Needs Assessment
Writing A Business Plan
Delivery Models
Inter-Disciplinary Team
Program Operations
Quality Assessment

Rethinking Fundamental Assumptions: SUPPORT's Implications for Future Reform

Explores the possibility that improved individual, patient-level decision-making is not the most effective strategy for improving end-of-life care and that improving routine practices may be more effective.

Joanne Lynn, MD, MA, MS

BACKGROUND: The intervention in SUPPORT, the Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments, was ineffective in changing communication, decision-making and treatment patterns despite evidence that counseling and information were delivered as planned. The previous paper, in this volume, shows that modest alterations in the intervention design probably did not explain the lack of substantial effects.

OBJECTIVE: To explore the possibility that improved individual, patient-level decision-making is not the most effective strategy for improving end-of-life care and that improving routine practices may be more effective.

DESIGN: This paper reflects our efforts to synthesize findings from SUPPORT and other sources in order to explore our conceptual models, their consistency with the data and their leverage for change.

RESULTS: Many of the assumptions underlying the model of improved decision-making are problematic. Furthermore, the results of SUPPORT suggest that implementing an effective intervention based on a normative model of shared decision-making can be quite difficult. Practice patterns and social expectations may be strong influences in shaping patients' courses of care. Innovations in system function, such as quality improvement or changing the financing incentives, may offer more powerful avenues for reform.

CONCLUSIONS: SUPPORT's intervention may have failed to have an impact because strong psychological and social forces underlie present practices. System-level innovation and quality improvement in routine care may offer more powerful opportunities for improvement. J Am Geriatr Soc 48:S214-S221, 2000.

The Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments . SUPPORT . has been at once productive and perplexing, especially with regard to its finding that an intervention intended to improve decision-making was completely ineffectual. In a companion article in this volume, we report and assess others' explanations of why the intervention did not succeed. We conclude that those explanations cannot adequately account for the intervention's failure. Clearly, the intervention did not improve communication about preferences and outcomes or practices concerning patient decision-making. Perhaps the SUPPORT intervention was not forceful enough or clever enough to persuade physicians to work with their patients to make better decisions. Alternatively, however, the complete ineffectiveness of the intervention could point to a different interpretation.

SUPPORT built on the widespread perception, both then and now, that (1) each patient's course of care results from the interaction between each patient's illness and the decisions about his or her care; (2) those decisions largely require physicians; and (3) improved individual, patient-level decision-making will enhance both process and outcomes for patients. Improved decision-making requires incorporating the patient's ratings of the desirability of various courses of action. In turn, the physician needs to know the patient's evaluation of the merits of alternatives, requiring effective communication between patient and physician.

However, we contend here that the course of care may be determined largely by pre-existing routine, that patient centered decision-making is often difficult to implement, and that improving the experience of patients might best be achieved by changing institutional and professional routines.

We will present the essential elements of this argument, acknowledging from the outset that our data cannot prove our contentions. Our aim is to show that a shared decision making model is not always the key goal, that another conceptual framework might offer greater opportunity for improvement, and that SUPPORT'S findings are consistent with this alternative model. We begin by characterizing the "decision making model" that underlay the SUPPORT intervention. We then examine its potential shortcomings and propose an alternative "patterned care model" that is consistent with our data. Finally, we will outline the implications of that model for reform.


At the time we designed SUPPORT, most research on outcomes focused on survival and prognosis for survival, and most studies used a simple model relating baseline status to outcome, e.g:

Baseline characteristics (disease and physiology) =>

Likelihood of survival to a particular time or event (1)

The SUPPORT team recognized a number of shortcomings in this model, including that it overlooked a "black box" of decisions and treatments between the baseline and the outcomes:

Baseline characteristics =>

Decisions and Treatments => Outcomes (2)

The intervention was motivated by the observation that, if one started from the same baseline but had the benefit of better decisions, one might achieve better outcomes.

Baseline characteristics =>

Usual Decisions and Treatments=> Usual Outcomes (3)

Baseline characteristics =>

Better Decisions and Treatments => Better Outcomes (4)

Medical, nursing, legal, and ethical literature, both then and now, assumed that decisions drive treatments and arise from decision-making that can be deficient and that can be improved. Those authorities generally espoused a model of shared decision-making in which good decision-making is marked by the patient's authority to decide and his or her understanding of likely outcomes. In that model, achieving better outcomes requires enhancing participants' understanding of outcomes and preferences before making and implementing shared decisions, reflecting patients' informed preferences and physician counseling. The SUPPORT intervention did not involve formal decision analysis but did follow the decision analysis paradigm: the best course to implement is the one with the highest product of (1) the probability of achieving a set of outcomes and (2) the patient's assessment of the desirability of those outcomes. Toward that end, the SUPPORT intervention aimed to facilitate these key components of decision-making:

  • Understanding of likely outcomes by each patient (and family) and physician;
  • Understanding of current patient preferences by each patient (and family) and physician;
  • Utilization of preferences expressed previously by the patient if a patient becomes incompetent; and
  • Counseling each patient (and family) so they can make choices aligned with patient preferences.

Many commentators have not realized that SUPPORT tried to improve decision-making, not end-of-life care per se. The observant leader may question why pain relief was a measured outcome of the project. In Phase I, the project uncovered a disturbingly high prevalence of serious pain affecting patients from all sites and involving most of the diseases. As investigators and clinicians, we felt obliged to address this issue, but we were limited by having planned a randomized trial. Addressing pain comprehensively would have contaminated the control group and would thus complicate evaluation of the intervention's effect on decision making. Furthermore, treating pain comprehensively would have diverted resources away from improving decision making. After consultation with advisors, we decided that addressing pain only among the intervention patients would be a suitable compromise. That strategy would allow SUPPORT to shed light on pain management, provide an entree into the clinical setting for the SUPPORT intervention nurses, and respond to the substantial rates of pain among patients. Thus, the intervention added a focus on pain relief by assessing pain, encouraging patients to seek better pain relief, and having the SUPPORT nurse ensure access to existing pain relief services. However, the intervention's primary aim remained to improve the decision-making process, which in turn was expected to improve decisions and treatments and thus improve outcomes.


We suggest that there may have been serious inadequacies in this model regarding how the course of care comes to be and, therefore, how an intervention might succeed. The shared decision-making model of patient care is predicated on several assumptions, among which are the following:

  1. Patients' preferences about outcomes are stable (for long enough to act on), important to them and expressible upon inquiry. Instead, we found that patients' preferences often evolved as they confronted situations and that their preferences were often quite labile and difficult to articulate completely. This was the case, in part, because people find it very hard to imagine themselves in a future state with unfamiliar symptoms and more severe illness and debilitation and to imagine or evaluate the desirability of death. With unstable and poorly articulated patient preferences, it is difficult to ascertain which course of action would maximize a patient's expected utility.
  2. All parties concerned would have to recognize the need to decide. Instead, as the patient's body failed, situations resolved themselves through predictable, routine behaviors. Indeed, certain expected behaviors were called "decisions" even though there was no consideration of any alternative. Do-not-resuscitate (DNR) "decisions" near the time of death were a prominent example: by the time questions about cardiopulmonary resuscitation were raised, DNR was likely to be a foregone conclusion in most participants' minds. With no actual decision to make, there would be no change with an intervention based on improving decision-making.
  3. Whoever is the responsible decision-maker would have to be willing and able to take responsibility for the resulting outcomes. Instead, we found that people often delayed or dodged making a choice, perhaps because the task was unfamiliar or because they feared subsequent regret. If no one wants to accept responsibility for making a decision, there may be less motivation to weigh alternatives, and an intervention based on improved decision-making might be harder to implement.
  4. All responsible decision-makers would have to be willing and able to deal with information in a rational way, to advance the patent health-related interests, and to consider all of the important outcomes, costs and benefits. Instead, we found multiple distorting elements that precluded considering these issues as dispassionately as the expected utility calculation for a decision analytic model requires. For example, patients and providers were often willing to let things unfold as they usually do (rather than to "rock the boat"). Patients were often concerned with how they would be seen by others or how their relationship with others would be remembered. Many people (patients and providers) did not want to talk about death, or they dealt with life, death and disease in non-rational terms. For example, some patients framed their experience in fatalistic or magical ways that would not be easy to incorporate in a decision analytic model.

We now discuss each of these four elements in more detail.

Decision-Making Model Assumption: Patient Preferences are Stable, Important and Expressible

The model used in SUPPORT and, indeed, normative decision-making models in general, assume that patients can articulate preferences and make choices congruent with them. In formal decision analysis, the utility of each outcome is multiplied by its likelihood of occurrence, and this product is compared with other options. Even if formal decision analysis is not performed, ascertaining a patient's relative preferences for various outcomes should be part of the decision-making procedure."

Unfortunately, the preferences of patients confronting end-of-life care often evolve or are constructed during the situation rather than accessed from a stable set of settled priorities." Preferences are often not fixed characteristics that people can report easily in a reliable and valid manner; preferences are often highly dependent on 13.14 For example, McNeil et al. "have shown that patient preference for radiation versus surgical treatment for cancer is influenced by whether the potential outcomes are expressed in terms of mortality or survival. These two variables are perfectly inversely related, so the patient's preference should be unaffected by the semantic framing. However, because patients' preferences are actually affected, a physician may find it very difficult to have confidence in an analysis based on a patient's preference assessed with only one phrasing. Semantic framing is not the only way in which preferences can be shown to be liable. Many SUPPORT patients were willing to answer questions such as whether they would prefer to die rather than live with some adverse condition.

However, most had not had that condition, many who were currently facing the condition had avoided contemplating it, and virtually no one who said that he or she would rather die than endure a particular outcome actually sought to die when that outcome occurred. Patients and families often seemed to be at a loss in attempting to express their preferences, often checking with people around them for cues as to how one should feel and what responses were conventional."

A second problem, which complicates assessing patients' preferences for outcomes of end-of-life care, pertains to the relationship between predicted utility and experienced utility. Predicted utility represents one's beliefs about the utility of some future outcome. Experienced utility is the actual utility of the experience associated with that outcome when it transpires." In end-of-life care, the patients usually have very little experience with the potential outcomes. Hence, a given patient's predicted utility may not be congruent with his or her eventual experienced utility, even though the predicted utility serves as the basis for the patients' decision-making. If the predicted utility were routinely different from the experienced utility, patients' interests would not be well served by ensuring their ability to make and enforce plans for future states. Finally, people near death may well have quite different values from persons who are not near death."

Practitioners working with people who are dying often describe patients who say that their "quality of life is very poor" but also that, "this is the best time of my life," or "these last few weeks have been so very valuable." Patients may be referring to the meaningfulness they have found in religion, relationships or contemplating the mystery of life. They may now see a value to their lives that was missing before or rediscover long-held tenets and commitments.'" In any case, they often attribute great worth to an existence that would seem to be of little value when evaluated by others in the usual ways. Although clinicians caring for terminally ill patients see this phenomenon often, it is not likely that a patient can anticipate such a substantial shift in this or other values when still remote from the experience.

