Racial Differences in PAM50 Subtypes in the Carolina Breast Cancer Study
JNCI J Natl Cancer Inst (
Racial Differences in PAM50 Subtypes in the Carolina Breast Cancer Study
Melissa A. Troester 0
Xuezheng Sun 0
Emma H. Allott 0
Joseph Geradts 0
Stephanie M. Cohen 0
Chiu-Kit Tse 0
Erin L. Kirk 0
Leigh B. Thorne 0
Michelle Mathews 0
Yan Li 0
Zhiyuan Hu 0
Whitney R. Robinson 0
Katherine A. Hoadley 0
Olufunmilayo I. Olopade 0
Katherine E. Reeder-Hayes 0
H. Shelton Earp 0
Andrew F. Olshan 0
Lisa A. Carey 0
Charles M. Perou 0
0 Affiliations of authors: Department of Epidemiology (MAT , XS, CKT, ELK, WRR, AFO) , Lineberger Comprehensive Cancer Center (MAT , CKT, YL, ZH, WRR, KAH, KERH, HSE, AFO, LAC , CMP); Department of Pathology and Lab Medicine (MAT , SMC, LBT, MM, CMP) , Department of Nutrition (EHA), and Department of Genetics (KAH, CMP), University of North Carolina at Chapel Hill, Chapel Hill, NC; Dana Farber Cancer Institute, Harvard University , Boston, MA (JG); Center for Clinical Cancer Genetics, Global Health Department of Medicine, The University of Chicago , Chicago, IL, OIO , USA
Background: African American breast cancer patients have lower frequency of hormone receptor-positive (HRþ)/human epidermal growth factor receptor 2 (HER2)-negative disease and higher subtype-specific mortality. Racial differences in molecular subtype within clinically defined subgroups are not well understood. Methods: Using data and biospecimens from the population-based Carolina Breast Cancer Study (CBCS) Phase 3 (2008-2013), we classified 980 invasive breast cancers using RNA expression-based PAM50 subtype and recurrence (ROR) score that reflects proliferation and tumor size. Molecular subtypes (Luminal A, Luminal B, HER2-enriched, and Basal-like) and ROR scores (high vs low/medium) were compared by race (blacks vs whites) and age ( 50 years vs > 50 years) using chi-square tests and analysis of variance tests. Results: Black women of all ages had a statistically significantly lower frequency of Luminal A breast cancer (25.4% and 33.6% in blacks vs 42.8% and 52.1% in whites; younger and older, respectively). All other subtype frequencies were higher in black women (case-only odds ratio [OR] ¼ 3.11, 95% confidence interval [CI] ¼ 2.22 to 4.37, for Basal-like; OR ¼ 1.45, 95% CI ¼ 1.02 to 2.06, for Luminal B; OR ¼ 2.04, 95% CI ¼ 1.33 to 3.13, for HER2-enriched). Among clinically HRþ/HER2- cases, Luminal A subtype was less common and ROR scores were statistically significantly higher among black women. Conclusions: Multigene assays highlight racial disparities in tumor subtype distribution that persist even in clinically defined subgroups. Differences in tumor biology (eg, HER2-enriched status) may be targetable to reduce disparities among clinically ERþ/HER2- cases.
