At the Intersection of Health, Health Care and Policy

doi: 10.1377/hlthaff.24.2.516 

, 24, no.2 (2005):516-526Health AffairsUse Of Geocoding In Managed Care Settings To Identify Quality Disparities

José J. Escarce, Thomas Horstman and Thomas RectorAllen M. Fremont, Arlene Bierman, Steve L. Wickstrom, Chloe E. Bird, Mona Shah,

Cite this article as:

  http://content.healthaffairs.org/content/24/2/516.full.html

available at: The online version of this article, along with updated information and services, is

 

For Reprints, Links & Permissions: http://healthaffairs.org/1340_reprints.php

http://content.healthaffairs.org/subscriptions/etoc.dtlE-mail Alerts : http://content.healthaffairs.org/subscriptions/online.shtmlTo Subscribe:

from the Publisher. All rights reserved.including photocopying or by information storage or retrieval systems, without prior written permission may be reproduced, displayed, or transmitted in any form or by any means, electronic or mechanical,

Health Affairs Foundation. As provided by United States copyright law (Title 17, U.S. Code), no part of by Project HOPE - The People-to-People Health2005Bethesda, MD 20814-6133. Copyright ©

is published monthly by Project HOPE at 7500 Old Georgetown Road, Suite 600,Health Affairs

Not for commercial use or unauthorized distribution

by guest on June 12, 2013Health Affairs by content.healthaffairs.orgDownloaded from

Use Of Geocoding InManagedCare Settings To IdentifyQuality DisparitiesHow indirect measures of race/ethnicity and socioeconomic statuscan be used by the nation’s health plans to demonstrate disparities.

by Allen M. Fremont, Arlene Bierman, Steve L. Wickstrom, Chloe E. Bird,Mona Shah, José J. Escarce, Thomas Horstman, and Thomas Rector

ABSTRACT: Tracking quality-of-care measures is essential for improving care, particularlyfor vulnerable populations. Although managed care plans routinely track quality measures,few examine whether their performance differs by enrollee race/ethnicity or socioeco-nomic status (SES), in part because plans do not collect that information. We show thatplans can begin examining and targeting potential disparities using indirect measures ofenrollee race/ethnicity and SES based on geocoding. Using such measures, we demon-strate disparities within both Medicare+Choice and commercial plans on Health Plan Em-ployer Data and Information Set (HEDIS) measures of diabetes and cardiovascular care, in-cluding instances in which race/ethnicity and SES have distinct effects.

Di s par it i e s in qual ity of care based on race/ethnicity and socioeco-nomic status (SES) are well documented. Eliminating these disparities isessential to improving health in the United States. Tracking measures of

the quality of care received by different racial/ethnic and SES groups to help targetquality improvement efforts and monitor progress represents a crucial step to-ward this goal. This is especially true for managed care plans, which now servemore than half of all Americans.1 Yet relatively little is known about disparities inmanaged care settings.

Although managed care plans routinely collect and monitor performance onquality measures, such as the National Committee for Quality Assurance (NCQA)Health Employer Data and Information Set (HEDIS), few plans examine thesedata for possible disparities based on enrollees’ race/ethnicity or SES. A key bar-

5 1 6 M a r c h / A p r i l 2 0 0 5

D a t a W a t c h

DOI 10.1377/hlthaff.24.2.516 ©2005 Project HOPE–The People-to-People Health Foundation, Inc.

Allen Fremont (fremont@rand.org), Chloe Bird, and José Escarce are with RAND Health in Santa Monica,California. Arlene Bierman is with the University of Toronto (Ontario). Steve Wickstrom, Mona Shah, ThomasHorstman, and Thomas Rector are with the Center for Health Care Policy and Evaluation, Minneapolis,Minnesota. Fremont is also affiliated with the Department of Medicine, West Los Angeles Veterans AffairsMedical Center and with the Division of General Internal Medicine, University of California, Los Angeles(UCLA) in Westwood. Escarce shares this latter affiliation as well.

by guest on June 12, 2013Health Affairs by content.healthaffairs.orgDownloaded from

rier is the fact that few plans routinely collect race/ethnicity and SES information,which could then be linked to performance on quality measures.2 The reasonsplans do not collect this information reflect concerns such as uncertainty aboutthe legality of collecting racial/ethnic data and fear that some consumers or advo-cacy groups would assume that plans were using this information inappropriately,perhaps to avoid enrolling or reenrolling minority patients.

