Does Working-Memory Training Improve Inhibition in Abstinent Methamphetamine Users? A 32-Patient Trial

Answer: In a small randomized study of 32 men undergoing residential treatment after long-term methamphetamine use, 20 sessions of adaptive working-memory updating training improved performance on the practiced n-back exercise and on a separate running-memory test. It did not produce a statistically significant advantage over an active control on the tested Stroop, flanker, go/no-go, or task-switching measures. The distinction is between near transfer to a similar memory operation and unproven far transfer to other executive skills. These laboratory findings do not establish that a named individual was intoxicated, impaired, capable of safe driving, or unlikely to relapse.

The primary evidence is the open-access study by Zhao, Wang, and Maes, Training and transfer effects of working memory updating training in male abstinent long-term methamphetamine users (2021), doi:10.1016/j.abrep.2021.100385. Its practical value lies as much in the outcomes that remained nonsignificant as in the memory improvement.

What exactly did the experiment test?

The question was whether improving an executive-function component called working-memory updating could generalize beyond the practiced activity. Updating involves keeping currently relevant information available and replacing obsolete information as a task unfolds. An adaptive n-back exercise continually asks a participant to compare the current location of a stimulus with its location a selected number of presentations earlier. Running memory instead requires retaining the last few items in a sequence whose length is not disclosed in advance. Both require information updating, even though their interfaces differ.

The investigators went further by measuring executive functions that are conceptually distinct. A Stroop color-word task and an arrow-flanker task indexed interference control. A go/no-go task sampled response inhibition. A digit-classification switching task measured the cost of changing rules and of mixing task sets. If memory updating training strengthened a broad, general executive capacity, improvements might also appear on those assessments. The experiment was designed to test that possibility, not to diagnose acute exposure to a drug.

Who participated, and what do the sample counts mean?

The study enrolled 32 independent male inpatients at a rehabilitation center in Zhejiang, China. Sixteen were randomized to the working-memory training condition and sixteen to an active comparison condition. The article describes long histories of methamphetamine use and several months of previous withdrawal treatment; participants were abstinent at the time of testing. The paper reports that institutional restrictions on substance use were checked using urine tests, but it does not present the analytical method, cutoffs, quantitative metabolite values, or a timed series of samples from which to reconstruct drug elimination.

These are 32 people, not 640 independent people because training was repeated over 20 days, and not thousands of independent observations because tasks involved many computerized trials. Repeated results within a participant are correlated. The two randomized groups are the biological comparison units. All participants were men, so applicability to women or broader community populations requires further evidence. Neither group underwent experimental methamphetamine administration as part of this intervention.

What did the training group do compared with controls?

Trained participants completed 20 daily sessions over approximately 20–21 days. A session lasted approximately 25 minutes and comprised 20 blocks of an adaptive spatial n-back activity. A blue square appeared at one of eight possible grid locations. Participants responded when its position matched the position a designated number of trials earlier, withholding a response on nonmatches. Training began at the two-back level and adapted to individual accuracy: accuracy above 90% increased difficulty, accuracy below 70% lowered it, and intermediate accuracy kept the current level unchanged.

The comparison participants spent equivalent periods making sand paintings, rather than merely waiting without any assigned activity. That active-control design is important because time, staff contact, expectancy, and repeated test familiarity can themselves influence scores. Everyone completed the same battery of executive tasks before and after the intervention. The trial therefore compares an adaptive cognitive practice activity against another structured activity within residential treatment; it does not compare brain training against withdrawal from methamphetamine, because the latter was not experimentally administered.

Did improvement on the practiced n-back task really occur?

Yes. Across the 20 practice sessions, achieved n-back difficulty rose substantially. The original repeated-measures analysis reported a session effect of F(19,285) = 37.18, p < .001, with partial eta squared of 0.71. The linear component was also significant, F(1,15) = 81.01, p < .001, and a quadratic component was reported, F(1,15) = 11.30, p = .004. The plotted progression begins near a two-back level and rises to approximately the five-to-six-back range by later sessions.

