# NOTES – evidence tables WCRF 1997 and IARC 114 (red meat, colorectal cancer)

Written 2026-10-30. Brief: `/home/claude/db/OPPDRAG_evidenstabeller.md`. Read-only on `/home/claude/db/database/`. Scripts are in the scratchpad `/tmp/claude-0/-home-claude/e1a1fb2e-b9c0-5883-98be-1225e3955a50/scratchpad/evidens/` (`dump.py`, `triples.py`, `refs.py`, `build_wcrf.py`, `build_iarc.py`).

## 1. What could not be read

- **WCRF/AICR 1997 full text.** The Internet Archive item `foodnutritionp00amer` is lending-only:
  - metadata OK;
  - `_djvu.txt` → 302 then 401;
  - `_hocr_searchtext.txt.gz` and `_page_numbers.json` → 401;
  - `fulltext/inside.php` search → "Item not available";
  - WebFetch of `archive.org/details/foodnutritionp00amer?q="Table 4.10.10"` → metadata only.

  Table 4.10.10 is **not confirmed**. All WCRF values come from the card `kort/25_WCRF1997.md`. Exposure labels per row and the full Table 4.10.11 (26 case-control studies) are unknown.
- **Primary papers not in the database** (values cannot be checked against them): Phillips 1985, Knekt 1994 (WCRF); [Thun 1992 was added 30.10.2026 (K13, R65): its red-meat Q5 values men 1.21 and women 1.05 (unadjusted, no CI printed) match the WCRF 1997 card]; Berndt 2006, Sørensen 2008, Andersen 2009, Figueiredo 2014, Ananthakrishnan 2015 (IARC Table 2.2.1).
- **Meta-analyses not on disk:** Pham 2014, Carr 2016, Qiao & Feng 2013 and Alexander 2010. IARC's numbers for these are transcribed, not checked.
- `36_IARC114.txt` is a two-column layout. Text lines interleave left and right columns, so a line range can contain parts of two studies. Line numbers are therefore approximate to ±5 lines for text passages. They are exact for Table 2.2.1 rows. The reference list (lines ~11440–12000 and others) is **confirmed interleaved by column**. I de-interleaved the entries used here by column position, with `refs.py` plus manual checks of Andersen 2009, Zhang 2011 and Sellers 1998.

## 2. Judgment calls

1. **"IARC's comments column"** (brief) does not exist in Table 2.2.1. I used the Working Group's bracketed notes in the text, with page and line, in `iarc_wg_comment`.
2. **One summary row per study** in `iarc114_cohorts.csv`, holding the IARC top-vs-bottom (or per-increment) value for each organ site. All other category values are verified in `iarc114_table_values_check.csv`.
3. **Matching rule:** a database row "matches" when the study_id is the same and the point estimate and both CI limits are identical after rounding to 2 decimals. `db_is_main` is reported as is. A match to a non-main row is labelled `in_db_nonmain`.
4. **superseded_by**: used only when the database holds a later publication of the same cohort and outcome. These are Takayama → wada2017 and NIH-AARP → zouiouich2025. Both appeared after IARC, so they do not count against IARC. They are noted as later evidence. Weaker links are noted under "later_followup_in_db": ATBC → tammi2024 (pooled Finnish), EPIC → papadimitriou2022 (DWAS, not like-for-like), JPHC/Miyagi → islam2019, and NHS/HPFS early papers → bernstein2015 (IARC's own choice).
5. **The 14 cohorts** (IARC line 5660) were reconstructed from the order of section (a). The 2 consortia are UKDCC and GECCO/CCFR.
6. **The "seven positive" cohorts** (5.2.1, p. 493) are reconstructed as point estimate >1 for highest vs lowest. This gives exactly 7 (Kato, Tiemersma, Singh, Bostick, English, Larsson, Norat). IARC does not list them, so this is an inference.
7. **Singh 1998**: I treat the Table 3 full-cohort estimate 1.41 (0.90–2.21) as the comparable value (db main) and IARC's 1.90 (1.16–3.11) as the restricted-subset estimate. The paper's abstract features the subset result (db notes on singh1998_012). The choice is therefore defensible, but IARC's table mislabels what it is.
8. **Bernstein 2015**: the database has no `is_main_model=yes` row for unprocessed red meat per serving. The main flag sits on the per-100 g row from supplement S5. I report the per-serving Table 2/3 rows that IARC prints.
9. **WCRF Bostick 1994**: "no increase" vs main-dish 1.21 (0.75–1.96) is logged as a minor discrepancy (verbal null vs point estimate >1, NS), following the example in the brief. The paper's "red meat" variable is 1.04, which fits the verdict.
10. Citations I gave from memory (journal/volume/pages for Willett 1990, Giovannucci 1994, Goldbohm 1994, Bostick 1994, Manousos 1983, Miller 1983, Singh 1998, Pietinen 1999) were **not re-checked** against a reference list. DOIs and PMIDs marked [IARC] come from the IARC reference list.

