Socioeconomic disparities in health change in a ... - Semantic Scholar

Social Science & Medicine 53 (2001) 29–40

Socioeconomic disparities in health change in a longitudinal study of US adults: the role of health-risk behaviors Paula M. Lantza,b,*, John W. Lyncha,b, James S. Housea,b,c, James M. Lepkowskia,b, Richard P. Meroa, Marc A. Musickd, David R. Williamsa,c a

b

Survey Research Center, Institute for Social Research, University of Michigan, Ann Arbor, MI, USA School of Public Health, University of Michigan, Room M3 116, 109 observatory Ann Arbor, MI 48109-2029, USA c Department of Sociology, University of Michigan, Ann Arbor, MI, USA d Department of Sociology, University of Texas, Austin, USA

Abstract This study investigated the hypothesis that socioeconomic differences in health status change can largely be explained by the higher prevalence of individual health-risk behaviors among those of lower socioeconomic position. Data were from the Americans’ Changing Lives study, a longitudinal survey of 3617 adults representative of the US noninstitutionalized population in 1986. The authors examined associations between income and education in 1986, and physical functioning and self-rated health in 1994, adjusted for baseline health status, using a multinomial logistic regression framework that considered mortality and survey nonresponse as competing risks. Covariates included age, sex, race, cigarette smoking, alcohol consumption, physical activity, and Body Mass Index. Both income and education were strong predictors of poor health outcomes. The four health-risk behaviors under study statistically explained only a modest portion of the socioeconomic differences in health at follow-up. For example, after adjustment for baseline health status, those in the lowest income group at baseline had odds of moderate/severe functional impairment in 1994 of 2.11 (95% C.I.: 1.40, 3.20) in an unadjusted model and 1.89 (95% C.I.: 1.23, 2.89) in a model adjusted for health-risk behaviors. The results suggest that the higher prevalence of major health-risk behaviors among those in lower socioeconomic strata is not the dominant mediating mechanism that can explain socioeconomic disparities in health status among US adults. # 2001 Elsevier Science Ltd. All rights reserved. Keywords: Socioeconomic disparities; Poverty; Health status change; Health-risk behaviors; Longitudinal studies; USA

Introduction Morbidity and mortality rates vary greatly by markers of socioeconomic position, including education, income, and occupation (Davey Smith, Shipley & Rose, 1990; Sorlie, Backlund & Keller, 1995; Pappas, Queen, Hadden & Fisher, 1993). The strong negative association between socioeconomic factors and health status has been observed across a wide variety of historical *Correspondence address. University of Michigan School of Public Health, 109 Observatory, Room M3116, Ann Arbor, MI, 48109-2029, USA. Tel.: 313-763-9903; fax: 313-764-4338. E-mail address: [email protected] (P.M. Lantz).

contexts, geographic locations, and populations (Blaxter 1987; Haan, Kaplan & Syme, 1989; Link & Phelan, 1995; Lynch & Kaplan, in press). In addition, results from a number of population-based studies in more developed countries suggest that socioeconomic differences in health status occur over the life course, yet are the greatest in the adult years (House et al., 1994; Mustard, Derksen, Berthelot, Wolfson & Roos, 1997). Empirical evidence from a number of longitudinal and cross-sectional studies demonstrates that health-risk behaviors such as tobacco use, alcohol consumption, having a sedentary lifestyle, and obesity are associated with a variety of health-risks and mortality (Healthy People 2000, 1990; McGinnis & Foege 1993;

0277-9536/01/$ - see front matter # 2001 Elsevier Science Ltd. All rights reserved. PII: S 0 2 7 7 - 9 5 3 6 ( 0 0 ) 0 0 3 1 9 - 1

30

P.M. Lantz et al. / Social Science & Medicine 53 (2001) 29–40

Fraser et al., 1997). There is also ample evidence that the prevalence of most health-risk behaviors is higher among those with lower levels of education and income (Liu et al., 1982; Winkleby, Fortman & Barrett, 1990; Lynch, Kaplan & Salonen, 1997; National Center for Health Statistics, 1998). Thus, a common yet also debated perspective is that the increased prevalence of major health-risk behaviors among people of lower socioeconomic position accounts for much of their increased risk for negative health outcomes. (Williams 1990; Krieger, Rowley, Herman, Avery, Phillips, 1993; McGinnis & Foege 1993; Blaxter 1997) In other words, a prominent hypothesis is that differences in personal health-risk behaviors (or ‘‘lifestyle choices’’) across socioeconomic strata can largely explain observed socioeconomic disparities in health (Macintyre, 1997). Previous longitudinal research has shown that selected health-risk behaviors account for only a modest proportion of socioeconomic differences in mortality (Davey Smith et al., 1990; Hirdes & Forbes, 1992; Lynch, Kaplan, Cohen, Tuomilehto & Salonen, 1996; Marmot et al., 1997; Lantz, House, Lepkowski, Williams, Maro, Chen, 1998). The role of major health-risk behaviors in mediating socioeconomic differences in health status, however, is less studied. Lynch, Kaplan and Shema (1997), using data from the Alameda County Study, found that those with incomes less than 200% of the poverty level in 1965, 1974 and 1983 had a significantly higher rate of problems with physical, and psychological functioning in 1994. After adjusting for baseline levels and changes in smoking, Body Mass Index, alcohol consumption, and physical activity level, the association between economic hardship and functioning attenuated somewhat, but was still strong and significant. Similarly, in a study of Swedish men, Mansson, Rastam, Eriksson and Israelsson (1998) found that socioeconomic status was strongly associated with being granted a disability pension in the future, adjusting for working conditions and for body weight and other disease-risk factors. Vita, Terky, Hubert & Fries (1998) found that smoking, obesity and exercise patterns in both mid-life and later years were associated with subsequent disability, and that the absence of healthrisk behaviors postponed and thus compressed disability into fewer years at the end of life. However, these researchers } while addressing the issue of health behaviors and subsequent health status } did not address socioeconomic differentials in disability. Overall, there does not appear to be much information on the contribution of major health-risk behaviors in explaining socioeconomic differentials in health status from longitudinal studies with nationally representative samples. It is possible that, compared to mortality, individual health-risk behaviors play a stronger role in explaining

