Journal of Gerontology: SOCIAL SCIENCES
Copyright 1992 by The Gerontologkal Society of America
1992. Vol. 47. No. 4.SI73-SI82
The Risk of Nursing Home Placement and Subsequent Death Among Older Adults Fredric D. Wolinsky,12 Christopher M. Callahan,12 John F. Fitzgerald,'23 and Robert J. Johnson4
This article examines the effects of the characteristics specified in the behavioral model of health services utilization and measured at baseline on the subsequent risk of nursing home placement and death within four years. Analyses of the 5,151 respondents in the Longitudinal Study on Aging indicate that the risk for nursing home placement is greater for older adults, Whites, those who lived alone, persons with telephones, those with fewer nonkin social supports, those who did not feel that they had much control over their future health, those with more household ADL or lower body limitations, and those who had been in the hospital during the year prior to baseline, or in a nursing home at any time before baseline. A mong the 549 respondents placed in nursing homes, the risk of dying there was greaterfor older adults, men, those who had not lived in multigenerational households, persons who did not worry about their health, individuals with more upper body limitations, and respondents having a history of valvular heart disease or cancer. The odds of dying were 2.74 times greater among the 549 respondents placed in nursing homes than among the 4,602 respondents who remained in the community.
any given day, 1.3 million, or 4.5 percent of all ONAmericans aged 65 or older, may be found in nursing homes (Lazenby and Letsch, 1990). Older adults constitute most (88.4%) of the nursing home population. In 1989, the total cost of nursing home care was $47.9 billion, or 8 percent of all health care expenditures. This was a 12 percent increase from the year before, and exceeded the 9 percent average annual increase observed for the 1980s overall. The bulk (64%) of nursing home expenditures are financed by public programs, either directly through Medicaid (43%) or Medicare (about 3% since the December 1989 repeal of the Medicare Catastrophic Coverage Act), or indirectly from transfer payments (i.e., the Medicaid mandated contributions from the resident's Social Security income; 18%). Given the projected considerable increase in the number and proportion of older Americans (Rice and Feldman, 1981), the use of and expense associated with nursing homes has become an issue of considerable national concern. That concern has stimulated research on nursing home utilization in recent years (see Greene and Ondrich [1990] for a current review). The majority of these studies focus either on admission rates (e.g., Wingard et al., 1990) or the lifetime likelihood of nursing home placement (e.g., Murtaugh, Kemper, and Spillman, 1990). Some, however, have examined transition patterns between the community, the nursing home, the hospital, and death (e.g., Garber and MaCurdy, 1989; Liang and Tu, 1986). Still others have identified profiles of older adults that are most predictive of nursing home placement (e.g., Morris, Sherwood, and Gutkin, 1988; Shapiro and Tate, 1988). Perhaps what characterizes this growing body of literature most is the marked variation in design and analysis (see
Greene and Ondrich [1990], Murtaugh, Kemper, and Spillman [1990], and Wingard et al. [1990] for thoughtful discussions of these issues). With respect to design, some studies are based on cross-sectional surveys of living persons (e.g., Hing and Sekscenski, 1987; Ingram and Barry, 1977), others involve cross-sectional reviews of decedents' records (e.g., Lesnoff-Caravaglia, 1978-79), and some prospectively follow members of the same cohort over time (e.g., Branch and Jette, 1982; Vicente, Wiley, and Carrington, 1979). With few exceptions (e.g., Cohen, Tell, and Wallack, 1986), most of these studies have involved local or state, rather than national samples (see Wingard et al.. 1990). In terms of analytic strategy, multivariate modeling has been a rather recent development (see Greene and Ondrich, 1990). Moreover, even when a variety of factors are used to predict the risk of nursing home placement, they are seldom embedded within a coherent, conceptual framework (for a notable exception, see Greene and Ondrich, 1990). As a result, although the literature on nursing home utilization is no longer limited in volume, much remains to be known about the use of this critically important health service. In this article we use the baseline characteristics obtained from the respondents to the Longitudinal Study on Aging (LSOA) in 1984 to model their risk of nursing home placement and subsequent death by 1988. Our approach differs from prior studies in four ways. First, we use a large, national probability sample of noninstitutionalized older adults who were prospectively followed for four years. Second, we include a rather comprehensive array of psychosocial, physical, and functional health status measures, as well as prior health services utilization patterns. Third, we S173
Downloaded from http://geronj.oxfordjournals.org/ at Queen Mary, University of London on July 11, 2014
'Department of Medicine, Indiana University School of Medicine. 2 Regenstrief Institute for Health Care. 3 Richard Roudebush Veterans Administration Medical Center. "Department of Sociology, Kent State University.
