A review of 40 trials of skin supplements found effects too large to believe

The market for oral supplements sold to improve skin — collagen powders, polyphenol capsules, "beauty from within" drinks — has grown well ahead of the evidence supporting it. A systematic review published this month set out to test that evidence formally, and the paper is genuinely useful, though not in the way its abstract suggests [s1].

Its headline conclusion is that collagen, flavanols and other polyphenols alleviate skin photoaging while hyaluronic acid, lycopene and carotenoids do not [s1]. Its own tables show why that conclusion should be read with care.

What was reviewed

The authors searched PubMed, Embase, Web of Science and the Cochrane Central Register up to October 2024, retrieving 1,449 records and, after screening, including 40 English-language randomised controlled trials [s1][s2]. Of those 40, 22 contributed to meta-analyses across six supplement groups [s2]. The review is registered on PROSPERO as CRD42023472473 [s1].

Trials were excluded if they were not randomised, lacked skin outcomes, had insufficient data, were unavailable, were not in English, or involved participants older than 75 [s2].

The supplements studied were collagen (14 studies, 35%), flavanols (5), other polyphenols (5), carotenoids (5), hyaluronic acid (4), lycopene (4) and astaxanthin (3) [s2]. Participants were mostly women aged 20 to 75 with signs of photoaging such as sunspots and reduced elasticity, and most trials ran 4 to 12 weeks [s2].

Two outcomes carry most of the analysis: R2, a measure of overall skin elasticity, and minimum erythema dose, the ultraviolet exposure needed to produce visible reddening — a proxy for sun tolerance.

The positive findings, with their intervals

For collagen, eight studies (n=306) pooled to a standardised mean difference in R2 of 1.93 (95% CI 0.65 to 3.20; p=0.003) [s2]. Split by population, the Asian cohort gave 0.49 (95% CI 0.25 to 0.72; p<0.0001) with no heterogeneity (I²=0%), and the European cohort gave 4.38 (95% CI 1.24 to 7.51; p=0.006) with I²=98% [s2].

For flavanols and minimum erythema dose, four studies (n=167) gave a standardised mean difference of 3.31 (95% CI 0.75 to 5.87; p=0.01), with I²=97% [s2]. Flavanols did not improve skin elasticity: two studies (n=121) gave 0.34 (95% CI −0.10 to 0.78; p=0.13) [s2].

For other polyphenols, three studies (n=160) gave 1.88 for R2 (95% CI 0.17 to 3.59; p=0.03), with I²=95% [s2].

Why those numbers should not be taken at face value

A standardised mean difference expresses an effect in units of pooled standard deviation. Values around 0.2 are conventionally small, 0.5 moderate, 0.8 large. A value of 4.38 would mean the treated group sat more than four standard deviations from the control group — an effect of a magnitude almost never seen in real biology outside of things like antibiotics for bacterial infection.

The accompanying heterogeneity statistics say the same thing in a different language. I² of 97% or 98% means that almost all of the variation between study results reflects genuine differences between the studies rather than chance. Pooling results that disagree that violently produces a number, but not an interpretable one. The authors ran leave-one-out sensitivity analyses and could not identify the source of the heterogeneity in the collagen, flavanol or polyphenol analyses [s2].

The one estimate in the positive set that looks like a normal biological effect is the Asian collagen subgroup: 0.49, tight interval, I²=0% [s2]. That is the finding in this paper most likely to survive.

A comparator problem

For flavanols, the control condition in three of the four minimum erythema dose studies was a low-flavanol group rather than a placebo; only one used a placebo [s2]. The same pattern recurs in the carotenoid analysis, where one of two studies used a low-dose control [s2].

Comparing a high dose against a low dose of the same compound is a legitimate design, but it answers a different question from "does this work?" and the pooled results mix the two designs together.

The null findings

Lycopene: two studies (n=162 in the pooled analysis) gave a standardised mean difference in R2 of 0.14 (95% CI −0.31 to 0.59; p=0.53), I²=21% [s2].

Carotenoids: two studies (n=71) gave 0.12 (95% CI −0.54 to 0.78; p=0.72), I²=47% [s2].

Hyaluronic acid: two studies (n=100) gave 1.88 (95% CI −1.06 to 4.83; p=0.21), I²=97% [s2] — a large point estimate with an interval spanning zero, driven by two studies that disagreed sharply.

Astaxanthin appears in the review with three studies but no pooled analysis, and the authors state there is insufficient evidence to recommend it [s1].

Note that these null results rest on two studies each. Absence of a demonstrated effect in two small trials is not demonstration of absence.

The quality of the underlying trials

The review's own risk-of-bias assessment is the most quietly damning section. Only 19 of 40 trials (47%) described their randomisation process in enough detail to judge it, and 21 were rated unclear [s2]. Adequate allocation concealment was reported in 15 trials (38%), with two rated high risk [s2]. Performance bias was unclear in 55% of studies and detection bias unclear in 83%, with only five trials rated low risk on the latter [s2].

Eight studies (21%) were rated high risk for "other biases", with the authors naming potential conflicts of interest among the causes, alongside uncontrolled participant habits and unreported environmental conditions during measurement [s2]. Attrition and reporting bias were the two domains that came out well: 90% and 92% respectively [s2].

An internal inconsistency

The collagen section states that the meta-analysis covered 8 studies with 306 participants, comprising 5 studies in an Asian population (n=291) and 3 studies in a European population (n=312) [s2]. Those subgroup totals sum to 603, not 306. The paper does not reconcile the figures, and readers should treat the collagen participant count as unresolved rather than pick one.

Safety

On safety the review is more reassuring, within its limits. Across the supplement categories, the great majority of trials that addressed safety reported no adverse events, and the authors state that all the orally administered supplements were found to be safe over the study periods, typically 24 weeks or less [s1][s2]. For collagen, 12 of 14 studies addressed safety, 11 reporting no adverse events and one reporting an event without serious or treatment-related adverse events; the authors note that documentation of incidence and severity was incomplete [s2].

Short trials in healthy adults are a weak test of safety, and "no adverse events reported" is not the same as "no adverse events occurred."

What a reader can take from it

Three things. There is some randomised evidence that oral collagen improves a measured index of skin elasticity, most credibly in the Asian-population subgroup with a moderate effect size. There is essentially no support here for lycopene, carotenoids or hyaluronic acid, and none for astaxanthin. And the field's trials are small, short, frequently under-described, and in a fifth of cases carry acknowledged conflict-of-interest concerns [s2].

The authors' own framing is that the review is an initial step and that more focused studies are needed [s1]. That is the right conclusion, and it sits awkwardly with the way results of this kind get quoted on product pages.

This article is informational and is not medical or cosmetic advice.

Sources

• [s1] Effectiveness of dietary supplements for skin photoaging in healthy adults: a systematic review and meta-analysis of randomized controlled trials — Frontiers in Medicine, published online 21 July 2025. https://doi.org/10.3389/fmed.2025.1582946
• [s2] Same paper, open-access full text, forest plots and risk-of-bias assessment — Europe PMC, PMC12318760. https://europepmc.org/article/PMC/PMC12318760