| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||
RESEARCH ARTICLE |
1 Department of Gerontology, University of Nebraska at Omaha.
2 Research Scientist, MEDTAP Institute at UBC, Bethesda, Maryland.
3 Department of Sociology, Case Western Reserve University, Cleveland, Ohio.
4 Gerontology Program, University of Maryland Baltimore County, Baltimore.
Address correspondence to Karl Kosloski, PhD, Department of Gerontology, University of Nebraska at Omaha, University of Nebraska–Omaha, Omaha, NE 68182. E-mail: kkoslosk{at}mail.unomaha.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
The RELATIONSHIP between self-assessed overall health (typically measured by ratings of excellent, very good, good, fair or poor) and depressive symptomatology is both well established and enigmatic. On the one hand, the existence of a correlation between measures of self-rated global health and depression has been repeatedly demonstrated in a wide variety of circumstances (Han, 2002
; Han, Small, & Haley, 2001; Lynch & George, 2002; Mills, 2001; Molarius & Janson, 2002; Mulsant, Ganguli, & Seaburg, 1997). On the other hand, the mechanism that underlies this correlation remains unclear. One possibility is that poor self-rated health or a perceived change in one's overall health leads to depression (Han et al.; Lynch & George; Mills). An alternative possibility is that a depressed state causes a change in the assessment of one's health (Han; Molarius & Janson; Mulsant et al.). A third, virtually unexamined, possibility is that both processes operate simultaneously in a reciprocal relationship that unfolds over time.
The large majority of studies that have examined the relationship between self-assessed overall health and depression have used cross-sectional data (Schnittker, 2003) that provide information about covariation but do not contain information about the temporal ordering among variables. Longitudinal data, with time-structured observations, provide information about stability and change in relationships and allow for more convincing pronouncements concerning causal effects (Finkel, 1995). Our goal in the present study is to evaluate possible reciprocal effects that may exist between self-assessed health and depressive symptoms by using time-structured longitudinal data in order to identify the relative reciprocal influences that may exist between these two variables over time.
Why Self-Assessed Global Health Might Affect Depression
One can derive the hypothesis that self-assessed overall health causally affects depressive symptoms by linking two distinct theoretical models. In the first model, by Liang (1986), overall physical health is viewed as a cascading process in which chronic disease affects functional disability, which in turn affects individuals' self-assessed ratings of their overall health. Accordingly, in this model Liang proposes a causal sequence among the distinct dimensions of physical health but does not discuss potential causal links to depression. The second model, by Lewinsohn, Hoberman, Teri, and Hautzinger (1985), is similar in that physical disease leads to functional impairment, but it differs in that functional impairment is thought to lead directly to depression as a consequence of declines in physical health that limit one's ability to engage in valued and rewarding activities. Thus, whereas the model by Liang links physical disease and functional disability to self-assessed health, the model by Lewinsohn and associates links physical disease and functional disability to depression. By integrating these two models, then, we can hypothesize that self-assessed health affects depression.
Extensions of Liang's basic model (Ferraro, Farmer, & Wybraniec, 1997; Ferarro & Farmer, 1999; Kempen, Miedema, van den Bos, & Ormel, 1998; Liang & Whitelaw, 1990; Stull, Kercher, & Kosloski, 1996) provide strong evidence for a multidimensional conception of physical health, and they highlight the need to distinguish conceptually and empirically between the separate constructs of physical disease, functional impairment, and overall self-assessments of physical health. Within this approach, self-assessed global health is more than just a summary variable that represents the effects of the other dimensions of physical health. It also contains unique information that is not captured by the other health dimensions (also see Idler & Benyamini, 1997; Idler & Kasl, 1995).
There is a substantial body of research supporting the model by Lewinsohn and colleagues that links chronic disease and functional limitations to depression (for reviews, see Zeiss, Lewinsohn, Rohde, & Seeley, 1996; also see Mills, 2001). As Zeiss and her colleagues pointed out, however, virtually all of these studies are cross-sectional. Furthermore, none of these studies examine the association between self-assessed overall health and depression, despite the potential central role of self-assessed health in linking other dimensions of physical health to depression. One of our goals in the present study is to use longitudinal data to provide an empirical test of the hypothesized direct effect of self-assessed global health on depression.
