In patients with breast cancer treated with neoadjuvant chemotherapy (NACT), a positive sentinel lymph node (SLN) usually requires completion axillary lymph node dissection (ALND). To enable de-escalation of this traumatic surgery, we aimed to develop a model to accurately estimate the likelihood of axillary disease after a positive SLN biopsy in the NACT setting. We retrospectively analyzed clinicopathological data from 237 patients composed of a training set of 150 patients from a single institution and a validation set of 87 patients from 2 other institutions. Variables that were collected included the histologic type, lymphovascular invasion, the number of lymph nodes (LNs) (SLN and non-SLN), positive and negative, with and without treatment effect, extranodal extension, and the calculated residual cancer burden of the largest SLN metastasis. Residual axillary disease was defined as ≥1 positive LNs in the completion ALND specimen. Univariable and multivariable statistical analyses were performed. Then, a formula for the risk of predicted probability of residual axillary disease was created using a stepwise feature selection based on the Akaike Information Criterion to select variables in the model. Residual axillary disease was identified in 120 out of 237 (50.6%) cases (73 [48.7%] in the training set and 47 [54%] in the validation set). Independent predictors of residual axillary disease in the multivariable model included the greatest dimension of the largest SLN metastasis, lymphovascular invasion, greater number of positive LNs with no treatment effect, greater number of positive LNs with treatment effect, greater number of negative LNs with treatment effect, and fewer number of negative LNs. These variables along with residual cancer burden of the largest SLN metastasis and histologic type were incorporated into the final model by stepwise feature selection. The predictive formula achieved an area under the curve of 77.6% for the training set and 69.7% for the validation set. A predicted probability value of ≤20% yielded a negative predictive value of 86.5% in the training set and 64.7% in the validation set. This corresponds to 37 (25.3%) patients who could be spared ALND in the training set and 17 (19.5%) in the validation set. Using the formula, a subset of patients treated with NACT could be spared unnecessary ALND.
扫码关注我们
求助内容:
应助结果提醒方式:
