Pathway to Hospice: How Has Place of Care before Hospice Changed with the Growth of Hospice in the United States?

Abstract

Background:

Hospice use among Medicare decedents increased from 21.6% in 2000 to 51.6% in 2019. Whether this growth has been accompanied by more referrals to hospice directly from the community is not known.

Objective:

To assess trends in place of care before hospice enrollment.

Design:

Retrospective cohort from 2011 to 2018.

Subjects:

Medicare decedents age ≥66 years.

Measure:

Location of care before hospice enrollment in the last 90 days of life, defined as: the community with and without home health, short- or long-term nursing facility, or inpatient hospital. A county-level random effects model examined changes in enrollment from the community after adjusting for admitting diagnosis, age, race/ethnicity, sex, and Medicaid participation.

Results:

Among hospice enrollees (N = 7,650,933), 27.7% transitioned to hospice from the community, 31.8% transitioned from the hospital, and 10.1% transitioned after short- or long-term nursing facility stay. Rates of enrollment to hospice from the community remained stable from 35.1% in 2011 to 34.3% in 2018. After adjustment, the proportion enrolling in hospice from the community decreased by 1.2% (95% confidence interval −1.0% to 1.4%). Place of care before hospice enrollment in 2018 varied by hospice admitting diagnosis, with patients with cancer more likely to enroll from the community (39.5%) and patients with cerebrovascular accidents from the hospital (53.2%). Prior place of care varied by state, with Florida having the highest rate of the enrollment following hospitalization (47.8%).

Conclusion:

Despite the growth of hospice, the site of care before hospice enrollment has remained relatively stable and was strongly influenced by region.

Introduction

The Medicare hospice benefit (MHB) was enacted in 1982 to fund a model of care that would better support beneficiaries and their family at the end of life.¹ Although initial utilization of the benefit was slow growing, policy changes, including allowing enrollment of nursing home residents, fostered tremendous growth. The proportion of decedents in Medicare who received the hospital benefit increased from 21.6% in 2000 to 51.6% in 2019.² Despite the growth in the MHB, there are ongoing concerns about the pathway that decedents take to hospice enrollment, and whether decedents and their family are accessing the full benefits of hospice. High and intensive health care utilization before hospice enrollment, especially acute hospital stays, is associated with short hospice length of stay,³ which is associated with a lower perceived benefit of hospice by families of decedents.⁴ Among persons with a short hospice length of stay, 40% were hospitalized in an intensive care unit before hospice admission.⁵

Multiple other policy changes separate from the growth in the MHB may have shifted patterns of care at the end of life in the last decade, including increased focus on value-based payment models, public reporting of hospital mortality rates and 30-day hospital readmissions, and advances in medical treatment. With the growth of the MHB and other secular changes, it is unknown whether the patterns of utilization before hospice enrollment have changed over time, by illness, and by place. As public awareness of hospice has increased, are persons more likely to enroll in hospice directly from the community rather than after a hospital stay?

In this research, we examined secular trends in the place of care immediately before hospice enrollment among Medicare decedents between 2011 and 2018. We aimed to assess the following: (1) how has the place of care before hospice enrollment care changed between 2011 and 2018; (2) how does this vary by the leading causes of death based on hospice admission primary diagnosis, including cancer, chronic obstructive pulmonary disease (COPD), congestive heart failure (CHF), dementia, and cerebral vascular accident (CVA), or traumatic brain injury (TBI); and (3) how does this vary by geographic region of the country?

Methods

Data and study population

We used Medicare beneficiary enrollment data, Medicare claims data, and assessment data from home health services and nursing homes between 2011 and 2018 to examine patterns of care and utilization before hospice. We included beneficiaries who were 66 years and older at the time of death based on Medicare beneficiary enrollment data. An age cutoff of 66 was chosen so all decedents would have one year prior utilization data. All traditional Medicare deaths and Medicare Advantage (MA) deaths were included in this analysis. Location of care in the last 90 days of life were identified with the Residential History File (RHF),⁶ which compiles information from assessment data and Medicare Provider Analysis and Review (MEDPAR) files. The MEDPAR file includes hospital encounters for Medicare beneficiaries, including those in MA, as it is used to allow the Centers for Medicare and Medicaid Services (CMS) to calculate disproportionate share payments and indirect and direct medical education adjustment.⁷ These claims include hospitalizations of beneficiaries in MA with the exception of those hospitals that do not take disproportionate share payments or medical education dollars. Prior research found these hospitals account for only 8% of the all Medicare hospital discharges.⁸

At the time of this study, hospice was a carve-out for beneficiaries with MA. Thus, hospice claims included data from the traditional Medicare and MA populations. Medicare and assessment data were utilized under a CMS data use agreement, and the Institutional Review Boards of Brown University and Oregon Health and Science University waived informed consent.

