EVIDENCE LIBRARY
Does Caffeine Prevent Attention Lapses After 26 Hours Awake? A 24-Person Trial Found a Benefit—and a Task-Fatigue Limit
A 24-person blinded crossover trial found caffeine reduced some vigilance deficits after 26 hours awake, but not the within-task rise in lapses or executive-task errors.
Answer: In a randomized, double-blind crossover experiment involving 24 healthy adults, caffeine reduced some objective vigilance deficits after approximately 26 hours awake. Yet it did not eliminate the progressive deterioration that developed during a single 10-minute attention task. The difference between caffeine and placebo in the increase in vigilance lapses from the beginning to the end of that task was not statistically significant. Nor did caffeine produce a general improvement in Go/no-go inhibitory control or two-back working-memory accuracy. These findings are useful for understanding fatigue-related performance, but they cannot establish an individual driver’s fitness, the timing of caffeine ingestion, or a measured blood caffeine concentration.
The primary evidence comes from Quiquempoix and colleagues’ 2022 randomized crossover study, Effects of Caffeine Intake on Cognitive Performance Related to Total Sleep Deprivation and Time on Task, published in Nature and Science of Sleep. Read the original primary study (DOI: 10.2147/NSS.S342922). The central distinction is between a caffeine-associated improvement in the overall level of vigilance and a statistically demonstrable change in how performance deteriorates while a repetitive task continues.
The experiment asked two questions, not one
The first question was whether deprivation of an entire night’s sleep would worsen vigilance, response speed, inhibition, and working-memory performance. The second was whether caffeine could counter those decrements, especially the decline that can occur as an apparently simple task continues. That second phenomenon is commonly called the time-on-task effect. A person might begin a monitoring assignment reasonably promptly and nevertheless respond less reliably near its end. A drug can improve average responses without changing that within-task trajectory.
That distinction matters in forensic interpretation. “Fewer lapses with caffeine” does not mean “no lapses after caffeine.” “Faster mean reaction time” is not equivalent to protection against each rare, consequential failure to react. And a finding measured on a computer vigilance task does not automatically transfer to a vehicle, industrial installation, operating room, or particular witness’s attention during a disputed event.
Who participated—and what counts as the sample size?
The investigators studied 24 distinct healthy adults, eight men and 16 women, with a reported average age of approximately 33.7 years. Participants were screened for relevant medical and sleep factors, and their habitual caffeine consumption was characterized. They completed both the caffeine and placebo conditions in a counterbalanced, double-blind crossover protocol. The same participant therefore served as their own comparison across experimental conditions.
The number of observations was much larger than 24 because each person completed several tasks, each task produced many stimulus responses, and the investigators examined early and late time windows. None of those repeated responses constitutes another independent human. Treating trial count, EEG epochs, conditions, or time windows as extra participants would artificially inflate the biological sample size and distort the study’s evidentiary weight. Its small, screened, predominantly female sample also limits generalization to diverse professional or clinical populations.
How much caffeine was administered?
The caffeine condition used an oral dose of 2.5 mg per kilogram of body weight supplied as caffeine powder in a decaffeinated drink; the comparison was the matched decaffeinated placebo drink. According to the experimental schedule, subjects received interventions at approximately 8:30 a.m. and 2:30 p.m. on the baseline day and at approximately 8:30 a.m. on the sleep-deprivation day. The principal cognitive testing session began at about 9:15 a.m. Participants underwent approximately 26 hours of continuous wakefulness before the deprivation-day testing.
The dose was prescribed according to body weight. It should not be converted into a universal fixed milligram dose for every participant. More importantly, a swallowed dose is not an analytically confirmed concentration in any collected human specimen. The study did not quantify caffeine in venous whole blood, separated serum or plasma, urine, or oral fluid; it therefore cannot define a toxicological threshold based on those matrices.
What exactly was measured?
Three ten-minute computerized cognitive tasks were used. The psychomotor vigilance test (PVT) measured speeded responses to visual signals and recorded delayed responses classified as lapses, including responses exceeding 500 milliseconds. A Go/no-go task measured response inhibition through erroneous responses when a response should have been withheld. A two-back working-memory task assessed accuracy when participants had to compare a current item with one presented two positions earlier.
Before testing, participants also rated sleepiness on the nine-point Karolinska Sleepiness Scale. Scalp electroencephalography measured electrical activity, with emphasis on theta-band spectral power in frontal, centro-temporal, and parieto-occipital regions. Those are different endpoints: a subjective rating, an electrical signal, task-specific behavior, and—if it had been performed—a chemical analysis of a biological specimen. Only the first three were measured here. No toxicological specimen measurement, road-driving outcome, or objectively verified collision was part of the study.
