Despite the importance of early communication, most patients and physicians have not communicated about end-of-life decisions, and most patients have not completed advance directives [19-21]. Hypothesized barriers to end-of-life discussions include physician discomfort about discussing these issues, perceived time constraints, and variation in physician attitudes about the appropriateness of such discussions [22, 23]. Little research has evaluated patient barriers to such communication, and little is known about how patient characteristics and perceptions affect their preferences for communicating with physicians about end-of-life care [24]. Identifying and exploring barriers to patient–physician communication about end-of-life issues [22, 25, 26] may help guide physicians and their patients toward more effective discussions and may help physicians provide their patients with more appropriate and useful information.

To improve our understanding of factors influencing patients' decisions about end-of-life care, we examined correlates and associated outcomes of patient communication and preferences for communication with physicians about end-of-life decisions. Specifically, we examined communication and preferences about cardiopulmonary resuscitation and prolonged mechanical ventilation among patients enrolled in the Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatment (SUPPORT), a prospective multicenter study of the preferences, treatments, and outcomes of seriously ill hospitalized patients.

Methods

Top Methods Results Discussion Author & Article Info References

Study Design

This analysis was performed using data collected during phase II of SUPPORT. The overall study objectives and methods have been published elsewhere [27]. This study had two phases: an observational phase (phase I) and a subsequent interventional phase (phase II). Phase II was a controlled trial of an intervention intended to improve care provided to seriously ill hospitalized patients [28]. During this phase, clinicians randomly assigned to the intervention were given information about their patients' prognoses and preferences for care and were assigned a clinical nurse specialist to facilitate symptom control and effective communication with patients. The study sample for the current analysis consisted of patients who were enrolled in phase II of SUPPORT between January 1992 and January 1994 and were hospitalized at one of five participating clinical sites (Beth Israel Hospital, Boston, Massachusetts; University of California Medical Center. Los Angeles, California; MetroHealth Medical Center, Cleveland, Ohio; Duke University Medical Center. Durham, North Carolina; and Marshfield Clinic-St. Joseph's Hospital, Marshfield Wisconsin).

Patients enrolled in phase I of SUPPORT were not included in this analysis because we did not ask specific questions about patient preferences for discussion with their physicians about end-of-life issues during phase I. Because the phase II intervention did not affect care processes or outcomes [28], the intervention and control patients are combined in this analysis. Patients were eligible if they met defined criteria for at least one of the following nine diagnostic categories: acute respiratory failure, chronic obstructive lung disease, congestive heart failure, cirrhosis, nontraumatic coma, metastatic colon cancer, advanced non-small-cell lung cancer, multiorgan system failure with sepsis, or multiorgan system failure with a malignant condition. The 6-month mortality rate in these patients was approximately 50%. Patients were excluded if they did not speak English; were nonresident foreign nationals, younger than 18 years of age, or pregnant; had sustained head trauma; had been hospitalized with an expected stay of less than 72 hours; had AIDS; or had died or were discharged within 48 hours of study entry.

Data Collection

Patients were screened for study entry at the time of hospital admission; patients in intensive care units were screened every day. Sixty-four percent of patients in phase II of SUPPORT were enrolled on the first day of their hospitalization. Data were gathered by medical record review and interviews with patients between days 2 and 6 after enrollment. Information collected by chart review included patient demographic data (age, sex, study site, type of insurance, attending physician service) and information on patient clinical characteristics (diagnosis, acute physiology score on day 3 after study entry, and comorbid conditions). The acute physiology score is the physiology-based component of Acute Physiology and Chronic Health Evaluation III (APACHE III) and includes vital signs, laboratory measurements, and the Glasgow Coma Scale score [29]. By using a list developed as part of the APACHE II scoring system, we assessed comorbid conditions by reviewing charts and calculated a comorbidity score by summing the number of comorbid conditions; the score ranged from 0 to 7 [30]. Objective estimates of patients' 2-month survival were calculated by using the SUPPORT prognostic model [31].

