A review of 19 skin trials found only a small effect on wrinkles

A review published in Frontiers in Medicine in March 2026 gathered the randomised trials of peptide preparations for aging skin, found 19 of them covering 1,341 adults, and added their wrinkle results into one figure. The figure is small, and the work of reading the paper is understanding what sits underneath it [1].

How the review was assembled

The search covered three databases and returned 2,479 records: 1,680 from MEDLINE, 527 from CENTRAL and 272 from Web of Science. Nineteen randomised controlled trials survived the screening. Such a trial assigns each participant by chance to the thing being tested or to a placebo, an inactive substitute, so the comparison is not decided by who volunteered for what. The review was registered in advance with PROSPERO, the public register for reviews of this kind, and its authors declare no funding and no commercial relationships [1].

What the pooled figures were

Across the trials that measured wrinkles, the pooled mean difference was 0.27, with a range of statistical uncertainty from 0.01 to 0.52 and a P value of 0.04. That P value means a difference this large would be expected by chance about four times in a hundred if the preparations did nothing at all. The bottom of the range, 0.01, sits barely above zero, which is another way of saying the result only just cleared the line [1].

Two other measurements came out larger: skin hydration at a pooled mean difference of 5.80 and skin brightness at 2.40, both at P below 0.01. Elasticity and skin density moved the same way but did not clear the threshold. No trial reported a serious side effect [1].

Why 0.27 has no unit

A mean difference is expressed in whatever units the original trials used, and here they did not all use the same ones. Some read wrinkle depth off an instrument. Some had graders score photographs. The review states that the methods behind those measurements were not consistently described, and that many trials gave no detail on which of the two produced the number. So 0.27 is an average across scales that share neither a zero point nor a step size [1].

Variation between the trials stayed high. The review treats an I squared value above 50 percent as substantial, and in one subgroup comparison it reached 96 percent. I squared estimates how much of the spread between studies comes from real differences rather than chance, so a figure that high describes trials that are not measuring the same thing [1].

The blinding problem

Pooling validity is further compromised by methodological weaknesses in the included RCTs, particularly inadequate blinding and inconsistent outcome assessment methods.

Oral and topical peptides for skin aging, Frontiers in Medicine, March 2026 [1]

Blinding is whether the participant and the person scoring the result know which group someone is in. Where the outcome is how skin looks, an assessor who knows can score generously without meaning to, and a participant who knows can report a change that is not on the photograph. The review rated several trials unclear on this point, one as high risk, and two as high risk for missing data [1].

Seventeen swallowed, two applied

The split inside the review matters more than the headline. Seventeen of the 19 trials tested a preparation taken by mouth. Two tested one applied to the skin, and those two covered 105 of the 1,341 participants. The pooled wrinkle effect came from the oral trials; the topical ones showed a smaller effect that did not clear the threshold. A great deal of marketing rests on creams. In this pooled record, creams rest on two studies [1].

One figure in the paper is worth checking against itself. The summary box at the top gives the oral subgroup a mean difference of 0.35 at P equals 0.05. The abstract and the results section give 1.5 at P equals 0.01 for what reads as the same comparison. Both are printed in the same paper, which is a reason to read the results section rather than the box above it [1].

Who ran the trials underneath

The review's own funding is clean. The studies it gathers are a separate question, and skin papers published since 2024 show the pattern. A 2024 study followed 135 women aged 45 to 65 for three months on a collagen peptide supplement, 116 of whom completed. Its expert grader wrinkle score fell from 5.9 to 5.0 at P below 0.0001, and 83.6 percent improved. It was open label, meaning everyone knew what they were taking, and it had no comparison group at all. Its five authors are employees of the company behind the product [2].

A 2025 trial of a preparation made from pig placenta was built more carefully: 90 participants aged 35 to 60, randomised, double blind and placebo controlled across 12 weeks, reporting reductions in wrinkle depth and in water loss through the skin. Two of its authors work in the research division of the company that supplies the ingredient [3].

A 2026 paper pairing a laboratory skin model with a clinical study of a swallowed combination is franker still. Four of its authors are employees of the company that paid for it, and its funding statement says that company was involved in the study design, in collecting, analysing and interpreting the data, in writing the article, and in the decision to submit it for publication [4].

What would settle it

None of that makes a result false, and a paper that states who paid and what they did is behaving better than one that does not. What it sets is how much independent confirmation a finding needs before it counts. The reviewers put their position plainly: the available data are not sufficient to support definitive conclusions about how well these preparations work [1].

Trials of creams, blinded on both sides, measured the same way in every study and run by people with nothing riding on the answer, would move this record. Nineteen randomised trials sounds like a closed question. Read one study at a time, it is not [1].