Purpose: Most node-negative breast cancer patients are older and postmenopausal and are increasingly being offered adjuvant chemotherapy despite their low overall risk of distant relapse. A molecular diagnostic test with high negative predictive value (NPV) for distant metastasis in this subgroup would spare many older breast cancer patients adjuvant treatment.

Experimental Design: We determined the NPV and positive predictive value of the MammaPrint assay in breast cancer patients who were consecutively diagnosed and treated at the Massachusetts General Hospital between 1985 and 1997. Primary tumors from 100 patients with node-negative, invasive breast cancer (median age, 62.5 years; median follow-up, 11.3 years) were subjected to MammaPrint analysis and classified as being at either low or high risk for distant metastasis.

Results: The MammaPrint 70-gene signature displayed excellent NPV as in previous studies, correctly identifying 100% of women at low risk for distant metastases at 5 years. However, this assay had a lower positive predictive value (12% at 5 years) than previously observed.

Conclusions: The MammaPrint assay was originally designed to identify younger breast cancer patients at low risk for distant metastasis, who might consequently be spared systemic treatment. We show here that the same signature has a very high NPV for distant recurrence after adjuvant treatment in older breast cancer patients.

Breast cancer is clinically heterogeneous. Patients differ widely with respect to natural history and response to treatment (1). This clinical heterogeneity is probably due to the varying mutational spectrum or cell type of origin of tumors and complex genetic differences among individuals (2). These factors in combination influence the expression of many genes involved in tumor growth, invasion, metastasis, and survival. Although consensus criteria, based on clinical and histopathologic features [age, tumor size, histologic tumor type, pathologic grade, estrogen receptor (ER) status, and axillary lymph node status], are currently used to assess risk of distant relapse in patients with breast cancer, these clinical measures are imperfect (3, 4). A large number of additional factors have been evaluated for determining likely prognosis or treatment response, but most have limited power and few reflect the genetics of the disease (5).

DNA microarrays have recently been used to obtain genome-wide views of human tumor gene expression, and these studies have identified cancer biomarkers with diagnostic, prognostic, and predictive potential in a wide variety of solid tumors (6). Breast cancer has proven particularly fertile ground for microarray-based biomarker discovery, and recent studies have suggested that the clinical behavior of a patient's cancer is encoded in the gene expression profile of their primary tumor (79).

Members of our group initially reported a 70-gene prognostic microarray expression signature in primary tumors from patients with breast cancer who were diagnosed and treated between 1983 and 1996 at the Netherlands Cancer Institute (NKI; ref. 9). Subsequently, retrospective validation studies showed that this 70-gene expression signature indeed can be used to classify younger patients, with either node-positive or node-negative disease, into groups with significantly different probabilities of remaining metastasis-free (10, 11). Moreover, the prognostic value of the 70-gene signature is additive to currently used St. Gallen (3) or NIH (4) consensus criteria for assessing risk of distant relapse. These initial observations suggest that the 70-gene signature is a powerful predictor of clinical outcome in younger women with early-stage breast cancer and that clinically defined high-risk patients with microarray-defined “good prognosis” disease might actually be at low risk for developing distant metastases. In theory, these patients may be spared adjuvant chemotherapy with its associated toxicity risk. These issues are currently being addressed in a prospective, multicenter, clinical trial in Europe (12).

Most patients with breast cancer, however, are older than the cohorts used to define and evaluate the 70-gene signature and usually present with smaller, early-stage, ER+ tumors that tend to metastasize less frequently. In addition, such patients often have comorbid disease that further complicates decisions about the use of adjuvant chemotherapy. Despite this concern, many older patients are increasingly being offered adjuvant chemotherapy. A molecular test that can accurately identify older patients at low risk of distant metastasis, who could then be spared chemotherapy toxicity, might therefore have clinical value in selected circumstances. To explore these issues, we did a retrospective evaluation of the 70-gene MammaPrint assay in 100 older patients with node-negative breast cancer who were initially diagnosed and treated at the Massachusetts General Hospital (MGH) between 1985 and 1997.