Breast cancer incidence is higher in young black women
compared with young white women, and while 2010
Surveillance, Epidemiology, and End Results data showed that
across all ages white women had higher incidence (
data from the American Cancer Society suggest that overall
incidence rates have converged (
). This convergence could
compound breast cancer mortality disparities. Mortality hazard
rates among black women vary by subtype but are 20% to 150%
higher relative to white women (
). Differences are
particularly pronounced among hormone receptor (HR)–positive,
human epidermal growth factor receptor 2 (HER2)–negative
). Interventions to reduce these disparities require
improved understanding of how tumor-level and patient-level
The Carolina Breast Cancer Study Phase 3 (CBCS3, 2008–
2013) was initiated to disentangle the role of health service and
tumor biological factors in breast cancer disparities (
Beginning in 1993, participants were enrolled in the study via
randomized recruitment that oversampled black women and
women younger than age 50 years at diagnosis. Research from
earlier phases of the Carolina Breast Cancer Study (CBCS1 and 2,
1993–2001) used a five-marker immunohistochemistry panel to
show higher prevalence of Basal-like breast cancer (defined as
HR- and HER2- and positive for either Ck5/6 or EGFR) and lower
prevalence of HRþ/HER2- breast cancer among young (age <
50 years) black women (
), findings that have been confirmed in
other studies (
Decreased HRþ/HER2- disease among young black women
could arise from lower screening utilization (
) or from
differences in risk factor profiles (
) or both. These factors could also
lead to differences in genomic characteristics, even within more
clinically homogeneous groups. Of particular interest is whether
biological differences in HRþ/HER2- tumors of black and white
women could account for mortality disparities by race (
Current guidelines recommend that these patients receive
genomic testing as a decision-making tool (
), but few studies
have utilized genomic data to characterize racial differences in
clinically defined groups (
), and these differences have not
been evaluated using biospecimens from population-based
To elucidate differences in tumor biology by race, we used
high-throughput RNA profiling methods to determine the
PAM50 molecular subtypes of invasive breast cancers from
more than 1000 women participating in the population-based
CBCS3, with nearly equal numbers of black and white women.
We identified PAM50-based intrinsic subtype and classified
patients for ROR score based on proliferation (ROR-P) or a
combination of proliferation and tumor size (ROR-PT) (
). We also
explored Oncotype DX scores by race for a subset of patients
who underwent clinical genomic testing. This work evaluates
racial differences in the relative frequency of molecular
subtypes and assesses whether differences in tumor genomics
persist even within clinically defined subgroups.
The Carolina Breast Cancer Study Phase 3 (CBCS3) is the third
phase (2008–2013) of a population-based study conducted in
North Carolina (NC) that began in 1993; study details and
sampling schemes have been described previously (
cases of invasive breast cancer between age 20 and 74 years
were identified using rapid case ascertainment in cooperation
with the NC Central Cancer Registry, with black and young
cases (age 20–49 years) oversampled using randomized
recruitment (18). Randomized recruitment allows sample weighting to
make inferences about the frequency of subtype in the NC
source population. Tumor size, stage, node status, estrogen
receptor (ER), progesterone receptor (PR), HER2, and Oncotype DX
data were abstracted from medical records, and tumor grade
was centrally assigned by a single pathologist (JG) using the
Nottingham breast cancer grading system (
). The study was
approved by the Office of Human Research Ethics at the
University of North Carolina at Chapel Hill, and informed
consent was obtained from each participant.
Molecular and Clinical Subtyping
Paraffin-embedded tumor blocks were requested from
participating pathology laboratories for each case. The study
pathologist (JG) reviewed hematoxylin and eosin (H&E) for each tumor,
selected a representative tumor block, and circled tumor areas
for coring. Ten 10 mM sections were cut from blocks after coring
(for immunohistochemistry), and an additional H&E section
was obtained and reviewed for tumor cellularity. Only cores
with adequate top and bottom tumor cellularity by manual
review were selected for RNA analyses. Nanostring assays were
performed using two separate 1.0 mm cores. RNA was isolated
from cores using Qiagen RNeasy formalin-fixed,
paraffinembedded kit and protocol, with 95% of tumors producing
quantifiable RNA. In total, 1122 samples from 1042 participants
with invasive breast cancer from CBCS Phase 3 were analyzed
using the PAM50 assay. Samples were randomly assigned to
three batches for RNA analyses, and the laboratory was blinded
to duplicates and all participant characteristics. Two standard
samples were included with each batch: an RNA sample
comprised of Stratagene (La Jolla, CA) Universal Human Reference
and a breast cell-line cocktail comprised of RNA isolates from
pooled breast cell lines. To assess batch variability, we
calculated Pearson correlation coefficients for pairwise comparisons
of these samples across the PAM50 genes. Pearson correlations
were 0.980, 0.987, and 0.992 for the Stratagene reference and
0.987, 0.989, and 0.997 for the breast cell line mix. All assays
were performed in the Rapid Adoption Molecular (RAM)
laboratory at the University of North Carolina.
