Does Methamphetamine Picture-Stroop Bias Track Withdrawal Craving More Than VR Cue Craving? A 134-Man Study

Short answer: In an observational study of 134 abstinent men with methamphetamine use disorder, responses to methamphetamine-related photographs were slower than responses to neutral photographs: 956.58 versus 921.90 milliseconds, respectively (p < .001). The calculated picture-Stroop delay correlated with ratings of ongoing withdrawal craving (r = .277, p = .001), whereas a smaller correlation with virtual-reality-evoked craving (r = .175, p = .043) did not meet the investigators’ Bonferroni-adjusted threshold of .025. In a joint regression, withdrawal craving remained associated with the delay (p = .006), but VR cue craving did not (p = .266). These are group-level associations among computerized reactions and self-reports—not measurements of drug in blood, reliable measures of current intoxication, or individual legal impairment tests.

The original empirical research is Huang, Liao, Wang and colleagues, Relationship between attentional bias and psychological craving in methamphetamine use disorder, Frontiers in Public Health (published January 5, 2026), volume 13, article 1659759. The complete primary study can be read through its original DOI. The distinction between a nominally small p-value and the authors’ planned correction matters just as much as the size of the mean reaction-time difference. Neither a brief task delay nor a strong subjective urge is a chemical result identifying a drug concentration in an actual biological specimen.

What specific scientific question did the investigators test?

The study separated two kinds of reported methamphetamine craving. One was an ongoing urge described during an abstinence period; the other was the urge elicited while viewing simulated drug-related settings through virtual reality. Researchers then compared both craving measures with a third variable: how much longer it took to identify a color when a methamphetamine-associated picture appeared behind the color target. This is a narrowly defined cross-sectional association question, not a trial testing an anti-craving treatment.

A photograph can compete for limited attentional resources without exposing the person to methamphetamine. A virtual-reality scene can provoke a subjective state without supplying a physical sample for chemical analysis. A computer response to the first task and a rating following the second can be related, but that relationship need not be instantaneous or causal. The paper’s most valuable methodological feature is that it assessed these distinct phenomena in the same study population rather than treating craving as a single undifferentiated outcome.

Who was studied, and what is the real biological denominator?

Investigators initially recruited 150 men with methamphetamine use disorder from a compulsory residential drug-rehabilitation setting in Hunan, China. Of those, 137 completed the picture-Stroop task, and three were excluded because their accuracy fell below 90 percent. The resulting analysis involved 134 independent men. No women, short-term abstinent group, voluntarily abstinent comparison cohort, or unaffected control group formed part of this primary analysis. It is not a controlled clinical population representative of every person using a stimulant.

The analyzed men had a reported mean age of 36.40 years (standard deviation 7.69) and a historical consecutive abstinence interval averaging 18.57 months (SD 1.65; range 14–23 months). They reported a mean period of methamphetamine use of 57.49 months and a historical mean amount of 0.70 grams per use episode. Those quantities came from reported history; they were not drug doses administered by investigators or amounts contemporaneously confirmed with chromatography. Repeated keyboard trials, craving ratings, and virtual scenes do not increase the independent human sample above 134.

What does the picture-based addiction Stroop task actually measure?

Instead of asking participants to read drug-related words, the investigators displayed either methamphetamine-related or neutral photographs with a small colored square. Participants pressed one of four fixed keys corresponding to the square’s color. The correct response concerned the color target, not the content of the photograph. A distracting picture could produce slower responding even when the person still identified the color accurately. The central endpoint was the mean correct-response reaction time in milliseconds for each background-image type.

Each session included practice followed by a formal task with 240 presentations, split by a break after 120 responses. Drug and neutral images were balanced, the same color was not shown in long successive runs, individual stimulus exposure could last up to 4,000 milliseconds, and intervals varied between 500 and 1,500 milliseconds. E-Prime 2.0 controlled the display. The difference score was calculated as drug-image reaction time minus neutral-image reaction time for each person. A positive score identifies longer color-response time against drug imagery in this task; it is not a direct measurement of where the participant looked, a validated dose-response relationship, or proof of an uncontrollable impulse.

How were ongoing withdrawal craving and VR-evoked craving distinguished?

Participants rated background craving using a visual analog craving measure during abstinence. For the cue-evoked measure, the researchers first familiarized them with a neutral virtual-reality environment and then presented standardized VR scenes, including drug-associated settings. The score recorded during the VR drug-use scenario served as the cue-craving variable in the subsequent analysis. Such ratings capture the person’s own reported experience at the time; they do not by themselves identify a physiological mechanism, predict future use, or quantify methamphetamine molecules.