Patients were interviewed between days 2 and 6 after study enrollment by trained interviewers who used standardized techniques to obtain information on patient characteristics; preferences; and perceptions of symptoms, function, and quality of life. Patients' self-reported race, marital status, religion, living situation, employment status, levels of income and education, and function 2 weeks before admission were assessed. Function was determined by using two different measures: 1) the number of dependencies among seven activities of daily living using a revised version of the Katz Activity of Daily Living Scale [32-34] and 2) a revised version of the Duke Activity Status Index [35, 36]. Quality of life was measured by asking patients to rate their overall quality of life on a five-point descriptive scale (1 = excellent; 5 = poor). Subjective estimates of prognosis were obtained by asking patients to estimate their probability of surviving 2 months beyond study entry. Possible responses included "90% or better," "about 75%," "about 50-50," "about 25%," and "10% or less." Patients' self-reports of having an advance directive and interest in participating in medical decisions [37, 38] were assessed. For advance directives, patients were asked whether they currently had a signed durable power of attorney or living will. Patients' desires for active involvement in decision making were assessed by using the Krantz scale [37, 38]. In a series of seven questions, patients were asked whether they preferred to rely on their physician or nurse to make decisions or whether they preferred to direct the decision-making process themselves. Patients were classified as wanting active involvement in decision making if they indicated a desire for active decision making in their answers to at least 4 of 7 questions.

Patients were asked the following question about their preference for cardiopulmonary resuscitation: "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 doctors to do if your heart ever stopped beating? Would you want your doctors to try to revive you or would you want your doctors not to try to revive you?" Responses were coded as: 1) patient wants resuscitation, 2) patient does not want resuscitation, 3) patient wants resuscitation but not intubation, or 4) patient doesn't know. For this analysis, patients who wanted resuscitation but no intubation (5%) were grouped with patients who wanted full cardiopulmonary resuscitation, because in both groups patients expressed preferences for some type of life-extending treatment. The 12% of patients who answered "don't know" were excluded from the analysis. Patients were next asked, "Have you specifically told your doctors that you want doctors to: (1) revive you, (2) not revive you, or (3) don't know, and were those doctors at this hospital?" Patients who had not discussed these preferences were then asked. "Would you like the opportunity to discuss this with your doctor[s] here?" Possible responses were 1) "yes," 2) "no," or 3) "don't know." Test-retest reliability (exact agreement) for the question on cardiopulmonary resuscitation, assessed within 24 hours of the initial interview for 90 patients, was 93% [39].

In a separate portion of the interview, we asked patients about their willingness to live while indefinitely attached to a breathing machine. The five possible responses were "very willing," "somewhat willing," "somewhat unwilling," "very unwilling," or "would rather die." For this analysis, responses for "very willing" and "somewhat willing" were grouped together, as were the last three responses. Patients were also asked whether they had discussed these preferences with their physicians and, if not, whether they would like to.

Outcome data were obtained by using medical record review, interviews with patients, and the National Death Index. Outcomes of interest were in-hospital, 2-month, and 6-month mortality rates; presence of do-not-resuscitate (DNR) orders in the charts of patients who did not want cardiopulmonary resuscitation; mean time to DNR orders; and number of unwanted resuscitation attempts. Figures for 2-month and 6-month mortality rates were cumulative and included patients who died during the study hospitalization. For patients who received a DNR order during hospitalization, the mean time to the order was defined as the average number of days to a DNR order from study entry. The number of unwanted resuscitation attempts was defined as the number of patients who did not want cardiopulmonary resuscitation and were subsequently resuscitated at least once.

Statistical Analysis

We focused our analysis on patient–physician communication about two end-of-life issues: preferences for discussion about cardiopulmonary resuscitation and preferences for discussion about prolonged mechanical ventilation. In our analysis, we sought to identify factors associated with patients who had discussed preferences for resuscitation and prolonged ventilation with physicians, those who wanted to discuss these preferences but had not yet done so, and those who did not want to discuss these preferences.

For analyses of these outcomes, we selected candidate independent variables that have been shown to be related to the dependent variables [40-47] or those that on the basis of our clinical experience, we thought were related to the dependent variables. These independent variables were age, sex, race, marital status religion, income, employment status, level of education, living situation, health insurance coverage, study site, attending physician service, diagnostic category, number of comorbid conditions, patient's subjective estimate of 2-month survival, patient's assessment of quality of life, self-report of advance directive, patient's preferences for cardiopulmonary resuscitation and prolonged mechanical ventilation, patient's interest in active involvement in medical decision making, functional status (scores on activities of daily living scale and Duke Activity Status Index), acute physiology score, and objective estimates of 2-month survival (according to the SUPPORT prognostic model).

