← Tilbake til oversikten (norsk)

Reproducing Larsson & Wolk (2006) on red meat and colorectal cancer: an exact reproduction, a stated exposure rule and an unverifiable dose–response estimate

A re-analysis under a pre-specified protocol

Abstract

Background. Larsson and Wolk (2006) pooled prospective studies of red meat and colorectal cancer. Highest versus lowest intake gave a relative risk (RR) of 1.28 (95 % CI 1.15–1.42; 15 studies), and the dose–response analysis gave 1.28 (1.18–1.39) per 120 g/day (14 studies). WCRF/AICR 2007 and the 2011 Continuous Update Project report cited the dose–response figure. We tried to reproduce both estimates.

Methods. Under a pre-specified protocol, we pooled the per-study values printed in Table I and Fig 1 by DerSimonian–Laird random effects and compared each with the primary paper. Deviations were classified and corrected one class at a time and together, and later follow-ups were substituted. AI agents did the extraction; independent AI controllers checked it, with a logged correction record.

Results. The highest-versus-lowest estimate reproduces exactly: 1.28 (1.15–1.42), Q = 4.86, I² 0 %. Each of the 14 printed values we could check matches the primary paper or a within-study pooling of its site-specific rows; the fifteenth cohort is not in our database. One of those poolings (Chao 2005) was done by random effects although footnote 6 describes inverse-variance weighting; redoing it as stated leaves the pooled estimate at 1.28 (1.16–1.41). The dose–response estimate could not be reproduced, because no per-study slopes are printed. Three rows (Pietinen 1999, Larsson 2005, Norat 2005) include processed meat, as the paper's stated rule prescribed when both definitions were reported. Substituting their unprocessed rows gives 1.22 (1.10–1.35). Also removing the mortality cohort and correcting the Chao pooling gives 1.22 (1.11–1.33). Replacing the two Harvard cohorts with their later follow-up gives 1.15 (1.06–1.25).

Conclusions. The highest-versus-lowest analysis is transparent and arithmetically correct. Its choices are stated, and its own sensitivity analyses anticipate ours. The dose–response estimate, which later reports carried forward, cannot be checked from the publication.

1. Introduction

Larsson and Wolk [1] restricted their meta-analysis to prospective studies published to March 2006. The red-meat analyses were:

The paper updated Norat et al. (2002) [2], whose reference list it checked and whose 120 g portion size it adopted (pp. 2657, 2660).

Later reports relied on the dose–response figure in ways that blur what it measures:

This paper asks whether the two estimates can be rebuilt from the published inputs. It also asks how far they depend on the exposure definition and on which publication of each cohort is used.

2. Methods

2.1 Protocol

The re-analysis follows a protocol fixed before any run (REANALYSE_PROTOKOLL.md, version 1.0, 30 October 2026).

Scope. Cohort and nested case–control studies; red meat as the synthesis defined it; colorectal, colon or rectal cancer incidence (§2).

Reproduction rules. - R1: use the synthesis's own model and increment. - R2: use exactly its study list. - R3: use its own per-study values where printed. Reproduction succeeds when the point estimate and both confidence limits match to two decimals. - R4: where values are not printed, derive them from the primary paper by Greenland–Longnecker, using the synthesis's dose rules.

Deviation classes (R5). exposure, dose, cases, model, duplicate, transcription and inclusion.

Corrected and follow-up runs (R6). Corrected runs apply one deviation class at a time, then all classes together. A further run substitutes later follow-ups. Leave-one-out is reported for every run.

Exposure correction. In corrected runs, unprocessed red meat is used "where the paper separates it" (§2).

2.2 Larsson & Wolk's stated method

Inclusion criteria (p. 2657). Included studies had to: - be prospective in design; - report RRs with confidence intervals "(or data to calculate them)" for red or processed meat and "incidence of or mortality from colon, rectal or colorectal cancer".

Studies that reported only total meat were omitted.

Study selection. - Four publications were excluded as superseded by later reports of the same cohorts (Larsson refs 14–17). - Of the two Iowa Women's Health Study publications, Bostick 1994 was preferred to Sellers 1998 [7]. Bostick "focused on meat consumption and adjusted for more covariates" (p. 2657).

