# synthesis_extraction.csv — notes

Built 30.10.2026 from the printed text of each synthesis. 161 rows: norat2002 17, larsson2006 50, wcrf2007 27, cup2010 67. Red meat only; processed meat and red-and-processed rows are not extracted, except where a synthesis uses such a row inside its red-meat analysis (flagged). `in_db=no` rows also carry `not_in_db` in extraction_notes. Wei 2004 rows carry cohort `nhs` or `hpfs` (db study wei2004 = nhs_hpfs).

Transcription check (DerSimonian–Laird on the CSV rows, not a reanalysis):
- CUP 2010 DR CRC 8 rows: 1.17 (1.05–1.31), as printed. Colon 9 rows: 1.12 (0.97–1.29), as printed. Rectum 7 rows: 1.18 (0.98–1.43); printed 1.18 (0.98–1.42).
- CUP 2010 HvL CRC 12 rows: 1.10 (1.00–1.21). Colon 12 rows: 1.18 (1.04–1.35). Rectum 7 rows: 1.14 (0.83–1.56). All as printed in Chan T1.
- Larsson 2006 HvL, 15 main-analysis rows: 1.28 (1.15–1.42), as printed.
- WCRF 2007 per time/week, 9 rows: 1.43 (1.05–1.94), as printed.

---

## 1. Norat 2002 (Int J Cancer 98:241–256), `27_Norat2002.txt`

**Published pooled estimates, red meat (case-control and cohort studies together unless stated)**
- Table I (p244), highest vs lowest: all 1.35 (1.21–1.51) n=23; case-control 1.36 (1.17–1.59) n=14; **cohort 1.27 (1.11–1.45) n=9, p het 0.45**; colon 1.32 (1.18–1.48) n=19; rectum 1.36 (1.17–1.57) n=7; males 1.40 (1.20–1.64) n=9; females 1.13 (0.85–1.50) n=8; Europe 1.46 (1.22–1.75) n=7; USA 1.30 (1.12–1.52) n=13; fresh only 1.28 (1.11–1.47) n=13; fresh+processed 1.49 (1.26–1.77) n=11.
- Table II (p244), per 120 g/day vs none: all 1.24 (1.08–1.41) n=17; case-control 1.26 (1.02–1.55) n=8; **cohort 1.22 (1.05–1.41) n=9, p het 0.17**; colon 1.23 (1.04–1.46) n=14; rectum 1.64 (0.64–4.21) n=2; males 1.36 (1.18–1.55) n=9; females 1.11 (0.78–1.56) n=8; Europe 1.56 (1.07–2.26) n=5; USA 1.22 (1.05–1.41) n=10; fresh only 1.19 (0.91–1.55) n=8; fresh+processed 1.28 (1.11–1.48) n=9.
- Text p244–245: North American cohorts HvL 1.45 (1.20–1.76); American fresh-only DR 1.05 (0.55–2.00); American fresh+processed DR 1.24 (1.07–1.43); European fresh-only DR 1.41 (0.91–2.20); European fresh+processed DR 2.07 (1.25–3.42). Begg p=0.52 (DR). Inter-quantile mean range 116 g/day (3 cohorts reporting g/day), 101 g/day (other cohorts, rescaled).
- Methods (p242): 120 g per portion of meat/red meat; GL with log-midpoint of category boundaries; open top category given width of the nearest interval; open bottom = 0; random effects "when there was evidence of heterogeneity".
- rm2010 p156 cites Norat 2002 as "fresh red meat per 120 g/day 1.19 (0.91–1.55), 8 studies" — this is the Table II all-designs fresh-only row, not a cohort-only estimate.

**Study list.** 14 cohort studies identified (refs 19, 40, 42, 44, 48, 50, 57, 96–102; Appendix III pp255–256): Phillips 1985, Hirayama 1990, Thun 1992, Willett 1990, Giovannucci 1994, Goldbohm 1994, Bostick 1994, Gaard 1996, Kato 1997, Chen 1998, Hsing 1998, Singh 1998, Knekt 1999, Pietinen 1999. Text: 9 USA (2 Adventist), 4 European, 1 Japanese.
- **Not named:** the 9 cohort studies in the red-meat HvL (Table I) and the 9 in the red-meat DR (Table II). Per-study estimates are only in Fig 2 (HvL, image) and Fig 4 (DR curve, image): **figure only**, not readable.
- Appendix III cohorts with a red-meat line and printed CIs: Willett, Giovannucci, Goldbohm (labelled "Fresh red meat and poultry"), Bostick, Kato, Chen, Hsing, Singh, Pietinen (9). Thun has a red-meat label with no CIs. Phillips, Hirayama and Gaard print total meat only; Knekt prints cured meat only. Which 9 were used is not stated in the text.
- The CSV codes the Appendix III highest-vs-lowest values (primary-study categories as Norat transcribed them), with `rr` = top category vs reference. These are not necessarily the values plotted in Fig 2.

