Objective To explore value of prediction model established based on ultrasound features combined with clinical data in evaluating axillary lymph node (ALN) metastasis in patients with early breast cancer.

Methods The ultrasonic characteristics and clinical data of 203 women with unilateral early breast cancer were retrospectively analyzed. The patients were divided into metastatic group and non-metastatic group, and were determined whether they had ALN metastasis or not according to the pathological results. Single factor screening was performed for each index of the two groups. Logistic multivariate regression analysis was performed again and a prediction model was established. Receiver operating characteristic (ROC) curve was used to detect its discrimination, and goodness-of-fit test was used to evaluate the degree of calibration. Another 78 unilateral early breast cancer patients in our hospital were selected for clinical validation of the model.

Results The pathological results of 203 breast cancer patients showed that ALN metastasis occurred in 71 cases(metastasis group), accounting for 34.98%. Tumor diameter ≥3 cm, blurred tumor margin, longer ALN short diameter, higher ALN short diameter to long diameter ratio, higher value of ALN cortical thickness, lower degree of differentiation, and higher level of serum microRNA-21(miRNA-21) expression were all risk factors for ALN metastasis in breast cancer patients (P<0.05). According to the risk factors, the prediction model expression equation was as follows. Logit(P)=1.912×tumor diameter ≥3 cm(yes=1, no=0)+2.040×tumor margin blur(yes=1, no=0)+1.582×ALN short diameter(measured value)+3.374×ALN short/long diameter(measured value)+2.264×ALN cortical thickness(measured value)+2.497×differentiation degree(yes=1, no=0)+2.921×miRNA-21 expression amount(measured value)-33.615. The area of the ROC curve of this model was 0.886 (95%CI, 0.838 to 0.933), the sensitivity and specificity corresponding to the maximum Youden index (0.736) were 88.50% and 83.60% respectively. Goodness of fit test showed that the model did not overfit (χ²=2.067, P=0.394). Clinical validation results showed that the sensitivity of the model was 87.10%, the specificity was 82.98%, and the accuracy was 84.62%.

Conclusion It is valuable in predicting the risk of ALN metastasis by constructing a predictive model based on degree of tumor differentiation, serum miRNA-21 expression, tumor diameter, tumor margin, and ALN short-diameter, short-diameter/long-diameter ratio, and cortical thickness in early breast cancer patients.