
AI Predicting Clinical Trial Outcomes
GPT-4o shows promise but struggles with negative outcome identification
This research evaluates how accurately Large Language Models can predict clinical trial outcomes, comparing performance across multiple metrics.
- GPT-4o achieved the best overall performance among tested LLMs
- All LLMs showed limitations in identifying negative outcomes
- The specialized HINT model demonstrated superior ability in negative sample recognition
- Performance varied across different clinical endpoints and trial characteristics
Why it matters: Accurate prediction of clinical trial outcomes could dramatically reduce pharmaceutical R&D costs, accelerate drug development timelines, and improve resource allocation in medical research.
Can artificial intelligence predict clinical trial outcomes?