Caffeine: Part Two: What Daily Intake Does

Abstract

Almost everything known about caffeine's acute effects comes from studies in habitual consumers tested after abstaining. That design has an obvious weakness: the placebo condition may be a withdrawal condition.

A 2021 study in Scientific Reports built the experiment that separates the two, running habitual coffee drinkers through caffeine, withdrawal and placebo conditions for ten days each, with polysomnography at the end.

Two of its results are worth knowing: daily caffeine did not measurably disturb sleep in these people, and one EEG measure was altered identically by taking caffeine and by stopping it. A section added after publication sets both against a later review of thirty-two studies that points the other way.

1.The Design

Janine Weibel and colleagues studied twenty young male habitual caffeine consumers in a double-blind, randomised, crossover experiment with three ten-day conditions.

Caffeine: 150 mg three times daily throughout. Placebo: three placebos daily throughout. And withdrawal: 150 mg three times daily for eight days, then a switch to placebo.

After nine days of continuous treatment, sleep was recorded by EEG across a scheduled eight-hour night — beginning 8 hours after the last caffeine intake in the caffeine condition, and 15 hours after it in the withdrawal condition. Subjective sleep quality and withdrawal symptoms were assessed on waking.

The three-condition structure is what makes this useful. Comparing caffeine against placebo confounds the drug with its absence in someone adapted to it. Adding a withdrawal arm separates them.

2.The First Result Is a Null

The authors report it with visible surprise: unexpectedly, neither polysomnography-derived total sleep time, sleep latency, sleep architecture nor subjective sleep quality differed among the placebo, caffeine and withdrawal conditions.

Three 150 mg doses a day — 450 mg, more than most people consume — taken in the morning and afternoon, produced no measurable disturbance of night-time sleep structure in habitual consumers who sleep well.

This cuts against a widespread belief, and it is worth stating the boundaries carefully. The participants were young men, they were habitual consumers, they were healthy good sleepers, and the caffeine was taken in the morning and afternoon rather than the evening. The authors say so explicitly.

Acute evening caffeine does delay sleep onset and reduce sleep intensity; that is not in dispute. What this study addresses is the different question of whether those disturbances persist when caffeine is consumed continuously during the day, as most coffee drinkers consume it. In this group, they did not.

3.The Second Result Is Stranger

One measure did change. EEG power density in the sigma frequencies — 12 to 16 Hz, the band containing sleep spindles — was reduced during non-REM sleep in both the caffeine and the withdrawal conditions, compared with placebo.

Both. The reduction was present 8 hours after the last dose and still present 15 hours after it, when the drug had largely cleared and the participants were, pharmacologically speaking, in withdrawal.

So this is not an effect of caffeine being present. It is an effect of having been taking caffeine — something that outlasts the compound itself.

Sleep spindles matter here because of a connection running through our other work: they are among the better-characterised physiological correlates of memory consolidation, and they are the primary outcome of the spermidine trial covered in our research notes.

We are not claiming that regular caffeine impairs memory consolidation. The study measured spindle-band power, not consolidation, and did not test memory. But a persistent change in that band, unaffected by stopping the drug for a day, is the kind of finding that does not appear in any discussion of whether coffee is good for you.

4.What This Does and Does Not Settle

The withdrawal-reversal question — how much of caffeine's apparent daytime benefit is restoration of a habitual user's own baseline rather than enhancement above it — is not settled by this study, because this study measured sleep rather than daytime performance.

We should be straightforward that we could not obtain a source that settles it. The meta-analysis originally filed for this topic, which addresses the question directly, is closed access with no open copy available.

What this design does establish is that the confound is real and separable. A withdrawal condition behaves differently from a placebo condition on at least one measure — which means treating the two as equivalent, as the acute literature must, introduces an error of unknown size.

That is a reason to hold the effect sizes in Part One more loosely, not to discard them.

5.A Later Review Points the Other Way

This section was added after publication, because a systematic review we had not read contradicts the finding above on its most specific point.

James Chmiel at the University of Szczecin and Donata Kurpas at Wroclaw Medical University synthesised thirty-two human studies of caffeine and sleep EEG, published between 1980 and January 2026, with formal risk-of-bias assessment.

Their most consistent finding is suppression of low-frequency non-REM activity — slow-wave activity and the lowest delta frequencies. That is a reduction in the electrical signature of deep sleep, and it is not something the trial above measured as an outcome.

And on the sigma band specifically, they report the opposite direction to the one described in Section 3. Across the literature, caffeine frequently increased faster EEG activity, including the sigma and spindle range and beta, producing what they describe as a lighter, more aroused, more wake-like sleep profile.

We cannot reconcile that with the single trial above, and we are not going to pretend otherwise. One three-condition crossover in twenty young men found spindle-band power reduced; a review of thirty-two studies reports it more often increased. The honest position is that the direction of caffeine's effect on spindle activity is unsettled, and that Section 3 should be read as one result rather than the result.

One thing the review does settle, and it works against the reassuring reading of this article rather than for it. Quantitative EEG measures were repeatedly more sensitive than conventional sleep-stage variables, often revealing physiological disruption where sleep architecture looked unchanged.

That reframes the null in Section 2. Finding no change in total sleep time, sleep latency or architecture is what this review would predict even when caffeine is measurably altering sleep — because architecture is the insensitive measure. Their overall conclusion is that caffeine reliably alters sleep neurophysiology in the direction of reduced sleep depth and weakened homeostatic recovery.

The review also lists what moderates the size of these effects: dose, timing, habitual use, withdrawal state, age, circadian context, and adenosinergic genetic variation — the last of these centred on ADORA2A. That final moderator matters for Part Three, where a 2025 analysis found that fast caffeine metabolisers show more pronounced withdrawal symptoms, and different caffeine–cognition associations as a result.

Editorial Comment

Boost contains 100 mg of caffeine. The formulation page for it already records, as the thing this literature does not establish, that a recurring open question is how much of caffeine's apparent benefit is relief of withdrawal in habitual users rather than an effect on a rested brain.

That sentence was written before we read this study, and this study does not resolve it. We are leaving it as it stands.

The practical reading of this part is narrower than either the optimistic or the alarmist version. In healthy young habitual consumers, daytime caffeine at a fairly high dose did not disturb sleep architecture or subjective sleep quality. It did leave a measurable trace in the spindle band that persisted after stopping. Neither finding tells you what to do, and both are more interesting than the advice usually offered.

Part Three takes the longest available view: what the observational literature says about caffeine consumption across decades and the risk of cognitive decline.