A Controlled Trial to Improve Care for Seriously Ill Hospitalized Patients
The Study to Understand Prognosis and Preferences for Outcomes and Risks of Treatments (SUPPORT)
The SUPPORT Principal Investigators
Objectives. - To improve end-of-life decision making and reduce the frequency
of a mechanically supported, painful, and prolonged process of dying.
Design. - A 2-year prospective observational study (phase 1) with 4301 patients
followed by a 2-year controlled clinical trial (phase II) with 4804 patients and their
physicians randomized by specialty group to the intervention group (n=2652) or control group
Setting. - Five teaching hospitals in the United States.
Patients. - A total of 9105 adults hospitalized with one or more of nine life
threatening diagnoses; an overall 6-month mortality rate of 47%.
Intervention.- Physicians-in the intervention group received estimates of the
likelihood of 6-month survival for every day up to 6 months, outcomes of cardiopulmonary
resuscitation (CPR), and functional disability at 2 months. A specially trained nurse had
multiple contacts with the patient, family, physician, and hospital staff to elicit
preferences, improve understanding of outcomes, encourage attention to pain control, and
facilitate advance care planning and patient-physician communication.
Results.-The phase I observation documented shortcomings in communication,
frequency of aggressive treatment, and the characteristics of hospital death: only 47% of
physicians knew when their patients preferred to avoid CPR; 46% of do-not-resuscitate (DNR)
orders were written within 2 days of death; 38% of patients who died spent at least 10 days
in an intensive care unit (ICU); and for 50% of conscious patients who died in the hospital,
family members reported moderate to severe pain at least half the time. During the phase II
intervention, patients experienced no improvement in patient-physician communication (e.g.,
37% of control patients and 40% Of intervention patients discussed CPR preferences) or in
the five targeted outcomes, i.e., incidence or timing of written DNR orders (adjusted ratio,
1.02:95% confidence interval [Cl], 0.90 to 1.15), physicians' knowledge of their patients'
preferences not to be resuscitated (adjusted ratio, 1.22; 95% Cl, 0.99 to 1.49), number of
days spent in an ICU, receiving mechanical ventilation, or comatose before death (adjusted
ratio, 0.97; 95% Cl, 0.87 to 1.07), or level of reported pain (adjusted ratio, 1.15; 95% Cl,
1.00 to 1.33). The intervention also did not reduce use of hospital resources (adjusted
ratio, 1.05; 95% Cl, 0.99 to 1.12).
Conclusions.-The phase I observation of SUPPORT confirmed substantial shortcomings
in care for seriously ill hospitalized adults. The phase II intervention failed to improve
care or patient outcomes. Enhancing opportunities for more patient-physician communication,
although advocated as the major method for improving patient outcomes, may be inadequate to
change established practices. To improve the experience of seriously ill and dying patients,
greater individual and societal commitment and more proactive and forceful measures may be
PUBLIC HEALTH and clinical medicine during this century have given Americans the
opportunity' to live longer and more productive lives, despite progressive illness. For some
patients, however this progress has resulted in prolonged dying, accompanied by substantial
emotional and financial expense. Many Americans today fear they will lose control over their
lives if they become critically ill and their dying will be prolonged and impersonal. This
has led to an increasingly visible right-to-die movement. Two years after voters in
California and Washington State narrowly defeated referenda on physician assisted
euthanasia. Oregon voters approved physician prescription of lethal medications for persons
with a terminal disease. Physicians and ethicists have debated when to use cardiac
resuscitation and other aggressive treatments for patients with advanced illnesses. Many
worry about the economic and human cost of providing life-sustaining treatment: near the end
In response, professional organizations, the judiciary, consumer organization. and a
president's commission have all advocated more emphasis on realistically forecasting
outcomes of life sustaining treatment and on improved communication between physician and
patient. Statutes requiring informed consent and communication, like the Patient
Self-determination Act' have been passed. Advance care planning and effective outgoing
communication among clinicians, patients, and families are essential to achieve these goals.
