Characteristics Associated with In-Hospital Death among Commercially Insured Decedents with Cancer

Introduction

For patients with cancer, preference-concordant end-of-life care is an important and widely accepted indicator of care quality. When asked, most respondents identify home as the preferred place of death in the setting of advanced cancer,^1–3 and recent data suggest that in-hospital deaths are decreasing ⁴ and are less common in the United States than in other countries. ⁵ Nevertheless, gaps between preferred and observed patterns of end-of-life care persist for patients with cancer, with implications for quality of life, mental health of surviving caregivers, and family perceptions of end-of-life care quality.^6,7 In addition, in-hospital deaths in cancer patients are costly. For all of these reasons, monitoring the proportion of in-hospital deaths among decedents with cancer has been proposed as a systems-level quality measure to guide efforts for improving end-of-life care in patients with cancer. ⁸

Although place of death has been studied among U.S. cancer patients treated by the Medicare program,^4,5 less is known about the incidence and predictors of in-hospital death in decedents younger than age 65 (who account for 31% of U.S. cancer deaths ⁹ ). In a 2002 study, Bruera et al. examined predictors of in-hospital death among patients seen at the M.D. Anderson Cancer Center (inclusive of patients with both public- and private-payer health insurance). In that analysis of 1993 decedents, 52% died in an acute hospital setting. Risk of in-hospital death was increased for patients with hematologic malignancy, black race, and non-Medicare payer sources. ¹⁰ More recently, Wright et al. reported information regarding place of death for patients enrolled in the multi-institutional Coping with Cancer study. Thirty-four percent of 333 patients died in acute-care hospitals (including 8% in an intensive care unit [ICU]). End-of-life discussions with physicians and better terminal illness understanding were associated with decreased likelihood of in-hospital death, whereas the ongoing receipt of palliative chemotherapy (at a median of four months before death) was associated with increased likelihood of in-hospital death.^6,11 Our own group recently reported that 29% of decedents from a cohort of commercially insured cancer patients experienced in-hospital death, and that in-hospital deaths were the most common in patients with hematologic malignancy. ¹² Taken together, these and other studies suggest that place of death is influenced by both demographic and clinical factors.

In the present study, we investigate the association between in-hospital death and multiple clinical and demographic characteristics. Our objectives in pursuing this analysis were to enhance our understanding about the characteristics of cancer patients and their clinical care that are associated with in-hospital death, and to facilitate clinically targeted efforts to improve end-of-life care for patients with cancer.

Materials and Methods

Results

We identified 904 commercially insured patients who received outpatient cancer care at our center and who died during the study period. The median age at death was 59 years; additional demographic and clinical characteristics are shown in Table 1. In-hospital deaths were observed in 254 of 904 decedents (28.1%).

Only cancer type and time from cancer diagnosis to death were significantly associated with in-hospital death in the bivariate analysis (Table 1). In-hospital death was the most common in patients with hematologic malignancy, who had a 2.57 times increased risk when compared with the reference group of patients with gastrointestinal cancer (95% CI 1.91–3.45). A similar RR was observed for patients with hematologic malignancy when compared with the larger reference group of all patients with solid tumors (RR 2.51, 95% CI 2.06–3.06). In-hospital death was also more common among patients dying within six months of cancer diagnosis, as compared with patients diagnosed two or more years after diagnosis (RR 1.41, 95% CI 1.02–1.95). However, the association between time from cancer diagnosis to death and in-hospital death was nonmonotonic, as the lowest rates of in-hospital death were observed for patients dying one to two years after cancer diagnosis (RR 0.61, 95% CI 0.41–0.91). We found no statistically significant association of in-hospital death with age, sex, marital status, race/ethnicity, or travel distance from place of residence to the cancer center.

In addition to clinical characteristics, we also evaluated the association of in-hospital death with three end-of-life outcome measures. We found significant associations of in-hospital death with each of the three measures, including >1 ED visit in the last 30 days of life, enrollment in hospice before death, and receipt of chemotherapy in the last 14 days of life (Table 1). Hospice enrollment exhibited the strongest association with in-hospital death; 537 of 854 decedents with evaluable hospice claims enrolled in hospice before death (62.9%), and 25 of these patients (4.7%) experienced in-hospital death. Conversely, 214 of 317 decedents who died without receiving hospice services (67.5%) experienced in-hospital death, corresponding with a 14.5 times increased risk of in-hospital death among hospice nonenrollees (95% CI 9.81–21.4).

We also performed a multivariable analysis of the association of baseline patient characteristics with in-hospital death (Table 2), as well as separate adjusted analyses of the association of in-hospital death with hospice enrollment and time from cancer diagnosis to death (reported in the text). The association of hematologic malignancy with in-hospital death persisted in the multivariable analysis (ARR 2.60, 95% CI 2.12–3.19). There was no association of age, race/ethnicity, marital status, or travel distance to the cancer center with in-hospital death in the adjusted analysis. Nonenrollment in hospice remained strongly associated with in-hospital death after adjustment for baseline patient characteristics (ARR 13.3, 95% CI 8.94–19.8). The associations of time from cancer diagnosis to death were also persistent in adjusted analysis, with increased in-hospital deaths among patients dying within six months of cancer diagnosis (ARR 1.54, 95% CI 1.11–2.15)—and fewer in-hospital deaths among patients dying at one to two years from cancer diagnosis (ARR 0.62, 95% CI 0.41–0.93)—compared with the reference group of patients surviving more than two years after cancer diagnosis.

