An artificial intelligence-driven screening tool developed by a research team at the University of Wisconsin School of Medicine and Public Health successfully identified hospitalized adults at risk for opioid use disorder and recommended referral to inpatient addiction specialists.
The AI-based method was just as effective as a health provider-only approach in initiating addiction specialist consultations and recommending monitoring of opioid withdrawal. Compared to patients who received provider-initiated consultations, patients identified for addiction medicine referrals by AI screening and who received consultations had 47% lower odds of being readmitted to the hospital within 30 days after their initial discharge. This reduction in readmissions translated to a total of nearly $109,000 in estimated health care savings during the study period.
The study, which was recently published in Nature Medicine, reports the results of a completed National Institutes of Health-funded clinical trial, demonstrating AI’s potential to affect patient outcomes in real-world health care settings. The study suggests investment in AI may be a promising strategy for health care systems seeking to increase access to addiction treatment while improving efficiencies and saving costs, according to Dr. Majid Afshar, lead author and principal investigator of the study and associate professor of medicine at the UW School of Medicine and Public Health.
“AI holds promise in medical settings, but many AI-based screening models have remained in the development phase, without integration into real-world settings,” he said. “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.”
The AI screener was built to recognize patterns in data, similar to how our brains process visual information. It analyzed information within all the documentation available in electronic health records in real time, such as clinical notes and medical history, to identify features and patterns associated with opioid use disorder. Upon recognition, the system issued repeated alerts to providers when they opened the patient’s medical chart with a recommendation to order addiction medicine consultation and to monitor and treat withdrawal symptoms.
When the AI screener detects a patient at risk of opioid misuse, a Best Practice Alert (BPA) recommends follow-up, including an evaluation using the Clinical Opiate Withdrawal Scale (COWS). Health care providers can dismiss the alert if it is not appropriate.
To analyze effectiveness, the research team compared data from provider-led addiction specialist consultations in two time periods, from March to October in 2021 and 2022, to the performance of their AI screening tool when it was deployed, from March to October 2023. The tool had been developed and validated in prior work.
From start to finish, the trial involved screening 51,760 adult hospitalizations, with two-thirds, about 66% of the screenings occurring during the pre-intervention period without the AI screener being deployed and about one-third, or around 34%, occurring when the AI screener was deployed hospital-wide. A total of 727 addiction medicine consultations were completed during the study period.
The trial found that AI-prompted consultation was equally effective to provider-initiated consultation, ensuring no decrease in quality while offering a more scalable and automated approach. Specifically, the study showed that 1.51% of hospitalized adults received an addiction medicine consultation when health care professionals used the AI screening tool, compared to 1.35% without the assistance of the AI tool. The AI screener was also associated with fewer 30-day readmissions, with approximately 8% of hospitalized adults in the AI screening group being readmitted to the hospital, compared to 14% in the traditional provider-led group.
The reduction in 30-day readmissions still held after accounting for patients’ age, sex, race and ethnicity, insurance status and comorbidities, as calculated via an odds ratio. When analyzing the results using the odds ratio, the researchers estimated a decrease of 16 readmissions by employing the AI screener.
AI has the potential to strengthen implementation of addiction treatment while optimizing hospital workflow and reducing health care costs.
Nora D. Volkow, MD
A subsequent cost-effectiveness analysis indicated a net cost of $6,801 per readmission avoided for the patient, health care insurer and the hospital. This amounted to an estimated total of $108,800 in health care savings for the eight-month study period in which the AI screener was used, even after accounting for the costs of maintaining the AI software. Nationally, the average cost of a 30-day hospital readmission is currently estimated at $16,300.
“Addiction care remains heavily underprioritized and can be easily overlooked, especially in overwhelmed hospital settings where it can be challenging to incorporate resource-intensive procedures such as screening,” said Dr. Nora D. Volkow, director of NIH’s National Institute on Drug Abuse. “AI has the potential to strengthen implementation of addiction treatment while optimizing hospital workflow and reducing health care costs.”
While the AI screener showed strong effectiveness, challenges remain, including potential alert fatigue among providers and the need for broader validation across different health care systems, according to the study.
The authors also note that while the various study periods – spanning multiple years – were seasonally matched, the evolving nature of the opioid crisis may have introduced residual biases. Future research will focus on optimizing the AI tool’s integration and assessing its longer-term impact on patient outcomes.
The opioid crisis continues to strain health care systems in the U.S., with emergency department admissions for substance use increasing by nearly 6% between 2022 to 2023 to an estimated 7.6 million. Opioid use was second only to alcohol as a leading cause of emergency department visits between 2018 and 2021, but screening for opioid use disorder in hospitals remains inconsistent. As a result, hospitalized patients with opioid use disorder frequently leave the hospital before seeing an addiction specialist, a factor linked to a tenfold increase in overdose rates.
AI technology has emerged as a novel, scalable tool to potentially overcome these barriers and improve opportunities for early intervention and linkage to medications for opioid use disorder, but more research is needed to understand how AI can be used effectively in health care settings, according to Afshar.
“This study lays the groundwork for broader adoption of electronic-health-record-embedded AI screeners for opioid use disorder,” he said.
This work was funded by the National Institute on Drug Abuse of the National Institutes of Health grants R01-DA051464, R01-LM012973, R01-HL157262 and UL1TR002373.