**Could not be read:** Figs 1–4 (images); per-study weights; per-study DR slopes; which cohorts are in each n=9.

**Internal inconsistencies**
- Red meat DR: text p244 says 18 studies (9 case-control + 9 cohort); Table II gives all n=17 and case-control n=8.
- Total meat, fresh+processed HvL: Table I 1.16 (1.08–1.34); text p243 1.16 (1.01–1.34).
- Table II USA red meat 1.22 (1.05–1.41) n=10 is identical to the cohort row (n=9).
- Appendix III printing: Willett 84–105 row "0.25 (0.73–2.13)"; Giovannucci RR column offset by one row; Goldbohm both-genders CI "0.84 (1.51–1.37)"; Chen lowest category "≤5" (servings/day); Singh total-meat block lists "≥1 time/wk" twice; Hirayama rows misaligned.

## 2. Larsson & Wolk 2006 (Int J Cancer 119:2657–2664), `31_Larsson2006.txt`

**Published pooled estimates, red meat**
- HvL, 15 studies, 7,367 cases (p2660): **1.28 (1.15–1.42)**, Q=4.86, p=0.99, I²=0%. Excluding the 2 nested case-control studies: 1.28 (1.15–1.42). Excluding the 2 mortality studies [7, 20]: 1.27 (1.14–1.42). Restricted to studies adjusting for PA, BMI, smoking and diet [4, 8, 9, 23, 24]: 1.29 (1.09–1.53). Egger p=0.42.
- Table II (p2661), HvL subgroups: colon 1.21 (1.05–1.40) n=9; proximal 1.15 (0.91–1.47) n=3; distal 1.31 (0.62–2.79) n=3, I² 83.8%; rectum 1.56 (1.25–1.95) n=7; men 1.26 (1.02–1.54) n=5; women 1.16 (1.01–1.34) n=6; Europe 1.32 (1.12–1.57) n=5; USA 1.23 (1.06–1.42) n=9; total red meat 1.24 (1.09–1.42) n=8; fresh only 1.22 (1.08–1.37) n=9 (plus the adjustment strata in the table).
- Table III (p2662), per 120 g/day: **all 1.28 (1.18–1.39) n=14**, Q 8.82, p 0.79, I² 0%; colon 1.24 (1.12–1.38) n=10; rectum 1.63 (1.24–2.14) n=7, I² 50.3%; Europe 1.33 (1.17–1.50) n=5; USA 1.24 (1.10–1.39) n=8.
- Methods: DL random effects. CRC from site-specific or sex-specific RRs pooled within study by inverse variance (fn6). Total red meat chosen over fresh when both are given (Table I fn4). GL (or VWLS without case counts), 120 g/portion red meat, median/mean or midpoint, open top category given the width of the preceding one.

**Study list.** HvL (Table I, Fig 1): Bostick 1994, Kato 1997, Chen 1998, Hsing 1998, Singh 1998, Pietinen 1999, Järvinen 2001, Tiemersma 2002, Flood 2003, Wei 2004 NHS, Wei 2004 HPFS, English 2004, Larsson 2005, Chao 2005, Norat 2005 (15; case sum 7,367 matches the text).
- DR (p2662): refs 4 (Wei, ×2 cohorts), 5, 8–13, 18–20, 23, 24 = 14 studies. Kato [21] excluded (not quantifiable).
- In Table I but red meat "NA" (processed meat only, not in the CSV): Gaard 1996, Knekt 1999, Kojima 2004 (JACC, fatal), Lüchtenborg 2005 (NLCS case-cohort).
- Superseded and excluded: refs 14–17 (Goldbohm 1994, Thun 1992, Willett 1990, Giovannucci 1994). Sellers 1998 excluded in favour of Bostick 1994.

