JNCI J Natl Cancer Inst, 2015, 1–8 doi:10.1093/jnci/dju424 First published online January 20, 2015 Article
Palliative Care and the Aggressiveness of End-of-Life Care in Patients With Advanced Pancreatic Cancer Raymond W. Jang, Monika K. Krzyzanowska, Camilla Zimmermann, Nathan Taback, Shabbir M. H. Alibhai Affiliation of authors:Division of Medical Oncology and Hematology (RWJ, MKK) and Department of Psychosocial Oncology and Palliative Care (CZ), Princess Margaret Cancer Centre/University Health Network, Toronto, Canada (RWJ, MKK); Department of Medicine (RWJ, MKK, CZ, SMHA), Institute of Medical Sciences (MKK, CZ, SMHA), Division of Biostatistics, Dalla Lana School of Public Health (NT), and Institute of Health Policy, Management, and Evaluation (MKK, SMHA), University of Toronto, Toronto, Canada; Institute for Clinical Evaluative Sciences, Toronto, Canada (MKK). Correspondence to: Raymond W. Jang, Princess Margaret Cancer Centre, 610 University Avenue, Rm 5–217, Toronto, ON M5G 2M9, Canada (e-mail: [email protected]
Methods: A retrospective population-based cohort study using administrative data was conducted in patients with advanced pancreatic cancer from 2005 to 2010 in Ontario, Canada. Multivariable logistic regression was performed with the above measures of aggressive care as the outcomes of interest and PC as the main exposure, adjusting for covariables. Secondary analyses examined intensity of PC as the main exposure defined in two ways: 1) absolute number of PC visits before the outcome of interest (0, 1, 2, 3+ visits) and 2) monthly rate of PC visits. Results: The cohort included 5381 patients (median survival 75 days); 2816 (52.3%) had received a PC consultation. PC consultation was associated with decreased use of chemotherapy near death (odds ratio [OR] = 0.34, 95% confidence interval [CI] = 0.25 to 0.46); lower risk of ICU admission: OR = 0.12, 95% CI = 0.08 to 0.18; multiple ED visits: OR = 0.19, 95% CI = 0.16 to 0.23; multiple hospitalizations near death: OR = 0.24, 95% CI = 0.19 to 0.31). A per-unit increase in the monthly rate of PC visits was associated with lower odds of aggressive care for all four outcomes. Conclusion: PC consultation and a higher intensity of PC were associated with less aggressive care near death in patients with advanced pancreatic cancer.
Pancreatic cancer is a common and lethal type of cancer (1). Surgery is the only curative treatment, but only around 20% of patients are surgical candidates because most have advanced disease at diagnosis. Even among those who have a potentially
curative surgical resection, the five-year survival is only 20% (2). Patients with pancreatic cancer also have a high burden of symptoms and are frequent users of the emergency department (ED). In a recent study, patients with pancreatic cancer were second
Received: April 18, 2014; Revised: October 13, 2014; Accepted: November 26, 2014 © The Author 2015. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: [email protected]
1 of 8
Background: We examined the impact of palliative care (PC) on aggressiveness of end-of-life care for patients with advanced pancreatic cancer. Measures of aggressive care included chemotherapy within 14 days of death; and at least one intensive care unit (ICU) admission, more than one emergency department (ED) visit, and more than one hospitalization, all within 30 days of death.
2 of 8 | JNCI J Natl Cancer Inst, 2015, Vol. 107, No. 3
only to those with lung cancer in the frequency of ED visits in both the last six months and the last two weeks of life (3). Given that the majority of patients with pancreatic cancer die in a short period of time and with a high symptom burden, how they die, their quality of life near death, and the quality of care that they receive are important. For these reasons, pancreatic cancer is a good prototype cancer to study the quality of care that patients receive near death. Aggressive care near death for patients with incurable cancers is an important endof-life quality indicator (4,5), as aggressive care has been linked with worse quality of life for the patient and worse bereavement adjustment for their caregivers (6,7). The most common indicators used to study aggressive care near death include the administration of chemotherapy near death, frequent ED visits near death, hospitalizations near death (including death in a hospital), and intensive care unit (ICU) admissions near death (4,5,8–13). However, quality indicators are only useful if they are linkable to interventions that can be shown to improve areas of suboptimal performance. Because of its patient-centered approach and focus on pain and symptom management (14), palliative care (PC) may be an ideal intervention to reduce aggressiveness of care near death. Several randomized controlled trials have examined the effect of different interventions on the use of health care resources related to aggressiveness of care, such as number of ED visits (15–18), hospitalizations (15–22), ICU admissions (15), advanced care planning (23–25), and death at home (19,22,26–28) in cancer patients. It is difficult to draw definitive conclusions from these studies, as the results were mixed with respect to the effects of the intervention on health care use (29,30), and only a few of those studies included a PC physician consultation as part of the intervention and a patient population consisting of mostly cancer patients (17,20,27,31). The best evidence comes from a single-center academic randomized controlled trial that used early PC as an intervention for a cohort of patients with metastatic non–small cell lung cancer and found that patients randomly assigned to early PC experienced better quality of life, better mood, and longer survival, while receiving less aggressive care near death (31). Some retrospective evidence on the effects of PC on aggressive care exists (11,32–35); however, no large scale, population-based studies have examined this issue. The purpose of this study was to evaluate the impact of PC, including intensity of PC, on the aggressiveness of care near death in patients with advanced pancreatic cancer. Our hypothesis was that PC would be associated with decreased aggressive care near death for these patients.
