| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
RESEARCH ARTICLE |
a Department of Community Health Services, University of California, Los Angeles
b Department of Sociology, University of Maryland, College Park
c Department of Health Promotion and Gerontology, University of Texas, Medical Branch, Galveston
Carol S. Aneshensel, Department of Community Health Science, School of Public Health, Box 951772, University of California, Los Angeles, CA 90095-1772. Email: anshnsl@ucla.edu.
 |
Abstract |
|---|
 TOP Abstract Mortality Among the...Social Causation or Social...MethodsResultsDiscussionReferences |
|---|
Methods. Data from a multiwave panel survey of caregivers to persons with Alzheimer's Disease 
are analyzed with proportional hazard models of time from illness onset to death of the care recipient and, for those admitted to a nursing home 
, time from admission until death .
Results. Relocation is associated with a two-fold increase in mortality risk net of health status. Social selection effects were found for poor health, advanced age, being male, and being White. Patients admitted for reasons other than poor health also experienced elevated mortality immediately following admission, which is inconsistent with a social selection interpretation. However, none of the specific indicators of stressful admission or unsatisfactory nursing home conditions are significantly related to mortality.
Discussion. These data demonstrate selection processes for postadmission mortality, but indicate that the admission of patients in poor health may not fully account for the elevation in mortality that occurs immediately following admission.
RESIDENTIAL family care for frail elders is typically considered far more desirable than care provided within the confines of a formal organization. This conviction reflects cultural values supporting family ties, social norms mandating the obligations of family members for one another, and economic constraints impinging on individual families and society. Indeed, support from kin enables many elderly persons who would otherwise be placed in long-term care facilities to continue living in their communities (Day 1985
; Doty 1986). Escalating care demands, however, eventually prompt numerous caregivers to place their relatives in institutional care, often with great reluctance and lingering concern for the welfare of the placed elder. Contrary to conventional wisdom, most families are averse to institutionalization and continue to provide in-home care long after it is in their own best interests to cease (Brody 1985; Colerick and George 1986). Kelman and Thomas 1990, for example, describe the transition to institutional living as unwelcomed by elderly persons and their families and as accepted only as a last resort.
Are negative sentiments about long-term care institutions justified? Perhaps. A disquieting family portrait emerges from recent research: a picture that portrays the interests of those who give and those who receive care as competing, not served equally well by the alternatives of family and institutional care. Numerous studies now document that family care is hardly a "cost-free" alternative to institutional care. Instead, caregivers incur a broad spectrum of "hidden" burdens, including economic hardship, curtailment of social activities, emotional strain, and psychological distress (Cantor 1983; George and Gwyther 1986; Pearlin, Mullan, Semple, and Skaff 1990; Schulz, Visintainer, and Williamson 1990; Seltzer and Li 1996; Stone, Cafferata, and Sangl 1987; Strommel, Collins, & Given, 1994; Zarit, Todd, and Zarit 1986). Many, although certainly not all, of these burdens are eased following the transition from in-home to institutional care (Aneshensel, Pearlin, Mullan, Zarit, and Whitlatch 1995; Aneshensel, Pearlin, and Schuler 1993; Zarit and Whitlatch 1992). Yet, recent research indicates quite clearly that care recipients who are institutionalized have substantially higher mortality rates than do persons who continue to be cared for at home (Aneshensel et al. 1993, Aneshensel et al. 1995; Nygaard and Laake 1990; van Dijk, van de Sande, Dippel, and Habbema 1992; Wolinsky, Callahan, Fitzgerald, and Johnson 1992; Wolinsky, Stump, and Callahan 1997). Does transfer to a long-term care facility elevate the risk of death, or is high mortality merely an artifact of the selective admission of high-risk persons?
