Published on 17 September 2026
Transcranial stimulation in opioid use disorder: does hitting a brain target mean treating the disorder?
Molecular Psychiatry · 2026; online publication of 26 June 2026, 13 pages, no volume or pagination assigned as of the date of consultation, full text and supplementary material freely available on the publisher’s site · Soleimani, Kuplicki, Paulus and Ekhtiari
DOI 10.1038/s41380-026-03694-1
PMID 42362768
Scientific 74
Editorial 71
The essentials
Sixty men with opioid use disorder, recruited during early abstinence at a residential centre in Tulsa, Oklahoma, were randomised equally, thirty per arm, to receive a single 20-minute session of 6 Hz transcranial alternating current stimulation, in a high-definition montage over the right frontal and right parietal regions, or sham stimulation. The trial is randomised, described as triple-blind, and its protocol was registered with ClinicalTrials.gov on 9 April 2019 under number NCT03907644, before the recruitment period reported by the authors began. The prespecified primary endpoint is not clinical: it is the change in BOLD signal on functional magnetic resonance imaging during a drug cue-exposure task. This endpoint is met: the time-by-group interaction is significant after family-wise correction at a threshold of 0.05, with reduced cue reactivity in the active arm compared with the sham arm, over a main cluster of 864 voxels covering the bilateral striatum, amygdala, ventral tegmental area and posterior cingulate cortex. Subjective craving, a secondary endpoint measured on a visual analogue scale, does not differ between groups: group effect F(1, 344) = 0.056, p = 0.8133, time-by-group interaction F(5, 344) = 0.386, p = 0.8581. The other prespecified secondary endpoint, resting-state connectivity between cortical and subcortical structures, does not reach significance, which the authors attribute to a lack of power. The positive connectivity results concern connectivity during the task, recorded as an exploratory endpoint: increased engagement of the right frontoparietal network (β = 2.63, p = 0.0308 after FDR correction) and altered coupling with the right ventral striatum (β = 1.99, p = 0.037) and the left medial amygdala (β = 1.97, p = 0.039), which the authors interpret as strengthened top-down inhibitory regulation while stating explicitly that their method allows no inference about directionality. In the active group alone, the simulated electric field intensity correlates with network engagement (r = 0.43, p = 0.02). The last measurement occurs the day after the session. This work demonstrates that a brain target can be hit. It demonstrates no clinical benefit, and the authors do not claim otherwise: their conclusion speaks of a potential intervention.
Context
The reference treatment for opioid use disorder is pharmacological and agonist-based. In France, the national health authority’s guidance sheet on the treatment of opioid use disorder, validated by its board on 10 March 2022, names only two substitution medicines, buprenorphine, alone or combined with naloxone, and methadone, restricts methadone initiation to physicians at specialised addiction centres or hospital physicians, and states that treatment should be considered over the long term, with the optimal duration unknown. No neuromodulation technique appears in it. Other health systems set out comparable agonist-based frameworks, with their own specific rules, and this sheet is cited here as a worked example rather than a universal standard.
The strongest argument for these treatments is not patient comfort, it is survival. The cohort meta-analysis by Sordo and colleagues, published in the BMJ in 2017, pools 19 cohorts, 122,885 patients on methadone and 15,831 on buprenorphine. All-cause mortality is 11.3 per 1,000 person-years during methadone treatment against 36.1 outside it, a rate ratio of 3.20 (95% CI 2.65 to 3.86). For buprenorphine, 4.3 against 9.5, rate ratio 2.20 (95% CI 1.34 to 3.61). For overdose mortality with methadone, 2.6 against 12.7, rate ratio 4.80 (95% CI 2.90 to 7.96). This is observational data, with the usual caveats about confounding by indication, but the convergence is strong, and it is the foundation of the guideline.
This is the necessary starting point for reading the trial analysed here. A stimulation device tested over a single session, with no clinical primary endpoint and no follow-up, is not compared against this foundation and does not substitute for it. The question it asks is different: can the brain circuit thought to underlie craving be modified in humans, in a controlled and measurable way?
