|Home | About | Journals | Submit | Contact Us | Français|
While neighborhood deprivation is associated with prevalence of chronic diseases, it is not well understood whether neighborhood deprivation is also associated with cardiometabolic risk factors among adults with chronic disease. Subjects (n=19,804) from the Diabetes Study of Northern California (DISTANCE) cohort study, an ethnically stratified, random sample of members of Kaiser Permanente Northern California (KPNC), an integrated managed care consortium, with type 2 diabetes who completed a survey between 2005 and 2007 and who lived in a 19 county study area were included in the analyses. We estimated the association between a validated neighborhood deprivation index (NDI) and four cardiometabolic risk factors: body mass index (BMI= kg/m2), glycosylated hemoglobin (A1c), low-density lipoproteins (LDL) and systolic blood pressure (SBP) using multilevel models. Outcomes were modeled in their continuous form and as binary indicators of poor control (severe obesity: BMI ≥35, poor glycemic control: A1c ≥9%, hypercholesterolemia: LDL ≥130 mg/dL, and hypertension: SBP ≥140 mmHg). BMI, A1c and SBP increased monotonically across quartiles of NDI (p<0.001 in each case); however, LDL was significantly associated with NDI only when comparing the most to the least deprived quartile. NDI remained significantly associated with BMI and A1c after adjusting for individual level factors including income and education. A linear trend (p<0.001) was observed in the relative risk ratios for dichotomous indicators of severe obesity, poor glycemic control, and 2 or more poorly controlled cardiometabolic risk factors across NDI quartile. In adjusted models, higher levels of neighborhood deprivation were positively associated with indicators of cardiometabolic risk among adults with diabetes, suggesting that neighborhood level deprivation may influence individual outcomes. However, longitudinal data are needed to test the causal direction of these relationships.
Cardiovascular disease is a major contributor to disease progression among adults with diabetes (Kannal & McGee, 1979; Koskinen, et al., 1992; Stevens, et al., 2001), and one in which there are pronounced health disparities (Duru, et al., 2009; Shah, Dolam, Gao, Kimball, & Urbina, 2011). In cross-sectional studies among adults in a number of countries, neighborhood deprivation has been consistently and positively associated with prevalence of cardiovascular disease (e.g., Mobley, et al., 2006; Cubbin, et al., 2006; Sundquist, Malmstrom, & Johansson, 2004; Sundquist, Winkleby, Ahlen, & Johansson, 2004; Diez-Roux, et al., 2001) and obesity (e.g., Black, Macinko, Dixon, & Fryer 2010; Grafova, Freedman, Kumar, & Rogowski, 2008; Matheson, Moineddin, & Glazier, 2008; Dragano, et al., 2007; Inagami, Cohen, Finch, & Asch, 2006; Mujahid, Diez Roux, Borrell, & Nieto, 2005); it has also been associated with increased incidence of diabetes (Cox, Boyle, Davey, & Morris, 2007; Cox, Boyle, Davey, Feng, & Morris, 2007), and with insulin resistance among healthy, young adults in the US (Auchincloss, Diez Roux, Brown, O'Meara, & Raghunathan, 2007; Diez Roux, Jacobs, & Kiefe, 2002). A causal relationship between neighborhood deprivation and cardiovascular disease has not been well established, particularly among patients with diabetes (Leal & Chaix, 2010). In this study, we sought to better understand the extent to which neighborhood deprivation is independently associated with specific modifiable risk factors for cardiovascular disease progression among adults with diabetes in the US.
