Racial Disparities in Chronic Conditions Outcomes in Primary Care Settings: Between- Versus Within-Practice Differences

Abstract

Racial disparities in chronic condition management lead to adverse outcomes such as increased emergency department (ED) visits and hospitalizations among minority patients. These disparities may arise from differences within or between primary care practices, but limited research has explored these disparities. To examine racial/ethnic disparities in chronic condition outcomes in primary care and determine if they are due to between- versus within-practice differences. We analyzed 2018 Medicare claims data for beneficiaries visiting primary care practices during 2018 to 2019. We used logistic regression models to assess racial and ethnic disparities in outcomes and the contribution of between- versus within- practice differences to the disparities. 1033 primary care practices in Arizona, California, New Jersey, Pennsylvania, Texas, and Washington. Medicare beneficiaries aged 65+ with one of the chronic conditions attributed to primary care practices. All-cause ED visits, ambulatory care sensitive (ACS) ED visits, and all-cause hospitalizations among older adults with specific chronic conditions (asthma, chronic obstructive pulmonary disease, hypertension, congestive heart failure, cardiovascular disease, and diabetes). Black patients were more likely to experience ED visits (Predicted probability: .371, 95% Confidence Interval (CI): 0.362-0.380), and ACS ED visits (Predicted probability: .248, CI: 0.241-0.256) but less likely to experience hospitalizations (Predicted probability: .124, CI: 0.116-0.133), compared to White patients. Hispanic patients showed similar trends in ED visits (Predicted probability: .357, CI: 0.350-0.365) and ACS ED visits (Predicted probability: .233, CI: 0.227-0.239). Minority patients were clustered within a small number of practices (Black: 50% in 6% of practices, Hispanic: 50% in 8% of practices). Disparities in ED visits and hospitalizations were largely explained by within-practice differences. As disparities primarily originate from within-practice differences, addressing racial and ethnic disparities requires improving care quality across all practices rather than targeting those with high proportions of minorities.

-The healthcare system is marked by significant racial and ethnic disparities in care and outcomes of chronic conditions

-These disparities may be explained by within or between practices differences-

-We find that disparities in chronic condition ED visits and hospitalizations originate from within-practice differences

Introduction

Over half of American adults have at least one chronic condition, which is one of leading causes of death in the United States, and more recently has contributed to an increased risk of death associated with COVID-19.^{1 -3} In addition, care for chronic conditions is costly. Those with multiple chronic conditions can have up to 14 times higher annual health care expenditures than patients with no chronic conditions. ⁴ A significant contributor to increased costs is vastly higher rates of health care utilization among patients with chronic conditions. Particularly, individuals with chronic conditions are significantly more likely than those without chronic conditions to have an emergency department (ED) visit or inpatient hospitalization. ⁴ High-quality primary care helps patients to manage their conditions and can decrease ED visits and hospitalizations.^5,6 ED visits and hospitalizations, especially those that are preventable with adequate primary care, not only increase costs but also are themselves important chronic condition outcomes that can serve as a signal of the quality of primary care.

The United States’ healthcare system is marked by significant racial and ethnic disparities in care and outcomes, particularly in chronic conditions.^{7 -9} Non-Hispanic Black patients (Black patients) and Hispanic patients have higher incidences of chronic conditions such as diabetes, asthma, hypertension, and heart disease than non-Hispanic White patients (White patients).^{8,10 -12} Furthermore, management of these conditions is worse for Black and Hispanic patients.^{13 -15} Disparities in the management of chronic conditions may contribute to more frequent avoidable, ED visits and hospitalizations among minorities, which are pervasive in U.S.^7,16 For example, Black patients are more likely to visit the ED for diabetes compared with White patients, ¹⁷ and Black and Hispanic patients are more likely to have unscheduled readmissions for diabetes than White patients. ¹⁸ Black patients are nearly 4 times more likely to have an ED visit or hospitalization due to hypertension, and their rate of preventable hospitalizations for heart disease can be nearly 2 times than White patients.^19,20 The proportion of at least one asthma-related ED visit is higher among Black and Hispanic patients, and they were more likely to have multiple readmissions than White patients.^11,21 A higher percentage of Black and Hispanic patients visit emergency departments and hospitals due to chronic obstructive pulmonary disease (COPD) compared to White patients. ¹²

