Can Resting EEG Microstates Identify Cognitive Differences in Methamphetamine Use Disorder? An 82-Person Study

Short answer: In a 2020 observational study, 55 adults with methamphetamine use disorder had shorter resting EEG microstate A and B durations than 27 healthy controls, while some measures of the time represented by microstate classes B and C also differed. The groups differed on several error-based cognitive tasks, and selected EEG characteristics were associated with reported methamphetamine-use frequency and task performance. These are group-level associations, not a blood test, a validated diagnostic signature, or proof that any particular person was acutely intoxicated or impaired.

The original empirical report is Chen and colleagues, Disrupted brain network dynamics and cognitive functions in methamphetamine use disorder: insights from EEG microstates, BMC Psychiatry 20:334 (2020). Read the original primary study and its methods. Its forensic value lies as much in the limitations of translating an electrophysiologic group difference into a claim about an individual as in the measured differences themselves.

What scientific question was actually tested?

The investigators asked whether the moment-to-moment spatial configurations of electrical activity at rest differed between people with an established methamphetamine use disorder and comparison participants without that disorder. They then examined associations with a recorded history of methamphetamine use, computerized cognitive tests, and a methamphetamine-word Stroop task. They did not administer methamphetamine under controlled conditions and then record a dose–response curve. They did not collect serial toxicology specimens to reconstruct drug absorption or elimination. Their design was cross-sectional: both the EEG and cognitive data were observed during the same study period.

That distinction matters because a long-term clinical category and an acute, time-specific effect are different forensic propositions. An EEG pattern associated with group membership cannot, without independent validation, identify whether someone consumed methamphetamine at a particular hour or could safely operate a vehicle. A participant’s prior use frequency is also not the same variable as a contemporaneously measured blood concentration.

Who contributed the independent observations?

Researchers initially recruited 57 adults with methamphetamine use disorder from two compulsory rehabilitation centers in China and 30 community comparison participants. Two participants in the disorder group and three controls were excluded from the subsequent EEG analysis because more than five electrodes or channels were judged unusable. The analyzed dataset therefore contained 55 different people with methamphetamine use disorder and 27 different healthy controls: 82 independent individuals. The patients met DSM-5 diagnostic criteria and reported methamphetamine use during the three months before recruitment; their interviews followed detoxification. The article does not establish a uniform exact last-use interval for each patient.

Among those analyzed, the mean age was 29.56 years in the disorder group and 30.24 years in controls. Male participants represented 67.3% and 66.7% of the two groups. Neither of these baseline demographic comparisons was statistically significant in the published analysis. Age and sex matching helps interpret the comparison but does not establish that education, sleep history, tobacco exposure, other health factors, withdrawal timing, or unmeasured differences were perfectly controlled.

The group with methamphetamine use disorder reported an average five-year history of methamphetamine use. Their self-reported dose was 0.68 ± 0.48 grams as presented in the source’s characteristics table, without a chemically verified dose or a standardized route-and-purity assessment. The authors also reported different frequencies of historical use, including daily use and several-days-per-week use. Those historical reports are exposures of interest, not analytical concentrations measured in specimens.

What does an EEG microstate measure?

An electroencephalogram records electrical potential differences at scalp electrodes. A microstate analysis classifies short intervals in which the spatial pattern of these scalp potentials stays comparatively stable. The investigators used four recurring configuration classes labeled A, B, C, and D. These letters refer to categories of scalp topography generated through their analytic pipeline; they are not chemical compounds, toxicology cutoffs, or separate anatomical lesions.

Three principal metrics are important here. Mean microstate duration (MMD) describes the average persistence of a class in milliseconds. Global explained variance (GEV) describes the fraction of measured EEG topographic variance accounted for by that class. Ratio of total time covered (RTT) describes the share of time attributed to a class. A fourth measure, occurrence, counts how frequently a class appears. Differences in duration and explained variance need not be identical, and a statistically significant change in one metric does not authorize a claim that every metric changed.

Microstate classification is sensitive to decisions about preprocessing, artifact rejection, reference, temporal filtering, and clustering. Different algorithms or datasets may not reproduce the same numeric boundaries. It would be inaccurate to read class B as a stand-alone measurement of a particular named brain region, or to treat differences in these scalp-derived metrics as a direct assay of dopamine release.

How was the EEG actually recorded and analyzed?

