Predicting axillary sentinel node status in patients with primary breast cancer.

The aim of this study is to determine the combination of characteristics in early breast cancer that could estimate the risk of occurrence of metastatic cells in axillary sentinel lymph node(s). If we were able to reliably predict the presence or absence of axillary sentinel involvement, we could spare a considerable proportion of patients from axillary surgery without compromising therapeutic outcomes of their disease. The study is based on retrospective analysis of medical records of 170 patients diagnosed with primary breast cancer. These women underwent primary surgery of the breast and axilla in which at least one sentinel lymph node was obtained. Logistic regression has been employed to construct a model predicting axillary sentinel lymph node involvement using preoperative and postoperative tumor characteristics. Postoperative model uses tumor features obtained from definitive histology samples. Its predictive capability expressed by receiver operating characteristic curve is good, area under curve (AUC) equals to 0.78. The comparison between preoperative and postoperative results showed the only significant differences in values of histopathological grading; we have considered grading not reliably stated before surgery. In preoperative model only the characteristics available and reliably stated at the time of diagnoses were used. The predictive capability of this model is only fair when using the data available at the time of diagnosis (AUC = 0.66). We conclude, that predictive models based on postoperative values enable to reliably estimate the likelihood of occurrence of axillary sentinel node(s) metastases. This can be used in clinical practice in case surgical procedure is divided into two steps, breast surgery first and axillary surgery thereafter. Even if preoperative values were not significantly different from postoperative ones (except for grading), the preoperative model predictive capability is lower compared to postoperative values. The reason for this worse prediction was identified in imperfect preoperative diagnostic.