Although more than one-third of Americans who die use hospice care, many choose aggressive, often futile treatments at the end of life.^14,15 Approximately 30% of lifetime Medicare resources are expended in the last year of life, but hospice care may reduce these costs, especially for individuals dying of cancer.^13,16–20 When patients do seek hospice care, their median hospice stay lasts only 26 days, in large measure because of late referrals by health care providers.⁶

Of the more than 28 000 American men who die of prostate cancer each year, only one-third enroll in hospice.^{5,7–10,21,22} We sought to characterize not only hospice use by men dying of prostate cancer but also differences in the care these men receive compared with men not enrolled in hospice. The primary purpose of our analysis was to identify patient characteristics associated with hospice use among men dying of prostate cancer. Our secondary aim was to compare the use of diagnostic and interventional procedures and physician visits between those who did or did not enroll in hospice. While hospice entry is generally associated with lower use of services, the types of studies and services that are and are not used when an individual enrolls in hospice are incompletely understood. Delineating use of medical care at the end of life, with and without hospice enrollment, can be important in developing policies to deliver high-quality, affordable care at the end of life. We hypothesized a priori that white ethnicity, higher household income, and more formal education would be associated with greater hospice use.^10–12,22 We also posited that hospice use would be correlated with fewer intensive care unit (ICU) and inpatient admissions and with fewer interventional procedures in the last 6 months of life, especially those procedures involving placement of decompressive ureteral stents or nephrostomy tubes.

Table 1 gives the bivariate associations between sociodemographic and clinical characteristics and hospice use. We included in our analysis those who used hospice for 7 to 180 days prior to death, which we had defined as appropriate hospice use. Sixteen percent of the men in our cohort were younger than 70 years, the majority were white, and most had localized or locoregional disease at diagnosis. The P value represents differences in hospice use by sociodemographic factors; age, ethnicity, partnership status, Charlson comorbidity index, stage, primary prostate cancer treatment, household income, education level, and time from diagnosis to death were associated with hospice use.

Our multivariable model evaluating factors associated with hospice use demonstrated that white ethnicity, primary treatment type, income, lower Charlson comorbidity index, being partnered, and more recent year of death were associated with higher odds of hospice use (Table 2). We also identified significant regional variation, with lower hospice use in the Los Angeles registry compared with several other SEER registries. This variation remained significant after incorporating an interaction between subject ethnicity and SEER region.

Table 3 displays procedures and physician visits during the last 6 months of life. In our bivariate analysis, subjects not enrolled in hospice had more imaging studies, electrocardiograms, and cardiopulmonary resuscitations performed but received less palliative radiation. They made more emergency department visits and had more inpatient hospital stays and ICU admissions but made a comparable number of outpatient office visits.

At the other end of the spectrum, the Medicare hospice benefit requires that a primary care physician and a hospice medical director certify than an individual’s expected prognosis does not exceed 180 days when he or she is enrolled in hospice.^13,30 Despite these guidelines, physicians routinely overestimate their patients’ survival, with prognostic accuracy declining the longer the clinician has known the patient.³¹ The use rate we report builds on previous population-based studies of men dying of prostate cancer, in which use rates ranged from 18% to 43%, with a temporal trend toward increased use.^{5,7–10,21,22}

While overall rates of hospice use by all patients with cancer, including those with prostate cancer, have improved over time, the timing of hospice referral has not changed significantly over the past 25 years.^5,13 Our findings highlight both the successes of increased hospice use, to a zenith of 62% of men dying of prostate cancer in 2003, as well as avenues for improvement in the timing of hospice referral.

Second, we found that in the last 6 months of life, men not enrolled in hospice underwent more imaging studies and received more high-intensity care, which comprised inpatient hospital stays, ICU admissions, multiple emergency department visits, chemotherapy, cystoscopy, and ureteral stent or nephrostomy tube placement. Our hypothesis that stent or nephrostomy tube placement would represent the most frequent end-of-life intervention in men dying of prostate cancer was incorrect. Men enrolled in hospice were at least as likely as those not enrolled to make outpatient visits and to receive palliative radiation, suggesting that hospice enrollment did not prevent individuals from receiving primary or palliative care. Medicare expenditures for imaging services have risen approximately 17% per year in the last decade, from $3.6 billion in 2000 to $7.6 billion in 2006.^32,33 No data endorse a connection between increased use of imaging studies and patient outcomes, especially at the end of life.³⁴ Although the mean number of imaging studies performed on all men in our analysis was high overall, it was substantially higher in those not enrolled in hospice.

