Effects of Medicare advantage on patterns of end‐of‐life care among Medicare decedents

Abstract

Objective

To examine the effects of Medicare Advantage (MA) enrollment on patterns of end‐of‐life care.

Data sources

We used data from the Master Beneficiary Summary File, the Medicare Provider Analysis and Review, hospice claims, the Minimum Data Set, the Outcome and Assessment Information Set, the Area Health Resources File, and Geographic Variation Public Use File for 2012–2014.

Study design

To address selective enrollment into MA, we exploited a discontinuity in payment rates by county population (urban floor payments) as an instrument.

Data collection/extraction methods

We identified Medicare beneficiaries continuously enrolled in MA or TM during their last year of life between 2012 and 2014 using Medicare administrative data.

Principal findings

We did not find evidence that MA enrollment led to a change in hospital admissions in the last 30 days of life, but MA enrollment decreased hospital as the site of death by 11.0 (95% CI: −13.9 to −8.1) percentage points. Once hospitalized, however, MA enrollment increased use of intensive care by 6.7 (95% CI: 0.3 to 13.1) percentage points and non‐invasive mechanical ventilation by 9.2 (95% CI: 5.5 to 12.9) percentage points. MA enrollment increased hospice use by 6.2 (95% CI: 2.3 to 10.1) percentage points at time of death and 7.7 (95% CI: 3.8 to 11.6) percentage points in the last 30 days of life. Particularly, MA enrollment increased hospice admissions among those who were admitted to the hospital within 30 days prior to hospice admission by 18.8 (95% CI: 13.8 to 23.8) percentage points. However, MA enrollment decreased hospice admissions among those who were admitted to home health within 30 days prior to hospice admission by 18.6 (95% CI: −21.9 to −15.2) percentage points.

Conclusions

MA plans may improve end‐of‐life care by reducing hospital death while also improving access to hospice, especially among recently hospitalized persons.

What is known on this topic

• Prior research has found that compared to traditional Medicare enrollees, Medicare Advantage (MA) enrollees had lower hospital admissions and higher hospice use at the end of life.

• However, the prior research did not fully account for selective enrollment into MA, possibly leading to biased estimates.

What this study adds

• This study examined the effects of MA on end‐of‐life care patterns among decedents while accounting for selective enrollment,

• MA enrollment decreased hospital as the site of death, however once hospitalized MA was not associated with decreased care intensity.

• We also found that MA enrollment increased hospice use, but only among those who were admitted to the hospital within 30 days prior to hospice admission.

1. INTRODUCTION

Research has shown that costs for end‐of‐life care are substantial. Mean per‐person Medicare spending for decedents was estimated to be $34,529 in 2014, more than triple the mean spending for non‐decedents ($9121). ¹ Despite these substantial costs, decedents often experience low‐quality care. For example, decedents suffer from poor palliation of symptoms, burdensome health care transitions, and unmet needs. ² , ³ , ⁴ This highlights the importance of offering patient‐ and family‐centered care at the close of life. Providing palliative care and preventing utilization that is inconsistent with patients' goals of care can improve quality of life while reducing unwanted health care costs. ⁵ , ⁶

Medicare Advantage (MA), the alternative to traditional fee‐for‐service Medicare (TM) run by private insurance companies, has financial incentives to decrease health care use overall, ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ with particular potential for saving in high‐cost settings such as the end of life. MA plans are paid on a capitated basis and thus are incentivized to manage the care of enrollees while reducing unnecessary treatments. This enables MA plans to improve care delivery by changing the setting and intensity of care. Evidence shows that MA enrollees are more likely to be transitioned from inpatient to outpatient settings than TM enrollees. ⁹ , ¹⁰ Also, MA enrollees are less likely to have high‐intensity care and specialist visits than TM enrollees. ⁹ , ¹⁰ However, a concern for both these comparisons and policymakers is that MA plans may selectively enroll healthier beneficiaries. MA plan payments have traditionally had only minimal adjustments for beneficiaries' clinical characteristics, resulting in overpayments for healthier beneficiaries and underpayments for sicker beneficiaries. Evidence suggests that healthier beneficiaries were more likely to enroll in MA than TM. ¹² , ¹³ , ¹⁴ Understanding how MA delivers end‐of‐life care is of important policy relevance as the proportion of Medicare decedents in MA has risen. Nearly one in three Medicare decedents is cared for by MA, and some states have one‐half of all Medicare deaths enrolled in MA. ¹⁵