Understanding contributing to these disparities involves system-level, patient-level, and clinical-level factors. The National Academy of Medicine identifies the following 3 categories of factors that might drive racial and ethnic disparities. ²² System-level factors are community-level or health-system factors that may result in minorities receiving care at primary care practices that deliver lower quality of care overall.^22,23 Patient-level factors include differences in disease severity and socioeconomic status (SES) between White and minority patients or the communities in which they reside that may lead to different outcomes.^22,23 Clinical-level factors include ways in which providers might treat patients differently during the clinical encounter or during long-term disease management which might be driven by racism or differences in cultural communication styles.^22,23

Disparities in chronic condition outcomes can be influenced by each of these factors, which are classified as between- or within-practice differences. Between-practice differences in outcomes reflect system-level factors that result in minority patients being cared for in lower quality primary care practices. Minority patients tend to be clustered in practices that lack the necessary resources and infrastructure for high quality care delivery. For example, minority-serving practices tend to employ physicians less likely to be board certified and be familiar with evidence-based care. ²⁴ Within-practice differences are differences in outcomes between minority and White patients that are cared for within the same practice. Within-practice differences can be attributed to clinical- or patient-level factors. Previous studies have shown that clinical-level factors such as patient and provider racial discordance, ethnicity-based stereotyping, bias, or discrimination, and quality of communication may drive disparities. ²⁵ Patient-level factors that may contribute to within-practice differences are varying SES between minority and White patients. Minority patients tend to be lower SES which put them at risk for ED visits and hospitalization.^7,26

Most previous studies that examined between- versus within-practice sources of racial and ethnic health disparities focused on inpatient care.^{27 -31} Only 3 studies have focused on primary care practices. These studies found that there were significant disparities in process measures of quality of care for diabetes mellitus care, cancer screening, or in overall patient experience in primary care, and the vast majority of disparities were accounted for by within-practice differences.^23,32,33 However, none of these studies focused on clinical outcomes of chronic condition management within primary care practices such as ED visits or hospitalization.

In this study, we extend current literature by examining racial and ethnic disparities in ED visits and hospitalizations among patients with diverse chronic conditions and whether these disparities were explained by within- versus between-practice differences.

Methods

Study Setting and Sample

This study is a secondary analysis of data assembled as part of a parent study funded by National Institute of Minority Health and Disparities (R01 MD011514) which aimed to study care delivery in primary care practices employing nurse practitioners (NPs). To do this, we collected survey data from 1244 NPs at 1033 primary care practices across 6 states, Arizona, California, New Jersey, Pennsylvania, Texas, and Washington during 2018-2019. These states were chosen because they vary in their scope of practice regulations for NPs which are state laws that govern the extent to which NPs can practice independent of physician supervision and/or collaboration. ³⁴ The survey methods are described elsewhere. ³⁵ We summarize relevant aspects of the sampling design here. For this paper, we do not use the survey data itself but instead the administrative claim data (ie, Medicare) from 2018 for patients attributed to practices with NPs who completed the survey.

Using IQVIA’s OneKey database, we identified primary care NPs in the 6 states. The IQVIA OneKey database includes data on ambulatory-based providers in the U.S., including names of providers, practice names, locations, contact information, network affiliations, and National Provider Identifiers (NPIs). ³⁶ We identified primary care practices in which at least 50% of providers had specialties in family practice, general practice, geriatrics, internal medicine, preventative medicine, or pediatrics. This is a common approach to define a primary care practice. We selected only those practices that employed at least one NP. ³⁷

For each of the practices, we obtained Medicare data on each patient that had at least one visit to any of the study practices. For this analysis, we retained patients ≥65 with at least one of the following chronic conditions: asthma, COPD, hypertension, congestive heart failure (CHF), ischemic heart disease, and diabetes. These are the most common chronic conditions affecting Medicare beneficiaries. ³⁸ We used the Centers for Medicare and Medicaid Services’ (CMS) Chronic Condition Data Warehouse to identify these conditions. ³⁹

We then used a common approach to attribute patients to practices where they received care. ⁴⁰ We first attributed patients to a single provider (either NP or physician) by NPI. For each beneficiary-provider dyad, we calculated the proportion of primary care evaluation and management (E&M) paid amounts in 2018. We assigned patients to the provider who billed the highest proportion of the paid amounts if that paid amount represented at least 30% of all paid amounts. ⁴⁰ We randomly selected one provider in rare cases (<1%) of ties. We then attributed patients to their practices using the unique practice identifier in IQVIA OneKey.

