Waypoint develops AI-powered early warning system for mental health

Waypoint Centre for Mental Health Care and its research institute have developed an AI-powered early warning system to anticipate when patients are at risk of mental health deterioration.

Led by Dr. Andrea Waddell, Waypoint’s medical director for clinical informatics and quality, the solution leverages integrated patient data from the organization’s MEDITECH Expanse electronic health record (EHR) to calculate risks. The project has been successfully piloted in four 20-bed units at the Penetanguishene, Ontario, hospital.

“Psychiatric admissions are associated with more incidents of preventable harm than other hospital admissions,” said Dr. Waddell, who is also associate director of adult psychiatry and health systems at the University of Toronto’s Temerty Faculty of Medicine and Ontario Health Central Region’s clinical lead for mental health and addictions.

“Mental health deterioration,” she said, “can result in suicidal ideation, self-harm and violence against hospital staff and other patients, so if we can identify patients who are worsening, we can intervene and prevent these kinds of events.”

Waypoint’s high-secure provincial forensics program cares for some of the most severely ill patients in the province. They include those who come in contact with the legal system and are referred to mental health care by the Ontario Review Board if found to be not criminally responsible or unfit to stand trial.

The hospital also offers mental health inpatient services for the Central Region and a forensic assessment program for patients admitted from the courts, correctional facilities and regional psychiatric hospitals.

Waypoint also consulted with St. Michael’s Hospital in Toronto to understand the principles they used to build and deploy their ChartWatch tool. “ChartWatch looks at patients on medical units and identifies if they’re deteriorating to avoid adverse outcomes and admission to the ICU,” she noted.

The AI early warning application, developed with the assistance of Dr. Christo El Morr and Dr. Elham Dolatabadi of York University. After approval from the hospital’s Research Ethics Board, the model was trained on four years of deidentified data from thousands of Waypoint patients to learn what risk factors are associated with adverse outcomes.

“There are what we call static risk factors and dynamic risk factors,” explained Dr. Waddell. “Static risk factors are things that happened to a person in the past and are predictive, but don’t change over time.

“So, if someone has been violent in the past, they’re much more likely to be violent in the future. If someone has had a suicide attempt in the past, they’re at much higher risk of a future suicide attempt.” Other static risk factors include co-morbidities, chronic pain and substance abuse.

Dynamic risk factors include changes in medication, refusing medication, changes in eating and sleeping patterns, as well as nursing assessments of a patient’s mental status and any signs of agitation or anxiety.

All of the data used by the model is routinely collected by clinical staff and displayed in Waypoint’s MEDITECH EHR system, so there’s no additional workload for staff.

During a pilot over the preceding five to six months, said Dr. Waddell, the model was approximately 98 percent accurate in identifying patients at low risk of mental health deterioration: “It was very effective as a screening tool to be able to identify patients who are proceeding along their expected trajectory.”

For people who were identified as high-risk, approximately one-third of them actually do go on to have an adverse event in the subsequent 24 hours.

“So, it’s not perfect,” she said. “Everyone flagged as high-risk doesn’t necessarily go on to have an event, but physicians found that rate acceptable because it does screen out the other patients” and it’s not unduly onerous to check up on those few identified as high-risk.

Physicians use the early warning system as a decision support tool and will typically review the patient’s care plan. They may increase a patient’s access to pleasurable activities, adjust their medication or increase engagement with nursing staff.

Dr. Waddell cited the example of one patient who was flagged as a high-risk for mental health deterioration the same day they were scheduled for discharge. “That was a fairly dramatic example of the tool’s value. The physician cancelled the patient’s discharge, avoiding a potentially adverse outcome and making a big difference in the patient’s recovery.”

The machine learning model uses data from MEDITECH, but the results identifying patients at risk are currently external to the EHR and communicated to physicians via email.

Manual distribution of the results is feasible with the limited deployment of the tool in the context of a pilot, but far from ideal once it’s rolled out hospital-wide to 25 physicians, so Waypoint is working with MEDITECH on an Application Programming Interface (API) that will distribute the results through the EHR.

“That will be much more seamless for the physicians as they’ll see the alerts with the names of the patients when they log in to MEDITECH every morning,” said Dr. Waddell.

“We are pleased to be a part of Waypoint’s effort to improve patient safety and workflow efficiency through integration with the Early Warning System,” said MEDITECH director of Canadian market and product strategy Bob Molloy. “By designing a cloud-native EHR that adheres to the latest FHIR standards, we provide our customers with a platform that is interoperable and scalable. This approach supports innovation now and, in the future, to ensure that organizations like Waypoint are able to continue to advance healthcare.”

A prototype of the API is currently in development. A clinical trial will then be conducted to determine if the alerts improve patient outcomes. The early warning system is expected to be deployed hospital-wide by the end of this year or early spring 2026.