S174
WOLINSKY ET AL.
frame our multivariate statistical modeling from the conceptual perspective of the behavioral model of health services utilization (Andersen, 1968). Finally, we identify those risk factors associated with dying after nursing home placement.
METHODS
The data are taken from the LSOA, which is a collaborative effort of the National Center for Health Statistics (NCHS) and the National Institute on Aging. The LSOA involves a special six-year follow-up to the 1984 Health Interview Survey (HIS). Because it is such a well-known data source, the HIS will not be discussed here. For the interested reader, a detailed description of its history, design, and logistics can be found in several readily available publications (NCHS, 1975, 1985). In 1984 the HIS was augmented by two special supplements. One involved the collection of detailed information on health insurance coverage among all respondents. The other (referred to as the Supplement on Aging [SOA]) in-
Downloaded from http://geronj.oxfordjournals.org/ at Queen Mary, University of London on July 11, 2014
Model The behavioral model of health services utilization was first presented by Andersen in 1968, and subsequently revised with his colleagues. Because it is the most widely used model for studying health services utilization (Wan, 1989), it need not be discussed here at length. Indeed, several detailed, historical reviews of the behavioral model already exist, including its application to the special case of older adults (e.g., Wolinsky, 1990). Because of its comprehensive approach, the behavioral model provides an excellent context within which to examine the effects of a variety of factors on nursing home placement and subsequent death. Basically, the behavioral model views the use of health services as a function of the predisposing, enabling, and need characteristics of the individual. The predisposing component is an abstraction from the proposition that some individuals have a greater propensity for using health services than do others. These propensities can be predicted from individual characteristics prior to an illness episode. The three dimensions of the predisposing characteristics include demographics, social structure, and health beliefs. Collectively, they represent the sociocultural element of the behavioral model. The enabling component is abstracted from the proposition that although the individual may be predisposed to use health services, he or she must nonetheless have some means for obtaining them. The enabling component, then, contains those factors which make health services available to the individual for consumption, and includes familial and community resources. These two dimensions of the enabling characteristics represent the economic component of the behavioral model. Although the predisposing and enabling components are necessary conditions for the use of health services, they are not sufficient ones. To use health services the individual must have or perceive some illness (or its possibility). This need has two dimensions. One represents the amount of illness that the individual perceives exists, and the other represents professionally evaluated need. These need indicators tap the individual's recognition that a health problem either exists or is in the making.
volved the collection of detailed information by way of a 30minute add-on interview concerning the health, social functioning, and living arrangements of 16,148 individuals aged 55 and older. A sample of 5,151 SOA individuals aged 70 years or more in 1984 was selected for follow-up interviews (by telephone if they had one, or by mail if they did not) in 1986, 1988, and 1990 (although the 1990 data are not yet available for public use). It is this sample, known as the LSOA, that is the source of data for our research. Greater detail on the design and execution of the LSOA can be found in Fitti and Kovar (1987). With the exception of nursing home placement and mortality status, all of the data used in the present analysis are based on self-reports given in the baseline interview, which involved household proxies in 8.0 percent of the cases and nonhousehold proxies in 1.6 percent of the cases. In this article we used the unweighted data. It has been empirically shown that the complex schemes necessary to take the disproportionately stratified multistage cluster sampling design of the LSOA into account have little impact on variance estimation (see Fitti and Kovar). Moreover, that impact is sufficiently attenuated by the inclusion of age and race as covariates in multivariate models to warrant analyses of the unweighted data (Fitti and Kovar, 1987). The measures of the predisposing, enabling, need, and health services utilization characteristics are generally straightforward, rather typical of the literature, have been described in detail elsewhere (Wolinsky and Johnson, 1991, 1992), and are