Erythritol is a 4-carbon polyol (sugar alcohol), a fast-growing “non-nutritive” / “natural” sweetener (EU additive E968), GRAS in the US and EU, marketed to exactly the diabetes / obesity / prior-CVD population most at risk for thrombotic events. One research program (Hazen lab, Cleveland Clinic) has argued it is pro-thrombotic. The whole appraisal turns on one distinction the headline elides: the cohorts measure fasting PLASMA erythritol, which is partly endogenous; the mechanism substudies measure DIETARY erythritol. These are two different exposures answering two different questions. For the decision-relevant claim — does ingesting erythritol raise hard cardiovascular outcomes? — the evidence state is insufficient, not benefit, harm, or no-effect: no RCT on hard outcomes exists, the cohort signal has a live confound, and the mechanism is a surrogate from a single lab. (the framing + evidence-state verdict is the wiki’s own synthesis over the two cited sources)
The two-exposure distinction (type-B — this is load-bearing)
Plasma/serum erythritol and dietary erythritol are distinct objects that the alarm conflates:
- Endogenous plasma erythritol. Erythritol is produced in the body from glucose via the pentose-phosphate pathway (PPP): «Erythritol is endogenously produced by the pentose phosphate pathway20,27, and the metabolite is readily observed in circulation. We speculate that erythritol levels in both validation cohorts originate from a combination of ingestion and endogenous production . While fasting samples in the US validation cohort (where enrollment largely preceded proliferation of erythritol in processed foods) likely reflect endogenous levels» (Witkowski et al., 2023). By the source’s own reading, the US validation cohort’s enrollment «largely preceded proliferation of erythritol in processed foods», so its fasting plasma erythritol is largely a metabolic readout, not a diet marker (Witkowski et al., 2023). (The source gives no explicit enrollment years for the US validation cohort; the 2001-2007 window it reports is the discovery cohort’s (Witkowski et al., 2023).)
- Dietary erythritol. The ingestion substudies dose 30 g and watch plasma rise >1000-fold. That is the exposure a consumer controls.
Why the distinction decides the case: PPP flux (hence endogenous erythritol) rises with hyperglycemia and oxidative stress, so fasting plasma erythritol is plausibly a marker of dysglycemia / poor cardiometabolic health — which independently drives MACE — rather than a signal of what someone ate. This is the reverse-causation / confounding-by-frailty family on The U-Shaped Association Artifact and sits alongside the endogenous metabolite markers on Insulin Resistance Surrogates and Cardiovascular Risk. Corroborating the marker reading, prior work (cited by the source) tied plasma erythritol to incident type-2-diabetes and to central-adiposity weight gain, and in the same discovery screen 1,5-anhydroglucitol — an established glycemic-control marker — was negatively associated with MACE (Witkowski et al., 2023). The whole polyol/glycemic axis is entangled with cardiometabolic health, which is exactly what a confound looks like.
The cohort signal (observational; PLASMA exposure)
Fasting plasma erythritol vs incident 3-year MACE (death / non-fatal MI / stroke), across a discovery and two validation cohorts, all at quaternary referral centers with high CVD / T2D / obesity burden:
| Cohort | n | Q4 vs Q1 crude HR [95% CI] | Q4 vs Q1 adjusted HR [95% CI] |
|---|---|---|---|
| Discovery (untargeted) | 1,157 | 3.22 [1.91-5.41] (top MACE metabolite) | 2.95 [1.70-5.12] |
| US validation | 2,149 | 2.64 [1.79-3.90] | 1.80 [1.18-2.77], P=0.007 |
| European validation | 833 | 4.48 [2.86-7.02] | 2.21 [1.20-4.07], P=0.010 |
(Witkowski et al., 2023). Adjusted = traditional CV risk factors; adding CAD history barely moved it (US 1.79 [1.17-2.74]; EU 2.14 [1.15-3.98]). Continuous: per 1 uM higher erythritol, +21% (US) / +16% (EU) adjusted HR for MACE.
Read the adjustment, not just the point estimate. Adjusting for measured CV risk factors roughly halved the hazard ratios (US 2.64 -> 1.80; EU 4.48 -> 2.21). A signal that large a fraction of which is explained by the covariates that were measured is a signal whose remainder is plausibly explained by the ones that were not — diet, dysglycemia severity, renal function, adiposity trajectory. The authors concede the design «can only show association, and not causation» and flag «the possibility of unmodelled confounding (e.g. diet)» (Witkowski et al., 2023). Transportability is also limited by construction: single fasting measurement at enrollment; quaternary referral population (the authors note general-population translatability «needs to be determined»).
The mechanism (SURROGATE; mechanism-directional, discounted)
The mechanistic leg is what argues toward causality, but it stops at surrogate endpoints — platelet reactivity, aggregation, granule release, animal thrombosis — never a hard human outcome. Admit it directionally, discounted, per the mechanism-directional rule, never as an outcome finding:
- In vitro / ex vivo (2023). Physiological erythritol left-shifted the ADP and TRAP6 aggregation dose-response and dose-dependently enhanced aggregation in platelet-rich plasma; raised intracellular Ca2+ (thrombin), P-selectin and GPIIb/IIIa activation; and increased collagen-dependent platelet adhesion / thrombus in whole blood. Glucose and 1,5-AHG (controls) did nothing (Witkowski et al., 2023).