Previous studies indicate, however, that communication is often absent or occurs only during
a crisis. Physicians today often perceive death as failure.' they tend to be too pessimistic
regarding prognoses, and they provide more extensive treatment to seriously ill patients
than they would choose for themselves
Phase I of the Study to Understand Prognoses and Preferences for Outcomes and Risks of
Treatments (SUPPORT) confirmed barriers to optimal management and shortfalls in
patient-physician communication. The phase II intervention sought to address these
deficiencies by providing physicians with accurate predictive information on future
functional ability, survival probability for each day up to 6 months." and patient
preferences for end-of-life care; a skilled nurse augmented the care team to elicit patient
preferences, provide prognoses, enhance understanding, enable palliative care, and
facilitate advance planning. We hypothesized that increased communication and understanding
of prognoses and preferences would result in earlier treatment decisions, reductions in time
spent in undesirable states before death, and reduced resource use. This article describes
the effect of the SUPPORT intervention on five specific outcomes: physician understanding of
patient preferences; incidence and time of documentation of do-not-resuscitate (DNR) orders;
pain; time spent in an intensive care unit (ICU), comatose, or receiving mechanical
ventilation before death: and hospital resource use (Figure
Phase I was a prospective observational study that described the process of decision
making and patient outcomes. Phase II was a cluster randomized controlled clinical trial to
test the effect of the intervention. Enrollment, data collection, and interviewing were
virtually identical during the two phases.
Qualified patients were in the advanced stages of one or more of nine illnesses: acute
respiratory failure, multiple organ system failure with sepsis, multiple organ system
failure with malignancy, coma, chronic obstructive lung disease, congestive heart failure,
cirrhosis, metastatic colon cancer, and non-small cell lung cancer. Patients were excluded
if they were younger than 18 years, were discharged or died within 48 hours of qualifying
for the study, were admitted with a scheduled discharge within 72 hours, did not speak
English, were admitted to the psychiatric ward, had acquired immunodeficiency syndrome, or
were pregnant or sustained an acute burn, head, or other trauma (unless they later developed
acute respiratory failure or multiple organ system failure). Nurses' trained in the SUPPORT
eligibility criteria reviewed hospital admissions and ICU patients daily to identify newly
Phase I enrolled patients from June 1989 to June 1991, and phase II enrolled patients
from January 1992 through January 1994. Patients were recruited from five medical centers:
Beth Israel Hospital, Boston. Mass: MetroHealth Medical Center, Cleveland, Ohio; Duke
University Medical Center, Durham, NC; Marshfield Clinic/St Joseph's Hospital. Marshfield.
Wis: and the University of California at Los Angeles Medical Center. An independent
committee monitored potential adverse events, including 6-month mortality for intervention
patients and changes in patient satisfaction with medical care. Mortality follow-up to 6
months was complete for all phase I patients. In phase II, 22 patients (0.5%) were
unavailable for follow-up at a median of 80 days.
Data Collection Methods
Data collection was based on both concurrent and retrospective medical record reviews and
on interviews with patients. Patient surrogates (defined as the person who would make
decisions if the patient was unable to do so}, and patients' physicians.
Medical Record-Based Data. -We collected physiological indicators of disease
severity, length of stay, a modified version of the Therapeutic Intervention Scoring
System,' and comorbidities from the medical records on days 1,3, 7, 14 and 25. The permanent
medical record was retrospectively reviewed for discussions or decisions concerning 18
important issues, such as the use of dialysis, withdrawal from a ventilator, and DNR orders.
Reliability testing on 10% of the medical records showed at least 90% agreement on
Interview Data. -Patients and their designated surrogates were interviewed in the
hospital between days 2 and 7 (median, day 4) and again between days 6 and 15 (median, day
12) after study enrollment, whether or not the patient remained hospitalized. The surrogate
was interviewed 4 to 10 weeks after the patient's death. Among the 45% of patients who were
able to communicate, the response rate for the first interview was 85%. The surrogate
response rate for the first interview was 87%. For the second-week interviews, the patient
response rate was 71%c and the surrogate response rate was 78%. The interviews collected
information on patient demographics, functional status, self-assessed quality of life,
communication with physicians, frequency and severity of pain, satisfaction with medical
care and the patient's preferences for cardiopulmonary resuscitation (CPR). When a patient
interview was not possible, the surrogate's responses were substituted, a strategy that
mirrors clinical practice. Important elements of the patient/surrogate questionnaires were
retested for reliability. Initial and repeat responses had greater than 80% agreement.
