Do Stimulants Change Arousal and Reward Networks More Than Attention Networks? An MRI Analysis of 5,795 Children

Short answer: In a large observational childhood imaging analysis, taking a prescribed stimulant on the morning of scanning was associated with detectable changes in functional connectivity involving sensorimotor, auditory, salience, and parietal-memory systems. Comparable statistically significant differences were not established in the canonical dorsal attention, ventral attention, or frontoparietal control networks after the researchers applied their network-level correction. A separate, intensively scanned five-adult methylphenidate experiment reproduced two main connectivity patterns. The authors also reported that sleep duration was associated with behavioral performance and that the relation between sleep and performance differed in some respects according to same-day stimulant use. These are population-level imaging and statistical observations—not direct measurements of a child’s blood stimulant concentration, a demonstration that attention never improves, or a test capable of establishing individual legal impairment.

The most important forensic distinction is between what an imaging-derived association actually measures and what a case-specific opinion might seek to infer. A functional connectivity edge is a statistical relationship between blood-oxygen-level-dependent time series. It is not a mass spectrometric measurement of methylphenidate, amphetamine, lisdexamfetamine, or any metabolite in human venous whole blood. The study’s controlled adult dose was recorded, but its imaging changes cannot be translated into a validated individual drug cutoff, ingestion clock, driving-capacity determination, or attribution of a particular incident.

What primary research generated these results?

Benjamin P. Kay and a multidisciplinary team published Stimulant medications affect arousal and reward, not attention networks in Cell, volume 188, pages 7529–7546, in 2025. The complete original empirical publication integrates a childhood brain-wide association analysis with an independently controlled, high-density adult imaging experiment and reference maps derived from separate sleep, electroencephalography, respiration, and molecular-imaging studies. The title states a mechanistic interpretation of the observed network pattern; the underlying methods test functional connectivity, not direct synaptic norepinephrine or dopamine release.

The researchers used resting-state functional MRI, a method that compares fluctuations in blood-oxygen-related brain signals across predefined regions. Their analysis compared relationships among many cortical networks rather than limiting the original question to a few preselected attention regions. They used family-wise-error-controlled network-level statistics to reduce the risk of treating a large set of correlated connectivity comparisons as independent discoveries. Even a corrected finding remains an estimate subject to sampling, modeling, head-motion correction, and medication-selection assumptions.

Were there 11,875 children, 5,795 children, or five adults?

All three counts describe different levels of evidence. The broader Adolescent Brain Cognitive Development (ABCD) dataset comprised 11,875 children aged approximately eight through eleven years at the relevant study stage. The imaging analysis used 5,795 of those children with the required low-motion scans and other complete data. Within that imaging subset, 337 children had taken a prescription stimulant on the morning of their MRI. Another 76 children were prescribed stimulants but had not taken one that morning. The remaining comparison category consisted of 5,382 children who were not prescribed and were not taking a stimulant.

These groups are not 11,875 separate randomized drug allocations. They come from the same observational cohort with different eligibility and medication histories. The independent controlled experiment enrolled five healthy adults without ADHD, each scanned extensively before and after a 40 mg methylphenidate exposure. Multiple scan minutes, network edges, voxels, parcels, drug-on/drug-off acquisitions, and model-resampling iterations do not increase the number of independently recruited people. The five adults validate a selected brain connectivity pattern, not the behavioral regression on all 5,795 children.

Which stimulant drugs and dosing information were actually known?

In the childhood cohort, stimulant exposure was based on whether a prescription medication had been taken on the morning of scanning. The category included medication types such as methylphenidate and lisdexamfetamine, but the main analysis did not have uniform, precise data for the children’s dose, release formulation, ingestion-to-scan interval, chemical stereoisomer exposure, or measured circulating parent drug. The researchers did not give a common investigational stimulant dose to all 337 children. Therefore, the observational comparison is properly called a same-day prescription stimulant association, not an experimentally assigned 40 mg methylphenidate trial in children.

For the adult precision-imaging validation, the specific experimental exposure was 40 mg methylphenidate (Ritalin). Each of the five individuals contributed 120–180 minutes of imaging off stimulant and 15–60 minutes on stimulant, with 165–210 minutes of usable imaging per person in the described protocol. Methylphenidate is chemically different from dextroamphetamine, methamphetamine, and the lisdexamfetamine prodrug. Neither the administered dose in milligrams nor the time series of magnetic resonance images is itself an analytical blood concentration.

What did the researchers mean by functional connectivity?

