Do Polyphenol Supplements Actually Work?

Ten years ago the advice was to eat your berries. Now the same molecules are sold in capsules, positioned around "cellular health," "healthy ageing," NAD, senolytics and inflammation. The category has grown fast, and the labels have grown vague along with it.

Part of the problem is the word itself. "Polyphenol" is not a compound; it is a structural category covering thousands of very different plant molecules. Flavonoids alone split into flavonols (quercetin), flavan-3-ols (the epicatechin in cocoa, the EGCG in green tea), anthocyanins (the pigments in berries), flavanones (citrus) and isoflavones (soy, red clover). Then there are stilbenes such as resveratrol, the ellagitannins in pomegranate and walnuts, lignans, and a long list of phenolic acids. These differ in how well they are absorbed, what the body converts them into, and how much human evidence exists behind them — often by an order of magnitude.

So the word on the front of the bottle tells you almost nothing about the evidence behind what is inside it. The only useful question is compound by compound.

The dietary evidence is genuinely good. In the Danish Diet, Cancer and Health cohort, 56,048 adults were followed for up to 23 years, during which 14,083 died.^[2] Compared with the lowest fifth of flavonoid intake, those in the second fifth had a 12% lower risk of death from any cause (HR 0.88; 95% CI 0.85 to 0.91) and those in the third fifth a 17% lower risk (HR 0.83; 0.80 to 0.86). The same inverse pattern held for cardiovascular and cancer mortality.

Two details in that study matter more than the headline. The first is that the benefit plateaued: beyond roughly 500 mg of total flavonoids a day, hazard ratios stopped falling. The second is how ordinary 500 mg is. The authors point out that a cup of tea, an apple, an orange, 100 g of blueberries and 100 g of broccoli would together supply most of the flavonoid subclasses and more than 500 mg in total. The median intake in the cohort was already 494 mg a day.

Newer work points the same way. In 109,711 UK adults followed for a median of 10.5 years, a flavonoid-rich diet — a median of six servings a day of common flavonoid-rich foods — was associated with a 16% lower risk of chronic kidney disease (HR 0.84; 0.77 to 0.92).^[3]

The honest caveat is the one that always applies here. These are observational studies. People who ate more flavonoids in the Danish cohort also had lower BMI, were more physically active, were less likely to smoke, were better educated and better paid, ate more fibre and less processed and red meat. The authors adjusted extensively and ran a falsification endpoint — flavonoid intake showed no association with hospital visits for burns or foreign bodies (HR 1.01; 0.92 to 1.12), which is reassuring against gross confounding — but no amount of adjustment turns an association into a causal claim.

What it does support is eating the foods. Whether a capsule inherits that benefit is a separate question, and it has been tested.

Three things separate a polyphenol in food from a polyphenol in a capsule.

Absorption. A landmark review pooled 97 human bioavailability studies across 18 major polyphenols.^[4] After a 50 mg aglycone-equivalent dose, peak plasma concentrations of total metabolites ranged from zero to 4 µmol/L depending on the compound, and urinary recovery ranged from 0.3% to 43% of the dose. Gallic acid and isoflavones absorbed best; proanthocyanidins, galloylated tea catechins and anthocyanins absorbed worst. What circulates afterwards is usually not the molecule on the label but a sulfated, glucuronidated or microbially transformed derivative of it.

Your gut bacteria. Some polyphenols only become biologically interesting after gut microbes convert them — ellagitannins to urolithins, isoflavones to equol, resveratrol to lunularin or dihydroresveratrol. Not everyone's microbiome does this. Two people can take the identical capsule and end up with different chemistry in their blood, which is one reason trials in this field contradict each other.

Hormesis. Many polyphenols work less as direct antioxidants than as mild stressors that nudge the cell's own defence systems. That is a dose-response curve with a top to it. More is not linearly better, and at high enough doses some of these compounds become liver-toxic rather than protective — which is exactly what the safety data below show.

Cocoa flavanols — the best case, and its ceiling

Cocoa is where the supplement evidence is strongest. A Cochrane review pooled 35 trials (40 treatment comparisons, 1,804 mostly healthy adults) giving 30 to 1,218 mg of flavanols a day for two to 18 weeks.^[5] Flavanol-rich cocoa lowered systolic blood pressure by 1.76 mmHg (95% CI −3.09 to −0.43) and diastolic by 1.76 mmHg (−2.57 to −0.94), rated moderate-certainty. In the subgroup of people who already had hypertension, systolic pressure fell by about 4 mmHg; in people with normal blood pressure there was no significant effect at all.

Two caveats from the reviewers themselves. The result was heterogeneous in ways their prespecified subgroups could not explain, which is why they downgraded the evidence from high to moderate. And when they excluded trials with authors employed by the sponsoring industry, the effect size shrank — "indicating some reporting bias," in their words.