|
Mortality Among the Institutionalized Aged |
|---|
|
TOPAbstractMortality Among the...Social Causation or Social...MethodsResultsDiscussionReferences |
|---|
; Wolinsky et al. 1992, Wolinsky et al. 1997) and for persons with dementia in particular (Aneshensel et al. 1993, Aneshensel et al. 1995; van Dijk et al. 1992). Mortality rates are highest when closely following entry into a nursing home (Shah, Banks, and Merskey 1969; Shapiro and Tate 1988; van Dijk et al. 1992) and in the presence of select clinical and social psychological factors. Among the general institutionalized population, the following appear to be related to high mortality: cognitive impairment, severe mental abnormality, urinary incontinence, need for intense nursing, functional impairment, physical dependence, and poor physical mobility (Brauer, Mackeprang, and Bentzon 1978; Engle and Graney 1993; Goldfarb 1969; Kelman and Thomas 1990; Shapiro and Tate 1988). With regard to psychosocial factors, mortality is predicted by apathy, lack of social support, and social isolation (Diesfeldt, van Houte, and Moerkens 1986; Kelman and Thomas 1990; Wolinsky et al. 1992). Being of advanced age and being male are also associated with higher mortality in most of these studies, reflecting well established life-expectancy patterns in the general population.
Cognitive impairment, especially Alzheimer's Disease (AD), is a prime contributor to both nursing home admission and subsequent mortality among nursing home residents (Belloni-Sonzogni, Tissot, Tettamanti, Frattura, and Spagnoli 1989; Branch and Jette 1982; Diesfeldt et al. 1986; Goldfarb 1969; Greene and Ondrich 1990; Temkin-Greener and Meiners 1995; Vitaliano, Peck, Johnson, Prinz, and Eisdorfer 1981). The following behaviors appear to precipitate the institutionalization of persons with dementia: extreme forgetful behaviors (Pruchno, Michaels, and Potashnik 1990), incontinence, excessive irritability, inability to walk, wandering, hyperactivity, nighttime misbehavior (Knopman, Kitto, Deinard, and Heiring 1988), combativeness or angry outbursts (Chenoweth and Spencer 1986), and declining functioning (Scott, Edwards, Davis, Cornman, and Macera 1997). Postadmission mortality is related to the severity of the dementia and behavioral impairments, dependency, inactivity, physical disability, age and, possibly, age of onset and comorbidity (Diesfeldt et al. 1986; van Dijk, Dippel, and Habbema 1991; van Dijk et al. 1992; Walsh, Welch, and Larson 1990). Mortality may be elevated because the underlying disease process interferes with brain function, including risks associated with decreased cognition, memory, activity, and performance, or leads to comorbid conditions (Molsa, Marttila, and Rinne 1986; Harkness, Bentley, and Roghmann 1990; Thomas, Bennett, Laughon, Greenough, and Bartlett 1990; van Dijk et al. 1991, van Dijk et al. 1992). The natural course of AD involves significant excess mortality compared with normal aging, other psychiatric disorders, and possibly other forms of dementia (Belloni-Sonzogni et al. 1989; Diesfeldt et al. 1986; Martin, Miller, Kapoor, Arena, and Boller 1987; Molsa et al. 1986; Schoenberg, Okazaki, and Kokmen 1981; van Dijk et al. 1991, van Dijk et al. 1992; Varsamis, Zuchowski, and Maini 1972; Vitaliano et al. 1981). However, institutionalized persons with dementia experience higher mortality than do similarly impaired persons who remain in the community (Aneshensel et al. 1993, Aneshensel et al. 1995; van Dijk et al. 1992).
|
Social Causation or Social Selection? |
|---|
|
TOPAbstractMortality Among the...Social Causation or Social...MethodsResultsDiscussionReferences |
|---|
The social selection hypothesis possesses strong face validity because persons who enter long-term care institutions are hardly a random sample of the population of elderly persons. The empirical evidence is equivocal, however: Some studies report that elderly persons admitted to long-term care facilities are indeed in worse health than those who remain at home; other investigations find health differences to be of secondary importance to sociodemographic factors and social isolation (e.g., Aneshensel et al. 1993, Aneshensel et al. 1995; Branch and Jette 1982; Cox and Verdieck 1994; Freedman 1996; Jetter, Tennstedt, and Crawford 1995; Montgomery and Kosloski 1994; Morris, Sherwood, and Gutkin 1988; Shapiro & Tate, 1985; Wolinsky et al. 1992). However, even when health status is statistically taken into consideration, persons admitted to nursing homes are substantially more likely to die than are those who remain in the community (Aneshensel et al. 1993, Aneshensel et al. 1995; Wolinsky et al. 1992, Wolinsky et al. 1997). Thus, the selective admission of very ill persons seems to be insufficient to fully account for elevated mortality among institutionalized elders.
