Goodlin SJ, Zhong Z, Lynn J, Teno JM, Fago JP, Desbiens N, Connors, Jr AF, Wenger NS, Phillips RS. Factors Associated With Use of Cardiopulmonary Resuscitation in Seriously Ill Hospitalized Adults. JAMA. 1999;282(24):2333-2339. doi:10.1001/jama.282.24.2333
Author Affiliations: Division of Geriatrics, LDS Hospital, Salt Lake City, Utah (Dr Goodlin); Center to Improve Care of the Dying, George Washington University, Washington, DC (Drs Zhong and Lynn); Center for Gerontology and Health Care Research, Brown University, Providence, RI (Dr Teno); Department of Medicine, Dartmouth Medical School, Hanover, NH (Dr Fago); University of Tennessee, College of Medicine, Chattanooga Unit, Chattanooga (Dr Desbiens); Department of Medicine, University of Virginia, Charlottesville (Dr Connors); Department of Medicine, University of California at Los Angeles (Dr Wenger); and Department of Medicine, Beth Israel Deaconess Medical Center, Boston, Mass (Dr Phillips). Dr Goodlin is a Faculty Scholar, Open Society Institute, Project on Death in America, New York, NY.
Caring for the Critically Ill Patient Section Editor: Deborah J. Cook, MD, Consulting Editor, JAMA. Advisory Board: David
Bihari, MD; Christian Brun-Buisson, MD; Timothy Evans, MD; John Heffner, MD;
Norman Paradis, MD.
Context The epidemiology of do-not-resuscitate (DNR) orders for hospitalized
patients has been reported, but little is known about factors associated with
the use of cardiopulmonary resuscitation (CPR).
Objective To identify factors associated with an attempt at CPR for patients who
experienced cardiopulmonary arrest.
Design Secondary analysis of data collected in 2 prospective cohort studies:
the Study to Understand Prognoses and Preferences for Outcomes and Risks of
Treatments (SUPPORT, 1989-1994) and the Hospitalized Elderly Longitudinal
Project (HELP, 1994).
Setting Five teaching hospitals across the United States.
Participants A total of 2505 seriously ill hospitalized patients and nonelectively
admitted persons aged 80 years or older who experienced cardiopulmonary arrest.
Main Outcome Measures Medical records data on CPR efforts, DNR orders, disease severity, age,
race, sex, length of stay, and survival; functional status and preferences
concerning CPR obtained by interviews with patients or surrogates; and 2-month
survival estimates provided by physicians.
Results Five hundred fourteen study subjects (21%) received CPR during their
index hospitalization. Among them, 327 (63.6%) had CPR within 2 days of death
and 93 (18.1%) had resuscitation and survived their index hospitalization.
Use of CPR was more likely in men (odds ratio [OR], 1.39; 95% confidence interval
[CI], 1.12-1.73), younger patients (OR per 10-year increase, 0.90; 95% CI,
0.84-0.96), African Americans (OR, 1.76; 95% CI, 1.33-2.34), patients whose
reported preferences were for CPR (OR, 2.60; 95% CI, 1.91-3.55), who reported
better quality of life (OR, 1.49; 95% CI, 1.10-2.03), or who had higher physician
estimates for 2-month survival (OR per 10% increase, 1.14; 95% CI, 1.09-1.19).
Rates varied significantly with geographic location and diagnosis; the adjusted
OR for patients with congestive heart failure was 3.31 (95% CI, 2.12-5.15)
compared with patients with acute respiratory failure or multiple organ system
Conclusions Our data suggest that a resuscitation attempt is more likely when preferred
by patients and when death is least expected. Further study is required to
understand variation in use of CPR among sites and for patients with different
diagnoses, race, sex, or age.
Initially designed to rescue patients experiencing a sudden cardiac
arrest due to arrhythmia,1 cardiopulmonary
resuscitation (CPR) has come to be seen as a procedure that should be used
for patients for whom there is reasonable chance of restoring cardiopulmonary
function and prolonging life.2 Many guidelines,
articles, and legal actions recommend circumstances under which resuscitation
should be attempted.