Outcome measures

We identified the place of care before hospice enrollment using the RHF algorithm, which uses place and dates of service from Medicare claims, Minimum Data Set assessments, and Outcome Assessments and Information Set assessments. Given that we are most interested in transitions at the very end of life, we separately examined those who were first enrolled in hospice ≥90 days before death. Among those enrolled in hospice <90 days before death, we categorized place of care before hospice enrollment as: hospitals, which includes observation stays in the emergency room and specialty hospital stays; skilled nursing facility stays and long-term nursing home stays without skilled services; and residing in the community with and without Medicare skilled home health. Community includes persons in individual homes, group homes, or in assisted living facilities. Given that individuals may be referred to hospice from the hospital but have a delay in hospice enrollment, we considered the hospital to be the place of care before hospice enrollment for individuals who transferred from hospital to home health and then enrolled in hospice but died within three days of discharge from hospital, or individuals in a nursing facility or the community but who died within seven days after discharge from hospital.

Independent variables

Secular changes in place of care before hospice admission was examined using year of death. We also examined trends in primary hospice diagnosis defined as: cancer, COPD, CHF, dementia, CVA/TBI, and all other diagnosis. These diagnoses were determined from the hospice primary International Classification of Diseases (ICD)-9 and ICD-10 code. These diseases were selected as they are the leading causes of death and prior work has demonstrated that variation in disease trajectories may impact the pattern of hospice referral. Given regional variation in hospice enrollment, we assessed the state and county of residence as determined from the Medicare beneficiary summary file. Additional covariates included age at death, race/ethnicity, an indicator of end-stage renal disease, age, sex, race/ethnicity, and dual Medicaid-eligible status. Race/ethnicity was based on reports to the security administration and augmented by Research Triangle Institute.

Statistical analysis

We first assessed trends in the place of care before hospice enrollment by year, primary hospice diagnosis, and state in the place of care before hospice enrollment using descriptive statistics. Our main dependent variable is a dichotomized variable for place of care before hospice enrollment that is defined as community with or without home health versus all other locations (hospital, skilled nursing stay, long-term nursing facility stay). We used that dichotomized measure for place of care before hospice enrollment to examine variation by state. Because of the large sample size, small differences will be statistically significant and therefore the focus in interpreting these results should be on the magnitude of association. We then used a multivariable logistic regression model with county-level random effects to assess characteristics associated with enrolling in hospice from the community (vs. all other settings) while accounting for clustering of decedents within geographical regions. This model included the year of death, primary hospice diagnosis, indicator of end-stage renal disease, age, sex, race/ethnicity, and dual Medicaid-eligible status. We report the results using the adjusted average marginal effect that provides a difference in predicted probability. Analyses were carried out in Stata 17.0.⁹

Results

Between 2011 and 2018, 15,203,083 Medicare decedents aged 66 years and older died and the proportion of hospice use at time of death increased from 47.1% in 2011 to 53.2% in 2018. The proportion of beneficiaries with continuous hospice stays of 90 or more days increased from 16.1% in 2011 to 18.2% in 2018. Table 1 provides decedents based on sociodemographic characteristics and primary hospice diagnosis for those persons who enrolled in hospice in the last 90 days of life. With the exception of the primary diagnosis, the demographic characteristics of hospice users remained relatively stable. Over time, dementia as the primary hospice diagnosis increased from 18.0% in 2011 to 34.3% in 2018 with reductions in all other leading causes of death except CHF.

Table 1.

Sample Description of Medicare Decedents Who Enrolled in Hospice in the Last 90 Days of Life Between 2011 and 2018^a

	All years (n = 6,330,633), %	2011 (n = 703,284), %	2013 (n = 754,872), %	2016 (n = 823,384), %	2018 (n = 884,803), %
Male	42.8	41.9	42.5	43.0	43.4
Dual	21.4	21.7	21.6	21.5	20.6
Race: White	85.3	86.7	85.9	84.7	84.0
Race: Black	7.3	7.1	7.2	7.5	7.6
Race: Hispanic	4.9	4.3	4.6	5.1	5.4
Race: Asian	2.5	2.0	2.2	2.7	3.0
Any hospice use
<90 days
90+ days
Hospice admitting primary diagnosis
Cancer	28.4	30.6	29.8	28.8	27.5
Dementia	26.7	18.0	21.9	32.8	34.3
COPD	8.7	6.1	6.5	6.7	6.6
CHF	10.6	7.7	8.9	12.0	12.8
CVA/TBI/stroke	3.6	3.9	4.0	3.3	3.1
All other diagnoses	27.3	34.8	30.7	22.9	22.5

^a Analyses was carried out on all years, but we only included 2011, 2013, and 2016, and 2018 in this table.

CHF, congestive heart failure; COPD, chronic obstructive pulmonary disease; CVA, cerebral vascular accident; TBI, traumatic brain injury.

