Effects of Low-Carbohydrate and Ketogenic Diets on Aerobic Performance in Trained Athletes: A Systematic Review and Meta-Analysis
This systematic review and meta-analysis pulled together trials of low-carbohydrate diets (≤130 g·day−1 or ≤25% of total energy) and ketogenic diets (<50 g·day−1 or <10% of total energy) and their effect on aerobic performance in trained athletes.
Five databases (PubMed, Web of Science, SCOPUS, SPORTDiscus and the Cochrane Central Register of Controlled Trials) yielded 33 aerobic-focused studies of trained athletes (≥6 months of structured training, aged 18–45 years). Study quality was rated with the Newcastle-Ottawa Scale.
Results are given as shares of studies: VO2max (maximal aerobic capacity) held steady in 50.0% of studies and rose in 11.1%; submaximal exercise economy was the most sensitive outcome, worsening in 50.0%; time to exhaustion was maintained in 69.2%; and fat oxidation went up in all 30 studies that measured it (+28% to +200%).
The abstract also describes a "1-week adaptation threshold": performance was impaired in studies that measured outcomes within ≤7 days, and, it says, was consistently maintained or improved in studies measuring after >1 week. The authors conclude in favour of periodized carbohydrate strategies. The full text does not support the threshold claim (see Limitations).
Participants
33 aerobic-focused studies in trained athletes (≥6 months structured training, age 18–45 years), per the abstract.
409 participants "based on 28 studies with available participant data"; 18 studies in endurance-trained cyclists, runners and triathletes, three in elite racewalkers, three in ultra-endurance athletes; among studies reporting sex, 85.8% of participants were male (full text, checked 2026-10-02: sections 3.1.1 and 6.6).
Protocol
Five databases plus preprint servers, searched through October 2025, English only; described as a "semantic search" that retrieved "the 500 papers most relevant" and left 190 records, 49 full texts and 33 included studies (full text, checked 2026-10-02: sections 2.1 and 2.3).
Data extraction by "a large language model", "followed by human verification" (full text, checked 2026-10-02: section 2.4).
Stated designs: 23 crossover or repeated-measures, 9 parallel-group, 1 "RCT" (full text, checked 2026-10-02: Table 1).
Stated analysis: outcomes coded as improved, maintained or impaired and tabulated; pooled Cohen's d "for continuous outcomes"; Egger, Begg, trim-and-fill; meta-regression on duration (full text, checked 2026-10-02: section 2.5).
Quality: Newcastle-Ottawa Scale, with scores ≥7 called high quality (full text, checked 2026-10-02: section 2.4).
Key Results
| Outcome (abstract) | What is reported |
|---|---|
| VO2max | preserved in 50.0% of studies; 11.1% documenting improvements |
| Submaximal exercise economy | 50.0% documenting impaired efficiency |
| Time to exhaustion | 69.2% maintaining performance |
| Fat oxidation | all 30 studies measuring it showed increases (+28% to +200%) |
| Timing | ≤7 days: performance impairment; >1 week: "maintained or improved performance" |
Every headline result is a count of study directions, not an effect size; the abstract contains no pooled estimate and no measure of heterogeneity.
The economy result rests on four studies (full text, checked 2026-10-02: Table 5): impaired in elite racewalkers on a 3-week diet (p < 0.001) and in 8 trained runners after 31 days of a ketogenic diet at intensities above 70% VO2max (p < 0.050); improved in 8 runners after a 5-day fat-adaptation plus carbohydrate-restoration protocol (p = 0.048); no difference in 9 runners.
Pooled effect sizes exist only as isolated sentences in the sensitivity and timing sections, with no per-study data, no forest plot of effect sizes and no I² (full text, checked 2026-10-02: sections 4.5 and 4.6): fat oxidation d = 1.82 (1.71 without the largest study); exercise economy d = −0.38 in the studies rated high quality; time to exhaustion d = −0.06; "acute phase" performance d = −0.36 (95% CI: −0.58 to −0.14).None of them can be traced to the studies in the tables, and the subgroup counts beside them do not match the tables (economy "impaired in 3 of 4" high-quality plus "1 of 1" moderate-quality studies, against two impaired of four in Table 5).
The "1-week adaptation threshold" comes from eight sources (full text, checked 2026-10-02: Table 10 and section 4.2.4): two at ≤7 days (Terink, 2 days; Melby, 5 days) and six at >1 week (Burke, 3 weeks; Shaw, 31 days; Phinney, 4 weeks; Prins, 6 weeks; McSwiney, 12 weeks; Volek, habitual diet for 9–36 months).
What those eight found, by the paper's own tables (full text, checked 2026-10-02: Tables 5, 6 and 9): the 3-week racewalker study is recorded as "2.3% slower" over 10,000 m with impaired walking economy, and the 31-day study as "Impaired >70% VO2max" — both sit on the ">1 week, maintained or improved" side of the threshold.
Fat-oxidation examples given in the text (full text, checked 2026-10-02: section 4.2.1 and Table 7): 0.6 to 1.3 g/min in elite race walkers after 3 weeks; a 2.3-fold higher rate in habitually low-carbohydrate ultra-endurance runners.
Practical Application
"Will a low-carbohydrate or ketogenic diet hurt my endurance performance, and does it pass after a week?" This review of 33 studies in trained athletes reports — as shares of studies, not pooled effects — that all 30 studies measuring fat oxidation showed increases (+28% to +200%), that 50.0% of the studies measuring submaximal exercise economy documented impaired efficiency, and that VO2max was preserved in 50.0% of studies; its claim that performance is maintained or improved after more than 1 week is not supported by its own tables. Fat burning goes up reliably; what it costs at race intensity is the open part, and this paper does not settle it.
