Factors Associated with Do-Not-Resuscitate Orders: Patients' Preferences, Prognoses, and Physicians' Judgments

A Presidential Commission, recent medical guidelines, and various court rulings have supported the conclusion that medical decisions should be based on the informed preferences of individual patients and on considerations of the expected outcomes of the therapies of interest [1-5]. Although these considerations apply to all patients, they are especially relevant for hospitalized patients who have poor short-term prognoses and thus may be more willing to limit the use of aggressive treatments, such as cardiopulmonary resuscitation [6]. Because consent for cardiopulmonary resuscitation is implied unless a specific order is written, do-not-resuscitate (DNR) orders provide a useful model with which to evaluate end-of-life decision making. It is now agreed that DNR orders should be based on patient preferences [7, 8], but studies have shown that patients, even when capable of communication [9], infrequently participate in decisions about resuscitation [10, 11]. One explanation for this is that discussions are often delayed until patients cannot participate [12]. Other studies have found that DNR practices vary according to physicians' clinical specialties [13], particular medical institutions [14], patient diagnoses [14-16], and patient ages [10, 14, 15]. Although most of these studies controlled for variations in patient characteristics, none explicitly evaluated both these characteristics and patient preferences. The Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatment (SUPPORT) was done in two phases in five teaching hospitals. Phase I was a baseline observational study, and phase II was a block-randomized clinical trial (that is, patients were randomly assigned to intervention on the basis of the attending physician's specialty) of an intervention intended to improve medical decision making and outcomes for severely ill hospitalized patients [17, 18]. A primary concern of the SUPPORT investigators was that patients' informed preferences and other influential patient characteristics support decisions about the use of life-sustaining treatment, including cardiopulmonary resuscitation. A fundamental premise of SUPPORT was that preferences should be ascertained early in the course of the medical illness and that the medical record should be updated after it is known that a patient prefers to forego cardiopulmonary resuscitation. In this report, we examine the role that patients' preferences and other factors related to the patient, physician, and institution had in the incidence and timing of DNR orders. Methods Phase I of SUPPORT (1989 to 1991) evaluated the decision-making process and outcomes for 4301 severely ill hospitalized patients. Phase II (1992 to 1994) was a controlled clinical trial of an intervention that gave physicians information on patient prognoses and preferences for end-of-life care. Nurse clinicians were trained to assist with and facilitate communication to the 2652 patients who received the intervention. Another 2152 persons served as controls and received usual medical care. Phase II enrolled a total of 4804 patients [18]. A major hypothesis of this trial was that accurate information and better communication would increase the frequency of DNR orders and decrease the time taken to write an order. The phase II intervention neither increased the frequency nor accelerated the timing of DNR orders [18]. Because no secular trends in the timing of DNR orders were seen during the entire study period (1989 to 1994), we could combine data from phase I and phase II for this analysis. Inclusion Criteria The SUPPORT trial enrolled patients who met entry criteria at five medical centers: Beth Israel Hospital, Boston, Massachusetts; MetroHealth Medical Center, Cleveland, Ohio; Duke University Medical Center, Durham, North Carolina; Marshfield Clinic-St. Joseph's Hospital, Marshfield, Wisconsin; and the University of California Medical Center, Los Angeles, California. To be eligible, patients had to have an anticipated aggregate 6-month mortality rate of approximately 50% and had to have one or more of the following diagnoses: acute respiratory failure, multiple organ-system failure with sepsis or malignant condition treated in an intensive care unit, coma, chronic obstructive lung disease, congestive heart failure, cirrhosis, metastatic colon cancer, or non-small-cell lung cancer (stage III or IV). The entry requirements of the study have been described elsewhere [17]. The study sample included all patients enrolled in SUPPORT (or their surrogates) who answered the question about patient preferences for resuscitation in the first interview. We excluded the following patients: those for whom a DNR order had been written before study entry, those who did not survive or were discharged during the first 48 hours of the study; those admitted to the hospital with a