Aspartame and the IARC classification: hazard, risk, and the gap between them

Summary

On 14 July 2023 the World Health Organization published two conclusions about aspartame in a single joint release: one body classified it as possibly carcinogenic to humans on evidence it described as limited in every stream it examined, and another body reaffirmed a safe daily intake that an adult would have to drink nine to fourteen cans of diet soft drink a day to exceed, and the distance between those two statements is the single most instructive thing in the entire sweetener debate.

Why Moderate

This entry sits at Moderate overall, and the confidence field reads "high on the regulatory facts, low on the cancer question" because a single tier cannot carry that split.

Why not the tier above (Strong Evidence). A Strong rating would require consistent human evidence supporting a real-world conclusion about aspartame and health outcomes. That does not exist in either direction. The human cancer evidence is a pooled null across seventeen cohorts measuring a proxy exposure, with one positive tumour-specific row inside it and one protective-looking one; three observational cohorts on liver cancer with an unexcluded reverse-causation pathway; and two molecule-level cohorts finding small associations with cancer overall, neither with a dose-response, neither on the liver endpoint, and one of them overlapping a cohort already counted. The reassurance side has genuinely improved — a pooled null is more than an absence — but it still rests on regulatory synthesis plus a confounded null, and no mechanistic work on the relevant endpoints has ever been done in humans. The only genuinely randomised human trials in the topic are small challenge studies on headache that disagree with each other, one of which lost nearly half its participants before the end.

Why not the tier below (Emerging). Emerging would understate a topic with two major regulatory risk assessments reaching the same intake figure a decade apart, a legally mandated contraindication grounded in confirmed human metabolic genetics, a formal hazard evaluation by a twenty-five-scientist working group with published reasoning, three prospective cohorts with hundreds of thousands of participants, two further cohorts measuring the molecule itself, and a seventeen-cohort pooled analysis. The regulatory and metabolic facts here are among the best-documented in the diet corpus. It is the health conclusion that is unsettled, not the record.

Per sub-area:
• The July 2023 classification and its evidence grades: Foundational. Documented, public, uncontested as a matter of record. What is contested is the adequacy of one of the three appraisals, not the grades themselves.
• Hazard versus risk as a distinction: Foundational. Definitional, stated by the agency itself, not in scientific dispute.
• Group 2B calibration examples: Foundational. Classification record. No headcount is given because it was not verified.
• What the mechanistic grade contained: Moderate. Two named key characteristics with supporting evidence in experimental systems, no high-quality human mechanistic study located, non-liver cancers rated inadequate. Held below Strong because the wording was taken from a summary rather than the primary text.
• The acceptable daily intake and the nine-to-fourteen-cans arithmetic: Foundational. Two major regulatory bodies concurring at 40 mg per kilogram, with arithmetic that checks; the US figure of 50 mg per kilogram and the American regulator's rejection of the classification are stated alongside it, because divergence between regulators is part of the record rather than an embarrassment to it. The regulator's rejection is carried at the strength of a second-hand characterisation, not a quotation.
• The three human cohorts as a body of evidence: Moderate. Real prospective data, real positive signals, judged high quality by IARC, structurally confounded, proxy exposure, and liver-specific with other sites rated inadequate. Moderate is chosen because three independent cohorts agreeing in direction is worth something even when each is weak.
• The two cohorts measuring aspartame itself: Moderate. Molecule-level exposure, which answers the strongest objection to the older cohorts; small effect sizes with intervals close to 1 in the one whose numbers are verified, no effect sizes printed at all for the other; a volunteer cohort in one case and a narrow, partly overlapping population in the other; no dose-response in either.
• The seventeen-cohort pooled analysis: Moderate. A genuine pooled null for cancer overall, which is the strongest human reassurance in the topic, carrying a leukaemia dose-response and a protective-looking colorectal row in the same table. Held at Moderate rather than higher because the exposure is the beverage category, the full text was blocked, and the per-cohort contributions are unverified.
• The Ramazzini rodent findings, and the general limits of high-dose lifetime bioassays: Emerging. Contested methodology, single institute, no dose–response in the reported tumours, rated limited by IARC itself — with that rating non-unanimous and its adequacy contested in print. Rodent evidence, never human harm.
• The saccharin reclassification as a precedent: Foundational. Classification and legislative record, verifiable in public documents. It says nothing about aspartame, and its inclusion is calibration, not evidence.
• The metabolite chemistry and the methanol quantity: Strong Evidence. Settled chemistry plus IARC's explicit conclusion on formaldehyde. The kinetic question about bolus versus food matrix is unaddressed, and the direction of the juice comparison is unverified, which keeps this off Foundational.
• Phenylketonuria as a contraindication: Foundational. Confirmed enzyme deficiency, known metabolite, legally mandated disclosure. The cleanest claim in the entry.
• Headache challenge trials: Moderate. Genuine randomised blinded human data, small, contradictory, dose-divergent, and in the positive case substantially depleted by dropout.
• The WHO 2023 guideline on non-sugar sweeteners: Strong Evidence. Systematic review with an explicit grading, whose conditional status is part of the finding.
• Diet soda as a step off sugar: Moderate. Supported by the randomised half of WHO's review for the short term, unsupported as a destination, untested as a transitional protocol.
• Incidence of any aspartame-attributable harm at ordinary intake: not established. No tier assigned, because no defensible estimate exists and none should be manufactured.

Practical takeaway

If you are drinking sugared soft drinks and want to stop, diet soda is a legitimate step. That is not a grudging concession; it is what the randomised evidence supports for the short term. Take the step. Then keep going — the destination is water, sparkling water, tea or coffee, and the honest reading of the long-term data is that stalling on the bridge is where the benefit disappears.

Know roughly where you sit against the intake level. For an adult of average weight the acceptable daily intake corresponds to something in the region of nine to fourteen cans of diet soft drink per day. Most people are not within an order of magnitude of it. If you are drinking one or two diet drinks a day, the intake question is settled for you and worrying about it is a poor use of attention that could go toward sleep, protein, or walking.

Aspartame is not the only source, and heavy users should count. Sugar-free chewing gum, tabletop sweetener sachets, flavoured yoghurts, sugar-free jelly and squash concentrates, protein powders, and some chewable and dissolvable medicines all contribute. Even stacked, most people remain far below the threshold, but a person consuming several litres of diet drink daily plus gum through the day is the one case where doing the arithmetic is worth ten minutes.

Read the label for the phenylalanine statement if you have phenylketonuria. It is legally required on foods containing aspartame, and on medicines it must state the amount per dose. This is the one place where the label is doing exactly the job the label was designed for.

If you believe aspartame gives you headaches, run the test properly or accept the uncertainty. The blinded evidence at realistic doses does not reproduce the effect, and self-identified sensitive people in the largest trial reported more symptoms in their first session regardless of which bar they were given. That does not prove you are wrong — the trials were small and short — but an unblinded self-test on a symptom that fluctuates day to day cannot tell you anything. If you want to know, remove it completely for two weeks, then reintroduce it without deciding in advance which day you are reintroducing on, ideally with someone else preparing the drinks. If that is impractical, the reasonable move is simply to avoid it, because nothing is lost by doing so.

Do not use the fruit juice comparison as permission to drink juice. The methanol point is a scale calibration only. Whole fruit and fruit juice diverge sharply in this platform's evidence base, with whole fruit associated with lower type-2 diabetes risk and juice with higher risk in the same cohorts. Juice is not the safe alternative in this comparison; it is only the familiar yardstick.

If you are reading a new headline about aspartame and cancer, three questions settle almost all of them. Did the study measure aspartame, or a category of drinks? Was there a dose-response — more exposure, more risk? And is the result being reported alongside the rest of the same analysis? The two cohorts that measured the molecule found small associations with no dose-response; the pooled analysis of seventeen cohorts is null for cancer overall and carries one positive tumour-specific row and one protective-looking one. Any report that gives you one of those numbers and none of the others is not describing the evidence.

The genuinely useful reframe. The strongest reason to move away from aspartame is not that it is dangerous. It is that the thing it was invented to do — help you control your weight — does not hold up over the long term in WHO's own review, while the safety questions have stayed open for five decades. You are accepting an unresolved risk in exchange for a benefit that has not been confirmed. That is a weak trade regardless of how the cancer question eventually resolves, and it is a reason to reduce sweetness overall rather than to swap one sweetener for another.

Evidence detail

Why This Entry Exists

Almost nobody who has an opinion about aspartame has read the document that opinion is supposedly based on. That document is short, it is public, and it says two things at once on purpose. The International Agency for Research on Cancer classified aspartame in Group 2B, "possibly carcinogenic to humans", based on limited evidence in humans, limited evidence in experimental animals, and limited mechanistic evidence. The Joint FAO/WHO Expert Committee on Food Additives, reporting on the same day in the same release, concluded that the evidence gave no reason to change the acceptable daily intake of 40 mg per kilogram of body weight, and WHO itself supplied the worked example that a 70 kg adult would need more than nine to fourteen cans of diet soft drink daily to exceed it.

Neither half is a spin. Both were published together, deliberately, in one joint release from WHO with the International Agency for Research on Cancer — an autonomous agency within WHO — and the Joint FAO/WHO Expert Committee on Food Additives, because they answer two different questions. One asks whether a substance is capable of causing cancer under any circumstances at all. The other asks whether it does so at the amounts people actually consume. The first is a hazard identification. The second is a risk assessment. A shark is a hazard; a shark in an aquarium is not a meaningful risk to you on the other side of the glass. That distinction is not a rhetorical trick invented by industry to soften a bad headline, and it is not a bureaucratic evasion invented by regulators to avoid acting. It is the structure of the discipline, and WHO stated it explicitly in the release itself.

This entry exists because the aspartame story is the best available teaching case for reading an evidence claim properly, and because both of the loud positions on it are wrong in ways that are easy to demonstrate. It also exists because the honest precautionary argument about sweeteners has almost nothing to do with cancer, and gets buried every time the cancer argument is made badly. Two months before the classification, in May 2023, WHO published a separate guideline conditionally advising against the use of non-sugar sweeteners for weight control at all, on the grounds that they do not produce long-term reductions in body fat. That is the argument that actually holds: the benefit these molecules were invented to deliver is itself in doubt, while the risk side is a set of open questions that have not closed in fifty years. Unknown risk weighed against unconfirmed benefit is a weak trade. That is a much stronger position than any harm assertion, and it needs no scare story to stand up.

