|Home | About | Journals | Submit | Contact Us | Français|
Patients with advanced-stage cancer are receiving increasingly aggressive medical care near death, despite growing concerns that this reflects poor quality care.
Assess the association of aggressive EOL care with bereaved family members’ perceptions of the quality of EOL care and patients’ goal attainment.
We surveyed 1,146 family members (median [IQR] days after death, 144.5, [IQR 85.0-551.0]) of elderly Medicare patients with advanced-stage lung or colorectal cancer in the Cancer Care Outcomes Research and Surveillance study (a multi-regional, prospective, observational study) who died by 2011.
Claims-based quality measures of aggressive EOL care (i.e., ICU admission or repeated hospitalizations or emergency department visits during the last month of life; chemotherapy ≤2 weeks of death; no hospice or ≤3 days of services; and deaths in hospital).
Family reported “excellent” quality of EOL care. Secondary outcomes included: patient goal attainment (EOL care congruent with patients’ wishes and death in preferred place).
Of 1,146 cancer patients (median [IQR] age, 76.0 [65.0-87.0] years, 55.8% male), bereaved family members reported excellent EOL care quality for 51.3%. Family of patients who enrolled in hospice >3 days before death reported excellent EOL care quality more often than those receiving no or ≤3 days [58.8% (352/599) vs. 43.1% (236/547); adjusted difference=16.5 percentage points; 95% CI=10.7 to 22.4]. In contrast, family of patients admitted to an ICU ≤30 days before death or who died in the hospital less often reported excellent EOL care quality than those who were not [45.0% (68/151) vs. 52.3% (520/995); adjusted difference=-9.4 percentage points; 95% CI -18.2 to -0.6; and 42.2% (194/460) vs. 57.4% (394/686); adjusted difference=-17.0 percentage points; 95% CI=-22.9 to -11.1, respectively]. Family of patients receiving ≤3 days of hospice services were less likely to report patients died in their preferred place [40.0% (152/380) vs. 72.8% (287/394); adjusted difference=-34.4 percentage points; 95% CI=-41.7 to -27.0).
Among family members of elderly patients who died with lung or colorectal cancer, perceptions of better end-of-life care were associated with earlier hospice enrollment, avoidance of ICU admissions ≤30 days of death, and death outside the hospital. These findings are supportive of advance care planning consistent with patients’ preferences.
Patients with advanced-stage cancer are receiving increasingly aggressive medical care at the end of life,1 despite growing evidence that high-intensity treatments may not be associated with better patient quality of life, outcomes, or caregiver bereavement.2 The National Quality Forum (NQF) and American Society of Clinical Oncology (ASCO) have jointly endorsed several indicators of overly “aggressive” end-of-life care, developed by Earle et al.,3 including: repeated hospitalizations or emergency department visits or admission to an intensive care unit (ICU) within the last month, chemotherapy ≤2 weeks before death, and late or absent hospice referrals.4
The NQF/ASCO end-of-life care measures are integrated into ASCO’s Quality Oncology Practice Initiative5 and will likely become more important because the Patient Protection and Affordable Care Act mandates public reporting of quality metrics and has linked incentive and penalty payments to similar indicators (e.g., the Centers for Medicare & Medicaid Services Oncology Care Model).6 Improving end-of-life care offers an important opportunity to improve cancer patients’ dying experience, use resources more effectively,7 and potentially prolong survival.8
However, few studies have examined whether these aggressive care measures reflect patients’ preferences or bereaved family members’ perceptions of the quality of care provided. In this study, we assessed the relationship between aggressive end-of-life care and family ratings of the quality of end-of-life care using data from the Cancer Care Outcomes Research and Surveillance (CanCORS) Consortium.9 We also examined whether patients treated with aggressive care near death received care that was congruent with their preferences and died in their preferred place (i.e. patient goal attainment).
Data came from the CanCORS and CanCORS II studies linked to Medicare claims from 2003-2011. As described previously,9,10 CanCORS enrolled patients newly diagnosed with lung or colorectal cancer during 2003-2005 from 5 regions (northern California, Los Angeles County, North Carolina, Iowa, and Alabama) and 5 integrated health systems. Trained interviewers surveyed participants or their family in English, Spanish or Chinese using computer-assisted telephone interview software at 3 time points: 1) approximately 4-6 months after diagnosis, 2) 1 year after diagnosis (for patients alive at the first survey), and 3) 5-7 years after diagnosis (for patients alive 1 year after diagnosis). Our study protocol was approved by Institutional Review Boards at all participating sites and all participants provided informed consent.
