Few data are available on how race/ethnicity, insurance, and socioeconomic status (SES) interrelate to influence breast cancer treatment. The authors examined care for a national cohort of breast cancer patients to assess whether insurance and SES were associated with racial/ethnic differences in care.
The authors used multivariate logistic regression to assess the probability of definitive locoregional therapy, hormone receptor testing, and adjuvant systemic therapy among 662,117 white, black, and Hispanic women diagnosed with invasive breast cancer during 1998-2005 at National Cancer Data Base hospitals. In additional models, the authors included insurance and area-level SES to determine whether these variables were associated with observed racial/ethnic disparities.
Most women were white (86%), 10% were black, and 4% were Hispanic. Most had private insurance (51%) or Medicare (41%). Among eligible patients, 80.0% (stage I/II) had definitive locoregional therapy, 98.5% (stage I-IV) had hormone receptor testing, and 53.1% and 50.2% (stage I-III) received adjuvant hormonal therapy and chemotherapy, respectively. After adjustment, black (vs white) women had less definitive locoregional therapy (odds ratio [OR], 0.91; 95% confidence interval [CI], 0.88-0.94), hormonal therapy (OR, 0.90; 95% CI, 0.87-0.93), and chemotherapy (OR, 0.87; 95% CI, 0.84-0.91). Hispanic (vs white) women were also less likely to receive hormonal therapy. Hormone receptor testing did not differ by race/ethnicity. Racial disparities persisted despite adjusting for insurance and SES.
Standard guidelines recommend a series of local and systemic treatments for women with breast cancer. These recommendations include treatment with either mastectomy or breast-conserving surgery (BCS) with whole breast radiation in women with stage I or II tumors1, 2 and postmastectomy radiation (PMRT) for large tumors or when >3 lymph nodes are involved.2, 3 Guidelines also recommend that women with endocrine-sensitive disease (defined as estrogen receptor [ER] or progesterone receptor [PR] positive) receive hormonal therapy (particularly for tumors ≥1 cm) and those with tumors >1 cm in size or with lymph node involvement be considered for adjuvant chemotherapy.2 Regardless of stage, ER/PR testing is required for all new cancers to guide selection of effective systemic treatments.2
Previous studies have demonstrated less recommended breast cancer care for black and Hispanic women versus white women, including lower rates of radiation after BCS and lower rates of systemic therapy.4-8 Black women also experience worse breast cancer outcomes even when matched for known prognostic features, and mortality differences by race have widened over time.9 Many factors contribute to these differences, including differences in receipt of beneficial treatments. Although socioeconomic status (SES) and insurance have been hypothesized to explain observed racial disparities in care, data are limited on the independent contributions of these factors to the receipt of appropriate breast cancer treatments. Large-scale studies using cancer registry data have not typically included insurance and detailed treatment information, and studies using Surveillance, Epidemiology and End Results (SEER)-Medicare data include only older patients insured by Medicare. Understanding the impact of insurance on receipt of care is particularly relevant given recent legislation to expand health insurance in the United States.
In this study, we examined receipt of stage-appropriate breast cancer care by race/ethnicity in a large national sample of women with breast cancer, and we assessed whether insurance and area-level SES were associated with any observed racial/ethnic differences.
MATERIALS AND METHODS
Data Source and Study Population
We used National Cancer Data Base (NCDB) registry data from 1998 to 2005 to examine receipt of appropriate diagnostic testing and treatments for breast cancer by race/ethnicity. The NCDB is a national, hospital-based, cancer registry sponsored by the American College of Surgeons' Commission on Cancer (CoC) and the American Cancer Society (ACS). Since 1989, data have been uniformly collected on patient demographics, tumor characteristics, first course of treatment, and outcomes for cancer patients treated at US hospitals. Currently, more than 1400 CoC-approved hospitals submit data to the NCDB, representing 75% of all newly diagnosed cancer cases.10-13 Since 1998, the NCDB has also collected insurance information. Although the NCDB is a hospital-based registry, the registrars record all available diagnostic and treatment information from patients' inpatient and outpatient medical records at the time of case abstraction. Registrars routinely update the abstract as appropriate, once they have identified missing information. This often includes continued monitoring with letters and telephone calls to treating facilities and providers.
