Caffeine

It is the most widely consumed psychoactive substance in the world. It occurs naturally in the coffee bean, the tea leaf, cacao and the guarana seed, and it is added to energy drinks, cola soft drinks, sports supplements and some painkillers. In Australia, where there is recent national data, around 90% of adults consume it regularly, with an average of about 175 mg a day: roughly two espressos.

At the doses at which it is consumed, only one mechanism has been demonstrated: occupying the site of adenosine, the molecule that builds up in the brain during waking hours and produces sleep pressure. Caffeine does not provide energy. It covers the tiredness signal for as long as its effect lasts.

Caffeine is a non-selective antagonist of the four adenosine receptors, A1, A2A, A2B and A3. After a 160-milligram coffee the plasma peak is around 20 micromoles, a concentration sufficient to displace adenosine from its receptors and sufficient for no other demonstrated mechanism. The other two mechanisms that appear in every review — the inhibition of phosphodiesterases and the release of calcium from the sarcoplasmic reticulum — require millimolar concentrations, between 50 and 500 times higher. They are real mechanisms, but irrelevant outside an intoxication.

The effect on dopamine is indirect and structural. A2A and D2 receptors form heteromers in the striatum, where occupancy of A2A exerts a tonic inhibition on D2. By blocking A2A, caffeine removes that inhibition. The difference from the classic psychostimulants is not one of degree but of mechanism: those increase the availability of dopamine; caffeine merely lifts a pre-existing brake.

Caffeine metabolism generates three active derivatives. Paraxanthine accounts for between 80% and 90% of the demethylation and is responsible for much of the effect attributed to the parent compound; theobromine and theophylline make up the rest. The plasma ratio of paraxanthine to caffeine serves as a phenotypic marker of CYP1A2 activity.

It sustains alertness and shortens reaction time. The clearest benefit appears when sleep is lacking: in candidates for the US Navy SEALs, after some 72 hours without sleep, a single dose improved vigilance and reaction time against placebo. Working memory is not among what improved.

In someone who drinks coffee every day and has slept well, much of what they notice with the first coffee is not an improvement on their normal state but the return to it after a night without a dose.

Does caffeine improve anyone, or does it simply put them back where they were? Since most of the population consumes it every day and its half-life is around five hours, someone who reaches the laboratory without caffeine since the night before reaches it in withdrawal, and what the trial measures as an effect may be the repair of that withdrawal.

Neither position has won. The largest trial found an improvement in alertness in medium-to-high consumers and almost nothing in low consumers, beyond taking away their sleepiness. Another group found improvements in alertness and mood in both and concluded that withdrawal reversal does not explain everything. They agree on one thing: pure speed, motor performance, improves in every group.

By domain, the map looks like this:

The reference trial on sleep deprivation worked with candidates for the US Navy SEAL program, put through some 72 hours without sleep and continuous stress, with 100, 200 or 300 milligrams against placebo. Vigilance, reaction time, repeated motor learning, and self-reported fatigue and sleepiness improved. Marksmanship, which demands fine motor control, was not affected, and working memory is not among what improved. The authors pointed to 200 milligrams as the optimal dose.

The debate on withdrawal reversal rests on two incompatible designs. One imposed a night of deprivation on 212 medium-to-high consumers and on 157 low or non-consumers, administered 250 milligrams in two doses and found that the improvement in alertness was concentrated in the former. The other used 75 or 150 milligrams, did not impose prolonged abstinence and found improvements in alertness and mood in consumers and non-consumers alike.

The methodological question neither of them settles is how much washout is needed for a habitual consumer to stop being in withdrawal. With a half-life of five hours and a documented syndrome duration of up to nine days, one night without caffeine is not enough, while a nine-day washout creates a subject who no longer represents the reader. As long as the designs force a choice between the two biases, the debate will stay open.

The Australian Institute of Sport considers that there is solid evidence of improvement in endurance lasting more than an hour, in intense efforts of 1 to 60 minutes, in team and intermittent sports, and in single efforts of strength or power, which is where the measured effect turns out to be smallest. The same source warns that the individual response varies: there are people who do not respond and people who respond worse.

