Ethnic and Socioeconomic Differences in Cardiovascular Disease Risk Factors:  Findings for Women From the Third National Health and Nutrition Examination Survey, 1988-1994 FREE

CARDIOVASCULAR disease (CVD) is the leading cause of death among women in the United States, accounting for a half-million deaths and 2.5 million hospitalizations annually.^1- 2 These deaths are linked with the primary CVD risk factors: elevated blood pressure, cigarette smoking, hypercholesterolemia, excess body weight, sedentary lifestyle, and diabetes, all of which are substantially influenced by behavioral, social, cultural, and economic factors.^3- 8

Ethnic minority women exhibit CVD risk factors to a greater extent than do white women. Many studies have documented greater prevalences of high blood pressure, physical inactivity, excess weight, and diabetes in African American women than in white women.^9- 15 Fewer studies have examined CVD risk factors among Hispanic Americans, who constitute 11% of the population in the United States. As one of the fastest-growing ethnic minority groups, the Hispanic share of the population is projected to increase to 96.5 million people, or 24% of the population, by 2050.^16- 18 Hispanic Americans are a heterogeneous group, with Mexican Americans constituting 64% of its population.¹⁹ Studies, with some inconsistencies, have shown higher prevalences of excess weight, diabetes, and untreated hypertension in Mexican American women than in white women.^13,20- 26 Studies have also shown that incidence²⁷ and mortality rates of CVD are higher for ethnic minority women than for white women.^28- 29

Some ethnic variations in CVD risk factors may be genetically based.³⁰ However, because ethnic minority women are more likely to be poor, have less education, and have less access to health resources than white women,³¹ socioeconomic status (SES) may explain many of the differences. Numerous empirical studies in the United States have shown strong associations among CVD risk factors and a broad array of SES indicators, including education, income, and occupational status.^32- 36 Additional studies in the United States and abroad have shown that rates of coronary heart disease and stroke morbidity and mortality are strongly associated with SES.^37- 42

Despite the convincing body of literature that links SES to CVD, most studies of ethnic differences in CVD have not taken SES into account. Therefore, erroneous findings can result, overestimating the effects of ethnicity without considering the effects of SES.^43- 44 Although distinguishing the interrelated effects of ethnicity and SES on health indicators has been strongly advocated,^45- 46 a recent review of CVD studies published in the United States during the last 20 years identified only 13 studies that included both ethnicity and SES in their analyses.⁴⁷ All of these studies were limited in scope; each included only 1 ethnic comparison, examined only 1 CVD risk factor, and/or was restricted to 1 geographic area.

We hypothesized that black and Mexican American women would exhibit CVD risk factors to a greater extent than white women but that the differences would be explained substantially by SES. To test this hypothesis, we evaluated the independent relationships of ethnicity and SES with 6 primary CVD risk factors: systolic blood pressure, cigarette smoking, body mass index (BMI), physical inactivity, non–high-density lipoprotein cholesterol (non–HDL-C), and non–insulin-dependent diabetes mellitus (NIDDM, chosen because of its relationship to lifestyle factors and its disproportionately high rates in ethnic minority groups).¹² We present data on 5266 black, Mexican American, and white women from the Third National Health and Nutrition Examination Survey (NHANES III). We also addressed 2 important measurement and analytic issues that contribute to understanding ethnic differences in CVD. First, we examined whether 2 indicators of SES, educational attainment (an individual-level marker of SES) and poverty-income ratio (a family-level marker), produced similar findings. Second, we examined whether 2 analytic procedures, a multivariate linear model and a matched-pairs analysis, produced similar findings.

NHANES III, conducted by the National Center for Health Statistics, Hyattsville, Md, assessed the health and nutrition of the civilian, noninstitutionalized population of the United States. NHANES III used the same stratified, multistage probability design used by previous NHANE surveys.^48- 50 It was conducted in two 3-year phases. Phase I was conducted from 1988 to 1991 and phase II from 1991 to 1994. It included a total sample of 33994 persons aged 2 months or older. By design, NHANES III oversampled the black and Mexican American populations to produce reliable health estimates for the 2 largest ethnic minority groups in the United States.

