Cancer Inputs:
What You Take In

What Was Classified, and What You Were Told

There is a public body whose entire job is to decide whether something causes cancer in humans. It is the International Agency for Research on Cancer (IARC), the cancer arm of the World Health Organization. Working groups of scientists review the published evidence on one agent at a time and place it in a category. The findings are public, free, and have been for decades.

Most people have never seen that list. Not because it is hidden, but because nothing in ordinary life brings it to them. It is not on the package. It is not in the pharmacy leaflet. It is not raised at the appointment. The information exists and the route from it to you was never built.

This page is that route, for the things you take in on purpose: what you drink, what you are prescribed, what goes in your mouth, what you breathe at home. Each entry gives the classification where one exists, the mechanism behind it, and what the evidence will and will not carry. Where this site has a full article on something, the entry links to it rather than repeating it.

Nothing here tells you what to do. Several of these are things reasonable people continue, knowing what they now know. That is what consent is.

First, What the Categories Actually Mean

The groups are frequently misreported in both directions, so it is worth being precise before reading any of them.

• ●Group 1, carcinogenic to humans. The evidence in people is sufficient. There is no stronger category. Tobacco smoke, alcohol, asbestos, arsenic and formaldehyde are here.
• ●Group 2A, probably carcinogenic. Strong enough that the working group found it probable, not merely possible. Glyphosate and drinks above 65 °C are here.
• ●Group 2B, possibly carcinogenic. Limited evidence in humans. Aspartame has been here since 2023.
• ●Group 3, not classifiable. The evidence is inadequate to say either way. This is not a clean bill of health, and it is routinely reported as one.

Classifications also change as evidence accumulates. Saccharin was removed from the list in 1999. Coffee was re-evaluated in June 2016 and its earlier classification withdrawn. A list that never changed would be a worse list. There is also no longer any category meaning "does not cause cancer": Group 4 was abolished in 2019.

Tobacco and Nicotine

Group 1, and the least surprising entry on this page. What is less known is how specific and how reversible the numbers are.

The largest review to date pooled 205 studies on stomach cancer alone. Current smokers carried 1.53 times the risk of never-smokers (95% confidence interval 1.44 to 1.62). For cancer at the gastric cardia, the part of the stomach joining the esophagus, it was 2.08 times (1.66 to 2.61). Former smokers sat at 1.30, and the figure kept falling with time stopped, reaching 0.65 at thirty years quit (0.44 to 0.95).

That is risk of getting it. The part almost nobody is told concerns people who already have a diagnosis. Reviewing studies that compared people who quit after diagnosis against those who continued, the 2020 US Surgeon General's report found quitting associated with a median 45% reduction in overall mortality, and continuing associated with roughly 50% higher overall and 61% higher cancer-specific mortality. The report is careful to call this suggestive rather than proven, because nobody can be randomized to keep smoking.

Timing mattered. Benefit was largest for those who entered a cessation program within three to six months of diagnosis, and no significant survival association was found for those starting more than five years out.

On vaping, the replacement question is its own subject. Vaping and nicotine covers what the aerosol contains, including formaldehyde, a Group 1 carcinogen.

Alcohol

Group 1. Not "linked to." Not "associated with." The same category as tobacco smoke and asbestos, and it has been there since 2012.

The mechanism is not mysterious. Ethanol is converted in the body to acetaldehyde, which damages DNA directly and is itself classified as carcinogenic to humans. Seven cancer sites are recognized. There is no amount the classification identifies as safe, which is a different statement from saying any amount is dangerous, and the difference matters.

Full detail, including the seven sites and what the industry-funded research did with the moderate-drinking question, is in Alcohol: What You Weren't Told.

Hot Drinks: The Temperature, Not the Drink

This is the cleanest actionable finding on the page and the one almost nobody has heard, partly because the headline in 2016 was about coffee being cleared and the actual finding was about something else entirely.

In 2016 a working group of 23 scientists classified drinking beverages above 65 °C as Group 2A, probably carcinogenic to humans, for cancer of the esophagus. In the same review IARC found insufficient evidence to link coffee itself to esophageal cancer and withdrew coffee's earlier classification. The agency said plainly that it is the temperature a drink is consumed at, not the type of drink, that carries the risk. The earlier, narrower finding about hot maté was generalized: the issue was never the plant.

