Does Caffeine Speed Attention at Rest but Not During Exercise? A 24-Person Crossover Study

Short answer: In a controlled experiment involving 24 healthy young men, an oral caffeine dose of 4 mg/kg was associated with faster responses on a computerized attention task only while resting in the subgroup accustomed to moderate daily caffeine intake. In that subgroup, mean responses were approximately 37 milliseconds faster with caffeine than with placebo at rest. The comparable caffeine advantage did not appear during moderate cycling, and investigators did not demonstrate an overall caffeine-related speeding effect across every participant and activity condition. Exercise itself shortened responses by approximately 50 milliseconds, but it was accompanied by a small increase in errors. This is evidence about a specific attention experiment—not proof of any person’s blood caffeine concentration, fitness to drive, or cause of an accident.

The original primary investigation is Huertas, Blasco, Moratal, and Lupiañez, Caffeine intake modulates the functioning of the attentional networks depending on consumption habits and acute exercise demands, Scientific Reports 9:10043 (2019), doi:10.1038/s41598-019-46524-x. Its particular value for forensic interpretation is the contrast between a prescribed experimental dose, measured response behavior, independently manipulated physical exertion, and biological variables that were not measured.

What was the principal scientific question?

The authors investigated whether caffeine alters three partly distinguishable components of attention—alerting, orienting, and executive control—and whether those effects depend jointly on a person’s habitual caffeine consumption and current physical activity. Their design addressed a common weakness of broad statements about stimulants: a change seen in a resting computer task may not reappear when that person is simultaneously engaged in physical exertion. Habit also matters. A participant arriving after an instructed caffeine-free interval might respond differently if normally accustomed to caffeine than if he ordinarily consumes little.

Rather than asking whether caffeine universally improves cognition, the study asked which combinations of person, exposure, and immediate testing context were associated with changes in specified attention outcomes. That narrower framing is crucial. A statistically detectable effect for one subgroup and situation cannot automatically be carried across to the other subgroup, to a driver at an incident, or to a medically relevant concentration threshold.

Who were the participants, and what is the actual sample size?

The investigators selected 24 healthy male undergraduate sport-science students, aged 21–25 years. Twelve were categorized as low habitual caffeine consumers and twelve as moderate consumers according to a questionnaire about typical daily intake. The authors set the dividing point at 50 mg per day: the low group reported 50 mg/day or less, and the moderate group reported more than 50 mg/day. Mean estimated habitual intake was 14 mg/day in the low group and 101 mg/day in the moderate group. These categories describe reported consumption habits; they are not measured circulating caffeine concentrations or diagnostic classifications of tolerance.

Each person experienced four experimental combinations on separate visits. Four visits, hundreds of keypresses, repeated cardiac readings, and multiple alerting cues do not convert this experiment into 96 independently randomized people. The biological participant denominator remains 24, with 12 in each habit-defined group. That accounting matters when assessing precision, subgroup interactions, and how far a small study should be generalized.

What did participants actually ingest?

On caffeine visits, each person took a capsule containing 4 mg of pure anhydrous caffeine per kilogram of body weight. Placebo capsules contained pregelatinized starch and were matched in appearance. The administered amount was based on the person’s weight rather than a fixed 200-mg or 400-mg dose. The article’s reported overall mean body weight of approximately 73.7 kg does not establish the exact dose given to each participant; individual weights and capsule quantities determined those amounts. The study did not investigate intravenous caffeine, energy-drink formulations, methamphetamine, amphetamine, or interactions with alcohol.

The researchers used a double-blind intake comparison. Participants were instructed to avoid caffeine, other stimulants, and exercise for 24 hours before each session. They took the study capsule and waited roughly 45 minutes before testing. That interval was chosen to permit absorption; it was not an analytically measured plasma peak in this participant group. The resulting nominal dose in milligrams per kilogram must not be mistaken for a quantified ng/mL result in human blood, plasma, serum, oral fluid, or urine.

How did the four rest-and-exercise conditions work?

After an initial visit to determine physical workload, participants returned for four experimental visits, separated by approximately 48 to 96 hours. Each experienced caffeine at rest, placebo at rest, caffeine during cycling, and placebo during cycling, with the condition order counterbalanced using a Latin-square procedure. The two habitual-consumption groups were not created by randomly assigning caffeine habits: those groups were defined by the participants’ pre-existing reported behavior.

In the exercise condition, the participant cycled on an ergometer while taking the computerized attention test. Workload was set at 80% of the individual’s workload at his measured lactate threshold, with an approximately 10-minute warm-up before the attention assessment. At rest, the person performed the test sitting in a chair. A design that measures task performance both during exertion and without exertion can directly test the context-dependence of a caffeine effect, but it is still a specialized laboratory activity rather than a real-road driving experiment.

