TrialGPT’s strongest result is that doctors still reviewed the match
The useful pattern is not replacing clinical judgment. TrialGPT 2.0 recommends cancer trials that deserve a closer look. It also gives structured explanations for expert review. In the public abstract, the gains come from narrowing the search while keeping clinicians responsible for the decision.
What the source reports
The arXiv abstract presents TrialGPT 2.0 as an AI-assisted system for matching cancer patients to clinical trials. It fits oncology job steps. It does more than screen for eligibility. It judges which trials deserve more review based on a patient’s needs and local priorities. It gives structured explanations that experts can inspect. The system was tested in retrospective and prospective settings across government, academic cancer-center, patient-advocacy, and NIH referral settings. In retrospective multicenter cohorts of 288 cases, it retrieved at least one clinician-recommended trial in the top 10 for about 91% of cases. It cut clinician screening time by 55.0%. In a six-month prospective tumor-board evaluation, it found trial options missed by routine practice and expanded patient access by 90.9%. NIH-TrialBench adds 126 synthetic patient vignettes from 11 NIH Institutes and Centers.
Original source
- Title
- Towards AI-Assisted Clinical Trial Matching: TrialGPT 2.0 - Real-World Deployment
- Author
- Fang, Jin, Tian, He & Geer
- Publication
- arXiv
- Date
- Tuesday, September 1, 2026