The most senior available physician acknowledging responsibility for the patient's
medical decisions was interviewed in the first and second weeks after patient enrollment
(median days. 3 and II, respectively). In both interviews, we asked the physician's
understanding of the patient's preferences for CPR. In the second interview, physicians
assigned to the intervention were queried about its influence on the patient's care.
Physician response rates were 86% for the first interview and 82% for the second interview.
Phase II Intervention
Presented with early findings from phase I documenting substantial shortcomings in
communication, decision making, and outcomes, physicians at the participating institutions
voiced interest in attempting change. Physician leaders and study investigators at the sites
met to discuss how decision making could be unproved to more closely reflect both probable
outcomes and patient preferences and ways to improve patient, family, and physician
communication. Physicians suggested that communication could improve if there were more
reliable and prompt information generated by the study and if study personnel would make it
more efficient to have conversations. In response to these suggestions, the phase II
intervention aimed to improve communication and decision making by providing timely and
reliable prognostic information, by eliciting and documenting patient and family preferences
and understanding of disease prognosis and treatment, and by providing a skilled nurse to
help carry out the needed discussions, convene the meetings, and bring to bear the relevant
information. The elements of the intervention and their timing are presented in (Table 1
In each ease the nurse was free to shape her role so as to achieve the best possible care
and outcome. For example, she sometimes engaged in extensive emotional support. Other times,
she mainly provided information and ensured that all parties heard one another effectively.
All of the nurses' involvement required approval of the attending physicians. In virtually
all cases, the physician approval came with no limits. Physicians were free, however, to
limit the intervention in any way that they felt was best for the patient, and there was no
requirement for them to share or discuss the information with the patient or family or to
allow the nurse's involvement to continue. The nurse was identified on her badge and in the
consent process as part of a research effort, but she had the role and appearance of a
typical clinical specialist.
Randomization. -To limit contamination, patients were assigned to intervention or
control (usual care) status based on the specialty of their attending physician. Physician
specialties were divided into five groups: internal medicine, pulmonology/medical ICU,
oncology, surgery, and cardiology. We used a cluster randomization scheme to assign the
intervention randomly to 27 physician group-site combinations, restricted by the conditions
that .50% to 60% of patients would be assigned intervention status, and that at least one
intervention and one control physician specialty group be at each of the five study
institutions. This resulted in II physician specialty groups assigned to control and 16
assigned to the intervention (Figure
1). Analyses were based on allocation to intervention
(i.e., intention to treat), irrespective of whether a given patient received the
intervention. Investigators were blinded to the phase II results during data collection. .
Analytic Methods. -Five measures were chosen to evaluate the intervention: (1) The
timing of written DNR orders was analyzed with a log-normal regression model to prepare
Kaplan-Meier predicted median time until the first DNR order was written. If a DNR order was
not written. DNR order timing was censored at the day of death or hospital discharge. (2)
Patient and physician agreement on preferences to withhold resuscitation was based on the
first interview of the patient (or surrogate if the patient was unable to be interviewed)
and the responsible physician. Agreement was defined as a response to forgo resuscitation
from both patient and physician, analyzed with binary logistic regression, and applied to
all interviewed patients or surrogates who had matching physician interviews. (3) Days spent
in an ICU, receiving mechanical ventilation, or comatose before death were analyzed using
ordinary' least-squares regression (after taking the log of 0.5 plus the number of days) and
only included phase II patients who died during the index hospitalization. (4) Frequency and
severity of pain analyses were based on all patients or surrogates interviewed in the second
week with a combined measure (moderate or severe pain all, most, or half the time) and
analyzed using a single, ordinal logistic regression model. (5) Hospital resource use was
defined as the log of the product of the average Therapeutic Intervention Scoring System
rating and length of hospital stay after the second day of the study. In regression analyses
on phase I data, this measure closely estimated hospital bills across the five study
institutions (Pearson R^2=0.93 on log product). We used ordinary least-squares regression to
model the log of resource use, which was then converted to 1993 dollars. This method allows
comparisons of groups and institutions across time without having to adjust for varying
Power and Safety Calculations
Power calculations based on phase I data indicated greater than 90% power (a=. 05) to
detect a 1-day decrease in days until a DNR order was written, a 5% increase in the
proportion of physicians and patients agreeing on a DNR order, a 20% decrease in undesirable
days, a 10% decrease in reported pain, and a 5% decrease in resource use. Effects of the
intervention on mortality rates were quantified by the estimated intervention, control
hazard ratio from adjusted Cox models.