The team used a 64-channel scalp EEG system with recordings during five to six minutes of eyes-closed rest. Continuous EEG was sampled at 1,000 Hz. They removed eye movement artifacts and other unacceptable segments, then filtered, downsampled, and classified the recordings. For microstate analysis they selected the first 20 artifact-free two-second epochs per person, meaning the microstate calculations were based on selected clean segments, not 5–6 minutes of equally usable EEG for every person. Twenty epochs from one person remain technical observations from that person; they do not increase the biological sample size from 82 to 1,640.

The analytical steps included k-means clustering of scalp maps and comparison of four microstate categories. The authors used a within-person microstate-class factor and a between-person group factor, with follow-up comparisons and a stated Bonferroni procedure for microstate-related outcomes. They also employed standardized low-resolution brain electromagnetic tomography, or sLORETA, to estimate possible intracerebral sources from the scalp signals. That last operation is a mathematical source-localization model, not surgically implanted electrodes measuring current inside a specific patient’s brain.

Which EEG differences survived the reported comparisons?

The most useful distinction is between effects that were reported as significant after correction, effects that were nominal or borderline before correction, and outright null comparisons. Microstates A and B showed shorter mean durations in the disorder group. Class B accounted for less explained variance and class C more. By contrast, the listed class D duration and class C RTT comparison did not support an equivalent finding. The measured numeric results below are presented as the article reports them; no new effect size, significance test, or diagnosis is being calculated.

Values are reported means ± standard deviations from the original Table 2. The original report supplies corrected p values for the selected differences but not for every null row. The class C time-coverage result illustrates why a nominal p value should not be promoted to a corrected positive result. The investigators also reported a statistically significant group-by-microstate-class interaction for several parameters; this does not mean that all four individual classes changed significantly.

Did the groups differ on cognitive performance?

Participants completed a CogState battery addressing memory, problem solving, social-emotional cognition, and spatial learning, as well as a color-naming task involving methamphetamine-associated and neutral words. The clearest group differences in the reported baseline table involved errors, rather than a universal slowing of response time. Patients made more errors on methamphetamine-associated Stroop words, more errors on neutral Stroop words, more Groton Maze Learning errors, and more Continuous Paired Association Learning errors. The reaction-time contrasts on the two word types were not significant.

The original table also labels a Two-Back measure an “accuracy rate,” although its reported mean values exceed one. That reporting ambiguity cautions against interpreting the printed number as a straightforward zero-to-one accuracy proportion. The appropriate source-grounded conclusion is that the study reported no significant group difference for that Two-Back index, not that the groups were proven cognitively equivalent.

These are cross-sectional comparisons of people, not acute methamphetamine challenge effects within the same participant. The Stroop color response is also not a lie-detection procedure, a standardized field sobriety assessment, or an estimate of driving speed. A greater error ratio on a laboratory task does not, without further evidence, establish what a person perceived, intended, or did during a disputed event.

What relationships were found between EEG classes, reported drug history, and task scores?

The investigators fitted generalized linear models after identifying microstate parameters of interest. More frequent self-reported methamphetamine use was associated with shorter class A mean microstate duration, with a published coefficient of −0.18 and reported p = .02. For class B, use frequency was associated with shorter mean duration (coefficient −0.57, confidence interval −0.96 to −0.19) and lower global explained variance (coefficient −0.11, confidence interval −0.17 to −0.06). The source prints some associated p values as “0.00,” a rounded display rather than a mathematical probability of exactly zero.

The Two-Back index and methamphetamine-word Stroop error ratio also entered the class B models. For example, class B mean duration had a reported positive association with the Two-Back index (coefficient 0.45, confidence interval 0.12 to 0.77; p = .01) and a negative association with methamphetamine-word errors (coefficient −5.35, confidence interval −8.52 to −2.17; p rounded to .00). Class B global explained variance was likewise associated with the Two-Back index (coefficient 0.06, confidence interval 0.01 to 0.11; p = .01) and methamphetamine-word errors (coefficient −0.58, confidence interval −1.05 to −0.10; p = .02).

These coefficients describe relationships within the modeled dataset. Their units and interpretive ranges depend on the source’s coding and task scales; they must not be transposed into a predictive equation for an unfamiliar person’s blood level or behavior. In particular, a model fit in recruited patients and matched controls was not externally validated as a forensic classifier of acute intoxication.

What important null outcomes limit the headline?

