Can a Four-Week Cognitive Training App Improve Methamphetamine Decision-Making Without Changing Drug-Cue Attention? A 40-Person Trial

Answer: In a small randomized trial involving 40 men undergoing residential treatment for methamphetamine use disorder, adding 20 sessions of a computerized cognitive-training program was associated with better performance on several laboratory learning and decision-making measures after four weeks. The reported treatment-by-time results favored the training group for verbal list learning, gambling-task choices, simulated risk-taking, and delay discounting. But the study did not demonstrate a treatment-specific change in its methamphetamine-related Stroop attention-bias outcome. Two other cognitive measures were not significantly different between groups. The trial did not measure acute methamphetamine intoxication, real driving, actual relapse prevention, or quantified methamphetamine concentrations in a forensic specimen.

The direct primary source is Zhu and colleagues’ 2018 randomized trial in JMIR mHealth and uHealth, doi:10.2196/10292. The distinction between a change in an experimental task and proof of a particular person’s behavior is essential when this research is considered in forensic toxicology or litigation.

What Was the Actual Scientific Question?

The investigators asked whether a tablet-delivered program combining working-memory exercises with methamphetamine-cue attention exercises could improve cognitive-task performance in men already receiving residential treatment. They were not testing an antidote for acute stimulant intoxication. Nor were they trying to establish a clinical blood-concentration threshold for impairment. The comparison was between usual residential treatment and that same usual treatment supplemented with a structured computerized training schedule.

This matters because the word effect can refer to many different things. A medication effect can be examined using a measured drug concentration and a pharmacodynamic response. A behavioral intervention can be assessed with repeated task scores. A forensic question may instead require the contemporaneous condition of a named person at an identified event. Those questions require different methods and cannot be collapsed into one conclusion.

Who Participated—and What Does N = 40 Mean?

The paper’s participant flow diagram records 56 assessed for eligibility, 16 excluded, and 40 randomized. Twenty were assigned to computerized cognitive addiction therapy (CCAT) plus routine care, and twenty to routine care alone. The diagram shows 20 analyzed in each group, with no reported loss to follow-up. Thus, the independent biological sample was forty adults, not forty training sessions, hundreds of game decisions, or the total number of repeated stimuli.

All participants were men from one compulsory drug rehabilitation center in Shanghai and met criteria for moderate or severe methamphetamine use disorder. They had a history of use and were already in treatment. Mean age was 32.70 years (SD 5.27) in the training group and 35.05 (SD 8.02) in the control group. The groups reported average abstinence periods of 4.30 and 4.10 months, respectively. These historical intervals were not measured as postdose drug-elimination times in a controlled toxicokinetic experiment.

Baseline self-reported methamphetamine-use quantities were 0.60 (SD 0.31) and 0.66 (SD 0.39) grams per day, respectively. These figures described previous use histories; they were not doses administered by the investigators. Neither the quantity reported in an interview nor an average duration since past consumption can be substituted for the methamphetamine concentration in a legally collected venous blood specimen.

Exactly What Intervention Was Compared?

The training group received a four-week program of 20 sessions, five days per week, approximately sixty minutes per session. The tasks ran on an iPad and were delivered in addition to routine rehabilitation. The control group received the standard residential program alone, including scheduled education, physical activity, and vocational activities. The control group did not receive a matched sham tablet intervention of the same duration and engagement.

CCAT included four components: a drug-cue interference task, a drug-related attention-control task, an N-back working-memory exercise, and a spatial memory-matrix exercise. The exercises incorporated images or words associated with methamphetamine use, sometimes as distractions and sometimes as stimuli the patient had to ignore. The research group used these cues to try to retrain automatic attention and strengthen deliberate cognitive control.

The active group completed each training component twice per session. A short relaxation period followed the training because the investigators recognized that drug-related imagery might provoke discomfort or craving. The paper says participants did not report discomfort during sessions, but this is a small, setting-specific safety observation, not a general guarantee about the psychological effects of drug-cue exposure.

Which Outcomes Were Actually Measured?

The scientists administered a collection of laboratory assessments before training and after four weeks. The CogState battery included verbal list learning, spatial paired-association learning, N-back working memory, a maze-learning/error-monitoring measure, and social-emotional cognition. Separate experimental tasks assessed delay discounting, Iowa Gambling Task choices, and simulated balloon inflation decisions. A methamphetamine-word Stroop procedure assessed the speed difference in responding to drug-related versus neutral words.

