Zhang B, Nilsson ME, Prigerson HG. Factors Important to Patients' Quality of Life at the End of Life. Arch Intern Med. 2012;172(15):1133-1142. doi:10.1001/archinternmed.2012.2364
Author Affiliations: Center for Psychosocial Epidemiology and Outcomes Research (Ms Zhang, Mr Nilsson, and Dr Prigerson) and Division of Population Sciences, Department of Medical Oncology (Dr Prigerson), Dana-Farber Cancer Institute, and Department of Psychiatry, Brigham and Women's Hospital, Harvard Medical School (Dr Prigerson), Boston, Massachusetts.
Background When curative treatments are no longer options for patients dying of cancer, the focus of care often turns from prolonging life to promoting quality of life (QOL). Few data exist on what predicts better QOL at the end of life (EOL) for advanced cancer patients. The purpose of this study was to determine the factors that most influence QOL at the EOL, thereby identifying promising targets for interventions to promote QOL at the EOL.
Methods Coping With Cancer is a US multisite, prospective, longitudinal cohort study of 396 advanced cancer patients and their informal caregivers who were enrolled from September 1, 2002, through February 28, 2008. Patients were followed up from enrollment to death a median of 4.1 months later. Patient QOL in the last week of life was a primary outcome of Coping With Cancer and the present report.
Results The following set of 9 factors, preceded by a sign indicating the direction of the effect and presented in rank order of importance, explained the most variance in patients' QOL at the EOL: 1 = (−) intensive care unit stays in the final week (explained 4.4% of the variance in QOL at the EOL), 2 = (−) hospital deaths (2.7%), 3 = (−) patient worry at baseline (2.7%), 4 = (+) religious prayer or meditation at baseline (2.5%), 5 = site of cancer care (1.8%), 6 = (−) feeding-tube use in the final week (1.1%), 7 = (+) pastoral care within the hospital or clinic (1.0%), 8 = (−) chemotherapy in the final week (0.8%), and 9 = (+) patient-physician therapeutic alliance at baseline (0.7%). The vast majority of the variance in QOL at the EOL, however, remained unexplained.
Conclusion Advanced cancer patients who avoid hospitalizations and the intensive care unit, who are less worried, who pray or meditate, who are visited by a pastor in the hospital/clinic, and who feel a therapeutic alliance with their physicians have the highest QOL at the EOL.
When curative treatments are no longer options for patients dying of cancer, the focus of care often turns from prolonging life to promoting the quality of life (QOL).1 In 1997, the Institute of Medicine issued a report on improving care at the end of life (EOL).2 The report stated that to ensure better care at the EOL, researchers needed to fill gaps in knowledge about the EOL. One gap has been data on the strongest predictors of higher QOL at the EOL. Data exist on what factors are considered important at the EOL by physicians, patients, and family members3 and the factors that predict the quality of EOL care.4 Few data exist on what predicts better QOL at the EOL for advanced cancer patients.5- 9 To our knowledge, there has not yet been a comprehensive model of the strongest predictors of QOL at the EOL for cancer patients.
Research has identified factors important to higher-quality EOL care, including adequate management of pain and symptoms, effective patient-physician communication and a strong therapeutic alliance, physicians' responsiveness to patients' treatment preferences, and care that enables patients to attain a sense of life completion.3,10- 20 Although these studies note factors that physicians, patients, and caregivers consider important to patient QOL and care, they were not designed to determine prospectively the most influential set of factors that predict QOL at the EOL. By establishing empirically the strongest set of predictors of QOL at the EOL for terminally ill advanced cancer patients, we can guide physicians, patients, and family members in focusing on what matters most for ensuring a high QOL for dying cancer patients.
