Abstract A31: Risk stratification of ductal carcinoma in situ: Analytical validation of a prognostic test analyzing live-primary cells via phenotypic biomarkers and machine learning at single-cell resolution

Largely due to tumor heterogeneity, risk stratification of patients diagnosed with ductal carcinoma in situ (DCIS) of the breast remains a significant challenge. Management of DCIS is also problematic as we wish to personalize treatment of a patient’s tumor in order to avoid overtreatment of lower-risk lesions or undertreatment of DCIS that may recur or progress into invasive cancer. Matching treatment to the underlying severity of the illness is key to practicing cost-effective cancer care in an era where this is a very large concern to society. The aim of this study was to analytically validate a precision risk-stratification tool based on phenotype, which is capable of predicting which patients will develop invasive cancer with greater than 80% sensitivity and specificity. Leveraging the novel capability to rapidly culture primary breast biopsy cells, we present a “biopsy-on-a-chip” microfluidic platform that quantifies dynamic and static phenotypic biomarkers via machine vision to generate predictive clinical scores via machine learning algorithms to determine if a DCIS patient will experience invasive cancer. 47 consecutive lumpectomy or mastectomy samples were collected and objectively analyzed in a blinded study, measuring 1000 phenotypic biomarkers with single-cell resolution using machine vision software. Biomarker measurements were input into machine learning algorithms to develop predictive statistical algorithms. Statistical algorithms were able to independently predict surgical adverse pathology features such as extranodal extension, grade, lymphovascular invasion, lymph invasion, lobular carcinoma in situ (LCIS), and DCIS with sensitivities and specificities greater than 90%. Additional machine learning based algorithms were able to predict if DCIS patients were more likely to develop subsequent metastasis as measured by lymphovascular invasion and/or lymphatic invasion with area under the curve (AUC) > 0.85. This study is the first study to demonstrate the prediction of breast cancer adverse pathology features from live primary biopsy cells and provides the basis to develop a powerful precision risk-stratification tool to risk-stratify DCIS. Furthermore, the methodology described and its ability to rapidly analyze primary breast biopsy tissue with single-cell resolution in a high-throughput manner engenders a powerful research tool to further understand tumor heterogeneity in breast cancer towards the development of personalized therapeutics. Applications of cost effectiveness analysis to our methodology will achieve the triple goal of providing cost-effective, patient-centered, and appropriate breast cancer and DCIS care. Note: This abstract was not presented at the conference. Citation Format: Ashok Chander, Michael Manak, Jonathan Varsanik, Brad Hogan, Grannum Sant, Kevin Knopf. Risk stratification of ductal carcinoma in situ: Analytical validation of a prognostic test analyzing live-primary cells via phenotypic biomarkers and machine learning at single-cell resolution [abstract]. In: Proceedings of the AACR Special Conference: Advances in Breast Cancer Research; 2017 Oct 7-10; Hollywood, CA. Philadelphia (PA): AACR; Mol Cancer Res 2018;16(8_Suppl):Abstract nr A31.