Moderate Diet

Erythritol and the Sugar Alcohols: What the Heart Signal Actually Shows

Summary

Blood erythritol genuinely predicts heart attacks and strokes, and now does so in ordinary older adults as well as in cardiac patients, but the human body manufactures erythritol itself and manufactures more of it when metabolism is unwell, so the association may be measuring the disease rather than the sweetener, and the strongest evidence-based reason to keep intake modest is not the heart at all but a European regulator's 2023 finding that children already exceed the safe daily amount on gastrointestinal grounds.

Why Moderate

This entry is assigned Moderate overall.

Why not Strong. The central question — whether eating erythritol causes cardiovascular events — has no randomised trial with a clinical endpoint, no cohort that measured dietary intake, and two genetic analyses pointing in opposite directions with the better-designed one null. The mechanistic evidence for the proposed harm is dominated by test-tube, cell-culture and rodent work. An entry whose headline question rests on an association with a measured, unresolved confound cannot carry Strong.

Why not Emerging. Several components are much better established than "preliminary." The absorption and excretion pathway is well characterised and regulator-confirmed. The endogenous synthesis pathway is demonstrated with isotope work in human blood. The European safety assessment is a full regulatory evaluation of pooled toxicology. The scope exclusion in the WHO guideline is a matter of reading the document. An entry containing tracer-demonstrated human biochemistry and a completed regulatory assessment sits above Emerging.

Per-sub-area split, because the overall tier flattens genuinely different levels of confidence:
• Erythritol's absorption, excretion and non-glycaemic character: Strong. Well characterised, regulator-confirmed, mechanistically coherent. The specific glucose-and-insulin-response claim is inferred from the pharmacokinetics here rather than cited to a dedicated trial.
• Endogenous synthesis via the pentose phosphate pathway: Strong. Stable-isotope demonstration in human blood, not inference. The attached enzyme identification and the oxidative-stress regulation are weaker and separately labelled — the latter is cell-line evidence and never carries this tier.
• The gradient of blood erythritol with worse glycaemic status and adiposity gain: Strong as an observation, with the fold-change magnitudes coming from pooled samples by phenotype and their confidence intervals removed as unverified.
• The cardiovascular association in cardiac-referral cohorts: Moderate. Consistent across three cohorts, structurally unable to establish causation, and measuring an exposure that partly reflects the outcome's cause.
• The cardiovascular association in a community cohort of older adults: Moderate. Larger and free of the cardiac-referral objection, still without dietary measurement, still showing the confound in its own baseline table, and framed by its authors as marker rather than cause. Held at abstract level for now.
• The sampling-window and missing-diet objection from the United States regulator: Moderate, confirmed in substance and unquoted, because the document's full text was not retrieved for this entry.
• The reverse-causation critique and the inborn-errors natural experiment: Moderate. Published, coherent, resting on small numbers and a narrative rather than systematic review.
• The null genetic result for coronary artery disease: Moderate. Good design, appropriate journal, and instruments that capture self-made rather than eaten erythritol.
• The positive 2025 genetic result: Emerging. Weak-instrument signature, lower-scrutiny venue, internally inconsistent across venous outcomes.
• The thousandfold plasma rise and its persistence: Emerging as reported, from a pilot of eight with the magnitude confirmed in a controlled study.
• The human platelet-reactivity finding: Emerging. Controlled with a glucose comparator, ten per group, single dose, surrogate endpoint, same laboratory, no independent replication.
• In-vitro platelet activation, cultured-cell endothelial effects and murine thrombosis: Experimental and species- or model-flagged. Never to be stated as human evidence.
• Erythritol's gastrointestinal tolerance advantage over sorbitol and xylitol: Moderate, with the gram-per-kilogram thresholds now verified against the primary study and the remaining caveat being a sample of thirty-eight.
• The European acceptable daily intake and the exceeded-exposure finding: Foundational. A completed regulatory assessment, and the entry's firmest ground.
• The WHO scope exclusion: Foundational, because it is definitional rather than evidential.
• Xylitol's cardiovascular signal: Moderate, inheriting every limitation of the erythritol cardiac cohorts with no genetic test yet run.
• Xylitol's dental benefit: Emerging. Cochrane's own certainty rating, not this entry's opinion.
• Xylitol toxicity in dogs: Strong as veterinary evidence, and explicitly not human evidence.

Practical takeaway

If you are using erythritol to get off sugar, that is a defensible use. The bridge logic holds. A sweetened drink or dessert made with erythritol instead of sugar is a real step down from the sugared version, and someone taking that step should not be talked out of it by an unresolved association. What should be clear in your own head is that it is a bridge. The destination for drinks is water, and for food it is a palate that no longer needs the sweetness. Treat erythritol as scaffolding you intend to remove, not as a permanent ingredient you have solved the problem with.

Keep an eye on the amount, because the gut limit is real and lower than most people assume. The European limit works out at half a gram per kilogram of body weight per day: about thirty-five grams for a seventy-kilogram adult, about twenty for a forty-kilogram child. That is one large serving of keto ice cream, or a keto baking recipe split two ways. The regulator's own exposure work found children and adolescents already going past it, adolescents at the high end by roughly threefold. If there are keto or sugar-free products in the house that children eat freely, that is the population the finding was about.

Expect the gut to complain in a predictable pattern, and use the pattern. Symptoms come from dose, from liquid delivery on an empty stomach and from the other polyols in the same product. If a product upsets you, check the ingredient list before blaming erythritol — maltitol, sorbitol and inulin are common travelling companions and all three are worse tolerated. If you have irritable bowel syndrome or a diagnosed FODMAP sensitivity, polyols are one of the categories that group covers, and erythritol being the mildest member of a difficult class is not the same as it being fine for you.

If you have established cardiovascular disease, or the risk factors that lead to it, this is worth a conversation rather than a decision made from a headline. The original association was measured in people undergoing cardiac evaluation, and the 2025 community cohort found it in ordinary older adults too — so "that was only heart patients" is no longer an available reassurance. What is still true is that in both settings the people with high blood erythritol were the people with worse metabolic health to begin with, nobody measured what anyone ate, and no study has shown eating erythritol causing an event. That is not grounds for alarm, and it is not grounds for stopping an antiplatelet medication or changing any treatment — all of that sits with cardiovascular_health_management and with your clinician. It is grounds for treating a daily keto-dessert habit as a discretionary thing you could reduce cheaply while the question stays open, rather than as a settled-safe ingredient.

Do not swap to fruit juice. This is the substitution people make when a sweetener scares them, and it is the wrong direction. Whole fruit lowers type-2 diabetes risk in large cohorts while juice raises it in the same cohorts. Whole fruit is the answer to a sweet craving. Juice is a sugared drink with a better reputation than it has earned.

If you have a dog, xylitol is a household hazard, not a nutrition question. Sugar-free gum, mints, some peanut butters and some baked goods contain it, and the dose that hurts a dog is small. Store it out of reach, check peanut butter labels before sharing, and if a dog eats any, that is an immediate call to a vet rather than a wait-and-see.

Evidence detail

Why This Entry Exists

In February 2023 a paper in Nature Medicine reported that people with high blood erythritol had roughly double the three-year risk of dying, having a heart attack or having a stroke, and that erythritol made platelets stickier in a test tube and made mice clot faster. The coverage wrote itself. Within a week the sentence in circulation was "the sweetener in your keto ice cream causes heart attacks," and that sentence is not what the paper showed.

