Does Brain Stimulation Reduce Methamphetamine Craving? A 60-Man Trial Changed fMRI Connectivity, Not Craving

Short answer: In a triple-blind, randomized, sham-controlled experiment involving 60 men with methamphetamine use disorder, one session of active transcranial direct current stimulation (tDCS) did not reduce self-reported drug craving significantly more than sham stimulation. Both groups reported lower craving after stimulation and repeat cue exposure. Researchers nevertheless detected an active-versus-sham change in a specific frontal-to-parietal functional magnetic resonance imaging (fMRI) connectivity measure using cluster-level false-discovery-rate correction. Five other candidate brain-activation clusters, although initially significant under permissive uncorrected thresholds, did not survive false-discovery-rate correction. A brain-connectivity result is not a demonstrated clinical craving benefit, a measured methamphetamine blood concentration, or evidence that any particular person was impaired.

The primary source is the original trial by Hamed Ekhtiari, Ghazaleh Soleimani, Rayus Kuplicki, Hung-Wen Yeh, Yoon-Hee Cha and Martin Paulus, Transcranial direct current stimulation to modulate fMRI drug cue reactivity in methamphetamine users: A randomized clinical trial, Human Brain Mapping 43 (2022): 5340–5357, original research DOI. The essential scientific question is whether neural response to drug-related images and a person’s felt craving respond to stimulation in the same way. They did not provide interchangeable answers.

What precise question did this experiment answer?

Researchers tested whether a twenty-minute application of weak electrical current over a frontal brain area changed the neural response to drug-related pictures and subjective craving more than a matching sham procedure. The imaging protocol compared responses to photographs of methamphetamine-related objects and neutral images, both before and after the intervention. Participants also repeatedly rated how much they craved the drug. The investigators asked whether electric fields modeled for each person’s head related to subsequent activity and connectivity patterns.

This was a study of a nonpharmacological intervention during residential treatment in early abstinence. It was not an acute methamphetamine administration experiment, not a trial of a pharmacological antidote to intoxication, and not a test of whether someone could drive safely after consuming a controlled drug dose. In particular, the intervention did not directly introduce methamphetamine, amphetamine, ethanol, or any other drug into the participants’ bloodstream. Treating the electrical current as an administered drug dose would be a fundamental category error.

How many independent people actually participated?

The investigators screened 80 people, obtained consent from 75, and randomized 60 men with methamphetamine use disorder into two groups: 30 active tDCS and 30 sham. The participants were assessed as independently randomized people, not as the number of images, fMRI voxels, picture blocks, repeated self-ratings, or statistically selected brain clusters. A person assessed at six time points is still one person. Likewise, a brain scan containing tens of thousands of voxel measurements does not increase the biological sample from 60 to tens of thousands.

Participants were men drawn from a residential recovery setting, with extended histories of methamphetamine use and variable prior exposure to other drugs. The source’s demographic table reported average ages of 34.96 years in active and 37.75 years in sham, with average current abstinence of approximately 63.27 and 58.33 days, respectively. These are population descriptors, not a precise last-use clock for any one participant. The study does not establish generalizability to women, emergency department patients, medically unstable persons, on-road drivers, or persons experiencing recent acute methamphetamine toxicity.

What does active tDCS mean—and what did sham deliver?

The active protocol used an approximately 2 milliampere current for 20 minutes across two scalp electrodes, with the anode at the right dorsolateral prefrontal scalp location F4 and the cathode near the left supraorbital location Fp1. Each electrode pad covered approximately 5 × 7 square centimeters. The sham procedure delivered current briefly, for about 40 seconds, to mimic initial sensations before sustained stimulation was stopped. Both groups therefore experienced an intervention setting and the sensory context of electrode placement.

The study was designed to mask the participant, stimulation personnel and data analysts to assignment. That design is stronger than an unmasked before-and-after intervention, but it cannot remove all uncertainty. Sham sensations, belief in treatment, repeat visual exposure, time and expectancy may influence symptom ratings. Nor should a simulated electrical field distribution be mistaken for a direct neuronal recording: the investigators’ individual head models estimated where stimulation-induced electric fields might be strongest, rather than chemically assaying drug or directly measuring neurotransmitter concentrations.

