Adjusted odds of intensive care unit (ICU) use during the last 6 months of life among patients with chronic obstructive pulmonary disease compared with patients with lung cancer (adjusted for prior hospitalization, Charlson index, and age). Veterans Affairs (VA) sites include the following: 405, White River Junction, Vt; 521, Birmingham, Ala; 598, Little Rock, Ark; 652, Richmond, Va; 662, San Francisco, Calif; 663, Seattle, Wash; and 691, West Los Angeles, Calif.
Variation among Veterans Affairs (VA) sites in the odds of being admitted to the intensive care unit in the 6 months prior to death, stratified by condition. Veterans Affairs sites include the following: 521, Birmingham, Ala; 598, Little Rock, Ark; 652, Richmond, Va; 662, San Francisco, Calif; 663, Seattle, Wash; and 691, West Los Angeles, Calif. COPD indicates chronic obstructive pulmonary disease; and asterisk, adjusted for prior hospitalization, Charlson index, and age.
Au DH, Udris EM, Fihn SD, McDonell MB, Curtis JR. Differences in Health Care Utilization at the End of Life Among Patients With Chronic Obstructive Pulmonary Disease and Patients With Lung Cancer. Arch Intern Med. 2006;166(3):326–331. doi:10.1001/archinte.166.3.326
We sought to examine health care resource utilization in the last 6 months of life among patients who died with chronic obstructive pulmonary disease (COPD) compared with those who died with lung cancer and to examine geographic variations in care.
We performed a retrospective cohort study of patients diagnosed as having COPD or lung cancer, who were seen in 1 of 7 Veteran Affairs medical centers primary care clinics and who died during the study period. Our outcome of interest was health care resource utilization in the last 6 months of life.
In the last 6 months of life, patients with COPD were more likely to visit their primary care providers but had fewer hospital admissions compared with patients with lung cancer. Patients with COPD had twice the odds of being admitted to an intensive care unit (ICU), 5 times the odds of remaining there 2 weeks or longer, and received fewer opiates and benzodiazepine prescriptions compared with patients with lung cancer. There were geographic variations in the use of ICUs for patients with COPD but not for those with lung cancer. Total health care costs were $4000 higher for patients with COPD because of ICU utilization.
In the last 6 months of life, patients with COPD were more likely to have had a primary care visit and been admitted to an ICU but less likely to receive palliative medications compared with patients with lung cancer. We found significant geographic variability in ICU utilization but only for patients with COPD.
Chronic obstructive pulmonary disease (COPD) is the fourth leading cause of death in the United States.1 In contradistinction to cardiovascular and cerebrovascular disease, mortality from COPD continues to rise.1 At present, oxygen therapy for patients who have chronic hypoxemia is the only treatment that improves survival.2 Most therapies available for COPD are prescribed to improve quality of life and palliate symptoms.3
Two previous studies compared patients with severe COPD and nonoperable lung cancer.4,5 These studies demonstrated that patients with COPD had worse health-related quality of life, had depression and anxiety, received fewer palliative care services, and, despite having similar pREFERENCES for indefinite mechanical ventilation, were more likely to receive invasive therapy.4,5 These discrepancies suggest that physicians may be deciding to provide palliative services based more on general characterizations of disease (ie, “malignant” vs “benign”) rather than more specific characteristics, such as severity of symptoms or quality of life. To better understand this potential inconsistency, we examined variations in health care resource utilization in the last 6 months of life among a cohort of Veteran Affairs (VA) internal medicine clinic patients who had COPD or lung cancer.6,7
We performed a cohort study using data collected as part of the VA-funded Ambulatory Care Quality Improvement Project (ACQUIP). Details of ACQUIP have been described elsewhere.8 In brief, ACQUIP was a randomized clinical trial designed to test whether monitoring patients' self-reported health and providing regular reports to primary care clinicians improved clinical outcomes and patient satisfaction. The ACQUIP investigators sought to enroll all patients actively participating in the primary care clinics of 7 VA medical centers (VAMCs) including VA Puget Sound Health Care System, Seattle, Wash; West Los Angeles VAMC, Los Angeles, Calif; Birmingham VAMC, Birmingham, Ala; Little Rock VAMC, Little Rock, Ark; San Francisco VAMC, San Francisco, Calif; Richmond VAMC, Richmond, Va; and White River Junction VAMC, White River Junction, Vt.
