Does Tai Chi Reduce Methamphetamine Cue Bias More Than Running? Evidence From 44 Men

Quick answer: In a 12-week randomized exercise comparison involving men undergoing residential methamphetamine rehabilitation, Tai Chi produced a lower drug-cue attention-bias score and a faster separate choice-reaction measure than treadmill running. These were group-level differences among the 44 men with analyzable follow-up data, not proof that Tai Chi eliminated craving, prevented relapse, or restored fitness for a particular legal task. The comparison on self-rated drug craving was not statistically significant. The trial measured blood-derived metabolic indicators, but it did not measure methamphetamine or amphetamine concentration in forensic whole blood, serum, or another specimen. Its results do not identify acute intoxication or explain a specific person’s driving or incident.

The primary study is Zhang, Zeng, Sun, Xue, Gao, and Zhu (2024), Effect of Tai Chi Compared to Running on Drug Cravings, Attention Bias, and Physical Fitness in Men with Methamphetamine Use Disorder, Healthcare 12(16):1653, doi:10.3390/healthcare12161653. The study is useful because it permits a narrow comparison between two supervised exercise programs while showing how easily cognitive-task improvement can be mistaken for a toxicological or legally relevant endpoint.

What question did the researchers actually test?

The investigators asked whether a structured Tai Chi program would improve craving, attention to methamphetamine-related visual cues, and selected fitness measures more than a program of running at a comparable moderate exercise intensity. They did not compare exercise against no treatment. They did not test a forensic screening instrument, administer methamphetamine experimentally, determine a pharmacokinetic curve, or enroll drivers following a collision. The scientifically supported comparison is therefore Tai Chi versus running during rehabilitation, not Tai Chi versus untreated methamphetamine use or Tai Chi versus a proven drug-based intervention.

The two exercise programs shared considerable time and activity. Each lasted twelve weeks, with three sixty-minute sessions each week and an intended exercise intensity of about 65–75% of maximum heart rate. The distinctive intervention was the 24-form Tai Chi sequence, combining movement, postural control, and sustained attention, rather than treadmill running. A difference after treatment cannot isolate mindfulness, instructor contact, sequence learning, motor coordination, expectations, or other elements of the combined program.

Who participated, and why does the denominator matter?

Forty-eight men with methamphetamine use disorder were randomly assigned: 24 to Tai Chi and 24 to running. Four participants assigned to Tai Chi could not complete the postintervention evaluation because of scheduling conflicts with other rehabilitation activities. The researchers’ comparative tables therefore describe 20 Tai Chi participants and 24 running participants, 44 men in all. That is the follow-up analysis population, not 48 fully observed participants. Those numbers cannot be multiplied by the number of exercise sessions, reaction-time trials, visual cues, or measured physiologic variables to enlarge the biological sample.

The men were recruited within a mandatory isolation or detoxification rehabilitation setting in Shaanxi Province, China. Their institutional schedule and diet were relatively controlled, a feature that can reduce some environmental variation but also limits direct generalization to voluntary outpatient programs. Mean ages in the analyzed groups were about 38 and 40 years. Both groups reported years of prior methamphetamine use, but the study did not establish identical use histories, abstinence duration, concurrent behavioral therapy, or baseline risk of future relapse for every participant.

The source describes random assignment; however, allocation alone does not guarantee that expectation, supervision, or adherence differed only by exercise modality. Moreover, the four missing follow-up participants were all allocated to Tai Chi. Any effect estimated from completers warrants caution because the analysis no longer preserves all randomized participants. This is particularly relevant when a reported difference is small relative to the broad person-to-person variation in attention measures.

How was attention to drug cues measured?

Attention bias was not inferred from blood chemistry. Participants completed a visual dot-probe paradigm with images associated with methamphetamine use and neutral images. A brief display and subsequent probe required responses by computer key, producing reaction-time information that the investigators used to derive an attention-bias score. A lower group score in this paradigm can indicate less preferential processing of a cue under the task’s scoring convention. It is not a drug concentration, a biological marker of use, or an observation of how a person would behave when confronted with an actual drug-related situation.

The analysis makes an important distinction between a derived bias score and the underlying reaction times to individual image categories. A significant difference in the composite does not logically entail that every component mean differed significantly. Nor is one laboratory response to photographs equivalent to continuous surveillance of attention, thought, or temptation outside the test session. Visual stimuli, learning, fatigue, test familiarity, and baseline response style can all influence such tasks.

This distinction matters for legal interpretation. An attention task might help describe a group difference in selective processing after an intervention; it cannot determine whether an individual was exposed to methamphetamine on a particular date, whether a witness remembered an event accurately, or whether a driver could respond safely to an unforeseen roadway hazard. Each of those claims would require separate measurements and case-specific evidence.

What changed in the visual dot-probe results?