Importantly, the color-response test was administered at least three days after the VR assessment. The investigators also used a brief, standardized guided-breathing exercise following VR exposure and offered additional support as needed. Any immediate drug-cue urge may rise and fall over minutes or hours, making a comparison separated by days difficult to interpret as a simultaneous relationship. The separation is particularly relevant when the study finds stronger associations for an ongoing craving measure than for a transitory VR cue response.

How large was the observed methamphetamine-picture delay?

Across the 134 analyzed participants, the source reports mean correct-response times of 921.90 ± 142.58 milliseconds with neutral photographs and 956.58 ± 164.47 milliseconds with methamphetamine-related photographs. Subtracting the two reported group means yields a descriptive delay of 34.68 milliseconds. The paired comparison was statistically significant (t = −6.114, p < .001). The standard deviations are the source’s dispersion values for each condition; they are not uncertainty bounds for the paired mean difference.

A significant group mean does not establish that every participant was slower with drug imagery. Nor does this comparison isolate an acute pharmacological effect. There was no randomized drug administration, and all analyzed participants had a history of methamphetamine use disorder. Distinct processes—picture salience, learned familiarity, context, motor response speed, attention, and situational motivation—can contribute to a task difference. The magnitude should remain in its original units of milliseconds, not be recast as an amount of methamphetamine exposure.

Numerical evidence table 1: human participants, stimulus times, and craving correlations

The table preserves the paper’s denominators and measured variables. Its two correlation tests of the calculated bias score were explicitly subject to a two-comparison adjustment; the raw condition-specific correlations answer different questions and were not evidence that methamphetamine concentration had been measured. Correlation coefficients describe the tendency of observations to vary together across people. They neither supply the amount of drug in a person nor establish a one-person clinical or legal threshold.

Why the Bonferroni correction changes the cue-craving conclusion

The paper states in its statistical methods that the Pearson analysis was evaluated using a Bonferroni-adjusted significance level of 0.05 divided by two = 0.025. At that specified level, the withdrawal craving association with the picture-delay score, r = .277 and p = .001, remains statistically significant. The corresponding VR cue-craving association, r = .175 and p = .043, does not. A sentence claiming two statistically supported correlations without mentioning the adjustment would overstate the evidentiary result.

The failure of .043 to meet .025 does not establish that cue craving and attentional allocation are unrelated in every possible circumstance. The estimated correlation is still positive, with uncertainty arising from sample selection, the rating instrument, the task, and separation in assessment time. The defensible statement is narrower: this dataset did not establish the VR-cue association at the authors’ specified corrected threshold. It should not be converted to evidence of a known zero effect, biological separation of craving systems, or the absence of a phenomenon in short-term abstinence.

Do correlations with raw reaction times tell the same story?

No. The calculation subtracts neutral-image reaction time from drug-image reaction time. A relationship with that difference need not appear when either component is examined alone. The source reports withdrawal craving correlated with raw drug-image reaction time at r = .120, p = .169, and with neutral-image reaction time at r = .010, p = .906. Neither is significant at the conventional .05 level.

The source also reports a nonsignificant association between VR cue craving and raw drug-image reaction time (r = .138, p = .113), and between VR craving and neutral-image reaction time (r = .078, p = .370). The derived difference can reduce some participant-level general-speed variation while introducing its own measurement noise. A significant bias-score correlation therefore should not be misrepresented as a proven general psychomotor slowing effect, or as a highly discriminating test of craving at the individual level.

Did different withdrawal-craving categories behave differently?

The paper grouped withdrawal craving into no craving (score 1; n = 39), low craving (scores 2–3; n = 48), and high craving (scores 4–10; n = 47). These are post-assessment subdivisions of the same 134 people. Within the no-craving category, neutral and drug-picture means were 888.88 and 917.35 milliseconds (p = .006); in the low category the means were 953.94 and 970.55 milliseconds, with a nonsignificant image-condition contrast (p = .073). In the high category the values were 916.56 and 974.87 milliseconds (p < .001).

The derived mean difference scores across withdrawal groups were 28.47, 16.61 and 58.31 milliseconds, respectively. A separate across-category comparison was significant (F = 5.364, p = .006); exploratory least-significant-difference comparisons favored higher bias in the high group versus no craving (p = .032) and low craving (p = .002). Such categorization throws away detail from a ten-point rating and can emphasize thresholds chosen for analysis rather than natural biological boundaries. No diagnostic classification accuracy, external validation or prospective relapse forecast was demonstrated.

What happened when craving was categorized after the VR drug-use scene?

