# Updated analysis, October 2026: unprocessed red meat and colorectal cancer incidence in prospective cohorts (CUP 2016 method)

**Main result: RR per 100 g/day = 1.10 (1.05–1.15)** (DerSimonian–Laird random effects; k = 19 cohorts; I² = 15 %; τ² = 0.0016). REML: 1.10 (1.05–1.16) (τ² 0.0016). HKSJ: 1.10 (1.04–1.16). Fixed effect: 1.10 (1.06–1.14). Egger intercept 0.44, p = 0.28.

| Cohort (publication) | RR per 100 g (95 % CI) | Weight % | Region | Sex | Exposure as defined | Cases | Provenance |
|---|---|---|---|---|---|---|---|
| Netherlands MP (Tiemersma 2002) | 1.92 (0.86–4.30) | 0.4 | Europe | both | fresh red meat | 102 | tiemersma2002_010;tiemersma2002_011;tiemersma2002_012 [GL_categories; conversion:cup:cup_rule] |
| MCCS (English 2004) | 1.19 (0.89–1.57) | 2.9 | Australia | both | fresh red meat | 451 | english2004_005 [continuous_scaled; per_serving_week x5.833 (portion 120.0 g)] |
| EPIC (Papadimitriou 2022) | 1.06 (0.95–1.14) | 16.3 | Europe | both | red meat (EPIC definition, fresh) | 5,069 | papadimitriou2022_001 [^2.762] |
| SMC (Larsson 2005) | 1.23 (0.90–1.67) | 2.4 | Europe | women | beef/pork main dish (unprocessed) | 733 | larsson2005_033;larsson2005_034;larsson2005_035;larsson2005_036 [GL_categories; conversion:cup:paper_median] |
| MEC (Ollberding 2012) | 0.97 (0.83–1.13) | 8.3 | N America | both | red meat excluding processed | 3,404 | ollberding2012_066;ollberding2012_067;ollberding2012_068;ollberding2012_069;ollberding2012_070 [GL_categories; conversion:cup:paper_median] |
| NLCS (Gilsing 2015) | 1.00 (0.81–1.23) | 4.9 | Europe | both | fresh red meat | 437 | gilsing2015_118 [continuous_scaled; per_50g] |
| JPHC + Takayama (Islam 2019) | 1.09 (0.91–1.31) | 6.2 | Asia | both | unprocessed red meat | 3,550 | islam2019_149 [continuous_scaled; per_100g]; islam2019_274 [continuous_scaled; per_100g] [strata combined, fixed] |
| UK Biobank (Knuppel 2020) | 1.50 (1.10–2.04) | 2.4 | Europe | both | unprocessed red meat | 3,201 | knuppel2020_101;knuppel2020_102;knuppel2020_103;knuppel2020_104 [GL_categories; paper_dose_point] |
| DANSDA (Mejborn 2021) | 1.04 (0.67–1.61) | 1.2 | Europe | both | unprocessed red meat | 127 | mejborn2021_006 [continuous_scaled; per_100g] |
| Israel pooled (Dankner 2024) | 1.15 (0.87–1.52) | 2.9 | Other | both | unprocessed red meat | 813 | dankner2024_016 [continuous_scaled; per_100g] |
| Sister Study (Mehta 2020) | 1.30 (0.77–2.19) | 0.9 | N America | women | unprocessed red meat | 210 | mehta2020_016 [continuous_scaled; per_50g] |
| NIH-AARP (Zouiouich 2025) | 1.13 (1.06–1.20) | 24.9 | N America | both | unprocessed red meat | 9,461 | zouiouich2025_134 |
| NHS + HPFS (Bernstein 2015) | 1.00 (0.89–1.11) | 13.5 | N America | both | unprocessed red meat (incl. hamburger) | 2,731 | bernstein2015_488 [continuous_scaled; per_100g] |
| Million Women (Papier 2025) | 1.28 (1.00–1.54) | 4.7 | Europe | women | red meat (unprocessed, partial) | 12,251 | papier2025_007 [continuous_scaled; per_20g] |
| Alberta's Tomorrow (Al Rajabi 2021) | 1.25 (0.68–2.28) | 0.7 | N America | both | unprocessed red meat | 257 | alrajabi2021_057;alrajabi2021_059;alrajabi2021_061;alrajabi2021_063 [GL_categories; conversion::cup_rule]; alrajabi2021_058;alrajabi2021_060;alrajabi2021_062;alrajabi2021_064 [GL_categories; conversion::cup_rule] [strata combined, fixed] |
| BWHS (Yiannakou 2022) | 1.63 (0.84–3.14) | 0.6 | N America | women | unprocessed red meat (partial) | 564 | yiannakou2022_025;yiannakou2022_026;yiannakou2022_027 [GL_categories; conversion:paper:cup_rule] |
| Finnish cohorts (Tammi 2024) | 1.23 (1.00–1.50) | 5.2 | Europe | both | unprocessed red meat | 1,124 | tammi2024_003 [continuous_scaled; per_100g_week] |
| Korea (Wie 2014) | 0.90 (0.48–1.79) | 0.6 | Asia | both | unprocessed red meat | 53 | wie2014_007 [continuous_scaled; per_10g] |
| SWHS (Lee 2009) | 0.81 (0.50–1.29) | 1.1 | Asia | women | red meat (undefined; mainly pork) | 394 | CUP 2016 cup2016_lee2009_colorectal |

