# synthesis_extraction.csv — figure pass (30.10.2026)

Task: fill the figure-only gaps by reading the numbers printed next to the forest plots on rendered PDF pages. No values were estimated from box or line positions.

Files:
- Backup of the pre-pass CSV: `database/synthesis_extraction_v1.csv` (161 rows). Updated CSV: `database/synthesis_extraction.csv` (170 rows; cup2010 has 67 → 76).
- Rendered pages (pdftoppm -r 220) and the native embedded figure images (pdfimages): `/home/claude/fig_in/` (chan_p-07, norat_p-03/05/06, norat_img-001..004, lars_p-4/6, lars_img-001/002).
- Scripts: `kontroll/scripts/update_synth_figs.py` (CSV edits), `kontroll/scripts/pool_synth.py` (DerSimonian–Laird).

## 1. Chan 2011, PLoS ONE 6:e20456 — Fig 4 p. 7 (red meat, per 100 g/day, RE weights printed)

The task brief said Figs 1–4 print per-study RR and weights. That is not quite right:
- Fig 1 is the flow diagram.
- Fig 2 is **red AND processed** meat (outside this CSV's scope).
- Fig 3 is the non-linear curve.
- Fig 4 is the red-meat forest plot, the only one relevant here.
- Fig 5 is processed meat.

Chan does not plot the red-meat HvL analyses.

| study | outcome | analysis | increment | rr | ci_low | ci_high | weight % | figure+page |
|---|---|---|---|---|---|---|---|---|
| Pietinen 1999 Male | colorectal | DR | per 100 g | 0.83 | 0.43 | 1.61 | 2.92 | Fig 4 p7 |
| Jarvinen 2001 Mixed | colorectal | DR | per 100 g | 1.37 | 0.92 | 2.06 | 7.74 | Fig 4 p7 |
| Tiemersma 2002 Mixed | colorectal | DR | per 100 g | 1.69 | 0.88 | 3.23 | 2.98 | Fig 4 p7 |
| English 2004 Mixed | colorectal | DR | per 100 g | 1.19 | 0.89 | 1.58 | 15.79 | Fig 4 p7 |
| Larsson 2005 Female | colorectal | DR | per 100 g | 1.23 | 0.90 | 1.67 | 13.11 | Fig 4 p7 |
| Norat 2005 Mixed | colorectal | DR | per 100 g | 1.21 | 1.02 | 1.43 | 44.43 | Fig 4 p7 |
| Lee 2009 Female | colorectal | DR | per 100 g | 0.80 | 0.52 | 1.23 | 6.78 | Fig 4 p7 |
| Nothlings 2009 Mixed | colorectal | DR | per 100 g | 1.00 | 0.64 | 1.57 | 6.25 | Fig 4 p7 |
| *Subtotal (I² 0.0%, p 0.483)* | colorectal | | | 1.17 | 1.05 | 1.31 | 100 | Fig 4 p7 |
| Bostick 1994 Female | colon | DR | per 100 g | 0.94 | 0.43 | 2.04 | 2.95 | Fig 4 p7 |
| Singh 1998 Mixed | colon | DR | per 100 g | 2.49 | 0.46 | 13.39 | 0.63 | Fig 4 p7 |
| Jarvinen 2001 Mixed | colon | DR | per 100 g | 1.30 | 0.77 | 2.17 | 6.70 | Fig 4 p7 |
| English 2004 Mixed | colon | DR | per 100 g | 1.00 | 0.69 | 1.46 | 12.51 | Fig 4 p7 |
| Wei 2004 Female (NHS) | colon | DR | per 100 g | 1.19 | 0.75 | 1.88 | 8.47 | Fig 4 p7 |
| Wei 2004 Male (HPFS) | colon | DR | per 100 g | 1.07 | 0.74 | 1.52 | 13.95 | Fig 4 p7 |
| **Larsson 2005 Female** | colon | DR | per 100 g | **1.68** | **1.11** | **2.55** | **10.33** | Fig 4 p7 |
| Norat 2005 Mixed | colon | DR | per 100 g | 1.20 | 0.96 | 1.48 | 38.12 | Fig 4 p7 |
| Oba 2006 Mixed | colon | DR | per 100 g | 0.74 | 0.28 | 1.96 | 1.91 | Fig 4 p7 |
| Lee 2009 Female | colon | DR | per 100 g | 0.85 | 0.45 | 1.61 | 4.43 | Fig 4 p7 |
| *Subtotal (I² 0.0%, p 0.646)* | colon | | | 1.17 | 1.02 | 1.33 | 100 | Fig 4 p7 |
| Jarvinen 2001 Mixed | rectum | DR | per 100 g | 1.32 | 0.69 | 2.53 | 8.24 | Fig 4 p7 |
| English 2004 Mixed | rectum | DR | per 100 g | 1.57 | 0.99 | 2.49 | 16.35 | Fig 4 p7 |
| Wei 2004 Female (NHS) | rectum | DR | per 100 g | 0.91 | 0.39 | 2.15 | 4.72 | Fig 4 p7 |
| Wei 2004 Male (HPFS) | rectum | DR | per 100 g | 0.95 | 0.47 | 1.89 | 7.30 | Fig 4 p7 |
| Larsson 2005 Female | rectum | DR | per 100 g | 0.98 | 0.55 | 1.72 | 10.87 | Fig 4 p7 |
| Norat 2005 Mixed | rectum | DR | per 100 g | 1.23 | 0.94 | 1.62 | 47.17 | Fig 4 p7 |
| Lee 2009 Female | rectum | DR | per 100 g | 0.74 | 0.33 | 1.66 | 5.34 | Fig 4 p7 |
| *Subtotal (I² 0.0%, p 0.666)* | rectum | | | 1.18 | 0.98 | 1.42 | 100 | Fig 4 p7 |

