Is six hours enough? Caffeine half-life and when to have your last coffee

The study people cite for one of the most repeated numbers in sleep advice — stop caffeine six hours before bed — gave twelve people in Detroit about four coffees’ worth of caffeine in pill form. And the ones who took it six hours before bed still lost more than an hour of sleep.

I’ll admit I’d been treating six hours as a safe line for years without once looking at what was underneath it. Turns out it isn’t a line at all. It’s the longest gap that study tested, and the problem hadn’t gone away by then.

Having now gone through the trials that came after it, my honest read is that six hours is the wrong shape for the rule. The evidence doesn’t point at a time of day. It points at a dose and at how fast your liver clears it — two numbers you can multiply out on a napkin — and at a slightly uncomfortable finding about how little you notice. So: the claim, where it gets its evidence, what the bigger studies say, and then the napkin.

One scope note before I start. This isn’t a post about whether coffee is good for you. The coffee and longevity post covers that, and the short version is that moderate drinking looks fine to good. I drink plenty. This is only about the clock.

The rule, as everyone states it

Stop caffeine six hours before bed. It turns up in sleep-hygiene advice everywhere, usually with caffeine’s half-life quoted next to it, as if the two numbers prove each other.

It deserves a fair hearing, because it has real things going for it. It’s simple. It needs no arithmetic and no idea of what’s in your cup. It points in the right direction — caffeine late in the day is bad for sleep, and almost nobody disputes that. And unlike most wellness rules, it has a randomised, double-blind, placebo-controlled trial attached to it.

The problem isn’t that it’s wrong, exactly. It’s that it answers a question about one dose and gets applied to whatever is in your mug.

Where the six hours gets its evidence

Sleep-hygiene advice was already saying something like this when Christopher Drake and colleagues at Henry Ford Hospital’s sleep centre in Detroit set out to test it. Their paper, in the Journal of Clinical Sleep Medicine in 2013, opens by pointing out that sleep-hygiene recommendations were being handed out widely with few systematic studies behind them, at least in people’s own homes. Which is a refreshingly honest way to start a paper.

The design was neat. Twelve healthy normal sleepers — six women and six men, aged 19 to 48, all regular but not heavy caffeine users — each did four experimental nights at home. On each of those nights they took pills six hours, three hours and zero hours before their usual bedtime. On three of the nights, one of those pills held 400 mg of caffeine; on the fourth, every pill was a placebo. Neither the participants nor the researchers knew which night was which. Sleep was scored by a wireless single-channel EEG headband and a morning sleep diary.

400 mg is a lot to take at once. It’s roughly four standard coffees, and it’s twice the 200 mg that the European Food Safety Authority (EFSA) considers unproblematic as a single dose for healthy adults. EFSA’s better-known 400 mg figure is a daily total “consumed throughout the day”, not one hit.

All three timings disrupted sleep compared with placebo. The line that gets quoted is the authors’ own: “Even at 6 hours, caffeine reduced sleep by more than 1 hour.” The line that rarely gets quoted is that the three timings didn’t differ significantly from one another. The study couldn’t show that taking the caffeine six hours out was any gentler than taking it at lights-out. With twelve people, it couldn’t show much of anything about the differences between timings — which is rather the point.

Then there are the diaries. At six hours, only the headband picked up a significant difference. The diaries leaned the same way — about 41 minutes less sleep reported — but didn’t reach significance. By the headband’s count people lost more than an hour, and their own reports didn’t register a significant change.

The caveats are the usual ones for a small study, and the authors list most of them themselves: twelve people, one dose, no blood caffeine measurements, young-to-middle-aged moderate drinkers only, and a headband so recently validated that they call the sleep-stage results preliminary. It was also funded by a grant from Zeo, a company that made home sleep headbands.

But the bit that matters most is in the conclusion. The authors say their results support advice to refrain from substantial caffeine “for a minimum of 6 hours prior to bedtime”. A minimum. Read that way — six hours as a floor for a big dose — the study holds up fine. Read the way it usually gets repeated — six hours as the point where you’re in the clear — it says nearly the opposite, because the six-hour group was still losing more than an hour.

What the bigger studies say

Two papers from one Australian group do most of the work here, and they’re worth reading side by side, because they don’t quite agree.

The first is a meta-analysis: Carissa Gardiner and colleagues, Sleep Medicine Reviews, 2023, pooling 24 controlled studies of caffeine and night-time sleep in healthy adults. On average, caffeine cost 45 minutes of total sleep, cut sleep efficiency by 7%, added 9 minutes to falling asleep and 12 minutes of lying awake in the night, and trimmed about 11 minutes of deep sleep. The authors treat 20 minutes of total sleep as the threshold for a clinically meaningful change, so 45 is not a rounding error — it’s a real chunk of the sleep you actually need.

