The association between hospice use and depressive symptoms in surviving spouses

Abstract

Importance

Family caregivers of individuals with serious illness are at risk for depressive symptoms and depression. Hospice includes the provision of support services for family caregivers, yet evidence is limited regarding the impact of hospice use on depressive symptoms among surviving caregivers.

Objective

To determine the association between hospice use and depressive symptoms in surviving spouses.

Design

We linked data from the Health and Retirement Study (HRS), a nationally representative longitudinal survey of community dwelling U.S. adults 50 years of age and older to Medicare claims. With propensity score matching on both decedent and spousal characteristics, we compared the spouses of individuals enrolled in hospice prior to death to the spouses of individuals who did not use hospice.

Setting

National study of decedents and surviving spouses.

Participants

Propensity score matched sample of 1016 HRS decedents with at least one serious illness and their surviving spouses interviewed 2002–2010.

Exposure(s) for observational studies

Hospice enrollment for at least 3 days in the year prior to death.

Main Outcome(s) and Measure(s)

Spousal depressive symptom scores measured 0–2 years post-death with the Center for Epidemiologic Studies Depression Scale (CES-D), which is scored 0 (no symptoms) to 8 (severe symptoms).

Results

Of the decedents in the matched sample, 305 (30.0%) used hospice services for >= 3 days in the year prior to death. Of the spouses, 52% had more depressive symptoms over time (mean change = 2.56; S.D = 1.65), with no difference related to hospice use. A minority, (28.2%) of spouses of hospice users had improved CES-D scores, compared to 21.7% of spouses of decedents who did not use hospice (p=0.06). Among the spouses who were the primary caregivers (n-662), 27.3% of spouses of hospice users had improved CES-D scores, compared to 20.7% of spouses of decedents who did not use hospice (p=0.10); in multivariate analysis, the odds ratio for the association of hospice enrollment with depressive symptoms after the spouse's death was 1.63 (95% CI = 1.00–2.65).

Conclusions and Relevance

After bereavement, depression symptoms increase overall for surviving spouses regardless of hospice use. A modest reduction in depressive symptoms was more likely among spouses of hospice users than among spouses of non-hospice users.

Background

The Institute of Medicine’s recent report on improving quality of care near the end-of-life highlights the need for supporting family caregivers.¹ Despite the fact that family caregivers play a critical role in the care of patients with serious illness, they are at increased risk of poor health. Although there are benefits to providing care,^2–4 caregivers are vulnerable to increased depression⁵ and other health problems^,6,7 and reduced preventive health behaviors.^8,9 As a consequence, caregiving is associated with increased individual and societal healthcare costs.^10,11

The growing use of hospice care (currently, 45% of decedents in U.S. die under hospice care with a 21% increase in the last decade)^12,13 reflects one effort to meet the challenge of addressing the multidimensional needs of individuals with terminal illness while simultaneously supporting family members during a patient’s illness and after their death. Hospice services are focused on palliative rather than curative care and include medical services, symptom management, spiritual counseling, social services and bereavement counseling delivered by an interdisciplinary team of professionals for dying patients (i.e., prognosis of six months or less). Core components of high quality hospice care include counseling services for family members before and after the patient’s death.¹⁴ This unique service is bundled into the hospice per diem payment for the patient and is almost universally implemented - according to a national survey, 98% of hospices reported making calls to bereaved family members.¹⁵

Studies to date have found beneficial effects of hospice use and bereavement support on caregivers including improved family functioning,¹⁶ better bereavement adjustment,^16,17,18 improved satisfaction,^17,19,20 decreased depression,^21–23 and fewer unmet needs.²⁴ Yet these studies have been largely limited to patients with cancer, have failed to adequately control for differences between patients who do or do not use hospice, or are subject to recall bias. Although surviving spouses of patients with a range of illnesses who die in hospice care have been shown to live longer than the spouses of patients who do not use hospice,²⁵ prospective national studies that address the potential for selection bias are lacking.

