Two guideline families read the same cohort evidence and issue opposite instructions — WCRF 2018 says limit it, NutriRECS 2019 says adults may continue. The disagreement is not about the numbers below; it is about what they warrant, and that lives on Should Adults Reduce Red and Processed Meat. This page holds the shared evidence.
(World Cancer Research Fund International, 2018) (Johnston et al., 2019)
The association — modest, and red meat’s is not even significant
| Exposure -> colorectal cancer | Relative risk (per stated dose) | Source |
|---|---|---|
| Red meat | RR 1.12 (1.00 to 1.25) per 100 g/day — lower bound touches the null | WCRF |
| Processed meat | RR 1.16 (1.08 to 1.26) per 50 g/day | WCRF |
| Red + processed combined | RR 1.12 (1.04 to 1.21) per 100 g/day | WCRF |
(World Cancer Research Fund International, 2018)
The red-meat pooled estimate is not statistically significant — “no statistically significant association between the risk of colorectal cancer and consumption of red meat (RR 1.12 [95% CI 1.00-1.25], per 100 grams increase per day)” (WCRF’s own words); significance did appear in the colon-only stratum (RR 1.22, 1.06 to 1.39) and the European stratum (RR 1.23, 1.08 to 1.41), not in the pooled point estimate. (World Cancer Research Fund International, 2018)
Processed meat’s association is significant and consistent. WCRF also records the confounding candidly: “an apparent effect of red meat could possibly be due, at least in part, to low intakes of these other foods… Further analysis of adjustment factors was not performed in the CUP.” (World Cancer Research Fund International, 2018)
The absolute effect — the number WCRF never states and NutriRECS does
NutriRECS re-pooled the same body of cohort evidence into absolute terms — its own de novo dose-response meta-analyses on its own outcome set, not a restatement of WCRF’s RRs (corrected 2026-08-08) — for a realistic 3-servings/week reduction (average intake in North America / Western Europe is 2-4 servings/week):
- Unprocessed red meat -> overall cancer mortality: 7 fewer per 1000 over a lifetime (9 fewer to 6 fewer), Low certainty — with no significant difference on 8 other cancer outcomes.
- Red meat -> cardiometabolic events: 1 to 6 fewer per 1000 over ~11 y; all-cause and CVD mortality non-significant.
- Processed meat -> cancer / cardiometabolic outcomes: ranges of 1 to 8 fewer and 1 to 12 fewer per 1000.
This is Baseline Risk and the Relative-Absolute Split doing decisive work — but on related, not identical, quantities (corrected 2026-08-08): WCRF’s “16% increased risk” (processed meat) is a colorectal-cancer-incidence RR per 50 g/day; NutriRECS’s handful-per-1000 figures are its own absolute estimates on different endpoints and a different increment (e.g. overall cancer mortality per 3 servings/week). They are not one finding restated two ways — they are the same evidence body re-pooled to different outcomes, so the relative and absolute numbers here are not a matched pair. What generalizes is the framing effect: whether a guideline foregrounds a relative or an absolute magnitude of the same endpoint largely determines whether it reads as alarming or trivial. WCRF states relative forms and no absolute; NutriRECS computes absolutes and builds its recommendation on them.
The certainty split — the crux, and it is a grading disagreement
Both bodies looked at the same observational evidence and graded it oppositely:
- WCRF: processed meat is a “convincing” cause and red meat a “probable” cause of colorectal cancer — both “strong evidence” that «generally justify making public health recommendations.» (World Cancer Research Fund International, 2018)
- NutriRECS: the same evidence is low-to-very-low certainty, because under strict GRADE “in the absence of a large effect or a compelling dose-response gradient, observational studies provide only low- or very low-certainty evidence for causation,” and they “did not rate up for dose-response, given the potential for residual confounding.” (Johnston et al., 2019)
Same evidence, opposite grade -> Certainty of Evidence vs Strength of Recommendation. This is a methodological joined issue, not a difference in the data, and it is the engine of the recommendation split on Should Adults Reduce Red and Processed Meat.
