Genomic Grade Index: An important tool for assessing breast cancer tumor grade and prognosis

1. Introduction 

The assessment of prognosis and the treatment decision-making process still generate controversy in discussions about individualizing breast cancer treatment. Decisions related to the use of adjuvant therapy in breast cancer have relied on traditional clinicopathologic features such as tumor size, node involvement, histologic grade, hormone receptors, and HER2 status, among others. In the post-genomic era a molecular classification of breast cancer was proposed, which lead to a better understanding of breast cancer heterogeneity. Gene expression signatures were subsequently developed to predict individual patient outcome, and also to predict response to hormone and chemotherapy treatment. The Genomic Grade Index (GGI) is one such signature and was designed to better evaluate whether histologic grade could be better defined by gene expression profiling.

In this review article we discuss the limitations inherent in the classical approach to assessing histological grade and the process that was followed to develop a gene expression profile defined to answer a biologic question. The applicability of GGI to assessing breast cancer tumor grade, prognosis, different subtypes of hormone-responsive breast cancer and response to chemotherapy is critically reviewed. A comparison of the prognostic performance of different signatures is also performed, followed by an analysis of the biological significance of the genes grouped in different signatures.

2. Histologic grade of breast cancer 

Histologic grade of breast cancer was first described in 1925 [1]. Although improvements in grade assessment have been made, the latest recommendations of the Breast Task Force, an advisory body constituted by the American Joint Committee on Cancer considered the available data too weak to support the addition of histologic grade to the TNM staging system for breast cancer [2].

Following the primordial definition of tumor grade at the beginning of the 20th century, tumor grade started to be defined as a combination of cellular factors or as a group of nuclear characteristics [3], [4], [5], [6], [7], [8]. Later, in 1957, Bloom and Richardson, aiming to increase tumor grade assessment objectivity, proposed a numerical scoring system (SBR) [9]. In order to increase reproducibility, Elston and Ellis then proposed a modification of the Bloom and Richardson grading system [10]. The proposed criteria, referred to as the Nottingham combined histologic grade, is based on the semi-quantitative evaluation of morphologic features related to tumor differentiation (tubule formation, degree of nuclear pleomorphism and mitotic count) [10]. The Nottingham combined histologic grade is the recommended method for breast cancer tumor grade assessment by the last College of American Pathologists Consensus Statement (Table 1) [11].

Table 1. Nottingham combined histologic grade. The semi-quantitative evaluation of tumor grade is based on the morphologic features described below (tubule formation, nuclear polymorphism and mitotic counts). Overall tumor grade is obtained adding the score obtained in each category. Grade I or well-differentiated tumors, 3–5 points; grade II or moderately differentiated, 6–7 points; grade III or poorly differentiated, 8–9 points.

		Score 1	Score 2	Score 3
	Tubule Formation	Majority of tumor (>75%)	Moderate degree (10–75%)	Little or none (<10%)
	Nuclear pleomorphism	Small, regular uniform cells	Moderate increase in size and variability	Marked variation
	Mitotic countsa	0–9/10 fields	10–19/10 fields	>20/10 fields

a Mitotic counts need to be adapted according to the microscope used.

2.1. Histologic tumor grade – prognostic impact 

The relationship between Nottingham combined histologic grade and prognosis was first evaluated in the Nottingham/Tenovus study [10]. A total of 1830 patients with primary operable breast cancer treated by mastectomy or local excision and radiotherapy, with loco-regional lymph node sampling were analyzed. Tumors were graded independently by two pathologists, with an overall agreement of 90%. The discordant cases were resolved by joint examination on a conference microscope. The tumor grade distributions were as follows: 342 cases (19%) grade 1, 631 (34%) grade 2 and 857 (47%) grade 3. A significant correlation between tumor grade and prognosis could be demonstrated. Grade 2 and 3 tumors have superimposed worse recurrence free interval compared to grade I (p<0.0001). In the same direction, grade 2 and grade 3 tumors had worse overall survival (OS) when compared to grade 1 tumors (p<0.0001). The analysis of other factors that could improve the prognostic information of histological grade has led to the development of the Nottingham Prognostic Index (NPI). NPI is calculated using tumor size, histological lymph node stage and histological grade (Table 2) [12].

