The Concentration of Hospital Care for BlackVeterans in Veterans Affairs Hospitals:Implications for Clinical OutcomesAshish K. Jha, Roslyn Stone, Judith Lave, Huanyu Chen, Heather Klusaritz, Kevin Volpp

Research QuestionsThrough our research we sought to describe(1) the concentration of care for Black veter-ans within the Veterans Affairs (VA) healthcaresystem; (2) the characteristics of hospitals thatdisproportionately care for Black veterans; (3)whether differences in 30-day mortality ratesfollowing admission differ between minority-and non-minority-serving VA hospitals; and (4)whether racial differences in 30-day mortalityvary significantly across all VA hospitals.

Review of LiteratureThe pervasiveness of racial disparities in health-care is widely known (Smedley, Stith, & Nelson,2003), yet the causes of these disparities are lessunderstood (Jha, Li, Orav, & Epstein, 2005;Vaccarino et al., 2005). Given that the exis-tence of racial disparities in care is well known,understanding why they occur is critically im-portant. The site where minorities receive carehas received increasing attention from policy-

makers as a promising target for interventionsaimed to reduce disparities.

Hospital care for Black Americans is highlyconcentrated: just 5% of U.S. hospitals care fora majority of Black Americans (Jha, Orav, Li, &Epstein, 2007), and the characteristics and per-formance outcomes of these hospitals differfrom hospitals that predominantly care forWhite populations. Furthermore, these minor-ity-serving hospitals care for more Medicaidand poor patients, have fewer nurses, and havemodestly lower quality of care than non-minor-ity-serving hospitals. Because hospital care canbe more expensive for poor patients (Epsteinet al., 1988), the financial stresses of treatingthis population may contribute to substandardquality and worse outcomes seen in minority-serving hospitals (Jha et al., 2007; Skinner,Chandra, Staiger, Lee, & McClellan, 2005).

It is possible that in a large, integratedhealthcare system, the concentration of minor-ities to a small number of providers may be lesspronounced and that such a system would dis-tribute resources more equitably. The VeteransHealth Administration in the Department ofVeterans Affairs in the United States runs thelargest integrated healthcare system in the na-tion. Prior studies have suggested thatdisparities in care may be less pronounced orare not prevalent in the VA (Jha, Shlipak, Ho-smer, Frances, & Browner, 2001; Volpp et al.,2007), although other work has found that caremay favor Black patients (Saha et al., 2008).Whether care for Black patients is similarlyconcentrated among all VA hospitals andwhether minority-serving hospitals within theVA have poorer quality of care is unknown.

Study Design and MethodsDataWe used data from the VA Patient TreatmentFile (PTF) for fiscal years (FYs) 1996–2002. ThePTF is a database of all discharges from VAhospitals. It contains the principal diagnosis(the diagnosis under which the patient wasadmitted), secondary diagnoses, age, gender,

Abstract: Where minorities receive their care may contribute to

disparities in care, yet, the racial concentration of care in the

Veterans Health Administration is largely unknown. We sought to

better understand which Veterans Affairs (VA) hospitals treat

Black veterans and whether location of care impacted disparities.

We assessed differences in mortality rates between Black and

White veterans across 150 VA hospitals for any of six conditions

(acute myocardial infarction, hip fracture, stroke, congestive heart

failure, gastrointestinal hemorrhage, and pneumonia) between

1996 and 2002. Just 9 out of 150 VA hospitals (6% of all VA

hospitals) cared for nearly 30% of Black veterans, and 42 hospitals

(28% of all VA hospitals) cared for more than 75% of Black

veterans. While our findings show that overall mortality rates

were comparable between minority-serving and non-minority-

serving hospitals for four conditions, mortality rates were higher

in minority-serving hospitals for acute myocardial infarction (AMI)

and pneumonia. The ratio of mortality rates for Blacks compared

with Whites was comparable across all VA hospitals. In contrast

to the private sector, there is little variation in the degree of racial

disparities in 30-day mortality across VA hospitals, although

higher mortality among patients with AMI and pneumonia

requires further investigation.

Keywordsmortality

racial and ethnic disparitiesVA quality

veterans

Journal for Healthcare Quality52

Journal for Healthcare QualityVol. 32, No. 6, pp. 52–61

& 2010 National Association forHealthcare Quality

Journal for Healthcare Quality

discharge disposition, transfer status, length ofstay, and other patient-level information. Welinked these data to the annual surveys of theAmerican Hospital Association, which provideshospital characteristics. We also linked the PTFfile to the VA Beneficiary Identification Re-cords Locator Subsystem (BIRLS), whichcontains the date of death of each deceasedveteran whose beneficiaries file for benefits.We verified the date of death for each veteranlisted as dead in the PTF or BIRLS and the vitalstatus for patients of unknown status using theNational Death Index.

