Predictors of Lymph Node Metastasis in Primary Breast Cancer - Risk Models for Tailored Axillary Management

Abstract: Most patients with breast cancer present with low-risk tumors, node-negative disease, and excellent prognosis. For these patients, routine axillary nodal staging by sentinel lymph node biopsy (SLNB) has no therapeutic benefit. For patients with limited sentinel lymph node metastasis, completion axillary nodal dissection is controversial. Furthermore, those with heavy-burden metastasis could benefit from preoperative selection for neoadjuvant treatment and/or more extensive axillary nodal excision rather than SLNB.This thesis present results on the utility of axillary ultrasonography (AUS), as well as novel prediction models for the estimating disease-free axilla, limited axillary nodal metastasis, and heavy-burden axillary nodal metastasis.Study I The sensitivity of AUS to detect metastatic nodal disease was low with a high false negative rate. Axillary metastatic burden, defined by metastatic size and number of involved nodes, was the most important predictor of an abnormal AUS. This suggest that AUS is unreliable in patients with low metastatic burden. Histological grade was found to be an independent factor that effected the accuracy of AUS performance. Patients with HER2-positive tumors were found to have higher rates of AUS abnormalities. The overall axillary metastatic burden was higher in patients with preoperative verified nodal metastasis by AUS-guided biopsy compared with those with normal AUS findings but with metastatic sentinel lymph node.Study II Breast cancer surrogate molecular subtypes, age, mode of detection, tumor size, multifocality, and vascular invasion were identified as predictors of nodal disease in patients with T1-T2 breast cancer. Three nomograms that included these predictors were developed to predict disease-free axilla N0, limited axillary nodal metastasis (1-2 positive lymph nodes), and heavy-burden axillary nodal metastasis (≥ 3 positive lymph nodes). Area under the ROC curves (AUCs) ranged from 0.70–0.81. The increase in tumor size was found to be less often associated with metastatic nodal involvement in the TNBC subtype than in other non-TNBC subtypes.Study III Clinicopathological characteristics were incorporated into artificial neural network models to predict disease-free axilla N0, low-burden metastasis (1-3 positive nodes), and heavy-burden metastasis (≥ 4 positive nodes) in patients with clinically node-negative breast cancer. Tumor size, LVI, and multifocality displayed linear correlation patterns to the nodal status end-points, while other predictors (age, histological type, ER status, PR status, Ki-67 values, mode of detection, and tumor localization in the breast) revealed non-linear dynamic associations. The clinical utility of reducing unnecessary SLNB was assessed; a cut-off value according to maximum negative predictive value or false-negative rate of 5–10% in a model to discriminate disease-free axilla yielded a SLNB reduction rate of 8–27%.Study IV Predictors of nodal metastasis were assessed using clinicopathological characteristics, gene expression data, and combined features. In the overall validation cohort, the predictor with combined features showed the highest discriminative performance (AUC 0.72). However, discriminatory performances were highly similar using clinicopathological predictors alone across the surrogate molecular subtypes based on the ER, PR, and HER2 status. Higher proportions of the luminal B intrinsic features and proliferation-related genes were observed in predicted node-positive ER+HER2- and HER2+ tumors, while low-expression of basal-like markers were observed in predicted node-positive TNBC tumors. In conclusion, these studies demonstrate the ability to estimate axillary nodal burden using preoperatively obtainable predictors and highlight nonlinear associations between clinicopathological variables and nodal metastasis. Preoperative prediction of the nodal status would facilitate individualized axillary management.