Downloaded from caonline.amcancersoc.org by on June 13, 2010 (American Cancer Society, Inc.

)
New Molecular Classifications of Breast Cancer
Mary Cianfrocca and William Gradishar
CA Cancer J Clin 2009;59;303-313
DOI: 10.3322/caac.20029

This information is current as of June 13, 2010

The online version of this article, along with updated information and services, is
located on the World Wide Web at:
http://caonline.amcancersoc.org/cgi/content/full/59/5/303

To subscribe to the print issue of CA: A Cancer Journal for Clinicians, go to (US
individuals only): http://caonline.amcancersoc.org/subscriptions/

CA: A Cancer Journal for Clinicians is published six times per year for the American Cancer
Society by Wiley-Blackwell. A bimonthly publication, it has been published continuously since
November 1950. CA is owned, published, and trademarked by the American Cancer Society, 250
Williams Street NW, Atlanta GA 30303. (American Cancer Society, Inc.) All rights reserved.
Print ISSN: 0007-9235. Online ISSN: 1542-4863.
 CA CANCER J CLIN 2009;59:303313

New Molecular Classications of Breast Cancer

Mary Cianfrocca, DO1 and William Gradishar, MD2

Abstract
Traditionally, pathologic determinations of tumor size, lymph node status, endocrine receptor status, and human
epidermal growth factor receptor 2 (HER2) status have driven prognostic predictions and adjuvant therapy
recommendations for patients with early stage breast cancer. However, these prognostic and predictive factors are

Downloaded from caonline.amcancersoc.org by on June 13, 2010 (American Cancer Society, Inc.)
relatively crude measures, resulting in many patients being overtreated or undertreated. As a result of gene
expression assays, there is growing recognition that breast cancer is a molecularly heterogeneous disease. Evidence
from gene expression microarrays suggests the presence of multiple molecular subtypes of breast cancer. The
recent commercial availability of gene expression proling techniques that predict risk of disease recurrence as well
as potential chemotherapy benet have shown promise in rening clinical decision making. These techniques will be
reviewed in this article. CA Cancer J Clin 2009;59:303313. 2009 American Cancer Society, Inc.

To earn free CME credit or nursing contact hours for successfully completing the online quiz based on this article, go
to http://CME.AmCancerSoc.org.

Introduction
Clinically, breast cancer is a remarkably heterogeneous disease. Traditionally, pathologic determinations of tumor
size, lymph node status, endocrine receptor status, and human epidermal growth factor receptor 2 (HER2) status
have driven prognostic predictions and, ultimately, adjuvant therapy recommendations for patients with early
stage breast cancer. However, these prognostic and predictive factors are relatively crude measures and many
patients are overtreated or undertreated as a result. Using data from the Surveillance, Epidemiology, and End
Results (SEER) database and the results of individual clinical trials, Ravdin et al. developed a widely used,
computerized model called Adjuvant! Online, an online source (available at: http://www.adjuvantonline.com,
Accessed July 27, 2009) to facilitate clinical decision making.1 In an independent validation using data from the
British Columbia Breast Cancer Outcomes Unit, Adjuvant! Online performed reliably. However, although
predicted and observed outcomes were within 2% for the majority of the demographic, pathologic, and
treatment-dened subgroups, Adjuvant! Online overestimated overall survival (OS), breast cancer-specic sur-
vival, and event-free survival for women aged younger than 35 years and for patients with tumors with
lymphovascular or vascular invasion.2 Recently, gene expression proling techniques that predict risk of disease
recurrence as well as potential chemotherapy benet have shown promise in rening clinical decision making.

Breast Cancer Molecular Subtypes

Evidence from gene expression microarrays suggests the presence of multiple molecular subtypes of breast cancer.
Using complementary DNA (cDNA) microarrays representing 8,102 human genes to characterize gene expres-

1
Assistant Professor, Division of Hematology/Oncology, Northwestern University, Feinberg School of Medicine, Robert H. Lurie Comprehensive Cancer
Center, Chicago, IL; 2Professor, Division of Hematology/Oncology, Northwestern University, Feinberg School of Medicine, Robert H. Lurie Comprehensive
Cancer Center, Chicago, IL.

Corresponding author: Mary Cianfrocca, DO, Division of Hematology/Oncology, Northwestern University, Feinberg School of Medicine, Robert H. Lurie Comprehensive
Cancer Center, 676 N. St. Clair, Suite 850, Chicago, IL 60611; [email protected]
DISCLOSURES: Dr. Cianfrocca has received honoraria from Genomic Health and Genentech. No other conict of interest relevant to this article was reported.
2009 American Cancer Society, Inc. doi:10.3322/caac.20029.
Available online at http://cajournal.org and http://cacancerjournal.org

VOLUME 59 NUMBER 5 SEPTEMBER/OCTOBER 2009 303

 New Molecular Classications of Breast Cancer

sion patterns in a set of 65 surgical specimens of Among the 626 ER-positive tumors studied, 73%
human breast tumors from 42 different individuals, were luminal A or B, 11% were HER2-enriched, 5%
Perou et al demonstrated that the phenotypic diver- were basal-like, and 12% were normal breast. In
sity of breast tumors was associated with correspond- contrast, among the ER-negative tumors, 11% were
ing gene expression diversity.3 From the genes in the luminal A or B, 32% were HER2-enriched, 50%
65 tissues samples, the investigators selected a subset were basal-like, and 7% were normal breast. The
of 456 genes, which were termed the intrinsic gene intrinsic subtypes as distinct entities were found to
subset, and consisted of genes with signicantly have a signicant impact on RFS in the untreated
greater expression variation between different tumors patients and remained signicant in multivariate

