Hundreds of thousands of patients annually die in or shortly after an ICU admission, generally after decisions to forego life-sustaining treatments (1–3). In most cases, patients are too ill to participate in these decisions and clinicians, therefore, turn to patients’ surrogate decision-makers to help make end-of-life decisions that align with the patient’s values and preferences (34). In order to be informed participants in decision-making, surrogates need a clear understanding of the patient’s prognosis with intensive treatment. However, numerous studies suggest that surrogates of patients with advanced illness often have overly optimistic expectations about prognosis (5–8). Although there is inherent uncertainty in physicians’ predictions for individual patients, physicians’ survival predictions are significantly more accurate than surrogates’ in the setting of critical illness (8). ICU physicians’ discriminant accuracy is also superior to existing risk prediction models (910). Physicians’ judgments of a poor prognosis are independently predictive of patient outcomes among patients who died receiving full life support, lessening concerns that physicians’ prognostic accuracy is solely a result of a self-fulfilling prophecy (11). Despite the imperfect accuracy of physicians’ prognostications, surrogate decision-makers highly value physicians’ predictions, and the vast majority of surrogates wish to hear physicians’ prognostications despite this uncertainty (12).
Although clinicians cite unrealistic expectations about prognosis by surrogates as a barrier to good decision-making in advanced illness (13), there are two important gaps in knowledge about this issue. First, there is a paucity of empirical data about the causes of surrogates’ optimistic expectations. The prevailing assumption is that surrogates’ optimistic expectations arise from miscomprehension of physicians’ prognostications. However, insights from decision science (14) as well as single-center studies in ICUs (815–18) suggest that cognitive biases—rather than solely misunderstandings—may also contribute. One such bias that may be relevant, termed the better-than-average effect, is the tendency of individuals to rate themselves as more likely to have better outcomes than their peers (14). Lack of knowledge of the causes of surrogates’ misperceptions about prognosis makes it difficult to know how to best intervene to improve communication about prognosis.
Second, it is uncertain whether surrogates’ optimistic expectations about prognosis contribute to more intensive treatment at the end of life. According to traditional decision theory (e.g., expected utility theory), individuals will be less willing to authorize intensive treatment as the likelihood of a good outcome diminishes (19). However, modern decision theory suggests that individuals do not behave in purely rational ways, instead that decisions may be influenced by strong emotions and a variety of cognitive biases (20). For example, some have argued that the emotional difficulty of authorizing treatment withdrawal for another person may make surrogates continue life support in the face of a poor prognosis (21). Using hypothetical vignettes with ICU surrogates, Zier et al (22) found that 25% of surrogates were unwilling to withdraw life support when informed that the treating physician judged there to be a less than 1% chance of survival to hospital discharge. Therefore, it remains uncertain whether improving the effectiveness of communication about prognosis should be expected to change end-of-life treatment decisions.
We, therefore, conducted this prospective, multicenter cohort study to determine the prevalence and causes of optimistic expectations about prognosis among surrogates of ICU patients and whether surrogates’ optimistic expectations are associated with healthcare usage at the end of life.
From October 2009 to October 2012, we conducted a prospective multicenter cohort study in 12 ICUs in the National Heart, Lung, and Blood Institute Acute Respiratory Distress Syndrome Clinical Trials Network. The medical centers were located in California, Massachusetts, North Carolina, Pennsylvania, and Washington.
Patients and Procedures
We enrolled incapacitated, mechanically ventilated, adult patients at high risk of death or severe, long-term functional impairment, their surrogate decision-makers, and their attending physician. Patient inclusion criteria included lack of decision-making capacity, a diagnosis of acute respiratory distress syndrome using traditional clinical criteria (23), an Acute Physiology and Chronic Health Evaluation (APACHE) II score greater than or equal to 25, or greater than or equal to 50% chance of severe long-term functional impairment as judged by their attending physician (defined as requiring ongoing assistance with at least two activities of daily living). Decisional incapacity was determined through clinical assessment by the patient’s treating physician. We excluded patients who were awaiting organ transplantation, actively dying, or had no surrogate available. We enrolled the patient’s legally designated surrogate decision-maker. If there was no legally designated surrogate, we enrolled the individual acting as the patient’s decision-maker for clinical care decisions. We excluded individuals who were not able to complete study procedures in English. We enrolled the patient’s attending physician of record or his/her designee, defined as a physician directly supervised by the attending physician. This study was approved by the Institutional Review Board of each participating institution. All participants provided written consent for all study procedures; the patient’s surrogate provided proxy consent for review of the patient’s medical record.
