Hospital Algorithm Flags Thousands, Yet Only a Fraction Start Buprenorphine

Hospitals are often the first place a person with opioid use disorder receives meaningful medical attention, yet they are also among the most frequently missed opportunities to start lifesaving addiction treatment. A new brief report published in Addiction Science & Clinical Practice offers an unusually candid look at just how difficult it is to convert that opportunity into actual treatment. Researchers led by Abigail Haber and Aaron D. Fox of Albert Einstein College of Medicine and Montefiore Medical Center in the Bronx examined screening data from a randomized controlled trial of in-hospital buprenorphine initiation strategies, and the numbers they report reveal a sobering funnel in which more than nine thousand patient records were reviewed to enroll just twenty-three people.

The study, part of a clinical trial registered on ClinicalTrials.gov as NCT 05118204, was designed to test strategies for starting buprenorphine during medical hospitalization. Buprenorphine is a partial opioid agonist that is considered one of the most effective medications for treating opioid use disorder, and it is also used for chronic pain management. Clinical guidelines increasingly encourage hospitals to initiate the medication before discharge, because patients who leave the hospital without treatment face elevated risks of overdose and readmission. Yet in practice, identifying which hospitalized patients might benefit from buprenorphine remains a major operational challenge, one that the Bronx team set out to address systematically.

Recruitment for the trial took place between October 2022 and October 2024 at two urban teaching hospitals. The research team built an automated electronic health record algorithm that generated a daily list of patients with likely opioid misuse or opioid use disorder. The idea was straightforward: rather than relying on busy clinicians to notice which patients might be candidates for addiction treatment, the algorithm would continuously scan hospital records and surface potentially eligible individuals. This approach reflects a broader trend in implementation science, in which health systems attempt to use their existing data infrastructure to close well-documented gaps between evidence and practice.

The scale of the screening effort was enormous. Over the two-year recruitment window, the algorithm produced 9,140 records for screening. Of those, 8,534 records, or 93 percent, were excluded during preliminary manual review of the electronic health record. In other words, for every record that advanced beyond the first stage of screening, more than thirteen were set aside. Staff members conducted this manual review to confirm whether the algorithm’s flags were accurate, a labor-intensive process that the authors acknowledge as a central limitation of the approach. The algorithm successfully identified patients who were potentially eligible for buprenorphine, but its precision was low enough that substantial human effort was required to separate true candidates from false positives.

From that initial pool of 9,140 records, only 434 patients were deemed potentially eligible for buprenorphine treatment. That figure alone illustrates how narrow the funnel becomes at each stage. But the steepest drop came next: of those 434 potentially eligible patients, only 43 agreed to be assessed for the trial. The remaining patients either declined assessment or could not be evaluated, and when patients were not assessed, research assistants documented the reasons using pre-specified categories. This attrition at the consent stage is particularly significant, because it suggests that the barriers to buprenorphine initiation are not solely technical or logistical. They are also deeply human, rooted in patient concerns and the stigma that continues to surround both addiction and its treatment.

Among the 43 patients who agreed to assessment, 29 were found to be eligible for the trial, and 23 ultimately enrolled and started buprenorphine. Taken together, that represents a 5 percent initiation rate among the pool of potentially eligible patients. The authors frame this outcome as evidence that while automated surveillance can identify candidates, the pathway from identification to treatment is fraught with attrition. Each step of the process, from algorithmic flagging to manual chart review to patient outreach to eligibility assessment to enrollment, loses a substantial proportion of the people who might benefit from the medication. The final yield of 23 patients across two years of screening underscores how much work remains to make hospital-based buprenorphine initiation routine.

The report’s conclusions point in two directions at once. On the technical side, the authors suggest that improving the precision of the screening algorithm, for example through artificial intelligence applications, could better identify patients who are genuinely eligible for buprenorphine. Machine learning models trained on richer clinical data might reduce the 93 percent exclusion rate at the manual review stage, freeing research and clinical staff to focus their outreach on patients who are truly candidates. The study’s own keyword list includes artificial intelligence, and the authors clearly view computational refinement as a promising avenue for future work. Better algorithms would not eliminate the need for human judgment, but they could dramatically reduce the burden of reviewing thousands of records to find a few hundred relevant ones.

On the human side, the authors are equally direct: patient concerns and potential stigma regarding buprenorphine present ongoing obstacles to uptake. Even when the system successfully identified a potentially eligible patient, fewer than one in ten agreed to be assessed. This finding resonates with a large body of addiction research showing that misconceptions about medication-assisted treatment, fear of being labeled as having a substance use disorder in the medical record, and ambivalence about starting buprenorphine during an acute hospitalization all suppress engagement. The study’s design, which included documenting pre-specified reasons when patients were not assessed, reflects an effort to characterize these barriers systematically rather than simply counting losses.

The trial was funded by the National Institutes of Health under award RM1DA055437 and was approved by the Einstein Institutional Review Board under protocol 2021-13311. Informed consent was obtained from all individuals who participated in the clinical trial, and the screening data reported here include individuals who were assessed for eligibility but did not enroll. The authors declare no competing interests, and the article is published open access under a Creative Commons Attribution 4.0 license, making the full screening dataset description available to any research team attempting similar surveillance-based recruitment. The corresponding author is Aaron D. Fox, and the author team includes collaborators from the Peer Network of New York and Kelly S. Ramsey Consulting, bringing peer and clinical expertise to the research.

For hospitals and health systems grappling with the overdose crisis, the report offers both a caution and a roadmap. The caution is that simply deploying an automated screening tool will not, by itself, translate into widespread buprenorphine initiation; the Bronx experience shows that even a well-resourced trial with dedicated staff achieved a 5 percent initiation rate among potentially eligible patients. The roadmap lies in the study’s dual diagnosis of the problem: algorithms need to be more precise, and the clinical encounter needs to be better designed to address patient concerns and reduce stigma. As health systems increasingly turn to electronic health record surveillance and artificial intelligence to identify candidates for addiction treatment, this brief report provides a rare, transparent accounting of what such efforts actually look like on the ground, complete with the attrition that most publications leave unreported. The gap between 9,140 flagged records and 23 treatment initiations is precisely the kind of number that should shape the next generation of hospital-based addiction care.

Subject of Research: Screening and recruitment for in-hospital buprenorphine initiation in patients with opioid use disorder

Article Title: Brief report on screening data from a randomized controlled trial of in-hospital buprenorphine initiation strategies

Article References: Haber, A., Sanchez-Fat, G., Buonora, M., Sabado, A., Ghiroli, M., Deng, Y., Khalid, L., Torres-Lockhart, K. E., Reyes, M., Ramsey, K. S., Starrels, J. L., & Fox, A. D. (2026). Brief report on screening data from a randomized controlled trial of in-hospital buprenorphine initiation strategies. Addiction Science & Clinical Practice. https://doi.org/10.1186/s13722-026-00731-w

Image Credits: AI Generated

DOI: 10.1186/s13722-026-00731-w

Keywords: buprenorphine, opioid use disorder, hospitalization, electronic health records, screening, randomized controlled trial, addiction treatment, stigma, artificial intelligence, chronic pain, implementation science, clinical trial recruitment