Patient PopulationWe used mortality rates for patients with sixconditions considered by the Agency for Health-care Quality and Research (AHRQ) to beinpatient quality indicators: acute myocardial in-farction (AMI), hip fracture, stroke, congestiveheart failure (CHF), gastrointestinal hemor-rhage (GI bleed), and pneumonia (Agency forHealthcare Research and Quality, 2006).

Of all veterans admitted to the VA hospitalsites during FYs 1996–2002, we limited ouranalysis to White and Black veterans due to therelatively small numbers of non-Black minori-ties and less reliable coding of other races(Kressin, Chang, Hendricks, & Kazis, 2003).We excluded hospitalizations for patients whowere o18 years old, nonveterans, treated atfacilities or resided outside of the 50 states, ad-mitted to nonacute facilities, admitted afterhospital transfer, who were female, or missing acode for race. Any patient that was admitted forthe same condition twice within 30 days wasonly counted once. Hospitalizations could beexcluded for more than one of these reasons.The unit of analysis was the hospitalization,with veterans being ‘‘at-risk’’ for mortalitywithin 30 days of admission each time theywere admitted. Further description of ourmethodology appear elsewhere (Volpp et al.,2007).

OutcomesThe primary outcome was all-cause mortalitywithin 30 days of hospital admission for each ofthe six conditions. While the AHRQ’s qualityindicators use only in-hospital mortality, we ex-amined any in-hospital or postdischarge deathswithin 30 days of hospital admission, to elim-inate bias due to length of stay differencesacross hospitals (Iezzoni, 2003).

PredictorsConcentration of care for Black veteransThe predictor of primary interest, the concen-tration of care for Black veterans, was definedby ranking each hospital by the average annualnumber of Black veterans discharged. Averag-ing accounted for hospital closures andmergers over the study period. We then cate-gorized hospitals into quartiles (i.e., thosehospitals in the top quartile were those hospi-tals that had the highest number of Blackpatients discharged and constituted 25% of allBlack patient hospitalizations). We consideredthe hospitals in the top three quartiles to beminority-serving VA hospitals as they cared forat least 75% of all Black veterans.

Hospital characteristicsWe examined the characteristics of hospitalsthat disproportionately cared for Black patientsand compared those characteristics with theother hospitals. A priori, we chose to examinenumber of beds, geographic region (based onthe four census regions), teaching status,the presence of a cardiac intensive care unit(ICU), and the availability of key technologies(angioplasty, coronary artery bypass surgery[CABG], and magnetic resonance imaging[MRI]).

Age and comorbidity adjustmentsBecause our prior work demonstrated that ra-cial disparities in mortality rates varied by agefor these six conditions (Volpp et al., 2007), weconducted all hospitalization-level analyses sep-arately for veterans aged o65 years and thoseaged � 65 years. In adjusted models, a linearage term was included to account for residualage effects. We used the Elixhauser approachfor risk-adjustment, which accounts for thepresence or absence of 30 comorbiditiesunrelated to the primary diagnosis (Elixhaus-er, Steiner, Harris, & Coffey, 1998). Thisapproach has shown better discriminatingpower than other approaches to risk adjust-ment using administrative data (Southern,Quan, & Ghali, 2004; Stukenborg, Wagner, &Connors, 2001).

Statistical AnalysisConcentration of VA hospital care forBlack veteransIn hospital-level analyses, the race-specificnumbers of hospitalizations were summarized

Vol. 32 No. 6 November/December 2010 53

by quartile of average annual hospitalizationsof Black veterans, both overall and separatelyfor each of the six medical conditions. Becausehospitals that merged during the study periodwere counted separately with possibly differentcharacteristics in different time periods, ourhospital-level descriptives show a total of 150hospitals rather than 138.

Characteristics of VA hospitals with high andlow volumes of BlacksPreliminary Chi-squared tests comparing thecharacteristics of hospitals across the four quar-tiles of Black patient volume revealed littleevidence of heterogeneity across the top threequartiles. Therefore, we aggregated the topthree quartiles into ‘‘high volume’’ hospitalsin subsequent analyses. The characteristicsof high- and low-volume hospitals were com-pared using Chi-squared tests with 1 degree offreedom.

Mortality rates following admission to VA hos-pitals in minority- versus non-minority-servingVA hospitalsSeparately for each age group and medicalcondition, we summarized mortality within 30days of admission to minority-serving and non-minority-serving hospitals. We estimated an un-adjusted odds ratio (OR) for the odds ofmortality in high- compared with low-volumehospitals and the corresponding 95% confi-dence interval (CI) using ordinary logisticregression.