What did the attention task actually require?

The Attention Network Test–Interactions, or ANT-I, required speeded left-versus-right keypresses to identify the direction of a central arrow. Other arrows flanked the target and sometimes pointed in the opposite direction. An auditory warning tone appeared on some trials; a spatial cue preceded the target on some others. These manipulations allow the researchers to estimate three different performance components rather than collapse them into a single label such as “alertness.”

Participants practiced 48 trials and completed five 48-trial experimental blocks at each experimental visit. The test lasted approximately 25 minutes per condition. Researchers measured reaction time in milliseconds and incorrect responses as percentages. A trial with a flanker conflict, a warning tone, or a misleading spatial cue is still a computerized arrow-discrimination event. It does not reproduce the perceptual, legal, fatigue, multitasking, or threat conditions of an actual traffic encounter.

How did the investigators measure alerting, orienting, and control?

Alerting was operationalized as the reaction-time advantage associated with a brief auditory warning tone. A smaller alerting index can arise when no-tone responses become faster, when tone responses change, or both; it should not automatically be translated into diminished general consciousness or worse vigilance. Orienting concerned the time effect of a spatial cue, contrasting cued, uncued, and no-cue trials. Executive control concerned the interference produced by flankers pointing opposite the target arrow. These indices are task-defined differences, not clinical impairment scores or drug concentrations.

The authors also recorded overall reaction time and error percentages. That dual reporting matters: faster average responses can coexist with more mistakes. An effect on one attentional network does not justify claiming a significant effect on the others, particularly when the reported interaction analyses are nonsignificant or specifically restricted to selected groups and physical conditions.

What were the strongest findings in the original results?

The most robust overall performance contrast concerned physical activity: reaction times were approximately 50 milliseconds faster during exercise than at rest, with an activity main effect F(1,22) = 91.70, p < .001. The researchers also observed statistically significant activity-by-intake and activity-by-intake-by-habit interactions, p = .038 and p = .011 respectively. Those interactions are the basis for looking beyond the overall group average at the particular contexts in which caffeine did, or did not, affect response speed.

Among moderate habitual caffeine consumers tested at rest, the caffeine-versus-placebo comparison indicated approximately 37 milliseconds faster average response time after caffeine, F(1,11) = 16.68, p = .002. The authors reported no corresponding caffeine-speed effect in the other habit-by-activity cells, characterizing those tests as F < 1. This should not be rewritten as a claim that caffeine had a proven biologically zero effect in each other situation: failure to detect a difference in twelve-person subgroups is not demonstration of exact equivalence.

What do the original outcome and contrast figures show?

The following figures are taken from the investigators’ stated analyses. The effect sizes and inferential tests refer to different contrasts; they should not be merged into a single “caffeine effectiveness” percentage. Approximate changes in milliseconds reflect group-level response time rather than a person’s time to brake a vehicle.

Did caffeine also speed responses during cycling?

Not in the tested comparisons reported by these investigators. Exercise produced faster responses regardless of caffeine intake, while the additional caffeine-related speed advantage seen in moderate consumers at rest was not reproduced during cycling. The authors did not identify a significant caffeine main effect on overall reaction time when the intake variable was considered without the relevant interactions. Thus the data support a condition-specific observation, not the broad statement that caffeine always accelerates motor responses or invariably improves divided attention.

One possible interpretation discussed by the authors is that exercise and caffeine operate against different baseline arousal states, making an additional caffeine effect easier to observe when a habitual consumer is resting after caffeine restriction. That is a mechanistic interpretation, not a direct measurement of central catecholamine release in these participants. No brain dopamine, norepinephrine, or caffeine pharmacokinetic analysis established the pathway for any particular person.

Was the faster exercise response entirely beneficial?

No simple accuracy-free benefit follows. The overall error rate was slightly higher during exercise, by approximately 0.73 percentage points, and that activity effect was statistically significant. On the executive-control task, flanker-related error interference was also larger during exercise: approximately 3.10% versus 2.16% at rest, F(1,22) = 8.31, p = .009. These findings underscore the well-known problem of interpreting reaction time in isolation. If a condition changes a participant’s emphasis on speed versus accuracy, faster responses do not necessarily mean safer decisions.

Neither an abstract “stimulant benefit” nor an equally broad “stimulant impairment” adequately describes the multidimensional outcomes. The task required a fast keypress and penalized incorrect directions; a legal crash reconstruction concerns substantially different demands. The laboratory results can raise hypotheses about response selection under exertion, but they cannot determine whether a specific motorist was capable of responding appropriately to a particular unexpected hazard.

What happened to the alerting network?

