The Diagnostic Accuracy of Magnetic Resonance Imaging in Predicting Pathologic Complete Response after Neoadjuvant Chemotherapy in Patients with Different Molecular Subtypes of Breast Cancer.
Xinfeng Zhang,Dandan Wang,Zhuangkai Liu,Zheng Wang,Qiang Li,Hong Xu,Bin Zhang,Ting Liu,Feng Jin
DOI: https://doi.org/10.21037/qims.2019.11.16
2020-01-01
Quantitative Imaging in Medicine and Surgery
Abstract:Background: Patients treated with neoadjuvant chemotherapy (NAC) who achieve a pathologic complete response (pCR) can be identified preoperatively and can potentially be spared the morbidity of surgery. The objective of this retrospective study was to estimate the diagnostic accuracy of preoperative magnetic resonance imaging (MRI) in predicting pCR in patients with different molecular subtypes of breast cancer and to provide a basis for the selection of surgical methods. Methods: We retrospectively reviewed breast MRI data from August 2015 to December 2018 of patients who underwent four or more cycles of NAC. Factors associated with radiological complete response (rCR) and pCR were analyzed in univariable and multivariable settings. The accuracy of MRI and the correlation between rCR and pCR were also analyzed in each tumor subtype. Results: A total of 177 women with a primary tumor fulfilled the study criteria; 18 of these patients (10.2%) achieved rCR, and 21 (11.9%) achieved a pCR. MRI diagnosis of rCR was significantly correlated with pCR with a Spearman's correlation coefficient of 0.686 in the entire population. The sensitivity, specificity, accuracy, pCR predictive value (PPV), and non-pCR predictive value (NPV) were estimated to be 66.67%, 97.44%, 93.79%, 77.78%, and 95.60%, respectively. Statistically significant correlations between rCR and pCR were found in Luminal B high Ki67% (P<0.001), HER2-positive (P=0.0035), and triple-negative (P<0.001) subtypes, but not in Luminal A and Luminal B low Ki67% subtypes. On univariate analysis, the tumor characteristics significantly associated with both rCR and pCR were small tumor, lymph node metastasis (LNM) negativity, early clinical stage, high grade, high Ki67% index, and different molecular subtype. On multivariate logistic regression analysis, grade 3 tumors (P=0.013), Ki67% >= 40% (P<0.000), and stage I tumor (P=0.006) were independently associated with rCR. However, grade 3 tumors (P=0.001), triple-negative breast cancer (TNBC), and clinical stages I and II tumors (P=0.003; P=0.030) were independently associated with the likelihood of attaining a pCR. Conclusions: The overall accuracy of MRI in predicting pCR in invasive breast cancer patients who received NAC was 93.8%. The performance of MRI differed among molecular subtypes, and the highest PPV was found in TNBC (100%) and Luminal B high Ki67% (75%) subtypes.