Therefore, we used a national sample to prospectively evaluate the impact of hospice use on depressive symptoms in surviving spouses and, more specifically, the subset of spouses identified as primary caregivers. We hypothesized that hospice use is associated with decreased depressive symptoms in surviving spouses.

Methods

Study population

The study cohort is from the Health and Retirement Study (HRS), a nationally representative longitudinal survey of community dwelling U.S. adults 50 years of age and older.²⁶ The majority of participants are interviewed via telephone every two years. If participants are married, spouses are recruited and surveyed regardless of their age. Survey data are linked for eligible participants with individual Medicare claims. Study participants provided informed consent upon enrollment and again for linkage to Medicare claims.

We included HRS participants, enrolled in fee-for-service Medicare for at least 6 months prior to death, with post-death interviews between 2002 and 2010, and linked to 2000–2010 Medicare claims (the most recently available Medicare claims data). Because our primary interest was the impact of hospice use on surviving spouses, we limited our analytic sample to decedents who were married at the time of death and whose spouse completed interviews both before and after the participant’s death. We limited the sample to those decedents with at least one chronic condition identified in claims data up to one year prior to death and excluded decedents whose cause of death was documented as an accident, suicide, or homicide.

Measures

Spousal depressive symptoms

The HRS assesses depressive symptoms (e.g., feeling sad, lonely, low energy) at each interview with an 8-item version of the Center for Epidemiologic Studies Depression Scale (CES-D), which is scored 0 (no symptoms) to 8 (severe symptoms).²⁷ A cut-off of 3 or more has been used as an indicator of clinical depression.^28,29 For our primary outcome, we determined whether scores improved (i.e., reduced depressive symptoms) from the interview before patient’s death to the next study assessment by at least one point (mean follow up time =25.5 months (SD=4.4)). We dichotomized the value of change from pre-death to post-death as improved CES-D score versus no change/worsening symptoms from time 1 (pre-death) to time 2 (post-death). Our focus was on depressive symptoms regardless of indicator of clinical depression because subclinical depression symptoms are associated with adverse outcomes among older adults, including lower quality of life and worsened physical function.^30,31 Because depressive symptoms increase during bereavement,³² our primary outcome was symptom reduction, not magnitude of change.

Hospice use

We determined length of hospice enrollment from individual Medicare hospice claims and determined date and cause of death from the National Death Index. Although any length of hospice use appears to be correlated with improved quality of care³³ and should afford family members bereavement services, the ability of hospices to help family members prepare for the death of a loved one may be hampered with a very short length of hospice stay.²¹ For this reason, we focused on hospice use >=3 days. Because the optimal length of stay in hospice for improvement in patient or caregiver outcomes is unknown, we performed sensitivity analyses using cutoffs of >=7 days and >=1 day.

Primary caregiver status

The spouse’s role as primary caregiver was determined by whether or not the spouse was reported to be the primary helper with decedents’ activities of daily living (ADLs) or instrumental activities of daily living (IADLs) in the last 3 months of life as per the post-death interview.

Analysis

We used both propensity score matching and multivariable regression adjustment. Based on previous work,³³ we developed propensity scores to control for factors hypothesized to be associated with both likelihood of patient hospice enrollment and spousal depressive symptoms. If two factors were highly correlated (e.g., decedent and spouse race), only one factor was retained. Decedent factors included were net worth, insurance coverage, functional status, residential status, self-reported health, medical conditions, number of chronic conditions, presence of advance directives, and discussion of end of life care preferences. Spousal factors included were age, sex, race/ethnicity, education, self-reported heath, level of comorbidity, presence of psychiatric illness, and importance of religion. Propensity scores were calculated as a logistic regression of hospice enrollment. The balance of propensity scores was established prior to matching.

