ARTICLE IN PRESS. 676 N. St. Clair Street, Rm. 255, Chicago, IL 60611, USA. Available online 2 August 2005

Size: px
Start display at page:

Download "ARTICLE IN PRESS. 676 N. St. Clair Street, Rm. 255, Chicago, IL 60611, USA. Available online 2 August 2005"


1 Social Science & Medicine 62 (2006) Explaining US racial/ethnic disparities in health declines and mortality in late middle age: The roles of socioeconomic status, health behaviors, and health insurance Joseph J. Sudano a,b,, David W. Baker c a Center for Health Care Research and Policy, Case Western Reserve University at The MetroHealth System, Rammelkamp 236, 2500 MetroHealth Drive, Cleveland, OH , USA b Departments of Medicine and Epidemiology and Biostatistics, Case School of Medicine, Case Western Reserve University, Cleveland, OH, USA c Division of General Internal Medicine, Feinberg School of Medicine, Northwestern University, 676 N. St. Clair Street, Rm. 255, Chicago, IL 60611, USA Available online 2 August 2005 Abstract Pervasive health disparities continue to exist among racial/ethnic minority groups, but the factors related to these disparities have not been fully elucidated. We undertook this prospective cohort study to determine the independent contributions of socioeconomic status (SES), health behaviors, and health insurance in explaining racial/ethnic disparities in mortality and health declines. Our study period was , and our study population consists of a US nationally representative sample of 6286 non-hispanic whites (W), 1391 non-hispanic blacks (B), 405 Hispanics interviewed in English (H/E), and 318 Hispanics interviewed in Spanish (H/S), ages in 1992 in the Health and Retirement Study. The main outcome measures were death; major decline in self-reported overall health (SROH); and combined outcome of death or major decline in SROH. Crude mortality rates over the 6-year study period for W, B, H/E and H/S were 5.8%, 10.6%, 5.8%, and 4.4%, respectively. Rates of major decline in SROH were 14.6%, 23.2%, 22.1% and 39.4%, for W, B, H/E and H/S, respectively. Higher mortality rates for B versus W were mostly explained by worse baseline health. For major decline in SROH, education, income, and net worth independently explained more of the disparities for all three minority groups as compared to health behaviors and insurance, reducing the effect for B and H/E to non-significance, while leaving a significant elevated odds ratio for H/S. Without addressing the as-yet undetermined and pernicious effects of lower SES, public health initiatives that promote changing individual health behaviors and increasing rates of insurance coverage among blacks and Hispanics will not eliminate racial/ethnic health disparities. r 2005 Elsevier Ltd. All rights reserved. Keywords: Racial/ethnic health disparities; Hispanics; Socioeconomic status; Health behaviors; Insurance; Self-reported health declines; USA; Cohort study Corresponding author. Center for Health Care Research and Policy, The MetroHealth System, Rammelkamp 236, 2500 MetroHealth Drive, Cleveland, OH , USA. Tel.: ; fax: addresses: (J.J. Sudano), (D.W. Baker) /$ - see front matter r 2005 Elsevier Ltd. All rights reserved. doi: /j.socscimed

2 910 J.J. Sudano, D.W. Baker / Social Science & Medicine 62 (2006) Introduction Over the past several decades, studies have accumulated overwhelming evidence of mortality and health status disparities between racial and ethnic groups (Bassett & Krieger, 1986; Feldman & Fulwood, 1999; Geronimus, Bound, Waidmann, Hillemeier, & Burns, 1996; Schulz et al., 2000; Smith & Kington, 1997). Conceptual models have emerged in the literature that attempt to explain these differences (Mendes de Leon & Glass, 2004; Schulz, Williams, Israel, & Lempert, 2002). Generally, these explanatory models focus on how upstream macrosocial factors (e.g., racism and racial discrimination, racial residential segregation, social stratification) structure and limit the personal resources (e.g., educational, occupational, financial) available to racial and ethnic minority group members, and how in turn various downstream social and individual level mechanisms (e.g., stress exposure, lack of medical care, unhealthy lifestyles) impact health outcomes over time. Several key factors have emerged and been identified as the primary social pathways and processes that affect health. These include low socioeconomic status (SES) (Lantz et al., 1998; Schulz et al., 2002; Williams & Collins, 1995), adverse health behaviors (Black, Ray, & Markides, 1999; Lantz et al., 2001), and lack of health insurance (Baker, Sudano, Albert, Borawski, & Dor, 2001; Monheit & Vistnes, 2000). To the extent that these adverse health factors tend to cluster together and are more prevalent among racial/ethnic groups (Lynch, Kaplan, & Salonen, 1997), they