The proposed mechanism is not muscular but nervous: caffeine reduces the perception of effort, which would allow pace and technique to be held for longer. Trained cyclists cut their time trial by 4.2% with 3 milligrams per kilogram of body weight. The Australian Institute of Sport classifies it in Group A, its category of solid evidence for specific situations in sport.

The effect size in sport runs from small to moderate, between 0.22 and 0.68 in the review that pools twenty-one meta-analyses, with the highest values in aerobic endurance and the lowest in maximal strength and power. In strength the pooled effect is 0.20 and in power 0.17, with intervals that graze zero.

At high doses, the trade-off is tremor and loss of fine motor control, a problem in precision sports. For very long events, such as an Ironman, the Australian Institute of Sport notes that doses above 3 milligrams per kilogram might be appropriate. There is no trial demonstrating it, and it is the only point at which this guide reports a figure above the ceiling it itself describes: the prudent course is to discuss it with a professional.

The umbrella review, which brings together twenty-one meta-analyses, places the effect sizes between 0.22 and 0.68 depending on the outcome. In maximal strength the pooled effect is 0.20 with a confidence interval of 0.03 to 0.36, and in power 0.17 with an interval of 0.00 to 0.34: the lower bound touches zero. The split by body segment adds another irregularity, because the effect appears in the upper body and not in the lower body, with no established physiological explanation.

The mechanisms proposed for the ergogenic effect are central: a reduction in the perception of effort and of pain, and easier recruitment of motor units. The peripheral contribution, by way of calcium release, requires the millimolar concentrations already described and does not operate at physiological doses.

The two reference institutional positions differ on the upper dose limit. The Australian Institute of Sport places the plateau at 3 milligrams per kilogram and proposes starting with the lowest effective dose, which it puts at 1.5 milligrams per kilogram or about 100 milligrams. The position stand of the sports nutrition society works with a range of 3 to 6 and allows up to 9 in some contexts; it comes from a group with declared ties to the supplement industry. This guide adopts the former as its reference.

With the most precise technique available, a dose of caffeine reduced insulin sensitivity by 15%. The rise in blood pressure was still significant after four weeks of daily consumption: tolerance to the pressor effect exists, but it is incomplete.

The first randomized trial to follow a hundred adults with continuous electrocardiogram monitoring for two weeks gave a mixed result. Coffee did not increase atrial premature beats, which was what had been feared, but it did increase ventricular ones. On coffee days, participants logged a thousand more steps and slept less.

A cup of coffee reduces the absorption of non-heme iron, the kind that comes from plants, by 39%. The effect is due to polyphenols, not to caffeine, which is why decaf inhibits it just the same. Taken an hour before the meal, the effect disappeared; an hour after, it stayed intact.

The 15% reduction in insulin sensitivity was measured with a hyperinsulinemic euglycemic clamp and intravenous caffeine: it is the cleanest design available for this outcome, but also the furthest from real consumption.

The blood pressure study administered caffeine daily for four weeks to participants stratified by previous consumption. The pressor elevation was still statistically significant at the end of the period in every stratum, which indicates that tolerance to the cardiovascular effect is partial, not complete, unlike tolerance to the subjective effect.

The randomized trial with continuous electrocardiogram monitoring for two weeks in a hundred adults is the most recent work in this block. Its result was mixed: there was no increase in atrial premature beats, which was the main hypothesis, but there was an increase in ventricular premature beats, which was not. On coffee days, participants logged 1,058 more steps and slept less. The design does not allow the effect of caffeine to be separated from that of the added activity.

On iron, the 39% inhibition was measured with labeled non-heme iron. The causal agents are the polyphenols, so decaf inhibits absorption to the same extent. The separation in time works in one direction only: coffee taken an hour before the meal produced no inhibition, while coffee taken an hour after produced it in full.