Table 2. Nottingham Prognostic Index. NPI is calculated using tumor size, lymph node stage and histological grade. In primary operable breast cancer, NPI is able to identify three prognostic groups.

	NPI score=0.2×tumor size+lymph node stage+histological grade
	Tumor size	Measured pathologically – results in cm
	Lymph node stage	1: node negative	2: 1–3 positive node	3: ≥4 positive nodes
	Histological grade	1: grade I	2: grade II	3: grade III

	Prognosis groups	NPI cut-off points	Annual mortality %
	Good	<3.4	3
	Moderate	≥3.4 and ≤5.4	7
	Poor	>5.4	30

Further studies were carried out to evaluate the performance of the Nottingham combined histologic grade to determine prognosis [13], [14], [15]. In consonance with previous results, grade 1 and grade 3 tumors were associated with different disease free survival (DFS) and/or OS. In the subgroup of patients with grade II tumors, the prognostic information obtained was not clear, sometimes grouped with grade III, and sometimes in an intermediate group (Table 3).

Table 3. Outcome of patients with early stage breast cancer according to tumor grade assessed by the Nottingham combined histologic grade (Elston–Ellis modification of Bloom and Richardson grading system). The prognostic information obtained with histological grade clearly differentiates two groups of patients with distinct clinical outcome (grade I versus grade III). The clinical relevance of a histological grade II result is debatable, because the clinical outcome is not consistently maintained in different series.

	Study	No. of patients; TNM staging	Follow-up (years)	Outcome evaluated	Outcome (%)
	1998 [13]	877; T1/2, N0/1	6	OS	96	88	80
	1999 [14]	318; T1a/b, N0	10	OS	95	91	91
	2000 [15]	228; T1/2, N0	10	RFS	90	70	69

OS, overall survival; RFS, relapse-free survival.

Another modification from the original SBR tumor grade evaluation was proposed based on a multivariate analysis of 1262 invasive ductal breast carcinomas [16]. The disproportional number of patients characterized as grade 2 (approximately two-thirds), coupled with the confusing prognostic information obtained with this result provided arguments for modifying the pre-defined SBR classification. The three components of histologic grade (tubule formation, nuclear pleomorphism, mitotic count) were analysed individually to better define the impact of each category. A modified SBR (MSBR) risk score was thus defined using the two nuclear components (nuclear pleomorphism, mitotic count). MSBR was able to divide the 694 SBR grade 2 patients into three groups with statistically significant different prognosis for metastasis-free survival. Although the MSBR classification could add valuable prognostic information to histological grade 2-classification, it was not subsequently validated in independent series. In addition, one of the concerns would be the lack of interobserver reproducibility, a major limitation of histological grade scoring.

3. Genomic Grade Index development 

GGI was developed to explore whether gene expression profiling could be used to grade tumors more accurately than the conventional histological grade [19]. DNA microarray technology renders it possible to assay the expression of thousands of genes simultaneously. Gene-specific probes (i.e. oligonucleotides) representing thousands of genes are arrayed on an inert substrate (microarray platform). Messenger RNA (mRNA) extracted from a tissue of interest is amplified, fluorescent labeled and further hybridized with the gene-specific probes in the microarray platform. Fluorescent detectors coupled with computers are used for scanning and are able to estimate the gene expression for a given sample [20].

4. Definition of breast cancer subtypes and application of Genomic Grade Index 

Gene expression profiling with the use of DNA microarrays has added valuable information to our understanding of breast cancer biology. In the seminal work of Perou et al., the ability to interrogate thousands of genes at the same time was translated into a “molecular portrait” of each tumor sample studied, and the concomitant analysis of the individual molecular portraits of breast cancer tumor samples made the definition of molecular subtypes of breast cancer possible [25]. In order to analyze this large quantity of information (thousands of genes per sample evaluated), a hierarchical clustering method was used to group genes according to similar patterns of expression [26]. The proposed molecular classification of breast cancer was divided into five classes: luminal-A, luminal-B, basal-like, HER2-positive and normal-like tumors [21], [25], [27], [28], [29], [30], [31]. Subsequently, the correlation between molecular subtypes and clinical data have shown a significant difference in overall survival between the subtypes [22].

