# K7 control – dankner2024 (Dankner et al., Int J Cancer 2024;155:2009–2020)

Full text: `drive/post2015/dankner2024.txt`. Table 4a is an image, so I rendered `pdf_in/dankner2024.pdf` page 7 (journal p. 2015) with pdftoppm at 220 dpi and transcribed it independently. Table 1 (p. 2011) was also rendered to check which cohort used which diet tool.
Rows checked: 43 (dankner2024_001–043). The 40 Table 4a rows were compared by script with my transcription: 0 mismatches in cohort, cases, participants, HR or CI.

## Discrepancies

| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| 043 | ci_low / ci_high (paper error) | 1.59, CI 1.02–1.78 (as printed) | Text p. 2014 prints this, but the CI is identical to the preceding gastrointestinal red-meat estimate, 1.35 (1.02–1.78), and is very asymmetric around 1.59 (upper/lower log ratio 0.25). | **critical for use; transcription correct.** Set verification_status = `disputed` and never use the CI (see below). |
| 043 | notes | "the subgroup may be FFQ" | My recomputation contradicts this. Pooling the beef HRs of the two 24-h recall cohorts (Table 1: Negev, MABAT) gives **1.59** (1.11–2.28); the five FFQ cohorts give 1.07 (0.51–2.23). | minor (note wording): the printed point estimate does belong to the 24-h recall subgroup. The subgroup label `diet_tool_24h_recall_as_printed` is therefore correct, and the note should be updated. |
| dankner2024 (studies) | n_participants / cases_colorectal | 27754 / 839 | The meat analysis (Table 4a) uses 26,889 participants and 813 cases. The codebook asks for the meat-analysis population. | minor (the notes give both) |
| dankner2024 (studies) | baseline_end | 2014 | The text says assessments ran "from 1963 until 2014", but Table 1's latest baseline year is 2005 (MABAT). | minor (paper inconsistency; the database follows the text) |
| 033–040 | exposure_includes_processed | `unclear` (poultry) | Total meat = red + processed + poultry (Table 3 footnote). Processed meat includes processed poultry, so "poultry" is probably unprocessed poultry. | minor: `unclear` is acceptable under the codebook; `no` is arguable. |

## Number check (Table 4a, rendered p. 2015)
For each of the five exposures (total meat, red meat and beef per 100 g; processed per 50 g; poultry per 100 g), the seven per-cohort rows and the random-effects row match exactly. Cohort-level cases/participants (122/10,936 NICCCS, 14/1090 HDS, 41/1694 MABAT, 53/1728 Ovary, 26/967 Negev, 20/617 GOH, 537/9863 IIHD) match on every row. As an independent check, my inverse-variance pooling of the seven beef rows gives 1.47 (1.06–2.03), against the printed 1.46 (1.06–2.02).

## Cohort ids
- Per-cohort ids `israel_nicccs`, `israel_hds`, `israel_mabat`, `israel_ovary`, `israel_negev`, `israel_goh` and `israel_iihd` each have parent `israel_hist` in cohorts.csv. The pooled rows (008, 016, 024, 032, 040) and the text rows 041–043 are `israel_hist`. Check rule 7 will therefore stop per-cohort rows and pooled rows from entering the same pool. Correct.
- Sex: IIHD rows are `men` (working men) and Ovary rows are `women` (controls of an ovarian cancer study). This is correct per Table 1. All other cohorts are `all`.
- IIHD rows use effect_measure = OR. This is correct: the IIHD model is logistic with imputed meat, even though the column is headed "HR".
- is_main_model = yes only for the pooled rows. This is reasonable: the paper's result is the meta-analysis, and the per-cohort rows use the same covariate model.

## The disputed row 043
- Printed (p. 2014): "59% … increased CRC … risk for each increment of 100 g/day beef … HR = 1.59 (95% CI: 1.02, 1.78) … for the 24-HR nutritional assessment type only (see Table S4)".
- The CI cannot belong to 1.59: the log distance above the estimate is 0.11 and below is 0.44. The same "(95% CI: 1.02, 1.78)" closes the previous sentence for GI red meat (FFQ), where it is symmetric around 1.35. This is a copy error of the same kind as the one in Chan 2005.
- The point estimate fits the 24-h subgroup exactly. Pooling Negev 1.39 (0.76–2.54) and MABAT 1.71 (1.09–2.69), with I² = 0 so RE = FE, gives 1.59 (1.11–2.28). This reconstruction comes from the rounded per-cohort CIs, so the true CI is approximately **1.1–2.3**. It can go in a note but must not replace the printed value.
- The Discussion (p. 2018) says "no association was found among studies using the 24-h recall with any type of meat". This contradicts the Results, where the 24-h beef estimate is positive and by any reconstruction excludes 1.
- Recommendation: keep the row as printed, set verification_status = `disputed`, and exclude it from all pools. The subgroup is 24-h recall, not FFQ. If n_cases is wanted for this row, it would be 67 (26 + 41).

## Rows 041–042
- Processed poultry 0.97 (0.75–1.27) (p. 2013) and processed non-poultry 0.95 (0.70–1.28) (p. 2014, "data not shown"). Confirmed. Both are coded `processed` with subgroup labels, which is good practice.

## Adjustments
m1 matches the Table 4a title and Methods: energy, sex, age (time axis), origin, education and smoking, with BMI and physical activity excluded from the pooled HRs. The flags are correct, and the per-cohort availability caveat is in the notes.

## Missing estimates
None in the main text. Table 4b (gastrointestinal) is not a colorectal outcome. Table S4 (by diet tool, which would give a correct CI for 043), Table S5a/b (unimputed IIHD) and Table S3 are not available. The text says the unimputed sensitivity analysis showed "the same results persisted" but gives no CRC numbers.

## Confirmed rows
001–042 are all confirmed. 043 is confirmed as printed, with the paper error described above.

## Internal inconsistencies in the paper
1. The 043 CI is a copy of the preceding GI estimate, and the Discussion contradicts the 24-h finding (above).
2. Red meat pooled: the text (p. 2013) gives "1.15, 95% CI: 0.97–1.52"; Table 4a and the abstract give 0.87–1.52. The database correctly uses the table.
3. Table 4a participants sum to 26,895, not the printed 26,889. Cases sum to 813 as printed.
4. Table 1 vs Table 4a case counts: NICCCS 142 vs 122, IIHD 543 vs 537. Table 1 sums to 839.
5. The 24-h recall sample is given as n = 2737 in the text, against 2679 in Table 1 and 2661 in Table 4a.
6. GI red meat: abstract 1.27 (1.02–1.58) vs text 1.27 (1.02–1.68). This is not CRC.