Customize your JAMA Network experience by selecting one or more topics from the list below.
Wenger NS, Kanouse DE, Collins RL, et al. End-of-Life Discussions and Preferences Among Persons With HIV. JAMA. 2001;285(22):2880–2887. doi:10.1001/jama.285.22.2880
Author Affiliations: Division of General Internal Medicine and Health Services Research, Department of Medicine (Drs Wenger, Liu, and Shapiro), and Departments of Pediatrics and Health Services (Dr Schuster), University of California, Los Angeles; RAND Health, Santa Monica, Calif (Drs Kanouse, Collins, Schuster, Bozzette, and Shapiro); and Department of Medicine, San Diego VA Healthcare System, and the University of California, San Diego, La Jolla (Drs Gifford and Bozzette).
Context Deficits in advance care planning leave many patients and their physicians
unprepared for decisions about end-of-life care. Even though the prognosis
has improved for many persons with human immunodeficiency virus (HIV) infection,
a need for planning remains.
Objective To evaluate prevalence of end-of-life discussions, use of advance directives,
and preferences concerning end-of-life care and their relationship with patient
demographics, clinical status, psychosocial variables, and practitioner characteristics
among HIV-infected persons.
Design, Setting, and Patients Cross-sectional survey of a US probability sample of 2864, which represents
231 400 adults receiving care for HIV, conducted from January 1996 to
Main Outcome Measures Communication with physician regarding end-of-life issues, completion
of an advance directive, preference for aggressiveness of care, and willingness
to tolerate future permanent adverse health states.
Results A total of 1432 patients (50%) discussed some aspect of end-of-life
care with their practitioner and 1088 (38%) completed an advance directive.
Patients were more likely to complete an advance directive after a physician
discussion (odds ratio [OR], 5.82; 95% confidence interval [CI], 4.50-7.52).
Practitioners discussed end-of-life care less with blacks (OR, 0.57; 95% CI,
0.39-0.83) and Latinos (OR, 0.74; 95% CI, 0.55-0.98) than with whites. Women
(OR, 1.39; 95% CI, 1.05-1.84) and patients with children in the household
(OR, 1.53; 95% CI, 1.12-2.10) communicated the most with practitioners about
end-of-life issues. Patients infected with HIV via injection drug use (OR,
0.64; 95% CI, 0.45-0.89) and those with less education communicated the least
with physicians about end-of-life issues. Less denial, greater trust in one's
practitioner, and longer patient-practitioner relationship were associated
with more advance care planning.
Conclusions Half of all persons infected with HIV are at risk of making end-of-life
decisions without prior discussions with their health care practitioners.
Blacks, Latinos, intravenous drug users, and less educated individuals need
advance care planning interventions in clinical HIV programs.
With the dramatic improvement in treatment modes that are available
for human immunodeficiency virus (HIV) infection and the decrease in mortality
among persons with HIV, less attention has been focused on end-of-life treatment
decisions for these individuals. However, HIV remains a prominent cause of
death among US individuals aged 18 to 35 years, and despite the drop in mortality
across the age spectrum, new treatment modes for many patients with HIV may
yield long-term survival with severe disability or incapacity. Additionally,
HIV demographics are shifting toward nonwhites and intravenous drug users,
groups with which health care practitioners communicate less well.1 Moreover, throughout many underdeveloped countries,
where newer HIV therapeutics are unavailable, prevalence and mortality continue
Prior studies demonstrated difficulties in decision making for HIV-infected
patients, including lack of identification of surrogate decision makers and
decision making with inadequate information. These studies showed that despite
the risk of complicated decision making at the end of life, most persons with
HIV had not completed advance directives2-4
and most had not discussed end-of-life issues with their physicians,3-5 even when they desired
such discussions.5 Recent studies have focused
on HIV-infected patients' interest in6,7
and some practitioners' willingness to provide euthanasia.8,9
As it becomes a chronic disease, HIV infection will require patients
and their health care practitioners to discuss prognosis and make decisions
about care goals over an extended time.10 Yet,
few researchers have investigated whether communication about preferences
and decision making has progressed with evolution of therapeutic options.
