Is Erythritol Bad For You? A Sweetener Worth Watching

This is one in the sweetener aisle we would not wave away.

Not because it has been shown to hurt anyone — it has not. But because this one, like the xylitol scare the same lab raised the following year [3], arrived with a mechanism attached, and a mechanism is a different kind of problem from a correlation.

It also arrived with a confound big enough to swallow the headline. The paper raised it in its discussion rather than its abstract [1], and the Cleveland Clinic’s press release gave it one sentence: the body makes “low amounts” of erythritol naturally, “so any additional consumption can accumulate” [15]. That turns the confound into a reason to worry about the food.

What the 2023 study found

Researchers at the Cleveland Clinic ran metabolomics — a broad sweep of the small molecules circulating in blood — on patients being assessed for heart problems, and asked which ones predicted heart attacks, strokes and death over the following three years [1]. For scale: of those first 1,157 patients, 116 — about one in ten — had a heart attack or a stroke, or died, within the three years [1].

Erythritol came out near the top. They then checked it in two more groups: 2,149 patients in the US and 833 in Europe. In each, the quarter of patients with the most erythritol in their blood had heart attacks, strokes and deaths pile up at 1.80 and 2.21 times the rate of the quarter with the least [1]. That ratio is what researchers call a hazard ratio — one group’s events divided by the other’s. In both groups the whole plausible range for it sat above 1, which is to say “the two quarters were no different” was not one of the answers the data left open.

Then they went looking for a mechanism, and found one. At levels you actually reach in blood, erythritol made platelets more reactive in the lab and increased clot formation in living animals [1]. Platelets are the tiny cell fragments in blood that form clots; clots are how heart attacks and most strokes actually happen.

And in eight healthy volunteers, a single drink containing 30 g of erythritol pushed plasma levels up and kept them above the threshold linked to that platelet effect for more than two days [1].

Laid out like that it sounds close to settled. It is not, and here is why.

The cohorts measured blood. Not diet.

Read the study design again: they measured how much erythritol was circulating in people’s blood. At no point did they measure how much erythritol those people ate [1].

That would not matter if blood erythritol only came from food. It does not.

Your body makes erythritol. It is synthesised internally through the pentose-phosphate pathway, a normal route of glucose metabolism [2]. Everyone has some in them, including people who have never knowingly eaten the stuff.

And now the detail that ought to have been in every headline. Elevated plasma erythritol has been associated with metabolic disorders — and weight loss has been linked to decreased plasma erythritol [2]. In one analysis of people who had bariatric surgery, the change in their BMI predicted the change in their fasting erythritol [2].

We should have printed the rest of that paper too, because two of its findings cut against us. Erythritol in those 15 surgery patients did not fall significantly afterwards, even as their weight did. And across all 218 people it measured, the thing that predicted fasting erythritol was age — not weight, blood sugar or insulin; lean adults in fact had more of it than adults with obesity [2].

So put it together. The study population was people already in a cardiac clinic. High blood erythritol tracks metabolic dysfunction in the large studies that have measured both [8][9]. Metabolic dysfunction causes heart attacks.

Which means a finding of “high circulating erythritol predicts cardiac events in cardiac patients” is entirely compatible with erythritol being a marker of the problem rather than a cause of it — a warning light rather than the fault.

Nobody has untangled this yet. The researchers studying endogenous production say plainly that the processes regulating it “remain unclear” [2].

There is a real counterpoint, and it belongs here. The same Cleveland Clinic group published a companion paper on xylitol the following year, and in it they state that sugar alcohols are made in the body “at levels over 1000-fold lower than observed following consumption as a sugar substitute” [3]. If that holds, what you eat swamps what you make, and the confound shrinks.

Does it settle it? No — because the cohorts drew blood after an overnight fast, and fasting is where the home-made fraction shows up. But it does cut the other way too: that eight-person dosing study found erythritol still elevated more than two days after a single drink, so a fasting sample from someone who eats the stuff daily is carrying a dietary signal as well.

The study’s own authors drew the same line, and it is the best support this page’s argument has. They wrote that the fasting samples from their US group, most of it enrolled before erythritol spread through processed food, “likely reflect endogenous levels” — the body’s own — while their European group, recruited into 2018, may be carrying some of what people ate [1].

