AI-assisted trial to predict delirium launches

TORONTO – A large-scale trial of an AI-driven solution to predict and prevent delirium is rolling out across 13 Ontario hospitals. The randomized-controlled trial, which reduces bias and is considered the gold standard in clinical testing, will involve approximately 15,000 patients.

This pioneering project was funded by Ontario’s Ministry of Health in partnership with Ontario Health, and is a collaboration between the GEMINI research network, based at Unity Health Toronto, the University of Toronto, the Centre for Quality Improvement and Patient Safety, Regional Geriatric Program of Toronto, the Centre for Advancing Collaborative Healthcare & Education, the Toronto Academic Health Sciences Network, and the participating hospitals (see list below).

“Artificial intelligence offers many promising opportunities in healthcare, but we need to ensure it is used safely and effectively. This is an extraordinary partnership between AI scientists, front-line healthcare teams, patients and caregivers, hospital leaders, and policymakers, studying whether AI can help us predict and prevent delirium, one of the most devastating problems in healthcare,” said Dr. Amol Verma (pictured), the project’s principal investigator.

Delirium is a serious condition that affects about 1 in 4 adults hospitalized with medical or surgical problems – about 500,000 Canadians annually. It is a state of new-onset confusion that can make patients agitated and disoriented, which can be profoundly distressing for them and their caregivers.

“In addition to the stress hospitalization was already causing, my husband’s delirium episode triggered fear, confusion, sadness, not just for him, but for our entire family,” said Nicole Lafreniere-Davis, a patient partner on the trial.

Delirium can lead to dementia and increased mortality. Beyond the significant human cost, delirium is also an economic burden on the healthcare system, significantly extending hospital stays and adding an estimated $11,000 in additional costs per hospital admission.

As Canadian healthcare systems grapple with an aging population, workforce shortages and burnout, there is a growing urgency to adopt innovative and scalable solutions. This project contributes to the broader shift toward using AI responsibly to support clinical decision-making and improve patient outcomes and hospital sustainability.

While it can be a devastating condition, up to 40 percent of cases of delirium are preventable. However, delirium prevention relies on frequent monitoring and interventions to address patients’ underlying medical condition, nutrition, sleep, and mental state. This is difficult to sustain with real-world staffing conditions in healthcare.

Dr. Barbara Liu, executive director of the Regional Geriatric Program of Toronto and co-investigator on the project, comments that, “although we have known how to prevent delirium for decades now, it has not been possible for hospitals to implement and sustain prevention efforts given the resources available.”

The “AI Models (AIM) to Prevent Delirium” trial uses AI to identify patients who are at the highest risk for delirium. This allows staff to prioritize the highest risk patients and direct delirium prevention efforts to those who need them the most.

“I am convinced that had our project been implemented, our AI tool would have identified the risk, preventative strategies would have been put into place, and this difficult situation would have been avoided,” said Lafreniere-Davis.

The AI model was developed using state-of-the-art methods that prioritized both prediction accuracy and the simplicity of the tool.

The engineering team led by professor Eldan Cohen at the University of Toronto identified just 10 routinely-available factors, such as a patients’ age, medical conditions and routine lab tests, that can predict delirium risk with 70 percent accuracy – which is impressive given how difficult the condition is to diagnose.

Keeping the number of inputs to the model small ensures that it could be used at every hospital, not just those with advanced IT systems.

Another benefit – keeping the number of inputs to the tool small and understandable ensures strong protection of patient privacy, as the AI does not require any patient identifying information.

“Advanced engineering methods can help us create usable and accessible health AI solutions. The mathematical equations within the model are still quite complex, but the number of inputs is small,” said Cohen.

As with many health AI interventions, the care team’s response to AI alerts is just as important as the AI alert itself. Over the last year, teams across all the participating hospitals have been co-designing the intervention so that delirium prevention efforts can be implemented well within regular workflows.

“It has been inspiring to see nurses, doctors, patients and caregivers, hospital administrators, and quality improvement specialists come together to design and implement this intervention,” said Dr. Brian Wong, director of the Centre for Quality Improvement and Patient Safety.

The trial is launching across 13 hospital sites in Ontario with the study period running through March 31, 2027. By evaluating the program across many different organizations, the project team aims to create AI and implementation tools that are ready to scale widely to protect patients and support clinicians across Canada.

Participating Hospitals:

1. Humber River Health – Hennick Humber Hospital
2. London Health Sciences Centre – University Hospital
3. London Health Sciences Centre – Victoria Hospital
4. North York General Hospital
5. Scarborough Health Network – Scarborough General
6. Scarborough Health Network – Birchmount Hospital
7. Unity Health Toronto – St. Joseph’s Health Centre
8. Unity Health Toronto – St. Michael’s Hospital
9. Sunnybrook Health Sciences Centre
10. University Health Network – Toronto Western
11. Trillium Health Partners – Credit Valley Hospital
12. Trillium Health Partners – Mississauga Hospital
13. Niagara Health Marotta Family Hospital

About GEMINI:

GEMINI is a not-for-profit research program based at Unity Health Toronto (and is unrelated to Google GEMINI). It is Canada’s largest hospital data sharing network for research and analytics to study and improve healthcare. Established in 2014, GEMINI collects and standardizes de-identified clinical data from more than 40 hospitals in Ontario, covering 60 per cent of inpatient adult medicine care, and 70 per cent of paediatric care. GEMINI-driven research has supported more than 150 projects, and enabled researchers across Canada to secure more than $210 million in grant funding.