Vigilance worsened substantially after sleep deprivation
On the sleep-deprivation day, participants made more vigilance lapses and had slower average responses than on the baseline day, whether they received caffeine or placebo. The mean PVT response time in the placebo condition increased from 260 to 310 milliseconds; the caffeine-condition values were 252 and 295 milliseconds. This is consistent with a substantial adverse effect of prolonged wakefulness even in a brief, controlled, repetitive detection task.
Caffeine improved some measures relative to placebo on the deprivation day, including overall PVT response time and a subset of delayed responses. But the caffeine-condition mean of 295 milliseconds remained slower than its own baseline mean of 252 milliseconds. The proper conclusion is an incomplete, task-specific mitigation of sleep-loss effects, not restoration to rested performance. An average response may improve even while a participant still makes occasional slow responses that could matter greatly in a particular setting.
Two evidence tables: group effects and the within-task limit
Values below are reported study group summaries, not subject-specific diagnostics. The PVT lapse entries reproduce the paper’s numerical table; the original presentation of their rate units requires separate scrutiny before anyone uses them as a literal per-minute count. Mean response times use milliseconds, and working-memory scores are percentages. Parentheses indicating significance refer to the authors’ reported tests and comparisons rather than proof of general real-world benefit.
| Measured endpoint | Placebo at baseline / after sleep loss | Caffeine at baseline / after sleep loss | Interpretation |
|---|---|---|---|
| PVT mean reaction time, ms | 260 ± 5 / 310 ± 6 | 252 ± 5 / 295 ± 6 | Reaction time worsened after deprivation; caffeine treatment main effect F(1,23)=10.92, p<.01 |
| PVT slowest 10% reaction time, ms | 354 ± 8 / 430 ± 7 | 341 ± 9 / 411 ± 8 | Caffeine treatment main effect F(1,23)=4.714, p<.05 |
| PVT lapse entries, as numerically tabulated | 0.4 ± 0.2 / 9.8 ± 1.7 | 0.1 ± 0.09 / 5.5 ± 1.2 | Lower caffeine-condition entries after deprivation; source lapse unit label is not internally consistent with its minute-by-minute figure |
| Go/no-go commission error rate, % | 3.0 ± 0.5 / 5.7 ± 0.9 | 3.8 ± 0.5 / 8.0 ± 1.3 | No established caffeine rescue of inhibition |
| Two-back correct responses, % | 92.0 ± 0.7 / 90.2 ± 0.8 | 91.2 ± 0.9 / 86.1 ± 1.0 | The deprivation-day caffeine group did not outperform placebo; reported comparison favored placebo |
| Karolinska subjective sleepiness, scale points | 3.2 ± 0.3 / 5.6 ± 0.4 | 2.7 ± 0.2 / 5.0 ± 0.4 | Sleepiness rose in both conditions; deprivation-day treatment contrast not significant |
One must not merge PVT, Go/no-go, and two-back scores into a single invented “cognitive impairment percentage.” These tests have different scoring rules, response demands, and sources of error. A stronger result in the easiest detection task could coexist with little change, or a different directional result, on a demanding memory task. This selective pattern is precisely why source-specific interpretation is superior to a blanket assertion that caffeine “fixes cognition.”
The critical result: late-task deterioration persisted with caffeine
For the time-on-task analysis, the researchers compared the first three minutes with the last three minutes of each 10-minute task. This design asks whether performance degrades as testing continues, not merely whether one treatment has a better overall average. On the deprivation day, PVT lapses became more frequent near the end of the test in both conditions. The caffeine curve was lower, but the increase from early to late was not demonstrably smaller under caffeine.
The authors reported p = 1.0 for the comparison of the early-to-late PVT lapse increases between the two treatment conditions. This is a nonsignificant difference, not evidence that the treatment prevented deterioration. The distinction between a downward-shifted curve and a changed slope is consequential: caffeine may make several minutes of monitoring better on average without supplying statistically supported protection against fatigue accumulating over those minutes.
Average reaction time shows the same distinction
Within the deprivation-day PVT, mean response time increased from approximately 301 to 316 milliseconds between the early and late windows with placebo; with caffeine it increased from approximately 278 to 306 milliseconds. The general time-on-task effect was statistically significant, F(1,23)=32.2, p<.001, and a general treatment effect was also reported, F(1,23)=11.20, p<.01. But the treatment-by-time-on-task interaction was not significant, F(1,23)=0.42, p=.7.