The social causation orientation is diametrically opposite to the selection orientation: Various conditions associated with placement and life in an institution influence whether and how long one continues to live. Two variants of the social causation hypothesis can be identified. Each mirrors dominant traditions in social stress research, one concerned with the deleterious health consequences of eventful life change and the other emphasizing ongoing stressful experience. Admission to a nursing home epitomizes the worst features of potentially health-threatening life events: Relocation completely disrupts the ordinary patterns of daily life, requiring readjustment of virtually all behavioral patterns, some of them life-long routines; nursing home admission is viewed almost universally as undesirable; the patient is likely to have little or no control over placement decisions, especially when health conditions impair cognitive functioning; moreover, the move may weaken or sever social connections to family and friends that might otherwise mitigate the full force of this trauma. The uncontrollable, undesirable, stressful event of relocation, then, could explain the elevation in mortality occurring around the time of relocation. The traumatic event perspective is less persuasive with respect to deaths that occur later, however, after a reasonable adjustment period has transpired. The explanation of these deaths, we believe, is more productively sought within ongoing conditions of institutionalized life, in particular, conditions exposing residents to undue bodily insult or diminishing resilience to normal encounters with infectious agents or injury. At issue is not the specific biological mechanism that causes death, but rather how social circumstances might shape whether such conditions are encountered, and if so, whether appropriate and timely treatment is received.
|
Methods |
|---|
|
TOPAbstractMortality Among the...Social Causation or Social...MethodsResultsDiscussionReferences |
|---|
). All data are from the caregiver interviews.
Sample
A total of 555 caregivers participated in the baseline interview in 1988: 300 in the San Francisco Bay Area and 255 in greater Los Angeles. A parent was the patient in 229 cases, and 326 spousal caregivers were interviewed. Subjects were recruited from lists of persons contacting local Alzheimer's Disease and Related Disorders Associations (ADRDA) in these two sites. Participants were primary caregivers, defined as the family member who provided the most care to the person with dementia. Although refusals were infrequent, it is not possible to calculate a meaningful response rate because the eligibility of most noninterviewed persons is unknown. The sample should not be construed as representative of all caregivers because it is self-selected. In particular, it underrepresents those who have not sought information or help from ADRDA, although this subgroup was accessed through secondary referrals (e.g., sons who contacted the agency on behalf of an elderly parent who was providing care). The sample entirely omits persons with dementia who are not cared for by a spouse or child, a group with an especially high risk of institutionalization.
However, the sample is heterogeneous in its characteristics. Most care recipients had been diagnosed with AD or senile dementia of the AD-type (74%); multiinfarct dementia (9%) and presenile, senile, or unspecified dementia (9%) were the next most common diagnoses. Three types of caregiver–patient dyads predominate: husbands caring for wives (25%), wives caring for husbands (34%), and daughters caring for mothers (29%). The majority of the patients are female (59%), as are the caregivers (69%). The majority of the patients were over age 70 at the onset of the study (Time 1; 77%), as were spousal caregivers (77%), whereas the majority of daughters and sons were under age 55 (66%). The caregiver sample is 84% White and, thus, underrepresents members of ethnic minority groups. Annual household incomes range from poverty to affluence, with a median income of $27,500.
A total of 272 patients were placed in long-term care institutions before Time 6: 155 (57%) subsequently died, 94 (35%) continued living through Time 6; and 23 (8%) caregivers were lost to follow-up while the patient was still living. Among those not institutionalized prior to Time 6 (N = 283), 136 (48%) died, 70 (25%) continued to receive in-home care through Time 6, and 77 (27%) were lost to follow-up while the patient was still living. Sample attrition is not associated with most sociodemographic factors, with the exception of ethnicity: Fewer White caregivers (16%) were lost than caregivers belonging to other ethnic groups (28%; p 
.05).
