Matching Patients to Clinical Trials with Large Language Models.

Area of Science:

• Clinical Informatics
• Artificial Intelligence in Medicine
• Biomedical Data Science

Background:

• Patient recruitment is a major bottleneck in clinical trial success.
• Efficiently matching patients to suitable clinical trials remains a significant challenge.

Purpose of the Study:

• To introduce TrialGPT, an end-to-end framework for zero-shot patient-to-trial matching using large language models.
• To evaluate the efficacy of TrialGPT in improving patient recruitment for clinical trials.

Main Methods:

• Developed TrialGPT with three modules: TrialGPT-Retrieval for large-scale filtering, TrialGPT-Matching for criterion-level eligibility prediction, and TrialGPT-Ranking for trial-level scoring.
• Evaluated on synthetic patient cohorts and real-world data, comparing against expert performance and existing models.

Main Results:

• TrialGPT-Retrieval recalled over 90% of relevant trials using <6% of the dataset.
• TrialGPT-Matching achieved 87.3% accuracy with faithful explanations.
• TrialGPT-Ranking outperformed competing models by 43.8% and reduced screening time by 42.6% in user studies.

Conclusions:

• TrialGPT demonstrates a powerful and efficient approach to patient-to-trial matching.
• The framework offers significant potential to accelerate clinical trial recruitment and improve patient access to trials.