What makes this topic unusually hard is that the counter-argument is not the usual industry hand-waving. It is a measured, published fact that predates the controversy entirely. Erythritol is not only something you eat. It is something you make. Human cells synthesise it from glucose through the pentose phosphate pathway, and a 2017 study of university students found blood erythritol twenty-one times higher in those with worse blood-sugar control and fifteen times higher in those gaining central fat, then demonstrated the synthesis directly by incubating human blood with labelled glucose. That cohort was studied years before erythritol became a controversy, and no dietary intake was recorded in it, which makes "the metabolically worse-off group was eating more diet products" an unlikely explanation for the gradient — though it cannot be formally excluded, because nobody asked what they ate. If your body makes more erythritol when your metabolism is deteriorating, then "high erythritol predicts heart attacks" and "erythritol causes heart attacks" become two very different sentences, and only the first one has support.

That does not settle it in the sweetener's favour either, and the entry has to resist the pull in that direction too. Eating thirty grams of erythritol raises blood levels roughly a thousandfold above baseline and keeps them raised for more than two days, which is nowhere near anything the body produces on its own. A follow-up study in humans, small and using a surrogate measurement rather than a clinical outcome, found that erythritol but not glucose made platelets more reactive after a single dose. In 2025 the association turned up again in a large community cohort of older adults who did not have heart disease when they were measured, which removes the "this was only ever cardiac patients" objection while leaving the marker-versus-cause question exactly where it was. And a European regulator, looking only at the gut and never at the heart, concluded in 2023 that current intakes already exceed the safe level in children and adolescents, and refused industry's request to drop the laxative warning from the label.

So the honest position is neither camp's. It is that a real association sits on top of a confound large enough to swallow it, that the mechanistic work is suggestive and mostly non-human, that the one hard regulatory finding is about diarrhoea rather than thrombosis, and that the long-term benefit these molecules are sold for — sustained weight loss or better cardiometabolic health over years — has never been demonstrated in a trial, even though the short-term substitution effect on calories and blood sugar is perfectly real. Unknown long-term risk set against unproven long-term benefit is a weak trade. That is a much stronger argument than a harm assertion, and unlike a harm assertion it is one the evidence actually supports.

What bad advice this protects against, in all directions:
• "Erythritol causes heart attacks and blood clots." Overstated in a way that will not survive contact with the papers. The human evidence is an association across four cohorts — three cardiac-referral, one community-based — in none of which dietary erythritol intake was measured, so the exposure variable is blood erythritol of unknown origin, much of which may be self-made. The clotting evidence is a test-tube platelet assay and an injected mouse injury model. No human clinical thrombotic event has been produced by any experiment in this literature. Saying "causes" here does not just overstate the case, it hands the other side a genuine correction to make, and they will make it.
• "Erythritol is the safe natural one, your body already makes it anyway." This is the mirror image of the "it's ninety-nine percent water" defence of diet drinks, and it fails the same way. That a molecule occurs endogenously says nothing about what a supraphysiological dose of it does. The thousandfold spike from a normal thirty-gram serving is the pro-erythritol camp's own inconvenient number, because it comes from the same measurements they cite for reassurance. Endogenous production is a reason to doubt the epidemiology. It is not a reason to assume a large dietary bolus is inert.
• "The World Health Organization said sweeteners don't help you lose weight, so stop eating erythritol." A citation error, and the one most likely to sink an argument in this topic. The 2023 WHO guideline explicitly excludes sugar alcohols from its scope, in its own words, because they contain calories and are therefore not classified as non-sugar sweeteners. Its recommendation covers aspartame, sucralose, stevia, saccharin and their relatives. It does not cover erythritol or xylitol. Anyone citing WHO about erythritol, in either direction, is citing outside the document.
• "Sweeteners are poison, drink juice instead, it's natural." The worst possible substitution. Our own fruit entry has whole fruit lowering type-2 diabetes risk while fruit juice raises it in the same cohorts. Juice is not a health drink. And on the class question generally, diet soda genuinely beats sugared soda as a step off sugar. That concession has to survive, because it is true and because an entry that reads as "sweeteners are poison" contradicts its own library. The framing that holds is a bridge, not a destination. Water is the finish line.
• "All the sugar alcohols are the same, just avoid them." Flattens a real and mechanistically explained difference. Roughly ninety percent of erythritol is absorbed in the small intestine and passed out in urine rather than reaching the colon to be fermented, which is why its laxative threshold is several times sorbitol's. If someone is going to use a polyol, erythritol is the best-tolerated member of the class by a wide margin. That advantage is real, and it is also the reason erythritol gets eaten in the large doses that produce the thousandfold plasma spike. Both halves of that sentence are true at once.
• "Xylitol is proven to prevent cavities." Weaker than almost everyone states it. The Cochrane review of ten trials in nearly six thousand participants rated the evidence low to very low quality and insufficient, with the one positive finding coming from two trials by the same authors in the same population. Xylitol has the strongest trial base of any polyol health claim, and Cochrane still rates that base low quality. Calling it established would be exactly the overclaim this entry criticises on the other side.
• "Erythritol has been debunked as a risk, industry sorted that out." Also wrong. The most careful critique in the literature, from a researcher whose career is built on documenting sugar harm rather than defending sweeteners, concludes only that it cannot be concluded that dietary erythritol promotes thrombosis until long-term clinical data exist. That is an absence of conclusion, not a proof of safety. The association has since appeared in a community population as well as in cardiac patients. And the European regulator's gastrointestinal finding stands entirely outside the cardiovascular argument, so no amount of reverse-causation reasoning touches it.

What this entry OWNS: the erythritol cardiovascular signal and how to read it; the endogenous-production confound and the evidence that establishes it; the gastrointestinal tolerance profile of erythritol against the other polyols and the regulatory limit built on it; the xylitol parallel including the dental claim and the canine hazard.

What this entry DEFERS: the class-level argument about non-sugar sweeteners, including the weight-control question, the microbiome question and the WHO guideline itself, to artificial_sweeteners_evidence; anything about assessing or managing actual cardiovascular risk, including lipids, blood pressure and antithrombotic therapy, to cardiovascular_health_management; the general question of how to read an observational association against a randomised one to rct_vs_observational_evidence; and blood-sugar management as a topic to blood_sugar_regulation.

Evidence

Read the tiers, not the thesis. This topic has an unusual shape: the finding with the biggest headline is observational and structurally unable to separate cause from consequence, the study design built specifically to break that deadlock came back null, a second study using the same design came back positive but with instruments too weak to trust, and the single piece of evidence a regulator was willing to act on has nothing to do with the heart. Every claim below is labelled by evidence type, because confusing an in-vitro or rodent result for a human harm finding is the defining error of this subject.