The likelihood of survival after CPR has been shown to vary with age
and disease.3 Decisions against attempting
CPR may be a component of care plans that limit aggressiveness of care.4 Yet decisions about CPR may reflect generally held
expectations about particular diseases and perceptions of the appropriateness
of specific treatments, rather than actual prognoses.5
Do-not-resuscitate (DNR) orders are more common in patients with more functional
compromise and with increased age and vary by diagnosis, sex, race, and location.6 In the Study to Understand Prognoses and Preferences
for Outcomes and Risks of Treatments (SUPPORT), a study of seriously ill hospitalized
patients, earlier DNR orders were associated with patient preferences to forego
resuscitation, worse overall prognosis, and age older than 75 years.7 In intensive care units, DNR order rates have increased
and rates of initiation of CPR have declined in the past decade.8- 10
Surprisingly, patient preferences about life-extending treatment as reflected
in living wills have not been shown to alter DNR decision making.11,12
SUPPORT and the Hospitalized Elderly Longitudinal Project (HELP) were
designed to describe and improve decision making and outcomes of care for
seriously ill or elderly hospitalized adults.13
In SUPPORT, all patients were hospitalized with 1 of 9 illnesses, severe enough
to have a high risk of dying in the ensuing 6 months. In HELP, all patients
were 80 years of age or older and were enrolled at the time of emergency hospital
admissions. Many patients received life-prolonging interventions, and sometimes
these interventions conflicted with patients' or surrogates' preferences for
In both SUPPORT and HELP, among patients who died or experienced successful
CPR during their enrollment hospitalization, we examined the features of the
patient, the physician, and the hospitalization to understand what factors
were associated with resuscitation attempts among patients who experienced
cardiopulmonary arrest. We expected that the likelihood of having CPR at the
end of life would reflect patients' prognoses and preferences. We aimed to
measure the strength of those associations and to assess the potential impact
of other clinical and demographic factors.
The data collection methods used in SUPPORT have been reported previously.13 From June 1989 to June 1991 (phase I) and from January
1992 to January 1994 (phase II), SUPPORT enrolled patients 18 years of age
or older who met specific severity criteria for 1 of 9 serious illnesses (nontraumatic
coma, acute respiratory failure, multiple organ system failure with sepsis
and multiple organ system failure with malignancy, chronic obstructive pulmonary
disease, congestive heart failure [CHF], cirrhosis, metastatic colon cancer,
or inoperable non–small cell lung cancer) who were admitted to 1 of
5 medical centers: Beth Israel Hospital, Boston, Mass; MetroHealth Medical
Center, Cleveland, Ohio; Duke University Medical Center, Durham, NC; St Joseph's
Hospital, Marshfield, Wis; and the University of California Medical Center
at Los Angeles. Some patients were eligible for the study at the time of admission;
others became eligible as their condition worsened while in an intensive care
unit. Patients were excluded if they died or were discharged within 48 hours
of study enrollment; were admitted with a planned discharge within 72 hours;
did not speak English; or had the acquired immunodeficiency syndrome (AIDS),
multiple trauma, or pregnancy. SUPPORT entry criteria and enrollment have
been reported in detail elsewhere.15 In these
analyses, we grouped patients with acute respiratory failure and patients
with multiple organ system failure and sepsis together because our previous
work had found them to be overlapping categories with similar prognoses and
HELP enrolled patients 80 years of age and older, with any diagnosis
except elective surgery, AIDS, or multiple trauma, who had an unplanned hospitalization
for 48 hours or more at 1 of 4 teaching hospitals (Beth Israel Hospital, MetroHealth
Medical Center, St Joseph's Hospital, and the University of California Medical
Center at Los Angeles) between February and November 1994. Patients were assessed
for eligibility for HELP at the time of hospital admission.
Institutional review committees at all participating hospitals approved
the study. Informed consent was obtained prior to interviews.