The mechanism
What “hitting the target” means
The working hypothesis is stated explicitly in the first sentence of the abstract: theta-band oscillation supports frontoparietal connectivity within the executive control network and the top-down regulation that network exerts over subcortical structures. Applying a 6 Hz alternating current simultaneously to a frontal and a parietal region of the right hemisphere aims to artificially synchronise these two nodes.
The approach belongs to a logic known as target engagement: before testing whether a technique treats a condition, one verifies that it actually reaches what it claims to reach. This step is legitimate and often skipped. It also carries a direct consequence for how the results should be read, which needs to be stated plainly.
| Link in the chain | Status in this trial | Reading |
|---|---|---|
| The current reaches the brain | Documented, with individual electric field modelling | In the active arm alone, the simulated field correlates with network engagement (r = 0.43, p = 0.02), which is an argument of plausibility |
| The brain signal changes | Primary endpoint, met | An intermediate imaging marker, prespecified as such in the registry from April 2019 and never modified since |
| The subjective experience changes | Secondary endpoint, no detectable difference between groups | |
| Opioid use changes | Not measured | Neither consumption, nor relapse, nor treatment retention, nor mortality. This study says nothing about these endpoints |
There is therefore no therapeutic demonstration here, and the phrase “treating opioid use disorder” cannot be attached to this device. What is established is that a session changes a measured brain response within the following hour.
The study at a glance
| Question (PICO) | |
|---|---|
| Population | |
| 74 candidates assessed, 14 excluded or ineligible at screening, 60 men randomised, 30 per arm, all analysed. Mean age 33.75 years (standard deviation 7.04; range 20 to 52 years). Recruitment from June 2019 to June 2023 during the early abstinence period, at a residential addiction treatment centre in Tulsa, Oklahoma. The registry requires a diagnosis made within the preceding twelve months, abstinence of at least three days and less than six months, and excludes schizophrenia, bipolar disorder, active suicidal ideation, a positive urine screen, and contraindications to MRI or stimulation. The publication’s supplementary material, however, states a minimum abstinence of one week: the registry and the publication diverge on this threshold. The same material indicates that the buprenorphine-naloxone combination is the most frequently reported current medication in both arms, without the proportion of participants on agonist treatment being reported as such | |
| Intervention | |
| A single 20-minute session of 6 Hz alternating current stimulation. Two high-definition 4×1 montages centred on F4, right dorsolateral prefrontal cortex, and on P4, right inferior parietal cortex. In-phase stimulation, intensity 2 mA peak to peak, return currents of 0.5 mA in phase opposition on the surrounding electrodes. These parameters appear in the Methods section of the published article | |
| Comparator | |
| Sham stimulation, using the same device | |
| Prespecified primary endpoint | |
| Change in BOLD signal on functional MRI during a drug cue-exposure task, craving-versus-neutral contrast, in a voxel-by-voxel analysis within regions of interest including the prefrontal cortex, insula, striatum, thalamus and extended amygdala, immediately before and after the intervention. Submitted to the registry on 5 April 2019, posted on 9 April 2019, and strictly identical across the six successive registry versions up to the last one, of 11 July 2023. The analysis foregrounded in the publication is, however, a whole-brain analysis, not the region-of-interest analysis as formulated in the registry | |
| Secondary endpoints | |
| In the registry: self-reported craving on a 0 to 100 visual analogue scale, and resting-state connectivity between subcortical and cortical regions. Connectivity during the cue-exposure task, which carries the positive connectivity results, is registered under exploratory endpoints, not secondary endpoints | |
| Design | |
| Randomised parallel-group trial, described as triple-blind in the publication, with the registry describing the masking as quadruple (participants, care providers, investigators, assessors). ClinicalTrials.gov registration NCT03907644, submitted 5 April 2019 and posted 9 April 2019, for a trial start date declared to the registry of 15 March 2019 and recruitment the publication places from June 2019 to June 2023. The stated purpose in the registry changed from “treatment” to “basic science” on 23 April 2019, and the phase from “1 and 2” to “not applicable” on 30 August 2021. Sponsor: Laureate Institute for Brain Research, with the Brain and Behavior Research Foundation as a partner. Six measurement time points, from baseline to the day after the session · CEBM 2b · single-centre RCT |