Decision-Making Model Assumption: A Decision Opportunity Will Be Recognized

In order for decision analysis to be applied, one must be aware that there is a decision to be faced. If no need for a decision is perceived at a particular juncture, then probabilities and utilities are not germane. For example, if a physician tells a patient that his or her disease is normally treated in a particular way, then the patient may never consider probabilities and utilities, because there seems to be no decision to be made. Such situations are quite common. While the course of care might be characterized, especially in retrospect, as resulting from a choice, it may be that the patient and everyone else involved really perceived it as simply "what is done." Joe Cruzan, commenting on his search to have artificial feeding stopped for his daughter Nancy said that the events felt more like a "wagon picking up speed going downhill." SUPPORT patients may have felt the same way - as if they were riders on the roller coaster of illness and their care system's responses. They did not often note that coming to the hospital was itself a decision or that there might exist alternatives to the "usual course of care."

The same description may apply to physicians' decision-making. Klein has promulgated the concept of "recognition-primed decision-making" to describe decision-making in situations involving firefighters who must make quick life-and-death decisions when they themselves are in conditions of extreme danger. Klein reports that firefighters do not ponder probabilities and utilities; they would have neither the time nor the inclination to use such information even if it were readily available. Instead firefighters attempt to characterize each situation. When they recognize a fire as a particular type, they then take the course of action that such a fire calls for. This recognition-primed decision-making does not explicitly employ the two analytic inputs to normative responses.

Even a poor prognostic estimate provided to the physician by the SUPPORT investigators will not be relevant to the daughter's decision-making. Decision-making based on rule-following is evaluated by how closely one follows the rule, not by expected utility. Is forgoing major surgery for one's parent seen as an appropriate decision for a loving clild? If not, and the child is motivated by rule following, then that option will not be considered seriously or its consideration will not be on the rational grounds of expected utility in terms of health outcomes.


The goal of SUPPORT was to improve individual, patient-level decision-making because we and others thought that enhanced decision-making would lead to better treatments and outcomes. Guiding the SUPPORT intervention were certain assumptions about decision-making: patients would be able to articulate preferences that would be stable and important; all parties would acknowledge that decisions were needed at particular junctures; patients or their surrogates would step forward and assume responsibility for the decision and its outcome; and decision-makers would appropriately incorporate all relevant information, including prognostic estimates and patient preferences.

We now conclude that these assumptions were naive. We do not deny that appropriately considering prognostic estimates and patients' utilities could lead to better treatments and outcomes. We suggest, however, that people tend not to consider end-of-life care in these terms, and the SUPPORT intervention failed to encourage decision-makers to do so. Hence, improving end-of-life care may require means other than enhanced, shared decision-making. We suggest that to achieve better treatments and outcomes, one should consider improving practice patterns directly rather than relying on enhanced decision-making.


A more fundamental re-interpretation of SUPPORT may help achieve better outcomes. Many have claimed that the culture of medicine is antithetical to good care at the end-of-life. In this view, improving patient-physician communication cannot ameliorate shortcomings in care; improvement requires restructuring the entire environment. Many of these commentators point to the physicians' drive to use technology to fight disease, even if against the wishes of patients. As Callahan put it, medical culture and the healthcare environment encourage heroic and aggressive treatment, which engenders deafness to patients' wishes and a tendency to discount prognostic information. Marshall's pointed to physicians' desire to preserve life (take action, cure disease, overcome helplessness incurred by illness) rather than to cultivate sensitivity to emotional aspects of a person's approaching death. In accord with Marshall's assessment, Annas contended that the pressure to use technology leaves the physician with little time for conversation with families or patients. He contends that such attitudes are so pervasive in hospitals that "the only realistic way to improve the care of dying patients in the short run is to get them out of the hospital, and to keep them from going to the hospital, at the end of their life." Both Dugan and Godec speculated that financial incentives often motivate hospitals and doctors to provide invasive interventions to dying patients whether they want them or not. "By contrast, relieving pain and allowing patients to die in peace and dignity are not significantly compensable in the US medical system. In essence, then, these critiques charge that, in a medical (hospital) culture that values technology over discussion, the SUPPORT intervention was too limited in scope to yield real change. These critics seem to agree that rather than working within the existing structure (the status quo), a more systematic, far-reaching, and deep-seated plan to restructure the hospital environment is needed. As Brody puts it, "The system is so entrenched, so powerful and (not coincidentally) so rich that present-day reformers who would try to alter it usually find the task much more difficult than they anticipated. This happened to the SUPPORT investigators."

Insofar as these critics envision rescuing dying persons from a hostile environment by discharging them from the hospital, they disregard SUPPORT'S findings regarding the difficulty of identifying this class of "dying" persons. SUPPORT-like patients with very poor prognoses ordinarily die quickly, often deliberately forgoing aggressive treatment. SUPPORT patients very near death usually did not undergo CPR; the overall rate of CPR was 11% by family reports and 14% by medical record review for those who died in SUPPORT hospitals. However, individuals just a little farther from death usually have quite uncertain prognoses. During what turned out to be their last week, most persons had a prognosis of at least a 50% likelihood to live two months or more. Even among individuals with progressive chronic illness such as colon cancer or congestive heart failure, the remaining life is often comfortable and valued. Thus, very few SUPPORT patients could readily be classified as "dying" in the sense that they - or anyone else - would have found it appropriate to accept death and be treated for symptoms at home. Instead, they were thought to have a substantial chance to leave the hospital and do well for a while if treatment were successful. Most of our patients were in the "middle muddle" of prognoses, bad enough to be at risk of death but good enough to hope for longer survival with the appropriate treatment. Thus, one cannot expect to improve on SUPPORTS findings simply by splitting off the dying and treating them somewhere else.

Nevertheless, the observation that the hospital or intensive care environment would need pervasive change for the SUPPORT intervention to succeed presents some considerations. First, it signals that the course of care arises from number of arrangements, many of which are not well articulated, most of which did not arise specifically from considerations of the patient's good, and few of which are expected to be responsive to different decisions by different patients. Second, the course of care is a function of the characteristics of the region's healthcare system. Third, these arrangements may be hard to change because they are generally efficient, predictable, and comforting to all involved. Most involved seem to cherish the comfort of predictable patterns, the efficiency of well-rehearsed behaviors and the reassurance that they are not taking on substantial responsibility for novel approaches.

Indeed, the narratives written by the intervention nurses in SUPPORT used some illuminating phrases. For example, the nurses repeatedly mentioned the "readiness" of patients or families when considering the opportunity to "make decisions." When we sought what the patient or family member was to be "ready" for, it seemed that they were getting "ready" for a change in course that signaled acknowledgement of impending death or some other undesired outcome. The nurses' claim seemed almost like etiquette: that it was gracious and humane to give patients and families the time to recognize what was in store. The patient's failing course would eventually force the issue, but the appearance of participants making choices was still valued, even if it only meant agreeing sooner rather than later that death or another undesired outcome was imminent.

In another example, we found many situations in which the patient or family member explains his or her motives as "trying to be a good _______," with the blank being father, daughter or some other social role. Although the process may have the same end result, reassurance that one is behaving virtuously is different from seeking the best expected utility for the patient's health-related outcomes. Family members were often unsure about what constituted reliable landmarks as they sought to define their roles. The guidance of their physician was often highly valued, but such guidance could be unintelligible if it spoke to utility rather than virtue or expected behavior. The SUPPORT nurse's guidance was probably also trusted, but it was subject to the same mismatch of language and, in addition, the potential for suspicion of an unfamiliar party providing counsel. In any case, if the patients and family members were looking for reassurance that their behavior would be seen as virtuous, accommodation to the usual pattern of care is one useful way to accomplish that. Myriad cues provide reassurance to a family member who is "going with the flow," and, conversely, myriad cues raise doubt in a family member who is "doing something differently."

We contend that the course of care for each patient operates more as a glide path in which all parties concerned watch the patient for signs that they should act in certain ways. When the patient becomes more difficult to arouse and acknowledges that the end is near, family and care providers feel that it "is time" for a DNR "discussion," which acts as the symbol that death cannot be averted. The family's refusal of a DNR order is, then, just a delay and is accepted as long as there is likely to be time for reconsideration. Most families eventually follow the script and "make a decision for DNR," which may be more an acknowledgement of the patient's physical decline than a decision.

The "default option," or what is "usually done," is very powerful. If the default option at a particular hospital is a certain pattern of aggressive care, then the salience and timing of various "decisions" will reflect that milieu. What keeps patients and families from deviating from strong glide paths? Patients and families may well be comforted by the assurance that they are good people that arises from following the expected pathways. Deviations foster such disincentives as anticipatory regret or worry about being a "good daughter."

The information available is not usually so revealing that it compels deviating from the usual glide path. Prognoses are "middling," and patient preferences are not firm or well known. A patient with a strong viewpoint can be accommodated, but the usual patient and family will "go along," trusting that the course of care has been honed over many patients. Patients may well trust that the usual course of care will serve patients "like me" better than anything that would result from trying to customize each choice to reflect their own values.

For SUPPORT patients with fatal illnesses, the intervention not only did not improve provider-patient communication; it also simply did not affect the assumptions about the appropriateness of the default option, whatever it may have been. Perhaps patients rarely had strong and relevant preferences; they usually had muddled prognoses for survival; their quality of life before hospitalization was generally acceptable, and rarely did alternatives seem clearly better or worse to the participants than usual care. The individual, patient-level decision-making model is constrained by system factors that are clearly beyond the control of the patient, such as availability of supportive care at home. The five SUPPORT sites differed dramatically in rates of dying at home versus in the hospital. Indeed, such variations were pervasive in the

Medicare population generally, and differences across regions (from 29% to 66%) correlated strongly with hospital bed supply and patterns of utilization (r2 = 0.83)." In the SUPPORT hospital that had the highest rate of out-of-hospital death, physicians usually followed their patients at home or in nursing home or hospice settings, employers were generally willing to grant leave for family caregivers, and active hospice services were readily available. In contrast, the hospital with the highest rate of in-hospital deaths had very few physicians who followed patients at home, had many family members faced with losing their jobs if they became primary caregivers and had underdeveloped, competitive and largely unfamiliar hospice and home-care services. These phenomena show the SUPPORT decision-making model to be naively optimistic. No patient could simply choose to have their care system behave like the first care system just described, nor, for that matter, like the last one. The range of choices was often limited to the timing of DNR orders, with virtually everyone (patient, family, physicians and others) agreeing that resuscitation at the time of death would not serve the patient. This is a weak choice, although alternatives with greater impact were largely determined by the care system.

How can we think about such change? Certainly, merely signing advance directives would not accomplish much," and neither would encouraging patients one by one to demand change. Instead, change could be conceptualized more as a public health issue: How could care be organized to serve seriously ill people best? This question is very different from SUPPORTS framework, which might be characterized as: How can each patient's course of care reflect that patient's likely outcomes and informed preferences?


