A Mobile Unit Cut Deaths by 71%. It Didn't Get More People on Treatment.

Four people died in one arm of a federally funded trial. Thirteen died in the other. The difference held up statistically, produced a hazard ratio of 0.29 and gave the study its headline: a 71 percent reduction in all-cause mortality among people who inject drugs, achieved by parking a medical unit where those patients already were.

The trial was not designed to measure that. It was designed to answer a narrower and more bureaucratic question — whether delivering HIV prevention and addiction medication on a van, with a peer navigator attached, gets more people onto those medications than referring them to existing clinics. On that question, the results published November 4 in JAMA Psychiatry are a flat no.

Both findings matter for New York, which was one of the five cities enrolled. One is a reason to fund mobile units. The other is a warning that proximity is not the same thing as treatment.

What the Trial Was Built to Prove

HPTN 094, also called INTEGRA, screened 842 people who inject drugs and randomized 409 of them between May 2021 and September 2024 in Houston, Los Angeles, New York, Philadelphia and Washington, D.C. To enroll, a person had to be 18 or older, HIV-negative, living with opioid use disorder, not currently on medication for it, and willing to consider medication. That last clause matters — this was not a sample that had already refused treatment.

The intervention arm, 203 people, got 26 weeks of manualized peer navigation plus integrated PrEP, medication for opioid use disorder, and limited clinical and supportive services delivered directly on a mobile unit. The comparison arm, 206 people, got 26 weeks of the same peer navigation, pointed at the community-based PrEP, medication, harm reduction and support services that already existed. Researchers followed both groups to 52 weeks.

This is an unusually honest design. The control group was not left to fend for itself; it received a human being whose job was to get it into care. Comparing a mobile unit against a good navigator is a much harder test than comparing it against nothing, and it is closer to how a health system would actually spend the money.

The Numbers That Didn't Move

The co-primary outcomes were confirmed PrEP use and confirmed medication for opioid use disorder at 26 weeks. Neither separated the arms.

Outcome	Mobile unit (n=203)	Navigation only (n=206)
PrEP use, 26 weeks	5.4%	2.9%
Opioid use disorder medication, 26 weeks	14.8%	10.7%
PrEP use, 52 weeks	2.0%	1.9%
Opioid use disorder medication, 52 weeks	13.3%	14.1%

By 52 weeks the medication difference had not merely narrowed, it had reversed by eight-tenths of a point. Nothing in the table clears the bar the study set for itself.

The authors' conclusion is blunt: mobile delivery of integrated HIV prevention, addiction and navigation services "did not improve primary outcomes beyond peer navigation alone." A press release from the NIH-funded network put the same result in softer language, noting no difference between groups in either PrEP or medication use while emphasizing the mortality signal.

Reading 14 percent honestly

Fourteen percent is easy to dismiss and easy to oversell. The investigators argue it is clinically meaningful because none of the 409 participants were on medication when the study began, so every one of those 30 people in the intervention arm represents a genuine start rather than a continuation. That is true and worth saying.

The other way to read it is that roughly six of every seven participants — people who had already said they were willing to consider medication, and who had a navigator and in some cases a clinic on wheels outside — were still not taking it a year later. Whatever keeps that number where it is, it is not mostly distance, parking or clinic hours.

Then There Were the Deaths

The mortality result was a secondary outcome, which is the technical way of saying the trial was not built to prove it. Four deaths in the mobile arm against 13 in the comparison arm, hazard ratio 0.29 with a 95 percent confidence interval running from 0.1 to 0.9, P = .03.

That interval is wide because the numbers are small: 17 deaths across 409 people. A secondary endpoint with 17 events can produce a real signal or an accident, and the authors say so plainly, writing that the lower death rate "bears replication." Anyone quoting 71 percent as settled science is quoting past the paper.

