← Tilbake til oversikten (norsk)

Red meat and colorectal cancer in the WCRF/AICR Continuous Update Project 2010 and Chan et al. 2011: a reproduction and re-analysis

Re-analysis series on the red meat – colorectal cancer syntheses, 2002–2016. Synthesis identifier: cup2010.

Abstract

Background. The 2010 systematic literature review (SLR) of the WCRF/AICR Continuous Update Project (CUP), published in peer-reviewed form as Chan et al. 2011 (PLoS ONE), reported a relative risk (RR) of colorectal cancer of 1.17 (95 % CI 1.05–1.31) per 100 g/day of red meat in eight prospective studies. It underpinned the CUP 2011 judgement that red meat is a convincing cause of colorectal cancer. We tested whether the estimate can be reproduced and how it depends on the extraction choices behind it.

Methods. We followed a protocol written before any re-analysis was run. We re-pooled the printed per-study values (SLR 2010 Figs 70, 73, 74 and 78; Chan 2011 Fig 4) by DerSimonian–Laird random effects. We compared each row with its primary paper and classified the deviations. We then re-ran the colorectal analysis with each class corrected, all classes corrected, and later publications of two cohorts substituted.

Results. All published summaries reproduce, two of them within 0.01 in one limit: colorectal 1.17 (1.05–1.31); colon 1.12 (0.97–1.29) in the SLR and 1.17 (1.02–1.33) in Chan (reproduced 1.17 (1.02–1.34)); rectum 1.18 (0.98–1.42) (reproduced 1.18 (0.98–1.43)); highest versus lowest 1.10 (1.00–1.21). Chan's colon analysis includes a Larsson 2005 colon estimate, 1.68 (1.11–2.55) with 10.33 % of the weight. This follows Chan's stated method, but it contradicts Chan's Table S2 and is absent from the SLR's colon analysis. The Multiethnic Cohort enters through a nested case-control report (Nöthlings 2009) whose red-meat variable probably includes processed red meat. EPIC's calibrated and English's mutually adjusted estimates give 1.15 (0.99–1.34); all corrections give 1.12 (0.95–1.32). Substituting the later EPIC and MEC publications gives 1.04 (0.96–1.13).

Conclusions. The CUP 2010/Chan 2011 estimates are computationally sound and reproducible. Their statistical significance depends on extraction choices that are defensible individually, several of them stated by the authors. When the alternatives the primary papers also print are used, the interval includes 1.

1. Introduction

The SLR. The CUP 2010 SLR on colorectal cancer [1] replaced the red-meat evidence base of the 2007 Second Expert Report with a dose-response meta-analysis on a common gram scale. It reported RRs per 100 g/day of red meat of:

• 1.17 (1.05–1.31) for colorectal cancer, eight studies;
• 1.12 (0.97–1.29) for colon cancer, nine studies;
• 1.18 (0.98–1.42) for rectal cancer, seven studies;

with I² 0 % in all three (SLR 2010 p155).

Chan et al. 2011. Chan et al. 2011 [2], written by the same Imperial College team, reported the same colorectal estimate in its abstract ("RR for 100 g/day increase = 1.17, 95% CI = 1.05–1.31"; p1). Its colon estimate differs: 1.17 (1.02–1.33) from ten studies (Table 1, p4).

What depends on it. These analyses replaced the 2007 per time/week and per 100 g/day summaries [3]. The CUP Colorectal Cancer 2011 Report [4] restated the Panel's judgement: "The evidence that red meat, processed meat, … are causes of colorectal cancer is convincing." The IARC Working Group evaluating red meat in Monograph 114 [5] discussed Chan 2011 alongside other meta-analyses. It noted that Chan's rescaling of servings to grams "may have increased the measurement error" (IARC 114, pp143–144).

This paper. The 2010 analysis is thus the first CUP gram-scale estimate and the template for the 2016–2017 update. We ask three questions:

1. Can its summaries be reproduced from its printed per-study values?
2. Do these values reflect the primary papers?
3. How do the summaries respond to correcting the deviations and to the cohorts' later publications?