At baseline the two groups’ mean attention-bias scores were close: 4.08 in the Tai Chi group and 5.94 in the running group. At twelve weeks the Tai Chi group’s mean was −3.56, whereas the running group’s mean was 13.82. The investigators’ adjusted group comparison was F(1,43)=6.023, p=.019, with a reported effect size of d=−0.42. That result supports a difference in the study’s calculated cue-bias measure between exercise programs after adjustment. It does not establish a zero or universally normal attentional response for every Tai Chi participant.

By contrast, the individual category-reaction-time comparisons did not reach the study’s statistical threshold. The neutral-picture comparison gave p=.232, and the drug-picture comparison p=.887. These are genuine nonsignificant outcomes, not omitted findings. The pattern is fully compatible with a difference in a derived within-task contrast despite broad overlap in the individual group response-time distributions.

The standard deviations for the cue-derived measures are large compared with the means. These data are group summaries with dispersion, not subject-level prediction intervals or sensitivity-and-specificity estimates. The attention-bias score should not be presented as a validated forensic diagnostic threshold simply because its group contrast met p<.05. The individual non-signal and signal images, response mapping, participant familiarity with the task, and image-selection protocol would all be important to replicate the finding.

Did participants respond faster on the separate choice-reaction test?

Yes, a separate electronic choice-reaction-time measure favored Tai Chi. In the completed Tai Chi group the reported mean was 0.54 seconds before training and 0.52 seconds afterward. In the running group it changed from 0.54 to 0.57 seconds. The adjusted comparison was F(1,43)=6.181, p=.017, d=−0.72. On the surface this suggests a relative advantage of roughly five hundredths of a second between the two groups’ changes, although the study’s inference rests on its adjusted statistical model rather than this simple arithmetic alone.

This choice-reaction task is not the same outcome as the response time to drug photographs in the dot-probe procedure. It uses a different setup and addresses speed of an instructed reaction to a controlled stimulus. It does not encompass divided attention in traffic, perception of unusual hazards, decision accuracy under stress, balance, executive control in a complex scene, or a driver’s actual visual search. A faster choice reaction in a laboratory is one piece of cognitive-performance evidence, not a global certification of psychomotor safety.

What happened to craving—and why is a trend insufficient?

Participants rated drug craving on a visual analog scale. The researchers described a tendency for craving to decrease after Tai Chi relative to running, but the adjusted comparison was F(1,43)=0.023, p=.880. By conventional criteria, this study did not establish a statistically reliable difference between the exercise programs in self-reported craving. The reported direction of change cannot be promoted to evidence that the intervention prevented future use.

Craving is a subjective report at the time of questionnaire administration. It is neither a measured methamphetamine concentration nor a verified absence of use. Even a larger statistically significant decline in craving would not automatically establish fewer subsequent uses, fewer legal violations, lower crash risk, or a change in a particular person’s intent. This study does not provide a measured relapse comparison as its primary endpoint, and it did not validate a predictive rule connecting the dot-probe score with later confirmed methamphetamine use.

Were other physical or metabolic outcomes affected?

The study measured anthropometric, strength, blood-pressure, reaction, and serum-biochemistry variables. Most adjusted between-group results were nonsignificant. There is, however, one important finding that a headline focused only on cognition could conceal: waist circumference increased more in the Tai Chi group than in the running group. The Tai Chi mean rose from 85.70 to 90.95 cm, whereas the running mean rose from 84.79 to 86.08 cm. The adjusted group comparison was F(1,43)=7.433, p=.010. That observation should not be characterized as an overall favorable cardiometabolic improvement from Tai Chi. The study does not show that Tai Chi caused clinically meaningful adiposity changes in each participant, but the direction is not a straightforward health benefit.

Other examples are informative because nonsignificant results constrain how broadly the intervention can be described: triglyceride comparison p=.052; systolic blood pressure p=.564; diastolic pressure p=.248; grip strength p=.786; glucose p=.401; and insulin p=.476. A p value of .052 does not become significant through rounding or by describing the result as nearly positive. Neither do multiple nonsignificant results prove the two programs were biologically identical.

The waist result, choice-reaction result, and bias score all deserve attention. Selectively quoting only the favorable cognitive contrasts would present an incomplete account of the experiment. Conversely, an isolated unfavorable anthropometric result does not invalidate the separately measured attention effects. Each endpoint has its own instrument, sampling variability, study model, and evidentiary meaning.

Blood was collected—but what did the laboratory actually measure?

The investigators collected fasting venous blood for clinical biochemical measurements and separated the material for serum-related analyses. Those assays addressed variables such as glucose, insulin, and lipids. They did not report quantitative methamphetamine or amphetamine parent-drug concentrations from venous whole blood, serum, plasma, or urine. The distinction is central: saying a study includes a blood draw is not the same as saying it measured the drug of forensic interest.