Study design. Study design, patient selection, histopathologic analysis of tumors, clinical annotation, clinical interpretation, RNA isolation, microarray profiling, and statistical analysis were carried out jointly at the MGH, the NKI, and Agendia.

Patient selection. We first identified those patients for whom frozen primary tumor material was available in the MGH Department of Pathology Breast Tumor Bank. We next identified patients who had been consecutively diagnosed and treated for lymph node–negative (pN0), invasive breast cancer at the MGH between 1985 and 1997 and for whom histopathologic and clinical information could be retrieved from the medical record (through September 2007). This information was obtained under a protocol approved by the MGH Institutional Review Board in accordance with federal human research study guidelines. All clinical and histopathologic data were stripped of personal identifiers. Information was collected on age and calendar year of diagnosis, surgery, tumor (size, grade, histologic type, and ER status), nodal status, radiation treatment, hormonal therapy or chemotherapy, and clinical follow-up, including local, locoregional, or distant recurrences, second primary malignancies, and death or date of last visit. Patients with prior malignancies (except nonmelanoma skin cancers) and those with questionable diagnoses about site of tumor origin were excluded. Tumors with >50% tumor cellularity were subjected to RNA isolation and microarray profiling (n = 100). Raw microarray data underwent further statistical analysis.

All patients in this study cohort had been treated with mastectomy or breast-conserving surgery, including axillary lymph node dissection, followed by radiotherapy if indicated. A proportion received adjuvant systemic therapy consisting of chemotherapy, hormonal therapy, or both. The median duration of follow-up was 11.8 y (range, 1.2-18.5) for the 91 patients without metastasis as a first event and 4.9 y (range, 1.6-13.0) for the 9 patients with metastasis as a first event. The median follow-up among all 100 patients was 11.3 y.

Histopathologic evaluation and RNA isolation. Tumors had been snap frozen in liquid nitrogen within 1 h after surgery and subsequently stored at −80°C. Histopathologic evaluation of all tumors was done centrally in the Department of Pathology at the MGH (by D.C.S.) using established and previously published clinical criteria (13). H&E-stained sections of frozen tumors, obtained before and after the sections used for RNA isolation, were evaluated for percent tumor cellularity and samples with <50% were excluded. Histologic grade was assessed using modified Scarff-Bloom-Richardson criteria. Vascular invasion was assessed as absent or present. ER expression was estimated using ER transcript levels on the microarray for each tumor. Fifteen to thirty 30-μm sections were used for RNA isolation. Total RNA was isolated with RNAzol B and dissolved in RNase-free water. Twenty-five micrograms of total RNA were treated with DNase using the Qiagen RNase-free DNase kit and RNeasy spin columns for cleanup. RNA was then dissolved in RNase-free water to a final concentration of 0.2 μg/μL.

Microarray expression profiling. RNA labeling, microarray hybridization, and scanning were done at Agendia using standardized methods and protocols (14). Briefly, cRNA was generated by in vitro transcription with T7 RNA polymerase (Low Input Fluorescent Labeling kit, Agilent Technologies) and labeled with Cy3 or Cy5 (Cy Dye, Perkin-Elmer). Cy-labeled cRNA from one breast cancer tumor was mixed with the same amount of reverse-color Cy-labeled product from a standard “mamma-reference” pooled RNA control. Labeled cRNAs were hybridized to an eight-pack Agendia MammaPrint microarray using standard protocols (Agilent Oligo Microarray kit, Agilent Technologies). This microarray contains eight identical subarrays, each containing 60-mer probes for the 70 prognosis genes in triplicate. Each sample was hybridized twice to do dye swaps. Fluorescence intensities on scanned images were quantified, and the values were corrected for the background level and normalized. Standard quality control measures were applied to determine technically acceptable hybridizations and poor-quality hybridizations were repeated.