Choice of estimate. From each study the authors took the RRs "that reflected the greatest degree of control for potential confounders" (p. 2657).

Exposure rule (Table I, footnote 4, p. 2659). The paper states its choice between definitions of red meat:

"Relative risks for total red meat (fresh red meat plus processed meat) were chosen when provided; otherwise, relative risks for fresh red meat were included."

Pooling (p. 2657). Pooling was by DerSimonian and Laird [8] random effects. Where a study reported separate RRs by site or sex, Larsson and Wolk "pooled the RRs, weighted by the inverse of the variance, within each study" (also Table I, footnote 6).

Dose–response analysis (p. 2660). - Method: Greenland–Longnecker [9, 10], or variance-weighted least squares where cases and non-cases by category were not reported. - Portion size: 120 g for red meat. - Category dose: the median or mean intake, otherwise the category midpoint. - Open upper category: assumed to have "the same amplitude as the preceding category". - Software: Stata 9.0.

2.3 Data

Larsson & Wolk's own values. We transcribed the per-study highest-versus-lowest RRs and 95 % CIs from Table I (pp. 2658–2659) into synthesis_extraction.csv. That file holds 50 rows for this synthesis: the 15 main-analysis rows, the site and subsite rows, and 14 dose–response membership rows with blank values. Fig 1 (p. 2660) was rendered and compared with Table I. Its 15 printed RRs and CIs are identical to Table I. Fig 1 shows weights only as box sizes, not as numbers.

Primary-paper values. These come from the project database (estimates.csv). Every row was verified against the full text by a controller who had not seen the extraction. Hsing 1998 is not in the database.

2.4 Follow-ups

For the follow-up run, the Nurses' Health Study (NHS) and Health Professionals Follow-up Study (HPFS) rows of Wei 2004 [11] were replaced by Bernstein et al. (2015) [12]. We used unprocessed red meat, >1 serving/day versus the lowest category, by sex (Supplementary Tables S1–S2).

No other cohort in Larsson's set has a later highest-versus-lowest estimate in the database that scripts/reanalyses.py could substitute.

2.5 Software and statistics

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

For the dose–response comparison, scripts/reanalyses.py took per-100 g estimates from derived_per100g.csv and rescaled them to 120 g (log RR × 1.2). Those estimates are either the primary paper's own continuous estimate or a Greenland–Longnecker slope from scripts/gl_doseresponse.py with the codebook dose rules (KODEBOK §4.3).

2.6 AI-assisted extraction and control

All extraction, coding and control were done by AI agents (Anthropic Claude Opus) under the author's direction, using a codebook with fixed vocabularies.

Every number in this paper comes from the run files kontroll/reanalyser/larsson2006.md/.json or from the cited source page.

3. Results

3.1 Per-study inputs against the primary papers

Table 1. Larsson & Wolk 2006, red-meat highest-versus-lowest inputs (Table I pp. 2658–2659 = Fig 1 p. 2660) compared with the primary papers.

