Racial and Ethnic Differences in End-Of-Life Costs: Why Do Minorities Cost More Than Whites?

Abstract

Background

Racial and ethnic minorities generally receive fewer medical interventions than Whites, but racial and ethnic patterns in Medicare expenditures and interventions may be quite different at life's end.

Methods

Based on a random, stratified sample of Medicare decedents (n=158,780) in 2001, we used regression to relate differences in age, sex, cause of death, total morbidity burden, geography, life-sustaining interventions (e.g., ventilators), and hospice to racial/ethnic differences in Medicare expenditures in the last 6 months of life.

Results

In the final 6 months of life, costs for Whites average $20,166; Blacks, $26,704 (32% more); Hispanics, $31,702 (57% more). Similar differences exist within sexes, age groups, all causes of death, all sites of death, and within similar geographic areas. Differences in age, sex, cause of death, total morbidity burden, geography, socioeconomic status, and hospice account for 53% and 63% of the higher costs for Blacks and Hispanics respectively. While Whites use hospice most frequently (Whites 26%, Blacks 20%, and Hispanics 23%), this affects racial and ethnic differences in end-of-life expenditures only minimally. However, fully 85% of the observed higher costs for non-Whites are accounted for after additionally modeling their greater end-of-life use of the ICU and various intensive procedures (such as, gastrostomies, used by 10.5% of Blacks, 9.1% of Hispanics, 4.1% of Whites).

Conclusions

At life's end, Black and Hispanic decedents have substantially higher costs than Whites. Over half of this is related to geographic, socio-demographic and morbidity differences. Strikingly greater use of life-sustaining interventions accounts for most of the rest.

Racial and ethnic disparities pervade US health care.^1-9 Many studies show Blacks and Hispanics receiving fewer medical services and spending less than Whites. For example, minorities receive fewer cardiac procedures, prescriptions for life-saving medications, and narcotic medications for pain relief. Despite attention, these disparities persist.^{5, 10}

At the end of life, however, this pattern may reverse.¹¹ Several studies find higher Medicare costs and service utilization for non-Whites at life's end.^{2, 12-16} These studies examine differences in socio-demographic and geographic factors as contributors to these disparities. Shugarman et al. reported that in the second and third years prior to the last year of life, spending by Blacks was significantly lower.¹⁴ However, in the last year, this deficit “flipped;” estimated final-year spending was 19% higher for Blacks than for Whites (P = 0.10). They did not study Hispanics.

In analyses restricted to Medicare Part A (hospital bills) and Hospital Referral Regions with substantial numbers of Blacks, several Dartmouth studies attribute most of Black-White cost differences to geography: basically, more Blacks live in regions with high hospital utilization. However, even after controlling for geography, costs in the last 6 months of life remained 29% higher for Blacks than for Whites.²

To sharpen understanding of racial differences in end-of-life care, and to extend comparisons to Hispanics and other minorities, we constructed a national random sample of nearly 160,000 Medicare decedents, oversampled for non-Whites. We tallied all Medicare costs (Parts A and B) in the final 6 months, and quantified the effects on racial and ethnic cost differences of a range of factors, including age, sex, pre-existing co-morbidities (from the penultimate 6 months), cause of death, geography, and socioeconomic indicators, as well as markers for conservative or aggressive end-of life utilization (e.g., hospice, ICU or ventilator use).

Methods

Data Source

From the 1.76 million Medicare beneficiaries aged 66 or older who died in 2001, we selected 241,655, including random samples of 85,000 each for Blacks and Whites, and all Hispanic (∼30,000) and other minority decedents (∼42,000). To ensure complete health and healthcare records, we required enrollment in “traditional” Medicare (the non-HMO program, where services are individually billed) throughout 2000, with both Medicare Parts A and B entitlement (for inpatient and ambulatory care) continuously for 12 months preceding death, and a positive match in the National Death Index (NDI). We excluded beneficiaries in the end stage renal disease (ESRD) program, and those residing in Puerto Rico or other non-mainland territories – because care for these groups is administered quite differently than for others in Medicare. This left 158,780 decedents in the analytic sample. The proportions excluded among Hispanics (55%) and Others (47%) were much higher than among Whites (24%) and Blacks (32%), due principally to excluding residents of Puerto Rico and to high rates of non-NDI- match for Hispanic decedents, and to higher rates of Part B coverage among Whites.