The proposed mechanism is worth understanding, because it explains why this interacts with everything else you take in. Thermal injury is thought to impair the barrier function of the lining of the esophagus, which increases that tissue's exposure to carcinogens such as N-nitroso compounds. Tobacco smoke is a major source of exactly those compounds.

One review argues the human evidence has strengthened enough that the classification is likely to be upgraded in coming years. That is the author's view and not an IARC decision.

What follows from it is free. Let a hot drink stand about four minutes, or until it can be drunk without wincing. The exposure is the temperature, so letting it cool removes it without giving anything up. Of everything on this page, this asks the least and rests on a formal classification.

Coffee, Tea and Caffeine

Separate from the temperature question, and the answers differ by what is being asked.

Coffee as a drink has been studied hard and comes out clean for stomach cancer. Twelve prospective studies pooled gave a relative risk of 1.12 for highest versus lowest intake, with a confidence interval of 0.93 to 1.36. An interval crossing 1 means no effect was detected. A dose-response analysis found 1.07 per three cups daily, and a nine-cohort analysis of about 1.25 million people found 1.05. The authors of the first noted that apparent signals in some subgroups were probably smoking hiding inside the coffee data, which is a recurring problem in this literature and a good reason to read the adjustments rather than the abstract.

What is in the coffee is a separate question from the coffee. Ochratoxin A, a mold toxin that accumulates in the kidneys, is classified Group 2B and is a storage and sourcing issue rather than an inherent one.

Tea carries an interaction that is rarely mentioned and matters most to the people least able to afford it. The tannins in black tea bind non-heme iron, the form found in plant foods, and prevent its absorption. A classic controlled meal study found reductions of 79 to 94% under its conditions; other trials put it in the 50 to 70% range depending on strength and quantity. The effect is dose-dependent, so all-day drinking is a high dose.

For anyone already losing iron, which includes people with heavy periods and people with any source of slow gastrointestinal bleeding, drinking tea through the day blocks the absorption of the iron they do eat. A controlled trial using traceable iron found that leaving about one hour between an iron-containing meal and the tea substantially blunts the interference. Heme iron, the kind in meat, is largely unaffected either way.

The fix is timing rather than abstinence, which is the whole point of knowing about it. Caffeine covers the rest, including what caffeine does to sleep, cortisol and appetite.

Pharmaceuticals: Some Are Formally Classified

This is the entry most people have never been shown any part of, and it needs stating carefully, because the honest version is neither "drugs cause cancer" nor "this has been ruled out."

A number of medicines are classified Group 1: carcinogenic to humans. Not suspected. Classified, in the same category as tobacco smoke. A 2023 systematic review and meta-analysis in Scientific Reports went through the Group 1 pharmaceuticals and pooled the human evidence for each. These are its summary relative risks:

And here is the part that keeps this honest. Most of that table is chemotherapy and transplant medicine, given to people who would die without it. The review's own conclusion was not that these should be withdrawn. It called for closer surveillance, better strategies for the patients who need them, and re-categorization of some agents where the evidence has not held up. It noted that few alternatives exist. The authors suggested cyclosporine may not belong in Group 1 at all, since its link to blood cancers did not reach significance.

That is what a real risk-benefit decision looks like, and it is the reader's to make with their prescriber. The consent gap is not that these drugs exist. It is that the classification is rarely part of the conversation in which they are accepted.

Everyday medicines, and what regulators have already done

The table above is oncology and transplant. These are not.

• ●Hydrochlorothiazide, a very widely prescribed blood pressure diuretic, and the clearest example on this page of a cancer signal found in people after approval rather than in rodents before it. Europe acted first: EMA and PRAC added warnings in 2018, on the strength of Danish registry case-control studies. The FDA followed in August 2020 after its own Sentinel cohort analysis, which located the signal mainly in squamous cell carcinoma, in White patients, at large cumulative doses.
  
