterns, weight control practices, along with resource use for weight loss purposes.

There were, however, a few limitations. First, using the census tract as a proxy for neighborhood has been criticized, however, many studies have used this indicator, allowing us to compare our findings across studies. Furthermore, the wealth of data available from the US Census provides a comprehensive view of this geographic entity. Similarly, the neighborhood data may not have represented the entire baseline time-period for the Look AHEAD study. Data used were from the 2000 Census and 2004 Consumer database; Look AHEAD participants were recruited from 2001-2004. Neighborhoods are constantly changing, however the time-frame for the data used was close to the study recruitment period. Second, given the eligibility criteria for entry into the study, the population was fairly homogeneous with respect to some factors. One example was weight, which may explain why there was little variation of BMI status by neighborhood. This may explain many of our negative findings. In a future study, we plan to conduct longitudinal analyses and determine how neighborhood influences response to the weight loss intervention. The longitudinal analyses should show more variation in the dependent variables as individuals respond differently to the intervention. Furthermore, as this was an exploratory study, there were many negative findings that may have been due to low statistical power.

Conclusion

Future studies should evaluate neighborhood in relation to weight loss behaviors in other populations and further explore the impact of various aspects of the physical and social neighborhood environment. Now that individual-level correlates of healthy weight and weight loss are fairly well understood, attention should be given to other social and environmental determinants that may have a substantial impact. In addition to policy changes such as those that regulate the unhealthy selections in restaurants on an environmental level, incorporating teaching points on influences such as portion control and choosing health options in restaurants should contribute to more successful weight-loss interventions.