Adjustment Methods for Phase II Results
Because patients were assigned to intervention or control status based on a limited
number of specialty groups, the resulting cohorts might be unbalanced in patient baseline
risk factors. Furthermore, practice patterns among the physician specialty groups in phase I
differed substantially. We controlled for these expected pre-intervention differences using
baseline multivariable risk scores that were derived by generating models to predict phase I
outcomes, each of which incorporated interactions between physician specialty and hospitals.
Observed imbalances in phase II baseline patient characteristics were also adjusted using a
propensity score that corrected for selection bias associated with being assigned to
intervention status:" Further details on the construction of both of these risk scores are
available on request. Imputation methods for missing data have been published. Finally, we
simulated the phase II randomization scheme on the phase I data to evaluate secular trends.
To adjust for multiple outcomes, our methods pre-specified adjusting confidence intervals
(CIs) using the method of Hochberg and Benjamini if more than one P value was less than .05.
All statistical analyses were done with UNIX S-Plus, version 3.2 software* and the Design
Phase I Observations
Phase I enrolled 4301 patients (Figure
1) with a median age of 65 years and other
characteristics summarized in Table 2. The mean predicted 6-month survival probability was
52% with an actual 6-month survival probabilitv of 48% (Table 2). Thirty-one percent of
phase I patients with interviews preferred that CPR be withheld, but only 47% of their
physicians accurately reported this preference during the first interview. Nearly half (49%)
of the 960 phase I patients who indicated a desire for CPR to be withheld did not have a DNR
order written during that hospitalization. Nearly one third of these patients (278 [29%])
died before discharge.
Among all phase I patients who died during the index hospitalization (n = 1150), 79% died
with a DNR order, but 46% of these orders were written within 2 days of death. Among all
phase I deaths, the median number of days spent in an ICU, comatose, or receiving mechanical
ventilation was 8, more than one third (38%) spent at least 10 days in an ICU, and 46%
received mechanical ventilation within 3 days of death. In the second week, 22% of patients
reported being in moderate to severe pain at least half the time. In interviews conducted
after a patient died, surrogates indicated that 50% of all conscious phase I patients who
died in the hospital experienced moderate or severe pain at least half the time during their
last 3 days of life.
We found substantial variation in the five outcomes among physician specialty groups and
across the five institutions. Across institutions, the median number of days spent in an ICU
before death varied from 5 to 9. The proportion of patients reporting moderate to severe
pain at least half the time varied by a factor of 2.7, from 12% to 32% across study
institutions. The predicted median number of days until a DNR order was written for a
standard patient varied by a factor of 3.5, from 73 days for patients on a surgical service
to 22 for oncology. One study institution had a predicted median time until a DNR was
written for a standard patient of 28 days, and another institution had a predicted median
time of 49 days. Agreement on DNR varied from 8% for cardiology patients to 24% for oncology
patients and from a low of 8% at one study institution to a high of 27% at another. The
median number of days spent in an ICU before death ranged from 14 in the surgical
specialties to 5 for patients in pulmonary/ ICU and oncology services.
Phase II Demographics
Phase II enrolled 4804 patients. 2152 assigned to usual medical care and 2652 assigned to
intervention status (Figure
1). Their characteristics were generally similar to those of
phase I patients (Table 2).
Delivery of the Intervention
Ninety-five percent of intervention patients received one or more patient-specific
components of the intervention. The SUPPORT nurse was involved in the care of all but 133
patients, and 75 of these were patients who died or were discharged on the day of
enrollment. The SUPPORT nurse communicated with the physician in virtually all cases. She
talked directly with the patient or family in most cases (e.g., with 84% concerning
prognosis. 77% about pain, 63% about likely outcomes or resuscitation, and 73% concerning
written advance directives). For patients in the hospital for 7 or more days after
qualifying for the study, the median number of SUPPORT nurse contacts with the patient,
family, or physician was four and the mean was six. Documentation in the progress notes of
discussions about patients' preferences with regard to resuscitation was increased from
38.5% in phase I to 50.3% among phase II intervention patients.
