Racial Disparity, Primary Care, and Specialty Referral Jayasree Basu and Carolyn Clancy Objective. The study examines the role of primary care physicians (PCP) in reducing racial disparities in referral-sensitive admissions. Data Sources/Study Setting. The study examined hospital discharges of New York residents in the age group 20 to 64 hospitalized either in New York or in any of three contiguous states—New Jersey, Pennsylvania, or Connecticut—using complete discharge files for the four states in 1995. The discharge data were linked to the Area Resource File and the American Hospital Association’s survey files for 1995. Study Design. The study used multivariate logistic models to compare the effect of PCP supply on referral-sensitive (discretionary) admissions versus marker admissions (urgent, insensitive to primary care) for whites, blacks, and Hispanics. Principal Findings. Compared with marker admissions, an increased PCP density in an area was associated with a higher probability of black admissions than white admissions for referral-sensitive procedures. Conclusions. Our analysis suggests that the supply of primary care may be important for admissions for referral-sensitive procedures. Nonwhites, especially blacks, may have greater access to these procedures through increased PCP supply. Increased PCP density may significantly narrow the racial disparity in specialty referrals and improve the referral process for these procedures for nonwhites. Key Words. Primary care, racial disparity, referral-sensitive admissions

INTRODUCTION A growing body of literature reveals that members of minority ethnic and racial populations consistently receive fewer invasive, “high-tech” interventions. These disparities have been most extensively studied for cardiac disease and blacks (Sheifer, Escarce, and Schulman 2000; Einbinder and Schulman 2000) but have also been found for a variety of other conditions such as lung cancer and renal transplants (King and Brunetta 1999; Ayanian et al. 1999; Ayanian and Epstein 2001). Most studies have used hospital discharge abstracts or billing claims to provide cross-sectional snapshots of care and have offered limited insights re-

64

Racial Disparity, Primary Care, and Specialty Referral––65

garding the pathways preceding the index admission (Horner, Oddone, and Matchar 1995). In recent years interest in the effect of referrals to specialists by primary care physicians (PCP) on overall costs and quality of care has increased in part because of incentives offered by managed care arrangements to rationalize the use of specialty services. Of particular note, selected studies have demonstrated that coordination between PCPs and specialists can improve the quality of the referral process as well as health outcomes (e.g., Epstein 1995; Forrest, Glade, Baker, et al. 2000). In other words, PCPs may function as both “gateways” and gatekeepers (Franks, Clancy, and Nutting 1992). In this study we examine the relationship between PCP supply and hospital admissions for certain high-cost hightechnology surgical procedures for which a specialty referral is needed (often referred to as referral-sensitive surgeries). Referral-sensitive surgeries are fairly discretionary, often elective, high-technology procedures, which often involve or require a referral from a PCP to a procedural specialist. These surgical procedures include hip or joint replacement, breast reconstruction after mastectomy, pacemaker insertion, organ and bone marrow transplantation, most coronary artery bypass graft surgery, and coronary angioplasty (Billings, Zeitel, Lukomnik, et al. 1993). While most high-tech procedures are referral sensitive, not all referral-sensitive conditions require high-tech procedures. Multiple studies have confirmed the existence of racial disparities in receipt of referral-sensitive procedures, but few studies have been able to identify the causes or clarify the mechanism. The premise of this study is that an increased supply of PCPs will be associated with decreased rates of disparities in referralsensitive admissions. This premise is based on the hypothesis that a relative PCP shortage will affect referrals for minorities more than other racial groups. There are essentially two pathways through which an individual receives specialized services. In one pathway a PCP refers a patient to a specialist (gatekeeper model); in the second an individual consults a specialist directly (self-referral model, e.g., “I

——— The views expressed in this article are those of the authors. No official endorsement by any agency of the federal government is intended or should be inferred. Address correspondence to Jayasree Basu, Ph.D., Senior Economist, Center for Primary Care Research, Agency for Healthcare Research and Quality, 6010 Executive Boulevard, Suite 201, Rockville, MD 20852. Carolyn Clancy, M.D., is Director, Center for Outcomes and Effectiveness Research. This article, submitted to Health Services Research on August 16, 2001, was revised and accepted for publication on November 1, 2001.

