Creatine, protein, omega-3: what 35 trials rank for what

Most sports-nutrition meta-analyses answer one question: does this ingredient beat placebo? That is a useful question for a scientist and a frustrating one for anyone building a product, because a brand launching a single SKU is not choosing between creatine and nothing. It is choosing between creatine, a protein system, an omega-3 oil and a flavour story, under a fixed cost per serving.

A network meta-analysis published in Nutrients in March 2026 is one of the few attempts to answer the second question. Wang, Qin and Kim pooled 35 randomised controlled trials in 1,211 trained athletes and placed protein, creatine and omega-3 in a single network, using placebo as the shared comparator, so that supplements never tested against each other could still be ranked. The finding is cleanly stated and worth reading carefully: the ranking changes completely depending on which outcome you measure.

What the network actually compared

Inclusion was narrow on purpose. Participants had to have trained in a structured programme for at least six months; studies in untrained populations, in people over 65 with no training history, and in metabolic disease were excluded. Dosing floors were set per ingredient: protein at ≥1.2 g/kg body weight per day for at least four weeks, creatine either as a 20 g/day load for 5–7 days followed by 3–5 g/day maintenance or as 3–5 g/day throughout, and omega-3 at ≥1 g/day combined EPA and DHA for at least two weeks. Multi-ingredient products were excluded unless the single-nutrient effect could be isolated.

Two details matter for anyone reading this as a sourcing document. First, every creatine trial in the network used creatine monohydrate — no other salt or form qualified. Second, the protein node was heterogeneous by construction: 10 whey trials, 3 casein, 1 amino acid. The 14 protein trials ran at 1.2 to 2.4 g/kg/day for a median of 8 weeks.

Participants were young (mean 24.6 years, range 18–35) and predominantly male (71.4%). Risk of bias was better than the field average: 23 of 35 trials at low risk, 9 with some concerns, 3 at high risk.

One ingredient, one outcome

For muscle strength, 28 trials and 942 participants in active arms produced a clear order. Creatine moved strength by 0.46 standard deviations against placebo (95% CI 0.29 to 0.63), protein by 0.33 (0.17 to 0.49), omega-3 by 0.16 (−0.06 to 0.38, not significant). In power-sport subgroups creatine's effect rose to 0.52 (0.31 to 0.73).

For endurance the order inverted. Protein was the only intervention with an interval excluding zero (0.28, 0.08 to 0.48), and the effect depended on duration: interventions longer than eight weeks gave 0.36 (0.12 to 0.60), shorter ones 0.18 (−0.08 to 0.44). Creatine was effectively flat at 0.05 (−0.26 to 0.36). That is a null result, and it is useful: it is a reason not to position a creatine ingredient against an endurance claim.

For recovery, omega-3 led at 0.40 (0.18 to 0.62), with a dose threshold — above 2 g/day combined EPA and DHA the effect was 0.48 (0.22 to 0.74); below it, 0.24 (−0.06 to 0.54), not significant. Creatine came second at 0.22 (0.01 to 0.43), an interval that only just clears zero. It should be read against a separate 2026 meta-analysis of eight randomised trials that found no effect of creatine on CRP or IL-6, acute or chronic. Whatever creatine is doing to soreness and creatine kinase, the inflammatory-marker evidence does not currently support an anti-inflammatory mechanism.

Why a ranking is not a result

The SUCRA percentages are the part most likely to be quoted out of context. Protein ranked 85.2% on endurance, omega-3 88.7% on recovery, creatine 82.4% on strength. Those are probabilities of being top-ranked, not effect sizes, and they are generated from a star-shaped network in which no trial compared two active supplements directly. Every active-vs-active number below is an indirect estimate.

Two of six exclude zero. Creatine is not statistically better than protein for strength, and omega-3 is not statistically better than creatine for recovery. The ranking order is real; the gaps between adjacent ranks mostly are not. The authors graded certainty as moderate for creatine-vs-placebo on strength and omega-3-vs-placebo on recovery, and low for every active-vs-active comparison, on grounds of imprecision.

What it means for a formulator

1. Build the SKU around one outcome, and let the specification follow. The data do not support a single product that claims strength, endurance and recovery with equal confidence. A strength-positioned product has a well-evidenced active at a defined dose; the engineering question becomes whether the labelled dose survives to end of shelf life, which is an assay and stability question, not an efficacy one. For a monohydrate like CREAPRIME™, that means a batch specification with an HPLC method that separates creatine from creatinine, not a single purity figure.

2. The protein node's breadth is a warning about your own blend. The network's protein arm pooled whey, casein and free amino acids at 1.2–2.4 g/kg/day. The pooled estimate therefore describes a category, not a formulation. If you are designing a multi-source system such as LACTOPRIME™, nothing in this analysis tells you which ratio to use — and the analytical trap is separate from the evidence one, because the Kjeldahl nitrogen-to-protein factor is 6.38 for dairy and 6.25 for soy, while EU nutrition declaration requires 6.25 (Reg. (EU) 1169/2011, Annex I). A blend specified with one factor will mis-declare.

3. Stacking is an assumption, not a finding. The review excluded multi-ingredient products precisely because their single-nutrient effects could not be isolated, and the authors list combined supplementation as future work. Adding omega-3 to a creatine SKU to borrow the recovery result is not supported by this dataset. It also multiplies the documentation: each added active brings its own specification, its own certificate of analysis and its own allergen statement.

4. Where this network stops, two narrower reads begin. The pooled sample was 71.4% male and could not be stratified by sex, so the strength estimate above says nothing about how the same active behaves in women: a separate meta-analysis of creatine in female athletes put the pooled effect across 21 trials in 424 women at Hedges g 0.23 (95% CI 0.09–0.37), at low certainty. And where this network collapses whey, casein and free amino acids into a single protein node, a network meta-analysis of protein sources takes them apart: whey ranked first for lean mass at +1.24 kg against placebo across 36 randomised trials, the blend node fifth of seven on three trials, with the dose response flattening at 30–35 g/day. Both are narrower questions than the one asked here, and both sit closer to the specification you have to write.

Where the analysis is thin

The review was not prospectively registered in PROSPERO; the authors state a protocol was written before analysis, which is weaker than a public record. The network is star-shaped, so every comparison between two supplements is indirect. The omega-3 node is the smallest: 7 trials and 211 participants overall, 5 trials for strength and 4 for endurance, which is why those confidence intervals are the widest in the figure. Heterogeneity was moderate throughout (I² 34–41%, τ² 0.06–0.09). Comparison-adjusted funnel plots showed mild asymmetry for strength (Egger's p = 0.08), which the authors describe as a possible small-study effect. The review also does not report a funding analysis of the 35 included trials, so industry sponsorship at the primary-study level is unaccounted for.

Generalisation is limited by design. Mean age 24.6 years, 71.4% male, and sex-stratified analysis was impossible because the included trials did not report sex-disaggregated outcomes. None of these results transfer to older or clinical populations, and the effect sizes are small to moderate in absolute terms — 0.46 is not a transformative number, it is a reliable one. A supplier claiming more than that from this paper is overreading it. If you want the batch documentation behind an ingredient rather than the marketing around it, request the specification and a lot certificate of analysis.