2. Methods

2.1 Protocol

The analysis follows a protocol fixed before the re-analyses were run (version 1.0, 30 October 2026, REANALYSE_PROTOKOLL.md). It has seven rules, R1–R7, covering model, increment, study set, reproduction criterion, deviation classes, corrected runs and leave-one-out.

2.2 Data

Synthesis sources. We used:

• the SLR 2010 red-meat section, §2.5.1.3, pp155–177: text, Tables 34–37 and Figs 70–81;
• Chan 2011, including its text, Table 1 (p4), Fig 4 (p7, the red-meat forest plot with random-effects weights) and supplementary Tables S1 and S2.

The printed per-study values, weights and contrasts are in database/synthesis_extraction.csv (rows with synth_id = cup2010, separated by analysis_set).

The colorectal dose-response analysis. It comprises eight studies:

The colon analysis. It adds Bostick 1994 [14], Singh 1998 [15], Wei 2004 [16] (NHS and HPFS entered as separate rows) and Oba 2006 [17]. It drops Pietinen, Tiemersma, Larsson and Nöthlings.

The highest-versus-lowest analysis. It adds four studies that could not enter the dose-response analysis: Kato 1997 [18], Chan 2005 [19], Butler 2008 [20] and Cross 2010 [21] (SLR 2010 p155).

Primary-paper estimates. These come from database/estimates.csv, with per-100 g conversions in database/derived_per100g.csv.

2.3 The synthesis's own method, as stated

Scale and conversions.

• "all results were converted to a common scale (grams per day) and 120 grams was used as a standard serving or portion size in studies that reported frequency of intake";
• results are "presented for an increment of 100 grams per day";
• intakes in g/1000 kcal/day were converted "using the mean or median energy intake reported by the studies" (SLR 2010 p155).

Category rules (Chan 2011 p2).

• Means or medians of categories were used where reported, otherwise midpoints.
• Zero was used as the lower boundary of an open lowest category.
• An open highest category was given the width of the adjacent category.
• Slopes were derived from categorical data by generalised least squares [22, 23].
• Results by sex were first pooled by fixed effects.
• "One study provided results for distal and proximal colon cancer [ref. 34] and we derived the relative risk for colon cancer using the same procedure." (Chan's ref. 34 is Larsson 2005.)

Model. Random effects (DerSimonian–Laird [24]) unless otherwise stated (Chan 2011 p2), with heterogeneity measured by Cochran's Q and I² [25].

Publication and estimate choices (SLR 2010 p156).

• "the studies by Willett, W., et al., 1990 and Giovannucci, E., et al., 1994 were replaced by the most recent publication by Wei, E.K. et al., 2004";
• Chen 1998 "was included under red and processed meat";
• "The observed result from EPIC (Norat, T. et al., 2005) was selected instead of the calibrated results … for comparison with other studies that had not corrected for measurement error."

Sensitivity analysis (SLR 2010 p155). The SLR excluded the three studies that "did not provide details on the definition of red meat" (Järvinen 2001, Lee 2009, Nöthlings 2009). This gave colorectal 1.21 (1.06–1.37), colon 1.12 (0.97–1.31) and rectum 1.20 (0.99–1.47).

2.4 Reproduction rule

The rule follows R1–R3:

• printed per-study RR and 95 % CI were used as given;
• log standard errors were derived from the CI;
• pooling was by DerSimonian–Laird random effects per 100 g/day.

Reproduction requires the point estimate and limits to match to two decimals. Because inputs are printed to two decimals, the script accepts a difference of 0.01 in a confidence limit as rounding. Where it does so, we state it.

2.5 Deviations, corrections and follow-ups

Each colorectal row was compared with its primary paper and classified under R5. The classes are exposure, dose, cases, model, duplicate, transcription and inclusion.

