Machine-learning algorithm led by UF researcher improves the prediction of opioid overdose by integrating health claims data with public human services data

In a study recently published by PLOS One, researchers at the University of Florida College of Pharmacy reported on a machine-learning approach for predicting opioid overdose risk that integrates public human services data with health claims.  In response to the growing epidemic of opioid use in the United States, health systems have implemented surveillance methods using simple criteria to identify those at high-risk of overdose and opioid use disorder.  To improve the surveillance methods, Weihsuan “Jenny” Lo-Ciganic, Ph.D., M.S., M.S.Pharm., an assistant professor of pharmaceutical outcomes and policy worked with researchers at the University of Pittsburgh and Pennsylvania’s Allegheny County Department of Human Services.  The team linked Medicaid health claims data with human services and criminal justice data.  This allowed them to account for important social determinants of opioid overdose, such as social services use and incarcerations.

Jenny Lo Ciganic Headshot

Lo-Ciganic’s study found that of the top 30 most important predictors, nine were human services and criminal justice variables. Of the individuals with overdoses, approximately 70% were members of the top risk decile in the team’s model.  Ultimately, Lo-Ciganic’s algorithms can be used to more accurately identify those at highest risk of opioid overdose allowing for timely and better allocation of treatment resources and preventative care.