Impact of Health Disparities Collaboratives on Racial/Ethnic and Insurance Disparities in US Community Health Centers FREE

Several studies^1- 5 examining the quality of health care provided in the United States have documented significant problems with quality and disparities according to patient race and socioeconomic status. These issues are of particular concern for publicly supported community health centers (CHCs), which are responsible for caring for more than 15 million Americans, many of whom are members of groups that have previously been documented to receive care of lower quality.^2,5- 6 Furthermore, federally qualified CHCs are likely to have an increasing role in providing care for these populations.⁷

A few studies^8- 10 have examined the effectiveness of quality improvement programs aimed at improving overall quality to determine whether broad improvements in quality of care reduce disparities in processes of care, with conflicting results reported. One important national initiative designed to narrow disparities by improving overall quality is the Health Disparities Collaboratives (HDCs) sponsored by the Health Resources and Services Administration (HRSA). These collaboratives bring CHCs together to learn and disseminate quality improvement techniques developed by the Institute for Healthcare Improvement.^11- 13 Since 1998, approximately two-thirds of the CHCs (645 centers) have voluntarily participated in a collaborative focusing on improving care for chronic medical conditions using the chronic care model.¹⁴

Previously, we conducted a controlled national evaluation¹⁵ of the HRSA HDCs and found that they significantly improved the extent to which processes of care were followed for asthma and diabetes mellitus. To date, however, whether the HDCs reduced previously documented racial/ethnic and insurance quality gaps in CHCs remains unknown.¹⁶ In this study, we examine racial/ethnic and insurance differences in quality of care for asthma, diabetes mellitus, and hypertension before and after participation in the HDCs to provide a better understanding of whether these programs differentially narrowed disparities in care in addition to improving overall quality.

We examined medical record data from patients receiving care in a nationally representative sample of 64 CHCs participating in this evaluation of the HRSA collaboratives to improve the care of patients with asthma, diabetes, or cardiovascular disease. For the cardiovascular disease collaborative, we focused the assessment on the improvement of care for patients with hypertension. We included 44 CHCs that participated in 1 of 3 HDCs and 20 CHCs that were not HDC participants for any condition; because HDC participants may be different from CHCs that never participated in a collaborative, each intervention center was compared with a participating center in a collaborative for 1 of the other conditions (eg, diabetic HDC participants for analyses of hypertension HDC outcomes) and were classified as internal controls because they would not reasonably be considered the same as other controls that never participated in an HDC. Therefore, CHCs that had never participated in a collaborative were classified as external controls.¹⁵

From each CHC, we abstracted medical record data from random samples of patients with the condition of interest before and after participation in a collaborative. For each condition, we defined disparities in care as the difference in quality of care between white patients and members of racial/ethnic minority groups and between those with commercial or Medicare insurance and those with Medicaid insurance or no insurance. We compared changes in disparities across time by race/ethnicity and insurance status for intervention clinics vs internal and external controls using a difference-in-differences design. We hypothesized that participating in an HDC would result in a narrowing of disparities, leading us to observe smaller disparities in intervention CHCs compared with both types of control clinics.

The methods of CHC selection have been described in detail previously.¹⁵ Of 238 eligible CHCs identified by the HRSA as participating in the asthma I or II, diabetes II or III, or cardiovascular I collaboratives, 138 (58%) agreed to participate in this independent evaluation. From these, 48 intervention centers were selected for participation: 17 for diabetes, 16 for asthma, and 15 for hypertension on the basis of region, location (rural/urban/mixed), number of sites, and caseload. Each of these CHCs was also asked to serve as an “internal” control for one of the other conditions under study. The earliest collaborative started on January 1, 2000, and the latest collaborative started on August 1, 2001. Potential “external” control centers that had never participated in a collaborative were then matched with intervention centers using the same variables as noted previously, yielding 34 potential external control centers, of which 22 (65%) agreed to participate in the study. Subsequent to sampling, 4 intervention centers and 2 control centers dropped out of the evaluation because they were unable to prepare appropriate patient lists, leaving a final study sample of 44 intervention and 20 external control centers. The Committee on Human Studies of Harvard Medical School, Boston, Massachusetts, approved the study protocol.

