Truace
Subscribe

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

Objective To investigate whether the Sydney Triage to Admission Risk Tool with Artificial Intelligence (START-AI) model could be used to predict specific inpatient admitting teams based on clinical notes. Methods This was a pre-trained language model analysis using electronic medical record data from an inner-city tertiary referral hospital emergency department (ED) in Sydney, Australia. All adult patients who were admitted to an inpatient ward between January 2023 and June 2025 were included. A fine-tuned Bidir…

By the Truvace desk1 min read1 sourcePeer-reviewedTRV-2026-1564· cite & verify
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
Hospital Universitari Doctor Peset, València 04 by 19Tarrestnom65. CC BY-SA 4.0 · https://creativecommons.org/licenses/by-sa/4.0

In brief

Researchers applied a Bio-ClinicalBERT model, previously used in the START-AI tool, to emergency department case history notes from an inner-city tertiary hospital in Sydney. They trained it to assign each admitted adult patient to one of twenty specialist inpatient teams, using 40,054 admissions between January 2023 and June 2025.

The approach could eventually help streamline the handoff from emergency care to inpatient wards, but the evidence so far is retrospective and from one site. The authors say further exploration and validation are needed before clinical usefulness and wider feasibility are known.

Main points

  1. The analysis covered 40,054 adult admissions from a single Sydney emergency department.
  2. Weighted average AUROC was 0.97, ranging from 0.91 for infectious diseases to 0.99 for psychiatry and gynaecology.
  3. Exact-match accuracy was 0.75, while weighted average precision was a moderate 0.71.
  4. The correct team was among the model's top two predictions 88 percent of the time.

The gain

A fine-tuned clinical language model predicted the correct admitting specialty team from ED case notes with high discrimination in a single Sydney hospital.

The rundown

The study used electronic medical record data and a fine-tuned Bio-ClinicalBERT language model trained on ED medical case history notes. The model chose among twenty specialist inpatient admitting teams for adults admitted to a ward.

Discrimination was strong across classes, with the lowest AUROC for infectious diseases and the highest for psychiatry and gynaecology. Precision was only moderate and about three quarters of cases were classified correctly, so errors remained common.

The authors conclude the model could accurately predict the admitting team in this single centre setting, and they call for further exploration and validation to assess clinical utility.

Sources

  1. Peer-reviewedEmergency Medicine Australasia2026-10-01

The debate