| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
Correspondence: Address correspondence to Dr. Beth Han, 3311 Toledo Road, Room 3409, Hyattsville, MD 20782. E-mail: hih9{at}cdc.gov
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Key Words: Change in self-rated health • Mortality • Disabled older women • Time-dependent covariates
Recently, researchers have suggested that the failure to examine self-rated health as a dynamic evaluation may lead to underestimation of the true association between self-rated health and mortality (Fayers & Sprangers, 2002; Ferraro & Kelly-Moore, 2001). Self-rated health is a dynamic rather than static perception, which is related to an individual's changes in health status over time (Fayers & Sprangers; Ferraro & Kelly-Moore; Han, 2002; Strawbridge & Wallhagen, 1999). Thus, self-rated health is likely to change during a long follow-up period. One study demonstrated that baseline self-rated health was associated only with 4-year mortality and not with 9-year mortality in older women (Benyamini, Blumstein, Lusky, & Modan, 2003). This is consistent with the notion that the longer the period of follow-up, the less likely the health status measured at baseline would stay the same over time.
Very few studies, such as those by Ferraro and Kelly-Moore (2001) and Strawbridge and Wallhagen (1999), have considered self-rated health as a dynamic evaluation and have investigated its association with mortality. Strawbridge and Wallhagen used self-rated health as a time-dependent (change over time) covariate in a Cox regression model and found that change in self-rated health was predictive of mortality among women; in their study, time intervals between follow-up interviews were at least 9 years and participants were aged 21 or older at baseline. Ferraro and Kelly-Moore found that change in self-rated health was predictive of mortality among Black and White adults; in their study, time intervals between follow-up interviews ranged from 5 to 10 years and participants were aged 25 to 74 at baseline (National Center for Health Statistics, 1987). Ferraro and Kelly-Moore mentioned that people adjust their health ratings in the last few years of their lives to reflect declines in health status. The intervals of these two studies could have been too long to detect some significant change in self-rated health, particularly if it occurred a year or two prior to death.
Thus, the association between change in self-rated health and mortality reported by the two studies may still be underestimated. Moreover, other studies have found that poor self-rated health at baseline is not related to mortality among elderly women (Deeg & Kriegsman, 2003; Hays et al., 1996; van Doorn & Kasl, 1998). The purpose of our study was to assess whether change in self-rated health was predictive of mortality among disabled older women, particularly when all available transitions in self-rated health were considered during 6-month intervals between follow-up interviews. In addition, we investigated whether change in self-rated health was a stronger predictor of mortality among disabled older women than either self-rated health at baseline or at the most recent observation.
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
The measure of self-rated health in WHAS was a single question asked at each interview: "At the present time, would you say that your health is excellent, very good, good, fair, or poor?" Answers were coded as 1 through 5, respectively. We used the linear form of self-rated health to avoid the coarseness caused by collapsing the five responses into fewer categories and to keep Cox regression models (with time-dependent self-rated health) parsimonious. This approach is consistent with the recent research by Ferraro and Kelly-Moore (2001), which helps us to better understand the association between change in self-rated health and mortality. We measured the number of diseases by the total number of self-reported diseases at each interview ("whether a doctor told you that you had..."). We assessed the number of instrumental activity of daily living (IADL) difficulties by the number of items for which the participant reported having difficulty; the items included doing light housework, preparing meals, shopping for groceries, managing money, making phone calls, and taking medications. Difficulty in walking was evaluated with this question: "By yourself, that is, without help from another person or special equipment, do you have any difficulty in walking for a quarter of a mile, which is about 2 or 3 blocks?" (Answers were coded as 1 = no difficulty, 2 = a little difficulty, 3 = some difficulty, 4 = a lot of difficulty, and 5 = not able to walk one fourth of a mile.) Depressive symptoms were assessed by using the 30-item Geriatric Depression Scale (coded as 0–30), which is a reliable and valid measure of depression among older adults (Yesavage et al., 1982–1983). These health indicators were assessed at all seven interviews.
Sociodemographic factors included age, race (White vs Black), years of education completed, marital status (married vs other), and annual household income with imputation for those not reporting income (Simonsick, Guralnik, & Fried, 1999). Cognitive function was evaluated by using the Mini-Mental State Examination (MMSE) (Folstein, Folstein, & McHugh, 1975) at baseline. Participants who had a MMSE score of less than 18 were ineligible to participate in the WHAS. Years of smoking were assessed only at baseline.
The use of time-dependent covariates in Cox regression models offers us opportunities to better understand dynamic relationships between investigated variables and the outcome, taking into account all available changes in self-rated health and variations in covariates at each interview during the follow-up period (Allison, 1995). In this way we can consider baseline health status, the most recent health status, and all available transitions in health status from interview to interview among those who survive. When the values of some time-dependent covariates are missing at some interviews among some participants, the time-dependent covariate option allows us to skip these interviews and to examine available data at next interviews. We used –2 log-likelihood statistics (–2LL) and Martingale residual methods to determine how our Cox regression models fit our data during the model building.
First, we examined changes in self-rated health among disabled older women over time. Second, we conducted bivariate analyses to examine the associations between self-rated health (baseline self-rated health, self-rated health at the most recent observation, or time-dependent self-rated health) and mortality by using three Cox regression models separately (Models 1–3).
