Can Clinical Interventions Change Care at the End of Life?

Methods

We conducted a systematic review of published interventions designed to change clinical practices and improve care at the end of life. We searched the MEDLINE database for English-language articles published from January 1990 to March 1996 using advance directive, terminal care, controlled trial, intervention, life-sustaining treatment, and prospective as key words or text words. To identify additional studies, we searched abstracts presented at recent national meetings and the reference lists of the selected articles.

We included original research articles if data were presented on the effects of a clinical trial designed to change the experience of patients or the practices of physicians who provide care near the end of life. Because human behavior may be influenced by education, incentives, or organizational influences, we did not restrict our review to one type of intervention. We accepted studies that used historical controls or did not define a control group, because relatively few randomized, controlled trials have been done. We excluded any study that did not have the potential to affect clinical care, such as those limited to changes in attitudes or knowledge. We also excluded studies of patients or physicians outside the United States because of important variations in practice norms.

Because of the heterogeneity of interventions, we grouped studies by descriptors of the interventions rather than by study design. Studies were first grouped by the target population: patients, physicians, or both. We then examined whether an intervention led to changes in four desirable clinical outcomes: use of patient preferences, pain control, use of life-sustaining treatments, or medical costs. Finally, we compared positive and negative trial results for differences in intervention, target group, or the strength of research design and methods. Funding sources for this review did not influence our analysis or interpretation of the data.

Data Synthesis

Sixteen studies met the inclusion criteria. Eight interventions targeted patients, five targeted only physicians, and three attempted to influence behavior in both groups. All studies attempted to increase use of patient preferences; some also examined changes in pain control, use of life-sustaining treatments, and cost.

Interventions for Patients

Eight studies used patient-education strategies to increase the recording of advance directives or proxy choices [6-13] (Table 1). Six of the eight trials were successful [6, 8-11, 13], and one achieved a particularly high rate of advance directive completion [10]. Five studies were randomized, controlled trials [7, 9, 11-13]; three used historical [10] or nonrandom [6, 8] control groups. Patients were enrolled in outpatient settings and were generally healthy. Five trials selected only elderly patients for study [7-10, 13]. Some sampling strategies also restricted the intervention to patients at higher-than-average risk for acute illness [13]. Baseline rates of documented preferences or proxies ranged from 0% to 31%. The rates of recorded proxies or advance directives after intervention ranged from 15% to 71%. The two studies in this group with negative results did not differ from the others in sampling strategies, but both were randomized, controlled trials.

All interventions provided patients with education about life-sustaining treatments and their right to express preferences about use of these treatments. Most successful trials did not rely on written materials alone. In the two studies with negative results, written materials were supplemented only by interviews with researchers [7, 12]. The only successful study to use educational materials alone was a randomized, controlled trial in a large, well-established health maintenance organization [9]. In response, 19% of patients in the intervention group completed proxy forms. This success rate may have resulted, in part, from the investigators' choice of outcome measure. Unlike in other studies, advance directives were counted if they were mailed back to the research team; the directives did not have to be documented in the patients' medical records. These patients may also have been more inclined to complete written advance directives than those who were uninsured or otherwise disenfranchised [14].

All other successful studies in this group used patient-education materials and discussions with a physician or other clinician. In two of the earliest trials [6, 8], researchers reported success rates of 15% to 46% with the use of clinical discussions after they had found no change in behavior with the use of written materials alone. Richter and colleagues [11] designed a randomized, controlled trial that combined an initial discussion with a physician and mailed educational materials, resulting in increased completion of advance directives (3% before the intervention and 23% after the intervention). The highest rates of use of advance directives (71%) and proxy appointment (68%) were achieved when a social worker held repeated discussions with patients during follow-up clinic visits; this gave patients the time to ask questions and to reach thoughtful decisions [10]. Meier and colleagues [13] found similarly high rates of proxy appointment (48%) in a randomized trial of patient-advocate intervention in which repeated discussions were permitted. In summary, most educational interventions for patients increased documentation of patient treatment preferences. More patients recorded preferences when they had repeated opportunities to discuss their decisions with a physician or a social worker.