Neighborhood deprivation may produce negative health effects such as increased cardiovascular disease, obesity, and diabetes through several pathways (Diez Roux & Mair, 2010, Cox, et al. 2007b). Briefly reviewed here, a paucity of resources in more deprived neighborhoods impede engagement in healthy behaviors (Macintyre, 2007), such as acquiring nutritious food and engaging in physical activity (Rundle, et al., 2009). Furthermore, limited access to nutritious foods because of household food insecurity, the inability to access nutritious foods in socially acceptable ways because of a lack of money to purchase foods, is significantly associated with diabetes prevalence (Seligman, Bindman, Vittinghoff, Kanaya, & Kushel, 2007) as well as poor diabetes management evidence by higher risk of elevated glycosylated hemoglobin (Seligman, Laraia, & Kushel, 2010). Second, more deprived neighborhoods have higher crime rates and physical incivilities—physical manifestations of neighborhood degradation such as litter, unkempt property, disheveled public space—which may increase threat and induce a stress response (Sundquist, et al., 2006). Stressful events, perceived as threat, are associated with non-homeostatic eating (eating for reasons other than caloric need), visceral fat accumulation and weight gain in animal and human research (Adam & Epel, 2007). Stress induced eating is associated with ingestion of highly palatable food—foods high in fat and refined sugars (Epel, et al., 2004; Epel, Lapidus, McEwen, & Brownell, 2001). Third, one's residential neighborhood setting may provide cues that support social norms governing smoking, inactivity, and poor diet (Stimpson, Ju, Raji, & Eschbach, 2007) that in turn influence individuals' health behaviors. In more deprived neighborhoods, billboards and other cues may undermine healthy behaviors. Since human interactions with neighborhood environments are complex, it is likely that more than one of these pathways is involved in promoting or impeding health. Neighborhoods may affect both the risk of developing a chronic disease and affect the course of management. This is especially true for diseases whose management is strongly impacted by health behavior. Intervention programs and policies may be more effective if they target not only the individual but also the neighborhood environments that shape individual risk.
The prevalence of diabetes continues to increase in the US with current estimates of 7.8% - 9.0% of the population affected (Centers for Disease Control and Prevention, 2008; Cowie, et al., 2006). Although family history, race/ethnicity, and socioeconomic status (SES) are predictors of diabetes, behavioral factors including weight gain, poor dietary intake, and sedentary life style are important modifiable risk factors that also contribute to the development of diabetes. Each of these modifiable factors has been associated with neighborhood deprivation (e.g., Cubbin & Winkleby, 2007; Giskes, van Lenthe, Turrell, Brug, & Mackenbach, 2006; Stimpson, Ju, et al., 2007; Stimpson, Nash, Ju, & Eschbach, 2007). Engaging in healthy behaviors is a key defense against disease progression for those diagnosed with diabetes. Diagnosed patients who live in a deprived neighborhood encounter the same resource constraints, stressors, and social norms that may have contributed to their risk for developing diabetes in the first place (Cox, et al., 2007a). Neighborhood deprivation may underlie disparities in diabetes prevalence and severity associated with social status and race/ethnicity. Lower SES individuals and ethnic minorities are more likely to live in deprived neighborhoods and are more exposed to environments that affect risk factors as described above. Identifying and modifying exposures that perpetuate diabetes-related disparities is a critical research and public health priority and understanding the role of neighborhood disparities should help in these effort.
To understand the association between neighborhood deprivation and cardiometabolic risk factors, we studied a large, ethnically-diverse cohort of adults from all socioeconomic levels with diabetes included in The Diabetes Study of Northern California (DISTANCE). In these analyses, confounding of the association between neighborhood deprivation and cardiometabolic risk factors with being uninsured was minimized since all respondents were insured members of a non-profit, health delivery organization (Karter, et al., 2002; Martin, Selby, & Zhang, 1995). We hypothesized that cardiometabolic risk factors were more likely to be poorly controlled in deprived neighborhoods. The aim of the present study was to estimate the associations between neighborhood deprivation and body mass index (BMI), glycosylated hemoglobin (A1c), low density lipoproteins (LDL) and systolic blood pressure (SBP), independent of individual-level demographic and socioeconomics characteristics, among a sample of insured adults with diabetes.