Existing evidence suggests that MA may have higher efficiency of care delivery at the end of life than TM. Compared to TM enrollees, MA enrollees are less likely to be hospitalized in the last month of life and are more likely to use hospice care. ¹⁶ , ¹⁷ , ¹⁸ However, findings from prior research are limited in two aspects. First, MA enrollees tend to be healthier than TM enrollees, suggesting selective enrollment into MA and making direct comparisons difficult. ¹⁴ As prior research only accounted for differences in observable factors, their findings may be subject to bias related to residual confounding due to unobservable factors. Second, little is known about detailed patterns of end‐of‐life care, especially for hospice settings. Prior research found that compared to TM enrollees, MA enrollees were less likely to have burdensome health care transitions in hospital and hospice settings. ¹⁹ , ²⁰ While MA may increase hospice use among Medicare decedents, less is known about the underlying pathways to hospice admission. Also, prior research found that family and friends of those in MA reported lower quality end‐of‐life care than family and friends of those in TM. ²¹ This may indicate that MA decedents enroll in low‐quality hospice.

In this study, we examined the effects of MA enrollment on patterns of end‐of‐life care in hospital and hospice settings among Medicare decedents. Specifically, we conducted several analyses. First, we investigated the effects of MA enrollment on the care settings. Then, we focused on the hospital setting and examined the effects of MA enrollment on care intensity during hospitalization and potentially burdensome transitions. Finally, we focused on the hospice setting and investigated the effects of MA enrollment on pathways to hospice. We conducted our analysis using a quasi‐experimental instrumental variable (IV) method to account for selective enrollment into MA.

2. METHODS

2.1. Data and study population

We used data from multiple sources between 2012 and 2014. Our primary data sources were Medicare enrollment and claims data. Specifically, the Medicare Master Beneficiary Summary File (MBSF) was used to obtain enrollee characteristics. The Medicare Provider Analysis and Review (MedPAR) and hospice claims data were used to identify end‐of‐life care in acute care hospitals and hospice settings. The hospice claims data provide data on all MA and TM enrollees because hospice care is currently carved out of the MA benefits and is reimbursed on a fee‐for‐service basis. However, the MedPAR includes claims from MA enrollees who were admitted to hospitals that receive Medicaid Disproportionate Share Hospital (MDSH) payments or graduate medical education hospitals, which accounts for 92% of all Medicare discharges from MA‐reporting hospitals. ¹⁰ Since MA encounter data was not available for our entire study period, the Minimum Data Set (MDS) and the Outcome and Assessment Information Set (OASIS) data were used to identify stays in the nursing home and home health settings. Finally, the Area Health Resources File and Geographic Variation Public Use File were used to obtain county‐level demographic and socioeconomic data. We linked enrollee‐level information using unique personal identifiers. We linked county‐level information using area identifiers (Social Security Administration state and county codes and/or Federal Information Processing Standard state and county codes).

Using these data, we identified Medicare beneficiaries (aged 66 years and older) continuously enrolled in MA or TM during their last year of life and died between 2013 and 2014. We excluded those who moved from one county to another at any point during the last year of life (N = 189), those whose original Medicare eligibility was attributable to end‐stage renal disease (N = 110,812), and those who lived in the U.S. overseas territories (N = 31,844). We also excluded those who were admitted to hospitals that do not receive MDSH payments or medical education credits as these hospitals may not completely report data for MA enrollees (N = 27,077). The unit of analysis is the decedent.

2.2. Outcomes

We had four types of outcomes: care setting, care intensity during hospitalization, potentially burdensome health care transitions, and pathways to hospice. Care setting was defined as the hospital as the site of death and whether the decedent was receiving hospice services. Among those hospitalized, we examined whether the person receives care in an intensive care unit (ICU), the use of invasive mechanical ventilation, and the use of non‐invasive mechanical ventilation. Invasive mechanical ventilation and non‐invasive mechanical ventilation were identified based on the International Classification of Disease (see Table A for specific codes used). Based on prior research, four measures of potentially burdensome transitions were created. Finally, we examined the utilization prior to hospice enrollment as the last place of care within 30 days prior to hospice admission being a hospital, nursing home, or home health agency.