66––HSR: Health Services Research 36:6, Part II (December 2001)

have chest pain so I should go directly to a cardiologist”). Expansions in the proportion of Americans enrolled in point-of-service or open-access plans reflect a market perception that a growing proportion of individuals prefer having, if not always exercising, this option. However, studies show that self-referral to specialists is not particularly common among minorities. Self-referral is associated with higher patient education level, but a complete explanatory framework has not yet been elucidated (Clancy and Franks 1997). While prior studies have suggested that PCPs may facilitate access to specialty care (Franks, Nutting, and Clancy 1993; Forrest, Glade, Starfield, et al. 1999), no attempt has yet been made to examine the relationship between PCP supply and observed disparities in the receipt of referral-sensitive procedures, that is, whether limited access to PCPs is a potential explanation for differential procedure rates for minority populations. This study will examine whether increased supply of PCPs could serve as a possible mediating factor to reduce racial disparities associated with increased access to referral-sensitive admissions among minorities. We focused on nonelderly adult residents (aged 20 to 64) in the state of New York. This age group was selected because studies found that barriers to access are higher among the young adult and middle aged populations (Billings, Zeitel, Lukomnik, et al. 1993). Children were excluded because conditions leading to these surgeries are much less frequent among children.

BACKGROUND AND HYPOTHESIS The incidence of hospitalization for referral-sensitive surgeries has traditionally been low for nonwhites. In 1995 only 8 percent of adult New Yorkers admitted for these procedures were blacks, and 5 percent were Hispanics. The proportions of these groups in the entire adult population of New York in 1995 were, respectively, 15 percent and 9 percent. Studies of referral patterns generally have found race to be an important predictor of disparity in the referral process (Shea et al. 1999). For example, McBean and Gornick (1994) reported that black Medicare beneficiaries are generally less likely than white beneficiaries to have received major procedures, and that the largest racial differences are seen for referralsensitive surgeries such as percutaneous transluminal coronary angioplasty, coronary artery bypass graft surgery, total knee replacement, and total hip replacement. A previous study using 1995 discharge data for New York adult residents also found that blacks are less likely than whites to be admitted for referralsensitive surgeries relative to other conditions such as ambulatory care–sensitive

Racial Disparity, Primary Care, and Specialty Referral––67

conditions (preventable admissions) or nonpreventable urgent conditions (Basu, Friedman, and Burstin, 2001). One study suggested that enrollment in an HMO may tend to reduce racial disparities (Clancy and Franks 1997) in the referral process, but the sample size in that study was limited. The hypothesis for this study was that increases in the number of PCPs per population in an area will be associated with proportionately higher rates of nonwhite admissions for referral-sensitive procedures relative to other admission types. To test this hypothesis, referral-sensitive admissions were compared with marker admissions. Marker admissions are usually urgent and nonelective. These conditions have been defined as diagnoses for which provision of timely and effective ambulatory care immediately prior to admission will likely have little effect on the need for hospital admission and where there is substantial agreement among practitioners on the clinical criteria for admissions (Billings, Zeitel, Lukomnik, et al. 1993). Included in such admissions are appendicitis with appendectomy, acute myocardial infarction, gastrointestinal obstruction, and fracture of hip/femur (Billings et al. 1993). Because marker admissions should be insensitive to primary care, they provide an appropriate comparison group.

METHODS AND DATA Study Design This study focused on three ethnic groups: whites, blacks, and nonwhite nonblack Hispanics. The last two groups represent nonwhite populations and consisted of 8 percent and 5 percent, respectively, of adult New York residents hospitalized for referral-sensitive procedures in 1995 (assuming that those with unknown races, 7.7 percent of total admissions, were evenly distributed across different racial groups). Of the remaining population, the majority were whites (80 percent of the total and 92 percent of the total excluding nonwhite Hispanics and blacks). We classified those counties above the statewide average as high PCP– density areas and those below that average as low PCP–density areas (see Table 1). The data for the population were obtained from the U.S. Census Bureau’s web site for 1995 New York resident population classified by age, race, sex, and county. For the multiple logistic model in Table 2, PCP density was used as a continuous variable.