Likewise, the significant burden of high-intensity procedures performed in the last 6 months of life, particularly in men not enrolled in hospice, raises concern over what Emanuel³⁵ has termed cost without benefit. While our population-based analysis did not allow us to characterize the specific reason for each individual emergency department visit or hospital admission, a high number of emergency department visits and hospital admissions, including ICU stays, indicates poor-quality end-of-life care, according to the quality of care indicators proposed by Earle et al.²⁸ Data from the National Hospice Study and other analyses suggest that hospice enrollment improves symptom management, quality of death, and quality of life at the end of life.^3,4 While prospectively collected data are lacking, observational data suggest that hospice enrollment significantly reduces the cost of caring for individuals at the end of life.^{5,7,13,16,18–20} If hospice enrollment is associated with lower cost and higher quality of life at the end of life, and if, as our analysis shows, hospice enrollment portends the performance of fewer costly high-intensity interventions, hospice use may minimize resources spent with limited benefit. Approximately one-third of the total Medicare budget is expended on individuals at the end of life.¹³ The United States currently spends approximately $2.2 trillion per year on Medicare and Medicaid, accounting for 16% of the gross domestic product, and, as the current Director of the Office of Management and Budget pointed out in 2007, “The long-term fiscal balance of the United States will be determined primarily by the future rate of growth of health care costs.”^36,37(p1885) The Obama Administration’s fiscal year 2010 budget aims at reducing hospital admission rates as one way of controlling cost without sacrificing quality.^38,39 Our findings suggest that increasing indicated hospice enrollment may aid in achieving this goal specifically. It also may provide other avenues for controlling cost and maximizing quality of life, by minimizing expensive interventions that improve neither the quality of care nor the quality of life at the end of life. Reducing the proportion of health care resources expended at the end of life might liberate resources that could then be used in other areas of need.

Third, several sociodemographic characteristics of our patients were associated with hospice use. In our multivariable model, white ethnicity, lower Charlson comorbidity index, higher socioeconomic status, and partnered relationship status were associated with increased hospice use. Retrospective analyses of population-based data reveal conflicting findings about sociodemographic variation in hospice use. Virnig et al¹² found no sociodemographic predictors of hospice use, but Lackan et al⁷ noted variations in age, marital status, insurance type, cancer type, education level, income level, race, and sex. The analysis by Lackan et al⁷ demonstrated decreasing variation over time for subgroups defined by type of insurance, marital status, urban residence, and income, but increasing variation over time as a function of age and cancer type. Other analyses suggest that men who have never been married are less likely to enroll in hospice.⁸ We were surprised to find an inverse relationship between Charlson comorbidity index and hospice referral. It is possible that men who were sicker with multiple illnesses had more difficulty coping in a home setting with hospice and therefore did not get referred; this would be of interest for further study.

Several authors have expressed concern over misuse of hospice to manage acute complications in the last few days before death, as opposed to more appropriate use earlier in the end of life arc.^7,9,15 That individuals with more comorbidity were less likely enroll in hospice suggests that they were not enrolled with the goal of palliating pain as death approached, although nuances in referral patterns stratified by comorbidity remains an intriguing area for future study. Inappropriate variation in use of medical treatments is considered poor-quality care,⁴⁰ and identifying subgroups that use hospice resources less frequently may help target interventions aimed at increasing appropriate hospice use.

Our findings are limited by several methodological considerations. First, because our sample was restricted to Medicare enrollees older than 65 years, our findings may not apply to younger individuals. However, SEER-Medicare data are representative of the US population and allow valuable temporal analyses that are not feasible in many other data sets. Similarly, although many men are diagnosed as having prostate cancer at younger ages, the majority of men dying of prostate cancer are of Medicare age.²¹ Second, our analysis missed all hospice costs paid out of pocket, although this accounts for less than 20% of total hospice expenditures.¹³ The lost opportunity cost as well as the actual out of pocket cost was not accounted for in our analysis. Third, we could not incorporate patient preference for end-of-life care into our analysis. Prospective studies to examine patient preferences for hospice care at the end of life would be invaluable. Fourth, use of death certificates to establish prostate cancer as the cause of death risks attribution bias, although death certificates for men dying of prostate cancer have been shown to be accurate. We captured men dying of prostate cancer, not men with prostate cancer, but could not ascertain the acute reason leading to their final hospital admission or death.⁴¹ Fifth, by using SEER-Medicare claims data we did not capture all individuals dying of prostate cancer, but SEER-Medicare claims have been shown to be representative of trends for the overall population. Sixth, our model did not examine the temporal relationship between hospice use and high-intensity care, and our results reflect associations, not causation. Elucidating the aforementioned temporal relationships would be important in improving our understanding of end-of-life care for men with prostate cancer. Although we did not prove causality, we believe that the marked differences in the care received by men who did and did not enroll in hospice are important to note and should be further examined. Seventh, during the study period, standards of care regarding chemotherapy for advanced prostate cancer changed. Specifically, the benefit of docetaxel for those with good performance status was not fully appreciated until late in our study period.