What bad advice this protects against, in all directions:
• "The WHO says aspartame causes cancer." It says the opposite of that, twice over. It says a positive association was seen in some human data but that chance, bias and confounding could not be ruled out with reasonable confidence, which is what "limited evidence" means as a technical term when IARC applies it to human data. And in the same release it reaffirmed the intake level it considers acceptable. A classification that carries no dose information cannot license a sentence containing the word "causes".
• "Group 2B is meaningless, it's the same category as pickled vegetables." The calibration is real and worth knowing, but it is calibration, not refutation. Group 2B also records that a WHO working group of twenty-five scientists looked at three independent human cohorts, judged them to be of high quality and adjusted for many confounders, found a positive signal pointing in the same direction on liver cancer, and could not explain it away. That is not nothing. Pointing at the company aspartame keeps in the category shows the category is broad; it does not show the signal is absent.
• "Ramazzini proved aspartame causes tumours." Those were rodent studies, and the word "rodent" has to be in the sentence. IARC in 2023 rated the animal evidence limited as well, which means the body that issued the Group 2B classification did not accept the Ramazzini findings as sufficient. That formal grade is a fact — but its finality is not, because a minority of the working group is recorded as not agreeing, and senior figures from inside the hazard-evaluation world have since argued in print that the appraisal was not done to the agency's own standard.
• "Ramazzini was discredited, so the animal question is closed." Also wrong, for the mirror reason. The methodological criticisms are real and specific, but "limited" is not "none", and one of the disputed points is that the studies ran the animals to natural death rather than stopping at the standard 104 weeks, which is at least arguably a design that sees more, not less. The single strongest objection — that the reported blood cancers showed no dose-response — has also been complicated by a human pooled analysis reporting a leukaemia dose-response in the same tumour family.
• "A big new study just found aspartame causes cancer." Two cohorts have now measured aspartame as a molecule rather than as a beverage category, and both report a small positive association with cancer risk overall. That is a real strengthening of the concern side and it should be said. It is also true that neither reported the liver endpoint that drove the classification, neither produced a dose-response, one is a volunteer cohort skewed female and highly educated, the other overlaps the cohort family the classification already used, and the largest pooled human analysis of the beverage category is null for cancer overall. A small association without a dose-response in confounded data is exactly the shape of an open question.
• "The American regulator rejected it, so the classification is dead." The disagreement is real and belongs on the page: the US Food and Drug Administration did not accept the classification, citing significant shortcomings in the studies it rested on. It is also true that an agency which has authorised a substance for four decades is not a disinterested reader of a classification implying it should not have. The same scepticism this entry applies to an agency that classifies applies to an agency that declines to.
• "It turns into formaldehyde, a known carcinogen." Chemically true and practically empty without the quantity. The IARC working group concluded there was no evidence that the formaldehyde arising from aspartame consumption would significantly alter normal endogenous formaldehyde concentrations, which is to say: your own metabolism produces far more of it, continuously, than a can of diet soda contributes.
• "It's 99% water, so how could it hurt you?" A category error, and it is the same category error as the scare version. Composition percentages say nothing about what the remaining fraction does. A medicine is mostly excipient. The proportion of the vehicle is not the question; the dose of the active molecule is. Both camps are substituting a percentage or a label for a dose.
• "It has a warning label, so it must be dangerous." The label reads "Phenylketonurics: Contains Phenylalanine" and it is a disclosure requirement for people with a specific inherited enzyme deficiency who must count phenylalanine from every source in their diet. It is not a general safety warning, and it implies nothing about anyone without that condition.
• "Diet soda is just as bad as regular soda, so you may as well drink the real thing." This is the advice that does actual damage. As a step off sugar, diet soda is a genuine improvement, and the randomised evidence in WHO's own review supports the short-term substitution effect. It is a bridge, not a destination. Water is the finish line. Telling someone reducing sugar that the substitution is pointless pushes them back toward the sugar, which is worse on any reading of the evidence.

What this entry OWNS: aspartame specifically, and the July 2023 IARC and JECFA decisions as a worked lesson in reading hazard against risk. The acceptable daily intake and the arithmetic that makes it concrete. The three human cohorts the classification rested on and what they actually measured, plus the two cohorts that have since measured the molecule itself and the pooled analysis of the beverage category. What the three "limited" grades actually contained, including the mechanistic stream. The Ramazzini dispute, the published challenge to how the working group appraised it, and where that stands. The saccharin reclassification as the one case in this chemical class where a rodent tumour finding was resolved rather than merely disputed. The metabolic breakdown into phenylalanine, aspartic acid and methanol. The phenylketonuria contraindication, which is the one non-negotiable in the topic. The controlled headache-challenge trials.

What this entry DEFERS: the class-level case for and against non-sugar sweeteners as a category, including the WHO 2023 guideline's implications for practice, to artificial_sweeteners_evidence. The general reasoning about why observational cohorts and randomised trials disagree, and how to weigh them, to rct_vs_observational_evidence. Why a marker moving is not an outcome changing, to surrogate_endpoints_vs_outcomes. Why the people who choose diet drinks differ systematically from those who do not, to healthy_user_bias. Molecule-specific evidence on other sweeteners to sucralose_evidence_and_safety and erythritol_and_sugar_alcohols. Additive risk framing beyond sweeteners to food_additives_to_avoid and food_colourings_evidence_and_claims.

Evidence

Read the tiers, not the thesis. This is a topic where a single vivid phrase — "possibly carcinogenic" — does all of the persuasive work in both directions, and where the underlying claim structure is much more specific and much less dramatic than the phrase. Every numbered claim below is labelled by evidence type, because on this topic the difference between a regulatory classification, a human observational cohort, a human randomised trial, a rodent bioassay and a cell-culture result is not a detail. It is the entire disagreement.

Sub-area 1: What was actually decided in July 2023

1. IARC classified aspartame as Group 2B, "possibly carcinogenic to humans", on limited evidence in humans, limited evidence in experimental animals, and limited mechanistic evidence. Foundational. Evidence type: regulatory hazard classification, not itself a study. Its inputs were human observational, rodent, and in-vitro. A working group of twenty-five scientists from twelve countries met at IARC in Lyon in June 2023; the summary appeared in The Lancet Oncology on 13 July 2023, and the full assessment was published later as IARC Monographs Volume 134, which appeared in 2024. All three evidence streams were rated limited. None reached sufficient. Applied to human data, "limited evidence" means that a positive association was observed, but chance, bias and confounding could not be ruled out with reasonable confidence. That definition belongs to the human stream only: the same word carries a different technical definition for animal evidence and a third for mechanistic evidence, so the fact that all three streams landed on the same word is a coincidence of vocabulary, not of meaning. In none of the three is it a weaker grade of "causes cancer"; each names an uncertainty rather than resolving it. (WHO/IARC/JECFA joint news release, 14 July 2023; The Lancet Oncology, 13 July 2023; IARC Monographs Volume 134.) Cui bono: campaigners and "clean label" brands gain enormously from compressing "Group 2B on limited evidence" into "the WHO says aspartame causes cancer" — it is the most shareable misreading in the sweetener debate. On the other side, beverage manufacturers and sweetener trade associations benefit from framing the whole exercise as a bureaucratic non-event, which understates that a WHO body found a human signal in three separate cohorts and could not account for it.

2. A Group 2B classification identifies a hazard. It carries no dose information and makes no statement about risk at real-world intake. Foundational. Evidence type: methodological and definitional, from the agency's own framework. The WHO release draws the distinction in its own words: IARC classifications identify carcinogenic potential as a hazard, whereas JECFA risk assessments determine the probability of a specific type of harm under specified conditions and levels of exposure. Group 2B is the third of four levels. In the preamble's own terms it is reached by several routes — limited evidence in humans with less than sufficient evidence in animals, or sufficient evidence in animals alone, or limited human evidence supported by strong mechanistic evidence. WHO's public gloss compresses that into "limited but not convincing evidence in humans, or convincing evidence in experimental animals, but not both", which is serviceable for a reader but is a summary of the criteria rather than the criteria. Either way, the question "can this cause cancer under any circumstance?" and the question "does it, at the amounts people consume?" are separate questions with separate answers, and on 14 July 2023 both were answered. (WHO news release, 14 July 2023; IARC Monographs Preamble.) Cui bono: the hazard-versus-risk distinction is exactly the sentence that must be cut for the scare headline to work, which is why it is almost never in the headline. But the same distinction is also used to imply Group 2B is empty, which it is not — it records a real, weak, unexplained human signal.

3. Group 2B also contains agents that calibrate how weak the category is. Foundational. Evidence type: regulatory classification. The category includes aloe vera whole leaf extract and traditional Asian-style pickled vegetables, the latter classified in IARC Monographs Volume 56 in 1993. The category is large, running to several hundred agents, and includes items nobody treats as a public health emergency. The specific headcount of Group 2B is not stated here because it moves and was not verified against a primary source; nor is any claim made about the current status of other commonly cited examples. The useful move is to state the confirmed company aspartame keeps and let the reader calibrate. (IARC Monographs, Volume 56, 1993; IARC list of classifications.) Cui bono: calibration-by-company is a rhetorical device and it cuts both ways. It is the sweetener industry's favourite line, and it is legitimate as far as it goes — but "pickled vegetables are in the same box" demonstrates that the box is broad, not that this particular item is safe.