The patient's physician received at least one prognostic report for 94% of patients, and
the report was put in the medical record of 80%. The patient's physician received at least
one printed report of patient or surrogate understanding and preferences in 78% of cases. No
physician refused to receive the printed reports or to have them shared with other
professional staff. The physicians for 43 patients refused to allow the SUPPORT nurse to
have contact with the patient and family, and seven patients or surrogates refused to speak
with the SUPPORT nurse.
Effect of Intervention on Outcomes
The prevalence or timing of documentation of DNR orders for the 2534 intervention
patients was the same as for the 2208 control patients (adjusted ratio of median time, 1.02;
95% Cl, 0.90 to 1.15) (Table 3). There was a small association of the intervention with
improved patient-physician DNR agreement for the 1480 intervention patients who had patient
or surrogate and matching physician interviews, compared with 1159 control patients
(adjusted ratio. 1.22; 95% Cl. 0.99 to 1.49). The number of days spent in an ICU, comatose,
or receiving mechanical ventilation before death for. the 680 intervention patients who died
in the hospital was the same as for the 530 control patients (adjusted ratio of median days,
0.97; 95% CI, 0.87 to1.07). Reported pain increased for the 1677 intervention patients and
surrogates interviewed in the second week, compared with the control group (adjusted ratio,
1.15: 95% Cl, 1.00 to 1.33) (Table 3). There was no change in hospital resources used for
2593 intervention patients not dead or discharged before the third study day compared with
2129 control patients (adjusted ratio of average resource use, 1.05; 95% Cl, 0.99 to 1.12).
The unadjusted differences between intervention and control patients for median days
until the first DNR order was written were large, especially for patients with colon cancer
and non-small cell lung cancer for whom the median number of days until a written DNR order
was 80% lower in intervention patients. Adjustment for baseline imbalances reduced much of
the difference in each category (Table 4). The differences that persist in the cancer
category are of uncertain importance, being one among multiple comparisons and being based
on a small number of patients (Table 4).
Figure 2 illustrates the secular trends of each outcome in the phase II intervention and
control groups, as well as in phase 1., using simulations of the physician specialty
groupings used in phase II. None of the five outcomes changed significantly during the 5
years of the study. The differences between those who would have been assigned to
intervention and control in phase I persisted throughout the SUPPORT study, unaffected by
time or by the intervention.
Communication and Preferences
The intervention did not change the unadjusted proportion of patients or surrogates
reporting a discussion about CPR: 37% of control patients and 40% of intervention patients
reported discussing their preference. Of patients who did not have such a discussion. 41% of
each group said they would like to discuss CPR. Seventeen percent of control patients and
20% of intervention patients changed their resuscitation preferences to forgo CPR by the
second week after enrollment, and 39% of control patients and 41% of intervention patients
reported having a discussion about their prognosis with a physician. Of those who did not
discuss their prognosis, 44% of control patients and 42% of intervention patients reported
that they would like to have such a discussion.
Physician's Perspective on Intervention
In the second physician interview, 59% acknowledged receiving the prognostic reports and
34% acknowledged receiving the preference reports. Fifteen percent reported discussing this
specific information with patients or families. Nearly a quarter of respondents (22%) said
they thought the SUPPORT nurses involvement improved patient care.
After adjusting for baseline differences, the 6-month mortality for phase II control
patients was the same as for intervention patients (adjusted relative hazard, 0.95:95% Cl,
0.87 to 1.04). Both control (68%) and intervention (69%) patients or surrogates rated their
care as excellent or very good.