Playing out the contrast between the patterned care model, which leads to a public health approach, and the shared decision-making paradigm has substantial implications for reform of end-of-life care. First, we have to acknowledge the limited role that patient self-determination is likely to have, not because self-determination is unimportant to patients but because the care system is not able to respond adequately." Instead, although accommodation can be made for a few patients with unusual, and unusually strong, convictions, the usual course of care will remain the usual course of care. Unlike the major reforms in obstetric care in the 1970s, in which enough patients were able to "vote with their feet" and refuse obstetric care not meeting their demands it seems unlikely that patients as sick as those in SUPPORT would be well served by refusing to go to the hospital.

Fortunately, however, the power of the "usual course of care" may also point toward avenues for improvement. For years, the legal profession tried to persuade Americans to write a will. Eventually, most states resolved most of the problem by changing the law concerning what happens to an estate with no will Most people have rather similar preferences, and those were codified into law. Now, for most modest estates, not much turns on whether a will was written. Perhaps those who want reform in health care will do well to think in the same ways. Rather than trying to have patients, one by one, state their preferences and shape their own health care, reform should first center on ensuring that persons who do not explicitly state their wishes still get treatment that is close to what most people would want. There would be room for customizing, but perhaps most of what can be achieved depends on getting the default glide path to be much more congruent with what serves most people best.

That default glide path arises from several forces, a particularly worthy one being evidence about which practices tend to serve the population best. One of the questions for reform is to ensure that the routine practices give the time and attention to individual decision-making at the junctures where doing so serves the patient population best. Rather than aiming to make customized decisions on every issue, a good default glide path would encourage decision-making primarily when the issue is important and carries quite divergent outcomes, the desirability of which is evaluated quite differently by affected patients. Thus, the care system might operate as if on auto-pilot to treat acute respiratory failure but might build in a substantial opportunity for considering a decision to reintubate a person struggling with advanced emphysema. The optimum default glide path would be defined, in a sense, by having the best overall utility for a population of people with that condition.

For example, this approach would take into consideration that patients with cancer generally want their pain treated promptly and appropriately. Reform would make competent pain management the usual course. Then, routine assessment of pain, prompt and pre-emptive drug and non-drug interventions and patient and family education about how to adjust treatments would become high priorities. The occasional patient who wants no opioids or who wants to try unusual approaches could still be accommodated in such a care system, but the person who is just "going with the flow," or is confused or uncertain, will nonetheless be assured pain relief.

In another example, the person who has very poor cardiac function and widely metastatic cancer would be advised that CPR is contraindicated and, therefore, will be withheld unless that person has a strong reason to want it. In order to solicit an early DNR order, however, the current symbolic uses of DNR would require attention. Perhaps other actions could symbolize that a person is now close to death: e.g., calling the family together, advising to consider "last words," and so forth. Perhaps there would need to be overt disarticulation of the symbolism of DNR, with physicians and other providers educating families and patients about the best uses and merits of CPR and its commonplace uselessness in some circumstances. In addition, one might set out to establish different and more useful rituals. Such suggestions would need to be tried out and assessed. The heart of strategic reform in end-of-life care may be widespread, carefully evaluated, well networked improvement in the routine functioning of care systems. Rather than assuming, as SUPPORT did, that utility maximization for each patient can be accomplished by close adherence to a decision analytic model at the individual level, reformers might do better to adopt a model of continuous quality improvement at the systems level.

Default paths can be structured that maximize utility summing over all patients, still allowing for customization of care for those patients who wish to do so. Such an approach requires repeated efforts at improvement, with evaluation and redirection. A number of recent reports show remarkable effects of such strategies, which stand in striking contrast to the utter failure of the SUPPORT intervention.

Innovation, evaluation and system change are not how our financing system, medical ethics and case law have construed reform for end-of-life care. Instead, social policy has construed patient care as resulting from decision-making concerning each patient. While Medicare pays for each service to each individual patient. Medicare generates no measures of overall quality of care for very sick populations.

While medical ethics addresses the individual's right to choose, there is strikingly little authoritative guidance on what choices the care system should make available and what the course of care should be in the absence of choice. Case law has added to this emphasis by posing issues as "the right to refuse treatment" or the "right to choose." General welfare, costs, social meanings or system function have little role in litigation arising from a particular patient's care.


The intervention in SUPPORT did not yield any measurable improvements in decision-making for seriously ill hospitalized adults. Most of the explanations offered seem unconvincing for such a thoroughly ineffective effort. However, we now question the fundamental assumption: that the course of care for the seriously ill hospitalized patient is the result of individual, patient-level decision-making that could be improved with better counseling and information. Instead, the course of care may well be shaped largely by how the care system is organized and by the interpersonal meanings ascribed to various cues and signals that shape the predictable patterns of care. Research might be more fruitful if it shed light on the psychological, sociological and anthropological factors that shape the course of care. Future interventions may be more effective if they address system change and quality improvement.


The authors acknowledge our debt to the thousands of patients, family members, and physicians who participated in the SUPPORT project, to the dozens of nurses and data collectors who made it work, and to the investigators and colleagues who have encouraged us to learn as much as possible from the project. We are deeply indebted as well to the project's sponsor, The Robert Wood Johnson Foundation, which not only provided the resources for SUPPORT but has also launched a major national campaign to direct public attention and clinical service innovation toward problems affecting the dying.


1. The SUPPORT Principal Investigators. A controlled trial to improve care for seriously ill hospitalized patients: The study to understand prognoses and preferences for outcomes and risks of treatment (SUPPORT).

2. Lynn J, DeVries KO, Arkes H et al. Ineffectiveness of the SUPPORT Intervention: Review of explanations. J Am Geriatr Soc 2000;48(iuppl):S206213.

3. Making Health Care Decisions. The President's Commission for the Study of Bioethical Problems in Medicine and Biomedical and Behavioral Research. US Government Printing Office, Washington, DC, October 1982.

4. New York State Task Force on Life and the Law. Do Not Resuscitate Orders: The Proposed Legislation and Report of the New York State Task Force on Life and the Law. 1986.

5. American Geriatrics Society Ethics Committee The care of dying patients. A position statement from AGS. J Am Geriatr Soc 1995;43:577-578.

6. American Medical Association, Guidelines for the appropriate use of do-not-resuscitate orders. JAMA 1991;265:18ff8-1871.

7. ACP Ethics Committee, American College of Physicians Ethics Manual, 3"* Ed. Ann Intern Med 1992:117:947-960.

8. Committee on Ethics in End-of-life Decision-Making. Understanding the goals of care. American College of Obstetrics and Gynecology Committee Opinion 19951156:1-6.

9. Kasper JF, Mulley AG Jr. Wennberg JE. Developing shared decision-making programs to improve the quality of health care. QRB 1992;18:183-190.

10. Destnens NA, Vu AW. Pain and suffering in seriously ill hospitalized patients. J Am Geriatr Soc 2000;48(suppl):S183-186.

11. Tsevat J, Cook F, Green ML et al. Health values of the seriously ill. Ann Intern Med 1995,122:514-520.

12. Slovic P. The construction of preferences. Am Psychol 1995;50:364-371.

13. Forrow L. The green eggs and ham phenomenon. Hastings Cent Rep 1994; 24(special supplement):S29-32.

14. Callahan D. Once again, reality: Now where do we go? Hastings Cent Rep 1995;25(special supplement):S33-36.

15. McNeil BJ, Pauker SG, Sox HC Jr. Tversky A. On the elicitation of preferences for alternative therapies. N EnglJ Med 1982;306:1259-1262.

16. Drought, TS, Koenig BA, Raffin TA, Advance directives. Changing our expectations (editorial). Chest 1996;110;589-591.

17. Kahneman D, Snell J. Predicting a change in taste: Do people know what they will like? J Behav Decis Making 199215:187-200.

18. Byock 1. Dying Well. New York: RiverHead Books, 1997.

19. Cruzan V. Director, Missouri Department of Health, 497 U.S. 261 (1990).

20. Klein GA. A recognition-primed decision (RPD) model of rapid decision making. In: Klein GA, Orasanu J, Calderwood R. Zsambok C, eds. Decision Making in Action: Models and Methods. Norwood, NJ: Ablex, 1993, pp 138-147.

21. Hammond KR. Human judgment and social policy: Irreducible Uncertainty, Inevitable Error, Unavoidable Justice. New York: Oxford University Press, 1996.

22. Hern HE Jr. Koenig BA, Moore LJ, Marshall PA. The difference that culture can make in end-of-life decision-making. Camb Q Health Ethics 1998:7:2740.

23. Paris JJ, MuirJ, Cameron, Reardon FE. Ethical and legal issues in intensive care. J Intensive Care Med 1997;12:298-309.

24. Samuelson W, Zeckhauser R. Status quo bias in decision making. J Risk Uncertainty 1988:1:7-59.

25. Prilchard RS, Fisher ES, Teno JM et al. Influence of patient preferences and local health system characteristics on the place of death. J Am Geriatr Soc 1998:46:1242-1250.

26. Landman J. Regret: The persistence of the possible. Oxford: Oxford University Press, 1993.

27. Bell-DE. Regret in decision making under uncertainty. Operat Res 1982:30: 961-981.

28. Bell DE- Disappointment in decision making under uncertainty. Operat Res 1984:32:1-27.

29. Loomes G, Sugden R. Regret theory: An alternative theory of rational choice under uncertainty. Econ J 1982:92:805-824.

30. Landman J. Regret and elation following action and inaction: Affective responses to positive versus negative outcomes. Personality Soc Psychol Bull 1987:13:524-536.

31. Spranca M, Minsk E, Baron J. Omission and commission in judgment and choice. J Exp Social Psychol 1991:27:76-105.

32. Hamm RM, Scheid DC, Smith WR, Tape TG. Opportunities for applying psychological theory to improve medical decision making. Two case histories. In G.B. Chapman, F. Sonnenberg, eds. Decision Making in Health Care: Theory, Psychology, and Applications. New York: Cambridge University Press, in press.

33. Pauker SG, Kassirer JP. Decision analysis. New Engi JMed 1987:5:250257.

34. Stiggelbout AM, Kiebert GM, Kievit J et al. The utility of the time trade-off method in cancer patients: Feasibility and proportional trade-off. J Clin Epidemiol 1995:48:1207-1214.

35. Stiggelbout AM, de Hae; JCJM, Kieben GM et al. Tradeoff between quality and quantity of life: Development of the QQ questionnaire for cancer patient attitude. Med Decis Making 1996:16:184-192.

36. Chapman GB, Elstein AS, Andrews A et at. Prostate cancer patients' utility for health states: How it looks depends on where you stand. Med Decis Making 1998:18:278-286.

37. Glasziou PP, Bromwich S, Siroes RJ. Quality of life six months after myocardial infarction treated with thrombolytic therapy. Med J Aust 1994:161: 532-536.

