REF-1690 · Wang, Zhang & Cho, 2026 · Frontiers in Physiology, · Paper · PubMed · Free full text

Caffeine before a cycling time trial made riders a little faster in both short (10 min or less) and long (20 min or more) efforts: across 23 randomised, double-blind, placebo-controlled trials (320 riders, mostly men), completion time improved by SMD −0.32 (95% CI −0.49 to −0.16; I² 0%), −0.30 in short and −0.37 in long trials, with no difference between them; mean power rose in short trials (SMD 0.27) but not significantly in long ones (0.07, 95% CI −0.23 to 0.37); GRADE moderate. Read in full

Main caveat: Small trials (7 to 46 riders); some crossover trials were analysed as independent groups, which widens intervals and changes weights.
fairly strong evidenceMeasured on recreational to trained athletesRead in full

Wang, Zhang and Cho (Keimyung University) asked whether a single caffeine dose before a cycling time trial helps, and whether the effect differs between short efforts, which sit above critical power, and longer ones, which are limited more by critical power, pacing and fatigue.

They prespecified two duration bands, 10 min or less and 20 min or more, and left out trials of 10-20 min as hard to place. Twenty-three randomised, double-blind, placebo-controlled trials qualified: 16 short (about 1.2 to 9 min) and 7 long (about 22 to 59 min), mostly men and mostly laboratory rides, with 3-6 mg/kg of caffeine given about an hour before in nearly all.

Caffeine shortened completion time overall (SMD −0.32, 95% CI −0.49 to −0.16) with no detectable heterogeneity, by similar amounts in short and long trials (−0.30 and −0.37; subgroup difference p = 0.70).

Mean power rose overall (SMD 0.20, 0.02 to 0.38) and in short trials (0.27), but the long-trial estimate (0.07) was not significant; that split was itself not significant (p = 0.28). No trial was rated high risk of bias in any domain, and GRADE certainty was moderate for both outcomes, marked down once because allocation concealment was often not described.

The authors call the benefit small but practically meaningful, and warn that the duration subgroups do not show that duration changes the effect.

Participants

23 trials, 320 riders by our sum of Table 1 (253 men, 54 women, 13 of unreported sex): 18 trials men only, 1 women only (14 women), 3 both sexes, 1 sex not reported.
Training status by the authors' McKay classification: 18 trials "trained" (recreationally trained or competitive), 5 recreational or non-trained; mean ages 16.3 to 36.2 years, two samples adolescent (16.3 and 17 years).
Habitual caffeine intake reported in 12 trials, from about 40 to about 460 mg/day where given in mg; caffeine withdrawal before testing in 21 trials (the text lists Alejo 2026 both among those with withdrawal and among the two that did not report it).
Sample sizes 7 to 46 per trial; countries Brazil (11 trials), Great Britain, Spain, New Zealand, the USA, Australia, Denmark.

Protocol

Search: five databases from inception to 14 January 2026, plus reference lists; 3,237 records, 1,845 screened, 194 full texts, 23 included (115 excluded for the task, which includes trials of 10-20 min).
Inclusion: healthy cyclists; one pre-exercise caffeine dose; placebo or control; a fixed-distance or fixed-time time trial reporting completion time or mean power; randomised trials, crossover or parallel; English, peer-reviewed.
Caffeine: 3 mg/kg in 5 trials, 5 mg/kg in 11, 6 mg/kg in 7; capsules in 18, solution in 4, form not reported in 1; about 60 min before in most, 50 min in 2, about 80-90 min where a preload preceded the time trial.
Time trials: short, 1 to 6.4 km (about 1.2-9 min); long, 10 to 40 km (about 22-59 min), duration from the reported or extractable mean completion time.
Analysis: random-effects SMD (Hedges' g); crossover trials without paired data analysed as independent groups, which the authors call conservative; I² and Cochran's Q; test for subgroup differences; leave-one-out; funnel plots and Egger's test; no meta-regression for dose, timing, training status or habitual intake (too few comparisons, uneven reporting).
Risk of bias and certainty: RoB 1 (seven domains) and PEDro (0-10); GRADE by two authors.
Funding and interests: no financial support received; no conflicts declared.

Key Results

Outcome (caffeine vs placebo)ComparisonsSMD (95% CI)pI²
Completion time, all22 (292 rides per condition)−0.32 (−0.49 to −0.16)0.00010%
Completion time, ≤10 minshort subgroup−0.30 (−0.50 to −0.10)0.0040%
Completion time, ≥20 minlong subgroup−0.37 (−0.65 to −0.09)0.010%
Mean power, all20 (243 rides per condition)0.20 (0.02 to 0.38)0.030%
Mean power, ≤10 minshort subgroup0.27 (0.05 to 0.50)0.020%
Mean power, ≥20 minlong subgroup0.07 (−0.23 to 0.37)0.640%
Negative SMD for time and positive SMD for power favour caffeine. Subgroup differences: time p = 0.70, power p = 0.28.
The review gives no result in seconds, watts or percent.
Risk of bias: no trial high risk in any RoB 1 domain; unclear allocation concealment in 14 trials, one trial each unclear for blinding, incomplete data, selective reporting and other bias; PEDro excellent in 18 trials and good in 5.
Publication bias: Egger's test p = 0.133 (time) and 0.588 (power) overall; subgroup p 0.083-0.340.
Sensitivity: removing any one trial did not change the pooled estimates or their direction.
GRADE: moderate for completion time and for mean power (one step down for risk of bias).

Practical Application

For a time trial, a tested caffeine dose of 3-6 mg/kg about an hour before is worth taking: across 23 placebo-controlled trials it made cyclists slightly faster (SMD −0.32), and about equally so in efforts of about 22-59 min (−0.37) as in efforts under 10 min (−0.30). The size is small by the usual benchmarks, and the review gives it in standard units only.
For a long time trial, judge caffeine by the clock rather than the power file. Completion time improved in the long trials while their mean-power estimate was small and not significant (0.07); the authors read this as gains beyond average power, and the difference between the duration bands was not significant, so it is not shown that caffeine works differently over longer efforts.
This agrees with [SYNTH-62]'s caffeine row and [SYNTH-67]'s default (3-6 mg/kg, tested in training): it adds a pooled estimate built only from cycling time trials of the two lengths, and its doses (3-6 mg/kg, mostly about 60 min before) sit inside that range. Inference, not tested here: a 40-km or 50-mile time trial lasting more than about an hour is beyond the longest trial pooled (about 59 min).
Not settled here: whether 3 mg/kg does as well as 6 mg/kg, whether habitual users gain less, or whether women respond the same (no meta-regression or sex subgroup was run); and time trials of 10-20 min were left out by design.

Limitations

Small trials (7 to 46 riders); some crossover trials were analysed as independent groups, which widens intervals and changes weights.
Only 7 long trials (86 riders by our sum), and 16 short; the subgroup tests had little power, and an I² of 0% does not mean the trials were alike in dose, training status, habitual intake, withdrawal or protocol.
Mostly men (253 of 320); one women-only trial.
No trial longer than about 59 min; trials of 10-20 min excluded.
Allocation concealment unclear in 14 of 23 trials; whether the riders guessed caffeine is not reported in the review.
Small inconsistencies in the text: one trial is listed both with and without a reported caffeine withdrawal.
Supplementary material (per-study risk of bias, search strings, sensitivity plots) not read.
How this entry was checked and changed
Why this level

a meta-analysis of controlled trials whose risk of bias we read in the full text (no trial high risk in any domain, allocation concealment unclear in 14; GRADE moderate), held at A- because every trial is small (7 to 46 riders, most 8-16)