To classify samples, the NanoStringNorm package in
Bioconductor was first used to eliminate samples that did not
have sufficient Nanostring data quality (39 of 1122, 3%). The
PAM50 predictor (
) was then used to categorize breast tumors
as Luminal A, Luminal B, HER2-enriched, Basal-like, and
normallike and to calculate the risk of recurrence (ROR) score with
proliferation (ROR-P) and proliferation plus tumor size (ROR-PT).
Briefly, each sample was classified based on the subtype centroid
with the highest Pearson correlation. Duplicate samples were
treated independently during classification; after classification,
the sample with the highest PAM50 confidence score was
selected for inclusion in patient-level analyses. After excluding
patients with normal-like subtype (ie, specimens with
insufficient tumor cellularity), patients missing clinical data, and
patients with race other than black or white, 980 patients were
included in the final analysis. Patients were roughly evenly
divided in four groups: black women younger than age 50 years
(n ¼ 232), black women age 50 years or older (n ¼ 268), white
women younger than age 50 years (n ¼ 236), and white women
age 50 years or older (n ¼ 244). Compared with cases not analyzed,
included samples were more likely to be at least age 50 years
(32.5% vs 52.2%, chi-square P ¼ 0.01) and have grade 2 tumors
(15.4% vs 34.3%; chi-square P ¼ 0.01). Participant characteristics
are presented in Supplementary Table 1 (available online).
Biomarker variables, including PAM50-based ROR-P and ROR-PT
and Oncotype DX, were used both continuously and
categorically. The cutoff points to define high levels were 52.9 for ROR-P,
*Frequency estimates (%) were adjusted for sampling weights from the randomized recruitment design. HER2 ¼ human epidermal growth factor receptor 2.
†Two-sided chi-square comparing Basal-like with non-Basal-like across all age/race categories (P < .001).
‡Two-sided chi-square comparing Luminal A to Non–Luminal A across all age/race categories (P < .001).
64.7 for ROR-PT, and 30 for Oncotype DX, per established
). Racial differences by categorical variables, such as
clinical characteristics and subtype, were assessed with
Pearson chi-square tests or Fisher exact tests. Sampling weights
were used to adjust subtype frequency estimates across all
estimates, consistent with the randomized recruitment design of
the study. Odds ratios (ORs) represent prevalence odds ratios
and were estimated using logistic regression, regressing tumor
characteristics on population subgroup (age or race groups).
Analysis of variance was used to assess statistical differences
for continuous variables. All statistical tests were two-sided
with an a of .05, and all analyses were performed using SAS
version 9.2 (SAS Institute, Inc, Cary, NC). P values were not
corrected for multiple comparisons because tumor characteristics
are not independent.
Breast Cancer Subtypes by Race and Age
In the CBCS Phase 3, racial differences in frequency of molecular
subtype were most pronounced for Basal-like and Luminal A
breast cancer. As shown in Table 1, the proportion of Basal-like
breast cancer as measured by RNA profiling was 25.4% overall
(n ¼ 249 cases), but was higher in black women; 36.6% and 31.3% of
younger (age < 50 years) and older (age 50 years) black women,
respectively, had Basal-like breast cancer. Basal-like breast cancer
comprised less than 20% of cases among white women. The
higher frequency of Basal-like tumors in black women was offset
by a decrease in the frequency of Luminal A breast cancer.