Findings from phase I of SUPPORT documented many shortcomings of care. The SUPPORT patients were all seriously ill, and their dying proved to be predictable, yet discussions and decisions substantially in advance of death were uncommon. Nearly half of all DNR orders were written in the last 2 days of life. The final hospitalization for half of patients included more than 8 days in generally undesirable states: in an ICU, receiving mechanical ventilation, or comatose. Families reported that half of the patients who were able to communicate in their last few days spent most of the time in moderate or severe pain. Based on a study in a defined population at our Wisconsin site, we estimate that patients meeting SUPPORT criteria account for approximately 400,000 admissions per year in the United States and that another 925,000 people are similarly ill but would not meet SUPPORT entry requirements of being hospitalized or in intensive care. Patients with SUPPORT illnesses and severity account for about 40% of persons dying in the defined population.
Building on the findings in phase I, observations of others, the opinions of physicians at the five sites, and the marked variation in their baseline practices, the phase II intervention aimed to make it easier to achieve better decision making for these seriously ill patients. The intervention gave physicians reliable prognostic information and timely reports of patient and surrogate perceptions, the two most important factors cited recently by physicians when considering life-support decisions for critically ill patients. The intervention nurse also undertook time-consuming discussions, arranged meetings, provided information, supplied forms, and did anything else to encourage the patient and family to engage in an informed and collaborative decision-making process with a well-informed physician (Table 1).
The intervention was limited by its application to a diverse group of physicians and patients, all of whom had to comply voluntarily. The intervention had to be perceived as helpful, polite, and appropriate. As an initial attempt to change outcomes for seriously ill patients, we did not seek authority to be coercive or more than minimally disruptive. As designed, however, the intervention was vigorously applied. The SUPPORT nurses were committed, energetic, and highly trained. They engaged in the care of virtually all our patients, and nearly everyone had printed reports delivered promptly.
Because we thought that changes in the decision-making processes that were not reflected in improved patient outcomes would not be worth much expense, we specified five outcomes, each indicating an important improvement in patient experience, as the main targets of the intervention.
The intervention had no impact on any of these designated targets (Table 3 and Table 4). Furthermore, even though the targeted outcomes are objectives of much ethical and legal writing and of some explicit social policy (such as informed consent statutes, the Patient Self-determination Act. and guidelines on pain). There were no secular trends toward improvement for intervention or control patients during the 5 years of SUPPORT data collection (Figure 2).
These results raise fundamental questions about the intent and design of this trial. Do patients and physicians see the documented shortcomings as troubling? Can enhanced decision making improve the experience of seriously ill and dying patients? Were the inevitable 'limitations of this project too great to draw strong conclusions?
Because there was no movement toward what would seem to be better practices, one could conclude that physicians, patients, and families are fairly comfortable with the current situation. Certainly, most patients and families indicated they were satisfied, no matter what happened to them. Physicians have their established patterns of care, and while they were willing to have the SUPPORT nurse present and carrying on conversations, physician behavior appeared unchanged. Perhaps physicians and patients in this study acknowledged problems with the care of seriously ill patients as a group. However, when involved with their own situation or engaged in the care of their individual patients, they felt they were doing the best they could, were satisfied they were doing well, and did not wish to directly confront problems or face choices. The study certainly casts a sail over any claim that if the health care system is given additional resources for collaborative decision making In the form of skilled professional time, improvements will occur. In phase II of SUPPORT, improved information, enhanced conversation, and an explicit effort to encourage use of outcome data and preferences in decision making were completely ineffectual, despite the fact that the study had enough power to detect small effects.
It is possible that the intervention would have been more effective if implemented in different settings, earlier in the course of illness or with physician leaders rather than nurses as implementers. Perhaps, it would have been effective if continued for more time or tested at later end points. However, the overall results of this study are not encouraging. No pattern emerged that implied that the intervention was successful for some set of patients or physicians or that its impact increased overtime. The five hospitals had been chosen for their diversity and their willingness to undertake a substantial and controversial challenge. Yet none showed a tendency toward improvement in these outcomes.
SUPPORT did demonstrate, however, that issues this complex can be studied with sufficient scientific rigor to be confident of the findings. We achieved good interview response rates among seriously ill patient, their families, and physicians' widespread acceptance of the intervention in diverse hospitals, and high-quality data. Consent and confidentiality issues were complex but amenable to solution. The analytic issues required application of relatively novel approaches but they proved effective. The study also demonstrated the need for such methods when performing evaluations of complex interventions in seriously ill patients. We would have concluded that the intervention positively influenced all outcomes had we not had phase I results for baseline adjustment and phase II control patients to evaluate secular trends (Table 4 and Figure 2).