Sub-area A: What erythritol is and what it does in the body

1. Erythritol is readily and dose-dependently absorbed in humans, converted to erythronate only to a small extent, and then excreted unchanged in the urine, which is why it is essentially non-glycaemic. [Strong] Evidence type: human — regulatory synthesis plus a human dose-ranging pharmacokinetic study. The absorption and excretion pathway is confirmed by EFSA's 2023 re-evaluation and by human dose-ranging work. Two honest caveats. First, absorption is saturable, so the fraction reaching the colon rises at higher doses — which is precisely why very large single servings cause gut symptoms despite good tolerance at ordinary ones. Second, the conversion to erythronate is a genuine open question; the authors of the dose-ranging study state that its implications for human health remain to be determined, a clause that appears nowhere in marketing, and it matters more now that a community cohort has found erythronate tracking more cardiovascular outcomes than erythritol itself. A note on citation fidelity: the pathway is verified, but no dedicated human trial reporting glucose and insulin response curves was retrieved for this entry, so the non-glycaemic property is stated here as following from the pharmacokinetics rather than as a separately measured outcome. EFSA FAF Panel, EFSA Journal 2023; Bordier et al., International Journal of Molecular Sciences 2022. Cui bono: the entire keto, diabetic and low-carbohydrate packaged-food category depends on this property — it is the commercial reason erythritol exists at scale. That does not make it false; the pharmacokinetics are well characterised and regulator-confirmed. It does explain why "excreted unchanged" is repeated everywhere and "metabolised to erythronate, implications undetermined" is repeated nowhere. Competing sweetener and sugar interests would benefit from muddying this and have very little to work with.

2. Erythritol is manufactured inside the human body from glucose through the pentose phosphate pathway. [Strong] Evidence type: human — ex-vivo stable-isotope incubation of human blood, plus in-vivo stable-isotope dried-blood-spot work. This is not an inference or an industry talking point. Incubating human blood with labelled glucose showed erythritol being synthesised via the pentose phosphate pathway, and in-vivo dried-blood-spot work in humans showed onward conversion to erythronate. Two attachments to this claim carry weaker evidence than the pathway itself and must not inherit its label. The first is the enzymology: the final step has been attributed to sorbitol dehydrogenase and alcohol dehydrogenase 1 in subsequent cell-based work rather than in the 2017 paper. Fidelity flag: that follow-up paper was not retrieved for this entry, so state the enzyme names without an author or a year. The second is the regulation: the finding that erythritol synthesis rises under oxidative stress and is governed by the non-oxidative branch of the pathway comes from a human lung-cancer cell line in culture, not from people — evidence type for that sub-claim: in-vitro cell line. What is human and observational, from the 2017 cohort itself, is that blood erythritol tracks upward with worse glycaemic status, which is consistent with flux rising under high blood glucose. Hootman et al., PNAS 2017; cell-line regulation work, Frontiers in Nutrition 2022. Cui bono: sweetener manufacturers benefit enormously from this fact and will over-state it into "debunked." The counterweight that matters is timing — this was published in 2017 for an unrelated reason, as part of an adiposity metabolomics study, before the controversy existed. It is not a defensive artefact produced after the 2023 paper.

Sub-area B: The cardiovascular association

3. Higher circulating erythritol is associated with major adverse cardiovascular events across three independent cardiac-referral cohorts. [Moderate] Evidence type: human observational. Untargeted metabolomics in a discovery cohort of 1,157 patients undergoing cardiac risk assessment linked several polyols, erythritol most strongly, to three-year death, non-fatal heart attack and stroke. Targeted validation in a United States cohort of 2,149 and a European cohort of 833 confirmed it, with top-versus-bottom quartile adjusted hazard ratios of 1.80 (95% CI 1.18–2.77) and 2.21 (95% CI 1.20–4.07). The confounds are not incidental, they are structural: every cohort here is a cardiac-referral population, so these are not general-population risk estimates; no dietary erythritol intake was measured anywhere in the study, so the exposure is plasma erythritol of unknown origin; and erythritol rises endogenously with high blood glucose, which independently predicts the same outcomes. The association is real. Its interpretation as dietary is not established by this paper. Witkowski et al., Nature Medicine 2023. Cui bono, both ways. Pro-alarm: the laboratory behind this runs a long-standing research programme on metabolite-driven atherothrombosis and holds related commercial and patent interests in cardiovascular diagnostics, and a headline metabolite finding extends that programme and attracts funding; media and low-carbohydrate-sceptic commentators get enormous traffic from "sweetener causes heart attacks." Pro-dismissal: erythritol is the backbone bulk sweetener of the entire keto packaged-food category, and manufacturers and the Calorie Control Council issued rapid rebuttals. Both camps have money in the answer. Neither camp's incentive changes what the confound analysis shows.

4. The association also appears in a community cohort of older adults who did not have cardiovascular disease when they were measured — and it is strongest for the body's own downstream metabolite rather than for erythritol itself. [Moderate] Evidence type: human observational, community-based prospective cohort. In 4,006 participants of the ARIC study with metabolomic profiling and no prevalent cardiovascular disease at the 2011–2013 visit, followed for a median of 8.4 years, higher circulating erythritol and higher erythronate were both significantly associated with hospitalisation for heart failure, and erythronate was additionally associated with coronary heart disease, stroke and heart failure with reduced ejection fraction. This claim must never be reported one-sidedly, because it cuts in both directions at once. It removes the "cardiac-referral population only" objection that the rest of this literature rests on — these were ordinary older adults without heart disease. It also displays the confound rather than resolving it: participants in the top tertiles were older and more likely to have diabetes, hypertension, hyperlipidaemia and microalbuminuria, with higher body mass index, higher cardiac biomarkers and lower kidney filtration, which is precisely the metabolic profile that raises the body's own erythritol production. Dietary intake was again not measured. The authors' own framing is that circulating erythritol and erythronate are markers of cardiometabolic health and cardiovascular outcomes. That erythronate — the product of the body's own processing of erythritol — tracks more outcomes than erythritol does sits more comfortably with a marker reading than with a dietary one. JACC: Advances 2025, ARIC study analysis. Fidelity flag: read at abstract level for this entry. Do not attach hazard ratios, effect sizes or a first-author name to it, and do not describe the metabolomics platform. Cui bono: a publicly funded long-running community cohort with no sweetener stake, which is why both camps will quote it selectively — the alarm side gets "confirmed outside cardiac patients," the dismissal side gets "the authors called it a marker." Both sentences are genuinely in the paper, and quoting either alone misrepresents it.

5. The United States regulator's evaluation of the 2023 paper flags the sampling window and the missing dietary measurement. [Moderate] Evidence type: regulatory document review. The agency's published evaluation notes that the 1,157-patient discovery cohort underwent cardiac risk assessment between 2001 and 2007, that none of these studies assessed dietary intake of erythritol, and that in some of them the blood in which erythritol was measured was collected before erythritol was approved for use in United States foods. If the erythritol being measured was largely self-made rather than eaten, the exposure variable becomes a metabolic-dysfunction readout almost by construction. Correction flag: an earlier version of this entry stated that "around eighty percent of the samples" fell in that window. That figure is not in the regulator's document and must never be used. What the document supports is the 2001–2007 discovery-cohort window, the absence of dietary measurement everywhere, and the observation that some sampling predates erythritol's United States food-additive approval. Fidelity flag: the document's substance was confirmed through its published summary rather than full-text retrieval, so do not attach a page number, a named author or a verbatim quotation. United States Food and Drug Administration, An evaluation of the article "The artificial sweetener erythritol and cardiovascular event risk." Cui bono: industry cites this point constantly and it happens to be a serious objection, which is the awkward combination this entry has to handle honestly — an argument does not become wrong because a self-interested party makes it. A regulator raising a sampling-window problem has no product to protect. Equally, the temptation to inflate it into a bigger number than the document contains is exactly how a good objection gets discredited.