Why Depression Might Affect Self-Assessed Global Health
Researchers have hypothesized that mental health problems operate as stressors that can potentially exert negative effects on physical health. For example, Kiecolt-Glaser and her colleagues (Kiecolt-Glaser, McGuire, Robles, & Glaser, 2002) described different pathways by which emotions can affect morbidity, including dysregulation in the endocrine and immune systems. In this regard, longitudinal studies have linked depression to physical health outcomes such as heart disease (Glassman & Shapiro, 1998; Pratt, Ford, Crum, Armenian, Gallo, & Eaton, 1996), cancer (Penninx, Leveille, Ferrucci, van Eijk, & Guralnick, 1999), and muscle strength (Rantanen et al., 2000).
In contrast, studies linking depression or depressive symptomatology expressly to self-assessed global health, as opposed to other physical health dimensions, are less numerous. Some studies are based on cross-sectional analyses that do not allow for a strict empirical test of the causal order (e.g., Han et al., 2001; Molarius & Janson, 2000; Mulsant et al., 1997; Schnittker, 2003). At least two studies, however, have been prospective. One study, by Miller and colleagues (1996) found that self-assessed ratings of overall health among older adults improved following treatment for depression, even when no change was observed in objectively rated health conditions. In a more recent study, controlling for demographic characteristics and prior levels of chronic disease, disability, and other health indicators, Han (2002) found that depressive symptoms predicted change in self-assessed health.
Although these two longitudinal studies suggest that depressive symptoms exert negative effects on self-rated global health, they are based on data collected only at two points in time. Moreover, the studies attempt to correct neither for measurement errors in the observed variables nor for correlated errors in the observed and latent variables that are likely to occur across time (cf. Kessler & Greenberg, 1981). Failure to do so can lead to biased estimates of the causal effect of depression on self-assessed health. One of our goals in the present study is to explore the link between depression and self-assessed global health, controlling for these potentially confounding factors.
Why Depression and Self-Assessed Global Health May Have Reciprocal Effects
Wrosch, Schulz, and Heckhausen (2004) described a theory of reciprocal influences between physical illness and depression. Specifically, they hypothesized physical illness to affect depression through two pathways. First, physical illness can cause depression directly through neurochemical and neuroanatomical changes. Second, physical illness can cause functional disabilities that, in turn, lead to depression and other forms of psychological distress. The extent to which depression accrues by means of the second pathway is moderated by the success of various control strategies designed to facilitate goal attainment. Successful control strategies serve as protective factors that mitigate the potentially negative effects of health on depression (e.g., realizing an important life goal, or disengaging oneself from unattainable goals). The theory also includes a feedback loop that creates reciprocal pathways by which depression affects physical illness. For example, depression can produce deficits in cognitive and behavioral abilities or compromised biological functioning that contribute to further health declines. However, in spite of the hypothesized reciprocal pathways that allow for the joint influence of health on depression and depression on health, we are aware of no studies that attempt to model these reciprocal influences by using this model or any other. One of our goals in the present study is to evaluate the possibility of these reciprocal effects, using simultaneous equations.
Purpose of the Present Study
The foregoing theoretical models and a very limited number of empirical studies using longitudinal designs suggest that there are compelling reasons to believe that (a) self-assessed health affects depression, (b) depression affects health, or (c) a reciprocal relationship exists between self-assessed health and depression. Our goal in the present study is to test for these potential individual and reciprocal effects by using longitudinal data within an autoregressive cross-lagged modeling framework. One of the strengths of the present analysis is the use of a structural equation model that simultaneously tests for cross-lagged effects across five waves of data, controlling for (a) random measurement error in the two key constructs of global health and depression, and (b) autocorrelated errors in the indicators of the latent variables.
