# Red meat and colon, rectal, proximal- and distal-colon cancer: site-specific 2026 analysis (protocol addendum R8)

Script `scripts/kolon2026.py`; numbers to 4 decimals in `kolon2026.json`. Protocol `PROTOKOLL_KOLON_R8.md` v1.0 with its change log (30 Oct 2026, 18:20 EDT); row choices from `kontroll/kolon2026/INVENTAR.md`.

## Methods

- **Cohort set (R8.1).** The cohorts of the 2026 colorectal pool (`update2026.json`; 19 after R100, 20 when R8 was written – ATBC was removed because it is inside Tammi 2024 and reports no site rows). A cohort enters a site analysis only if its publication or supplement reports that site for unprocessed red meat, or the nearest definition under the R7 hierarchy. Red meat including processed meat is never used.
- **R8.1 substitution.** EPIC site rows come from Norat 2005, because Papadimitriou 2022 has no red-meat site analysis.
- **Row choice (R8.2).**
  - Main model, highest adjustment, not calibrated.
  - Continuous per-unit rows are scaled to per 100 g/day (per 50 g → ^2; per serving/week × 5.833 at 120 g/serving).
  - Category-only rows use Greenland–Longnecker (GL), with the pipeline's documented covariance variant (K19). Values come from `database/derived_per100g.csv` where present.
  - The SMC proximal and distal sets are not in that file, because the pipeline has no women-specific site total. They were computed here with `gl_doseresponse.gl_trend` and the same rules: the CUP dose rule (paper medians × 120 g/7), and the Table II site totals (234 and 155) split in proportion to RR.
- **Colon from subsites (R8.3).** SMC, NIH-AARP and BWHS do not report colon overall. Their colon row is the fixed-effect combination of the proximal and distal rows, treated as independent. These rows are flagged.
- **Sex strata (R8.4).** JPHC + Takayama strata are combined by fixed effect.
- **Pooling (R8.5).**
  - Main method: DerSimonian–Laird random effects (`pool.py`).
  - Also reported: REML, Hartung–Knapp–Sidik–Jonkman (HKSJ, on the DL τ²), fixed effect, I², τ² and leave-one-out.
  - Egger's test is run only when k ≥ 10.
  - Cohort overlap is checked through `cohorts.csv` parents.
- **Ratio tests (R8.6e).** Ratio of the pooled RRs. The CI is exp(Δ ± 1.96·√(se₁² + se₂²)), from the pooled DL log-RRs treated as independent.
  - This is an approximation. Both pools come from largely the same cohorts and participants.
  - Positive within-cohort correlation would make the true CI narrower, so the test is conservative under that assumption. No within-cohort covariances are published.
- **Change-log rows used only as sensitivity analyses:** UK Biobank subsites from Bradbury 2020; EPIC right/left colon from Norat 2005; calibrated and regression-dilution-corrected rows.
- **Islam 2019 case counts** are taken from Suppl. S3/S4 inside the script. `estimates.csv` is not edited.

## Per-cohort rows (RR per 100 g/day, 95 % CI)

| Cohort (site publication) | Region | Women only | Colon | Rectum | Proximal | Distal | Cases C / R / P / D | Colon row |
|---|---|---|---|---|---|---|---|---|
| MCCS (English 2004) | Australia | no | 1.00 (0.70–1.48) | 1.57 (0.94–2.38) | – | – | 283 / 169 / – / – | direct |
| EPIC (Norat 2005) | Europe | no | 1.20 (0.96–1.48) | 1.23 (0.94–1.62) | – | – | 855 / 474 / – / – | direct |
| SMC (Larsson 2005) | Europe | yes | 1.68 (1.11–2.54) | 0.98 (0.56–1.72) | 1.17 (0.66–2.05) | 2.59 (1.40–4.79) | 389 / 230 / 234 / 155 | constructed (R8.3) |
| NLCS (Gilsing 2015) | Europe | no | 0.94 (0.74–1.21) | 0.90 (0.56–1.42) | – | – | 307 / 92 / – / – | direct |
| JPHC + Takayama (Islam 2019) | Asia | no | 1.05 (0.88–1.26) | 1.27 (0.90–1.78) | 0.80 (0.62–1.03) | 1.15 (0.95–1.39) | 2,552 / 1,127 / 1,200 / 1,150 | direct |
| UK Biobank (Knuppel 2020) | Europe | no | 1.86 (1.27–2.73) | 0.97 (0.59–1.61) | – | – | 2,114 / 1,109 / – / – | direct |
| Sister Study (Mehta 2020) | N America | yes | 1.23 (0.62–2.46) | 1.61 (0.45–5.71) | – | – | 116 / 30 / – / – | direct |
| NIH-AARP (Zouiouich 2025) | N America | no | 1.13 (1.05–1.22) | 1.14 (1.00–1.29) | 1.09 (0.99–1.19) | 1.21 (1.07–1.37) | 7,002 / 2,208 / 4,565 / 2,437 | constructed (R8.3) |
| NHS + HPFS (Bernstein 2015) | N America | no | 0.99 (0.87–1.12) | 1.08 (0.86–1.35) | 1.12 (0.94–1.32) | 0.79 (0.65–0.97) | – / 589 / 1,151 / 817 | direct |
| SWHS (Lee 2009) | Asia | yes | 0.86 (0.43–1.74) | 0.74 (0.33–1.66) | – | – | 236 / 158 / – / – | direct |
| BWHS (Yiannakou 2022) | N America | yes | 0.74 (0.30–1.80) | 3.12 (0.62–15.81) | 0.58 (0.20–1.67) | 1.34 (0.26–7.01) | 314 / 104 / 222 / 92 | constructed (R8.3) |

