Baseline Health, Socioeconomic Status, and 10

4 downloads 0 Views 266KB Size Report
housework, including ''walking, dancing, gardening, golfing, or bowling.'' Vigorous exercise ..... paratively healthier in older age (Beckett, 2000). However, re-.
Journal of Gerontology: SOCIAL SCIENCES 2007, Vol. 62B, No. 4, S209–S217

Copyright 2007 by The Gerontological Society of America

Baseline Health, Socioeconomic Status, and 10-Year Mortality Among Older Middle-Aged Americans: Findings From the Health and Retirement Study, 1992–2002 Joe Feinglass,1 Suru Lin,1 Jason Thompson,1 Joseph Sudano,2 Dorothy Dunlop,1 Jing Song,1 and David W. Baker1 1

Northwestern University Feinberg School of Medicine, Chicago, Illinois. 2 Case Western University, Cleveland, Ohio.

Objectives. This study analyzed whether socioeconomic status in older middle age continues to be associated with 10-year survival after data are controlled for baseline health status. Methods. We confirmed deaths through 2002 for 9,759 participants in the Health and Retirement Study, aged 51 to 61 in 1992. We used discrete time survival models to examine hazard ratios over 10 years of follow-up. We examined associations of demographic characteristics and socioeconomic status measures before and after adjustment by health status and behavioral risk factors. Results. The 10-year mortality rate was 10.9%, ranging from 4.7% for respondents reporting excellent health to 35.8% for those reporting poor health at baseline. Lower levels of education, income, and wealth were strongly associated with higher mortality risk after we controlled for just demographic characteristics. After further adjustment for health status and behavioral risk factors, only household income remained significant. Discussion. Baseline health by age 50 is an important pathway in the association between midlife socioeconomic status and mortality risk to age 70. The continuing effect of low household income on mortality risk was concentrated among respondents reporting excellent to good health at baseline. Socioeconomic disparities in middle-age health continue to limit disability-free life expectancy at older ages.

O

LDER Americans are living longer, more disability-free lives. Americans aged 45 and older have accounted for two thirds of the overall increase in U.S. life expectancy since 1960 (Cutler & Meara, 2001). However, it is well known that the benefits of the compression of morbidity and increased life expectancy in America are very unevenly distributed by social class. Especially among older Americans, the gradient of health status and survival outcomes across educational, occupational, income, and wealth inequalities may explain much of the more highly publicized racial disparities in health. Despite overall improvement, socioeconomic inequalities in disability-free life expectancy in the United States have widened over the past 20 years (Schoeni, Martin, Andreski, & Freedman, 2005; Singh & Siahpush, 2006). In interpreting these trends, social epidemiologists have begun to construct what Alwin and Wray described as a lifespan developmental perspective on the relationship of socioeconomic status (SES) to health (Alwin & Wray, 2005). This approach emphasizes the particular timing across the life span of economic, social, and psychological exposure pathways, related to social status, that affect population health and mortality across generations. Within just the past few years, this theoretical framework has produced a wave of studies documenting the enduring effects of early life social status on adult health and life expectancy. This literature has ranged from the effects of parental social class and birth weight, to early

childhood exposures, to cognitive or physical deprivation (Barker, Osmond, Forsen, Kajantie, & Eriksson, 2005; Batty, Der, Macintyre, & Deary, 2006; Galobardes, Lynch, & Davey Smith, 2004; Krieger, Chen, Coull, & Selby, 2005; Luo & Waite, 2005; O’Rand & Hamil-Luker, 2005; Pollitt, Rose, & Kaufman, 2005). Researchers now know that by young adulthood, there are already very substantial SES differences in both traditional cardiovascular disease risk and inflammatory biomarkers (Banks, Marmot, Oldfield, & Smith, 2006; Koster et al., 2006; Krieger, Chen, Coull, et al., 2005; Yan, Liu, et al., 2006). However, the specific effects of SES in later life on older age health and mortality remain controversial (Hoffman, 2005). This is, in part, because none of the landmark cohort studies that originally documented the association of SES with mortality among older Americans included or controlled for baseline health in midlife. Most commonly, these studies relied exclusively on educational attainment as a marker for social status or, more rarely, household income or wealth. Midlife SES measures were then associated with mortality over highly variable follow-up periods (Backlund, Sorlie, & Johnson, 1999; Bond Huie, Krueger, Rogers, & Hummer, 2003; Daly, Duncan, McDonough, & Williams, 2002; Liao, McGee, Kaufman, Cao, & Cooper, 1999; Pappas, Queen, Hadden, & Fisher, 1993; Sorlie, Backlund, & Keller, 1995; Wong, Shapiro, Boscardin, & Ettner, 2002). However, none of these studies included detailed

S209

S210

FEINGLASS ET AL.

