EVIDENCE LIBRARY
Did Dexamphetamine Improve Inhibitory Control in 47 Adults with Obesity? The Null Treatment Interaction
A randomized dexamphetamine obesity study found no treatment-specific improvement in stop-signal inhibition or executive-function ratings. Understand n=47 and why p=.039 does not settle causation.
The direct answer: the treatment-by-time result was not significant
No demonstrated treatment-specific improvement in executive function or response inhibition emerged from this pilot randomized trial. The authors tested dexamphetamine against an otherwise similar placebo regimen in adults receiving care for obesity and obstructive sleep apnoea. Although some test scores changed across repeated visits and the treatment arm had a lower mean stop-signal reaction time overall, the decisive group-by-session interactions were not significant. They reported p = 0.659 for the executive-function questionnaire and p = 0.707 for the stop-signal outcome. Those findings do not establish that taking dexamphetamine caused an improvement in either measure.
This is a more careful answer than reading one favorable group comparison in isolation. A group can perform better on average even when its trajectory is not more favorable after treatment begins. The clinical and forensic question is whether the trajectory differed in a way consistent with treatment. The trial does not show that. It also does not establish that dexamphetamine impaired the study participants, and the absence of a significant interaction is not proof that no effect could exist in another population or dose schedule.
The source is Poulton and colleagues’ original 2024 primary research article. It is a small, longitudinal, placebo-controlled experiment in a selected clinical population, not an experiment on driving, intoxication, drug blood concentrations, or criminal conduct.
Who participated, and why 312 measurements are not 312 patients
The investigators initially recruited 52 adults whose body mass index was 25–60 and who met criteria for nocturnal respiratory support for obstructive sleep apnoea. The complete baseline analytic dataset included 47 distinct individuals, after stated eligibility and participation exclusions: 22 assigned to dexamphetamine and 25 to placebo. The average age of those analyzed was approximately 47.7 years, and approximately 68% were male. Study enrollment and valid observations are different denominators, and neither should be silently substituted for the other.
Investigators gathered repeated survey responses and computerized task sessions from many of the same people. Their executive-function model used 279 observations; the stop-signal model used 312 observations. Those are repeated measurement occasions nested within the 47-person clinical dataset, not 279 or 312 biologically independent research participants. Some people contributed data on more occasions than others, and missing sessions were common. Mixed-effects modeling included participant identifiers to account for within-person dependence; it did not magically turn each visit into a new human subject.
This distinction matters to scientific inference. A larger observation count may improve estimation when the model is appropriate, but between-person treatment assignments and unmeasured participant differences still constrain generalizability. The smallest late visit samples also restrict the meaning of a seemingly smooth group-average trajectory. For instance, the questionnaire table reports 47 observations at baseline but only 33 at follow-up; the validated stop-signal series reports 40 at baseline and 31 at follow-up. These numbers are successive visit counts, not a sample attrition calculation for a single drug measurement.
What the experimental treatment actually involved
The experimental drug was dexamphetamine, not methamphetamine, racemic amphetamine, or the prodrug lisdexamfetamine. The authors prescribed oral tablets in an escalating regimen. The active group began at 5 mg twice daily, a nominal total of 10 mg each day, and the amount increased by 5 mg twice daily on successive weeks to a maximum 30 mg twice daily, or 60 mg/day. Controls received matching placebo tablets. The protocol encompassed a six-week titration phase, an extended maintenance period, and a later withdrawal/taper and follow-up period.
These are scheduled treatment amounts, not individually proven absorbed doses. The research paper does not provide quantitative dexamphetamine assay results from a legally collected whole-blood specimen, a separately centrifuged plasma or serum sample, or an oral-fluid sample for these visits. Accordingly, one cannot calculate a specimen-specific peak, a measured blood half-life, a particular person’s exposure, or a forensic detection window from the assigned dose schedule. Medication logs and clinical oversight have different evidentiary functions from an analytically confirmed biological concentration.