Despite these concerns, plans’ interest in collecting these data to identify andtarget potential disparities is increasing.3 At least one large managed care organi-zation (MCO) has already embarked on a well-publicized and -received effort tocollect race/ethnicity information to identify and address disparities. Althoughspecific results of this effort are not yet available, the plan’s initial experiences sug-gest that it takes several years for even a committed plan to obtain this informa-tion from a sufficient sample of enrollees to adequately assess disparities.

Plans’ interest in examining disparities in care is being facilitated by the Cen-ters for Medicare and Medicaid Services (CMS), which decided to provideMedicare+Choice (M+C) plans with race/ethnicity information on their enrollees.This decision was supported by several studies that linked the CMS race/ethnic-ity information to HEDIS performance data on selected measures from a nationalsample of M+C plans.4 These studies showed substantial disparities on a varietyof measures and helped demonstrate the feasibility of using CMS race/ethnicitydata and HEDIS measures to identify disparities. Nevertheless, since these studiesinvolved only M+C enrollees, it remains unclear whether disparities also existamong commercial enrollees. Further, although CMS race/ethnicity data may offera reasonable alternative to collecting data directly from enrollees, most managedcare enrollees are in commercial plans, which lack access to CMS data. These datado not include information about enrollees’ SES other than Medicaid eligibility.Finally, these studies used data pooled across a national sample of several hundredplans, so it is unclear whether disparities are limited to a minority of plans or areconsistent across plans.

For this study we estimated enrollees’ race/ethnicity and SES indirectly basedon geocoding and then used that information to examine disparities on HEDISmeasures of quality of care for cardiovascular disease and diabetes.

Study Data And MethodsGeocoding is a method in which certain characteristics of a person are esti-

mated based on characteristics of the person’s area or neighborhood. Plans canperform this straightforward process quickly and inexpensively. Although geo-coding does not provide precise estimates of a given person’s characteristics, it canprovide reasonably accurate estimates when applied to a group or population.5

We used geocoded measures to examine racial/ethnic and SES disparities forM+C and commercial plan enrollees. We analyzed HEDIS process-of-care mea-sures for diabetes and cardiovascular care and also conducted secondary analyses

I n d i r e c t M e a s u r e s

H E A LT H A F F A I R S ~ V o l u m e 2 4 , N u m b e r 2 5 1 7

by guest on June 12, 2013Health Affairs by content.healthaffairs.orgDownloaded from

of disparities on HEDIS intermediate outcome measures. To assess the validity ofusing geocoded measures in estimating enrollees’ race/ethnicity and associateddisparities in care, we compared results on a subset of measures for M+C enrolleeswhen race is based on geocoding versus individual-level race information from theCMS. To assess how disparities are distributed across plans, we also examineddisparities on selected measures separately within individual plans.

� Data sources. For our primary analyses (disparities on selected HEDIS proc-ess measures), we obtained data from nine M+C and ten commercial plans. Whendata were obtained in 2000, these plans served approximately 195,116 Medicare and2,151,050 commercial enrollees across four regions. Data on enrollment, patient ad-dress, medical care encounters, and claims were contained in a research databasemaintained by the MCO of which the nineteen plans were part.

The research database lacked chart data needed to calculate HEDIS intermedi-ate outcome measures. For these analyses, we obtained enrollee-level performancedata directly from plans. Unfortunately, only one in the research database used forour primary analyses was able to provide intermediate outcome data during thedata collection phase. However, we obtained intermediate outcome data fromnine additional plans in the same MCO. Together these ten plans included threeM+C and seven commercial plans, representing 92,483 M+C and 1,763,292 com-mercial enrollees, respectively.