  The Danish dose-response is striking. For squamous cell carcinoma, ever-use came out at 1.75 times the risk of non-users (1.66 to 1.85, so solidly detected), rising to 7.38 times at the highest cumulative dose. Basal cell carcinoma was far weaker, 1.08 ever-use rising to 1.54. Lip squamous cell carcinoma ran on its own, lower dose thresholds.
  
  Now the number that puts those in proportion, and it is the one almost never quoted alongside them. FDA characterised the overall excess in its study population as roughly one additional squamous cell carcinoma per 16,000 treated patients per year. Both figures are true. The 7.38 applies to people at the top of twenty years of cumulative dosing; the one-in-16,000 is what it averages to across everyone taking it. This is exactly why a ratio without a baseline is half a fact.
  
  The agency's own advice was to continue the medication and use sun protection and skin surveillance unless a provider directs otherwise. And it does not generalise to the class: the same review found no comparable signal for bendroflumethiazide or indapamide.
• ●Ranitidine, sold as Zantac. In 2020 the FDA requested removal of all ranitidine products from the market over contamination with NDMA, a nitrosamine. Not a label change. Withdrawal. The reason it went further than a contaminated-batch recall is that NDMA formed in the product over time, rising with storage age and heat, so passing a test on release could not guarantee compliance through shelf life. And it is back: the FDA approved a reformulated ranitidine on 24 November 2025, after five years off the market. Read the caveat though. Complete elimination of NDMA formation was not confirmed, and the new labeling carries strict storage and handling requirements to keep it down.
• ●Valsartan, losartan and irbesartan, common blood pressure medicines. Recalls began in July 2018 over NDMA contamination, with NDEA, a second nitrosamine, found in later batches. The contamination came from manufacturing, not from the drug molecule, which is a different kind of problem and one a patient has no way to see. These were selected manufacturers and lots, not a withdrawal of the class.
• ●And a distinction that matters for all of the contamination cases. A nitrosamine recall rests on analytical testing against a calculated exposure limit, not on anyone having been shown to develop cancer. FDA's acceptable intakes for these compounds are small, 96 nanograms a day for NDMA and 26.5 for NDEA, and they are derived from carcinogenic-potency assumptions. They are deliberately conservative. Exceeding one is a reason to act; it is not a finding that recipients were harmed, and it is not the same category of evidence as the skin cancer or bladder cancer signals above.
• ●Pioglitazone, for type 2 diabetes. Bladder cancer signal, and the regulators split on what to do about it. France suspended prescribing in 2011. Germany recommended not starting new patients, which is not the same thing as a suspension. The EMA kept it authorized with contraindications. And the US did act, contrary to how this is often told: FDA warnings in 2010 and 2011, a labeling change in August 2011, and another update after its December 2016 review. So the disagreement was never about whether anyone noticed the signal. It was about what to do under uncertainty. Worth knowing too that bladder tumors had already appeared in pre-approval rat studies, so this is not a risk that surfaced only after approval.

And one where a warning was taken back off

Becaplermin, a topical growth factor for diabetic foot ulcers, received an FDA boxed warning in 2008 for increased cancer mortality, on the strength of a retrospective postmarketing cohort, with the signal concentrated in patients dispensed three or more tubes.

In 2018 the boxed warning was removed. Longer follow-up and additional cohorts did not sustain the original finding, with later cancer-mortality estimates sitting essentially at no effect. Sparse events and unstable estimates in a small high-exposure subgroup are the plausible explanations, though non-replication on its own does not establish which one produced the first result.

And one where the frightening number shrank

Including this is the point of the page rather than an exception to it.

Proton pump inhibitors, the long-term acid blockers, have been reported as raising stomach cancer risk by 1.3 to nearly 3 times, and a 2025 analysis in high-prevalence regions found an odds ratio of 2.02, meaning twice the odds compared with non-users, with an interval of 1.35 to 3.01 that sits entirely above 1. Alarming, and widely circulated.

Then the adjustments. A meta-analysis that pooled only the six studies with comprehensive adjustment for confounding found no association: 1.07 (0.97 to 1.19), and reported explicitly that weaker adjustment went with larger effect sizes. The main confounder is Helicobacter pylori, a stomach bacterium that is itself the principal known cause of this cancer and is also a reason people are prescribed acid blockers in the first place. One review concluded it seems unlikely that these drugs are an independent carcinogen, and more likely that they act on a pathway H. pylori has already started. The single randomized trial with cancer as an outcome found no increased risk.

