Choices of Seriously Ill Patients About Cardiopulmonary Resuscitation: Correlates and Outcomes

CLINICAL STUDIES

Choices of Seriously 111Patients About Cardiopulmonary Resuscitation: Correlates and Outcomes Russell S. Phillips, MD, Boston,Massachusetts, Neil S. Wenger, MD, LOSAngeles, California, Joan Teno, MD, Hanover, New Hampshire, Robert K. Oye, MD, Los Angeles, California, Stuart Youngner, MD, Cleveland, Ohio, Robert Califf, MD, Durham, North Carolina, Peter Layde, MD, Norman Desbiens, MD, Marshfield. Wisconsin. Alfred F. Connors. Jr.. MD. Cleveland, Ohio, Joanne Lynn, MD, Hanover, New Hampshire, for the SUPPORTinvestigators ’ ’ ’ PURPOSE: For patients hospitalized with serious illnesses, we identified factors associated with a stated preference to forgo cardiopulmonary resuscitation (CPR), examined physician-patient communication about these issues, and deter mined the relationship of patients’ preferences to intensity of care and survival. PATIENTS AND METHODS: The study was a crosssectional evaluation of patient preferences. The setting was five geographically diverse academic acute-care medical centers participating in the SUPPORT (Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments) project. Study participants were hospitalized patients 218 years of age with 1 of 9 serious illnesses who were interviewed between days 3 and 6 after qualifying for the study. Using standardized interviews, patients provided information on demographics, preferences for CPR and other treatments, quality of life, functional status, perceptions of prognosis, and whether the patient had discussed CPR preferences with his or her physician. Data abstracted from the medical record included physiologic measures, therapeutic intensity, whether CPR was provided, and whether there was a do-not-resuscitate order. RESULTS: Of 1,955 eligible patients, 84% were interviewed (mean age 62 years, 58% men, in-

From the Division of General Medicine and Primary Care (RSP),Beth Israel Hospital and Harvard Medical School, Boston, Massachusetts; UCLAMedical Center (NSW,RKO),Los Angeles, California; DartmouthHitchcock Medical Center (JT. JL), Hanover, New Hampshire; University Hospital of Cleveland (SY), Case Western Reserve University Medical School, Cleveland, Ohio; Duke University Medical Center (RC),Durham, North Carolina; Marshfield Clinic (PL, ND), Marshfield, Wisconsin; Cleveland MetroHealth Medical Center (AFC),Case Western Reserve University Medical School, Cleveland, Ohio; and the ICU Research Group, George WashingtonUniversity, Washington,DC. Presented at the meeting of the Societv for General Internal Medicine, May l-4, 1991, Seattle, Washington. Supported by the Robert Wood Johnson Foundation. The opinions and findings contained in this manuscript are those of the authors and do not necessarily represent the views of the Robert Wood Johnson Foundation or their Board of Trustees. Requests for reprints should be addressed to Russell S. Phillips, MD, Division of General Medicine and Primary Care, Beth Israel Hospnal, 330 Brookline Avenue, Boston, Massachusetts 02215. Manuscript submitted April 5, 1994 and accepted in revised form 4ugust 1, 1995.

128

February

hospital mortality 7%‘ 6-month mortalhy 33%). Of the respondents, 28% did not want CPR. Factors associated independently with not wanting CPR included: hospital site; diagnosis; being older; being more functionally impaired; and patient perception of a worse prognosis. Only 29% of patients had discussed their preferences with their plhysician; 48% of those who did not want CPR reported such discussions. After adjusting for illness severity and factors associated with CPR preferences, patients not wanting CPR had lower intensity of care; similar inhospital mortality; and higher rnottality at 2 and 6 months following study entry.. CONCLUSIONS: The diagnosis, patients’ perception of the prognosis, and hospital site were significantly associated with patients’ resuscifation preferences after adjusting for patient cdemographics, severity of illness, and functional status. The rate of discussing CPR was low even for patients who did not want CPR. Patient preferences not to receive CPR were associated with a small decrease in intensity of care but no difference in hospital survival. he principle of patient autonomy af6rm.spatients’ rights to make decisions regarding their own medT ical treatment.’ Increasingly in the United States, patients are asking and being encouraged to assert this right as principal participants in clinical decision making. The responsibilities of physicians and other members of the health care team include facilitating informed decisions2 Essential to this decision-making process is an understanding of the patients’ preferences for care, and factors that intluence these preferences. A decision commonly faced by seriousl!y ill hospitalized adults is whether to attempt or to?forgo car-

diopulmonary resuscitation (CPR) in the event of cardiac arrest. As information has become available on patients’ outcomes following CPR, it has become clear that only a small minority will be returned to their prior state of health.“7 Previous studies have demonstrated much variation in CPR decisions for hospitalized patients, including variation by diagnosis and age.“13Nevertheless, few data exist on how patients’ demographic and medical characteristics, and their attitudes and perceptions affect prefer-

1996 The American Journal of Medicine@ Volume 100

PATIENT PREFERENCES FOR CARDIOPULMONARY RESUSCITATION/PHILLIPS ET AL

ences. For example, in formulating preferences for CPR, how important are factors such as patients’ perceptions of their prognosis and quality of life, their age, gender, race, and their underlying illness? An improved understanding of factors influencing patients’ preferences might help physicians to address resuscitation with patients more effectively. To improve our understanding of factors influencing patients’ resuscitation decisions, we studied the prevalence of patients’ preferences to forgo CPR and the correlates of these preferences among patients enrolled in the SUPPORT (Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments) project, a national multicenter study of seriously ill hospitalized patients’ preferences and outcomes. Also, we examined the rate of CPR discussions between patients and their physicians, and evaluated the relationships among patients’ preferences, the intensity of medical care they received, and survival.