That combination is not contradictory. A treatment can affect the overall elevation of response time while exerting no convincing effect on the within-test deterioration rate. In a forensic report, those two questions should be kept separate. One concerns whether group responses tended to be faster after caffeine; the other concerns whether caffeine altered the pattern of slowing from the start to the end. The experiment supports a difference on the former, not a clear differential treatment effect on the latter.
Go/no-go inhibition did not show a parallel benefit
On a Go/no-go task, participants had to respond to target stimuli and withhold responses to other stimuli. Commission errors are failures of inhibition; they are not the same thing as delayed PVT responses. After sleep loss, errors increased under both treatments. In the original numerical table, the deprivation-day Go/no-go error rate was 5.7% with placebo and 8.0% with caffeine. The paper did not establish that caffeine improved inhibition simply because it helped some PVT outcomes.
The within-task commission error burden also increased as minutes elapsed. The authors reported a p=.98 comparison of the treatment conditions for the change across the task, a genuine null contrast. Faster vigilance is not synonymous with better restraint, decision quality, or accuracy on a task where sometimes the correct response is not to act. An examiner who substitutes reaction speed for inhibitory control would be comparing different psychological constructs.
Two-back working memory likewise resisted a simple rescue narrative
The two-back task required subjects to track a short sequence and decide whether each new item matched one seen two steps earlier. That demands monitoring, updating, and accurate selection, not merely a rapid motor response to one cue. The original table reported 90.2% correct answers after deprivation with placebo and 86.1% with caffeine. The authors indicated a statistically significant unfavorable comparison for caffeine in that deprivation-day aggregate.
Both conditions also showed a loss of correct responses over the 10-minute task. The between-treatment difference in the early-to-late change was not significant (p=.48). Consequently, it would be inaccurate to portray caffeine’s vigilance advantage as a demonstrated improvement in all facets of executive functioning. The study is better understood as evidence of endpoint-specific caffeine effects under experimental sleep deprivation, with some remaining deficits and potentially unfavorable differences in a more complex task.
Why the time-on-task interaction matters more than a striking average
When considering a long monitoring period, an overall group mean may conceal how a participant’s response pattern changes with sustained demand. The first few minutes may differ from the last even if the stimulus type is unchanged. A statistically significant average drug effect cannot be used to assert a statistically significant difference in the shape of a decline; that requires its own appropriate interaction or difference-in-change test.
The study’s comparison windows were not arbitrary extra clinical encounters. They were repeated segments within one short task. Nor do these windows permit extrapolation to an eight-hour flight, a night shift, a 30-minute drive, or a critical incident involving a rare unpredictable signal. Those settings differ in workload, motivation, competing cues, stressors, circadian timing, and opportunities to respond. The proper evidentiary use is to explain a laboratory limitation, not to calculate an individual event probability from these group means.
Which outcomes actually changed over a ten-minute task?
The second table separates statistically supported time-on-task patterns from unsupported assertions about treatment-by-time interactions. Its entries refer to the original analyses of the deprivation-day observations. In the table, “nonsignificant” means that the reported comparison did not establish an effect at the study’s chosen statistical criterion; it does not prove perfect equivalence between the treatments.
| Endpoint and comparison | Original reported result | What it establishes |
|---|---|---|
| PVT lapse early-to-late increase, caffeine vs placebo | Between-condition change p=1.0 | No demonstrated smaller within-test lapse accumulation with caffeine |
| PVT mean response time, time-on-task main effect | F(1,23)=32.2, p<.001 | Responses slowed later in the test across conditions |
| PVT mean response time, caffeine treatment main effect | F(1,23)=11.20, p<.01 | Overall treatment-associated response-time advantage |
| PVT mean response time, treatment × time-on-task | F(1,23)=0.42, p=.7 | No established treatment difference in the time-on-task slope |
| Go/no-go commission error change, caffeine vs placebo | p=.98 | No established treatment rescue of growing inhibition errors |
| Two-back correct-response decline, caffeine vs placebo | p=.48 | No established treatment mitigation of within-task memory accuracy loss |
These are not six replications of one outcome. The tests differ and the models vary. Exact confidence intervals for a legal subject-level prediction were not furnished for these cross-task claims, and multiple exploratory endpoints increase the importance of reading the original methods and numerical tables. A defensible summary records both significant and nonsignificant results rather than turning the most favorable p value into a global conclusion.
Sleepiness ratings remained elevated
On the Karolinska Sleepiness Scale, reported sleepiness was higher after sustained wakefulness in both treatment conditions. The deprivation-day mean was 5.6 under placebo and 5.0 with caffeine, but the authors did not find a statistically significant caffeine-versus-placebo difference at that point. The values are ordinal questionnaire ratings, not an assay of caffeine or a direct measure of the frequency of unsafe actions.