Downloaded from caonline.amcancersoc.org by on June 13, 2010 (American Cancer Society, Inc.)
than between paired samples from the same tumor. analysis incorporating standard prognostic factors
Using this subset, the authors were then able to such as ER status, histologic grade, tumor size, and
identify 4 different molecular subtypes of breast can- lymph node status. Furthermore, the intrinsic sub-
cer: estrogen receptor (ER)-positive/luminal-like, type model predicted the likelihood of a pCR after
basal-like, ErbB2-positive, and normal breast. Sub- neoadjuvant chemotherapy, with a sensitivity and
sequent data expanded the classication to distin- specicity of 94% and 57%, respectively. The positive
guish between luminal A and luminal B.4 These 5 and negative predictive values were 43.2% and
molecular subtypes have been conrmed in indepen- 96.6%, respectively. However, there were signicant
dent data sets5 and, importantly, the gene expression discrepancies between the clinical classication of the
subtype appears consistent between primary tumors tumors and the classication based on intrinsic sub-
and subsequent metastatic lesions occurring years types. For example, of the 626 ER-positive tumors
later.6 Furthermore, the subtypes are associated with analyzed in the microarray test set, 5% were found to
differences in clinical outcome. Sorlie et al examined be basal-like. Of the 33 HER2-positive tumors, only
a subset of 49 patients with locally advanced breast 64% were classied as HER2-enriched by gene ex-
cancer who were treated with doxorubicin and had a pression and 6% were classied as basal-like. Fur-
median follow-up of 66 months and found that the thermore, 9% of the HER2-negative tumors were
recurrence-free survival (RFS) and OS differed sig- classied as HER2-enriched by gene expression. The
nicantly among the breast cancer subtypes, with the authors conclusions were that ER and HER2 status
luminal A tumors having the longest survival times, are not accurate surrogates for the true intrinsic sub-
the basal-like and HER2-positive subtypes having type status.7 However, this raises important questions
the shortest survival times, and the luminal B tumors with regard to the optimal classication system to
having an intermediate survival time.4 guide therapeutic decision making.
Recently, a risk model incorporating the gene ex-
pression-based luminal A and B, HER2-positive, Gene Expression Proling Assays
and basal-like subtypes was developed by Parker et
al.7 Using microarray and quantitative reverse tran- The 70-Gene Assay (MammaPrint)
scriptase-polymerase chain reaction (RTPCR) data Using inkjet-synthesized oligonucleotide microarrays
from 189 samples, a 50-gene subtype predictor was on primary breast tumors from 117 patients aged
developed and evaluated in 2 cohorts of patients: a younger than 55 years, investigators from the Neth-
cohort of patients receiving no adjuvant systemic erlands Cancer Institute identied a gene expression
therapy and a cohort of patients undergoing neoad- prole based on 70 genes associated with prognosis
juvant chemotherapy with paclitaxel, uorouracil, in patients with lymph node-negative breast cancer.8
doxorubicin, and cyclophosphamide. Test sets from The odds ratio for the development of metastatic
761 patients who did not receive systemic therapy disease from a tumor with a poor-prognosis gene
were evaluated for prognosis and 133 samples from signature compared with a tumor with a good-prog-
patients who received neoadjuvant chemotherapy nosis gene signature was approximately 15. To vali-
were evaluated for prediction of a pathologic com- date the prole, a cohort of 295 consecutive patients
plete response (pCR) after neodjuvant chemother- aged younger than 53 years with stage I or II breast
apy. cancer (151 with lymph node-negative disease and

304 CA: A Cancer Journal for Clinicians

 CA CANCER J CLIN 2009;59:303313

144 with lymph node-positive disease) were evalu- 70-gene prole did not appear to derive a signicant
ated and classied as having either a poor or good benet from chemotherapy (P .962). However, a
prognosis prole.9 There were 69 ER-negative tu- limitation of this analysis was the relatively small
mors and 226 ER-positive tumors. Among the 295 number of events in the low-risk group of patients.
patients, 180 were classied as having a poor-prog- The 70-gene assay requires fresh mRNA for anal-
nosis signature and 115 as having a good-prognosis ysis (fresh-frozen tumor samples or tissues collected
signature, with mean (/ SE) overall 10-year sur- in an RNA preservative solution).
vival rates of 54.6% 4.4% and 94.5% 2.6%, re-
spectively. At 10 years, the probability of remaining 76-Gene Assay
free of distant metastasis was 50.6% 4.5% for the

Downloaded from caonline.amcancersoc.org by on June 13, 2010 (American Cancer Society, Inc.)
Investigators from Rotterdam, the Netherlands, identi-
group with a poor-prognosis signature and 85.2% ed a 76-gene signature (60 genes for patients with
4.3% for the group with a good-prognosis signa-
ER-positive disease and 16 genes for patients with
ture. The hazard ratio (HR) for distant metastases in
ER-negative disease) in a training set of 115 tumors.12
the poor-prognosis group compared with the good-
In an independent testing set of 171 patients with
prognosis group was 5.1 (95% condence interval
lymph node-negative disease, this signature demon-
[95% CI], 2.9 9.0; P .001) and this ratio remained
strated 93% sensitivity and 48% specicity in identifying
signicant when analyzed according to lymph node
patients who developed distant metastatic disease
status. Furthermore, on multivariate analysis, the
prognosis prole was found to be a strong indepen- within 5 years (HR, 5.67; 95% CI, 2.59 12.4). At 80
dent predictor of the likelihood of distant metastases months, the absolute difference between the patients
(HR, 4.6; 95% CI, 2.39.2 [P .001]). with a good and those with a poor prognosis was 39%
The assay was further validated by the Transla- (88% vs 49%) for DDFS and 27% (97% vs 70%) for
tional Breast International Group (TRANSBIG) re- OS. Subgroup analysis demonstrated the prole to be a
search consortium in a retrospective study of frozen, strong prognostic factor for both premenopausal and
archival tumor material collected from 302 patients postmenopausal women, as well as women with small
with lymph node-negative disease from 5 non-Dutch tumors (those measuring 12 cm).
cancer centers.10 All the patients were aged 60 years This signature was subsequently validated in an in-
or younger and had lymph node-negative, T1 or T2 dependent, multi-institutional set of tumor samples
tumors, and the majority of patients had not received from 180 patients with lymph node-negative disease
systemic adjuvant therapy. The median follow-up who did not receive adjuvant systemic therapy.13 In
was 13.6 years. The 70-gene prognosis prole was this group, the 5-year and 10-year DDFS rates were
found to be a signicant prognostic indicator of both 96% and 94%, respectively, for the good-prole
distant disease-free survival (DDFS) and OS in this group and 74% and 65%, respectively, for the poor-
group of patients. prole group. The sensitivity and specicity for
There are emerging data addressing the ability of 5-year DDFS were 90% and 50%, respectively. This
the 70-gene assay to predict chemotherapy benet. analysis conrmed the signature to be a strong prog-
Recently, a pooled analysis of 1,637 patients collected nostic factor in the subgroups of ER-positive patients
from 7 large data sets at multiple institutions across and both premenopausal and postmenopausal pa-
Europe was reported.11 In this meta-analysis, the tients, as well as those with a tumor size 20 mm.
70-gene assay assigned 772 patients (47%) to the However, the subgroup of patients with ER-negative
low-risk category and 865 patients (53%) to the tumors was too small for analysis.
high-risk category. Among these patients, 349 were The 76-gene assay also requires fresh or frozen
treated with endocrine therapy alone, whereas 226 extracted mRNA, similar to the 70-gene assay.
were treated with both chemotherapy and endocrine
therapy. Patients with a poor-prognosis 70-gene pro- The HOXB13:IL17BR Assay
le appeared to derive a signicant benet from the Ma et al performed microarray gene expression anal-
addition of chemotherapy. DDFS was improved ysis of 60 tumors identied from a total of 103
from 69% to 88% (HR, 0.28; 95% CI, 0.14 0.56; [P patients with ER-positive, early stage breast cancer
.001]). Conversely, patients with a good-prognosis who presented to Massachusetts General Hospital