NHANES III staff conducted surveys in households, administering questionnaires to families, adults, and children. Standardized medical examinations followed, conducted in NHANES III mobile examination centers. The household surveys included demographic, socioeconomic, dietary, and health history questions; the medical examinations included measurements of blood pressure, lipid levels, and glucose levels.

The sample for our analyses included black, Mexican American, and white women aged 25 to 64 years who completed both the home questionnaire and medical examination. We used the lower age cut point of 25 years to ensure that most individuals had completed their highest level of education (our primary indicator of SES) and the upper age cut point of 64 years to avoid problems of selection effects due to non–CVD-caused morbidity and mortality.⁵¹ We excluded data for women who were pregnant (n=168), women from other ethnic groups (n=262), women whose surveys were coded as unreliable (n=8), and women who completed the home questionnaire but not the medical examination (n=402). The 402 women who completed only the home questionnaire were similar to the 5266 women who completed both the home questionnaire and medical examination in age (43.6 vs 42.2 mean years, respectively), years of education (11.9 vs 11.3 mean years, respectively), and country of birth (86% vs 84% born in the United States, respectively).

Respondents to NHANES III were asked to classify their race or ethnicity as black; Mexican or Mexican American; white; Asian or Pacific Islander; Aleut, Eskimo, or American Indian; or other Latin American or other Spanish. Women who chose 1 of the first 3 categories were included in our analyses.

The NHANES III data included several individual-level and family-level indicators of SES. We used educational attainment as our primary indicator of SES because education level, unlike income level and occupation, is available regardless of employment status, a more constant measure of lifelong SES, and not affected by the CVD outcomes we analyzed.^52- 53 Women were asked to give the highest grade or year of regular public or private school that they had completed. Poverty-income ratio, our secondary indicator of SES, is calculated from family income and family size and is based on US Bureau of the Census tables. Years of education and poverty-income ratio were moderately correlated (r=0.47, P <.001), a finding consistent with previous research.⁵³

We used the following 6 CVD risk factors as outcome variables: (1) Systolic blood pressure, measured in millimeters of mercury. We report the mean of the second and third of 3 readings, measured on the right arm by a physician while the participant was seated during the medical examination. (2) Current cigarette smoking status. Participants reported whether they had smoked at least 100 cigarettes during their lifetimes and whether they were currently smoking cigarettes. We used serum cotinine levels (>74 nmol/L [>13 ng/mL])⁵⁴ available in phase I participants to validate self-reported smoking status. (3) Body mass index, a measure of weight in kilograms divided by the square of height in meters. The calculations provide a measure of relative weight. (4) Leisure-time activity. Questions on leisure-time physical activity were adapted from the 1985 National Health Interview Survey,⁵⁰ which asked participants whether they had engaged in any leisure-time physical activity in the past month, including exercises, sports, or physically active hobbies. Women who reported no leisure-time activities were considered physically inactive. (5) Non–high-density lipoprotein cholesterol, measured in millimoles per liter (milligrams per deciliter). Measurements were taken from serum specimens, analyzed by standardized protocols, and calculated as the difference between total cholesterol and HDL-C.⁵⁵ Non–high-density lipoprotein cholesterol, which may be a better indicator of atherogenic lipoprotein particles than indirectly estimated low-density lipoprotein cholesterol,⁵⁵ does not require fasting blood samples, and therefore, allowed the use of the entire NHANES III sample. (6) Non–insulin-dependent diabetes mellitus. Fasting plasma glucose levels, available on the entire NHANES III sample, were determined using a microadaptation of the national glucose oxidase reference method.⁵⁶ Women were classified as having NIDDM if they fasted for 8 or more hours and had plasma glucose levels of 7.0 mmol/L (126 mg/dL) or more (cut point based on the new diagnostic criteria for epidemiological studies by the American Diabetes Association)⁵⁷ or if they reported a medical history of diabetes (other than during pregnancy) with an age of onset older than 25 years.