Kaiser Permanente Northern California (KPNC) is a non-profit, group practice, health plan that is one of the largest and oldest managed care organizations in the United States. KPNC currently provides comprehensive medical services to over 3.3 million health plan members through 17 hospitals and 32 outpatient clinics located in a 15-county region of Northern California. Approximately 30% of the general population in the region are members of KPNC. The sociodemographic characteristics of KPNC members are generally representative of the overall population, other than for income, where the very poor and very wealthy are enrolled but somewhat under-represented (Gordon, 2006; Karter, et al., 2002).
KPNC maintains the Kaiser Permanente Northern California Diabetes Registry (Registry) that included 227,421 active members with diabetes (as of 1/1/2008). The Registry has been updated on an annual basis continuously since 1993 and has been the basis for extensive epidemiologic and health services research (Karter, et al., 2002; Karter, et al., 2005; Selby, Karter, Ackerson, Ferrara, & Liu, 2001). At the time of the study, registry eligibility was based on multiple data sources of case ascertainment, including pharmacy prescription records (diabetes medications dispensed), laboratory test results (A1c ≥ 7% and/or ≥2 elevated fasting glucoses), and outpatient, emergency room and hospitalization records listing a primary diagnosis of diabetes. Sensitivity is about 99% based on a survey-based gold standard of self-reported diabetes. Turnover is very low with an average registry membership period of about 9 years.
The DISTANCE survey was conducted between 2005 and 2007 in an ethnically-stratified random sample of KPNC members who were Diabetes Registry members (n = 40,735) with approximately equal samples from the 5 largest ethnic groups (African American, Chinese, Filipino, Latino and White). A total of 20,188 people responded to the survey, giving a survey response rate of 62%, after accounting for eligibility and people who were unable to be contacted (The American Association for Public Opinion, 2008). A more complete description of the study methods, cohort and survey has been published previously (Moffet, et al., 2009). The inclusion criteria for this analysis included: established DISTANCE study participant with Type 2 diabetes (based on a typing algorithm that combine age (over 20 years), weight status and medication that was published as an appendix to the following reference: (Karter, et al., 2001) and a geocodable address within a 19 county study area. The study area included 15 counties that have a Kaiser facility and an additional four adjacent counties with more than 30 DISTANCE respondents (a total sample of 19,804). This study was approved by the Institutional Review Boards of Kaiser Division of Research and the University of California, San Francisco School of Medicine.
Based on previous research (Messer, et al., 2006), we created a neighborhood deprivation index (NDI) using 2000 US Census of Housing and Population data. The 2007 home address data for each of the 19,804 eligible DISTANCE survey respondents was geo-referenced, assigning the address to the block face level (street segment) using MapMarker (Version 11) geocoding software. The NDI was created at the census tract level for 2,250 tracts within the 19 counties using principal components analysis to estimate the total variance in preference to factor analysis which estimates the shared variance among identified variables. We selected census variables that are thought to contribute to an area's overall or total variance of deprivation based on six domains—income, poverty, housing, education, employment and occupation. Based on our previous work of contextual indicators associated with adverse birth outcomes (Elo, et al., 2009; Holzman, et al., 2009; O'Campo, et al., 2008), we used 8 census-derived variables to measure the six domains: percentage of households below the 2000 income to poverty ratio, percentage of households on public assistance, percentage of female-headed households with dependent children, percentage of households with annual income < $30,000 per year, percentage of adults not completing high school, percentage living in crowded housing (> 1 person/room), percentage of unemployed adults, and percentage of males in management or professional occupations. Although the NDI has been previously applied in studies of pregnancy outcomes, it was developed based on the literature assessing neighborhood-level attributes and health regardless of the outcome, and was recently used in an analysis and publication associated with mortality outcomes (Doubeni, et al., 2011). The NDI was based on all census tracts in the 19 county study region, regardless of whether a Kaiser member lived in the census tract. The NDI was standardized with a mean of 0 and standard deviation of 1; the more negative the score, the less deprived the neighborhood; the more positive the score, the more deprived. We determined quartile cutpoints (Q1-Q4) from the continuous NDI measure for all census tracts, with Q1 being the least and Q4 being the most deprived neighborhoods (census tracts).
