Copyright 1992 by The Gemniologiail Society ofAmei
hmrnul of Gerontology: PSYCHOLOGICAL SCIENCES 1942. Vol. 47. No. 6, P373-P.184
Determinants of Change in Perceived Health in a Longitudinal Study of Older Adults Judith Rodin and Gail McAvay Psychology Department, Yale University.
perception of their health appears to be imporPEOPLE'S tant in predicting their subsequent mortality. This holds true even after controlling for the contribution of objective health status. Two studies of older people (Kaplan, Barell, & Lusky, 1988; Mossey & Shapiro, 1982) found that subjective health ratings predicted survival, independent of the effects of objective health, sex, health care utilization and, in the Kaplan etal. (1988) study, functional status. Idler, Kasl, and Lemke (1990) questioned the rigor of the objective health status measures used in the earlier studies and applied a variety of methodological and statistical controls to measures in their study. Self-reported health still predicted survival, independent of physical health status, disability, pain, medications, demographic factors, and health practices in an older age sample. Kaplan and Camacho (1983) also found strong support for the importance of self-rated health in predicting 9-year mortality in the Alameda County sample, which included adults of all ages. Their findings are impressive because they reported that perceived health was an independent predictor of mortality even when the regression equation included some measures of psychological functioning and social networks, as well as actual health, health practices, sex, and age. Most recently, Idler and Angel (1990) found that after controlling for the contribution of medical diagnoses derived by a medical examination, as well as demographic factors and health-related behaviors, self-rated health was significantly associated with mortality for middle-aged males in the sample, with a weaker trend for the older age group. Thus, the weight of the evidence suggests that many individuals, when asked how they feel, are evaluating something with important prognostic value regarding their probability of survival.
Such demonstrations lead quite naturally to asking what it is about perceived health that makes it so predictive, sometimes more so than presumably "harder" health-relevant data. Two bodies of literature relate to this question. The first study, much of which preceded the studies of perceived health and mortality, is used to validate perceived health as a measure of health when more "objective" indicators were unavailable. These correlational studies attempted to evaluate which classes of variables — health-relevant, demographic, mood state and general well being, and functional status — correlated with subjects' ratings of their own health. The second smaller and more recent literature is composed of studies attempting to determine which previously measured variables predict changes in perceived health over time. The first group of correlational studies shows that perceived health is clearly related to objective health status. Several find strong correlations between perceived health and long-standing chronic illness, especially in the elderly (Fillenbaum, 1979; Goldstein, Siegel, & Boyer, 1984; Jylha, Leskinen, Alanen, Leskinen, & Heikkinen, 1986; Liang, 1986; Linn & Linn, 1980; Tissue, 1972; Wan, 1976; Zonderman, 1986) and with other health indicators, e.g., number of medications, sick days, or hospitalizations (Fillenbaum, 1979; Linn & Linn, 1980; Wan, 1976). In the correlational studies, the relationship between perceived health and psychological indicators of well-being appears more inconsistent. In some work, poorer perceived health is significantly associated with greater neuroticism (Zonderman et al., 1986), hypochondriasis (Blazer & Houpt, 1979), depression (Blazer & Houpt, 1979; Goldstein & Hurwicz, 1989; Levkoff, Cleary, & Wetle, 1986), and lower life satisfaction (Blazer & Houpt, 1979; George & P373
Downloaded from http://geronj.oxfordjournals.org/ at New York University on June 2, 2015
To determine the factors that are predictive of a negative decline in perceived health in a longitudinal study of 251 men and women aged 62 and older, we developed a "synthetic" cohort of individuals who experienced a decline in perceived health between two time points and a control cohort of those who did not. The longitudinal design of the study made it possible to evaluate the effects of "baseline" predictors, gathered prior to the time interval during which the reported change in perceived health occurred, in addition to measures assessed at the two time points between which the change in health was reported. Hierarchical logistic regression models were used to examine the effect of demographic, psychosocial, and illness measures on change in perceived health. These analyses revealed that increases in new illnesses, increased physician visits, and worsening of preexisting conditions (in subjects who had higherfeelings of self-efficacy) were all associated with a decline in perceived health, after controlling for the effects of baseline illnesses and medication use. After accounting for these effects of changes in objective illness indicators, we found that changed psychosocial factors also predicted a decline in perceived health. Lower life satisfaction and higher depression at baseline were predictive of a later decline in perceived health. In addition, data collected at the third consecutive time point were evaluated to assess determinants of a sustained vs transient decline in perceived health. A sustained decline in perceived health was associated with increased depression and decreased self-efficacy.
