Artificial intelligence flags behavioral health issues

Sanford Health model designed to spot risk of self-harm, psychosis or chemical dependency

Female doctor typing on computer.

Artificial intelligence is helping Sanford Health physicians predict potential behavioral health conditions in patients based on changes to their medical history.

Clark Casarella, a data scientist with the organization’s enterprise data and analytics team, developed the model. It’s called Predicting Health Outcome in a Behavioral Health Outpatient Setting, or PHOBOS.

The goal is to predict whether a person might develop three common behavioral health conditions: self-harm, psychosis-type events or chemical dependency.

“Those are, honestly, life-destroying in some cases,” Casarella said. “You might have slipped through the cracks before. And now with this model that chance is lowered.”

Find a provider: Behavioral health specialists at Sanford

The AI notifies the health care provider of the patient’s PHOBOS score for each of the three conditions. It displays as a percentage for the likelihood they could suffer from one or more of them in the coming year. The health care provider can then independently review the basis for that recommendation along with other relevant inputs to determine if changes in a patient’s care are needed.

“The health care provider is not looking through 25 years of medical records,” Casarella said. “They’re not able to see all of that data all at once, whereas as an artificial intelligence can plow through it in a couple of hours.”

Just as behavioral health issues can cause physical health problems, serious medical conditions also can impact a person’s mental well-being, he said.

“We’re looking at how a person’s total diagnosis history changes over time to help us catch maybe a bad cancer diagnosis that causes a person’s mental health to decline. Or any type of traumatic experience that one might have. Some diagnoses are hard to mentally cope with, so we wanted to make sure that we’re catching all of that. If somebody’s been ticking every single year with some bad diagnosis, that could eventually break a person’s psyche,” Casarella said.

Difficult to predict

Jeff Leichter, Ph.D., is a Sanford Health psychologist in Detroit Lakes, Minnesota, and the project’s provider champion. He said the PHOBOS score is another tool in a challenging area of health care.

“By and large in our profession of behavioral health, we are not very good at predicting future behavior. Even in the research on suicide and self-harm, the predictive analytics that we use historically have been only slightly better than chance,” he said.

Some people who have had chronic suicidal ideation for years never actually act on those feelings. Other patients who are seemingly happy and healthy can compulsively react to extreme acute challenges and hurt themselves.

“What’s attractive about this algorithm is it has the potential to use multiple data points to help us predict these negative outcomes, suicide being one,” Leichter said.

Three benefits

Leichter said the AI tool has the potential to help in three ways.

Prioritize 

First, it could prioritize access to behavioral health services.

Leichter likens it to vehicles trying to merge onto a busy city freeway onramp during rush hour. If the onramp could rate the vehicles based on urgency, it could give higher priority to a pregnant woman whose water just broke and is hurrying to the hospital over the guy picking up a pizza, for example.

Inform

The second potential benefit could be to inform primary care physicians, Leichter said.

Most clinic visits last 15 minutes, so if a patient comes in with multiple issues, a high PHOBOS score would help the doctor focus on their mental health.

“If you’ve got a PHOBOS score near the top, I need to spend a good chunk of that 15 minutes with you assessing where your mood is at, where your emotional health is at. And I might not be aware of that without this tool,” Leichter said.

Just as the physician uses lab results to identify potential heart trouble, diabetes or other disease, they may be able to use the PHOBOS score as another marker to spot potential behavioral health concerns.

“It doesn’t tell the doctor what to do,” Leichter said. “It doesn’t replace clinical judgment. It augments and supports clinical judgment.”

Measure improvement

Finally, Leichter said the AI also might provide a way to measure improvement in care.

The provider must still ask important questions about the patient’s work schedule, sleeping habits, relationships and other factors. But the AI can help assess if the care plan is working.

“Six months into treatment, what we would hope to see is your PHOBOS score would reduce, if what we’re doing is being helpful,” he said.

Pandemic effect

Casarella finished the model in March, just as the pandemic hit. That’s obviously one of those possibly life-altering situations that will likely show up in medical records this year, he said.

“We’re going to see a lot of mental health declines. People have been at home. People are scared. The virus itself has undetermined mental health effects. I had it and there’s been weeks where you can hardly concentrate. That’s a real tangible effect,” he said.

“So seeing how all of the 2020 data that’s going to go into this model will be super interesting because that’s a whole other thing that we can add into what we’re predicting for this model.”

Read more

Posted In Behavioral Health, Chemical Dependency, Detroit Lakes, Innovations

Leave A Reply