Participants were assessed at baseline, weekly through the first six treatment weeks, at maintenance and at follow-up. The source’s Table 3 reports average intervals between dosing and the computerized stop-signal test during active treatment, generally around 148–190 minutes across sessions. There was wide variation within visits. That scheduling information establishes approximate time from assigned administration to testing; it is not a blood-collection timestamp and does not identify the biological concentration at which a task was performed.
What does the Barkley short form measure?
The Barkley Deficits in Executive Functioning Scale–Short Form (BDEFS-SF) is a structured self-report measure of aspects of everyday executive functioning. The primary summed scores represent reported difficulties involving organization, restraint, regulation and related abilities. This is not a gas chromatogram, chemical biomarker, direct observation of driving, or courtroom reliability measure. It asks people to evaluate their own everyday difficulties over an assessment period.
At baseline, the original table gives a whole-sample mean score of 27.51 (SD 7.05). That mean was 30.15 (7.97) at Time 1 and 29.56 (9.19) at Time 2. Late in the study, it was 27.14 (7.98) at maintenance and 31.64 (12.51) at follow-up. These are pooled group values from different available respondents; the table should not be read as the repeated trajectory of one illustrative person or as a direct treated-minus-placebo effect.
The model found a session main effect (p = 0.024) but neither a significant treatment-group main effect (p = 0.440) nor a treatment-group-by-session interaction (p = 0.659). Individual time-contrast comparisons reported an increase at Time 2 compared with later visits (p = 0.011) and a maintenance-versus-follow-up contrast (p = 0.022). These do not establish a dexamphetamine-specific gain, because the decisive differential change by assigned treatment was missing.
What a stop-signal task can and cannot capture
The computerized Stop-Signal Task asks participants to respond rapidly to a go stimulus on many trials while withholding responses on some trials following a stop signal. The tested constructs include go accuracy, ordinary go reaction time, a varying stop-signal delay, and a derived stop-signal reaction time (SSRT). The authors defined a longer SSRT as weaker response inhibition for this task. An SSRT expressed in milliseconds is a model-derived behavioral timing estimate, not a measured drug concentration or a general judicial finding about self-control.
The delay between the go and stop cues changed in 50-millisecond steps depending on each participant’s preceding inhibition success. Prespecified performance exclusions eliminated sessions with extreme stop accuracy, low go accuracy, too many go errors, or implausibly short SSRT. That screening is relevant: it helps maintain task quality, but creates unequal visit samples. The total session-level data used for stop-signal modeling were therefore not the same as the number of adults enrolled.
SSRT did decline with repeat testing, but practice and familiarity with the task offer a plausible competing interpretation of such a pattern. Repeated testing is an intervention on behavior in its own right. When both arms improve or change similarly after the first task encounter, an overall time effect is not sufficient to attribute the gain to the medication.
The result that is easiest to misinterpret: p = 0.039
The authors reported a treatment-group main effect for stop-signal reaction time, F(1, 44.39) = 4.52, p = 0.039, with a model coefficient of −14.70 milliseconds (95% CI −28.26 to −1.14). On average across sessions, the dexamphetamine assignment group had a lower SSRT than the placebo group. This is a between-group average; it is not the same as proof that the difference newly appeared because dexamphetamine was taken.
To ask whether the treatment made the difference, compare changes through time. Here the group-by-session test was F(8, 255.74) = 0.68, p = 0.707. The analysis therefore did not identify distinct temporal responses to active drug versus placebo. The authors expressly considered baseline or other preexisting group differences as a possible explanation for the lower overall treated-group mean.
It would be incorrect to headline the first result as “dexamphetamine improved inhibition by 14.7 milliseconds.” The coefficient is a pooled group separation in the model, not a validated individual pre/post causal treatment response. The source’s own plotted trajectories further illustrate why separating these questions is essential: the active arm does not have an unmistakable divergent improvement that would resolve the null interaction.