P374
RODIN AND McAVAY
The focus of this study was on determining the relative importance of demographic, psychosocial, and health-status factors that are predictive of a decline in self-rated health. The longitudinal design of the study made it possible to evaluate the effects of "baseline" predictors gathered prior to the time interval during which, the reported change in
perceived health occurred, in addition to measures assessed at the two time periods between which the change in health was reported. METHODS
Subjects Subjects were 264 participants in a 3-year longitudinal study of psychosocial, dispositional, and biological factors related to changes in health. Participants were males and females aged 62 and older who lived in South Central Connecticut. All of the subjects lived in the community rather than in specialized care facilities, and all spoke English. Names were drawn from census tract data to sample from a broad cross-section of the community. Full recruitment procedures and response rates have been detailed previously (Rodin, McAvay, & Timko, 1988). Procedure The longitudinal design of this study included an initial, extensive baseline interview, followed by eight interviews spaced at varying time intervals over a 3-year period. The first four interviews were spaced at 2-month intervals; the next two took place one year after the fourth interview, and the last two took place one year after the sixth interview. Subjects who had complete baseline data on all variables relevant to the present questions (N = 251 or 95% of the entire sample) were included in the analyses. These subjects had completed an average of seven interviews following the baseline. Subjects were categorized into groups on the basis of their ratings of perceived health. Subjects rated their own health (on a 5-point scale: very good, good, fair, poor, and very poor) at each interview in response to the question, "How have you been feeling since I last talked to you?" To make the criteria for change more stringent and conceptually meaningful, these data were recoded into three categories: 1 = very good or good, 2 = fair, and 3 = poor or very poor. For example, using the 3-point scale, ratings that went from very good to good or poor to very poor were not considered a decline in perceived health. Changes in perceived health were then calculated for each of the seven possible consecutive time periods (i.e., between interviews 1-2, 2-3, 3-4, etc.). Subjects who had a negative change (decline in perceived health) between any two consecutive time points were included in Group 1. The first change in perceived health might have occurred from Interview 1 to Interview 2 for some cases, whereas for others it may have occurred from Interview 6 to Interview 7. Using this procedure, 145 subjects were categorized as experiencing a negative change in perceived health over the course of the study. The column in Table 1 for Group 1 indicates in which time period the first negative change in perceived health occurred for this sample. The availability of a third consecutive time point made it possible to examine the stability of this initial decline in perceived health and to assess determinants of the stability or instability of decline. Therefore, for some analyses, we divided the initial group of subjects who experienced a
Downloaded from http://geronj.oxfordjournals.org/ at New York University on June 2, 2015
Landerman, 1984). Other studies show no correlation with morale or other psychological indicators of well-being (Tissue, 1972; Wan, 1976). Self-perceptions of health also appear uncorrelated with social resource and network variables (Blazer & Houpt, 1979; Kutner, 1987). Self-assessed health does, on the other hand, correlate with functional capacity and instrumental activities (Liang, 1986; Linn & Linn, 1980). There are four reports in the second group of studies: those attempting to predict changes in perceived health over time. Supporting the earlier work of Maddox and Douglass (1973), most authors are impressed by the remarkable stability in perceived health in the overall samples studied. Three studies of older aged subjects — two prospective (Markides & Lee, 1990; Minkler & Langhauser, 1988) and one retrospective (Weinberger et al., 1986) — found that people whose self-rated health declined were older, had greater financial need, and had fewer social support resources five years earlier (Minkler & Langhauser, 1988), had lower levels of education and functional health eight years earlier (Markides & Lee, 1990), and had more negative life events one year earlier (Weinberger et al., 1986). Unlike Minkler and Langhauser (1988), Weinberger et al. found no direct or interactive (with changes in life events) effects of social support. Studying predictors of change over one year in a broader age range of subjects, Goldstein et al. found that changes in perceived health were not sensitive to short-term changes in objective health status (e.g., health beliefs, health services). Although these studies are quite informative, they lack measures of demographic factors, objective health status, psychological well-being, and social support together in any single study. Given the important relationship between many of these classes of variables and perceived health in the earlier, correlational studies, such an analysis is critical. Furthermore, none took multiple measures of perceived health over time. It is possible for subjects to change their ratings several times but appear quite stable when only two ratings are obtained. This longitudinal study attempted to address these issues by evaluating a full range of predictors in individuals who underwent a decline in perceived health, providing a more powerful examination of the determinants of negative changes in perceived health. The availability of eight time points with repeated measures of self-rated health allowed us to create "synthetic" groups (Campbell & Hudson, 1985) of individuals who experienced a sustained or nonsustained deterioration in perceived health between three consecutive time points. This type of comparison is powerful because each subject serves as his or her own control. In addition, however, we also developed a third synthetic cohort as a different type of control group. It was made up of individuals who did not experience a negative change in perceived health at any time point through the entire course of the study.
DETERMINANTS OF CHANGE IN PERCEIVED HEALTH
Health Measures
Table 1. Interviews Used in Statistical Analysis Group 1 Negative Change in Perceived Health
Group 2 No Change in Perceived Health
Interviews Used
N
%
N
%
1 and 2 2 and 3 3 and 4 4 and 5 5 and 6 6 and 7 7 and 8
38 25 31 16 18 11 6
26.2 17.2 21.4 11.0 12.4 7.6 4.1
28 18 22 12 13 4
26.4 17.0 20.8 11.3 12.3 8.5 3.8
145
100.0
106
100.0
Total
9
P375
Doctor visits. — At each visit following the background interview, subjects were questioned about any doctors they had seen since the previous interview. The type of physician (e.g., cardiologist, general practitioner, etc.), reason for the visit, and any treatments prescribed by the physician were recorded. A score reflecting the number of doctor visits since the previous interview was constructed for each subject. Change in the number of doctor visits reported at the two time periods studied was then calculated for each subject, such that a positive score represented an increase and a negative score a decrease in the number of physicians seen. Medications. — Use of medications prescribed by the subject's physician was assessed at the background and at each interview. The number of new medications prescribed plus the number of preexisting, episodic (taken on an asneeded basis) medications taken at each interview were totaled for each subject. Social Support Measures Emotional support availability. — The presence (coded as 1) of emotional support was measured at the background interview in response to the item, "Whom can you really count on to listen when you need to talk?" The absence of
Downloaded from http://geronj.oxfordjournals.org/ at New York University on June 2, 2015
negative change in perceived health over the first two time points into two groups based upon their perceived health rating at the third time point. Subjects who reported the same or a lower level of perceived health at the interview following the initial decline in perceived health were included in the negative change group (/V = 51), whereas subjects whose ratings went back up were included in the transient negative change group (N = 72). Subjects missing a third time point of data were excluded from these analyses. The 106 respondents who did not report a negative change in perceived health between any consecutive time points of the study were included in Group 2. The absence of negative change in this group cannot be attributed to a "floor" effect, because through the study, only four of these subjects rated their health at the worst level (poor or very poor). Three of these subjects reported improvement in perceived health at the following interview, whereas the other subject's health rating was poor at the first two interviews and showed improvement during the remainder of the study. For the second group of subjects, any of the two consecutive time points of data where their ratings of perceived health were stable were available for use. We selected the time points that would make the percentage of subjects with data from each of the seven possible time intervals the same in Groups 1 and 2. Subjects in Group 2 were assigned at random to a particular interval. Assigning the subjects in the second group to correspond to the percentages in the first group for each interval reduces the possibility of bias due to the use of different time intervals. The column for Group 2 in Table 1 presents the interviews used in the statistical analysis for this group. There was a somewhat higher percentage of respondents who did not have the last interview in Group 2 (N = 28,26%) than in Group 1 (N = 25,17%). Although the 28 subjects in the "no change group" who did not reach the final interview could possibly have declined in perceived health if they had been interviewed again, we decided to include them in the control group. We believe that this represents a conservative approach, because they might later have experienced a negative change that would have placed them in the other group. A possible misclassification into the control group reduced the possibility of finding differences between the two groups.