Provenance of every row (estimate ids, method and dose source, then table and page from `estimates.csv`):

- **MCCS (English 2004)**: fresh red meat; per serving/week x 5.833 (120 g/serving); subsites 'data not shown' (147 proximal, 111 distal).
  - Colon: 1.00 (0.70–1.48). Ids: `english2004_010 [continuous_scaled; per_serving_week x5.833 (portion 120.0 g)]`. Source: Table 3, 1512.
  - Rectum: 1.57 (0.94–2.38). Ids: `english2004_015 [continuous_scaled; per_serving_week x5.833 (portion 120.0 g)]`. Source: Table 3, 1512.
- **EPIC (Norat 2005)**: R8.1 substitution: colorectal row is Papadimitriou 2022 (no red-meat site analysis); observed (not calibrated) per 100 g; right/left colon = sensitivity only (change log 3).
  - Colon: 1.20 (0.96–1.48). Ids: `norat2005_033 [continuous_scaled; per_100g]`. Source: Table 4, 912.
  - Rectum: 1.23 (0.94–1.62). Ids: `norat2005_035 [continuous_scaled; per_100g]`. Source: Table 4, 912.
- **SMC (Larsson 2005)**: beef/pork as main dish; GL with CUP dose rule (paper medians x 120 g); per-category site cases not printed (site totals split proportional to RR).
  - Colon: 1.68 (1.11–2.54). Ids: `larsson2005_037;larsson2005_038;larsson2005_039;larsson2005_040 [GL_categories (pipeline K19 covariance); conversion:cup:paper_median; cases ASSUMED: site total 234 (Table II) split proportional to RR]; larsson2005_041;larsson2005_042;larsson2005_043;larsson2005_044 [GL_categories (pipeline K19 covariance); conversion:cup:paper_median; cases ASSUMED: site total 155 (Table II) split proportional to RR] [R8.3: proximal + distal combined, fixed]`. Source: Table II, 831.
  - Rectum: 0.98 (0.56–1.72). Ids: `larsson2005_045;larsson2005_046;larsson2005_047;larsson2005_048 [GL_categories; conversion:cup:paper_median]`. Source: Table II, 831.
  - Proximal colon: 1.17 (0.66–2.05). Ids: `larsson2005_037;larsson2005_038;larsson2005_039;larsson2005_040 [GL_categories (pipeline K19 covariance); conversion:cup:paper_median; cases ASSUMED: site total 234 (Table II) split proportional to RR]`. Source: Table II, 831.
  - Distal colon: 2.59 (1.40–4.79). Ids: `larsson2005_041;larsson2005_042;larsson2005_043;larsson2005_044 [GL_categories (pipeline K19 covariance); conversion:cup:paper_median; cases ASSUMED: site total 155 (Table II) split proportional to RR]`. Source: Table II, 831.
- **NLCS (Gilsing 2015)**: fresh red meat; per 50 g -> ^2; case-cohort; subsites 'data not shown'.
  - Colon: 0.94 (0.74–1.21). Ids: `gilsing2015_128 [continuous_scaled; per_50g]`. Source: Table 4, Sci Rep 5:13484 (PMC text).
  - Rectum: 0.90 (0.56–1.42). Ids: `gilsing2015_138 [continuous_scaled; per_50g]`. Source: Table 4, Sci Rep 5:13484 (PMC text).
- **JPHC + Takayama (Islam 2019)**: per 100 g; sex strata (Tables 4/5) combined, fixed; case counts from Suppl. S3/S4 (change log 4; n_cases empty in estimates.csv).
  - Colon: 1.05 (0.88–1.26). Ids: `islam2019_174 [continuous_scaled; per_100g]; islam2019_299 [continuous_scaled; per_100g] [strata combined, fixed]`. Source: Table 4, page not marked in txt; Table 5, page not marked in txt.
  - Rectum: 1.27 (0.90–1.78). Ids: `islam2019_199 [continuous_scaled; per_100g]; islam2019_324 [continuous_scaled; per_100g] [strata combined, fixed]`. Source: Table 4, page not marked in txt; Table 5, page not marked in txt.
  - Proximal colon: 0.80 (0.62–1.03). Ids: `islam2019_224 [continuous_scaled; per_100g]; islam2019_349 [continuous_scaled; per_100g] [strata combined, fixed]`. Source: Table 4, page not marked in txt; Table 5, page not marked in txt.