We also include covariates that have been shown to be related to both self-assessed global health and to depression in order to control for possible spurious or intervening effects. These covariates include chronic illness and functional disability (Meador & Blazer, 1998), education and income (Franks, Gold, & Fiscella, 2003; Kim, Wolde-Tsadik, & Reuben, 1997; Martikainen, Adda, Ferrie, Smith, & Marmot, 2003; Mitchell, Mathews, & Yesavage, 1993), gender (Blazer, 1993; Franks et al.), race (Ferraro et al.,1997; Miller, Campbell, Farran, Kaufman, & Davis, 1995), and age (Mirowsky & Ross, 1992).
|
METHODS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
) of the Health and Retirement Survey (HRS). The HRS is a nationally representative, biennial, panel survey of 12,652 individuals aged 51 to 61 years of age at Wave 1 and their spouses (Juster & Suzman, 1995), with initial interviews conducted in 1992. The RAND HRS was developed to be a user-friendly version of a subset of the HRS. In addition, RAND created composite variables from subsets of variables with consistent names and uniform coding that facilitate longitudinal analyses. In the present analyses we use data from primary respondents who provided complete data for all five waves from 1992 through 2000 (N = 7,475).
Measures
Depression
There were 8 items from the full 20-item version of the Center for Epidemiological Studies–Depression Scale (CES-D; Radloff, 1977) included in the RAND version of the HRS. Respondents were asked about negative affect ("I felt depressed," "I felt lonely," and "I felt sad"), positive affect ("I was happy" and "I enjoy life"), and somatic symptoms ("I felt everything I did was an effort," "My sleep was restless," and "I could not get going"). Respondents indicated whether each symptom was present or absent all or most of the time. This set of items has been used successfully in previous studies (Han, 2002; Soldo, Hurd, Rodgers, & Wallace, 1997) and results from past research indicated that using similarly truncated sets of CES-D items, in this format, detracts very little from the precision of the full scale (Kohout, Berkman, Evans, & Coroni-Huntley, 1993). The CES-D has a well-established second-order factor structure (e.g., see Hertzog, Van Alstine, Usala, Hultsch, & Dixon, 1990). Accordingly, we created three separate additive-composite indicators for the present study by summing the responses to the items comprising each dimension (i.e., three items for depressive affect, three items for somatic complaints, and two items for positive affect). Simulation studies suggest that, in situations in which items have a known second-order factor structure, combining indicators from primary factors results in less bias in structural parameters than using individual items, especially in cases in which items are coarsely categorized (e.g., dichotomous) and nonnormally distributed (Bandalos, 2002).
Self-assessed global health
Only a single indicator of self-assessed health was available. It was the respondent's "graded" assessment of his or her global health status (i.e., "Would you say your health is excellent, very good, good, fair, or poor?"), with higher scores reflecting better health. Because the reliability of this measure is unknown, we performed a sensitivity analysis (described in the Analysis section) to identify an appropriate level of reliability in order to correct this variable for measurement error.
Covariates
We assessed chronic illness by adding together the number of positive responses to the question, "Has a doctor ever told you that you had ..." for seven diseases: high blood pressure, diabetes, cancer, lung disease, heart disease, stroke, and arthritis. Past research has shown that a composite variable representing multiple chronic illnesses performs comparably to a set of dummy variables, but it is more parsimonious in complex models (cf. Ferraro et al., 1997).
We assessed functional disability by asking whether the respondent had difficulty on each of five activities of daily living (ADLs): bathing, eating, dressing, walking across a room, and getting in or out of bed. In the RAND version of the HRS data, a set of dummy variables was created in which item responses were coded 1 if the respondent reported any difficulty and 0 otherwise, and then formed into an additive composite scale (range: 0–5) with an estimated reliability of 
= 0.70.
We measured educational attainment on a 5-point ordinal scale ranging from 1 (high school dropout) to 5 (college graduate and above). We represented income on a 30-point ordinal scale ranging from 1 (no income) to 30 ($145,000 and above) in $5,000 increments. We coded gender as 0 for men and 1 for women. In the case of race, we coded Caucasians as 0 and African Americans as 1. We dropped other racial or ethnic categories because of small subsample sizes.