Why an early improvement can appear in both groups
The model found an overall visit effect for SSRT, F(8, 255.74) = 5.17, p < 0.001. Its principal reported contrast placed baseline SSRT 25.06 milliseconds higher than the average of all subsequent occasions, with 95% CI 16.29 to 33.84 and p < 0.001. This is evidence that the task measurements differed over repeated sessions. It is not evidence, by itself, that the drug altered response inhibition.
The source’s Table 2 shows an overall baseline SSRT mean of 283.21 milliseconds, compared with 265.24 at Time 1, 253.21 at Time 5, and 255.30 at follow-up. These entries summarize whichever valid respondents were available at each stage; do not subtract the overall means as though they represented a fixed matched-pair response in all 47 people. Participants may gain familiarity with the instructions, apparatus, timing and attentional demands. Both the modeled main effect and the descriptive means are compatible with that concern.
A stronger future study would include practice sessions before the baseline intended for treatment-effect estimation. It could also test whether measures are stable before randomization and pre-specify clinically or functionally meaningful effect sizes. That approach would help distinguish a change in cognitive capacity from learning the rules of one computerized task.
Two newly organized tables of the original evidence
The following two summaries separate the tested between-group contrasts from descriptive observations. They reorganize primary-study figures without copying publisher table images. All values and uncertainty are tied to a specific outcome rather than presented as universal drug-response rules.
| Outcome / model test | Original statistic | Interpretive meaning |
|---|---|---|
| Questionnaire group main effect | F(1,45.32) = 0.61; p = 0.440 | No significant overall treatment-group difference in this model |
| Questionnaire time main effect | F(8,234.82) = 2.26; p = 0.024 | Visits differ overall; not a drug-specific gain |
| Questionnaire group × session | F(8,234.82) = 0.74; p = 0.659 | No detected treatment-specific trajectory |
| SSRT group main effect | F(1,44.39) = 4.52; p = 0.039; β = −14.70 ms; 95% CI −28.26 to −1.14 | Average arm separation, not proof of change caused by dosing |
| SSRT session main effect | F(8,255.74) = 5.17; p < 0.001 | Performance varies across test occasions |
| SSRT group × session | F(8,255.74) = 0.68; p = 0.707 | No detected treatment-specific change |
| Research unit or visit | Source value | What the number represents |
|---|---|---|
| Recruited adults | 52 | Eligible-contact/initial study population; not all in analytic dataset |
| Baseline analytic people | 47: 22 active, 25 placebo | Independent assigned adults after stated exclusions |
| Longitudinal observations in questionnaire model | 279 | Repeated within-person survey occasions; not 279 people |
| Longitudinal observations in SSRT model | 312 | Repeated valid task occasions; not 312 people |
| Questionnaire baseline / follow-up mean | 27.51 (SD 7.05), n=47 / 31.64 (SD 12.51), n=33 | Different visit samples, pooled across randomization arms |
| SSRT baseline / follow-up mean | 283.21 ms (SD 33.70), n=40 / 255.30 ms (SD 35.65), n=31 | Valid task-session aggregates, not one constant paired sample |
| Mean drug-to-SSRT testing interval at Times 1 and 4 | 169.14 (SD 99.34) / 190.00 (SD 125.44) minutes | Scheduled timing variability; not observed drug concentrations |
The genuine nonsignificant findings are scientifically important
A null interaction should not be rewritten as an experimental success because another comparison has a small p value. For executive-function questionnaire scores, the treatment-by-visit p = 0.659 is the central check of different change patterns. For SSRT, the analogous p = 0.707 fails that check. Neither result justifies labeling the active drug an enhancer of this task in this trial.
At baseline, the groups also were not statistically different on age (p = 0.145), weight (p = 0.833), and biological sex (p = 0.522). Those are actual reported tests, but nonsignificant balance tests do not guarantee two small randomized groups are identical in every unmeasured trait. Relevant traits such as previous task familiarity, mood, education, sleep quality and medication adherence may vary.