socioeconomic disparities in individual health status. All-cause mortality is a more general health outcome, including many causes not known to be related to individual health-risk behaviors. Thus, certain measures of health status may be more proximately related to major behavioral risk factors than all-cause mortality and as such may account for more of the socioeconomic gradient in health status. In addition, the impact of some health-risk behaviors on health status is much stronger and clearer than their impact on mortality. For example, the impact of being overweight on mortality has not been demonstrated consistently across population-based studies, yet obesity has a strong and significant association with many chronic diseases and debilitating conditions, some of which are not necessarily life threatening (Anonymous, 1998; Bender, Trautner, Spraul & Berger, 1998). The role of individual healthrisk behaviors in explaining socioeconomic disparities may vary across health outcomes. The degree to which the higher prevalence of risky health practices among people of lower income and education levels explains or contributes to socioeconomic disparities in health status is an important issue for public health policy. A greater understanding of the potential for interventions focused on the health behaviors of individuals to reduce social inequalities in health is needed. Thus, the purpose of this study was to examine socioeconomic differences in health status among adults in the United States over a 7.5-year period, using two different measures of general health status (functional impairment regarding physical health and self-rated health), focusing on the degree to which major health-risk behaviors explain the observed association between socioeconomic characteristics and changes in health status over time. This research contributes to existing literature by employing a longitudinal study design and a nationally representative sample (including both males and females) to address issues that are of great importance both to epidemiological research and to public health policy.

Methods Study population The Americans changing lives (ACL) survey is a stratified, multistage area probability sample of noninstitutionalized civilian adults age 25 years and older living in the coterminous United States, with an oversampling of blacks and persons aged 60 years and older. The ACL wave 1 survey, conducted in 1986, interviewed 3617 persons face-to-face (representing 70% of sampled household and 68% of sampled individuals). Wave 2 was conducted in 1989, and involved face-to-face interviews with 83% (n=2867) of the wave 1 survivors.


Wave 3 was conducted approximately 7.5-years after Wave1 in 1994, and involved face-to-face or telephone interviews with 83% of wave 1 survivors or proxies (n=2562 of which 164 were proxies). Additional information on the ACL study design is provided elsewhere (House et al., 1994; Lantz et al. 1998). This study uses a longitudinal research design, through which a nationally representative populationbased cohort was followed over time. Data on demographic control variables (age, sex and race), socioeconomic variables (education and income), baseline health status and health-risk behaviors were taken from the 1986 Wave 1 survey. Data on subsequent health status were taken from the 1994 wave 3 survey. Deaths among study subjects from baseline through the third interview period in 1994 were identified using information from the National Death Index and informants, and were further verified with death certificate review in approximately 95% of cases. Reports of deaths not yet verified with a death certificate have been reviewed carefully, and death appears certain in all cases. Demographic and socioeconomic measures Age was based on self-reported information from the ACL wave 1 survey, and grouped into 6 categories: 25– 34, 35–44, 45–54, 55–64, 65–74, and 75 years and older. Other demographic factors included as control variables included sex (male vs female) and race (black vs nonblack). Since these variables are related to socioeconomic position, health-risk behaviors, and health status, they were included in the analysis as potential confounders. The two socioeconomic factors } education and income } were measured using self-reported information from the ACL wave 1 survey. Education was measured as total years of completed education, and categorized as 0–11, 12–15, and 16 or more years. Income was measured as the combined income from all sources for both respondents and their spouses in the preceding year. Income also was grouped into three categories: $0–$9999, $10,000–$29,999, and $30,000 or more. (More refined categories of education and income produced virtually identical results.) Behavioral risk factor measures Indicators of health-risk behavior were based on selfreported information from the wave 1 data. As reported in previous research using ACL survey data, there is a high degree of stability in self-reported health behaviors across study waves (Lantz et al., 1998). Regarding cigarette smoking, respondents were classified as never smokers, former smokers and current smokers. Alcohol consumption was coded using 3 categories based on the number of drinks consumed in the past month:

31

nondrinkers (0 drinks in the past month), moderate drinkers (1–89 drinks), and heavy drinkers (90 or more drinks). Body weight was measured using the body mass index (BMI), based on self-reported height and weight. Following the method of Berkman and Breslow (1983), those in the highest 15% of the sex-specific distributions were coded as overweight, while those in the lowest 5% were coded as underweight. A physical activity index was created based on answers to questions regarding how often the respondent took walks, did gardening or yard work, or engaged in sports or exercise. Physical activity index scores were divided into quintiles to create 5 groups of approximately the same size. The group in the top quintile represents those who are the most physically active.

