Emergency medicine Australasia : EMA · 2026
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.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Dinh Michael, Corbett Elizabeth, Ngo Thuy Truc, Salmon Eliot, Khan Saleem Ahmed, Pethani Farhana, Moore Nicholas, Koprinska Irena
Аннотация
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 Bidirectional Encoder Representation of Transformer (Bio-ClinicalBERT) used in the original START-AI analysis was trained on ED medical case history notes to predict one of twenty specialist inpatient admitting teams.
RESULTS: A total of 40,054 cases were analysed. The overall weighted average area under receiver operating characteristic curve (AUROC) was 0.97 (95% CI: 0.96, 0.98). For individual classes of inpatient admitting teams, AUROC ranged from 0.91 for infectious diseases to 0.99 for psychiatry and gynaecology. The overall weighted average precision (positive predictive value) was moderate at 0.71 (95% CI: 0.71, 0.72). The model correctly classified around three quarters of the dataset (accuracy 0.75 (95% CI: 0.74, 0.076)). Accuracy in correctly assigning one of the two most probable admitting teams by the model was 0.88 (95% CI: 0.88, 0.89).
CONCLUSIONS: A pre-trained language model used in START-AI was able to accurately predict the inpatient admitting team for ED patients requiring admission based on ED medical case history notes in a single centre study. Further exploration and validation are warranted to assess clinical utility and broader feasibility of the modelling approach.
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 Bidirectional Encoder Representation of Transformer (Bio-ClinicalBERT) used in the original START-AI analysis was trained on ED medical case history notes to predict one of twenty specialist inpatient admitting teams.
RESULTS: A total of 40,054 cases were analysed. The overall weighted average area under receiver operating characteristic curve (AUROC) was 0.97 (95% CI: 0.96, 0.98). For individual classes of inpatient admitting teams, AUROC ranged from 0.91 for infectious diseases to 0.99 for psychiatry and gynaecology. The overall weighted average precision (positive predictive value) was moderate at 0.71 (95% CI: 0.71, 0.72). The model correctly classified around three quarters of the dataset (accuracy 0.75 (95% CI: 0.74, 0.076)). Accuracy in correctly assigning one of the two most probable admitting teams by the model was 0.88 (95% CI: 0.88, 0.89).
CONCLUSIONS: A pre-trained language model used in START-AI was able to accurately predict the inpatient admitting team for ED patients requiring admission based on ED medical case history notes in a single centre study. Further exploration and validation are warranted to assess clinical utility and broader feasibility of the modelling approach.