Comparison of Prognostic Gene Expression signatures for Breast Cancer

Haibe-Kains B, C. Desmedt, G. Bontempi and C. Sotiriou

Background: During the last years, several groups have identified prognostic gene expression signatures that consistently outperform traditional clinical parameters (e.g. tumor size, age, and histological grade) and guidelines (e.g. Adjuvant! Online). Previous publications reported that gene signatures exhibited similar classification and performance. Although these studies yielded promising conclusions, some issues remained open: (i) the dataset which was considered for these studies was the one employed also for the identification of some gene sets and as such it could not be considered as an independent test set  and/or (ii) since some of these signatures were developed on another platform, the initial algorithms could not be applied due to different or missing probe sequences and difference in gene expressions measurement. Therefore, signatures were never compared on an independent population of untreated breast cancer patients, where classification was computed using the original algorithms and microarray platforms.

Patients and methods: We compared three gene expression signatures, the 70-gene, the 76-gene and the genomic grade signatures, in terms of predicting time to distant metastasis (TDM), distant metastasis free survival (DMFS), and overall survival (OS) for the individual patient. To this end, we used the previously published TRANSBIG independent validation series of node-negative untreated primary breast cancer patients. We used Cramer's V statistic to quantify the strength of the association between two gene signature classifications, i.e. the low- and high-risk groups of patients. We estimated hazard ratios between the low- and high-risk groups, as well as the hazard ratios adjusted for clinical risk as defined by Adjuvant! Online, using the Cox model. We used the concordance index to quantify the predictive ability of a survival model. Standard errors, confidence intervals and p-values for the concordance index were computed making an assumption of asymptotic normality. The difference in hazard ratios and concordance indices were computed using a Student t test for two dependent samples.

Results: The three evaluated signatures had high rates of concordance in their prediction of clinical outcome, with the 70-gene and genomic grade signatures showing the highest rate, 89%, compared to 71% and 78 % for the 76- and the 70-gene signatures, and the 76-gene and genomic grade signatures, respectively. The three signatures had similar capabilities of predicting TDM, DMFS, and OS in adding significant prognostic information to that provided by the classical parameters.

Conclusions: Despite the difference in development of these signatures and the small overlap in gene identity, they showed similar prognostic performance, confirming that these prognostic signatures are of wide clinical relevance. Moreover, the nature of the genomic grade signature suggested that proliferation might be the common driving force of several prognostic signatures.