Casarett D, Karlawish J, Morales K, Crowley R, Mirsch T, Asch DA. Improving the Use of Hospice Services in Nursing HomesA Randomized Controlled Trial. JAMA. 2005;294(2):211-217. doi:10.1001/jama.294.2.211
Author Affiliations: Center for Health Equity
Research and Promotion at the Philadelphia Veterans Affairs Medical Center,
Philadelphia, Pa (Drs Casarett and Asch); the Division of Geriatric Medicine
(Drs Casarett and Karlawish), Center for Bioethics (Drs Casarett, Karlawish,
and Asch), Leonard Davis Institute of Health Economics (Drs Casarett, Karlawish,
and Asch), and the Center for Clinical Epidemiology and Biostatistics (Dr
Morales) at the University of Pennsylvania, Philadelphia; Temple University
Law School, Philadelphia, Pa (Ms Crowley); and Holy Redeemer Home Health and
Hospice Care, Philadelphia, Pa (Ms Mirsch).
Context Hospice care may improve the quality of end-of-life care for nursing
home residents, but hospice is underutilized by this population, at least
in part because physicians are not aware of their patients’ preferences.
Objective To determine whether it is possible to increase hospice utilization
and improve the quality of end-of-life care by identifying residents whose
goals and preferences are consistent with hospice care.
Design, Setting, and Participants Randomized controlled trial (December 2003-December 2004) of nursing
home residents and their surrogate decision makers (N=205) in 3 US nursing
Intervention A structured interview identified residents whose goals for care, treatment
preferences, and palliative care needs made them appropriate for hospice care.
These residents’ physicians were notified and asked to authorize a hospice
Main Outcome Measures The primary outcome measures were (1) hospice enrollment within 30 days
of the intervention and (2) families’ ratings of the quality of care
for residents who died during the 6-month follow-up period.
Results Of the 205 residents in the study sample, 107 were randomly assigned
to receive the intervention, and 98 received usual care. Intervention residents
were more likely than usual care residents to enroll in hospice within 30
days (21/107 [20%] vs 1/98 [1%]; P<.001 [Fisher
exact test]) and to enroll in hospice during the follow-up period (27/207
[25%] vs 6/98 [6%]; P<.001). Intervention residents
had fewer acute care admissions (mean: 0.28 vs 0.49; P = .04
[Wilcoxon rank sum test]) and spent fewer days in an acute care setting (mean:
1.2 vs 3.0; P = .03 [Wilcoxon rank sum
test]). Families of intervention residents rated the resident’s care
more highly than did families of usual care residents (mean on a scale of
1-5: 4.1 vs 2.5; P = .04 [Wilcoxon rank
Conclusion A simple communication intervention can increase rates of hospice referrals
and families’ ratings of end-of-life care and may also decrease utilization
of acute care resources.
At least 1 in 4 Americans dies in a nursing home,1 and
considerable evidence indicates that nursing home residents do not receive
optimal end-of-life care. Approximately 25% of residents with daily cancer
pain receive no pain medications,2 and residents
are often transferred to an acute care setting to receive aggressive treatment
in the last weeks of life.3,4 It
is not surprising, therefore, that families often express dissatisfaction
with the end-of-life care that their loved ones receive in nursing homes.5
Hospice care has been associated with improvements in these and other
outcomes. For instance, nursing home residents who receive hospice care are
more likely to have their pain assessed, receive better pain management, and
have lower rates of inappropriate medications and physical restraint use.6,7 Families perceive that hospice improves
nursing home care,8 and hospice is associated
with improved satisfaction in other settings as well.5
However, only 1 in 4 residents enrolls in hospice care before death.9 Although hospice is not the right choice for all nursing
home residents, it is likely that more residents and their families could
benefit from hospice care.10 We hypothesized
that more residents might enroll in hospice care if physicians, residents,
and families were encouraged to discuss hospice as an option for care. Therefore,
the goals of this study were to determine whether an intervention that promotes
communication about hospice care can increase hospice enrollment and improve
the quality of care that residents receive at the end of life.