  - Distal colon: 1.15 (0.95–1.39). Ids: `islam2019_249 [continuous_scaled; per_100g]; islam2019_374 [continuous_scaled; per_100g] [strata combined, fixed]`. Source: Table 4, page not marked in txt; Table 5, page not marked in txt.
- **UK Biobank (Knuppel 2020)**: GL from category means, model 2 (Suppl. S5); proximal/distal only in Bradbury 2020 (different follow-up; sensitivity only, change log 2).
  - Colon: 1.86 (1.27–2.73). Ids: `knuppel2020_116;knuppel2020_117;knuppel2020_118;knuppel2020_119 [GL_categories; paper_dose_point]`. Source: Suppl Table S5, suppl p. 19.
  - Rectum: 0.97 (0.59–1.61). Ids: `knuppel2020_131;knuppel2020_132;knuppel2020_133;knuppel2020_134 [GL_categories; paper_dose_point]`. Source: Suppl Table S5, suppl p. 19.
- **Sister Study (Mehta 2020)**: per 50 g -> ^2; subsites not analysed (few cases).
  - Colon: 1.23 (0.62–2.46). Ids: `mehta2020_235 [continuous_scaled; per_50g]`. Source: Suppl Table 1, suppl.
  - Rectum: 1.61 (0.45–5.71). Ids: `mehta2020_240 [continuous_scaled; per_50g]`. Source: Suppl Table 1, suppl.
- **NIH-AARP (Zouiouich 2025)**: addition model, per 50 g/1000 kcal = per 100 g/day at 2000 kcal/day (same assumption as the colorectal row); colon NOS (C18.8-18.9) in neither subsite.
  - Colon: 1.13 (1.05–1.22). Ids: `zouiouich2025_142; zouiouich2025_182 [R8.3: proximal + distal combined, fixed]`. Source: Fig 2, 6.
  - Rectum: 1.14 (1.00–1.29). Ids: `zouiouich2025_214`. Source: Fig 2, 6.
  - Proximal colon: 1.09 (0.99–1.19). Ids: `zouiouich2025_142`. Source: Fig 2, 6.
  - Distal colon: 1.21 (1.07–1.37). Ids: `zouiouich2025_182`. Source: Fig 2, 6.
- **NHS + HPFS (Bernstein 2015)**: Suppl. S5 per 100 g (unprocessed incl. hamburger); colon cases not printed (>= 1151 + 816); distal 816 in main text vs 817 (S3 + S4).
  - Colon: 0.99 (0.87–1.12). Ids: `bernstein2015_494 [continuous_scaled; per_100g]`. Source: Suppl Table S5, suppl.
  - Rectum: 1.08 (0.86–1.35). Ids: `bernstein2015_512 [continuous_scaled; per_100g]`. Source: Suppl Table S5, suppl.
  - Proximal colon: 1.12 (0.94–1.32). Ids: `bernstein2015_500 [continuous_scaled; per_100g]`. Source: Suppl Table S5, suppl.
  - Distal colon: 0.79 (0.65–0.97). Ids: `bernstein2015_506 [continuous_scaled; per_100g]`. Source: Suppl Table S5, suppl.
- **SWHS (Lee 2009)**: CUP 2016's extraction (as the colorectal row); red meat undefined (mainly pork), lowest rung of the R8.1 hierarchy; own GL not possible (Q2-Q4 CIs not printed).
  - Colon: 0.86 (0.43–1.74). Ids: `CUP 2016 cup2016_lee2009_colon (SLR 2016 F132)`. Source: SLR 2016 F132.
  - Rectum: 0.74 (0.33–1.66). Ids: `CUP 2016 cup2016_lee2009_rectum (SLR 2016 F139)`. Source: SLR 2016 F139.
- **BWHS (Yiannakou 2022)**: GL from tertiles (85 g serving, CUP midpoint rule); unprocessed, partial (incl. 'dinner ham'); log10 slope not usable.
  - Colon: 0.74 (0.30–1.80). Ids: `yiannakou2022_061;yiannakou2022_062;yiannakou2022_063 [GL_categories; conversion:paper:cup_rule]; yiannakou2022_065;yiannakou2022_066;yiannakou2022_067 [GL_categories; conversion:paper:cup_rule] [R8.3: proximal + distal combined, fixed]`. Source: Table 3, 1258.
  - Rectum: 3.12 (0.62–15.81). Ids: `yiannakou2022_069;yiannakou2022_070;yiannakou2022_071 [GL_categories; conversion:paper:cup_rule]`. Source: Table 3, 1258.
  - Proximal colon: 0.58 (0.20–1.67). Ids: `yiannakou2022_061;yiannakou2022_062;yiannakou2022_063 [GL_categories; conversion:paper:cup_rule]`. Source: Table 3, 1258.
  - Distal colon: 1.34 (0.26–7.01). Ids: `yiannakou2022_065;yiannakou2022_066;yiannakou2022_067 [GL_categories; conversion:paper:cup_rule]`. Source: Table 3, 1258.