Sample attrition is substantially greater among caregivers who continued to provide in-home care (27%) than it is among those who institutionalized their relatives (9%, p .001), in part because of the high mortality among care recipients who were institutionalized; for this analysis, it is immaterial whether the caregivers for these persons were subsequently lost because the persons with dementia had already reached the "failure" event, death. In addition, concern about sample attrition is minimized in the hazard analysis (see below) because these subjects are retained in the analysis. Care recipients who were alive at the last observation—both those lost to follow-up and those living at home—are "censored" because follow-up was too short to observe their eventual deaths. Information about their survival through their last observation is retained in the analysis because these groups contain a disproportionate number of long-term survivors. From this perspective, those lost to follow-up are similar to those who were alive at the end of data collection. Sample weights to adjust for attrition are not appropriate because these subjects are included in the analysis through their last observation.
Measurement
There are two dependent variables in this analysis. The first outcome is illness duration, which is defined as the interval from illness onset to death of the person with dementia; it is assessed for all care recipients irrespective of residential status. Illness onset was measured at baseline with the following question: "How long ago did you realize that something was wrong with your (relative)?" It represents the perceived onset of AD, a recognition that occurs well after the true onset of the underlying disease, meaning that disease duration is longer than illness duration. The retrospective interval reported at baseline is incremented by time from the baseline interview to death. The date of death was ascertained in the bereavement interview with the question, "First, when did your (relative) pass away?" recorded as month and year. For persons living at their last follow-up, illness duration is not yet known. For this group, illness duration is incremented until the person's last observation, at which time the case is "censored"; this value is the month and year of the last interview completed with the caregiver. It is necessary to use this partial information to avoid bias due to the disproportionate exclusion of long-term survivors (Willett and Singer 1991; Yamaguchi 1991). It is not appropriate to simply compare those who are alive at last observation with those who have died, because the sample is heterogeneous with regard to the length of time they have been at risk of dying and because all of these persons will eventually die.
The second dependent variable is post admission mortality, which is necessarily assessed only for those who have been admitted to a nursing home. It is defined as the interval of time from relocation until death, calculated as the difference between the dates of these two events. Month and year of admission were obtained from either (a) the first institutional care interview for patients who were alive at the next annual caregiver interview or (b) the first bereavement interview for patients who had died before the next annual caregiver interview. For patients living at their last observation, the interval between relocation and death is not yet known. For this group, postadmission mortality is calculated as the number of months between relocation and the last interview with their caregiver; the case is censored at this time. This situation parallels that just described for illness duration.
Measures of health status are especially important because they address the core issue of whether poor health accounts for the high mortality of care recipients who are admitted to nursing homes. Four measures were used to operationalize health status, as shown in Table 1 . The caregiver's subjective assessment of the care recipient's health status represents the caregiver's overall assessment of the patient's health, encompassing not only dementia-related conditions but also comorbidity and frailty. Hospitalization can be a pivotal transition point, because dementia often appears to worsen in response to the strange environment, a potentially temporary reaction that can be misinterpreted as a permanent deterioration that necessitates admission to a nursing home, and because the physical removal of the person from the home can help the caregiver overcome emotional resistance to institutionalization (Aneshensel et al. 1995). Finally, for relatives admitted to a nursing home, caregivers were asked whether this relocation was the result of several catalysts, including the care recipient's poor health, as also shown in Table 1 . Other reasons are sorted into two categories: aspects of the dementia and poor health of the caregiver.
|
It should be noted that the caregiver's health reports are not limited specifically to the care recipient's dementia, but also encompass comorbid conditions and age-related frailty. This feature is important because the cause of death for persons with AD often is not the AD itself, but rather its complications or comorbid conditions. The analysis of selection effects should take into consideration all mortality risks that precede admission to a nursing home, not just risks specific to dementia.
The indicators of admission-related stress available in this study emphasize the impact of this event on the caregiver rather than the recipient. Although not capturing the patient's appraisal of the event, these variables do reflect the social psychological context of relocation, a key contributor to the experience of the patient. In any event, it is not realistic to obtain ratings from persons with advanced dementia.