Leave-one-out: without tiemersma2002: 1.10 (1.04–1.15); without english2004: 1.10 (1.04–1.16); without papadimitriou2022: 1.11 (1.05–1.17); without larsson2005: 1.10 (1.04–1.15); without ollberding2012: 1.11 (1.06–1.16); without gilsing2015: 1.10 (1.05–1.16); without islam2019: 1.10 (1.04–1.16); without knuppel2020: 1.09 (1.05–1.14); without mejborn2021: 1.10 (1.04–1.16); without dankner2024: 1.10 (1.04–1.16); without mehta2020: 1.10 (1.04–1.16); without zouiouich2025: 1.09 (1.03–1.16); without bernstein2015: 1.11 (1.06–1.17); without papier2025: 1.09 (1.04–1.14); without alrajabi2021: 1.10 (1.04–1.16); without yiannakou2022: 1.10 (1.04–1.15); without tammi2024: 1.09 (1.04–1.15); without wie2014: 1.10 (1.05–1.16); without lee2009: 1.10 (1.05–1.16)

## Subgroups

| Subgroup | k | RR (95 % CI) | I² |
|---|---|---|---|
| region: Europe | 8 | 1.16 (1.05–1.28) | 32 % |
| region: N America | 6 | 1.07 (0.98–1.17) | 38 % |
| region: Asia | 3 | 1.04 (0.88–1.22) | 0 % |
| women-only cohorts | 5 | 1.22 (1.05–1.43) | 0 % |
| mixed/men cohorts | 14 | 1.09 (1.03–1.14) | 15 % |
| explicitly unprocessed (database flag = no; excludes partial/undefined rows) | 13 | 1.11 (1.03–1.20) | 0 % |
| partial / undefined exposure rows only | 6 | 1.09 (1.01–1.18) | 45 % |
| published ≥ 2015 | 13 | 1.11 (1.05–1.17) | 13 % |
| published < 2015 | 6 | 1.06 (0.91–1.23) | 20 % |
| published ≤ 2015 (overview article split) | 8 | 1.01 (0.94–1.09) | 0 % |
| published > 2015 (overview article split) | 11 | 1.13 (1.08–1.18) | 0 % |
| ≥ 1000 cases | 8 | 1.10 (1.03–1.18) | 49 % |
| < 1000 cases | 11 | 1.11 (1.00–1.25) | 0 % |

## Sensitivity

| Analysis | k | RR (95 % CI) | I² |
|---|---|---|---|
| without Million Women (calibrated) | 18 | 1.09 (1.04–1.14) | 12 % |
| without NIH-AARP (energy-density assumption) | 18 | 1.09 (1.03–1.16) | 14 % |
| without nested case-control (Tiemersma) | 18 | 1.10 (1.04–1.15) | 12 % |
| without the two largest (MWS, NIH-AARP) | 17 | 1.07 (1.02–1.14) | 8 % |
| UK Biobank → Hoang 2023 (frequency categories, CUP dose rules 120 g/serving) | 19 | 1.11 (1.05–1.17) | 23 % |
| with Malmö DCS (Vulcan 2017) added despite EPIC overlap | 20 | 1.10 (1.05–1.15) | 11 % |
| NHS/HPFS → Wang 2020 (NHS+NHSII+HPFS, per 3 servings/wk → per 100 g at 120 g/serving; age strata combined) | 19 | 1.10 (1.06–1.15) | 5 % |