We selected random samples of patients with one of the diagnoses of interest during the 1 year before the beginning of the applicable quality improvement collaborative and 1 year after completion of the collaborative. Each CHC used administrative data to generate lists of unique patients who had at least 1 visit to the center during the appropriate 12-month period and in the year before the study period and who had received care for asthma, diabetes, or hypertension. From each list, we randomly selected 40 patients for each condition after excluding patients with end-stage renal disease, malignant neoplasm, and human immunodeficiency virus infection. For diabetes and hypertension, we excluded patients younger than 18 years and pregnant women; for asthma, we excluded patients younger than 2 years.

One to 4 clinic staff members at each health center were trained to be medical record abstractors in condition-specific conference calls led by a clinical consultant. Abstractors then completed 2 sample medical record abstracts that were compared with a criterion gold standard completed by the consultant. Abstractors who scored 90% or more correct were certified to abstract medical records at their centers. Data abstracted from the medical records included sociodemographic information (such as age, sex, race-ethnicity, insurance status, and zip code), comorbid medical and psychiatric illnesses, and disease-specific quality indicators in the areas of preventive care and screening, disease monitoring and treatment, and intermediate outcomes of care for each condition.

We chose quality-of-care indicators based on national guidelines or standards, such as the Health Plan Employer Data and Information Set, the Diabetes Quality Improvement Project, and the National Asthma Education and Prevention Program (Table 1). We supplemented these with indicators identified by the Bureau of Primary Health Care HDCs and additional measures of quality we considered important in the care of asthma, diabetes, and hypertension. Because relatively few patients qualified for the smoking cessation advice measure, we created a composite measure that included assessment and advice. To qualify as having met this measure, patients had to be documented as nonsmokers or have received smoking cessation advice if documented as smokers.

We created composite scores for quality in the clinical domains of prevention and screening, disease monitoring and treatment, and outcomes for each condition by averaging the scores across all the indicators applicable to that patient, with higher mean scores indicating better quality of care.^15- 16 We define disparity score as the difference in mean quality scores between 2 racial/ethnic groups or between 2 groups with different levels of insurance coverage. We focus on disparities because they negatively affect ethnic minority groups and the uninsured; therefore, disparity score is the difference in quality between white patients and some other ethnic group or between a group with Medicare or private insurance coverage and uninsured individuals or Medicaid recipients. A positively valued disparity score indicates that a disparity is present (eg, a white-Hispanic diabetes disparity score of +6 indicates a 6% higher mean quality score for white patients compared with Hispanic patients). The change in disparity score is the disparity score for the postcollaborative period minus the disparity score for the precollaborative period. The difference-in-difference disparity is the change in disparity score for the intervention group minus the change in disparity score for a control group (internal or external control).

We compared the characteristics of the intervention centers and the internal and external control centers and their patient populations for each condition under study. Disparity scores were computed for each period. Using center-level quality scores as the outcome, we then compared mean changes in quality from baseline to follow-up for the intervention clinics and both types of control clinics using hierarchical linear regression models that controlled for patient characteristics and that accounted for the grouped error structure of the longitudinal data. Results are reported in terms of the change in disparity score for each treatment group. To determine whether disparities in quality narrowed or widened across time and whether changes in disparities across time differed by clinic type, we included a 3-way interaction term among period (preintervention vs postintervention), patient characteristic of interest (race/ethnicity or insurance status), and treatment group (intervention vs control) in each model. Patient control variables included age, sex, race/ethnicity, insurance status, and an adapted version of the Charlson Comorbidity Index.^15- 16 Results are reported in terms of the difference-in-difference disparity of the intervention group vs each control group.