As shown in Table 1 , we examined six transitional stressors. Two of these measures are prospective ratings obtained at each continuing care interview concerning the possible relocation of the care recipient in the future: caregiver distress at the thought of institutionalization and preparations for eventual placement. The analytic values of these variables are taken from the interview immediately preceding admission. Retrospective information about actual relocations was obtained at the first interview following this transition. These data are from the institutional care questionnaire or the bereavement questionnaire, depending on whether the care recipient died before the next regularly scheduled caregiver interview. Family disagreements are measured in this manner. Several additional indicators of transitional stress are taken from the institutional care interview, which is limited to patients who lived long enough for this caregiver interview to take place: worry about care recipient distress, concerns about placement, and problems in arranging a placement.
Finally, five ongoing stressors were assessed. Two of these measures were contained in both the institutional care and bereavement interviews and therefore pertain to all instances of institutionalization: the caregiver's assessment of the quality of (a) medical care and (b) nursing and attendant care. The other ongoing stressors are available from the institutional care interviews only: satisfaction and problems encountered with the nursing home facility or its staff and the amount of supplemental or "invisible" care provided by the caregiver.
Analytic Strategy
The data are analyzed using the Cox proportional hazard model for two outcomes: (a) illness duration (illness onset until death) for the full sample and (b) postadmission mortality (nursing home admission until death) for persons relocated to nursing homes. The data are reformatted into an event-history format with time counted as months from a transitional event (illness onset or nursing home admission) to either a failure event (death) or a censoring event (last observation). The hazard rate or hazard function [h(t)] is the risk of experiencing the failure event at time t given that the event did not occur before time t. It is analogous to an incidence rate or a conditional probability: the proportion of patients dying within an interval (among those living at the beginning of the interval). The hazard is the slope of the survival function, which is analogous to a cumulative prevalence distribution. The basic proportional hazard model is
 | (1A) |
 | (1B) |
 | (2A) |
i is a dummy variable for the ith subject, which takes the value 1 if the event occurred and the value 0 if the observation was censored at ti (Yamaguchi 1991, p. 106). When hi(t) from 1b is substituted into 2a, the baseline hazard function is canceled out between the numerator and the denominator, meaning that the PL function can be written solely as a function of parameters for covariates:
 | (2B) |
Thus, although the presence of a baseline hazard function that reflects the time dependence of hazard rates is assumed, its functional form need not be specified, making the proportional hazard model semiparametric (Yamaguchi 1991, p. 107). The hazard is expressed proportional to the baseline hazard, which also means that there is no intercept.
In this study, hi(t) is the hazard of individual i dying at time t—where time is counted as months from illness onset until death or months from admission to a nursing home until death—given that one is still at risk of dying at time ti (i.e., among persons who are still living at that time). Three types of covariates are analyzed. Fixed covariates are constants for the entire period of observation, such as gender. Other covariates, like declining cognitive functioning, change values over the course of observations. The values of these time-varying covariates are updated from their initial values when the patient lives longer than the next observation. Quasi-fixed covariates vary prior to institutionalization but do not change thereafter; an example is whether the caregiver finds the prospect of eventually institutionalizing his or her relative distressing. The time-varying and quasi-fixed covariates are lagged rather than contemporaneous covariates that change according to step functions of time (Petersen 1995). In this case, the time function corresponds to data collection points, except for institutionalization, which is calendar time.
For those admitted to a nursing home, we supplement the hazard model with a logistic regression analysis. In this analysis, the dependent variable is whether the care recipient died within the first 6 months following admission versus lived at least that long. The logistic model is used because this interval appears to be the peak period of risk for admission-related mortality (see below) and because thereafter the hazard is more or less constant.
|
Results |
|---|
|
TOPAbstractMortality Among the...Social Causation or Social...MethodsResultsDiscussionReferences |
|---|
). As shown in Model 1, Table 2 , which contains basic sociodemographic covariates only, the risk of dying increases as patient health status declines, as would be expected. Hospitalization in the previous year almost doubles this risk, and each decrement in physical health status, rated by caregivers in four categories ranging from poor to excellent, is associated with an increase in risk of about 25%. The severity of cognitive impairment elevates the risk of death over and above the contribution of poor physical health status and hospitalization. Advancing age also independently elevates mortality; male patients are substantially more likely to die than are female patients, and, there is also a mortality disadvantage for Whites. Of the variables considered in this analysis, only family income does not independently contribute to the risk of death, even though socioeconomic status is usually inversely associated with mortality in the general population.