How were methamphetamine cues, neutral pictures and craving measured?

The fMRI task showed pictorial methamphetamine cues interspersed with neutral pictures. A conventional BOLD imaging sequence recorded blood-oxygenation-linked MRI signal changes as participants watched the images. The methamphetamine-versus-neutral contrast was derived statistically from these signal series. Outside the scanner, craving was assessed six times, including baseline before the stimulation day, before and after the first scan, before and after the second scan, and on the following day. Some inside-scanner scores used a different, shorter scale; they must not be numerically combined with the outside-scanner 0–100 visual analogue scores.

The participants also performed a simple response to a yellow box placed around images. Reaction times and image-category responses were task outcomes, not observed errors on an actual highway or validated operational impairment. Repetition mattered: participants viewed drug pictures in the initial scan, then again after stimulation. Familiarity and habituation were plausible alternate explanations for why several responses diminished on the second exposure, including among those given sham stimulation. Comparing active with sham is consequently more probative of an intervention-specific effect than noting a simple decline over time.

Was any biological specimen used to quantify methamphetamine?

The original report describes screening related to abstinence and eligibility, including drug testing and alcohol-breath screening, but does not report a quantitative methamphetamine or amphetamine result from legally collected human venous whole blood. It provides no validated parent-drug whole-blood concentration in ng/mL, no serial forensic blood pharmacokinetic curve, and no method-specific instrument calibration or detection threshold that would establish a participant’s contemporaneous concentration. It also does not provide a paired chemical analysis in centrifuged plasma or serum, urine, oral fluid, sweat, hair, vitreous, or breast milk for interpreting these fMRI responses.

Whole blood, serum, plasma and oral fluid are not interchangeable biological matrices. An eligibility urine test may address a limited screening question while saying very little about a precisely timed concentration in a separate specimen. A breath alcohol prescreen is not a forensic whole-blood ethanol assay. Likewise, fMRI BOLD signal is blood-oxygenation-related imaging—not quantitative mass spectrometry of methamphetamine circulating in venous blood. For background on the distinction, see the site’s separately published biological-matrices guide and confirmatory drug-testing guide.

Did active brain stimulation reduce self-reported craving more than sham?

No statistically significant active-versus-sham advantage in craving was demonstrated. The mixed-effects model found a strong main effect of measurement time, reported at p < .0001, but no significant group main effect and no significant group-by-time interaction. Before the first cue scan, mean 0–100 craving ratings were 39.4 in sham and 42.50 in active. After that first scan, means rose to 59.07 and 64.40. Both groups subsequently fell to lower values after the intervention and second exposure. A significant change occurring in both groups does not establish that the active component produced the decrease.

When compared at the later day-after assessment, reported means were 11.17 for sham and 12.67 for active—again not an intervention-specific superiority result. Because the study reports means with standard errors for these craving scales, it is wrong to relabel the ± figures as between-person standard deviations. The raw values describe participants’ reported experiences at specific measurement times; they do not represent a chemical exposure or an estimate of the probability of relapse in a future month.

What did the actual numerical craving data show?

The first source-grounded evidence table preserves measurement time, units, group allocation and dispersion type rather than combining unrelated scales. The pronounced first-cue increase and later reduction are visible in both trial arms. No row in this table establishes a statistically significant active-versus-sham craving effect, because the relevant modeled interaction was nonsignificant.

*The source prints an unusually large standard error, 21.55, for the sham T3 value. This is reproduced faithfully and not silently corrected. Because individual participant data were not made available for reanalysis, its cause cannot be determined here. The main inference relies on the reported nonsignificant group and interaction effects, not on speculative repair of one dispersion entry.

How should the five apparent fMRI activation clusters be interpreted?

