Whey protein for sarcopenia: 235 trials, 20,980 older adults

Most protein-source comparisons in sarcopenia research are small and inconclusive. Whey beats soy in one trial and ties it in the next. For anyone deciding which protein system to build a product around, that literature has been close to useless: unstable estimates, inconsistent comparators.

A network meta-analysis published in Nutrients in April 2026 is the largest attempt so far to fix that. It pools 235 randomised controlled trials and 20,980 participants, and it compares 10 protein sources against three exercise modalities in a single network — 24 treatment arms in total. It is worth reading closely, and it is also worth reading sceptically, because the headline ranking and the underlying effect sizes point in slightly different directions.

How the analysis was built

The authors searched six databases through December 2025 and included randomised trials of any protein supplementation combined with resistance training (RET), aerobic training (AET) or multicomponent training (MET) in untrained older adults. Outcomes were lean mass, handgrip and leg strength, and mobility measures: gait speed, chair rise, timed up-and-go, and the SPPB score.

Effects were expressed as standardised mean differences (SMD) with 95% confidence intervals, pooled with frequentist random-effects models. Regimens were ranked by SUCRA, a 0–1 score for how often a treatment finishes near the top of the network. Certainty was graded with GRADE, risk of bias with Cochrane RoB2.

The protein arms were not evenly populated. Whey carried 111 trials, milk protein 56, protein-enriched meals 50, soy 28, casein 10, meat 9, collagen 6, rice 4, and insect protein and oat one each. That imbalance matters for how much weight any single ranking deserves.

Where whey ranked, and by how much

For muscle mass, whey plus resistance training was ranked first of the 24 regimens (SMD 1.29, SUCRA 0.93), followed by collagen plus resistance training (SMD 1.23, SUCRA 0.88) and meat plus resistance training (SMD 1.16, SUCRA 0.87). Whey plus resistance training held the top rank across short-term (SMD 1.24), medium-term (SMD 1.26) and long-term (SMD 0.81) follow-up.

For leg strength, whey plus resistance training and whey plus multicomponent training tied at the top (both SUCRA 0.86), with soy plus resistance training immediately behind (SUCRA 0.85). For handgrip strength the order flipped: soy plus resistance training ranked first (SUCRA 0.86), whey second (0.85). For mobility outcomes — chair rise, timed up-and-go, SPPB — whey plus multicomponent training ranked first, and those were the outcomes GRADE rated as high certainty.

The gap between first and third place in the muscle-mass ranking is 0.13 SMD. On a network with I² = 78.6%, that is not a margin anyone should treat as a product decision.

Why the pairwise numbers are smaller

A network ranking answers a different question from a pairwise meta-analysis. The ranking asks which regimen finishes near the top against a common reference — here routine care, which in many trials meant no supplement and no training. The pairwise estimate asks what supplementation adds on top of the training the control group was already doing. The second number is always smaller, and it is the one that maps onto a product.

Three other recent reviews make that visible. A 2025 network meta-analysis of 19 trials in healthy older adults found protein plus resistance training improved strength by SMD 0.45 (95% CI 0.20–0.69) over resistance training alone; in the same analysis creatine monohydrate showed no significant strength effect (SMD 0.03, 95% CI −0.35 to 0.42) but the largest muscle-mass effect (MD 2.18, 95% CI 0.92–3.44). A 2025 pairwise review of 25 trials and 1,454 adults aged 64–84 found whey increased lower-body strength by SMD 0.16 (95% CI 0.04–0.28) and changed nothing in body composition, upper-body strength or lipid profile — while raising fasting insulin and HOMA-IR. A 2026 review of 14 trials and 763 women across reproductive stages found no effect on skeletal muscle mass (g = 0.065, 95% CI −0.353 to 0.482) and modest gains in bench press (g = 0.279) and handgrip (g = 0.412). The creatine literature in the same population narrows the same way: pooled across seven trials, lean-mass and strength gains in postmenopausal women appeared only at 5 g/day or more with resistance training.

In young trained and untrained adults the picture is the same shape. A 2025 systematic review of 12 trials found no significant effect of whey or soy on lean body mass, but whey improved bench press by 8.87 kg (95% CI 5.95–11.79) and squat by 9.60 kg (95% CI 5.61–13.60).

Dose and origin: the two variables that moved

The meta-regression is the most transferable part of this paper, because it describes intervention design rather than ranking. Three intervention variables were associated with larger effects.