We distinguish a deviation (the row departs from what the paper prints, or from the synthesis's own stated rule) from an assumption or stated choice. Model choices that the SLR announced (observed rather than calibrated EPIC) or that follow standard practice (the paper's main model) are reported as alternative specifications, not errors.

Corrected runs (R6) for colorectal cancer.

• model.
• EPIC: the observed 1.21 (1.02–1.43) was replaced by the calibrated 1.49 (0.91–2.43) (Norat 2005 Table 4, p912).
• English 2004: the model not adjusted for processed meat was replaced by the model with fresh red and processed meat entered together. That model gives 1.02 (0.98–1.08) per additional serve/week (English 2004, Results, p1511), which is 1.12 (0.89–1.57) per 100 g/day after the CUP scaling (×7 × 100/120 = ×5.833 on the log scale).
• exposure. Järvinen 2001 was removed (§3.3, D3).
• All classes. Both of the above together.
• Follow-ups.
• EPIC → Papadimitriou et al. 2022 [26]: red meat per 1 SD (36.2 g/day) 1.02 (0.98–1.05), Suppl. Table 2. This is 1.06 (0.95–1.14) per 100 g/day.
• MEC → Ollberding et al. 2012 [27], the full-cohort analysis (3,404 cases) using the paper's row for red meat excluding processed red meat (Table 2). This was converted by Greenland–Longnecker with the SLR's energy conversion: 0.97 (0.83–1.13) per 100 g/day.

Deviations with no replacement available are documented only. These are Nöthlings' exposure definition and design, and the Lee 2009 transcription question.

The colon, rectum and highest-versus-lowest analyses were reproduced but not corrected in this paper.

2.6 Software

All runs come from scripts/reanalyses.py, which calls scripts/pool.py. It provides:

• DerSimonian–Laird random effects and inverse-variance fixed effect;
• I²;
• leave-one-out;
• a cohort-uniqueness check.

Per-100 g conversions come from scripts/derive_per100.py, with Greenland–Longnecker slopes [22] computed by scripts/gl_doseresponse.py following Orsini et al. [23].

An independent controller (K11) recomputed every published summary with a separately written script, kontroll/meta_check.py.

2.7 Extraction and control

All extraction and checking were carried out by AI agents (Anthropic Claude, Opus models) under the author's direction, to a written codebook (KODEBOK.md).

• Primary-paper database. Every row of estimates.csv was checked against the full text by an independent agent instance that had not seen the extraction. Corrections are recorded in kontroll/rettelogg.md.
• Synthesis extraction. The CUP 2010/Chan 2011 extraction (76 rows) was checked by a separate controller (K11, kontroll/K11/K11_rapport.md) against the printed pages and the rendered Chan Fig 4. K11 confirmed the RR, CI and printed weight in all 76 rows. It flagged citation and labelling issues, which were corrected (correction log R57–R58).

3. Results

3.1 Reproduction

All five published summaries reproduce. Details are in Table 2.

• Colorectal. The eight rows of SLR Fig 70 give 1.17 (1.05–1.31), I² 0 %, k = 8 (run: CRC – published per-study values).
• This equals the SLR (p155) and Chan (abstract; Table 1, p4).
• Fixed and random effects coincide because τ² = 0.
• Printed and recomputed weights agree within 0.39 percentage points (K11).
• The printed weights in SLR Fig 70 and Chan Fig 4 are identical.
• Colon (SLR). The nine rows of Fig 74 give 1.12 (0.97–1.29) (run: Colon – SLR 2010 Fig 74 (9 rows)).
• Colon (Chan). The ten rows of Chan Fig 4 give 1.17 (1.02–1.34) against the printed 1.17 (1.02–1.33) (run: Colon – Chan 2011 Fig 4). The difference of 0.01 in the upper limit is within input rounding.
• Rectum. The seven rows give 1.18 (0.98–1.43) against the printed 1.18 (0.98–1.42) (run: Rectum – SLR 2010 Fig 78 (7 rows)). The same rounding applies.
• Highest versus lowest, colorectal. Twelve rows give 1.10 (1.00–1.21), I² 22 % (run: CRC – highest vs lowest (SLR 2010 Fig 73, 12 rows)).

