REF-1517 · Martinho et al., 2025 · Journal of the International Society of Sports Nutrition · Paper · PubMed · Free full text
Does caffeine or carbohydrate help in a mixed workout?
CrossFit trials: caffeine g 0.37, not significant (the abstract says significant); carbohydrate g −0.13 (CI −0.65 to 0.39), no shown effect. None described at HYROX length.
Main caveat: Abstract contradicts results. "the effect of caffeine was significant" and "Caffeine should be used prior to or during" vs "moderate (Hedges' g = 0.371) but…
Martinho and colleagues searched four databases and grey literature on 11 June 2024 for nutrition in adult CrossFit participants. On intake (self-report), mean energy was 2247 ± 606 kcal a day, carbohydrate 3.6 and protein 1.7 g/kg/day; the authors judge carbohydrate below the 6-12 g/kg/day athlete range, and protein adequate. Supplement use averaged 72% (63-79%).
Three studies of higher-carbohydrate diets favoured them (g = 0.487). A ketogenic diet showed no benefit in two unpooled studies. Acute carbohydrate had no effect.
For caffeine, four crossovers gave g = 0.371, reported as not significant, falling to 0.144 (−0.409 to 0.696) after trim-and-fill; the abstract then calls the effect significant and says caffeine should be used, while the paper's own conclusions say it "may be utilized". Bicarbonate was not pooled because findings were inconsistent.
Our reading: this is the best single map of CrossFit nutrition, and its caffeine headline should not be quoted from the abstract. It does not overturn REF-350 or SYNTH-67 on caffeine in endurance.
It gives no evidence on carbohydrate during a one-hour mixed race, and the advice to use creatine and protein rests on how common they are and on outside literature, since it reviews no creatine trial.
Protocol
Search: PubMed, Web of Science, Scopus, SportDiscus, plus theses and trial registers; 11 June 2024; English, Portuguese, Spanish; n = 49 studies per the abstract (57 records in the flow: 51 papers plus 6 from other sources).
Meta-analysis: Hedges' g, random effects (Comprehensive Meta-Analysis v2.2); funnel plot, Egger's test, trim-and-fill.
Quality: NIH-type tool for observational studies; PEDro for trials: 19 excellent, 9 good, 4 fair; no study excluded for risk of bias.
Trials pooled: caffeine 60 to 70 min before, 3 to 6 mg/kg in the discussion, 5 to 6 by Table 3 with Główka's medium dose (Główka 2024; Caetano 2023; Stein 2020; Fogaça 2020; pooled n = 73, which needs Table 3's 30 for Stein, while Stein's own abstract says 20, making 63 by our arithmetic); carbohydrate 30 g during, 16 g before and during, or ad libitum (Grijota 2024; Rountree 2017; Howarth 2017; which three is inferred from Table 3 and the pooled n = 40); higher-carbohydrate diets (Ficarra, Mediterranean, 8 weeks; Durkalec-Michalski, vegan, 4 weeks; Escobar, 6–8 g/kg, 3 days; pooled n = 31).
Funding and interests: Portuguese Recovery and Resilience Program (PRR), IAPMEI/ANI/FCT, Agenda C645022399–00000057 (eGamesLab). No conflicts reported. A co-author, Durkalec-Michalski, led several included trials.
Key Results
| Question | Effect | CI / p | Note |
|---|---|---|---|
| Caffeine before a workout | g = 0.371 ("moderate" in the results, "not significant") | no CI in the text for the unadjusted value; forest plot image not read | 4 crossovers, pooled n = 73 |
| Caffeine, trim-and-fill | g = 0.144 | −0.409 to 0.696 | Egger's intercept = 2.506, p = 0.05, read by the authors as no bias |
| Carbohydrate before or during | g = −0.130 | −0.652 to 0.391, p = 0.624 | 3 crossovers, pooled n = 40; Egger p = 0.003 |
| Carbohydrate, trim-and-fill | 0.063 | − 0.492 to 0.613 | |
| Higher-carbohydrate diet | g = 0.487 | 0.110 to 0.886 | 3 studies, pooled n = 31; heterogeneity low; called "moderate" though the paper's own scale has 0.2–0.6 small |
| Ketogenic diet | not pooled | 2 studies, no benefit in either | |
| Bicarbonate | not pooled | inconsistent |
| Intake and use | Value |
|---|---|
| Energy, overall | 2247 ± 606 kcal.day−1; females 1746 ± 40, males 2360 ± 174 |
| Carbohydrate | 3.6 g.kg−1.day−1 overall; females 3.4, males 3.2 |
| Protein | 1.7 g.kg−1.day−1 |
| Energy balance against one Polish study's expenditure | −468 males, −852 females; with a 19% under-reporting correction −20 and −520 kcal |
| Supplement use | 72% (95% CI: 63% to 79%), range 25% to 98%, I² 93% |
| Creatine, protein | 26% to 61%; 100% to 28% |
Mechanisms
Measured: performance effect sizes in trials, and self-reported intake and use.