We examined four cardiometabolic risk factors and indicators of clinical control (body mass index, glycosylated hemoglobin, low density lipoproteins and systolic blood pressure) in relation to neighborhood deprivation. All clinical measures were ascertained during the one-year prior to the survey date. Body mass index (BMI, in kg/m2) was calculated using a clinical measurement of weight and height recorded during an outpatient visit within one year prior to or after administration of the survey (n=18,413). For individuals with no measured weight and height in the year prior to the survey, we used the self-reported weight and height from the survey (n=884). Correlations between administrative and self-reported BMI were ≥ 0.89 among those who had both measures, and the average clinically measured BMI was nearly 1 unit greater than self-report. We created and tested a BMI flag variable indicating the source of data (clinical vs. self-report). While the BMI source flag was significant in a few models, it introduced no substantive changes to the coefficients on the NDI quartiles, so we did not include it in the models. Glycosylated hemoglobin (A1c, an integrated measure of blood glucose control over approximately 3 months), and low-density lipoprotein cholesterol (LDL, mg/dL, a measure of cholesterol) were analyzed by a single, central KPNC laboratory and downloaded from electronic medical records. Systolic blood pressure (SBP) measures were available from two electronic data sources, one of which recorded SBP as a continuous measure (n = 3,384) in the year prior to the survey date, and these values were used preferentially. The second, earlier source recorded narrow ranges of measures in a categorical structure with 10 point increments. For the latter, we used the mid-point value in the range. While the SBP source flag was significant in some models, it introduced no substantive changes to the coefficients on the NDI quartiles, so we didn't include it in the models. Finally, we dichotomized each cardiometabolic outcome at well-accepted clinical thresholds to demarcate poor control: severe obesity (≥ 35 kg/m2), A1c (≥ 9%), LDL (≥ 130 mg/dL) and SBP (≥ 140 mmHg). We also created a dichotomous variable indicating that two or more of the above outcomes were poorly controlled for an individual.
We identified the following potential confounders that were associated with both the exposure and outcomes, but that were not on the pathway between the exposure and outcomes: age (categorized as 30-51, 52-64, >65 years) (Tabassum, Breeze, & Kumari, 2010), sex, race/ethnicity (defined as African American, Latino, Asian, or other, compared to White non-Latino), marital status (defined as divorced/separated, widowed, never married, refused/don't know/missing, compared to being married/living together) and time lived in the US (defined as 1/3 to 2/3 of life in US, >2/3 of life in US, born in the US compared to living less than 1/3 of life in US). Objective socioeconomic indicators included: education (defined as no high school degree, high school/GED/technical school, associate degree, or college graduate, compared to post graduate education) and family-level income/poverty line (categorical > 600%, 301-600%, and 101-300% compared with 0-100% of poverty line). We defined income as self-reported family income divided by the poverty line income for a given age and household size based on the US Department of Health and Human Services 2005 Poverty Guidelines (Department of Health and Human Services, 2005). As a subjective social status indicator, we used the 10-step MacArthur Subjective Social Status scale (“social ladder”) which is designed to measure relative standing within society in a way that spans socioeconomic indicators (Adler, Epel, Castellazzo, & Ickovics, 2000; Adler & Newman, 2002). The social ladder has been associated with poor self-rated health, higher mortality, cardiovascular risk, diabetes and respiratory illness in previous studies (Adler, et al., 2000; Kopp, Skrabski, Rethelyi, Kawachi, & Adler, 2004; Singh-Manoux, Marmot, & Adler, 2005). A missing indicator was included for each covariate to account for unavailable data.