The two things worth keeping agree with the controlled trial the base already holds (REF-486): more fat oxidation, and a worse oxygen cost at race-relevant intensity, with no performance gain from a 3-week ketogenic block in elite race walkers.
Do not tell an athlete "the first week is hard, then you are adapted". Two of the controlled diet studies on the "more than 1 week" side of the review's own timing table (3 weeks and 31 days) are the ones it records as impaired.
Inference, not tested here: a higher fat-oxidation rate is a change in fuel mix, not evidence of a faster athlete; no entry in the base shows a performance gain from it at race intensity.
Limitations
Vote-counting presented as a meta-analysis. The outcome tables (Tables 4–10) list each study's direction and p-value and end in percentages of studies; the figures called forest plots are grouped by outcome category. Counting studies ignores sample size and effect size: a trial of 5 cyclists weighs the same as one of 28 race walkers (full text, checked 2026-10-02: Tables 4–10, Figures 1–6).
The wrong quality tool. The Newcastle-Ottawa Scale is written for cohort and case-control studies; the review applies it to crossover and parallel dietary trials and reports "adequate case definitions" in 30 studies. Randomisation, allocation concealment, carry-over and washout — the risks that matter in these designs — are not rated. The mean score is given as 7.2 ± 1.1 in the Methods and 7.4 ± 0.9 in the Results (full text, checked 2026-10-02: sections 2.4 and 4.4).
The designs are not what the abstract and conclusion say. The abstract describes athletes "randomized" to diets and the Conclusions speak of "33 randomized controlled trials and crossover studies".The tables include two rows labelled "Cross-sectional" (athletes on a habitual diet for 9–36 months) and a 12-week study whose own abstract says athletes "self-selected" their diet, listed here as a 47-participant crossover lasting "12 days" in one table and "12 weeks" in another (full text, checked 2026-10-02: Tables 4, 6 and 9; source abstract via Europe PMC).
Items counted as studies that are not studies of a low-carbohydrate diet. Checked against Crossref and Europe PMC: reference 22 (Bock & Kruse 2017) and reference 44 (Melby) are two-page commentaries in The Journal of Physiology; references 11, 17, 35 and 43 are single-page conference abstracts, which the review's own criteria exclude; reference 38 is a narrative review, entered as a second "cross-sectional" study of 20 ultra-endurance athletes; reference 37, entered in the VO2max table as a 4-week crossover in 13 cyclists, is the IOC consensus statement on REDs; reference 41, entered as a low-carbohydrate "5 h time trial" study in 13 cyclists, is an EEG study of neural efficiency with no diet arm; reference 34 is a trial of periodic carbohydrate restriction on three days a week, not a low-carbohydrate diet.Every DOI checked resolves to a real paper; several do not match what the tables say they tested (full text, checked 2026-10-02: Tables 4, 6, 7 and the reference list).
The adaptation-threshold claim fails on its own evidence. Its ≤7-day side is two sources — a conference abstract (2 days) and a commentary (5 days, sample size "NR"). Its >1-week side of six includes the two studies the review itself records as impaired (3 weeks: "2.3% slower"; 31 days: "Impaired >70% VO2max") and one cross-sectional comparison.Elsewhere in the same tables, short protocols did not show impairment: a 5-day fat-adaptation protocol improved running economy (p = 0.048) and a fat-adaptation study found no difference in a 1 h time trial (p = 0.110).The text also says both "all studies with adaptation periods exceeding 1 week (n = 6)" maintained or improved and "five of six" did (full text, checked 2026-10-02: Tables 5, 6, 9 and 10; sections 4.2.3, 4.2.4 and 5.1).
Counts in text and tables do not agree. Section 3.1.2 says "32 studies (96.9%) implemented short-term interventions of 7 days or less" and only one lasted longer; the outcome tables list durations of 3 weeks to 24 weeks for most rows, and section 4.5 analyses 20 "chronic (≥4 weeks)" studies.VO2max is "maintained" in 9 of 18 studies (50.0%) in the Results and abstract, in 13 in the Figure 1 caption, and in 83.3% in the Conclusions. Time to exhaustion is "Impaired (n = 2, 15.4%); Maintained (n = 9, 69.2%); Improved (n = 3, 23.1%)" — fourteen results for thirteen studies — and 7.7% impaired in the Discussion.Fat oxidation rose "+28% to +200%" in the abstract and "+34% to +230%" in section 4.3; Table 3 says fat oxidation was measured in 11 studies, Table 7 lists 30 rows, with one 1983 study entered twice (once dated 1982). Table 2 finds no female-only study; section 6.6 counts two (full text, checked 2026-10-02: sections 3.1.2, 4.1, 4.3, 4.5, 6.1, 6.6; Tables 2, 3, 4, 6, 7).
Registration and search. The protocol is called "prospectively registered"; the PROSPERO number given begins CRD42026, while the search ran through October 2025. The search is described as retrieving "the 500 papers most relevant", which is not a reproducible database search (full text, checked 2026-10-02: section 2.1).
No GRADE or other certainty rating. Heterogeneity is described in words ("variation in p-values") rather than measured.
Funding: none; conflicts of interest: none declared. One of the authors' own trials is discussed but is not among the pooled studies (full text, checked 2026-10-02: back matter; section 4.1.1).
Almost all participants are men; the review itself says the findings need testing in women.
How this entry was checked and changed
Why this level
lowered 2026-10-02 from the nominal A- of a systematic review after reading the full text: no pooled estimate in the abstract or in any outcome table (vote-counting), a quality tool built for observational studies, a "study" list that includes commentaries, conference abstracts, a narrative review, a consensus statement and an unrelated EEG study, and counts that contradict each other between text and tables