Despite this progress, the clinical applicability of molecular classification is limited by the tight correlation between the molecular subtypes and currently available immunohistochemical markers (ER, PR, HER2 ki67) [32]. For example, the molecular subtype HER2-positive is clinically detected by IHC or fluorescent in situ hybridization (FISH) according to published guidelines [33]. Although a good correlation has been established between the molecular subtype HER2 and clinically assessed HER2-positive breast cancer, the opposite is not true, because 30% of HER2-positive breast cancers are molecularly characterized as luminal-B [34]. Luminal-A and luminal-B molecular subtypes are, by definition, hormone receptor positive tumors, but the distinction between these two subtypes is controversial.

One of the proposed clinical definitions characterizes luminal-A and luminal-B tumors using hormone receptor status, HER2 status and the Ki67 index (percentage of Ki67-positive nuclei by IHC). Luminal-A is defined as being ER- and/or PR-positive, HER2-negative and Ki67-low (ki67 index<14%). Luminal-B is defined as ER- and/or PR-positive, HER2-negative and Ki67-high (ki67 index>14%). Another luminal-B subtype has also been proposed, namely luminal HER2 enriched, with tumors being ER- and/or PR-positive, HER2-positive and Ki67-high (ki67 index>14%) [34]. Although the proposed classification allows for broader application, due to the widespread use of IHC, some inherent limitations raise concern: IHC evaluation is limited by interobserver variability, qualitative readouts and technical reproducibility [35]. For example, false positive and negative rates for distinguishing luminal-A and luminal-B tumors have been reported to be approximately 25%, which suggests that using IHC classification for therapeutic decision-making may be limited [34]. Considering the worse outcome for luminal-B tumors and consequent therapeutic implications, the precise definition of the luminal-A and luminal-B breast cancer subtypes is clinically important.

In Perou et al. a biological interpretation of the genes involved in the different subtypes was also performed. The largest group of genes was related to cell proliferation, and their expression varied among the tumor subgroups [25]. This observation, as well as the fact that the high grade tumors identified by GGI revealed a preponderance to overexpress cell cycle and proliferation genes, contributed to the thinking that GGI could potentially find application in better stratifying hormone receptor positive breast cancer patients [36].

When GGI was applied to previously defined breast cancer molecular subtypes, its ability to classify the subgroups in high or low risk was demonstrated. In the data set of van de Vijver, even the population that was previously classified molecularly as “undefined” could be classified by the application of GGI [23]. Luminal-A and normal-like subtypes were categorized as low-GGI. HER2, basal-like, luminal-B and the subgroup of patients previously unclassified were categorized as high-GGI [36].

The prognostic ability of GGI to stratify luminal subtypes was further evaluated in an ER-positive data set of 666 tumor samples, deriving from two smaller data sets. The first was a group of 417 samples from ER-positive tumors that had received no adjuvant treatment; these samples had been used to evaluate the prognostic information that could be obtained with GGI without interference with hormonal treatment. The second, complimentary data set consisted of 249 samples of ER-positive tumors treated with tamoxifen, and had been evaluated to measure the effect of tamoxifen according to GGI. For each tumor sample from the combined set of 666 ER-positive patients, a high or low-GGI value was assessed according to the expression of 97 genes. At 10 years of median follow-up, time to distant metastases (TDM) was significantly worse for ER-positive patients with a high-GGI score. In the untreated data set (417 patients), multivariate analysis demonstrated that genomic grade, progesterone at mRNA level and histologic grade were significantly associated with prognosis (GGI-high versus -low: HR=1.96, 95% CI 1.27–3.02, p=0.0024). In the tamoxifen-treated dataset (249 patients), only GGI was associated with prognosis in multivariate analysis (GGI-high versus -low: HR=2.50, 95% CI 1.28–4.90, p=0.0074). Taking GGI as a continuous variable, the rate of metastases increased with increasing GGI value, indicating a high-GGI discrimatory value in the ER subgroup [36].

