AI Algorithm Could Help Predict SUD, Personalize Rehab Treatment Plans

Researchers at Pennsylvania State University have developed a new algorithm that uses artificial intelligence (AI) to identify homeless youth at risk of developing substance use disorder (SUD).

The hope is that the formula will eventually help homeless centers intervene with personalized rehab programs before and after at-risk youth develop SUD.

“If it brings about even a small change in the prevalence [of SUD], I’ll be satisfied,” Amulya Yadav, principal investigator on the project, told Behavioral Health Business.


Yadav, who is also an assistant professor of information sciences and technology at Penn State, got the idea for the algorithm from his past work, which involved using AI and machine learning to try to help homeless youth in another way.

Before the SUD project, the focus of Yadav’s research was on encouraging safer sex practices among that population.

“Now we are trying to shift focus,” Yadav said. “The reason people engage in risky sex practices many times is because they’re under the influence of drugs. … We want to do something about the root cause.”

The algorithm is based on a dataset of more than 1,400 homeless people ages 18 to 26 across six US states. The data was collected by Anamika Barman-Adhikari, an assistant professor of social work at the University of Denver.

“She had collected data about homeless youth, … [from] their education status … to their income level,” Yadav said. “Have they been involved in the foster care system? Have they ever had any involvement with the justice system or in gangs? Do they have a history of mental health issues, etc?”

The data also contained information about each persons’ substance use behavior.

Yadav looked for associations in the data in order to determine which factors were most commonly linked to SUD. From there, he built a machine learning model to predict which homeless youth were most likely to have SUD issues.

“Basically, we tried to build a map of all the factors that were associated with substance abuse, and then also determine what the strength of association between these factors and substance abuse were,” Yadav said.

Yadav and his team went on to apply the data to rehabilitation, helping to develop personalized rehab and pre-SUD interventions to address the root causes of each patient’s addiction.

While the model works in theory, Yadav and his team still have to figure out if it works in practice. They’re currently in talks with a few homeless youth centers across the country to do a small pilot test of the AI tech. If that is successful, the next step is to put the software in an app that could be easily used by homeless youth centers.

While that possibility is still years away from reality, Yadav is hopeful.

“All of this will happen only if we find through our pilot studies and initial deployments that this algorithm and these techniques truly work and that they are providing a benefit above and beyond what homeless youth centers are able to do by themselves,” he said. “Our software has to be able to run on limited low-end devices so that [homeless centers] have an incentive to use the systems. That’s the long-term vision.”