We then matched hospice users to one or many decedents who did not use hospice within 0.02 of the standard deviation of the logit of the propensity scores using matching techniques with replacement. We accounted for multiple matches with weighting.³⁴ For sensitivity analyses of alternate hospice lengths of stay (1 day and 7 day hospice use), matching techniques were repeated retaining two groups in each comparison: (1) those that met the specific hospice length of stay requirement and (2) those that had no hospice or had a shorter hospice length of stay. Finally, matching techniques were repeated in samples restricted to spousal primary caregivers. After balancing, matching was verified by examining: standardized differences across treatment groups, variance of covariates before and after matching, and standardized differences across covariates. Standardized differences less than 10% were considered to be adequate balance.^35,36

We examined bivariate comparisons of unadjusted measures of depressive symptoms using the matched, weighted samples using chi square and t-tests. We then conducted a series of multivariate logistic regressions for all spouses and the subgroup of spouses who were primary caregivers measuring reduction in spousal depressive symptoms overall including all covariates described above and an indicator for year of death to account for secular trends over time. Finally, to examine whether results were sensitive to length of bereavement, we limited our sample to those spouses interviewed more than 1 year after the decedent’s death.

Results

A total of 6,814 HRS participants died between 2002 and 2010 and had a spouse who completed the postmortem interview. We limited our sample to those with linked Medicare data, fee for service Medicare for at least the last 6 months of life, and who were married at the time of death (n=1617). We further excluded decedents for the following reasons: spouse was missing a pre-death and/or post-death interview (n=302); the decedent had no documented chronic illness in the past year (n=49); or the decedent cause of death was accident, suicide, or homicide (n=54). Finally, dyads were excluded because they were missing CES-D spousal data pre or post patient death because they had proxy interviews (n=135) or because of non-response (n=28), or they did not have sufficient covariate data available for propensity score matching (n=24).

The final unmatched sample included 1025 decedents with mean age of 78.1 years at death. Seventy-four percent were men, 84.2% were non-Hispanic white, and 64.8% had at least a high school education. Thirty percent used hospice at least 3 days in their last year of life and individuals who used hospice at least 3 days did so for a mean of 55.1 days (median = 22) (See Table 1). On average spouses were interviewed 13.9 months before and 11.6 months after the decedent’s death. Surviving spouses were mostly female (74.0%), mean age 74.9, and 70.7% had at least a High School education. The vast majority of spouses (95.3%) were independent in all ADLs at baseline.

Table 1.

Characteristics of study sample (n=1025)

Decedent			Spouse
Demographics	Age at death, years, mean (SD)	78.12 (7.93)	Demographics	Age at spouse's death, years, mean (SD)	74.87 (8.95)
	Female, %	25.95		Female, %	74.05
	Race, Non-Hispanic, White, %	84.18		Race, Non-Hispanic, White, %	83.32
	Education, High School Grad, %	64.78		Education, High School Grad, %	70.73
	Net Worth, 1 (low) Quartile, %	23.02		Medicaid, %	6.56
	Net Worth, 2 Quartile, %	30.54		Religion very important, %	95.32
	Net Worth, 3 Quartile, %	26.73	Functional status	ADL Independent pre-death, %	72.98
	Net Worth, 4 (high) Quartile, %	19.71	Self-rated health	Poor/Fair
	Medicaid, %	12.00	Chronic conditions	Cancer, %	15.90
	VA Insurance, %	7.32		Lung disease, %	11.12
	Medigap, %	57.27		Heart disease, %	26.63
	Nursing home resident pre-death, %	8.68		CHF, %	4.00
Advanced care planning	Had advanced directive, %**	57.46		Stroke, %	6.05
	Discussion of EOL care, %	56.00		Memory disease, %	1.37
Functional status	ADL Independent pre-death, %	64.64		Hypertension, %	58.73
	Impr or Indep in ADLs n2 to n1, %	64.59		Diabetes, %	19.12
	Stable Moderate 1–3 ADL, %	8.00		Comorbidities, none, %	20.68
	Stable Severe 4–6 ADL, %	5.66		Mild (1–3), %	71.22
	Declined, Indep to Moderate, %	12.00		Moderate (4–6), %	8.10
	Declined Moderate to Severe, %	3.71		Psychological problems, %	16.10
	Declined Independent to Severe n2-n1, %	5.27	Health behaviors	Currently smoker, %	10.24
Self-rated health	poor or fair, %	64.78		Vigorous physical activity, %	27.80
Chronic conditions	Alzheimer's or Dementia	28.49	Depression	CES-D score pre-death, mean(SD)	1.76 (2.03)
	Chronic kidney disease	43.02		CES-D score post-death, mean(SD)	2.67 (2.35)
	AMI or Ischemic heart failure	58.93
	CHF	53.17
	Diabetes	42.44
	COPD	50.83
	Stroke/TIA	29.37
	Cancer, all types	27.12
	Atrial fibrillation	36.88
	Hip fracture	6.15
	Depression	19.90
	Osteoporosis	11.22
	Arthritis	29.27
	Number of chronic conditions, mean (SD)	4.69 (2.18)