4. "Limited mechanistic evidence" was not an empty box. The working group found supporting evidence for two mechanisms in experimental systems, and recorded that no high-quality human mechanistic studies existed. Moderate. Evidence type: MECHANISTIC, in experimental systems only — in-vitro and animal — as appraised by the working group. Explicitly NOT human. Two of the key characteristics of carcinogens in the agency's own evaluation framework were judged to have supporting evidence for aspartame: that it induces oxidative stress, and that it induces chronic inflammation. Both were supported in experimental systems. Alongside that, the working group recorded that there were no high-quality studies of mechanistic endpoints associated with those key characteristics in exposed humans, named metabolic pathways and gut-microbiome effects among the work it wanted done, and rated the human evidence for cancers other than liver cancer inadequate. The inference the label hides runs in both directions at once. Toward concern: oxidative stress plus chronic inflammation is a coherent liver-carcinogenesis story sitting directly alongside a human liver-cancer signal, and that cross-stream coherence is a large part of why a working group acts at all on evidence it grades as weak in every stream — three weak streams pointing the same way is a different situation from three unrelated weak streams. Toward reassurance: none of that mechanistic work was done in humans, which is a cleaner and more damning statement of the gap than any general hedge, and it is the specific reason the evaluation stopped at Group 2B rather than climbing. Sourcing caveat, stated because it matters: this account was taken from a summary of the General Remarks rather than read off the primary text, so no wording is quoted here and the two named characteristics stand as substance pending re-verification against the monograph itself. (IARC Monographs Volume 134, General Remarks.) Cui bono: reducing this stream to the single word "limited" suits the reassurance side, because a named mechanism is more alarming than a grading label. Reporting the two characteristics without the words "experimental systems" and "no high-quality human studies" suits the alarm side, because it converts a laboratory finding into a description of a person.

Sub-area 2: The risk assessment published the same day

5. JECFA reaffirmed the acceptable daily intake at 40 mg per kilogram of body weight, and WHO's own worked example is that an adult would need more than nine to fourteen cans of diet soft drink daily to exceed it. Foundational. Evidence type: regulatory risk assessment, synthesising human randomised, human observational, rodent and toxicokinetic data. JECFA concluded the data evaluated indicated no sufficient reason to change the previously established acceptable daily intake of 0 to 40 mg per kilogram of body weight. The arithmetic is WHO's, not a reconstruction: a 70 kg adult has an allowance of 2,800 mg per day; at 200 to 300 mg of aspartame per can, that is between roughly nine and fourteen cans, assuming no other dietary source. The European Food Safety Authority independently reached the same 40 mg per kilogram figure in its 2013 re-evaluation, from a different review process. Regulators do not all land in the same place, and the divergence is instructive: the United States Food and Drug Administration sets its acceptable daily intake for aspartame at 50 mg per kilogram, a quarter higher than the European and international figure. That regulator also went further than a numerical difference — it did not accept the classification itself, citing significant shortcomings in the studies the human stream relied on, and the joint expert committee for its part concluded that aspartame has not been found to have adverse effects after ingestion and that no mechanism for cancer induction had been identified. Sourcing caveat: the American regulator's position on the classification is reported here as characterised by the US National Cancer Institute's public summary; the regulator's own statement was not retrieved in this pass, so no wording of it is quoted and the characterisation should be sourced directly before being repeated as a quotation. (WHO/IARC/JECFA joint news release, 14 July 2023; FAO/WHO JECFA 96th meeting; EFSA ANS Panel, EFSA Journal 2013;11(12):3496; US FDA aspartame guidance; National Cancer Institute, artificial sweeteners and cancer.) Confounds worth stating: the acceptable daily intake is a conservative benchmark derived from animal no-effect levels with a safety factor applied, not a cliff edge at which harm begins — being under it is not a certificate and being briefly over it is not an injury. The 40-versus-50 spread between major regulators is the clearest demonstration available that this number is a judgement about how much margin to demand, not a measured threshold. The worked example also assumes no other source, which is unrealistic for someone stacking sugar-free gum, tabletop sweetener, flavoured yoghurt and certain medications on the same day, though even generous stacking leaves most people far below 2,800 mg. Cui bono: the nine-to-fourteen-cans figure is the sweetener sector's most-cited line and it is genuinely accurate. It is deployed to close the conversation rather than open the more interesting question of whether the benefit is real at all. The alarm side must suppress the figure entirely, because a hazard with no plausible exposure route is not a story. And the regulator's rejection of the classification cuts both ways: it is a real expert disagreement that belongs on the page, and it comes from the agency that authorised the substance in the first place, which is not a neutral position to read a classification from.

Sub-area 3: The human evidence underneath the classification, and what has arrived since

6. The limited human evidence was three prospective cohort analyses reporting a positive association between artificially sweetened beverages and liver cancer. Moderate. Evidence type: HUMAN OBSERVATIONAL (prospective cohort). No human randomised trial exists for cancer endpoints with aspartame, and none is feasible. The three the working group identified as showing a positive association between artificially sweetened beverage consumption and liver cancer risk, overall or in important subgroups, were: Stepien and colleagues in 2016, on soft drinks and juices and risk of liver and biliary tract cancers in a European cohort; McCullough and colleagues in 2022, on sugar- and artificially-sweetened beverages and cancer mortality in a large United States prospective cohort of 934,777 cancer-free participants enrolled in 1982 with deaths tracked through 2016; and Jones and colleagues in 2022, a pooled analysis of sweetened beverage consumption and liver cancer risk stratified by diabetes status. The working group's own appraisal should travel with the criticism that follows: it recorded that these were high-quality studies which controlled for many potential confounders, and still concluded that chance, bias or confounding could not be ruled out with reasonable confidence. It is worth stating what this stream did not cover, because it is a point in the reassurance side's favour that is usually left unmade: the human evidence was liver-specific, and evidence for cancers at other sites was rated inadequate. (IARC Monographs Volume 134, General Remarks; Stepien M et al., Eur J Nutr 2016;55(1):7-20; McCullough ML et al., Cancer Epidemiol Biomarkers Prev 2022;31(10):1907; Jones GS et al., Cancer Epidemiol 2022.) Cui bono: "three separate cohorts found a liver cancer link" is true, sounds decisive, and works only if the next paragraph is omitted. "Purely observational, proves nothing" is also true as far as it goes, and is used to dismiss a signal that three independent groups produced in the same direction — which is precisely why IARC acted at all.

7. The exposure measured in those cohorts was "artificially sweetened beverages", not aspartame. Moderate, and this is the load-bearing limitation. Evidence type: HUMAN OBSERVATIONAL, exposure-assessment critique. The drinks in these cohorts contained varying blends of aspartame, acesulfame K, saccharin and sucralose, differing by product, by country and by decade. A cohort recruited in 1982 and followed to 2016 is measuring a category whose contents changed underneath it. Attributing a signal from that category to one molecule is an inference the data cannot support on its own. The two cohorts that have measured aspartame as a molecule are the next two claims, and the honest summary of them is that they answer the exposure objection and raise a different one: neither reported the liver endpoint that drove the classification. Nobody has yet published molecule-level aspartame exposure against liver cancer, which means the specific finding the classification rests on has never been tested with a clean exposure measure. (IARC Monographs Volume 134, General Remarks.) Cui bono: nobody's marketing depends on exposure misclassification, which is why it is the most reliable fact in the sub-area and the least reported.

8. The largest cohort to measure aspartame itself rather than a beverage category reported a small positive association with overall cancer risk. Moderate. Evidence type: HUMAN OBSERVATIONAL (prospective volunteer cohort), molecule-level exposure across dietary sources. The NutriNet-Santé cohort followed 102,865 French adults, 78.5 percent of them women, for a median 7.7 years, with 3,358 incident cancers. Comparing higher consumers against non-consumers, total artificial sweeteners carried a hazard ratio of 1.13 (1.03 to 1.25) with a trend p of 0.002; aspartame specifically 1.15 (1.03 to 1.28); acesulfame K 1.13 (1.01 to 1.26); and sucralose showed no significant association. Among sites, breast cancer with aspartame was 1.22 (1.01 to 1.48) and obesity-related cancers 1.15 (1.01 to 1.32), with prostate null. This entry previously described those as "isolated signals", which inverted the paper: the headline result was total sweeteners against cancer overall, and the site findings were secondary analyses. The reason it did not drive the classification is not that it was isolated — it is that the working group's human stream was liver-specific and this cohort did not report a liver signal. Four limits must travel with it and they are substantial: this is a volunteer cohort skewed toward women and the highly educated, with reported intake lower than national estimates and a real possibility of selection bias; exposure came from self-reported 24-hour dietary records; several confidence intervals sit close to 1; and the authors themselves name residual confounding and reverse causality and state that causal links cannot be established from this single study. (Debras C, Chazelas E, Srour B et al., "Artificial sweeteners and cancer risk: Results from the NutriNet-Santé population-based cohort study", PLOS Medicine 2022;19(3):e1003950.) Note against this entry's own change-of-mind list: the study this entry asks for is a large cohort measuring aspartame specifically with repeated dietary assessment. This meets the exposure-precision and repeated-assessment conditions and fails only the liver-endpoint one. That partial satisfaction is stated rather than buried. Cui bono: the alarm side quotes the aspartame hazard ratio without the volunteer-cohort limits or the interval's proximity to 1. The reassurance side has an interest in this study staying a parenthetical, because it is the one cohort that cannot be answered with the exposure-misclassification objection — and a symmetrical interest in the sucralose null inside it, which is genuine reassurance the corpus should not throw away.

9. A second molecule-level cohort, published in February 2026, reported a higher cancer risk with aspartame and no dose-response. Moderate. Evidence type: HUMAN OBSERVATIONAL (prospective cohort), molecule-level exposure. No effect size is printed here — see the sourcing note. The American Cancer Society's nutrition cohort assessed aspartame intake in 100,004 US adults in 1999 and again in 2003, at molecule level, from low-calorie carbonated beverages (70 to 180 mg per 12-ounce serving depending on type) and from single-serving packets (20 mg per packet), with 21,356 verified cancer cases through June 2017 including 8,087 obesity-related. The stated conclusion was that aspartame consumption was associated with higher cancer risk, both obesity-related and non-obesity-related, regardless of weight status — with the authors themselves writing that the lack of a dose-response relationship underscores the need for cautious interpretation and replication. Sourcing note, and it is a hard constraint: the hazard ratios and confidence intervals could not be retrieved — both the publisher and the journal page returned access errors — so no effect size appears here and none may be supplied downstream. The design, the scale, the direction and the authors' own caveat are what is verified. Three limits, one of them serious. This is not a fourth independent replication: it is a subcohort of the same cohort family as one of the three studies the classification already rests on, so what it adds is better exposure measurement and an incidence rather than a mortality endpoint on partly the same people. The population is narrow — 98 percent non-Hispanic White, mean age 69. And there is no dose-response, which is the same objection this entry applies to the rodent tumour findings and must be applied here with exactly equal force. (Journal of Nutrition, February 2026, article S0022-3166(26)00109-4.) Cui bono: a 2026 cohort in a major nutrition journal is a strong-sounding citation and will circulate as "new study confirms", with the missing dose-response and the cohort overlap dropped. The reassurance side will cite the missing dose-response alone, which is fair only from someone who applies it to the rodent data too.

