INTRODUCTION
End-of-life care strives to honor terminally ill patients'
preferences regarding the way they die (Stewart et al.,
1999). Recently, Patrick et al. (2001)
defined quality of dying and death as the degree to which a person's
preferences for dying and the moment of death are consistent with how the
person actually dies. They identified one component of quality of dying
and death as dying in the place of one's choice. The actual place of
death can provide a measure of whether this proposed goal of end-of-life
care is being achieved. Therefore, place of death has been increasingly
recognized as an outcome to evaluate end-of-life care interventions (Hearn & Higginson, 1998; Higginson & Sen-Gupta, 2000; Cobbs, 2001).
Since 1977, researchers have investigated the determinants of place of
death for terminally ill cancer patients. Except for Hunt and colleagues
(Roder et al., 1987; Hunt et al., 1993, 2001), who
systematically tracked the trend and determinants of deaths at home,
hospital, hospice, and nursing home in South Australia over time, and,
recently, Gatrell et al. (2003), who analyzed
cancer deaths at different settings, the majority of published studies
compared either home versus “institutionalized” deaths by
combining hospital and inpatient hospice deaths together (Mor & Hiris, 1983; Higginson et
al., 1998; Costantini et al., 2000; Ryan, 2000; Gallo et al.,
2001) or deaths in hospitals versus deaths in other places (Pritchard et al., 1998; Bruera et
al., 2002). However, research showed that each place has unique
meanings for terminally ill cancer patients to choose as the place that
they would like to spend the rest of their life and to die (Fried et al., 1999; Tang,
2003a). Each health care setting also has its own mission,
philosophy, and practices. Undistinguishingly categorizing deaths at a
hospital, inpatient hospice, or nursing home as institutional deaths
cannot provide the whole picture about place of death and also cannot
retain the dying experiences of terminally ill cancer patients at each
setting.
Although freestanding facilities and dedicated units currently provide
only a small fraction (3%) of the hospice care now delivered in the United
States (National Hospice and Palliative Care
Organization, 2002), inpatient hospice services have always been
recognized as an important and essential element in the continuum of
services that make up a fully comprehensive hospice program (Doyle, 1998; Emanuel & Emanuel,
1998). Despite the strong emphasis on home-based end-of-life care
in the United States and the recognition of dying at home as a gold
standard of quality of care (Grande et al.,
1998; Cobbs, 2001), home hospice care is
not a panacea, and death at home may not be feasible for every terminally
ill cancer patient. Admission to an inpatient hospice at some point during
the dying trajectory may occur because dying there may become a necessary
and appropriate solution to distressing patients or exhausted families.
Understanding what factors predispose terminally ill cancer patients to
dying in an inpatient hospice may shed light on the future directions for
developing end-of-life care tailored to the needs of cancer patients who
are admitted to hospices and eventually die there. A prospective cohort
study was therefore conducted to investigate the determinants of dying in
an inpatient hospice for terminally ill cancer patients.
METHODS
Study Sample
Terminally ill cancer patients were recruited from six study sites in
Connecticut by a convenience sampling strategy. These study sites included
four tertiary care hospitals (Yale New Haven Hospital, VA Hospital,
Hospital of Saint Raphael, Hartford Hospital) and two home care programs
(VNA Home Care HOPE program and Hospice and Palliative Care of
Connecticut). The Yale University Human Subjects Research Review Committee
approved this study and the Institute Review Board of the six study sites
granted formal access to patients. Cancer patients were eligible for
participation if they (1) had a disease at a terminal stage based on the
clinical judgments of patients' primary physicians, (2) had a disease
that continued to progress with distant metastases and was unresponsive to
current curative cancer treatment, (3) were aged 21 years or older, (4)
knew the terminal status of their disease, and (5) could communicate in
English. Patient recruitment lasted from March 1 to December 15, 2001.
Subjects were continually followed until death or through March 31, 2004
(for participants who died during the study period, median survival time
after data collection was 52 days (range: 1–761 days). On average,
subjects were followed for 169 days (range: 1–883; median = 63
days). Written informed consent was obtained from all subjects.
