Purpose: Three models have been developed to predict four or more involved axillary lymph nodes (ALNs) in patients with breast cancer with one to three involved sentinel lymph nodes (SLNs). Two scores were developed by Chagpar et al (Louisville scores excluding or including method of detection), and a nomogram was developed by Katz et al. The purpose of our investigation was to compare these models in a prospective, multicenter study.
Patients and methods: Our study involved a cohort of 536 patients having one to three involved SLNs who underwent ALN dissection. We evaluated the area under the receiver operating characteristic curve (AUC), calibration (for the Katz nomogram only), false-negative (FN) rate, and clinical utility of the three models. Results were compared with the optimal logistic regression (OLR) model that was developed from the validation cohort.
Results: Among the 536 patients, 57 patients (10.6%) had > or = four involved ALNs. The AUC for the Katz nomogram was 0.84 (95% CI, 0.81 to 0.86). The Louisville score excluding method of detection was 0.75 (95% CI, 0.72 to 0.78). The Louisville score including method of detection was 0.77 (95% CI, 0.74 to 0.79). The FN rates were 2.5% (eight of 321 patients), 1.8% (two of 109 patients), and 0% (zero of 27 patients) for the Katz nomogram and the Louisville scores excluding and including method of detection, respectively. The Katz nomogram was well calibrated. Optimism-corrected bootstrap estimate AUC of the OLR model was 0.86. Using this result as a reasonable target for an external model, the performance of the Katz nomogram was remarkable.
Conclusion: We validated the three models for their use in clinical practice. The Katz nomogram outperformed the two other models.