Does 9 mg/kg Caffeine Improve Attention More Than 3 mg/kg? A 14-Athlete Crossover Trial

Answer: No additional benefit of the highest tested dose was demonstrated. In a randomized, double-blind, four-condition crossover study of 14 young female team-sport athletes who habitually consumed little caffeine, single oral caffeine doses of 3 and 6 mg/kg improved simple reaction time, choice reaction time and a short number-cancellation attention score relative to placebo. The 9 mg/kg condition did not significantly improve those three outcomes versus placebo, and participants reported more undesirable symptoms at that higher dose. The two lower active doses did not significantly differ from each other on the three main tasks. Importantly, mental-rotation response time also improved at 3 and 6 mg/kg, while mental-rotation errors showed no significant treatment effect. The findings concern scheduled computerized and paper-based tasks, not chemically measured caffeine concentrations, lawful driving fitness or any individual’s impairment.

This analysis is based on the full original experiment by Bougrine and colleagues, Effects of Various Caffeine Doses on Cognitive Abilities in Female Athletes with Low Caffeine Consumption, published in Brain Sciences in 2024. Read the primary study, DOI 10.3390/brainsci14030280. Its central lesson is that dose, task, observed adverse symptoms and physical specimen must be described separately. A significant group-level improvement on a one-minute attention exercise cannot establish a toxicological blood concentration or what happened during an unrelated real-world event.

What scientific question did the study actually answer?

The investigators compared four experimental conditions in the same participants: an inert cellulose placebo, followed on separately randomized occasions by caffeine at 3, 6 or 9 milligrams per kilogram of body mass. The objective was not to diagnose caffeine intoxication. It was to determine whether cognitive-test performance improved monotonically as nominal caffeine dose increased and whether undesirable symptoms became more common. That distinction matters because a drug may improve a selected task at one dose but show diminishing performance gains, changed speed–accuracy tradeoffs or greater symptoms at another.

All three active levels were compared with the placebo session. The lower-dose comparisons were generally favorable for processing speed and short attention tasks; the highest dose was not. Consequently, the experiment does not support a simple claim that more caffeine reliably produces faster or better cognition. It also does not establish a universal optimal dose for athletes, adolescents, patients or motorists. The study has a particular sample, task battery, exposure schedule and set of measured outcomes.

How many independent people were tested?

Fourteen independent athletes completed all four sessions. The original screening narrative describes 41 surveys reviewed, 26 apparently eligible athletes, three withdrawals and nine exclusions involving menstrual-cycle scheduling, leaving the analyzed 14. Their mean age was 17.4 ± 0.8 years, so this was a predominantly adolescent cohort, not a randomized sample of adult drivers. Mean body mass was 59.7 ± 9 kg. Participants had several years of team-ball-sport experience and regular weekly training.

Each athlete could contribute observations under four conditions. Those are repeated measures on the same individual; they do not create 56 independently recruited biological subjects. Four pages of cancellation stimuli, repeated reaction-time signals and individual side-effect questions are likewise not additional people. Although a crossover design reduces some stable between-person differences, its precision still depends heavily on only fourteen individuals and on how well treatment order, practice, testing conditions and expectations were controlled.

Why does habitual caffeine consumption matter?

The researchers selected athletes who usually consumed relatively little caffeine, reporting a mean daily intake of 42.4 ± 9.4 mg, approximately 0.7 ± 0.2 mg/kg/day. They excluded participants classified as caffeine-naïve and those exceeding the protocol’s low-use threshold. This selection was deliberate: habitual use, withdrawal and tolerance could otherwise complicate comparisons among acute doses. Selection also restricts generalizability. The same response should not simply be assumed in a person who habitually takes large amounts of caffeine, a person who avoids it entirely or someone using additional psychoactive medications.

Reported prior intake was obtained through a dietary-frequency assessment, not by repeatedly measuring caffeine and its metabolites in biological specimens. Participants said they avoided caffeine-containing food and drinks for the 24 hours preceding testing. A declared abstinence instruction helps standardize the protocol but does not provide a verified concentration of zero in blood, plasma, serum, saliva or urine at baseline. Exposure history, biological level and behavioral effect are different pieces of evidence.

How were the four doses administered and timed?

The study used opaque oral capsules containing placebo or caffeine powder, with nominal active doses calculated from each participant’s body mass. Testing began about 60 minutes after swallowing the assigned capsule. At least 72 hours separated visits. Each trial began at approximately 09:00 under controlled environmental conditions, with standardized testing order and a prescribed pretrial meal. Participants were familiarized with testing beforehand to limit, but not eliminate, learning and practice effects.