The research reported five left-hemisphere brain clusters in a time-by-treatment contrast for drug-versus-neutral cue responses: the middle frontal gyrus, anterior insula, inferior parietal lobule, precuneus and inferior frontal gyrus. These initially passed the analysis’s exploratory uncorrected voxel threshold of p < .005 with a minimum cluster size greater than 40 voxels. The numerical sizes in the original Table 4 were 88, 87, 105, 47 and 49 voxels, respectively. Those are counts of spatially correlated image voxels, not counts of independent biological participants.

Most importantly, the original Results section expressly states that none of these five clusters survived more stringent false-discovery-rate correction. Therefore, this article does not advertise those activation maps as five separately validated brain biomarkers, diagnostic tests or reliable treatment targets. The paper’s interpretive prose sometimes refers to significant activation, but that wording must be read in light of the correction failure. Spatial multiple testing across many brain locations can generate seemingly compelling patterns that are more fragile than a pre-specified, adequately corrected comparison. The distinction between exploratory and correction-surviving findings should be stated in the answer, not hidden in a methods footnote.

Did any connectivity result survive multiple-comparison control?

Yes, one task-modulated frontoparietal connectivity result did. For a generalized psychophysiological interaction (gPPI) analysis, investigators selected a right superior frontal gyrus seed based on where their modeled electrical field was greatest, then examined changes in connectivity during methamphetamine-versus-neutral pictures. A single right parietal cluster of 330 voxels, centered in standardized MNI coordinates approximately (36, −60, 38), met the source’s voxel-level p < .001 uncorrected threshold and cluster-level false-discovery-rate corrected p < .05.

In the source’s standardized task-modulated connectivity estimates, the active arm changed from 0.25 ± 0.14 before stimulation to 0.02 ± 0.13 afterward, while sham changed from 0.11 ± 0.15 to 0.20 ± 0.15. These are imaging-derived functional-connectivity units and standard errors, not milligrams, nanograms per milliliter, volts measured inside neurons, or direct drug levels. Statistical evidence for different imaging responses does not by itself demonstrate a clinically meaningful reduction in the participants’ own craving, much less a validated effect on relapse or behavior outside the research scanner.

What did the second evidence table separate?

The next table combines several distinct kinds of numerical evidence to make visible why they should not be treated as one outcome. The imaging activation screening used a threshold that failed FDR correction; the selected connectivity interaction survived cluster-level correction; and the behavioral craving time-by-group test was nonsignificant. The table is not a new statistical meta-analysis and does not invent confidence intervals that the source did not supply.

What is a gPPI signal—and what can it not establish?

Generalized psychophysiological interaction modeling estimates how the statistical relationship between two regions’ BOLD signals varies with the task condition. It is useful for hypotheses about the functional organization of networks during an experiment. It does not directly prove anatomical wiring, synaptic transmitter flux, a measurable electrical current reaching every relevant neuron, conscious desire, or drug administration. Even a positive task-dependent frontal-to-parietal connectivity result cannot be substituted for the actual endpoint of interest if that endpoint is reduced craving, drug use, dangerous conduct or recovery.

The imaging study used standard preprocessing and statistical modeling choices. A “cluster” contains neighboring voxels that are not independent sampling units, and correction methods address only specified families of tests under their assumptions. The individual head models provide estimates of current distribution influenced by tissue geometry and model specifications, not a direct measurement of dopamine, GABA, glutamate, methamphetamine, or cardiac effects. Scientific interpretation is strongest when one identifies which measurement was directly acquired, which quantity was estimated through modeling and which proposed mechanism remains hypothetical.

Did the modeled electric field predict connectivity?

The original investigators reported that the normal component of the estimated electrical field in the right superior frontal region correlated with pre-to-post connectivity change in active participants (r = .44, p = .021) and with the frontal functional-activity change (r = .47, p = .013). They stated that the latter association differed between active and sham arms (p = .0078). Tangential-field associations were not comparably established. These are exploratory associations among modeled fields and imaging-derived responses within the same experiment.