may explain the observed disparities in health outcomes between minority group members when compared to majority group members, i.e., typically non-hispanic whites in the United States (Mendes de Leon & Glass, 2004). However, several limitations in the studies listed above have prevented a more complete understanding and elucidation of the causal relationships between these factors and the observed racial/ethnic disparities. These include: variations in study design (e.g., cross-sectional vs. longitudinal); focus on only one particular health outcome (typically mortality); the coarse or inadequate measurement of key factors (usually SES but also baseline health status); aggregating very distinct racial/ ethnic sub-groups (e.g., more or less acculturated Hispanic groups); or the omission of key factors from the analysis (e.g., insurance status). In the background section that follows, we first elaborate on several of these issues, and then finish with the research questions we investigate in this study. Background Any examination of racial/ethnic disparities in health outcomes must necessarily include an investigation of socioeconomic (SES) determinants because of the strong correlations between SES and health, and SES and race/ ethnicity. For example, because African Americans are disproportionately poorer and have less education compared to non-hispanic whites in the United States, SES differentials confound the relationship between race and health outcomes. Therefore, if inadequate or coarse controls for SES are employed, researchers will underestimate the effect of SES. This results in a residual race effect, leaving the causes of health disparities only partly explained, and often implicitly attributing the unexplained portion to some underlying genetic difference (Kaufman, Cooper, & McGee, 1997). Among the variables available to measure SES, education and income are the most used. Education is relatively easy to measure and stable over time; and it is positively related to a variety of health outcomes, both perceived and objectively measured (Ross & Mirowsky, 1999). Education contributes to an array of resources that are salutary to health, including cognitive and problem solving skills; self-directedness, self-efficacy and personal sense of control; healthy lifestyles; and more lucrative and health-sustaining occupations (Mirowsky & Ross, 1998; Ross & Wu, 1996). Moreover, these salutary relationships persist notwithstanding some salient racial and gender variations in financial returns from educational attainment (Williams & Collins, 1995). Regarding financial resources, income is the most commonly used whether measured at the individual or household level. There is a longstanding positive relationship between income and health in the literature (Lynch, Kaplan, & Salonen, 1997; Lynch, Kaplan, & Shema, 1997; Lynch & Kaplan, 2000), but the use of contemporaneous income (i.e., typically income measured in the year prior to a survey) does pose several measurement challenges when studying health outcomes over time. First, it is highly volatile over the lifecourse, particularly for those persons in the service and labor sectors of the economy, who are also disproportionately members of racial/ethnic minority groups. Second, current income may be susceptible to reverse causation in its relationship to health, where declines in health may produce lower income. Hence, some authors suggest measures of accumulated wealth may be better indicators of financial resources available for health and health-related decisions across the lifecourse, particularly for aging adults (Duncan, Daly, McDonough, & Williams, 2002). There are several pathways by which wealth may affect health. First, it may be a marker for an intergenerational transfer of advantage including health stock and lifechances during the developmental years, through childhood and adolescence, and into early adulthood (Lynch & Kaplan, 2000). Alternatively, the lack of wealth when measured in adulthood may be a marker for past economic hardship, and poorer health stock and fewer

