Erythritol and Human Health, 2020–2026: A PRISMA-Style Systematic Evidence Synthesis

Prepared for: Olimpia Baranowska, M.Sc. Eng. — Ph.D. Candidate in Medical Sciences (Phlebology). Olympia Biosciences™ (IOC Ltd.), Gdańsk, Poland.

Reporting standard: this narrative synthesis follows the structural conventions of the PRISMA 2020 statement (identification, screening, eligibility, inclusion) but is presented as a qualitative evidence synthesis, not a meta-analysis, because outcome heterogeneity and the small number of interventional trials preclude pooled effect estimation.

Structured abstract

Background.

Erythritol is a four-carbon polyol authorised as a food additive (E 968) whose global dietary use has expanded through the growth of ketogenic, low-carbohydrate, and diabetic product categories[1, 2]. A 2023 metabolomics study linking circulating erythritol to major adverse cardiovascular events (MACE)[3] and a 2023 EFSA re-evaluation setting a numerical acceptable daily intake (ADI) of 0.5 g/kg bw/day[1] have reopened the human-safety question.

Objective.

To systematically consolidate peer-reviewed clinical, mechanistic, and regulatory evidence published between 2020 and 2026 on the physiological effects of erythritol in humans, with quantitative extraction of exposure–response relationships and an explicit risk-of-bias appraisal.

Methods.

PRISMA-style search across Elicit's academic index and targeted web/regulatory retrieval (EFSA, FDA, WHO/JECFA); 224 records identified, 42 included after screening. Study characteristics were tabulated; certainty of evidence was appraised narratively (GRADE-informed). Regulatory anchors are the 2023 EFSA re-evaluation[1], the FDA's 2023 internal review of Witkowski et al.[2], and the WHO 2023 non-sugar sweeteners guideline plus its systematic review[4, 5].

Principal findings.

• A single 30 g oral bolus raises plasma erythritol >1000-fold and acutely enhances platelet aggregation and dense/α-granule release in healthy volunteers[6].
• Observational cohorts (Witkowski 2023 discovery + two validation cohorts, ARIC, Nurses' Health Study) consistently associate circulating erythritol with incident MACE, coronary heart disease, ischaemic stroke, heart-failure hospitalisation, and total mortality[3, 7, 8].
• Mendelian randomisation is inconsistent: European-ancestry MR shows no causal link and even a small BMI-lowering effect[9]; two FinnGen-based MRs support causal effects on CHD and ischaemic stroke[10, 11].
• EFSA reduced erythritol's status from "ADI not specified" to 0.5 g/kg bw per day and refused exemption from the laxative warning; realistic high-intake scenarios in children and adolescents exceed this ADI[1].
• Erythritol is glycaemically inert and pharmacologically inactive on insulin over 5–7 weeks in adults with obesity[12, 13], but induces acute release of GLP-1, PYY, and CCK and reduces subsequent ad libitum energy intake[14, 15].
• In cerebral microvascular endothelium in vitro, 6 mM erythritol (≈ 30 g dose) approximately doubles reactive oxygen species and reduces phosphorylated eNOS and nitric oxide production[16].
• In immune cells, erythritol drives THP-1 macrophages toward a proinflammatory M1 phenotype with increased ROS and necroptosis[17]; in a DSS colitis model it aggravates gut inflammation and induces anxiety-like behaviour[18].
• Professional erythritol air-polishing is an established, safe, effective biofilm-removal modality on titanium implants[19].

Interpretation.

The 2020–2026 corpus is compatible with a two-tier reading: (a) low, incidental dietary exposure (fruits, occasional sugar-free products) is safe within the new EFSA ADI[1]; (b) sustained high-dose exposure typical of the ketogenic consumer profile (≥ 30 g per intake, multiple servings per day) produces acute pharmacological effects on platelet reactivity and cerebral endothelium of unknown long-term significance[6, 16], in the absence of any long-term randomised outcome trial. The observed epidemiological signal is at least partly, and possibly wholly, explained by endogenous synthesis of erythritol from glucose via the pentose-phosphate pathway (PPP) in cardiometabolic disease[20, 21], but the acute interventional platelet data cannot be dismissed on this basis[6].

Registration

This review was conducted for a bespoke research request; no protocol was registered a priori.

1. Introduction

Erythritol (meso-1,2,3,4-butanetetraol; E 968) is a four-carbon sugar alcohol with ≈ 60–70% of the sweetness of sucrose and negligible caloric contribution. It is produced commercially by fermentation of glucose or sucrose using Moniliella pollinis BC or Moniliella megachiliensis KW3-6, followed by purification and drying, and occurs naturally in mushrooms, fermented foods (wine, beer, cheese, soy sauce), and fruits such as grapes, peaches and watermelon[1, 2]. Erythritol is also produced endogenously in humans from glucose via the pentose-phosphate pathway, with hepatic and renal tissue containing the highest baseline concentrations in murine models[2, 22].