To account for clustering on site, we fit two-level random intercept (RI) logistic models us-ing MLwiN vers. 2.02 (Rasbash, Steele, Brown,& Prosser, 2004). The restricted iterative gen-eralized least squares (RIGLS) second-orderpenalized quasilikelihood (PQL2) algorithmwas used to fit these models; RIGLS PQL1 wasused when the PQL2 model would not con-verge (Goldstein, 1991). We compared theodds of overall mortality in minority-servingand non-minority-serving hospitals both unad-justed (RI Model 1) and adjusted for theElixhauser comorbidities and a linear termfor age (RI Model 2). We compared the oddsof race-specific mortality in minority-servingand non-minority-serving hospitals by addingan indicator variable for Black race to RI model2 (RI Model 3).

Racial disparities in 30-day mortality acrossVA hospitalsTo assess whether the effect of Black race var-ied by hospitals’ concentration of Blackpatients, we fit another version of RI Model 3by adding a variable that interacted the minor-ity-serving status of the hospital and patient-level Black race (RI Model 4). To assesswhether the effect of Black race varied acrossall hospitals after adjusting for concentration ofBlacks, the Elixhauser comorbidities, and a lin-ear age term, we also fit a variant of RI Model 3that included a random coefficient for Blackrace (RC Model 1) (Longford, 1993).

All of these random effects models were fitseparately for each age group and medicalcondition. Repeated hospitalizations for thesame patient for the same condition were con-sidered to be conditionally independent. Fixedeffects were tested using Wald statistics, andrandom effect variances were tested using theappropriate Chi-squared mixture distributions(Fitzmaurice, Laird, & Ware, 2004). Interactionterms were assessed using linear contrasts.

Institutional Review Board ApprovalThis study was approved by the University ofPennsylvania Institutional Review Board.

ResultsConcentration of VA Hospital Care ForBlack VeteransOur study sample consisted of 318,610 hospi-talizations of White veterans and 87,927hospitalizations of Black veterans (Table 1).Nine VA hospitals cared for nearly 30% of allhospitalized Black veterans and just 42 hospi-tals (28% of all VA hospitals) cared for 77.2%of hospitalized Black veterans. Conversely, 108VA hospitals (72% of all VA hospitals) cared for65% of all hospitalized White veterans but only22.8% of Black veterans.

Among the 42 minority-serving hospitals,there were 832 Black veterans and 1155 Whiteveterans discharged annually with the six med-ical conditions of interest. In contrast, amongthe other 108 non-minority-serving VA hospi-tals, there were 32 Black veterans and 326White veterans discharged annually with thesesix conditions (Table 1). Most of the minority-serving VA hospitals are located in the easternhalf of the United States, many in the south(Figure 1).

Journal for Healthcare Quality54

The Characteristics of VA Hospitals thatDisproportionately Care for Black VeteransCompared with the 108 non-minority-servinghospitals, the 42 minority-serving VA hospitalswere more often major teaching hospitals(86% vs. 38%, po.001), likely to have cardiacICUs (74% vs. 32%, po.001), and able to pro-vide coronary angioplasty (69% vs. 36%,po.001) and CABG (57% vs. 27%, po.001,Table 2). The minority-serving VA hospitalswere less likely to be small (17% o100 beds,83% 4100 beds, po.001) and more often lo-cated in the South (53% vs. 24%, po.001).

Outcomes in VA Hospitals that Dispropor-tionately Care for Black VeteransWe examined 30-day mortality for six commonmedical conditions for two age groups (thoseunder 65 and those 65 years of age or older) in

the 42 minority-serving and the 108 non-mi-nority-serving VA hospitals. We found that forfour of the six conditions, mortality rates werecomparable for both age groups in minority-serving and non-minority-serving hospitals(Table 3). For example, elderly patients admit-ted with hip fracture had a 10.2% 30-daymortality rate in minority-serving hospitalsand a comparable 10.2% mortality rate innon-minority-serving hospitals (OR 0.99, 95%CI 0.88–1.12). Even after adjusting for otherhospital characteristics, minority and non-mi-nority-serving hospitals had comparablemortality for hip fracture (Table 3, first set ofrows). The mortality rates were also compara-ble in all models for GI Bleed.

