Simple prediction models for breast cancer patients with solitary positive sentinel nodes--are they valid?

Identification and prediction of breast cancer patients with metastases isolated to the sentinel lymph node(s) would potentially allow avoidance of axillary dissection and its complications. In this study, we evaluate the performance of two recently published models (Alkhatib et al. and Chagpar et al.) that attempt to predict patients who have isolated sentinel lymph node metastases. Both of these models reported a 5% rate of positive nonsentinel nodes in their respective lowest risk category. From 1997 to 2004, 465 breast cancer patients had a positive sentinel node and underwent axillary lymph node dissection at Mayo Clinic. To evaluate the Alkhatib model, patients were assigned to the following groups: group 1: 1 positive sentinel node and > or =1 negative sentinel node(s); group 2: >1 positive sentinel node and > or =1 negative sentinel node(s); group 3: 1 positive sentinel node and no negative sentinel node(s); group 4: >1 positive sentinel node and no negative sentinel node(s). To evaluate the Chagpar model, patients were assigned a score based on the sum of three factors: tumor size (T1a = 1, T1b or T1c = 2, T2 = 3, and T3 = 4 points), number of positive sentinel nodes (>1 positive sentinel node = 1 point), and ratio of positive/total sentinel nodes (>50% positive = 1 point). The chi-square test was used to compare our results to those of the original studies. For the Alkhatib model, we found that 30% (p < 0.0001) of Group 1 patients had nonsentinel node metastases. Using the Chagpar model, the percentage of patients with a cumulative score of 1 with a nonsentinel node metastasis was 17% (p = 0.19). In our cohort, neither model accurately predicted which patients have a < or = 5% chance of having nonsentinel metastases. These models are not adequate to identify patients in whom axillary dissection can be omitted.