The useful part is that both the dose and the timing turned out to matter, which let the authors model a cut-off: the latest time a given dose can be taken without a statistically significant loss of total sleep. For a 107 mg coffee, 8.8 hours before bed. For a 217.5 mg serving of pre-workout, 13.2 hours. For a 47 mg cup of black tea, no cut-off at all. On a 10 PM bedtime, that’s coffee before 1:12 PM and pre-workout before 8:50 AM.

Bit nerdy, but this matters: the cut-off is where the model’s confidence band crosses zero, the point beyond which it can no longer tell the effect apart from no effect. That isn’t the same as the point where the effect stops. The authors say as much themselves — the cut-offs are “based on statistical significance”, on the assumption that a statistically significant loss is a clinically significant one — and they suggest treating them as “a starting point” to adjust against your own sleep.

The second paper is the trial the first one was missing. Gardiner and colleagues again, in Sleep in 2025: 23 healthy young men (average age 25, moderate habitual caffeine drinkers, non-smokers) each did seven conditions at home — placebo, or a 100 mg or 400 mg capsule taken 12, 8 or 4 hours before their usual bedtime — with sleep measured by a forehead EEG device and a diary, one night per condition, 48 hours apart.

100 mg made no significant difference to anything, objective or subjective, at any of the three times. 400 mg changed something significant at all three. Taken four hours before bed, it cost about 51 minutes of sleep and 30 minutes of deep sleep. At eight hours, people spent about 29 more minutes awake in the night and lost about 15 minutes of deep sleep. At twelve hours — mid-morning, for most people — they took about 15 minutes longer to reach their first stretch of steady sleep and lost about 21 minutes of deep sleep. Total sleep time dipped at eight and twelve hours too, but not significantly. Within the trial, the dose comparison was blunt: at four hours, 400 mg cost about 46 minutes more sleep than 100 mg did.

And the perception problem again. Perceived sleep quality only dropped when the 400 mg came four hours before bed. When they were asked each morning which dose they’d had and when, they got both right only 22% of the time.

Which leaves one awkward fact on the table. Same research group, two answers for a single cup: the meta-analysis says to keep a 107 mg coffee 8.8 hours clear of bed, and the trial found nothing significant from 100 mg at four. I don’t think either is simply wrong. The meta-analysis is a straight-line model fitted across studies that mostly used doses above a single cup, and its cut-off is a significance boundary rather than a measured loss — the trial’s authors point out that the absolute sleep loss behind the 8.8 hours couldn’t be determined from it. The trial is 23 young men, one night per condition, and a non-significant result in 23 people is a long way from proof that nothing happened.

So for one ordinary cup, the honest window runs from early afternoon to early evening, and nobody can currently tell you where in it you sit. For a big dose, both papers point the same way, and it’s earlier than most people assume.

If you’d rather have evidence from real life than from capsules, there’s a lovely 2023 trial in the New England Journal of Medicine. Gregory Marcus and colleagues texted 100 adults every day for two weeks, telling them at random either to drink caffeinated coffee that day or to avoid caffeine. By a wrist accelerometer, they slept about 36 minutes less on coffee days: 397 minutes a night against 432.

One more mechanism, because caffeine doesn’t only block the adenosine that builds up sleep pressure through the day. In a seven-week, placebo-controlled study, Burke and colleagues (Science Translational Medicine, 2015) found that a double espresso’s worth of caffeine three hours before bedtime pushed the body’s melatonin rhythm about 40 minutes later — nearly half the shift produced by three hours of bright evening light. A later body clock is its own problem if you care about keeping your sleep timing regular.

The napkin: a half-life sum

Right, so here’s the arithmetic the rule skips.

Caffeine from a drink typically peaks in your blood somewhere between half an hour and two hours after you have it. After that, your liver clears a roughly constant fraction of what’s left every hour, which is what a half-life describes. After one half-life, half the caffeine is still there. After two, a quarter. After three, an eighth. The formula fits on a napkin:

what's left = dose × 0.5^(hours ÷ half-life)

The half-life is the slippery bit. Five hours is the number you’ll see most, and it’s a reasonable middle, but the studies scatter around it: an average of 3.4 hours in men and non-pregnant women in one 1981 study (Knutti and colleagues, European Journal of Clinical Pharmacology), and 6.0 hours in non-smokers in another (Parsons and Neims, Clinical Pharmacology and Therapeutics, 1978). Gardiner’s review describes the typical range as three to six hours.

