Jump to the Article ( DOI: 10.13140/RG.2.2.22237.17129) here
“The NRC is simply exaggerating the wrong issues to attract attention.”
Rhe cause Alcohol, Smoking, Overweigt and nitrate in the drinking water
J.Konstapel, Leiden,31-10-2026.
De beïnvloedbare oorzaken (samen 40% van de gevallen):
- Roken — 19,3%
- Overgewicht — 7,6%
- Alcohol — 5,4%
- UV-straling — 4,6%
- Dieet — 4,2%
- Infecties (HPV, hepatitis, etc.) — 3,4%
- Beweegtekort — 3,1%
Plus mindere maar meetbare omgevingsfactoren: nitraat in drinkwater (>10 mg/L), radon, en bepaalde beroepsexposities.
Kern: roken + overgewicht + alcohol alleen al = 32,3% van alle gevallen.
Executive Summary
Cancer research pursues two questions separately: what causes the disease, and what cures it. Both efforts are stalling. And they are stalling for the same reason. In both cases, the field reports an average over a population that contains effects running in opposite directions. The measurement that would separate those effects already exists in data already collected. Nobody has performed the separation.
This essay makes four claims. First, driver mutations cannot by themselves be the cause of cancer, because they are not scarce. Second, therapeutic removal — the operation oncology relies on — has a structural ceiling, and that ceiling is visible within a single patient over eighteen months. Third, the largest measured lever against cancer is not therapeutic at all: 40.0% of cases in US adults aged 30 and over are attributable to exposures that can be removed from the environment. Fourth, the causal and the therapeutic questions fail for the same averaging error, at the scale of a trial arm and at the scale of a county.
1. Mutations Are the Occasion, Not the Cause
The standard account holds that cancer arises when a cell accumulates enough driver mutations. This account makes a prediction: driver mutations should be rare in healthy tissue. They are not.
In normal eyelid skin, 18% to 32% of cells carry positively selected driver mutations. TP53 and FAT1 mutations appear in 3% to 5% of cells. Researchers describe aged skin as a patchwork of thousands of evolving clones. More than a quarter of cells carry cancer-causing mutations — in tissue that remains healthy.
The oesophagus is more striking still. In tissue from nine people aged 20 to 75, none with any history of oesophageal cancer, more than half of the normal epithelium had been taken over by mutant clones by middle age.
And within that one tissue, two classical drivers behave in opposite ways. TP53 fits the standard account: it is uncommon in healthy tissue and near-universal in the disease. NOTCH1 behaves backwards. It is several times more frequent in healthy oesophagus than in oesophageal cancer. A mutation more common in healthy tissue cannot, on its own, be a step toward the disease.
The conclusion is narrow but consequential. Driver mutations are not the scarce ingredient. By middle age, a normal oesophagus is mostly mutant clones that remain a normal oesophagus. Something holds them in the tissue. The right question is not how many drivers a clone carries, but what decides whether it keeps settling.
Caveat. These are small, cross-sectional studies. They do not follow clones into malignancy, and they do not refute the somatic mutation theory. They refute the sufficiency of counting mutations — the specific assumption the research programme has been built on.
2. Removal Has a Ceiling, and One Body Shows It
There are four ways a disease state ends. It resolves on its own. One party is removed. Something missing is supplied. Or both parties remain, the relation never settles, and what must change is the proportion.
Oncology is organised almost entirely around the second: removal. The question is how far removal reaches.
The FLAURA trial gives the cleanest available answer. It tested the best first-line removal agent in its class against the previous generation, in EGFR-mutated non-small-cell lung cancer. The results:
| Measure | Osimertinib | Comparator | Effect |
|---|---|---|---|
| Median progression-free survival | 18.9 months | 10.2 months | HR 0.46 — near-doubling |
| Median overall survival | 38.6 months (34.5–41.8) | 31.8 months (26.6–36.0) | HR 0.80, P = 0.046 — +6.8 months |
| Patients still on regimen at 3 years | 28% | — | 72% had left treatment |
The distance between those figures is the finding. Time before return nearly doubles. Survival gains a fifth. Yet nearly three quarters of patients have left the treatment within three years. What is removed is removed. What remains is what the removal did not reach — and what remains is what continues.