Note: Chronic conditions for Decedent are from the Medicare Claims. Chronic conditions for spouse are self-reported in HRS core interviews; EOL= end-of-life; CES-D = Center for Epidemiologic Studies Depression Scale

Prior to propensity score matching, hospice users and non-users differed substantially (>10% standardized difference) on rates of several confounders. In the un-matched sample, individuals who used hospice were more likely than those who did not use hospice to be older (mean age 78.8 vs 77.8), white (87.5% vs 82.7%), have an advance directive (64.4% vs 54.4%) and a diagnosis of cancer (39.7% vs. 21.6%) or dementia (32.7% vs. 26.7%).

The final propensity score matched sample included 1016 spouses: 305 spouses of decedents who used hospice and 711 spouses of decedents who did not use hospice (99.1% of complete sample). In our matched sample, observed confounders were balanced (<10% standardized difference) (See Table 2).³⁶ Variables that were not included in the matching because of collinearity (e.g., decedent race) were also found to have adequate balance in the matched sample. Median duration of hospice use among those who used hospice at least 3 days was 22 days and 43% used hospice for at least 30 days.

Table 2.

Decedent and spouse characteristics used in propensity score matching, by hospice use status

		Unmatched Sample			Matched Sample
		Hospice users	No hospice use	Standardized Difference(%)	Hospice users	No hospice use	Standardized Difference(%)
	Sample n	312	713		305	711
Decedent	Net Worth, 1 (low) Quartile, %	23.08	23.00	<0.01	22.62	22.50	<0.01
	Net Worth, 2 Quartile, %	30.13	30.72	−0.01	29.51	30.24	−0.02
	Net Worth, 3 Quartile, %	26.60	26.79	−0.01	27.21	28.00	−0.02
	Net Worth, 4 (high) Quartile, %	20.19	19.50	0.02	20.66	19.26	0.03
	Medicaid, %	14.10	11.08	0.09	13.11	13.35	−0.01
	VA Insurance, %	8.01	7.01	0.04	8.20	7.69	0.02
	Medigap, %	59.29	56.38	0.06	59.67	59.62	<0.01
	Nursing home resident pre-death, %	7.37	9.26	−0.07	7.54	7.91	−0.01
	Had advanced directive, %**	64.42	54.42	0.20*	63.93	64.65	−0.02
	Did not have advance directive, %**	34.29	44.04	−0.20*	34.75	34.28	0.01
	Discussion of EOL care, %	61.54	53.58	0.16*	60.66	61.05	−0.01
	SRH poor or fair, %	68.59	63.11	0.12*	67.87	69.65	−0.04
	Improve or Indep in ADLs n2 to n1, %	60.58	66.34	−0.12*	60.98	62.34	−0.03
	Stable Moderate 1–3 ADL, %	8.65	7.71	0.03	8.85	8.52	0.01
	Stable Severe 4–6 ADL, %	6.09	5.47	0.03	6.23	7.40	−0.05
	Declined, Indep to Moderate, %	12.50	11.78	0.02	11.80	11.51	0.01
	Declined Moderate to Severe, %	4.81	3.23	0.08	4.92	4.48	0.02
	Declined Independent to Severe n2-n1, %	6.09	4.91	0.05	5.90	6.17	−0.01
	Alzheimer's or Dementia	32.69	26.65	0.13*	32.46	32.41	<0.01
	Chronic kidney disease	34.29	46.84	−0.26*	35.08	34.61	0.01
	AMI or Ischemic heart failure	50.00	62.83	−0.26*	51.15	51.84	−0.01
	CHF	44.23	57.08	−0.26*	45.25	44.81	0.01
	Diabetes	37.50	44.60	−0.14*	37.70	37.89	>−0.01
	COPD	52.56	50.07	0.05	52.13	50.94	0.02
	Stroke/TIA	29.81	29.17	0.01	29.51	31.58	−0.04