In conclusion, we are left with a troubling situation. The picture we describe of the care of seriously ill or dying persons is not attractive. One would certainly prefer to envision that when confronted with life-threatening illness the patient and family would be included in discussions, realistic estimates of outcome would be valued, pain would be treated, and dying would not be prolonged. That is still a worthy vision.
However, it is not likely to be achieved through an intervention such as that implemented by SUPPORT. Success will require reexamination of our individual and collective commitment to these goals. More creative efforts at shaping the treatment process, and. perhaps, more proactive and forceful attempts at change.
1. Callahan-D. The Troubled Dream of Life.- Living With Mortality. New York. NY: Simon & Schuster, 1993.
2. McCue JD. The naturalness of dying. .JAMA. 1995:213:1039-1043.
3. Annas GJ. Death by prescription: the Oregon initiative. N Engl J Med. 1994:301-40-1243.
4. Brown D. Medical community still divided on Oregon's assisted suicide law. Washington Post. November 13.1994:A20.
5. Tomlinson T, Brody H. Futility and the ethics of resuscitation. JAMA. 1990:364:1276-1280.
6. Blackhall LJ. Must we always use CPR? N Engl J Med. 1987:317:1281-1285.
7. Wanzer SH. Federman DD, Adelstein SJ. et al. The physician's responsibility toward hopelessly ill patients. .V Engi J Med. 1989;320:884-49.
8. Ginzberg E. The high cost of dying, Inquiry. 1980-.17:293-295.
9. Hill TP, Shirley D.A Good Death. Reading, Mass: Addison-Wesley; 1992.
10. President's Commission for the Study of Ethical Problems in Medicine and Biomedical and Behavioral Research. Deciding to Forego Life-Sustaining Treatment. Washington. DC: US Government Printing Office; 1983.
11. Hastings Center. Guidelines on the Termination of Life-Sustaining Treatment and the Care of the Dying. New York, NY: Hastings Center. 1987.
12. Coordinating Council on Life-Sustaining Medical Treatment Decisions by the Court. Guidelines far State Court Decicions Making in Authorizing or Withholding Life-Sustaining Medical Treatment. Williamsburg, Va: West Publishing Company; 1991.
13. Crozan V. director, Missouri Dept of Health. 110 SCt 284. l99M8552856.
14. Council on Ethical and Judicial .affairs. Guidelines for the appropriate use of do-not-resuscitate orders. JAMA. 1991265:188-1871.
15. American Thoracic Society Bioethics Task Force. Withholding and withdrawing life-sustaining therapy. Ann Intern Med. 1991:115:478-185.
16. Cancer Pain Relief and Palliative Care. Report of a WHO Expert Committee, Technical Report W. Geneva. Switzerland: World Health Organization: 1990.
17. Omnibus Budget Reconciliation Act of 1990 (OBRA-90). Pub L 101-508, 4206 and 4751 (Medicare and Medicaid, respectively). 42 USC 1395cc (a)(I)(Q), 1395mm (c)(8). 1395ccff. 139a(a)(57). 1396a(a)(58), and 1396a(w) (supp 1991).
18. Bedell SE. Delbanco TL. Choices about cardiopulmonary resuscitation in the hospital. N Engl J Med. 1984: 10:1088-1093.
19. Katz J. The Silent World of Doctor & Patient. New York, NY: Free Press: 1986.
20. Blackhall U. Cobb J. Moskowitz MA. Discussions regarding aggressive care with critically ill patients. J Gen Intern .Med. 1989:4:399-102.
21. Arkes HR. Dawson NV. Speroff T. et al. The covariance decomposition of the probability score and its use in evaluating prognostic estimates. Med Decis Making. 1995:15:120-131.
22. Molloy DW, Guyatt G..Alemayehu E. McIloy W. Treatment preferences, attitudes toward advance directives, and concerns about health care. Humane Med. 1991:7:285-290.