The NutriRECS RCT + cohort arms, now held, confirm the “not the numbers” reading [2026-07-29]. The
RCT SR (Zeraatkar) found diets lower in red meat had «little or no effect» on cancer — total cancer
mortality HR 0.95 (0.89-1.01), colorectal-cancer incidence HR 1.04 (0.90-1.20) — but at low/very-low
certainty and «rated down for serious indirectness» (the evidence is one low-fat trial, not a red-meat (Zeraatkar et al., 2019)
trial). (Zeraatkar et al., 2019) The cohort SR (Vernooij, 70
cohorts / 6M people) found «very small or possibly small decreases» in cancer incidence/mortality from
lower-meat patterns, low/very-low certainty. (Vernooij et al., 2019) So across RCTs and cohorts the cancer signal is small/uncertain and (in the RCT) null-but-
indirect — reinforcing that the WCRF-vs-NutriRECS split is a grading-and-standpoint disagreement, not a
dispute about the effect size.
Mechanism
WCRF’s mechanistic account: haem iron “promote[s] colorectal tumorigenesis by stimulating the endogenous formation of carcinogenic N-nitroso compounds,” and high-temperature cooking forms heterocyclic amines and polycyclic aromatic hydrocarbons. WCRF rated mechanistic evidence “robust” for processed meat but only “moderate” for red meat — a split that tracks the convincing/probable grade gap. (World Cancer Research Fund International, 2018)
NutriRECS does not engage WCRF’s biological mechanism — its silence on the haem-iron / NOC account is not a dispute of it, but neither is it acceptance (corrected 2026-08-08, F1). And the two bodies do differ on causal status, not only on certainty and magnitude (corrected 2026-08-08, D2): the concession that «this does not preclude the possibility that meat has a very small causal effect» is the tail of a sentence whose head runs the other way — NutriRECS’s own intake-vs-pattern comparison suggests red and processed meat are «unlikely to be causal factors for adverse health outcomes», and it judges true causation of the observed effect sizes to be one «which we believe to be implausible». So WCRF asserts a convincing/probable cause while NutriRECS holds causation implausible — a genuine disagreement on existence, softened only by NutriRECS declining to rule a very small effect out entirely. (Johnston et al., 2019)
The intrinsic-heme channel — a within-category attribute, and what it bounds [2026-08-22, Bastide 2011]
WCRF names haem iron as a mechanism (above). Bastide 2011 (gold MA + mechanism review) makes heme the leading candidate mechanism and — more usefully — turns it into a within-category attribute that carries a specific decision consequence: heme is intrinsic to red meat regardless of curing.
The heme -> colon-cancer association. Five prospective cohorts (566,607 people, 4,734 colon-cancer cases): «The summary relative risk of colon cancer was 1.18 [95%C.I.: 1.06-1.32] for subjects in the highest category of heme iron intake compared with those in the lowest category» (Bastide et al., 2011). Read the magnitude with its two bounds: it is a highest-vs-lowest categorical contrast (mostly quintiles), not an RR per gram — the paper gives no human dose-response curve and no threshold — and it is colon cancer only (rectal data missing in two cohorts). And the heme exposure is partly a red-meat proxy: two cohorts computed heme as «a factor of 0.4 to the total iron content of all meat items which essentially is reporting an overall red meat effect» (Bastide et al., 2011), so the cohorts alone cannot separate heme from red meat. The heme-specific attribution rests on mechanism
- the white-meat contrast + a rat dose-response (aberrant-crypt effect size 1.73 [1.33-2.14]; heme concentration explains R-squared 0.62 of the variance) (Bastide et al., 2011).
Why heme, and why it names the red/white boundary. Heme content of red meat is «10-fold higher than that of white meat» — the proposed reason red meat carries CRC risk and white meat does not (Bastide et al., 2011). Heme acts as a catalyst on two pathways: endogenous N-nitroso (ATNC) formation — «Heme iron, and not inorganic iron or meat proteins, may be responsible for the nitrosation observed in the gut of volunteers fed red meat» — and lipid peroxidation to genotoxic aldehydes (MDA, 4-HNE) (Bastide et al., 2011). Competing mechanisms are down-weighted in-source: heterocyclic amines (chicken is HCA-rich yet carries no risk) and saturated fat («several studies, including a recent meta-analysis, showed no effect of saturated fat on colorectal carcinogenesis») (Bastide et al., 2011).