2.3. Instrumental variable

We adopted an IV approach to estimate the effect of MA enrollment on end‐of‐life care among Medicare decedents while accounting for selective enrollment in MA. Following prior research, ²² , ²³ we took advantage of a discontinuity in the rules governing MA payments to health plans that give greater payments to plans operating in counties in Metropolitan Statistical Areas (MSAs) with populations of 250,000 or more (described as “urban floor” counties) compared to smaller populations. The difference in payment rates creates a greater incentive for plans to enroll more beneficiaries in MA in counties just above versus just below the cutoff. The validity of our IV approach depends on the assumption that residency in counties in MSAs with populations of 250,000 or more is not correlated with unmeasured determinants of end‐of‐life care patterns. To make this assumption as plausible as possible, we followed prior research and limited it to TM and MA enrollees in counties in MSA with populations of 100,000 and 400,000. ²² , ²³ We verified that this bandwidth was appropriate by conducting a data‐driven bandwidth selection procedure. ²⁴ Our IV estimates represent the local average treatment effect (LATE) of MA enrollment on end‐of‐life care, which is the average causal effect for those whose enrollment status is sensitive to the value of the instrument (those whose enrollment in MA was influenced by county of residence), also known as “compliers.” To assess the instrument strength, we tested the association with MA enrollment and then examined Kleibergen Paap F statistics, where a value greater than approximately 100 indicates a strong instrument. ²⁵

2.4. Control variables

To control for differences in individual‐level characteristics, we included age, sex, race/ethnicity, dual eligibility for Medicare and Medicaid, original Medicare eligibility due to disability, (unnormalized) Hierarchical Condition Category (HCC) risk scores, and the month of death. We used the Center for Medicare and Medicaid Services (CMS)‐HCC risk adjustment model to estimate HCC risk scores, but only used diagnosis codes from the MedPAR and hospice claims data. To control for county‐level characteristics that might be correlated with end‐of‐life care patterns, we included population size, percentage of those older than 65 years old, percentage of dual‐eligible Medicare beneficiaries, county‐level mean HCC risk score, median household income, percentage of residents with incomes below poverty, unemployment rates, number of hospital beds, medical doctors, primary care physicians, specialist physicians per 1000 residents, and number of deaths in hospice per 1000 Medicare beneficiaries in a prior year.

2.5. Statistical analysis

We first summarized sample characteristics of Medicare decedents in MSAs with a population between 100,000 to 250,000 and a population between 250,000 to 400,000. We then estimated standardized differences to quality the differences in characteristics between the two groups. Next, we estimated unadjusted outcomes by insurance coverage and residence in urban floor counties. To see whether patterns of end‐of‐life care were similar between those included and not included in our analysis, we estimated unadjusted outcomes for those in counties in MSAs with populations of 100,000 to 400,000, those in counties in MSAs with populations of 100,000 or less, and those in counties in MSAs with populations of 400,000 or more.

We then performed two‐stage least squares regression while controlling for individual‐level and county‐level characteristics described above. In the first stage, we conducted the following form of analysis to obtain the estimated likelihood of deceased beneficiaries being enrolled in MA plans due to residing in counties in MSAs with populations of 250,000 or more:

$${\mathrm{MA}}_{\mathrm{ict}}=\alpha +{\text{Floor}}_c+{\mathrm{X}}_{\mathrm{ict}}\mathrm{\gamma }+{\mathrm{V}}_{\mathrm{ct}}\delta +{\mathrm{\epsilon }}_{\mathrm{ict}}$$

where $i$ indexes beneficiaries, $c$ indexes counties, and $t$ indexes year, and $\mathrm{MA}$ is a binary indicator for 12‐month continuous enrollment in MA. $\mathit{\text{Floor}}$ is a binary indicator for residence in counties in MSAs with populations of 250,000–400,000. $X$ is a vector of individual‐level characteristics. $V$ is a vector of county‐level characteristics. $\epsilon$ is an error term that we allow to be arbitrarily correlated within each county.