Across the task, the auditory warning tone was associated with responses about 31 milliseconds faster than when no tone was present, F(1,22)=109.75, p<.001. Investigators found that alerting interacted with intake and activity, and reported a four-way interaction involving habitual consumption, F(1,22)=5.13, p=.034. In the moderate-consumption group at rest, caffeine reduced the alerting index, F(1,11)=5.15, p=.044. Among low consumers during cycling, a corresponding comparison reached p=.058, which does not satisfy a conventional .05 significance threshold.

It would be misleading to say that a “reduced alerting index” proves drug-induced inability to perceive warning sounds. The index is calculated from the advantage of a tone over no tone in a defined timing paradigm. When baseline/no-tone responding becomes faster, the additional advantage supplied by the tone may decrease. The source’s interpretation rests on tonic versus phasic activation; the experiment did not provide a clinical diagnosis of reduced alertness or document failures to notice an actual siren or traffic signal.

What happened to orienting and executive control?

Physical activity significantly changed orienting indices, F(2,44)=13.48, p<.001, while reaction-time orienting was not shown to be specifically modified by caffeine in the manner reported for moderate consumers’ overall resting speed. The source reported a statistically significant intake-by-orienting interaction for error percentages, F(2,44)=3.39, p=.043. The distinction between error-rate orienting and reaction-time orienting must be preserved: significance for one response measure is not evidence that the same drug-specific effect existed on the other.

Executive-control interference in reaction time was roughly 70 milliseconds, with F(1,22)=672.79, p<.001. The authors state that this interference effect was not modulated by the other experimental variables. The observed rise in executive-control-related errors during exercise is a different outcome. One cannot assign a single “executive functioning improved” label when the drug, workload, task definition, and measured performance index do not point to the same conclusion.

Which genuine nonsignificant findings prevent overstatement?

A rigorous reading gives the null results at least as much prominence as the positives. The researchers did not demonstrate a general caffeine-related heart-rate rise under these conditions (intake effect p=.321). They did not find a caffeine-related acceleration of average ANT-I reaction time during cycling. The low-consumer exercise alerting comparison at p=.058 was not significant. The executive-control reaction-time conflict effect did not show modulation by caffeine intake or consumption habit. These are bounded statements about specific tested comparisons; the absence of a result cannot be converted into proof of pharmacological inactivity in every person.

A relevant comparison table helps distinguish true statistical nulls from entirely unmeasured clinical variables. No statistical test can be called “nonsignificant” for a whole-blood caffeine concentration or roadside crash outcome that the study never measured in the first place.

Did the investigators analyze any blood samples?

Yes, but the analyte matters. The experimental protocol describes earlobe capillary blood collected for a portable lactate measurement during workload determination and around the attention task. Blood lactate in mmol/L helped confirm and characterize the physical exercise condition. The investigators also tracked heart rate in beats per minute. Neither measurement quantifies caffeine. A blood sample taken for lactate is not evidence that caffeine was assayed, and its result must not be relabeled as a whole-blood caffeine concentration.

For broader background on why a laboratory conclusion must name both its analyte and its biological compartment, see biological matrices in forensic toxicology and confirmatory drug testing. Those context pages are not additional experimental sources for the Huertas results.

What is the difference between the nominal 4 mg/kg dose and a forensic result?

The 4 mg/kg quantity is a record of administration. It does not reveal the quantity that entered the bloodstream at any specific time, the extent of caffeine metabolism, the distribution between plasma and cellular blood fractions, or its elimination rate for a particular participant. A concentration measurement would require an appropriate collected specimen, documented collection time, validated analytical procedure, and calibrated result. Even such a concentration cannot on its own classify individual impairment without a separate validated effect framework.

The source’s reference to a 45-minute postcapsule waiting period is likewise not proof that every participant reached a particular peak concentration at that time. The study used that delay as an experimental scheduling choice. The absence of direct chemical results rules out reconstructing an individual serum or whole-blood caffeine level from the experimental performance data. For context about the very different clinical use of timed concentrations, see therapeutic drug monitoring.

Could withdrawal explain why moderate consumers showed the clearest resting effect?

It is a plausible alternative mechanism, not an experimentally demonstrated diagnosis of withdrawal. Participants were asked to avoid caffeine for 24 hours before each session, while the moderate group ordinarily reported more caffeine exposure. The authors specifically discussed the possibility that the placebo-at-rest condition produced relatively slow performance in habitual consumers and that caffeine or physical activity diminished that disadvantage. This mechanism competes with an explanation framed as caffeine adding the same intrinsic speeding benefit to everyone’s baseline performance.

Importantly, there was no independently randomized assignment to lifelong low versus moderate caffeine intake, no separate prospective diagnostic assessment of caffeine withdrawal severity, and no quantified caffeine plasma curve. It would therefore overstate the source to conclude that withdrawal was the proven cause of the 37-ms contrast. The safer conclusion is that habitual intake modified the observed experimental effect under a specific rest condition and that the underlying explanation needs independent study.

