Twelve weeks of vouchers or six: what STAR-OM settled in fifteen Vietnamese methadone clinics

Methamphetamine arriving on top of opioid agonist treatment is where this field makes both of its mistakes. One is to read a stimulant-positive urine as proof that methadone has failed and to taper the patient out of the programme. The other is to leave the stimulant alone because no medication is approved for it, and to treat the clinic's work as finished once the opioid is covered. Vietnam has the problem at scale: methamphetamine co-use among methadone patients there has been associated with missed methadone doses and with returning to heroin. STAR-OM, run by Hanoi Medical University with the University of California, Los Angeles and funded by the US National Institute on Drug Abuse, asked what the clinic should do next, in a design that lets the answer depend on how the patient responds.

What the protocol consists of

Everything happened inside the methadone clinic and was delivered by the clinic's own trained providers. Every participant received two motivational interviewing sessions, one before stage one at Week 0 and one at the midpoint, Week 13. The first randomization split the frontline stage. The low-intensity arm received six weeks of contingency management, with escalating-value vouchers for consecutive methamphetamine-negative urine samples and a reset to base value after any positive sample, up to $40 in total, then six weeks of group education across six topics: addiction mechanism, road to recovery, coping with triggers, boredom, building trust and relapse prevention. The high-intensity arm received 12 weeks of contingency management with the same escalating structure, up to $150 in total, which the authors describe as comparable to half the average monthly income and about three times the monthly cost of methadone treatment. Vouchers were redeemable only for methadone treatment fees or for personal goods such as shampoo, body wash, tea, coffee and children's toys.

Urine was screened twice weekly on randomly selected days for 25 weeks, and it is the screen that drives the algorithm. A treatment response was defined strictly: four methamphetamine-negative screens out of a possible four in Weeks 11 to 12. Responders moved to 12 weeks of unidirectional scripted text messages, twice daily. Non-responders, meaning anyone with at least one positive screen in that window, were re-randomized 1:1 to either the Matrix model alone, 12 weekly counselling sessions shortened from the original 24 with the 12-step content and family education dropped as culturally unsuited or infeasible, or the Matrix model plus contingency management on the same $150 structure. The whole thing is reproducible in a clinic that already runs methadone: what it needs is twice-weekly urine testing, a voucher budget, a counsellor and a phone number.

The word in the trial's name, and the outcome on the table

STAR-OM stands for Screen, Treat and Retain people with opioid use disorders who use methamphetamine in methadone clinics. Retention in methadone treatment is neither among the outcomes reported nor among the sixteen registered on ClinicalTrials.gov, six primary and ten secondary, and the registry's other-outcomes field is empty. The retention figures in the paper, 96.7% and 90.1%, are retention in the trial. The reason methamphetamine co-use is a clinical problem at all, in the authors' own introduction, is missed methadone doses, return to heroin and overdose risk, and heroin-negative urine and frequency of opioid overdose were both registered on ClinicalTrials.gov as secondary outcomes. Neither appears in this report. The paper says so plainly in its statistical analysis section and promises them, along with the four registered HIV primary outcomes, in forthcoming publications. That is honest reporting of a surrogate, not concealment of one, and it is also the reason the result cannot yet be carried where a clinician would most want to carry it.

One further mismatch is worth naming. The registry's operational definition of the methamphetamine primary outcome is the four-of-four treatment response at Weeks 11 to 12 and 24 to 25. In the paper that measure appears under secondary outcomes, and the primary analysis is the twice-weekly negative screen modelled over time. The paper explains the difference between the numbers as a reporting convention: the registry requires raw counts and unadjusted percentages, whereas the published percentages are marginalized from mixed-effects models. It does not explain, or anywhere mention, the move of the registered primary measure into the paper's secondary outcomes.

What held, what did not, and where the failure sits

What the authors call the main result is that consistent exposure to contingency management predicted greater reductions in methamphetamine use, that text messaging supported maintenance in early responders, and that combining Matrix with contingency management improved outcomes in those who did not initially respond. Their summary box states that 12 weeks of contingency management produced "more rapid and sustained" reductions than six weeks. Table 2 supports the first half and not the second: the two arms' gains from Week 0 to Week 25 differed by 1.7 percentage points, 95% CI -1.5 to 4.9, p = 0.286, leaving the Week 25 levels 3.7 points apart, 77.2% against 73.5%, and the discussion itself says only that the high-intensity arm maintained lower use through Week 12. I followed the table.