68––HSR: Health Services Research 36:6, Part II (December 2001)

Table 1:–– Density of PCPs, Racial Distribution, and Admission Rates for Referral-sensitive Surgeries in 1995 New York Residents Aged 20 to 64 Counties Grouped by Primary Care Density* Percentage Distribution of Admissions –Below statewide average –Above statewide average Admission Rates per 1,000 Population† –Below statewide average –Above statewide average

White

Black

Hispanic

60.71 39.29

64.32 35.68

69.92 30.08

3.01 2.72

1.03 1.29

0.94 0.88

*Defined as PCPs less pediatricians per 1,000 total population less half of children in a county. † Population data were obtained from the U.S. Census Bureau’s web site, http://www.census.gov /population/estimates/county/casrh, for 1995 New York resident population classified by age, race, and county.

Table 2:–– Adjusted Odds Ratios of Admissions for Referral-sensitive Surgeries Compared to Marker Admissions, Logistic Regression Results for 1995 New York Residents† Aged 20 to 64 Adjusted Odds Ratio (95% Confidence Interval)

Adjusted Odds Ratio— New York City Excluded as a Predictor (95% Confidence Interval)

Black Hispanic‡ Male Age Severity score§ Source of admission = ER Teaching hospital HMO Medicaid fee for service Uninsured Nonadjacent rural Metro New York City Median family income PCPs per 1,000 Per capita inpatient days Specialists per 1,000

0.579* (0.444, 0.754) 0.845 (0.622, 1.148) 1.319* (1.254, 1.389) 1.049* (1.047, 1.052) 1.294* (1.275, 1.314) 0.048* (0.045, 0.051) 2.492* (2.359, 2.633) 1.347* (1.085, 1.673) 0.893 (0.701, 1.137) 0.543* (0.403, 0.731) 1.083 (0.892, 1.314) 0.984 (0.863, 1.122) 0.719* (0.666, 0.777) 1.005** (1.001, 1.01) 0.423* (0.247, 0.726) 0.873* (0.790, 0.965) 1.239* (1.067, 1.439)

0.489* (0.377, 0.636) 0.707** (0.522, 0.958) 1.317* (1.251, 1.386) 1.050* (1.047, 1.052) 1.297* (1.278, 1.317) 0.048* (0.045, 0.050) 2.364* (2.241, 2.494) 1.390* (1.120, 1.725) 0.845 (0.664, 1.075) 0.542* (0.403, 0.730) 1.060 (0.873, 1.286) 0.844* (0.744, 0.957) — 1.015* (1.011, 1.019) 0.336* (0.214, 0.626) 0.822* (0.745, 0.907) 1.211** (1.044, 1.406)

Primary care x Hispanic Primary care x Black

0.946 (0.579, 1.543) 1.668** (1.085, 2.560)

1.118 (0.685, 1.821) 2.002* (1.305, 3.069) Continued

Racial Disparity, Primary Care, and Specialty Referral––69

Adjusted Odds Ratio— Adjusted Odds Ratio (95% Confidence Interval) Primary care x Medicaid Primary care x HMO Primary care x uninsured

0.821 (0.543, 1.237) 0.723*** (0.549, 1.034) 1.337 (0.813, 2.193)

New York City Excluded as a Predictor (95% Confidence Interval) 0.852 (0.564, 1.284) 0.730*** (0.510, 1.044) 1.345 (0.819, 2.202)

*p < .01; ** p < .05; *** p < .10. † Number of cases under referral-sensitive and marker admissions were 24,606 and 24,491, respectively. The admissions with both referral-sensitive and marker conditions were assigned to the marker admission category based on clinical review; thus, the two categories are mutually exclusive. ‡ Hispanic admission will be significant when New York City is not a predictor. § A predicted change based on disease staging classification of Medstat, Inc., normalized to a scale of 1.0.

