Comprehensive decision support tool (DST) in breast cancer: Automated National Comprehensive Cancer Network (NCCN) guidelines linked to living systematic literature review (SLR).

Abstract

e13676 Background: An SLR of interventional clinical trials projects > 7,500 publications in 2024, alongside > 6 NCCN updates and numerous FDA approvals. For oncologists with demanding schedules, staying updated is a challenge. A clinical data system integrating automated NCCN treatment paradigm with a real-time SLR, FDA approvals, and ongoing trials can support evidence-based decisions. This study leveraged AI and automation to create a comprehensive DST for breast cancer (BC). Methods: A Cochrane-compliant SLR protocol was automated for daily updates. A Large Language Model trained on > 20,000 annotated clinical trial records (across six cancer types, including BC) extracted 32 variables per publication (e.g., TNM staging, histology, biomarker, risk factors, treatment line, interventions, interventions; randomization, study phase and sample size; analysis type, follow-up period; reported outcomes (median and landmark overall survival (OS), progression-free survival (PFS) and other progression measures, response data, quality of life); subgroup analyses, safety/toxicity). Extracted data were standardized and integrated with established guidelines treatment paradigm. A software tool was built to filter this database, linking treatment pathways to real-time SLR, FDA approvals, and ongoing clinical trials. Results: The real-time SLR identified 317 interventional BC trial publications in 2024. In a test-case query for treatments in HER2-negative metastatic BC patients progressing after CDK4/6 inhibitors, 89 relevant publications were identified in < 1 minute. Of these, 24 focused on post-CDK 4/6 progression with 10 reporting on 8 Phase 2 or 3 RCTs of targeted therapies, antibody-drug conjugates, CDK4/6 inhibitors, and chemotherapy (3, 2, 2, and 1 studies, respectively). Five studies showed significant PFS improvement vs. comparators, of which four were included in NCCN guidelines. Traditional SLR would require 3 weeks and multiple researchers to achieve similar results. Conclusions: By selecting key patient criteria (e.g., biomarkers, metastatic stage, treatment path), our Decision Support System links real-time clinical data to established guidelines. Implementation in clinical practice could enable oncologists to make timely, informed decisions, potentially improving patient outcomes.