The authors explicitly reported no significant main group effect across all four microstate parameters, alongside selected group-by-class interactions. Class D comparisons did not reveal the same corrected changes as A, B, and C. Mean duration of class C did not differ significantly, and class C time coverage did not survive the stated correction. Several occurrences, time coverage, and other class-specific EEG comparisons were nonsignificant. These null findings are an important part of the pattern, not editorial exceptions to hide.

The word-based Stroop reaction times did not differ significantly between the disorder and control groups: p = .42 for methamphetamine words and p = .75 for neutral words. The reported Two-Back between-group result had p = .70, while the social-emotional cognition score comparison was also nonsignificant (p = .25). The International Shopping List comparison was printed at p = .05, a boundary value requiring caution rather than a firm positive claim. A nonsignificant comparison does not prove functional equivalence or the absence of an effect; it reflects the resolution of the available study.

The source stated that the class C explained-variance measure showed no correlation with the examined drug history and cognitive measures. No claim should be made that every microstate parameter tracked use frequency or cognitive errors. A study reporting several related tests also deserves attention to multiplicity, choice of comparison, effect direction, and whether the authors applied the appropriate adjustment to the specific number being quoted.

Were any methamphetamine concentrations or metabolites measured?

No quantitative biological methamphetamine concentration is reported as an outcome in this study. There is no measured parent methamphetamine in legal venous whole blood, no matching separated plasma or serum, and no drug-quantitation series in urine, oral fluid, hair, vitreous humor, milk, or tissues. There is also no measured amphetamine metabolite concentration to compute a parent-to-metabolite ratio. The reported gram quantity and frequency are participant history variables, not concentrations recovered with a validated analytical instrument.

The actual measured physical signals were scalp EEG voltage patterns and their modeled features, together with button-press responses and error counts. Nothing here compares a plasma-to-whole-blood distribution ratio, detection windows, sample preservation, instrumental limits of detection, or toxicological cutoffs. When a case requires evidence that a drug was present in a particular specimen, the correct evidentiary route is the laboratory’s specimen-specific analytic record, not an EEG network association.

What does source localization add—and what can it not establish?

The authors also used sLORETA to estimate the possible source distribution of scalp electrical activity. They reported differences in modeled current-density patterns for classes A, B, and C in cortical regions that included frontal, temporal, parietal, and occipital locations. These model-based regional findings provided hypotheses about the neural networks associated with the EEG observations.

However, inverse modeling an internal electrical source from voltage measured on the scalp is not direct localization of molecular drug action. The team did not record simultaneously acquired participant-level functional MRI in this experiment. They did not demonstrate that the observed source model discriminated MUD patients from all other relevant neurologic or psychiatric conditions. They did not isolate exposure duration, sleep disruption, withdrawal, or other competing causes as experimental variables. The localization results therefore should be framed as explanatory associations, not as anatomical proof that methamphetamine damaged a particular brain region in a specific person.

Why is the cross-sectional design a central forensic limitation?

Cross-sectional case-control research is useful for detecting differences worthy of further study, but it cannot by itself establish a temporal causal chain. Here, individuals were grouped by disorder history and compared once; the study was not a randomized methamphetamine administration study. Preexisting differences in education, sleep, comorbid health, nutrition, substance history, and laboratory setting could influence cognitive or EEG measures. Even when an investigator has matched age and sex, those other influences do not automatically disappear.

The compulsory rehabilitation-center setting introduces additional generalizability questions. Treatment status, the length of abstinence, and environmental characteristics could differ from those of people encountered in workplaces, emergency departments, roadside investigations, or court cases. The researchers themselves cautioned that their participant population may not represent people with methamphetamine use disorder at all treatment stages and recommended longitudinal follow-up.

One laboratory measure may be elevated in a population yet overlap extensively across individuals. The paper did not provide a court-ready threshold, externally validated sensitivity or specificity, a positive predictive value at real-world prevalence, or established likelihood ratios for a defendant, patient, employee, or driver. The absence of that individual-level validation matters more to case-specific use than whether a particular comparison’s p value crossed .05.

How should an expert distinguish a group association from impairment?

A defensible opinion must specify its endpoint. Association with diagnosed methamphetamine use disorder is the question this study addressed. Recent exposure requires appropriate evidence of a drug’s presence and timing. Acute impairment ordinarily requires a separate evaluation of relevant performance and circumstances; a specimen result alone may still be insufficient. Event causation asks an even more demanding question about actions or omissions during a particular incident.

