Questioning Claims of Collagen Superiority in a Protein‐Supplement Network Meta‐Analysis

SP has received grant or contract funding (paid to McMaster University) from the Canadian Institutes of Health Research, the National Science and Engineering Research Council of Canada, the US National Institutes of Health, Nestlé Health Sciences, FrieslandCampina, the US National Dairy Council, and Dairy Farmers of Canada and Cargill. SP has received travel expenses and speaking honoraria from Nestlé Health Sciences, Optimum Nutrition, Nutricia, and Danone. SP is an advisory board member for WndrHlth and LiquidIV. SP holds patents licensed to Exerkine Inc. but reports no financial gains from these patents or otherwise. Dear Editor, Drummond et al. recently published a network meta-analysis (NMA) concluding that collagen is the “most effective” protein supplement for increasing strength and fat-free mass (FFM) during resistance training and that collagen is “superior” to whey for both outcomes [1]. I highlight several methodological and interpretive problems that, in my view, materially undermine the claim of collagen superiority. 1) “Most effective” and “superior” are not justified by NMA ranking. The authors conflate ranking with proven superiority. SUCRA values summarize the probability that an intervention occupies a given rank, but SUCRA is not a statistical test of superiority and is unstable when evidence is sparse or indirect [7, 8]. A treatment can “rank” first while remaining statistically indistinguishable from key comparators. Yet Drummond et al. state in the abstract and conclusion that “collagen shows a superior effect compared to whey protein for both outcomes” [1]. This statement is not supported for the primary outcome (strength): their own league table shows the collagen–whey contrast is not statistically significant (SMD 0.26, 95% CI −0.06 to 0.59) [1]. The NMA therefore does not demonstrate that collagen improves strength more than whey; at most, it suggests a higher point estimate and a higher rank probability. Presenting this as a claim of superiority overstates what NMA can legitimately infer. 2) The collagen evidence base is small, and the FFM base is extremely small. The authors’ network plot indicates that only 41 participants contribute to the collagen estimate for FFM [1]. From the supporting tables, this appears to reflect essentially two small trials (Kirmse et al. and Oertzen-Hagemann et al.) using 15 g/day of “specific collagen peptides” versus a noncaloric placebo over 12 weeks [7]. Claims that collagen is “the most effective” supplement for FFM in healthy adults are therefore driven by a narrow evidence base and are highly vulnerable to small-study effects, product-specific effects, and publication/reporting biases. While the authors note CINeMA downgrades due to within-study bias and imprecision [1], the take-home conclusion does not reflect this uncertainty. 3) Transitivity/clinical exchangeability is not credible for the collagen node as assembled. Balshaw et al.: collagen 15 g/day vs placebo, ∼15 weeks, young adults [9]. Kirmse et al.: collagen 15 g/day vs noncaloric placebo, 12 weeks, recreationally active young men [6]. Oertzen-Hagemann et al.: collagen 15 g/day vs noncaloric placebo, 12 weeks, recreationally active young men [4]. Rindom et al.: whey vs collagen (25 g doses) in a controlled-diet crossover focused on short-term recovery after an intense training microcycle, not chronic adaptation, and concluding recovery was “not markedly influenced by the type of protein supplement” [3]. Kuwaba et al.: 33-day crossover in middle-aged men “unfamiliar with exercise” with a single squat-bout load (soreness/fatigue primary endpoints), not a resistance-training program [5]. Including acute recovery trials and single-bout protocols within a network framed as “healthy adults undergoing resistance training” [1] undermines transitivity because “strength” does not measure the same construct (training adaptation vs recovery), and effect modifiers (training status, diet control, exercise dose, and placebo type) are not balanced across nodes. Under these conditions, rank hierarchies are particularly prone to artifactual “winners.” 4) The FFM endpoint is vulnerable to measurement artifact and misinterpretation in collagen trials. The two collagen trials [4, 6] that appear to drive the FFM finding rely on bioelectrical impedance analysis (BIA) rather than DXA [6]. BIA-derived FFM is sensitive to hydration and glycogen/water shifts, which are changes that can occur with training and supplementation, independent of muscle hypertrophy. Critically, Kirmse et al. report greater FFM with collagen than with placebo but no between-group difference in muscle fiber cross-sectional area, and explicitly interpret the discrepancy as likely reflecting “enhanced passive connective tissue adaptations” rather than contractile hypertrophy [6]. Thus, interpreting collagen as the “most effective” supplement for hypertrophy (or implying a muscle-specific advantage) is not aligned with the mechanistic and measurement limitations of the trials that generated the signal. 5) Search/selection issues undermine confidence, especially given the reliance on citation searching. Drummond et al. report database searches through May 2024, followed by the addition of 35 studies via citation searching/reference list screening (an extraordinary number in itself) [1]. However, a directly relevant randomized, double-blind head-to-head trial was somehow missed: Jacinto et al. (2022) [2]. This study provides rare comparative evidence and reports greater hypertrophy (muscle thickness) with whey than with (leucine-content-matched) collagen peptides, while performance outcomes were broadly similar [2]. Even if not all endpoints align with the NMA’s FFM outcome, the trial is clearly relevant to the collagen-vs-whey question and would strengthen and potentially shift the direct/indirect evidence structure for the primary strength network. Its omission is difficult to reconcile with the published PubMed strategy in the supporting information, which includes collagen terms plus resistance training/strength/hypertrophy terms with an RCT filter [1]. If Jacinto et al. were excluded due to a prespecified reason (e.g., free leucine co-supplementation), that exclusion should be explicitly documented and justified; otherwise, the unexplained absence raises the possibility of selection bias that favors collagen by excluding direct contrary evidence. Given these concerns, I respectfully request that the authors consider the following: independent statistical review of the NMA; a corrected, fully reproducible search plus a list of excluded full texts with explicit reasons; sensitivity analyses restricted to chronic resistance-training interventions (excluding single-bout and recovery-only designs), DXA-only FFM, and separation of collagen peptides from collagen-plus-cointervention formulations; and revising the conclusion to match what the evidence supports: collagen and whey may show benefits versus placebo in some contexts, but collagen superiority over whey for strength is not demonstrated, and the apparent FFM “advantage” rests on sparse, indirect, and nonmuscle-specific evidence. If these issues cannot be resolved transparently via correction, an expression of concern, or retraction if errors are irreparable, should be considered to prevent ongoing dissemination of a misleading “collagen is most effective” claim. No funding was received for this manuscript. Stuart M. Phillips has received grant or contract funding (paid to McMaster University) from the Canadian Institutes of Health Research, the National Science and Engineering Research Council of Canada, the US National Institutes of Health, Nestlé Health Sciences, FrieslandCampina, the US National Dairy Council, and Dairy Farmers of Canada and Cargill. Stuart M. Phillips has received travel expenses and speaking honoraria from Nestlé Health Sciences, Optimum Nutrition, Nutricia, and Danone. Stuart M. Phillips is an advisory board member for WndrHlth and LiquidIV. Stuart M. Phillips holds patents licensed to Exerkine Inc. but reports no financial gains from these patents or otherwise. Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.