Authors (including presenting author) :
Mak KH(1), Lai SO(1), Lui CT(1), Leung WY(1), Chan YW (1), Lee KY(1), Tang H(1)
Chan CW(2)
Chan MC(3)
Affiliation :
(1) Accident & Emergency Department, NTWC
(2) Information and Technology Section
(3) Hong Kong Applied Science And Technology Research Institute
Introduction :
Triage is the key-process in emergency department for initial risk stratification of patients, to ensure appropriate resources allocation and timely care to patients in-need. The current emergency triage are based on structural training and guideline from COC A&E. Currently, there is lack of decision support tool integrated with clinical systems and workflow.
Objectives :
To develop an innovation using ASR voice recognition technology and AI to support emergency triage
Methodology :
TSWH collaborated with the Hong Kong Applied Science And Technology Research Institute in iSTriage project, which was funded by the Innovation and Technology Commission.
The ASR-engine was multilingual compatible with Cantonese and English, which was based on a common language model enhanced with 50-hours specific medical-domain specific contents. The engine was linked to a large-language-model to translate Chinese to English for documentation. On hitting corresponding keywords or context, structural decision-tree algorithm, which was based on the COC A&E Triage guideline, would be triggered to guide through the triage process.
iSTriage was piloted in TSWH A&E since October 2024 with evaluation. Convenient sampling of 100 AED patients was evaluated. 18 nurses had participated in the trial including senior (61%) and junior (39%) nurses.
Result & Outcome :
For standardization and quality of documentation, 86% cases could hit the iSTriage to trigger the follow up questions of the decision tree. Only 36% could commit a triage category by iSTriage based on COC A&E guideline. The accuracy of iStriage was 97% with cases of overtriage. The ASR demonstrated around 70% accuracy which is inadequate to impact on the operation of triage.
Time-motion analysis on iSTriage versus traditional manual documentation did not show up benefit on efficiency, with limitation in the coverage of decision-tree support as well as lack of eAED system integration which requires manual transcription.
Staff survey demonstrated 78% of participants agreed that iSTriage and decision-support tool can improve patient safety and most of the staff agree that the triggered question can guide them in triage process. However, there was feedback on limited applicability in operation in ED triage with lack of eAED integration as well as coverage of decision-tree.
Conclusion
An intelligence smart triage with decision-support is needed to envisage standardized and efficient ED triage. The iSTriage trial provide valuable insight on the key success factor of adequate decision-tree coverage, need of strengthening of COC triage guideline, as well as the importance of eAED system integration to facilitate clinical operation in triage.