MammaPrint tumor classification. Tumor classification was done using the previously reported 70-gene classification model (9). Briefly, for each of the 100 tumors, we calculated the cosine correlation coefficient of the level of expression of the 70 prognosis genes with the previously determined average profile of these genes in breast tumors from patients with a good prognosis (9, 14). A patient with a correlation coefficient >0.4 (the threshold in the original NKI study of 78 tumors that resulted in a 10% rate of false-negative results) was then assigned to the “good” signature/low-risk group, and all others were assigned to the “poor” signature/high-risk group.

Adjuvant! Online tumor classification. Analysis of available clinicopathologic information for each of the 100 tumors with Adjuvant! Online v8.07

yielded population-based estimates of 10-y relapse risk for each individual MGH patient. Input variables included age, comorbidity (set to “average for age”), ER status, and tumor grade and size.

Statistical analysis. To determine the probability that patients would remain free of distant metastases, we defined distant metastases as a first event to be a treatment failure; data on all other patients were censored on the date of the last follow-up visit, local or regional disease recurrence, the development of a second primary cancer, including contralateral breast cancer, and death from causes other than breast cancer.

Data on patients were analyzed from the date of diagnosis to either the time of the first event or the date on which data were censored. Survival data were compared using the log-rank test using the exact form of the Mantel-Haenszel test. Confidence intervals for Kaplan-Meier survival curves were constructed using the log method and SE was calculated by the method of Tsiatis. Confidence intervals for predictive values were computed by the method of Clopper and Pearson. All P values characterizing differences between proportions were determined by Fisher's exact test. All calculations were done using the R statistical package.8

Clinicopathologic information and Adjuvant! Online and MammaPrint results are available as Supplementary Table S1.

The NKI lymph node–negative cohort referred to throughout included 151 consecutive patients originally reported in 2002 (11), which included 61 lymph node–negative patients originally reported separately in 2002 (9).

Clinical characteristics of the MGH cohort. The MGH cohort had a median age of 62.5 years, which was significantly older than previously studied patients (P < 0.001; Table 1; ref. 11). Statistical analysis revealed additional differences in tumor size, histologic grade, and treatment with both adjuvant chemotherapy and hormonal therapy between the 100 MGH and 151 NKI node-negative patients (P < 0.005), but censoring rates in the two cohorts did not differ appreciably. These differences were associated with a strikingly lower rate of distant metastasis as a first event in the MGH cohort compared with the NKI node-negative cohort (P < 0.001; Fig. 1). Notably, the MGH and NKI node-negative cohorts did not differ significantly in overall survival, despite the low metastasis rate in MGH patients. This was due to death from causes other than breast cancer in the older MGH population (data not shown).

Table 1.

Clinicopathologic features of MGH and node-negative NKI patients

CharacteristicsMGH, n (%)NKI, n (%)P
Age (y)    
    <40 4 (4) 36 (24) <0.001 
    40-44 6 (6) 42 (28)  
    45-49 14 (14) 49 (32)  
    50-54 7 (7) 24 (16)  
    ≥55 69 (69) 0 (0)  
Tumor size (cm)    
    ≤2 72 (72) 82 (54) 0.005 
    >2 28 (28) 69 (46)  
Histologic grade    
    I 5 (5) 34 (23) <0.001 
    II 54 (54) 46 (30)  
    III 40 (40) 71 (47)  
    Medullary carcinoma 1 (1) 0 (0)  
ER    
    Negative 20 (20) 42 (28) 0.180 
    Positive 80 (80) 109 (72)  
Surgery    
    Breast conserving 44 (44) 90 (60) 0.020 
    Mastectomy 56 (56) 61 (40)  
Chemotherapy    
    Yes 21 (21) 6 (4) <0.001 
    No 79 (79) 145 (96)  
Hormonal therapy    
    Yes 24 (24) 6 (4) <0.001 
    No 76 (76) 145 (96)  
CharacteristicsMGH, n (%)NKI, n (%)P
Age (y)    
    <40 4 (4) 36 (24) <0.001 
    40-44 6 (6) 42 (28)  
    45-49 14 (14) 49 (32)  
    50-54 7 (7) 24 (16)  
    ≥55 69 (69) 0 (0)  
Tumor size (cm)    
    ≤2 72 (72) 82 (54) 0.005 
    >2 28 (28) 69 (46)  
Histologic grade    
    I 5 (5) 34 (23) <0.001 
    II 54 (54) 46 (30)  
    III 40 (40) 71 (47)  
    Medullary carcinoma 1 (1) 0 (0)  
ER    
    Negative 20 (20) 42 (28) 0.180 
    Positive 80 (80) 109 (72)  
Surgery    
    Breast conserving 44 (44) 90 (60) 0.020 
    Mastectomy 56 (56) 61 (40)  
Chemotherapy    
    Yes 21 (21) 6 (4) <0.001 
    No 79 (79) 145 (96)  
Hormonal therapy    
    Yes 24 (24) 6 (4) <0.001 
    No 76 (76) 145 (96)  
Fig. 1.