Ethical approval was obtained from the Institutional Review Boards of the relevant Institutions. All survey participants provided implied informed consent by returning the survey, and their data were anonymized to ensure privacy and confidentiality.

Variables

We measured 3 outcomes at the patient level: (1) all-cause ED visits, (2) ambulatory care sensitive (ACS) ED visits, and (3) all-cause hospitalizations. The variables were coded as categorical (zero events or one/ more events). We used Part B carrier claims to identify ED visits as any visit for Healthcare Common Procedure Coding System codes 99281, 99282, 99283, 99284, and 99285. ⁴¹ ACS ED visits were unique ED visits with evidence of being avoidable or primary care treatable according to the “NYU ED Algorithm”.^42,43 For each ED visit, the algorithm assigns a probability, based on the primary ICD-10-CM diagnosis, that the visit is in 1 of 5 categories: 1- Non-Emergent; 2- Emergent, Primary Care Treatable; 3- Emergent, ED Care Needed, Preventable/Avoidable; 4- Emergent, ED Care Needed, Not Preventable/Avoidable; 5- All other. We counted an ED visit as ACS if it had a nonzero probability of belonging in any of the first 3 categories based on the principal ICD-10-CM diagnosis from the Medicare Part B claims data. We defined hospitalization as any record in the CMS Part A inpatient claims file with length of stay 1 day or more during the study period.

Patients’ race and ethnicity was measured as either (1) White, (2) Black, or (3) Hispanic using the Research Triangle Institute (RTI) Race Code available in the Medicare Beneficiary Summary File (MBSF). We excluded other racial categories due to small sample sizes and samples that were too heterogenous to meaningfully interpret the results. The RTI Race Code uses first and last name to enhance reported race and ethnicity categories.^44,45 This variable uses the beneficiary race code available in the Medicare data and then applies the RTI Race Code algorithm which utilizes first and last name of beneficiaries to identify additional beneficiaries that are likely to be from minority populations.

We included control for patient age and sex, which were measured using data from MBSF. We also included 15 practically useful but not overly inclusive comorbidities that we accessed via the CMS Chronic Condition Data Warehouse. These 15 specific conditions are listed in Table 1. Importantly, we did not include SES in the models. The National Academy of Medicine’s (formerly the Institutes of Medicine) definition defines disparities as differences in outcomes not explained by differences in health status and preferences. So, it is appropriate to control for health status, age, and sex. However, SES is likely a mechanism through which race and ethnicity function to produce disparities. Therefore, by controlling for SES, we would be controlling away part of the disparity.^22,46

Table 1.

Patient Characteristics and Outcomes Overall and by Race (N = 417 832).

	Overall, N = 417 832	White, N = 359 180 (85.96%)	Black, N = 21 645(5.18%)	Hispanic, N = 37 007(8.86%)	P-value
Patient characteristics
Age, Mean (SD)	76.01 (7.64)	76.25 (7.65)	74.42 (7.49)	74.62 (7.40)	<.001
Female (%)	56.89%	56.01%	64.36%	61.12%	<.001
Chronic conditions
Alzheimer’s disease and dementia	9.92%	9.60%	11.68%	11.96%	<.001
Acute myocardial infarction	1.11%	1.10%	1.16%	1.17%	.35
Arthritis	43.83%	43.98%	46.00%	41.06%	<.001
Asthma	8.02%	7.88%	9.97%	8.22%	<.001
Atrial fibrillation	13.47%	14.46%	6.90%	7.67%	<.001
Cancer a	12.12%	12.37%	14.11%	8.51%	<.001
Congestive heart failure	18.53%	18.21%	22.89%	19.11%	<.001
Chronic kidney disease	33.46%	31.64%	45.98%	43.80%	<.001
Chronic obstructive pulmonary disease	15.25%	15.67%	15.56%	11.00%	<.001
Depression	19.30%	19.58%	16.13%	18.45%	<.001
Diabetes	36.75%	33.66%	51.06%	58.32%	<.001
Hyperlipidemia	67.83%	68.02%	63.22%	68.68%	<.001
Hypertension	86.68%	86.13%	92.54%	88.57%
Ischemic heart disease	38.68%	39.23%	34.60%	35.75%	<.001
Osteoporosis	9.10%	9.22%	5.76%	9.96%	<.001
Stroke/TIA	5.04%	5.01%	6.15%	4.70%	<.001
Patient outcomes
ED visits (%)	29.58%	28.50%	37.98%	35.18%	<.001
ACS ED visits (%)	19.38%	18.23%	28.20%	25.40%	<.001
Hospitalizations (%)	19.11%	19.13%	19.85%	18.41%	<.001

Note. SD = standard deviation; ED = emergency department; ACS = ambulatory care sensitive.