The exposure labels here are 3, 6 and 9 mg of caffeine per kilogram, not 3, 6 or 9 milligrams in total and certainly not 3, 6 or 9 ng/mL in blood. The article itself sometimes abbreviates the dose groups as “3 mg CAF” and similar labels; the protocol unambiguously defines milligrams per kilogram. The actual milligram amount therefore varied with body mass. Absorption, distribution and elimination were not quantified in individual participants, and no dose-to-blood conversion was measured.

Which laboratory tasks were actually observed?

The simple reaction-time task recorded how quickly an athlete pressed a response control after a visual signal. The choice reaction-time task asked the participant to identify a matching colored geometric target among alternatives, adding stimulus discrimination. Both primarily generated a time measure rather than a holistic assessment of thought, judgment or movement safety.

The attention test was a one-minute number-cancellation exercise: participants scanned four printed pages of numbers and marked three-digit targets, with a maximum of 187 correct target identifications. This measure can depend on attention, visual scanning and rapid hand responses simultaneously. The mental-rotation task required comparisons of differently oriented objects, with distinct measurements for completion time and answer errors. The battery was administered at rest, not during a sports match, while steering a vehicle or amid actual hazards.

How did simple reaction time change?

The source reports an overall caffeine-dose effect on simple reaction time of F(3,39) = 40.46, p < .001, partial η² = .75. Relative to placebo, mean simple response time was 8.2% shorter at 3 mg/kg and 7.7% shorter at 6 mg/kg, with each reported post hoc comparison p < .001. The direct comparison between those two active levels was not significant (p > .05). Nor was the 9 mg/kg treatment significantly different from placebo for this outcome.

Thus, the graph and statistical text support a localized benefit at two lower levels, not a continuously improving dose response. A decrease in milliseconds on this task represents faster button presses following a predictable visual event. It does not directly measure braking in traffic, perception of multiple hazards, a person’s decision about right of way, balance, endurance or the probability of preventing a collision. Even a reproducible average reaction-time improvement is not a validated legal finding of unimpaired driving.

Did the choice reaction-time results follow the same pattern?

Yes, for the conditions evaluated. The omnibus choice reaction-time effect was F(3,39) = 27.69, p < .001, partial η² = .68. Compared with placebo, mean choice response time improved by 4.2% at 3 mg/kg and 3.7% at 6 mg/kg; both post hoc comparisons were reported at p < .001. Again, 3 and 6 mg/kg did not significantly differ from one another. The 9 mg/kg condition was not significantly better than placebo.

Choice response is often described casually as decision-making, but the participant was choosing whether to click for a matching shape, not solving a legal, safety or interpersonal problem. More complicated real-world decisions include context, competing goals, error consequences and long periods of attention. The trial did not directly measure those abilities. An individual who responds quickly in this tightly defined visual task may still make an error during a complex event; conversely, a slower response in an artificial task does not by itself establish legally relevant impairment.

What happened on the one-minute attention test?

The reported omnibus difference on the number-cancellation attention score was F(3,39) = 23.74, p < .001, partial η² = .64. Relative to the placebo session, correctly marked targets increased by 6.4% at 3 mg/kg (p < .001) and 4.6% at 6 mg/kg (p < .01). No significant benefit versus placebo was reported at 9 mg/kg, and the low-versus-moderate comparison did not reach statistical significance.

The evidence is specific to a short task. A higher total could reflect improved sustained visual attention over the minute, quicker scanning or a change in how rapidly the participant marks numbers. It does not necessarily imply improved accuracy of all observation, fewer important misses over a prolonged shift or faster recognition of emotionally salient events. The source does not establish a validated correlation between its cancellation score and the ability of an identified person to drive, testify accurately, monitor machinery or perform an emergency response.

Evidence table 1: what improved, and what did not?

There was no significant difference between 3 and 6 mg/kg on the simple, choice or cancellation tasks in the source’s reported direct comparisons. “Not significant” does not mean the true effects were identical, especially in a sample of fourteen. This table intentionally uses the printed percent differences and test statistics, rather than inventing exact unprinted means from graphical bar heights.

Did mental-rotation performance truly remain unchanged?

The proper answer depends on which component is being discussed. The study measured both time to complete its mental-rotation task and the number of errors. For completion time, its Results section reports F(3,39) = 13.36, p < .001; 3 mg/kg and 6 mg/kg shortened task time by 3.1% (p < .01) and 2.5% (p < .05), respectively, compared with placebo. At 9 mg/kg, no significant improvement was reported. These are observed time outcomes.