Overall clinical outcome of node-negative MGH versus NKI patients. Kaplan-Meier analysis of time to metastasis as a first event for the MGH and node-negative NKI cohorts. Green, MGH; black, NKI node negative; thin lines, 95% confidence interval (95% CI); black lines, previously published data (11).

Fig. 1.

Overall clinical outcome of node-negative MGH versus NKI patients. Kaplan-Meier analysis of time to metastasis as a first event for the MGH and node-negative NKI cohorts. Green, MGH; black, NKI node negative; thin lines, 95% confidence interval (95% CI); black lines, previously published data (11).

Close modal

Primary tumor classification. We next applied the 70-gene predictive model to gene expression profiles from all 100 patients in the MGH cohort. This model classified 27 patients into low-risk group and 73 patients into high-risk group (Table 2). As observed in previous studies, prediction values (i.e., correlations) for these tumors fell along a continuum rather than into discrete pools, suggesting complex, nondiscrete differences in gene expression among individual tumors (data not shown; refs. 9, 11). Patients classified as low risk by tumor gene expression tended to have smaller, lower-grade, ER+ tumors as a group compared with high-risk patients in univariate analysis but, importantly, did not differ significantly in the use of adjuvant treatment.

Table 2.

Clinicopathologic features of MGH patients classified as either high or low risk for distant metastasis using the 70-gene signature

CharacteristicsHigh-risk signature (n = 73), n (%)Low-risk signature (n = 27), n (%)P
Age (y)    
    <40 4 (5) 0 (0) 0.548 
    40-44 5 (7) 1 (4)  
    45-49 8 (11) 6 (22)  
    50-54 5 (7) 2 (7)  
    ≥55 51 (70) 18 (67)  
Tumor size (cm)    
    ≤2 48 (66) 24 (89) 0.025 
    >2 25 (34) 3 (11)  
Histologic grade    
    I 2 (3) 3 (11) 0.011 
    II 35 (48) 19 (70)  
    III 35 (48) 5 (19)  
    Medullary carcinoma 1 (1) 0 (0)  
ER    
    Negative 20 (27) 0 (0) 0.001 
    Positive 53 (73) 27 (100)  
Surgery    
    Breast conserving 33 (45) 11 (41) 0.821 
    Mastectomy 40 (55) 16 (59)  
Chemotherapy    
    Yes 17 (23) 4 (15) 0.420 
    No 56 (77) 23 (85)  
Hormonal therapy    
    Yes 18 (25) 6 (22) 1.000 
    No 55 (75) 21 (78)  
Radiotherapy    
    Combination 1 (1) 1 (4) 0.221 
    External beam 30 (41) 7 (26)  
    None 42 (58) 19 (70)  
CharacteristicsHigh-risk signature (n = 73), n (%)Low-risk signature (n = 27), n (%)P
Age (y)    
    <40 4 (5) 0 (0) 0.548 
    40-44 5 (7) 1 (4)  
    45-49 8 (11) 6 (22)  
    50-54 5 (7) 2 (7)  
    ≥55 51 (70) 18 (67)  
Tumor size (cm)    
    ≤2 48 (66) 24 (89) 0.025 
    >2 25 (34) 3 (11)  
Histologic grade    
    I 2 (3) 3 (11) 0.011 
    II 35 (48) 19 (70)  
    III 35 (48) 5 (19)  
    Medullary carcinoma 1 (1) 0 (0)  
ER    
    Negative 20 (27) 0 (0) 0.001 
    Positive 53 (73) 27 (100)  
Surgery    
    Breast conserving 33 (45) 11 (41) 0.821 
    Mastectomy 40 (55) 16 (59)  
Chemotherapy    
    Yes 17 (23) 4 (15) 0.420 
    No 56 (77) 23 (85)  
Hormonal therapy    
    Yes 18 (25) 6 (22) 1.000 
    No 55 (75) 21 (78)  
Radiotherapy    
    Combination 1 (1) 1 (4) 0.221 
    External beam 30 (41) 7 (26)  
    None 42 (58) 19 (70)  