Means and standard deviations for all the variables are shown in Table 1. Of the individuals making up the sample, 85% were White and and 60% were female. The average household income was in the range of $50,000 to $55,000. The average age at Wave 1 was 54.9 years. We assessed physical health on three dimensions: level of functional impairment, number of chronic conditions, and self-assessed physical health; and physical health was quite good on every dimension assessed. For example, respondents reported virtually no functional impairment, and there was less than one chronic condition, on average, per respondent. Self-assessed health was also rated between "good" and "very good" at the outset of the study and remained in that range throughout. In addition, symptoms of depression were generally low and remained low.
|
; Kessler & Greenberg, 1981; Taris, 2000). We estimated model parameters by using maximum likelihood estimation. We used only Wave 1 values of the covariates in an attempt to reduce the complexity of the model. Because self-assessed health has only a single indicator, an important analytic issue was determining its reliability. The assumption that this single indicator can measure the latent variable of self-assessed health perfectly is dubious. Therefore, we estimated the single factor loading as a free parameter. The estimated standardized factor loading was 0.85, with very little variability across measurement waves. Next, we undertook a sensitivity analysis in which we varied the factor loading of the indicator over a range from 1.0 to 0.80 (i.e., reliability of.64). Factor loadings of 0.8 or higher, including modeling the factor loading as a free parameter, yielded a plausible and consistent set of findings, with the strongest cross-lagged effects of depressive symptoms and self-assessed health on each other occurring when we used a value of 0.8 for self-assessed health. As a conservative measure, however, we subsequently report our results by using a value of 1.0.
Overall, the proportion of missing data on individual items was generally quite small. When aggregated over all of the variables in the study, however, and compounded across five waves, 16% of the cases were missing at least some data. To evaluate the effect of missing data, we performed analyses with listwise deletion of missing data and then replicated this by using casewise (full information) maximum likelihood estimation of model parameters available in EQS 6 software (Bentler, 2001). Differences between the two analytic approaches were extremely small, and results are reported with listwise deletion.
|
RESULTS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
2(198) = 4433; p <.05]. More relevant were the subjective fit indices (Tucker-Lewis index =.90; comparative fit index or CFI =.95; standardized root mean residual or SRMR =.07; root mean square error of approximation or RMSEA =.05), which generally met the criteria suggested by Hu and Bentler (1999) for a well-fitting model. Releasing the equality constraints on the measurement models across waves produced only a small improvement in CFI goodness of fit of.01, suggesting that the null hypothesis of measurement invariance should not be rejected (Cheung & Rensvold, 2002). To simplify the presentation of results, we report the correlations among the exogenous covariates separately in Table 2. As the age of study participants increased, so did the number of chronic health conditions. In contrast, with increasing age, income and education tended to be lower. ADL impairment was positively associated with both African-American and female status. Not surprisingly, as ADL impairment increased, so did the number of chronic health conditions. ADL impairment was also associated with lower income and education. Although the correlations in Table 2 generally make intuitive sense, even trivial correlations attained statistical significance, given the large sample size.
|
|
|
An important consideration in a cross-lagged panel analysis is the length of the causal lags (see Taris, 2000, for a fuller discussion of this issue). In the models tested to this point, we have assumed that the length of time needed for health to affect depressive symptoms is the same as the time needed for depressive symptoms to affect health, and 2 years is an appropriate interval for both effects. Intuitively, one might expect a substantial lag to occur (e.g., of the order of 2 years) before depressive symptoms might affect one's assessed overall health, but that the effect of perceived health on depressive symptoms might be more immediate (e.g., contemporaneous). If so, the previous analysis would represent the fairest test of whether depressive symptoms have any nontrivial effect on self-assessed health, but it might not use the most appropriate lag period for self-assessed health on depressive symptomatology.
