Prognostic Disclosure to Patients with Cancer near the End of Life

Forty years ago, physicians did not inform most patients with cancer of their diagnoses (1, 2). This practice of nondisclosure is now generally considered out of date, primarily because it may represent physician paternalism that compromises patient autonomy. Indeed, almost all patients with cancer are now informed of their diagnoses (3). Nevertheless, it is not clear how many understand the survival implications, that is, the associated prognosis. Because survival estimates often strongly affect decisions about cancer treatment, especially at the end of life, patients need and often rightly request prognoses when making such decisions (4-7). Studies that compare physicians' prognostic estimates with those of patients often show a substantial discrepancy between the two. In a study of 100 patients with cancer who were undergoing treatment, Mackillop and colleagues (8) found that one third of those with metastatic cancer thought that they had local or regional disease and were being treated for cure. Similarly, Eidinger and Schapira (9) studied 190 patients being treated for incurable metastatic cancer and found that approximately one third thought that the treatment would cure them. Weeks and colleagues (10), in their analysis of 917 patients with metastatic colon cancer or advanced nonsmall-cell lung cancer in the Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments, found that patients who had optimistic misperceptions of their prognosis often requested medical therapies that most physicians would consider futile (10). Their study suggests that in patients with terminal cancer, optimistic prognostic estimates may lead to choices of invasive but ineffective medical therapies rather than perhaps more appropriate supportive care. Previous studies do not clarify the reason for the asymmetry between comparatively optimistic survival estimates made by patients with cancer and the estimates made by their physicians. Do patients misinterpret or deny the poor prognoses that physicians give them? Do physicians think one thing about patients' prognoses but tell the patients something different? Although several factors may be operating simultaneously, some research suggests that at least part of the discrepancy may be due to the generally optimistic prognostic estimates that physicians give their patients or to omission of prognostic discussions altogether (11). Our principal objective was to evaluate how often physicians favor communicating frank survival estimates to patients with terminal cancer who request them. An additional objective was to examine how specific patient and physician characteristics are associated with physicians' preferences (that is, their stated intentions) for prognostic communication. To meet these objectives, we interviewed physicians from several specialties who referred their own patients with cancer to hospice terminal care. We asked them to provide us with their most accurate estimate of how long their patients had to live (formulated prognosis). We then asked them what they would tell their patients if the patients insisted on obtaining an estimate of how long they had to live (communicated prognosis). We compared these two survival estimates and sought to explain discrepancies by evaluating several patient and physician variables in which we had a substantive interest. Methods Study Sample We assembled a cohort of all patients with cancer admitted to five outpatient hospice programs in Chicago, Illinois, during 130 consecutive days in winter and spring 1996. We approached all hospices in the Chicago area that admitted more than 200 patients per year, at least 70% from within the Chicago city limits. Six hospices met these criteria, and five agreed to participate; we estimate that most hospice patients in Chicago were captured in our sample. Our research was approved by the institutional review board at each participating hospice and was conducted in accordance with the regulations of these boards. Participating hospices usually notified us about patients on the day of admission. We contacted referring physicians promptly to administer a 4-minute telephone survey about patient prognosis and to collect other information. A total of 767 patients were referred by 502 physicians during the study period and consented to the study. The five hospices contributed 13%, 14%, 17%, 22%, and 34% of the sample, respectively. Of the 767 patients, 325 did not meet the entry criteria: Two hundred eighty-five had a noncancer diagnosis (an expected percentage based on national data) (12, 13), and 40 had physicians who were not appropriate participants (that is, they had already responded to several previous cases in the study). Thirty patients died before we were notified of admission. Because they died within a few hours and their physicians' predictions of survival would be meaningless, we did not include them in our cohort. For the remaining 412 eligible patients, we reached 38 physicians (9.2%) after the patient had died (and therefore could not get a meaningful prognostic estimate); we reached 8 physicians (1.9%) before the patient died, but the physician declined to participate; and we failed to reach 40 physicians (9.7%). However, for these 86 patients, we obtained basic physician and patient information and time of death. We therefore successfully completed surveys with physicians who cared for 326 of the 412 eligible patients (a completion rate of 79.1%). Our analytic sample consists of these 326 patients, who were referred by 258 physicians. When we compared the 326 patients with the 86 excluded patients, we did not find important differences in age, sex, ethnicity, cancer type, or disease duration or in their physicians' sex, practice experience, or specialty. While most participating physicians (83% [214 of 258]) referred only 1 patient, a small number referred more than 1 (range, 2 to 6 patients). The average number of patients per physician was 1.26. Variables and Data Sources We acquired information about patient age, sex, ethnicity, religion, marital status, cancer diagnosis, and comorbid conditions from the hospices. From the physician telephone survey, we obtained patients' Eastern Cooperative Oncology Group (ECOG) performance status scores (a measure of debilitation that ranges from 0 to 4) (14) and duration of illness. We obtained patients' dates of death from publicly available death registries or from the hospices. As of 30 June 1999, dates of death were known for 96% of the cohort (313 of 326). From the physician telephone survey, we also determined physicians' experience with similar patients and how well they knew the study patients (that is, the duration, recency, and frequency of their contact). From publicly available records, we determined physicians' specialty, years in practice, and board certification. Our key questions involved 1) an estimate of the patient's prognosis [by asking the physicians to provide your best estimate of how long you think this patient has to live] and 2) a comparable statement about what the physician would tell the patient if the patient or family insisted on receiving a specific estimate of survival. We refer to the first prognosis (the estimates of survival given to us by the physicians) as the formulated prognosis and the second prognosis (the estimates physicians would give to patients) as the communicated prognosis. By design, these two questions were separated by 20 questions that required approximately 2 minutes to answer. Although physicians were not reminded of their formulated prognosis when asked for the communicated prognosis, it was provided if they requested it. Physicians were not asked to explain discrepancies between their formulated and communicated prognoses. We also asked physicians to quantify their confidence in their formulated prognosis as a percentage, from 0% (no confidence) to 100% (complete confidence). The instrument is available from the investigators upon request. Statistical Analysis We created a multinomial disclosure variable capturing the four possible categories of prognostic disclosure that could result from comparison of the formulated and communicated prognoses. The categories were 1) no disclosure [the physicians formulated a prognosis for the investigators but would not communicate any temporally specific prognosis to the patient], 2) frank disclosure [formulated prognosis was the same as communicated prognosis], 3) optimistically discrepant disclosure [formulated prognosis was shorter than communicated prognosis], and 4) pessimistically discrepant disclosure (formulated prognosis was longer than communicated prognosis). To evaluate associations between the multinomial disclosure variable and categorical and continuous variables, we used chi-square tests and analysis of variance, respectively. We used multinomial logistic regression to model the multivariate effect of patient and physician variables on the intended strategy of prognostic disclosure (15). This type of model describes the relative odds, through conditional odds ratios, of being in one category compared with another (the omitted category, which was frank disclosure). Although 83% of physicians referred only one patient to the cohort, we adjusted our regression model to account for clustering of patients within physicians (16). All analyses were performed by using Stata 6.0 (Stata Corp., College Station, Texas). Odds ratios may present difficulties when used to characterize relationships, because they may seem to overstate the relative risk when the frequency of an outcome is high. Therefore, we used a variation of a method described elsewhere (17-19) to transform odds ratios into relative risks for selected key comparisons. These relative risks provide an additional, easier to appreciate characterization of the relationship between predictors of interest and the outcome being examined. Such relative risks depend o