measures of midlife health and were thus unable to directly assess the potential confounding of SES with health at midlife cohort inception. As a result, this literature captured the entire cumulative effect of social status on mortality occurring across the entire life course, including long before baseline SES data were collected. Researchers’ new understanding of the health effects of childhood social status, as well as the growing awareness of the importance of early-life subclinical disease progression, have raised questions about the extent to which traditional midlife SES measures indeed remain independent predictors of mortality after data are controlled for health conditions already embodied by age 50 (Krieger, 2005). It may be that social and economic inequalities have already taken their toll by older middle age. Health status may already be so well established across the socioeconomic gradient by middle age, and poor health already so confounded with low SES, that individuals’ social position matters little by older middle age (J. P. Smith, 1998). From a life-span developmental perspective, another major weakness of the historic literature on midlife SES and mortality is that by middle age, factors such as education, income, and wealth may operate very differently for individuals already at different levels of baseline health. Educational disparities have often been the sole measure of SES in mortality studies, especially after education was included in U.S. death certificate data. However, education effects on health can operate through many facets of cumulative economic opportunity, for example, exposure to occupational risk factors or the availability of health insurance and preventive care from secure employment. Educational differences in social networks, community influences, individual self-efficacy, and other protective psychosocial attributes are also closely linked to the prevalence of behavioral risk factors over many years of adulthood (Wray, Alwin, & McCammon, 2005). Data from the Americans’ Changing Lives Study and other studies have indicated that although education is strongly associated with the onset of functional disability, only income remains associated with progression of functional impairment at older ages (House, Lantz, & Herd, 2005; Zimmer & House, 2003). Researchers have very rarely studied health disparities related to the distribution of household wealth, but have found them to be strongly associated with mortality (Bond Huie et al., 2003; Daly et al., 2002). By older middle age, when the incidence of symptomatic chronic conditions increases dramatically, persons with higher levels of income and wealth may be more successful at self-management and engaging social support, have better access to higher quality medical care, and have a more comfortable retirement lifestyle, all of which may serve to limit disease progression (Goldman & Smith, 2002; Pincus, Esther, DeWalt, & Callahan, 1998). However, even this apparent association of older middle-age SES with mortality has yet to be tested when controlling for baseline health status. The present study describes all-cause mortality over a 10-year follow-up period, using data from 9,759 participants in the Health and Retirement Study (HRS), aged 51 to 61 in 1992, with mortality follow-up across five biannual survey waves through 2002. Findings from discrete time survival models address the question of the relative association of all-cause mortality risk with individuals’ years of school

completed, household income, and net assets (wealth) after the data were controlled for respondents’ demographic characteristics. We then tested these associations after further controlling for baseline health status and behavioral risk factors. The results provide insight into the extent to which the wellknown association of SES with mortality is primarily due to earlier life effects on health status and behaviors versus ongoing exposure to economic and social inequality during older middle age itself.

METHODS

Data Source The HRS, a nationally representative, longitudinal sample of U.S. households, uses a multistage area probability design. The study population consists of noninstitutionalized adults living in the contiguous United States, aged 51 to 61 in 1992. We based the data used here on face-to-face interviews conducted with the 9,759 participants in 1992 with biannual telephone follow-up over 5 consecutive waves. Mortality was determined through matching to the National Death Index or from information from household members through 2002. HRS tracking studies have indicated a 98.8% validation of deaths with essentially zero false positives. HRS data are publicly available and can be found at http://hrsonline.isr.umich.edu/. The Northwestern University Institutional Review Board ruled the study exempt.

SES Measures We fixed years of school completed as reported in 1992 and grouped education levels as 0 to 8 years, 9 to 11 years, high school graduate (or General Equivalency Diploma), and more than 12 years. We analyzed time-varying values for household income and household wealth measures using data from each of four (1992–2000) biannual waves of the HRS to predict survival through 2002. We computed an income-to-needs ratio (INR) in 1992 by dividing household income by the 1991 poverty guideline for a given household size; we categorized INR levels as less than or equal to 1.9, 2 to 2.9, 3 to 4.9, and greater than or equal to 5.0 at each wave. We imputed missing data for survivors by forward replacement of INR level. RAND investigators calculated household wealth, which represented the net value of all assets (including primary residence, other real estate, businesses, individual retirement accounts and Keogh accounts, stocks and bonds, pensions, and checking and savings accounts) minus debts (St. Clair et al., 2006). We divided wealth into three categories: less than 20th percentile (from , $15,000 in 1992 to , $23,500 in 2002), 20th to 60th percentile (from $15,000–$126,000 in 1992 to $23,500– $177,000 in 2002), and 60th to 100th percentile (from . $126,000 including the top 20th percentile to . $260,000 in 1992, to . $177,000 including the top 20th percentile at . $399,800 in 2002). Education, income, and wealth were only weakly correlated with one another at 1992 categorical levels; the highest correlation was between levels of household income and household wealth (r ¼ .46). This relative lack of correlation reinforced the importance of analyzing separate SES effects during older middle age (Braveman et al., 2005).

BASELINE HEALTH AND SOCIOECONOMIC STATUS

Health Status Measures All health status measures were time-varying values with forward imputation of missing data for survivors through 2000. Self-reported overall health was measured in five categories: excellent, very good, good, fair, and poor. The number of chronic conditions included self-reported hypertension, diabetes, heart disease, chronic lung disease, cancer, arthritis, stroke, or visual difficulties and ranged from 0 to 8. We empirically merged the small group of respondents reporting four or more chronic conditions into a single highest severity category. Physical impairment was measured by difficulties in climbing stairs and walking various distances; pushing, pulling, and stretching; and performing activities of daily living such as bathing, eating, and dressing, without help. We summed each item rated as a physical impairment or difficulty across nine HRS questions, with a maximum score of 9 difficulties (Feinglass et al., 2005). We empirically grouped the small number of respondents with five or more difficulties into a single high disability category.

Sample Demographic Characteristics We categorized age as 51 to 55, or 56 to 61 in 1992. Marital status categories included married, divorced or separated, widowed, or never married. Race and ethnicity were categorized as White, Black, English-speaking Hispanic, or Spanishspeaking Hispanic, due to well-known differences in health and access to health care between these groups. We dichotomized parents’ mean age at death by whether respondents’ parents’ average age at death was 75 years or older, versus younger than 75 or still alive in 1992 (Hoffman, 2005). Findings from separate models for men and women indicated that the only significant difference was higher male mortality for persons who were separated or divorced, and higher female mortality among those who were widowed at baseline. All data presented here therefore combined male and female respondents while controlling for gender as an indicator variable.

