Assessment of the performance of the Stanford online calculator for the prediction of non-sentinel lymph node metastasis in sentinel node positive breast cancer patients.

Abstract

Abstract Abstract #1005 Background: Several models for the prediction of non-sentinel lymph node metastasis (+NSLN) in sentinel lymph node positive (+SLN) breast cancer patients have been published. Multiple variables have been shown to be important in prediction of +NSLN. To this date, models that utilize 7-8 variables have been significantly more discriminatory than models using only 2-3 variables. In this study, we evaluate the performance of the newly created Stanford online calculator (SOC) to predict the likelihood of +NSLN. The SOC uses 3 variables: primary tumor size, sentinel lymph node metastasis size, and presence/absence of lymphovascular invasion.
 Methods: From 1997-2004, 465 breast cancer patients with clinically negative axillae were found to have a +SLN and underwent completion axillary lymph node dissection (ALND) at the Mayo Clinic. Using the SOC the probability of having +NSLN was determined for each patient. Complete variable data was available for 464 patients which make up the cohort reported here. A receiver-operating characteristic (ROC) curve was created and the area-under-the-curve (AUC) was calculated. Mean probabilities of patients with and without +NSLN were compared. Additionally, patients with a low Stanford probability were examined to determine the model's accuracy in discriminating patients who could potentially be spared ALND. These results were compared to the results of the Mayo Clinic and Memorial Sloan Kettering Cancer Center (MSKCC) nomograms which have been previously applied to this same cohort of patients.
 Results: The AUC of the Stanford model was 0.72 (95% CI 0.67-0.77). The mean Stanford probabilities for patients with and without NSLN disease were 0.75 (range 0.06 to 1.0) and 0.50 (range 0.05 to 1.0), respectively (p < 0.0001). Examining patients with a Stanford probability <= 10%, we found that 47 patients met this criterion; 6 had +NSLN and 41 did not, for a false negative (FN) prediction rate of 13%. There was only 1 patient with Stanford probability <= 5%, and this patient did not have +NSLN. The AUC of the MSKCC and Mayo nomograms in this same patient population was 0.74 and 0.77, respectively. The AUCs did not differ significantly (p = 0.13) among the 3 models.
 
 Conclusion: Despite using only three variables, the Stanford nomogram appears to perform as well as, but not better than the MSKCC and Mayo nomograms in our patient cohort. Further validation in other patient populations is needed prior to widespread utilization of this nomogram. Citation Information: Cancer Res 2009;69(2 Suppl):Abstract nr 1005.