shows characteristics for the full sample at baseline in 1982, for the subsample in the LYOL, and for the subsample at baseline in 1982. For 142 respondents, the baseline interview was the one that took place during their LYOL; hence, their values are the same; for the other 357, the baseline measure is taken approximately 4, 7, or 13 years prior to death. Because of overlap between the samples, the differences are for the purposes of illustration only, and no statistical tests are performed. Compared with the full baseline sample, the LYOL subsample attends religious services less frequently, about 18 times per year by comparison with 26. But on the more subjective measures of religiousness, there are no differences. The LYOL subsample is more likely to be male, is older by almost 7 years, has slightly less education, is slightly more likely to be White, is slightly more likely to have low income, and is much less likely to be married. As one would expect, on nearly every indicator of HRQOL, the LYOL sample has poorer health, with much higher functional disability scores, more prescription medications, a greater history of hospitalization in the past year or ever having been in a nursing home, and a higher percent with more than 1 week in bed in the last 3 months. When we compared the functional disability scores for the LYOL with the respondents’ scores at the previous round of interviews 3 years earlier (data not shown), about 33% had experienced a major decline in functioning, whereas 44% had persisted with relatively little change in mild to moderate (scores 1–40) or severe (scores 41–150) disability during the previous 3 years. Among respondents with proxies, 69% experienced a major decline in functioning and just 30% stayed at approximately the same level of disability, whereas among nonproxy respondents, only 19% declined, 69% remained stably disabled, and 11% remained nondisabled. We could thus characterize the end-of-life trajectories of our subsample as similar to those found among elderly persons in the U.S. population as a whole, the majority of whom experience long periods of moderate to severe disability in the years prior to death.
Descriptive Means and Percentages for Baseline Sample and LYOL Subsample, New Haven Established Populations for Epidemiologic Studies of the Elderly
With respect to QOL, the differences between the LYOL subsample and the full sample are inconsistent. The LYOL subsample is slightly less likely to report having a confidant, and reports seeing kin less often, but a greater percentage reports seeing friends. The number of holidays celebrated (question asked only in 1982 and 1985) was about the same. CESD and subscale scores are on average higher for the LYOL subsample, though both are well below the clinical depression approximation cut point of 16. Self-rated health is poorer for the LYOL subsample, and a smaller number were rated as “finds life exciting” by the interviewer.
In all the following analyses, we consider only the subsample of those in their LYOL. describes the religiousness of our respondents and addresses the issue of health selection for the ability to attend services. Perhaps surprisingly, there are no differences by gender, age, education, race, income, or marital status within our LYOL subsample with respect to the frequency of attendance at service. However, those with higher levels of functional disability attend services less frequently; for every additional 10 points on our disability scale of 150, a respondent attends services almost two times per year less frequently. On the other hand, those who were patients in a nursing home attended services on average 8.1 times more often than those who had not been institutionalized; this is possibly an indication of the frequency with which religious services are held in the nursing homes where respondents were residents. There are no other significant coefficients; functional disability and nursing home residence explain about 7% of the overall variance in religious service attendance. We note that the larger N for this model indicates that the attendance variable was answered in proxy interviews.
Unstandardized Coefficients From General Linear Models for Religion Variables
There are more demographic differences in how deeply religious respondents report themselves to be. Males, Whites, and those with more education are significantly less likely than females, non-Whites, and those with less education to say they are deeply religious. None of the health status or disability indicators are associated with feelings of religiousness. Demographic factors explain 10% of the variance in this measure. For the third measure of religiousness—receiving strength and comfort from religion—there is yet a different pattern. Males and Whites are less likely than females and non-Whites to say that they have received a great deal of strength and comfort from their religion, whereas those who had been in a hospital in the last year reported more strength and comfort. The R2 for this third model is .08. Proxies did not answer the two subjective religiousness items, so the sample sizes are smaller.
shows results for the test of the hypothesis that religious involvement in the LYOL is associated with a better social QOL. These models adjust for all demographic and health variables. The number of kin seen is fewer if a respondent is not married or is in a nursing home. There is no significant association of any religion variable with seeing kin. With respect to seeing friends, however, those who are deeply religious are 62% more likely to have seen any friend than those who are less religious. None of the religion variables is related to reporting having a confidant, although being unmarried and/or disabled reduces the chances of having a confidant. The final social QOL indicator is the number of holidays celebrated; those respondents who attended services more frequently also attended significantly more holiday parties. The models show only modest, albeit positive associations of religious involvement with social QOL during the LYOL.
Unstandardized Coefficients and ORs From General Linear and Logistic Regression Models, for Indicators of Social Quality of Life
examines the hypothesis that religiousness during the LYOL will be associated with better psychological well-being, including CESD depressive symptoms and its subscales, self-rated health, and the interviewer's rating of how exciting the informant finds his/her life. For most of these models, having a low income and being disabled and, to a lesser extent, spending more than a week in bed is associated with poorer well-being, whereas there are no significant differences by gender, age, education, or race. Attendance at religious services has no main effect associations with any of the outcomes. Being deeply religious is associated with lower total CESD scores; for every increased level of religiousness, CESD scores decline by 2.6 points. Among the CESD subscales, the positive affect and somatic symptoms subscales are also associated with religiousness; thus, it appears that the association of being deeply religious with fewer depressive symptoms is due particularly to the lower level of somatic symptoms and better positive affect experienced by the more deeply religious. Specific items include fewer difficulties sleeping, less poor appetite, and more feelings of happiness. Being deeply religious is also associated with better self-rated health and with better interviewer–observer ratings of the respondent finding his/her life exciting (respondents were 51% less likely to be rated by the interviewer as “does not find life exciting and enjoyable”). Getting a great deal of strength and comfort from religion shows a different picture, including associations with poorer outcomes; it is associated with more, not fewer, somatic symptoms and with poorer, not better, self-reported health.
Unstandardized Coefficients From General Linear and Logistic Regression Models for Indicators of Subjective Health and Psychological Well-being
To test the idea that religious involvement might carry special benefits for the most disabled respondents in the sample, we tested interactions of functional disability with each of the religion variables for each of the outcomes. The significant interactions were all for self-rated health. When tested singly in models, there are significant, positive-sign interactions of all three religion variables, meaning that among the most disabled, those who say they are deeply religious, get a lot of strength and comfort, and attend service (marginally) rate their health better than those who are less religiously involved, and these relationships are stronger than they are for the nondisabled. We show the composite of these three models in the second set of columns under self-rated health. Overall, we see that one or more of the religion variables is associated in the hypothesized direction with each of the dependent variables, with the exception that getting strength and comfort from religion was associated with poorer, not better, self-rated health. This finding had been anticipated in , where more reported strength and comfort was reported by those who had a hospital stay, suggesting that health crises may have elicited a religious response. The other finding in is that people with disabilities show a stronger association of all three of the religion variables with better self-rated health than do the nondisabled.
We conducted a further set of analyses, subdividing the subsample into those who were in their last 6 months of life and those who had 7–12 months of life remaining. The findings (available on request) were largely similar but showed a larger number of significant associations for the 7- to 12-month group than for the 0- to 6-month group; all the significant interactions with self-rated health for the sample as a whole pertained only to the 7- to 12-month group.