Outcomes

Our primary outcome is total Medicare-covered health care expenditures at the “end of life:” specifically, in the 180 days (“6 months”) preceding death. For each beneficiary we added Medicare-allowed payments for all covered services, including hospital and skilled nursing facility care, hospice and home health services, physician services and durable medical equipment purchases. We also examined intermediate outcomes, such as total costs by type of service, and any use of hospice or of selected life-sustaining procedures, such as ventilators.¹¹

Other Variables

Covariates included age, sex, race, total morbidity burden in the penultimate 6 months of life, geographic location, and socioeconomic status, as well as direct concurrent contributors to, or proximate causes of, final-6-month cost, such as cause of death, hospice use, and receipt of specific, intensive life sustaining procedures.

Age, sex, and race were obtained from Medicare's denominator file, using its racial/ethnic categories of “White,” “Black,” and “Hispanic” and grouping everyone else into “Other,” including Asian (37%), North American Natives (8%), Other (32%), and Unknown (33%). Socioeconomic status was proxied by 1) median income of the patient's zip code of residence, and 2) whether the patient received Medicaid assistance (buy-in) to pay Medicare Part B premiums. Morbidity burden was summarized using: 1) the Charlson comorbidity score and 2) a DCG prospective relative risk score (DxCG version 6.1 for SAS Windows). Each score was based on ICD-9-CM diagnoses recorded during the 6 months preceding the last 6 months of life.^{17, 18} The Charlson score assigns points to 19 disease conditions; scores in these data range from 0 to 37. The DxCG software organizes all ICD-9-CM codes into 184 “condition categories” and summarizes their expected impact on future expenditures via a relative risk score (RRS); a RRS of 1.00 refers to average expected next-year expenditures among all Medicare beneficiaries (not just decedents) observed for one year. Because the choice of morbidity measure did not affect estimates of racial/ethnic differences, we report analyses only using the more predictive RRS.

To see the effect of aggressive end-of-life care on differences in expenditures, we used surgical procedure codes and other markers in the Medicare inpatient file during the last 6 months to identify decedents with any (non-psychiatric) ICU admission, and each of ten intensive life-sustaining interventions: cardiac catheterization, implantation of a cardiac assist device, pulmonary artery wedge monitoring, cardio-pulmonary resuscitation or cardiac conversion, gastrostomy, blood transfusions, dialysis, and the use of mechanical ventilators, intravenous antibiotics, and cancer chemotherapies.¹³ We identified these procedures using AHRQ's Clinical Classifications Software, excluding carotid sinus stimulation (code 99.64); finally, we identified chemotherapy use, using diagnosis and procedure codes as described elsewhere.¹⁹

We examined differences in expected costs by where decedents had lived – in three ways. First, we classified county of residence descriptively, using Beale rural-urban continuum codes to distinguish “metropolitan counties by size and nonmetropolitan counties by degree of urbanization and proximity to metro areas”.²⁰ We also mapped zip-code-of-residence into Dartmouth Atlas-based hospital referral regions (HRRs) and hospital service areas (HSAs). The 306 HRRs are aggregations of over 3000 HSAs, that distinguish geographic regions whose residents access the same hospital(s) and doctors.²¹ We rely principally on models that use HSA (as so-called “fixed effects” indicators), because each HSA defines a small geographic cluster of people who typical rely upon the same hospital systems. In a sensitivity analysis we quantify the modest differences associated with coding place of residence by county, HRR or HSA.