|
Caregivers who institutionalize their relatives are substantially more likely to become bereaved than those whose relatives continue to reside at home. The zero-order odds of patient death more than double following admission to a nursing home 
. This elevation in risk is virtually unchanged (2.21) when physical health status and sociodemographic characteristics are statistically controlled, as shown in Model 2, Table 2 . This increment is over and above the contribution of health status, meaning that it cannot be attributed to either the caregiver's subjective appraisal of the health of the care recipient prior to institutionalization or to preadmission hospitalizations. Moreover, the observed impact of institutionalization is not an artifact of confounding with sociodemographic characteristics that are likely to influence the probability of being admitted to a nursing home.
In Model 3, Table 2 nursing home admissions are separated into admissions that caregivers attribute to poor patient health and those attributed exclusively to reasons other than poor health. The elevation in risk associated with institutionalization is especially pronounced among those whose admissions embody the selection argument (2.76). However, the risk of death is also significantly elevated among those whose admissions are seemingly incompatible with a selection interpretation (1.80). Although the increased risk for persons admitted for reasons other than poor health (relative to those who remain at home) is smaller than the increased risk for persons admitted specifically because they became ill, the elevation in risk is substantial for both groups. As in Model 2, the effects for the two types of admissions in Model 3 are independent of other indicators of care-recipient health and of sociodemographic characteristics usually associated with mortality.
The addition of the variable for relocation in Model 2 and addition of the reason for relocation in Model 3 does not alter appreciably most of the coefficients for the health and sociodemographic variables that comprise Model 1. The impact of hospitalization is reduced somewhat, suggesting that some of the deaths associated with hospitalization occur in the nursing homes to which hospitalized dementia patients are often discharged. Similarly, the coefficient for cognitive impairment is somewhat larger in Model 1 than in Model 2 or Model 3. This reduction suggests that some of the impact of institutionalization on mortality is in actuality an indirect effect of cognitive impairment. The addition to the model of nursing home admission and the reason for admission decreases somewhat the association between being male and dying, but increases somewhat the impact of being White. Some of the impact of institutionalization, then, reflects the mortality risks associated with these demographic characteristics.
The between-group analysis, therefore, provides some support for the social selection hypothesis. Model 3 is especially informative because it shows the exceptionally large increment in mortality that is directly attributable to the selective admission of persons in poor health. However, Model 3 also provides compelling evidence that contradicts a selection interpretation insofar as those who are institutionalized for reasons other than poor health also encounter a substantial elevation in risk compared with those who remain at home. Moreover, the impact of both types of nursing home admissions is over and above the impact of other indicators of poor health and sociodemographic characteristics that are known risk factors for a wide range of causes of death.
The Postadmission Life Course
Mortality varies considerably as a function of time since admission to a nursing home. The survival function (not shown) displays an immediate, rapid decline in the proportion of patients surviving, followed by slow but steady decline through the remaining period of observation. Taking into account censored cases, estimated survival rates for one, two, and three years postadmission are 66%, 54% and 47%. On the other hand, a quarter (26%, adjusted for censoring) do not survive the first 6 months. This descriptive picture is quite clear: If one survives the first few months of institutional life, then long-term survival is more likely than not; however, the chances of surviving these first few months are not good.
The hazard, which is the slope of the survival distribution, is shown in Fig. 1. As can be seen, the chances of dying are especially high immediately following admission and thereafter decline until reaching a plateau about 9 months after admission. This pattern is most pronounced among those admitted for poor health but is also present among those admitted for other reasons. These two strata converge after the first few months following admission. In other words, after the immediate postadmission period, the risk of dying in the next interval, given that one has survived up until the start of that interval, does not depend upon whether poor health was cited as a reason for institutionalization. These trajectories suggest that the period of maximum risk for admission-related mortality is limited to the immediate postadmission period and extends no longer than the first 6 months of institutional life.
|
Of the two time variables, only age at admission is associated with mortality, both in bivariate and in multivariate analysis. Preadmission illness duration does not independently contribute to the risk of postadmission mortality. The pattern for sociodemographic characteristics is similar to that reported above for time from illness onset until death (see Table 2 ). Specifically, the risk of dying is elevated among males and among Whites, but no income effect is evident.