What any individual drug has documented, including depletions and warnings drawn from the labels themselves, is in the drug library.

Peptides: No Classification, and That Is the Problem

Therapeutic peptides have moved from research into ordinary use, often bought online and often taken by people managing a chronic condition. None of the research peptides appear on the IARC list at all, and the reason matters: nobody has evaluated them. As the companion page on reading classifications explains, absence from the list is a statement about the research, not about the substance.

What can be said is mechanistic, and for one class it is specific enough to be worth stating plainly.

The angiogenesis problem

BPC-157 is sold for tissue repair, and the mechanism it is sold on is the problem. It is pro-angiogenic: it promotes the growth of new blood vessels, substantially through VEGF signaling. That is precisely why it is marketed for healing.

What the evidence does and does not show. No human study has shown BPC-157 causing or accelerating tumors. There is also no long-term human safety data at all, and no human cancer studies in either direction. Animal work in tumor-bearing models looked at cachexia rather than tumor growth, and reported tumors continuing to grow without measurable acceleration. A 2004 cell study suggested reduced melanoma cell proliferation and appears not to have been replicated. The prevailing advice is to avoid it where there is a cancer history until human data exists.

A note on the sources, because it bears on how to read anything about this. Almost all the available writing on BPC-157 is published by companies selling it, and the most substantial review in the scientific literature is written by the peptide's own proponents. One claim circulating is that an FDA briefing document cited tumor promotion as a reason for excluding it from the compounding bulks list; that could not be verified and is left here as unconfirmed rather than repeated as fact.

The same question applies to growth-hormone secretagogues, the peptides that raise growth hormone and therefore IGF-1. Anything that raises a growth signal deserves the same question in anyone with a cancer history, and the specific evidence on IGF-1 is outside what this page has verified.

The one peptide class that does carry a cancer warning

GLP-1 receptor agonists, the weight and diabetes drugs, are peptides, and they are the exception: they carry a formal warning.

• ●Semaglutide carries a boxed warning for thyroid C-cell tumors. In rodents it caused dose-dependent and duration-dependent C-cell tumors at clinically relevant exposures. The label states plainly that whether it causes such tumors, including medullary thyroid carcinoma, in humans is unknown.
• ●It is contraindicated in anyone with a personal or family history of medullary thyroid carcinoma, or with multiple endocrine neoplasia type 2. Papillary or differentiated thyroid cancer alone is not a listed contraindication.
• ●Liraglutide carries the same boxed warning, and cases of medullary thyroid carcinoma have been reported after marketing. The regulatory language is careful: the data are insufficient to establish or exclude a causal relationship.

And the human evidence has now arrived, which complicates it

The boxed warning rests on rodents. Since then a great deal of human data has accumulated, and it does not line up neatly in either direction.

The first-year figure is the interesting one. A risk that appears in year one and then flattens is what you would expect if the drug were finding cancers rather than causing them. People starting a new medication see clinicians more, get more imaging, and get more thyroid nodules looked at. The authors of that study named enhanced early detection as a plausible explanation themselves. Timing alone does not prove it, and a null result on all thyroid cancer cannot rule out a small increase in medullary cancer specifically, because medullary cancer is rare enough that these studies have few cases of it.

And the regulator looked and declined to act. In 2023 the EMA's pharmacovigilance committee reviewed the French signal and concluded the data did not establish a causal association, making no change to the product information. That is not a finding of no risk. It is a finding that the evidence did not support a change, which is a different statement.

What any individual drug's label documents is in the drug library, and the logic of absent evidence is in How to Read an IARC Classification.

Supplements: The Zinc Case

Supplements are an input almost nobody classifies, and zinc is the one with the most carefully studied cancer signal attached to it. It is worth going through properly, because the shape of this finding is unusual and almost always reported wrongly in both directions.

The numbers

The main evidence is the Health Professionals Follow-up Study, roughly 47,000 men followed from 1986. Its 30-year update reported supplement dose against four different outcomes:

Duration behaved the same way. At 15 years or more of use, advanced disease came out at 1.93 (1.08 to 3.44) and lethal disease at 1.91 (1.28 to 2.85), while the risk of getting prostate cancer at all stayed flat at 1.04.