## Multiverse over extraction choices

1152 combinations (Netherlands MP (Tiemersma 2002): 2 × MCCS (English 2004): 2 × EPIC (Papadimitriou 2022): 3 × SMC (Larsson 2005): 2 × MEC (Ollberding 2012): 2 × JPHC + Takayama (Islam 2019): 2 × UK Biobank (Knuppel 2020): 2 × NIH-AARP (Zouiouich 2025): 3 × model). Median RR 1.11 (range 1.09–1.14); lower CI limit > 1 in 1152 of 1152 (100 %).

| Choice | n | significant | median RR |
|---|---|---|---|
| Netherlands MP (Tiemersma 2002) = main | 576 | 576 (100 %) | 1.11 |
| Netherlands MP (Tiemersma 2002) = CUP value | 576 | 576 (100 %) | 1.11 |
| MCCS (English 2004) = main | 576 | 576 (100 %) | 1.11 |
| MCCS (English 2004) = adjusted for processed meat | 576 | 576 (100 %) | 1.11 |
| EPIC (Papadimitriou 2022) = main | 384 | 384 (100 %) | 1.10 |
| EPIC (Papadimitriou 2022) = Norat 2005 observed | 384 | 384 (100 %) | 1.12 |
| EPIC (Papadimitriou 2022) = Norat 2005 calibrated | 384 | 384 (100 %) | 1.11 |
| SMC (Larsson 2005) = main | 576 | 576 (100 %) | 1.11 |
| SMC (Larsson 2005) = total red meat per 100 g (paper) | 576 | 576 (100 %) | 1.11 |
| MEC (Ollberding 2012) = main | 576 | 576 (100 %) | 1.11 |
| MEC (Ollberding 2012) = total red meat (CUP's row) | 576 | 576 (100 %) | 1.11 |
| JPHC + Takayama (Islam 2019) = main | 576 | 576 (100 %) | 1.11 |
| JPHC + Takayama (Islam 2019) = Wada 2017 (Takayama only) | 576 | 576 (100 %) | 1.12 |
| UK Biobank (Knuppel 2020) = main | 576 | 576 (100 %) | 1.11 |
| UK Biobank (Knuppel 2020) = Bradbury 2020 (same cohort) | 576 | 576 (100 %) | 1.11 |
| NIH-AARP (Zouiouich 2025) = main | 384 | 384 (100 %) | 1.11 |
| NIH-AARP (Zouiouich 2025) = at 1800 kcal | 384 | 384 (100 %) | 1.12 |
| NIH-AARP (Zouiouich 2025) = at 2200 kcal | 384 | 384 (100 %) | 1.11 |
| model = random | 576 | 576 (100 %) | 1.11 |
| model = fixed | 576 | 576 (100 %) | 1.11 |

## Negative controls (same method, main-model rows, one per cohort)

- poultry: 0.95 (0.80–1.13), k = 6, I² 23 % (dankner2024, english2004, gilsing2015, mejborn2021, norat2005, bradbury2020); weights dankner2024 41 %, english2004 12 %, gilsing2015 6 %, mejborn2021 6 %, norat2005 22 %, bradbury2020 12 %; leave-one-out without dankner2024: 0.88 (0.71–1.10); without english2004: 1.02 (0.89–1.17); without gilsing2015: 0.96 (0.79–1.16); without mejborn2021: 0.93 (0.77–1.11); without norat2005: 0.94 (0.74–1.19); without bradbury2020: 0.93 (0.75–1.15); dropped for cohort overlap (R101): spencer2010
- fish: 0.76 (0.64–0.90), k = 4, I² 0 % (english2004, gilsing2015, norat2005, bradbury2020); weights english2004 12 %, gilsing2015 9 %, norat2005 65 %, bradbury2020 14 %; leave-one-out without english2004: 0.73 (0.61–0.88); without gilsing2015: 0.75 (0.63–0.90); without norat2005: 0.87 (0.65–1.17); without bradbury2020: 0.74 (0.62–0.89); dropped for cohort overlap (R101): spencer2010

## Absolute risk (illustration, assumes the whole association is causal)

Formula: risk(g) = 6.5 % x RR^((g - 89)/100); per 1,000 = 10 x difference in %; cases/year = 4876 x (1 - 1/RR^((89-50)/100)). Baseline 6.5 % before age 80 at 89 g/day (Norkost 4 mean, red meat incl. processed, product weight; the RR is for unprocessed meat, so this is approximate).

