Abstract P6-01-06: Identification of Molecular Subtypes of DCIS and Invasive Breast Cancer Using Computerized Image Analysis for Data Integration

Background: Breast cancer is a heterogeneous disease with different putative therapeutic targets based upon the particular subtype. The particular subtype can have profound implications for therapy, as in the case of basal-like carcinomas which tend to be more aggressive with less therapeutic options. Although many studies have examined the transition from Ductal Carcinoma in-situ (DCIS) to Invasive Breast Cancer (IBC), the mechanisms involving the transformation are poorly understood. DCIS cases that are more likely to progress to IBC can be treated more aggressively. This study used image analysis to examine the association between predictive molecular biomarkers between pure DCIS, DCIS associated with IBC and IBC. Methods: With approval by the Institutional Review Board, we evaluated tissue microarrays comprising 453 tissue cores from 149 patients. Five immunohistochemical biomarkers were utilized: estrogen (ER), progesterone (PR), human epidermal growth factor receptor 2 (HER2), epidermal growth factor receptor (EGFR), and cytokeratin 5/6 (CK-5/6). We then employed the Automated Cellular Image Analysis (ACIS) to score the biomarker status. We compared regional scoring (rated 0, 1, 2, or 3) between an experienced pathologist9s subjective assessment and an automated score derived from ACIS. An image library was generated and integrated to our Translational Data Mart (TraM: http://tram.uchicago. edu). Statistical analyses included a comparison of histological grade, race, age, tumor size, and lymph node status associated with the prevalence of subtypes between pure DCIS tumors and those that advanced to IBCs. Results: The concordance between ACIS-based and pathologist-based scoring was moderate (kappa = 0.38). There was more variability in the pathologist9s scoring with a standard deviation of 1.14 as compared to 0.85 for ACIS. Subtypes associated with IBC were also present in DCIS. In the pure DCIS cases, there was a higher proportion of the luminal A subtype (74.1%, n = 54) as compared to DCIS adjacent to IBC (54.2%, n=24) or pure IBC cases (57.4%, n = 115) (p=0.05). In contrast, there was more basal-like subtypes in DCIS juxtaposed with IBC (20.8%) and IBC (27.8%) than in pure DCIS tumors (9.3%). Discussion: Molecular subtype analysis of DCIS could be useful in predicting DCIS with high risk of progressing to invasive cancer. The resemblance between subtypes bolsters the hypothesis that IBCs and DCIS originate from the same precursor lesion. Although the automated scoring featured by ACIS would be useful in expediting readings, it must be validated by a pathologist9s assessment. Citation Information: Cancer Res 2010;70(24 Suppl):Abstract nr P6-01-06.