Methods Study Design We conducted a population-based, retrospective study of an incident cohort of all patients who were diagnosed with advanced pancreatic cancer in Ontario, Canada, between Jan 1, 2005 and Dec 31, 2010 and who subsequently died. Follow-up data were available until March 2011. We included only patient cases with a valid provincial health insurance number and age 18 years or older at the time of diagnosis. Ethics approval was obtained from the University Health Network and the University of Toronto. Patient informed consent was waived, as the data was deidentifed and anonymized before given to the investigators.
OCR is a population-based cancer registry and is approximately 95% complete (36). Index cases from the OCR were linked using encrypted provincial health card numbers across different data sources. The data sources included: 1) Ontario Health Insurance Plan (OHIP) (37) claims database, which contains information on billing claims for physicians’ services provided to Ontario residents, 2) the Canadian Institute for Health Information (CIHI) Discharge Abstract Database (DAD) (38), which contains diagnostic and procedure information on all discharges from acute care facilities and information on same-day surgeries, 3) the CIHI National Ambulatory Care Reporting System (NACRS) (39), which captures information on emergency department visits, and 4) the Registered Persons Database (RPDB), which provides basic sociodemographic information.
Cohort Selection The OCR does not contain staging information for pancreatic cancer. Therefore, in order to avoid the bias of including patients with potentially curable disease (40), we created a proxy definition of advanced stage (ie, locally advanced and metastatic) by restricting patients with pancreatic cancer using the following criteria. Patients were excluded if they had: 1) a diagnosis of another cancer (either before or after the diagnosis of pancreatic cancer), as it would not be known which cancer contributed to the death, 2) major pancreatic cancer surgery with no chemotherapy, or 3) major pancreatic cancer surgery with chemotherapy within 120 days of surgery. Excluding patients who have had cancer-directed surgery to define a cohort of patients with advanced cancer has been used in a previous study of end-oflife care (12).
Study Objectives Our main objective was to evaluate the impact of an inpatient or outpatient PC consultation in patients with advanced pancreatic cancer on the proportions of: 1) chemotherapy near death (received the last dose of chemotherapy within 14 days of death), 2) multiple ED visits near death (more than one ED visit within 30 days of death), 3) multiple hospitalizations near death (more than one hospitalization within 30 days of death), and 4) ICU admission near death (at least one ICU admission within 30 days of death). A secondary objective was to study the impact of the intensity of PC using the same four measures of aggressive care. The intensity of PC was defined in two ways: 1) absolute number of PC visits before the outcome of interest (0, 1, 2, 3+ visits) and 2) monthly rate of PC visits defined as: number of PC visits ÷ (survival time in days ÷ 30). Patients who lived less than 14 days were excluded, as their limited survival would have artificially inflated the monthly rate (eg, if a person has one PC visit but only lived for one day, the monthly rate would be 1/[1/30] = 30 visits). The PC visits were counted from the day of diagnosis until each outcome of interest. The four outcome measures of aggressive end-of-life care were based on previous studies (5,41,42) and used the same codes as a previous Ontario study (41). PC was defined using OHIP physician billing data. Data sources and definitions of important variables are provided in Supplementary Table 1.
Statistical Analysis Data Sources We used the Ontario Cancer Registry (OCR) to identify index cases of patients with a diagnosis of pancreatic cancer. The
We calculated the frequency of each of the measures of aggressive care for all patients and conducted subset analyses for those who received or did not receive PC. The denominator
R. W. Jang et al. | 3 of 8
Sensitivity Analysis We conducted several types of sensitivity analyses. First, we limited the cohort to those who had received some palliative chemotherapy. These patients are different in that they are likely to have better performance status and have been seen by a medical oncologist. Second, we excluded patients diagnosed in 2010, as these will have had a shorter follow-up. Third, we included survival duration as a covariable in the logistic regression, as patients who lived for a longer time have more exposure time to receive PC, and differing survival times may have different effects on aggressive care near death. The multivariable logistic regression analyses were repeated by adding survival duration as a continuous covariable (in days) and as a categorical covariable (short duration: fewer than 90 days; medium duration: 90 or more days and less than 180 days; long duration: 180 or more days) in separate models. Fourth, we performed a hierarchical random effects model (44) with the patient’s specific medical oncologist as the clustering variable (level 1), while keeping all other covariables unchanged (level 2, patient-level), except for removal of the binary covariable of having seen a medical oncologist from level 2. This was done using each physician’s uniquely anonymized identifier from the OHIP database. Medical oncologists were chosen as the clustering variable, as some medical oncologists may be more likely to refer to palliative care, while recommending less aggressive treatment, creating a possible bias for our study. We considered both palliative care and primary care physicians as alternative clustering variables. Palliative care physicians were felt to be an inappropriate clustering variable since they constituted the main exposure. We felt that family physicians usually are not the ones involved in decision-making such as chemotherapy near death and ICU admission near death. Some family physicians are also the local palliative care doctors and do home visits—when these doctors billed the palliative care billing codes, the patient was considered as having received palliative care. Analysis of within-cluster variance was also performed. The random effects model was chosen to be a sensitivity analysis as opposed to the main analysis, as approximately 40% of the patients in our cohort had not seen a medical oncologist. We also performed a propensity score analysis based on the receipt of PC, as those receiving PC may be different in ways that cannot be measured than those who did not. We created
a propensity score for the likelihood of receiving PC, ranked all the patients, created quintiles based on this score, and included the scores as an independent variable in our four main regression analyses. Lastly, we examined the effect of the medical oncologist on each of our outcomes by using a two-level hierarchical random effects model. The specific methodology and results are outlined in the supplemental file.