These outcomes should not be described collectively as a single validated “real-life decision-making score.” Each task has its own instructions, possible practice effects, scoring direction, and inferential scope. The Iowa Gambling Task uses repeated selections among decks with different payoffs. The Balloon Analog Risk Task simulates tradeoffs between accumulating points and risking a balloon’s explosion. The delay-discounting exercise asks participants to choose between smaller immediate and larger delayed hypothetical monetary rewards. None directly records a patient’s actual driving, purchase behavior, workplace performance, or criminal intent.

The same person could complete 150 card choices or 100 balloon trials, but these nested observations do not turn one participant into 150 or 100 independent people. Treatment assignment remained at the person level, with repeated measurements contributing to the tested group-by-time comparisons.

Study Design and Measurement Inventory

Which Cognitive Results Favored Training?

The investigators reported a significant group-by-time interaction for the International Shopping List, with p < .001. Their plotted results showed the active group increasing its number of correct verbal-learning responses, while the control group changed comparatively little. This is evidence of a differential four-week task trajectory, not evidence that one participant was cured of methamphetamine-related cognitive dysfunction.

For Continuous Paired Association Learning, the study’s original figure reports a difference at p = .01. Here the outcome was the number of errors; fewer errors represented better performance. The plotted direction was favorable to the trained participants. The researchers also reported group and time effects around p = .02. The precise hypothesis being tested matters: a main group effect, main time effect, and treatment-by-time interaction are not interchangeable statements.

Not all measures improved. The investigators stated that the Groton maze-learning and two-back tasks did not show a significant between-group change. That is especially important because an intervention may train N-back exercises without establishing transferable improvement on every independently administered working-memory assessment. One cannot use the existence of an N-back training module as proof that the independent N-back evaluation improved significantly.

Social-emotional cognition presented a further nuance. The between-group time interaction was reported significant, but the active participants themselves did not show a statistically significant improvement (p = .56 in the figure); the control group showed a decline (p = .02). A favorable group contrast produced partly by deterioration in the comparison group should not be retold as a direct demonstrated increase in the active group’s social cognition.

Did Participants Make Different Laboratory Risk Decisions?

Yes, the study described significant differential changes in several structured decision tasks. The Iowa Gambling Task showed a treatment-by-time interaction reported at p < .001. The training group’s choices tended to become more favorable under the task’s payoff structure. This finding concerns the arithmetic of laboratory deck choices, rather than a demonstrated reduction in real-world harmful decisions.

On the Balloon Analog Risk Task, the reported group-by-time interaction was also p < .001. Its outcome involved how far participants inflated a virtual balloon before ending a trial. Interpreting a BART score requires the exact task scoring and payoff rules; a changed score should not automatically be called safer conduct in an automobile, less aggression, or a reduced risk of relapse.

The delay-discounting analysis likewise favored CCAT. The source describes negative changes in the logarithm of the discount parameter across multiple offered delays in the training group, consistent with less steep discounting of a future monetary reward in that particular task. The original figure depicts the pattern across delays of two days, one week, one month, three months, six months, and one year. It does not measure participants making actual financial transfers at those future dates, nor does it establish an individualized biomarker of inhibitory control.

These results are noteworthy because decision tasks and memory tasks were both measured. But multiple measured endpoints also create multiple opportunities for statistically significant comparisons. A reader should evaluate which analyses were planned, how repeated measures were modeled, and whether claims remain supported after considering the entire set of reported findings, including negative ones.

What Happened to Methamphetamine-Cue Attention Bias?

The treatment-specific change was not statistically significant. In the methamphetamine Stroop comparison, the paper reported a group-by-time statistic corresponding to p = .34. A time effect at p = .01 was reported, but this did not demonstrate that the active CCAT group improved significantly more than the usual-care group. Those two statistical questions must be kept separate.

The app deliberately presented methamphetamine-related words and photographs as part of the exercises. Accordingly, the study provides a particularly useful caution against equating the intervention’s theoretical target with confirmed success on the independent matching outcome. Adding a cognitive-bias modification module does not establish that the bias itself changed more than under the comparator condition.

This also distinguishes the study from observational eye-tracking experiments. A dot-probe, first-fixation duration, or drug-word Stroop interference score is a different measurement. Results obtained with one procedure cannot be transferred to another without empirical validation. Nor does an attention-bias test provide a validated method for establishing the truthfulness of a witness, the moment someone last consumed methamphetamine, or why a particular collision occurred.

Which Findings Were Significant, Nonsignificant, or Not Measured?

