REF-1705 · Yu et al., 2026 · BMC Sports Science, Medicine and Rehabilitation, · Paper · PubMed · Free full text
Cold water after a session did not speed the return of strength: in a network meta-analysis of 87 randomised trials (2,313 healthy athletes and active adults, 85% of studies men only, outcomes within 72 h), no cold-water protocol differed from passive rest for strength after resistance exercise (moderate certainty; medium-duration moderate-temperature bath g 0.00, 95% CrI −0.38 to 0.38), and none differed for aerobic performance after endurance exercise (low certainty); only very cold, long baths (5-8 °C, over 15 min) came out ahead for jump and power measures (g 0.80 after resistance exercise, 1.00 after team sport; low certainty). The review measured acute recovery, not training adaptation, and its own strength table contradicts its abstract on which option ranked best, while its stated coding rule and rankings conflict with how it reads its soreness, muscle-damage and endurance-recovery results
Main caveat: The authors': star-shaped networks with passive rest at the centre, so most comparisons are indirect and node-splitting was not possible.
Yu and colleagues asked which cold-water immersion protocol, by water temperature and time in the water, best restores performance after three kinds of exercise: resistance training, endurance exercise and team sports.
They pooled 87 randomised trials published from 2014 to mid-2025 in healthy athletes and physically active adults, coded each bath into a temperature and a duration band, and ran a separate Bayesian network for each modality and outcome that had at least 10 studies, ranking options by SUCRA.
After resistance exercise, no cold-water protocol, and no other recovery option, differed from passive rest for strength (24 studies, 598 participants; moderate certainty); a cold, long bath was the one option ahead of passive rest for power (17 studies; g 0.80, 95% CrI 0.03-1.57; low certainty).
After endurance exercise, nothing differed from passive rest for aerobic performance (11 studies; low certainty), and for endurance recovery (time to exhaustion or time-trial time) only an unspecified "control" condition came out worse than passive rest.
After team-sport exercise, the cold, long bath was ahead of passive rest for power (18 studies; g 1.00, 0.10-1.89; low certainty). Pooled across modalities, a moderate-temperature, medium-duration bath differed from passive rest for muscle-damage markers (g −0.51, −0.86 to −0.16) and perceived recovery (g −0.66, −0.98 to −0.33), both low certainty.
The authors conclude that the best protocol depends on the modality, that passive recovery is preferable after strength-focused resistance training, and that 9-12 °C for 10-15 min may help endurance recovery.
Our reading: the strength null is the most trustworthy result; the "passive recovery was highest" line, the endurance recommendation and the direction of the soreness and muscle-damage effects are contradicted by the paper's own Table 6 and coding rule, and nothing here measures adaptation.
Participants
87 randomised trials; the abstract and Discussion give 2,313 participants, the Results text 2,847; trial sizes from 6-8 per condition in small crossovers to 96 across five groups.
Study mean ages 16.9 ± 1.1 to 35.9 ± 7.8 years, most 20-26 (two trials had mean ages under the 18-year inclusion floor: 16.9 and 17.9).
74 studies (85%) men only, 2 women only, the rest mixed.
Training status by the primary studies' labels: recreational 44 studies, elite 21, competitive 14, physically active 5 (with trained and untrained in Table 1); 23 countries, most from Australia (19) and Brazil (18).
Modality: resistance training 32 studies, team sports 26, endurance exercise 29, assigned by the session's main demand.
Protocol
Search: four databases, publications 1 January 2014 to 30 June 2025, run after registration; English only; two reviewers screened (full-text agreement κ = 0.89) and extracted.
Inclusion: randomised parallel or crossover trials; healthy athletes or active adults aged 18 or over in organised training; cold water ≤ 20 °C within 1 h after exercise; a comparison condition; outcomes within 72 h after exercise. Excluded: clinical or injured samples, non-randomised designs, cold water combined inseparably with another intervention, and chronic protocols without acute outcomes.
Protocols seen: water from 5 to 20 °C, most often 9-12 °C, most often 10-15 min; passive rest was the comparator in 62% of trials.
Networks: seven analysed (the paper says six; team-sport power is missing from its Table 3) out of 32 candidates: resistance-training strength (1RM, maximal voluntary contraction) and power (jumps, peak power); endurance aerobic performance (VO2max, maximal aerobic power) and endurance recovery (time to exhaustion, time-trial time); team-sport power; and, across all modalities, muscle damage (CK, LDH, myoglobin) and perceived recovery (soreness, recovery scales).
