How data helps doctors make life-saving decisions in emergency care
Experts in digital health at the Faculty of Information Technology (IT) at Monash University are working on advanced technological systems to aid clinicians in making these vital decisions, especially during COVID-19 and cardiac arrest scenarios.
Minutes make all the difference in an emergency. The right drug given early to interrupt dangerous inflammation, or a life saved due to timely ambulance dispatch.
But sometimes humans need help in making fast, accurate and effective decisions.
“Computer systems will never replace humans in healthcare,” says Chris Bain, Professor of Practice in Digital Health at Monash University.
“But what digital health tools can do is complement the skills of doctors, particularly when it comes to interpreting and acting on lots of complex data.”
A model of COVID-19 to guide treatment
One of the most difficult aspects of COVID-19 is that clinicians are still learning how people respond to this infection. Some patients manage OK, while others progress to serious illness and death – and doctors can’t always pick who ends up where.
Recently appointed Interim Dean in the Faculty of Information Technology at Monash University, Professor Ann Nicholson is developing a computer model to help.
“The point of modelling is that you capture something about the real world in the computer, which means you can then ask the system questions and get useful answers,” Professor Nicholson explains.
“It’s like a miniature, summarised version of reality.”
The end goal of this work is to create a system that allows clinicians to make use of what they know about a new COVID-19 patient – things like age, gender, pre-existing conditions like asthma or diabetes, and clinical features such as oxygen levels and blood pressure. The model applies this data plus other information to map out the most likely path forward for that particular person. Then clinicians are able to make informed and timely treatment decisions.
This type of approach is called Bayesian modelling, as it’s founded on Bayes theorem of statistical probability. It is particularly useful in hospital situations because it’s still reliable even if data is patchy, as is often the case with patient information.
“When there’s data, we use it,” explains Professor Nicholson. “When there’s limited data, we can still make the model work by including expert knowledge and information from published literature.”
Professor Nicholson and her colleagues are optimistic they will soon have a prototype COVID-19 decision support model ready for trials in hospital settings.
Work on the COVID-19 model started in March 2020, and is being developed with real patient data from overseas and Australia. It’s a project the Faculty of IT is conducting with colleagues at the Sydney University, Queensland University of Technology and other members of the national Digital Health Cooperative Research Centre.
When to send an ambulance
In addition to COVID-19 modelling, Professor Nicholson is working with Monash IT’s Associate Professor Burak Turhan to develop AI technology to improve response times to cardiac arrest. The project is supported by Ambulance Victoria and Safer Care Victoria.
Cardiac arrest results from an electrical malfunction that stops the heart from pumping blood around the body effectively.
“Seconds are the difference between life and death in these cases,” says Associate Professor Turhan.
“We’re creating a system to support an ambulance being sent as fast as possible to a cardiac arrest once a triple zero call comes in.”
Triple zero calls are typically intense conversations, with the caller often stressed and unsure about what they are witnessing. It can be difficult for the ESTA triple zero call taker to collect all of the key information, and sometimes this translates into delays in making the decision to send an ambulance.
AI can help speed things up.
Associate Professor Turhan says they’re using data from thousands of triple zero calls to train their system. For this training data set, the outcomes of all the patients are known – true cardiac arrest or not, and survival versus death.
“We’re using the outcome data, plus text data from the conversations between caller and call-taker, as well as time information, the order of events and even sounds from the scene of the emergency,” he says.
Patients in severe cardiac distress can experience something called agonal breathing, which creates a distinctive, gasping noise that can be audible through a phone.
The system learns which features of the phone call are reliable indicators of a cardiac arrest. Applying a Bayesian approach, the model can then be used as a predictor in real, live triple zero phone calls.
“Working in a live scenario, the system will use the data from a triple zero phone call in real time, and then generate a simple flag for the call taker if it looks like a cardiac arrest has taken place,” explains Professor Nicholson.
“It’s the human who will make the final decision about calling an ambulance, but this system can support them in making that decision accurately and early.”
Fast identification of a cardiac arrest also means CPR guidance can be provided over the phone to keep a person alive until an ambulance arrives.
Ambulance Victoria says up to 185 lives could be saved per year with such an AI system.
Helping patients feel informed
Associate Professor Turhan and his colleagues, Dr Chakkrit Tantithamthavorn and Dr Aldeida Aleti, are also working on an information system to improve the patient experience of healthcare. If you’ve ever sat in a hospital emergency room, you’ll likely be familiar with the frustration of not knowing when you’ll be seen.
“The idea is to create a readout that gives a live indication of how long patients can expect to wait to see a clinician,” he says.
Waiting time depends on staffing, on the medical condition of patients who are currently being cared for, on the sudden arrival of ambulances with urgent cases, and on seasonal factors such as flu or COVID-19.
“Our system will also include data such as when patients have arrived, how they arrived, demographic information like gender and age, and the triage category they are placed in after being assessed by a nurse,” Associate Professor Turhan says.
“We aggregate all the data, and then create a single number that indicates wait time.”
As well as being displayed in the hospital, wait time could also be published on a website or through an app to help people make important decisions before arriving at the emergency ward.
“Families might be able to use the information to choose one hospital over another, or to make a decision about where to park,” Associate Professor Turhan says.
“After the initial presentation, they could find the waiting time indicator useful for organising childcare, or planning when to eat.”
Options such as this can help reduce stress in the waiting room.
The waiting time prediction model may also prove useful for ambulance routing, as drivers could choose to avoid the most busy emergency departments if their patient wasn’t critically ill.
All in all, the Faculty of IT’s digital health innovation is founded in collaboration – and not just amongst technology experts. Research and product development processes include doctors, nurses, ambulance workers, emergency call takers and other professionals working in healthcare. This approach ensures any new tools not only deliver better health outcomes for patients, but also are designed in a way to complement the skills and workplace setups in hospitals and other settings.