# K7 control – alrajabi2021 (Al Rajabi et al., Br J Nutr 2021, accepted manuscript)

Full text: `drive/post2015/alrajabi2021.txt`. Pages were checked against the form-feed page breaks in the file and against `pdf_in/alrajabi2021.pdf`.
Rows checked: 64 (alrajabi2021_001–064). Table 5 was parsed by script, and all 64 cells (point, CI, p-value) were matched to their rows: 0 mismatches, 0 unmatched cells.

## Discrepancies

| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| 033–048 (and by extension 001–016) | is_main_model / subgroup | is_main_model = yes on **three** different processed-meat variants that share exposure = `processed`, sex and outcome (red-derived 033–040, non-red-derived 041–048, combined 049–056). subgroup is empty, so the variants differ only by exposure_text and category_set_id. | Table 5 reports three separate processed-meat exposures. Codebook 4.2 allows one main model per exposure × outcome × sex. | **major (flag)**: a filter on exposure = processed & is_main_model = yes returns three sets per sex, including poultry-based "non-red" processed meat. Recommend: keep the combined variant (049–056) as main, and mark the red-derived and non-red-derived variants with a subgroup label (as Dankner does with `processed_poultry_only`) and/or is_main_model = no. |
| alrajabi2021 (studies) | cases_colorectal | 265 | Table 1: 119 men + 147 women, printed total 265 (sum = 266). Table 5 Model 1: 118 + 147 = 265. Model 2: 116 + 141 = 257. | minor: the value equals the Model 1 analysis total, which is acceptable. The inconsistency is noted in studies.notes. |

## Number check
All 64 rows match Table 5 (ms p. 37–38) exactly: point, CI and p-value. Reference rows are 1 with empty CI. p_value is stored for each non-reference row, and p_trend is empty, which is correct because no trend test is printed. The source_page split is also correct: Model 1 and Model 2 "processed meat from red meat" are on p. 37, and the rest of Model 2 is on p. 38.

## Processed-meat variants
- "Processed meat from red meat", "Processed meat from non-red meat" and "…red meat & non-red meat combined" are all coded `processed` with includes_processed = yes. Their category_set_ids are distinct (`_processed_red`, `_processed_nonred`, `_processed_all`). The quartile bounds (g/week, sex-specific) match the table for every row.
- "Red meat" → `red_unprocessed`, includes_processed = no. The definition is WCRF/AICR unprocessed mammalian muscle meat (Introduction). Supplementary Table SA1 is not available, so this follows the paper's statement.
- Only the combined variant corresponds to the conventional "processed meat". The non-red variant is mostly poultry-based and should not enter a red/processed-meat pool.

## Model 2 mutual adjustment
- The Table 5 footnote c lists "red meat, and all processed meat" for **every** row of Model 2.
- For red-meat rows (057–064): is_mutually_adjusted_processed = yes. Correct.
- For the processed rows (033–056): `unclear`. This is correct. Adjusting the combined variant for "all processed meat" would be adjusting for the exposure itself, so the generic footnote cannot be taken literally. Whether the red-derived and non-red-derived variants were adjusted for the other component, or for red meat, cannot be determined.
- adjustments m2 has adj_processed_meat = yes. This is literally correct for the red-meat rows and ambiguous for the processed rows. The notes say so.
- Model 2 is a complete-case analysis (116/141 vs 118/147 cases), so Model 1 and Model 2 are not fitted on identical samples.

## Other flags
- adj_age = no for both models is **correct and important**. For CRC the models were neither age-stratified nor age-adjusted (methods: PH assumption met, "without any age stratified adjustment"). The time scale was follow-up from enrollment, so age is uncontrolled in every CRC estimate.
- m2 flags confirmed against footnote c: energy, BMI, smoking, alcohol, physical activity, fruit/veg, education, family history (all cancers) and processed meat are yes; fibre is no (pulses and whole grains only); diabetes is unclear (composite health-condition variable).
- Firth penalised Cox models; sex-stratified. adj_sex = no is acceptable.
- studies fields confirmed: 26,218 participants, ages 35–69, enrolled 2000–2008, CDHQ-I with 257 items, registry linkage to June 2018, median follow-up 13.3 y.

## Missing estimates
None in the available text. Supplementary B (Tables SB1–SB4, adherence cut-offs of 500 g/wk red and 50 g/wk processed, with CRC included) and Supplementary C (competing risk) are not in the file, so they could not be checked. No per-category case counts, no person-years, no continuous estimates and no colon/rectum split are printed.

## Confirmed rows
All 64 rows are confirmed numerically. The flag issue on 033–048 is described above.

## Internal inconsistencies in the paper
1. Table 1 CRC: 119 + 147 is printed as a total of 265.
2. Results describe Q1 (the reference) in women as "mild intake (i.e., 1st quartile)" associated with increased risk for all cancers and fifteen cancers. Q1 is the reference category, so this is probably a mislabel of Q2. This does not concern the CRC rows.
3. The Results say the CRC association "persisted even after adjustment (Table 5 – Model 2)". Model 1 Q4 in women is 1.89 (1.16–3.06) and Model 2 is 1.90 (1.12–3.22), which is consistent.