Analysis: Hedges' g with 95% credible intervals, vague priors, 4 chains; τ < 0.2 low, 0.2-0.5 moderate, > 0.5 substantial heterogeneity; SUCRA rankings; crossover trials without reported within-person correlations (51, 59%) analysed as independent groups; no node-splitting in most networks (no closed loops); Egger's test where 10 or more studies allowed; sensitivity analyses without high-risk trials and by design; GRADE per network.
Funding and interests: two Anhui provincial university research projects (sports intangible cultural heritage; badminton performance characteristics); no competing interests declared; no equipment maker named.
Key Results
Each option against passive rest (Table 6; Hedges' g, 95% CrI; positive = better by the paper's stated coding):
| Network (studies, participants; certainty) | Result |
|---|---|
| Resistance training, strength (24, 598; moderate) | No option differed from passive rest (all p > 0.05): cold-long 0.57 (−0.11 to 1.24); moderate-medium 0.00 (−0.38 to 0.38); mild-medium −0.12 (−0.52 to 0.28); cold-short −0.22 (−0.87 to 0.43); moderate-long −0.43 (−1.05 to 0.19); thermoneutral water −0.51 (−1.14 to 0.12). Passive rest ranked 8th of 13 (SUCRA 47.1%). τ 0.065, I² 2.4%; Egger p = 0.415 |
| Resistance training, power (17, 389; low) | Cold-long 0.80 (0.03 to 1.57), SUCRA 85.0%, the only option ahead of passive rest; moderate-medium 0.18 (−0.25 to 0.61); passive rest 10th of 13 (34.6%). τ 0.000 |
| Endurance, aerobic performance (11, 235; low) | No option differed: cold-medium 0.17 (−0.71 to 1.05), SUCRA 75.1%; passive rest 2nd (69.2%); moderate-medium −0.26 (−0.65 to 0.12). τ 0.000 |
| Endurance, recovery: time to exhaustion, time trial (17, 463; low) | Only "control" differed, worse than passive rest: −1.58 (−2.62 to −0.53), p = 0.003; moderate-medium −0.47 (−0.97 to 0.04); moderate-long −0.79 (−1.69 to 0.11); passive rest 4th of 12 (70.2%). τ 0.321, I² 42.7% |
| Team sport, power (18, 418; low) | Cold-long 1.00 (0.10 to 1.89), SUCRA 93.3%; moderate-medium 0.15 (−0.28 to 0.57); passive rest 7th of 9 (38.9%). τ 0.316, I² 35.5% |
| All modalities, muscle-damage markers (37, 862; low) | Moderate-medium −0.51 (−0.86 to −0.16), p = 0.005, yet ranked last (SUCRA 24.1%); passive rest 2nd (72.0%). τ 0.388, I² 44.0%; design-by-treatment Q p = 0.009 |
| All modalities, perceived recovery (47, 1,114; low) | Moderate-medium −0.66 (−0.98 to −0.33), p < 0.001 (SUCRA 30.6%); cold-long −0.81 (−1.51 to −0.11), p = 0.024 (24.2%, last); passive rest 2nd (78.2%). τ 0.507, I² 57.5% |
Indirect evidence: 69.2% of strength comparisons, 78.2% of power, 66.7% of aerobic performance, 75.8% of endurance recovery, 75.0% of team-sport power, 52.6% of muscle damage and 84.8% of perceived recovery rested on indirect comparisons through passive rest.
Crossover against parallel trials: for the most-studied protocol (moderate-medium) the design subgroups differed by 0.15 g for strength and by 0.26-0.53 g in every other network (power 0.42 vs 0.08; team-sport power 0.32 vs −0.12; perceived recovery −0.96 vs −0.43); GRADE downgraded those networks for inconsistency on this ground.
Risk of bias (RoB 2): overall 2 trials low, 75 some concerns, 10 high; deviations from intended interventions were low risk in 2 trials only (blinding a cold bath is not possible); randomisation low in 74, outcome measurement low in 63, selective reporting low in 78.
Practical Application
An ice bath does not bring strength back faster after lifting. Across 24 trials no cold-water protocol differed from passive rest for strength within 72 h (moderate certainty; the most common 9-12 °C, 10-15 min bath g 0.00, −0.38 to 0.38), and none differed for aerobic performance after endurance work (low certainty).A very cold, long bath (5-8 °C, more than 15 min) was ahead for jump and power measures after resistance and team-sport sessions (g 0.80 and 1.00), at low certainty with intervals reaching near zero.