Clinical prediction. Kaplan-Meier analysis for time to metastasis as a first event revealed nonoverlapping confidence intervals between low- and high-risk patients, suggesting that the 70-gene microarray test discriminates among tumors with differing metastatic propensity in this cohort despite use of adjuvant treatment (Fig. 2). Despite this trend, and most probably due to the low metastasis event rate in this cohort, there was no statistically significant difference in time to distant metastasis between patients classified with low-risk or high-risk tumor signatures in the MGH cohort. In contrast, microarray classification of tumors from node-negative NKI patients previously revealed statistically significant differences between high- and low-risk signature patients for time to distant metastasis (P < 0.001; ref. 11).

Fig. 2.

Molecular classification of MGH patients. Kaplan-Meier analysis of time to metastasis as a first event for the MGH cohort based on classification with the 70-gene signature. Red, MammaPrint high risk; blue, MammaPrint low risk; thin lines, 95% CI.

Fig. 2.

Molecular classification of MGH patients. Kaplan-Meier analysis of time to metastasis as a first event for the MGH cohort based on classification with the 70-gene signature. Red, MammaPrint high risk; blue, MammaPrint low risk; thin lines, 95% CI.

Close modal

Detailed examination of prediction results revealed that the negative predictive value (NPV) of the 70-gene signature was 100% in the MGH cohort, which was comparable with a NPV of 88% in the node-negative NKI cohort (Table 3A). Of 100 patients, 27 were classified as low risk and 73 as high risk by MammaPrint assay. Of the 27 low-risk patients, none had distant metastasis as a first event after initial diagnosis. Notably, the NPV of the 70-gene signature also remained 100% when strictly considering 5- or 10-year distant metastasis-free survival (Table 3B and C). Importantly, all MGH patients who developed distant metastasis as a first event had primary tumors with high-risk 70-gene expression signatures (n = 9). However, 64 additional patients were also classified as poor prognosis by the 70-gene test, which was unexpectedly high given what is generally thought to be the low-risk nature of older breast cancer patients. The positive predictive value (PPV) of the 70-gene signature was therefore only 12%, significantly lower than a previously observed PPV of 52% in node-negative NKI patients.

Table 3.

NPV and PPV [overall (A), 5-y (B), and 10-y (C)] for time to distant metastasis in MGH compared with node-negative NKI patients classified as either high or low risk for distant metastasis using the 70-gene signature

A
GroupPatientsDistant metastasesDisease-freePPredictive value
PPV (95% CI)NPV (95% CI)
MGH       
    High-risk signature 73 64 0.108 12% (6-22%) 100% (87-100%) 
    Low-risk signature 27 27    
NKI lymph node negative       
    High-risk signature 91 47 44 <0.001 52% (41-62%) 88% (77-95%) 
    Low-risk signature 60 53    
       
B
 
      
Group Patients Distant metastases (≤5 y) Disease-free (>5 y) P 5-y predictive value
 
