Bronx Hospitals Screened 9,140 Patients for Addiction Care. 23 Started Treatment.
A study from Albert Einstein College of Medicine and Montefiore tracked how 9,140 Bronx hospital records produced just 23 patients started on buprenorphine.
Two Bronx teaching hospitals spent two years trying to answer a practical question: of all the people lying in hospital beds who could benefit from addiction medication, how do you actually find them? Their answer is published, and it is uncomfortable. An automated system flagged 9,140 patient records. Twenty-three people started buprenorphine.
The numbers come from a brief report in Addiction Science & Clinical Practice published on 8 October, led by Abigail Haber with Aaron D. Fox as senior author. The team works out of Albert Einstein College of Medicine on Morris Park Avenue and Montefiore Medical Center on East 210th Street, both in the Bronx, with co-authors from Peer Network of New York and a consulting group in Olivebridge. The trial itself was registered as NCT 05118204 on ClinicalTrials.gov and funded by a grant from the National Institute on Drug Abuse, RM1 DA055437.
The authors do not frame the result as a failure of medicine. They frame it as a map of where a hospital's addiction care leaks.
The Funnel Behind a Single Number
Screening ran in four stages, and each stage shed most of the people the previous one had found. The arithmetic is the story.
| Stage of screening | Patients |
|---|---|
| Records flagged by the automated algorithm | 9,140 |
| Excluded during manual chart review | 8,534 |
| Judged potentially eligible | 434 |
| Successfully approached in person | 201 |
| Agreed to be assessed | 43 |
| Met trial eligibility | 29 |
| Enrolled and started buprenorphine | 23 |
Read that table from the middle down. The algorithm narrowed 9,140 records to 434 candidates, a reduction of more than 95 percent that the authors attribute mostly to imprecision rather than to clinical judgment. But the steepest single drop came later and had nothing to do with software: of the 201 patients a research assistant actually reached and spoke with, only 43 agreed to even be evaluated.
That is roughly one in five. Of the 434 patients the study considered potentially eligible for buprenorphine — people who were hospitalized, had signs of opioid misuse or opioid use disorder, and were not already on treatment — about 5 percent ended up on the medication before discharge.
Buprenorphine is a partial opioid agonist that treats opioid use disorder and is widely considered one of the most effective tools available for it. National prescriptions for it have plateaued even as opioid use disorder has become more common, a gap the study's authors note plainly in their introduction. A hospital admission is one of the few moments when a person with untreated addiction is already inside the health system, under observation, with a physician assigned to them. The study measures how often that moment converts into treatment. The answer, in two Bronx hospitals, was rarely.
How the Screening System Worked
Recruitment ran from October 2022 through October 2024 at two urban teaching hospitals that share a single Epic electronic health record. Every hospitalized patient the algorithm surfaced was included in the analysis, whether or not they went on to enroll in the trial. The Einstein institutional review board approved the protocol under number 2021-13311.
The algorithm was deliberately broad. It generated a daily list of hospitalized patients with likely opioid misuse by checking five separate triggers, and a patient needed only one of them to appear:
- at least one of twelve drug-related terms in clinical notes, such as "IVDU"
- a urine toxicology screen positive for opioids
- an opioid-related ICD-10 diagnosis code
- an existing prescription for medication to treat opioid use disorder
- any documented treatment or assessment for opioid withdrawal
From there the process turned manual. A research assistant read each flagged chart to determine preliminary eligibility. An addiction specialist reviewed the selected records again to rule out patients who were too unstable to approach, such as someone hospitalized with sepsis, and to confirm that buprenorphine was clinically warranted. Only then did a research assistant approach the patient and the attending physician, first to discuss the trial, and finally to assess eligibility and treatment.
What the Algorithm Got Wrong
Nearly forty-three percent of the 8,534 excluded records were misidentified — the algorithm claimed opioid misuse where a human reviewer found none. Another 21 percent were patients already receiving medication for opioid use disorder before admission, which the trial excluded, and 16 percent hit study-specific exclusions such as documented hypoxia.
The precision problem was not a footnote in the paper. It was the workload. Staff read thousands of charts to find hundreds of candidates, and the authors describe that manual review as a central limitation of using electronic health record surveillance this way. A screening tool that flags the wrong patients does not just waste effort; in a busy hospital it competes for the same limited staff time that would otherwise be spent talking to the right ones.
Why Patients Said No
The consent stage is where the study becomes more than a technical report. Of 434 potentially eligible patients, 201 could be approached. Of those 201, 14 percent declined any addiction service at all, and 48 percent declined buprenorphine specifically — a rejection rate the authors connect to patient concerns and the stigma that still surrounds both addiction and its treatment.