We carried out our primary statistical analyses, based on linear models, using SUDAAN, Version 7.11 (Research Triangle Institute, Research Triangle Park, NC), a software program that adjusts for the complex NHANES III sample design for calculating variance estimates.⁵⁸ All analyses incorporated sampling weights that adjusted for unequal probabilities of selection. We used SUDAAN to calculate weighted means and percentages and SEs, and we used SAS statistical software, Version 6.12 (SAS Institute Inc, Cary, NC) to calculate weighted SDs. We used multiple linear regression models for continuous outcomes and logistic regression models for binary outcomes. The outcome variables in our models were the 6 risk factors described herein. The predictor variables were age (in years, centered at the sample mean to aid in the interpretation of the regression coefficients), race or ethnicity (black and Mexican American women compared separately with white women; each comparison used white women as the reference group), and years of education (continuous and centered at 12 years). We included all first-order interactions between predictor variables (age and ethnicity, age and education, ethnicity and education). Because of multiple outcome variables, we selected a conservative level of statistical significance (P<.01, 2-tailed). In a secondary analysis, we substituted poverty-income ratio (continuous, centered at the sample mean) for years of education to examine the extent to which this indicator of SES might produce different results.

We used a matched-pairs analysis to confirm findings from the linear models because matching can overcome potential limitations such as unmet linear model assumptions, problems of collinearity (eg, a strong correlation between ethnicity and SES), and exclusion of higher-order interaction terms.^59- 60 For the matched-pairs analysis, we paired each black woman and each Mexican American woman with a different white woman and matched the pairs on educational level (<9, 9-11, 12, and >12 years) and age (±2 years). To test for overall differences in CVD risk factors between the matched pairs, we used paired t tests for continuous risk factor variables and the McNemar test for binary variables.

The selected NHANES III sample included 1762 black, 1481 Mexican American, and 2023 white women, aged 25 to 64 years, who completed both the home questionnaire and the medical examination (Table 1). Less educated women were older, less likely to live in urban areas (especially white women), and poorer than women who were more highly educated. Mexican American women were younger, had completed less education, and were less likely to be born in the United States and to speak English at home than black or white women. There were adequate numbers of women in 2 subgroups that are commonly underrepresented in studies of ethnic differences in CVD: Mexican American women with higher levels of education (n=236 with >12 years of education) and white women with lower levels of education (n=120 with <9 years of education).

The data did not confirm our hypothesis that SES would substantially explain differences in CVD risk factors. Clearly, both ethnicity and SES have important and independent associations (P<.001) with CVD risk factors (Table 2). Ethnic minority status was significantly associated with BMI, systolic blood pressure, NIDDM, and physical inactivity. Socioeconomic status was significantly associated with BMI, physical inactivity, cigarette smoking, and high non–HDL-C. As expected, age was significantly associated with all risk factors. The magnitude of the ethnic and SES differences was large for many comparisons. For example, the β coefficient was 2.8 for the black-white comparison for BMI. Each BMI unit is equivalent to about 2.7 kg (6.0 lb) for comparison of persons with average heights. Thus, the BMI β coefficient of 2.8 for black women indicates that they were, on average, 7.6 kg (16.8 lb) heavier than white women of comparable SES and age.

When poverty-income ratio was substituted for years of education in the linear models, results were almost identical; all risk factors that were significantly related to years of education were also significantly related to the poverty-income ratio. In addition, NIDDM was also significantly related to the poverty-income ratio.

Table 3 presents descriptive statistics, displaying means with SDs and percentages. As confirmed by the linear models (Table 2), black and Mexican American women had significantly higher levels of BMI and blood pressure and higher prevalences of diabetes and physical inactivity than did white women of comparable SES. The main exceptions were for smoking prevalences and non–HDL-C levels; Mexican American women had lower smoking prevalences and black women had lower non–HDL-C levels than did white women. Women with lower SES from each of the 3 ethnic groups had significantly higher levels of BMI and non–HDL-C and higher prevalences of physical inactivity and smoking than women with higher SES.

