Open AccessThe Molecular Subtype Classification Is a Determinant of Sentinel Node Positivity in Early Breast Carcinoma
Author(s) -
Fabien Reyal,
Roman Rouzier,
Berenice Depont-Hazelzet,
Marc A. Bollet,
JeanYves Pierga,
S. Alran,
Rémy Salmon,
Virginie Fourchotte,
Anne VincentSalomon,
Xavier Sastre-Garau,
Martine Antoine,
Serge Uzan,
Brigitte SigalZafrani,
Yann De Rycke
Publication year - 2011
Publication title -
plos one
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.99
H-Index - 332
ISSN - 1932-6203
DOI - 10.1371/journal.pone.0020297
Subject(s) - breast cancer , sentinel node , covariate , logistic regression , medicine , multivariate statistics , multivariate analysis , biopsy , breast carcinoma , oncology , stage (stratigraphy) , node (physics) , metastasis , cancer , statistics , biology , mathematics , physics , paleontology , quantum mechanics
Several authors have underscored a strong relation between the molecular subtypes and the axillary status of breast cancer patients. The aim of our work was to decipher the interaction between this classification and the probability of a positive sentinel node biopsy. Materials and Methods Our dataset consisted of a total number of 2654 early-stage breast cancer patients. Patients treated at first by conservative breast surgery plus sentinel node biopsies were selected. A multivariate logistic regression model was trained and validated. Interaction covariate between ER and HER2 markers was a forced input of this model. The performance of the multivariate model in the training and the two validation sets was analyzed in terms of discrimination and calibration. Probability of axillary metastasis was detailed for each molecular subtype. Results The interaction covariate between ER and HER2 status was a stronger predictor (p = 0.0031) of positive sentinel node biopsy than the ER status by itself (p = 0.016). A multivariate model to determine the probability of sentinel node positivity was defined with the following variables; tumour size, lympho-vascular invasion, molecular subtypes and age at diagnosis. This model showed similar results in terms of discrimination (AUC = 0.72/0.73/0.72) and calibration (HL p = 0.28/0.05/0.11) in the training and validation sets. The interaction between molecular subtypes, tumour size and sentinel nodes status was approximated. Discussion We showed that biologically-driven analyses are able to build new models with higher performance in terms of breast cancer axillary status prediction. The molecular subtype classification strongly interacts with the axillary and distant metastasis process.