Can We Predict Which Women with a Negative Sentinel Lymph Node Biopsy Are at High Risk of a False Negative Result?

Abstract Background: Standard of care for women with a positive sentinel lymph node biopsy (SLNB) is to have a completion axillary lymph node dissection (cALND). For women with a negative SLNB, cALND is often not performed, accepting that a proportion will have a false negative (FN) result. FN rates are commonly reported based on surgeon experience. In the absence of cALND, information on FN rates will not be available, and other means of assessing the risk for FN SLNB is needed.Materials and Methods: Between May 1999 and December 2006, 1661 women with early-stage breast cancer that had undergone SLNB followed by cALND, were identified from our provincial database: 77 FN, 560 true positive (TP), and 1024 true negative (TN). FN cases were matched 1:3 with TN cases by date of SLNB. Chi-square and Wilcoxon Rank-sum tests were used to screen variables and those with moderate association were identified and included in subsequent models. Logistic regression was used to develop a multivariable model to predict the probability of FN vs. TN status. ROC curves were used to estimate the optimal probability cut-off, at which sensitivity (SN) and specificity (SP) were maximized. The model's performance was then assessed using a cross-validation technique.Results: Factors examined that did not significantly affect FN vs. TN status rate included: age, body mass index, previous breast surgery, histology, estrogen receptor status, margin status, tumor palpability, injection technique (peritumoral, periareolar), mapping agent used (yes/no), and SLNB done pre vs. post breast surgery (all p=NS). Factors identified that significantly affected FN vs. TN status (p<0.05) and thus included for potential use in the model included: tumor size, tumor grade, lymphovascular invasion (+/-LVI), tumor site (central/medial, lateral, other), type of breast surgery (mastectomy vs. lumpectomy), pre-operative lymphoscintiscan (yes/no), colloid (yes/no), number of SLN (1 vs. >1). The final model contained 5 variables: T stage (1 vs. 2), tumor grade, number of SLN removed, tumor site, and LVI. ROC identified an optimal probability cut-point for this model of 0.217 (21.7% risk of FN) with a corresponding SN of 71% and SP of 70%. In the cross-validation, the model correctly classified 66% of cases (SN of 73%, SP of 64%) with an AUC of 0.76. With increasing FN risk, SN declined and SP increased such that at a FN risk of 30%, this model had a SN 56%, SP 77%, and accuracy 72%. Using this model a woman with a T1, lateral, grade 3, 1 SLN and LVI- had a predicted FN risk of 21.5%, increasing to 52.2% if she were LVI+.Discussion: For women with early breast cancer, a negative SLNB result has a significant impact on prognosis and recommendations for further systemic and radiation therapy, this model (T size, tumor grade, number of SLN removed, tumor site, and LVI) is the first that would offer a quantitative prediction of FN risk in this setting, which could influence further discussion and therapeutic decision making. Potential refinements of this model will be explored, incorporating 'lower-priority' variables. Citation Information: Cancer Res 2009;69(24 Suppl):Abstract nr 303.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.