The re-randomization is the cleanest positive finding in the trial, and it applies to the patients who had already failed 12 weeks of treatment. Matrix plus contingency management rose from 54.2% (42.9% to 65.4%) negative screens at Week 14 to 63.2% (52.8% to 73.6%) at Week 25, a gain of 9.1 percentage points (5.8 to 12.3), p < 0.001; Matrix alone did not change, -1.0 points (-4.3 to 2.3), p = 0.553. The abstract's headline of 10.1 percentage points is the difference between those two within-arm changes. On the registered four-of-four measure the effect is 33.3% (19.6% to 47.1%) against 15.9% (6.0% to 25.8%), a difference of 17.4 points (5.1 to 29.7), p = 0.005, and the registry gives the same comparison in counts, 28 of 107 against 13 of 108, with an odds ratio of 2.98 (95% CI 1.37 to 6.47), p = 0.006, from its own logistic regression rather than a recalculation of the paper's percentages. The frontline comparison at Weeks 24 to 25 is posted as odds ratio 1.20 with a 95% CI of 0.84 to 1.71, alongside a p value of 0.001 that cannot belong to that interval: an interval that wide implies p near 0.32, Supplemental Table S2 gives p = 0.313 for the same comparison, and the posted 0.001 appears to have been duplicated from the Weeks 11 to 12 row above it. The registry results were posted on 23 January 2026. The maintenance stage held on the authors' own headline measure: among the 428 responders reassigned to text messaging, methamphetamine-negative screens stood at 91.8% (88.6% to 94.9%) at Week 14 and 88.5% (84.8% to 92.2%) at Week 25, a fall of 3.3 percentage points (-4.9 to -1.6), p < 0.001, and 75.2% (66.1% to 84.2%) still met the four-of-four criterion at Weeks 24 to 25, although that stage had no comparison group and every responder was sent to it. Self-reported methamphetamine use fell in every group: by 24.8 percentage points in the low-intensity arm and 27.5 in the high-intensity arm by Week 13, and by 11.7 in the Matrix arm and 14.8 in the Matrix plus contingency management arm by Week 26.

Two things the paper does not report follow from its own Table 1. Of the 643 patients who reached stage two, 352 attended Hanoi clinics and 291 attended Ho Chi Minh City clinics, and of the 428 responders 291 were from Hanoi. That puts the response rate at 82.7% in Hanoi against 47.1% in Ho Chi Minh City, which pulls the two stage-two subsamples towards opposite cities: the re-randomized comparison is 154 of its 215 patients from Ho Chi Minh City, while the text-messaging group is 291 of its 428 from Hanoi. Clinic was carried as a random intercept in every model. Separately, Table 1 records 252 of 665 baseline screens as negative, 37.9%, while the Week 0 figures of 58.4% and 60.4% are marginalized out of the regression rather than counted; the paper does not bring the two quantities together.

What changes in the clinic

Harms were counted independently of anyone's opinion and belong in the record: 13 incarcerations, 11 compulsory drug rehabilitation placements and seven hospitalisations as non-fatal serious adverse events, two of the seven for psychosis, which was itself an exclusion criterion at entry, and six deaths, three in each stage, from kidney failure, asthma, non-HIV pneumonia, bowel obstruction, pleural effusion and one suicide described as possibly methamphetamine-related. No adverse event was attributed to participation.

What a clinician can act on is narrower than the abstract and more useful than nothing. Twelve weeks of vouchers get patients to a stimulant-free urine faster than six, and by six months that head start has gone; if the budget is fixed, there is a case for spending it early and no clear case for spending twice as much. For the roughly one patient in three who is still testing positive after 12 weeks, adding contingency management to a manualized counselling programme produced a real, randomized, registry-confirmed gain, while the counselling alone produced none. And the intervention was delivered by ordinary methadone providers in ordinary clinics, with vouchers that could be spent on the methadone fee itself. What nobody should say yet is that any of this kept a patient in methadone treatment or prevented an overdose. Overdose frequency was registered as a secondary outcome and is promised in a later paper; retention in methadone treatment was never registered at all.