The original discussion explains why a clinically plausible treatment theory can coexist with a null cognitive outcome. Some participants may not have had pronounced executive-function deficits at baseline. A short self-report scale may miss changes in the particular behavior of interest. The stop-signal task can be affected by learning and by the validity thresholds that change the contributing sample. The legitimate conclusion is a failure to detect the hypothesized differential improvement with these instruments and this sample, not proof that every individual under all amphetamine regimens will respond identically.
Weight reduction and inhibition should not be conflated
The investigators did describe greater weight reduction in the active-treatment arm. Their original model reports a baseline-to-follow-up weight interaction, F(1,45) = 160.06, p < 0.001. Weight is a clinical outcome and the research was embedded in a broader obesity/obstructive sleep apnoea treatment program. Yet weight change cannot be treated as a validated proxy for response inhibition, cortical dopamine release, driver safety, or addiction risk.
Nor can the mean duration of dosing or a substantial body-weight difference establish how much dexamphetamine was in any participant’s blood. A prescribed tablet dose and a kilogram of weight loss are different kinds of measurements. A mechanistic claim about concentration and cognitive response would require paired pharmacokinetic sampling and a defensible pharmacodynamic analysis; this pilot did not provide that direct link.
Forensic reports often encounter this kind of category error: a measurable treatment effect in one domain is carried over to a separate domain as if they were causally equivalent. The appropriate response is to name each measured endpoint, identify the instrument and its uncertainty, and separately examine which outcome the experiment actually tested.
Does a dose of 60 mg establish an individual blood concentration?
No. The maximum scheduled regimen was 30 mg dexamphetamine taken twice per day, but an administered or prescribed oral dose is not a directly measured concentration. Absorption, adherence, sampling time, metabolism, repeated dosing and individual variability can affect drug exposure. The paper reports neither postdose parent-dexamphetamine concentrations nor metabolite concentrations from a specified human forensic sample. It does not present a dose–blood concentration calibration, a quantitative analytical method, matrix validation, a limit of detection, a limit of quantification or individual result units such as ng/mL.
Human venous whole blood contains plasma and cells. Plasma is the separated liquid component prepared with anticoagulant, and serum is liquid prepared after clotting. They are not identical reported matrices, and a plasma value cannot be silently entered into a whole-blood evidentiary column. Urine, oral fluid, sweat, hair and vitreous humor address still different sampling and timing questions. None of these matrices was quantitatively assayed for the stimulant in this trial as reported.
For the scientific reasons these material distinctions matter, the site’s existing biological-matrices guide explains how specimen selection limits inference. A nominal dose can be a valid exposure assignment for a randomized clinical study while being insufficient to reconstruct an individual’s legal toxicology result.
Why this cannot determine driving impairment or drug use timing
The trial did not use an on-road driving examination, a standardized driving simulator, field sobriety examination, collision reconstruction, witness evaluation, or objective legal impairment endpoint. It used a self-rated everyday executive function scale and a specific computerized stop-signal task. A participant’s performance in that narrow test environment cannot be substituted for the totality of driving skills or impairment at a different time.
Even if a laboratory task effect had been demonstrated, one would still need independent evidence to connect that group-level result to a particular driver. Actual observed behavior, specimen collection timing, a validated chemical method, alternative causes of poor performance, medical conditions and other drugs might matter. A null treatment-specific interaction here does not certify safe operation of a vehicle after drug ingestion; neither does the group main effect establish unsafe operation.
Likewise, these models do not calculate when a person last used dexamphetamine. A time since assigned dose in a protocol is known from researchers’ planned medication schedule; a real case might have no such independently established dosing record. There is no measured decay curve or specimen concentration in this article to reverse engineer into a clock.
Why a medical history of obesity and sleep apnoea matters
This population was specifically recruited for obesity and an obstructive sleep apnoea care program. Many people needed nocturnal respiratory support. That clinical context does not make the participants equivalent to healthy younger volunteers or to unselected drivers with a forensic amphetamine-positive blood sample. Baseline sleep, fatigue, task motivation, treatment compliance and other clinical factors may affect performance.