On patients’ 5th day of mechanical ventilation and within 1 hour of each other, surrogates and physicians independently estimated the likelihood that the patient would survive the hospitalization using a previously validated question: “What do you think are the chances that your loved one/the patient will survive this hospitalization if the current plan of care stays the same?” (1824) Figure 1 illustrates the probability scale used to record subject’s prognostic estimates. To minimize the chance of response errors among participants with limited numeracy, the anchors on the response scale contained non-numeric expressions of risk (i.e., “No chance of survival” and “Will definitely survive”). Such scales are well validated to assess risk perceptions and may be less affected by limited numeracy than other methods to elicit quantitative risk estimates (25–27). Surrogates and physicians were blinded to other’s response. Surrogates also recorded what they perceived to be the physician’s prognostic expectation with the following question: “If you had to guess, what do you think the doctor thinks is the chance that your loved one will survive this hospitalization if the current plan of care stays the same?” All data were collected after at least one family meeting had occurred.
Appendix C (Supplemental Digital Content 1, http://links.lww.com/CCM/E572) describes other covariates recorded from surrogate decision-makers and physicians, including surrogates’ trust in physicians, health literacy, religiosity, quality of communication with physicians, and symptoms of depression.
We quantified the proportion of surrogates with optimistic expectations about prognosis compared with the treating physicians defined as the surrogate’s prognostic estimate being at least 20% more optimistic than the physician’s. The rationale for choosing a 20% difference as clinically significant is that prior studies using hypothetical cases suggest that changes in prognostic expectations of roughly this magnitude are associated with changes in patients’ willingness to accept life support (2829).
To determine how often surrogates’ optimistic expectations about prognosis arose from misunderstanding the physician’s prognostic expectations, we determined the proportion of cases in which there was a difference between the physician’s estimate and the surrogate’s best guess of the physicians’ prognostic expectations, which indicated misunderstanding. To determine the frequency with which prognostic discordance arose from different beliefs about prognosis by surrogates and physicians, we determined the proportion of cases in which there was a difference between the surrogate’s perception of the physicians’ prognostic expectations and the surrogate’s own prognostic estimate.
At the time of hospital discharge, we abstracted from the medical record the ICU and hospital length of stay, the duration of mechanical ventilation, the timing of withdrawal of life support, and patients’ vital status.
We calculated that a sample size of 275 patients with APACHE II score greater than or equal to 25 would result in a sample with 105 patients who died, which would yield 80% power to detect a correlation of at least 0.26 between optimistic expectations by surrogates and length of stay on a two-sided 0.05 level test. This sample size also gives 80% power to detect a multiple R-squared as small as 0.11, equivalent to multiple correlation of 0.33, in a multiple linear regression with five predictors using a two-sided alpha error of 0.05.
We performed linear regressions to evaluate the relationship between optimistic expectations by surrogates and 1) duration of ICU stay, 2) duration of mechanical ventilation, and 3) time to withdrawal of life support. These variables were right-skewed using the Shapiro-Wilk test; therefore, log-transformation of the outcome variables was required to normalize the variables before fitting them into linear regression models. We also performed multilevel modeling to assess for physician-level clustering, which did not reveal significant clustering. To assess for clustering by study site, we performed likelihood ratio tests, which showed no statistical difference in using ordinary least square regression as opposed to multilevel regression clustering by sites. We, therefore, present results of standard linear regressions. We identified confounding variables at the patient, clinician, and surrogate level using significance test methods (30). We first identified factors associated with length of stay in the univariate analyses with a p value of less than 0.20 (Appendix A, Supplemental Digital Content 1, http://links.lww.com/CCM/E572). Then we used stepwise regression to select the subset of factors to be incorporated into the multivariable model. We performed similar linear regressions for patients who survived to hospital discharge.