Coping With Cancer is a prospective, multi-institutional study of advanced cancer patients and their caregivers. Coping With Cancer was designed to examine the effect of mental and physical health and health service use, patient-physician relationships, patient and caregiver coping, social support, spirituality, and other relevant psychosocial factors on 2 primary patient outcomes: the care patients receive at the EOL and their QOL at the EOL. Previous Coping With Cancer reports have examined QOL at the EOL as it relates to the intensity of care provided and family dynamics. For example, we have shown that higher QOL at the EOL is associated with longer hospice stays6 and lower QOL at the EOL is associated with more life-prolonging care in the last week of life,6 having a dependent child in the home,7 and dying in a hospital, particularly in the intensive care unit (ICU).8 Patients' peaceful awareness of their terminal illness21,22 and pastoral care visits in the hospital23 have also been shown to relate positively to QOL at the EOL. Nevertheless, to our knowledge, no study has simultaneously examined a wide variety of the advanced cancer patient's experience, from medical care received to social and spiritual support, to determine the set of predictors that best accounts for QOL at the EOL.
The aim of this study was to derive parsimonious models of the set of factors that have the greatest influence on QOL at the EOL. On the basis of our conceptual model of determinants of EOL outcomes,24 we posited that in addition to the negative effects of intensive life-prolonging care, modifiable psychosocial factors would be of paramount importance. Specifically, we hypothesized that the therapeutic alliance between patients and their physicians, patients' and caregivers' mental health, and support of patients' spiritual needs would be the most significant modifiable contributors to higher QOL at the EOL.
Patients were recruited from September 1, 2002, through February 28, 2008, as part of the federally funded Coping With Cancer study. Participating sites included Yale Cancer Center (New Haven, Connecticut), Veterans Affairs Connecticut Healthcare Systems Comprehensive Cancer Clinics (West Haven), the Parkland Hospital and Simmons Comprehensive Cancer Center (Dallas, Texas), Massachusetts General Hospital and Dana-Farber Cancer Institute (Boston), and New Hampshire Oncology-Hematology (Hooksett). Trained interviewers assessed patients and caregivers at baseline, and physicians and caregivers completed the postmortem evaluations. All study protocol and contact documents were approved by the human subjects committee at each participating institution.
Eligibility criteria included (1) the presence of distant metastases, disease refractory to first-line chemotherapy, and an oncologist estimate of life expectancy less than 6 months; (2) age at least 20 years; (3) identified unpaid, informal caregiver; and (4) clinic staff and interviewer assessment that the patient had adequate stamina. Patient-caregiver dyads in which either person met criteria for significant cognitive impairment25 or did not speak either English or Spanish were excluded. Potentially eligible patients were identified from medical records, and their eligibility was confirmed by their physicians. Trained research staff approached each identified patient to offer participation in the study. Once the patient's written informed consent was obtained, medical records and physicians were consulted to confirm eligibility.
Of the 1015 patients approached for participation and confirmed eligible, 289 (28.5%) declined participation. Reasons for nonparticipation included “not interested” (n = 120), “caregiver refuses” (n = 37), and “too upset” (n = 20). Nonparticipants reported significantly more distress than participants on a scale ranging from 1 (minimal/nonexistent) to 5 (distraught) (mean score, 2.72 vs 2.34; P < .001). Latinos were more likely to participate than were other ethnic groups (12.5% vs 5.6%; P = .002). Nonparticipants did not differ significantly from participants in sex, age, or educational level. Of the 726 patients who completed the baseline survey, 414 patients died at the time of data analysis and had postmortem assessments. This cohort did not differ significantly (P < .05) by cancer type, psychological distress, or rates of psychiatric disorders from the study participants at large. However, the deceased cohort had worse baseline QOL, symptom burden, and performance status, as would be expected in patients closer to death.
Baseline interviews were conducted in English or Spanish and took approximately 45 minutes to complete. Patients and caregivers received $25 as compensation for completing the interview.
