Mark W. Friedberg, Kathryn L. Coltin, Dana Gelb Safran, Marguerite Dresser, Eric C. Schneider. Medical Home Capabilities of Primary Care Practices That Serve Sociodemographically Vulnerable Neighborhoods. Arch Intern Med. 2010;170(11):938–944. doi:10.1001/archinternmed.2010.110
Under current medical home proposals, primary care practices using specific structural capabilities will receive enhanced payments. Some practices disproportionately serve sociodemographically vulnerable neighborhoods. If these practices lack medical home capabilities, their ineligibility for enhanced payments could worsen disparities in care.
Via survey, 308 Massachusetts primary care practices reported their use of 13 structural capabilities commonly included in medical home proposals. Using geocoded US Census data, we constructed racial/ethnic minority and economic disadvantage indices to describe the neighborhood served by each practice. We compared the structural capabilities of “disproportionate-share” practices (those in the most sociodemographically vulnerable quintile on each index) and others.
Racial/ethnic disproportionate-share practices were more likely than others to have staff assisting patient self-management (69% vs 55%; P = .003), on-site language interpreters (54% vs 26%; P < .001), multilingual clinicians (80% vs 51%; P < .001), and multifunctional electronic health records (48% vs 29%; P = .01). Similarly, economic disproportionate-share practices were more likely than others to have physician awareness of patient experience ratings (73% vs 65%; P = .03), on-site language interpreters (56% vs 25%; P < .001), multilingual clinicians (78% vs 51%; P < .001), and multifunctional electronic health records (40% vs 31%; P = .03). Disproportionate-share practices were larger than others. After adjustment for practice size, only language capabilities continued to have statistically significant relationships with disproportionate-share status.
Contrary to expectations, primary care practices serving sociodemographically vulnerable neighborhoods were more likely than other practices to have structural capabilities commonly included in medical home proposals. Payments tied to these capabilities may aid practices serving vulnerable populations.
In the United States, racial minority and economically disadvantaged groups experience poorer health outcomes than others.1 Many factors contribute to these disparities. The characteristics of primary care practice sites may play an important role.2 Relatively few physicians serve large shares of minority patients, and minority-serving primary care practice sites have reported lower levels of payment and more problems delivering high-quality health care than other practice sites.3- 5 The current crisis in primary care further threatens access and clinical quality for these vulnerable patient populations.6
The “patient-centered medical home” has emerged as a prominent response to challenges facing primary care, promoting a new ambulatory practice model for high-quality primary care that is comprehensive, coordinated, culturally and linguistically appropriate, and accessible to all patients.7,8 In addition to raising health care quality, lowering costs, and revitalizing the primary care workforce, the medical home may also reduce sociodemographic disparities in care.9- 12 Medical home pilot projects in at least 13 states and a Medicare medical home demonstration are planned or under way.13,14 These pilot projects designate practice sites as medical homes based on site-level structural capabilities such as reminder systems and electronic health records (EHRs).15- 17 Practice sites designated as medical homes may qualify for enhanced payments, generally in the form of a monthly per-patient fee.18
Whether primary care practices that disproportionately serve sociodemographically vulnerable neighborhoods will be able to qualify for medical home payments is unclear. For these “disproportionate-share” practice sites, the cost of adopting medical home capabilities may constitute an especially high barrier to medical home designation, particularly if such practices currently lack qualifying structural capabilities.4,19 If disproportionate-share practices are less likely than others to qualify for enhanced medical home payments, then the distributive effect of these payments might be to exacerbate, rather than ameliorate, disparities in access to medical home capabilities. Our aim in this study was to describe the distribution of structural capabilities among primary care practices, measuring relative access to these capabilities by members of sociodemographically vulnerable groups and foreshadowing the potential distribution of enhanced medical home payments.