scheduled discharge within 72 hours; those younger than 18 years of age; those with the acquired immunodeficiency syndrome (AIDS); those who did not speak English; those who were admitted to the psychiatric service; those who were pregnant; and those who had sustained an acute burn, head trauma, or other trauma (unless they later developed acute respiratory or multiple organ-system failure). We selected disease categories on the basis of their prevalence among dying patients in the acute care hospital. One of the goals of the SUPPORT study was to develop prognostic models; thus, at the time of the conceptualization and pilot testing of the study (1986 to 1988), we decided that the treatment and prognoses of patients with AIDS were changing too rapidly to allow us to include such patients. We remained in direct contact with surviving patients for 6 months, and we used the National Death Index to follow patients thereafter. Data Collection We interviewed patients and their surrogates (a surrogate was defined as the person who would make decisions if the patient was unable to do so) on the third day after study enrollment (95% of interviews took place between the second and seventh study days). Patients and surrogates were interviewed again during the second study week and at 2 and 6 months. For our analysis, we used only the data obtained from the first interview, during which patients were asked about their socioeconomic status, functional status before hospitalization, self-assessed quality of life, and preferences for cardiopulmonary resuscitation [19]. The question on preferences for cardiopulmonary resuscitation was worded As you probably know, there are a number of things doctors can do to try 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 stops beating? Would you want your doctors to revive you, or would you want your doctors not to try to revive you? Responses were coded as wanted resuscitation, wanted resuscitation but no ventilator, wanted no resuscitation, or did not know. In all analyses, a surrogate interview was substituted when no patient interview was possible. Functional status was measured by using a slightly modified version of the Katz Activities of Daily Living scale [20, 21]. The scale proposed by Katz measured impairment in bathing, dressing, eating, continence, transferring, and toileting; we added a question about impairment in walking. We summed the number of dependencies, which could range from 0 to 7 (the latter indicated dependency in all seven functions). For study days 1, 3, 7, 14, and 25, patients' medical records were abstracted concurrently for physiologic variables that were known to be predictors in the Acute Physiology and Chronic Health Evaluation (APACHE) II and APACHE III [22, 23] prognostic systems (as used in the SUPPORT prognostic model [17]) that predicted survival probability for as long as 180 days after enrollment. We report the acute physiology scores of APACHE III; these scores range from 0 to 299 (a higher number indicates greater acuity of medical illness). In this analysis, we used only the 2-month SUPPORT survival estimate, which was based on data collected on the first study day [17]. After discharge or death, medical records were abstracted for discussions and treatment decisions about cardiopulmonary resuscitation. We considered the attending physician to be the physician of record at the end of the second study day (patients had to survive 48 hours to qualify for the study). On the basis of an interview during which demographic information was collected, we grouped physician specialties into the following categories: oncology, pulmonary or intensive care medicine, cardiology, surgery, and general medicine (which included other medical subspecialties, such as rheumatology and gastroenterology). Each study institution had policies stating that a DNR order had to be discussed and that the attending physician had to sign or co-sign the order. Statistical Analysis We did univariate analyses to determine the unadjusted incidence of DNR orders by patient and institutional characteristics. We did a multivariable analysis using a log-normal regression model that contained the following predictor variables: patient preferences about cardiopulmonary resuscitation; age, sociodemographic factors, and diagnosis; scores on the Katz Activities of Daily Living scale [20, 21] and a 5-point scale that rated the patient's quality of life from excellent to poor; comorbid conditions; the Glasgow coma score [24]; the probability of surviving for 2 months based on the SUPPORT model; and the hospital and the attending physicians' specialty and the relation between the two. The number of days between study enrollment and a DNR order was the dependent variable. To avoid taking the logarithm of zero, we added one half-day to the time the DNR order was written. We used cubic spline functions for all continuous variables to avoid assumptions of linearity b