We used a multivariate logistic model to compare referral-sensitive admissions with marker admissions. The odds ratios in the logistic model represent the odds of admissions for referral-sensitive conditions relative to the odds of admissions for marker conditions. For example, in Table 2, odds ratio for referralsensitive admissions, odds(Y = referral-sensitive admission)/odds(Y = marker), is 131 percent for men compared to women. As a shorthand in some of the discussion below, a reduction in odds of referral-sensitive admission will be relative to the odds of a marker admission. In the case of a continuous variable the odds ratio is the change in relative odds for one-unit change in the independent variable. To examine the effect of PCP supply on racial disparity in referral-sensitive admissions the model includes several variables indicating the interactions of PCP supply with race/ethnicity and with different types of insurance. In addition, the model controls for the effects of several patient-level, hospital-level, and countylevel factors and was run both including and excluding New York City as a predictor (to avoid confounding because New York City had a high concentration of nonwhite patients). Source of Data and Description of Variables Information on hospital discharges during 1995 for New York residents aged 20 to 64 was drawn from complete hospital discharge files for four states: New York, New Jersey, Pennsylvania, and Connecticut. These records were assembled, edited, and standardized as part of the Healthcare Cost and Utilization Project (HCUP) database of the Agency for Healthcare Research and Quality. All New

70––HSR: Health Services Research 36:6, Part II (December 2001)

York residents were included in the analysis, and they were hospitalized either in New York or in any of three contiguous states (New Jersey, Pennsylvania, and Connecticut). Any admissions to Massachusetts or Vermont hospitals were not included (those are expected to be few because of the distance of those borders from population centers in NY). To create the analytic file, inpatient discharge records of New York residents from HCUP files were linked to the 1995 Area Resource File (ARF) for sociodemographic and other information on the patient’s county of residence and to the American Hospital Association’s (AHA) survey files for 1995 for information on the hospital where a patient was treated. The study compared two groups of conditions: referral-sensitive conditions and marker conditions. Groups were defined on the basis of past research by Billings, Zeitel, Lukomnik, et al. (1993). The conditions are usually defined by principal diagnosis codes from the International Classification of Diseases, Ninth Revision (ICD-9-CM) system. In several cases specific exclusion criteria based on age, sex, and selected procedures were used. For example, fracture of hip/femur (a marker condition) is applied to persons over age 45 only, while breast reconstruction after mastectomy (a referral-sensitive procedure) is obviously used for women (for more details see Billings, Zeitel, Lukomnik, et al. 1993, or contact the authors). Independent variables used in these regressions included patient characteristics, county characteristics, and characteristics of the hospitals where the patients were treated. Patient characteristics included age, race, ethnicity, gender, insurance status, source of admission, and severity of illness. Race/ethnicity was grouped into four categories: white, black, Hispanic, and other. We used “white and others” as the omitted category in the logistic model. Insurance status of patients was grouped into Medicare, Medicaid fee for service, Medicaid HMO, selfpay, commercial HMO, and all other types of insurance, which includes principally commercial insurance and a small group of other types of public programs. Three major sources of admissions were considered: admission from emergency rooms, transfer from another facility, and all others. The first two sources are indicators of a relatively high severity of illness. A more direct measure of severity of illness was calculated using a variable called RDSCALE, which is a later development of the Disease Staging System (Gonnella, Hornbrook, and Louis 1984; Coffey and Goldfarb 1986) designed by Medstat, Inc. RDSCALE is a single resource-based predictor assigned to each patient and represents a patient’s within–diagnosis-related group (DRG) severity and the complexity of his or her DRG (Christoffersson, Conklin, and Gonnella 1988).

