Objective To analyze the factors influencing axillary lymph-node metastasis (ALNM) in microinvasive breast cancer (MIBC) patients, as well as to establish the risk-prediction model of ipsilateral ALNM in MIBC patients.

Methods A total of 4475 primary female MIBC diagnosed by pathology from 2010 to 2015 were searched and screened from the SEER database. The obtained data were used to establish a prediction model for ALNM of MIBC. A total of 2266 primary female MIBC patients diagnosed by pathology from 2018 to 2020 in the SEER database were screened as the external validation cohort. The clinicopathological data of the enrolled MIBC patients were collected. Univariate analysis was used to screen out the factors affecting ALNM in MIBC patients. Statistically significant variables in univariate analysis were included in multivariate logistic regression analysis. The independent factors influencing ALNM in MIBC patients were screened out, and a nomogram was established. The area under the curve (AUC) was calculated by plotting the ROC curve. After plotting the calibration curve, the model was evaluated by Hosmer-Lemeshow goodness-of-fit test.

Results A total of 309 patients were diagnosed with ipsilateral ALNM among the 6741 MIBC patients, accounting for about 4.58%. Univariate analysis of the modeling group showed that age, ethnicity, histological grade, pathological type, molecular subtype, and lateral side were associated with ALNM in MIBC patients (P<0.05). Results of multivariate analysis showed that the risk of ALNM was higher in MIBC patients ≤40 years old, black people, histological grade Ⅱ and Ⅲ, and primary right side (P<0.05). Subtype is an independent factor influencing ALNM in MIBC patients. However, the difference in ALNM risk was not statistically significant between the subtype of HR+HER2+, HR−HER2+, HR−HER2− and HR+HER2− subtypes. The AUC of the modeling group was 0.716 (95%CI: 0.682-0.750), the best cut-off value was 0.045, the sensitivity was 0.733, and the specificity was 0.608. The newly established nomogram model was used for the validation cohort, and its AUC was 0.722 (95%CI: 0.667-0.777). The P values of the Hosmer-Lemeshow goodness-of-fit test of the calibration curves in the modeling and validation groups exceeded 0.05.

Conclusion The risk-prediction model of ALNM in MIBC patients established by the SEER database has good predictive ability and can thus be expected to serve as a reference for clinical practice.