5. Genomic Grade Index and response to chemotherapy 

The ability of GGI to predict response to neoadjuvant chemotherapy was evaluated in 229 tumor samples collected before neoadjuvant chemotherapy with paclitaxel, fluorouracil, doxorubicin and cyclophosphamide (T/FAC) [37]. In general, pathologic complete response (pCR) is associated with better disease outcome regardless of hormone receptor status [38], [39]. Histologic grade is known to be a predictor of pathologic complete response, but inherent limitations to histological grade assessment limits its applicability (discussed in Section 2.2) [2], [40].

In the evaluation of GGI as a predictor of response, a more precise method for evaluating pathologic response called residual cancer burden (RCB) was used as a comparator [41]. RCB better defines different ranges of pathologic response after neoadjuvant chemotherapy. It is calculated as continuous variable, using pathologic measurements of the primary tumor and nodal metastases. In post-treatment surgical resection specimens a bidimensional diameter of primary tumor bed and the proportion of invasive tumor cells in the same area are measured. The number of nodes containing metastases and the diameter of the large lymph node metastases are also components of RCB. The prognostic information obtained with RCB was evaluated in 382 patients treated with neoadjuvant chemotherapy. In a multivariate analysis containing age, clinical stage, hormone receptor status, hormone treatment and pathologic response (pCR versus residual disease), RCB was an important prognostic factor associated with distant relapse-free survival (HR=2.50; 95% CI 1.70–3.69; p<0.001). Minimal residual cancer burden (RCB-I) and pCR (RCB-0) were associated with similar favourable long-term relapse-free prognosis. RCB adds to a better understanding of response to primary chemotherapy.

In the evaluation of GGI compared to RCB to predict chemotherapy response, a data set comprising 229 samples from 132 ER-positive and 97 ER-negative patients was used. All patients had HER2 non-amplified tumors, which avoided the interference of chemotherapy and trastuzumab response. Pathologic response was assessed as follows: RCB-0 indicating pCR, and RCB-I, RCB-II, RCB-III for minimal, intermediate and extensive residual disease respectively. GGI was assessed for each tumor sample and assigned as low or high risk, as in the original publication, but also as a continuous variable. The GGI evaluation characterized 84.6% of grade 1 tumors as low risk and 88.3% of grade 3 as high risk. The histological grade 2 group was divided into 62.7% low risk and 37.3% high risk. For the ER-positive and ER-negative subgroups, 44.8% and 89.6% respectively were assigned to the GGI high-risk category. For the overall group treated with neoadjuvant T/FAC, high-risk GGI was associated with higher response than low-risk GGI (40% versus 12%; p<0.001). A positive correlation was observed between GGI high-risk category and the level of observed response to neoadjuvant chemotherapy, with 85.8% of patients with RCB-0 or RCB-I, characterized as GGI high-risk.

GGI evaluated as a continuous variable also correlated with response to neoadjuvant chemotherapy in the ER-positive and ER-negative evaluated patients. In a multivariate analysis including age, ER status, tumor grade, tumor stage and continuous GGI, the presence of a high-GGI score predicted independently for increased probability of response to chemotherapy (OR, 1.86; 95% CI 1.15–3.00; p=0.011). In the ER-positive and GGI high-risk subgroup, a worse distant relapse-free survival was observed, despite higher chemotherapy sensitivity.

A complex balance between baseline prognosis, chemotherapy sensitivity and hormonal responsiveness (among the ER-positive tumors) dictates patients’ overall survival after neoadjuvant or adjuvant therapy. A worse prognosis in the subpopulation with a high-GGI index is probably explained by a balance favouring the prognostic information of the index itself over the higher chemotherapy sensitivity.

