Predictions for positive nodes in breast cancer using computerized modeling.

Abstract

16 Background: The accuracy and consistency of sentinel lymph node (SLN) biopsy for axillary staging in breast cancer has been well established by multiple studies. Although the standard of care has been to perform a completion axillary lymph node dissection (ALND) on patients who have a positive sentinel node, recent data (ACOSG Z0010, Z0011 and NSABP B-32) suggests that an ALND may be avoided in certain subsets of patients. To further study this question, we used a Bayesian model to predict the probability of finding disease in the ALND for breast cancer. Methods: All SLN procedures performed by a single surgeon, for clinical T1-T2 N0 disease between September 2000 and March 2009 were retrospectively reviewed under an IRB approved protocol. Demographic, disease and surgical procedural variables were collected. Values are reported as mean ± standard deviation. A Bayesian model, a standard statistical model frequently used in medicine, was used to identify variables that could predict a positive ALND. Results: A total number of 235 SLN procedures were reviewed. The mean patient age was 58.6 ± 11.8 (range 28-88), tumor size was 1.8 ± 1.3 cm (range 0.2-7.4), and BMI was 28 ± 6.1. The number of SLN found was 3.6 ± 1.9 (range 0-10), and in 2 cases no SLN were found, for an overall failure rate of 0.85%. A total of 73 ALND were performed in this group; 27 cases had ALND as national study participants randomized to ALND, 44 cases for positive SLN, and 2 cases in which no SLN could be identified. The mean number of axillary lymph nodes removed was 14.8 ± 7 and the mean number of positive axillary nodes was 2.7 ± 4.1. Tumor size, tumor grade, number of positive SLN by hematoxylin and eosin staining, and a low technetium count of the first SLN were found to be significant predictors of positive axillary nodes during a completion ALND. There was also a trend for progesterone receptor expression as a predictor of axillary disease. Conclusions: A model to predict a positive ALND for breast cancer is presented. We found tumor characteristics, number of positive SLN and low technetium count for first SLN to be predictors for finding axillary disease. Further studies are needed to validate our model as a means of sparing women an ALND who have a low probability of finding additional breast disease.