10. Pooled across seventeen prospective cohorts, artificially sweetened beverages showed no association with cancer overall — and a leukaemia signal that did carry a dose-response. Moderate. Evidence type: HUMAN OBSERVATIONAL, dose-response meta-analysis of prospective cohorts. Exposure is again the beverage category, not the molecule. Lead with the null, because it is the single strongest piece of human reassurance available in this topic and this entry previously did not have it: for cancer overall, highest against lowest intake, the pooled relative risk was 1.03 (0.96 to 1.11), p=0.407. That upgrades the reassurance case from "no demonstrated effect" — which is an absence — to a pooled null, which is a finding. In the same analysis, non-lymphoid leukaemia came out at 1.35 (1.03 to 1.77), p=0.030, with a positive linear dose-response of about fifteen percent per 355-millilitre daily serving, 1.15 (1.02 to 1.30); colorectal cancer ran the other way at 0.78 (0.62 to 0.99). Four counters must travel with the leukaemia strand and they are substantial: it is one site among many tested, so multiple comparisons are live as an explanation; the exposure is the beverage category again, not aspartame; the signal is likely carried by a small number of cohorts whose own authors treated their leukaemia finding cautiously; and a protective-looking colorectal result in the same analysis is a classic signature of noise rather than of biology. Sourcing caveat: effect sizes came from the published abstract plus a corroborating retrieval; the full text was blocked, so per-cohort contributions and heterogeneity statistics are unverified and the multiple-comparisons explanation is offered as a candidate rather than asserted as the answer. Why this matters beyond its own numbers: the rodent tumours this entry discounts were lymphomas and leukaemias, and the stated reason for discounting them is a missing dose-response — so a human pooled leukaemia signal with a dose-response is concordance on tumour type across two evidence streams plus the arrival of the exact element said to be absent. Whether a dose-response inside a proxy exposure counts as evidence, or merely as a better-shaped version of the same confound, is a judgement this entry does not pretend to settle; it is carried as a lead, not as a receipt. (Yin S et al., "Artificially Sweetened Beverage Consumption and Cancer Risk: A Comprehensive Dose–Response Meta-Analysis of Prospective Studies", Nutrients 2022;14(21):4445.) Cui bono: the reassurance side wants the overall null and not the leukaemia row; the alarm side wants the leukaemia row and not the null. They are in the same table, from the same authors, on the same data.

11. Reverse causation runs directly through the liver-cancer signal, and IARC did not claim to have excluded it. Moderate. Evidence type: HUMAN OBSERVATIONAL, confounding structure. People with obesity, metabolic syndrome, fatty liver disease and type-2 diabetes switch preferentially to diet drinks, and those same conditions are leading causes of hepatocellular carcinoma. The diet drink can therefore be a marker of the disease process rather than a cause of it. That Jones and colleagues stratified by diabetes status is itself an acknowledgement that diabetes sits in the middle of the association. Add to that the reliance in most cohorts on a single baseline food-frequency questionnaire administered sometimes decades before diagnosis; residual confounding by alcohol, hepatitis B and C, and total energy intake; the fact that McCullough measured cancer mortality rather than incidence, which is a different endpoint; and the low incidence of liver cancer, which leaves subgroup estimates resting on small case counts. IARC's own stated research need was repeated dietary questionnaires and longer follow-up, which is a direct statement that the current data cannot settle the question. What should not be said is that nobody has tried to separate association from reverse causation. Attempts exist and do not settle it: genetic instrumental-variable analyses have been run in this field, but their instruments proxy a questionnaire answer about whether someone adds sweetener to coffee, cereal or tea rather than an actual dose, so they carry over much of the confounding they were designed to remove, and their results have been inconsistent. Meanwhile the observational literature keeps growing in the same shape, which adds volume rather than identification. Guard against the most likely future misattribution in this literature: the one instrumental-variable study in this space with a positive cardiometabolic signal is for erythritol, a sugar alcohol, and it belongs to erythritol_and_sugar_alcohols. It must never be imported into an aspartame claim in either direction. The general form of this problem is covered in healthy_user_bias and rct_vs_observational_evidence; what is specific here is that the confounder and the outcome are the same disease. (IARC Monographs Volume 134, General Remarks.) Cui bono: the alarm side needs reverse causation unmentioned; the industry side needs it treated as a complete answer, when in fact cohorts pointing the same way is what you would expect from either reverse causation or a real effect, and no design yet applied has separated them.

Sub-area 4: The animal evidence

12. The Ramazzini Institute lifespan rodent bioassays reported aspartame-related tumours, regulators discounted them on methodological grounds, IARC rated the animal evidence limited — and the adequacy of that appraisal is now formally contested in print. Emerging. Evidence type: RODENT lifetime bioassay, plus a published methodological critique of the hazard evaluation. This is animal evidence and must never be reported as human harm. The Ramazzini Institute, under Soffritti and colleagues, ran bioassays in which exposure continued to natural death rather than terminating at the standard 104 weeks, reporting increased tumour incidence including lymphomas and leukaemias. EFSA did not accept the findings as evidence of carcinogenicity. Its stated grounds were the high background incidence of chronic inflammatory changes in the lungs and other organs in that colony, affecting control animals not fed aspartame as well as treated ones, which EFSA argued could account for the observed lesions; the absence of a positive dose–response relationship for the reported lymphomas and leukaemias, which is the point that bears most directly on whether the tumours were treatment-related; and deviation from internationally established study protocols, compounded by the loss of tissue to autolysis that follows from holding animals to natural death. EFSA concluded the results did not change its conclusions and that the 40 mg per kilogram intake needed no revision. The counter-view, stated fairly: running to natural death is arguably more sensitive to late-onset tumours than a protocol that stops at 104 weeks, since rodent tumours cluster in late life, and Ramazzini's defenders argue a design difference is being treated as a defect. On the dose–response ground specifically, one complication now has to be stated: a pooled analysis of seventeen human cohorts has since reported a leukaemia signal with a positive dose-response — see the pooled-analysis claim above — which is the same tumour family as the rodent findings and supplies the element whose absence was the strongest objection. That is cross-stream concordance on tissue, in a confounded proxy exposure, and it is carried as a lead rather than as a receipt. One commonly repeated criticism — that the pathology slides never received independent review — could not be confirmed as one of EFSA's stated grounds in this pass and is therefore not asserted here. The published dispute this entry previously flagged as unverified has resolved, and it resolves toward concern. Landrigan and colleagues, in the Annals of Global Health on 6 June 2025, argue that the IARC working group failed to articulate and justify its negative evaluation of these studies — that is, they argue the animal evidence was under-graded and imply the defensible grade was sufficient. Their specific allegations: strong statements documenting tumours in multiple organs and tissues sitting underneath a conclusion of merely limited evidence; a broad-brush summary assessment lacking the study-by-study appraisal the agency's own guidance prescribes; divergence from the agency's stated procedures for animal evidence; and a failure to discuss differing opinions, including a recorded minority opinion in which a minority of the working group did not consider that the stated concerns would substantially change an interpretation of sufficient evidence. The paper also answers the litter-effects criticism, noting that the rat lifetime study restricted one male and one female pup per litter to each dose group. Two things make this more than advocacy. The authorship is a former head of the agency's own monographs programme, a former director of the US environmental health institute and its national toxicology programme, a former senior scientist at that programme, and a very senior environmental epidemiologist — not the institute defending its own findings, but people who used to run evaluation programmes of this kind saying the evaluation was not done to the programme's own standard. And the documented minority means that "the agency did not accept them as sufficient" is true of the formal grade while concealing that the working group was not unanimous. The counterweight belongs in the same breath: this is a commentary, not new data; it does not answer all of EFSA's specific 2013 objections, and on the litter question it answers one; two of its authors declare that they serve on the journal's editorial team; and its funding is not declared in the accessible content and could not be established. Exact quotations are not offered here: the minority-opinion and sufficient-evidence wording was taken from the journal's article page after the archive mirror was blocked, so the substance is reported and the phrasing awaits verification against the primary text. (EFSA ANS Panel, EFSA Journal 2013;11(12):3496; EFSA evaluation of the Ramazzini Institute aspartame studies; Soffritti M et al., Ramazzini Institute lifespan bioassays; IARC Monographs Volume 134, animal evidence rated limited; Landrigan PJ, Straif K, Birnbaum LS, McDiarmid M, Perry MJ, Forastiere F, Comba P, Bucher JR, Annals of Global Health, 6 June 2025, doi 10.5334/aogh.4771.) The framing that survives, replacing the one that did not: the formal grade of limited is a fact, and the finality of that grade is not. IARC did not accept Ramazzini as sufficient; a documented minority of its own working group did not agree with that; and senior figures from inside the hazard-evaluation world have argued in print that the appraisal departed from the agency's stated procedure. Neither camp gets to end the argument by citing the grade alone. Cui bono: Ramazzini is the load-bearing citation for every "aspartame causes cancer" claim, and its findings circulate without the word "rodent" and without the colony-infection problem. On the other side, sweetener trade bodies fund and amplify the methodological critiques, and some of the re-analyses in the toxicology literature have funding that was not verified in this pass — treat any critique whose funding you cannot see with the same suspicion you would apply to the original, and that instruction now cuts at the 2025 critique too, whose funding is equally invisible.

13. Rodent lifetime bioassays at high doses have limited and contested predictive value for human cancer at dietary exposures. Moderate. Evidence type: methodological, drawing on comparative toxicology. This is a general caution rather than an aspartame-specific finding, and it cuts against both camps: it weakens the Ramazzini case, and it also weakens any claim that clean rodent data would settle the question in the other direction. Species differences in aspartame metabolism are real. Whatever reading of run-to-natural-death designs the corpus settles on has to be the same reading everywhere, including where a regulator has used that design objection to reject a bioassay on a different sweetener. Cui bono: whichever side has the inconvenient animal data cites this argument, and the side with the convenient animal data forgets it.

