Neighborhood Characteristics and Components of the Insulin Resistance Syndrome in Young Adults: The Coronary Artery Risk Development in Young Adults (CARDIA) Study

CARDIA is a cohort study of cardiovascular risk factors in young adults. A total of 5,115 adults aged 18–30 years at baseline were recruited primarily through telephone contact from community lists in Birmingham, Alabama; Chicago, Illinois; and Minneapolis, Minnesota and from membership in a prepaid health plan in Oakland, California (11). The goal of recruitment was to obtain at each field center nearly equal numbers of blacks and whites, women and men, persons <25 and ≥25 years of age, and persons with a high-school education or less and more than a high school education. Baseline interviews and examinations were conducted in 1985–1986. Follow-up examinations were conducted every 2–3 years. Overall retention of the cohort at the year 10 examination (1995–1996), on which these analyses are based, was 79%. Year 10 data were used because earlier addresses for the full cohort were not available for geocoding.

Six components of the IRS were investigated as outcomes in these analyses: BMI; fasting plasma HDL cholesterol, triglycerides, insulin, and glucose; and systolic blood pressure. BMI was estimated as weight (in kilograms) divided by height in meters squared. Body weight and height were measured with the participant in light clothing and without shoes. HDL cholesterol was measured using an enzymatic method after dextran sulfate-magnesium precipitation (12), and triglyceride levels were determined using enzymatic procedures (13). Fasting insulin was measured by radioimmunoassay (14). Fasting glucose was measured by the hexokinase-ultraviolet method. Blood pressure was measured on the right arm in a sitting position after a 5-min rest using a random-zero sphygmomanometer. The mean of the second and third measurements was used in the analyses.

The six components were investigated separately as well as combined into a summary IRS score. For each variable, a Z score was calculated by subtracting the sample mean and dividing by the sample standard deviation. Triglycerides and insulin were log transformed before the calculation of Z scores because of their skewed distribution. Z scores for the six variables were then summed and divided by 6 to obtain an average Z score (HDL cholesterol Z scores were multiplied by −1 before summing). The average Z score was then divided by its standard deviation to obtain a standardized normal deviate score. This score (the IRS score) was the outcome variable investigated in analyses. An increasing score signifies increasing levels of components of the insulin resistance syndrome. The absolute value of the score reflects deviations from the mean score in standard deviation units. For example, a score of 1 means that the average Z score of the six components is 1 standard deviation above the average score for the sample as a whole. Values of the IRS score may be transformed into an average percentile ranking of the six variables using a table of the standard normal distribution. The use of this summary score allows investigation of how exposures are related to the syndrome as a single continuous variable and affords more power to detect associations (especially in relatively young populations) than the use of arbitrary cut-off points. The use of summary scores is appropriate when dealing with multiple intercorrelated variables. These six factors have been shown to be correlated with each other in other samples (15) and in CARDIA participants (16).

Census-defined block groups (mean population 1,000) were used as proxies for neighborhoods (17). Study participants were linked to their census-defined area of residence using their year 10 home address. Six area variables collected by the 1990 U.S. Census reflecting the dimensions of wealth/income, education, and occupation were investigated (Table 1). The 1990 U.S. Census currently provides the most appropriate socioeconomic data available to characterize residential areas at the time of the year 10 CARDIA visit. For each of the census variables, a Z score was estimated by subtracting the mean for the total sample of block-groups and dividing by the standard deviation. These six variables were then combined into a neighborhood summary score constructed by summing Z scores for each of the six variables, with increasing neighborhood score signifying increasing neighborhood advantage (18). Neighborhood scores for the year 10 CARDIA sample used in these analyses ranged from −11.4 to 17.4. Baseline residence information was available for 96% of the Chicago sample; the correlation between baseline and year 10 neighborhood socioeconomic score was high (Pearson’s correlation 0.70, P < 0.0001). Information on personal income and education was reported by study participants as part of the year 10 exam and categorized as shown in Table 1. Categories for income and block-group score differed for blacks and whites because of large differences in the distribution of these variables in both groups.