Interventions for Physicians

A second group of educational interventions was designed for physicians, who must elicit and respect patient preferences if they are to influence the treatments the physicians order [15-19] (Table 2). Investigators used sophisticated educational or motivational techniques in these trials, all of which achieved some positive results. However, these studies also required less rigorous standards of evidence. No trial in this group used randomly assigned controls, and some used patterns of practice for physician groups as a proxy for behavioral changes in individual physicians. All studies were conducted in a single institution or practice, where concerns about statistical power and cross-contamination make randomization less feasible.

Two interventions were designed to increase use of patient preferences by increasing the number of written advance directives in outpatient records. Markson and colleagues [15] studied the impact of a 5-week didactic course for physicians. After the course, physicians still did not initiate discussions of advance directives; therefore, the trial was modified. In the final intervention, only patients receiving continued care were enrolled, extra clinic time was allowed for discussions, and expert modeling was available during clinic sessions. New advance directives were then documented for 22% of eligible patients or 74% of those asked. The researchers concluded that didactic education alone was insufficient to change physician behavior. In a similar study, Meier and colleagues [16] reported that 32% of geriatric outpatients documented a proxy decision maker after their physicians took part in a multistep intervention. Physicians heard one lecture and then received detailed reminders and proxy forms on the charts of selected patients during each clinic visit. As in a similar patient-based intervention [10], a high success rate was achieved when patients and physicians had repeated opportunities for discussion before completing written directives.

Three other physician-based interventions were designed to increase use of patient preferences and change use of life-sustaining treatment. All targeted physicians cared for hospitalized patients. Holloran and colleagues [17] began a housestaff ethics seminar in a surgical intensive care unit that involved a case-based curriculum and adult education techniques. The researchers examined trends during a 3-year period and found that the length of stay in the surgical unit decreased by 46%, compared with a concurrent 13% decrease in a medical unit. Treatment decisions were discussed earlier with patients who were hospitalized for more than 30 days and then died. Interpretation of these findings was limited by lack of adequate controls and failure to measure patient- or physician-specific outcomes.

Sulmasy and colleagues [18] reported modest success with a housestaff program composed of intensive ethics education and directed feedback [18]. This study considered only the quality of discussions and documentation of do-not-resuscitate (DNR) orders and did not examine whether the intervention affected their frequency. Qualitative measures of DNR documentation did not change, partly because existing housestaff practices already met the quality guidelines of this study. However, housestaff did take personal responsibility for more DNR orders than did historical controls. Discussions of “concurrent care concerns” or decisions about treatments other than cardiopulmonary resuscitation increased from 68% to 86% of patients who had a discussion about DNR orders.

Reilly and colleagues [19] studied the most comprehensive educational intervention for physicians. Didactic sessions were supplemented by chart reminders, group feedback, and the use of a new advance directive form in all inpatient charts. This multistep intervention took 3 years to complete. The use of DNR orders did not change (22% compared with 19%), but the number of recorded patient treatment preferences increased from 25% to 63%. The researchers also measured a change in “impact directives” from 3% to 13%; these directives were defined as preferences that would change treatment orders in the event of life-threatening complications. This intervention was successful in broadening the range of treatments discussed, and more patients recalled having discussions after the intervention (51% compared with 33%; P < 0.001).

Interventions for Physicians and Patients

Only three studies used interventions that had components for patients and physicians [5, 20, 21] (Table 3). Two were randomized, controlled trials [5, 21]; all studies used carefully described outcomes in defined patient populations. All three interventions increased expression of patient preferences but showed no effect on the use of life-sustaining treatments or other outcome measures.

All patients in these studies were chosen to be at increased risk for death because of life-threatening acute illness [5], severe life-limiting chronic illness [21], or advanced age and functional impairment [20]. Patients were asked to express choices of treatment that were then communicated to their physicians. Compared with studies of patient education methods, all of these studies were successful at the initial step of documenting patient preferences. In the SUPPORT study [5] and the intervention designed by Danis and colleagues [20], the investigators assumed that patients or surrogates who agreed to participate would be willing and able to express treatment preferences. These studies had participation rates of 85% and 83%, respectively, suggesting that severely ill patients were eager to discuss their treatment options. In the third study, by Schneiderman and colleagues [21], physicians selected clinic patients who were believed to have a limited life expectancy and offered them information on advance directives. Combined proxy appointment and advance-directive instructions were recorded by 66% of patients. In all three studies [5, 20, 21], preferences were communicated to treating physicians in the hospital by using chart documentation. In the SUPPORT study, discussions among physician, patient, and family were also facilitated by a research nurse and detailed prognostic information was provided to all decision makers.