Two important and significant interactions emerged in the linear models, between ethnicity and age for both blood pressure and smoking (Figure 1). Black and Mexican American women had steeper increases in blood pressure than white women across age groups, resulting in significantly greater ethnic differences for the older than younger age groups (eg, the black-white difference of 4 mm Hg at age 25-34 years increased to 11 mm Hg at age 55-64 years). Black and Mexican American women had stable smoking rates across age groups, in contrast with white women, who showed a sharp decrease in smoking rates, especially between the 2 youngest age groups (eg, 41% of the white women aged 25-34 years smoked compared with 26% of the white women aged 35-44 years).

The matched-pairs analysis confirmed the results from the linear models. Results were almost identical for all risk factors, indicating no bias in applying the linear model to the full sample. The only result not confirmed was for non–HDL-C; the black-white comparison was significant in the linear model but not significant in the matched-pairs analysis. This difference raises some doubts about the validity of the finding of significantly higher non–HDL-C values for black vs white women in the unmatched sample.

To our knowledge, this article is one of the first to document higher levels of CVD risk factors among black and Mexican American women than among white women of comparable age and SES. We hypothesized that ethnic minority status would be associated with higher levels of CVD risk factors, but that the associations would be explained substantially by SES. Our hypothesis was not confirmed. After adjustment for age and SES, highly significant differences in BMI, blood pressure, diabetes, and physical inactivity remained between white women and both black and Mexican American women. In addition, we found large differences in CVD risk factors by SES, a finding that illustrates the high-risk status of both ethnic minority women as well as white women with low SES.

Only 1 previous study examined the joint relationship of SES and ethnicity with a set of CVD risk factors. Shea and colleagues¹¹ examined the independent associations of educational attainment and ethnicity (white, black, and Hispanic) with the primary CVD risk factors. Findings were similar to ours; however, the study was confined to New York State, relied on self-reported data from a telephone survey, and was limited by potential collinearity between SES and ethnicity (eg, 5% of whites vs 22% of blacks and 38% of Hispanics had not completed high school). Other CVD studies that have been more limited in scope (eg, restricted to 1 ethnic comparison, 1 CVD risk factor, and/or 1 geographic area) have, in general, supported our findings of higher-risk factors among ethnic minority groups after adjustment for SES.⁴⁷

NHANES III is the most comprehensive national survey to date with data on CVD risk factors for black, Mexican American, and white women. The survey included an oversampling of black and Mexican American women, the 2 largest groups of ethnic minority women in the United States. Because NHANES I and II did not include large numbers of Mexican American women and a related survey, the Hispanic Health and Nutrition Examination Survey, did not include white women, NHANES III is the first such survey to allow a comparison of Mexican American and white women. It also included large numbers of women at the extremes of educational attainment. Extensive and complete data are available from both the home survey and medical examination, including individual-level and family-level indicators of SES and standardized measures of blood pressure, lipid levels, and glucose levels.

Despite these strengths, our results should be interpreted with caution because of several design and measurement limitations. First, the cross-sectional design of NHANES III limited our ability to draw inferences about causal pathways. For example, the onset of smoking, excess weight gain, and physical inactivity usually occurs at early ages, often before the completion of formal education. Although we could not determine from the data the temporal association between level of education and these CVD risk factors, it is more likely that SES influences risk of disease than that risk of disease influences SES.³⁸

Second, the NHANES III data contained a potential for bias from self-reported questions, such as those questions about smoking and physical activity. To verify self-reported smoking, we used a biochemical measure, serum cotinine, that was available for all women from phase I of the survey. Among women who reported being nonsmokers, the following percentages had positive results for cotinine: 8.2% of black women, 2.0% of Mexican American women, and 3.3% of white women, and 6.2%, 3.4%, and 2.5% of women completing less than 12, 12, and more than 12 years of education, respectively. Thus, the minimal underreporting of smoking by Mexican American women and women with higher SES lends credence to our finding of their low smoking rates. Few studies have assessed reporting bias by ethnicity for physical activity measures; however, we have seen no indications of differential underreporting or overreporting by black, Mexican American, or white women.⁶¹