In the bivariable analyses of factors, we used the chi-square test for categorical variables and the Wilcoxon rank-sum test for continuous variables. This analysis was performed on all independent variables to determine eligibility for inclusion in multivariable models (inclusion criterion, P 0.20). To adjust for potential confounding factors, we chose a backward selection model (stay criterion, P 0.05). For the multivariable analyses, logistic regression was used to identify factors independently associated (two-tailed P < 0.05) with each of the three categories of dependent (outcome) variables adjusted for diagnosis, physician specialty group, study site, and objective estimates of prognosis. The multivariable models reported included all factors significant at a P value less than 0.05 in either the cardiopulmonary resuscitation or prolonged mechanical ventilation models for each of the three comparisons. Inclusion of the confounding variables that we identified had no substantial effect.

Patient outcomes associated with communication and preferences for communication were compared for the three groups defined above; we used the chi-square test to compare proportions.

Role of Sponsor

This study was funded by the Robert Wood Johnson Foundation. Representatives of the Foundation had no role in gathering, analyzing, or interpreting the data and did not review the manuscript before it was submitted for publication.

Results

Top Methods Results Discussion Author & Article Info References

Response Rates

Of 4804 patients enrolled in phase II of SUPPORT, 2162 (45%) were eligible for interview between the second day and the sixth day after study entry. Patients were ineligible if they were comatose or intubated (n = 1391), could not communicate for other reasons (n = 366), had died (n = 368), had been discharged during the interview window (n = 239), or were cognitively impaired (n = 278).

Among the 2162 eligible patients, interviews were completed for 1832 (response rate, 85%). Of eligible patients who were not interviewed, 298 (90%) refused to be interviewed and 32 (10%) were not interviewed for other reasons. When compared with those interviewed, the 330 patients who were not interviewed were more likely to be of an ethnicity other than black (94% compared with 84%; P < 0.001) and to have worse prognoses (SUPPORT prognostic estimates for survival at 2 months, 71% compared with 76%; P < 0.001). The two groups were similar in age, sex, and religious preferences.

Patient Preferences for Cardiopulmonary Resuscitation

Data on patients' desires to discuss preferences for cardiopulmonary resuscitation were available for 1589 of 1832 patients interviewed (87%). Data on these patients are given in Table 1.

Of the 1589 patients who responded to the question on cardiopulmonary resuscitation preferences, 366 (23%) had discussed these preferences with their physicians before their initial interview and 1223 (77%) had not (Figure 1, top). Of the 1223 patients who had not had such a discussion, 516 (42%) said that they wanted to and 707 (58%) said that they did not want to. Of the 1589 patients with available data, 1113 (70%) wanted physicians to try to revive them and 476 (30%) did not.

View larger version (26K): [in this window] [in a new window]   Figure 1. Flow chart of patients' responses to questions about cardiopulmonary resuscitation (CPR) and prolonged mechanical ventilation. Shown are the results of patients' responses to three questions on preferences about cardiopulmonary resuscitation and prolonged mechanical ventilation (asked in the order shown, which were part of a larger patient questionnaire about preferences for care administered 2 to 6 days after study enrollment.

As shown in Table 2, patients who had discussed preferences for cardiopulmonary resuscitation were about twice as likely to have an advance directive and to want to forego resuscitation than patients who had not discussed these preferences. Compared with patients who had not discussed preferences for resuscitation, patients who had discussed preferences were less likely to estimate a 2-month survival probability of 90% or greater and were more likely to have a lower mean SUPPORT prognostic estimate for surviving 2 months. Patients who had not discussed and did not want to discuss their preferences for cardiopulmonary resuscitation had preferences for this intervention similar to those of patients who wanted to discuss resuscitation but had not done so.