Approximately half of the clinics were located in urban areas, and the clinics were well distributed throughout the country. No statistically significant differences were noted in clinic characteristics in intervention and control centers for any of the 3 conditions of interest (Table 2).

Overall, we studied 10 153 patients with 1 of the 3 target conditions in the experimental and control groups (3887 with asthma, 2904 with diabetes, and 3362 with hypertension). The mean number of patients per center in the precollaborative and postcollaborative periods was 41 (range, 16-84; median, 40) (Table 3). Approximately 60% of the patients were female in all 3 groups, and the mean age was approximately 46 years. Approximately 30% of the patients were covered by Medicaid insurance, and another 19% to 20% had no insurance. Except for insurance status and Charlson Comorbidity Index score for asthma (P < .01), no significant differences were noted between the intervention group and either of the control groups for any of the 3 conditions.

We observed differences in patterns of racial/ethnic disparities in care comparing intervention and control CHCs (Table 4). For all of the conditions, white patients received significantly higher-quality care than did black or Hispanic patients in control CHCs. For example, white patients in external control clinics received 67% of the recommended care for hypertension compared with 63% for Hispanic patients and 62% for black patients (P < .001). However, this pattern of racial/ethnic disparity existed only at baseline for asthma care in intervention CHCs.

For diabetes and hypertension, we observed significant disparities in care by insurance status at baseline for the intervention and control clinics (Table 4). In most instances, uninsured patients received lower-quality care than did those insured privately, by Medicare, or by other public funding. For example, uninsured individuals in collaborative clinics received 34% of the recommended care for diabetes compared with 39% for the insured groups (P < .001).

Comparing disparity scores in the preintervention and postintervention periods, we found that the intervention had a variable effect on narrowing disparities (Table 4). For asthma care, there was a significant reduction in the Hispanic-white disparity score in collaborative clinics, and this change was significantly greater than the disparity seen in the external controls (P = .04). However, there were greater reductions in Hispanic-white disparities for diabetes care in external control centers and for hypertension care in internal control clinics compared with changes in disparities in collaborative participating centers for these conditions across time (P < .004 for all).

When examining each domain of quality, we found that the observed improvement in Hispanic-white disparities in asthma care in collaborative clinics was because of a nonsignificantly greater reduction in these disparities in all 3 domains of care compared with external controls (Figure). For diabetes care, the greater reduction in Hispanic-white disparity in external controls was because of more significant reductions in quality gaps in the domains of monitoring and treatment and outcomes in these clinics (P ≤ .01 for both) compared with collaborative clinics. For hypertension care, the better performance of internal control clinics in reducing the Hispanic-white disparity compared with collaborative clinics was because of an increasing Hispanic-white disparity in disease monitoring and treatment in collaborative participants during the study (P = .007). We found no significant reductions in black-white disparities in overall care for any of the conditions.

Comparing disparity scores in the preintervention and postintervention periods, we found that the intervention had little effect on insurance disparities. There were no improvements in insurance disparities in overall care for any of the 3 conditions comparing collaborative clinics with controls (Table 4). For diabetes care, there was a significant increase in the Medicaid-private insurance disparity across time in collaborative clinics that led to a significantly greater increase in disparity compared with external controls (P = .01).

In this large controlled evaluation of a national program examining improvements in disparities in care for CHC patients with asthma, diabetes, and hypertension, we found that although overall quality of care improved across all involved CHCs, barriers according to race/ethnicity or insurance status did not change appreciably. This evaluation of the effectiveness of the HRSA HDCs in reducing disparities is particularly relevant today because of substantial growth in the number of CHCs during the past decade and the possibility that the current administration may further increase their number to increase access to care for the uninsured.

