Can Heart-Rate Variability Predict Stimulant-Related Creativity Gains in ADHD? A 17-Adult Pilot

Short answer: Not reliably on the evidence available. In a small within-person study of adults with attention-deficit/hyperactivity disorder (ADHD), the investigators’ preselected heart-rate-variability measure, pNN50, did not predict the difference in anagram-solving performance between periods on and off prescribed stimulants. Several other ECG-derived frequency measures were associated with changes in divergent-thinking scores at unadjusted statistical thresholds, but those exploratory associations did not survive correction for multiple comparisons. The study provides a useful demonstration of why a physiological recording, a behavioral test and a measured drug concentration are three different kinds of evidence.

The primary study is Appling and colleagues’ 2025 investigation, Heart Rate Variability Prediction of Stimulant-Induced Creativity Gains in Attention-Deficit/Hyperactivity Disorder. Its participants were already prescribed stimulant medication; researchers did not randomize them to a standardized amphetamine or methylphenidate dose. They compared cognitive performance during a session after usual medication with performance following an instructed medication-free interval. A brief electrocardiogram (ECG) supplied autonomic indices, not a toxicological measurement of medication exposure.

What question did this experiment actually answer?

The scientific question was whether a person’s resting autonomic pattern—especially the fraction of successive normal heartbeat intervals differing by more than 50 milliseconds—could predict how much that person’s creativity-related task performance changed between medication conditions. The proposal was not that an ECG identifies a drug molecule. It was that one quantifiable physiological characteristic might account for some between-person variability in cognitive response to treatment. The researchers considered two different kinds of creativity-related task, because generating many original possibilities and finding a single correct solution place different demands on cognition.

This distinction matters to forensic interpretation. A variable can correlate with a group-level response to a behavioral test yet remain unsuitable for identifying an individual’s dose, blood concentration, exact time of ingestion or functional impairment. In this study the primary predictive hypothesis was not supported. The nominally significant exploratory observations invite replication, not a diagnostic cutoff or a legal inference about a particular person.

Who participated, and what does the sample size mean?

Seventeen adults with an existing ADHD diagnosis entered the study. Their reported mean age was 27.9 years, with a standard deviation of 6.7 years; the listed ages ranged from 19 to 42. The sample included nine women and eight men. Every participant already had a prescribed stimulant regimen. These were not seventeen stimulant-naïve volunteers assigned a common investigational dose, and the design did not create a separate parallel control group of untreated adults without ADHD.

One participant’s ECG-derived series required correction of approximately 76% of heartbeat observations, and that record was excluded from the predictive analyses. Accordingly, enrollment was seventeen people, while the usable HRV-regression sample was smaller. The repeated on-medication and off-medication sessions were measurements of the same people, not thirty-four independent participants. Likewise, the numerous anagram items, Torrance responses and ECG intervals were nested observations within participants; they do not multiply the biological sample size.

These sample-size distinctions affect precision. A regression built on roughly sixteen usable records has limited room to characterize the effects of multiple candidate HRV predictors, participant characteristics and different medication formulations. It can identify a preliminary association, but its estimated slope may change substantially in a larger sample.

What medications were involved?

The medication category was clinically prescribed psychostimulants, not a single standardized experimental formulation. The participant table includes amphetamine-based treatment, lisdexamfetamine, methylphenidate-based medication and dexmethylphenidate. These agents differ pharmacologically, and some are prodrugs or contain different active stereochemical components. Pooling them is reasonable for a small exploratory question about prescribed stimulant treatment, but it is not equivalent to a controlled pharmacokinetic comparison of one chemical substance.

One session occurred within approximately two hours of taking the person’s usual prescribed stimulant medication. Another occurred after the person was instructed to abstain from that medication for at least twenty-four hours. This is a reported dosing-condition distinction, not laboratory confirmation that every parent compound and active metabolite had disappeared from every biological compartment. Formulation, individual elimination, adherence and the timing of the last dose could all influence actual exposure. The original study does not provide concentration-time curves to resolve those uncertainties.

Was any stimulant concentration measured in blood or urine?