Multivariable Correlates of Having Discussed Preferences for Cardiopulmonary Resuscitation

In multivariable analyses that adjusted for diagnosis, study site, physician specialty group, and objective estimate of prognosis, patient factors independently associated with having discussed preferences for cardiopulmonary resuscitation included not wanting resuscitation (adjusted odds ratio [OR], 2.15 [95% CI, 1.64 to 2.83]), having an advance directive (OR, 2.24 [CI, 1.66 to 3.01]), desiring active involvement in medical decisions (OR, 1.48 [1.13 to 1.94]), estimating a poor prognosis (OR, 1.90 [CI, 1.32 to 2.74]), having more dependencies in activities of daily living (OR, 1.12 per dependency [CI, 1.04 to 1.21]), living alone (OR, 1.47 [CI, 1.08 to 2.00]), having an income of $11 000 to $25 000 per year (OR, 1.41 [CI, 1.03 to 1.94]), and having more comorbid conditions (OR, 1.12 per comorbid condition [CI, 1.10 to 1.24]).

By adjustment for patient diagnosis, study site, physician specialty group, and objective estimate of prognosis, the effects of these potential confounders were accounted for during examination of potential associations between selected patient factors and a specific outcome (such as having discussed preferences for cardiopulmonary resuscitation). For example, the data presented in the preceding paragraph show that among our patients who did not want cardiopulmonary resuscitation, and after adjustment for the above-mentioned factors, the odds of having discussed resuscitation were approximately twice as great as the odds among patients who wanted resuscitation.

Multivariable Correlates of Wanting To Discuss Preferences for Cardiopulmonary Resuscitation but Not Having Done So

In multivariable analyses that adjusted for diagnosis, study site, physician specialty group, and objective estimate of prognosis, being black (adjusted OR, 1.53 [CI, 1.11 to 2.11] and being younger (OR, 1.14 per 10-year interval younger [CI, 1.04 to 1.25]) were independently associated with wanting to discuss preferences for cardiopulmonary resuscitation but not having done so.

Multivariable Correlates of Not Wanting To Discuss Preferences for Cardiopulmonary Resuscitation

In multivariable analyses that adjusted for patient diagnosis, study site, physician specialty group, and objective estimate of prognosis, patient factors independently associated with not wanting to discuss preferences for cardiopulmonary resuscitation included being of an ethnicity other than black (adjusted OR, 1.48 [CI, 1.10 to 1.99]), not desiring active involvement in medical decisions (OR, 1.33 [CI, 1.07 to 1.65]), not having an advance directive (OR, 1.35 [CI, 1.04 to 1.76]), estimating an excellent prognosis (OR, 1.72 [CI, 1.32 to 2.59]), and reporting fair to excellent quality of life (OR, 1.36 [CI, 1.05 to 1.76]) (Table 3).

Outcomes Associated with Patient Preferences for Communication about Cardiopulmonary Resuscitation

Patients who wanted to discuss preferences for cardiopulmonary resuscitation but had not yet done so and patients who had not discussed and did not want to discuss their preferences were similar with respect to SUPPORT prognostic estimates and in-hospital, 2-month, and 6-month mortality rates. For most outcome measures (except in-hospital mortality), patients who had discussed their preferences for resuscitation were significantly different from patients who had not discussed these preferences (Table 4). For example, patients who had discussed their preferences were significantly more likely to die within 2 or 6 months after study entry. However, patients who did not want a discussion (or who wanted a discussion but had not yet had one) had a substantial risk for death; their in-hospital mortality rate was 4.3%, and their 6-month mortality rate was 26.2%. Patients who did not want cardiopulmonary resuscitation and had discussed this preference with their physician or physicians were substantially more likely to have DNR orders written than were patients who did not want resuscitation but had not discussed their preferences (45% compared with 21%; P < 0.001). Only 8 (<2%) of the 476 patients who did not want resuscitation underwent this procedure; this number was too small to allow meaningful comparisons between patient groups.

Patient Preferences for Prolonged Mechanical Ventilation

Patients who responded to the questions about mechanical ventilation were similar to those who responded to the questions about cardiopulmonary resuscitation (Table 1). Of 1573 patients who responded to the question on preferences for prolonged mechanical ventilation, 185 (12%) had discussed these preferences with their physicians before their initial interview and 1388 (88%) had not (Figure 1, bottom). Of the 1388 patients who had not had such a discussion, 279 (20%) said that they wanted to and 1109 (80%) said that they did not want to.