The VR cue ratings were classified as no craving (n = 32), low craving (n = 45) and high craving (n = 57). Picture-condition contrasts were nonsignificant in the no-cue group (p = .088), but significant within low (p = .001) and high (p < .001) cue groups. These within-group results compare responses to photographs inside each category and do not themselves show that bias differs among the three categories.

The more relevant between-group test on the calculated bias scores was nonsignificant: F = 1.413, p = .247. Mean scores rose descriptively from 19.88 to 33.33 to 44.07 milliseconds across no, low, and high cue-craving groups, but the source did not support that ordering as a reliable between-category contrast. Selecting only the significant within-group tests while omitting the nonsignificant across-group test would improperly turn a descriptive pattern into a confirmed group difference. Moreover, these VR ratings preceded the Stroop task by at least three days.

Numerical evidence table 2: adjusted models, genuine nulls, and group differences

The source presents its regression coefficients with a beta symbol but values such as 8.135 are not plausible standardized coefficients. This discussion therefore describes them as source-reported coefficients without assigning a standardization, a causal slope, or an individual diagnostic interpretation that the publication does not adequately establish. The adjusted results are also sensitive to how the ten-point scales behave and how the pictured task difference was computed.

Why does the two-predictor regression matter more than an isolated nominal correlation?

Two craving scores may overlap. If individuals with higher withdrawal ratings also report more VR-evoked urge, a simple correlation between either score and task slowing can include some shared variation. The investigators therefore fitted a multiple regression that entered both craving measures simultaneously. The overall model was significant, F(2,131) = 6.130, p = .003. Withdrawal craving retained a reported coefficient of 8.135 with p = .006, while the VR cue-rating coefficient of 2.643 was nonsignificant at p = .266. The reported variance-inflation factor was 1.108 for each predictor.

Conditional association does not establish that withdrawal craving causes picture-related delay or that VR cue craving has no effect on attention. Temporal ordering is uncertain, measurement dates differed, and confounders were not eliminated by the two-variable model. The data do not provide an externally validated individual prediction equation. In particular, one cannot translate a given craving rating into a precise expected reaction delay with legally defensible error limits, let alone into a measured dose of methamphetamine or an assessment of impairment during an unrelated event.

Why do three days between the VR task and Stroop test matter?

Cue-induced craving is inherently context dependent. A simulated drug-use room might elicit a strong urge during exposure and little urge at the next appointment; the reverse could occur following a real-life reminder. The investigators scheduled the picture-based Stroop at least three days after the VR session and administered brief relaxation exercises following VR. The resulting measurements are not simultaneous observations of a single fleeting state. This alone offers a plausible noncausal explanation for a weaker cue-score association.

The study did not randomize assessment spacing to estimate that bias, did not continuously follow craving for three days, and did not pair repeated Stroop tests immediately after the same VR scene. Consequently the work cannot identify a time constant for craving decay, a biological half-life for a subjective impulse, or a person-specific interval since stimulant consumption. The three-day gap is a methodological interpretive issue—not a demonstrated pharmacokinetic elimination interval. A future within-day repeated-measure experiment would address a different, more tightly synchronized hypothesis.

What is—and is not—an independent observation in this experiment?

There were 134 analyzed people, each contributing repeated image responses. The 240 trials increase measurement resolution within a person but do not create 240 independent human subjects. Similarly, partitioning the same people into three withdrawal categories or three VR categories does not create two separate biological cohorts. Tests using task averages, correlations, group categories and a two-predictor regression are multiple analyses of related underlying measurements. Treating all those statistics as independent confirmations would exaggerate replication.

A group-level reaction-time mean is not the distribution of within-person paired differences. Without the complete paired raw data, it is not legitimate to recompute participant-level classification performance or infer how many individuals slowed by more than a chosen number of milliseconds. None of these study statistics comes with a validation study of screening sensitivity, specificity, false-positive rates, known forensic cutoffs, or admissible inference about a specific accused person. Those questions require their own design and sampling plan.

Was methamphetamine administered or chemically measured?

No experimental dose of methamphetamine was administered. The nominal amount of 0.70 ± 0.54 grams per use episode appears in the retrospective history. It must not be confused with a controlled dose given at the test or a laboratory result. The study reports no quantitative parent methamphetamine or amphetamine measurement by GC–MS, LC–MS/MS, or another validated chemical assay in a contemporaneously collected legal human venous whole-blood sample. It did not publish a drug chromatogram, specimen collection time tied to the task, calibration range, limit of detection, limit of quantitation, or ng/mL measured concentration for any participant.