Among all patients admitted with CHF andelderly patients admitted with a diagnosis ofstroke, unadjusted mortality rates were lower atminority-serving hospitals compared with non-

Table 1. Distribution of Hospitals, Number of Hospitalizations, and Race-SpecificAverage Annual Number of Hospitalizations by Quartile of AverageAnnual Number of Black Hospitalizations

Characteristic

Minority-Serving Hospitals Non-Minority-Serving HospitalsN 5 42 N 5 108

Top Quartile Second Quartile Third Quartile Bottom Quartile

Number of hospitals 9 12 21 108% of hospitals 6.0 8.0 14.0 72.0Number of

hospitalizationsOverall 52,462 42,836 83,293 227,946% of hospitals 12.9 10.5 20.5 56.1Black 25,836 19,523 22,505 20,063% of Black patients 29.4 22.2 25.6 22.8White 26,626 23,313 60,788 207,883% of White patients 8.4 7.3 19.1 65.3Average annual number of Black hospitalizations overall and by medical conditionOverall 410 264 158 32AMI 31 18 14 3HIP 8 5 3 1Stroke 64 34 21 4CHF 131 89 51 10GI bleed 69 43 25 5Pneumonia 107 75 44 9Average annual number of White hospitalizations overall and by medical conditionOverall 423 309 423 326AMI 60 36 62 47HIP 18 13 19 12Stroke 54 33 47 33CHF 115 84 107 85GI bleed 60 42 60 44Pneumonia 116 101 128 105

Vol. 32 No. 6 November/December 2010 55

minority-serving hospitals. However, after ac-counting for site, comorbidity, and residual ageeffects (RI Model 2), overall 30-day mortalityrates were comparable for both of these con-ditions in minority- and non-minority-servinghospitals.

In contrast, for patients admitted with AMIand pneumonia, unadjusted mortality rateswere significantly higher in minority-servinghospitals for veterans in both age groups. Afteraccounting for site, comorbidity, and residualage effects (RI Model 2), overall 30-day mor-

tality for AMI and pneumonia was still higherin veterans admitted to minority-serving hospi-tals with a high volume of Blacks (Table 3).Even when we compared veterans of the samerace hospitalized in minority-serving and non-minority-serving sites (RI Model 3), we foundthat outcomes were worse in minority-servinghospitals for all patients with AMI and elderlypatients with pneumonia. The odds of deathfor younger patients with pneumonia werenot significantly elevated in minority-servinghospitals.

Figure 1. Geographic Location of VA Hospitals by Quartiles of AverageAnnual Black Patient Volume. The Larger Dots Represent theMinority-Serving Sites and the Smallest Dots Represent the SitesWhere Relatively Few Black Veterans are Hospitalized

Table 2. Characteristics of Minority-Serving and Non-Minority-Serving Hospitals

CharacteristicMinority-Serving Hospitals Non-Minority-Serving Hospitals

P-valueN 5 42 (%) N 5 108 (%)

Number of bedso100 17 45 o.001100–400 71 424400 12 13

RegionWest 17 20 o.001Midwest 24 31South 53 24Northeast 7 21

Major teaching hospital 86 38 o.001Presence of cardiac ICU 74 32 o.001Angioplasty available 69 36 o.001CABG available 57 27 o.001MRI available 44 74 o.001

Journal for Healthcare Quality56

Tabl

e3.

Mor

talit

yRa

tes

atM

inor

ity-S

ervi

ngan

dN

on-M

inor

ity-S

ervi

ngVA

Hosp

itals

,by

Med

ical

Cond

ition

and

Age

Grou

p

Con

ditio

nA

geM

inor

ity-S

ervi

ngH

ospi

tals

N5

42(%

)N

on-M

inor

ity-S

ervi

ngH

ospi

tals

N5

108

(%)

Cru

de�a

Odd

sR

atio

(95%

CI)

RI

Mod

el1�

bO

dds

Rat

io(9

5%C

I)R

IM

odel� 2

cO

dds

Rat

io(9

5%C

I)R

IM

odel

3�d

Odd

sR

atio

(95%

CI)

Hip

�65

10.2

10.2

0.99

0.98

1.02

1.07

(0.8

8,1.

12)

(0.8

6,1.

12)

(0.9

5,1.

09)

(0.9

3,1.

24)

o65

3.0

3.2

0.95

0.93

0.95

1.02

(0.6

0,1.

51)

(0.5

5,1.

58)

(0.7

0,1.

29)

(0.5

5,1.

88)

AM

I�

6516

.014

.91.

09��

1.06

1.12�

1.19��

(1.0

3,1.

16)

(0.9

7,1.

16)

(1.0

6,1.

18)

(1.0

7,1.

33)

o65

5.5

4.8

1.17�

1.16�

1.20��

1.17�

(1.0

3,1.

33)

(1.0

2,1.

32)

(1.1

3,1.

29)

(1.0

2,1.

35)

Stro

ke�

6511

.912

.90.

91��

1.17��

0.90

0.95

(0.8

5,0.

97)

(1.0

8,1.

27)

(0.8

4,0.

95)

(0.8

4,1.

07)

o65

6.7

6.3

1.08

1.09

1.07

1.03

(0.9

5,1.

24)

(0.9

8,1.

22)

(0.9

7,1.