14. Saccharin is the one case in this chemical class where the whole arc has already run: a rodent tumour finding was pursued to mechanism, the mechanism proved species-specific, and the classification, the listing and the warning label were all undone. Foundational. Evidence type: regulatory and classification record. Not evidence about aspartame. In 1999 IARC moved saccharin and its salts from Group 2B to Group 3, not classifiable as to carcinogenicity to humans, on the basis that sodium saccharin produces bladder tumours in rats by a mechanism not relevant to humans because of critical interspecies differences in urine composition. In 2000 the US National Toxicology Program's ninth report on carcinogens removed saccharin from the list of substances reasonably anticipated to be a human carcinogen. Federal legislation removing the mandated warning label was signed on 21 December 2000, and California delisted it effective 6 April 2001. The National Cancer Institute states that no clear evidence for an association between saccharin use and bladder cancer incidence in humans has emerged from epidemiological studies. Deliberately not asserted: the finer mechanistic account commonly given — high-dose sodium salt raising urinary pH, calcium phosphate microcrystals, cytotoxicity and regenerative hyperplasia, largely in male rats — was not verified against primary text, so only the general interspecies-urine-composition wording is used. The honest limit is the whole point of including it: different molecule, different tumour site, different mechanism. This is a precedent for the system working, not evidence about aspartame. It earns its place for two reasons. It is the strongest available demonstration that a hazard classification is not permanent and can be undone when a mechanism is resolved, which is exactly what this entry's own reassurance trigger asks for. And it is the sharpest available contrast with the Ramazzini situation, because it shows what a rodent tumour finding looks like when it is resolved rather than merely disputed — a named mechanism, a named species difference, a downgrade, a delisting, a repealed label. None of those five things has happened for aspartame. (IARC reclassification of saccharin and its salts to Group 3, 1999; US National Toxicology Program, Ninth Report on Carcinogens, 2000; National Cancer Institute, artificial sweeteners and cancer.) Cui bono: the sweetener sector cites the saccharin reversal as proof that sweetener scares always deflate, which is a prediction dressed as a precedent. The alarm side has to leave it out entirely, because it is the clearest counterexample to the claim that agencies never move backwards.

Sub-area 5: Metabolism, and the one real contraindication

15. Phenylketonuria is the one genuinely established, non-negotiable contraindication, and it is why the label warning exists. Foundational. Evidence type: established human metabolic genetics plus a regulatory requirement. Mechanism confirmed in humans. Aspartame is a methyl ester of a dipeptide, so digestion releases free phenylalanine. People with phenylketonuria carry a genetic deficiency in phenylalanine hydroxylase and cannot metabolise phenylalanine; accumulation causes neurological damage, so intake must be tightly restricted for life. United States regulation therefore mandates the statement "Phenylketonurics: Contains Phenylalanine" on aspartame-containing foods, and a parallel declaration on over-the-counter and prescription drugs that additionally requires the amount per dosage unit, so patients can count it into their daily allowance. Incidence is roughly one in ten to fifteen thousand births, and newborn screening detects it in most high-income countries. (21 CFR 172.804; 21 CFR 201.21.) The teaching value is the contrast: this is what a real contraindication looks like — a named enzyme defect, a known metabolite, a dose that matters, a legally mandated disclosure. It looks nothing like the cancer discourse. Cui bono: the label is routinely presented as proof that aspartame is dangerous to everyone, which inverts what the label is for. Industry has no incentive to obscure this one, because it is narrow, genetic, and screened for at birth.

16. Aspartame breaks down into phenylalanine, aspartic acid and methanol, and the methanol quantity is small in absolute terms. Strong. Evidence type: established chemistry and human toxicokinetics; the formaldehyde judgement is IARC's reading of mechanistic and human metabolic data. The IARC working group confirmed the hydrolysis products and concluded explicitly that there was no evidence that the overall amount of formaldehyde formed as a result of aspartame consumption would significantly alter normal endogenous formaldehyde concentrations. The working group declined to assess the individual metabolites separately in the carcinogenic hazard evaluation. The methanol arithmetic is derivable rather than borrowed: methanol accounts for roughly eleven percent of aspartame by mass, so a can containing 200 to 300 mg of aspartame yields roughly 22 to 33 mg of methanol. Fruit juice also contains methanol, produced by pectin demethylation, and the quantity from a glass of juice is of the same order of magnitude — tens of milligrams. Which of the two is higher is not asserted here: the widely circulated milligram-per-litre comparison figures trace only to industry and advocacy fact sheets, could not be traced to a primary measurement, and are not used, so the comparison stands only as a statement that both quantities sit in the same range. (IARC Monographs Volume 134, General Remarks; 21 CFR 172.804 for the molecular formula; mass fraction derived stoichiometrically.) Important guard: this is a scale calibration, not an endorsement of juice. The platform's own fruit evidence has whole fruit lowering type-2 diabetes risk while juice raises it within the same cohorts, so juice is a poor benchmark for health — which is exactly what makes it a useful benchmark for "this quantity of methanol is unremarkable". Confound: the quantity argument answers dose but not kinetics, and there is no strong human data either way on whether a bolus from a rapidly absorbed liquid behaves identically to methanol released slowly from a food matrix. The formaldehyde pathway is real chemistry; IARC's judgement is about scale, not about the pathway not existing. Do not overstate it as "aspartame produces no formaldehyde". Cui bono: "aspartame turns into formaldehyde, a Group 1 carcinogen" is chemically true and completely misleading without the quantity, and it is one of the most effective scare framings in circulation. The fruit-juice rebuttal is an industry staple, and the version of it that asserts diet drinks contain less methanol than juice is asserted far more often than it is sourced.

Sub-area 6: Headaches, and the only clean randomised human evidence in the topic

17. Controlled challenge trials on aspartame and headache give mixed results and have not established a reproducible effect. Moderate. Evidence type: HUMAN RCT, double-blind randomised crossover challenge. Schiffman and colleagues in 1987 ran a double-blind crossover in 40 subjects who reported repeated headaches after aspartame, challenged with 30 mg per kilogram of body weight or placebo; headache incidence was 35 percent after aspartame against 45 percent after placebo, not significantly different. Sathyapalan and colleagues in 2015 ran the largest study of its kind: a double-blind randomised crossover in 48 self-reported aspartame-sensitive individuals against 48 age- and gender-matched controls, with 100 mg of aspartame in a cereal bar, roughly one diet drink's worth, and symptoms rated over four hours, sessions at least seven days apart. None of the rated symptoms differed between aspartame and control bars, or between sensitive and control participants. Self-identified sensitive participants did rate more symptoms, but concentrated in the first test session and regardless of which bar they received that day, which points at novelty and expectation rather than at the molecule. Against those, Van den Eeden and colleagues in 1994 ran a double-blind two-treatment four-period crossover at approximately 30 mg per kilogram per day, and reported headaches on 33 percent of aspartame days against 24 percent of placebo days, p=0.04; that trial randomised 32 self-identified aspartame-headache subjects but only 18 completed the full protocol, with seven withdrawing because of adverse effects and the remainder for other reasons. (Sathyapalan T et al., PLOS ONE 2015;10(3):e0116212; Van den Eeden SK et al., Neurology 1994;44(10):1787-93, PMID 7936222; Schiffman SS et al., NEJM 1987;317(19):1181-5, PMID 3657889.) Honest synthesis: the positive trial used repeated dosing at roughly 30 mg per kilogram per day, which is close to the entire acceptable daily intake and far above ordinary consumption, in a small sample preselected for believing they react, and fewer than twenty of them finished. The largest and best-controlled sensitivity study, at a realistic single-can dose, found no discriminable response. The defensible line is that some individuals may react, that blinded testing at ordinary doses has not reproduced it, and that self-reported sensitivity has repeatedly failed to survive blinding — not a dismissal, and not a confirmation. Confounds: all these trials are small — Van den Eeden randomised 32 and 18 completed all four periods, Schiffman ran 40, Sathyapalan 48 against 48; all recruit people who already believe aspartame gives them headaches, which maximises nocebo; doses diverge by more than an order of magnitude; headache is subjective and fluctuates; a 33 against 24 percent difference at p=0.04 with that much attrition, where seven of the withdrawals were for adverse effects and so are unlikely to be missing at random, is fragile in more than one way; Sathyapalan's four-hour window would miss delayed reactions; and no trial is large enough to detect a rare true-responder subgroup, so "no average effect" does not exclude "a few real reactors". The funding of these individual trials was not verified in this pass and no inference is drawn from it in either direction. Cui bono: Van den Eeden is cited alone as proof of an effect, without its dose, its attrition or its sample size. Sathyapalan is cited alone as proof that sensitivity is imaginary, without noting the single low dose and the short observation window. Citing all three and letting the disagreement stand is more credible than either resolution.