It likewise did not report matched quantitative separated plasma or serum drug concentrations, urine, oral fluid, hair, sweat, vitreous, milk, breath aerosol, or tissue results for forensic interpretation. A video scene is not a physical specimen, a VR rating is not an analyte, and a keyboard latency is not a drug amount. Nor can a historical abstinence interval be converted into a known contemporaneous drug concentration without actual sampling, analytical data and scientifically supportable time relationships. The only direct numeric milliseconds are task outcomes, not blood-toxicology results.

Why should forensic interpretation distinguish whole blood from plasma and serum?

Even if a future investigator collects biological specimens, the physical matrix will matter. Legal venous whole blood includes cells and liquid components, whereas separated plasma and serum are processed fractions with different collection and distribution properties. Urine primarily provides information about excretion or the presence of an analyte or metabolite, not a direct whole-blood drug concentration at a particular earlier event. Oral fluid, sweat patches, hair and vitreous each answer other questions and require their own validated methods and interpretive cautions.

No intermatrix conversion ratio can be inferred from this paper, because it did not produce any of those measurements. For a general review of why specimen choice and analytical purpose must remain linked, see the independently published biological matrices guide. The guide supplies contextual definitions, not missing evidence for these 134 participants. An assay result from some unrelated population could not retroactively transform their recorded Stroop latencies into drug levels.

Is longer reaction time evidence of intoxication or unsafe driving?

Not by itself. This experiment did not put anyone behind the wheel, conduct an on-road test, measure lane position, document braking behavior, assess divided attention under traffic load, record crashes, or determine whether anybody was acutely pharmacologically affected during the computer task. The stimuli consisted of background pictures and a color-response target in a controlled room. A mean increase of 34.68 milliseconds on this task cannot be converted to stopping distance, blood concentration, legal driving impairment, or a particular motorist’s fitness to drive.

There was no contemporary comparison with unaffected controls that could isolate a history-of-use group effect, and no blinded chemical exposure verification in the published outcomes. The lack of acute drug measurements is especially important because the sample reported roughly a year and a half of abstinence. A finding in an abstinent selected rehabilitation group should not be generalized as a validated intoxication test for a newly exposed driver, pilot, witness, employee or patient. Legal case conclusions demand contemporaneous evidence and task-specific validation that this publication does not supply.

Could a toxicologist estimate when an individual last used from these findings?

No. The study contains retrospective group histories of methamphetamine use and the duration of abstinence, not drug disposition profiles. No documented acute dose, collection series, half-life estimation, elimination slope, verified formulation, actual whole-blood concentration, or paired excretion data were obtained. Craving and cue processing can persist for reasons unrelated to detectable circulating parent drug. A score of five on a visual analog craving instrument does not timestamp use any more than a thirty-five-millisecond reaction difference does.

A valid discussion of last-use timing would require a clearly identified analyte, validated assay and matrix, time of specimen collection, relevant pharmacokinetic uncertainties, and independent case circumstances. The present paper does not meet any of those requirements. Readers should also not substitute a generic testing detection window, symptom duration, or an average clinical half-life for the missing actual sample data. Those would be outside assumptions, not empirical findings from Huang and colleagues.

What strengths support use of the study as scientific background?

The authors described an objective computerized latency instrument, distinct drug-related and neutral conditions, an explicit formula for the bias score, concrete screening exclusions, a prespecified multiple-comparison correction for correlations, and a regression separating two subjective craving dimensions. Presenting these features with the true nonsignificant outcomes provides a more useful scientific account than a general claim that methamphetamine changes attention. Its comparatively large within-group sample of 134 also offers a more stable description than an anecdote from one participant.

Yet precision and external validity remain different questions. Repeated tests of related outcomes were reported, subgroup categories were derived from ratings, the adjusted regression was observational, and the paper supplied no independent replication cohort. The authors acknowledged limits arising from men only, compulsory rehabilitation, long-term rather than short-term abstinence, and noncontemporaneous VR and Stroop testing. Even a well-executed cognitive experiment does not automatically validate an evidentiary measurement, an expert conclusion about a defendant, or a causal explanation of a real-world accident.

What would a stronger follow-up experiment need?

To test whether the VR state actually drives immediate attentional change, investigators could measure cue craving immediately before and after a standardized scene, interleave a preplanned picture-Stroop test, and repeat observations after controlled delays. Including appropriate comparison groups, women, short-term and voluntary abstainers, and replication participants would help determine whether the pattern generalizes. Analyses should specify primary outcomes and multiplicity adjustments in advance, quantify uncertainty around estimates, and distinguish nested within-person trials from the independent biological sample.