The mechanism is identical to antimicrobial resistance, where attributable deaths are forecast to rise from 1.14 million in 2021 to 1.91 million in 2050. Removing a self-replicating party selects for whatever was not removed. The difference is only scale: across a population over decades in infection, inside one body over eighteen months in oncology. The ceiling is structural, not technical. Cancer is therefore not a removal problem awaiting a better molecule.
3. The Averaging Error, at the Scale of a County
Iowa has had the second-highest cancer incidence in the United States for three consecutive years. It is the fastest-rising state while national rates fall. The state projects about 21,700 diagnoses and 6,400 deaths in 2026. The public argument blames agricultural chemistry.
The registry’s own March 2026 report tested that claim where it is most testable. Comparing Agricultural Health Study farmers with the general Iowa population over 1994–2015, adjusted for age and sex, farmers had 13% fewer cancers than expected. Their spouses had 10% fewer. The most exposed group had less cancer than the population around it.
That single number is now quoted by both sides of a political argument. Neither should quote it. It is an average of two opposite things:
| Direction | Sites |
|---|---|
| Lower than expected in farmers | Colon and rectum, lung, bladder, oral cavity and pharynx, pancreas, oesophagus, larynx, liver, tongue |
| Higher in farmers | Prostate (~20% above expected); lip (~65% above expected) |
| Higher in spouses | Melanoma, thyroid |
All nine low sites are associated with tobacco, alcohol, or both. The registry attributes the deficit to lower smoking and drinking, plus the healthy-worker effect. The two high sites are the two not driven down by abstinence: lip, where outdoor work and smokeless tobacco are candidates, and prostate — where the exposure signal in this population actually sits.
A large behavioural deficit has been averaged against a specific exposure excess. Reporting the result as one number supports neither conclusion. This is the same pooled error that RECOVERY made with dexamethasone and HARP-2 with simvastatin. Only here, the strata are tissues rather than patients.
Cautions. The cohort has aged, many participants stopped farming years ago, and Iowa lacks a system for tracking current compound use. Melanoma and thyroid are also the two cancers most sensitive to increased detection, which makes the spousal excess hard to read.
Nitrate: the tractable case
Drinking-water nitrate above 10 mg/L is associated with increased risk of aggressive prostate cancer in this same population, and ovarian cancer risk rises with average nitrate concentration. Prostate is also the site that stands out among farmers. Three things therefore exist at once: a measured exposure gradient (nitrate is assayed per public water supply), a proposed threshold (10 mg/L), and a complete population cancer registry covering the same geography. The stratifier, the dose, and the outcome are all already collected — by different institutions — and have never been linked. That is a records exercise, not a discovery.
Glyphosate: the exhausted case
| Analysis | Risk estimate for non-Hodgkin lymphoma |
|---|---|
| Zhang 2019, fixed effect | 1.41 (1.13–1.75) |
| Zhang 2019, random effects | 1.56 (1.12–2.16) |
| Agricultural Health Study, highest quartile, 20-year lag | 1.12 (0.83–1.51) |
| EPA meta-analysis, random effects | 1.14 (0.87–1.50) |
| High-quality studies only | 1.03 (0.61–1.75) |
| AGRICOH pooled, all NHL | 0.95 (0.77–1.18) |
| AGRICOH pooled, DLBCL subtype | 1.36 (1.00–1.85) |
The spread across analyses (0.95 to 1.56) is now larger than any effect they argue about. IARC classifies glyphosate as probably carcinogenic; the EPA concludes it is not likely to be carcinogenic. Both read the same papers. The county-level maps circulating in the Iowa campaign cannot break the deadlock: the most-cited overlay reports no effect size, no confidence interval, and no adjustment for age, smoking, or other compounds. Herbicide is applied where row crops are — which is also where the age structure, the radon, the nitrate, and the farming population are. A map cannot separate those.
The recommendation is unglamorous: leave glyphosate to the institutions already locked on it, and spend available effort on the exposure that has a number, a threshold, and a registry.
4. The One Removal Without a Ceiling
Environmental exposures cannot acquire resistance. Nitrate does not evolve around a catchment rule. Tobacco does not evolve around a tax. There is exactly one application of removal in cancer with no structural ceiling, and it operates before the disease rather than on it.
