Nucleus of the cancer-prevention cluster. Opens the cancer domain for the wiki (previously near-
empty — only the red/processed-meat pages). Source: the WCRF/AICR Third Expert Report (2018), the
gold-standard continuously-updated synthesis (Continuous Update Project — 18 systematic reviews graded
by an independent panel). Single-source scaffolding — provisional (confidence: low) until a second
cancer-evidence body intersects it.
The decision this page changes. By breadth of sites and strength of evidence (not a per-person effect magnitude, which WCRF does not quantify), the leading diet-adjacent cancer exposures — after tobacco — are body fatness, alcohol, and physical activity, not the specific “superfoods” and “carcinogens” that dominate discourse (attention is an anti-signal). WCRF estimates «Between 30 and 50 per cent of all cancer cases are estimated to be preventable through healthy lifestyles» — but that figure is broader than diet: the full source sentence adds «avoiding exposure to occupational carcinogens, environmental pollution and certain long-term infections», and “healthy lifestyles” includes not smoking. (World Cancer Research Fund & American Institute for Cancer Research, 2018) It is also a population-attributable fraction, not a per-person risk reduction. WCRF warns «If current trends continue, overweight and obesity are likely to overtake smoking as the number one risk factor for cancer.» (World Cancer Research Fund & American Institute for Cancer Research, 2018)
The grading system — the decision-relevant object, not the raw effect
WCRF grades each exposure -> cancer-site relationship for strength of causal evidence, and only strong grades drive recommendations. The grade, not a bare association, is what to read:
| Grade | Band | Drives a recommendation? |
|---|---|---|
| Convincing (increases / decreases risk) | strong | yes |
| Probable (increases / decreases risk) | strong | yes |
| Substantial effect on risk unlikely | strong (a null) | no — a strong null; reassurance use is the wiki’s gloss, not a stated WCRF rec basis |
| Limited — suggestive | limited | generally no |
| Limited — no conclusion | limited | no |
«They enable evidence to be categorised as being ‘strong’ (‘convincing’, ‘probable’ or ‘substantial effect on risk unlikely’) or ‘limited’ … Only evidence judged to be strong is usually used as the basis for Recommendations.» (World Cancer Research Fund & American Institute for Cancer Research, 2018) The criteria are Bradford-Hill-derived (study quality, number of study types/cohorts, heterogeneity, dose-response, plausible mechanism at typical human exposures). This is WCRF’s own system, distinct from GRADE — the same evidence graded “convincing/probable” here is graded “low certainty” under strict GRADE, which is the engine of the meat controversy -> Certainty of Evidence vs Strength of Recommendation, Should Adults Reduce Red and Processed Meat.
The graded matrix — decision-relevant big rocks (STRONG evidence only)
From the §5 recommendation justifications (prose) cross-checked against the color-coded matrix asset
(raw/assets/WCRF-All-Cancers-Matrix-2018.pdf). Grades stated where prose pins them; the full
57-exposure × 20-site grid lives in the asset.