Medical conditions. — At the background interview, subjects were questioned about illnesses and medical conditions using a standardized form that included diseases from all of the major categories (Fetter, Shin, Freeman, Averill, & Thompson, 1980). At each interview following the background, subjects were asked about any new illnesses or accidents that had occurred since the previous interview, and the status of each of the medical conditions reported at the background was evaluated. Each condition was assigned a code number from the diagnosis categorization system, Diagnosis-Related Groups (Fetter et al.). There were two types of illness measures constructed from these data. First, changes in existing chronic conditions (i.e., conditions that were ongoing) and recurrences of preexisting episodic conditions (i.e., conditions that reappeared from time to time with no symptoms in the intervals between) were scored. For each subject, the number of chronic conditions that got worse and the number of episodic conditions that recurred were totaled to represent the number of preexisting medical conditions that worsened. A change score was calculated so that a positive number reflected a greater number of preexisting conditions that worsened at the second time point, compared with the first time point. Second, the number of new illnesses never reported before was calculated at each interview. A change score was calculated such that a positive number reflected the number of new illnesses reported at the second time point was greater than the number reported at the first time point. Based on the bivariate distribution of change scores at each time point, a dichotomous variable was created to indicate whether or not a subject had experienced an increase in new illnesses during the time periods analyzed.
P376
RODIN AND MeAVAY
emotional support was coded as 0. The use of a dichotomous scale for emotional support availability was chosen based on the work by Seeman and Berkman (1988) on social support in the elderly. An additional score was constructed based on responses to this item to reflect the number of different sources of emotional support available. The distinct sources counted included spouse, parents/parents-in-law, children, grandchildren, siblings/other relatives, and friends and other people (e.g., social worker, housecleaner, etc.).
Psychological Well-Being Measures Efficacy. — At each interview following the background, subjects were questioned about their feelings of self-efficacy for the current month. These questions assessed efficacy in eight domains of living: health, transportation, family relationships, finances, safety, relationships with friends, living arrangements, and productivity (see Appendix for these items). Each item was scaled using a Likert scale with the following four categories: strongly agree, agree, disagree, and strongly disagree. A total efficacy score was created by summing the responses to each of the eight domains of living. (For subjects who had one or two items missing, an average of the nonmissing items was calculated and weighted by 8.) Scores could range from 8 to 32, with higher scores reflecting lower levels of self-efficacy. Changes in self-efficacy were also calculated for each subject, with a positive score reflecting decreases in feelings of self-efficacy, whereas a negative score reflected increases in selfefficacy for the two time periods studied. Depression. — The Depression Adjective Checklist (DACL) (Lubin, 1966) was used to measure depressive affect at each interview. There are seven different forms of the brief DACL, each consisting of 16 self-descriptive adjectives. Intercorrelations among the lists range from .80 to .93, and internal consistencies range from .79 to .90 (Lubin, 1967). To administer the DACL, the interviewer asked subjects to think about how they had been feeling during the past day and to respond "yes" to words shown one at a time on cards, if the words accurately reflected their feelings. The depression score was calculated by adding the number of negative items chosen to the number of positive items not
Life events. — At the background interview, subjects were questioned about significant life events that happened during the past 5 years. In addition, at each interview following the background, they were asked about any new life events that had occurred between visits. The life events reported were categorized as negative, positive, or ambiguous using the PERI Life Events Scale (Dohrenwend, Krasnoff, Askenasy, & Dohrenwend, 1978). The number of negative, health-related, and nonhealth-related life events involving the respondent was then calculated for each interview. For the two time periods studied, subjects who reported more negative life events at the second as compared with the first interview were categorized as experiencing an increase in negative life events. Life satisfaction. — Life satisfaction was measured at the background interview using items from the Life Satisfaction Index of Neugarten, Havighurst, and Tobin (1961). Each item was scaled as strongly agree, agree, disagree, or strongly disagree, and a life satisfaction score was calculated as the sum of the items. Higher scores on this index reflect lower levels of life satisfaction. Demographic Measures Sex (0 = male, 1 = female) and age (0 = =S74, 1 = ^75) were also included in the analysis, as previous research has shown differences in perceived health between men and women, and between the young-old (^74) and the old-old (^75) (Ferraro, 1980; Linn & Linn, 1980; Stoller, 1984). Statistical Analyses The stability of the ordinal measures over time was assessed using the measure gamma (Agresti, 1984). 7-tests, analysis of covariance (ANCOVA), and chi-square tests were used to assess simple differences between the different perceived health groups (Daniel, 1983). These calculations were performed using SAS statistical software (SAS Institute, 1987). Hierarchical logistic regression models were used to examine simultaneously the effect of the demographic, psychosocial, and illness measures on change in perceived health (Hosmer & Lemeshow, 1989). A hierarchical approach was used so that variables could be entered in predetermined sets and their effect on the model assessed at each step. Two-way interactions between some of the variables in the final model were also examined. BMDPLR, a logistic regression procedure in BMDP (BMDP Statistical Software, 1988), was used to obtain the estimated regression coefficients by maximum likelihood methods. RESULTS
Stability of Measures The first thing notable in the data is the high degree of stability of perceived health across eight different measure-
Downloaded from http://geronj.oxfordjournals.org/ at New York University on June 2, 2015
Contact with social supports. — At the background interview, subjects were questioned in detail about children, siblings, relatives, friends, and other people whom they saw or talked to often. A score was calculated using this information to represent the potential number of supports available for three groups: children, siblings/relatives, and friends. At each interview following the background, subjects were questioned about the amount of face-to-face or phone contact they had in the past month with each person mentioned at the background interview. In addition, changes in the number of potential supports available were followed (e.g., new friendships developed, loss of a relative, etc.). A contact score was calculated for each interview to reflect the proportion of available people seen or talked to in each of the three groups. A proportion was used because the two perceived health groups differed initially on the baseline number of available supports.