We previously developed a survey to assess the use of 13 structural capabilities of primary care practice sites that were believed to enable high-quality health care and that could be reported by practicing physicians (see the eAppendixes for survey items contributing data to the current analysis).20 These capabilities fell into the following 4 domains: (1) patient assistance and reminders (ie, assistance of patient self-management, system for contacting patients for preventive services, and paper-based physician reminder systems); (2) culture of quality (ie, physician awareness of performance on quality and patient experience, new initiatives on quality and patient experience, frequent meetings on quality performance, and the presence of a practice leader for quality improvement); (3) enhanced access (ie, language interpreters, multilingual clinicians, and regular appointment hours on weekends); and (4) EHRs. We defined frequently used, multifunctional EHRs as those that were used usually or always during clinical care and featured laboratory and radiology results, notes from consultants, problem and medication lists, and electronic reminders.
Practice sites were defined as at least 2 physicians plus other clinical and nonclinical staff providing primary care at a single address. Data on solo practitioners were unavailable. We classified sites as community health centers (CHCs) if they were so identified by the Massachusetts League of Community Health Centers.21 We designated sites affiliated with any of the 9 large physician networks in the state as network affiliated and others as nonaffiliated. We classified practice sites as within metropolitan Boston if their street addresses were located in the Boston-Cambridge-Quincy metropolitan statistical area as defined by the US Census Bureau.22 All other practice sites were classified as outside metropolitan Boston.
Our survey sample consisted of all 412 Massachusetts practice sites with a physician eligible for the survey (ie, who provided substantial primary care services at the target practice but not elsewhere).20 We surveyed 1 eligible physician selected at random from each of the 412 practices. We conducted the survey by mail during May to October 2007 and received survey responses for 308 sites (75% response rate). This study was approved by the Human Subjects Committee of the Harvard School of Public Health.
To describe the sociodemographic case-mix of each practice site, we used a geocoding approach to construct 2 indices, one representing minority race/ethnicity and the other representing economic disadvantage. These indices were empirically derived using factor analysis, a technique that identifies mutually correlated sets of variables. We applied this technique twice, first at the neighborhood level (to eliminate essentially duplicative variables) and then at the practice site level (to combine correlated sets of variables).
To describe patients' neighborhoods, we obtained the 9-digit zip codes of residence for 1 009 932 patients who were (1) aged 18 to 64 years, (2) continuously enrolled for 12 months in a managed health care product offered by 1 of the 5 health plans participating in Massachusetts Health Quality Partners' statewide performance reporting initiatives,23 and (3) recorded by the enrolling health plan as having a primary care physician in 1 of the study practice sites during 2005. Residential data on patients without commercial insurance were not available to us. We matched the patients' 9-digit zip codes to year 2000 Census data at the block group level, obtaining a successful match for 87%. The remaining patients were matched using their 5-digit zip codes. In supplementary analyses, exclusion of patients matched on the 5-digit rather than the 9-digit zip code did not substantially alter our results. This geocoding approach, previously described as “geographic retrofitting” of practices to the residential addresses of their patients, has been used to improve the validity of practice-specific community estimates (relative to community estimates based solely on practice site locations).24
We identified 17 census variables representing sociodemographic characteristics previously found to be associated with quality of health care.1,25- 30 These variables included the proportions of residents within each census block group in categories of race and ethnicity (non-Hispanic white, black, Asian, or Hispanic), income (<100% and <200% of the federal poverty line), receipt of public assistance, spoken language (English, Spanish without English, or other language without English), education (less than a high school education), employment (unemployment or labor force participation), occupation (professional or other), and citizenship and country of birth (United States or other).
To reduce colinearity among neighborhood-level sociodemographic variables (eg, Spanish language and Hispanic ethnicity), we included all Massachusetts census block groups in a principal components analysis of the census variables. This analysis identified 4 factors with eigenvalues of greater than 1. We chose a single variable to represent each of the following 4 factors: black race, Hispanic ethnicity, less than a high school education, and non-US citizenship. For completeness, we also included 3 variables that did not load on any of the 4 factors: receipt of public assistance, income of less than 200% of the federal poverty line, and unemployment.
We then used these geocoded data to calculate the prevalence of the 7 census variables in the geographically retrofitted community served by each practice site. For example, if a site had 100 patients—50 living in a census block group with 40% black residents and 50 living in a block group with 20% black residents—the estimated proportion of black patients in the practice's geographically retrofitted community would be 30%. The median number of patients included in these sociodemographic case-mix calculations was 1586 per practice site (range, 127-19 949).