The decision-relevant bound — how much can nitrite/curing removal help? Because heme is present in red meat whether or not it is cured, the intrinsic-heme channel bounds the benefit of removing nitrite/curing: a genuinely nitrite-free processed-meat product still carries heme, so it cannot fall below the intrinsic-heme risk floor. But the two channels are coupled, not additive-independent. «In processed red meat, heme iron is nitrosylated, because curing salt contains nitrate or nitrite», and Pierre «demonstrated that the nitrosylation of heme was a key event in the promoting effect of processed meat in rats» (Bastide et al., 2011) — curing acts partly through heme (nitrosyl-heme > native heme in toxicity). So this is one side of an open attribution question: is processed-meat CRC harm intrinsic-heme-driven or curing-nitroso-driven? The nitroso side is addressed in the next section — but it splits into two exposures, only one of which Said Abasse 2022 measures. (inferred from Bastide et al., 2011)
The three-exposure decomposition — free dietary nitrate/nitrite is NOT the colorectal driver [2026-08-22, Said Abasse 2022]
The intrinsic-heme section leaves an attribution question: heme vs curing-nitroso. Said Abasse 2022 (gold SR+MA, 41 articles, 13 cancer sites) forces a finer cut than that binary, because the term “nitrite” in the diet is not one exposure. The correct decomposition of processed-meat cancer risk is three exposures, not two:
- Intrinsic heme iron — present regardless of curing (Bastide, above); promotes colon cancer (RR 1.18 categorical).
- In-matrix curing-nitroso / nitrosyl-heme — nitrite added to the meat matrix, nitrosylating heme and driving endogenous NOC (Bastide’s curing coupling); the channel most specific to processed meat.
- Free dietary nitrate/nitrite — total ingested nitrate/nitrite as measured by diet questionnaires, which is vegetable-dominated: «fruits and vegetables contribute over 80% of the daily dietary intake of nitrate … and nitrite … which represent the primary sources of exposure» (Said Abasse et al., 2022).
Said Abasse measures only exposure (3), and finds NO colorectal signal for it. Across both categorical approaches and both molecules, colon and rectal risk are null: Colon (nitrate) 0.99 (0.91, 1.08) / 1.00 (0.96, 1.04); Colon (nitrite) 1.02 (0.92, 1.11) / 1.02 (0.93, 1.11); Rectal (nitrate) 1.01 (0.88, 1.14) / 1.10 (0.96, 1.24); Rectal (nitrite) 1.09 (0.79, 1.39) / 1.06 (0.87, 1.26) (Said Abasse et al., 2022). The only significant categorical associations anywhere in the 13 sites were thyroid (nitrate) OR 1.40 (1.02, 1.77) and glioma (nitrite) OR 1.12 (1.03, 1.22) — both I2 = 0%, both non-GI (Said Abasse et al., 2022). A dose meta-regression adds only directional (no OR) hints — nitrite -> stomach/bladder up, pancreatic down; nitrate -> kidney/bladder down — several of which are publication-bias-flagged (Egger/Begg: kidney, stomach, pancreatic, colon-nitrite) (Said Abasse et al., 2022).
Why this is a DISTINCTION, not a contradiction of Bastide (not-joined check ii — different exposure/unit). A null on free dietary nitrite -> colorectal does not rebut Bastide’s in-matrix nitrosyl-heme -> colon mechanism, because they are different quantities: Bastide’s harm runs through heme-bound, matrix-embedded curing nitrite plus endogenous NOC; Abasse’s exposure is total ingested nitrate/nitrite dominated (>80%) by protective vegetable sources, so aggregate dietary nitrite is a poor proxy for the meat-matrix channel and even carries the opposite confounding. Said Abasse keeps the meat channel conceptually separate itself: «It is the presence of nitrite, amides, and amines … in processed meats and heme iron in fresh meat … that is considered to be responsible for these risk effects» (Said Abasse et al., 2022). And the one in-matrix interaction it reports runs the other way from the protective aggregate: among high-nitrate individuals, high heme carried a bladder signal (Catsburg 2014, OR 1.76 [1.21, 2.55]) (Said Abasse et al., 2022).
The decision consequence. “Cut the nitrite” is ambiguous across three exposures with different signs. Reducing free dietary nitrite by eating fewer vegetables would be net-harmful (it is the protective, vegetable-borne channel) and does nothing for CRC (null). The colorectal lever, if any, is the in-matrix curing-nitroso channel (exposure 2) — which this SR does not isolate — bounded below by the intrinsic-heme floor (exposure 1) that curing-removal cannot touch. So the fabric holds: dietary-nitrate guidance and processed-meat-CRC guidance are about different molecules, and a broad site-specific synthesis (41 articles, 13 sites) shows the vegetable-borne aggregate is not the colorectal driver.
