AI predicting patient discharge times could manage hospital overcrowding, researchers say

Hospitals operating at 95% occupancy rate, according to Vitalité Health Network

Researchers at the Université de Moncton believe artificial intelligence is the answer to improve the way patients are processed in and out of hospitals which would reduce overcrowding in acute care beds.

"We know that the hospitals need some tools to improve the process, to optimize the flow of patients," said Moulay Akhloufi, a computer science professor at the university.

It's why Vitalité Health Network is working with Akhloufi, whose team has developed a predictive AI program that can forecast how long a patient will need hospital care.

The idea is that if nurses and doctors know how long a patient will be in a bed, they can plan discharge and track the availability of beds for the long term.

"Ultimately, this can contribute to reducing access delays, particularly in the emergency department, while maintaining the quality and safety of care," Jenny Toussaint, vice-president of clinical logistics with the network, wrote in an email.

Could AI help manage overcrowding in N.B. hospitals?

According to Toussaint, hospitals across the network are operating with an occupancy rate of 95 per cent.

"Occupancy rates remain a concern and are difficult to maintain at the safe target level of 85%," she said.

Toussaint also said more than a third of patients in acute care beds could be getting the care they need outside the hospital and in the community. But in many cases, those services are not available, she said.

"This creates significant pressure on hospital capacity and affects patient flow across the system," she said.

The AI is still in its infancy and has yet to be implemented into hospital systems.

Toussaint said it's essential for staff to plan for a patient's discharge the moment they occupy a bed. She believes the program will help busy nurses and managers to do that planning more seamlessly.

AI program combing decades of anonymous medical data

Since last fall, Akhloufi and another researcher, Oumeima Thaalbi, have gathered years worth of medical data from hospitals across the province.

The program has examined data such as when a patient came in, the symptoms they presented, and their medical history. It then notes when a patient was discharged.

Combing through years data allows AI to find patterns and accurately predict how long a new patient is likely to need a hospital bed.

If a patient presents with a new symptom, or a doctor orders a test to be run, AI can adjust its prediction accordingly.

"So the algorithm doesn't predict once and stops, it predicts along the time where the person is in the hospital,"Akhloufi said.

Although the program can produce errors, Akhloufi said early results of the study show AI is likely to predict a person's discharge time as accurately as a nurse with 30 years experience, for instance.

Plus, it's more objective in its decision-making, he said.

"As a human, we can make errors because we are sometimes in a rush, probably we will not have all the factors considered," he said. "The algorithms don't do that."

Still, Akhloufi said, it's meant to give staff evidence and more confidence on the decisions they make when allocating resources.

"This tool is not to replace a person. It's a tool to help the person."

The health network anonymized the patient data it shared with Akhloufi's team, according to the professor.

"There is no way we can go back to the specific person," he said, adding that the program identifies patients as numbers which hospitals can only relate to a patient on their own.

Tailoring the algorithm for use at the Dr. Georges-L.-Dumont University Hospital Centre in Moncton is the research team's next step.

Once a pilot program is successfully implemented by Vitalité at the hospital, the goal is to expand the program across the network.

Katelin Belliveau is a CBC reporter based in Moncton.