Toronto hospital prescribes AI to cure ER wait times

At the corner of Queen Street and Victoria Street in downtown Toronto lies one of the busiest hospitals in the entire country – St. Michael’s Hospital, the large trauma Centre within Unity Health Toronto. With a prime location in the city’s downtown core, it’s no surprise that St. Mike’s sees a massive influx of patients and visitors each year. And without the ability to accurately forecast the number of daily visitors to the hospital’s ER, wait times can be just as unpredictable.

However, as Muhammad Mamdani, Vice President, Data Science and Advanced Analytics and Jennifer DeCaria, Senior Director of Business Development at Unity Health Toronto explain, it doesn’t have to be that way. “In a year, we have over 75,000 emergency department visits, 25,000+ inpatient visits, over half a million ambulatory visits and about 15,000 surgery procedures,” says Mamdani. “So, if you can imagine, patients who walk in need labs, medications, orders, vitals monitoring, medical imaging – and from that, we have substantial data that’s constantly coming in. Unfortunately, like every other hospital in the country, we haven’t been using it very well to manage patient care and hospital processes.”

Taking a turn for the better

This is the challenge for Mamdani’s Data Science and Advanced Analytics team at Unity Health Toronto. How can they use all this data to drive better cost-efficiency while improving patient care? The first step was to build the infrastructure. “Over a 3-year period, we worked with IBM to create an enterprise data warehouse and an analytics environment that we could leverage to enable AI,” says Mamdani. “I think it’s fair to say that we have the most advanced applied analytics and computing environment of any hospital in the country now - our analytics database is always updating.”

Next on the agenda was hiring a group of data scientists who could work directly with hospital clinicians and program directors to put the data to work in order to create efficiencies. “Mamdani’s team doesn’t start any projects that haven’t been driven from the bottom-up,” says DeCaria. “That’s right. Instead we say, ‘Who wants to work with us? What is your problem? How is data going to help you make a difference?’” Mamdani adds.

AI in Practicum

Mamdani and his team have been approached by multiple areas of the hospital to help find solutions to everyday problems, creating over 20 different AI tools to date. One challenge in particular has a colossal effect on the way the hospital is staffed. “At St. Mike’s alone, we have 1,600 to 1,700 nurses,” says Mamdani. “Here’s the issue; a nurse gets sick or takes vacation - what happens? In a lot of cases, we must replace that nurse’s shift, otherwise we’re going to be short staffed.” Historically, a nurse’s shift would be covered by another hospital employee, usually one who had already worked a 12-hour shift meaning, from a workforce perspective they may be overworked and from a hospital financial impact would be paid considerably more in overtime. Alternatively, an external agency could be hired to outsource nurses, which has an even bigger negative financial impact to the hospital system.

To rectify this, the AI team studied the data, analyzing how many nurses should be on staff at a time. “By looking at just 18 months of historical data we saw patterns,” says Mamdani. “We saw predictable spikes allowing us to forecast for the next year to see how many nurses are going to get sick or go on vacation and when that is going to happen. Our optimization model suggested that a ‘floater’ team needs to consist of 37 nurses instead of the planned 22 – and even specified how many of those should be regular and critical care nurses.” This model has now been implemented at St. Michaels and St. Josephs hospitals, with an estimated cost reduction of over a million dollars per year.

But how has the team worked to improve patient care? One of the biggest concerns Ontarians have about healthcare has to do with hospital wait times, and fortunately, the Unity Health AI team has also found a way to better predict overcrowding in the ER. “Our experience has been that, every year, we have two to three major surges, and two to three minor surges every month,” says Mamdani. ‘It becomes a lot for the ER team to handle.”

The team used three years of historical data on ER visits, weather patterns, and city planning data (are the Raptors playing tonight? Is there a marathon on the Lakeshore this weekend?) and put them into an AI algorithm that uses a variety of machine learning models called an “Ensemble Model.” The algorithm figures out how many patients are going to visit the ER on any given day.

The result: the tool predicts with well over 90 per cent accuracy and can forecast any increment of time. This information helps the hospital to schedule enough staff, ensuring shorter wait times. “[Ironically], we have a list of hospitals lining up for this tool,” says DeCaria. “We’re now getting ready to move to commercialization and have organizations across the country that want to collaborate.”

A Healthy Range of Opportunity

With a technology so powerful, and a corridor of talent right outside their doorstep, Mamdani and his team know the possibilities for their Advanced Analytics team are endless.

“I think there are a lot of opportunities” says Mamdani. “Vector, Ryerson and U of T are major partners for us. We get many students that spend months with us working on projects that are often first in the world. It’s impressive to be able to leverage all that talent.”

The government also plays a key role in supporting innovations to healthcare.

“We have a government that has identified they want to revamp the space,” says Mamdani. “The Ontario health board is eager and hungry to learn and listen and Ontario is moving very quickly with respect to both healthcare and big data. For us, it couldn’t be a better combination.”