Weekly interrogations of the VA computerized medical record system were performed to determine inpatient and outpatient visits. Inpatient and outpatient VA medical record information was available from January 1995 through September 2000. Among Medicare-eligible veterans, additional inpatient and outpatient visits were collected from the Centers for Medicare and Medicaid Services from January 1997 through December 1999. Information from both Medicare and VA data files included diagnoses based on the International Classification of Diseases, Ninth Revision (ICD-9), hospital ward location, and length of stay. We obtained all outpatient medications dispensed from VA facilities during the study period. All-cause mortality was assessed using the VA computerized medical record system and the Beneficiary Identification and Records Locator Subsystem. This combined assessment has been recommended by others9 and has been demonstrated to be accurate at identifying deaths.10
The Human Subjects Committee of the University of Washington, Seattle, approved this study that included patients enrolled in ACQUIP between December 1996 and October 1999. To be eligible for ACQUIP, patients must have been seen in the primary clinic, been assigned a primary care provider, have a future clinic visit scheduled, and have a valid mailing address.
We identified all participants who had died between April 1997 and September 2001 and had a diagnosis of COPD or lung cancer. We defined a patient's index date as that date 6 months prior to the date of death. We sought to identify patients who had been actively receiving care for either COPD or lung cancer. We identified patients as having COPD or lung cancer if they met any one of the following criteria: (1) 3 or more outpatient visits for COPD (ICD-9 codes 491.x, 492.x, 493.2, and 496.x) or lung cancer (ICD-9 codes 162.x, 195.1, and 165.x) during the 2 years prior to their index date, (2) had a primary inpatient ICD-9 diagnosis of COPD or lung cancer in the 2 years prior to their index date, or (3) had 1 or more clinic visits for COPD or lung cancer in the year prior to their index date and received medications specific for either COPD (≥4 canisters of ipratropium bromide or ≥6 canisters of albuterol) or lung cancer (platinum derivative).
Our primary outcome measure was health care resource utilization in the 6 months prior to death. We examined the use of all outpatient visits and inpatient hospitalizations and outpatient pharmacy services that were obtained through VAMCs. For eligible patients, we supplemented VA inpatient and outpatient visits with data from Medicare. Baseline health care utilization was assessed in the 12 months prior to the index date. As a proxy to cause of death, we assessed the primary ICD-9 discharge diagnosis of patients who were hospitalized within the 6-weeks prior to death. Costs associated with direct health care resource utilization were based on VA Health Economics Resource Center algorithms (http://www.herc.research.med.va.gov/home/default.asp; accessed December 16, 2005) and adjusted to year 2000 dollars.
We assessed overall comorbidity using the Deyo modification of the Charlson comorbidity index, which is derived from administrative diagnoses and is a valid predictor of hospitalization and mortality.11 We defined the presence of cardiovascular disease as an ICD-9 diagnosis of atherosclerosis (ICD-9 code 440.x), myocardial ischemia (ICD-9 codes 412-414 and 429.x), acute coronary syndromes (ICD-9 codes 410.x and 411.x), or chronic heart failure (ICD-9 codes 428.x, 398.91, and 425.x). We defined atrial fibrillation as an inpatient or outpatient ICD-9 diagnosis code of 427.3, 427.31, or 427.32. Patients were considered to have diabetes if they had an ICD-9 diagnosis code of 250.x. Distance to VAMCs was calculated as the straight-line distance from the geographic center of ZIP code of residence to the patient's VAMC. The correlation between travel times and straight-line distance has been demonstrated to be 0.987 in upstate New York and 0.826 for distances less than 15 miles.12
We performed bivariate analysis using χ2 tests for dichotomous variables. We used the Kruskal-Wallis test for variables with nonparametric distributions and t tests for normally distributed variables to assess differences between patients with COPD and those with lung cancer. We used multivariate logistic regression to estimate the relative risk associated with admission to the ICU or for a prolonged stay (>14 days) in the ICU and to adjust for potential confounding factors, including age, coexisting illnesses, distance to the VAMC, and prior hospitalizations. All statistical tests were 2 tailed, and P<.05 was used to define statistical significance.
Of 1949 patients who had died and met our cohort definitions, 1490 had COPD, 349 had lung cancer, and 110 had both lung cancer and COPD. Because patients with a diagnosis of COPD and lung cancer exhibited patterns of care similar to patients with lung cancer alone, we combined these 2 groups (Table 1). Consistent with national VA enrollment (Table 1), the cohort was mainly elderly white men, who lived a considerable distance from the VAMC. Almost all patients had a primary care clinic visit in the 12 months prior to their index date. Of the patients with lung cancer, 67% were hospitalized in the year prior to their index date compared with 50% of the patients with COPD (P<.001). Among Medicare-eligible patients who were hospitalized, those with COPD were nearly twice as likely to have been admitted to a non-VA facility (COPD, 24.9%; lung cancer, 10.7%; both, 19.2%; P<.001). Except for congestive heart failure, patients with COPD and those with lung cancer had similar prevalence of coexisting illnesses. In addition, Charlson scores were higher among patients with lung cancer, consistent with the higher weighting associated with those diagnosis codes.