Health status measures Two variables measuring different dimensions of health status were used as dependent variables: (1) functional status in regard to physical health; and (2) self-rated health. Self-reported data on perceived health status has been shown to be highly predictive of mortality and other health outcomes (Idler & Benyamini, 1997; Bosworth et al., 1999). Self-rated health as a measure of health status has also been shown to have high test–retest reliability (Lundberg & Manderbacka, 1996). In addition, both self-rated health and selfassessments of physical functioning are believed to be valid and very useful indicators for measuring population health (Avlund, 1997; Miilunpalo, Vuari, Oja, Pasanen & Urponen, 1997; Lundberg & Manderbacka, 1996). Thus, these two measures of health status were used as the outcome variables of interest in our analyses. An index of functional status was created, based on respondent answers to several questions. A low score of 1 represents confinement to a bed or chair, and a high score of 4 represents the ability to do heavy work inside or outside the home. These scores were transformed into a 3-category variable: (1) no functional impairment; (2) low impairment (i.e. unable to do heavy physical labor at work or around the house); and (3) moderate/severe impairment (i.e. unable to walk a few blocks or climb several stairs, or confined to a bed or chair). Self-rated health was measured with a single 5-category scale, which was collapsed into a 3-category variable for analysis: (1) excellent/very good; (2) good; and (3) fair/ poor. Analyses (not reported) showed that very little information was lost by collapsing categories. Data on functional impairment and self-rated health for the dependent variables were taken from the wave 3 data. In addition, data on these variables from the wave 1 survey (operationalized in the same way) were used to provide controls for baseline health status.

32


Statistical analysis The data were weighted in all analyses to adjust for differential response rates and design-based variation in probabilities of selection into the sample by age and race. Poststratification weights adjust the ACL wave 1 sample results to Bureau of the Census population estimates by sex, age and region of the country for July 1, 1986. Descriptive statistics were obtained through the Statistical Analysis System, SAS Institute, Inc, Cary NC. Health status in a longitudinal sample can be investigated in a variety of ways, including the following approaches: (1) analyzing predictors of health status at a later point in time controlling for baseline health; (2) analyzing predictors of health status at a later point in time for subsets of the population with different levels of health at baseline (i.e. the analysis is stratified by baseline health status); and (3) analyzing ‘‘change scores’’ or the difference in health status between two observational periods (which leads to the same interpretations about predictors of health status at a later point in time as the first approach described above). We employed the first two methods in the analyses presented here. Multinomial logistic regression analyses were performed to estimate the relative risk of specific health status outcomes at wave 3 in terms of demographic, socioeconomic and health-behavior variables, controlling for health status at wave 1. For analyses regarding functional health, the outcome variable had 5 possible values, all of which were measured at wave 3: (1) no functional impairment (the referent category); (2) low functional impairment; (3) moderate/severe functional impairment; (4) mortality between wave 1 and wave 3; and (5) survey nonresponse at wave 3. For analyses regarding self-rated health, the outcome variable also had 5 possible values: (1) excellent/very good self-rated health (the referent category); (2) good health; (3) fair/ poor health; (4) mortality between wave 1 and wave 3; and (5) survey nonresponse at wave 3. The models predict the relative risk of being in a specific health category, having died or not participating in the wave 3 survey compared with being in the best or optimum health category (i.e. no functional impairment or excellent/very good health). Because omitting those who died between wave 1 and wave 3 (n=546), or those who did not respond to the wave 3 survey (n=509) would bias the population base of the study sample, attrition due to death and survey nonresponse also were considered as outcomes. However, because the focus of this analysis is on predictors of health status, and because mortality findings are reported elsewhere, results regarding the risk factors for mortality and survey nonresponse are not presented; the results presented are adjusted for these competing risks (Lantz et al., 1998).

For each of the two health status outcomes under study, a series of multiple predictor models was estimated. First, the relative risk of being in a specific health status category was estimated for education and income groups (both separately and together), controlling for age, sex, race and Wave 1 health status. Second, the behavioral risk factors under investigation were added to the base model to see how much of the observed education and income differentials in health status could be attributed to these factors. Additional models were also run to test the sensitivity of the results to different specifications of the behavioral risk factors, and to investigate whether or not the impact of socioeconomic variables and health-risk behaviors varied across age, sex or racial groups. We also conducted stratified analyses to see if the general pattern of results produced with the approach described above held across subgroups with different levels of health status at baseline. We therefore performed the same series of multiple predictor models separately for six subgroups based upon their health status at wave 1: (1) no functional impairment; (2) low functional impairment; (3) moderate/severe functional impairment; (4) excellent/very good self-rated health; (5) good self-rated health; and (6) fair/poor self-rated health. This approach allowed us to, among other things, identify the relationship between socioeconomic factors and major health-risk behaviors in explaining negative changes in health status over time among those with good health at baseline.

Results At the time of the baseline (wave 1) survey, 84. 7% of the weighted sample (which represents the noninstitutionalized population of the United States in 1986) reported no functional impairment, 6.8% had low functional impairment, and 8.5% had moderate/severe functional impairment (Table 1). In addition, 64.2% of the weighted sample rated their own health to be ‘‘excellent’’ or ‘‘very good’’, 20.6% rated their health to be ‘‘good’’, and 15.2% rated their health to be ‘‘fair’’ or ‘‘poor’’ at the time of the wave 1 survey (Table 1). As reported previously, the ACL participants in the lowest income and education groups were significantly more likely to report having some level of functional impairment and fair/poor self-rated health at baseline (House et al., 1994). A significant proportion of the weighted study sample could be classified as socioeconomically disadvantaged (Table 1). One-fourth of the study population reported less than 12 years of education, and almost one-fifth reported incomes of less than $10,000. In addition, education and income both were significantly related to self-reported health-risk behavior, adjusting for age

33

P.M. Lantz et al. / Social Science & Medicine 53 (2001) 29–40 Table 1 Distribution of study variables in ACL population at wave 1, 1986 Variable

Unweighted (N=3617)

Unweighted percent

Weighted percent

740 591 390 685 765 446

(20.5) (16.3) (10.8) (18.9) (21.2) (12.3)

(29.0) (23.2) (14.6) (13.8) (12.5) (7.0)