**Could not be read / not printed:** Fig 1 (HvL forest, image) and therefore per-study weights. **Per-study per-120 g estimates are not printed anywhere.** Table I prints HvL only, despite the task brief saying it lists per 120 g. The CSV DR rows therefore have `rr` empty. The colon (n=10) and rectum (n=7) DR study lists are not given. Contrasts (category bounds) are not printed: Table I fn3 says only "highest vs lowest".

**Internal inconsistencies**
- The mortality-exclusion sensitivity cites refs 7 and 20 (Kojima, Hsing), but Kojima has red meat "NA" in Table I.
- Table II fn4 classifies refs 4, 10, 13, 24 as "fresh only" and 5, 8, 18, 23 as "both". Counting Wei as 2 studies gives fresh n=9 and total n=8, consistent. But Table I fn4 says total red meat was chosen when provided, so the main-analysis rows for refs 5, 8, 18 and 23 are total red meat (including processed).
- Table II "Other" location n=1 for red meat, but fn3 names Australia [10] and Japan [7], and Japan has no red-meat estimate.
- Pietinen's red-meat RR has no site label.
- Flood 2003 has 487 CRC here and 311 CRC in Chan 2011 S1.

## 3. WCRF/AICR 2007 Second Expert Report, `32_WCRF2007.txt` (section 4.3.5.1.1, pp120–121); rm2010 pp156–157, 165

**Published pooled estimates (cohort studies; colorectal or, if not available, colon)**
- Per time/week, 7 studies: **1.43 (1.05–1.94)**, "moderate heterogeneity" (Fig 4.3.2 p121; text p121).
- Per 100 g/day, 3 studies: **1.29 (1.05–1.59)** in Fig 4.3.3 and **1.29 (1.04–1.60)** in the text p121. rm2010 p156 repeats 1.29 (1.04–1.60) but calls the other summary "per time/day". rm2010 Table 36 p157 prints **1.29 (0.94–1.78)** per 100 g/day, I²=32%, and per time/day 1.43 (1.05–1.94), I²=58%, p=0.01. Sex strata from Table 36 (2005 SLR, per time/day): men 1.79 (0.98–3.27), 3 studies, I² 47%, p 0.13; women 1.27 (0.74–2.19), 3 studies, I² 72%, p 0.03.
- HvL (Fig 4.3.1 p120) is plotted only; no summary. Case-control studies were "not summarised". Larsson 2006's 1.28 (1.18–1.39) per 120 g is cited as support.

**Study list.** Text: "Sixteen cohort studies [8–24]" (17 reference numbers). Refs: 8 Bostick 1994, 9 Willett 1990, 10 Giovannucci 1994, 11 Singh 1998, 12 Fraser 1999, 13 Järvinen 2001, 14 English 2004, 15 Norat 2005, 16 Kato 1997, 17 Pietinen 1999, 18 Larsson 2005, 19 Wei 2004, 20 Chen 1998, 21 Tiemersma 2002, 22 Feskanich 2004, 23 Phillips 1975, 24 Sellers 1998. Six cohorts published after the SLR (refs 27–32: Chao 2005, Lin 2004, Chan 2005 (NAT2), Brink 2005, Sato 2006, Oba 2006) did not change the judgement.
- Fig 4.3.1 HvL (13 rows, 11 studies): Philips 1975, Slattery 2001 W, Kato 1997 W, Singh 1998, Pietinen 1999 M, Järvinen 2001, Tiemersma 2002 M/W, Wei 2003 M/W, Chen 2003, English 2004, Norat 2004.
- Fig 4.3.2 per time/week (9 rows, 7 studies): Willett 1990 W, Bostick 1994 W, Giovannucci 1994 M, Singh 1998, Chen 1998 M NAT rapid/slow, Tiemersma 2002 M/W, English 2004.
- Fig 4.3.3 per 100 g/day (3): Pietinen 1999, Norat 2005, Larsson 2005 W.
- Fig 4.3.4 dose-response graph (no values): Kinlen 1983 M/W, Giovannucci 1994 M, Bostick 1994, English 2004.
- rm2010 Table 37 "SLR" column (2005 SLR use): Bostick Dose & HvL; Willett Dose; Giovannucci Dose; Singh Dose & HvL; Pietinen Dose & HvL; Kato HvL; Järvinen HvL; Tiemersma M and F Dose & HvL; English Dose & HvL; Norat Dose & HvL; Larsson Dose; Wei 2004 HvL; Feskanich 2004 and Fraser 1999 "reviewed in text"; Sellers 1998, Khan 2004, Chan 2005 not used. rm2010 p165: Phillips 1975 and Chen 2003 were in the SLR and dropped in 2010. rm2010 Table 35 footnote: Phillips 1975 is called a case-control study.