Rates of hospice enrollment, stratified by patient characteristics, are shown in Table 3. Compared with the overall hospice enrollment rate of 62.9%, the lowest rates of hospice use were seen among patients aged 75 years and older (48.0% enrolled in hospice), and among patients with hematologic malignancy (32.7% enrolled in hospice).

In addition to identifying characteristics associated with in-hospital death, we also sought to describe the characteristics of terminal hospitalizations preceding in-hospital death. Of 254 in-hospital deaths, 175 (68.9%) occurred at hospital sites that are affiliated with our cancer center, and 79 deaths (31.1%) occurred at nonaffiliated hospitals. The median length of stay for these terminal hospitalizations was 7 days (interquartile range, 4–15 days). The most common hospital diagnoses for terminal hospitalizations were septicemia, respiratory failure, pneumonia, and renal failure. One hundred fourteen patients were admitted to an ICU during the hospitalization preceding in-hospital death, and 110 of these patients died in the ICU (12.2% of 904 decedents). Associations between clinical characteristics and death in the ICU were similar to the associations observed for in-hospital death (Supplementary Table S1; Supplementary Data are available online at www.liebertpub.com/jpm). Patients with hematologic malignancy had a 4.24 times increased risk of death in the ICU, compared with the reference group of patients with gastrointestinal cancer.

Discussion

In this study, we review the association of patient characteristics with in-hospital death among commercially insured decedents with cancer. In this cohort of 904 patients who died at a median age of 59 years, 28.1% died in an acute-care hospital. Both hematologic malignancy and time from cancer diagnosis to death were associated with in-hospital death in the bivariate analysis; whereas age, sex, race/ethnicity, marital status, and travel distance from the cancer center were not associated with the study outcome. All three end-of-life outcome measures tested were strongly associated with in-hospital death. Among these measures, hospice enrollment exhibited the greatest magnitude of association; compared with hospice enrollees, nonenrollees had a 14.5 times increased risk of in-hospital death. All significant associations persisted in analyses that adjusted for baseline patient characteristics.

Notwithstanding differences in patient populations, the proportion of in-hospital deaths reported in our study is roughly similar to proportions observed in other recent U.S. cohorts,^5,6,20,21 and it is substantially lower than figures of in-hospital death reported in older studies ¹⁰ and from other countries.^5,8 Our findings regarding the association of patient characteristics with in-hospital death are largely consistent with a prior systematic review, ² and the increased risk of in-hospital death associated with hematologic malignancy has also been previously reported.^10,12 Nearly half (43%) of all in-hospital deaths in our study occurred in the ICU, and hematologic malignancy again stood out as the baseline characteristic most strongly associated with that outcome.

With regard to the strong association between in-hospital death and hospice enrollment, the observed relationship supports prior findings that increased and timely use of hospice care is associated with fewer hospitalizations and fewer in-hospital deaths.^22–24 The 4.6% risk of in-hospital death that we observed among hospice-enrolled decedents in our cohort is slightly higher than the 2.3% rate recently reported in a large cohort of older, Medicare-insured hospice enrollees. ¹⁷ That study also reports that in-hospital deaths were less common in decedents cared for by hospices that routinely assess patients' preferred place of death.

Our findings regarding time from cancer diagnosis to death are nuanced, showing an inverse J-shaped relationship with in-hospital death. In both crude and adjusted analyses, the RR of in-hospital death declined sharply between the first six months after diagnosis and 1–2 years after diagnosis (from an RR of 1.41 to 0.61 in the crude analysis), before increasing again among patients surviving more than two years from diagnosis (RR 1.0). The same pattern of association was seen for the nested outcome of death in the ICU. Increased intensity of care associated with longer survival after cancer diagnosis was also reported in a recent analysis of Medicare enrollees, ¹⁸ and this finding may suggest that patients who become experienced with aggressive life-sustaining medical care (and their doctors) have difficulty identifying the point at which further aggressive care is no longer life-prolonging.

A strength of our study is the use of cancer center administrative and clinical data merged with commercial insurance claims to identify all end-of-life hospitalizations in our study population. This included 31% of in-hospital deaths occurring at hospitals that were unaffiliated with our cancer center. The key limitation of our study is that we did not assess patient preferences for place of death, and we cannot determine when the inpatient hospital may have been the preferred place of death. Furthermore, even when the hospital is not the preferred place of death, it may still be the patient's preferred site of care as he or she approaches death, as some patients may prioritize access to hospital-based care over achieving their preferred place of death. ²⁵ Additional limitations of our study include the small size of some analytic subgroups (especially among race/ethnicity subgroups and subgroups of cancer type) and the study population of patients drawn from a single tertiary cancer center.

How can we use these findings to improve end-of-life care for patients with cancer? Our findings further reinforce that physicians should discuss preferences for end-of-life care with all patients receiving treatment for advanced cancer, including the potential benefits of hospice care and patient preferences regarding place of death. These conversations are challenging, and efforts are ongoing to develop and evaluate a patient-centered “serious illness conversation guide” to assist physicians in end-of-life conversations (clinicaltrials.gov, NCT01786811). In addition, our findings highlight the need to learn more about the reasons that certain patient subgroups, such as patients with hematologic malignancy, experience particularly low rates of hospice referral and correspondingly high rates of in-hospital death. ²⁶ An improved understanding of the factors driving current patterns of end-of-life care will help identify the best patient-centered approaches for delivering high-quality care that aligns with patient preferences. Such care will maximize quality of life while also seeking to honor patients' preferences regarding their preferred site of death.