New AI tool will predict patients at high risk for opioid use disorder and overdose

AI tool

New AI tool will predict patients at high risk for opioid use disorder and overdose

 

Newswise — University of Florida researchers are developing a new artificial intelligence tool that will help clinicians identify patients at high risk for opioid use disorder and overdose.

The tool will use data from patients’ electronic medical records to guide clinicians in safely and effectively prescribing opioid medications. The project is supported by a five-year, $3.2 million grant from the National Institute on Drug Abuse, or NIDA, and aims to reduce the unprecedented rise in opioid overdose and opioid use disorder in the United States.

“In 2019, almost 12 million Americans reported misuse of prescription opioids, and this public health crisis is placing a substantial burden on individuals, families, society and our health care system,” said Wei-Hsuan “Jenny” Lo-Ciganic, Ph.D., M.S., M.S.Pharm., an associate professor of pharmaceutical outcomes and policy in the UF College of Pharmacy and the principal investigator of the NIDA grant. “If we can more accurately identify patients who are at a high risk for opioid use disorder and overdose, then we can better allocate resources and provide timely and targeted interventions.”

For UF researchers, identifying high-risk patients begins by leveraging ongoing NIDA-funded work and an analysis of health care claims data using a type of AI called machine-learning. This NIDA grant will use electronic health records or integrated health care data to help clinicians identify patients most susceptible to opioid use disorder and overdose. The data analysis requires advanced AI technology, which UF provides researchers through its HiPerGator AI supercomputer.

“Machine-learning is an innovative analytic technique that handles complex interactions in large data, discovers hidden patterns through modeling and generates more accurate prediction algorithms in real-time that are often superior to traditional statistical techniques,” Lo-Ciganic said. “Although AI techniques are widely used in activities from fraud detection to genomic studies, this is one of the first examples where machine learning will be applied and implemented in a clinical setting to address the nation’s opioid epidemic.”

Lo-Ciganic estimates the new algorithm will accurately identify between 70-90% of high-risk patients. The algorithm will exclude the large majority of prescription opioid users with negligible opioid use disorder or overdose risk while evaluating the benefits and risk tradeoffs of prescription opioid use for high-risk patients.

The second part of the NIDA grant involves designing and developing a clinical decision support tool that integrates AI-based risk scores to warn clinicians about high-risk patients. The tool will be integrated into patients’ electronic health records and provide clinicians with early warnings and risk mitigation strategies. UF researchers, in collaboration with primary care providers and information technology experts at UF Health, will create the dashboard that lives in the electronic health record and provides real-time information to clinicians.

“We want the tool to help clinicians, but the ultimate goal is to improve patient outcomes and care,” Lo-Ciganic said. “Usually, patients need opioids because they are in pain. If our platform can better inform strategies around pain management and not increase the risk of a bad outcome, like overdose or addiction, then we have accomplished our goal.”

When fully developed, the new clinical decision support tool will be piloted at three UF Health primary care clinics. UF researchers will study its usability, accessibility and feasibility in deciding whether the tool warrants broader use.

The research team supporting the NIDA grant includes faculty from the UF colleges of Pharmacy and Medicine, as well as collaborators from the University of Pittsburgh, the University of Utah, the University of Arizona and Carnegie Mellon University.



Publicación más antigua Publicación más reciente