PATIENTS AND METHODS Study Design This analysis was performed using data collected as part of the SUPPORT project. The objectives of this project have been published previ~usly.‘~ Inpatients were enrolled prospectively at five geographically diverse academic medical centers. Patients were eligible if they were at least 18 years of age, and met defined criteria for at least one of the following nine diagnostic categories: acute respiratory failure; chronic obstructive lung disease; congestive heart failure; cirrhe sis; nontraumatic coma; metastatic colon cancer; advanced non-small-cell lung cancer; multiorgan system failure with sepsis, or multiorgan system failure with malignancy. Specific diagnostic criteria were designed to identify patients at the late or advanced stages of their illnesses, when the average probability of &month survival was estimated to be 50%. Patients were screened for eligibility upon hospital admission and those in intensive care units were screened daily. Patients were excluded if, at the time of hospital admission, they were pregnant, nonEnglish speaking, nonresident foreign nationals, transferred from another hospital to a nonintensive care setting, diagnosed as having AIDS, hospitalized with an expected stay of less than 72 hours, or admitted with brain wury following head trauma Data from eligible patients who were discharged or had died within 48 hours of study entry were excluded. Patients with AIDS were excluded because a primary objective of the first phase of SUPPORT was to develop prognostic models, and the investigators believed that the prognosis for patients with AIDS would change substantially during the time of the study. This analysis includes only patients who were eligible for interview. Patients enrolled in the study were

excluded from interview if they were unable to communicate because of coma or cognitive impairment, or for other reasons. Additionally, intubated patients were excluded from interview. Cognitive function was assessed using a revised version of the orientation and information subtest of the Wechsler Memory Sca.le.15

Data Collection Data were collected from chart abstraction and interviews with patients. The data gathered by chart review for this analysis of patient preferences included: diagnosis; comorbid conditions; acute physiology score (APS) on day 1 and day 3 following study entry; whether a do-not-resuscitate (DNR) order was written; whether CPR was provided; and intensity of resource utilization. The APS is the physiology-based component of APACHE III and includes ph,ysiologic measurements as well as the Glasgow coma score, a measure of neurologic function.16 The APS (ranging in this sample from 3 to 102) has been shown previously to predict in-hospital mortality, with a higher score indicating increased risk of in-hospital death. l6 Intensity of resource utilization was measured using a :modified version of the therapeutic intensity scoring system (TISS), a weighted score of medical care services during the index hospitalization. This score is a valid and reliable measure of resource utilization and highly correlated with hospital charges.17 The TISS score (which ranges from 1 to 68 for the sample, with a higher score indicating greater resource use) was collected on study days 1, 3, 7, 14, and 25 if the patient remained hospitalized. Comorbid conditions were obtained by chart review using a list developed as part of the APACHE II scoring system; a comorbidity score was calculated by a simple count of the comorbid conditions, which ranged from 0 to 7.18In addition to chart review data, a Cox-model probability estimat’e for survival at 2 months based on the experience of lhe study cohort was used as an additional measure of severity of illness in the analysis.1g Patients were interviewed between hospital day 3 and 6 following study enrollment by trained interviewers using standardized techniques to elicit information on patients’ function 2 weeks prior to admission, quality of life, expectations for survival and function, willingness to tolerate adverse outcomes, preferences for aggressive versus comfort care, and demographics. Function was measured as the number of dependencies among 7 activities on the activities of daily living (ADL) scale using a revised version of the Katz ADL scale.20 Quality of life was measured by asking patients to rate their overall quality of life using a 5-point descriptive scale (with 1 being excellent and 5 being poor). Prognostic estimates were obtained by asking patients to estimate their probability of survival 2 months following study en-

February

1996 The American Journal of Medicine’

Volume ,100

129

PATIENT PREFERENCES FOR CARDIOPULMONARY RESUSCITATION/PHILLIPS ET AL

try and the probability of independent function 2 months following study entry. Possible responses included “90% or better, n “about 75%,” “about 50-50,” “about 25%,” and “10% or less.” Patients were also asked about their willingness to live permanently with certain situations. These included being in pain, being attached to breathing machines, and being unconscious or in a coma. Possible responses included ‘Very willing,” ‘lwilling,” “unwilling,” “very unwilling,” or “rather die.” To learn about patients’ preferences for aggressiveness of care, they were asked the following: “If you had to make a choice at this time, would you prefer a course of treatment that focuses on extending life as much as possible, even if it means having more pain and discomfort, or would you want a plan of care that focuses on relieving pain and discomfort as much as possible, even if it means not living as long?” Possible responses included: “extend life as much as possible, ” “relieve pain or discomfort as much as possible,” or “don’t know.” Demographics included age, gender, race, education, marital status, religion, and whether the patient was living alone. Patients were asked about their CPR preference using the following question: “As you probably know, there are a number of things doctors can do to try to revive someone whose heart has stopped beating, which usually includes a machine to help breathing. Thinking of your current condition, what would you want your doctors to do if your heart ever stopped beating? Would you want your doctors to try to revive you or would you want your doctors not to try to revive you?” Possible responses were (1) “I would want my doctor to try to revive me,” (2) “I would not want my doctor to try to revive me,” (3) “I would want CPR but no intubation,” or (4) “Don’t know.” The first two choices were responses offered by the interviewer while the latter two were recorded only if volunteered by the patient. Interviewers were instructed not to provide any additional information about CPR to avoid interviewer-related bias. For this analysis, patients who wanted CPR but no intubation (5%) were grouped with those patients who wanted full CPR. Test/retest reliability (exact agreement) assessed within 24 hours of the initial interview for 90 patients on the CPR preference question was 93%. Lastly, patients were asked whether they had discussed their resuscitation preference with their physician. We collected data on survival to 6 months using a standard follow-up protocol that included frequent telephone contact with the patient or the family. Sixmonth outcomes were available for 96% of enrolled patients. For the remaining 4% of patients, we searched the National Death Index to ascertain survival to 6 months. The study protocol was approved by the institutional review board at each clinical site. 130

