Substance and claimed effects

Caffeine is the world's most-used psychoactive substance, an adenosine-receptor antagonist absorbed almost completely from the gut and metabolised in the liver, primarily by the CYP1A2 enzyme Thorn 2012. The substance of this entry is not caffeine in the abstract but the half-life dimension of caffeine pharmacokinetics: the time it takes a single dose to fall to half its peak plasma concentration, and the considerable inter-individual variation in that number. Population mean is roughly 5 hours in healthy non-smoking adults, but credible clinical sources document a real-world range of about 2 to 10+ hours depending on CYP1A2 genotype, smoking, oral contraceptive use, pregnancy, and concurrent medications Carrillo and Benitez 2000. The claimed effects this entry covers are the downstream consequences of that variation: timing of evening cut-off for sleep onset and sleep architecture Drake 2013 Landolt 1995, anxiety expression in genetically sensitive consumers Alsene 2003 Childs 2008, and downstream daytime energy, focus and mood mediated through restored sleep Clark and Landolt 2017.

Evidence by addressing question

mechanism

Caffeine is absorbed within 30–45 minutes of ingestion and reaches peak plasma concentration around the same time EFSA 2015. About 95% of clearance proceeds through hepatic CYP1A2, which N-3-demethylates caffeine to paraxanthine; minor pathways via CYP2A6, NAT2 and xanthine oxidase handle the remainder Thorn 2012. The CYP1A2 gene at 15q24 carries a common intron-1 polymorphism (rs762551, often labelled CYP1A2*1F) where the variant A allele is associated with higher inducibility. Homozygous AA "rapid metabolisers" clear caffeine roughly twice as fast as homozygous CC "slow metabolisers" after smoking or other CYP1A2 induction; in non-smokers without induction the genotype-only effect on baseline clearance is smaller but measurable Sachse 1999 Cornelis 2011. The pharmacodynamic side runs through adenosine antagonism: caffeine competitively blocks A₁ and A_2A receptors, suppressing the buildup of sleep pressure that adenosine normally signals over the waking day Bjorness and Greene 2009 Reichert 2022. The longer caffeine stays above its effective receptor-occupancy threshold (commonly cited as roughly half-saturating around 100 mg plasma exposure), the longer adenosine signalling stays attenuated. That means a slow metaboliser drinking a 2 pm latte still has clinically meaningful receptor occupancy at 10 pm; a rapid metaboliser has cleared most of it. The variation in half-life is therefore not a curiosity — it directly maps to how long after a dose adenosine remains blocked at the cortex.

evidence

The pharmacokinetic-variation literature is older and reasonably mature. Knutti 1981 documented the dramatic pregnancy effect — caffeine half-life rising from ~3 hours pre-pregnancy to ~10.5 hours by the third trimester in the same individuals, a finding since replicated across cohorts Knutti 1981. Abernethy and Todd 1985 showed oral contraceptives roughly double caffeine half-life via CYP1A2 inhibition by ethinyl oestradiol Abernethy and Todd 1985; this is reinforced by general CYP1A2 inhibition data on ethinyl oestradiol/gestodene preparations Granfors 2005. Smoking is the largest induction effect — about 50% reduction in half-life in heavy smokers, normalising over 4–6 days after cessation Carrillo and Benitez 2000. Fluvoxamine, ciprofloxacin and fluconazole inhibit CYP1A2 strongly; fluvoxamine 100 mg/day raises caffeine AUC roughly 5-fold and extends half-life from ~5 to ~30+ hours Jeppesen 1996 Carrillo and Benitez 2000. On the sleep-architecture side, Drake 2013 is the canonical RCT: 400 mg caffeine ingested at bedtime, 3 hours before bed, and 6 hours before bed all produced significant disruption of total sleep time measured by polysomnography, with the 6-hour-before condition still costing roughly an hour of objective sleep Drake 2013. Landolt 1995 showed that a single 200 mg morning dose taken roughly 13 hours before bedtime still measurably altered overnight EEG spectral power — fewer slow waves, reduced sleep efficiency — establishing that effects on sleep architecture persist well past plasma clearance because the receptor occupancy and homeostatic system take longer to recover Landolt 1995. Clark and Landolt 2017 reviewed the broader RCT and epidemiological literature: consistent dose-dependent reductions in total sleep time and slow-wave sleep, increased sleep latency, and the consistent finding that habitual users underestimate their own sleep impairment compared to objective polysomnography Clark and Landolt 2017. The anxiety side is anchored by Alsene 2003 and Childs 2008: variants in ADORA2A (the A_2A receptor gene), particularly rs5751876, are associated with higher caffeine-induced anxiety at typical doses (150–450 mg), with carriers more likely to report panicky responses to caffeine independent of clearance speed Alsene 2003 Childs 2008.