Indeed, the nontrivial contemporaneous correlations between the disturbances of global health and depressive symptoms suggest the possibility of some relationship between the two variables that is not yet modeled (e.g., a contemporaneous effect of global health on depression, or possibly even contemporaneous reciprocal effects between the two variables). Given that the contemporaneous reciprocal effects model subsumes the theoretically more compelling recursive model in which the causal order is from global health to depression only, the former (larger) model would be the one most appropriate to test initially. In theory, with five waves of data, such a model is estimable; in practice, however, it proves to be much harder. With the present data, even though the model converged, we obtained inadmissible (i.e., out-of-range) parameter estimates. A closer inspection indicated that the problem might stem from multicollinearity among some parameter estimates, which is a common problem in these types of autoregressive models (Burkholder & Harlow, 2003). In an effort to circumvent this problem, we performed follow-up analyses in which the contemporaneous effects for each latent variable on the other were estimated in two separate models.
In the first model, we estimated the contemporaneous effect of self-assessed health on depressive symptoms, along with the lagged effect of depressive symptoms on self-rated overall health and the other covariates. The structural parameters between the latent variables are shown in Figure 2. The model in Figure 2 is identical to the model in Figure 1, except that contemporaneous effects have been substituted for lagged effects of self-assessed health on depressive symptoms. As a result, the models in Figures 1 and 2 have identical degrees of freedom. The overall fit of this minimally constrained model (i.e., only factor loadings and stability coefficients are constrained to be equal) was acceptable (Tucker-Lewis index =.90; CFI =.95; SRMR =.08; RMSEA =.05). The standardized contemporaneous effects of self-assessed health on depressive symptoms shown in Figure 2 are slightly larger than the lagged effects shown in Figure 1. In addition, the correlations between disturbance terms of the latent variables are generally much smaller in Figure 2, suggesting that adding a contemporaneous effect of self-rated overall health on depressive symptomatology may be appropriate as well and, logically, may be more appropriate than a lagged effect. Empirically, both types of effects remain plausible.
|
Because five waves of data were available, with potential lagged effects of up to 8 years, we explored the possibility that the effect of depressive symptoms on self-rated overall health might take longer than 2 years. Specifically, we reestimated the full model in Figure 1 three times: with three cross-lags of 4 years each, two cross-lags of 6 years each, and a single cross-lag of 8 years. One of the three 4-year cross-lags attained statistical significance (B = –.028), as did one of the two 6-year lags (B = –.038). The 8-year lag was nonsignificant. Although the effects are small, there is some evidence that, if depressive symptoms affect self-assessed health, the lag may be longer than 2 years.
All the analyses thus far have examined the effect of depressive symptoms on a self-assessed global health variable from which chronic health conditions and functional disability have been partitioned. Self-assessed global health, however, has a substantial correlation with both the number of medical conditions and functional disability (see Table 3). To evaluate the possibility that depressive symptoms may exert some influence on the component of variance that self-assessed global health shares with these other dimensions of health, we dropped chronic medical conditions and functional disability as control variables and reestimated the model in Figure 1. In the resulting model, the stability coefficients representing the regressions of self-assessed health on itself at the previous wave increased slightly, as did the effect of self-assessed health on depression; however, the effect of depressive symptoms on self-assessed health remained negligible.
|
DISCUSSION |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
These findings have important implications for the conceptualization of self-assessed global health. Researchers from varied disciplines have been intrigued by the interesting properties that self-assessed health displays, including predicting long-term decline in functional ability (Idler & Kasl, 1995), health service utilization (Stull et al., 1996; Whitelaw & Liang, 1991), and even mortality (Idler & Benyamini, 1997). Indeed, self-assessed health predicts these outcomes better than ostensibly more objective indicators of health and physical functioning (Ferraro & Farmer, 1999). Not surprisingly, self-assessments of global health by patients have been advocated as providing clinically useful information (Kivinen, Halonen, Eronen, & Nissinen, 1998; Lansky, Butler, & Waller, 1992; Parkerson, Broadhead, & Tse, 1995). When used in this manner, the present findings indicate that clinicians can be confident that global pronouncements of overall health on the part of patients are not simply manifestations of depressed affect.