Analysis

We used STATA version 9.2 for all analyses, which are weighted to adjust for oversampling non-Whites.²² We summarize outcome and covariate variables for all Medicare decedents, and compare them by race and ethnic group (using chi-square and analysis of variance to test for differences).

We used regression, to estimate differences by race/ethnicity in total end-of-life expenditures after accounting for differences in covariates. Models successively added covariate sets: age and sex only (Model B); underlying cause of death determined from death certificates and morbidity (using RRSs) in the 6 months prior to the final 6 months (Model C); geography (using HSAs, Model D); socioeconomic status (Model E); hospice use (Model F); and, use of the ICU and 10 intensive procedures during the end-of-life period itself (Model G). The order in which explanatory variables are added affects the size of the contribution attributed to each. We first adjusted for pure patient characteristics (age, sex and medical problems), then geographic and socioeconomic indicators. We included geography early in the sequence to focus on differences by race among (otherwise similar) people who face the same health care delivery systems.23 Finally, we examined differences in the use of specific services only after all these factors that might otherwise confound associations between race and procedure use had been accounted for. For example, to the extent that non-Whites use more aggressive interventions because they have the same utilization patterns as their White neighbors, we wanted that to be attributed to geography. By entering utilization variables last, the estimated effects focus exclusively on cost differences that arise because non-Whites have different utilization patterns than their neighbors. We also explored the effect of sequencing on the apparent importance of different sets of variables.

Due to large sample sizes, racial/ethnic differences are almost always statistically significant at the P < 0.05 level. Thus, we only explicitly remark on it, when race and ethnic differences are not significant.

Results

Among the 158,780 decedents, Blacks and Hispanics are younger than Whites, and more often live in metropolitan areas with over 1 million people (Table 1). Cause of death is similar across racial groups. However, site of death differs, with Blacks and Hispanics dying more often than Whites in hospitals and less often in nursing homes.

Table 1.

Characteristics of 2001 Medicare Decedents within Racial/Ethnic Groups

Characteristic		White	Black	Hispanic	Other minorities	All
Number of cases analyzed		64,819	58,182	13,634	22,145	158,780
Sex & age (%)
Women	66 to 74	9	13	12	10	10
	75 to 84	21	21	23	17	21
	85 or older	26	22	18	29	26
Men	66 to 74	11	16	16	12	12
	75 to 84	19	18	23	17	19
	85 or older	13	10	9	15	12
Mean Age (Std. Deviation)		81.8 (8.0)	80.2 (8.4)	79.7 (7.3)	82.3 (8.5)	81.7 (8.0)
DCG Prospective Risk Score > 3 (%)		14	18	21	16	14
Mean DCG Score (Std. Deviation)		1.7 (1.3)	1.7 (1.3)	1.8 (1.5)	1.8 (1.6)	1.8 (1.5)
Cause of Death(%)	Heart disease	36	37	36	34	36
	Cancer	22	23	20	21	22
	Stroke & brain diseases	13	11	11	13	13
	COPD	6	4	4	5	6
	Pneumonia & influenza	3	4	6	3	3
	Diabetes	3	3	4	4	3
	Chronic liver disease & cirrhosis	0.5	0.3	1.7	0.6	0.5
	Injuries, homicide, suicide	2	2	2	3	2
	Other	15	17	15	16	15
Site of Death (%)	Hospital	38	46	52	45	39
	Nursing home	31	21	17	25	31
	Residence	21	19	20	19	21
	Other	10	14	11	11	10
Urbanicity: County of Residence (%)	Metropolitan >1 million	40	54	57	52	42
	Metropolitan 250K to 1 million	21	17	23	21	21
	Town < 250K	12	9	9	7	12
	Non-metro (incl. rural)	26	19	11	18	26
Median Income quartile: Residence zip code (%)	Lowest	10	38	33	15	12
	Second lowest	26	27	26	21	26
	Third lowest	30	21	23	26	30
	Highest	34	14	18	39	33
Medicaid Buy-in (%)	No	80	51	32	52	77
	Yes	20	49	68	48	23
Hospice Use (%)	No	74	80	77	80	74
	Yes	26	20	23	20	26

Notes:

1. ‘All’ column denotes rates for all Medicare decedents (in 2001), obtained by adjusting for the stratified sampling.

