Home › Operations & systems › AI predicts ED admitting team from triage notes

Operations & systems

AI predicts ED admitting team from triage notes

Using Sydney Triage to Admission Risk Tool With Artificial Intelligence (START-AI) to Predict Inpatient Admitting Specialty for Emergency Department Patients: Single Centre Deep Learning Analysis
Emergency Medicine Australasia · Added 8 September 2026

Sydney single-centre deep learning study of 40,054 admitted ED patients — a fine-tuned clinical language model predicted the inpatient admitting specialty from triage notes with AUROC 0.97, though precision was a moderate 0.71.

Study type

Modelling

Design

Single-centre deep learning analysis of electronic medical record data

Participants

40,054 admitted adult ED patients in Sydney

Main finding

Bio-ClinicalBERT predicted one of 20 inpatient admitting teams with weighted AUROC 0.97 (95% CI 0.96-0.98), precision 0.71, accuracy 0.75 (0.88 when top two teams accepted)

Caveat

Single centre; moderate precision and only ~75% single-label accuracy; needs external validation

Bottom line

Hypothesis-generating

Learning goal

Understand how artificial intelligence triage tools may support streaming decisions in Australian emergency care.

More

• Topic: Operations & systems
• Daily scan: 8 September 2026

Related papers on Operations & systems

• 2026 review of prehospital guideline quality

  Prehospital Emergency Care · 12 Sep 2026 · Systematic review · Context only

  The 2026 update of the systematic review appraising evidence-based guidelines for prehospital care — abstract not yet available (ahead-of-print), but this recurring review is the reference map of what guideline-level evidence exists in prehospital practice.

• ED volume not linked to mortality

  Scandinavian Journal of Trauma, Resuscitation and Emergency Medicine · 8 Sep 2026 · Cohort · Context only

  Cohort of 724,256 ED presentations across Dutch and Danish hospitals — annual ED volume had no association with in-hospital mortality, while mortality was lower on busier days, driven by excess deaths on quiet ones.

• Ramping driven by hospital 'back door'

  Emergency Medicine Australasia · 8 Sep 2026 · Other · Hypothesis-generating

  A mixed-methods Queensland study of six years of ambulance and hospital records found access block is largely a hospital discharge ('back door') problem, with recommendations spanning discharge metrics, primary care support and prehospital flow.

• ED ultrasound archiving works in Australia

  Emergency Medicine Australasia · 3 Sep 2026 · Cohort · Context only

  Four-year implementation study of a middleware ED ultrasound archiving and QA workflow at Nepean Hospital — 7,281 studies archived across 15 modalities with zero data loss or failures, automated trainee logbooks saving each trainee ~2 h/week, and clinician time (not software) the dominant recurring cost.