Indeed, increased optimism about HIV treatment may make it easier to avoid
planning for future clinical deterioration and death. Furthermore, most studies
of advance directives and end-of-life preferences have focused on small convenience
samples. To gain an understanding of behaviors and preferences regarding end-of-life
care among HIV-infected persons and to identify areas that need improvement,
we asked a nationally representative sample of HIV-infected patients about
end-of-life discussions with physicians, advance directive use, and willingness
to tolerate future adverse health states. Because these questions were included
in a broad survey of HIV-infected persons' characteristics and care, we were
able to evaluate in depth the factors associated with end-of-life preferences
The HIV Cost and Services Utilization Study (HCSUS) cohort is a nationally
representative probability sample of HIV-infected adults receiving care in
the contiguous United States. The reference population consists of persons
aged at least 18 years with known HIV infection who made at least 1 visit
for regular or ongoing care to a nonmilitary, nonprison medical care provider
other than an emergency department in early 1996.11,12
The HCSUS used a multistage design in which geographic areas, health care
practitioners, and patients were sampled. Of the 4402 eligible participants
sampled, 2864 (71%) completed the interview used in this analysis. The overall
coverage rate (the ratio of the population directly represented to the population
that would have been directly represented if we had complete responses at
all levels) is about 68%. Data were weighted to construct a representative
Most interviews were conducted in person using computer-assisted personal
interviewing, with 9% completed by telephone.13
Interviews began in January 1996 and ended 15 months later. Patients were
approached after their health care practitioners obtained their permission.
The study protocol was approved by the RAND (Santa Monica, Calif) and local
institutional review boards.
In addition to gathering data on patients' demographic characteristics
and clinical circumstances (contained in the HCSUS baseline interview), we
asked about their preferences and communication about end-of-life issues,
their willingness to tolerate future health states, and whether they had completed
an advance directive. Communication with one's practitioner about end-of-life
care was addressed by 3 questions: "Did your doctor or nurse ever discuss
with you . . . " " . . . how you feel about the use of machines and other
medical treatments to prolong your life when there is no chance of you getting
better?" " . . . whether you want to be resuscitated if you stopped breathing?"
and " . . . living wills?" These items were collapsed into a dichotomous variable
indicating any end-of-life communication between patient and practitioner.
Respondents were asked about their preferences for aggressiveness of
care: "If you had to make a choice at this time, would you prefer a course
of treatment that focuses on extending life as much as possible, even if it
means having more pain and discomfort, or would you prefer a plan of care
that focuses on relieving pain and discomfort as much as possible, even if
it means not living as long?" Response options included "definitely extend
life as much as possible," "probably extend life as much as possible," "probably
relieve pain and discomfort as much as possible," and "definitely relieve
pain and discomfort as much as possible." This item was previously shown to
have adequate test-retest reliability.14 In
addition, respondents were queried about willingness to tolerate future adverse
health states with 5 items modified from the Study to Understand Prognoses
and Preferences for Outcomes and Risks of Treatment (SUPPORT)14
that asked whether they "would prefer to live as long as possible in this
condition, even if you would not get any better, or if you would prefer to
die" than to live permanently (1) in pain, (2) attached to a ventilator, (3)
fed through a tube into your stomach, (4) unconscious or in a coma, or (5)
forgetting or being confused. Patients' responses to the 5 "willingness-to-tolerate"
items were combined into a willingness-to-tolerate index (internal consistency
= 0.77) ranging from 0 (not willing to tolerate any) to 5 (willing to tolerate
Based on prior literature and clinical experience, 4 groups of covariates
were hypothesized to be related to these end-of-life variables: demographics,
clinical characteristics, psychosocial variables, and practitioner characteristics.
