Dr. Peter Voros, vice-president of clinical services, inside the Command Centre at Humber River Health in Toronto on Oct. 10. On a tour this week, Voros showed off how the command centre and its staff keep patients flowing.Yader Guzman/The Globe and Mail
Shortly after he co-won the Nobel Prize in Physics this week for pioneering modern artificial intelligence, Geoffrey Hinton digressed from his well-known warnings about the dangers of AI to praise the technology’s promise for medical care.
“I want to emphasize that AI is going to do tremendous good,” the British-Canadian computer scientist told The Globe and Mail. “In areas like health care, it’s going to be amazing.”
One part of Canadian health care where AI could make a genuine difference is in reducing emergency-department waiting times, a perennial problem that has grown worse since the COVID-19 pandemic, according to hospital-level data that The Globe obtained through access-to-information requests for its Secret Canada: Your Health project.
The project revealed that data on emergency-department waiting times is often incomplete or inaccessible, making it difficult for average Canadians to judge the performance of their local hospital.
Where The Globe got its hands on waiting times for individual facilities, it was able to identify the laggards and the standouts – including one Toronto hospital that already uses machine learning in a NASA-style command centre to get patients from the emergency department to an in-patient bed faster than any another hospital in the city.
At Humber River Health in northwest Toronto, emergency patients requiring admission have waited just more than 12 hours, on average, for a bed since the start of 2023, according to the administrative agency Ontario Health. That’s notably shorter than the Ontario average of about 17 hours, and significantly shorter than Canada’s worst-performing facilities, where patients can be stuck in emergency departments for two or three days.
Humber River achieved those results while running the busiest emergency department in Ontario.
Barb Collins, the chief executive officer of Humber River and a veteran nurse, credits the hospital’s ER waiting times to the way staff work with advanced technology that was embedded across the Wilson Avenue site when it opened in 2015.
Rudimentary machine learning – a branch of artificial intelligence in which computers trained on mountains of data teach themselves to get better at tasks over time – was critical to the hospital’s command centre from the beginning, she said.
An early goal was to improve “patient flow” so that sick people in the ER wouldn’t be stuck waiting for beds to be vacated by longer-tenured patients who could have been discharged earlier.
“One of the root problems in health care is this never-ending flow of patients,” Ms. Collins said. “What you’re always battling every day in your work environment is, who needs a bed? How quickly can I get them there?”
Barbara Collins, president and CEO of Humber River Health, credits the hospital’s ER waiting times to the way staff work with advanced technology.Yader Guzman/The Globe and Mail
On a tour this week, Peter Voros, Humber River’s vice-president of clinical programs, showed off how the command centre and its staff keep patients flowing. One of the big-screen TVs displayed how long patients had been sitting in the ER’s “O-zone,” where patients wait if they are expected to be examined and sent home. Experienced nurses known as clinical expediters monitor the screens and intervene if a patient waits longer than the targeted time for their health complaint.
Another screen showed the length of stay for patients who came to Humber from nursing homes or complex-continuing-care facilities. The numbers turned orange if patients were in danger of losing their long-term care beds for staying in hospital for more days than allowed under Ontario rules. The orange warning prompts staff to put extra effort into getting those patients discharged back to their nursing homes, if possible.
Many of the command centre’s features use older digital technology, but recent upgrades rely on modern AI, including one feature that analyzes reams of data to predict when and where the hospital will need the most housekeepers and porters. If empty rooms aren’t cleaned right away, or patients wait too long for a porter to bring them down for a CT scan, the hospital gets backed up.
Humber, in partnership with consulting firm Deloitte, is now preparing to roll out an AI-enabled tool to predict traffic into the ER and to allow patients not in need of immediate emergency care to reserve slots when the department is quiet. The plan is for patients to book those slots from a kiosk in the ER beginning in November, and from a smartphone by next March. The project is backed by a $1.5-million investment from SCALE AI, a Canadian public-private AI accelerator.
“This really is the pure AI piece,” Dr. Voros explained. “If we see it’s really busy, and we throw on an extra physician, the AI will change that prediction, because it now knows there’s another doc working. If there is an accident on the 401 and we have a large influx of ambulances from that accident, the AI knows that and will change the prediction.”
Other emergency departments have found ways to use AI to see more patients and reduce physician burnout.
At Michael Garron Hospital in eastern Toronto, eight emergency physicians recently tested AI scribe services to automate their charting. With permission from their patients or next of kin, the doctors turned on a smartphone app to record their interactions. The scribe transformed their interactions into clinical notes that the doctors could check over and edit.
“People were convinced this could never work in the resuscitation room. It’s too chaotic, too busy,” said David Rosenstein, an emergency physician and IT lead for the department. “In fact, it’s kind of the opposite, in the sense that the one time that you can’t sit there taking notes is when you’re actually with a very sick patient at the bedside.”
Dr. Rosenstein found that the AI scribe gave him time to see two or three extra patients on every shift. His colleague, Kyle Vojdani, chief of the emergency department, said it saved him hours of time that he used to spend charting at the end of every shift. All 65 of the hospital’s emergency physicians have decided to spring for the cost of an off-the-shelf AI scribe out of their own billings to improve the department’s efficiency as they prepare for a surge of winter illnesses.
At Toronto’s Hospital for Sick Children, Devin Singh, an emergency physician and computer scientist, helped to create a platform called Hero AI that monitors the waiting room to ensure that high-risk patients aren’t left in the queue dangerously long.
“For most people, a prolonged wait time is just an inconvenience, right – and it’s a terrible inconvenience at that,” Dr. Singh said. “But for some people, a prolonged wait time is literally the difference between life and death.”
Promising as AI is in some ERs, most of these innovations are limited to pilot projects or to big-city hospitals for now.
For smaller emergency departments, especially in rural Canada, AI could help around the margins, but won’t solve a fundamental shortage of staff and in-patient beds, said Alan Drummond, a veteran emergency doctor in Perth, a town of 6,500 in eastern Ontario. His hospital is perpetually full of elderly patients with multiple chronic illnesses who can’t get into long-term care or get enough home care.
“So what” if AI can help predict ER traffic, Dr. Drummond asked. “Where are you going to put them? There are no beds.”
With reports from Yang Sun and Ivan Semeniuk