What the official limits are, which is a separate question

• ●United States: 40 mg/day of total zinc from everything combined, food included. Set from the dose at which a copper-dependent enzyme in red blood cells started to fall, not from any cancer data.
• ●European Union: 25 mg/day total, derived the same way with a wider safety margin because the underlying studies were small and short.
• ●UK: no more than 25 mg/day from supplements unless a doctor advises otherwise.

Note that these are total intake limits and that they were set on copper status, not cancer. A reader comparing them with the doses in the table above will notice that the exposures carrying the lethal-disease signal are roughly two to seven times the US limit, which is a useful thing to know independently of whether the cancer association is causal.

The biology runs the other way, which is the strange part

Healthy prostate tissue is one of the most zinc-rich tissues in the body. It uses a transporter called ZIP1 to concentrate zinc, which then blocks an enzyme in the cell's energy cycle and lets the prostate make and secrete citrate.

In prostate cancer, that transporter is switched off and tissue zinc falls. Malignant cells lose the ability to accumulate it. Which produces the obvious-sounding inference, "so take more zinc," and the inference does not hold: once the transporter is down, more zinc in the blood does not become more zinc in the cell. Oral dose and tissue concentration are not the same exposure, and the gap between them is where most supplement reasoning about the prostate goes wrong.

Animal work points the same way as the human dose data, with both zinc-deficient and high-zinc diets producing larger tumors than an intermediate diet. A middle range, in other words, rather than more being better.

The copper problem, which is established on its own

High oral zinc reliably reduces copper absorption, by inducing a gut protein that binds copper and carries it out in shed cells. Sustained high doses can cause copper deficiency with anemia, low white cells and neurological injury. That is a documented toxicity at these intakes, separate from any cancer question.

But it is not established as the mechanism behind the prostate finding, and adding copper should not be assumed to cancel the concern. The large eye-disease trial that used 80 mg of zinc paired it with 2 mg of copper for exactly this reason, and found no increase in prostate cancer. It also was not built to answer this question.

Why this is a signal and not a settled fact

• ●It rests mainly on one cohort. Impressive follow-up, repeated measurement, careful adjustment, but one population of mostly White male health professionals.
• ●The high-dose groups are small. 18 advanced and 29 lethal cases in the top category, which is why intervals stay wide even where they clear 1.
• ●Another cohort found the opposite. The VITAL study found lower advanced-disease risk with supplemental zinc, though its highest category topped out near the dose where the other study's concern begins, so the two are not measuring the same exposure.
• ●Pooled analysis of total incidence is flat: 1.00 (0.79 to 1.27). Which is consistent with the pattern rather than contradicting it, since the signal was never in total incidence.
• ●Men take zinc for prostate symptoms. So some of the association could run backwards, with early disease driving the supplement rather than the reverse. The researchers tested for this with lag periods of up to ten years and the pattern held, which weakens that explanation without eliminating it.
• ●No randomized trial was ever built for this question. The trials that exist tested 80 mg in an eye-disease population for about six years, or 20 mg inside a five-nutrient cocktail. Neither tested decades of very high intake against fatal prostate cancer.

The defensible reading: chronic intake above the official limits has no demonstrated prostate benefit, carries a recognized copper risk, and sits in the dose range where the lethal-disease signal appears. That holds whether or not the association ultimately proves causal, which is the useful kind of conclusion.

Sweeteners

In July 2023 IARC classified aspartame as Group 2B, possibly carcinogenic to humans. That is the limited-evidence category, and it is the same category as lead and DDT, which tells you something about how broad a category can be.

There is a separate chemical route worth knowing: at body temperature methanol from aspartame converts to formaldehyde, which is Group 1, and then onward to formate. Saccharin, by contrast, was removed from the list in 1999, which is a reminder that these evaluations move.

Detail, including the approval history, is in Artificial Sweeteners.

Pesticide Residue

In 2015 IARC classified glyphosate as Group 2A, probably carcinogenic. It has since been detected in wine and beer in independent testing, which is to say that an input you think you are choosing arrives inside inputs you did not connect to it.