chosen. Scores could range from 0 to 16. In order to assess mood fluctuations during the two time periods studied, changes in depression were calculated such that a positive score reflected an increase and a negative score reflected a decrease in depression level.
DETERMINANTS OF CHANGE IN PERCEIVED HEALTH
ment points in a 3-year time frame. In pairwise comparisons between each of the two consecutive time points (e.g., T l T2, T2-T3), the gamma measures ranged from .726 to .807. Comparing the two most distant points, the first and last interviews, 75% of the sample showed no change, 11% improved and 14% declined (gamma = .776). Because of the high overall stability or perceived health, we chose to analyze the data as described previously, selecting any two time points where there was a negative change as the unit of analysis.
Predictors of Negative Change in Perceived Health To determine whether negative changes in perceived health between any two time points were related to changes
Table 2. Measures at Background Interview by Change in Perceived Health Negative Change in Perceived Health (N = 145) N
No Change in Perceived Health {N = 106) N
40 105
27.6 72.4
42 64
39.6 60.4
.045
Age =S74yr 2*74yr
102 43
70.3 29.7
64 42
60.4 39.6
.100
Medical conditions present
X
SD
N
145 4.7
2.4
106 3.7 2.6
Perceived health Very good/good Fair Very poor/poor Emotional support available No Yes
Sources of emotional support available
SD
%
A'
%
87 41 16
60.4 28.5 11.1
82 20 4
77.4 18.9 3.8
10 134
6.9 93.1
10 95
SD
X
144 1.4
.8
105 1.3
%
109 36 W
X
A'
75.2 24.8 SD
Negative life events in past 5 years
145 1.0
1.0
Life satisfaction
144 4.4
.9
86 20 W 106
.004
.009
9.5 90.5
X
N Negative health life events in past 5 years No Yes
X
A'
N
Table 3. Measures at Prior Interview by Change in Perceived Health"
p- value
Sex Male Female
JV
ceived health, and have lower levels of life satisfaction (evidenced by higher numerical scores on the life satisfaction scale). The groups did not differ from one another at baseline in age, number of negative life events in the past five years, or on the two different measures of perceived emotional support. Another way to see how the groups may have differed prior to any decline in perceived health is to evaluate differences in measures taken just before the change in self-rated health took place. These data are presented in Table 3. Importantly, the two groups did not differ prior to declining in self-rated health in the number of new illnesses present, number of preexisting conditions that got worse, or visits to the doctor. They did not differ in the number reporting a negative life event, had similar contact with children, siblings/relatives, and friends, and were equal in level of depression. On the other hand, they did take significantly more medications than the group that would later experience no change in perceived health, and they had significantly lower feelings of self-efficacy.
.462
SD .8
.619
Negative Change in Perceived Health (A' = 145)
New illness present No Yes
No Change in Perceived Health {N = 106)
N
%
N
125 20
86.2 13.8
86 20
W
X
Preexisting conditions that got worse
145
.9 1.1
New or episodic medications taken
145 1.2
No. of doctor visits
145
Negative life event reported No Yes
.280
81.1 18.9
A'
X
SD
106
.9
I.I
.634
1.6
106
.7
.9
value
82.8 17.2
91 15
85.8 14.2
.507
A'
X
Contact with children (proportion)
SD
A'
X
SD
145
Contact with siblings/ relatives (proportion)
.78
.4
106
.82
.3
.338
Contact with friends (proportion)
145
.67
.3
106
.64
.4
.442
Self-efficacy
145
145 13.3
.77
2.0
105 17.7
106
.73
1.9
.3
.408 .026
Depression
145 3.6
3.2
106 3.2
2.7
.282
%
81.1 18.9 X
.260
SD
.9
1.0
.480
106 4.1
1.1
.008
.3
"Interview prior to the change in perceived health varied across subjects.