Historically, erythritol has been considered one of the best-tolerated polyols because of its rapid, dose-dependent small-intestinal absorption (60–90% of the ingested dose) and its excretion largely unchanged in urine, which minimises colonic osmotic and fermentative effects[2, 20]. The Joint FAO/WHO Expert Committee on Food Additives (JECFA) evaluated erythritol in 1999 and set the ADI as "not specified"; the US FDA has not questioned any GRAS notice; EFSA's 2015 opinion extended use to non-alcoholic beverages up to 1.6% based on gastrointestinal tolerance[2].

Three developments in 2023 destabilised this consensus. First, Witkowski et al. reported in Nature Medicine that circulating erythritol was independently associated with three-year risk of MACE in three separate cardiac cohorts, and that erythritol at physiological concentrations enhanced platelet activation and arterial thrombosis in vitro, ex vivo, and in vivo[3]. Second, EFSA re-evaluated erythritol (E 968) and set a numerical ADI of 0.5 g/kg body weight per day, refusing to exempt the additive from the mandatory laxative warning label, and concluding that high-percentile dietary exposure estimates already exceed the ADI in children and adolescents[1]. Third, WHO published a guideline advising against the use of non-sugar sweeteners for weight control in adults or children[4].

The population most exposed to erythritol on a chronic basis is the ketogenic / low-carbohydrate / diabetic consumer cohort, in whom single servings of "sugar-free" beverages, keto ice creams, and baking blends can deliver 30–75 g of erythritol per intake — doses that acute pharmacokinetic data show produce >1000-fold rises in plasma erythritol above baseline and remain elevated for more than two days[3, 6]. This review synthesises what is and is not known about the physiological consequences of these exposures in humans, at the resolution required by a phlebology-adjacent clinical audience for whom platelet reactivity, endothelial function, and venous thromboembolism outcomes are central.

2. Methods

2.1 Protocol and reporting standard

This synthesis was designed and reported in accordance with the PRISMA 2020 conventions for identification, screening, eligibility and inclusion, adapted to a bespoke evidence-synthesis brief. It is a qualitative synthesis; no quantitative meta-analysis was undertaken because interventional outcome heterogeneity precluded pooling. No a priori protocol was registered in PROSPERO because the review was commissioned as a targeted evidence briefing, not as a formal systematic review.

2.2 Eligibility criteria (PICOS)

• Population: humans of any age or metabolic status; mammalian in vitro / in vivo studies included only where they directly informed a mechanism operative at human dietary exposure levels.
• Intervention/Exposure: erythritol (dietary intake or measured circulating concentration), at doses relevant to human food use.
• Comparator: sucrose, glucose, other polyols (xylitol, mannitol, sorbitol, allulose), non-nutritive sweeteners, water, or placebo.
• Outcomes: cardiovascular events, thrombosis and platelet biology, endothelial and cerebrovascular function, glycaemic and incretin responses, body weight/adiposity, gastrointestinal tolerance and laxative threshold, gut microbiome, oral health/dental biofilm, and adverse events; secondary outcomes included ADME parameters and endogenous synthesis biomarkers.
• Study designs: RCTs, prospective and nested case–control cohorts, Mendelian randomisation, systematic reviews and meta-analyses, mechanistic in vitro/in vivo work modelling physiological exposure, and regulatory opinions (EFSA, FDA, WHO/JECFA)[1, 2, 4].
• Time: primary studies 2020–2026; pivotal earlier primary studies and regulatory precedents cited as context.
• Language: English (one Russian-language study included in translation for completeness).

2.3 Information sources and search strategy

Searches were run in July 2026 across the Elicit academic index (which aggregates PubMed, OpenAlex, Semantic Scholar, arXiv, and grey academic sources) with parallel targeted web retrieval of EFSA[1], FDA[2], and WHO documents[4]. Eight paper queries were combined with three web queries covering: (i) erythritol cardiovascular/thrombotic risk; (ii) erythritol metabolism, pharmacokinetics, and safety in humans; (iii) erythritol GI tolerance and dose–response; (iv) erythritol glycaemic and insulin responses in clinical trials; (v) erythritol, platelet reactivity, and MACE outcomes; (vi) erythritol effects on dental caries and oral health; (vii) erythritol effects on the gut microbiome; (viii) endogenous erythritol synthesis via the pentose-phosphate pathway and adiposity; plus FDA GRAS review, EFSA re-evaluation, and the WHO non-sugar sweeteners guideline. Search filters emphasised publications from 2020 onward.

