Can a computer program help identify stroke patients faster in the Emergency Department? A study of automated triage note analysis.

A pre-trained natural language algorithm will be applied retrospectively to de-identified Emergency Department triage notes from Monash Medical Centre. The algorithm analyses free-text triage notes for stroke-related indicators and produces a binary classification output (activate Code Stroke pathway: yes/no) with a generated justification. The algorithm will observe recorded patient symptoms and observations such as blood pressure and GCS then determine likelihood of fulfilling code stroke pathway. Stateless processesing is used - the model receives only the current patient's data and produces an output without any information being carried over from previous patients. Therefore each patients data will be processed by the model individually. The algorithm is locally deployed on secure departmental infrastructure, requires no internet access, and performs no model training on patient data. The duration of (retrospective) observation covers ED presentations over a 1 month period between 14/9/2023 and 2/10/2023. Data collection will occur in April 2026 and be completed within one month. Given the observational design results will not alter clinical outcomes and the analysis is retrospective.