38. March JG. A Primer of Decision Making: How Things Happen. New York: FrA Press, 1994.

39. Annas GJ. How we die. Hastings Cent Rep 1995;25 (special supplement): S12-14.

40. Brody H. The best system in the world. Hastings Cent Rep 1995;25(special supplement's 18-21.

41. Dugan DO. SUPPORT. Asking different questions. Making the rounds in healthcare. Faith Ethics 1996:1:1-2.

42. Codec MS. The SUPPORT Project and Improving Care for Seriously III Patients. JAMA (letter) 1996:275:1228.

43. Lo B. End-of-life care after termination of SUPPORT. Hastings Cent Rep 1995;25(special supplement):S6-8.

44. Lo B. Improving care near the end of life: Why is it so hard? JAMA 1995; 272:1634-1636.

45. Marshal PA The SUPPORT Study: Who's talking? Hastings Cent Rep 1995; 25(special supplement):S9-ll.

46. Miller FG, Fins JJ. A proposal to restructure hospital care for dying patients. New Engi J Med 1996:334:1740-1742.

47. Moskowin EH, Nelson JL. The best laid plans. Hastings Cent Rep 1995; 25(special supplement):S3-5.

48. Sachs GA. The SUPPORT Project and Improving Care for Seriously III Patients. JAMA (letter) 1996:275:1229.

49. Schneider CE. From consumer choice to consumer welfare. Hastings Cent Rep 1995;25(special supplement):S25-28.

50. Solomon MZ. The enormity of the task: SUPPORT and changing practice. Hastings Cem Rep 1995;25(special supplement):S28-32.

51. Solomon MZ, Jennings B, Dickey N et al. The SUPPORT Project and Improving Care for Seriously III Patients. JAMA (letter) 1996;275.

52. Berwick DM The SUPPORT Project. Lessons for action. Hastings Cent Rep 1995;25(special supplement):S21-22.

53. Lynn J, Harrell F Jr. Cohn F et al. Prognoses of seriously ill hospitalized patients on the days before death: Implications for patient care and public policy. NHoriz 1997:5:56-61.

54. Lynn J. An 88-year-old woman facing the end of life. JAMA 1997:277:1633-1640.

55. Teno JM, Murphy D, Lynn J et al, for the SUPPORT Investigators. Prognosis-based futility guidelines: Does anyone win? J Am Geriatr Soc 1994:42:1202-1207.

56. Lynn J, Teno JM, Phillips RS et al. Perceptions by family members of the dying experience of older and seriously ill patients. Ann Intern Med 1997; 126:97-106.

57. Goodlin S, Zhong Z, Lynn J et al. Factors associated with use of cardiopulmonary resuscitation in seriously ill hospitalized adults. JAMA 1999:282: 2333-2339.

58. Lynn J, Harrell FE Jr. Cohn F et al. Defining the "Terminally Bl:" Insights from SUPPORT. Duq Law Rev 1996:25:311-336.

59. Jaagosild P, Dawson NV, Thomas C et al. Outcomes of acute exacerbation of severe congestive heart failure: Quality of life, resource use, and survival. Arch Intern Med 1998:158:1081-1089.

60. Murphy P, Kreling B, Kathtyn E et al. Description of the SUPPORT Intervention. J Am Geriatr Soc 2000;48(suppl):S154-161.

61. Teno JM, Lynn J, Wenger N, Phillips RS et al. Advance directives for seriously ill hospitalized patients: Effectiveness with the patient selfdetermination act and the SUPPORT Intervention. J Am Geriatr Soc 1999; 45:500-507.

62. DuPen SL, DuPen AR, Polissa N et aL Implementing guidelines for cancer pain management. Results of a randomized controlled clinical trial. J Clin Oncol 1999:17:361-370. .

63. Lynn J, Shuster JL, The Center to Improve Care of the Dying, and The Institute for Healthcare Improvement. Improving Care at the End of Life-. A Sourcebook for Managers and Providers. New York: Oxford University Press, 1999, in press.

64. The Quality Letter for Health Care Leaders, Vol. 10, No. 10, October 1998. Canitol Publications- a division of Asoen Publishers. Inc.. Alexandria, VA.

Annotated Bibliography

The Collected Reports from SUPPORT and HELP: An Annotated Bibliography of Manuscripts in print as of December 31, 1999, and SUPPORT Prognostic Formulas
Nancy Freeborne, MPH, PA-C

"The Study to Understand Prognoses and Preferences and Risks of Treatment, SUPPORT, enabled a variety of researchers, including those from five clinical sites and the National Coordinating Center at George Washington University, to observe and comment on many aspects of the care and outcomes of very sick hospitalized patients. An allied project, the Hospitalized Elderly Longitudinal Project, HELP, studied the course of hospitalized persons aged 80 years or older.

SUPPORTS basic approach and methods were outlined in a special edition of a journal that included data collection instruments such as questionnaires and medical record abstraction forms: Murphy DJ, Cluff LE, eds. SUPPORT: Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments, Study Design. J Clin Epidemiol 1990;43:iS-123S.

Public Use Data regarding SUPPORT is available from the Inter-University Consortium for Political and Social Research (http://www.icpsr.umich.edu/index.html).

The SUPPORT Physiology Score and Prognostic Models are presented in Appendix B. The original prognostic model had a typographical error in the formula which is corrected in this notation. A national public opinion poll concerning experiences with end-of-life care is reported in Appendix A.

Before the collection of articles published in a supplement to the Journal of the American Geriatric Society, one overview and interpretation was published by a lead investigator: LynnJ. Unexpected returns: Insights from SUPPORT. In: Isaacs SL et aL, eds. To Improve Health and Health Care: The Robert Wood Johnson Foundation Anthology (JosseBass Health Series). New York: Josse-Bass, 1997, pp 161186.

Following below are citations to all of the articles published from SUPPORT and HELP before December 31,1999, along with brief summaries of their approach and findings.

These are listed alphabetically by first author. Despite having a common database, the reader will find that the research topics are varied and that many journals are represented.


1. Arkes HR, Dawson NV, Speroff T et al. and the SUPPORT Investigators. The covariance decomposition of the probability score and its use in evaluating prognostic estimates. Med Decis Making 1995;15:120-131.

The covariance decomposition of the mean probability score has not previously been used to evaluate medical judgment. Authors analyzed prognostic estimates made by three groups within SUPPORT: physicians, their patients, and patients' surrogate decision-makers. The decomposition reveals that while the physicians have the best overall estimation performance, their levels of bias and scatter are not always

Superior to those of the other groups. This is because physicians' prognostic estimates are pessimistic, and the patients' estimates are generally optimistic (thereby achieving low scatter). Authors conclude that the decomposition is a useful tool with which to analyze medical judgment.

2. Connors AF Jr., Dawson NV, Thomas C et al. for the SUPPORT Investigators. Outcomes following acute exacerbation of severe chronic obstructive lung disease. Arn J Respir Crit Care Med 1996;154:959-967.

A cohort of patients with exacerbation of COPD (n = 1,016) were studied for 6 months. Mortality was high at 60 days, 180 days, I year, and 2 years (20%, 33%, 43%, and 49%, respectively) despite a relatively low index hospitalization mortality rate (11%). At 6 months, only 26% of the cohort were alive and able to report a good, very good, or excellent quality of life.

3. Connors AF, Speroff T, Dawson NV et al. for the SUPPORT Investigators. The effectiveness of right heart catheterization in the initial care of critically ill patients.JAMA 1996;276:889-897.

Investigators reviewed records of SUPPORT patients having undergone right heart catheterization (RHC) and compared them with comparable matched controls. This observational study suggested that patients with RHC had an increased 30-day mortality. Also, the mean cost per hospital stay was $49,300 in patients with RHC and $35,700 in patients without RHC.

4. Cooper GS, Bellamy P, Dawson NV et al. A prognostic model for patients with end-stage liver disease. Gastroenierology 1997;113:1278-1288.

Charts were reviewed on patients with end-stage liver disease in Phase 1 (295) and Phase II (243). Cumulative incidence of death was 30% at 30 days and 50% at 6 months. In 295 Phase I patients, time until death was associated independently (P < .01) with five factors measured on study Day 3: renal insufficiency, cognitive dysfunction, ventilatory insufficiency, age >= 65 years, and prothrombin time >= 16 seconds. A model based on these risk factors stratified 243 patients in Phase II into three groups, with cumulative incidences of death at 30 days of 12%, 40%, and 74%.

5. Covinsky KE, Goidman L, Cook EF et al. The impact of serious illness on patients' families. JAMA 1994;272:1839-1844.

Seriously ill patients and their family members (n = 2129) were interviewed about the impact of the patient's illness on the family. Severe care-giving and financial burdens were common. One-third (34%) of patients required substantial care-giving help from a family member. For 20% of the patients, a family member had to quit work or make another major life change because of the time required to care for the patient. Loss of most or all of the family savings was reported by 31%; 29% reported loss of the major source of family income and 17% reported that the illness required a major change in family plans. Factors associated independently with the loss of family savings included poor functional status, lower family income, and young age.

6. Covinsky KE, Landefeld S, Teno J et al. for the SUPPORT Investigators. Is economic hardship on the families of the seriously ill associated with patient and surrogate care preferences? Arch Intern Med 1996;156:1737-1741.

Two months after index hospitalization, patients and surrogates (n == 3158) were interviewed about whether the patient's preference for care focused on maximizing comfort or on maximizing life-expectancy. Answers were compared with responses about whether the illness had caused financial hardship within the family. Respondents who reported family financial stress as a result of the illness were significantly more likely to express a preference for comfort care over life-extending care in two out of three age strata (42% vs 32% in those aged <45 years; 69% vs 62% in those aged 265 years). However, economic hardship on the family did not affect either the frequency or direction of patient-surrogate disagreements about whether the goal of care should be focused on maximizing comfort or prolonging life (73.7% agreement between patients and surrogates in those not reporting economic hardship, and 72% agreement in those reporting economic hardship).

7. Covinsky KE, Wu AW, Landefeld CS et al. Health status versus quality of life in older patients: Does the distinction matter? Am J Med 1999;106:435-440.

Two months after a hospitalization, 493 cognitively intact patients older than age 80 were asked to describe their health status using scales that described four domains (physical capacity, limitations in performing activities of daily living, psychological distress and pain). Patients were also asked to rate their quality of life. Each of the four health status scales was significantly correlated with patients' global quality of life. However, for a substantial number of patients, scores on the health status scales did not accurately reflect their global quality of life. The results suggest that whereas health status may be a marker of the quality of life for populations of patients, assumptions about the quality of life of individual patients should not be based on their health status alone.

8. Desbiens NA, Wu AW, Broste SK et al. for the SUPPORT Investigators. Pain and satisfaction with pain control in seriously ill hospitalized adults: Findings from the SUPPORT research investigations. Crit Care Med 1996:24:1952-1961.

A total of 5176 of 9105 patients or surrogates were interviewed about their experiences with pain while in the study. Nearly 50% of patients reported pain. About 15% reported extremely severe pain or moderately severe pain occurring at least half the time, and nearly 15% were dissatisfied with pain control.

