The Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments (SUPPORT) and the Hospitalized Elderly Longitudinal Project (HELP) have primarily sought to understand and improve decision making for seriously ill and elderly hospitalized patients. In SUPPORT, all patients who were hospitalized at one of five participating centers during 4 years (1989-1991, 1992-1994) and were in the advanced stages of at least one of nine medical conditions were enrolled [9-11]. In HELP, a random sample of persons 80 years of age and older who were hospitalized in one of four of these centers during 1994 was enrolled.

The Hospitalized Elderly Longitudinal Project was intended only to describe the largely unexamined experience of elderly hospitalized persons. However, SUPPORT also studied an intervention-improved prognostic and subjective information and the efforts of specially trained nurse specialists-to enhance communication and accelerate decision making. The SUPPORT intervention targeted improvement in the following areas: time before orders against resuscitation; agreement between physician and patient about orders against resuscitation; time in the intensive care unit, in a coma, or on a ventilator before death; alleviation of pain; and use of resources. The intervention was ineffectual in improving any of these measures [10].

Both studies included very sick and elderly persons, many of whom died during the follow-up period. Therefore, these databases included much information that described the experience of patients who turned out to be near death. Here, we report the ways in which the patients' surrogate decision makers, who were usually close family members, recalled the patients' health, symptoms, functional state, emotional state, preferences for care, use of life-sustaining treatments, and site of death.

Methods

Top Methods Results Discussion Author & Article Info References

Study Sites and Study Sample

The patient selection and data collection methods used by SUPPORT are reported elsewhere [10-13]. Our analysis focused primarily on patients who died and for whom a surrogate decision maker completed an interview about the experience of dying. For patients in SUPPORT, we examined deaths that occurred within 6 months of enrollment; patient deaths in HELP occurred within 1 year of enrollment.

From June 1989 to June 1991 and from January 1992 to January 1994, SUPPORT enrolled all patients who met the study entry criteria at five hospitals: Beth Israel Hospital, Boston; MetroHealth Medical Center, Cleveland; Duke University Hospital, Durham, North Carolina; St. Joseph's Hospital, Marshfield, Wisconsin; and the University of California at Los Angeles Hospital, Los Angeles. Entry criteria required that patients be in the advanced stages of one of the following conditions: coma, acute respiratory failure, multiple organ system failure with sepsis or a malignant condition, chronic obstructive pulmonary disease, congestive heart failure, cirrhosis, colon cancer, or non-small-cell lung cancer [11-13]. These criteria were designed so that the cohort would have an aggregate 6-month mortality rate of about 50%. Patients were excluded if they died or were discharged within 48 hours of hospital admission; if they were younger than 18 years of age; if, at the time of hospitalization, they were scheduled to be discharged within 72 hours; if they had the acquired immunodeficiency syndrome (AIDS) or multiple trauma; if they were pregnant; or if they did not speak English.

From February to November 1994, HELP enrolled persons who were 80 years of age or older at four institutions: Beth Israel Hospital, MetroHealth Medical Center, St. Joseph's Hospital, and the University of California at Los Angeles Hospital. Patients were included regardless of diagnosis, but they were excluded if they had elective surgery, were discharged within 48 hours of admission, had AIDS, or had multiple trauma. Institutions with a large volume of elderly admissions applied random sampling so that each of the four institutions enrolled an equal number of elderly patients.

Data Collection

Between 4 and 10 weeks after a patient's death, the patient's surrogate decision maker was interviewed about the circumstances of the death. The surrogate was defined as the person who would have made decisions if the patient had been unable to do so. During this interview, the surrogate reported the location in which death occurred; the patient's health status; the patient's ability to perform self-care activities, as measured by a modified version of the Katz Activities of Daily Living (ADL) scale [14]; the patient's consciousness and ability to communicate; the patient's symptoms; the presence of anxiety or depression in the patient, as measured by the Profile of Mood State subscales [15]; the patient's preferences about medical care; and the medical treatments used in the last 3 days of the patient's life. The interviews done for patients in SUPPORT phase I, SUPPORT phase II, and HELP were identical in timing and administration and were similar in content. Variations in interviews primarily reflect our increasing experience and are outlined below and in the Appendix Table.

