Customize your JAMA Network experience by selecting one or more topics from the list below.
Steinhauser KE, Christakis NA, Clipp EC, McNeilly M, McIntyre L, Tulsky JA. Factors Considered Important at the End of Life by Patients, Family, Physicians, and Other Care Providers. JAMA. 2000;284(19):2476–2482. doi:10.1001/jama.284.19.2476
Author Affiliations: Program on the Medical Encounter and Palliative Care (Drs Steinhauser and Tulsky), Center for Health Services Research in Primary Care (Drs Steinhauser, McIntyre, and Tulsky), and Geriatric Research Education and Clinical Center (Dr Clipp), Durham Veterans Affairs Medical Center, Department of Medicine (Drs Steinhauser, Clipp, and Tulsky), Center for the Study of Aging and Human Development (Drs Steinhauser, Clipp, McNeilly, and Tulsky), and Institute on Care at the End of Life (Dr Tulsky), Duke University, and the Institute for Multiculturalism (Dr McNeilly), Durham, NC; and the Departments of Medicine and Sociology, University of Chicago, Chicago, Ill (Dr Christakis).
Context A clear understanding of what patients, families, and health care practitioners
view as important at the end of life is integral to the success of improving
care of dying patients. Empirical evidence defining such factors, however,
Objective To determine the factors considered important at the end of life by
patients, their families, physicians, and other care providers.
Design and Setting Cross-sectional, stratified random national survey conducted in March-August
Participants Seriously ill patients (n = 340), recently bereaved family (n = 332),
physicians (n = 361), and other care providers (nurses, social workers, chaplains,
and hospice volunteers; n = 429).
Main Outcome Measures Importance of 44 attributes of quality at the end of life (5-point scale)
and rankings of 9 major attributes, compared in the 4 groups.
Results Twenty-six items consistently were rated as being important (>70% responding
that item is important) across all 4 groups, including pain and symptom management,
preparation for death, achieving a sense of completion, decisions about treatment
preferences, and being treated as a "whole person." Eight items received strong
importance ratings from patients but less from physicians (P<.001), including being mentally aware, having funeral arrangements
planned, not being a burden, helping others, and coming to peace with God.
Ten items had broad variation within as well as among the 4 groups, including
decisions about life-sustaining treatments, dying at home, and talking about
the meaning of death. Participants ranked freedom from pain most important
and dying at home least important among 9 major attributes.
Conclusions Although pain and symptom management, communication with one's physician,
preparation for death, and the opportunity to achieve a sense of completion
are important to most, other factors important to quality at the end of life
differ by role and by individual. Efforts to evaluate and improve patients'
and families' experiences at the end of life must account for diverse perceptions
Dying patients confront complex and unique challenges that threaten
their physical, emotional, and spiritual integrity. The Study to Understand
Prognosis and Preferences for Outcomes and Risks of Treatments (SUPPORT) documented
that many patients die prolonged and painful deaths, receiving unwanted, expensive,
and invasive care.1 Patients' emotional suffering
at the end of life can be profound, yet physicians are too frequently ill
equipped to address this suffering.2,3
In response, medical societies, health care organizations, and the public
have identified improved end-of-life care as a high national priority. The
American Medical Association and the Institute of Medicine have outlined goals
for improved care of the dying, and The Robert Wood Johnson Foundation has
devoted millions of dollars to public education on this issue through the
Last Acts initiative.4-6
These efforts depend, in part, on certain presumptions regarding how
dying patients and their families define quality at the end of life. During
the latter part of the 20th century, advances in biomedical technology propelled
us to see a "good" death as one involving the fight against disease. Partly
in response to this view, the modern hospice movement emerged, redefining
a good death as one that included acceptance and closure, most often at home.
Unfortunately, empirical support for a notion of a good death that might best
structure end-of-life care is lacking, as is a comprehensive understanding
about how the definition of a good death might vary across relevant constituencies.
Empirical evidence defining a good death would assist efforts to improve
end-of-life care by documenting the breadth of preferences of dying patients
and their families. Such data would provide clinicians with information to
help guide patients through this challenging and uncertain time. Therefore,
we investigated what patients, family members, physicians, and others consider
to be important attributes at the end of life.
