Prediction of High Nodal Burden in Patients With Sentinel Node-Positive Luminal ERBB2-Negative Breast Cancer

Abstract

Importance: In patients with clinically node-negative (cN0) breast cancer and 1 or 2 sentinel lymph node (SLN) macrometastases, omitting completion axillary lymph node dissection (CALND) is standard. High nodal burden (≥4 axillary nodal metastases) is an indication for intensified treatment in luminal breast cancer; hence, abstaining from CALND may result in undertreatment.

Objective: To develop a prediction model for high nodal burden in luminal ERBB2-negative breast cancer (all histologic types and lobular breast cancer separately) without CALND.

Design, setting, and participants: The prospective Sentinel Node Biopsy in Breast Cancer: Omission of Axillary Clearance After Macrometastases (SENOMAC) trial randomized patients 1:1 to CALND or its omission from January 2015 to December 2021 among adult patients with cN0 T1-T3 breast cancer and 1 or 2 SLN macrometastases across 5 European countries. The cohort was randomly split into training (80%) and test (20%) sets, with equal proportions of high nodal burden. Prediction models were developed by multivariable logistic regression in the complete luminal ERBB2-negative cohort and a lobular breast cancer subgroup. Nomograms were constructed. The present diagnostic/prognostic study presents the results of a prespecified secondary analysis of the SENOMAC trial. Herein, only patients with luminal ERBB2-negative tumors assigned to CALND were selected. Data analysis for this article took place from June 2023 to April 2024.

Exposure: Predictors of high nodal burden.

Main outcomes and measures: High nodal burden was defined as ≥4 axillary nodal metastases. The luminal prediction model was evaluated regarding discrimination and calibration.

Results: Of 1010 patients (median [range] age, 61 [34-90] years; 1006 [99.6%] female and 4 [0.4%] male), 138 (13.7%) had a high nodal burden and 212 (21.0%) had lobular breast cancer. The model in the training set (n = 804) included number of SLN macrometastases, presence of SLN micrometastases, SLN ratio, presence of SLN extracapsular extension, and tumor size (not included in lobular subgroup). Upon validation in the test set (n = 201), the area under the receiver operating characteristic curve (AUC) was 0.74 (95% CI, 0.62-0.85) and the calibration was satisfactory. At a sensitivity threshold of ≥80%, all but 5 low-risk patients were correctly classified corresponding to a negative predictive value of 94%. The prediction model for the lobular subgroup reached an AUC of 0.74 (95% CI, 0.66-0.83).

Conclusions and relevance: The predictive models and nomograms may facilitate systemic treatment decisions without exposing patients to the risk of arm morbidity due to CALND. External validation is needed.

Trial registration: ClinicalTrials.gov Identifier: NCT02240472.

Figure 1.. Discrimination and Calibration of the Prediction Models for High Nodal Burden

Threshold refers to the predicted probability of high nodal burden (≥pN2). In the calibration plot, the dashed oblique reference line indicates perfect calibration with an intercept of 0 and a slope of 1. The cluster of points in the lower left corner of the figure indicates that observed, as well as predicted, risk of high nodal burden is low for most patients in the cohort. In addition, the absence of a linear trend at this end of the risk spectrum suggests that the discrimination is poor for low-risk patients. However, the points in the upper right corner indicate that the model might be useful for identification of patients at high risk of high nodal burden, which was the aim of the present work. AUC indicates area under the ROC curve; FN, false negative; FP, false positive; N, nodal status; ILC, invasive lobular carcinoma; NPV, negative predicted value; PPV, positive predicted value; TN, true negative; TP, true positive; ROC, receiver operating characteristic curve.

Figure 2.. Nomogram for Predicting the Probability of High Nodal Burden

SLN ratio refers to (SLN micrometastases + SLN marcrometastases) / (all removed lymph nodes at SLN biopsy). SLN indicates sentinel lymph node.

Figure 3.. Prognostic Value of the High Nodal Burden Prediction Model

Patients were followed up from randomization to the date of first recurrence or death, defined according to the Standardized Definitions for Efficacy End Points (STEEP) criteria or, for event-free patients, to the date of last follow-up. Contralateral breast cancer was not considered a recurrence-free survival (RFS) event. Follow-up times for recurrence-free and alive participants were censored at the date of last visit. No follow-up times were censored for other reasons. The proportional hazards assumption was fulfilled for both Cox models (P = .48 in model with continuous score and P = .41 with score categories). Risk groups were created by dividing the total nomogram score into 3 groups: low (quartile 1), medium (quartiles 2-3), and high (quartile 4). The hazard ratio (HR) was calculated for a 10-unit nomogram score increase. ^aDifferent spline functions were assessed using Akaike Information Criterion (AIC), but a linear model for total points gave the best fit to the data after adjustment for model complexity. ^bInterpretation of this HR: for each 10-step increase of total nomogram points, the incidence of RFS events increases with 7%.