� Measures of race/ethnicity and SES. Our measures of race/ethnicity andSES were derived from geocoded data. This approach takes advantage of the correla-tion between neighborhoods’ sociodemographic characteristics and certain charac-teristics of their residents. We geocoded to the census block group level, which cor-responds to a small neighborhood with approximately 1,000 residents and tends toprovide more precise estimates than larger areas such as ZIP codes.

Some characteristics are well suited to estimating from geocoding; others arenot. For example, geocoding can be used to identify people likely to be black, sinceracially segregated neighborhoods persist in many U.S. regions and, thus, peopleliving in predominantly black neighborhoods are likely to be black themselves.6

Geocoding is less well suited for identifying likely Asians and Hispanics, sincethese groups tend to live in less segregated neighborhoods. Other techniques suchas surname analyses have proved to be more reliable methods of determining therace/ethnicity of Hispanics and Asians when this information is not otherwiseavailable.7 Thus, in this study we focus on a single geocoded measure of race/eth-nicity: whether or not an enrollee lives in a predominantly black block group, de-fined as a neighborhood where more than 66 percent of residents are black.

Geocoding also can provide reasonable estimates of SES. For example, enrolleesliving in poor neighborhoods tend to be far less affluent than those living in neigh-borhoods without poverty. We computed several different SES measures, butsince results were similar for each measure, we report only one measure, living in apoor neighborhood, which we defined as a block group where more than 20 per-

5 1 8 M a r c h / A p r i l 2 0 0 5

D a t a W a t c h

by guest on June 12, 2013Health Affairs by content.healthaffairs.orgDownloaded from

cent of the residents have an income below the federal poverty level.� Geocoding process. We used a vendor, Mapping Analytics, to convert en-

rollees’ addresses to the corresponding census block group number. All but 8 per-cent of addresses were matched to the block group level, which is an excellent rate.Enrollees with matched and unmatched addresses were similar with respect to age,sex, insurance type, plan, medical condition, and race/ethnicity (for M+C enrollees).We then linked each enrollee’s block group number to selected information aboutthat block group from 1990 census data (2000 census data were not yet available)and calculated the measure of race/ethnicity and SES described above.

Although our geocoded measures of SES were previously validated, the geo-coded measure of race/ethnicity was not.8 To help validate this measure, we alsoobtained race/ethnicity information on individual M+C enrollees from the CMS.Individual-level measures of race/ethnicity (such as self-reported race/ethnicity)for commercial enrollees were unavailable.

� Quality measures. Our analyses focused on six HEDIS 2000 measures ofwhether specific processes of care were performed for eligible enrollees: in diabetics,an annual check of glycosolated hemoglobin (HbA1c), low density lipoprotein(LDL), and urine protein levels, as well as a dilated eye exam; beta-blocker prescrip-tion for myocardial infarction (MI) patients; and LDL check in patients after a car-diac event.9 We focused on the process measures because the NCQA permits plansto calculate these from administrative data alone. Thus, plans can easily calculatethese measures for all eligible enrollees, increasing the likelihood of detecting dis-parities. We used this approach, calculating performance (that is, percentage of eli-gible enrollees receiving an indicated service) for all eligible enrollees.

We also assessed disparities for four HEDIS intermediate outcome measures:adequately controlled LDL cholesterol after a cardiac event, blood pressure inhypertensives, and HbA1c and LDL in diabetics. For this paper we treated assess-ments with these measures as a secondary analysis, since we had data from only afew plans and enrollees for them. In contrast to the process measures, for whichwe could assess care for all eligible enrollees directly from administrative data, wewere limited to samples of approximately 411 enrollees (the NCQA’s suggestedsample size) or fewer from each of the ten plans that provided us with these data.Consequently, our capacity to detect racial/ethnic and SES disparities was gener-ally more limited than for the process measures.

� Analysis. In our primary analyses, we used chi-square tests to compare thepercentage of eligible patients in each racial/ethnic or SES subgroup who receivedthe service specified by each process measure. We conducted all comparisons sepa-rately for M+C and commercial enrollees. Since specifications were designed so thatall eligible patients should receive the indicated care regardless of other factors, wefocused on unadjusted comparisons as recommended by the NCQA.