The data collection methods used in HELP were identical to those in
SUPPORT for the purposes of this analysis. The following descriptive variables
were recorded from medical records: diagnoses including comorbid conditions,17 vital signs, common laboratory values, and a clinical
assessment of neurological status using the Glasgow Coma Scale. We obtained
the date of any attempt at CPR (documentation in the medical record of a call
for the CPR team and their having initiated CPR) and the date of death from
the record. Disease severity was assessed using the Acute Physiology, Age,
Chronic Health Evaluation (APACHE) III Acute Physiology Score (APS).13,18
Patients and their surrogate decisionmakers were interviewed in the
first week of entry into SUPPORT or HELP. Interview data included demographic
factors, perceived quality of life (using a 5-point scale), and functional
status 2 weeks prior to admission using a modified Katz activity of daily
living (ADL) scale ranging from 1 to 7 points.19
Patient interview data were available for 233 subjects; for 1486 subjects,
the surrogate's response was calibrated to patients' responses and substituted
as described previously.20 When both patient
and surrogate responses were unavailable, we imputed ADL (for 814 patients)
and quality of life values (for 868 patients) using a logistic regression
model adjusting for patients' interview status, diagnosis, age, coma score,
APS, comorbidities, dementia, and sites.20
When we restricted our analysis to patients without imputed data, our results
were unchanged; therefore, we present models including imputed data as our
Preferences about resuscitation were assessed with the following question:
"As you probably know, there are a number of things doctors can do to try
to revive someone whose heart has stopped beating, which usually includes
a machine to help breathing. Thinking of your current condition, what would
you want your doctor to do if your heart ever stops beating?: (1) would want
doctors to try to revive; (2) would want doctors not to try to revive; (3)
would want CPR but no ventilator; and (4) ‘don't know'"(the last 2 responses
were recorded if the patient or surrogate volunteered them). For our analyses,
responses 1 and 3 were included as showing a preference for CPR. Patients
who responded "don't know" were also considered as if they had stated a preference
for CPR, as would be done in clinical practice. Surrogate response for patients'
preferences about CPR were used for 1479 patients in this analysis who could
not be interviewed; 601 patients died before their interview and 1584 patients
were intubated, in coma, or otherwise unable to communicate. All surrogate
interviews reported here were conducted prior to the patient's death. No data
about preferences were available for 820 subjects (32.7%), either because
a surrogate could not be identified or the patient or their surrogate declined
to answer the question.
In a separate interview within 5 days of enrollment, but prior to patient
death, the most senior physician available on the treatment team was asked
to state the probability (0-100) that the patient would live for 2 months
or more. Physician interview data were available for 1972 patients.
In this secondary analysis of SUPPORT and HELP data, we identified all
patients in the SUPPORT and HELP studies who died during their index admission
and/or received CPR during their enrollment hospitalization. We considered
CPR to have been provided if we found chart documentation of CPR on or after
the third day after study admission. Since patients were excluded from SUPPORT
if death occurred during the first 48 hours of study enrollment, we excluded
all CPR attempts tried on the first 2 calendar days of study entry.
Patients were characterized with descriptive statistics. To study the
bivariable association between CPR attempts and patients' characteristics,
we used χ2 tests to analyze discrete variables and the Wilcoxon
test for continuous variables. We used logistic regression models to determine
which factors were associated with a CPR attempt at the time of cardiopulmonary
arrest. Independent variables included in the model were patient age, race,
sex, disease group, preferences for CPR, APS on day 3, number of ADL dependencies,
patient reported quality of life, physician's estimate of patients' 2-month
survival, and study site. We used the c statistic as a measure of the ability
of the model to discriminate between those who did and those who did not receive
a CPR attempt (0.5 = no discrimination, 1 = perfect discrimination).21 A 2-sided α of .05 was used as the criterion
for significance. In a secondary analysis, we added to the model whether the
patient had a DNR order prior to cardiopulmonary arrest to explore whether
variation in DNR orders explained the variation in CPR we observed, but found
that DNR orders were highly collinear with use of CPR, resulting in substantial
increases in SEs and an unstable model. Separately, to assess whether secular
trends confounded our results, we also adjusted for year of study entry and
results were similar to our primary analyses.