In the second stage, we conducted the following form of analysis to estimate the association between estimated enrollment in MA from the first stage and the outcomes of interest:

$${\mathrm{Y}}_{\mathrm{ict}}=\mathrm{\zeta }+{\widehat{\mathrm{MA}}}_{\mathrm{ict}}+{\mathrm{X}}_{\mathrm{ict}}\eta +{\mathrm{V}}_{\mathrm{ct}}\mathrm{\theta }+{\mathrm{ϵ}}_{\mathrm{ict}}$$

where $Y$ is an outcome measure for beneficiary $i$ in county $c$ at year $t$.

We conducted linear regression for all outcomes. For all analyses, we included year‐fixed effects and adjusted the standard errors for clustering within county.

As findings from IV analyses are only applied to the “complier” (those whose MA enrollment status is sensitive to residing in counties in MSAs with populations of 250,000 or more), we described the characteristics of the complier ²⁶ and compared to our full sample.

3. RESULTS

3.1. Sample description

Our sample included a total of 457,440 Medicare decedents were studied (see Table 1). We found that 28% of the decedents were long‐term nursing home residents. Standardized differences in individual‐level and county‐level characteristics around this population cut‐off were small (standardized differences less than 0.10). However, MSAs with populations of 250,000 to 400,0000 had higher median household income and more hospital beds per 1000 residents than MSAs with populations of 100,000 to 250,000. Outcomes and sample characteristics of Medicare decedents by insurance coverage and residence in urban floor counties are presented in Tables B and C. Unadjusted outcomes were relatively similar between those included and not included in our analysis (see Table D).

TABLE 1.

Sample characteristics of Medicare decedents between 2013 and 2014

	Beneficiaries from MSAs with 100 k‐400 k population
Variable	Below 250 k population	Above 250 k population	Standardized differences
Number of counties, N (%)	272	126
Number of beneficiaries, N (%)	265,510	191,930
Individual‐level characteristics
Age, N (%)
65–69	22,857 (8.6)	16,575 (8.6)	0.00
70–74	32,970 (12.4)	23,891 (12.4)	0.00
75–79	38,998 (14.7)	27,420 (14.3)	0.01
80–84	48,279 (18.2)	34,310 (17.9)	0.00
85+	122,406 (46.1)	89,734 (46.8)	−0.01
Female, N (%)	144,030 (54.2)	104,714 (54.6)	0.00
Race/ethnicity, N (%)
Non‐Hispanic White	241,979 (91.1)	175,788 (91.6)	−0.01
Non‐Hispanic Black	16,460 (6.2)	12,115 (6.3)	0.00
Hispanic	3239 (1.2)	1009 (0.5)	0.07
Non‐Hispanic Asian/Pacific Islander	899 (0.3)	965 (0.5)	−0.02
Non‐Hispanic American Indian/Alaska Native	1296 (0.5)	804 (0.4)	0.01
Other	1199 (0.5)	878 (0.5)	0.00
Dual eligibility for Medicare and Medicaid, N (%)	59,558 (22.4)	39,330 (20.5)	0.04
Medicare eligibility due to disability, N (%)	955 (0.4)	651 (0.3)	0.00
(Unnormalized) HCC risk score	1.8 (1.8)	1.8 (1.8)	0.00
County‐level characteristics
Population, mean (SD)	137574.0 (63397.7)	205848.6 (98927.0)	−0.82
Percent of 65 year and older, mean (SD)	154463.5 (42607.7)	162333.6 (41046.3)	−0.18
Percent of dual eligible Medicare beneficiaries, mean (SD)	21.6 (6.8)	20.8 (5.4)	0.13
HCC risk score, mean (SD)	1.0 (0.1)	1.0 (0.1)	−0.10
Median household income, mean (SD)	47081.5 (7226.7)	49100.8 (9340.5)	−0.24
Percent of residents with incomes below poverty, mean (SD)	17.2 (4.7)	16.6 (4.1)	0.14
Unemployment rate, mean (SD)	6.8 (2.8)	7.0 (1.6)	−0.11
Number of hospital beds per 1000 residents, mean (SD)	535.1 (401.0)	747.5 (491.2)	−0.47
Number of MDs per 1000 residents, mean (SD)	2.5 (2.3)	2.5 (1.5)	0.02
Number of primary care physicians per 1000 residents, mean (SD)	0.9 (0.5)	0.9 (0.4)	−0.04
Number of specialists per 1000 residents, mean (SD)	3.2 (2.4)	3.1 (1.7)	0.00
Number of deaths in hospice per 1000 Medicare beneficiaries, mean (SD)	9.2 (39.1)	8.7 (42.2)	0.01