3 J.J. Sudano, D.W. Baker / Social Science & Medicine 62 (2006) life-chances. Second, it can also be a measure of the ability to sustain unanticipated economic downturns such as job loss, lay offs, health shocks, and other events that require large expenditures (Chiteji, 2004). A discussion of the impact of wealth on health would be moot if there were no variation across racial/ethnic groups in the United States. That is far from reality: large differences in both income and wealth exist. For example, median household incomes for 1998 were $42,440, $25,350 and $28,330 for non-hispanic whites, non-hispanic blacks, and Hispanics, respectively (US- Census Bureau, 2004). Even more dramatic inequalities in wealth statistics existed in 1998: median net worth (including home equity) was $71,000, $6200, and $7190 and median non-home equity assets (liquid assets, savings, etc.) were $20,100, $1020 and $1820 for non- Hispanic whites, non-hispanic blacks, and Hispanics, respectively (Orzechowski & Sepielli, 2003). For these reasons, it is imperative that investigations of racial/ ethnic disparities in health include several measures of SES, including education, income and wealth. For residents of the United States where universal health insurance coverage does not yet exist, insurance status represents another important factor in health outcomes, particularly for older adults. Previous studies have shown that adults age years old who lack health insurance have higher risk-adjusted rates of decline in their overall health and physical functioning (Baker et al., 2001) and higher risk-adjusted mortality compared to individuals with private insurance (McWilliams, Zaslavsky, Meara, & Ayanian, 2004). Likewise, reduced access to care can have serious consequences for health outcomes via lack of preventive services use, delayed diagnosis of disease, and poor monitoring and control of chronic diseases (Committee on the Consequences of Uninsurance, 2002). Non-elderly Hispanics had the highest rates of uninsurance among any ethnic group in the United States in 1998: 32%, compared to 12% for whites and 19% for blacks (National Center for Health Statistics, 2004). These rates have improved only modestly over the past several years, with public insurance replacing most of the employer-sponsored decreases in coverage for blacks and Hispanics. Parenthetically, even with higher rates of insurance coverage albeit public in nature blacks and Hispanics continue to report greater barriers and reduced access to care when compared to white (Hargraves, 2004). Coupling these factors with lower satisfaction with the quality of care they do receive (Johnson, Saha, Arbelaez, Beach, & Cooper, 2004), Hispanics have higher insurance-related risks for adverse health outcomes compared to their non-hispanic white and black counterparts. This trend is expected to continue as employersponsored insurance coverage continues to drop and more individuals obtain public coverage or are forced into the ranks of the uninsured (Hargraves, 2004). Finally, the relationships between racial/ethnic group membership, SES, and health outcomes is sometimes paradoxical and often complex. For example, many members of Hispanic sub-groups (particularly Mexican- Americans) traditionally have an SES profile similar to that of African Americans, report poorer self-assessed health than non-hispanic whites, and yet have mortality rates similar to the non-hispanic white population (Franzini, Ribble, & Keddie, 2001; Markides & Coreil, 1986). This relationship seems to be even more pronounced for Hispanics who respond to interviews in Spanish (Finch, Hummer, Reindl, & Vega, 2002). No longitudinal study to date has simultaneously examined the contributions of (1) health status, (2) SES, (3) health behaviors, and (4) health insurance for explaining racial/ethnic disparities in mortality and health decline among a diverse, nationally representative sample of older adults. This study uses data from the Health and Retirement Study (HRS) to go beyond prior studies and to answer the following research questions: (1) For older adults, do health status, SES, health behaviors and health insurance differ significantly across US racial/ethnic groups? (2) How much of the observed racial/ethnic disparity in health outcomes is explained by each set of factors above and do these factors completely explain the observed disparity? (3) Is this relationship consistent across different health outcomes (mortality and health declines)? Methods Data and study population For this study we used data from Wave 1 (1992) and Wave 4 (1998) of the HRS, a longitudinal, nationally representative, multistage area probability sample of US households. Supported by the National Institute on Aging and conducted by the University of Michigan s Institute for Social Research, the first wave of data collection targeted non-institutionalized persons in the contiguous US ages (born between ) and their spouses. For the initial interview, in-home, face-to-face interviews were conducted for 7608 households yielding 12,652 individual respondents (an overall response rate of 82%). Subjects were subsequently interviewed every two years. By design, the HRS staff provided sampling and analytical weights only for the targeted age-eligible respondents and their ageeligible spouses/partners to adjust for over sampling and non-response bias. A more detailed description of the overall study design and sampling methods are provided elsewhere (Heeringa & Connor, 1995).