6. Genomic Grade Index and other gene expression signatures 

Gene expression signatures have been developed to provide better prediction of clinical outcome than conventional clinicopathologic criteria. Most signatures are developed to identify a higher proportion of low-risk patients that can be treated without systemic adjuvant treatment, while still correctly identifying the high-risk patients [42]. Three different strategies are commonly used to developing gene expression prognostic signatures. According to the “top-down” approach, gene expression data from cohorts of patients with known clinical outcome are compared to identify genes associated with prognosis without any previous biological assumptions. In the “bottom-up” approach, gene patterns associated with a specific biological phenotype are first identified and subsequently correlated with clinical outcome. The candidate gene approach utilizes genes known to be related to a specific biological processes and combines them in a multivariate predictive model [32].

The Amsterdam 70-gene signature (Mammaprint®, Agendia) is an example of a gene signature developed using the top-down approach. A retrospective series of 78 patients with node negative breast cancer who had received no adjuvant systemic therapy was used and a group of 70 genes could be selected. The expression measure of the 70 genes was used to calculate a correlation score that divides patients into good or poor risk groups [43]. Retrospective validation studies subsequently demonstrated the ability of the 70 genes to define breast cancer patients with node-positive and node-negative disease as being either low or high risk [23], [44].

The 21-gene recurrence score (Oncotype Dx®, Genomic Health) uses quantitative reverse-transcriptase-polymerase-chain-reaction (RT-PCR) assay to measure the expression of genes known to be related with a biological process, such as ER, HER2 and proliferation genes. The gene expression measure is combined in a recurrence score (RS) that can be used to predict disease recurrence [18]. The retrospective evaluation of RS in 668 patients with ER-positive and node negative breast cancer treated with tamoxifen was able to predict disease recurrence at 10 years and divide the studied population into three risk categories. The benefit of chemotherapy was also evaluated with the RS, where larger recurrence scores were associated with benefit from chemotherapy [45], [46].

The GGI was developed using the bottom-up approach, whereby genes associated with histologic grade were identified and subsequently related to clinical outcome [19]. This signature expresses 97 genes and is able to better define tumor grade, patient prognosis, and breast cancer subtypes, as well as to predict chemotherapy sensitivity (discussed in Sections 4, 5, Fig. 1).

View Large Image Download to PowerPoint

Fig. 1. Genomic Grade Index (GGI) is a valuable tool for assessing patient prognosis, response to chemotherapy and endocrine therapy. The chemotherapy responsiveness increases with higher GGI values. Better prognosis and higher endocrine responsiveness are observed in the subgroup of patients with lower GGI values.

The prognostic information obtained with GGI was compared directly with the prognostic information obtained with the Amsterdam 70-gene signature, by using the original Amsterdam data set [43]. In the NKI2 dataset, 93 of 97 GGI-associated genes were identified and a similar separation in distant metastasis-free survival between low and high-risk groups was found in both the 70-gene signature (HR=3.41, 95% CI 2.09–5.58, p<0.001) and GGI (HR=4.68, 95% CI 2.74–8.0, p<0.001) [19].

GGI also performed like the 21-gene RS to stratify hormone receptor positive breast cancer into two subtypes with corresponding distinct outcomes. With a data set of 249 ER-positive, tamoxifen-treated tumors, GGI and RS were performed using microarray data and the 21-gene RS algorithm. The RS categories of low and intermediate risk were combined into one lower risk category due to similar outcome. The comparison between risk groups generated by GGI and RS demonstrated that the classifications were significantly correlated (p<0.0001). Although a similar prognostic performance could be observed between GGI and 21-gene RS, no direct comparison between the 21-gene RS performed in paraffin-fixed samples with GGI is available [36].

7. Gene expression signatures, prognosis and biological significance 

The development of microarray technology and the classification of breast cancer into molecular subtypes have made it possible for different research groups to develop different gene expression signatures. Breast cancer prognosis that is usually assessed by using common histopathologic variables now can be done in conjunction with gene expression signatures that add to outcome prediction.