Third, the NHANES III data used only leisure-time activities to measure overall physical activity. It is not surprising that black and Mexican American women, who often have fewer resources and less time to engage in leisure-time physical activity than white women, reported lower levels of activity. We did not anticipate, however, that higher-educated black and Mexican American women (those who had completed >12 years of education) would be substantially more likely than white women to report no leisure-time activity (27.8% and 18.0% compared with 12.3%, respectively). The few questions available in NHANES III to assess physical activity highlight the need for future surveys to include measures such as total activity recall or energy expenditure that better reflect overall activity levels of women, regardless of ethnic and income status.

Last, our primary measure of SES (educational attainment) did not measure core beliefs and practices in relation to specific CVD risk factors or medical conditions. Furthermore, it did not provide the same economic benefits to black and Mexican American women as to white women.³¹ As shown in Table 1, the likelihood of living below the poverty level decreased with increasing education, but less so for black and Mexican American women than for white women.

The striking ethnic and SES differences in CVD risk factors highlight the need for reform in public health policies, health care systems, and intervention programs.^19,62- 63 Although past primary and secondary CVD preventive efforts have been effective in reaching white women and women with higher SES, they have been less effective in reaching ethnic minority women and women with lower SES.^64- 65 For example, of the women aged 25 to 64 years who participated in the 3 large CVD community intervention programs of the 1980s, more than 90% were white (non-Hispanic), and more than 75% had completed a high school education or more.⁶⁵ In addition, there is strong evidence that white women and women with higher SES obtain earlier diagnoses for CVD-related conditions, have better access to care, and are more likely to be treated effectively than other women.^66- 69 These findings are consistent with results from health services research and clinical trials that show that white women and women with higher SES are more likely than African American women and women with lower SES to be screened, diagnosed, and treated for hypercholesterolemia,^70- 72 and that white women are more likely than Mexican American women to receive treatment for diabetes,⁷³ hypercholesterolemia,⁷⁴ and hypertension.^67,75- 76

The challenge for public health professionals is to identify and understand groups of women with high prevalences of risk factors and disease, and to design effective interventions at the individual and societal levels that will benefit these women. Public policies need to ensure that ethnic minority women and women with low SES from all ethnic groups have heart-healthy food choices, smoke-free environments at the workplace, and safe and convenient places to exercise.⁷⁷

Health care systems need effective protocols to reach ethnic minority women and women with low SES who often experience differential access to health care services^17,78- 79 because of cost barriers, unavailability of health insurance, and discrimination in health care.^80- 81 Health care systems need to reach high-risk women with screening programs for cholesterol, blood pressure, and diabetes; provide more care (and more effective care) for chronic conditions such as diabetes and hypertension; and create better access to programs that address smoking onset or cessation, weight problems, high fat intake, and high cholesterol levels.

Intervention programs, based on targeted subgroup approaches, need to identify high-risk populations, develop strong and enduring partnerships with communities, and tailor strategies to the language and literacy needs, values, and cultures of the populations.^82- 83 A targeted subgroup approach should not overshadow the essential need for population-wide approaches to CVD prevention, such as those shown to be important during the 1980s and 1990s.^84- 87 The public health challenge is to continue effective approaches to lowering CVD in the overall population and to accelerate changes in high-risk populations.⁴⁷

Our findings are especially relevant in light of the global rise of CVD and other chronic diseases that are influenced by increasing prevalences of hypertension, cigarette smoking, hypercholesterolemia, obesity, physical inactivity, and diabetes. Our work builds on previous studies that have documented disproportionate suffering, disability, and premature death from CVD among ethnic minority women and those with low SES. Fortunately, CVD risk factors are well established and their occurrence can be largely prevented. Unfortunately, CVD is not curable, a fact that highlights the critical need for effective primary and secondary preventive efforts to lessen the substantial health disadvantages faced by ethnic minority women and women with low SES.