This study recruited residents and their surrogates from 3 nursing homes
selected for the diversity of their resident populations. Site 1 is an urban
facility with a high proportion of African American residents; site 2 is a
suburban facility with a largely white, affluent population; and site 3 is
a Veterans Affairs (VA) nursing home with an ethnically diverse population.
Sites 1 and 2 contract with outside hospice organizations to provide end-of-life
care, and site 3 uses an in-house hospice program. The intervention was applied
throughout all units of sites 1 and 2. In site 3, where virtually all residents
are male, 2 units (out of 4) were selected to participate to preserve a representative
sex distribution in the sample. This study was approved by the Philadelphia
VA institutional review board (site 3), and by the University of Pennsylvania
institutional review board on behalf of sites 1 and 2. Participants were enrolled
and followed up between December 2003 and December 2004.
All residents in a unit at the time of the initial chart review were
eligible. Residents were excluded if they were admitted for a respite stay,
if they were already receiving hospice care, or if the resident was too cognitively
impaired to complete the interview and did not have a surrogate. Letters sent
to each resident’s surrogate described the study and gave them the opportunity
to refuse participation by calling a toll-free number. Residents underwent
cognitive screening using the orientation items from the Mini-Mental State
Examination (MMSE)11 to determine their ability
to participate. When residents could not answer screening questions or could
not answer questions to assess their understanding during the consent process,
a surrogate was invited to participate instead. Both residents and surrogates
provided oral informed consent.
Clinical and Demographic Characteristics. Two
research assistants interviewed the resident in person and the surrogate by
telephone. Interviews assessed resident and surrogate demographic characteristics,
the resident’s deficits in activities of daily living,12 and
cognitive function using the MMSE.11 To better
define the representativeness of the sample, residents and surrogates were
asked to define their ethnicity using predefined categories. Medical charts
were reviewed for current medications, type of insurance, and existing diagnoses,
which were used to calculate the resident’s Charlson comorbidity score.13 Chart review also identified existing orders to limit
life-sustaining treatment (eg, do not resuscitate [DNR] orders) and advance
directives or chart documentation of preferences to limit such treatment in
Hospice Appropriateness. For all residents
and surrogates in the study in both the usual care and intervention groups,
scripted interview questions assessed the resident’s appropriateness
for hospice care. Residents were determined to be appropriate for hospice
care if either the resident or surrogate: (1) expressed goals for care that
focused on comfort; (2) refused both cardiopulmonary resuscitation and mechanical
ventilation; and (3) identified at least 1 need for palliative care. Prognosis
was not used to determine study eligibility because in nursing homes, as in
other settings, even the best prognostic models are often inaccurate.14,15
We defined goals for care using a single question adapted from the SUPPORT
(Study to Understand Prognosis and Preferences for Outcomes and Risks of Treatment)
study.16 This question offered a choice between
a course of life-prolonging treatment (“A course of treatment that focuses
on extending life as much as possible. But with this course of treatment you
might have more pain and discomfort.”) and palliative care (“A
course of treatment that focuses on relieving pain and discomfort as much
as possible. But with this course of treatment you might not live as long.”).
Single items assessed preference regarding cardiopulmonary resuscitation and
mechanical ventilation. When a resident or surrogate could not decide on a
single answer for these questions, we inferred a preference for life-sustaining
treatment, which is the default in most health care settings.
We assessed 10 needs for symptom management using the Global Distress
Index (GDI) of the Memorial Symptom Assessment Scale.17 The
GDI assesses the presence and severity of 6 common physical symptoms (pain,
constipation, lack of appetite, lack of energy, drowsiness, dry mouth) and
4 common psychological symptoms (feeling sad, worrying, feeling nervous, feeling
anxious). We also assessed 8 needs for palliative care services, corresponding
to core hospice services18 that previous research
has shown that patients and families value at the time of hospice enrollment19,20: (1) additional nursing support;
(2) physician care focused on comfort; (3) practical support with personal
care needs; (4) help with advance care planning; (5) counseling and emotional
support; (6) bereavement support for family members; (7) spiritual support;
and (8) visits from a volunteer to provide company for the resident.
