| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
TOPIC 6. CUMULATIVE ADVERSITY AND HEALTH INEQUALITIES |
Department of Sociology, Duke University, Durham, North Carolina.
Address correspondence to Angela M. O'Rand, Department of Sociology, Box 90088, Duke University, Durham, NC 27708-0088. E-mail: aorand{at}soc.duke.edu
Abstract
Objectives. This article examines how processes of cumulative adversity shape heart attack risk trajectories across the life course.
Methods. Our sample includes 9760 Health and Retirement Study respondents born between 1931 and 1941. Using self-reported retrospective measures of respondents' early background, we first identify three latent classes with differential exposure to childhood disadvantage. Intervening covariates associated with educational attainment, employment status, income attainment, marital history, and health behaviors are added to capture sequential processes of adversity. Final latent-class cluster models estimate the cumulative impact of these covariates on three different heart attack risk trajectories between 1992 and 2002: high, increasing, and low.
Results. Early disadvantage and childhood illness have severe enduring effects and increase the risk for heart attack. Adult pathways, however, differentially influence trajectories of heart attack risk and mediate the effects of early disadvantage.
Discussion. Findings suggest that future research should consider how processes of cumulative adversity initiated in childhood influence health outcomes in older ages.
CUMULATIVE adversity connotes a protracted chain of life course "insults" (Hayward & Gorman, 2004
), wherein childhood disadvantage is compounded or amplified across the life course by successively contingent structural constraints, life course transitions, and health behaviors that increase disease risk. Hence, it operates as a sequentially contingent process, that is, as a process whereby earlier disadvantages lead to later disadvantages and to an increasingly compromised capacity to respond to and manage new insults. However, the mechanisms by which baseline socioeconomic disadvantage persists to increase life-threatening illness are not well understood.
The following article examines processes of cumulative adversity using six waves of the Health and Retirement Study (HRS) between 1992 and 2002. It takes advantage of the addition of childhood recall variables added to the full panel in 1998 to construct latent classes of childhood advantage and disadvantage. It also constructs latent trajectories of heart attack risk in adulthood. The latent-class approach permits the identification of heterogeneity in life course origins and outcomes and controls for unobserved heterogeneity at both points in the life course. Intervening covariates associated with educational attainment, employment status, income attainment, marital history, and health behaviors are included to capture the sequential process of adversity.
Cumulative Adversity: Latent and Pathway Processes
Life course research focuses on heterogeneity and inequality in social origins and outcomes and on the mechanisms that link them. Early life inequalities begin with inherited or in utero vulnerabilities, and quickly widen as a result of differential exposure to environmental hazards including disease, poverty, and family distress (e.g., Elo & Preston, 1992). The rapidly developing child is thus embedded in an early biological and social environment with fateful effects (Kuh & Davey-Smith, 1997). Hertzman (1999) argues that this biological embedding establishes an underlying, or latent, condition with lifelong implications. These origins also set individuals on diverse life course pathways in adolescence and beyond that expose them differentially to life course risks or protections that can lead to the acquisition of resources (such as higher education) to overcome early disadvantages or to progressively deteriorating experiences (such as unemployment or detrimental health behaviors) that can accelerate adult morbidity or mortality. Latent conditions exert a biosocial gravity on the human development process, and pathway processes are constrained by these forces although, by chance and by choice, they can mediate the effects of early life conditions and redirect the life course.
A myriad of studies linking socioeconomic status (SES) and health warns us that a complex array of causal processes drives this relationship over the life course. Because social processes, such as those underlying the association between SES and health, are intrinsically selective processes, we expect that selectivity occurs both between and within salient population subgroups. Selection, however, plays only a partial role (Goldman, 2001). Preston, Hill, and Drevenstedt's (1998) schema of the array of possible associations between childhood inequality and mortality and/or longevity effectively demonstrates this complexity. They propose ways that childhood hardship can be linked to at least four different adult mortality effects: direct and positive, implying a fateful effect; direct and negative, implying an immunity effect; indirect and positive, operating through intervening pathways; and indirect and negative, implying a robust survivor process. The plausibility of these competing processes across subgroups of the population motivates much current research.
A large and growing body of research is identifying the sources of childhood adversity and the specific processes by which they predict cumulative hardship (Robert & House, 2000). Three major sources of childhood adversity include childhood poor health, economic hardship, and family instability. Poor child health has been found in the HRS to predict poor self-assessed health in adulthood (Elo, 1998) as well as higher risks for a variety of chronic diseases such as cancer, lung disease, cardiovascular conditions, and arthritis and rheumatism in late middle-age (Blackwell, Hayward, & Crimmins, 2001). Poor nutrition, childhood chronic illness, and exposure to environmental hazards (e.g., lead) have been specifically implicated in early and late health outcomes (e.g., Holland et al., 2000).