The increasing number of gene signatures does not completely elucidate the biological significance of different genes, however. The ability of three different signatures to predict outcome in a same population was evaluated in the TRANSBIG series (198 patients), where long patient follow-up (up to 25 years) made it possible to evaluate performance over time [47]. The 70-gene signature, a 76-gene signature (Rotterdam) and the GGI were built in different microarray platforms, with different development strategies (the 70- and 76-gene-signatures both top-down; GGI – bottom-up), with limited overlapping genes, and with different statistical plans. Despite these discrepancies between the signatures, they demonstrated similar capabilities of predicting distant metastases-free survival. In accordance with previous findings, strong time dependence could be observed for up to 10 years of observation [44], [47], [48]. After 10 years, the ability of gene expression signatures to predict outcome was diminished, probably due to pre-specified sensitivity and the specificity of these signatures to predict early relapse.

The similar outcome predictions between the different signatures motivated a search for underlying biologic processes that could be represented by different genes in non-overlapping signatures. A large meta-analysis of publicly available breast cancer gene expression and clinical data evaluated the contribution of known biological processes to the performance of different gene signatures. “Coexpression modules” for ER signaling, ERBB2 amplification and proliferation were generated, putting together a comprehensive list of genes with highly correlated expression [24]. The meta-analysis was able to confirm in 2833 patients that the initial classification of breast cancer molecular subtypes was highly conserved, with the exception of normal-like breast cancer, which could not be identified. In all breast cancer subtypes (HER2, basal-like, luminal-A, luminal-B) the coexpression module proliferation was the most important determinant of prognosis. Although the coexpression module HER2 could be identified in the subtype HER2, the prognostic information was mainly driven by genes related to proliferation. HER2 and basal-like subtypes were consistently characterized as high proliferative tumors. In the luminal subtypes, the module of genes related to proliferation could divide this group into a low-proliferative subtype with better prognosis and a highly proliferative group with poorer prognosis. In luminal tumors the coexpression module of ER signaling was observed, but again the genes related to proliferation were determinants of prognosis. The evaluation of clinical variables demonstrated that tumor size and nodal status still have independent prognostic value and need to be evaluated together with the information obtained from gene-signatures.

Moving further with the attempt to investigate the biological significance of gene signatures, a meta-analysis of more than 2100 breast cancer patients containing gene expression data and clinicopathologic data was performed [42]. Gene expression modules were created, selecting a “prototype gene” for each chosen biological process, followed by the identification of genes highly correlated with each process. Gene modules were built representing proliferation, tumor invasion/metastasis, immune response, angiogenesis, apoptosis phenotypes, ER and HER2 signaling, respectively. The molecular classification of breast cancer was evaluated according to ER and HER2 module scores, identifying three main clusters of tumors. The basal-like, HER2 and luminals were identified as ER−/HER2−, HER2+ and ER+/HER2− respectively, which corresponds to the initial definition of breast cancer molecular portraits [25]. Subsequently, the different gene modules were evaluated according to breast cancer subgroups defined by ER and HER2 modules. In the ER−/HER2− subgroup only the immune response module was associated with outcome in multivariate analysis (HR=0.70, 95% CI 0.50–0.98, p=0.04). The outcome of the subpopulation defined as ER+/HER2− was predicted in multivariate analysis by proliferation (HR=2.68, 95% CI 2.02–3.55, p=9×10−12) and histologic grade (HR=2.00, 95% CI 1.18–3.37, p=0.01). For the HER2+ tumors, tumor invasion/metastasis (HR=2.07, 95% CI 1.32–2.25, p=0.001) and immune response (HR=0.56, 95% CI 0.36–0.86, p=0.009) were associated with prognosis in the multivariate model. The difference in coexpression modules emphasized the importance of proliferation genes in the ER+/HER2− subset. The absence of a significant impact of the proliferation module in the ER−/HER− and HER2+ does not preclude the importance of proliferation, considering that proliferating genes were present in all coexpression modules.