These data show that people improved on an exercise they repeatedly practiced under an adaptive rule. They do not by themselves demonstrate restoration of everyday planning, occupational judgment, or resistance to an intoxicant. Practice effects are expected on many trained tasks. The more important scientific challenge is determining how far any benefit extends beyond the trained interface and whether any apparent improvement survives an active-control comparison.

Was there transfer to a different working-memory test?

The strongest transfer finding was on the running-memory task, which was not the practice program. Participants listened to or viewed sequences of digits and had to return the final three in order without knowing in advance when the sequence would stop. Results from versions with different presentation speeds were combined because their transfer patterns were similar. This task shares an updating requirement with n-back but does not require deciding whether a square matched one displayed several trials earlier.

The trained group’s mean proportion correct increased from 0.83 to 0.94; the comparison group’s mean changed from 0.85 to 0.82. An analysis adjusting post-treatment performance for baseline reported F = 13.96, p = .001, partial eta squared 0.33. The Bayesian analysis gave a Bayes factor for the model including treatment group of BF10 = 29.95, described by the authors as strong evidence for a group-related effect. This is evidence of near transfer, not proof that all cognitive systems or real-world functions improved.

Evidence table 1: enrollment, procedure and the two positive findings

Accuracy is a proportion correct, not a percent blood concentration, score of drug effect, or clinical diagnostic probability. The running-memory analysis used the preintervention score as a covariate. A small-sample effect estimate may be unstable even when a reported comparison crosses the chosen significance threshold.

Did training improve Stroop interference control?

No statistically significant differential benefit was demonstrated. Stroop interference is commonly measured by comparing response times for conflicting color-word information against more congruent trials. The original table reports an interference-time difference of 29.42 milliseconds before and 16.30 milliseconds after training. The active-control group changed from 30.55 to 21.65 milliseconds. Both sets of raw means moved downward, which illustrates why merely observing a favorable within-group direction cannot establish a training-specific effect.

The baseline-adjusted between-group comparison was F = 0.17, p = .69, with a reported BF10 of 0.34. A smaller numerical interference cost in the trained group after practice was not sufficient to demonstrate that the n-back intervention caused superior Stroop performance. The outcome is a laboratory response-time contrast, not a standardized assessment of roadside drug impairment or a validated measure of the veracity of a witness’s recollection.

What happened on the flanker task?

The flanker task asked participants to identify a central arrow while surrounding arrows were congruent or conflicting. The trained group improved its mean interference-time contrast from 39.12 to 15.97 milliseconds; the comparison group’s values were 25.23 to 25.39 milliseconds. Those descriptive directions can look compelling when considered in isolation. However, the study’s comparison adjusted for the baseline level, and the between-group difference did not meet its prespecified significance criterion.

For the flanker outcome, the reported adjusted result was F = 1.17, p = .29, with BF10 = 0.51. The appropriate reading is that this experiment failed to establish a specific advantage on that measure. Failure to reject the null is not a mathematical proof of no possible effect in larger trials, different tasks, or people with different histories. It does mean that a firm claim of improved interference resistance would overrun the evidence actually generated.

Did participants gain demonstrable go/no-go inhibition?

A go/no-go task measures performance under rules requiring a response to some signals and response withholding to others. In this experiment the mapping of X and Y signals reversed partway through the procedure. The dependent score reflected the difference between the proportion of go hits and no-go false alarms. A programming error affected the last 100 trials of a 400-trial task for eight people; the researchers excluded that block for those individuals and stated that this did not change the null pattern.

The trained group’s score changed from 0.87 to 0.92; the active control changed from 0.93 to 0.95. The adjusted treatment comparison was F = 1.63, p = .21, with BF10 = 0.76. These findings do not demonstrate that training made participants better at withholding dangerous real-world actions. Computerized withholding of a keypress is a narrow performance variable, and generalizing it to drug-impaired driving, the ability to resist a substance, or criminal intent would require separate validation.

Did switching between rules show far transfer?