6. The same research group reported a comparable cardiovascular signal for xylitol. [Moderate] Evidence type: human observational, plus in-vitro, rodent and small human ingestion work. Metabolomics on fasting plasma from over three thousand patients undergoing cardiac evaluation found elevated circulating xylitol associated with increased three-year major adverse cardiovascular events; laboratory and animal work showed enhanced platelet reactivity and clot formation, and a small group of healthy volunteers drinking xylitol beverages showed large plasma rises with increased platelet responsiveness — a single-dose surrogate measurement, not an outcome. The structural point matters more than the result: this is the same group, the same design, the same cardiac-referral population type, and xylitol is also produced endogenously in humans, so the identical reverse-causation problem applies. A second finding from the same laboratory using the same method on a second molecule is not independent confirmation. It inherits every limitation of the first, and no Mendelian randomisation has been run for xylitol to test it. Witkowski et al., European Heart Journal 2024. Cui bono: extending the finding to a second polyol substantially expands the programme's reach and media footprint; on the other side the chewing gum, oral care and sugar-free confectionery industries have direct exposure. Neither incentive resolves the confound, which is identical to erythritol's and equally unaddressed.

Sub-area C: The confound, which is measured rather than speculative

7. Blood erythritol is dramatically higher in people with worse glycaemic status and in people gaining central fat. [Strong] Evidence type: human observational, in the same paper as the tracer work above. In a nine-month prospective study of 264 university freshmen, participants who gained central adiposity had fifteen-fold higher blood erythritol on the targeted assay than those with stable adiposity, and participants with baseline HbA1c above 5.05% had twenty-one-fold higher blood erythritol than those below it. Statistic flag: the two fold-changes are verified against the paper's reported findings and may be quoted. The confidence intervals carried in an earlier version of this entry were not verifiable and have been removed rather than repeated — do not reintroduce them without the primary text in hand. Two honest limits beyond that: the fold-change magnitudes come from pooled plasma by phenotype rather than individual-level distributions, and the study cannot tell us the dietary-versus-endogenous split in a modern population that also eats erythritol. That split is the open question of this entry. Hootman et al., PNAS 2017. Cui bono: as above — industry gains from emphasis, the harm-narrative side gains from omission, and the pre-controversy publication date is what makes the finding hard for either side to dismiss.

8. The reverse-causation objection has been made formally in the peer-reviewed literature, and it comes with a natural experiment. [Moderate] Evidence type: narrative review of human observational, human inborn-error case series and rodent evidence. Mazi and Stanhope raise three specific objections. People with inborn errors of metabolism who run chronically elevated erythritol for life do not show higher platelet activation or thrombosis — the closest available lifetime high-exposure human model, and it is negative. Most long-term animal studies of high erythritol intake do not support a role in platelet activation or thrombosis. And erythritol may, in their words, be merely a marker of pentose phosphate pathway dysregulation caused by impaired glycaemia. Their conclusion is explicitly not exoneration: they state that until long-term clinical trial data exist, it cannot be concluded that dietary erythritol promotes platelet activation, thrombosis and cardiometabolic risk. The natural experiment is suggestive rather than decisive — the relevant conditions alter several pathway metabolites at once rather than erythritol in isolation, the numbers are small, and absence of reported thrombosis in a rare-disease population is partly a question of what was looked for. Mazi and Stanhope, Nutrients 2023;15(18):4011. Cui bono: Stanhope's research career is built substantially on documenting sugar and fructose harm, which cuts directly against an "industry apologist" reading and makes the critique more credible rather than less. Manufacturers nonetheless cite this paper heavily. On the other side, the harm-narrative ecosystem benefits from treating it as industry pushback rather than engaging the inborn-errors evidence.

Sub-area D: The design built to break the deadlock, run twice, with opposite answers

9. Mendelian randomisation did not support erythritol causing coronary artery disease. [Moderate] Evidence type: human genetic. Citation trap, and it is the second-biggest in this topic: the paper is titled "Erythritol as a Potential Causal Contributor to Cardiometabolic Disease" and its result is null. Bidirectional analysis using genome-wide significant variants from three cohorts, in European-ancestry populations, found no supportive evidence that increased erythritol increases coronary artery disease, and an indication that erythritol may decrease body mass index. The authors conclude they did not find supportive evidence from this design that erythritol increases cardiometabolic disease, and that the findings await confirmation in well-designed prospective studies. The limitation cuts in an underappreciated direction: genetic instruments for plasma erythritol overwhelmingly capture endogenous synthesis capacity, not dietary intake, so this is strong evidence against self-made erythritol being causal — which is exactly the reverse-causation question — and much weaker evidence about what a thirty-gram dietary dose does. A null here does not license "proven safe to eat." Khafagy, Paterson and Dash, Diabetes 2024;73(2):325. Cui bono: industry will cite this as exoneration while ignoring the endogenous-versus-dietary instrument problem; the harm-narrative side benefits from omitting the paper entirely, and its misleading title makes that omission easy to disguise as inclusion. The authors are academic diabetes researchers with no evident sweetener stake.

10. A 2025 analysis using the same design reported the opposite, and should not be weighted equally. [Emerging] Evidence type: human genetic. Two-sample analysis using sixty variants against Finnish biobank outcomes found genetically predicted higher erythritol significantly associated with coronary heart disease and ischaemic stroke, with deep vein thrombosis suggestive and venous thromboembolism and pulmonary embolism inconsistent in direction. Two reasons to discount it relative to the null result above. Sixty independent genome-wide significant variants drawn from a discovery study of only 8,167 people is implausibly many for a metabolite trait at that sample size, which is the classic signature of weak instruments and winner's curse — a bias that pushes estimates back toward the observational association the design was supposed to escape. And the journal sits far below the other paper's level of editorial scrutiny. The internal inconsistency across the venous outcomes is also awkward, given that clotting is the proposed mechanism. Sun et al., Medicine (Baltimore) 2025. Cui bono: authors and journals gain citations from a result aligned with a high-profile controversy, and the harm-narrative ecosystem gains a "genetics confirms it" headline. Industry benefits from pointing at the weak instruments — and here the methodological objection is legitimate rather than merely convenient. That distinction is worth naming, because dismissing a paper for good reasons and dismissing it for paid reasons look identical from the outside.

Sub-area E: The mechanism experiments, and the error to avoid

11. Erythritol enhanced platelet reactivity in a test tube and increased clot formation in mice. [Experimental] Evidence type: in-vitro and rodent. This is where the defining error of the topic gets made. The mechanistic claim in the 2023 paper rests entirely on non-human models: platelet reactivity in vitro at physiological concentrations, and thrombosis in a mouse carotid-injury model with injected erythritol. No human clinical thrombotic outcome was produced by any experiment in that paper. Mouse models use acute injected exposure with induced arterial injury, which is not chronic dietary exposure, and in-vitro platelet assays remove the endothelium, blood flow and the body's own anticoagulant regulation. A 2025 report that erythritol adversely affects brain microvascular endothelial cell function produced a second wave of coverage and belongs in exactly this tier and no higher: it is cultured cells in a dish, not a person, not a stroke, and not a dose that anyone ate. Evidence type for that report: in-vitro human cell culture. **Fidelity flag: it was identified at title and journal level only for this entry — Journal of Applied Physiology, 2025 — so do not quote its concentrations, methods or effect sizes, and never let it be described as evidence about human brains.** The counterweight from the critique literature is that most longer-term animal feeding studies do not replicate the thrombosis effect, and that discrepancy between acute injection and chronic feeding is itself informative — it points to something route- or peak-concentration dependent rather than a stable property of the molecule. Witkowski et al., Nature Medicine 2023; counterweight in Mazi and Stanhope, Nutrients 2023. Cui bono: mechanistic plausibility is what converts an association into a story, and "clots faster in mice" is the sentence that made this global news. Industry correctly notes that in-vitro and rodent thrombosis results have a poor record of predicting human cardiovascular outcomes — a general truth that also happens to serve them here.

