SENTRYLINE answers evolving cancer care questions with updated guideline checks
SentryLine: Evidence-Grounded Question Answering over Evolving Documents in Oncology Care
Computation and Language
Summary
Cancer care guidelines change frequently as new medical evidence emerges, which makes it hard for doctors to stay current. The authors created SENTRYLINE, a system that answers clinical questions by searching multiple versions of these guidelines and highlights when information has changed. It provides answers tailored to different roles in healthcare, shows exact citations, and verifies if the facts and timing are correct. The system was tested on a set of cancer-related questions and performed better than other methods, especially for complicated questions that need combining multiple pieces of information.
What this means in practice
- •For clinical decision support teams: Provide up-to-date, role-tailored answers from evolving oncology guidelines to support clinicians' treatment decisions.
- •For medical guideline publishers: Integrate the system to alert users about changes and updates across multiple versions of living guidelines.
Authors
Tampu Ravi Kumar, Gaurav Najpande, Muhammad Ali Khan, Kaneez Zahra Rubab Khakwani, Karan Kathuria, Shorya Azriel Moses, Yuvraj Kalia, M Bassam Sonbol, Irbaz Bin Riaz, Vivek Gupta
Abstract
Oncology care operates at constant pressure of absorbing rapidly evolving evidence base in biomedicine. The American Society of Clinical Oncology (ASCO) addresses this through living guidelines, but the format introduces a new burden: any recommendation can change at any point, across multiple versioned documents. We present SENTRYLINE, a living guideline-aware clinical question answering system. SENTRYLINE retrieves guideline passages through a vectorless hierarchical RAG pipeline and returns a role-specific answer with inline citations, factual and temporal verification reports, and drift detection notes that surface when a guideline has been updated. We construct ASCOBENCH, a benchmark of 405 three-turn conversations across four question categories with gold answers from expert annotators(clinicians), and use test set to evaluate SENTRYLINE against five baselines under an LLM-as-judge framework. Experiments across three generation backbones show consistent improvements over four retrieval baselines and ASCO's guideline assistant, with particularly strong gains on Reasoning and Role-Specific questions where multi-hop synthesis and register adaptation are required