Accuracy of a Preoperative Model for Predicting Invasive Breast Cancer in Women with Ductal Carcinoma-in-situ on Vacuum-Assisted Core Needle Biopsy

Abstract

Core needle biopsy (CNB) diagnoses of ductal carcinoma-in-situ (DCIS) may represent understaged invasive breast cancer (IBC). We aimed to develop a model that helps identify preoperatively women with IBC after a CNB diagnosis of DCIS. Retrospective study of all women with DCIS on vacuum-assisted CNB of microcalcifications (1999-2008), with prospective classification of imaging variables independently by two radiologists. Variables included lesion size and level of suspicion on imaging, morphology and distribution of microcalcifications, DCIS nuclear grade on CNB, number of cores, and age. Multivariate logistic regression models of the probability of IBC were developed; the accuracy of these models was examined for each radiologist. Excision histology showed IBC in 77 (17.4%) of 442 subjects with DCIS on CNB. Lesion size on imaging yielded the best model fit and highest accuracy, and had the highest agreement between radiologists. Addition of grade to a model which included size improved model fit (P<0.0001). However, model fit and accuracy were not improved by inclusion of any other variables. A model based on size and grade had similar areas under the receiver operating characteristic curve (accuracy of 74%) for each radiologist. Modeled sensitivity, specificity, and predictive values for different combinations of size and grade thresholds are reported. If the imaging lesion is >50mm and the CNB grade is high, the model's positive predictive value is ≥50%. A model based on imaging size of microcalcifications and CNB nuclear grade can identify women at high risk of having IBC with moderate accuracy and may be used to guide informed preoperative discussion in women with newly diagnosed DCIS on CNB.