Study (cohort) Cases (Table I) Larsson RR (95 % CI) Definition per Larsson Table II fn 4 Primary paper (table, page) Primary paper value Processed meat in the row used?
Bostick 1994 (IWHS) [13] 212 colon 1.04 (0.62–1.76) both reported → total Table 3, p. 43 1.04 (0.62–1.76) unclear (not defined)
Kato 1997 (NYUWHS) [14] 100 1.23 (0.68–2.22) items not specified Table 3, p. 279 1.23 (0.68–2.22) unclear
Chen 1998 (PHS, nested c–c) [15] 212 1.17 (0.68–2.02) total only Table 2, p. 3309 1.17 (0.68–2.02) yes
Hsing 1998 (Lutheran Brotherhood) [16] 145 (fatal, fn 7) 1.90 (0.90–4.30) total only not in database – –
Singh 1998 (AHS) [17] 157 colon (incl. 22 rectosigmoid, fn 8) 1.41 (0.90–2.21) fresh only Table 3, p. 766 1.41 (0.90–2.21) unclear
Pietinen 1999 (ATBC) [18] 185 1.10 (0.70–1.70) both reported → total Table 5, p. 393 total 1.10 (0.70–1.70); beef, pork, lamb 0.80 (0.50–1.20) yes
Järvinen 2001 (FMCHES) [19] 109 1.50 (0.77–2.94) items not specified Table 3, p. 359 1.50 (0.77–2.94) unclear
Tiemersma 2002 (Dutch Monitoring Project, nested c–c) [20] 102 1.60 (0.90–2.90) fresh only Table 2, p. 388 1.6 (0.9–2.9) no
Flood 2003 (BCDDP) [21] 487 1.10 (0.83–1.45) total only Table 4, p. 64 1.10 (0.83–1.45) yes
Wei 2004, NHS [11] 876 1.21 (0.72–2.03) fresh only Table II, p. 439: colon 1.31 (0.73–2.36), rectum 0.92 (0.31–2.71) pooled by Larsson (fn 6) "main dish", no hamburger
Wei 2004, HPFS [11] 602 1.24 (0.78–1.96) fresh only Table II, p. 439: colon 1.35 (0.80–2.27), rectum 0.90 (0.34–2.45) pooled by Larsson (fn 6) "main dish", no hamburger
English 2004 (MCCS) [22] 451 1.40 (1.00–1.90) fresh only Table 3, p. 1512 1.4 (1.0–1.9) no
Larsson 2005 (SMC) [23] 733 1.32 (1.03–1.68) both reported → total Table II, p. 831 total ≥94 g/day 1.32 (1.03–1.68); beef and pork main dish ≥4/week 1.22 (0.98–1.53) yes
Chao 2005 (CPS-II Nutrition) [24] 1,667 1.36 (0.93–2.00) total only Table 2, pp. 176–177: colon 1.15 (0.90–1.46); Table 3, p. 178: rectum 1.71 (1.15–2.52) pooled by Larsson; matches a random-effects pool, whereas the fixed-effect pool described in fn 6 gives 1.28 (1.04–1.58) yes
Norat 2005 (EPIC) [25] 1,329 1.35 (0.96–1.88) both reported → total Table 3, p. 911 total ≥160 vs <20 g/day 1.35 (0.96–1.88); fresh ≥80 vs <10 g/day 1.17 (0.92–1.49) yes

What the comparison shows. - Single estimates. Every Larsson value that corresponds to a single primary-paper estimate matches that estimate exactly. - Pooled rows. Wei NHS, Wei HPFS and Chao were pooled within study by Larsson, who states this in footnote 6. The site-specific rows Larsson prints for these three studies (Table I) match the primary papers. For Wei NHS and Wei HPFS the pooled row is consistent with the inverse-variance pooling footnote 6 describes. For Chao it is not: Larsson's 1.36 (0.93–2.00) is a random-effects pool of the colon and rectum rows, while a fixed-effect inverse-variance pool gives 1.28 (1.04–1.58) (deviation table, chao2005 / model). - Case count. The case column sums to 7,367, as stated on p. 2660. - Fig 1 against Table I. K11 confirmed that the two agree for all 15 rows (K11 §1b).

3.2 Pooled estimates

Table 2. Published, reproduced, corrected and follow-up estimates (DerSimonian–Laird random effects).

Analysis Run name in larsson2006.md k RR (95 % CI) Q I² Reproduced?
Published HvL (p. 2660; Fig 1) – 15 1.28 (1.15–1.42) 4.86 0 % –
Reproduced from Table I / Fig 1 CRC HvL – published per-study values (Table I / Fig 1) 15 1.28 (1.15–1.42) 4.86 0 % yes
Corrected: inclusion CRC HvL – corrected: inclusion 14 1.27 (1.14–1.42) 3.87 0 % –
Corrected: exposure CRC HvL – corrected: exposure 15 1.22 (1.10–1.35) 8.44 0 % –
Corrected: model CRC HvL – corrected: model 15 1.28 (1.16–1.41) 4.76 0 % –
Corrected: all classes CRC HvL – corrected: all classes 14 1.22 (1.11–1.33) 7.12 0 % –
Corrected + follow-ups CRC HvL – corrected + follow-ups 14 1.15 (1.06–1.25) 12.10 0 % –
Published per 120 g/day (Table III, p. 2662) – 14 1.28 (1.18–1.39) 8.82 0 % could not be reproduced
Our approximation per 120 g/day, 8 of 14 studies DR reconstruction per 120 g – our GL values for Larsson's DR studies where gram doses exist 8 1.27 (1.10–1.46) 4.01 0 % not a reproduction

