# Why coffee and sleep studies disagree

## Bottom line

The studies disagree mainly because they do not test the same exposure or the same outcome. A 100 mg morning or afternoon dose in a healthy, caffeine-tolerant young adult is not equivalent to 400 mg near bedtime, four self-reported “cups” spread through a day, or caffeine taken during 38 hours of sleep deprivation. Likewise, a questionnaire asking whether sleep felt good can miss changes in sleep latency, fragmentation, deep-sleep EEG power, or total sleep time detected by polysomnography.

The most defensible synthesis is dose- and time-dependent: large doses reliably disrupt subsequent sleep, sometimes many hours later; a single 100 mg dose may have little detectable effect when taken at least four hours before bed in healthy moderate users. Habitual-intake studies are less consistent because tolerance, reverse causation, beverage size, smoking, work schedules, and participant differences blur the acute pharmacological effect.

## Representative human evidence

| Study | Dose and timing | Participants | Sleep measure | Main finding |
|---|---|---|---|---|
| [Gardiner et al., 2025](https://pubmed.ncbi.nlm.nih.gov/39377163/), [DOI](https://doi.org/10.1093/sleep/zsae230) | 100 or 400 mg at 12, 8, or 4 h before bed; randomized crossover | 23 healthy men, mean age 25, habitual intake under 300 mg/day | In-home partial polysomnography plus diary | 100 mg caused no significant measured impairment; 400 mg altered sleep initiation/architecture even 12 h before bed, with more fragmentation closer to bed. Subjective quality worsened only at 4 h. |
| [Drake et al., 2013](https://pubmed.ncbi.nlm.nih.gov/24235903/), [DOI](https://doi.org/10.5664/jcsm.3170) | 400 mg at bedtime, 3 h, or 6 h before bed; placebo-controlled | 12 healthy habitual sleepers | Portable objective monitor plus diary | Sleep was disrupted at all three times, including 6 h before bed. |
| [Pauchon et al., 2024](https://pubmed.ncbi.nlm.nih.gov/39458438/), [DOI](https://doi.org/10.3390/nu16203442) | 2.5 mg/kg twice during 38 h awake; last dose 6.5 h before recovery sleep | 41 adults in double-blind crossover total-sleep-deprivation protocol | Connected-headband polysomnography | Caffeine shortened recovery sleep by about 30 min and N3 by about 36 min, and increased awakenings and stage transitions. This is recovery sleep, not an ordinary night. |
| [Baur et al., 2024](https://pubmed.ncbi.nlm.nih.gov/38221756/), [DOI](https://doi.org/10.1111/jsr.14140) | 160 mg delayed-release at habitual bedtime; sleep opportunity 4.5 h later | 21 healthy young men; randomized crossover | Standard polysomnography, plasma caffeine, EEG spectra, autonomic measures | Effects tracked plasma concentration: higher levels reduced NREM delta activity, with very large person-to-person concentration variation after the same dose. |
| [Stucky et al., 2025](https://pubmed.ncbi.nlm.nih.gov/41109744/), [DOI](https://doi.org/10.1177/02698811251368364) | Habitual intake: at least 4 versus at most 3 caffeinated drinks/day; no clock-time dose | UK Biobank genetic/intake data (485,511) and HypnoLaus (1,702) | At-home full polysomnography, questionnaires, Mendelian-randomization and matching analyses | High habitual intake was linked to shorter objective sleep, while self-rated sleep quality did not differ. Causal estimates ranged implausibly widely (11–229 min), so direction is more credible than magnitude. |
| [Watson et al., 2016](https://pubmed.ncbi.nlm.nih.gov/27527212/), [DOI](https://doi.org/10.3390/nu8080479) | Habitual intake estimated by food-frequency questionnaire | 80 adults, broad age range; cross-sectional | Pittsburgh Sleep Quality Index | Poor sleepers consumed more caffeine, but most PSQI components were not associated. Direction of causation cannot be established. |
| [Cusick et al., 2020](https://pubmed.ncbi.nlm.nih.gov/32386419/), [DOI](https://doi.org/10.1093/jpepsy/jsaa033) | Number and time-of-day of caffeinated drinks | 302 adolescents aged 12–14, with and without ADHD | Actigraphy plus adolescent/parent reports | Afternoon/evening use related to reported sleep problems, with subgroup differences; actigraphy and reports did not necessarily tell the same story. |
| [Riera-Sampol et al., 2022](https://pubmed.ncbi.nlm.nih.gov/35458223/), [DOI](https://doi.org/10.3390/nu14081661) | Self-reported caffeine from coffee and other products | 886 university students aged 18–25 | Cross-sectional survey/self-reported sleep quality | Associations coexist with smoking, alcohol, cannabis use and performance-motivated caffeine use, leaving substantial confounding and reverse causation. |