This does not change [SYNTH-70]'s line that cold water after lifting costs gains, and does not add to it. That line rests on training studies (REF-011, REF-914); this review measured recovery within 72 h, not adaptation, so its abstract's phrase about passive recovery protecting "adaptive responses" was not tested here.It agrees with SYNTH-70's cold-water row that cold water does not speed strength recovery (REF-250). It does not show that passive rest beats cold water either: in the paper's own strength table passive rest ranked 8th of 13 with no option different from it, which contradicts the abstract's "passive recovery was highest".
Soreness and muscle-damage markers: direction not settled by this paper. The authors read the moderate-temperature, medium-duration bath as lowering markers (g −0.51) and soreness (g −0.66); by the coding rule in their Methods (lower-is-better outcomes reversed so that positive means benefit) and the SUCRA rankings, which put that bath near the bottom, the same numbers would favour passive rest; but Table 3 labels both outcomes 'Lower = Better' and the muscle-damage Results say negative values mean lower concentrations, which supports the authors' reading.For soreness, use the base's earlier pooled evidence that cold water lowers CK and soreness at 24 h (REF-250); muscle-damage markers are markers in any case.
Endurance athletes: no performance benefit shown. Nothing beat passive rest for aerobic performance, and the endurance-recovery estimates for the recommended 9-12 °C, 10-15 min bath (−0.47, −0.97 to 0.04) do not support the authors' recommendation under their stated coding. Inference, not tested here: for a cyclist or runner, cold water stays a comfort or heat tool between race days (SYNTH-70), not a performance aid.
Injury or medical cold-water questions go to a clinician; these trials excluded injured athletes and anyone with a medical reason to avoid cold immersion.
Limitations
The authors': star-shaped networks with passive rest at the centre, so most comparisons are indirect and node-splitting was not possible; moderate heterogeneity in most primary networks (τ 0.32-0.51 by their Discussion, though Tables 5-6 give τ 0.065 for strength and 0.000 for resistance-training power and aerobic performance); time points from immediately after to 72 h pooled, so no advice on timing; few women; crossover trials analysed as independent groups because correlations were missing in 59%; standardised effects across different instruments; no long-term adaptation data.
Contradictions inside the paper (our reading): (1) the abstract says passive recovery ranked highest for resistance-training strength and the Discussion gives it a SUCRA of 91.5%, but Table 6a ranks it 8th (47.1%) with no option different from it; (2) the Methods reverse-code lower-is-better outcomes so that positive means benefit, yet Table 3 labels muscle damage and soreness 'Lower = Better' (not reversed) and the Results read negative muscle-damage, soreness and endurance-recovery values as benefits, and the text calls "EN_Exercise (g = 0.92)" a positive effect where Table 6d prints −0.92, so the direction of these three networks cannot be settled from the paper; (3) participants are 2,313 in the abstract and 2,847 in the Results; aerobic performance has 235 participants in the abstract and text but 283 in Table 3, and endurance recovery 17 studies and 463 participants against 19 and 497; (4) the moderate temperature band is 9-12 °C in the Methods and 11-15 °C in the Discussion; (5) the Discussion's statement that moderate-temperature protocols were "systematically worse" than passive rest for strength is not what Table 6a shows (moderate-medium 0.00).
Classification: several endurance sessions were assigned to the team-sport or resistance networks (an Ironman, cycling time trials, a 90-min ride, a downhill-running damage protocol), so the networks are not clean by sport.
Inclusion drift: two trials had mean ages under 18; one swimming trial is listed with 27 °C water, against a ≤ 20 °C criterion (it was dropped from the aerobic network as disconnected).
Not read: the risk-of-bias figure, network plots, search appendix and the supplementary data and code.
Funding from two provincial university research projects; no competing interests declared; no industry funding stated.
How this entry was checked and changed
Why this level
a systematic review of randomised trials whose risk of bias and GRADE certainty we read in the full text (2 of 87 trials low risk, 75 some concerns, 10 high; certainty moderate for strength after resistance exercise and low for every other network), capped at A- by the low certainty on most outcomes, star-shaped networks with 69-85% indirect comparisons, crossover trials analysed as independent groups, and reporting contradictions between the abstract, text and tables (Limitations)