Caffeine cost about 35 minutes of sleep in pooled lab trials

That caffeine disturbs sleep is not in question. What has been harder to pin down is how much, at what dose, and whether it lands differently on a 20-year-old than on a 50-year-old. A meta-analysis published in Sleep Medicine on October 18 set out to quantify exactly that, and the honest summary of its result is that the main effect is solid and the subgroup story is not [s1].

What was pooled

The authors searched PubMed, Web of Science, CINAHL and the Cochrane Library through March 25, 2025, for controlled crossover trials that measured night-time polysomnography after acute administration of caffeine or placebo in healthy adults [s1]. Twenty-two trials, covering 956 participants, met the inclusion criteria [s1].

The crossover design matters. Each participant serves as their own control, which removes between-person variation in sleep — the largest source of noise in this literature — and polysomnography means the outcome is measured electrophysiologically rather than self-reported.

The pooled effects

In random-effects meta-analysis, caffeine reduced total sleep time by a mean difference of 34.67 minutes, reduced sleep efficiency by 4.74%, and reduced the proportion of slow-wave sleep by 1.01% [s1]. It lengthened sleep onset latency by 8.35 minutes [s1]. Effects on the proportion of REM sleep were not significant [s1].

Leave-one-out sensitivity testing confirmed the stability of the pooled estimates — that is, no single trial was driving them [s1].

These numbers sit close to, though somewhat below, an earlier meta-analysis of 24 studies published in Sleep Medicine Reviews in 2023, which found caffeine reduced total sleep time by 45 minutes and sleep efficiency by 7%, increased sleep onset latency by 9 minutes and wake after sleep onset by 12 minutes, with light sleep up by 6.1 minutes and deep sleep down by 11.4 minutes [s2]. Two independent syntheses landing in the same region, from overlapping but not identical trial sets, is the more persuasive fact here.

Where the analysis stops short of its own title

The paper is titled for age- and dose-specific effects, and the subgroup findings are the part that gets quoted. They should be read carefully.

Caffeine reduced total sleep time in both young and middle-aged groups [s1]. In the young group it additionally reduced sleep efficiency, lengthened sleep onset latency and decreased the slow-wave sleep proportion [s1]. Both high- and low-dose caffeine reduced total sleep time and sleep efficiency and increased sleep onset latency; only high-dose caffeine additionally reduced slow-wave sleep [s1].

But the between-group differences were not statistically significant [s1]. And mixed-effects meta-regression found no significant linear relationship with continuous age or with continuous dose [s1].

That combination — an effect reaching significance in one subgroup and not another, with no significant difference between them — is one of the most commonly misread patterns in evidence synthesis. It is compatible with a real age or dose gradient that this dataset is too small to detect. It is equally compatible with no gradient at all, where the subgroup with more trials or more participants simply had more power. The meta-regression, which uses age and dose as continuous variables rather than splitting them, is the more sensitive test, and it found nothing.

The authors' own conclusion is that caffeine disrupted multiple sleep parameters in younger and high-dose consumers, and they argue this underlines the need for age- and dose-specific public-health guidance [s1]. The data support the first half of that sentence more firmly than the second.

Timing, which this analysis does not address

The question most people actually have is when to stop drinking coffee. This meta-analysis pooled acute administration trials; it did not model a cut-off time [s1].

The 2023 review did attempt that translation, concluding from its pooled estimates that to avoid reductions in total sleep time, coffee at 107 mg per 250 mL should be consumed at least 8.8 hours before bedtime, and a standard serving of pre-workout supplement at 217.5 mg at least 13.2 hours before [s2]. Those thresholds are model-derived from pooled averages rather than tested directly, and they describe group means, not individual sensitivity, which varies substantially between people for reasons including caffeine metabolism.

The limits worth keeping in view

Twenty-two trials and 956 participants is a respectable but not large evidence base for a substance consumed by most of the world's adults [s1]. All included studies were in healthy adults [s1], so the findings do not speak to people with insomnia disorder, shift workers, or anyone with a sleep or psychiatric condition.

These were acute administration trials [s1]. Habitual consumption, tolerance and withdrawal are different questions, and a single laboratory dose in someone who does not normally consume caffeine is not the same exposure as a daily habit.

What to watch

Whether larger individual-participant-data syntheses can resolve the dose-response question the study-level meta-regression could not; whether trials in people with insomnia show larger effects than in healthy sleepers; and whether genotype-stratified work on caffeine metabolism explains some of the variation between individuals.

This article describes pooled research findings and is informational only. It is not medical advice and does not recommend any change in caffeine consumption.

Sources

• [s1] Age- and dose-specific effects of caffeine on sleep: A meta-analysis of controlled crossover trials, Sleep Medicine, 2025;136:106874, published online 2025-10-18.
• [s2] The effect of caffeine on subsequent sleep: A systematic review and meta-analysis, Sleep Medicine Reviews, 2023;69:101764, published online 2023-02-06.