For mental-rotation errors, however, the dose comparison was nonsignificant: F(3,39) = 1.00, p = .40. A shorter completion time without an established error difference should not be described as a verified improvement in all spatial reasoning or all cognitive accuracy. The paper’s broad discussion and conclusion say that mental-rotation performance did not benefit, but that wording is too sweeping when read against the statistically positive time outcome. The defensible account states both results explicitly and does not erase either one.

Were there more undesirable symptoms at the highest dose?

Participants completed a symptom questionnaire immediately after testing and again 24 hours later. At the 9 mg/kg level, the paper reports more complaints such as feeling a rapid heartbeat, headache, gastrointestinal discomfort and increased urine output. For example, 35.71% reported tachycardia or palpitations immediately after the 9 mg/kg session, corresponding to five of fourteen persons if the percentage uses the full cohort. The placebo-session proportion for that item was 7.14%, about one of fourteen.

The same high-dose condition was associated with 35.71% reporting increased urine output 24 hours afterward, compared with 7.14% after placebo. These observations are important because a mean test-score benefit should not be considered independently of symptom burden. Yet the questionnaire is self-report: a complaint labeled “tachycardia” does not provide an electrocardiogram, a medically confirmed arrhythmia, a measured heart rate, or proof that caffeine directly caused a particular clinical episode. This was not a formal cardiovascular safety trial powered to identify uncommon serious events.

Evidence table 2: actual symptom-report percentages

*The source prints 12.5% for the immediate 6 mg/kg gastrointestinal item, even though its other percentages imply increments of approximately one fourteenth. That single entry should not be converted into a fabricated number of people. The table preserves the study’s printed value rather than silently “fixing” the paper. No original inferential p-value was provided for these descriptive symptom percentages.

Can the authors claim that 3 mg/kg is definitively safer or better than 6 mg/kg?

Only with care. Although the report characterizes 3 mg/kg as a potentially favorable low-dose option, the direct comparisons of 3 versus 6 mg/kg on the main reaction-time and attention outcomes were not statistically significant. A larger percent change at 3 mg/kg is not itself proof of superiority. The safety questionnaire also cannot precisely quantify a population-level difference in adverse-event risk between those doses in fourteen athletes.

Clinical or sports supplementation advice additionally requires consideration of age, individual tolerance, other stimulant exposures, underlying cardiac conditions, pregnancy or other physiological circumstances, sleep debt and the reason for taking caffeine. This research should not be repurposed into a prescription, a safe universal maximum or a claim that one dose prevents dangerous behavior. The investigators did not enroll a broad patient group or test medical outcomes over weeks or months.

What makes the study’s statistical evidence useful yet limited?

The within-person randomization is valuable because each athlete served as her own comparison. Testing at a consistent morning time and using a washout between sessions also addresses two possible confounders. The study reports strong omnibus differences for several cognitive endpoints and specifies Tukey-type comparisons, rather than only drawing subjective impressions from graphs. That makes it possible to distinguish positive from null findings in a transparent way.

Nonetheless, the sample is very small; multiple cognitive endpoints, pairwise dose comparisons and symptom questions were analyzed. The authors’ initial power calculation depends on an assumed effect size and does not make the study precise for every endpoint, person or adverse symptom. Some original statements conflate “no error effect” with “no mental-rotation effect,” and the symptom table includes an unusual percentage not aligned with a denominator of fourteen. These limitations counsel against turning point estimates into universal findings or reconstructing individual scores from the published summaries.

Was caffeine actually measured in whole blood, serum or plasma?

No. The study describes capsule administration and subsequent task outcomes, but does not report chemically quantified caffeine or paraxanthine in any participant’s legally collected venous whole blood. It also does not provide a serial postdose caffeine profile in separated plasma or serum. The authors expressly identify the lack of blood sampling as a limitation to confirming serum peaks across doses. Therefore, a comparison between 3, 6 and 9 mg/kg cannot be translated into particular nanogram-per-milliliter concentrations.

Nor did the investigators quantify parent caffeine or metabolites in urine, oral fluid, hair, sweat, vitreous humor, exhaled breath or another specimen. Their urine-output item was a symptom question, not a urine drug test. A hypothetical clinical or forensic laboratory analysis would need the exact specimen type, collection time, validated assay, calibration, reporting range, analytes, quality controls and the applicable interpretive purpose. One may consult the site’s guide to biological matrices to understand why different sample types are not interchangeable.

