Transferable Self-Harm Surveillance from Emergency Department Triage Notes Using an Evidence-Augmented Machine Learning Approach
Two new research papers explore the challenges and potential solutions for using NLP models to predict self-harm from emergency department triage notes. The first paper identifies lexical and semantic variations across different hospitals as a key reason for model generalizability issues. The second paper proposes an evidence-augmented machine learning approach, combining traditional methods with LLM-based screening, to improve model transferability across institutions and accurately identify self-harm methods. AI
IMPACT These studies highlight the need for more robust and transferable AI models in healthcare, particularly for sensitive applications like self-harm prediction, which could improve patient safety and resource allocation.