REF-422 · Deng et al., 2026 (CMR meta-analytic review) · Journal of the International Society of Sports Nutrition · Paper · PubMed
Revisiting the evidence on caffeine mouth rinse: effects on exercise and cognitive performance
A meta-analytic review
Eligibility & Search
The search covered six electronic databases through 2 October 2025, looking for Caff-MR studies with exercise or cognitive outcomes
Study quality: modified PEDro and RoB-2; certainty graded with GRADE
Moderators set in advance: rinse duration, total oral exposure, sex, training status, feeding state, habitual caffeine use, and type of exercise or cognitive task
Sensitivity analyses for assumed within-subject correlations, outliers and influential cases
Studies Retrieved
31 studies (k = 167 effects) met inclusion
Exercise outcomes: k = 114 effects; cognitive outcomes: k = 53 effects
Rationale: Caff-MR may stimulate receptors in the mouth and throat (oropharyngeal) and quickly switch on central (brain) networks for motivation, attention and pacing, without the caffeine being absorbed into the body; the one earlier meta-analysis did not find a stable ergogenic effect
Key Findings — Effect Sizes
| Outcome | Pooled effect | Certainty (GRADE) |
|---|---|---|
| Exercise performance (k=114) | g = 0.12, p = 0.01 (trivial-to-small) | Moderate |
| Cognitive performance (k=53) | g = 0.23, p = 0.07 (inconsistent overall); after outlier removal the overall and speed-based effects reached significance, accuracy remained variable | Very low – low |
Moderators (from the abstract): the benefit was most reliable in aerobic endurance tasks and when athletes had eaten (fed state); short ~5-s rinses beat longer ones. A U-shaped dose-response (32–133 mg window) appeared in the primary analysis but did NOT survive removal of outliers; with 5-s rinses, more total oral exposure went with worse performance. Most studies were rated "some concerns" for risk of bias.
Conclusion (from source): the authors see Caff-MR as a practical option that needs no swallowing and gives small benefits that depend on context, best with short (~5 s) rinses and moderate exposure, and mainly for aerobic endurance. Cognitive applications are inconsistent. They call for standardised, adequately powered trials to settle dosing, timing and cognitive uses.
Mechanism
The authors' proposed pathway: oropharyngeal (bitter-taste) receptor activation rapidly engaging central networks for motivation, attention and pacing, without systemic absorption
Inference, not tested in this review: a reduced perception of effort is the most likely psychophysiological correlate; any minor buccal absorption of caffeine is not the proposed route
Because the route is oral-sensory rather than systemic, the usual costs of swallowed caffeine (GI load, sleep disruption) are not expected — a plausible advantage, not a measured outcome of this review
Practical Application
Across 31 studies, rinsing (not swallowing) a caffeine solution improved exercise performance by a trivial-to-small pooled g = 0.12; the effect was most consistent for aerobic endurance, in the fed state and with ~5-s rinses, while cognitive effects were inconsistent — a small, context-dependent tool, not a substitute for ingested caffeine
When Caff-MR is a reasonable choice (the base's own practice, not tested head-to-head in this review): a late-race top-up when there is no time for ingested caffeine to absorb; GI-sensitive athletes in long events; habitual users wanting a mild perceptual lift without raising blood caffeine
Practical protocol — Not from this paper; the base's own practice, shaped by the abstract's moderators: a brief ~5-s rinse with a moderate caffeine amount (the abstract's non-robust U-shaped window is 32–133 mg per rinse), spat out, not swallowed; longer rinses and higher total oral exposure did not help
Fed state: the abstract found the most consistent benefit in the fed state, so there is no reason to fast to "unlock" the rinse effect
Limitations
Heterogeneity in concentration, rinse duration, volume and timing across the 31 studies — the pooled effect masks protocol-specific magnitude
Predominantly male samples: female responses to Caff-MR understudied (sex was a prespecified moderator)
The dose-response (U-shape) was not robust to outlier removal
Risk of bias predominantly "some concerns"; GRADE certainty moderate for exercise, very low/low for cognition
Placebo design challenges: a bitter caffeine solution is hard to blind fully
Long-duration and elite-cohort data are thin (inference from the general picture; not quantified in the abstract)
No long-term safety/tolerance data for chronic Caff-MR use
Audit note (2026-07-06)
WAS NOT FOUND → FOUND: the entry is verified as real — Deng et al. 2026, JISSN 23(1):2638903, DOI 10.1080/15502783.2026.2638903, PMID 41774525 (open access).
Level A → B+: the source reports a pooled effect on endurance of g=0.12 (trivial-to-small), certainty exercise = moderate, cognition = very-low–low.
Corrected: number of studies ("~46 RCT" → 31; k=167 effects); effect sizes (were overstated ~3-fold).
How this entry was checked and changed
Why this level
regraded 2026-09-27 from B+: meta-analysis of 31 controlled studies with GRADE; heterogeneous, mostly small trials (the earlier B+ graded the size of the effect, not the design)
Content correction 2026-09-27
The whole "[unconfirmed]" moderator block (concentration ≥1.2% / <1% / 0.3%, fasted g≈0.45-0.65 vs fed g≈0.20-0.35, ≥75% VO2max / 30-60 min, sprint <30 s, rinse 5-10 s vs 20-30 s, repeated every 10-15 min, trained > untrained, habitual users) -> not in the abstract and partly contradicting it (fed state won) — deleted.
Mechanism numbers (plasma rise within 5-15 min, onset 1-5 min vs 30-60 min, TMS corticospinal excitability) -> not in the abstract — deleted; mechanism restated as the authors' proposal plus labelled inference.
Practical protocol (1.2-1.6%, 120-160 mg/100 mL, 25-50 mL, every 10-15 min) and "Magnitude vs ingested caffeine" (g≈0.30-0.55 for CMR, which contradicts the pooled g=0.12) -> deleted; replaced by a protocol labelled as the base's own practice built on the abstract's moderators (~5 s, 32-133 mg window, fed state).
Limitations numbers (0.3-2.0%, 5-60 s, 5-50 mL, <60 min, >3 h) -> not in the abstract — removed. Level B+ -> A- (design-based regrade); added Tier and Sport lines.
Cross-reference pointed at an entry whose content does not match the label (checked against the cited entry); repointed or relabelled.