Collaborative framework on responsible AI in LLM-driven CDSS for precision oncology leveraging real-world patient...

Area of Science:

• Oncology
• Biomedical Informatics
• Artificial Intelligence

Background:

• Precision oncology relies on real-world data for biomarker and therapy identification.
• Unstructured data presents a significant bottleneck in current precision oncology practices.
• Large language models (LLMs) offer potential solutions for data structuring challenges.

Purpose of the Study:

• To propose a framework for the responsible integration of LLMs into precision oncology.
• To address current data-related bottlenecks in precision oncology research and application.
• To guide multidisciplinary experts in leveraging LLMs for enhanced precision cancer care.

Main Methods:

• Development of a framework for responsible LLM integration, co-created by multidisciplinary experts.
• Establishment of five thematic dimensions and ten guiding principles for LLM implementation.
• Application of the framework to a thought experiment involving uterine carcinosarcoma data.

Main Results:

• The proposed framework provides a structured approach to utilizing LLMs in precision oncology.
• The framework addresses key challenges in handling unstructured real-world data.
• Illustrative application demonstrates the framework's utility in a specific cancer type.

Conclusions:

• Responsible LLM integration is crucial for advancing precision oncology.
• The framework facilitates the effective use of LLMs for biomarker discovery and therapy identification.
• This approach supports overcoming data bottlenecks and enhances cancer care through AI.