Demographics included age (18-35, 36-49, or ≥50 years), sex, race/ethnicity
(white, black, Latino, or other), HIV exposure group (men who have sex with
men, intravenous drug users, men who have sex with men/intravenous drug users,
heterosexual contact, or other), education level (less than high school graduate,
high school graduate, completed some college, or college graduate), whether
there were children in the household, health insurance status (private, Medicare,
Medicaid, or none), region (Northeast, South, Midwest, or West) and employment
(employed, unemployed, disabled [and, thus, unable to work], or not working
[and not seeking employment]). Clinical variables included lowest CD4 cell
count (≤50, 51-200, 201-500, or >500/µL), HIV stage (asymptomatic,
symptomatic, or diagnosed as having acquired immunodeficiency syndrome [AIDS]),
symptom intensity score (grouped in tertiles), current heavy alcohol use or
drug dependence in the past year, scales assessing physical and mental health–related
quality of life (grouped in tertiles), and number of hospitalizations in the
past 6 months (0, 1, or >1). Psychosocial variables (all measured in tertiles)
included scales measuring social support (3 items; α = .72), positive
coping (4 items; α = .57), denial (2 items in which a higher score indicates
less denial), emotional well-being (5 items; α = .89), and whether the
respondent desired to participate in treatment decisions (2 items; α
= .76). Practitioner-related variables included regular practitioner type
(physician, nurse, or none), length of the relationship (<1, 1-3, or >3
years), a 2-item scale of respondents' trust in their practitioner (complete
or less than complete), health care practice setting (public, private, small,
or other) and teaching or nonteaching site.
For each respondent, an analytic weight was constructed to adjust the
sample to represent the entire reference population. Each analytic weight
can be interpreted as the number of persons in the underlying population represented
by that respondent. Each weight is the product of the sampling weight, which
adjusts for differential sampling probabilities, a multiplicity weight, which
adjusts for patients who could have entered the sample through visits to multiple
practitioners, and a nonresponse weight, which adjusts for differences in
rates of cooperation.15
The principal dependent variables in the analysis were communication
with practitioner about end-of-life issues, completion of an advance directive,
aggressiveness of care preference, and willingness to tolerate future permanent
adverse health states. We evaluated the relationship between each of these
dependent variables and respondent demographic factors, clinical factors,
psychosocial variables, and practitioner variables. The relationships among
dependent variables also were explored. Significance testing, with statistical
significance at P<.01, was performed by regression
and χ2 tests for bivariate analyses. All analyses incorporated
sample weights to estimate population prevalence parameters.
Multivariable logistic regression models were constructed to evaluate
the independent relationship of respondent demographics, clinical factors,
psychosocial variables, and practitioner variables with end-of-life communication
with one's practitioner and whether the patient had an advance directive.
The advance directive model also included whether the practitioner discussed
advance directives with the patient and both models included the willingness-to-tolerate
index. Multivariable linear regression was performed on the aggressiveness
of care preference score (ranging from 1 [definitely relieve pain] to 4 [definitely
extend life]) and the willingness-to-tolerate index. Respondent demographics,
clinical factors, psychosocial variables, and practitioner variables served
as the predictor variables for these models. Statistical analyses were performed
using SAS (Cary, NC) and Stata (College Station, Tex) statistical software.
The HCSUS sample of 2864 patients represents 231 400 persons in
the United States (95% confidence interval [CI], 162 800-300 000)16; 98% answered the questions about end-of-life communication
and preferences. Most of the population had symptomatic HIV disease; 59% had
AIDS and only 10% had asymptomatic disease. Seventy-seven percent of the HCSUS
population was male, 89% were younger than 50 years, and 52% had a high school
education or less. Forty-nine percent of the population was white, 33% was
black, and 15% was Latino. Thirty-seven percent were employed; 46% had an
annual household income below $10 000; and 32% had private health insurance,
19% had Medicare, and 39% had Medicaid alone.16
Slightly more than half of the physicians caring for these patients were primary
care physicians and 40% were infectious disease specialists. Three quarters
described themselves as HIV specialists.