Why Are Drug Dose and Drug Concentration Separate Questions?

No methamphetamine was administered as a study drug. The men’s previously reported consumption histories helped characterize the sample, but the randomized exposure was the training program. The study consequently was not designed to draw a dose-response curve for injected, smoked, swallowed, or insufflated methamphetamine. It supplied no measured parent methamphetamine concentration at the moment of the cognitive evaluations.

Forensic toxicology reports should identify the actual physical matrix: venous whole blood includes cellular and liquid compartments; plasma is separated from cells using anticoagulated blood; serum is recovered after clotting; urine, oral fluid, hair, and postmortem tissues have different collection and interpretive properties. The trial’s experimental answers were recorded behavioral responses, not reported milligrams per liter of parent methamphetamine in those matrices. A preexisting diagnosis or a use-history interview cannot bridge that analytical gap.

Even when a separate legal case contains an accurately measured stimulant concentration, it would require independent consideration of collection time, analytical method and uncertainty, drug history, behavioral evidence, alternative explanations, and jurisdiction-specific rules. This pilot trial does not provide a conversion from a list-learning score or a balloon-task outcome to a whole-blood ng/mL value.

For additional background on why specimen identity and measurement purpose matter, see the published guide to biological matrices in forensic toxicology and the companion guide to confirmatory drug testing.

Could These Results Establish Reduced Methamphetamine Relapse?

No direct evidence of relapse prevention was produced by this four-week assessment. The trial enrolled people undergoing treatment in a controlled residential setting and evaluated cognitive and decision tasks. It did not randomize people to a long-term community relapse-prevention comparison with prospective toxicological verification of drug abstinence. Therefore, a statement that CCAT prevented relapse would require a distinct study and outcome set.

Improving an experimental ability may be clinically interesting. It may also support the rationale for further trials of mobile interventions. But a clinical hypothesis about improved resilience is not equivalent to documenting fewer subsequent methamphetamine-positive specimens, longer confirmed abstinence, reduced emergency department visits, or improved vocational performance. These propositions were not established by the analyzed comparison.

Likewise, one participant’s improved gambling-task score cannot adjudicate whether that person consumed methamphetamine immediately before a charged incident, understood a specific agreement, or formed a legally relevant state of mind. Those are particularized factual questions beyond the randomized task data.

Does No Reported Discomfort Mean the App Is Harmless?

The authors reported that participants did not disclose discomfort during training sessions. They used a short relaxation procedure at the end of sessions to mitigate any possible response to drug cues. This is useful descriptive information, but the sample included only twenty active participants and excluded several groups with complicating medical or psychiatric histories. Some undesirable events may be uncommon, delayed, underreported, or not assessed by a short self-administered form.

A rigorous safety claim would require a prespecified adverse-event taxonomy, active ascertainment, adequate exposure and follow-up, and enough participants to estimate uncommon risks. The paper’s zero reported training discomfort is not proof of zero potential risk or general applicability outside the original rehabilitation context.

What Limits Generalization to Other Patients?

All forty randomized participants were men, treated at a single compulsory residential facility. The setting created consistent access to the intervention, but also differs from unsupervised home deployment. Participants with certain major psychiatric, neurological, or medical comorbidities were excluded. The original research thus does not directly establish the same effects in women, adolescents, outpatients, people with active severe co-occurring conditions, or individuals using multiple drugs at the time of a legal event.

The trial was single-blind: outcome assessors were masked, but people spending an hour with a tablet likely knew whether they were receiving the added intervention. Because the control group did not receive an attention-matched sham activity, participant expectancy, increased staff contact, or practice effects could contribute to the observed between-group change. Random assignment improves causal inference about the offer of the combined program in this setting; it does not isolate each active module or establish that a particular neural mechanism changed.

Twenty participants per arm also limits precision. Multiplicity is relevant because the authors tested many task families and reported repeated contrasts. Several printed statistics express test degrees of freedom unusually; the central result should be evaluated from the complete method and the actual figure captions, not by inventing cleaner statistics or assuming that a significant group contrast is interchangeable with every within-group result.

How Should Attorneys and Expert Witnesses Use This Experiment?

First identify the proposition for which the paper is offered. If it is introduced as evidence that a structured four-week training program may change some cognitive tests in a small residential sample, that is broadly within the investigated design. If it is offered to prove that an individual with a history of methamphetamine use was incapable of driving, was dishonest, was safer after an intervention, or could not have caused an accident, the evidentiary bridge is missing.

