Defining Health Disparity Groups

The Minority Health and Health Disparities Research and Education Act of 2000 defines health disparity populations (or groups) as those for which “there is a significant disparity in the overall rate of disease incidence, prevalence, morbidity, mortality or survival rates.” As discussed earlier, which group is identified as being a disparity group will differ depending on which of the above health indicators is used. For example, men could be viewed as a disparity group relative to women if mortality rates are used, but women would be seen as a disparity group if some measures of morbidity are used. Similarly, differences in disease incidence or prevalence depend on the disease examined. Some differences between men and women derive from biology: There is a greater prevalence of breast cancer among women than men. Only women experience cervical or ovarian cancer; only men experience prostate cancer. Differences in these disease rates do not fit the definition of a disparity because they are unavoidable.

When examining other diseases, the question of whether a difference in prevalence represents a disparity becomes more complex. For example, women have a lower prevalence of cardiovascular disease than men. It can be debated whether this difference represents a disparity. The female advantage may reflect the protective effects of female hormones—a biological difference that is not modifiable by social policy. However, the extent of the male-female difference could also be due to modifiable conditions that reflect social disadvantage. It is possible that greater hardships faced by women as the result of discrimination in the workplace, exposure to sexual harassment and abuse, and so forth lessen the biological advantage they might otherwise enjoy. To the extent that this is the case, the disparity is the reduction in the “natural” female advantage. Others might argue that men face greater lifetime stresses than women and that the difference in cardiovascular diseases represents some combination of an unavoidable biological difference and a modifiable difference. However, given that, on average, men have greater social advantages than women, this would not fit the definition of a health disparity, if one assumes that only avoidable differences experienced by disadvantaged groups qualify as a disparity.

Lung cancer is an example of a disease whose patterns have grown more similar between women and men, but one would not see this as a reduction in disparities. Men once had a relatively greater prevalence of lung cancer, compared with women, than they do today. The change in the relative prevalence reflects changes in rates of smoking among men and women: Over time, as women’s rate of smoking has increased, so has their rate of lung cancer. The difference in the prevalence of smoking, and the resulting difference in lung cancer rates, meet the criterion of an avoidable difference. However, the perceived fairness of this change depends upon one’s explanation for the greater rates of smoking among men than women in the 20th century. Ironically, the greater social equality of women may have provided more opportunity for women to take on male risk factors, such as smoking.

The example of gender suggests that a further definition of health disparities may be useful. Even under ideal social and environmental conditions, there will be differences in rates of some diseases and in longevity due to genetic, and other biological, factors. Both individuals and groups may differ in vulnerability to specific diseases, due to this variation. Eventually, medical treatments for genetic risk may equalize individuals’ capacities for a healthy lifespan. If so, failures to reach the same end point could be considered a disparity because anything less would be potentially avoidable and inequitable. Thus, disparities could be defined as the extent to which individuals, or segments of the population, fail to achieve their highest potential state of health, at a given age, given currently available medical treatments.

Current Approaches to Disparity Groups: Race/Ethnicity

Several definitions of health disparities equate disparities with differences among racial and ethnic groups. The NIH Strategic Plan Volume 1 (2002, pp. 19–20) presents data on health among several selected populations. These data are reproduced in Table D-4. They show marked differences in such diverse health indicators as infant mortality, cancer mortality, coronary heart disease mortality, and the prevalence of diabetes, end-stage renal disease, and stroke. More recent data are available from Health, United States, 2004 (National Center for Health Statistics, 2004), recently released by the Centers for Disease Control and Prevention (CDC). Table D-5 presents the overall death rate, as well as death rates for the two leading causes of death—heart disease and malignant neoplasms. Table D-6 presents data on infant deaths.

TABLE D-4

Health Disparities of Certain Conditions in Selected Populations.

TABLE D-5

Death Rates per 100,000 by Race/Ethnicity, 2002.

TABLE D-6

Infant Deaths per 1,000 Live Births, 2000–2002, Overall by Education.

There are two clear observations that can be made about the data presented in both of these tables. One is that African Americans show more adverse health outcomes on each one of the indicators. They have the greatest morbidity and mortality on every reported indicator, and the gap is often substantial. For example, compared with Asians or Pacific Islanders who experience 4.8 deaths for every 1,000 live births, African Americans experience 13.6 deaths. The next highest group, American Indians or Alaskan Natives, have a rate of 8.9 deaths.

The second observation is that no other group shows consistently poor health outcomes across indicators. Whites show poorer outcomes than groups other than African Americans on most of the reported health indicators (e.g., overall cancer mortality as well as death rate for breast and lung cancer, coronary heart disease and stroke mortality, and prevalence of AIDS). American Indians or Alaskan Natives have the second highest rates of infant mortality, and Hispanics or Latinos have the second highest prevalence of diabetes. Asian Americans or Pacific Islanders show the most favorable profile. They experience the lowest rates of infant mortality, overall cancer mortality and death from lung and breast cancer, and coronary heart disease mortality. They have a markedly lower prevalence of AIDS than any group other than Native Americans/Alaskan Natives. They show intermediate rates of stroke mortality and end-stage renal disease.