Sub-area 7: The benefit side, which is where the real argument is

18. WHO's May 2023 guideline conditionally recommends against using non-sugar sweeteners to control body weight or reduce the risk of noncommunicable disease. Strong. Evidence type: systematic review and guideline, synthesising HUMAN RCT evidence on short-term weight outcomes and HUMAN OBSERVATIONAL evidence on long-term disease outcomes. Released on 15 May 2023, two months before the classification, the guideline found that non-sugar sweeteners do not confer any long-term benefit in reducing body fat in adults or children, while longer-term observational data suggested possible associations with increased type-2 diabetes, cardiovascular disease and mortality. The sweeteners covered include acesulfame K, aspartame, advantame, cyclamates, neotame, saccharin, sucralose, and stevia and its derivatives. The caveats must travel with it: the recommendation is conditional, not strong, because WHO judged that the observed links might be confounded by the baseline characteristics of sweetener consumers and by reverse causation; it does not apply to people with pre-existing diabetes; and it excludes personal care products and sugar alcohols. WHO's own rationale is that non-sugar sweeteners are not essential dietary factors, have no nutritional value, and that people should reduce the overall sweetness of the diet. (WHO, "WHO advises not to use non-sugar sweeteners for weight control in newly released guideline", 15 May 2023; WHO guideline: Use of non-sugar sweeteners, 2023.) The disagreement between the two evidence types is the finding, not a flaw in it: the randomised stream, which is the higher-quality one, showed modest short-term reductions in body weight when sweeteners replaced free sugars; the observational stream, which is more confounded, ran the other way over the long term. Reverse causation is the obvious candidate for much of that, and WHO acknowledged it by grading the recommendation conditional. The underlying effect sizes were not independently verified here, so the short-term benefit is stated qualitatively. The class-level benefit question, including trials published after this review's searches closed, belongs to artificial_sweeteners_evidence rather than to this entry. This is the structural backbone of the entry: unknown risk weighed against unconfirmed benefit is a weak trade. That argument needs no harm claim at all, which is why it survives scrutiny that harm claims do not. Cui bono: the beverage sector pushed back hard on this guideline and emphasises its conditional grading and the substitution trials. The alarm side quotes it as "WHO says sweeteners cause diabetes and heart disease", dropping both the conditional grading and WHO's own statement that the association may be confounded.

19. Diet soda genuinely beats sugared soda as a step off sugar, and that concession is load-bearing. Moderate. Evidence type: HUMAN RCT (short-term substitution trials) contrasted with HUMAN OBSERVATIONAL (long-term cohorts). The pattern in WHO's own review — a real short-run substitution effect on body weight that does not persist as a long-run body-fat benefit — is precisely the shape of a bridge and not a destination. Combined with the reaffirmed intake level, which ordinary consumption sits an order of magnitude below, and with human cancer evidence that is null overall in pooled analysis and rests on proxy exposure or on cohorts without a dose-response where the molecule was measured, swapping a sugared soda for a diet soda remains a defensible intermediate step for someone reducing sugar. Water is the finish line. (WHO guideline, 15 May 2023; FAO/WHO JECFA reaffirmation, 14 July 2023.) Confounds: substitution trials are short, often industry-funded, and measure weight rather than metabolic health or long-run dietary pattern. The bridge argument also assumes the person crosses it — part of the long-term observational signal may reflect people who substituted and then stayed on diet drinks indefinitely, which is the failure mode the "not a destination" framing exists to name. No trial has tested "diet soda as a transitional step to water" as an intervention, so the bridge is a reasoned position, not a tested protocol. Cui bono: "diet beats sugar" is the sweetener sector's core commercial claim and is broadly true, but it is sold as an endpoint rather than a transition. The anti-sweetener position benefits from denying the substitution benefit entirely, which requires ignoring the randomised evidence and pushes people back toward sugar.

20. The "it's 99% water" rebuttal is a category error, and so is its mirror image. Not an evidence claim. Evidence type: reasoning and methodology; no empirical assertion made. Composition by percentage says nothing about what the remaining fraction does. Pharmacology is dose-response by molecule, not by proportion of the vehicle — a medication is mostly excipient, and the excipient percentage is irrelevant to what the active ingredient does. The same error runs in the other direction: "it contains a chemical classified by the WHO" is equally uninformative without the quantity. The two most popular arguments in this debate are the same mistake made by opposite camps, and both substitute a proportion or a label for a dose. Take care not to invert it: the point is not that the remaining fraction must therefore be harmful, only that the question is open and must be answered by dose-response data — which is exactly what the JECFA assessment addresses.

Mechanism

Aspartame is a methyl ester of a dipeptide built from two amino acids, aspartic acid and phenylalanine. It is not absorbed intact. In the gut it is hydrolysed into three things: aspartic acid, phenylalanine, and methanol from the cleaved methyl ester. Everything downstream of that hydrolysis is ordinary metabolism of ordinary compounds, which is both the strongest argument in its favour and the reason the topic is difficult — there is no exotic novel molecule circulating to point at.

The phenylalanine is the branch that produces the one certain contraindication. In a person with normal phenylalanine hydroxylase activity, it enters the same pool as the phenylalanine from any dietary protein and is metabolised without incident; a can of diet soda supplies far less than a modest serving of chicken. In a person with phenylketonuria, that enzyme is deficient, phenylalanine accumulates, and the accumulation is neurotoxic. This is why the mandated label exists, and why it specifies a quantity per dose on medicines: it is a counting aid for people already running a lifelong phenylalanine budget.

The methanol branch is where the scare mechanism lives. Methanol is metabolised to formaldehyde and then to formate. Both steps are real; neither is exotic. The quantity is the whole question, and the quantity is small: methanol is roughly eleven percent of aspartame by mass, so the methanol from a can of diet soda is in the region of twenty to thirty milligrams. That is the same order of magnitude as the methanol that pectin demethylation produces in fruit juice, which is the only reason juice is mentioned here — as a familiar yardstick for an unremarkable quantity, not as a recommended drink, since this platform's own evidence has whole fruit associated with lower type-2 diabetes risk while juice runs the other way in the same cohorts. Which of the two contains more methanol is not asserted, because the circulating comparison figures could not be traced to a primary measurement. Human metabolism also generates formaldehyde continuously as a normal intermediate, in quantities that dwarf any dietary contribution, which is the point IARC's conclusion is making when it says there is no evidence the amount formed from aspartame would significantly alter endogenous concentrations. The pathway exists. The flux through it from a soft drink does not stand out against background.

The aspartic acid branch, the excitotoxicity argument, is the weakest of the three in mechanistic terms, because aspartate from aspartame is a trivial addition to dietary aspartate and the blood-brain barrier regulates its entry. It is mentioned here only because it circulates.

The carcinogenic mechanism the classification was actually about is none of those, and it is worth saying what the working group found rather than only what it graded. Two of the key characteristics of carcinogens in the agency's framework were judged to have supporting evidence for aspartame — that it induces oxidative stress and that it induces chronic inflammation — and the evidence for both came from experimental systems. Alongside that, the working group recorded that no high-quality studies of those mechanistic endpoints existed in exposed humans, and it named metabolic pathways and gut-microbiome effects among the work still to be done. That is the honest shape of the mechanistic stream: a coherent story about how oxidative stress and chronic inflammation could produce a liver cancer, sitting next to a human liver-cancer signal, with nothing in humans to connect them. It is also why the evaluation stopped where it did. Read either half alone and you get one of the two loud positions; read both and you get a classification. The wording of this account was taken from a summary of the monograph's General Remarks rather than the primary text, so the substance is reported and the phrasing is unverified.

That leaves the mechanisms that would explain a metabolic effect rather than a cancer effect, and these are the genuinely open ones: whether sweet taste without accompanying calories alters appetite regulation through cephalic-phase responses, whether sweeteners alter the gut microbiome in ways that matter for glucose handling, and whether habitual sweetness maintains a preference that keeps total sugar intake high. These are the substance of the class-level debate and belong to artificial_sweeteners_evidence, not here; what matters for this entry is that none of them is the mechanism the IARC classification was about.

Risks And Contraindications

Phenylketonuria: absolute contraindication. Not a caution, not a warning, an exclusion. People with phenylketonuria must count phenylalanine from every source, and aspartame is a source. This applies to foods and to medicines, and the medicine labelling specifies the quantity per dose precisely so it can be counted. Carriers, meaning heterozygotes without the condition, are not affected. The condition is detected by newborn screening in most high-income countries, so most people who have it already know.

Everything else is uncertainty rather than established harm, and that should be said plainly rather than dressed up either way. There is no established causal harm from aspartame at ordinary intakes. There is also no positive proof of long-term safety at the level people intuitively want, because that proof would require a decades-long randomised trial that has never been run and never will be.

Possible headache reaction in a subset of people. Blinded challenge at a realistic one-can dose did not discriminate sensitive individuals from controls. A smaller trial at repeated high doses reported a modest increase, on a completed sample of eighteen. The trials are too small to exclude a rare genuine responder. If you react, avoiding it costs you nothing.

Cancer: an open question, not an established risk. The classification records a hazard on limited evidence in every stream examined. Pooled across seventeen prospective cohorts there is no association with cancer overall, which is the strongest single piece of human reassurance in the topic. Within that same analysis one blood-cancer row is positive with a dose-response, and two cohorts that measured aspartame as a molecule report small positive associations with cancer overall without a dose-response. The human data measure either a proxy exposure or a molecule in populations where the main confounder is also a main cause of the outcome. Nothing in that set supports a personal risk estimate, in either direction.

What is not established, and must not be inferred here. No claim is made in this entry about aspartame in pregnancy, in children, in migraine as a diagnosed condition, in mood or depression, or in seizure disorders. Those claims circulate; none was verified in this pass; the absence of a claim here is not a safety clearance and is not a warning.

The population-level caution that does hold. WHO's guideline advises against using non-sugar sweeteners for weight control, and explicitly does not apply that advice to people with pre-existing diabetes, for whom the substitution calculus is different. That is a guideline about expected benefit, not a warning about harm, and it should not be quoted as the latter.

Controversy

Nature of the dispute. There are two disputes running at once and they are usually conflated. The narrow scientific dispute is whether aspartame poses any carcinogenic risk at dietary exposures — a question on which the human data are observational and confounded, the rodent data are contested, and the mechanistic data were rated limited and exist only in experimental systems. There is now a third, narrower dispute sitting inside the first: whether the hazard evaluation itself was conducted to the agency's own procedural standard, which senior former officials of that programme have argued in print that it was not. The broad public dispute is about what a hazard classification means, and it is not really a scientific dispute at all; it is a literacy problem, in which one side reads "possibly carcinogenic" as "carcinogenic" and the other reads "limited evidence" as "no evidence". The remarkable thing about July 2023 is that WHO pre-empted both misreadings in a single document, and both happened anyway.