A 24-study meta-analysis estimated that caffeine reduced total sleep time by about 45 minutes and sleep efficiency by 7%, increased sleep-onset latency by 9 minutes and wake after sleep onset by 12 minutes, shifted sleep toward lighter stages, and suggested longer cutoffs for larger doses ([Gardiner et al., 2023](https://pubmed.ncbi.nlm.nih.gov/36870101/), [DOI](https://doi.org/10.1016/j.smrv.2023.101764)). Those pooled averages combine heterogeneous doses, timings, participants and measurement systems; they describe an average response, not a universal cutoff.

## Why conclusions diverge

### Dose is often mislabeled

“Coffee” is not a standardized dose. Cup volume, bean, brew, and serving strength vary; tea, cola, energy drinks and pre-workout products add other ingredients. Observational studies often count cups, whereas trials administer anhydrous caffeine in milligrams. The 2025 crossover result—no detectable effect for 100 mg but clear disruption for 400 mg—shows why conclusions change when papers use the same word, “caffeine,” for a fourfold exposure difference.

Body-size normalization matters too. A fixed 400 mg is a much larger mg/kg dose for a smaller person. Plasma levels also differ after identical doses because caffeine clearance varies with CYP1A2 genetics, smoking, pregnancy, oral contraceptives, liver function and interacting drugs.

### Timing changes residual exposure

Caffeine has a typical half-life of several hours, but variation is large. “Afternoon” is therefore too crude: noon caffeine before a midnight bedtime differs from 5 p.m. caffeine before a 10 p.m. bedtime. Studies aligned to each participant’s bedtime are more interpretable than studies using clock time. Multiple small doses can also maintain exposure differently from one bolus.

### Participants are not interchangeable

Many rigorous laboratory trials are tiny crossovers in healthy young men, chosen to reduce noise. They estimate a clean acute effect but generalize poorly to women, older adults, adolescents, shift workers, pregnant people, people with insomnia or ADHD, and medication users. Habitual consumers may develop partial tolerance or select their dose based on known sensitivity. Genetic “fast” and “slow” metabolizers can have different bedtime concentrations.

A particularly important bias is reverse causation: poor sleep causes next-day caffeine use, while caffeine can then worsen the following night. Cross-sectional studies cannot reliably separate those directions. Smoking, alcohol, stress, workload and chronotype influence both caffeine use and sleep.

### Sleep is not one endpoint

Polysomnography measures stage architecture, awakenings and EEG activity; actigraphy infers sleep from movement; consumer headbands use proprietary algorithms; diaries record perceived latency and quality; PSQI summarizes sleep over a longer period. These measures are related but not equivalent.

The 2025 crossover and community studies both found objective changes that participants did not necessarily perceive. A paper can therefore truthfully conclude “no change in subjective sleep quality” while another finds less N3 sleep or lower delta power. Small studies also lack power for every endpoint, and testing many sleep variables raises the chance that only some reach statistical significance.

## Practical interpretation

For an ordinary healthy adult, the evidence does not support either extreme—“any coffee ruins sleep” or “coffee does not matter.” A small dose early enough may have little measurable effect, particularly in a habitual consumer. A large dose can impair objective sleep even when taken 6–12 hours before bed and even when the person does not feel that sleep was worse.

The most informative personal test is to hold bedtime stable, track the actual caffeine dose and time, and compare several caffeine-free or early/low-dose days with usual-use days. Subjective ratings alone may miss disruption; a validated clinical assessment is preferable if insomnia, marked daytime sleepiness, pregnancy, medication interactions, or other health concerns are present.

## Scope and limitations

This review prioritizes recent human studies indexed in PubMed through 23 September 2026 and full texts available in PubMed Central, plus an older benchmark timing trial. The newest directly relevant studies are not all traditional randomized trials; causal-inference and observational findings depend on stronger assumptions. Small crossover trials establish acute effects well but have narrow samples, whereas large community studies are more representative but measure exposure less precisely.