The lowest frequency of Luminal A breast cancer was
observed among young black women (25.4%, n ¼ 59), followed
by older black women (33.6%, n ¼ 90), young white
women (42.8%, n ¼ 101), and older white women (52.1%,
n ¼ 127) (Table 1). The frequency of Luminal B cancers was
relatively stable across all four age- and race-defined groups,
at approximately 20%. There was a suggestion that
HER2enriched tumors may be more frequent among young black
women (16.0%, n ¼ 37) compared with all other groups (8.9%–
11.1%); however, the sample size was small, and this
difference was not statistically significant. But considering all age
groups together, Table 2 shows that compared with white
women, black women had statistically significantly higher
odds of all three non–Luminal A subtypes: Basal-like
(OR ¼ 3.11, 95% confidence interval [CI] ¼ 2.22 to 4.37), Luminal
B (OR ¼ 1.45, 95% CI ¼ 1.02 to 2.06), and HER2-enriched breast
cancer (OR ¼ 2.04, 95% CI ¼ 1.33 to 3.13). All of these odds ratios
were attenuated after adjusting for other clinical covariates
(size, node, stage, grade), but the odds ratios for Basal-like
breast cancer remained statistically significant. Table 2 also
illustrates age associations for breast cancer subtypes. Women
younger than age 50 years had higher odds of Basal-like and
Luminal B breast cancer (relative to Luminal A), but age
associations with PAM50 subtype were not statistically significant
after adjusting for clinical covariates.
Mortality disparities are greatest among HRþ/HER2- cancers (
so racial differences in clinical, histopathologic, and biomarker
data are particularly important for this group. Among HRþ/
HER2- patients, tumor size and grade varied statistically
significantly by race and age (Table 3). Compared with older white
women, younger black women had more than twice the odds of
a large tumor (OR ¼ 2.60, 95% CI ¼ 1.51 to 4.45), and both younger
and older black women had twice the odds of having a
highgrade tumor (younger black: OR ¼ 2.66, 95% CI ¼ 1.47 to 4.84;
older black: OR ¼ 2.24, 95% CI ¼ 1.27 to 3.93). Stage and node
status were not statistically significantly different within strata
defined by race or age, although when cross-classifying on both
race and age, younger black women had higher odds of being
Table 4 shows racial differences in biomarker levels and
class distributions among HRþ/HER2- cases (n ¼ 492), including
PAM50 subtype, ROR-P and ROR-PT, and Oncotype DX. ROR-P is a
risk of recurrence score based on correlation to PAM50 subtype
and a proliferation term. ROR-PT utilizes these correlations and
additionally incorporates tumor size (
). ROR-PT is a research
version of a test that is utilized clinically in the United States
and internationally. PAM50 subtype classification is not reported
clinically in the United States but is utilized in Europe. While
ROR-P and ROR-PT are correlated, approximately 20% of cases
have different categorical scores for the two metrics. PAM50
subtype differed statistically significantly across all race and age
strata, and there were statistically significant differences in the
percentage of Luminal A breast tumors between black and white
women with HRþ/HER2- disease. Both older and younger black
women were more likely to have non–Luminal A PAM50 subtype,
but the association was attenuated and not statistically
significant in older black women. The ROR-P and ROR-PT scores
differed by race and age, with black women having statistically
significantly higher frequency of high-risk tumors for both ROR
scores (OR > 1.5 for ROR-P and OR > 2.0 for ROR-PT in both
younger and older black women) (Table 5). On a continuous
scale, average scores for ROR-P and ROR-PT also differed by race
and age (P < .001). Associations between age-/race-defined
groups and PAM50, ROR-PT, and Oncotype DX were not
substantially altered by adjustment for size, grade, node status, or stage.
To evaluate differences in genomic subtype based on race
(without further stratifying on age), several supplemental
analyses were conducted (Supplementary Table 2, available online).
Considering black women vs white women, only about half of
black women (51.0%) had Luminal A PAM50 subtype, compared
with 59.9% of white women (P ¼ .05). Black women overall
(combining age groups) had higher odds of a high risk score by ROR-P
(OR ¼ 1.31, 95% CI ¼ 0.82 to 2.08), ROR-PT (OR ¼ 1.87, 95% CI ¼ 1.11
to 3.16), and Oncotype DX (OR ¼ 1.63, 95% CI ¼ 0.63 to 4.24), but
these racial differences were statistically significant only for
ROR-PT (Supplementary Table 2, available online). The higher
frequency of ROR-P high among black women was
independently validated using data from The Cancer Genome Atlas
Project (TCGA) (Supplementary Table 2, available online); in the
TCGA, the magnitude of association was similar (OR ¼ 2.87, 95%
CI ¼ 1.24 to 6.65). ROR-PT could not be assessed in TCGA because
of missing data on tumor size.