 

 

 

 

 

 
PPV (95% CI)
 
NPV (95% CI)
 
MGH       
    High-risk signature 60 53 0.192 12% (5-23%) 100% (81-100%) 
    Low-risk signature 18 18    
NKI lymph node negative       
    High-risk signature 70 36 34 <0.001 51% (39-64%) 93% (82-98%) 
    Low-risk signature 54 50    
       
C
 
      
Group Patients Distant metastases (≤10 y) Disease-free (>10 y) P 10-y predictive value
 
 

 

 

 

 

 
PPV (95% CI)
 
NPV (95% CI)
 
MGH       
    High-risk signature 49 42 0.330 14% (6-27%) 100% (72-100%) 
    Low-risk signature 11 11    
NKI lymph node negative       
    High-risk signature 58 42 16 <0.001 72% (59-83%) 75% (55-89%) 
    Low-risk signature 28 21    
A
GroupPatientsDistant metastasesDisease-freePPredictive value
PPV (95% CI)NPV (95% CI)
MGH       
    High-risk signature 73 64 0.108 12% (6-22%) 100% (87-100%) 
    Low-risk signature 27 27    
NKI lymph node negative       
    High-risk signature 91 47 44 <0.001 52% (41-62%) 88% (77-95%) 
    Low-risk signature 60 53    
       
B
 
      
Group Patients Distant metastases (≤5 y) Disease-free (>5 y) P 5-y predictive value
 
 

 

 

 

 

 
PPV (95% CI)
 
NPV (95% CI)
 
MGH       
    High-risk signature 60 53 0.192 12% (5-23%) 100% (81-100%) 
    Low-risk signature 18 18    
NKI lymph node negative       
    High-risk signature 70 36 34 <0.001 51% (39-64%) 93% (82-98%) 
    Low-risk signature 54 50    
       
C
 
      
Group Patients Distant metastases (≤10 y) Disease-free (>10 y) P 10-y predictive value
 
 

 

 

 

 

 
PPV (95% CI)
 
NPV (95% CI)
 
MGH       
    High-risk signature 49 42 0.330 14% (6-27%) 100% (72-100%) 
    Low-risk signature 11 11    
NKI lymph node negative       
    High-risk signature 58 42 16 <0.001 72% (59-83%) 75% (55-89%) 
    Low-risk signature 28 21    

NOTE: In determining 5- and 10-y values, the following were excluded: (a) patients with distant metastasis as a first event (after 5 or 10 y), (b) patients without distant metastasis as a first event but with <5 (or 10) y of follow-up, and (c) patients with local or regional disease recurrence or the development of a second primary cancer as a first event.

The 70-gene signature was originally optimized for sensitivity by choosing a classification threshold of 0.4 for microarray correlation values, a cutoff that resulted in a 10% false-negative rate for positive prediction of distant metastasis (9). However, reanalysis using a range of alternative cutoffs for the 70-gene predictive model did not reveal any that could yield a high PPV in the MGH cohort (data not shown).

Finally, using the Adjuvant! Online clinical decision-making tool, we determined population-based estimates of 10-year relapse risk for each individual MGH patient. By applying a threshold, we defined patients as being at low or high risk for distant relapse. We chose the threshold that simultaneously maximized the number of patients classified as low risk while maintaining a NPV of 100%. With this threshold, we found that MammaPrint classified an additional 21 patients (beyond Adjuvant! Online) as low risk, and importantly, none of these patients developed distant metastasis as a first event (Fig. 3A). Moreover, even when strictly considering 10-year distant metastasis-free survival, we found that MammaPrint classified an additional five patients (beyond Adjuvant! Online) as low risk, and again, none developed distant metastasis as a first event (Fig. 3B). These results suggest that the 70-gene signature might provide useful and additive information to an important clinicopathologic risk model in older breast cancer patients.

Fig. 3.

