PURPOSE: To compare the outcomes of the available systems that predict the risk of non-sentinel lymph node (non-SLN) metastasis and to evaluate the variability within a group of SLN-positive breast cancer patients. METHODS: Predicted probabilities and scores for non-SLN metastasis were calculated with nine predictive systems for 120 SLN-positive patients who underwent a completion axillary lymph node dissection. The number of patients was calculated that were considered low risk or had a probability of ≤ 10% by at least one of the systems. For each nomogram, a box plot was constructed. All patients with a predicted probability of ≤ 10% according to the Memorial Sloan-Kettering Cancer Center (MSKCC) nomogram were selected, and a comparison was made with the probabilities predicted by the other systems. RESULTS: Nearly two-thirds (64.2%, n = 77) of patients with SLN-positive breast cancer were allocated to a low-risk or low-probability group by at least one of the predictive systems. No patients were uniformly classified as low risk by all nine prediction models. At the group level, a considerable variation in the distribution of the predicted probabilities was observed. At the individual level, calculation of the predicted probabilities for the selected patients who were considered low risk (≤ 10%) according to the MSKCC nomogram, showed even larger variations, ranging from 4 to 94%. CONCLUSIONS: This study shows that there is an unacceptably high variability in individual predictions when the predictive systems that are currently available are used to predict non-SLN metastasis in patients with SLN-positive breast cancer.
PURPOSE: To compare the outcomes of the available systems that predict the risk of non-sentinel lymph node (non-SLN) metastasis and to evaluate the variability within a group of SLN-positive breast cancerpatients. METHODS: Predicted probabilities and scores for non-SLN metastasis were calculated with nine predictive systems for 120 SLN-positive patients who underwent a completion axillary lymph node dissection. The number of patients was calculated that were considered low risk or had a probability of ≤ 10% by at least one of the systems. For each nomogram, a box plot was constructed. All patients with a predicted probability of ≤ 10% according to the Memorial Sloan-Kettering Cancer Center (MSKCC) nomogram were selected, and a comparison was made with the probabilities predicted by the other systems. RESULTS: Nearly two-thirds (64.2%, n = 77) of patients with SLN-positive breast cancer were allocated to a low-risk or low-probability group by at least one of the predictive systems. No patients were uniformly classified as low risk by all nine prediction models. At the group level, a considerable variation in the distribution of the predicted probabilities was observed. At the individual level, calculation of the predicted probabilities for the selected patients who were considered low risk (≤ 10%) according to the MSKCC nomogram, showed even larger variations, ranging from 4 to 94%. CONCLUSIONS: This study shows that there is an unacceptably high variability in individual predictions when the predictive systems that are currently available are used to predict non-SLN metastasis in patients with SLN-positive breast cancer.