A multilingual chatbot to help bilingual patients receive better emergency department triage assessments

In moments of acute pain, physical trauma and psychological distress, every minute spent in a hospital emergency department (ED) waiting room can seem like a lifetime. But what happens when the language barrier prevents triage staff from properly understanding patients’ medical conditions?

This is the problem facing hospitals in areas like south-western Sydney, Australia, where more than half of the local population—or 55% according to the South Western Sydney Local Health District—speaks a language other than English at home.

Dr. Padmanesan Narasimhan is a UNSW Sydney academic researcher and an emergency department clinician who focuses on integrating digital health into acute services. He knows first-hand the difficulties facing hospital admission staff and triage nurses when communication is thwarted by language and cultural barriers.

He is leading a team developing an AI system that functions as an interpreter between hospital staff and patients which also addresses the cultural nuances in symptom description that can further complicate communication.

“The first thing admission staff will want to do is allocate patients with a score from 1 to 5 that measures the acuity—or severity—of the patient’s illness and the level of care required to treat them, with a score of 1 meaning a doctor will prioritize to see them within 30 seconds,” says Dr. Narasimhan, who is senior lecturer in UNSW’s School of Population Health.

“If there’s a language barrier and triage staff have difficulty understanding the person presenting to ED, it can lead to people with really severe or urgent medical conditions being assigned a lower acuity score and potentially being made to wait, whereas people with a non-urgent condition can be misclassified as urgent and bumped up to see a doctor straight away.”

For example, says Dr. Narasimhan, imagine an Arabic-speaking patient coming into ED with acute abdominal pain.

“Beyond language barriers, cultural norms around stoicism might lead them to downplay discomfort, describing severe pain as mere ‘tiredness.’ A triage nurse, unaware of these nuances, could misinterpret this, assigning a lower priority to a potentially critical condition like appendicitis.”

This highlights how the interplay of language and cultural differences has the potential to dangerously delay diagnosis and treatment in emergency departments if the wrong interpretation is made.

While there are official interpretation services available 24/7 to NSW hospitals, Dr. Narasimhan says the short interaction times and unpredictable nature of medical emergencies make it difficult to engage professional interpreters in the space of time required.

“Unfortunately, in an emergency setting the interaction is only three to five minutes. It’s really hard for a triage nurse to call the interpreter services and then engage them for translation and interpretation.”

AI interpreter

Together with linguistics specialists, artificial intelligence engineers and emergency clinicians, Dr. Narasimhan is working on a project that aims to help triage staff overcome language and cultural differences to make more accurate assessments on the spot.

The team is developing an AI chatbot with multilingual capabilities, allowing it to process different languages through machine learning.

“The idea is that this chatbot will be listening in at the registration point on a computer in an ED and will be able to interpret a patient’s description of their symptoms in real time, allowing triage staff to more quickly and accurately assess the severity of a patient’s condition,” Dr. Narasimhan says.

“So if you speak Arabic, it will be able to interpret and translate your Arabic into English. And because it has natural language processing and machine learning capabilities, it will also be able to give an appropriate triage recommendation.

“And of course, a human would always be overseeing this process. If there is any discrepancy between the AI and the nurse’s triage recommendations, it will be referred immediately to the senior consultant in the emergency department.”

Getting the system to that level of sophistication will involve three phases of development over the next few years. The first phase involves training the AI system on datasets based on other languages and medical terminology typically used in hospital EDs. The next phase involves simulating triage in a controlled environment to put the chatbot through its paces.

The final phase will be to road-test the technology in real-life emergency departments in areas like Western Sydney where patients are more likely to come from multicultural backgrounds.

Dr. Narasimhan says while there is plenty of research aimed at optimizing triage workflows, to his knowledge there are no other studies examining the effect of multi-lingual communication in triage settings.

“While there are some commercial apps that aim to break down language barriers in hospital settings, we’re doing it for the public good,” he says.

“We’re trying to access patented algorithms to use in specific settings, such as an ED. If it works in acute settings, it should be easy to adapt for other hospital settings, and even non-hospital settings, such as your GP’s office.

“We think it has multiple uses, and hope it removes one of the major barriers that can get in the way of multilingual people accessing health care services in Australia.”