We included all women aged ≥18 years with a first diagnosis of breast cancer during the period 1998-2005 (n = 1,420,471). To maximize ascertainment of treatments, we focused on patients whose cancers were diagnosed at the reporting institution and received all or part of their first course of treatment at the same facility (n = 966,976). Noninvasive breast cancers (n = 198,388), those treated in Puerto Rico (n = 162), and those without histologies likely to be treated by standard breast cancer guidelines (n = 7861) were excluded. Next, we excluded patients with miscoded primary tumor site (n = 31), missing zip code (n = 686), missing histology (n = 45), missing both tumor size and stage (n = 9565), and cases reported by Veterans Affairs Medical Centers (excluded from the data use agreement) (n = 764). Lastly, we restricted the sample to white, black, and Hispanic women (n = 662,117) because other racial/ethnic subgroups were small, and disparities have been primarily documented for black and Hispanic women.
Dependent variables included definitive locoregional therapy, hormone receptor testing, adjuvant hormonal therapy, and adjuvant chemotherapy (Table 1).
Table 1. Selected Cohorts for Outcomes of Interest
AJCC clinical then pathological stage.
AJCC pathological then clinical stage.
Definitive locoregional therapy
Stagea I, II who underwent cancer-directed surgery (1998-2005); n=550,453
Stageb I, II, III with tumors ≥1 cm or with node-positive disease; All women underwent cancer-directed surgery and had estrogen receptor or progesterone receptor-positive tumors (1998-2005); n=238,745
Stageb I, II, III who underwent cancer-directed surgery and had tumors >1 cm or node-positive disease (1998-2005); n= 410,124
Definitive locoregional therapy
We assessed definitive locoregional therapy in women with stage I/II breast cancers, defined as receipt of: 1) BCS with radiation, 2) PMRT in those with tumor size >5 cm or with >3 positive lymph nodes,2 or 3) mastectomy with or without PMRT women who did not meet strict criteria for PMRT. Pathologic margin status was not included as a criterion for PMRT because detailed data were not available. Mastectomy included mastectomies characterized as modified radical, radical, extended radical, subcutaneous, total, or not otherwise specified (NOS). BCS included partial mastectomy NOS, less than total mastectomy NOS, lumpectomy, and segmental mastectomy. Radiation was documented by the registrars; women were categorized as not receiving radiation when radiation was recommended but not received, unknown if received, refused by the patient, or unknown when recommended or received (n = 100,601). We excluded women who underwent mastectomy but had unknown indications for PMRT (unknown tumor size or unspecified number of positive nodes) (n = 1017).
Hormone receptor testing
We defined receipt of hormone receptor testing among women with stage I-IV cancers as documentation of ER and PR status, regardless of testing results. Unknown or incomplete ER or PR testing was defined as not receiving hormone receptor testing (n = 7539). Before 2004, data were missing for >20% patients; thus, we restricted these analyses to patients with cancers diagnosed in 2004-2005 (367,543 women excluded).
Adjuvant hormonal therapy
We assessed initiation of hormone therapy among women with hormone receptor-positive (ER- or PR-positive) stage I-III cancers with tumors ≥1 cm or lymph node-positive disease. Women with missing information on hormonal therapy were excluded (n = 34,790).
We assessed chemotherapy among women with pathologic stage I-III cancers who underwent primary surgery and had tumors >1 cm or lymph node-positive disease. Women with missing information on chemotherapy were excluded (n = 46,119 of the initial cohort of 662,117 patients).
Independent Variables of Interest
Independent variables of interest included race/ethnicity, insurance at initial diagnosis and/or treatment, and SES. Race/ethnicity was categorized as non-Hispanic white, non-Hispanic black, or Hispanic. Insurance was categorized as private (health maintenance organization, preferred provider organizations, managed-care NOS, Tricare, military, and insured NOS), uninsured (not insured-NOS, charity write-off, self-pay), Medicaid, Medicare (including Medicare with supplements, categorized as beneficiaries aged ≤64 and those aged ≥65), and unknown.14 Area-level SES (median household income and percentage without high school diploma) was ascertained by linking patient zip codes with 1990 US Census Bureau data and was categorized into quartiles.
Control variables differed slightly for each outcome of interest and were chosen a priori on the basis of clinical relevance or known associations with receipt of therapy. All models included age, histology, grade, hospital type,15 diagnosis year, US census division,16 treatment facility volume, and American Joint Commission on Cancer (AJCC) stage (clinical or pathologic as appropriate). We also included tumor size to further distinguish risk within stage. We coded stage using the AJCC fifth edition17 for diagnoses during 1998-2002 and the AJCC sixth edition18 for diagnoses during 2003-2005. Models for definitive locoregional therapy and chemotherapy also included ER/PR status. Models for chemotherapy and hormonal therapy included surgical and radiation treatments received. Variables were categorized as shown in Table 2.
