Clinical Trial Matching
Clinical trial matching has traditionally been a very manual and time-consuming process. Preliminary use of AI in clinical trial matching was accomplished through Trial GPT, a proof-of-concept system developed by researchers at NIH. Trial GPT boasted an 87.3% accuracy in matching while reducing screening time by 42.6%1. While these results are encouraging, the last year has seen significant advancements in the reasoning abilities of LLMs which can be leveraged to increase accuracy and provide more nuanced explanations.
To this end, CAAI hosted a hackathon in February 2025, bringing together developers and researchers to develop a proof-of-concept for a more advanced clinical trial matching model that uses initial filtering based on minimum criteria such as age, sex, and location to scan trials for relevancy, then compares FHIR bundles to patient information. These results are run through an LLM that incorporates better reasoning for explaining match eligibility which assists with increasing patient understanding and reducing misconceptions, adhering to legal requirements for transparency, and validating accurate matching.
This effort is ongoing and future-state expectations is a tool that matches patients on par with human experts, surpassing Trial GPT benchmarks, while offering more detailed, rational explanations. Collaboration with the Markey Cancer Center will allow pilot testing.
Jin, Q., Wang, Z., Floudas, C.S. et al. Matching patients to clinical trials with large language models. Nat Commun 15, 9074 (2024). https://doi.org/10.1038/s41467-024-53081-z
From <https://www.nature.com/articles/s41467-024-53081-z#citeas>
Photo courtesy of NIH/NLM https://www.ncbi.nlm.nih.gov/research/trialgpt/