We also conducted several secondary analyses. First, we compared results ofthe racial disparities analyses above using the geocoded measure of race/ethnicity

I n d i r e c t M e a s u r e s

H E A LT H A F F A I R S ~ V o l u m e 2 4 , N u m b e r 2 5 1 9

by guest on June 12, 2013Health Affairs by content.healthaffairs.orgDownloaded from

with those obtained when a more direct measure of race/ethnicity from CMS datawas used. Second, we used multivariable logistic regression to compare the ad-justed probability of M+C and commercial plan enrollees in each subgroup receiv-ing specified care. To address the possibility that any disparities observed reflectdifferential enrollment of minority or low-SES enrollees into plans with lower av-erage performance, we initially adjusted for the specific plan from which theenrollee obtained care, and then we added sociodemographic characteristics in-cluding enrollees’ age, sex, and, depending on the model, race or SES (or both).Third, to assess the extent to which disparities varied within plans, we repeatedthe primary analyses at the individual plan level. We limited these analyses to thefour diabetes process measures, which had the largest denominators (the numberof eligible enrollees). In our final analyses, we examined unadjusted and adjustedracial/ethnic and SES disparities on the intermediate outcomes measures.

Study ResultsThe number of enrollees eligible for each HEDIS process measure and their

characteristics are shown in Exhibit 1. Far more enrollees met NCQA eligibilityrequirements for the diabetes measures than for the two cardiac measures. M+Cenrollees were older than commercial enrollees, with a higher percentage living inpredominantly black or poor neighborhoods. A higher percentage of diabetic-than cardiac-eligible enrollees also lived in these neighborhoods. Exhibit 2 showssimilar information for the HEDIS intermediate outcome measures.

Exhibit 3 shows the percentage of enrollees across all plans studied who re-

5 2 0 M a r c h / A p r i l 2 0 0 5

D a t a W a t c h

EXHIBIT 1Eligible Enrollee Characteristics, By HEDIS Process Measure And Plan Type, 2000

Characteristic

Beta-blockerafter MI

LDLcheckafter cardiacevent

HbA1ccheck fordiabetics

LDLcheck fordiabetics

Eyeexam fordiabetics

Urineproteincheck fordiabetics

Medicare+ChoiceNumber of eligible enrolleesAge (mean years)Percent femalePercent living in predominantly

black neighborhoodPercent living in poor neighborhood

3337340%

11%13%

9926939%

6%13%

15,2526853%

21%25%

15,2526853%

21%25%

15,2526853%

21%25%

15,2526853%

21%25%

CommercialNumber of eligible enrolleesAge (mean years)Percent femalePercent living in predominantly

black neighborhoodPercent living in poor neighborhood

7275321%

3%11%

2,3045623%

3%11%

34,9815245%

9%15%

34,9815245%

9%15%

34,9815245%

9%15%

34,9815245%

9%15%

SOURCES: Managed care organization’s administrative database; and data from the U.S. Census Bureau.

NOTES: HEDIS is Health Plan Employer Data and Information Set. MI is myocardial infarction. LDL is low density lipoprotein.HbA1C is hemoglobin A1c or glycosolated hemoglobin. Eligibility criteria for the four diabetic measures were the same.

by guest on June 12, 2013Health Affairs by content.healthaffairs.orgDownloaded from

ceived indicated processes of care by racial and SES subgroup and plan type. Ra-cial and SES disparities among M+C enrollees were present for the majority ofprocess measures. The largest disparities were between racial groups on beta-blocker after MI and LDL checks. Sizable SES disparities were evident as well.

The pattern of disparities on process measures was similar for commercial en-rollees. However, disparities were generally smaller than for M+C enrollees; thelargest disparity for commercial enrollees was for LDL check after a cardiac event.In addition, SES disparities were somewhat more common than racial disparitiesacross measures, whereas the converse was the case for M+C enrollees.

When we compared estimated performance and disparities in M+C plansbased on CMS measures of race/ethnicity for individual enrollees with results us-ing our geocoded measure, the results were essentially the same (Exhibit 4). In-deed, disparities differed by less than one percentage point for all six measures.