To combine correlated sets of sociodemographic variables at the practice site level, we performed a second principal components analysis of the 7 census variables using the geographically retrofitted communities. This analysis yielded 2 factors with eigenvalues of greater than 1, which we labeled minority race/ethnicity case-mix and economic disadvantage case-mix. Three of the census variables (black race, Hispanic ethnicity, and non-US citizenship) loaded on the race/ethnicity factor, and the remaining 4 census variables (less than a high school education, receipt of public assistance, income <200% of the federal poverty line, and unemployment) loaded on the economic factor. We then created index scores for each factor by calculating the mean of its constituent census variables, standardizing the mean and variance of each variable to ensure equal weighting.
To identify disproportionate-share practice sites, we ranked the sites on each sociodemographic case-mix index (race/ethnicity and economic). Because the distribution of the case-mix indices was skewed, with little variation among practices in the bottom 80% of each distribution, we classified practice sites with index scores above the 80th percentile on each index as disproportionate share. Supplementary analyses using alternative cutoffs at the 75th and 90th percentiles yielded results substantively similar to our main findings (results not shown).
Data from the survey of practice site structural capabilities were classified as described elsewhere.20 For survey items assessing the presence of a structural capability, we collapsed responses of no and don't know into a single no category. For each survey item with more than 2 ordinal response categories, we created a dichotomous variable using the median of the response distribution. For ease of presentation, we dichotomized practice site size at 4 or fewer or at more than 4 physicians (the median).
We assessed the effect of survey nonresponse using Wilcoxon rank sum and Fisher exact tests to compare responding and nonresponding practices on median size, percentage with network affiliation, and median prevalence of patients with each of the 7 geocoded sociodemographic characteristics. Responding and nonresponding practices were similar with respect to median size, percentage with network affiliation, and estimated prevalence of each sociodemographic case-mix variable (eTable).
To assess the bivariate relationships between practice structural capabilities and sociodemographic case-mix, we constructed logistic models, each with the presence of a single structural capability as the dependent variable and minority race/ethnicity disproportionate-share status as the independent variable. Because network-affiliated practices were clustered within networks, statistical tests that assume independence of observations (eg, χ2 tests of categorical data) were unavailable to us. The main analyses were therefore based on regression models fit using generalized estimating equations with exchangeable working correlation structures and empirical standard errors to adjust estimated variances for clustering of observations within networks.31,32
We then repeated these models using economic disadvantage disproportionate-share status as the independent variable. To address potential confounding by other practice site characteristics (size, network affiliation, teaching status, multispecialty status, and practice geographic location), we repeated the regression models, controlling for 1 potential confounder at a time. Put another way, these models account for the possibility that if observable practice characteristics mediate a relationship between disproportionate-share status and structural capabilities in Massachusetts, these mediating characteristics may have different distributions in other geographic settings.
Because CHCs may participate in collaboratives that support investment in specific capabilities, we repeated our analyses after excluding CHCs from the practice site sample.33 Results of these supplementary analyses were substantively similar to those of the main analysis. All statistical analyses were performed with SAS statistical software, version 9.1.3 (SAS Institute Inc, Cary, North Carolina). P < .05 was considered statistically significant for all comparisons.
Practice sites serving disproportionate shares of patients residing in neighborhoods with high racial/ethnic minority prevalence were more likely than others to have more than 4 physicians (66% vs 38%; P < .001), to be teaching practices (72% vs 40%; P < .001), and to be located within metropolitan Boston (84% vs 56%; P = .03) (Table 1). Practices serving disproportionate shares of patients residing in economically disadvantaged neighborhoods were also more likely than others to have more than 4 physicians (63% vs 38%; P < .001) but were less likely to be located in metropolitan Boston (38% vs 67%; P < .001).