One problem with the conclusions reached above is that they are based on large groupings by race and ethnicity. These broad categories may mask substantial variation in health within some of the groups. Members of the same ethnic group from different countries and areas of origin have different degrees of disadvantage and health risk. For example, as shown in Table D-6, Asians/Pacific Islanders as a group have the lowest rate of infant deaths (4.8 per 1,000 births) compared with other groups. However, this masks substantial variation among Asians and Pacific Islanders. The rate for Hawaiians (8.7) is more than double that of Chinese (3.2), with intermediate rates shown by Japanese (4.5) and Filipinos (5.7). A similarly large span in outcomes is shown among Hispanics and Latinos. As a group, they show the second lowest rates of infant deaths (5.5 per 1,000 births). Within this category, however, Puerto Ricans experience 8.3 deaths, whereas Cubans experience 4.2 deaths per 1,000, with intermediate rates shown by Central and South Americans and by Mexicans. Within both whites and blacks, removal of Hispanics has little impact on rates, probably because Hispanics make up a relatively small proportion of the larger group. Recent analyses reported by Zsembik and Fennell (2005) using National Health Interview data from 1997 to 2001 compared a number of medical conditions, functional impairment, and overall self-rated health for Mexicans, Puerto Ricans, Cubans, and Dominicans in the United States along with blacks and whites. The pattern of health advantage depended, in part, on the health outcome examined; however, overall, Mexicans reported better health outcomes than others, while Puerto Ricans reported poorer outcomes. The relative position of Cubans and Dominicans differed by outcome.

Data from Palaniappan, Wang, and Fortmann (2004) also show variation in disparities when examining subgroups in relation to specific diseases. They examined rates of death from coronary heart disease and from all causes broken down into more precise subgroups of Asians. Although Asian Indians had the lowest rates of all-cause mortality, as can be seen in Table D-7, they had relatively high rates of coronary heart disease compared with other Asian groups. Among blacks as well, subgroups vary substantially. Fang, Madhavan and Alderman (1996) reported significant differences in the rates of mortality from cardiovascular disease among blacks born in different parts of the United States or in the Caribbean. Mortality rates among blacks residing in New York City were markedly higher than among those residing in the South, intermediate among those born in the Northeast, and lowest among those born in the Caribbean.

TABLE D-7

Mortality Ratios for Coronary Heart Disease (CHD) and All-Cause Mortality in California, 1996–2000.

These data illustrate the importance of looking at subgroups within large ethnic categories. However, it is often difficult to obtain adequate data to evaluate health disparities in these subgroups because of their relatively small numbers. This becomes even more acute when studying smaller populations, such as those from specific countries or ethnic groups. For example, Yang, Mills, and Riordan (2004) reported markedly higher incidence rates and mortality from cervical cancer among Hmong women than among other Asians/Pacific Islanders, and Cho and Hummer (2001) reported substantial variations in disability status among subpopulations of Asians/Pacific Islanders, with the Hmong, Laotians, and Cambodians showing the poorest outcomes. These groups differ by SES, as well. For example, within the Asian immigrant group, more than 60 percent of those from India or Taiwan are college graduates, compared with roughly 5 percent of those from Cambodia or Laos (Rumbaut, 1996). Even within a given group, subpopulations may experience greater disadvantage and poorer health (e.g., Native Americans living on reservations versus other Native Americans).

Further complicating ethnic group differences in health, health status appears to vary by length of time in the United States. First-generation immigrants appear to have a health advantage across virtually every group (Singh and Miller, 2004). This may be due, in part, to the healthy immigrant effect, in which there is differential selection for those who have the characteristics (including better health) that allows them to immigrate to the United States (Thomas and Karagas, 1996). It may also reflect protective effects of traditional diets, supportive social networks, or other health practices of first-generation immigrants. Supporting this view, Eschbach, Ostir, Patel, Markides, and Goodwin (2004) report lower mortality among older Mexican Americans living in neighborhoods with a high density of Mexican Americans. They attributed this difference to the protective effects of the concentration, which may buffer Mexican Americans from “unhealthful aspects of U.S. culture” (Eschbach et al., 2004, p. 1810).

Finally, as shown in recent analyses by Williams (2005), the extent of disparities also varies depending on the measure used. Disparities will differ not only between different diseases, but also within mortality rates, depending on the measure. For example, Williams showed greater disparities between African Americans and whites when age-specific comparisons were made, rather than age-adjusted comparisons. Looking at age-specific rates also shows differences that occur only at some points across the lifespan.

The approach to disparities suggested earlier—which frames disparity as the gap between current health status and biologically feasible health—suggests a strategy of using the group with the best health outcome as the comparison group. This group presumably represents the highest achievable outcome under current social and health care conditions, though one would need to evaluate potential genetic factors. Research could then be directed to understanding the other factors responsible for the gap between the optimal outcome and the groups with the poorest outcomes. These may be disease-specific mechanisms. At the same time, the large, persistent, and consistent disadvantage suffered by African Americans across diseases suggests that some common mechanisms systematically affect this group’s health. It also suggests that more attention should be paid to cross-cutting factors that systematically affect African Americans’ health. Potential factors are described below.

Current Approaches to Disparity Groups: SES

Early research on socioeconomic factors and health tended to report health outcomes of those in poverty versus those above the poverty line. In the past two decades, research has increasingly examined health differences across the full socioeconomic spectrum. The shift in perspective was stimulated by data from the Whitehall Study of British civil servants (Marmot et al., 1978). These data demonstrated a graded association of occupational grade and 10-year mortality. Even within this relatively homogenous group of individuals, all of whom were employed, higher occupational grade conferred a lower risk for a range of diseases and for subsequent mortality.