Principal components analysis was used to estimate the amount of the total variance in the underlying construct NDI explained by the eight items. Component loadings ranged from 0.29 to 0.39, suggesting that each item loaded almost equally onto the first component, and were used to weight the contribution of each item to the overall index (Messer, et al., 2006). Cronbach alpha was used to test internal reliability of the NDI. The continuous NDI measure was then regressed on individual-level socioeconomic indicators of income/poverty line, education and marital status to assess the level of overlap between individual and neighborhood socioeconomic status. We calculated the intra-class correlation coefficient (ICC) to assess the proportion of total variance in the outcomes of interest associated with NDI quartile. We estimated the unadjusted and covariate adjusted values for continuous BMI, A1c, LDL and SBP by quartile of neighborhood deprivation, Q1 being the least and Q4 being the most deprived neighborhoods, using a random effects model that included expansion weights by race/ethnicity strata (due to non-proportional sampling fractions). For dichotomous outcomes, we used a series of modified fixed Poisson models, adjusting for the same individual covariates, to estimate the relative risks, between NDI quartiles and an outcome variables indicating whether the individual had poor control for each of the four cardiometabolic risk factors, or had two or more risk factors poorly controlled. (Barros & Hirakata, 2003; Zou, 2004). All models were restricted to the total number of individuals who had the outcome variable, therefore the sample size for the regression models varied for each outcome. We also conducted a sensitivity analysis restricting the sample to complete cases, meaning that respondents had a value for all four outcomes (n=16,129). Data were managed and analyzed using Stata, version 10.1 using the commands XTREG for continuous and GLM for dichotomous multivariate fixed effects analysis (Rabe-Hesketh, Skrondal, & Pickles, 2005; Rabe-Hesketh, Skrondal, & Pickles, 2002).
The principal components analysis showed that the first component explained 67% of the total variance among the eight items, suggesting that together they relate to the underlying construct of neighborhood deprivation. The items that comprise the neighborhood deprivation index (NDI) were highly correlated and had excellent internal reliability (Cronbach alpha = 0.93). Individual-level indicators of socioeconomic status—income, education and marital status— explained only 10% of the variance of the NDI (R2=0.10) suggesting that the NDI is a distinct construct from individual level socioeconomic status and may serve as an adequate proxy for neighborhood deprivation.
Demographics and outcome values for the analysis sample are presented in Table 1. A significantly (p<0.05) higher percent of missing outcome variables was found for men (BMI, A1c, SBP), younger age adults (A1c, LDL, SBP), those who reported unknown race (BMI), unknown income (BMI) or higher income (A1c, LDL, SBP), and lower and unknown education (BMI) or who held associates and high school degrees (LDL). No difference was found in the percent missing for A1c or SBP by education. Asians were least likely to have missing outcome values for A1c, LDL. The range of percent missing was most often within three percentage points, but never higher than five percentage points. The mean BMI for individuals residing in the 19 county study area was above 30, a value that meets the definition for obesity (NHLBI 1998). The mean values for A1c, LDL and SBP fell within the range suggestive of being in relatively good clinical control. The mean values for the four cardiometabolic risk factors, the percentage of participants above the clinical cutpoint for each risk factor, and the percentage with two or more risk factors were all significantly higher for NDI Q4, the most deprived quartile, compared to NDI Q1, the least deprived.
Based on the intra-class correlation coefficient (ICC) from the multi-level models, a modest amount of between group variance at the neighborhood level was found for BMI (5.5%) and LDL (1.8%), there was no variance at the neighborhood level for A1c or SBP. In unadjusted models accounting for the sampling design and possible clustering by census tract, NDI quartiles were significantly associated with all four cardiometabolic risk factors (BMI, A1c, LDL, SBP) (Table 2). In the crude models, the association between NDI quartiles and BMI, A1c and SBP, but not LDL, was monotonic. The results from the unadjusted regression models can be interpreted such that the intercept is the mean value for Q1 and the point estimate for each higher quartile is the change from the constant. The values for BMI and A1c were higher with each quartile, and all quartile were significantly greater than Q1. For SBP, Q3 and Q4 were significantly greater than Q1. The association between NDI quartiles and LDL was significant only when comparing the least deprived quartile to the most deprived quartile (β: 2.48, 95% CI: 0.98, 3.97). The test for trend was significant for all unadjusted models.