VOLUME 59 NUMBER 5 SEPTEMBER/OCTOBER 2009 305

 New Molecular Classications of Breast Cancer

TABLE 1. Validation Studies for the HOXB13:IL17BR Assay

TRIAL POPULATION RESULTS

North Central Cancer Treatment 211 postmenopausal women with ER- Lymph node-positive patients:
Group (NCCTG) 89-30-52 (Goetz positive, early stage breast cancer who were no association between the HOXB13:IL17BR
2006)16 treated with tamoxifen. expression ratio and RFS or OS
Lymph node-negative patients:
a high HOXB13:IL17BR expression ratio was
associated with worse RFS and OS
compared with a low ratio.
Tumor Bank and Data Network 852 patients with stage I or stage II breast The HOXB13:IL17BR expression ratio
Core at the Breast Center of cancer who were treated with tamoxifen and predicted clinical outcome independently of
Baylor College of Medicine (Ma 286 patients with stage I or stage II breast tamoxifen treatment in the patients with

Downloaded from caonline.amcancersoc.org by on June 13, 2010 (American Cancer Society, Inc.)
2006)17 cancer who did not receive tamoxifen who ER-positive disease. Its prognostic ability
were diagnosed between 1973 and 1993. was stronger in patients with lymph node-
negative disease. In the subgroup of
patients with ER-positive, lymph node-
negative disease, multivariate analysis
demonstrated the expression ratio to be a
signicant predictor of RFS (HR, 3.9; 95%
CI, 1.5-10.3 [P .007]).
Rotterdam cohort (Jansen 2007)18 1,252 patients with ER-positive, operable breast The HOXB13:IL17BR expression ratio was
cancer. A total of 468 patients with ER-positive, found to be signicantly associated with
primary breast cancer were analyzed, 217 poor DFS and OS. The expression ratio was
(46%) of whom developed disease recurrence found to be signicantly associated with
during the follow-up period. Expression levels poor response to tamoxifen (P .027) and
were also evaluated in 193 patients with ER- short PFS (P .001).
positive, primary breast cancer who developed
disease recurrence and were treated with rst-
line tamoxifen therapy.

ER indicates estrogen receptor; RFS, recurrence-free survival; OS, overall survival; HR, hazard ratio; 95% CI, 95% condence interval; DFS, disease-free survival; PFS,
progression-free survival.

between 1987 and 1997.14 All the women were uoxymesterone for 1 year. Tumor blocks were ob-
treated with adjuvant tamoxifen alone. A 2-gene tained from 211 of the 256 patients treated with
expression ratio comprised of the homeobox gene tamoxifen alone and RTPCR proles for HOXB13
HOXB13 and the interleukin17B receptor IL17BR and IL17BR were obtained from 206 patients. The
(HOXB13:IL17BR) was generated and found to be HOXB13:IL17BR expression ratio was not found to
predictive of disease-free survival (DFS). HOX genes be associated with relapse or survival in the lymph
control morphogenesis and also play a role in main- node-positive (n 86) group of patients. However,
taining tissue specicity.15 HOXB13 may interact in the lymph node-negative group (n 130), a high
with the ER receptor and therefore overexpression ratio was associated with a signicantly worse RFS
may contribute to tamoxifen resistance. The role of (HR, 1.98; P .031), DFS (HR, 2.03; P .015),
IL17BR in breast cancer is less clear. The IL17BR and OS (HR, 2.4; P .014) compared with a low
gene, located at 3p21, is frequently lost in breast HOXB13:IL17BR expression ratio.
cancer. It has been hypothesized that one explanation The HOXB13:IL17BR expression ratio was sub-
for the correlation between IL17BR and prognosis is sequently validated in a larger, independent patient
that low expression of the gene correlates with loss of cohort using data from the Tumor Bank and Data
tumor suppressor genes at 3p21. Network Core at the Breast Center of the Baylor
The HOXB13:IL17BR ratio was validated in sev- College of Medicine in Houston, Texas.17 Expres-
eral population data sets (Table 1).16 18 It was ini- sion of HOXB13 and IL17BR were quantied using
tially validated using the North Central Cancer RTPCR in 852 patients not treated with tamoxifen
Treatment Group (NCCTG) 89 30 52 trial, an and 286 patients who did receive tamoxifen treat-
adjuvant tamoxifen trial.16 The NCCTG 89 30 52 ment. The HOXB13:IL17BR expression ratio was
trial randomized 541 postmenopausal women with found to be predictive of clinical outcome indepen-
ER-positive, early stage breast cancer to receive ta- dent of tamoxifen treatment in the group of patient
moxifen for 5 years or tamoxifen for 5 years plus with ER-positive disease; however, its prognostic

306 CA: A Cancer Journal for Clinicians

 CA CANCER J CLIN 2009;59:303313

TABLE 2. Validation Studies for Prognostic Value of the 21-Gene Recurrence Score Assay
RATE OF DISTANT RISK OF BREAST
DISEASE CANCER-
RECURRENCE AT RELATED DEATH
VALIDATION STUDY POPULATION RISK CATEGORY RESULTS 10 YEARS AT 10 YEARS

NSABP B14 (Paik 2005) 24 2,892 women with ER-positive, lymph Tamoxifen-treated patients: Low risk: 6.8%
node-negative breast cancer were Low risk: 51% (RS 18), Intermediate risk:
randomized to 5 y of tamoxifen or Intermediate risk: 22% (RS, 14.3%
placebo; an additional 1,235 women were 18-31), High risk: 30.5%.
assigned to receive an additional 5 y of High risk: 27% (RS 31).
tamoxifen.
RT-PCR was performed on 668 samples.

Downloaded from caonline.amcancersoc.org by on June 13, 2010 (American Cancer Society, Inc.)
Kaiser Permanente Study Case-control study of 4,964 women not Among 55 cases, 150 ER-positive
(Habel 2006)25 treated with adjuvant chemotherapy. controls, tamoxifen-treated patients treated
220 patients who died of breast cancer patients: with tamoxifen:
and 570 matched controls. Low risk: 29% (RS 18). Low risk: 2.8%
Intermediate risk: 40% (RS, Intermediate
18-31). risk: 10.7%
High risk: 31% (RS 31). High risk:
15.5%
The University of Texas M. D. 149 patients with lymph node-negative RS was not found
Anderson Cancer Center breast cancer who had not received to be predictive of
Study (Esteva 2005)26 adjuvant systemic therapy. distant disease
recurrence.