Racial Disparity, Primary Care, and Specialty Referral––71

County-level variables included geographic location of the county, sociodemographic condition, and county resources. Data for these variables were obtained from the 1995 ARF. In terms of geographic location, residents were grouped into three categories: metropolitan, nonmetropolitan adjacent (to metropolitan areas), and nonadjacent rural. We also used New York City, comprising four large counties (Kings, Queens, Brooklyn, Bronx), as a separate independent variable (NYC) because of the high concentration of minorities in this area. To capture county demographic factors, median county family income and the proportion of the nonwhite population were included as two independent variables. To account for the effects of differences in the PCP supply while controlling for local inpatient capacity and specialty capabilities, four county resource variables were used: inpatient days per capita, the number of total specialists per 1,000 population, hospital outpatient visits per capita, and the number of PCPs per 1,000 population. Although reflecting utilization, both outpatient visits and inpatient days per capita represent, respectively, total hospital outpatient and inpatient capacities in the county. Because beds are not always reflective of hospital inpatient capacity, we used total hospital inpatient days per capita. The data on outpatient visits were obtained from survey files of the AHA and summed over short-stay community hospitals in a county. Other county variables were obtained from the ARF. PCPs outside hospitals usually include general pediatricians, general internists, and general practice and family practice physicians (this does not include nurse practitioners in primary care, who may be particularly important for HMOs). However, in this study we excluded general pediatricians because their availability does not reflect the resources available to the adults. PCP availability was expressed as the total number of PCPs per 1,000 total population less half of children (in the 19-and-under age group). The denominator included age groups in addition to adults because the availability of PCPs to adults is dependent on the potential use by other age groups. Because some children see PCPs other than pediatricians, half of them were included in the denominator as potentially using PCPs available to adults. According to data released by the National Center of Health Statistics (NCHS 1999) for children who are less than 15 years old, the proportion of office visits to pediatricians was 61 percent in 1995–96. Children above that age group are more likely to compete with adults to get primary care from PCPs other than pediatricians. We also tested alternative models with a primary care density variable using as a denominator the population over age 19 as well as total population.

72––HSR: Health Services Research 36:6, Part II (December 2001)

Other variables that reflect the supply of PCPs include emergency room visits per capita and the proportion of PCPs among total physicians, but these are highly correlated with other variables and were dropped in the final analysis. The number of specialist physicians (including both medical and surgical specialists) in office-based practice was also expressed per 1,000 total population in a county. Another resource variable not used in the analysis because of lack of detailed county data was full-time equivalent employment in community health centers. These could be particularly important resources in a few rural counties classified as “underserved.” The data on county resources and demographic information were used as continuous variables in the model. We also examined selected hospital attributes from the AHA file of the hospitals where patients were treated. These attributes included urban versus rural, teaching status, and number of beds. Except for teaching status, the other major descriptors were subsequently dropped because of collinearity with other variables and low predictive power. Teaching status was indicated by membership in the Council of Teaching Hospitals and was a bivariate categoric variable.

RESULTS Descriptive Results The descriptive data in Table 1 show that compared with blacks and Hispanics, a higher proportion of white New Yorkers admitted for referral-sensitive procedures were from areas with a PCP density higher than the statewide average. However, high PCP–density areas were associated with higher admission rates for blacks, whereas the same was not found to be true for whites or Hispanics. Data showed that black adults from high-PCP areas had a 25 percent higher rate of admissions per 1,000 adult black population of the area than those from low-PCP areas. In contrast, white and Hispanic adults from low PCP–density areas were more likely to have higher referral-sensitive admission rates. Referral-sensitive Admissions Vs. Marker Admissions The results of the logistic regression in Table 2 confirm that, even after adjusting for various patient and area characteristics, increased PCP density in an area was associated with a higher probability of black admissions than white admissions for referral-sensitive procedures compared with the corresponding comparison group (marker admissions). This finding is indicated by the coefficient of interac-