Age in years

25–34 35–44 45–54 55–64 65–74 75+

Sex

Male Female

1358 2259

(37.5) (67.5)

(47.1) (52.9)

Race

White Black

2243 1174

(67.5) (32.5)

(89.0) (11.0)

Education

0–11 years 12–15 years 16+ years

1349 1768 500

(37.3) (48.9) (13.8)

(25.6) (54.7) (14.7)

Income

5$10K $10 K–$29K >$30K

1176 1475 966

(32.5) (40.8) (26.7)

(19.2) (40.5) (40.3)

Smoking

Current Past Never

1060 941 1616

(29.3) (26.0) (44.7)

(30.4) (27.5) (42.1)

Alcoholic drinking (Past month)

None Moderate High

1837 1650 130

(50.8) (45.6) (3.6)

(41.2) (54.5) (4.3)

Body Mass Index

Overweight Normal Underweight

679 2752 186

(18.8) (76.1) (5.1)

(15.3) (79.6) (5.1)

Physical activity

Quintile Quintile Quintile Quintile Quintile

1037 540 952 439 649

(28.7) (14.9) (26.3) (12.1) (17.9)

(21.3) (14.9) (27.4) (15.2) (21.3)

None Low Moderate/severe

2777 348 492

(76.8) (9.6) (13.6)

(84.7) (6.8) 8.5)

Excellent/very good Good Fair/Poor

2045 788 784

(56.5) (21.8) (21.7)

(64.2) (20.6) (15.2)

1 (Low) 2 3 4 5 (High)

Functional impairment

Self-rated health

differences in education and income subgroups. As reported in previous research using ACL data, lower age-adjusted levels of income and education were both associated with smoking, being overweight, being underweight, and having a low level of physical activity (Lantz et al., 1998). For example, 42.0% those with 0–11 years of completed education were current smokers compared with 33.1% of those with 12–15 years and 19.6% of those with 16 or more years ( p50.001). Similarly, 24.4% of those in the lowest income group had a BMI indicating overweight compared with 18.0%

of those in the middle income group and 14.0% of those in the lowest income group ( p50.001). Of the total wave 1 weighted sample, 9.9% died between wave 1 and wave 3 in 1994, and 12.6% were alive but did not participate in the wave 3 survey. In addition to the outcomes of being dead or a nonresponder, 64.7% of the weighted sample reported the same level of functional health, 9.1% experienced a decline in functional health, and 3.7% experienced an improvement at wave 3. Regarding self-rated health, 50.0% reported the same level of health, 18.8% reported

34


a decline in health, and 8.7% reported an increased level of health at wave 3. The socioeconomic variables of income and education were individually strong predictors of health change between wave 1 and wave 3. When education and income were included in the same model, both of these socioeconomic variables maintained a significant relationship with health status changes, adjusting age, sex, race and baseline health status (Model 1, Tables 2 and 3). Regarding physical functioning, those in the lowest

and middle educational groups were significantly more likely to have moderate/severe impairment at wave 3 (as compared to having no impairment), as were those in the lowest income group (Table 2, Model 1), adjusting for wave 1 health status. Regarding self-rated health, those in the lowest education group and those in the lowest income group were significantly more likely to have fair/poor health at wave 3 (as opposed to excellent/ very good health), adjusting for the fact that they were also more likely to have fair/poor health at baseline

Table 2 Odd ratios (95% CI) for functional impairment at ACL wave 3 from models adjusting for age, sex, race, and wave 1 functional impairment

Education 16+years 12–15 years 0–11 years Income $30,000 + $10-29, 999 $5$10,0000 Smoking status Never Current Former Alcohol past month Moderate Heavy None BMI Normal Underweight Overweight Physical activity Quintile 5 (high) Quintile 4 Quintile 3 Quintile 2 Quintile 1 (low)

Model 1 low functional impairment

Model 1 moderate/severe functional impairment

Model 2 low functional impairment

Model 2 moderate/severe functional impairment

1.0 1.18 (0.71, 1.96) 1.57 (0.86, 2.83)

1.0 1.72 (1.06, 2.78) 2.96 (1.75, 4.98)

1.0 1.20 (0.72, 2.01) 1.54 (0.84, 2.84)

1.0 1.49 (0.92, 2.42) 2.21 (1.30, 3.77)

1.0 0.82 (0.55, 1.24 0.98 (0.57, 1.66)

1.0 1.38 (0.98, 1.94) 2.11 (1.40, 3.20)

1.0 0.81 (0.54, 1.23) 0.99 (0.58, 1.71)

1.0 1.19 (0.84, 1.69) 1.89 (1.23, 2.89)

1.0 1.16 (0.77, 1.75) 0.81 (0.51, 1.30)

1.0 2.19 (1.56, 3.08) 1.42 (1.0, 2.03)

1.0 0.79 (0.55, 1.16) 0.97 (0.32, 2.88)

1.0 0.80 (0.37, 1.73) 1.33 (0.98, 1.79)

1.0 1.67 (0.74, 3.79) 1.44 (0.92, 2.24)

1.0 0.99 (0.51, 1.95) 1.93 (1.39, 2.66)

1.0 0.70 (0.38, 1.05 (0.65, 0.84 (0.47, 1.11 (0.66,

1.0 0.80 (0.46, 1.40 (0.91, 1.66 (1.04, 1.53 (0.98,

1.28) 1.70) 1.51) 1.86)

1.39) 2.15) 2.66) 2.39)