NSABP B14 indicates National Surgical Adjuvant Breast and Bowel Project Protocol B14; ER, estrogen receptor; RTPCR, reverse transcriptate-polymerase chain
reaction; RS, recurrence score.

ability was stronger in the patients with lymph node- and were treated with rst-line tamoxifen therapy. The
negative disease. In the subgroup of patients with HOXB13:IL17BR expression ratio was found to be
ER-positive, lymph node-negative disease, multivar- signicantly associated with a poor response to tamox-
iate analysis including age, progesterone receptor sta- ifen (P .027) and a short progression-free-sur-
tus, tumor size, S-phase fraction, and tamoxifen vival (P .001).
treatment demonstrated the 2-gene ratio to be a The HOXB13:IL17BR (H/I) index uses forma-
signicant predictor of RFS (HR, 3.9; 95% CI, 1.5 lin-xed, parafn-embedded tissue and is commer-
10.3 [P .007]). As was noted in the NCCTG cially available in the United States.
validation study,16 the ratio was a better indicator of
prognosis in the patients with lymph node-negative The 21-Gene RTPCR Assay (Oncotype DX)
disease than in those with lymph node-positive dis- Prognostic Data
ease. Although to our knowledge the mechanism for Table 2 shows prognostic data.
this disparity is unclear, the authors noted that lymph Although the signatures based on DNA arrays (eg,
node-positive tumors tend to have a higher HOXB13: the 70-gene assay) have prognostic value, their clinical
IL17BR expression ratio than lymph node-negative applicability has been limited by the need for fresh-
tumors.17 frozen tissue. In an attempt to circumvent this issue,
Jansen et al evaluated the HOXB13:IL17BR expres- Cronin et al developed a real-time RT-PCR method to
sion ratio in 1,252 patients with operable breast cancer quantify gene expression in sections of xed, parafn-
and demonstrated that the ratio was associated with embedded tumor tissue.19 By using the published liter-
both tumor aggressiveness as well as the likelihood of ature and genomic databases, as well as experiments
tamoxifen failure.18 A total of 468 patients with ER- based on DNA arrays in fresh-frozen tissue, 250 can-
positive, primary breast cancer were analyzed, 217 didate genes were selected3,8,20,21 and their correlation
(46%) of whom relapsed during the follow-up period. with breast cancer recurrence was examined in 3 inde-
The HOXB13:IL17BR expression ratio was found to pendent clinical breast cancer trials with a combined
be signicantly associated with a poor DFS and OS. total of 447 patients.22,23 Samples from these 3 trials
Expression levels were also evaluated in 193 patients were used to select a panel of 16 cancer-related and 5
with ER-positive, primary breast cancer who relapsed reference genes (Table 3),24 26 and an algorithm based

VOLUME 59 NUMBER 5 SEPTEMBER/OCTOBER 2009 307

 New Molecular Classications of Breast Cancer

TABLE 3. Panel of 21 Genes

PROLIFERATION OTHER CANCER-
GENES INVASION GENES HER2 ESTROGEN RELATED GENES REFERENCE GENES

Ki-67 MMP11 (stromolysin 3) GRB7 ER GSTM1 ACTB (-actin)

STK15 CTSL2 (cathepsin L2) HER2 PR CD68 GAPDH
Survivin BCL2 BAG1 RPLPO
CCNB1 (cyclin B1) SCUBE2 GUS
MYBL2 TFRC
MYBL2 indicates v-myb myeloblastosis viral oncogene homolog (avian)-like 2; MMP11, matrix metalloproteinase 11; GRB7, growth factor receptor-bound protein 7;
HER2, human epidermal growth factor receptor 2; ER, estrogen receptor; PR, progesterone receptor; GSTM1, glutathione S-transferase Mu 1; GAPDH, glyceraldehyde
3-phosphate dehydrogenase; RPLPO, ribosomal large protein; TFRC, transferrin receptor (p90, CD71).

Downloaded from caonline.amcancersoc.org by on June 13, 2010 (American Cancer Society, Inc.)
on the expression of these genes was devised to compute controls were patients with breast cancer who were
a recurrence score (RS) for each tumor sample. individually matched to cases with regard to age,
The 21-gene RTPCR assay and the RS algorithm race, adjuvant tamoxifen use, medical facility, and
were validated in a population of patients with lymph year of diagnosis and who were alive at the date of
node-negative disease who were treated with tamoxifen death of their matched case. After adjustment for
on a large, multicenter trial, the National Surgical Ad- grade and tumor size, the RS was found to be asso-
juvant Breast and Bowel Project Protocol B14 ciated with the risk of breast cancer death in patients
(NSABP B14) trial.24 NSABP B14 randomized with ER-positive disease who were treated and those
2,892 patients to receive either 5 years of treatment with not treated with tamoxifen. The risks of death from
tamoxifen or placebo and enrolled an additional 1,235 breast cancer at 10 years in the patients treated with
patients to receive an additional 5 years of tamoxifen tamoxifen for the low-risk, intermediate-risk, and
treatment. Parafn blocks with sufcient tumor tissue high-risk groups were 2.8% (95% CI, 1.73.9%),
were available for 675 of the 2,617 patients treated with
10.7% (95% CI, 6.314.9%), and 15.5% (95% CI,
tamoxifen. RTPCR was successful in 668 of the 675
7.622.8%), respectively. In the patients not treated
samples. The expression levels of the 21 genes were
with tamoxifen, these risks were 6.2% (95% CI, 4.5
used to calculate an RS and assign each patient to either
7.9%), 17.8% (95% CI, 11.823.3%), and 19.9%
a low-risk (RS 18), intermediate-risk (RS, 18 30), or
(95% CI, 14.225.2%), respectively. As was observed
high-risk (RS 31) group. The percentage of patients
in the NSABP B14 trial, approximately half of the
assigned to the low-risk, intermediate-risk, and high-
patients had a low-risk RS.
risk RS groups were 51%, 22%, and 27%, respectively.
We believe that the data presented above validate
The KaplanMeier estimates of the distant recurrence
the use of the 21-gene assay in patients with ER-
rate at 10 years were 6.8% (95% CI, 4.0 9.6%) in the
positive, lymph node-negative disease. Esteva et al
low-risk group, 14.3% (95% CI, 8.320.3%) in the
evaluated the assay in a population of patients with
intermediate-risk group, and 30.5% (95% CI, 23.6
lymph node-negative breast cancer who were treated
37.4%) in the high-risk group. The difference in the
at The University of Texas M. D. Anderson Cancer
distant recurrence rate between the low-risk and high-
Center and who did not receive adjuvant chemother-
risk groups was statistically signicant (P .001). The
apy and had been followed for a minimum of 5
RS was also found to be predictive of OS (P .001). In
years.26 Of the 149 eligible patients, 69% had tumors
a multivariate Cox model, the RS was found to be a
that were ER positive. In this mixed group of pa-
signicant predictor of distant recurrence independent
tients, in terms of hormonal status, the RS was not
of age and tumor size (P .001).
found to be predictive of distant disease recurrence.
The results from the NSABP B14 trial were
independently conrmed in a community hospital Predictive Data
setting.25 A case-control study was performed among The ability of gene expression proling assays to predict
4,964 patients from Kaiser Permanente who were benet from chemotherapy in the neoadjuvant as well
diagnosed between 1985 and 1994 and not treated as adjuvant setting also has been evaluated (Table 4).
with adjuvant chemotherapy. The 220 cases were Gianni et al evaluated the assay in 89 patients receiving
patients who died from breast cancer and the 570 neoadjuvant chemotherapy with three 3-week cycles of

