Abstract
Introduction
Molecular assays such as the 70-gene signature are increasingly used as prognostic indicators to select chemotherapy in individual patients. These assays are typically derived from postoperative excision specimens and require several weeks to complete. Earlier assessment of the results of such assays could open up new therapeutic options in subgroups of patients, potentially avoiding overtreatment of early breast cancer. Although molecular assays may be derived from biopsied tissue, tumor heterogeneity may cause uncertainty. Dynamic contrast-enhanced MRI (DCE-MRI) depicts some of the hallmarks of cancer that are tested by these molecular essays. The goal of this study was to investigate the association between the postoperatively derived 70-gene signature and computer-derived DCE-MRI features of the ipsilateral breast prior to surgery.
Material and Methods
Sixty-nine patients with node-negative invasive breast cancer were enrolled between 2003 and 2006. These patients received a preoperative MRI in study setting and a postoperative 70-gene signature assay. Association between preoperative features and the 70-gene signature was evaluated using a computer prediction model combining clinical features and automatically extracted MRI features. The clinical features were age at diagnosis and largest tumor diameter on MRI. The MRI features were rate of contrast uptake in the tumor, rate of wash-out, tumor volume, and two features from the intramammary blood vessel tree (total length and mean rate of contrast uptake). The features were transformed into an orthogonal feature set using principal component analysis. Association with the 70-gene signature (positive or negative indication for systemic therapy) was evaluated using binary logistic regression. Model performance was measured using the area under the receiver operating characteristics curve (AUC) after bootstrap validation using 200 iterations. Two operating points were examined: one to predict a positive 70-gene signature with high certainty (i.e., at high positive-predictive value (PPV)) and one to predict a negative signature with high certainty (i.e., at high negative-predictive value (PPV)).
Results
The average patient age at diagnosis was 48 years (range: 32-58). The median largest tumor diameter on MRI was 17 mm (range: 5-40). The 70-gene signature was positive in 29/69 (42%) patients. The computer prediction model achieved an AUC of 0.72 after bootstrap validation. At high PPV, 10/29 (34.5%) positive 70-gene signatures were identified preoperatively at the expense of 2/40 (5.0%) false-positive. The PPV was 10/12 (83.3%). At high NPV, 12/40 (30.0%) negative 70-gene signatures were identified at the expense of 1/29 (3.5%) false-negative signature. The NPV was 12/13 (92.3%).
Conclusion
Computer-derived DCE-MRI features from the ipsilateral breast in combination with clinical parameters show potential to preoperatively assess a negative outcome of the 70-gene signature in approximately one-third of the total patient group.
Citation Format: van der Velden BHM, Schmitz AMTh, Loo CE, Gilhuijs KGA. Association between computer-derived features of the ipsilateral breast on DCE-MRI and the 70-gene signature in patients with invasive breast cancer. [abstract]. In: Proceedings of the Thirty-Eighth Annual CTRC-AACR San Antonio Breast Cancer Symposium: 2015 Dec 8-12; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2016;76(4 Suppl):Abstract nr P4-02-07.