Using RNA expression data, this analysis classified breast
cancer PAM50 subtype in a population-based sample of black and
*Two-sided chi-square P values exclude participants with missing data, and for PAM50 subtype, normal-like cases are excluded. HER2 ¼ human epidermal growth
factor receptor 2; HR ¼ hormone receptor; ROR-P ¼ risk of recurrence based on proliferation; ROR-PT ¼ risk of recurrence based on tumor size.
†P value calculated using a two-sided analysis of variance test.
white women. As has been previously reported using other
molecular methods (
), the results show a strong racial disparity in
the frequency of Basal-like breast cancer, affecting both younger
and older black women. The higher relative frequency of
Basallike breast cancer in black women is offset by a decreased
frequency of Luminal A breast cancers. Black women also had
higher frequency of Luminal B and HER2-enriched PAM50
subtype. Mortality disparities among clinically defined
HRþ/HER2cases are greater than in other clinical subgroups (
), and these
analyses suggest that tumor biology may contribute; black
women with clinically defined HRþ/HER2- disease are more
likely to have aggressive PAM50 subtypes (Luminal B,
HER2enriched, or Basal-like) and high risk of recurrence (ROR) scores.
The potential of genomic biomarkers to guide clinical
decision-making has the largest impact among HRþ/HER2- cases,
where genomic tests are most frequently utilized. Although
clinically indistinguishable by standard tests, outcomes vary
widely and racial disparities are pronounced in this group (
Several commercial genomic tests are available clinically
(Oncotype DX, Mammaprint, Prosigna), and relative to standard
clinical markers or immunohistochemical surrogates (
least one previous observational study has shown that
RNAbased subtyping more accurately predicts recurrence and
survival (23). Based on the current analysis, all four tumor PAM50
subtypes occur in HRþ/HER2- cases, with black women having
higher frequency of non–Luminal A (Basal-like, HER2-enriched,
and Luminal B) cancer relative to white women with the same
clinical profile. The ROR-P and ROR-PT scores that track
proliferation and proliferation plus tumor size, respectively, were
higher in black women of all ages relative to white women.
Quantitative Oncotype DX scores did not vary statistically
significantly by race, but there were increased odds of a high risk
score among black women overall. When considering
categorical risk scores, only the ROR-PT showed statistically
significantly higher odds of “high-risk” relative to “low- and
mediumrisk” tumors. While previous studies have evaluated uptake of
Oncotype DX by race (
), few studies have compared
categorical Oncotype DX scores by race among HRþ/HER2- cases.
Two previous studies found increased odds of high Oncotype
DX recurrence score for black vs white women (
), with both
reporting similar proportions of patients with high Oncotype
DX. While racial differences for Oncotype DX scores were not
statistically significant in this population, the magnitude and
direction of our findings agree with those previous reports.
Across multiple genomic biomarkers, high-risk tumors are
more common (roughly 10% of the HRþ/HER2- population) in
The current analysis suggests that genomic subtyping could
have important clinical implications. Among
HRþ/HER2patients, higher risk scores in black patients suggest that
adherence to American Society of Clinical Oncology guidelines (
may result in chemotherapy or novel therapeutic approaches
for a larger proportion of black patients (ie, because high risk
scores are an indication for chemotherapy). Detection of
HER2-enriched status among clinically HER2- cases may also be
targetable, given a recent clinical trial suggesting that these
cases benefit from HER2-targeted therapies (
). Disparities in
frequency of positive HER2 status by race have been evaluated
previously in Surveillance, Epidemiology, and End Results data,
and differences between black and white women have been
). However, the current findings suggest that clinical
tests may miss some HER2-enriched cancers that would benefit
from biologic therapy. In our study, approximately 5% of
clinically defined ERþ/HER2- cases were HER2-enriched by PAM50
The distribution of PAM50 subtype detected in our analysis
differs from a previous population-based analysis conducted in
the Life After Cancer Epidemiology (LACE) and Pathways studies
) and suggests a poorer prognostic profile for blacks overall.