The regulatory history, the independent research, and what reduces exposure are in GMOs and Pesticides.

Water, the Mouth, and the House

Three inputs nobody experiences as inputs, because none of them feel like a decision.

• ●Municipal water. Chlorination reacts with organic matter to form trihalomethanes, a disinfection byproduct that the treatment process creates rather than removes. See Water.
• ●What goes in the mouth daily. Independent testing of oral care products has found arsenic, cadmium and lead, all three Group 1, with cadmium holding a kidney half-life measured in decades. See Toothpaste and Dental Toxins.
• ●Indoor air and furnishings. Formaldehyde, Group 1, off-gasses from engineered wood, textiles and foam. It is the single clearest overlap between what is in a house and what is on the IARC list. See The Toxic Home and Toxic Beds.

These matter for a reason the single-agent studies cannot capture. Every classification above was evaluated on its own, one agent at a time. Nobody lives one agent at a time, and no classification, label or clinical trial accounts for the total load across a life. That gap is structural and it is not going to be closed by better labels.

The One Place It Should Reach You, and Does Not

Everything above concerns information that never travels from a published finding to the person it describes. There is one setting where that journey should be shortest, because the person is already in the room, already diagnosed, and already being handed things to put in their body. It is the oncology office.

Patients there are commonly told that diet does not matter, to eat whatever they can keep down. Sugary snacks and juice are handed out in infusion suites. Many people come away understanding that food is simply not part of this.

The guidelines say something else entirely. The European Society for Clinical Nutrition and Metabolism states that all cancer patients should be screened regularly for malnutrition, early in care, and regardless of body mass index and weight history.

Here is the gap between that and practice, measured in a multi-site study of real referral behavior:

A nineteen-fold staffing shortfall is not indifference from the person in front of you. It is a service that was never built. That distinction matters, because understaffing is auditable and fixable, and it also tells a patient what actually works: not a general question about diet, which gets a general answer, but a specific request for a nutrition screen and a named referral, written in the chart.

The same study noted, in its own words, that weight loss may be masked by pre-existing obesity. Someone who began heavier can lose a dangerous amount of muscle and still land in a body mass index range that reads as fine, so the alarm never sounds. That is precisely why the guideline says to screen regardless of BMI, and precisely the person least likely to be screened.

And the argument against the cookie is not the one usually made

This page would be dishonest if it only corrected in one direction, so here is the correction that goes the other way.

"Sugar feeds cancer" does not hold up as stated. Tumor cells do consume glucose rapidly, which is how PET scans find them, but there is no clear evidence that dietary sugar preferentially feeds a tumor over the rest of the body, and an established tumor cannot be starved by removing carbohydrate, because the body maintains blood glucose regardless. A 2017 systematic review of ketogenic diets in cancer found no studies with methodologically rigorous designs, randomized trial evidence is absent, and no major cancer organization recommends a ketogenic diet as treatment. Sugar's established link to cancer runs mainly through body weight, which is a real pathway and a slower one.

But notice what that does to the defense of the cookie, rather than to the objection. If sugar is not a tumor-feeder, then it also cannot be justified on the grounds that calories are calories for a wasting patient. What that patient has lost is muscle, and muscle is rebuilt from protein. Sweet liquid and refined carbohydrate deliver fullness and almost no amino acids, which means they do not merely fail to help. They occupy the appetite that the next real meal needed.

That version survives a conversation with an oncologist. The other one ends it.

What This Page Is For

Not fear. Several things on this list are worth continuing, and at least one of them, chemotherapy, is the reason some readers are alive to read this.

What was missing was never the information. It was the route to it. A classification decided by a working group in Lyon and published for free never arrives at the person drinking the tea, taking the diuretic, or sleeping on the foam. Between the evidence and you there was no path, and the absence of a path is easy to mistake for the absence of a finding.

Some of what follows from this page costs nothing at all. Letting a drink cool. Putting an hour between the tea and the meal. Asking which of these a prescription belongs to, before accepting it rather than after.

The rest is yours to weigh. It always was. The only thing that changes is whether you were weighing it with the evidence in front of you.