	Cancer, all types	39.74	21.60	0.40*	38.36	37.62	0.02
	Atrial fibrillation	34.29	38.01	−0.08	35.08	35.08	>−0.01
	Hip fracture	5.77	6.31	−0.02	5.90	6.23	−0.01
	Depression	21.79	19.07	0.07	22.30	21.57	0.02
	Osteoporosis	12.18	10.80	0.04	12.13	12.31	−0.01
	Arthritis	26.92	30.29	−0.07	27.54	27.74	>−0.01
	Number of chronic conditions, mean (SD)	4.54 (2.31)	4.76 (2.12)	−0.10*	4.57 (2.32)	4.57 (2.15)	>−0.01
Spouse	Age at spouse's death, years, mean (SD)	75.49 (8.74)	74.60 (9.04)	0.10*	75.59 (8.70)	75.66 (8.72)	−0.01
	Female, %	73.72	74.19	−0.01	73.11	73.86	−0.02
	Race, Non-Hispanic, White, %	85.90	82.19	0.10*	85.90	85.30	0.02
	Education, High School Grad, %	74.04	69.28	0.11*	74.10	72.06	0.05
	Self Rated Health: Poor/Fair	25.32	32.26	−0.15*	25.25	24.64	0.01
	Religion very important, %	76.60	71.39	0.12*	76.07	76.30	−0.01
	Comorbidities, none, %	20.51	20.76	−0.01	20.66	19.25	0.04
	Mild (1–3), %	72.44	70.69	0.04	72.13	73.73	−0.04
	Moderate (4–6), %	7.05	8.56	−0.06	7.21	7.03	0.01
	Psychological problems, %	16.67	15.85	0.02	16.72	16.03	0.02

Note: SRH = Self-reported health; hospice users defined as individuals who used hospice>= 3 days in last year of life

^*Absolute value of mean standardized difference above 10%

^**Not all categories add up to 100% due to rounding and missing data.

^***n2-n1 refers to change in ADL status over 2 pre-death assessments

Within the matched sample, depressive symptoms were common among all spouses pre-death (mean = 3.2 years before death) and did not vary based on hospice use (See Table 3). Regardless of hospice use, the majority of spouses had more depressive symptoms over time. CES-D scores increased from 1.8 (SD=2.0) before death to 2.6 (SD=2.3) after death. Fifty-two percent of the sample had increased CES-D scores over time (mean change =2.56; S.D= 1.65) with no difference across groups.

Table 3.

Spousal depressive symptoms by hospice use before and after death (matched sample, n=1016)

		Hospice use	No hospice use	p-value
2nd assessment before death	Days from assessment to spouse’s death, mean	1182	1163	.36
	CES-D score, mean (S.D)	1.48 (1.90)	1.58 (1.87)	.49
	Clinical depression, %	23.84	22.76	.75
1st assessment before death	Days from assessment to spouse’s death, mean	439	425	.45
	CES-D score, mean (S.D.)	1.76 (1.88)	1.71 (2.06)	.73
	Clinical depression, %	30.49	27.19	.37
1st assessment after death	Days from spouse’s death to assessment, mean	328	351	.25
	CES-D score, mean (S.D)	2.61 (2.32)	2.53 (2.33)	.67
	Clinical depression, %	41.64	42.54	.82
CES-D symptom reduction	% of subjects with reduced CES-D symptoms	28.20	21.68	.06
	Change in CES-D among those with reduced symptoms, mean, S.D.	1.86 (1.34)	2.16 (1.33)	.17

Note: CES-D= Center for Epidemiologic Studies Depression Scale; hospice users defined as individuals who used hospice>= 3 days in last year of life; Clinical depression measured via CES-D score >=3 ; sample size reduce to n=1000 2 assessments before death because of missing CES-D data.

