Murray-García, Jann L. MD, MPH; García, Jorge A. MD, MS; Schembri, Michael E.; Guerra, Lisa M. MD, MS
The U.S. population's increasing racial, ethnic, and linguistic diversity is well documented,1,2 but the nation's physician workforce has not kept pace with these dramatic demographic shifts.3–5 While underrepresented groups (African Americans, Latinos, and Native Americans) make up 25% of the population,1 members from these groups compose only 7% to 8% of the physician workforce.6,7
At the same time, increasing evidence in the medical literature documents that physicians of color provide a disproportionate amount of primary care services to diverse and poor communities.3,8–12 For example, using data from the 1987 National Medical Expenditure Survey, Moy et al.11 found that nonwhite patients were more than four times as likely as were white patients to receive their care from nonwhite physicians, the latter including Asian physicians. Nonwhite physicians were also more likely to serve low-income, Medicaid, and uninsured patients. Using the same database and controlling for patients' socioeconomic factors, Gray and Stoddard9 found that African-American patients were more than three times as likely and Latino patients were more than nine times as likely as were white patients to see a physician of the same race or ethnicity.
These and other studies3,8–12 have relied on both physicians' and patients' self-reports. We know of no study that has analyzed actual utilization or visit-level data, independent of these self-reports. We also know of no study that has explored whether these patterns begin after residency training or are apparent before physicians leave their training institutions. Further, although language differences in the patient—physician encounter can have profound implications for both the quality and the outcome of health care services,13–16 none of the studies documenting disproportionate service patterns has controlled for second-language proficiencies of physicians in their relationships with patients.
We analyzed computerized utilization and administrative data from the resident continuity clinics at Children's Hospital Oakland (CHO). Located approximately 15 miles east of San Francisco, CHO is the oldest, largest, and busiest exclusively children's medical center between Seattle and Los Angeles. Both its large primary care patient population and its housestaff (approximately 70 pediatric residents in total) are racially, ethnically, and linguistically diverse. The goals of our study were to determine whether the patterns of racial or ethnic concordance between physicians and their patients that had been described among practicing physicians were present in residency training, and to explore to what extent these patterns could be explained by the second-language proficiencies of the pediatric residents. Our data set does not include information on how patients were assigned or chose resident physicians for each visit. Thus, this is a study focused solely on residents' service patterns, and not on patients' choices or staff-referral patterns.
Data and Sample
The study was approved by the CHO Institutional Review Board.
We obtained gender, race or ethnicity, and second-language proficiencies of residents from administrative data collected at the initiation of employment. Our analyses included residents' gender because women physicians have been shown to serve higher proportions of poor and minority patients.3,10 Residents self-identified their racial or ethnic backgrounds by responding to a forced-choice question with the options: American Indian, Asian/Pacific Islander, African American, Latino (a mutually exclusive category for “Hispanics,” irrespective of race), white, and unknown. Before employment, residents reported the secondary languages in which they were fluent to the appointment-scheduling staff, with the understanding that they would provide ongoing primary care in these non-English languages. While these residents reported fluency in a number of second languages, our final analyses only included those that accounted for at least 50 patient visits during the study period (viz., Spanish, Cambodian, and Vietnamese). Residents who began their employment at CHO as second- or third-year residents were coded by level of training rather than by their lengths of employment at CHO.