3.2 Table 1. Colorectal dose-response inputs: as printed vs the primary papers

Sources.

• Printed values: SLR 2010 Fig 70 (p166), identical to Chan 2011 Fig 4 (p7).
• Contrasts and cases: SLR 2010 Fig 73 and Tables 34 and 37, and Chan S1.
• Re-derivations: kontroll/test2_report.md. These were run for the identical CUP 2016 rows; Lee 2009 differs between the two reports.

3.3 Deviations

Six deviations were recorded for the colorectal analysis (cup2010.md, deviation table). The colon and descriptive issues follow in §3.4–3.5.

D1 – Nöthlings 2009, exposure (documented only).

• What our deviation table says. MEC's "red meat" probably includes processed red meat; the table marks this as an inference, not a printed fact.
• What the paper itself says. Nöthlings et al. analyse "red meat" and "processed red meat" as separate exposures (Methods) but do not define red meat. Our database codes the inclusion of processed meat as unclear.
• Indirect evidence. The later full-cohort MEC paper, with the same questionnaire, distinguishes "total red meat" from "red meat excluding processed red meat" (Ollberding 2012, Table 2). This implies that MEC's undifferentiated "red meat" variable contains processed red meat.
• Conclusion. We regard the deviation as probable, not demonstrated.
• The SLR's own handling. The SLR itself flagged Nöthlings among the studies without a red-meat definition and excluded it in its sensitivity analysis (p155).

D2 – Nöthlings 2009, inclusion (documented only). The MEC enters through a nested case-control analysis of 1,009 cases and 1,522 controls (SLR 2010 Table 34, p156; Chan S1). Its blood samples were mostly drawn after diagnosis, but diet came from the baseline questionnaire. The full-cohort analysis with 3,404 cases (Ollberding 2012) appeared after both the SLR's search (end date December 2009, with one later study added up to May 2010; p155) and Chan's (to March 2011). This is therefore not a deviation at the time of publication. It is relevant to the follow-up run.

D3 – Järvinen 2001, exposure (inferred; removed in the corrected run).

• What CUP used. Järvinen's Table 3 reports a row labelled "Red meat", and CUP used it: Q4 vs Q1 1.50 (0.77–2.94).
• The concern. The quartile cut-points for this row (men <94, 94–141, 142–206, >206 g/day) are 91–94 % of those for "Meat, all" (<101, …, >220 g/day). The paper gives no definition.
• Classification. We infer that the variable covers nearly all meat, including processed products. This is an inference from quantities, not a printed definition.
• The SLR's own handling. The SLR anticipated the problem by listing Järvinen among the undefined exposures (p155).

D4 – English 2004, model (alternative specification).

• What CUP used. The paper's main model for fresh red meat (Table 3).
• The alternative. The paper also reports, in the text, that with fresh red and processed meat "modeled together" the fresh red meat estimate is 1.02 (0.98–1.08) per serve/week (p1511). That is 1.12 (0.89–1.57) per 100 g/day.
• Classification. Choosing the main model is standard practice and not an error. The alternative is relevant because the question addressed is unprocessed red meat.

D5 – Norat 2005, model (alternative specification, stated by the SLR). EPIC prints both the observed estimate, 1.21 (1.02–1.43), and the calibrated estimate, 1.49 (0.91–2.43) (Table 4, p912). The SLR chose the observed estimate and explained why (p156, quoted in §2.3). This is a stated, reasoned choice. The calibrated estimate is larger but much less precise, and it would cut EPIC's weight from 44.4 % to about 9 %.