summarized in Table 1. Accordingly, we shall not review them here in great detail. As indicated, there are 11 measures of the predisposing characteristics. These include age, sex, race, living alone, living in a multigenerational household, having a telephone, educational attainment, kin social supports, nonkin social supports, feeling a sense of control over one's health, and worrying about one's health. There are five measures of the enabling characteristics. These include having private insurance that covers at least some portion of hospital and physician charges, whether the respondent has a valid Medicaid card, residential stability, population density based on county adjacency codes, and sole financial dependence on public retirement sources of minimal magnitude (i.e., Social Security). As shown, there are 13 measures of the need for health services. Because of their traditionally pivotal role as risk factors for nursing home placement and mortality, they will be described at somewhat greater length. One is a dichotomous index of perceived health, indicating whether the respondent is in excellent, very good, or good health vs being in fair or poor health. Five of the remaining measures of health need are multiple-item scales that emerged from theoretically directed factor and principal components analyses of 21 questions routinely taken from or modeled after various measures of activities of daily living. We call the first the basic activities of daily living scale (basic ADL; minimum factor loading [the smallest correlation between an item in the scale and the underlying factor] = .718; alpha [an internal consistency based reliability coefficient with a lower limit of zero and an upper limit of unity] = .827). It consists of four items from the traditional ADL (Katz et al., 1963), including having any difficulties with such personal activities as bathing, dressing, getting out of bed, and
S175
NURSING HOME PLACEMENT AND DEATH
Table 1. Means, Standard Deviations, Coding Algorithms, and Psychometric Properties of the Baseline LSOA Data Coding Algorithms
SD
78.113 .640 .106 .378 .188 .969 9.755 1.618 2.347 .330 .697
5.917 .480 .308 .485 .391 .174 3.738 .644 1.338 .470 .460
iictual number of years = yes, 0 = no = yes, 0 = no = yes, 0 = no = yes, 0 = no = yes, 0 = no iictual number of years :'-item scale, 1 = has support, •>-item scale, 1 = has support, = worries about health, 0 = = has control over health, 0
.645 .056 .838 7.390
.478 .230 .369 2.526
.653
.476
= has private physician and hospital insurance, 0 = no = has Medicaid card, 0 = no = same address for 5 or more years, 0 = no 0 point 1980 county adjacency code, ranging from () = thinly populated not adjacent to 9 = core SMSA county = yes, 0 = no
Need Characteristics Perceived health Basic activities of daily living Household activities of daily living Advanced activities of daily living Lower body limitations Upper body limitations Ever had atherosclerosis Ever had valvular heart disease Ever had osteoporosis Ever had a fractured hip Ever had cerebrovascular disease Ever had cancer Ever had Alzheimer's disease
.660 .605 .656 .163 1.959 .442 .192 .025 .036 .050 .478 .120 .005
.474 1.222 1.167 .515 1.953 .821 .394 .156 .185 .218 .500 .326 .073
= excellent, very good, or good, 0 = fair or poor !5-item scale, 1 = has difficulty, 0 = no difficulty 4-item scale, 1 = has difficulty, 0 = no difficulty :5-item scale, 1 = has difficulty, 0 = no difficulty
Copyright 1992 by The Gerontologkal Society of America
1992. Vol. 47. No. 4.SI73-SI82
The Risk of Nursing Home Placement and Subsequent Death Among Older Adults Fredric D. Wolinsky,12 Christopher M. Callahan,12 John F. Fitzgerald,'23 and Robert J. Johnson4
This article examines the effects of the characteristics specified in the behavioral model of health services utilization and measured at baseline on the subsequent risk of nursing home placement and death within four years. Analyses of the 5,151 respondents in the Longitudinal Study on Aging indicate that the risk for nursing home placement is greater for older adults, Whites, those who lived alone, persons with telephones, those with fewer nonkin social supports, those who did not feel that they had much control over their future health, those with more household ADL or lower body limitations, and those who had been in the hospital during the year prior to baseline, or in a nursing home at any time before baseline. A mong the 549 respondents placed in nursing homes, the risk of dying there was greaterfor older adults, men, those who had not lived in multigenerational households, persons who did not worry about their health, individuals with more upper body limitations, and respondents having a history of valvular heart disease or cancer. The odds of dying were 2.74 times greater among the 549 respondents placed in nursing homes than among the 4,602 respondents who remained in the community.