A mitigation lever that does not require abstention. Bastide’s own decision move is to «inhibit red and processed meat toxicity instead of stopping meat intake» (Bastide et al., 2011): calcium and chlorophyll trap heme, vitamin C/E block ATNC formation, polyphenols block lipid peroxidation (hence «eat a yogurt after your steak»; ascorbate is already added during curing). This is directional/mechanistic (rat + human-biomarker), with no outcome RCT — held as a candidate lever, not a recommendation.
Supersession + independence. This 2011 MA (5 cohorts, categorical only) is the newest heme-CRC MA the wiki holds; a larger/newer heme-specific dose-response MA would upgrade it (G-gap). Bastide draws on the same red-meat cohort literature as WCRF/NutriRECS, so it is not independent backing for the association — its value is the mechanistic attribution and the intrinsic-vs-curing bound, not a second independent count (no [E-independent]).
The same channel decomposition on a MORTALITY endpoint [2026-09-02, Etemadi 2017]
The three-exposure logic above was built on cancer endpoints. An independent US mega-cohort — NIH-AARP, 536 969 adults aged 50-71, 16-year follow-up, 128 524 deaths — runs the same heme-iron / nitrate-nitrite decomposition on all-cause and cause-specific MORTALITY and reaches the same channel structure: «Heme iron and processed meat nitrate/ nitrite were independently associated with increased risk of all cause and cause specific mortality» (Etemadi et al., 2017). (The mortality dose-response itself lives on Food Groups and Health Outcomes - A Dose-Response Matrix; this page holds it only for the channel attribution it shares with the cancer decomposition above.)
| Exposure (highest vs lowest fifth) -> all-cause mortality | Hazard ratio |
|---|---|
| Red meat | 1.26 (1.23 to 1.29) |
| Heme iron | 1.15 (1.13 to 1.17) |
| Processed-meat nitrate | 1.15 (1.13 to 1.17) |
| Processed-meat nitrite | 1.16 (1.14 to 1.18) |
Etemadi’s nitrate/nitrite is exposure (2), NOT exposure (3). The measured nitrate/nitrite here is the in-matrix processed-meat additive, not the vegetable-dominated free-dietary aggregate Said Abasse measured: «The meat associated nitrate/nitrite intake is almost exclusively from the additives used in meat processing, as levels in unpro- cessed red meat are low» (Etemadi et al., 2017). So this cohort populates on a mortality endpoint the in-matrix curing-nitroso channel (exposure 2) that the cancer SR could not isolate, and finds it carries risk — a type-F extension of the channel decomposition from CRC incidence to all-cause and cause-specific death. It says nothing about the vegetable-borne exposure (3), so the CRC distinction above (nitrite-the-molecule has opposite signs across channels) is untouched.
Mediation is «accounted for, in part», not «explained by». SAS-macro mediation estimated that of the processed-red-meat -> all-cause-mortality association, nitrate statistically accounted for 50.1% (72.0% of the CVD-death association) and heme iron 20.9-24.1%; for unprocessed red meat -> all-cause mortality, heme iron accounted for 20.8% (13.7 to 30.3) (Etemadi et al., 2017). The two channels are mechanistically coupled, not additive-independent in origin: «the endogenous N-nitroso compound production is in fact stimulated by heme iron and not by protein residues» (Etemadi et al., 2017).
Read as one confounded observational cohort, not a causal-channel proof. Every exposure is FFQ-measured (124-item NCI DHQ) and the design is a single prospective cohort — the association is confounded and the mediation is a modelled decomposition, not an experiment. The authors flag the bias direction themselves: «measurement error may have biased our results toward null» (Etemadi et al., 2017) — i.e. the true channel gradients are, if anything, steeper than reported (Measurement Error in Dietary Assessment). This is not independent-backing (type-E) for the cancer association: NIH-AARP/Sinha overlaps the cohort literature the matrix and WCRF draw on, and a highest-vs-lowest-fifth mortality HR is a different estimand from a per-100 g/day CRC-incidence slope.