protocol

The conservative population default — 8 to 10 hours of caffeine-free time before bedtime — derives from the Drake 2013 finding that even 6 hours of clearance is insufficient to prevent measurable sleep disruption for the average drinker Drake 2013. Multiplying observed half-life by 5 gives ≥97% clearance: for a typical 5-hour half-life that is ~25 hours of full washout, but useful sleep onset is achievable once plasma is below the threshold that produces receptor occupancy >25%, which is roughly 3 half-lives or 15 hours after a typical 200 mg dose. The practical translation: an average metaboliser drinking 200 mg of caffeine at noon still has receptor-relevant levels at 11 pm; cutting off by 10 am puts most of the clearance before bedtime. Slow metabolisers (oral contraceptive users, non-smoking CC homozygotes, pregnant women, those taking fluvoxamine/ciprofloxacin) should target morning-only or eliminate entirely. Fast metabolisers (heavy smokers, AA homozygotes) can tolerate later cut-offs without measurable polysomnography effects, but the felt-experience boundary still tends to be around 6–8 hours pre-bed Clark and Landolt 2017. EFSA's safety opinion settled on 400 mg/day for healthy adults and 200 mg single doses with no near-bedtime intake, which is the regulatory expression of the same logic EFSA 2015.

contraindications

Pregnancy is the strongest individual modifier of half-life: by the third trimester caffeine clearance is roughly one-third of pre-pregnancy values, and EFSA recommends no more than 200 mg/day during pregnancy on safety grounds independent of sleep Knutti 1981 EFSA 2015. Anxiety disorders, panic disorder and ADORA2A genotype carriers tolerate caffeine poorly; reduction or elimination is a standard recommendation in those populations Childs 2008. Concurrent fluvoxamine (SSRI used for OCD/depression) and ciprofloxacin (fluoroquinolone antibiotic) require either caffeine cessation or major dose reduction during the course of treatment, because plasma AUC can rise 5–10-fold Jeppesen 1996 Carrillo and Benitez 2000. Untreated atrial fibrillation, uncontrolled hypertension, and known severe arrhythmia warrant clinician input — caffeine acutely raises plasma catecholamines and renin activity, although the long-term cardiovascular effect in habituated users is modest Robertson 1978.

misconceptions

Three persistent ones. First: "I can drink coffee right up to bedtime and still sleep fine." Subjective sleep onset is a weak detector of caffeine effects; polysomnography in habitual users shows reduced slow-wave sleep, increased sleep fragmentation and decreased sleep efficiency that the drinker does not consciously perceive Drake 2013 Clark and Landolt 2017. Second: "Half-life is 5 hours so I'm clear after 5 hours." Half-life is the time to halve the dose, not eliminate it; three to five half-lives are needed for clinical clearance, and the receptor effect on sleep architecture outlasts even that Landolt 1995. Third: "I'm tolerant, so caffeine doesn't disrupt my sleep any more." Tolerance affects the subjective alerting and anxiogenic effects more than the sleep-architecture effect; chronic users show some adaptation in adenosine receptor density but the basic sleep-disruption signal persists Weibel 2020.

audience

Several populations live at the slow-clearance end of the distribution and need a more conservative cut-off than the population default. Women using combined oral contraceptives have roughly doubled half-life (~10 hours) Abernethy and Todd 1985. Pregnancy clearance approaches one-third of baseline by the third trimester Knutti 1981. Non-smoking CYP1A2 CC homozygotes (estimated 10–15% of European-ancestry populations) clear caffeine measurably slower than AA homozygotes Sachse 1999. ADORA2A risk-genotype carriers (~30%+ of the population) are at elevated risk of anxiety responses regardless of clearance speed Childs 2008. Older adults clear caffeine slightly slower on average; adolescents, who are still circadian-shifted later than adults, are particularly vulnerable to bedtime intrusion. By contrast, heavy smokers clear caffeine ~50% faster — but the moment they quit, half-life renormalises over a few days and they typically find their habitual caffeine dose now keeps them awake Carrillo and Benitez 2000.