4 912 J.J. Sudano, D.W. Baker / Social Science & Medicine 62 (2006) Because of the HRS complex survey design and our desire to produce nationally representative estimates for this age group in our final analysis, we were necessarily limited to using only those respondents for whom sampling and analysis weights were provided. These were the targeted age-eligible respondents and spouses/partners mentioned above, ages 51 61, numbering 9824 individuals. Next, 75 respondents had incomplete records for either the 1992 or 1998 interview. An additional 1349 (13.7%) were lost to follow-up over the 6-year period. This left us with 8400 respondents in our final sample. Statistical analysis All analyses used the survey subroutines available in Stata 7.0 (Stata Corporation, College Station, TX) to account for sample weighting and to adjust variances for the complex survey design of the HRS. Multivariate logistic regression modeling was conducted using the svylogit procedure in Stata version 7.0. A p-value of 0.05 was used to determine statistical significance. Analytic scheme Our analysis was conducted in three stages. First, we conducted bivariate analysis of demographic, socioeconomic, health behavior and health status characteristics by race/ethnicity (Table 1). Second, we analyzed the distributional characteristics of outcome measures across racial/ethnic groups (Table 2). Third, multivariate analysis of each outcome measure employed 5 models. We first specified the unadjusted relative risk (model 1) for outcomes by racial/ethnic group compared to whites. Next, we specified a core model equation (model 2) that includes 1992 baseline control variables and covariates: age, sex, marital status, self-reported overall health (SROH), number of physical limitations, and number of chronic diseases. We then employed a series of nested logistic regression models where independent variables of interest (health behaviors, insurance, and SES) are sequentially added to the core model (models 3 5). This procedure allows us to examine the incremental reduction effects of these variables on the unadjusted racial/ethnic relative risks (Tables 3 5). Outcome measures Three outcome measures were used: death (all cause mortality) between , major decline in SROH between 1992 and 1998, and a combined measure of major decline or death over the study period. First, death was determined using National Death Index and household contacts. Second, SROH was assessed with the Excellent/Very Good/Good/Fair/Poor questionnaire response format at each interview. A major decline in SROH was defined as follows: (1) a decline from excellent/very good/good health in 1992 to fair/poor health in 1998; or (2) a decline from fair health in 1992 to poor in 1998 (Baker et al., 2001). Because this variable excludes individuals who died between 1992 and 1998, as well as individuals at the floor of health status in 1992, we also used a third outcome variable a combined major decline or death. We also used these three variables because it is possible that the effects of the independent variables of interest may be different for all-cause mortality compared to declines in health. Adjustments were made to odds ratios to more accurately reflect true relative risk using a published formula (Zhang & Yu, 1998). Main Independent variable: A short excursus into race and ethnicity Our operationalization of an individual s race/ethnicity was classification based on two self-reported selfidentification items in the HRS questionnaire. First, respondents were asked Do you consider yourself Hispanic or Latino? and second, Do you consider yourself primarily white or Caucasian, black or African American? If an individual answered yes