| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||
RESEARCH ARTICLE |
1 Department of Psychology, University of Southern California, Los Angeles.
2 The Karolinska Institute, Stockholm, Sweden.
Address correspondence to Amy Fiske, University of Southern California, Department of Psychology, Los Angeles, CA 90089-1061. E-mail: fiske{at}usc.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
DEPRESSIVE symptoms, which are more common in late life than in middle age (Kessler, Foster, Webster, & House, 1992
), are associated with serious negative outcomes such as increased risk of a depressive disorder (Beekman et al., 1997) and increased morbidity and mortality (Bruce, Seeman, Merrill, & Blazer, 1994). Thus, it is important for us tounderstand the association between aging and depressive symptoms. Investigating this critical issue can be difficult, however, because of the need for large, population-based, longitudinalsamples covering a wide range of ages. The current study represents an unusual opportunity to examine depressive symptoms and aging in such a sample.
In cross-sectional studies, levels of depressive symptoms are highest in the youngest and oldest adults, yielding a U-shaped curve (Kessler et al., 1992). The higher levels of symptoms in older compared with middle-aged adults thus represent the right-hand portion of that curve.
We need longitudinal data, however, to discern whether depressive symptoms actually change with age or whether differences between midlife and old age represent a cohort effect or some sort of selection effect. We found only two longitudinal studies of depressive symptoms and aging that included any middle-aged participants. Rothermund and Brandtstädter (2003) found modest increases over 8 years in a sample of persons aged 54–77, and Barefoot and colleagues (Barefoot, Mortensen, Helms, Avlund, & Schroll, 2001) found increases only for men in their sample of men and women interviewed at ages 50, 60, and 80. Several longitudinal investigations have been conducted in samples limited to older adults. Most have not detected changes in depressive symptoms over time (Dent et al., 1999; Haynie, Berg, Johansson, Gatz, & Zarit, 2001; Henderson et al., 1997). However, Wallace and O'Hara (1992) found increases in depressive symptoms over 6 years of follow-up. Thus, existing longitudinal studies provide limited support for the increase in depressive symptoms with age that is suggested by cross-sectional studies. Studies with wider age ranges and longer follow-up periods are needed.
We may ask whether increases in depressive symptoms in old age may be accounted for by the inclusion of somatic symptoms on depressive symptom checklists, because many somatic symptoms of depression are also commonly found in nondepressed older individuals who are physically ill. The answer may depend on the measure used. Compared with younger adults, older adults endorsed disproportionately more somatic than nonsomatic items on several instruments (e.g., Bolla-Wilson & Bleecker, 1989) but not on the Center for Epidemiologic Studies–Depression (CES-D) scale (e.g., Kessler et al., 1992).
Beyond characterizing age-related trends in depressive symptoms, many authors have examined covariates to test the extent to which age-related patterns might be explained by other variables. Health status consistently predicts depressive symptoms in cross-sectional studies of older adults (e.g., Kraaij, Arensman, & Spinhoven, 2002). Longitudinal investigations, in contrast, have yielded mixed results. Although numerous investigators report that health at one time point predicts the emergence of a clinically significant level of depressive symptoms at the next (e.g., Geerlings, Beekman, Deeg, & Van Tilburg, 2000; Prince, Harwood, Thomas, & Mann, 1998; Wallace & O'Hara, 1992), studies controlling not only for baseline depressive symptoms but also for baseline health status have not found additional longitudinal effects of change in health on depressive symptoms (Dent et al., 1999; Henderson et al., 1997). Therefore, although there is much evidence that health status is related to depressive symptoms in late life, it is not clear whether change in health is associated with change in depressive symptoms.
The relationship between health status and depressive symptoms in late life may be reciprocal (e.g., Meeks, Murrell, & Mehl, 2000). Recent research in mixed age samples shows thatboth depressive disorders and depressive symptoms predict later development of specific medical conditions, particularly cardiovascular disease (reviewed by Musselman, Evans, & Nemeroff, 1998).