Table 2. Patient Characteristics by Race/Ethnicitya
ER indicates estrogen receptor; PR, progesterone receptor.
P < .0001 (for each characteristic analyzed by race/ethnicity) for all categories using chi-square testing.
American Joint Commission on Cancer (AJCC) clinical stage augmented by pathological stage when not recorded.
Median income in zip code of residence ($, quartiles)
Year of diagnosis
West North Central
East North Central
West South Central
East South Central
Comprehensive cancer center
Hospital volume (tertiles)
We compared rates of definitive locoregional therapy, hormone receptor testing, adjuvant hormonal therapy, and adjuvant chemotherapy by race/ethnicity by using chi-square tests and repeated these tests stratified by insurance. We then used multivariate logistic regression to assess the probability of each dependent variable, adjusting for the control variables described above. In a second set of logistic regression models, we added insurance and SES to assess whether the expected association of race with treatment was sensitive to inclusion of these variables. We used generalized estimating equations for all models to account for clustering at the hospital level.
We performed several sensitivity analyses to address missing data for the dependent variables. For definitive locoregional therapy, we first excluded women for whom radiation was recommended but might not have been received (n = 10,226) and those with unknown recommendation or receipt of radiation (n = 55,014). Second, we included these women after redefining these variables as having received radiation. For the hormone receptor testing and adjuvant hormonal therapy outcomes, we repeated all models coding women with unknown testing status or unknown hormonal therapy status as not having received these procedures (n = 7539 and n = 34,790, respectively). Finally, for the chemotherapy models, we repeated all analyses restricting to stage II/III tumors (n = 256,642) because practice guidelines often allow avoidance of chemotherapy in early stage, endocrine-sensitive tumors.
All statistical analyses were performed with SAS version 9.2 (SAS Institute, Cary, NC). Because this study used pre-existing data with no identifying patient information, the Partners Human Research Committee granted exemption from review.
The cohort of potentially eligible patients included 662,117 women. Patient, tumor, and hospital characteristics by race/ethnicity are shown in Table 2. Compared with white women, black and Hispanic women with cancer were younger, were more often uninsured or insured by Medicaid, and had more advanced stage at presentation. Black and Hispanic women were more likely than white women to have poorly differentiated, larger, and ER-negative tumors, and to live in areas with lower high school graduation rates and lower median incomes.
Rates of receipt of definitive locoregional therapy, hormone receptor testing, adjuvant hormonal therapy, and adjuvant chemotherapy among eligible patients were 80.0%, 98.5%, 53.1%, and 50.2%, respectively. Unadjusted rates of each outcome by race/ethnicity are shown in Table 3. Compared with white women, black and Hispanic women had lower rates of receipt of definitive locoregional therapy and adjuvant hormonal therapy and higher rates of chemotherapy. All groups had high rates of hormone receptor testing.
Table 3. Unadjusted Rates (%) for Each Outcome of Interest by Race/Ethnicitya
Outcome of Interest
Chi-square testing was used to examine differences in treatment rates by race/ethnicity; P < .0001 for all.
Definitive locoregional therapy, n=550,453
Hormone receptor testing, n=139,203
Adjuvant hormonal therapy, n=238,745
Adjuvant chemotherapy, n=410,124
After stratifying by insurance, white women had higher rates of definitive locoregional therapy and hormonal therapy and lower rates of chemotherapy than black and Hispanic women (Fig. 1). Racial/ethnic differences persisted regardless of insurance coverage. For example, privately insured black women had lower rates of definitive locoregional therapy (76.4%) than white women who were uninsured (77.7%) or insured by Medicaid (80.4%). Similar findings were noted for receipt of adjuvant hormonal therapy.
In adjusted analyses (Table 4), among women with stage I/II breast cancer, black women were less likely than white women to receive definitive locoregional therapy (odds ratio [OR], 0.91; 95% confidence interval [CI], 0.88-0.94). In the model that included insurance and SES, uninsured patients, those with Medicaid, younger Medicare beneficiaries, and those with unknown insurance were less likely to undergo definitive locoregional therapy than privately insured women. Area-level measures of income and education were not associated with receipt of definitive locoregional therapy. After adjustment for SES and insurance, the findings for black women were slightly attenuated (OR, 0.93; 95% CI, 0.89-0.96).