Prior to interviews, patients were asked to give verbal informed consent, using a standardized.procedure.

Statistical Analysis For the bivariable analysis of factors associated with CPR preferences, we used the &i-square test for categorical variables, and the Wiicoxon rank-sum test for continuous or ordered variables. Paltients who preferred to forgo CPR were compared with those who preferred to undergo CPR or who answered “don’t know.” Patients who were uncertain were combined with patients who wanted CPR since CPR is generally provided in the event of uncertainty. For the multivariable analysis of CPR preferences, we selected variables shown in the literature21m2* or by clinical experience to be related to patient resuscitation preferences or the assignment of DNR ordlers: patient age; race; marital status; gender; religion; income; whether the patient lived alone; level of education; health insurance coverage; study site; dia,gnostic category; number of comorbid conditions; whether the patient had cancer (either as a primary diagnosis or comorbid condition); patient’s estimate of 2-month survival; patient’s assessment of quality of life; function (ADL); APS; and the Cox-model probability estimate for survival at 2 months based on information available on the third study day.lg Logistic regression was used to examine the association of individual factors with CPR preference while controlling for other variables included in the mode1.2gFactors considered significant in the multivariable model were those with a P valule 10.05. To examine the effect of including patients with “don’t know” responses to the CPR question, the analysis was repeated after excluding these patients. We used the area under the receiver operator ch.aracteristic (ROC) curve (c statistic) to assess the model.30The ROC curve area is a measure of overall predictive discrimination, which, for this study, is defined as the ability to distinguish patients who did not want CPR from those who wanted CPR. An ROC curve area of 0.5 indicates no discrimination, whereas an area of 1.0 indicates perfect discrimination. The association of patients’ preferences for CPR with actual treatment decisions was evaluated by examining the relation between their preferences and the following outcomes: in-hospital death; DNR orders; the use of CPR; mortality at 2 and 6 months following study entry; and average intensity of resource utilization (average TISS measured during hospitalization). To determine the independent effect of CPR preference on these outcomes, we adjusted for study site, diagnosis, age, gender, APS, functional status, race, Cox-model probability estimate of prognosis, and the patient’s own estimate of prognosis.

February 1996 The American Journal of Medicine@ Volume 100

TABLE I Comparison of Patients Interviewed Between Study Days 3 and 7 and Noninterviewed Patients Interviewed Not Interviewed Factor (n = 1,650) (n = 2,651) Median age (y) (25th, 75th percentile) 63.5(53,72) 65.1(53,75) Gender (% male) 58.2 56.2 Median APS’, day 1 (25th, 75th percentile) 35 (23,49) 57 (39,75) Diagnostic category (%I CHFor COPD 43.4 19.0 Cirrhosis 10.2 5.5 ARF, MOSF,or coma 23.2 62.4 Lung or colon cancer 23.2 13.1 Median number of comorbid conditions (25th, 75th percentile) 2 (1, 3) 111,2) Median length of hospital stay (d) (25th, 75th percentile) 13 (7, 24) 9 (6, 141 l&hospital mortality 1%) 6.9 39.2 Six-monthmortality (%I 33.2 57.4 DNRorders written by study day 3 (%I 9.1 17.3

P Value

0.02 0.18


‘Acute Physiology Score from APACHEIll. CHF = congestive heart failure; COPD = chronic obstructive pulmonary disease; ARF = acute respiratory failure; MOSF = multiorgan system failure; DNR = do not resuscitate.

RESULTS

Patient Preferences for CPR

Of 1,650 interviewed patients, 1,585 ((96%) responded to the question regarding CPIR preferOf 4,301patients enrolled in SUPPORT, 1,955(45%) ences. Of these, 59% (930) wanted doctors to try to met eligibility criteria for interview between the third revive them, 5% (81) wanted CPR but no intubation and sixth day following study entry. The primary rea- or respirator, 28% (438) did not want CPR, and 9% sons for ineligibility were the following: being co- (136) were uncertain. Responses were missing matose or intubated (55%); unable to communicate from 53 patients because the questions were for other reasons (15%); having died (10%) or having skipped or the interview was not completed; 11 pabeen discharged (11%) during the interview window; tients refused to answer the question. One patient and failing the cognitive screen (9%).Among the 1,955 was excluded as the only patient with a qualifying eligible patients, interviews were completed for 1,650 diagnosis of coma who had been able to respond (response rate 84%).Of eligible patients who were not to the initial interview. Excluded patients
Response Rates