9. Desbiens NA, Mueller-Rizner N, Connors AF, Wenger NS. The relationship of nausea and dyspnea to pain in seriously ill patients. Pain 1997;71:149-156. A cohort of patients (n = 1556) were interviewed approximately 8 days after study admission about the frequency and severity of their nausea, dyspnea and pain.

Ordinal logistic regression was used to test the independent association of pain with nausea and dyspnea. With increasing levels of nausea and dyspnea, patients reported more pain.

10. Desbiens NA, Mueller-Rizner N, Connors AF Jr, et al. for the HELP Investigators. Pain in the oldest-old during hospitalization and up to one year later. J Am Geriatr Soc 1997;45:1167-1172.

Patients 80 years and older or their surrogates were interviewed about level of pain during hospitalization (n = 806 patients), at 2 months (n = 614 patients), and at 12 months (n = 416 patients). In this cohort of oldest-old hospitalized patients, 45.8% complained of pain, and 19.0% reported extremely severe pain or moderately severe pain occurring at least half of the time. Pain levels appeared to be equal to those of SUPPORT patients who were younger and sicker.

11. Desbiens NA, Wu AW, Aizola C et al.for the SUPPORT Investigators. Pain during hospitalization is associated with continued pain six months later in survivors of serious illness. ArnJ Med 1997;102:269-276.

Patients or surrogates were interviewed about patient level of pain during hospitalization. Investigators used separate ordinal logistic regressions to compare levels of pain over time. Patients who reported high levels of pain during hospitalization were likely to report high levels after hospitalization.

12. Desbiens NA, Wu AW, Masui Y et al. and the SUPPORT Investigators. Patient empowerment and feedback does not decrease pain in seriously ill hospitalized adults. Pain 1998;75:237-246.

Of 4804 Phase n patients, 2652 received the care of an intervention nurse who assessed pain) educated patients and families about pain control, empowered patients to expect pain relief, informed patients' nurses and physicians about level of pain, and suggested or used other pain management resources. Interviewers assessed patient pain level from interviews with patients or surrogates at baseline and 2 and 6 months later. Overall, 50.9% of patients reported some pain. Intervention patients did not report a significant improvement in level of pain compared to control patients.

13. Desbiens NA, Mueller-Rizner BS, Hamel MB, Connors AF Jr. for the SUPPORT Investigators. Preference for comfort care does not affect the pain experience of seriously ill patients. J Pain Symptom Manage 1998; 16:281-289.

Patients (n = 2820) responded to interview questions about willingness to live permanently in pain, and choice of a course of treatment aimed at extending life or relieving pain.

They answered these questions at an early interview time (between study days 2 and 6) and at a later time (between days 8 and 10 in Phase I, and 8 and 12 in Phase II). Of the 2820 patients, 1388 (49.2%) reported that they preferred a course of care aimed at relieving pain, and 1426 (50.6%) were very unwilling or would rather die than be in permanent pain. At the later interview, 655 (23.2%) reported that they had extremely severe or moderately severe pain.

14. FitzGerald JD, Wenger NS, Califf RM et al. for the SUPPORT Investigators, functional status among survivors of in-hospital cardiopulmonary resuscitation. Arch Intern Med 1997;157:72-76.

Functional status data were available both before and after CPR for 162 SUPPORT patients. Investigators performed logistic regression using specific variables to evaluate predictors of worse functional status 2 months after CPR. Of the 162 patients, 91 (56%) had preserved or improved functional status and 71 (44%) had worse functional status after CPR. Patients with worse functional status were extremely disabled and less likely to survive to hospital discharge or to 6 months.

15. Galanos AN, Pieper CP, Kussin PS et al. for the SUPPORT Investigators. Relationship of body mass index to subsequent mortality among seriously ill hospitalized patients. Crit Care Med 1997;25:19621966

Body mass index (BMI) was calculated for 3002 Phase I patients. Investigators reported that even when controlling for multiple disease states and physiologic variables and removing from analysis all patients with significant prior weight loss, a BMI <= 15th percentile remained a significant and independent predictor of mortality.

16. Goodlin SJ, Zhong Z, LynnJ et al. factors associated with use of cardiopulmonary resuscitation in seriously ill hospitalized adults. JAMA 1999; 282:2333-2339.

Investigators performed a secondary analysis of SUPPORT and HELP data to identify factors associated with use of cardiopulmonary resuscitation(CPR). Of 10,281 SUPPORT and HELP patients, 2505 experienced cardiopulmonary arrest and 514 of these received CPR. Multivariable analysis revealed that use of CPR was more likely in men (odds ratio (OR) 1.39; 95% confidence interval (Cl), 1.121.73), younger patients (OR per 10-year increase 0.90; 95%d, 0.84-0.96), blacks (OR 1.76; 95%CI, 1.33-2.34), patients whose reported preferences were for CPR (OR = 2.60; 95%CI, 1.91-3.55), who reported better quality of life (OR 1.49; 95%CI, 1.10-2.03), or who had higher physician estimates for 2-month survival (OR per 10% increase 1.14; 95%CI, 1.09-1.19). The likelihood of having CPR varied by hospital site and diagnosis. Patients with congestive heart failure had a much higher likelihood of having CPR (adjusted OR 3.31; 95%CI, 2.12-5.15) compared with patients with acute respiratory failure or multiple organ system failure.

17. Haidet P, Hamel MB, Davis RB et al. for the SUPPORT Investigators. Outcomes, preferences for resuscitation, and physician-patient communication among patients with metastatic colorectal cancer. Am J Med 1998;105:222229.

Researchers studied a cohort of 520 patients with metastatic colon cancer. Patients were interviewed about functional status, whether any weight loss had occurred, anxiety, depression, quality of life and the patient's estimates for their probability of 2- and 6- month survival. Patients or surrogates were interviewed about patient CPR preference and whether they had discussed their preference with their doctor. Quality of life and functional status were high at study entry and remained so at 2 and 6 months. Of the 339 patients or surrogates who expressed a preference about CPR, 127 (37%) noted a preference not to have CPR; of these, 69 did not have do-not-resuscitate order.

18. Hakim RB, Teno JM, Harrell FE Jr. et al. for the SUPPORT Investigators. Factors associated with do-not resuscitate orders: Patients' preferences, prognoses and physicians judgment. Ann Intern Med 1996:125:284-293.

Patients or surrogates (n = 6802) were interviewed about patients' cardiopulmonary resuscitation preferences and answers were correlated with disease severity and timing of DNR orders. Investigators concluded that patients' preferences for cardiopulmonary resuscitation and poor short-term prognosis were associated with timing of DNR order. Increased age was frequently associated with DNR order regardless of prognosis.

19. Hamel MB, Goidman L, Teno JM et al. Identification of comatose patients at high risk for death or severe disability. JAMA 1995;273:1842-1848.

A total of 596 of the SUPPORT patients had non-traumatic coma. Primary causes of coma were cardiac arrest (31%) or cerebral infarction or intra-cerebral hemorrhage (36%). On the third day after study enrollment, five clinical variables were independently associated with 2-month mortality. Mortality at 2 months for patients with four or five of the following risk factors was 97%: abnormal brain stem response, absent verbal response, absent withdrawal response to pain, creatinine level greater than or equal to 1.5 mg/dL, and aged 70 years or older.

20. Hamel MB, Phillips RS, Teno JM et al for the SUPPORT Investigators. Seriously ill hospitalized adults: Do we spend less on older patients? J Am Geriatr Soc 1996;44:1043-1048.

To assess resource utilization, records were reviewed for 4301 patients using a modified version of the Therapeutic Intervention Scoring System (TISS), for performance of three invasive procedures (major surgery, dialysis, and right heart catheterization) and for estimated hospital costs. Results were compared in patients aged older and younger than SO. Investigators noted that older patients were less likely than younger patients to undergo major surgery (3% vs 8%), dialysis (1% vs 8%), and right heart catheterization (17% vs 34%). Older patients had median estimated hospital costs that were lower than those of patients who were younger ($9,422 vs $22,339). These differences persisted after adjustment, suggesting that the results were not related to differences in patients' severity of illness or preferences for life-extending care.

21. Hamel MB, Phillips RS, Davis RB et al. for the SUPPORT Investigators. Outcomes and cost-effectiveness of initiating dialysis and continuing aggressive care in seriously ill hospitalized adults. Ann Intern Med 1997;127:195-202.

Records were reviewed to identify patients who had begun dialysis while in SUPPORT. Charts of this cohort (490 patients) were reviewed and investigators determined that the median survival after dialysis initiation was 32 days. Only 27% of patients were alive 6 months after dialysis was begun.

Patients who survived had a median of one dependency in activities of daily living, and 62% of survivors rated their quality of life as "good" or better. The overall estimated cost per quality-adjusted life-year (QALY) saved by providing dialysis was $128,200, which is well above $50,000 per QALY, an amount cited commonly for cost-effective care. However, the cost per QALY for the 94 patients with the best prognosis was $61,900.

22. Hamel MB, Teno JM, Goldman L et al. Patient age and decisions to withhold life-sustaining treatments from seriously ill, hospitalized adults. Ann Intern Med 1999; 130:116-125.

Among SUPPORT patients for whom each treatment issue arose, decisions were made to withhold ventilator support for 30% (1600 of 3571), to withhold surgery for 13% (375 of 2982) and to withhold dialysis for 29% (380 of 1298). Analysis revealed that decisions were more common in older age groups even after adjustment for socio-demographic characteristic, prognoses, baseline function, patients' preferences for life-extending care and physicians' understanding of patients' preferences for life-extending care.

23. Hamel MB, Davis RB, Teno JM et al. Older age, aggressiveness of care, and survival for seriously ill hospitalized adults. Ann Intern Med 1999;131:721-728.

Investigators reviewed the records of the 9105 SUPPORT patients to test the hypothesis that older age is associated with less aggressive treatment and higher short-term mortality caused by serious illness. Authors noted that mean patient age was 63, 44% of patients were female, and 16% were black. They found that adjusted estimates of age specific 6-month mortality rates were 44% for 55-year-old patients, 48% for 65-year-old patients, 53% for 75-year-old patients, and 60% for 85-year-old patients. Acute physiology and patients' diagnoses were more highly correlated with prognosis than patient age.

24. Harrell FE, Jr., Lee KL, Mark DB. Multivariable prognostic models: Issues in developing models, evaluating assumptions and adequacy and measuring and reducing errors. Stat Med 1996;15:361-387.

Authors discuss an easily interpretable index of predictive discrimination and methods for assessing calibration of predicted survival probabilities. Since multivariable regression models are used often in clinical outcome studies, authors review the drawbacks to having poorly fitted or over-fitted regression models and recommend a modeling strategy that avoids many of the problems discussed.

25. Hiltimen EF, Puopolo AL, Marks GK et al. The nurse's role in end-of-life treatment discussions: Preliminary report from the SUPPORT project. J Cardiovasc Nurs 1995;9:68-77.