Measures
Outcome Variable
The outcome variable was death in an inpatient hospice and was
considered censored for patients who were alive at the last day of
follow-up or who died at places other than an inpatient hospice. Inpatient
hospices included freestanding facilities and dedicated units located in
hospitals as defined by Lupu (1996) for the
National Hospice Organization. Information about place of death was
obtained from medical records, obituaries in local newspapers, or family
reports after the subjects' deaths.
Independent Variables
Factors influencing place of death as proposed by Mor and Hiris (1983) were organized into (1) sociodemographics, (2)
support network, (3) clinical condition and care needs, and (4) health
system factors. Preference of place of death was another independent
variable explored in this study as suggested by researchers as an
important factor related to place of death (Hinton,
1994; McWhinney et al., 1995; Karlsen & Addington-Hall, 1998; Cantwell et al., 2000). To elicit preferences of place
of death, patients were asked: “Some people prefer to die at home,
whereas other people prefer to die in a hospital, a hospice, or a nursing
home. Considering your current situation, where would you prefer to
be?”
Sociodemographic variables included gender, marital status,
age, race, and family financial status. Marital status was categorized as
married or unmarried. Race was dichotomized into native-born white
Americans or others based on statistics showing that the predominant users
of hospice care in the United States are non-Hispanic white people (National Hospice and Palliative Care Organization,
2002). Self-reported family income was characterized by three
categories: <$20,000, $20,000–$40,000, and >$40,000, each of
which contained about one-third of the income distribution.
Support network variables included living status and
availability and identity of primary family caregivers. The identity of
primary family caregivers was dichotomized into spouse and others.
Subjects were asked how supportive and how capable they perceived their
families were to help them achieve their preference of place of death
using a five-point Likert scale (1 = not at all supportive or
capable; 5 = very supportive or capable).
Disease-related variables, including diagnosis,
comorbidities, cancer-related treatment, and health care needs, were
obtained from the subjects' medical records and were supplemented by
self-reported information from subjects and their family caregivers when
necessary. Length of survival postinitial diagnosis was categorized as
<1 year, 1–5 years, and >5 years to reflect the
well-established cancer survival intervals. Health care needs at the time
of data collection interviews included home care services and hospice
care, advance directives, and do-not-resuscitate (DNR) preference. After
the data collection interview had been conducted, information on the use
of hospice care was retrieved from the patient's medical records.
Clinical conditions of subjects were self-reported at the
time of the data collection interviews as levels of symptom distress and
functional dependency. Symptom distress was measured by the Symptom
Distress Scale (SDS; McCorkle & Young,
1978), an instrument assessing 13 common symptoms of cancer
patients. Each symptom was rated by the patient on a scale ranging from 1
(normal or no distress) to 5 (extensive distress).
Responses to symptoms range from 13 to 65, with higher scores indicating
greater distress. Reported Cronbach's alphas of the SDS ranged from
0.72–0.89 for cancer patients and evidence of both concurrent and
predictive validity was reported in various studies on cancer patients
(McCorkle et al., 1998). In the current study,
the standardized Cronbach's alpha of the SDS was 0.75. Functional
dependency was measured by the Enforced Social Dependency Scale (ESDS;
Benoliel et al., 1980). The ESDS measures
personal and social competence. Personal competency includes six
activities: eating, dressing, walking, traveling, bathing, and toileting.
Dependency in each activity was reported by the patient and rated by the
interviewer on a 6-point scale. Social competence includes home, work, and
recreational activities, rated on 4-point scales, and communication, rated
on a 3-point scale. Total scores of the ESDS range from 10 to 51, with
higher scores reflecting greater dependency. Reported Cronbach's
alphas of the ESDS for various types of cancer ranged from 0.73 to 0.96.
Validity of the ESDS has been established by concurrent and predictive
validity, criterion validity, factor analysis, and responsiveness of this
instrument to detect the effectiveness of a home care intervention in
adults with lung cancer (McCorkle et al. 1989).