Age also matters for generalizability. The analyzed group averaged approximately 47.7 years with a broad adult age span. The authors examined age in relation to baseline questionnaire scores and baseline SSRT and did not find significant correlations there. Those results do not prove age is irrelevant in all human amphetamine effects. The absence of a no-obesity/no-apnoea healthy control group further prevents estimating how far this selected patient population departed from community cognitive norms before drug treatment.
The source also excluded persons with several conditions or treatments, including some psychiatric illnesses and psychotropic medication use. A trial’s safety-related recruitment restrictions can improve internal control while reducing its reach to complex medico-legal cases with comorbid disease, co-medications and heterogeneous exposure histories.
Why repeated tests, missing visits and COVID disruption matter
Repeated administrations of the same questionnaire and stop-signal task create dependence among measurements. The mixed models appropriately used subject-level effects, but model quality still depends on assumptions about missingness and the validity of each session. During COVID-19 disruptions, access to the hospital varied and participants missed testing opportunities. The source explicitly identifies these interruptions as a contributor to missing data and uncertainty.
The analytic set also varies because the stop-signal task has predefined rules for unacceptable performance. It is scientifically appropriate to omit technically invalid trials, but the number of eligible observations then differs by visit. Simple subtraction of a baseline group mean from a late mean is not equivalent to an individual-level within-person treatment-effect estimate.
Authors describe an exploratory post-hoc power analysis suggesting high power to identify medium effects but materially less power for small effects. Such calculations do not transform a nonsignificant interaction into a proven absence of any effect. For judicial or clinical communication, state the observed test, the uncertainty, the sample size and the model’s targeted contrast rather than declaring the study “conclusive” or claiming a small universal drug effect.
What the result suggests—and what remains unmeasured
The most defensible positive statement is that the investigators conducted randomized repeated assessments and estimated average scores under a titrated prescription regimen. They observed group-level weight change, overall session-related shifts in the cognitive measurements and an SSRT average group separation. The more clinically specific question—did cognitive trajectory improve more because people received dexamphetamine—was not supported by the two reported treatment-by-session tests.
The results do not establish dopamine receptor occupancy, brain dopamine release, tolerance, treatment dependence, substance use disorder onset, or individual inability to inhibit a dangerous act. Those outcomes were not directly studied with a validated measure. In particular, the article’s discussion suggests the regimen did not worsen the tested cognitive measures, but the experiment was not designed or powered to determine later addiction incidence. Absence of observed cognitive deterioration is not a prospective addiction-risk analysis.
There is also no assay of parent amphetamine in human legal whole blood, plasma, serum or oral fluid, no documented LOD/LOQ and no quantitative last-use window. The assignment of oral medication supplies pharmacological context; it is not the missing analytical observation. This distinction is central when scientific literature is used in legal opinions.
A five-part framework for interpreting similar stimulant experiments
First, identify the independent person-level sample. Record the number recruited, the number assigned, the number analyzed and how many valid observations contribute to each instrument. Do not inflate statistical certainty by calling repeated trials new participants.
Second, name the tested experimental contrast. A within-group trend, treatment-arm main effect and treatment-by-time interaction answer different questions. When the claimed mechanism requires differential change after randomization, the interaction usually deserves special scrutiny.
Third, separate subjective and objective behavioral measures. A short executive-function self-report and a stop-signal timing estimate are useful but imperfect and not interchangeable. Their limitations are not repaired merely because both are included in the same paper.
Fourth, identify actual drug testing, if any. If no biological stimulant concentration was assayed, make that explicit. If there is one, document the real specimen, method, units and collection context. The site’s existing confirmatory drug testing guide explains why chemical identity and quantitative validity are separate from broader behavioral inference.
Fifth, limit individual conclusions. An average task association is not proof of individual driving impairment, medical causation, criminal intent or witness reliability. The record in a specific matter controls the facts needed for those opinions.