18)

(0.8

4,1.

26)

CH

F�

658.

39.

30.

88��

0.87��

0.94

1.04

(0.8

4,0.

93)

(0.7

9,0.

95)

(0.9

0,0.

99)

(0.9

5,1.

15)

o65

4.3

4.9

0.86��

0.87��

0.97

1.07

(0.7

7,0.

96)

(0.7

6,1.

00)

(0.9

0,1.

05)

(0.9

2,1.

24)

GI

Ble

ed�

657.

47.

41.

001.

011.

051.

08(0

.93,

1.09

)(0

.92,

1.11

)(1

.00,

1.10

)(0

.97,

1.20

)o

655.

44.

91.

101.

091.

101.

13(0

.97,

1.24

)(0

.93,

1.28

)(1

.01,

1.20

)(0

.95,

1.34

)Pn

eum

onia�

6517

.815

.71.

16��

0.90

1.12��

1.15��

(1.1

2,1.

21)

(0.7

9,1.

01)

(1.0

8,1.

17)

(1.0

6,1.

25)

o65

7.8

7.2

1.10�

1.07

1.13�

1.11

(1.0

2,1.

19)

(0.9

0,1.

28)

(1.0

6,1.

19)

(0.9

8,1.

24)

� Odd

sof

deat

haf

ter

the

pati

ent

isdi

sch

arge

dfr

oma

min

orit

y-se

rvin

gh

ospi

tal.

��Po

0.05

.a U

nad

just

edlo

gist

icre

gres

sion

.bR

IM

odel

1:un

adju

sted

ran

dom

inte

rcep

tm

odel

.c R

IM

odel

2:ra

ndo

min

terc

ept

mod

elin

clud

ing

hos

pita

lm

inor

ity-

serv

ing

stat

us,

pati

ent

com

orbi

dity

,an

dag

e.dR

IM

odel

3:ra

ndo

min

terc

ept

mod

elin

clud

ing

hos

pita

lm

inor

ity-

serv

ing

stat

us,

pati

ent

com

orbi

dity

,ag

e,an

dra

ce.

Vol. 32 No. 6 November/December 2010 57

Tabl

e4.

Adju

sted

Odds

Ratio

sof

30-D

ayM

orta

lity

for

Blac

kRe

lativ

eto

Whi

teVe

tera

ns

Con

ditio

nA

ge

RI

Mod

el3a

RC

Mod

el1b

Odd

sR

atio

(bla

cks

vs.

whi

tes)

Var

iabi

lity

inm

orta

lity

acro

sssi

tes

Odd

sR

atio

(bla

cks

vs.

whi

tes)

Var

iabi

lity

inm

orta

lity

acro

sssi

tes

Var

iabi

lity

indi

spar

ityac

ross

site

sP-

valu

eP-

valu

eP-

valu

e

Hip

�65

0.72��

40.

30.

72��

40.

34

0.9

(0.5

8,0.

90)

(0.5

8,0.

90)

o65

0.60

40.

20.

594

0.3

o0.

005

(0.2

5,1.

45)

(0.1

6,2.

16)

AM

I�

650.

74��

o0.

001

0.73��

o0.

001

40.

3(0

.67,

0.83

)(0

.64,

0.82

)o

651.

164

0.3

1.11

40.

34

0.3

(0.9

7,1.

40)

(0.9

0,1.

38)

Stro

ke�

650.

83��

o0.

001

0.82��

o0.

005

40.

7(0

.75,

0.91

)(0

.75,

0.91

)o

651.

15o

0.00

51.

050.

1–.2

40.

2(0

.97,

1.35

)(0

.86,

1.27

)C

HF

�65

0.69��

o0.

001

0.68��

o0.

001

40.

3(0

.65,

0.75

)0.

62,

0.73

)o

650.

72��

o0.

025

0.72��

o0.

025

40.

9(0

.62,

0.83

)(0

.62,

0.83

)G

IB

leed

�65

0.90

o0.

050.

904

0.05

40.

7(0

.80,

1.00

)(0

.81,

1.01

)o

650.

924

0.05

0.92

40.

054

0.9

(0.7

7,1.

09)

(0.7

7,1.

09)

Pneu

mon

ia�

650.

89��

o0.

001

0.89��

o0.

005

40.

9(0

.84,

0.94

)(0

.84,

0.94

)o

651.

07o

0.00

51.

04o

0.05

40.

3(0

.96,

1.18

)(0

.92,

1.17

)

aR

IM

odel

3:ra

ndo

min

terc

ept

mod

elin

clud

ing

hos

pita

lm

inor

ity-

serv

ing

stat

us,

pati

ent

com

orbi

dity

,ag

e,an

dra

ce.

bR

CM

odel

1:ra

ndo

mco

effi

cien

tm

odel

incl

udin

gh

ospi

tal

min

orit

y-se

rvin

gst

atus

,pa

tien

tco

mor

bidi

ty,

age,

and

race

.��

Po0.