## 3. Discrepancies found (report vs primary paper / database)

### IARC 114
| Sev. | Where | What |
|---|---|---|
| major | T2.2.1 p. 145, lines 7366–7378 (Singh & Fraser 1998) | Organ site printed "colon and rectum" (paper: colon only). RRs 1.40/1.90 come from the subset with white meat <1/wk (Table 4, 112 cases) but are printed with full-cohort case counts 42/40/45 and Table 3 covariates. Full-cohort RR is 1.41 (0.90–2.21) |
| major | 2.2.3 p. 143, lines 7266–7268 (Alexander 2015) | "1.02 (1.00–1.14) for 1 serving/day". Alexander prints 1.02 (1.00–1.04) per serving per **week** |
| moderate | 5.2.1 p. 493 line 23978 and Bouvard 2015 line 106 | Chan 2011 per 100 g CRC 1.17 (1.05–1.31) attributed to "10 cohort studies". Chan's table and IARC's own 2.2.3 (line 7252) say **8** |
| moderate | T2.2.1 p. 160, lines 8040–8046 (Bernstein 2015) | Rectum rows labelled "Processed red meat". The values are unprocessed red meat |
| moderate | T2.2.1 p. 156 (Sørensen 2008 + Andersen 2009) | The same DCH cohort is tabulated twice, against IARC's own rule (line 5658) |
| minor | T2.2.1 pp. 149–150 (Chao 2005) | Unit "g/day" should be g/week. Men Q5 cases 125 (paper 124) |
| minor | T2.2.1 p. 145 (Pietinen 1999) | Q2 0.6 (0.4–1.1); paper 0.6 (0.4–1.0) |
| minor | text p. 114 line 5840 (Norat 2005) | Pork CI "0.95–0.48"; paper 0.95–1.48 |
| minor | text p. 124 (Flood 2003) | Quintile medians described as cut-points (">52.2 vs <6.1 g/1000 kcal") |
| minor | T2.2.1 p. 145 vs text p. 123 (Chen 1998) | Table lists BMI as a covariate; text says BMI was not controlled for |
| minor | T2.2.1 p. 154 vs text p. 123 (Berndt 2006) | 272 vs 250 cases; covariates differ; text says tertiles |
| minor | T2.2.1 p. 156 (Sørensen 2008) | "Exposed cases 105" in a 379-case study |
| structural | T2.2.1 vs text p. 111 | IWHS (Bostick 1994) and UKDCC (Spencer 2010) are counted among the red-meat cohorts in the text but are absent from Table 2.2.1. The table instead includes 7 combined red+processed studies that IARC calls "less informative" |
| structural | T2.2.1 p. 159 | Figueiredo 2014 and Ananthakrishnan 2015 (mainly retrospective case-control pools) sit in the "Cohort studies" table |
| interpretive | text p. 115 (Wei 2004) | "generally showing no association". The database has colon ≥5/wk 1.43 (1.00–2.05) for Wei 2004 |
| model choice | Flood 2003, Chan AT 2005, Bernstein 2015 | IARC's value matches a non-main database row (a total-meat-adjusted model, the Table II category, and per serving instead of per 100 g respectively) |

All other values agree: 166 of 178 Table 2.2.1 triples match the database exactly.

### WCRF 1997 (card values only)
- No numeric discrepancy for the 2 cohorts with printed numbers that we hold (Willett 1990, Giovannucci 1994).
- Minor: Bostick 1994 "no increase" vs main dish 1.21 (0.75–1.96).
- Context the report does not give (per the card): both positive cohorts are age (+energy) adjusted only and cover colon only, and both come from the same group.
- Thun 1992: men 1.21 described as "no difference" (the report's own number).