Studies from the United States and several other countries repeatedly find that economic hardship in childhood is clearly associated with later economic status, poorer physical and mental health in adulthood, and higher death rates (e.g., Kuh & Davey-Smith, 1997). A recent study of a British postwar cohort reports that economic hardship in childhood, measured by father as manual laborer, living in substandard housing, and receiving the poorest care, doubles the rate of mortality among adults living in the best SES conditions. The death rate for those for whom disadvantage continues into adulthood is between three and five times higher than that for those from the most advantaged conditions (Kuh, Hardy, Langenberg, Richards, & Wadsworth, 2002). A U.S. study using a panel of adults from Alameda County reports similar cumulative effects of early hardship on health (Lynch, Smith, Kaplan, & Shema, 1997). This study tracked the frequency with which adults' household incomes fell to less than 200 percent of the poverty level in 1965, 1974, and 1983. Those adults with sustained economic hardship were much more likely to have difficulties with instrumental activities of daily living, activities of daily living, and clinical depression.
The third source of childhood adversity identified across countries is family distress. Family instability and conflict appear to have enduring effects on some health outcomes (Lundberg, 1993). Divorce and marital distress among parents creates an unstable environment of mistrust, distraction, and parental inattention. Resources for the support and nurturance of healthy and unhealthy children are strained and stretched thin as a result of family distress, even in intact households (Wickrama, Lorenz, & Conger, 1997). Health-related behaviors such as regular bedtimes and home-prepared meals are often sacrificed in these environments. Hence, family distress directly affects childhood health and childhood opportunities for upward mobility, and probably directly and indirectly affects health and mortality risk in adulthood.
Adulthood experiences also introduce differential opportunities for the maintenance of health. Occupational environments and roles present rewards and hazards to workers that influence health. Physically demanding and/or stressful jobs have negative effects on health and mortality (Moore & Hayward, 1990). Access to health insurance is an additional work-related resource that mediates health risks. And, the risk of unemployment exacerbates the cumulative process of adversity through the loss of health insurance and income (Cubbins & Parmer, 2001). Similarly, the family experiences of adults can present hazards and protections for health maintenance. Adult experiences with family adversity, including divorce and household poverty, have short-term and long-term impacts on mental and physical health (Ku & Davey-Smith, 1997). Accordingly, the cumulative effects of latent and pathway processes are initiated in childhood and compounded or ameliorated in adulthood.
Challenges to Life Course Research on Cumulative Adversity
Most previous research on the enduring effects of childhood socioeconomic origins and experiences on adult health and mortality has faced conceptual and methodological challenges. First, most studies do not clearly specify causal mechanisms linking childhood SES and adult health. Cross-sectional and aggregate level designs reveal robust correlations of these variables but are hampered in the identification and specification of fundamental and intervening mechanisms associated with the adult life course (such as sustained economic hardship beyond childhood or marital history) and with health behaviors (smoking, obesity, and exercise). However, recent longitudinal examinations have introduced models that permit the integration of latent and pathway processes and the identification of direct and indirect selection effects (Goldman, 2001). There is evidence, for example, that upward social mobility diminishes the effects of early childhood hardship on adult health. This has been found in studies of adult cognitive functioning (e.g., Turrell et al., 2002) and physical health and mortality (e.g., House, Kessler, Herzog, Mero, & Breslow, 1990). Higher educational attainment, and its subsequent impact on occupational status, earnings, and household income, is perhaps the major pathway out of sustained hardship in the U.S. context and, arguably, the most critical long-term protective factor in health maintenance (Ross & Wu, 1995).
How education operates to affect health outcomes is still unclear. Education can be considered a stock of life course capital that operates in complex ways: As human capital it enriches cognitive and technical skills, improves problem-solving, and provides higher returns for work; as social capital it facilitates higher levels of social integration and social support; and as personal capital it motivates healthier lifestyles, flexibility, and greater personal efficacy and control (O'Rand, 2002). However, measured years of schooling can mediate the effects of early life circumstances in diverse ways. Higher educational attainment can attenuate the effects of early childhood adversity on health through the provision of protective resources noted above. Alternatively, lower educational attainment can amplify the effects of early childhood adversity through a process of cumulative adversity.