05.

Journal for Healthcare Quality58

Racial Differences in 30-Day Mortality inMinority-Serving HospitalsWe found no systematic racial variation in 30-daymortality between minority and non-minority-serving hospitals. For each medical conditionand age group, we found that mortality betweenBlacks and Whites did not vary by site. Further-more, the site-specific 30-day mortality for Blackrelative to White veterans did not vary signifi-cantly across sites for any medical condition orage group except for hip fracture in veteransunder age 65 (RC Model 1; Table 4). The esti-mated 30-day mortality for Black versus Whiteveterans at a ‘‘typical’’ site was similar to the cor-responding estimates in RI Model 3. Even afteraccounting for hospital minority-serving status,patient comorbidity, age, and race, and allowingthe effect of race to vary by site, significant vari-ation in 30-day mortality remained across sites forCHF (both age groups) and AMI among veter-ans aged � 65 years.

LimitationsThere are important limitations to our study.First, we used administrative data to determinerisk-adjusted mortality rates across differenttypes of hospitals. The lack of detailed clinicaladministrative data limits our ability to risk ad-just. However, this was unlikely to bias ourcomparisons of outcomes in minority versusnon-minority-serving hospitals. Second, thereare limitations to the race data in the PTF files,though much of the concern has focused onthe accuracy of data on non-Black minorities.While the coding of race in most datasets haslimitations, the PTF file is generally consideredaccurate for Whites and Blacks (Volpp et al.,2007). It is possible that minority-serving Blackhospitals may in fact serve a substantial numberof non-Black minorities. Third, 30-day mortal-ity rates may not exclusively capture hospital-level effects, as mortality rates may be affectedby community and ambulatory-related factors.Fourth, our categorization of minority- andnon-minority-serving hospitals by the volume ofBlack patients is likely to favor larger institu-tions and would cause some small hospitalswith large proportion of Black patients to beclassified as as non-minority serving. Whileboth approaches to designating minority-serv-ing institutions (proportion based and volumebased) have their strengths and weaknesses, wechose to use volume because the high-volumeinstitutions have the biggest impact on the mi-

nority population. Finally, we could notdetermine if minority-serving hospitals at-tracted more Black patients because of areputation for providing good care to this pop-ulation.

Directions for Future ResearchIn the non-VA setting, hospitals with relativelymore Black patients do seem to have lowerperformance on AMI process measures, al-though these differences are small (Jha et al.,2007). Whether the higher overall AMI mor-tality in minority-serving VA hospitals is due toworse quality of care, greater use of procedureswith known short-term risks, or unmeasuredclinical confounders, is unknown but clearlywarrants further investigation. However, higherrisk due to more interventions would not ex-plain the higher mortality rates among patientswith pneumonia, as most pneumonia patientsrequire no invasive procedures. Understandingwhy these differences exist and how to improveoutcomes in hospitals with large Black popula-tions is critically important.

DiscussionWe examined where Black veterans receive hos-pital care within the VA healthcare system, thelargest integrated healthcare provider in thenation, and found a high degree of concentra-tion: just nine VA hospitals provide care for29.4% of Black veterans, and 28% of VA hospi-tals provide care to 77.2% of Black veterans.Hospitals that care for Black veterans tended tobe mid-sized teaching hospitals, located in theSouth, and generally offered advanced technol-ogies such as cardiac ICUs, angioplasty, andCABG. These minority-serving hospitals hadcomparable outcomes among elderly patientsto non-minority-serving VA hospitals for four ofthe six conditions studied but had higher mor-tality rates for both AMI and pneumonia. Wefound no evidence that racial differences inoutcomes varied between minority- and non-mi-nority-serving VA hospitals.

The degree of concentration within the VAsystem may be surprising given that eligibleveterans can receive care in any VA hospital.However, the degree of concentration almostsurely reflects geographic concentration ofBlack veterans in a small number of commu-nities across the country. Interestingly, care iseven more concentrated in the private sector: arecent study using similar methodology found

Vol. 32 No. 6 November/December 2010 59

that 5% of U.S. hospitals care for over 45% ofelderly Blacks and 25% of hospitals care fornearly 90% of elderly Black Americans (Jhaet al., 2007). Others have found similar levels ofcare concentration in private hospital and am-bulatory care settings (Bach, Pham, Schrag,Tate, & Hargraves, 2004; Skinner et al., 2005).One important difference between these stud-ies and ours is that we included patients of allages, while these others only focused on Med-icare patients.

