Disparities in Health Care Are Driven by Where Minority Patients Seek Care:  Examination of the Hospital Quality Alliance Measures FREE

Differences in the quality of health care for minority patients are well documented. Two reports from the Institute of Medicine, Crossing the Quality Chasm: A New Health System for the 21st Century¹ and Unequal Treatment: Confronting Racial and Ethnic Disparities in Health Care,² concluded that minority patients experience a lower quality of care across a wide range of conditions even after adjusting for health insurance coverage and socioeconomic status. Some studies suggest that disparities in receipt of certain health care services are a result of bias or racism.^3- 4 An alternative hypothesis is that disparities can be explained, in part, by where care is received. Thus, the relevant research and policy question might be restated as follows: Are racial disparities a result primarily of discrimination, bias, or lack of cultural understanding by providers and practitioners so that minority patients receive worse quality care than nonminority patients in the same institutions, or are they a result of patterns of care seeking and service delivery resulting in minority patients receiving care from lower quality providers? In other words, are racial disparities the result of who you are or where you seek care?

Relatively few studies have addressed this question. Working with national data from the University HealthSystem Consortium (UHC), we used the Hospital Quality Alliance (HQA) 10 core indicators, which address recommended treatments for 3 clinical conditions: acute myocardial infarction (AMI), congestive heart failure (CHF), and community-acquired pneumonia (CAP). Seven other measures were added to the core set, and these 17 indicators have been endorsed by the National Quality Forum, a national standard-setting entity.

We hypothesized that the greatest disparities occur in processes of care that require patient counseling compared with some of the other measures (eg, aspirin at arrival for AMI). We posited that documentation for counseling measures would not be better for people of one race/ethnicity over another within a hospital. Thus, we hypothesized that a documentation effect would occur between hospitals rather than within hospitals. We examined whether disparities in all HQA measures were within hospitals, between hospitals, or both.

Recent studies suggest that at least some of the disparities found in large national databases can be attributed to site of care and geographic factors. Schneider et al⁵ found that clinical quality of care for Medicare beneficiaries in managed care plans was significantly lower for black patients than for white patients across a range of services. Both socioeconomic characteristics and differential enrollment of black patients in poor-performing health plans seemed to contribute to these disparities. Barnato et al⁶ found that black and white patients tend to go to different hospitals; thus, quality differences across hospitals could have a role in observed racial disparities in care. Bach et al⁷ noted that minority patients are treated by a group of physicians who may differ from physicians who treat primarily nonminority patients. Lucas et al⁸ found that black patients have higher operative mortality risks across a range of surgical procedures because of higher mortality at the hospitals in which they seek care. Other studies with teaching and nonteaching hospitals suggest that geographic variations in health care are responsible for some of the observed disparity nationally because minorities live disproportionately in parts of the country that have lower quality hospitals and primary care physicians.^9- 10

These studies have provided important contributions to our understanding of disparities. However, several studies examined only a single condition or procedure and none examined quality measures that focus on direct patient counseling using a national data set.

The HQA data are likely to be the foundation of hospital quality assessments for the foreseeable future and provide a unique opportunity to better understand the quality of care patients receive nationwide. Through the HQA, more than 3700 hospitals report quality measures to the Centers for Medicare & Medicaid Services. We analyzed data for a subset of 123 teaching hospitals reporting to the UHC, providing important information about the quality of care and disparities within and between this subset of teaching hospitals

This study used patient-level HQA quality measures obtained from the UHC, an alliance of academic medical centers and their affiliate hospitals in the United States. We present data, including patient characteristics, for 320 970 patients in 123 hospitals.

We examined hospital performance on 13 of the 17 measures from the third quarter of 2002 to the first quarter of 2005. Four measures (AMI–inpatient mortality; CAP–blood cultures before administration of first antibiotic therapy; CAP–antibiotic therapy started within 4 hours of arrival; and CAP–antibiotic selection) were excluded because there were not a sufficient number of cases at the individual hospital level to make meaningful comparisons. Table 1 gives measure specifications. The first 3 groups correspond to clinical measures that refer to processes of care for AMI, CHF, and CAP; the last group contains 4 counseling measures for each of the conditions.