For this analysis, a family member or close friend of the deceased patient was interviewed after the patient’s death about their treatment preferences,11 symptoms near death, end-of-life wishes, preferred place of death, quality of care, and place of death. If the patient was deceased at the initial study contact the next-of-kin was invited to participate. Patients who were alive for the baseline survey were asked to identify a primary family member or friend “familiar with your care since diagnosis” and a secondary respondent (in case the first could not be reached). Because 95% of surveys were completed by family, we will hereafter refer to respondents as family members.
We measured the intensity of end-of-life care received by participants enrolled in traditional Medicare by obtaining claims data with approval from the Privacy Board of the Centers for Medicare and Medicaid Services. Potentially eligible enrollees were linked using a common matching algorithm based on Social Security number, date of birth and gender,12 which matched 81.5% of CanCORS enrollees aged ≥65 to Medicare data.
CanCORS participants who were diagnosed or recurred with advanced lung or colorectal cancers, died by 2011, were aged ≥65 and continuously enrolled in Parts A&B fee-for-service Medicare for ≥3 months before death were included.
We examined family-reported quality of end-of-life care and two measures of patient goal attainment. To assess care quality, family members were asked: “Overall, how would you rate the care received [at the last place where of care was provided]? Would you say it was excellent, very good, good, fair, or poor?” We defined high quality end-of-life care as care that family members rated as “excellent.”13
We assessed patient goal attainment using two measures. First, family members were asked “In your opinion, to what extent were [the patient’s] wishes followed in the medical care received in the last month of life?” Goal attainment was defined as care that “followed patients’ wishes a great deal.” In addition, family were asked where the patient’s preferred and actual places of death were; by design, these questions were separated by >40 questions to minimize conscious comparison, but included similar categories. Categories of preferred place of death included: hospital, home, relative’s home, nursing facility, other; categories of actual place of death included hospital, home, nursing home, hospice inpatient unit, other (specify). Patients whose actual place of death matched their preferred place were considered to have achieved goal attainment. Preferences for home were considered to match if the patient died at home or a relative’s home. For the small number of patients who died in a hospice inpatient unit, or “other” that could not be assigned to pre-existing preference category, the variable was coded as unknown.
We previously developed 5 measures of aggressive end-of-life care that address: 1) overuse of chemotherapy very near death, 2) possible misuse of treatments resulting in emergency room visits, hospitalizations or ICU stays and 3) underuse of hospice. These measures were developed through literature reviews and focus groups with patients, families, and an expert panel of providers.3 We identified aggressive end-of-life care using Medicare claims data.14,15 We examined deaths in hospitals using family-reported place of death since these have been associated with worse family-reported patient quality of life near death, compared with home deaths, and more psychiatric morbidity among caregivers.16
We obtained information about patients’ sociodemographic characteristics (age, marital status, education and income), overall health status, comorbid conditions, depression, and treatment preferences from decedent surveys with family. Race/ethnicity was analyzed as a potential determinant of end-of-life care because patients may vary in their treatment preferences, advance care planning, and health utilization by race/ethnicity.17 Stage at diagnosis and cancer type was obtained through medical record abstraction or, if unavailable, from cancer registries. We also adjusted for geographic region and enrollment in an integrated health system, which included 5 sites and Kaiser Permanente of Northern or Southern California. Finally, we adjusted for time from diagnosis to death as a linear variable, family respondent (e.g., spouse/partner, child, other relative, or friend), and time from death to survey completion as a linear variable. In additional analyses, we tested for an interaction of time from death to survey completion and respondent role; none of the interactions were statistically significant, so models included only main effects.