35


(Table 3, Model 1). These results suggest that those in the lowest income and education groups were significantly more likely to experience a health decline over time, even adjusting for health status at the baseline observation. When health-risk behaviors were added to the models, the relationship between the socioeconomic variables and the two health status outcomes attenuated somewhat, but remained strong and significant (Tables 2 and 3). For example, the relative risk of low education on moderate/severe functional impairment at wave 3 was 2.96 (95% confidence interval [CI]: 1.75, 4.98) in the

base Model 1, and 2.21 (95% CI: 1.30, 3.77) in Model 2 controlling for health-risk behaviors (Table 2). Similarly, the effect of low income on fair/poor health status was 2.16 (95% CI: 1.47, 3.17) in the base model, and 2.07 (95% CI: 1.40, 3.05) in the model controlling for health-risk behaviors (Table 3, Models 1 and 2). The attenuation of the socioeconomic effects when healthrisk behaviors were considered suggests that these behaviors do explain some of the association between socioeconomic position and health status. However, because a significant portion of the education and income effects remained after taking health behaviors

Table 3 Odd ratios (95% CI) for self-rated health at ACL wave 3 from models adjusting for age, sex, race, and wave 1 self-rated health

Education 16+years 12–15 years 0–11 years Income $30,000 + $10–29,999 $5$10,0000 Smoking status Never Current Former Alcohol past month Moderate Heavy None BMI Normal Underweight Overweight Physical activity Quintile 5 (high) Quintile 4 Quintile 3 Quintile 2 Quintile 1 (low)

Model 1 good health

Model 1 fair/poor health

Model 2 good health

Model 2 fair/poor health

1.0 1.22 (0.97, 1.54) 1.65 (1.2, 2.25)

1.0 1.46 (1.0, 2.13) 2.75 (1.78, 4.26)

1.0 1.16 (0.92, 1.47) 1.45 (1.05, 2.0)

1.0 1.36 (0.92, 1.99) 2.22 (1.42, 3.48)

1.0 0.99 (0.80, 1.21) 1.20 (0.88, 1.65)

1.0 1.26 (0.93, 1.70) 2.16 (1.47, 3.17)

1.0 0.96 (0.78, 1.18) 1.19 (0.87, 1.63)

1.0 1.19 (0.88, 1.62) 2.07 (1.40, 3.05)

1.0 1.12 (0.97, 1.52) 0.99 (0.79, 1.24)

1.0 1.43 (1.06, 1.94) 0.93 (0.68, 1.28)

1.0 1.52 (0.99, 2.34) 1.21 (0.97, 1.52)

1.0 0.79 (0.40, 1.56) 1.18 (0.90, 1.54)

1.0 0.73 (0.45, 1.18) 1.47 (1.14, 1.90)

1.0 0.58 (0.31, 1.09) 1.56 (1.14, 2.13)

1.0 1.39 (1.05, 1.16 (0.89, 1.46 (1.08, 1.32 (0.98,

1.0 0.97 (0.61, 1.64 (1.14, 1.28 (0.83, 1.82 (1.22,

1.86) 1.51) 1.97) 1.79)

1.54) 2.27) 1.98) 2.69)

36


into account, the results also suggest that the major health-risk behaviors considered in this study are not likely to be the dominant mechanism by which socioeconomic disparities in important measures of health status are produced. In addition to education and income, the variables of age, sex, race and baseline health status were all significantly associated with one or both measures of health status at wave 3 (results not shown). Regarding functional impairment, those with some impairment at baseline were significantly more likely to be have some level of impairment at wave 3. In addition, older age, being female and being black were all associated with a higher risk of impairment at wave 3. Regarding selfrated health, those who rated their health as good or fair/poor at baseline were more likely to rate their health as fair/poor at follow-up. In addition, black respondents had a higher risk of fair/poor health at wave 3, even controlling for socioeconomic factors (OR=1.85, 95% CI: 1.15, 2.96). Age and sex were not related to self-rated health at wave 3, controlling for baseline health status. Despite the finding that the major health-risk behaviors under study explained only a modest portion of the association between socioeconomic variables and declines in health status over time, several of these behavioral variables were significantly related to health change. Current smokers, those who were overweight, and those with lower levels of physical activity were more likely to experience declines in health, both in terms of moderate/severe impairment and fair/poor selfrated health (Model 2, Tables 2 and 3). For example, those who were overweight at baseline were almost twice as likely to report moderate/severe functional impairment at wave 3 (OR=1.93, 95% CI: 1.39, 2.66), adjusting for age, race, sex, socioeconomic variables and baseline health status. Similarly, current smokers had an adjusted elevated risk of moderate/severe functional impairment at wave 3 (OR=2.19, 95% CI: 1.56, 3.08). The results of analyses testing for interactions suggest that the general pattern of results presented in Tables 2 and 3 was consistent across age, sex and race subgroups. As discussed above, we also conducted stratified analyses to see if the general pattern of results reported above held across subgroups with different levels of health status at baseline, including those with excellent/ very good self-reported health and among those with no functional impairment at baseline. In general, the results showed strong income and education effects on health status at wave 3 in each of the health status subgroups, and that these effects attenuated a modest amount when the health-risk behavior variables were added to the models. In particular, analyses conducted on respondents in good health at wave 1 demonstrated strong income and education effects on health status at wave 3, and these effects remained strong and significant when

controlling for health-risk behaviors. For example, among those with no functional impairment at wave 1 (n=2777), the odds of having moderate/severe functional impairment at wave 3 were 3.9 (95% CI: 2.13, 7.3) for those with 0–11 years of education and 1.85 (95% CI: 1.12, 3.05) for those with incomes less than $10,000 per year. When controlling for health-risk behaviors, the odds of having moderate/severe functional impairment at wave 3 attenuated somewhat but still were statistically significant for those with low levels of education (OR=3.1, 95% CI: 1.62, 5.67) and low incomes (OR=1.73, 95% CI: 1.04, 2.89).