Health Insurance Previous HRS studies have shown that lack of health insurance is associated with forgone preventive services and major declines in health and mortality (Ayanian, Weissman, Schneider, Ginsburg, & Zaslavsky, 2000; Baker, Sudano, Albert, Borawski, & Dor, 2001; Baker et al., 2006; McWilliams, Zaslavsky, Meara, & Ayanian, 2004). However, because insurance status is so confounded by baseline health and SES in this cohort, we did not include it in the mortality models presented below. Large numbers of HRS respondents gained and lost health insurance coverage over the 10-year follow-up period. This volume of insurance status transitions, particularly among the previously uninsured who became disabled and transitioned to public insurance before death, represented a formidable methodological challenge beyond the scope of this investigation.

S211

meters squared, within four categories: underweight (, 18.5 kg/m2), normal weight (18.5–24.9 kg/m2), overweight (25.0– 29.9 kg/m2), and obese (. 30 kg/m2). We calculated physical activity level in 1992 from several questions about weekly or monthly light and vigorous leisure time exercise and heavy housework, including ‘‘walking, dancing, gardening, golfing, or bowling.’’ Vigorous exercise included ‘‘aerobics, running, swimming, or bicycling.’’ We scaled these items using previously published methods (Feinglass et al., 2005) to approximate the Surgeon General report categories of met recommendations ( 30 min/day of moderate activity on  5 days/week or  20 min of vigorous activity on  3 days/week), insufficient (less than recommended but some physical activity), and inactive (engaged in leisure time physical activity fewer than three times per month). We found that measures of alcohol consumption and dependence were not associated with mortality risk in multivariate models and therefore do not include them here.

Statistical Analysis and Discrete Interval Survival Models We analyzed data using the Stata 9.0 (StataCorp, College Station, TX) complex survey design module. We performed bivariate analyses of 10-year survivors versus decedents by risk factor (Table 1) using chi0square tests with person-level analytic weights. Because HRS mortality data are publicly reported only by biannual survey waves, we used survival analysis for discrete data, an analog of the Cox proportional hazard model for continuous data (Prentice & Gloeckler, 1978). These analyses model the discrete hazard rate, which is the probability of dying in each subsequent wave given a person’s risk factor profile at the previous interview. We estimated a discrete hazards model, which accounts for repeated measures on the same individual and does not require a proportional hazard assumption, using SAS Proc GenMod (SAS Institute, Carey, NC) to fit a generalized linear model with a complementary log log link. To account for the complex sampling design, we estimated variance using balanced repeated replication, a form of bootstrapping (Korn, 1999). We report results as hazard ratios with 95% confidence intervals. We present three sequential analyses to estimate the separate and joint associations of baseline health and SES measures with the likelihood of death through 2002. We estimated an initial model to measure the association of SES measures and demographic factors only, beginning with education, and adding income and then wealth (Model 1). Results from this model approximated the many previously published studies of cumulative SES effects on mortality that did not control for baseline health. We then estimated a second model (Model 2) for demographic characteristics, health status, and SES measures combined, testing whether SES factors remained significant after inclusion of time-varying baseline health measures. Finally, a third model (Model 3) added behavioral risk factors to assess any further effects on the significance of SES measures.

Behavioral Risk Factors Behavioral risk factors were both correlated with SES and, in the cases of sedentary activity or being underweight, linked to health status. Smoking status in 1992 was categorized as never smoked, previous smoker, or current smoker. Body mass index was calculated by weight in kilograms divided by height in

RESULTS

Bivariate Associations With 10-Year Mortality Total mortality over 10 years of follow-up for all 1992 HRS participants was 10.9%, reflecting a 13.1% rate for men and an

FEINGLASS ET AL.

S212

Table 1. Mortality Rates, Hazard Ratios, and 95% Confidence Intervals for All-Cause Mortality, 1992–2002 Health and Retirement Study Variable

Estimated Percent of U.S. Population Aged 51–61

Percent Died, 1992–2002

Model 1a

Model 2b

Model 3c

46.7 53.3

8.9 13.4

1.58 (1.31–1.79)

1.47 (1.26–1.69)

1.50 (1.28–1.71)

48.1 51.9

13.1 8.9

0.53 (0.46–0.60)

0.54 (0.47–0.59)

0.58 (0.50–0.66)

74.1 3.9 15.4 6.4

9.5 15.4 12.3 16.6

0.86 (0.56–1.16) 1.28 (1.05–1.46) 1.37 (1.06–1.67)

0.97 (0.61–1.32) 1.16 (0.96–1.33) 1.34 (1.01–1.63)

1.01 (0.67–1.34) 1.06 (0.88–1.25) 1.22 (0.89–1.55)

83.6 2.6 3.5 10.0

10.1 8.2 9.3 17.6

0.35 (0.24–0.46) 0.64 (0.46–0.82) 1.09 (0.92–1.25)

0.44 (0.28–0.59) 0.69 (0.49–0.89) 1.06 (0.89–1.24)

0.57 (0.34–0.78) 0.78 (0.55–1.00) 1.17 (0.96–1.37)

33.1 66.9

14.3 9.2

1.36 (1.16–1.54)

1.20 (1.02–1.37)

1.19 (1.01–1.37)

10.0 44.7

18.2 14.7

1.24 (0.97–1.49) 1.12 (0.93–1.30)

0.79 (0.58–0.95) 0.86 (0.69–1.02)

0.76 (0.55–0.97) 0.84 (0.70–1.04)

36.8 38.5

10.1 8.3

0.99 (0.84–1.14)

0.89 (0.76–1.01)

0.90 (0.76–1.03)

18.2 13.4 27.8 42.6

19.5 13.7 10.1 6.9

2.34 (1.88–2.81) 1.58 (1.27–1.90) 1.27 (1.02–1.52)

1.40 (1.40–1.72) 1.30 (1.04–1.57) 1.13 (0.89–1.38)

1.34 (1.04–1.65) 1.26 (1.01–1.51) 1.12 (0.87–1.36)

17.4 37.8 44.8

19.4 11.3 7.3

2.02 (1.65–2.38) 1.35 (1.14–1.56)

1.16 (0.93–1.38) 1.05 (0.87–1.24)

