|Home | About | Journals | Submit | Contact Us | Français|
The reasons for race/ethnicity (R/E) differences in breast cancer survival have been difficult to disentangle.
Surveillance, Epidemiology, and End Results (SEER)-Medicare data were used to identify 41,020 women aged ≥68 years with incident breast cancer between 1994–1999 including African American (2479), Hispanic (1172), Asian/ Pacific Island (1086), and white women (35,878). A Cox proportional hazards model assessed overall and stage-specific (0/I, II/III, and IV) R/E differences in breast cancer survival after adjusting for mammography screening, tumor characteristics at diagnosis, biologic markers, treatment, comorbidity, and demographics.
African American women had worse survival than white women, although controlling for predictor variables reduced this difference among all stage breast cancer (hazards ratio [HR], 1.08; 95% confidence interval [95% CI], 0.97–1.20). Adjustment for predictors reduced, but did not eliminate, disparities in the analysis limited to women diagnosed with stage II/III disease (HR, 1.30; 95% CI, 1.10–1.54). Screening mammography, tumor characteristics at diagnosis, biologic markers, and treatment each produced a similar reduction in HRs for women with stage II/III cancers. Asian and Pacific Island women had better survival than white women before and after accounting for all predictors (adjusted all stages HR, 0.61 [95% CI, 0.47–0.79]; adjusted stage II/III HR, 0.61 [95% CI, 0.47–0.79]). Hispanic women had better survival than white women in all and stage II/III analysis (all stage HR, 0.88; 95% CI, 0.75–1.04) and stage II/III analysis (HR, 0.88; 95% CI, 0.75–1.04), although these findings did not reach statistical significance. There was no significant difference in survival by R/E noted among women diagnosed with stage IV disease.
Predictor variables contribute to, but do not fully explain, R/E differences in breast cancer survival for elderly American women. Future analyses should further investigate the role of biology, demographics, and disparities in quality of care.
Breast cancer is the most common cancer and the second leading cause of cancer death among women in the U.S.1 Although there has been an overall reduction in breast cancer mortality rates in the U.S. since the 1990s, most of this benefit has been experienced by white women.2 African American (AA) and Hispanic (H) women remain more likely to be diagnosed with poor prognostic breast cancers (ie, late stage, large size, lymph node-positive, estrogen receptor-negative) with AA women experiencing worse survival than white women, a disparity that has increased rather than decreased over time.3 In contrast, Asian/Pacific Island (A/PI) women tend to have better prognostic breast cancers (ie, early stage, small size, lymph node-negative, estrogen receptor-positive) and better survival than white women.4
The reasons for the persistence of these racial and ethnic disparities have been difficult to disentangle. Possible explanations include differences in screening mammography leading to differences in the stage and size of tumors at diagnosis,5,6 tumor biology, inadequate receipt of appropriate breast cancer treatment,5 and underlying patient comorbidities and socioeconomic factors.7–11
Several studies have used Surveillance, Epidemiology, and End Results (SEER) program information to explore these issues.3,4,12–18 Although SEER data include valuable regional information on a large, geographically diverse population, it lacks detailed information concerning screening mammography use and underlying comorbidities that may impact survival.19 This study aims to overcome these problems by using the combined SEER-Medicare dataset to explore the contribution of screening, tumor characteristics, biology, treatment, comorbidities, and demographics to race/ethnicity (R/E) differences in breast cancer survival for elderly women.
Data were obtained from the SEER-Medicare database that combines clinical information with claims information from Medicare, the primary health insurer for 97% of the U.S. population aged ≥65 years.19 SEER cancer registry data includes information on cancer characteristics and treatment from 11 SEER sites representing ~14% of the U.S. population.19 Medicare data include information that allow assessment of measurements not possible in SEER data alone, including mammography utilization, comorbidities, and treatment.