In LACE/Pathways, roughly half of the tumors were Luminal A,
whereas only 38% were Luminal A in CBCS3. Frequency of
Luminal B (both studies approximately 20%) and HER2-enriched
(13% in LACE/Pathways, 12% in CBCS3) tumors were similar, but
CBCS3 had a much higher frequency of Basal-like breast cancer
(9.8% in LACE/Pathways vs 25% in CBCS3). The differences
between these two studies may reflect national geographic trends
or differences in population genetics; a recent Report to the
Nation has emphasized geographic variation in incidence of
triple-negative breast cancer, with highest incidence in the
southeastern United States (
). However, the most compelling
differences between LACE/Pathways and CBCS are in race and
age composition; namely, LACE/Pathways was predominantly
(>75%) older women and fewer than 10% were black. A strength
of the CBCS3 for disparities research is oversampling of young
women (age < 50 years) and black women.
Our findings should be interpreted in light of some
limitations. We used a research version of the PAM50, ROR-P, and
ROR-PT and not the clinically approved Prosigna assay (which
reports the ROR-PT score). Clinical data for these tests were
unavailable. Oncotype DX data were available in the clinical
record; however, data were missing for about 60% of
HRpositive/HER2-negative patients in our study, limiting statistical
power to detect small differences by race and age. While
missing data limit comparisons of Oncotype DX and ROR-P/PT
results, the Oncotype DX association with race is similar to
those reported previously. Furthermore, the statistically
significant association between race and ROR-PT persisted when
restricting to the group of patients with clinically available
Oncotype DX data. Considering PAM50 subtype, we lacked the
statistical power to assess race- and age-associated differences
in the frequency of HER2-enriched and Luminal B breast cancer.
We did observe statistically significant differences in the
frequency of both PAM50 subtypes by race, but could not further
evaluate whether these differences were stronger for younger
vs older black women. Finally, the CBCS3 did not collect
screening records for the years prior to diagnosis, so we were unable
to classify patients as to mode of detection (screen detected,
interval detected, medically detected). Given that screening
adherence patterns may differ by race and age and could lead to
higher rates of indolent, screen-detected cancers in older white
women, relationships between screening patterns and subtype
frequency distributions merit further investigation.
Despite some limitations, these data clearly show that even
within clinically defined subgroups, there are important
biological differences between black and white women’s tumors. A
persistent high-priority research question is how tumor
biological characteristics balance with patient-level variables (such
as treatment adherence or access to quality diagnostic
information that informs therapy) in the progression of
HRþ/HER2breast cancers. CBCS3 recruitment ended in 2013, and survival
data are not yet mature enough for analyses of how PAM50 or
clinical subtype may mediate survival disparities, but future
work will leverage the biological data collected herein, together
with detailed risk factor and treatment data, to elucidate how
multiple factors work together to produce differences in
frequency distributions and overall poorer outcomes in all
subtypes of breast cancer for black women.
KAH and KERH received Career Catalyst Awards from the Susan
G. Komen Foundation. Komen Foundation also provided
financial support for CBCS study infrastructure. Funding was
provided by the National Institutes of Health, National Cancer
Institute P50-CA058223 to HSE, U54-CA156733 to HSE,
P01CA151135, and U01-CA179715 to MAT and CMP.
The funders had no role in the design of the study; the
collection, analysis, or interpretation of the data; the writing of the
manuscript; or the decision to submit the manuscript for
MAT, XS, EHA, CKT, WRR, KAH, KERH, HSE, JG, AFO, LAC, and
CMP contributed to data collection, analysis, and review of the
manuscript. FO, LT, MM, SC, EK, YL, and ZH contributed to data
collection and review of the manuscript. MAT and XS had full
access to all of the data in the study and take responsibility for
the integrity of the data and the accuracy of the data analysis.