In models adjusted for the above mentioned individual level SES and demographic indicators, each NDI quartile was significantly associated with higher BMI (β: 0.54 to 1.06) and levels of A1c (β: 0.08 to 0.18), but only marginally associated in Q3 and Q4 with higher systolic blood pressure. Furthermore, the test for trend was significant across NDI quartile for BMI, A1c and SBP (Table 2). The association between the highest quartile of NDI and LDL was attenuated and no longer significant after adjustment for individual characteristics. Inclusion of NDI quartiles reduced the between group variance for BMI from 5.5% to 3.1%. In the fully adjusted model, individual level factors further reduced the neighborhood level variance to 0, suggesting that all of the between group variance in BMI was accounted for by both neighborhood and individual factors. Sensitivity analysis, restricting the cohort to complete cases, yielded very similar point estimates and confidence intervals for BMI, A1c and LDL models, but attenuated the marginal results for SBP with no significant trend (data not shown). This suggests that the magnitude of association and precision of the estimate between quartiles of NDI and BMI, A1c and LDL were internally reliable and were not influenced by the using a dataset restricted to only those who had information on all four outcomes. SBP, however, had a weak and marginal association with NDI which became insignificant and attenuated with the dataset that was restricted to only those adults with complete data for all outcomes.
We used modified fixed Poisson models to estimate the relative risks between NDI quartiles and dichotomous indicators of poor control for each of the four cardiometabolic risk factors, plus having two or more risk factors in poor control, adjusting for the same covariates.
The adjusted relative risk ratio for severe obesity (BMI ≥ 35) increased significantly for each increment in NDI quartile compared to NDI Q1; a 22% greater risk at the fourth quartile of NDI compared to the first (RRR: 1.22, 95% CI: 1.12, 1.31). The adjusted relative risk ratio for poor glycemic control (A1c ≥ 9%) was significantly higher (RRR: 1.33, 95% CI: 1.16, 1.52), as was the adjusted relative risk for having two or more poorly controlled risk factors (RRR: 1.23, 95% CI: 1.10, 1.37) when comparing the least to the most deprived quartile (See Table 3). There was also a significant trend (p < 0.001) in the percentage of adults with poor control for BMI, A1c and having two or more cardiometabolic risk factors in poor control with higher NDI quartiles. Sensitivity analysis, restricting the cohort to complete cases, yielded very similar point estimates and confidence intervals for all models, leaving the inferences unchanged (data not shown).
This study contributes to the wider literature on links between neighborhood deprivation and health by finding associations of neighborhood deprivation with poor cardiovascular risk factor control in a population of adults with diabetes who are in managed care, after extensively adjusting for several individual socioeconomic and demographic (SES) characteristics. We stress two important aspects of our findings, 1) NDI quartiles were associated with differences in diet sensitive cardiometabolic factors independent of a number of individual level SES indicators, and 2) the NDI quartiles appeared to have excellent statistical reliability and validity as a measure in California. In models adjusted for individual SES indicators, comparing the least deprived to the most deprived neighborhood was associated with a 1 unit higher BMI value and about 0.2 % point higher A1c. While there were modest effect sizes for each risk factor individually, the cumulative effect of poor control for multiple cardiometabolic factors deserves consideration. Evaluating summary statistics (e.g., mean A1c at each quartile of NDI) may mask the effect of neighborhood deprivation on the most vulnerable individuals. Findings were more substantive when evaluating indicators of poor (vs. adequate) control; there was a significant and substantive trend of higher risk of poorly controlled cardiometabolic risk factors with higher levels of neighborhood deprivation. The adjusted relative risk ratio for residing in a census tract in the highest quartile of NDI was significantly associated 22% greater risk for severe obesity, 33% greater risk for poorly controlled A1c (>9%), and 23% greater risk for having two or more poorly controlled cardiometabolic risk factors compared to adults residing in census tracts from the lowest NDI quartile. These data point out the relationship between neighborhood deprivation and high risk clinical profiles. Importantly, given the consistently poorer control of these four important risk factors, the cumulative neighborhood impact on health may be substantial.
