Automated digital oncologic data review for breast cancer research: AI-enabled mCODE and field-of-interest extraction framework.

Abstract

e12647 Background: Enhancing outcomes in cancer care and research can be optimally achieved by accelerating data extraction from unstructured electronic health resources. Understanding the association between treatment response markers and long-term outcomes is crucial for refining cancer therapeutic approaches and improving patient prognosis. In this study, we explore the association of oncologic outcomes with pathologic complete response (pCR) in high-risk early-stage breast cancer treated with neoadjuvant chemotherapy (NACT) using automated oncologic data extraction framework. Methods: Patients with primary stage I-III breast cancer who received NACT followed by mastectomy, with or without immediate breast reconstruction from January 2010 to August 2023 were retrospectively analyzed. We extracted pCR results and data on clinical and pathological stage, tumor subtype, grade, and treatment regimens and factors affecting postoperative complications from electronic medical records based on the AI enabled-mCODE (Minimal Common Oncology Data Elements) and fields of interest extraction framework. Employing mixed-effects Cox models, we explored the association between pCR and event-free survival (EFS), distant recurrence free Survival (DRFS) and overall survival (OS). Survival curves of patients were estimated using Kaplan-Meier method. Multivariate analysis was performed to evaluate the predictive factors for pCR. Results: Overall, 2725 patients were analyzed. The median age was 50.6 years. The median follow-up was 67 months. pCR was achieved in 31.0%. Hazard ratio for pCR vs. non-pCR were 0.34 for EFS (95% CI, 0.25-0.45), 0.32 for DRFS (95% CI, 0.23-0.43), 0.21 for OS (95% CI, 0.13-0.34), varying from 0.13 to 0.56 for EFS, 0.12 to 0.48 for DRFS and 0.07 to 0.40 for OS. The presence of pCR remained prognostic for event-free survival in multivariate models adjusted for age, tumor size, and nodal status at baseline. Conclusions: This study confirms that patients achieving pCR after NACT have better long-term survival outcomes that patients who do not. Oncologic data extraction from unstructured electronic medical records using AI-enabled extraction framework can be expected to enhance robust cancer research and generate valuable insights from real-world practice settings for tailored therapeutic decision-making. [Table: see text]