Mammography is an early detection modality for breast cancer that is implemented widely in the United States, has established benchmarks of performance, and in most studies throughout the world has been demonstrated to reduce mortality due to the disease. This relatively inexpensive x-ray imaging of the breast also provides a location that can be directly sampled through needle biopsy which leads generally to an unambiguous pathologic diagnosis of invasive cancer, carcinoma in situ, or benign findings. No system is perfect and mammographic screening, particularly in the US, prompts over 1.6 million biopsies per year detecting approximately 230,000 invasive and 60,000 non-invasive cancers for a positive predictive value of less than 20%. There may be substantial room to improve on this and reduce the number of biopsies but this improvement must not sacrifice detection rates so the negative predictive value (NPV, identification of true negatives) must remain very high. In this Biomarker Development Laboratory application, we propose to test whether a combination of mammographic feature analysis and candidate biomarkers that we have identified can achieve an NPV that would be acceptable to patients and providers to prevent unnecessary breast biopsies. One of the biomarkers is a type of circulating giant cell termed ?Cancer Associated Macrophage Like? (CAML) that can only be detected using freshly drawn whole blood, we propose to conduct a prospective trial at Duke University in women undergoing breast cancer diagnosis. Our realistic goal is to accrue ~1000 women over the course of 4 years for which full field digital mammography has been performed. The images will undergo feature extraction for decision modeling. Blood will be analyzed for the presence and type of CAML cells, immunosignaturing using the high density peptide arrays developed by Stephen Johnston at Arizona State, and measurements of two specific analytes that have the highest sensitivity and specificity for basal type cancers, CA125 and TP53 autoantibodies. As feature analysis from imaging alone can achieve, at least for masses on mammography, an AUC of ~0.9, the study is designed to determine whether the biomarkers have sufficient complementary information to the imaging and each other to increase the AUC to 0.95 allowing us to identify a threshold where there is a 98% NPV. We will make use of the most careful and consistent standard operating procedures, the best candidate biomarkers, and the most well developed imaging algorithms to make this a definitive study.
Statement Over 1.6 million breast biopsies are performed each year in the United States detecting less than 300,000 cancers (total of invasive and non-invasive cancers). The vast majority of these biopsies occur after mammographic detection of a suspicious area in the breast. The relatively low positive predictive value of mammography based on these numbers results from the careful avoidance of missing a cancer leading to a vast number of negative biopsies. The goal of our proposal is to evaluate and quantify the most promising approaches to discriminate benign versus breast cancer in a prospective study. We intend to combine imaging features with a series of candidate biomarkers to determine whether sufficient negative predictive value can be reached to substantially reduce the number of breast biopsies. Out study will have major public health and societal impact if a multi-modal assessment can increase the predictive accuracy of standard imaging to 98% negative predictive value in women who are recommended for breast biopsy.