It is interesting to note the relationship between NDI and risk factors that are most influenced by behaviors and hardest to control pharmacologically (BMI and A1c) were stronger than the relationship between NDI and risk factors that are less easily modified by behaviors and more easily controlled pharmacologically (LDL and BP). This suggests that the association between neighborhood and BMI and A1c may be mediated through health behaviors, such as diet and physical activity, rather than clinical factors including medication adherence. Furthermore, poor control of these clinical factors place people at risk for complications of diabetes, such as heart disease, stroke, lower extremity amputations, blindness and kidney failure.
In the international realm there is a long history of using a composite indicator to measure area level disadvantage or deprivation. With regard to studies focused on adults, examples include a number of studies from Canada (Matheson, Moineddin, & Glazier, 2008), the United Kingdom (Stafford, Brunner, Head, & Ross, 2010), Sweden (Sundquist, Malmstrom, et al., 2004; Sundquist, Winkleby, et al., 2004), and Australia (Adams, et al., 2009) that have found area-level deprivation associated with body mass index, especially for women. Previously in the US, the NDI was found to be an internally reliable measure of socioeconomic disadvantage and significantly predictive of adverse birth (Elo, et al., 2009; O'Campo, et al., 2008) and mortality outcomes (Doubeni, et al., 2011). We found similar factor loadings and variance explained, despite using populations from a different geographical region (Northern California). Conceptually, neighborhood deprivation captures a range of socioeconomic conditions faced by individuals living within a given neighborhood. More than an aggregate of individual level poverty or wealth, NDI is a mechanistic construct comprised of a number of domains that contribute to area-level socioeconomic disadvantage. In this study, individual socioeconomic indicators only explained 10% of the variance in neighborhood deprivation. In addition to area level income and wealth, we and others have shown the importance of including education, employment, occupation and housing conditions in its construction (Messer, et al., 2006; Singh & Siahpush, 2002; Winkleby, Sundquist, & Cubbin, 2007).
Our results are consistent with the results of several studies from the US, Canada, Australia and Sweden, indicating an association between neighborhood deprivation and BMI among adults (e.g., Grafova, et al., 2008; Cubbin, et al., 2006; Inagami, et al. 2006; Mujahid, et al., 2005) and cardiometabolic risk factors (e.g., Dragano, et al., 2007; Mobley, et al., 2006). As hypothesized, higher levels of neighborhood deprivation were associated with higher levels of cardiometabolic risk among adults with diabetes, similar to the association between neighborhood deprivation and diabetes prevalence (Cox, et al., 2007a; Cox, et al., 2007b). In our adjusted models, increased neighborhood deprivation remained significantly associated with higher levels of BMI and A1c, both of which are sensitive to a number of health behaviors such as diet, physical activity and smoking. Previous US studies have found neighborhood deprivation and distance from wealthy areas associated with insulin resistance among adults without prevalent chronic disease (Auchincloss, et al., 2007; Diez Roux, et al., 2002), and have found neighborhood deprivation and poor eating environments associated with worse dietary patterns and higher BMI among African Americans with diabetes (Millstein, Yeh, Brancati, Batts-Turner, & Gary, 2009).
Individuals living in deprived neighborhoods may be at higher risk for poor health outcomes because of fewer area-level resources and employment opportunities, more stress or more environmental health contaminants (Adler, 2006). Living in a deprived neighborhood may also accelerate the progression of existing disease and lead to poorer disease-related outcomes. In population-based studies, deprived neighborhoods are often associated with decreased access to and quality of health care, factors that amplify residents' risk for poor health outcomes. Our study is unusual in that all subjects were all insured and had relatively uniform access to quality health care conferred by their membership in an integrated health delivery system. Thus concerns in the US context regarding confounding due to insurance membership status or variations in quality of care due to a variety of insurance providers, variables that are difficult to quantify, were minimized. For that reason these findings should provide a nuanced view of the association of neighborhood deprivation with cardiometabolic risk factors that can complement existing population based studies.