Intervention assignment was determined by blocked randomization using
a random-number list generated by the study statistician (K.M.), stratified
by site of care. Research assistants revealed each resident’s assignment
at the beginning of the interview, assigning each resident and his/her surrogate
to the same treatment group. Participants were not informed of their assignment
until the end of the interview, when they were asked for permission to share
the results of the interview with the physician. Research assistants could
not be blinded to the resident’s treatment assignment because they were
required to assign residents and surrogates to the same treatment group. However,
they determined hospice appropriateness by reading a script verbatim and applying
a priori criteria (described below), making it unlikely that their knowledge
of treatment assignment could influence the participant’s responses.
The intervention was designed to help physicians identify those residents
whose goals, preferences, and palliative care needs would make them appropriate
for hospice care. This strategy is suggested by previous studies that have
shown that physicians are often not aware of their patients’ preferences,21- 25 and
that simple interventions can improve communication.26- 31 In
the intervention group, the results of resident and/or surrogate interviews
were sent by fax to the resident’s physician when either the resident
or surrogate met all 3 criteria for appropriateness for hospice care. These
faxes described the study aims and informed them that an interview had identified
one of their patients who might be appropriate for hospice care. We did not
require agreement between residents and surrogates because we reasoned that
the resident’s physician would be best able to reconcile resident’s
and surrogate’s responses when they differed. Physicians were asked
to reply by fax, indicating whether the resident had a prognosis of 6 months
or less and, if so, whether nursing home staff should arrange a hospice visit.
In the usual care group, hospice appropriateness was assessed in the
same way, but interview results were not communicated to the resident’s
physician. Instead, all residents and families were given a brief description
of hospice and hospice services. In addition, hospice-appropriate residents
and families were also told that they could learn more about hospice care
by speaking with the resident’s health care team.
Residents were followed up for 6 months or until death. Medical records
were reviewed biweekly to identify deaths, hospitalizations, new orders to
limit life-sustaining treatment, and hospice enrollment. When residents died
during the follow-up period, the surrogate was asked to complete a brief interview
approximately 2 months after the resident’s death. These interviews
used items adapted from the Toolkit Afterdeath Survey,32 a
widely used instrument to measure families’ perceptions of the quality
of end-of-life care. The surrogate evaluated the resident’s care in
the last week of life on a scale from 1 to 5 (poor, fair, good, very good,
excellent). Surrogates were also asked whether a member of the health care
team discussed the resident’s prognosis with them and what to expect
during the dying process, whether the resident had pain or shortness of breath
that was not managed as well as it could have been, and whether the resident
died where he/she would have wanted.
Baseline differences and final outcomes of the intervention and usual
care groups were evaluated using the Wilcoxon rank sum test or χ2 test. Mortality rates were compared using Kaplan-Meier survival curves
and the log-rank test. Sample size was planned to provide at least 80% power
to detect a 15% absolute difference in 30-day hospice referral rates (0% vs
15%) and a 0.5-point difference in overall ratings of end-of-life care (1-5
scale; SD = 0.5; assuming a 15% mortality at 6 months) (α = .05
for each comparison). Stata software version 8.0 (StataCorp LP, College Station,
Tex) was used for all statistical analyses.
Of 400 eligible residents, 14 died and 3 were discharged before an interview
could be completed, 14 were already receiving hospice care, and 26 could not
complete the interview and had no surrogate (Figure). Of the remaining 343 residents, 138 residents or their
surrogates refused, leaving a total sample size of 205 (60%). Participating
residents were similar to all residents at the 3 sites in terms of age, sex,
ethnicity, presence of a DNR order, primary insurance, and Charlson comorbidity
score. Participating residents were also similar to the national population
of nursing home residents with respect to age, sex, and ethnicity, but were
less likely to have Medicaid (24% vs 55%) and less likely to have a DNR order
(37% vs 53%) (Table 1).33
Thirty-two interviews (16%) were conducted only with the resident, 45
interviews (22%) were conducted with both the resident and his/her surrogate,
and 128 interviews (62%) were conducted only with the resident’s surrogate.