Downloaded from http://geronj.oxfordjournals.org/ at New York University on June 2, 2015
Differences Between Subjects With and Without Declines in Perceived Health It is important to determine how subjects with and without later negative changes in perceived health may have differed at baseline. As Table 2 shows, at the background interview, those who would later decline in self-rated health differed from those with no decline in that they were more likely to be female, have more medical conditions, have poorer per-
P377
RODIN AND MCAVAY
P378
Table 4. Changes in Health and Psychosocial Measures by Change in Perceived Health Negative Change in Perceived Health (N = 145)
No Change in Perceived Health (N = 106)
N
p-value
N
Increase in new illness No Yes
102 43
70.3 29.7
96 10
90.6 9.4
hmrnul of Gerontology: PSYCHOLOGICAL SCIENCES 1942. Vol. 47. No. 6, P373-P.184
Determinants of Change in Perceived Health in a Longitudinal Study of Older Adults Judith Rodin and Gail McAvay Psychology Department, Yale University.
perception of their health appears to be imporPEOPLE'S tant in predicting their subsequent mortality. This holds true even after controlling for the contribution of objective health status. Two studies of older people (Kaplan, Barell, & Lusky, 1988; Mossey & Shapiro, 1982) found that subjective health ratings predicted survival, independent of the effects of objective health, sex, health care utilization and, in the Kaplan etal. (1988) study, functional status. Idler, Kasl, and Lemke (1990) questioned the rigor of the objective health status measures used in the earlier studies and applied a variety of methodological and statistical controls to measures in their study. Self-reported health still predicted survival, independent of physical health status, disability, pain, medications, demographic factors, and health practices in an older age sample. Kaplan and Camacho (1983) also found strong support for the importance of self-rated health in predicting 9-year mortality in the Alameda County sample, which included adults of all ages. Their findings are impressive because they reported that perceived health was an independent predictor of mortality even when the regression equation included some measures of psychological functioning and social networks, as well as actual health, health practices, sex, and age. Most recently, Idler and Angel (1990) found that after controlling for the contribution of medical diagnoses derived by a medical examination, as well as demographic factors and health-related behaviors, self-rated health was significantly associated with mortality for middle-aged males in the sample, with a weaker trend for the older age group. Thus, the weight of the evidence suggests that many individuals, when asked how they feel, are evaluating something with important prognostic value regarding their probability of survival.
Such demonstrations lead quite naturally to asking what it is about perceived health that makes it so predictive, sometimes more so than presumably "harder" health-relevant data. Two bodies of literature relate to this question. The first study, much of which preceded the studies of perceived health and mortality, is used to validate perceived health as a measure of health when more "objective" indicators were unavailable. These correlational studies attempted to evaluate which classes of variables — health-relevant, demographic, mood state and general well being, and functional status — correlated with subjects' ratings of their own health. The second smaller and more recent literature is composed of studies attempting to determine which previously measured variables predict changes in perceived health over time. The first group of correlational studies shows that perceived health is clearly related to objective health status. Several find strong correlations between perceived health and long-standing chronic illness, especially in the elderly (Fillenbaum, 1979; Goldstein, Siegel, & Boyer, 1984; Jylha, Leskinen, Alanen, Leskinen, & Heikkinen, 1986; Liang, 1986; Linn & Linn, 1980; Tissue, 1972; Wan, 1976; Zonderman, 1986) and with other health indicators, e.g., number of medications, sick days, or hospitalizations (Fillenbaum, 1979; Linn & Linn, 1980; Wan, 1976). In the correlational studies, the relationship between perceived health and psychological indicators of well-being appears more inconsistent. In some work, poorer perceived health is significantly associated with greater neuroticism (Zonderman et al., 1986), hypochondriasis (Blazer & Houpt, 1979), depression (Blazer & Houpt, 1979; Goldstein & Hurwicz, 1989; Levkoff, Cleary, & Wetle, 1986), and lower life satisfaction (Blazer & Houpt, 1979; George & P373
Downloaded from http://geronj.oxfordjournals.org/ at New York University on June 2, 2015
To determine the factors that are predictive of a negative decline in perceived health in a longitudinal study of 251 men and women aged 62 and older, we developed a "synthetic" cohort of individuals who experienced a decline in perceived health between two time points and a control cohort of those who did not. The longitudinal design of the study made it possible to evaluate the effects of "baseline" predictors, gathered prior to the time interval during which the reported change in perceived health occurred, in addition to measures assessed at the two time points between which the change in health was reported. Hierarchical logistic regression models were used to examine the effect of demographic, psychosocial, and illness measures on change in perceived health. These analyses revealed that increases in new illnesses, increased physician visits, and worsening of preexisting conditions (in subjects who had higherfeelings of self-efficacy) were all associated with a decline in perceived health, after controlling for the effects of baseline illnesses and medication use. After accounting for these effects of changes in objective illness indicators, we found that changed psychosocial factors also predicted a decline in perceived health. Lower life satisfaction and higher depression at baseline were predictive of a later decline in perceived health. In addition, data collected at the third consecutive time point were evaluated to assess determinants of a sustained vs transient decline in perceived health. A sustained decline in perceived health was associated with increased depression and decreased self-efficacy.