Thirty-five percent of respondents reported that they had discussed
use of life-sustaining machines with their practitioner, 32% discussed resuscitation,
and 44% discussed living wills. Patients who had one such discussion tended
to have others; 26% discussed all 3 topics with their practitioner and 9%
discussed 2 topics. Fifty percent discussed none of these topics with their
In bivariate analysis, white patients were more likely to discuss end-of-life
issues, as were patients with Medicare insurance and disabled patients. Patients
with more advanced disease were more likely to have had a discussion with
their practitioner. Discussions were more prevalent among patients with more
symptoms, more hospitalizations in the last 6 months, and worse health status.
A longer relationship with and greater trust in one's practitioner were associated
with more communication
(See online appendix available in pdf format
In the multivariable analysis predicting any discussion (Table 1), physicians communicated less with black (odds ratio [OR],
0.57; 95% CI, 0.39-0.83) and Latino (OR, 0.74; 95% CI, 0.55-0.98) patients
about end-of-life issues compared with white patients. Women communicated
more with practitioners (OR, 1.39; 95% CI, 1.05-1.84), as did patients who
had children in their household (OR, 1.53; 95% CI, 1.12-2.10). Patients whose
route of HIV infection was via intravenous drug use (OR, 0.64; 95% CI, 0.45-0.89)
and those with less education were less likely to have discussions, but age
and employment status were not associated with end-of-life communication.
Patients with an AIDS diagnosis (OR, 1.72; 95% CI, 1.05-2.83) and lowest
CD4 cell counts of 50/µL or less were more likely to have discussed
end-of-life issues with their practitioner. Patients who were admitted to
the hospital more than once in the past 6 months had more than twice the odds
of discussion compared with those who had not been hospitalized. However,
health-related quality of life and emotional well-being were unrelated to
Practitioner type and practice setting were not important predictors
of discussion; however, patients with more trust in their practitioner were
more likely to have had an end-of-life discussion. Compared with patients
who had had a relationship with their practitioner for less than 1 year, those
whose relationships exceeded 3 years were more likely to have had a discussion
(OR, 1.40; 95% CI, 1.00-1.80). Patients who desired a greater role in medical
decision making, those who were less willing to tolerate adverse health states,
and those who coped more positively were more likely to report end-of-life
discussions. Denial and perceived social support were unrelated to discussion.
Thirty-eight percent of respondents reported that they had completed
an advance directive. In bivariate analyses, older patients, men, white patients,
and those with more education and higher income were more likely to have completed
an advance directive. Sicker patients also were more likely to have an advance
directive (Online Appendix available in pdf format
In the multivariable logistic regression model (Table 1), the most important predictor of advance directive completion
was respondent report that the practitioner had discussed the topic (OR, 5.82;
95% CI, 4.50-7.52). Black and Latino patients had about half the odds of completing
an advance directive, but disabled patients did so more often (OR, 1.38; 95%
CI, 1.06-1.80). Education level, health insurance, age, and sex were not significantly
associated with having an advance directive. After adjusting for covariates,
most clinical variables also were not associated with having an advance directive.
Psychosocial variables played a prominent role in advance directive completion;
higher social support, more positive coping, and less denial were associated
with having an advance directive. Longer patient-practitioner relationships
were associated with having an advance directive; patients whose relationships
with their practitioners exceeded 3 years had 1.5-fold odds of having an advance
directive compared with those who had had relationships of less than 1 year.
When asked whether they would prefer life-extending care or care focused
on comfort, HCSUS respondents expressed widely divergent opinions. Thirty
percent of respondents stated that they would definitely desire to extend
life and 14% probably would want to extend life, even if it meant having more
discomfort. Thirty-two percent said they would definitely desire to relieve
pain and 23% would probably want to relieve pain, even if it meant not living
Concerning whether they would want to remain alive in adverse health
states, 91% of respondents reported that they would prefer to die than to
live permanently in a coma. Eighty-seven percent would prefer to die than
remain permanently mechanically ventilated; 78%, 70%, and 45% would rather
die than live permanently fed through a tube, in pain, or confused, respectively.