Residents in the study received care from a total of 33 primary physicians
(range: 1-39 residents per physician). Residents in the intervention (n = 107)
and usual care (n = 98) groups were similar with respect to site
of care and other clinical and demographic characteristics (Table 1).
The section of the interview that determined hospice appropriateness
typically required between 5 and 10 minutes to administer to residents, and
somewhat less for surrogates. There were fewer hospice-appropriate residents
in the intervention group than in the usual care group (35/107 [33%] vs 49/98
[50%]; P = .01 [χ2]) (Table 2). Surrogates in the usual care group
were more likely to express goals of comfort care, and residents in the usual
care group were more likely to refuse life-sustaining treatment.
Faxes were sent to 26 physicians responsible for the 35 residents in
the intervention group who were hospice appropriate. Physicians did not respond
to repeated faxes for 6 patients (17%). Others declined to authorize a referral
because they believed the resident did not have a prognosis of 6 months or
less (n = 2 residents [6%]) or would not benefit from hospice care
(n = 2 residents [6%]). Of the remaining physicians who authorized
a hospice visit for a resident (25/35 [71%]), 4 residents did not enroll because
the hospice team believed they did not have a prognosis of 6 months or less.
The remaining 21 residents enrolled in hospice. In the intervention group,
all referrals in the first 30 days were for residents whose physician agreed
to a hospice visit.
Hospice referral rates at 30 days were significantly higher in the intervention
group (21/107 [20%] vs 1/98 [1%]; P<.001 [Fisher
exact test]) (coefficient, 3.16; 95% confidence interval [CI], 1.34-5.19)
(number needed to treat [NNT] = 5). The single resident referred
from the control group had been found to be hospice appropriate during the
interview, and the referral was made after the family contacted the physician
directly. A similar effect was observed in all 3 interview categories, although
the smallest subgroup (resident only) did not reach statistical significance
(all results using Fisher exact test: resident only, 6/19 [32%] vs 0/13 [0%], P = .06; surrogate only, 9/69 [13%] vs 1/59 [2%], P = .02; resident and surrogate, 6/19 [32%] vs
0/26 [0%], P = .003). Similar proportions
of hospice-appropriate residents in the intervention group were referred to
hospice across the 3 sites (site 1: 5/26 [19%]; site 2: 10/65 [15%] ; site
3: 6/16 [38%]; P = .16 [Fisher exact test]).
By the end of the 6-month follow-up period, 33 residents (16%) had enrolled
in hospice care. Intervention residents were more likely to enroll in hospice
than were usual care residents (27/107 [25%] vs 6/98 [6%]; P<.001 [χ2]). Intervention residents also had fewer
acute care admissions (mean [SD]: 0.28 [0.70] vs 0.49 [0.89]; P = .04 [Wilcoxon rank sum test]) and spent fewer days in
an acute care setting (mean [SD]: 1.17 [3.32] vs 2.99 [6.07]; P = .03 [Wilcoxon rank sum test]) during the follow-up period
(Table 3). Almost all of the admissions
in the intervention group occurred when the resident was not enrolled in hospice
(70/78 [90%]). Residents with longer lengths of stay in hospice had fewer
admissions (Spearman ρ = −0.38; P = .04)
and spent fewer days in an acute care setting (Spearman ρ = −0.38; P = .03).
Twenty-three residents (11%) died during the 6-month follow-up period.
Mortality rates were similar in the intervention and control groups (15/107
[14%] vs 8/98 [8%]; P = .18 [χ2]). Kaplan-Meier survival curves were also similar in the 2 groups
(P = .20 [log rank test]). Six-month survival
was similar for those residents who were determined to be hospice appropriate
and those who were not (8/84 vs 15/121; 9% vs 12%; P = .52
Intervention residents received more days of hospice care than usual
care residents (mean: 64 vs 14 days; P = .01
[rank sum test]). However, intervention residents were not more likely to
be enrolled in hospice at the time of death (5/14 [36%] vs 3/8 [38%]; P = .93 [χ2]) nor were they more
likely to die in the nursing home rather than in an acute care setting (9/15
[60%] vs 5/8 [62%]; P >.99 [Fisher exact test]).