Comparison of Interviewed and Noninterviewed Patients

BIVARIABLECORRELATES OF NOT WANTING CPR Demographic Factors and

A comparison of noninterviewed patients and inter- Characteristics of Patient’s Illness In the bivariable analyses, patients who did not viewed patients is shown in Table I. The age and genwant CPR were older (mean age 65.6 years versus der distributions were similar in the two groups. Compared with noninterviewed patients, those inter- 60.3,P
131

TABLE II Bivariable Analysis of Factors Associated With Patients not Wanting Cardiopulmonary Number of InterviewedPatients PefcentofPa8ents Factor Wti Factor I%)’ Not Wantine CPR Demographics Hospital site 1

2 3 4 5 Age <50 50-75 >75 Race Nonwhite White Marital status Married Widowed Divorced/separated Single Gender Female Male Insurance class None or Medicaid Medicare Private (commercial) Income $25K or more $1 l-$25K Under SllK Characteristics of patient’s illness Diagnosis Cirrhosis CHF MOSF with sepsis ARF MOSF with malignancy Lung cancer Colon cancer COPD ADL dependencies Less than 2 2 or more Cancer None Localized Metastatic

248 346 407 366 218

Resuscitation (n = 1,585) Relative Risk for Not Wanting CPR 195% cl1

(16) (22) (26) (23) (141

14% 24% 29% 30% 43%

0.47 0.83 1.0 1.03 1.49

305 (19) 990 (62) 290 (18)

16% 28% 38%

1.0 1.76 (1.3, 2.3)’ 2.34 (1.7, 3.11+

343 (22) 1,238 (78)

19% 30%

1.0 1.62 (1.3, 2.1)+

834 (53) 291 (19) 290 (18) 151 (10)

28% 35% 23% 18%

1.0 1.24 (1.0, 1.5)+ 0.83 (0.7, 1.1) 0.63 (0.4, 0.9)

656 (41) 929 (59)

34% 23%

0.70 (0.6, 0.817

373 (24) 404 (25) 808 (51)

19% 32% 29%

1.0 1.64 (1.3, 2.1)+ 1.52 (1.2, 1.91+

361 (28) 326 (26) 586 (46)

20% 32% 30%

1.0 1.6 (11.2,2.1)’ 1.5 (11.2,1.9v

159 476 79 228 54 218 152 219

(10) (301 15) (14) (3) (14) (10) (14)

14% 22% 25% 31% 31% 34% 34% 35%

0.47 0.72 0.81 1.0 1.01 1.08 1.10 1.11

1,143 (72) 442 (28)

25% 34%

1.0 1.37 (1.2, 1.6)+

1,024 (65) 134 (8) 427 (27)

25% 27% 35%

1.0 1.09 (0.8, 1.5) 1.41 (1.2, 1.7)’

193 1,102 244 46

20% 27% 33% 39%

1.0 1.39 (1.0, 1.9Y 1.67 (1.2, 2.3)t 1.99 (1.3, 3.11+

(0.3, 0.717 (0.7, 1.1) (0.8, 1.31 (1.2, 1.8P

1.0

(0.3, 0.71t (0.6, 0.91’ (0.5, 1.2) (0.7, (0.8, (0.8, (0.9,

1.6) 1.4) 1.5) 1.5)

Cox model probability estimate of prognosis (Z-month survival) 90% or better 65% to 89%

36% to 64% 35% or less

(12) (70) (15) (31

‘Missing data: complete information was not available on all patients. The number of patients with missing data for each variable is as follows: site (0); age (0); race (4); marital status (19); gender (0); insurance class (0); income (312); diagnosis 10);ADL dependencies(6); cancer (0); Coxmodel probability estimate of survival (0). t95% confidence Interval (Cl) does not include 1, indicating that the group is significantly different than the reference group. CPR = cardiopulmonary resuscitation; CHF = congestive heart failure; MOSF= muitiorgan system failure; ARF = acute respiratory failure; COPD = chronic obstructive pulmonary disease; ADL = activities of daily living.

.32

February 1996 The American Journal of Medicine* Volume 100

TABLE III Bivariable Analysis of the Relation Between Cardiopulmonary Resuscitation (CPR) Preferences, Perceptions of Prognosis, and Quality of Life (n = l,Ei85) Number of ReMve Risk lntervlewed Patients Patients Not for Not Wanting Factor Wti Factor (%)’ Wanting CPR (%) CPR (95% Cl) Patient estimates of 2month survival 90% or better 955 (60) About 75% 119 (8) About 50-50 128 (8) About 25% or less 42 (3) Response “don’t know’ 283 (181 Patient view of quality of life Excellent 105 (7) Very good 209 (14) Good 338 (22) Fair 405 (27) Poor 456 (30)

22 29 44 48 37

0.73 (0.5, 1.0) 1.0 1.49 (1.1, 2.1) 1.62 (1.1, 2.5) 1.25 (0.9, 1.7)

25 22 27 26 32

1.0 0.89 10.6, 1.4) 1.09 10.7, 1.6) 1.07 1:0.7,1.5) 1.28 1:0.9,1.8)

‘Missing data: complete information was not available on all patients. The number of patients with missing data for each variable is as follows: survival estimate (58); quality of life (72). Thus percentages do not total 100%. Cl = confidence interval.

Perceptions of Prognosis and Quality of Life Patients with a higher expectation for survival at 2 months were more likely to want CPR (Table III). Of those with an expected 2-month survival of 90% or more, 22%did not want CPR, whereas for those who predicted a 2-month survival of 25%or less, 48%didn’t want CPR. There was no clear relationship between perceived quality of life on the &point scale and resuscitation preference.