Of the 1573 patients with available data, 192 (12%) were willing to accept prolonged mechanical ventilation (Figure 1, bottom). Patients who had discussed preferences for mechanical ventilation were more likely to have an advance directive and to want to forego prolonged ventilation than patients who had not discussed these preferences (P 0.006). Patients who had not discussed and did not want to discuss their preferences for prolonged ventilation had preferences for this intervention that were similar to those of patients who wanted to discuss mechanical ventilation but had not done so (P > 0.2). Patients who had discussed preferences for prolonged ventilation were less likely to estimate a 90% or greater 2-month survival rate and were more likely to have a lower mean SUPPORT prognostic estimate for surviving 2 months than patients who had not discussed these preferences (P 0.001).

In multivariable analyses that adjusted for diagnosis, study site, physician specialty group, and objective estimate of prognosis, patient factors independently associated with having discussed preferences for prolonged mechanical ventilation included not wanting cardiopulmonary resuscitation (adjusted OR, 1.96 [CI, 1.36 to 2.83]), having an advance directive (OR, 2.40 [CI, 1.62 to 3.56]), estimating a poor prognosis (OR, 2.12 [CI, 1.33 to 3.39]) or an unknown prognosis (OR, 1.94 [CI, 1.23 to 3.06]), and having more comorbid conditions (OR, 1.24 per comorbid condition [CI, 1.10 to 1.39]). In similar analyses, not being Jewish (OR, 2.23 [CI, 1.05 to 4.73]) and reporting a poor quality of life (OR, 1.59 [CI, 1.16 to 2.16]) were independently associated with wanting to discuss preferences for prolonged ventilation but not having done so. Patient factors independently associated with not wanting to discuss preferences for prolonged ventilation included not having an advance directive (OR, 1.66 [CI, 1.26 to 2.19]), estimating an excellent prognosis (OR, 1.70 [CI, 1.22 to 2.35]), and reporting fair to excellent quality of life (OR, 1.63 [CI, 1.26 to 2.12]).

Discussion

Top Methods Results Discussion Author & Article Info References

For these seriously ill hospitalized adults, communication about preferences for cardiopulmonary resuscitation and prolonged mechanical ventilation was uncommon. Only 23% of patients had discussed preferences for resuscitation with their physicians at the beginning of their hospitalization, and only 12% had discussed preferences for prolonged ventilation. These data are consistent with findings showing that physicians often do not communicate with patients about patient preferences for cardiopulmonary resuscitation [4, 48, 49] and that patients often do not discuss end-of-life issues with their physicians [3, 4, 49, 50]. Surprisingly, of patients who had not discussed cardiopulmonary resuscitation preferences with their physicians, only 42% expressed a desire to talk about their preferences with their physicians. Of patients who had not discussed their preferences about prolonged mechanical ventilation with their physicians, only 20% said that they wanted to talk about their preferences.

Unlike a previous study by Reilly and colleagues [1], which showed that only 19% of hospitalized patients had not discussed and did not want to discuss advance directives, our analysis indicates that a majority of seriously ill inpatients had not discussed and were not interested in discussing preferences for cardiopulmonary resuscitation (58%) or prolonged mechanical ventilation (80%). Patients who had not discussed and did not want to discuss their preferences for resuscitation were similar to patients who had discussed or were interested in discussing these preferences in terms of their SUPPORT prognostic estimates and mortality rates, but their subjective estimates of survival at 2 months were somewhat higher. Patients who had not discussed and did not want to discuss their preferences for resuscitation did not, in fact, have better prognoses, but they may have perceived that they had better prognoses than patients who had discussed or wanted to discuss these preferences. One might understand patients' reluctance to discuss end-of-life issues if, in fact, these discussions are irrelevant. Yet both patients who did and those who did not want to discuss these issues had the same risk for dying, as measured by the SUPPORT prognostic estimates of surviving 2 months. Furthermore, many of these patients who had not discussed and did not want to discuss end-of-life issues preferred to forego resuscitation (25%) and did not want prolonged ventilation (87%). Eighty percent of patients who did not want to discuss end-of-life issues reported not having an advance directive.