Racial Disparity, Primary Care, and Specialty Referral––73

tion between PCP density and black race (adjusted odds ratio = 2.002, p < .01). This was true even when New York City, with its high minority population, was added as a predictor to the model (adjusted odds ratio = 1.668, p < .05). PCP density, however, was not found to be a significant factor for Hispanic admissions, possibly as a result of the confounding effects of New York City. (While 60 percent of referral-sensitive admissions among blacks were from New York City, the corresponding ratio for Hispanics was 76 percent.) With New York City excluded as a predictor, Hispanics were less likely than whites (p < .05) to be admitted for referral-sensitive procedures, but PCP density was not a significant predictor for these admissions. These findings indicate that PCP density may explain racial differences in referral-sensitive admissions, particularly between blacks and whites. Analysis of data in Table 2 indicates that at the mean PCP level of the county (0.58 per 1,000 relevant population), both blacks and Hispanics were less likely than whites to be hospitalized for referral-sensitive admissions compared to marker admissions. Adjusted odds ratios calculated at the mean PCP level were 0.73 for blacks versus whites, and 0.75 for Hispanics versus whites. However, further analysis (data not shown) showed that an addition of one PCP per 1,000 (a fairly large increase since PCP density would increase by nearly three times) would result in a 102 percent increase in odds of referral-sensitive admissions among blacks, 64 percent among Hispanics, and 36 percent among whites, relative to marker admissions.1 PCP density was not associated with increased referral-sensitive admissions among Medicaid patients relative to the control group. Irrespective of PCP availability, Medicaid patients did not differ significantly from privately insured patients in seeking referral-sensitive versus marker admissions. (A model run without interactions, however, showed that Medicaid patients were significantly more likely to be hospitalized for marker conditions than for referral-sensitive conditions.) Uninsured patients, on the other hand, were significantly less likely to undergo referral-sensitive surgeries, and primary care referral did not cause a significant increase in admissions for these patients as compared to the privately insured patients. HMO enrollees exhibited a higher likelihood of referralsensitive admissions. Other findings were consistent with previous research showing that male patients, aged patients, patients with higher severity, and those with higher income are more likely to be admitted for referral-sensitive procedures than for marker conditions. Specialist density was also found to be a significant factor discriminating between referral-sensitive and marker admissions.2 Among location

74––HSR: Health Services Research 36:6, Part II (December 2001)

variables, New York City showed a significant likelihood for more urgent admissions (odds ratio = 0.719, p < .01).

DISCUSSION Our study findings strongly suggest that PCP supply is associated with reductions in racial disparities in referral patterns for certain high-cost surgical procedures for blacks. Increased PCP supply is associated with a much larger increase in odds of black admissions than white admissions and thus reduces disparity in referralsensitive admissions compared with other admissions. These findings are derived from unadjusted descriptive data as well as data adjusted for various patient, area, and hospital characteristics. Different test runs, for example, those using other measures of PCP density, also confirm that these findings are fairly robust. The findings for blacks and Hispanics are somewhat different. PCP supply had a stronger association with referral-sensitive admissions among blacks compared with Hispanics. However, both blacks and Hispanics, compared with whites, were less frequently admitted for referral-sensitive procedures than for marker conditions. The association between PCP supply and admissions for Hispanics was not significant. On the other hand, admissions among blacks for referral-sensitive procedures were likely to increase substantially in response to increased PCP referrals, which in fact may reverse the utilization bias compared with the control group. A major reason for lack of significant effects among Hispanics could be the relatively small number being admitted for these procedures. A majority of Hispanics were also covered under Medicaid insurance in New York. Medicaid patients were less likely to be affected by PCP supply in this study (see discussion below) and more likely to be admitted for urgent conditions. High PCP supply was not found to be associated with increased referralsensitive admissions for Medicaid or uninsured patients. In fact, separate tests showed that increased PCP supply significantly increased the likelihood of preventable admissions among Medicaid patients compared with referral-sensitive admissions. The finding indicates that a traditionally higher rate of preventable admissions among Medicaid patients may not necessarily stem from a lower PCP supply alone and that receiving a referral may be less of an issue for Medicaid patients than it is for nonwhites. A likely explanation is that patients hospitalized as Medicaid enrollees often are previously uninsured and therefore less likely to seek primary care (Friedman and Basu 2001). The study found HMO enrollees likely to be admitted more frequently for referral-sensitive than for marker condi-