Multivariable Correlates of Not Wanting CPR In a multivariable analysis, seven factors were associated independently (P ~0.05) with patients’ CPR preferences while controlling for all other factors included in the analysis (Table IV’). Patients hospitalized at one site were more likely to want CPR, as were patients with heart failure or chronic liver disease. Factors associated with not wanting CPR included: being hospitalized at one of the other study hospitals; being older; being female; being more functionally impaired; and patient estimate of a worse prognosis. Race, although important in the bivariable analysis, was not significant (P = 0.06) after adjusting for being self-pay or having Medicaid coverage. Neither the APS, the Cox-model probability estimate of the patient’s survival, or the number of comorbidities had any independent effect on patients’ preferences for CPR. The area under the ROC curve (c statistic) was 0.72. In a separate analysis, patients who responded “don’t know” to the CPR preference question were excluded, and results were similar.

TABLE IV Multivariable Analysis of Factors Associated With Patient Not Wanting Cardiopulmonary Resuscitation Adjusted Odds Ratio Factor 95% Cl 0.46 Site 1 (0.29-0.72)’ 0.77 4 (0.54-1.10) 2 0.81 (0.56-1.17) 1.67 (1.13-2.47)’ 5 1.00 3 Diagnosis Cirrhosis 0.41 (0.23-0.73)’ 0.65 CHF (0.43-0.97)’ 0.84 MOSF and sepsis (0.45-l ,591 0.85 MOSF and malignancy (0.40-l ,781 0.91 (0.58-l ,431 COPD 1.oo ARF 1.00 Lung cancer (0.61-1.65) 1.21 (0.72- 2.03) Colon cancer Age 1.19 (1.07-l .32)’ Per lo-year increase Gender 1.79 (1.40-2.30)’ Female 1.00 Male Functional status 1.10 (1.03-1.181’ Per increase by 1 in ADL dependencies Patient 2-month prognostic estimate 0.63 (0.40498)’ 90% or better 1.00 About 75% 1.77 (1.04-3.01)’ 50% or less 1.10 (0.67-l ,801 Don’t know ‘95% confidence interval (Cl) does not include 1, indicating that the group is significantly different than the reference group. CHF = congestive heart failure; MOSF= muitiorgan systerr failure; COPD = chronic obstructive pulmonary disease; ARF = acute respiratory failure; ADL = activities of daily living.

February 1996 The American Journal of Medicine” Volume 100

133

PATlENT PREFERENCES FOR CARDIOPULMONARY RESUSClTATION/PHILLIPS ET AL

TABLE V Relation Between Cardiopulmonary Resuscitation (CPR) Preferences and Preferences for Other Treatments Number of Relative Risk Interviewed Patients Percent of Patients for not Wantine With Factor (%)* not Wanting CPR CPR (95% Cl)-

Factor Patient prefers Life-extending treatments 537 (39) Comfort measures 845 (61) Willingness to live in pain Would tolerate condition 922 (64) Would rather die 511 (36) Willingness to be attached to breathing machine Would tolerate condition 614 (42) Would rather die 833 (58) Willingness to be unconscious or in coma Would tolerate condition 395 (27) Would rather die 1,065 (73)

9% 40%

1.0 4.57 (3.4, 6.1)

20% 45%

1.0 2.22 (1.9, 2.6)

15% 40%

1.0 2.69 (2.2, 3.3)

15% 33%

2.17 (1.7, 2.8)

1.0

‘Missing data: complete information was not available for all patients. The number of patients with missing data for each variable is as follows: preference for life extension versus comfort (203); willingness to live with adverse outcomes, pain all the time (1521; attached to breathing machines (138); unconscious or in coma (125). Cl = confidence interval.

did not want CPR were more likely to have discussed their CPR preferences. Of patients who did not want CPR, 48% reported they had discussed it with their doctors compared with 23% of patients who wanted CPR (odds ratio [95% confidence interval] = 2.0 [1.8, 2.41). During hospitalization, 32% of the patients who stated that they preferred not to receive CPR were assigned a DNR order. Patients’ preferences for CPR were available for 96 of the 114 interviewed patients who died during their initial hospitalization (Table VI). Of patients who did not want CPR (n = 34), 26 (76%) had a DNR order and did not receive CPR ‘For 2 of the 9 patients, notes in the medical record documented the whereas for the 8 patients without a DNR order, CPR decision not to provide CPR, but no DNR order was written. For 2 of the 3 patients, notes in the medical record documented the was provided to 5 patients. Of those patients who decision not to provide CPR, but no DNR order was written. wanted CPR initially (n = 62), 39 (63Oh)hald a DNR order written during their hospitalization and did not receive CPR. For the remaining 23 patients who iniCPR Preferences and Preferences tially wanted CPR and did not have DNR orders, 14 for Other Treatments had CPR attempted prior to death. Of the 12 patients In response to the forced-choice question about preference for life-extending treatments, 91% of pa- without DNR ordeF who did not receive CPR, 4 had tients who favored life-extending treatments over chart documentation of the decision not to provide comfort wanted CPR (Table V). However, 9% of pa- CPR. For nearly all of the 39 patients who reported wanting CPR and subsequently had DNR orders writtients would not want CPR even when they favored life-extending treatments. Among patients who pre- ten, there was chart documentation of discussion ferred comfort measures over extending life, 60% among the patient, family, and physician of the decision not to provide CPR. The disparity between pawould still want CPR. Among patients who would rather die than be forced to live permanently with sit- tients’ preferences and care decisions was due largely uations such as being in pain, being attached to ma- to changes in patients’ preferences between the inichines for breathing, or being unconscious or in a tial interview and the date the DNR order was written (mean time between patient interview and DNR coma, more than 50??still wanted CPR. order was 9.5 days). Patients who did not want CPR, compared with Process and Outcomes of Care those who preferred CPR, were more likely to die Overall, 29% of patients reported discussing their within 6 months of study entry (45% versus 27%, CPR preferences with their physician. Patients who TABLE VI Relation Between Do-Not-Resuscitate (DNR) Orders and Attempts at Cardiopulmonary Resuscitation (CPR) Wiiin 6 Days of Death, Stratified by Patients’ Preferences for CPR at Initial Interview (n = 96) DNR Order No DNR Order 1 qatient prefers CPR (n = 62) 0 14 CPR attempted 39 . 9’ No CPR Patient prefers DNR (n = 341 5 0 CPR attempted 26 3t No CPR