We found previously that participation in an HDC improved processes of care related to prevention and screening and disease monitoring and treatment.¹⁵ In this study, however, we found that these global improvements in quality had minimal effect on any baseline racial/ethnic and insurance disparities in care in individual CHCs. Specifically, although collaborative participation reduced disparities in quality of asthma care between Hispanics and non-Hispanic whites, participation was not effective in reducing racial/ethnic disparities for diabetes and hypertension and was associated with an increase in disparity of diabetes care between Medicaid and insured patients compared control CHCs. These findings may, in part, reflect the significantly smaller disparity present in HDC-participating CHCs compared with controls. Theoretically, participating CHCs had much less room for improvement compared with controls and, thus, were less likely to differentially improve. Note, however, that for many of the conditions, disparities differentially increased in HDC-participating CHCs during the intervention period compared with controls. Whether these differences in disparity trends across outcomes and diseases are related to the variable improvement across time for intervention and control centers for most individual measures is unknown.¹⁵

The HDCs were designed to improve the processes and outcomes of care for CHC patients with prevalent medical conditions and, in so doing, to decrease disparities in care. There are 2 potential mechanisms through which the HDCs could reduce disparities. The collaboratives themselves could lead to a narrowing of disparities in the target health centers akin to a rising tide lifting all boats. Alternatively, these collaboratives could reduce disparities nationally by raising performance for CHCs overall because CHCs care for disproportionate shares of minority and underserved populations. The present findings suggest that the HDCs did not reduce disparities in the targeted CHCs compared with a contemporaneous control group. This finding raises concerns that generalized quality improvement programs may be unsuccessful in reducing disparities, as many had believed, and that such programs may need to specifically target underserved populations if reducing intracenter disparities is an explicit goal.

These findings are supported by previous research examining the effects of similar broad quality improvement initiatives and have yielded mixed results, with some studies showing a reduction in disparities in processes of care with little effect on differences in clinical outcomes.^8,10 For example, in a longitudinal study examining the effectiveness of a Centers for Medicare and Medicaid Services national initiative to improve the intensity and outcomes of hemodialysis, Sehgal¹⁰ observed that the proportion of patients receiving adequate hemodialysis dosing increased 2-fold during a 7-year period and that the gap in intensity between white and black patients significantly decreased during that time. However, racial gaps in hemoglobin and albumin levels persisted.¹⁰

Similarly, in a study that examined managed care plans participating in Medicare, Trivedi et al⁸ found that reporting performance on quality-of-care measures from the Health Plan Employer Data and Information Set was associated with a reduction in disparities between white and black patients for 7 of 9 measures studied and an increase or no change in disparity in clinical outcomes related to blood glucose and cholesterol control. More recently, Moylan et al¹⁷ found that previously documented racial disparities in wait listing for liver transplantation were eliminated after introduction of the Model for End-Stage Liver Disease score. These previous studies, however, either used a historical cohort or examined general trends in quality and, thus, did not isolate specific effects from a quality improvement intervention. The present study results demonstrate a similar narrowing of black-white disparities across time at centers participating in the HDCs; however, these changes were not significantly different than those that occurred in control clinics. Furthermore, because CHCs provide care for many Hispanic patients,^5- 6,18 we were able to go beyond these previous studies by examining disparities in quality of care between Hispanic and non-Hispanic patients.¹⁹

In addition, because CHCs provide care for many uninsured and Medicaid patients, we were able to address limitations of previous assessments of the impact of broad quality improvement programs on disparities that have been limited to populations who were adequately insured.^8- 10 In a previous study¹⁶ examining quality of care for patients with chronic disease in CHCs, we found that Medicare patients received significantly higher quality of asthma and diabetes care compared with Medicaid and uninsured patients. In the present study, we found that the collaboratives had minimal effect on insurance disparities in the CHCs. We also found that although secular trends in control centers were associated with persistent insurance-based gaps in quality, the gap in diabetes quality between Medicaid and insured patients increased with collaborative participation.