Those numbers describe something specific. A person admitted for a heart problem or a broken leg is asked, in the middle of an acute medical crisis, to accept a diagnosis and a medication they may not have sought. Some worry about what a substance use disorder note in their record will mean for future care, insurance, or employment. Others are simply not ready to consider treatment during the days they are focused on surviving a hospitalization.
The study's design tried to capture this systematically. When a patient was not assessed, research assistants logged the reason using pre-specified categories rather than leaving the loss unexplained, so the attrition could be described rather than merely counted.
The Chronic Pain Complication
This trial was not limited to people with opioid use disorder. Chronic pain was an inclusion criterion, defined as pain present on most or all days for at least three months with at least moderate intensity and interference, measured at 4 or higher on the Pain, Enjoyment of Life and General Activity scale. That definition matched how the hospitals actually saw patients, and it reflects a real clinical problem: buprenorphine is used for chronic pain as well as for addiction, and it carries lower overdose risk than full-agonist opioids.
But it also complicated every conversation. Switching someone from a full-agonist opioid to buprenorphine is a clinical decision with withdrawal risk attached, and it requires an addiction specialist's sign-off. In practice, the study found that more than half of the patients it could reach declined addiction services or buprenorphine itself. The authors point to high illness acuity and the difficulty of approaching patients before discharge as factors working alongside stigma, which is a reminder that a hospital is a hard place to start a chronic-care treatment even when the medication is available and the clinician is willing.
Where the Study Points Next
The authors' own recommendation is to sharpen the tool. They argue that better algorithm precision, potentially with artificial intelligence trained on richer clinical data, could reduce the volume of false positives and free staff to focus outreach on genuine candidates. The paper's keyword list puts artificial intelligence front and center alongside opioid use disorder, chronic pain, and addiction treatment.
That is a reasonable engineering goal, and the study is careful not to oversell it. A more accurate algorithm would fix the misidentification problem and shorten the chart-review burden. It would not fix the part of the funnel where 158 of 201 approached patients said no. Software can find the right patient; it cannot make a frightened person accept a diagnosis on a Tuesday morning when they were admitted for something else entirely.
Why the Bronx Numbers Travel
The study sites are not incidental. Einstein and Montefiore sit in the borough that has carried the heaviest overdose burden in New York City for years, and the research team includes a peer network based in the city, which is the kind of staffing that shapes how patients hear what they are being offered.
The broader state picture makes the finding harder to dismiss. New York reported a 32 percent drop in overdose deaths in 2024, the largest single-year decline the state had recorded, and state officials said the decline continued into 2025. Deaths are falling. Treatment initiation inside hospitals, by this study's measure, is still not keeping up — and the two trends can coexist, because the people dying and the people declining buprenorphine are not necessarily the same population.
New York has the infrastructure to close some of that gap. Buprenorphine is covered by Medicaid, which insures a large share of the patients in this study's population, and the state's Office of Addiction Services and Supports licenses and funds treatment programs across the five boroughs. The state publishes its own overdose death data through the OASAS overdose dashboard, which makes it possible to track whether hospital-based initiation moves at all.
The Hour Before Discharge
There is a reason researchers keep returning to hospitals. Clinical guidelines increasingly favor starting buprenorphine before discharge precisely because patients who leave without treatment face elevated overdose and readmission risk. The window is narrow: a few days of inpatient care, a patient whose attention is on an acute problem, and a discharge order that arrives whether or not anyone resolved the addiction question.
The Bronx report suggests that the barrier is not mainly a shortage of medication or a rule against prescribing it. Both hospitals had addiction specialists available, an institutional review board-approved protocol, and a funded trial designed to make initiation easier. What they did not have was a reliable way to identify the right patients cheaply, or a reliable way to persuade them once identified.
The authors put the next steps in two directions at once, and the pairing is deliberate: better tools for finding candidates, and better interventions for earning their trust. Health systems across New York are moving toward electronic surveillance of their patient populations to spot addiction and offer treatment. This study is a rare transparent accounting of what that looks like before anyone has solved it, down to the gap between 9,140 and 23.
Getting Help in New York
- NYS HOPEline, staffed 24 hours a day: 1-877-8-HOPENY (1-877-846-7369)
- 988 Suicide & Crisis Lifeline, for mental health and substance use crises
- OASAS treatment program search and information: oasas.ny.gov
- Trial record and protocol for the Bronx screening study: NCT 05118204
Written by
MTNYC Editorial TeamThe MTNYC Editorial Team is a group of healthcare writers, researchers, and addiction specialists dedicated to providing accurate, compassionate, and evidence-based information about addiction treatment and recovery resources in New York State.