23. Teno JM. Lynn J. Phillips RS. et al. Do formal advance directives affect resuscitation decisions and the use of resources for seriously ill patients? J Clin Ethics. 1994:5:23-30.
24. Teno JM, Hakim RB. Knaus WA, et al. Preferences for CPR: physician-patient agreement and hospital resource use. / Gen Intern Med. 1995:10: 179-186.
25. Wu AW, Damiano AM, Lynn J. et al. Predicting future functional status for seriously ill hospitalized adults. Ann Intern.Med. 1995:122-42-350.
26. Knaus WA. Harrell FE. Lynn J, et al The SUPPORT prognostic model: objective estimates of survival for seriously ill hospitalized adults. Ann Intern Med. 1995:122:191-203.
27. Lynn J, Knaus WA. Background for SUPPORT. J Clin Epidemiol. 1990;43<suppl):1S-4S.
28. Murphy DJ. Knaus WA, Lynn J. Study population in SUPPORT. J Clin Epidemiol. 1990:43 (suppl):IIS-28S.
29. Knaus WA. Wagner DP. Draper EA. et al. The APACHE III prognostic system. Chest. 1991:100: 1619-16.36.
30. Knaus WA. Draper EA, Wagner DP, Zimmerman JE. APACHE II: a severity of disease classification system. Crit Care Med. 1985:13:818-829.
31. Cullen DJ, Civetta JM. Briggs BA, Ferrara LC. Therapeutic Intervention Scoring System. Crit Care Med. 1974 2:57-60.
32. Ware JE Jr. Synder MK, Wright WR. Davies AR. Defining and measuring patient satisfaction with medical care. Eral Program Plann. 1983;6:247-263.
33. Cox DR. Regression models and life tables. J Royal Stat Society B. 1972:34:187-220.
34. Cook EF, Goldman L. Performance of tests of significance based on stratification by a multivariate confounder score or by a propensity score. J Clin Epidemiol. 1989:-42:317-324.
35. Hochberg Y. Benjamin Y. More powerful procedures for multiple significance testing. Stat Med. 1990:811-818.
36. Statistical Sciences. S-Plus User's Manual Version 3.2 Seattle, Wash: StatSci. MathSoft Inc: 1993.
37. Harrell FE. Design: S-Plus functions for biostatistical/epidemiologic modeling. testing. estimation. validation, graphics. prediction and typesetting. 1994. Available on the Internet lib.stat.cnu.edu.
38. Layde PM. Broste SK. Desbiens N. at al. Generalizability of clinical studies conducted as tertiary care medical centers. J Clin Epidemiol. In press.
39. Covinsky KE. Goldman L. Cook EF. et al. The impact of serious illness on patients' families. JAMA. 1994:272:1839-1844.
40. Wachter RM. Luce JM. Hearst N. Lo B. Decisions about resuscitation: inequities among patients with different but similar prognosis. Ann Intern M. 1989.
41. Davidson KW. Hackler C. Caradine DR. McCord RS. Physicians' attitudes on advance directives. JAMA. 1989-262:2415-2419.
42. Brody DS. Miller SM. Lerman CE. et al Patient perception of involvement medical care. J Gen Intern Med. 1989:4:506-511.
43. Uhlmann RF. Pearlman RA. Cain KC. Physicians' and spouses' predictions of elderly patients' resuscitation preferences. J Gerontol. l988:43:M115-M121.
44. Cook DJ. Guyatt. GH. Jaeschke R.et al. Determinants in Canadian health care workers of the decision to withdraw life support from the critically ill. JAMA 1995:275:703-708.
45. Danis M. Patrick DL. Southerland LI. et al. Patients' and families' preferences for medical intensive care. JAMA. 1988:260:797-802.
46. Tversky A. Shafir E. Choice under conflict: the dynamics of deferred decision. Psychol Sci. 1992: 3:358-361.
47. Redelmeir DA. Shafir E. Medical Decision making in situations that offer multiple alternatives. JAMA 1995:273:302-305.
48. Greco PJ. Eisenberg JM. Changing physicians' practices. N Engl J Med 1993:329:1271-1274.
49. Neuland S. How We Die: Reflections on Life's Final Chapter. New York. NY: .Alfred A Knopf Inc: 1994.