Electronic Records
Fraser Health, in British Columbia, aims to become an AI leader
June 27, 2024
Fraser Health, in British Columbia, is emerging as one of Canada’s most technologically innovative health regions – especially when it comes to deploying artificial intelligence (AI) across the enterprise.
“We have 40-plus AI project in various stages of development, and we’ve deployed 12 of them,” said Sheazin Premji, executive director of Fraser Health’s Centre for Advanced Analytics, Data Science and Innovation (CAADSI).
“Our team is scouring the world for best practices. We don’t want to be following, we want to be leading,” she added.
The CAADSI team consists of engineers, analysts, consultants and data scientists. For its part, Fraser Health serves a population of nearly 2 million people and is the largest of BC’s five health regions.
One of the key projects under way is the Digital Twin, which recreates the workings of the health region as a digital model. It’s said to be the first ‘enterprise wide’ deployment of a digital twin system in Canada.
While the solution is in its initial stages, it will ultimately enable health region administrators and clinicians to test different scenarios and by doing so, to improve patient flow, work patterns and hopefully, medical outcomes, too.
These sophisticated ‘what-if’ scenarios could include everything from seeing what happens when you change the location of a service – such as the medical imaging department – to altering the course of a treatment.
One could even experiment with designing and adding an entire building to the region.
But by using a computerized, AI-driven model, it can be done much more efficiently and at less cost.
“It would be very expensive to test in real life, from a capital-intensive point of view,” said Jennifer MacGregor, vice president, Digital Health and Provider Experience at Fraser Health.
MacGregor and colleagues from Fraser Health gave an hour-long presentation about the Digital Twin and other AI projects at the recent e-Health conference in Vancouver.
Creating the Digital Twin is not a simple exercise, they noted, as it requires massive stores of data – in the case of Fraser Health, the region has built up 16 terabytes of data said senior director Casper Shyr. And the volume is quickly rising each day.
The project also relies upon the ability of the team to make the right connections between data, so that real relationships are modeled and accurate predictions can be made.
It’s an iterative process, involving trial and error and getting the tweaks right. “You’ve got to plan, implement, measure, plan, implement, measure,” said MacGregor, stressing the process is continued until a satisfactory result is obtained.
MacGregor noted that the Digital Twin is initially much like a 2-dimensional dashboard, but this isn’t the final interface. Instead, she envisions a 3-dimensional model – perhaps like a Sims-style simulation of buildings and people, where the model changes as various parameters are altered.
That 3D interface should be even easier to use than the current dashboard, with its charts, speedometers and buttons. “We want a self-serve model,” she asserted. “So that the end-user can use it to answer their own questions, like with ChatGPT. We want it to be conversational. That’s where we want it to go.”
While Fraser Health has a team dedicated to technological advances – which includes work on the region’s new implementation of MEDITECH Expanse – it is also collaborating with members of the private sector to build the Digital Twin. They include Verto and Deloitte.
Toronto-based Verto Health is supplying a good deal of the technology for the Digital Twin. The company has created digital twins in small pieces at many hospitals across Canada, but this is the first enterprise-wide deployment.
On the AI side, Millar explained that the Digital Twin is consuming a lot of unstructured data – such as encounter notes and discharge summaries – and making sense of it, said Michael Millar, CEO and co-founder of Verto.
Using this data, and other feeds, the system will be able to identify and flag numerous issues and bring them to the attention of caregivers.
For example, said Millar, “If a person goes for treatment to the ER, and the system sees that the person also has mental health issues but no regular provider, it will alert the hospital that the person needs supports once he is discharged and back in the community.”
Those kinds of supports can reduce the likelihood of a return visit to the ED and can promote the health of the individual over the long term.
The system, in future, will be able to mine unstructured and structured data and will be able to make predictions about the care needs of people – even before they enter hospital.
“It’s a quantum leap in the way health data is used,” said Millar.
Right now, the Digital Twin is focused on the Emergency Departments at Fraser Health, as the project gets off the ground. “We’ve got a lot of wicked problems that we want to solve,” said MacGregor.
The virtual model will track how ED patients arrive at the 12 regional hospitals, how they move through the facilities, and what tests, medications and treatments they are given.
The system will also analyze how they are getting to the emergency departments, whether they are virtual care patients, and the time taken for them to move through the ED, into the hospital or to be discharged.
In this way, staff and clinicians can use the Digital Twin for problem-solving – finding out how patients and staff are affected by various factors and changes.
While the Digital Twin is just getting off the ground, Fraser Health’s AI-driven tool for predicting potential discharges is well-established in the region. And the system is making a significant difference.
“We want to be at a 95 percent occupancy rate in acute care beds, not at the 115 percent that we sometimes see,” said Teresa O’Callaghan, regional executive director at Fraser Health. “Then we ask, how many discharges would that take? And we also ask, realistically, how many discharges can we expect? That’s where the dashboard comes in.”
Recently, this solution helped the region discharge 500 patients on one day – reducing acute care bed occupancy from 3,100 beds down to 2,600. “Imagine that, 500 patients discharged in a single day,” she mused.
The discharge dashboard tracks all inpatients across the region and shows users whether these patients have completed all their tests. It tracks their vital signs and shows whether the patients have met certain benchmarks and are theoretically ready to leave.
“It’s not meant to usurp the human individual, and we’re not telling doctors who to discharge,” said O’Callaghan. “But it does predict when a person could reasonably be expected to be discharged. We now rely on this at Fraser Health.”
She emphasized that doctors are the ones in charge of discharge, but the tool can be used to alert and remind them when patients need their attention for this purpose. “Why have patients in beds who could go, if they don’t have to be there,” she asked.
Finally, Dr. Amyeen Hassanali, CIO at the health region, discussed Fraser Health’s innovative chatbot for assisting clinicians and staff with questions about the new MEDITECH Expanse electronic health record.
He explained that a good deal of work went into training Fraser Health’s 50,000 clinicians, staff and volunteers about using the new system.
As part of the effort, special trainers were employed to give users assistance after the Expanse system went online. However, people have only so much memory, and having trainers on the ground to provide support with a new system while navigating patient care was valuable.
For example, a clinician might forget how to transfer a patient from Fraser Health to a health region using a different system. For some time, the trainers were on hand to help.
But after a while, “It becomes expensive to keep the trainers employed,” said Dr. Hassanali. So, Fraser Health decided to build its own AI-driven bot that could answer the questions of clinicians and staff, making it easy for them to solve their problems.
The bot makes use of AWS artificial intelligence and AI tools, including natural language processing. There were online manuals available that could be searched, but as Dr. Hassanali observed, clinicians don’t want to stop what they’re doing to log into a different system and answer a question. With the bot, they could simply ask their questions – ChatGPT style – and quickly receive an answer.