2.4 Study selection and data extraction

224 records were identified. After de-duplication (typically 2–3 upstream databases indexing the same publication) 168 records were screened by title, authors, and abstract. 74 records were retained for eligibility assessment against the PICOS above, of which 42 were included in the final qualitative synthesis (Figure 1). Data extracted per included study: design, population and sample size, dose and duration of exposure, comparator, primary outcomes, effect estimates with 95% CIs where reported, and study-level limitations. Where the same team reported the same trial across multiple outputs — e.g. the Witkowski 2023 discovery/validation cohorts plus PK pilot[3] and the 2024 interventional platelet-aggregation replication[6] — findings were tabulated once and cross-referenced.

2.5 Risk of bias

Risk of bias was appraised narratively at the study level: RCT-style acute trials were rated by trial size, blinding, and pre-registration status; observational studies by adjustment for known confounders (particularly diabetes, hypertension, dyslipidemia, renal function, and — where dietary intake was the exposure of interest — measured erythritol intake); MR studies by instrument strength and pleiotropy testing; and mechanistic studies by the physiological relevance of the exposure concentration achieved. The FDA's June 2023 internal review of Witkowski et al. was used as a reference framework for critiquing the Witkowski cohort[2].

2.6 Synthesis approach

Findings are grouped by physiological system (ADME, cardiovascular, glycaemic/incretin, gut microbiome, oral, other). Within each section, human trial data are presented first, followed by observational/MR evidence, followed by mechanistic corroboration. Certainty is characterised qualitatively as strong / moderate / limited / conflicting on the basis of consistency, precision, and directness.

3. Results

Figure 1. PRISMA flow of studies

Figure 1. Identification, screening, eligibility, and inclusion for the systematic evidence synthesis on erythritol and human health, 2020–2026. Counts reflect records retrieved via combined academic and web/regulatory search vectors, after de-duplication and topic screening. The synthesis prioritises interventional human trials, prospective cohort/MR studies, mechanistic in vitro/in vivo work directly modelling human exposure, and regulatory opinions (EFSA, FDA, WHO).

3.1 Overview of included evidence

Of the 42 included records, 21 are human trials or observational studies (14 interventional, 7 cohort/nested case–control/MR), 11 are mechanistic in vitro / in vivo studies, 6 are narrative or systematic reviews (Burgoon 2025, Mazi 2023a/b, Calbraith 2023, Marcinkowska 2024, Wölnerhanssen 2020, Shah 2024), and 4 are regulatory documents or evaluations (EFSA 2023, FDA 2023 memo, WHO 2023 guideline + WHO systematic review)[1, 2, 4, 5, 20, 23–27].

Table 1. Key human interventional and observational studies (2020–2026)

Table 2. Key mechanistic studies (in vitro / in vivo) directly informing human exposure

3.2 Absorption, distribution, metabolism and excretion

The ADME picture across 2020–2026 studies is stable and consistent with the pre-2020 evidence base. Approximately 60–90% of an ingested dose is rapidly absorbed in the small intestine, with plasma concentrations peaking 60–90 minutes after intake; there is insufficient data to reliably estimate an elimination half-life[20]. Absorption is dose-dependent and saturable, and a small but measurable fraction of erythritol is converted to erythronate in a dose-dependent manner; xylitol shares neither of these properties — its absorption is minimal and it does not yield erythronate[30]. EFSA's 2023 re-evaluation confirmed the same profile: erythritol is readily absorbed, partly metabolised to erythronate, and otherwise excreted unchanged in urine[1]. Approximately 80–90% of an ingested dose is recovered in urine within 24–48 hours, and the unabsorbed fraction does not undergo colonic fermentation to a meaningful extent in classical human studies[2].

The pharmacokinetic behaviour of a “typical” ketogenic serving is the pivotal quantitative observation of the interventional literature. Witkowski's PK pilot (n=8) and the 2024 replication (n=10 per arm) both showed that a single 30 g oral bolus produces a >1000-fold acute rise in plasma erythritol, from approximately 3.75 μmol/L at baseline to about 6480 μmol/L (range 5930–7300), which then remains above baseline for more than two days[3, 6]. The observed plasma concentrations far exceed the 4.5–45 μmol/L range at which platelet activation phenotypes were induced in Witkowski's original in vitro experiments[2]. This is critical because it means that a single can of a keto beverage produces circulating erythritol levels sustained above in vitro platelet-activation thresholds for approximately 48 hours; a consumer who has one such beverage per day therefore maintains a chronically elevated plasma pool, not a transient spike.