^aIncludes any breast, colorectal, endometrial, lung, and prostate cancer from the 27 chronic condition data warehouse.

Data Analysis

We first computed descriptive statistics for all patient characteristics and outcomes and compared these variables across each racial category using chi-square tests or t-tests. We then examined the extent to which minority patients cluster in specific primary care practices. We created histograms that graph the number of Black and Hispanic patients within each primary care practice in descending order. We then counted the proportion of practices that accounted for 50% of Black and Hispanic patients.

We then estimated 2 logistic regression models to assess the relative contribution of within- versus between-practice differences to overall racial and ethnic disparities in the outcomes. In Model 1, we estimated a logistic model that includes patient race, age, sex, and chronic conditions. From these models, we estimated predicted probabilities for each race and ethnicity category.

In Model 2, we incorporated a separate intercept for each practice in our sample (ie, practice fixed effects) and produced updated predicted probabilities. The differences in the predicted probability on each race category in Model 2 represents the within-practice differences that remain after accounting for practice fixed effects which control for all between-practice differences. The within-practice differences can be thought of as the difference in outcomes of minority patients compared to White patients within the same practice. In other words, within-practice difference is the disparity that remains after we account for fixed differences in quality between practices, accounting for the fact that minorities may cluster in lower performing practices. Previous studies have used this fixed-effects approach to calculate within- versus between-practice differences.^23,32

We then calculated the proportion of disparities explained by within- and between-practice differences. To do this, we used post-estimation commands to calculate predicted probabilities for each outcome by racial and ethnic category. We then calculated the difference in the mean predicted probabilities between Black patients and White patients as well as Hispanic patients and White patients for Models 1 and 2. The proportion of disparities accounted for by between differences was the percent change in the mean differences in the proportions between Model 1 and Model 2. Within-practice differences was calculated as 1 minus the proportion explained by between-practice differences. If the mean proportional difference did not change or increased magnitude (away from zero) in Model 2, we considered 100% of the disparity to be explained by within-practice differences.

Results

Our study involved 417 832 patients from 1033 primary care practices. Of these patients, 85.96% (n = 359 180) were White while 5.18% (n = 21 645) were Black and 8.86% (n = 37, 007) were Hispanic (Table 2). We identified significant differences between White, Black, and Hispanic patients. Black (74.42 years) and Hispanic patients (74.62 years) were younger than White patients (76.25 years). Black (64.36%) and Hispanic (61.12%) patients were also more likely than White (56.01%) patients to be female.

Table 2.

Results of Regression Models for the Contribution of Within- Versus Between-Practice Differences to Racial and Ethnic Disparities in the Outcomes.

		Model 1: Overall disparities			Model 2: Within practice effects
		Predicted probability	CI		Predicted probability	CI
			Lower limit	Upper limit		Lower limit	Upper limit
ED	White	.269	0.267	0.270	.283	0.279	0.288
	Black	.369*	0.362	0.376	.371*	0.362	0.380
	Hispanic	.352*	0.347	0.357	.357*	0.350	0.365
ACS ED	White	.167	0.165	0.168	.166	0.163	0.169
	Black	.262*	0.256	0.268	.248*	0.241	0.256
	Hispanic	.244*	0.240	0.249	.233*	0.227	0.239
HOSP	White	.155	0.154	0.157	.132	0.125	0.139
	Black	.149*	0.144	0.154	.124*	0.116	0.133
	Hispanic	.155	0.151	0.159	.135	0.127	0.144

Note. OR = odds ratio; CI = confidence interval; ED = emergency department visits; ACS ED = ambulatory care sensitive emergency department visits; HOSP = hospitalizations; Model 1 includes race and patient characteristics (eg, age, sex, chronic conditions), Model 2 includes race, patient characteristics, and practice fixed effects. White race is referent in each model.

^*Differences significant at P < .05 compared to White.