| Comparison | Difference per 1,000 before age 80 (RR point, lower, upper) |
|---|---|
| 0 vs 100 g/day | 5.90 (2.84–8.84) |
| 50 vs 100 g/day | 3.02 (1.44–4.58) |
| 89 (mean) down to 50 g/day | 2.34 (1.12–3.54) |
| cases per year avoided, 89 → 50 g/day (4,876 cases in 2025) | 176 (84–266) |
| Europe subgroup RR, 89 → 50 g/day, per 1,000 | 3.66 (1.19–6.03) |

## Non-linear dose-response (restricted cubic spline, 3 knots, two-stage, multivariate REML)

Knots at 9, 38, 84 g/day; 12 category sets (pietinen1999 men, tiemersma2002 all, larsson2005 women, norat2005 all, ollberding2012 all, knuppel2020 all, mehta2020 women, wada2017 men, wada2017 women, alrajabi2021 men, alrajabi2021 women, yiannakou2022 women). Wald test for non-linearity: χ² = 0.38, df = 1, p = 0.54. Heterogeneity Q = 20.0 on 22 df, p = 0.58.

Linear two-stage slope on the same 12 sets: 1.24 (1.06–1.44) per 100 g (DL).

| Dose g/day | RR vs 0 (95 % CI) |
|---|---|
| 25 | 1.07 (1.00–1.16) |
| 50 | 1.14 (1.01–1.28) |
| 75 | 1.19 (1.04–1.35) |
| 100 | 1.23 (1.06–1.44) |
| 125 | 1.28 (1.05–1.55) |
| 150 | 1.32 (1.04–1.69) |

## Excluded from the main pool (and why)

- Wang 2020 (NHS + NHSII + HPFS): new in the 2026 search (fetched by the author); overlaps NHS/HPFS; no whole-population all-CRC estimate for unprocessed red meat (only by age at onset) and no serving size defined — shown as sensitivity replacing Bernstein 2015
- Nguyen 2020 (Shanghai Women's Health Study): fetched; red and processed meat only as a diet-score component
- Vulcan 2017 (Malmö Diet and Cancer Study): new in the 2026 search; MDCS is the Malmö centre of EPIC, already inside Papadimitriou 2022 — overlap; shown as sensitivity
- Andersen 2019 (Danish DCH), Beslay 2021 (NutriNet-Santé): new in the 2026 search; red and processed meat combined only (Beslay full text fetched and confirmed: HR per 50 g/day red+processed 1.42 (1.03–1.94) for CRC)
- Pan 2023 (China Kadoorie): new in the 2026 search; early-onset CRC only, red meat frequency undefined
- Järvinen 2001 (Finnish Mobile Clinic): exposure is total meat
- Chao 2005 (CPS-II): red meat includes processed
- Flood 2003 (BCDDP): red meat includes processed; superseded by nothing
- Spencer 2010 (UK Dietary Cohort Consortium): nested case-control overlapping EPIC-Norfolk/EPIC-Oxford
- Takachi 2011, Oba 2006, Wada 2017: replaced by Islam 2019 (same cohorts)
- Bradbury 2020: UK Biobank, earlier/overlapping analysis of the cohort used through Knuppel 2020 (multiverse alternative)
- Hoang 2023 (UK Biobank): later and larger UK Biobank analysis (4,681 cases), but only frequency categories (times/week) with no gram intake; Knuppel 2020 gives category means in grams and is kept. Hoang converted with CUP's rules (120 g/serving, midpoints) is a sensitivity analysis, not a multiverse option (R100; K26d)
- Cross 2010: NIH-AARP earlier follow-up (Zouiouich 2025 used)
- Etemadi 2018: incidence paper pooling NIH-AARP, PLCO (604 CRC cases) and the Agricultural Health Study (299); substitution-type models adjusted for total meat; cohort-specific estimates only in Suppl. Table 1, not on disk, so PLCO and AHS are not represented. NIH-AARP is covered by Zouiouich 2025. Open item: Etemadi 2018 Suppl. Table 1 not yet obtained (R102; K26d). Earlier stated reason 'NIH-AARP earlier follow-up / mortality' was wrong
- Jin 2023: per SD with no SD in grams
- Butler 2008 (Singapore): highest vs lowest only
- Key 2014: diet-group comparison
- Bostick 1994, Singh 1998, Sellers 1998: colon only (colon pool)
- Willett 1990, Giovannucci 1994, Wei 2004, Chan 2005: NHS/HPFS earlier
- Norat 2005: EPIC earlier (multiverse)
- Kato 1997, Chen 1998, Phillips 1975, Goldbohm 1994, Fraser 1999, Kabat 2007, Nöthlings 2009: no per-gram unprocessed estimate / superseded