SUPPORT and HELP enrolled 10,281 subjects. Of these, 2505 subjects experienced
cardiopulmonary arrest, and 514 (21%) received CPR on or after the third study
day during their index hospitalization. Of the 514 patients who received CPR,
93 (18.1%) survived the index hospitalization, and 327 (63.6%) had CPR attempted
within the last 2 days of life (293 on the calendar day of death and 34 on
the day before death). Table 1
shows the age, sex, race, primary diagnosis, APS on the third study day, ADL
score, and quality of life 2 weeks prior to study entry, and the number of
comorbidities for patients with and without an attempt at resuscitation. Patient
or surrogate preference to attempt CPR was associated with having CPR, yet
13% of patients (or surrogates) who received CPR preferred to forego CPR,
and 38% of patients who wanted CPR or were unsure did not receive CPR at the
end of life. Patients who had CPR were younger, more often African American,
and more often male. We found substantial variation by site and diagnosis.
Patients with fewer ADL dependencies and better APS were more likely to receive
a CPR attempt.
Of the 514 patients who received CPR on or after day 3 of the study
entry, 15 (3%) had DNR orders written on or before the day prior to the CPR
attempt, and 50 had DNR orders written on the day of the attempt, while 449
had no DNR orders written. Of the 93 patients who survived CPR, 1 had a DNR
order written the day of the CPR attempt, 8 had DNR orders written subsequently,
and 1 had a DNR order prior to the CPR attempt. Of the 421 patients who had
CPR but died during the index hospitalization, 293 died on the same day of
the last CPR attempt, and 128 lived more than 1 day after the last CPR attempt.
Among the 293 patients who died on the same day of the last CPR attempt, 29
had DNR orders written that day. Among the 128 who lived more than 1 day after
the last CPR attempt, 93 had DNR orders written before death.
Of the 1991 patients who died without CPR attempts, 1802 (91%) had DNR
orders in place. Of the 189 patients who died without DNR orders or a CPR
attempt, 75 had DNR decisions documented in the medical record without a specific
DNR order. Among the 114 remaining patients who died without a CPR attempt
prior to death or a DNR order or note, 42 had preferred CPR, 24 had preferred
to forego CPR, 9 were unsure, and 39 had no data available.
Of the 514 patients who had CPR, 74 (14%) had more than 1 CPR attempt
during the hospitalization. Of the 421 patients who had CPR but died during
the index hospitalization, 62 (15%) had more than 1 CPR attempt. Of the 93
patients who had CPR and survived the hospitalization, 12 (13%) had more than
1 CPR attempt.
Table 2 shows the adjusted
odds ratios (ORs) for the factors that were significant in the logistic regression
model. The multivariable model shows that the adjusted likelihood of a resuscitative
attempt decreased with increasing age. Men were more likely to have an attempt
at resuscitation than were women. African Americans were more likely to undergo
a CPR attempt. Patients who expressed clear preference against CPR were less
likely to undergo CPR than those who had no preference or who wanted CPR.
Patients who reported better quality of life prior to hospitalization were
more likely to receive CPR. Neither APS nor functional status were associated
with the use of CPR.
Patients with CHF were most likely to receive CPR. Patients in coma
were least likely to have a resuscitative effort. Use of CPR varied substantially
across sites. For patients at one site, the OR for having CPR was 2.53 compared
with patients at another study institution. Physician prognostic estimates
were strongly associated with receiving CPR; the OR for having CPR increased
by 0.14 for each 10% increase in the likelihood of survival at 2 months. The
c statistic for our multivariable model is 0.746.
Resuscitation was attempted for 514 (21%) of the 2505 patients who experienced
cardiac arrest during their index hospitalization and 93 (18%) of 514 patients
who had CPR survived their index hospitalization. Most patients who died more
than 1 day following a CPR attempt had a DNR order written prior to death.