Almost half of the caregivers specifically cited poor patient health as a reason for the relocation of their spouse or parent. As shown in Table 3 , their relatives faced almost twice the risk of death as patients whose caregivers cited only nonhealth-related reasons for admission. This difference is diminished substantially in multivariate analysis that takes health status and sociodemographic risk factors into consideration ( p .07), which validates the caregivers' reports of motivations for institutionalization. Two other sets of reasons also were analyzed—poor caregiver health and symptoms related to AD—but these measures did not significantly influence postadmission patient mortality (not shown).
|
.06), a trivial difference between models. In addition, reason for admission attains significance in the logistic regression, but was marginally nonsignificant in the hazard model ( p .07), again, a trivial difference between models. These results support the selection hypothesis because the risk of death is greatest among those in poor health. They are uninformative about the social causation hypothesis, however, because selection and causation are not necessarily mutually exclusive processes.
Traumatic Admission, Unsatisfactory Care, and Survival
Consequently, we turn now to the possibility that early postadmission mortality is evidence of processes of social causation, specifically processes indicative of acute and chronic stress. The basic hypothesis is that stressful admissions and unsatisfactory care should be associated with high patient mortality. As before, the dependent variable is whether the patient died within the first 6 months following admission.
All of the indicators of transitional stress (see Table 1 ) lack any discernible bivariate association with early postadmission mortality. Similarly, none of the variables tapping ongoing stressors are significantly related to the odds of early postadmission mortality. The detailed information about the nursing home available only from the institutional care interview should be viewed with caution, however, because analyses using this subset of variables necessarily omit the most problematic deaths, those occurring immediately after admission.
These results are consistent with a social selection argument insofar as the only factors related to mortality are indicators of poor health or generic mortality risks (i.e., older age and being male) and not indicators of transitional or ongoing stressors.
|
Discussion |
|---|
|
TOPAbstractMortality Among the...Social Causation or Social...MethodsResultsDiscussionReferences |
|---|
A second caveat pertains to the possible effects of undetected pathology on postadmission mortality. As discussed earlier, some persons suffer from undetected conditions that may result in their deaths. We argue that these conditions are likely to be present, but at similar rates among those who remain at home and those who are placed in institutions precisely because these conditions are undetected, meaning that they do not exert a manifest effect on the caregiver's admission decision. However, there remains the possibility that such conditions make a latent contribution to relocation, contributing to the caregiver's subjective appraisal of which environment is best for his or her relative. Some of this effect would seem to be captured by the caregiver's subjective rating of the care recipient's health, but this measure might not fully represent latent influences of undetected pathology. Furthermore, it is possible that these conditions become more lethal when the person is relocated from home to nursing home. For example, the life-long familial connection between the caregiver and the person with dementia may enable the family caregiver to detect forewarnings that the patient is ill that might be missed by a formal care provider, even one with advanced medical training. In other words, there may be an interaction between unobserved pathology, living arrangements, and mortality. It is not possible to directly test this possibility because the conditions are unobserved. However, the continuing impact of cognitive impairment on early mortality when admission for poor health is held constant is consistent with this possibility. Cognitive impairment portends difficulties in the patient's ability to interact with his or her physical and social environment, difficulties that may be compensated for by vigilant caregivers in home settings.
Nevertheless, we are unable to rule out the possibility that undetected pathology produces the results observed here, or more generally, that unobserved heterogeneity biases results. Unobserved heterogeneity refers to individual-level omitted variables. Zohoori and Savitz 1997 proposed the use of instrumental variables to eliminate confounding associated with unobserved heterogeneity. This approach is not feasible for this application, however, because the four exogenous variables (age, gender, ethnicity, and family income) have only weak correlations with the relevant instrumental variable, nursing home admission. Zohoori and Savitz also proposed using lagged endogenous variables as pseudo-exogenous variables, but this approach requires the assumption that measurement error is not serially correlated, an untenable assumption with self-report data. Thus, our results may be biased by unobserved heterogeneity, in particular the possibility of unmeasured pathology.