For patients in SUPPORT phase I, surrogates were asked about the following symptoms: pain, nausea, confusion, shortness of breath, sadness, anxiety, and fatigue. The surrogate stated the amount of time in the patient's last 3 days of life during which the patient had the symptom; the response categories used were "not at all," "occasionally," "about half of the time," "most of the time," and "all of the time." For patients who had a given symptom during the last 3 days of life, the surrogate characterized the severity of the symptom as "not at all severe," "moderately severe," or "extremely severe." We classified a symptom as severe if the surrogate noted that the patient had the symptom one half of the time or more and the surrogate considered the symptom to be of at least moderate severity. For patients in SUPPORT phase II and HELP, surrogates were asked about only three symptoms: pain, nausea, and confusion. Only the SUPPORT phase I interviews asked whether the patients felt alone and isolated and how the patients felt about their illnesses during the last 3 days of life. To find out how the patients felt about their illnesses, we asked five questions about the degree to which the patients felt concern about needing help from others, were satisfied with their lives, had difficulty tolerating physical symptoms and problems, and thought themselves a burden. Surrogate responses stating that patients were "somewhat" or "very much" troubled are reported together to simplify presentation of the data. Questions about the patients' emotional state and symptoms were asked only for patients who were reported to have been conscious and able to communicate for at least some of the time during the last 3 days of life.

The SUPPORT phase II and HELP interviews included questions about the use and role of advance directives in medical decision making and about patients' preferences regarding care. Preferences about care were ascertained with the following question: "Thinking of the time from (patient's) last hospitalization to (his/her) death, did (he/she) prefer a course of treatment that focused on extending life as much as possible, even if it meant more pain and discomfort, or on a plan of care that focused on relieving pain and discomfort as much as possible, even if it meant not living as long?"

As follow-up to this question, surrogates were asked to state the degree to which the patients' preferences about medical treatment were followed. Surrogates chose one of the following responses: "a great deal," "very much," "moderately," "very little," or "not at all." We classified care as "potentially inconsistent" with patient preferences if the surrogate chose one of the latter three responses.

In all three data collection efforts, surrogates reported whether resuscitation was attempted at the patient's death and whether the patient received tube feedings in the last 3 days of life. A question about whether the patient was attached to a ventilator in the last 3 days of life was available only for the SUPPORT phase II and HELP cohorts.

Medical Record Data

Hospital records were abstracted by nurses who were trained at a central location and received ongoing supervision to ensure reliable data collection. Concurrent abstraction included the physiologic indicators of disease severity necessary for calculating a SUPPORT prognostic model [13]; the Glasgow coma scale score [16]; length of stay; and date of death, if the participant died in the hospital.

Statistical Analysis

We compared patients about whom interviews were given with patients about whom interviews were not given with regard to time course to death, sociodemographic characteristics, diagnosis, and disease severity. The chi-square test was used for nominal variables, and the Wilcoxon test was used for continuous variables. We examined surrogate reports of symptoms and patient conditions for the eight condition categories (acute respiratory failure and multiple organ system failure with sepsis were combined into one category because their survival trajectories are similar) and for patients 80 years of age and older.

Results

Top Methods Results Discussion Author & Article Info References

A total of 9105 seriously ill patients were enrolled in SUPPORT, and HELP enrolled 1176 elderly patients. A total of 4214 patients in SUPPORT died within 6 months (46%); 2338 of these deaths (55%) occurred during the enrollment hospitalization. Older persons enrolled in HELP were more likely than patients in SUPPORT to survive the enrollment hospitalization; only 72 (6%) died during this hospitalization. However, 336 patients in HELP (28%) died later, for a total of 408 deaths among patients in HELP (35%) within 1 year.

Previously reported analyses help to establish a perspective for the frequency of SUPPORT patients in hospitalized populations. About 3% of the patients admitted to the participating hospitals were enrolled in SUPPORT, and these patients accounted for about 19% of deaths in those hospitals [10]. Within a defined geographic area, the number of persons who would probably have been sick enough to be eligible for SUPPORT extrapolates to about 400 000 in the United States each year. Patients with SUPPORT-eligible diseases and severity also accounted for approximately 40% of deaths in that geographic area [17].

A total of 4622 patients died, and 3357 surrogates (73%) were interviewed. Patients did not have surrogates in 228 cases. To limit respondent burden, we excluded 368 surrogates who had completed an interview within 7 days before the patient's death. For the remaining 669 cases, either the surrogate refused the interview (n = 516), the physician refused the interview on behalf of the surrogate (n = 125), or the interview could not be done for other reasons (n = 28). The response rate was 83% after we excluded patients without surrogates and surrogates who were not approached for an interview because they had given an interview just before the patient's death.

The surrogate was a member of the patient's immediate family (spouse, child, parent, or sibling) in 89% of cases, another relative in 6% of cases, and someone other than a relative in 5% of cases. In general, patients without a surrogate interview were younger and had more ADL impairments than patients with an interview (Table 1). Otherwise, patients with interviews were similar to patients without interviews.