Thirty-seven percent of respondents were unwilling to stay alive in any of
the 5 permanent adverse health states, 31% were unwilling to stay alive in
all but 1, and 15% were unwilling to stay alive in all but 2. Only 6% of respondents
desired to remain alive in all 5 health states.
In bivariate analyses, respondents' aggressiveness of care preference
was related to willingness to live permanently in future adverse health states.
The willingness-to-tolerate index demonstrated a gradient from a mean of 1.69
tolerable states among patients who definitely desired to extend life to a
mean of 0.74 tolerable states among patients who definitely desired to relieve
pain (Table 2). Patients who desired
to relieve pain reported more communication about end-of-life issues with
their practitioners and were more likely to have completed an advance directive;
54% of respondents who definitely desired to relieve pain reported having
an end-of-life discussion, whereas a discussion occurred with 44% of those
who definitely desired to extend life. Patients with a less aggressive care
preference were more likely to have completed an advance directive (Table 2).
The multivariable models of factors associated with patients' aggressiveness
of care preference and willingness-to-tolerate index score are presented in Table 3. Demographic factors were independently
associated with these preferences, whereas clinical variables, psychosocial
variables, and descriptors of the patient-practitioner relationship tended
to be unrelated.
Older patients were less willing to tolerate future adverse health states;
however, age was not significantly related to aggressiveness of care preference.
Women were less likely to desire life extension and were less willing to tolerate
future adverse health states. Black patients were willing to tolerate more
adverse health states, but race/ethnicity was not related to aggressiveness
of care preference. Human immunodeficiency virus exposure group was related
to both aggressiveness of care preference and willingness-to-tolerate index.
Compared with men who have sex with men, intravenous drug users, heterosexual
respondents, and those infected via other sources more often desired life
extension and expressed greater willingness to live in adverse health states.
Patients with children in their household desired to extend life, and those
with Medicare or no health insurance desired to relieve pain at the expense
of life extension. Patients with drug addiction or heavy alcohol use preferred
comfort to life extension (Table 3).
Most clinical, psychosocial, and practitioner variables were not significantly
associated with aggressiveness of care preference or the willingness-to-tolerate
index. The exception was that patients with greater desire to participate
in medical decision making held a preference for less aggressive care.
Despite nearly 2 decades of premature deaths due to HIV infection, communication
about end-of-life issues and completion of advance directives are still reported
by fewer than half of a nationally representative sample of patients in care
for HIV. During this time, much has been written about the difficulties of
making end-of-life decisions for HIV-infected patients because of inadequate
legal mechanisms and poor information exchange.17-19
The person who best knows a patient's preferences is not always the legal
decision maker.20 Without a proxy directive,
that job, at times, falls to estranged family members who may not even be
aware of the patient's diagnosis, let alone their preferences for end-of-life
care. Intravenous drug users, the second largest exposure group, commonly
have fractured social situations and often have poor social support. The deficits
demonstrated are likely to grow because those least likely to participate
in advance care planning—black patients, Latino patients, and intravenous
drug users—are rapidly growing segments of the HIV-positive population.
The data in this study reveal that there has been little improvement
in the proportion of HIV-infected individuals who complete advance directives
compared with earlier studies of more limited populations in the late 1980s
and early 1990s.2,3 Haas et al5 showed that among those who had not discussed end-of-life
care with their practitioners, 72% desired to do so. Demographic characteristics
were not associated with desire for discussion in that study5
or in studies conducted in non–HIV-positive samples.21,22
As demonstrated in a recent smaller study,23
the prevalence of discussion among individuals with HIV infection has not
increased, suggesting a persistent unmet need for patient-practitioner communication
about end-of-life issues. Black and Latino race/ethnicity and low education
level are associated with lack of communication and not completing an advance
directive. Consistent with prior studies,22,24
clinical need was associated with end-of-life discussions and completing advance
directives. The strongest facilitator of end-of-life discussions was having
2 or more hospital admissions in the past 6 months. This may be simply an
indicator of recent illness that stimulated discussion about end-of-life issues.