## Results (R8.6 a–d)

| Site | k | DL random (main) | REML | HKSJ | Fixed | I² % | τ² (DL) | Q (p) | Egger intercept (p) |
|---|---|---|---|---|---|---|---|---|---|
| Colon | 11 | **1.11 (1.01–1.23)** | 1.11 (1.01–1.23) | 1.11 (0.98–1.27) | 1.10 (1.05–1.16) | 46.06 | 0.0099 | 18.54 (0.05) | 0.19 (0.78) |
| Rectum | 11 | **1.14 (1.04–1.25)** | 1.14 (1.04–1.25) | 1.14 (1.05–1.25) | 1.14 (1.04–1.25) | 0.00 | 0.0000 | 7.24 (0.70) | 0.14 (0.76) |
| Proximal colon | 5 | **1.03 (0.89–1.18)** | 1.02 (0.87–1.19) | 1.03 (0.83–1.27) | 1.06 (0.99–1.15) | 42.72 | 0.0093 | 6.98 (0.14) | k < 10 |
| Distal colon | 5 | **1.17 (0.89–1.53)** | 1.21 (0.84–1.76) | 1.17 (0.72–1.88) | 1.11 (1.02–1.22) | 80.42 | 0.0584 | 20.43 (0.00) | k < 10 |

τ² is shown to 4 decimals because 2 decimals would round it to 0.00.

**Colon.** Random-effects weights: MCCS (English 2004) 5.62 %; EPIC (Norat 2005) 11.97 %; SMC (Larsson 2005) 4.83 %; NLCS (Gilsing 2015) 10.30 %; JPHC + Takayama (Islam 2019) 14.44 %; UK Biobank (Knuppel 2020) 5.52 %; Sister Study (Mehta 2020) 1.99 %; NIH-AARP (Zouiouich 2025) 23.35 %; NHS + HPFS (Bernstein 2015) 18.81 %; SWHS (Lee 2009) 1.93 %; BWHS (Yiannakou 2022) 1.23 %.

Leave-one-out:

- without MCCS (English 2004): 1.12 (1.01–1.25)
- without EPIC (Norat 2005): 1.10 (0.99–1.24)
- without SMC (Larsson 2005): 1.09 (0.99–1.19)
- without NLCS (Gilsing 2015): 1.13 (1.02–1.26)
- without JPHC + Takayama (Islam 2019): 1.13 (1.00–1.27)
- without UK Biobank (Knuppel 2020): 1.08 (1.00–1.17)
- without Sister Study (Mehta 2020): 1.11 (1.00–1.23)
- without NIH-AARP (Zouiouich 2025): 1.12 (0.98–1.28)
- without NHS + HPFS (Bernstein 2015): 1.14 (1.02–1.28)
- without SWHS (Lee 2009): 1.12 (1.01–1.24)
- without BWHS (Yiannakou 2022): 1.12 (1.01–1.24)

**Rectum.** Random-effects weights: MCCS (English 2004) 3.82 %; EPIC (Norat 2005) 11.03 %; SMC (Larsson 2005) 2.54 %; NLCS (Gilsing 2015) 3.83 %; JPHC + Takayama (Islam 2019) 7.03 %; UK Biobank (Knuppel 2020) 3.20 %; Sister Study (Mehta 2020) 0.51 %; NIH-AARP (Zouiouich 2025) 50.39 %; NHS + HPFS (Bernstein 2015) 16.07 %; SWHS (Lee 2009) 1.25 %; BWHS (Yiannakou 2022) 0.31 %.

Leave-one-out:

- without MCCS (English 2004): 1.13 (1.03–1.24)
- without EPIC (Norat 2005): 1.13 (1.03–1.24)
- without SMC (Larsson 2005): 1.15 (1.05–1.26)
- without NLCS (Gilsing 2015): 1.15 (1.05–1.26)
- without JPHC + Takayama (Islam 2019): 1.13 (1.03–1.24)
- without UK Biobank (Knuppel 2020): 1.15 (1.05–1.26)
- without Sister Study (Mehta 2020): 1.14 (1.04–1.25)
- without NIH-AARP (Zouiouich 2025): 1.14 (1.00–1.30)
- without NHS + HPFS (Bernstein 2015): 1.15 (1.04–1.27)
- without SWHS (Lee 2009): 1.15 (1.05–1.26)
- without BWHS (Yiannakou 2022): 1.14 (1.04–1.25)

**Proximal colon.** Random-effects weights: SMC (Larsson 2005) 5.47 %; JPHC + Takayama (Islam 2019) 19.37 %; NIH-AARP (Zouiouich 2025) 43.60 %; NHS + HPFS (Bernstein 2015) 29.88 %; BWHS (Yiannakou 2022) 1.68 %.

Leave-one-out:

- without SMC (Larsson 2005): 1.01 (0.86–1.19)
- without JPHC + Takayama (Islam 2019): 1.09 (1.01–1.19)
- without NIH-AARP (Zouiouich 2025): 0.97 (0.75–1.24)
- without NHS + HPFS (Bernstein 2015): 0.96 (0.76–1.22)
- without BWHS (Yiannakou 2022): 1.04 (0.91–1.19)

**Distal colon.** Random-effects weights: SMC (Larsson 2005) 12.06 %; JPHC + Takayama (Islam 2019) 27.72 %; NIH-AARP (Zouiouich 2025) 30.30 %; NHS + HPFS (Bernstein 2015) 27.46 %; BWHS (Yiannakou 2022) 2.46 %.