WCRF’s quantified recommendation — the gram target, now held [2026-08-05]
The WCRF Recommendations part (the Third Expert Report Summary) supplies the number the meat-fish-dairy exposure part did not: «If you eat red meat, limit consumption to no more than about three portions per week. Three portions is equivalent to about 350 to 500 grams … cooked weight of red meat. Consume very little, if any, processed meat.» The target «was chosen to provide a balance between the advantages of eating red meat (as a source of essential macro- and micronutrients) and the disadvantages (an increased risk of colorectal cancer and other NCDs)», and for processed meat «there is no level of intake that can confidently be associated with a lack of risk of colorectal cancer». (World Cancer Research Fund & American Institute for Cancer Research, 2018)
Same body, not a second witness. This is WCRF corroborating WCRF (a different work — the Recommendations part vs the meat-fish-dairy exposure part), so it adds the quantified target but no independent grade support for the convincing/probable judgements above. (inferred from World Cancer Research Fund & American Institute for Cancer Research, 2018)
Cooked-to-raw yield factor, now held. «500 grams of cooked red meat is roughly equivalent to 700–750 grams of raw meat, but the exact conversion depends on the cut … and the method and degree of cooking.» (World Cancer Research Fund & American Institute for Cancer Research, 2018) So cooked weight ≈ 0.67-0.71 × raw — a cooked-weight numeral corresponds to ~40-50% more raw meat.
Limits
- The gram target is now held — from the WCRF Recommendations part (
350 to 500 g cooked weight/week, ~3 portions), distinct from the meat-fish-dairy exposure part which gave only “no more than moderate amounts”. Attribute the gram number toWCRF - Diet Nutrition Activity Cancer 2018, not to the exposure-part source. - Cross-body commensurability — the yield factor now lets the direction be stated (partially cashed). NNR 2023 states 350 g cooked weight; WCRF gives 350-500 g cooked; ESC 2021’s 350-500 g leaves the basis unspecified. With WCRF’s yield factor (cooked ≈ 0.67-0.71 × raw) the direction is now statable: NNR’s 350 g cooked equals WCRF’s lower bound, and if ESC’s figure were raw it would be a materially lower true intake than the same cooked numeral. ESC’s weight basis remains the one open leg — a matching numeral is a matching recommendation only once ESC’s basis is confirmed. (inferred from World Cancer Research Fund & American Institute for Cancer Research, 2018)
- Cancer only for WCRF; NutriRECS spans cardiometabolic + cancer + mortality. The two do not cover identical outcome sets — matched only where both report colorectal cancer.
- The bigger cancer lever sits elsewhere (Layer-1). On WCRF’s own cancer-prevention scale the meat limit is a single-site colorectal recommendation, whereas body fatness is graded a convincing/probable cause across 12-of-17 sites and «one of the most important ways to protect against cancer» -> Body Fatness and Cancer Risk. Same grader, same scale: for a person carrying excess adiposity the meat decision is a small-lever refinement of the broader and more strongly graded adiposity lever — rank the big rock first. (Meat carries the public controversy while adiposity gets the quiet recommendation — attention is an anti-signal.) (inferred from World Cancer Research Fund & American Institute for Cancer Research, 2018)
- Not independent on the evidence. Both draw on largely the same observational cohort literature; their divergence is in appraisal and standpoint, not in separate data (no [E-independent]).
- Coherence, not validity (R1): the associations are what the cohorts report; whether reducing meat reduces a given person’s cancer is not established by either.
Appraising this observational evidence — the instrument [2026-07-31]
The per-serving cancer associations (Vernooij/Zeraatkar 2019) are observational; ROBINS-I (Risk of Bias Assessment Tools) is the appraisal instrument, with domain 1 (confounding) and domain 7 (selective reporting — the many-model problem) the likely caps. Flagged as a re-appraisal candidate there; not re-graded here. (inferred from Vernooij et al., 2019)
Non-cancer outcomes live in the matrix (2026-08-28)
This page stays cancer-scoped. The red/processed-meat associations with the OTHER outcome families — mortality, T2D, hypertension, CVD, adiposity — are held in the DIfE/Boeing dose-response grid, where processed meat is the most consistently harmful food in the whole matrix (harmful in all five outcome families; T2D 1.37 per 50 g/d, the single largest effect, HIGH grade) and red meat mirrors it one notch smaller. See Food Groups and Health Outcomes - A Dose-Response Matrix; not re-extracted here. (Schwingshackl et al., 2017) (Bechthold et al., 2017)