to the first question, they were classified as Hispanic regardless of their answer to the subsequent question. In the dataset used for this analysis, 18 people identified themselves as Hispanic or Latino and black or African American and were classified as Hispanic. Many authors suggest caution when using race and ethnicity in health care research because it may be unclear what these variables represent. For example, studies may inadvertently use race/ethnicity as surrogates for income, education, occupation, health behaviors, or health insurance coverage if they do not have data to explicitly adjust for these variables. The detailed information in the HRS on SES, health behaviors and insurance status allows us to conceptualize race and ethnicity as complex social variables (Hummer, 1996). In essence, the labeling of persons as white, black, or Hispanic suggests a particular social status that to some degree marks the social contexts in which persons live out their lives. Elements of these social contexts might include variations in exposure to prejudice, discrimination and segregation, cultural differences and language barriers, health beliefs, and health care seeking behaviors (Herman, 1996; Williams, 1996). However, our ability to measure these domains is clearly limited by the HRS data and hence we use race/ ethnicity here as a surrogate for this complex of important social factors. Likewise, we recognize that the race/ethnicity categories available in our data imply a measure of homogeneity within groupings that may belie the empirical record. For example, there are important

5 J.J. Sudano, D.W. Baker / Social Science & Medicine 62 (2006) Table 1 Baseline 1992 study population characteristics by racial/ethnic group in the HRS (N ¼ 8400) Characteristic Whites (N ¼ 6286) Blacks (N ¼ 1391) Hispanic/English (N ¼ 405) Hispanic/Spanish (N ¼ 318) Age, mean years (SE) (0.05) (0.10) (0.20) (0.16) Female sex (%) ***,a Marital status (%) Never married ***,a 5.5**,a 3.7**,a Sep., div., widowed Married Education, mean years ***,b 10.7***,b 6.2***,b (SE) (0.07) (0.15) (0.29) (0.38) Income-to-needs ratio, mean ***,b 3.98***,b 2.23***,b (SE) (0.14) (0.11) (0.26) (0.19) Net worth in quintiles, mean ***,b 2.6***,b 1.7***,b (SE) (0.04) (0.05) (0.08) (0.09) Insurance status 1992, 1994, and 1996 Insured at all 3 interviews ***,a 67.3***,a 46.6***,a Intermittently insured Uninsured at all 3 interviews Smoking (%) Never ***,a ***,a Past smoker Current smoker Alcohol consumption pattern (%) Abstainer ***,a ***,a Moderate Heavy CAGE score (%) None ***,a One indicator Two or more indicators BMI quintiles (%) Lowest ***,a 15.5**,a 10.9***,a 2 nd rd th Highest Self-reported overall health (%) Poor ***,a 7.5***,a 16.7***,a Fair Good Very good Excellent Chronic disease count, mean ***,b (SE) (0.02) (0.05) (0.06) (0.09) Physical limitations count, mean ***,b 3.3**,b 4.0***,b (SE) (0.07) (0.15) (0.29) (0.26) Notes: Results are from weighted analyses using survey estimation procedures in Stata 7.0. Significance tests for categorical variables are based on the Pearson w 2 -statistic corrected for survey design and the test for continuous variables are based on the t-statistic. Income-to-needs ratio is based on household income in 1991 divided by the poverty guideline for a given household size. A value of 5.00 for example indicates a household income 5 times (or alternative, 500% of) the poverty guideline for a given household size. BMI ¼ Body mass index (kg/m 2 ) and was used in the construction of quintiles. *po0:05, **po0:01, ***po0:001. a Statistical significance is for trends compared to whites across all categories of the variables. b Statistical significance is compared to whites.

