You’re caught up.

Home/Issues/Issue 07/Study

Systematic review and meta analysis with meta regression Publication date verified at source

Across sixty years of loading trials, the gram dose and the athlete's fitness predicted nothing

AI narration, generated on first listen
Journal
Frontiers in Physiology 16:1620943
Authors
Solem K, Clauss M, Jensen J
Institution
the Norwegian School of Sport Sciences
Published
18 August 2025
Source
PMID 40901614 · DOI 10.3389/fphys.2025.1620943
Design
Systematic review and random effects meta analysis of muscle glycogen supercompensation after a cycling or running bout followed by three to five days of high carbohydrate intake, with meta regressions on seven moderator variables run on the cycling studies only. Searches ran in PubMed and Web of Science in March 2025. The review protocol was not prospectively registered.
Sample
Thirty studies published between 1966 and 2020, 319 participants, 271 male and 48 female.

What a meta regression is

A meta analysis pools results from many studies into one number. A meta regression goes a step further and asks which features of those studies predicted how large each individual result was.

That is what makes this paper unusual here. It lets a protocol variable, such as how much carbohydrate was fed or how empty the muscle was before loading started, be tested against the whole literature at once rather than one trial at a time.

Drawn from background method, not from this paper.

Why they ran it

Many studies had confirmed that glycogen overshoots after exercise and a carbohydrate rich diet, but they used different exercise and different diets and reached very different sized answers, and the authors write that the mechanisms behind the elevated glycogen remain unclear.

Their aim was to put a single number on the size of the overshoot for cycling and for running, and then to run meta regressions to find which factors were influencing it.

Drawn from the paper's introduction.

Thirty studies published between 1966 and 2020 were included, covering 319 participants, of whom 271 were male and 48 were female. Muscle glycogen rose by 269.7 millimoles per kilogram of dry weight after cycling and by 156.5 after running.

Three moderators reached significance in the cycling studies. Carbohydrate as a percentage of total energy intake was positively associated with the overshoot and explained the largest share of the variation. Glycogen measured immediately after the depleting exercise was negatively associated with it, and so was glycogen before loading began. The last two are the same headroom story told twice: the emptier the muscle, the bigger the subsequent overshoot.

Three moderators came out flat. Carbohydrate intake in grams per kilogram per day was not associated with the overshoot. Neither was the amount of glycogen broken down during the exercise bout. Neither was maximal oxygen uptake, which explained none of the variation at all.

Heterogeneity was very high, above 90 percent for both cycling and running. Adding the two significant moderators to a single model brought it down from 92.4 percent to 66.1 percent, which is the clearest evidence in the paper that protocol differences are what the trials have been disagreeing about.

The numbers

Glycogen increase after cycling269.7 +/- 29.2 mmol/kg dw, 95% CI 212.4 to 327.0, p<0.001, I2 92.4%
Glycogen increase after running156.5 +/- 48.6 mmol/kg dw, 95% CI 61.3 to 251.7, p=0.001, I2 93.5%
Males294.3 +/- 32.0 mmol/kg dw, n=168
Females151.6 +/- 70.9 mmol/kg dw, 95% CI 12.8 to 290.5, p=0.032, n=32
Moderator: carbohydrate as percent of total energyestimate 15.25, 95% CI 9.86 to 20.65, p<0.001, R2 0.56
Moderator: glycogen immediately after exerciseestimate -2.25, 95% CI -3.42 to -1.09, p<0.001, R2 0.49
Moderator: basal glycogenestimate -0.80, 95% CI -1.42 to -0.18, p=0.011, R2 0.18
Moderator: carbohydrate in g/kg/dayp=0.177, R2 0.03, not significant
Moderator: glycogen broken down during exercisep=0.574, R2 0.00, not significant
Moderator: maximal oxygen uptakep=0.949, R2 0.00, not significant
Heterogeneity after adding the two significant moderatorsI2 fell from 92.4% to 66.1%, p<0.001

Why this might happen

Proposed by the authors This is the explanation the authors offer in their discussion. This study did not test it.

The authors read the gram dose null as a ceiling rather than an absence. Every study in the pool fed enough carbohydrate to saturate resynthesis, so there was no low dose left in the literature for the regression to detect, and they put the threshold at more than eight grams per kilogram per day being sufficient when no further exercise is done.

They read the depletion association as a matter of headroom. Muscle glycogen appears to have an upper limit, since the supercompensated values cluster far more tightly than the increases do, so a muscle that starts emptier has further to travel and shows a larger overshoot without necessarily ending up any fuller.

On the sex comparison they point at energy rather than biology: the one study that found supercompensation in men and not women had fed the women far less energy, and its own authors found no difference once energy intake was raised.

Drawn from Discussion, PMC12399638.

What this does not show

  • The outcome is glycogen, not performance. Not one of these thirty studies is pooled for a race time. This is the only quantitative meta analysis of loading protocols that exists, and it cannot say whether moving any of these knobs makes anyone faster.
  • The design cannot see the one day protocols. To be included a study had to contain a prior exercise session and had to measure glycogen after three to five days of high carbohydrate. That rules out, by construction, the very trials that tested whether the depletion phase and the extra days are needed at all.
  • The depletion finding is an association across studies, not a randomised comparison. Studies that happened to empty the muscle further happened to show a larger overshoot. Nobody in this analysis was randomly assigned to a depletion phase, and the authors' own statement that depletion is required rests on a one legged cycling design from the 1960s rather than on a head to head trial.
  • The studies disagree with each other more than chance explains. An I squared above 90 percent means the trials are not estimating one common effect. The authors say so and warn that the pooled numbers should be read with caution.
  • The female estimate rests on very few people. Forty eight women appear in the whole analysis and thirty two of them in the sex comparison, across five study groups. The authors report the overshoot in women as significant and immediately warn that the magnitude may not generalise.

Where this leaves us

Three procedural factors move muscle glycogen and three do not. How carbohydrate rich the diet is matters, and how much headroom the muscle has matters, which shows up twice as the glycogen left after the bout and the glycogen there before it. Grams per kilogram per day, the size of the depleting bout, and the athlete's maximal oxygen uptake all came out flat. That last one is the one worth sitting with, because the received idea is that better trained athletes have less to gain, and across sixty years of these trials aerobic fitness predicted none of the difference in how much glycogen was stored.

Thirty studies of cycling and running published between 1966 and 2020, with the moderator analysis run on the cycling studies alone.

The same analysis with performance as the outcome and the depletion phase, the loading duration and the dose coded as moderators. It does not exist, and the paper that examined the performance literature most closely argues the underlying evidence may not currently support one.

Caveats worth holding

  • Dry weight units throughout, roughly four times the wet weight numbers used in several of the other studies in this issue.
  • The meta regressions were run on cycling studies only. The authors judged the running data too sparse and too poorly reported to support them.
  • The number of study groups behind each moderator varies from 18 to 30, so the moderators are not all estimated on the same evidence.
  • The review protocol was not prospectively registered, and the search terms centred on the established vocabulary of supercompensation and carbohydrate loading, which the authors note may have missed studies using other words.
  • Egger's test found no funnel plot asymmetry for cycling and a non significant trend toward it for running.

Newsletter

Each issue by email, when the newsletter launches. Leaving your address puts you on the list, nothing is sent yet.