P374
RODIN AND McAVAY
The focus of this study was on determining the relative importance of demographic, psychosocial, and health-status factors that are predictive of a decline in self-rated health. The longitudinal design of the study made it possible to evaluate the effects of "baseline" predictors gathered prior to the time interval during which, the reported change in
perceived health occurred, in addition to measures assessed at the two time periods between which the change in health was reported. METHODS
Subjects Subjects were 264 participants in a 3-year longitudinal study of psychosocial, dispositional, and biological factors related to changes in health. Participants were males and females aged 62 and older who lived in South Central Connecticut. All of the subjects lived in the community rather than in specialized care facilities, and all spoke English. Names were drawn from census tract data to sample from a broad cross-section of the community. Full recruitment procedures and response rates have been detailed previously (Rodin, McAvay, & Timko, 1988). Procedure The longitudinal design of this study included an initial, extensive baseline interview, followed by eight interviews spaced at varying time intervals over a 3-year period. The first four interviews were spaced at 2-month intervals; the next two took place one year after the fourth interview, and the last two took place one year after the sixth interview. Subjects who had complete baseline data on all variables relevant to the present questions (N = 251 or 95% of the entire sample) were included in the analyses. These subjects had completed an average of seven interviews following the baseline. Subjects were categorized into groups on the basis of their ratings of perceived health. Subjects rated their own health (on a 5-point scale: very good, good, fair, poor, and very poor) at each interview in response to the question, "How have you been feeling since I last talked to you?" To make the criteria for change more stringent and conceptually meaningful, these data were recoded into three categories: 1 = very good or good, 2 = fair, and 3 = poor or very poor. For example, using the 3-point scale, ratings that went from very good to good or poor to very poor were not considered a decline in perceived health. Changes in perceived health were then calculated for each of the seven possible consecutive time periods (i.e., between interviews 1-2, 2-3, 3-4, etc.). Subjects who had a negative change (decline in perceived health) between any two consecutive time points were included in Group 1. The first change in perceived health might have occurred from Interview 1 to Interview 2 for some cases, whereas for others it may have occurred from Interview 6 to Interview 7. Using this procedure, 145 subjects were categorized as experiencing a negative change in perceived health over the course of the study. The column in Table 1 for Group 1 indicates in which time period the first negative change in perceived health occurred for this sample. The availability of a third consecutive time point made it possible to examine the stability of this initial decline in perceived health and to assess determinants of the stability or instability of decline. Therefore, for some analyses, we divided the initial group of subjects who experienced a
Downloaded from http://geronj.oxfordjournals.org/ at New York University on June 2, 2015
Landerman, 1984). Other studies show no correlation with morale or other psychological indicators of well-being (Tissue, 1972; Wan, 1976). Self-perceptions of health also appear uncorrelated with social resource and network variables (Blazer & Houpt, 1979; Kutner, 1987). Self-assessed health does, on the other hand, correlate with functional capacity and instrumental activities (Liang, 1986; Linn & Linn, 1980). There are four reports in the second group of studies: those attempting to predict changes in perceived health over time. Supporting the earlier work of Maddox and Douglass (1973), most authors are impressed by the remarkable stability in perceived health in the overall samples studied. Three studies of older aged subjects — two prospective (Markides & Lee, 1990; Minkler & Langhauser, 1988) and one retrospective (Weinberger et al., 1986) — found that people whose self-rated health declined were older, had greater financial need, and had fewer social support resources five years earlier (Minkler & Langhauser, 1988), had lower levels of education and functional health eight years earlier (Markides & Lee, 1990), and had more negative life events one year earlier (Weinberger et al., 1986). Unlike Minkler and Langhauser (1988), Weinberger et al. found no direct or interactive (with changes in life events) effects of social support. Studying predictors of change over one year in a broader age range of subjects, Goldstein et al. found that changes in perceived health were not sensitive to short-term changes in objective health status (e.g., health beliefs, health services). Although these studies are quite informative, they lack measures of demographic factors, objective health status, psychological well-being, and social support together in any single study. Given the important relationship between many of these classes of variables and perceived health in the earlier, correlational studies, such an analysis is critical. Furthermore, none took multiple measures of perceived health over time. It is possible for subjects to change their ratings several times but appear quite stable when only two ratings are obtained. This longitudinal study attempted to address these issues by evaluating a full range of predictors in individuals who underwent a decline in perceived health, providing a more powerful examination of the determinants of negative changes in perceived health. The availability of eight time points with repeated measures of self-rated health allowed us to create "synthetic" groups (Campbell & Hudson, 1985) of individuals who experienced a sustained or nonsustained deterioration in perceived health between three consecutive time points. This type of comparison is powerful because each subject serves as his or her own control. In addition, however, we also developed a third synthetic cohort as a different type of control group. It was made up of individuals who did not experience a negative change in perceived health at any time point through the entire course of the study.
DETERMINANTS OF CHANGE IN PERCEIVED HEALTH
Health Measures
Table 1. Interviews Used in Statistical Analysis Group 1 Negative Change in Perceived Health
Group 2 No Change in Perceived Health
Interviews Used
N
%
N
%
1 and 2 2 and 3 3 and 4 4 and 5 5 and 6 6 and 7 7 and 8
38 25 31 16 18 11 6
26.2 17.2 21.4 11.0 12.4 7.6 4.1
28 18 22 12 13 4
26.4 17.0 20.8 11.3 12.3 8.5 3.8
145
100.0
106
100.0
Total
9
P375
Doctor visits. — At each visit following the background interview, subjects were questioned about any doctors they had seen since the previous interview. The type of physician (e.g., cardiologist, general practitioner, etc.), reason for the visit, and any treatments prescribed by the physician were recorded. A score reflecting the number of doctor visits since the previous interview was constructed for each subject. Change in the number of doctor visits reported at the two time periods studied was then calculated for each subject, such that a positive score represented an increase and a negative score a decrease in the number of physicians seen. Medications. — Use of medications prescribed by the subject's physician was assessed at the background and at each interview. The number of new medications prescribed plus the number of preexisting, episodic (taken on an asneeded basis) medications taken at each interview were totaled for each subject. Social Support Measures Emotional support availability. — The presence (coded as 1) of emotional support was measured at the background interview in response to the item, "Whom can you really count on to listen when you need to talk?" The absence of
Downloaded from http://geronj.oxfordjournals.org/ at New York University on June 2, 2015
negative change in perceived health over the first two time points into two groups based upon their perceived health rating at the third time point. Subjects who reported the same or a lower level of perceived health at the interview following the initial decline in perceived health were included in the negative change group (/V = 51), whereas subjects whose ratings went back up were included in the transient negative change group (N = 72). Subjects missing a third time point of data were excluded from these analyses. The 106 respondents who did not report a negative change in perceived health between any consecutive time points of the study were included in Group 2. The absence of negative change in this group cannot be attributed to a "floor" effect, because through the study, only four of these subjects rated their health at the worst level (poor or very poor). Three of these subjects reported improvement in perceived health at the following interview, whereas the other subject's health rating was poor at the first two interviews and showed improvement during the remainder of the study. For the second group of subjects, any of the two consecutive time points of data where their ratings of perceived health were stable were available for use. We selected the time points that would make the percentage of subjects with data from each of the seven possible time intervals the same in Groups 1 and 2. Subjects in Group 2 were assigned at random to a particular interval. Assigning the subjects in the second group to correspond to the percentages in the first group for each interval reduces the possibility of bias due to the use of different time intervals. The column for Group 2 in Table 1 presents the interviews used in the statistical analysis for this group. There was a somewhat higher percentage of respondents who did not have the last interview in Group 2 (N = 28,26%) than in Group 1 (N = 25,17%). Although the 28 subjects in the "no change group" who did not reach the final interview could possibly have declined in perceived health if they had been interviewed again, we decided to include them in the control group. We believe that this represents a conservative approach, because they might later have experienced a negative change that would have placed them in the other group. A possible misclassification into the control group reduced the possibility of finding differences between the two groups.
