Abstract P1-01-09: Which nomograms may be the best for predicting nonsentinel lymph node metastasis in breast cancer patients: a meta-analysis.

Abstract

Abstract Background: Axillary lymph node dissection (ALND) is the standard treatment for breast cancer patients with positive sentinel lymph nodes (SLNs). Several nomograms were developed to identify SLN-positive patients with low risk of nonsentinel lymph nodes (NSLNs) metastasis. These nomograms were validated in different populations and it is still unknown which is the best. This study is to present a systemic review and perform a meta-analysis to obtained the pooled AUC (Area Under the receiver-operator Curve) value of each models. Methods: This review focused on six models: Cambridge, MSKCC, Mayo, MDA, Tenon and Stanford models. A “Pubmed” search and “Web of science” search were conducted and 35 literatures were ultimately included. AUC and the number of patients with positive NSLNs were extracted. Publication bias and heterogeneity were analyzed. AUCs were converted to odds ratios (ORs) for combination. The combined ORs were converted back to AUCs to represent the integrated discriminative capabilities of each models. Findings: In total, the Cambridge, Mayo, MDA, MSKCC, Stanford and Tenon models were validated in 8, 6, 4, 39, 14 and 15 studies, with 2156, 2431, 843, 8143, 3700 and 3648 patients included, respectively. There were no publication bias or heterogeneity observed in the Cambridge, Mayo, MDA and Tenon models (Table 1). The combined ORs and the corresponding AUCs of each models were listed as follow: Cambridge (OR = 3.86, AUC = 0.71), Mayo (OR = 3.71, AUC = 0.71), MSKCC (OR = 3.47, AUC = 0.70), MDA(OR = 3.44, AUC = 0.70), Tenon (OR = 3.46, AUC = 0.70) and Stanford (OR = 2.92, AUC = 0.67). For each of the predictive models, both fixed and random effect models were used to calculate the combined OR. The presence of larger difference between the fixed and random effect analysis suggests small study effects, rendering the meta-analysis relatively less reliable. The combined ORs were identical when fixed and random effect models were used in the Cambridge and MDA models, suggesting that there was no small study effects in these two models. Conclusions: All of the included models are all better than random chance but not provide excellent discriminative capabilities. The Cambridge and Stanford models were relatively superior and inferior when compared with the other models, respectively. Citation Information: Cancer Res 2012;72(24 Suppl):Abstract nr P1-01-09.