How Toronto's SickKids hopes to use artificial intelligence to predict cardiac arrests
Computer model can predict up to 70% of cardiac arrests 5 minutes before, researcher says
Toronto's Hospital for Sick Children is pioneering the use of artificial intelligence (AI) to help save infants at risk of imminent cardiac arrest.
Anna Goldenberg, the hospital's first-ever chair of biomedical informatics and AI, has developed a computer model that can tell whether a patient will go into cardiac arrest in the next five minutes.
This machine-learning research, she told The Current's Anna Maria Tremonti, relies on the patient's current and historic medical data — like blood pressure and heart rate — to make the prediction.
"We are seeing ... a lot of physiological signals being measured and recorded, and clinicians, doctors and nurses have to kind of monitor it," said Goldenberg.
- Robo-doctors? How AI could do jobs we once thought couldn't be automated
The more physiological signals there are, however, means it is increasingly difficult for doctors to make quick decisions about their patients' heart health, she explained
"We are using more of the historical data to help to make a prediction," Goldenberg said of the type of AI algorithm being used in patient care in the hospital's emergency department.
Model has 70% success rate
So far, she says, the computer model has been able to predict up to 70 per cent of cardiac arrests just five minutes before and is on the frontier of machine-learning research in medicine.
Most of this research remains in medical journals, Goldenberg explains, but her team at SickKids is trying to "bring that power into the hospital" by deciphering what medical data is valuable to doctors.
Health care, right now, is a really rich place for us to be really trying to understand this [AI].- Jennifer Gibson
In some cases five minutes might not be enough for the doctor to determine the appropriate intervention to prevent the child from going into cardiac arrest, she adds, noting that her department is working on a model that is able to predict cardiac arrest earlier.
"What we're aiming is to really say we have this wealth of information of the patients that we have observed in the past and based on all that knowledge, we can deduce some patterns that may be useful for the condition," Goldenberg said.
'The human connection is so clear'
Jennifer Gibson, director of the University of Toronto's Joint Centre for Bioethics, describes the present-time as a "transitional moment" for medicine.
Yet many questions about how best to use AI in the medical field and the long-term impacts still hover over its application.
Gibson says this uncertainty is bringing people together from all different fields — computer scientists to legal experts, clinicians and policymakers — to find answers.
"We all have a stake in how we manage this," Gibson said.
"Health care, right now, is a really rich place for us to be really trying to understand this [AI], because the human connection is so clear."
Click 'listen' near the top of this page to hear the full conversation.
Written by Kirsten Fenn. Produced by Imogen Birchard.