Main analysis. The highest-versus-lowest analysis reproduces exactly: point estimate, both confidence limits and Q (run: "CRC HvL – published per-study values (Table I / Fig 1)"). With τ² = 0, the fixed-effect estimate is identical.

Weights. Larsson & Wolk do not print weights. The random-effects weights in our reproduction are largest for Larsson 2005 (18.5 %), Flood 2003 (14.3 %), English 2004 (10.8 %) and Norat 2005 (9.8 %).

Subgroups. K11 also reproduced several of Larsson's printed subgroup estimates exactly (K11 §3): - Rectum: 1.56 (1.25–1.95), 7 studies. - Proximal colon: 1.15 (0.91–1.47). - Distal colon: 1.31 (0.62–2.79). - Colon: 1.21 (1.05–1.40), 9 studies, Q = 2.35 (Table II, p. 2661). This reproduces only when Singh 1998 is left out of the ten colon rows of Table I. The paper does not say why Singh is absent from the colon subgroup.

3.3 Deviations

The deviation list in larsson2006.md ("Deviations") contains eleven entries. For each, we note whether the paper itself disclosed the point.

  1. Hsing 1998: inclusion (removed). - Fatal cases only (Table I, fn 7). - Larsson & Wolk's inclusion criteria explicitly admit mortality, and their own sensitivity analysis excluding "the 2 studies based on colorectal cancer mortality" gives 1.27 (1.14–1.42) (p. 2660). - Our inclusion-corrected run gives the same value. Of the two references cited for that sensitivity analysis (refs 7 and 20), Kojima 2004 has no red-meat estimate in Table I ("NA"). The published sensitivity analysis is therefore in effect the removal of Hsing alone.
  2. Pietinen 1999: exposure (substituted). - Larsson's row is ATBC "total red meat", 203 vs 79 g/day, which includes processed meat (Pietinen Table 5, p. 393; Chan et al. 2011, Table S1 [26]). - Beef, pork and lamb alone give 0.80 (0.50–1.20). - Larsson's footnote 4 rule prescribes the total row, and Table II fn 4 lists Pietinen (ref. 23) among the four studies reporting both definitions.
  3. Norat 2005: exposure (substituted). - Total red meat including processed, ≥160 vs <20 g/day: 1.35 (0.96–1.88). - Fresh red meat, ≥80 vs <10 g/day: 1.17 (0.92–1.49) (Norat 2005, Table 3, p. 911). - The total row was chosen by the stated rule.
  4. Larsson 2005: exposure (substituted). - Total red meat including processed, ≥94 vs <47 g/day: 1.32 (1.03–1.68). - Beef and pork as a main dish, ≥4 servings/week: 1.22 (0.98–1.53) (Larsson 2005, Table II, p. 831). - The total row was chosen by the stated rule.
  5. Chao 2005: exposure (documented only). "Red meat" includes processed items, and no unprocessed row exists.
  6. Chao 2005: model (substituted). - Larsson's CRC row, 1.36 (0.93–2.00), is not printed by Chao; Larsson derived it from Chao's colon row, 1.15 (0.90–1.46), and rectum row, 1.71 (1.15–2.52). - Footnote 6 says such rows were pooled "weighted by the inverse of the variance". A fixed-effect inverse-variance pool gives 1.28 (1.04–1.58); Larsson's value is the random-effects pool of the same two rows. - Substituted in the model and all-classes runs.
  7. Bostick 1994: exposure (documented only). Table II fn 4 lists Bostick (ref. 18) among the studies reporting both total and fresh red meat. Larsson's row is the paper's "red meat", >11 vs <4 servings/week, and the main-dish variable includes ham; neither is a clean unprocessed row, so Larsson's row is kept.
  8. Flood 2003: exposure (documented only). As for Chao.
  9. Singh 1998: cases (documented only). The 157 "colon" cases include 22 rectosigmoid cancers. This is disclosed in Table I, fn 8.
  10. Chen 1998: inclusion (documented only). A nested case–control study. Disclosed, with a sensitivity analysis excluding both nested studies (Chen, Tiemersma): 1.28 (1.15–1.42) (p. 2660).
  11. Wei 2004: exposure (documented only). "Red meat as main dish" excludes hamburger and mixed dishes, in both the NHS and HPFS.