6 914 J.J. Sudano, D.W. Baker / Social Science & Medicine 62 (2006) cultural differences between Mexican Americans, Cuban Americans, and Puerto Ricans that could affect their health outcomes. However, in the HRS the number of Hispanics other than Mexican-American was too small to allow subgroup analysis. Combining these groups into a single category of Hispanics would ignore these differences, and hence we chose to divide Hispanics into two groups based on their choice of language for the initial interview in Wave 1 (1992). This method (a) differentiates two very different populations (Arcia, Skinner, Bailey, & Correa, 2001; Fiscella, Franks, Doescher, & Saver, 2002), and (b) uses language as a proxy for acculturation/assimilation (Arcia et al., 2001). We label these two Hispanic subgroups Hispanic/ English and Hispanic/Spanish. We also recognize that there are likely to be important cultural differences between whites and blacks of different ethnic backgrounds that may affect their health care use and that are not captured by other variables in the data. In sum, we recognize that our aggregation of persons into the socially constructed groupings provided is less than ideal. We chose therefore to include as many factors present in the HRS data that could be considered covariates of race/ethnicity (e.g., education, income, wealth, etc.) both between and within groups, and to conceptualize race/ethnicity as a surrogate for unmeasured social and cultural factors. Demographic variables Respondents age in years (continuous) was recorded from the 1992 interview. Marital status was categorized as married (referent group), separated/divorced/widowed, or never married; sex was categorized as female or male (referent group). SES measurement Educational attainment was measured in number of years of formal schooling completed. We computed an adjusted income measure for each respondent an income-to-needs ratio (INR) that incorporates total household income divided by the poverty guideline for a given size household. As an example, an INR of 3.0 is equivalent to an income 300% of the poverty guidelines for a given family size. Net worth as of 1992 was first measured as a continuous variable and then collapsed into quintiles. Health behavior and health insurance variables Health behaviors include a smoking behavior measure (never smoked as referent); past problem drinking behavior measured using the CAGE index (none as referent) (Mayfield, McLeod, & Hall, 1974); a measure of current drinking pattern based on number of drinks per day categorized as abstainer, moderate (referent), and heavy drinking; and body mass index (BMI) quintiles used as indicators of weight problems/obesity. Finally, we used a measure of intermittent insurance status, first determining the existence of any kind of health insurance coverage (public or private) in 1992, 1994, and 1996, and then categorizing individuals as (a) insured all three interviews (referent), (b) 1 2 reports of uninsurance, or (c) uninsured all three interviews. Health status variables All multivariate models adjusted for three health status variables based on self-reports from the 1992 interview. These included SROH (excellent health as referent), a count of total number of chronic diseases (hypertension, diabetes, health disease, chronic lung disease, cancer, arthritis, stroke, and visual difficulties) (Verbrugge, Lepkowski, & Imanaka, 1989), and number of physical limitations. Two sets of questions previously described by HRS investigators assess the number of physical limitations (Fillenbaum, Burchett, & Welsh, 1993) and include four items assessing difficulties with mobility (activities requiring large muscle strength) and six items assessing difficulties with agility (activities required to perform instrumental activities of daily living). We have reported details on the items in a previous study (Baker et al., 2001). Results Study population characteristics Our final sample included 6286 non-hispanic whites, 1391 non-hispanic blacks, 405 Hispanics interview in English, and 318 Hispanics interviewed in Spanish. Those lost to follow up were slightly more likely to be minority group members, and to have lower incomes and lower educational attainment compared to those who remained in the study. No other significant differences were found between these two populations in any other characteristics we investigated. There were large and significant differences between the racial/ethnic groups in SES and insurance status (Table 1). Blacks, Hispanic/English and Hispanic/ Spanish were more socioeconomically disadvantaged compared to Whites in each of the SES-component variables. Whites were somewhat more likely than blacks and Hispanic/English to be insured at all three interviews (82.3%, 71.7%, and 67.3%, respectively); Hispanic/Spanish were much less likely to be insured at all interviews (46.6%). The percentage of people who were uninsured at all three interviews showed the opposite trend (4.1%, 7.1%, 10.2%, and 27.7% for