1.02 0(0.82–1.22 1.01 (0.83–1.18)

23.7 29.4 26.6 12.8 7.3

4.7 6.2 10.2 20.4 35.7

1.03 1.58 3.08 5.89

(0.73–1.31) (1.09–2.08) (1.97–4.18) (2.30–8.16)

1.01 1.51 2.81 5.11

(0.73–1.28) (1.05–1.98) (1.84–3.80) (3.16–7.11)

40.7 34.2 19.3 7.6 3.8

5.4 9.5 21.6 13.4 38.2

1.35 1.56 1.74 2.19

(0.96–1.75) (1.02–2.09) (1.08–2.41) (1.40–2.97)

1.42 1.70 1.94 2.55

(1.01–1.83) (1.12–2.27) (1.20–2.67) (1.60–3.47)

48.5 25.2 10.6 5.5 4.0 6.2

6.1 9.8 14.3 17.9 21.2 33.5

1.21 1.35 1.41 1.61 2.18

(0.99–1.44) (1.01–1.71) (1.04–1.78) (1.09–2.12) (1.59–2.76)

1.18 1.28 1.34 1.47 2.00

(0.94–1.41) (0.93–1.63) (0.98–1.69) (0.93–1.99) (1.38–2.59)

Demographics Age, 1992* 51–55 (reference) 56–61 Gender* Male (reference) Female Marital status, 1992* Married (reference) Never married Separated/divorced Widowed Race/ethnicity* White/other (reference) Spanish-speaking Hispanic English-speaking Hispanic Black Parents’ age at death* , 75 years old  75 years old (reference) Socioeconomic status Education* 0–8 years 9–11 years High school graduate or General Equivalency Diploma . 12 years (reference) Household income-to-needs ratio, 1992*  1.9 2.0–2.9 3.0–4.9  5.0 (reference) Household wealth, 1992* , 20th percentile 20th–60th percentile 60th–100th percentile (reference) Baseline health Self-reported overall health, 1992* Excellent (reference) Very Good Good Fair Poor Number of chronic conditions, 1992* 0 (reference) 1 2 3 4 Number of physical difficulties, 1992* 0 (reference) 1 2 3 4 5

(Table 1 continues)

BASELINE HEALTH AND SOCIOECONOMIC STATUS

S213

Table 1. Mortality Rates, Hazard Ratios, and 95% Confidence Intervals for All-Cause Mortality, 1992–2002 Health and Retirement Study (Continued) Variable

Estimated Percent of U.S. Population Aged 51–61

Percent Died, 1992–2002

35.9 37.0 27.1

5.4 10.3 18.7

1.3 36.1 40.7 21.9

40.7 10.7 6.1 9.9

42.5 40.2 17.3

7.6 10.5 19.7

Model 1a

Model 2b

Model 3c

Behavioral risk factors Smoking status, 1992* Never smoked (reference) Past smoker Current smoker

1.55 (1.26–1.84) 2.08 (1.64–2.15)

Body mass index, 1992* Underweight Normal (reference) Overweight Obese

2.20 (1.23–3.17) 0.67 (0.60–0.84) 0.73 (0.55–0.78)

Physical activity, 1992* Meets recommendations (reference) Insufficient Inactive

1.12 (0.92–1.31) 1.29 (1.04–1.52)

Notes: N ¼ 9,759 survey-weighted respondents from the Health and Retirement Study. Demographic characteristics, education, income, and wealth. b Demographic characteristics, baseline health, socioeconomic status measures. c Demographic characteristics, baseline health, socioeconomic status measures, and behavioral risk factors. *p , .0001, mortality rate risk factor categorical comparison to reference group. a

8.9% rate for women. Biannual death rates grew from approximately 1.5% in 1992/1994 to 2.8% by 2000/2002. Table 1 provides bivariate associations of 1992 demographic, SES, baseline health, and behavioral measures with cumulative 10-year mortality. There were statistically significant differences in 10-year death rates for all measures. Higher mortality was associated with older age, male gender, not being currently married at baseline, and having parents whose average age at death was less than 75 years old. Although Blacks had higher 10-year mortality (17.6%) than Whites, Spanish-speaking Hispanics had the lowest mortality, followed by English-speaking Hispanics, another demonstration of the Hispanic paradox (Palloni & Arias, 2004). Bivariate differences in mortality are graded across all SES and baseline health comparisons in Table 1 with differences significant at the p , .0001 level. Higher mortality was also associated with being a current smoker, being underweight (often related to illness), and being physically inactive (which also may have been a proxy for baseline health). Respondents who reported being overweight or obese had better 10-year mortality than normal-weight respondents.

Model 1 Results for SES Measures The final three columns of Table 1 present discrete time survival model results, beginning with demographic and SES measures in Model 1. Older age; male gender; being separated, divorced, or widowed; and having parents who died at less than 75 years old were all significantly associated with mortality. Hispanics had significantly lower mortality than Whites, and Black respondents no longer had significantly higher mortality than Whites after adjustment for SES. Lower levels of educational attainment, although initially significantly associated with higher mortality risk, became nonsignificant when first income and then wealth were entered into sequential models (data not shown). Individuals of lower income levels all had significantly higher mortality with hazard ratios only

modestly attenuated when wealth was included, and respondents from lower wealth households also had a higher mortality risk net of income.