Discussion These results from a nationally representative sample of US adults show that both education and income have strong, graded associations with change in self-rated and functional health status in a 7.5 year prospective study. Men and women with less than 12 years of schooling, and those with incomes below $10,000 were two to three times more likely to have functional limitations and fair/ poor self-rated health over time, compared to more advantaged men and women. Baseline levels of smoking, physical activity, and overweight were important predictors of poorer health status after 7.5 years of followup, net of baseline health status. Despite this, the increased risks of adverse changes in health status associated with lower education and income were not substantially reduced by adjustment for these behavioral risk factors. It appears that, in this representative sample of US adults, higher levels of smoking, physical inactivity and overweight played a minor role in understanding how, over a 7.5-year time period, socioeconomic disadvantage translated into poorer functional and self-rated health. Moreover, in this study, education and income appeared to make unique contributions to changes in functional and self-rated health. Men and women with less than 12 years of schooling were more than twice as likely to have poorer functional and self-rated health, even after adjustment for their income levels. Similarly, men and women with incomes less than $10,000 were significantly more likely to have poorer functional and self-rated health, even after adjustment for their levels of education. While education and income reflect two temporally distinct stages of the lifecourse, this finding may also reflect some unique pathways between education and income and health status. For instance, while educational success generally translates into higher incomes, higher levels of education may also confer knowledge and cognitive assets that are health protective. Conversely, income may imply more than just returns on education. It may also represent access to tangible resources at both the individual and community


level that have implications for health status, including such things as better housing, working conditions, food and health care, and increased social support and community cohesion. The findings reported here differ in some important ways from a study in this same population that examined the association between education, income, the same four health-risk behaviors, and mortality (Lantz et al., 1998). In that study, education was unrelated to mortality after adjustment for income. In the present study, both education and income made unique contributions to predicting levels of health status after 7.5 years of follow-up. Education may be more strongly related to measures of functioning and selfrated health because cognitive assets gained through formal education could be important in managing, interpreting and reporting on-going health problems. Additionally, from a lifecourse perspective, the period of formal education may be important in the development, adoption and maintenance of certain health protective and health-risk behaviors. Thus, it is possible that educational experiences may be more sensitive measures of socioeconomic conditions at an important developmental stage of the lifecourse, and thus may be more strongly linked to functional and self-rated health status through behavioral mechanisms than are earnings in later life. Indeed, there is some evidence for this in these data, since the attenuating effect of adjustment for health behaviors was more marked for the effects of education than for income. For instance, odds ratios for moderate/severe functional impairment were reduced by approximately 25% in the least-educated group, compared to 10% in the lowest income group, after adjustment for health behaviors. Thus, there is some evidence for a stronger link between education, risk behavior and health status indicators than for income. However, studies in Europe, while not directly comparable with these data, have shown that occupation is a better predictor of smoking than is education (Davey Smith et al., 1998). This approach to understanding the pathways that link particular socioeconomic indicators with health might suggest that the association between income and health status may be partly mediated by factors associated with the nature of the work environment. Our main finding that the behavioral risk factors of smoking, alcohol consumption, physical inactivity and overweight statistically account for only a small part of the increased risk of poor health status is consistent with other recent studies with more restrictive samples (Lynch et al., 1997; Mansson et al., 1998). A limitation of our work is that some important personal health-risk behaviors (e.g. illicit drug use, specific dietary habits, risky sexual behaviors, seat belt use, etc.) were not included in the ACL study. Thus, we were able to investigate the contribution of only a subset of health-

37

risk behaviors to the socioeconomic gradient in health status. However, our set of four health-risk behaviors includes the major behavioral risk factors associated with heart disease, cancer and other leading causes of morbidity (McGinnis & Foege, 1993). In addition, it is these specific risk factors that are most frequently invoked in the behavioral explanation or hypothesis under study. Another limitation is that the behaviors included in these analyses are measured with error (as they are in all studies), and their potential to confound associations between health status and socioeconomic indicators depends on how they are specified in the model and the extent of measurement error. Thus, any estimate of their confounding effects should always be considered within the context of the specified regression model, and the extent and type of measurement error involved. It is likely that these analyses underestimate the confounding effects of health behaviors. In addition, a further limitation of our approach is that it is possible that one-time measures of health-risk behaviors do not fully capture exposure over the lifecourse and thus they are an incomplete marker for the impact of behavioral factors on health status. Nevertheless, the modest impact of four major health behaviors on the strong association between socioeconomic position and health observed in these longitudinal data suggests that while these behaviors are clearly important determinants of health status and play a mediating role in health disparities, they do not shed an enormous amount of light on our understanding of health status inequalities among adults. This should not be taken to mean that major healthrisk behaviors are not important determinants of individual or population health. For example, it is clear that smoking has serious negative consequences for health, and that different rates of tobacco use across socioeconomic groups significantly accounts for differentials in associated cancer incidence and mortality (Harrell et al., 1998; Stellman & Resnicow, 1997). In addition, the impressive declining trends in heart disease in many countries attest to the importance of reductions in smoking and dietary fat (Tunstall-Pedoe et al., 1999). However important these secular changes in health behaviors have been in driving down overall population rates of disease, they nonetheless do not seem to explain the persistent mortality and morbidity differences that continue to exist across socioeconomic groups. Thus, while changing individual behaviors in the areas of smoking, alcohol consumption, exercise, and body weight remain critically important public health priorities, the results of this study imply that even targeted behavioral interventions may have limited impact on reducing inequalities in health among socioeconomic groups. The policy implications of this research are important. The results suggest that if we only focus on individual health-risk behaviors in our policy responses