We obtained patients' data from CHO's Hospital Information System. This centralized computer data bank contains information from a variety of sources, including appointment-scheduling phone contacts, outpatient and inpatient registration interviews, and billing processes. Patients' insurance status, race or ethnicity, and language preferences were obtained from this source. Patients' visits were coded as Medi-Cal (Medicaid) or non—Medi-Cal to control for the potential confounding effects of economic status.3,10,11
Data on patients' race or ethnicity are entered into the computer system at the time of each patient's initial visit to CHO. Hospital protocol requires appointment and registration clerks to ask parents or guardians about the patient's race or ethnicity, theoretically preventing the clerks from independently assigning these categories to patients. Patients or guardians who do not provide spontaneous responses are offered a list of 49 options from which to choose. The parents or guardians of biracial children must designate one race or ethnicity or choose the “other” category. All patients are further classified as either “Hispanic” or “non-Hispanic,” so that Hispanics actually constitute a heterogeneous group of individuals with diverse racial backgrounds. For the analyses in the current study, we placed all Hispanics into a mutually exclusive “Latino” category, with this ethnicity taking precedence over any other race or ethnicity listed. Asian ethnic groups were designated “Asian” to match the corresponding aggregated category of residents of this race. We recognize that this aggregated category of Asian obscures substantial differences in languages, immigration histories, health practices and beliefs, etc.17,18 The other categories used in our analyses were African American, white, and other. The language preferences of parents and guardians or older patients were obtained by appointment or registration clerks, who asked, “What is the language spoken at home?”
In May 1998, CHO began subcontracting all of its outpatient clinic scheduling to an outside agency, initiating a slightly different though highly standardized system for obtaining patients' demographic data. Race or ethnicity and language data obtained prior to this conversion remained unchanged in the computerized data bank. The protocol for recording which resident saw the patient also changed and, reportedly, became more consistent. Hence, we selected a database that began with May 1998.
From May 1998 through October 1999, there were a total of 14,117 visits to the resident continuity clinics at CHO. The patients' race or ethnicity variables were missing on 3% of the records for these visits, and these visits were excluded from our study. The final study set was 13,681 visits.
Studney and Hakstian19 have reported that billing records may not accurately reflect who and what were involved in the clinical encounter. Concerned that such inaccuracies could compromise our analysis of race or ethnicity among patients and their physicians, we randomly selected 140 visits from the final study set to assess the accuracy of the physicians' race or ethnicity as the variable in the billing data. A professional medical coder, blinded to the study hypotheses, was hired to perform the chart abstraction, obtaining from each visit's record (chart) the identity of the resident who actually saw the patient. While very familiar with the signatures of the residents who were employed during the study period, the coder had not been previously responsible for entering data into the computerized outpatient billing records (the database utilized in our study). Using race concordance between residents and their patients as the outcome in our chart abstraction, we obtained a kappa value of .79 (95% CI, 0.75–0.83), well above the acceptable level of .61.20
We performed statistical analyses using a standard statistical software package. We computed the distribution of patients by race or ethnicity, insurance status, and language preference; of residents by gender, race or ethnicity, and second-language proficiency; and of patient visits by race or ethnicity, language preference, and residents' level of training. For each racial or ethnic group of patients, we used the chi-square test to compare the proportion of patients' visits to residents of the same race or ethnicity with the proportion of visits to all residents who were of a different race than their patients.
We then used multivariate logistic regression to further explore the extent of concordance between patients' and residents' races or ethnicities while adjusting for potential confounders. We constructed a separate but structurally similar model for each racial or ethnic group of patients. In each case, the dependent variable was a clinic visit made by a patient from the racial or ethnic group of interest. All models included the same covariates identified in the literature as potentially influential: resident's gender, resident's level-of-training, and patient's insurance status (Medi-Cal versus non—Medi-Cal status).3,8–12 However, each model also included a unique “concordance covariate” for the resident's racial or ethnic group corresponding to the patient's group of interest, with all residents for whom race or ethnicity did not match that of their patients as the reference group. Where appropriate, we examined models with and without variables for residents' second-language proficiencies to explore whether these variables mediated the effects of residents' race or ethnicity on visits with patients of the same race or ethnicity. Because African-American and Latino patient visits were very common outcomes in our sample (>10%), we calculated adjusted relative risk ratios (aRRs) as better estimates of the true risk ratios.21
Finally, because 16 non-Latino residents and all 13 Latino residents reported second-language proficiency in Spanish, we were able to further explore the independent association of residents' ethnicity on patient—resident dyads. We performed two additional logistic regression analyses on the subset of Spanish-speaking residents to determine whether Latino patients were more likely to be seen by Latino or non-Latino residents. The dependent variable in the first model was a clinic visit made by a Latino patient, regardless of language preference. In the second model, the dependent variable was a clinic visit made by a Spanish-speaking Latino patient. The key independent variable in each model was Spanish-speaking Latino resident, with Spanish-speaking, non-Latino residents serving as the reference group. Both models included the potential confounders listed above, and we computed aRRs as previously noted.
