Abstract
BACKGROUND: Breast fibroepithelial lesions are biphasic tumors and include fibroadenomas and phyllodes tumors. Preoperative distinction between fibroadenomas and phyllodes tumors is pivotal to clinical management. Fibroadenomas are clinically benign while phyllodes tumors are more unpredictable in biological behavior, with potential for recurrence. Differentiating the tumors may be challenging when they have overlapping clinical and histological features especially on core biopsies. Current molecular and immunohistochemical techniques have a limited role in the diagnosis of breast fibroepithelial lesions. We aimed to develop a practical molecular test to aid in distinguishing fibroadenomas from phyllodes tumors in the pre-operative setting. METHODS: We profiled the transcriptome of a training set of 48 formalin-fixed, paraffin-embedded fibroadenomas and phyllodes tumors and further designed 43 quantitative polymerase chain reaction (qPCR) assays to verify differentially expressed genes. Using machine learning to build predictive regression models, we selected a five-gene transcript set (ABCA8, APOD, CCL19, FN1, and PRAME) to discriminate between fibroadenomas and phyllodes tumors. We validated our assay in an independent cohort of 230 core biopsies obtained pre-operatively. RESULTS: Overall, the assay accurately classified 92.6 % of the samples (AUC = 0.948, 95 % CI 0.913-0.983, p = 2.51E-19), with a sensitivity of 82.9 % and specificity of 94.7 %. CONCLUSIONS: We provide a robust assay for classifying breast fibroepithelial lesions into fibroadenomas and phyllodes tumors, which could be a valuable tool in assisting pathologists in differential diagnosis of breast fibroepithelial lesions.