The second major challenge to research in this area is reliance on data sets that provide only retrospective accounts of a few aspects of childhood circumstances. Besides parental education, parental occupational statuses, and exposure to poverty in childhood, other characteristics of the parental household are not regularly measured. Such data cannot account for all early life sources of heterogeneity pertinent to the problem at hand. In addition, retrospective data can obviously suffer from problems of recall. Thus, life course researchers face the stubborn problem of unmeasured selective processes that may influence the outcome(s) of interest and bias the true effects of measured control variables. In the current study, we confront some of the problems of selectivity by using eight manifest indicators of childhood environment and latent-class cluster analyses.
Risk of Heart Attack
Exposure to early and midlife economic hardship has been found to influence overall mortality (Link & Phelan, 1995) and specific mortality associated with coronary heart disease risks among women and men, although most research has focused on middle-aged males (Kaplan & Keil, 1993). A recent study based on the National Longitudinal Survey of Older Men reports that mortality among men is positively influenced by early deprivation, but this influence is mediated by adult achievement levels and lifestyle or health-related factors (Hayward & Gorman, 2004). Childhood circumstances were indicated by the education of the household head, parents' nativity, living arrangements with parents, mother's employment status, and rural or urban residence. The addition of adulthood occupational and marital status, asset and income levels, obesity, and lifestyle behaviors to the analyses mediated the effects of some childhood conditions.
Childhood conditions also affect women's mortality risk. A recently published study of women's all-cause, cardiovascular, and noncardiovascular mortality based on four waves of the Alameda County Study (1965, 1974, 1983, 1994) reports that childhood hardship contributes to women's mortality due to cardiovascular disease, but was unrelated to deaths associated with other causes (Beebe-Dimmer et al., 2004). Childhood household variables were limited to recalled father's education and occupation. Intervening variables included respondent's education, occupation, and household income and spouse's (husband's) occupation. Overall mortality was most highly associated with the lowest educational levels, but education most strongly affected noncardiovascular deaths. Household income, in contrast, strongly increased the rates of mortality overall.
Cardiovascular diseases lend themselves to this study because they have been closely tied to the effects of cumulative stress over the life course (e.g., Singer & Ryff, 1999, using the Wisconsin Longitudinal Study; Seeman, McEwen, Rowe, & Singer, 2001, using the MacArthur Studies of Successful Aging) and because they appear to occur in men and women with enough frequency in late middle-age to permit a multivariate longitudinal analysis. Until recently, our understanding of the etiology of risk factors for heart disease centered on adult health behaviors, such as diet and physical exercise, with little attention paid to childhood factors (Leon & Ben-Shlomo, 1997). Thus, research is needed to increase our understanding of the extent to which early-life disadvantage increases the risk for heart attacks through adulthood.
METHODS
The data we use for this analysis come from the HRS, a national panel study launched in 1992 to collect information on the physical, mental, social, and financial well-being of Americans over the age of 50. Every other year since then, respondents have reported stability and changes in their health status, employment, and income, with proxy interviews conducted after death. Our sample includes the 9760 respondents born between 1931 and 1941 who were 51–61 years old in 1992 and who aged into their 60s and early 70s by 2002. We selected the HRS because it is one of the rare data sets that include information on childhood health and living conditions and health outcomes during young, middle, and older adulthood.
Between 1992 and 2002, respondents reported whether or not a doctor had ever told them they had had a heart attack. For each survey year, we created a dichotomous indicator that equals 1 if respondents had a heart attack by that year, 0 otherwise. To test the hypothesis that those disadvantaged early in life would be at greater risk of heart attack throughout adulthood, we used eight self-reported retrospective measures of respondents' childhood circumstances. In 1996, the HRS included a module that gathered information on respondents' childhood health and family background. They were asked to think about their health and home life when they were growing up, from birth to age 16. Elo's (1998) examination of these data and findings of the significant relationship between childhood circumstances and adult health prompted the inclusion of childhood measures for all HRS respondents in 1998. We created dichotomous variables comparing respondents who reported poor childhood health to those who said that their health was excellent, very good, good, or fair. We compared those who reported that their families were pretty well off or average financially to those with a poor family of origin. Respondents reported whether or not financial difficulties while they were growing up caused the family to move (1 = yes; 0 = no) or to receive help from relatives (1 = yes; 0 = no). Our models also included measures for mothers' education and fathers' occupation (1 = farming, forestry, fishing, mechanics and repair, machine/equipment operator, construction trade; 0 otherwise). Finally, respondents reported whether their father was unemployed for several months or more and whether or not they never lived with their father. In preliminary analyses, we tested alternative methods of variable coding for nondichotomous indicators, such as childhood health and family's socioeconomic position. We tested more detailed distinctions for father's occupation, which included manager, professional, salesman, clerk, service worker, farmer, machine/equipment operator, and other manual laborer. These alternative measures of childhood conditions were not statistically significant predictors of latent-class membership. Thus, we chose the above scoring to test the core hypotheses of this study because it identified clusters of respondents who shared similar early-life environments.