All 123 hospitals were teaching hospitals. Sixty-two percent had at least 300 beds. They were distributed by region, with 10.8% in the Northeast, 19.2% in the Mid-Atlantic, 18.3% in the Southeast, 35.9% in the Midwest, and 15.8% in the West. Of the approximately 1100 teaching hospitals in the United States, 245 are members of the UHC. To be eligible for UHC membership, organizations must be one of the following: a nonprofit hospital or health system having common ownership with a college of medicine; a nonprofit or government hospital or health system in which most medical school chairs and hospital chiefs of service are the same; or a specialty hospital or health system that satisfies either of the first 2 criteria.

The UHC consists of 96 academic medical center members and an additional 149 affiliate hospital members (ie, related community hospitals). The American Association of Medical Colleges defines 117 hospitals as academic medical centers; of these, 8 are ineligible for UHC as for-profit entities. The UHC represents 96 of the 109 not-for-profit academic medical centers. Approximately 123 member hospitals use the UHC clinical database.

We excluded patients who were younger than 18 years, classified as “unknown” for race/ethnicity, and from hospitals that reported fewer than 50 total cases or fewer than 15 minority cases for any of the 3 conditions, to conduct reasonable logistic regression analysis. Thus, the number of hospitals differs for each of the measures (Table 2) because of individual patient eligibility criteria for denominator cases in addition to the hospital-level exclusion criteria. After exclusions, the study sample consisted of 320 970 patients.

The UHC merged the Centers for Medicare & Medicaid Services data with inpatient discharge data to obtain information about sociodemographic characteristics and comorbidities. We classified patients according to the following categories for analysis: age (>60 years), sex, race/ethnicity (minority or nonminority), and payer (Medicaid/indigent vs other). Medicaid/indigent payer included Medicaid, county medically indigent services, and no charge. Other payer included Medicare, Blue Cross/Blue Shield, managed care, self-pay, and third-party insurance. We defined minority patients as those identified as black or African American, Hispanic, Asian, and other. The number of minority patients in the specific racial/ethnic groups was too small to analyze separately at the individual hospital level. We examined a subset of the entire population for each of the measures to ensure that we had adequate numbers of patients at the individual hospital level to generate stable and reasonable estimates. Many of the hospitals had few patients in each of the subgroups. Preliminary analyses showed that the patients in each of these subgroups displayed similar trends. Therefore, we combined the minority subgroups into an overall minority category for analytical purposes.

The patient-level data were adjusted for severity of illness using the 3M APR-DRG (All Patient Refined–Diagnosis-Related Groups)¹¹ methods with 4 subclasses of severity of illness: minor, moderate, major, and extreme. Each patient was assigned a single all patient refined–diagnosis-related groups severity-of-illness subclass. In addition, the patient-level data were also adjusted for comorbid conditions.¹² The UHC adjusts patient-level data using both methods to account for inherent risk in chronic and comorbid conditions.

For each HQA measure, we tabulated the number of patients in the denominator affected (ie, those for whom the measure was relevant), then defined the performance rate as the percentage of eligible patients who received the indicated service. A disparity measure was derived as the nonminority rate minus the minority rate. Next, we divided the hospitals into the top quintile (top performers) and the bottom quintile (low performers) for each of the 13 measures, then determined the percentage of minorities in the top-performing and low-performing hospitals.

We used patient-level multivariate logistic regression models to estimate the likelihood of receiving the specified service. In all multivariate models, we adjusted standard errors for clustering of outcomes within hospitals. In the first model, we examined unadjusted bivariate relationships by including a dummy for race/ethnicity. In the second model, we controlled for the individual characteristics of age, sex, payer, severity of illness, and comorbid conditions. The 4 severity categories and the 30 comorbid conditions were entered into the model as dummies. In the third model, to address the possibility of confounding by site of care, we fit a set of models that included a dummy variable for each hospital.¹³ Differences in the adjusted probability of receiving the specified service among minorities compared with nonminorities were computed using the coefficient associated with minority status, setting all other covariates at their mean value. Some of the measures in models 2 and 3 (CAP–oxygenation assessment and CAP–antibiotic therapy started within 8 hours of arrival; AMI–aspirin at arrival, AMI–aspirin at discharge, and AMI–β-blocker at discharge; and CHF–left ventricular function assessment) did not converge; thus, we showed results not adjusted for clustering for these measures.

Of the 320 970 patients included in these analyses, 40% were of racial/ethnic minorities. However, the proportion of minority patients varied across measures, ranging from 25% for AMI–β-blocker at hospital discharge to 61% for CHF–smoking cessation counseling (Table 2).