Medical conditions. — At the background interview, subjects were questioned about illnesses and medical conditions using a standardized form that included diseases from all of the major categories (Fetter, Shin, Freeman, Averill, & Thompson, 1980). At each interview following the background, subjects were asked about any new illnesses or accidents that had occurred since the previous interview, and the status of each of the medical conditions reported at the background was evaluated. Each condition was assigned a code number from the diagnosis categorization system, Diagnosis-Related Groups (Fetter et al.). There were two types of illness measures constructed from these data. First, changes in existing chronic conditions (i.e., conditions that were ongoing) and recurrences of preexisting episodic conditions (i.e., conditions that reappeared from time to time with no symptoms in the intervals between) were scored. For each subject, the number of chronic conditions that got worse and the number of episodic conditions that recurred were totaled to represent the number of preexisting medical conditions that worsened. A change score was calculated so that a positive number reflected a greater number of preexisting conditions that worsened at the second time point, compared with the first time point. Second, the number of new illnesses never reported before was calculated at each interview. A change score was calculated such that a positive number reflected the number of new illnesses reported at the second time point was greater than the number reported at the first time point. Based on the bivariate distribution of change scores at each time point, a dichotomous variable was created to indicate whether or not a subject had experienced an increase in new illnesses during the time periods analyzed.
P376
RODIN AND MeAVAY
emotional support was coded as 0. The use of a dichotomous scale for emotional support availability was chosen based on the work by Seeman and Berkman (1988) on social support in the elderly. An additional score was constructed based on responses to this item to reflect the number of different sources of emotional support available. The distinct sources counted included spouse, parents/parents-in-law, children, grandchildren, siblings/other relatives, and friends and other people (e.g., social worker, housecleaner, etc.).
Psychological Well-Being Measures Efficacy. — At each interview following the background, subjects were questioned about their feelings of self-efficacy for the current month. These questions assessed efficacy in eight domains of living: health, transportation, family relationships, finances, safety, relationships with friends, living arrangements, and productivity (see Appendix for these items). Each item was scaled using a Likert scale with the following four categories: strongly agree, agree, disagree, and strongly disagree. A total efficacy score was created by summing the responses to each of the eight domains of living. (For subjects who had one or two items missing, an average of the nonmissing items was calculated and weighted by 8.) Scores could range from 8 to 32, with higher scores reflecting lower levels of self-efficacy. Changes in self-efficacy were also calculated for each subject, with a positive score reflecting decreases in feelings of self-efficacy, whereas a negative score reflected increases in selfefficacy for the two time periods studied. Depression. — The Depression Adjective Checklist (DACL) (Lubin, 1966) was used to measure depressive affect at each interview. There are seven different forms of the brief DACL, each consisting of 16 self-descriptive adjectives. Intercorrelations among the lists range from .80 to .93, and internal consistencies range from .79 to .90 (Lubin, 1967). To administer the DACL, the interviewer asked subjects to think about how they had been feeling during the past day and to respond "yes" to words shown one at a time on cards, if the words accurately reflected their feelings. The depression score was calculated by adding the number of negative items chosen to the number of positive items not
Life events. — At the background interview, subjects were questioned about significant life events that happened during the past 5 years. In addition, at each interview following the background, they were asked about any new life events that had occurred between visits. The life events reported were categorized as negative, positive, or ambiguous using the PERI Life Events Scale (Dohrenwend, Krasnoff, Askenasy, & Dohrenwend, 1978). The number of negative, health-related, and nonhealth-related life events involving the respondent was then calculated for each interview. For the two time periods studied, subjects who reported more negative life events at the second as compared with the first interview were categorized as experiencing an increase in negative life events. Life satisfaction. — Life satisfaction was measured at the background interview using items from the Life Satisfaction Index of Neugarten, Havighurst, and Tobin (1961). Each item was scaled as strongly agree, agree, disagree, or strongly disagree, and a life satisfaction score was calculated as the sum of the items. Higher scores on this index reflect lower levels of life satisfaction. Demographic Measures Sex (0 = male, 1 = female) and age (0 = =S74, 1 = ^75) were also included in the analysis, as previous research has shown differences in perceived health between men and women, and between the young-old (^74) and the old-old (^75) (Ferraro, 1980; Linn & Linn, 1980; Stoller, 1984). Statistical Analyses The stability of the ordinal measures over time was assessed using the measure gamma (Agresti, 1984). 7-tests, analysis of covariance (ANCOVA), and chi-square tests were used to assess simple differences between the different perceived health groups (Daniel, 1983). These calculations were performed using SAS statistical software (SAS Institute, 1987). Hierarchical logistic regression models were used to examine simultaneously the effect of the demographic, psychosocial, and illness measures on change in perceived health (Hosmer & Lemeshow, 1989). A hierarchical approach was used so that variables could be entered in predetermined sets and their effect on the model assessed at each step. Two-way interactions between some of the variables in the final model were also examined. BMDPLR, a logistic regression procedure in BMDP (BMDP Statistical Software, 1988), was used to obtain the estimated regression coefficients by maximum likelihood methods. RESULTS
Stability of Measures The first thing notable in the data is the high degree of stability of perceived health across eight different measure-
Downloaded from http://geronj.oxfordjournals.org/ at New York University on June 2, 2015
Contact with social supports. — At the background interview, subjects were questioned in detail about children, siblings, relatives, friends, and other people whom they saw or talked to often. A score was calculated using this information to represent the potential number of supports available for three groups: children, siblings/relatives, and friends. At each interview following the background, subjects were questioned about the amount of face-to-face or phone contact they had in the past month with each person mentioned at the background interview. In addition, changes in the number of potential supports available were followed (e.g., new friendships developed, loss of a relative, etc.). A contact score was calculated for each interview to reflect the proportion of available people seen or talked to in each of the three groups. A proportion was used because the two perceived health groups differed initially on the baseline number of available supports.