A subgroup of the sample (25%) had improved CES-D scores over time. In the matched sample, 28.2% of spouses of hospice users had improved CES-D scores, compared to 21.7% of spouses of decedents who did not use hospice (p=0.06). Of those who improved, the mean change was 1.99 points (S.D.= 1.34).

In multivariate analysis of propensity score matched decedents, hospice enrollment was associated with improved spousal depressive symptoms (OR=1.64; 95% CI=1.09 – 2.48) (See Table 4). With a hospice enrollment cutoff of at least 7 days, hospice enrollment was associated with improve spousal depressive symptoms (OR=1.65; 95% CI=1.02–2.65) but there was no association with a cutoff of at least one day of hospice use (OR=1.31; 95% CI=0.91–1.90).

Table 4.

Change in spousal depressive symptoms following the death of individuals who used and did not use hospice (matched sample)

					Alternate hospice LOS definitions**
		>=3 day hospice use (vs 0–2 day use)			>=1 day hospice use			>=7 day hospice use (vs 0–6 day use)
		OR	95% CI	p-value	OR	95% CI	p-value	OR	95% CI	p-value
All Spouses (n=1,016)	Any reduction in depressive symptoms	1.64	1.09 – 2.48	0.02	1.31	0.91 – 1.90	0.15	1.65	1.02 – 2.65	0.04
	Reduction in clinical depression	1.75	0.94 – 3.23	0.08	1.32	0.76 – 2.28	0.32	1.51	0.73 – 3.12	0.27
Primary caregiver spouses only (n=662)	Any reduction in depressive symptoms	1.63	1.00 – 2.65	0.05	1.52	0.97 – 2.37	0.07	1.33	0.76 – 2.33	0.32
	Reduction in clinical depression	1.46	0.74 – 2.87	0.27	1.30	0.70 – 2.43	0.40	1.21	0.55 – 2.66	0.64

Note: LOS= length of stay; clinical depression measured via Center for Epidemiologic Studies Depression Scale (CES-D) >=3

^*using doubly robust propensity score matching and multivariate regression models controlling for decedent net worth, insurance coverage, functional status, residential status, medical conditions, number of chronic conditions, presence of advance directives, discussion of end of life care preferences and spousal factors included were age at respondent’s death, sex, race/ethnicity, education, self-reported heath, level of comorbidity, presence of psychiatric illness, and importance of religion

^**Sample size varies slightly in alternate hospice LOS definitions due to matching as follows: 1 day hospice use includes 1020 spouses and 659 caregivers;7 day hospice use includes 1010 spouses and 655 caregivers

About two-thirds (65.2%) of spouses were identified as primary caregivers. Among spousal caregivers, 27.3% of spouses of hospice users had improved CES-D scores, compared to 20.7% of spouses of decedents who did not use hospice (p=0.10); in multivariate analysis, the odds ratio for the association of hospice enrollment with depressive symptoms after the spouse's death was 1.63 (95% CI = 1.00–2.65).

Based on previous studies,^28,29 we created a variable to indicate reduction in clinical depression to identify those spouses whose CES-D score changed from clinically significant depression (>=3) to no depression (<3) following the decedent’s death. Thirty percent of spouses of hospice users and 27% of spouses of individuals who did not use hospice had clinical depression pre-death. In multivariate models we found no reduction in clinical depression for spouses of individuals who used hospice compared to those who did not (OR=1.75; 95%CI=0.94–3.23) (See Table 4).

