Moreover, because patients may not always be willing or able to make decisions about their care, it is important to know whether legally authorized proxy decision makers [1] are reasonable substitutes. Thus, for decision making that involves individual patients, both patients and their caregivers must know not only the salient possible outcomes but the "value" of each outcome to the patient. This is particularly germane for seriously ill patients, for whom weighty decisions arise frequently.

Individual preferences are also important for public policy. If particular treatment practices are judged undesirable by many persons, it would be useful to incorporate that information into the establishment of "default" policies [2, 3] so that, in the absence of clearly stated preferences, certain interventions might or might not be provided on the basis of a previously established consensus. The issue here is whether groups within which preferences are consistent can be clinically defined, or whether predictors of health values can be identified.

Finally, there are calls to direct scarce resources toward the most worthwhile health care endeavors. From a societal perspective, it may be important to compare preference data across conditions to set priorities for diverse treatments. An understanding of the experiences of large groups of persons potentially affected by allocation decisions will be essential to rational priority setting. Clearly, for both clinical decisions and policymaking, we need to measure health-related quality of life and to assess preferences accurately.

The time-tradeoff utility is a quantification of a person's preference for quality rather than quantity of life [4]. As used here, it ascertains the strength of a person's preference for a shorter but healthy life by asking how much (if any) of his or her remaining life expectancy he or she would be willing to exchange for a shorter life in excellent health. Time-tradeoff utilities are particularly useful for the evaluation of complex decisions and for cost-effectiveness analyses [5-7]. Our goal was to examine the preferences of a group of seriously ill persons and their surrogate decision makers. In addition to preferences for individual treatments and outcomes, we focused on ratings and health values [utilities] for the patients' current state of health. Specifically, we set out to 1) assess the health values and health ratings of a group of seriously ill patients; 2) compare patients' health values and health ratings with those of their surrogate decision makers and physicians; 3) assess relations between health values and health status measures, preference measures, measures of perceived prognosis, demographic variables, and the effects of the patient's illness on family members to identify potential predictors of health values and to establish construct validity; and 4) assess whether health values change over time.

Methods

Top Methods Results Discussion Author & Article Info References

The Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments (SUPPORT) was a prospective 5-center study of the prognoses, preferences, and decision making of seriously ill hospitalized adults, their surrogates, and their physicians. In phase I of SUPPORT, we gathered descriptive data; in phase II, we tested an intervention that consisted of giving physicians prognostic information and feedback about patient preferences. Analyses in this article involve patients enrolled in phase I. The study population, characteristics of patients and collaborating hospitals, data collection strategies, and statistical methods of SUPPORT have been described previously [8-22] and are summarized below.

Study Participants

From June 1989 to June 1991, phase I of SUPPORT enrolled consecutive patients who had been diagnosed with one of nine conditions and who had clinical characteristics predictive of an aggregate 6-month mortality rate of approximately 50%. These clinical characteristics included the Acute Physiology score component of the APACHE (Acute Physiology, Age, Chronic Health Evaluation) III prognostic scoring system as a measure of severity of acute illness. Acute Physiology scores can range from 0 to 252; higher scores indicate more severe illness [23]. The nine conditions were acute respiratory failure; acute exacerbation of severe chronic obstructive pulmonary disease; acute exacerbation of severe chronic congestive heart failure; chronic liver failure with cirrhosis; nontraumatic coma; colon cancer metastatic to the liver; metastatic non-small-cell carcinoma of the lung; multiorgan system failure with malignancy; and multiorgan system failure with sepsis. A surrogate was identified by the patient, attending physician, or medical record reviewer as the person who would make decisions about care if the patient was unable to do so (in some cases, no surrogate was identified) [10].

Interviews and Instruments

Patients who passed a cognitive screening test [22] were eligible to be interviewed. Interviews of patients and surrogates were done serially at approximately days 3 (the initial interview "window" spanned 4 days), 8, 14, and 25 and at months 2 and 6 after study entry. Physicians were interviewed at day 3 and, if the patient was still in the hospital, at day 25 [9].

The utility measure used was the time tradeoff [4, 18, 22]. Patients and surrogates were independently asked whether the patient would prefer living 1 year in his or her current state of health or 11 months in excellent health, or whether the two would be considered equivalent. If 11 months of excellent health was chosen, the respondent was asked whether the patient would prefer 1 year in his or her current state of health or 9 months in excellent health. The series continued until the point at which the choices were equivalent was ascertained. The time-tradeoff score was calculated as the fraction of a year of excellent health that was considered equivalent to a year in the current state of health. For example, if the patient equated living 12 months in his or her current state of health with living only 9 months in excellent health, the time-tradeoff utility would be 9/12, or 0.75. Possible scores ranged from 0.04 (if 2 weeks in excellent health was equivalent to 1 year in the current state of health) to 1.0.