D6 – Lee 2009, transcription (documented only). CUP 2010 prints 0.80 (0.52–1.23). The CUP 2016 SLR prints 0.81 (0.50–1.29) for the same study and table (SLR 2016 Fig 125 [28]). We could not re-derive either value: the manuscript version available to us lacks the confidence intervals for Q2–Q4.

Minor, no effect on pooled estimates.

• Wording. The SLR's summary paragraph describes the red-meat result as "for each 100g/day increase in red and processed meats" (p155). This is a slip in a red-meat section.
• Case totals. Printed case totals do not equal the sums of the per-study counts:

(synthesis_extraction_NOTES.md §4). - Unit label. The SLR describes the 2007 time-based summary as "per time/day" (p156), whereas the 2007 report printed "per time/week". This is discussed in the companion paper on WCRF 2007.

3.4 The Larsson 2005 colon row in Chan 2011

Chan's colon forest plot (Fig 4, p7) contains ten rows: the SLR's nine (Fig 74, p170) plus Larsson 2005, 1.68 (1.11–2.55), weight 10.33 %. This accounts for the ten studies and 3,561 cases in Chan Table 1, against nine studies and 3,172 cases in the SLR. The difference of 389 cases equals Larsson's colon cases.

The inconsistency. The inclusion is consistent with Chan's Methods, which state that the colon RR for Larsson 2005 (Chan's ref. 34) was derived from the proximal and distal results (p2). It is inconsistent with Chan's Table S2, which says of Larsson 2005:

"Outcome was proximal and distal colon cancer; meta-analysis was not performed as this is the only study provided sufficient information (results on colorectal and rectal cancer from article were included in the meta-analyses)."

So the paper describes the same study's colon result as both included and not included. The SLR, by contrast, did not include it.

Why it matters. The choice moves the colon estimate across the significance threshold:

• SLR, without Larsson: 1.12 (0.97–1.29);
• Chan, with Larsson: 1.17 (1.02–1.33).

Both are internally reproducible. Including the row is defensible under Chan's stated method. The S2 statement is the error, or else the Methods sentence is. A reader cannot tell which version the authors intended.

Related HvL findings (K11). Two related observations come from the independent control:

1. Chan's colon highest-versus-lowest summary, n = 10 with p(heterogeneity) = 0.70, reproduces only if Oba's two sex strata and Larsson's two subsites are each first combined within study. That is consistent with Chan's stated method.
2. Chan's red-meat proximal and distal highest-versus-lowest summaries (n = 2 each) could not be reproduced from the rows that claim them: Larsson 2005 and Chao 2005 [29]. S1 lists the Chao values under red and processed meat. This remains unresolved.

3.5 Table 2. Published, reproduced, corrected and follow-up estimates (per 100 g/day)

In the colon and rectum analyses, k counts rows; NHS and HPFS (Wei 2004) are entered as two rows, as in the SLR.

3.6 Corrected and follow-up runs (colorectal)

Model. With EPIC's calibrated estimate and English's processed-meat-adjusted estimate, the summary is 1.15 (0.99–1.34). The point estimate barely moves. Precision falls because EPIC's weight drops from 44.4 % to 8.9 %, and the interval now includes 1.

Exposure. Removing Järvinen gives 1.16 (1.03–1.30). On its own this hardly matters.

All classes. 1.12 (0.95–1.32), I² 9 %. Leave-one-out ranges from 1.09 (0.89–1.34) without Larsson 2005 to 1.18 (1.00–1.39) without Lee 2009. No single omission gives a lower limit clearly above 1.

Follow-ups. Replacing EPIC 2005 (1,329 cases) by Papadimitriou 2022 (5,069 cases) and MEC's nested case-control by Ollberding 2012's full-cohort unprocessed estimate gives 1.04 (0.96–1.13), I² 6 %.