Findings:
- **CRC and rectum:** every RR, CI and weight is identical to SLR 2010 Figs 70 and 78. The CSV rows were already complete; a confirmation note has been added to them.
- **Colon:** Chan's 10-row set is the SLR's 9 rows plus **Larsson 2005 colon, 1.68 (1.11–2.55), weight 10.33%**.
  - This confirms the earlier inference from the 389-case difference.
  - It contradicts Chan S2, which says only the CRC and rectal results of Larsson 2005 were used.
  - The other 9 rows have the same RR (CI) as SLR Fig 74. Their weights are uniformly smaller (×≈0.90) because Larsson 2005 is added.
- **CSV changes:**
  - The empty `cup2010 / larsson2005 / colon / DR` row has been filled.
  - 9 rows were added for the rest of Chan's colon set (`source_page = Chan Fig 4 p7`, notes "from Fig 4 p. 7 (rendered); Chan2011 colon DR set").
  - These duplicate the SLR Fig 74 rows in RR/CI. Filter on `source_page` to choose the SLR set or the Chan set.

## 2. Larsson & Wolk 2006, Int J Cancer 119:2657 — Fig 1 p. 2660 (red-meat HvL)

Fig 1 prints RR (95% CI) for each study but **no weights** (box size only). Fig 2 (p. 2662) is processed meat. There is no dose-response figure.