Other health-related quality-of-life instruments included an overall health rating scale, in which the respondent was asked to rate the patient's current state of health on a scale anchored by 0 (death) and 100 (perfect health) [18, 22, 24]; a revised measure of dependence in activities of daily living over the previous 2 weeks [13, 22, 25, 26]; a revised version of the Duke Activity Status Index [12, 22, 27], which assesses ability to do strenuous activities; a shortened version of the Profile of Mood States [20, 28, 29], which assesses anxiety and depression; and the Sickness Impact Profile [20, 30, 31], a comprehensive functional status measure. We also asked about preferences for cardiopulmonary resuscitation (CPR); willingness to tolerate any of six lifelong adverse outcomes (pain, mechanical ventilation, tube feeding, coma, confusion, and living in a nursing home); preferences for care that focuses on extending life as much as possible rather than for care that focuses on relieving pain and discomfort as much as possible; perceived prognoses for surviving for 2 and 6 months, for being free of pain and discomfort in 2 and 6 months, and for being able to take care of oneself in 2 and 6 months [15, 22]; and the effect of the patient's illness on family members in terms of assistance needed and savings depleted [20, 32].

Reliability

Test-retest reliability of several measures, including the time-tradeoff utility, was assessed at one of the sites by re-interviewing 23 respondents within 4 to 32 hours of their initial interview.

Statistical Analysis

Means are expressed ±SD and medians are given with 25th and 75th percentiles. Continuous variables were compared using the Wilcoxon rank-sum test. Within-patient changes over time were assessed using the Wilcoxon signed-rank test. Test-retest reliability of the time tradeoff was assessed using mean absolute differences in scores and Spearman correlation coefficients.

We compared health value scores across disease categories using the Kruskal-Wallis test after first combining into one category acute respiratory failure and multiorgan system failure with sepsis. Concordance between patients' and surrogates' time-tradeoff scores and between patients' and their surrogates' and physicians' health ratings was assessed using the Wilcoxon signed-rank test. Associations between the time tradeoff and the other health-related quality-of-life scales, measures of perceived prognosis, willingness to tolerate adverse outcomes, and demographic variables were assessed using Spearman correlation coefficients. Because time-tradeoff scores were not normally distributed, we used ordinal logistic regression to identify significant predictors of time-tradeoff scores at day 3 and month 2. Because patients' responses were missing for certain predictor variables, in separate models we eliminated nonsignificant variables, imputed missing values, or substituted surrogates' responses to increase power; the results of these models were similar to those from the models in which no substitution of responses was done and thus are not reported. The summary statistic for the ordinal logistic regression models is Somers D, a linear function of the area under the receiver operating characteristic (ROC) curve (D = 2 x [area under ROC curve –0.5]). Finally, within-patient changes in time-tradeoff scores were compared with changes in other measures using the Kruskal-Wallis test and Spearman correlation coefficients.

Results

Top Methods Results Discussion Author & Article Info References

Patients and Interviews

In phase I of SUPPORT, 4301 patients were enrolled over 2 years and were followed for 6 months. The nature of the study was such that many patients could not be interviewed at study day 3: Twelve hundred eighty-nine (30.0%) were intubated, comatose, or both; 351 (8.2%) could not be interviewed for reasons such as dementia or aphasia; 214 (5.0%) failed the cognitive screening test; 228 (5.3%) died before they could be interviewed; 266 (6.2%) were discharged before they could be interviewed; 247 (5.7%) either refused or were not permitted by their physicians or surrogates to be interviewed; and 56 (1.3%) were not approached for an interview. The remaining 1650 (38.4%) patients were interviewed at study day 3; 114 of these (6.9%) terminated their interview before the time-tradeoff question was asked. Of the remaining 1536 patients, 98 (6.4%) refused to answer the time-tradeoff question, answered it with "don't know," or had missing or incomplete answers.

The remaining 1438 patients formed our main analytic sample (Table 1). Both in-hospital mortality rates (6.1% compared with 37.1%; P < 0.0005) and 6-month mortality rates (31.8% compared with 56.2%; P < 0.0005) were significantly lower for the 1438 patients who completed the time-tradeoff question than for the others. Similarly, on the basis of the Acute Physiology score of APACHE III, acute illness on day 3 was less severe among patients who answered the time-tradeoff question than among those who did not (mean scores, 25.0 and 43.9, respectively; P < 0.0001). Both the patient and his or her surrogate completed the time-tradeoff question at day 3 in 1041 cases, thus creating 1041 patient-surrogate pairs.