38


to social inequalities in health, we will likely not achieve great success in reducing them. Observed socioeconomic disparities in adult levels of functioning and self-rated health may be best understood as resulting from an accumulating cascade of events, environments, and experiences from childhood to adulthood that eventually leave their imprint on levels of health and functioning (Lynch et al., 1997). Individual health-risk behaviors clearly play a role in this process, but they are not the dominant mediating mechanism that explains socioeconomic disparities in health status. This is despite the popularity of this hypothesis and how it has been frequently invoked to ‘‘explain away’’ socioeconomic inequalities in health (Lynch et al., 1997; Macintyre, 1997). Macintyre (1997), in her overview of Britain’s Black Report, wrote that there are both ‘‘hard’’ and ‘‘soft’’ interpretations of various explanations for socioeconomic differentials in health, including the behavioral explanation investigated here. The ‘‘hard’’ version of this explanation is that socioeconomic disparities in health can be accounted for by risky or health-damaging behaviors (including a broader array of behaviors than are available in the ACL data). The ‘‘soft’’ version of this explanation is that health-related individual behaviors ‘‘do not explain away class differences, but contribute to them, and push the explanatory task further back to ask why such behaviours are persistently more common in poorer groups’’ (Macintyre, 1997). In other words, the position that individual health-risk behaviors explain some portion of socioeconomic differentials in health still begs the question of why these risk behaviors are all patterned by social position in the first place (Lynch et al., 1997). From this perspective, it also could be argued that even if healthrisk behaviors empirically explained more of the association between socioeconomic position and health than the ACL results suggest, that because health-risk behaviors themselves are influenced by socioeconomic position and context, their statistical control represents some over-adjustment for the total effects of socioeconomic position on health (Lynch et al., 1996; Davey Smith et al., 1998b). Explanations for the link between socioeconomic position and health ultimately must do more than describe how socioeconomic disadvantage can negatively influence health through behavioral, psychological, social, or medical care pathways at the individual level. While this approach may provide valuable insights into the mechanisms involved, and in some cases even suggest potential interventions, it obviously cannot answer questions of why socioeconomic inequalities in health exist across societies and historical contexts (Kaplan & Lynch, 1997; Muntaner, 1999; Muntaner & Lynch, 1999; Lynch, Davey Smith, Kaplan & House, 2000). To answer questions of why socioeconomic

disparities in health exist at all, we must continue to investigate the historical, political, economic, and social processes that differentially allocate negative health exposures and protective health resources across different groups in the population (Link & Phelan, 1995; Davey Smith 1996; Lynch, 2000).

Acknowledgements This study was supported by Grants P01AG05561 and R01AG09978-01 from the National Institute on Aging, National Institutes of Health, Bethesda, MD, and by a Health Investigator Award (to Dr. House) from the Robert Wood Johnson Foundation, Princeton, NJ.

References Anonymous. (1998). Executive summary of the clinical guidelines on the identification, evaluation and treatment of overweight and obesity in adults. Archives Internal Medicine, 158, 1855–1998. Avlund, K. (1997). Methodological challenges in measurements of functioning ability in gerontological research: A review. Aging, 9(3), 164–174. Bender, R., Trautner, C., Spraul, M., & Berger, M. (1998). Assessment of excess mortality in obesity. American Journal of Epidemiology, 147, 42–48. Berkman, L. F., & Breslow, . L. (1983). Health and ways of living. New York, NY: Oxford University Press. Blaxter, M. (1987). Evidence on inequality in health from a national survey. Lancet, 2(8549), 30–33. Blaxter, M. (1997). Whose fault is it? People’s own conceptions of the reasons for health inequalities. Social Science and Medicine, 44, 747–756. Bosworth, H. B., Siegler, I. C., Brummett, B. H., Barefoot, J. C., & Williams, R. B., et al. (1999). The association between self-rated health and mortality in a well-characterized sample of coronary artery disease patients. Medical Care, 37(12), 1226–1236. Davey Smith, G., 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. Davey Smith, G. (1996). Income inequality and mortality: Why are they related? Income inequality goes hand in hand with underinvestment in human resources. British Medical Journal, 312, 987–988. Davey Smith, G., Hart, C., Blane, D., & Hole, D. (1998). Adverse socioeconomic conditions in childhood and cause specific adult mortality: Prospective observational study. British Medical Journal, 316, 1631–1651. Davey Smith, G., Hart, C., Hole, D., et al. (1998). Education and occupational social class: Which is the more important indicator of mortality risk? Journal of Epidemiology and Community Health, 52, 153–160.