Table 1 shows summary statistics for the patients and patient visits in our sample by race or ethnicity, language preference, and insurance status. Most of the patients in each racial or ethnic group had Medi-Cal insurance. In the majority of visits, patients preferred to speak in English (83.5%) or Spanish (12.2%).
Table 2 shows the characteristics of the residents. As expected, third-year residents saw more patients than did second-year residents, and second-year residents saw more patients than did first-year residents. A total of 68% of residents were women, comparable to the percentage of women in pediatrics residency programs nationwide (61%).22 None of the residents in our sample was designated “American Indian” or “unknown.” A fourth (24.8%) of the residents were fluent in Spanish, 5.1% were fluent in Vietnamese, and 8.5% were fluent in one of several other languages.
Figure 1 shows the unadjusted proportion of patient visits by race or ethnicity to each racial or ethnic group of residents. The highest proportions of visits by African-American, Asian, and Latino patients were made to residents of the same racial or ethnic group. For example, Latino patients composed 60% of the total number of patient visits to Latino residents. The next highest proportion of visits by Latino patients was made to white residents (17%). For visits by African Americans, Asians, and Latinos, the proportion of visits to all residents of a different race or ethnicity was significantly less than the proportion of visits to residents of the same race or ethnicity. This pattern was not seen among white patients.
Additional calculations indicated that African-American residents made up 15% of all residents and saw 19% of all African-American patients. Asian residents made up 30% of the residents and saw 46% of the Asian patients. Latino residents made up 11% of all residents and saw 34% of all Latino patients. White residents made up 44% of all residents and saw 42% of all white patients.
Multivariate analyses showed that these service patterns persisted after adjusting for residents' gender, level of training, and patients' Medi-Cal status (see Table 3). Compared with patients from other racial or ethnic backgrounds, African-American patients were 1.23 times more likely to see African-American residents (95% CI, 1.19–1.26); Asian patients were 1.86 times more likely to see Asian residents (95% CI, 1.61–2.16); and Latino patients were 4.58 times more likely to see Latino residents (95% CI, 4.35–4.79). White patients were not more likely than were non-white patients to see white residents (aRR 1.03; 95% CI, 0.89–1.20).
After controlling for residents' second-language proficiencies, risk ratios remained statistically significant: Latino patients were over twice as likely to see a Latino resident (aRR, 2.36; 95% CI, 2.12–2.61), and Asian patients were 1.56 times as likely to see an Asian resident (aRR 1.56; 95% CI, 1.32–1.84; see Table 3).
Table 4 shows the results of logistic regression analyses that included Spanish-speaking residents only. After adjusting for residents' gender, level of training, and patients' Medi-Cal status, Latino patients, regardless of language preference, were 1.78 times as likely to be seen by a Latino resident (all of whom spoke Spanish) than by a non-Latino Spanish-speaking resident (95% CI 1.66–1.89). Latino patients with a Spanish-language preference were also nearly twice as likely to be seen by Latino residents than by non-Latino Spanish-speaking residents (aRR, 1.96; 95% CI 1.80–2.11).
We examined the service patterns of a diverse group of pediatric residents serving a large, multiethnic population. Our study used data on over 13,000 patient visits during an 18-month period and five different classes of residents at CHO. We found that African-American, Asian, and Latino residents provided primary care services to disproportionately higher numbers of patients from their own racial or ethnic backgrounds. These findings persisted after controlling for residents' gender, level of training, fluency in the primary language of their patients, and patients' Medi-Cal status. In fact, among patients' visits to all residents who spoke Spanish, Spanish-speaking Latino patients were nearly twice as likely to be seen by Latino residents than by non-Latino residents. Further, among all visits to Spanish-speaking providers, the proportion of Latino patient visits, Spanish-speaking or not, was nearly two times greater among Latino residents than it was among non-Latino residents. Together these findings suggest a distinct pattern of racial or ethnic concordance between residents and patients that is, to a substantial extent, independent of the residents' linguistic abilities.