| study | outcome | analysis | rr | ci_low | ci_high | weight % | figure+page |
|---|---|---|---|---|---|---|---|
| Bostick 1994 | colon | HvL | 1.04 | 0.62 | 1.76 | not printed | Fig 1 p2660 |
| Kato 1997 | colorectal | HvL | 1.23 | 0.68 | 2.22 | not printed | Fig 1 p2660 |
| Chen 1998 | colorectal | HvL | 1.17 | 0.68 | 2.02 | not printed | Fig 1 p2660 |
| Hsing 1998 | colorectal | HvL | 1.90 | 0.90 | 4.30 | not printed | Fig 1 p2660 |
| Singh 1998 | colon | HvL | 1.41 | 0.90 | 2.21 | not printed | Fig 1 p2660 |
| Pietinen 1999 | colorectal | HvL | 1.10 | 0.70 | 1.70 | not printed | Fig 1 p2660 |
| Järvinen 2001 | colorectal | HvL | 1.50 | 0.77 | 2.94 | not printed | Fig 1 p2660 |
| Tiemersma 2002 | colorectal | HvL | 1.60 | 0.90 | 2.90 | not printed | Fig 1 p2660 |
| Flood 2003 | colorectal | HvL | 1.10 | 0.83 | 1.45 | not printed | Fig 1 p2660 |
| Wei 2004* (NHS) | colorectal | HvL | 1.21 | 0.72 | 2.03 | not printed | Fig 1 p2660 |
| Wei 2004** (HPFS) | colorectal | HvL | 1.24 | 0.78 | 1.96 | not printed | Fig 1 p2660 |
| English 2004 | colorectal | HvL | 1.40 | 1.00 | 1.90 | not printed | Fig 1 p2660 |
| Larsson 2005 | colorectal | HvL | 1.32 | 1.03 | 1.68 | not printed | Fig 1 p2660 |
| Chao 2005 | colorectal | HvL | 1.36 | 0.93 | 2.00 | not printed | Fig 1 p2660 |
| Norat 2005 | colorectal | HvL | 1.35 | 0.96 | 1.88 | not printed | Fig 1 p2660 |
| *Summary* | | | 1.28 | 1.15 | 1.42 | Q 4.86, p 0.99, I² 0% | Fig 1 p2660 |

- All 15 values are identical to Table I, which the CSV already holds. The notes on those rows have been updated.
- The figure has a typo, "Järvinen et al., 200¹²": the year is truncated.
- **Remaining gap:** the per-study per-120 g dose-response values (14 rows) are printed nowhere, neither in a figure nor in a table.

## 3. Norat 2002, Int J Cancer 98:241 — Fig 2 p. 245 (red-meat HvL), Fig 4 p. 246 (dose-response)

**Fig 2 prints no per-study numbers.** It shows lines and boxes only, plus the printed summaries:
- cohort 1.27 (1.11–1.45)
- case-control 1.36 (1.17–1.59)
- all 1.35 (1.21–1.51)

Values cannot be read without estimating them from positions, which this pass does not do.

It does **name the 9 cohorts** in the red-meat HvL. The text did not name them, so this is new information:

Willett, USA, 90 · Chen, USA, 92 · Thun, USA, 92 · Giovannucci, USA, 94 · Bostick, USA, 94 · Kato, USA, 97 · Singh, USA, 98 · Hsing, USA, 98 · Pietinen, Finland, 99

- "Chen, USA, 92" is presumably Chen 1998 (ref 57, Physicians' Health Study). The year label looks wrong; this is an inference.
- **Not** in Fig 2 red meat: Phillips 1985, Hirayama 1990, Goldbohm 1994, Gaard 1996, Knekt 1999. Goldbohm and Knekt appear in Fig 3 (processed meat) instead.
- Thun is plotted with a CI although Appendix III prints none, so Norat must have derived one. It is not printed.

Fig 4 (p. 246) shows only the pooled dose-response lines (processed, red all, meat all, red fresh, meat fresh). It has no per-study values.

Fig 1 (p. 243, total meat) and Fig 3 (processed meat) likewise print only summaries:
- Fig 1: cohort 1.03 (0.81–1.32), case-control 1.18 (0.99–1.40), all 1.14 (0.99–1.31)
- Fig 3: cohort 1.39 (1.09–1.76), case-control 1.29 (1.09–1.52), all 1.31 (1.13–1.51)

**CSV changes:**
- Each norat2002 HvL row now states whether it is IN or NOT IN Fig 2.
- The old "figure not readable / 9 cohorts not named" wording has been replaced.
- No rr/ci values were changed, because none are printed.

## 4. WCRF/AICR 2007 Second Expert Report — Fig 4.3.4 (and Fig 4.3.3)

**No PDF was found.**
- Searched in Drive: the root folder [Drive-id], Arbeidsfiler/wcrf_kriterier, and title searches for Second / Expert / SER / WCRF.
- Only `32_WCRF2007.txt` and `32_WCRF2007.md` exist; WCRF_AICR_1999 is the first report.