3.2. Instrument validity

Across all outcomes, our instrument was statistically strong. We found that there were discontinuities in MA benchmarks and MA penetration in counties just above versus just below the cutoff (Figures B and C). Results from the first stage regression showed that Medicare decedents from MSAs with populations of 250,000 to 400,000 were 7.7 (95% CI: 7.5 to 8.3) to 8.3 (95% CI: 8.0 to 8.6) percentage points more likely to enroll in MA than those in MSAs with populations of 100,000 to 250,000 (Table E). The Kleibergen Paap F‐statistics were generally higher than 1700. Findings from ordinary least squares regression are presented in Table F.

3.3. Care setting

Our IV analyses showed that MA enrollment led to changes in a care setting for Medicare decedents (Table 2). MA enrollment did not lead to a significant change in admissions in the last 30 days of life but significantly decreased hospital deaths by 11.0 (95% CI: −13.9 to −8.1) percentage points. MA enrollment significantly increased hospice use by 6.2 (95% CI: 2.3 to 10.1) percentage points for hospice use at the time of death and 7.7 (95% CI: 3.8 to 11.6) percentage points for hospice use in the last 30 days of life. There were no significant changes in hospital days and hospice days in the last 30 days of life. Full regression results are presented in Table G.

TABLE 2.

Effect of MA enrollment on care setting at death and place of care during last 30 days of life among Medicare decedents

Outcomes	Effect of MA enrollment, percentage points a (95% CI)
Care setting at death (N = 457,440)
Hospital	−11.0 (−13.9 to −8.1)
Hospice	6.2 (2.3 to 10.1)
Place of care received during last 30 days of life
Hospital admission (N = 457,440)	−1.9 (−5.2 to 1.4)
Hospital days (conditional on hospitalization) (N = 184,173)	−1.1 (−3.5 to 1.2)
Hospice use (N = 457,440)	7.7 (3.8 to 11.6)
Hospice days (conditional on hospice use) (N = 235,500)	−2.5 (−5.6 to 0.5)

^a We estimated the effects of Medicare Advantage (MA) enrollment on end‐of‐life care using an instrumental variable (IV) approach. We exploited a discontinuity in the rules governing MA payments to health plans that give greater payments to plans operating in counties in Metropolitan Statistical Areas (MSAs) with populations of 250,000 or more. We used the Medicare Provider Analysis and Review (MedPAR) and hospice claims data to identify care setting at death and place of care during the last 30 days of life.

3.4. End‐of‐life care in Hospital

We found mixed evidence that MA enrollment increased care intensity once hospitalized (Table 3). Specifically, MA enrollment significantly increased ICU use in the last 30 days of life by 6.7 (95% CI: 0.3 to 13.1) percentage points, but also significantly increased use of non‐invasive mechanical ventilation in the last 30 days of life by 9.2 (95% CI: 5.5 to 12.9) percentage points. There was no significant change in the use of invasive mechanical ventilation in the last 30 days of life.

TABLE 3.