Black and Hispanic patients had significantly higher levels of comorbidities compared to White patients with some notable exceptions. Hispanic were less likely than White patients to be diagnosed with arthritis (41.06% vs 43.98%), atrial fibrillation (7.67% vs 14.46%), cancer (8.51% vs 12.37%), chronic obstructive pulmonary disorder [COPD] (11.00% vs 15.67%), depression (18.45% vs 19.58%), ischemic heart disease (35.75% vs 39.23%), and stroke (4.70% vs 5.01%). Likewise, Black patients were less likely than White patients to be diagnosed with atrial fibrillation (6.90% vs 14.46%), COPD (15.56% vs 15.67%), and hyperlipidemia (63.22% vs 68.02%).

Black (37.98%) and Hispanic (35.18%) patients were more likely to experience an ED visit compared to White patients (28.50%) as well as ACS ED visits (Black: 28.20%, Hispanic: 25.40%, White:18.23%). Black (19.85%) patients had higher rates of hospitalization compared to White (19.13%) patients, but Hispanic patients (18.41%) had lower rates of hospitalizations.

Among minority patients, we identified significant levels of clustering into a small set of primary care practices. The histograms (Figure 1) show a significant right-tailed distribution for both Black and Hispanic populations. Among Black patients, 50% were clustered in 6% of practices while 50% of Hispanic patients were clustered in 8% of practices.

Figure 1.

Counts of minority patients per primary care practice. Panel A shows the number of Black patients per practice. Panel B shows the number of Hispanic patients per practice. (A) Black patients. (B) Hispanic patients.

In Model 1, we found significant overall racial and ethnic disparities in patient outcomes. Black patients were significantly more likely to experience ED visits (Predicted probability: .369, 95% Confidence Interval (CI): 0.362-0.376), and ACS ED visits (Predicted probability: .262, CI: 0.256-0.268) than White patients. They were, however, less likely to experience hospitalizations (Predicted probability: .149, CI: 0.144-0.154). Hispanic patients were significantly more likely to experience ED visits (Predicted probability: .352, CI: 0.347-0.357), ACS ED visits (Predicted probability: .244, CI: 0.240-0.249), and no difference in hospitalizations.

After controlling for between-practice differences with practice fixed effects (Model 2), Black patients were more likely to experience ED visits (Predicted probability: .371, CI: 0.362-0.380) and ACS ED visits (Predicted probability: .248, CI: 0.241-0.256) but less likely to experience hospitalizations (Predicted probability: .124, CI: 0.116-0.133). Hispanic patients were more likely to experience ED visits (Predicted probability: .357, CI: 0.350-0.365), and ACS ED visits (Predicted probability: .233, CI: 0.227-0.239) and no difference in hospitalizations.

Based on changes in the mean differences between the racial groups. we found that the vast majority (though not all) of racial and ethnic disparities can be explained by within-practice differences (Table 3). For ED visits (both overall and ACS-specific), approximately 85% to 100% of all disparities can be explained by within-practice differences meaning that Black and Hispanic patients have significantly higher ED visit rates compared to White patients within the same practice with one exception that Black patients having lower hospitalization rates than Black patients. There were no disparities observed among Hispanic patients in hospitalizations.

Table 3.

Proportion of Disparities Explained by Between- Versus Within- Practice Differences.

		Marginal difference in means		Proportion of disparities explained by between- versus within- practice differences
		Model 1	Model 2	% Between	% Within
ED	Black	0.100	0.088	12.1%	87.9%
	Hispanic	0.083	0.074	10.8%	89.2%
ACS ED	Black	0.095	0.082	13.5%	86.5%
	Hispanic	0.078	0.067	14.1%	85.9%
HOSP	Black	−0.006	−0.008	0.00%	100.00%
	Hispanic	N/A	N/A	N/A	N/A

Note. ED = emergency department visits; ACS ED = ambulatory care sensitive emergency department visits; HOSP = hospitalizations; Model 1 includes race and patient characteristics (eg, age, sex, chronic conditions), Model 2 includes race, patient characteristics, and practice fixed effects. White race is referent in each model.