Pharmacology does not remove the need for a measured specimen
Dexamphetamine has recognizable stimulant pharmacology, but a receptor-based explanation of a clinical theory is not a specimen measurement. The paper used a prescribed oral titration scheme, not a controlled forensic blood pharmacokinetic investigation. No subject-level measured concentration-versus-SSRT curve was reported. None of its significance tests converts a dose from milligrams per day into milligrams per liter or nanograms per milliliter of legal whole blood.
Clinical therapeutic monitoring can sometimes connect a timed measurement to dosing, but it requires an actual relevant analyte and matrix. The same site’s existing therapeutic drug monitoring guide distinguishes clinical concentration interpretation from a legal assertion about a person’s functioning. Without primary analytical data, the right answer remains “not measured,” not a numerical blood estimate borrowed from another population.
Another common inferential error would be to substitute methamphetamine evidence for this dexamphetamine experiment. The molecules are not the same, and neither the clinical trial’s outcome pattern nor its nominal dosing schedule authorizes a methamphetamine-specific detection cutoff or behavioral claim.
How to explain the evidence in an expert report
A concise expert summary would identify the original trial as a double-blind randomized pilot with 47 analyzable adults and repeated measures. It would note that dexamphetamine was titrated over six weeks, that follow-up extended beyond titration, and that both questionnaire and stop-signal outcomes were tested through mixed-effects models. It should give the two interaction p values rather than relying on the favorable SSRT group main effect alone.
The limitations belong beside the result: a modest clinical sample selected for obesity/OSA, missing visits, short self-report instrument, task-learning effects, no direct stimulant blood quantitation, and no driving or prospective addiction outcome. That combination supports a narrow scientific conclusion without dismissing the value of a thoughtfully designed trial.
For source identification, the complete citation is Poulton A, Gauci N, Khalifa H, Hibbert EJ, Poulton AS. The Impact of Dexamphetamine Treatment for Obesity on Executive Function: A Double-Blind Randomised Controlled Pilot Study. Brain Sciences. 2024;14(12):1274. Read the original DOI publication alongside the numerical tables and study limitations when the distinction is consequential.
Frequently asked questions
Did dexamphetamine improve inhibitory control in this obesity trial?
Not demonstrably as a drug-specific change. The stop-signal group-by-session test was nonsignificant (p = 0.707). An overall treatment-arm mean difference (p = 0.039) is not the same question and does not establish a treatment-induced trajectory.
Why does the paper mention 312 stop-signal observations if it studied 47 adults?
There were repeated valid computerized tests from the same people. The 312 observations were statistically nested within participants; they are not 312 independent human volunteers. Repeated measures must be accounted for when calculating uncertainty.
What dexamphetamine dose did the researchers prescribe?
The active oral regimen began at 5 mg twice daily and increased by weekly steps to a scheduled maximum of 30 mg twice daily. It was a clinical titration, not a measured blood-drug concentration protocol.
Did the trial measure how much dexamphetamine was in whole blood or plasma?
No quantitative parent drug or metabolite assay in a specified human whole-blood, plasma or serum specimen was reported. Consequently no individual drug concentration, calibration curve, analytical detection limit or concentration-based impairment threshold can be inferred.
Does the shorter stop-signal time prove safer driving?
No. This study did not observe driving behavior or validate stop-signal reaction time against a collision or traffic impairment endpoint. Any individual driving opinion requires independent facts and methods appropriate to that question.
Was the null result evidence that stimulants are harmless or cannot cause addiction?
No. The investigators did not prospectively test addiction onset or all potential adverse outcomes. Failure to detect treatment-specific change in two cognitive measures is not evidence of universal medication safety or absence of dependence risk.
What should a forensic reviewer take from this experiment?
Keep the endpoints separate. The study supports a randomized comparison of two cognitive measures under an oral dexamphetamine regimen; it does not support a measured legal whole-blood concentration or case-specific intoxication and causation. Report both the significant main effects and the genuinely nonsignificant treatment-by-time tests.