Krux
May 11, 2026
Multi-Agent AI Cuts Cancer Trial Matching From Two Hours to Thirty Minutes
Published: May 11, 2026 at 12:09 AM
Updated: May 11, 2026 at 12:09 AM
100-word summary
A new study of 3,804 cancer patients shows that a multi-agent AI system matched patients to clinical trials in 30 minutes versus 120 for manual review, while hitting an accuracy score of 0.82 compared to GPT-4's 0.47. The platform processed 157,367 clinical pages and screened nearly 24,000 patient-trial pairs over 12 months, with oncologists confirming more than 17,900 matches. The system combines specialized agents for data extraction, an oncology knowledge graph, and mandatory clinician oversight for every recommendation. A separate radiotherapy application showed physicians preferred the AI's tumor-volume outlines over traditional supervised models. The real advantage isn't speed alone but adaptability: when cancer treatment guidelines change, the system updates instantly...
What happened
A new study of 3,804 cancer patients shows that a multi-agent AI system matched patients to clinical trials in 30 minutes versus 120 for manual review, while hitting an accuracy score of 0.82 compared to GPT-4's 0.47. The platform processed 157,367 clinical pages and screened nearly 24,000 patient-trial pairs over 12 months, with oncologists confirming more than 17,900 matches. The system combines specialized agents for data extraction, an oncology knowledge graph, and mandatory clinician oversight for every recommendation. A separate radiotherapy application showed physicians preferred the AI's tumor-volume outlines over traditional supervised models.
Why it matters
The real advantage isn't speed alone but adaptability: when cancer treatment guidelines change, the system updates instantly without retraining.