The median estimated prevalence of each census variable was higher among practice sites identified as disproportionate share on the corresponding case-mix index than among other sites. Disproportionate-share practice sites were also significantly more likely than other sites to be CHCs. Approximately 50% of the practices identified as disproportionate share on minority race/ethnicity were also identified as disproportionate share on economic disadvantage (and vice versa) (Table 2).
In bivariate analyses, minority disproportionate-share practice sites were significantly more likely than others to have the following 4 structural capabilities: staff to assist patient self-management (69% vs 55%; P = .003), on-site language interpreters (54% vs 26%; P < .001), multilingual clinicians (80% vs 51%; P < .001), and frequently used, multifunctional EHRs (48% vs 29%; P = .01) (Table 3). However, physicians in these practices were less likely than others to report awareness of clinical quality performance (80% vs 90%; P = .01). Economic disproportionate-share practices were significantly more likely than others to have 4 structural capabilities: physician awareness of patient experience ratings (73% vs 65%; P = .03), on-site language interpreters (56% vs 25%; P < .001), multilingual clinicians (78% vs 51%; P < .001), and frequently used, multifunctional EHRs (40% vs 31%; P = .03).
After adjustment for practice size, minority disproportionate-share practice sites were more likely than others to have only 2 structural capabilities: on-site language interpreters and multilingual clinicians (Table 4). This adjustment also revealed that these sites were less likely than others to have frequent meetings to discuss quality and to provide weekend care. Similarly, economic disproportionate-share practice sites were more likely than other sites to have language interpreters and multilingual clinicians after adjustment for practice size (Table 5). The other unadjusted relationships we observed were no longer statistically significant. Other practice characteristics (network affiliation, teaching status, multispecialty status, and geographic location) also appeared to mediate some relationships between disproportionate-share status and structural capabilities. However, adjusting for these characteristics did not narrow the gap between practices with and without disproportionate-share status as dramatically as adjustment for practice size (results not shown).
In a statewide sample of primary care practice sites, we found that practices serving disproportionate shares of patients residing in neighborhoods with high prevalences of minority race/ethnicity or economic disadvantage were more likely than others to have the following 3 key structural capabilities: on-site language interpreters, multilingual clinicians, and frequently used, multifunctional EHRs. In addition, minority disproportionate-share practices were more likely to have staff to assist patient self-management of chronic disease, and economic disproportionate-share practices were more likely to report physician awareness of patient experience ratings. None of the structural capabilities we examined was less likely to be found in disproportionate-share practices than in other practices.
These findings were contrary to our expectations, and they differ from previous studies suggesting that disproportionate-share practices may lack the resources necessary to support high-quality primary care.4,5 There are several potential explanations for this divergence. First, our practice site sociodemographic profiles were based on commercially insured patient populations, so we could not include practice sites serving only publicly insured and uninsured patients (eg, some CHCs) in our analyses. Second, the practice capabilities we assessed differ from those investigated in prior studies, in which the principal findings were based on disparities in access to specialty care, quality of physician-patient interactions (eg, continuity of care), and workplace stress. Third, Massachusetts disproportionate-share practices tended to be larger than others. Practice size appeared to mediate some observed relationships between disproportionate-share status and structural capabilities. Disproportionate-share practices may not be larger than others in states with different characteristics (eg, those with significant rural populations).3
Our findings may be especially relevant to medical home demonstration projects funded by commercial health plans. By assessing structural capabilities at the practice site level, our study anticipates the kind of criteria that may qualify practices for medical home enhanced payments.16,17 To the extent that the medical home designation will be based on structural capabilities like those we studied, it appears that practice sites serving sociodemographically vulnerable neighborhoods may not face unusually high barriers to enhanced payments. In Massachusetts, the medical home movement may in fact represent a mechanism for increasing the relative resources available to care for these patient populations.
The persistence of the positive association between practice sites' language capabilities and disproportionate-share status, even after stratifying by practice size, is striking. This robust finding suggests that disproportionate-share practices may reliably be expected to benefit from the inclusion of language capabilities in medical home definitions. In a sense, medical home enhanced payments may function as a means of reimbursing primary care practices for the investments necessary to serve patients with limited English proficiency. As the criteria used to qualify practices for medical home payments evolve across demonstration projects, payers and policy makers may wish to take these distributional considerations into account. The degree of emphasis on language capabilities may emerge from negotiations among constituencies serving different shares of patients with limited English proficiency.