Dose: a higher protein dose predicted greater lean mass gain (β = 0.28, 95% CrI 0.001–0.57) and greater handgrip gain (β = 0.36, 95% CrI 0.09–0.64). Origin: animal-derived protein predicted greater lean mass gain than plant-derived (β = 0.27, 95% CrI 0.03–0.50). Duration: longer treatment (β = 0.24) and longer follow-up (β = 0.22) predicted greater handgrip gain. Participant factors ran the other way — older age and female sex predicted smaller lean mass gains, and higher BMI predicted smaller handgrip gains.

Two details matter for a serving protocol. About 43% of trials (101 RCTs) dosed around training sessions at a mean of 25 g; the rest dosed daily at a mean of 28.5 g, or used protein-enriched meals at 0.8–1.8 g per kg body weight. And whey carried a higher rate of minor adverse events — OR 2.18 alone, OR 1.73 with resistance training — almost all gastrointestinal: nausea, diarrhoea, bloating, reflux. Withdrawal was also higher in the whey arms.

What it means for a formulator

Three practical consequences, in order of how much they should change a specification.

1. Dose per serving is the variable with the clearest support. The dose–response coefficients here are small but consistent, and the modal trial dose is 25–28.5 g of protein. A 20 g serving sits below the centre of the evidence base; a 30 g serving sits inside it. If a format constrains you below 25 g, the honest position is that the product is below the range where these trials found their effects, not that the effect scales down proportionally. Nothing in this analysis supports extrapolating downward.

2. Source substitution is defensible; source superiority is not. The animal-protein coefficient is real but small (β = 0.27), and the soy arm outranked whey on handgrip. A blended or plant system built to an equivalent protein dose and a comparable leucine profile is a supportable formulation choice on this evidence — whereas a marketing claim that one source produces materially more muscle than another is not, given a 0.13 SMD gap between the first and third ranked regimens. How protein blends rank against whey in a 2026 network meta-analysis sets the boundary on that argument: across 36 randomised trials, 1,552 participants and 89 arms in 8 nodes, whey ranked first for lean body mass (+1.24 kg versus placebo, 95% CrI 0.61 to 1.87) while the blend node ranked fifth of seven on three trials (SUCRA 0.48; +0.62 kg, 95% CrI −0.34 to 1.58). Our LACTOPRIME™ protein systems exist as WPC, MPI and soy isolate specifically because that substitution question is a formulation decision, not a settled one.

3. Tolerability is a formulation problem, not a footnote. An OR of 2.18 for minor gastrointestinal events in the whey arms, with higher withdrawal, is the kind of signal that shows up as repeat-purchase failure long before it shows up as a complaint. Lactose load, protein concentration, and the free amino acid fraction are all levers here. If you are reformulating for tolerability, the comparison you need is between finished systems at matched protein content, on your own subjects — not between raw material datasheets.

One thing this literature cannot tell you: none of these trials controlled for raw material identity. Protein content, amino acid profile and contaminant load were assumed from the label. If you are benchmarking a system against published trial doses, your own lot-level analysis is the only thing connecting the two. Every LACTOPRIME™ lot ships with its certificate of analysis, and samples for a formulation trial can be requested here. Lot-level identity is not a protein-only problem: on the creatine side a single word covers both a ppm impurity limit and a clinical blood analyte.

Limits of this evidence

The authors are explicit, and their caveats deserve to be reproduced rather than buried. Only 39 of 235 trials (16.6%) were rated low risk of bias; 70 (29.8%) were rated high risk. Global heterogeneity was substantial for every primary outcome: I² = 78.6% for muscle mass, 76.8% for handgrip, 70.1% for leg strength. Node-splitting found significant inconsistency between direct and indirect estimates in several arms, including whey plus aerobic training and soy plus resistance training.

Publication bias was detected for muscle mass and leg strength, which is why GRADE certainty for muscle mass came out low to moderate rather than high. Funding was a further concern: 29 trials (12.3%) reported a financial source with no conflict-of-interest disclosure, and 30 (12.8%) reported neither.

Finally, the comparator. Some trials used a placebo supplement and some used nothing, and placebo effects were not separated out of the routine-care reference — which the authors note may bias every arm. Their own conclusion is that the findings "are based on uncertain results and should be cautiously interpreted." That sentence belongs in any commercial summary of this paper, including this one.