No quantitative post-dose drug assay is reported for a human legal venous whole-blood specimen, separated plasma, serum, urine or oral fluid. The study therefore cannot compare an ECG variable against measured dextroamphetamine, lisdexamfetamine-derived amphetamine, methylphenidate, dexmethylphenidate or metabolite concentrations in nanograms per milliliter. Neither a drug-detection limit nor a validated toxicological decision cutoff can be derived from the available data.

Physically collected specimens and behavioral or physiological measurements must remain separate. Plasma or serum involves a separated blood fraction; whole blood includes cellular components. Urine often addresses exposure or metabolite excretion; oral fluid samples a different compartment. Here, the principal biological signal was the electrical timing between heartbeats, acquired from surface ECG electrodes. It was not a direct assay of parent stimulant drug, a dopamine measurement or a validated substitute for any of those tests.

For a practical explanation of why these distinctions matter, see the published guide to biological matrices in forensic toxicology. It should not be presumed that a response in one compartment or measurement domain numerically converts into another.

How was heart-rate variability obtained?

Participants rested while an ECG was recorded for eight minutes. Approximately three minutes were allocated to acclimation, followed by a five-minute recording interval used for analysis. Electrode recordings were processed into successive heartbeat timing intervals, with artifact correction applied before HRV calculations. The group used a dedicated ECG acquisition system and HRV analysis software to extract both time-domain and frequency-domain measures.

Time-domain measures include SDNN, the standard deviation of normal-to-normal intervals; RMSSD, which summarizes successive interval differences; NN50, a count of interval pairs differing by more than 50 milliseconds; and pNN50, the corresponding percentage. These terms refer to rhythm variability. They do not indicate how many milligrams of stimulant were taken. Their values can also respond to respiration, posture, movement, stress, disease, time of day and the recording protocol.

The exploratory frequency measures addressed oscillatory components of the ECG-derived interval series. The study specifically considered the high-frequency fast Fourier transform peak and the share of very-low-frequency power. Those quantities are related to autonomic physiology but are not selective, direct molecular assays of norepinephrine or dopamine. Claims that an observed spectral association proves activation of one neurotransmitter pathway would go beyond what was chemically measured.

How was convergent creativity measured?

Convergent performance was tested using anagrams. Participants encountered twenty scrambled-word problems in a session: fourteen five-letter items and six seven-letter items. The task demanded finding the intended solution, not generating an unlimited range of answers. Participants had up to 120 seconds per item. An unsuccessful trial was assigned the maximum latency, and response-time information underwent logarithmic transformation for analysis.

The study’s dependent variable for the predictive question was a change in task performance across the medication conditions. The authors then asked whether a particular HRV feature tracked that change across participants. A longer or shorter solution time on this laboratory exercise is meaningful within the experiment, but it is not a comprehensive measure of real-world executive function, safe driving, the accuracy of testimony or occupational competence.

Did the preregistered-style pNN50 prediction work?

The preselected pNN50 variable did not significantly predict the stimulant-associated change in anagram performance: the reported regression coefficient was 0.112, with p = 0.691 and R² = 0.013. Several other time-domain indices were likewise nonsignificant. The numerical table below retains the original study’s model-level regression outputs rather than converting them into unsupported sensitivity, specificity or individual diagnostic probabilities.

These results should be read as failed evidence for the proposed prediction, not proof that autonomic physiology can never influence cognition. A small pilot may have inadequate precision to estimate modest effects, while a larger future experiment could study medication dose, respiration and time-of-day effects more directly. Nevertheless, the correct statement about this dataset is that none of the illustrated ECG metrics established a statistically significant predictor of the change in anagram performance.

How did the researchers evaluate divergent thinking?

Divergent thinking was assessed with the verbal Torrance Test of Creative Thinking, or TTCT. Participants completed structured prompts encouraging questions, interpretations, unusual uses and possible consequences. Responses were evaluated for fluency, originality and flexibility. Fluency refers to the number of relevant ideas, originality to uncommon responses under the scoring criteria, and flexibility to diversity across conceptual categories. Two evaluators scored the responses, and the analysis used their mean scores.