There are no dose, duplicate or transcription deviations in the highest-versus-lowest analysis.

3.4 Corrected and follow-up runs

Exposure corrections. The three exposure substitutions give 1.22 (1.10–1.35) (run: "CRC HvL – corrected: exposure").

Model correction. Replacing Larsson's Chao row with the fixed-effect pool of its colon and rectum rows gives 1.28 (1.16–1.41) (run: "CRC HvL – corrected: model"). The point estimate is unchanged; Chao's weight rises from 7.6 % to 22.1 %.

All classes. Applying the exposure, model and inclusion corrections together gives 1.22 (1.11–1.33) (run: "CRC HvL – corrected: all classes").

Follow-ups. We then replaced the Wei 2004 NHS and HPFS rows with Bernstein 2015: - NHS: 0.96 (0.76–1.21), weight 13.1 %. - HPFS: 0.98 (0.75–1.27), weight 10.2 %.

The estimate becomes 1.15 (1.06–1.25) (run: "CRC HvL – corrected + follow-ups").

Heterogeneity. I² is 0 % in every run.

[Figure 1. Forest plots for every run in Table 2. See the HTML report, kontroll/reanalyser/html/index.html, panel forest-larsson2006.]

3.5 Leave-one-out

Reproduced run. Leave-one-out estimates range from 1.27 to 1.32. The largest shift comes from omitting Flood 2003: 1.32 (1.17–1.47).

All-classes run. The range is 1.20, without English 2004 (1.09–1.32) or without Chao 2005 (1.08–1.33), to 1.24 (1.13–1.36), without Pietinen 1999.

Follow-up run. The range is 1.13 to 1.18. Every lower confidence limit stays at or above 1.03: - without Chao 2005: 1.13 (1.03–1.24) - without English 2004: 1.14 (1.04–1.24) - without Wei NHS: 1.18 (1.08–1.30)

No single study carries any run.

3.6 Dose–response per 120 g/day

The dose–response estimate of 1.28 (1.18–1.39) could not be reproduced. What the paper prints: - the 14 studies, by reference number: refs 4, 5, 8–13, 18–20, 23 and 24 (p. 2662); - the exclusion of Kato 1997 "because red meat consumption could not be quantified"; - the summary in Table III.

What it does not print: - any study's slope; - the category doses it assigned; - which studies were analysed by Greenland–Longnecker and which by variance-weighted least squares; - the study lists behind the colon (n = 10) and rectum (n = 7) dose–response estimates.

Fig 1 is the red-meat highest-versus-lowest plot and Fig 2 the processed-meat one; there is no dose–response forest plot.

Why R4 cannot substitute for the missing values. Applying R4 (derivation from the primary papers with Larsson's rules) would reproduce Larsson's method only if we also knew which RR row Larsson used for each study. Footnote 4 implies the total-red-meat row where both exist. It also requires knowing whether Larsson used medians or midpoints for each study. Neither is recorded for the dose–response analysis.

Our approximation. For comparison we report what the database's own per-100 g estimates give when rescaled to 120 g: 1.27 (1.10–1.46), k = 8 (run: "DR reconstruction per 120 g – our GL values for Larsson's DR studies where gram doses exist"). This is not a reproduction: - Missing studies. It lacks six of the 14 studies: Chen, Hsing, Flood, Wei NHS, Wei HPFS and Chao. Our table has no gram-based categories for them. - Different exposure rows. Several of the eight use rows that Larsson's rule would not have chosen. For example, Pietinen, Norat 2005 and Tiemersma use unprocessed red meat, and Bostick and Larsson 2005 use the main-dish rows. - Different methods. The eight mix the primary papers' own continuous estimates with our Greenland–Longnecker slopes.