134

February 1996 The American Journal of Medicine@ Volume 100

PATIENT PREFERENCES FOR CARDIOPULMONARY RESUSCITATION/PHILLIPS ET AL

P
DISCUSSION

TABLE VII Patient Outcomes Associated Wiih Not Wanting Cardiopulmonary Resuscitation’ Odds Ratio P Value Mortality In-hospital death 1.19 NS Died within 2 months of study entry 1.77
I Average TISS

Mean differenoet -1.27 points co.01

‘Adjustedfor factors tested in multivariable model of CPR preferences (hospital sate, diagnosis, gender, age, having no health insurance or having Medicaid, race, objective estimate of prognosis, patient’s estimate of prognosis, and patients ability to perform activities of daily living). ‘Mean difference calculated as value for patients who wanted CPR sub tracted from value for those not wanting CPR. TISS = therapeutic intensity scoring system.

Even among seriously ill hospitalized patients, a minority of patients (29%) reported discussing their CPR postdischarge mortality rates across hospitals.32 preferences with their physician. Among patients who When entering into discussions of patients’ resusdid not want CPR, only 48% said they had discussed citation preferences, physicians should review the pathis preference with their physician. Preferences to tients’ diagnosis and prognosis. It should be noted forgo CPR were associated with study enrollment that patients’ prognostic estimates were :important site, having a diagnosis other than heart failure or cir- correlates of preference even after adjusting for an objective estimate of survival. This finding suggests rhosis, being older, being female, being more functionally impaired, and having an expectation for a that it is the patient’s perception rather than an obworse prognosis. Although CPR preferences were not jective prognostic estimate that determines the paassociated with mortality at the index hospitalization, tient’s preference. Since patients may be overly optinot wanting CPR was associated with a higher mor- mistic about their prognosis,33 providing more realistic estimates to patients might increase the protality at 2 and 6 months and a less resource-intensive portion of patients from whom an authentic preferindex hospitalization. The impact of hospital site as an independent cor- ence not to receive CPR is elicited. Previous studies have found that older patients and relate of patients’ preferences is particularly notable. This finding is consistent with previous work that women were less willing to undergo CPR, although demonstrated substantial variation in the rate of DNR other studies of DNR orders have not shown such a relationship.” Although we show variation in resusciorders across different intensive care units21 Markers of disease severity such as APS, and patients’ charac- tation preferences by age and gender, it should be emteristics such as age and gender, do not explain this phasized that preferences cannot be predicted from variation. Possible explanations for this variation are patient characteristics, but are dependent on a patient’s perceptions of prognosis and functional status. as follows: First, to the extent that patients’ preferResuscitation preferences cannot be inferred, but ences are influenced by their caretakers, it is possible must be elicited by the physician discussing the issue that unmeasured physician or nurse attitudes toward CPR may explain these differences. Second, there may with the patient.28 Further work is needed to elucidate why there are differences in resuscitation preferences be unmeasured differences in patients’ characteristics that account for the observed variation, either because by age and gender, even after accounting for funcof differences in the populations served by the study tional status and perceived prognosis. Does this rephospitals or differences in factors leading patients to resent differences in physician-patient interaction by gender and age or is this variation more fundamental? be admitted to study hospitals. For example, a hospiDifferences in attitudes and perceptions about CPR tal that offers aggressive treatments such as heart or liver transplants may attract a larger percentage of pa- according to physician specialty may explain variatients seeking aggressive care, while a hospital that of- tion in DNR order rates across diagnoses. Previous fers hospice services may be more likely to attract pa- research has demonstrated that decisions :not to retients seeking terminal care, Since mortality after suscitate may vary by diagnosis even when prognohospital discharge was affected by patients’ CPR pref- sis may be similar. g~loFor example, Wachter et al9 erences, the observed variation in resuscitation pref- studied DNR orders among patients with different erences across sites may help explain variation in diseases. Their disease groups included AIDS, unreFebruary 1996 The American Journal of Medicine@ VOlUme100