Its size is measured. In US adults aged 30 and over in 2019, excluding non-melanoma skin cancer, 713,340 of 1,781,649 cases — 40.0% — were attributable to modifiable risk factors, and 262,120 of 595,737 deaths — 44.0%.
| Factor | Share of cases | Attributable cases |
|---|---|---|
| Cigarette smoking | 19.3% | 344,070 |
| Excess body weight | 7.6% | 135,910 |
| Alcohol consumption | 5.4% | 96,730 |
| Ultraviolet radiation | 4.6% | 82,710 |
| Dietary factors | 4.2% | — |
| Infections | 3.4% | — |
| Physical inactivity | 3.1% | — |
Smoking, excess weight, and alcohol together account for 32.3% of all cases — 81% of the attributable total. Smoking alone accounts for 65% of attributable deaths.
Three consequences follow:
- The lever is larger than anything in the therapeutic pipeline. No agent in development will move 40% of incidence. The comparison is not close.
- Prevention does not require the causal question to be settled first. Removing tobacco does not wait on knowing which mutation it produced in which cell.
- The Iowa argument is inverted. The registry names binge drinking among the worst in the nation and a cigarette tax unchanged for nearly twenty years, in a state rated high-risk for radon. These are the factors with the largest measured attributable fractions — and they are almost entirely absent from the public argument, which is about herbicide. The exposure with the strongest evidence is the one nobody campaigns on, because it implicates the electorate rather than an industry.
5. Where the Work Should Go
Six directions follow. None requires a discovery. Three require no new data collection at all.
| # | Direction | What is built | Judged by |
|---|---|---|---|
| 1 | Report occupational and geographic cancer comparisons by site, never pooled | Site-level tables as the default registry output | Sign reversals visible, not averaged away |
| 2 | Link the nitrate gradient to the registry | Cohort analysis using measured water-supply nitrate, 10 mg/L as stratifier | Dose-response by site, prostate first |
| 3 | Stop funding ecological map overlays as evidence | Nothing; the saving is the point | Claims withdrawn from the public argument |
| 4 | Study normal tissue, not only tumours | Longitudinal surveys of mutant clones in healthy tissue across exposures | Which clones stop settling, and on what |
| 5 | Name a boundary marker for cancer | A measurable quantity dividing clones that continue from clones that do not | Prospective separation before intervention |
| 6 | Spend the prevention budget in proportion to attributable fractions | Tobacco, weight, and alcohol policy at the measured 32.3% | Incidence, not awareness |
Direction 5 matters most, and nothing stands behind it. The other “fourth-row” diseases became tractable the moment somebody named a marker: suPAR at 6 ng/mL for sepsis, a baseline inflammatory class for ARDS, hsCRP at 2 mg/L for atherosclerosis. In each case the marker did not explain the disease. It divided the population into halves in which the same intervention ran in opposite directions. That was enough to recover effects that pooled trials had discarded. Cancer has no such quantity. The nearest candidate is not a property of the tumour but of the tissue around it.
Direction 2 could start this quarter. Iowa holds a complete population cancer registry going back decades, and nitrate is assayed per public water supply across the same geography. Linking them is a records exercise. If the association with aggressive prostate cancer above 10 mg/L holds in the full registry, that is an exposure with a threshold — the first thing in this entire argument a regulator could act on.
Limitations
- Attributable-fraction estimates are counterfactual modelling, not experiments. They depend on assumed exposure prevalences and relative risks, and they are US figures that do not transfer unchanged to Iowa or Europe.
- The normal-tissue mutation studies are small (nine individuals for the oesophagus) and cross-sectional. The possibility that NOTCH1 protects is the investigators’ stated suggestion, not a result.
- The Iowa farmer comparison is against the general state population, not matched non-farmers. The cohort has aged out of current exposures, and the healthy-worker effect is acknowledged rather than adjusted away.
- FLAURA compares against first-generation inhibitors, not against no removal, and crossover was permitted. It says nothing about whether removal was the right operation.
- The four-operation scheme is exhaustive by construction and cannot be refuted as a scheme. What is refutable is every row assignment in it, including this one.
Annotated References
Driver mutations in normal tissue
- Martincorena, I., et al. “High burden and pervasive positive selection of somatic mutations in normal human skin.” Science (2015).