Previous studies have documented that physicians of color provide disproportionately more care to patients of color.3,8–12 Our study extends these findings in several important ways. To our knowledge, ours is the first to report that these disproportionate service patterns are present in residency training, and that they are, to some extent, independent of a resident's proficiency in the language of his or her patient. Our study also used actual volumes of patient visits, whereas previous studies had been unable to verify physicians' self-reports and estimates of service patterns. In addition, while prior studies had relied on physicians' racial or ethnic designations of patients and vice versa, our study relied on patients' and residents' self-designations of race or ethnicity from independent sources.
Finally, this study is unique in that all patients and residents were associated with a single clinical site. Theoretically, residents had equal access to the same diverse set of patients, and patients had equal access to the same diverse set of residents. While patients within a busy resident continuity clinic are not completely free to choose their regular resident physicians or even a “substitute resident” on a specific day, residents' availability and assignment patterns should not vary significantly by residents' race or ethnicity. This is in contrast to previous national3,9,11,12 and regional studies8,10 wherein the likelihood of a physician serving a diverse population base was biased by that physician's choice of practice location. Only the study by Komaromy and colleagues10 controlled for the demographic profile of each physician's practice location. These investigators found that African-American and Latino physicians in California served higher proportions of patients from their respective racial or ethnic groups over and above what the demographic profiles of their practice locations would predict.
In attempts to explain these service patterns, one might reason that patients, residents, or appointment staff assume that demographic similarities between patients and residents enhance the therapeutic alliance. This is perhaps most easily seen in the matching of patients and residents by language proficiency. Evidence suggests that when physicians are proficient in the languages of their non-English-speaking patients, these patients miss fewer appointments,13 are more compliant with medications,13 ask their physicians more questions,15 recall more of their physicians' recommendations,15 and report better health status, psychological well-being, and physical functioning.14
However, our regression analyses suggest that the degree to which Asian and Latino residents serve patients of their own race or ethnicity is to some extent independent of language concordance (see Tables 3 and 4). And of course, second-language proficiency does not explain the concordance patterns we found among African-American patients and residents.
With regard to race and ethnicity, just as some women patients prefer to see women physicians, here too perhaps patients matched themselves or were matched on another dimension of their identities (i.e., ethnic identity) that was believed to have an impact on the encounter with the health care provider. Communication styles, world views, health beliefs and practices, dietary practices, assumptions of equitable service delivery, understanding of immigration histories, or other collective experiences may be optimized when the physician is of the same race or ethnicity as the patient. In pediatric populations, the choice or assignment of a resident physician who is of the same race or ethnicity as the patient may be influenced by parents' disciplinary goals and practices, educational goals and learning styles, and even the desire for role models for their children who are of the same race or ethnicity.
Some researchers have suggested that, at least for a subset of patients from diverse backgrounds, having a physician of the same race or ethnicity may enhance care.14,23–25 For instance, in a study of nearly 300 AIDS patients in Boston, nonwhite patients were half as likely as were white patients to have discussed end-of-life decisions with their physicians.24 However, nonwhite patients with nonwhite physicians were more than four times as likely as were nonwhite patients with white physicians to have discussed end-of-life care. Cooper-Patrick et al.23 report higher levels of physicians' participatory decision making styles and higher patient satisfaction in physician—patient relationships of the same race or ethnicity. A national study of patient satisfaction25 documented higher patient satisfaction among African Americans whose physicians were African-American compared with those whose physicians were not.