Classification of MGH patients using Adjuvant! Online compared with MammaPrint. Adjuvant! Online–based relapse risk for individual patients classified as MammaPrint low or high risk for (A) all 100 patients, (B) strictly considering 10-y distant metastasis-free survival. Solid square, patient with distant metastasis as a first event; circle, patient without distant metastasis as a first event; + in circle, patient identified as low risk by MammaPrint beyond Adjuvant! Online–based classification; horizontal line, threshold that simultaneously maximizes the number of patients classified as low risk while maintaining a NPV of 100%.

Fig. 3.

Classification of MGH patients using Adjuvant! Online compared with MammaPrint. Adjuvant! Online–based relapse risk for individual patients classified as MammaPrint low or high risk for (A) all 100 patients, (B) strictly considering 10-y distant metastasis-free survival. Solid square, patient with distant metastasis as a first event; circle, patient without distant metastasis as a first event; + in circle, patient identified as low risk by MammaPrint beyond Adjuvant! Online–based classification; horizontal line, threshold that simultaneously maximizes the number of patients classified as low risk while maintaining a NPV of 100%.

Close modal

The 70-gene MammaPrint assay has been retrospectively validated as a prognostic tool, suggesting that younger breast cancer patients with intrinsically low-risk disease can be molecularly identified and that this information can be combined with clinical risk profiling to more accurately identify patients who might be spared adjuvant chemotherapy with its associated toxicity (10, 11). Most patients with breast cancer, however, are older and have comorbid disease that complicates decisions about the use of adjuvant chemotherapy. The potential role of the MammaPrint assay in the clinical care of these patients remains unclear.

We studied the 70-gene signature in an older breast cancer cohort who were diagnosed and treated at the MGH, purposely focusing on patients with lymph node–negative disease like those with whom MammaPrint was initially discovered and validated (911). These MGH patients had a median age of 62.5 years with predominantly small and low-grade tumors. Forty-five percent of MGH patients also received some form of systemic treatment, reflecting known practice patterns among U.S. oncologists. Consistent with these differences, the MGH cohort had low distant recurrence rates.

The 70-gene signature displayed a remarkably high NPV (100%) in this cohort, which is consistent with earlier findings in younger cohorts (10, 11). Interestingly, MammaPrint-based classification also seemed to provide additive information to Adjuvant! Online–based risk determination. Although the MammaPrint assay was originally designed to identify younger breast cancer patients at low risk for distant metastasis who might consequently be spared systemic treatment, these results show that the same signature might identify older breast cancer patients at lower risk for distant recurrence at initial diagnosis.

In contrast, the MammaPrint assay had an extremely low PPV (12%), significantly lower than that previously observed in node-negative NKI patients (52%), thus resulting in statistically insignificant differences in overall survival between low- and high-risk patients. Remarkably, 73% of MGH patients were classified as high risk (including all nine patients who eventually developed metastasis), which was strikingly discordant with the low overall metastasis rate in this cohort (9%).

Because 45% of all patients received some adjuvant treatment, this discrepancy between molecular classification and clinical outcome could in part be explained by the significant use of adjuvant treatment in these patients. However, the proportional risk reduction for recurrent disease from adjuvant chemotherapy has been estimated to be ∼20%, and for hormonal therapy ∼30%, in postmenopausal patients (1). Thus, even in the complete absence of adjuvant treatment, the event rate in the MGH cohort would likely only have doubled to ∼18%, still far below what might be predicted through molecular classification with the MammaPrint assay.

This finding that the 70-gene signature molecularly classifies a significant percentage of older age breast cancer patients as “high risk,” of which few develop metastatic disease, may offer insight into the process of metastases. Following Paget's hypothesis that metastasis depends both on the “seed” (the cancer cell itself) and the “soil” (the “host”), our findings raise the intriguing hypothesis that most breast cancers in older patients are intrinsically high risk but that this intrinsic metastatic capacity does not become manifest due to other (possibly host) factors in these predominantly postmenopausal patients.

