LLM-augmented ML can extract clinical signals from unstructured text (triage notes) for public health surveillance, achieving high accuracy while remaining transferable across hospitals—showing practical value for real-world healthcare deployment.
Researchers built a machine learning system to detect self-harm cases from emergency department triage notes by combining traditional ML with large language models. The system identifies self-harm incidents and the specific method used, and works reliably across different hospitals without needing retraining for each location.