Discussion

To our knowledge, this is the first study to examine the extent to which disparities in chronic condition utilization outcomes in primary care were attributable to within- versus between-practices differences. Our study provides important contributions to the health care disparities literature overall. We found significant racial differences in ED visits and hospitalizations among Black and Hispanic patients compared to White patients. Black patients had consistently higher rates of ED visits, ACS ED visits, and ACS hospitalizations but lower rates of hospitalizations compared to White patients, whereas Hispanic patients had higher rates of ED visits, ACS ED visits, ACS hospitalizations, and not significantly different hospitalization rates than White patients. These findings confirm that racial disparities in chronic condition-related, potentially avoidable ED visits and hospitalizations are widespread in the United States,^7,16 and that patterns of these disparities in ED visits and hospitalizations for chronic conditions are different by racial groups.^{11,12,17 -21}

We also found that minority patients were notably clustered within a small number of practices. Our findings showed minorities were slightly more clustered compared to those from previous studies that examined clustering both among Medicare beneficiaries and commercially-insured patients.^24,32 Patient clustering suggests that we might be able to target interventions to improve care delivery to a small number of practices with high proportions of minorities to effectively address disparities. This would be especially true if disparities were driven by between-practice differences. However, we found that only a relatively small proportion of disparities can be explained by minorities clustering in low performing practices (between-practice differences). Instead, racial and ethnic disparities were primarily explained by within-practice differences. Between-practice differences explained at most 14% of observed disparities, while within-practice effects explained the remaining proportion of disparities. This means that minority patients still had worse outcomes within the same practices than White patients, and that efforts to improve minorities’ health care experiences across all practices are likely necessary.

These within-practice differences may be explained by certain patient factors such as age and disease severity, which we partly controlled for.^26,47 There are likely other patient-level factors that we did not control for that may be driving these disparities including social capital and patient preferences that may especially drive the use of emergency departments. Additionally, provider behavior within practices, potentially influenced by race or ethnicity, could contribute to these disparities. This may take the form of overt racism or unconscious bias. There is suggestive evidence that providers treat patients of different races differently within the clinical setting. ⁴⁸

Our study is especially important as it focuses on ambulatory care outcomes (ie, ED visits and hospitalizations) which are less under provider control compared to other settings (ie, acute care) or other variables (ie, processes of care and patient experience). For example, one previous study of inpatient care attributes disparities to between-hospital differences meaning that minorities are more likely to get care from lower performing hospitals. ²⁸ However, once within the hospital, minorities and White patients may receive the same quality of care. Because hospitals are more closed systems than ambulatory settings and likely have more clinical “control” over patients, the impact of external factors driving health care quality may be less determinative compared to more open systems like ambulatory care. ⁴⁹ Studies that have focused primarily on the ambulatory care found that the magnitude of the between-within effects varied by the type of measure that was used in the study.^23,32,33 One study examining the impact of between-within practice impact on diabetes care measures (ie, hemoglobin A1C control, low-density lipoprotein control) found that within-physician differences accounted for approximately 66% to 75% of all variation in disparities which was smaller than the within-practice effects we found on outcome measures. ³³

Our study has several limitations. First, we focused only on practices that employ NPs, which might not represent all primary care practices. In 2016, approximately 26% of all primary care practices employed at least one NP. ⁵⁰ Second, our analysis was limited to Medicare Fee-for-Service patients, which may not be generalizable to Medicare Advantage population, or other publicly (eg, Medicaid) and privately insured (eg, employer-sponsored) patients. In addition, our analyses could not control for provider, practice, and healthcare system characteristics. To include healthcare-related characteristics might have helped us to address the contextual circumstances of patient outcomes in care. Another limitation is our use of a provider-centric approach to identifying ED visits (Part B carrier claims) which can potentially undercount ED visits in a small number of cases in which a facility may bill the visit but not the clinician (ie, suture or wound packing removal). Conversely, the facility-based approach to counting ED visits likely counts visits considered “emergent” but occurring outside of an ED. We decided to use the more conservative provider-based approach. ⁴¹ Lastly, although we originally planned on analyzing ACS hospitalizations as an additional outcome variable, we excluded these results from the final analysis due to instability across various estimation models. We were not confident in the accuracy of the results.

Conclusions and Relevance

This study examined the influences of between- versus within-practice differences on disparities in ED visits and hospitalizations for patients with chronic conditions in primary care settings. Our results suggest that racial and ethnic disparities in ED visits and hospitalizations are largely explained by within-practice differences. To address these disparities, efforts to improve care of minorities across all practices are likely necessary, rather than targeting specific practices with high proportions of minorities.