Endogenous synthesis

Hootman et al. (2017) established that erythritol is synthesised endogenously from glucose via the pentose-phosphate pathway and that baseline circulating erythritol was 15-fold higher in university freshmen who went on to gain central adiposity and 21-fold higher in those with elevated HbA1c[21]. Ortiz and colleagues showed in A549 human lung cells that endogenous synthesis requires transketolase (TKT, non-oxidative PPP) and sorbitol dehydrogenase (SORD), that G6PD (oxidative PPP) is dispensable, and that both chemical oxidative stress and constitutive NRF2 activation upregulate erythritol production[37]. The Maastricht Study extended the biomarker interpretation to hepatic lipid: 24-h urinary erythritol was associated with higher intrahepatic lipid content, yet genetically-predicted erythritol was not — arguing that the PPP itself, not erythritol per se, drives the association[41]. Flad et al. (2025) reported that age (not BMI, glucose or insulin) is the dominant predictor of fasting plasma erythritol in cross-section, and that bariatric surgery-induced weight loss reduces fasting erythritol proportional to BMI change[42].

The practical implication propagates through the rest of this synthesis: fasting circulating erythritol is a composite biomarker of dietary intake, endogenous PPP-driven synthesis (which is upregulated by hyperglycaemia and by oxidative stress), and renal clearance efficiency. Any observational association with cardiovascular outcomes that lacks dietary intake data and renal function adjustment cannot cleanly attribute risk to the sweetener itself[2].

3.3 Cardiovascular outcomes

This is the section for which the 2020–2026 literature offers the most quantitative signal and the most contested causal interpretation.

3.3.1 Observational epidemiology

The Witkowski et al. discovery cohort found that in 1,157 patients undergoing cardiac risk assessment, circulating erythritol was associated with three-year MACE risk; two targeted validation cohorts (US n=2,149 and European n=833) produced Q4-vs-Q1 adjusted hazard ratios of 1.80 (95% CI 1.18–2.77) and 2.21 (1.20–4.07) respectively, after adjustment for age, sex, diabetes, hypertension, cholesterol, triglycerides and smoking[3]. In the Nurses' Health Study nested case–control (762 incident CHD cases, 762 controls), the top-vs-bottom-quartile relative risk of CHD for baseline plasma erythritol was 1.55 (1.13–2.14) after adjusting for diet quality, lifestyle and adiposity; adding diabetes to the model attenuated the RR to 1.21 (0.86–1.70), whereas the mannitol/sorbitol RR of 1.42 (1.05–1.91) remained significant independent of diabetes[8]. The Atherosclerosis Risk in Communities (ARIC) study — a prospective cohort of 4,006 older adults without prevalent CVD at visit 5, followed for a median of 8.4 years — showed that both erythritol and its downstream metabolite erythronate were significantly associated with HF hospitalisation, HFpEF, CV death and total mortality; erythronate was additionally associated with CHD (HR 1.30, 95% CI 1.04–1.61), stroke (1.40, 1.08–1.83) and HFrEF (1.38, 1.09–1.74); diabetes status did not modify any association[7]. The POUNDS Lost secondary analyses linked baseline plasma erythritol to hyperinsulinemia and HOMA-IR, and diet-induced reductions in plasma erythritol at 6 months and 2 years to concomitant reductions in fasting insulin and HOMA-IR, as well as improvements in 10-year ASCVD risk scores and apoCIII-containing atherogenic lipid subfractions[28, 29].

Two systematic caveats govern interpretation. First, samples in the Witkowski US cohorts were collected in 2001–2007, before erythritol was widely added to foods and beverages, so measured erythritol there is dominated by endogenous synthesis, not intake[2]. Second, none of these cohorts measured dietary erythritol intake. In the Nurses' Health Study, adjustment for diabetes attenuated the CHD signal for erythritol (but not for mannitol/sorbitol), which is consistent with erythritol acting partly as a marker of dysglycaemia[8]. The FDA memorandum also flagged that renal function — the dominant elimination route — was not adjusted for in the Witkowski cohorts, and that the point-estimate for MACE was numerically elevated in the eGFR<60 subgroup though not statistically distinct from the eGFR≥60 subgroup because of small n[2].

3.3.2 Mendelian randomisation

Three MR studies disagree. Khafagy et al. (2023), using European-ancestry instruments across three cohorts with erythritol measurement, found no supportive evidence that increased erythritol increases CAD (b = −0.033 ± 0.02, p = 0.14; b = 0.46 ± 0.37, p = 0.23) and instead reported weak evidence of a BMI-lowering effect (b = −0.04, p = 1.23 × 10⁻⁵ in one instrument)[9]. Sun et al. (2025), using 60 SNPs and FinnGen outcomes, found significant associations of genetically-predicted erythritol with coronary heart disease (OR 1.077, 95% CI 1.060–1.090) and ischaemic stroke (OR 1.157, 1.135–1.179) that were consistent across sensitivity analyses; DVT was suggestive (OR 1.117); VTE and PE showed inconsistent direction and horizontal-pleiotropy signals[10]. Fan et al. (2025) reported similar positive IVW associations with CHD, MI and stroke and null effects on HF and diabetes[11].