This study was a cross-sectional, stratified random national survey
of seriously ill patients, recently bereaved family members, physicians, and
other care providers (nurses, social workers, chaplains, and hospice volunteers).
Patients were randomly selected from the national Veterans Affairs (VA)
Patient Treatment File database using International Classification
of Diseases, Ninth Revision diagnostic codes for a variety of advanced
chronic illnesses (lung, colon, gastric, esophageal, pancreatic, head and
neck, and lymphatic cancer; end-stage renal disease; advanced chronic obstructive
pulmonary disease; and congestive heart failure). All patients had been hospitalized
for these diagnoses within the prior year. Family members were randomly selected
from national samples of relatives of VA patients who had died 6 months to
1 year earlier. We chose this period so that family members would be past
the immediate stages of grief, yet the death would not be so distant that
the risk of retrospective bias would be introduced.7-10
Patient and family samples reflected the racial/ethnic and socioeconomic composition
of VA patients. However, we oversampled female patients (20%).
Individuals involved in end-of-life care were randomly selected from
membership lists of national professional associations (American College of
Physicians–American Society of Internal Medicine, American Nurses Association,
National Association of Social Workers, Association of Professional Chaplains,
and National Hospice Volunteers). The sample composition reflects the demographic
profile of each association's membership.
We mailed surveys to 500 subjects from each of the 4 groups (total potential
n = 2000). Sample size was calculated to provide adequate power to detect
differences among groups. We used several well-established techniques to maximize
response rates and data quality.11,12
Participants who did not respond to the initial survey within 5 weeks received
a second survey that included return postage. In the first wave only, we provided
a nominal financial incentive. Survey completion time was less than 15 minutes.
The institutional review boards of the Durham VA and Duke University Medical
Centers, Durham, NC, approved the study.
The survey asked respondents to rate the importance of 44 attributes
of experience at the end of life (survey available at
Survey items were generated on the basis of 12 previously conducted
focus groups and in-depth interviews with patients, family members, physicians,
and other care providers in which participants were asked to define attributes
of a good death.13 Participants rated the importance
of each item on a 5-point scale: strongly disagree, disagree, neither agree
nor disagree, agree, and strongly agree. Respondents also were presented with
the 9 items most frequently identified in the focus groups and asked to rank-order
them from 1 (most important) to 9 (least important). The rank sum for each
respondent for the 9 items was constrained to be 45; tied ranks were permitted.
We examined the distribution of responses for all 44 items, including
frequency, mean, median, and range. For descriptive parsimony, we collapsed
the 5 response categories into 3: agree, disagree, and neither agree nor disagree.
Based on natural breaks in the distribution of the data, items for which more
than 70% of respondents in all 4 groups chose "agree" or "strongly agree"
were identified as having substantial agreement. We used χ2
and, when appropriate, Fisher exact tests of independence to compare responses
among groups. We also assessed mean rank score for the 9 ranked items and
conducted Friedman tests to examine whether rankings within groups were different
than would be expected by chance alone. Wilcoxon tests were used to examine
the significance of specific response differences among groups.
To identify independent correlates of 4 selected attributes with the
most varied ratings of importance, we conducted multivariate exploratory analyses
of factors associated with item response. We used logistic regression to model
the likelihood of responding "disagree" or "neither agree nor disagree" vs
"agree" for a given item. We used a stepwise procedure to identify covariates
strongly associated with response. Our final models also included variables
in which we had a substantive interest, regardless of their precise significance
level. Each question was initially evaluated in the pooled sample using the
same set of covariates. We present odds ratios (ORs) and 95% confidence intervals
(CIs) only for significant covariates. Because income and education were strongly
associated with role, socioeconomic status effects on the full sample could
not be evaluated; hence, the effects of income and education on patient and
bereaved family member responses were tested in separate analyses. Exploratory
analyses revealed no differences in patients' responses by diagnosis; therefore,
it was excluded from the multivariate models. Because only female patients
were oversampled, we present results of unweighted analyses. In analyses not
shown, we also modeled items as a 3-category response variable, using multinomial
logistic regression. Because the results were nearly identical, we report
the less cumbersome binary models. Statistical analyses were conducted using
SAS Version 8.0 software (SAS Institute Inc, Cary, NC).
