Utilizing Natural Language Processing (NLP) to identify breast cancer associated-lung metastases from pathology reports to delineate characteristics and challenges of this common site of breast cancer recurrence.

Abstract

e13592 Background: NLP (artificial intelligence) can automate the identification of records in large datasets. The purpose of this study was to evaluate the feasibility of NLP to identify breast cancer-associated lung metastases to understand clinical characteristics and challenges posed by this common site of breast cancer recurrence. Methods: Patients with pathologically confirmed breast cancer associated-lung metastases seen at a large academic center between 3/2012-5/2019 were identified using NLP of institutional pathology reports, with an IRB approved protocol. Chart review was performed to confirm breast cancer associated-lung metastases and determine clinical and pathological features. Results: Using NLP, 32 patients with pathology reports denoting breast cancer associated-lung metastases were identified, with pathologic confirmation of lung biopsy tissue in the majority of cases (24), and pleural fluid specimens (8) on the remainder. Ten of 32 (31%) were HR+/HER2-, 3/32 (9.3%) HER2+, and 19/32 (59%) TNBC. The majority were invasive ductal carcinoma (21/26) with the remainder invasive lobular carcinoma (2/26) or mixed histology (3/26). Median age at lung metastasis diagnosis was 62 years (range 31-88). The median time to development of lung metastasis following primary breast cancer was 5.6 years (range 0-24.8 years). Fifty six percent of lung metastases were detected on imaging and 44% by symptoms including dyspnea, cough, or pain. Tumor tissue genotyping results on the lung metastases were available for 8 patients showing PI3KCA (5), TP53 (3), SMARCA4 (2), ERBB2 (1), FGFR3 (1), ATM (1), CDK4 (1), MYC (1), and ESR1 (1). Treatment after diagnosis of lung metastases included hormone therapy (61%), chemotherapy (84%), lung irradiation (26%), and surgical resection of lung metastases (6%). Lung metastases were associated with considerable morbidity including pleural effusion (15%), dyspnea (6%), pneumothorax (3%), hemothorax (3%), and atelectasis (3%). Patients diagnosed with lung metastases had brain (32%), bone (35%), renal (6%), skin (3%) and adrenal (3%) metastases during disease course. Conclusions: NLP can help identify organ specific metastases from pathology reports, such as breast cancer associated-lung metastases, which can then facilitate observational, translational, and clinical research to characterize and address challenges posed by this common site of breast cancer recurrence. This cohort of patients highlights the morbidity of breast cancer associated-lung metastases and potential role of NLP for disease characterization and clinical research. (Support from ASCO Medical Student Rotation for Underrepresented Populations Award.)