A multi-agent artificial intelligence system designed to mimic the workflow of a multidisciplinary tumor board may improve the interpretation of complex oncology guidelines, according to a study published in JCO Clinical Cancer Informatics.
Researchers developed an AI framework in which several specialised “agents” work together to answer clinical questions based on oncology guidelines. The system was structured to reflect the collaborative decision-making process used in tumour boards, with different agents responsible for identifying relevant guidelines, extracting key information, and synthesising final responses.
CLINICAL SUMMARY
What was examined
Researchers developed a multi-agent artificial intelligence system designed to interpret oncology guidelines using a workflow modelled on multidisciplinary tumor boards.
Key findings
- The AI framework used specialised agents to identify relevant guidelines, extract information, and generate answers.
- In testing with 100 oncology guideline questions, the system achieved 94% accuracyin selecting the correct guideline and 90% accuracy in answering questions.
- Performance exceeded that of several standalone large language models.
Clinical implications
- AI systems may help clinicians navigate increasingly complex oncology guidelines and support clinical decision-making.
- Multi-agent architectures could improve the reliability of clinical decision-support tools.
- Further validation will be needed before clinical implementation.
The framework included a coordinator agent that selected the most relevant guideline, several tumour board agents that extracted information from guideline text, tables, and figures, and a review agent that generated the final answer.
The investigators evaluated the system using 100 guideline-based oncology questions in a benchmark testing dataset. The multi-agent system achieved 94% accuracy in selecting the correct guideline and 90% accuracy in answering clinical questions, outperforming several standalone large language models.
Featured Event
Additional experiments showed that removing specific agents reduced performance, suggesting that the collaborative multi-agent architecture contributed to the system’s accuracy.
The authors note that such systems may help clinicians navigate increasingly complex oncology guidelines. However, they emphasise that further validation will be required before these tools can be used in clinical decision-support settings.
Paper: Jiasheng Wang et al. Tumor Board–Inspired Multiagent Artificial Intelligence System for Interpreting Oncology Guidelines. JCO Clin Cancer Inform 10, e2500286(2026).DOI:10.1200/CCI-25-00286 Access online here.