Authors' interpretation: carbohydrate during short workouts adds nothing since hours-before stores suffice and the protocols did not deplete glycogen; they ask for trials of longer sessions. They note that carbohydrate during resistance training tends to help when sessions exceed 45 minutes (outside work).
Authors' interpretation: the non-significant caffeine result "could be explained by the demands of CrossFit", which mixes endurance, strength and power. This is a suggestion; the review does not test it.
Inference: the four caffeine trials differ in test and dose; Table 3 reports Stein (printed as 30 men; its own abstract, PMID 32340121, says 20) as no difference and Główka (26) as mixed, so a positive pooled point estimate leans on the small studies, which fits the trim-and-fill drop.
Practical Application
Does caffeine or carbohydrate help in a mixed workout? CrossFit trials: caffeine g 0.37, not significant (the abstract says significant); carbohydrate g −0.13 (CI −0.65 to 0.39), no shown effect. None described at HYROX length.
Carbohydrate in HYROX: no evidence here for or against fuel during a race of an hour or more; keep SYNTH-02's rows. What the review suggests is daily carbohydrate intake often runs low, and diets that raise it did better, in very few people.
Women's energy intake: a low reported intake (1746 kcal) stays a signal to ask about, with the caveat that intake is self-reported and expenditure rests on one study. Not from this paper; the base's own practice: use the REDs screen in SYNTH-54.
Limitations
Abstract contradicts results. "the effect of caffeine was significant" and "Caffeine should be used prior to or during" vs "moderate (Hedges' g = 0.371) but not significant", then "small and nonsignificant"; no pooled trial gave caffeine during a workout.The conclusions say "6 mg.kg−1 caffeine 60–70 minutes before exercise had a small effect on performance", from a pool whose doses the discussion gives as 3 to 6 mg/kg and whose result the same paper reports as not significant.
Bicarbonate told three ways. For Fran: Table 3 improved; Results no difference; Limitations improved. For Cindy: Table 3 no difference; Results positive; Limitations no effect. Do not quote this review on bicarbonate; read the trials.
Small pools. 3-4 studies each; the diet pool's n = 31 is below Table 2's sample sizes for those three studies (22, 20, 18), with no explanation. Diet labels in the Dietary interventions text pair the wrong references (high carbohydrate [58], Mediterranean [62]; the reference list is the other way round).
Diets differ in more than carbohydrate. Vegan and Mediterranean diets change food quality and possibly energy as well; 3 days of high carbohydrate is not 8 weeks.
Self-report. Intake from recalls, diaries and questionnaires; energy expenditure from one heart-rate-based study, and wearable error is stated by the authors as 15.1% to 57.0%. Means combine small samples, and the overall carbohydrate mean (3.6) exceeds both sex means (3.4, 3.2).
Numbers that do not add up. 784 records minus 281 duplicates is 503, printed 502. Abstract 49, flow 57. PEDro "5 to 11 points" on a scale scored to 10. Creatine range 26% to 61%, while Table 1 lists Brisebois (n = 2576) at 23%.
A recovery citation. Its introduction credits 48 to 72 hours to reference 12, an RPE paper, and repeats CK higher at 72 hours after Karen (REF-1513's limitation).
Population: mainly Brazilian in the intake studies (~70%); Stein 2020 is the valid version, and the retracted 2019 one is not cited.
How this entry was checked and changed
Why this level
meta-analyses of mostly small trials would reach A- by design; held at B because each pool has 3-4 small studies and a pooled n of 31-73, the abstract's caffeine statement contradicts its own results, bicarbonate is reported three different ways, and intake data are self-report
Full text (checked 2026-10-30)
Four crossover studies, pooled n = 73: Hedges' g = 0.371, not significant; adjusted 0.144 (95% CI: −0.409 to 0.696) (full text)
The effect of caffeine ingestion on performance was small and nonsignificant (full text)
Three crossover studies of CHO, pooled n = 40: Hedges' g = −0.130; 95% CI: −0.652 to 0.391, p = 0.624 (full text)
Three diet studies, pooled n = 31: Hedge's g = 0.487; 95% CI: 0.110 to 0.886 (full text)
Mean energy 2247 ± 606 kcal.day−1; average CHO 3.6 and protein 1.7 g.kg−1.day−1 (full text)
Supplement prevalence: mean 72% (95% CI: 63% to 79%), I2 93% (full text)
Forty-nine papers from the four databases, two additional references and six other records: the present review included 57 records (full text)
Funded by the Portuguese Recovery and Resilience Program (PRR) (eGamesLab); no conflict of interest reported (full text)