Creatine and serum creatinine: what the 2026 evidence shows

The question that reaches a technical team about creatine is rarely about performance. It is about a lab report. Someone taking the ingredient sees serum creatinine above the reference range, or an eGFR that has dropped since the last blood draw, and asks the brand what happened. Until this year the answer rested on physiology and a handful of small trials. Two meta-analyses published in 2026 now put pooled numbers on it.

Both are indexed in PubMed and both reach the same place: the marker moves, and the outcomes that do not depend on that marker do not. That distinction is the whole of this article. It is a measurement question, not a safety claim, and the difference matters for anyone who has to write a technical dossier that will be read by a customer's quality department.

Two 2026 meta-analyses, one consistent number

The larger of the two, by de Souza Almeida and colleagues in International Urology and Nephrology, pooled 26 randomised studies and 1,036 participants, drawn from both healthy individuals and people with chronic kidney disease, with databases searched to March 2025 (PROSPERO CRD420251015042). Its primary outcome was serum creatinine: a mean difference of +0.14 mg/dL (95% CI 0.05 to 0.22; P = .002), with I² = 93.9%.

The second, by Tsiaras and colleagues in the Journal of Renal Nutrition, included 19 randomised controlled trials plus one double-blind randomised crossover study. Serum creatinine rose by +0.13 mg/dL (95% CI 0.07 to 0.18, from 19 comparisons). Serum urea did not change (−0.60 mg/dL, 95% CI −2.15 to 0.96, n = 12), and eGFR did not reach significance (−5.20 mL/min/1.73 m², 95% CI −15.00 to 4.60, n = 8). Splitting trials at one month of supplementation changed none of the three (P = .15, .15 and .88).

Two teams, different cut-off dates, overlapping but not identical trial sets, and point estimates 0.01 mg/dL apart. That is about as close to replication as meta-analysis gets.

For context on why people take the ingredient at all, a separate 2026 three-level meta-analysis of 11 trials in adults aged 60 and over found that adding creatine to resistance training improved muscle strength by a small margin (g = 0.31, 95% CI 0.18 to 0.45, moderate certainty), while effects on muscle mass and physical function were not statistically significant. Small, and reported as small.

Why estimated filtration falls when measured filtration does not

Creatinine is the breakdown product of creatine. Supplementing creatine enlarges the body’s creatine pool; a roughly fixed fraction of that pool converts non-enzymatically to creatinine each day, so the steady-state serum concentration rises. CKD-EPI, MDRD and Cockcroft–Gault all read serum creatinine as a proxy for filtration. A higher creatinine is therefore reported as a lower filtration rate, whether or not anything changed at the glomerulus.

Almeida separated the two approaches. Estimated from creatinine, pooled GFR fell by 10.75 mL/min (95% CI −17.48 to −4.02; P = .002; I² = 0%). Measured with Cr-EDTA, a filtration tracer with no creatine in its lineage, the pooled difference was +5.89 mL/min (95% CI −0.30 to 12.08; P = .06; I² = 6.3%): the interval crosses zero and the point estimate runs in the opposite direction.

The outcomes that do not depend on creatinine behaved the same way. Serum urea, albuminuria, proteinuria and urinary creatinine showed no significant difference from placebo in Almeida, and Tsiaras found the same for urea. In the haemodialysis subgroup Almeida reported serum creatinine up and serum urea down. None of this is a safety conclusion about any individual, and the limits below are not decoration.

Two different creatinines, one word

There is a second confusion worth heading off, because it turns up in purchasing specifications. The word “creatinine” appears in two entirely different documents in this industry, and they measure different things.

Same molecule, different measurement. A certificate of analysis says nothing about anyone’s blood result, and a blood result says nothing about material quality. CREAPRIME™ specifies creatinine at ≤30 ppm by HPLC per USP <621>, tested lot by lot, because creatinine in a powder is the marker of thermal and hydrolytic degradation during manufacture and storage — a purity control, unrelated to the clinical analyte discussed above.

What it means for a formulator

Put the interference in the technical dossier, not the marketing copy. Brands field these questions from consumers and occasionally from physicians, and the defensible answer is descriptive: creatine intake raises serum creatinine, creatinine-based eGFR equations will therefore read lower, and directly measured filtration did not change in the pooled data. Stop there. Do not offer reassurance the evidence does not support, and do not advise on whether a particular person should take the ingredient.

Separate the trial doses from your label dose. The pooled trials span loading and maintenance protocols in populations from healthy adults to dialysis patients, and Tsiaras found no difference between short and long supplementation. A product delivering 3 g/day sits at the bottom of that range, and 3 g/day is also the intake the only authorised EU claim is tied to. Trials run on other endpoints sit higher: in postmenopausal women the pooled lean-mass and strength effects appeared only at 5 g/day or more alongside resistance training. The female-specific literature is wider still: the protocols pooled in a 2026 multilevel meta-analysis of creatine in female athletes ran from 2 g/day to 25 g/day loading, and the average effect across 170 outcomes was small (Hedges g = 0.23). If your dossier cites this literature, cite the dose range with it.

Keep the two creatinines apart in every document you ship. One is a purity control in ppm on a certificate of analysis; the other is a clinical analyte in mg/dL. Conflating them on a spec sheet is the kind of error a quality reviewer notices, and it costs more credibility than the shortcut saves. Impurity limits, test methods and CoA fields are worth agreeing before the first pallet ships — that is a conversation to have at specification stage, alongside the protein and sweetener systems in the same formulation. The same distance between what a trial used and what a label declares runs through the protein literature, where none of 235 pooled trials controlled for raw material identity. Creatine turns up there as a confounder to be removed: how protein blends rank against whey in a 2026 network meta-analysis rests on 36 randomised trials in which arms combining protein with creatine were treated as co-interventions and stripped out in a sensitivity analysis, leaving whey first for lean body mass at +1.24 kg versus placebo (95% CrI 0.61 to 1.87) and the blend node fifth of seven.

Limits of this evidence

The case above is coherent, but it is not strong evidence, and it should not be quoted as if it were.

• Heterogeneity for the primary outcome was extreme in Almeida (I² = 93.9%). The direction of the creatinine shift is consistent across analyses; the magnitude is not well estimated by any single pooled figure.
• The Cr-EDTA comparison carries the argument and is the weakest part of it. It rests on a smaller subset of studies and its interval crosses zero. Failing to detect a difference is not the same as showing there is none.
• Populations were pooled across healthy adults and people with chronic kidney disease, including a haemodialysis subgroup, under a single question. Those groups do not share a baseline.
• Neither analysis extends much beyond a year, and Tsiaras explicitly calls for trials longer than that. Long-term renal safety in supplemented populations is an open question and this article does not close it.

Both papers appeared in 2026 and have not yet accumulated independent critique. Almeida’s search closed in March 2025, so trials published since then are not in the pool. Anyone building a regulatory or technical file on this should read both papers in full rather than rely on a summary, including this one.