Another focus in the study of late life depressive symptoms has been the role of negative life events. Prior research has yielded mixed findings. A meta-analysis of cross-sectional research concluded that the total number of negative life events is correlated with depressive symptoms in older adults, although the effect is modest (r =.15; Kraaij et al., 2002). Notably, however, several large, population-based studies in older adults have found no association (e.g., Beekman et al., 1995). In a prospective study, Glass, Kasl, and Berkman (1997) found that life events predicted depressive symptoms. Life events may have a time-limited effect, lasting less than 6 months (Norris & Murrell, 1987).
The relationship between life events and depressive symptoms may also be reciprocal. Several researchers have found, primarily among younger adults, that depressive status predicts future life events (e.g., Daley et al., 1997), which is consistent with Hammen's (1991) stress-generation hypothesis. We found no studies that examined this question specifically in older adults.
Thus, existing empirical literature suggests that depressive symptoms increase with age. There is ample evidence that health status is associated with depressive symptoms in late life, but longitudinal evidence that declining health explains increasing depressive symptoms is thus far lacking. Prior research also supports the conclusion that life events may lead to depressive symptoms among older adults, although the effect may be weak and time limited.
The present study tested several hypotheses regarding age and depression. Our first hypothesis was that depressive symptoms would increase between midlife and old age, longitudinally as well as cross-sectionally, and that the increase would not be fully explained by comorbid physical illness. We further hypothesized that both declining health and non-health-related negative life events would explain longitudinal increases in depressive symptoms.
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
), which is a subset of the population-based Swedish Twin Registry (Lichtenstein et al., 2002). The base sample for SATSA was composed of all pairs of twins in the registry who were reared apart and matched twins who were reared together. The 2,845 surviving individuals from these pairs were contacted in 1984 by means of a mailed questionnaire. Responses were received from 2,018 individuals (71% response rate). Three additional questionnaires were mailed at 3-year intervals to all surviving members who had not asked to discontinue participation in the study. The latter three questionnaires were used in the current study (referred to as Times 1–3). Responses were received from 1,637 individuals at Time 1, 1,496 individuals at Time 2, and 1,450 individuals at Time 3. The proportion of surviving, prior-wave respondents who participated at each time point was high (81% at Time 1, 87% at Time 2, and 89% at Time 3). Additional responses were received at each wave from individuals who had not participated in prior waves. Among those who responded at Time 1, 12% had died by the end of the study, a further 13% dropped out of the study, and 5% skipped Time 2 but resumed at Time 3; thus, 70% of the sample completed questionnaires at all three times (1,148). Of these respondents, 89% completed the CES-D portion of the questionnaire, yielding a sample of 1,017 individuals for longitudinal analyses. Sample sizes for individual analyses were reduced somewhat by missing data on covariates.
Although data for the current study were gathered from twins, pairs were randomly divided into two samples of unrelated individuals to satisfy the assumption of independent observations for statistical purposes. Sample 1 is the primary sample for analysis. Sample 2 serves as a nonindependent replication. Results are given for Sample 1. Except where noted, Sample 2 results were similar. For some analyses, the sample was divided into middle-aged (<60 years) and older adults (
60 years).
The mean age of the participants was 60 years (range = 29–93). Most were women (58%) and married (74%). Educational attainment was low, reflecting the socioeconomic situation in Sweden experienced by this cohort (58% completed or attended elementary school; 27% went to O-level, trade, or folk school; 8% attended academic high school; and 7% attended a university). All participants were Caucasian.
By design, half the twins in SATSA were separated early in life. To determine whether rearing status might bias results, we performed an analysis of variance. We found few differences. Twins reared apart had slightly higher illness scores at one time of measurement, F(1,573) = 4.74 and p <.05, but these findings were not replicated in the second sample. Twins reared apart did not differ from twins reared together on CES-D scores, age, or gender.
Instruments
CES-D
The CES-D is a widely used self-report measure of depressive symptoms (Radloff, 1977). Respondents rate how often they experienced each of 20 symptoms within the past week on a 0–3 scale, with four items that are reverse scored. Reliability and validity of the scale have been demonstrated in mixed age and older adult samples (e.g., Beekman et al., 1997). Items were translated into Swedish for the SATSA project, and they were backtranslated by a different individual to ensure the original meanings were retained. The CES-D scale has demonstrated measurement invariance across older and younger participants in the SATSA data set (Gatz, Johansson, Pedersen, Berg, & Reynolds, 1993). It is highly internally consistent; Cronbach's alpha was.89 at each wave.