2. (*) A measure of total morbidity burden based on diagnoses in the billing data during the 6 months that precede the last 6 months of life and quantified using the DCG prospective relative risk score.

3. (**) As indicated by Beale urbanization score applied to county of residence in 2001.

4. Similarity of each measure across racial and ethnic cohorts was rejected (p < 0.05).

Do expenditures differ by race and ethnicity?

Black and Hispanic decedents have significantly higher end-of-life Medicare Parts A and B expenditures than Whites (Table 2). White decedents average $20,166 in the last 6 months of life; Blacks, $26,704 (32% more); and Hispanics, $31,702 (57% more). These racial disparities persist within strata defined by age, sex, morbidity-burden level, cause and site of death.

Table 2.

Medicare Expenditures in the Last 6 Months of Life by Decedent Characteristics and Racial/Ethnic Group

Characteristic		White	Black	Hispanic	Other minorities	All
		Mean Expenses in Thousands of Dollars
All		20.2	26.7	31.7	25.5	1.20.7
Sex & age
Women	66 to 74	26.7	32.9	38.0	32.8	27.3
	75 to 84	21.8	28.5	34.9	28.6	22.3
	85 or older	14.3	22.5	25.0	17.5	14.8
Men	66 to 74	24.4	27.5	33.2	31.6	24.9
	75 to 84	22.6	26.3	31.1	30.6	23.0
	85 or older	17.2	23.5	27.4	22.0	17.6
DCG Score Categoy	< 1.0	17.2	20.5	26.1	21.0	17.5
	1.0 - 3.0	19.6	26.8	30.9	24.1	20.1
	> 3.0	29.7	39.9	42.7	39.6	30.7
Cause of Death	Heart disease	17.9	23.2	29.0	22.7	18.3
	Cancer	23.8	27.5	32.3	29.3	24.2
	Stroke & brain diseases	14.7	24.1	26.0	20.5	15.3
	COPD	23.5	32.7	38.7	30.2	23.9
	Pneumonia & influenza	20.2	27.9	32.4	27.7	21.3
	Diabetes	22.3	35.9	38.9	33.1	23.3
	Chronic liver disease & cirrhosis	23.9	25.5	28.2	28.0	24.2
	Injuries, homicide, suicide	18.5	24.5	22.3	22.0	18.9
	Other	23.1	32.0	39.0	27.1	23.8
Site of Death	Hospital	28.0	34.8	40.4	35.7	28.7
	Nursing home	15.8	24.6	27.8	18.7	16.2
	Residence	14.9	15.7	19.1	15.4	15.0
	Other	14.8	17.8	19.9	16.5	15.0
Urbanicity (County of Residence)	Metropolitan >1 million	23.8	31.7	36.7	31.7	24.7
	Metropolitan 250K to 1 million	18.4	21.9	28.4	20.0	18.7
	Town < 250K	17.9	21.1	22.6	18.0	18.0
	Non-metro (incl. rural)	17.0	19.8	20.4	17.1	17.1
Income quartile (Residence zip code median)	Lowest	20.4	27.9	34.1	27.0	22.0
	Second lowest	18.9	25.2	28.7	23.1	19.4
	Third lowest	19.2	25.7	32.1	24.9	19.6
	Highest	21.9	27.9	31.1	26.5	22.2
Medicaid Buy-in	No	20.7	27.0	27.7	22.1	21.0
	Yes	17.9	26.4	33.6	29.2	19.7
Hospice Use	No	19.8	26.9	32.9	26.0	20.5
	Yes	21.1	25.9	27.7	23.3	21.3

Note:

1. ‘All’ column denotes rates for all Medicare decedents (in 2001), obtained by adjusting for stratified sampling.