4. Discussion

4.1 What the reproduction shows

The highest-versus-lowest analysis is unusually transparent. - Each study's RR and CI is printed twice, in Table I and Fig 1, consistently. - Within-study pooling and the choice between exposure definitions are both stated in footnotes. - Every value can be traced to the primary paper (for Chao, to a within-study pooling done by a different model from the one footnote 6 describes). - The pooled estimate and Q reproduce exactly. - Larsson & Wolk anticipated the two inclusion questions we raise, mortality and nested case–control designs, and ran the sensitivity analyses themselves.

4.2 Errors or defensible choices?

Only one listed deviation departs from the authors' own stated method, and it does not change the result. Footnote 6 describes inverse-variance pooling within studies, but the Chao row is a random-effects pool. Pooling it as stated changes Chao's row from 1.36 (0.93–2.00) to 1.28 (1.04–1.58) and leaves the summary at 1.28 (1.16–1.41). This is a small inconsistency between method and execution, not a material error. The other deviations are stated choices or properties of the primary studies.

Exposure. The three exposure substitutions reverse a rule the authors stated openly. "Relative risks for total red meat (fresh red meat plus processed meat) were chosen when provided" (Table I, fn 4). - Preferring total red meat is a defensible choice for a question framed as "red meat" in 2006. - Larsson & Wolk also reported the alternative: their fresh-red-meat-only stratum gives 1.22 (1.08–1.37), n = 9, and their total-red-meat stratum 1.24 (1.09–1.42), n = 8 (Table II, p. 2661). - Our exposure-corrected run, 1.22 (1.10–1.35), agrees with their own fresh-only stratum to two decimals. - In other words, the paper already contained the answer to our exposure correction.

Inclusion. The mortality and nested-design questions are equally transparent. The authors' inclusion criteria admitted both, and their sensitivity analyses, 1.27 and 1.28, show that neither choice matters.

Remaining definitional differences. Two exposure points can only be documented: - Chao 2005 and Flood 2003 have no unprocessed row. - Wei 2004's "main dish" variable omits hamburger.

These are properties of the primary studies, not of the synthesis.

4.3 What the corrected and follow-up runs mean

Corrected runs. Under the protocol's definitions the estimate moves from 1.28 to 1.22 (1.11–1.33). Precision is retained, and no leave-one-out omission removes it.

Follow-up run. Substituting Bernstein 2015 for Wei 2004 gives 1.15 (1.06–1.25). The NHS and HPFS follow-ups are not definitionally identical to Wei 2004: - Bernstein's "unprocessed red meat" includes hamburger, which is coded partial (KODEBOK §4.2b), whereas Wei's main-dish variable did not. - The top category is >1 serving/day rather than ≥5 times/week.

So this run shows how the same cohorts' later reports bear on the pooled estimate. It is not a corrected version of Larsson & Wolk's analysis.

Effect of the exposure definition. In this set of cohorts, the definition moves the highest-versus-lowest estimate by about 0.06. Updating the two Harvard cohorts moves it by a further 0.07.

4.4 What later reports did with the estimate

The dose–response estimate is the one that travelled, and it is the one that cannot be checked: - WCRF/AICR 2007 paired it with the study count of the highest-versus-lowest analysis. - CUP 2011 paired it with that analysis's case count. - The 2010 SLR and CUP 2011 described the exposure as "red (and processed) meat" or "red and processed meat". This is a closer description of the main-analysis rows chosen under footnote 4 than of the paper's own label, "red meat". As far as the printed text shows, it is not a description of a separate analysis.

Section 1 gives the page references for these citations. The misdescriptions are the later reports', not Larsson & Wolk's.