Site of Death

Mortality rates during the enrollment hospitalization varied across the condition categories: These rates were 70% to 81% for patients with nontraumatic coma and organ system failure and only 13% to 18% for patients 80 years of age and older and patients with colon or lung cancer (Table 2). Overall, 54% of the deaths occurred during the enrollment hospitalization. Almost half of the deaths (46%) among persons who survived the first hospitalization occurred during a subsequent hospitalization. The proportion of deaths that occurred in hospitals, nursing homes, hospices, and homes varied strikingly by condition category. Patients with colon cancer died at home in 45% of cases; in contrast, 16% of patients with acute respiratory failure or multiple organ system failure with sepsis, 19% of patients with cirrhosis, and 15% of elderly persons died at home. Patients enrolled in HELP were more likely to die in a nursing home (34%). An inpatient hospice was the site of death for about one tenth of patients with either colon cancer (11%) or lung cancer (10%).

General Condition

Almost half of the patients (45%) were unconscious throughout the last 3 days of life. The rate of unconsciousness varied between 20% (for patients with congestive heart failure or lung cancer) and 87% (for patients who entered the study with nontraumatic coma) (Figure 1). Most conscious patients were able to communicate with their surrogates (range, 48% for patients with acute respiratory failure or multiple organ system failure with sepsis to 77% for patients with congestive heart failure). Only 38 patients (12%) who had coma when they were initially enrolled were conscious before death, and only 13 of these patients (4%) were able to communicate with their surrogates.

View larger version (27K): [in this window] [in a new window]   Figure 1. Rates of consciousness and ability to communicate in patients 3 days before death, by medical condition, as reported by family members. MOSF = multiple organ system failure.

Family members reported that patients were "extremely" or "very" ill or disabled in 85% of cases (range, 64% for patients with congestive heart failure to 95% for patients with coma). With the exception of patients with congestive heart failure, patients were gravely ill in at least four of five cases, according to surrogate reports. For patients who were able to communicate, quality of life had been poor in two of three cases (67%).

Symptoms

Figure 2 shows the rates at which surrogates reported that patients had a given symptom at least half of the time at moderate or extreme severity. Almost 40% of conscious patients had severe pain (range, 34% for patients with chronic obstructive pulmonary disease to 46% for patients with colon cancer). With the exception of patients with colon cancer (and elderly persons, whose surrogates were not asked about pain), more than half of patients with serious illness had severe dyspnea in the last 3 days of life. Severe confusion was found in almost one fourth of all cases. In all condition categories, fewer than one in eight patients had severe nausea. Fatigue was the most prevalent symptom, affecting almost 80% of patients in each of the categories. Surrogates reported that almost three fourths of patients (73%) found it difficult to tolerate these physical symptoms.

View larger version (21K): [in this window] [in a new window]   Figure 2. Rates of severe symptoms in patients 3 days before death, as reported by family members. Family members of elderly patients were not asked about dyspnea. MOSF = multiple organ system failure.

Emotional State

In the Profile of Mood State subscales, patients are reported to be troubled by anxiety and dysphoria "not at all," "a little," "moderately," "quite a bit," or "extremely." They receive a score of 0 to 4 on each item, and these scores are then averaged over the items in the appropriate subscale. A score of 4 would result if the surrogate were to answer "extremely" to all of the items in one subscale. Although the average patient had scores that indicated only modest anxiety or depression (range, 0.7 to 1.7), about one fourth of patients in each category had scores greater than 2 (indicating at least moderate anxiety or depressed affect in the last 3 days of life). The most anxious or most depressed one tenth of the study sample had a mean score between "quite a bit" (a score of 3) and "extremely" (a score of 4). The elderly cohort had less anxiety and less depressed affect than the seriously ill patients enrolled in SUPPORT. In addition, 17% of family members reported that patients felt alone and isolated in the last 3 days of life. The proportion of patients who had this sense of isolation ranged from fewer than 6% (for patients with colon cancer) to 21% (for patients with acute respiratory failure or multiple organ system failure or chronic obstructive pulmonary disease). Almost two thirds of family members (62%) reported that patients found it difficult to tolerate these emotional symptoms.