Health
We measured health status in two ways: current illnesses and new illnesses. Participants were asked whether they have or have had each of 38 medical conditions. Symptoms of angina, myocardial infarct, claudication, chronic bronchitis, and asthma were also assessed, and diagnostic algorithms were applied to supplement the list of medical conditions endorsed. The current illness scale was a weighted sum of illnesses endorsed (or inferred by diagnostic algorithm). The new illness scale was a subset of the current illness measure, including only illnesses that had not been reported at a prior wave (including the 1984 questionnaire not otherwise used in the current study). Weighting was based on ratings by a panel of seven physicians of likely disability that would result from each condition. Agreement among raters was reasonable, with at least five of the seven doctors agreeing on the majority of items. If one or more items were missing, the entire scale wasconsidered missing, resulting in missing scores for 7% of individuals who completed the CES-D scale. Older adults more frequently reported most cardiovascular disorders, metabolic disorders, ulcers, urinary tract disorders, and shingles, whereas middle-aged adults more frequently reported migraines.
Life events
Life events were measured at each wave with a version of the Holmes and Rahe Social Readjustment Rating Scale (Holmes & Rahe, 1967) that was modified for use with older adults (Persson, 1980). For each of 14 events, participants rated whether it had occurred and, if it had, whether it was of little, some, or great importance. The present study included only negative items, non-health-related events (to create a scale complementary to the illness scales), and items that did not primarily reflect mental illness in the participant (to avoid confounding life events and depressive symptom scales). The scale only included events not reported at the preceding wave (or any prior wave for death of a child, death of a spouse, and placement of spouse in an institution). Responses to divorce and widowhood items were augmented with demographic information. The total scale score is the sum of item scores, weighted by mean importance ratings across the entire sample. Missing items were imputed by using the sample item mean. If more than two items were missing, a total scale score was not computed. The most frequent event reported for both age groups was deterioration in financial status. Most events that differed in frequency were reported more often in older participants (including deterioration in financial status, death or illness in relatives or friends, and caregiving), supporting the intent of the scale, which was designed for use with older adults. Only divorce was more frequent in middle-aged adults.
Analysis
We tested our first hypothesis, that depressive symptoms would increase with age, by using both cross-sectional (linear regression) and longitudinal (repeated measures analysis of variance) analyses. Because we anticipated the greatest age differences in the oldest ages, we tested for quadratic (age2) as well as linear effects. To determine whether increases in depressive symptoms could be an artifact of counting symptoms of physical illness, we tested specifically for changes in three of the four CES-D subscales validated in previous work (e.g., Gatz, Johansson, Pedersen, Berg, & Reynolds, 1993; Hertzog, Van Alstine, Usala, Hultsch, & Dixon, 1990). The psychomotor retardation subscale includes lack of motivation and somatic symptoms of depression. The depressed mood subscale includes affective symptoms. The lack of well-being subscale is a reverse-scored measure of positive affect. A fourth subscale, interpersonal difficulties, was not used because of its weak association with the overall scale (Hertzog et al., 1990). We tested for Age x Gender interactions in both cross-sectional and longitudinal analyses.
Before testing our hypothesis that declining health and other negative life events would each predict change in depressive symptoms over time, we first examined the combined and independent effects of current illnesses and negative life events on depressive symptoms by using a series of multivariate linear regression models. We then tested the hypothesis by using structural equation modeling (with Mx structural equation modeling software; Neale, 1999). For each variable, we first tested whether the longitudinal data were best represented by a simplex pattern (specifying a transmission path, which allows for stronger correlations between adjacent compared with distal time points), a common factor model (correlations between time points that do not follow the simplex pattern), or a model assuming no correlation between time points. A nonsignificant 
2 was taken to indicate absolute model fit, whereas the Akaike's information criterion (AIC) statistic was used to determine the best model on the basis of both fit and parsimony. We then tested a series of cross-lagged longitudinal models. Analyses other than structural equation modeling used SAS software (Version 8.02, SAS, Cary, NC).