chosen. Scores could range from 0 to 16. In order to assess mood fluctuations during the two time periods studied, changes in depression were calculated such that a positive score reflected an increase and a negative score reflected a decrease in depression level.
DETERMINANTS OF CHANGE IN PERCEIVED HEALTH
ment points in a 3-year time frame. In pairwise comparisons between each of the two consecutive time points (e.g., T l T2, T2-T3), the gamma measures ranged from .726 to .807. Comparing the two most distant points, the first and last interviews, 75% of the sample showed no change, 11% improved and 14% declined (gamma = .776). Because of the high overall stability or perceived health, we chose to analyze the data as described previously, selecting any two time points where there was a negative change as the unit of analysis.
Predictors of Negative Change in Perceived Health To determine whether negative changes in perceived health between any two time points were related to changes
Table 2. Measures at Background Interview by Change in Perceived Health Negative Change in Perceived Health (N = 145) N
No Change in Perceived Health {N = 106) N
40 105
27.6 72.4
42 64
39.6 60.4
.045
Age =S74yr 2*74yr
102 43
70.3 29.7
64 42
60.4 39.6
.100
Medical conditions present
X
SD
N
145 4.7
2.4
106 3.7 2.6
Perceived health Very good/good Fair Very poor/poor Emotional support available No Yes
Sources of emotional support available
SD
%
A'
%
87 41 16
60.4 28.5 11.1
82 20 4
77.4 18.9 3.8
10 134
6.9 93.1
10 95
SD
X
144 1.4
.8
105 1.3
%
109 36 W
X
A'
75.2 24.8 SD
Negative life events in past 5 years
145 1.0
1.0
Life satisfaction
144 4.4
.9
86 20 W 106
.004
.009
9.5 90.5
X
N Negative health life events in past 5 years No Yes
X
A'
N
Table 3. Measures at Prior Interview by Change in Perceived Health"
p- value
Sex Male Female
JV
ceived health, and have lower levels of life satisfaction (evidenced by higher numerical scores on the life satisfaction scale). The groups did not differ from one another at baseline in age, number of negative life events in the past five years, or on the two different measures of perceived emotional support. Another way to see how the groups may have differed prior to any decline in perceived health is to evaluate differences in measures taken just before the change in self-rated health took place. These data are presented in Table 3. Importantly, the two groups did not differ prior to declining in self-rated health in the number of new illnesses present, number of preexisting conditions that got worse, or visits to the doctor. They did not differ in the number reporting a negative life event, had similar contact with children, siblings/relatives, and friends, and were equal in level of depression. On the other hand, they did take significantly more medications than the group that would later experience no change in perceived health, and they had significantly lower feelings of self-efficacy.
.462
SD .8
.619
Negative Change in Perceived Health (A' = 145)
New illness present No Yes
No Change in Perceived Health {N = 106)
N
%
N
125 20
86.2 13.8
86 20
W
X
Preexisting conditions that got worse
145
.9 1.1
New or episodic medications taken
145 1.2
No. of doctor visits
145
Negative life event reported No Yes
.280
81.1 18.9
A'
X
SD
106
.9
I.I
.634
1.6
106
.7
.9
value
82.8 17.2
91 15
85.8 14.2
.507
A'
X
Contact with children (proportion)
SD
A'
X
SD
145
Contact with siblings/ relatives (proportion)
.78
.4
106
.82
.3
.338
Contact with friends (proportion)
145
.67
.3
106
.64
.4
.442
Self-efficacy
145
145 13.3
.77
2.0
105 17.7
106
.73
1.9
.3
.408 .026
Depression
145 3.6
3.2
106 3.2
2.7
.282
%
81.1 18.9 X
.260
SD
.9
1.0
.480
106 4.1
1.1
.008
.3
"Interview prior to the change in perceived health varied across subjects.
Downloaded from http://geronj.oxfordjournals.org/ at New York University on June 2, 2015
Differences Between Subjects With and Without Declines in Perceived Health It is important to determine how subjects with and without later negative changes in perceived health may have differed at baseline. As Table 2 shows, at the background interview, those who would later decline in self-rated health differed from those with no decline in that they were more likely to be female, have more medical conditions, have poorer per-
P377
RODIN AND MCAVAY
P378
Table 4. Changes in Health and Psychosocial Measures by Change in Perceived Health Negative Change in Perceived Health (N = 145)
No Change in Perceived Health (N = 106)
N
p-value
N
Increase in new illness No Yes
102 43
70.3 29.7
96 10
90.6 9.4