Preferences and Treatments

Most patients reportedly preferred an approach to care that focused on comfort, even if this approach shortened life (Figure 3). Surrogates reported that actual care was at odds with patient preferences in about 1 in 10 cases (range, 3% in patients with cirrhosis to 17% in patients with chronic obstructive pulmonary disease). Life-sustaining treatments were often used in the last 3 days of life (Figure 4). About 1 in 10 patients (11%) had a resuscitation attempt (range, 3% for patients with colon cancer to 15% for patients with congestive heart failure); 1 in 4 were on a ventilator (range, 2% for patients with colon cancer to 37% for patients with acute respiratory failure or multiple organ system failure with sepsis); and 4 in 10 patients had a feeding tube (range, 10% for patients with lung cancer to 61% for patients with acute respiratory failure or multiple organ system failure with sepsis). Overall, 56% of patients received at least one of these three life-sustaining treatments during the 3 days before death.

View larger version (21K): [in this window] [in a new window]   Figure 3. Rates of preferring comfort care rather than life-extending care and rates of perceiving that care was not in accord with patient preference for comfort care in the 3 days before death, as reported by family members. MOSF = multiple organ system failure.

View larger version (28K): [in this window] [in a new window]   Figure 4. Treatments administered during the last 3 days of life, as reported by family members. MOSF = multiple organ system failure.

Discussion

Top Methods Results Discussion Author & Article Info References

Most of the elderly and seriously ill patients in our study died in an acute care hospital, often with severe pain, shortness of breath, dysphoria, and other symptoms that were difficult to tolerate. Surrogates usually reported that the patients preferred treatments that focused on comfort, even if the use of these treatments shortened life. Nevertheless, many patients were treated aggressively. Only 44% died without being placed on a ventilator, receiving a feeding tube, or undergoing an attempt at resuscitation. Most, by far, died in hospitals. Compared with younger patients, elderly persons died in hospitals less often but in nursing homes more often. Of course, deaths that occur at home may not be better than deaths that occur elsewhere. Only hospice programs have ensured a focus on symptoms, control by the patient and family, and attentiveness to emotional needs; however, hospice was used by only a small proportion of patients.

Except for data on mortality rates, the medical literature includes little information about dying. The last substantial study of sequential hospital deaths in the United States was done in 1908 by Osler, who studied 486 deaths at John Hopkins Hospital [6, 18]. Osler summarized his findings with this reassuring statement: "Ninety suffered bodily pain or distress ... eleven showed mental apprehension. ... The great majority gave no sign one way or the other; like their birth, their death was a sleep and a forgetting." A British study [4] described the experience of dying in a geographically defined population, and the National Institute on Aging researched the experience of dying in a small area of Connecticut [5]. Little else has been done. Data with which to answer reasonable patients' questions (such as, "How often do people with my illness die on a ventilator?" or "How does my fatal illness affect families?") are not available. Only the growing literature on palliative care for patients with cancer [19, 20] provides tested and reliable scientific information. Our cultural inclination to disavow dying may have encouraged inattention to this research.

Since the turn of century, the usual time courses and causes of death have changed dramatically. At that time, physicians could do little to alter the rapidly fatal course of infectious disease or injury. The risk for death was almost equal in every decade of life after the first. Even cancer and heart disease appeared suddenly and killed quickly. Tuberculosis killed slowly, but its infectivity led patients with this condition to be excluded from the mainstream of the community [21]. Today, most persons will have lived for months, if not years, with the illness that results in their death. Most die when they are old. Dying has become a gradual period of deterioration, and the once familiar boundaries between "being sick" and "dying" have been obscured. The "final common pathway" of disability and decline makes establishing causes and correcting physiology less important than maintaining function and mitigating spiritual, social, and emotional needs.

Our study relies on the reports of surrogates, most of whom were close family members. These surrogates may have under- or over-reported patients' symptoms and preferences for palliative care. The turmoil of bereavement may have colored the families' perceptions in ways that we cannot estimate. Because most patients could not be interviewed near the time of death, our knowledge of their experiences inevitably relies on reports by family members or caregivers. Reports from family members are important in themselves as well as in estimating patients' experiences. Dying is not only what the patient experiences but also what the family remembers. The words of Dame Cicely Saunders, the founder of the modern hospice movement in England, underscore this observation: "How people die remains in the memories of those who live on" [22]. The problems that families report warrant attention, regardless of whether the patients would have described their dying differently.

Our study has other limitations. We interviewed family members for 73% of the patients who died, and the interviewed sample over-represents older persons and persons with less functional impairment. Patients were recruited at five teaching hospitals; different results might have been obtained at community hospitals or among persons who were not hospitalized at the time of a serious illness. However, a recent national survey [23] found that the final days of life for 502 persons were marked by similar rates of pain and limited physician-patient communication, regardless of whether patients were treated at teaching or community hospitals.