Our study population included all female Medicare beneficiaries diagnosed with incident breast cancer between January 1, 1994, and December 31, 1999. All histologic subtypes were included in this cohort. We limited the study to women with at least 3 years of Medicare enrollment before their cancer diagnosis to assess the use of screening mammography and presence of comorbidities, thus restricting the sample to women aged ≥68 years. Medicare does not receive claims for women enrolled in health maintenance organization (HMO) plans, making it impossible to assess the predictor variables in these women. Therefore, we restricted the sample to women enrolled in Medicare's Fee For Service plans.19 Similarly, mammography coverage is included as part of Part B enrollment and women without Part B enrollment were excluded as we could not assess mammography screening, outpatient treatment, or comorbidity status.19 Overall, 17,364 women were excluded because they had either HMO enrollment or Part B nonenrollment. A total of 41,020 women met inclusion criteria and were included in the analysis.
Age was calculated using SEER data and grouped into 5 categories (68–70 years, 71–75 years, 76–80 years, 81–85 years, and >85 years). R/E was classified according to SEER data as African American, Hispanic, Asian/Pacific Island, Other/Unknown, and white. The SEER ethnicity variable used to identify H women is based on a surname-matching algorithm that has greater sensitivity than ethnicity recorded in Medicare data.20 Race in SEER data are determined from medical records and registration information. The A/PI category includes Chinese, Japanese, Filipino, and Hawaiian women. The Other/Unknown category includes Native American data that were not isolated due to small numbers.
Each woman was characterized based on her mammography utilization in the 3 years before her breast cancer diagnosis.21 Because most women with a new diagnosis of breast cancer will undergo mammography around the time of diagnosis, either as a screening examination or as evaluation of known or suspected cancer, we characterized a woman's screening history based on the timing of mammography before this pericancer mammogram.21 The intervals were categorized as within 1 year, 1 to 2 years, 2 to 3 years, and >3 years (including women who had no screening mammography before diagnosis). Adequate screening was defined as screening received ≤2 years and inadequate screening as screening received >2 years before the pericancer mammography, or no screening before diagnosis.
Tumor characteristics were determined using SEER data and included American Joint Committee on Cancer (AJCC) categories describing cancer stage (0–IV, tumor size [mm], and lymph node status [positive vs negative/unknown]).
We assessed estrogen receptor status using SEER data as 1) positive versus 2) negative, 3) borderline, 4) unknown, and 5) none done.
Histologic grade was characterized using SEER data as per the AJCC categories as grade 1 through 4.22
Treatment receipt was determined using both SEER and Medicare data.5,23–26 In cases of disagreement on surgery type, the most invasive surgery, ie, mastectomy as opposed to lumpectomy, noted in either dataset was used. Categories included 1) breast-conserving surgery without radiation, 2) breast-conserving surgery with radiation, 3) mastectomy, and 4) no surgery.
R/E differences in comorbidities may produce disparities in the use of mammography screening and treatment.27 Comorbidity was measured with the Charlson comorbidity index derived from Medicare inpatient and physician/supplier claims.28
Income was used as a marker of socioeconomic status (SES) and obtained from SEER data using the median income of residence as recorded by the ZIP code. The type of community a woman lives in may influence her access to breast cancer services and was determined via SEER data using the assigned metropolitan statistical area of 1) rural, 2) less rural, 3) urban, 4) metropolitan, and 5) big metropolitan.29
A Cox proportional hazards model was used to determine time from breast cancer diagnosis to cancer-specific death among all women with breast cancer and stratified by stage at diagnosis (0/I, II/III, and IV). These groups were created because we expected to see similar outcomes among women with stage 0/I and stage IV disease and different outcomes for women with stage II/III breast cancer (ie, excellent prognosis with early stage, poor prognosis with late stage disease).
The base model controlled for age and SEER site. Additional variables were sequentially added to increasingly adjusted models, including utilization of screening mammography (adequate vs inadequate); tumor characteristics including size (≤15 mm vs ≥16 mm), lymph node status (positive vs negative/ unknown), stage (0/I vs II/III/IV, except in the stage-specific models), biological measures including grade (1/2 vs 3/4) and estrogen receptor status (positive vs negative), breast cancer treatment (adequate vs inadequate), comorbidities, and demographic variables (income/community type).