Leave-one-out:

- without SMC (Larsson 2005): 1.05 (0.82–1.33)
- without JPHC + Takayama (Islam 2019): 1.22 (0.81–1.84)
- without NIH-AARP (Zouiouich 2025): 1.21 (0.79–1.84)
- without NHS + HPFS (Bernstein 2015): 1.28 (1.04–1.58)
- without BWHS (Yiannakou 2022): 1.17 (0.88–1.54)

## Tests of difference (R8.6 e)

| Contrast | Ratio of pooled RRs (95 % CI) | z | p |
|---|---|---|---|
| colon / rectum | 0.98 (0.85–1.12) | -0.37 | 0.72 |
| proximal / distal | 0.88 (0.65–1.19) | -0.82 | 0.41 |
| rectum / proximal (supplementary, for R8.7) | 1.11 (0.94–1.31) | 1.24 | 0.21 |
| rectum / proximal, same 5 cohorts (supplementary, for R8.7) | 1.11 (0.93–1.32) | 1.15 | 0.25 |

For the same-5-cohort comparison, the rectum pool restricted to the five subsite cohorts is 1.14 (1.03–1.26).

Approximation: the two pooled log-RRs come largely from the same cohorts and participants and are treated as independent; positive within-cohort correlation (shared confounder adjustment, shared exposure measurement) would make the true CI of the ratio narrower, so these tests are conservative under that assumption. No within-cohort covariances are published.

## Subgroups (R8.6 f; run only where k ≥ 3)

**Colon**

| Subgroup | k | RR (95 % CI) | I² % |
|---|---|---|---|
| region: Europe | 4 | 1.32 (0.99–1.77) | 73.48 |
| region: N America | 4 | 1.07 (0.97–1.18) | 25.86 |
| region: Asia | 2 | not run (k < 3) | – |
| women-only cohorts | 4 | 1.20 (0.82–1.76) | 31.02 |
| mixed-sex cohorts | 7 | 1.10 (0.99–1.22) | 54.60 |
| published <= 2015 | 6 | 1.07 (0.93–1.24) | 40.05 |
| published > 2015 | 5 | 1.17 (0.99–1.39) | 50.83 |

**Rectum**

| Subgroup | k | RR (95 % CI) | I² % |
|---|---|---|---|
| region: Europe | 4 | 1.09 (0.89–1.33) | 0.00 |
| region: N America | 4 | 1.13 (1.02–1.27) | 0.00 |
| region: Asia | 2 | not run (k < 3) | – |
| women-only cohorts | 4 | 1.04 (0.68–1.58) | 0.00 |
| mixed-sex cohorts | 7 | 1.15 (1.04–1.26) | 0.00 |
| published <= 2015 | 6 | 1.12 (0.97–1.30) | 0.00 |
| published > 2015 | 5 | 1.15 (1.03–1.29) | 0.00 |

**Proximal colon**

| Subgroup | k | RR (95 % CI) | I² % |
|---|---|---|---|
| region: Europe | 1 | not run (k < 3) | – |
| region: N America | 3 | 1.09 (1.01–1.18) | 0.00 |
| region: Asia | 1 | not run (k < 3) | – |
| women-only cohorts | 2 | not run (k < 3) | – |
| mixed-sex cohorts | 3 | 1.03 (0.88–1.20) | 64.40 |
| published <= 2015 | 2 | not run (k < 3) | – |
| published > 2015 | 3 | 0.92 (0.69–1.24) | 68.70 |

**Distal colon**

| Subgroup | k | RR (95 % CI) | I² % |
|---|---|---|---|
| region: Europe | 1 | not run (k < 3) | – |
| region: N America | 3 | 1.00 (0.68–1.48) | 84.24 |
| region: Asia | 1 | not run (k < 3) | – |
| women-only cohorts | 2 | not run (k < 3) | – |
| mixed-sex cohorts | 3 | 1.04 (0.81–1.34) | 84.57 |
| published <= 2015 | 2 | not run (k < 3) | – |
| published > 2015 | 3 | 1.19 (1.07–1.32) | 0.00 |

The publication period is that of the publication supplying the site rows. EPIC is Norat 2005 here, so it falls in the ≤ 2015 group.