Outpatient prescriptions for opiates and benzodiazepines, but not selective serotonin reuptake inhibitors, were less common for patients with COPD compared with those with lung cancer (Table 2). Compared with patients with lung cancer, patients with COPD were half as likely to receive an outpatient prescription for an opiate and 65% less likely to have received an outpatient prescription for benzodiazepines (Table 2).
A higher proportion of patients with COPD had made at least 1 primary care visit in the 6 months before death, but a lower proportion was hospitalized at a VA facility (Table 2). Among Medicare-eligible patients, those with COPD were more likely to have been hospitalized at a non-VAMC (39.8% for COPD vs 18.5% for lung cancer; P<.001). Although length of hospital stay among patients with COPD and lung cancer were similar, patients with COPD were twice as likely to be admitted to an ICU (Figure 1) and accrued more ICU days (Table 2). Patients with COPD had 5 times the odds of spending more than 2 weeks in an ICU and had twice the odds of being readmitted to the ICU (Figure 1). These results did not change significantly after adjusting for age, coexisting illnesses, and distance to the VAMC.
As shown in Figure 1, after adjustment, all sites were more likely to admit patients with COPD to the ICU compared with patients with lung cancer. For patients with COPD, there was statistically significant geographic variability in ICU admissions, with some sites having 3 times the odds of admitting a patient with COPD to the ICU (Figure 2). In contrast, no sites had statistically significant differences in the odds of admitting patients with lung cancer to the ICU (Figure 2), although there appeared to be a similar pattern of relative use of ICU care among sites.
While patients with cancer who were hospitalized during the last 6 weeks of life were admitted for their cancers, those with COPD were just as likely to have been admitted for an unrelated condition as they were for their lung disease (Table 3). Among the 597 patients hospitalized in the last 6 months of life but discharged prior to death, subsequent median survival was similar for both conditions (lung cancer, 34.5 days [interquartile range, 12.8-89.5 days]; COPD, 37.0 days [interquartile range, 13.0-78.2 days]; P = .90) (Table 2). Patients with COPD and lung cancer who were hospitalized in the last 6 months of life had a similar proportion of in-hospital deaths (lung cancer, 47.6%, vs COPD, 52.8%; P = .11).
In the last 6 months of life, median VA costs were approximately $4000 higher for patients with COPD than for patients with lung cancer (Table 2). Because length of stay was similar between patient groups, the difference in cost was attributable to ICU costs, which were approximately $5000 higher among patients with COPD.
Our results suggest that compared with patients with lung cancer, patients with COPD received care that was more consistent with prolongation of life than palliation of symptoms during their last 6 months of life, such as more frequent admissions to and greater time spent in an ICU and less use of outpatient opiates and benzodiazepines. Although all sites were more likely to admit patients with COPD than admit patients with lung cancer to the ICU, there were significant variations among sites in ICU admissions among patients with COPD. The cause of geographic variation in ICU admission is likely complex, including variations in patient pREFERENCES, case mix, and local organizational structure and staffing ratios. Geographic variation in utilization of services does not necessarily imply poor-quality care but suggests that understanding and addressing the cause of this variation may be an opportunity to improve quality of care.13
The greater use of ICUs for patients with COPD offset the higher outpatient costs of patients with lung cancer and accounted for substantially higher total costs. These differences in cost are likely conservative because patients with COPD were more likely to be admitted to non-VA facilities, potentially for acute indications including COPD exacerbations, and we did not capture these costs. Our results may suggest that providing palliative care to patients with COPD may improve end-of-life care, while reducing costs by minimizing unwanted ICU care. There are a number of potential limitations to this interpretation. First, we do not have reliable means to discriminate which patients will benefit from palliative care and not from ICU care.14- 16 Second, limiting ICU services without assessing pREFERENCES for such care would limit life-saving interventions to those patients who would survive their ICU admission. Finally, most total hospital budgets are fixed,17 suggesting limited cost savings associated with shifting to lower cost services.18,19
Previous studies have demonstrated that pREFERENCES for medical therapy at the end of life do not significantly differ from patients with COPD or lung cancer, and, in the context of our findings, raise the question of whether patients with COPD receive care that is in excess of their pREFERENCES for care.4 Furthermore, variations in care suggest a potential target for improving the delivery of care. One potential reason for these differences may be that COPD is not thought of as a terminal condition and patient pREFERENCES for care are not being addressed. Heffner and colleagues20 demonstrated that most patients with COPD undergoing rehabilitation wanted to discuss advance care planning with their physicians, but these conversations rarely occurred. Similarly, Curtis et al21 identified important physician skills for end-of-life care from patients with COPD, cancer, or AIDS and found that patients with COPD had similar pREFERENCES for communication, emotional support, and access and continuity of care but had a greater need for education about COPD prognosis including the dying process. In addition, unlike patients with lung cancer, patients with COPD may not recognize their trajectory toward death and may exercise pREFERENCES that are more consistent with preservation of life. Because patients with COPD are less likely to discuss advance care planning and have unmet educational needs, they may receive care as a reaction to acute deteriorations in symptoms in lieu of care that was based on long-term goals. Our results suggest that utilization differences were not attributable to patients' access to primary care because patients with COPD had a greater number of outpatient visits.