What makes the signal interesting anyway is that it is all-cause mortality, not overdose mortality. A mobile unit that reduced deaths only by reversing overdoses would show up as an overdose effect. A reduction this broad hints at something less specific — more contact with clinicians, faster attention to infections and wounds, transport to appointments, someone noticing when a patient's health is sliding. None of that is proven here. All of it is plausible, and all of it is the kind of thing that gets stripped out of a program when a grant runs out and the van stops coming.

Why New York Was in This Trial

The New York site was one of five, and the study's co-chair was Nabila El-Bassel, the Willma and Albert Musher Professor of Social Work at Columbia University. Her explanation for why the primary outcomes failed is less about medicine than about logistics.

"People who inject drugs often face fragmented systems, unstable housing, stigma, and other barriers that make sustained care difficult," El-Bassel said in the announcement. She described the findings as a case for models that "meet people where they are" and connect HIV prevention, addiction treatment and supportive care.

New York is, by the numbers, a good place to test that claim. In the 12 months ending June 2026, the CDC's provisional count was 1,777 overdose deaths in New York City and 1,693 in the rest of the state — about 3,470 statewide, down from roughly 3,885 a year earlier and less than half the 2022 peak, when the combined 12-month figure passed 6,500. Those city figures come with an asterisk: roughly 40 percent of records in the latest window were still pending investigation, so the true toll is higher than the reported one.

Access has expanded alongside that decline. OASAS Commissioner Chinazo Cunningham said during the agency's statewide tour this fall that there are now more than 130 methadone treatment locations in New York, that access to methadone for people in rural upstate areas has risen about 50 percent, and that medication is available in all of the state's jails and prisons. Those are real accomplishments, and they describe a state that has largely solved the question of whether medication exists nearby.

HPTN 094 suggests that was the easier question.

PrEP Is the Harder Sell

The PrEP numbers are worse than the medication numbers and less discussed. Confirmed use at 26 weeks was 5.4 percent in the mobile arm and 2.9 percent in the comparison arm; by 52 weeks it had fallen to about 2 percent in both groups. One HIV acquisition occurred in the intervention arm and two in the comparison arm, an incidence rate of 0.9 per 100 person-years — too few events to draw a conclusion, but not a reassuring direction of travel.

The investigators read this as a problem of product, not just of delivery. A daily pill to prevent a virus you do not have competes badly against the immediate concerns of someone without stable housing, and the paper points toward long-acting injectable PrEP as the next thing worth testing in this population. That is a reasonable bet with an uncomfortable implication: the intervention that produced the mortality signal may have worked through contact and attention rather than through the medicines it was carrying.

What a Replication Would Have to Answer

The authors call for replication, and the design of that study is not obvious. Three questions stand between this paper and a policy.

The first is dose. Twenty-six weeks of a van and a navigator is a short exposure to a long condition, and the 52-week medication numbers suggest the advantage of any intensive model decays quickly once contact stops. Whether continuous, low-intensity access beats a six-month burst is an empirical question nobody has answered.

The second is cost and staffing. Peer navigation was present in both arms, and the control arm's results were good enough to erase the mobile unit's advantage. That makes the marginal value of the vehicle itself, rather than the human being inside it, the thing a replication has to isolate — and the thing a state deciding next year's budget will actually want to know.

The third is the outcome everyone now cares about. Mortality was measured here almost by accident. A trial powered for deaths among people who inject drugs would need to be far larger, or far longer, or both, and would raise hard questions about what to do for the control group while it runs.

Where This Leaves a Patient in New York

For someone trying to get into treatment this week, none of this changes the practical map. Medication for opioid use disorder is available through OASAS-licensed programs across the state, in jails, and through a growing number of telehealth and mobile arrangements; SAMHSA's treatment locator and OASAS both list options by county.

What the trial changes is the expectation attached to those options. New York has spent a decade buying access — more sites, more prescribers, more vans — and the return on that spending shows up in the overdose curve. It does not automatically show up in the number of people who stay on medication. Nearly 85 percent of a group that had already agreed to consider treatment was not taking it a year later, in a study that handed them a navigator.

The 71 percent is the part that will be cited in funding requests. The 14.8 percent is the part that should shape them.