2. Similarity of each measure across racial and ethnic cohorts was rejected (p < 0.05).

End-of-life expenditures for Blacks are significantly higher than for Whites in almost every state (Figure 1) and in cities as well as rural areas (Table 2). Interestingly, several southeastern states have both the lowest overall spending and the smallest Black/White differences in cost; cost disparities were largest in urban areas, where more minorities live and end-of-life costs are also high for Whites. Specifically, average end-of-life spending in the largest metropolitan areas for all races is $24,700; in areas with population 250,000 to 1 million, it is $18,700; and in rural areas, $17,100. Finally, these racial and ethnic cost differences occur in each of the last 6 months (Figure 2).

Figure 1.

Comparison of Black and White Average Medicare expenditures 6 months before death, by State (thousand dollars)

Figure 2.

30-Day Medicare expenditures for 2001 decedents, by months before death

How much of the racial and ethnic cost disparity is accounted for by differences in demographics, morbidity, geography and socioeconomic factors?

In Table 3 we quantify the importance of various factors on these raw cost differences by sequentially controlling, first for demographic, health, geographic variables. Model A shows the raw differences in costs for the last 6 months of life among the racial and ethnic groups. Blacks' costs are $6,538 higher than Whites, Hispanics', $11,536 higher, and Other minorities, $5,307 higher.

Table 3.

Models Examining Racial/Ethnic Differences in Total Medicare Expenditures per Capita in the Last 6 Months of Life

	Model A Race only	Model B A + Age, gender	Model C B + Cause of death + Illness Burden	Model D C + HSA Effect	Model E D + SES	Model F E + Hospice	Model G F + Medical Interventions
Reference (Constant)	20,166	14,462	4,901	6,178	6,243	6,455	2,488
Race
White	Ref.	Ref.	Ref.	Ref.	Ref.	Ref.	Ref.
Black	6,538	5,840	5,395	2,924	3,068	3,031	997
Hispanic	11,536	10,504	9,524	3,705	4,282	4,278	1,902
Other	5,307	5,613	5,203	2,103	2,477	2,454	717
Sex & age
Women
66 to 74		12,258	11,427	11,777	11,640	11,607	4,156
75 to 84		7,438	6,922	6,973	6,839	6,813	2,075
85 or older		Ref.	Ref.	Ref.	Ref.	Ref.	Ref.
Men
66 to 74		9,690	9,358	9,865	9,543	9,478	2,716
75 to 84		7,959	7,115	7,538	7,177	7,131	1,691
85 or older		2,783	1,687	2,022	1,717	1,695	-231*
Cause of death
Stroke & brain diseases			Ref.	Ref.	Ref.	Ref.	Ref.
Heart disease			1,915	1,051	941	856	110*
Cancer, all			5,333	4,790	4,555	4,797	4,178
COPD			5,586	5,345	5,294	5,278	4,135
Pneumonia & influenza			1,048	1,203	1,182	1,110	1,145
Diabetes			7,891	6,937	6,943	6,814	3,494
Chronic liver disease & cirrhosis			4,021	3,541	3,443	3,531	2,683
Unintentional injuries, homicide, suicide, etc			3,371	3,489	3,310	3,176	1,585
Other			7,503	7,292	7,269	7,213	4,264
DCG Risk Score Geography (HSA) Income quartile (Residence Zip code Median Income)			3,731	3,327	3,393	3,410	2,907
Lowest					1,851	1,831	772
Second lowest					997	982	251*
Third lowest					171*	162*	-287*
Highest					Ref.	Ref.	Ref.
Medicaid Buy-In					-1,963	-1,984	-616
Enrolled for hospice in last 6 months						-784	3,470
End-of-life Interventions
ICU							12,094
Resuscitation, cardiac conversion							-5,501
Ventilation							15,208
Gastrostomy							22,827
Vascular transfusion							11,023
Dialysis							13,072
Chemotherapy							7,039
Cardiac Catheterization							20,377
Heart assist/balloon							7,470
PAP, wedge							2,927
Antibiotic							12,360

All coefficients are significant at p<0.05 except those starred (*).