Reliability

Test-retest reliability was generally excellent. The mean difference for time-tradeoff scores was 0.00, and the mean absolute difference was 0.06 (r = 0.95). The mean difference for the 0 to 100 rating scale question was 2.26, and the mean absolute difference was 8.78 (r = 0.85).

Health Values and Ratings

The mean time-tradeoff score for the 1438 patients at day 3 was 0.73 ±0.32 (median [25th, 75th percentile], 0.92 [0.63, 1.0]), indicating that, on average, patients equated living 1 year in their current state of health with living 8.8 months (0.73 x 12 months) in excellent health. But time-tradeoff scores varied widely from patient to patient Figure 1: A total of 34.8% had utilities of 1.0, meaning that they were unwilling to give up any time in exchange for a shorter life in excellent health, and 9.0% had utilities of 0.04, indicating that they preferred living 2 weeks or less in excellent health to living 1 year in their current state of health. Mean time-tradeoff scores varied little across disease categories, but large interpatient variation was seen in scores within disease categories.

View larger version (16K): [in this window] [in a new window]   Figure 1. Histogram of patients' time-tradeoff scores at day 3 (n = 1438).

The patients' mean score on day 3 for the 0 to 100 overall health rating scale was 57.8 ±24.0 (median, 60 [50, 75]; n = 1327)). Twenty-three (1.7%) patients indicated that their state of health was equal to death (health rating = 0), and two patients indicated that their current state of health was worse than death (health rating < 0).

Patients Compared with their Surrogates and Physicians

Scores varied widely among the 1041 patient-surrogate pairs: A total of 34.7% of patients and 28.2% of surrogates had utilities of 1.0; 9.6% of patients and 14.5% of surrogates had utilities of 0.04. The mean patient utility was 0.73 ±0.33 (median, 0.92 [0.63, 1.0]) and was higher than the mean surrogate utility, which was 0.65 ±0.35 (median, 0.83 [0.38, 1.0]; P < 0.0001) (Figure 2). Differences were also statistically significant within several disease categories: acute exacerbation of severe chronic obstructive pulmonary disease, acute exacerbation of chronic congestive heart failure, chronic liver failure with cirrhosis, and metastatic non-small-cell carcinoma of the lung (Figure 3).

View larger version (14K): [in this window] [in a new window]   Figure 2. Histogram of differences between the time-tradeoff scores of patients and those of their surrogates at day 3 (n = 1041). Differences are expressed as the fraction of a year of excellent health that the patient equated with 1 year in his or her current state of health minus the corresponding value estimated by the surrogate. A difference of zero indicates exact agreement between patient and surrogate. Positive values indicate that patients' time-tradeoff scores were higher; negative values indicate that they were lower.

View larger version (25K): [in this window] [in a new window]   Figure 3. Comparison of the mean time-tradeoff scores of patients and their matched surrogates, overall and by disease category (n = 1041). An asterisk indicates a statistically significant difference (P < 0.05) between the scores of patients and their surrogates. ARF = acute respiratory failure; MOSF = multiorgan system failure; COPD = chronic obstructive pulmonary disease; CHF = congestive heart failure; Ca = cancer.

Patients gave higher ratings to their state of health than did their surrogates and physicians. Among the 959 patient-surrogate pairs that provided rating scores of the patient's health at day 3, the patients' mean rating score was 57.9 ±23.9 (median, 60 [50, 75]), and the surrogates' mean rating score was 49.6 ±22.2 (median, 50 [30, 65]; P < 0.0001). Seventeen patients indicated that their state of health was the same as or worse than death (health rating = 0). The corresponding mean surrogates' rating was 41.0, and no surrogate gave a score of 0. On the other hand, 11 other surrogates rated a patient's health as 0, but the patients themselves gave a mean rating of 56.7. None gave a score of less than 30.

The difference in health ratings at day 3 was even greater for the 1079 patient–physician pairs. The mean patient rating score was 58.1 ±24.2 (median, 60 [50, 75]); the mean physician rating score was 42.5 ±23.5 (median, 40 [25, 60]; P < 0.0001). Among the patient–physician pairs, 20 patients rated their own health as the same as or worse than death, but none of their physicians rated it so; the mean physician rating for those 20 patients was 29.3.