any given day, 1.3 million, or 4.5 percent of all ONAmericans aged 65 or older, may be found in nursing homes (Lazenby and Letsch, 1990). Older adults constitute most (88.4%) of the nursing home population. In 1989, the total cost of nursing home care was $47.9 billion, or 8 percent of all health care expenditures. This was a 12 percent increase from the year before, and exceeded the 9 percent average annual increase observed for the 1980s overall. The bulk (64%) of nursing home expenditures are financed by public programs, either directly through Medicaid (43%) or Medicare (about 3% since the December 1989 repeal of the Medicare Catastrophic Coverage Act), or indirectly from transfer payments (i.e., the Medicaid mandated contributions from the resident's Social Security income; 18%). Given the projected considerable increase in the number and proportion of older Americans (Rice and Feldman, 1981), the use of and expense associated with nursing homes has become an issue of considerable national concern. That concern has stimulated research on nursing home utilization in recent years (see Greene and Ondrich [1990] for a current review). The majority of these studies focus either on admission rates (e.g., Wingard et al., 1990) or the lifetime likelihood of nursing home placement (e.g., Murtaugh, Kemper, and Spillman, 1990). Some, however, have examined transition patterns between the community, the nursing home, the hospital, and death (e.g., Garber and MaCurdy, 1989; Liang and Tu, 1986). Still others have identified profiles of older adults that are most predictive of nursing home placement (e.g., Morris, Sherwood, and Gutkin, 1988; Shapiro and Tate, 1988). Perhaps what characterizes this growing body of literature most is the marked variation in design and analysis (see
Greene and Ondrich [1990], Murtaugh, Kemper, and Spillman [1990], and Wingard et al. [1990] for thoughtful discussions of these issues). With respect to design, some studies are based on cross-sectional surveys of living persons (e.g., Hing and Sekscenski, 1987; Ingram and Barry, 1977), others involve cross-sectional reviews of decedents' records (e.g., Lesnoff-Caravaglia, 1978-79), and some prospectively follow members of the same cohort over time (e.g., Branch and Jette, 1982; Vicente, Wiley, and Carrington, 1979). With few exceptions (e.g., Cohen, Tell, and Wallack, 1986), most of these studies have involved local or state, rather than national samples (see Wingard et al.. 1990). In terms of analytic strategy, multivariate modeling has been a rather recent development (see Greene and Ondrich, 1990). Moreover, even when a variety of factors are used to predict the risk of nursing home placement, they are seldom embedded within a coherent, conceptual framework (for a notable exception, see Greene and Ondrich, 1990). As a result, although the literature on nursing home utilization is no longer limited in volume, much remains to be known about the use of this critically important health service. In this article we use the baseline characteristics obtained from the respondents to the Longitudinal Study on Aging (LSOA) in 1984 to model their risk of nursing home placement and subsequent death by 1988. Our approach differs from prior studies in four ways. First, we use a large, national probability sample of noninstitutionalized older adults who were prospectively followed for four years. Second, we include a rather comprehensive array of psychosocial, physical, and functional health status measures, as well as prior health services utilization patterns. Third, we S173
Downloaded from http://geronj.oxfordjournals.org/ at Queen Mary, University of London on July 11, 2014
'Department of Medicine, Indiana University School of Medicine. 2 Regenstrief Institute for Health Care. 3 Richard Roudebush Veterans Administration Medical Center. "Department of Sociology, Kent State University.
S174
WOLINSKY ET AL.
frame our multivariate statistical modeling from the conceptual perspective of the behavioral model of health services utilization (Andersen, 1968). Finally, we identify those risk factors associated with dying after nursing home placement.
METHODS
The data are taken from the LSOA, which is a collaborative effort of the National Center for Health Statistics (NCHS) and the National Institute on Aging. The LSOA involves a special six-year follow-up to the 1984 Health Interview Survey (HIS). Because it is such a well-known data source, the HIS will not be discussed here. For the interested reader, a detailed description of its history, design, and logistics can be found in several readily available publications (NCHS, 1975, 1985). In 1984 the HIS was augmented by two special supplements. One involved the collection of detailed information on health insurance coverage among all respondents. The other (referred to as the Supplement on Aging [SOA]) in-
Downloaded from http://geronj.oxfordjournals.org/ at Queen Mary, University of London on July 11, 2014
Model The behavioral model of health services utilization was first presented by Andersen in 1968, and subsequently revised with his colleagues. Because it is the most widely used model for studying health services utilization (Wan, 1989), it need not be discussed here at length. Indeed, several detailed, historical reviews of the behavioral model already exist, including its application to the special case of older adults (e.g., Wolinsky, 1990). Because of its comprehensive approach, the behavioral model provides an excellent context within which to examine the effects of a variety of factors on nursing home placement and subsequent death. Basically, the behavioral model views the use of health services as a function of the predisposing, enabling, and need characteristics of the individual. The predisposing component is an abstraction from the proposition that some individuals have a greater propensity for using health services than do others. These propensities can be predicted from individual characteristics prior to an illness episode. The three dimensions of the predisposing characteristics include demographics, social structure, and health beliefs. Collectively, they represent the sociocultural element of the behavioral model. The enabling component is abstracted from the proposition that although the individual may be predisposed to use health services, he or she must nonetheless have some means for obtaining them. The enabling component, then, contains those factors which make health services available to the individual for consumption, and includes familial and community resources. These two dimensions of the enabling characteristics represent the economic component of the behavioral model. Although the predisposing and enabling components are necessary conditions for the use of health services, they are not sufficient ones. To use health services the individual must have or perceive some illness (or its possibility). This need has two dimensions. One represents the amount of illness that the individual perceives exists, and the other represents professionally evaluated need. These need indicators tap the individual's recognition that a health problem either exists or is in the making.