Because older women with newly diagnosed breast cancer are increasingly being offered adjuvant treatment, in contrast with historical practice patterns, larger studies of the 70-gene signature should be done to determine whether the MammaPrint assay (with its excellent NPV) is clinically useful for pretreatment determination of intrinsic risk in older breast cancer patients. An open question in light of the poor PPV of MammaPrint in older patients is whether there exist additional, clinically useful gene expression signatures for positively predicting distant metastasis risk in the postmenopausal breast cancer population. Clearly, larger studies on postmenopausal breast cancer cohorts will be required to address this issue.

L.J. Van't Veer, A.M. Glass, and T. Bruinsma are employed by Agendia BV. L.J. Van't Veer has an ownership interest in MammaPrint. L.J. Van't Veer and R. Bernards are named inventors on a patent to use microarray technology to ascertain breast cancer prognosis and hold equity interests in Agendia BV.

Grant support: Avon Foundation (Massachusetts General Hospital Cancer Center) and Dutch National Genomics Initiative “Cancer Genomics Center” (Netherlands Cancer Institute).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).

We thank Marc van de Vijver for helpful discussions and Arno Floore and Agendia laboratory staff for excellent technical work in generating microarray data.

1
Early Breast Cancer Trialists' Collaborative Group (EBCTCG). Effects of chemotherapy and hormonal therapy for early breast cancer on recurrence and 15-year survival: an overview of the randomised trials.
Lancet
2005
;
365
:
1687
–717.
2
Perou CM, Sorlie T, Eisen MB, et al. Molecular portraits of human breast tumours.
Nature
2000
;
406
:
747
–52.
3
Goldhirsch A, Glick JH, Gelber RD, Coates AS, Senn HJ. Meeting highlights: International Consensus Panel on the Treatment of Primary Breast Cancer. Seventh International Conference on Adjuvant Therapy of Primary Breast Cancer.
J Clin Oncol
2001
;
19
:
3817
–27.
4
Eifel P, Axelson JA, Costa J, et al. National Institutes of Health Consensus Development Conference statement: adjuvant therapy for breast cancer, November 1-3, 2000.
J Natl Cancer Inst
2001
;
93
:
979
–89.
5
Isaacs C, Stearns V, Hayes DF. New prognostic factors for breast cancer recurrence.
Semin Oncol
2001
;
28
:
53
–67.
6
Ramaswamy S, Golub TR. DNA microarrays in clinical oncology.
J Clin Oncol
2002
;
20
:
1932
–41.
7
Ramaswamy S, Perou CM. DNA microarrays in breast cancer: the promise of personalised medicine.
Lancet
2003
;
361
:
1576
–7.
8
Ramaswamy S, Ross KN, Lander ES, Golub TR. A molecular signature of metastasis in primary solid tumors.
Nat Genet
2003
;
33
:
49
–54.
9
van 't Veer LJ, Dai H, van de Vijver MJ, et al. Gene expression profiling predicts clinical outcome of breast cancer.
Nature
2002
;
415
:
530
–6.
10
Buyse M, Loi S, van't Veer L, et al. Validation and clinical utility of a 70-gene prognostic signature for women with node-negative breast cancer.
J Natl Cancer Inst
2006
;
98
:
1183
–92.
11
van de Vijver MJ, He YD, van't Veer LJ, et al. A gene-expression signature as a predictor of survival in breast cancer.
N Engl J Med
2002
;
347
:
1999
–2009.
12
Bogaerts J, Cardoso F, Buyse M, et al. Gene signature evaluation as a prognostic tool: challenges in the design of the MINDACT trial.
Nat Clin Pract Oncol
2006
;
3
:
540
–51.
13
Ma XJ, Wang Z, Ryan PD, et al. A two-gene expression ratio predicts clinical outcome in breast cancer patients treated with tamoxifen.
Cancer Cell
2004
;
5
:
607
–16.
14
Glas AM, Floore A, Delahaye LJ, et al. Converting a breast cancer microarray signature into a high-throughput diagnostic test.
BMC Genomics
2006
;
7
:
278
.

Supplementary data