308 CA: A Cancer Journal for Clinicians

 CA CANCER J CLIN 2009;59:303313

TABLE 4. Validation Studies for Predictive Value of the 21-Gene Recurrence Score Assay
STUDY POPULATION RESULTS

Gianni 2005 27 89 patients with locally advanced breast cancer who RS was positively associated with the likelihood of
were receiving neoadjuvant paclitaxel and achieving a pCR (P .005).
doxorubicin.
NSABP B-20 (Paik 2006)28 2,299 women with ER-positive, lymph node-negative Signicant interaction between RS and
breast cancer randomized to receive tamoxifen alone vs chemotherapy benet.
tamoxifen MF vs tamoxifen CMF. High RS: RR, 0.26; mean absolute decrease in
Gene expression results were available in 651 patients. 10-year distant recurrence rate, 27.6%
Low RS: RR, 1.31; mean absolute decrease in
10-year distant recurrence rate, 1.1%
Estimates in the intermediate RS group were too

Downloaded from caonline.amcancersoc.org by on June 13, 2010 (American Cancer Society, Inc.)
uncertain to exclude a clinically signicant benet.
RS indicates recurrence score; pCR, pathologic complete response; NSABP B-20, National Surgical Adjuvant Breast and Bowel Project B-20 trial; ER, estrogen receptor;
MF, methotrexate and uorouracil; CMF, cyclophosphamide, methotrexate, and uorouracil; RR, relative risk.

doxorubicin (60 mg/m2) and paclitaxel (200 mg/m2) RS did not appear to have a large benet, the esti-
followed by 12 weeks of weekly paclitaxel (80 mg/m2) mates were too uncertain to exclude a clinically sig-
(Table 4).27 Adjuvant cyclophosphamide, methotrexate, nicant benet.
and uorouracil (CMF) were administered after sur-
gery.27 RNA was extracted from pretreatment, forma- Lymph Node-Positive Patients and
lin-xed, parafn-embedded core needle biopsies. Us- Patients Treated with an Aromatase
ing RTPCR, the expression of 384 genes was
Inhibitor
quantied and correlated with pCR. Eighty-six genes
were found to correlate with a pCR and a pCR was To our knowledge, the 21-gene assay has been most
more likely to occur with a higher expression of prolif- extensively validated in women with lymph node-
eration-related and immune-related genes and with a negative disease who were treated with tamoxifen.
lower expression of ER-related genes. The RS, calcu- However, data are emerging regarding the prognostic
lated from the 21-gene assay, was found to be positively and predictive value of the assay in women with
associated with the likelihood of achieving a pCR lymph node-positive disease and in women treated
(P .005).27 with an aromatase inhibitor (AI). Goldstein et al29
The 21-gene assay has been evaluated for its ability evaluated the prognostic value of the assay in a group
to predict adjuvant chemotherapy benet using the of patients with ER-positive, early stage breast can-
NSABP B20 trial (Table 4).28 The NSABP B20 cer, all of whom received adjuvant chemotherapy.
trial evaluated the benet of adding CMF or meth- The Eastern Cooperative Oncology Group (ECOG)
otrexate and uorouracil to treatment with 5 years of study E2197 randomized 2,885 patient with operable
tamoxifen in women with lymph node-negative, ER- breast cancer and 0 to 3 positive lymph nodes to four
positive breast cancer. Of 2,299 eligible patients, 3-week cycles of doxorubicin (60 mg/m2) plus cyclo-
blocks with sufcient tumor tissue were available for phosphamide (600 mg/m2) or docetaxel (60 mg/m2)
670 patients, from which gene expression results plus cyclophosphamide (600 mg/m2). After chemo-
were obtained in 651. There was a signicant inter- therapy, the patients with ER-positive disease re-
action noted between RS and benet from chemo- ceived 5 years of tamoxifen treatment, although the
therapy. Patients with a high RS were found to have trial was subsequently amended to allow the use of
a large benet from chemotherapy (relative risk AIs. With a median follow-up of 76 months, there
[RR], 0.26; 95% CI, 0.13 0.53 [mean absolute de- was no signicant difference noted between the 2
crease in the 10-year distant recurrence rate, 27.6%]) treatment arms with regard to either DFS or OS.29 A
whereas patients with a low RS derived little or no sample of 465 patients with ER-positive tumors with
benet (RR, 1.31; 95% CI, 0.46 3.78 [mean abso- available tissue underwent the 21-gene assay, and
lute decrease in the 10-year distant recurrence rate, also had their recurrence risk estimated by Adjuvant!
1.1%]). Although the patients with an intermediate Online.30 The 5-year recurrence estimates were com-

VOLUME 59 NUMBER 5 SEPTEMBER/OCTOBER 2009 309

 New Molecular Classications of Breast Cancer

puted by Adjuvant! Online and patients were classied risk group, and 43% for tamoxifen alone versus 55%
as being at low, intermediate, or high risk using the for CAFT in the high-risk group.
previously dened criteria. The prognostic utility of RS The majority of the data regarding the 21-gene
was evaluated for each Adjuvant! Online risk group. assay in patients with ER-positive disease have been
Similar to data previously reported in patients with derived from patients treated with tamoxifen. How-
lymph node-negative disease, 46% of the patients ever, currently, tamoxifen is not the only drug avail-
had a low-risk RS, 30% had an intermediate-risk RS, able for the adjuvant treatment of postmenopausal
and 24% had a high-risk RS. The RS was found to be women with early stage breast cancer and many pa-
a highly signicant predictor of local as well as dis- tients are in fact receiving an AI instead. The ability
of the 21-gene assay to predict the risk of distant

