Clinical Note Triage with Privacy-Preserving Natural Language Processing

Abstract

Emergency departments generate unstructured clinical notes faster than staff can review them. We describe a privacy-preserving NLP system that runs entirely on-premise to prioritise notes by acuity, using de-identification and federated fine-tuning to keep patient data local. Deployed across two hospital networks, the system surfaced high-acuity cases a median of 22 minutes earlier while meeting institutional data-governance requirements, illustrating responsible applied AI in healthcare.

Keywords