12. A thirty-gram dose raises blood erythritol roughly a thousandfold, and a controlled human study found erythritol but not glucose increased platelet reactivity. [Emerging] Evidence type: human intervention, small, surrogate endpoint. The original pilot gave eight healthy volunteers a thirty-gram beverage; plasma erythritol rose steeply and stayed elevated for more than two days, above the concentrations that produced effects in the non-human experiments. The 2024 follow-up, with ten volunteers per group and a glucose comparator, confirmed the thousandfold magnitude and found enhanced stimulus-dependent platelet aggregation and increased release of platelet granule markers, with no comparable effect from glucose. This is the most important development since 2023, because the glucose control shows the effect is not simply "ingesting a sweet carbohydrate," and because it moves the platelet finding out of the test tube into living people. It remains a surrogate: enhanced aggregation measured after a single dose is not a clot, a heart attack or a stroke, no clinical outcome was measured, the groups were parallel rather than crossover, and it comes from the same laboratory as the original finding. Independent replication by an unaffiliated group is the single thing most worth watching for. The authors themselves frame it as warranting reconsideration of the ingredient's regulatory status, not as demonstrating harm. Witkowski et al., Nature Medicine 2023 (pilot, n=8); Witkowski et al., Arteriosclerosis, Thrombosis, and Vascular Biology 2024. Cui bono: the thousandfold figure is rhetorically devastating and travels far better than its caveats, and a brief report in a specialty journal is a lower bar than the original paper — that is not an accusation, it is the reason independent replication matters. Industry emphasises the surrogate endpoint and the tiny sample, both legitimate objections that do not explain why the elevation persists past two days.

Sub-area F: The gut, and the only regulator that acted

13. Erythritol tolerates far better than the other polyols, with a laxative threshold roughly three to four times sorbitol's. [Moderate] Evidence type: human intervention, dose-ranging tolerance trials. Dose-response work in healthy subjects — 14 male office workers and 24 female students, with sorbitol and sucrose dosed for comparison in the same study — estimated erythritol's laxative threshold at about 0.80 g per kg body weight in women and 0.66 in men, against sorbitol at about 0.24 and 0.17. For an eighty-kilogram man that is roughly fifty grams of erythritol against roughly fourteen of sorbitol. Head-to-head in liquid, fifty grams of erythritol significantly increased only nausea and stomach rumbling, with twenty and thirty-five grams provoking no significant symptoms, while fifty grams of xylitol significantly increased nausea, bloating, rumbling, colic, watery stools and bowel movement frequency. The mechanism explains the gap cleanly: most erythritol is absorbed in the small intestine and never reaches colonic bacteria to be fermented. Verification note: the gram-per-kilogram thresholds were checked against the primary study's reported values for this entry and the earlier open flag on them is now closed — they are not transposed. The remaining caveat is sample size rather than fidelity: thirty-eight subjects is a thin base for a number that gets quoted as a constant. Substantive caveats regardless: these are single boluses in liquid on an otherwise normal gut, so they overstate tolerance for anyone with irritable bowel syndrome, small intestinal bacterial overgrowth or FODMAP sensitivity — polyols are the P in FODMAP; liquid on an empty stomach tolerates worse than the same dose inside a meal; and many commercial products combine erythritol with other polyols or with inulin, so real-world complaints often cannot be pinned on erythritol alone. Oku and Okazaki, Nutrition Research 1996; Storey, Lee, Bornet and Brouns, European Journal of Clinical Nutrition 2007. Cui bono: manufacturers fund and cite tolerance research heavily and "better tolerated than sorbitol" is a core marketing claim, with the head-to-head study carrying the food-industry author affiliations typical of this literature; competing categories such as stevia, allulose and monk fruit benefit from erythritol's gut reputation. This platform's interest aligns with neither, because the tolerance advantage is real and mechanistically explained, and it is also the reason erythritol gets consumed in the doses that produce the thousandfold spike.

14. EFSA set an erythritol acceptable daily intake of 0.5 g per kg body weight based on diarrhoea, found real-world intake already exceeds it in children and adolescents, and refused to drop the laxative warning label. [Foundational] Evidence type: regulatory safety assessment of pooled human and animal toxicology. This is the strongest evidence-based precautionary point available, and it has nothing to do with the cardiovascular controversy — which means no amount of reverse-causation argument touches it. The 2023 re-evaluation identified diarrhoea as the concern and set a limit the Panel considered protective against both the immediate laxative effect and chronic effects secondary to diarrhoea. Its own exposure modelling makes the exceedance concrete rather than rhetorical: the highest chronic estimates were in children at 742 mg per kg body weight per day and adolescents at 1,532, against a limit of 500, with acute exposure reaching 3,531 mg per kg body weight per meal for children at the ninety-ninth percentile. Minor fidelity note: the retrieved summary did not separate cleanly which of those chronic figures is a mean and which a ninety-fifth-percentile estimate, so quote them as the Panel's high-end chronic estimates rather than assigning each a percentile. The Panel concluded that high-intake individuals may be at risk of adverse effects after single and repeated exposure. Industry had applied for exemption from the laxative warning requirement; the Panel determined the available data did not support the proposal. For a forty-kilogram child the limit is twenty grams, which is roughly one generous serving of keto ice cream or a handful of sugar-free sweets. Meanwhile the United States regulator has not revised erythritol's generally-recognised-as-safe status in response to the 2023 paper, so the two major regulators have diverged. Do not present this as a regulator responding to the thrombosis question. It is not. EFSA FAF Panel, EFSA Journal 2023. Cui bono: EFSA has no commercial stake and issued this against an active industry application — refusing an exemption request costs a regulator nothing but tells you the submitted data did not clear the bar. On the other side, the applicant was industry, and the unchanged United States position is cited by industry as the authoritative view. EFSA is the better citation here precisely because it is the harder one to dismiss.

Sub-area G: The dental claim and the dog

15. Xylitol's cavity-prevention evidence is rated low to very low quality by Cochrane. [Emerging] Evidence type: human randomised trials, systematic review, low certainty. Ten studies, nearly six thousand participants, comparing xylitol-containing toothpaste, syrup, lozenges, sweets and tablets against controls. The conclusion was that there is little high-quality evidence of benefit against caries and that the remaining evidence is insufficient to determine whether these products prevent decay in infants, older children or adults. The one positive signal — fluoride toothpaste containing xylitol outperforming fluoride-only toothpaste in children's permanent teeth — is explicitly flagged as low quality, high risk of bias, and derived from two studies by the same authors in the same population, which is close to a single finding. The defensible statement is that xylitol has the strongest randomised base of any polyol health claim and Cochrane still rates that base low quality. The review is from 2015 and later work exists that was not retrieved here, so check whether certainty has been revised before leaning on it. Riley, Moore, Ahmed, Sharif and Worthington, Cochrane Database of Systematic Reviews 2015. Cui bono: gum and oral-care manufacturers have marketed the caries claim for decades and fund much of the trial literature, which is a large part of why the bias ratings are poor; fluoride-focused dental bodies and makers of competing products benefit from the claim being downgraded. Cochrane's own certainty rating is the safest ground because it is not a stakeholder position.