Effect of MA enrollment on care intensity during hospitalization and potentially burdensome transitions among Medicare decedents

Outcomes	Effect of MA enrollment, percentage points a (95% CI)
Care intensity during hospitalization (conditional on hospitalization) (N = 184,173)
ICU use during last 30 days of life	6.7 (0.3 to 13.1)
Non‐invasive mechanical ventilation during last 30 days of life	9.2 (5.5 to 12.9)
Invasive mechanical ventilation during last 30 days of life	−1.2 (−6.3 to 3.9)
Potentially burdensome transitions (N = 457,440)
Transition from hospital to hospice during last 3 days of life	5.3 (3.5 to 7.1)
Transition from hospital to nursing home during last 3 days of life	0.1 (−0.8 to 0.9)
Transition from hospital to another hospital during last 3 days of life	−1.1 (−1.7 to −0.5)
3+ transitions to hospital during last 90 days of life	−8.3 (−12.1 to −4.4)

^a We estimated the effects of Medicare Advantage (MA) enrollment on end‐of‐life care using an instrumental variable (IV) approach. We exploited a discontinuity in the rules governing MA payments to health plans that give greater payments to plans operating in counties in Metropolitan Statistical Areas (MSAs) with populations of 250,000 or more. We used the Medicare Provider Analysis and Review (MedPAR) to identify care intensity during hospitalization. We used the MedPAR, hospice claims, the Minimum Data Set (MDS), and the Outcome and Assessment Information Set (OASIS) data to identify potentially burdensome transitions.

The effects of MA enrollment on potentially burdensome transitions were mixed (Table 3). Specifically, MA enrollment significantly increased transitions from hospital to hospice in the last 3 days of life by 5.3 (95% CI: 3.5 to 7.1) percentage points. However, MA enrollment significantly decreased transitions from hospital to another hospital in the last 3 days of life and more than three hospitalizations in the last 90 days of life by 1.1 (95% CI: −1.7 to −0.5) percentage point and 8.3 (95% CI: −12.1 to −4.4) percentage points. MA enrollment did not lead to a significant change in transitions from hospital to a nursing home in the last 3 days of life.

3.5. Pathway to hospice enrollment

MA decedents were more likely to use hospice when they had an inpatient hospitalization within 30 days prior to hospice admission, suggesting that hospitalization served as a gateway to hospice (Table 4). Specifically, MA enrollment hospitalized increased hospice referral by 18.8% percentage points (95% CI: 13.8–23.8). However, MA enrollment significantly decreased hospice admissions among those who were admitted to home health by 18.6 (95% CI: −21.9 to −15.2) percentage points. There was no change in hospice admissions among those who were admitted to nursing home within 30 days prior to hospice.

TABLE 4.

Effect of MA enrollment on pathway to hospice among Medicare decedents

Outcomes	Effect of MA enrollment, percentage points or coefficient a (95% CI)
Pathway to hospice (30 days prior to hospice admission) (N = 237,742)
From hospital	18.8 (13.8 to 23.8)
From nursing home	2.1 (−0.8 to 5.0)
From home health	−18.6 (−21.9 to −15.2)

^a We estimated the effects of Medicare Advantage (MA) enrollment on end‐of‐life care using an instrumental variable (IV) approach. We exploited a discontinuity in the rules governing MA payments to health plans that give greater payments to plans operating in counties in Metropolitan Statistical Areas (MSAs) with populations of 250,000 or more. We used the MedPAR, hospice claims, the Minimum Data Set (MDS), and the Outcome and Assessment Information Set (OASIS) data to identify pathway to hospice.

3.6. Characteristics of the compliers

There were differences in sample characteristics between the compliers and the full sample (Table 5). Compared to the full sample, the compliers were less likely to be older than 85 years (46.3% vs. 34.8%), to be non‐Hispanic White (91.3% vs. 87.8%), and more likely to be dual‐eligible for Medicare and Medicaid (29.4% vs. 21.6%). There were marginal differences in other characteristics including HCC risk scores. Sample characteristics between the compliers and the full sample were also similar among those who were hospitalized and those who had a hospice admission (Table H).

TABLE 5.