Quality control
| Criterion | Status |
|---|---|
| Prior registration | Solid |
| Finding NCT03907644 posted on 9 April 2019, recruitment starting in June 2019 according to the publication. The registry’s version history was checked: the primary imaging endpoint was never modified, from the initial submission of 5 April 2019 to the last update of 11 July 2023, and self-reported craving appears as a secondary endpoint in every version. No swapping of the primary endpoint. Worth noting: the trial start date declared to the registry, 15 March 2019, precedes the registration posting by three weeks, which is consistent with registration ahead of the first enrolment only if the authors’ reported recruitment date is used | |
| Hierarchy of endpoints | Caveat |
| Finding The prespecified secondary neural endpoint, resting-state connectivity, gives no significant result. The connectivity results foregrounded instead concern connectivity during the task, which the registry classes among exploratory endpoints and which the publication presents as a secondary endpoint. This slippage is flagged here because it changes the inferential status of these results | |
| Nature of the primary endpoint | Intermediate marker |
| Finding An imaging signal, not a clinical outcome. The choice is stated openly and disclosed, but it sets the ceiling of what this trial can conclude | |
| Blinding integrity | Caveat |
| Finding Blinding success was tested, and tested statistically, which is to the authors’ credit. After the functional MRI, 24 participants in the sham arm and 25 in the active arm believed they had received real stimulation; the next day, 27 and 23. The difference between arms on these guesses is not significant (p = 0.509 then p = 0.464), and the authors judge the blind to be effective overall. The supplementary material, however, contains a result the main text does not carry forward: the next day, reported confidence in this guess is significantly higher in the active arm than in the sham arm (83.11 ± 17.48 against 70.37 ± 24.54; p = 0.028). The blind on the response itself is therefore not broken, but subjective certainty differs between arms. Neither the Bang blinding index nor the James blinding index is reported | |
| Number of sessions | Insufficient for a therapeutic question |
| Finding A single 20-minute session. No neuromodulation protocol aimed at treatment can be conceived on a single exposure | |
| Follow-up duration | Very short |
| Finding Last measurement the day after the session. No data beyond 24 hours | |
| Size and representativeness | Caveat |
| Finding 74 candidates assessed, 60 randomised, 30 per arm, all analysed, with no loss to follow-up. Men only, a single centre, in residential abstinence, with negative urine screens as an exclusion. The sample size calculation targeted 80% power for a Cohen’s d effect size of 0.74 on cue reactivity, that is, a large effect | |
| Independence | Solid |
| Finding Funded by a NARSAD Young Investigator Award no. 27305 from the Brain and Behavior Research Foundation and by a MnDRIVE grant from the University of Minnesota, with the Medical Discovery Team on Addiction. The published version states “The authors declare no competing interests.” No industry funding, no device manufacturer among the declared funders | |
| Access to the full text | Solid |
| Finding The published version is freely available on the publisher’s site, under a Creative Commons Attribution, Non-Commercial, No-Derivatives licence, with its supplementary material. A preprint version deposited on medRxiv is also accessible. The two versions were compared for this analysis: no discrepancy was found in the sample sizes, stimulation parameters, craving statistics, connectivity coefficients, correlations, or conclusions | |
Results
| Endpoint | Published result |
|---|---|
| Brain activity to drug-related cues (primary endpoint) | Significant time-by-group interaction in whole-brain analysis, voxel threshold p < 0.005 and cluster extent threshold set by simulation for family-wise correction at α = 0.05, that is, at least 40 contiguous voxels. Main cluster of 864 voxels over the bilateral striatum, amygdala, ventral tegmental area and posterior cingulate cortex, peak at MNI coordinates [19, -35, 7]. The direction of the effect is a fall in cue-related activity after stimulation in the active arm compared with the sham arm |