P.M. Lantz et al. / Social Science & Medicine 53 (2001) 29–40 Fraser, G. E., Sumbureru, D., & Pribis, P., et al. (1997). Association among health habits, risk factors, and all-cause mortality in a black California population. Epidemiology, 8(2), 168–174. Haan, M. N., Kaplan, G. A., & Syme, S. L. (1989). Socioeconomic status and health: Old observations and new thoughts. In J. P. Bunker, D. S. Gomby, & B. H. Kehrer (Eds.), Pathways to health: The role of social factors (pp. 76–135). Menlo Park, CA: HJ Kaiser Family Foundation. Harrell, J. S., & Bangdiwala, S. I., et al. (1998). Smoking initiation in youth: the roles of gender, race, socioeconomics and developmental status. Journal of Adolescent Health, 23, 271–279. Healthy People 2000. National health promotion and disease prevention objectives. Washington, DC: US Dept of Health and Human Services, 1990. DHHS publication 91–50212. Hirdes, J. P., & Forbes, W. F. (1992). The importance of social relationships, socioeconomic status and health practices with respect to mortality among Ontario males. Journal of Clinical Epidemiology, 554, 175–182. House, J. S., Lepkowski, J. M., Kinney, A. N., & Mero, R. P., et al. (1994). The social stratification of aging and health. Journal of Health and Social Behavior, 35, 213–234. Idler, E. L., & Benyamini, Y. (1997). Self-rated health and mortality: A review of twenty-seven community studies. Journal of Health and Social Behavior, 38, 21–37. Kaplan, G. A., & Lynch, J. W. (1997). Whither studies on the socioeconomic foundations of population health. (Editorial). American Journal of Public Health, 87, 1409–1411. Krieger, N., Rowley, D. L., Herman, A. A., Avery, B., & Phillips, M. T. (1993). Racism, sexism and social class: Implications for studies of health, disease and well-being. American Journal of Preventive Medicine, 9(Suppl), 82–122. Lantz, P. M., House, J. S., Lepkowski, J. M., Williams, D. R., Mero, R. P. Chen, Jieming. (1998). Socioeconomic factors health behaviors and mortality: Results from a nationally representative prospective study of US adults. Journal of American Medical Association, 279, 1703–1708. Link, B. G. & Phelan, J. C. (1995). Social conditions as fundamental causes of disease. Journal of Health and Social Behavior, (extra issue), 80–94. Liu, K., Cedres, L. B., & Stamler, J., et al. (1982). Relationship of education to major risk factors and death from coronary heart disease, cardiovascular diseases, and all causes. Circulation, 66, 1308–1314. Lundberg, O., & Manderbacka, K. (1996). Assessing reliability of a measure of self-rated health. Scandinavian Journal of Social Medicine, 24(3), 218–224. Lynch, J. W. (2000). Income inequality and health: Expanding the debate. Social Science and Medicine, 51, 1001–1005. Lynch, J. W., & Kaplan, G. A. (2000). Socioeconomic position. In L. S. Berkman, & I. Kawachi (Eds.), Social epidemiology (pp. 13–35). New York: Oxford University Press. Lynch, J. W., Davey Smith, G., Kaplan, G. A., & House, J. (2000). Income inequality and mortality: Importance to health of individual income, psychosocial environment, or material conditions. British Medical Journals, 320, 1200– 1204.

39

Lynch, J. W., Kaplan, G. A., & Salonen, J. T. (1997). Why do poor people behave poorly? Variations in adult health behaviour and psychosocial characteristics, by stage of the socioeconomic lifecourse. Social Science and Medicine, 44, 809–820. Lynch, J. W., Kaplan, G. A., & Shema, S. J. (1997). Cumulative impact of sustained economic hardship on physical, cognitive, psychological and social functioning. New England Journal of Medicine, 337, 1889–1895. Lynch, J. W., Kaplan, G. A., Cohen, R. D., Tuomilehto, J., Salonen, J. T. (1996). Do cardiovascular risk factors explain the relation between socioeconomic status, risk of all-cause mortality, cardiovascular mortality and acute myocardial infarction? American Journal of Epidemiology, 144, 934–942. Macintyre, S. (1997). The Black report and beyond: What are the issues? Social Science and Medicine, 44, 723–746. Mansson, N. O., Rastam, L., Eriksson, F. K., & Israelsson, B. (1998). Socioeconomic inequalities and disability pension in middle-aged men. International Journal of Epidemiology, 27, 1019–1025. Marmot, M., Ryff, C. D., & Bumpass, L. L., et al. (1997). Social inequalities in health: next questions and converging evidence. Social Science and Medicine, 44, 901–910. McGinnis, M. J., & Foege, W. H. (1993). Actual causes of death in the United States. Journal of American Medical Association, 270, 2207–2212. Miilunpalo, S., Vuari, I., Oja, P., Pasanen, M., & Urponen, H. (1997). Self-rated health status as a health measure: the predictive value of self-reported health status on the use of physician services and on mortality in the working-age population. Journal of Clinical Epidemiology, 50(5), 517–528. Muntaner, C. (1999). Invited commentary: Social mechanisms, race and social epidemiology. American Journal of Epidemiology, 150, 121–126. Muntaner, C., & Lynch, J. W. (1999). Income inequality and social cohesion versus class relations: A critique of Wilkinson’s neo-Durkheimian research program. International Journal of Health Services, 29, 59–81. Mustard, C. A., Derksen, S., Berthelot, J. M., Wolfson, M., & Roos, L. L. (1997). Age-specific education and income gradients in morbidity and mortality in a Canadian province. Social Science and Medicine, 45, 383–397. National Center for Health Statistics (1998). Health, United States, 1998 With Socioeconomic Status and Health Chartbook. Hyattsville, Maryland: NCHS. 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. 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. Stellman, S. D., Resnicow, K. (1997). Tobacco smoking, cancer and social class. vol. 138. (pp. 229–250). Moscow: IARC Scientific Publications. Tunstall-Pedoe, H., Kuulasmaa, K., & Mahonen, M., et al. (1999). Contribution of trends in survival and coronaryevent rates to changes in coronary heart disease mortality: 10-year results from the 37 WHO MONICA project populations. Lancet, 353, 1547–1557.

40


Vita, A. J., Terry, R. B., Hubert, H. B., & Fries, J. F. (1998). Aging, health-risks, and cumulative disability. New England Journal of Medicine, 338(15), 1035–1041. Williams, D. R. (1990). Socioeconomic differentials in health: A review and redirection. Social Psychology Quarterly, 53, 81–88.

Winkleby, M. A., Fortmann, S. P., & Barrett, D. C. (1990). Social class disparities in risk factors for disease: Eight-year prevalence patterns by level of education. Preventive Medicine, 19, 1–12.