An endocrine or metabolic test responds to a different scientific question from a toxicological confirmation. A glucose result expressed in mmol/L cannot be converted into a methamphetamine concentration expressed in ng/mL. Heart-rate monitoring during exercise likewise provides information about exertion and program intensity, not direct identification of a psychostimulant. Nor is a history of methamphetamine use a substitute for an analytically characterized specimen from a specified collection time.

For background on why a report must identify the physical compartment, see the site’s existing guide to biological matrices in forensic toxicology. If an actual case requires drug identification, testing would need a scientifically validated method designed for the intended substance and specimen, with known collection timing, calibration, and controls. The site’s published confirmatory drug testing guide describes that fundamentally different evidentiary undertaking.

Why cannot a dot-probe score be treated like a toxicology concentration?

A dot-probe score is computed from a participant’s task behavior. It depends on what images appear, how their spatial positions change, how the probe is displayed, and how reaction-time differences are summarized. A forensic drug result instead depends on a physical specimen, target analyte identity, preparation, validated analytical response, and units. The two quantities are not interchangeable, and neither a cue-attention score nor a serum cholesterol value can stand in for a validated drug assay.

This also means that no concentration-response curve was established here. The participants were not administered increasing experimental methamphetamine doses, and there are no matched contemporaneous parent-drug concentrations with which to correlate attentional scores. The source cannot determine whether the men who had the lowest bias scores had the lowest methamphetamine concentrations, because such concentrations were not reported. It cannot estimate a toxicology cutoff for recognizing a particular cognitive profile.

Does a better group average establish that every participant benefited?

No. The principal posttraining bias score had a standard deviation of approximately 23 in the Tai Chi group and 22 in the running group. These distributions imply appreciable individual variability around the displayed averages. One group mean cannot be assigned to all group members. A significant group difference does not disclose whether a particular person improved, worsened, or remained unchanged, absent that participant’s measured baseline, follow-up, scoring record, and measurement reliability.

Nor is it enough to compare baseline and endpoint means without regard to missing observations. Four Tai Chi-assigned men lacked follow-up data; their outcomes are not present in the 44-person final comparison. A complete intention-to-treat analysis with transparently handled missing observations could produce a different estimate. The study’s observed result remains scientifically useful, but should not be described as if all 48 randomized men completed the intervention and were proven to improve.

Could the controlled rehabilitation setting influence the findings?

Yes. Residential restrictions may constrain access to methamphetamine, meals, outside stressors, and daily routine compared with an ordinary community setting. They can make attendance more predictable while narrowing the types of people and situations to which an effect generalizes. The regimen also exposed both groups to exercise, supervised time, regular social contact, and program expectations. Accordingly, the study cannot isolate the effect of adding any exercise to a completely inactive lifestyle, much less establish the causal role of a specific meditation component within Tai Chi.

Institutional recruitment also raises questions of selection, motivation, and consent context, even though the paper states that participants agreed to the study and an ethics committee approved it. Those considerations do not erase the observed statistics; they inform whether the statistics can be transported to a different population. The original sample consisted of men, not women or adolescents. It was not a general adult community sample, a clinical trial of medical treatment alone, or a cohort selected by forensic blood concentration.

Which findings were significant, which were null, and which were never studied?

Keeping these categories separate prevents a common interpretive error. The Tai Chi-versus-running attention-bias score comparison was statistically significant (p=.019). The separate choice-reaction-time comparison was statistically significant (p=.017). The waist-circumference comparison was also statistically significant (p=.010), although its direction cannot be marketed as universally favorable. By contrast, the drug-craving scale comparison was not significant (p=.880), the direct neutral and drug-image reaction-time comparisons were not significant (p=.232 and .887), and multiple clinical-chemistry and strength comparisons were nonsignificant.

Different still are endpoints that the study did not measure: active drug concentrations in a forensic specimen, a validated drug concentration-to-task relationship, a verified date of last ingestion, formally adjudicated relapse after treatment, real-world driving performance, collision involvement, eyewitness accuracy, or a specific adverse event caused by any participant. An unmeasured outcome has no meaningful study p value. Describing it as nonsignificant would falsely imply that the investigators actually tested it.

Why do several measured outcomes make interpretation more difficult?

The investigators evaluated attention, craving, choice reaction, anthropometrics, strength, blood pressure, lipid chemistry, and other endpoints in a modest sample. More endpoints create more opportunities to see an isolated small p value by chance, especially if the analysis plan does not specify a stringent familywise adjustment. The reported p values should be preserved accurately, but interpreted in light of the number and dependence of tested measures, follow-up loss, and the magnitude and precision of each estimated group difference.