Despite the degree of racial concentration inthe VA system, overall 30-day mortality inminority-serving hospitals was comparable tonon-minority-serving hospitals for veteranshospitalized with hip fracture, stroke, CHF,and GI bleeding. These results suggest that formany common medical conditions, minorityand non-minority-serving VA hospitals providecomparable care. Also reassuring is that careseems to be concentrated at major teachinghospitals with advanced technological capabil-ities, as many studies suggest that patientoutcomes are better in teaching hospitals(Hartz et al., 1989; Keeler et al., 1992; Rosent-hal, Harper, Quinn, & Cooper, 1997).

VA hospitals that disproportionately care forBlacks had higher mortality for AMI and pneu-monia care. Given that minority-servinghospitals were more likely to have technolo-gies associated with better AMI outcomes, suchas the presence of cardiac catheterization andangioplasty; these findings raise concernsabout the effectiveness of those treatmentsand the selection of patients for these inter-ventions. Hospitals with greater technologicalcapabilities generally perform more interven-tions, and it is possible that higher mortality inminority-serving hospitals could be due to risksassociated with catheterization and revasculari-zation (Barnato, Lucas, Staiger, Wennberg, &Chandra, 2005).

One important finding is that Black patientsdid not have worse outcomes than White pa-tients for any of the six conditions for either ofthe two age groups, within either minority-servingor non-minority-serving hospitals. In fact,among the elderly, Black veterans had consis-tently lower mortality rates compared withWhite veterans, a finding that has been de-scribed recently (Jha et al., 2001; Volpp et al.,2007). One important contribution of ourstudy is the finding that the lower mortalityfor elderly Black veterans and comparablemortality rates for nonelderly Black veterans

was consistent across all VA hospitals. Thesefindings suggest that differences in treatmentbetween Blacks and Whites, if they exist, maylead to worse outcomes rather than better out-comes for Blacks in VA hospitals. The strikinglack of variation is consistent with the notionthat the VA provides standardized care regard-less of race in all its facilities.

ConclusionsThe pattern of racial concentration within VAhospitals appears similar to the non-VA setting.Mortality rates were predominantly similaroverall in minority-serving and non-minority-serving hospitals. The lack of racial variation in30-day mortality across VA hospitals stands instark contrast to the variability observed in non-VA hospitals and may reflect the efforts of theVA to provide equal access to treatment re-gardless of race.

ReferencesAgency for Healthcare Research and Quality. (2006). In-

patient quality indicators overview. Retrieved 3 August 2007,from www.qualityindicators.ahrq.gov/iqi_overview.htm

Bach, P. B., Pham, H. H., Schrag, D., Tate, R. C., & Har-graves, J. L. (2004). Primary care physicians who treatblacks and whites. New England Journal of Medicine, 351,575–584.

Barnato, A. E., Lucas, F. L., Staiger, D., Wennberg, D. E., &Chandra, A. (2005). Hospital-level racial disparities inacute myocardial infarction treatment and outcomes.Medical Care, 43, 308–319.

Elixhauser, A., Steiner, C., Harris, D. R., & Coffey, R. M.(1998). Comorbidity measures for use with administra-tive data. Medical Care, 36, 8–27.

Epstein, A. M., Stern, R. S., Tognetti, J., Begg, C. B., Hartley,R. M., Cumella, E. Jr., et al. (1988). The association ofpatients’ socioeconomic characteristics with the lengthof hospital stay and hospital charges within diagnosis-related groups. New England Journal of Medicine, 318,1579–1585.

Fitzmaurice, G. M., Laird, N. M., & Ware, J. H. (2004).Applied longitudinal analysis. Hoboken, NJ: John Wileyand Sons Inc.

Goldstein, H. (1991). Nonlinear multilevel models, with anapplication to discrete response data. Biometrika, 78, 45–51.

Hartz, A. J., Krakauer, H., Kuhn, E. M., Young, M., Jacob-sen, S. J., Gay, G., et al. (1989). Hospital characteristicsand mortality rates. New England Journal of Medicine, 321,1720–1725.

Iezzoni, L. (Ed.). (2003). Risk adjustment for measuring health-care outcomes (3rd ed.). Chicago, IL: Health Administra-tion Press.

Jha, A. K., Li, Z., Orav, E. J., & Epstein, A. M. (2005). Carein U.S. hospitals—the hospital quality alliance program.New England Journal of Medicine, 353, 265–274.

Jha, A. K., Orav, E. J., Li, Z., & Epstein, A. M. (2007). Con-centration and quality of hospitals that care for elderlyblack patients. Archives in Internal Medicine, 167, 1177–1182.

Journal for Healthcare Quality60

Jha, A. K., Shlipak, M. G., Hosmer, W., Frances, C. D., &Browner, W. S. (2001). Racial differences in mortalityamong men hospitalized in the Veterans Affairs healthcare system. JAMA, 285, 297–303.