Challenged by findings from the Whitehall Study, researchers began to evaluate the patterning of health disparities by socioeconomic variables beyond the bivariate association with poverty or nonpoverty (and even beyond the categorization into poor, near poor, and nonpoor). Data to evaluate whether a similar SES-health gradient occurred in the United States were difficult to find, as most public health monitoring and epidemiologic surveys provide data on race and ethnicity but not on income; many sources do not even include education. Those that collected or reported data on income were often coded only in terms of being above or below the poverty line.

The research that has been done on a broader SES spectrum has generally shown a graded, but nonlinear, association between income and mortality. Data on income and mortality from the National Longitudinal Mortality Study reported by Backlund, Sorlie, and Johnson (1999) were best characterized by a two-slope model, with a sharp drop in mortality as income increased from $2,500 a year to $22,500 (in 1980 dollars) and a continued but more gradual drop above that income level up to $57,500—the highest income level reported. The graphic representation of the association of income and mortality provided by Wolfson, Kaplan, Lynch, Ross, and Backlund (1999) shown in Figure D-1 similarly shows a steeper drop in the relative risk of mortality as income increases at the lower part of the income distribution. Superimposed on the graph of the relative risk of mortality is a graph showing the population distribution. A relatively small proportion of the U.S. population is at the very bottom of the income distribution, where the steepest drop occurs. Most of the population resides in the distribution between about $15,000 and $40,000 of income, which is above the poverty line. Thus, if one limited research to those below the poverty line as a disparity group, such research would not incorporate an understanding of the largest segment of the population in which disparities occur. There is a continued drop in the relative risk of mortality even above $50,000, but it is much shallower and encompasses a smaller portion of the population.

FIGURE D-1

Relative risk of dying and population distribution for U.S. individuals by household income ($). SOURCE: Wolfson et al., 1999.

These data suggest the difficulty in defining a disparity group by income. In terms of individual need and alleviating the most extreme disparity, those in poverty should be the focus. At the same time, a much higher proportion of the population is above the poverty line, yet experiences higher mortality than those who are relatively more affluent, and attention must also be paid to this broader segment. Given that the increased burden is greatest among those in the bottom 30 to 40 percent of the income distribution, this segment of the population should be of particular concern.

One caveat about these data is that they were collected more than 15 years ago, so the absolute levels of income will have a different meaning. In addition, the distribution of income has changed over time, with increasing income inequality. There is debate in the field regarding whether absolute versus relative income is more important for health and whether the distribution of income across the population has an additional impact. Studies have shown that greater income inequality at the level of states (Kaplan et al., 1996; Kennedy et al., 1996) and countries (Wilkinson, 1996) is associated with higher mortality. These data are at the population level, and it may be that such inequality adversely affects the health of the entire population. It may also be that greater income inequality has a proportionately greater impact on those at the bottom of the income hierarchy, though the empirical work on this is inconclusive. More recent data suggest that the association between income inequality and health may occur in the United States, but to a lesser extent in other countries, and may vary by time period.

In contrast with income, which shows a continued drop in mortality as income increases (though with a much sharper drop in the lower portion of the distribution), the associations of mortality with education are less continuous. Table D-8 provides information on death rates for those who did not graduate from high school (< 12 years), high school graduates (12 years), and those with some post-secondary education (13+ years). For all-cause mortality and for each of the specific causes, the death rates are lower for those with more education. The gap is larger between high school graduates and those with some post-secondary education. The latter are a mixed group, as some have a college and/or advanced degree and others have either some post-secondary education but no degree or have earned an associate’s degree. Other research shows differences between those with some post-secondary education versus those with a bachelor’s degree and also, for white men, between those with a bachelor’s degree versus those with advanced degrees. The health benefit of an advanced degree versus a college degree, however, may be less for African American men and women and for white women, consistent with lower social and economic returns on education.

TABLE D-8

Age-Adjusted Death Rates per 100,000 by Educational Level (Age 25–64), 2002.

Education provides a mixture of resources. To the extent that education provides information, knowledge, and skills that improve health, each additional year of education should contribute somewhat equally to improved health. However, educational attainment also serves as a credentialing function. As a result, there will be a greater benefit for achieving years of schooling that result in a degree or credential than for additional years that do not. Thus, the benefits of completing the 12th year of schooling that results in a high school degree is greater than the benefit of the 10th or 11th year of schooling. This is referred to as the sheepskin effect.

There may also be some effect due to the personality traits of those who complete their schooling. Conscientiousness, for example, has been linked to greater longevity (e.g., Friedman et al., 1995). Those who are more conscientious may exhibit this trait both in relation to their schooling (leading to greater attainment) and in relation to their health. Similarly, as Victor Fuchs has suggested, those who are willing to delay gratification to invest in their schooling to obtain a degree may do the same in relation to their health. These are not mutually exclusive explanations to the sheepskin effect, as both may be operating and contribute to the discontinuous effect of education at credentialing milestones. Yet another explanation for the association of education and health is that social networking (and associated social norms) change as one moves on to the next level of education.

The empirical data are consistent with the hypothesis that education confers benefits due to credentialing and/or due to the related characteristics associated with completion, rather than simply due to knowledge accumulation or social norms. Figure D-2 from Backlund et al. (1999) shows that for both men and women, the relative risk of mortality is essentially flat for those with 11 years or less of education, drops for those who are high school graduates, and remains unchanged until 16 years of schooling, which indicates college graduation, at which point there is another drop in mortality.