Of the CARDIA participants who attended the year 10 follow-up, 89% (3,531) were matched to block-group data. An additional 181 participants were excluded for missing information on income, education, BMI, HDL cholesterol, triglycerides, fasting insulin, fasting glucose, or systolic blood pressure. An additional 257 individuals were excluded because they were diabetic (on diabetes medication or with fasting glucose ≥126 mg/dl and not pregnant) or had not been fasting at least 8 h. The total of 3,093 participants included in these analyses resided in 2,260 different block groups dispersed in 45 U.S. states. Seventy-nine percent of the block groups had only one study participant, and only 2% had five or more participants.

We first estimated age-adjusted means for the summary IRS score and its components by race-specific quartiles of neighborhood score. Linear regression was then used to estimate mean differences in the IRS score by neighborhood characteristics after adjustment for personal income and education. Before adjustment, however, we used stratified analyses to determine whether associations of personal income and education with the IRS score were similar across neighborhood categories (i.e., whether there was interaction between neighborhood and personal socioeconomic indicators). These interactions were tested by including appropriate interactions in regression models. Final estimates of neighborhood differences after controlling for personal income and education were obtained from models including neighborhood characteristics, personal income and education, and their interactions when appropriate. Neighborhood characteristics and personal socioeconomic indicators were investigated as both categorical and continuous variables. The use of socioeconomic categories as continuous variables allows estimation of the “average” difference in the outcome across categories. As the vast majority of block-groups had only one study participant, no special methods were needed to account for within-neighborhood correlations in outcomes. Adjustment for original study site did not modify the results.

The mean IRS score was similar in white and black men but was higher in black women than in white women (P < 0.0001) (Table 1). Black participants were more likely to be of lower income and educational attainment than white participants and were also more likely to live in neighborhoods with lower socioeconomic scores (Table 1).

The mean IRS score was inversely associated with neighborhood socioeconomic characteristics in white men and women and in black women (Table 2). For the most part, patterns were consistent across IRS score components. The strongest and most consistent associations were observed in white women. In black men, neighborhood characteristics were not clearly associated with neighborhood score. In fact, patterns for HDL cholesterol, triglycerides, and fasting glucose were the opposite of those observed in the other three groups: HDL cholesterol tended to decrease, and triglycerides and fasting glucose generally increased, with increasing neighborhood score.

In white men and women, increasing personal income and education were associated with decreasing IRS score across categories of neighborhood score (Table 3). In black men, however, personal income and education were inversely associated with the IRS score in the “better-off” neighborhoods but positively associated with the IRS score in the “worse-off” neighborhoods. In black women, the inverse associations of income and education with the IRS score were consistently weaker in “worse-off” than in “better-off” neighborhoods (Table 3).

In white men and women, neighborhood score remained inversely associated with IRS score after adjustment for personal income and education (mean differences in IRS score ± SE per unit increase in neighborhood score: −0.016 ± 0.008 in men and −0.038 ± 0.007 in women). When extrapolated to the 95th percentile of neighborhood score compared to the 5th percentile, differences were −0.24 standard deviation units (SE 0.12) in men and −0.56 standard deviation units (SE 0.10) in women. Patterns were weaker and less consistent in men than in women (Table 4).

In black participants, mean differences by neighborhood characteristics are shown stratified by income and education because (as shown in Table 3) there was evidence of interaction between neighborhood characteristics and personal socioeconomic indicators. For illustrative purposes, associations of neighborhood score with the IRS score are shown for extreme categories of income and education. In men, neighborhood score was positively associated with IRS score in men in the lowest categories of income and education (mean difference 0.048 ± 0.017 per unit increase in neighborhood score, 0.60 ± 0.21 standard deviation units when extrapolated to 95th vs. 5th percentile of neighborhood score) but negatively associated with IRS score in the highest categories of income and education (mean difference −0.047 ± 0.023 per unit increase and −0.54 ± 0.26 standard deviation units for 95th vs. 5th percentile). In black women, no association between neighborhood score and IRS score was observed in the lower income and education categories (mean difference 0.00 ± 0.16), but an inverse association was apparent in the higher income and education categories (mean difference per unit increase −0.037 ± 0.018, −0.52 ± 0.25 standard deviation units for 95th vs. 5th percentile). All results were similar when BMI was replaced by waist-to-hip ratio in the IRS score. Dropping individual items from the score one at a time did not substantially modify results.