16. Xylitol is severely toxic to dogs. [Strong] Evidence type: veterinary clinical, species-specific, explicitly not human evidence. In dogs, xylitol triggers rapid dose-dependent insulin release causing profound low blood sugar. Doses above roughly 100 mg per kg are associated with hypoglycaemia; above roughly 500 mg per kg, severe liver failure with hepatic necrosis can follow. Signs can appear within thirty minutes or be delayed twelve to eighteen hours if food slows absorption, and the United States regulator reports effects within ten to sixty minutes and deaths in as little as an hour. Cats do not show the response and are not considered at risk, and clinical toxicosis reports are essentially confined to dogs — treat "dogs only" as a practical household rule rather than a claim about every species, since xylitol-induced insulin release has been described in some other animals. Fidelity flag: the cross-species detail was not re-verified for this entry. What is firm is that this is a canine hazard, that cats are not considered at risk, and that humans do not show the canine insulin response. The thresholds come from case series and poison-control data rather than controlled dosing, so treat them as practical action levels rather than precise constants. The species specificity is the entire point: humans do not show the insulin response dogs do, and this must never be allowed to imply human liver toxicity. Its practical value is that xylitol-sweetened peanut butter, gum and baked goods in a household with a dog are a genuine emergency risk. Merck Veterinary Manual; Dunayer, xylitol toxicity in dogs; FDA consumer update. Cui bono: veterinary bodies and poison-control services have straightforward public-safety motivation and no stake in suppressing xylitol; gum and confectionery manufacturers prefer this not be prominent on packaging. This is one of the rare claims here where the incentives are lopsided and the evidence is not seriously contested.

Sub-area H: What the WHO guideline does and does not cover

17. The 2023 WHO guideline on non-sugar sweeteners explicitly excludes sugar alcohols, so it cannot be cited about erythritol or xylitol. [Foundational] Evidence type: human — guideline synthesising randomised and observational evidence; the scope exclusion itself is definitional. The guideline defines its subject as sweeteners not classified as sugars, and states that because low-calorie sugars and sugar alcohols contain calories they are not considered non-sugar sweeteners and the recommendation does not apply to them. Its named scope is acesulfame K, aspartame, advantame, cyclamates, neotame, saccharin, sucralose and stevia derivatives. Within that scope, the recommendation against using these sweeteners for weight control or disease-risk reduction is conditional and rests on low-certainty evidence: short trials under three months showed reduced sugar and energy intake and lower body weight, the small number of trials running six to eighteen months showed no effect on body weight, and observational data suggested possible increased risk of type 2 diabetes, cardiovascular disease and mortality. Note the symmetry the entry gains by naming it — those observational harm signals face exactly the same reverse-causation problem as the erythritol story, because people at higher cardiometabolic risk choose diet drinks. Applying scepticism to one and not the other is the tell of an argument built backwards. World Health Organization, Use of non-sugar sweeteners: WHO guideline, 2023. Cui bono: WHO has no commercial stake but does hold an institutional position on sugar reduction that a "sweeteners don't help either" finding complicates rather than serves; the sweetener industry funded rapid rebuttals, and independent nutrition scientists also published methodological objections, so criticism of the guideline is not purely industry-driven. For polyols specifically, anyone citing WHO in either direction is citing outside the document.

Mechanism

Three separate mechanisms are running here and they are routinely mixed up.

Why erythritol does nothing to blood sugar. It is a four-carbon polyol. Roughly ninety percent is absorbed intact in the small intestine, it is not metabolised for energy to any meaningful extent, and it leaves in the urine essentially unchanged. Nothing enters glycolysis, so there is no glucose rise and no insulin response. The same property explains its gut tolerance: what is absorbed high in the intestine never reaches colonic bacteria, and it is bacterial fermentation of unabsorbed polyol, plus the osmotic water it drags with it, that produces the bloating, rumbling and diarrhoea associated with sorbitol and maltitol. Absorption is saturable, however, so at large enough doses the unabsorbed fraction grows and the symptoms appear anyway.

Why the body makes it. Glucose entering the pentose phosphate pathway can be diverted to erythrose-4-phosphate and onward, via dehydrogenase enzymes, to erythritol. In people, blood erythritol tracks upward with worse glycaemic control and with expanding visceral fat, which is consistent with flux rising as metabolic health deteriorates; the finding that synthesis also rises under oxidative stress comes from cultured cells rather than from people and should be labelled that way whenever it is used. So a person with poor glycaemic control and expanding visceral fat is, by ordinary biochemistry, running a small erythritol factory. This is the mechanistic reason the epidemiology is ambiguous: the exposure being measured is partly an output of the disease being predicted. It is also the reason the genetic studies are less decisive than they look, since the genetic variants that predict blood erythritol are largely variants in this synthesis machinery rather than in dietary habit. And it is why the downstream metabolite erythronate tracking more cardiovascular outcomes than erythritol itself, in the community cohort, reads more naturally as metabolic-state signalling than as a dietary effect.

Why platelets are the proposed link. The claimed pathway is that erythritol, at concentrations reached after ingestion, lowers the threshold at which platelets respond to activating stimuli, so they aggregate and release their granule contents more readily, which in an artery with existing disease would favour clot formation. The evidence stack for this is one tier of test-tube work, one tier of injected-mouse work, and one small human study measuring platelet behaviour after a single dose. There is no tier showing a clot, a heart attack or a stroke caused by eaten erythritol in a person. A mechanism with a plausible chain and no clinical endpoint is a hypothesis with a good story, and this corpus treats it as exactly that.

The dog mechanism, kept separate on purpose. In dogs, xylitol is a potent stimulus to insulin release from the pancreas, which is why a modest dose crashes their blood sugar and a larger one can destroy liver tissue. Humans do not have this response. Two different species, two different physiologies, one shared molecule.

Risks And Contraindications

• Gastrointestinal symptoms at dose. The best-established adverse effect and the one a regulator acted on. Bloating, rumbling, nausea and, past the threshold, osmotic diarrhoea. Worse in liquid, worse on an empty stomach, worse in combination with other polyols.
• Children and adolescents specifically. The European exposure assessment found this group already exceeding the acceptable daily intake, and body weight is why: the limit scales with size, and a child eating an adult-sized portion of a keto or sugar-free product can clear it in one sitting.
• Irritable bowel syndrome, small intestinal bacterial overgrowth and FODMAP sensitivity. Polyols are a recognised trigger category. Erythritol is the best tolerated of them, which is a relative statement inside a class this group is often advised to limit.
• Existing cardiovascular disease, and the metabolic risk factors that precede it. Not a contraindication and not evidence of harm — an area of genuine uncertainty, now measured both in cardiac patients and in a community population of older adults. The reasonable response is moderation of a discretionary intake, not fear, and never a change to prescribed treatment.
• A note on what is not a risk. There is no evidence that erythritol causes cancer, damages the liver in humans, or causes heart attacks. The platelet finding is a laboratory measurement after a single dose in a handful of healthy volunteers, the clotting result is a mouse, and the brain-cell result is a dish of cultured cells. Anyone telling you otherwise has flattened a rodent, a cell culture or a test tube into a person.
• Dogs and xylitol. Not a human risk at all, and the most concrete hazard in this entry. Low blood sugar within thirty to sixty minutes, potential liver failure at higher doses, deaths documented. Cats are not considered at risk.