FIGURE D-2

Predicted relative risks using the best-fitting models for education and income for men and women aged 25–64. Solid line with squares adjusted for demographic variables. Solid line adjusted for socioeconomic status and demographic variables. SOURCE: (more...)

Current Approaches to Disparity Groups: Rural Health

Those who live in rural areas show more adverse health outcomes than those in metropolitan areas. Hartley (2004) noted that in the Urban and Rural Health Chartbook (National Center for Health Statistics, 2001), populations in rural areas showed poorer outcomes on 21 of the 23 health indicators, including mortality. Health, United States, 2004 (National Center for Health Statistics, 2004) provides more recent data on mortality by level of urbanization, as well as by region. As can be seen in Table D-9, the age-adjusted death rate increases as urbanization decreases. The highest death rate (914.3) is in nonmetropolitan areas, while the lowest (833.1) is in large metropolitan areas. When broken down by race, this pattern holds for whites. For African Americans, however, death rates are less different across levels of urbanization, although they are still higher in non-metropolitan than in medium or large metropolitan areas. There are differences in health indicators such as self-rated health and limitation of activity by residence area. In relation to the former, 8.7 percent of those residing within a metropolitan area report being in fair or poor health compared with 11.7 percent of those outside a metropolitan area. Similarly, 11.4 percent of those in metropolitan areas report experiencing a limitation of activity due to a chronic condition, compared with 15.9 percent of those outside a metropolitan area.

TABLE D-9

Age-Adjusted Death Rates by Region and Level of Urbanization, 2000–2002.

Death from injury and suicide are more common in rural areas. Figure D-3, taken from Peek-Asa, Zwerling, and Stallones (2004), shows the mortality rate from unintentional injuries for three types of metropolitan areas and two types of nonmetropolitan areas. The highest rate occurs in the most rural settings (i.e., nonmetropolitan areas without a city). The next highest rate occurs in non-metropolitan areas with a city, followed by small metropolitan areas. Importantly, the death rate in large fringe areas is somewhat lower than in large central cities. A similar pattern was reported by Eberhardt and Pamuk (2004) for mortality rates for suicide and for a range of risk factors (e.g., smoking, limitation of activity, and obesity). The difference between the central city and the large fringe was particularly great for sedentary activity during leisure, with rates in the central city second only to the most rural area.

FIGURE D-3

Unintentional traumatic injury death rates, by urbanization level: United States, 1996–1998. SOURCE: Peek-Asa et al., 2004.

Consistent with the data presented above, Hartley (2004) noted that because suburban counties show the most positive health profiles, with better health status than either the rural areas or urban centers, finer differentiations than simply urban and rural should be made. This is the case for some indicators but not for others. On some health indicators, suburban areas show an intermediate level of health between urban and rural (e.g., Weeks et al., 2004).

In addition to differences by degree of urbanicity or rurality, health status differs by geographic region. Tables D-10 and D-11 show rates of poor/fair self-rated health and limitations of activity, respectively, for different groups. As described above, there are differences between those within or outside a metropolitan area, but there are also regional differences. A greater proportion of those living in the South than in other regions report poor or fair health, whereas more in the Midwest report a limitation of activity due to chronic conditions, compared with other regions. There is no cross-reference for rural/urban status by geographic area. One confounding factor is that some areas of the country are more rural than others, so geographic differences may, in part, be due to the degree of rural residence. Additionally, the conditions of rural life may differ in various parts of the country. Rural areas may differ in terms of physical environments such as water quality, the type of agriculture, and even the built environment. Rural residents also vary in sociodemographic characteristics.

TABLE D-10

Self-Rated Health: Percent Reporting Fair or Poor Health, 2002, by Race/Ethnicity and Geographic Locale.

TABLE D-11

Limitation of Activity Due to Chronic Conditions, 2002, by Race/Ethnicity and Geographic Locale.

Interaction Among Factors Associated with Disparities

The sections above present data showing health disparities by race and ethnicity, education, income, and rural/urban residence. However, these characteristics of individuals and groups do not exist in isolation. Groups that are disadvantaged in one domain of life are often disadvantaged in others.

The interconnections among different factors associated with health disparities are clearest in the overlap of race/ethnicity with SES. Both African Americans and Hispanics are overrepresented in lower-SES categories. Data from the 2000 census summarized in Table D-12 show that 49 to 50 percent of Hispanics and 26 to 30 percent of African Americans have less than a high school education, compared with 13 to 14 percent of whites. Differences by gender in education are greater for Asians than for the other groups. Table D-13 shows the differences in median income, ranging from $29,645 for African Americans to $47,777 for whites; the differences in wealth are even more dramatic.

TABLE D-12

Education by Gender and Race/Ethnicity.

TABLE D-13

Median Income by Racial/Ethnic Group.

For some health outcomes, differences between African Americans and whites become insignificant once income is controlled for (Haan and Kaplan, 1985). For other health outcomes, although the difference between groups is substantially reduced, a residual effect associated with race/ethnicity remains. Whether this residual effect reflects poor measurement of economic status or the importance of other factors is unknown. The residual effect is particularly strong for birth outcomes. Data presented in Table D-14 show life expectancy at age 25 for U.S. males by race and income (Lin et al., 2003). The data show that the overall difference in life expectancy is greater by SES than by race: There is a 4.4-year difference between blacks and whites, compared with a 7.9-year and 8.6-year difference between those earning less than $10,000 and those earning more than $25,000 within whites and blacks, respectively. Although the data show that the gap between races diminishes slightly with higher incomes, a difference between whites and blacks remains at each income level. Some part of this difference may have to do with other socioeconomic differences between blacks and whites not captured by household income. At each level of income, for example, African Americans and Hispanics have lower net worth and live in worse neighborhoods than whites. For example, among the lowest quintile of income, whites have a net worth of almost $50,000, whereas the net worth of African Americans and Hispanics is a little more than $7000. Thus, controlling for income will not adequately control for differences in wealth.

