Development of a Model to Predict Invasiveness in Ductal Carcinoma In Situ Diagnosed by Percutaneous Biopsy-Original Study and Critical Evaluation of the Literature

Abstract

Our aim was to develop a model to predict invasiveness in patients with a diagnosis of ductal carcinoma in situ found at percutaneous biopsy. The calculated sample size was 296 patients. We used Nagelkerke's R-2 and Hosmer-Lemeshow goodness-of-fit tests to improve statistical analysis. We evaluated 354 patients and developed 2 models that have the best discrimination reported to date. Background: Approximately 30% of ductal carcinoma in situ (DCIS) cases have an invasive component discovered on the final analysis that could affect surgical management. The aims of the present study were to determine the risk factors associated with the underestimation of DCIS and to develop a model to predict the probability of invasiveness. Materials and Methods: A retrospective analysis was performed on the data for all patients with a diagnosis of DCIS found by percutaneous biopsy from January 2008 to February 2016. Thirteen potential predictors of invasiveness were examined. The statistical analysis of the present study was improved using Nagelkerke's R-2, the area under the receiving operating characteristic (AUC) curve, and the Hosmer-Lemeshow goodness-of-fit test. Results: Of 354 biopsy specimens deemed to be DCIS on initial biopsy, 100 (28.2%) were recategorized as invasive carcinoma after surgery. On multivariate analysis, the strongest predictors of invasiveness were comedonecrosis, size on mammography, suspected microinvasion, histologic grade, and younger patient age. The model had a good discriminative ability, with an AUC of 0.764. The overall performance of the model was fair, with a Nagelkerke's R-2 of 40.9%. A separate analysis performed on 274 specimens obtained through vacuum-assisted biopsy revealed different variables were associated with underestimation; however, a similar AUC (0.743) and Nagelkerke's R-2 (45.7% ) were obtained. Conclusion: Our model had the best AUC for predicting DCIS invasiveness reported to date. However, further statistical analysis showed only a fair overall performance. The currently known clinical, radiographic, and pathologic features might be insufficient to identify which patients with DCIS have underestimated disease.