Although we believe the findings of this study are important from both clinical and public health perspectives, there are some limitations. Our analysis was cross-sectional and causal inferences cannot be drawn from the results. Furthermore, the sample for this study was from a managed care population of adults with diabetes and may not be directly generalizable to uninsured populations. Selection bias from the modest response rate of 62% and from missing outcomes may have biased our findings since participants with poorer health outcomes and deprived neighborhoods were more likely to have declined to participate. The different survey modes (telephone interview, mailed questionnaire, or web questionnaire) may also variably influence participant responses. Moreover, our results likely provide a conservative picture of the association between neighborhood deprivation and indicators of clinical risk expected for the general population where access to and quality of care may covary with neighborhood deprivation. While the majority (92%) of US citizens diagnosed with diabetes are insured (Harris, Cowie, & Eastman, 1994), adult Latinos with diabetes are over-represented among the uninsured. Also, while exposures (neighborhood deprivation) and outcomes (e.g., BMI) may differ for managed care members with diabetes compared to uninsured patients or those insured by different organizations, observed relationships between exposures and outcomes are typically much less variable. Finally, we used census tract as a proxy for neighborhood boundaries, which may not reflect an individual's actual neighborhood. We also did not measure neighborhood specific resources (e.g., distance to supermarket, density and type of food outlets, etc.) that may be more proximal causes of poor cardiometabolic risk control.
In summary, the fact that (a) we studied a diabetic cohort with relatively uniform access to and quality of care and (b) the neighborhood deprivation index was operationalized at census tract level (rather than contextually-defined neighborhoods), suggests that these findings are likely conservative, underestimating the true association detected when using more meaningful, finer-scaled boundaries of neighborhoods or activity spaces.
We found that the neighborhood deprivation index (NDI) had excellent internal reliability across 19 Northern California counties, based on the Cronbach's alpha, factor loadings and percent variance explained. Furthermore, individual-level indicators of socioeconomic status explained only 10% of the variance in neighborhood deprivation. While our primary goal was to identify whether there was a neighborhood-level association with cardiometabolic risk independent of (adjusted for) individual characteristics, from the public health policy, assessment and intervention standpoint, the unadjusted results are equally important. Vulnerable populations cluster in deprived neighborhoods, making targeted community interventions potentially efficient and effective. The fact that the association between neighborhood deprivation and cardiometabolic risk factors, namely higher levels of BMI and A1c, persisted independent of the characteristics of the individuals who live in the neighborhood provides stronger evidence. If health care providers and educators counsel patients with diabetes to improve their diet, increase physical activity and decrease stress without an appreciation of the contextual influences of life in low-resourced communities, these encouragements run the risk of being less effective. These findings warrant further exploration into modifiable factors that might mediate the relationship between neighborhood deprivation and cardiometabolic risk factors, followed by testing interventions that in some way act on these health-related neighborhood-level mediators.
This study was supported by the NIH National Institute of Diabetes and Digestion and Kidney study “Ethnic disparities in diabetes complications” (PI Andrew Karter, R01 DK065664-01-A1); “Neighborhood Effects on Weight Change and Diabetes Risk Factors (PI Barbara Laraia, R01 DK080744); and the NIH National Institute of Child Health and Human Development study “Socioenvironmental Influences on Nutrition and Obesity (PI Barbara Laraia, K01 HD047122).
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Barbara A. Laraia, Department of Medicine, Division of Prevention Sciences; Center for Health & Community, University of California, San Francisco, CA.
Andrew J. Karter, Kaiser Permanente Division of Research, Oakland, CA; Department of Epidemiology and Health Services, School of Public Health & Community Medicine, University of Washington, Seattle, WA.
E. Margaret Warton, Kaiser Permanente Division of Research, Oakland, CA.
Dean Schillinger, Department of Medicine, Center for Vulnerable Populations, University of California, San Francisco, CA; California Diabetes Program, UCSF Institute for Health and Aging and California Department of Public Health.
Howard H. Moffet, Kaiser Permanente Division of Research, Oakland, CA.
Nancy Adler, Department of Psychiatry; Center for Health & Community, University of California, San Francisco, CA.