On the other hand, hospitalization may present a point in illness conducive
to discussion of prognosis and end-of-life care.
The patient-physician relationship plays a large role in end-of-life
communication, as described in a qualitative analysis of patients with HIV.25 In this study, patients with longer relationships
with their physician and those with more trust were more likely to have had
end-of-life discussions. Patients who desired a greater role in decision making
also reported more discussions. Ongoing relationships infused with trust may
be an important promoter of end-of-life discussion. Of course, direction regarding
trust cannot be discerned from these data; increased trust in practitioners
might be a product of end-of-life discussions and similar communication.26
It should be noted that even after adjusting for trust, willingness
to tolerate adverse health states, and characteristics of the patient-practitioner
relationship, race/ethnicity remained a strong predictor of end-of-life discussion.
These findings suggest that attention to trust and the patient-practitioner
relationship alone will not overcome barriers to end-of-life discussion and
decisions among ethnically diverse groups. Research to elucidate these obstacles
among HIV-infected individuals in minority and disadvantaged communities is
The most important factor associated with whether patients reported
having an advance directive was whether their practitioners discussed advance
directives with them. Patients who reported such a discussion had more than
5-fold odds of completing an advance directive compared with individuals who
stated that their practitioner had not raised the issue. This finding, reported
previously by Teno et al,2 is consistent with
smoking interventions27 showing that physician
communication and education can influence health behaviors. Providers of HIV
care can identify patients at high risk of difficult end-of-life decision
making (eg, intravenous drug users, persons who engage in denial or less positive
coping, and those with the least social support—the groups least likely
to have an advance directive) and help such patients understand the need to
specify a surrogate and to discuss preferences for care.
The scale measuring willingness to tolerate adverse health states demonstrates
why it is important for practitioners to elicit preferences from patients
concerning end-of-life care. This index and the closely related aggressiveness
of care preference demonstrate the wide variation in preferences held by HIV-infected
persons. Few factors were associated with willingness to tolerate adverse
health states. This finding is consistent with the lack of correlates to the
"reluctance to give up life" scale.28
The generally low willingness to tolerate adverse health states may
lead some to suggest that comfort care, rather than prolongation of survival,
should be the default clinical course in such circumstances. However, fully
17% of patients indicated that they would be willing to live permanently in
at least 3 of these health states. Having a preference to preserve survival
in these states was negatively associated with communication about end-of-life
issues with one's practitioner. Thus, a default posture not to provide aggressive
care in these clinical situations would override the preferences of many of
the patients least likely to communicate their wishes. As an alternative,
this study shows that information about care preferences can be elicited prospectively.
The data presented here not only emphasize the groups who are at risk of poor
communication about these issues but also suggest that clinicians can play
a role in improving communication with these individuals.
This study has several limitations. Decision making toward the end of
life is often most problematic for the patient newly entering into the health
care system; the HCSUS sample includes only persons in care for HIV infection.
In addition, the findings that we present are from a cross-sectional data
collection. As such, the relationships between variables, such as that of
patient-practitioner communication with advance directive completion, must
be considered associations rather than causal. Interventions are needed to
evaluate whether practitioner communication can increase advance directive
completion and ultimately improve end-of-life care. The focus of this study
is on communication and preferences in guiding end-of-life care. These cover
only 2 of the domains of quality end-of-life care.29,30
As HIV care enters its third decade, attention is focused on novel therapeutic
approaches and new treatment modes. Providers of HIV care should not forget
that despite the renaissance in therapeutics, many patients will die of HIV.
Health care practitioners have an important opportunity to encourage discussions
about end-of-life care preferences and surrogate decision making. This is
particularly important for the marginalized groups that make up an increasing
proportion of the HIV-infected population and who are least likely to have
engaged in advance care planning.