The study measured change between the on-medication and off-medication test conditions. An individual might increase an originality score while showing little change in anagrams, because the two tasks represent different cognitive operations. A high or low score is not a generic property of whether the participant was chemically impaired. It is an operationalized response to a particular task, under particular instructions and scoring rules.

Which exploratory HRV relationships appeared significant?

Four frequency-domain relationships met the authors’ initial unadjusted threshold in models of TTCT change. Higher high-frequency FFT peak frequency was associated with greater change in fluency and flexibility. Greater very-low-frequency power, expressed as a share of the measured spectral power, was associated with lower change in fluency and originality. The numerical results are shown separately because they were exploratory and not interchangeable with the preselected pNN50 hypothesis.

Crucial qualification: The authors state that these secondary findings did not survive correction for multiple measurements. A graph label showing an unadjusted p less than 0.05 does not reverse that conclusion. When many associations are tested in a small cohort, some may cross a conventional threshold by chance. The reported R² values describe model fit in these observations; they are not validated predictive accuracy in an independent population.

What is the difference between a nominal association and a replicated biomarker?

A useful biomarker must do more than produce a visually interesting scatterplot. Its purpose, collection protocol, reproducibility, uncertainty, analytical validity, target population and decision rule need to be established. If the intended claim is that an ECG score can determine a medication-related cognitive change in an individual, evidence would need prospective external validation and clinically meaningful error estimates. Those elements are not supplied by these pilot regressions.

Here, the significance corrections are especially consequential because many HRV indices and creativity domains were explored. HRV features may also correlate with one another, and their estimates can be sensitive to breathing and short-recording artifacts. With one participant excluded for heavy ECG artifact correction, the distinction between a usable biological recording and a computed number is particularly visible. There is no basis for constructing a forensic decision threshold from the significant unadjusted associations.

Can the experiment distinguish dopamine from norepinephrine effects?

Not directly. The authors discuss possible dopaminergic and noradrenergic contributions to attention, autonomic function and creativity, but did not chemically quantify synaptic dopamine or norepinephrine in a human brain. They also did not experimentally isolate receptor systems with a standardized active drug and selective antagonists. ECG spectral bands are indirect measures that can reflect multiple physiological processes. A claim that a particular participant’s creativity score was caused by a measured change in one catecholamine would require more specific evidence.

The paper offers a reason to investigate autonomic–cognitive relationships further. It does not demonstrate that prescribed stimulants universally improve creativity, nor that any one psychostimulant mechanism is necessarily responsible for a person’s observed score. Different medication classes were present in a heterogeneous clinical sample, and the on/off comparison does not fully eliminate expectations or carryover effects.

Why do medication condition and measured exposure require different wording?

It is tempting to write that each participant was either ‘drug-positive’ or ‘drug-free.’ That wording would overstate the evidence. A documented session conducted after taking prescribed medication is a dosing-history condition. A reported interval of at least twenty-four hours without medication is an instructed behavioral condition. Neither establishes a confirmed laboratory concentration. Without physical assay data, the study cannot determine whether a specific drug or active metabolite was quantifiable at the instant of ECG or cognitive testing.

Ordinary forensic confirmation involves a specific specimen, target compounds, calibration and quality controls. The current article contains none of those drug-assay results, so no comparison against a forensic laboratory’s analytical limits can be made. A general discussion of drug identification versus inferred effect appears in the published confirmatory drug testing guide.

What would a stronger follow-up experiment measure?

A larger, prospectively planned design could define one primary HRV predictor and one primary creativity change score before analysis. It could specify treatment formulation and dosing schedule, collect timed biological specimens for quantitative medication analysis, record respiration and sleep, and prespecify appropriate multiplicity corrections. It could separately study medication-naïve participants, participants stable on treatment, and people undergoing a controlled washout, without treating those populations as equivalent.

Where legally and ethically feasible, repeated timed venous whole-blood samples could characterize concentrations directly; a separate plasma arm would need its own validated method. Those measurements would still not by themselves prove creativity, intent or safe driving. Independent validation would be needed before any ECG-and-task predictive model could support an individual-level statement. Sample size planning should account for repeated observations within each biological participant rather than assuming that dozens of tasks create dozens of independent people.