## Sensitivity analyses (R8.6 g)

| Analysis | k | RR (95 % CI) | I² % |
|---|---|---|---|
| colon: directly reported only | 8 | 1.09 (0.96–1.23) | 43.09 |
| colon: constructed from subsites only (R8.3) | 3 | 1.22 (0.88–1.68) | 53.33 |
| proximal_colon: + UK Biobank (Bradbury 2020 Suppl. S9, GL) | 6 | 1.04 (0.92–1.17) | 28.50 |
| distal_colon: + UK Biobank (Bradbury 2020 Suppl. S9, GL) | 6 | 1.27 (0.97–1.67) | 80.39 |
| proximal_colon: + EPIC (Norat 2005 right side, GL) | 6 | 1.05 (0.92–1.18) | 33.45 |
| proximal_colon: + UK Biobank (Bradbury) + EPIC (Norat) | 7 | 1.05 (0.95–1.17) | 20.21 |
| distal_colon: + EPIC (Norat 2005 left side, GL) | 6 | 1.16 (0.92–1.47) | 75.71 |
| distal_colon: + UK Biobank (Bradbury) + EPIC (Norat) | 7 | 1.25 (0.98–1.58) | 76.56 |
| colon: calibrated EPIC (Norat) only | 11 | 1.11 (0.99–1.24) | 45.64 |
| colon: calibrated EPIC (Norat) + regression-dilution-corrected UK Biobank (Knuppel) | 11 | 1.11 (1.00–1.24) | 46.08 |
| rectum: calibrated EPIC (Norat) only | 11 | 1.14 (1.04–1.25) | 0.00 |
| rectum: calibrated EPIC (Norat) + regression-dilution-corrected UK Biobank (Knuppel) | 11 | 1.14 (1.04–1.26) | 0.00 |
| proximal_colon: + UK Biobank (Bradbury 2020, regression-dilution corrected per 50 g -> ^2) | 6 | 1.04 (0.92–1.17) | 28.45 |
| distal_colon: + UK Biobank (Bradbury 2020, regression-dilution corrected per 50 g -> ^2) | 6 | 1.28 (0.97–1.68) | 80.75 |
| colon: textbook GL covariance for the GL rows (UK Biobank (Knuppel 2020), SMC (Larsson 2005), BWHS (Yiannakou 2022)) | 11 | 1.11 (1.01–1.23) | 44.51 |
| rectum: textbook GL covariance for the GL rows (UK Biobank (Knuppel 2020), SMC (Larsson 2005), BWHS (Yiannakou 2022)) | 11 | 1.14 (1.04–1.25) | 0.00 |
| proximal_colon: textbook GL covariance for the GL rows (SMC (Larsson 2005), BWHS (Yiannakou 2022)) | 5 | 1.03 (0.89–1.18) | 41.86 |
| distal_colon: textbook GL covariance for the GL rows (SMC (Larsson 2005), BWHS (Yiannakou 2022)) | 5 | 1.16 (0.89–1.51) | 79.70 |
| colon: BWHS with the cohort's printed person-years instead of pseudo person-years | 11 | 1.11 (1.01–1.23) | 46.13 |
| rectum: BWHS with the cohort's printed person-years instead of pseudo person-years | 11 | 1.14 (1.04–1.25) | 0.00 |
| proximal_colon: BWHS with the cohort's printed person-years instead of pseudo person-years | 5 | 1.03 (0.89–1.18) | 42.80 |
| distal_colon: BWHS with the cohort's printed person-years instead of pseudo person-years | 5 | 1.17 (0.89–1.53) | 80.41 |

Rows used only in sensitivity analyses:

- UK Biobank, Bradbury 2020 (Suppl. S9), GL:
  - Proximal 1.10 (0.61–1.99). Ids: `bradbury2020_209;bradbury2020_210;bradbury2020_211;bradbury2020_212 [GL_categories; paper_dose_point]`.
  - Distal 2.32 (1.23–4.38). Ids: `bradbury2020_214;bradbury2020_215;bradbury2020_216;bradbury2020_217 [GL_categories; paper_dose_point]`.
- UK Biobank, Bradbury 2020, regression-dilution-corrected:
  - Proximal 1.04 (0.59–1.85). Ids: `bradbury2020_213 [continuous_scaled; per_50g]`.
  - Distal 2.40 (1.28–4.54). Ids: `bradbury2020_218 [continuous_scaled; per_50g]`.
- EPIC, Norat 2005 (Table 3, p. 911), GL:
  - Right side 1.25 (0.82–1.91). Ids: `norat2005_016;norat2005_017;norat2005_018;norat2005_019;norat2005_020 [GL_categories (pipeline K19 covariance); conversion::cup_rule; cases ASSUMED: site total 351 (Table 3) split proportional to RR]`.
  - Left side 1.21 (0.82–1.78). Ids: `norat2005_021;norat2005_022;norat2005_023;norat2005_024;norat2005_025 [GL_categories (pipeline K19 covariance); conversion::cup_rule; cases ASSUMED: site total 391 (Table 3) split proportional to RR]`.
  - Neither side has per-category cases.
  - The left side excludes the splenic flexure.
- Calibrated or regression-dilution-corrected rows:
  - EPIC (Norat 2005), colon: 1.36 (0.74–2.50). Ids: `norat2005_034 [continuous_scaled; per_100g]`. Source: Table 4, 912.
  - UK Biobank (Knuppel 2020), colon: 1.85 (1.28–2.69). Ids: `knuppel2020_120 [continuous_scaled; per_50g]`. Source: Suppl Table S5, suppl p. 19.
  - EPIC (Norat 2005), rectum: 1.75 (0.93–3.30). Ids: `norat2005_036 [continuous_scaled; per_100g]`. Source: Table 4, 912.
  - UK Biobank (Knuppel 2020), rectum: 0.98 (0.59–1.61). Ids: `knuppel2020_135 [continuous_scaled; per_50g]`. Source: Suppl Table S5, suppl p. 19.
- Textbook GL covariance:
  - UK Biobank (Knuppel 2020): colon 1.90 (1.29–2.81); rectum 0.95 (0.57–1.57).
  - SMC (Larsson 2005): proximal colon 1.11 (0.64–1.92); distal colon 2.33 (1.32–4.11); rectum 0.94 (0.54–1.62); colon 1.59 (1.07–2.36).
  - BWHS (Yiannakou 2022): proximal colon 0.59 (0.21–1.66); distal colon 1.38 (0.27–7.04); rectum 4.38 (0.83–23.19); colon 0.75 (0.31–1.81).
- BWHS with the printed cohort person-years (Table 2) in place of pseudo person-years: proximal colon 0.58 (0.20–1.67); distal colon 1.32 (0.25–6.89); rectum 2.88 (0.58–14.32); colon 0.74 (0.30–1.79).