Of the 2000 mailed surveys, 1885 potential subjects could be reached
(ie, had not moved or died). We received responses from 340 (77%) of 444 patients,
332 (71%) of 465 bereaved family members, 361 (74%) of 486 physicians, and
429 (88%) of 490 other care providers(120 chaplains [96%], 105 hospice volunteers
[84%], 107 social workers [86%], and 97 nurses [78%]), for a total of 1462
participants. The average age of respondents was 57 years; 51% were men and
82% were white (Table 1a). Nonrespondents
did not differ from respondents with regard to sex, race/ethnicity, diagnosis,
or geographic location.
Based on responses to the survey, we classified the 44 attributes into
3 categories: (1) items with strong agreement regarding importance among the
entire sample (>70% of every group agreed that the attribute is important);
(2) items with strong agreement regarding importance (>70%) among patients
but less agreement among physicians; and (3) items with broad response variation
among the entire sample (ie, large percentages of respondents agreeing, disagreeing,
and neither agreeing nor disagreeing that the item is important).
Attributes Rated As Important Among All Participants.
Twenty-six survey items displayed strong agreement in all 4 groups (Table 2). Of the 26 items, 5 were associated
with symptoms or personal care: freedom from pain, freedom from anxiety, freedom
from shortness of breath, being kept clean, and having physical touch. Four
items related to preparation for the end of life: having financial affairs
in order, feeling prepared to die, believing that one's family is prepared
for one's death, and knowing what to expect about one's physical condition.
Three items related to achieving a sense of completion about one's life: saying
good-bye to important people, remembering personal accomplishments, and resolving
unfinished business. Two items involved decisions about treatment preferences:
having treatment preferences in writing and naming someone to make decisions
in the event that one cannot. Seven items were associated with what focus
group participants in a prior study13 called
"being treated as a whole person": maintaining one's dignity, maintaining
a sense of humor, having a physician who knows one as a whole person, presence
of close friends, not dying alone, and having someone who will listen. Finally,
5 items were linked to patients' relationships with health care professionals:
receiving care from one's personal physician, trusting one's physician, having
a nurse with whom one feels comfortable, knowing that one's physician is comfortable
talking about death and dying, and having a physician with whom one can discuss
Attributes Important Among Patients.
The second category included 8 items that were consistently rated as
important among patients (>70%) but were significantly less important to physicians
3). These included being mentally aware, having funeral arrangements
planned, feeling that one's life was complete, not being a burden to family
or society, being able to help others, coming to peace with God, and praying.
These differences persisted after conducting multivariate analyses controlling
for sex, race/ethnicity, socioeconomic status, household composition, religion,
and religiosity (data available on request).
Attributes With Broad Response Variation Among All Participants.
A final category comprised 10 items with a broad distribution of responses
among the 4 groups. These items included attributes relating to treatment
preferences, preparation, and completion or spirituality (Table 4). For example, the groups showed wide response variation
regarding the importance of knowing the timing of death. A slight majority
of the sample agreed with the importance of meeting with a clergy member,
having a chance to talk about the meaning of death, and discussing spiritual
beliefs with one's physician. However, a sizable percentage of each group
disagreed or neither agreed nor disagreed about the importance of these items
(Table 4). Compared with patients,
bereaved family members more frequently agreed with and physicians less frequently
agreed with the importance of meeting with clergy.
Multivariate models were created controlling for role, sex, race/ethnicity,
income, education, religion, religiosity, being present during the last hour
of someone's life, household composition, self-reported health status, and,
for patients, diagnosis. Only ORs that were significant are presented herein.
Use of All Available Treatments.
Physicians (OR, 0.1; 95% CI, 0.1-0.2) and other care providers (OR,
0.08; 95% CI, 0.04-0.14) were significantly less likely than patients to agree
with the importance of using all available treatments no matter what the chance
of recovery, whereas bereaved family members were equally likely to agree
(OR, 0.8; 95% CI, 0.6-3.3). African American (OR, 3.3; 95% CI, 2.0-4.0) and
other nonwhite ethnic groups (OR, 2.5; 95% CI, 1.4-3.3) were significantly
more likely than white participants to agree with the importance of using
all available treatments. Persons who had not been present during the last
hour of another person's life were also more likely to agree (OR, 1.7; 95%
CI, 1.0-2.5). Sex, religion, and the other variables were not associated with
the response to this question. Among patients and bereaved family members,
respondents with more education (bachelor's degree, OR, 0.5; 95% CI, 0.3-0.8;
graduate/professional degree, OR, 0.4; 95% CI, 0.2-0.9 vs no college) and
higher annual income ($20,000–$40,000, OR, 1.0; 95% CI, 0.6-1.7; ≥$50,000,
OR, 0.3; 95% CI, 0.2-0.8 vs <$20,000) were significantly less likely to
agree with the importance of use of all available treatments.