Relation of Time-Tradeoff Utilities to Other Measures

Time-tradeoff utilities were related to patients' preferences for CPR (Tables 2 and 3). Patients who wanted CPR had a mean utility score at day 3 of 0.79 ±0.29 (median, 0.92 [0.63, 1.0]), whereas patients who preferred not to have CPR had a mean utility of 0.61 ±0.37 (median, 0.75 [0.25, 0.92]); P < 0.0001). Time-tradeoff scores were related to patients' preferences for care that focused on extending life as much as possible rather than care that focused on relieving pain and discomfort as much as possible. Scores were also related to the burden of the patient's illness on the family: The greater the burden, the less the patient valued his or her current state of health. The domain-specific health status measures that correlated best with the time-tradeoff scores were the Sickness Impact Profile psychosocial dimension score (r = 0.33) and the Profile of Mood States depression score (r = 0.36): The better the patient's mental health, the higher he or she rated his or her state of health. The health rating scale score also showed moderate correlation with the time-tradeoff score (r = 0.34 to 0.35).

Time-tradeoff scores varied slightly according to race but did not vary according to sex, religion, or employment status. In univariate analyses, time-tradeoff scores did not correlate with income, age, or level of education, and they correlated significantly but weakly (r = 0.11 to 0.29) with patients' own estimates of their prognoses for survival, independent functioning, and comfort and their willingness to tolerate the six adverse outcomes. Severity of acute illness at day 3 and the health status measures of independence in activities of daily living and physiologic reserve also correlated only weakly with patients' utilities (r = 0.09 to 0.20).

In multivariable analyses, time-tradeoff scores at day 3 were related to 0 to 100 health ratings; desire for resuscitation and extension of life rather than relief of pain and discomfort; degree of willingness to live with constant pain; perceived prognosis for survival and independent functioning; and race (Somers D = 0.35). Time-tradeoff utilities at month 2 were related to health ratings, to desire for resuscitation and extension of life rather than relief of pain and discomfort, and to psychosocial well-being (Somers D = 0.42).

Change over Time

Eight hundred fifty-five patients completed time-tradeoff questions at both day 3 and month 2. By month 2, patients' time-tradeoff utilities had improved by an average of 0.06 ±0.32 (P < 0.0001). Utilities of the 732 patients who completed interviews at day 3 and month 6 improved by 0.08 ±0.35 (P < 0.0001). The mean scores of 650 patients interviewed at day 3, month 2, and month 6 increased from 0.78 at day 3 to 0.85 at month 2 and 0.86 at month 6. Scores for surrogates interviewed serially showed similar changes. The mean health rating scale score of surviving interviewed patients increased by 4.0 ±26.5 points from day 3 to month 2 (n = 800; P < 0.0001) and by 7.8 ±25.8 points from day 3 to month 6 (n = 672; P < 0.0001). The mean rating scale scores of patients interviewed at day 3, month 2, and month 6 (n = 593) increased from 59.6 at day 3 to 65.3 at month 2 and 68.0 at month 6.

Changes in patients' time-tradeoff scores from day 3 to month 2 were only weakly associated with changes in ability to do activities of daily living (r = 0.10) and with changes in health ratings (r = 0.25). The mean time-tradeoff score of patients who had wanted CPR at day 3 but who had changed their minds by month 2 increased by 0.02 (n = 95); the mean score of those that did not change their minds increased by 0.06 (n = 589); and the mean score of those who changed to wanting CPR increased by 0.21 (n = 59; P < 0.0012 for change by category). The mean time-tradeoff score of patients who at day 3 preferred treatments that focused on extending life as much as possible but who at month 2 preferred treatments that focused on relieving pain or discomfort as much as possible decreased by 0.02 (n = 132); the mean score of patients who did not change their minds increased by 0.05 (n = 476); and the mean score of patients who initially preferred pain relief but later preferred life extension increased by 0.17 (n = 105; P < 0.0002). Finally, changes in time-tradeoff utilities between months 2 and 6 were weakly related to changes in level of anxiety (r = 0.19) and depression (r = 0.24): Improvements in mental health weakly correlated with increases in health values.

Discussion

Top Methods Results Discussion Author & Article Info References

In recent years, great progress has been made in treating seriously ill patients. Those who may have died quickly in the past can now be kept alive much longer. Yet, advances in the treatment of seriously ill patients have outpaced our understanding of those patients' preferences and the health-related quality of their lives. Are preferences shared by most seriously ill patients or are they highly individual? Our findings suggest that the health values of the seriously ill vary greatly from patient to patient. This implies that the average preference of a group should not be used in a decision involving an individual patient, especially because most patients who can be interviewed can state their preferences themselves. We found the time tradeoff score to have not only excellent test-retest reliability but also construct validity because it strongly correlated with preferences for care that focused on extending life even at the expense of pain and discomfort, and with preferences for CPR.

