Advancing Clinical Trial Outcome Prediction with Domain-Specific Large Language Models
The results suggest that pretrained biomedical representations already encode signal relevant to trial outcome, and that probing is a computationally cheaper alternative to fine-tuning for this task.