In the bustling ward of St. Michael’s Hospital in downtown Toronto, one of Shirley Bell’s patients was suffering from a cat bite and a fever but seemed otherwise healthy — until an AI-powered early warning system alert revealed he was more serious than he appeared.
While the nursing team would typically check blood test results around noon, the technology would flag the results coming in hours earlier. That alert showed that the patient’s white-cell count was “really, really high,” recalls Bell, a clinical nurse educator in the hospital’s general medicine program.
The cause turned out to be cellulitis, a bacterial skin infection. Without prompt treatment, it can lead to extensive tissue damage, amputation and even death. Bell said the patient was quickly given antibiotics to avoid the worst-case scenario, thanks in large part to the team’s in-house AI technology, called Chartwatch.
“There are many, many other scenarios where the patient’s condition is flagged earlier, the nurse is notified earlier, and interventions are put in place earlier,” she said. “It doesn’t replace the nurse at the bedside; it actually improves your nursing care.”
A year and a half of Chartwatch research, published on Monday The Canadian Medical Association Journal found that the use of an AI system led to a significant 26% reduction in unexpected deaths among hospitalized patients.
“We’re excited to be saving lives,” said co-author Dr. Muhammad Mamdani, vice president of data science and advanced analytics at Unity Health Toronto and director of the Centre for Artificial Intelligence Research and Medical Education at the Temerty Faculty of Medicine at the University of Toronto.
“A promising sign”
The research team analyzed more than 13,000 admissions to the general medicine department at St. Michael’s Hospital — an 84-bed unit that treats some of the hospital’s most challenging patients — to compare the tool’s impact on this group of patients with thousands of admissions to other specialty departments.
“During this time period, in other areas of our hospital that weren’t using Chartwatch, we didn’t see any change in the number of these unexpected deaths,” said lead author Dr. Amol Verma, a clinician-scientist at St. Michael’s, one of three hospitals in the Unity Health Toronto network, and a Temerty professor of artificial intelligence research and medical education at the University of Toronto.
“It was a promising sign.”
The Unity Health AI team began working on Chartwatch in 2017 based on suggestions from employees that predicting death or serious illness could be a key area where machine learning could have a positive impact.
The technology underwent rigorous testing and development for several years before being launched in October 2020, Verma said.
Chartwatch measures about 100 inputs from [a patient’s] medical records that are now routinely collected as part of the care delivery process,” he explained. “So the patient’s vital signs, their heart rate, their blood pressure… all the lab results that are done every day.”
Working in the background, in conjunction with clinical teams, the tool monitors any changes to a person’s medical record “and makes a dynamic prediction every hour as to whether a given patient’s health is likely to deteriorate in the future,” Verma told CBC News.
This could mean someone is going to feel worse, need intensive care, or even be on the brink of death. This gives doctors and nurses a chance to intervene.
In some cases, these interventions include escalating a person’s level of treatment in order to save their life or providing early palliative care in situations where patients can no longer be saved.
In both cases, the researchers say, Chartwatch appears to complement doctors’ judgment and lead to better outcomes for frail patients, helping to avoid sudden and potentially avoidable deaths.
Artificial intelligence in healthcare is gaining popularity
Beyond its applications in medicine, artificial intelligence has been generating a lot of buzz — and opposition — in recent years.
From the controversy surrounding the use of machine learning software to write academic essays to concerns about the ability of AI to create realistic audio and video content that imitates real celebrities, politicians, or average citizens, there are many reasons to approach this new technology with caution.
Verma himself said he has long been cautious. But in health care, he stressed, these tools have enormous potential to combat staffing shortages that plague Canada’s health care system by supplementing traditional bedside care.
Many of these efforts are still in the early stages of development. Various research teams, including private companies, are exploring ways to use AI to detect cancer earlier. Some studies suggest it has the potential arterial hypertension just by listening to someone’s voice; others show that it can scan brain patterns, detect symptoms of concussion.
Verma emphasized that Chartwatch is noteworthy for its effectiveness in keeping real patients alive.
“Very few AI technologies have actually been implemented in a clinical setting. This is, as far as we know, one of the first in Canada that has actually been implemented to help us with day-to-day patient care in our hospital,” he said.
A ‘Real’ Look at AI’s Impact on Healthcare
The research at St. Michael’s has some limitations. The study was conducted during the COVID-19 pandemic, a time when the health care system is facing an unusual set of challenges. The team acknowledged that the city’s hospital’s patient population is also diverse, with a high level of complexity among patients, including homelessness, addictions and overlapping health issues.
“Our study was not a randomized controlled trial in multiple hospitals. It was done within one organization, within one unit,” Verma said. “So before we say this tool can be widely used everywhere, I think we need to do research on its use in multiple contexts.”
Dr. John-Jose Nunez, a psychiatrist and researcher at the University of British Columbia — who was not involved in the study — agreed that the research needs to be replicated elsewhere to better understand how well Chartwatch might work in other settings. He added that patient privacy also needs to be considered when using any new AI technology.
Still, he praised the research team for providing a “real-world” example of how machine learning can improve patient care.
“I really think AI tools are going to become another member of the clinical care team,” he said.
The Unity Health team hopes their technology will be used more widely in the future within their own network of hospitals in Toronto and beyond.
Much of this work is done through TWINSCanada’s largest hospital data-sharing network for research and analytics,” said Mamdani, vice president of data science at Unity Health.
He added that more than 30 hospitals in Ontario are working together to offer testing opportunities for Chartwatch and other AI tools across clinical settings and hospitals.
“This just lays the groundwork for implementing these solutions far beyond our four walls,” Mamdani said.