The standardized Cronbach's alpha of the ESDS in this study was
0.93.
Health-system factors used in this study included dummy
variables indicating residence in New Haven County and the availability of
acute inpatient beds and hospice care providers as proposed by Gallo et
al. (2001). The variable of residence in New
Haven Country was used to indicate residence in close proximity to the
largest university-based hospital and the only inpatient hospice and
Veterans Affairs (VA) hospital in Connecticut. The number of staffed acute
care hospital beds was used to measure the availability of inpatient
hospital care within 15 miles of the center of subjects' town of
residence. The number of hospice care programs within the same area
reflected the availability of hospice care. With one exception, these were
home care based services. Information on hospice services was obtained
from the 1997–8 Guide to the Nation's Hospices (National Hospice Organization, 1999) and data on
staffed acute adult care hospital beds were obtained from State of
Connecticut Office of Health Care Access Health System Data—acute
care staffed beds, FY 1998–2000 (State of
Connecticut, 2002). Both measures were dichotomized at their median
values for analysis.
Statistical Analysis
The study sample and the distribution of place of death were described
by means and standard frequency analyses. Bivariate analyses were
performed to assess the crude associations between likelihood of dying in
an inpatient hospice and various potential determinants including the
variables of sociodemographics, support network, clinical conditions and
care needs, and health-system related factors. The chi-square test and the
two independent sample Student's t test were used in the
bivariate analyses to declare the statistical significance of the crude
associations. For binary variables in which there were significant
differences between the inpatient-hospice death and the censored group,
unadjusted relative risk ratios (RR) were calculated to explore the
unadjusted associations between the outcome variable and the selected
independent variables in bivariate analyses.
Survival analysis by Cox proportional hazard model with backward
selection was chosen as the statistic to model the determinants of death
in an inpatient hospice to allow inclusion of those who survived at the
end of follow-up as censored observations. Only the variables that showed
significantly crude associations with likelihood of dying in an inpatient
hospice were included in the initial model. The effect of the individual
explanatory variable on the outcome variable was estimated by the
covariate adjusted hazard ratio (aHR) and the corresponding 95% confidence
intervals (CI). The significance level of the above analyses was set to be
5%, and the statistical analyses were conducted using Statistical Analysis
System (SAS) version 8.2.
RESULTS
Of 207 eligible patients, 180 patients (87% response rate) agreed to
participate. Disabling physical conditions (55.6%) and lack of interest
(25.9%) were the two primary reasons that potential subjects gave for
their nonparticipation in the study. Due to the restriction of access to
patients' medical records for those who refused to participate in the
study, no further comparison of characteristics of participants and
nonparticipants could be performed. Demographic characteristics of these
180 patients are listed in Table 1. At the end
of the study period, 164 (91.1%) subjects had died. The place of death for
12 subjects could not be identified; therefore, those subjects were
treated as censored observations. For those whose actual place of death
could be obtained, they most frequently died in an inpatient hospice
(n = 62, 40.8%), followed by in a hospital (n = 38,
25.0%), at home (n = 36, 23.7%), and in a nursing home
(n = 16, 10.5%).
Demographic and medical characteristics of the study
sample
Bivariate analyses showed that except for gender, the
inpatient-hospice-death group and the censored group were comparable on
all sociodemographic variables. For female terminally ill cancer patients,
the relative risk of dying in an inpatient hospice was 1.55 (95% CI =
1.02–2.36) times greater than male patients (Table 2).
Significant unadjusted associations between covariates and deaths in
inpatient hospices
There were no differences in variables related to living status and
the identity of primary caregivers between subjects who died in an
inpatient hospice and those in the censored group. Terminally ill cancer
patients who died in an inpatient hospice perceived significantly lower
levels of support (mean ± SD: 3.4 ± 1.4) for their
family to help achieve their preference of place of death than those who
did in the censored group (mean ± SD: 3.9 ± 1.4;
Table 3).
Discrete covariates with significant associations with likelihood of
death at an inpatient hospice
There were major differences in disease-related variables or care
needs between subjects who died in an inpatient hospice and those in the
censored group. Terminally ill cancer patients in the
inpatient-hospice-death group were significantly more likely to receive
hospice care services before death (Table 2).