Descriptive statistics were used to characterize family-reported end-of-life care quality, goal attainment, and rates of aggressive end-of-life care according to patients’ preferences. We examined associations between aggressive end-of-life care and family-reported 1) “excellent” quality of end-of-life care, 2) care consistent with wishes, and 3) death in preferred place using regression models, adjusted for all other covariates, regardless of statistical significance. For each dependent variable, separate models were fit for each indicator of aggressive end-of-life care, adjusting for the patient characteristics described. We fit linear-binomial models with an identity link to estimate adjusted differences in the likelihood of each outcome.18
In each model, we included patients with non-missing data for that outcome measure. Since covariate information was missing for between 0.8% (education) to 10% (treatment preferences) of the cohort, we used multiple imputation methods to create 5 complete datasets and repeated all analyses on each imputed dataset, combining results using standard methods for multiply imputed data19.
Sensitivity analyses were conducted to examine associations between patients’ preference for life-extending treatments and receipt of aggressive care over shorter time periods (i.e. ≤90 days and ≤180 days from diagnosis to death). We examined the effect of re-categorizing family members’ quality of care ratings as “excellent and very good” vs. “good, fair, and poor” and as an ordinal model. We also fit models that included family members’ contact with patients during the last week as a proxy for closeness. We repeated these analyses among a sub-set of the total cohort with information from the baseline and 1 year follow-up surveys, both with and without family-reported frequency of discussing important medical decisions with the patient. Next, we examined the effect of including patients’ symptoms at the end-of-life (pain and difficulty breathing) and how well these were controlled since uncontrolled symptoms often precipitate hospitalizations near death. Finally, we repeated all analyses stratified by disease site. Two-sided P values <0.05 were considered statistically significant. All statistical analyses were performed with SAS software (version 9.3; SAS Institute, Cary, NC).
Among patients who were diagnosed with or recurred with advanced cancer (N=3620), 2011 were linked to medical claims; of these, 1847 had died by the end of 2011 and 1713 were aged 65 and older as of 3 months before death (linkage rate with claims was >90% across all sites for patients aged 65 and older). We excluded 449 patients for whom we did not have an after death survey from a family member (after death surveys were available for 73.8% of potentially eligible patients). We also excluded 92 patients who were not continuously enrolled in parts A and B of fee-for-service Medicare as of 3 months before death and 26 patients for whom the family member did not rate the quality of end of life care, leaving a final study population of 1146 patients.
Most of the study population had stage IV disease at diagnosis (Table 1). Lung cancer decedents had much shorter median survivals than colorectal cancer decedents (122.5 vs. 572.0 days), and were more likely to die in 2003-2005 compared with later (85.0% vs. 50.0%, P<0.001, Supplemental Table 1). More than 80% of family members were spouses or children; 15.2% were other relatives or friends. Family reported a median of 7.0 days (IQR 6.0-7.0) of contact with patients in the last week of life. The median time between patient deaths and the decedent interview was 144.5 days (IQR 85.0, 551.0). Patients with a post-mortem family interview did not differ significantly from patients without by cancer type or gender; however, they were more likely to be older, white, and have died <3 months of diagnosis (Supplemental Table 2).
Overall, family members reported that the quality of end-of-life care was, “excellent” for 51.3% of decedents, “very good” for 27.8%, and “good,” “fair,” or “poor” for 20.9% (Table 2). Among family members, 81.1% indicated that patients’ end-of-life wishes were followed a “great deal,” while 18.9% reported that end-of-life care was “somewhat” or “not at all” consistent with patients’ wishes; only 56.7% of patients died in their preferred place.
Patients’ treatment preferences were significantly associated with the medical care received near death (Table 3). For example, patients who preferred life-extending therapies were more likely to receive chemotherapy ≤2 weeks of death than those who preferred comfort care (12.4% vs. 3.9%, P<0.001), and were more likely to receive ≤3 days of hospice (64.2% vs. 41.0% P<0.001). Results were similar in patients with shorter survival times (Supplementary Tables 3a-b)
In adjusted analyses (Table 4), family of patients who received no hospice or ≤3 days of services before death were less likely to report excellent quality end-of-life care than those who received more [43.1% (236/547) vs. 58.8% (352/599); adjusted difference: -16.5 percentage points; 95% CI: -22.4 to -10.7]. Similarly, family of patients admitted to an ICU ≤30 days of death were less likely to report excellent quality [45.0% (68/151) vs. 52.3% (520/995); adjusted difference: -9.4 percentage points; 95% CI: -18.2 to -0.6], compared with those who were not, as were family of patients who died in the hospital [42.2% (194/460) vs. 57.4% (394/686); adjusted difference: -17.0 percentage points; 95% CI: -22.9 to -11.1]. Family-reported ratings of end-of-life care were similar in patients with and without frequent use of the emergency department in the last month of life [51.1% (46/90) vs. 51.3% (542/1056); adjusted difference: -1.5 percentage points; 95% CI: -12.8 to 9.7]; and in patients with and without chemotherapy ≤2 weeks of death [49.3% (37/75) vs. 51.5% (551/1071); adjusted difference -2.6 percentage points; 95% CI: -14.8 to 9.6].