Controversy

The nature of the disagreement. Unusually, both sides accept the same primary data. Nobody disputes that blood erythritol predicts cardiovascular events in these cohorts, and nobody disputes that the body makes erythritol and makes more when metabolism is poor. The entire fight is over which of those two facts explains the other. That makes this a dispute about study design rather than about findings, which is both why it is tractable and why it has not been resolved.

Position A: erythritol is a cardiovascular hazard and its safety status should be reconsidered.

Where it has the best evidence. Four cohorts pointing the same way is not nothing, the effect sizes are large for a metabolite, and the 2025 community cohort answers the strongest objection to the first three by finding the association in people who did not have heart disease when they were sampled. The dose-response step is genuinely striking: a single ordinary serving produces a thousandfold plasma rise that persists past two days, which is not a subtle exposure. The mechanistic chain is coherent, and the 2024 human study, with a glucose comparator, showed the platelet effect in living people rather than only in glass. A second polyol, xylitol, produced a parallel signal. And the position has the structural advantage of being the one asking a question that has not been answered, which is a legitimate position for a safety argument to occupy.

Where it overreaches. It routinely presents the mouse, cell-culture and test-tube results as though they were human harm findings, which is the single most common failure in coverage of this topic. It has no clinical endpoint from eaten erythritol in a human, anywhere. It leans on cohorts that never measured dietary intake, using an exposure variable that the body produces in proportion to metabolic dysfunction — including the community cohort, where the people with high levels were also the older, more diabetic, more hypertensive, more kidney-impaired ones, and where the authors' own conclusion was that these metabolites are markers of cardiometabolic health. It generally omits the null genetic study, aided by that study's misleading title. And the human platelet work is small, acute, unblinded as described, and from the same laboratory as the original claim, with no independent replication yet.

Position B: the association is confounded and erythritol is a marker rather than a cause.

Where it has the best evidence. The endogenous-production finding is the strongest single piece of evidence in this entry, it is mechanistic rather than correlational, and it was published before the controversy for an unrelated reason. The genetic study designed specifically to break reverse causation came back null for coronary artery disease. People with lifelong high erythritol from inborn errors of metabolism do not show excess clotting. Most long-term animal feeding studies fail to reproduce the effect. The original cohorts' sampling window partly predates widespread dietary erythritol in the United States, which would make much of the measured exposure self-made by construction. And in the community cohort the downstream metabolite of the body's own processing tracked more outcomes than erythritol itself did, which is what a metabolic-state marker looks like.

Where it overreaches. It slides from "not established" to "debunked," which the literature does not support — the most careful critique explicitly declines to conclude safety. It leans on a null genetic result whose instruments capture endogenous synthesis rather than dietary intake, so the null is much weaker evidence about eating erythritol than it appears. It treats the inborn-errors argument as decisive when it rests on small numbers and conditions that shift several metabolites at once. It has been caught inflating the sampling-window objection beyond what the regulator's document actually says. Its "these were all cardiac patients" line no longer holds now that a community cohort exists. And it has nothing whatsoever to say about the gastrointestinal finding, which is where the actual regulatory action happened.

The funding and bias dimension — cui bono, both ways. On the alarm side, the research programme that produced these findings has a long-running institutional and commercial interest in metabolite-driven cardiovascular risk, including patent and diagnostic interests, and a headline metabolite result extends that programme; media and low-carbohydrate-sceptic commentators monetise "sweetener causes heart attacks" at enormous scale; and the follow-up confirming the original finding came from the same laboratory in a lower-scrutiny venue. On the dismissal side, erythritol is the structural sweetener of an entire packaged-food category, manufacturers and their trade body issued rebuttals within days, and the keto industry has a direct revenue stake in the confound argument winning. Three asymmetries are worth holding on to. First, the key confound evidence was published in 2017 by an unrelated group for an unrelated purpose, so it cannot be dismissed as industry-generated. Second, the most rigorous critique comes from a researcher whose life's work documents sugar harm, and who therefore has no obvious motive to defend a sugar substitute. Third, the community cohort that extends the association comes from a long-running publicly funded epidemiological study with no stake in either answer, and its own authors framed the metabolites as markers. None of those facts settles the question. All of them mean neither side's core evidence can be waved away as bought.

Realised Position: blood erythritol is a real predictor of cardiovascular events — now in ordinary older adults as well as cardiac patients — and probably a poor candidate for a cause of them, and neither of those statements can currently be made with confidence. We do not say erythritol causes heart attacks, because no human clinical outcome has ever been produced by eating it. We do not say it has been cleared, because the study built to clear it tested self-made erythritol rather than eaten erythritol, and because a controlled human study did find a platelet effect that glucose did not produce. Our practical stance rests on a different argument entirely, and a stronger one: the long-term benefit these molecules are sold for has never been demonstrated for polyols in any trial, while the risk side is a growing set of open questions and one confirmed regulatory limit that current intakes already exceed in children. The short-term substitution benefit against sugar is real, and we say so. Unknown long-term risk against unproven long-term benefit is still a weak trade, and moderating a discretionary sweetener costs almost nothing. As a step off sugar, erythritol remains defensible and we will not talk anyone out of it. As a permanent daily fixture in large amounts, it is a bet we would not take.

Cross-Pillar Connections

• artificial_sweeteners_evidence — the class-level hub, and it owns the general argument about non-sugar sweeteners: the weight-control question, the WHO guideline, the microbiome literature, and the bridge-not-destination framing as a general principle. This entry defers all of that and retains only what is specific to the polyols, which the WHO guideline explicitly excludes from its scope.
• cardiovascular_health_management — owns risk assessment and management in full, including lipids, blood pressure, antithrombotic therapy and what someone with established disease should actually be doing. This entry retains only the observation that the erythritol association was measured first in cardiac patients and later in older community-dwelling adults, and that in both settings the people with high levels were the metabolically less healthy ones. It must never generate a treatment implication.
• blood_sugar_regulation — owns glycaemic control as a topic. The connection here runs in an unexpected direction: worse glycaemic control appears to raise the body's own erythritol production, which means the metabolic state a person brings to the question partly determines the biomarker they will show. Managing blood sugar is upstream of this entire debate.
• rct_vs_observational_evidence — the hub for how to read study designs against each other, and the lesson this entry exists to teach in miniature. A large, internally consistent observational signal cannot establish causation when the exposure is plausibly produced by the disease, moving from a referral population to a community population strengthens generalisability without touching that problem, and a genetic design that appears to settle it may be testing a different exposure than the one you eat.
• healthy_user_bias — the general phenomenon of the people who choose an exposure differing systematically from those who do not. It applies here in a specific and important form: the observational harm signals for sweeteners generally, including WHO's, are vulnerable to people at higher cardiometabolic risk being the ones who choose diet products. Applying that scepticism to the sweetener literature but not to the erythritol literature, or the reverse, is the tell of an argument built backwards.
• diet_gut_microbiome — owns fermentation, the FODMAP framework and the gut ecology of unabsorbed carbohydrate. This entry retains only the narrow mechanical point that erythritol's high small-intestinal absorption is what keeps it away from colonic bacteria and therefore what explains its tolerance advantage.
• food_additives_to_avoid — the general additive-risk framing. This entry is a worked example of why that framing needs per-molecule evidence rather than category-level suspicion, since the polyols differ from each other more than they resemble the non-sugar sweeteners they are usually filed alongside.
• sucralose_evidence_and_safety, aspartame_and_the_iarc_classification, food_colourings_evidence_and_claims — sibling entries covering the individual molecules within the sweetener and additive class. Each carries its own evidence and its own controversy; nothing in this entry generalises to them, and nothing in them generalises here.