## 4. Database issues noticed (not fixed; read-only)
- `butler2008`: no row has `is_main_model=yes` (butler2008_001 red meat HvL is `no`). studies.csv `cases_colorectal`=961; IARC text says 941. Not checked against the paper.
- `bernstein2015`: per-serving unprocessed estimates (Table 2/3) are all `is_main_model=no`; the main flag is on the per-100 g S5 rows.
- `singh1998`, `bostick1994`, `kato1997`: no DOI/PMID in studies.csv. Kato has DOI only; PMID 9343837 is in the IARC ref list.

## 5. Studies in either report that are missing from the database

Cohort / nested designs (red meat, CRC):
| Study | Report | DOI / PMID (source) |
|---|---|---|
| Phillips & Snowdon 1985, Adventist mortality | WCRF | J Natl Cancer Inst 1985;74:307–17 [Norat 2002 ref 96]; DOI/PMID not on disk. PDF on Drive `NED/14b_PhillipsSnowdon1985.pdf` (per fase0) |
| Thun 1992, CPS-II mortality | WCRF | J Natl Cancer Inst 1992;84:1491–500 [Norat 2002 ref 98]; no DOI/PMID on disk |
| Knekt 1994, Finland fried meat | WCRF | doi:10.1002/ijc.2910590608; PMID 7989114 [IARC line 16275] |
| Berndt 2006, CLUE II | IARC T2.2.1 | doi:10.1158/1055-9965.EPI-06-0449; PMID 17119055 [IARC line 11447] |
| Sørensen 2008, DCH | IARC T2.2.1 | doi:10.1016/j.canlet.2008.02.046; PMID 18372103 [IARC line 11876] |
| Andersen 2009, DCH | IARC T2.2.1 | doi:10.1186/1471-2407-9-407; PMID 19930591 [IARC line 11469] |
| Figueiredo 2014, GECCO/CCFR | IARC T2.2.1 | doi:10.1371/journal.pgen.1004228; PMID 24743840 [IARC line 11579] |
| Ananthakrishnan 2015, GECCO/CCFR | IARC T2.2.1 | doi:10.1158/1055-9965.EPI-14-0897; PMID 25342387 [IARC line 11460] |
| Kantor 2014, GECCO/CCFR | IARC text | doi:10.1158/1055-9965.EPI-14-0062; PMID 24994789 [IARC line 11660] |
| Hansen 2007, DCH (XPC) | IARC text | doi:10.1016/j.mrfmmm.2007.02.002; PMID 17363013 [IARC line 11611] |
| Wei 2009, NHS | IARC text | doi:10.1093/aje/kwp210; PMID 19723749 [IARC line 11976] |
| Fung 2010, NHS/HPFS | IARC text | doi:10.3945/ajcn.2010.29242; PMID 21097651 [IARC line 11603] |
| Zhang 2011, NHS/HPFS (haem iron) | IARC text | doi:10.1007/s10552-011-9839-z; PMID 21909950 [IARC line 11968] |
| Gaard 1996, Norway | IARC text | Eur J Cancer Prev 1996;5:445–54; PMID 9061275 [IARC line 11608] |
| Lin 2004, WHS | IARC text | doi:10.1093/aje/kwh319; PMID 15522858 [IARC line 11731] |
| Lüchtenborg 2005, NLCS | IARC text | doi:10.1007/s10552-005-0239-0; PMID 16184469 [IARC line 11758] |
| Sato 2006, Miyagi | IARC text | doi:10.1097/01.cej.0000197455.87356.05; PMID 16679863 [IARC line 5318] |
| Egeberg 2013, DCH (mentioned in 2.2.3 WG note) | IARC text | doi:10.3945/jn.112.168799; PMID 23427329 [IARC line 4982] |
| Cross 2007 / Ruder 2011, NIH-AARP | IARC text (combined section) | Cross 2007 doi:10.1371/journal.pmed.0040325, PMID 18076279; Ruder 2011 doi:10.3945/ajcn.111.020701, PMID 22071715 |

Case-control: Manousos 1983 and Miller 1983 (WCRF; texts on disk under `HT/19_…` and `HT/12b_…`). The other 24 WCRF case-control studies are unidentified. The 55 IARC Table 2.2.3 rows are listed in `kontroll/fase0_inventar.md` §6b; none are in the database.

Already in the database (previously listed as missing in fase0 §6a): kato1997, chen1998, flood2003, kabat2007, sellers1998, willett1990, giovannucci1994, goldbohm1994, brink2005, nothlings2009.