How AI shapes emergency visits | MEDICUS 2024 Issue 2

Artificial Intelligence: How is it shaping your A&E visit?
By Alice Chia, Senior media and content specialist

Assoc Prof Liu Nan (right) with third-year medical student Yvonne Wong, who seeks to bring AI to the Emergency Department // Credit: Norfaezah Abdullah, Duke-NUS

Can Artificial Intelligence (AI) assess how quickly patients should be seen by doctors and predict their mortality risk when they arrive at the Emergency Department?

A team led by researchers from Duke-NUS Medical School has developed a machine learning model to do just that. It can flag patients at risk of death, as well as those who may be safe for discharge, helping healthcare professionals to prioritise patients more accurately according to the severity of their conditions.

Currently, assessing and sorting patients who arrive at the Emergency Department is done by a triage nurse, who will ask about the patient’s condition and assess their vital signs before assigning a score from one to four from the Patient Acuity Category (PAC) scale.

PAC 1 is the most severe, where patients require resuscitation, while PAC 4 is the least severe and often a non-emergency.

“But it can be difficult to gauge the risk of death accurately based on the triage nurse’s initial assessment, which could also vary depending on the nurse’s experience,” said Yvonne Wong, lead author of the project and a third-year medical student at Duke-NUS Medical School (Duke-NUS).

“Based on our observations, most patients are categorised under PAC 2, which means that they appear to be in a stable state but still require early attention due to the severity of their symptoms. This can lead to longer wait times at the Emergency Department,” Wong added.

“Derived using a machine learning framework, SERP showed better prediction performance than existing triage scores while maintaining easy implementation and ease of ascertainment at the Emergency Department. It has the potential to be widely applied in different circumstances and healthcare settings,” said Liu, who is also from Duke-NUS’ Centre for Quantitative Medicine. In addition, he is the Director of the Duke-NUS AI + Medical Sciences Initiative, which connects AI and data science specialists with clinicians to address healthcare problems.

Overcrowding in Emergency Departments is a significant challenge in Singapore and around the world, contributing to prolonged wait times and high hospital bed occupancy. To tackle this challenge, an effective triage system is crucial.

Photo of Assoc Prof Kenneth Tan

Assoc Prof Kenneth Tan feels AI can be used to augment the work of clinicians // Credit: Singapore General Hospital

In addressing the increasing volume of patients and the limited resources available, Wong and the team developed a tool called PAC+. Using data from more than 190,000 Singapore General Hospital ED patients from 2018 to 2019, the model can more accurately triage patients based on information such as their vital signs and medical history. This is done via integrating both conventional clinical triage and an AI-generated score.

“Derived using a machine learning framework, SERP showed better prediction performance than existing triage scores while maintaining easy implementation and ease of ascertainment at the Emergency Department. It has the potential to be widely applied in different circumstances and healthcare settings,” said Liu, who is also from Duke-NUS’ Centre for Quantitative Medicine. In addition, he is the Director of the Duke-NUS AI + Medical Sciences Initiative, which connects AI and data science specialists with clinicians to address healthcare problems.

“The development of such a tool is timely given that our population is getting older and their healthcare needs are more complex. The use of PAC+ as a more objective triage tool... will help us make better judgements and have better use of our resources. But of course, this must augment the work of our clinicians. It cannot take over our clinicians at all.”
Assoc Prof Kenneth Tan

Overcrowding in Emergency Departments is a significant challenge in Singapore and around the world, contributing to prolonged wait times and high hospital bed occupancy. To tackle this challenge, an effective triage system is crucial.

In addressing the increasing volume of patients and the limited resources available, Wong and the team developed a tool called PAC+.

Using data from more than 190,000 Singapore General Hospital ED patients from 2018 to 2019, the model can more accurately triage patients based on information such as their vital signs and medical history. This is done via integrating both conventional clinical triage and an AI-generated score.

This Score for Emergency Risk Prediction (SERP) was created by Professor Marcus Ong and Associate Professor Liu Nan from the Health Services and Systems Research Programme at Duke-NUS.

It makes use of data such as a patient’s age, blood pressure readings and heart rate—information that is easily obtainable in the emergency department’s care process—to generate a reading on a scale of 0 to 100, that provides a simple yet accurate estimate of a patients’ risk of death.

In addition to SERP, Ong and Liu also developed FedScore, a framework that can be used to create accurate scoring systems across multiple institutions without the need for data sharing. This means that data from various sites does not have to be pooled in a central location or server, allowing cross-institution collaboration without sacrificing data privacy and risking the breach of personal information.

Both SERP and FedScore were licensed to Anhui Hansin Intelligent Care Technology, a Duke-NUS spin-off, in February 2024 for commercialisation. These are just some of the projects in the School’s pipeline, as it seeks to cement its position as a leader in medical innovation.

The PAC+ project is also gaining momentum. A team from the Singapore General Hospital’s emergency department is looking to conduct a clinical trial for the tool at their hospital.

Associate Professor Kenneth Tan, Head and Senior Consultant, Department of Emergency Medicine, Singapore General Hospital, shared: “The development of such a tool is timely given that our population is getting older and their healthcare needs are more complex. The use of PAC+ as a more objective triage tool, on top of the clinical experience of our nurses and doctors, will help us make better judgements and have better use of our resources. But of course, this must augment the work of our clinicians. It cannot take over our clinicians at all.”

question-mark

Have a question? Send it in and it may be answered in the next issue of MEDICUS!

ASK MEDICUS

“Patients may generally look well, but if their risk of death is high based on the AI model, then the clinician may reassess – should they be assigned a higher priority level and be seen within a shorter period of time,” Wong added.

Wong’s study offered an encouraging picture of how this approach could improve resource management and patient care in the emergency department. Her study, conducted as part of her third-year training, also showed that PAC+ performed better than PACS alone in identifying critically ill patients at high risk of death, and could safely identifying low risk patients who may be suitable for discharge and potentially be referred to primary or urgent care settings.

“In a retrospective study conducted using PAC+, almost half the number of patients could be safely down-triaged from higher (P1-P2) to lower priority levels (P3-P4). This could help to ease the emergency department’s caseload, allowing more resources to be diverted to looking after critically ill patients,” explained Wong.

The PAC+ team hopes that their model can help identify patients who may be suitable for discharge, referred to primary care, or admitted under Mobile Inpatient Care @ Home (MIC@Home), a programme that provides hospital-level care to patients in their homes.

Associate Professor Christopher Laing, Vice-Dean for Innovation and Entrepreneurship at Duke-NUS, said: “Artificial intelligence is the future of medicine, with immense potential to solve complex clinical problems. Our scientists and students are working with industry partners and entrepreneurs to bring new AI technologies to medical practice, where they can make a meaningful impact to patients.”

Duke-NUS commercialisation impact has included therapeutics, vaccines, diagnostics, medical devices, and digital health technologies, developed to combat our society’s greatest emerging health challenges.