Most of the clearing is done by one liver enzyme, CYP1A2. A 2022 systematic analysis of 141 pharmacokinetic publications (Grzegorzewski and colleagues, Frontiers in Pharmacology) notes that it is estimated to account for about 95% of caffeine’s clearance and that its activity varies five- to six-fold between people, and its own pooled analysis found smoking and the contraceptive pill shifting clearance consistently across more than 50 years of studies. A few things move the half-life:

• Smoking speeds it up. 3.5 hours in smokers against 6.0 in non-smokers (Parsons and Neims, 1978).
• The contraceptive pill slows it down. 7.9 hours against 5.4 in women taking low-dose oestrogen pills (Abernethy and Todd, European Journal of Clinical Pharmacology, 1985).
• Pregnancy slows it most. An average of 8.3 hours, with individual values from 3 to 16 (Knutti, 1981); a follow-up from the same group found an average of 10.5 hours in the last four weeks.
• Genes play a part, though less tidily than you’d hope. Gardiner’s 2025 trial genotyped everyone for commonly studied CYP1A2 and adenosine-receptor variants and found no significant effect on sleep — while noting it wasn’t powered to find one.

Here’s what the sum gives for a single drink at different times, with an 11 PM bedtime. The three columns are illustrative half-lives from the range above: three hours for a quick clearer, five for the middle, eight for a slow one.

Multiply by your dose. EFSA’s rough figures are about 90 mg for a 200 ml filter coffee, 80 mg for an espresso or a 250 ml energy drink, and 50 mg for a cup of black tea — and real cups vary a lot.

A couple of things fall out of this that the six-hour rule hides.

A big dose takes more half-lives to come down. 200 mg drops below 100 mg after exactly one half-life: three, five or eight hours, depending on who you are. 400 mg needs two: six, ten or sixteen hours. A slow clearer who has 400 mg at 7 AM still has about 100 mg on board at 11 PM, on this arithmetic.

The bedtime number tracks the total, not only the clock. I usually have two or three filter coffees. On a three-cup day (8, 10 and half twelve, at EFSA’s 90 mg a cup) they leave about 47 mg at 11 PM on a five-hour half-life — more than a single 4 PM coffee would leave, at about 34 mg. That doesn’t make my mornings worse than a 4 PM coffee; it’s 270 mg against 90. It means a clock time on its own can’t tell you where you are.

The studies, run through the same sum

This next bit is my own arithmetic rather than anyone’s finding, so treat it with suspicion. If what’s left at bedtime is what matters, the studies above should line up when you run their doses through the same sum.

One more study first. In 1995 Hans-Peter Landolt and colleagues (Brain Research) gave nine healthy men 200 mg of caffeine at ten past seven in the morning and recorded their sleep from 11 PM. Saliva caffeine fell from 17 µmol/l an hour after the dose to 3 µmol/l just before sleep — under a fifth of the peak. Compared with placebo, total sleep time and sleep efficiency were still significantly reduced.

The middle column below is my calculation from each study’s test dose alone, at a five-hour half-life, with the three-to-eight-hour range in brackets. These are estimates, not measurements: Drake’s team took no caffeine measurements, and the other two studies measured saliva concentrations, which don’t convert neatly into milligrams in the body. In two of the studies, participants also had their usual caffeine earlier in the day, which these figures leave out.

Read top to bottom, the sum does a decent job on the big effects. The conditions that left the most caffeine on board produced the biggest losses, and the 2025 trial’s own dose comparisons agree: 400 mg against 100 mg at the same hour cost more deep sleep at four hours and again at twelve.

It does a poor job at the bottom. Landolt’s morning dose and the trial’s 100 mg at twelve hours both come out at about 20 mg on the sum; one produced a significant loss and the other produced nothing significant. Different decades, devices, people and doses — and “significant” against “not significant” is not itself a significant difference, so this is a hint, not a finding. It isn’t the only hint, though. In the 2025 trial, 400 mg taken twelve hours out left, on paper, roughly what 100 mg left at four (~75 against ~55 mg), and only the bigger dose showed significant effects. The authors never compared those two conditions directly, so the same caveat applies.

Landolt’s own explanation would fit both. Either even a small leftover amount acts on sleep, or — the version I find more interesting — caffeine circulating during the day blunts the build-up of sleep pressure, so a big early dose costs you something that a bedtime calculation can’t see. Gardiner’s team reached a similar practical place by a different route: don’t take 400 mg as one dose within 12 hours of bed. Whether spreading the same amount across the day is any gentler is, in their own word, unclear.