We used logistic regression and the variable selection strategy described above to examine whether optimistic expectations by surrogates were associated with higher odds of patient survival to hospital discharge. All analyses were performed with Stata Version 14.0 (StataCorp, College Station, TX).
Among 405 eligible patients, surrogate decision-makers for 275 agreed to participate, for an overall enrollment rate of 68%. All 150 physicians who treated these patients agreed to participate in the study. There were no differences in the demographic characteristics of enrolled versus nonenrolled patients (Appendix B, Supplemental Digital Content 1, http://links.lww.com/CCM/E572). Tables 1–3 list characteristics of the enrolled patients, surrogates, and physicians. The in-hospital mortality rate was 44% (122/275; 95% CI, 39–50%)
Prevalence of Optimistic Prognostic Expectations by Surrogates
Overall, 45% of surrogate decision-makers (109/245; 95% CI, 38–51%) held prognostic expectations that were at least 20% more optimistic that the physicians. Among these patients, the average prognostic estimates for hospital survival among surrogates and physicians were 86% (± 19) and 48% (± 26), respectively. Among the 107 patients who died, optimistic expectations by surrogates were present in 42% (45/107; 95% CI, 33–51%) of surrogates. Among these cases, the mean ± sd prognostic estimates for hospital survival from surrogates and physicians were 79.4% ± 21.0 and 36.4% ± 26.5, respectively. Appendix C (Supplemental Digital Content 1, http://links.lww.com/CCM/E572) contains a summary of the discriminant accuracy and calibration of physicians’ and surrogates’ predictions.
Sources of Optimistic Prognostic Expectations by Surrogates
In 52 of 109 patients (48%), the discordance arose from the surrogate misunderstanding the physician’s prognostic expectations. In 49 patients (45%), the discordance arose from both surrogates misunderstanding physicians’ prognostic expectations and surrogates holding systematically more hopeful beliefs about the patient’s prognosis compared with what they heard from the physicians. In seven patients (6%), the prognostic discordance was caused only by surrogates holding more hopeful beliefs about the patient’s prognosis compared with what they heard from the physician. Data were missing from one surrogate who did not respond to the question eliciting what they thought the physicians’ prognosis estimate was.
Optimism by Surrogates and Patient Survival
In a multivariable model adjusted for patient age and APACHE II, there was no significant association between higher levels of optimism by surrogates and patients’ odds of survival to hospital discharge (odds ratio, 1.39; 95% CI, 0.80–2.41; p = 0.25).
Association Between Optimistic Prognostic Expectations and Healthcare Usage
Among patients who died in the hospital, the median unadjusted ICU length of stay and duration of mechanical ventilation were 8 days (interquartile range [IQR], 4–15 d) and 8 days (IQR, 3–15 d), respectively; 75% (91/122) of deaths occurred after life support was withdrawn and an additional 13% (16/122) occurred after life support was withheld (Appendix D, Supplemental Digital Content 1, http://links.lww.com/CCM/E572). In a multivariable model after log-transformation and adjustment for severity of illness (APACHE II), surrogate race, and relationship to the patient, optimistic expectations by surrogates were associated with a significantly longer ICU stay (β coefficient = 0.44; 95% CI, 0.05–0.83; p = 0.027), corresponding to a 56% longer hospital stay before death (Table 4). Optimistic expectations were also associated with a significantly longer time to withdrawal of life support among dying patients (β coefficient = 0.61; 95% CI, 0.16–1.07; p = 0.009), corresponding to an average 6.5 more days of life support before death. Among patients who survived their hospitalization, optimistic expectations by surrogates were not associated with longer hospital ICU stay (p = 0.78) (Appendix E, Supplemental Digital Content 1, http://links.lww.com/CCM/E572).