| PEB readingTarget reached the result is clear and properly corrected, but it concerns a brain signal, not the patient’s state | |
| Subjective craving, visual analogue scale (secondary) | Group effect F(1, 344) = 0.056, p = 0.8133. Time-by-group interaction F(5, 344) = 0.386, p = 0.8581. Time effect F(5, 344) = 14.21, p < 0.0001 |
| PEB readingNo difference between groups with 60 participants in a trial sized for a large effect on a different endpoint, the absence of a difference does not prove the absence of an effect. No favourable trend appears on this scale either | |
| Resting-state connectivity between cortex and subcortical regions (prespecified secondary) | An increase beneath the stimulation sites and a decrease with the left medial amygdala, without reaching significance. The authors attribute this result to a lack of power. No significant change in striatal subregions |
| PEB readingNot significant this is the only neural endpoint prespecified as secondary, and it is negative. Lack of power is one possible explanation among others, it is not a demonstration | |
| Right frontoparietal network engagement during the task | Time-by-group interaction on connectivity in a generalised psychophysiological interaction analysis: β = 2.16 for the cluster of connections, p = 0.0308 after FDR correction, and β = 2.63 for the right parieto-frontal connection highlighted in the abstract |
| PEB readingBarely significant a p value three hundredths short of the threshold, on an endpoint registered as exploratory, corrected within one family of tests but not across the three families analysed | |
| Coupling with the right ventral striatum during the task | β = 1.99, p = 0.037 after FDR correction. The measured quantity is a more marked fall in parieto-striatal coupling in the active arm than in the sham arm, which the authors interpret as strengthened top-down inhibitory regulation |
| PEB readingConsistent with the hypothesis the same statistical fragility applies. The authors themselves state that this type of analysis allows no inference about the direction of influence between regions | |
| Coupling with the left medial amygdala during the task | β = 1.97, p = 0.039 after FDR correction, also in the direction of reduced coupling in the active arm |
| PEB readingConsistent with the hypothesis same caveat, and the same impossibility of inferring a direction of influence | |
| Simulated electric field and network engagement | Significant group-by-electric-field interaction (p = 0.03). Correlation r = 0.43, p = 0.02 in the active arm, absent in the sham arm (r = 0.08, p = 0.69) |
| PEB readingA plausibility argument a relationship between the physical dose received and the measured effect, present under active stimulation and absent under sham stimulation, makes the result less likely to be an artefact | |
| Link between connectivity and craving | Significant group-by-connectivity interaction (p = 0.04). Brain-behaviour correlation r = -0.37, p = 0.05 in the active arm, absent in the sham arm (r = 0.08, p = 0.69). Result present in the published full text, absent from the abstract |
| PEB readingExploratory an uncorrected within-group correlation, at the exact edge of the threshold, in 30 participants. It suggests a lead, it does not establish one | |
| Other craving scale | No time-by-group interaction on the total score of the desire-to-use questionnaire. Post hoc analyses on the desire subscale: p = 0.071 immediately after the post-stimulation MRI (sham 1.92 ± 1.03; active 1.49 ± 0.81) and p = 0.0613 the next day |
| PEB readingPost hoc and not significant the authors describe a lower score in the active arm the next day while reporting 1.68 ± 0.82 for the active arm against 1.34 ± 0.5 for the sham arm, figures that say the opposite. This inconsistency appears identically in the preprint and the published version. The next-day figure is therefore unusable | |
| Tolerability | One severe adverse event in each arm: difficulty concentrating after the session under sham stimulation, skin redness the next day under active stimulation. No other severe event reported |
| PEB readingNo concerning signal over a single exposure and sixty participants, this documents only immediate tolerability | |
One methodological point deserves to be isolated, because it holds well beyond this trial. The fall in craving over the course of the session is clear and highly significant in both arms. A study with no control group, using the same device and the same before-and-after measurement, would have produced a spectacular and false result. This is precisely what the sham arm allows to be ruled out.