Both of those are true at once. Which is the whole reason this page says unresolved instead of picking a side.

Did anyone outside the lab find it?

First, a word on the 2023 study’s own “validation”. Its two extra groups of patients, in the US and Europe, were separate people, but the same research team measured them [1]. That is a check, not an independent repeat.

Independent repeats exist, though, and this page went up without them. Three teams with no authors in common with the Cleveland Clinic group have measured erythritol in stored blood and followed people for years. None of them measured how much erythritol anyone ate.

The Nurses’ Health Study (Harvard and Tulane): 762 women who went on to develop coronary heart disease, each matched with a woman who did not. Their blood was drawn in 1989–90 or 2000–02 — before erythritol’s first approval as a food additive, the authors note, so they think it mostly reflects the body’s own production. The top quarter for erythritol had 1.55 times the risk of the bottom quarter after accounting for diet quality, lifestyle and weight, the plausible range running from 1.13 to 2.14. Then the authors accounted for diabetes, and the figure fell to 1.21, with a range from 0.86 to 1.70 — which puts “no difference at all” back on the table [8]. The 2023 study had accounted for diabetes from the start [1].

ARIC (Baylor, Johns Hopkins and others): 4,006 older Americans free of cardiovascular disease, blood drawn in 2011–13, followed for a median of 8.4 years. After accounting for the usual risk factors, diabetes and weight included, higher erythritol went with more hospital stays for heart failure, more deaths from cardiovascular disease and more deaths overall. It did not go with coronary heart disease or ischaemic stroke — the heart attacks and clot-driven strokes behind the 2023 headlines [9]. Its breakdown product, erythronate, did track both, so this is not a clean null [9].

ATBC (US National Cancer Institute researchers): 4,468 Finnish men who smoked, aged 50 to 69 when they joined a vitamin trial, their blood drawn in 1985–93 and followed for an average of 19 years. Against the bottom quarter, the top quarter for erythritol had a 15% higher death rate from any cause, 21% higher from cardiovascular disease and 17% higher from cancer, after accounting for smoking, blood pressure, cholesterol, weight and diabetes [10]. Deaths from heart disease and from stroke, taken separately, pointed the same way without clearing the bar for “probably not chance” [10].

So hold two things apart, because the evidence supports both. The association reproduces. Higher erythritol in stored blood went with worse outcomes years later in all three studies, in the US and in Finland, and in two of them in blood drawn before erythritol was a common additive.

And in those same data it behaves like a marker. It tracked weight, waistlines and blood sugar [8][9]. In the Nurses’ study the coronary signal faded once diabetes was accounted for, and in ARIC there was none for coronary disease or stroke [8][9]. And it predicted deaths from cancer, and from causes other than the heart, about as strongly as deaths from cardiovascular disease [9][10]. A clotting problem should show up as heart attacks and strokes. A warning light for metabolic ill health would show up everywhere. The pattern looks more like the second. That is our reading, and the authors of all three studies raise the same possibility — the Finnish team calling fasting erythritol possibly “a pre-existing indicator of endogenous metabolic production in individuals at risk of early death” [10].

What none of the three can tell you is whether the erythritol is doing the damage or just reporting it. Two other kinds of study have tried.

Somebody did test the direction. Four times now. They still disagree.

You cannot randomly assign people to have high erythritol for twenty years. But you can do the next best thing, and researchers have.

It is called Mendelian randomisation, and the trick is elegant: the gene variants you inherit were dealt at birth, before any lifestyle could confound them. If the variants that nudge erythritol levels up also track with heart disease, that points at cause. If they do not, the association was probably something else riding alongside.

Four such studies now exist — three when this page first went up, and a fourth published at the end of August 2026. Here are all four, including the one that makes our own argument look better and the three that do not. We got two of the first three wrong the first time, and we say where.

The earliest, in Diabetes in 2024, ran the analysis in both directions and concluded: “we did not find supportive evidence from MR that erythritol increases cardiometabolic disease” [4]. Its two estimates for heart disease pointed in opposite directions and neither was far enough from zero to rule out chance, and the authors add that their findings “await confirmation in well-designed prospective studies” [4]. It also found erythritol may decrease body mass index. That cuts against erythritol driving weight gain. On its own it does not show erythritol is a marker — we said it did, and that was a step too far.