Downloaded from caonline.amcancersoc.org by on June 13, 2010 (American Cancer Society, Inc.)
tant recurrence in both the patients with lymph
node-negative (P .0007) and lymph node-positive recurrence in postmenopausal women receiving an
(P .0004) disease. Furthermore, a low RS pre- adjuvant AI has been evaluated in the TransATAC
dicted a low risk of recurrence (5%), irrespective of analysis of the Arimidex, Tamoxifen, Alone or in
lymph node status. The RS provided additional Combination (ATAC) trial.32 The ATAC trial ran-
prognostic information to Adjuvant! Online, partic- domized 9,366 patients with early stage breast cancer
ularly with regard to those patients projected to have to 5 years of treatment with tamoxifen, 5 years of
treatment with anastrazole, or 5 years of treatment
better outcomes.
with both tamoxifen and anastrazole. Of the 9,366
All the patients in the ECOG study E2197 re-
women, 3,486 were either negative for ER or were
ceived chemotherapy. However, Albain et al evalu-
randomized to the combination arm and therefore
ated the 21-gene assay in patients with lymph node-
were not included in the TransATAC analysis. Of
positive, ER-positive disease who were treated with
the remaining 5,880 patients who were eligible for
adjuvant tamoxifen alone.31 The Southwest Oncol-
the TransATAC analysis, blocks were available with
ogy Group Intergroup Trial S8814 was a phase 3 trial sufcient tumor in 1,856 patients and a reportable
of postmenopausal women with lymph node-posi- RS was obtained in 1,308 patients, of whom 1,231
tive, ER-positive breast cancer that demonstrated were evaluable. In the prospectively dened, primary,
that the addition of 6 cycles of cyclophosphamide, multivariate analysis, tumor size, tumor grade, and
doxorubicin, and uorouracil (CAF) added a signif- RS were each found to be separately statistically
icant benet with regard to DFS and OS compared signicant in predicting time to distant recurrence in
with tamoxifen alone, particularly if CAF and ta- patients with lymph node-negative disease (P
moxifen were administered sequentially (CAFT). .001, P .003, and P .001, respectively) with
Of the 927 patients randomized to receive either similar results observed in patients with lymph node-
tamoxifen alone or CAFT, 45% provided speci- positive disease. For the patients with lymph node-
mens, with 367 patients (148 treated with tamoxifen negative disease, the 9-year distant recurrence rates
alone and 219 treated with CAFT) found to have for the low-risk, intermediate-risk, and high-risk RS
sufcient RNA for RTPCR analysis. The RS risk groups were 4%, 12%, and 25%, respectively; and
distribution was somewhat different from that noted those for the patients with lymph node-positive dis-
in patients with lymph node-negative disease: 40% in ease were 17%, 28%, and 49%, respectively. The RS
the low-risk group, 28% in the intermediate-risk demonstrated statistically signicant prognostic value
group, and 32% in the high-risk group. The RS was beyond that provided by Adjuvant! Online with re-
found to be prognostic for DFS and OS in the gard to both lymph node-negative (P . 001) and
patients treated with tamoxifen alone (P .006). lymph node-positive patients (P .003). The data
There was a large benet noted for CAFT com- were not predictive of a differential benet between
pared with tamoxifen alone in the high-risk RS sub- tamoxifen and anastrazole.
set but no apparent benet was observed in the
low-risk RS group. The 10-year DFS estimates (95%
CI) were 60% for tamoxifen alone versus 64% for
Comparison of Gene Expression Assays
CAFT in the low-risk group, 49% for tamoxifen Several of the gene expression assays discussed have
alone versus 63% for CAFT in the intermediate- been compared. Fan et al used 295 samples to com-

310 CA: A Cancer Journal for Clinicians

 CA CANCER J CLIN 2009;59:303313

pare predictions from 5 gene expression assays: in- patients being reassigned to a different risk classi-
trinsic subtypes, the 70-gene prole, the 2-gene cation and chemotherapy was avoided in 9% of the
HOXB13:IL17BR expression ratio, the 21-gene RS patients. In the base case, the 70-gene signature
assay, and the wound response assay.33 The wound strategy was found to be cost-neutral. Lifetime costs
response signature, identied by testing the correla- per patient were $178,811 and $178,893, respec-
tion between tumor progression and a gene expres- tively, for the 70-gene assay and Adjuvant! Online
sion program identied in an experimental wound strategies. Use of the 70-gene assay was associated
response model, was previously validated in 295 con- with an increase of 0.13 life-years and 0.16 quality-
secutive patients with early stage breast cancer.34,35 adjusted life-years.

Downloaded from caonline.amcancersoc.org by on June 13, 2010 (American Cancer Society, Inc.)
Despite the absence of gene overlap, the assays, with
the exception of the HOXB13:IL17BR expression Current Use of the Assays
ratio, demonstrated high concordance rates in pre- Currently, the 70-gene assay and the 21-gene RS assay
dicting outcome, suggesting that the assays identify are the most commonly used genomic proling assays
common biologic characteristics that are predictive of in Europe and the United States. MammaPrint was the
patient outcomes. rst assay in the United States to receive US Food and
Drug Administration (FDA) approval under the FDAs
Cost-Effectiveness of the Assays new, in vitro, diagnostic, multivariate index assay clas-
The emergence of commercially available gene pro- sication as a prognostic test for women aged younger
ling assays has raised the question of the cost- than 61 years with ER-positive or ER-negative, lymph
effectiveness of these techniques. Cost-effectiveness node-negative breast cancer. Oncotype DX has been
analyses comparing 2 of the commercially available exempt from this approval process.
assays in the United States, the 21-gene RS assay and In the United States, Oncotype DX is currently the
the 70-gene assay, with other methods of assessment most commonly used assay in clinical practice for a
have been performed recently. variety of reasons, including the nding that it can be
A cost-utility analysis was conducted using the 21- performed on formalin-xed, parafn-embedded tis-
gene RS assay in patients previously classied as having sue. It is unclear at the present time how the need for
a low or high risk of distant recurrence based on clinical fresh tissue will affect the adoption of MammaPrint
guidelines published by the National Comprehensive in the United States. In addition, the use of the
Cancer Network (NCCN).36 The cost of the assay was Oncotype DX assay to predict the risk of recurrence
estimated at $3,460. The analysis demonstrated that and the benets of tamoxifen and CMF chemother-
using the assay to guide chemotherapy decisions pro- apy in newly diagnosed patients with lymph node-
vided a net savings of $2,256 compared with chemo- negative, ER-positive breast cancer are included in
therapy and tamoxifen, with an incremental cost-effec- the 2007 American Society of Clinical Oncology
tiveness ratio of $1,944 per life saved with treatment (ASCO) tumor marker guidelines.38 The ASCO
with tamoxifen alone. Furthermore, using the assay to panel believed that the precise clinical utility and
guide therapy was associated with a gain in individual appropriate application of other assays (eg, the 70-
life expectancy of 2.2 years compared with tamoxifen gene or 76-gene assays) were under investigation.
alone and a similar life expectancy compared with the However, there are limitations to the use of the
use of tamoxifen and chemotherapy. Oncotype DX assay, including the lack of a data-
A similar analysis estimating the costs and cost- driven answer regarding the optimal treatment of
effectiveness of the 70-gene assay versus Adjuvant! patients with an intermediate-risk RS. In addition,
Online in deciding whether to use adjuvant chemo- use of the Oncotype DX assay is limited to patients
therapy for women aged 61 years and younger with with ER-positive disease, unlike other assays (includ-
lymph node-negative, HER2-negative, early stage ing the MammaPrint assay), which have been vali-
breast cancer with ER-positive or negative status was dated in patients with both ER-positive and ER-
recently reported.37 Compared with Adjuvant! On- negative disease. Currently, both assays cost
line, using the 70-gene assay resulted in 35% of approximately $3,000 to $4,000 in the United States.