Mendelian randomization is the instrument that has done most to erode coffee’s protective reputation. The method takes advantage of the random allocation of alleles at conception to approximate a natural randomized trial. A review of fifteen such works finds no consistent support for a causal role of coffee in the diseases where observational epidemiology attributes protection to it. The method has its own limits, which the review declares: low power, pleiotropy and collider bias. It does not demonstrate that no causality exists; it shows that the available evidence does not support it.

It has been studied above all in people who have to perform while awake when they should be sleeping: night shifts, on-call duty, driving in the small hours. It does not replace sleep, and its effect on the sleep that follows is documented.

On a real motorway in the small hours, 200 mg of caffeine in coffee reduced drivers’ lane departures, as did a thirty-minute nap. For shift work disorder, trials in real workers are scarce.

In children and adolescents with ADHD, the meta-analysis of the available trials finds no benefit on symptoms.

In older adults, the effect depends on the task and does not lend itself to a summary. With seven days of prior washout and 100 milligrams, performance on interference tasks and reaction time improved, while in a rapid attention test accuracy improved in one condition and worsened in another.

In team and racket sports the improvement is split between physical work and concentration.

On ultra-endurance there is no answer: no solid randomized trial has been located in ultramarathon or ultratrail.

The sex gap is structural, not anecdotal. An audit of 1,826 studies on supplements with evidence found that only 23% of participants were women and that between 0% and 2% of the works were designed to compare responses by sex. Recent meta-analyses that look specifically at female athletes find no clear differences between phases of the menstrual cycle, but none of the included studies verified the phase by hormonal measurement, the only reliable method.

The Australian Institute of Sport places the ceiling of the effect at 3 milligrams per kilogram, about 210 mg for someone weighing 70 kg, and proposes starting with 1.5 mg per kilogram, around 100 mg. In the reference trial, doubling the 3 mg/kg did not improve time trial performance and did increase adverse effects.

The timing allows for variants: up to an hour before, spread out during the effort, or well into it when fatigue appears. The three schedules have given equivalent results in trials. Effects have been observed with capsules, coffee, gel, bars, chewing gum and dissolvable oral strips, which dissolve in the mouth and are swallowed. Mouth rinsing is a different matter: it is spat out without being swallowed and is the route with the least backing, as explained further down.

The Australian Institute of Sport holds that abstaining from caffeine in the days before a competition does not increase the benefit.

Trials have administered the dose between 60 and 90 minutes before the effort, with equivalent results. Format does change the kinetics: chewing gum reaches its peak between 44 and 80 minutes, against the 84 to 120 of the capsule, because part of the caffeine is absorbed through the lining of the mouth. Mouth rinsing, spat out without being swallowed, is what promises most and delivers least: of eleven crossover trials, eight found no effect.

It is worth undoing a confusion that this very guide may sow. In sports performance, the Australian Institute of Sport holds that consumers and non-consumers respond alike and that abstaining before competing adds nothing. In mental alertness, the literature points to much of the effect in habitual consumers being withdrawal reversal. They do not contradict each other: they are talking about the muscle and about the head.

The individual response varies a great deal: there are people who do not respond and people who respond worse. It has not been banned in competition since 2004, although it remains on the World Anti-Doping Agency’s monitoring program.

The pharmacokinetics by format have been measured. Chewing gum reaches its peak between 44 and 80 minutes, against the 84 to 120 of the capsule, because a fraction of the caffeine is absorbed by the oral route and avoids first-pass metabolism. The functional equivalence between coffee and caffeine in capsules is documented for endurance outcomes.

Mouth rinsing is the case in which the distance between mechanism and result is greatest. Of eleven crossover trials, eight found no effect on performance, and the most recent meta-analysis places the effect between trivial and small. The existence of oral receptors for caffeine is established; their practical relevance is not.

On the combination with creatine an antagonism circulates that the evidence does not support. The foundational study, with nine participants, observed that adding caffeine cancelled out the improvement obtained with creatine loading, and it has not been replicated in more than twenty-five years. The later systematic review concludes that creatine loading followed by an acute dose of caffeine does not interfere.