Can this study establish that caffeine makes driving safer?

No. The experiment did not place participants behind the wheel, monitor actual collisions, administer field sobriety testing, or link chemical blood results to driving behavior. A central-arrow keypress, even while a person pedals, differs from lane maintenance, hazard perception, steering coordination, divided attention across mirrors, and speed-distance judgments in traffic. A group-level reaction-time advantage of 37 milliseconds in one subgroup is not a validated vehicle stopping-distance reduction. A separate increase in error rate under exercise illustrates why one speed measure alone would be an inadequate safety metric.

Equally, the study is not evidence that caffeine makes every driver unsafe. Its behavioral findings concern a task, an exposure, and a context. Forensic application would require the original incident’s chronology, medications and other exposures, actual scene and witness evidence, validated task-to-driving transfer, and any properly obtained chemical specimen. None is supplied by this paper.

What scientific limitations should accompany every citation?

The authors explicitly relied on 24 healthy male sport-science students and did not conduct a prospective power analysis; that restricts precision and population generalizability. Habitual-consumption groups were established by self-report rather than randomized assignment. All participants were relatively young, healthy, and experienced with physical exercise; the findings should not automatically be extended to older adults, women, fatigued shift workers, people with cardiopulmonary disease, or drivers under acute medical stress. The tasks were repeated on four visits, and practice, carryover, and prior caffeine abstinence can influence performance despite scheduling controls.

Multiple attention outcomes and interactions were tested, so isolated nominally significant subgroup comparisons should be considered alongside the full pattern, study design, and uncertainty. The paper does not report a chemically verified absorbed caffeine dose or plasma level, and any explanation involving withdrawal or monoamine mechanisms remains interpretive. In particular, exercise improved average keypress speed while mildly worsening error frequency: a reason to avoid reduction of attention to one supposedly universal stimulant-effect number.

Frequently asked questions

1. How many independent people were studied?

Twenty-four healthy male participants, twelve low habitual caffeine consumers and twelve moderate consumers. Four repeated testing visits and many within-visit trials were repeated observations of the same people, not independent new subjects.

2. Was the caffeine dose 200 mg?

No fixed 200-mg dose is specified. The investigators administered 4 mg/kg of oral anhydrous caffeine in a capsule, compared with a starch placebo. The actual milligram amount depended on each participant’s body weight.

3. Did caffeine significantly speed everyone up?

No. The reported caffeine-associated roughly 37-ms speed advantage was limited to moderate habitual caffeine consumers tested at rest, with p=.002. Investigators did not demonstrate the same additional speed effect during cycling or in the low-consumption subgroup.

4. Did exercise improve every measure of attention?

No. It reduced mean reaction time by about 50 ms overall but increased overall errors by about 0.73 percentage points. A faster keypress does not prove uniformly better accuracy, judgment, or driving safety.

5. Did a lower warning-tone alerting index prove reduced consciousness?

No. The alerting index is a mathematical difference in computerized response times with and without a tone. It is not a clinical measure of consciousness, perception of real alarms, or intoxication severity.

6. Was caffeine actually measured in whole blood or plasma?

No quantitative caffeine or metabolite assay is reported for venous whole blood, separated plasma or serum, urine, or oral fluid. Earlobe blood was sampled for lactate related to exercise physiology; that is a different analyte and scientific question.

7. Can these results explain a particular accident or establish impairment?

No. No real driving event or individualized legal impairment assessment was investigated. The study supports context-sensitive interpretation of group-level ANT-I behavior, not an opinion on the cause of a specific collision or an individual’s caffeine concentration or functional state at an incident.

What is the evidence-based bottom line?

Huertas and colleagues found that physical exertion, pre-existing caffeine habits, and oral caffeine administration interacted in meaningful but nonuniform ways in a 24-man attention experiment. Cycling accelerated average reaction time and modestly increased errors. An additional caffeine-associated speed advantage appeared among moderate habitual caffeine consumers only at rest; the authors did not identify a general caffeine-speed effect across all experimental conditions. Changes in tone-dependent alerting and orienting measures further illustrate why different endpoints should not be collapsed into one abstract concept of “better attention.”

For a forensic scientist, the defensible statement is the measured one: 4 mg/kg oral caffeine and moderate exertion produced context-dependent changes in computer-task performance in this selected group; capillary lactate and heart rate characterized exertion, but no forensic caffeine concentration or actual driving outcome was measured. Any stronger attribution requires separate evidence. The original peer-reviewed source remains Huertas et al., 2019, DOI 10.1038/s41598-019-46524-x.