Multivariate adjustments for the specific M+C or commercial plan in which en-rollees received their care reduced some disparities, but they did not alter the ba-sic results. Additional adjustments for sociodemographic factors further reduceddisparities’ size, and some became insignificant (results not shown). However,significant racial and SES disparities were still present on several measures forM+C and commercial enrollees. For commercial enrollees, SES disparities weresignificant on five of the six measures after adjustments.

Measures of race and SES each remained significant on several measures despite

I n d i r e c t M e a s u r e s

H E A LT H A F F A I R S ~ V o l u m e 2 4 , N u m b e r 2 5 2 1

EXHIBIT 2Eligible Enrollee Characteristics by HEDIS Intermediate Outcome Measure And PlanType, 2000

CharacteristicLDL controlled aftercardiac eventa

Blood pressurecontrolled inhypertensives

HbA1C controlledin diabetics

LDL controlledin diabetics

Medicare+ChoiceNumber of eligible enrolleesAge (mean years)Percent femalePercent living in predominantly

black neighborhoodPercent living in poor neighborhood

5587039%

7%15%

1,2637362%

11%15%

1,2426954%

13%17%

1,2426954%

13%17%

CommercialNumber of eligible enrolleesAge (mean years)Percent femalePercent living in predominantly

black neighborhoodPercent living in poor neighborhood

1,6215620%

3%8%

2,5465648%

6%11%

2,8125143%

7%14%

2,8125143%

7%14%

SOURCES: Health Plan Employer Data and Information Set (HEDIS) performance data supplied by individual plans; managedcare organization’s administrative database; and data from the U.S. Census Bureau.

NOTES: LDL is low density lipoprotein. HbA1C is hemoglobin A1c or glycosolated hemoglobin. Eligibility criteria for the twodiabetic measures were the same.a The relatively low number of eligible enrollees for LDL control after a cardiac event reflects the fact that because of stricteligibility criteria, most plans had fewer eligible enrollees for this measure than the sample size of 411 recommended by theNational Committee for Quality Assurance (NCQA) for other measures.

by guest on June 12, 2013Health Affairs by content.healthaffairs.orgDownloaded from

5 2 2 M a r c h / A p r i l 2 0 0 5

D a t a W a t c h

EXHIBIT 3Racial And Socioeconomic Status (SES) Disparities By Plan Type: Process Measures

Racial disparity SES disparity

Process measure Other (%)

Lives inpredominantlyblackneighborhood (%) Difference Other (%)

Lives inpoorneighborhood(%) Difference

Medicare+ChoiceBeta-blocker after MILDL after cardiac eventHbA1c check for diabeticsLDL check for diabeticsEye exam for diabeticsUrine protein check for

diabetics

63.061.964.459.132.3

18.3

40.045.349.937.838.6

19.4

23.1***16.5***14.5****21.3****7.3****

–1.1

63.261.864.259.446.3

18.3

44.454.152.840.239.0

19.3

18.8**7.7

11.4****19.2****–0.2

1.0

CommercialBeta-blocker after MILDL after cardiac eventHbA1c check for diabeticsLDL check for diabeticsEye exam for diabeticsUrine protein check for

diabetics

–a

67.666.261.836.1

20.1

–a

57.161.655.435.0

21.7

–a

10.44.6****6.4****1.1

–0.8

71.168.666.862.536.6

21.3

74.756.360.253.532.5

19.0

3.612.3****6.6****9.0****4.2****

2.3****

SOURCES: Managed care organization’s administrative database; and data from the U.S. Census Bureau.

NOTES: Difference is in percentage points. MI is myocardial infarction; LDL is low density lipoprotein; HbA1C is hemoglobin A1cor glycosolated hemoglobin.a Too few enrollees eligible for this measure who live in a predominantly black census block group (< 30 cases) for reliableestimates.

