Neurofeedback has a long history in addiction care, with encouraging results on abstinence and psychological profile — but only as an add-on to a structured treatment programme, never on its own. More recently, HRV biofeedback (slow breathing guided by the heart rhythm) reduced craving and use days in one large trial.
Updated :
What the research shows
EEG neurofeedback — standard amplitude training
Alpha-theta neurofeedback (Peniston protocol, Scott-Kaiser modification: SMR/beta then alpha-theta) as an adjunct to a treatment programme — alcohol, opiate, stimulant and mixed substance use
Ages
Adult (18+)
Techniques
Alpha-theta, Sensorimotor rhythm (SMR), Beta training
AAPB3Probably efficaciousNeuroLogic4Efficacioushigher vs AAPB
Level 4, raised from the AAPB's Level 3. Randomised superiority over treatment as usual, replicated by independent groups, always as an add-on to a care programme: Scott 2005 (n = 121 randomised inpatients, SMR-beta then alpha-theta: better retention, higher one-year sustained abstinence, TOVA normalisation, 5 of 10 MMPI-2 scales — AAPB ch. 2); Rostami & Dehghani-Arani 2015 (n = 100 methamphetamine-dependent patients, 30 sessions plus pharmacotherapy vs pharmacotherapy alone: lower addiction severity, better psychological health and quality of life); Dehghani-Arani 2010/2013 (n = 20 on methadone or buprenorphine: improved general health, reduced heroin craving); Peniston & Kulkosky 1989/1990 (n = 10 + 10 + 10, blind outcome evaluation, sustained relapse prevention at 13 months). Wan 2026 pools 17 RCTs (n = 662): Hedges' g = 0.85 on addiction symptoms, larger for substance than behavioural addiction, session number a main moderator. AAPB stopped at Level 3 ('do not completely meet the criteria for Level 4'); NeuroLogic reads three positive randomised TAU-controlled trials in independent settings, plus the pooled estimate, as meeting the Level 4 replication criterion. No 2022-2026 RCT of clinician-delivered alpha-theta exists (Rempala 2022: 15 enrolled, 2 completed); the level rests on the earlier trials. Level 5 is not assigned: no superiority over a bona fide comparator and no sham-controlled positive.
Other amplitude EEG protocols without an alpha-theta phase (SMR/beta only, theta/SMR, qEEG-guided alpha) as an adjunct to a treatment programme
AAPB2Possibly efficaciousNeuroLogic2Possibly efficacioussame vs AAPB
Level 2. The AAPB's printed Level 2 for 'other randomised clinical studies using different NFB or biofeedback protocols': Keith 2015 (n = 95, three arms — automated SMR/low-beta, clinician-guided, or additional therapy sessions: attention improved in both neurofeedback arms with no difference between them, no change in the therapy arm), Fielenbach 2018 (19 completers, theta/SMR in forensic patients: one in five changed the target bands; only craving was associated with response), Ko & Park 2018 (n = 36 inpatients with alcohol use disorder, 10 qEEG-based sessions: higher alpha, lower high beta, better abstinence self-efficacy), Lackner 2015. The 2022-2026 window adds Hack 2024 (n = 56, add-on EEG neurofeedback vs inpatient rehabilitation alone: better emotional competences and life satisfaction, resting low-beta decrease; drinking outcomes not reported in the abstract, protocol unspecified) and Fadaei 2025 (four arms of 8: neurofeedback alone normalised EEG power but did not reduce cue-induced craving). Small samples, short courses, mixed results: Level 2 held.
Other neurofeedback methods
Cognition-guided / decoded EEG neurofeedback (deactivating a multivariate cue-reactivity pattern)
Ages
Adult (18+)
Techniques
Other neurofeedback
AAPB0Not ratedNeuroLogic3Probably efficacious
Level 3. Gou 2025: cognition-guided closed-loop EEG neurofeedback, 99 men with methamphetamine use disorder in two samples — 10 sessions deactivating each participant's own methamphetamine-cue MVPA pattern vs yoked feedback (n = 33 + 33): improved go/no-go response inhibition (Cohen's f = 0.31), neurofeedback performance correlated with the gain; an internal validation sample (17 real vs 16 standard rehabilitation) replicated the effect and exceeded standard rehabilitation. Meng 2023: double-blind yoked-controlled RCT, n = 60 nicotine-dependent, two sessions; the parent trial reported reduced cigarette consumption at 4 months, and higher pre-training frontal theta predicted response (secondary analysis). Sham-controlled and positive, with an internal replication, but the primary outcome is cognitive rather than use or craving, both reports come from one research programme, and no independent group has replicated — Level 3 rather than 4.