1. Anderson WF , Rosenberg PS , Menashe I , Mitani A , Pfeiffer RM . Age-related crossover in breast cancer incidence rates between black and white ethnic groups . J Natl Cancer Inst . 2008 ; 100 ( 24 ): 1804 - 1814 .
2. DeSantis CE , Fedewa SA , Goding Sauer A , Kramer JL , Smith RA , Jemal A . Breast cancer statistics, 2015 : Convergence of incidence rates between black and white women . CA Cancer J Clin . 2016 ; 66 ( 1 ): 31 - 42 .
3. Menashe I , Anderson WF , Jatoi I , Rosenberg PS . Underlying causes of the black-white racial disparity in breast cancer mortality: A population-based analysis . J Natl Cancer Inst . 2009 ; 101 ( 14 ): 993 - 1000 .
4. Warner ET , Tamimi RM , Hughes ME , et al. Racial and ethnic differences in breast cancer survival: Mediating effect of tumor characteristics and sociodemographic and treatment factors . J Clin Oncol . 2015 ; 33 ( 20 ): 2254 - 2261 .
5. O 'Brien KM , Cole SR , Tse CK , et al. Intrinsic breast tumor subtypes, race, and long-term survival in the Carolina Breast Cancer Study . Clin Cancer Res . 2010 ; 16 ( 24 ): 6100 - 6110 .
6. McGee SA , Durham DD , Tse CK , Millikan RC . Determinants of breast cancer treatment delay differ for African American and White women . Cancer Epidemiol Biomarkers Prev . 2013 ; 22 ( 7 ): 1227 - 1238 .
7. Hair BY , Hayes S , Tse CK , Bell MB , Olshan AF . Racial differences in physical activity among breast cancer survivors: Implications for breast cancer care . Cancer . 2014 ; 120 ( 14 ): 2174 - 2182 .
8. Roberts MC , Weinberger M , Dusetzina SB , et al. Racial variation in the uptake of Oncotype DX testing for early-stage breast cancer . J Clin Oncol . 2016 ; 34 ( 2 ): 130 - 138 .
9. Carey LA , Perou CM , Livasy CA , et al. Race, breast cancer subtypes, and survival in the Carolina Breast Cancer Study . JAMA . 2006 ; 295 ( 21 ): 2492 - 2502 .
10. Parise CA , Bauer KR , Caggiano V. Variation in breast cancer subtypes with age and race/ethnicity . Crit Rev Oncol Hematol . 2010 ; 76 ( 1 ): 44 - 52 .
11. Parise CA , Bauer KR , Brown MM , Caggiano V . Breast cancer subtypes as defined by the estrogen receptor (ER), progesterone receptor (PR), and the human epidermal growth factor receptor 2 (HER2) among women with invasive breast cancer in California, 1999 - 2004 . Breast J. 2009 ; 15 ( 6 ): 593 - 602 .
12. Howlader N , Altekruse SF , Li CI , et al. US incidence of breast cancer subtypes defined by joint hormone receptor and HER2 status . J Natl Cancer Inst . 2014 ; 106 ( 5 ): dju055 .
13. Dawson SJ , Duffy SW , Blows FM , et al. Molecular characteristics of screendetected vs symptomatic breast cancers and their impact on survival . British J Cancer . 2009 ; 101 ( 8 ): 1338 - 1344 .
14. Millikan RC , Newman B , Tse CK , et al. Epidemiology of basal-like breast cancer . Breast Cancer Res Treat . 2008 ; 109 ( 1 ): 123 - 139 .
15. Harris LN , Ismaila N , McShane LM , et al. Use of biomarkers to guide decisions on adjuvant systemic therapy for women with early-stage invasive breast cancer: American Society of Clinical Oncology Clinical Practice Guideline . J Clin Oncol . 2016 ; 34 ( 10 ): 1134 - 1150 .
16. D'Arcy M , Fleming J , Robinson WR , Kirk EL , Perou CM , Troester MA . Raceassociated biological differences among Luminal A breast tumors . Breast Cancer Res Treat . 2015 ; 152 ( 2 ): 437 - 448 .