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
|
|
Longitudinally, there was a significant interaction between age and time (Wilks's lambda = 3.99, df = 2, and p =.02). Therefore, we performed analyses separately by age group. CES-D scores increased in the older group, F(1,270) = 10.24 and p <.01, but not in the middle-aged group, which is consistent with the right-hand portion of the U-shaped curve reported in earlier studies. The mean CES-D score among older participants increased 1.6 points over 6 years. There were no interactions between time and gender, indicating that rates of change in depressive symptoms over time did not differ significantly between men and women.
We next considered age changes in subscales as a way of testing whether increases in CES-D scores could be explained by physical illness. As would be expected, the psychomotor retardation subscale, which includes symptoms that could reflect physical illness, was higher in older respondents in cross-sectional analyses adjusting for sex. That is, Time 1, F(2,616) = 21.77 and p <.0001, Time 2, F(2,608) = 13.67 and p =.0002, and Time 3, F(2,633) = 46.13 and p <.0001, with significant quadratic effects at Time 1, F(3,615) = 4.11 and p =.04, and Time 2, F(3,607) = 10.38 and p =.001. Longitudinal increases were significant only among older participants at F(1,252) = 11.95 and p <.001. In Sample 2, longitudinal increases were significant in both age groups.
However, depressed mood subscale scores also increased with age. That is, Time 1, F(2,625) = 12.67 and p =.0004; Time 2, F(2,614) = 9.28 and p =.002; and Time 3, F(2,639) = 33.23 and p <.0001. There was a significant quadratic effect at Time 2, with F(3,613) = 12.67 and p =.0004. Longitudinal analyses indicated significant increases in the older group only, F(1,260) = 13.43 and p <.001, with no gender interaction. In Sample 2, depressed mood scores increased significantly with time in middle-aged participants and there was a nonsignificant trend in the older group (results not shown).
Lack of well-being showed no consistent age differences in cross-sectional or longitudinal analyses and no gender differences.
Are Illnesses or Life Events Associated With Depressive Symptoms?
Because depressive symptoms increased with age, we next tested the extent to which current illnesses and non-health-related negative life events could explain the increases. In hierarchical linear regression, both current illnesses and non-health-related events were significant independent predictors of CES-D score at each wave, when sex was controlled for (Table 3 shows Wave 1). The additional variance explained by life events, beyond that predicted by current illnesses, provides further evidence that age differences are not solely due to comorbid physical illness. As seen in Table 3 (Eq. 5), age remained a significant predictor of CES-D score after covariates were adjusted for. Thus, age differences are greatly but not completely explained by a combination of current health status and negative life events.
|
2(1) = 0.17, ns, and AIC = -1.83; common factor, 2(2) = 0.81, ns, and AIC = -3.19. In Sample 2, however, the common factor model provided a less good fit: 2(2) = 5.04, p =.08, and AIC = 1.04. Thus, the simplex model was chosen for CES-D data in the multivariate model. The common factor model fit the newillness data, 2(2) = 2.81, ns, whereas the simplex did not, 2(2) = 4.10 and p =.04; the same pattern was seen in Sample2.Neither model fit the life event data. Therefore, a multivariate cross-lagged model was fit that combined the simplex model for CES-D data and the common factor model for new illness, with life events specified as unrelated influences ateach time point.
Estimated path coefficients for the full model (Table 4, M0) are included in Figure 1. A series of models was fit to test for significance of directional paths by examining loss of fit (
2) when sets of paths were dropped from the model (Table 4). InTwin 1, new illnesses did not predict depressive symptoms atthe next time of measurement (M2 - M0), and depressive symptoms did not predict new illnesses (M1 - M0). The pattern of path coefficients indicates that (a) depressive symptoms are highly stable across time points, with few new influences beyond Time 1; (b) new illnesses also show stability, with a common process influencing occurrence of new illnesses; and (c) the only indication of an association between illness and CES-D score was at the first time point. In separate models in middle-aged and older participants, as in the full sample, new illnesses did not predict depressive symptoms and depressive symptoms did not predict new illnesses. In Sample 2, in contrast, new illnesses did significantly predict depressive symptom scores and vice versa. Notably, as in Sample 1, the strongest association between new illnesses in the past 3 years and CES-D score occurred at the first time point, suggesting that these findings reflect, to some extent, the baseline correlation between the measures rather than a truly prospective effect. This interpretation is consistent with results we obtained in additional cross-lagged longitudinal models we fit testing the relationship between current illnesses and CES-D score (results not shown). In those models, the relationship between current illness and CES-D score was fully explained by baseline correlation in both Samples 1 and 2, with no significant paths from illness to later CES-D score or from CES-D score to later illness.