In the anti-doping arena, caffeine came off the list of prohibited substances in 2004 and remains on the monitoring program, which records urinary concentrations without sanctioning them. The NCAA keeps a limit of its own of 15 micrograms per milliliter in urine, a threshold that requires intakes far above the usual ergogenic doses.

To sustain alertness without sleep, 200 mg was the optimal dose of the three tested in the SEAL study. EFSA places its reference limits for healthy adults at 200 mg per dose and 400 mg a day. Above the ergogenic ceiling, the time trial measured no further performance, and the Australian Institute of Sport lists among the effects of high doses anxiety, digestive discomfort, lack of concentration and disturbed sleep.

Half the dose is still in the body about five hours after taking it. Four hundred milligrams taken four hours before bed took away 50 minutes of total sleep and 30 of deep sleep; at eight and at twelve hours before, deep sleep was still impaired. One hundred milligrams did not move any measure, not even at four hours.

Participants only said they had slept worse when they had taken it four hours before: at eight and at twelve hours, deep sleep degraded without their noticing.

The body of trials yields about 45 minutes less total sleep and 7% lower efficiency, with no effect on REM. From that come two rough cut-offs: about nine hours before bed for a coffee and about thirteen for a pre-workout serving.

You sleep worse, you need more caffeine the next day, and that caffeine ruins the night again. Long trials that quantify the whole loop are lacking, although each stretch of it has been measured separately.

The meta-analysis of the sleep trials quantifies the pooled effect: 45 minutes less total sleep, a 7% drop in efficiency, nine minutes more latency, twelve minutes more wake after sleep onset and 11.4 minutes less deep sleep, with no detectable effect on REM. From those figures come the two cut-offs that circulate as a practical recommendation: 8.8 hours before bed for a dose equivalent to a coffee and 13.2 hours for a pre-workout serving.

The at-home trial with polysomnography and seven conditions provides the resolution that was missing, by crossing two doses with three timings. Its distinctive finding is not the magnitude of the effect but the dissociation between what was measured and what was perceived: the degradation of deep sleep at eight and at twelve hours was not reported by the participants. An earlier trial that also found alteration at six hours used a consumption device whose manufacturer was among the declared conflicts of interest of the work.

The half-life is around five hours, with a documented population range of 2.3 to 9.9: the same coffee can last up to four times longer in one person than in another. More than 90% is metabolized by a single liver enzyme, CYP1A2. Smoking shortens the half-life by up to 50%; oral contraceptives roughly double it and pregnancy stretches it to about fifteen hours in the third trimester.

The CYP1A2 variant marketed as a marker of caffeine response comes from a finding that independent groups have not reproduced in basketball, sprinting, strength or adolescent populations, and the systematic review that brings them together rates the evidence as inconsistent. The Australian Institute of Sport states that it remains unclear whether genetic differences explain the opposing responses.

The documented population half-life runs from 2.3 to 9.9 hours, and CYP1A2 activity varies up to sixfold between individuals. Known inducers and inhibitors act on that baseline variability. Tobacco induces the enzyme and shortens the half-life by up to 50%, so quitting smoking doubles exposure to the same dose and can produce symptoms of excess without any change in consumption. Oral contraceptives inhibit it and roughly double the half-life. During pregnancy the lengthening is progressive and reaches about fifteen hours in the third trimester, while the fetus lacks the enzyme.

The case of commercial genotyping deserves attention because it is the clearest example of how funding can shape a line of evidence. The original trial, with 101 athletes and three doses, reported an overall improvement of 3% which, when stratified by the rs762551 polymorphism, became a 6.8% improvement in one genotype and a 13.7% worsening in another. The article declares as its funding sources the Canadian Institutes of Health Research, the Canadian Foundation for Dietetic Research, Nutrigenomix Inc., The Coca-Cola Company and Mitacs. It further declares that, at the time of publication, the senior author was a founder and shareholder of Nutrigenomix Inc., the company that markets the test, and that the first author sat on its scientific advisory board. Replication attempts in basketball, sprinting, strength and adolescent populations did not reproduce the interaction; a specific trial also found no interaction between genotype and caffeine, and the systematic review that brings the set together rates the evidence as inconsistent.

