CHUM to use AI to reduce overcrowding in ERs

MONTREAL – The emergency room at the University of Montreal Hospital (CHUM) is testing a new triage system using artificial intelligence that is designed to move people in need of care through the waiting room faster. Index Santé listed the CHUM as operating at 131 percent capacity in late December.

For example, “if you enter the ER, you’re 88 years old, you have bad vital signs, you’re short of breath, and the algorithm predicts a 95 percent chance that you need to be admitted to the hospital,” said emergency room physician Dr. Elyse Berger-Pelletier (pictured), “we want to use that knowledge to do all the steps quicker, so the admission goes earlier in the patient perspective. So, we have less waste of time and, probably, better care for the patient.”

Dr. Berger-Pelletier is working as a consultant on the project, which the team hopes will roll out at the CHUM ER sometime in 2023. After that, they say, they want to expand it to other departments, and then other hospitals.

The AI system uses mass amounts of ER data to predict the patients’ needs. If all goes well, they’ll be able to allocate resources to departments to receive patients before they arrive.

“What we want to do is to predict the admissions every day so that doctors will be aware of it, so that nurses will be aware of it, and we could plan the resources differently,” Dr. Berger-Pelletier told CTV News.

“This is as big a no-brainer as you can imagine,” said tech expert Carmi Levy. “This is what AI was born to do, and it’ll be interesting to see how that plays out, but really it’s the smart thing to do considering where we’re at in healthcare today.”

During the system’s rollout across the network, experts say the AI will have to be monitored and reviewed regularly so that any bias present in the data – for example, if certain groups were not provided equal care – can be ironed out.

“Artificial intelligence is only as smart as the data it is fed,” said Levy. “So, if you already had infrastructure that was, for example, biased towards not providing services properly to one group versus another, then that’s simply going to be amplified.”

“It’s really a problem in all the world of AI, particularly in healthcare,” said Dr. Berger-Pelletier, who added that a variety of healthcare professionals would be involved in reviewing the data as the system is tested.

“If, as an emergency doctor, I see [a trend] and it doesn’t make sense, like ‘that kind of population usually gets admitted, how come the computer didn’t figure that out?’ ‘Oh, it’s because we have a bias in the data’.”

If implemented properly, Levy said AI could serve to further avoid bias during triage.

“In other words, introducing artificial intelligence into the equation gives us an opportunity to fix some of the bias problems that we might have had in healthcare delivery in the past,” he said.