**Could not be read / not available:** the 2006 SLR document itself (protocol §7). Per-study weights and case numbers (not printed in Figs 4.3.1–4.3.3). All Fig 4.3.4 values. The upper CI of Larsson 2005 in Fig 4.3.3.

**Internal inconsistencies**
- "Sixteen cohort studies" is followed by 17 reference numbers.
- The text says significant in "four", citing refs 9, 10, 12, 18, 23 (five refs), "one of these was specific to rapid-acetylator genotypes". The rapid-acetylator result is Chen 1998 (ref 20), which is not in that citation list. In Fig 4.3.1 three rows have CIs excluding 1 (Philips, Tiemersma M, English).
- "All of the cohort studies that reported [HvL] showed increased risk", yet rm2010 Table 35 (2005 SLR) says 9 of 11 showed increased risk and 1 showed no association. That same Table 35 cell says "processed meat comparison" in the red-meat row.
- Figure labels differ from the reference list: Wei 2003 (ref 2004), Norat 2004 (ref 2005), Chen 2003 (ref 20 is Chen 1998). Slattery 2001 and Kinlen 1983 have no reference number in ch 4.3. The "Chen 2003" value 1.48 (0.85–2.59) equals Chan S2's Chen 2003 Jiashan pork value 1.481 (0.846–2.593).
- Bostick: Table 37 says "Dose & HvL", but Bostick is not in Fig 4.3.1. Larsson 2005 and Willett/Giovannucci are also absent from Fig 4.3.1.
- Fig 4.3.3: Larsson 2005 CI is printed "1.05–0.95". The summary CI differs between the figure, the text and rm2010 Table 36 (three versions, listed above).
- Järvinen upper CI is 2.93 here and 2.94 in Larsson 2006 and SLR 2010.
- Norat 2005 per 100 g is the calibrated value 1.49 (0.91–2.43). The observed value 1.21 (1.02–1.43) is used in CUP 2010 (rm2010 p156).

## 4. CUP 2010 (SLR 2010 section 2.5.1.3, `rm2010.txt` = `slr2010.txt` pp155–177) and Chan 2011 (PLoS ONE 6:e20456)

**Published pooled estimates, red meat**
- SLR 2010 / Chan T1 (p4), per 100 g/day: **CRC 1.17 (1.05–1.31)**, 8 studies, I² 0%, p 0.48, 4,314 cases. **Colon: SLR 1.12 (0.97–1.29), 9 studies, I² 0%, p 0.89, 3,172 cases. Chan 1.17 (1.02–1.33), 10 studies, I² 0%, p 0.65, 3,561 cases.** Rectum 1.18 (0.98–1.42), 7 studies, I² 0%, p 0.67, 1,477 cases (both sources).
- Sensitivity without Järvinen, Lee and Nöthlings (SLR p155): CRC 1.21 (1.06–1.37); colon 1.12 (0.97–1.31); rectum 1.20 (0.99–1.47).
- Sex (Chan T1; SLR Figs 71, 75, 79): men CRC 1.28 (0.49–3.35), 2 studies; women CRC 1.05 (0.78–1.42), 3 studies; men colon 1.06 (0.75–1.50), 2; **women colon: Chan 1.16 (0.84–1.61), 5 studies; SLR 1.00 (0.72–1.38), 4 studies**; women rectum 0.90 (0.60–1.35), 3; men rectum 0.95 (0.47–1.89), 1.
- Region (Chan T1): Europe CRC 1.23 (1.08–1.40) n=5, colon 1.29 (1.08–1.54) n=3 (1,307 cases), rectum 1.20 (0.95–1.50) n=3. North America CRC n=1, colon 1.11 (0.86–1.44) n=4 (1,476), rectum 0.93 (0.54–1.60) n=2. Asia-Pacific CRC 1.01 (0.69–1.48) n=2, colon 0.94 (0.69–1.27) n=3 (732), rectum 1.16 (0.57–2.39) n=2.
- HvL (Chan T1): CRC **1.10 (1.00–1.21) n=12**, I² 22%; colon 1.18 (1.04–1.35) n=10; proximal 1.13 (0.83–1.54) n=2; distal 1.57 (0.98–2.49) n=2; rectum 1.14 (0.83–1.56) n=7.
- Methods: 120 g/serving; g/1000 kcal converted using mean or median energy intake; per 100 g/day; random effects.