Residents in both groups who were enrolled in hospice care at the time of
death were more likely to die in the nursing home rather than in an acute
care setting (8/9 [89%] vs 6/14 [43%]; P = .04
[Fisher exact test]).
Of the 23 residents who died during the follow-up period, 3 surrogates
could not be reached for a second interview, 2 refused, and 1 had died. A
retrospective interview was completed with 17 surrogates (74%) approximately
2 months after the resident’s death. Of these, 4 of the 11 intervention
residents and 3 of the 6 control residents died in hospice (P = .64 [Fisher exact test]). Surrogates’ ratings of
the care that residents received in the last week of life were significantly
higher in the intervention group (mean [SD]: 4.3 [1.01] vs 2.2 [1.47]; P = .01 [Wilcoxon rank sum test]) (coefficient,
3.33; 95% CI, 0.69-5.97) (Table 3).
Surrogates of residents with a longer length of stay in hospice rated the
resident’s end-of-life care more highly (Spearman correlation coefficient,
0.75; P = .05).
The results of this study have 5 implications for efforts to improve
the care of the growing numbers of older adults who die in nursing homes.
First, we found that by increasing access to hospice care we were able to
improve families’ ratings of the care that residents received at the
end of life. This finding is consistent with those of an early controlled
trial34 and a more recent case-control study5; however, the early trial was conducted in a single
health care setting, and the interpretation of case-control studies of hospice
care is limited by selection bias because enrollment in hospice is influenced
by a variety of patient characteristics that may affect satisfaction. In contrast,
the results reported herein provide novel evidence that increased use of hospice
can improve families’ ratings of end-of-life care.
Second, this intervention could feasibly be implemented in most long-term
care settings. Unlike several previous studies that have evaluated intensive
or multistep communication interventions,35,36 this
intervention can be administered in several minutes by any member of the health
care team. The brief scripted interview and simple criteria to determine hospice
appropriateness ensure that this intervention can be used for an unselected
resident population in virtually any long-term care setting, with minimal
staff training. These results add to growing evidence that simple interventions
may offer some of the most promising opportunities to improve communication
about goals and preferences.27,28,37
Third, this intervention was efficient, resulting in hospice referral
for approximately 20% of residents within 30 days. The effect size of this
intervention, and its NNT, compare favorably with other patient- and family-focused
interventions.26,37 In fact, it
is possible that the intervention’s true effects are underestimated
by the results reported herein because residents in the intervention group
were less likely to be hospice appropriate. In addition, some physicians cared
for residents in both groups, creating contamination that could have increased
rates of hospice enrollment in the usual care group. Together, these factors
suggest that the true effect of the intervention may have been even larger
than that observed.
Fourth, this study suggests that the benefits of hospice care may be
greater for those patients who are referred earlier. Indeed, the principal
effect of this intervention may be earlier referral rather than greater rates
of hospice use. In this study, families of residents in the intervention group
were more satisfied with care in the last week of life, even though intervention
residents were not more likely to receive hospice care at the time of death.
This suggests that the improvements in families’ perceptions of care
observed in the intervention group were due to earlier hospice enrollment
and longer lengths of hospice service. This interpretation is consistent with
the finding that intervention residents received more days of hospice care
and that a longer period of hospice care prior to death was associated with
higher family ratings of the quality of care in the resident’s last
week of life. These results are consistent with those of several descriptive
studies that have found associations between length of stay in hospice and
provision of services38 and families’
perceptions of the value that hospice offers.39 These
findings also provide data to support the recommendations of experts in the
field,40- 43 referring
clinicians,44 and government agencies,45,46 that some patients should be enrolling
in hospice care sooner.