Results Cohort Description and Patient Demographics In our initial cohort, 7928 patients were diagnosed with pancreatic cancer in Ontario. Of these, 670 were excluded because they had a diagnosis of a second cancer, leaving 7258 patients. Of the 7258 patients, 1182 were excluded because they underwent pancreatic cancer surgery (533 without adjuvant chemotherapy, and 649 with adjuvant chemotherapy). The remaining 6076 patients were considered to have advanced disease. Of these, 5381 (88.6%) had died by March 2011 and comprised our cohort. Table 1 shows the patient characteristics. Patients age 80 years and older made up 29.1% of the final cohort. Males and females were equally represented, 14.6% of patients lived in rural areas, 81.5% of patients had a primary care physician, and 63.1% of patients had seen a medical oncologist.
Palliative Care Visit Characteristics and Survival Times Table 1 also displays characteristics of PC visits and survival times. Approximately half of the patients (2816 of 5381 or 52.3%) had a PC consultation: 9.5% had one visit, 10.3% had two visits, and 32.5% had three or more visits. For patients who had seen a PC physician, the median and mean numbers of PC visits were 3 and 5.7, respectively (interquartile range [IQR] = 2 to 7). The median and mean survival times of patients with advanced pancreatic cancer were 75 and 147 days, respectively. Of those who received some palliative chemotherapy, the median and mean survival times were 199 and 264 days, respectively. Patients who received PC also had longer survival times (eg, median survival of 100 days vs 51 days) compared with those who did not.
Frequency of Aggressive Care Near Death and Association With PC Of the 5381 patients, 4.1% received chemotherapy near death, 4.3% had ICU admissions near death, 18.5% had multiple ED visits near death, and 8.3% had multiple hospitalizations near death; 74.1% of patients did not experience any of the four measures of aggressive care near death (Table 2). Table 2 also shows the unadjusted frequency of each outcome for all patients and by receipt of PC. Patients who had received a PC consultation had lower unadjusted frequencies of aggressive care than those who did not: 2.6% vs 5.6% (χ2 = 29.9, P < .001) for chemotherapy near death, 1.1% vs 7.8% (χ2 = 142.3, P < .001) for ICU near death, 7.4% vs 28.5% (χ2 = 393.5, P < .001) for multiple ED visits, and 3.8% vs 12.8% (χ2 = 141.8, P < .001) for multiple hospitalizations. Figure 1 illustrates the relationship between aggressive care and number of PC visits. Table 3 shows the results of the multivariable logistic regression analyses. PC consultation was associated with lower odds of receiving any of the four measures of aggressive care near
included all patients who died with a diagnosis of an advanced pancreatic cancer. Chi-squared (χ2) tests were performed as a first test to determine differences in aggressiveness of care among patients who did and did not receive palliative care. All statistical tests were two-sided and a P value of less than .05 was considered statistically significant. Multivariable logistic regression analyses were used to predict the likelihood of experiencing each of the outcomes of aggressive care, with PC as the main exposure and the following variables as covariables: age, sex, rural residence, health region, year of diagnosis, income quintile, primary care physician involvement, medical oncologist involvement (for outcomes other than chemotherapy near death), chemotherapy (for outcomes other than chemotherapy near death), and Charlson-Deyo Comorbidity Index Score (0, 1, 2+) (43). The index was obtained from hospitalizations and same-day surgery admissions only. Similar analyses were performed for the secondary objective, using the two definitions of intensity of PC in separate models. All analyses were performed with the R statistical software (http://www.r-project.org; version 2.14.0).
4 of 8 | JNCI J Natl Cancer Inst, 2015, Vol. 107, No. 3
death: chemotherapy near death (OR = 0.34, 95% CI = 0.25 to 0.46); ICU admission (OR = 0.12, 95% CI = 0.08 to 0.18), multiple ED visits (OR = 0.19, 95% CI = 0.16 to 0.23), and multiple Table 1. Pancreatic cohort demographics (n = 5381) Characteristic