Artificial intelligence enhances hospital screening for opioid use disorder, reducing readmissions and costs

Opioid use disorder (OUD) continues to pose a major public health challenge in the United States. Individuals living with OUD who are admitted to hospital face an elevated risk of overdose, repeated hospitalisation, and a host of related complications. Research has shown that providing addiction treatment during hospital stays can improve health outcomes and reduce mortality. Yet, despite its potential benefits, screening for OUD within hospitals remains inconsistent. Many individuals at high risk are discharged before having the opportunity to see an addiction specialist—a factor strongly associated with increased rates of overdose following discharge.

In an effort to address these shortcomings, a research team led by Dr Majid Afshar at the University of Wisconsin–Madison, with support from the National Institutes of Health (NIH), explored whether artificial intelligence (AI) could play a meaningful role. The team had previously developed an AI-driven screening tool designed to identify hospitalised individuals at risk of OUD. This tool analyses data from electronic health records (EHRs) to detect patterns that may suggest the presence of OUD. When such patterns are identified, the system notifies healthcare providers and may recommend a referral to an addiction medicine specialist, alongside other appropriate interventions.

In their latest study, published in Nature Medicine on 3 April 2025, the researchers assessed whether the AI tool could increase the likelihood that individuals at risk would be seen by an addiction specialist during their hospital stay. The study also examined whether use of the AI system had any effect on hospital readmission rates within 30 days of discharge.

The study encompassed over 51,000 adults admitted to the University of Wisconsin Hospital. It was conducted in two phases. The initial, or baseline, phase took place between March and October in both 2021 and 2022, involving roughly 34,000 patients. During this period, data were collected on how clinicians assessed patients’ risk of OUD and the outcomes of those assessments, without the use of AI tools.

The second phase, from March to October 2023, focused on evaluating the AI screener in practice. During this period, the AI tool was employed to assist in evaluating over 17,000 patients. In total, 727 consultations with addiction medicine specialists were conducted during the study’s entire duration.

Findings indicated that the AI tool was just as effective as clinician-led assessments in prompting referrals to addiction specialists. Notably, individuals assessed with the help of the AI screener were 47% less likely to be readmitted to hospital within 30 days of discharge compared to those assessed without it.

The study also demonstrated potential cost savings. Each avoided readmission was estimated to reduce healthcare expenditure by approximately $6,800. These results suggest that integrating AI into hospital workflows could increase access to addiction support, enhance clinical efficiency, and lead to substantial cost savings for healthcare systems.

“AI holds promise in medical settings, but many AI-based screening models have remained in the development phase, without integration into real-world settings,” said Dr Afshar. “Our study represents one of the first demonstrations of an AI screening tool embedded into addiction medicine and hospital workflows, highlighting the pragmatism and real-world promise of this approach.”

As hospitals grapple with the ongoing burden of opioid use disorder, this research provides compelling evidence for the adoption of AI-driven screening tools. By identifying at-risk individuals earlier and linking them to appropriate care, such technology may not only improve outcomes for individuals living with OUD but also help health systems operate more efficiently and equitably.