Table 4. Odds Ratios and Confidence Intervals for Receipt of Diagnostic Testing or Therapya
Hormone receptor testing: adjusted for same variables above with exception of ER and PR status.
Adjuvant hormonal therapy: adjusted for same variables in addition to type of surgery performed (mastectomy or breast-conserving surgery) and radiation (yes/no). ER and PR status information was not included.
Adjuvant chemotherapy: adjusted for same variables as adjuvant hormonal therapy models but ER and PR status were included.
Model 2: Race, Variables, SES, and Insurance Characteristics
Medicare (age ≤64 y)
Medicare (age ≥65 y)
% Without high school diploma (quartiles)
Median income in zip code of residence ($, quartiles)
During 2004-2005, hormone receptor testing rates were high (>97%) for all women and did not differ by race/ethnicity (Table 4). After adjustment for insurance and SES, Hispanic women had higher odds of being tested than white women (OR, 1.25; 95% CI, 1.01-1.54).
In adjusted analyses, black and Hispanic women were less likely to receive hormonal therapy than white women (OR, 0.90; 95% CI, 0.87-0.93; and OR, 0.94; 95% CI 0.89-0.98) (Table 4). When insurance and SES variables were included in the models, the odds ratios remained similar, but the Hispanic race effect was no longer statistically significant. Patients who were uninsured, covered by Medicaid, covered by Medicare and ≤64 years of age, or had unknown insurance had lower odds of receiving adjuvant hormonal therapy than privately insured patients. Area-level SES was not associated with adjuvant hormonal therapy.
Black, but not Hispanic, women also had lower adjusted odds of receiving adjuvant chemotherapy than white women (OR, 0.87; 95% CI, 0.84-0.91). These lower odds were slightly attenuated when the insurance and SES variables were included in the model (OR, 0.92; 95% CI, 0.88-0.95). Lack of insurance, Medicaid, Medicare (both age groups), and unknown insurance were all associated with lower odds of chemotherapy (vs privately insured patients). Area-level SES was not associated with receipt of adjuvant chemotherapy (Table 4).
Results of various analyses assessing the sensitivity of findings to different means of coding missing treatment information were similar for all outcomes examined with the exception of hormone receptor testing, where findings were no longer significant for Hispanic women (data not shown). In addition, results of the adjuvant chemotherapy analysis restricting the cohort to patients with stage II/III cancer were similar to the analysis of women with stage I-III disease.
We examined guideline-recommended breast cancer diagnostic procedures and treatments by race/ethnicity in a large, diverse sample of women to assess whether insurance and area-level SES were associated with any racial/ethnic differences. Compared with white women, we observed modest differences in receipt of definitive locoregional therapy, chemotherapy, and hormonal therapy for black women and hormonal therapy for Hispanic women. We observed no association of area-level SES with any of the treatments studied, but we observed lower odds of receiving definitive locoregional therapy and adjuvant systemic treatments for uninsured women, Medicaid enrollees, and younger Medicare beneficiaries (vs privately insured patients). In addition, Medicare beneficiaries aged ≥65 years had lower odds of adjuvant chemotherapy receipt. Although insurance status was an important independent predictor of lower rates of guideline therapy, only slight attenuations in the odds of locoregional definitive therapy and chemotherapy were observed for black women after addition of insurance and SES to the models.
Racial disparities in receipt of treatments for breast cancer have been well documented,4-6, 19-21 but the reasons for these differences remain uncertain. We have attempted to further understand these disparities in the context of a diverse, generalizable patient sample. Although previous studies have examined the individual effects of race, SES, or insurance on the receipt of treatments, to our knowledge, this analysis is the first to include all of these variables in a large-scale, national study. In addition, we were able to control for treating-facility characteristics, such as hospital type and volume. Studies using SEER-Medicare data have demonstrated racial disparities in receipt of recommended breast cancer treatments and some variations based on area-level income,6, 21, 22 but these studies included only older women with Medicare insurance. Other studies have also reported lower rates of recommended care for younger Medicare beneficiaries (who qualify for coverage because of disability or end-stage renal disease).23, 24 Comorbid illness and poor tolerance of therapy may contribute to these differences.