The demographic variables in our models include gender, race (white vs all others), and age (aged 55 and older in 1992 = 1; 0 otherwise). Because we are interested in how both educational marginalization and each additional year of schooling influence health trajectories, we measured education both as the number of years of schooling completed by 1992 and a dichotomous measure of whether or not respondents lacked a high school diploma. Our employment measures included three dichotomous indicators, including whether or not respondents in 1992 were employed, had a self-defined stressful job, and lacked employer-provided health insurance. We measured income both as total household income during 1991, logged, and as a dichotomous indicator of whether or not 1991 household income was less than half the median income, which we define as low income. Similar to that for education, this coding scheme allowed us to test whether the relationship between income and heart attack risk is linear, with each additional dollar of household income predictive of low risk heart attack trajectories, or whether location on the extreme low end of the income distribution more strongly predicts increased risk for heart attack. We compared those who had ever been divorced by 1992 to those who had not gone through a divorce. Finally, we included three dichotomous measures for respondents' health behaviors in 1992, comparing those who had ever smoked, never exercised, and were overweight (body mass index > 27), to the reference group. Table 1 provides descriptive statistics for the measures.
|
). Traditional models, such as conventional regression or discriminant analyses, contain parameters that describe only relationships between observed variables. We are interested, however, in associations among variables that may arise because of an unobserved, latent source of variation. In this study, the seven observed measures of childhood environment could not measure directly nor capture completely without error early exposure to disadvantage, so we tried to detect this unobservable characteristic through latent-class analyses. Computationally, latent-cluster membership for respondents showing various observed response vectors were predicted using Bayes' theorem (Vermunt & Magidson, 2000; see also Hamil-Luker, Land, & Blau, 2004). We first estimated one-cluster models, then estimated two-, three-, and four-cluster models. We assessed model fit by comparing likelihood ratio chi-square statistics (L2), Bayes information criteria (BIC), and bivariate residuals. We concluded that a three-cluster model—the clusters we identified as early disadvantaged, fatherless, and early advantaged—provided the best fit to the data. Then all cases with a given response vector were classified into the latent class for which the posterior (Bayesian) probability is highest (Dayton, 1998).
In the next stage of analyses, latent-class cluster models used measured indicators of whether respondents had a heart attack in each survey year to identify different trajectories of risk for heart attack over time. We first estimated a one-cluster model, which tests the idea that all respondents share similar heart attack trajectories. We compared this baseline model to models identifying two, three, and four different heart attack trajectories. Because BIC statistics balance a reward for goodness-of-fit with a penalty for model complexity, we chose the model with the lowest absolute value for the BIC statistic (1-cluster = 35019.06; 2-cluster = 13301.64; 3-cluster = 10841.50; 4-cluster = 11872.54). Classification statistics from the selected three-cluster model indicated that only 0.004 percent of cases may be misclassified. As the dependent variables are dichotomous, we used a finite mixture of logistic regression models. Finite mixture regression models allow for the incorporation of covariates as regressors directly into the estimation of a latent-class model, so we then extended the model to include various covariates to predict latent-class membership. We estimated the models by use of the Latent Gold program (Vermunt & Magidson, 2000).
Researchers studying health inequalities, especially at older ages, must pay careful attention to unmeasured selective survivorship. In panel studies, sample attrition may bias results in the same direction as hypothesized for morbidity or mortality selection. Because sicker people die at younger ages, the samples on which investigators base their conclusions disproportionately include healthier older adults. An additional source of sample selection bias comes from the exclusion of people who are institutionalized or otherwise unavailable for an interview for health reasons. Thus, characteristics of respondents in panel studies who are lost to follow-up tend to be related to poor health, especially at older ages (Goldman, Korenman, & Weinstein, 1995).
Proxy interviews conducted after the deaths of most HRS respondents partially reduce such selection bias. To account for additional unmeasured selective survivorship, we used a modified pseudo-variables approach similar to that implemented by Beckett (2000). First, we estimated binary logistic regression models in which the likelihood of having a heart attack in each survey year is predicted by health status in 1992, age, education, race, marital status, and income. Next, we applied the estimated coefficients obtained from the observed sample to the out-of-sample respondents to obtain their predicted probability of heart attack. This allowed us to assess the sensitivity of our results to the exclusion of people due to death, institutionalization, or loss to follow-up. Comparison of results including and excluding respondents lost to follow-up revealed only slight variations in parameter estimates, none of which altered levels of statistical significance. Because the same conclusions may be drawn from both samples, we presented only the results that adjust for unmeasured selective survivorship.