For patients who were conscious and able to communicate, our results raise serious concerns. Severe fatigue and dyspnea were common in the last 3 days of life. Previous research found that 70% of patients with cancer had dyspnea in the last 6 weeks of life [24], although this symptom was of moderate severity in only 28%. Our study found that almost 70% of conscious patients who died with lung cancer or multiple organ system failure with a malignant condition had severe dyspnea (that is, dyspnea of moderate or extreme severity for at least half the time). Although fatigue might be mitigated by attentive care, dyspnea can be alleviated more directly [25, 26].

More than one in three conscious patients had severe pain, even in disease categories in which severe pain might not have been expected. Pain, unlike fatigue and dyspnea, is amenable to treatment. The rates of pain relief that could be achieved with optimal care are unclear, but two hospice physicians report that hospice programs fail to relieve pain in no more than 2% of cases [27, 28]. Our study did not fully characterize the pain experienced by study patients, nor did it evaluate the alleviation of that pain. Previous research suggests that inadequate professional training and dysfunctional aspects of delivery system organization are important barriers to effective pain management [29, 30]. The pain associated with cancer has been discussed publicly and in the professional literature. We found that the rates of severe pain with other fatal illnesses were similar. Developing treatments for pain in patients with fatal illnesses other than cancer and implementing effective palliative care should become priorities.

Mild dysphoria or anxiety is common in the last 3 days of life. One fourth of our seriously ill patients had moderate anxiety or dysphoria, and the most afflicted one tenth were very troubled. These psychological symptoms might be alleviated with social support or pharmacologic treatment.

Reports about whether hospice care improves quality of life during dying are mixed [31-33]. The one randomized trial that has been done showed no difference in pain control, symptoms, or levels of depression or anxiety, although hospice patients were more satisfied than nonhospice patients [32]. A reanalysis of the National Hospice Study found that hospice patients had a better "quality of dying" in the last 3 days of life than did patients receiving conventional care [31]. Our study had too few hospice patients to allow an examination of the merits of hospice care, although the low rates of hospice use are troubling in themselves.

Family members stated that patients generally wanted medical treatment that focused on comfort in the last days of life. However, many patients wanted more treatment, even at the risk of discomfort. If this reflects informed and deeply committed preferences, public policy will have to come to terms with the expenses and suffering that will be engendered by honoring these wishes. However, communication about patient preferences and prognoses was uncommon [10]. If preferences for aggressive care commonly arise from a misunderstanding of or an inattention to likely outcomes, then improving decision making might be a high priority. Further inquiry is needed to understand the moral importance of preferences for aggressive care on the part of dying patients and their families.

Improving the experience of dying will undoubtedly require several simultaneous reforms. Expanding hospice programs [34, 35] and incorporating hospice concepts into broader programs targeted at persons with disabling, eventually fatal conditions [36] might be effective. Giving a prominent role to care of the dying in caregiver education might enable change [37]. It is likely that quality of care at the end of life will have to be routinely measured and that the public and other health care purchasers will have to demand and pay for adequate performance before lasting and sustained improvements are possible.

From George Washington University Medical Center, Washington, D.C.; Beth Israel Hospital, Boston, Massachusetts; Johns Hopkins University, Baltimore, Maryland; Marshfield Clinic and Marshfield Medical Research Foundation, Marshfield, Wisconsin; Department of Veterans Affairs Hospital, White River Junction, Vermont; University of California at Los Angeles Medical Center, Los Angeles, California; and MetroHealth Medical Center, Cleveland, Ohio.

Dr. Phillips: Department of Clinical Epidemiology, Department of Medicine, Beth Israel Hospital, 330 Brookline Avenue, Boston, MA 02215.

Dr. Wu: Health Services Research Center, Johns Hopkins University, 624 North Broadway, Room 633, Baltimore, MD 21205.

Dr. Desbiens: Department of General Internal Medicine, Marshfield Medical Research Foundation, 1000 North Oak Avenue-1R5, Marshfield, WI 54449-5790.

Dr. Claessens: Outcomes Research Group, Department of Veterans Affairs Hospital, White River Junction, VT 05009.

Dr. Wenger: Center for Health and Ethics, University of California at Los Angeles School of Medicine, Room B-564/Factor, Los Angeles, CA 90095.

Dr. Connors: Division of Pulmonary Disease and Critical Care, Cleveland MetroHealth Medical Center, 2500 MetroHealth Drive, H-323, Cleveland, OH 44109-1998.