practicalities

Self-identifying as a slow metaboliser without genotyping uses three signals: (1) caffeine subjectively "lasts longer" than peers describe; (2) afternoon coffee predictably costs sleep that night; (3) anxiety or palpitations from a moderate dose. Genotyping (23andMe historically reported rs762551; consumer pharmacogenomic panels still do) is cheap, but the behavioural test — moving the cut-off back by 2 hours for two weeks and tracking sleep — is more informative because it captures the joint effect of genotype, contraceptive use, smoking status and current medications. Caffeine doses come in widely variable forms: 8 oz drip coffee is roughly 95 mg, espresso shot 60–80 mg, black tea 40–70 mg, green tea 25–50 mg, energy drinks 80–300 mg, pre-workout supplements often 200–400 mg per serving. Decaf is not zero — it usually retains 2–15 mg per cup. The 400 mg/day EFSA ceiling translates to roughly 4 standard coffees per day, with single doses kept under 200 mg and none in the post-noon window for average metabolisers EFSA 2015.

failure-modes

Common screwups. Trying to cut off at 6 pm and concluding "caffeine doesn't affect me" because subjective sleep latency feels normal — but objective sleep architecture is still degraded for several more hours Drake 2013. Quitting smoking without adjusting caffeine, then attributing the resulting insomnia and anxiety to nicotine withdrawal when much of it is unmasked caffeine effect at the new (slower) clearance rate Carrillo and Benitez 2000. Starting an oral contraceptive or fluvoxamine without realising baseline caffeine dose has become functionally doubled. Treating decaf as zero-caffeine when it isn't, particularly for very slow metabolisers stacking 4–5 decafs through the evening. Trusting tolerance: habitual users still show measurable sleep disruption — the felt experience habituates faster than the physiology Clark and Landolt 2017 Weibel 2020.

stakes

Felt-experience consequence of ignoring half-life over months and years. In the population-default slow metaboliser drinking 2–3 coffees per day with the last around 3 pm: objective sleep efficiency drops several percentage points, slow-wave sleep is reduced by 15–30% on caffeine-exposed nights, total sleep time falls by 30–60 minutes Clark and Landolt 2017. This is the substrate for the chronic daytime fatigue that drives the next morning's coffee — a self-perpetuating loop in which caffeine compensates for the sleep deficit caffeine caused, with the drinker unaware of the causal direction. In ADORA2A-sensitive users, the anxiety load is constant and attributed to "stress" or "anxiety disorder" when its proximal cause is the daily dose Childs 2008. Over years, the chronic sleep restriction tracks with the longevity-relevant outcomes that follow chronic sleep restriction generally, but the dose-response is shallow at the individual level. Brain imaging shows reversible reductions in grey matter volume in the medial temporal lobe over 10 days of daily caffeine intake, normalising during washout Lin 2021.

payoff

Adoption of half-life-aware timing — typically cutting off 8–10 hours before bed, earlier for slow metabolisers — produces measurable polysomnography changes within the first sober night: shorter sleep latency, more slow-wave sleep, less fragmentation Landolt 1995. The subjective shift takes 1–2 weeks because habitual users have rebuilt their baseline expectations around the impaired sleep state. Reported timecourse from clinical trials and behavioural sleep interventions: within 3–5 days, easier sleep onset and fewer 3 am wakings; within 2 weeks, an unmasked baseline daytime alertness that does not require afternoon caffeine to maintain; within months, a stable lower daily caffeine total (because the morning dose is now sufficient). The anxiety reduction for ADORA2A carriers is faster — within hours of removing the dose. The catch: an established 4-coffee-per-day habit going to 1-coffee-at-7am produces 1–3 days of withdrawal — headache, low mood, sluggishness — and a clean two-week trial is what reveals the steady-state benefit.

out-of-scope

Adjacent topics not covered here: chronic caffeine withdrawal as a clinical syndrome; caffeine's metabolic effects (glucose tolerance, lipolysis); ergogenic dosing for endurance and strength sport; coffee as a beverage (polyphenols, chlorogenic acids); long-term cardiovascular epidemiology of habitual coffee consumption; caffeine and pregnancy beyond clearance kinetics; caffeine in children/adolescents.

The credibility range

Optimist case

The pharmacokinetic story is solid: CYP1A2 variation is real, large, and clinically actionable; the modifiers (smoking, OCP, pregnancy, fluvoxamine/ciprofloxacin) are documented across decades of pharmacology research Carrillo and Benitez 2000. The sleep-architecture effect is consistently demonstrated at polysomnography, with the canonical Drake 2013 trial showing 400 mg caffeine 6 hours pre-bed costs roughly an hour of objective sleep Drake 2013. Multiple replications from the same labs (Landolt and colleagues) confirm that morning caffeine alters evening EEG spectral power Landolt 1995. The recommendation — cut off 8–10 hours pre-bed, earlier if you have a known slow-clearance modifier — is mechanistically grounded, empirically supported, and zero-cost to test individually. The ADORA2A-anxiety link replicates across independent samples Childs 2008. The benefit at the population level is non-trivial: sleep restriction is a major mediator of fatigue, mood and metabolic complaints in the working population, and a meaningful fraction of that is iatrogenic — caused by mis-timed caffeine.