**Checks.**

- SMC subsite GL against CUP 2016's own rows:
  - Proximal: 1.17 (0.66–2.05) here, against CUP 1.17 (0.66–2.05).
  - Distal: 2.59 (1.40–4.79) here, against CUP 2.59 (1.40–4.79).
  - Rectum: 0.98 (0.56–1.72) here, against CUP 0.98 (0.55–1.72).
- Pipeline GL recomputed here against the rows used: UK Biobank (Knuppel 2020) colon: 1.86 (1.27–2.73) vs 1.86 (1.27–2.73); UK Biobank (Knuppel 2020) rectum: 0.97 (0.59–1.61) vs 0.97 (0.59–1.61); SMC (Larsson 2005) proximal_colon: 1.17 (0.66–2.05) vs 1.17 (0.66–2.05); SMC (Larsson 2005) distal_colon: 2.59 (1.40–4.79) vs 2.59 (1.40–4.79); SMC (Larsson 2005) rectum: 0.98 (0.55–1.72) vs 0.98 (0.56–1.72); BWHS (Yiannakou 2022) proximal_colon: 0.58 (0.20–1.67) vs 0.58 (0.20–1.67); BWHS (Yiannakou 2022) distal_colon: 1.35 (0.26–7.01) vs 1.34 (0.26–7.01); BWHS (Yiannakou 2022) rectum: 3.12 (0.62–15.81) vs 3.12 (0.62–15.81).

## Reporting-bias check (R8.8)

- 11 of 19 cohorts report a usable colon and rectum row.
- 5 of 19 report usable proximal and distal rows.
- The 8 cohorts with no usable site row hold 18,131 of 44,932 colorectal cases (40.35 %).
- Those cohorts carry 23.99 % of the colorectal random-effects weight.
- Inventory claim (written for 20 cohorts: 9 non-reporting, 18,316 cases; after R100 8 and 18,316 − 185 = 18,131) verified: yes.
- Cohorts without subsite rows carry 52.41 % of the colorectal weight (27,893 cases).
- Cases in the site pools: colon 14,168; rectum 6,290; proximal colon 7,372; distal colon 4,651.
- The colon total excludes NHS + HPFS, whose colon case count is not printed.
- No imputation.

| Cohort (colorectal row) | CRC cases | CRC weight % | Site source | Colon | Rectum | Proximal | Distal | Why not |
|---|---|---|---|---|---|---|---|---|
| Netherlands MP (Tiemersma 2002) | 102 | 0.37 | – | no | no | no | no | site analysis for fish only |
| MCCS (English 2004) | 451 | 2.85 | MCCS (English 2004) | 283 | 169 | no | no |  |
| EPIC (Papadimitriou 2022) | 5,069 | 16.28 | EPIC (Norat 2005) | 855 | 474 | no | no |  |
| SMC (Larsson 2005) | 733 | 2.40 | SMC (Larsson 2005) | 389 | 230 | 234 | 155 |  |
| MEC (Ollberding 2012) | 3,404 | 8.35 | – | no | no | no | no | colon/rectum analysed, 'data not shown' |
| NLCS (Gilsing 2015) | 437 | 4.92 | NLCS (Gilsing 2015) | 307 | 92 | no | no |  |
| JPHC + Takayama (Islam 2019) | 3,550 | 6.24 | JPHC + Takayama (Islam 2019) | 2,552 | 1,127 | 1,200 | 1,150 |  |
| UK Biobank (Knuppel 2020) | 3,201 | 2.40 | UK Biobank (Knuppel 2020) | 2,114 | 1,109 | no | no |  |
| DANSDA (Mejborn 2021) | 127 | 1.24 | – | no | no | no | no | colorectal only |
| Israel pooled (Dankner 2024) | 813 | 2.94 | – | no | no | no | no | colorectal and GI only (suppl. S1-S6 checked) |
| Sister Study (Mehta 2020) | 210 | 0.89 | Sister Study (Mehta 2020) | 116 | 30 | no | no |  |
| NIH-AARP (Zouiouich 2025) | 9,461 | 24.89 | NIH-AARP (Zouiouich 2025) | 7,002 | 2,208 | 4,565 | 2,437 |  |
| NHS + HPFS (Bernstein 2015) | 2,731 | 13.50 | NHS + HPFS (Bernstein 2015) | yes (n not printed) | 589 | 1,151 | 817 |  |
| Million Women (Papier 2025) | 12,251 | 4.69 | – | no | no | no | no | sites only for red + processed meat (excl. R8.1) |
| Alberta's Tomorrow (Al Rajabi 2021) | 257 | 0.67 | – | no | no | no | no | no site breakdown in main text (Suppl. A-C not on disk) |
| BWHS (Yiannakou 2022) | 564 | 0.56 | BWHS (Yiannakou 2022) | 314 | 104 | 222 | 92 |  |
| Finnish cohorts (Tammi 2024) | 1,124 | 5.17 | – | no | no | no | no | colorectal only (Online Resources 1-5 checked) |
| Korea (Wie 2014) | 53 | 0.57 | – | no | no | no | no | colorectal only |
| SWHS (Lee 2009) | 394 | 1.07 | SWHS (Lee 2009) | 236 | 158 | no | no |  |