Because the size cutoff between tumor stages I and II is 20 mm, we categorized tumor size as >15 mm or <15 mm so as to avoid collinearity with the tumor stage variable and present unique clinical information into the models that may be prognostically significant. Breast conservation without radiation and no surgery were defined as ‘inadequate treatment.’ Breast conservation with radiotherapy and mastectomy were defined as ‘adequate treatment’ based on clinical trial data demonstrating similar long-term survival rates.30
We investigated the interaction between 1) screening and age (age ≤75 years vs >75 years), as screening for very elderly women is not routinely recommended and the survival impact may vary by age; 2) screening and stage, as the impact of screening on survival could matter differently for different stage disease; and 3) stage and radiation to account for the finding that radiotherapy for early stage 0/I is not indicated. These potential interactions were included in the model to account for differences in 1) the use of mammography screening by age and stage and 2) appropriate use of radiation by stage.
Women with unknown R/E were included in the statistical modeling. All analyses were conducted using SAS software (version 8.2; SAS Institute, Inc, Cary, NC).
A total of 41,020 women were included: 6% (n = 2479) were AA, 2.7% (n = 1086) were A/PI, 2.9% (n = 1172) were H, and 87.5% (n = 35,878) were white W (Table 1). In all, 40.1% of the women were aged ≤75 years, with A/PI (54.1%) and H women (46.4%) having a greater proportion of younger women compared with AA (41.4%) and white women (40.0%). AA and H women had a greater proportion of women living in areas with income levels <$30,000 (47% and 21%, respectively) compared with A/PI and white women (5% each).
Overall, greater than half of this cohort (58%) received inadequate mammography screening, with AA (64%) and H (68%) women being more likely to be inadequately screened than A/PI (57%) and white (57%) women. AA and H women were also more likely to have poor prognostic tumors at the time of diagnosis.
AA women had the highest proportion of women who received no surgical treatment (11.5% vs 8.2% for H, 6.8% for white and 3.3% for A/PI) or who received breast-conserving surgery without radiation (19.5% vs 15.8/% for white , 11.9% for H, and 11.9% for A/PI).
Hazards ratios (HRs) for cancer-specific death are provided in Table 2. The baseline model adjusts for age and SEER site only, whereas the fully adjusted model adjusts for mammography utilization, tumor characteristics, tumor biology, treatment type, comorbidities, and demographics. Results are presented for all stages of disease and then stratified by stage (0/I, II/III, and IV).
For all stages, AA women had a significantly higher risk of death than white women at baseline (HR, 1.63; 95% confidence interval [95% CI], 1.48–1.80); however, this was considerably reduced, and no longer significant (HR, 1.08; 95% CI, 0.97–1.20) after full adjustment. H women had a significantly higher risk of death at baseline (HR, 1.24; 95% CI, 1.06–1.46), yet after full adjustment these women had a nonsignificant 12% lower risk of death than white women (HR, 0.88; 95% CI, 0.75–1.04). A/PI women had a 41% significantly lower risk of death than white women at baseline (HR, 0.59; 95% CI, 0.35–0.77), a finding that changed very little after full adjustment (HR, 0.61; 95% CI, 0.47–0.79).
In the fully adjusted models for stage 0/I, only A/PI women had a statistically significant different risk of death than white women (HR, 0.44; 95% CI, 0.24–0.81) with no significant differences for all other groups (AA: HR, 1.19 [95% CI, 0.91–1.55]; and H, HR, 0.85 [95% CI, 0.57–1.27]).
For stage II/III, AA women had a 66% higher risk of mortality than white women at baseline (HR, 1.66; 95% CI, 1.43–1.93). Although controlling for all variables reduced the mortality risk substantially, it did not eliminate the difference, leaving a 30% increased risk after full adjustment (HR, 1.30; 95% CI, 1.10–1.54). H women had a nonsignificant similar risk of death at baseline (HR, 1.07; 95% CI, 0.84–1.36) that reduced to a 10% nonsignificant lower risk with full adjustment (HR, 0.90; 95% CI, 0.71–1.15). A/PI women had a 37% lower risk of mortality than white women after full adjustment (HR, 0.63; 95% CI, 0.43–0.93), showing little change from baseline (HR, 0.69; 95% CI, 0.46–1.10).