Characteristics of the complier and full sample

Characteristics	Complier a , %	All, %	Prevalence ratio of complier to all
Age
65–69	9.5	8.6	1.1
70–74	15.8	12.4	1.3
75–79	19.2	14.5	1.3
80–84	20.4	18.0	1.1
85+	34.8	46.3	0.8
Female	52.4	54.3	1.0
Race/ethnicity
Non‐Hispanic White	87.8	91.3	1.0
Non‐Hispanic Black	8.5	6.2	1.4
Hispanic	3.0	0.1	30.0
Non‐Hispanic Asian/Pacific Islander	0.3	0.4	0.8
Non‐Hispanic American Indian/Alaska Native	0.7	0.4	1.8
Dual eligibility for Medicare and Medicaid	29.4	21.6	1.4
Medicare eligibility due to disability	0.4	0.3	1.3
(Unnormalized) HCC risk scores
> 0–0.3	1.6	2.2	0.7
> 0.3–0.5	6.2	5.7	1.1
> 0.5–1.0	8.9	9.1	1.0
> 1.0–2.0	21.0	20.1	1.0
> 2.0	37.8	37.7	1.0

^a The complier was those whose MA enrollment status is sensitive to residing in counties in MSAs with populations of 250,000 or more.

4. DISCUSSION

Prior studies found that compared to TM enrollees, MA enrollees had lower hospital admissions and higher hospice use at the end of life. ¹⁶ , ¹⁷ , ¹⁸ However, the prior studies did not fully account for selective enrollment into MA, possibly leading to biased estimates. In this study, we addressed this issue by adopting a quasi‐experimental IV method. Consistent with prior studies, we found that MA enrollment decreased hospital death and increased hospice use. We further explored the effects of MA enrollment on more detailed patterns of end‐of‐life care finding hospice use was only higher after a hospitalization, suggesting a hospitalization may be a trigger for hospice referral. However once hospitalized, there was no difference in the ICU use or use of non‐invasive mechanical ventilation.

Our findings suggest that MA plans may achieve high efficiency of care delivery by avoiding deaths in the hospital setting. This may indicate that MA plans may not prevent hospitalizations in the last 30 days of life, but may play a critical role in the hospital discharge to hospice, resulting in a reduction in hospital deaths. As most older adults prefer palliation of symptoms at home rather than invasive therapies or hospitalization at the end of life, ²⁷ this may potentially improve quality of life at the end of life. Our finding is inconsistent with findings from prior research that MA enrollment decreased hospital admissions for all enrollees after accounting for selective enrollment into MA. ²² This finding discrepancy may be because we focused on decedents who tend to have an advanced illness.

However, there was limited evidence in improving the efficiency of care delivery once a person was hospitalized. Additionally, MA enrollment decreased two measures of potentially burdensome transitions (transitions from hospital to another hospital in the last 3 days of life and more than three hospitalizations in the last 90 days of life). Once hospitalized, however, MA decedents had more ICU use. There may be multiple potential explanations for this finding. One potential explanation is that those who were hospitalized may have different preference patterns compared to those who were not hospitalized, leading to differences in care intensity during hospitalization. Another potential explanation is that MA's traditional approaches to care management may focus on avoiding hospitalizations. Further research is warranted to examine more closely the underlying mechanisms for increased care intensity among those hospitalized.

Our results indicate that MA enrollment improves access to hospice care. Hospice care is currently carved out of the MA benefits and it is reimbursed on a fee‐for‐service basis. There are some concerns that this carve‐out may have offered an incentive to MA plans to refer potentially costly enrollees to hospice. However, prior research found little evidence of this. ¹⁸ , ¹⁹ Rather, we found that MA enrollment significantly increased admission into hospice among Medicare decedents. This finding suggests that MA plans may play a critical role in improving quality of life at the end of life by improving access to hospice. However, prior research found that family and friends of those in MA reported slightly lower quality end‐of‐life care than family and friends of those in TM. ²¹ As we did not examine whether MA enrollment leads to admission into high‐quality hospice, further research is warranted to explore the underlying mechanism.

There were two other notable findings for the effects of MA enrollment on hospice. First, inpatient hospitalization may be a critical pathway to hospice for MA enrollees. This suggests that an inpatient setting may be a trigger for MA plans to initiate care management. Second, an increase in transitions from hospital to hospice in the last 3 days of life is an opportunity to improve. This may suggest that the timing of transitions to hospice care is suboptimal, and thus patients may have received care that does not align with their needs and preferences. Indeed, about 25% of Medicare beneficiaries with hospice services received these services only in the last week of life. ²⁸ Further research is required to investigate how to improve access to hospice in a timely manner.