They do not show that a given person’s modeled field can predict whether that person will cease methamphetamine use or maintain abstinence. The relationship could be sensitive to anatomical model assumptions, signal noise, participant selection and the statistical model’s form. The authors expressly described unresolved mechanistic links between induced fields, BOLD changes and subjective craving. Without a separate prospective validation cohort and clearly specified predictive performance, it would be premature to market these correlations as an individualized clinical or forensic test.

Could repeated picture exposure explain declining craving?

Yes, cue habituation is a substantial interpretive concern. The first methamphetamine-cue scan tended to elicit more craving, but the second scan produced a muted or absent rise in both study arms. The sham group also showed widespread decreases in some brain responses during the repeated task. Thus both the procedure’s timing and a second exposure to familiar cues could contribute to within-person reductions unrelated to sustained active tDCS. A short real current pulse at the start of sham and participant beliefs about having received real stimulation present further alternate explanations.

This does not prove that active stimulation has no possible therapeutic effects in every configuration; one right-prefrontal twenty-minute protocol in sixty selected men is a narrower test. It also does not demonstrate a durable clinical treatment benefit. The source did not follow participants for enough time with objective relapse adjudication to validate the prediction that a changed image-network measure would correspond to sustained absence of methamphetamine use. Clinical efficacy requires outcomes appropriate to the claim, not only a striking anatomical map or a significant before-after comparison.

Were side effects and blinding meaningfully assessed?

The source reported that all 60 randomized participants tolerated the protocol during data collection, with no conventionally significant between-group differences on the measured adverse-effect categories. In the post-session questionnaire, 25 of 30 active participants and 26 of 30 sham participants guessed that their treatment had been active; that comparison was nonsignificant (p = .718). This is evidence that assignment guesses did not differ detectably between the groups, but it also shows that most participants in both arms believed they had received active stimulation.

The same questionnaire asked whether stimulation seemed to reduce craving: 22/30 active and 18/30 sham responded yes, with a nonsignificant group difference (p = .273). Perceived efficacy is therefore not equivalent to independently verified active treatment benefit. The absence of a statistically significant adverse-event difference in sixty people should not be advertised as proof of zero risk in all medical populations or with repeated, differently targeted electrical stimulation schedules.

Can one infer drug exposure, impairment or last-use timing from the brain maps?

No. A colored BOLD activation cluster has no validated calibration against a chemically measured concentration of parent methamphetamine in legally collected human venous whole blood. Nor can a modeled gPPI connectivity coefficient be converted into ng/mL in plasma or serum, into the time since a drug was smoked or injected, or into a laboratory cutoff. The experiment’s images are evidence about selected participants’ neural responses to a visual task during early abstinence, not proof of their chemical toxicological state at a later hypothetical roadside incident.

Even where a separate case has valid quantitative blood, oral-fluid or urine evidence, the toxicologist must specify the exact specimen matrix, assay, identified analyte, collection time, validation, uncertainty and biological interpretation. The leap from population-average imaging effects to a named person’s driving errors, intention, memory reliability, violence, crash cause or credibility cannot be made from this source. Such determinations require case-specific facts and appropriate validated tests, and sometimes cannot be made confidently at all.

What study features most limit generalization?

The trial involved only sixty men undergoing residential recovery; it was a single session of one tDCS montage with a parallel active-and-sham design. Multiple brain locations and imaging models were explored. An important activation family did not survive the stated FDR correction. The subjective outcome did not show a specific treatment-group interaction. Repeat cue exposure and participant expectancy created plausible sources of non-drug and nonstimulation change. Changes in mood, attention or comfort during a scan may be relevant but cannot be automatically treated as an objectively validated addiction outcome.

Other boundaries concern how predictors are built. The electric field was individualized by computational head models, but electrical-field maps are estimates rather than independent proof of which causal neurons changed. Functional-connectivity indices can change even when self-reported craving does not. The experiment lacked a separately randomized extended treatment course and an external validation sample establishing the sensitivity, specificity or calibration of imaging predictions for real relapse. Generalization to women, older adults, acute intoxication and specific legal incidents remains unsupported by the source’s actual design.

