Predicting Sentinel Lymph Node And Non-sentinel Lymph Node Metastasis In A Chinese Breast Cancer Population:Development And Assessment Of Two New Predictive Nomograms

Part I:Predicting sentinel lymph node metastasis in a Chinese breast cancer population:development and assessment of a new predictive nomogram[Objective] The Memorial Sloan-Kettering Cancer Center (MSKCC) has developed a nomogram to predict the presence of sentinel lymph node (SLN) metastasis in breast cancer patients. In our study, we assessed the MSKCC nomogram performance in predicting SLN metastases in a Chinese breast cancer population. A new model (the Shanghai Cancer Hospital sentinel lymph node Nomogram, SCH-SLN nomogram) was developed; it employs clinically relevant variables and offers possible advantages over the MSKCC nomogram.[Methods] Data were collected from1545patients who had a successful SLN biopsy and were treated between March2005and November2011. Touch imprint cytology (TIC) and serial sectioning with H&E staining were routinely performed on each sentinel node. A total of966SLN biopsy procedures, conducted between March2005and December2010were used as the modeling group for validating the MSKCC nomogram. Clinical and pathologic features of SLN biopsy in modeling group were assessed with multivariable logistic regression to predict the presence of SLN metastasis in breast cancer.The SCH-SLN nomogram was created from the logistic regression model. This new model was subsequently applied to545consecutive SLN biopsies from January2011to November2011, and these biopsies were used as the validation group. The predictive accuracy of the MSKCC and SCH-SLN nomogram was measured by calculating the area under the receiver-operating characteristic (ROC) curve (AUC).[Results] Age, tumor size, tumor location, tumor type, histological grade, lymphovascular invasion and neural invasion were correlated with the likelihood of SLN metastasis in the univariate analyses (P<0.05). Using the multivariate analysis, tumor size, tumor location, tumor type and lymphovascular invasion were identified as independent predictors of SLN metastasis. The SCH-SLN nomogram was then developed using the five variables associated with SLN metastasis:age, tumor size, tumor location, tumor type and lymphovascular invasion. The new model was accurate and discriminating (with an AUC of0.7649when applied to the modeling group) compared to the MSKCC nomogram (which had an AUC of0.7105in the modeling group). The area under the ROC curve for the SCH-SLN nomogram in the validation population is0.7587. The actual probability trends for the various deciles were comparable to the predicted probabilities. The false-negative rates of the SCH-SLN nomogram were3.54%and8.20%for the predicted probability cut-off points of10%and15%, respectively.[Conclusion] To our knowledge, this is the first study designed to evaluate the MSKCC nomogram and to develop a new nomogram in the early breast cancer population in China. Compared with the MSKCC nomogram, the SCH-SLN nomogram has a better AUC with fewer variables and has lower false-negative rates for the low-probability subgroups. The SCH-SLN nomogram could serve as a more acceptable clinical tool in preoperative discussions with patients, especially very-low-risk patients. When applied to these patients, the SCH-SLN nomogram could be used to safely avoid a SLN procedure, thereby reducing the postoperative morbidity rate (which has been reported in the literature to be as high as8%). Although the SCH- SLN nomogram performed well at predicting sentinel lymph node metastasis in the Chinese breast cancer population, the nomogram is imperfect. For example, the SCH-SLN nomogram was developed and validated at a single institute; before it is applied widely, the nomogram should be validated in various patient populations to demonstrate its reproducibility.Part ⅡPredicting non-sentinel lymph node metastasis in a Chinese breast cancer population with a positive sentinel node:development and assessment of a new predictive nomogram[Objective] We assessed the MSKCC nomogram performance in predicting non-SLN metastases in a Chinese breast cancer population with a positive sentinel lymph node. A new model (the Shanghai Cancer Hospital non-sentinel lymph node Nomogram, SCH-NSLN nomogram) was developed; it employs clinically relevant variables and offers possible advantages over the MSKCC nomogram.[Methods] Data were collected from1545patients who had a successful SLN biopsy and were treated between March2005and November2011.443among them were diagnosis of a positive lymph node and thus underwent axillary lymph node dissection. TIC and serial sectioning with H&E staining were routinely performed on each sentinel node. A total of300SLN biopsy procedures, conducted between March2005and December2010were used as the modeling group for validating the MSKCC nomogram. Clinical and pathologic features of patients in modeling group were assessed with multivariable logistic regression to predict the presence of non-SLN metastasis in breast cancer.The SCH-NSLN nomogram was created from the logistic regression model. This new model was subsequently applied to143patients from January2011to November2011, and these patients were enrolled in the validation group. The predictive accuracy of the MSKCC and SCH-NSLN nomogram was measured by calculating the area under the receiver-operating characteristic (ROC) curve (AUC).[Results]Tumor size, tumor type, lymphovascular invasion, neural invasion, number of positive SLNs, number of negative SLNs and the SLN metastasis size were correlated with the likelihood of non-SLN metastasis in the univariate analyses (P<0.05). Using the multivariate analysis, tumor size, lymphovascular invasion,number of positive SLNs, number of negative SLNs and the SLN metastasis size were identified as independent predictors of SLN metastasis. The SCH-NSLN nomogram was then developed using the five variables associated with non-SLN metastasis: tumor size, lymphovascular invasion,number of positive SLNs, number of negative SLNs and the SLN metastasis size. The new model was accurate and discriminating (with an AUC of0.7900when applied to the modeling group) compared to the MSKCC nomogram (which had an AUC of0.7843in the modeling group). The area under the ROC curve for the SCH-NSLN nomogram in the validation population is0.7777. The actual probability trends for the various deciles were comparable to the predicted probabilities. The false-negative rates of the SCH-NSLN nomogram were8.40%for the predicted probability cut-off points of14%, when applied to the SLN positive patients. The false-negative rates of the SCH-NSLN nomogram were only5.20%for the predicted probability cut-off points of7%, when applied to the breast cancer patients with a SLN micrometastasis or isolate tumor cells (ITC) only.[Conclusion]Compared with the MSKCC nomogram, the SCH-NSLN nomogram has a better AUC with fewer variables and has lower false-negative rates for the low-probability subgroups. The SCH-NSLN nomogram could serve as a more acceptable clinical tool in clinical discussions with patients, especially very-low-risk patients. When applied to these patients, especially for patients with SLN micrometastasis or ITC, the SCH-NSLN nomogram could be used to safely avoid ALND procedure, thereby reducing the postoperative morbidity rate.