Among those interviewed more than 12 months after their spouse’s death (n=438), 38% of the hospice group had symptom improvement compared to 23% of spouses of those who did not use hospice (p= 0.01).The likelihood of reduced depressive symptoms was greater for spouses who were exposed to hospice use as compared to those who were not (OR=2.15; 95%CI=1.10–4.21). For spouses who were primary caregiver the comparable odds ratio was 2.54 (95%CI=1.09–5.94).

Discussion

Using linked personal interviews and Medicare claims for married decedents in a national cohort, we found that depressive symptoms increased for the majority of surviving spouses after the death of their spouse. About a quarter, however, had decreased symptoms. Specifically, surviving spouses of individuals who used hospice at least 3 days prior to death were more likely to have some reduction in depressive symptoms after their spouses died than surviving spouses of those who did not use hospice. This finding applied to all spouses (not just those identified as primary caregivers) and was most evident at least one year after death.

Because HRS interviews are conducted at two-year intervals regardless of timing of death, we were able to assess the association between hospice use and depressive symptoms across various time periods. The relationship between hospice use and reduced depression symptoms was stronger 1–2 years after death, consistent with previous research.³⁷ Because grief is normal and expected the first year after the death of a spouse, we may only be able to capture meaningful differences for families of hospice users after one year. Additionally, caregivers have not received the full dose of bereavement benefits provided to caregivers under Medicare until one year after death. Future evaluation of the impact of hospice use on spousal depression and other outcomes should include longer follow up periods.

Given that hospice is an existing comprehensive care package that provides benefits to patients,^23,38,39 and achieves cost savings³³, any improvements in health for spouses would be an added benefit. Because the majority of surviving spouses regardless of decedent’s hospice use had increased depressive symptoms after death, our findings suggest that additional supports are needed if the increased depressive symptoms associated with death of spouse are to be substantially reduced.

We focused on the surviving spouses of patients who used hospice for at least 3 days prior to death. Although there is concern about whether patients receive the full benefit of hospice care when their referral is late (<7, <3 or even only 1 day before death),^22,40 our findings suggest that caregivers may nonetheless benefit from short hospice stays. Perhaps that is because the caregiver is entitled to a full year of bereavement follow-up care regardless of the patient’s length of stay in hospice.

Our findings have important limitations. Because we used Medicare claims data to measure hospice use and comorbidities, we were limited to decedents with fee-for-service Medicare. Because bereavement services are bundled into hospice payments and do not generate separate claims, we were unable to determine whether and how frequently spouses actually used these services. A recent national survey of hospices, however, indicates that hospices almost always reach out to bereaved family members after death.¹⁵ Similarly, although we examined whether patients had advance directives, we did not have information on individual patient and family care preferences, or on their satisfaction with the care received. Additionally, although the mortality follow-back sampling method introduces potential selection bias,⁴¹ this approach is feasible for studying end-of-life health care services. And our use of propensity score matching and multivariable regression, although robust, cannot adjust for unmeasured factors.

We were unable to determine from claims data whether there was any use of inpatient or community programs that provided palliative care services.⁴² Although use of such services among patients who did not use hospice would have biased our effect size to the null, research should examine the impact of the receipt of palliative care services other than hospice on caregiver depression. We only studied surviving spouses able to independently complete the CES-D survey; hospice may have different effects on depressive symptoms in spouses with poorer health. Further research should examine other caregiver factors that may impact depression after the patient's death such as the role of spousal social resources and support networks.⁴³

We focused on patients with at least one serious illness at the end of life. We did not consider how long each patient had been ill with any particular disease. Controlling for both the need for informal caregiving before death and reports by next-of-kin of the expectation of death as proxy measures of burden of illness did not change our findings (data not shown).

It is unknown which aspect or aspects of hospice are associated with improved symptoms for spouses. High intensity treatments such as intensive care unit admissions may negatively impact family members.^44–46 Next-of-kin report worse health when more invasive treatments are performed at the end of life,²² experiences that are typically avoided among hospice beneficiaries. Future work should determine which components of hospice care are most beneficial for spouses. Finally, we were only able to examine outcomes among spouses. Hospice use may have beneficial effects for other family members regardless of their direct involvement in end-of-life care.