Racial Disparity, Primary Care, and Specialty Referral––75

tions. The finding is in agreement with other studies reporting higher referral rates for patients in HMOs (Forrest and Reid 1997, Franks and Clancy 1997) than for those in indemnity plans. However, in areas with high PCP density, commercially insured patients may be more likely (p < .10) than HMO patients to be admitted for referral-sensitive conditions. A possible hypothesis is that in areas with high primary care density, PCPs may be better able than HMO plans to direct patients for a referral-sensitive admission. This finding further contributes to our understanding of the role of primary care in the referral process. Our study has some limitations that should be acknowledged. First, the analysis was based on an assumption that, on average, most individuals are likely to receive primary care in their county of residence. Second, the results are the product of cross-sectional analyses, so inferences regarding causal relationships are limited. Third, PCP supply as defined here is a proxy for availability of primary care services because there are no reliable data available regarding length of hours, availability of advice after hours, and other factors likely to be important in considering PCP availability.

CONCLUSIONS PCP supply may be very important for admissions for referral-sensitive procedures, especially for blacks. Barriers to primary care access are found to be associated with significantly lower rates of such admissions. Controlling for other factors, increased PCP supply may significantly enhance the referral process for nonwhites. Individuals who receive such procedures can do so by being referred by a PCP or by self-referral to specialists. While our study did not assess whether there are differences associated with race and ethnicity, access to referral-sensitive procedures through a PCP may be particularly important for members of racial and ethnic minority groups. As managed care arrangements continue to offer increased options for direct self-referral to specialists associated with higher copayments, the role of PCP availability may become even more important than in the results indicated here. Previous studies suggest that racial disparity often goes beyond clinical appropriateness and accessibility issues and can be addressed only through tailored interventions (Epstein, Ayanian, Keogh, et al. 2000). This study indicates that increased PCP density may be used as a possible intervention strategy to narrow racial disparity in specialty referrals and to improve the referral process for highcost, high-technology surgical procedures. Areas with higher concentrations of

76––HSR: Health Services Research 36:6, Part II (December 2001)

privately insured nonwhite adults may derive greater benefit from this intervention. Future research should examine whether the relationships presented here are robust in longitudinal studies, whether policy changes that increase the supply of PCPs are associated with reduced disparities in referral-sensitive admissions, and whether the results can be confirmed in other geographic areas.

NOTES 1. These numbers were derived from a separate logistic model in which whites, in addition to blacks and Hispanics, were added as independent variables (other race was the default category). The percentages were calculated relative to the race-specific overall odds of referral-sensitive admissions relative to marker admissions. 2. Because “primary care density” and “specialist density” variables are significantly correlated, we replaced “specialist density” by “proportions of specialists” in test runs for Table 2. The results were fairly comparable, and we therefore retained the density variable. The specialists in Table 2 represent total specialists (medical and surgical). Test runs replacing them with medical specialists yielded similar results. We also ran models using medical specialist density and surgical specialist density as two separate variables. We found that surgical specialists were associated with more referral-sensitive admissions and medical specialists were associated with fewer referral-sensitive admissions compared with marker admissions.

REFERENCES Ayanian, J. Z., P. D. Cleary, J. S. Weissman, and A. M. Epstein. 1999. “The Effect of Patients’ Preferences on Racial Differences in Access to Renal Transplantation.” New England Journal of Medicine 341 (22): 1661–69. Ayanian, J. Z., and A. M. Epstein. 2001. “Racial Disparities in Medical Care.” New England Journal of Medicine 344 (May 10): 1443–49. Basu, J., B. Friedman, and H. Burstin. 2001. “Primary Care, HMO Enrollment and Hospitalization for Ambulatory Care Sensitive Conditions: A New Approach.” Billings, J., L. Zeitel, J. Lukomnik, T. S. Carey, A. E. Blank, and L. Newman. 1993. “Impact of Socioeconomic Status on Hospital Use in New York City.” Health Affairs 12: 162–73. Christoffersson, J. G., J. E. Conklin, and J. Gonnella. 1988. “The Impact of Severity of Illness on Hospital Costs.” DRG Monitor 6 (1): 1–8. Clancy, C. M., and P. Franks. 1997. “Utilization of Specialty and Primary Care: The Impact of HMO Insurance and Patient-related Factors.” Journal of Family Practice 45 (6): 500–508. Coffey, R. M., and M. G. Goldfarb. 1986. “DRGs and Disease Staging for Reimbursing Medicare Patients.” Medical Care 24 (9): 814–29.