Resting-state functional connectivity is an estimate of the degree to which measured brain-region time series vary together. The MRI signal used here depends on local changes in blood oxygenation and associated hemodynamic responses. It is spatially and temporally different from an electrode’s direct voltage recording, a positron-emission estimate of receptor binding, a blood laboratory test, or a behavioral reaction-time measure. A negative or positive connectivity difference does not automatically mean inhibition or excitation at a synapse, and a network label does not signify a person’s competence in all tasks associated with that network.

The study discussed somato-cognitive action and sensorimotor regions (SM), auditory cortex systems (AUD), salience areas (SAL), and parietal-memory areas (PMN). Attention-related comparison categories included dorsal attention (DAN), ventral attention (VAN), and frontoparietal control (FPN) networks. Some networks were combined for statistical power or anatomical analysis, notably SAL with PMN. As a result, a detected composite SAL/PMN change should not be silently described as a separate proven effect in each small subnetwork or translated into individual reward seeking.

Which connectivity findings were statistically supported?

In the 5,795-child analysis, taking a stimulant on the scanning morning was associated with lower within- and between-network connectivity involving SM and AUD, and higher cross-network connectivity between SM and the combined SAL/PMN group. These selected comparisons survived the authors’ network-level family-wise-error correction at p < .05. The relatively strongest regional and network contrasts clustered around action, motor, auditory, salience, and parietal-memory territories. The paper reports a representative high-percentile stimulant connectivity contrast of beta −0.030, corresponding to Cohen’s d approximately −0.17; that effect estimate is small in standardized terms, even when a large dataset supports rejecting a null hypothesis.

By contrast, the study did not establish significant network-level stimulant differences in the canonical DAN, VAN, or FPN systems. The investigators specifically examined whether this might be an underpowered comparison and reported adequate power for effects of the size seen in some earlier small studies. Nevertheless, an absence of a significant corrected network difference is not proof of no drug-related change at every location, on every task, at every dose, in every age group, or in a particular individual. The meaningful source-supported finding is the distribution of detectable effects under a specified model.

What did the five-person methylphenidate experiment independently show?

The five adults underwent dense imaging within persons under medication-off and medication-on conditions. This matters because the childhood observational data could reflect why some children receive prescriptions, sleep differently, or have other characteristics rather than medication alone. In the adult experiment, the primary connectivity directions agreed with the childhood comparisons: decreased within-SM connectivity after methylphenidate, mixed-effects p = .008, and increased SM-to-SAL/PMN connectivity, p = .013. Spatial magnitude maps of the adult and child stimulant contrasts correlated at r = .41, with a spatial-permutation p < .0001.

That concordance is encouraging for reproducibility of the particular imaging signature, but the sample size remains five independently recruited adults. Hundreds of minutes per adult enhance within-person sampling density; they cannot establish between-person sensitivity, specificity, or a clinical reference interval suitable for identifying drug exposure in a new individual. The adult study also did not experimentally replicate the childhood sleep–school-grade interaction or provide an individualized clinical driving threshold.

What does the source actually quantify? First evidence table

The p values above pertain to different hypotheses, models, and sampled populations. They are not interchangeable with an accuracy rate for predicting a person’s drug use or impairment. The 337 children and the five adult participants contributed different observational and experimental evidence; pooling them into one artificially large randomized stimulant cohort would be methodologically unsound.

How did sleep duration relate to the brain-network patterns?

Parent-reported habitual sleep duration served as the child’s sleep-exposure measure. In the sample that was not taking stimulants, longer reported sleep was associated with lower connectivity within somatomotor, auditory, and visual networks, a pattern overlapping the direction of several stimulant-related effects. The authors compared that spatial distribution with other research-derived templates involving concurrent EEG and functional MRI, respiration-related MRI fluctuations, and norepinephrine transporter availability estimated by PET in different adult samples. For example, the sleep-derived map correlated with an EEG-fMRI arousal map at r = .49, spatial-permutation p = .0015, and with a respiration-derived map at r = .51, p = .0015.

These are correlations among brain maps or among regional signals, not actual measurements of a child’s synaptic norepinephrine concentration. In the subset taking a stimulant, the authors did not observe the same significant relationship between sleep duration and canonical network-level connectivity that appeared among children not taking a stimulant. That comparison motivated their proposal that stimulants can partially offset some immediate correlates of shorter sleep. It does not show that a prescribed stimulant substitutes for sufficient sleep, prevents cumulative sleep loss, or normalizes sleep-related physiology in every person.

Were school grades and cognitive test results consistently improved?