There were no statistically significant differences by R/E for stage IV results.
Tumor severity accounted for approximately one-third of the overall mortality reduction from baseline to full adjustment (or 29% reduction from Model B to C) between AA and white women in the all stages analysis, followed by screening (14%), treatment (5%), biology (3%), comorbidities (2%), and demographics (2%) (Fig. 1). For H women, screening (16%) and tumor severity (15%) accounted approximately one-third of the difference.
For the stage II/III results, screening mammography (7%), tumor severity (6%), biology (9%), and breast cancer treatment (8%) all produced a similar-sized reduction in the mortality difference for AA women (Fig. 2). A similar pattern was observed for H women.
The order the variables were placed in the model would have affected the percentage reduction in hazard ratio attributable to each variable. However, when we made changes in the order of the variables within the model, we saw relatively little difference in the magnitude of effect.
This study found large differences by R/E in breast cancer survival for elderly women. Screening mammography, tumor severity, biology, treatment, comorbidities, and demographics all contributed to these differences. Controlling for these variables reduced nearly all of the differences between AA and white women in the all stages analysis and reduced, but did not eliminate, disparities in the stage II/III analysis.
In contrast, A/PI women had a consistently better survival profile than white women in all analyses that remained with the addition of predictor variables. This may be because there were no large differences between A/PI women and white women with regard to the variables measured. No statistically significant differences in survival were observed between H and white women, although the overall pattern demonstrated that H women have better survival than white women after full adjustment in both all stages and stage II/III analyses.
All women diagnosed with stage IV breast cancers had a similar risk of death regardless of R/E. This likely reflects the poor prognosis and small number of women in this group.
Screening mammography is known to reduce mortality from breast cancer in the general U.S. population31 with disparities in screening mammography heavily contributing to R/E disparities in breast cancer survival.6,32,33 Our findings show persistent differences by R/E in the utilization of screening mammography. These results are consistent with recent reports21,34 and may be due to persistent underutilization of mammography in this elderly population.
The findings of the current study demonstrate that, although the use of adequate screening will reduce differences in survival between AA and white women, it will not eliminate them. Despite not being the sole remedy, screening nevertheless remains important. It is possible that if H women had greater utilization of adequate screening, then their risk of breast cancer death could drop further below white women, as seen for A/PI women. AA differences in screening interval accounted for a considerable portion of the mortality difference with white women. Addressing this disparity should therefore reduce differences between AA and white women with regard to breast cancer survival.35
Stage at diagnosis, the strongest predictor of breast cancer survival, is known to contribute to R/E survival disparities.4 Our findings support this hypothesis, given the importance of tumor severity in reducing mortality disparities between AA and white women.
Tumor grade and estrogen receptor status are known to be different between AA and white women, suggesting that they may also contribute to differences in survival.18 In this study, biologic markers were found to have an effect of similar magnitude on survival results as did screening, tumor characteristics, and treatment. In exploring the role of biology further, it is important to understand the determinants of tumor biology given the ongoing debate as to whether biology reflects genetic causes versus exposure to less favorable environmental conditions.36,37
We found persistent differences in treatment by R/E, with AA women more likely to have poor-quality treatment. Adjusting for differences in treatment receipt reduced differences between AA and white women with regard to survival. This finding has been observed elsewhere and may reflect differential access to optimal care, including receipt of adjuvant radiotherapy with breast-conserving surgery.21,26,38,39 R/E disparities in access to optimal treatment, and therefore survival, may reflect differences in healthcare access, regional variations, or a higher disease burden.5 However, our findings among a Medicare population accounted for insurance type, SEER site, type of community, comorbidities, and tumor severity and still found differences.