We found that the likelihood of attempting resuscitation at the time of cardiac
or pulmonary arrest varies substantially with prognosis and preferences, quality
of life, diagnosis, site of hospitalization, and patient age, sex, and race.
Most medical ethicists suggest that decisions about resuscitation should
be guided by patient preferences and by the likelihood of success of the resuscitative
effort or the patient's prognosis for survival. In our study, the physician
estimate of 2-month survival and patient preference have an important effect
on use of CPR, but are similar in magnitude to other variables. Although less
than 60% of patients who had DNR orders died during their initial hospitalization
in SUPPORT,7 similar factors were associated
with the timing of DNR orders among all patients in SUPPORT and the use of
CPR among patients experiencing cardiopulmonary arrest. Previous work in SUPPORT
identified a strong association between patient preferences and DNR orders,7 and in our study, patients with a DNR order generally
did not receive an attempt at resuscitation at the end of life. Although there
was a strong correlation between preferences, DNR orders, and use of CPR,
9% of patients who died without CPR did not have a DNR order in place. Of
the 114 patients who died without a CPR attempt, DNR order or note, 42 patients
or their surrogates had expressed a preference for CPR.
Patients for whom we had no data about preferences had a rate of CPR
attempt similar to those who preferred CPR. Many of these subjects lacked
a ready surrogate who might have made health care decisions to limit interventions.
Of patients or surrogates who wanted CPR or were unsure, only 28% received
CPR, while 11% of patients who did not want CPR received an attempt at CPR.
This lack of congruency between preferences and treatment may reflect changes
in preferences or prognoses in the time between interviews and cardiopulmonary
arrest or misunderstood communication between patients and physicians. Physician
decisions about CPR may be heavily influenced by factors such as their own
values and preferences.22,23 Some
clinical situations may appear to offer so little hope of effectiveness of
CPR that physicians may not offer an attempt at CPR to patients or surrogates.
In other work, patient preferences were not associated with the use of life-sustaining
Our data show higher rates of CPR attempts for patients for whom physicians
felt there was reasonable likelihood of surviving 2 months. In a recent study
of intensive care unit patients, severity of illness (APACHE III score) was
the most important variable associated with a DNR order.25
Conversely, when death is not expected imminently, physicians may be reluctant
to forego resuscitation. However, even in end-stage disease, physicians may
have difficulty recognizing that death is imminent.26,27
It may seem that decisions to forego an attempt at CPR should not vary
with the type of disease after adjustment for physician estimate of prognosis.
However, characteristics associated with disease type may explain different
patterns of attempted resuscitation. For example, reasonably well-defined
prognostic markers are available for patients with incurable, metastatic cancer
or with coma, and when poor prognosis is clear, many physicians and patients
avoid CPR. In fact, some question the merits of making CPR available to terminally
ill cancer patients.28,29
However, for patients with other diseases, the prognosis may be more
uncertain. Patients with end-stage CHF follow a less predictable course,26,30 and this diagnosis may be a more
important influence on the decision to attempt CPR than the survival prognosis
itself. Furthermore, despite a poor long-term prognosis, patients with advanced
CHF may be quickly resuscitated from an event caused by a cardiac arrhythmia.
Physicians may also have different thresholds of perceived chance of survival
of patients for recommending CPR for patients with CHF or chronic obstructive
pulmonary disease compared with those with cancer.31
The ambiguity in prognosis for CHF and the ability of patients with CHF to
respond well to treatment even late in the course may explain the increased
use of CPR for patients with CHF compared with patients with other diseases
included in our study.
The patients who received resuscitation attempts did not significantly
differ in severity of illness (as measured by APS) on the third study day
or in functional status 2 weeks earlier, in the adjusted analysis, when compared
with those who did not have a resuscitation attempt. This contrasts with other
reports in which more functionally dependent patients received less aggressive
care than others. However, our measure for functional status may be inaccurate,
in part, because it was imputed (rather than reported directly by patient
or surrogate) for one third of our subjects. Alternatively, for the seriously
ill patients included in SUPPORT, there may have been less variation in functional
status compared with those described in other studies. Additionally, the strong
correlation between prognostic estimate and measures of disease severity may
have dampened any possible effect of functional status and APS on use of CPR
in our multivariable model. In contrast to our findings on functional status,
we found that patients were more likely to receive CPR if they reported better
quality of life prior to hospitalization.