Although 70% of patients in our study indicated that they would want cardiopulmonary resuscitation, only 12% were willing to accept prolonged mechanical ventilation. Other studies have shown that most acutely ill patients desire resuscitation and mechanical ventilation and are more likely to want these treatments if the outcome is perceived to be favorable [9, 51, 52]; fewer patients want either measure if they are faced with prolonged ventilation [53, 54].

In an examination of factors that are independently associated with patients having discussed preferences for cardiopulmonary resuscitation, multivariable analyses showed that patient characteristics indicative of desiring a more "active" patient role (having an advance directive and desiring active involvement in medical decisions) were strongly associated with patients not wanting resuscitation. Characteristics indicative of a worsening prognosis were also independently associated with patients having discussed preferences for resuscitation. Similar patient characteristics were associated with having discussed preferences for prolonged ventilation. Although most studies indicate that patients are less inclined to opt for aggressive treatment as their perceived level of future cognitive and physical function and quality of life declines [10, 51, 53, 55, 56], previous work by Uhlmann and Pearlman [57] and Danis and colleagues [58] shows that patients' perceptions of current quality of life do not seem to affect their desire for intensive, life-sustaining medical care.

Previous investigators have found similar percentages of patients who are interested in discussing end-of-life care preferences with their physicians but have not done so [1, 3, 39]. Our analysis showed that being black and being younger were strongly associated with wishing to discuss preferences for cardiopulmonary resuscitation but not having done so. Although studies have shown racial differences with respect to access to medical care [59, 60], management of patients with acute [61] and chronic [62] life-threatening conditions, and assignment of DNR orders [63], few studies have documented racial differences in patient–physician communication. In a study of patients with AIDS, Haas and colleagues [2] found that being black was a correlate of having an unmet need to discuss preferences for end-of-life care. Our findings are consistent with those of Caralis and colleagues [64], who showed that among patients who had not discussed preferences for life-prolonging treatment, more black (63%) and Hispanic persons (62%) desired such discussions than did non-Hispanic white persons (39%). Just as nonwhites receive less intensive use of resources [65-70], they may also be less likely to have their needs met for discussions about care preferences at the end of life. The association between younger age and unmet need for end-of-life discussions may indicate that physicians are more open to discussion and more likely to bring up this topic with older patients who have chronic progressive illness than with younger patients whose severity of illness is similar. Previous studies [9, 71] suggest that physicians often believe that patients share their goals and values with respect to cardiopulmonary resuscitation and therefore do not ask patients about their opinions and preferences.

Our analysis has several limitations. First, we relied on patients' self-reports as to whether (and when) they had discussed their preferences for cardiopulmonary resuscitation or prolonged mechanical ventilation with their physician. Such an approach is subject to recall bias and the patient's interpretation of what constitutes such a discussion. However, what may be most important are patients' perceptions of whether or not a conversation took place, as well as the belief that their physician understands their preferences. Second, we ascertained whether a discussion about end-of-life preferences had occurred by using a single interview question at a single time point, whereas discussions about prognoses, resuscitation, and other end-of-life preferences often evolve over several days, weeks, or months. Third, the order of questions about these two interventions focused first on patients' choices for treatment and not on their preferences for discussing various end-of-life therapies. If patients had first been asked about their preferences for discussing these options and if we had provided more detailed explanations about these interventions and included patient preferences over time, our results may have been different. Fourth, our study sample was a group of hospitalized, seriously ill patients whose initial preferences for end-of-life care and discussions about this care were obtained during acute illness. These patients were a subset of the patients enrolled in SUPPORT; they had a better prognosis overall and a lower severity of illness than patients who could not be interviewed, although all were seriously ill (6-month mortality rate, 29%). We did not collect data on preferences of patients who were too sick to be interviewed or on preferences of outpatients; therefore, our findings may not be generalizable to patients less ill than our study patients. Finally, we did not collect specific information from patients about why they did or did not want to discuss preferences for end-of-life care.

