Protein blends vs whey: what a 2026 network meta-analysis says

Protein blends are sold on a formulation argument: a fast fraction and a slow fraction, complementary amino acid profiles, a better cost per gram of delivered protein. Until now that argument has rested on pairwise trials, each comparing one product against one comparator. A Bayesian network meta-analysis published in Frontiers in Nutrition in August 2026 put seven protein sources and placebo into a single connected network of 36 randomised controlled trials, and produced something the category has not had before: a rank for blends, with an interval around it.

Blends came fifth of seven. That is not a verdict that blends do not work. The blend node rests on three trials, and the authors say so repeatedly. But it is the number a formulator should have in front of them before putting 'advanced multi-source blend' on a pack, because it shows how little published evidence currently sits under that phrase. What follows is what the study measured, what it found, and where the estimate is soft.

What the network actually compared

The search ran to February 2026 across six databases and was pre-registered on PROSPERO (CRD420261340515). Eligible trials randomised adults to a protein supplement or placebo during supervised resistance training of at least six weeks. Thirty-six RCTs met the criteria: 1,552 participants, 89 trial arms, eight network nodes (whey, milk, casein, rice, blend, soy, pea, placebo).

The trials ran from 2000 to 2024. Supplemental protein doses ranged from 15 to 75 g/day, intervention length from 6 to 24 weeks (median 12), and age from 19 to 72 years. Lean body mass was the primary outcome, measured by DXA in 89% of studies. Every trial included a placebo or non-protein control arm, so the network was fully connected through placebo. Risk of bias was rated low in 15 trials (42%), with some concerns in 16 (44%) and high in 5 (14%); the recurring problem was blinding, because protein and placebo drinks differ in taste and texture.

The ranking, and where blends fall

Three sources produced a credible increase in lean body mass against placebo: whey (+1.24 kg), milk (+1.18 kg) and casein (+0.92 kg). Rice, blend, soy and pea all had positive point estimates whose 95% credible intervals crossed zero. The ranking is driven as much by how much evidence sits behind each node as by the point estimates themselves.

Two robustness checks matter more than the ranking itself. First, when the authors removed trial arms containing co-interventions (multi-ingredient formulas, creatine co-supplementation, added amino acids), whey, milk and casein barely moved, while the blend node fell from SUCRA 0.48 to 0.30 — two of its arms had carried co-supplements. Second, threshold analysis estimated that eight additional null trials would be needed to displace whey from first place, but only one or two to move rice or blend. A rank built on three trials is a rank that can change next year.

Prediction intervals were wide for everything. Even whey, the best-supported node, had a 95% predictive interval of −0.18 to 2.66 kg, meaning a future trial could plausibly report no effect. Between-study heterogeneity for lean mass was τ = 0.62 kg.

Where the dose curve flattens

Restricted cubic splines fitted to ten whey dose-arm data points (20–75 g/day) rose steeply from 15 to 30 g/day and flattened at approximately 30–35 g/day. Above that, the incremental lean-mass response diminished. The authors label this estimate exploratory, and with ten data points that label is appropriate, but it converges with the meta-regression: each extra gram per day was worth 0.024 kg of lean mass (95% CrI 0.004 to 0.044), or +0.24 kg per 10 g/day. Programme duration mattered too, at 0.035 kg per additional week (95% CrI 0.002 to 0.068).

In participants aged 65 and over, the fitted plateau shifted right, to roughly 35–40 g/day. That direction is consistent with the older-adult literature: a network meta-analysis of 235 trials in 20,980 older adults ranked whey plus resistance training first of 24 regimens for muscle mass. Casein and milk showed a broadly similar shape with more uncertainty, favouring about 25–35 g/day. Soy had a shallower slope, which the authors attribute to leucine density: about 7–8 g per 100 g of protein in soy isolate against about 13 g per 100 g in whey. One caveat runs through all of it: the model used supplemental dose as the predictor and could not adjust for habitual dietary protein intake, which the included trials did not report consistently.

What it means for a formulator

A blend is a specification, not a category. The network treated 'blend' as one node fed by three trials of different compositions. Nothing in this dataset supports or refutes any particular ratio of whey protein concentrate, milk protein isolate and soy isolate. What it does suggest is that the word on the pack carries no evidential weight of its own. The parameters that can be defended are the ones you can measure per lot: protein content by a stated method, the fraction split, and leucine per serving. Our LACTOPRIME™ protein systems are specified that way for exactly this reason — the ratio is a design input, not a marketing adjective.

Serving size is a dose decision, and most servings sit below the plateau. If the curve flattens at 30–35 g/day of supplemental protein and the label serving delivers 20 g taken once a day, the product is on the steep part of the curve, where the trials show the response still climbing. Designing to the plateau means either a larger single serving or a two-serving daily pattern — and in a blend, the leucine contribution of each fraction has to be calculated against that serving, not against 100 g of powder.

Declared protein depends on the nitrogen conversion factor. Dairy protein is conventionally calculated as nitrogen × 6.38, soy as nitrogen × 6.25, while EU labelling requires protein to be declared as total Kjeldahl nitrogen × 6.25 (Regulation (EU) No 1169/2011, Annex I). A multi-source blend specified internally on a dairy factor and declared on the labelling factor will not reconcile: an 80.0% figure at 6.38 becomes 78.4% at 6.25. The factor belongs in the purchase specification and on the certificate of analysis, next to the assay method. If you want to see how we write that, request the specification and a lot certificate.

One of the co-intervention categories the authors stripped out of the sensitivity analysis was creatine, which is often formulated alongside a protein blend. What creatine does to routine kidney markers — and why a raised serum creatinine reading is not by itself evidence of impaired filtration — is covered separately in creatine and serum creatinine: what the 2026 evidence shows. What that same co-intervention is worth on its own in women — a population absent from the 72% of trials here that recruited men only — is the subject of a 2026 multilevel meta-analysis of creatine in female athletes, which pooled 170 outcomes in 424 women to a small effect: Hedges g 0.23 (95% CI 0.09 to 0.37), at low certainty.

Where this evidence is thin

The authors are unusually direct about the weaknesses, and they are worth repeating. Several nodes rest on two or three trials, so their SUCRA values should not be read as settled. Transitivity is not established: whey was studied mainly in young trained men (mean age 34.1 years), while the casein and milk nodes carried more older and untrained participants (casein mean age 48.8 years), which can bias indirect comparisons. Blinding was imperfect across the network. Whey concentrate, isolate and hydrolysate were pooled into a single node despite differing in processing and purity. Seventy-two per cent of trials recruited men only, and just nine trials studied adults aged 65 or over, so the age subgroup results are hypothesis-generating. Full-text screening was restricted to English. The model-fit comparison favoured the consistency model by 2.8 DIC points, below the 5-point threshold the authors had pre-specified, so consistency is not strongly demonstrated. The record retrieved from PubMed carried no funding or conflict-of-interest statement, so we have not verified sponsorship.

The honest summary is that a +1.24 kg difference over a median of 12 weeks is real but modest, that the hierarchy among dairy sources is narrow, and that blends and plant proteins are under-studied rather than disproven.