None of the social psychological factors we examined explained excess mortality among the institutionalized aged, but it is premature to conclude that the elevation in mortality is solely the result of selection factors. First, the indicators of acute and chronic stress examined here are but a limited portion of the universe of social factors that may contribute to premature death. Moreover, these indicators pertain to the experiences of the caregiver, which may be too peripheral to the conditions harmful to the patient. Also, these indicators are all perceptions of the environment as distinct from direct assessment of the environment. These perceptions may or may not be accurate. Finally, the conditions that trouble or dismay caregivers may not be the conditions relevant to understanding morbidity and mortality among patients.
One other measurement issue merits mention. Illness duration, a key dependent variable, is based on retrospective baseline reports of illness onset. This report underestimates actual disease onset and duration because the underlying disease is present long before symptoms are recognized. Moreover, these caregiver reports are subject to recall bias and are without doubt approximations. Many caregivers mark the onset of the illness by a discrete event that clearly indicated that something was seriously the matter with their spouse or parent. These markers are likely to be preceeded by signs and symptoms that were ignored or interpreted as something other than dementia. This limitation is most directly relevant to the point estimates of average illness, producing a probable underestimate of illness duration and imprecision in these estimates. It is less relevant to the analysis of factors related to mortality, we believe, because these reports were obtained at baseline, whereas all deaths occurred after the baseline interview. In other words, the subjectivity of the illness onset measure is more relevant to absolute rather than relative estimates of survival time.
Finally, the most convincing reason for continuing to consider a social causation perspective is the trajectory of mortality among those institutionalized for reasons other than poor health. Although the initial risk of death is substantially greater among those admitted for poor health, those in better health also evidence an initial elevation in mortality. Because this peak in mortality does not appear to be due to preexisting poor health, it does not lend itself to a social selection interpretation. In addition, cognitive impairment is associated with early mortality even when admission specifically for poor health is statistically controlled. This result suggests that sources of early postadmission mortality might productively be sought in aspects of the relocation that tax or exceed the understanding of the patient, his or her ability to adapt to a new environment, and difficulties in communicating one's needs, distress, or ill health. Our unwillingness to reject outright a social causation perspective notwithstanding, it must be emphasized that with the data available for our examination we observed no discernible evidence that social factors cause elevated mortality.
|
Acknowledgments |
|---|
Received for publication June 14, 1999. Accepted for publication November 4, 1999.
|
References |
|---|
|
TOPAbstractMortality Among the...Social Causation or Social...MethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  R. K. Scherer, N. Scarmeas, J. Brandt, D. Blacker, M. S. Albert, and Y. Stern The Relation of Patient Dependence to Home Health Aide Use in Alzheimer's Disease J. Gerontol. A Biol. Sci. Med. Sci., September 1, 2008; 63(9): 1005 - 1009. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. E. Gaugler, A. M. Pot, and S. H. Zarit Long-Term Adaptation to Institutionalization in Dementia Caregivers Gerontologist, December 1, 2007; 47(6): 730 - 740. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. Adamis, A. Treloar, F.-Z. Darwiche, N. Gregson, A. J. D. Macdonald, and F. C. Martin Associations of delirium with in-hospital and in 6-months mortality in elderly medical inpatients Age Ageing, November 1, 2007; 36(6): 644 - 649. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. S. Mittelman, W. E. Haley, O. J. Clay, and D. L. Roth Improving caregiver well-being delays nursing home placement of patients with Alzheimer disease. Neurology, November 14, 2006; 67(9): 1592 - 1599. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. J. McClendon, K. A. Smyth, and M. M. Neundorfer Long-term-care placement and survival of persons with Alzheimer's disease. J. Gerontol. B. Psychol. Sci. Soc. Sci., July 1, 2006; 61(4): P220 - P227. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R Portelli, D Lowe, P Irwin, M Pearson, and A. Rudd Institutionalization after stroke Clinical Rehabilitation, January 1, 2005; 19(1): 97 - 108. [Abstract] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
