PD02-08: Validation over Time of a Nomogram Predicting the Sentinel Node Positivity in Early Breast Carcinoma According to the Molecular Subtypes Classification

Background: The molecular subtypes of breast cancer have different axillary status. A new nomogram including the interaction covariate between estrogen receptor (ER) and HER2 status has been recently published (Reyal et al. PLOSone, may 2011) and allows to identify before surgery the patients with a high risk of positive sentinel lymph node (SLN). The purpose of our study was to validate this model on an independent population. Patients and methods: We studied 755 consecutive patients treated for operable breast cancer with sentinel node biopsies in 2009, from the Institut Curie breast cancer prospective database. Baseline characteristics were compared between our population and the population used to build the model, using Chi-square test for categorical variables and Kruskal-Wallis test for continuous variables. The multivariate model, including age, tumor size, lymphovascular invasion and interaction covariate between ER and HER2 status, was used to calculate the theoretical risk of positive sentinel lymph node (SLN) for all patients. The performance of the model on our population was then evaluated in terms of discrimination (area under the curve AUC) and of calibration (Hosmer-Lemeshow HL test). Results: Characteristics of our population were significantly different from the training population for the following variables: tumor size (median 12mm [1-60] versus 13mm [1-100] p=0.005), lymphovascular invasion (18.6% versus 23.7% p=0.006), positive ER (91.4% versus 87% p=0.002) and age as followed: 56.7% of patients ≤ 60 versus 63.1%, 17.5% of patients between 60 and 65 versus 14.1% and 25.8% of patients above 65 versus 22.8% p=0.01. The nomogram showed similar results in our population than in the training population in terms of discrimination (AUC= 0.72 [0.68−0.76] versus 0.73 [0.7−0.75] and calibration (HL p= 0.4 versus p=0.35). Conclusions: Despite significant differences between the two populations concerning variables which are part of the nomogram, the model was validated in our population. Our study shows that this nomogram is efficient and robust over time to predict the likelihood of positive SLN according to molecular subtypes defined by surrogate markers ER and HER2 determined by immunohistochemistry in clinical practice. Citation Information: Cancer Res 2011;71(24 Suppl):Abstract nr PD02-08.