LORETA and variants (z-score LORETA)
Ages
Adult (18+)
Techniques
LORETA, qEEG-guided (incl. z-score)
AAPB0Not ratedNeuroLogic2Possibly efficacious
Level 2. One three-arm randomised trial, n = 30 men with opioid use disorder on methadone maintenance (10 per arm), 20 sessions of LORETA z-score neurofeedback vs 15 sessions of attention-bias modification vs methadone alone, published as three papers: craving reduced more than both comparators at post-test and one month (Faridi 2022); anxiety and depression (BAI, BDI) reduced more than in the cognitive arm, maintained at follow-up (Faridi 2024); quality of life (WHOQOL-BREF) and go/no-go response time improved more, the response-time gain holding at one month only in the LORETA arm (Faridi 2025). One small sample, one group, no sham, no independent replication: Level 2, not 3.
Level 2. Repeated demonstration of target engagement without clinical superiority. Karch 2022: double-blind placebo-controlled rt-fMRI, n = 52 alcohol-dependent inpatients, three sessions as an add-on to standard treatment — no superior 3-month abstinence, but greater medial frontal / caudate down-regulation in active patients who stayed abstinent. Taebi 2024: double-blind sham-controlled, preregistered, n = 60 nicotine users, one session of connectivity-based feedback — no difference on inhibitory control, smoking or impulsivity; both groups regulated the target network similarly. Fede 2023: n = 30 with alcohol use disorder, single session comparing three feedback methods — ACC and dlPFC down-regulation predicted craving reduction, no no-feedback control. Murphy 2024 systematic review (16 studies, 446 participants): consistent between-condition effects, little intervention-by-time effect, evidence 'early and inconsistent'. Dose is one to three sessions throughout: mechanistic work rather than a course of training; under Parsons 2026 these sham nulls do not show absence of effect, but nothing shows presence either.
Level 1. The single adequately powered trial is null: Gabrielsen 2022, n = 93 outpatients with substance use disorders in Norway, randomised to 20 sessions (30 min) of ILF neurofeedback plus treatment as usual or treatment as usual alone — no difference in quality of life (primary), drug use, sleep, anxiety or depression; lower restlessness only (mean difference −1.8, 95 % CI −3.1 to −0.5). This is a treatment-as-usual comparison at a plausible clinical dose, not a sham comparison, so the Parsons 2026 caveat does not apply. With no positive controlled study, the indication-specific level is 1, without prejudice to the ILF levels set for other indications.
Biofeedback
HRV biofeedback (resonance-frequency breathing) — craving and substance use, as an adjunct to treatment
Level 4. Two randomised treatment-as-usual-controlled positives from independent groups, plus a positive pooled craving effect. Eddie 2025 (phase 2 RCT, n = 120 treatment-seeking adults with substance use disorder, 8 weeks of wearable HRV biofeedback + TAU vs TAU, ecological momentary assessment): reduced negative affect and craving (P = .001 and P < .001) while controls increased on both; lower proportion of alcohol-and-other-drug use days (OR 0.36, 95 % CrI 0.25-0.54, a 64 % reduction); the within-person craving → later use link weakened (OR 0.84). Yen 2022 (pilot RCT, n = 61 men in an outpatient methamphetamine programme, four weeks of HRVB + TAU vs TAU): greater reductions in craving, dependence severity and positive urine tests at end of the intervention and four-week follow-up. Roxburgh 2026 (15 studies meta-analysed): craving g = 0.30 (95 % CI 0.12-0.48, I² = 23.5 %, 12 studies, 696 participants), dependence g = 0.85 (2 studies), withdrawal uncertain, no difference between HRVB and other controlled-breathing interventions. Not counted: van Dijk 2024 (n = 156 pregnant smokers, HRVB as one component of an eHealth package vs active control — smoking fell in both arms), a package null rather than a test of biofeedback. No trial has relapse or 6-month sustained abstinence as its primary endpoint.
HRV biofeedback — comorbid depression and sleep in substance use disorder
Ages
Adult (18+)
Techniques
HRV — resonance-frequency breathing
AAPB0Not ratedNeuroLogic3Probably efficacious
Level 3. Yen 2023, the same n = 61 methamphetamine sample randomised to HRV biofeedback + TAU or TAU: greater decrease in depressive symptoms and greater improvement in sleep quality than TAU at end of intervention and at follow-up, with different HRV-symptom associations in the two groups. Eddie 2025 adds reduced negative affect (positive affect unchanged) in an independent sample. Secondary outcomes from a single small trial, effect sizes not given in the abstract — Level 3, not 4.
