Abstract
BACKGROUND: Individualized decisions are required in early-stage breast cancer patients. We aimed to establish a novel model for predicting non-sentinel lymph node (SLN) metastases in patients with positive SLNs, using preoperative and intraoperative characteristics and inflammatory indicators. METHODS: The data of 489 patients with invasive breast cancer were retrospectively collected from Xuanwu Hospital between 2014 and 2021. Among them, 96 patients with at least one positive SLN were used to build the predictive model. Univariate and multivariate analyses were performed to identify the risk factors of non-SLN metastases. A nomogram was developed using these risk factors and was validated by calibration curves. The area under the receiver operating characteristics curve (AUC) and decision curve analyses (DCA) were used to compare our novel nomogram with the Memorial Sloan Kettering Cancer Center (MSKCC) nomogram. Cross-validation was performed for further internal validation of the predictive model. External validation was conducted using another treatment group (n=46 patients) in Xuanwu Hospital. RESULTS: Non-SLN metastases occurred in 42 of the 83 patients with positive SLNs (50.6%). Multivariate stepwise logistic regression indicated that the risk factors were age (P=0.032), number of positive SLNs (P=0.020), number of negative SLNs (P=0.011), resected tumor size (P=0.038), and monocyte count (P=0.012). A predictive model was developed and virtualized by nomogram using these five risk factors. The AUC of our nomogram was 0.867, which was significantly higher than that of the MSKCC model. DCA also showed a superior clinical value for our novel nomogram. After 10-fold cross-validation with 400 times repetitions, the AUC of our model was still 0.830. External validation of our model showed an AUC of 0.727. The model was well-calibrated in the internal and external validation series. CONCLUSIONS: A five-factor nomogram was developed for predicting non-SLN metastases in early-stage breast cancer patients. This novel tool exhibited good accuracy and could assist clinicians with intraoperative decisions in breast cancer patients with positive SLNs.