Controlling Time and Place of Death.
Religiosity was the only covariate significantly associated with preference
for controlling the time and place of death. Participants who considered faith
or spirituality not at all important were significantly more likely (OR, 1.7;
95% CI, 1.1-2.0) than were those who considered it very important to agree
with the importance of such control.
Dying at Home.
Other care providers (OR, 1.7; 95% CI, 1.1-2.0) were significantly more
likely to agree with the importance of dying at home compared with patients.
Physicians (OR, 1.4; 95% CI, 1.0-2.0) and bereaved family members (OR, 0.8;
95% CI, 0.6-1.3) were not significantly different from patients. Roman Catholic
(OR, 1.4; 95% CI, 1.0-2.0) and "other" respondents were significantly less
likely than Protestants (the reference point) or Jews (OR, 0.8; 95% CI, 0.5-1.3)
to disagree. Separate logistic analyses showed no significant covariates among
patients and families.
Talking About the Meaning of Death.
Physicians (OR, 2.0; 95% CI, 1.3-2.5), other care providers (OR, 1.7;
95% CI, 1.1-2.0), and bereaved family members (OR, 1.7; 95% CI, 1.1-2.7) were
significantly more likely than patients to agree that talking about the meaning
of death is important. Those for whom faith or spirituality was not at all
(OR, 0.3; 95% CI, 0.2-0.5) or somewhat (OR, 0.4; 95% CI, 0.3-0.6) important
were significantly less likely than those for whom spirituality was very important
to agree that this attribute was important. Among patients and family members,
women were significantly more likely to agree with talking about the meaning
of death (OR, 2.0; 95% CI, 1.3-2.5).
We measured the mean rank scores for the 9 preselected attributes, with
1 being most important and 9 being least important (Table 5). Friedman tests were significant (P<.001),
suggesting that the rankings by each group were different than would be expected
by chance alone. Freedom from pain was ranked as most important (ie, received
the lowest mean score) by patients (3.07), bereaved family members (2.99),
physicians (2.36), and other care providers (2.83). Coming to peace with God
and presence of family were ranked second or third in importance in all groups.
For patients and families, the difference between the ranking of freedom from
pain and being at peace with God was trivial (0.09 and 0.12 difference, respectively),
suggesting that these items are nearly identical in importance for both groups.
Physicians' mean score difference between the items was 2.46; other care providers
had a difference of 0.88 in mean score (P<.001).
Of note, dying at home received the least important relative ranking by all
groups except other care providers, who ranked it second to last.
Our results reveal areas of strong agreement and variation among end-of-life
care participants' definitions of what constitutes a good death. More than
half of the survey items showed consensus among all 4 groups. For example,
in concert with previous findings in the palliative care literature, survey
participants overwhelmingly endorsed pain and symptom management.14-16 Regardless of role,
respondents also converged on the importance of preparation for the end of
life. These findings echo the results of a recent study that showed that many
patients wish to plan ahead for their own deaths and support the importance
of prognostication in clinical practice.17-20
Additionally, respondents expressed a strong preference for having an opportunity
to gain a sense of completion in their lives. Life review, saying good-bye,
and resolving unfinished business provide both patients and their families
with an opportunity for human development at the end of life.21
Finally, all groups advocated strong relationships between patients and health
care professionals that emphasized more than just the patient's disease.