Of staff nurses interviewed (n = 1369), 65% reported having knowledge of their patients' CPR preference. Nurses were significantly more likely to have discussed CPR preferences when a patient did not want CPR (32%) than when a patient did want CPR (13%). They were also significantly more likely to have discussed CPR when they had known the patient longer and when the patient was conversant and cognitively intact. This article describes the purposes of Phase II and the role of the nurse facilitator but does not describe

Phase II results. After 18 months of Phase II, SUPPORT nurse activities aimed at increasing communication were focused on participating in team rounds, family conferences and chart documentation.

26. Hiltunen EF, Medich C, Chase S et al. Family decision making for end-of-life treatment: The SUPPORT nurse narratives. J Clin Ethics 1999; 10:126-134.

During SUPPORT, 18 specially trained nurses had contact with 2519 seriously ill patients and their families. Authors of this article review narratives from these encounters, focusing on 75 incidents representing decisional conflict.

They found that families were involved in three-quarters of the decisions, often because of the severity of patients' illnesses. In addition, authors noted that nurses found that the decision-making process was often quite complex.

27. Hofmann JC, Wenger NS, Davis RB et al. for the SUPPORT Investigators. Patient preferences for communication with physicians about end-of-life decisions. Ann Intern Med 1997;127:l-12.

Phase II patients (n = 1832) were interviewed about their preferences for CPR. Researchers found that of those who responded to the question on CPR preferences (n = 1589), only 366 (23%) had discussed preferences with their physicians. Of the 1223 patients who had not discussed their preferences with their physician, 707 (58%) were not interested in doing so.

28. Jaagosild P, Dawson NV, Thomas C et al. Outcomes of acute exacerbation of severe congestive heart failure: Quality of life, resource use, and survival. SUPPORT Investigators. The Study to Understand Prognosis and Preferences for Outcomes and Risks of Treatments. Arch Intern Med 1998;158:1081-1089.

Of the 9105 patients enrolled in SUPPORT, 1404 patients were admitted with acute exacerbation of severe congestive heart failure. Researchers reviewed the charts and interviewed 1390 patients or surrogates. Analysis revealed that survival was 93.4% at discharge from the index hospitalization, 72.9% at 180 days, and 61.5% at I year. Of patients interviewed at 180 days, the median health rating on a scale of 0 to 100 (0 indicates death; 100, excellent health) was 60, and 59.7% were independent in their activities of daily living. However, patients with worse functional status were more likely to die.

29. Kennard MJ, Speroff T, Puopolo AL et al. Participation of nurses in decision making for seriously ill adults. Clin Nurs Res 1996:5:199-219.

In the Phase I portion of SUPPORT, a cohort of nurses (n = 696) were interviewed about their knowledge of patients' preferences for care. Patients, surrogates and physicians were also interviewed. Many nurses reported having no (31%) or little (36%) knowledge of their patients' preferences. While nurses and physicians did not think that the nurse had much influence on decision-making for the patient, patients and surrogates were more likely to report that conversations with nurses were helpful in deciding which type of care the patient would receive.

30. Knaus WA, Harrell FE, Lynn Jet al. The SUPPORT prognostic model: Objective estimates of survival for seriously ill hospitalized adults. Ann Intern Med 1995;122:191203.

SUPPORT investigators hypothesized that accurate prediction of risk for death might help physicians in clinical decision making. Researchers developed a prognostic model using the following variables: diagnosis, age, number of days in hospital before study entry, presence of cancer, neurologic function and II physiologic measures recorded on Day 3 after study entry. Analysis revealed that the prognostic model was generally equal to physician clinical prediction, and that the two used together yielded more accurate estimates.

31. Krumholz HM, Phillips RS, Hamel MB et al. for the SUPPORT Investigators. Resuscitation preferences among patients with severe congestive heart failure: Results from the SUPPORT project. Circulation 1998; 98:648-655.

Of 9105 patients, 1404 were hospitalized with severe exacerbation of congestive heart failure (New York Heart Association class IV, or clinical heart failure and an ejection fraction < 20%). Of the 1404 patients hospitalized, 936 participated in interviews about resuscitation preferences. Of these patients, 215 (23%) did not wish to be resuscitated in case of cardiac arrest. Only 25% of patients had discussed their preferences with their physician. Physicians did not perceive their patient's preference correctly 24% of the time.

32. Layde PM, Beam CA, Broste SK et al. Surrogates' predictions of seriously ill patients' resuscitation preferences. Arch Fam Med 1995;4:518-523.

Researchers compared patient preferences for CPR with surrogate belief of CPR choice in 1226 patient-surrogate pairs. Surrogates stated patient wishes incorrectly in 50% of the cases where the patient did not wish to have CPR. For patients who preferred CPR, only 16% had discordance with their surrogate.

33. Layde PM, Broste SK, Desbiens N et al. Generalizability of clinical studies conducted at tertiary care medical centers: A population-based analysis. J Clin Epidemiol 1996;49:835-844

This article discusses the results of a population-based ancillary study in the Marshfield Epidemiologic Study Area (MESA) in which one of the SUPPORT hospitals resides. Researchers attempted to identify nursing home residents, hospice patients and outpatients who were residents of the MESA who would have been eligible for the main SUPPORT study if they had been hospitalized. Reviewing their data, they were able to estimate that 400,000 patients per year would fulfill SUPPORT eligibility criteria in the United States, but an estimated 925,000 patients each year have SUPPORT-like illnesses and do not receive intensive care.

34. Lynn J,Johnson J, Levine RJ. The ethical conduct of health services research: A case study of 55 institutions' applications to the SUPPORT project. Clin Res 1994;42:310.

Fifty institutions unsuccessfully applied to be participants in the multicenter health services research project, SUPPORT. Authors analyze these institutions' approach to defining informed consent and comment on perceived risks to participants: patients, surrogates, physicians. The troubling boundary between quality improvement and research and the ethics of human subjects protection is a special focus.

35. LynnJ, Teno JM, Harrell FE. Accurate prognostications of death: Opportunities and challenges for clinicians.

This article describes the importance of prognostication in seriously ill patients and makes recommendations for translating a prognostic model into clinical practice.

36. Lynn J, Harrell FE Jr., Cohn F et al. for the SUPPORT Investigators. Defining the "Terminally Ill:" Insights from SUPPORT. Duquesne Law Rev 1996;2S:311-336.

This article describes three possible strategies for defining persons who are terminally ill. Definitions of terminal illness are helpful in developing public policy, but this article suggests that any definition will exclude some persons with true terminal illness and include some without such illness. Because of the inability to accurately prognosticate, the authors conclude that it would be inappropriate to allow for physician-assisted suicide using a standard definition of terminal illness.

37. Lynn J, Harrell F Jr., Cohn F et al. Prognoses of seriously ill hospitalized patients on the days before death: Implications for patient care and public policy. New Horiz 1997;5:56-61.

The median SUPPORT model-generated likelihood of surviving 2 months was 17% the day before death and 51% I week before death in SUPPORT. For an APACHE HI based model, the predicted probability of survival was 14% the day before death and 45% I week before death. Different diseases had dramatically different prognoses near death.

38. Lynn J, Teno JM, Phillips RS et al. for the SUPPORT Investigators. Perceptions by family members of the dying experience of older and seriously ill patients. Ann Intern Mcd 1997;126:97-106.

Surrogate decision-makers of patients in SUPPORT, Phases I and II, and HELP were interviewed after a patient's death about the circumstances of the death. Overall, surrogates reported that 55% of patients were conscious in the last 3 days of life and that pain, dyspnea and fatigue were common in these patients. In 59% of the cases, patients had preferred a treatment plan that focused on comfort but actual care was not consistent with preferred care 10% of the time.

39. Lynn J, Zhong Z, Dawson NV et al. Physician experience caring for dying patients and its relationship to patient outcomes. J Palliat Care 1998;337-346.

Attending physicians (n = 765) were interviewed about their experiences caring for dying patients. Oncologists and pulmonologists or critical care physicians were more likely to have cared for dying patients. Physician interview responses were compared with interviews of surrogates. Analyses showed that physicians' experience with dying had no significant effect on patients' experience with pain, anxiety or depressive symptoms in their last 3 days of life.

40. Marbella AM, Desbiens NA, Mueller-Rizner N, Layde PM. Surrogates' agreement with patients' resuscitation preferences: Effect of age, relationship and SUPPORT intervention. J Crit Care 1998;13:140-145.

This article reviews 717 Phase II patient-surrogate pairs and their agreement or disagreement about CPR preferences. Of these, 386 were assigned to specially trained nurses who spent extra time with patients and families explaining prognoses and treatments. Rates of agreement in resuscitation preferences increased by 4.6% from the study Day 3 interview to the Month 2 interview; however, there was no significant difference between intervention and control groups.

41. Matrimore 1J, Wenger NS, Desbiens NA et al. Surrogate and physician understanding of patients' preferences for living permanently in a nursing home. J Am Geriatr Soc 1997:45:818-824.

Patients' willingness to live permanently in a nursing home was measured on a 5-point scale. The following results were noted in 3262 Phase I and II patients: 7% were "very willing" to live permanently in a nursing home, 19% "somewhat willing," 11% "somewhat unwilling," 26% "very unwilling," and "30% would rather die." Overall, surrogate perceptions of patient willingness to live permanently in a nursing home were similar to patient preferences. Physicians were able to identify patient preference correctly 67% of the time.

42. Oliverio R, Fraulo B. SUPPORT Revisited: The Nurse Clinician's Perspective. Holist Nurse Pract 1998;13:l-7.

This article presents personal accounts of the nurse clinician's role in SUPPORT. The two authors, who worked as nurse clinicians at one of the SUPPORT hospitals, describe dealing with seriously ill patients in SUPPORT. Reflecting on their experiences, they make recommendations for education about death and dying and for the role of nurse clinicians in the hospital setting.

43. Phillips RS, Hamel MB, Tcno JM et al. for the SUPPORT Investigators. Race, resource use, and survival in seriously ill hospitalized adults. J Gen Intern Med 1996;ll:387-396.

Investigators measured resource use in the SUPPORT cohort. They found that, in adjusted analyses, blacks were less likely to receive five procedures (operation, dialysis, pulmonary artery catheterization, endoscopy, and bronchoscopy) on study Days I and 3, and blacks had lower Therapeutic Intervention Scoring System scores, indicating lower resource use. However, blacks had slightly better survival, adjusted for severity.

44. Phillips RS, Wenger NS, Teno JM et al. for the SUPPORT Investigators. Choices of seriously ill patients about cardiopulmonary resuscitation: Correlates and outcomes. Am J Med 1996;100:128-137.

A cohort of patients was interviewed about preferences for CPR, quality of life, functional status, perceptions of prognosis and whether patients had discussed CPR with physicians. Of those interviewed (n = 1650), 28% did not want CPR. Of these, only 48% reported having discussions with their physicians about CPR. Overall, only 29% of patients had discussed their preferences with their physicians.

Factors associated with not wanting CPR included: hospital site, diagnosis, being older, being more functionally impaired and patient perception of a worse prognosis. Patients who did not prefer CPR had a lower intensity of care, similar inhospital mortality and higher mortality at 2 and 6 months following study entry.