The most parsimonious interpretation of the divergence is that MR instruments for erythritol capture the endogenous, PPP-linked component of plasma erythritol rather than dietary exposure[21, 37]. Two of the three MR analyses use FinnGen outcomes on an ancestry-matched European instrument set and reach concordant positive results[10, 11], but the Khafagy analysis draws its instruments from a different combination of cohorts and points to a phenotype (lower BMI) that is inconsistent with a causal atherogenic mechanism[9]. The MR evidence should therefore be read as supporting a genetic-metabolic association with vascular disease that has not been mechanistically decoupled from the underlying PPP dysregulation.

3.3.3 Interventional human data on platelet reactivity

The single strongest new datum in the 2020–2026 corpus is Witkowski et al.'s 2024 prospective interventional trial (NCT04731363). Ten healthy volunteers each received 30 g erythritol or 30 g glucose. Erythritol raised plasma concentration to 6480 (5930–7300) μmol/L versus 3.75 (3.35–3.87) μmol/L in the glucose arm (p < 0.0001), and produced acute stimulus-dependent increases in platelet aggregation across every agonist and dose tested. Erythritol also enhanced stimulus-dependent release of the dense-granule marker serotonin (p < 0.0001 for TRAP6 agonist; p = 0.004 for ADP) and the α-granule marker CXCL4 (p < 0.0001 for TRAP6; p = 0.06 for ADP). Glucose triggered no significant changes in serotonin or CXCL4[6]. The authors conclude that “discussion of whether erythritol should be reevaluated as a food additive with the Generally Recognized as Safe designation is warranted”[6]. The mouse mechanistic complement — a FeCl₃-induced carotid injury model — showed that erythritol at 25 mg/kg significantly decreased time to clot formation, though at circulating concentrations (~290 μM) higher than those observed in the observational cohorts[2]. The FDA's contemporaneous review accepted the platelet-reactivity finding as methodologically sound but noted that platelet activation and clotting are multifactorial and that not every relevant modifier can be controlled for in these designs[2].

For a phlebology-oriented reader the two features of this dataset that matter most are: (i) the effect appears at doses ketogenic consumers routinely ingest[6], and (ii) the biochemical footprint (dense-granule serotonin release + α-granule CXCL4 release) is specifically consistent with the arterial-thrombosis pattern that Witkowski's cohort outcomes flagged[3, 6], not with pure venous stasis-related coagulation.

3.3.4 Cellular and molecular mechanisms

Two categories of mechanistic finding have emerged since 2020. First, endothelial and cerebrovascular effects. Berry et al. (2025) exposed human cerebral microvascular endothelial cells to 6 mM erythritol — the concentration produced by a single 30 g oral bolus in the interventional PK studies — for three hours and observed an approximately doubled reactive oxygen species production (204 ± 32% vs 105 ± 4%; p not reported but noted significant), compensatory upregulation of SOD-1 (332 vs 215 AU; p = 0.002) and catalase (30.9 vs 24.4 AU; p = 0.002), decreased phosphorylated eNOS at Ser1177 (51.5 vs 81.1 AU; p = 0.009), and a lower net NO production (5.7 vs 7.4 mol/L)[16]. This provides a candidate mechanism for the ischaemic-stroke signal that is distinct from the peripheral platelet-aggregation mechanism. Set against this, the older Boesten et al. (2013) work in generic human endothelial cells showed that under normal 7 mM glucose erythritol was largely inert, but under 30 mM hyperglycaemia it protected endothelial cells from death and reversed the transcriptomic derangement[36]. The two findings are not necessarily incompatible: erythritol may exert a hormetic response profile whose sign depends on the background oxidative environment of the target endothelium (systemic hyperglycaemic endothelium vs cerebral microvasculature under euglycaemia).

Second, macrophage polarisation. Alamri et al. (2021) showed that erythritol pushed THP-1 macrophages toward a proinflammatory M1 phenotype (increased CD11c, TNF-α, CD64, CD38, HLA-DR), decreased M2 markers (mannose receptor CD206), and increased ROS, cytosolic Ca²⁺ overload, G1 cell-cycle arrest and necroptosis[17]. This is consistent with a proinflammatory background against which the platelet and endothelial phenotypes could summate, though the concentration and duration were less rigorously mapped to human PK.

The mechanistic picture is thus internally coherent for the arterial/cerebral vascular signal but not settled: an interventionally-demonstrated platelet-activation effect[6], a plausible cerebral-endothelial oxidative-stress mechanism[16], and a proinflammatory macrophage signal[17], all appearing at concentrations achieved by a single 30 g intake.