• Dominance check. The two substituted rows carry 53.2 % (EPIC) and 25.1 % (MEC) of the weight, so the run is driven by these two cohorts, EPIC above all. Without the EPIC row the result is 1.04 (0.90–1.20); without the MEC row it is 1.07 (0.97–1.17).
• Caveats. Papadimitriou 2022 does not define red meat, and processed meat is a separate variable in the same table. Its estimate is per standard deviation (36.2 g/day) and was rescaled to 100 g/day. The other five cohorts are represented by their 2010-era publications, because no later publication was substituted for them in this run.

Reproduction-run leave-one-out. In the published data, the colorectal result already depends on EPIC: without Norat 2005 it is 1.14 (0.97–1.33). The other single omissions leave the lower limit above 1. The highest is 1.21 (1.07–1.35), without Lee 2009.

3.7 Figure

Figure 1. Forest plots for each run (colorectal published, corrected per class, all classes, follow-ups; colon SLR and Chan; rectum; highest versus lowest), with random-effects weights. See the interactive page kontroll/reanalyser/html/index.html, tab "CUP 2010" (#forest-cup2010; tables #tbl-cup2010, #loo-cup2010, #dev-cup2010).

4. Discussion

4.1 A careful analysis, with consequential choices

The CUP 2010 SLR is a carefully documented analysis. Its rules are stated:

• 120 g per serving;
• energy conversion of g/1000 kcal;
• category-dose rules;
• random effects;
• the most recent publication per cohort.

It corrected the most consequential problem of its predecessor by replacing Willett 1990 and Giovannucci 1994 with Wei 2004 (p156). It anticipated the definitional problem by running a sensitivity analysis without the three undefined exposures (p155). Every printed summary reproduces from its printed rows, and every printed weight reproduces within rounding.

The deviations we found are, with one exception, not errors in the ordinary sense. They are choices among the estimates the primary papers offer:

• EPIC observed rather than calibrated. A stated and argued choice.
• English's main model rather than the processed-meat-adjusted model. Standard practice.
• Järvinen's row labelled "red meat". Taken at face value, while the quantities suggest near-total meat.
• The MEC nested case-control report. The only MEC publication available at the time; its red-meat variable is probably not restricted to unprocessed meat.

The exception is the contradiction within Chan 2011 over the Larsson 2005 colon row (§3.4), which is an inconsistency in the published record.

4.2 What the corrected and follow-up results mean

Individually, the alternatives move the colorectal summary only slightly: 1.15, 1.16. Together they move it to 1.12 with an interval that includes 1. The reason is largely EPIC.

With the observed estimate, EPIC holds 44 % of the weight and its narrow interval excludes 1. With the calibrated estimate, its point estimate rises but its weight collapses. The significance of the published 1.17 (1.05–1.31) therefore rests substantially on one cohort's observed estimate, and leave-one-out in the published data shows the same thing: 1.14 (0.97–1.33) without EPIC.

The follow-up run asks a different question: what the same eight-cohort framework yields when two cohorts are represented by their later, larger and, for MEC, explicitly unprocessed analyses. The answer, 1.04 (0.96–1.13), is not an updated meta-analysis. Only two of seven cohorts were updated, and EPIC's 2022 exposure is undefined. It does show that the later evidence from the two largest cohorts in this set is closer to null than the estimates CUP 2010 used.

None of these runs addresses whether red meat causes colorectal cancer. They concern how the 2010 estimate depends on its inputs.

4.3 The colon discrepancy

The colon result is the clearest case of a published estimate depending on an undocumented, or contradictorily documented, inclusion. The SLR and Chan, from the same group and the same data, report 1.12 (0.97–1.29) and 1.17 (1.02–1.33). The difference is one row, which Chan's Methods include and Chan's supplement excludes. A reader of Chan's Table S2 would conclude that the colon summary does not contain Larsson 2005, but Fig 4 shows that it does.