**p < .05 ***p < .01 ****p < .001

EXHIBIT 4Performance On Process Measures For Medicare+Choice (M+C) Plan Enrollees AndRacial Disparities When Race/Ethnicity Is Based On CMS Data Versus GeocodedInformation, 2000

CMS data Geocoded data

Process measure, M+C Other (%) Black (%) Difference Other (%)

Lives inpredominantlyblackneighborhood (%) Difference

Beta-blocker prescribed after MILDL after cardiac eventHbA1c check for diabeticsLDL check for diabeticsEye exam for diabeticsUrine protein check for diabetics

60.662.465.660.946.618.0

37.845.251.139.339.019.8

22.7***17.2****14.5****21.6****7.6****

–1.9***

63.161.964.459.145.918.3

40.045.349.937.838.619.4

23.1***16.5***14.5****21.3****7.3****

–1.1

SOURCES: Person-level race data from the Centers for Medicare and Medicaid Services (CMS); other data are from themanaged care organization’s administrative database and from the U.S. Census Bureau.

NOTES: Difference is in percentage points. We coded race/ethnicity information from the CMS as either “black” or “other”(nonblack) to make the coding consistent with the geocoded measures. In addition, coding CMS race/ethnicity data in this wayis reasonable, since the “black” category is accurate in CMS data, but the “white” category includes a sizable number ofnonwhites, such as nonblack Hispanics.

***p < .01 ****p < .001

by guest on June 12, 2013Health Affairs by content.healthaffairs.orgDownloaded from

statistical adjustments for both. For example, although the patterns of racial andSES disparities were similar, the adjusted analyses for LDL check in diabetic M+Cenrollees not only showed a disparity of eight percentage points associated withrace (p < .01), but also revealed a distinct effect of SES that was associated with anadditional disparity of eight percentage points (p < .01).

Exhibits 5 and 6 illustrate the results of our disparities analysis within individ-ual plans. When we observed a disparity on a given measure (such as LDL checkfor diabetics) in the analyses of data pooled across the plans (that is, the primaryanalysis), we found that about half of the individual plans had a significant dispar-ity as well. Several other plans showed similar but insignificant differences in care.Interestingly, disparity size was not consistently associated with a plan’s overallperformance on a given measure.

The final results (Exhibit 7) focus on disparities on the intermediate outcomemeasures across the M+C and commercial plans. Despite limited sample sizes, sig-nificant—and generally large—racial disparities were evident for M+C enrolleeson all but one measure, and SES disparities were present for all four. The patternwas less consistent for commercial enrollees, with fewer disparities evident, al-though some of those were large. For example, there was a racial disparity ofnearly fifteen percentage points (p < .05) on the LDL control after cardiac eventmeasure.

Discussion� Promise of geocoding. Geocoding offers managed care plans and other health

care providers the opportunity to obtain the race/ethnicity and SES data they needto begin addressing care disparities now. With the data geocoding provides, the

I n d i r e c t M e a s u r e s

H E A LT H A F F A I R S ~ V o l u m e 2 4 , N u m b e r 2 5 2 3

EXHIBIT 5Racial Disparities In LDL Checks Among Diabetic Enrollees In Medicare+Choice PlansIn The Managed Care Organization Studied, 2000

SOURCE: Data from managed care organization's administrative database.NOTES: LDL is low density lipoprotein. Difference between “other” and “predominantly black neighborhood” significant at the.05 level for plans A–F.

80

60

40

20

0

Percent receiving check

A B C D EMedicare+Choice plan

F G H I

Other

Live in predominantly black neighborhood

by guest on June 12, 2013Health Affairs by content.healthaffairs.orgDownloaded from

plans can monitor quality of care and target best practices to the sites and groups ofpeople who need them the most. Later, if and when the plans have individual-level

5 2 4 M a r c h / A p r i l 2 0 0 5

D a t a W a t c h

EXHIBIT 7Racial And Socioeconomic Status (SES) Disparities By Plan Type: IntermediateOutcome Measures, 2000

Racial disparity SES disparity

Measure Other (%)

Lives inpredominantlyblackneighborhood (%) Difference Other (%)

Lives inpoorneighborhood(%) Difference

Medicare+ChoiceLDL controlled after

cardiac eventBP controlled in

hypertensivesHbA1c controlled in

diabeticsLDL controlled in diabetics

42.4

33.5

62.138.9

29.3

21.8

46.524.5

13.1

11.7***

15.5****14.3****

43.1

33.9

62.038.5

31.2

23.0

50.529.7

11.4**

10.8***

11.6**8.8***

CommercialLDL controlled after

cardiac eventBP controlled in

hypertensivesHbA1c controlled in

diabeticsLDL controlled in diabetics

38.9

27.9

43.530.7

24.1

29.9

35.729.5

14.9**

–2.0

7.8**1.2

38.4

28.0

43.731.0

37.5

28.2

38.228.3

0.9

–0.2

5.6**2.7

SOURCES: Health Plan Employer Data and Information Set (HEDIS) performance data supplied by individual plans; managedcare organization’s administrative database; and data from the U.S. Census Bureau.