4.5 Limitations

  1. The dose–response analysis could not be reproduced. No per-study slopes are printed, so it could not be checked or corrected (§3.6).
  2. Hsing 1998 is not in the database. Its value is Larsson's, cross-checked only against Norat 2002 Appendix III.
  3. Fig 1 weights are not printed. Our weights are recomputed, not read.
  4. The follow-up run substitutes only the two Harvard cohorts. These are the only ones with a later highest-versus-lowest estimate that the script could use.
  5. OCR and AI extraction. The primary-paper checks rest on AI extraction verified by independent AI controllers. The published database, control reports and correction log allow every value to be re-checked by a human reader.

5. Data and code availability

The following files are published with this paper:

References

  1. Larsson SC, Wolk A. Meat consumption and risk of colorectal cancer: a meta-analysis of prospective studies. Int J Cancer 2006;119:2657–64.
  2. Norat T, Lukanova A, Ferrari P, Riboli E. Meat consumption and colorectal cancer risk: dose-response meta-analysis of epidemiological studies. Int J Cancer 2002;98:241–56.
  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. WCRF/AICR Systematic Literature Review Continuous Update Project Report: The Associations between Food, Nutrition and Physical Activity and the Risk of Colorectal Cancer. London: Imperial College London; 2010.
  5. World Cancer Research Fund/American Institute for Cancer Research. Continuous Update Project Report. Food, Nutrition, Physical Activity, and the Prevention of Colorectal Cancer. 2011.
  6. International Agency for Research on Cancer. Red Meat and Processed Meat. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans, Vol. 114. Lyon: IARC; 2018.
  7. Sellers TA, Bazyk AE, Bostick RM, Kushi LH, Olson JE, Anderson KE, et al. Diet and risk of colon cancer in a large prospective study of older women: an analysis stratified on family history (Iowa, United States). Cancer Causes Control 1998;9:357–67.
  8. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986;7:177–88.
  9. 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.
  10. Orsini N, Bellocco R, Greenland S. Generalized least squares for trend estimation of summarized dose-response data. Stata J 2006;6:40–57.
  11. 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.
  12. Bernstein AM, Song M, Zhang X, Pan A, Wang M, Fuchs CS, et al. Processed and unprocessed red meat and risk of colorectal cancer: analysis by tumor location and modification by time. PLoS ONE 2015;10(8):e0135959.
  13. Bostick RM, Potter JD, Kushi LH, Sellers TA, Steinmetz KA, McKenzie DR, 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.
  14. 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.
  15. Chen J, Stampfer MJ, Hough HL, Garcia-Closas M, Willett WC, Hennekens CH, et al. A prospective study of N-acetyltransferase genotype, red meat intake, and risk of colorectal cancer. Cancer Res 1998;58:3307–11.
  16. Hsing AW, McLaughlin JK, Chow WH, Schuman LM, Co Chien HT, Gridley G, et al. Risk factors for colorectal cancer in a prospective study among U.S. white men. Int J Cancer 1998;77:549–53.
  17. Singh PN, Fraser GE. Dietary risk factors for colon cancer in a low-risk population. Am J Epidemiol 1998;148:761–74.
  18. Pietinen P, Malila N, Virtanen M, Hartman TJ, Tangrea JA, Albanes D, et al. Diet and risk of colorectal cancer in a cohort of Finnish men. Cancer Causes Control 1999;10:387–96.
  19. 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.
  20. Tiemersma EW, Kampman E, Bueno de Mesquita HB, Bunschoten A, van Schothorst EM, Kok FJ, et al. 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.
  21. Flood A, Velie EM, Sinha R, Chaterjee N, Lacey JV Jr, Schairer C, et al. Meat, fat, and their subtypes as risk factors for colorectal cancer in a prospective cohort of women. Am J Epidemiol 2003;158:59–68.
  22. 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.
  23. 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.
  24. Chao A, Thun MJ, Connell CJ, McCullough ML, Jacobs EJ, Flanders WD, et al. Meat consumption and risk of colorectal cancer. JAMA 2005;293:172–82.
  25. Norat T, Bingham S, Ferrari P, Slimani N, Jenab M, Mazuir 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.
  26. Chan DSM, Lau R, Aune D, Vieira R, Greenwood DC, Kampman E, et al. Red and processed meat and colorectal cancer incidence: meta-analysis of prospective studies. PLoS ONE 2011;6(6):e20456.

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.