Second distinguish general background from case-specific proof. The reported endpoints were laboratory task outcomes before and after treatment, not incident reconstruction. One would need contemporaneous facts and independently admissible evidence to analyze an actual event. Third inspect both positive and negative findings. A source showing better simulated gambling choices but no treatment-specific change in drug-cue Stroop interference cannot be accurately summarized as uniformly improving all impulsive or automatic cognitive processes.

Finally, consider the exact population and exposure. These were residential participants with established MUD histories, not healthy volunteers receiving a controlled acute methamphetamine dose. The comparator was usual care rather than an hour-matched sham program. A defensible account of the science should preserve those distinctions even when discussing the practical promise of computerized rehabilitation.

What Would a Stronger Follow-Up Trial Need?

A larger multicenter trial could enroll a more diverse population, compare CCAT to a credible time-and-attention-matched digital control, preregister a smaller set of primary outcomes, and follow participants after leaving residential care. It could separately assess adherence, cue-provoked distress, real-world functioning, and objectively verified drug-use outcomes. Blinded outcome evaluation and appropriately prespecified multiple-comparison methods would strengthen interpretation.

Forensic concentration-effect questions would require a distinct protocol: actual exposure timing, specimen collection, validated analyte-specific quantitative assays with collection and analytical uncertainty, and an outcome measuring the real capacity or task at issue. The behavioral intervention here was not a drug-challenge study and should not be retroactively used as one. Similarly, an assessment of sustained patient-specific improvement requires longitudinal evidence beyond a pre/post comparison in a small pilot sample.

Frequently Asked Questions

1. Did the training app improve all cognitive tests?

No. The paper reported benefits on verbal list learning and paired-association performance, but also described nonsignificant changes in independent two-back and maze-learning evaluations. The findings are selective rather than a universal improvement across cognition.

2. Was methamphetamine administered during the experiment?

No. Participants had historical methamphetamine use disorder and were in rehabilitation. The randomized intervention was the four-week tablet training program added to ordinary treatment, not a new methamphetamine dose.

3. Did the app eliminate methamphetamine-related attention bias?

No treatment-specific effect was demonstrated by the independent drug-word Stroop comparison: the reported group-by-time p value was .34. A general passage-of-time effect cannot substitute for evidence that one randomized intervention outperformed the other on that measure.

4. Did improved gambling scores prove the men were less likely to reoffend or relapse?

No. The Iowa Gambling and balloon outcomes were computer tasks. The study did not prospectively establish a reduction in methamphetamine relapse, criminal behavior, traffic crashes, or other real-world events.

5. How many independent participants were studied?

Forty men were randomized and analyzed, twenty per group. Repeated cognitive trials, individual cards, balloon decisions, or training sessions should not be counted as additional independent participants.

6. Did investigators determine the participants’ blood methamphetamine concentrations?

No contemporaneous quantitative venous whole-blood concentration-effect analysis was reported. History-based grams-per-day estimates and duration of abstinence are not equivalents of an analytically confirmed parent-drug concentration in a collected specimen.

7. Can this study decide whether someone was impaired in a particular lawsuit?

No. It can inform discussion of cognitive rehabilitation evidence and its limitations, but it contains no individual incident investigation, road-driving test, accident analysis, or validated conversion from task performance to a legally determinative level of impairment.

Bottom Line: What Does This Primary Study Actually Support?

The study supports a preliminary, setting-limited conclusion: among forty men in residential methamphetamine treatment, adding four weeks of CCAT produced more favorable change on several learning and laboratory risk-decision outcomes compared with usual care alone. It also yielded important negative results, particularly no additional change in the targeted methamphetamine Stroop interference measure, alongside nonsignificant results in independent cognitive tasks.

The careful scientific question is not whether computerized rehabilitation is promising—it may be—but exactly which measured functions improved, which did not, and what remained untested. No contemporaneous parent methamphetamine concentration was established in whole blood, serum, plasma, urine, or oral fluid. No actual driving, relapse event, or individual impairment threshold was validated. For related interpretation principles, see the published discussion of timed drug measurements and therapeutic drug monitoring.

Original study: Zhu Y, Jiang H, Su H, et al. A Newly Designed Mobile-Based Computerized Cognitive Addiction Therapy App for the Improvement of Cognition Impairments and Risk Decision Making in Methamphetamine Use Disorder: Randomized Controlled Trial. JMIR mHealth and uHealth. 2018;6(6):e10292. https://doi.org/10.2196/10292.