There is also a nomenclature problem that is rarely declared: different primary sources assign the fast and slow metabolizer labels to the same alleles in an apparently contradictory way, to the point of prompting a formal critical letter in the literature. It is one more reason not to build a protocol on genotype.

The rs5751876 variant of the ADORA2A receptor has been associated with greater sleep sensitivity to caffeine, although the evidence is preliminary and the allele assignment is not free of paradoxes either.

“A cup of coffee” is not a quantity. When twenty espressos from twenty different coffee shops were measured, the caffeine per serving ran from 51 to 322 mg: a sixfold difference for ordering the same thing.

A 250 mL cup of filter coffee is around 80 mg, with a range of 40 to 110; instant, about 60. Decaf is not zero either: it contains from 0 to 13.9 mg per large serving in the samples analyzed.

Tea has less, though with an equally wide range: from 25 to 110 mg per cup in black tea and from 30 to 50 in green. Herbal infusions, such as chamomile or rooibos, contain none: they do not come from the tea plant. A 60-gram bar of dark chocolate provides 10 to 50 mg, and a 250 mL can of energy drink, about 80.

The studies that associate coffee with less diabetes, less liver disease or lower mortality do not allow that effect to be attributed to caffeine: decaf shows similar associations. They are also observational: they describe who drinks what and who falls ill, without demonstrating that the one causes the other.

The variability of coffee does not stop between coffee shops. In the Australian Institute of Sport reference table, the same drink bought at the same outlet on different days ranges between 130 and 282 milligrams. With that spread, coffee is a poor vehicle when a specific dose is what you are after. In that same table, a 600 mL filter coffee from an international chain is listed at 415 milligrams on average, more than double EFSA’s reference limit for a single serving.

What coffee provides and caffeine does not are polyphenols, chlorogenic acids and diterpenes, and it is the last of these that are responsible for the only effect of coffee on lipids demonstrated in a randomized trial.

Tea is another matrix. It contains L-theanine, an amino acid credited with a synergy with caffeine that the data do not support at the proportions of a real cup: in the trial that used them, the combination cancelled out the behavioral effects of caffeine instead of enhancing them. The trials with a positive result use L-theanine doses far above those of any tea. The most recent meta-analysis on the subject declares that, at the time of publication, the doctoral grant of its first author was funded by Lipton Teas and Infusions and two of its co-authors were employees of Lipton Teas and Infusions and of Unilever.

The data on real content come from two kinds of source that are best not mixed. Chromatography measurements on drinks bought in real establishments give the spread: 51 to 322 milligrams per serving in twenty espressos from twenty coffee shops. Institutional tables give mean reference values with their range, and those are the ones used in this guide’s table.

That distinction explains a discrepancy that would otherwise look like an error. The institutional table attributes 25 to 110 milligrams per 250-milliliter cup to black tea, while a recent analytical study measured about 14 milligrams per 200 milliliters under standardized brewing conditions. They do not contradict each other: the laboratory fixes time, temperature and leaf quantity, and the table records what people brew. Brewing time is the dominant factor: a yellow tea brewed for ten minutes reached 53 milligrams per 200 milliliters in that same study.

Roasting does not destroy caffeine in any appreciable amount, but it does change the density of the bean: the loss of water and the gain in volume mean that a dark-roast bean contains slightly more per unit of weight and slightly less per unit of volume. Botanical variety weighs more than roast: robusta contains roughly twice as much as arabica.

Diterpenes are the point at which coffee and caffeine part company completely. Cafestol and kahweol are responsible for the only effect of coffee on lipids demonstrated in a randomized trial. Their concentration depends entirely on the brewing method: a home paper filter leaves around 12 milligrams of cafestol per liter, while boiled coffee reaches 939. Paper retains them; the French press, the moka pot, Turkish coffee and Scandinavian boiled coffee do not. Decaf changes nothing on this point, because diterpenes are not caffeine.