Someone can feel somewhat more alert and still make long-latency responses. Conversely, a report of feeling sleepy is not a chemical identification of an ingested substance or a diagnosis of legal impairment. A forensic evaluation should specify whether the asserted endpoint is self-perceived sleepiness, monitored cognitive performance, scalp brain activity, objectively recorded sleep, or an actual event. This paper measured some of these endpoints but not all; conflating them produces a narrative stronger than the underlying data.
What did the scalp EEG theta signal mean?
The investigators recorded theta-frequency electrical power from electrodes grouped by scalp region. They observed sleep-deprivation-associated changes and some regional differences between the caffeine and placebo conditions. A selected parieto-occipital PVT comparison illustrates the issue: the table reported higher theta values after deprivation under placebo than under caffeine. But a regional theta response is an electrophysiological correlate during a particular task, not a chemical analysis of circulating caffeine or an exclusive readout of adenosine-receptor activity.
In one parieto-occipital PVT time-on-task analysis, the caffeine treatment main effect itself was nonsignificant (p=.07), as was the treatment-by-time interaction (p=.67). A visually compelling brain waveform cannot substitute for those inferential results. Theta spectral power is also not a validated individual threshold for driving or witness reliability. It may aid a mechanistic research discussion, but a specific person’s alertness, drug concentration, and moment-by-moment response should not be reconstructed from group EEG averages.
The toxicology distinction: administered milligrams are not measured ng/mL
The experimental intervention was an oral caffeine dose assigned by body weight. That establishes a study protocol and, assuming compliance, nominal exposure. It does not establish the concentration of parent caffeine at any time in a legally relevant physical specimen. None of the reported figures supplies a calibration curve, specimen collection and preservation protocol, instrument-specific limits of detection, analyte identification transitions, extraction recovery, measurement uncertainty, or a human whole-blood drug concentration.
Clinical and forensic samples are not interchangeable without validation. Venous whole blood contains cells and plasma; separated serum and plasma are distinct matrices, and measured concentrations can differ depending on drug distribution and laboratory method. Urine can reflect elimination, oral fluid can be affected by secretion and collection, and exhaled breath requires an analyte-specific validated assay. For background on these differences, see the independently published biological matrices in toxicology guide. A controlled oral intervention can be scientifically rigorous even when it does not collect the specimens a forensic case requires.
No measured caffeine blood concentration means no concentration–performance rule
Because the researchers did not assay participant caffeine concentrations, they could not estimate a within-person relationship between chemically confirmed caffeine levels and the late-task lapse slope. They also could not provide a serum-to-whole-blood conversion or calculate the caffeine concentration at the moment of a real incident. Variables such as distribution, absorption, prior regular consumption, timing, and metabolism would matter in any attempted retrospective inference, and those quantities cannot simply be filled in from group assignment.
In a toxicological dispute, confirmatory drug testing ordinarily asks whether a particular molecular analyte was identified in a specified specimen using a method fit for that matrix. See the independently published confirmatory drug-testing discussion. This experiment answers a different question: whether a controlled administered caffeine condition changes laboratory task outcomes in a crossover comparison. Both can be valid sciences without one proving the other’s central proposition.
Can the study establish that a tired person was safe to drive?
No. There was no licensed on-road driving, no actual collision, no standardized individual fitness-to-drive adjudication, and no event-level link between a person’s PVT lapse and a traffic hazard. The experiment also did not establish a concentration threshold at which caffeine renders sleep-deprived drivers safe. Some PVT differences favored caffeine while degradation within the task remained and other cognitive measures did not consistently benefit.
An individual assessment would require independent evidence about the actual event and its timeline, conditions of wakefulness, relevant medications, clinical state, environmental demands, observations, and any properly collected and validated chemical measurements. The absence of a significant interaction in 24 screened volunteers cannot exonerate or condemn another individual in a distinct setting. Equally, it would be wrong to conclude that caffeine has no benefit whatsoever: the controlled vigilance outcomes show a qualified group effect. Accuracy means retaining both halves of that conclusion.
How should an expert report use these findings?
A useful report identifies the biological population, exact intervention, physical specimens actually assayed, response definitions, and the statistical comparison corresponding to each proposition. It should say that 24 healthy adults completed a within-person caffeine/placebo crossover and that PVT response speed and overall lapses after approximately 26 hours awake were more favorable under caffeine on some comparisons. It should then say that caffeine did not significantly reduce the *increment* in lapses within the ten-minute task and did not reliably improve inhibitory control or working-memory task results.