In the baseline interview, patients and caregivers reported their sociodemographic characteristics, including age, sex, race/ethnicity, family structure, religious faith, educational level (years of schooling), yearly family income (≥$31 000 vs <$31 000), and health insurance coverage. Diagnostic information from the patient's medical chart and clinic was recorded. Self-efficacy,26 coping styles,27,28 religious coping,29,30 religiousness/spirituality,31 and preferences regarding EOL care32 were assessed in patients and caregivers. Patients were asked whether they had completed a do-not-resuscitate order and whether they had discussed their EOL care preferences with their physician. Patients were asked about pastoral care visits in the clinic or hospital23 and their use of mental health services.33 Structured Clinical Interview for the DSM-IV Axis I modules34 were administered by trained interviewers to diagnose current major depressive disorder, generalized anxiety disorder, posttraumatic stress disorder, and panic disorder among patients and caregivers. The Structured Clinical Interview for the DSM-IV has proven reliability and validity.35 Patients completed validated assessments of physician-patient relationships.20 Therapeutic alliance was coded equal to 1 when the patient reported being seen by the physician as a whole person, being treated with respect, respecting and trusting the physician, and feeling comfortable asking the physician questions about health care.6,20 Caregivers completed established measures of social support.36 Patients' performance status and comorbid medical conditions were assessed with the Karnofsky scale37 and the Charlson comorbidity index.38 The McGill Quality of Life Questionnaire's physical and psychological functioning (eg, how nervous or worried the patient felt in the last 2 days where 0 indicates not at all and 10, extremely), symptom burden, and social support subscales were administered to the patient (coded so higher scores reflected better QOL).39 Patients' peacefulness was assessed from an item from the National Institute on Aging/Fetzer Multidimensional Measure of Religiousness/Spirituality.31 Patients were asked to describe their current health status; response options were “relatively healthy,” “relatively healthy but terminally ill,” “seriously but not terminally ill,” and “seriously and terminally ill.” Patients who described themselves as “terminally ill” were coded as acknowledging their terminal illness.
Health care received in the last week of life was obtained in the postmortem assessment completed by the patient's formal (49.0%) or informal (51.0%) caregiver 2 to 3 weeks after the death. These retrospective assessments recorded the location of the patient's death, the types of care received in the last week of life, the patient's QOL at the EOL, whether the patient was enrolled in inpatient or outpatient hospice, and the length of hospice enrollment. The postmortem assessment contained the following questions regarding QOL at the EOL: “Just prior to the death of the patient (his/her last week, or when you last saw the patient), how would you rate his/her level of psychological distress?” (0-10, with 0 indicating none and 10, extremely upset), “Just prior to the death of the patient (his/her last week, or when you last saw the patient), how would you rate his/her level of physical distress?” (0-10, with 0 indicating none and 10, extremely distressed), and “How would you rate the patient's overall quality of life in the last week of life/death?” (0-10, with 0 indicating worst possible and 10, the best possible). The sum of the 3 questions was our primary outcome measure. At baseline, caregivers completed the McGill QOL measure for the patient; this score was significantly (P < .001) associated with the patient's self-reported McGill QOL scores, suggesting caregivers were capable of evaluating the QOL of the patient for whom they cared.
Random-effects modeling40 was used to examine the univariate and multivariate associations between the potential predictors and QOL at the EOL, treating recruitment site as a random effect. Univariate analyses determined whether patients' QOL in the last week differed significantly by patient and caregiver background characteristics and the hypothesized set of predictors. Variables significant at P < .20 in the univariate analyses were entered into the multivariate random-effects models.
Cross-validation (CV)41 provides a way to measure the predictive performance of a statistical model. One way to measure the predictive ability of a model is to test it on a set of data not used in the estimation. The data used to test for the model's predictive ability are called the test sets, and the data used for model estimation are called the training sets. The predictive accuracy of a model can be measured by a CV statistic (eg, mean squared error [MSE]) for the test set. Minimizing the CV statistic is a recommended41 method of model selection. Based on the sample size (n = 396), 9-fold CV model selection was used to determine the best model predicting QOL at the EOL. The study sample was randomly partitioned into 9 subsamples, 8 of them used as the training set and the other 1 as the test set. The process was repeated 9 times and the 9 results were then averaged to produce a single estimate, the average MSE. The advantage of this method is that all observations are used for training and validation, and each observation is used for validation exactly once.
