Jakarta, INTI - The use of artificial intelligence (AI) is becoming increasingly widespread across various sectors, including healthcare. One such implementation is in the triage process in emergency departments (ED), which determines patient care priorities based on the severity of their condition.
Nuraisa, a researcher at the Research Center for Data and Information Science (PRSDI) of the National Research and Innovation Agency (BRIN), developed a Large Language Model (LLM)-based Triage Decision Support System to assist healthcare professionals in the ED.
According to Nuraisa, triage is a crucial initial stage in emergency services as it dictates the priority of patient care. Therefore, technology is needed to support healthcare professionals in ensuring the assessment process is consistent and efficient.
"The system we developed serves as a tool to assist healthcare professionals in conducting triage screening based on the Australasian Triage Scale (ATS) standards. The recommendations generated are not final decisions. They must still be validated by a doctor or healthcare professional," she said on Wednesday, July 29, as written in an official statement.
The application was built using the Qwen3-32B model, which underwent fine-tuning with a dataset of triage cases. The system utilizes patient clinical information, such as health complaints, medical history, vital signs, and initial examination results, to generate triage category recommendations in accordance with ATS standards.
Guardrail to Minimize Undertriage
To enhance patient safety, the research also incorporates a guardrail mechanism aimed at minimizing the risk of undertriage, which is a situation where a patient's severity level is assessed lower than their actual condition. This mechanism acts as an additional layer of safety, ensuring that the system's recommendations prioritize caution in clinical decision-making.
Nuraisa added that system development is ongoing, involving the expansion and curation of datasets, clinical validation with doctors and healthcare professionals, refinement of guardrail mechanisms, and the strengthening of data security and system reliability before wider implementation.
"Collaboration with healthcare professionals is vital to ensure that the developed system is not only technically capable but also aligned with service needs and clinical practices in the field," she said.
Conclusion
Researcher from BRIN developed a Large Language Model (LLM)-based triage decision support system to assist healthcare professionals in prioritizing patient care in emergency departments, in accordance with the Australasian Triage Scale (ATS) standards. Utilizing the Qwen3-32B model, the system leverages patient clinical data to provide triage recommendations, which serve as a reference requiring doctor validation.
Read more: TBScreen.AI can Analyze Chest X-Ray Images to Detect Tuberculosis in Less than Five Seconds