The foundational study for the claim that driver mutations are not scarce. Sequenced normal eyelid skin and found positively selected driver mutations in 18–32% of cells, with NOTCH1, TP53, and FAT1 frequencies and clone densities reported per cell. Why it matters: it breaks the prediction of the accumulation account at its most testable point. - Wellcome Sanger Institute / Martincorena, I., et al. “Somatic mutant clones colonize the human esophagus with age.” Science (2018).
Extends the finding to the oesophagus: more than half of normal epithelium is mutant clones by middle age. Supplies the TP53-versus-NOTCH1 contrast that shows two classical drivers behaving in opposite directions. Why it matters: it is the strongest single piece of evidence that mutation burden alone does not cause cancer.
Attributable fractions
- Islami, F., et al. “Proportion and number of cancer cases and deaths attributable to potentially modifiable risk factors in the United States, 2019.” CA: A Cancer Journal for Clinicians (2024).
The source of every figure in the 40.0% argument: 713,340 of 1,781,649 cases and 262,120 of 595,737 deaths attributable to modifiable factors, with the site-by-site breakdown by factor. Why it matters: it quantifies the only removal with no structural ceiling.
Iowa incidence and the farmer comparison
- 2026 Cancer in Iowa Report, Iowa Cancer Registry, University of Iowa.
The registry’s own farmer-versus-population comparison (13% deficit overall, site-level detail) and its attribution of the deficit to smoking and drinking. Why it matters: it is the primary evidence in the agricultural-chemistry debate, and it repays stratification far better than pooling. - Iowa Capital Dispatch, coverage of the 2026 report.
Quantifies the prostate (~20%) and lip (~65%) excesses and records the cautions on cohort age and missing pesticide tracking. Why it matters: it supplies the caveats any honest use of the farmer data must carry. - The Cancer Letter, on Iowa’s cancer rate as a campaign issue.
Context on Iowa’s second-place national ranking, its rising trend against falling national rates, and the nitrate findings. Why it matters: it frames the political stakes of the measurement question. - PBS NewsHour, “As cancer rates fall nationally, Iowa sees a rise.”
The registry director on alcohol, tobacco taxation, and radon. Why it matters: it documents the three largest-lever exposures that the public argument ignores.
Glyphosate and non-Hodgkin lymphoma
- EPA Office of Chemical Safety and Pollution Prevention, epidemiological review of Zhang et al. and Leon et al. (January 2020).
The per-analysis risk estimates, the Agricultural Health Study results (575 cases, no significant association across quartiles, P = 0.62), and the EPA’s “not likely carcinogenic” conclusion. Why it matters: it shows that the spread across analyses now exceeds the effect under dispute. - Food & Water Watch, “Glyphosate Use and Cancer Clusters” (March 2026).
The county overlay driving the Iowa campaign: 82% of heaviest-spraying counties above the national lymphoma average. Why it matters (negatively): no effect size, no confidence intervals, no adjustment, 31–42% of counties dropped for missing data — the pooled error in cartographic form.
Oncology outcomes
- FLAURA trial, first-line osimertinib in EGFR-mutated NSCLC.
Progression-free survival (18.9 vs 10.2 months, HR 0.46), overall survival (38.6 vs 31.8 months, HR 0.80, P = 0.046), and three-year regimen retention (28%). Why it matters: it makes the ceiling of removal visible inside a single body over eighteen months. - GRAM Project, University of Oxford.
Forecasts of antimicrobial-resistance mortality rising from 1.14 million (2021) to 1.91 million (2050). Why it matters: it establishes the selection-for-what-was-not-removed mechanism that oncology shares with infection.
Framework and companion essays
- J. Konstapel, Research Directions to Cure All Diseases, Constable Research, Leiden (29 October 2026).
The four operations by which a disease state ends, and the ten research directions. Why it matters: it supplies the structural framework the essay tests cancer against. - J. Konstapel, The Order of Questions, Constable Research, Leiden (30 October 2026).
The ordering of marker, endpoint, and agent, and the six gates. Why it matters: it explains why naming a boundary marker precedes, and does not follow, intervention design. - J. Konstapel, Depth of Discharge, Constable Research, Leiden (29 October 2026).
The stratification argument in full. Why it matters: it is the methodological core of the averaging critique applied to trials and registries alike. - J. Konstapel, The Vacuum.Net Theory, Ninth Edition, Constable Research, Leiden (27 October 2026).
The underlying theoretical framework of the research programme. Why it matters: it is the context in which the four-operation scheme is exhaustive by construction.