Among persons who enrolled into their final hospice stay in the last 90 days of life, Figure 1 provides the last place of care for persons who enrolled in hospice in the last 90 days of life finding that hospital and community without home health services were the predominant place of care before hospice enrollment, with little variation over time. Figure 2 characterizes the place of care before hospice enrollment in 2011 and 2018 among leading causes of hospice death. Figure 2 in contrast to Figure 1 reports all decedents that utilized hospice services including those that enrolled in hospice for 91 days or longer. Patients with dementia, heart failure, and COPD had the highest proportions of persons in hospice for at least 90 continuous days in hospice. In 2018, about one in four beneficiaries joined hospice only after an acute care hospital stay. For patients with CVA/TBI, the predominant place of care before hospice enrollment was the acute care hospital, reflecting the catastrophic nature of the diagnosis. This did decrease from 57.3% (2011) to 51.8% (2018). Among those with cancer dying in 2018, the two most common locations of care before hospice enrollment were the community (40.2%) and the acute care hospital (31.8%). The trend among decedents with cancer was a slight increase in persons enrolling in hospice after a hospital stay and a slight decrease in enrollment from the community (Fig. 2a, b).

FIG. 1.

National trends in pathway to hospice from 2011 to 2018 for those decedents that entered hospice in the last 90 days of life.

FIG. 2.

(a) Pathway to hospice by disease in 2011 for all decedents that utilized hospice services. (b) Pathway to hospice by disease in 2018 for all decedents who utilized hospice services. *Note in these graphs those decedents who used hospice services, including those persons in hospice longer than 90 days, whereas Figure 1 documents only those persons who enrolled in final hospice stay in the last 90 days of life.

The pathway to hospice was associated with hospice length of stay for those enrolling in hospice in the last 90 days of life. Persons entering hospice after an acute care hospital stay had a median hospice length of stay of three days (mean 5.9 days), whereas those enrolling from the community had a median length of stay of eight days (mean of 10.5 days).

Although nationally the pathway to hospice has changed little in this nine-year period of time, there is striking geographic variation by state. Decedents in Iowa and Rhode Island were the least likely to enter hospice from the community (29.0% and 30.6%, respectively), whereas Vermont, Oregon, Idaho, and Alaska had >60% of decedents enrolling in hospice from a community setting with or without prior home health services. Figures 3 and 4 provide two maps of state variation in place of care before hospice in the last 90 days of life. Figure 3 provides the proportion of decedents who entered hospice from the community, whereas Figure 4 provides those who entered through an acute care hospital. Vermont and Utah had the lowest rates of decedents entering hospice from the hospital (20.2% for Vermont; 24.3% for Utah), whereas Florida had nearly one in two entering from an acute care hospital (47.8%).

FIG. 3.

State variation in hospice enrollment from the community with or without prior home health services.

FIG. 4.

State variation in the hospital as the place of enrollment to hospice.

Table 2 provides the absolute adjusted marginal difference in probability of entering hospice before being in the community versus all other places of care, including all covariates and county-level random effects. For example, decedents in 2018 versus 2011 had an adjusted marginal reduction in the proportion enrolling in hospice from the community versus other settings of 1.2% (95% confidence interval, −1.0% to −1.4%). The adjusted marginal difference for race and ethnicity were small, whereas decedents with noncancer diagnosis were less likely to enter hospice from the community. The largest difference was in those with CVA/TBI or stroke who had 30.4% lower adjusted proportion of hospice enrollment from the community versus other settings. In addition, dual-eligibles had 22.1% lower adjusted proportion of hospice enrollment from the community versus other settings compared with nonduals. Overall, county explained 41.0% of the variance.

Table 2.

Adjusted Marginal Difference in Probability of Entering Hospice from the Community in the Last 90 Days of Life from Multivariate Logistic Random Effects Model Clustered at the County

	Adjusted marginal difference	95% confidence intervals
Year of death:2011	Reference
Year of death: 2012	−0.38	−0.23 to −0.54
Year of death: 2013	−0.50	−0.34 to −0.66
Year of death: 2014	−0.87	−0.70 to −1.04
Year of death: 2015	1.27	1.08 to 1.45
Year of death: 2016	−1.48	−1.31 to −1.66
Year of death: 2017	−1.66	−1.47 to −1.85
Year of death: 2018	−1.24	−1.06 to −1.44
Age	0.10	0.09 to 0.10
Male	0.54	0.46 to 0.62
Dual status	−22.14	−22.0 to −22.3
Race and ethnicity
White	Reference
Black	−0.38	−0.22 to −0.54
Asian	0.45	0.18 to 0.72
Hispanic	2.83	2.63 to 3.02
Native	−1.03	−0.30 to −1.76
Hospice primary diagnosis
Cancer	Reference
Dementia	−9.89	−9.78 to −10.00
COPD	−6.71	−6.57 to −6.86
CHF	−12.01	−11.87 to −12.14
CVA/TBI/stroke	−30.41	−30.18 to −30.65
End-stage renal disease	−3.40	−3.14 to −3.67
All other diagnoses	−18.19	−18.08 to −18.30