Critical appraisal
| Domain | Judgement |
|---|---|
| Match between claim and evidence | Solid |
| Finding The published abstract explicitly states the absence of a difference on subjective craving, and the conclusion speaks of a potential intervention, not a treatment. This restraint is real and deserves to be noted in a field where it is rare | |
| Internal validity | Solid |
| Finding Randomisation by computer-generated codes, sham arm with 30-second ramps, groups coded A and B for analysts, unblinding only after the prespecified analysis, family-wise correction on the primary imaging endpoint, no loss to follow-up among sixty randomised participants. Two different MRI scanners were used, of identical model and with the same acquisition protocols | |
| Scope of the primary endpoint | Major caveat |
| Finding An imaging endpoint does not predict clinical benefit. The recent history of biological psychiatry is full of brain targets successfully hit with no translation for the patient. Treating this as an efficacy result would be a misreading | |
| Single exposure | Major caveat |
| Finding A twenty-minute session. Neuromodulation protocols aimed at treatment are counted in dozens of sessions. This trial does not test a treatment, it tests an acute physiological response | |
| Observation horizon | Insufficient |
| Finding Twenty-four hours. The persistence of the brain effect is unknown, and the question of relapse is not addressed | |
| Multiplicity and fragility | Caveat |
| Finding The connectivity results are FDR-corrected, and the authors say so: this is a positive point. But the correction is applied within each of the three families of connections tested, not across them, and the follow-up correlations (r = 0.43 and r = -0.37) are uncorrected. Several values sit between p = 0.02 and p = 0.05. These effects require replication | |
| Blinding and expectations | Caveat |
| Finding The strength here is that the guesses do not discriminate between groups: in both arms, a large majority believe they received real stimulation, with no significant difference. Two weaknesses follow. First, what this reveals, a strong and uniform expectation: on a self-reported endpoint like craving, this expectation pushes both arms in the same direction and plausibly contributes to the observed time effect. Second, the supplementary material shows that the next day, participants in the active arm are significantly more confident in their guess than those in the sham arm (p = 0.028). Blinding is not limited to the response given, it also concerns certainty, and on this point symmetry between the arms is not complete | |
| External validity | Caveat |
| Finding Men only, abstinent, residentially housed, negative urine screens, a single centre in the United States. The authors explain the exclusively male recruitment by the very small number of women housed at the centre, and acknowledge that generalisation to women remains to be established. Active users and outpatients are not represented. The exact proportion of participants on agonist treatment is not reported, even though the buprenorphine-naloxone combination is the most frequently reported current medication in both arms, which limits interpretation of the observed craving level | |
| Priority claim | To be handled precisely |
| Finding The introduction claims priority, but narrowly defined: the first registered, randomised, triple-blind, sham-controlled trial applying high-definition alternating current stimulation at two in-phase sites on the frontoparietal network in opioid use disorder. Stated this way, the claim is defensible and identical across both versions of the text. It does not mean this would be the first trial of non-invasive brain stimulation in this condition at all: such work exists and is the subject of a 2025 update review. PEB therefore does not relay any shortened formulation along the lines of “first stimulation trial in opioid addiction” | |
The dividing line is clear. What is demonstrated is that a single session of frontoparietal stimulation at 6 Hz changes, immediately after stimulation, the brain response to opioid-related cues in sixty abstinent men, and that no difference in subjective craving is detected between the two arms under these conditions. What is suggested, by the relationship between the simulated electric field and network engagement, present under active stimulation and absent under sham stimulation, and by a within-group correlation at the edge of the threshold, is that the observed effect does depend on the physical dose delivered. What amounts to hypothesis or expert opinion is the idea that a repeated, extended protocol combined with agonist treatment might one day translate into clinical benefit. This hypothesis is legitimate. It is not tested here. And nothing in this data allows one to say that stimulation has no effect on craving: a trial sized to detect a large effect on a different endpoint cannot settle this question.
Level of evidence
PEB assessment: reasonable confidence in the existence of an acute brain effect of active stimulation compared with sham stimulation, low confidence in the stability of the connectivity effects taken individually, and a complete absence of data on any clinical outcome whatsoever. The quality of the experimental design is above the field average: a sham arm, multiple blinding tested and tested statistically, a primary endpoint registered before recruitment and never modified, individual electric field modelling, freely available full text and supplementary material, a null self-reported result reported without hedging in the abstract. What limits the reach is not the rigour of execution, it is the scope of the question asked: one session, one imaging marker, twenty-four hours of observation.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ They showed that a twenty-minute session moves the MRI signal, and they showed it cleanly, with a real placebo arm. Except on craving, nothing separates the two groups, and follow-up stops the next day. That wasn’t the question they were asking anyway. For my patients, the question remains methadone or buprenorphine, and retention in care. I’ll look again when there are twenty sessions and a clinical endpoint at six months. ”
What this means in practice: for any neuromodulation trial, systematically ask whether the primary endpoint is a brain signal or a patient state. The two are not interchangeable.