What would change our mind

Falsifiability: explicit upgrade/downgrade criteria from source

Toward "erythritol is a genuine dietary cardiovascular hazard":
• A randomised controlled trial of dietary erythritol with clinical cardiovascular endpoints, or a large prospective cohort that measures dietary erythritol intake rather than plasma level and finds the association surviving full adjustment for glycaemic status and adiposity. The 2025 community cohort met the population half of this and not the dietary half, which is why it moves the picture without settling it.
• A Mendelian randomisation using instruments for dietary intake rather than endogenous synthesis capacity, returning a positive result for coronary artery disease. This is the specific design gap the current genetic evidence leaves open, and it is the most informative study nobody has run.
• Independent replication of the human platelet finding by a laboratory unaffiliated with the original group, ideally crossover rather than parallel, with a repeated-dose rather than single-dose exposure.
• Long-term animal feeding studies, as opposed to acute injection models, consistently reproducing the thrombosis effect.

Toward "erythritol is safe at ordinary intakes":
• A well-powered prospective cohort measuring dietary intake directly and finding a null association after adjustment for glycaemic status.
• Demonstration that plasma erythritol in modern populations is overwhelmingly endogenous, which would substantially collapse the dietary interpretation of every existing cohort.
• Failure to replicate the human platelet effect in an independent, better-designed trial.
• A regulatory re-evaluation that revisits the gastrointestinal exposure finding with better consumption data and concludes current intakes sit within the limit.

What would NOT move us:
• More in-vitro platelet work, more cell-culture endothelial work, or more acute-injection mouse thrombosis models. These already exist, they are the weakest tier in the stack, and adding to them does not change what tier the argument sits at.
• Another observational cohort measuring plasma erythritol without measuring diet, in any population. That has now been done four times, including once in a community cohort, and the limitation is structural rather than statistical. A fifth would add sample size to a design that cannot answer the question.
• Industry-funded reviews concluding safety on the basis of the null genetic study without addressing the endogenous-versus-dietary instrument problem, or campaigning press releases from either side.
• Another Mendelian randomisation with weak instruments from a small discovery study, in either direction. Instrument quality is the load-bearing part of that design, and results built on inadequate instruments are not evidence regardless of which way they point.
• A regulator declining to act. Absence of regulatory action reflects the evidentiary bar and the lag built into the process, not a positive finding of safety — and note that the two major regulators currently disagree with each other.
• Anything citing the WHO non-sugar-sweetener guideline about erythritol, since the guideline says in its own text that it does not apply to sugar alcohols.

Industry bias note

Structural incentives the evidence base may reflect

The pro-erythritol end. Erythritol is not a minor ingredient with a marketing budget; it is the structural bulk sweetener that makes the entire keto and sugar-free packaged-food category physically possible. It provides volume and mouthfeel that high-intensity sweeteners cannot, which is why it appears in the ingredient list of products whose branding foregrounds monk fruit or stevia. That means the commercial exposure to this question is far larger than the sweetener's public profile suggests, and the response has been correspondingly fast and well organised. The industry's two favoured arguments are both legitimate — the reverse-causation confound and the weakness of rodent thrombosis models as predictors — which is precisely what makes this a difficult bias analysis. The tell is not that the arguments are bad. It is where they stop, and where they stretch. Industry communication reliably converts "not established" into "debunked," cites the null genetic study without mentioning that its instruments measure self-made erythritol, quotes "excreted unchanged in the urine" without the accompanying "converted to erythronate, implications undetermined," inflates the regulator's sampling-window note into bigger numbers than the document contains, and is entirely silent on the European regulator's finding, which the confound argument cannot reach. The "natural, your body already makes it" line deserves a specific flag, because it is the same category error as the "it's ninety-nine percent water" defence of diet drinks: a statement about composition or origin standing in for a statement about effect.

The anti-erythritol end. Less obviously commercial and not therefore clean. The research programme behind the finding has an institutional stake in metabolite-driven cardiovascular risk, including diagnostic and patent interests, and this line of work extends a franchise rather than opening one. That is not misconduct, and the work is published in serious venues; it is the reason independent replication rather than same-laboratory confirmation is the thing to watch for, and it is worth noting that the confirmatory human study appeared in a specialty journal at a lower evidentiary bar than the original. Outside academia, the incentive is much starker. "Sweetener causes heart attacks" is one of the highest-performing health headlines available, and an ecosystem of commentators, supplement sellers offering "clean" alternatives such as allulose and monk fruit, and sugar-positive influencers all convert this story directly into revenue. Every fresh cell-culture result gets a second run at those headlines with the dish quietly removed from the sentence. The competing-sweetener angle is underrated: every category that loses share to erythritol gains from the scare, and those categories fund content.

The clean signal. Four findings in this entry have little or no commercial constituency, and they are the ones to weight most heavily.

The first is the 2017 endogenous-production work. It was designed to study adiposity metabolomics, it was published six years before the controversy, and no sweetener money is anywhere near it. It is the single most important fact in the topic and it arrived for entirely unrelated reasons.

The second is the European regulator's decision. It was issued against an active industry application to remove a warning label, and the panel's conclusion was that the submitted data did not support the request. A regulator declining an application costs the regulator nothing and tells you something real about the quality of what was submitted. That the finding concerns diarrhoea rather than heart attacks is not a weakness — it is why it is the most robust precautionary point available, because no reverse-causation argument reaches it.

The third is the most careful critique of the cardiovascular claim, which comes from a researcher whose published career is built on documenting the harms of sugar and fructose. She has no plausible motive to defend a sugar substitute, and her conclusion is not that erythritol is safe but that safety cannot be concluded. When someone with every incentive to condemn a sweetener declines to condemn it and declines to clear it, that restraint is information.

The fourth is the community-cohort analysis, which comes out of a decades-old publicly funded epidemiological study with no stake in sweeteners at all, and which hands each camp exactly one useful sentence and one inconvenient one.

One structural observation to close. The position this entry lands on — a real association, a confound large enough to explain it, an unresolved question, and moderation on grounds that have nothing to do with the headline — sells nothing to anyone. No product follows from it, no headline is generated by it, and neither industry can quote it. In a topic this commercially contested on both sides, a position with no constituency is usually the one that survived the incentives rather than being shaped by them.

Sources (16)

MODERATE (unresolved cardiovascular question)Last updated August 2026
Open in the Library: search, filter, every entry →

We set no cookies and run no ad trackers. We count visits with Cloudflare's cookieless, privacy-first analytics. The only thing stored on your device is which example you last viewed.