Clinical trial matching, identifying suitable trials for patients, is labor-intensive. This study presents PRISM, an end-to-end system using Large Language Models (LLMs) and real-world Electronic Health Records (EHRs) to automate this process. A custom-tuned model, OncoLLM, outperforms GPT-3.5 and achieves performance comparable to medical doctors in clinical trial matching. The system is scalable, handles long-context scenarios, and addresses privacy concerns by using a smaller, fine-tuned model deployable on private infrastructure. Experiments demonstrate OncoLLM's superior ranking ability and cost-effectiveness compared to GPT-4.

Shashi Gupta, Aditya Basu, Mauro Nievas, Jerrin Thomas, Nathan Wolfrath, Adhitya Ramamurthi, Bradley Taylor, Anai N. Kothari, Regina Schwind, Therica M. Miller, Sorena Nadaf-Rahrov, Yanshan Wang, Hrituraj Singh