One more translation, because it bears on the 100 mg line you’ll meet in a minute. Run the meta-analysis’s two cut-offs through the same sum — a 107 mg coffee at 8.8 hours, a 217.5 mg pre-workout at 13.2 — and both land at roughly 30–35 mg still on board at bedtime, on a five-hour half-life. Translated that way, the cautious model’s line sits well below 100 mg; the newest trial saw nothing significant at around 55. There’s no clean threshold in this literature, and anyone who sells you one — us included — is rounding.

The honest verdict

Six hours is the wrong shape for a caffeine rule.

For a big dose it’s too generous: the study it leans on found more than an hour of lost sleep at exactly six hours, and 400 mg still dented deep sleep from twelve. For a single ordinary cup it may be stricter than it needs to be — the 2025 trial found nothing significant at four hours — though the cautious meta-analysis disagrees, and I wouldn’t bet much on 23 young men speaking for everyone. What the rule leaves out is the dose, and then your clearance rate. And what it quietly assumes — that you’d notice if your coffee were costing you sleep — is contradicted by both trials.

What I do instead: three questions

I’ve stopped asking what time it is and started asking three other things.

1. How big is this one dose? Single-dose size seems to matter on its own, beyond what’s left at bedtime. I use EFSA’s 200 mg single-dose level as my ceiling for one go. It’s a safety figure, not a sleep figure — Landolt’s row shows that 200 mg isn’t free even at breakfast — so it’s a cap on any one dose rather than a sleep-safe line, and smaller is better if your sleep is fragile. What it does do is keep me well away from the 400 mg doses that turned up in the results above.

2. What will be left at bedtime, at my clearance? Run the napkin sum with your likely half-life, and assume a slower one if you’re on the pill, pregnant, or simply know you’re sensitive. Sarvita draws a line at 100 mg (more on that below). I treat it as a ceiling, not a target, because nothing above gives any reason to think 99 is fine and 101 isn’t.

3. Am I using how I slept as evidence? If so, stop. In both trials, people’s own reports missed changes the EEG picked up once the caffeine was taken further from bed. And if you’re tempted to check your tracker’s deep-sleep figure instead: the losses in these studies were measured with EEG, and deep sleep is a number consumer wearables estimate poorly.

In practice that’s my two or three filter coffees before about one, and on the odd day an espresso at four, which leaves about 30 mg at 11 PM if I’m typical. And if you’ve been eyeing a sleep supplement, try moving the afternoon coffee first — it’s free — before working out whether magnesium or melatonin fits your particular problem.

What Sarvita 1.22 does with this

I’m biased, obviously — I work here. But the arithmetic above is tedious, and 1.22 does it for you.

Sarvita 1.22 adds water and caffeine tracking. There’s nothing separate to log: when you photograph or type a meal that includes a drink, the app estimates the caffeine and water in it. A plain glass of water is one tap in the app, or from the Home Screen widget, Control Center or Siri. The water half is the quieter sibling — hydration has its own post, and yes, your coffee counts towards it.

On the caffeine side, the app estimates what’s in your body now with a half-life model like the one above and draws the curve through the day to your usual bedtime, which it takes from your sleep data. Without sleep data, it assumes 11 PM. So you get a line like “≈33 mg left at 11:00 PM”. It shows when you’ll drop below 100 mg — EFSA’s figure for a single dose close to bedtime, which the app borrows as its bedtime line — and today’s total against EFSA’s 400 mg daily level for healthy adults. You can set how fast you clear caffeine — Fast, Typical or Slow, where Typical is a half-life of about five hours — and the setting reminds you that smoking shortens it while pregnancy and the contraceptive pill lengthen it. Caffeine and water both go to Apple Health (caffeine once you’ve allowed it).

It also says, in as many words, that this is an estimate, not a measurement: caffeine peaks about 45 minutes after a drink, then halves every half-life. Given everything above, that’s the right amount of confidence. It tells you roughly what’s still in you. What your sleep does with that is the part nobody — us included — can read off a curve. And if you’re pregnant: EFSA’s figure is 200 mg a day, the half-life stretches a long way, and that’s a conversation for your midwife or doctor rather than an app.

The short version

The six-hour rule has a real study behind it, and that study’s six-hour group still lost more than an hour of sleep on four coffees’ worth of caffeine. The newer evidence says the cut-off moves with the dose and with how fast you clear it: somewhere from early afternoon to early evening for a single ordinary cup, depending on whose method you trust, and not even mid-morning for 400 mg in one go. The half-life sum — the dose, halved once per half-life until bedtime — gets you most of the way, and it’s a better guide than how you think you slept. It’s still a guide, not a verdict.

Which is a slightly deflating answer for a rule that promised to be simple. But “smaller, earlier, and don’t trust the feeling” isn’t complicated either. Fair enough.