Keeler, E. B., Rubenstein, L. V., Kahn, K. L., Draper, D.,Harrison, E. R., McGinty, M. J., et al. (1992). Hospitalcharacteristics and quality of care. JAMA, 268, 1709–1714.

Kressin, N. R., Chang, B. H., Hendricks, A., & Kazis, L. E.(2003). Agreement between administrative data andpatients’ self-reports of race/ethnicity. American Journalof Public Health, 93, 1734–1739.

Longford, N. T. (1993). Random coefficient models. New York:Oxford University Press.

Rasbash, J., Steele, F., Brown, W., & Prosser, B. (2004). Auser’s guide to MLwiN version 2.0. London: Institute ofEducation.

Rosenthal, G. E., Harper, D. L., Quinn, L. M., & Cooper, G.S. (1997). Severity-adjusted mortality and length of stayin teaching and nonteaching hospitals. Results of a re-gional study. JAMA, 278, 485–490.

Saha, S., Freeman, M., Toure, J., Tippens, K. M., Weeks, C.,& Ibrahim, S. (2008). Racial and ethnic disparities in theVA health care system: A systematic review. Journal ofGeneral Internal Medicine, 23, 654–671.

Skinner, J., Chandra, A., Staiger, D., Lee, J., & McClellan,M. (2005). Mortality after acute myocardial infarction inhospitals that disproportionately treat black patients.Circulation, 112, 2634–2641.

Smedley, B. D., Stith, A. Y., & Nelson, A. R. (Eds.). (2003).Unequal treatment: Confronting racial and ethnic disparitiesin health care. Washington, DC: National Academy ofSciences Press.

Southern, D. A., Quan, H., & Ghali, W. A. (2004). Com-parison of the Elixhauser and Charlson/Deyo methodsof comorbidity measurement in administrative data.Medical Care, 42, 355–360.

Stukenborg, G. J., Wagner, D. P., & Connors, A. F. Jr.(2001). Comparison of the performance of two comor-bidity measures, with and without information fromprior hospitalizations. Medical Care, 39, 727–739.

Vaccarino, V., Rathore, S. S., Wenger, N. K., Frederick, P.D., Abramson, J. L., Barron, H. V., et al. (2005). Sex andracial differences in the management of acute myocar-dial infarction, 1994 through 2002. New England Journalof Medicine, 353, 671–682.

Volpp, K. G., Stone, R., Lave, J. R., Jha, A. K., Pauly, M.,Klusaritz, H., et al. (2007). Is thirty-day hospital mortalityreally lower for black veterans compared with white vet-erans? Health Service Research, 42, 1613–1631.

Authors’ BiographiesAshish K. Jha, MD, MPH, is an Associate Professor ofHealth Policy at the Harvard School of Public Health, As-sistant Professor of Medicine at Harvard Medical School,and Staff Physician at VA Boston Healthcare System andBrigham and Women’s Hospital. He is currently serving asa senior advisor to the Under Secretary for Health of theVeterans Health Administration, focusing on areas of clin-ical quality and patient safety.

Roslyn Stone, PhD, is Associate Professor of Biostatistics atthe University of Pittsburgh. Her research interests includegeneralized linear models, survival analysis, multilevelmodels, statistical methods for occupational and environ-mental epidemiology, guideline implementation, andcluster-randomized studies.

Judith Lave, PhD, is the Chair of the Department of HealthPolicy and Management and Director of the MHA and JD/MPH Programs at University of Pittsburgh School of PublicHealth. She is also the Director of the Pennsylvania Med-icaid Policy Center.

Huanyu Chen, PhD, is a Project Coordination at the Cen-ter for Health Equity Research and Promotion, VAPittsburgh Healthcare System. She is also at the School ofSocial Policy and Practice, University of Pennsylvania andthe Department of Medicine, Center for Biomedical In-formatics, University of Pittsburgh School of Medicine.

Heather Klusaritz, MSW, is a doctoral student at the Uni-versity of Pennsylvania, School of Social Policy andPractice, where she also received her MSW. Her researchinterests focus on the intersection between healthcare andsocial welfare policy, specifically access to healthcare fordisadvantaged populations.

Kevin Volpp, MD, PhD, is an Associate Professor of Med-icine and Healthcare Management and Director of theLeonard Davis Institute of Health Economics Center forHealth Incentives at the Wharton School of the University ofPennsylvania. His research interests include health services,health economics, delivery of healthcare, quality of health-care, and behavioral economics.

For more information on this article, contact Ashish K. Jhaat ajha@hsph.harvard.edu.

Vol. 32 No. 6 November/December 2010 61