NOTES: Difference is in percentage points. LDL is low density lipoprotein. BP is blood pressure. HbA1C is hemoglobin A1c orglycosolated hemoglobin.

**p < .05 ***p < .01 ****p < .001

EXHIBIT 6

Racial Disparities In LDL Checks Among Diabetic Enrollees In Commercial Plans In

The Managed Care Organization Studied, 2000

by guest on June 12, 2013Health Affairs by content.healthaffairs.orgDownloaded from

data (for example, self-reported race/ethnicity by enrollee), they can target their ef-forts to individuals and continue to use geocoding simultaneously to monitor qual-ity in neighborhoods and groups of enrollees who may differ in their race/ethnicitybut share other characteristics, such as low SES or limited access to public transpor-tation, that can undermine their care.

� Both race and SES effects. Our study also demonstrates the importance ofexamining the effects of both race and SES. Although race/ethnicity and SES are cor-related and disparities associated with these two characteristics linked, our studyshows that on several quality measures, race and SES exert independent effects.That is, although the patterns of racial and SES disparities were similar, both charac-teristics remained significant on a number of measures, even though the effect of theother characteristics was taken into account. Although some of the causes of thesedisparities may be the same for low-income and minority enrollees, in other in-stances different factors may be responsible. Plans can use this information to iden-tify and target these contributing factors.

� Medicare and commercial plans. Our study also highlights potentially im-portant differences in the pattern of disparities for M+C and commercial plans. Al-though the overall pattern of disparities for M+C and commercial enrollees was sim-ilar, there were distinct differences on some measures, and disparities weregenerally smaller for enrollees in commercial plans.

Commercial enrollees were younger than M+C enrollees, so it is possible thatphysicians’ practice styles differed for these two groups. However, HEDIS mea-sures are designed to take clinical factors into account so that all eligible enrolleesshould have received appropriate care as specified by these measures. Commercialand M+C enrollees also may differ in terms of their SES and the resources availableto them, since commercial enrollees either are employed by or have a family mem-ber with an employer that provides relatively high levels of health insurance, orthey have the resources to buy such insurance themselves. These and related fac-tors could have contributed to some of the differences in results. More research isneeded to clarify these issues.

� Study limitations. Our study had several limitations. We used a geocoded (in-direct) measure of race/ethnicity. However, we found relatively little misclassifica-tion or bias when we used a geocoded versus a direct measure of race/ethnicity fromCMS data. Less than 11 percent of our sample was misclassified; 9.5 percent of thosewere blacks living outside predominantly black neighborhoods. Moreover, disparityresults were the same regardless of which race/ethnicity measure was used. Otherstudies suggest higher rates of misclassification among enrollees in commercialplans, particularly in regions where more blacks live outside of predominantly blackneighborhoods or tend to be more affluent.10 Nonetheless, these studies also showedreasonably accurate estimates of enrollees’ racial/ethnic composition and care dis-parities when the geocoded data were analyzed at the group level.

Because we compared “blacks” with “other,” rather than exclusively with

I n d i r e c t M e a s u r e s

H E A LT H A F F A I R S ~ V o l u m e 2 4 , N u m b e r 2 5 2 5

by guest on June 12, 2013Health Affairs by content.healthaffairs.orgDownloaded from

whites, our analyses may have underestimated racial disparities, since the minori-ties included in the “other” category may have received worse care than whites.We may have further underestimated disparities because of our limited power todetect disparities on several measures. Finally, since all study plans were from oneMCO, results may not be generalizable to other MCOs.