Although the present results indicate that depressive symptomatology has little, if any, effect on self-assessed health, self-assessed health clearly affects depressive symptomatology. Past research has found that chronic medical conditions (e.g., heart disease, cancer) and functional limitations (ADL and instrumental ADL) affect measures of depression (see Mills, 2001). The present findings add to the growing evidence that self-assessed overall health does as well. Thus, clinicians need to be aware of the potential affect of health problems on screening instruments designed to detect depression.
Given our findings that global self-assessed health has effects on depression, net of chronic disease and functional limitations, the present results also reaffirm findings from earlier studies indicating that health is a multidimensional construct. That is, physical health includes subjective assessments of overall health as well as other factors such as chronic disease states and functional disability (c.f. Liang, 1986; Liang & Whitelaw, 1990; Stull et al., 1996). Because these different dimensions of health have substantial correlations with self-assessed global health, self-assessed global health serves as a reasonable proxy for them. However, self-assessed global health would also appear to provide its own unique effects.
With respect to the conceptualization of health, some researchers have hypothesized that the factors that define health are themselves causally related in a cascading manner in which chronic illness affects functional ability and both, in turn, affect ratings of self-assessed health (e.g., Ferraro et al., 1997; Liang, 1986; Liang & Whitelaw, 1990; Stull et al., 1996). From this perspective, one possibility is that depression, as a chronic stressor, affects aspects of health that occur earlier in the causal cascade and that any effect of depression on self-assessed health occurs indirectly through intervening variables such as chronic disease or functional disability. To test this possibility, we dropped chronic disease and functional disability from the model as control variables, and we reestimated the model. When the controls were dropped, any indirect effect through these variables should have been transmitted as a direct effect of depression on self-assessed health. Interestingly, the effect of depressive symptoms on self-assessed health remained negligible, suggesting that we did not obscure the effect of depressive symptomatology on self-assessed global health by including controls for the potential intervening variables of chronic disease and functional disability. However, this conclusion might change if our measures of chronic disease and functional disability had greater variability.
Our conclusions also might change if self-assessed global health had greater variability. It is important to note that the individuals in the HRS sample are relatively young and healthy. The average global health rating at the start of the study is 3.6 on a scale of 1 to 5, that is, somewhere between "good" and "very good," and remained stable over the course of the study. Indeed, by the fifth wave (8 years later), the average health rating declined less than one third of a scale point. As a result, most of the variation in global health at time (t) is explained by global health at (t – 1), with very little variation in global health left for depressive symptoms to explain. In contrast, restricted variance should also reduce the estimated effects of self-assessed global health on depressive symptoms, and the current study finds a nontrivial effect. Nevertheless, a thorough test of the hypothesized relationships in this study requires, at a minimum, that there be reasonable variation in self-assessed global health, as well as other health variables. A sample made up of older individuals, with greater variation on all health measures, would provide a more rigorous test of the manner in which depression may affect physical health, and this remains an avenue for future research.
Our conclusions might also be affected by the manner in which depression was measured. We assessed depressive symptomatology by using a subset of eight items from the CES-D. One possible reason that depressive symptoms had no effect on self-rated overall health is that these items are dominated by milder and more transitory states of depressed affect that rarely rise to the level of clinical depression or even to a level that could be characterized as chronic psychological stress.
With these caveats in mind, we find that the present findings suggest a small, but meaningful, effect of self-assessed health on depressive symptoms over time. There was no meaningful effect of depressive symptoms on self-assessed health. Thus, with respect to theories that hypothesize that depression causes declines in self-assessed global health, our findings suggest that researchers using cross-sectional data must be cautious, and models hypothesizing such a causal link should be considered carefully. Although there is a substantial correlation between self-assessed health and depression, it appears to be due to a unidirectional effect that flows primarily from self-assessed health to depression.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
Received for publication March 16, 2004. Accepted for publication May 13, 2005.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
| ||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
.05.