# K7 control – wie2014 (Wie et al., Br J Nutr 2014;112:238–247)

Full text: `wie2014.txt` and `wie2014_supplement.txt`.
Rows checked: 9 (wie2014_001–009). Each was checked against Table 3 (p. 244), Table 1 continued (p. 242), Supplement 1 and Supplement 5.

## Discrepancies

| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| – | – | – | – | none found |

## Number check
- 002: crude 1.05 (0.60, 1.82), 004: MV 1.31 (0.60, 2.61). These are Table 3, colorectal column, red meats ≥43 g/d (P = NS). Confirmed.
- n_cases 22 / 31 come from the Table 1 colorectal column (41.5 % / 58.5 %) and sum to 53, which is the colorectal n. Confirmed.
- person_years 23049.8 / 30976.8 come from Table 1 and are whole-cohort figures, not colorectal-specific. Confirmed; the notes say so.
- 004 log-symmetry: upper 0.69 vs lower 0.78 (12 %), which is below the 15 % threshold.
- 005: Suppl 1 colorectal, red meats, MV 0.98 (0.73, 1.31). The footnote gives the increment as **50 g/d for red meat**. Confirmed, with dose_unit = per_50g.
- 006–009: Suppl 5 colorectal, **per 10 g/d**:
  - fresh red meat: crude 0.98 (0.93 1.03), MV 0.99 (0.93 1.06)
  - processed red meat: crude 0.97 (0.86 1.08), MV 1.01 (0.90 1.14)

  All four confirmed with dose_unit = per_10g.

## Per-10 g vs per-50 g rows
The units are correctly separated. Total red meat (which includes processed meat and offal) is per 50 g/d (Suppl 1). Fresh and processed red meat are each per 10 g/d (Suppl 5) and are not mutually adjusted. **These rows must not be combined without rescaling.** A per-10 g HR of 0.99 corresponds to about 0.95 per 50 g. The notes record this correctly.

## Definitions, models and flags
- "Red meats" = beef, pork, lamb, goat, processed meats and offal (p. 240) → `red_total`, includes_processed = yes. Correct.
- "Fresh red meats" → `red_unprocessed`. Correct; whether offal is included is unstated, as the notes say.
- m1 is crude. m2 adjusts for age, sex, energy, BMI, physical activity, smoking, alcohol, income, education and marital status. This matches the footnotes of Table 3 and Suppl 1/5, and the flags are correct.
- is_main_model = yes for both the multivariable category set (003–004) and the continuous rows (005, 007, 009). These are the same model in different parameterisations, so this is acceptable.
- studies: 8024 participants, 53 CRC cases, median follow-up 7.0 y, baseline Sept 2004–Dec 2008, 3-day food record. Confirmed.
- The Table 3 header for colorectal reads "Person-years 53, median 0.1, range 0.0–6.0". The extractor reads this as follow-up among the cases, which I agree with: the thyroid column (136 cases, 223 person-years, median 0.7) fits the same reading. **This means most CRC cases were diagnosed at or right after baseline screening.** That is a substantive validity caveat for this study, and it is recorded.

## Missing estimates
None that the codebook requires. Table 3 also gives red meat ≥43 g/d for the combined outcome "gastric and colorectal cancers" (crude 1.11 (0.74, 1.65), MV 1.48 (0.83, 2.64)). That outcome is not in the vocabulary, so I did not count it as missing. Supplements 2–4 cover other outcomes.

## Confirmed rows
All 9 rows confirmed.

## Internal inconsistencies in the paper
1. Total person-years: 54 027 in the text vs 54 076 in Table 2.
2. Discussion: processed red meat in men is given as "HR 1·05, 95 % CI 1·00, 1·92"; Suppl 5 gives 1.05 (1.00 1.09). This is a typo in the Discussion and concerns all cancers, not CRC.
3. Table 1 lists female colorectal cases as 30 at "37.7 %", but 30/53 = 56.6 %. The count fits the gastric+CRC sum (45), so the percentage is wrong. This does not affect meat rows.