Model B controls for age and sex. These two factors reduce the cost differences between Blacks and Hispanics and Whites by about 10%. Model C additionally controls for morbidity burden and cause of death, and reduces these differences by another 7 to 9%, with nearly all of the reduction due to morbidity burden. Using the Charlson instead of the DCG morbidity measure produced very similar reductions (Appendix Table A1).

Additionally accounting for place of residence by including hospital service areas (HSAs) in Model D brings unexplained extra costs for Blacks down to only $2,924 (together eliminating 55% of the raw difference for Blacks), only $3,705 (eliminating 68% of the raw difference) for Hispanics, and only $2,103 (eliminating 60%) for decedents of Other races. Using either health care referral region (HRR) or county as the geographical unit instead of HSA in Model D, yields similar findings (Appendix Table A2).

Adding socioeconomic indicators (zip code, median income and Medicaid buy-in) yields very modest further reductions in cost differences by race and ethnicity (Model E).

In total, between 55% and 68% of the raw differences in end-of-life expenditures between Whites and the other three groups are accounted for by differences in demographics, morbidity, geography and socioeconomic indicators, with geography contributing the largest part.

How much of the remaining cost disparity is accounted for by specific end-of-life interventions?

Most of the remaining racial and ethnic differences in end-of-life costs can be attributed to differences in the use of hospital-based, life-sustaining interventions (Table 4). Blacks and Hispanics are significantly more likely to be admitted to the intensive care unit (32.5% for Blacks, 39.6% for Hispanics, 27.0% for Whites). Minorities also receive significantly more intensive procedures, such as resuscitation and cardiac conversion (4.4% of Blacks, 4.0% of Hispanics, 2.7% of Whites), mechanical ventilation (18.0% Blacks, 21.0% Hispanics, 11.6% Whites) and gastrostomy for artificial nutrition (10.5% Blacks, 9.1% Hispanics, 4.1% Whites).

Table 4.

Use of Life-Sustaining Interventions in the Last 6 Months of Life by Racial/Ethnic Group

	White	Black	Hispanic	Other	All
	Percentage of column with any such utilization at the end of life
ICU	27.0	32.5	39.6	30.6	27.5
Resuscitation & cardiac conversion	2.7	4.4	4.0	3.6	2.8
Ventilation	11.6	18.0	21.0	16.6	12.1
Gastrostomy	4.1	10.5	9.1	6.6	4.5
Vascular transfusion	10.4	16.6	18.0	13.7	10.9
Dialysis	1.2	2.7	3.4	2.9	1.3
Chemotherapy	7.9	7.6	7.4	6.8	7.9
Cardiac Catheterization	3.4	2.6	4.6	3.0	3.4
Heart assist/balloon	0.6	0.3	0.8	0.6*	0.5
PAP & wedge	1.3	1.1	1.7	1.5*	1.3
Antibiotic	0.9	1.2	1.6	1.9	1.0
Hospice enrollment, last 6 months	25.9	20.1	22.9	20.3	25.5

Notes:

1. All White / non-White differences in utilization rates, except those starred (*), are statistically significant (p < 0.05).

2. ‘All’ column denotes rates for all Medicare decedents (in 2001), obtained by adjusting for stratified sampling.

In contrast, Whites are slightly more likely to receive inpatient cancer chemotherapy (7.9%) than either Blacks (7.6%) or Hispanics (7.4%), and are more likely than Blacks, but not Hispanics, to receive cardiac catheterization, cardiac balloon assist devices, and pulmonary artery pressure measurements.^{5, 9}

Hospice is used more frequently by Whites (26%) than Blacks (20%) or Hispanics (23%) and is associated with an average reduction in end-of-life expenditures of $784 per beneficiary overall (Model F, Table 3). However, differential hospice use has essentially no effect on racial and ethnic differences in end-of-life costs (Model F).