135

sectable non-small-cell lung cancer, cirrhosis, and congestive heart failure; the survival for these patients was similar to that of patients in SUPPORT. Overall, 28% of their patients had DNR orders written. However, the rate of DNR orders varied significantly by diagnosis (ranging from 5% to 52%) and, as shown in our study, patients with congestive heart failure and cirrhosis were less likely to receive DNR orders than were patients with malignancies. These results suggest that DNR decisions are related to patients’ preferences, their perceptions of prognosis, and diagnosis-specific factors. These latter factors may be related to differences in the burden of suffering imposed by illness, or differences in physicians’ attitudes across specialties. Few studies have reported CPR preferences for hospitalized seriously ill adults.” Most studies have emphasized preferences given hypothetical scenarios such as functional incapacity, stroke, or terminal cancer. In studies that asked questions about preferences in their current situation, 59% to 89% of outpatients expressed preferences in favor of attempting CPR.3i’:36 Fewer studies report patients’ preferences at the time of hospitalization. In one study reporting preferences during hospitalization (n = 200), Frankl et al” found that 99% of patients desired CPR if they could be restored to their usual level of health, but preferences were strongly influenced by the perceived outcome of CPR. Patients’ preferences for CPR given their current health and their own understanding of their prognosis were not reported. The impact of quality of life on patients’ preferences for CPR has been investigated previously. In a small study of 105 elderly outpatients, Uhlmann and Pearlman, like us, found that patients’ report of their quality of life did not influence their desire for life-sustaining treatments. Our data have policy implications for plans to reduce the costs of aggressive care by increasing patient participation in health care decisions. Despite the observation that patients interviewed for this study were seriously ill, had a 33% chance of dying within 6 months, and the majority wanted comfort care, only 28% did not want to receive resuscitation in the case of a cardiopulmonary arrest. It will be important to evaluate what patients expect to gain from CPR and how they believe this therapy might further their goals. Whether patients’ preferences would change if their prognostic estimate were reoriented38 and why there are differences in preferences by diagnosis must be understood in order to design interventions to integrate the elicitation of patient preferences into medical decision making for seriously ill inpatients. Our analysis is limited in several ways. First, the CPR decision itself is complex, and the relatively straightforward question we asked may have oversimplified the issue. As indicated by the large pro136

February

portion of patients who stated that they preferred a comfort-oriented medical care plan but aIso would undergo CPR, the exact meaning a patie:nt ascribes to an aggressiveness-of-care decision must be carefully evaluated by their physician. In addition, we did not provide patients with prognostic information that may have changed their resuscitation preference.38 Furthermore, we did not provide information on outcomes of resuscitation or ask patients about their expectations regarding recovery from CPR. Most importantly, it should be noted that the interviewed patients represented a subset of the patients enrolled in SUPPORT, namely those who were capable of participating in an interview. These patients h.ad a better prognosis overall and a lower severity of illness than patients who could not be interviewed. The overall proportion of patients who wished to forgo CPR may have been substantially higher if information on the preferences of noninterviewable SUPPORT patients were known. Also, since we examined patients’ preferences at one point in time, our analysis d.oes not account for changes in patients’ preferences during the course of their illness. Finally, although we used a standardized interview with centralized interviewer training and evaluations, interviewer bias could have contributed to the site-to-site variation w’e observed in patients’ preferences. We conclude that a number of factors are associated with patients’ preferences and their willingness to undergo CPR. Further research is required to identify reasons CPR preferences are influenced by such factors as patient gender, hospital site, and diagnosis, and to determine if reorienting patients’ prognostic estimates leads to different preferences for care and treatment decisions. Comparative studies of hospital mortahty and patient outcomes must take into account the goals of treatment, as expressed directly by the patient or as reflected by resuscitation decisions. Further research is required on factors associated with changes in patients’ preferences for CPR and the relation between patients’ preferences for CPR and do-not-resuscitate orders. Most importantly, patients’ preferences are not easily predicted and often are not discussed with their physicians. Interventions are needed to increase and to evaluate the impact of such discussions.

ACKNOWLEDGMENT The authors thank Marilyn Bergner, PhD (deceased), Lee Goldman, MD, MPH, and two anonymous reviewers who made thoughtful conhibutions to this manuscript; Jane Soukup, MS, and E. Francis Cook, ScD, who assisted in the analysis; Carmen Alicea for preparation of the manuscript; and the SUPPORT project staff and patients who contributed generously of their trme and energy.

REFERENCES 1. Wanzer SH, Adelstin SJ, Crawford RE, et al. The physicians’ responsibility toward hopelessly ill patients. NEJM. 1989;320:844-849, 2. Perkins HS. Ethics at the end of life: practical principles for making resuscitation decisions. J GenIntern Med. 1986;1:17&176.

1996 The American Journal of Medicine@ Volume 100

3. Bedell SE, Delbanco TL. Cook EF, Epstein FH. Survival after cardtopulmonary resuscltatton in the hospital. NEJM. 1983;309:569-576. 4. Longstreth Wl Jr., Cobb LA, Fahrenbruch CE, Copass MK. Does age affect outcomes of out-of-hospital cardiopulmonary resuscltatlon? JAMA.

1990;31:2109-2110. 5. Kyff J, Purl VK, Raheja R, Ireland T. Cardiopulmonary resuscitation in hosprtalized patients. continuing problems of deciston-making. Crit Care Med.

1987;15:41-43. 6. Roberts D, Landolfo K, Light RB, Dobson K. Early predictors of mortality for hospitalized patients suffering cardiopulmonary arrest. Chest. 1990;97:

413-419.