The present assessment of the effectiveness of the HDCs in reducing disparities has limitations. First, we evaluated quality improvement collaboratives that were based on the Institute for Healthcare Improvement model, the most prevalent and reproducible type of quality improvement program, but there are many variations on this model and multiple other approaches to quality improvement that have been tried.^20- 21 Consequently, there is still much to know about quality improvement collaboratives, and a broader understanding of the tools and methods used for quality improvement and their potential effectiveness is missing.²² As a result, we cannot determine which aspects of collaborative participation were associated with increasing insurance disparities in diabetes care and improving racial disparities in asthma care. More research would likely be helpful regarding the operations and effectiveness of individual collaboratives and the broad range of organizational contextual factors that can impact the success of quality improvement collaboratives.^20- 23 Furthermore, we had only 44 HDC-participating CHCs in this study, limiting our ability to examine whether there are significant differences between collaborative participating centers for each condition to determine whether there are activities conducted in settings that may be more effective at disparity reduction than others.

Second, we could not perform a pure randomized trial of the intervention, instead relying on matching and statistical models to adjust for potential confounding variables. Third, although we assessed important markers of care quality that were the main focus of the collaboratives, some clinics might have improved in areas of care that we did not measure (eg, patient experiences with care). Fourth, we note that the cardiovascular collaboratives (the collaboratives involved in hypertension management) were focused on a broader set of goals than just hypertension, so there might have been improvement in other areas of care or for other cardiovascular disease populations that we did not study. Fifth, some of the improvements we observed might have resulted from improved documentation rather than from improved care. We¹⁵ previously noted, however, that measures that might be more sensitive to documentation effects (eg, smoking related) did not improve more than did measures that required an action (eg, glycohemoglobin assessment).

The HRSA HDCs are among the most important national health initiatives attempting to reduce disparities by targeting the quality of care for underserved populations. In this controlled study, these collaboratives significantly improved the quality of care in intervention centers, which affects disparities broadly because of the large share of underserved patients being cared for by CHCs, but had minimal effect on racial/ethnic disparities and insurance disparities in the health centers. These findings suggest that approaches that specifically target racial/ethnic and insurance disparities should be included as part of broad quality improvement initiatives and that future initiatives should assess disparity reduction as an outcome in addition to examining overall improvement in quality.

Accepted for Publication: September 30, 2009.

Correspondence: LeRoi S. Hicks, MD, MPH, Division of General Internal Medicine, Brigham and Women's Hospital, 1620 Tremont St, Boston, MA 02120 (lhicks1@partners.org).

Author Contributions: Dr Hicks had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Hicks, O’Malley, Lieu, McNeil, Guadagnoli, and Landon. Acquisition of data: Hicks, Keegan, McNeil, Guadagnoli, and Landon. Analysis and interpretation of data: Hicks, O’Malley, Keegan, and Landon. Drafting of the manuscript: Hicks, O’Malley, and Guadagnoli. Critical revision of the manuscript for important intellectual content: Hicks, O’Malley, Lieu, Keegan, McNeil, and Landon. Statistical analysis: O’Malley. Obtained funding: O’Malley, Lieu, McNeil, Guadagnoli, and Landon. Administrative, technical, and material support: Keegan, McNeil, and Guadagnoli. Study supervision: Hicks, Keegan, and Landon.

Financial Disclosure: Dr Hicks serves as a consultant for Health Management Corp and on the board of directors to Health Resources in Action.

Funding/Support: This project was supported by grant 1 U01 HS13653 from the Agency for Healthcare Research and Quality, with support from the Health Resources and Services Administration, and by grant 20030185 from The Commonwealth Fund.

Previous Presentation: An abstract of this project was accepted for oral presentation at the Society of General Internal Medicine 2007 annual meeting; April 25-28, 2007; Toronto, Ontario, Canada.

Additional Contributions: Yang Xu, MA, provided statistical programming; Rebecca Ouellette, MPA, assisted with project management; Mary Ly, MPH, Lynn Huynh, BA, and Adam Lessler, BA, provided research assistance, and Laura Peterson, RN, MS, assisted with developing chart abstraction instruments and training of abstractors.