- In vivo (2023). Mouse FeCl3 carotid-injury: erythritol (25 mg/kg) sped clot formation and shortened time to cessation of flow vs saline or 1,5-AHG (Witkowski et al., 2023). Animal != human (transportability caveat).
- Human interventional surrogate (2024). n=10/group healthy young volunteers (~30 y, no CVD /
diabetes / hypertension, normal renal), 30 g erythritol vs 30 g glucose. Erythritol raised plasma
1000-fold (6480 [5930-7300] uM vs baseline 3.75 [3.35-3.87] uM); glucose did not change plasma erythritol. Erythritol acutely enhanced ADP- and TRAP6-induced aggregation in every subject and enhanced release of the dense-granule marker serotonin (P<0.0001 TRAP6; P=0.004 ADP) and the alpha-granule marker CXCL4 (P<0.0001 TRAP6; P=0.064 ADP); glucose did neither (Witkowski et al., 2024). Conclusion: erythritol «but not glucose, enhances platelet reactivity in healthy volunteers, raising concerns that erythritol consumption may enhance thrombosis potential» (Witkowski et al., 2024).
The dietary->surrogate bridge is the postprandial substudy (2023 n=8; 2024 n=10): a single 30 g dose (= one artificially-sweetened can, or a pint of keto ice cream) drives plasma erythritol above the platelet-reactivity thresholds seen in vitro (Ca2+ 45 uM; aggregometry 18 uM; P-selectin 18 uM; GPIIb/IIIa 4.5 uM) and keeps it there for >2 days (Witkowski et al., 2023). This is the strongest part of the causal story — it connects a controllable exposure to the mechanism — but it lands on a surrogate, in healthy people, with no outcome follow-up.
Independence ceiling — this is type-F, not type-E
Both papers are the same Hazen / Cleveland-Clinic program (overlapping authors, same COSETTE trial NCT04731363, same assays). The 2024 human study refines and extends the 2023 paper it cites — that is claim-refinement (type-F), not independent corroboration (type-E). The mechanism agreeing with the cohort raises confidence within one program’s evidence, but the confidence ceiling is one lab: there is no independent replication of either the cohort association or the platelet mechanism. (The parallel xylitol finding the 2024 paper invokes — «consistent with similar clinical and mechanistic findings recently reported with the sugar alcohol xylitol» (Witkowski et al., 2024) — is also the same lab, so it does not add independence.)
Evidence-state verdict and decision relevance
- Dietary erythritol -> hard CV outcomes (MI / stroke / death): INSUFFICIENT EVIDENCE. No RCT on hard outcomes; the human evidence is observational (on a confounded plasma exposure) plus a single-lab surrogate mechanism. Do not state that erythritol causes MACE. (the four-state verdict + decision framing below is the wiki’s own synthesis over the two cited sources)
- Plasma erythritol -> MACE: a real observational association with a live endogenous-production confound — as consistent with erythritol-as-dysglycemia-marker as with erythritol-as-cause.
- Direction of concern is real but small-lever and unproven. Erythritol is not a big rock; by the Layer-1 ranking this is a peripheral question. The precautionary weight is highest precisely where the sweetener is marketed — high baseline thrombotic risk (diabetes, obesity, prior CVD, impaired renal clearance) — because if the effect is real, absolute risk scales with baseline (route (a), Baseline Risk and the Relative-Absolute Split); but the “if” is unresolved.
- Substitution framing (the comparator is the effect). Judge erythritol against its realistic alternative — sugar, another non-sugar sweetener, or water — not against nothing; this is the comparator logic on Non-Sugar Sweeteners. Replacing sugar-sweetened beverages (the well-evidenced SSB harm on Free Sugars Intake) with an erythritol drink trades a well-evidenced adiposity/T2D harm for an unproven thrombotic concern — a trade the current evidence cannot price.
- Do not generalize across polyols/sweeteners. Allulose has essentially no CV/thrombosis evidence; xylitol has a same-lab parallel signal, not independent confirmation. Each sweetener is its own exposure — the erythritol finding transports to none of them without its own evidence.
Gaps type-G
- No RCT on hard outcomes (the state that would move this off “insufficient”); intervention trials with «appropriate duration of follow-up for clinically relevant outcomes» are what the authors themselves call for (Witkowski et al., 2023).
- No independent (non-Hazen-lab) replication of either the cohort association or the platelet mechanism.
- No study that separates endogenous from dietary plasma erythritol against outcomes (e.g. a Mendelian-randomization or a diet-quantified prospective design) — the one design that would break the confound. AWAITS an independent erythritol->outcome cohort or MR study — resolves the endogenous-vs-dietary confound and the single-program ceiling.