|
|
To aid in interpreting the finding that illness was only an inconsistent predictor of depressive symptoms, we tested post hoc whether illness would predict "incidence" or "persistence" of clinically significant depressive symptoms. Incidence, which was examined in participants with baseline CES-D scores below 16, was defined as an increase of at least 5 points in CES-D score to a level of 16 or greater at either Time 2 or Time 3. Persistence, examined in participants with baseline CES-D scores of 16 or greater, was defined as a CES-D score of 16 or greater at Times 2 and 3. In logistic regression analyses, baseline current illness scores predicted incidence, that is, parameter estimate =.07 (SE =.03), LR (1, n = 354) = 4.05, p =.04, odds ratio (OR) = 1.07, and 95% confidence interval (CI) = (1.00, 1.14), but did not significantly predict persistence, that is, parameter estimate =.06 (SE =.06), LR (1, n = 76) = 1.08, p =.31, OR = 1.06, and 95% CI = (0.94, 1.20). In both analyses, effects were small.
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
The longitudinal increase in CES-D scores in the present sample demonstrates that cross-sectional results in other studies (e.g., Kessler et al., 1992) cannot be explained fully by cohort effects; rather, they represent within-person change over time. Results extend the findings of Wallace and O'Hara (1992) and Rothermund and Brandtstädter (2003), who used more limited age ranges. The current findings support the right-hand portion of the U-shaped curvilinear relationship between CES-D score and age found in other studies (e.g., Kessler et al., 1992), as longitudinal increases were found only in older participants. Age-related increases seen in the current study have not been consistently detected in the few other longitudinal studies, perhaps because effects are modest and may not be reliably detected with smaller samples or shorter follow-up periods. It should be noted that the magnitude of increase in depressive symptoms does not imply clinically significant increases in depression among most older adults. Rather, the mean increase of 1.6 points over 6 years is equivalent to reporting a slightly increased frequency of one or two symptoms.
Although some have suggested that increasing depressive symptoms with age may largely reflect comorbid physical illness (Bolla-Wilson & Bleeker, 1989), this is not true in the current study. Age effects remained in the cross-sectional analyses after illnesses were controlled for. Furthermore, age differences were found on both mood and psychomotor retardation subscales.
Despite a growing literature suggesting that the gender gap in levels of depressive symptoms may narrow with age (e.g., Barefoot et al., 2001), neither our cross-sectional nor our longitudinal results supported this conclusion. Our findings are consistent with others (e.g., Kessler et al., 1992) who have found higher levels of depressive symptoms among women than men throughout late life.
The use of both cross-sectional and longitudinal analyses enabled us to gain a more complete picture of the relationship between health status and depressive symptoms than has been possible in studies limited to cross-sectional samples. Consistent with existing literature, current health status was significantly associated with depressive symptoms in cross-sectional analyses. Longitudinal results, however, provided only inconsistent support for the prediction that increases in depressive symptoms would be predicted by health events, as effects were small and not significant in both samples. A consistent finding, however, is that depressive symptoms were highly stable and the relationship between depressive symptoms and health was evident at the outset. These analyses and results are similar to those in the report by Meeks and colleagues (2000). Like Geerlings and colleagues (2000), we found that baseline health predicted incidence of clinically significant depressive symptoms. Unlike Geerlings and colleagues, we found that baseline illness did not predict persistent depressive symptoms. However, Geerlings and associates noted that health status did not predict a course marked by frequent fluctuations within a 3-year interval. These results raise the question of whether short-term fluctuation about a stable CES-D score could have occurred undetected in our study with its 3-year measurement intervals. Then, lack of consistent support for the hypothesis that change in illnesses would predict change in depressive symptoms could stem largely from high levels of measured stability of depressive symptoms.
The stability of depressive symptoms was particularly impressive when we consider the nature of the scale, which assesses for symptoms over the past week. Although depressive symptoms are thought to reflect a state, this study and others (e.g., Rothermund & Brandtstädter, 2003) suggest they may indicate an underlying trait. The stability may also reflect the continuity of depressive symptoms commonly found prodromal to, or persistent following, a depressive episode (Fava, 1999).