What is the properly limited evidence-based conclusion?

The strongest conclusion is selective and somewhat counterintuitive: one twenty-minute active tDCS session produced a correction-surviving task-dependent frontal-to-parietal imaging connectivity difference, while self-reported craving did not improve significantly more than sham. Five separately highlighted functional-activation clusters were exploratory because their statistics failed FDR correction. The repeat-exposure findings and sham responses explain why within-person improvement should not be confused with treatment-specific efficacy. These conclusions are not contradictory; they arise because neural measurements, subjective experiences and actual clinical behavior are different endpoints.

For scientific and forensic readers, source integrity requires retaining the negative results alongside the imaging signal. The sample size must remain sixty independent men, not hundreds of brain clusters; the source intervention must remain scalp electrical stimulation, not methamphetamine dosing; and BOLD or modeled electric fields must remain physiological proxies, not chemical assays. The original research offers mechanistic hypotheses that justify further careful trials, not a drug blood threshold, validated intoxication biomarker, safe-driving determination, or case-specific causal conclusion.

Seven frequently asked questions

1. Did active tDCS reduce methamphetamine craving significantly more than sham?

No. Participants in both groups reported lower craving at later assessments, but the original mixed-effects analysis found no significant active-versus-sham group effect or group-by-time interaction for the craving endpoint. A time effect at p < .0001 is not proof of a treatment-specific benefit.

2. Did any brain-imaging finding survive correction for multiple testing?

The right frontoparietal task-modulated gPPI connectivity cluster of 330 voxels satisfied the paper’s cluster-level FDR-corrected p < .05 criterion. The five exploratory regional activation clusters did not survive their more stringent FDR correction. They must not be reported as five independently validated clinical biomarkers.

3. Were 330 connectivity voxels the same as 330 study participants?

No. The study randomized sixty independent men, thirty to each arm. The 330-voxel cluster is a set of nearby MRI image locations arising from a statistical analysis. It is not an additional sample of independent people, blood specimens, or treatment events.

4. Did participants receive an experimentally administered dose of methamphetamine?

No. The experimental intervention was 20 minutes of scalp tDCS or a sham control. The participants had histories of methamphetamine use disorder, but researchers did not assign methamphetamine doses and were not measuring acute methamphetamine pharmacokinetic effects.

5. Did the experiment establish methamphetamine concentration in venous whole blood?

No. No quantitative forensic whole-blood concentration, analyte-specific GC–MS/LC–MS method, drug clearance curve, or legally relevant individual blood threshold is reported. Brain BOLD fMRI and modeled electric fields are different measurements with no validated conversion into blood ng/mL.

6. Why did craving fall in the sham group too?

The study itself highlights repeated exposure to drug pictures, habituation, a brief initial current in the sham procedure, and participant expectations as possible influences. These explanations are plausible, not independently proven causal mechanisms. They reinforce the importance of comparing outcomes between randomized arms.

7. Could these results determine whether a particular driver or witness was impaired?

No. The researchers did not perform validated actual driving, witness-memory or incident reconstruction tests and did not chemically establish acute drug exposure. Neither group-level imaging changes nor craving questionnaires establish an individual’s last drug use, legal intoxication, safe-driving capacity, truthfulness or causation of a specific event.

Primary research citation

Ekhtiari H, Soleimani G, Kuplicki R, Yeh H-W, Cha Y-H, Paulus M. Transcranial direct current stimulation to modulate fMRI drug cue reactivity in methamphetamine users: A randomized clinical trial. Human Brain Mapping. 2022;43(17):5340–5357. Original full research DOI: 10.1002/hbm.26007. This explanation independently analyzes the primary Methods, Results, Tables, Figures and Discussion; original publisher figures are not reproduced. The evidence is mechanistic and group-level, not a validated individual forensic toxicology or legal-impairment determination.