45. Pritchard RS, Fisher ES, Teno JM ct al. Influence of patient preferences and local health system characteristics on the place of death. J Am Geriatr Soc 1998; 46:1242-1250.

Researchers reviewed places of death (hospital or non-hospital) in SUPPORT patients and compared these with those of a sample of Medicare beneficiaries who died in 1992-1993. The percent of SUPPORT patients dying inhospital varied from 29 to 69% across the five SUPPORT sites, and the percent of Medicare beneficiaries dying inhospital varied from 23% to 54% across US Hospital Referral Regions (HRRs). The likelihood of in-hospital death was increased for residents of regions with greater hospital bed availability and use, and the likelihood of in-hospital death was decreased in regions with greater nursing home and hospice availability and use. This effect explained 80% of the variation across regions. Within the HRRs, the number of acute care hospital beds per 1000 population ranged from 1.86 in Mesa, Arizona to 5.30 in Monroe, Louisiana (mean, 3.3).

46. Puopolo AL, Kennard MJ, Mallatratt L et al. Preferences for cardiopulmonary resuscitation. Image. J Nurs Scholarship 1997:29:229-235.

This study reviews interview responses from nurses (n = 1427) regarding discussions about CPR with their patients. Nurses reported having discussions about CPR with 13% of their patients. Discussions were more likely if the nurse thought that the patient did not want CPR, if the nurse had spent more time with the patient, or if the patient was in an intensive care unit at time of study entry. Nurses understood preferences accurately for 74% of patients.

47. Regueiro CR, Hamel MB, Davis RB et al.. A comparison of generalist and pulmonologist care for patients hospitalized with severe chronic obstructive pulmonary disease: resource intensity, hospital costs and survival. Am J Med 1998; 105:366-372.

The study described in this article compared generalists and pulmonologists in their care of the seriously ill COPD patient. Investigators reviewed the charts of 866 patients with severe COPD. Resource intensity was measured using a modified version of the Therapeutic Intervention Scoring System and estimated hospital costs. Of the 866 patients studied, 512 were treated by generalist physicians, and 354 were treated by pulmonologists. Investigators did not find any differences in resource intensity and hospital costs in-patients treated by pulmonologists compared with generalists. Patients treated by pulmonologists did not have better survival.

48. Roach MJ, Connors AF, Dawson NV et al. for the SUPPORT Investigators. Depressed mood and survival in seriously ill hospitalized adults. Arch Intern Med 1998;158:397-404.

Phase I and II patients (n = 3,529) were interviewed using a shortened version of the Profile of Mood States depression scale. Researchers used a stratified Cox proportional hazards model to study association of depressed mood with worse levels of physical functioning. Depressed mood was associated with reduced survival time even after adjusting for patients' demographics and health status.

49. Rosenfeld KE, Wenger NS, Phillips RS et aLfor the SUPPORT Investigators. Factors associated with change in resuscitation preference of seriously ill patients. Arch Intern Med 1996;156:1558-1564.

Investigators interviewed patients (n = 1590) about CPR preference on study Day 3 (baseline) and 2 months later. Overall, 80% of patients had stable preferences over 2 months. For patients initially preferring DNR, patients with substantial improvements in depression scores were more likely to change to a preference for CPR at follow-up.

50. Schroeder SA. The legacy of SUPPORT. Ann Intern Med 1999,131:780-782.

The author discusses his thoughts on the impact of SUPPORT. He asserts that despite the failure of the study to show improvement of end-of-life care in the seriously ill, SUPPORT has proven itself to be a worthy study.

51. The SUPPORT Principal Investigators. A controlled trial to improve care for seriously ill hospitalized patients: The study to understand prognoses and preferences for outcomes and risks of treatments (SUPPORT). JAMA 1995;274:1591-1598,

This article reviews the primary goals of SUPPORT and describes the findings of the Phase I observational trial and the Phase II clinical trial. Phase I analysis of 4301 patients revealed that communication about CPR preference was uncommon, moderate to severe pain affected half of dying patients able to communicate, and final hospitalizations for half of the patients included more than 8 days in the ICU.

Also, nearly half of DNR orders were written in the last 2 days of life. In Phase II, trained nurses were provided to intervention patients to provide prognostic information to patients and physicians, and to elicit and document patient and family preferences and understanding of disease prognosis and to improve pain management. Analysis revealed that intervention patients did not show an improvement in any of the targeted outcomes.

52. Teno J, Lynn J, Phillips RS et al. for the SUPPORT Investigators. Do formal advance directives affect resuscitation decisions and the use of resources for seriously ill patients? J Clin Ethics 1994;5:23-30.

This analysis of advance directives reveals that 20.2% of patients (618 out of 3058) for whom information was available had written advance directives, 11% had living wills, 13.4% had Durable Powers of Attorney, and 4.9% had both. The investigators noted that there was no significant association between the existence of advance directives and decisions about resuscitation.

53. TenoJM, Murphy D, Lynn J et al. for the SUPPORT Investigators. Prognosis-based futility guidelines: Does anyone win? J Am Geriatr Soc 1994;42:1202-1207. This article estimates the savings in hospital costs for SUPPORT patients that would be realized if cases deemed futile had forgone or withdrawn life-sustaining treatment.

Approximately $1.2 million would have been saved if, on study Day 3, life-sustaining treatment had been stopped or not initiated for the 115 subjects with estimated 2-month survival probability of less than or equal to 1%. However, only 12 patients accounted for 75% of the potential savings, and they were often young, acutely ill or posttransplant patients with good prior functional status.

54. Teno JM, Hakim RB, Knaus WA et al. for the SUPPORT Investigators. Preferences for cardiopulmonary resuscitation: Physician-patient agreement and hospital resources. J Gen Intern Med 1995;10:l 79-186. Patients or surrogates (n = 2636) and matched physicians were interviewed about patients' CPR preferences. Answers were correlated with hospital resource use. Adjusted hospital resource consumption was lowest when the patient and physician agreed on a do-not-resuscitate preference and highest when patients did not have a preference and the physician believed patients wanted resuscitation in the case of cardiopulmonary arrest.

55. Teno J, Lynn J, Wenger N for the SUPPORT Investigators. Advance directives for seriously ill hospitalized patients' Effectiveness with the patient self-determination act and the SUPPORT intervention. J Am Geriatr Soc 1997;45:500-507.

Researchers reviewed interviews conducted with patients, surrogates, and physicians about awareness, completion, and impact of Advance Directives (ADs). The authors compared responses before (PRE) and after (POST) the Patient Self-Determination Act (PSDA) and after the SUPPORT intervention (POST+SUPPORT). Before the PSDA (PRE), 62% were familiar with a living will, and 21% had an Advance Directive. These rates were similar for the POST and POST-SUPPORT cohorts, but advance directive documentation increased by 78% in these groups. The POST patients with and without Advance Directives had no significant differences in the rates of medical record documentation of discussion about resuscitation, DNR orders among those who wanted to forgo resuscitation, and attempted resuscitations at death. The POST+SUPPORT patients had similar results.

56. Teno J, Lynn J, Connors AF, Jr. et al. for the SUPPORT Investigators. The illusion of end-of-life resource savings with advance directives. J Am Geriatr Soc 1997;45:513-518.

Three analyses were performed to help answer the question of whether increasing advance directives (ADs) would lead to a reduction in resource utilization. Investigators reviewed charts of patients both before enactment of the Patient Self Determination Act (PSDA) and after. In the latter group, 2652 patients received an intervention: a nurse was provided to facilitate communication among patients, surrogates and physicians about preferences for and outcomes of treatments. Control patients (n = 2152) received no intervention. Researchers found increased chart documentation of existing advance directives with both PSDA and SUPPORT. Intervention patients were more likely to have preexisting ADs documented, but there was no corresponding change in hospital resource use for those who died during hospitalization. Also, ADs documented without further intervention by study Day 3 were associated with decreased hospital resource use, but the same was not true for cases where intervention was used to increase AD documentation. The authors concluded that ADs are not likely to lead to a reduction in resource use.

57. Teno JM, Licks S, Lynn J et al. for the SUPPORT investigators. Do advance directives provide instructions that direct care? J Am Geriatr Soc 1997;45:508-512.

For the 4804 patients in Phase II, authors found 688 Advance Directives (ADs) written by 569 patients. Almost all of the documents were witnessed (88%), and 38% were notarized. Of these directives, 453 were durable powers of attorney for health care, 214 were living wills, and 21 were other directives. Only 90 (13%) directives provided additional instructions for medical care beyond naming a proxy or staring preferences of a standard living will, only 36 (5%) gave specific instructions to guide the use of life-sustaining treatment, and only 22 (3%) might have been applicable to the patient's current situation.

58. Teno JM, Stevens M, Spernak S, Lynn J. Role of written advance directives in decision making: Insights from qualitative and quantitative data. J Gen Intern Med 1998:13:439-446.

Investigators reviewed quantitative and qualitative data in 14 patients to better understand the role of Advance Directives (ADs) in decision making. The 14 patients studied were selected based on the report of having a written Advance Directive and meeting at least one of the following criteria: comatose as defined by a Glasgow coma score of 9 or less and 2-month prognosis for survival of 40% or less; or death during enrollment hospitalization. Investigators extensively reviewed qualitative data: narratives written by intervention nurse written shortly after the patient's death or discharge from the hospital. These data suggested that advance directives played an important role in decision-making in five of the 14 cases.

59. Teno JM. Lessons learned and not learned from the SUPPORT project (editorial). Palliat Med 1999;13:91-93. In this editorial, the author (a SUPPORT investigator) notes the failure of SUPPORT to enhance communication or physician understanding of patient preferences. The author discusses important lessons learned through SUPPORT which include: (1) seriously ill patients are often able to communicate preferences; (2) there are opportunities for improvement in care of the seriously ill, especially for communication regarding preferences and pain control; (3) there may be a chance to affect change in the care of the terminally ill by working at the local level; and (4) we cannot assume that all seriously ill patients want less care. More research needs to be done regarding this fundamental assumption. 60. Tsevat J, Cook F, Green ML et al. for the SUPPORT Investigators. Health values of the seriously ill. Ann Intern Med 1995;122:514-520.

Investigators measured time-tradeoff (TTO) utilities and health ratings in 1438 patients. TTO utilities indicate whether seriously ill patients would prefer a shorter, healthier life to their current state of health. Patients' answers varied widely. 34.8% of patients were unwilling to give up any time in their current state of health for a shorter life in excellent health, and 9.0% were willing to live 2 weeks or less in excellent health rather than I year in their current state of health. Health rating scales averaged 57.8 24.0 on a scale of 0 (death) to 100 (perfect health), and patient's health values exceeded that of their paired surrogate.

61. Tsevat J, Dawson NV, Wu AW, Lynn J, Soukup JR, Cook EF, VidaiUet H, Phillips RS, for the HELP Investigators. Health values of hospitalized patients 80 years or older. JAMA 1998;279:371-375.