The study measured none of those latter three categories as validated individual outcomes. It had no on-road assessment, simulator driving-performance endpoint, collision analysis, or observed incident. It did not establish that a shorter class B microstate duration predicts unsafe driving, delayed reaction to a hazard, credibility of testimony, or intent. The source is best used to discuss the research state and why proposed EEG biomarkers remain research findings rather than stand-alone forensic determinations.

Which underlying records would matter in a contested case?

If the question concerns this experiment itself, an independent reviewer would want the group inclusion and exclusion records, original EEG preprocessing settings, artifact-rejection decisions, topographic clustering parameters, task scripts, individual task distributions, and model specifications. A figure showing an average difference is not the same as a reproducible classification rule. Exact sample accounting is important because 87 recruits became 82 analyzed people, and because within-person EEG epochs cannot be mistaken for new participants.

If the question concerns a separate forensic drug investigation, specimen chain of custody, collection times, validated analytical identification and quantitation, uncertainty, possible metabolite findings, contemporaneous conduct, medical information, and alternative causes should be considered on their own terms. The site’s published discussions of biological matrices and confirmatory drug testing explain why specimen-based questions require methods different from scalp EEG. Therapeutic drug monitoring provides additional context for distinguishing concentration interpretation from cognitive observations.

What is the narrow, defensible takeaway?

Chen and colleagues reported a measurable pattern in a relatively small, selected sample: shorter mean EEG durations for classes A and B, less explained variance for class B, more for class C, and selected differences in error-based cognitive tasks. Some class B measures correlated with reported methamphetamine-use frequency and task performance. Other comparisons were genuinely nonsignificant, including class D duration and Stroop response-time differences. The observations do not constitute chemical proof of methamphetamine exposure or a validated one-person impairment test.

A useful scientific conclusion is that resting EEG microstates are candidates for additional mechanistic and longitudinal research in methamphetamine use disorder. Any stronger claim needs independently validated diagnostic performance, controls for confounding, reproducible EEG processing, and outcome-specific evidence that this one study did not provide. The original source, including its complete comparison tables and discussion, remains the appropriate starting point: Chen et al., primary full-text study.

Frequently asked questions

1. Did the EEG study test people immediately after a controlled methamphetamine dose?

No. Participants were recruited based on an existing DSM-5 methamphetamine use disorder diagnosis and relevant prior-use history. This was a cross-sectional comparison rather than a placebo-controlled acute dosing experiment. The self-reported history should not be translated into an observed dose, time since last ingestion, or concentration at EEG recording.

2. Does a shorter class B microstate prove methamphetamine intoxication?

No. Group-average class B duration was shorter, but the study did not establish a validated individual diagnostic threshold or acute intoxication classifier. Different people may have overlapping values, and EEG patterns can be influenced by factors other than methamphetamine exposure.

3. How many people were actually analyzed, and why does that matter?

Eighty-two independent people were analyzed: 55 with methamphetamine use disorder and 27 controls. Five of 87 recruited people were excluded because their EEG recordings had excessive bad channels. Multiple EEG epochs, scalp sensors, and cognitive trials within an individual do not increase the number of independent humans.

4. Were all microstate classes significantly altered?

No. Selected differences in classes A, B, and C were reported after correction, but class D did not show the same pattern. The class C ratio of total time covered had corrected p = .20, and its apparent uncorrected boundary result should not be recast as a corrected significant difference.

5. Did people with the disorder have slower reaction times on methamphetamine words?

Not significantly in the reported comparison. The methamphetamine-word Stroop reaction-time result had p = .42. The significant result concerned the error ratio, which was higher in the patient group. Error rate and response latency are distinct behavioral endpoints and should not be conflated.

6. Can the findings be converted into a whole-blood methamphetamine level?

No. The experiment did not quantitatively assay methamphetamine, amphetamine metabolites, or another drug in venous whole blood, plasma, serum, urine, or oral fluid. Scalp microstate durations, questionnaire histories, and cognitive-error ratios cannot be expressed as an analytically validated blood-drug concentration.

7. Can these observations determine why an individual driver crashed or whether a witness was credible?

No. No crash, real driving, field performance, or credibility endpoint was investigated. Establishing an individual person’s behavior, impairment, or causation requires evidence specific to the event and the relevant validated methods. This study supports only a carefully bounded discussion of group-level EEG and cognitive associations.