What you can take from this on Monday morning. Nothing changes in prescribing, but several things can be said, documented and monitored differently.
- Reaffirm the foundation, with figures rather than with a position of principle. The French national health authority’s guidance sheet, validated on 10 March 2022, names only buprenorphine, alone or combined with naloxone, and methadone for opioid use disorder, and sets out long-term treatment. The cohort data pooled by Sordo and colleagues place all-cause mortality at 11.3 per 1,000 person-years during methadone treatment against 36.1 outside it. The first task of a consultation is not to add a technique, it is not to interrupt what protects.
- Know how to answer a patient or relative who has read that addiction is treated by stimulating the brain. The honest answer fits in three sentences: yes, we can now modify the activity of the involved circuit in a controlled way; no, this has not been shown to reduce craving or use; research is at the stage of verifying that the technique reaches its target, which is a useful step, not a therapy.
- Do not overestimate an improvement observed after any uncontrolled intervention. In this trial, craving falls sharply over the course of the session in both arms, including under sham stimulation. Any before-and-after assessment with no comparator, in addiction medicine, produces this kind of curve.
- Document craving with a simple, dated tool, a visual analogue scale or equivalent, at fixed time points. This trial shows that the measurement moves quickly and substantially: a single measurement has no value, a dated series does.
- On the question of access: a search carried out on 11 August 2026 on the French national health authority’s website found no opinion, evaluation, or reimbursed-procedure listing concerning transcranial alternating current stimulation in opioid use disorder. This absence is an absence of search results, not proof that no document exists. What can be affirmed, however, is that the authority’s March 2022 guidance sheet on this disorder mentions no neuromodulation technique. Such a device also falls under medical device regulation rather than medicine marketing authorisation. Authorisation, commercial availability and reimbursement are three distinct questions, and in this indication none of the three has a documented answer to date. Readers outside France should expect a comparable three-way distinction in their own jurisdiction and verify its current specifics rather than assume this example transfers directly.
- What to watch for in the literature going forward: multiple-session protocols, with a clinical primary endpoint, a maintained sham comparator, follow-up of at least three to six months, and stimulation tested as an addition to agonist treatment rather than in its place.
Frequently asked questions
Does this stimulation treat opioid use disorder?
No, and the trial was not designed to answer this question. It measures an immediate brain response after a single session. Neither use, nor relapse, nor treatment retention, nor mortality was assessed. The reference treatment remains pharmacological and agonist-based.
Was the primary endpoint negative?
No, and this is the main misreading to avoid. The primary endpoint prespecified in the registry was an imaging endpoint, the change in BOLD signal during a drug cue-exposure task, and it is met. It is subjective craving, a secondary endpoint, that does not differ between groups. The other prespecified secondary endpoint, resting-state connectivity, is not significant either. A primary endpoint met on an intermediate marker is not a clinical result on its own.
Did the blind hold?
The authors tested it, and tested it statistically, which is to their credit. Most participants in both groups believed they had received real stimulation: 24 in the sham arm and 25 in the active arm after the functional MRI, 27 and 23 the next day, with no significant difference between arms (p = 0.509 then p = 0.464). On the response given, the blind therefore held. One nuance appears in the supplementary material: the next day, reported confidence in this guess is significantly higher in the active arm (83.11 ± 17.48) than in the sham arm (70.37 ± 24.54), p = 0.028. Neither the Bang nor the James blinding index is reported.
Why does craving fall in the sham group too?
Several mechanisms combine: expectation of an effect, regression to the mean after a high initial measurement, habituation to the repeated exposure task, and the residential care context. This is exactly the reason a sham arm exists, and it is why before-and-after studies with no comparator are uninterpretable in this field.
Is this device available or reimbursed anywhere today?
A search carried out on 11 August 2026 on the French national health authority’s website found no opinion, evaluation, or reimbursed-procedure listing concerning transcranial alternating current stimulation in opioid use disorder, and the authority’s March 2022 guidance sheet on this disorder mentions no neuromodulation technique. This is an absence of search results, not to be confused with proof that no document exists anywhere. In practice, in this indication, this technique currently belongs to clinical research rather than to a validated treatment option, and the same is likely true in most jurisdictions, though readers should verify their own.