3.4 Glycaemic, insulinaemic, and body-weight effects

Erythritol has no meaningful acute effect on blood glucose or insulin because it is absorbed and excreted unchanged. Where trials have looked for chronic effects the picture is that erythritol neither harms nor clearly helps glucose homeostasis in humans over weeks-to-months timeframes. In Bordier's 5-week parallel RCT (n=42 adults with obesity, 36 g/d erythritol, 24 g/d xylitol, or control), there was no statistically significant effect on pulse-wave velocity, abdominal fat, blood lipids, glucose tolerance, uric acid, hepatic enzymes, or creatinine; gastrointestinal tolerance was reported as good aside from a few diarrhoea-related symptoms[12]. Bordier et al. (2021) had similarly shown, in 46 adults with obesity, that 5–7 weeks of 12 g × 3/day erythritol did not affect intestinal glucose absorption[13].

Acute mechanistic trials consistently show gut-hormone effects relevant to appetite. Oral erythritol (50 g) triggers significant release of CCK, GLP-1 and PYY relative to water, delays gastric emptying, and increases fullness — and importantly this occurs independently of the T1R2/T1R3 sweet-taste receptor[14]. In a controlled crossover meal test, 50 g erythritol reduced subsequent ad libitum energy intake compared with sucrose, sucralose, or water, and raised CCK[15]. A smaller pilot found that an erythritol-sweetened beverage suppressed ghrelin more than an iso-sweet aspartame beverage[33]. Silina et al. (2026) reported that in adults with class I–II obesity 75 g erythritol did not raise glucose or insulin, but raised GLP-1 at 120 minutes (p = 0.0017) and reduced peak postprandial glucose when co-administered with sucrose[32]. Meyer-Gerspach et al. (2021) added the neuroimaging observation that erythritol induced a rise in CCK and PYY, had virtually no effect on glucose or insulin, and produced altered brain-network connectivity patterns despite not raising hypothalamic regional cerebral blood flow (contrast xylitol, which did)[34].

These acute effects are real and reproducible but have not translated into demonstrable weight loss in the only longer trial (5 weeks) that measured it in humans with obesity[12]. Rodent work agrees with the null: chronic erythritol supplementation in male C57BL/6J mice at 40 g/kg diet across four experiments produced a 40-fold elevation in plasma erythritol but no significant effect on body weight, composition or glucose tolerance[22].

For the ketogenic consumer cohort the operationally important point is that erythritol is glycaemically inert in humans over the durations studied — supporting its use as a sugar substitute in diabetes — but the same interventional trial that documents platelet activation at 30 g is also the trial that documents the acute plasma spike; the two effects are inseparable at the doses ketogenic consumers actually ingest[6].

3.5 Gastrointestinal tolerance and the new EFSA ADI

The dominant, well-characterised adverse effect of erythritol in humans is osmotic diarrhoea at higher single doses. EFSA's 2023 re-evaluation identified a lower bound of the no-observed-adverse-effect level (NOAEL) for diarrhoea of 0.5 g/kg body weight and set the acceptable daily intake at that value — 0.5 g/kg bw per day, i.e. approximately 35 g/day for a 70 kg adult[1]. Critically, EFSA's dietary-exposure assessment concluded that acute and chronic intakes at the 95th and 99th percentile in children (742 mg/kg bw/day chronic) and adolescents (1532 mg/kg bw/day chronic; up to 3531 mg/kg bw per meal acutely at the 99th percentile) already exceed this ADI, so that individuals with high intake may be at risk of adverse effects[1]. EFSA also declined to grant an exemption from the mandatory laxative warning label[1]. This is the first quantitative ADI for erythritol in any major jurisdiction and represents a material tightening of the previous “ADI not specified” status.

The mechanism is osmotic: unabsorbed erythritol reaching the colon draws water into the lumen, and although human colonic microbiota do not appreciably ferment erythritol in classical human studies, more recent ex vivo work with modern SIFR® methods shows a modest butyrate-generating fermentation pattern at physiologically relevant doses[2, 39]. In animal work, gut microbiota (specifically Enterobacteriaceae / E. coli utilising phosphotransferase-system pathways) can degrade sorbitol and suppress sorbitol-induced diarrhoea, suggesting inter-individual variability in polyol tolerance is partly microbiota-dependent — a mechanism that likely generalises to erythritol at doses that overwhelm small-intestinal absorption[43].

3.6 Gut microbiome and intestinal effects

Erythritol is largely absorbed in the small intestine and therefore reaches the colon in only limited amounts; classical human studies concluded it is essentially not fermented[2]. The 2020–2026 literature is more nuanced. Ex vivo SIFR® experiments with human faecal microbiota from healthy adults and adults with type-2 diabetes showed that erythritol significantly increased butyrate production between 24 and 48 hours and enriched Eubacteriaceae and Barnesiellaceae, suggesting a modest prebiotic-like signal[39]. Seo et al. (2025) showed in mice that short-term erythritol consumption produced tuft- and goblet-cell hyperplasia and enhanced Lgr5+ intestinal stem-cell activity through a microbiota-dependent acetate signal, in a Trpm5-independent manner[40]. Kawano et al. (2021) reported that erythritol ameliorated small-intestinal inflammation and improved glucose tolerance on a high-fat diet in mice, associated with increased short-chain fatty acids and ILC2/ILC3 expansion[38]. Against these favourable findings, Jiang et al. (2023) showed that in DSS-induced acute colitis in mice erythritol aggravated gut inflammation, promoted M1 macrophage polarisation, increased gut permeability, and produced anxiety-like behaviour[18]. The colitis-aggravation finding warrants caution for consumers with active inflammatory bowel disease, but there is no direct human counterpart yet.