8. Conclusions 

Understanding breast cancer molecular heterogeneity has made it possible to develop gene signatures that can be applied to predict prognosis and response to therapies in daily practice. The superiority of gene signatures to classic histopathologic variables is related to their ability to better define a greater proportion of low-risk patients that do not need to be treated with systemic adjuvant therapy, while still correctly identifying those patients who fall into a high-risk group. Clinical variables related to the measurement of tumor progression such as tumor size and nodal involvement remain significantly associated with prognosis and should therefore continue to be evaluated in conjunction with gene signatures. Models to assist physicians in the treatment decision-making process, such as Adjuvant! Online (AOL), have also been compared with gene signatures. The 70-gene and 76-gene signatures, which have the same performance as the 97-gene GGI, outperformed AOL, demonstrating their better ability to prognosticate for breast cancer patients [44], [48].

The diminished accuracy of gene expression signatures to predict prognosis after 10 years of follow-up does constitute an important limitation, however. In the subgroup of hormone-responsive tumors, disease progression after 10 years is not unusual and cannot be predicted by the gene expression signatures available [44], [48].

Most of gene signatures developed depend on fresh or frozen tissue being obtained, which is not possible at all hospitals. The availability of RNA preserving solutions eliminates the need to immediately freeze samples and allows hospitals without freezing capabilities to have the genomic tests performed. One important point favouring fresh tissue over paraffin-fixed samples is that frozen tissue can be stored long-term in biological specimen banks, which are fundamental to the subsequent development of breast cancer treatments. The commercially available GGI signature (MapQuant Dx™ Genomic Grade-Ipsogen®) and 70-gene signature (Mammaprint®, Agendia) are examples of tests performed with fresh or frozen tissue. Oncotype Dx® Genomic Health utilizes formalin-fixed paraffin-embedded tissue, facilitating their applicability.

GGI can be considered a valuable tool for the assessment of breast cancer prognosis. The intrinsic prognostic information of proliferation genes is better evaluated with GGI than with classic histologic grade. The association between high-GGI and higher sensitivity to neoadjuvant chemotherapy (T/FAC) can be useful for designing neoadjuvant trials in breast cancer, not just with chemotherapy, but also with new biologic agents and hormonal therapy [19], [36], [37].

Oncotype DX® is recommended by the American Society of Clinical Oncology and National Comprehensive Cancer Network Guidelines in Oncology [49], [50]. The last Saint Gallen expert consensus on the primary therapy for early breast cancer considers appropriate the integration of molecular assays to pathology assessment [51]. Two ongoing clinical trials were designed to address the question of outcome, for which the gene signature was selected. The Microarray in Node Negative and 0 to 3 Positive Lymph Node Disease May Avoid Chemotherapy Trial (MINDACT) is being conducted in Europe with the 70-gene assay [52]. The Trial Assigning Individualized Options for Treatment (TAILORx) uses the 21-gene assay and is being run in the United States [53]. The tissue bank that has been established for the MINDACT trial, together with the publicly available genomic library, will be of important scientific value for future research.

The development of gene signatures must be done together with cost-effectiveness analysis. Increasing health care costs is nowadays not just a problem of developing countries, but a global issue. An economic analysis demonstrated that Oncotype Dx®, if applied appropriately, is predicted to increase quality-adjusted survival and save costs [54]. The application of GGI to a restricted population of hormone receptor positive and histologic grade 2 breast cancer could be a cost-effective strategy, limiting the use of this tool to the population for which the treatment decision-making process remains most challenging.

Conflict of interest statement 

Otto Metzger Filho has no conflict of interest to be declared. Michail Ignatiadis has no conflict of interest to be declared. Christos Sotiriou is the inventor of Genomic Grade Index.

Reviewers 

Dr. Joseph Gligorov, APHP Tenon, University of Paris VI, Dept. of Medical Oncology CancerEst, 4 rue de la Chine, FR-75970 Paris Cedex 20, France.

Dr. Edi Brogi, Memorial Sloan-Kettering Cancer Center, Department of Pathology, New York, NY 10021, United States.