Terminally ill cancer patients dying in an inpatient hospice scored
significantly higher on the ESDS and experienced significantly greater
symptom distress than those in the censored group (Table 3). No other different patterns of health care
needs and health care resources utilization were found between the two
groups in terms of use of home or hospice care or with advance directives
or DNR orders at the time of data collection.
Except for the variable of residing in New Haven County, health care
system factors (availability of acute inpatient beds and hospice care
providers) did not play a significant role in distinguishing terminally
ill cancer patients who died in an inpatient hospice from those who were
in the censored group. Subjects residing in New Haven County were about
twice (RR = 1.84, 95% CI = 1.18–2.87) as likely as their
counterparts to die in an inpatient hospice (Table
2).
Preference of place of death influenced the probability of dying in an
inpatient hospice. If terminally ill cancer patients expressed a
preference to die in an inpatient hospice, they had a 2.38 times (95% CI =
1.60–3.53) higher likelihood to die in an inpatient hospice than
those who preferred to die elsewhere (Table
2).
In summary, the results of bivariate analyses indicated that
terminally ill cancer patients who died in an inpatient hospice (1)
experienced higher levels of functional dependency and greater symptom
distress at the time of interview, (2) were referred to hospice care
services more frequently during the final days of their life, (3) were
more likely to live in New Haven County, (4) were more likely to be
female, (5) preferred to die in an inpatient hospice, and (6) perceived
their families as lacking the support to achieve their preferences of
place of death. These variables were examined by a multivariate Cox
proportional hazards regression model to select the determinants of dying
in an inpatient hospice.
Variables that were retained in the final best fitting multivariate
Cox proportional hazards regression model are presented in Table 4. Important determinants of deaths in an
inpatient hospice included receiving hospice care before death, inpatient
hospice as preferred place of death, higher levels of functional
dependency, and residence in New Haven County. The adjusted hazard ratio
showed that, if terminally ill cancer patients received hospice care
services before death, preferred to die in an inpatient hospice, and
resided in New Haven County, they had a 7.32 (95% CI: 3.21–16.67),
4.86 (95% CI: 2.24–10.51), and 1.70 (95% CI: 1.00–2.93) times,
respectively, higher likelihood than their counterparts to die in an
inpatient hospice. At the time of the interview, if terminally ill cancer
patients experienced higher levels of functional dependency, they were
also more likely to die in an inpatient hospice (aHR [95% CI] =
1.05 [1.02–1.08] with each unit increase in the score of
the ESDS).
Covariate-adjusted determinants of inpatient hospice death derived
from Cox proportional hazards model
DISCUSSION
Terminally ill cancer patients in this study most frequently died in
an inpatient hospice, followed by in a hospital, at home, and in a nursing
home. The prevalence of deaths in inpatient hospices was much higher than
national statistics (Higginson et al., 1998;
Hunt et al., 2001; National
Center of Health Statistics, 2002) and those reported from
large-scale studies (Gallo et al., 2001; Bruera et al., 2002). In South Australia, Hunt et al.
(2001) reported that from 1990 to 1999, 54.6% of
cancer patients died in a hospital, 19.9% in a hospice, 15.8% at home, and
9.7% in a nursing home. In England, Higginson et al., (1998) documented that from 1985 to 1994, 48.9% of
cancer patients died in a hospital (including National Health Service
[NHS] and non-NHS all types of hospitals and nursing homes),
16.7% in a hospice, 26.3% at home, and 3.1% at communal establishments
(data derived from the table and information provided in the article). In
the United States, according to the National Center for Health Statistics,
in 2000, for all causes of deaths, 50.0% of deaths occurred in a hospital,
22.7% at home, 22.3% in a nursing home, and 5.1% occurred elsewhere
(including inpatient hospices). For subjects in the Study to Understand
Prognoses and Preferences for Outcomes and Risks of Treatments (SUPPORT),
the distribution of place of death was as follows: 56% in an acute care
hospital, 25% at home, and 9% in either a nursing home or an inpatient
hospice (Pritchard et al., 1998). Gallo et al.