Then cocoa got the trial the rest of this category has never had. COSMOS randomised 21,442 older US adults — 12,666 women aged 65 and over, 8,776 men aged 60 and over — to 500 mg of cocoa flavanols a day (including 80 mg of epicatechin) or placebo, with NIH support and a median 3.6 years of follow-up.^[6] The primary endpoint, total cardiovascular events, was not met: HR 0.90 (0.78 to 1.02), P = 0.11.

Among the secondary endpoints, cardiovascular death was lower (HR 0.73; 0.54 to 0.98), a 27% relative reduction. Myocardial infarction (0.87; 0.66 to 1.16), stroke (0.91; 0.70 to 1.17), coronary revascularisation (0.95; 0.77 to 1.17) and all-cause mortality (0.89; 0.77 to 1.03) were all neutral. A per-protocol analysis censoring at non-adherence did favour the supplement on total events (0.85; 0.72 to 0.99). There were no safety concerns.

That result deserves to be read in both directions. It is the most encouraging hard-outcome signal the polyphenol supplement category has produced, and it is also a secondary endpoint inside a trial that was negative on its primary — the kind of finding that justifies another trial, not a marketing claim.

Resveratrol — a lot of meta-analyses, very little signal

Resveratrol has been studied heavily, which makes its umbrella review unusually informative. Researchers pooled 11 meta-analyses covering 29 outcomes in 1,476 people with type 2 diabetes, 17 meta-analyses covering 26 outcomes in 727 people with metabolic syndrome, and 10 meta-analyses covering 24 outcomes in 271 people with fatty liver disease.^[7]

Resveratrol did show benefits on scattered outcomes — blood pressure, lipids, glycaemic control, insulin resistance, waist circumference, inflammatory markers. But as the authors put it, "for almost all outcomes, the magnitude of the effect was trivial, the certainty of evidence was very low to low, or the number of trials was too few." The single exception was HbA1c in short trials under 12 weeks (mean difference −1.05%; −2.09 to −0.02; six trials; GRADE moderate) — clinically meaningful, but short-term, small and not replicated over longer durations. Their conclusion was unambiguous: current evidence does not support resveratrol supplementation for managing cardiometabolic risk.

Quercetin — one marker, barely

An umbrella review of quercetin pooled five meta-analyses covering 18 randomised trials plus five further trials.^[8] Quercetin did not affect diastolic blood pressure, lipid profile, inflammation, body composition, fasting glucose or insulin resistance. It did lower systolic blood pressure by 1.9 mmHg (−3.2 to −0.6) and fasting insulin modestly. Heterogeneity was very high (I² of 88% and 75% respectively) and the certainty of evidence ranged from very low to moderate.

The most interesting polyphenol result published this year is also the best teaching case in the field, because it shows two markers of the same disease moving in opposite directions at the same time.

The PolyPAUSE trial was a randomised, double-masked, placebo-controlled, two-period crossover study.^[1] Ninety mildly hypercholesterolaemic, non-medicated postmenopausal women aged 45 to 59 were recruited in Murcia and Madrid between September 2023 and January 2025; 78 completed. Each took three capsules a day delivering 133.2 mg of trans-resveratrol (from a stilbene-rich extract), 312.0 mg of ellagitannins and ellagic acid (from punicalagin-enriched pomegranate extract) and 166.3 mg of isoflavones (from red clover extract) for eight weeks, then four weeks of washout, then eight weeks of an identical-looking microcrystalline cellulose placebo. Everyone followed a nutritionist-supervised low-polyphenol diet throughout, and adherence was 100%.

Here is what happened to the standard lipid panel. Total cholesterol rose 7%, LDL cholesterol rose 9.5%, and triglycerides rose 16% — all with P < 0.001, in roughly 80% of completers, with moderate-to-large within-subject effect sizes (Cohen's dz 0.73, 0.61 and 0.63). The rises in total and LDL cholesterol were larger in older women. On the lipid panel alone, this blend looked actively unhelpful.

Then the researchers measured the particles rather than the cholesterol they carried. Apolipoprotein B — the count of atherogenic particles, which is what current dyslipidaemia guidelines treat as the primary driver of risk — fell by about 11%, a borderline result (P = 0.054, dz −0.54). The LDL-cholesterol-to-ApoB ratio rose significantly (P < 0.001), meaning each particle was carrying more cholesterol rather than there being more particles. And oxidised LDL fell by 39% (P < 0.001; dz −0.71), with the oxidised-LDL-to-ApoB ratio showing a very large reduction (dz −3.97). Myeloperoxidase activity, an enzyme that oxidises LDL in vivo, trended down without reaching significance (P = 0.08).