TABLE D-14

Life Expectancy at Age 25, U.S. Men: Race and Income Differences.

The meaning of specific indicators of SES may differ across groups, and it may be fruitful to examine SES effects within groups. In addition to the problems of equating the implications of a given income across groups as discussed above, the meaning of educational attainment may vary as well. This is most problematic for groups that have received their education in other countries, with different educational systems and levels of accreditation. Even within the United States, a given level of educational attainment may not confer the same resources and advantages, due to differences in the quality of the schooling. High school graduates from underfunded inner-city schools do not have the same opportunities as graduates of more affluent suburban schools or private schools. Due to residential segregation and local funding of schools, African Americans are especially likely to experience poor-quality schooling. As a result, the economic and health benefits of high school graduation may be diminished.

The data presented above suggest that one cannot adequately study racial and ethnic disparities in health without considering socioeconomic factors and vice versa. Similarly, race/ethnicity and SES need to be considered in evaluating disparities in rural and urban health. Probst, Moore, Glover, and Samuels (2004) noted that generalizations about rural health largely capture the experience and outcomes of whites, because whites make up 84 percent of rural populations, African Americans comprise 8 percent, nonblack Hispanics comprise 5 percent, and Asians/Pacific Islanders and American Indians/Native Alaskans comprise less than 2 percent each. There are differences in racial/ethnic composition of rural populations in different regions of the country. Rural African Americans are predominantly in the South, and of these, half are living in Mississippi, Georgia, and North and South Carolina. In contrast, rural Hispanics are living primarily in the West, with over one quarter of rural Hispanics in Texas alone (Probst et al., 2004).

Conclusions about the nature of disparities may differ if one looks at characteristics individually rather than in conjunction with associated factors. For example, Coughlin (2002) presented a deeper analysis of a finding of lower rates of cancer screening for African Americans versus whites in the overall population. They broke down the overall rates to look at rates in three different types of counties: black counties in the South, other counties in South, and all other counties in United States. In this analysis, they found no differences by race within each of those categories; the differences by race were due to the different distribution of populations across the three types of counties.

Similarly, Probst et al. (2004) noted that “health issues among rural racial/ ethnic minorities cannot be separated from educational and economic issues” (p. 1697). As shown in Table D-15, there are substantial differences in educational attainment by both race/ethnicity and area of residence. Of working-age adults, 40 percent of African Americans in rural areas lack a high school diploma, compared with 19 percent in urban areas; comparable figures are 50 percent and 42 percent for Hispanics and 15 percent and 9 percent for whites in rural versus urban areas. Differences are less marked among older adults, for whom a lack of high school graduation was more common, and show smaller discrepancies by either race/ethnicity or rural/urban residence. Rural residence is also associated with an increased likelihood of being in a low-paying job. The increased likelihood holds for all three ethnic groups, although the difference between urban and rural rates is less for Hispanics than for whites or African Americans. Given the predominance of whites in rural areas, the data suggest that to understand the relatively poorer health outcomes of whites compared with other groups on some health indicators, area of residence should be examined.

TABLE D-15

Socioeconomic Profile of Urban and Rural African Americans, Hispanics, and Whites.

Gender

Considerable literature exists on gender and health. As discussed earlier, the comparisons by gender are complex, reflecting both biological and social differences. A review of this area is beyond the scope of this review. However, it is important to note that health disparities associated with race/ethnicity, SES, and rural residence may differ by gender. As with race and ethnicity, the actual meaning of these variables may differ for men and women. Women’s status may be determined more by a husband’s socioeconomic characteristics than vice versa so that household, rather than personal, income may be particularly salient for women. Women’s roles differ in various ethnic groups, so gender differences may vary by ethnicity, as well. As will be seen in the section on conceptual models below, most portray gender as a modifying factor in the pathways linking socioeconomic factors with health.

Approaches to Understanding Causes of Health Disparities

One approach to understanding health disparities is to examine them in light of what is known more generally about determinants of health. It is logical to assume that health disparities operate through factors that determine health. The following discussion is organized around the categories of determinants first discussed by Lee and Paxman (1997) and updated by McGinnis, Williams-Russo, and Knickman (2002). Based on analyses from the CDC, they present the relative contribution of a range of factors to premature mortality. Sociodemographic factors including SES, race/ethnicity, gender, and area of residence were not evaluated. Rather, they represent more proximal determinants that may mediate the impact of the sociodemographic variables on health. The factors are health care deficiencies, genetic vulnerabilities, social environment, physical environment, and behavior and lifestyle.

Health Behaviors

According to the CDC’s analysis, the single category of factors that contribute most to premature mortality is health behavior and lifestyle, estimated to account for 40 to 50 percent of premature mortality. On virtually every risk factor, one finds differences by SES and, to a lesser extent, by race/ethnicity.