Participants classified digits under alternating decision rules: whether a number was larger or smaller than five, or whether it was odd or even. Two contrasts capture different costs. A switch cost compares trials that change task rule against repetitions within a mixed block; a mixing cost contrasts repeated-rule trials in mixed blocks against trials from single-rule blocks. Neither cost is identical to the working-memory updating exercise, though switching also places demands on executive control.

The reported switching-time cost for training changed from 151.96 to 125.91 milliseconds and for controls from 179.96 to 152.55 milliseconds. Its adjusted comparison was p = .59. The corresponding mixing-cost change was 112.28 to 90.06 milliseconds with training and 110.36 to 45.83 milliseconds among controls; its adjusted comparison was p = .21. Both were nonsignificant. The substantial dispersion shown in standard deviations should discourage attempts to interpret a modest change in a mean as a reliable prediction for one person.

Evidence table 2: why positive-looking means are not far-transfer proof

RT means are in milliseconds; hit-minus-false-alarm and running-memory scores are proportions. The tabulated p and BF10 values concern a baseline-adjusted group comparison, not simply the direction of the two raw means. The Bayes factors for the nonsignificant outcomes were mostly described as weak support for the null model, not definitive proof that no effect exists.

How strong is the distinction between near and far transfer?

Near transfer here means a measurable benefit on a different task drawing on a closely related updating operation. Far transfer would require improvement on different executive processes or in functioning outside the practiced format. This trial supports the former and does not demonstrate the latter. Its measured far-transfer outcomes were interference control, go/no-go response withholding, and task-switching; it did not measure every conceivable aspect of executive functioning. Neither general intelligence nor actual professional driving performance was a primary endpoint.

There is also a second caution: an effect on an untrained but similar task does not disclose the mechanism. Participants might have learned a strategy for keeping recent sequences organized, improved a general updating capacity, developed better familiarity with computerized tasks, or benefited through a combination. The researchers examined the relationship between individual training improvement and running-memory transfer. The correlation was essentially zero: Spearman’s ρ = 0.007, p = .98. That null association does not distinguish all competing explanations, but it weighs against a simple linear theory that those who improved most on n-back necessarily improved most on running memory.

How does abstinence affect forensic interpretation?

The participants were undergoing residential rehabilitation after extended histories of methamphetamine use, not undergoing a controlled methamphetamine dose challenge. Reported abstinence and the facility’s urine-check procedures frame the setting. They do not replace a contemporaneous, validated laboratory report for any particular court case. The publication supplies no quantified parent methamphetamine or amphetamine concentrations from venous whole blood, separated plasma, serum, oral fluid, or urine, and it does not present a blood collection time from which to build a pharmacokinetic reconstruction.

An institution’s urine screen may have a role in eligibility or compliance, but such a screen is a different evidentiary object from confirmed quantitative blood testing. For background about matrix specificity, see biological matrices in forensic toxicology, and for the distinction between screening and definitive identification see confirmatory drug testing. Those general references do not add unreported specimen measurements to this trial.

Can these findings validate impairment, relapse, or witness credibility?

No. The experiment did not include observed driving, a validated standardized field sobriety examination, a laboratory drug-concentration response curve, clinical relapse verification, road collisions, or witness perception and memory of a disputed event. It did not demonstrate that running-memory accuracy predicts any of those outcomes. Its measured causal contrast is assignment to a cognitive-training activity versus sand painting, in a small selected cohort undergoing ongoing rehabilitation.

Even a significant laboratory result should be situated within the measurement and intervention actually performed. A stronger running-memory score is not a legal threshold for competence, fitness to drive, degree of culpability, or sincerity. A nonsignificant go/no-go comparison cannot prove a named patient had intact judgment at a particular time. To examine an individual episode, the relevant evidence would include the validated case-specific specimen, collection chronology, medication and exposure history, physiological or clinical findings, and conduct being assessed. The necessary components depend on the actual question; this paper does not supply them.

What limitations should an expert or editor identify?

The first is sample size: sixteen people in each group provide limited precision and leave open uncertainty regarding smaller effects. Second, participation was limited to long-term male abstinent inpatients at one center; generalization to women, shorter histories, untreated individuals, emergency patients, and people acutely exposed to methamphetamine was not tested. Third, the task battery covers specific laboratory constructs, not every possible executive skill. Fourth, the study did not follow the trained patients to determine how long the near-transfer benefit persisted or whether it improved daily functioning, rehabilitation retention, or objectively documented relapse outcomes.