Model 2 and Model 3 Results for SES, Health Status, and Behavioral Risk Factors Model 2 in Table 1 presents results after including baseline health with SES measures in the same model. Education and wealth became nonsignificant, and only the lowest two income categories remained significantly different after inclusion of health status measures. As expected, time-varying self-reported overall health proved to be a dynamic evaluation, with respondents reporting poor health almost 6 times as likely to subsequently die as those reporting excellent health. As compared to those with zero chronic conditions or functional difficulties, having two or more conditions or impairments was associated with progressively greater mortality risk. These results suggest the extent to which baseline health and SES were already confounded at cohort inception in 1992. More than 40% of respondents in the lowest INR and education categories reported fair or poor health, more than double the overall rate. Not only did individuals reporting excellent health at baseline have the lowest 10-year mortality (4.7%), but the vast majority (58.2%) of those in excellent health also lived in the highest INR households. Respondents in poor health in 1992, who had the worst 10-year mortality rate (35.8%), provided a stark contrast. More than half of this high-risk group were from the lowest INR households. The primary influence of baseline health was illustrated by the fact that respondents in fair to poor health in 1992 were at least twice as likely to die as respondents in good, very good, or excellent health across all SES strata. Model 3 added 1992 smoking, body mass index, and physical activity measures. Respondents who were inactive, smoked, or were underweight were at significantly higher risk. Although there was only very modest change in the effects of baseline

S214

FEINGLASS ET AL.

health variables, addition of behavioral risk factors further attenuated the effects of being separated or divorced, widowed, and English-speaking Hispanic to nonsignificance. Hazard ratios for income levels changed little between Models 2 and 3.

DISCUSSION Our primary question was whether SES measures continued to be associated with mortality after data were controlled for baseline health status, given that middle-age health is such a powerful predictor of life expectancy. As confirmed by the results presented here, baseline health in older middle age mediates much of the relationship between SES and mortality to about age 70. After the inclusion of both baseline health and behavioral risk factors, SES measures like education and wealth were no longer significantly associated with mortality risk. However, household income, although attenuated, remained significant.

Decreasing Health Inequalities in Older Age? These findings are relevant to the debate about the apparent reduction of SES differences in health at older ages. There has been relatively little empirical support for the selective mortality hypothesis, which asserts that lower SES individuals die earlier, thus survivors from lower SES backgrounds are comparatively healthier in older age (Beckett, 2000). However, researchers have suggested that more equal access to health care, given near-universal Medicare coverage, and the importance of Social Security income in eliminating poverty among the elderly population, may have reduced health disparities among more recent cohorts of older Americans (House et al., 2005). In one of the only attempts to directly address the confounding of SES measures with baseline health in the older American population, Hoffmann (2005) analyzed 8-year allcause mortality for White adults aged 59 to 107 belonging to a merged cohort from the HRS and the Study of Asset and Health Dynamics. Interaction effects in this study indicated that once individuals were in poor health, there were few subsequent SES differences in older age mortality, and that SES factors primarily influenced mortality rates for older adults in excellent to good health at baseline. Post hoc analyses of our data (not shown) confirmed Hoffman’s analysis: The lowest levels of household income were only significantly associated with mortality for respondents reporting excellent to good health at baseline. As health status generally deteriorates with age, social disparities in mortality appear to converge. Thus, the weakened association between some SES measures (particularly education) and mortality after age 65 may be related to the overwhelming influence of the increasing proportion of all individuals in poor baseline health, with transitions from serious chronic illness to death being more biologically than socially driven at older ages.

Association of Education, Household Income, and Wealth With Mortality Risk Education, or years of school completed, is probably the most frequently used SES measure in analyses of social determinants of health in the United States. Schooling can translate into quality of job opportunities and income, fewer risky behaviors, and availability of social support and high-

quality medical care. This is what Ross and Wu (1996) termed education’s cumulative advantage, which they found tends to increase health disparities with age. Cognitive ability, or intelligence test score measures, are also closely bound up with childhood educational experience. Childhood learning disabilities (Noble & McCandliss, 2005) and low health literacy (Baker et al., 2002) appear to have substantial independent effects on older age health and mortality (Batty & Deary, 2004). This has led some to call for using education as a cardiovascular risk factor equal to the widely utilized traditional clinical measures (Fiscella & Franks, 2004). In this study, years of school completed had a clear monotonic relationship with mortality risk, even after the inclusion of income and wealth (Model 4). However, education failed to predict mortality at all after the data were controlled for baseline health. This is not unexpected (Davey Smith, Hart, et al., 1998). With few exceptions (Winkleby, Jatulis, Frank, & Fortmann, 1992), studies that have reported significant mortality differences by educational level failed to include other SES covariates (Elo & Preston, 1996; Liao et al., 1999; Pappas et al., 1993; Sorlie et al., 1995; Steenland, Henley, & Thun, 2002; Wong et al., 2002). Among those studies that did use multivariable economic and social measures, almost all found that after adjustment, education is only weakly or nonsignificantly associated with later life mortality (Daly et al., 2002; Davey Smith, Hart, et al., 1998; Lantz et al., 1998). The findings presented here confirm those from the Americans’ Changing Lives Study and other studies that education influences the onset, but not the progression, of chronic conditions (House et al., 2005). These findings run counter to widely held beliefs about the superior importance of education over race and income in predicting adult health and life expectancy (Kolata, 2007). Finally, an important additional problem for evaluating educational attainment is that schooling quality varies so greatly across the United States and across generations that it may be a poor proxy for directly measured adult health literacy (Baker et al., 2002). Household income at the individual or community level also has a well-known association with Americans’ health and life expectancy (Davey Smith, Neaton, Wentworth, Stamler, & Stamler, 1998; Krieger, Chen, Waterman, Rehkopf, & Subramanian, 2005; Shishehbor, Litaker, Pothier, & Lauer, 2006). However, several important caveats remain when considering how income may directly affect health and mortality. Among the older middle-aged adults in the HRS, employment status is closely tied to household income. Of the 68.4% of respondents who reported being employed in 1992, half were in the highest INR category and only 10% in the lowest. By contrast, more than one third of those respondents who reported being unemployed due to disability or just being unemployed were in the lowest INR category. Employment or retirement status is obviously related to chronic health problems for many. Thus, some of the continuing effect of household income in our models undoubtedly represents reverse confounding with conditions acquired earlier in life that caused premature disability, unemployment, or retirement. As indicated by the changing associations of car ownership and employment status with postretirement mortality in the Whitehall Study (Marmot & Shipley, 1996), it is likely that higher household wealth becomes increasingly protective