VOLUME 59 NUMBER 5 SEPTEMBER/OCTOBER 2009 311

 New Molecular Classications of Breast Cancer

Future Directions both methods classify the relapse risk as high, che-
motherapy will be administered. If the methods are
As a result of gene expression assays, there is growing
discordant, the patient will be randomly assigned to
recognition that breast cancer is a molecularly heter-
follow the results of Adjuvant! Online or the 70-gene
ogeneous disease. However, there are multiple unre-
assay.39 In North America, the 21-gene assay is cur-
solved issues with regard to the adoption of these
rently being evaluated in a prospective clinical fash-
assays. For example, there are very few data regarding
ion in the Trial Assigning IndividuaLized Options
the biologic reproducibility of these assays; the effect
for Treatment (Rx) (TAILORx) trial. Patients with
of variable tumor cellularity as well as intratumoral
lymph node-negative, ER-positive breast cancer will
heterogeneity on the molecular classication; and the

Downloaded from caonline.amcancersoc.org by on June 13, 2010 (American Cancer Society, Inc.)
be divided into 3 treatment arms depending on their
potential for contamination by normal breast tissue
RS. However, the RS categories are different from
or in situ carcinomas, particularly in small invasive
those previously validated. A low-risk RS on the
tumors. Similarly, the effect of a prior biopsy on gene
TAILORx trial is 11, an intermediate-risk RS is
expression results has to our knowledge been under-
between 11 and 25, and a high-risk RS is 25. The
explored. Furthermore, the usefulness of these assays
purpose of these adjustments was to minimize the
in other clinical settings (eg, in patients with locally
potential for undertreatment in the intermediate-risk
advanced or metastatic breast cancer) has not been
and high-risk groups.40 Patients with a low-risk RS
adequately examined.
will receive endocrine therapy without chemother-
We await further data to clarify the optimal use of
apy, patients with a high-risk RS will receive chemo-
these assays, particularly prospective, randomized therapy followed by endocrine therapy, and patients
data. In Europe, the 70-gene assay is currently being in the intermediate-risk RS category will be random-
evaluated in a prospective clinical fashion in the Eu- ized to receive either endocrine therapy without che-
ropean Organization for Research and Treatment of motherapy or chemotherapy followed by endocrine
Cancer (EORTC) Microarray In Node-negative and therapy. The choice of chemotherapy regimen and
1 to 3 positive lymph node Disease may Avoid Che- endocrine therapy (tamoxifen or an AI) will be at the
moTherapy (MINDACT) trial.39 This trial aims to discretion of the treating physician. Exclusion criteria
enroll 6,000 patients with lymph node-negative include HER2-positive tumors. These trials, along
breast cancer who will have their risk assessed both with the incorporation of tissue collection and
by Adjuvant! Online and the 70-gene prole. If both genomic proling into general clinical trial design,
methodologies assess the patient as having a low will improve our ability to optimally tailor therapy for
relapse risk, no chemotherapy will be administered. If individual patients.

across distinct ethnic populations. Clin prognostic signature for women with node-
References Cancer Res. 2004;10:5508 5517. negative breast cancer. J Natl Cancer Inst.
1. Ravdin PM, Siminoff LA, Davis GJ, et al. 6. Weigelt B, Hu Z, He X, et al. Molecular 2006;98:11831192.
Computer program to assist in making portraits and 70-gene prognosis signature 11. Bender RA, Knauer M, Rutgers EJ, et al.
decisions about adjuvant therapy for women are preserved throughout the metastatic The 70-gene prole and chemotherapy
with early breast cancer. J Clin Oncol. process of breast cancer. Cancer Res. 2005; benet in 1600 breast cancer patients
2001;29:980 991. 65:91559158. [abstract]. J Clin Oncol. 2009;27(suppl):
2. Olivotto IA, Bajdik CD, Ravdin PM, et al. 7. Parker JS, Mullins M, Cheang MCU, et al. 15s. Abstract 512.
Population-based validation of the prognos- Supervised risk predictor of breast cancer 12. Wang Y, Klijn JGM, Sieuwerts AM, et al.
tic model ADJUVANT! for early stage breast based on intrinsic subtypes. J Clin Oncol. Gene-expression proles to predict dis-
cancer. J Clin Oncol. 2005;23:2716 2725. tant metastasis of lymph node-negative
2009;27:1160 1167.
3. Perou CM, Sorlie T, Eisen MB, et al. primary breast cancer. Lancet. 2005;365:
Molecular portraits of human breast tu- 8. vant Veer LJ, Dai H, van de Vijver MJ, et al. 671 679.
mours. Nature. 2000;406:747752. Gene expression proling predicts clinical
outcome of breast cancer. Nature. 2002;415: 13. Foekens JA, Atkins D, Zhang Yi, et al.
4. Sorlie T, Perou AM, Tibshirani R, et al. 530 536. Multicenter validation of a gene expression-
Gene expression patterns of breast carcino- based prognostic signature in lymph node-
mas distinguish tumor subclasses with 9. van de Vijver, He Y, vant Veer LJ, et al. A negative primary breast cancer. J Clin
clinical implications. Proc Natl Acad Sci U S gene-expression signature as a predictor of Oncol. 2006;24:16651671.
A. 2001;98:10869 10874. survival in breast cancer. N Engl J Med. 14. Ma XJ, Wang Z, Ryan PD, et al. A two-gene
5. Yu K, Lee CH, Tan PH, et al. Conservation of 2002;347:1999 2009. expression ratio predicts clinical outcome
breast cancer molecular subtypes and tran- 10. Buyse M, Loi S, vant Veer L, et al. in breast cancer patients treated with tamox-
scriptional patterns of tumor progression Validation and clinical utility of a 70-gene ifen. Cancer Cell. 2004;5:607 616.