(2001) reported cancer deaths in Connecticut and
showed that 42% of the study sample died in a hospital, 29% at home, 17%
in a nursing home, and 11% in an inpatient hospice (only freestanding
hospices included). The high prevalence of inpatient hospice deaths for
terminally ill cancer patients in this study may be related to the local
health care system characteristics—a factor that has well been
documented as a powerful determinant of place of death (Pritchard et al., 1998; Wennberg
& Cooper, 1998). Over half (53.3%) of the study participants
resided in New Haven County in Connecticut. The first inpatient hospice in
the United State was founded in New Haven County and this institute is one
of the few inpatient hospices existing in the United States. The influence
of this inpatient hospice on the place of death for the study participants
is substantial. The State of Connecticut, or, specifically, New Haven
County, provides groundwork for a natural experiment on the impact of
institutionalized hospice care on directing end-of-life care away from
hospitals to inpatient hospices for terminally ill cancer patients.
Findings from this study make a valuable contribution to our understanding
of how inpatient hospice care can influence the place of death for
terminally ill cancer patients.
Reported relationships among age (Hunt et al., 1993, 2001; Gilbar & Steiner, 1996), race (Hunt et al., 2001), socioeconomic status (Sims et al., 1997; Hunt et al.,
2001), presence of family caregivers (Roder et
al., 1987; Dunphy & Amesbury, 1990),
and hematological malignancies (Hunt et al.,
1993; Gatrell et al., 2003), and the
likelihood of death in an inpatient hospice could not be detected in this
study. The positive association of female gender and inpatient hospice
deaths in bivariate analysis was consistent with the conclusion made by
Hunt et al. (2001). There was a significant
trend that more women than men were admitted to the inpatient hospice and
this finding was not surprising because women have traditionally assumed
caregiving roles at home more than men but have lacked caregivers when
they become sick. However, gender was not retained in the Cox proportional
hazards regression model. The negligible influence of sociodemographic and
social support network variables on home deaths asserted by Mor and Hiris
(1983) was also observed for inpatient hospice
deaths. Due to the lack of knowledge about impact of gender on the place
of death in an inpatient hospice from the existing studies, these data
support the need to look closer at this issue in future studies, including
the influencing of gender on the preference of place of death and health
of the caregivers (Bradley, 2003).
Findings from the bivariate analysis showed that perceived support of
their families to help achieve their preference of place of death was
inversely associated with the likelihood of dying in an inpatient hospice.
Consistent with the conclusion made from a systematic review of
preferences of place of death from terminally ill cancer patients (Higginson & Sen-Gupta, 2000), participants in this
study overwhelmingly indicated a preference of dying at home (Tang, 2003a). Further analyses indicated
that, when study subjects perceived their families having stronger support
and ability to help them achieve their preference of dying at home, they
were more likely to recognize dying at home as a feasible and achievable
goal and to use hospice home care services as a strategy to achieve their
preference of dying at home (Tang,
2003b). Consequently, the probability of dying in an
inpatient hospice was decreased; however, this variable was not
significantly related to site of death in the fully adjusted model.
In line with the literature that preference of dying at home
facilitates home death (Hinton, 1994; McWhinney et al., 1995; Lupu,
1996; Karlsen & Addington-Hall, 1998;
Cantwell et al., 2000), the findings of this
study suggest that terminally ill cancer patients who prefer to die in an
inpatient hospice have a greater likelihood of actually dying there.
However, the small number of subjects who indicated a preference for dying
in an inpatient hospice (n = 12, 6.7%) should be acknowledged.
The impact of a preference for dying in an inpatient hospice on inpatient
hospice deaths needs to be validated with a larger and representative
population.