Of 1,266 HELP patients, 414 successfully completed interview questions pertaining to health value. These patients were asked whether they would prefer living only I year in their current state of health or less time in excellent health. The mean time-trade-off score was .81 which indicated that, on average, patients equated living I year in their current state of health with living 9.7 months in excellent health. However, time-trade-off scores varied widely. In 300 patient surrogate pairs, the mean patient time-trade off scale was .80 while the mean surrogate value was .05 less. 61 (20.3%) of surrogates underestimated the patient's time-trade off score by .25 (3 months of 12) or more.

62. Weeks JC, Cook EF, O'Day SJ, Peterson LM, Wenger N, Reding D, Han-ell FE, Kussin P, Dawson NV, Connors AF, Jr., Lynn J, Phillips RS. Relationship between cancer patients' predictions of prognosis and their treatment preferences. JAMA 1998;279:1709-1714.

Phase I and II patients with stage III or IV non-small cell lung cancer or colon cancer metastatic to the liver (n = 977) were interviewed for estimates of their own prognoses and for their treatment preferences. Patients who thought they had a high probability of surviving for at least 6 months were more likely to favor life-extending therapy over comfort care, compared to patients who thought there was at least a 10% chance that they would not live 6 months. Patients overestimated their chances of 6-month survival while physicians' predictions were more accurate. Patients' overly optimistic views of their prognosis led to adverse outcomes.

63. Wenger NS, Oye RK, Bellamy PE, Lynn J, Phillips RS, Desbiens NA, Kussin P, Youngner SJ. Prior capacity of patients lacking decision making ability early in hospitalization: Implications for advance directive administration. J Gen Intern Med 1994;9:539-543.

SUPPORT investigators reviewed the ability of seriously ill patients to make decisions regarding advance directives at hospitalization. Forty percent of patients were not considered interviewable due to reasons such as coma, intubation, or failing a cognitive screen. However, 83% of those that could not participate in an interview could have participated in treatment decisions 2 weeks before hospitalization.

64. The SUPPORT Investigators: Wenger NS, Oye RK, Desbiens NA, Phillips RS, Teno JM, Connors AF, Liu H, Zemsky MF, Kussin P. The stability of DNR orders on hospital readmission. J Clin Ethics 1996;7:48-54. Researchers reviewed patients' charts for DNR orders.

Of 3,710 patients with DNR orders, 543 were subsequently readmitted to the same hospital during the six-month study period. In 157 cases (29%), the DNR order differed from first admission to the next admission. Multivariable analysis revealed that patients with unstable DNR orders were younger and more likely to be non-white. Patients from one hospital, white patients, and patients with metastatic cancer and COPD had greater stability of DNR orders.

65. Wenger NS, Greengold NL, Oye RK, Kussin P, Phillips RS, Desbiens NA, Lin H, Hiatt JR, Teno JM, and Connors AF, Jr. for the SUPPORT Investigators. Patients with DNR orders in the operating room: Surgery, resuscitation, and outcomes. J Clin Ethics 1997:8:250-257.

Of 4,301 patients in Phase 1,1,251 were considered for surgery. Of these, 119 (10%) had DNR orders recorded on the chart. Of the patients considered for surgery who had DNR orders, 48% (57) underwent surgical procedures. In contrast, 61% of patients without a DNR order who were considered for surgery had surgery.

66. Wilson ffi. Green ML, Goidman L, TsevatJ, Cook EF, Phillips RS, for the SUPPORT Investigators. Is experience a good teacher? How interns and attending physicians understand patients' choices for end-of-life care. Med Deds Making 1997:17:217-227.

Medical interns and attending physicians at one SUPPORT hospital were interviewed about their Phase I patients' preferences for CPR and their thoughts about patient willingness to live with a series of undesirable outcomes. Answers were compared to responses from patients or surrogates. Compared with interns, attending physicians had known the patients longer, had talked with patients more frequently about prognosis, and felt they knew more about their patients' preferences. However, attending physicians were no more accurate than interns in predicting patients' preferences.

67. Wn AW, Damiano AM, Lynn J, Aizola C, Teno J, Landefdd CS, Desbiens N, Tsevat J, Mayer-Oakes A, Harrell HE, Knaus WA. Predicting future functional status for seriously ill hospitalized adults: The SUPPORT model. Ann Intern Med 1995;122:342-350.

At 2 months after admission, one third of patients available for assessment (n = 1746) had severe functional limitations defined as Sickness Impact Profile scores >= to 30 or activities of daily living scores >= 4. Researchers also noted that patients had frequent functional change: 21% who had no baseline dependencies were limited at 2 months, and 30% of those with 4 or more baseline limitations had improved at 2 months. Using clinical and questionnaire data, it is possible to predict future functional status with acceptable accuracy. 68. Zhong Z., Lynn J. The Lamont/Christakis Article Reviewed. Oncology 1999; 13: 1172-1173.

Authors discuss the suggestion by Lamont and Christakis that physicians' survival estimates in advanced cancer patients are relatively inaccurate and biased in the direction of excessive optimism in light of SUPPORT findings. Authors analyzed SUPPORT data of patients with advanced cancer (colon n = 406; lung n = 764; and multiple-organ system failure with malignancy n = 594). They compared physician estimates of patient prognosis (surviving 2 and 6 months) with prognosis estimated by the SUPPORT prognostic model. They found that the SUPPORT model, mean physician estimates, and actual survival are virtually identical at 2 and 6 months. Also, they noted some variation in estimates depending on time and disease type but noted that physicians and models always erred in the same direction.


National Perspective on Dying in America: Does Place of Death Matter?

John M. Benson, MA,* Joel C. Cantor, ScD,** Joanne Lynn, MD, MA, MS,*** and Joan Teno, MD, MS **** First presented at the New York Symposium on Health Services Research, December 10, 1996.

SUPPORT (Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatment) focused on specific diagnostic categories and enrolled patients in five major medical centers, thus potentially limiting its generalizability. We present here results of a national telephone survey of participants (defined below) in end-of-life decisions, aimed at assessing the degree to which the findings of SUPPORT may be generalized beyond major teaching hospitals.


Between September 8 and October 8, 1995, telephone interviews were conducted with 502 participants in end-of-life decisions. Respondents were screened from a national probability sample of adults, with a screening rate of 9% and a response rate of 50%. Respondents gave affirmative answers to two questions: "In the past two years, have you had a close family member or other loved one who died after a serious illness?" and, if yes: "Did you have contact with the medical staff caring for this person, or not?"

Respondents were asked about their relationship to the decedent, the timing of the person's death, the degree to which the death was expected, the cause and place of death, the quality of communication with medical providers, the use of life support, the extent of patient consciousness and pain, patient preferences for treatment focused on comfort versus extending life and the degree to which those wishes were honored, advance directives, family economic impact, and sociodemographic characters of the respondent and the decedent.

Data were analyzed using univariate and multivariant methods. Characteristics of survey respondents and decedents and respondent-reported outcomes of care were compared by place of death (hospital, home, or other place). Findings among patients who died in a hospital were further stratified by whether the death took place in a "major medical center affiliated with a university" or in another type of hospital. Statistical significance was examined using chi-square tests of homogeneity and Kruskal-Wallis tests of differences among medians. Univariate associations of outcomes of care with place of death and type of hospital were confirmed using multiple regression analyses, controlling for potential confounders.


Characteristics of the Sample Median age of the 502 respondents was 46 years; 57% were female; 83% were white and not Hispanic; 30% had a high school level education, and another 58% had at least some college; annual family income split almost evenly between those earning less than $25,000, between $25,000 and $50,000, and more than $50,000; 35% were the deceased person's child, 13% were the spouse, and 48% were another relative. Decedents had a median age of 72 years; 42% died of cancer and 23% of heart disease; 85% wanted relief of pain and discomfort (rather than extending life as much as possible); and the death was reported to be completely unexpected 22% of the time and expected for at least the last two months in 40% of the cases.

Place of Death

More than half (58%) of the respondents died in a hospital. Among those died in hospital, about one in four (27%) were reported to have died in a "major medical center affiliated with a university." Home deaths accounted for 26%, and all other for 16%.

Place of death (hospital, home, other) was not associated with respondent characteristics but was associated with several decedent characteristics. Those who died at home or in other settings (including hospice and nursing home) settings were more likely to have died of cancer (55% of those at home, 44% for other, and 36% of hospital deaths, P < .001) and to have had their death expected for a longer period (57% expected for at least 2 months for home deaths, 45% for other, and 31% for hospital, P < .001). Those who died in other settings also had a significantly higher median age of death (78 years vs 71 at home and 70 in hospital, P < .001) and a slightly more common preference for palliation over efforts to extend life (93% for other, 79% for home, and 85% for hospital, P = .033)

Among hospital decedents, age and preferences varied significantly by major medical center versus other. The median age of decedents in major medical centers was 5 years less (67 vs 72 years), and these patients were more than twice as likely to prefer extension of life over pain relief compared with those who died in other hospitals.(24% vs 10%, P = .025) Decedents in major medical centers were also more likely to have been on mechanical life support during the last 3 days of life (38% vs 23%, P = .022). Survey respondents whose loved ones died in a major medical center were somewhat less likely to be non-Hispanic-whites (70% vs 86%, P = .009).

Outcomes of Care

Experiences of patients and families varied little by place of death or by type of hospital for seven outcomes examined.

Only having an advance directive (AD) was significantly associated with place of death or type of hospital in univariate statistics. Compared with hospital decedents, those who died in other settings (including nursing home and hospice) were about 2.4 times as likely to have an AD; and compared with decedents in major medical centers, those in other hospitals were about twice as likely to have an AD. However, both of these associations became statistically insignificant in multivariate analysis controlling for potential confounders (decedent age, expectedness of death, respondent education, and whether the respondent was one of the main people consulting with the medical staff).

The other six measures of experiences near the end of life were not associated with place of death or type of hospital in univariate analysis. These experiences included statements by respondents that: (1) patient/family preferences were followed (overall rate = 91% of those with valid answers); (2) the medical staff explained the patient's condition adequately (overall, 80%); (3) the medical staff explained treatment options adequately (overall, 80%); (4) the patient/family had adequate influence over treatment decisions (overall, 76%); (5) the patient (if conscious) was in severe pain most of the time during the last 3 days of life (overall, 46%); and (6) the medical staff did as much as the patient/family wanted to control pain (overall, 86%). The proportion of patients who were on mechanical life support was 18% overall.

Two of these measures became significantly associated with type of hospital in multivariate analysis: opportunity to influence treatment (AOR 0.23; 95% Cl, 0.07-0.74) and presence of severe pain (AOR= 0.36; 95% Cl, 0.14-0.93). Controlling for potential confounders (enumerated above), these outcomes were worse in other hospitals than in major medical centers.

Additional Resources

The Dartmouth Atlas Report: End-of Life Care Future Challenges
A CAPC Fall Forum 2000 Workshop
POWERPOINT Presentation Link

Other Resources By

Joanne Lynn, MD, MA, MS

back to top