Caffeine chewing gum, mostly chewed 5-10 min and taken within 15 min of exercise, improved performance only slightly: in a three-level meta-analysis of 25 placebo-controlled crossover trials (390 participants, 359 of them men) the pooled effect was SMD 0.195 (95% CI 0.115 to 0.275; GRADE low), 0.258 for aerobic endurance, 0.228 in trained and −0.038 (not significant) in recreationally active people; no trial was at low risk of bias. Read in full
Main caveat: No trial at low overall risk of bias; blinding often unclear and rarely checked, so expectancy may have added to the effect (the authors' caution).
Miraftabi and colleagues (University of Tehran, Australian Athletics, Universidad Pontificia Comillas, University of Western Australia) updated a 2023 meta-analysis on caffeine chewing gum, which delivers part of its caffeine through the lining of the mouth and so raises blood caffeine within minutes, and added chewing time and habitual intake as moderators.
They found 25 randomised, double-blind, placebo-controlled crossover trials in 390 healthy adults, mostly trained young men in team and strength sports, with doses of 1.35-6.4 mg/kg (most near 3 mg/kg) chewed for 5 or 10 min and taken mostly 0-15 min before exercise.
Across 47 outcomes, gum improved performance by SMD 0.195 (95% CI 0.115 to 0.275), a trivial effect by the usual benchmarks, with very low heterogeneity (I² 4%). Trained participants gained (0.228) and recreationally active ones did not (−0.038), and that difference was significant.
Exercise type, chewing time and habitual intake did not significantly moderate the effect, although the separate estimates were significant for aerobic endurance (0.258), anaerobic tests (0.186), jumps (0.183), 10-min chewing (0.262), and low and moderate habitual users, and not significant for strength and muscular endurance, 5-min chewing and the two high-intake outcomes.
Larger doses went with larger effects (β 0.100 per mg/kg), a result that depended on the assumed correlation and on two trials; timing did not matter within the narrow range tested. No trial was rated at low overall risk of bias and certainty was low to very low. The authors call the gain statistically significant but trivial, possibly useful where margins are small.
Participants
25 trials, 390 participants (359 men, 31 women), 8 to 27 per trial; 22 trials men only, 2 mixed, 1 women only (18 table-tennis players); mean ages about 18 to 34 years.
Training: 20 trials trained (k = 40), 4 recreationally active (k = 6), 1 not specified (k = 1).
Sports by Table 1: resistance training, soccer, rugby, basketball, volleyball, ice hockey, table tennis, judo, wrestling, CrossFit, sprinting, running (a recreational 5-km sample and well-trained runners) and cycling (trained cyclists in a 30-km time trial and 8 recreational cyclists riding to exhaustion).
Habitual caffeine: low (0-150 mg/day) in 10 trials, moderate (150-300) in 8, high (over 300) in 1 (14 men), not reported in 6; how intake was assessed was not reported for 12.
Protocol
Search: four databases to September 2025, updated March 2026; reference lists checked; grey literature, registries and theses not searched; 31 full texts assessed, 25 included; one eligible trial excluded for missing data.
Inclusion: randomised trials of caffeine gum against water or placebo gum in healthy adults with a physical performance outcome; co-ingested ingredients, swallowed caffeine and mouth rinse excluded; outcomes such as rate of force development, bar velocity and skill accuracy left out of the synthesis.
Doses: 1.35-6.4 mg/kg, absolute doses converted with each sample's mean body mass; most 2-4 mg/kg.
Chewing and timing: 5 min in 12 trials (k = 24), 10 min in 10 (k = 17), not reported in 3; mostly 0-10 min before exercise, k = 35 outcomes within 15 min and k = 3 at longer intervals (30-35 and 60 min).
Outcome groups: aerobic endurance (9 trials, k = 11), anaerobic (sprints, agility, rowing, RAST; 10, k = 14), jumps (12, k = 13), muscular strength and endurance (6, k = 9).
Analysis: Hedges' g from paired data with an assumed within-person correlation of 0.5 (0.2 and 0.8 in sensitivity, since no trial reported it); three-level REML model with study and effect-size levels; categorical moderators in mixed-effects models, dose and timing as continuous meta-regressions; leave-one-out; Egger's test where 10 or more effect sizes.
Risk of bias and certainty: RoB 2, five domains, by two reviewers; a modified Quality Criteria Checklist (its results are not given in the text); GRADE by one reviewer, checked by a second.
Funding and interests: no funding for the article; no competing interests declared.