We conducted two sensitivity analyses to verify that the relationship between optimistic expectations and duration of intensive treatment at the end of life is robust to different analytic approaches. Both yielded qualitatively similar conclusions to those from the main analysis (Appendix F, Supplemental Digital Content 1, http://links.lww.com/CCM/E572).
First, we used an alternative definition of optimistic expectations: a survival expectancy ratio (SER) exceeding 1.2. The SER is the ratio of the surrogate’s prognostic expectations to the physician’s expectations. The use of such ratios has previously been used to quantify discrepancies between physicians’ and patients’ prognostic estimates (24). A potential advantage of the SER compared with choosing a single absolute difference to signify prognostic discordance is that it may better account for the possibility that as the patient’s prognosis worsens smaller difference between clinician’s and surrogate’s expectations may be important.
Second, we maintained the original definition of optimistic expectations and excluded patients for whom the treating physician judged that the patient had greater than 80% chance of survival; the rationale for this analysis is that the lack of optimistic expectations in these cases arises because the physician’s expectations of survival are so high in these patients that surrogates cannot be 20% more optimistic.
We found that nearly half of surrogate decision-makers for incapacitated ICU patients at high risk of death or severe disability held substantially more optimistic expectations about prognosis compared with the treating physicians. These expectations arose from both misunderstandings by surrogates of physicians’ prognoses as well as from surrogates holding more optimistic beliefs about the patient’s prognosis compared with what they heard from the physician. Optimistic expectations by surrogates were associated with significantly longer duration of intensive treatment at the end of life without an increase in survival.
Our results suggest that problems persist with clinician-family communication in ICUs, despite substantial efforts in the last 2 decades to improve this aspect of ICU care. The prevalence of misperceptions about prognosis observed in our study was similar to that documented by Azoulay et al (5) and also similar to those observed in more recent single-center studies (68). The lack of improvement in this particular aspect of end-of-life care is consistent with published reports that patients’ and families’ perceptions of the quality of end-of-life care have not improved in the last decade (31).
The results of the present study suggest that the findings from prior single-center (8) and simulation-based studies (16–18) on this topic are broadly generalizable. Specifically, our data indicate that surrogates’ optimism about prognosis arises from both miscomprehension by surrogates of physicians’ expectations as well as surrogates holding systematically more optimistic beliefs than what they heard from the physicians. These findings fit with insights from decision psychology about optimism bias, which is the tendency of individuals to view themselves as more likely to have good outcomes compared with others (14). Taken together, these studies suggest that interventions to improve communication about prognosis need to address both the comprehensibility of prognostic information and also attend to the psychologic and affective complexities for surrogates coming to terms with news of a poor prognosis (15–1732–34).
It is uncertain how to intervene to improve the accuracy of surrogates’ prognostic expectations. In fact, expert clinicians such as palliative care consultants often avoid direct attempts to change surrogates’ prognostic expectations, instead align with surrogates’ optimism, attend to their emotions, and gently help them move in the direction of acceptance (35). A recent trial testing the effect of trained interventionists providing clear prognostic information to surrogates in ICUs found no impact on treatment decisions, but more psychologic distress among surrogates in the intervention arm (36). One possible explanation for these findings is that the protocolized intervention did not allow the interventionists to longitudinally support surrogates through the emotional difficulty of receiving news of a loved one’s poor prognosis.
The finding that optimism about prognosis by surrogates is associated with more invasive treatment at the end of life highlights the potential clinical and economic consequences of ineffective communication in ICUs. Our results may partially explain the results of the study to understand prognoses and preferences for outcomes and risks of treatments (SUPPORT) trial, which showed no effect on healthcare usage near the end of life from an intervention in which physicians were provided with model-derived prognostic estimates for patients with serious illness (37). In the SUPPORT trial physicians rarely shared the prognostic estimates with patients/surrogates, yet our data suggest that surrogates’ prognostic expectations—which differ from physicians’—may substantially influence usage. Our results are consistent with studies of nursing home residents and elderly outpatients, which reported that individuals’ were less willing to authorize intensive treatment when they had accurate expectations about prognosis (2938).