What forensic questions cannot be answered from this pilot?

The study offers no validated inference about a specific person’s ingestion time, quantitative drug concentration, therapeutic adherence, intoxication, malingering, witness truthfulness, driving performance or the cause of an accident. In addition, its task endpoints do not test motor-vehicle operation, visual scanning during driving, emergency responses or real-world errors. Nothing in a group-level regression supplies those missing outcomes.

A clinician might use timed drug levels for a therapeutic question, but medication response and toxicology interpretation remain distinct. The published explanation of therapeutic drug monitoring illustrates why dose history, analytical matrix and sample time must be considered together. In a particular legal case, the relevant record would require actual specimen testing, collection history, validated methods, contemporaneous behavior and the precise question being litigated.

How should the principal evidence be described in an expert report?

A scientifically narrow description would state that a small within-subject pilot enrolled seventeen prescribed-stimulant users with ADHD, that the usable ECG-derived prediction sample was reduced by one artifact-heavy record, and that the investigators’ selected pNN50 metric did not predict anagram-change scores. Four exploratory HRV frequency associations with changes in TTCT measures crossed unadjusted thresholds but were not significant after multiple-testing correction. No post-dose concentration of a specified stimulant was quantitatively measured in forensic whole blood or other chemically assayed biological matrix.

That formulation leaves separate issues separate: participants’ treatment circumstances, recorded autonomic variability, task-specific creativity scores and any future forensic chemical evidence. It avoids turning an exploratory physiological association into a claim of a measured toxicological fact. It is also more informative than simply saying that stimulants either help or harm creativity, because the tested predictors, analytical correction and endpoint definitions matter.

Frequently asked questions

1. Did the study prove prescribed stimulants improve creativity in everyone with ADHD?

No. The study tested whether HRV metrics predicted changes in specific creativity-related tasks in a small clinical sample. It was not a population-level proof of a uniform stimulant benefit, and it combined different prescription medications rather than a common randomized exposure. The results must be interpreted for the particular laboratory endpoints measured.

2. Was there a statistically significant result for pNN50?

Not for the anagram-change regression highlighted by the authors: β = 0.112, p = 0.691 and R² = 0.013. This is an observed nonsignificant association; it is not equivalent to a demonstration that pNN50 has exactly zero physiological relevance in every setting.

3. Did any exploratory HRV result cross p < 0.05?

Yes, four reported frequency-domain associations did so before multiplicity adjustment, including high-frequency peak associations with fluency and flexibility changes. The authors explicitly state that those secondary signals did not survive multiple-comparison correction. They should be described as preliminary, not as confirmed biomarkers.

4. Were there seventeen independent people or thirty-four?

Seventeen were enrolled. Each could contribute observations in two medication conditions, but sessions are repeated measurements of the same individual. One participant was excluded from the HRV regression analysis because extensive ECG artifact correction affected the record. Repeated trials and heartbeat intervals also do not create new independent people.

5. Was amphetamine or methylphenidate chemically detected in whole blood?

No quantitative post-dose whole-blood assay for either agent is reported in the study. The recorded ECG was a cardiac electrical signal, and cognitive task scores were behavioral results. Neither substitutes for a validated drug identification or concentration measurement.

6. Can this paper determine whether an ADHD patient was impaired while driving?

No. It did not administer a driving test, establish an impairment classification rule, or correlate physical blood drug concentrations with on-road performance. It cannot establish the cause of an individual collision or determine the reliability of particular testimony.

7. What is the most defensible takeaway?

The selected autonomic measure did not support the authors’ primary predictive hypothesis; a few exploratory frequency-domain relationships appeared before correction but lacked adjusted significance. The findings warrant careful study of physiology and creativity, while giving no valid shortcut from a brief ECG to forensic drug levels or individual legal conclusions.

Primary original research

Appling C, Nuraini N, Hart E, Wang D, Tosh A, Beversdorf D, Ferguson B. 2025. Heart Rate Variability Prediction of Stimulant-Induced Creativity Gains in Attention-Deficit/Hyperactivity Disorder. Journal of Clinical Medicine 14(10):3570. https://doi.org/10.3390/jcm14103570.