Skeptic case

Effect sizes are modest in absolute terms. Drake 2013's 6-hour cut-off cost ~1 hour of total sleep at 400 mg — but at 200 mg or 100 mg the effect is smaller, and habitual heavy users may have partially compensated. The CYP1A2*1F (rs762551) polymorphism's effect size on baseline (non-induced) caffeine clearance is real but small — most variance comes from smoking, OCP and pregnancy, all of which are observable without genotyping Sachse 1999. The polygenic GWAS hits for caffeine consumption (AHR, CYP1A2 region) capture habit not individual half-life precisely Cornelis 2011. Many users report perfectly adequate sleep with afternoon caffeine; while objective architecture may differ from their subjective report, the long-term clinical consequence of a 10% reduction in slow-wave sleep on a person who is otherwise healthy is plausibly small. Tolerance does shift the sleep-disruption magnitude in chronic users Weibel 2020. The anxiety effect via ADORA2A is dose-dependent and shows up most clearly at supra-habitual doses; the typical 1–2 cup/day user without anxiety symptoms does not need genotyping. The 8–10 hour rule is therefore a defensible default rather than a clean dose-response prescription, and population variation means many users will find shorter or longer cut-offs match their own observation.

Author's call

The half-life story is settled science with directly actionable downstream recommendations. The strongest two interventions a typical reader can act on are: (1) cut caffeine off at least 8 hours before bedtime, and earlier (morning-only) if they use OCPs, are pregnant, recently quit smoking, or take a CYP1A2 inhibitor; (2) keep daily total under 400 mg and single doses under 200 mg. The behaviour test — moving the cut-off back two hours for two weeks and tracking sleep — is more informative than genotyping for most readers. Evidence rating: 4 — multiple consistent RCTs anchor the sleep effect; pharmacokinetic variation has decades of clinical pharmacology behind it; ADORA2A-anxiety link is replicated; the only reason it isn't 5 is that the field has not produced large pragmatic trials of "move your cut-off" as an intervention. Controversy: 1 — the disagreement is mostly at the margin of whether everyone needs an aggressive cut-off, not whether caffeine timing matters.

Stakeholder and incentive map

• Coffee industry and energy-drink manufacturers have a clear commercial incentive to downplay timing and dose limits and emphasise tolerance. National Coffee Association communications consistently emphasise the moderate-consumer safety story EFSA 2015 and rarely lead with sleep architecture.
• Sleep clinicians and AASM-aligned researchers (Landolt, Drake, Roth, Reichert) have driven the polysomnography literature and consistently recommend conservative cut-offs Clark and Landolt 2017. Sleep medicine as a specialty is structurally aligned with surfacing sleep harms.
• Pharmacogenomics consumer companies (23andMe historically, current PGx panels) have a commercial interest in foregrounding the CYP1A2 genotype story, sometimes overstating actionability.
• Performance / wellness communities (sport nutrition, biohacking podcasts) have surfaced the half-life framing aggressively over the past decade — credit where due, the half-life and cut-off advice that reached general audiences came from this community before mainstream guidelines caught up. Risk: the community-amplified version sometimes treats the population default as universal law without acknowledging the variation it is trying to manage.
• Regulators (EFSA, FDA) sit at 400 mg/day for healthy adults with caveats; they have not codified a bedtime cut-off in regulatory language though sleep researchers consistently recommend one.

Population variability

• CYP1A2 rs762551 genotype. AA fast metabolisers vs CC slow metabolisers — population frequencies vary by ancestry; in European-ancestry samples roughly 45% AA, 45% AC, 10% CC. Effect on caffeine clearance is largest under induction (smoking) and smaller in non-smokers Sachse 1999 Cornelis 2011.
• Smoking status. The single largest non-genetic modifier — heavy smokers clear caffeine ~50% faster; effect reverses within 4–6 days of cessation Carrillo and Benitez 2000.
• Oral contraceptives. Roughly double half-life via ethinyl oestradiol CYP1A2 inhibition Abernethy and Todd 1985 Granfors 2005.
• Pregnancy. Third-trimester clearance is one-third of pre-pregnancy Knutti 1981.
• Age. Mild slowing in older adults. Adolescents have similar clearance to adults but more bedtime vulnerability from later circadian phase.
• Concurrent medications. Fluvoxamine (5–10x AUC increase), ciprofloxacin (~2x), fluconazole, mexiletine — all CYP1A2 inhibitors; carbamazepine, rifampicin, phenytoin — inducers Carrillo and Benitez 2000 Jeppesen 1996.
• ADORA2A rs5751876. Risk allele carriers (~30%+ of population) show stronger anxiety response at typical doses Alsene 2003 Childs 2008.
• Tolerance. Chronic users develop partial tolerance to alerting and anxiety effects; sleep-architecture effects are more resistant to tolerance Weibel 2020.