This study has several limitations. First, all study ICUs had academic affiliations; therefore, it is uncertain whether our findings can be generalized to nonacademic settings and regions of the United States not represented in the trial sites. Second, the cohort was predominantly Caucasian, and these results may not generalize to other racial and ethnic groups. Third, the cohort study design does not allow us to make causal inferences from the observed associations. Fourth, due to the logistical challenges of studying surrogate decision-makers in crisis situations, we did not make serial measurements of prognostic estimates and, therefore, were not able to quantify the natural history of surrogates’ prognostic expectations during a terminal hospitalization and how this relates to decisions about life support.
This multicenter study shows that optimistic expectations about prognosis are prevalent among surrogates of patients with advanced critical illness, arise both from misunderstandings by surrogates and from surrogates holding more hopeful beliefs than what they heard from physicians and are associated with a longer duration of intensive treatment at the end of life. These findings underscore the need to develop strategies to improve the comprehensibility of physicians’ prognostications, and also to attend to the emotional and psychologic challenges surrogates face when confronted with news of a poor prognosis.
1. Sprung CL, Cohen SL, Sjokvist P, et al.; Ethicus Study Group: End-of-life practices in European intensive care units: The Ethicus Study. JAMA 2003; 290:790–797
2. Angus DC, Barnato AE, Linde-Zwirble WT, et al.; Robert Wood Johnson Foundation ICU End-Of-Life Peer Group: Use of intensive care at the end of life
in the United States: An epidemiologic study. Crit Care Med 2004; 32:638–643
3. Prendergast TJ, Claessens MT, Luce JM. A national survey of end-of-life care for critically ill patients. Am J Respir Crit Care Med 1998; 158:1163–1167
4. Silveira MJ, Kim SY, Langa KM. Advance directives and outcomes of surrogate decision making before death. N Engl J Med 2010; 362:1211–1218
5. Azoulay E, Chevret S, Leleu G, et al. Half the families of intensive care unit patients experience inadequate communication
with physicians. Crit Care Med 2000; 28:3044–3049
6. Cox CE, Martinu T, Sathy SJ, et al. Expectations and outcomes of prolonged mechanical ventilation. Crit Care Med. 2009; 37:2888–2894; quiz 2904
7. Teno JM, Fisher E, Hamel MB, et al. Decision-making
and outcomes of prolonged ICU stays in seriously ill patients. J Am Geriatr Soc 2000; 48:S70–S74
8. White DB, Ernecoff N, Buddadhumaruk P, et al. Prevalence of and factors related to discordance about prognosis
between physicians and surrogate decision makers of critically ill patients. JAMA 2016; 315:2086–2094
9. Sinuff T, Adhikari NK, Cook DJ, et al. Mortality predictions in the intensive care unit: Comparing physicians with scoring systems. Crit Care Med 2006; 34:878–885
10. Detsky ME, Harhay MO, Bayard DF, et al. Discriminative accuracy of physician and nurse predictions for survival and functional outcomes 6 months after an ICU admission. JAMA 2017; 317:2187–2195
11. Rocker G, Cook D, Sjokvist P, et al.; Level of Care Study Investigators; Canadian Critical Care Trials Group: Clinician predictions of intensive care unit mortality. Crit Care Med 2004; 32:1149–1154
12. Evans LR, Boyd EA, Malvar G, et al. Surrogate decision-makers’ perspectives on discussing prognosis
in the face of uncertainty. Am J Respir Crit Care Med 2009; 179:48–53
13. You JJ, Downar J, Fowler RA, et al.; Canadian Researchers at the End of Life
Network: Barriers to goals of care discussions with seriously ill hospitalized patients and their families: A multicenter survey of clinicians. JAMA Intern Med 2015; 175:549–556
14. Alicke MD, Klotz ML, Breitenbecher DL, et al. Personal contact, individuation, and the better-than-average effect. J Pers Soc Psychol 1995; 68:804
15. Schenker Y, White DB, Crowley-Matoka M, et al. “It hurts to know… and it helps”: Exploring how surrogates in the ICU cope with prognostic information. J Palliat Med 2013; 16:243–249
16. Zier LS, Burack JH, Micco G, et al. Doubt and belief in physicians’ ability to prognosticate during critical illness
: The perspective of surrogate decision makers. Crit Care Med 2008; 36:2341–2347
17. Zier LS, Sottile PD, Hong SY, et al. Surrogate decision makers’ interpretation of prognostic information: A mixed-methods study. Ann Intern Med 2012; 156:360–366
18. Lee Char SJ, Evans LR, Malvar GL, et al. A randomized trial of two methods to disclose prognosis
to surrogate decision makers in intensive care units. Am J Respir Crit Care Med 2010; 182:905–909
19. von Neumann J, Morgenstern O. Theory of Games and Economic Behavior. 2004Princeton, NJ, Princeton University Press.