Increases risk (strong):
| Exposure | Cancer site(s) | Grade | Note |
|---|---|---|---|
| Adult body fatness | oesophagus (adeno), pancreas, liver, colorectum, kidney, endometrium, postmeno breast, + more | Convincing / Probable | «12 of the 17 cancers … linked»; IARC adds thyroid, myeloma, meningioma -> Body Fatness and Cancer Risk |
| Alcohol | mouth/pharynx/larynx, oesophagus (SCC), breast, colorectum, liver, stomach | Convincing / Probable | «no … threshold» for cancer overall (breast: none identified); the CRC and liver/stomach judgements are based on intakes above ~30 / ~45 g/d — the edge of the evidence, NOT a safe threshold -> Alcohol and Mortality and Vascular Disease |
| Processed meat | colorectum | Convincing | «no level of intake … associated with a lack of risk» |
| Red meat | colorectum | Probable | limit ~350-500 g cooked/week -> Red and Processed Meat and Cancer |
| High-dose beta-carotene supplements | lung (current/former smokers) | Convincing (harm) | the supplement-harm case -> Vitamin and Mineral Supplements for Disease Prevention |
| Glycaemic load | endometrium | Probable | «is a cause of endometrial cancer» |
| Aflatoxins | liver | Convincing | (regional) |
| Adult attained height (marker) | several | Convincing / Probable | a marker for growth factors, «unlikely to directly … cause» |
Decreases risk (protective, strong):
| Exposure | Cancer site(s) | Grade | Note |
|---|---|---|---|
| Physical activity | colorectum (colon), endometrium, breast | Convincing (colon) / Probable | «the greater the amount … the greater the benefit» -> Physical Activity Dose and Mortality |
| Wholegrains | colorectum | Probable | -> Whole Grains Refined Grains and Pulses |
| Dietary fibre | colorectum | Probable | goal >=30 g/day -> Dietary Fibre and Health |
| Non-starchy veg/fruit (aggregated) | aerodigestive | Probable (aggregate) | individually limited; aggregate «more persuasive» |
| Coffee | liver, endometrium | Probable | |
| Dairy / calcium (incl. supplements) | colorectum | Probable | supplements >200 mg/d; adverse effects elsewhere |
| Body fatness in young adulthood | breast (pre + postmeno) | Probable | the paradox — a protective arm, direction opposite to adult body fatness |
| Lactation | breast (mother) | Probable / Convincing |
(World Cancer Research Fund & American Institute for Cancer Research, 2018)
Two structural cautions the report itself raises
1. The holistic shift — the pattern, not the food, is the causal unit. The Panel now judges «it appears increasingly unlikely that specific foods, nutrients or other components of foods are themselves important singular factors … rather, different patterns of diet and physical activity combine to create a metabolic state» conducive to cancer. (World Cancer Research Fund & American Institute for Cancer Research, 2018) This is the wiki’s own observed-healthy-population-is-not-evidence-for-a- component rule arriving from a cancer body: confidence in the causal exposure is «always … greater for the aggregated set of dietary factors … than for any single food». The adherence studies bear it out — «the more people adhere to the … Recommendations, the greater the reductions» — so «confidence in the protective effect from following all of the Recommendations is greater than that for any individual Recommendation.» (World Cancer Research Fund & American Institute for Cancer Research, 2018)
2. Population judgements, and the loop is open. These are population-level causal grades, not per-person effects, and they rest on observational cohorts (few lifetime-diet RCTs are feasible — a fit-to-question limit, not a defect). No source here shows that an individual’s weight loss or meat cut reduces their cancer — the survivor evidence is explicitly «limited» on whether changing exposures «will alter the clinical course». (World Cancer Research Fund & American Institute for Cancer Research, 2018) (inferred from World Cancer Research Fund & American Institute for Cancer Research, 2018)
The 2025 DLP evidence — caution #1 now evidenced, not just asserted (type-F)
The 2018 report asserted the holistic shift (caution #1 above) but graded whole dietary patterns as
too weak to act on: for breast and colorectal cancer they were «graded as ‘limited-no conclusion’», the
exception being mouth/pharynx/larynx («‘limited-suggestive’»).
(World Cancer Research Fund International & American Institute for Cancer Research, 2025)
The 2025 CUP Global pattern report (breast: 84 pubs; CRC: 86 pubs) upgrades that same claim — the
pattern-as-causal-unit shift is now a strong-probable grade, not a Panel intuition. This is a type-F
refinement of caution #1 by the same body (WCRF/AICR), so it raises the grade WCRF itself assigns
but earns no [E-independent] corroboration — the confidence stays low on this page until a
non-WCRF cancer-evidence family intersects it.
The grade movement.
- CRC: «The WCRF/AICR Score was graded as having a ‘strong-probable’ likelihood of causality for a decreased risk of colorectal cancer.» (World Cancer Research Fund International & American Institute for Cancer Research, 2025)
- Breast: «WCRF/AICR score and the American Cancer Society score were graded as having a ‘strong-probable’ likelihood of being causally associated with a decreased risk of breast cancer» — strong-probable for overall and post-menopausal, «‘limited-suggestive’» for pre-menopausal. (World Cancer Research Fund International & American Institute for Cancer Research, 2025)
The lifestyle leg carries the evidence — a DLP beats a diet-only pattern. «patterns incorporating modifiable risk factors (namely markers of adiposity, smoking and physical activity) showed stronger associations with breast cancer incidence than patterns scoring dietary factors alone» (same for CRC). (World Cancer Research Fund International & American Institute for Cancer Research, 2025) This sharpens caution #1: the aggregate that matters is not diet alone but diet plus body fatness, activity, alcohol and smoking — the same big rocks already ranked above, now shown to dominate the pattern signal too, consistent with «the more components of our DLP adhered to, the better for cancer risk reduction.» (World Cancer Research Fund International & American Institute for Cancer Research, 2025)
Hybrid biomarker patterns run the other way (harm). Patterns built to predict insulin / inflammation biomarkers were «graded as ‘strong-probable’ for their likelihood of causality for an increased colorectal cancer risk. These patterns are: Empirical Dietary Index for Hyperinsulinaemia … Empirical Dietary Inflammatory Pattern … Empirical Lifestyle Index for Hyperinsulinaemia.» (World Cancer Research Fund International & American Institute for Cancer Research, 2025) The report flags these are not clean causal maps of their named mechanism — «alcohol consumption is negatively scored in these patterns but has strong evidence of an increased risk for colorectal cancer» (World Cancer Research Fund International & American Institute for Cancer Research, 2025) -> Inflammation as a Modifiable Lever.
Site-specific decision splits. For breast, «the evidence suggests that any amount of alcohol increases risk» -> avoid alcohol. (World Cancer Research Fund International & American Institute for Cancer Research, 2025) For CRC the recommended DLP adds «the consumption of coffee and calcium containing foods, such as dairy products» and «Avoiding processed meat.» (World Cancer Research Fund International & American Institute for Cancer Research, 2025)
What the upgrade does NOT buy — no magnitudes, mostly observational, high RoB. «meta-analyses were not conducted for either breast or colorectal cancer», so the report gives grades, not effect sizes (no HR / RR / absolute risk to rank or dose). (World Cancer Research Fund International & American Institute for Cancer Research, 2025) The base is near-entirely observational — «the scarcity of randomised controlled trials for dietary patterns and DLPs, with one intervention trial found for both breast and colorectal cancer outcomes» (World Cancer Research Fund International & American Institute for Cancer Research, 2025) — and carries critical risk of bias: «in the 86 papers included in the colorectal cancer review, 31 studies were considered to have critical risk of bias for confounding.» (World Cancer Research Fund International & American Institute for Cancer Research, 2025) The report is explicit it «is not intended to supersede the 2018 WCRF/AICR Cancer Prevention Recommendations, but to support them.» (World Cancer Research Fund International & American Institute for Cancer Research, 2025)
Yin 2025 umbrella review — the first ranked magnitudes, and the independence test failed (type-F)
The 2018 and 2025 WCRF reports grade the pattern->cancer relationship but give no effect sizes (the
2025 DLP report «meta-analyses were not conducted»). Yin’s umbrella review of 74 meta-analyses of
prospective cohorts supplies the first magnitudes with CIs — but on the strict type-E test it is
not independent backing; it is a type-F refinement plus a distinct-pattern extension, so the nucleus
confidence stays low.
The magnitudes (the type-F payoff). Only 3 of 74 MAs (4.1%) reached convincing evidence, and they are: «adherence to the 2007 World Cancer Research Fund/American Institute for Cancer Research (WCRF/AICR) dietary recommen- dations (per 1-unit score increase) and lower risk of all cancers (relative risk [RR] ¼ 0.93, 95% confidence interval [CI]: 0.92, 0.95), whereas negative associations were found comparing the highest compared with lowest categories for a prudent diet (RR ¼ 0.89, 95% CI: 0.85, 0.93) and vegetable-fruit- soybean diet (RR ¼ 0.87, 95% CI: 0.83, 0.92) in relation to breast cancer.» (Yin et al., 2025) On GRADE, only 4 of 74 (5.4%) were high — «adherence to the 2007 WCRF/AICR dietary recommendations and decreased risks of all cancers, breast cancer, colorectal cancer, and prostate cancer.» (Yin et al., 2025) The effect envelope across all 74 MAs is modest — «the magnitude of the observed summary random-effects estimates ranged from 0.59 to 2.23» (Yin et al., 2025) — and the base is mostly weak (only 13/74 prediction intervals excluded null).
Why it is NOT [E-independent]. The convincing/GRADE-high result scores adherence to WCRF’s own
2007 index — WCRF is the antecedent being tested, not a separate route to the same claim (F, not E).
And Yin pools the same prospective-cohort literature the WCRF CUP draws on: «Because the WCRF/AICR
recommendations are not dietary pat- terns in the strictest sense, studies recommending an overall
lifestyle pattern were excluded … whereas studies only referencing dietary recommendations were
included.» (Yin et al., 2025) The
author list shares no WCRF panellist but sits in the same Harvard/Channing/Tulane nutritional-epidemiology
lineage that generates much of the pooled cohort evidence — shared cohorts and shared school, so
shared-source agreement, not independent backing.
(inferred from Yin et al., 2025)
One genuinely new direction — DII harm (very low GRADE). Yin re-grades a harmful pattern: «The present UR also found a harmful association between the DII score and the risk of CRC comparing the highest score to the lowest, as well as a 1-unit increment in the score» (Yin et al., 2025) — but «classified as very low quality in terms of GRADE, mainly owing to the presence of heterogeneity and publication bias.» (Yin et al., 2025) This runs the same direction as the 2025 DLP hybrid- biomarker (EDIH/EDIP/ELIH) harm signal above, from a second body but on very-low-grade evidence -> Inflammation as a Modifiable Lever.
Layer-1 ranking note
(inferred from World Cancer Research Fund & American Institute for Cancer Research, 2018) The ranking is stratum-dependent. For a lean, non-drinking, active non-smoker the diet-adjacent cancer levers are mostly already pulled, and the remaining ones (specific foods) are small-and-limited by construction — reporting that ceiling is itself the result. Where a big rock is present (obesity, heavy alcohol, inactivity) it dominates the cancer ranking as it does the mortality one, and the same exposure typically moves cancer, cardiometabolic and all-cause outcomes together (the recommendations «promote an overall way of life» beneficial across NCDs) -> Layer 1 - Ranking Interventions for a Stratum.
Limits
- Still single-body for
[E-independent], even after Yin. WCRF/AICR is one (authoritative) body; its related reports (this master, Body-Fatness, Physical-Activity, the 2025 DLP) are the SAME body. The Yin 2025 umbrella review was tested as the candidate independent family and failed the strict test — its headline finding scores WCRF’s own 2007 index and pools the same cohort literature (shared-source, same epidemiology lineage), so it refines (type-F) but does not corroborate independently. A genuinely disjoint cancer-evidence family (IARC monographs beyond the labels held; an umbrella built on a non-overlapping cohort base) is still owed before any grade here earns[E-independent]. - Grade granularity. Exact convincing-vs-probable per cell is read from the color-coded matrix figure; where the §5 prose does not pin a grade, the band (strong) is stated rather than a guessed cell.
- Coherence, not validity (R1): the grades are causal judgements on cohort associations; the wiki cannot verify them against realized outcomes.