Key Results
| Analysis | k, n | SMD (95% CI) | p | GRADE |
|---|---|---|---|---|
| All outcomes | 47, 390 | 0.195 (0.115 to 0.275) | 0.001 | low |
| Trained | 40, 304 | 0.228 (0.146 to 0.310) | — | low |
| Recreationally active | 6, 59 | −0.038 (−0.244 to 0.167) | 0.676 | very low |
| Aerobic endurance | 11, 122 | 0.258 (0.081 to 0.436) | 0.002 | very low |
| Anaerobic performance | 14, 156 | 0.186 (0.042 to 0.340) | 0.005 | very low |
| Jumps | 13, 195 | 0.183 (0.037 to 0.330) | 0.008 | very low |
| Muscular strength and endurance | 9, 97 | 0.142 (−0.038 to 0.322) | 0.372 | very low |
| Chewed 10 min | 17 | 0.262 (0.151 to 0.384) | 0.001 | low |
| Chewed 5 min | 24 | 0.142 (−0.030 to 0.262) | 0.091 | very low |
| Low habitual intake | 20, 153 | 0.207 (0.091 to 0.323) | 0.003 | very low |
| Moderate habitual intake | 12, 130 | 0.187 (0.038 to 0.336) | 0.013 | very low |
| High habitual intake | 2, 14 | −0.203 (−0.670 to 0.260) | 0.290 | very low |
Moderator tests: training status QM(1) = 5.587, p = 0.018 (the abstract attaches this p to the trained subgroup); exercise type p = 0.834; chewing time p = 0.142; habitual intake p = 0.176.
Heterogeneity: I² 4% (Q = 37.53, df 46, p = 0.809); 64.6% of the variance between studies, none within studies.
Dose: β = 0.100 per mg/kg (95% CI 0.002 to 0.192), p = 0.031; not significant with a correlation of 0.2 (β 0.083, p = 0.063), significant at 0.8 (β 0.162, p = 0.008); no longer significant when either of two trials was removed.
Timing: β = −0.002 per min (−0.010 to 0.006), p = 0.683.
Sensitivity: overall SMD 0.156 (correlation 0.2) and 0.282 (0.8), both p < 0.001; leave-one-out 0.182 to 0.215; without a trial with a possibly overlapping sample, 0.183 (0.103 to 0.263). The strength and endurance estimate became significant when one trial was removed.
Risk of bias: no trial low overall; some concerns mostly for randomisation and selective reporting; high risk in a few trials for deviations from the intended intervention, mainly unclear blinding; few trials checked whether blinding worked; carryover and period effects not assessed in any; washouts 3-14 days.
Publication bias: Egger z = −0.19, p = 0.848 overall, no asymmetry in subgroups where testable; the authors note low power with few independent trials.
Practical Application
Caffeine gum is a usable fast option, not a stronger one: chewed for 5-10 min shortly before exercise it gave a trivial overall gain (SMD 0.195) and a small one for aerobic endurance (0.258), in trained athletes rather than recreational ones. The authors place it where an hour-ahead capsule is impractical (half-time, a substitution, early-morning training) and in longer events; trying it in training first is the base's practice (SYNTH-67).
Inference, not tested here: a dose near 3 mg/kg chewed for the full 10 min is the safer choice. From this pool: most trials gave about 3 mg/kg (2-4 mg/kg), larger doses went with larger effects, and only 10-min chewing reached significance, though chewing time was not a significant moderator and no trial compared 5 and 10 min directly.
[SYNTH-67] says gum needs a shorter lead than capsules; this review supports the short lead as far as it goes, since gum chewed within 15 min of exercise (k = 35 outcomes, against 3 with a longer lead) gave the pooled gain, but timing did not moderate the effect (p = 0.683) and too few outcomes had a longer lead to show that gum needs a short one; the authors add that swallowed caffeine from gum gives a second, later rise in blood caffeine, so a longer lead is untested rather than ruled out. It did not compare gum with capsules.
Habitual users: low and moderate users gained; the one high-intake trial (14 men, 2 outcomes) showed a non-significant −0.203, too little to set a rule; REF-909 found habitual intake did not change the capsule response in trained cyclists.
Not settled by this design: gain in women (31 of 390 participants), in long endurance events (no trial resembles a race of hours), and whether gum taken well before the start, or during an event, works as well.
Limitations
No trial at low overall risk of bias; blinding often unclear and rarely checked, so expectancy may have added to the effect (the authors' caution).
Certainty low to very low; small trials (8 to 27); many subgroups rest on few trials (high habitual intake: one trial; recreationally active: 4 trials, 59 people).
Mostly men (359 of 390); one women-only trial, so no sex subgroup was possible.
Outcome groups are broad and mostly short tests; few endurance trials, the longest named task a 30-km cycling time trial (durations of the treadmill run and the ride to exhaustion not given).
The dose-response result depends on the assumed correlation and on two trials; timing was tested over a narrow range.
Doses converted with mean body mass; habitual intake measured in varied or unreported ways.
Our reading: Table 1 describes reference 19 (Paton) as a 30-km cycling time trial, while that reference's title reports repeated sprints in competitive cyclists; the review does not explain the difference.
Supplementary tables and figures not read.
How this entry was checked and changed
Why this level
a meta-analysis of randomised trials whose risk of bias we read in the full text, capped at A- because that assessment is serious: no trial at low overall risk, several high (mostly unclear blinding), GRADE low for the main effect and very low for most subgroups