Not only do the health values of seriously ill patients vary widely but, as other investigators have found [24, 33-41], a patient's value for his or her current state of health cannot easily be predicted from his or her current state of health. One new finding with potential clinical implications is that utilities correlate inversely, albeit modestly, with depression, indicating that a patient's mental health may be an important component of how he or she values his or her state of health and that treating depression may affect how a patient values his or her state of health.

If the health values of seriously ill patients cannot readily be predicted from clinical, demographic, or health status information, can surrogate decision makers or physicians offer reliable substitutes? The health values and ratings of participating patients were generally higher than their surrogates and physicians realized. This finding is tempered by the fact that only the minority of patients in SUPPORT could be interviewed. Thus, one can only speculate on whether our findings would apply to patients unwilling or too ill to participate. Furthermore, patient, surrogate, and physician interviews may have been separated by a few days. The consistent discordance between patients and their surrogates has, nevertheless, potential legal and ethical ramifications: It suggests that simply substituting a surrogate's preferences when a patient cannot or will not participate in decision making may be far from an ideal practice, unless the patient prefers family decision making to alternative strategies such as advance planning.

If patients' preferences are the only "valid" ones, however, the issue then is when to assess them: in anticipation of illness, in the heat of an acute illness, or after the patient has had time to recover and reflect? Unlike other populations that have been studied, in which health values remained stable despite changes in health status (and thus could be assessed at any one point) [33, 34, 42-44], utilities of surviving patients in our study increased over time, suggesting that the preferences of seriously ill patients evolve and may need to be obtained serially.

For decisions that involve resource allocation on a local, state, or national level, it may be helpful to compare preferences across conditions. The mean utility of participating patients in this study was 0.73 (median, 0.92). By comparison, previous studies have found mean utility scores of 0.87 in survivors of myocardial infarction [24], of 0.79 in patients with the acquired immunodeficiency syndrome [35], and of 0.43 to 0.56 in patients with renal failure on dialysis [36]. Utilities such as these can be combined with prognosis to calculate quality-adjusted life expectancy that, in turn, can be used as the denominator of a cost-effectiveness ratio to calculate incremental costs per incremental quality-adjusted year of life saved for various programs [5]. As resources become scarcer, payers may base reimbursement decisions on cost-effectiveness. Indeed, many analysts [45-48], although certainly not all [48-52], have advocated the use of quality-adjusted life years and cost-effectiveness analyses for just such purposes.

Before such information is incorporated into resource allocation decisions, several issues should be resolved. First, the issue of whose values to use—those of the patient or those of someone else—has not been settled [53-58]. The case for using those of the patient is that the patient is the true "expert" when it comes to understanding what it is like to live with a certain disease; the case for using the values of "society" is that taxpayers or insurance premium payers often foot the health care bill. When the health values of patients differ substantially from those of others, the issue is magnified. Another concern is that it is difficult to calculate quality-adjusted life expectancy when utilities are changing [59-62]. Finally, although the practice is widely accepted and done, use of a mean utility score in the denominator of a cost-effectiveness ratio when there is wide interpatient variation [63] and poor correlation with actual health status raises both methodologic [37] and ethical concerns. In practice, the concerns of particularly needy populations or inclinations to fund heroic life-saving procedures may play as great a role in allocation decisions as information about health values and cost-effectiveness does [47, 52, 64].

We show that the health values of the seriously ill vary widely from patient to patient; that they are higher than their surrogates believe; that they are related to only a few other preference and health status measures; and that they increase over time. Findings from this study, if replicated, have ramifications for decision making in both clinical practice and policymaking [6]. Future researchers should address two key areas involving decision making in seriously ill patients. First, they should explore ways to improve understanding of patients' health values among surrogates and caregivers. As a corollary, we may also need to understand the importance patients attach to surrogates' accuracy in replicating the preferences patients state in advance. Second, for situations in which patients will not or cannot participate in decision making and in which surrogate decision makers are not available, we need to understand the determinants of health values and of changes in health values among both healthy and ill persons.

Portions of this paper were presented in abstract form at the 14th and 15th annual meetings of the Society of General Internal Medicine, Seattle, Washington and Washington, D.C.