Thermal biofeedback (skin temperature with autogenic phrases)
Ages
Adult (18+)
Techniques
Thermal biofeedback
AAPB0Not ratedNeuroLogic2Possibly efficacious
Level 2. Not free-standing here: skin-temperature training with autogenic phrases is the mandated pre-training phase of the Peniston protocol (at least five sessions, 94 °F criterion before EEG training — AAPB ch. 2) and therefore sits inside the alpha-theta package rated above; Scott 2005 omitted it. Independent evidence is thin: one skin-temperature smoking-cessation study within Keilani 2022 (three-study PRISMA review, poor-to-good methodological quality, no pooling possible) and Pandria 2023 (17 smokers, five thermal biofeedback sessions then 20 neurofeedback sessions, uncontrolled: lower exhaled carbon monoxide and Fagerström scores, within-session temperature learning). The AAPB's Level 2 sentence for 'other biofeedback protocols' is the nearest printed rating; the AAPB block shows 0 because no thermal study is rated separately.
In short
Clinical reading
NeuroLogic Level 4 (AAPB 3) for the Scott-Kaiser modification of the Peniston protocol combined with residential or office-based rehabilitation (including opioid replacement therapy): randomised superiority over treatment as usual in Scott 2005 (n = 121), Rostami 2015 (n = 100) and Dehghani-Arani 2010/2013, pooled g = 0.85 (Wan 2026). Other amplitude EEG protocols stay at the AAPB's Level 2. HRV biofeedback: Level 4 (not rated by AAPB) on Eddie 2025 (n = 120) and Yen 2022 (n = 61). Decoded neurofeedback 3, LORETA 2, fMRI 2, ILF 1.
Protocols
Occipital or parietal alpha-theta training (O1, Pz) with auditory feedback, often preceded by SMR/beta work (Cz) and thermal pre-training; long series (20-40 sessions) embedded in the care pathway. Resonance-frequency HRV biofeedback, 4-8 weeks with daily home practice.
Limits
Other protocols rest on small samples, short interventions and mixed results; the specific effect of neurofeedback remains confounded with rehabilitation; no 2022-2026 RCT of clinician-delivered alpha-theta neurofeedback; no trial with relapse as primary endpoint; no evidence in children or adolescents.
Study base
One 121-participant controlled study, pilot RCTs and several observational studies, all pre-2020; a newer and methodologically stronger HRV literature. 2022-2026 base: 23 publications indexed in the archive (17 neurofeedback, 5 HRV biofeedback, 1 biofeedback review).
Brendan's perspective
Level 4 here is a raise, and I am comfortable with it. Scott 2005, Rostami 2015 and Dehghani-Arani 2010/2013 are three randomised comparisons against treatment as usual, run in independent settings, and Wan 2026 pools them at g = 0.85. What they show is an add-on effect: alpha-theta training inside a rehabilitation programme, never instead of one. That is also how I use it — SMR/beta work around Cz first, then occipital or parietal alpha-theta with auditory feedback. Cz is a landmark in that sentence, not a prescription: where the electrode actually goes follows the qEEG profile in front of me, twenty to forty sessions, agreed with the team carrying the rest of the care. I would bring in HRV biofeedback early: an unregulated autonomic system is noise in the EEG learning loop, and Eddie 2025 and Yen 2022 make it more than a comfort measure. ILF sits at 1 because Gabrielsen 2022 tested it properly, at a clinical dose and against usual care, and found nothing. What nobody has shown: relapse, as a primary endpoint, at a year.
Sokhadze et al. (2008) EEG biofeedback as a treatment for substance use disorders: review, rating of efficacy and recommendations doi:10.1007/s10484-007-9047-5
Scott et al. (2005) Effects of an EEG biofeedback protocol on a mixed substance abusing population doi:10.1081/ada-200056807
Rostami & Dehghani-Arani (2015) Neurofeedback training as a new method in treatment of crystal methamphetamine dependent patients: A preliminary study doi:10.1007/s10484-015-9281-1
Wan et al. (2026) Effect of EEG neurofeedback therapy on addiction disorders: A systematic review and meta-analysis doi:10.1111/add.70164
Eddie et al. (2025) Heart Rate Variability Biofeedback for Substance Use Disorder: A Randomized Clinical Trial doi:10.1001/jamapsychiatry.2025.2700
Roxburgh et al. (2026) Controlled breathing and heart rate variability biofeedback for substance use-related outcomes: A systematic review and meta-analysis doi:10.1111/add.70542
Gou et al. (2025) Improved Response Inhibition Through Cognition-Guided EEG Neurofeedback in Men With Methamphetamine Use Disorder doi:10.1176/appi.ajp.20240475
Gabrielsen et al. (2022) Infralow neurofeedback in the treatment of substance use disorders: a randomized controlled trial doi:10.1503/jpn.210202