Several studies suggest that hospitalized older persons who are seriously
ill have similar rates of short- and long-term success of resuscitation attempts
compared with younger patients.32,33
Long-term survival and functional status were worse after successful CPR for
individuals older than 70 years in another study.34
In our study, older patients were less likely to receive CPR, even after adjustment
for patient preferences and severity of illness. This observation is consistent
with other observations that suggest that less aggressive care is provided
to older patients.35 Less aggressive care for
older persons may reflect patient and physician values.36,37
Other analyses of SUPPORT data found that older seriously ill patients preferred
CPR less often than did younger patients.38
While differing preferences are not sufficient to explain the effect we document,
older persons or their families may be more readily persuaded to avoid CPR
as their clinical situation worsens.
African American patients had higher rates of attempts at CPR than non–African
Americans, even after adjustments for their greater preference for CPR. Differing
priorities in end-of-life care have previously been observed between various
cultural and racial groups,39,40
but it is unclear what effects these differences have on decisions to attempt
CPR at the end of life.
Variation in rates of attempted resuscitation by site suggests that
decisions to attempt resuscitation are influenced by the culture of local
medical practices. The influence of the institution on the likelihood of attempting
CPR was more powerful than patients' preferences, which is difficult to justify.
Profound geographic variation in care for the dying has been demonstrated
in other work.41,42 Understanding
how decisions about attempting resuscitation are made in different settings
and what influences guide these decisions may help to identify ways to improve
care for the seriously ill.
This study has several limitations. First, our analysis focuses on factors
associated with use of CPR but because we restricted our analysis to the index
hospitalization, our data cannot be used to estimate the longer-term survival
of patients following CPR. Additionally, in our analysis, we do not account
for variation in success of CPR in different patient populations or in different
institutions, nor do we present data on factors associated with successful
CPR. However, the likelihood of successful CPR is generally low in hospitalized
patients, especially in those similar to patients in SUPPORT.3,32,33
Both SUPPORT and HELP were based in academic referral hospitals and may have
attracted patients more likely to desire aggressive care. Limiting our description
to those who had successful CPR or died during their initial hospitalization
excludes information on outcomes experienced shortly after hospital discharge
or in other settings. Since patients who died or were discharged within the
first 48 hours of enrollment were excluded, our results do not generalize
to patients' entire hospital stay. Since we did not collect data on the exact
time DNR orders were written, we are unable to tell which occurred first when
a CPR attempt and DNR order occurred the same day. However, it is likely that
DNR orders were generally written after CPR attempts when they occurred on
the same day.
Our reliance on surrogate responses and imputed interview data in both
SUPPORT and HELP may introduce bias. Surrogates may not accurately understand
or predict patients' preferences.43 Yet in
many clinical settings, surrogates function in a role of decision maker. Functional
status data derived by imputing the number of functional deficits from a logistic
regression model may reduce our ability to find differences in functional
status among patient groups. Finally, our analysis uses data from SUPPORT
and HELP collected from 1989 to 1994. Practice patterns regarding CPR and
end-of-life care for seriously ill patients may have changed since these data
We found that a CPR attempt was associated with patient or surrogate
preferences for CPR and the physician's estimate of better survival at 2 months.
However, the size of these effects is modest and comparable to the effect
of other factors. The variation in use of CPR across different diagnoses may
be explained by variability in the clinical course of disease and prognostic
uncertainty. Patients' race, sex, age, and site of hospitalization, however,
had significant effects that are more difficult to explain. Future work to
improve CPR decision making should include perceptions of CPR effectiveness
and seek to understand cultural forces in medical and lay communities that
cause the variations we observed. Additionally, systems to ensure that patients'
preferences are addressed and that DNR orders are written and followed need
further attention and improvement.