The two greatest risk factors in this domain are tobacco use and obesity. Smoking is more common among those with less education and less income (Pierce et al., 1989), though it has not always been so. The most recent data from Health, United States, 2004 (National Center for Health Statistics, 2004), shown in Table D-16, shows a marked effect of education. Rates of smoking are more than 3 times higher among those with less than a high school education, compared with those who have graduated from college. The drop-off in rates of smoking is most marked between those with some college and those who graduate from college; the latter have smoking rates that are less than half those of the former. Smoking rates do not differ substantially between African Americans and whites; at most educational levels, rates are slightly higher for whites. Thus, although tobacco use may serve as an important mechanism in disparities associated with SES, it does not appear to be a major contributor to the relatively poorer health of African Americans.

TABLE D-16

Percentage Smokers, 2002.

Rates of obesity show a very different pattern, varying both by race/ethnicity, SES, and gender. As seen in Table D-17, in 2001, 31.1 percent of blacks and 23.7 percent of Hispanics were obese compared with 19.6 percent of non-Hispanic whites and 15.7 percent of “other” (which includes Asians). A comparable range is shown by education, where 15.7 percent of college graduates are obese compared with 27.4 percent of those who did not graduate from high school. Obesity is an outcome, not a behavior. The relevant behaviors for obesity are food consumption (nutrition) and amount of exercise. Lower SES is related to lower rates of exercise and lower consumption rates of fresh fruits and vegetables (Krebs-Smith et al., 1995; Pamuk et al., 1998). These behaviors are multidetermined. For example, those with less education and income may have less access to information about the hazards of smoking or the benefits of a low-fat diet and exercise, as well as information about potential methods for quitting smoking or reducing weight. At the same time, those with less education and income are more likely to be living in neighborhoods that provide less access to the resources needed to maintain their weight or quit smoking. As noted above, low-income neighborhoods have fewer resources for achieving good nutrition or undertaking exercise. In such neighborhoods, concerns about safety may inhibit outdoor exercise. These neighborhoods may also be differentially targeted for advertising and marketing for cigarettes, alcohol, and junk foods (Morland et al., 2002; Stoddard et al., 1998).

TABLE D-17

Percentage Obese, 2001, by Race/ Ethnicity and Educational Level.

Exposure to stress. The contribution of the domains described above to premature mortality have all been quantified by the CDC. One domain was not part of their analysis, most likely because its effects on mortality are more difficult to quantify and overlap the other determinants. This domain is exposure to stress and the resultant behavioral and biological responses that put individuals at risk for a range of diseases (McEwen, 1998a). A number of studies document greater stress exposure for more disadvantaged groups, whether by race and ethnicity and/or by SES. Given this, differential stress exposure may be an important mechanism by which social disadvantage affects health. Stress can directly affect health through biological pathways as discussed below. It can also have indirect effects through its impact on health behaviors. One of the ways that individuals cope with stress is through the use of substances, including tobacco and alcohol, as well as through eating. Recent animal research (Dallman et al., 2003) has demonstrated that the ingestion of fat and sucrose may help to downregulate the biological cascade that occurs in response to stress. Chronic stress may thus foster weight gain and obesity. Moreover, in the presence of cortisol, which is part of the stress response, the resulting fat appears to be differentially deposited in the abdominal region. Abdominal adiposity is a risk factor for cardiovascular disease, stroke, and diabetes above and beyond the risk associated with overweight itself.

Much of the research on the physiological effects of stress exposure has examined responses to acute stress. This type of research is easier to model in laboratory settings because researchers can simulate an acute threat. However, the kind of stress associated with social disadvantage is more chronic. This includes the kinds of daily stressors that are more common for those with fewer resources. It also includes the chronic stress of experiences of discrimination. It is not simply the dramatic single events that signal discrimination but the day-today slights and ambiguities of unequal treatment (e.g., being stopped more often for routine traffic checks, being treated rudely by others, being followed by a store security guard, losing out on a promotion without explanation).

The concept of allostatic load was developed by McEwen and colleagues (McEwen, 1998b; McEwen and Stellar, 1993) to describe the biological processes involved in responses to chronic stress. This concept posits that over time, with repeated exposure to stress and engagement of the stress response systems, the affected systems may become dysregulated. Central to this process is the hypothalamic-pituitary-adrenal (HPA) axis and the regulation of cortisol. With repeated stress exposures, the HPA axis may become less flexible (McEwen and Seeman, 1999). Rather than showing the typical stress response (increased level of cortisol, which then downregulates the HPA axis to return cortisol to baseline), there may be either a prolonged response with a delayed return to baseline or a dampened initial response. The first pattern (delayed recovery) results in prolonged exposure to cortisol, which can increase visceral fat and raise the risk of diseases such as diabetes and cardiovascular disease. The second pattern (under-response) is less effective in resolving the stress response and may increase vulnerability to autoimmune diseases, including rheumatoid arthritis. The concept of allostatic load provides a summative measure of the cumulative effects of stress and may reflect the multiple biological pathways by which social disadvantage can affect a range of health outcomes. Initial research testing this concept has shown that allostatic load is related, on the one hand, to race/ethnicity and SES and, on the other, to health. Among participants in the CARDIA sample, allostatic load scores are higher among African Americans than among whites and are greater among those with less education than among the more educated (Seeman, 2004). Allostatic load scores, in turn, have been shown in a sample of older adults to predict physical and cognitive decline, the onset of new cardiovascular disease, and mortality over a 7-year period (Karlamangla et al., 2002; Seeman et al., 2001).