20. Kahneman D. A perspective on judgment and choice: Mapping bounded rationality. Am Psychol. 2003; 58:697–720
21. Rubenfeld GD, Elliott M. Evidence-based ethics? Curr Opin Crit Care 2005; 11:598–599
22. Zier LS, Burack JH, Micco G, et al. Surrogate decision makers’ responses to physicians’ predictions of medical futility. Chest 2009; 136:110–117
23. Bernard GR, Artigas A, Brigham KL, et al. The American-European Consensus Conference on ARDS. Definitions, mechanisms, relevant outcomes, and clinical trial coordination. Am J Respir Crit Care Med 1994; 149(3 Pt 1):818–824
24. Cox CE, Carson SS. Medical and economic implications of prolonged mechanical ventilation and expedited post-acute care. Semin Respir Crit Care Med 2012; 33:357–361
25. Gigerenzer G, Edwards A. Simple tools for understanding risks: From innumeracy to insight. BMJ 2003; 327:741–744
26. Huskisson EC. Measurement of pain. Lancet 1974; 2:1127–1131
27. Holland JC. Now we tell–but how well? J Clin Oncol 1989; 7:557–559
28. Lloyd CB, Nietert PJ, Silvestri GA. Intensive care decision making in the seriously ill and elderly. Crit Care Med 2004; 32:649–654
29. Fried TR, Bradley EH, Towle VR, et al. Understanding the treatment preferences of seriously ill patients. N Engl J Med 2002; 346:1061–1066
30. Maldonado G, Greenland S. Simulation study of confounder-selection strategies. Am J Epidemiol 1993; 138:923–936
31. Teno JM, Freedman VA, Kasper JD, et al. Is care for the dying improving in the United States? J Palliat Med 2015; 18:662–666
32. Apatira L, Boyd EA, Malvar G, et al. Hope, truth, and preparing for death: Perspectives of surrogate decision makers. Ann Intern Med 2008; 149:861–868
33. Schenker Y, Crowley-Matoka M, Dohan D, et al. I don’t want to be the one saying ‘we should just let him die’: Intrapersonal tensions experienced by surrogate decision makers in the ICU. J Gen Intern Med 2012; 27:1657–1665
34. Cox CE, Lewis CL, Hanson LC, et al. Development and pilot testing of a decision aid for surrogates of patients with prolonged mechanical ventilation. Crit Care Med 2012; 40:2327–2334
35. Jackson VA, Jacobsen J, Greer JA, et al. The cultivation of prognostic awareness through the provision of early palliative care in the ambulatory setting: A communication
guide. J Palliat Med 2013; 16:894–900
36. Carson SS, Cox CE, Wallenstein S, et al. Effect of palliative care-led meetings for families of patients with chronic critical illness
: A randomized clinical trial. JAMA 2016; 316:51–62
37. 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). The SUPPORT Principal Investigators. JAMA 1995; 274:1591–1598
38. Mitchell SL, Teno JM, Kiely DK, et al. The clinical course of advanced dementia. N Engl J Med 2009; 361:1529–1538