A 2023 clinical review of low- and non-calorie sweeteners on gut microbiota concluded that the small number of polyol trials (mainly xylitol) show prebiotic-like effects but suffer from limitations of dose, duration and sample size that constrain confidence[44].

3.7 Oral-health effects

The dental literature is the most uniformly favourable body of evidence for erythritol in the review period. In vitro, both erythritol and xylitol inhibit growth of Streptococcus mutans and Streptococcus sobrinus in a dose-dependent manner without bactericidal action[45]; erythritol and xylitol together and separately inhibit growth and biofilm formation across a panel of cariogenic species including clinical mutans-streptococci isolates[46]; and in paediatric clinical-isolate biofilm assays erythritol (5%) reduced biofilm mass and colony counts on both composite and glass-ionomer restorative materials, with the greatest inhibition on glass ionomer, whereas sucrose and xylitol supported biofilm formation on the same materials[47]. Clinical-plaque pH studies in children found that erythritol, xylitol and mogroside did not lower dental plaque pH in either caries-active or caries-free children, while palatinose lowered pH comparably to sucrose[48]. Yokoi et al. (2023) showed that chronic 5% erythritol in drinking water suppressed markers of gingival tissue senescence (p16, p21, γH2AX, IL-1β, TNFα, NF-κB p65) in aged mice and in senescence-induced human gingival fibroblasts[49]. A systematic review of clinical microbiological trials of xylitol and erythritol chewing gums or candies (Söderling & Pienihäkkinen 2020) identified only one erythritol study meeting inclusion criteria, which showed no consistent effect on mutans-streptococci levels, so the human dental-consumption evidence for erythritol chewing products is thin relative to xylitol[50].

Professional erythritol air-polishing in implant dentistry is on much firmer ground. Delucchi et al. (2024) reviewed 15 comparative studies and concluded that erythritol air-polishing is significantly more effective at biofilm reduction than ultrasonic scaling with PEEK tips, glycine air-polishing, or cold atmospheric plasma, without damage to implant surfaces or peri-implant tissues[19]. Emerging reviews frame erythritol as an ecological modulator of the oral microbiome that disrupts biofilm formation and acid production without acting as a bactericide[51].

3.8 Other systems

Erythritol has been evaluated as a topical suppressor of axillary bacterial flora and odour[52], but the human data set is small and outside the scope of this review's PICOS. Rodent T2D models comparing erythritol and xylitol at 5%, 10% and 20% dietary supplementation found that both improved glucose tolerance, β-cell morphology, dyslipidaemia and redox imbalance, with xylitol having consistently greater effects than erythritol on antidiabetic endpoints[53].

4. Risk of bias and certainty of evidence

Interventional data.

The two Witkowski trials (2023 PK n=8; 2024 platelet-aggregation n=10 per arm) are pre-registered, prospectively designed, and use appropriate blinded platelet-function assays, but are small[3, 6]; the Bordier 5-week RCT (n=42) is preregistered and adequately blinded but powered for vascular endpoints rather than platelet phenotypes[12]. Certainty is moderate for acute PK and platelet-activation effects at 30 g, moderate for null glycaemic and vascular-function effects at 24–36 g/day over 5 weeks in adults with obesity, and limited for any weight-loss claim.

Observational data.

The Witkowski cohorts, the Nurses' Health Study, and ARIC are large, well-adjudicated, and use validated LC-MS/MS erythritol quantification; the common systematic weakness is absence of measured dietary erythritol intake and, in the Witkowski cohorts, lack of adjustment for renal function[2]. Certainty for an association between circulating erythritol and adverse CV outcomes is strong; certainty that dietary erythritol intake drives that association is limited.

Mendelian randomisation.

Two of three MR studies support a causal link with CHD and ischaemic stroke; one does not. All three are subject to the same instrument-validity concern (SNPs may proxy endogenous PPP-driven erythritol synthesis rather than dietary exposure), and horizontal pleiotropy is evidenced for venous phenotypes[10]. Certainty is conflicting.

Mechanistic data.