A recent study examined care for more than 3000 women insured by a Southeastern US health plan and observed that black women had lower rates of hormonal therapy than white women despite uniform insurance coverage.25 In addition, 2 large meta-analyses demonstrated that African American ethnicity was an independent predictor of higher mortality even after adjustment for SES, although these analyses did not examine contributions of insurance or the receipt of specific breast cancer treatments.26, 27 Other analyses have examined the impact of insurance on receipt of recommended breast cancer care. A study of breast cancer care in New York City found that black or Hispanic race/ethnicity and lack of insurance were significantly associated with underutilization of adjuvant breast cancer care; the study did not examine SES and included 6 local hospitals only.4
The modest attenuation of disparities we observed, after accounting for differences in insurance status, suggests that simply providing access to insurance is unlikely to have a major impact on breast cancer disparities by race/ethnicity. A better understanding of the factors other than insurance and SES that contribute to racial disparities is essential. Breast cancer care has become increasingly complex, with multiple providers and office visits required to complete recommended adjuvant therapy. Patient factors, such as difficulty accessing care even with insurance,28 mistrust of the medical system,28-30 and differing preferences may be important.30, 31 Disparities in care are also influenced by the characteristics of physicians32 and hospitals33 from which patients obtain care.
Once breast cancer is diagnosed, patients must navigate through surgical, radiation oncology, and medical oncology visits and must carefully consider adjuvant treatment options. Decisions are often time-intensive, complicated, and nuanced, and effective communication is crucial.30, 31 Moreover, such complex care may be particularly challenging for patients with limited social support, superseding life stressors, or medical comorbidities.4, 8 Although we measured area-level SES, this is an imperfect measure of individual SES, and lower levels of education and financial resources may also contribute to disparities in care. In addition, we could not examine the reasons therapies were not used nor could we adjust for potential unobserved confounders because we lacked information on comorbidity, treatment preferences, access to care, and changes in insurance over time. Nevertheless, research suggests very high rates of continuous enrollment among patients covered by Medicaid at the time of a cancer diagnosis.34
Considering the high rates of hormone receptor testing, the lack of disparities we observed during the years 2004-2005 is reassuring. Prior studies reported lower ER documentation rates for black and Hispanic women living in SEER areas with breast cancer diagnosed during 1992-2002.21, 35 Our findings may be due to improved documentation in recent years or the increased clinical emphasis on hormone receptor status information because of its implications for treatment selection.
Our study has several limitations. First, hospital registry data may not fully capture systemic therapy administered in the outpatient setting. However, the NCDB strives to maximize the collection of inpatient and outpatient primary treatment and updates information within 5 years of initial case abstraction. In addition, a previous analysis comparing cancer registry data with medical record data demonstrated lower sensitivity of registry data for ascertaining chemotherapy and hormonal therapy administration, but there was high agreement between data sources among patients identified as having had chemotherapy.36 Although differential ascertainment of chemotherapy could influence our results, our findings were robust to several sensitivity analyses.
Second, we used strict criteria for adjuvant chemotherapy eligibility (tumors >1 cm or lymph node-positive disease). Although current guidelines for adjuvant chemotherapy consider chemotherapy optional for lower risk, ER-positive, breast cancers, treatment guidelines during the study period recommended chemotherapy for all lymph node-positive cancers or tumors >1 cm.37 Furthermore, our findings were similar when restricted to stage II/III patients. We were unable to examine whether standard chemotherapy agents were administered because the NCDB does not collect this information.
A third limitation is that the findings may not be generalizable to non-CoC-approved hospitals. Although the NCDB includes data for greater than 75% of new cancer cases in the United States annually, CoC-approved hospitals are more often larger, urban centers that provide more cancer related services (including cancer screening, chemotherapy, and radiation) than non-CoC-approved hospitals.38 Future research should focus on potential disparities at the institutional level. Finally, the absolute differences we observed in receipt of treatment by race/ethnicity were relatively modest; however, because breast cancer is so frequently diagnosed, small differences affect large numbers of women.
In summary, despite efforts to eliminate disparities in cancer care in recent years, our study of a large, hospital-based sample suggests that modest racial differences in receipt of recommended breast cancer care persist even after adjustment for insurance and area-level SES. Although health insurance expansion may resolve disparities in treatment by health insurance status, this study suggests that expansion of insurance coverage alone is unlikely to have a major impact on disparities in breast cancer care among black women. Multifaceted efforts are needed to ensure receipt of effective treatments and to improve outcomes in women with potentially curable disease.
The authors thank Michael Halpern for his thoughtful input on the study concept and the National Cancer Data Base staff for the development, maintenance, and improvement of this registry.
CONFLICT OF INTEREST DISCLOSURES
Research support was provided by Susan G. Komen for the Cure.