In each training set, backward model selection was used to generate the best model fitting the training data set, and then the 9 best models were compared to select the final model with the lowest average MSE of the test set. We used SAS, version 9.2 (SAS Institute, Inc), as the statistical software for the analyses.
Characteristics of the 396 patients who enrolled with no missing site information, died, and had their postmortem data collected revealed that patients were predominantly white (65.0%), Christian (71.3%), and insured (60.8%), and almost half had an educational level of high school (52.4%). Their mean (SD) age was 58.7 (12.5) years. Patients survived a median of 125 days after baseline. Patients closer to death and younger patients had worse QOL at the EOL. Caregivers' better overall health was associated with patients' better QOL at the EOL. Informal caregivers (family) rated the QOL of patients marginally significantly worse than did formal caregivers (professional/clinical staff) (Table 1).
In the analyses of our conceptual model's potential predictors of QOL at the EOL using random-effects models (Table 2), patients with major depressive disorder, posttraumatic stress disorder, or panic disorder and those who were worried at baseline had significantly worse QOL at the EOL, whereas those with a sense of inner peacefulness at baseline had much better QOL at the EOL. Caregiver's panic disorder was associated with worse patient's QOL at the EOL.
Patients who reported having received pastoral care services within the clinic or hospital had better QOL. Those whose religious beliefs or activities helped them cope with their illness and who participated in private religious activities before receiving their cancer diagnosis and at baseline had much better QOL at the EOL. Analyses of physician-patient relationships revealed a significant positive effect for patients who had a strong therapeutic alliance with their physician. Receipt of any life-prolonging procedure in the last week and an ICU stay predicted significantly worse QOL. Deaths in the ICU and hospital were associated with significantly worse QOL, whereas deaths at home were associated with significantly better QOL at the EOL.
Table 3 includes the best models identified in each of the 9 training sets and the average MSE values using all of the 9 training sets and the 9 test sets. The second model had the lowest average MSE values for training sets (49.93) and test sets (38.36) and therefore was selected as the final model.
Table 4 displays the estimation parameters in the best model identified in 1 training set (n = 352). The model included patient's receipt of pastoral care services within the clinic or hospital, therapeutic alliance, ICU stay, hospital death, patient's participation in private religious activities before receiving the cancer diagnosis, patient being worried, and chemotherapy and feeding tube in the last week of life. Because of the significant amount of missing data associated with the variables of informal caregiver as the source of the postmortem assessment (n = 311) and survival time (n = 310), these 2 variables were not included in the adjusted analyses. However, sensitivity analyses were performed to examine the impact of controlling for these 2 variables. When these 2 variables were included, all of the variables remained significant at P < .05 except for therapeutic alliance (P = .11), informal caregiver (P = .32), and survival (P = .26), as shown in Table 5. Table 6 presents the results by applying the final model to the full study sample, where receiving pastoral care services and therapeutic alliance were borderline significant while other predictors remained significant at P < .05.
The MSE for the best overall model was 51.40, with 17.7% of the variance explained by the predictors included in the final model estimated using the full study sample. The residuals account for the majority of the total variance, followed by an ICU stay, hospital death, worried patients, random effects of site, pastoral care services reported at baseline, chemotherapy in the last week of life, and therapeutic alliance (Table 7).
The aim of this study was to identify the best set of predictors of QOL of patients in their final week of life. By doing so, we identify promising targets for health care interventions to improve the QOL of dying patients.