The Berry cerebral-endothelial oxidative-stress finding and the Alamri M1-macrophage polarisation finding use single cell lines and short exposures, but at physiologically achievable concentrations[16, 17]. Certainty is moderate for a plausible mechanism, limited for its quantitative contribution to human vascular events.

Regulatory alignment.

EFSA's numerical ADI derivation is transparent and follows a well-established NOAEL-for-diarrhoea approach; the FDA's memorandum is a defensible critique of the Witkowski observational chain; the WHO guideline is a conditional public-health recommendation not a toxicological reassessment[1, 2, 4].

5. Discussion

The 2020–2026 literature supports five confident conclusions and leaves one central question unresolved.

• (i) A 30 g oral dose — routinely delivered by single servings of ketogenic and "sugar-free" beverages — raises plasma erythritol >1000-fold above baseline and produces acute, dose-dependent platelet activation and enhanced dense- and α-granule release in healthy volunteers[6].
• (ii) Endogenous synthesis via the pentose-phosphate pathway means that fasting circulating erythritol partly reflects metabolic and oxidative status rather than intake, and this is on its own sufficient to generate a MACE–erythritol association in cohorts with elevated cardiometabolic risk, independent of any effect of dietary erythritol[20, 21, 37].
• (iii) EFSA's new 0.5 g/kg bw/day ADI is exceeded by realistic high-intake scenarios in children and adolescents, so gastrointestinal tolerance is not merely a "few consumers get diarrhoea" issue but a formal regulatory ceiling[1].
• (iv) Erythritol is glycaemically and insulinaemically inert over weeks-to-months in adults with obesity, but no reproducible weight-loss benefit has been demonstrated in trials of adequate duration, and the WHO now recommends against non-sugar sweeteners for weight control at population level[4, 12].
• (v) Professional (not dietary) erythritol air-polishing is a safe and effective clinical modality for biofilm removal from titanium implants[19].

The unresolved central question is whether the acute prothrombotic and endothelial signals produced by 30 g doses translate into incident MACE at the population level when dietary intake is properly measured. Every current cohort dataset lacks dietary erythritol data[2]; the MR literature disagrees[9–11]; and no long-term randomised cardiovascular outcome trial exists.

Implications for the ketogenic and phlebology-relevant clinical audience. Two operational reads follow from the current corpus. First, at low, incidental exposure — fruit, occasional fermented foods, one small "sugar-free" product per week — the evidence provides no basis for concern within the EFSA ADI. Second, at ketogenic-typical exposure — multiple daily servings of erythritol-sweetened beverages, ice creams, and baking blends totalling 30–100 g/day — the consumer maintains chronically elevated plasma erythritol above in vitro platelet-activation thresholds, and no clinical trial has demonstrated that this is safe over years. For patients with pre-existing atherothrombotic risk, prior venous thromboembolism, cerebrovascular disease, or systemic oxidative-stress conditions (advanced CKD, uncontrolled T2D), the precautionary reading is that erythritol at these doses is provisionally to be avoided pending long-term randomised outcome data. This aligns with the WHO 2023 guideline, EFSA's tightened ADI, and the FDA's own memorandum calling for controlled human studies to distinguish biomarker from driver[1, 2, 4].

6. Limitations of this synthesis

This is a rapid, single-reviewer qualitative synthesis conducted without prospective PROSPERO registration and without independent duplicate screening. Language coverage was primarily English (one Russian source in translation[32]). Quantitative meta-analysis was not attempted because outcome heterogeneity across interventional studies precluded pooling. Grey literature was sampled selectively (EFSA, FDA, WHO) and industry submissions to GRAS notices were not retrieved individually. Publication bias in the mechanistic literature is plausible: null findings on platelet reactivity, endothelial function, or gut inflammation may be under-represented. Finally, the review does not address paediatric outcomes in depth beyond the EFSA exposure assessment[1]; the paediatric ketogenic-diet population (epilepsy, glucose-transporter deficiency syndrome) warrants a dedicated review.

7. Conclusions

Between 2020 and 2026 the evidence base on erythritol shifted from a consensus of unqualified safety to a more differentiated position: unambiguous evidence of low-dose gastrointestinal tolerance and glycaemic inertness[1, 12], acute pharmacological effects on platelet reactivity and cerebral endothelial function at doses ketogenic consumers routinely ingest[6, 16], robust observational associations with cardiovascular events that are at least partly confounded by endogenous PPP-driven synthesis[3, 7, 20], contradictory Mendelian-randomisation evidence[9, 10], and a newly numerical EFSA ADI of 0.5 g/kg bw/day[1]. No long-term randomised outcome trial has been reported. The defensible clinical position is that erythritol is safe at low, incidental dietary exposure and as a professional dental modality[19], uncertain at habitual high daily intakes typical of ketogenic product patterns, and specifically flagged for further study in individuals with pre-existing atherothrombotic, cerebrovascular, or venous-thromboembolic risk.