Finally, controlling for the use of all 10 life-sustaining interventions, such as ICU admissions, ventilators, and gastrostomies (Model G) eliminates more than half of the remaining differences between Whites and each of the other groups (Table 3). Of the original $6,538 excess cost for Black over White decedents, only $997 (15%) remains after this final adjustment, while for Hispanics, only $1,902 (16%) of the original $11,536 in excess costs over Whites remains “unexplained.” Most of the life-sustaining interventions are associated with strikingly higher total costs. For example, among otherwise similar people, those who use the ICU cost $12,000 more than those who do not, use of gastrostomy adds $22,800, and mechanical ventilation, $15,200.

Discussion

This study of nearly 160,000 Medicare decedents and their total (Parts A and B) costs in the last 6 months of life shows substantial differences by race and ethnicity. While there are differences in the magnitude, and relative attribution to causes, of their cost differences with Whites, all three groups of non-White decedents are similar to Whites in their causes of death, but incur substantially higher Medicare expenditures. In each group, prior to considering differences in specific services used, the largest part (between 38% and 58% across the three non-White groups) of the total excess cost over Whites is explained by geography. After controlling for all other factors, use of aggressive end-of-life interventions, such as ICU care, ventilators, and gastrostomies accounts for between 21% and 33% of the difference in total end of life costs. Differences in hospice use contribute little to racial and ethnic differences in total end-of-life costs, both because estimated savings from hospice are small (less than $800) and differences in utilization by race are modest (20% for Blacks; 23% for Hispanics; 26% for Whites).

Using a large and nationally representative sample that linked Medicare data with cause-of-death data from the National Death Index, this study confirms and extends findings regarding Black-White differences in costs at the end of life. Compared to White decedents, raw costs of health care in the last 6 months of life for Black Medicare beneficiaries are 32% higher. Our findings contrast with those of numerous non-end-of-life studies, where minorities received fewer services, and especially fewer technologically intensive interventions.^1-10

Extending end-of-life analyses to Hispanics, their Medicare expenditures are even higher: fully 57% higher than for Whites.

Since racial and ethnic differences in cause of death are minimal, they do not contribute to racial differences in end-of life-costs. However, geography is very important, whether measured as hospital referral regions (HRRs), hospital service areas (HSAs), or county of residence. Blacks and Hispanics are far more likely than Whites to live in large urban areas, where medical care in general, and end-of-life care in particular, is more expensive than in smaller cities and rural areas. Medicare costs for Black and Hispanic decedents are higher largely because more of them live and die in higher cost locations.

However, even within the same geographic locations, Black and Hispanic decedents have notably higher end-of-life Medicare costs than their White neighbors. In contrast to previous studies, our method of using indicators for each geographic unit instead of area-level measures, such as county hospital beds and physician supply, adjusted not only for measured geographical factors related to variations in practice patterns, but also for unknown, and therefore unmeasured, factors.²³

Despite the cumulative importance of age, sex, cause of death, geography, morbidity burden and socioeconomic status on decedent costs, 45% of the excess costs of blacks and 32% of the excess costs for Hispanics are not explained by these factors. Most of this residual difference is accounted for by more end-of-life ICU admissions and life-sustaining interventions for non-Whites. Black Medicare decedents are significantly more likely to receive resuscitation, mechanical ventilation, and gastrostomy for artificial feedings than White decedents—even when residing in the same hospital service area (HSA).¹¹ Hispanics were even more likely than Blacks to receive ICU care, mechanical ventilation, dialysis, and cardiac catheterization.