volved the collection of detailed information by way of a 30minute add-on interview concerning the health, social functioning, and living arrangements of 16,148 individuals aged 55 and older. A sample of 5,151 SOA individuals aged 70 years or more in 1984 was selected for follow-up interviews (by telephone if they had one, or by mail if they did not) in 1986, 1988, and 1990 (although the 1990 data are not yet available for public use). It is this sample, known as the LSOA, that is the source of data for our research. Greater detail on the design and execution of the LSOA can be found in Fitti and Kovar (1987). With the exception of nursing home placement and mortality status, all of the data used in the present analysis are based on self-reports given in the baseline interview, which involved household proxies in 8.0 percent of the cases and nonhousehold proxies in 1.6 percent of the cases. In this article we used the unweighted data. It has been empirically shown that the complex schemes necessary to take the disproportionately stratified multistage cluster sampling design of the LSOA into account have little impact on variance estimation (see Fitti and Kovar). Moreover, that impact is sufficiently attenuated by the inclusion of age and race as covariates in multivariate models to warrant analyses of the unweighted data (Fitti and Kovar, 1987). The measures of the predisposing, enabling, need, and health services utilization characteristics are generally straightforward, rather typical of the literature, have been described in detail elsewhere (Wolinsky and Johnson, 1991, 1992), and are summarized in Table 1. Accordingly, we shall not review them here in great detail. As indicated, there are 11 measures of the predisposing characteristics. These include age, sex, race, living alone, living in a multigenerational household, having a telephone, educational attainment, kin social supports, nonkin social supports, feeling a sense of control over one's health, and worrying about one's health. There are five measures of the enabling characteristics. These include having private insurance that covers at least some portion of hospital and physician charges, whether the respondent has a valid Medicaid card, residential stability, population density based on county adjacency codes, and sole financial dependence on public retirement sources of minimal magnitude (i.e., Social Security). As shown, there are 13 measures of the need for health services. Because of their traditionally pivotal role as risk factors for nursing home placement and mortality, they will be described at somewhat greater length. One is a dichotomous index of perceived health, indicating whether the respondent is in excellent, very good, or good health vs being in fair or poor health. Five of the remaining measures of health need are multiple-item scales that emerged from theoretically directed factor and principal components analyses of 21 questions routinely taken from or modeled after various measures of activities of daily living. We call the first the basic activities of daily living scale (basic ADL; minimum factor loading [the smallest correlation between an item in the scale and the underlying factor] = .718; alpha [an internal consistency based reliability coefficient with a lower limit of zero and an upper limit of unity] = .827). It consists of four items from the traditional ADL (Katz et al., 1963), including having any difficulties with such personal activities as bathing, dressing, getting out of bed, and
S175
NURSING HOME PLACEMENT AND DEATH
Table 1. Means, Standard Deviations, Coding Algorithms, and Psychometric Properties of the Baseline LSOA Data Coding Algorithms
SD
78.113 .640 .106 .378 .188 .969 9.755 1.618 2.347 .330 .697
5.917 .480 .308 .485 .391 .174 3.738 .644 1.338 .470 .460
iictual number of years = yes, 0 = no = yes, 0 = no = yes, 0 = no = yes, 0 = no = yes, 0 = no iictual number of years :'-item scale, 1 = has support, •>-item scale, 1 = has support, = worries about health, 0 = = has control over health, 0
.645 .056 .838 7.390
.478 .230 .369 2.526
.653
.476
= has private physician and hospital insurance, 0 = no = has Medicaid card, 0 = no = same address for 5 or more years, 0 = no 0 point 1980 county adjacency code, ranging from () = thinly populated not adjacent to 9 = core SMSA county = yes, 0 = no
Need Characteristics Perceived health Basic activities of daily living Household activities of daily living Advanced activities of daily living Lower body limitations Upper body limitations Ever had atherosclerosis Ever had valvular heart disease Ever had osteoporosis Ever had a fractured hip Ever had cerebrovascular disease Ever had cancer Ever had Alzheimer's disease
.660 .605 .656 .163 1.959 .442 .192 .025 .036 .050 .478 .120 .005
.474 1.222 1.167 .515 1.953 .821 .394 .156 .185 .218 .500 .326 .073
= excellent, very good, or good, 0 = fair or poor !5-item scale, 1 = has difficulty, 0 = no difficulty 4-item scale, 1 = has difficulty, 0 = no difficulty :5-item scale, 1 = has difficulty, 0 = no difficulty
