To the Editor:

Breast cancer microarray analyses have generated a large volume of data allowing for the introduction of new concepts in genetic classification of breast tumors and a better understanding of tumor pathogenesis. Nevertheless, more debatable at this stage is the clinical relevance for the application of genetic prognostic signatures to predict the outcome of the individual patient in everyday practice.

In the long process of prognostic factor validation,¹ some steps in the demonstration that gene expression signatures are superior to classical prognostic factors may have been overlooked in many of the reported microarray studies.

In an interesting analysis, Loi et al² shed a new light on the molecular luminal subtype classification by using a genomic grade index (GGI) based on the expression of 97 genes, differentially expressed between histologic grade 1 and 3 tumors. In this study, two subtypes of estrogen receptor–positive tumors could be distinguished roughly reproducing luminal A and B classification. Furthermore, the authors stated that GGI was the strongest prognostic factor for metastasis, both in univariate and in multivariate analyses involving traditional clinical prognostic factors. GGI was initially performed to individualize two different prognostic subgroups among patients with histologic grade 2 (hG2: differentiation, nuclear plemorphism and mitotic activity total score of 6 and 7) tumors.³ In a similar approach, Ivshina et al separated hG2 tumors into two highly discriminant classes based on a gene expression signature.⁴

It is interesting to note that most of the genes included in these signatures are associated with cell progression and proliferation. Consequently, and quite logically, the prognostic value of GGI is not surprising. However, demonstrating that GGI is superior to classical prognostic factors by comparing it with each single individual prognostic factor only, may not be sufficient. In our opinion, this multigene technology should have been tested against a combinative classical factor index such as the Nottingham Prognostic Index (NPI), St Gallen criteria, or Adjuvant! Online. As previously demonstrated, when classical factors are correctly combined, such as the NPI or in Artificial Neural Network indexes, they show similar power as cDNA profilers in predicting breast cancer prognosis.^5,6

While there is no evidence enough to use GGI in clinical practice at the present time, these results may provide new insights on the use of good old clinicopathologic parameters. For example, comparing conventional pathological factors according to genomic classification, Ivshina et al stated on the importance of proliferation, nuclear pleomorphism and progesterone receptor expression for this subclassification.⁴

We tried to use these results to individualize two different prognostic subgroups among patients with estrogen receptor–positive hG2 tumors without any other adverse prognostic factors, in order to optimize indication of adjuvant chemotherapy.

We analyzed 398 patients primarily operated on at our institute for a G2 pT1N0M0 estrogen receptor–positive breast carcinoma without lymphovascular invasion who did not receive any adjuvant medical treatment. With a 139-month median follow-up, those with negative progesterone receptor status and a total grade score equal to 7 (14% of the patients) had a significant worse prognosis than positive progesterone receptor and/or total grade score equal to 6: 10-year metastases-free survival were 69.0% versus 87.7% (log-rank hazard ratio: 2.2; 95% CI, 1.19 to 4.09; P = .011; Fig 1).

Consequently, we would be interested if Loi et al investigated the differences between these two molecular subtypes using conventional clinicopathologic variables, especially proliferation factors.

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The author(s) indicated no potential conflicts of interest.