Position A: the precautionary reading, and where it overreaches. The best of this case: three independent prospective cohorts, in different populations, produced a positive association with liver cancer pointing the same direction, IARC judged those studies to be of high quality and adjusted for many confounders, and a WHO working group of twenty-five scientists judged the association substantial enough to act on. Two cohorts that measured aspartame as a molecule rather than as a beverage category have since reported small positive associations with cancer overall. A pooled analysis of seventeen cohorts reports a leukaemia association with a dose-response, in the same tumour family as the rodent findings. The mechanistic stream was not empty: oxidative stress and chronic inflammation both had supporting evidence in experimental systems, which is a coherent route to a liver cancer. The Ramazzini bioassays reported tumours in a design that arguably sees late-onset cancers a standard 104-week protocol cannot, a documented minority of the working group did not agree that the objections would change an interpretation of sufficient evidence, and senior figures who used to run hazard-evaluation programmes have argued in print that the animal appraisal departed from the agency's own procedure. Aspartame has never been subjected to a long-term randomised human trial and never will be, so the observational data are the only human data that will ever exist. And most importantly, WHO's own May 2023 guideline found no long-term body-fat benefit, meaning the reason to accept any residual uncertainty is itself unconfirmed. Where it overreaches: the same pooled analysis that carries the leukaemia row is null for cancer overall, and carries a protective-looking colorectal row that nobody in this camp cites; neither molecule-level cohort found the liver endpoint or any dose-response, and one of them overlaps a cohort the classification already used; the exposure in the leukaemia strand is the beverage category again; the critique of the appraisal is a commentary rather than new data, does not answer all of the European regulator's 2013 objections, and comes with an undeclared funding position and a declared editorial conflict; the mechanistic evidence has no human component at all; the formaldehyde argument collapses on quantity; and "possibly carcinogenic" is repeatedly and knowingly reported as "carcinogenic", which is the point at which a defensible caution becomes a false claim.

Position B: the reassurance reading, and where it overreaches. The best of this case: two regulatory bodies, JECFA and EFSA, reviewing separately and years apart, arrived at the same acceptable daily intake of 40 mg per kilogram, and ordinary consumption sits roughly an order of magnitude below it. The American regulator, reviewing the same evidence, did not accept the classification at all, citing significant shortcomings in the studies it relied on. Pooled across seventeen prospective cohorts there is no association between artificially sweetened beverages and cancer overall, which is a positive null rather than a mere absence of data. No mechanistic study of the relevant endpoints has ever been done in exposed humans. The rodent studies driving alarm came from one institute and failed methodological review at EFSA on specific, stateable grounds, including a colony in which control animals not fed aspartame were also diseased and an absence of dose–response in the reported tumours. The human evidence is observational, uses a proxy exposure in the studies the classification rested on, and has a confounder that is also the disease. The metabolites are ordinary compounds in unremarkable quantities. The largest blinded sensitivity trial found that people who identify as aspartame-sensitive report symptoms whether or not they received aspartame. And there is a precedent in this exact chemical class for a rodent tumour scare being resolved rather than merely rebutted: saccharin's bladder tumours were traced to a rat-specific mechanism, the classification was downgraded, the listing removed and the warning label repealed. Where it overreaches: treating Group 2B as an administrative nothing understates that a WHO working group found a human signal in high-quality cohorts and could not explain it; the animal grade was not unanimous and its adequacy is now contested by people who used to run that evaluation programme; the same pooled analysis that supplies the overall null also supplies a leukaemia dose-response in the tumour family the reassurance case dismisses for lacking one; two cohorts measuring the molecule have found small positive associations; the regulatory agreement is less unanimous than it sounds, since the US FDA has long set its own intake level 25 percent higher at 50 mg per kilogram, which shows the number is a judgement about margin rather than a measured threshold; a regulator that has authorised a substance for four decades is not a disinterested reader of a classification implying it should not have, which is the same asymmetry argument this entry makes about agencies that classify; the saccharin precedent is a different molecule, site and mechanism and predicts nothing; "the dose makes the poison" is true but does not answer a question about lifetime low-dose exposure that no trial design has ever addressed; the fruit-juice methanol comparison borrows credibility from a beverage with its own problems, and is usually asserted in a stronger form than any traceable measurement supports; and the whole position is usually deployed to defend the product's benefit, which is the part WHO's own guideline declines to support.

The funding and bias dimension — cui bono, both ways.

Running toward aspartame: the global sweetener and diet beverage market is enormous, aspartame is off-patent and cheap, and reformulating away from it costs money and shelf space. The International Sweeteners Association and equivalent trade bodies fund research, fund methodological critiques of unfavourable studies, and run public information sites; the nine-to-fourteen-cans figure is their most effective single line and it is accurate. Some of the published re-analyses of the aspartame bioassays have funding that was not verified in this pass, which is a reason to read them with the same scepticism as the studies they criticise, not a reason to dismiss them. The beverage sector also pushed back hard against WHO's May 2023 guideline, which threatens the product's entire commercial rationale far more than a hazard classification does. And there is a subtler interest: as long as the public argument is about cancer, an argument the industry can win on the pooled human evidence, it is not about whether the product works.

Running against aspartame: the "clean label", natural sweetener, stevia and monk fruit categories all gain directly from aspartame fear, and are commercial actors with the same incentives in the opposite direction. Wellness media monetises the scare headline, which is close to costless to produce and reliably shared. Campaign organisations gain organisational relevance from a WHO classification and have little incentive to explain what "limited evidence" means. The Ramazzini Institute has an institutional stake in the significance of its own findings, as any laboratory does. The 2025 critique of the working group's appraisal is not from that institute, which is what makes it worth citing, but two of its authors serve on the publishing journal's editorial team and its funding is not declared — and a scepticism this entry applies to industry-funded critiques has to apply here too. Regulators, meanwhile, have their own asymmetry in both directions: an agency that classifies gets safety credit at low cost, while an agency that reaffirms an intake level, or that publicly rejects a classification of a substance it authorised forty years ago, absorbs the blame if anything is later found — which is a reason to read every institutional statement in this file as cautious in its own direction rather than as one honest position and one captured one.

The genuinely neutral parties are thinner on the ground than either camp claims. The epidemiologists who built the cohorts were mostly studying diet and cancer generally, not aspartame specifically. The funding of the blinded headache challenge trials was not verified in this pass, and at least one of those investigators has documented ties to the sweetener sector, so no claim is made here that this small literature is unmotivated — only that it produced results in both directions, which is the reason all three trials are cited rather than the convenient one.

Realised Position: the aspartame cancer question is open, weak, and not where the decision should be made. What is known: the classification is a hazard identification on limited evidence in all three streams, published alongside a reaffirmed intake level that ordinary consumption sits far below, and there is exactly one established contraindication, which is phenylketonuria. We will not say aspartame causes cancer, because the evidence does not say that and the entry would deserve to be discarded if it did — the largest pooled human analysis of this exposure is null for cancer overall, and that is now the first thing we say about the harm side rather than the last. We will also not say the question is settled, because a WHO working group found a human signal in three cohorts it rated high quality and could not account for it, because two cohorts that measured the molecule itself have since reported small positive associations with cancer overall, because the same pooled analysis that supplies the null also reports a leukaemia dose-response in the exact tumour family the rodent studies produced, and because the animal grade those studies received was not unanimous and its adequacy is contested in print by people who used to run that kind of evaluation. Each of those concern strands is narrow, confounded, and short of a dose-response where it matters most; none of them licenses a harm claim; and a reassurance case built by omitting them is not a reassurance case we would sign. The decision that matters is the other one: the benefit these molecules were invented for does not hold up over the long term in WHO's own systematic review, so a person is accepting unresolved uncertainty in exchange for an unconfirmed benefit. That is a weak trade. It argues for reducing overall sweetness rather than for swapping sweeteners, and it argues just as strongly against going back to sugar. Diet soda remains a real improvement on sugared soda for someone stepping off sugar. It is a bridge. Water is the finish line, and the bridge is only worth building if you cross it.

Cross-Pillar Connections

Diet. This entry is a worked example of a reasoning pattern that recurs across the diet corpus: a proxy exposure measured in an observational cohort, a confounder that is also the outcome's main cause, and a regulatory statement compressed past the point of meaning. The same structure appears wherever a food category is studied rather than a molecule, and the general machinery lives in rct_vs_observational_evidence and healthy_user_bias. The liver cancer signal here is the cleanest illustration of reverse causation in the corpus, because the people most likely to drink diet soda and the people most likely to develop hepatocellular carcinoma are, to a large extent, the same people for the same underlying reason. It is also a worked example of what it looks like when the proxy problem is finally fixed: two cohorts measured the molecule, and the answer they returned was smaller, different in site, and without a dose-response — which is how evidence usually improves, not by confirming the headline but by changing the question.

Diet, again, on the marker-versus-outcome axis. Much of the broader sweetener literature reports movements in glucose response, insulin, or microbiome composition rather than in anything a person experiences, and the reasoning about why that is weak evidence belongs to surrogate_endpoints_vs_outcomes. The aspartame case adds a specific wrinkle: the endpoint that actually got measured in the cohorts, cancer mortality in one case and incidence in another, are not the same endpoint and were pooled in public discussion as if they were.

Diet, on the substitution question. The bridge argument here is the same argument the platform makes about whole food versus isolated components, run in reverse: a substitution can be a genuine improvement at the moment it is made and a poor place to stop. The fruit comparison in this entry is a scale calibration only — this platform's evidence has whole fruit associated with lower type-2 diabetes risk while juice runs the other way in the same cohorts, so nothing here should be read as an endorsement of juice as a beverage.

Mental. The headache trials are one of the better nocebo demonstrations available in a dietary context: people who identified as aspartame-sensitive reported more symptoms than controls, concentrated in their first test session and regardless of whether they received aspartame. That does not mean their symptoms are imaginary — reported symptoms are real symptoms — but it does mean the attribution failed under blinding. The general lesson is that expectation is a genuine physiological input, not a rhetorical dismissal, and it is why unblinded self-experiments on subjective symptoms produce confident conclusions that do not replicate.

Mental, on the anxiety cost. A meaningful share of the harm this topic causes is not metabolic. It is the low-grade dread produced by believing that a drink consumed daily for years is doing something irreversible, on the basis of a headline that misreported a document that said the opposite. Correcting that is a genuine intervention, and it costs nothing. Correcting it honestly means also saying what has not been ruled out, because reassurance that later turns out to have been selective is worse than uncertainty stated plainly.