4.4 Limitations

• Corrections applied to colorectal only. Corrections and follow-ups were applied only to the colorectal analysis. Colon and rectum were reproduced but not corrected here; the CUP 2016 paper in this series treats colon in detail.
• Inferred deviations. The exposure deviations for Järvinen and Nöthlings rest on inference (cut-point ratios; a later paper's variable names), not on printed definitions.
• Re-derivation gaps. Two rows could not be re-derived from the primary papers: Tiemersma 2002 (CUP's top-category dose is unpublished) and Lee 2009 (incomplete manuscript tables).
• Follow-up estimates. Papadimitriou 2022 does not define red meat. The Ollberding conversion uses the SLR's energy-based method and is therefore an assumption, not a printed value.
• Figure transcription. Chan Fig 4 was read from rendered page images. The values were checked by an independent controller.
• AI extraction. All extraction and control were done by AI agents. Residual error is possible despite independent verification, which is why all inputs are published.

5. Conclusions

The CUP 2010 and Chan 2011 red-meat estimates are reproducible from their printed inputs. The colorectal estimate of 1.17 (1.05–1.31) rests mainly on EPIC's observed estimate. Using alternatives that the primary papers also print, and that the protocol prefers for unprocessed red meat, gives 1.12 (0.95–1.32). With EPIC and MEC represented by their later publications, the estimate is 1.04 (0.96–1.13).

Chan's colon estimate of 1.17 (1.02–1.33) differs from the SLR's 1.12 (0.97–1.29) only through a Larsson 2005 colon row that Chan's own supplement says was not used.

6. Data and code availability

All inputs, scripts and control reports are published alongside this article:

References

1. World Cancer Research Fund / American Institute for Cancer Research. Continuous Update Project. Systematic Literature Review: The Associations between Food, Nutrition and Physical Activity and the Risk of Colorectal Cancer. London: Imperial College London; 2010. (Section 2.5.1.3, pp155–177.)
2. Chan DSM, Lau R, Aune D, Vieira R, Greenwood DC, Kampman E, Norat T. Red and processed meat and colorectal cancer incidence: meta-analysis of prospective studies. PLoS ONE 2011;6(6):e20456. (Supplementary Tables S1 and S2.)
3. World Cancer Research Fund / American Institute for Cancer Research. Food, Nutrition, Physical Activity, and the Prevention of Cancer: a Global Perspective. Washington DC: AICR; 2007.
4. World Cancer Research Fund / American Institute for Cancer Research. Continuous Update Project Report. Food, Nutrition, Physical Activity, and the Prevention of Colorectal Cancer. 2011.
5. IARC Working Group on the Evaluation of Carcinogenic Risks to Humans. Red Meat and Processed Meat. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans, Vol. 114. Lyon: International Agency for Research on Cancer; 2018.
6. Pietinen P, Malila N, Virtanen M, Hartman TJ, Tangrea JA, et al. Diet and risk of colorectal cancer in a cohort of Finnish men. Cancer Causes Control 1999;10:387–96.
7. Järvinen R, Knekt P, Hakulinen T, Rissanen H, Heliövaara M. Dietary fat, cholesterol and colorectal cancer in a prospective study. Br J Cancer 2001;85:357–61.
8. Tiemersma EW, Kampman E, Bueno de Mesquita HB, Bunschoten A, van Schothorst EM, Kok FJ, Kromhout D. Meat consumption, cigarette smoking, and genetic susceptibility in the etiology of colorectal cancer: results from a Dutch prospective study. Cancer Causes Control 2002;13:383–93.
9. English DR, MacInnis RJ, Hodge AM, Hopper JL, Haydon AM, Giles GG. Red meat, chicken, and fish consumption and risk of colorectal cancer. Cancer Epidemiol Biomarkers Prev 2004;13:1509–14.
10. Larsson SC, Rafter J, Holmberg L, Bergkvist L, Wolk A. Red meat consumption and risk of cancers of the proximal colon, distal colon and rectum: the Swedish Mammography Cohort. Int J Cancer 2005;113:829–34.
11. Norat T, Bingham S, Ferrari P, Slimani N, Jenab M, et al. Meat, fish, and colorectal cancer risk: the European Prospective Investigation into Cancer and Nutrition. J Natl Cancer Inst 2005;97:906–16.
12. Lee SA, Shu XO, Yang G, Li H, Gao YT, Zheng W. Animal origin foods and colorectal cancer risk: a report from the Shanghai Women's Health Study. Nutr Cancer 2009;61:194–205.
13. Nöthlings U, Yamamoto JF, Wilkens LR, Murphy SP, Park SY, et al. Meat and heterocyclic amine intake, smoking, NAT1 and NAT2 polymorphisms, and colorectal cancer risk in the Multiethnic Cohort Study. Cancer Epidemiol Biomarkers Prev 2009;18:2098–106.
14. Bostick RM, Potter JD, Kushi LH, Sellers TA, Steinmetz KA, et al. Sugar, meat, and fat intake, and non-dietary risk factors for colon cancer incidence in Iowa women (United States). Cancer Causes Control 1994;5:38–52.
15. Singh PN, Fraser GE. Dietary risk factors for colon cancer in a low-risk population. Am J Epidemiol 1998;148:761–74.
16. Wei EK, Giovannucci E, Wu K, Rosner B, Fuchs CS, Willett WC, Colditz GA. Comparison of risk factors for colon and rectal cancer. Int J Cancer 2004;108:433–42.
17. Oba S, Shimizu N, Nagata C, Shimizu H, Kametani M, et al. The relationship between the consumption of meat, fat, and coffee and the risk of colon cancer: a prospective study in Japan. Cancer Lett 2006;244:260–7.
18. Kato I, Akhmedkhanov A, Koenig K, Toniolo PG, Shore RE, Riboli E. Prospective study of diet and female colorectal cancer: the New York University Women's Health Study. Nutr Cancer 1997;28:276–81.
19. Chan AT, Tranah GJ, Giovannucci EL, Willett WC, Hunter DJ, Fuchs CS. Prospective study of N-acetyltransferase-2 genotypes, meat intake, smoking and risk of colorectal cancer. Int J Cancer 2005;115:648–52.
20. Butler LM, Wang R, Koh WP, Yu MC. Prospective study of dietary patterns and colorectal cancer among Singapore Chinese. Br J Cancer 2008;99:1511–6.
21. Cross AJ, Ferrucci LM, Risch A, Graubard BI, Ward MH, et al. A large prospective study of meat consumption and colorectal cancer risk: an investigation of potential mechanisms underlying this association. Cancer Res 2010;70:2406–14.
22. Greenland S, Longnecker MP. Methods for trend estimation from summarized dose-response data, with applications to meta-analysis. Am J Epidemiol 1992;135:1301–9.
23. Orsini N, Bellocco R, Greenland S. Generalized least squares for trend estimation of summarized dose-response data. Stata J 2006;6:40–57.
24. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986;7:177–88.
25. Higgins JPT, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat Med 2002;21:1539–58.
26. Papadimitriou N, Bouras E, van den Brandt PA, Muller DC, Papadopoulou A, et al. A prospective diet-wide association study for risk of colorectal cancer in EPIC. Clin Gastroenterol Hepatol 2022;20:864–73.
27. Ollberding NJ, Wilkens LR, Henderson BE, Kolonel LN, Le Marchand L. Meat consumption, heterocyclic amines and colorectal cancer risk: the Multiethnic Cohort Study. Int J Cancer 2012. doi:10.1002/ijc.27546.
28. World Cancer Research Fund / American Institute for Cancer Research. Continuous Update Project. Systematic Literature Review: The Associations between Food, Nutrition and Physical Activity and the Risk of Colorectal Cancer. London: Imperial College London; 2016 (published 2017).
29. Chao A, Thun MJ, Connell CJ, McCullough ML, Jacobs EJ, et al. Meat consumption and risk of colorectal cancer. JAMA 2005;293:172–82.

Generated from the open database and scripts described in the article; extraction and control performed by AI agents (Claude Opus/Fable) under the author's direction. 30 October 2026.