Family of patients who enrolled in hospice >3 days before death were much more likely to report that patients died in their preferred place [72.8% (287/394) vs. 40.0% (152/380); adjusted difference: 34.4 percentage points; 95% CI: -41.7 to -27.0; Table 5]. In contrast, patients who were admitted to an ICU in the last month were much less likely to die in their preferred place [38.0% (35/92) vs. 59.2% (404/682); adjusted difference: -22.8 percentage points; 95% CI: -33.5 to -12.0] as were those who received chemotherapy ≤2 weeks of death [42.1% (16/38) vs. 57.5% (423/736); adjusted difference: -17.7 percentage points; 95% CI: -34.0 to -1.4].
In sensitivity analyses that re-categorized family-reported quality of care as “excellent” and “very good” vs. “good,” “fair,” and “poor,” and as an ordinal model, the results were consistent with the main findings. There were few differences in models which included family-reported contact with patients in the last week or frequency of discussions of important medical decisions with patients. We also found few differences in results from analyses that included family-reported measures of patients’ symptoms, and whether they were adequately controlled (Supplemental Tables 4&5). Finally, in models stratified by cancer type, the results were comparable, except that ICU admissions were not associated with family-reported quality of end-of-life care. Among patients with colorectal cancers, those who enrolled in hospice ≤3 days before death were less likely to report that patients’ wishes were followed “a great deal” (66.6% vs. 84.4%; adjusted difference=-17.8 percentage points; 95% CI=-28.9 to -6.7; P=0.002, data not shown).
In this large, diverse population-based cohort, we found that three measures of aggressive end-of-life care—ICU admission within 30 days of death, no or ≤3 days of hospice services, and deaths in hospital—were associated with relatively large differences in family reported end-of-life care quality and a lower likelihood that advanced-stage cancer patients received care congruent with their preferences. Our findings suggest that efforts to increase earlier hospice enrollment and avoid ICU admissions and hospital deaths that are not consistent with patients’ informed preferences might improve the quality of end-of-life care. For example, implementation of multi-faceted approaches (e.g., enhanced counseling of patients and families, early palliative care referrals, and an audit-and-feedback system to monitor physicians’ use of aggressive end-of-life care) might result more preference-sensitive and improved quality of care.2,16
To our knowledge, this study is one of the first empirical validations of these aggressive end-of-life care indicators using patient- and family-centered outcomes in a population-based cohort. These findings may have policy implications because administrative-database indicators4 are becoming increasingly feasible to assess with the expansion of electronic health records under the Health Information Technology for Economic and Clinical Health Act. This is particularly noteworthy given growing evidence that physicians’ characteristics and beliefs are one of the strongest predictors of end-of-life care,20,21 suggesting that modifications to physicians’ practices may result in significant improvements.
Several studies have previously documented an association between less-aggressive end-of-life care and improved patient quality of life near death.2,22-27 Our study extends this research by demonstrating similar associations with family-reported quality of end-of-life care in a large cohort of patients treated in community settings in multiple regions of the United States. In addition, we assessed patients’ preferences for care and found that most elderly patients wanted palliative treatments over life-extending care, and preferred to die at home, similar to other studies.28,29 Such preferences were strongly associated with the aggressiveness of end-of-life care received.
Although existing quality measures characterize the repeated use of emergency visits near death as poor quality care, we did not observe a difference in family ratings by this measure, perhaps because some patients or caregivers may be unprepared or unable to manage acute symptoms at home.30 The lack of validation of this indicator suggests that some aggressive care measures may be less salient to patients and families.