Fifth, repeated assessments themselves can change performance. An active comparison helps separate broad exposure-to-testing effects from a training-specific benefit, but it does not eliminate all expectancy or task-structure differences. Finally, Bayes factors less than one in the several null results generally offer weak—not conclusive—support for the simpler model given this sample. The combination of positive near-transfer evidence and inconclusive far-transfer evidence is more precise than either saying the intervention “worked across cognition” or saying it “did nothing.”

What should follow-up research measure?

A stronger trial could randomize a larger, more diverse group; retain a credible active comparison; prespecify a small number of primary outcomes; and test whether effects persist at later follow-ups. It could employ multiple independent instruments per executive domain and avoid relying on a single go/no-go or switching paradigm. Assessing whether training improves rehabilitation engagement, clinical decision-making, and verified relapse endpoints would add practical meaning beyond a computerized accuracy score. Measuring strategy use could help disentangle transferable capacity from learning a narrow procedure.

If a proposed study were also intended to make pharmacological or forensic claims, it would require a different design, with appropriately consented and timed chemical assays, validated analyte/matrix methods, and independently defined clinical or driving outcomes. Those additional data should not be retrofitted into this trial. The useful next question is whether an intervention that improved running-memory accuracy can produce a durable, reproducible benefit that matters to a person’s real functioning without turning laboratory scores into unsupported forensic labels.

Seven frequently asked questions

1. How many distinct participants were tested?

Thirty-two male residential-treatment patients: sixteen in adaptive n-back training and sixteen in the sand-painting active-control condition. The twenty training sessions and numerous computer trials do not create more independent participants.

2. Did the intervention significantly improve a different memory task?

Yes. Running-memory accuracy increased from .83 to .94 in the trained group while the control moved from .85 to .82. The baseline-adjusted group test was p = .001, with BF10 = 29.95. The evidence supports near transfer between related updating activities.

3. Did it improve Stroop or go/no-go inhibition?

Not at the study’s significance criterion. The adjusted Stroop comparison was p = .69 and the go/no-go comparison p = .21. The flanker and switching outcomes were also nonsignificant. Nonsignificance should not be misreported as a proven zero effect in every possible population.

4. Was methamphetamine administered or quantified in blood?

No experimental methamphetamine was given. The authors reported urine-based checks of treatment-setting abstinence, but not validated quantitative methamphetamine concentrations in forensic venous whole blood, plasma, serum, or urine for constructing a concentration-response relationship.

5. Is a computerized n-back score an impairment test?

No. It measures performance on an adaptive memory exercise. The trial did not validate n-back or running-memory accuracy as a threshold for intoxication, safe driving, occupational fitness, or a person’s actions in a legal incident.

6. Did better n-back practice predict more running-memory improvement?

No significant monotonic relationship was reported: Spearman’s ρ = .007, p = .98. Training and group-level near transfer both occurred, but their individual improvement magnitudes were not linearly linked in this small sample.

7. Did the experiment prove relapse prevention or lasting recovery?

No. The investigators did not supply follow-up evidence of long-term maintenance, objectively verified relapse prevention, collision reduction, or improved community functioning. Those important endpoints were outside the measurements reported here.

Bottom line for scientific and legal readers

The defensible conclusion is narrow and useful: among 32 abstinent male methamphetamine-treatment patients, adaptive n-back training improved a practiced task and a related running-memory assessment, while the randomized comparison did not show statistically significant benefits on Stroop, flanker, go/no-go, or switching tasks. The original study is Zhao, Wang, and Maes, 2021, doi:10.1016/j.abrep.2021.100385. Quantitative blood methamphetamine, individual intoxication, verified relapse, driving and incident causation were not measured. For the broader distinction between monitoring a timed drug concentration and interpreting a behavioral result, the therapeutic drug monitoring guide offers a complementary framework.