Researchers at the University of Hawaiʻi at Mānoa are using artificial intelligence and machine learning (AI/ML) to better understand what improves outcomes for individuals receiving treatment for substance use disorders.
A recent study published in The Journal of Prevention Science analyzed more than 7.9 million publicly available treatment records across the U.S. to identify patterns in services, recovery and outcomes. The research was led by Treena Becker, an assistant researcher with the UH Center on Aging , and Alberto Gonzalez-Martinez, a UH computer scientist.
"We believe our research findings can help states and local organizations better understand how to support people in substance use disorder treatment and their long-term recovery journey at a time when drug overdose deaths continue to be a major public health concern across the U.S.," Becker said.
Top predictors of positive treatment outcomes
"We developed and used an ensemble machine learning model called Random Forest Model with the aim to predict the 10 most important features that increase the likelihood of positive treatment outcomes," Becker said.
The analysis found the most important factor associated with positive outcomes was how long an individual remains in treatment, regardless of setting. According to Becker, longer engagement significantly increases the likelihood of reducing or stopping substance use.
Other key factors included treatment accessibility, depending on clinical need, treatment type at entry and at discharge, housing status, participation in self-help groups, employment status and referral source.
Mapping disparities in treatment services
AI/ML tools also allowed researchers to map and visualize the data, revealing patterns difficult to detect using traditional methods. Using the Machine Learning Random Forest Model, the team found that states with the highest overdose death rates tend to have fewer clinically appropriate treatment services available.
"It would have been virtually impossible to analyze so many treatment records without AI/ML assistance," Becker said.
Based on the findings, Becker recommends that state governments prioritize behavioral health services and work collaboratively to expand access to longer-duration, clinically appropriate treatment programs. Increasing availability—especially in states with limited treatment infrastructure—could significantly improve recovery outcomes nationwide.
Becker, who recently received a pilot project award from PIKO (Center for Pacific Innovations, Knowledge and Opportunities), plans to build on the research by examining local data on addiction treatment and recovery among Native Hawaiians and Pacific Islanders.
The post Using AI to identify key factors in substance use recovery first appeared on University of Hawaiʻi System News .