7 J.J. Sudano, D.W. Baker / Social Science & Medicine 62 (2006) whites, blacks, Hispanic/English and Hispanic/Spanish, respectively). Differences in health behaviors across the three racial/ ethnic minority groups were more complex and countervailing when compared to whites. Blacks were more likely to be current smokers, while rates for Hispanic/ English and Hispanic/Spanish were similar to whites. There were no significant differences between Hispanic/ English respondents in either alcohol consumption patterns or in CAGE indicator scores compared to whites. Both blacks and Hispanic/Spanish respondents were more likely to report alcohol abstinence; however, they also were more likely to have histories of past drinking problems. All three minority groups were more likely to be overweight/obese compared to whites. Finally, all three minority groups were more likely to report fair or poor health and elevated mean numbers of physical limitations compared to whites. Only blacks reported elevated mean numbers of chronic diseases compared to Whites. Synopsis of outcomes: Death, major decline, and major decline or death A total of 581 deaths occurred over the study period, including 383 whites, 153 blacks, 29 Hispanic/English and 16 Hispanic/Spanish (Table 2). Blacks had unadjusted relative risks nearly twice that of whites (1.82, 95% CI ¼ ), while the relative risks for the 2 Hispanic groups were equal to or less than that for whites (0.99, 95% CI ¼ and 0.75, 95% CI ¼ , for Hispanic/English and Hispanic/ Spanish, respectively). Across the study period, 1285 persons reported a major decline in overall health, including 828 whites, 263 blacks, 80 Hispanic/English and 94 Hispanic/Spanish. All three minority groups had elevated risks for a major decline compared to whites. Relative risks for blacks and Hispanic//English were elevated compared to whites, and were similar in magnitude to each other (1.56, 95% CI ¼ and 1.52, 95% CI ¼ , respectively). Those for Hispanic/Spanish were more than 2.5 times that for whites (2.70, 95% CI ¼ ). A total of 1846 combined deaths and major declines occurred over the study period: 1211 for whites, 416 for blacks, 109 and 110 for Hispanic/English and Hispanic/ Spanish, respectively, with elevated relative risks for the three minority groups as expected (1.53, 95% CI ¼ ; 1.34, 95% CI ¼ ; and 1.93, 95% CI ¼ for blacks, English and Spanishspeaking Hispanics, respectively). Nested models for death For blacks, adding baseline variables (model 2) reduced the unadjusted relative risk to a statistically Table 2 Sample sizes, percentages and unadjusted relative risks for death, major decline in self-reported overall health (SROH), and combined outcome (major decline or death) by racial/ethnic group, in the HRS, Outcome Racial/Ethnic group membership White Black Hispanic/English Hispanic/Spanish Death as of 1998 (N ¼ 8400) Unweighted sample size No. of deaths (n ¼ 581) Weighted %, (95% CI) 5.8 ( ) 10.6*** ( ) 5.8 ( ) 4.4 ( ) Unadjusted RR (95% CI) Referent 1.82*** ( ) 0.99 ( ) 0.75 ( ) Major decline in SROH (N ¼ 7297) Unweighted sample size No. with major decline (n ¼ 1265) Weighted %, (95% CI) 14.6 ( ) 23.2*** ( ) 22.1** ( ) 39.4*** ( ) Unadjusted RR (95% CI) Referent 1.56*** ( ) 1.52** ( ) 2.70*** ( ) Major decline or death (N ¼ 8400) Unweighted sample size No. with major decline or death (n ¼ 1846) Weighted %, (95% CI) 18.9 ( ) 29.0*** ( ) 25.3** ( ) 36.4*** ( ) Unadjusted RR (95% CI) Referent 1.53*** ( ) 1.34** ( ) 1.93*** ( ) Notes: RR ¼ relative risk, CI ¼ confidence interval, Unadjusted RR is corrected relative risk following the formula of Zhang and Yu (see Methods section for citation), derived from logistic regression. All statistical tests are compared to the referent group ( white ). *po0:05; **po0:01; ***po0:001.