312 CA: A Cancer Journal for Clinicians

 CA CANCER J CLIN 2009;59:303313

15. Jerevall PL, Brommesson S, Strand C, et al. 24. Paik S, Shak S, Tang G, et al. A multigene 32. Dowsett M, Cuzick J, Wales C, et al. Risk of
Exploring the two-gene ratio in breast assay to predict recurrence of tamoxifen- distant recurrence using oncotype DX in
cancer-independent roles for HOXB13 and treated, node-negative breast cancer. N Engl postmenopausal primary breast cancer pa-
IL 17BR in prediction of clinical outcome. J Med. 2004;351:28172826. tients treated with anastrozole or tamox-
Breast Cancer Res Treat. 2008;107:225234. ifen: a TransATAC study [abstract]. Cancer
25. Habel LA, Shak S, Jacobs MK, et al. A
16. Goetz MP, Suman VJ, Ingle JN, et al. A population-based study of tumor gene ex- Res 2009;69(suppl). Abstract 53.
two-gene expression ratio of homeobox 13 pression and risk of breast cancer death 33. Fan C, Oh DS, Wessels L, et al. Concordance
and interleukin17B receptor for prediction among lymph node-negative patients. among gene-expression-based predictors for
of recurrence and survival in women receiv-
Breast Cancer Res. 2006;8:115. breast cancer. N Engl J Med. 2006;355:
ing adjuvant tamoxifen. Clin Cancer Res.
2006;12:2080 2087. 26. Esteva FJ, Sahin AA, Cristofanilli M, et al. 560 569.
17. Ma XJ, Hilsenback SG, Wang W, et al. The Prognostic role of a multigene reverse 34. Chang HY, Sneddon JB, Alizadeh AA, et al.
HOXB13:IL17BR expression index is a prog- transcriptase-PCR assay in patients with Gene expression signature of broblast
nostic factor in early-stage breast cancer. node-negative breast cancer not receiving serum response predicts human cancer
J Clin Oncol. 2006;24:4611 4617. adjuvant systemic therapy. Clin Cancer Res. progression: similarities between tumors
2005;11:33153319. and wounds. PLoS Biol. 2004;2:E7.

Downloaded from caonline.amcancersoc.org by on June 13, 2010 (American Cancer Society, Inc.)
18. Jansen MP, Siewerts AM, Look MP, et al.
HOXB13toIL17BR expression ratio is re- 27. Gianni L, Zambetti M, Clark K, et al. Gene 35. Chang HY, Nuyten DSA, Sneddon JB, et al.
lated to tumor aggressiveness and response expression proles in parafn-embedded Robustness, scalability, and integration of a
to tamoxifen of recurrent breast cancer: a core biopsy tissue predict response to che- wound-response gene expression signature
retrospective study. J Clin Oncol. 2007;25: motherapy in women with locally ad- in predicting breast cancer survival. Proc
662 668. vanced breast cancer. J Clin Oncol. 2005;23: Natl Acad Sci U S A. 2005;102:3738 3743.
19. Cronin M, Pho M, Dutta D, et al. Measure- 72657277. 36. Lyman GH, Cosler LE, Kuderer NM, et al.
ment of gene expression in archival parafn- 28. Paik S, Tang G, Shak S, et al. Gene Impact of a 21-gene RTPCR assay on
embedded tissues: development and perfor- expression and benet of chemotherapy in treatment decisions in early-stage breast
mance of a 92-gene reverse transcriptase- women with node-negative, estrogen recep- cancer: an economic analysis based on
polymerase chain reaction assay. Am J tor-positive breast cancer. J Clin Oncol. prognostic and predictive validation stud-
Pathol. 2004;164:35 42. 2006;24:3726 3734. ies. Cancer. 2007;109:10111018.
20. Golub TR, Slonim DK, Tamayo P, et al. 29. Goldstein LJ, ONeill A, Sparano JA, et al.
Molecular classication of cancer: class 37. Tong KB, Chen E, Brink G, et al. Cost-
Concurrent doxorubicin plus docetaxel is effectiveness of targeting chemotherapy
discovery and class prediction by gene not more effective than concurrent doxoru-
expression monitoring. Science. 1999;286: with the 70-gene prognostic signature in
bicin plus cyclophosphamide in operable early-stage breast cancer (ESBC) patients
531537. breast cancer with 0 to 3 positive axillary
[abstract]. J Clin Oncol. 2009;27(suppl):
21. Sorlie T, Perou CM, Tibshirani R, et al. Gene nodes: North American Breast Cancer Inter-
expression patterns of breast carcinomas group Trial E 2197. J Clin Oncol. 2008;26: 15s. Abstract 6570.
distinguish tumor subclasses with clinical 4092 4099. 38. Harris L, Fritsche H, Mennel R, et al.
implications. Proc Natl Acad Sci U S A. 30. Goldstein L, Gray R, Badve S, et al. American Society of Clinical Oncology 2007
2001;98:10869 10874. Prognostic utility of the 21-gene assay in update of recommendations for the use of
22. Paik S, Shak S, Tang G, et al. Multi-gene hormone receptor-positive operable breast tumor markers in breast cancer. J Clin
RTPCR assay for predicting recurrence in cancer compared with classical clinico- Oncol. 2007;25:52875312.
node negative breast cancer patientsNSABP pathologic features. J Clin Oncol. 2008;26: 39. Cardoso F, vant Veer L, Rutgers E, et al.
studies B20 and B14 [abstract]. Proc Am 4063 4071. Clinical application of the 70-gene prole:
Soc Clin Oncol. 2003;22:850. 31. Albain K, Barlow W, Shak S, et al. Prognos- the MINDACT trial. J Clin Oncol. 2008;26:
23. Cobleigh MA, Tabesh B, Bitterman P, et al. tic and predictive value of the 21-gene 729 735.
Tumor gene expression and prognosis in recurrence score assay in postmenopausal, 40. Sparano JA, Paik S. Development of the
breast cancer patients with 10 or more node-positive, ER-positive breast cancer 21-gene assay and its application in clinical
positive lymph nodes. Clin Cancer Res. (S8814, INT0100) [abstract]. Br Cancer Res practice and clinical trials. J Clin Oncol.
2005;11:8623 8631. Treat 2007;106(suppl 1). Abstract 10. 2008;26:721728.

VOLUME 59 NUMBER 5 SEPTEMBER/OCTOBER 2009 313