A notable finding of the current study is the lack of interaction between age and health, indicating that health is not more important in predicting depressive symptoms in late life than it is earlier in life. Although health status has often been cited as a particularly prominent risk factor for depressive symptoms in old age, current results indicate that health status assumes a larger role with age primarily as a result of the increasing frequency of health problems.
The current results offer new insights into the relationship between negative life events and depressive symptoms. Longitudinally, direct effects of life events did not carry over beyond the 3-year measurement interval, suggesting recovery and supporting the findings of previous research that indicates short-lived effects for life events (e.g., Norris & Murrell, 1987). There may be indirect effects, however, because change in depressive symptoms also predicted later life events and thus a recursive cycle may develop in which negative events and depressive states become self-perpetuating.
The finding that depressive symptoms predicted later life events is intriguing and has not, to our knowledge, been previously reported specifically in an older adult sample. Brostedt and Pedersen (2003) found that depressive disorders were associated with later stressful life events in a mixed age sample. In addition, a handful of investigators have found that depressive diagnosis or depressive symptoms predict later stressful events in younger samples (e.g., Daley et al., 1997). The current results support Hammen's (1991) stress-generation model, which predicts that some of the stressful events that befall depressed individuals—in particular, events associated with interpersonal conflict—are directly attributable to their own characteristics and behaviors and the contexts in which they live. Poor social functioning, observed among individuals with elevated depressive symptoms (see Fava, 1999), may partly explain this effect.
Limitations of the Study
The original scale on which our life event measure was based, the Social Readjustment Rating Scale (SRRS), has been the subject of criticism (e.g., Chiriboga, 1989), although adjustments for the present study mitigate concerns. The SRRS remains a widely used self-report checklist of stressful events, however, and has repeatedly demonstrated utility in predicting distress (Scully, Tosi, & Banning, 2000).
Because life events were reported retrospectively, it is possible that current depressive symptoms may have biased the reporting of past events. Nonetheless, the finding that depressive symptoms affect future life events was based on prospective data.
As in any longitudinal study, selective attrition is a concern. Attrition analyses tested whether death or dropout from the study was predicted by CES-D score. With a link to the Swedish Death Register, we determined that death accounted for nearly half of study attrition. Individuals who died or dropped out had higher baseline CES-D scores than those who completed all three times. Attrition could not be explained by depressive symptom levels, however, because it was better predicted by advanced age. Other investigators have concluded that such attrition may affect descriptive results but have little impact on findings regarding association between variables (e.g., Kempen & van Sonderen, 2002). In the current study, the main result of such a pattern of attrition would be to underestimate age-related increases in depressive symptoms.
Several findings differed between Samples 1 and 2. Longitudinal changes in psychomotor retardation and depressed mood symptoms showed different patterns of significance across samples. Because the magnitude of longitudinal change in CES-D scores was small (1.6 points over 6 years), it is not surprising that effects were not consistently significant. In the same way, new illnesses did not significantly predict depressive symptoms longitudinally in Sample 1, but they did in Sample 2. Differences may reflect the small magnitude of effects. It is notable that, longitudinally, the non-health-related life event score was a much stronger predictor of depressive symptoms than was the new illness score, even in Sample 2. Thus, lack of consistency in certain findings across samples highlights the small magnitude of these particular effects.
Some may question the use of twins. The fact that SATSA twins are similar in health and sociological characteristics to a cross section of the Swedish population of the same age (Pedersen et al., 2002) suggests that use of this data set to test non-twin research questions is appropriate. Further, this large, population-based, longitudinal data set offers advantages over other longitudinal studies published to date.
Conclusions
In summary, we conclude that depressive symptoms increase slightly with age and that the increase is not merely a cohort effect or an artifact of measuring symptoms of comorbid physical illness. Certain negative life events, which are reported more frequently in late life, explain some of the increases in depressive symptoms. Physical illnesses, which also become more common in late life, are correlated with depressive symptoms, but new illnesses only inconsistently predict change in depressive symptoms over time.
|
Acknowledgments |
|---|
The authors gratefully acknowledge the consultation provided by Chandra Reynolds.
Received for publication September 12, 2002. Accepted for publication July 18, 2003.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
| ||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