Among patients hospitalized in the 6 weeks prior to death, those with lung cancer were more likely to have cancer coded as the primary reason for hospitalization. In contrast, patients with COPD had nearly an equal distribution of primary reasons for hospitalization. These data suggest that COPD contributes to but may not be the immediate cause of death for many patients.22,23 Some have recommended developing patient profiles for patients for whom discussions about treatment pREFERENCES or end-of-life care are especially important.24 This profile could include a forced expiratory volume in 1 second of less than 30% of predicted norms, declining functional status, frequent hospitalizations, significant comorbid conditions, or advanced age. Having more than 1 of these indicators has been suggested as criteria for discussing advance treatment pREFERENCES.24 Although these characteristics are associated with morbidity and mortality, none reliability predict survival. Rather, because of the ambiguity in cause and trajectory to death, advance care planning may be appropriate for patients with severe airflow limitation regardless of other markers of severity. Advance care planning would certainly include patients who will not have severe exacerbations and death but would potentially fill unmet needs for education around advanced lung disease.21
Patients with COPD have been demonstrated to have similar degrees of severe pain and depression and worse dyspnea, functional status, and anxiety compared with patients with lung cancer.4,5 For patients with advanced COPD, maximal therapy produces modest resolution of symptoms, leaving significant residual disability.3 In this situation, treatment of pain and anxiety with benzodiazepines and/or opioids has been recommended.3 Although we did not assess the indication for these medications, our finding that patients with COPD received less palliative medications raises concern about unmet palliative care needs. Additional studies are needed to assess the unmet need based on symptom severity. Finally, because we did not have inpatient pharmacy records and most patients with COPD died during a hospitalization, we may have underestimated the use of these medications.
There were a number of strengths to this study. First, the cohort was drawn from a large and geographically diverse sample that reflected the demographic characteristics of the 5.5 million patients who receive care within the VA system. Second, we achieved relatively complete ascertainment of key variables such as mortality, utilization of services within the VA system, and medications. Third, the study was conducted in the setting of patients' regular care rather than in a terminal care setting.
There were also a number of limitations to this study. First, we had no assessment of pREFERENCES for care at the end of life or whether discussions had occurred between health care providers and their patients and therefore cannot determine the concordance between pREFERENCES and care. Second, we did not have information about the number of patients who were referred to hospice. Our estimates are conservative because patients with COPD are rarely referred to hospice and any differential would likely favor palliative care for patients with lung cancer. Third, we did not know lung cancer stage nor did we have spirometry to confirm or assess COPD severity. We hypothesized that frequent visits for COPD or lung cancer would accurately identify patients who had symptomatic disease. Finally, although we had no measure of severity of disease among patients with COPD or lung cancer, our study design accounted for some of the potential confounding by trajectory toward death by assessing the 6 months prior to death.
This study found that in the VA health care system, where COPD and lung cancer occur more commonly than in US general populations, patients with COPD receive care that may be more focused on preservation of life, and resultant costs are statistically significantly greater. Furthermore, we found significant geographic variability in the level of care provided to patients with COPD that was not present for patients with lung cancer. Our findings, in the setting of prior research, imply that to ensure that clinicians provide care consistent with patient wishes, discussions between patients, families, and health care providers should occur prior to an acute deterioration in clinical status. Additional research is needed to assess the concordance between patients' desire for treatment and the treatment received and to identify barriers, facilitators, and means to improve the occurrence of advance care planning. Because of its prevalence and burden, COPD is an important paradigm for future research about advance care planning and decision making regarding the balance and integration of palliative and life-prolonging therapies.
Correspondence: David H. Au, MD, MS, Health Services Research and Development (MS 152), Division of Pulmonary and Critical Care Medicine, 1660 S Columbian Way, Seattle, WA 98108 (email@example.com).
Accepted for Publication: July 7, 2005.
Financial Disclosure: None.
Funding/Support: This study was funded by grants IIR 99-376 and RCD 00-018 from the VA Health Services Research and Development, Washington, DC, and an award from the CHEST Foundation, Northbrook, Ill. Dr Au is funded by a VA Health Services Research and Development Career Development Award.
Disclaimer: The views expressed in this article are those of the authors and do not necessarily represent the views of the Department of Veteran Affairs.