Despite the success at eliciting and recording patient preferences, none of these intervention studies showed a difference in use of life-sustaining treatment or in other desirable outcomes. Danis and colleagues [20] found that treatments were consistent with patient preferences 75% of the time, but the presence of an advance directive in the medical record did not necessarily result in a treatment that was more consistent with preferences. When it was inconsistent with a patient's preference, treatment was usually less aggressive than that requested. Schneiderman and colleagues [21] found that offering advance directives to patients made no difference in the use of medical treatments, survival, health status, satisfaction, or the cost of care in the final month of life. Although the SUPPORT intervention was both intensive and flexible, it had no effect on clinical decisions to use or to withhold life-sustaining treatments, pain control, or cost of terminal care [5]. Despite the clear success in recording patient treatment preferences near the end of life, these interventions could not alter other clinically desirable outcomes.

Discussion

Several interventions have successfully increased the use of patient treatment preferences in end-of-life care. With appropriate education and reminders, physicians and patients will increase their discussion of treatment choices. In nearly all of the studies reviewed, many patients became willing to commit their proxy choices or treatment preferences to paper once they understood the need for advance planning. The success rates of interventions varied with patient characteristics and the educational techniques used. Healthier patients, similar to the general population [22], are willing to complete advance directives only about 15% of the time. For outpatients, written materials combined with repeated clinical discussions led to higher rates of use. When only severely ill patients are approached, most are willing to record advance directives. Discussions can be initiated by a physician but are also effective when conducted by a social worker or patient advocate. If physicians are expected to initiate such discussions, they may require extra clinic time, chart reminders, convenient forms, or peer modeling and feedback.

Increased expression of patient treatment preferences is one of the common goals of improving care at the end of life. This first step is considered necessary by contemporary ethical standards but may not be sufficient to precipitate changes in use of treatments, experience of pain, and cost of care. Three studies of more intensive interventions with physicians offered some suggestive evidence that use of treatment could be modified. None, however, used randomly assigned controls, and outcome measures were few and methodologically limited [17-19]. Three interventions that allowed patients to express preferences and then communicated these preferences to the treating physicians had no effect on the use of treatments [5, 20, 21], pain control, satisfaction with care, or cost [5, 21].

Careful review of these intervention studies is difficult. The behavior under study is complex, and study designs varied greatly. Authors had many similar goals but targeted different groups and used various intervention strategies and outcome measures. Such variation limits the comparison of studies; however, we attempted to decrease this variation by classifying interventions by target group. Educational and motivational interventions are more difficult to describe or replicate than other clinical trials, particularly if they rely on a charismatic leader or committed teacher. These studies can be compared qualitatively, but only as the study of human behavior advances will development of more rigorous evidence standards be possible.

Many intervening steps occur between the time that a patient expresses a desire for less aggressive treatment and the actual use of treatments that determine comfort and cost of care. One simple explanation is that patient treatment choices are documented but are poorly communicated to physicians. In a complex health care system, even a simple transfer of advance directives is difficult [22, 23]. All three physician–patient studies found this to be true. For example, Danis and colleagues [20] reported that written advance directives were transferred from nursing home to hospital only 35% of the time. Although nurses in the SUPPORT study said that they communicated with treating physicians in nearly all cases, the physicians recalled receiving information on patient preferences only 34% of the time [5]. These findings from inpatient studies suggest that advance directives completed in outpatient clinics would have no impact on hospital-based medical care. New mechanisms, such as electronic transfer, may help to ensure continuity of records between sites.

A second explanation may be that patients and their proxies change their minds and pressure physicians to prolong life at all cost. Although this scenario may be critically important in individual cases, it does not seem to be common in large observational studies. Two prospective cohort studies of seriously ill patients found no relation among advance directives, patient preferences, and the type and cost of treatment received after controlling for severity of illness [24, 25]. A 2-year study of the attitudes of Medicare recipients showed that although a small subset of patients desired more aggressive treatment as the illness became more severe, 80% of older adults had stable preferences or desired less aggressive treatment over time [26]. Some studies of family proxies have found that families want more aggressive treatment than do patients [27], but this finding is inconsistent [28]. In fact, when patients are unable to communicate preferences, families and physicians become far more likely to limit life-sustaining treatment [18, 29]. In light of this evidence, it seems doubtful that the demands of patients and their families drive decisions for aggressive treatment against prior advance directives.