What would it take for this to change practice?
A multiple-session trial, with a clinical primary endpoint rather than an imaging one, a credible sham arm, several months of follow-up, a population including women and outpatients, and stimulation assessed as an addition to agonist treatment. None of these conditions is met by the work analysed here, which does not diminish its methodological interest.
Annotated bibliography
Soleimani G, Kuplicki R, Paulus MP, Ekhtiari H (2026). Targeting cortico-striatal-amygdalar networks via theta-band frontoparietal synchronization in opioid use disorder: a randomized tACS-fMRI trial. Molecular Psychiatry, online publication of 26 June 2026, 13 pages. DOI 10.1038/s41380-026-03694-1 · PMID 42362768. Source study analysed here, read in its full published version, supplementary material included. Free access on the publisher’s site under a Creative Commons Attribution, Non-Commercial, No-Derivatives licence. No volume, issue or final pagination appear on the record as of the date of consultation, and no PubMed Central identifier is attached to it: open access here is provided by the publisher, not by a repository. No erratum or correction attached as of the date of consultation.
Soleimani G, Kuplicki R, Paulus MP, Ekhtiari H (2026). Targeting cortico-striatal-amygdalar networks via theta-band frontoparietal synchronization in opioid use disorder: a randomized tACS-fMRI trial. Preprint version, medRxiv. DOI 10.64898/2026.02.10.26346048. Preprint version, not peer reviewed, cited here for a single purpose: it was compared line by line with the published version on sample sizes, stimulation parameters, craving statistics, connectivity coefficients, correlations, funding statements and conclusions. No discrepancy was found. No figure in this analysis is drawn from the preprint alone.
ClinicalTrials.gov. Frontoparietal Synchronization to Modulate Drug Craving in Opioid Use Disorder: A Randomized Experimental Trial, NCT03907644. clinicaltrials.gov. Registration submitted 5 April 2019, posted 9 April 2019, sponsor Laureate Institute for Brain Research. Source used to verify the prespecified primary endpoint, the hierarchy of secondary and exploratory endpoints, the eligibility criteria, the declared masking, and the final sample of 60 participants. This document is what allows the statement that the primary endpoint was indeed an imaging endpoint and not craving. The six successive versions of the record, from version 0 of 5 April 2019 to version 5 of 9 July 2023 posted 11 July 2023, were compared: the primary endpoint was never modified; the stated purpose changed from “treatment” to “basic science” on 23 April 2019 and the phase from “1 and 2” to “not applicable” on 30 August 2021.
Haute Autorite de sante (2022). Treatment of opioid use disorder, guidance sheet validated by the board on 10 March 2022. has-sante.fr. Source of the French framework: buprenorphine and methadone as substitution medicines, initial methadone prescription restricted to physicians practising at specialised addiction centres or hospital physicians, buprenorphine prescribable in general practice, treatment to be considered over the long term, naloxone made available. No neuromodulation technique is mentioned in it.
Sordo L, Barrio G, Bravo MJ, Indave BI, Degenhardt L, Wiessing L, et al. (2017). Mortality risk during and after opioid substitution treatment: systematic review and meta-analysis of cohort studies. BMJ, 357, j1550. DOI 10.1136/bmj.j1550 · PMID 28446428. Reference used for the mortality argument: 19 eligible cohorts, 122,885 patients followed on methadone and 15,831 on buprenorphine, figures and confidence intervals verified against the published abstract. This is a synthesis of cohort studies, not of randomised trials, with a risk of confounding by indication that requires these rate ratios to be read as robust, convergent associations, not as experimental-level estimates of causal effect.
Blyth SH, et al. (2025). Noninvasive brain stimulation for opioid use disorder: an update of the literature. Current Behavioral Neuroscience Reports, 12(1), 8. DOI 10.1007/s40473-024-00293-8 · PMID 40343327. Update review cited here for a single purpose: to establish that non-invasive brain stimulation work in opioid use disorder predates the trial analysed, which rules out any absolute priority claim, without contradicting the narrow claim the authors themselves make. Its content was not critically evaluated for this analysis.