A review of a hundred and nineteen supplement trials graded almost all of its own comparisons as very low certainty

What a certainty rating is

A certainty rating is a judgement about how far a pooled result can be trusted, made separately from the result itself. It takes account of how well the underlying trials were run, how precise the combined estimate is, how directly those trials answer the question being asked, and whether the direct and indirect evidence agree with each other. A large effect carrying a very low certainty rating is not yet a finding.

Drawn from background research methodology, not from this paper.

Why they ran it

The authors set out to compare supplements against each other rather than one at a time, and to report alongside each comparison how much confidence it deserves and whether it survives being tested from a different angle. The question they treat as unresolved is not which supplement is best but whether the evidence base can support a ranking at all.

Drawn from the stated aims of the paper.

Systematic review with four separate Bayesian random effects network meta-analyses, one each for peak power, relative peak power, vertical jump and countermovement jump. PubMed, Web of Science, Embase and the Cochrane Library, final search 8 September 2026, eligible trials published in English through 31 August 2026. Risk of bias with RoB 2, certainty of evidence with CINeMA.

A total of 119 studies were included. The peak power, relative peak power, vertical jump and countermovement jump networks comprised 47, 33, 31 and 44 studies respectively, with some studies contributing to more than one outcome.

Against placebo, beta hydroxy beta methylbutyrate produced a mean difference of 64.70 W, 95 percent credible interval 19.15 to 110.50, for peak power, and 0.84 W per kg, 95 percent credible interval 0.42 to 1.28, for relative peak power. Both were rated moderate certainty, and the relative peak power finding remained relatively stable across multiple sensitivity analyses.

Iron produced a mean difference of 6.81 cm, 95 percent credible interval 0.84 to 12.83, for vertical jump, supported by only one direct study. For countermovement jump, betaine produced 3.72 cm, 95 percent credible interval 1.24 to 6.18, and the credible intervals for palmitoylethanolamide, caffeine combined with Rhodiola rosea, creatine, multi-ingredient products and caffeine alone also excluded zero. Favourable effects were observed for beta alanine on peak power and for creatine and vitamin D on relative peak power.

Of 1,636 pairwise comparisons across the four networks, 17 were rated as moderate certainty, 55 as low and 1,564 as very low.

The numbers

Studies included	119 across four networks
HMB versus placebo, peak power	64.70 W, 95% CrI 19.15 to 110.50, moderate certainty
HMB versus placebo, relative peak power	0.84 W/kg, 95% CrI 0.42 to 1.28, moderate certainty
Betaine, countermovement jump	3.72 cm, 95% CrI 1.24 to 6.18
Iron, vertical jump	6.81 cm, 95% CrI 0.84 to 12.83, one direct study
Certainty across all comparisons	17 moderate, 55 low, 1,564 very low, of 1,636
Registration	PROSPERO CRD420261283835, retrospective

Why this might happen

Open question The authors do not claim to know why.

A review of this kind judges an evidence base rather than explaining a physiological effect, and the authors stop at a verdict on the trials underneath their own estimates.

Drawn from the stated conclusions of the paper.

What this does not show

This does not show that the supplements it names do not work. Most of the 1,636 comparisons are uninformative rather than negative, and the authors attribute that to how little direct evidence exists for each pairing. An uninformative comparison and a demonstrated absence of effect are different things, and this review contains almost entirely the first kind.
This does not establish a ranking anyone should act on. The authors state that differences between networks cannot establish that outcome type modifies supplementation effects, and that rankings alone should not be used to guide individualised supplementation regimens.
This is not a prospectively planned review. The authors record that the review was registered in PROSPERO retrospectively, so the four network structure and the outcome choices were not committed to before the data were seen.
This says nothing about endurance. All four outcome networks are measures of power or jump height. No endurance outcome appears anywhere in the review.

Where this leaves us

The question shifts from which supplement raises power to whether the trials have been run well enough for that question to have an answer, and for almost every comparison the answer is not yet.

Trials of dietary supplements in healthy physically active adults and athletes, measuring peak power, relative peak power, vertical jump or countermovement jump.

Prospectively registered trials of the handful of agents whose credible intervals excluded zero, sized to the effects actually claimed.

Caveats worth holding

Registration was retrospective, so the primary outcome and network structure were not pre-specified.
The iron result for vertical jump rests on a single direct study and should not be repeated as a finding.
Only the HMB comparisons survived sensitivity analysis. Other favourable comparisons either had prediction intervals including zero or were attenuated.
Exploratory associations with training level and sport were limited by sparse evidence and imbalanced covariate distributions, which the authors state.
No participant total is reported, so the weight behind each network cannot be judged from the paper alone.
All outcomes are power or jump measures. The review carries no endurance outcome.
No conflict of interest or funding statement is present in the indexed record. All listed affiliations are academic.