A separate forensic investigation, if it sought drug exposure rather than task associations, would need validated sampling and quantitative drug analyses of actual specified specimens. If driving impairment were the question, measured driving outcomes, appropriately designed exposure contrasts and independent validation would be indispensable. A craving treatment trial would instead require longitudinal clinical outcomes and suitable control conditions. These are different research programs; none of them can be answered merely by adding an appealing interpretive headline to the present cross-sectional study.

How should experts quote the positive and negative results together?

The faithful scientific formulation is: methamphetamine-related photographs were associated with significantly slower correct color-response times in 134 long-term-abstinent men; calculated bias was associated with withdrawal craving after the planned two-test correction and in the adjusted model; the smaller VR cue-craving correlation failed that correction and the VR term did not independently predict bias in the joint regression. Between-category withdrawal differences were significant, whereas between-category VR cue differences were not. This formulation does not equate statistical nonsignificance with proof of exact zero.

A careful case review should identify the sample, measurement dates, tasks, units, direction of the actual comparisons, multiple-testing threshold, absent matrices, and the absence of individual event outcomes. The source can illuminate an experimental distinction between ongoing and cue-evoked craving. It cannot establish dose, recent use, intoxication, person-specific ability to drive, truthfulness of testimony, competence, or a causal explanation of a specific injury or incident. Where drug identification is actually at issue, the independently published confirmatory testing guide provides general laboratory context; it does not supply chemical results missing from this experiment.

Frequently asked scientific questions

1. Did the study prove that methamphetamine was still in the participants’ blood?

No. The paper reports no quantified methamphetamine or amphetamine in contemporaneous legal human venous whole blood or separated serum or plasma. Historical reported abstinence and a reaction-time difference are not chemical assays and cannot establish a detectable parent compound.

2. Were there 134 people or 240 people in the study?

There were 134 independent analyzed men. The formal Stroop included 240 stimulus presentations within each participant, with repeated measurements contributing to individual reaction-time averages. Counting task trials as separate biological subjects would inflate the denominator.

3. Was the relationship with VR-induced craving statistically significant after correction?

No. Its unadjusted Pearson result was r = .175, p = .043, which did not meet the authors’ prespecified Bonferroni threshold of .025. The jointly adjusted VR coefficient was also nonsignificant, p = .266. Neither null test proves the absence of any effect in other settings.

4. Did withdrawal craving show the same null result?

No. The withdrawal craving association with the calculated picture-delay score was r = .277, p = .001, which passed the specified .025 correlation threshold. Its source-reported joint regression coefficient was 8.135 with p = .006, although that observational association cannot establish causality.

5. Why is the three-day interval between VR and Stroop important?

Cue-induced urge can be transient and context dependent. VR craving was assessed at least three days before the picture-Stroop task. This separation means the two tests did not measure the same short-lived state simultaneously, and the study did not quantify how the urge changed during the intervening days.

6. Can the result distinguish a drug-impaired driver from a sober driver?

No. There were no directly measured driving tasks, contemporaneous drug concentrations, legal whole-blood analyses, classifier sensitivity/specificity tests, or validated individual driving cutoffs. The experiment studied color-key reactions to photographs in selected long-term-abstinent men.

7. Can the paper establish that craving caused a relapse or particular incident?

No. Its design was cross-sectional, with no randomized craving manipulation establishing causality and no case-level incident or prospective relapse outcomes. It describes associations between reported craving and a computerized task, not a causal finding about any one person.

Bottom line for forensic toxicology

The scientifically defensible result is specific. In 134 analyzed long-term-abstinent men with methamphetamine use disorder, methamphetamine photographs increased average color-key response time; withdrawal craving was the more robust correlate of that calculated delay, while the VR cue-craving findings failed key corrected or adjusted tests. The design included genuine null results, a three-day or longer mismatch between measurement sessions, and neither acute drug administration nor contemporaneous chemically assayed drug concentrations. Nothing in these group measurements is a validated individual test for intoxication, last use, driver impairment, witness credibility, relapse, or the cause of a particular legal event.

Full original empirical source: Huang Q, Liao Z, Wang X, Wang W, Chao L, Nie Y, Peng S, Zhao L, Shen H, Qi J, Chen X. Relationship between attentional bias and psychological craving in methamphetamine use disorder. Frontiers in Public Health. 2026;13:1659759. Read the primary published research (DOI 10.3389/fpubh.2025.1659759). This discussion is an independent synthesis of the reported numerical endpoints, not reproduction of publisher tables or graphics.