The disproportionately higher levels of service rendered by physicians of color to patients of color may also reflect long-standing commitments on the part of underrepresented group members to diverse and underserved communities. Several studies document that medical students and residents from underrepresented backgrounds enter training with an intent to serve these communities and fulfill these commitments when they complete their training.3,8,26
Our findings must be interpreted in light of several important limitations. First, we had little if any information with which to determine who in the chain of potential decision makers was primarily responsible for the patterns of racial or ethnic concordance demonstrated in this study. In fact, there are several ways in which patients are assigned to residents. Patients or parents or guardians may call for an appointment to establish a relationship with a resident and be assigned to the first available resident. Alternatively, patients or parents or guardians may request to see a specific resident, perhaps on the referral of a relative or friend. Appointment staff are encouraged to accommodate families' need for residents who speak the same language. A list of residents' second-language proficiencies—including those of interns and other new residents—is compiled at the beginning of each academic year at CHO and is readily available to appointment staff. Anecdotally, these staff also note that it is not unusual for parents or guardians to request a resident of the same race or ethnicity. Patients or family members may also solicit residents to undertake their primary care during the course of acute care or other clinic encounters. Patients or family members are often given the opportunity to choose from the roster of residents' photographs at the time of a resident's graduation. Conversely, graduating residents may “hand-pick” a junior resident to assume a patient's care. Patients may be referred from an inpatient service or other clinic by a resident, attending or community physician, social worker, nurse specialist, etc. Finally, residents may “recruit” patients into their primary care clinics from their various outpatient and inpatient rotations.
The complex interplay of factors that ultimately determines which patients and residents enter into therapeutic relationships precludes definitive explanations for the service patterns seen in our study. Nonetheless, it is clear that some combination of decision makers felt that it was desirable to have patients see residents of the same race or ethnicity.
Uncertainties about the integrity of the race and ethnicity data represent an additional limitation of this study. In particular, systematic errors in collecting information about race or ethnicity may have introduced bias into the study. For example, such errors could have resulted if registration clerks were more likely to label a patient as being of the same race or ethnicity as the resident, in lieu of asking the patient. Billing personnel might have been more likely to record a resident's name according to race or ethnicity of the patient if, for example, the name of the resident on the chart's visit record was not legible or missing. Our abstraction study argues against this latter possibility, with a high kappa value of .79. In fact, the data source in general seems to be of exceptional quality, in that only 3% of the over 14,000 visits were missing the race or ethnicity variable, and less than 0.1% were missing the language-preference variable. Finally, we cannot know how many patients or parents, as dictated by protocol, listed only one racial or ethnic descriptor, or how they came to choose that one descriptor when they or their children were in fact biracial.
Limitations in the generalizability of our findings must also be considered. In this regard, studies with patient and resident populations in different specialties should be undertaken. Because we focused on a pediatric study population, our results must be considered in light of an important caveat: Contained within the unit of analysis in this study—the patient visit—are varying degrees and combinations of child—parent—resident, adolescent—parent—resident, or adolescent—resident influences. For each visit, the nature of this dynamic depends on the cognitive abilities, psychosocial maturity and initiative, and degree of ethnic consciousness or level of acculturation of the child, as well as the extent of parental authority, prevalance of increasing English proficiency in children, and, perhaps, the decreasing dependence on parents' language preferences. Intriguing questions inevitably arise about whose choices are reflected in the patient—resident pairings we have described. Although the answers to these questions clearly exceed the capacity of our data set and study design, they are certainly worth pursuing in future investigations that include qualitative data.
Studies with patient populations in different geographic regions are also needed. In particular, it would be interesting to study a population made up of proportionately fewer African-American patients to see whether there is a greater difference between African-American and non—African-American residents in the patterns of service rendered. All CHO residents necessarily see substantial proportions of African-American patients, who make up over 60% of patient visits to residents' continuity clinics. Also of interest would be a study in which the service patterns of Spanish-speaking, non-Latino residents could be compared with those of Latino residents who do not speak Spanish. All of the Latino residents in our sample spoke Spanish.
Our inability to demonstrate the fluency of the residents in non-English languages or to determine whether patients were truly not proficient in English represents additional and related limitations. Were the excess number of “English-proficient” Latino patients who were seen by Latino residents truly proficient in English? In the analyses of Spanish-speaking residents, did proportionately fewer Spanish-speaking patients visit non-Latino residents compared with Latino residents because of lower levels of Spanish-speaking proficiency in the non-Latino residents? Of course it is difficult to measure and model the ways language and culture may be inextricably linked due to cultural nuances in communication (e.g., choice of words, slang, nonverbal communication, etc.). Indeed, the patient's sense of “being heard and understood” might encompass much more than technical linguistic competence, including whatever is entailed in ethnocultural membership or experience.
