Matching patients to clinical trials with large language models
Problem
Patient recruitment for clinical trials is challenging; matching patients to eligible trials manually does not scale.
Why it matters
Poor recruitment delays trials and prevents patients from accessing potentially beneficial experimental treatments.
What the AI did
TrialGPT uses large language models in three modules to retrieve candidate trials, predict criterion-level eligibility, and rank trials, all zero-shot.
Results
Recall over 90% of relevant trials using <6% of collection; 87.3% accuracy on 1,015 patient-criterion pairs; outperforms best competing models by 43.8% in ranking
Impact
Benchmarked on 183 synthetic patients with over 75,000 trial annotations, with performance close to expert level.
Record
| Venue | Nature Communications |
|---|---|
| Published | 2024 |
| Work type | article |
| Authors | Qiao Jin, Zifeng Wang, Charalampos S. Floudas, Fangyuan Chen, Changlin Gong, Dara Mark Bracken-Clarke, Elisabetta Xue, Yifan Yang et al. |
| Citations | 283 (OpenAlex, at import time) |
| References | 47 |
| Model or method | TrialGPT (LLM-based framework) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41467-024-53081-z |
| Open access | gold |
| Keywords | clinical trials, patient matching, large language models, TrialGPT, eligibility prediction |
Importance rating
significant — thousands to millions affected
Clinical trial recruitment is a broad healthcare bottleneck; abstract shows 42.6% screening time reduction and near-expert matching, implying significant benefit to trials and patients.