The final model showed that providers with this aim should strive to reduce intensive life-prolonging care. Two of the most important determinants of poor patient QOL at the EOL were dying in a hospital and ICU stays in the last week of life. Therefore, attempts to avoid costly9 hospitalizations and to encourage transfer of hospitalized patients to home or hospice might improve patient QOL at the EOL. Because chemotherapy and feeding tube use also appeared in the final model, limiting these types of aggressive EOL care may be an effective strategy to enhance QOL at the EOL.
The best model also demonstrated that patient worry at baseline was one of the most influential predictors of worse QOL at the EOL. These results highlight the reduction of patient anxiety as a top priority for care aimed at enhancing QOL at the EOL. Patients who reported engaging in religious prayer or meditation had better QOL at the EOL. Pastoral care services within the clinic or hospital were significantly associated with better QOL at the EOL. These findings are consistent with other studies that have shown significant associations between spirituality and peacefulness and QOL in patients with life-threatening diseases.42,43 Evidently, terminally ill patients who participate in religious/spiritual activities privately and within the medical setting have better QOL near death than those who do not.
The best model in the training set found therapeutic alliance to be among the most important predictors of patient QOL at the EOL. Therapeutic alliance included measures of patients feeling treated with respect and as a “whole person” by their physician, trusting and respecting their physician, and feeling comfortable asking their physician questions about their care. When “survival” and “informal caregiver reporting of EOL QOL” were forced into the final model for conceptual reasons, the sample size dropped, and therapeutic alliance became marginally statistically significant. Although therapeutic alliance may be one of the weaker predictors, it nevertheless was among the top 9 factors predicting QOL at the EOL. These results suggest that physicians who are able to remain engaged and “present” for their dying patients—by inviting and answering questions and by treating patients in a way that makes them feel that they matter as fellow human beings—have the capacity to improve a dying patient's QOL.
As is always the case, this study is constrained by the data available. Even the best models explained less than 20% of the variance in QOL at the EOL, leaving much to learn about other influences on this outcome. There are undeniably many unmeasured factors (eg, provider and hospital characteristics) that contribute substantially to QOL. Future research with assessments of hospital (eg, number of ICU beds and number of clinical trials) and provider (eg, communication and treatment styles) characteristics and more comprehensive, prospective, repeated measures, particularly of therapeutic alliance and QOL, is needed.
Taken together, these results indicate that when medicine is no longer able to cure, physicians may still positively and significantly influence the lives of their patients. By reducing patient worry, encouraging contemplation, integrating pastoral care within medical care, fostering a therapeutic alliance between patient and physician that enables patients to feel dignified,44 and preventing unnecessary hospitalizations and receipt of life-prolonging care, physicians can enable their patients to live their last days with the highest possible level of comfort and care.
Correspondence: Holly G. Prigerson, PhD, Center for Psychosocial Epidemiology and Outcomes Research and Division of Population Sciences, Department of Medical Oncology, Dana-Farber Cancer Institute, Dana 1134, 450 Brookline Ave, Boston, MA 0221 (firstname.lastname@example.org).
Accepted for Publication: April 17, 2012.
Published Online: July 9, 2012. doi:10.1001/archinternmed.2012.2364
Author Contributions:Study concept and design: Zhang and Prigerson. Acquisition of data: Prigerson. Analysis and interpretation of data: Zhang, Nilsson, and Prigerson. Drafting of the manuscript: Zhang, Nilsson, and Prigerson. Critical revision of the manuscript for important intellectual content: Zhang, Nilsson, and Prigerson. Statistical analysis: Zhang, Nilsson, and Prigerson. Obtained funding: Prigerson. Administrative, technical, and material support: Prigerson. Study supervision: Prigerson.
Financial Disclosure: None reported.
Funding/Support: This research was supported in part by grant MH63892 from the National Institute of Mental Health (Dr Prigerson), grants CA 106370 and CA 156732 from the National Cancer Institute (Dr Prigerson), and the Center for Psychosocial Epidemiology and Outcomes Research, Dana-Farber Cancer Institute.