Remaining Issues

This review provides one overview of the vast data on health disparities. Different perspectives undoubtedly highlight different issues. This review demonstrated that the term health disparities is not being used in any single way. The papers on health disparities encompass research on socioeconomic factors, race and ethnicity, sex/gender, and rural health. The data point to the difficulty in specifying specific disparity groups. In examining a range of reports on overall mortality and on the prevalence of specific diseases, including the data used in the NIH Health Disparities Strategic Plan, Fiscal Years 2004–2008, the only racial/ethnic group that shows consistently poorer health across a range of indicators is African Americans. A caveat, however, is that the available data may not provide a full and accurate estimate of disparities. For some groups, for example, ethnicity may not be accurately captured in mortality data or in surveys, and this may lead to an undercounting of deaths or disease prevalence in these groups. There may be biases introduced for specific populations (e.g., a Mexican American health advantage may be due, in part, to the return of those who are ill or dying to Mexico). Also, as noted earlier, the identification of race and ethnicity into broad categories (e.g., Asian) may miss specific groups (e.g., Hmong) whose health status is markedly worse. This argues for greater attention paid to the nature of the data and the sampling being used to establish the degree of health disparities.

The data on disparities among other groups shows some consistency, although here, too, finer differentiations could yield more informative findings. Health in rural areas appears to be poorer than in more populated areas, though finer differentiators may be helpful in examining suburban versus urban areas. Health is also worse for those who are poorer and less educated, as well as those in low-SES occupations, and these factors account for much, but not all, of the health disadvantage experienced by different racial and ethnic groups and those in rural areas.

Policies to eliminate health disparities need to be informed by scientific understanding of their causes. The empirical and conceptual approaches to date have revealed that the poorer health of African Americans is largely, but not wholly, accounted for by socioeconomic disadvantage. This raises questions. What accounts for the remaining effect? One candidate is exposure to discrimination and racism, which may increase stress responses with their attendant health effects. Another candidate is inadequate measurement of socioeconomic disadvantage and the implications of SES for a range of environmental exposures. A second question: What aspects of socioeconomic disadvantage contribute to health disparities (for those that account for racial/ethnic disparities as well as those that operate for members of all racial and ethnic groups)? SES includes various aspects, each of which confers different resources and has different implications for health. In addition to individual-level factors (e.g., income, education, wealth/assets, occupation), both race/ethnicity and SES shape the area of residence and work environments, each of which has an additional effect on health. For example, residential segregation of African Americans has resulted in areas of concentrated poverty that have health-damaging effects. At the same time, recent research suggests that for Latinos, the barrio effect of greater ethnic density may be health-protective, despite the greater poverty in these areas (Eschbach et al., 2004). Within the work environment, physical conditions may contribute to the risk of injury or disease, as does the social organization of work and particularly the degree of control over demands (Bosma et al., 1997).

It should be noted that virtually all research on health disparities shows associations but does not establish causality. The challenges of establishing causality differ for various sociodemographic variables. Race/ethnicity and sex are determined at birth, and it is not plausible that these are affected by their own health status. With regard to socioeconomic factors, however, mutual causation between SES and health is possible, especially for income. When people become ill, they not only incur medical expenses but may also be less able to work. Smith (1999) demonstrated the adverse effect of poor health on income among participants in the Health and Retirement Study. In this sample of older adults, the most important factor in diminished income and wealth was early retirement due to health problems. Reverse causation is less plausible for education; disease later in life does not change earlier educational attainment. However, the life course models presented above suggest that health disparities early in life may, in turn, affect educational attainment that could both limit later SES and affect adult health. This suggests a reciprocal causal chain between SES and health, but one in which the socioeconomic factors are likely to be more fundamental (Link and Phelan, 1995).

There are some innovative approaches to establishing causal direction by forming links to experimental programs in which specific aspects of social disadvantage are reduced, and the health impact can be examined. For example, several Central and Latin American countries are embarking on anti-poverty programs that may yield health benefits. The PROGRESA project (now called OPORTUNIDADES) in Mexico has shown in a randomized social experiment that income supplementation tied to incentives for health-promoting behaviors—such as using prenatal care and pediatric check-ups and additional cash incentives to keep one’s children in school—has resulted in improved growth and decreased anemia in children (Gertler, 2004). Programs in other countries may tease apart the beneficial effects of income supplementation versus income supplementation linked to behavioral incentives. In the United States, the Moving to Opportunity program called for randomized housing-project residents to receive a voucher to allow them to move elsewhere only to a low-poverty area, or to a control condition with no voucher. Both children and adults randomized to the low-poverty neighborhood condition subsequently showed better mental health outcomes but not other health outcomes; there were more favorable outcomes for girls than for boys (Kling et al., 2004).

Other social experiments have not explicitly examined health effects but could be used to do so. For example, a few early childhood education programs, such as the Perry Preschool, had a randomized design that showed economic and social benefits for the children randomized to the experimental condition (Barnett, 1996; Reynolds et al., 2001). However, long-term health effects of enriched early education have not yet been demonstrated.

Social experiments such as those described above require collaboration across sectors and links between health research and housing, education, labor, and so forth. With the exception of massive national programs like PROGRESA, these will necessarily be on a small scale because they are expensive to implement. In addition, there will be a continuing need for sophisticated and creative approaches to examining causal effects in the context of observational studies, which will likely comprise the bulk of research. In this work, longitudinal studies will be important to help establish temporal ordering, as well as cross-sectional studies to provide initial evidence of associations and identify possible mediators. If the National Children’s Study is launched, it will be critical to have adequate measures of sociodemographic factors at each time point of data collection and measures of the psychosocial and environmental factors likely to shape health disparities in this population.