Knowledge gaps

Several. The dose-response for cut-off timing has been pinned at 400 mg in Drake 2013 but no comparable trial spans the full grid of dose (100–400 mg) × clearance speed (fast/slow metaboliser) × cut-off (0–10 h pre-bed); the population default of 8–10 hours is a reasonable interpolation rather than a directly trialed point. Long-term outcomes of moving caffeine to morning-only have not been studied in a pragmatic RCT — most evidence is mechanistic and short-term. The ADORA2A genotype work has been replicated in modest-sized samples but lacks a large-scale population biobank confirmation. The interaction between CYP1A2 induction (smoking, cruciferous vegetables, grilled meat) and adenosine pharmacodynamics is incompletely mapped. Reversibility of grey matter changes with caffeine cessation has been shown short-term Lin 2021 but long-term consequences are unclear. Finally, the question of whether "morning-only" cleanly translates to the chronotype-shifted young adult population (whose biological morning is, say, 11 am) is undocumented.

Scope and narrowing. The brief named four consequences — sleep onset, sleep architecture, anxiety, timing of consumption — and the entry covers all four end-to-end. Sleep onset and architecture are the headline (anchored to Drake 2013 and Landolt 1995). Anxiety lands in the audience and contraindications sections via the ADORA2A variant work. Timing relative to bedtime is the protocol. Holistic scoring then adds the downstream dimensions — energy, focus, mood (sleep-mediated), modest health-short-term and longevity, marginal beauty-cumulative — that follow from the substance even though the brief didn't name them. No silent dropping.

Rating difficulties. Sleep was scored 4 rather than 5 because the substance here is the timing intervention, not caffeine cessation outright — strong but not dominant. Energy at 3 is the dominant felt-experience win for typical users, separable from the morning-coffee energy that the intervention preserves. Effort burden at 2 (not 1) reflects the persistent daily decision plus the 1–3 day withdrawal window for established habits; calling it 1 would understate the typical user's experience. Evidence at 4 (not 5) because no large pragmatic RCT of "move your cut-off" has run end-to-end — the evidence is anchored on canonical sleep-lab work that proves the mechanism rather than on a pragmatic effectiveness trial.

Hard decision on action verb. Considered know (since half-life is fundamentally a piece of knowledge) but landed on do because the actionable surface — set a personal cut-off and hold it — is the load-bearing reader takeaway. The article reflects that with the two-week behavioural test as the central protocol.

Excluded from this entry.

• Caffeine cessation as a strategy. Different substance, different decision; this entry is about timing the dose, not removing it.
• Caffeine in sport performance. Ergogenic dosing and the genetics of athletic response to caffeine are a separate frame. Future entry candidate.
• Coffee as a beverage. Polyphenols, chlorogenic acids, the long-term cardiovascular epidemiology — those ride on the coffee matrix, not the caffeine molecule. Separate entry.
• Caffeine in pregnancy beyond the timing rule. The 200 mg/day pregnancy ceiling is mentioned in contraindications; the broader fetal-exposure literature warrants its own entry.
• Adolescent caffeine. Circadian-shifted bedtimes plus developing brains is a separate population worth its own treatment.
• Chronic caffeine withdrawal as a clinical syndrome. Mentioned briefly under out-of-scope; would benefit from its own entry covering DSM-5 criteria, treatment, and timecourse.

Future-link candidates. Sleep debt; circadian alignment / morning light; slow-wave sleep biology; ADORA2A and individual stimulant response (if such an entry is ever spun up); coffee-the-beverage; pre-workout supplement safety.

Notes on evidence used. The Reichert 2022 review and Lin 2021 grey-matter finding sit in the dossier as supporting context but didn't earn a place in the article — the article keeps the focus narrowly on timing rather than on adenosine receptor plasticity and brain-imaging endpoints. The Yang 2010 and Smith 2002 reviews and Granfors 2005 on OCP CYP1A2 inhibition similarly back the dossier rather than the article; Cornelis 2006 and Cornelis 2011 are kept in the dossier as the population-genetics anchor.

Status set to draft pending review.