No: the behavioral patterns varied by diagnosis, sleep, and endpoint. The paper’s regression results for the 5,795-child cohort, adjusted for demographic factors, reported significant negative associations between ADHD and school grades, NIH Toolbox score, and n-back accuracy. Longer reported sleep was positively associated with grades, NIH Toolbox, and n-back accuracy. Simply taking a stimulant that morning was not associated with a statistically significant main effect on school grades (p = .154), NIH Toolbox score (p = .98), or n-back correct response fraction (p = .64). A negative estimated n-back reaction-time difference of 101 milliseconds was statistically significant (p = .0025), but reaction speed is not synonymous with accuracy or academic attainment.

The same regression identified an ADHD-by-stimulant interaction for school grades (p = .0057), NIH Toolbox (p = .000024), and n-back accuracy (p = .039). A stimulant-by-shorter-sleep interaction was significant for school grades (p = .025) and n-back reaction time (p = .00083), while the corresponding NIH Toolbox (p = .32) and n-back accuracy (p = .90) interactions were not. These are conditional regression interactions in observational child data, not controlled observations that administering a drug causes higher school grades at an individual level.

Which results were genuinely nonsignificant? Second evidence table

The table distinguishes true tested null results from statistically supported interaction terms. A significant interaction can coexist with a nonsignificant overall main effect because the two tests answer different questions. Neither type of estimate constitutes a blood concentration or a validation study for classifying a single driver or patient. Nor should a significant reaction-time shift be assumed to demonstrate an improvement in correctness; the two measured outcomes require separate examination.

Why does the sleep analysis not establish an acute pharmacological rescue?

First, sleep duration was ordinarily reported by parents, not established for every child by prospectively monitored polysomnography. Second, stimulant-taking children were not randomly assigned to sleep deprivation or a standardized drug regimen. Third, ADHD symptoms, medication access, clinical severity, comorbidities, school context, socioeconomic differences, and sleep habits are related in ways a regression may not completely eliminate. Fourth, even an association seen under adjustment can be affected by selection into a sample of low-motion, analyzable fMRI recordings.

The authors appropriately proposed a future washout design to evaluate persistence and reversal after drug levels decline. That proposal highlights what the published design could not directly determine: whether the short-term network pattern caused better cognition, whether delayed drug cessation produces compensatory decrements, or whether changes would hold under measured objective sleep deprivation across an experimentally assigned population. Forensic interpretation should not transform an observed conditional population association into proof of the timing or magnitude of an effect after a named person’s ingestion.

What does the study say about dopamine, norepinephrine, and reward?

The physiological interpretation invokes norepinephrine-associated wakefulness and dopamine-associated reward salience. However, the primary childhood and five-adult stimulant analyses did not sample synaptic transmitter release, measure receptor occupancy after dosing in those same child participants, or administer a selective receptor antagonist to prove a neurotransmitter pathway. Instead, the investigators compared their connectivity maps against existing template maps from independent research, including PET measures of norepinephrine transporter density. A degree of spatial concordance supports compatibility with a hypothesis but does not isolate biochemical causation.

The salience and parietal-memory systems may relate to anticipated reward, effort, and persistence. A plausible interpretation is that some observed behavioral benefits involve the motivation to continue an otherwise unrewarding task rather than a universal enhancement of attention-network activity. That is a model of how the overall findings fit together. The study did not separately measure, in each child, the exact psychological value of a particular classroom assignment or demonstrate that reward processing caused a better report-card grade.

Were drug molecules actually quantified in a forensic biological specimen?

No quantitative forensic drug concentration is reported in human venous whole blood for these participants. The primary pediatric imaging cohort records medication use, and the adult validation records an administered oral methylphenidate amount; neither supplies a corresponding validated parent-drug/metabolite concentration-time series in legal venous whole blood. The study also does not report quantitative parent methylphenidate or amphetamine in separated plasma or serum, or paired oral fluid, urine, hair, sweat, vitreous, or breath drug concentrations. No source-specific whole-blood analytical method, internal standard, detection or quantitation limit, chromatographic calibration, or specimen-collection time was used to link the published imaging effects to person-specific chemical concentrations.

The blood-oxygen-level-dependent fMRI acronym BOLD includes the word “blood,” but that does not make BOLD a chemical blood toxicology assay. MRI detects hemodynamic signal variation, not drug molecules. Similarly, a PET map of norepinephrine transporter availability from separate participants is not a parent stimulant blood measurement. A court interpretation that converts a connectivity coefficient or a known oral dose into a numerical venous whole-blood value would invent data not provided by the primary experiment.

Why can the findings not determine individual driving impairment or event causation?

The childhood study did not use an on-road course, validated impaired-driving simulator, controlled traffic hazard, standardized field sobriety examination, crash reconstruction, or witness memory-verification test. The adult controlled methylphenidate experiment assessed dense resting-state imaging, not individual road performance at a known forensic blood concentration. Therefore the investigators generated no empirically validated operating characteristic, sensitivity, specificity, positive predictive value, or discriminating cutoff for impaired driving from this neural signature.