In summary, our findings suggest that for seriously ill hospitalized adults, communication about preferences for cardiopulmonary resuscitation and mechanical ventilation is uncommon. Most patients do not discuss these end-of-life preferences with their physicians, although many patients are interested in such discussions. Some patient characteristics, such as being black, being younger, and reporting poor quality of life, are associated with wanting to discuss these preferences with one's physician but not having done so. Our findings also suggest that many patients who have not discussed their preferences do not wish to do so. Not having an advance directive, estimating an excellent prognosis, and reporting fair to excellent quality of life are associated with not wanting to discuss preferences. Patients who do not want to talk about their preferences often do not want life-extending treatments and are at substantial risk for undergoing these treatments. For patients who have not discussed and do not want to discuss their preferences, as well as patients with an unmet need for such discussions, failure to discuss and understand patient preferences for cardiopulmonary resuscitation and mechanical ventilation may result in the provision of unwanted interventions.

Appendix

The following are the SUPPORT investigators: George Washington University, Washington, D.C.: Rose Baker, MSHyg, Rosemarie Hakim, PhD, William A. Knaus, MD, Barbara Kreling, BA, Detra K. Robinson, MA, and Douglas P. Wagner, PhD; Dartmouth Medical School, Hanover, New Hampshire: Jennie Dulac, BSN, RN, Joanne Lynn, MD, MA, MS, Joan Teno, MD, MS, and Beth Virnig, PhD; John Hopkins University, Baltimore, Maryland: Marilyn Bergner, PhD (deceased), Albert W. Wu, MD, MPH, and Yutaka Yasui, PhD; Beth Israel Deaconess Medical Center, Boston, Massachusetts: Lee Goldman, MD, MPH, E. Francis Cook, ScD, Mary Beth Hamel, MD, Lynn Peterson, MD, Russell S. Phillips, MD, Joel Tsevat, MD, Lachlan Forrow, MD, Linda Lesky, MD, and Roger Davis, ScD; Cleveland MetroHealth Medical Center, Cleveland, Ohio: Alfred F. Connors Jr., MD, Neal V. Dawson, MD, Claudia Coulton, PhD, C. Seth Landefeld, MD, Theodore Speroff, PhD, and Stuart Youngner, MD; Duke University Medical Center, Durham, North Carolina: William J. Fulkerson Jr., MD, Robert M. Califf, MD, Anthony N. Galanos, MD, Peter Kussin, MD, Lawrence H. Muhlbaier, PhD, Maria Winchell, MS, Carlos Alzola, MS, and Frank E. Harrell Jr., PhD; Marshfield Medical Research Foundation, Marshfield, Wisconsin: Norman A. Desbiens, MD, Steven Broste, MS, Michael Kryda, MD, Douglas J. Reding, MD, Humberto J. Vidaillet Jr., MD, and Marilyn Follen, RN, MSN; University of California, Los Angeles, California: Robert K. Oye, MD, Paul E. Bellamy, MD, Gill Cryer, MD, James W. Davis, MD, Jonathan Hiatt, MD, Neil S. Wenger, MD, MPH, Honghu Liu, PhD, and Margaret Leal-Sotelo, MSW; Medical College of Wisconsin, Milwaukee, Wisconsin: Peter M. Layde, MD, Msc; Ohio University, Athens, Ohio: Hal Arkes, PhD; Presbyterian-St. Luke's Medical Center, Denver, Colorado: Donald J. Murphy, MD.

From Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, Massachusetts; University of California, Los Angeles, Medical Center, Los Angeles, California; George Washington University, Washington, D.C.; Case Western Reserve University School of Medicine, Cleveland, Ohio; University of Virginia School of Medicine, Charlottesville, Virginia; Duke University Medical Center, Durham, North Carolina; and Marshfield Clinic, Marshfield, Wisconsin.

Dr. Wenger: Department of Medicine, University of California, Los Angeles, School of Medicine, B-564 Factor Building, 10833 Le Conte Avenue, Los Angeles, CA 90024-1736.

Drs. Teno and Lynn: Center to Improve Care of the Dying, George Washington Medical Center, 1001 22nd Street, NW, Suite 870, Washington, DC 20037.

Dr. Connors: Department of Health Evaluation Sciences, University of Virginia School of Medicine, Box 600, Charlottesville, VA 22908.

Dr. Desbiens: Department of General Medicine, Marshfield Medical Research Foundation, 1001 North Oak Avenue-1RF, Marshfield, WI 54449-5790.