Racial Disparity, Primary Care, and Specialty Referral––77

Einbinder, L. C., and K. A. Schulman. 2000. “The Effect of Race on the Referral Process for Invasive Cardiac Procedures.” Medical Care Research and Review 57 (Suppl. 1): 162–80. Epstein, R. M. 1995. “Communication Between Primary Care Physicians and Consultants.” Archives of Family Medicine 4 (5): 403–09. Epstein, A. M., J. Z. Ayanian, J. H. Keogh, S. J. Noonan, N. Armistead, P. D. Cleary, J. S. Weissman, J. A. David-Kasdan, D. Carlson, J. Fuller, D. Marsh, and R. M. Conti. 2000. “Racial Disparities in Access to Renal Transplantation—Clinically Appropriate or Due to Underuse or Overuse?” New England Journal of Medicine 343 (21): 1537–44. Forrest, C. B., G. B. Glade, A. E. Baker, A. Bocian, S. V. Schrader, and B. Starfield. 2000. “Coordination of Specialty Referrals and Physician Satisfaction with Referral Care.” Archives of Pediatric and Adolescent Medicine 154 (5): 499–506. Forrest, C. B., G. B. Glade, B. Starfield, A. E. Baker, M. Kang, and R. J. Reid .1999. “Gatekeeping and Referral of Children and Adolescents to Specialty Care.” Pediatrics 104 (1, Pt 1): 28–34. Forrest, C. B., and R. J. Reid. 1997. “Passing the Baton: HMOs’ Influence on Referrals to Specialty Care.” Health Affairs (Millwood) 16 (6):157–62. Franks, P., and C. M. Clancy. 1997. “Referrals of Adult Patients from Primary Care: Demographic Disparities and Their Relationship to HMO Insurance.” Journal of Family Practice 45 (1): 47–53. Franks, P., C. M. Clancy, and P. A. Nutting. 1992. “Gatekeeping Revisited—Protecting Patients from Overtreatment.” New England Journal of Medicine 327 (6): 424–29. Franks, P., P. A. Nutting, and C. M. Clancy. 1993. “Health Care Reform, Primary Care, and the Need for Research.” Journal of the American Medical Association 270 (12): 1449–53. Friedman, B., and J. Basu. 2001. “Health Insurance, Primary Care and Preventable Hospitalization of Children in a Large State.” American Journal of Managed Care 7 (5): 473–81. Gonnella, J. S., M. C. Hornbrook, and D. Z. Louis. 1984. “Staging of Disease: A Case-mix Measurement.” Journal of the American Medical Association 251 (5): 637–44. Horner, R. D., E. Z. Oddone, and D. B. Matchar. 1995. “Theories Explaining Racial Differences in the Utilization of Diagnostic and Therapeutic Procedures for Cerebrovascular Disease. Milbank Quarterly 73 (3): 443–62. King, T. E., Jr., and P. Brunetta. 1999. “Racial Disparity in Rates of Surgery for Lung Cancer.” New England Journal of Medicine 341 (16): 1231–33. McBean, A. M., and M. Gornick. 1994. “Differences by Race in the Rates of Procedures Performed in Hospitals for Medicare Beneficiaries.” Health Care Financing Review 5 (4): 77–90. National Center of Health Statistics. 1999. “U.S. NCHS Vital and Health Statistics.” Series 13, No. 142. Shea, D., B. Stuart, J. Vasey, and S. Nag. 1999. “Medicare Physician Referral Patterns.” Health Services Research 34 (1): 331–48. Sheifer, S. E, J. J. Escarce, and K. A. Schulman. 2000. “Race and Sex Differences in the Management of Coronary Artery Disease.” American Heart Journal 139 (5): 848–57.