While Black, Hispanic and other minority decedents receive more intensive life sustaining interventions at the end of life, Blacks receive less cancer chemotherapy, cardiac catheterization, and other aggressive cardiac interventions, such as balloon assist devices, than Whites. This may be because such interventions require access to sub-specialists—oncologists and cardiologists—where Blacks may have fewer prior relationships. It is not clear why the same is not true for Hispanics. Indeed, why Blacks, Hispanics and other minorities receive so much more of many intensive, life-sustaining interventions now emerges as an important area for further research.

Differences in the use of aggressive end-of-life interventions may reflect patient preferences;^24-26 since some studies have found minorities to be more reluctant than Whites to have “Do Not Resuscitate” orders, more likely to prefer life-sustaining treatments at the end-of-life, and less likely to use hospice.^27-35 Such differences in preferences and utilization of high-technology interventions at the end-of-life contrasts with other life stages, where Whites get more intensive interventions.¹⁰ Even if such racial preferences are real, they are not a “first cause;” they raise important policy issues.

Are health care resources for non-Whites mis-allocated over a lifetime, with racial and ethnic minorities receiving fewer life-extending and enhancing interventions than Whites throughout their lives, ^1-10 but more at the end, when there is less opportunity to improve the quantity and quality of life? Perhaps the use of aggressive, hospital-based interventions at the end of life is a well-considered preference. However, even if a choice, this may stem from distrust of the medical care system or from economic constraints;¹⁰ non-Whites who receive timely, effective care throughout their lives may find it easier to reject cardiac resuscitation, mechanical ventilation, and artificial nutrition at the end.^{36, 37} We also know that Blacks receive lower quality primary care than whites^{38, 39} and fewer preventive services.^38-40 Perhaps not having a usual source of care and an established relationship with a physician does not allow for an expression of preferences for less intensive treatments at the end of life. Indeed, we found (in this Medicare population) that more primary care visits just prior to the last 6 months of life are associated with lower costs and less hospital utilization at the end of life.⁴¹

This study has limitations. First, the Medicare claims files contain few clinical descriptors. The findings may not generalize to Medicare managed care, to patients without Medicare coverage, or to patients who do not self-identify as “Black” or “Hispanic.” Previous work has indicated that sensitivity of Medicare data in identifying minorities is good for Blacks but less so for Hispanics.⁴² But specificity (the proportion of identified minorities correctly classified) is high for both. The Hispanic effect that we found on end-of-life costs only pertains directly to mainland Hispanics identified as such in Medicare's database. Percentage reductions in racial and ethnic expenditure differences attributed to individual covariates are only rough guides to their relative importance, especially since how a variable affects a model depends upon what other factors have previously been accounted for. Nonetheless, the primacy of “geography” and “aggressive interventions” in accounting for differences in end-of-life costs is a robust finding (Appendix Table A3). Medicare expenditures do not capture all health care expenditures, especially pharmaceuticals and long term nursing home costs, which may displace some hospital care. Medicare data also do not capture the resources expended by private or government supplemental insurers, or financial or in-kind support from families.⁴³ Importantly, we had no direct data on patient preferences for the various interventions at the end of life. While the models control for most other variables known to influence medical service use—age, sex, race, geography, cause of death, morbidity burden—patient preference, overt or covert racism in how the same providers treat patients, and differences in which providers and systems of care are accessed, might all contribute to these differences in health care system use at the end of life.

Conclusion

Blacks and Hispanics die of similar causes but cost significantly more than Whites in the last 6 months of life. Although 40 to 60% of these excess differences are associated with geography, that is, living in high-medical-expenditure areas, substantial differences remain, even after adjusting for many patient characteristics in addition to where people live. Strikingly higher rates of use of intensive end-of-life treatments such as ICU and ventilators account for most of these residual differences. Thus, at life's end, minorities often receive more expensive, but not necessarily life-enhancing care. It is unclear how much of this was actively sought, or the extent to which racial and ethnic differences are principally driven by how choices are presented or how they are “heard.” These would be fruitful questions for future research.