Finally, this study found that intervention residents were admitted
to the hospital less frequently and spent fewer days in an acute care setting.
Previous research has also found that residents receiving hospice care spend
less time in an acute care setting.4 This finding
has important implications for efforts to improve the quality of end-of-life
care in nursing homes. Because of the discontinuity of care that transfers
produce,47 there is general agreement that
residents should be able to die in the nursing home without transfers, hospitalizations,
and aggressive interventions, if that is consistent with their goals and preferences.10,48- 51 In
addition, a reduction in hospitalizations could have important policy implications
because care in the last year of life constitutes approximately 30% of total
Medicare expenditures.52 Further research is
needed to determine whether efforts to increase rates of hospice referral
in nursing homes can offer cost savings as well as improved care.
This study has 3 main limitations. First, all 3 nursing homes had their
own hospice programs or relationships with community hospice programs. Nursing
homes face numerous barriers in forming relationships with community hospices,10 and few have developed their own hospice programs.
This intervention would likely prove to be less effective in a nursing home
without these resources. Second, diagnostic data resident characteristics
were extracted from the medical record and therefore may be inexact. For instance,
MMSE testing revealed that not all residents with cognitive impairment carried
a diagnosis of dementia. Third, this study did not examine processes of care
during the follow-up period. Therefore, it was not possible to determine how
the intervention changed care, and how hospice enrollment changed the care
that residents received. Although it is clear that this intervention was successful
in improving families’ perceptions of the quality of care that residents
received, it cannot shed light on the mechanisms of this effect. Nor is it
clear whether the intervention alone was entirely responsible for the observed
reduction in hospitalizations. It is highly plausible that this effect was
due primarily to increased rates of hospice enrollment in the intervention
group, particularly since almost all admissions occurred when residents were
not enrolled in hospice. However, it is also possible that subtle differences
in baseline characteristics may also have contributed to the observed difference.
The results reported herein show that a simple communication intervention
can improve the quality of end-of-life care and decrease resource utilization
by promoting earlier access to hospice care in nursing homes. If there is
a surprise in these findings, it is that such a simple, inexpensive, and easily
exportable intervention can contribute so much to improving care. As Americans
spend more time in nursing homes near the end of life, it will become increasingly
important to emphasize simple, low-cost interventions like this one that can
help to ensure that residents and their families have access to the best possible
quality of care.
Corresponding Author: David Casarett, MD,
MA, 3615 Chestnut St, Philadelphia, PA 19104 (email@example.com).
Author Contributions: Dr Casarett had full
access to all of the data in the study and takes responsibility for the integrity
of the data and the accuracy of the data analysis.
Study concept and design: Casarett, Karlawish,
Acquisition of data: Casarett, Karlawish, Crowley,
Analysis and interpretation of data: Casarett,
Karlawish, Morales, Asch.
Drafting of the manuscript: Casarett.
Critical revision of the manuscript for important
intellectual content: Casarett, Karlawish, Morales, Crowley, Mirsch,
Statistical analysis: Casarett, Karlawish,
Obtained funding: Casarett.
Administrative, technical, or material support:
Study supervision: Casarett, Mirsch.
Financial Disclosures: None reported.
Funding/Support: Dr Casarett is the recipient
of an Advanced Research Career Development Award from the Department of Veterans
Affairs and a Paul Beeson Physician Faculty Scholars Award.
Role of the Sponsors: Neither funding agency
had any role in the design and conduct of the study; in the collection, management,
analysis, and interpretation of the data; or in the preparation, review, or
approval of the manuscript.
Acknowledgment: We thank Thomas ten Have, PhD,
for critical review of the manuscript and are grateful for the advice and
assistance of staff at participating nursing homes including Darlene Pidgeon,
CRNP, Ernestine Steiner, CRNP, Patricia Chriss, CRNP, Bette McNee, RN, NHA,
Robin Frankwich, MEd, NHA, Barry Fabius, MD, and Joan Weinryb, MD, and the
residents, families, and physicians who participated.