Why can’t a dose in mg/kg establish individual impairment or dose timing?

Milligrams per kilogram describes a mass administered relative to body weight; a toxicology concentration describes the quantity detected in a specific physical specimen at a specific time. Even when an administered oral dose is known, actual absorbed amount and concentration-time behavior vary. Different tasks might also respond differently at comparable exposure. Since no postdose chemical concentrations were determined here, the paper cannot estimate an individual participant’s legal whole-blood level, establish a last-use clock or supply a validated concentration linked to impairment.

The study contains no on-road driving assessment, simulator lane-position measure, emergency stop outcome, crash, eyewitness-identification test, or legal impairment classification. Better mean reaction time on a predictable computer signal is not a direct measurement of those outcomes. Likewise, a headache or heartbeat complaint is not a validated proof that an individual was too impaired to drive. The site discusses why chemical identification and quantitation, when available, require method-specific validation in its confirmatory drug-testing guide.

How should this experiment be used in a forensic or expert setting?

It can support a bounded explanation: acute oral caffeine affected particular performance scores in a small, selected, low-caffeine-use group, with no additional reported benefit at the highest dose and more descriptive symptom complaints. It also demonstrates that treatment effects can be task-specific, because mental-rotation time and mental-rotation errors produced different statistical conclusions. The experiment is a useful example of how nominal dosing, laboratory speed and symptomatic experience need not move in unison.

It cannot establish what was in a defendant’s or witness’s blood, whether another person had recently consumed caffeine, whether an accused driver’s performance was impaired, or whether a collision or injury would have occurred differently. Those claims require independently documented facts, qualified forensic assays when relevant, timing evidence, observed conduct and a case-specific scientific framework. For broader context about how actual timed biological concentrations differ from dosing assumptions, see therapeutic drug monitoring and interpretation; this particular caffeine experiment did not perform that monitoring.

Seven frequently asked questions about caffeine dose and attention

1. Did 9 mg/kg improve reaction time more than 3 mg/kg?

No such benefit was demonstrated. Simple and choice reaction-time comparisons at 9 mg/kg were not significantly better than placebo, whereas the 3 mg/kg comparisons were positive. That is a group result from these particular tasks, not evidence that every person has exactly the same dose response.

2. Were 3 mg/kg and 6 mg/kg significantly different on attention?

No significant direct difference was reported between those two doses on the short number-cancellation task, nor on simple or choice reaction time. The fact that 3 mg/kg had numerically larger percent changes does not establish statistically proved superiority.

3. Did caffeine improve mental rotation?

Mental-rotation time was shorter at 3 and 6 mg/kg in the reported comparisons, but mental-rotation errors showed no significant overall difference, p = .40. Treating those distinct variables as a single all-or-none “mental-rotation” result would misstate the source.

4. Were the side effects medically confirmed?

No. Symptoms were reported on questionnaires immediately and 24 hours later. The study did not establish electrocardiographic arrhythmias or a rate of clinical caffeine poisoning. The small number of participants also limits confidence in rare-event risk.

5. Does a 6 mg/kg caffeine capsule imply 6 ng/mL in blood?

No. The former is a dose per unit body mass; the latter would be a chemical concentration in a specified specimen. The trial did not quantify caffeine or its metabolites in human blood, plasma, serum or other samples, so no such conversion is supported.

6. Did the experiment show that caffeine makes driving safer?

No actual driving, crash avoidance or validated forensic impairment outcome was measured. A short, controlled reaction-time or number-cancellation task does not establish how a particular driver responded to an unexpected real-world hazard.

7. Is no significant benefit at 9 mg/kg proof that high-dose caffeine never affects cognition?

No. A nonsignificant comparison in fourteen selected young athletes means that this experiment did not demonstrate the specified benefit under its own conditions. It does not establish exact equivalence or exclude different effects in larger studies, other participants, other times or different tests.

Primary research reference and interpretive boundary

Bougrine H, Ammar A, Salem A, Trabelsi K, Jahrami H, Chtourou H, Souissi N. Effects of Various Caffeine Doses on Cognitive Abilities in Female Athletes with Low Caffeine Consumption. Brain Sciences. 2024;14(3):280. Original primary empirical source, DOI 10.3390/brainsci14030280. The study measured group-level cognitive tasks and self-reported symptoms following scheduled oral dosing. It did not establish participant-specific blood caffeine concentrations or legally calibrated conclusions about intoxication, driver behavior, event timing or causation.