Our results provide key policy implications. Starting in 2021, CMS has allowed MA plans to participate in a demonstration to test the inclusion of the hospice benefit component as part of the MA Value‐Based Insurance Design model. Implementing “hospice carve‐in” to MA is likely to give MA plans full financial and care responsibility for their enrollees at the end of life, possibly reducing care intensity as well as potentially burdensome transitions. However, this requires well‐designed policies and quality measures. Otherwise, there may be unintended adverse consequences, possibly offsetting the effect of MA enrollment on end‐of‐life care observed in our study. If reimbursement rates for end‐of‐life care are set too low, then this may induce MA plans not to provide appropriate patients with access to hospice services. If quality measures for end‐of‐life care are not reflected in estimating quality performance, then this may discourage MA plans to improve care quality for advanced illness. To maximize the benefit of MA enrollment in the context of “hospice carve‐in”, policymakers need to consider how to incentivize care that aligns with the patient and family needs in designing the hospice carve‐in policy and avoids potential unintended consequences. ¹⁵ As policies that merely rely on financial incentives and do not focus on quality may contribute to suboptimal end‐of‐life care, it is critical to ensure that decisions on end‐of‐life care are not solely based on financial incentives but also on the values and goals of those with advanced illness. Second, MA enrollment has been increasing over the past decade. However, enrollment growth has been more pronounced among racial/ethnic minority groups. ²⁹ Addressing the diverse needs of racial/ethnic minority groups will be a priority for MA plans.

Our study has several limitations. First, our findings may not be generalizable. Specifically, our sample was limited to Medicare decedents in counties in MSA with populations of 100,000–400,000. This may be particularly relevant to Medicare decedents in rural areas as there may be differences in plan availability and benefits between rural and urban areas. Relatedly, our IV analyses estimated the LATE of MA enrollment on end‐of‐life care for those whose MA enrollment has been changed due solely to residing in counties in MSAs with populations of 250,000 or more. Thus, the pattern of MA enrollment in these counties may not be applicable to other counties. Second, our instrument operates at the county level, not at the individual level. Thus, our findings may reflect a combination of the direct effect of MA enrollment as well as the spillover effect of MA enrollment, possibly leading to overestimating the impact of MA enrollment. Indeed, evidence suggests that improvements in care delivery by MA may provide spillover effects by improving care delivery for TM enrollees. ³⁰ , ³¹ Third, our findings may be subject to survival bias. We analyzed retrospectively from death, but prior research showed that MA enrollment not only reduced hospitalizations but also mortality rates, ²² suggesting that time to death may be influenced by MA enrollment. However, the magnitude of the effect on mortality was relatively modest (0.2–0.7 percentage point decreases), and thus the magnitude of the bias may be marginal. Fourth, we only had information available in administrative data. Thus, our findings do not necessarily indicate that decreases in hospital admissions and/or increases in hospice use are consistent with a person's preference and goals of care. Finally, we excluded those without continuous enrollment either in TM or MA. However, evidence suggests that disenrollment rates are higher among MA enrollees than TM enrollees, possibly raising some concerns about the sickest patients, disenrolling from MA prior to death. ³² In our data, we found disenrollment rates were slightly higher among MA decedents than TM decedents (5.6% vs. 2.5), but overall disenrollment rates were relatively low (3.2%). However, our findings may not be generalizable to this proportion of the population if switching between TM and MA in the last year of life reflects substantial differences in unobservable health status.

5. CONCLUSIONS

Our findings suggest that MA plans may improve end‐of‐life care by reducing hospital death while improving access to hospice. However, we also identified potential opportunities to improve. For example, there was limited evidence in improving the efficiency of care delivery once a person was hospitalized. As we debate changes in MA payment policy, understanding the opportunities and the potential unintended consequences is important to guide policy for a frail, vulnerable population dying in the United States.

Supporting information

Data S1 Supporting information.

Park S, Teno JM, White L, Coe NB. Effects of Medicare advantage on patterns of end‐of‐life care among Medicare decedents. Health Serv Res. 2022;57(4):863‐871. doi: 10.1111/1475-6773.13953