BASELINE HEALTH AND SOCIOECONOMIC STATUS

relative to employment income and occupational status as retirement (including early retirement) becomes common for this age group. The association of household wealth with 10-year mortality in the HRS was initially very significant, reflecting a crude death rate of 19.4% for those with less than the 20th percentile of net assets as compared to 7.3% for those above the 60% percentile of wealth. This association was greatly reduced by the inclusion of baseline health in Model 2 and, unlike income, much further reduced with the inclusion of behavioral risk factors in Model 3. This latter finding may be related to the large baseline differences between lowest and highest wealth category respondents in current smoking (42% vs 19%, respectively) and physical inactivity (33% vs 22%, respectively).

Behavioral Risk Factors and the Association of SES With Mortality Our findings replicate U.S. (Lantz et al., 1998) and British studies (Marmot, 2006) in demonstrating that SES differences in mortality remain independent of the distribution of behavioral risks; inclusion of these factors had little effect on SES mortality risk ratios. However, it is important to note that smoking and physical activity are strongly patterned across the SES gradient and are thus really an integral part of SES disparities in health (Barbeau, Krieger, & Soobader, 2004; Richardson, Kriska, Lantz, & Hayward, 2004). Nevertheless, it is striking that smoking and leisure time physical activity risks had little effect on the association of household income and mortality. An anomalous, but not unprecedented (Lantz et al., 1998), finding was that respondents who were overweight and obese at baseline did not have an elevated relative risk of mortality. Subanalysis (results not shown) of the obese subgroup revealed that these individuals were significantly more likely to be younger (less than 56 years old) and less likely to be current smokers, both of which confer a mortality advantage. The results of a recent National Health Interview Survey follow-up study of individuals aged 18 to 64 years old showed that the body mass index associated with the lowest mortality is actually around 26 or 27 (which would have traditionally been considered overweight) for both sexes combined, and that only beyond a body mass index of 27 does mortality start to increase appreciably (Gronniger, 2006). It would be surprising, however, if the effects of obesity did not have an independent effect on mortality rates above and beyond prevalence of specific chronic conditions over a longer follow-up period (Yan, Daviglus, et al., 2006).

Limitations Perhaps the most important limitation to these findings is that the HRS provides no estimates for traditional clinical mortality risk factors such as blood pressure and cholesterol. This may have led to underestimation of the true effects of unreported conditions. Other research has indicated that traditional clinical risk factors add little prognostic power once populations are stratified by SES characteristics, but this remains to be studied (Brindle et al., 2005; Fiscella & Franks, 2004; G. D. Smith, Shipley, & Rose, 1990). Our models also could not separately analyze the effects of other potentially significant measures of SES, such as childhood SES, occupational job strain, or

S215

environmental health exposures. We did not directly examine the potentially large number of interactions between the SES and baseline health variables. Although important to flesh out and clarify the data presented here, this investigation was far too great in scope to present here. Finally, the HRS does not include institutionalized persons, which may have slightly biased the results.

Future Epidemiologic Surveillance of SES Disparities Uncertainty about concurrent versus historical effects of economic and social inequality is inevitable given major demographic shifts as a new generation of Americans ages. Such changes include the huge historical increase in college education of both men and women entering older middle age, large shifts in the American workforce, changes in diet and exercise patterns, and the notable increases in life expectancy for the oldest old Americans. These changes alone make it difficult to extrapolate to contemporary American society findings from older cohort studies of social determinants of health. Although expensive to field, additional populationbased, prospective longitudinal studies with more sophisticated SES indicators are essential to public health surveillance in an era of increasing social and economic inequality. ACKNOWLEDGMENTS This work was supported in part by Grant RO1 HS10283 to the Northwestern University Feinberg School of Medicine from the U.S. Agency for Healthcare Research and Quality. We are grateful to three anonymous reviewers for their remarkably helpful suggestions and comments. CORRESPONDENCE Address correspondence to Joe Feinglass, PhD, Division of General Internal Medicine, Northwestern Feinberg School of Medicine, 676 St. Clair No. 200, Chicago, IL 60611. E-mail: [email protected] REFERENCES Alwin, D. F., & Wray, L. A. (2005). A life-span developmental perspective on social status and health. Journals of Gerontology: Psychological Sciences and Social Sciences, 60B(Special Issue II), 7–14. Ayanian, J. Z., Weissman, J. S., Schneider, E. C., Ginsburg, J. A., & Zaslavsky, A. M. (2000). Unmet health needs of uninsured adults in the United States. Journal of the American Medical Association, 284, 2061–2069. Backlund, E., Sorlie, P. D., & Johnson, N. J. (1999). A comparison of the relationships of education and income with mortality: The National Longitudinal Mortality Study. Social Science and Medicine, 49, 1373–1384. Baker, D. W., Gazmararian, J. A., Williams, M. V., Scott, T., Parker, R. M., Green, D., et al. (2002). Functional health literacy and the risk of hospital admission among Medicare managed care enrollees. American Journal of Public Health, 92, 1278–1283. Baker, D. W., Sudano, J. J., Albert, J. M., Borawski, E. A., & Dor, A. (2001). Lack of health insurance and decline in overall health in late middle age. New England Journal of Medicine, 345, 1106–1112. Baker, D. W., Sudano, J. J., Durazo-Arvizu, R., Feinglass, J., Witt, W. P., & Thompson, J. (2006). Health insurance coverage and the risk of decline in overall health and death among the near elderly, 1992–2002. Medical Care, 44(3), 277–282. Banks, J., Marmot, M., Oldfield, Z., & Smith, J. P. (2006). Disease and disadvantage in the United States and in England. Journal of the American Medical Association, 295, 2037–2045. Barbeau, E. M., Krieger, N., & Soobader, M. J. (2004). Working class matters: Socioeconomic disadvantage, race/ethnicity, gender, and