In such research, explicit models of health disparities should be specified. This will guide not only the selection of independent and mediating variables but also the health outcome to be studied. The latter may include mortality rates, life expectancy at birth or at different ages, the incidence or prevalence of specific diseases, functional status, and/or self-rated health. One of the questions in such research is whether to examine single diseases or multiple outcomes. Understanding disease-specific pathways is useful for delineating pathophysiological processes. However, data showing similar disparities across a range of diseases suggest that there may be some common pathways to multiple health outcomes. Some of the models and research reviewed above propose that exposure to stress is one such common pathway. Recent research linking greater stress to cell aging (Epel et al., 2004) provides some evidence that chronic stress may, indeed, lead to a type of accelerated aging that can increase risk for a number of diseases. Risk factors such as tobacco use and obesity (which are more common in more disadvantaged groups) may also serve as a common risk factor, as may environmental exposures. The linkage across diseases points to the need for greater cooperation across NIH institutes in supporting disparities research.

At the same time that research is needed on common pathways to multiple outcomes, some mechanisms may be unique to specific diseases. Not every disease shows the same associations with race/ethnicity, SES, etc. For some diseases, such as breast cancer and malignant melanoma, the usual SES gradient is reversed; these diseases are actually more common among more advantaged groups. One unexplained finding is why African Americans show more adverse outcomes in relation to physical health but often show lower rates of mental illness.

The pattern of associations with SES and race/ethnicity can also vary for different stages of disease. For example, higher-SES women are more likely to be diagnosed with breast cancer than women who are less well educated or affluent; this is a real difference in rates of onset and not simply due to better diagnosis. However, once diagnosed, higher-SES women have a greater length of survival, even when controlling for the stage of disease at diagnosis. Thus, it may be useful to look at predictors of different components of mortality associated with a given disease and take into account disparities in incidence and survival. More common than reversals in associations is the finding that the degree of disparities varies for different diseases. For example, the SES gradient is steeper for cardiovascular disease than for many cancers. Within cancer, the gradient is steeper for cervical cancer than for other types of cancer. As researchers identify disease-specific pathways that may account for disparities, they may also learn much by comparing the nature and degree of disparities across diseases. Finally, new approaches to measuring health outcomes also exist. Social disadvantage has a pervasive impact on a variety of risk factors and diseases. In addition to identifying common pathways, it would be helpful to have a valid measure of health capital, a summative measure of the overall health and functioning of individuals that could be aggregated to assess the health stock of groups. This would require operationalizing the World Health Organization (1948) definition of health. There are early efforts to do so, primarily by health services researchers who have developed instruments such as the 36-Item Short-Form Health Survey (SF-36). The next generation of measures should be done with an eye to their applicability to evaluating health disparities. Collaboration between NIH and the Agency for Healthcare Research and Quality (AHRQ), which uses such measures more frequently, may also prove beneficial.

Implications of This Analysis for the NIH Research Agenda

Researchers would find it difficult to implement the idea that the healthiest group be taken as the standard against which other groups could be evaluated, in terms of the extent of their health disparity. However, this strategy may be worth discussing. It shares common ground with that suggested by Murray et al. (1999), but it incorporates a focus on groups that may address the concern raised by Braveman et al. (2001) that the health status of disadvantaged groups could be overlooked. Such an approach may have the potential to stimulate novel research and provide information on the strengths of groups that could help inform others (e.g., understanding the Hispanic paradox may provide clues to health-protective social and cultural processes).

The data presented in this paper underline the importance of collaboration across NIH institutes, because health disparities cross-cut multiple diseases and populations. These data also suggest that a strategy based on disparity groups is not as likely to be fruitful as one based on disparity processes. Specifically, understanding the interrelationships and interactions among different sources of social disadvantage (which includes race/ethnicity, SES, gender, and area of residence) will provide a fuller explanation of the mechanisms by which disparities occur. The existing data suggest that socioeconomic disadvantage is a key pathway by which racial/ethnic disparities emerge. At the same time, African Americans show poorer health outcomes even when SES is adjusted for. There may be more impact from research on socioeconomic disadvantage because it is the more powerful effect and is more amenable to intervention. However, it is also important to understand what it is about the experiences of African Americans that places them at heightened risk above and beyond that associated with their socioeconomic position. This review makes clear the importance of encompassing the measurement of race/ethnicity, SES, and gender in research.

To achieve the dual goals of Healthy People 2010 (U.S. Department of Health and Human Services, 2000), we will need more research—and, importantly, more sophisticated research—on understanding the pathways by which health disparities are created. This work will be facilitated by greater inclusion of appropriate measures of SES as well as race and ethnicity in national data sets and public health monitoring measures in addition to gender and area of residence. Additionally, to the extent possible, measures of psychosocial and behavioral variables that are likely to mediate these effects should be included. Strategies that involve the measurement of risk factors and preclinical indicators of disease states will be particularly important, as these may provide information on common underlying pathways to multiple diseases, as well as information on disease-specific risk states that can suggest strategies for earlier intervention. The examination of common pathways to multiple diseases underlines the importance of coordinating health disparities research across the NIH institutes, as well as the AHRQ.