A context-augmented large language model for accurate precision oncology medicine recommendations

The rapid expansion of molecularly informed therapies in oncology, coupled with evolving regulatory food and drug administration (FDA) approvals, poses a challenge for oncologists seeking to integrate precision oncology medicine into patient care. Large language models (LLMs) have clinical potential, but their reliance on general knowledge limits their ability to provide up-to-date and niche treatment recommendations. Here, we developed a retrieval-augmented generation (RAG)-LLM workflow using the molecular oncology almanac (MOAlmanac) and benchmarked it against an LLM-only approach for biomarker-driven treatment recommendations. Our RAG-LLM achieved up to 95% accuracy on synthetic queries and 93% on real-world queries collected from practicing oncologists. Finally, our study explored several prompting and retrieval strategies to enhance performance. Taken together, this approach may serve as valuable guidance for deploying LLMs to support cancer patients' treatment decisions in precision oncology clinical settings.