The two statistically favorable cognitive comparisons are worth follow-up precisely because they are narrow and replicable. A subsequent adequately powered trial could prespecify the principal cue-bias contrast, measure adherence, blind outcome assessment where possible, publish the individual score distributions and confidence intervals, and track independent verified use outcomes. That would test whether the observed benefits generalize. It would not retroactively turn the current task scores into forensic analyte measurements.

What would stronger forensic evidence require?

For a drug-exposure question, stronger evidence would start with a specified collection of whole blood, plasma, serum, urine, or another appropriate matrix, alongside an analyte-specific validated procedure and defensible collection chronology. The relevant records would include specimen identity, preservation, chain of custody where applicable, calibration, controls, detection and quantitation characteristics, and any circumstances affecting interpretation. The article’s therapeutic drug monitoring primer provides complementary context about why time and specimen type matter, although therapeutic monitoring does not itself decide a legal impairment question.

For a rehabilitation-outcome question, investigators would need an independently defined relapse endpoint, consistent retention tracking, and preplanned follow-up. For a driving or incident question, one would additionally need contemporaneous observation or validated performance assessment relevant to the event, without pretending that a dot-probe experiment supplies the missing information. A properly verified concentration might support exposure; it would still not automatically prove impairment, intent, reliability of testimony, or accident causation.

How should clinicians, toxicologists, and attorneys use this study?

Clinicians may view the trial as preliminary evidence that a structured Tai Chi program can outperform an equally timed running program on a particular drug-cue bias score and choice-reaction measure in a supervised male rehabilitation population. The nonsignificant craving difference should temper any claim that Tai Chi independently prevents use. The waist-circumference result reminds readers to examine all measured outcomes rather than merely favorable ones.

Toxicologists should keep exposure and effect separate. The trial contains no quantified methamphetamine test result to interpret, and its serum metabolic tests do not fill that gap. Attorneys assessing an expert claim can ask whether the claimed conclusion relates to a measured endpoint, whether the denominator is 44 completers or 48 randomized participants, whether response-time and bias-score measurements were conflated, and whether the witness is presenting a group-level rehabilitation result as proof about an individual event. These questions are more useful than treating a single p value as a universal forensic rule.

Frequently asked questions

1. Did Tai Chi significantly lower methamphetamine-related attention bias compared with running?

Yes, within the final analysis of 20 Tai Chi and 24 running participants, the study’s derived dot-probe bias score favored Tai Chi, with an adjusted comparison of p=.019. It is a specific task result, not a validated methamphetamine exposure or impairment test.

2. Did Tai Chi significantly reduce drug craving?

No. The between-program visual-analog craving comparison gave p=.880. The authors described a downward tendency, but their data did not establish a statistically significant advantage over running or prove a reduction in objectively verified relapse.

3. Were all 48 randomized men included in the follow-up analysis?

No. Four men assigned to Tai Chi lacked the posttest. The study tables report 20 Tai Chi and 24 running men, for 44 analyzed individuals. Repeated exercise sessions and computer responses do not create new independent participants.

4. Does a lower dot-probe attention-bias score mean faster reaction time to drug pictures?

Not necessarily. The derived bias score differed significantly, but the separate drug-picture reaction-time comparison was nonsignificant (p=.887), as was neutral-picture reaction time (p=.232). The separate general choice-reaction-time test was significant (p=.017).

5. Did the researchers measure methamphetamine concentration in blood?

No quantitative methamphetamine or amphetamine measurement was reported. Venous blood was collected for clinical metabolic biochemistry, which is a different assay target. Neither a serum glucose or lipid measurement nor exercise heart-rate monitoring provides a parent-drug concentration in forensic whole blood.

6. Did every physical-health measurement favor Tai Chi?

No. Most adjusted fitness and metabolic comparisons were nonsignificant, and waist circumference rose more in the Tai Chi group than in the running group (p=.010). A favorable cognitive outcome should not obscure a different or unfavorable outcome in the same study.

7. Can these results prove that a particular person was sober, unimpaired, truthful, or not responsible for an incident?

No. The trial did not test individual acute methamphetamine concentrations, real driving, eyewitness credibility, or event causation. Its group-level rehabilitation-task outcomes do not determine a person’s drug exposure, impairment at a specific time, or responsibility for any event.

Bottom line

In the analyzed 44-person comparison, twelve weeks of Tai Chi versus running was associated with a lower methamphetamine-cue dot-probe attention-bias score and faster performance on a separate choice-reaction task, but not with a significant difference in visual-analog craving. The study also found a greater increase in waist circumference with Tai Chi and mostly nonsignificant clinical-chemistry outcomes. This combination of positive, null, and potentially unfavorable endpoints supports a limited rehabilitation research inference—not a universal assertion of recovery, a chemical toxicology finding, or an individual legal impairment conclusion. The complete original primary study is available at Zhang et al., 2024, DOI 10.3390/healthcare12161653.