Among young white men and women, the summary IRS score was inversely related to neighborhood socioeconomic score. There was no evidence that associations differed in persons of high or low income or education. In white women, neighborhood score remained inversely and significantly associated with the summary IRS score after controlling for personal income and education. Strikingly different patterns were observed in young black men: neighborhood score was inversely associated with the IRS score in black men of relatively high income and education but positively associated with the IRS score in black men of relatively low income and education. Similar interactions appeared to be present in black women, although differences were not as great as they were in black men.

Numerous studies have examined socioeconomic differences in cardiovascular risk factors (19), but the social patterning of the IRS has been infrequently investigated (20,21). Although neighborhood deprivation has been linked to cardiovasular risk factors and mortality in persons with diabetes (22–24), the relation of neighborhood characteristics to the IRS has not been examined.

Our study is unique in focusing on components of the IRS in young, asymptomatic adults and in demonstrating associations with neighborhood characteristics, after controlling for personal measures of income and education.

There are several proximate mechanisms through which neighborhood characteristics could be hypothesized to influence the development of components of the IRS. Recent reviews have highlighted the possible role of residential environments in influencing behaviors linked to diet and physical activity (25–26), both of which may be related to insulin resistance (5–7). Although evidence in humans is not yet conclusive, chronic stress may be related to the development of the IRS through endocrine pathways involving the hypothalamo-pituitary-adrenal (HPA) axis (27–29) or activation of the sympathetic nervous system (28,30). Because sources of chronic stress (such as noise, violence, and poverty itself) are also likely to vary across neighborhoods, this proximate pathway may also be involved in linking residential environments to the development of components of the IRS.

Contrary to our second hypothesis, the inverse associations of neighborhood socioeconomic characteristics with the IRS score were not stronger in persons of low income or education. In fact, among black men the opposite was true: neighborhood socioeconomic advantage was positively, rather than negatively, associated with the IRS score in persons of low income and education (and personal income and education were positively associated with the IRS score in disadvantaged neighborhoods). We have observed similar interactions between neighborhood characteristics and personal socioeconomic indicators for other cardiovascular risk factors in black men in other samples (31). Explanations for this finding remain speculative. It is plausible that among young black men living in the most disadvantaged neighborhoods, the behavioral and psychosocial correlates of higher personal income are different from those in other groups. These differences in socioeconomic gradients based on the larger context are reminiscent of differences in the socioeconomic patterning of cardiovascular risk in rich and poor countries, as well as over time (32). Consistently with some (33,34) but not all (20) previous work, we found stronger and more consistent socioeconomic gradients for components of the IRS syndrome in women than in men. The reasons for this sex difference remain to be determined.

Our findings are cross-sectional and need to be confirmed in longitudinal analyses relating neighborhood characteristics over a lifetime to the development of insulin resistance before conclusions can be drawn regarding causal relations between neighborhood characteristics and insulin resistance. However, the high correlation between baseline and year 10 neighborhood scores for the subsample with available baseline residence information suggests that the cross-sectional associations we observed are likely to be representative of associations of past place of residence with current outcomes. Our study is also limited by the use of the neighborhood socioeconomic score as an indirect proxy for the specific features of neighborhoods that may be relevant. These results highlight the need to identify specific features of neighborhood environments that may be important and empirically test their effects. Overall, our findings suggest that residential environments may play a role in facilitating recent increases in diabetes, and that broad-based, multidisciplinary strategies for disease prevention (ranging from medical care to urban planning) may be necessary.

This work was supported by the National Heart, Lung, and Blood Institute (NHLBI) Grant R29 HL59386 (to A.V.D.R.). The CARDIA Study was supported by NHLBI contracts N01-HC-48047, N01-HC-48048, N01-HC-48049, N01-HC-48050, and N01-HC-95095.