The gut microbiome mattered. Women who produce equol from isoflavones had a 2.3-fold reduction in oxidised LDL (65%), while equol non-producers had a non-significant 1.2-fold change. The best-responding metabotype cluster showed a 2.5-fold reduction, but with only 15 women in it — a caveat the authors state themselves.

Who ran it is worth saying plainly. The authors are based at CEBAS-CSIC — the Spanish National Research Council's food and health laboratory — together with IMDEA-Food and Virgen de la Arrixaca University Hospital. In a category where most trials of a specific commercial blend are run by the company selling it, this one was not, and that makes the result more credible, not less.

And here is what it does not show. Oxidised LDL, ApoB and the ratios are all surrogate markers. Not one cardiovascular event was prevented or counted in this trial, because it was neither designed nor sized to count any. It ran for eight weeks in 78 unmedicated postmenopausal Spanish women with mild hypercholesterolaemia. Because the three botanicals were given as a single blend, the study cannot say which of them did what. The 9.5% rise in LDL cholesterol is not something to wave away either: the reassurance rests on the ApoB result, and ApoB only reached P = 0.054. The authors are careful about this, calling the lipid rise "unexpected" and saying the mechanism was not addressed.

The right reading is that this is a well-run surrogate-endpoint finding that should generate an outcome trial. It is not evidence that a polyphenol blend prevents heart attacks, and nobody should sell it as though it were. If you want the background on why oxidised LDL and ApoB are not interchangeable with LDL cholesterol, our cardiovascular biomarker guide covers what each one measures. The same surrogate-versus-outcome problem shows up in our reviews of nattokinase and endothelial glycocalyx supplements.

The assumption that a plant extract is harmless because the plant is food does not survive contact with the data.

High-dose green tea extract and the liver. After reviewing interventional trials, the European Food Safety Authority concluded that EGCG taken as a food supplement at 800 mg a day or more produced a statistically significant rise in serum transaminases compared with controls.^[9] Traditionally prepared green tea infusions were considered safe at the intakes reported across EU member states — mean EGCG exposure of roughly 90 to 300 mg a day, up to around 866 mg in high consumers — although rare idiosyncratic liver injury has been reported even with tea. Supplements on the market span roughly 5 to 1,000 mg of EGCG per day. Drink the tea; be careful with the extract.

Turmeric and curcumin. The US Drug-Induced Liver Injury Network reported 10 adjudicated cases of turmeric-associated liver injury, all enrolled since 2011 and six of them since 2017.^[10] Eight were women, median age 56. The injury was hepatocellular in nine, with a latency of one to four months. Five patients were hospitalised and one died of acute liver failure. Seven of the ten carried the HLA-B*35:01 allele — an allele frequency of 0.450 against 0.056 to 0.069 in population controls — pointing to a genetic susceptibility nobody is screened for. Three of the seven products analysed also contained piperine, the black-pepper alkaloid added specifically to raise curcumin absorption.

Resveratrol, red clover and medication interactions. Resveratrol and several other polyphenols affect platelet function in laboratory work, which is reason enough to raise a concentrated stilbene supplement with your clinician if you take an anticoagulant or antiplatelet drug, or have surgery coming up. Red clover isoflavones are phytoestrogens; anyone with a hormone-sensitive condition or on endocrine therapy should take that decision to their own clinician rather than to a label.

What is actually in the capsule. Researchers analysed 20 resveratrol supplements sold on the Slovenian market by high-performance thin-layer chromatography.^[11] Ninety-five per cent contained an amount of resveratrol different from the declared value, and 55% contained more than declared. In 25% of products, the amount per unit exceeded the EU novel-food maximum of 150 mg a day of trans-resveratrol. Label errors ranged from typographical mistakes to misleading claims. It is worth noting that the PolyPAUSE trial dose, 133 mg a day, sits just under that EU ceiling — so a product that overshoots its label is not a trivial problem.

There is no single answer to "do polyphenol supplements work," because polyphenols are not one thing. But the pattern across compounds is consistent enough to be useful: the diet evidence is strong and the supplement evidence is modest, surrogate-heavy and compound-specific. Cocoa flavanols have the best randomised case and still missed their primary endpoint in the one trial large enough to test it. Resveratrol has been studied for two decades without accumulating a convincing effect. Quercetin moves one number slightly. And the most striking recent result — a 39% drop in oxidised LDL — came alongside a 9.5% rise in LDL cholesterol in a trial that could not tell us what either change meant for anyone's heart.

That is not a reason to dismiss the field. It is a reason to be specific. If you are taking a capsule that lists fifteen botanicals at unstated doses because the bottle says "longevity," the evidence is not behind you. If you are eating a diet that gets you past 500 mg of flavonoids a day, it largely is.