Our study has limitations. First, because our data on practice sites' patient panels were derived from commercial health plan enrollee files, other patients (covered by Medicare, Medicaid, or no insurance) could not be included in the geographic retrofitting of practices to the neighborhoods they serve. Therefore, the prevalences of minority race/ethnicity and economic disadvantage in the retrofitted communities of the practices may be higher than our estimates, particularly for practices with high percentages of patients without commercial health insurance. Practices exclusively serving patients without commercial health insurance could not be included in our analysis. Second, it is possible that some practices attract patients whose individual sociodemographic characteristics do not reflect their neighborhoods of residence. However, the classification of most CHCs as disproportionate share lends validity to the geocoding approach. Third, our practice survey was designed before current medical home standards emerged and did not anticipate all of the structural capabilities subsequently advanced by others.16,17 Fourth, our reliance on a single physician respondent to represent each practice site may generate imprecision in the measurement of structural capabilities. Fifth, we were unable to collect reliable data on practices' financial performance or on forms of external support that might have contributed to practices' structural capabilities.
Finally, findings from Massachusetts may be difficult to extrapolate to other states, especially those with differing systems for delivering primary care to sociodemographically vulnerable patients. For example, Massachusetts has a strong culture of not-for-profit health care, and a large proportion of care is delivered by academic health centers and their affiliated practices. Future analyses of how medical home capabilities (including those promulgated by the National Committee for Quality Assurance) are distributed in other areas of the country may provide additional information.
Medical home proposals include financial rewards for primary care practices that implement selected structural capabilities. Although these resources may be critical to sustaining primary care, they will only accrue to qualifying practices and, by extension, to the communities of patients served by these practices. The concentration of sociodemographically vulnerable patients in disproportionate-share practice sites therefore raises concerns about the distributive effects of these proposed payments: could they worsen disparities in care? Our analysis of Massachusetts primary care practices suggests that paying for structural capabilities may instead benefit practices serving patients from sociodemographically vulnerable neighborhoods. Whether medical home payments will produce these beneficial effects on disparities will depend on the exact criteria used to qualify practices as medical homes and on the organizational characteristics of disproportionate-share practices, which may vary across the country. Given this uncertainty, policy makers should carefully monitor the distributive impact of medical home payments as demonstration projects unfold.
Correspondence: Eric C. Schneider, MD, MSc, RAND Health, 20 Park Plaza, Seventh Floor, Ste 720, Boston, MA 02116 (email@example.com).
Accepted for Publication: December 5, 2009.
Author Contributions: Dr Friedberg 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: Friedberg, Coltin, Safran, and Schneider. Acquisition of data: Friedberg, Coltin, Safran, and Dresser. Analysis and interpretation of data: Friedberg, Coltin, Safran, and Schneider. Drafting of the manuscript: Friedberg. Critical revision of the manuscript for important intellectual content: Coltin, Safran, Dresser, and Schneider. Statistical analysis: Friedberg, Safran, and Schneider. Obtained funding: Safran and Schneider. Administrative, technical, and material support: Coltin and Dresser. Study supervision: Schneider.
Financial Disclosure: None reported.
Funding/Support: This study was supported by the Commonwealth Fund and National Research Service Award 5 T32 HP11001 20 from the Health Resources and Services Administration (Dr Friedberg).
Previous Presentation: Preliminary results from this study were presented at the Annual Meeting of the Society of General Internal Medicine; May 15, 2009; Miami, Florida.
Role of the Sponsors: The funding sources had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or in the preparation, review, or approval of the manuscript.
Online-Only Material: eAppendixes and an eTable are available at http://www.archinternmed.com.
Additional Contributions: Katherine Howitt, MA, assisted in fielding the survey. Elaine Kirshenbaum, MPH, of the Massachusetts Medical Society advised us regarding development of the physician survey.