## Comparison

| Analysis | k | RR (95 % CI) | I² % |
|---|---|---|---|
| CUP 2016 colon (published 1.22 (1.06-1.39); reproduced) | 11 | 1.22 (1.06–1.39) | 11.89 |
| CUP 2016 colon set, corrected + later follow-ups (Islam 2019, Bernstein 2015) | 8 | 1.07 (0.94–1.20) | 16.07 |
| 2026 colorectal pool (update2026.json) | 19 | 1.10 (1.05–1.15) | 14.94 |
| **2026 colon, this analysis** | 11 | **1.11 (1.01–1.23)** | 46.06 |

CUP 2016's own pools restricted to the five cohorts common to its colorectal, colon and rectum analyses (SWHS, SMC, EPIC, MCCS, Järvinen; CUP's per-study values; SMC colon = proximal + distal, fixed): colorectal 1.19 (1.06–1.34); colon 1.21 (1.02–1.44); rectum 1.22 (1.00–1.49) (K23 claim 11).

## R8.7: are the two pre-declared predictions met?

*(Wording revised after independent control K22, which reproduced every number to 4 decimals; see kontroll/K22_kolon/RAPPORT.md.)*

**Prediction 1 (colon RR significantly above 1, as CUP 2017 relied on): met, but not robustly.**

- **Main result.** Across 11 cohorts the colon RR per 100 g/day of unprocessed red meat is 1.11 (1.01–1.23), p = 0.04 (DerSimonian–Laird; REML 1.11 (1.01–1.23)), with moderate heterogeneity (I² 46 %).
- **The size is stable.** It is about 1.1 under every choice we tested: leave-one-out 1.08–1.14; textbook GL 1.11; calibrated EPIC 1.11; alternative NIH-AARP and BWHS rows 1.11–1.12 (sensitivity_rows).
- **Statistical significance is not stable.**
  - The Hartung–Knapp interval is 0.98–1.27, and the 95 % prediction interval is 0.86–1.43.
  - The lower limit falls to 1 or below if any one of 3 cohorts is left out: EPIC, SMC, NIH-AARP.
  - NIH-AARP and SMC do not report colon cancer at all. Their colon estimates are built here from proximal and distal results.
  - EPIC enters with its 2005 follow-up (855 colon cases), because its current analysis (3,143 colon cases) gives no red-meat estimate by site.
  - Cohorts that report colon directly give 1.09 (0.96–1.23).
- **Compared with CUP and colorectal.** The colon estimate is about half of CUP 2016's 1.22 (1.06–1.39) on the log scale, and no larger than the colorectal estimate (1.10). Colon and rectum do not differ (ratio 0.98, 0.85–1.12).
- **Conclusion.** The updated evidence is compatible with a small colon association of about 10 % per 100 g/day, but it does not robustly confirm a colon-specific association.

**Prediction 2 (stronger association in the distal colon and rectum than in the proximal colon): met in direction only, not statistically.**

- **Rectum.** 1.14 (1.04–1.25), I² 0 %. This is the most consistent site result: it stays significant whichever single cohort is removed. About half its weight is NIH-AARP.
- **Distal colon.** 1.17 (0.89–1.53), I² 80 %. This rests on 5 cohorts that disagree strongly (NHS + HPFS 0.79, SMC 2.59). Without SMC it is 1.05 (0.82–1.33), the same as the proximal colon at 1.03 (0.89–1.18).
- **Tests of difference.** Neither difference is significant: proximal/distal 0.88 (0.65–1.19); rectum/proximal 1.11 (0.94–1.31).

**What is missing.**

- 8 of the 19 cohorts (40 % of colorectal cases, 24 % of the weight) publish no usable site result for unprocessed red meat. They include Million Women (12,251 cases) and MEC (3,404).
- MEC states that it analysed colon and rectum and found no association, but did not publish the numbers.

**Additional sensitivity analyses added after K22:** BWHS paper-printed continuous rows (colon 1.11 (1.01–1.22)); NIH-AARP substitution model (colon 1.12 (1.01–1.25)); Hartung–Knapp with truncation at the DL interval width gives rectum 1.03–1.26 (untruncated 1.05–1.25).

R8.9: independent recomputation and full-text check done as K22 (kontroll/K22_kolon/RAPPORT.md).