While no stringent determination of language fluency exists in this system, all CHO residents are asked to disclose their second-language proficiencies at the start of residency, with the understanding that they will establish longterm primary care relationships with patients. If residents do not differ in their tendencies to over- or underestimate their language abilities under these circumstances, then it is unlikely that disparate levels of Spanish proficiency completely explain these findings. The disproportionate service patterns we found among the African-American and Asian residents, the latter of whom were all native speakers of their Asian languages, also argue against levels of Spanish fluency as a complete explanation for the disproportionate service patterns among the Latinos.
In addition, because our data are from a single pediatric medical center, institution-specific factors should be considered. It is possible that CHO residents feel less comfortable in cross-cultural encounters, and thus disproportionately seek out patients of the same race. The CHO residency program, however, unabashedly advertises its commitment to providing excellent service to the diverse surrounding communities, and it boasts an innovative multicultural training component to its residents' training. The hospital, as a community institution, has a long-standing practice of training a diverse group of pediatrics housestaff, with the explicit goal of improving access to care for the medically underserved populations in the San Francisco Bay Area and beyond. With all this, CHO is continually a popular choice for pediatrics training across the country, with over 600 applications for 25 internship positions annually. One might thus reasonably postulate that those residents who do decide to train at CHO are actually more comfortable with and actively seek out cross-cultural encounters. If this is true, the striking patterns of racial or ethnic concordance between residents and their patients here take on even greater significance.
A final limitation of the study lies in the fact that visit-level data do not indicate the designated primary care providers of patients. It is debatable whether it is more important to know the demographic profile of residents' patient panels, or to know who actually did the work of seeing specific patients over time. Given that our analyses allow us to demonstrate who actually rendered quantifiable service to diverse groups of patients, we see this study “limitation” as an important and unique contribution to the existing literature.
Our findings have important implications for physician work-force policy. First, the contributions that racially, ethnically, and linguistically diverse housestaff make to their training institutions and surrounding communities should not be understand. In addition to the disproportionate service they render to patients of color, a diverse housestaff, with attendant cultural experiences and insights, enhances the educational environment of the institution.4,27 At the same time, it should be recognized that a fine line exists between extolling the unique cultural and linguistic skills of residents, and exploiting residents from groups that remain underrepresented in the physician workforce.
Long-term solutions begin with diversifying the “pipeline” of elementary through college-age students interested in entering medicine,4,28 recognizing the economic class and intragroup diversity of individuals. Indeed, one of the objectives of Healthy People 2010, the nation's action agenda for preventive health, mandates an increase in “the proportion of all [health professions] degrees awarded to members of underrepresented racial and ethnic groups … as an integral part of the solution to improving access to care.”29 In the shorter term, if patients are making the conscious choice to see physicians of the same race or ethnicity, and if health care systems are assigning patients to achieve the same end, we need to explore the idea that race and ethnicity may reflect a unique set of skills or qualifications for matriculation to medical training programs, even if only from the patient's perspective of what is desirable in a physician.4,6,30,31
Our results also have implications for the training of all physicians in effective and respectful cross-cultural clinical practice. Though racial and ethnic parity in the physician workforce is a laudable goal, it remains a long way from being achieved.5,6 And yet, as early as the first year at one medical school, medical students had a sense of which racial or ethnic groups of patients they would be more comfortable with, and which of these patients would be most likely to comply with their treatment recommendations.26,32 To know what to teach most specifically, and to optimize cross-cultural dyads, more needs to be known about what it is in the race-concordant dyad that can be transferred to the discordant pairing. How do racially and ethnically diverse patients define optimal communication, issues of trust, and desirable health outcomes? How can current and future physicians achieve the ideal of “cross-cultural competency”? Answers to these questions are beyond the scope of this study, but should receive urgent and substantive attention as physician workforce and medical education policies are refined in the 21st century.