Forensic opinions require a scientifically defensible bridge between the sampled population and a legally material proposition. A person may have a drug identified in a particular specimen without being shown impaired; a person may perform poorly on one task without proof of drug causation; and a group-level neural association cannot identify whether a particular crash or action resulted from a chemical exposure. The Kay study addresses population imaging and cognitive correlations. The legal questions of drug identity, ingestion timing, individual performance, culpability, and event-specific causality remain unmeasured.

How should this paper be compared with other stimulant neuroimaging studies?

Compare each study’s participants, experimental assignment, compound, specimen, device, statistical contrast and outcome before aggregating conclusions. Kay and colleagues used a large observational sample of prescribed children and a five-adult imaging validation with 40 mg methylphenidate. Other research may employ a 60 mg methylphenidate crossover, PET dopamine-binding tracers, and a ball-tracking task in 37 healthy adults. Those are different source populations, doses, imaging estimands, and questions; they should not be merged into an invented single trial or portrayed as replication of every endpoint.

The strongest comparison focuses on what was actually directly measured. Here the meaningful original distinctions are sensorimotor/salience versus canonical attention-network connectivity, objectively analyzed MRI signals versus reported sleep and task ratings, main effects versus interactions, and observational childhood data versus small within-person controlled adult imaging. The findings can inform hypotheses for future quantitative pharmacological research, but independent drug assays and targeted functional validation are necessary before such hypotheses can support individual forensic conclusions.

Seven frequently asked questions

1. Did the researchers prove that prescription stimulants never improve attention?

No. Their corrected resting-state network analysis found no significant stimulant association in three canonical attention/control networks. That is narrower than claiming the compounds never influence attention in any context, dose, task, or person. Other behavioral and regional mechanisms remain scientifically possible.

2. Were there 5,795 randomized children given 40 mg methylphenidate?

No. The 5,795 children were the analyzable imaging subset of a broader observational study; 337 had taken various prescribed stimulants that morning. A separate controlled imaging study examined five adults exposed to 40 mg methylphenidate. These designs must not be conflated.

3. Did the adult experiment reproduce the main imaging directions?

Yes, for two selected network patterns: within-sensorimotor connectivity was lower at p = .008, and sensorimotor-to-salience/parietal-memory connectivity was higher at p = .013. That does not establish diagnostic accuracy for individual people.

4. Did stimulant-taking children all have better school grades?

No. The overall adjusted stimulant main effect on grades was nonsignificant, p = .154. An ADHD-by-stimulant interaction, p = .0057, and a stimulant-by-shorter-sleep interaction, p = .025, addressed conditional statistical associations rather than a universal benefit.

5. Was n-back correctness significantly increased for all children on medication?

No. The overall morning-stimulant main-effect test for n-back correctness was nonsignificant, p = .64, even though an ADHD-by-stimulant interaction was significant, p = .039. Accuracy and reaction time were separately tested, and the latter had a significant negative coefficient.

6. Can fMRI connectivity determine a drug concentration in venous whole blood?

No. The study did not calibrate connectivity against measured parent stimulant or metabolites in contemporaneously collected forensic venous whole blood. MRI BOLD covariance and a medication history are not chemical drug measurements and supply no quantitative conversion.

7. Does a stimulant-associated sleep pattern identify an impaired or unimpaired driver?

No. Sleep was primarily parent-reported in the childhood cohort, and the study assessed no validated person-level driving outcome. Differences in brain-network maps and selected task scores cannot establish how a particular individual drove or what caused an actual event.

Bottom line: what does the original evidence—and only that evidence—support?

This primary empirical work supports a specific group-level account: morning prescription stimulant exposure in a large childhood imaging subset was associated with altered action/motor, auditory, salience and memory-network functional connectivity, whereas corrected differences in canonical attention/control networks were not established. Five densely sampled adults exposed to 40 mg methylphenidate reproduced selected network directions. Sleep duration and ADHD status modified several behavioral associations, but original main-effect and interaction tables also contained meaningful nonsignificant results. Those distinctions make the study informative for neuroscience without making it a validated chemical toxicology assay or legal-impairment test.

Readers evaluating a specific case should keep experimental dose, prescription history, parent-reported sleep, fMRI-derived covariance, computer-task performance and directly assayed forensic specimens in separate evidentiary categories. None can be silently exchanged for another. The full statistical definitions, corrections, source limitations and reference maps should be reviewed in Kay and colleagues’ original complete primary paper before adopting any inference beyond the populations and measurements it actually studied.