21. Zimmerman JE, Knaus WA, Sharpe SM, et al. The use and lmpllcations of do not resuscitate orders In intensive care units. JAMA. 1986;25!?351-356. 22. Bedell SE, Pelle D, Maher PL, Cleary PD. Denot-resuscitate orders for criiically ill patients In the hospital JAMA. 1986;256:233-237. 23. Jonsson PV, McNamee M, Campion EW. The “do not resuscitate” order: a profile of its changtng use. Arch Intern Med. 1988;148:2373-2375. 24. Youngner SJ, Lewandowskl W, McClish DK, et al. “Do not resuscitate” orders: incidence and implicatlans in a medical intensive care unit. JAMA.

1985;253:54-57. 25. Levy MR, Lambe ME, Shear CL. Do-not-resuscitate orders in a county hospital. West J Med. 1984;140:11 l-l 13. 26. Gleeson K, Wise S. The d@notfesuscitate order: still too little too late. Arch intern Med. 1990;150:1057-1060. 27. Uhlmann RF, McDonald WJ, lnu~TS. Epidemiology of nocod,e orders n an academtc hospital. West J Med. 1984;140:114-116. 28. Garrett JM, Harris RP, Norburn JK, et al. Life-sustaining treatments during terminal illness: who wants what? J Gen Intern Med. 1993;8:361-368. 29. Walker SH, Duncan 05: Estimation of the probability of an event as a function of several independent variables. Biometrika. 1967;54:1#57-179. 30. Harrell FE, Califf RM, Pryor DP, et al. Evaluating the yield of medical tests. JAMA. 1982;247:2543-2546. 31. Zimmerman JE, Shortell SM. Knaus WA, et al. Value and cost of teaching hospitals: a prospective, multtcenter, inception cohort study. Crit Care Med.

7. Peterson MW, Gelst LJ, Schwartz DA, et al. Outcome after cardiopulmonary resuscitation In a medical lntenslve care unit. Chest 1991:100:168-174. 8. Lo B, SalkaG, StrullW, et al. “Donot resuscitate”decisions.A prospecttve study at three teaching hospitals. Arch Intern Med. 1985;145:1115-1117. 9. Wachter RM, Lute JM, Hearst N, Lo B. Decisions about resuscitation: inequities among patients with different diseases but similar prognoses. Ann Intern Med. 1989;15;111:525-532. 10. Solman CJ, Gregory JJ, Dunn D, Ripley B. Evaluation of do not resuscitate orders at a community hospital. Arch Intern Med. 1989;149:1851-1856. 11. Frank1 D, Oye RK. Bellamy PE. Attitudes of hospitalized patients toward life support: a survey of 200 medical inpabents. Am J Med. 1989;86:645-648. 12. Lawrence VA, Clark GM. Cancer and resuscitabon. Does the dtagnosis 1993;21:1432-1442. affect the decision? Arch Intern Med. 1987;147:1637-1640. 32. Kahn KL, Brook RH, Draper E, et al. Interpreting hospital rnortality data. 13. Hanson LC, Danis M. Use of Ilfe-sustaining care for the elderly. J Am How can we proceed? JAMA. 1988;260:3625-3628. Gerratr Sot. 1991;39:772-777. 33. Arkes HR, Dawson NV, Speroff T, et al. The covariance decomposition of 14. Lynn J, Knaus WA. Background for SUPPORT. J Clan Epldemrol. 1990;43; the probability score and its use in evaiuating prognostic estimates. Med Decks ls-4S. Makmg. 1995;15:12~131. 15. Philllps RS. Goldman L, Eergner M. Patient charactenstxs in SUPPORT: 34. Wagner A. Cardiopulmonary resuscitation in the aged. A prospective activity status and cognitive function. J C/in Epidemiol. 1990;43:33S-365. survey. NEJM. 1984;310:1129-1130. 16. Knaus WA, Wagner DP, Draper EA, et al. The APACHE Ill prognostic 35. Starr TJ, Pearlman RA, Uhlmann RF. Quality of life and resuscltatlon system. Risk predictlon of hospital mortality for critically ill hospitalized adults. decisions in elderly patients. J Gen Intern Med. 1986;1:373-379. Chest. 1991;100:1018-1036. 36. Shmerling RH, Bedell SE, Lilienfeld A, Delbanco TL. Discussing 17. Keene AR, Cullen DJ. Therapeutic intervention scoring system: update cardiopulmonary resuscitation: a study of elderly outpatients. J Glen Intern Med. 1983. Crit Care Med. 1983;11:1-3.

18. Knaus WA, Draper EA, Wagner DP, Zimmerman JE. APACHE II: a severity of disease classification system. Crit Care Med. 1985;13:818-829. 19. Knaus WA, Harrell FE, Lynn J, et al, for the SUPPORT Investigators. The SUPPORT prognosbc model: prediction of survival for seriously III haspitallzed adults. Ann Intern Med. 1995;122:191-203. 20. Landefeld CS, Phillips RS, Bergner M. Patient charactenstlcs in SUPPORT: functional status. J C/in Epidemiol. 1990;43:37WOS.

1988;3:317-321. 37. Uhlmann RF, Pearlman RA. Perceived quality of life and preferences for life sustaining treatment in older adults. Arch intern Med. 1991;151:495-497. 38. Schonwetter RS, Walker RM. Kramer DR, Robinson BE. Resuscltatlon decision making In the elderly: the value of outcome data. J Gen fntern Med.

1993;8:295-300.

February

1996 The American Journal of Medicine”

Volume 100

137