L-theanine deserves a clarification, because the supposed synergy with caffeine circulates as a fact. The trial that used the real proportions of a cup of tea observed that the combination cancelled out the behavioral effects of caffeine instead of enhancing them. The works with positive results use amounts of L-theanine far above what is attainable by drinking tea. The 2025 meta-analysis that brings together fifty trials declares that the doctoral grant of its first author was then funded by Lipton Teas and Infusions and that two of its co-authors were employees of Lipton Teas and Infusions and of Unilever. Its own authors point out that the confidence intervals leave the direction and the magnitude of the differences found in doubt.

In chocolate, theobromine exceeds caffeine by ten to one: about 800 milligrams per 100 grams of dark chocolate against about 80. Its profile is different, with a peak between two and three hours and a half-life of six to ten hours, which explains a milder and longer-lasting effect.

Two gaps are declared in this block. There is no intervention study comparing ground coffee with instant coffee on performance or health outcomes, despite it being the most frequent comparison in real life. And no figure for caffeine per serving of matcha anchored in a peer-reviewed source has been located.

What the label says and what the product contains do not always match. In an analysis of 53 supplements against reference materials, 89% of those that declared an amount were within a 16% margin. In an examination of fifteen pre-workouts sold in Australia, by contrast, only six declared caffeine on the label, and the real content ran from 59% to 176% of what was stated.

No coffee certification tells you anything about its caffeine. The Specialty Coffee Association (SCA) score rates sensory quality, fair trade rates socioeconomic conditions and the organic seal rates the growing method.

In supplements, the two most widespread certifications in the United States, Informed Sport and NSF Certified for Sport, deal with substances banned in competition and with the label matching the contents. That is their scope: they do not evaluate the efficacy of the product.

The agreement between what is declared on the label and the real content depends on the market and on the type of product. An analysis of 53 supplements against certified reference materials found that 89% of those that declared an amount stayed within a 16% margin. An examination of fifteen pre-workouts sold in Australia found that only six of them declared caffeine on the label and that the real content ranged between 59% and 176% of what was stated, with appreciable variation within a single container. A Polish study validating an analytical method, with a small sample, found contents between 4% and 35% above what was declared.

In terms of dose, the documented pre-workout servings run from 91 to 387 milligrams and weight-control supplements from 47 to 250. The practical consequence is that the dose of one product can be four times that of another without any label making it clear.

European regulation requires drinks with more than 150 milligrams per liter to declare the amount and warn of their high content, with the exception of drinks based on coffee or tea, which fall outside this obligation.

The evidence base is older than it looks: of the twenty-odd main trials, only two are later than 2020 and most are earlier than 2015.

The most serious methodological problem is blinding: caffeine is noticeable and participants can guess what they have taken. In sport it is worse, because the improvement being sought is 2 to 4%, of the same order as the placebo effect. In a trial designed for the purpose, the performance of trained cyclists also varied according to what they believed they had taken.

The second problem is who has been studied. In an audit of 1,826 studies on supplements with evidence, only 23% of participants were women. A recent meta-analysis finds improved performance in women with no clear differences between phases of the menstrual cycle, but none of its studies verified the phase with hormones.

Gaps remain uncovered: ultra-endurance, weight loss, bone density, reflux, interference with training adaptations and the comparison between ground and instant coffee, on which no intervention study exists.

The age of the documentary base is evident: of the twenty-odd main trials, two are later than 2020 and most earlier than 2015. Caffeine is a compound that has been studied for decades, and much of its canonical trial base was designed before preregistration and the systematic declaration of conflicts of interest became widespread.

Blinding is the structural defect of the field. Caffeine produces identifiable subjective effects, so participants can infer their assignment. In sport the problem is made worse because the improvement being sought, 2 to 4%, matches in magnitude the documented placebo effect. A trial designed specifically to measure it administered caffeine and placebo while manipulating the information given to forty-two trained cyclists, and performance also varied according to what the participant believed they had taken. No meta-analysis in the field corrects this bias, because there is no accepted procedure for correcting it.