Discussion

Despite the overall growth in hospice utilization and other important secular changes, there have been relatively minor changes in the place of care of decedents before hospice enrollment. Overall, there was a slight increase from 2011 to 2018 in hospice enrollment after an acute hospitalization, whereas rates of hospice enrollment from the community remained unchanged. The most notable exception to these consistent patterns is the increased enrollment of persons with dementia diagnosis with longer hospice lengths of stay. The proportion of hospice enrollees receiving hospice care for >90 days increased from 16.1% in 2011 to 18.2% in 2018, with dementia being the predominant diagnosis among those with long hospice stays. Compared with other leading causes of death, hospice enrollees with CVA demonstrated more variation in place of care before hospice enrollment in recent years: among this population, hospital as the place of care before hospice enrollment decreased from 57.3% to 51.8%. Earlier research in four communities participating in the Atherosclerosis Risk in Communities Study found that prior hospital use is associated with increased likelihood of short hospice length of stay³ and bereaved family reporting a lower rating of the quality of care. Understanding these patterns is important to shaping interventions to improve quality of care and reflecting on the impact of policy.

Between 2011 and 2018, important policy changes occurred, including the expansion of hospital-based and community-based palliative care teams, growth of value-based payment models that provided incentives to control costs, continued public reporting of hospital mortality and 30-day readmission rates, and expansion of hospice services. Despite these changes, we did not detect meaningful increases in the proportion of patients who were referred to hospice directly from the community.

In part, the place of care before hospice reflects the functional trajectory in the last year of life. Gill and colleagues previously reported that trajectories of functional decline in the last year of life varied among community-dwelling older persons with the same cause of death.¹⁰ The main exceptions were dementia, categorized by patterns of persistent severe disability in the last year of life, and CVA/TBI/stroke, categorized by sudden onset of a functional decline after a catastrophic event. Our findings suggest that, with the exception of CVA/TBI, there is not a predominant place of care before hospice for any individual disease. The common hospice referral pattern for patients with solid tumor cancer is based on disease progression or toxicity of treatment resulting in referral to hospice from the community after a decision is made to stop cancer treatment. These persons experience a rapid functional decline. However, there is heterogeneity within diseases: in solid tumor cancers, sudden death from massive pulmonary embolism or intracranial bleed represents a catastrophic event that results in hospitalization, followed by hospice enrollment just before death.

The striking variation by geographic region suggests that geographic variation may be a more influential factor than illness trajectory. Differences by county accounted for 41.0% of the variance in community referrals to hospice. This suggests that differences in culture, local state regulations, and local market characteristics are important to determining the pathway that individuals take to hospice. Further research is needed to disentangle what regional characteristics and policies are most important in driving this variation.

Limitations

Our study has limitations. An important caveat is that the focus of this article is on the place of care before hospice enrollment in the last 90 days of life. Given that we focus on transitions potentially associated with short hospice stays or end-of-life transitions, this is appropriate, and yet does not capture all hospice enrollment. There should be caution in the interpretation of the map of state variation in referrals to hospice from the community versus hospital in that the results are only among those decedents that utilized hospice services and does not capture rates of hospice use. Thus, Alaska, which we reported as having a high number of decedents enrolled in hospice from the community, is also one of the states with the lower rate of hospice enrollment in the United States.

A high proportion of decedents enrolled in hospice from the community need to be interpreted with other potential markers of the quality of hospice care, including the region's proportion of decedents using hospice as well as short hospice length of stays. Furthermore, these results include only those persons with Medicare coverage. These results may not be generalizable to those who either have no insurance, Medicaid only, or private commercial insurance. We relied on hospice claims submitted to CMS of the primary diagnosis, which may not be accurate. Finally, our focus has been on national trends that may miss important changes occurring in smaller geographic regions. This should be a consideration for future research.

Conclusion

We described patterns of the place of care before hospice enrollment at the national and state level from 2011 to 2018. During this time period, there was a continued increase in hospice use among Medicare decedents, especially those persons with long hospice stays with dementia, but only small changes in the place of care before hospice enrollment for those enrolling in the last 90 days before death. Understanding patterns of hospice referral is important to the design of payment policy to ensure that Medicare decedents receive high quality of care at the close of life.

Disclaimer

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health (NIH).

Funding Information

The sponsor did not have any role in the design, conduct, writing, or review of the submitted version of the article. Research in this article was supported by the National Institute on Aging of the NIH (2P01AG027296-11).

Author Disclosure Statement

No competing financial interests exist.