Physical and sleep, practically. Many diet colas carry caffeine, and for someone drinking several through the afternoon the caffeine is a far more defensible thing to change than the sweetener, with an effect on sleep that is immediate, measurable, and not in scientific dispute. When the sweetener question is consuming the attention, the caffeine timing is usually the intervention actually available.

What would change our mind

Industry bias note

Sources (30)

• *Primary regulatory documents**↗
• World Health Organization / IARC / FAO-WHO JECFA, "Aspartame hazard and risk assessment results released", joint news release, 14 July 2023. https://www.who.int/news/item/14-07-2023-aspartame-hazard-and-risk-assessment-results-released — the single document containing both the Group 2B classification and the reaffirmed acceptable daily intake, including the nine-to-fourteen-cans worked example and the hazard-versus-risk statement.↗
• IARC Monographs on the Identification of Carcinogenic Hazards to Humans, Volume 134 (Aspartame, methyleugenol and isoeugenol), published 2024, General Remarks. https://www.ncbi.nlm.nih.gov/books/NBK609284/ — the evidence grades, the identified cohorts and the working group's appraisal of their quality, the mechanistic key characteristics with supporting evidence, the absence of high-quality human mechanistic studies, the inadequate grade for non-liver cancers, the metabolite and formaldehyde conclusion, and the stated research needs. The mechanistic detail used here was taken from a summary of the General Remarks rather than read off the primary text and is flagged as such in the claim.↗
• Summary of the IARC evaluation published in The Lancet Oncology, 13 July 2023 — the working group of 25 scientists from 12 countries, meeting at IARC in Lyon in June 2023, and the three "limited" gradings for aspartame.↗
• IARC Monographs Preamble and list of classifications. https://monographs.iarc.who.int/ — the definitions of the evidence categories, which differ by stream, and the classification groups.↗
• IARC Monographs Volume 56 (1993), Pickled Vegetables (Traditional Asian). https://inchem.org/documents/iarc/vol56/02-pick.html↗
• IARC reclassification of saccharin and its salts from Group 2B to Group 3, 1999 — on the basis of a bladder-tumour mechanism in rats judged not relevant to humans owing to interspecies differences in urine composition.↗
• United States National Toxicology Program, Ninth Report on Carcinogens (2000) — removal of saccharin from the list of substances reasonably anticipated to be a human carcinogen; federal legislation removing the mandated saccharin warning label signed 21 December 2000; California delisting effective 6 April 2001.↗
• World Health Organization, "WHO advises not to use non-sugar sweeteners for weight control in newly released guideline", 15 May 2023, and the accompanying guideline Use of non-sugar sweeteners. https://www.who.int/news/item/15-05-2023-who-advises-not-to-use-non-sugar-sweeteners-for-weight-control-in-newly-released-guideline↗
• EFSA ANS Panel, "Scientific Opinion on the re-evaluation of aspartame (E 951) as a food additive", EFSA Journal 2013;11(12):3496. https://efsa.onlinelibrary.wiley.com/doi/10.2903/j.efsa.2013.3496↗ — the independent European review reaching the same 40 mg per kilogram figure, and the assessment of the Ramazzini studies.
• EFSA, "Evaluation of Ramazzini Institute Aspartame Studies and EFSA's Assessment". https://www.efsa.europa.eu/sites/default/files/event/documentset/130409-p06.pdf — the stated grounds, including the background of chronic inflammatory changes in the colony and the absence of a positive dose–response relationship.↗
• United States Food and Drug Administration, aspartame and sweetener guidance — the US acceptable daily intake of 50 mg per kilogram of body weight per day, higher than the JECFA and EFSA figure of 40.↗
• National Cancer Institute, artificial sweeteners and cancer — the summary from which this entry takes the American regulator's disagreement with the classification and the statement that no clear evidence of an association between saccharin use and human bladder cancer has emerged from epidemiological studies. This is the cancer institute's characterisation of the regulator's position; the regulator's own statement was not retrieved.↗
• 21 CFR 172.804 (Aspartame). https://www.ecfr.gov/current/title-21/chapter-I/subchapter-B/part-172/subpart-I/section-172.804 — the mandated "Phenylketonurics: Contains Phenylalanine" statement and the molecular formula.↗
• 21 CFR 201.21 (Declaration of presence of phenylalanine as a component of aspartame in over-the-counter and prescription drugs). https://www.ecfr.gov/current/title-21/chapter-I/subchapter-C/part-201/subpart-A/section-201.21↗
• *Human observational studies**↗
• Stepien M et al., "Consumption of soft drinks and juices and risk of liver and biliary tract cancers in a European cohort", Eur J Nutr 2016;55(1):7-20.↗
• McCullough ML et al., "Sugar- and artificially-sweetened beverages and cancer mortality in a large US prospective cohort", Cancer Epidemiol Biomarkers Prev 2022;31(10):1907 — 934,777 cancer-free participants enrolled 1982, deaths tracked through 2016. Endpoint is mortality, not incidence.↗
• Jones GS et al., "Sweetened beverage consumption and risk of liver cancer by diabetes status", Cancer Epidemiol 2022 — a pooled analysis stratified by diabetes status. Note for downstream use: this study is sometimes attributed to UK Biobank in secondary summaries. That attribution was not verifiable and should not be repeated.↗
• Debras C, Chazelas E, Srour B et al. (Touvier M, senior), "Artificial sweeteners and cancer risk: Results from the NutriNet-Santé population-based cohort study", PLOS Medicine 2022;19(3):e1003950 — 102,865 French adults, 78.5 percent women, median follow-up 7.7 years, 3,358 incident cancers; molecule-level exposure; total sweeteners 1.13 (1.03–1.25), aspartame 1.15 (1.03–1.28), acesulfame K 1.13 (1.01–1.26), sucralose null; breast 1.22 (1.01–1.48) and obesity-related 1.15 (1.01–1.32) with aspartame; volunteer cohort, 24-hour dietary records, authors name reverse causality.↗
• "Association between Aspartame Consumption and Cancer Risk: Evidence from a Large Prospective Cohort", Journal of Nutrition, February 2026, article S0022-3166(26)00109-4 — American Cancer Society nutrition cohort, 100,004 US adults, intake assessed 1999 and 2003 at molecule level, 21,356 verified cancer cases through June 2017 including 8,087 obesity-related; association with higher cancer risk regardless of weight status, with the authors noting the lack of a dose-response. Hazard ratios and confidence intervals could not be retrieved and are deliberately not printed. Not independent of the cohort family the classification already used.↗
• Yin S et al., "Artificially Sweetened Beverage Consumption and Cancer Risk: A Comprehensive Dose–Response Meta-Analysis of Prospective Studies", Nutrients 2022;14(21):4445 — 14 articles across 17 cohorts, search window January 1967 to September 2022; overall cancer 1.03 (0.96–1.11) p=0.407; non-lymphoid leukaemia 1.35 (1.03–1.77) p=0.030 with a linear dose-response of 1.15 (1.02–1.30) per 355 mL daily serving; colorectal 0.78 (0.62–0.99). Effect sizes from the published abstract plus a corroborating retrieval; full text blocked, so per-cohort contributions and heterogeneity statistics are unverified.↗
• *Human randomised challenge trials**↗
• Sathyapalan T et al., "Aspartame Sensitivity? A Double Blind Randomised Crossover Study", PLOS ONE 2015;10(3):e0116212. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0116212↗ — 48 self-reported sensitive against 48 matched controls, 100 mg aspartame in a cereal bar, sessions at least seven days apart, four-hour rating window, null; sensitive participants rated more symptoms particularly in the first session, whichever bar they received.
• Van den Eeden SK et al., "Aspartame ingestion and headaches: a randomized crossover trial", Neurology 1994;44(10):1787-93, pubmed.ncbi.nlm.nih.gov/7936222↗/" target="_blank" rel="noopener">PMID 7936222↗ — 32 randomised, 18 completing the full four-period protocol, seven withdrawing for adverse effects, approximately 30 mg/kg/day, 33 percent against 24 percent headache days, p=0.04.
• Schiffman SS et al., "Aspartame and susceptibility to headache", NEJM 1987;317(19):1181-5, pubmed.ncbi.nlm.nih.gov/3657889↗/" target="_blank" rel="noopener">PMID 3657889↗ — 40 subjects, 30 mg/kg, 35 percent against 45 percent, null.
• *Rodent evidence and the dispute over its appraisal**↗
• Soffritti M et al., Ramazzini Institute lifespan aspartame bioassays. Assessed and not accepted in the EFSA 2013 opinion above; rated as limited animal evidence by IARC in 2023.↗
• Landrigan PJ, Straif K, Birnbaum LS, McDiarmid M, Perry MJ, Forastiere F, Comba P, Bucher JR, "Irregularities in the IARC Working Group Evaluation of Ramazzini Institute Aspartame Studies", Annals of Global Health, 6 June 2025, doi 10.5334/aogh.4771↗ — argues the working group did not articulate or justify its negative evaluation of the animal evidence, alleges divergence from the agency's own procedures, and records a minority opinion within the working group. Declared competing interest: two authors serve on the journal's editorial team. Funding is not declared in the accessible content and could not be established. The minority-opinion wording was taken from the journal's article page after the archive mirror was blocked, so no phrasing is quoted here.
• *Deliberately not cited, or cited only at stated strength.** Milligram-per-litre comparisons of methanol content between diet soft drinks and fruit juice, which trace only to industry and advocacy fact sheets and could not be traced to a primary measurement — which is also why no direction is claimed for that comparison. Any count of the number of agents in Group 2B. Any claim about the current classification status of other named substances. The hazard ratios and confidence intervals of the February 2026 cohort, which were blocked at both the publisher and the journal page and must not be supplied from any secondary source. The American regulator's own statement on the classification, which was not retrieved; only the cancer institute's characterisation of it is used. Verbatim wording from the 2025 critique of the working group's evaluation, and verbatim wording of the working group's mechanistic key-characteristic statements, both of which came from summaries or article pages rather than primary text. The finer mechanistic account of the saccharin bladder findings — urinary pH, calcium phosphate microcrystals, cytotoxicity, regenerative hyperplasia, sex specificity — which was not verified to primary text. The funding sources of the three headache challenge trials, and of the 2025 critique, none of which were verified and from which no inference is drawn.↗