Table 3 gives crude success rates for HQA measures for nonminority and minority patients. Overall, HQA performance rates were greater than 90% for AMI measures, with statistically significant (P<.05) but small absolute differences, the exception being angiotensin-converting enzyme (ACE) inhibitor use. The CHF measures showed small but statistically significant (P<.001) differences between minority and nonminority patients. Minority patients were significantly more likely to receive ACE inhibitors. The CAP measures showed significant differences (P<.001) for oxygenation assessment and receipt of antibiotic therapy within 8 hours of arrival. The counseling measures had lower overall performance rates (well below 90%). Minority patients were significantly less likely to receive these services (P<.001), and the overall absolute differences were consistently large.

Table 4 gives unadjusted and adjusted racial/ethnic disparities in the HQA measures. Adjusting for individual patient characteristics had little effect on success rates for most measures. The exception was the receipt of ACE inhibitors in patients with CHF, where the difference (with minorities receiving the service more often) was reduced from 4.5% to 1.9%. The unadjusted differences ranged from 1.54% for AMI–β-blocker at discharge (95% CI, 0.5-2.68) to 14.58% for CAP–smoking cessation counseling (95% CI, 2.74-26.1). The magnitudes of the racial/ethnic disparities decreased substantially when adjusting for site of care (model 3). The 10 percentage point difference (unadjusted, 9.61; 95% CI, 4.44-14.74; P<.001) in the likelihood of receiving discharge instructions among patients with CHF was eliminated once we controlled for site of care (adjusted, 0.49; 95% CI, −0.92 to 1.89; P = .50). Adding the hospital effect to the counseling measures explained a large proportion of the disparities, and although disparities in smoking cessation counseling were decreased for all 3 conditions, they remained statistically significant (P<.05).

Hospitals that perform less well on the HQA measures tend to serve a higher percentage of minority patients (Figure). For example, only 20% of patients were minorities in top-performing hospitals compared with almost 70% of patients in lower performing hospitals for AMI–smoking cessation. Still notable is the reversal of this trend for the 2 ACE inhibitor measures; the top-performing hospitals had a higher percentage of minority patients than did the lower performing hospitals, which mirrors the disparities analyses given in Table 3 and Table 4.

Our analysis confirms that quality of care for minority patients is often lower than for nonminority patients. Small but statistically significant disparities were noted in the bivariate analysis for 12 of the 13 HQA measures. In model 1, which adjusted for clustering of outcomes within hospitals, 8 measures showed significant racial/ethnic disparities. None of these disparities were accounted for by age, sex, payer, severity of illness, or comorbidities (with the exception of ACE inhibitor for CHF). After adjusting for site of care, the magnitudes of the 8 disparities were reduced and 3 disparities were entirely eliminated, suggesting that an underlying cause of disparities may be that minority patients are more likely to receive care in lower performing hospitals. We found that hospitals that were lower performers tended to serve a larger proportion of minority patients; another explanation may be that minority patients seek care in underresourced hospitals. A number of factors can characterize low-performing hospitals as underresourced including nurse staffing shortages, inadequate budgets, lack of technical support such as health information systems, and lack of capital.^14- 15

These data offer some clues as to the nature and potential causes of racial/ethnic disparities in quality of care. The reasons for disparities may vary with the services being provided. Some services may be of low quality because of where they are provided and others because of bias, racism, or difficulties with intercultural communication.

First, the magnitude of the racial/ethnic disparity varies among measures. Several of the HQA measures, such as aspirin and β-blocker use at arrival and on hospital discharge, show small disparities. These services are neither highly technical nor complex and require little interaction between the patient and the clinician. This may be why performance is greater than 90% for these measures.

Disparities are much greater for the counseling measures, which are communication sensitive. In contrast to aspirin and β-blocker use in AMI, the effectiveness of smoking cessation counseling might not be so straightforward. Counseling requires time, interaction with patients, and documentation. Previous research has suggested that the benefits of communication training for clinicians and patients can be enhanced by considering patient characteristics such as race/ethnicity and culture.¹⁶ Communication training may improve the rates of the counseling measures for minority patients. Shrank et al¹⁷ found that patient race/ethnicity was associated with modest quality differences in measures such as medication education and documentation. Further, Baker et al¹⁸ found that patient-provider communication and education in a heart failure learning collaborative resulted in participants substantially more likely to have knowledge about managing their cardiovascular health.