The finding that terminally ill cancer patients who experienced higher
levels of functional dependency at the time of the data collection
interview were more likely to be admitted to an inpatient hospice and
eventually died there was consistent with the literature in regard to the
impact of accelerating symptom distress and dramatic deterioration of
physical function and clinical conditions on admission to inpatient
hospices (Mor et al., 1985; Addington-Hall et al., 1998). Recent studies also
observed that admitting to an inpatient hospice at the end of life
predisposes terminally ill cancer patients finally to dying there (Addington-Hall et al., 1998; Bruera
et al., 2000; Hogan et al., 2001). The
patient's functional status may have represented a demand for greater
levels of care. Home-based care may become nonviable when the patient
passes a threshold of significantly uncontrolled pain and with other
symptoms resulting in higher demands of care. If the only alternative to
home care is to receive care in a hospital, patients, families, and
physicians have no choice but to rehospitalize the patient. With the
availability of an inpatient hospice and the promise of providing
state-of-the-art end-of-life care, admission to an inpatient hospice and
dying there serve as an attractive third option to all stakeholders to
address the dilemma of preference for where to die.
Results from the Cox model indicated that receiving hospice care
services at the end of life increased the likelihood of dying in an
inpatient hospice. Hogan et al. (2001) suggested
that Medicare beneficiaries' likely place of death was strongly
associated with use of hospice care. In addition, Ryan (2000) observed that the presence of dedicated
inpatient hospice units increased the likelihood of inpatient deaths
(including hospital and inpatient hospice deaths). All of the referred
hospice services in this study were affiliated with either a freestanding
hospice or dedicated hospice units. Availability and ease of access to the
parent organization's inpatient hospice services might encourage
their use and predispose subsequent inpatient hospice deaths.
A major limitation of this study was related to the generalizability
of the findings due to the convenience sampling of subjects from six study
sites in a single state. In addition, this study did not explore family
caregivers' preferences of place of death for the patient as well as
the impact of caregiving burden on inpatient hospice deaths. The numerous
distressing symptoms, a great burden of sustained personal care, and
tremendous financial strain posed by advanced cancer may overwhelm the
capability of family caregivers to take care of dying patients at home and
therefore may result in admitting patients into inpatient hospices.
Despite these limitations, there are substantial implications of this
study for end-of-life care and health policy. For terminally ill cancer
patients who choose an inpatient hospice as their preferred place of
death, dying and death in an inpatient hospice is in accord with their
wishes—an important component of a “good death” and high
quality of dying and death as defined by the Institute of Medicine (1997) and Patrick et al. (2001), respectively. At a time when everything seems
gradually to fall apart, the ability to achieve the preference of dying at
a desirable place may breed a feeling of empowerment. Every effort should
be made to develop effective interventions and to modify health care
systems to help terminally ill cancer patients spend their final days of
life and eventually die at a place that they prefer.
Deaths in inpatient hospices may represent a shift of end-of-life
care; this shift represents dying away from acute care hospitals to an
institutionalized hospice. Considering the well-known unsatisfactory
quality (McCarthy et al., 2000) and high cost
(Scitovsky, 1994) of end-of-life care provided
in acute care hospitals, inpatient hospices may present a potential for
improving quality of end-of-life care and controlling the spiraling costs
at the end of life. More research is needed to compare the outcomes,
including quality and costs of care for patients with different
constellations of needs that are provided in an acute hospital and
inpatient hospice. Consistent findings across several studies are
warranted before the promotion of large-scale development of inpatient
hospice programs.
For the majority of terminally ill cancer patients who prefer to die
at home (Higginson & Sen-Gupta, 2000; Ryan, 2000) and live in an area where a variety of
end-of-life care services are available but who experience high levels of
symptom distress and functional dependency, does the presence of an
inpatient hospice serve as a deterrent to hospice home care referral or
does an inpatient hospice serve as a viable option? Further research is
needed to investigate whether this pattern of admission is appropriate
given the symptoms, dependency levels, and social situation of patients.
More research efforts need to focus on illustrating whether the inpatient
and home hospice care programs serve different populations and can be
regarded as strategies that are directed toward distinct sets of problems.
Knowledge gained from such research will guide the planning of end-of-life
care to develop each program to best serve the needs of its particular
constituency.
