Breast Cancer Molecular Subtype Classification: A Comparison of Three Microarray-Based Single Sample Predictors
B Weigelt, A Mackay, R Natrajan, DS Tan, A Ashworth, JS Reis-Filho. CRUK London Research Institute, London, United Kingdom; ICR, London, United Kingdom
Background: Microarray gene expression profiling has led to the identification of five molecular subtypes of breast cancer: luminal A, luminal B, HER2, normal breast-like and basal-like. Three single sample predictors (SSPs) have been described by the proponents of this taxonomy for the identification of molecular subtypes. The aim of this study was to determine the agreement between these SSPs in the identification of breast cancer molecular classes.
Design: Microarray-based SSPs were applied to one in-house and three publicly available breast cancer microarray datasets based on the methods described by the proponents of the molecular taxonomy. Analysis of agreement between each pair of SSPs was performed for the whole classification system and for each molecular subtype individually in each cohort. Hazard ratios for outcome for each molecular subtype according to each SSP in each cohort were calculated. For the in-house dataset, HER2 status was defined by immunohistochemistry (IHC) and fluorescence in situ hybridisation (FISH).
Results: A fair-to-substantial agreement between each pair of SSPs in each cohort was observed (Kappa scores ranging from 0.238 to 0.740). The proportion of cases classified as basal-like subtype in each cohort was consistent regardless of the SSP employed, however the percentages of the remaining molecular subtypes varied significantly. Of the five molecular subtypes, only basal-like cancers consistently showed an almost perfect agreement (Kappa scores >0.810). Most importantly, the significance of the associations with outcome of each molecular subtype other than basal-like and luminal A varied significantly depending on the SSP employed. The identification of HER2-positive tumours by SSPs was unreliable, given that one SSP assigned only 54% of the IHC/FISH HER2-positive cases to the HER2 molecular subtype, whereas the other two SSPs assigned all IHC/FISH HER2-positive cases to the luminal B subtype.
Conclusions: Basal-like cancers can be reproducibly identified by microarray-based gene expression profiling. The classification of breast cancers into molecular subtypes shows considerable variation depending on the SSP used. Before molecular classification by microarray profiling can be incorporated in routine clinical practice and treatment decision-making, accurate definitions of the methods used to identify the molecular subtypes of breast cancer should be standardised and rigorously validated.
Monday, March 22, 2010 1:00 PM
Poster Session II # 54, Monday Afternoon