Two studies in 2025 found the opposite direction. The one in Medicine took 60 gene variants linked to erythritol in the blood of 8,167 people and checked them against Finnish health records on some 400,000 others. Genetically higher erythritol went with more coronary heart disease — an odds ratio of 1.077, about 8% higher odds — and more ischaemic stroke, the kind caused by a clot: 1.157, about 16% higher [5]. Both results passed the study’s own checks for gene variants acting through some route other than erythritol [5].

We told you this study came with an acknowledged flaw, and it did. But the flaw belonged to its results for clots in the veins — deep-vein thrombosis and the lung clots that can follow it. Those failed the same checks, two of them flipped direction depending on the method, and the authors concluded there was no reliable evidence of an effect on them [5]. The heart-disease and stroke findings were the clean ones. We had pinned the flaw on the wrong results.

The other 2025 study, in the American Journal of Preventive Cardiology, called its evidence “suggestive” [6], and it is the one this page got most wrong. It printed its coronary heart disease result as 1.0020 [6], and we told you that meant two-tenths of one percent higher odds — a number “no human being could ever experience”. It is not an odds ratio at all. The heart-disease data it used, 361,194 people from UK Biobank, a large British health study, were analysed with a model that measures change in probability, not odds [6][14]. So 0.0020 is two-tenths of a percentage point of extra risk, in a group where only about 2.8% had the disease at all [14]. That is roughly 7% more heart disease in relative terms, or an odds ratio of about 1.08 by the standard conversion. The arithmetic is ours, not the paper’s, but it is the scale we were reading the number on, and on it the result is not tiny [14].

Two more things about that study, so you can weigh it properly. Its stroke result, from Finnish data analysed on the odds scale, was 1.0463 — about 5% higher odds [6][14]. And its genetic stand-in for erythritol came from a study that measured erythritol in the spinal fluid of 291 people, although the paper calls it circulating erythritol [6][13]. That is a small base, from a different body fluid, for a proxy of what is in your blood.

The fourth, published on 31 August 2026, after this page first went up, found genetically higher erythritol going with more ischaemic stroke — an odds ratio of 1.246, about 25% higher odds — and the link held when the authors also accounted for blood pressure, blood sugar and blood fats [11]. It names the same spinal-fluid study as the source of its erythritol genetics [11][13], so it and the Preventive Cardiology study are not two independent looks. And its authors add the caveat everyone in this story should: the result “should not be interpreted as direct evidence that dietary erythritol intake causes stroke” [11].

So: the one design built to answer “marker or cause” has now been run four times, for one null and three positives — though two of those positives share one genetic yardstick, so there are fewer independent looks than there are papers. The positive built on the most people, 8,167 of them, is the one we had marked down. (Each study reports its effect per step up in its own genetic measure, and the steps are not necessarily the same size, so read the directions more than the numbers.)

That leans toward erythritol in the blood being more than a bystander. It does not settle it. The authors of the Preventive Cardiology study and of the newest one both note that their genetic stand-ins cannot tell the erythritol your body makes from the erythritol you eat [6][11], and none of the four tested eating it at all. And all four drew mainly or entirely on people of European ancestry [4][5][6][11]. That is not a settled answer in either direction. It is the definition of the tier at the top of this page.

What that does and does not excuse

It does not dispose of the platelet work. That part did not rely on the cohorts at all — it put erythritol on platelets and watched what happened, and then watched clots form in animals [1]. The confound explains the epidemiology. It does not explain the biology.

Nor does it dispose of the dosing finding. Eight people is a small study, and it measured blood levels rather than health outcomes — but it does establish that eating the stuff raises circulating levels a great deal, and for longer than you would guess [1].

Then, in 2024, the same group ran the experiment the cohorts never could, and this page should have reported it from the start. Ten healthy volunteers drank 30 g of erythritol in water; ten others drank 30 g of glucose. Half an hour later the erythritol drinkers had more than a thousand times their starting level in their blood, and their platelets clumped more readily when prompted — in every one of the ten, with every trigger and every dose the lab tried. The glucose drinkers’ platelets did not [7]. Your body making its own erythritol cannot explain that result. The erythritol came out of the glass.