S216

FEINGLASS ET AL.

smoking in NHIS 2000. American Journal of Public Health, 94(2), 269–278. Barker, D. J., Osmond, C., Forsen, T. J., Kajantie, E., & Eriksson, J. G. (2005). Trajectories of growth among children who have coronary events as adults. New England Journal of Medicine, 353, 1802–1809. Batty, G. D., & Deary, I. J. (2004). Early life intelligence and adult health. British Medical Journal, 329, 585–586. Batty, G. D., Der, G., Macintyre, S., & Deary, I. J. (2006). Does IQ explain socioeconomic inequalities in health? Evidence from a population based cohort study in the west of Scotland. British Medical Journal, 332, 580–584. Beckett, M. (2000). Converging health inequalities in later life—an artifact of mortality selection. Journal of Health and Social Behavior, 41(1), 106–119. Bond Huie, S. A., Krueger, P. M., Rogers, R. G., & Hummer, R. A. (2003). Wealth, race, and mortality. Social Science Quarterly, 84, 667–684. Braveman, P. A., Cubbin, C., Egerter, S., Chideya, S., Marchi, K. S., Metzler, M., et al. (2005). Socioeconomic status in health research: One size does not fit all. Journal of the American Medical Association, 294, 2879–2888. Brindle, P. M., McConnachie, A., Upton, M. N., Hart, C. L., Davey Smith, G., & Watt, G. C. (2005). The accuracy of the Framingham risk-score in different socioeconomic groups: A prospective study. British Journal of General Practice, 55, 838–845. Cutler, D. M., & Meara, E. (2001). Changes in the age distribution of mortality over the 20th century (National Bureau of Economic Research Working Paper No. 8556). Retrieved June 6, 2004 from http:// www.nber.org/papers/w8556. Daly, M. C., Duncan, G. J., McDonough, P., & Williams, D. R. (2002). Optimal indicators of socioeconomic status for health research. American Journal of Public Health, 92, 1151–1157. Davey Smith, G., Hart, C., Hole, D., MacKinnon, P., Gillis, C., Watt, G., et al. (1998). Education and occupational social class: Which is the more important indicator of mortality risk? Journal of Epidemiology and Community Health, 52(3), 153–160. Davey Smith, G., Neaton, J. D., Wentworth, D., Stamler, R., & Stamler, J. (1998). Mortality differences between black and white men in the USA: Contribution of income and other risk factors among men screened for the MRFIT. Lancet, 351, 934–939. Elo, I. T., & Preston, S. H. (1996). Educational differentials in mortality: United States, 1979–85. Social Science and Medicine, 42, 47–57. Feinglass, J., Thompson, J. A., He, X. Z., Witt, W., Chang, R. W., & Baker, D. W. (2005). Effect of physical activity on functional status among older middle-age adults with arthritis. Arthritis and Rheumatism, 53, 879–885. Fiscella, K., & Franks, P. (2004). Should years of schooling be used to guide treatment of coronary risk factors? Annals of Family Medicine, 2(5), 469–473. Galobardes, B., Lynch, J. W., & Davey Smith, G. (2004). Childhood socioeconomic circumstances and cause-specific mortality in adulthood: Systematic review and interpretation. Epidemiologic Reviews, 26, 7–21. Goldman, D. P., & Smith, J. P. (2002). Can patient self-management help explain the SES health gradient? Proceedings of the National Academy of Sciences, USA, 99, 10929–10934. Gronniger, J. T. (2006). A semiparametric analysis of the relationship of body mass index to mortality. American Journal of Public Health, 96, 173–178. Hoffman, R. (2005). Does the impact of socioeconomic status on mortality decrease with increasing age? Demographic Research, 13, 35–62. House, J. S., Lantz, P. M., & Herd, P. (2005). Continuity and change in the social stratification of aging and health over the life course: Evidence from a nationally representative longitudinal study from 1986 to 2001/ 2002 (Americans’ Changing Lives Study). Journals of Gerontology: Psychological Sciences and Social Sciences, 60B(Special Issue), 15–26. Kolata, G. (2007, January 3). A surprising secret to long life: Stay in school. The New York Times, pp. A1, A16. Korn, E. (1999). Analysis of health surveys. New York: Wiley. Koster, A., Bosma, H., Penninx, B. W., Newman, A. B., Harris, T. B., van Eijk, J. T., et al. (2006). Association of inflammatory markers with socioeconomic status. Journal of Gerontology: Medical Sciences, 61A, 284–290. Krieger, N. (2005). Embodiment: A conceptual glossary for epidemiology. Journal of Epidemiology and Community Health, 59, 350–355.

