John Davis1, Tanya L Hoskin2, Courtney N Day2, Mark Wickre3, Mara A Piltin1, Abigail S Caudle4, Judy C Boughey5. 1. Department of Surgery, Mayo Clinic, Rochester, MN, USA. 2. Department of Health Science Research and Clinical Statistics, Mayo Clinic, Rochester, MN, USA. 3. Department of Radiology, Mayo Clinic, Rochester, MN, USA. 4. Department of Breast Surgical Oncology, University of Texas MD Anderson Cancer Center, Houston, TX, USA. 5. Department of Surgery, Mayo Clinic, Rochester, MN, USA. Boughey.judy@mayo.edu.
Abstract
INTRODUCTION: Prediction models are useful to guide decision making. Our goal was to compare three published nomograms predicting axillary response to neoadjuvant chemotherapy (NAC), clinically node-positive breast cancer. METHODS: Patients with cT1-T4, cN1-N3 breast cancer treated with NAC and surgery from 2008 to 2019 were reviewed. The predicted probability of pathologic node-negative (ypN0) status was estimated for each nomogram. Area under the curve (AUC) was compared across models, overall and by biologic subtype. RESULTS: Of 581 patients, 253 (43.5%) were ypN0. ypN0 status varied by subtype: 23.9% for estrogen receptor-positive (ER+)/human epidermal growth factor receptor 2-negative (HER2-), 68.9% for HER2-positive (HER2+), and 47.2% for ER-negative (ER-)/HER2-. The three nomograms had similar AUC values (0.761-0.769; p = 0.80). The Mayo model-predicted probability was significantly lower (p < 0.001) than the observed probability of ypN0 status, while the MD Anderson Cancer Center (MDACC) 1- and 2-predicted probabilities were similar to the observed probability. At a predicted probability threshold of 50%, the Mayo model had the highest sensitivity (89.6%) for detecting ypN+ patients compared with MDACC models 1 and 2 (76.5%; p < 0.001). However, both MDACC models had higher specificity in identifying ypN0 status among HER2+ (81.7%) and ER-/HER2- (75.9-77.6%) patients compared with the Mayo model (59.5% and 43.1%; each p < 0.001). None of the models identified the ER+/HER2- patients with ypN0 status well at the ≥ 50% threshold (specificity 0-9.4%). CONCLUSION: All three models predicting nodal response to NAC performed well overall with respect to discrimination, but differed with respect to calibration and performance at a 50% probability threshold. However, none of the models performed well at the 50% threshold for ER+/HER2- patients.
INTRODUCTION: Prediction models are useful to guide decision making. Our goal was to compare three published nomograms predicting axillary response to neoadjuvant chemotherapy (NAC), clinically node-positive breast cancer. METHODS:Patients with cT1-T4, cN1-N3 breast cancer treated with NAC and surgery from 2008 to 2019 were reviewed. The predicted probability of pathologic node-negative (ypN0) status was estimated for each nomogram. Area under the curve (AUC) was compared across models, overall and by biologic subtype. RESULTS: Of 581 patients, 253 (43.5%) were ypN0. ypN0 status varied by subtype: 23.9% for estrogen receptor-positive (ER+)/humanepidermal growth factor receptor 2-negative (HER2-), 68.9% for HER2-positive (HER2+), and 47.2% for ER-negative (ER-)/HER2-. The three nomograms had similar AUC values (0.761-0.769; p = 0.80). The Mayo model-predicted probability was significantly lower (p < 0.001) than the observed probability of ypN0 status, while the MD Anderson Cancer Center (MDACC) 1- and 2-predicted probabilities were similar to the observed probability. At a predicted probability threshold of 50%, the Mayo model had the highest sensitivity (89.6%) for detecting ypN+ patients compared with MDACC models 1 and 2 (76.5%; p < 0.001). However, both MDACC models had higher specificity in identifying ypN0 status among HER2+ (81.7%) and ER-/HER2- (75.9-77.6%) patients compared with the Mayo model (59.5% and 43.1%; each p < 0.001). None of the models identified the ER+/HER2- patients with ypN0 status well at the ≥ 50% threshold (specificity 0-9.4%). CONCLUSION: All three models predicting nodal response to NAC performed well overall with respect to discrimination, but differed with respect to calibration and performance at a 50% probability threshold. However, none of the models performed well at the 50% threshold for ER+/HER2- patients.