Second, our results suggest that individual patient characteristics did not explain the observed disparities, however, disparities in some measures were eliminated after adjusting for hospital effects. The one exception was the relatively higher proportion of minority patients who received ACE inhibitors for CHF, which may reflect the higher prevalence of comorbid conditions, such as kidney disease, in which ACE inhibitors have a role.¹⁹

About half of the racial/ethnic disparities for ACE inhibitors for AMI, oxygenation assessment for CAP, and antibiotic therapy within 8 hours of arrival for CAP seemed to reflect between-hospital rather than within-hospital factors. These results were particularly striking for the counseling measures, where well more than half of the disparity could be accounted for by between-hospital effects. In the case of hospital discharge instructions for CHF, the initial disparity was strong but entirely disappeared when controlling for hospital effect. It is possible that documentation of this measure in most hospitals is done by a nurse as part of standard orders; thus, whether this performance measure is checked off may be related to nurse-patient ratios, size and funding of quality improvement activities, and availability and use of electronic medical records. This finding may indicate that minority patients may be more likely to go to hospitals that are underresourced. Yet, these same hospitals may have providers who are hardworking and efficient, providing care with fewer resources to more disadvantaged patients.

The policy question that motivated this study was whether we should target resources to facilities that serve a high percentage of minority patients or direct them toward reducing potentially biased treatment patterns within facilities. Our findings suggest that the issue of where care is administered has a role in the quality of care received. Further research should try to identify which factors contribute to low performance in hospitals and how performance may be related to disparities in care.

Our study has several strengths. To our knowledge, this is the first study to use the HQA patient-level measures to assess racial/ethnic disparities within and between hospitals and to examine racial/ethnic disparities in counseling measures, which are communication sensitive. We were able to control for potential confounding between race/ethnicity and other socioeconomic factors, and hospital effects. By including hospital effects in our multivariable models, we controlled for both measurable and implicit hospital characteristics.

Our study has limitations. We lacked detailed information about patient knowledge, beliefs, and attitudes toward health care, their physicians, and the hospitals they used. Also, minority patients in teaching hospitals may be more likely to be on the teaching service while nonminority patients may be on a nonteaching service, which may be related to whether, for example, patients with CHF receive detailed discharge instructions from a physician who knows them and will follow up. We did not have this level of detail for analysis. We cannot exclude the possibility that inadequacies in the clinical and administrative records for the populations that we studied may have biased our results. However, controlling for site of care should at least partially address this potential bias. Our analysis focused only on UHC hospitals, a subset of all teaching hospitals, which may limit the generalizability of our findings.

Future research should address the reasons for disparities. Pay-for-performance is a policy tool to drive quality improvement; however, it is important for providers serving disadvantaged communities to be given incentives to improve quality metrics. Policy recommendations may need to focus on pay-for-improvement metrics for those underresourced providers caring for the most disadvantaged populations. Identifying and targeting the underlying causes of lower performance in hospitals and how these relate to health care disparities should inform programs and policies to reduce and potentially eliminate disparities in health care.

Correspondence: Romana Hasnain-Wynia, PhD, Health Research and Educational Trust, American Hospital Association, 1 N Franklin St, 30th Floor, Chicago, IL 60606 (rhasnain@aha.org).

Accepted for Publication: December 21, 2006.

Author Contributions:Study concept and design: Hasnain-Wynia, Baker, Nerenz, Feinglass, Beal, Behal, and Weissman. Acquisition of data: Hasnain-Wynia, Behal, and Weissman. Analysis and interpretation of data: Hasnain-Wynia, Baker, Nerenz, Beal, Landrum, and Weissman. Drafting of the manuscript: Hasnain-Wynia, Nerenz, and Weissman. Critical revision of the manuscript for important intellectual content: Hasnain-Wynia, Baker, Nerenz, Feinglass, Beal, Landrum, Behal, and Weissman. Statistical analysis: Baker, Feinglass, and Landrum. Obtained funding: Hasnain-Wynia, Baker, Beal, and Weissman. Administrative, technical, and material support: Hasnain-Wynia, Nerenz, and Behal. Study supervision: Hasnain-Wynia, Nerenz, Beal, and Weissman. Principal investigator: Hasnain-Wynia.

Financial Disclosure: None reported.

Funding/Support: This study was supported by grant 20040675 from The Commonwealth Fund.

Acknowledgment: We thank Raymond Kang, MA, of the Health Research and Educational Trust; Jodi Neikirk, MA, and Stacy Wang, MS, of the University HealthSystem Consortium for providing statistical and analytical support; and Debbie Pierce, The Health Research Trust, for assistance in project management.