Third, we conducted multivariate analyses to investigate the associations between self-rated health (baseline self-rated health, self-rated health at the most recent observation, or time-dependent self-rated health) and mortality with an additional five Cox regression models. Models 4–6 tested the associations between self-rated health (self-rated health at baseline, self-rated health at the most recent observation, or time-dependent self-rated health) and mortality after adjusting for covariates assessed at baseline. Model 7 examined the association between self-rated health at the most recent observation and mortality after controlling for covariates examined only at baseline (years of education, marital status, race, income, cognitive function, and years of smoking) and covariates at the most recent observation (age, the number of diseases, the number of IADL difficulties, walking difficulty, and depressive symptoms). Model 8 investigated the association between the most recent self-rated health and mortality after controlling for the same baseline covariates as in Model 7 and time-dependent covariates (age, the number of diseases, the number of IADL difficulties, walking difficulty, and depressive symptoms). Model 9 examined the association between time-dependent self-rated health and mortality after adjusting for the same covariates as in Model 8. We performed analyses by using SAS statistical software (SAS Institute, 1996).
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Table 1 shows the results of Models 1–3 on the associations between self-rated health and mortality at the bivariate level. Relative hazard (RH), 95% confidence intervals (CIs), and p values are presented. Baseline self-rated health, self-rated health at the most recent observation, and time-dependent self-rated health are predictive of mortality. Compared with self-rated health at baseline and at the most recent observation, time-dependent self-rated health is the strongest predictor of mortality among disabled older women at the bivariate level. With each additional 1 unit of decline in self-rated health, a disabled older woman was 1.29 times more likely to die (RH = 1.29, 95% CI = 1.14–1.46, p =.0001). If a disabled older woman's self-rated health changed from excellent to poor (4 units of decline), she was 2.77 times (1.294 = 2.77) more likely to die.
|
Table 2 shows the results of Model 7 on the association between self-rated health at the most recent observation and mortality after we controlled for covariates assessed only at baseline and covariates at the most recent observation. Self-rated health at the most recent observation was no longer predictive of mortality among disabled older women. Higher mortality was associated being married, presenting a lower cognitive function at baseline, and having more years of smoking at baseline, and with having more walking difficulty at the most recent observation.
|
|
|
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
If self-rated health of disabled older women were a current evaluation of their health conditions, self-rated health at the most recent observation (prior to death or loss of follow-up) would have been a stronger predictor of mortality than both self-rated health at baseline and change in self-rated health. Our results do not support this assumption. The use of time-dependent self-rated health and other time-dependent covariates in Cox regression models takes into account both the most recent health status and transitions in health status over time. Consistent with the study by Strawbridge and Wallhagen (1999), our study found that change in self-rated health is predictive of mortality among older women. Ferraro and Kelly-Moore (2001) found that change in self-rated health is a stronger predictor of mortality than baseline self-rated health among Black and White adults. Our study provided further evidence that change in self-rated health is a stronger predictor of mortality than self-rated health at baseline and at the most recent observation among disabled older women. In addition to self-rated health at any point, change in self-rated health has significant value in predicting mortality. Our results confirm the clinical notion that, although what older women rate their current health (e.g., excellent or fair) is important, how they arrive at their current health state (e.g., decline from excellent to fair) is even more important. For example, older women with "fair" health are worse off if they are on a declining health trajectory (e.g., from excellent to fair) than if their "fair" health is stable (no change in self-rated fair health). This basic phenomenon can also be seen in several other variables in our study (e.g., the number of diseases, the number of IADL difficulties, and walking difficulty), showing that their associations with mortality are stronger with the use of time-dependent covariates than the most recent observation.
Some studies support the earlier research of Maddox and Douglass (1973), suggesting the stability of self-rated health in terms of insignificant mean change over time. Our results show that mean-level stability in self-rated health over time among disabled older women is a result of multidirectional changes. In our study, self-rated health of some participants improved, whereas for others it stayed the same or declined over the follow-up period. If self-rated health of disabled older women were a static perception of their health status, self-rated health at baseline should have had the same prognostic value on mortality as self-rated health at the most recent observation. Our results do not support this assumption either. We found that change in self-rated health is a stronger predictor of mortality than self-rated health at baseline and at the most recent observation. Moreover, consistent with previous studies (Deeg & Kriegsman, 2003; Hays et al., 1996; van Doorn & Kasl, 1998), in our study we did not find the association between baseline self-rated health and mortality among disabled older women. Disabled older women are conscious of changes in their health status and adjust their perception of their health accordingly.
It is not difficult to detect change in self-rated health over time. Family caregivers and health professionals should pay attention to not only self-rated health per se (e.g., excellent or fair; see Maddox, 1999) but also change in self-rated health over time (e.g., decline from excellent to fair) among community-dwelling disabled older women. Change in self-rated health can be a useful and simple screening tool for individuals with a high risk of mortality. A decline in self-rated health over time indicates that health status is deteriorating and mortality risk is increasing. Our study provides evidence that time intervals of 6 months between follow-up interviews allow us to effectively detect change in self-rated health among our participants. Therefore, at least every 6 months, family caregivers and health professionals should assess the self-rated health of community-dwelling older women who are moderately to severely disabled, and they should be cautious about their decline in self-rated health over time. If there is no apparent reason for the decline (e.g., no changes in walking difficulty or no changes in identified disease status), family caregivers should work with health professionals to find out why their loved ones rate their health worse than they did previously.
|
Footnotes |
|---|
1 National Center for Health Statistics, Hyattsville, MD.
2 National Institute on Aging, Bethesda, MD.
3 Department of Health Policy and Management, Johns Hopkins University, Baltimore, MD.
4 School of Public Health, University of Tampere, Finland.
Decision Editor: Linda S. Noelker, PhD
Received for publication May 3, 2004. Accepted for publication October 29, 2004.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  A. Bowling and E. Grundy Differentials in mortality up to 20 years after baseline interview among older people in East London and Essex Age Ageing, January 1, 2009; 38(1): 51 - 55. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
