Axillary lymph node dissection in the case of sentinel lymph node micrometastatic invasion: evaluation of three predictive models.

Objective: To evaluate three predictive risk models of non-sentinel lymph node (NSLN) involvement in the case of micrometastatic sentinel node (SLN) involvement for breast cancer.

Study Design: This retrospective study included 72 successive patients with micrometastatic SLN involvement who had surgery between March 1996 and October 2007. All patients had undergone immediate or delayed axillary lymph node dissection (ALND). The Memorial Sloan-Kettering Cancer Center (MSKCC) nomogram, the Stanford nomogram and the Tenon score were applied to the population to calculate the probability of NSLN involvement.

Results: For the MSKCC nomogram with a threshold value of 10%, sensitivity was 50%, specificity was 70% and the negative predictive value (NPV) was 89%. The area under the receiver operating characteristic curve (AUC) was 0.6 (significant). Use of this nomogram would have avoided ALND in 49 out of 72 (68%) patients, but five out of 10 (50%) patients with NSLN involvement would not have been detected. With a threshold value of 7%, the AUC was 0.69, sensitivity was 90% and NPV was 97%. ALND would have been avoided in 31 out of 72 (43%) patients, with a 3% chance of leaving metastases when abstaining from ALND. For the Tenon score with a threshold value of 3.5, sensitivity was 50%, specificity was 72% and the AUC was 0.62. This was not clinically applicable because eight out of 10 (80%) patients with NSLN involvement would not have been detected. For the Stanford nomogram, the results could not be interpreted because the AUC was not significant.

Conclusion: None of the tested models are sufficiently reliable for use in daily practice. The MSKCC nomogram showed the most encouraging results, especially for a threshold value of 7%, but this has not been validated in the literature. Complete axillary dissection should be performed in the case of micrometastatic SLN involvement until more data become available.