Krieger, N., Chen, J. T., Coull, B. A., & Selby, J. V. (2005). Lifetime socioeconomic position and twins’ health: An analysis of 308 pairs of United States women twins. PLoS Medicine, 2(7), 645–653. Krieger, N., Chen, J. T., Waterman, P. D., Rehkopf, D. H., & Subramanian, S. V. (2005). Painting a truer picture of US socioeconomic and racial/ ethnic health inequalities: The Public Health Disparities Geocoding Project. American Journal of Public Health, 95, 312–323. Lantz, P. M., House, J. S., Lepkowski, J. M., Williams, D. R., Mero, R. P., & Chen, J. (1998). Socioeconomic factors, health behaviors, and mortality: Results from a nationally representative prospective study of US adults. Journal of the American Medical Association, 279, 1703–1708. Liao, Y., McGee, D. L., Kaufman, J. S., Cao, G., & Cooper, R. S. (1999). Socioeconomic status and morbidity in the last years of life. American Journal of Public Health, 89, 569–572. Luo, Y., & Waite, L. J. (2005). The impact of childhood and adult SES on physical, mental, and cognitive well-being in later life. Journal of Gerontology: Social Sciences, 60B, S93–S101. Marmot, M. G. (2006). Status syndrome: A challenge to medicine. Journal of the American Medical Association, 295, 1304–1307. Marmot, M. G., & Shipley, M. J. (1996). Do socioeconomic differences in mortality persist after retirement? 25 year follow up of civil servants from the first Whitehall study. British Medical Journal, 313, 1177–1180. McWilliams, J. M., Zaslavsky, A. M., Meara, E., & Ayanian, J. Z. (2004). Health insurance coverage and mortality among the near-elderly. Health Affairs, 23(4), 223–233. Noble, K. G., & McCandliss, B. D. (2005). Reading development and impairment: Behavioral, social, and neurobiological factors. Journal of Developmental and Behavioral Pediatrics, 26(5), 370–378. O’Rand, A. M., & Hamil-Luker, J. (2005). Processes of cumulative adversity: Childhood disadvantage and increased risk of heart attack across the life course. Journals of Gerontology: Psychological Sciences and Social Sciences, 60B(Special Issue II), 117–124. Palloni, A., & Arias, E. (2004). Paradox lost: Explaining the Hispanic adult mortality advantage. Demography, 41(3), 385–415. Pappas, G., Queen, S., Hadden, W., & Fisher, G. (1993). The increasing disparity in mortality between socioeconomic groups in the United States, 1960 and 1986. New England Journal of Medicine, 329, 103–109. Pincus, T., Esther, R., DeWalt, D. A., & Callahan, L. F. (1998). Social conditions and self-management are more powerful determinants of health than access to care. Annals of Internal Medicine, 129(5), 406–411. Pollitt, R. A., Rose, K. M., & Kaufman, J. S. (2005). Evaluating the evidence for models of life course socioeconomic factors and cardiovascular outcomes: A systematic review. BMC Public Health, 5(1), 1–13. Prentice, R. L., & Gloeckler, L. A. (1978). Regression analysis of grouped survival data with application to breast cancer data. Biometrics, 34(1), 57–67. Richardson, C. R., Kriska, A. M., Lantz, P. M., & Hayward, R. A. (2004). Physical activity and mortality across cardiovascular disease risk groups. Medicine and Science in Sports and Exercise, 36, 1923– 1929. Ross, C. E., & Wu, C. L. (1996). Education, age, and the cumulative advantage in health. Journal of Health and Social Behavior, 37(1), 104–120. Schoeni, R. F., Martin, L. G., Andreski, P. M., & Freedman, V. A. (2005). Persistent and growing socioeconomic disparities in disability among the elderly: 1982–2002. American Journal of Public Health, 95, 2065–2070. Shishehbor, M. H., Litaker, D., Pothier, C. E., & Lauer, M. S. (2006). Association of socioeconomic status with functional capacity, heart rate recovery, and all-cause mortality. Journal of the American Medical Association, 295, 784–792. Singh, G. K., & Siahpush, M. (2006). Widening socioeconomic inequalities in US life expectancy, 1980–2000. International Journal of Epidemiology, 35, 969–979. Smith, G. D., Shipley, M. J., & Rose, G. (1990). Magnitude and causes of socioeconomic differentials in mortality: Further evidence from the Whitehall Study. Journal of Epidemiology and Community Health, 44, 265–270.

BASELINE HEALTH AND SOCIOECONOMIC STATUS

Smith, J. P. (1998). Socioeconomic status and health. American Economic Review, 88(2), 192–196. Sorlie, P. D., Backlund, E., & Keller, J. B. (1995). US mortality by economic, demographic, and social characteristics: The National Longitudinal Mortality Study. American Journal of Public Health, 85, 949–956. St. Clair, P., Blake, D., Bugliari, D., Chien, S., Hayden, O., Hurd, M., et al. (2006). Rand HRS data documentation, Version F. Labor and Population Program, Rand Center for the Study of Aging. Retrieved December 5, 2006 from http://www.rand.org/labor/aging. Steenland, K., Henley, J., & Thun, M. (2002). All-cause and cause-specific death rates by educational status for two million people in two American Cancer Society cohorts, 1959–1996. American Journal of Epidemiology, 156(1), 11–21. Winkleby, M. A., Jatulis, D. E., Frank, E., & Fortmann, S. P. (1992). Socioeconomic status and health: How education, income, and occupation contribute to risk factors for cardiovascular disease. American Journal of Public Health, 82, 816–820. Wong, M. D., Shapiro, M. F., Boscardin, W. J., & Ettner, S. L. (2002). Contribution of major diseases to disparities in mortality. New England Journal of Medicine, 347, 1585–1592. Wray, L. A., Alwin, D. F., & McCammon, R. J. (2005). Social status and

S217

risky health behaviors: Results from the Health and Retirement Study. Journals of Gerontology: Psychological Sciences and Social Sciences, 60B(Special Issue II), 85–92. Yan, L. L., Daviglus, M. L., Liu, K., Stamler, J., Wang, R., Pirzada, A., et al. (2006). Midlife body mass index and hospitalization and mortality in older age. Journal of the American Medical Association, 295, 190–198. Yan, L. L., Liu, K., Daviglus, M. L., Colangelo, L. A., Kiefe, C. I., Sidney, S., et al. (2006). Education, 15-year risk factor progression, and coronary artery calcium in young adulthood and early middle age: The Coronary Artery Risk Development in Young Adults study. Journal of the American Medical Association, 295, 1793–1800. Zimmer, Z., & House, J. S. (2003). Education, income, and functional limitation transitions among American adults: Contrasting onset and progression. International Journal of Epidemiology, 32, 1089–1097.

Received August 31, 2006 Accepted March 2, 2007 Decision Editor: Kenneth F. Ferraro, PhD