RESULTS
To examine pathways between early environment and heart attack risk across the life course, we first identified latent clusters of childhood circumstances that could condition the probability of future health problems. Table 2 presents results from the latent-class cluster analysis that identified three groups with differential exposure to early disadvantage. Large positive estimates are associated with predicted membership into that cluster, and negative numbers are predictive of nonmembership.
|
Respondents who never lived with their father are most likely to be assigned membership in cluster 2. As shown in Table 3 below, none of the members of this cluster lived with their father or identified their father as unemployed, disabled, or a manual worker. Thus, we label this cluster the fatherless. Because our sample was born between 1931 and 1941, many fathers probably died during World War II. The identification of the fatherless cluster demonstrates that we cannot simply conceive of childhood disadvantage along a one-dimensional continuum and locate individuals along a line that stretches from the least to the most disadvantaged. There are categorical differences in kind, not only degree, of exposure to adverse life circumstances.
|
Figure 1 graphically depicts the three heart attack risk trajectories identified by latent-class cluster analysis. Eighty-six percent of the sample is clustered in the low risk trajectory, with virtually no probability of experiencing a heart attack through 2002. The model also identifies a high risk trajectory, comprised of 8 percent of the sample with elevated risk for heart attack. In 1992, three-quarters of this cluster had experienced a heart attack; 10 years later, they all had. Finally, the 6 percent of the sample clustered in the increasing risk trajectory were indistinguishable from those at low risk until 1996 when their probability of heart attack began to rise dramatically. By 2002, all of those in the increasing risk trajectory had experienced a heart attack.
|
|
|
DISCUSSION
The latent-class approach applied to this problem reveals several sources of heterogeneity in the SES–health relationship. First, childhood hardship is constituted by multiple factors including illness, parental SES, poverty, family distress, and unmeasured heterogeneity related to inherited, in utero, and other environmental conditions (both physical and social) to which children are exposed. Hardship not only reflects the material conditions of life, but also its social conditions. Early disadvantage and childhood illness have severe enduring effects and predict high risks for heart disease. Second, heart attack risk is also heterogeneous. High risks and increasing risks for heart attack reflect different health trajectories with variable latent and pathway effects. Early hardship positively affects placement in the high risk for heart attack category and negatively affects placement in the increasing risk category. Indeed, early disadvantage and poor child health are major factors that increase the relative probabilities of being in the high heart attack risk group. A clearly discernible pattern of cumulative adversity emerges.
Our models were specified in terms of cumulative risk, beginning with childhood adversity and continuing with the introduction of no high school diploma (controlling for years of schooling), adverse employment experiences, low household income (controlling for household income), ever divorced, and health risks. This approach revealed complex cumulative processes. Strong effects on outcomes often occurred at the margins of covariate distributions (e.g., schooling and household income). And the effects of education were not symmetrical on heart attack risk trajectories: Education amplified the effects of childhood adversity on the high risk trajectory and attenuated its effects on other trajectories.
However, the issue of omitted variables is still a working concern. The omission of other variables may not improve predictability, but would improve the specification of diverse pathways. For example, we know that comorbidity and other associative causes of fatal disease are omitted from our models. But their interactions with early hardship and pathway variables are worthy of study in their own right. So too are psychosocial variables that can contribute to our understanding of how material conditions operate through perceptual, cognitive, and relational patterns in aging populations.
Conclusion
Latent and pathway factors interact and accumulate to set individuals on diverse trajectories of heart attack risk. They reveal considerable heterogeneity, which is likely to be evident in the study of other health outcomes. Our efforts to account for problems of unmeasured heterogeneity with latent-class models improve the state of our knowledge.
References
This article has been cited by other articles:
 |
|
 |
|
  M. McEniry, A. Palloni, A. L. Davila, and A. G. Gurucharri Early Life Exposure to Poor Nutrition and Infectious Diseases and Its Effects on the Health of Older Puerto Rican Adults J. Gerontol. B. Psychol. Sci. Soc. Sci., November 1, 2008; 63(6): S337 - S348. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
Z. Zhang, D. Gu, and M. D. Hayward Early Life Influences on Cognitive Impairment Among Oldest Old Chinese J. Gerontol. B. Psychol. Sci. Soc. Sci., January 1, 2008; 63(1): S25 - S33. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 N. E. Schoenberg, D. Shenk, and C. S. Kart Food for Thought: Nourishing the Publication of Qualitative Research Journal of Applied Gerontology, February 1, 2007; 26(1): 4 - 16. [Abstract] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