Such an analysis should not convert a source table’s inconsistent lapse-unit labeling into a newly precise per-minute case estimate. Where a number or denominator is internally ambiguous, the source should be checked and the ambiguity made explicit rather than silently repaired. The article itself is evidence of research findings, not an authorized expert opinion about a litigant or an independent laboratory report. Therapeutic drug monitoring and concentration interpretation provide useful adjacent context but do not create missing concentrations for this trial.
Why the study matters despite its limitations
The most scientifically valuable contribution is its separation of three questions that are often blurred: Does sleep deprivation harm attention? Does caffeine improve a task’s average response profile? Does caffeine specifically change the process of task-related decline as the work continues? The answers were broadly yes, sometimes yes, and not convincingly on the central within-task comparisons. A sophisticated interpretation preserves those distinctions rather than declaring one stimulant either a universal antidote or entirely ineffective.
Forensic science benefits from research that reports null as well as positive endpoints. Here, the mixture constrains what could responsibly be said about performance under extreme wakefulness. The source was a small, highly controlled experiment, not a personal toxicology test, dose-response blood study, aviation certification, or reconstruction of a particular driver’s or witness’s decisions. Those are boundaries of application, not defects to be hidden.
Frequently Asked Questions
1. Did caffeine prevent all attention lapses after 26 hours awake?
No. Compared with placebo, caffeine was associated with fewer PVT lapses overall on the deprivation day, but lapses still occurred and performance remained affected by sleep loss. The key within-task comparison found no significant between-treatment difference in the increase in lapses from the first three to the last three minutes (p=1.0). Fewer average failures cannot be restated as zero failures or as complete restoration of rested vigilance.
2. Is the reported PVT lapse count a legally usable lapses-per-minute rate?
Not without further reconciliation. The printed source table labels its entries as a rate while its numerical values and minute-by-minute figure do not align cleanly with that unit. The entries may be discussed as originally reported group summaries, but their literal per-minute interpretation should not be used to derive a subject-specific danger frequency. The statistical finding of a caffeine-associated overall lapse benefit is a separate claim from a precise calibrated rate.
3. Why did caffeine improve PVT performance but not working memory?
These are different cognitive demands. The PVT tests responding promptly to a simple cue; Go/no-go tests response inhibition; two-back requires continuous updating and matching of items in working memory. A drug’s average effect on one endpoint does not predict equal effects on all others. In this original study, the deprivation-day two-back accuracy was not improved under caffeine, and the within-task accuracy decline was not significantly smaller than under placebo.
4. Did the EEG recordings prove that caffeine corrected the brain’s sleep-loss state?
No. Regional theta power changed in some task conditions, but it remained a scalp electrical signal, not a specific biochemical measurement or validated individual neurodiagnostic. Source-level effects depended on region, test, and model. At least one regional caffeine treatment contrast was nonsignificant, as was the relevant interaction. The experiment did not directly measure adenosine-receptor occupancy or declare a particular person’s brain function restored.
5. Were caffeine concentrations measured in venous whole blood or plasma?
No. The protocol specified caffeine amounts in milligrams per kilogram delivered orally. It did not report a quantitative laboratory assay of caffeine in venous whole blood, centrifuged plasma, serum, urine, or oral fluid. Consequently, the paper cannot establish an individual concentration, an analytical cutoff, an elimination curve, or an evidence-based conversion between matrices. A measured behavioral effect is not a substitute for a measured chemical analyte.
6. Can this experiment establish the cause of a driving collision or an unreliable witness?
No. The researchers measured computer tasks, perceived sleepiness, and scalp EEG in 24 selected volunteers—not a real driving episode, collision, witness identification, or particular legal proceeding. Even statistically significant group differences cannot alone determine whether a specific person’s reaction was delayed when a hazard appeared or whether testimony is accurate. Those propositions require additional event-specific evidence and appropriate inferential limits.
7. What is the precise finding an expert can cite?
In a blinded within-person crossover involving 24 healthy adults, oral caffeine at 2.5 mg/kg improved selected psychomotor vigilance outcomes after approximately 26 hours awake, but the authors did not demonstrate a caffeine-dependent attenuation of the increase in lapses over a ten-minute task. Go/no-go and two-back time-on-task changes were likewise not significantly different between treatments. The experiment did not measure caffeine in forensic biological specimens or assess real-world individual driving safety. Consult the complete original research publication (DOI: 10.2147/NSS.S342922) for methods, tables, figures, and the original statistical comparisons.