A third reason that interventions fail to affect use of treatment or cost is that they do not promote meaningful communication between patients and physicians. Some commentators have argued that advance directives are not specific enough to direct actual clinical decisions [30]. Schneiderman and colleagues [31] found that the general statement in a standard advance directive form could not predict the patients' specific choices of treatment. If physicians also find these written statements vague, it may be difficult to use them to change their standard approach to care. Physicians believe that they know what patients want, yet their understanding of patient preferences remains poor even after face-to-face discussions [28, 32-34]. Physicians tend to be coercive and to focus on treatment descriptions rather than listening to patients' concerns [34-38]. When physicians discuss withholding a specific life-sustaining treatment, they rarely document broader goals of care. Therefore, patients with written DNR orders continue to receive many other types of costly medical treatment [39, 40]. Physicians may need better communication skills to listen to patient preferences and respond to their patients' needs near the end of life. Advance planning discussions might be more effective if they focused on goals of care rather than on specific treatment choices. The range of options might also be broadened to include consideration of home- or hospice-based care.

Finally, these interventions may fail because many determinants of physicians' behavior are more influential than patient preferences. Decisions may be driven by beliefs about treatment efficacy, perceived standards of care, or prognostic uncertainty [41-43]. In practice, physicians have different norms of care for patients dying of different diseases. For example, patients who have cardiac failure receive more life-sustaining treatment than do patients with the same life expectancy who have cancer [44-46]. Prognostic uncertainty drives physicians to continue aggressive treatment and is a powerful predictor of the cost of hospital care [47]. Depending on how a physician defines terminal illness, he or she might feel legally and morally bound to treat until death is proximate and certain. Predicting death is an uncertain art, and computerized prognostic systems are not much better than physicians' estimates [48]. Decisions to limit aggressive care are often made after a trial of aggressive treatment, when the physician has better evidence that an illness is irreversible and terminal [5]. When decisions are made late in the course of illness, treatments have been given at substantial discomfort and cost. Many powerful determinants of physicians' practices supersede consideration of patient preferences. An increasing body of literature supports the need for sophisticated educational and motivational strategies to change physician practices [49-51].

Although clinical interventions using patient preferences have not changed medical care at the end of life, some policy and organization changes have affected treatment use, pain, and cost. Observational studies have shown changes in treatment practices after implementation of new DNR policies [52, 53] and after new institutional programs designed to comply with the Patient Self-Determination Act [54-56]. In these studies, it is unclear whether treatments changed because of individual patient preferences. In fact, some measures of physician–patient communication did not improve [56], especially for socially disenfranchised patients [55]. Insurance strategies might also change the treatment given to dying patients. After the U.S. government began using a prospective payment system for Medicare beneficiaries in 1983, researchers noted a decrease in the use of intensive care for Medicare and non-Medicare patients [57] and a change in the location of death from hospitals to nursing homes or a patient's own home [58, 59]. One observational study found an association between managed care insurance and reduced use of intensive care and ventilator support [60]. Again, it is unclear whether these changes were consistent with patient preferences and whether they improved patient comfort or satisfaction. A few innovative programs of care for chronically and critically ill patients [61] and patients who have advanced Alzheimer disease [62] used hospice techniques to meet the needs of family and patient while using fewer acute care resources. Although hospice care is deeply satisfying to patients and families [63, 64], a randomized, controlled trial of hospice care [65] showed no clear improvement in patient outcomes or decrease in total cost of care.

Medical treatment decisions are the result of complex human behaviors. Clinical interventions can increase the use of advance directives but have rarely examined other important clinical outcomes. When clinical outcomes are studied, patient preferences have little or no influence on the use of life-sustaining treatments, on patient comfort and satisfaction, or on the cost of care. Future interventions will provide new insights only if they have an impact beyond the initial step of advance directive discussions. Educational approaches must be creative and complemented by motivational and organizational strategies to change physician behavior. To build on previous research, interventions should prove that they affect patient-based measures of good clinical care, such as pain control or satisfaction, or result in substantial changes in use or cost of treatment. Renewed and creative interventions will show a commitment to the quality of terminal care that is equal to our commitment to extend life.

Dr. Tulsky: Veterans Affairs Medical Center (152) Division of General Medicine, 508 Fulton Street, Durham, NC 27705.