So Fig 4.3.4 (a dose-response graph that, per the text extraction, prints no values) could not be rendered. The garbled Larsson 2005 upper CI in Fig 4.3.3 ("1.05–0.95") could not be checked either. Notes saying this have been added to the kinlen1983 rows and the Fig 4.3.3 rows.

## 5. DerSimonian–Laird re-pooling (kontroll/scripts/pool_synth.py)

| pooled estimate | k | published | reproduced (DL) | max abs diff, printed vs DL weight (pp) |
|---|---|---|---|---|
| CUP/Chan DR per 100 g, CRC | 8 | 1.17 (1.05–1.31), I² 0% | 1.17 (1.05–1.31), I² 0% | 0.39 |
| SLR 2010 DR per 100 g, colon (9) | 9 | 1.12 (0.97–1.29), I² 0% | 1.12 (0.97–1.29), I² 0% | 0.24 |
| **Chan 2011 DR per 100 g, colon (10, incl. Larsson 2005)** | 10 | 1.17 (1.02–1.33), I² 0% | 1.17 (1.02–1.34), I² 0% | 0.22 |
| CUP/Chan DR per 100 g, rectum | 7 | 1.18 (0.98–1.42), I² 0% | 1.18 (0.98–1.43), I² 0% | 0.12 |
| CUP/Chan HvL CRC | 12 | 1.10 (1.00–1.21), I² 22% | 1.10 (1.00–1.21), I² 22% | no weights printed |
| CUP/Chan HvL colon (12 plotted rows) | 12 | 1.18 (1.04–1.35) | 1.18 (1.04–1.35), I² 0% | no weights printed |
| CUP/Chan HvL rectum | 7 | 1.14 (0.83–1.56) | 1.14 (0.83–1.56), I² 38% | no weights printed |
| Larsson 2006 HvL CRC | 15 | 1.28 (1.15–1.42), Q 4.86, I² 0% | 1.28 (1.15–1.42), Q 4.86, I² 0% | no weights printed |
| WCRF 2007 per time/week | 9 | 1.43 (1.05–1.94); rm2010 T36 I² 58% | 1.43 (1.05–1.94), I² 58% | no weights printed |
| Norat 2002 red-meat cohort HvL | 8 of 9 | 1.27 (1.11–1.45) | **not reproducible**. Indicative only: Appendix III values for the 8 Fig 2 cohorts with CIs give 1.38 (1.16–1.64) | – |
| Larsson 2006 DR per 120 g | 14 | 1.28 (1.18–1.39) | not reproducible (no per-study values) | – |
| WCRF 2007 per 100 g/day | 3 | 1.29 (1.05–1.59) fig / (1.04–1.60) text / (0.94–1.78) rm2010 T36 | not reproducible (Larsson 2005 upper CI garbled; no PDF) | – |
| Norat 2002 DR per 120 g (cohort) | 9 | 1.22 (1.05–1.41) | not reproducible (no per-study values) | – |

- The differences of 0.01 in the upper CI for Chan colon and for rectum are rounding: the inputs are printed to 2 decimals.
- Printed weights match the recomputed DL weights to within 0.4 percentage points, which is consistent with rounded inputs.
- The Norat indicative pool (1.38 against the published 1.27) shows that Norat did not simply use the top-vs-bottom Appendix III values. It probably used different contrasts, its own derived CIs (Thun) and/or a fixed-effect model. That cannot be checked without per-study figure values.

## Remaining gaps

1. Norat 2002: the per-study RR/CI/weights behind Fig 2 (HvL) and the DR slopes behind Table II/Fig 4. Not printed anywhere.
2. Larsson 2006: per-study per-120 g DR values (not printed) and per-study HvL weights (not printed).
3. WCRF 2007: Fig 4.3.4 and the Fig 4.3.3 Larsson 2005 upper CI. No Second Expert Report PDF is available in Drive.
4. Chan 2011: the red-meat HvL forest plot is not in the paper (T1 summaries only), so the HvL weights remain unavailable. Fig 2 (red + processed) was not transcribed because it is outside the CSV's scope.