We recommend that plans use indirect methods such as geocoding to obtainrace/ethnicity and SES data on plan enrollees and use these data to begin identify-ing care disparities and targeting interventions to eliminate them. Plans should si-multaneously move toward developing direct data collection methods such as en-rollees’ self-reports.

This research was supported by Agency for Healthcare Research and Quality Contract no. 290-00-0012. The au-thors thank Nancy Krieger for her sage advice during the initial stages of this project and Nicole Lurie for con-structive comments on an earlier draft. They also thank Shelley Wiseman and Judy Bearer for helpful editing andassistance preparing the manuscript. Portions of this work were presented at the Society of General Internal Medi-cine Annual Meeting, Atlanta, May 2002, and the AcademyHealth Annual Meeting, Washington, June 2002.

NOTES1. A. Fremont and N. Lurie, “The Role of Racial and Ethnic Data Collection in Eliminating Disparities in

Health Care,” in Eliminating Health Disparities: Measurement and Data Needs, ed. M. Ver Ploeg and E. Perrin(Washington: National Academies Press, 2004), 202–231.

2. A.S. Bierman et al., “Addressing Racial and Ethnic Barriers to Effective Health Care: The Need for BetterData,” Health Affairs 21, no. 3 (2002): 91–102; and K. Fiscella et al., “Inequality in Quality: Addressing Socio-economic, Racial, and Ethnic Disparities in Health Care,” Journal of the American Medical Association 283, no. 19(2000): 2579–2584.

3. Agency for Healthcare Research and Quality, “Major Health Plans and Organizations Join AHRQ to Re-duce Racial and Ethnic Disparities in Health Care,” Press Release, 14 December 2004, www.ahrq.gov/news/press/pr2004/dispcolpr.htm (4 January 2005).

4. E.C. Schneider, A.M. Zaslavsky, and A.M. Epstein, “Racial Disparities in the Quality of Care for Enrolleesin Medicare Managed Care,” Journal of the American Medical Association 287, no. 10 (2002): 1288–1294; B.A.Virnig et al., “Racial Variation in Quality of Care among Medicare+Choice Enrollees,” Health Affairs 21, no. 6(2002): 224–230; and A.M. McBean et al., “Racial Variation in the Control of Diabetes among ElderlyMedicare Managed Care Beneficiaries,” Diabetes Care 26, no. 12 (2003): 3250–3256.

5. N. Krieger, D.R. Williams, and N.E. Moss, “Measuring Social Class in U.S. Public Health Research: Con-cepts, Methodologies, and Guidelines,” Annual Review of Public Health 18 (1997): 341–378.

6. D.S. Massey and N.A. Denton, “Hypersegregation in U.S. Metropolitan Areas: Black and Hispanic Segre-gation along Five Dimensions,” Demography 26, no. 3 (1989): 373–391.

7. D.R. Nerenz et al., “Eliminating Racial/Ethnic Disparities in Health Care: Can Health Plans Generate Re-ports?” Health Affairs 21, no. 3 (2002): 259–263; and A. Abrahamse, P. Morrison, and N. Bolton, “SurnameAnalysis for Estimating Local Concentration of Hispanic and Asians,” Population Research and Policy Review13, no. 4 (1994): 383–398.

8. N. Krieger et al., “On the Wrong Side of the Tracts? Evaluating the Accuracy of Geocoding in PublicHealth Research,” American Journal of Public Health 91, no. 7 (2001): 1114–1116.

9. A complete listing of process measures is available in Online Appendix 1, content.healthaffairs.org/cgi/content/full/hlthaff/24/2/516/DC1.

10. W. Chen, D.B. Petitti, and S. Enger, “Limitations and Potential Uses of Census-based Data on Ethnicity in aDiverse Community,” Annals of Epidemiology 14, no. 5 (2004): 339–345; and R. Kwok and B. Yankaskas, “TheUse of Census Data for Determining Race and Education as SES Indicators: A Validation Study,” Annals ofEpidemiology 11, no. 3 (2001): 171–177.

5 2 6 M a r c h / A p r i l 2 0 0 5

D a t a W a t c h

by guest on June 12, 2013Health Affairs by content.healthaffairs.orgDownloaded from