17. Parker JS , Mullins M , Cheang MC , et al. Supervised risk predictor of breast cancer based on intrinsic subtypes . J Clin Oncol . 2009 ; 27 ( 8 ): 1160 - 1167 .
18. Newman B , Moorman PG , Millikan R , et al. The Carolina Breast Cancer Study: Integrating population-based epidemiology and molecular biology . Breast Cancer Res Treat . 1995 ; 35 ( 1 ): 51 - 60 .
19. Elston CW , Ellis IO . Pathological prognostic factors in breast cancer. I. The value of histological grade in breast cancer: Experience from a large study with long-term follow-up . Histopathology . 1991 ; 19 ( 5 ): 403 - 410 .
20. Paik S , Tang G , Shak S , et al. Gene expression and benefit of chemotherapy in women with node-negative, estrogen receptor-positive breast cancer . J Clin Oncol . 2006 ; 24 ( 23 ): 3726 - 3734 .
21. Nielsen TO , Parker JS , Leung S , et al. A comparison of PAM50 intrinsic subtyping with immunohistochemistry and clinical prognostic factors in tamoxifen-treated estrogen receptor-positive breast cancer . Clin Cancer Res . 2010 ; 16 ( 21 ): 5222 - 5232 .
22. Allott EH , Cohen SM , Geradts J , et al. Performance of three-biomarker immunohistochemistry for intrinsic breast cancer subtyping in the AMBER Consortium . Cancer Epidemiol Biomarkers Prev . 2016 ; 25 ( 3 ): 470 - 478 .
23. Caan BJ , Sweeney C , Habel LA , et al. Intrinsic subtypes from the PAM50 gene expression assay in a population-based breast cancer survivor cohort: Prognostication of short- and long-term outcomes . Cancer Epidemiol Biomarkers Prev . 2014 ; 23 ( 5 ): 725 - 734 .
24. Roberts MC , Weinberger M , Dusetzina SB , et al. Racial variation in adjuvant chemotherapy initiation among breast cancer patients receiving oncotype DX testing . Breast Cancer Res Treat . 2015 ; 153 ( 1 ): 191 - 200 .
25. Sheppard VB , O'Neill SC , Dilawari A , Horton S , Hirpa FA , Isaacs C . Patterns of 21-gene assay testing and chemotherapy use in black and white breast cancer patients . Clin Breast Cancer . 2015 ; 15 ( 2 ): e83 - e92 .
26. Jasem J , Amini A , Rabinovitch R , et al. 21 -gene recurrence score assay as a predictor of adjuvant chemotherapy administration for early-stage breast cancer: An analysis of use, therapeutic implications, and disparity profile . J Clin Oncol . 2016 ; 34 ( 17 ): 1995 - 2002 .
27. Lund MJ , Mosunjac M , Davis KM , et al. 21 -Gene recurrence scores: Racial differences in testing, scores, treatment, and outcome . Cancer . 2012 ; 118 ( 3 ): 788 - 796 .
28. Prat A , Cheang MC , Galvan P , et al. Prognostic value of intrinsic subtypes in hormone receptor-positive metastatic breast cancer treated with letrozole with or without lapatinib . JAMA oncology . 2016 ; 2 ( 10 ): 1287 - 1294 .
29. Kurian AW , Fish K , Shema SJ , Clarke CA. Lifetime risks of specific breast cancer subtypes among women in four racial/ethnic groups . Breast Cancer Res . 2010 ; 12 ( 6 ): R99 .
30. Sweeney C , Bernard PS , Factor RE , et al. Intrinsic subtypes from PAM50 gene expression assay in a population-based breast cancer cohort: Differences by age, race, and tumor characteristics . Cancer Epidemiol Biomarkers Prev . 2014 ; 23 ( 5 ): 714 - 724 .
31. Kohler BA , Sherman RL , Howlader N , et al. Annual report to the nation on the status of cancer, 1975-2011, featuring incidence of breast cancer subtypes by race/ethnicity, poverty, and state . J Natl Cancer Inst . 2015 ; 107 ( 6 ): djv048 .