Similarly, we found no differences in family-rated care quality among patients who received chemotherapy ≤2 weeks of death, compared with those who did not, although we were underpowered to detect these differences since <7% of patients received this. Nevertheless, chemotherapy use ≤2 weeks of death was associated with a decreased likelihood that patients’ died in their preferred place, consistent with another study.31
Our study had some limitations. Although the patients who enrolled in CanCORS were representative of patients diagnosed with lung and colorectal cancer in the US regions covered by Surveillance, Epidemiology, and End Results Registries,32 we studied older patients insured by Medicare with lung and colorectal cancers who died predominantly before 2008 and had to exclude many patients who were not Medicare fee-for-service enrollees or whose family members could not be reached. Our findings may not generalize to patients who are younger, commercially insured, uninsured, treated in Veteran’s Administration hospitals, have other diseases, or died more recently. The data are older, particularly for patients who died soon after diagnosis, due to the time involved with linking survey and claims data. However, one of the paper’s strengths is it includes patients who were longer-term survivors and may have very different experiences near death. Additionally, recent data demonstrates that care at the end of life among Medicare beneficiaries remains intensive (ICU admissions and late hospice enrollments are increasing over time);1 thus patients’ and family members’ experiences are likely to be similar today. Our results may be subject to unmeasured confounding related to unsuccessful linkages to Medicare claims. However, our rates of family-reported “excellent” quality of end-of-life care and receipt of care that was not consistent with patients’ preferences are similar to another population-based study of decedents with fee-for-service and HMO Medicare plans who died from diverse diseases.33 Family members reported on patients’ end-of-life experiences at variable time points after death, and some may have forgotten details over time. Although a recent study demonstrates moderate to high stability in bereaved family members’ assessments of patients’ end-of-life experiences over time,34 future research should examine family members’ recall over longer durations of time. We relied on surrogate reports of patients’ preferred and actual place of death to determine goal-attainment. Nevertheless, these questions were separated by at least 40 questions to minimize conscious comparisons. Finally, we were underpowered to examine some associations; e.g., we had 80% power to detect a 17% absolute difference in the quality of end-of-life care between patients who were and were not hospitalized ≥2 times in the last month of life, but this outcome was relatively infrequent.
Among family members of elderly patients who died with lung or colorectal cancer, perceptions of better end-of-life care were associated with earlier hospice enrollment, avoidance of ICU admissions in the last 30 days of life, and death outside the hospital. These findings are supportive of advance care planning consistent with patients’ preferences.
This work was supported by the National Cancer Institute (NCI): grant R01 CA91753 (Earle) and the Cancer Care Outcomes Research and Surveillance (CanCORS) Consortium (grant numbers U01 CA093344 to Dana-Farber Cancer Institute, U01 CA093324 to Harvard Medical School and Northern California Cancer Center, U01 CA093332 to Dana-Farber Cancer Institute and Cancer Research Network, U01 CA093348 to RAND Corporation and University of California, Los Angeles, U01 CA093329 to University of Alabama at Birmingham, U01 CA093339 to University of Iowa, U01 CA093326 and University of North Carolina) and R01 CA1604021 (Keating). The CanCORS II work was funded by grant 5U01CA093344-08 from the National Cancer Institute to the CanCORS Consortium Statistical Coordinating Center. Dr. Wright is supported by grant K07 166210 from the NCI. Dr. Keating is also supported by K24 CA181510 from the NCI.
Authorship acknowledgements: The authors would like to thank Lin Ding, PhD, for conducting statistical analyses. Dr. Ding is employed by Harvard Medical School and her work on this project was supported by research grants to Harvard Medical School from the National Cancer Institute.
Role of the Sponsor
The study was funded by the National Cancer Institute (NCI). Researchers from the NCI were involved in design and conduct of the study, collection, management, analysis, and interpretation of the data. NCI researchers were not involved in preparation, review, or approval of this manuscript, or in the decision to submit this manuscript for publication.
Authorship contributions: Drs. Wright and Keating had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.Study concept and design: Wright, Keating, Ayanian, Weeks, Earle, Landrum
Analysis and interpretation of data: All authors
Drafting of the manuscript: Wright, Keating, Landrum
Critical revision of the manuscript for important intellectual content: All authors
Final approval of the manuscript: All authors.
Conflicts of interest: The authors have no relevant financial disclosures to report.