BACKGROUND: Axillary treatment of breast cancer patients is undergoing a paradigm shift, as completion axillary lymph node dissections (ALNDs) are being questioned in the treatment of patients with tumor-positive sentinel nodes. This study aims to develop a novel multi-institutional predictive tool to calculate patient-specific risk of residual axillary disease after tumor-positive sentinel node biopsy. METHODS: Breast cancer patients with a tumor-positive sentinel node and a completion ALND from five European centers formed the original patient series (N = 1000). Statistically significant variables predicting nonsentinel node involvement were identified in logistic regression analysis. A multivariable predictive model was developed and validated by area under the receiver operating characteristics curve (AUC), first internally in 500 additional patients and then externally in 1068 patients from other centers. All statistical tests were two-sided. RESULTS: Nine tumor- and sentinel node-specific variables were identified as statistically significant factors predicting nonsentinel node involvement in logistic regression analysis. A resulting predictive model applied to the internal validation series resulted in an AUC of 0.714 (95% confidence interval [CI] = 0.665 to 0.763). For the external validation series, the AUC was 0.719 (95% CI = 0.689 to 0.750). The model was well calibrated in the external validation series. CONCLUSIONS: We present a novel, international, multicenter, predictive tool to assess the risk of additional axillary metastases after tumor-positive sentinel node biopsy in breast cancer. The predictive model performed well in internal and external validation but needs to be further studied in each center before application to clinical use.
BACKGROUND: Axillary treatment of breast cancerpatients is undergoing a paradigm shift, as completion axillary lymph node dissections (ALNDs) are being questioned in the treatment of patients with tumor-positive sentinel nodes. This study aims to develop a novel multi-institutional predictive tool to calculate patient-specific risk of residual axillary disease after tumor-positive sentinel node biopsy. METHODS:Breast cancerpatients with a tumor-positive sentinel node and a completion ALND from five European centers formed the original patient series (N = 1000). Statistically significant variables predicting nonsentinel node involvement were identified in logistic regression analysis. A multivariable predictive model was developed and validated by area under the receiver operating characteristics curve (AUC), first internally in 500 additional patients and then externally in 1068 patients from other centers. All statistical tests were two-sided. RESULTS: Nine tumor- and sentinel node-specific variables were identified as statistically significant factors predicting nonsentinel node involvement in logistic regression analysis. A resulting predictive model applied to the internal validation series resulted in an AUC of 0.714 (95% confidence interval [CI] = 0.665 to 0.763). For the external validation series, the AUC was 0.719 (95% CI = 0.689 to 0.750). The model was well calibrated in the external validation series. CONCLUSIONS: We present a novel, international, multicenter, predictive tool to assess the risk of additional axillary metastases after tumor-positive sentinel node biopsy in breast cancer. The predictive model performed well in internal and external validation but needs to be further studied in each center before application to clinical use.