Weigh it for what it is, though. Twenty people, not randomly assigned and not blinded, run by the same lab as the 2023 study. It measured how readily platelets clump in a test tube half an hour after one drink — a tendency, not a clot, and not a heart attack or a stroke. The authors say it plainly: “the study size was small and replication of results with different cohorts is needed” [7]. When we searched PubMed on 18 September 2026 we found no published attempt to repeat it. It is the most direct human evidence on the harm side, and it is still one small study.

The authors themselves conclude more than we would: that erythritol is associated with heart attack, stroke and death risk and “fosters enhanced thrombosis” — makes clotting more likely [1]. The sentence of theirs we would stand behind is the one that follows it: “Studies assessing the long-term safety of erythritol are warranted” [1]. Warranted is not the same word as demonstrated.

One disclosure, because we report these everywhere. We first printed it incomplete, so here it is by role. The study’s senior author declares Cleveland Clinic patents in cardiovascular diagnostics and therapeutics; paid consulting, formerly for Procter & Gamble and now for Zehna Therapeutics; research funds from Procter & Gamble, Zehna Therapeutics and Roche Diagnostics; and eligibility for royalties from Cleveland HeartLab (part of Quest Diagnostics), Procter & Gamble and Zehna [1]. The research itself was paid for by public and charitable grants, and the European group of patients it drew on came from a study partly funded by the drug company Sanofi-Aventis [1]. That is not an accusation and it does not make the platelet experiments wrong. It is context you are entitled to have.

And a technicality that matters

Erythritol is a sugar alcohol, not a non-sugar sweetener. The WHO’s 2023 guideline — the one behind every “WHO says avoid sweeteners” headline — explicitly excludes polyols.

So if you switched to erythritol because of that guidance, you switched to something the guidance never covered, on the basis of a document that does not mention it. Which is its own small comment on how this information reaches people.

What to actually do

If you use erythritol occasionally, there is no demonstrated harm to act on. What has been shown in people is a short-term change — platelets clumping more readily half an hour after a single 30 g drink, in one small study from one lab [7] — not a heart attack or a stroke. The long-term signal in stored blood is one your body’s own erythritol can explain [1][8][9][10]. The short-term one is not.

If you use a lot of it daily, this is a sweetener where we would say the question is genuinely open, and moderating is a reasonable response to an open question. For scale, the 30 g dose in both Cleveland Clinic drink studies is, in the 2023 paper’s own comparison, about what is in “a single can of commercially available artificially sweetened beverage, a pint of keto ice cream” [1].

Europe’s food-safety agency, EFSA, set a limit of 0.5 g per kilogram of body weight a day — about 34 g for a 150 lb (68 kg) adult — and set it to prevent diarrhoea, not on the heart question [12]. It also estimated that the heaviest consumers in every age group could reach or pass that limit, while calling its own estimates overestimates [12]. Notice what the limit does not do: the 30 g drink that changed platelets sits under it for anyone who weighs more than about 132 lb (60 kg). The 34 g and the 132 lb are our arithmetic from EFSA’s number, not figures EFSA printed.

If you already have cardiovascular disease, that is a conversation for your cardiologist rather than a website. The people in these cohorts were cardiac patients, and they are the group for whom the finding, if it is real, would matter most.

And note what this page is not doing: telling you a scare is nonsense. That is what we concluded about aspartame, where two expert bodies had actually looked. We used to tell you that here they had not looked yet. That was wrong. EFSA re-evaluated erythritol in December 2023, reviewed this very study, and concluded that a causal link between eating erythritol and heart disease “has not been demonstrated by the available studies” — adding that fasting blood erythritol “may be a biomarker of metabolic disturbances”, such as type 2 diabetes and fat gained around the middle [12]. It judged the lab platelet experiments too thin a basis to predict what happens in people [12]. That is the regulator reading the evidence much as this page does.

What still separates this from aspartame is timing. EFSA’s review came before the 2024 drink study and the three independent cohorts, so the most direct human experiment on the harm side is one it could not weigh. Those are different situations, and they still deserve different answers.