The main gaps the evidence leaves open are ultra-endurance, where no solid randomized trial has been located despite it being the setting where most is consumed; sustained weight loss; bone density; gastroesophageal reflux; interference with chronic training adaptations; and shift work studied in real workers, not in laboratory simulations.

There are situations in which the prudent course is to consult a healthcare professional.

For minors, the American Academy of Child and Adolescent Psychiatry places the maximum at 100 mg a day between the ages of 12 and 18, and sets no safe dose below that age. The societies that have taken a position agree that energy drinks are not appropriate for minors.

Pure caffeine powder is a case apart: a teaspoon contains about 3,200 mg, close to 28 cups of coffee. The doses described as lethal lie above 5 to 10 grams, with considerable individual variability.

Caffeine withdrawal is a formal diagnosis. It appears between twelve and twenty-four hours without the habitual dose, peaks between twenty and fifty-one hours and lasts from two to nine days. Half of those who go through it suffer headache, and 13% show significant functional impairment. An intake of 100 milligrams a day is enough for it to appear.

Among the interactions with medicines, the largest is with fluvoxamine: in a crossover trial, the half-life of caffeine went from five to thirty-one hours. Quinolones block the same enzyme. Caffeine can raise concentrations of clozapine and olanzapine, and it shares a pathway with theophylline, which has a narrow therapeutic margin. In stress tests with adenosine or dipyridamole, caffeine antagonizes them, and protocols call for stopping it between twelve and twenty-four hours beforehand.

The international classification of diseases records two clinical entities related to caffeine. Withdrawal appears with its own code and defined criteria: it sets in between twelve and twenty-four hours after the last dose, peaks between twenty and fifty-one hours, lasts from two to nine days and causes headache in approximately half of cases, with significant functional impairment in 13%. The intake threshold capable of producing it lies around 100 milligrams daily, well below average consumption. Caffeine intoxication requires recent consumption, usually above 250 milligrams, and five or more symptoms from the following: restlessness, nervousness, insomnia, facial flushing, increased diuresis, gastrointestinal disturbance, muscle twitching, tachycardia or arrhythmia, and psychomotor agitation.

Almost all the relevant pharmacological interactions derive from the CYP1A2 pathway. Fluvoxamine is the most potent inhibitor documented: in a crossover trial, caffeine clearance fell from 107 to 21 milliliters per minute and the half-life went from five to thirty-one hours. Quinolones, in particular ciprofloxacin and enoxacin, inhibit the same enzyme. In the opposite direction, caffeine competes for the pathway with clozapine and olanzapine and can raise their concentrations; it also shares metabolism with theophylline, whose therapeutic margin is narrow. In pharmacological stress tests with adenosine or dipyridamole, caffeine directly antagonizes the agent used, so protocols call for stopping it between twelve and twenty-four hours beforehand to avoid a false negative.

The institutional references for specific populations are three and they do not agree on the unit of measurement. The European Food Safety Authority sets 200 milligrams a day during pregnancy and breastfeeding, and 3 milligrams per kilogram in children and adolescents. The American Academy of Child and Adolescent Psychiatry places the maximum at 100 milligrams daily between the ages of 12 and 18 and establishes no safe dose below that age. The Australian Institute of Sport places the limit at less than 2.5 milligrams per kilogram a day in those under 18. The clinical report of the American Academy of Pediatrics that is frequently cited in this context was retired in July 2021 and should not be treated as current.

Pure caffeine powder constitutes a risk category of its own. A level teaspoon contains around 3,200 milligrams. Deaths have been described with blood concentrations above 70 micrograms per milliliter, and the lethal dose lies above 5 to 10 grams, with considerable individual variability. The US Food and Drug Administration considers adulterated those products that are sold in bulk to consumers, and the Australian Institute of Sport records that the Australian regulator acted in 2019 to prevent its sale.