**Study list.** 17 cohorts / 24 publications (SLR p155; Table 37 pp159–165). DR CRC: Lee 2009, Nöthlings 2009, Larsson 2005, Norat 2005, English 2004, Tiemersma 2002, Järvinen 2001, Pietinen 1999. DR colon (SLR): Lee, Oba 2006, Norat, English, Wei HPFS, Wei NHS, Järvinen, Singh 1998, Bostick 1994. DR rectum: Lee, Larsson, Norat, English, Wei HPFS, Wei NHS, Järvinen. HvL only: Cross 2010, Butler 2008, Chan 2005, Kato 1997.
- Chan's 16 HvL studies [18, 20, 21, 25–27, 34, 50, 52, 53, 57, 59–63]; DR exclusions [20, 21, 59, 62] (Butler, Cross 2010, Chan 2005, Kato).
- Excluded (Table 37 / S2): Willett 1990 and Giovannucci 1994 (superseded by Wei 2004); Sellers 1998 (Bostick chosen); Fraser 1999 (Singh chosen); Feskanich 2004; Khan 2004 (mortality); Wu 2004; Sorensen 2008 (EPIC component). Chen 1998 was moved to red and processed meat. Phillips 1975 and Chen 2003 were dropped.
- Chan S1 puts Willett, Giovannucci, Chen 1998, Flood 2003, Lin 2004, Chao 2005, Berndt 2006, Cross 2007, Kabat 2007, Cross 2010 and Fung 2010 under "red and processed meat" (Chan Fig 2), not red meat.

**Chan vs SLR differences (separate Chan2011 rows in the CSV)**
- Colon DR: Chan has 10 studies / 3,561 cases, SLR 9 / 3,172. The difference of 389 cases equals Larsson 2005 colon (389). Europe colon 1,307 = 855 + 63 + 389. **The Larsson colon per-100 g value is in Chan Fig 4 only (image), so it is not readable.** Chan S2, however, says only the CRC and rectal results of Larsson 2005 were used.
- English 2004: S1 prints per 1 time/week (CRC 1.03, CC 1.00, RC 1.08). The SLR uses per-100 g values (1.19 / 1.00 / 1.57).
- Proximal/distal HvL (Chan T1 n=2, refs [34, 54]): Larsson 2005 (1.10; 1.99) and Chao 2005 (1.27; 0.71). S1 lists Chao under red and processed meat.
- Chan's per-study DR values in Fig 4 are an image throughout, so it cannot be confirmed that the other values equal SLR Figs 70/74/78. Where S1 prints a value (Norat 1.21 / 1.20 / 1.23), it matches.

**Internal inconsistencies / arithmetic**
- CRC DR cases: the sum of the 8 Fig 70 studies (Table 34/S1 counts) is 4,313; published 4,314.
- Colon DR cases: the SLR's 9 rows sum to 3,126 using Järvinen 63; published 3,172 (= 3,126 − 63 + 109). Chan's regional sums give 1,307 + 1,476 + 732 = 3,515; published 3,561 (same 46 gap).
- Rectum DR cases: the 7 rows sum to 1,415; published 1,477.
- The HvL colon SLR plot (Fig 77) has 12 rows: Larsson proximal and distal are both entered as colon rows, and Oba is entered as M and F. Chan T1 says n=10. Pooling the 12 rows reproduces 1.18 (1.04–1.35).
- Larsson HvL contrast is ">=4 vs <2 servings/wk" (SLR Fig 73) and "5.5 vs. 1.5 servings/week" (S1).
- Chan S1 contrasts that are internally inconsistent with the SLR: Nöthlings "≥26.0 vs <10.4" (S1) vs "26 vs 0–10.3" (SLR); Järvinen (S1) "Men >206 vs <94; Women 134 vs <61".
- Parsed S2 text loses the point estimates for Butler 2008 (CI 0.82–1.26), Sorensen 2008 (0.97–1.09) and Spencer 2010 (0.84–1.22).
- SLR p156 "Possible reasons for differences" calls the 2007 time-based summary "per time/day". WCRF 2007 prints "per time/week".