Results of this study also highlight one of the challenges of comprehensive
end-of-life care: attending to aspects of care that are not intuitively important
to clinicians but are critical to patients and their families. For example,
in contrast with physicians, patients strongly endorsed the importance of
being mentally aware. When forced to choose between attributes (Table 5), patients ranked pain control higher than mental awareness;
however, the mean rank difference was only 1.51. In contrast, the average
difference between the same items among physicians was 3.76, suggesting physicians
may be more willing than patients to sacrifice lucidity for analgesia. Similarly,
other care providers generally emphasize what patients need to receive, but
our results indicate that being able to help others is central to patients'
conceptions of quality at the end of life.13
Finally, patients highly valued attention to spirituality; in particular,
the importance of coming to peace with God and praying. Rank-ordered responses
showed that coming to peace with God and pain control were nearly identical
in importance for patients and bereaved family members.
Perhaps the most interesting findings of our study are items for which
there was broad response variation within and across all groups. They serve
as a reminder that there is no one definition of a good death; quality end-of-life
care is a dynamic process that is negotiated and renegotiated among patients,
families, and health care professionals, a process moderated by individual
values, knowledge, and preferences for care. We choose to illustrate this
point with discussion of 4 critical issues raised in the survey.
Consistent with previous research,22,23
African Americans had higher odds than white participants of wanting all available
treatments, which may reflect a preference for life-sustaining treatment or
distrust of the predominantly white medical culture.22
The disagreement by physicians and other care providers with use of all available
treatments may reflect greater familiarity with life-sustaining treatments.
In one study, patients were less likely to want cardiopulmonary resuscitation
after receiving additional information about the procedure.24
In addition, despite many patients valuing use of all treatments, most disagreed
with the importance of being connected to machines. In contrast, physicians
equate these interventions and disagreed with both.
Respondents displayed broad variation in their desire to control time
and place of death. Those with less religiosity were most likely to want control.
Given the strong public support for the hospice movement and its emphasis
on home care, we expected to find overwhelming preference for dying at home.25 However, fewer than half of all participants in our
sample agreed that this was an important attribute in quality of dying. Moreover,
dying at home was consistently ranked least in importance among 9 selected
attributes. Religion and role were associated with a preference for dying
at home. Recently, Fried et al demonstrated a similar preference among older
adults.26 The notion of dying at home may be
romantic among health care professionals who want to provide a good death.
However, as symptoms accelerate in the last 24 to 48 hours, some patients
and families may feel overwhelmed by concerns about symptom control or a dead
body in the home and, therefore, prefer a skilled care environment.26 Therefore, although for many patients an appropriate
goal is to allow them to die at home, this should not be assumed.
While we anticipated that religiosity and female sex would be associated
with a desire to talk about the meaning of death, we did not expect that physicians,
other care providers, and bereaved family members would be more likely to
agree with its importance than were patients. Similarly, patients as a group
were least likely to rate discussing personal fears as important. All groups
lacked consensus in assigning importance to meeting with clergy and discussing
spiritual beliefs with one's physician. Spirituality, however, was clearly
important to patients, as illustrated by their strong consensus surrounding
the need for coming to peace with God and praying. These findings suggest
that for some patients, issues of faith that are resolved with oneself are
more important than social or interpersonal expressions of spirituality.27
This study has several limitations. Patients and family members were
recruited from VA medical centers; therefore, generalizations to other groups
should be made cautiously. However, participants' preferences reflected death
in a variety of settings, and patients and families participating in the study
represented broad age, educational, and socioeconomic ranges. The individuals
comprising other care providers are a diverse group and are not necessarily
expected to form a cohesive whole. However, given their role in end-of-life
care, it is important to determine their viewpoints, although future studies
should evaluate differences within these groups.
The results of this survey suggest that for patients and families, physical
care is expectedly crucial, but is only one component of total care. Whereas
physicians tend to focus on physical aspects, patients and families tend to
view the end of life with broader psychosocial and spiritual meaning, shaped
by a lifetime of experiences. While physicians' biomedical focus is a natural
outgrowth of medical care that emphasizes the physical self, physicians should
recognize patients' other needs and facilitate means for them to be addressed.
Physicians also should recognize that there is no one definition of a good
death. Quality care at the end of life is highly individual and should be
achieved through a process of shared decision making and clear communication
that acknowledges the values and preferences of patients and their families.
Patients, families, and care providers each play a critical role in
shaping the experience at the end of life. As our cultural lexicon of death
and dying expands, further research is needed to define both the common ground
and areas for negotiation as participants gather to construct quality at the
end of life. A challenge to medicine is to design flexible care systems that
permit a variety of expressions of a good death.