It is interesting to note that comorbidities and demographics appeared to contribute relatively little to R/E differences noted. However, these factors could each act in a common pathway to other factors, as underserved/poor women are less likely to undergo mammography.40 Because the exact contribution of each variable to the reduction in hazard ratio remains dependent on the model order, it is important to consider these findings within the larger context that highlights a similar reduction in hazard risk for screening, biology, and treatment.
Despite the large number of variables considered in our model, there were persistent differences in stage II/III disease. Other untested hypotheses should be explored, particularly the possibility of R/E differences in the timeliness and quality of breast cancer care.41 Such differences may reflect the patient-physician relationship, patient preferences, and/or institutional and interpersonal racism across healthcare service provision.41–43 Differences in quality may occur in some contexts more often than others, partially explaining why differences persist within stage II/III only for AA and white women.41 For example, R/E differences in the quality of breast cancer care are unlikely to occur when all women are expected to do well or less well (ie, early/late stage) but become unmasked in situations in which timeliness and quality are more likely to affect outcomes (ie, stage II/III cancers).
Due to small numbers of mortality events we chose R/ E categories that may unfortunately mask disparities within each category. This is of particular concern for A/PI, H, and Other categories because subgroups within these categories (ie, Native American, Hawaiian, Mexican, South/Central American, Puerto Rican) have greater breast cancer mortality and poorer survival than white women.4,44 The potential for underserved subgroups within the AA population also exists, with Caribbean women having a different screening profile than other women identified as AA.45,46 Future analyses should disaggregate the R/E categories so that we can better understand these findings.11,44
The use of aggregate-based measures to assign individual SES status is suboptimal.20 However, aggregate measures of income at the community level remains valuable because community-level predictors of SES have been shown to be a strong predictor of healthcare services and health outcomes for all individuals living in that community (both high and low income).47 Our population all have health insurance and we expect that the importance of demographics may differ in a less insurance homogenous sample.
Unfortunately, data regarding hormonal therapy is not available in SEER or Medicare data and information on chemotherapy has not been included in the SEER public use dataset and is of uncertain validity within Medicare.48 Quality indicators such as timeliness are also difficult to measure within the SEER-Medicare database.19 Additional biologic markers were not available (ie, HER-2 status, CYP1A1, or P53 mutations), although there is no conclusive evidence of marker differences between AA and white women.9
This analysis improves on studies using SEER data alone because a more accurate assessment of screening exposure was used (rather than self-report that overestimates exposure), and comorbidity and treatment information were included.21,49,50 Therefore, the current study responds to calls in the literature for analyses that include these variables.7,9,10,37,51
Understanding the reasons for R/E disparities in breast cancer survival remains complex. Despite the complexity, there are still areas in which health policy should clearly intervene. Enough evidence now exists such that policy initiatives should urgently be applied to improve access to adequate screening for AA and H women and increase access to appropriate treatment for AA women in particular. Further multidisciplinary investigation into the role of biology, demographics, and potential disparities in quality of care are required in both young and elderly cohorts of women.
We acknowledge the efforts of the Applied Research Branch, Division of Cancer Prevention and Population Science, National Cancer Institute; the Office of Information Services and the Office of Strategic Planning, Health Care Finance Adminstration; Information Management Services, Inc; and the SEER program tumor registries in the creation of the SEER-Medicare database.
Supported by the Commonwealth Fund, Mount Zion Health Care Fund Dean's Account, Division of Internal Medicine (University of California at San Francisco), California Breast Cancer Research Program, National Cancer Institute (K07 CA 86032), Department of Defense (DAMD 17-99-1-9112), and Ngā Pae o Te Māramatanga (University of Auckland, New Zealand).
Dr. Curtis was a Visiting Harkness Fellow in Healthcare Policy (The Commonwealth Fund) between 2004 and 2005.
We thank Dr. Andrew Bindman (UCSF), Dr. Ashley Bloomfield (MoH, NZ), and Ms Bridget Robson (Te Ropū Rangahau Hauora a Eru Pōmare, NZ).
This study used the linked Surveillance, Epidemiology, and End Results (SEER)-Medicare database. The interpretation and reporting of these data are the sole responsibilities of the authors.